2 * Copyright 2000, International Business Machines Corporation and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
9 * Portions Copyright (c) 2005-2008 Sine Nomine Associates
12 /* 1/1/89: NB: this stuff is all going to be replaced. Don't take it too seriously */
17 Institution: The Information Technology Center, Carnegie-Mellon University
21 #include <afsconfig.h>
22 #include <afs/param.h>
30 #ifdef HAVE_SYS_FILE_H
35 #include <afs/afsint.h>
38 #if !defined(AFS_SGI_ENV)
41 #else /* AFS_OSF_ENV */
42 #ifdef AFS_VFSINCL_ENV
45 #include <sys/fs/ufs_fs.h>
47 #if defined(AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV)
48 #include <ufs/ufs/dinode.h>
49 #include <ufs/ffs/fs.h>
54 #else /* AFS_VFSINCL_ENV */
55 #if !defined(AFS_AIX_ENV) && !defined(AFS_LINUX20_ENV) && !defined(AFS_XBSD_ENV) && !defined(AFS_DARWIN_ENV)
58 #endif /* AFS_VFSINCL_ENV */
59 #endif /* AFS_OSF_ENV */
60 #endif /* AFS_SGI_ENV */
61 #endif /* !AFS_NT40_ENV */
69 #if defined(AFS_SUN_ENV) || defined(AFS_SUN5_ENV)
71 #include <sys/mnttab.h>
72 #include <sys/mntent.h>
78 #if defined(AFS_SGI_ENV)
81 #ifndef AFS_LINUX20_ENV
82 #include <fstab.h> /* Need to find in libc 5, present in libc 6 */
85 #endif /* AFS_SGI_ENV */
87 #endif /* AFS_HPUX_ENV */
91 #include <afs/errors.h>
94 #include <afs/afssyscalls.h>
96 #include <afs/afsutil.h>
97 #include "daemon_com.h"
102 #include "partition.h"
103 #include "volume_inline.h"
108 #ifdef AFS_PTHREAD_ENV
109 pthread_mutex_t vol_glock_mutex;
110 pthread_mutex_t vol_trans_mutex;
111 pthread_cond_t vol_put_volume_cond;
112 pthread_cond_t vol_sleep_cond;
113 pthread_cond_t vol_init_attach_cond;
114 pthread_cond_t vol_vinit_cond;
115 int vol_attach_threads = 1;
116 #endif /* AFS_PTHREAD_ENV */
118 /* start-time configurable I/O parameters */
119 ih_init_params vol_io_params;
121 #ifdef AFS_DEMAND_ATTACH_FS
122 pthread_mutex_t vol_salvsync_mutex;
125 * Set this to 1 to disallow SALVSYNC communication in all threads; used
126 * during shutdown, since the salvageserver may have gone away.
128 static volatile sig_atomic_t vol_disallow_salvsync = 0;
129 #endif /* AFS_DEMAND_ATTACH_FS */
132 * has VShutdown_r been called / is VShutdown_r running?
134 static int vol_shutting_down = 0;
137 extern void *calloc(), *realloc();
140 /* Forward declarations */
141 static Volume *attach2(Error * ec, VolId volumeId, char *path,
142 struct DiskPartition64 *partp, Volume * vp,
143 int isbusy, int mode, int *acheckedOut);
144 static void ReallyFreeVolume(Volume * vp);
145 #ifdef AFS_DEMAND_ATTACH_FS
146 static void FreeVolume(Volume * vp);
147 #else /* !AFS_DEMAND_ATTACH_FS */
148 #define FreeVolume(vp) ReallyFreeVolume(vp)
149 static void VScanUpdateList(void);
150 #endif /* !AFS_DEMAND_ATTACH_FS */
151 static void VInitVolumeHeaderCache(afs_uint32 howMany);
152 static int GetVolumeHeader(Volume * vp);
153 static void ReleaseVolumeHeader(struct volHeader *hd);
154 static void FreeVolumeHeader(Volume * vp);
155 static void AddVolumeToHashTable(Volume * vp, int hashid);
156 static void DeleteVolumeFromHashTable(Volume * vp);
158 static int VHold(Volume * vp);
160 static int VHold_r(Volume * vp);
161 static void VGetBitmap_r(Error * ec, Volume * vp, VnodeClass class);
162 static void VReleaseVolumeHandles_r(Volume * vp);
163 static void VCloseVolumeHandles_r(Volume * vp);
164 static void LoadVolumeHeader(Error * ec, Volume * vp);
165 static int VCheckOffline(Volume * vp);
166 static int VCheckDetach(Volume * vp);
167 static Volume * GetVolume(Error * ec, Error * client_ec, VolId volumeId,
168 Volume * hint, const struct timespec *ts);
170 int LogLevel; /* Vice loglevel--not defined as extern so that it will be
171 * defined when not linked with vice, XXXX */
172 ProgramType programType; /* The type of program using the package */
173 static VolumePackageOptions vol_opts;
175 /* extended volume package statistics */
178 #ifdef VOL_LOCK_DEBUG
179 pthread_t vol_glock_holder = 0;
183 #define VOLUME_BITMAP_GROWSIZE 16 /* bytes, => 128vnodes */
184 /* Must be a multiple of 4 (1 word) !! */
186 /* this parameter needs to be tunable at runtime.
187 * 128 was really inadequate for largish servers -- at 16384 volumes this
188 * puts average chain length at 128, thus an average 65 deref's to find a volptr.
189 * talk about bad spatial locality...
191 * an AVL or splay tree might work a lot better, but we'll just increase
192 * the default hash table size for now
194 #define DEFAULT_VOLUME_HASH_SIZE 256 /* Must be a power of 2!! */
195 #define DEFAULT_VOLUME_HASH_MASK (DEFAULT_VOLUME_HASH_SIZE-1)
196 #define VOLUME_HASH(volumeId) (volumeId&(VolumeHashTable.Mask))
199 * turn volume hash chains into partially ordered lists.
200 * when the threshold is exceeded between two adjacent elements,
201 * perform a chain rebalancing operation.
203 * keep the threshold high in order to keep cache line invalidates
204 * low "enough" on SMPs
206 #define VOLUME_HASH_REORDER_THRESHOLD 200
209 * when possible, don't just reorder single elements, but reorder
210 * entire chains of elements at once. a chain of elements that
211 * exceed the element previous to the pivot by at least CHAIN_THRESH
212 * accesses are moved in front of the chain whose elements have at
213 * least CHAIN_THRESH less accesses than the pivot element
215 #define VOLUME_HASH_REORDER_CHAIN_THRESH (VOLUME_HASH_REORDER_THRESHOLD / 2)
217 #include "rx/rx_queue.h"
220 VolumeHashTable_t VolumeHashTable = {
221 DEFAULT_VOLUME_HASH_SIZE,
222 DEFAULT_VOLUME_HASH_MASK,
227 static void VInitVolumeHash(void);
231 /* This macro is used where an ffs() call does not exist. Was in util/ffs.c */
235 afs_int32 ffs_tmp = x;
239 for (ffs_i = 1;; ffs_i++) {
246 #endif /* !AFS_HAVE_FFS */
248 #ifdef AFS_PTHREAD_ENV
250 * disk partition queue element
252 typedef struct diskpartition_queue_t {
253 struct rx_queue queue; /**< queue header */
254 struct DiskPartition64 *diskP; /**< disk partition table entry */
255 } diskpartition_queue_t;
257 #ifndef AFS_DEMAND_ATTACH_FS
259 typedef struct vinitvolumepackage_thread_t {
260 struct rx_queue queue;
261 pthread_cond_t thread_done_cv;
262 int n_threads_complete;
263 } vinitvolumepackage_thread_t;
264 static void * VInitVolumePackageThread(void * args);
266 #else /* !AFS_DEMAND_ATTTACH_FS */
267 #define VINIT_BATCH_MAX_SIZE 512
270 * disk partition work queue
272 struct partition_queue {
273 struct rx_queue head; /**< diskpartition_queue_t queue */
274 pthread_mutex_t mutex;
279 * volumes parameters for preattach
281 struct volume_init_batch {
282 struct rx_queue queue; /**< queue header */
283 int thread; /**< posting worker thread */
284 int last; /**< indicates thread is done */
285 int size; /**< number of volume ids in batch */
286 Volume *batch[VINIT_BATCH_MAX_SIZE]; /**< volumes ids to preattach */
290 * volume parameters work queue
292 struct volume_init_queue {
293 struct rx_queue head; /**< volume_init_batch queue */
294 pthread_mutex_t mutex;
299 * volume init worker thread parameters
301 struct vinitvolumepackage_thread_param {
302 int nthreads; /**< total number of worker threads */
303 int thread; /**< thread number for this worker thread */
304 struct partition_queue *pq; /**< queue partitions to scan */
305 struct volume_init_queue *vq; /**< queue of volume to preattach */
308 static void *VInitVolumePackageThread(void *args);
309 static struct DiskPartition64 *VInitNextPartition(struct partition_queue *pq);
310 static VolId VInitNextVolumeId(DIR *dirp);
311 static int VInitPreAttachVolumes(int nthreads, struct volume_init_queue *vq);
313 #endif /* !AFS_DEMAND_ATTACH_FS */
314 #endif /* AFS_PTHREAD_ENV */
316 #ifndef AFS_DEMAND_ATTACH_FS
317 static int VAttachVolumesByPartition(struct DiskPartition64 *diskP,
318 int * nAttached, int * nUnattached);
319 #endif /* AFS_DEMAND_ATTACH_FS */
322 #ifdef AFS_DEMAND_ATTACH_FS
323 /* demand attach fileserver extensions */
326 * in the future we will support serialization of VLRU state into the fs_state
329 * these structures are the beginning of that effort
331 struct VLRU_DiskHeader {
332 struct versionStamp stamp; /* magic and structure version number */
333 afs_uint32 mtime; /* time of dump to disk */
334 afs_uint32 num_records; /* number of VLRU_DiskEntry records */
337 struct VLRU_DiskEntry {
338 afs_uint32 vid; /* volume ID */
339 afs_uint32 idx; /* generation */
340 afs_uint32 last_get; /* timestamp of last get */
343 struct VLRU_StartupQueue {
344 struct VLRU_DiskEntry * entry;
349 typedef struct vshutdown_thread_t {
351 pthread_mutex_t lock;
353 pthread_cond_t master_cv;
355 int n_threads_complete;
357 int schedule_version;
360 byte n_parts_done_pass;
361 byte part_thread_target[VOLMAXPARTS+1];
362 byte part_done_pass[VOLMAXPARTS+1];
363 struct rx_queue * part_pass_head[VOLMAXPARTS+1];
364 int stats[4][VOLMAXPARTS+1];
365 } vshutdown_thread_t;
366 static void * VShutdownThread(void * args);
369 static Volume * VAttachVolumeByVp_r(Error * ec, Volume * vp, int mode);
370 static int VCheckFree(Volume * vp);
373 static void AddVolumeToVByPList_r(Volume * vp);
374 static void DeleteVolumeFromVByPList_r(Volume * vp);
375 static void VVByPListBeginExclusive_r(struct DiskPartition64 * dp);
376 static void VVByPListEndExclusive_r(struct DiskPartition64 * dp);
377 static void VVByPListWait_r(struct DiskPartition64 * dp);
379 /* online salvager */
380 static int VCheckSalvage(Volume * vp);
381 #if defined(SALVSYNC_BUILD_CLIENT) || defined(FSSYNC_BUILD_CLIENT)
382 static int VScheduleSalvage_r(Volume * vp);
385 /* Volume hash table */
386 static void VReorderHash_r(VolumeHashChainHead * head, Volume * pp, Volume * vp);
387 static void VHashBeginExclusive_r(VolumeHashChainHead * head);
388 static void VHashEndExclusive_r(VolumeHashChainHead * head);
389 static void VHashWait_r(VolumeHashChainHead * head);
392 static int ShutdownVByPForPass_r(struct DiskPartition64 * dp, int pass);
393 static int ShutdownVolumeWalk_r(struct DiskPartition64 * dp, int pass,
394 struct rx_queue ** idx);
395 static void ShutdownController(vshutdown_thread_t * params);
396 static void ShutdownCreateSchedule(vshutdown_thread_t * params);
399 static void VLRU_ComputeConstants(void);
400 static void VInitVLRU(void);
401 static void VLRU_Init_Node_r(Volume * vp);
402 static void VLRU_Add_r(Volume * vp);
403 static void VLRU_Delete_r(Volume * vp);
404 static void VLRU_UpdateAccess_r(Volume * vp);
405 static void * VLRU_ScannerThread(void * args);
406 static void VLRU_Scan_r(int idx);
407 static void VLRU_Promote_r(int idx);
408 static void VLRU_Demote_r(int idx);
409 static void VLRU_SwitchQueues(Volume * vp, int new_idx, int append);
412 static int VCheckSoftDetach(Volume * vp, afs_uint32 thresh);
413 static int VCheckSoftDetachCandidate(Volume * vp, afs_uint32 thresh);
414 static int VSoftDetachVolume_r(Volume * vp, afs_uint32 thresh);
417 pthread_key_t VThread_key;
418 VThreadOptions_t VThread_defaults = {
419 0 /**< allow salvsync */
421 #endif /* AFS_DEMAND_ATTACH_FS */
424 struct Lock vol_listLock; /* Lock obtained when listing volumes:
425 * prevents a volume from being missed
426 * if the volume is attached during a
430 /* Common message used when the volume goes off line */
431 char *VSalvageMessage =
432 "Files in this volume are currently unavailable; call operations";
434 int VInit; /* 0 - uninitialized,
435 * 1 - initialized but not all volumes have been attached,
436 * 2 - initialized and all volumes have been attached,
437 * 3 - initialized, all volumes have been attached, and
438 * VConnectFS() has completed. */
440 static int vinit_attach_abort = 0;
442 bit32 VolumeCacheCheck; /* Incremented everytime a volume goes on line--
443 * used to stamp volume headers and in-core
444 * vnodes. When the volume goes on-line the
445 * vnode will be invalidated
446 * access only with VOL_LOCK held */
451 /***************************************************/
452 /* Startup routines */
453 /***************************************************/
455 #if defined(FAST_RESTART) && defined(AFS_DEMAND_ATTACH_FS)
456 # error FAST_RESTART and DAFS are incompatible. For the DAFS equivalent \
457 of FAST_RESTART, use the -unsafe-nosalvage fileserver argument
461 * assign default values to a VolumePackageOptions struct.
463 * Always call this on a VolumePackageOptions struct first, then set any
464 * specific options you want, then call VInitVolumePackage2.
466 * @param[in] pt caller's program type
467 * @param[out] opts volume package options
470 VOptDefaults(ProgramType pt, VolumePackageOptions *opts)
472 opts->nLargeVnodes = opts->nSmallVnodes = 5;
475 opts->canScheduleSalvage = 0;
476 opts->canUseFSSYNC = 0;
477 opts->canUseSALVSYNC = 0;
479 opts->interrupt_rxcall = NULL;
480 opts->offline_timeout = -1;
481 opts->offline_shutdown_timeout = -1;
484 opts->unsafe_attach = 1;
485 #else /* !FAST_RESTART */
486 opts->unsafe_attach = 0;
487 #endif /* !FAST_RESTART */
491 opts->canScheduleSalvage = 1;
492 opts->canUseSALVSYNC = 1;
496 opts->canUseFSSYNC = 1;
500 opts->nLargeVnodes = 0;
501 opts->nSmallVnodes = 0;
503 opts->canScheduleSalvage = 1;
504 opts->canUseFSSYNC = 1;
514 * Set VInit to a certain value, and signal waiters.
516 * @param[in] value the value to set VInit to
521 VSetVInit_r(int value)
524 CV_BROADCAST(&vol_vinit_cond);
528 VLogOfflineTimeout(const char *type, afs_int32 timeout)
534 Log("VInitVolumePackage: Interrupting clients accessing %s "
535 "immediately\n", type);
537 Log("VInitVolumePackage: Interrupting clients accessing %s "
538 "after %ld second%s\n", type, (long)timeout, timeout==1?"":"s");
543 VInitVolumePackage2(ProgramType pt, VolumePackageOptions * opts)
545 int errors = 0; /* Number of errors while finding vice partitions. */
550 #ifndef AFS_PTHREAD_ENV
551 if (opts->offline_timeout != -1 || opts->offline_shutdown_timeout != -1) {
552 Log("VInitVolumePackage: offline_timeout and/or "
553 "offline_shutdown_timeout was specified, but the volume package "
554 "does not support these for LWP builds\n");
558 VLogOfflineTimeout("volumes going offline", opts->offline_timeout);
559 VLogOfflineTimeout("volumes going offline during shutdown",
560 opts->offline_shutdown_timeout);
562 memset(&VStats, 0, sizeof(VStats));
563 VStats.hdr_cache_size = 200;
565 VInitPartitionPackage();
567 #ifdef AFS_DEMAND_ATTACH_FS
568 if (programType == fileServer) {
571 VLRU_SetOptions(VLRU_SET_ENABLED, 0);
573 osi_Assert(pthread_key_create(&VThread_key, NULL) == 0);
576 MUTEX_INIT(&vol_glock_mutex, "vol glock", MUTEX_DEFAULT, 0);
577 MUTEX_INIT(&vol_trans_mutex, "vol trans", MUTEX_DEFAULT, 0);
578 CV_INIT(&vol_put_volume_cond, "vol put", CV_DEFAULT, 0);
579 CV_INIT(&vol_sleep_cond, "vol sleep", CV_DEFAULT, 0);
580 CV_INIT(&vol_init_attach_cond, "vol init attach", CV_DEFAULT, 0);
581 CV_INIT(&vol_vinit_cond, "vol init", CV_DEFAULT, 0);
582 #ifndef AFS_PTHREAD_ENV
584 #endif /* AFS_PTHREAD_ENV */
585 Lock_Init(&vol_listLock);
587 srandom(time(0)); /* For VGetVolumeInfo */
589 #ifdef AFS_DEMAND_ATTACH_FS
590 MUTEX_INIT(&vol_salvsync_mutex, "salvsync", MUTEX_DEFAULT, 0);
591 #endif /* AFS_DEMAND_ATTACH_FS */
593 /* Ok, we have done enough initialization that fileserver can
594 * start accepting calls, even though the volumes may not be
595 * available just yet.
599 #if defined(AFS_DEMAND_ATTACH_FS) && defined(SALVSYNC_BUILD_SERVER)
600 if (programType == salvageServer) {
603 #endif /* AFS_DEMAND_ATTACH_FS */
604 #ifdef FSSYNC_BUILD_SERVER
605 if (programType == fileServer) {
609 #if defined(AFS_DEMAND_ATTACH_FS) && defined(SALVSYNC_BUILD_CLIENT)
610 if (VCanUseSALVSYNC()) {
611 /* establish a connection to the salvager at this point */
612 osi_Assert(VConnectSALV() != 0);
614 #endif /* AFS_DEMAND_ATTACH_FS */
616 if (opts->volcache > VStats.hdr_cache_size)
617 VStats.hdr_cache_size = opts->volcache;
618 VInitVolumeHeaderCache(VStats.hdr_cache_size);
620 VInitVnodes(vLarge, opts->nLargeVnodes);
621 VInitVnodes(vSmall, opts->nSmallVnodes);
624 errors = VAttachPartitions();
628 if (programType != fileServer) {
629 errors = VInitAttachVolumes(programType);
635 #ifdef FSSYNC_BUILD_CLIENT
636 if (VCanUseFSSYNC()) {
638 #ifdef AFS_DEMAND_ATTACH_FS
639 if (programType == salvageServer) {
640 Log("Unable to connect to file server; aborted\n");
643 #endif /* AFS_DEMAND_ATTACH_FS */
644 Log("Unable to connect to file server; will retry at need\n");
647 #endif /* FSSYNC_BUILD_CLIENT */
652 #if !defined(AFS_PTHREAD_ENV)
654 * Attach volumes in vice partitions
656 * @param[in] pt calling program type
659 * @note This is the original, non-threaded version of attach parititions.
661 * @post VInit state is 2
664 VInitAttachVolumes(ProgramType pt)
666 osi_Assert(VInit==1);
667 if (pt == fileServer) {
668 struct DiskPartition64 *diskP;
669 /* Attach all the volumes in this partition */
670 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
671 int nAttached = 0, nUnattached = 0;
672 osi_Assert(VAttachVolumesByPartition(diskP, &nAttached, &nUnattached) == 0);
676 VSetVInit_r(2); /* Initialized, and all volumes have been attached */
677 LWP_NoYieldSignal(VInitAttachVolumes);
681 #endif /* !AFS_PTHREAD_ENV */
683 #if defined(AFS_PTHREAD_ENV) && !defined(AFS_DEMAND_ATTACH_FS)
685 * Attach volumes in vice partitions
687 * @param[in] pt calling program type
690 * @note Threaded version of attach parititions.
692 * @post VInit state is 2
695 VInitAttachVolumes(ProgramType pt)
697 osi_Assert(VInit==1);
698 if (pt == fileServer) {
699 struct DiskPartition64 *diskP;
700 struct vinitvolumepackage_thread_t params;
701 struct diskpartition_queue_t * dpq;
702 int i, threads, parts;
704 pthread_attr_t attrs;
706 CV_INIT(¶ms.thread_done_cv, "thread done", CV_DEFAULT, 0);
708 params.n_threads_complete = 0;
710 /* create partition work queue */
711 for (parts=0, diskP = DiskPartitionList; diskP; diskP = diskP->next, parts++) {
712 dpq = (diskpartition_queue_t *) malloc(sizeof(struct diskpartition_queue_t));
713 osi_Assert(dpq != NULL);
715 queue_Append(¶ms,dpq);
718 threads = MIN(parts, vol_attach_threads);
721 /* spawn off a bunch of initialization threads */
722 osi_Assert(pthread_attr_init(&attrs) == 0);
723 osi_Assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
725 Log("VInitVolumePackage: beginning parallel fileserver startup\n");
726 Log("VInitVolumePackage: using %d threads to attach volumes on %d partitions\n",
730 for (i=0; i < threads; i++) {
733 osi_Assert(pthread_create
734 (&tid, &attrs, &VInitVolumePackageThread,
736 AFS_SIGSET_RESTORE();
739 while(params.n_threads_complete < threads) {
740 VOL_CV_WAIT(¶ms.thread_done_cv);
744 osi_Assert(pthread_attr_destroy(&attrs) == 0);
746 /* if we're only going to run one init thread, don't bother creating
748 Log("VInitVolumePackage: beginning single-threaded fileserver startup\n");
749 Log("VInitVolumePackage: using 1 thread to attach volumes on %d partition(s)\n",
752 VInitVolumePackageThread(¶ms);
755 CV_DESTROY(¶ms.thread_done_cv);
758 VSetVInit_r(2); /* Initialized, and all volumes have been attached */
759 CV_BROADCAST(&vol_init_attach_cond);
765 VInitVolumePackageThread(void * args) {
767 struct DiskPartition64 *diskP;
768 struct vinitvolumepackage_thread_t * params;
769 struct diskpartition_queue_t * dpq;
771 params = (vinitvolumepackage_thread_t *) args;
775 /* Attach all the volumes in this partition */
776 while (queue_IsNotEmpty(params)) {
777 int nAttached = 0, nUnattached = 0;
779 if (vinit_attach_abort) {
780 Log("Aborting initialization\n");
784 dpq = queue_First(params,diskpartition_queue_t);
790 osi_Assert(VAttachVolumesByPartition(diskP, &nAttached, &nUnattached) == 0);
796 params->n_threads_complete++;
797 CV_SIGNAL(¶ms->thread_done_cv);
801 #endif /* AFS_PTHREAD_ENV && !AFS_DEMAND_ATTACH_FS */
803 #if defined(AFS_DEMAND_ATTACH_FS)
805 * Attach volumes in vice partitions
807 * @param[in] pt calling program type
810 * @note Threaded version of attach partitions.
812 * @post VInit state is 2
815 VInitAttachVolumes(ProgramType pt)
817 osi_Assert(VInit==1);
818 if (pt == fileServer) {
820 struct DiskPartition64 *diskP;
821 struct partition_queue pq;
822 struct volume_init_queue vq;
824 int i, threads, parts;
826 pthread_attr_t attrs;
828 /* create partition work queue */
830 CV_INIT(&(pq.cv), "partq", CV_DEFAULT, 0);
831 MUTEX_INIT(&(pq.mutex), "partq", MUTEX_DEFAULT, 0);
832 for (parts = 0, diskP = DiskPartitionList; diskP; diskP = diskP->next, parts++) {
833 struct diskpartition_queue_t *dp;
834 dp = (struct diskpartition_queue_t*)malloc(sizeof(struct diskpartition_queue_t));
835 osi_Assert(dp != NULL);
837 queue_Append(&pq, dp);
840 /* number of worker threads; at least one, not to exceed the number of partitions */
841 threads = MIN(parts, vol_attach_threads);
843 /* create volume work queue */
845 CV_INIT(&(vq.cv), "volq", CV_DEFAULT, 0);
846 MUTEX_INIT(&(vq.mutex), "volq", MUTEX_DEFAULT, 0);
848 osi_Assert(pthread_attr_init(&attrs) == 0);
849 osi_Assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
851 Log("VInitVolumePackage: beginning parallel fileserver startup\n");
852 Log("VInitVolumePackage: using %d threads to pre-attach volumes on %d partitions\n",
855 /* create threads to scan disk partitions. */
856 for (i=0; i < threads; i++) {
857 struct vinitvolumepackage_thread_param *params;
860 params = (struct vinitvolumepackage_thread_param *)malloc(sizeof(struct vinitvolumepackage_thread_param));
864 params->nthreads = threads;
865 params->thread = i+1;
868 osi_Assert(pthread_create (&tid, &attrs, &VInitVolumePackageThread, (void*)params) == 0);
869 AFS_SIGSET_RESTORE();
872 VInitPreAttachVolumes(threads, &vq);
874 osi_Assert(pthread_attr_destroy(&attrs) == 0);
876 MUTEX_DESTROY(&pq.mutex);
878 MUTEX_DESTROY(&vq.mutex);
882 VSetVInit_r(2); /* Initialized, and all volumes have been attached */
883 CV_BROADCAST(&vol_init_attach_cond);
890 * Volume package initialization worker thread. Scan partitions for volume
891 * header files. Gather batches of volume ids and dispatch them to
892 * the main thread to be preattached. The volume preattachement is done
893 * in the main thread to avoid global volume lock contention.
896 VInitVolumePackageThread(void *args)
898 struct vinitvolumepackage_thread_param *params;
899 struct DiskPartition64 *partition;
900 struct partition_queue *pq;
901 struct volume_init_queue *vq;
902 struct volume_init_batch *vb;
905 params = (struct vinitvolumepackage_thread_param *)args;
911 vb = (struct volume_init_batch*)malloc(sizeof(struct volume_init_batch));
913 vb->thread = params->thread;
917 Log("Scanning partitions on thread %d of %d\n", params->thread, params->nthreads);
918 while((partition = VInitNextPartition(pq))) {
922 Log("Partition %s: pre-attaching volumes\n", partition->name);
923 dirp = opendir(VPartitionPath(partition));
925 Log("opendir on Partition %s failed, errno=%d!\n", partition->name, errno);
928 while ((vid = VInitNextVolumeId(dirp))) {
929 Volume *vp = (Volume*)malloc(sizeof(Volume));
931 memset(vp, 0, sizeof(Volume));
932 vp->device = partition->device;
933 vp->partition = partition;
935 queue_Init(&vp->vnode_list);
936 queue_Init(&vp->rx_call_list);
937 CV_INIT(&V_attachCV(vp), "partattach", CV_DEFAULT, 0);
939 vb->batch[vb->size++] = vp;
940 if (vb->size == VINIT_BATCH_MAX_SIZE) {
941 MUTEX_ENTER(&vq->mutex);
942 queue_Append(vq, vb);
943 CV_BROADCAST(&vq->cv);
944 MUTEX_EXIT(&vq->mutex);
946 vb = (struct volume_init_batch*)malloc(sizeof(struct volume_init_batch));
948 vb->thread = params->thread;
957 MUTEX_ENTER(&vq->mutex);
958 queue_Append(vq, vb);
959 CV_BROADCAST(&vq->cv);
960 MUTEX_EXIT(&vq->mutex);
962 Log("Partition scan thread %d of %d ended\n", params->thread, params->nthreads);
968 * Read next element from the pre-populated partition list.
970 static struct DiskPartition64*
971 VInitNextPartition(struct partition_queue *pq)
973 struct DiskPartition64 *partition;
974 struct diskpartition_queue_t *dp; /* queue element */
976 if (vinit_attach_abort) {
977 Log("Aborting volume preattach thread.\n");
981 /* get next partition to scan */
982 MUTEX_ENTER(&pq->mutex);
983 if (queue_IsEmpty(pq)) {
984 MUTEX_EXIT(&pq->mutex);
987 dp = queue_First(pq, diskpartition_queue_t);
989 MUTEX_EXIT(&pq->mutex);
992 osi_Assert(dp->diskP);
994 partition = dp->diskP;
1000 * Find next volume id on the partition.
1003 VInitNextVolumeId(DIR *dirp)
1009 while((d = readdir(dirp))) {
1010 if (vinit_attach_abort) {
1011 Log("Aborting volume preattach thread.\n");
1014 ext = strrchr(d->d_name, '.');
1015 if (d->d_name[0] == 'V' && ext && strcmp(ext, VHDREXT) == 0) {
1016 vid = VolumeNumber(d->d_name);
1020 Log("Warning: bogus volume header file: %s\n", d->d_name);
1027 * Preattach volumes in batches to avoid lock contention.
1030 VInitPreAttachVolumes(int nthreads, struct volume_init_queue *vq)
1032 struct volume_init_batch *vb;
1036 /* dequeue next volume */
1037 MUTEX_ENTER(&vq->mutex);
1038 if (queue_IsEmpty(vq)) {
1039 CV_WAIT(&vq->cv, &vq->mutex);
1041 vb = queue_First(vq, volume_init_batch);
1043 MUTEX_EXIT(&vq->mutex);
1047 for (i = 0; i<vb->size; i++) {
1053 dup = VLookupVolume_r(&ec, vp->hashid, NULL);
1055 Log("Error looking up volume, code=%d\n", ec);
1058 Log("Warning: Duplicate volume id %d detected.\n", vp->hashid);
1061 /* put pre-attached volume onto the hash table
1062 * and bring it up to the pre-attached state */
1063 AddVolumeToHashTable(vp, vp->hashid);
1064 AddVolumeToVByPList_r(vp);
1065 VLRU_Init_Node_r(vp);
1066 VChangeState_r(vp, VOL_STATE_PREATTACHED);
1079 #endif /* AFS_DEMAND_ATTACH_FS */
1081 #if !defined(AFS_DEMAND_ATTACH_FS)
1083 * attach all volumes on a given disk partition
1086 VAttachVolumesByPartition(struct DiskPartition64 *diskP, int * nAttached, int * nUnattached)
1092 Log("Partition %s: attaching volumes\n", diskP->name);
1093 dirp = opendir(VPartitionPath(diskP));
1095 Log("opendir on Partition %s failed!\n", diskP->name);
1099 while ((dp = readdir(dirp))) {
1101 p = strrchr(dp->d_name, '.');
1103 if (vinit_attach_abort) {
1104 Log("Partition %s: abort attach volumes\n", diskP->name);
1108 if (p != NULL && strcmp(p, VHDREXT) == 0) {
1111 vp = VAttachVolumeByName(&error, diskP->name, dp->d_name,
1113 (*(vp ? nAttached : nUnattached))++;
1114 if (error == VOFFLINE)
1115 Log("Volume %d stays offline (/vice/offline/%s exists)\n", VolumeNumber(dp->d_name), dp->d_name);
1116 else if (LogLevel >= 5) {
1117 Log("Partition %s: attached volume %d (%s)\n",
1118 diskP->name, VolumeNumber(dp->d_name),
1127 Log("Partition %s: attached %d volumes; %d volumes not attached\n", diskP->name, *nAttached, *nUnattached);
1132 #endif /* !AFS_DEMAND_ATTACH_FS */
1134 /***************************************************/
1135 /* Shutdown routines */
1136 /***************************************************/
1140 * highly multithreaded volume package shutdown
1142 * with the demand attach fileserver extensions,
1143 * VShutdown has been modified to be multithreaded.
1144 * In order to achieve optimal use of many threads,
1145 * the shutdown code involves one control thread and
1146 * n shutdown worker threads. The control thread
1147 * periodically examines the number of volumes available
1148 * for shutdown on each partition, and produces a worker
1149 * thread allocation schedule. The idea is to eliminate
1150 * redundant scheduling computation on the workers by
1151 * having a single master scheduler.
1153 * The scheduler's objectives are:
1155 * each partition with volumes remaining gets allocated
1156 * at least 1 thread (assuming sufficient threads)
1158 * threads are allocated proportional to the number of
1159 * volumes remaining to be offlined. This ensures that
1160 * the OS I/O scheduler has many requests to elevator
1161 * seek on partitions that will (presumably) take the
1162 * longest amount of time (from now) to finish shutdown
1163 * (3) keep threads busy
1164 * when there are extra threads, they are assigned to
1165 * partitions using a simple round-robin algorithm
1167 * In the future, we may wish to add the ability to adapt
1168 * to the relative performance patterns of each disk
1173 * multi-step shutdown process
1175 * demand attach shutdown is a four-step process. Each
1176 * shutdown "pass" shuts down increasingly more difficult
1177 * volumes. The main purpose is to achieve better cache
1178 * utilization during shutdown.
1181 * shutdown volumes in the unattached, pre-attached
1184 * shutdown attached volumes with cached volume headers
1186 * shutdown all volumes in non-exclusive states
1188 * shutdown all remaining volumes
1191 #ifdef AFS_DEMAND_ATTACH_FS
1197 struct DiskPartition64 * diskP;
1198 struct diskpartition_queue_t * dpq;
1199 vshutdown_thread_t params;
1201 pthread_attr_t attrs;
1203 memset(¶ms, 0, sizeof(vshutdown_thread_t));
1206 Log("VShutdown: aborting attach volumes\n");
1207 vinit_attach_abort = 1;
1208 VOL_CV_WAIT(&vol_init_attach_cond);
1211 for (params.n_parts=0, diskP = DiskPartitionList;
1212 diskP; diskP = diskP->next, params.n_parts++);
1214 Log("VShutdown: shutting down on-line volumes on %d partition%s...\n",
1215 params.n_parts, params.n_parts > 1 ? "s" : "");
1217 vol_shutting_down = 1;
1219 if (vol_attach_threads > 1) {
1220 /* prepare for parallel shutdown */
1221 params.n_threads = vol_attach_threads;
1222 MUTEX_INIT(¶ms.lock, "params", MUTEX_DEFAULT, 0);
1223 CV_INIT(¶ms.cv, "params", CV_DEFAULT, 0);
1224 CV_INIT(¶ms.master_cv, "params master", CV_DEFAULT, 0);
1225 osi_Assert(pthread_attr_init(&attrs) == 0);
1226 osi_Assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
1227 queue_Init(¶ms);
1229 /* setup the basic partition information structures for
1230 * parallel shutdown */
1231 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1233 struct rx_queue * qp, * nqp;
1237 VVByPListWait_r(diskP);
1238 VVByPListBeginExclusive_r(diskP);
1241 for (queue_Scan(&diskP->vol_list, qp, nqp, rx_queue)) {
1242 vp = (Volume *)((char *)qp - offsetof(Volume, vol_list));
1246 Log("VShutdown: partition %s has %d volumes with attached headers\n",
1247 VPartitionPath(diskP), count);
1250 /* build up the pass 0 shutdown work queue */
1251 dpq = (struct diskpartition_queue_t *) malloc(sizeof(struct diskpartition_queue_t));
1252 osi_Assert(dpq != NULL);
1254 queue_Prepend(¶ms, dpq);
1256 params.part_pass_head[diskP->index] = queue_First(&diskP->vol_list, rx_queue);
1259 Log("VShutdown: beginning parallel fileserver shutdown\n");
1260 Log("VShutdown: using %d threads to offline volumes on %d partition%s\n",
1261 vol_attach_threads, params.n_parts, params.n_parts > 1 ? "s" : "" );
1263 /* do pass 0 shutdown */
1264 MUTEX_ENTER(¶ms.lock);
1265 for (i=0; i < params.n_threads; i++) {
1266 osi_Assert(pthread_create
1267 (&tid, &attrs, &VShutdownThread,
1271 /* wait for all the pass 0 shutdowns to complete */
1272 while (params.n_threads_complete < params.n_threads) {
1273 CV_WAIT(¶ms.master_cv, ¶ms.lock);
1275 params.n_threads_complete = 0;
1277 CV_BROADCAST(¶ms.cv);
1278 MUTEX_EXIT(¶ms.lock);
1280 Log("VShutdown: pass 0 completed using the 1 thread per partition algorithm\n");
1281 Log("VShutdown: starting passes 1 through 3 using finely-granular mp-fast algorithm\n");
1283 /* run the parallel shutdown scheduler. it will drop the glock internally */
1284 ShutdownController(¶ms);
1286 /* wait for all the workers to finish pass 3 and terminate */
1287 while (params.pass < 4) {
1288 VOL_CV_WAIT(¶ms.cv);
1291 osi_Assert(pthread_attr_destroy(&attrs) == 0);
1292 CV_DESTROY(¶ms.cv);
1293 CV_DESTROY(¶ms.master_cv);
1294 MUTEX_DESTROY(¶ms.lock);
1296 /* drop the VByPList exclusive reservations */
1297 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1298 VVByPListEndExclusive_r(diskP);
1299 Log("VShutdown: %s stats : (pass[0]=%d, pass[1]=%d, pass[2]=%d, pass[3]=%d)\n",
1300 VPartitionPath(diskP),
1301 params.stats[0][diskP->index],
1302 params.stats[1][diskP->index],
1303 params.stats[2][diskP->index],
1304 params.stats[3][diskP->index]);
1307 Log("VShutdown: shutdown finished using %d threads\n", params.n_threads);
1309 /* if we're only going to run one shutdown thread, don't bother creating
1311 Log("VShutdown: beginning single-threaded fileserver shutdown\n");
1313 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1314 VShutdownByPartition_r(diskP);
1318 Log("VShutdown: complete.\n");
1321 #else /* AFS_DEMAND_ATTACH_FS */
1331 Log("VShutdown: aborting attach volumes\n");
1332 vinit_attach_abort = 1;
1333 #ifdef AFS_PTHREAD_ENV
1334 VOL_CV_WAIT(&vol_init_attach_cond);
1336 LWP_WaitProcess(VInitAttachVolumes);
1337 #endif /* AFS_PTHREAD_ENV */
1340 Log("VShutdown: shutting down on-line volumes...\n");
1341 vol_shutting_down = 1;
1342 for (i = 0; i < VolumeHashTable.Size; i++) {
1343 /* try to hold first volume in the hash table */
1344 for (queue_Scan(&VolumeHashTable.Table[i],vp,np,Volume)) {
1348 Log("VShutdown: Attempting to take volume %u offline.\n",
1351 /* next, take the volume offline (drops reference count) */
1352 VOffline_r(vp, "File server was shut down");
1356 Log("VShutdown: complete.\n");
1358 #endif /* AFS_DEMAND_ATTACH_FS */
1364 osi_Assert(VInit>0);
1371 * stop new activity (e.g. SALVSYNC) from occurring
1373 * Use this to make the volume package less busy; for example, during
1374 * shutdown. This doesn't actually shutdown/detach anything in the
1375 * volume package, but prevents certain processes from ocurring. For
1376 * example, preventing new SALVSYNC communication in DAFS. In theory, we
1377 * could also use this to prevent new volume attachment, or prevent
1378 * other programs from checking out volumes, etc.
1383 #ifdef AFS_DEMAND_ATTACH_FS
1384 /* make sure we don't try to contact the salvageserver, since it may
1385 * not be around anymore */
1386 vol_disallow_salvsync = 1;
1390 #ifdef AFS_DEMAND_ATTACH_FS
1393 * shutdown control thread
1396 ShutdownController(vshutdown_thread_t * params)
1399 struct DiskPartition64 * diskP;
1401 vshutdown_thread_t shadow;
1403 ShutdownCreateSchedule(params);
1405 while ((params->pass < 4) &&
1406 (params->n_threads_complete < params->n_threads)) {
1407 /* recompute schedule once per second */
1409 memcpy(&shadow, params, sizeof(vshutdown_thread_t));
1413 Log("ShutdownController: schedule version=%d, vol_remaining=%d, pass=%d\n",
1414 shadow.schedule_version, shadow.vol_remaining, shadow.pass);
1415 Log("ShutdownController: n_threads_complete=%d, n_parts_done_pass=%d\n",
1416 shadow.n_threads_complete, shadow.n_parts_done_pass);
1417 for (diskP = DiskPartitionList; diskP; diskP=diskP->next) {
1419 Log("ShutdownController: part[%d] : (len=%d, thread_target=%d, done_pass=%d, pass_head=%p)\n",
1421 diskP->vol_list.len,
1422 shadow.part_thread_target[id],
1423 shadow.part_done_pass[id],
1424 shadow.part_pass_head[id]);
1430 ShutdownCreateSchedule(params);
1434 /* create the shutdown thread work schedule.
1435 * this scheduler tries to implement fairness
1436 * by allocating at least 1 thread to each
1437 * partition with volumes to be shutdown,
1438 * and then it attempts to allocate remaining
1439 * threads based upon the amount of work left
1442 ShutdownCreateSchedule(vshutdown_thread_t * params)
1444 struct DiskPartition64 * diskP;
1445 int sum, thr_workload, thr_left;
1446 int part_residue[VOLMAXPARTS+1];
1449 /* compute the total number of outstanding volumes */
1451 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1452 sum += diskP->vol_list.len;
1455 params->schedule_version++;
1456 params->vol_remaining = sum;
1461 /* compute average per-thread workload */
1462 thr_workload = sum / params->n_threads;
1463 if (sum % params->n_threads)
1466 thr_left = params->n_threads;
1467 memset(&part_residue, 0, sizeof(part_residue));
1469 /* for fairness, give every partition with volumes remaining
1470 * at least one thread */
1471 for (diskP = DiskPartitionList; diskP && thr_left; diskP = diskP->next) {
1473 if (diskP->vol_list.len) {
1474 params->part_thread_target[id] = 1;
1477 params->part_thread_target[id] = 0;
1481 if (thr_left && thr_workload) {
1482 /* compute length-weighted workloads */
1485 for (diskP = DiskPartitionList; diskP && thr_left; diskP = diskP->next) {
1487 delta = (diskP->vol_list.len / thr_workload) -
1488 params->part_thread_target[id];
1492 if (delta < thr_left) {
1493 params->part_thread_target[id] += delta;
1496 params->part_thread_target[id] += thr_left;
1504 /* try to assign any leftover threads to partitions that
1505 * had volume lengths closer to needing thread_target+1 */
1506 int max_residue, max_id = 0;
1508 /* compute the residues */
1509 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1511 part_residue[id] = diskP->vol_list.len -
1512 (params->part_thread_target[id] * thr_workload);
1515 /* now try to allocate remaining threads to partitions with the
1516 * highest residues */
1519 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1521 if (part_residue[id] > max_residue) {
1522 max_residue = part_residue[id];
1531 params->part_thread_target[max_id]++;
1533 part_residue[max_id] = 0;
1538 /* punt and give any remaining threads equally to each partition */
1540 if (thr_left >= params->n_parts) {
1541 alloc = thr_left / params->n_parts;
1542 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1544 params->part_thread_target[id] += alloc;
1549 /* finish off the last of the threads */
1550 for (diskP = DiskPartitionList; thr_left && diskP; diskP = diskP->next) {
1552 params->part_thread_target[id]++;
1558 /* worker thread for parallel shutdown */
1560 VShutdownThread(void * args)
1562 vshutdown_thread_t * params;
1563 int found, pass, schedule_version_save, count;
1564 struct DiskPartition64 *diskP;
1565 struct diskpartition_queue_t * dpq;
1568 params = (vshutdown_thread_t *) args;
1570 /* acquire the shutdown pass 0 lock */
1571 MUTEX_ENTER(¶ms->lock);
1573 /* if there's still pass 0 work to be done,
1574 * get a work entry, and do a pass 0 shutdown */
1575 if (queue_IsNotEmpty(params)) {
1576 dpq = queue_First(params, diskpartition_queue_t);
1578 MUTEX_EXIT(¶ms->lock);
1584 while (ShutdownVolumeWalk_r(diskP, 0, ¶ms->part_pass_head[id]))
1586 params->stats[0][diskP->index] = count;
1587 MUTEX_ENTER(¶ms->lock);
1590 params->n_threads_complete++;
1591 if (params->n_threads_complete == params->n_threads) {
1592 /* notify control thread that all workers have completed pass 0 */
1593 CV_SIGNAL(¶ms->master_cv);
1595 while (params->pass == 0) {
1596 CV_WAIT(¶ms->cv, ¶ms->lock);
1600 MUTEX_EXIT(¶ms->lock);
1603 pass = params->pass;
1604 osi_Assert(pass > 0);
1606 /* now escalate through the more complicated shutdowns */
1608 schedule_version_save = params->schedule_version;
1610 /* find a disk partition to work on */
1611 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1613 if (params->part_thread_target[id] && !params->part_done_pass[id]) {
1614 params->part_thread_target[id]--;
1621 /* hmm. for some reason the controller thread couldn't find anything for
1622 * us to do. let's see if there's anything we can do */
1623 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1625 if (diskP->vol_list.len && !params->part_done_pass[id]) {
1628 } else if (!params->part_done_pass[id]) {
1629 params->part_done_pass[id] = 1;
1630 params->n_parts_done_pass++;
1632 Log("VShutdown: done shutting down volumes on partition %s.\n",
1633 VPartitionPath(diskP));
1639 /* do work on this partition until either the controller
1640 * creates a new schedule, or we run out of things to do
1641 * on this partition */
1644 while (!params->part_done_pass[id] &&
1645 (schedule_version_save == params->schedule_version)) {
1646 /* ShutdownVolumeWalk_r will drop the glock internally */
1647 if (!ShutdownVolumeWalk_r(diskP, pass, ¶ms->part_pass_head[id])) {
1648 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));
1661 params->stats[pass][id] += count;
1663 /* ok, everyone is done this pass, proceed */
1666 params->n_threads_complete++;
1667 while (params->pass == pass) {
1668 if (params->n_threads_complete == params->n_threads) {
1669 /* we are the last thread to complete, so we will
1670 * reinitialize worker pool state for the next pass */
1671 params->n_threads_complete = 0;
1672 params->n_parts_done_pass = 0;
1674 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1676 params->part_done_pass[id] = 0;
1677 params->part_pass_head[id] = queue_First(&diskP->vol_list, rx_queue);
1680 /* compute a new thread schedule before releasing all the workers */
1681 ShutdownCreateSchedule(params);
1683 /* wake up all the workers */
1684 CV_BROADCAST(¶ms->cv);
1687 Log("VShutdown: pass %d completed using %d threads on %d partitions\n",
1688 pass, params->n_threads, params->n_parts);
1691 VOL_CV_WAIT(¶ms->cv);
1694 pass = params->pass;
1708 /* shut down all volumes on a given disk partition
1710 * note that this function will not allow mp-fast
1711 * shutdown of a partition */
1713 VShutdownByPartition_r(struct DiskPartition64 * dp)
1719 /* wait for other exclusive ops to finish */
1720 VVByPListWait_r(dp);
1722 /* begin exclusive access */
1723 VVByPListBeginExclusive_r(dp);
1725 /* pick the low-hanging fruit first,
1726 * then do the complicated ones last
1727 * (has the advantage of keeping
1728 * in-use volumes up until the bitter end) */
1729 for (pass = 0, total=0; pass < 4; pass++) {
1730 pass_stats[pass] = ShutdownVByPForPass_r(dp, pass);
1731 total += pass_stats[pass];
1734 /* end exclusive access */
1735 VVByPListEndExclusive_r(dp);
1737 Log("VShutdownByPartition: shut down %d volumes on %s (pass[0]=%d, pass[1]=%d, pass[2]=%d, pass[3]=%d)\n",
1738 total, VPartitionPath(dp), pass_stats[0], pass_stats[1], pass_stats[2], pass_stats[3]);
1743 /* internal shutdown functionality
1745 * for multi-pass shutdown:
1746 * 0 to only "shutdown" {pre,un}attached and error state volumes
1747 * 1 to also shutdown attached volumes w/ volume header loaded
1748 * 2 to also shutdown attached volumes w/o volume header loaded
1749 * 3 to also shutdown exclusive state volumes
1751 * caller MUST hold exclusive access on the hash chain
1752 * because we drop vol_glock_mutex internally
1754 * this function is reentrant for passes 1--3
1755 * (e.g. multiple threads can cooperate to
1756 * shutdown a partition mp-fast)
1758 * pass 0 is not scaleable because the volume state data is
1759 * synchronized by vol_glock mutex, and the locking overhead
1760 * is too high to drop the lock long enough to do linked list
1764 ShutdownVByPForPass_r(struct DiskPartition64 * dp, int pass)
1766 struct rx_queue * q = queue_First(&dp->vol_list, rx_queue);
1769 while (ShutdownVolumeWalk_r(dp, pass, &q))
1775 /* conditionally shutdown one volume on partition dp
1776 * returns 1 if a volume was shutdown in this pass,
1779 ShutdownVolumeWalk_r(struct DiskPartition64 * dp, int pass,
1780 struct rx_queue ** idx)
1782 struct rx_queue *qp, *nqp;
1787 for (queue_ScanFrom(&dp->vol_list, qp, qp, nqp, rx_queue)) {
1788 vp = (Volume *) (((char *)qp) - offsetof(Volume, vol_list));
1792 if ((V_attachState(vp) != VOL_STATE_UNATTACHED) &&
1793 (V_attachState(vp) != VOL_STATE_ERROR) &&
1794 (V_attachState(vp) != VOL_STATE_DELETED) &&
1795 (V_attachState(vp) != VOL_STATE_PREATTACHED)) {
1799 if ((V_attachState(vp) == VOL_STATE_ATTACHED) &&
1800 (vp->header == NULL)) {
1804 if (VIsExclusiveState(V_attachState(vp))) {
1809 DeleteVolumeFromVByPList_r(vp);
1810 VShutdownVolume_r(vp);
1820 * shutdown a specific volume
1822 /* caller MUST NOT hold a heavyweight ref on vp */
1824 VShutdownVolume_r(Volume * vp)
1828 VCreateReservation_r(vp);
1830 if (LogLevel >= 5) {
1831 Log("VShutdownVolume_r: vid=%u, device=%d, state=%hu\n",
1832 vp->hashid, vp->partition->device, V_attachState(vp));
1835 /* wait for other blocking ops to finish */
1836 VWaitExclusiveState_r(vp);
1838 osi_Assert(VIsValidState(V_attachState(vp)));
1840 switch(V_attachState(vp)) {
1841 case VOL_STATE_SALVAGING:
1842 /* Leave salvaging volumes alone. Any in-progress salvages will
1843 * continue working after viced shuts down. This is intentional.
1846 case VOL_STATE_PREATTACHED:
1847 case VOL_STATE_ERROR:
1848 VChangeState_r(vp, VOL_STATE_UNATTACHED);
1849 case VOL_STATE_UNATTACHED:
1850 case VOL_STATE_DELETED:
1852 case VOL_STATE_GOING_OFFLINE:
1853 case VOL_STATE_SHUTTING_DOWN:
1854 case VOL_STATE_ATTACHED:
1858 Log("VShutdown: Attempting to take volume %u offline.\n",
1861 /* take the volume offline (drops reference count) */
1862 VOffline_r(vp, "File server was shut down");
1869 VCancelReservation_r(vp);
1873 #endif /* AFS_DEMAND_ATTACH_FS */
1876 /***************************************************/
1877 /* Header I/O routines */
1878 /***************************************************/
1880 /* open a descriptor for the inode (h),
1881 * read in an on-disk structure into buffer (to) of size (size),
1882 * verify versionstamp in structure has magic (magic) and
1883 * optionally verify version (version) if (version) is nonzero
1886 ReadHeader(Error * ec, IHandle_t * h, char *to, int size, bit32 magic,
1889 struct versionStamp *vsn;
1904 vsn = (struct versionStamp *)to;
1905 if (FDH_PREAD(fdP, to, size, 0) != size || vsn->magic != magic) {
1907 FDH_REALLYCLOSE(fdP);
1912 /* Check is conditional, in case caller wants to inspect version himself */
1913 if (version && vsn->version != version) {
1919 WriteVolumeHeader_r(Error * ec, Volume * vp)
1921 IHandle_t *h = V_diskDataHandle(vp);
1931 if (FDH_PWRITE(fdP, (char *)&V_disk(vp), sizeof(V_disk(vp)), 0)
1932 != sizeof(V_disk(vp))) {
1934 FDH_REALLYCLOSE(fdP);
1940 /* VolumeHeaderToDisk
1941 * Allows for storing 64 bit inode numbers in on-disk volume header
1944 /* convert in-memory representation of a volume header to the
1945 * on-disk representation of a volume header */
1947 VolumeHeaderToDisk(VolumeDiskHeader_t * dh, VolumeHeader_t * h)
1950 memset(dh, 0, sizeof(VolumeDiskHeader_t));
1951 dh->stamp = h->stamp;
1953 dh->parent = h->parent;
1955 #ifdef AFS_64BIT_IOPS_ENV
1956 dh->volumeInfo_lo = (afs_int32) h->volumeInfo & 0xffffffff;
1957 dh->volumeInfo_hi = (afs_int32) (h->volumeInfo >> 32) & 0xffffffff;
1958 dh->smallVnodeIndex_lo = (afs_int32) h->smallVnodeIndex & 0xffffffff;
1959 dh->smallVnodeIndex_hi =
1960 (afs_int32) (h->smallVnodeIndex >> 32) & 0xffffffff;
1961 dh->largeVnodeIndex_lo = (afs_int32) h->largeVnodeIndex & 0xffffffff;
1962 dh->largeVnodeIndex_hi =
1963 (afs_int32) (h->largeVnodeIndex >> 32) & 0xffffffff;
1964 dh->linkTable_lo = (afs_int32) h->linkTable & 0xffffffff;
1965 dh->linkTable_hi = (afs_int32) (h->linkTable >> 32) & 0xffffffff;
1967 dh->volumeInfo_lo = h->volumeInfo;
1968 dh->smallVnodeIndex_lo = h->smallVnodeIndex;
1969 dh->largeVnodeIndex_lo = h->largeVnodeIndex;
1970 dh->linkTable_lo = h->linkTable;
1974 /* DiskToVolumeHeader
1975 * Converts an on-disk representation of a volume header to
1976 * the in-memory representation of a volume header.
1978 * Makes the assumption that AFS has *always*
1979 * zero'd the volume header file so that high parts of inode
1980 * numbers are 0 in older (SGI EFS) volume header files.
1983 DiskToVolumeHeader(VolumeHeader_t * h, VolumeDiskHeader_t * dh)
1985 memset(h, 0, sizeof(VolumeHeader_t));
1986 h->stamp = dh->stamp;
1988 h->parent = dh->parent;
1990 #ifdef AFS_64BIT_IOPS_ENV
1992 (Inode) dh->volumeInfo_lo | ((Inode) dh->volumeInfo_hi << 32);
1994 h->smallVnodeIndex =
1995 (Inode) dh->smallVnodeIndex_lo | ((Inode) dh->
1996 smallVnodeIndex_hi << 32);
1998 h->largeVnodeIndex =
1999 (Inode) dh->largeVnodeIndex_lo | ((Inode) dh->
2000 largeVnodeIndex_hi << 32);
2002 (Inode) dh->linkTable_lo | ((Inode) dh->linkTable_hi << 32);
2004 h->volumeInfo = dh->volumeInfo_lo;
2005 h->smallVnodeIndex = dh->smallVnodeIndex_lo;
2006 h->largeVnodeIndex = dh->largeVnodeIndex_lo;
2007 h->linkTable = dh->linkTable_lo;
2012 /***************************************************/
2013 /* Volume Attachment routines */
2014 /***************************************************/
2016 #ifdef AFS_DEMAND_ATTACH_FS
2018 * pre-attach a volume given its path.
2020 * @param[out] ec outbound error code
2021 * @param[in] partition partition path string
2022 * @param[in] name volume id string
2024 * @return volume object pointer
2026 * @note A pre-attached volume will only have its partition
2027 * and hashid fields initialized. At first call to
2028 * VGetVolume, the volume will be fully attached.
2032 VPreAttachVolumeByName(Error * ec, char *partition, char *name)
2036 vp = VPreAttachVolumeByName_r(ec, partition, name);
2042 * pre-attach a volume given its path.
2044 * @param[out] ec outbound error code
2045 * @param[in] partition path to vice partition
2046 * @param[in] name volume id string
2048 * @return volume object pointer
2050 * @pre VOL_LOCK held
2052 * @internal volume package internal use only.
2055 VPreAttachVolumeByName_r(Error * ec, char *partition, char *name)
2057 return VPreAttachVolumeById_r(ec,
2059 VolumeNumber(name));
2063 * pre-attach a volume given its path and numeric volume id.
2065 * @param[out] ec error code return
2066 * @param[in] partition path to vice partition
2067 * @param[in] volumeId numeric volume id
2069 * @return volume object pointer
2071 * @pre VOL_LOCK held
2073 * @internal volume package internal use only.
2076 VPreAttachVolumeById_r(Error * ec,
2081 struct DiskPartition64 *partp;
2085 osi_Assert(programType == fileServer);
2087 if (!(partp = VGetPartition_r(partition, 0))) {
2089 Log("VPreAttachVolumeById_r: Error getting partition (%s)\n", partition);
2093 vp = VLookupVolume_r(ec, volumeId, NULL);
2098 return VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2102 * preattach a volume.
2104 * @param[out] ec outbound error code
2105 * @param[in] partp pointer to partition object
2106 * @param[in] vp pointer to volume object
2107 * @param[in] vid volume id
2109 * @return volume object pointer
2111 * @pre VOL_LOCK is held.
2113 * @warning Returned volume object pointer does not have to
2114 * equal the pointer passed in as argument vp. There
2115 * are potential race conditions which can result in
2116 * the pointers having different values. It is up to
2117 * the caller to make sure that references are handled
2118 * properly in this case.
2120 * @note If there is already a volume object registered with
2121 * the same volume id, its pointer MUST be passed as
2122 * argument vp. Failure to do so will result in a silent
2123 * failure to preattach.
2125 * @internal volume package internal use only.
2128 VPreAttachVolumeByVp_r(Error * ec,
2129 struct DiskPartition64 * partp,
2137 /* check to see if pre-attach already happened */
2139 (V_attachState(vp) != VOL_STATE_UNATTACHED) &&
2140 (V_attachState(vp) != VOL_STATE_DELETED) &&
2141 (V_attachState(vp) != VOL_STATE_PREATTACHED) &&
2142 !VIsErrorState(V_attachState(vp))) {
2144 * pre-attach is a no-op in all but the following cases:
2146 * - volume is unattached
2147 * - volume is in an error state
2148 * - volume is pre-attached
2150 Log("VPreattachVolumeByVp_r: volume %u not in quiescent state\n", vid);
2153 /* we're re-attaching a volume; clear out some old state */
2154 memset(&vp->salvage, 0, sizeof(struct VolumeOnlineSalvage));
2156 if (V_partition(vp) != partp) {
2157 /* XXX potential race */
2158 DeleteVolumeFromVByPList_r(vp);
2161 /* if we need to allocate a new Volume struct,
2162 * go ahead and drop the vol glock, otherwise
2163 * do the basic setup synchronised, as it's
2164 * probably not worth dropping the lock */
2167 /* allocate the volume structure */
2168 vp = nvp = (Volume *) malloc(sizeof(Volume));
2169 osi_Assert(vp != NULL);
2170 memset(vp, 0, sizeof(Volume));
2171 queue_Init(&vp->vnode_list);
2172 queue_Init(&vp->rx_call_list);
2173 CV_INIT(&V_attachCV(vp), "vp attach", CV_DEFAULT, 0);
2176 /* link the volume with its associated vice partition */
2177 vp->device = partp->device;
2178 vp->partition = partp;
2181 vp->specialStatus = 0;
2183 /* if we dropped the lock, reacquire the lock,
2184 * check for pre-attach races, and then add
2185 * the volume to the hash table */
2188 nvp = VLookupVolume_r(ec, vid, NULL);
2193 } else if (nvp) { /* race detected */
2198 /* hack to make up for VChangeState_r() decrementing
2199 * the old state counter */
2200 VStats.state_levels[0]++;
2204 /* put pre-attached volume onto the hash table
2205 * and bring it up to the pre-attached state */
2206 AddVolumeToHashTable(vp, vp->hashid);
2207 AddVolumeToVByPList_r(vp);
2208 VLRU_Init_Node_r(vp);
2209 VChangeState_r(vp, VOL_STATE_PREATTACHED);
2212 Log("VPreAttachVolumeByVp_r: volume %u pre-attached\n", vp->hashid);
2220 #endif /* AFS_DEMAND_ATTACH_FS */
2222 /* Attach an existing volume, given its pathname, and return a
2223 pointer to the volume header information. The volume also
2224 normally goes online at this time. An offline volume
2225 must be reattached to make it go online */
2227 VAttachVolumeByName(Error * ec, char *partition, char *name, int mode)
2231 retVal = VAttachVolumeByName_r(ec, partition, name, mode);
2237 VAttachVolumeByName_r(Error * ec, char *partition, char *name, int mode)
2240 struct DiskPartition64 *partp;
2245 #ifdef AFS_DEMAND_ATTACH_FS
2246 VolumeStats stats_save;
2248 #endif /* AFS_DEMAND_ATTACH_FS */
2252 volumeId = VolumeNumber(name);
2254 if (!(partp = VGetPartition_r(partition, 0))) {
2256 Log("VAttachVolume: Error getting partition (%s)\n", partition);
2260 if (VRequiresPartLock()) {
2261 osi_Assert(VInit == 3);
2262 VLockPartition_r(partition);
2263 } else if (programType == fileServer) {
2264 #ifdef AFS_DEMAND_ATTACH_FS
2265 /* lookup the volume in the hash table */
2266 vp = VLookupVolume_r(ec, volumeId, NULL);
2272 /* save any counters that are supposed to
2273 * be monotonically increasing over the
2274 * lifetime of the fileserver */
2275 memcpy(&stats_save, &vp->stats, sizeof(VolumeStats));
2277 memset(&stats_save, 0, sizeof(VolumeStats));
2280 /* if there's something in the hash table, and it's not
2281 * in the pre-attach state, then we may need to detach
2282 * it before proceeding */
2283 if (vp && (V_attachState(vp) != VOL_STATE_PREATTACHED)) {
2284 VCreateReservation_r(vp);
2285 VWaitExclusiveState_r(vp);
2287 /* at this point state must be one of:
2297 if (vp->specialStatus == VBUSY)
2300 /* if it's already attached, see if we can return it */
2301 if (V_attachState(vp) == VOL_STATE_ATTACHED) {
2302 VGetVolumeByVp_r(ec, vp);
2303 if (V_inUse(vp) == fileServer) {
2304 VCancelReservation_r(vp);
2308 /* otherwise, we need to detach, and attempt to re-attach */
2309 VDetachVolume_r(ec, vp);
2311 Log("VAttachVolume: Error detaching old volume instance (%s)\n", name);
2314 /* if it isn't fully attached, delete from the hash tables,
2315 and let the refcounter handle the rest */
2316 DeleteVolumeFromHashTable(vp);
2317 DeleteVolumeFromVByPList_r(vp);
2320 VCancelReservation_r(vp);
2324 /* pre-attach volume if it hasn't been done yet */
2326 (V_attachState(vp) == VOL_STATE_UNATTACHED) ||
2327 (V_attachState(vp) == VOL_STATE_DELETED) ||
2328 (V_attachState(vp) == VOL_STATE_ERROR)) {
2330 vp = VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2336 osi_Assert(vp != NULL);
2338 /* handle pre-attach races
2340 * multiple threads can race to pre-attach a volume,
2341 * but we can't let them race beyond that
2343 * our solution is to let the first thread to bring
2344 * the volume into an exclusive state win; the other
2345 * threads just wait until it finishes bringing the
2346 * volume online, and then they do a vgetvolumebyvp
2348 if (svp && (svp != vp)) {
2349 /* wait for other exclusive ops to finish */
2350 VCreateReservation_r(vp);
2351 VWaitExclusiveState_r(vp);
2353 /* get a heavyweight ref, kill the lightweight ref, and return */
2354 VGetVolumeByVp_r(ec, vp);
2355 VCancelReservation_r(vp);
2359 /* at this point, we are chosen as the thread to do
2360 * demand attachment for this volume. all other threads
2361 * doing a getvolume on vp->hashid will block until we finish */
2363 /* make sure any old header cache entries are invalidated
2364 * before proceeding */
2365 FreeVolumeHeader(vp);
2367 VChangeState_r(vp, VOL_STATE_ATTACHING);
2369 /* restore any saved counters */
2370 memcpy(&vp->stats, &stats_save, sizeof(VolumeStats));
2371 #else /* AFS_DEMAND_ATTACH_FS */
2372 vp = VGetVolume_r(ec, volumeId);
2374 if (V_inUse(vp) == fileServer)
2376 if (vp->specialStatus == VBUSY)
2378 VDetachVolume_r(ec, vp);
2380 Log("VAttachVolume: Error detaching volume (%s)\n", name);
2384 #endif /* AFS_DEMAND_ATTACH_FS */
2388 strcpy(path, VPartitionPath(partp));
2392 strcat(path, OS_DIRSEP);
2396 vp = (Volume *) calloc(1, sizeof(Volume));
2397 osi_Assert(vp != NULL);
2398 vp->hashid = volumeId;
2399 vp->device = partp->device;
2400 vp->partition = partp;
2401 queue_Init(&vp->vnode_list);
2402 queue_Init(&vp->rx_call_list);
2403 #ifdef AFS_DEMAND_ATTACH_FS
2404 CV_INIT(&V_attachCV(vp), "vp attach", CV_DEFAULT, 0);
2405 #endif /* AFS_DEMAND_ATTACH_FS */
2408 /* attach2 is entered without any locks, and returns
2409 * with vol_glock_mutex held */
2410 vp = attach2(ec, volumeId, path, partp, vp, isbusy, mode, &checkedOut);
2412 if (VCanUseFSSYNC() && vp) {
2413 #ifdef AFS_DEMAND_ATTACH_FS
2414 if ((mode == V_VOLUPD) || (VolumeWriteable(vp) && (mode == V_CLONE))) {
2415 /* mark volume header as in use so that volser crashes lead to a
2416 * salvage attempt */
2417 VUpdateVolume_r(ec, vp, 0);
2419 /* for dafs, we should tell the fileserver, except for V_PEEK
2420 * where we know it is not necessary */
2421 if (mode == V_PEEK) {
2422 vp->needsPutBack = 0;
2424 vp->needsPutBack = VOL_PUTBACK;
2426 #else /* !AFS_DEMAND_ATTACH_FS */
2427 /* duplicate computation in fssync.c about whether the server
2428 * takes the volume offline or not. If the volume isn't
2429 * offline, we must not return it when we detach the volume,
2430 * or the server will abort */
2431 if (mode == V_READONLY || mode == V_PEEK
2432 || (!VolumeWriteable(vp) && (mode == V_CLONE || mode == V_DUMP)))
2433 vp->needsPutBack = 0;
2435 vp->needsPutBack = VOL_PUTBACK;
2436 #endif /* !AFS_DEMAND_ATTACH_FS */
2438 #ifdef FSSYNC_BUILD_CLIENT
2439 /* Only give back the vol to the fileserver if we checked it out; attach2
2440 * will set checkedOut only if we successfully checked it out from the
2442 if (VCanUseFSSYNC() && vp == NULL && checkedOut) {
2444 #ifdef AFS_DEMAND_ATTACH_FS
2445 /* If we couldn't attach but we scheduled a salvage, we already
2446 * notified the fileserver; don't online it now */
2447 if (*ec != VSALVAGING)
2448 #endif /* AFS_DEMAND_ATTACH_FS */
2449 FSYNC_VolOp(volumeId, partition, FSYNC_VOL_ON, 0, NULL);
2452 if (programType == fileServer && vp) {
2453 #ifdef AFS_DEMAND_ATTACH_FS
2455 * we can get here in cases where we don't "own"
2456 * the volume (e.g. volume owned by a utility).
2457 * short circuit around potential disk header races.
2459 if (V_attachState(vp) != VOL_STATE_ATTACHED) {
2463 VUpdateVolume_r(ec, vp, 0);
2465 Log("VAttachVolume: Error updating volume\n");
2470 if (VolumeWriteable(vp) && V_dontSalvage(vp) == 0) {
2471 #ifndef AFS_DEMAND_ATTACH_FS
2472 /* This is a hack: by temporarily setting the incore
2473 * dontSalvage flag ON, the volume will be put back on the
2474 * Update list (with dontSalvage OFF again). It will then
2475 * come back in N minutes with DONT_SALVAGE eventually
2476 * set. This is the way that volumes that have never had
2477 * it set get it set; or that volumes that have been
2478 * offline without DONT SALVAGE having been set also
2479 * eventually get it set */
2480 V_dontSalvage(vp) = DONT_SALVAGE;
2481 #endif /* !AFS_DEMAND_ATTACH_FS */
2482 VAddToVolumeUpdateList_r(ec, vp);
2484 Log("VAttachVolume: Error adding volume to update list\n");
2491 Log("VOnline: volume %u (%s) attached and online\n", V_id(vp),
2496 if (VRequiresPartLock()) {
2497 VUnlockPartition_r(partition);
2500 #ifdef AFS_DEMAND_ATTACH_FS
2501 /* attach failed; make sure we're in error state */
2502 if (vp && !VIsErrorState(V_attachState(vp))) {
2503 VChangeState_r(vp, VOL_STATE_ERROR);
2505 #endif /* AFS_DEMAND_ATTACH_FS */
2512 #ifdef AFS_DEMAND_ATTACH_FS
2513 /* VAttachVolumeByVp_r
2515 * finish attaching a volume that is
2516 * in a less than fully attached state
2518 /* caller MUST hold a ref count on vp */
2520 VAttachVolumeByVp_r(Error * ec, Volume * vp, int mode)
2522 char name[VMAXPATHLEN];
2524 struct DiskPartition64 *partp;
2528 Volume * nvp = NULL;
2529 VolumeStats stats_save;
2533 /* volume utility should never call AttachByVp */
2534 osi_Assert(programType == fileServer);
2536 volumeId = vp->hashid;
2537 partp = vp->partition;
2538 VolumeExternalName_r(volumeId, name, sizeof(name));
2541 /* if another thread is performing a blocking op, wait */
2542 VWaitExclusiveState_r(vp);
2544 memcpy(&stats_save, &vp->stats, sizeof(VolumeStats));
2546 /* if it's already attached, see if we can return it */
2547 if (V_attachState(vp) == VOL_STATE_ATTACHED) {
2548 VGetVolumeByVp_r(ec, vp);
2549 if (V_inUse(vp) == fileServer) {
2552 if (vp->specialStatus == VBUSY)
2554 VDetachVolume_r(ec, vp);
2556 Log("VAttachVolume: Error detaching volume (%s)\n", name);
2562 /* pre-attach volume if it hasn't been done yet */
2564 (V_attachState(vp) == VOL_STATE_UNATTACHED) ||
2565 (V_attachState(vp) == VOL_STATE_DELETED) ||
2566 (V_attachState(vp) == VOL_STATE_ERROR)) {
2567 nvp = VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2573 VCreateReservation_r(nvp);
2578 osi_Assert(vp != NULL);
2579 VChangeState_r(vp, VOL_STATE_ATTACHING);
2581 /* restore monotonically increasing stats */
2582 memcpy(&vp->stats, &stats_save, sizeof(VolumeStats));
2586 /* compute path to disk header */
2587 strcpy(path, VPartitionPath(partp));
2591 strcat(path, OS_DIRSEP);
2596 * NOTE: attach2 is entered without any locks, and returns
2597 * with vol_glock_mutex held */
2598 vp = attach2(ec, volumeId, path, partp, vp, isbusy, mode, &checkedOut);
2601 * the event that an error was encountered, or
2602 * the volume was not brought to an attached state
2603 * for any reason, skip to the end. We cannot
2604 * safely call VUpdateVolume unless we "own" it.
2608 (V_attachState(vp) != VOL_STATE_ATTACHED)) {
2612 VUpdateVolume_r(ec, vp, 0);
2614 Log("VAttachVolume: Error updating volume %u\n", vp->hashid);
2618 if (VolumeWriteable(vp) && V_dontSalvage(vp) == 0) {
2619 #ifndef AFS_DEMAND_ATTACH_FS
2620 /* This is a hack: by temporarily setting the incore
2621 * dontSalvage flag ON, the volume will be put back on the
2622 * Update list (with dontSalvage OFF again). It will then
2623 * come back in N minutes with DONT_SALVAGE eventually
2624 * set. This is the way that volumes that have never had
2625 * it set get it set; or that volumes that have been
2626 * offline without DONT SALVAGE having been set also
2627 * eventually get it set */
2628 V_dontSalvage(vp) = DONT_SALVAGE;
2629 #endif /* !AFS_DEMAND_ATTACH_FS */
2630 VAddToVolumeUpdateList_r(ec, vp);
2632 Log("VAttachVolume: Error adding volume %u to update list\n", vp->hashid);
2639 Log("VOnline: volume %u (%s) attached and online\n", V_id(vp),
2643 VCancelReservation_r(nvp);
2646 if (*ec && (*ec != VOFFLINE) && (*ec != VSALVAGE)) {
2647 if (vp && !VIsErrorState(V_attachState(vp))) {
2648 VChangeState_r(vp, VOL_STATE_ERROR);
2657 * lock a volume on disk (non-blocking).
2659 * @param[in] vp The volume to lock
2660 * @param[in] locktype READ_LOCK or WRITE_LOCK
2662 * @return operation status
2663 * @retval 0 success, lock was obtained
2664 * @retval EBUSY a conflicting lock was held by another process
2665 * @retval EIO error acquiring lock
2667 * @pre If we're in the fileserver, vp is in an exclusive state
2669 * @pre vp is not already locked
2672 VLockVolumeNB(Volume *vp, int locktype)
2676 osi_Assert(programType != fileServer || VIsExclusiveState(V_attachState(vp)));
2677 osi_Assert(!(V_attachFlags(vp) & VOL_LOCKED));
2679 code = VLockVolumeByIdNB(vp->hashid, vp->partition, locktype);
2681 V_attachFlags(vp) |= VOL_LOCKED;
2688 * unlock a volume on disk that was locked with VLockVolumeNB.
2690 * @param[in] vp volume to unlock
2692 * @pre If we're in the fileserver, vp is in an exclusive state
2694 * @pre vp has already been locked
2697 VUnlockVolume(Volume *vp)
2699 osi_Assert(programType != fileServer || VIsExclusiveState(V_attachState(vp)));
2700 osi_Assert((V_attachFlags(vp) & VOL_LOCKED));
2702 VUnlockVolumeById(vp->hashid, vp->partition);
2704 V_attachFlags(vp) &= ~VOL_LOCKED;
2706 #endif /* AFS_DEMAND_ATTACH_FS */
2709 * read in a vol header, possibly lock the vol header, and possibly check out
2710 * the vol header from the fileserver, as part of volume attachment.
2712 * @param[out] ec error code
2713 * @param[in] vp volume pointer object
2714 * @param[in] partp disk partition object of the attaching partition
2715 * @param[in] mode attachment mode such as V_VOLUPD, V_DUMP, etc (see
2717 * @param[in] peek 1 to just try to read in the volume header and make sure
2718 * we don't try to lock the vol, or check it out from
2719 * FSSYNC or anything like that; 0 otherwise, for 'normal'
2721 * @param[out] acheckedOut If we successfully checked-out the volume from
2722 * the fileserver (if we needed to), this is set
2723 * to 1, otherwise it is untouched.
2725 * @note As part of DAFS volume attachment, the volume header may be either
2726 * read- or write-locked to ensure mutual exclusion of certain volume
2727 * operations. In some cases in order to determine whether we need to
2728 * read- or write-lock the header, we need to read in the header to see
2729 * if the volume is RW or not. So, if we read in the header under a
2730 * read-lock and determine that we actually need a write-lock on the
2731 * volume header, this function will drop the read lock, acquire a write
2732 * lock, and read the header in again.
2735 attach_volume_header(Error *ec, Volume *vp, struct DiskPartition64 *partp,
2736 int mode, int peek, int *acheckedOut)
2738 struct VolumeDiskHeader diskHeader;
2739 struct VolumeHeader header;
2742 int lock_tries = 0, checkout_tries = 0;
2744 VolumeId volid = vp->hashid;
2745 #ifdef FSSYNC_BUILD_CLIENT
2746 int checkout, done_checkout = 0;
2747 #endif /* FSSYNC_BUILD_CLIENT */
2748 #ifdef AFS_DEMAND_ATTACH_FS
2749 int locktype = 0, use_locktype = -1;
2750 #endif /* AFS_DEMAND_ATTACH_FS */
2756 if (lock_tries > VOL_MAX_CHECKOUT_RETRIES) {
2757 Log("VAttachVolume: retried too many times trying to lock header for "
2758 "vol %lu part %s; giving up\n", afs_printable_uint32_lu(volid),
2759 VPartitionPath(partp));
2763 if (checkout_tries > VOL_MAX_CHECKOUT_RETRIES) {
2764 Log("VAttachVolume: retried too many times trying to checkout "
2765 "vol %lu part %s; giving up\n", afs_printable_uint32_lu(volid),
2766 VPartitionPath(partp));
2771 if (VReadVolumeDiskHeader(volid, partp, NULL)) {
2772 /* short-circuit the 'volume does not exist' case */
2777 #ifdef FSSYNC_BUILD_CLIENT
2778 checkout = !done_checkout;
2780 if (!peek && checkout && VMustCheckoutVolume(mode)) {
2782 memset(&res, 0, sizeof(res));
2784 if (FSYNC_VolOp(volid, partp->name, FSYNC_VOL_NEEDVOLUME, mode, &res)
2787 if (res.hdr.reason == FSYNC_SALVAGE) {
2788 Log("VAttachVolume: file server says volume %lu is salvaging\n",
2789 afs_printable_uint32_lu(volid));
2792 Log("VAttachVolume: attach of volume %lu apparently denied by file server\n",
2793 afs_printable_uint32_lu(volid));
2794 *ec = VNOVOL; /* XXXX */
2802 #ifdef AFS_DEMAND_ATTACH_FS
2803 if (use_locktype < 0) {
2804 /* don't know whether vol is RO or RW; assume it's RO and we can retry
2805 * if it turns out to be RW */
2806 locktype = VVolLockType(mode, 0);
2809 /* a previous try says we should use use_locktype to lock the volume,
2811 locktype = use_locktype;
2814 if (!peek && locktype) {
2815 code = VLockVolumeNB(vp, locktype);
2817 if (code == EBUSY) {
2818 Log("VAttachVolume: another program has vol %lu locked\n",
2819 afs_printable_uint32_lu(volid));
2821 Log("VAttachVolume: error %d trying to lock vol %lu\n",
2822 code, afs_printable_uint32_lu(volid));
2829 #endif /* AFS_DEMAND_ATTACH_FS */
2831 code = VReadVolumeDiskHeader(volid, partp, &diskHeader);
2841 DiskToVolumeHeader(&header, &diskHeader);
2843 IH_INIT(vp->vnodeIndex[vLarge].handle, partp->device, header.parent,
2844 header.largeVnodeIndex);
2845 IH_INIT(vp->vnodeIndex[vSmall].handle, partp->device, header.parent,
2846 header.smallVnodeIndex);
2847 IH_INIT(vp->diskDataHandle, partp->device, header.parent,
2849 IH_INIT(vp->linkHandle, partp->device, header.parent, header.linkTable);
2852 /* only need to do this once */
2854 GetVolumeHeader(vp);
2858 #if defined(AFS_DEMAND_ATTACH_FS) && defined(FSSYNC_BUILD_CLIENT)
2859 /* demand attach changes the V_PEEK mechanism
2861 * we can now suck the current disk data structure over
2862 * the fssync interface without going to disk
2864 * (technically, we don't need to restrict this feature
2865 * to demand attach fileservers. However, I'm trying
2866 * to limit the number of common code changes)
2868 if (VCanUseFSSYNC() && (mode == V_PEEK || peek)) {
2870 res.payload.len = sizeof(VolumeDiskData);
2871 res.payload.buf = &vp->header->diskstuff;
2873 if (FSYNC_VolOp(vp->hashid,
2875 FSYNC_VOL_QUERY_HDR,
2878 goto disk_header_loaded;
2881 #endif /* AFS_DEMAND_ATTACH_FS && FSSYNC_BUILD_CLIENT */
2882 (void)ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
2883 sizeof(V_disk(vp)), VOLUMEINFOMAGIC, VOLUMEINFOVERSION);
2885 #ifdef AFS_DEMAND_ATTACH_FS
2888 IncUInt64(&VStats.hdr_loads);
2889 IncUInt64(&vp->stats.hdr_loads);
2891 #endif /* AFS_DEMAND_ATTACH_FS */
2894 Log("VAttachVolume: Error reading diskDataHandle header for vol %lu; "
2895 "error=%u\n", afs_printable_uint32_lu(volid), *ec);
2899 #ifdef AFS_DEMAND_ATTACH_FS
2900 # ifdef FSSYNC_BUILD_CLIENT
2902 # endif /* FSSYNC_BUILD_CLIENT */
2904 /* if the lock type we actually used to lock the volume is different than
2905 * the lock type we should have used, retry with the lock type we should
2907 use_locktype = VVolLockType(mode, VolumeWriteable(vp));
2908 if (locktype != use_locktype) {
2912 #endif /* AFS_DEMAND_ATTACH_FS */
2917 #if defined(AFS_DEMAND_ATTACH_FS) && defined(FSSYNC_BUILD_CLIENT)
2918 if (!peek && *ec == 0 && retry == 0 && VMustCheckoutVolume(mode)) {
2920 code = FSYNC_VerifyCheckout(volid, partp->name, FSYNC_VOL_NEEDVOLUME, mode);
2922 if (code == SYNC_DENIED) {
2923 /* must retry checkout; fileserver no longer thinks we have
2929 } else if (code != SYNC_OK) {
2933 #endif /* AFS_DEMAND_ATTACH_FS && FSSYNC_BUILD_CLIENT */
2936 /* either we are going to be called again for a second pass, or we
2937 * encountered an error; clean up in either case */
2939 #ifdef AFS_DEMAND_ATTACH_FS
2940 if ((V_attachFlags(vp) & VOL_LOCKED)) {
2943 #endif /* AFS_DEMAND_ATTACH_FS */
2944 if (vp->linkHandle) {
2945 IH_RELEASE(vp->vnodeIndex[vLarge].handle);
2946 IH_RELEASE(vp->vnodeIndex[vSmall].handle);
2947 IH_RELEASE(vp->diskDataHandle);
2948 IH_RELEASE(vp->linkHandle);
2961 #ifdef AFS_DEMAND_ATTACH_FS
2963 attach_check_vop(Error *ec, VolumeId volid, struct DiskPartition64 *partp,
2964 Volume *vp, int *acheckedOut)
2968 if (vp->pending_vol_op) {
2972 if (vp->pending_vol_op->vol_op_state == FSSYNC_VolOpRunningUnknown) {
2974 code = VVolOpLeaveOnlineNoHeader_r(vp, vp->pending_vol_op);
2976 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOnline;
2977 } else if (code == 0) {
2978 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOffline;
2981 /* we need the vol header to determine if the volume can be
2982 * left online for the vop, so... get the header */
2986 /* attach header with peek=1 to avoid checking out the volume
2987 * or locking it; we just want the header info, we're not
2988 * messing with the volume itself at all */
2989 attach_volume_header(ec, vp, partp, V_PEEK, 1, acheckedOut);
2996 if (VVolOpLeaveOnline_r(vp, vp->pending_vol_op)) {
2997 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOnline;
2999 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOffline;
3002 /* make sure we grab a new vol header and re-open stuff on
3003 * actual attachment; we can't keep the data we grabbed, since
3004 * it was not done under a lock and thus not safe */
3005 FreeVolumeHeader(vp);
3006 VReleaseVolumeHandles_r(vp);
3009 /* see if the pending volume op requires exclusive access */
3010 switch (vp->pending_vol_op->vol_op_state) {
3011 case FSSYNC_VolOpPending:
3012 /* this should never happen */
3013 osi_Assert(vp->pending_vol_op->vol_op_state != FSSYNC_VolOpPending);
3016 case FSSYNC_VolOpRunningUnknown:
3017 /* this should never happen; we resolved 'unknown' above */
3018 osi_Assert(vp->pending_vol_op->vol_op_state != FSSYNC_VolOpRunningUnknown);
3021 case FSSYNC_VolOpRunningOffline:
3022 /* mark the volume down */
3024 VChangeState_r(vp, VOL_STATE_UNATTACHED);
3026 /* do not set V_offlineMessage here; we don't have ownership of
3027 * the volume (and probably do not have the header loaded), so we
3028 * can't alter the disk header */
3030 /* check to see if we should set the specialStatus flag */
3031 if (VVolOpSetVBusy_r(vp, vp->pending_vol_op)) {
3032 /* don't overwrite specialStatus if it was already set to
3033 * something else (e.g. VMOVED) */
3034 if (!vp->specialStatus) {
3035 vp->specialStatus = VBUSY;
3047 #endif /* AFS_DEMAND_ATTACH_FS */
3050 * volume attachment helper function.
3052 * @param[out] ec error code
3053 * @param[in] volumeId volume ID of the attaching volume
3054 * @param[in] path full path to the volume header .vol file
3055 * @param[in] partp disk partition object for the attaching partition
3056 * @param[in] vp volume object; vp->hashid, vp->device, vp->partition,
3057 * vp->vnode_list, vp->rx_call_list, and V_attachCV (for
3058 * DAFS) should already be initialized
3059 * @param[in] isbusy 1 if vp->specialStatus should be set to VBUSY; that is,
3060 * if there is a volume operation running for this volume
3061 * that should set the volume to VBUSY during its run. 0
3062 * otherwise. (see VVolOpSetVBusy_r)
3063 * @param[in] mode attachment mode such as V_VOLUPD, V_DUMP, etc (see
3065 * @param[out] acheckedOut If we successfully checked-out the volume from
3066 * the fileserver (if we needed to), this is set
3067 * to 1, otherwise it is 0.
3069 * @return pointer to the semi-attached volume pointer
3070 * @retval NULL an error occurred (check value of *ec)
3071 * @retval vp volume successfully attaching
3073 * @pre no locks held
3075 * @post VOL_LOCK held
3078 attach2(Error * ec, VolId volumeId, char *path, struct DiskPartition64 *partp,
3079 Volume * vp, int isbusy, int mode, int *acheckedOut)
3081 /* have we read in the header successfully? */
3082 int read_header = 0;
3084 #ifdef AFS_DEMAND_ATTACH_FS
3085 /* should we FreeVolume(vp) instead of VCheckFree(vp) in the error
3089 /* in the case of an error, to what state should the volume be
3091 VolState error_state = VOL_STATE_ERROR;
3092 #endif /* AFS_DEMAND_ATTACH_FS */
3096 vp->vnodeIndex[vLarge].handle = NULL;
3097 vp->vnodeIndex[vSmall].handle = NULL;
3098 vp->diskDataHandle = NULL;
3099 vp->linkHandle = NULL;
3103 #ifdef AFS_DEMAND_ATTACH_FS
3104 attach_check_vop(ec, volumeId, partp, vp, acheckedOut);
3106 attach_volume_header(ec, vp, partp, mode, 0, acheckedOut);
3109 attach_volume_header(ec, vp, partp, mode, 0, acheckedOut);
3110 #endif /* !AFS_DEMAND_ATTACH_FS */
3112 if (*ec == VNOVOL) {
3113 /* if the volume doesn't exist, skip straight to 'error' so we don't
3114 * request a salvage */
3115 goto unlocked_error;
3121 /* ensure that we don't override specialStatus if it was set to
3122 * something else (e.g. VMOVED) */
3123 if (isbusy && !vp->specialStatus) {
3124 vp->specialStatus = VBUSY;
3126 vp->shuttingDown = 0;
3127 vp->goingOffline = 0;
3129 #ifdef AFS_DEMAND_ATTACH_FS
3130 vp->stats.last_attach = FT_ApproxTime();
3131 vp->stats.attaches++;
3135 IncUInt64(&VStats.attaches);
3136 vp->cacheCheck = ++VolumeCacheCheck;
3137 /* just in case this ever rolls over */
3138 if (!vp->cacheCheck)
3139 vp->cacheCheck = ++VolumeCacheCheck;
3142 #ifdef AFS_DEMAND_ATTACH_FS
3143 V_attachFlags(vp) |= VOL_HDR_LOADED;
3144 vp->stats.last_hdr_load = vp->stats.last_attach;
3145 #endif /* AFS_DEMAND_ATTACH_FS */
3149 struct IndexFileHeader iHead;
3151 #if OPENAFS_VOL_STATS
3153 * We just read in the diskstuff part of the header. If the detailed
3154 * volume stats area has not yet been initialized, we should bzero the
3155 * area and mark it as initialized.
3157 if (!(V_stat_initialized(vp))) {
3158 memset((V_stat_area(vp)), 0, VOL_STATS_BYTES);
3159 V_stat_initialized(vp) = 1;
3161 #endif /* OPENAFS_VOL_STATS */
3163 (void)ReadHeader(ec, vp->vnodeIndex[vSmall].handle,
3164 (char *)&iHead, sizeof(iHead),
3165 SMALLINDEXMAGIC, SMALLINDEXVERSION);
3168 Log("VAttachVolume: Error reading smallVnode vol header %s; error=%u\n", path, *ec);
3173 struct IndexFileHeader iHead;
3175 (void)ReadHeader(ec, vp->vnodeIndex[vLarge].handle,
3176 (char *)&iHead, sizeof(iHead),
3177 LARGEINDEXMAGIC, LARGEINDEXVERSION);
3180 Log("VAttachVolume: Error reading largeVnode vol header %s; error=%u\n", path, *ec);
3184 #ifdef AFS_NAMEI_ENV
3186 struct versionStamp stamp;
3188 (void)ReadHeader(ec, V_linkHandle(vp), (char *)&stamp,
3189 sizeof(stamp), LINKTABLEMAGIC, LINKTABLEVERSION);
3192 Log("VAttachVolume: Error reading namei vol header %s; error=%u\n", path, *ec);
3195 #endif /* AFS_NAMEI_ENV */
3197 #if defined(AFS_DEMAND_ATTACH_FS)
3198 if (*ec && ((*ec != VOFFLINE) || (V_attachState(vp) != VOL_STATE_UNATTACHED))) {
3200 if (!VCanScheduleSalvage()) {
3201 Log("VAttachVolume: Error attaching volume %s; volume needs salvage; error=%u\n", path, *ec);
3203 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_NO_OFFLINE);
3208 /* volume operation in progress */
3209 goto unlocked_error;
3211 #else /* AFS_DEMAND_ATTACH_FS */
3213 Log("VAttachVolume: Error attaching volume %s; volume needs salvage; error=%u\n", path, *ec);
3214 goto unlocked_error;
3216 #endif /* AFS_DEMAND_ATTACH_FS */
3218 if (V_needsSalvaged(vp)) {
3219 if (vp->specialStatus)
3220 vp->specialStatus = 0;
3222 #if defined(AFS_DEMAND_ATTACH_FS)
3223 if (!VCanScheduleSalvage()) {
3224 Log("VAttachVolume: volume salvage flag is ON for %s; volume needs salvage\n", path);
3226 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_NO_OFFLINE);
3229 #else /* AFS_DEMAND_ATTACH_FS */
3231 #endif /* AFS_DEMAND_ATTACH_FS */
3237 vp->nextVnodeUnique = V_uniquifier(vp);
3239 if (VShouldCheckInUse(mode) && V_inUse(vp) && VolumeWriteable(vp)) {
3240 if (!V_needsSalvaged(vp)) {
3241 V_needsSalvaged(vp) = 1;
3242 VUpdateVolume_r(ec, vp, 0);
3244 #if defined(AFS_DEMAND_ATTACH_FS)
3245 if (!VCanScheduleSalvage()) {
3246 Log("VAttachVolume: volume %s needs to be salvaged; not attached.\n", path);
3248 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_NO_OFFLINE);
3251 #else /* AFS_DEMAND_ATTACH_FS */
3252 Log("VAttachVolume: volume %s needs to be salvaged; not attached.\n", path);
3254 #endif /* AFS_DEMAND_ATTACH_FS */
3259 if (programType == fileServer && V_destroyMe(vp) == DESTROY_ME) {
3260 /* Only check destroyMe if we are the fileserver, since the
3261 * volserver et al sometimes need to work with volumes with
3262 * destroyMe set. Examples are 'temporary' volumes the
3263 * volserver creates, and when we create a volume (destroyMe
3264 * is set on creation; sometimes a separate volserver
3265 * transaction is created to clear destroyMe).
3268 #if defined(AFS_DEMAND_ATTACH_FS)
3269 /* schedule a salvage so the volume goes away on disk */
3270 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_NO_OFFLINE);
3271 VChangeState_r(vp, VOL_STATE_ERROR);
3274 #endif /* AFS_DEMAND_ATTACH_FS */
3275 Log("VAttachVolume: volume %s is junk; it should be destroyed at next salvage\n", path);
3280 vp->vnodeIndex[vSmall].bitmap = vp->vnodeIndex[vLarge].bitmap = NULL;
3281 #ifndef BITMAP_LATER
3282 if (programType == fileServer && VolumeWriteable(vp)) {
3284 for (i = 0; i < nVNODECLASSES; i++) {
3285 VGetBitmap_r(ec, vp, i);
3287 #ifdef AFS_DEMAND_ATTACH_FS
3288 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_NO_OFFLINE);
3290 #endif /* AFS_DEMAND_ATTACH_FS */
3291 Log("VAttachVolume: error getting bitmap for volume (%s)\n",
3297 #endif /* BITMAP_LATER */
3299 if (VInit >= 2 && V_needsCallback(vp)) {
3300 if (V_BreakVolumeCallbacks) {
3301 Log("VAttachVolume: Volume %lu was changed externally; breaking callbacks\n",
3302 afs_printable_uint32_lu(V_id(vp)));
3303 V_needsCallback(vp) = 0;
3305 (*V_BreakVolumeCallbacks) (V_id(vp));
3308 VUpdateVolume_r(ec, vp, 0);
3310 #ifdef FSSYNC_BUILD_CLIENT
3311 else if (VCanUseFSSYNC()) {
3312 afs_int32 fsync_code;
3314 V_needsCallback(vp) = 0;
3316 fsync_code = FSYNC_VolOp(V_id(vp), NULL, FSYNC_VOL_BREAKCBKS, FSYNC_WHATEVER, NULL);
3320 V_needsCallback(vp) = 1;
3321 Log("Error trying to tell the fileserver to break callbacks for "
3322 "changed volume %lu; error code %ld\n",
3323 afs_printable_uint32_lu(V_id(vp)),
3324 afs_printable_int32_ld(fsync_code));
3326 VUpdateVolume_r(ec, vp, 0);
3329 #endif /* FSSYNC_BUILD_CLIENT */
3332 Log("VAttachVolume: error %d clearing needsCallback on volume "
3333 "%lu; needs salvage\n", (int)*ec,
3334 afs_printable_uint32_lu(V_id(vp)));
3335 #ifdef AFS_DEMAND_ATTACH_FS
3336 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_NO_OFFLINE);
3338 #else /* !AFS_DEMAND_ATTACH_FS */
3340 #endif /* !AFS_DEMAND_ATTACh_FS */
3345 if (programType == fileServer) {
3346 if (vp->specialStatus)
3347 vp->specialStatus = 0;
3348 if (V_blessed(vp) && V_inService(vp) && !V_needsSalvaged(vp)) {
3349 V_inUse(vp) = fileServer;
3350 V_offlineMessage(vp)[0] = '\0';
3354 #ifdef AFS_DEMAND_ATTACH_FS
3355 /* Put the vol into PREATTACHED state, so if someone tries to
3356 * access it again, we try to attach, see that we're not blessed,
3357 * and give a VNOVOL error again. Putting it into UNATTACHED state
3358 * would result in a VOFFLINE error instead. */
3359 error_state = VOL_STATE_PREATTACHED;
3360 #endif /* AFS_DEMAND_ATTACH_FS */
3362 /* mimic e.g. GetVolume errors */
3363 if (!V_blessed(vp)) {
3364 Log("Volume %lu offline: not blessed\n", afs_printable_uint32_lu(V_id(vp)));
3365 FreeVolumeHeader(vp);
3366 } else if (!V_inService(vp)) {
3367 Log("Volume %lu offline: not in service\n", afs_printable_uint32_lu(V_id(vp)));
3368 FreeVolumeHeader(vp);
3370 Log("Volume %lu offline: needs salvage\n", afs_printable_uint32_lu(V_id(vp)));
3372 #ifdef AFS_DEMAND_ATTACH_FS
3373 error_state = VOL_STATE_ERROR;
3374 /* see if we can recover */
3375 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, 0 /*flags*/);
3378 #ifdef AFS_DEMAND_ATTACH_FS
3384 #ifdef AFS_DEMAND_ATTACH_FS
3385 if ((mode != V_PEEK) && (mode != V_SECRETLY))
3386 V_inUse(vp) = programType;
3387 #endif /* AFS_DEMAND_ATTACH_FS */
3388 V_checkoutMode(vp) = mode;
3391 AddVolumeToHashTable(vp, V_id(vp));
3392 #ifdef AFS_DEMAND_ATTACH_FS
3393 if (VCanUnlockAttached() && (V_attachFlags(vp) & VOL_LOCKED)) {
3396 if ((programType != fileServer) ||
3397 (V_inUse(vp) == fileServer)) {
3398 AddVolumeToVByPList_r(vp);
3400 VChangeState_r(vp, VOL_STATE_ATTACHED);
3402 VChangeState_r(vp, VOL_STATE_UNATTACHED);
3411 #ifdef AFS_DEMAND_ATTACH_FS
3412 if (!VIsErrorState(V_attachState(vp))) {
3413 VChangeState_r(vp, error_state);
3415 #endif /* AFS_DEMAND_ATTACH_FS */
3418 VReleaseVolumeHandles_r(vp);
3421 #ifdef AFS_DEMAND_ATTACH_FS
3428 #else /* !AFS_DEMAND_ATTACH_FS */
3430 #endif /* !AFS_DEMAND_ATTACH_FS */
3434 /* Attach an existing volume.
3435 The volume also normally goes online at this time.
3436 An offline volume must be reattached to make it go online.
3440 VAttachVolume(Error * ec, VolumeId volumeId, int mode)
3444 retVal = VAttachVolume_r(ec, volumeId, mode);
3450 VAttachVolume_r(Error * ec, VolumeId volumeId, int mode)
3453 VGetVolumePath(ec, volumeId, &part, &name);
3457 vp = VGetVolume_r(&error, volumeId);
3459 osi_Assert(V_inUse(vp) == 0);
3460 VDetachVolume_r(ec, vp);
3464 return VAttachVolumeByName_r(ec, part, name, mode);
3467 /* Increment a reference count to a volume, sans context swaps. Requires
3468 * possibly reading the volume header in from the disk, since there's
3469 * an invariant in the volume package that nUsers>0 ==> vp->header is valid.
3471 * N.B. This call can fail if we can't read in the header!! In this case
3472 * we still guarantee we won't context swap, but the ref count won't be
3473 * incremented (otherwise we'd violate the invariant).
3475 /* NOTE: with the demand attach fileserver extensions, the global lock
3476 * is dropped within VHold */
3477 #ifdef AFS_DEMAND_ATTACH_FS
3479 VHold_r(Volume * vp)
3483 VCreateReservation_r(vp);
3484 VWaitExclusiveState_r(vp);
3486 LoadVolumeHeader(&error, vp);
3488 VCancelReservation_r(vp);
3492 VCancelReservation_r(vp);
3495 #else /* AFS_DEMAND_ATTACH_FS */
3497 VHold_r(Volume * vp)
3501 LoadVolumeHeader(&error, vp);
3507 #endif /* AFS_DEMAND_ATTACH_FS */
3509 /**** volume timeout-related stuff ****/
3511 #ifdef AFS_PTHREAD_ENV
3513 static struct timespec *shutdown_timeout;
3514 static pthread_once_t shutdown_timeout_once = PTHREAD_ONCE_INIT;
3517 VTimedOut(const struct timespec *ts)
3522 if (ts->tv_sec == 0) {
3523 /* short-circuit; this will have always timed out */
3527 code = gettimeofday(&tv, NULL);
3529 Log("Error %d from gettimeofday, assuming we have not timed out\n", errno);
3530 /* assume no timeout; failure mode is we just wait longer than normal
3531 * instead of returning errors when we shouldn't */
3535 if (tv.tv_sec < ts->tv_sec ||
3536 (tv.tv_sec == ts->tv_sec && tv.tv_usec*1000 < ts->tv_nsec)) {
3545 * Calculate an absolute timeout.
3547 * @param[out] ts A timeout that is "timeout" seconds from now, if we return
3548 * NULL, the memory is not touched
3549 * @param[in] timeout How long the timeout should be from now
3551 * @return timeout to use
3552 * @retval NULL no timeout; wait forever
3553 * @retval non-NULL the given value for "ts"
3557 static struct timespec *
3558 VCalcTimeout(struct timespec *ts, afs_int32 timeout)
3568 ts->tv_sec = ts->tv_nsec = 0;
3572 code = gettimeofday(&now, NULL);
3574 Log("Error %d from gettimeofday, falling back to 'forever' timeout\n", errno);
3578 ts->tv_sec = now.tv_sec + timeout;
3579 ts->tv_nsec = now.tv_usec * 1000;
3585 * Initialize the shutdown_timeout global.
3588 VShutdownTimeoutInit(void)
3590 struct timespec *ts;
3592 ts = malloc(sizeof(*ts));
3594 shutdown_timeout = VCalcTimeout(ts, vol_opts.offline_shutdown_timeout);
3596 if (!shutdown_timeout) {
3602 * Figure out the timeout that should be used for waiting for offline volumes.
3604 * @param[out] ats Storage space for a local timeout value if needed
3606 * @return The timeout value that should be used
3607 * @retval NULL No timeout; wait forever for offlining volumes
3608 * @retval non-NULL A pointer to the absolute time that should be used as
3609 * the deadline for waiting for offlining volumes.
3611 * @note If we return non-NULL, the pointer we return may or may not be the
3614 static const struct timespec *
3615 VOfflineTimeout(struct timespec *ats)
3617 if (vol_shutting_down) {
3618 osi_Assert(pthread_once(&shutdown_timeout_once, VShutdownTimeoutInit) == 0);
3619 return shutdown_timeout;
3621 return VCalcTimeout(ats, vol_opts.offline_timeout);
3625 #else /* AFS_PTHREAD_ENV */
3627 /* Waiting a certain amount of time for offlining volumes is not supported
3628 * for LWP due to a lack of primitives. So, we never time out */
3629 # define VTimedOut(x) (0)
3630 # define VOfflineTimeout(x) (NULL)
3632 #endif /* !AFS_PTHREAD_ENV */
3640 retVal = VHold_r(vp);
3647 VIsGoingOffline_r(struct Volume *vp)
3651 if (vp->goingOffline) {
3652 if (vp->specialStatus) {
3653 code = vp->specialStatus;
3654 } else if (V_inService(vp) == 0 || V_blessed(vp) == 0) {
3665 * Tell the caller if a volume is waiting to go offline.
3667 * @param[in] vp The volume we want to know about
3669 * @return volume status
3670 * @retval 0 volume is not waiting to go offline, go ahead and use it
3671 * @retval nonzero volume is waiting to offline, and give the returned code
3672 * as an error to anyone accessing the volume
3674 * @pre VOL_LOCK is NOT held
3675 * @pre caller holds a heavyweight reference on vp
3678 VIsGoingOffline(struct Volume *vp)
3683 code = VIsGoingOffline_r(vp);
3690 * Register an RX call with a volume.
3692 * @param[inout] ec Error code; if unset when passed in, may be set if
3693 * the volume starts going offline
3694 * @param[out] client_ec @see GetVolume
3695 * @param[in] vp Volume struct
3696 * @param[in] cbv VCallByVol struct containing the RX call to register
3698 * @pre VOL_LOCK held
3699 * @pre caller holds heavy ref on vp
3704 VRegisterCall_r(Error *ec, Error *client_ec, Volume *vp, struct VCallByVol *cbv)
3707 #ifdef AFS_DEMAND_ATTACH_FS
3709 /* just in case the volume started going offline after we got the
3710 * reference to it... otherwise, if the volume started going
3711 * offline right at the end of GetVolume(), we might race with the
3712 * RX call scanner, and return success and add our cbv to the
3713 * rx_call_list _after_ the scanner has scanned the list. */
3714 *ec = VIsGoingOffline_r(vp);
3720 while (V_attachState(vp) == VOL_STATE_SCANNING_RXCALLS) {
3721 VWaitStateChange_r(vp);
3723 #endif /* AFS_DEMAND_ATTACH_FS */
3725 queue_Prepend(&vp->rx_call_list, cbv);
3730 * Deregister an RX call with a volume.
3732 * @param[in] vp Volume struct
3733 * @param[in] cbv VCallByVol struct containing the RX call to deregister
3735 * @pre VOL_LOCK held
3736 * @pre caller holds heavy ref on vp
3741 VDeregisterCall_r(Volume *vp, struct VCallByVol *cbv)
3743 if (cbv && queue_IsOnQueue(cbv)) {
3744 #ifdef AFS_DEMAND_ATTACH_FS
3745 while (V_attachState(vp) == VOL_STATE_SCANNING_RXCALLS) {
3746 VWaitStateChange_r(vp);
3748 #endif /* AFS_DEMAND_ATTACH_FS */
3754 /***************************************************/
3755 /* get and put volume routines */
3756 /***************************************************/
3759 * put back a heavyweight reference to a volume object.
3761 * @param[in] vp volume object pointer
3763 * @pre VOL_LOCK held
3765 * @post heavyweight volume reference put back.
3766 * depending on state, volume may have been taken offline,
3767 * detached, salvaged, freed, etc.
3769 * @internal volume package internal use only
3772 VPutVolume_r(Volume * vp)
3774 osi_Assert(--vp->nUsers >= 0);
3775 if (vp->nUsers == 0) {
3777 ReleaseVolumeHeader(vp->header);
3778 #ifdef AFS_DEMAND_ATTACH_FS
3779 if (!VCheckDetach(vp)) {
3783 #else /* AFS_DEMAND_ATTACH_FS */
3785 #endif /* AFS_DEMAND_ATTACH_FS */
3790 VPutVolume(Volume * vp)
3798 * Puts a volume reference obtained with VGetVolumeWithCall.
3800 * @param[in] vp Volume struct
3801 * @param[in] cbv VCallByVol struct given to VGetVolumeWithCall, or NULL if none
3803 * @pre VOL_LOCK is NOT held
3806 VPutVolumeWithCall(Volume *vp, struct VCallByVol *cbv)
3809 VDeregisterCall_r(vp, cbv);
3814 /* Get a pointer to an attached volume. The pointer is returned regardless
3815 of whether or not the volume is in service or on/off line. An error
3816 code, however, is returned with an indication of the volume's status */
3818 VGetVolume(Error * ec, Error * client_ec, VolId volumeId)
3822 retVal = GetVolume(ec, client_ec, volumeId, NULL, 0);
3828 * Get a volume reference associated with an RX call.
3830 * @param[out] ec @see GetVolume
3831 * @param[out] client_ec @see GetVolume
3832 * @param[in] volumeId @see GetVolume
3833 * @param[in] ts How long to wait for going-offline volumes (absolute time).
3834 * If NULL, wait forever. If ts->tv_sec == 0, return immediately
3835 * with an error if the volume is going offline.
3836 * @param[in] cbv Contains an RX call to be associated with this volume
3837 * reference. This call may be interrupted if the volume is
3838 * requested to go offline while we hold a ref on it. Give NULL
3839 * to not associate an RX call with this reference.
3841 * @return @see GetVolume
3843 * @note for LWP builds, ts must be NULL
3845 * @note A reference obtained with this function MUST be put back with
3846 * VPutVolumeWithCall
3849 VGetVolumeWithCall(Error * ec, Error * client_ec, VolId volumeId,
3850 const struct timespec *ts, struct VCallByVol *cbv)
3854 retVal = GetVolume(ec, client_ec, volumeId, NULL, ts);
3855 VRegisterCall_r(ec, client_ec, retVal, cbv);
3861 VGetVolume_r(Error * ec, VolId volumeId)
3863 return GetVolume(ec, NULL, volumeId, NULL, NULL);
3866 /* try to get a volume we've previously looked up */
3867 /* for demand attach fs, caller MUST NOT hold a ref count on vp */
3869 VGetVolumeByVp_r(Error * ec, Volume * vp)
3871 return GetVolume(ec, NULL, vp->hashid, vp, NULL);
3875 * private interface for getting a volume handle
3877 * @param[out] ec error code (0 if no error)
3878 * @param[out] client_ec wire error code to be given to clients
3879 * @param[in] volumeId ID of the volume we want
3880 * @param[in] hint optional hint for hash lookups, or NULL
3881 * @param[in] timeout absolute deadline for waiting for the volume to go
3882 * offline, if it is going offline. NULL to wait forever.
3884 * @return a volume handle for the specified volume
3885 * @retval NULL an error occurred, or the volume is in such a state that
3886 * we cannot load a header or return any volume struct
3888 * @note for DAFS, caller must NOT hold a ref count on 'hint'
3890 * @note 'timeout' is only checked if the volume is actually going offline; so
3891 * if you pass timeout->tv_sec = 0, this will exhibit typical
3892 * nonblocking behavior.
3894 * @note for LWP builds, 'timeout' must be NULL
3897 GetVolume(Error * ec, Error * client_ec, VolId volumeId, Volume * hint,
3898 const struct timespec *timeout)
3901 /* pull this profiling/debugging code out of regular builds */
3903 #define VGET_CTR_INC(x) x++
3904 unsigned short V0 = 0, V1 = 0, V2 = 0, V3 = 0, V5 = 0, V6 =
3905 0, V7 = 0, V8 = 0, V9 = 0;
3906 unsigned short V10 = 0, V11 = 0, V12 = 0, V13 = 0, V14 = 0, V15 = 0;
3908 #define VGET_CTR_INC(x)
3910 #ifdef AFS_DEMAND_ATTACH_FS
3911 Volume *avp, * rvp = hint;
3915 * if VInit is zero, the volume package dynamic
3916 * data structures have not been initialized yet,
3917 * and we must immediately return an error
3923 *client_ec = VOFFLINE;
3928 #ifdef AFS_DEMAND_ATTACH_FS
3930 VCreateReservation_r(rvp);
3932 #endif /* AFS_DEMAND_ATTACH_FS */
3940 vp = VLookupVolume_r(ec, volumeId, vp);
3946 #ifdef AFS_DEMAND_ATTACH_FS
3947 if (rvp && (rvp != vp)) {
3948 /* break reservation on old vp */
3949 VCancelReservation_r(rvp);
3952 #endif /* AFS_DEMAND_ATTACH_FS */
3958 /* Until we have reached an initialization level of 2
3959 * we don't know whether this volume exists or not.
3960 * We can't sleep and retry later because before a volume
3961 * is attached, the caller tries to get it first. Just
3962 * return VOFFLINE and the caller can choose whether to
3963 * retry the command or not. */
3973 IncUInt64(&VStats.hdr_gets);
3975 #ifdef AFS_DEMAND_ATTACH_FS
3976 /* block if someone else is performing an exclusive op on this volume */
3979 VCreateReservation_r(rvp);
3981 VWaitExclusiveState_r(vp);
3983 /* short circuit with VNOVOL in the following circumstances:
3986 * - VOL_STATE_SHUTTING_DOWN
3988 if ((V_attachState(vp) == VOL_STATE_ERROR) ||
3989 (V_attachState(vp) == VOL_STATE_SHUTTING_DOWN) ||
3990 (V_attachState(vp) == VOL_STATE_GOING_OFFLINE)) {
3997 * short circuit with VOFFLINE for VOL_STATE_UNATTACHED and
3998 * VNOVOL for VOL_STATE_DELETED
4000 if ((V_attachState(vp) == VOL_STATE_UNATTACHED) ||
4001 (V_attachState(vp) == VOL_STATE_DELETED)) {
4002 if (vp->specialStatus) {
4003 *ec = vp->specialStatus;
4004 } else if (V_attachState(vp) == VOL_STATE_DELETED) {
4013 /* allowable states:
4020 if (vp->salvage.requested) {
4021 VUpdateSalvagePriority_r(vp);
4024 if (V_attachState(vp) == VOL_STATE_PREATTACHED) {
4025 if (vp->specialStatus) {
4026 *ec = vp->specialStatus;
4030 avp = VAttachVolumeByVp_r(ec, vp, 0);
4033 /* VAttachVolumeByVp_r can return a pointer
4034 * != the vp passed to it under certain
4035 * conditions; make sure we don't leak
4036 * reservations if that happens */
4038 VCancelReservation_r(rvp);
4040 VCreateReservation_r(rvp);
4050 if (!vp->pending_vol_op) {
4065 if (VIsSalvaging(vp) || (*ec == VSALVAGING)) {
4067 /* see CheckVnode() in afsfileprocs.c for an explanation
4068 * of this error code logic */
4069 afs_uint32 now = FT_ApproxTime();
4070 if ((vp->stats.last_salvage + (10 * 60)) >= now) {
4073 *client_ec = VRESTARTING;
4082 #ifdef AFS_DEMAND_ATTACH_FS
4084 * this test MUST happen after VAttachVolymeByVp, so vol_op_state is
4085 * not VolOpRunningUnknown (attach2 would have converted it to Online
4089 /* only valid before/during demand attachment */
4090 osi_Assert(!vp->pending_vol_op || vp->pending_vol_op->vol_op_state != FSSYNC_VolOpRunningUnknown);
4092 /* deny getvolume due to running mutually exclusive vol op */
4093 if (vp->pending_vol_op && vp->pending_vol_op->vol_op_state==FSSYNC_VolOpRunningOffline) {
4095 * volume cannot remain online during this volume operation.
4098 if (vp->specialStatus) {
4100 * special status codes outrank normal VOFFLINE code
4102 *ec = vp->specialStatus;
4104 *client_ec = vp->specialStatus;
4108 /* see CheckVnode() in afsfileprocs.c for an explanation
4109 * of this error code logic */
4110 afs_uint32 now = FT_ApproxTime();
4111 if ((vp->stats.last_vol_op + (10 * 60)) >= now) {
4114 *client_ec = VRESTARTING;
4119 VChangeState_r(vp, VOL_STATE_UNATTACHED);
4120 FreeVolumeHeader(vp);
4124 #endif /* AFS_DEMAND_ATTACH_FS */
4126 LoadVolumeHeader(ec, vp);
4129 /* Only log the error if it was a totally unexpected error. Simply
4130 * a missing inode is likely to be caused by the volume being deleted */
4131 if (errno != ENXIO || LogLevel)
4132 Log("Volume %u: couldn't reread volume header\n",
4134 #ifdef AFS_DEMAND_ATTACH_FS
4135 if (VCanScheduleSalvage()) {
4136 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, 0 /*flags*/);
4141 #else /* AFS_DEMAND_ATTACH_FS */
4144 #endif /* AFS_DEMAND_ATTACH_FS */
4149 if (vp->shuttingDown) {
4156 if (programType == fileServer) {
4158 if (vp->goingOffline) {
4159 if (timeout && VTimedOut(timeout)) {
4160 /* we've timed out; don't wait for the vol */
4163 #ifdef AFS_DEMAND_ATTACH_FS
4164 /* wait for the volume to go offline */
4165 if (V_attachState(vp) == VOL_STATE_GOING_OFFLINE) {
4166 VTimedWaitStateChange_r(vp, timeout, NULL);
4168 #elif defined(AFS_PTHREAD_ENV)
4169 VOL_CV_TIMEDWAIT(&vol_put_volume_cond, timeout, NULL);
4170 #else /* AFS_PTHREAD_ENV */
4171 /* LWP has no timed wait, so the caller better not be
4173 osi_Assert(!timeout);
4174 LWP_WaitProcess(VPutVolume);
4175 #endif /* AFS_PTHREAD_ENV */
4179 if (vp->specialStatus) {
4181 *ec = vp->specialStatus;
4182 } else if (V_inService(vp) == 0 || V_blessed(vp) == 0) {
4185 } else if (V_inUse(vp) == 0 || vp->goingOffline) {
4196 #ifdef AFS_DEMAND_ATTACH_FS
4197 /* if no error, bump nUsers */
4200 VLRU_UpdateAccess_r(vp);
4203 VCancelReservation_r(rvp);
4206 if (client_ec && !*client_ec) {
4209 #else /* AFS_DEMAND_ATTACH_FS */
4210 /* if no error, bump nUsers */
4217 #endif /* AFS_DEMAND_ATTACH_FS */
4220 osi_Assert(vp || *ec);
4225 /***************************************************/
4226 /* Volume offline/detach routines */
4227 /***************************************************/
4229 /* caller MUST hold a heavyweight ref on vp */
4230 #ifdef AFS_DEMAND_ATTACH_FS
4232 VTakeOffline_r(Volume * vp)
4236 osi_Assert(vp->nUsers > 0);
4237 osi_Assert(programType == fileServer);
4239 VCreateReservation_r(vp);
4240 VWaitExclusiveState_r(vp);
4242 vp->goingOffline = 1;
4243 V_needsSalvaged(vp) = 1;
4245 VRequestSalvage_r(&error, vp, SALVSYNC_ERROR, 0);
4246 VCancelReservation_r(vp);
4248 #else /* AFS_DEMAND_ATTACH_FS */
4250 VTakeOffline_r(Volume * vp)
4252 osi_Assert(vp->nUsers > 0);
4253 osi_Assert(programType == fileServer);
4255 vp->goingOffline = 1;
4256 V_needsSalvaged(vp) = 1;
4258 #endif /* AFS_DEMAND_ATTACH_FS */
4261 VTakeOffline(Volume * vp)
4269 * force a volume offline.
4271 * @param[in] vp volume object pointer
4272 * @param[in] flags flags (see note below)
4274 * @note the flag VOL_FORCEOFF_NOUPDATE is a recursion control flag
4275 * used when VUpdateVolume_r needs to call VForceOffline_r
4276 * (which in turn would normally call VUpdateVolume_r)
4278 * @see VUpdateVolume_r
4280 * @pre VOL_LOCK must be held.
4281 * for DAFS, caller must hold ref.
4283 * @note for DAFS, it _is safe_ to call this function from an
4286 * @post needsSalvaged flag is set.
4287 * for DAFS, salvage is requested.
4288 * no further references to the volume through the volume
4289 * package will be honored.
4290 * all file descriptor and vnode caches are invalidated.
4292 * @warning this is a heavy-handed interface. it results in
4293 * a volume going offline regardless of the current
4294 * reference count state.
4296 * @internal volume package internal use only
4299 VForceOffline_r(Volume * vp, int flags)
4303 #ifdef AFS_DEMAND_ATTACH_FS
4304 VChangeState_r(vp, VOL_STATE_ERROR);
4309 strcpy(V_offlineMessage(vp),
4310 "Forced offline due to internal error: volume needs to be salvaged");
4311 Log("Volume %u forced offline: it needs salvaging!\n", V_id(vp));
4314 vp->goingOffline = 0;
4315 V_needsSalvaged(vp) = 1;
4316 if (!(flags & VOL_FORCEOFF_NOUPDATE)) {
4317 VUpdateVolume_r(&error, vp, VOL_UPDATE_NOFORCEOFF);
4320 #ifdef AFS_DEMAND_ATTACH_FS
4321 VRequestSalvage_r(&error, vp, SALVSYNC_ERROR, 0 /*flags*/);
4322 #endif /* AFS_DEMAND_ATTACH_FS */
4324 #ifdef AFS_PTHREAD_ENV
4325 CV_BROADCAST(&vol_put_volume_cond);
4326 #else /* AFS_PTHREAD_ENV */
4327 LWP_NoYieldSignal(VPutVolume);
4328 #endif /* AFS_PTHREAD_ENV */
4330 VReleaseVolumeHandles_r(vp);
4334 * force a volume offline.
4336 * @param[in] vp volume object pointer
4338 * @see VForceOffline_r
4341 VForceOffline(Volume * vp)
4344 VForceOffline_r(vp, 0);
4349 * Iterate over the RX calls associated with a volume, and interrupt them.
4351 * @param[in] vp The volume whose RX calls we want to scan
4353 * @pre VOL_LOCK held
4356 VScanCalls_r(struct Volume *vp)
4358 struct VCallByVol *cbv, *ncbv;
4360 #ifdef AFS_DEMAND_ATTACH_FS
4361 VolState state_save;
4364 if (queue_IsEmpty(&vp->rx_call_list))
4365 return; /* no calls to interrupt */
4366 if (!vol_opts.interrupt_rxcall)
4367 return; /* we have no function with which to interrupt calls */
4368 err = VIsGoingOffline_r(vp);
4370 return; /* we're not going offline anymore */
4372 #ifdef AFS_DEMAND_ATTACH_FS
4373 VWaitExclusiveState_r(vp);
4374 state_save = VChangeState_r(vp, VOL_STATE_SCANNING_RXCALLS);
4376 #endif /* AFS_DEMAND_ATTACH_FS */
4378 for(queue_Scan(&vp->rx_call_list, cbv, ncbv, VCallByVol)) {
4380 struct rx_peer *peer;
4382 peer = rx_PeerOf(rx_ConnectionOf(cbv->call));
4384 Log("Offlining volume %lu while client %s:%u is trying to read "
4385 "from it; kicking client off with error %ld\n",
4386 (long unsigned) vp->hashid,
4387 afs_inet_ntoa_r(rx_HostOf(peer), hoststr),
4388 (unsigned) ntohs(rx_PortOf(peer)),
4391 (*vol_opts.interrupt_rxcall) (cbv->call, err);
4394 #ifdef AFS_DEMAND_ATTACH_FS
4396 VChangeState_r(vp, state_save);
4397 #endif /* AFS_DEMAND_ATTACH_FS */
4400 #ifdef AFS_DEMAND_ATTACH_FS
4402 * Wait for a vp to go offline.
4404 * @param[out] ec 1 if a salvage on the volume has been requested and
4405 * salvok == 0, 0 otherwise
4406 * @param[in] vp The volume to wait for
4407 * @param[in] salvok If 0, we return immediately with *ec = 1 if the volume
4408 * has been requested to salvage. Otherwise we keep waiting
4409 * until the volume has gone offline.
4411 * @pre VOL_LOCK held
4412 * @pre caller holds a lightweight ref on vp
4417 VWaitForOfflineByVp_r(Error *ec, struct Volume *vp, int salvok)
4419 struct timespec timeout_ts;
4420 const struct timespec *ts;
4423 ts = VOfflineTimeout(&timeout_ts);
4427 while (!VIsOfflineState(V_attachState(vp)) && !timedout) {
4428 if (!salvok && vp->salvage.requested) {
4432 VTimedWaitStateChange_r(vp, ts, &timedout);
4435 /* we didn't time out, so the volume must be offline, so we're done */
4439 /* If we got here, we timed out waiting for the volume to go offline.
4440 * Kick off the accessing RX calls and wait again */
4444 while (!VIsOfflineState(V_attachState(vp))) {
4445 if (!salvok && vp->salvage.requested) {
4450 VWaitStateChange_r(vp);
4454 #else /* AFS_DEMAND_ATTACH_FS */
4457 * Wait for a volume to go offline.
4459 * @pre VOL_LOCK held
4461 * @note non-DAFS only (for DAFS, use @see WaitForOfflineByVp_r)
4464 VWaitForOffline_r(Error *ec, VolumeId volid)
4467 const struct timespec *ts;
4468 #ifdef AFS_PTHREAD_ENV
4469 struct timespec timeout_ts;
4472 ts = VOfflineTimeout(&timeout_ts);
4474 vp = GetVolume(ec, NULL, volid, NULL, ts);
4476 /* error occurred so bad that we can't even get a vp; we have no
4477 * information on the vol so we don't know whether to wait, so just
4481 if (!VIsGoingOffline_r(vp)) {
4482 /* volume is no longer going offline, so we're done */
4487 /* If we got here, we timed out waiting for the volume to go offline.
4488 * Kick off the accessing RX calls and wait again */
4494 vp = VGetVolume_r(ec, volid);
4496 /* In case it was reattached... */
4500 #endif /* !AFS_DEMAND_ATTACH_FS */
4502 /* The opposite of VAttachVolume. The volume header is written to disk, with
4503 the inUse bit turned off. A copy of the header is maintained in memory,
4504 however (which is why this is VOffline, not VDetach).
4507 VOffline_r(Volume * vp, char *message)
4510 #ifndef AFS_DEMAND_ATTACH_FS
4511 VolumeId vid = V_id(vp);
4514 osi_Assert(programType != volumeUtility && programType != volumeServer);
4519 if (V_offlineMessage(vp)[0] == '\0')
4520 strncpy(V_offlineMessage(vp), message, sizeof(V_offlineMessage(vp)));
4521 V_offlineMessage(vp)[sizeof(V_offlineMessage(vp)) - 1] = '\0';
4523 vp->goingOffline = 1;
4524 #ifdef AFS_DEMAND_ATTACH_FS
4525 VChangeState_r(vp, VOL_STATE_GOING_OFFLINE);
4526 VCreateReservation_r(vp);
4528 VWaitForOfflineByVp_r(&error, vp, 1);
4529 VCancelReservation_r(vp);
4530 #else /* AFS_DEMAND_ATTACH_FS */
4532 VWaitForOffline_r(&error, vid);
4533 #endif /* AFS_DEMAND_ATTACH_FS */
4536 #ifdef AFS_DEMAND_ATTACH_FS
4538 * Take a volume offline in order to perform a volume operation.
4540 * @param[inout] ec address in which to store error code
4541 * @param[in] vp volume object pointer
4542 * @param[in] message volume offline status message
4545 * - VOL_LOCK is held
4546 * - caller MUST hold a heavyweight ref on vp
4549 * - volume is taken offline
4550 * - if possible, volume operation is promoted to running state
4551 * - on failure, *ec is set to nonzero
4553 * @note Although this function does not return any value, it may
4554 * still fail to promote our pending volume operation to
4555 * a running state. Any caller MUST check the value of *ec,
4556 * and MUST NOT blindly assume success.
4558 * @warning if the caller does not hold a lightweight ref on vp,
4559 * then it MUST NOT reference vp after this function
4560 * returns to the caller.
4562 * @internal volume package internal use only
4565 VOfflineForVolOp_r(Error *ec, Volume *vp, char *message)
4568 osi_Assert(vp->pending_vol_op);
4574 if (V_offlineMessage(vp)[0] == '\0')
4575 strncpy(V_offlineMessage(vp), message, sizeof(V_offlineMessage(vp)));
4576 V_offlineMessage(vp)[sizeof(V_offlineMessage(vp)) - 1] = '\0';
4578 vp->goingOffline = 1;
4579 VChangeState_r(vp, VOL_STATE_GOING_OFFLINE);
4580 VCreateReservation_r(vp);
4583 if (vp->pending_vol_op->com.programType != salvageServer) {
4584 /* do not give corrupted volumes to the volserver */
4589 VWaitForOfflineByVp_r(ec, vp, salvok);
4591 VCancelReservation_r(vp);
4593 #endif /* AFS_DEMAND_ATTACH_FS */
4596 VOffline(Volume * vp, char *message)
4599 VOffline_r(vp, message);
4603 /* This gets used for the most part by utility routines that don't want
4604 * to keep all the volume headers around. Generally, the file server won't
4605 * call this routine, because then the offline message in the volume header
4606 * (or other information) won't be available to clients. For NAMEI, also
4607 * close the file handles. However, the fileserver does call this during
4608 * an attach following a volume operation.
4611 VDetachVolume_r(Error * ec, Volume * vp)
4613 #ifdef FSSYNC_BUILD_CLIENT
4615 struct DiskPartition64 *tpartp;
4616 int notifyServer = 0;
4617 int useDone = FSYNC_VOL_ON;
4619 if (VCanUseFSSYNC()) {
4620 notifyServer = vp->needsPutBack;
4621 if (V_destroyMe(vp) == DESTROY_ME)
4622 useDone = FSYNC_VOL_LEAVE_OFF;
4623 # ifdef AFS_DEMAND_ATTACH_FS
4624 else if (!V_blessed(vp) || !V_inService(vp))
4625 useDone = FSYNC_VOL_LEAVE_OFF;
4628 # ifdef AFS_DEMAND_ATTACH_FS
4629 if (V_needsSalvaged(vp)) {
4631 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, 0);
4634 tpartp = vp->partition;
4636 #endif /* FSSYNC_BUILD_CLIENT */
4638 *ec = 0; /* always "succeeds" */
4639 DeleteVolumeFromHashTable(vp);
4640 vp->shuttingDown = 1;
4641 #ifdef AFS_DEMAND_ATTACH_FS
4642 DeleteVolumeFromVByPList_r(vp);
4644 VChangeState_r(vp, VOL_STATE_SHUTTING_DOWN);
4646 if (programType != fileServer)
4648 #endif /* AFS_DEMAND_ATTACH_FS */
4650 /* Will be detached sometime in the future--this is OK since volume is offline */
4652 /* XXX the following code should really be moved to VCheckDetach() since the volume
4653 * is not technically detached until the refcounts reach zero
4655 #ifdef FSSYNC_BUILD_CLIENT
4656 if (VCanUseFSSYNC() && notifyServer) {
4657 if (notifyServer == VOL_PUTBACK_DELETE) {
4658 /* Only send FSYNC_VOL_DONE if the volume was actually deleted.
4659 * volserver code will set needsPutBack to VOL_PUTBACK_DELETE
4660 * to signify a deleted volume. */
4661 useDone = FSYNC_VOL_DONE;
4664 * Note: The server is not notified in the case of a bogus volume
4665 * explicitly to make it possible to create a volume, do a partial
4666 * restore, then abort the operation without ever putting the volume
4667 * online. This is essential in the case of a volume move operation
4668 * between two partitions on the same server. In that case, there
4669 * would be two instances of the same volume, one of them bogus,
4670 * which the file server would attempt to put on line
4672 FSYNC_VolOp(volume, tpartp->name, useDone, 0, NULL);
4673 /* XXX this code path is only hit by volume utilities, thus
4674 * V_BreakVolumeCallbacks will always be NULL. if we really
4675 * want to break callbacks in this path we need to use FSYNC_VolOp() */
4677 /* Dettaching it so break all callbacks on it */
4678 if (V_BreakVolumeCallbacks) {
4679 Log("volume %u detached; breaking all call backs\n", volume);
4680 (*V_BreakVolumeCallbacks) (volume);
4684 #endif /* FSSYNC_BUILD_CLIENT */
4688 VDetachVolume(Error * ec, Volume * vp)
4691 VDetachVolume_r(ec, vp);
4696 /***************************************************/
4697 /* Volume fd/inode handle closing routines */
4698 /***************************************************/
4700 /* For VDetachVolume, we close all cached file descriptors, but keep
4701 * the Inode handles in case we need to read from a busy volume.
4703 /* for demand attach, caller MUST hold ref count on vp */
4705 VCloseVolumeHandles_r(Volume * vp)
4707 #ifdef AFS_DEMAND_ATTACH_FS
4708 VolState state_save;
4710 state_save = VChangeState_r(vp, VOL_STATE_OFFLINING);
4715 DFlushVolume(vp->hashid);
4717 #ifdef AFS_DEMAND_ATTACH_FS
4721 /* DAFS: VCloseVnodeFiles_r drops the glock internally */
4722 VCloseVnodeFiles_r(vp);
4724 #ifdef AFS_DEMAND_ATTACH_FS
4728 /* Too time consuming and unnecessary for the volserver */
4729 if (programType == fileServer) {
4730 IH_CONDSYNC(vp->vnodeIndex[vLarge].handle);
4731 IH_CONDSYNC(vp->vnodeIndex[vSmall].handle);
4732 IH_CONDSYNC(vp->diskDataHandle);
4734 IH_CONDSYNC(vp->linkHandle);
4735 #endif /* AFS_NT40_ENV */
4738 IH_REALLYCLOSE(vp->vnodeIndex[vLarge].handle);
4739 IH_REALLYCLOSE(vp->vnodeIndex[vSmall].handle);
4740 IH_REALLYCLOSE(vp->diskDataHandle);
4741 IH_REALLYCLOSE(vp->linkHandle);
4743 #ifdef AFS_DEMAND_ATTACH_FS
4744 if ((V_attachFlags(vp) & VOL_LOCKED)) {
4749 VChangeState_r(vp, state_save);
4753 /* For both VForceOffline and VOffline, we close all relevant handles.
4754 * For VOffline, if we re-attach the volume, the files may possible be
4755 * different than before.
4757 /* for demand attach, caller MUST hold a ref count on vp */
4759 VReleaseVolumeHandles_r(Volume * vp)
4761 #ifdef AFS_DEMAND_ATTACH_FS
4762 VolState state_save;
4764 state_save = VChangeState_r(vp, VOL_STATE_DETACHING);
4769 DFlushVolume(vp->hashid);
4771 #ifdef AFS_DEMAND_ATTACH_FS
4775 VReleaseVnodeFiles_r(vp); /* DAFS: releases the glock internally */
4777 #ifdef AFS_DEMAND_ATTACH_FS
4781 /* Too time consuming and unnecessary for the volserver */
4782 if (programType == fileServer) {
4783 IH_CONDSYNC(vp->vnodeIndex[vLarge].handle);
4784 IH_CONDSYNC(vp->vnodeIndex[vSmall].handle);
4785 IH_CONDSYNC(vp->diskDataHandle);
4787 IH_CONDSYNC(vp->linkHandle);
4788 #endif /* AFS_NT40_ENV */
4791 IH_RELEASE(vp->vnodeIndex[vLarge].handle);
4792 IH_RELEASE(vp->vnodeIndex[vSmall].handle);
4793 IH_RELEASE(vp->diskDataHandle);
4794 IH_RELEASE(vp->linkHandle);
4796 #ifdef AFS_DEMAND_ATTACH_FS
4797 if ((V_attachFlags(vp) & VOL_LOCKED)) {
4802 VChangeState_r(vp, state_save);
4807 /***************************************************/
4808 /* Volume write and fsync routines */
4809 /***************************************************/
4812 VUpdateVolume_r(Error * ec, Volume * vp, int flags)
4814 #ifdef AFS_DEMAND_ATTACH_FS
4815 VolState state_save;
4817 if (flags & VOL_UPDATE_WAIT) {
4818 VCreateReservation_r(vp);
4819 VWaitExclusiveState_r(vp);
4824 if (programType == fileServer)
4826 (V_inUse(vp) ? V_nextVnodeUnique(vp) +
4827 200 : V_nextVnodeUnique(vp));
4829 #ifdef AFS_DEMAND_ATTACH_FS
4830 state_save = VChangeState_r(vp, VOL_STATE_UPDATING);
4834 WriteVolumeHeader_r(ec, vp);
4836 #ifdef AFS_DEMAND_ATTACH_FS
4838 VChangeState_r(vp, state_save);
4839 if (flags & VOL_UPDATE_WAIT) {
4840 VCancelReservation_r(vp);
4845 Log("VUpdateVolume: error updating volume header, volume %u (%s)\n",
4846 V_id(vp), V_name(vp));
4847 /* try to update on-disk header,
4848 * while preventing infinite recursion */
4849 if (!(flags & VOL_UPDATE_NOFORCEOFF)) {
4850 VForceOffline_r(vp, VOL_FORCEOFF_NOUPDATE);
4856 VUpdateVolume(Error * ec, Volume * vp)
4859 VUpdateVolume_r(ec, vp, VOL_UPDATE_WAIT);
4864 VSyncVolume_r(Error * ec, Volume * vp, int flags)
4868 #ifdef AFS_DEMAND_ATTACH_FS
4869 VolState state_save;
4872 if (flags & VOL_SYNC_WAIT) {
4873 VUpdateVolume_r(ec, vp, VOL_UPDATE_WAIT);
4875 VUpdateVolume_r(ec, vp, 0);
4878 #ifdef AFS_DEMAND_ATTACH_FS
4879 state_save = VChangeState_r(vp, VOL_STATE_UPDATING);
4882 fdP = IH_OPEN(V_diskDataHandle(vp));
4883 osi_Assert(fdP != NULL);
4884 code = FDH_SYNC(fdP);
4885 osi_Assert(code == 0);
4887 #ifdef AFS_DEMAND_ATTACH_FS
4889 VChangeState_r(vp, state_save);
4895 VSyncVolume(Error * ec, Volume * vp)
4898 VSyncVolume_r(ec, vp, VOL_SYNC_WAIT);
4903 /***************************************************/
4904 /* Volume dealloaction routines */
4905 /***************************************************/
4907 #ifdef AFS_DEMAND_ATTACH_FS
4909 FreeVolume(Volume * vp)
4911 /* free the heap space, iff it's safe.
4912 * otherwise, pull it out of the hash table, so it
4913 * will get deallocated when all refs to it go away */
4914 if (!VCheckFree(vp)) {
4915 DeleteVolumeFromHashTable(vp);
4916 DeleteVolumeFromVByPList_r(vp);
4918 /* make sure we invalidate the header cache entry */
4919 FreeVolumeHeader(vp);
4922 #endif /* AFS_DEMAND_ATTACH_FS */
4925 ReallyFreeVolume(Volume * vp)
4930 #ifdef AFS_DEMAND_ATTACH_FS
4932 VChangeState_r(vp, VOL_STATE_FREED);
4933 if (vp->pending_vol_op)
4934 free(vp->pending_vol_op);
4935 #endif /* AFS_DEMAND_ATTACH_FS */
4936 for (i = 0; i < nVNODECLASSES; i++)
4937 if (vp->vnodeIndex[i].bitmap)
4938 free(vp->vnodeIndex[i].bitmap);
4939 FreeVolumeHeader(vp);
4940 #ifndef AFS_DEMAND_ATTACH_FS
4941 DeleteVolumeFromHashTable(vp);
4942 #endif /* AFS_DEMAND_ATTACH_FS */
4946 /* check to see if we should shutdown this volume
4947 * returns 1 if volume was freed, 0 otherwise */
4948 #ifdef AFS_DEMAND_ATTACH_FS
4950 VCheckDetach(Volume * vp)
4955 if (vp->nUsers || vp->nWaiters)
4958 if (vp->shuttingDown) {
4960 if ((programType != fileServer) &&
4961 (V_inUse(vp) == programType) &&
4962 ((V_checkoutMode(vp) == V_VOLUPD) ||
4963 (V_checkoutMode(vp) == V_SECRETLY) ||
4964 ((V_checkoutMode(vp) == V_CLONE) &&
4965 (VolumeWriteable(vp))))) {
4967 VUpdateVolume_r(&ec, vp, VOL_UPDATE_NOFORCEOFF);
4969 Log("VCheckDetach: volume header update for volume %u "
4970 "failed with errno %d\n", vp->hashid, errno);
4973 VReleaseVolumeHandles_r(vp);
4975 ReallyFreeVolume(vp);
4976 if (programType == fileServer) {
4977 CV_BROADCAST(&vol_put_volume_cond);
4982 #else /* AFS_DEMAND_ATTACH_FS */
4984 VCheckDetach(Volume * vp)
4992 if (vp->shuttingDown) {
4994 if ((programType != fileServer) &&
4995 (V_inUse(vp) == programType) &&
4996 ((V_checkoutMode(vp) == V_VOLUPD) ||
4997 (V_checkoutMode(vp) == V_SECRETLY) ||
4998 ((V_checkoutMode(vp) == V_CLONE) &&
4999 (VolumeWriteable(vp))))) {
5001 VUpdateVolume_r(&ec, vp, VOL_UPDATE_NOFORCEOFF);
5003 Log("VCheckDetach: volume header update for volume %u failed with errno %d\n",
5007 VReleaseVolumeHandles_r(vp);
5008 ReallyFreeVolume(vp);
5009 if (programType == fileServer) {
5010 #if defined(AFS_PTHREAD_ENV)
5011 CV_BROADCAST(&vol_put_volume_cond);
5012 #else /* AFS_PTHREAD_ENV */
5013 LWP_NoYieldSignal(VPutVolume);
5014 #endif /* AFS_PTHREAD_ENV */
5019 #endif /* AFS_DEMAND_ATTACH_FS */
5021 /* check to see if we should offline this volume
5022 * return 1 if volume went offline, 0 otherwise */
5023 #ifdef AFS_DEMAND_ATTACH_FS
5025 VCheckOffline(Volume * vp)
5029 if (vp->goingOffline && !vp->nUsers) {
5031 osi_Assert(programType == fileServer);
5032 osi_Assert((V_attachState(vp) != VOL_STATE_ATTACHED) &&
5033 (V_attachState(vp) != VOL_STATE_FREED) &&
5034 (V_attachState(vp) != VOL_STATE_PREATTACHED) &&
5035 (V_attachState(vp) != VOL_STATE_UNATTACHED) &&
5036 (V_attachState(vp) != VOL_STATE_DELETED));
5040 * VOL_STATE_GOING_OFFLINE
5041 * VOL_STATE_SHUTTING_DOWN
5042 * VIsErrorState(V_attachState(vp))
5043 * VIsExclusiveState(V_attachState(vp))
5046 VCreateReservation_r(vp);
5047 VChangeState_r(vp, VOL_STATE_OFFLINING);
5050 /* must clear the goingOffline flag before we drop the glock */
5051 vp->goingOffline = 0;
5056 /* perform async operations */
5057 VUpdateVolume_r(&error, vp, 0);
5058 VCloseVolumeHandles_r(vp);
5061 if (V_offlineMessage(vp)[0]) {
5062 Log("VOffline: Volume %lu (%s) is now offline (%s)\n",
5063 afs_printable_uint32_lu(V_id(vp)), V_name(vp),
5064 V_offlineMessage(vp));
5066 Log("VOffline: Volume %lu (%s) is now offline\n",
5067 afs_printable_uint32_lu(V_id(vp)), V_name(vp));
5071 /* invalidate the volume header cache entry */
5072 FreeVolumeHeader(vp);
5074 /* if nothing changed state to error or salvaging,
5075 * drop state to unattached */
5076 if (!VIsErrorState(V_attachState(vp))) {
5077 VChangeState_r(vp, VOL_STATE_UNATTACHED);
5079 VCancelReservation_r(vp);
5080 /* no usage of vp is safe beyond this point */
5084 #else /* AFS_DEMAND_ATTACH_FS */
5086 VCheckOffline(Volume * vp)
5090 if (vp->goingOffline && !vp->nUsers) {
5092 osi_Assert(programType == fileServer);
5095 vp->goingOffline = 0;
5097 VUpdateVolume_r(&error, vp, 0);
5098 VCloseVolumeHandles_r(vp);
5100 if (V_offlineMessage(vp)[0]) {
5101 Log("VOffline: Volume %lu (%s) is now offline (%s)\n",
5102 afs_printable_uint32_lu(V_id(vp)), V_name(vp),
5103 V_offlineMessage(vp));
5105 Log("VOffline: Volume %lu (%s) is now offline\n",
5106 afs_printable_uint32_lu(V_id(vp)), V_name(vp));
5109 FreeVolumeHeader(vp);
5110 #ifdef AFS_PTHREAD_ENV
5111 CV_BROADCAST(&vol_put_volume_cond);
5112 #else /* AFS_PTHREAD_ENV */
5113 LWP_NoYieldSignal(VPutVolume);
5114 #endif /* AFS_PTHREAD_ENV */
5118 #endif /* AFS_DEMAND_ATTACH_FS */
5120 /***************************************************/
5121 /* demand attach fs ref counting routines */
5122 /***************************************************/
5124 #ifdef AFS_DEMAND_ATTACH_FS
5125 /* the following two functions handle reference counting for
5126 * asynchronous operations on volume structs.
5128 * their purpose is to prevent a VDetachVolume or VShutdown
5129 * from free()ing the Volume struct during an async i/o op */
5131 /* register with the async volume op ref counter */
5132 /* VCreateReservation_r moved into inline code header because it
5133 * is now needed in vnode.c -- tkeiser 11/20/2007
5137 * decrement volume-package internal refcount.
5139 * @param vp volume object pointer
5141 * @internal volume package internal use only
5144 * @arg VOL_LOCK is held
5145 * @arg lightweight refcount held
5147 * @post volume waiters refcount is decremented; volume may
5148 * have been deallocated/shutdown/offlined/salvaged/
5149 * whatever during the process
5151 * @warning once you have tossed your last reference (you can acquire
5152 * lightweight refs recursively) it is NOT SAFE to reference
5153 * a volume object pointer ever again
5155 * @see VCreateReservation_r
5157 * @note DEMAND_ATTACH_FS only
5160 VCancelReservation_r(Volume * vp)
5162 osi_Assert(--vp->nWaiters >= 0);
5163 if (vp->nWaiters == 0) {
5165 if (!VCheckDetach(vp)) {
5172 /* check to see if we should free this volume now
5173 * return 1 if volume was freed, 0 otherwise */
5175 VCheckFree(Volume * vp)
5178 if ((vp->nUsers == 0) &&
5179 (vp->nWaiters == 0) &&
5180 !(V_attachFlags(vp) & (VOL_IN_HASH |
5184 ReallyFreeVolume(vp);
5189 #endif /* AFS_DEMAND_ATTACH_FS */
5192 /***************************************************/
5193 /* online volume operations routines */
5194 /***************************************************/
5196 #ifdef AFS_DEMAND_ATTACH_FS
5198 * register a volume operation on a given volume.
5200 * @param[in] vp volume object
5201 * @param[in] vopinfo volume operation info object
5203 * @pre VOL_LOCK is held
5205 * @post volume operation info object attached to volume object.
5206 * volume operation statistics updated.
5208 * @note by "attached" we mean a copy of the passed in object is made
5210 * @internal volume package internal use only
5213 VRegisterVolOp_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
5215 FSSYNC_VolOp_info * info;
5217 /* attach a vol op info node to the volume struct */
5218 info = (FSSYNC_VolOp_info *) malloc(sizeof(FSSYNC_VolOp_info));
5219 osi_Assert(info != NULL);
5220 memcpy(info, vopinfo, sizeof(FSSYNC_VolOp_info));
5221 vp->pending_vol_op = info;
5224 vp->stats.last_vol_op = FT_ApproxTime();
5225 vp->stats.vol_ops++;
5226 IncUInt64(&VStats.vol_ops);
5232 * deregister the volume operation attached to this volume.
5234 * @param[in] vp volume object pointer
5236 * @pre VOL_LOCK is held
5238 * @post the volume operation info object is detached from the volume object
5240 * @internal volume package internal use only
5243 VDeregisterVolOp_r(Volume * vp)
5245 if (vp->pending_vol_op) {
5246 free(vp->pending_vol_op);
5247 vp->pending_vol_op = NULL;
5251 #endif /* AFS_DEMAND_ATTACH_FS */
5254 * determine whether it is safe to leave a volume online during
5255 * the volume operation described by the vopinfo object.
5257 * @param[in] vp volume object
5258 * @param[in] vopinfo volume operation info object
5260 * @return whether it is safe to leave volume online
5261 * @retval 0 it is NOT SAFE to leave the volume online
5262 * @retval 1 it is safe to leave the volume online during the operation
5265 * @arg VOL_LOCK is held
5266 * @arg disk header attached to vp (heavyweight ref on vp will guarantee
5267 * this condition is met)
5269 * @internal volume package internal use only
5272 VVolOpLeaveOnline_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
5274 return (vopinfo->vol_op_state == FSSYNC_VolOpRunningOnline ||
5275 (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
5276 (vopinfo->com.reason == V_READONLY ||
5277 (!VolumeWriteable(vp) &&
5278 (vopinfo->com.reason == V_CLONE ||
5279 vopinfo->com.reason == V_DUMP)))));
5283 * same as VVolOpLeaveOnline_r, but does not require a volume with an attached
5286 * @param[in] vp volume object
5287 * @param[in] vopinfo volume operation info object
5289 * @return whether it is safe to leave volume online
5290 * @retval 0 it is NOT SAFE to leave the volume online
5291 * @retval 1 it is safe to leave the volume online during the operation
5292 * @retval -1 unsure; volume header is required in order to know whether or
5293 * not is is safe to leave the volume online
5295 * @pre VOL_LOCK is held
5297 * @internal volume package internal use only
5300 VVolOpLeaveOnlineNoHeader_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
5302 /* follow the logic in VVolOpLeaveOnline_r; this is the same, except
5303 * assume that we don't know VolumeWriteable; return -1 if the answer
5304 * depends on VolumeWriteable */
5306 if (vopinfo->vol_op_state == FSSYNC_VolOpRunningOnline) {
5309 if (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
5310 vopinfo->com.reason == V_READONLY) {
5314 if (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
5315 (vopinfo->com.reason == V_CLONE ||
5316 vopinfo->com.reason == V_DUMP)) {
5318 /* must know VolumeWriteable */
5325 * determine whether VBUSY should be set during this volume operation.
5327 * @param[in] vp volume object
5328 * @param[in] vopinfo volume operation info object
5330 * @return whether VBUSY should be set
5331 * @retval 0 VBUSY does NOT need to be set
5332 * @retval 1 VBUSY SHOULD be set
5334 * @pre VOL_LOCK is held
5336 * @internal volume package internal use only
5339 VVolOpSetVBusy_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
5341 return ((vopinfo->com.command == FSYNC_VOL_OFF &&
5342 vopinfo->com.reason == FSYNC_SALVAGE) ||
5343 (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
5344 (vopinfo->com.reason == V_CLONE ||
5345 vopinfo->com.reason == V_DUMP)));
5349 /***************************************************/
5350 /* online salvager routines */
5351 /***************************************************/
5352 #if defined(AFS_DEMAND_ATTACH_FS)
5355 * offline a volume to let it be salvaged.
5357 * @param[in] vp Volume to offline
5359 * @return whether we offlined the volume successfully
5360 * @retval 0 volume was not offlined
5361 * @retval 1 volume is now offline
5363 * @note This is similar to VCheckOffline, but slightly different. We do not
5364 * deal with vp->goingOffline, and we try to avoid touching the volume
5365 * header except just to set needsSalvaged
5367 * @pre VOL_LOCK held
5368 * @pre vp->nUsers == 0
5369 * @pre V_attachState(vp) == VOL_STATE_SALVAGE_REQ
5372 VOfflineForSalvage_r(struct Volume *vp)
5376 VCreateReservation_r(vp);
5377 VWaitExclusiveState_r(vp);
5379 if (vp->nUsers || V_attachState(vp) == VOL_STATE_SALVAGING) {
5380 /* Someone's using the volume, or someone got to scheduling the salvage
5381 * before us. I don't think either of these should be possible, as we
5382 * should gain no new heavyweight references while we're trying to
5383 * salvage, but just to be sure... */
5384 VCancelReservation_r(vp);
5388 VChangeState_r(vp, VOL_STATE_OFFLINING);
5392 V_needsSalvaged(vp) = 1;
5393 /* ignore error; updating needsSalvaged is just best effort */
5394 VUpdateVolume_r(&error, vp, VOL_UPDATE_NOFORCEOFF);
5396 VCloseVolumeHandles_r(vp);
5398 FreeVolumeHeader(vp);
5400 /* volume has been effectively offlined; we can mark it in the SALVAGING
5401 * state now, which lets FSSYNC give it away */
5402 VChangeState_r(vp, VOL_STATE_SALVAGING);
5404 VCancelReservation_r(vp);
5410 * check whether a salvage needs to be performed on this volume.
5412 * @param[in] vp pointer to volume object
5414 * @return status code
5415 * @retval 0 no salvage scheduled
5416 * @retval 1 a salvage has been scheduled with the salvageserver
5418 * @pre VOL_LOCK is held
5420 * @post if salvage request flag is set and nUsers and nWaiters are zero,
5421 * then a salvage will be requested
5423 * @note this is one of the event handlers called by VCancelReservation_r
5425 * @note the caller must check if the volume needs to be freed after calling
5426 * this; the volume may not have any references or be on any lists after
5427 * we return, and we do not free it
5429 * @see VCancelReservation_r
5431 * @internal volume package internal use only.
5434 VCheckSalvage(Volume * vp)
5437 #if defined(SALVSYNC_BUILD_CLIENT) || defined(FSSYNC_BUILD_CLIENT)
5440 if (!vp->salvage.requested) {
5444 /* prevent recursion; some of the code below creates and removes
5445 * lightweight refs, which can call VCheckSalvage */
5446 if (vp->salvage.scheduling) {
5449 vp->salvage.scheduling = 1;
5451 if (V_attachState(vp) == VOL_STATE_SALVAGE_REQ) {
5452 if (!VOfflineForSalvage_r(vp)) {
5453 vp->salvage.scheduling = 0;
5458 if (vp->salvage.requested) {
5459 VScheduleSalvage_r(vp);
5462 vp->salvage.scheduling = 0;
5463 #endif /* SALVSYNC_BUILD_CLIENT || FSSYNC_BUILD_CLIENT */
5468 * request volume salvage.
5470 * @param[out] ec computed client error code
5471 * @param[in] vp volume object pointer
5472 * @param[in] reason reason code (passed to salvageserver via SALVSYNC)
5473 * @param[in] flags see flags note below
5476 * VOL_SALVAGE_NO_OFFLINE do not need to wait to offline the volume; it has
5477 * not been fully attached
5479 * @pre VOL_LOCK is held.
5481 * @post volume state is changed.
5482 * for fileserver, salvage will be requested once refcount reaches zero.
5484 * @return operation status code
5485 * @retval 0 volume salvage will occur
5486 * @retval 1 volume salvage could not be scheduled
5490 * @note in the fileserver, this call does not synchronously schedule a volume
5491 * salvage. rather, it sets volume state so that when volume refcounts
5492 * reach zero, a volume salvage will occur. by "refcounts", we mean both
5493 * nUsers and nWaiters must be zero.
5495 * @internal volume package internal use only.
5498 VRequestSalvage_r(Error * ec, Volume * vp, int reason, int flags)
5502 * for DAFS volume utilities that are not supposed to schedule salvages,
5503 * just transition to error state instead
5505 if (!VCanScheduleSalvage()) {
5506 VChangeState_r(vp, VOL_STATE_ERROR);
5511 if (programType != fileServer && !VCanUseFSSYNC()) {
5512 VChangeState_r(vp, VOL_STATE_ERROR);
5517 if (!vp->salvage.requested) {
5518 vp->salvage.requested = 1;
5519 vp->salvage.reason = reason;
5520 vp->stats.last_salvage = FT_ApproxTime();
5522 /* Note that it is not possible for us to reach this point if a
5523 * salvage is already running on this volume (even if the fileserver
5524 * was restarted during the salvage). If a salvage were running, the
5525 * salvager would have write-locked the volume header file, so when
5526 * we tried to lock the volume header, the lock would have failed,
5527 * and we would have failed during attachment prior to calling
5528 * VRequestSalvage. So we know that we can schedule salvages without
5529 * fear of a salvage already running for this volume. */
5531 if (vp->stats.salvages < SALVAGE_COUNT_MAX) {
5533 /* if we don't need to offline the volume, we can go directly
5534 * to SALVAGING. SALVAGING says the volume is offline and is
5535 * either salvaging or ready to be handed to the salvager.
5536 * SALVAGE_REQ says that we want to salvage the volume, but we
5537 * are waiting for it to go offline first. */
5538 if (flags & VOL_SALVAGE_NO_OFFLINE) {
5539 VChangeState_r(vp, VOL_STATE_SALVAGING);
5541 VChangeState_r(vp, VOL_STATE_SALVAGE_REQ);
5542 if (vp->nUsers == 0) {
5543 /* normally VOfflineForSalvage_r would be called from
5544 * PutVolume et al when nUsers reaches 0, but if
5545 * it's already 0, just do it ourselves, since PutVolume
5546 * isn't going to get called */
5547 VOfflineForSalvage_r(vp);
5550 /* If we are non-fileserver, we're telling the fileserver to
5551 * salvage the vol, so we don't need to give it back separately. */
5552 vp->needsPutBack = 0;
5556 Log("VRequestSalvage: volume %u online salvaged too many times; forced offline.\n", vp->hashid);
5558 /* make sure neither VScheduleSalvage_r nor
5559 * VUpdateSalvagePriority_r try to schedule another salvage */
5560 vp->salvage.requested = vp->salvage.scheduled = 0;
5562 VChangeState_r(vp, VOL_STATE_ERROR);
5571 * update salvageserver scheduling priority for a volume.
5573 * @param[in] vp pointer to volume object
5575 * @return operation status
5577 * @retval 1 request denied, or SALVSYNC communications failure
5579 * @pre VOL_LOCK is held.
5581 * @post in-core salvage priority counter is incremented. if at least
5582 * SALVAGE_PRIO_UPDATE_INTERVAL seconds have elapsed since the
5583 * last SALVSYNC_RAISEPRIO request, we contact the salvageserver
5584 * to update its priority queue. if no salvage is scheduled,
5585 * this function is a no-op.
5587 * @note DAFS fileserver only
5589 * @note this should be called whenever a VGetVolume fails due to a
5590 * pending salvage request
5592 * @todo should set exclusive state and drop glock around salvsync call
5594 * @internal volume package internal use only.
5597 VUpdateSalvagePriority_r(Volume * vp)
5601 #ifdef SALVSYNC_BUILD_CLIENT
5606 now = FT_ApproxTime();
5608 /* update the salvageserver priority queue occasionally so that
5609 * frequently requested volumes get moved to the head of the queue
5611 if ((vp->salvage.scheduled) &&
5612 (vp->stats.last_salvage_req < (now-SALVAGE_PRIO_UPDATE_INTERVAL))) {
5613 code = SALVSYNC_SalvageVolume(vp->hashid,
5614 VPartitionPath(vp->partition),
5619 vp->stats.last_salvage_req = now;
5620 if (code != SYNC_OK) {
5624 #endif /* SALVSYNC_BUILD_CLIENT */
5629 #if defined(SALVSYNC_BUILD_CLIENT) || defined(FSSYNC_BUILD_CLIENT)
5631 /* A couple of little helper functions. These return true if we tried to
5632 * use this mechanism to schedule a salvage, false if we haven't tried.
5633 * If we did try a salvage then the results are contained in code.
5637 try_SALVSYNC(Volume *vp, char *partName, int *code) {
5638 #ifdef SALVSYNC_BUILD_CLIENT
5639 if (VCanUseSALVSYNC()) {
5640 Log("Scheduling salvage for volume %lu on part %s over SALVSYNC\n",
5641 afs_printable_uint32_lu(vp->hashid), partName);
5643 /* can't use V_id() since there's no guarantee
5644 * we have the disk data header at this point */
5645 *code = SALVSYNC_SalvageVolume(vp->hashid,
5658 try_FSSYNC(Volume *vp, char *partName, int *code) {
5659 #ifdef FSSYNC_BUILD_CLIENT
5660 if (VCanUseFSSYNC()) {
5661 Log("Scheduling salvage for volume %lu on part %s over FSSYNC\n",
5662 afs_printable_uint32_lu(vp->hashid), partName);
5665 * If we aren't the fileserver, tell the fileserver the volume
5666 * needs to be salvaged. We could directly tell the
5667 * salvageserver, but the fileserver keeps track of some stats
5668 * related to salvages, and handles some other salvage-related
5669 * complications for us.
5671 *code = FSYNC_VolOp(vp->hashid, partName,
5672 FSYNC_VOL_FORCE_ERROR, FSYNC_SALVAGE, NULL);
5675 #endif /* FSSYNC_BUILD_CLIENT */
5680 * schedule a salvage with the salvage server or fileserver.
5682 * @param[in] vp pointer to volume object
5684 * @return operation status
5685 * @retval 0 salvage scheduled successfully
5686 * @retval 1 salvage not scheduled, or SALVSYNC/FSSYNC com error
5689 * @arg VOL_LOCK is held.
5690 * @arg nUsers and nWaiters should be zero.
5692 * @post salvageserver or fileserver is sent a salvage request
5694 * @note If we are the fileserver, the request will be sent to the salvage
5695 * server over SALVSYNC. If we are not the fileserver, the request will be
5696 * sent to the fileserver over FSSYNC (FSYNC_VOL_FORCE_ERROR/FSYNC_SALVAGE).
5698 * @note the caller must check if the volume needs to be freed after calling
5699 * this; the volume may not have any references or be on any lists after
5700 * we return, and we do not free it
5704 * @internal volume package internal use only.
5707 VScheduleSalvage_r(Volume * vp)
5711 VolState state_save;
5712 VThreadOptions_t * thread_opts;
5715 osi_Assert(VCanUseSALVSYNC() || VCanUseFSSYNC());
5717 if (vp->nWaiters || vp->nUsers) {
5721 /* prevent endless salvage,attach,salvage,attach,... loops */
5722 if (vp->stats.salvages >= SALVAGE_COUNT_MAX)
5726 * don't perform salvsync ops on certain threads
5728 thread_opts = pthread_getspecific(VThread_key);
5729 if (thread_opts == NULL) {
5730 thread_opts = &VThread_defaults;
5732 if (thread_opts->disallow_salvsync || vol_disallow_salvsync) {
5736 if (vp->salvage.scheduled) {
5740 VCreateReservation_r(vp);
5741 VWaitExclusiveState_r(vp);
5744 * XXX the scheduling process should really be done asynchronously
5745 * to avoid fssync deadlocks
5747 if (!vp->salvage.scheduled) {
5748 /* if we haven't previously scheduled a salvage, do so now
5750 * set the volume to an exclusive state and drop the lock
5751 * around the SALVSYNC call
5753 strlcpy(partName, vp->partition->name, sizeof(partName));
5754 state_save = VChangeState_r(vp, VOL_STATE_SALVSYNC_REQ);
5757 osi_Assert(try_SALVSYNC(vp, partName, &code) ||
5758 try_FSSYNC(vp, partName, &code));
5761 VChangeState_r(vp, state_save);
5763 if (code == SYNC_OK) {
5764 vp->salvage.scheduled = 1;
5765 vp->stats.last_salvage_req = FT_ApproxTime();
5766 if (VCanUseSALVSYNC()) {
5767 /* don't record these stats for non-fileservers; let the
5768 * fileserver take care of these */
5769 vp->stats.salvages++;
5770 IncUInt64(&VStats.salvages);
5775 case SYNC_BAD_COMMAND:
5776 case SYNC_COM_ERROR:
5779 Log("VScheduleSalvage_r: Salvage request for volume %lu "
5780 "denied\n", afs_printable_uint32_lu(vp->hashid));
5783 Log("VScheduleSalvage_r: Salvage request for volume %lu "
5784 "received unknown protocol error %d\n",
5785 afs_printable_uint32_lu(vp->hashid), code);
5789 if (VCanUseFSSYNC()) {
5790 VChangeState_r(vp, VOL_STATE_ERROR);
5795 /* NB: this is cancelling the reservation we obtained above, but we do
5796 * not call VCancelReservation_r, since that may trigger the vp dtor,
5797 * possibly free'ing the vp. We need to keep the vp around after
5798 * this, as the caller may reference vp without any refs. Instead, it
5799 * is the duty of the caller to inspect 'vp' after we return to see if
5800 * needs to be freed. */
5801 osi_Assert(--vp->nWaiters >= 0);
5804 #endif /* SALVSYNC_BUILD_CLIENT || FSSYNC_BUILD_CLIENT */
5806 #ifdef SALVSYNC_BUILD_CLIENT
5809 * connect to the salvageserver SYNC service.
5811 * @return operation status
5815 * @post connection to salvageserver SYNC service established
5817 * @see VConnectSALV_r
5818 * @see VDisconnectSALV
5819 * @see VReconnectSALV
5826 retVal = VConnectSALV_r();
5832 * connect to the salvageserver SYNC service.
5834 * @return operation status
5838 * @pre VOL_LOCK is held.
5840 * @post connection to salvageserver SYNC service established
5843 * @see VDisconnectSALV_r
5844 * @see VReconnectSALV_r
5845 * @see SALVSYNC_clientInit
5847 * @internal volume package internal use only.
5850 VConnectSALV_r(void)
5852 return SALVSYNC_clientInit();
5856 * disconnect from the salvageserver SYNC service.
5858 * @return operation status
5861 * @pre client should have a live connection to the salvageserver
5863 * @post connection to salvageserver SYNC service destroyed
5865 * @see VDisconnectSALV_r
5867 * @see VReconnectSALV
5870 VDisconnectSALV(void)
5873 VDisconnectSALV_r();
5879 * disconnect from the salvageserver SYNC service.
5881 * @return operation status
5885 * @arg VOL_LOCK is held.
5886 * @arg client should have a live connection to the salvageserver.
5888 * @post connection to salvageserver SYNC service destroyed
5890 * @see VDisconnectSALV
5891 * @see VConnectSALV_r
5892 * @see VReconnectSALV_r
5893 * @see SALVSYNC_clientFinis
5895 * @internal volume package internal use only.
5898 VDisconnectSALV_r(void)
5900 return SALVSYNC_clientFinis();
5904 * disconnect and then re-connect to the salvageserver SYNC service.
5906 * @return operation status
5910 * @pre client should have a live connection to the salvageserver
5912 * @post old connection is dropped, and a new one is established
5915 * @see VDisconnectSALV
5916 * @see VReconnectSALV_r
5919 VReconnectSALV(void)
5923 retVal = VReconnectSALV_r();
5929 * disconnect and then re-connect to the salvageserver SYNC service.
5931 * @return operation status
5936 * @arg VOL_LOCK is held.
5937 * @arg client should have a live connection to the salvageserver.
5939 * @post old connection is dropped, and a new one is established
5941 * @see VConnectSALV_r
5942 * @see VDisconnectSALV
5943 * @see VReconnectSALV
5944 * @see SALVSYNC_clientReconnect
5946 * @internal volume package internal use only.
5949 VReconnectSALV_r(void)
5951 return SALVSYNC_clientReconnect();
5953 #endif /* SALVSYNC_BUILD_CLIENT */
5954 #endif /* AFS_DEMAND_ATTACH_FS */
5957 /***************************************************/
5958 /* FSSYNC routines */
5959 /***************************************************/
5961 /* This must be called by any volume utility which needs to run while the
5962 file server is also running. This is separated from VInitVolumePackage2 so
5963 that a utility can fork--and each of the children can independently
5964 initialize communication with the file server */
5965 #ifdef FSSYNC_BUILD_CLIENT
5967 * connect to the fileserver SYNC service.
5969 * @return operation status
5974 * @arg VInit must equal 2.
5975 * @arg Program Type must not be fileserver or salvager.
5977 * @post connection to fileserver SYNC service established
5980 * @see VDisconnectFS
5981 * @see VChildProcReconnectFS
5988 retVal = VConnectFS_r();
5994 * connect to the fileserver SYNC service.
5996 * @return operation status
6001 * @arg VInit must equal 2.
6002 * @arg Program Type must not be fileserver or salvager.
6003 * @arg VOL_LOCK is held.
6005 * @post connection to fileserver SYNC service established
6008 * @see VDisconnectFS_r
6009 * @see VChildProcReconnectFS_r
6011 * @internal volume package internal use only.
6017 osi_Assert((VInit == 2) &&
6018 (programType != fileServer) &&
6019 (programType != salvager));
6020 rc = FSYNC_clientInit();
6028 * disconnect from the fileserver SYNC service.
6031 * @arg client should have a live connection to the fileserver.
6032 * @arg VOL_LOCK is held.
6033 * @arg Program Type must not be fileserver or salvager.
6035 * @post connection to fileserver SYNC service destroyed
6037 * @see VDisconnectFS
6039 * @see VChildProcReconnectFS_r
6041 * @internal volume package internal use only.
6044 VDisconnectFS_r(void)
6046 osi_Assert((programType != fileServer) &&
6047 (programType != salvager));
6048 FSYNC_clientFinis();
6053 * disconnect from the fileserver SYNC service.
6056 * @arg client should have a live connection to the fileserver.
6057 * @arg Program Type must not be fileserver or salvager.
6059 * @post connection to fileserver SYNC service destroyed
6061 * @see VDisconnectFS_r
6063 * @see VChildProcReconnectFS
6074 * connect to the fileserver SYNC service from a child process following a fork.
6076 * @return operation status
6081 * @arg VOL_LOCK is held.
6082 * @arg current FSYNC handle is shared with a parent process
6084 * @post current FSYNC handle is discarded and a new connection to the
6085 * fileserver SYNC service is established
6087 * @see VChildProcReconnectFS
6089 * @see VDisconnectFS_r
6091 * @internal volume package internal use only.
6094 VChildProcReconnectFS_r(void)
6096 return FSYNC_clientChildProcReconnect();
6100 * connect to the fileserver SYNC service from a child process following a fork.
6102 * @return operation status
6106 * @pre current FSYNC handle is shared with a parent process
6108 * @post current FSYNC handle is discarded and a new connection to the
6109 * fileserver SYNC service is established
6111 * @see VChildProcReconnectFS_r
6113 * @see VDisconnectFS
6116 VChildProcReconnectFS(void)
6120 ret = VChildProcReconnectFS_r();
6124 #endif /* FSSYNC_BUILD_CLIENT */
6127 /***************************************************/
6128 /* volume bitmap routines */
6129 /***************************************************/
6132 * allocate a vnode bitmap number for the vnode
6134 * @param[out] ec error code
6135 * @param[in] vp volume object pointer
6136 * @param[in] index vnode index number for the vnode
6137 * @param[in] flags flag values described in note
6139 * @note for DAFS, flags parameter controls locking behavior.
6140 * If (flags & VOL_ALLOC_BITMAP_WAIT) is set, then this function
6141 * will create a reservation and block on any other exclusive
6142 * operations. Otherwise, this function assumes the caller
6143 * already has exclusive access to vp, and we just change the
6146 * @pre VOL_LOCK held
6148 * @return bit number allocated
6154 VAllocBitmapEntry_r(Error * ec, Volume * vp,
6155 struct vnodeIndex *index, int flags)
6159 #ifdef AFS_DEMAND_ATTACH_FS
6160 VolState state_save;
6161 #endif /* AFS_DEMAND_ATTACH_FS */
6165 /* This test is probably redundant */
6166 if (!VolumeWriteable(vp)) {
6167 *ec = (bit32) VREADONLY;
6171 #ifdef AFS_DEMAND_ATTACH_FS
6172 if (flags & VOL_ALLOC_BITMAP_WAIT) {
6173 VCreateReservation_r(vp);
6174 VWaitExclusiveState_r(vp);
6176 state_save = VChangeState_r(vp, VOL_STATE_GET_BITMAP);
6177 #endif /* AFS_DEMAND_ATTACH_FS */
6180 if ((programType == fileServer) && !index->bitmap) {
6182 #ifndef AFS_DEMAND_ATTACH_FS
6183 /* demand attach fs uses the volume state to avoid races.
6184 * specialStatus field is not used at all */
6186 if (vp->specialStatus == VBUSY) {
6187 if (vp->goingOffline) { /* vos dump waiting for the volume to
6188 * go offline. We probably come here
6189 * from AddNewReadableResidency */
6192 while (vp->specialStatus == VBUSY) {
6193 #ifdef AFS_PTHREAD_ENV
6197 #else /* !AFS_PTHREAD_ENV */
6199 #endif /* !AFS_PTHREAD_ENV */
6203 #endif /* !AFS_DEMAND_ATTACH_FS */
6205 if (!index->bitmap) {
6206 #ifndef AFS_DEMAND_ATTACH_FS
6207 vp->specialStatus = VBUSY; /* Stop anyone else from using it. */
6208 #endif /* AFS_DEMAND_ATTACH_FS */
6209 for (i = 0; i < nVNODECLASSES; i++) {
6210 VGetBitmap_r(ec, vp, i);
6212 #ifdef AFS_DEMAND_ATTACH_FS
6213 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, 0 /*flags*/);
6214 #else /* AFS_DEMAND_ATTACH_FS */
6215 DeleteVolumeFromHashTable(vp);
6216 vp->shuttingDown = 1; /* Let who has it free it. */
6217 vp->specialStatus = 0;
6218 #endif /* AFS_DEMAND_ATTACH_FS */
6222 #ifndef AFS_DEMAND_ATTACH_FS
6224 vp->specialStatus = 0; /* Allow others to have access. */
6225 #endif /* AFS_DEMAND_ATTACH_FS */
6228 #endif /* BITMAP_LATER */
6230 #ifdef AFS_DEMAND_ATTACH_FS
6232 #endif /* AFS_DEMAND_ATTACH_FS */
6233 bp = index->bitmap + index->bitmapOffset;
6234 ep = index->bitmap + index->bitmapSize;
6236 if ((*(bit32 *) bp) != (bit32) 0xffffffff) {
6238 index->bitmapOffset = (afs_uint32) (bp - index->bitmap);
6241 o = ffs(~*bp) - 1; /* ffs is documented in BSTRING(3) */
6243 ret = ((bp - index->bitmap) * 8 + o);
6244 #ifdef AFS_DEMAND_ATTACH_FS
6246 #endif /* AFS_DEMAND_ATTACH_FS */
6249 bp += sizeof(bit32) /* i.e. 4 */ ;
6251 /* No bit map entry--must grow bitmap */
6253 realloc(index->bitmap, index->bitmapSize + VOLUME_BITMAP_GROWSIZE);
6254 osi_Assert(bp != NULL);
6256 bp += index->bitmapSize;
6257 memset(bp, 0, VOLUME_BITMAP_GROWSIZE);
6258 index->bitmapOffset = index->bitmapSize;
6259 index->bitmapSize += VOLUME_BITMAP_GROWSIZE;
6261 ret = index->bitmapOffset * 8;
6262 #ifdef AFS_DEMAND_ATTACH_FS
6264 #endif /* AFS_DEMAND_ATTACH_FS */
6267 #ifdef AFS_DEMAND_ATTACH_FS
6268 VChangeState_r(vp, state_save);
6269 if (flags & VOL_ALLOC_BITMAP_WAIT) {
6270 VCancelReservation_r(vp);
6272 #endif /* AFS_DEMAND_ATTACH_FS */
6277 VAllocBitmapEntry(Error * ec, Volume * vp, struct vnodeIndex * index)
6281 retVal = VAllocBitmapEntry_r(ec, vp, index, VOL_ALLOC_BITMAP_WAIT);
6287 VFreeBitMapEntry_r(Error * ec, Volume *vp, struct vnodeIndex *index,
6288 unsigned bitNumber, int flags)
6290 unsigned int offset;
6294 #ifdef AFS_DEMAND_ATTACH_FS
6295 if (flags & VOL_FREE_BITMAP_WAIT) {
6296 /* VAllocBitmapEntry_r allocs bitmap entries under an exclusive volume
6297 * state, so ensure we're not in an exclusive volume state when we update
6299 VCreateReservation_r(vp);
6300 VWaitExclusiveState_r(vp);
6307 #endif /* BITMAP_LATER */
6309 offset = bitNumber >> 3;
6310 if (offset >= index->bitmapSize) {
6314 if (offset < index->bitmapOffset)
6315 index->bitmapOffset = offset & ~3; /* Truncate to nearest bit32 */
6316 *(index->bitmap + offset) &= ~(1 << (bitNumber & 0x7));
6319 #ifdef AFS_DEMAND_ATTACH_FS
6320 VCancelReservation_r(vp);
6322 return; /* make the compiler happy for non-DAFS */
6326 VFreeBitMapEntry(Error * ec, Volume *vp, struct vnodeIndex *index,
6330 VFreeBitMapEntry_r(ec, vp, index, bitNumber, VOL_FREE_BITMAP_WAIT);
6334 /* this function will drop the glock internally.
6335 * for old pthread fileservers, this is safe thanks to vbusy.
6337 * for demand attach fs, caller must have already called
6338 * VCreateReservation_r and VWaitExclusiveState_r */
6340 VGetBitmap_r(Error * ec, Volume * vp, VnodeClass class)
6342 StreamHandle_t *file;
6343 afs_sfsize_t nVnodes, size;
6344 struct VnodeClassInfo *vcp = &VnodeClassInfo[class];
6345 struct vnodeIndex *vip = &vp->vnodeIndex[class];
6346 struct VnodeDiskObject *vnode;
6347 unsigned int unique = 0;
6351 #endif /* BITMAP_LATER */
6352 #ifdef AFS_DEMAND_ATTACH_FS
6353 VolState state_save;
6354 #endif /* AFS_DEMAND_ATTACH_FS */
6358 #ifdef AFS_DEMAND_ATTACH_FS
6359 state_save = VChangeState_r(vp, VOL_STATE_GET_BITMAP);
6360 #endif /* AFS_DEMAND_ATTACH_FS */
6363 fdP = IH_OPEN(vip->handle);
6364 osi_Assert(fdP != NULL);
6365 file = FDH_FDOPEN(fdP, "r");
6366 osi_Assert(file != NULL);
6367 vnode = (VnodeDiskObject *) malloc(vcp->diskSize);
6368 osi_Assert(vnode != NULL);
6369 size = OS_SIZE(fdP->fd_fd);
6370 osi_Assert(size != -1);
6371 nVnodes = (size <= vcp->diskSize ? 0 : size - vcp->diskSize)
6373 vip->bitmapSize = ((nVnodes / 8) + 10) / 4 * 4; /* The 10 is a little extra so
6374 * a few files can be created in this volume,
6375 * the whole thing is rounded up to nearest 4
6376 * bytes, because the bit map allocator likes
6379 BitMap = (byte *) calloc(1, vip->bitmapSize);
6380 osi_Assert(BitMap != NULL);
6381 #else /* BITMAP_LATER */
6382 vip->bitmap = (byte *) calloc(1, vip->bitmapSize);
6383 osi_Assert(vip->bitmap != NULL);
6384 vip->bitmapOffset = 0;
6385 #endif /* BITMAP_LATER */
6386 if (STREAM_ASEEK(file, vcp->diskSize) != -1) {
6388 for (bitNumber = 0; bitNumber < nVnodes + 100; bitNumber++) {
6389 if (STREAM_READ(vnode, vcp->diskSize, 1, file) != 1)
6391 if (vnode->type != vNull) {
6392 if (vnode->vnodeMagic != vcp->magic) {
6393 Log("GetBitmap: addled vnode index in volume %s; volume needs salvage\n", V_name(vp));
6398 *(BitMap + (bitNumber >> 3)) |= (1 << (bitNumber & 0x7));
6399 #else /* BITMAP_LATER */
6400 *(vip->bitmap + (bitNumber >> 3)) |= (1 << (bitNumber & 0x7));
6401 #endif /* BITMAP_LATER */
6402 if (unique <= vnode->uniquifier)
6403 unique = vnode->uniquifier + 1;
6405 #ifndef AFS_PTHREAD_ENV
6406 if ((bitNumber & 0x00ff) == 0x0ff) { /* every 256 iterations */
6409 #endif /* !AFS_PTHREAD_ENV */
6412 if (vp->nextVnodeUnique < unique) {
6413 Log("GetBitmap: bad volume uniquifier for volume %s; volume needs salvage\n", V_name(vp));
6416 /* Paranoia, partly justified--I think fclose after fdopen
6417 * doesn't seem to close fd. In any event, the documentation
6418 * doesn't specify, so it's safer to close it twice.
6426 /* There may have been a racing condition with some other thread, both
6427 * creating the bitmaps for this volume. If the other thread was faster
6428 * the pointer to bitmap should already be filled and we can free ours.
6430 if (vip->bitmap == NULL) {
6431 vip->bitmap = BitMap;
6432 vip->bitmapOffset = 0;
6434 free((byte *) BitMap);
6435 #endif /* BITMAP_LATER */
6436 #ifdef AFS_DEMAND_ATTACH_FS
6437 VChangeState_r(vp, state_save);
6438 #endif /* AFS_DEMAND_ATTACH_FS */
6442 /***************************************************/
6443 /* Volume Path and Volume Number utility routines */
6444 /***************************************************/
6447 * find the first occurrence of a volume header file and return the path.
6449 * @param[out] ec outbound error code
6450 * @param[in] volumeId volume id to find
6451 * @param[out] partitionp pointer to disk partition path string
6452 * @param[out] namep pointer to volume header file name string
6454 * @post path to first occurrence of volume header is returned in partitionp
6455 * and namep, or ec is set accordingly.
6457 * @warning this function is NOT re-entrant -- partitionp and namep point to
6458 * static data segments
6460 * @note if a volume utility inadvertently leaves behind a stale volume header
6461 * on a vice partition, it is possible for callers to get the wrong one,
6462 * depending on the order of the disk partition linked list.
6466 VGetVolumePath(Error * ec, VolId volumeId, char **partitionp, char **namep)
6468 static char partition[VMAXPATHLEN], name[VMAXPATHLEN];
6469 char path[VMAXPATHLEN];
6471 struct DiskPartition64 *dp;
6474 name[0] = OS_DIRSEPC;
6475 snprintf(&name[1], (sizeof name) - 1, VFORMAT,
6476 afs_printable_uint32_lu(volumeId));
6477 for (dp = DiskPartitionList; dp; dp = dp->next) {
6478 struct afs_stat_st status;
6479 strcpy(path, VPartitionPath(dp));
6481 if (afs_stat(path, &status) == 0) {
6482 strcpy(partition, dp->name);
6489 *partitionp = *namep = NULL;
6491 *partitionp = partition;
6497 * extract a volume number from a volume header filename string.
6499 * @param[in] name volume header filename string
6501 * @return volume number
6503 * @note the string must be of the form VFORMAT. the only permissible
6504 * deviation is a leading OS_DIRSEPC character.
6509 VolumeNumber(char *name)
6511 if (*name == OS_DIRSEPC)
6513 return strtoul(name + 1, NULL, 10);
6517 * compute the volume header filename.
6519 * @param[in] volumeId
6521 * @return volume header filename
6523 * @post volume header filename string is constructed
6525 * @warning this function is NOT re-entrant -- the returned string is
6526 * stored in a static char array. see VolumeExternalName_r
6527 * for a re-entrant equivalent.
6529 * @see VolumeExternalName_r
6531 * @deprecated due to the above re-entrancy warning, this interface should
6532 * be considered deprecated. Please use VolumeExternalName_r
6536 VolumeExternalName(VolumeId volumeId)
6538 static char name[VMAXPATHLEN];
6539 snprintf(name, sizeof name, VFORMAT, afs_printable_uint32_lu(volumeId));
6544 * compute the volume header filename.
6546 * @param[in] volumeId
6547 * @param[inout] name array in which to store filename
6548 * @param[in] len length of name array
6550 * @return result code from afs_snprintf
6552 * @see VolumeExternalName
6555 * @note re-entrant equivalent of VolumeExternalName
6558 VolumeExternalName_r(VolumeId volumeId, char * name, size_t len)
6560 return snprintf(name, len, VFORMAT, afs_printable_uint32_lu(volumeId));
6564 /***************************************************/
6565 /* Volume Usage Statistics routines */
6566 /***************************************************/
6568 #if OPENAFS_VOL_STATS
6569 #define OneDay (86400) /* 24 hours' worth of seconds */
6571 #define OneDay (24*60*60) /* 24 hours */
6572 #endif /* OPENAFS_VOL_STATS */
6575 Midnight(time_t t) {
6576 struct tm local, *l;
6579 #if defined(AFS_PTHREAD_ENV) && !defined(AFS_NT40_ENV)
6580 l = localtime_r(&t, &local);
6586 /* the following is strictly speaking problematic on the
6587 switching day to daylight saving time, after the switch,
6588 as tm_isdst does not match. Similarly, on the looong day when
6589 switching back the OneDay check will not do what naively expected!
6590 The effects are minor, though, and more a matter of interpreting
6592 #ifndef AFS_PTHREAD_ENV
6595 local.tm_hour = local.tm_min=local.tm_sec = 0;
6596 midnight = mktime(&local);
6597 if (midnight != (time_t) -1) return(midnight);
6599 return( (t/OneDay)*OneDay );
6603 /*------------------------------------------------------------------------
6604 * [export] VAdjustVolumeStatistics
6607 * If we've passed midnight, we need to update all the day use
6608 * statistics as well as zeroing the detailed volume statistics
6609 * (if we are implementing them).
6612 * vp : Pointer to the volume structure describing the lucky
6613 * volume being considered for update.
6619 * Nothing interesting.
6623 *------------------------------------------------------------------------*/
6626 VAdjustVolumeStatistics_r(Volume * vp)
6628 unsigned int now = FT_ApproxTime();
6630 if (now - V_dayUseDate(vp) > OneDay) {
6633 ndays = (now - V_dayUseDate(vp)) / OneDay;
6634 for (i = 6; i > ndays - 1; i--)
6635 V_weekUse(vp)[i] = V_weekUse(vp)[i - ndays];
6636 for (i = 0; i < ndays - 1 && i < 7; i++)
6637 V_weekUse(vp)[i] = 0;
6639 V_weekUse(vp)[ndays - 1] = V_dayUse(vp);
6641 V_dayUseDate(vp) = Midnight(now);
6643 #if OPENAFS_VOL_STATS
6645 * All we need to do is bzero the entire VOL_STATS_BYTES of
6646 * the detailed volume statistics area.
6648 memset((V_stat_area(vp)), 0, VOL_STATS_BYTES);
6649 #endif /* OPENAFS_VOL_STATS */
6652 /*It's been more than a day of collection */
6654 * Always return happily.
6657 } /*VAdjustVolumeStatistics */
6660 VAdjustVolumeStatistics(Volume * vp)
6664 retVal = VAdjustVolumeStatistics_r(vp);
6670 VBumpVolumeUsage_r(Volume * vp)
6672 unsigned int now = FT_ApproxTime();
6673 V_accessDate(vp) = now;
6674 if (now - V_dayUseDate(vp) > OneDay)
6675 VAdjustVolumeStatistics_r(vp);
6677 * Save the volume header image to disk after every 128 bumps to dayUse.
6679 if ((V_dayUse(vp)++ & 127) == 0) {
6681 VUpdateVolume_r(&error, vp, VOL_UPDATE_WAIT);
6686 VBumpVolumeUsage(Volume * vp)
6689 VBumpVolumeUsage_r(vp);
6694 VSetDiskUsage_r(void)
6696 #ifndef AFS_DEMAND_ATTACH_FS
6697 static int FifteenMinuteCounter = 0;
6701 /* NOTE: Don't attempt to access the partitions list until the
6702 * initialization level indicates that all volumes are attached,
6703 * which implies that all partitions are initialized. */
6704 #ifdef AFS_PTHREAD_ENV
6705 VOL_CV_WAIT(&vol_vinit_cond);
6706 #else /* AFS_PTHREAD_ENV */
6708 #endif /* AFS_PTHREAD_ENV */
6711 VResetDiskUsage_r();
6713 #ifndef AFS_DEMAND_ATTACH_FS
6714 if (++FifteenMinuteCounter == 3) {
6715 FifteenMinuteCounter = 0;
6718 #endif /* !AFS_DEMAND_ATTACH_FS */
6730 /***************************************************/
6731 /* Volume Update List routines */
6732 /***************************************************/
6734 /* The number of minutes that a volume hasn't been updated before the
6735 * "Dont salvage" flag in the volume header will be turned on */
6736 #define SALVAGE_INTERVAL (10*60)
6741 * volume update list functionality has been moved into the VLRU
6742 * the DONT_SALVAGE flag is now set during VLRU demotion
6745 #ifndef AFS_DEMAND_ATTACH_FS
6746 static VolumeId *UpdateList = NULL; /* Pointer to array of Volume ID's */
6747 static int nUpdatedVolumes = 0; /* Updated with entry in UpdateList, salvage after crash flag on */
6748 static int updateSize = 0; /* number of entries possible */
6749 #define UPDATE_LIST_SIZE 128 /* initial size increment (must be a power of 2!) */
6750 #endif /* !AFS_DEMAND_ATTACH_FS */
6753 VAddToVolumeUpdateList_r(Error * ec, Volume * vp)
6756 vp->updateTime = FT_ApproxTime();
6757 if (V_dontSalvage(vp) == 0)
6759 V_dontSalvage(vp) = 0;
6760 VSyncVolume_r(ec, vp, 0);
6761 #ifdef AFS_DEMAND_ATTACH_FS
6762 V_attachFlags(vp) &= ~(VOL_HDR_DONTSALV);
6763 #else /* !AFS_DEMAND_ATTACH_FS */
6766 if (UpdateList == NULL) {
6767 updateSize = UPDATE_LIST_SIZE;
6768 UpdateList = (VolumeId *) malloc(sizeof(VolumeId) * updateSize);
6770 if (nUpdatedVolumes == updateSize) {
6772 if (updateSize > 524288) {
6773 Log("warning: there is likely a bug in the volume update scanner\n");
6777 (VolumeId *) realloc(UpdateList,
6778 sizeof(VolumeId) * updateSize);
6781 osi_Assert(UpdateList != NULL);
6782 UpdateList[nUpdatedVolumes++] = V_id(vp);
6783 #endif /* !AFS_DEMAND_ATTACH_FS */
6786 #ifndef AFS_DEMAND_ATTACH_FS
6788 VScanUpdateList(void)
6793 afs_uint32 now = FT_ApproxTime();
6794 /* Be careful with this code, since it works with interleaved calls to AddToVolumeUpdateList */
6795 for (i = gap = 0; i < nUpdatedVolumes; i++) {
6797 UpdateList[i - gap] = UpdateList[i];
6799 /* XXX this routine needlessly messes up the Volume LRU by
6800 * breaking the LRU temporal-locality assumptions.....
6801 * we should use a special volume header allocator here */
6802 vp = VGetVolume_r(&error, UpdateList[i - gap] = UpdateList[i]);
6805 } else if (vp->nUsers == 1 && now - vp->updateTime > SALVAGE_INTERVAL) {
6806 V_dontSalvage(vp) = DONT_SALVAGE;
6807 VUpdateVolume_r(&error, vp, 0); /* No need to fsync--not critical */
6815 #ifndef AFS_PTHREAD_ENV
6817 #endif /* !AFS_PTHREAD_ENV */
6819 nUpdatedVolumes -= gap;
6821 #endif /* !AFS_DEMAND_ATTACH_FS */
6824 /***************************************************/
6825 /* Volume LRU routines */
6826 /***************************************************/
6831 * with demand attach fs, we attempt to soft detach(1)
6832 * volumes which have not been accessed in a long time
6833 * in order to speed up fileserver shutdown
6835 * (1) by soft detach we mean a process very similar
6836 * to VOffline, except the final state of the
6837 * Volume will be VOL_STATE_PREATTACHED, instead
6838 * of the usual VOL_STATE_UNATTACHED
6840 #ifdef AFS_DEMAND_ATTACH_FS
6842 /* implementation is reminiscent of a generational GC
6844 * queue 0 is newly attached volumes. this queue is
6845 * sorted by attach timestamp
6847 * queue 1 is volumes that have been around a bit
6848 * longer than queue 0. this queue is sorted by
6851 * queue 2 is volumes tha have been around the longest.
6852 * this queue is unsorted
6854 * queue 3 is volumes that have been marked as
6855 * candidates for soft detachment. this queue is
6858 #define VLRU_GENERATIONS 3 /**< number of generations in VLRU */
6859 #define VLRU_QUEUES 5 /**< total number of VLRU queues */
6862 * definition of a VLRU queue.
6865 volatile struct rx_queue q;
6872 * main VLRU data structure.
6875 struct VLRU_q q[VLRU_QUEUES]; /**< VLRU queues */
6878 /** time interval (in seconds) between promotion passes for
6879 * each young generation queue. */
6880 afs_uint32 promotion_interval[VLRU_GENERATIONS-1];
6882 /** time interval (in seconds) between soft detach candidate
6883 * scans for each generation queue.
6885 * scan_interval[VLRU_QUEUE_CANDIDATE] defines how frequently
6886 * we perform a soft detach pass. */
6887 afs_uint32 scan_interval[VLRU_GENERATIONS+1];
6889 /* scheduler state */
6890 int next_idx; /**< next queue to receive attention */
6891 afs_uint32 last_promotion[VLRU_GENERATIONS-1]; /**< timestamp of last promotion scan */
6892 afs_uint32 last_scan[VLRU_GENERATIONS+1]; /**< timestamp of last detach scan */
6894 int scanner_state; /**< state of scanner thread */
6895 pthread_cond_t cv; /**< state transition CV */
6898 /** global VLRU state */
6899 static struct VLRU volume_LRU;
6902 * defined states for VLRU scanner thread.
6905 VLRU_SCANNER_STATE_OFFLINE = 0, /**< vlru scanner thread is offline */
6906 VLRU_SCANNER_STATE_ONLINE = 1, /**< vlru scanner thread is online */
6907 VLRU_SCANNER_STATE_SHUTTING_DOWN = 2, /**< vlru scanner thread is shutting down */
6908 VLRU_SCANNER_STATE_PAUSING = 3, /**< vlru scanner thread is getting ready to pause */
6909 VLRU_SCANNER_STATE_PAUSED = 4 /**< vlru scanner thread is paused */
6910 } vlru_thread_state_t;
6912 /* vlru disk data header stuff */
6913 #define VLRU_DISK_MAGIC 0x7a8b9cad /**< vlru disk entry magic number */
6914 #define VLRU_DISK_VERSION 1 /**< vlru disk entry version number */
6916 /** vlru default expiration time (for eventual fs state serialization of vlru data) */
6917 #define VLRU_DUMP_EXPIRATION_TIME (60*60*24*7) /* expire vlru data after 1 week */
6920 /** minimum volume inactivity (in seconds) before a volume becomes eligible for
6921 * soft detachment. */
6922 static afs_uint32 VLRU_offline_thresh = VLRU_DEFAULT_OFFLINE_THRESH;
6924 /** time interval (in seconds) between VLRU scanner thread soft detach passes. */
6925 static afs_uint32 VLRU_offline_interval = VLRU_DEFAULT_OFFLINE_INTERVAL;
6927 /** maximum number of volumes to soft detach in a VLRU soft detach pass. */
6928 static afs_uint32 VLRU_offline_max = VLRU_DEFAULT_OFFLINE_MAX;
6930 /** VLRU control flag. non-zero value implies VLRU subsystem is activated. */
6931 static afs_uint32 VLRU_enabled = 1;
6933 /* queue synchronization routines */
6934 static void VLRU_BeginExclusive_r(struct VLRU_q * q);
6935 static void VLRU_EndExclusive_r(struct VLRU_q * q);
6936 static void VLRU_Wait_r(struct VLRU_q * q);
6939 * set VLRU subsystem tunable parameters.
6941 * @param[in] option tunable option to modify
6942 * @param[in] val new value for tunable parameter
6944 * @pre @c VInitVolumePackage2 has not yet been called.
6946 * @post tunable parameter is modified
6950 * @note valid option parameters are:
6951 * @arg @c VLRU_SET_THRESH
6952 * set the period of inactivity after which
6953 * volumes are eligible for soft detachment
6954 * @arg @c VLRU_SET_INTERVAL
6955 * set the time interval between calls
6956 * to the volume LRU "garbage collector"
6957 * @arg @c VLRU_SET_MAX
6958 * set the max number of volumes to deallocate
6962 VLRU_SetOptions(int option, afs_uint32 val)
6964 if (option == VLRU_SET_THRESH) {
6965 VLRU_offline_thresh = val;
6966 } else if (option == VLRU_SET_INTERVAL) {
6967 VLRU_offline_interval = val;
6968 } else if (option == VLRU_SET_MAX) {
6969 VLRU_offline_max = val;
6970 } else if (option == VLRU_SET_ENABLED) {
6973 VLRU_ComputeConstants();
6977 * compute VLRU internal timing parameters.
6979 * @post VLRU scanner thread internal timing parameters are computed
6981 * @note computes internal timing parameters based upon user-modifiable
6982 * tunable parameters.
6986 * @internal volume package internal use only.
6989 VLRU_ComputeConstants(void)
6991 afs_uint32 factor = VLRU_offline_thresh / VLRU_offline_interval;
6993 /* compute the candidate scan interval */
6994 volume_LRU.scan_interval[VLRU_QUEUE_CANDIDATE] = VLRU_offline_interval;
6996 /* compute the promotion intervals */
6997 volume_LRU.promotion_interval[VLRU_QUEUE_NEW] = VLRU_offline_thresh * 2;
6998 volume_LRU.promotion_interval[VLRU_QUEUE_MID] = VLRU_offline_thresh * 4;
7001 /* compute the gen 0 scan interval */
7002 volume_LRU.scan_interval[VLRU_QUEUE_NEW] = VLRU_offline_thresh / 8;
7004 /* compute the gen 0 scan interval */
7005 volume_LRU.scan_interval[VLRU_QUEUE_NEW] = VLRU_offline_interval * 2;
7010 * initialize VLRU subsystem.
7012 * @pre this function has not yet been called
7014 * @post VLRU subsystem is initialized and VLRU scanner thread is starting
7018 * @internal volume package internal use only.
7024 pthread_attr_t attrs;
7027 if (!VLRU_enabled) {
7028 Log("VLRU: disabled\n");
7032 /* initialize each of the VLRU queues */
7033 for (i = 0; i < VLRU_QUEUES; i++) {
7034 queue_Init(&volume_LRU.q[i]);
7035 volume_LRU.q[i].len = 0;
7036 volume_LRU.q[i].busy = 0;
7037 CV_INIT(&volume_LRU.q[i].cv, "vol lru", CV_DEFAULT, 0);
7040 /* setup the timing constants */
7041 VLRU_ComputeConstants();
7043 /* XXX put inside LogLevel check? */
7044 Log("VLRU: starting scanner with the following configuration parameters:\n");
7045 Log("VLRU: offlining volumes after minimum of %d seconds of inactivity\n", VLRU_offline_thresh);
7046 Log("VLRU: running VLRU soft detach pass every %d seconds\n", VLRU_offline_interval);
7047 Log("VLRU: taking up to %d volumes offline per pass\n", VLRU_offline_max);
7048 Log("VLRU: scanning generation 0 for inactive volumes every %d seconds\n", volume_LRU.scan_interval[0]);
7049 Log("VLRU: scanning for promotion/demotion between generations 0 and 1 every %d seconds\n", volume_LRU.promotion_interval[0]);
7050 Log("VLRU: scanning for promotion/demotion between generations 1 and 2 every %d seconds\n", volume_LRU.promotion_interval[1]);
7052 /* start up the VLRU scanner */
7053 volume_LRU.scanner_state = VLRU_SCANNER_STATE_OFFLINE;
7054 if (programType == fileServer) {
7055 CV_INIT(&volume_LRU.cv, "vol lru", CV_DEFAULT, 0);
7056 osi_Assert(pthread_attr_init(&attrs) == 0);
7057 osi_Assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
7058 osi_Assert(pthread_create(&tid, &attrs, &VLRU_ScannerThread, NULL) == 0);
7063 * initialize the VLRU-related fields of a newly allocated volume object.
7065 * @param[in] vp pointer to volume object
7068 * @arg @c VOL_LOCK is held.
7069 * @arg volume object is not on a VLRU queue.
7071 * @post VLRU fields are initialized to indicate that volume object is not
7072 * currently registered with the VLRU subsystem
7076 * @internal volume package interal use only.
7079 VLRU_Init_Node_r(Volume * vp)
7084 osi_Assert(queue_IsNotOnQueue(&vp->vlru));
7085 vp->vlru.idx = VLRU_QUEUE_INVALID;
7089 * add a volume object to a VLRU queue.
7091 * @param[in] vp pointer to volume object
7094 * @arg @c VOL_LOCK is held.
7095 * @arg caller MUST hold a lightweight ref on @p vp.
7096 * @arg caller MUST NOT hold exclusive ownership of the VLRU queue.
7098 * @post the volume object is added to the appropriate VLRU queue
7100 * @note if @c vp->vlru.idx contains the index of a valid VLRU queue,
7101 * then the volume is added to that queue. Otherwise, the value
7102 * @c VLRU_QUEUE_NEW is stored into @c vp->vlru.idx and the
7103 * volume is added to the NEW generation queue.
7105 * @note @c VOL_LOCK may be dropped internally
7107 * @note Volume state is temporarily set to @c VOL_STATE_VLRU_ADD
7108 * during the add operation, and is restored to the previous
7109 * state prior to return.
7113 * @internal volume package internal use only.
7116 VLRU_Add_r(Volume * vp)
7119 VolState state_save;
7124 if (queue_IsOnQueue(&vp->vlru))
7127 state_save = VChangeState_r(vp, VOL_STATE_VLRU_ADD);
7130 if ((idx < 0) || (idx >= VLRU_QUEUE_INVALID)) {
7131 idx = VLRU_QUEUE_NEW;
7134 VLRU_Wait_r(&volume_LRU.q[idx]);
7136 /* repeat check since VLRU_Wait_r may have dropped
7138 if (queue_IsNotOnQueue(&vp->vlru)) {
7140 queue_Prepend(&volume_LRU.q[idx], &vp->vlru);
7141 volume_LRU.q[idx].len++;
7142 V_attachFlags(vp) |= VOL_ON_VLRU;
7143 vp->stats.last_promote = FT_ApproxTime();
7146 VChangeState_r(vp, state_save);
7150 * delete a volume object from a VLRU queue.
7152 * @param[in] vp pointer to volume object
7155 * @arg @c VOL_LOCK is held.
7156 * @arg caller MUST hold a lightweight ref on @p vp.
7157 * @arg caller MUST NOT hold exclusive ownership of the VLRU queue.
7159 * @post volume object is removed from the VLRU queue
7161 * @note @c VOL_LOCK may be dropped internally
7165 * @todo We should probably set volume state to something exlcusive
7166 * (as @c VLRU_Add_r does) prior to dropping @c VOL_LOCK.
7168 * @internal volume package internal use only.
7171 VLRU_Delete_r(Volume * vp)
7178 if (queue_IsNotOnQueue(&vp->vlru))
7184 if (idx == VLRU_QUEUE_INVALID)
7186 VLRU_Wait_r(&volume_LRU.q[idx]);
7187 } while (idx != vp->vlru.idx);
7189 /* now remove from the VLRU and update
7190 * the appropriate counter */
7191 queue_Remove(&vp->vlru);
7192 volume_LRU.q[idx].len--;
7193 vp->vlru.idx = VLRU_QUEUE_INVALID;
7194 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
7198 * tell the VLRU subsystem that a volume was just accessed.
7200 * @param[in] vp pointer to volume object
7203 * @arg @c VOL_LOCK is held
7204 * @arg caller MUST hold a lightweight ref on @p vp
7205 * @arg caller MUST NOT hold exclusive ownership of any VLRU queue
7207 * @post volume VLRU access statistics are updated. If the volume was on
7208 * the VLRU soft detach candidate queue, it is moved to the NEW
7211 * @note @c VOL_LOCK may be dropped internally
7215 * @internal volume package internal use only.
7218 VLRU_UpdateAccess_r(Volume * vp)
7220 Volume * rvp = NULL;
7225 if (queue_IsNotOnQueue(&vp->vlru))
7228 osi_Assert(V_attachFlags(vp) & VOL_ON_VLRU);
7230 /* update the access timestamp */
7231 vp->stats.last_get = FT_ApproxTime();
7234 * if the volume is on the soft detach candidate
7235 * list, we need to safely move it back to a
7236 * regular generation. this has to be done
7237 * carefully so we don't race against the scanner
7241 /* if this volume is on the soft detach candidate queue,
7242 * then grab exclusive access to the necessary queues */
7243 if (vp->vlru.idx == VLRU_QUEUE_CANDIDATE) {
7245 VCreateReservation_r(rvp);
7247 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
7248 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
7249 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7250 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7253 /* make sure multiple threads don't race to update */
7254 if (vp->vlru.idx == VLRU_QUEUE_CANDIDATE) {
7255 VLRU_SwitchQueues(vp, VLRU_QUEUE_NEW, 1);
7259 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7260 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
7261 VCancelReservation_r(rvp);
7266 * switch a volume between two VLRU queues.
7268 * @param[in] vp pointer to volume object
7269 * @param[in] new_idx index of VLRU queue onto which the volume will be moved
7270 * @param[in] append controls whether the volume will be appended or
7271 * prepended to the queue. A nonzero value means it will
7272 * be appended; zero means it will be prepended.
7274 * @pre The new (and old, if applicable) queue(s) must either be owned
7275 * exclusively by the calling thread for asynchronous manipulation,
7276 * or the queue(s) must be quiescent and VOL_LOCK must be held.
7277 * Please see VLRU_BeginExclusive_r, VLRU_EndExclusive_r and VLRU_Wait_r
7278 * for further details of the queue asynchronous processing mechanism.
7280 * @post If the volume object was already on a VLRU queue, it is
7281 * removed from the queue. Depending on the value of the append
7282 * parameter, the volume object is either appended or prepended
7283 * to the VLRU queue referenced by the new_idx parameter.
7287 * @see VLRU_BeginExclusive_r
7288 * @see VLRU_EndExclusive_r
7291 * @internal volume package internal use only.
7294 VLRU_SwitchQueues(Volume * vp, int new_idx, int append)
7296 if (queue_IsNotOnQueue(&vp->vlru))
7299 queue_Remove(&vp->vlru);
7300 volume_LRU.q[vp->vlru.idx].len--;
7302 /* put the volume back on the correct generational queue */
7304 queue_Append(&volume_LRU.q[new_idx], &vp->vlru);
7306 queue_Prepend(&volume_LRU.q[new_idx], &vp->vlru);
7309 volume_LRU.q[new_idx].len++;
7310 vp->vlru.idx = new_idx;
7314 * VLRU background thread.
7316 * The VLRU Scanner Thread is responsible for periodically scanning through
7317 * each VLRU queue looking for volumes which should be moved to another
7318 * queue, or soft detached.
7320 * @param[in] args unused thread arguments parameter
7322 * @return unused thread return value
7323 * @retval NULL always
7325 * @internal volume package internal use only.
7328 VLRU_ScannerThread(void * args)
7330 afs_uint32 now, min_delay, delay;
7331 int i, min_idx, min_op, overdue, state;
7333 /* set t=0 for promotion cycle to be
7334 * fileserver startup */
7335 now = FT_ApproxTime();
7336 for (i=0; i < VLRU_GENERATIONS-1; i++) {
7337 volume_LRU.last_promotion[i] = now;
7340 /* don't start the scanner until VLRU_offline_thresh
7341 * plus a small delay for VInitVolumePackage2 to finish
7344 sleep(VLRU_offline_thresh + 60);
7346 /* set t=0 for scan cycle to be now */
7347 now = FT_ApproxTime();
7348 for (i=0; i < VLRU_GENERATIONS+1; i++) {
7349 volume_LRU.last_scan[i] = now;
7353 if (volume_LRU.scanner_state == VLRU_SCANNER_STATE_OFFLINE) {
7354 volume_LRU.scanner_state = VLRU_SCANNER_STATE_ONLINE;
7357 while ((state = volume_LRU.scanner_state) != VLRU_SCANNER_STATE_SHUTTING_DOWN) {
7358 /* check to see if we've been asked to pause */
7359 if (volume_LRU.scanner_state == VLRU_SCANNER_STATE_PAUSING) {
7360 volume_LRU.scanner_state = VLRU_SCANNER_STATE_PAUSED;
7361 CV_BROADCAST(&volume_LRU.cv);
7363 VOL_CV_WAIT(&volume_LRU.cv);
7364 } while (volume_LRU.scanner_state == VLRU_SCANNER_STATE_PAUSED);
7367 /* scheduling can happen outside the glock */
7370 /* figure out what is next on the schedule */
7372 /* figure out a potential schedule for the new generation first */
7374 min_delay = volume_LRU.scan_interval[0] + volume_LRU.last_scan[0] - now;
7377 if (min_delay > volume_LRU.scan_interval[0]) {
7378 /* unsigned overflow -- we're overdue to run this scan */
7383 /* if we're not overdue for gen 0, figure out schedule for candidate gen */
7385 i = VLRU_QUEUE_CANDIDATE;
7386 delay = volume_LRU.scan_interval[i] + volume_LRU.last_scan[i] - now;
7387 if (delay < min_delay) {
7391 if (delay > volume_LRU.scan_interval[i]) {
7392 /* unsigned overflow -- we're overdue to run this scan */
7399 /* if we're still not overdue for something, figure out schedules for promotions */
7400 for (i=0; !overdue && i < VLRU_GENERATIONS-1; i++) {
7401 delay = volume_LRU.promotion_interval[i] + volume_LRU.last_promotion[i] - now;
7402 if (delay < min_delay) {
7407 if (delay > volume_LRU.promotion_interval[i]) {
7408 /* unsigned overflow -- we're overdue to run this promotion */
7417 /* sleep as needed */
7422 /* do whatever is next */
7425 VLRU_Promote_r(min_idx);
7426 VLRU_Demote_r(min_idx+1);
7428 VLRU_Scan_r(min_idx);
7430 now = FT_ApproxTime();
7433 Log("VLRU scanner asked to go offline (scanner_state=%d)\n", state);
7435 /* signal that scanner is down */
7436 volume_LRU.scanner_state = VLRU_SCANNER_STATE_OFFLINE;
7437 CV_BROADCAST(&volume_LRU.cv);
7443 * promote volumes from one VLRU generation to the next.
7445 * This routine scans a VLRU generation looking for volumes which are
7446 * eligible to be promoted to the next generation. All volumes which
7447 * meet the eligibility requirement are promoted.
7449 * Promotion eligibility is based upon meeting both of the following
7452 * @arg The volume has been accessed since the last promotion:
7453 * @c (vp->stats.last_get >= vp->stats.last_promote)
7454 * @arg The last promotion occurred at least
7455 * @c volume_LRU.promotion_interval[idx] seconds ago
7457 * As a performance optimization, promotions are "globbed". In other
7458 * words, we promote arbitrarily large contiguous sublists of elements
7461 * @param[in] idx VLRU queue index to scan
7465 * @internal VLRU internal use only.
7468 VLRU_Promote_r(int idx)
7470 int len, chaining, promote;
7471 afs_uint32 now, thresh;
7472 struct rx_queue *qp, *nqp;
7473 Volume * vp, *start = NULL, *end = NULL;
7475 /* get exclusive access to two chains, and drop the glock */
7476 VLRU_Wait_r(&volume_LRU.q[idx]);
7477 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
7478 VLRU_Wait_r(&volume_LRU.q[idx+1]);
7479 VLRU_BeginExclusive_r(&volume_LRU.q[idx+1]);
7482 thresh = volume_LRU.promotion_interval[idx];
7483 now = FT_ApproxTime();
7486 for (queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
7487 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7488 promote = (((vp->stats.last_promote + thresh) <= now) &&
7489 (vp->stats.last_get >= vp->stats.last_promote));
7497 /* promote and prepend chain */
7498 queue_MoveChainAfter(&volume_LRU.q[idx+1], &start->vlru, &end->vlru);
7512 /* promote and prepend */
7513 queue_MoveChainAfter(&volume_LRU.q[idx+1], &start->vlru, &end->vlru);
7517 volume_LRU.q[idx].len -= len;
7518 volume_LRU.q[idx+1].len += len;
7521 /* release exclusive access to the two chains */
7523 volume_LRU.last_promotion[idx] = now;
7524 VLRU_EndExclusive_r(&volume_LRU.q[idx+1]);
7525 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
7528 /* run the demotions */
7530 VLRU_Demote_r(int idx)
7533 int len, chaining, demote;
7534 afs_uint32 now, thresh;
7535 struct rx_queue *qp, *nqp;
7536 Volume * vp, *start = NULL, *end = NULL;
7537 Volume ** salv_flag_vec = NULL;
7538 int salv_vec_offset = 0;
7540 osi_Assert(idx == VLRU_QUEUE_MID || idx == VLRU_QUEUE_OLD);
7542 /* get exclusive access to two chains, and drop the glock */
7543 VLRU_Wait_r(&volume_LRU.q[idx-1]);
7544 VLRU_BeginExclusive_r(&volume_LRU.q[idx-1]);
7545 VLRU_Wait_r(&volume_LRU.q[idx]);
7546 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
7549 /* no big deal if this allocation fails */
7550 if (volume_LRU.q[idx].len) {
7551 salv_flag_vec = (Volume **) malloc(volume_LRU.q[idx].len * sizeof(Volume *));
7554 now = FT_ApproxTime();
7555 thresh = volume_LRU.promotion_interval[idx-1];
7558 for (queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
7559 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7560 demote = (((vp->stats.last_promote + thresh) <= now) &&
7561 (vp->stats.last_get < (now - thresh)));
7563 /* we now do volume update list DONT_SALVAGE flag setting during
7564 * demotion passes */
7565 if (salv_flag_vec &&
7566 !(V_attachFlags(vp) & VOL_HDR_DONTSALV) &&
7568 (vp->updateTime < (now - SALVAGE_INTERVAL)) &&
7569 (V_attachState(vp) == VOL_STATE_ATTACHED)) {
7570 salv_flag_vec[salv_vec_offset++] = vp;
7571 VCreateReservation_r(vp);
7580 /* demote and append chain */
7581 queue_MoveChainBefore(&volume_LRU.q[idx-1], &start->vlru, &end->vlru);
7595 queue_MoveChainBefore(&volume_LRU.q[idx-1], &start->vlru, &end->vlru);
7599 volume_LRU.q[idx].len -= len;
7600 volume_LRU.q[idx-1].len += len;
7603 /* release exclusive access to the two chains */
7605 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
7606 VLRU_EndExclusive_r(&volume_LRU.q[idx-1]);
7608 /* now go back and set the DONT_SALVAGE flags as appropriate */
7609 if (salv_flag_vec) {
7611 for (i = 0; i < salv_vec_offset; i++) {
7612 vp = salv_flag_vec[i];
7613 if (!(V_attachFlags(vp) & VOL_HDR_DONTSALV) &&
7614 (vp->updateTime < (now - SALVAGE_INTERVAL)) &&
7615 (V_attachState(vp) == VOL_STATE_ATTACHED)) {
7618 V_attachFlags(vp) |= VOL_HDR_DONTSALV;
7619 V_dontSalvage(vp) = DONT_SALVAGE;
7620 VUpdateVolume_r(&ec, vp, 0);
7624 VCancelReservation_r(vp);
7626 free(salv_flag_vec);
7630 /* run a pass of the VLRU GC scanner */
7632 VLRU_Scan_r(int idx)
7634 afs_uint32 now, thresh;
7635 struct rx_queue *qp, *nqp;
7639 osi_Assert(idx == VLRU_QUEUE_NEW || idx == VLRU_QUEUE_CANDIDATE);
7641 /* gain exclusive access to the idx VLRU */
7642 VLRU_Wait_r(&volume_LRU.q[idx]);
7643 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
7645 if (idx != VLRU_QUEUE_CANDIDATE) {
7646 /* gain exclusive access to the candidate VLRU */
7647 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7648 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7651 now = FT_ApproxTime();
7652 thresh = now - VLRU_offline_thresh;
7654 /* perform candidate selection and soft detaching */
7655 if (idx == VLRU_QUEUE_CANDIDATE) {
7656 /* soft detach some volumes from the candidate pool */
7660 for (i=0,queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
7661 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7662 if (i >= VLRU_offline_max) {
7665 /* check timestamp to see if it's a candidate for soft detaching */
7666 if (vp->stats.last_get <= thresh) {
7668 if (VCheckSoftDetach(vp, thresh))
7674 /* scan for volumes to become soft detach candidates */
7675 for (i=1,queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue),i++) {
7676 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7678 /* check timestamp to see if it's a candidate for soft detaching */
7679 if (vp->stats.last_get <= thresh) {
7680 VCheckSoftDetachCandidate(vp, thresh);
7683 if (!(i&0x7f)) { /* lock coarsening optimization */
7691 /* relinquish exclusive access to the VLRU chains */
7695 volume_LRU.last_scan[idx] = now;
7696 if (idx != VLRU_QUEUE_CANDIDATE) {
7697 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7699 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
7702 /* check whether volume is safe to soft detach
7703 * caller MUST NOT hold a ref count on vp */
7705 VCheckSoftDetach(Volume * vp, afs_uint32 thresh)
7709 if (vp->nUsers || vp->nWaiters)
7712 if (vp->stats.last_get <= thresh) {
7713 ret = VSoftDetachVolume_r(vp, thresh);
7719 /* check whether volume should be made a
7720 * soft detach candidate */
7722 VCheckSoftDetachCandidate(Volume * vp, afs_uint32 thresh)
7725 if (vp->nUsers || vp->nWaiters)
7730 osi_Assert(idx == VLRU_QUEUE_NEW);
7732 if (vp->stats.last_get <= thresh) {
7733 /* move to candidate pool */
7734 queue_Remove(&vp->vlru);
7735 volume_LRU.q[VLRU_QUEUE_NEW].len--;
7736 queue_Prepend(&volume_LRU.q[VLRU_QUEUE_CANDIDATE], &vp->vlru);
7737 vp->vlru.idx = VLRU_QUEUE_CANDIDATE;
7738 volume_LRU.q[VLRU_QUEUE_CANDIDATE].len++;
7746 /* begin exclusive access on VLRU */
7748 VLRU_BeginExclusive_r(struct VLRU_q * q)
7750 osi_Assert(q->busy == 0);
7754 /* end exclusive access on VLRU */
7756 VLRU_EndExclusive_r(struct VLRU_q * q)
7758 osi_Assert(q->busy);
7760 CV_BROADCAST(&q->cv);
7763 /* wait for another thread to end exclusive access on VLRU */
7765 VLRU_Wait_r(struct VLRU_q * q)
7768 VOL_CV_WAIT(&q->cv);
7773 * volume soft detach
7775 * caller MUST NOT hold a ref count on vp */
7777 VSoftDetachVolume_r(Volume * vp, afs_uint32 thresh)
7782 osi_Assert(vp->vlru.idx == VLRU_QUEUE_CANDIDATE);
7784 ts_save = vp->stats.last_get;
7785 if (ts_save > thresh)
7788 if (vp->nUsers || vp->nWaiters)
7791 if (VIsExclusiveState(V_attachState(vp))) {
7795 switch (V_attachState(vp)) {
7796 case VOL_STATE_UNATTACHED:
7797 case VOL_STATE_PREATTACHED:
7798 case VOL_STATE_ERROR:
7799 case VOL_STATE_GOING_OFFLINE:
7800 case VOL_STATE_SHUTTING_DOWN:
7801 case VOL_STATE_SALVAGING:
7802 case VOL_STATE_DELETED:
7803 volume_LRU.q[vp->vlru.idx].len--;
7805 /* create and cancel a reservation to
7806 * give the volume an opportunity to
7808 VCreateReservation_r(vp);
7809 queue_Remove(&vp->vlru);
7810 vp->vlru.idx = VLRU_QUEUE_INVALID;
7811 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
7812 VCancelReservation_r(vp);
7818 /* hold the volume and take it offline.
7819 * no need for reservations, as VHold_r
7820 * takes care of that internally. */
7821 if (VHold_r(vp) == 0) {
7822 /* vhold drops the glock, so now we should
7823 * check to make sure we aren't racing against
7824 * other threads. if we are racing, offlining vp
7825 * would be wasteful, and block the scanner for a while
7829 (vp->shuttingDown) ||
7830 (vp->goingOffline) ||
7831 (vp->stats.last_get != ts_save)) {
7832 /* looks like we're racing someone else. bail */
7836 /* pull it off the VLRU */
7837 osi_Assert(vp->vlru.idx == VLRU_QUEUE_CANDIDATE);
7838 volume_LRU.q[VLRU_QUEUE_CANDIDATE].len--;
7839 queue_Remove(&vp->vlru);
7840 vp->vlru.idx = VLRU_QUEUE_INVALID;
7841 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
7843 /* take if offline */
7844 VOffline_r(vp, "volume has been soft detached");
7846 /* invalidate the volume header cache */
7847 FreeVolumeHeader(vp);
7850 IncUInt64(&VStats.soft_detaches);
7851 vp->stats.soft_detaches++;
7853 /* put in pre-attached state so demand
7854 * attacher can work on it */
7855 VChangeState_r(vp, VOL_STATE_PREATTACHED);
7861 #endif /* AFS_DEMAND_ATTACH_FS */
7864 /***************************************************/
7865 /* Volume Header Cache routines */
7866 /***************************************************/
7869 * volume header cache.
7871 struct volume_hdr_LRU_t volume_hdr_LRU;
7874 * initialize the volume header cache.
7876 * @param[in] howMany number of header cache entries to preallocate
7878 * @pre VOL_LOCK held. Function has never been called before.
7880 * @post howMany cache entries are allocated, initialized, and added
7881 * to the LRU list. Header cache statistics are initialized.
7883 * @note only applicable to fileServer program type. Should only be
7884 * called once during volume package initialization.
7886 * @internal volume package internal use only.
7889 VInitVolumeHeaderCache(afs_uint32 howMany)
7891 struct volHeader *hp;
7892 if (programType != fileServer)
7894 queue_Init(&volume_hdr_LRU);
7895 volume_hdr_LRU.stats.free = 0;
7896 volume_hdr_LRU.stats.used = howMany;
7897 volume_hdr_LRU.stats.attached = 0;
7898 hp = (struct volHeader *)(calloc(howMany, sizeof(struct volHeader)));
7899 osi_Assert(hp != NULL);
7902 /* We are using ReleaseVolumeHeader to initialize the values on the header list
7903 * to ensure they have the right values
7905 ReleaseVolumeHeader(hp++);
7908 /* get a volume header off of the volume header LRU.
7910 * @return volume header
7911 * @retval NULL no usable volume header is available on the LRU
7913 * @pre VOL_LOCK held
7915 * @post for DAFS, if the returned header is associated with a volume, that
7916 * volume is NOT in an exclusive state
7918 * @internal volume package internal use only.
7920 #ifdef AFS_DEMAND_ATTACH_FS
7921 static struct volHeader*
7922 GetVolHeaderFromLRU(void)
7924 struct volHeader *hd = NULL, *qh, *nqh;
7925 /* Usually, a volume in an exclusive state will not have its header on
7926 * the LRU. However, it is possible for this to occur when a salvage
7927 * request is received over FSSYNC, and possibly in other corner cases.
7928 * So just skip over headers whose volumes are in an exclusive state. We
7929 * could VWaitExclusiveState_r instead, but not waiting is faster and
7931 for (queue_Scan(&volume_hdr_LRU, qh, nqh, volHeader)) {
7932 if (!hd->back || !VIsExclusiveState(V_attachState(hd->back))) {
7940 #else /* AFS_DEMAND_ATTACH_FS */
7941 static struct volHeader*
7942 GetVolHeaderFromLRU(void)
7944 struct volHeader *hd = NULL;
7945 if (queue_IsNotEmpty(&volume_hdr_LRU)) {
7946 hd = queue_First(&volume_hdr_LRU, volHeader);
7951 #endif /* !AFS_DEMAND_ATTACH_FS */
7954 * get a volume header and attach it to the volume object.
7956 * @param[in] vp pointer to volume object
7958 * @return cache entry status
7959 * @retval 0 volume header was newly attached; cache data is invalid
7960 * @retval 1 volume header was previously attached; cache data is valid
7962 * @pre VOL_LOCK held. For DAFS, lightweight ref must be held on volume object.
7964 * @post volume header attached to volume object. if necessary, header cache
7965 * entry on LRU is synchronized to disk. Header is removed from LRU list.
7967 * @note VOL_LOCK may be dropped
7969 * @warning this interface does not load header data from disk. it merely
7970 * attaches a header object to the volume object, and may sync the old
7971 * header cache data out to disk in the process.
7973 * @internal volume package internal use only.
7976 GetVolumeHeader(Volume * vp)
7979 struct volHeader *hd;
7981 static int everLogged = 0;
7983 #ifdef AFS_DEMAND_ATTACH_FS
7984 VolState vp_save = 0, back_save = 0;
7986 /* XXX debug 9/19/05 we've apparently got
7987 * a ref counting bug somewhere that's
7988 * breaking the nUsers == 0 => header on LRU
7990 if (vp->header && queue_IsNotOnQueue(vp->header)) {
7991 Log("nUsers == 0, but header not on LRU\n");
7996 old = (vp->header != NULL); /* old == volume already has a header */
7998 if (programType != fileServer) {
7999 /* for volume utilities, we allocate volHeaders as needed */
8001 hd = (struct volHeader *)calloc(1, sizeof(*vp->header));
8002 osi_Assert(hd != NULL);
8005 #ifdef AFS_DEMAND_ATTACH_FS
8006 V_attachFlags(vp) |= VOL_HDR_ATTACHED;
8010 /* for the fileserver, we keep a volume header cache */
8012 /* the header we previously dropped in the lru is
8013 * still available. pull it off the lru and return */
8016 osi_Assert(hd->back == vp);
8017 #ifdef AFS_DEMAND_ATTACH_FS
8018 V_attachFlags(vp) &= ~(VOL_HDR_IN_LRU);
8021 hd = GetVolHeaderFromLRU();
8023 /* LRU is empty, so allocate a new volHeader
8024 * this is probably indicative of a leak, so let the user know */
8025 hd = (struct volHeader *)calloc(1, sizeof(struct volHeader));
8026 osi_Assert(hd != NULL);
8028 Log("****Allocated more volume headers, probably leak****\n");
8031 volume_hdr_LRU.stats.free++;
8034 /* this header used to belong to someone else.
8035 * we'll need to check if the header needs to
8036 * be sync'd out to disk */
8038 #ifdef AFS_DEMAND_ATTACH_FS
8039 /* GetVolHeaderFromLRU had better not give us back a header
8040 * with a volume in exclusive state... */
8041 osi_Assert(!VIsExclusiveState(V_attachState(hd->back)));
8044 if (hd->diskstuff.inUse) {
8045 /* volume was in use, so we'll need to sync
8046 * its header to disk */
8048 #ifdef AFS_DEMAND_ATTACH_FS
8049 back_save = VChangeState_r(hd->back, VOL_STATE_UPDATING);
8050 vp_save = VChangeState_r(vp, VOL_STATE_HDR_ATTACHING);
8051 VCreateReservation_r(hd->back);
8055 WriteVolumeHeader_r(&error, hd->back);
8056 /* Ignore errors; catch them later */
8058 #ifdef AFS_DEMAND_ATTACH_FS
8063 hd->back->header = NULL;
8064 #ifdef AFS_DEMAND_ATTACH_FS
8065 V_attachFlags(hd->back) &= ~(VOL_HDR_ATTACHED | VOL_HDR_LOADED | VOL_HDR_IN_LRU);
8067 if (hd->diskstuff.inUse) {
8068 VChangeState_r(hd->back, back_save);
8069 VCancelReservation_r(hd->back);
8070 VChangeState_r(vp, vp_save);
8074 volume_hdr_LRU.stats.attached++;
8078 #ifdef AFS_DEMAND_ATTACH_FS
8079 V_attachFlags(vp) |= VOL_HDR_ATTACHED;
8082 volume_hdr_LRU.stats.free--;
8083 volume_hdr_LRU.stats.used++;
8085 IncUInt64(&VStats.hdr_gets);
8086 #ifdef AFS_DEMAND_ATTACH_FS
8087 IncUInt64(&vp->stats.hdr_gets);
8088 vp->stats.last_hdr_get = FT_ApproxTime();
8095 * make sure volume header is attached and contains valid cache data.
8097 * @param[out] ec outbound error code
8098 * @param[in] vp pointer to volume object
8100 * @pre VOL_LOCK held. For DAFS, lightweight ref held on vp.
8102 * @post header cache entry attached, and loaded with valid data, or
8103 * *ec is nonzero, and the header is released back into the LRU.
8105 * @internal volume package internal use only.
8108 LoadVolumeHeader(Error * ec, Volume * vp)
8110 #ifdef AFS_DEMAND_ATTACH_FS
8111 VolState state_save;
8115 if (vp->nUsers == 0 && !GetVolumeHeader(vp)) {
8116 IncUInt64(&VStats.hdr_loads);
8117 state_save = VChangeState_r(vp, VOL_STATE_HDR_LOADING);
8120 ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
8121 sizeof(V_disk(vp)), VOLUMEINFOMAGIC,
8123 IncUInt64(&vp->stats.hdr_loads);
8124 now = FT_ApproxTime();
8128 V_attachFlags(vp) |= VOL_HDR_LOADED;
8129 vp->stats.last_hdr_load = now;
8131 VChangeState_r(vp, state_save);
8133 #else /* AFS_DEMAND_ATTACH_FS */
8135 if (vp->nUsers == 0 && !GetVolumeHeader(vp)) {
8136 IncUInt64(&VStats.hdr_loads);
8138 ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
8139 sizeof(V_disk(vp)), VOLUMEINFOMAGIC,
8142 #endif /* AFS_DEMAND_ATTACH_FS */
8144 /* maintain (nUsers==0) => header in LRU invariant */
8145 FreeVolumeHeader(vp);
8150 * release a header cache entry back into the LRU list.
8152 * @param[in] hd pointer to volume header cache object
8154 * @pre VOL_LOCK held.
8156 * @post header cache object appended onto end of LRU list.
8158 * @note only applicable to fileServer program type.
8160 * @note used to place a header cache entry back into the
8161 * LRU pool without invalidating it as a cache entry.
8163 * @internal volume package internal use only.
8166 ReleaseVolumeHeader(struct volHeader *hd)
8168 if (programType != fileServer)
8170 if (!hd || queue_IsOnQueue(hd)) /* no header, or header already released */
8172 queue_Append(&volume_hdr_LRU, hd);
8173 #ifdef AFS_DEMAND_ATTACH_FS
8175 V_attachFlags(hd->back) |= VOL_HDR_IN_LRU;
8178 volume_hdr_LRU.stats.free++;
8179 volume_hdr_LRU.stats.used--;
8183 * free/invalidate a volume header cache entry.
8185 * @param[in] vp pointer to volume object
8187 * @pre VOL_LOCK is held.
8189 * @post For fileserver, header cache entry is returned to LRU, and it is
8190 * invalidated as a cache entry. For volume utilities, the header
8191 * cache entry is freed.
8193 * @note For fileserver, this should be utilized instead of ReleaseVolumeHeader
8194 * whenever it is necessary to invalidate the header cache entry.
8196 * @see ReleaseVolumeHeader
8198 * @internal volume package internal use only.
8201 FreeVolumeHeader(Volume * vp)
8203 struct volHeader *hd = vp->header;
8206 if (programType == fileServer) {
8207 ReleaseVolumeHeader(hd);
8212 #ifdef AFS_DEMAND_ATTACH_FS
8213 V_attachFlags(vp) &= ~(VOL_HDR_ATTACHED | VOL_HDR_IN_LRU | VOL_HDR_LOADED);
8215 volume_hdr_LRU.stats.attached--;
8220 /***************************************************/
8221 /* Volume Hash Table routines */
8222 /***************************************************/
8225 * set size of volume object hash table.
8227 * @param[in] logsize log(2) of desired hash table size
8229 * @return operation status
8231 * @retval -1 failure
8233 * @pre MUST be called prior to VInitVolumePackage2
8235 * @post Volume Hash Table will have 2^logsize buckets
8238 VSetVolHashSize(int logsize)
8240 /* 64 to 268435456 hash buckets seems like a reasonable range */
8241 if ((logsize < 6 ) || (logsize > 28)) {
8246 VolumeHashTable.Size = 1 << logsize;
8247 VolumeHashTable.Mask = VolumeHashTable.Size - 1;
8249 /* we can't yet support runtime modification of this
8250 * parameter. we'll need a configuration rwlock to
8251 * make runtime modification feasible.... */
8258 * initialize dynamic data structures for volume hash table.
8260 * @post hash table is allocated, and fields are initialized.
8262 * @internal volume package internal use only.
8265 VInitVolumeHash(void)
8269 VolumeHashTable.Table = (VolumeHashChainHead *) calloc(VolumeHashTable.Size,
8270 sizeof(VolumeHashChainHead));
8271 osi_Assert(VolumeHashTable.Table != NULL);
8273 for (i=0; i < VolumeHashTable.Size; i++) {
8274 queue_Init(&VolumeHashTable.Table[i]);
8275 #ifdef AFS_DEMAND_ATTACH_FS
8276 CV_INIT(&VolumeHashTable.Table[i].chain_busy_cv, "vhash busy", CV_DEFAULT, 0);
8277 #endif /* AFS_DEMAND_ATTACH_FS */
8282 * add a volume object to the hash table.
8284 * @param[in] vp pointer to volume object
8285 * @param[in] hashid hash of volume id
8287 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
8290 * @post volume is added to hash chain.
8292 * @internal volume package internal use only.
8294 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
8295 * asynchronous hash chain reordering to finish.
8298 AddVolumeToHashTable(Volume * vp, int hashid)
8300 VolumeHashChainHead * head;
8302 if (queue_IsOnQueue(vp))
8305 head = &VolumeHashTable.Table[VOLUME_HASH(hashid)];
8307 #ifdef AFS_DEMAND_ATTACH_FS
8308 /* wait for the hash chain to become available */
8311 V_attachFlags(vp) |= VOL_IN_HASH;
8312 vp->chainCacheCheck = ++head->cacheCheck;
8313 #endif /* AFS_DEMAND_ATTACH_FS */
8316 vp->hashid = hashid;
8317 queue_Append(head, vp);
8318 vp->vnodeHashOffset = VolumeHashOffset_r();
8322 * delete a volume object from the hash table.
8324 * @param[in] vp pointer to volume object
8326 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
8329 * @post volume is removed from hash chain.
8331 * @internal volume package internal use only.
8333 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
8334 * asynchronous hash chain reordering to finish.
8337 DeleteVolumeFromHashTable(Volume * vp)
8339 VolumeHashChainHead * head;
8341 if (!queue_IsOnQueue(vp))
8344 head = &VolumeHashTable.Table[VOLUME_HASH(vp->hashid)];
8346 #ifdef AFS_DEMAND_ATTACH_FS
8347 /* wait for the hash chain to become available */
8350 V_attachFlags(vp) &= ~(VOL_IN_HASH);
8352 #endif /* AFS_DEMAND_ATTACH_FS */
8356 /* do NOT reset hashid to zero, as the online
8357 * salvager package may need to know the volume id
8358 * after the volume is removed from the hash */
8362 * lookup a volume object in the hash table given a volume id.
8364 * @param[out] ec error code return
8365 * @param[in] volumeId volume id
8366 * @param[in] hint volume object which we believe could be the correct
8369 * @return volume object pointer
8370 * @retval NULL no such volume id is registered with the hash table.
8372 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
8375 * @post volume object with the given id is returned. volume object and
8376 * hash chain access statistics are updated. hash chain may have
8379 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
8380 * asynchronous hash chain reordering operation to finish, or
8381 * in order for us to perform an asynchronous chain reordering.
8383 * @note Hash chain reorderings occur when the access count for the
8384 * volume object being looked up exceeds the sum of the previous
8385 * node's (the node ahead of it in the hash chain linked list)
8386 * access count plus the constant VOLUME_HASH_REORDER_THRESHOLD.
8388 * @note For DAFS, the hint parameter allows us to short-circuit if the
8389 * cacheCheck fields match between the hash chain head and the
8390 * hint volume object.
8393 VLookupVolume_r(Error * ec, VolId volumeId, Volume * hint)
8397 #ifdef AFS_DEMAND_ATTACH_FS
8400 VolumeHashChainHead * head;
8403 head = &VolumeHashTable.Table[VOLUME_HASH(volumeId)];
8405 #ifdef AFS_DEMAND_ATTACH_FS
8406 /* wait for the hash chain to become available */
8409 /* check to see if we can short circuit without walking the hash chain */
8410 if (hint && (hint->chainCacheCheck == head->cacheCheck)) {
8411 IncUInt64(&hint->stats.hash_short_circuits);
8414 #endif /* AFS_DEMAND_ATTACH_FS */
8416 /* someday we need to either do per-chain locks, RWlocks,
8417 * or both for volhash access.
8418 * (and move to a data structure with better cache locality) */
8420 /* search the chain for this volume id */
8421 for(queue_Scan(head, vp, np, Volume)) {
8423 if ((vp->hashid == volumeId)) {
8428 if (queue_IsEnd(head, vp)) {
8432 #ifdef AFS_DEMAND_ATTACH_FS
8433 /* update hash chain statistics */
8436 FillInt64(lks, 0, looks);
8437 AddUInt64(head->looks, lks, &head->looks);
8438 AddUInt64(VStats.hash_looks, lks, &VStats.hash_looks);
8439 IncUInt64(&head->gets);
8444 IncUInt64(&vp->stats.hash_lookups);
8446 /* for demand attach fileserver, we permit occasional hash chain reordering
8447 * so that frequently looked up volumes move towards the head of the chain */
8448 pp = queue_Prev(vp, Volume);
8449 if (!queue_IsEnd(head, pp)) {
8450 FillInt64(thresh, 0, VOLUME_HASH_REORDER_THRESHOLD);
8451 AddUInt64(thresh, pp->stats.hash_lookups, &thresh);
8452 if (GEInt64(vp->stats.hash_lookups, thresh)) {
8453 VReorderHash_r(head, pp, vp);
8457 /* update the short-circuit cache check */
8458 vp->chainCacheCheck = head->cacheCheck;
8460 #endif /* AFS_DEMAND_ATTACH_FS */
8465 #ifdef AFS_DEMAND_ATTACH_FS
8466 /* perform volume hash chain reordering.
8468 * advance a subchain beginning at vp ahead of
8469 * the adjacent subchain ending at pp */
8471 VReorderHash_r(VolumeHashChainHead * head, Volume * pp, Volume * vp)
8473 Volume *tp, *np, *lp;
8474 afs_uint64 move_thresh;
8476 /* this should never be called if the chain is already busy, so
8477 * no need to wait for other exclusive chain ops to finish */
8479 /* this is a rather heavy set of operations,
8480 * so let's set the chain busy flag and drop
8482 VHashBeginExclusive_r(head);
8485 /* scan forward in the chain from vp looking for the last element
8486 * in the chain we want to advance */
8487 FillInt64(move_thresh, 0, VOLUME_HASH_REORDER_CHAIN_THRESH);
8488 AddUInt64(move_thresh, pp->stats.hash_lookups, &move_thresh);
8489 for(queue_ScanFrom(head, vp, tp, np, Volume)) {
8490 if (LTInt64(tp->stats.hash_lookups, move_thresh)) {
8494 lp = queue_Prev(tp, Volume);
8496 /* scan backwards from pp to determine where to splice and
8497 * insert the subchain we're advancing */
8498 for(queue_ScanBackwardsFrom(head, pp, tp, np, Volume)) {
8499 if (GTInt64(tp->stats.hash_lookups, move_thresh)) {
8503 tp = queue_Next(tp, Volume);
8505 /* rebalance chain(vp,...,lp) ahead of chain(tp,...,pp) */
8506 queue_MoveChainBefore(tp,vp,lp);
8509 IncUInt64(&VStats.hash_reorders);
8511 IncUInt64(&head->reorders);
8513 /* wake up any threads waiting for the hash chain */
8514 VHashEndExclusive_r(head);
8518 /* demand-attach fs volume hash
8519 * asynchronous exclusive operations */
8522 * begin an asynchronous exclusive operation on a volume hash chain.
8524 * @param[in] head pointer to volume hash chain head object
8526 * @pre VOL_LOCK held. hash chain is quiescent.
8528 * @post hash chain marked busy.
8530 * @note this interface is used in conjunction with VHashEndExclusive_r and
8531 * VHashWait_r to perform asynchronous (wrt VOL_LOCK) operations on a
8532 * volume hash chain. Its main use case is hash chain reordering, which
8533 * has the potential to be a highly latent operation.
8535 * @see VHashEndExclusive_r
8540 * @internal volume package internal use only.
8543 VHashBeginExclusive_r(VolumeHashChainHead * head)
8545 osi_Assert(head->busy == 0);
8550 * relinquish exclusive ownership of a volume hash chain.
8552 * @param[in] head pointer to volume hash chain head object
8554 * @pre VOL_LOCK held. thread owns the hash chain exclusively.
8556 * @post hash chain is marked quiescent. threads awaiting use of
8557 * chain are awakened.
8559 * @see VHashBeginExclusive_r
8564 * @internal volume package internal use only.
8567 VHashEndExclusive_r(VolumeHashChainHead * head)
8569 osi_Assert(head->busy);
8571 CV_BROADCAST(&head->chain_busy_cv);
8575 * wait for all asynchronous operations on a hash chain to complete.
8577 * @param[in] head pointer to volume hash chain head object
8579 * @pre VOL_LOCK held.
8581 * @post hash chain object is quiescent.
8583 * @see VHashBeginExclusive_r
8584 * @see VHashEndExclusive_r
8588 * @note This interface should be called before any attempt to
8589 * traverse the hash chain. It is permissible for a thread
8590 * to gain exclusive access to the chain, and then perform
8591 * latent operations on the chain asynchronously wrt the
8594 * @warning if waiting is necessary, VOL_LOCK is dropped
8596 * @internal volume package internal use only.
8599 VHashWait_r(VolumeHashChainHead * head)
8601 while (head->busy) {
8602 VOL_CV_WAIT(&head->chain_busy_cv);
8605 #endif /* AFS_DEMAND_ATTACH_FS */
8608 /***************************************************/
8609 /* Volume by Partition List routines */
8610 /***************************************************/
8613 * demand attach fileserver adds a
8614 * linked list of volumes to each
8615 * partition object, thus allowing
8616 * for quick enumeration of all
8617 * volumes on a partition
8620 #ifdef AFS_DEMAND_ATTACH_FS
8622 * add a volume to its disk partition VByPList.
8624 * @param[in] vp pointer to volume object
8626 * @pre either the disk partition VByPList is owned exclusively
8627 * by the calling thread, or the list is quiescent and
8630 * @post volume is added to disk partition VByPList
8634 * @warning it is the caller's responsibility to ensure list
8637 * @see VVByPListWait_r
8638 * @see VVByPListBeginExclusive_r
8639 * @see VVByPListEndExclusive_r
8641 * @internal volume package internal use only.
8644 AddVolumeToVByPList_r(Volume * vp)
8646 if (queue_IsNotOnQueue(&vp->vol_list)) {
8647 queue_Append(&vp->partition->vol_list, &vp->vol_list);
8648 V_attachFlags(vp) |= VOL_ON_VBYP_LIST;
8649 vp->partition->vol_list.len++;
8654 * delete a volume from its disk partition VByPList.
8656 * @param[in] vp pointer to volume object
8658 * @pre either the disk partition VByPList is owned exclusively
8659 * by the calling thread, or the list is quiescent and
8662 * @post volume is removed from the disk partition VByPList
8666 * @warning it is the caller's responsibility to ensure list
8669 * @see VVByPListWait_r
8670 * @see VVByPListBeginExclusive_r
8671 * @see VVByPListEndExclusive_r
8673 * @internal volume package internal use only.
8676 DeleteVolumeFromVByPList_r(Volume * vp)
8678 if (queue_IsOnQueue(&vp->vol_list)) {
8679 queue_Remove(&vp->vol_list);
8680 V_attachFlags(vp) &= ~(VOL_ON_VBYP_LIST);
8681 vp->partition->vol_list.len--;
8686 * begin an asynchronous exclusive operation on a VByPList.
8688 * @param[in] dp pointer to disk partition object
8690 * @pre VOL_LOCK held. VByPList is quiescent.
8692 * @post VByPList marked busy.
8694 * @note this interface is used in conjunction with VVByPListEndExclusive_r and
8695 * VVByPListWait_r to perform asynchronous (wrt VOL_LOCK) operations on a
8698 * @see VVByPListEndExclusive_r
8699 * @see VVByPListWait_r
8703 * @internal volume package internal use only.
8705 /* take exclusive control over the list */
8707 VVByPListBeginExclusive_r(struct DiskPartition64 * dp)
8709 osi_Assert(dp->vol_list.busy == 0);
8710 dp->vol_list.busy = 1;
8714 * relinquish exclusive ownership of a VByPList.
8716 * @param[in] dp pointer to disk partition object
8718 * @pre VOL_LOCK held. thread owns the VByPList exclusively.
8720 * @post VByPList is marked quiescent. threads awaiting use of
8721 * the list are awakened.
8723 * @see VVByPListBeginExclusive_r
8724 * @see VVByPListWait_r
8728 * @internal volume package internal use only.
8731 VVByPListEndExclusive_r(struct DiskPartition64 * dp)
8733 osi_Assert(dp->vol_list.busy);
8734 dp->vol_list.busy = 0;
8735 CV_BROADCAST(&dp->vol_list.cv);
8739 * wait for all asynchronous operations on a VByPList to complete.
8741 * @param[in] dp pointer to disk partition object
8743 * @pre VOL_LOCK is held.
8745 * @post disk partition's VByP list is quiescent
8749 * @note This interface should be called before any attempt to
8750 * traverse the VByPList. It is permissible for a thread
8751 * to gain exclusive access to the list, and then perform
8752 * latent operations on the list asynchronously wrt the
8755 * @warning if waiting is necessary, VOL_LOCK is dropped
8757 * @see VVByPListEndExclusive_r
8758 * @see VVByPListBeginExclusive_r
8760 * @internal volume package internal use only.
8763 VVByPListWait_r(struct DiskPartition64 * dp)
8765 while (dp->vol_list.busy) {
8766 VOL_CV_WAIT(&dp->vol_list.cv);
8769 #endif /* AFS_DEMAND_ATTACH_FS */
8771 /***************************************************/
8772 /* Volume Cache Statistics routines */
8773 /***************************************************/
8776 VPrintCacheStats_r(void)
8778 struct VnodeClassInfo *vcp;
8779 vcp = &VnodeClassInfo[vLarge];
8780 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);
8781 vcp = &VnodeClassInfo[vSmall];
8782 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);
8783 Log("Volume header cache, %d entries, %"AFS_INT64_FMT" gets, "
8784 "%"AFS_INT64_FMT" replacements\n",
8785 VStats.hdr_cache_size, VStats.hdr_gets, VStats.hdr_loads);
8789 VPrintCacheStats(void)
8792 VPrintCacheStats_r();
8796 #ifdef AFS_DEMAND_ATTACH_FS
8798 UInt64ToDouble(afs_uint64 * x)
8800 static double c32 = 4.0 * 1.073741824 * 1000000000.0;
8802 SplitInt64(*x, h, l);
8803 return (((double)h) * c32) + ((double) l);
8807 DoubleToPrintable(double x, char * buf, int len)
8809 static double billion = 1000000000.0;
8812 y[0] = (afs_uint32) (x / (billion * billion));
8813 y[1] = (afs_uint32) ((x - (((double)y[0]) * billion * billion)) / billion);
8814 y[2] = (afs_uint32) (x - ((((double)y[0]) * billion * billion) + (((double)y[1]) * billion)));
8817 snprintf(buf, len, "%d%09d%09d", y[0], y[1], y[2]);
8819 snprintf(buf, len, "%d%09d", y[1], y[2]);
8821 snprintf(buf, len, "%d", y[2]);
8827 struct VLRUExtStatsEntry {
8831 struct VLRUExtStats {
8837 } queue_info[VLRU_QUEUE_INVALID];
8838 struct VLRUExtStatsEntry * vec;
8842 * add a 256-entry fudge factor onto the vector in case state changes
8843 * out from under us.
8845 #define VLRU_EXT_STATS_VEC_LEN_FUDGE 256
8848 * collect extended statistics for the VLRU subsystem.
8850 * @param[out] stats pointer to stats structure to be populated
8851 * @param[in] nvols number of volumes currently known to exist
8853 * @pre VOL_LOCK held
8855 * @post stats->vec allocated and populated
8857 * @return operation status
8862 VVLRUExtStats_r(struct VLRUExtStats * stats, afs_uint32 nvols)
8864 afs_uint32 cur, idx, len;
8865 struct rx_queue * qp, * nqp;
8867 struct VLRUExtStatsEntry * vec;
8869 len = nvols + VLRU_EXT_STATS_VEC_LEN_FUDGE;
8870 vec = stats->vec = calloc(len,
8871 sizeof(struct VLRUExtStatsEntry));
8877 for (idx = VLRU_QUEUE_NEW; idx < VLRU_QUEUE_INVALID; idx++) {
8878 VLRU_Wait_r(&volume_LRU.q[idx]);
8879 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
8882 stats->queue_info[idx].start = cur;
8884 for (queue_Scan(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
8886 /* out of space in vec */
8889 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
8890 vec[cur].volid = vp->hashid;
8894 stats->queue_info[idx].len = cur - stats->queue_info[idx].start;
8897 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
8905 #define ENUMTOSTRING(en) #en
8906 #define ENUMCASE(en) \
8907 case en: return ENUMTOSTRING(en)
8910 vlru_idx_to_string(int idx)
8913 ENUMCASE(VLRU_QUEUE_NEW);
8914 ENUMCASE(VLRU_QUEUE_MID);
8915 ENUMCASE(VLRU_QUEUE_OLD);
8916 ENUMCASE(VLRU_QUEUE_CANDIDATE);
8917 ENUMCASE(VLRU_QUEUE_HELD);
8918 ENUMCASE(VLRU_QUEUE_INVALID);
8920 return "**UNKNOWN**";
8925 VPrintExtendedCacheStats_r(int flags)
8928 afs_uint32 vol_sum = 0;
8935 struct stats looks, gets, reorders, len;
8936 struct stats ch_looks, ch_gets, ch_reorders;
8938 VolumeHashChainHead *head;
8940 struct VLRUExtStats vlru_stats;
8942 /* zero out stats */
8943 memset(&looks, 0, sizeof(struct stats));
8944 memset(&gets, 0, sizeof(struct stats));
8945 memset(&reorders, 0, sizeof(struct stats));
8946 memset(&len, 0, sizeof(struct stats));
8947 memset(&ch_looks, 0, sizeof(struct stats));
8948 memset(&ch_gets, 0, sizeof(struct stats));
8949 memset(&ch_reorders, 0, sizeof(struct stats));
8951 for (i = 0; i < VolumeHashTable.Size; i++) {
8952 head = &VolumeHashTable.Table[i];
8955 VHashBeginExclusive_r(head);
8958 ch_looks.sum = UInt64ToDouble(&head->looks);
8959 ch_gets.sum = UInt64ToDouble(&head->gets);
8960 ch_reorders.sum = UInt64ToDouble(&head->reorders);
8962 /* update global statistics */
8964 looks.sum += ch_looks.sum;
8965 gets.sum += ch_gets.sum;
8966 reorders.sum += ch_reorders.sum;
8967 len.sum += (double)head->len;
8968 vol_sum += head->len;
8971 len.min = (double) head->len;
8972 len.max = (double) head->len;
8973 looks.min = ch_looks.sum;
8974 looks.max = ch_looks.sum;
8975 gets.min = ch_gets.sum;
8976 gets.max = ch_gets.sum;
8977 reorders.min = ch_reorders.sum;
8978 reorders.max = ch_reorders.sum;
8980 if (((double)head->len) < len.min)
8981 len.min = (double) head->len;
8982 if (((double)head->len) > len.max)
8983 len.max = (double) head->len;
8984 if (ch_looks.sum < looks.min)
8985 looks.min = ch_looks.sum;
8986 else if (ch_looks.sum > looks.max)
8987 looks.max = ch_looks.sum;
8988 if (ch_gets.sum < gets.min)
8989 gets.min = ch_gets.sum;
8990 else if (ch_gets.sum > gets.max)
8991 gets.max = ch_gets.sum;
8992 if (ch_reorders.sum < reorders.min)
8993 reorders.min = ch_reorders.sum;
8994 else if (ch_reorders.sum > reorders.max)
8995 reorders.max = ch_reorders.sum;
8999 if ((flags & VOL_STATS_PER_CHAIN2) && queue_IsNotEmpty(head)) {
9000 /* compute detailed per-chain stats */
9001 struct stats hdr_loads, hdr_gets;
9002 double v_looks, v_loads, v_gets;
9004 /* initialize stats with data from first element in chain */
9005 vp = queue_First(head, Volume);
9006 v_looks = UInt64ToDouble(&vp->stats.hash_lookups);
9007 v_loads = UInt64ToDouble(&vp->stats.hdr_loads);
9008 v_gets = UInt64ToDouble(&vp->stats.hdr_gets);
9009 ch_gets.min = ch_gets.max = v_looks;
9010 hdr_loads.min = hdr_loads.max = v_loads;
9011 hdr_gets.min = hdr_gets.max = v_gets;
9012 hdr_loads.sum = hdr_gets.sum = 0;
9014 vp = queue_Next(vp, Volume);
9016 /* pull in stats from remaining elements in chain */
9017 for (queue_ScanFrom(head, vp, vp, np, Volume)) {
9018 v_looks = UInt64ToDouble(&vp->stats.hash_lookups);
9019 v_loads = UInt64ToDouble(&vp->stats.hdr_loads);
9020 v_gets = UInt64ToDouble(&vp->stats.hdr_gets);
9022 hdr_loads.sum += v_loads;
9023 hdr_gets.sum += v_gets;
9025 if (v_looks < ch_gets.min)
9026 ch_gets.min = v_looks;
9027 else if (v_looks > ch_gets.max)
9028 ch_gets.max = v_looks;
9030 if (v_loads < hdr_loads.min)
9031 hdr_loads.min = v_loads;
9032 else if (v_loads > hdr_loads.max)
9033 hdr_loads.max = v_loads;
9035 if (v_gets < hdr_gets.min)
9036 hdr_gets.min = v_gets;
9037 else if (v_gets > hdr_gets.max)
9038 hdr_gets.max = v_gets;
9041 /* compute per-chain averages */
9042 ch_gets.avg = ch_gets.sum / ((double)head->len);
9043 hdr_loads.avg = hdr_loads.sum / ((double)head->len);
9044 hdr_gets.avg = hdr_gets.sum / ((double)head->len);
9046 /* dump per-chain stats */
9047 Log("Volume hash chain %d : len=%d, looks=%s, reorders=%s\n",
9049 DoubleToPrintable(ch_looks.sum, pr_buf[0], sizeof(pr_buf[0])),
9050 DoubleToPrintable(ch_reorders.sum, pr_buf[1], sizeof(pr_buf[1])));
9051 Log("\tVolume gets : min=%s, max=%s, avg=%s, total=%s\n",
9052 DoubleToPrintable(ch_gets.min, pr_buf[0], sizeof(pr_buf[0])),
9053 DoubleToPrintable(ch_gets.max, pr_buf[1], sizeof(pr_buf[1])),
9054 DoubleToPrintable(ch_gets.avg, pr_buf[2], sizeof(pr_buf[2])),
9055 DoubleToPrintable(ch_gets.sum, pr_buf[3], sizeof(pr_buf[3])));
9056 Log("\tHDR gets : min=%s, max=%s, avg=%s, total=%s\n",
9057 DoubleToPrintable(hdr_gets.min, pr_buf[0], sizeof(pr_buf[0])),
9058 DoubleToPrintable(hdr_gets.max, pr_buf[1], sizeof(pr_buf[1])),
9059 DoubleToPrintable(hdr_gets.avg, pr_buf[2], sizeof(pr_buf[2])),
9060 DoubleToPrintable(hdr_gets.sum, pr_buf[3], sizeof(pr_buf[3])));
9061 Log("\tHDR loads : min=%s, max=%s, avg=%s, total=%s\n",
9062 DoubleToPrintable(hdr_loads.min, pr_buf[0], sizeof(pr_buf[0])),
9063 DoubleToPrintable(hdr_loads.max, pr_buf[1], sizeof(pr_buf[1])),
9064 DoubleToPrintable(hdr_loads.avg, pr_buf[2], sizeof(pr_buf[2])),
9065 DoubleToPrintable(hdr_loads.sum, pr_buf[3], sizeof(pr_buf[3])));
9066 } else if (flags & VOL_STATS_PER_CHAIN) {
9067 /* dump simple per-chain stats */
9068 Log("Volume hash chain %d : len=%d, looks=%s, gets=%s, reorders=%s\n",
9070 DoubleToPrintable(ch_looks.sum, pr_buf[0], sizeof(pr_buf[0])),
9071 DoubleToPrintable(ch_gets.sum, pr_buf[1], sizeof(pr_buf[1])),
9072 DoubleToPrintable(ch_reorders.sum, pr_buf[2], sizeof(pr_buf[2])));
9076 VHashEndExclusive_r(head);
9081 /* compute global averages */
9082 len.avg = len.sum / ((double)VolumeHashTable.Size);
9083 looks.avg = looks.sum / ((double)VolumeHashTable.Size);
9084 gets.avg = gets.sum / ((double)VolumeHashTable.Size);
9085 reorders.avg = reorders.sum / ((double)VolumeHashTable.Size);
9087 /* dump global stats */
9088 Log("Volume hash summary: %d buckets\n", VolumeHashTable.Size);
9089 Log(" chain length : min=%s, max=%s, avg=%s, total=%s\n",
9090 DoubleToPrintable(len.min, pr_buf[0], sizeof(pr_buf[0])),
9091 DoubleToPrintable(len.max, pr_buf[1], sizeof(pr_buf[1])),
9092 DoubleToPrintable(len.avg, pr_buf[2], sizeof(pr_buf[2])),
9093 DoubleToPrintable(len.sum, pr_buf[3], sizeof(pr_buf[3])));
9094 Log(" looks : min=%s, max=%s, avg=%s, total=%s\n",
9095 DoubleToPrintable(looks.min, pr_buf[0], sizeof(pr_buf[0])),
9096 DoubleToPrintable(looks.max, pr_buf[1], sizeof(pr_buf[1])),
9097 DoubleToPrintable(looks.avg, pr_buf[2], sizeof(pr_buf[2])),
9098 DoubleToPrintable(looks.sum, pr_buf[3], sizeof(pr_buf[3])));
9099 Log(" gets : min=%s, max=%s, avg=%s, total=%s\n",
9100 DoubleToPrintable(gets.min, pr_buf[0], sizeof(pr_buf[0])),
9101 DoubleToPrintable(gets.max, pr_buf[1], sizeof(pr_buf[1])),
9102 DoubleToPrintable(gets.avg, pr_buf[2], sizeof(pr_buf[2])),
9103 DoubleToPrintable(gets.sum, pr_buf[3], sizeof(pr_buf[3])));
9104 Log(" reorders : min=%s, max=%s, avg=%s, total=%s\n",
9105 DoubleToPrintable(reorders.min, pr_buf[0], sizeof(pr_buf[0])),
9106 DoubleToPrintable(reorders.max, pr_buf[1], sizeof(pr_buf[1])),
9107 DoubleToPrintable(reorders.avg, pr_buf[2], sizeof(pr_buf[2])),
9108 DoubleToPrintable(reorders.sum, pr_buf[3], sizeof(pr_buf[3])));
9110 /* print extended disk related statistics */
9112 struct DiskPartition64 * diskP;
9113 afs_uint32 vol_count[VOLMAXPARTS+1];
9114 byte part_exists[VOLMAXPARTS+1];
9118 memset(vol_count, 0, sizeof(vol_count));
9119 memset(part_exists, 0, sizeof(part_exists));
9123 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
9125 vol_count[id] = diskP->vol_list.len;
9126 part_exists[id] = 1;
9130 for (i = 0; i <= VOLMAXPARTS; i++) {
9131 if (part_exists[i]) {
9132 /* XXX while this is currently safe, it is a violation
9133 * of the VGetPartitionById_r interface contract. */
9134 diskP = VGetPartitionById_r(i, 0);
9136 Log("Partition %s has %d online volumes\n",
9137 VPartitionPath(diskP), diskP->vol_list.len);
9144 /* print extended VLRU statistics */
9145 if (VVLRUExtStats_r(&vlru_stats, vol_sum) == 0) {
9146 afs_uint32 idx, cur, lpos;
9151 Log("VLRU State Dump:\n\n");
9153 for (idx = VLRU_QUEUE_NEW; idx < VLRU_QUEUE_INVALID; idx++) {
9154 Log("\t%s:\n", vlru_idx_to_string(idx));
9157 for (cur = vlru_stats.queue_info[idx].start;
9158 cur < vlru_stats.queue_info[idx].len;
9160 line[lpos++] = vlru_stats.vec[cur].volid;
9162 Log("\t\t%u, %u, %u, %u, %u,\n",
9163 line[0], line[1], line[2], line[3], line[4]);
9172 Log("\t\t%u, %u, %u, %u, %u\n",
9173 line[0], line[1], line[2], line[3], line[4]);
9178 free(vlru_stats.vec);
9185 VPrintExtendedCacheStats(int flags)
9188 VPrintExtendedCacheStats_r(flags);
9191 #endif /* AFS_DEMAND_ATTACH_FS */
9194 VCanScheduleSalvage(void)
9196 return vol_opts.canScheduleSalvage;
9202 return vol_opts.canUseFSSYNC;
9206 VCanUseSALVSYNC(void)
9208 return vol_opts.canUseSALVSYNC;
9212 VCanUnsafeAttach(void)
9214 return vol_opts.unsafe_attach;