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>
29 #ifdef HAVE_SYS_FILE_H
34 #include <afs/afsint.h>
37 #if !defined(AFS_SGI_ENV)
40 #else /* AFS_OSF_ENV */
41 #ifdef AFS_VFSINCL_ENV
44 #include <sys/fs/ufs_fs.h>
46 #if defined(AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV)
47 #include <ufs/ufs/dinode.h>
48 #include <ufs/ffs/fs.h>
53 #else /* AFS_VFSINCL_ENV */
54 #if !defined(AFS_AIX_ENV) && !defined(AFS_LINUX20_ENV) && !defined(AFS_XBSD_ENV) && !defined(AFS_ARM_DARWIN_ENV)
57 #endif /* AFS_VFSINCL_ENV */
58 #endif /* AFS_OSF_ENV */
59 #endif /* AFS_SGI_ENV */
60 #endif /* !AFS_NT40_ENV */
68 #if defined(AFS_SUN_ENV) || defined(AFS_SUN5_ENV)
70 #include <sys/mnttab.h>
71 #include <sys/mntent.h>
77 #if defined(AFS_SGI_ENV)
80 #ifndef AFS_LINUX20_ENV
81 #include <fstab.h> /* Need to find in libc 5, present in libc 6 */
84 #endif /* AFS_SGI_ENV */
86 #endif /* AFS_HPUX_ENV */
90 #include <afs/errors.h>
93 #include <afs/afssyscalls.h>
95 #include <afs/afsutil.h>
96 #include "daemon_com.h"
101 #include "partition.h"
102 #include "volume_inline.h"
104 #include "afs/afs_assert.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, VPartitionPath(partp), 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, VPartitionPath(partp), 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 vp->specialStatus = VBUSY;
3043 #endif /* AFS_DEMAND_ATTACH_FS */
3046 * volume attachment helper function.
3048 * @param[out] ec error code
3049 * @param[in] volumeId volume ID of the attaching volume
3050 * @param[in] path full path to the volume header .vol file
3051 * @param[in] partp disk partition object for the attaching partition
3052 * @param[in] vp volume object; vp->hashid, vp->device, vp->partition,
3053 * vp->vnode_list, vp->rx_call_list, and V_attachCV (for
3054 * DAFS) should already be initialized
3055 * @param[in] isbusy 1 if vp->specialStatus should be set to VBUSY; that is,
3056 * if there is a volume operation running for this volume
3057 * that should set the volume to VBUSY during its run. 0
3058 * otherwise. (see VVolOpSetVBusy_r)
3059 * @param[in] mode attachment mode such as V_VOLUPD, V_DUMP, etc (see
3061 * @param[out] acheckedOut If we successfully checked-out the volume from
3062 * the fileserver (if we needed to), this is set
3063 * to 1, otherwise it is 0.
3065 * @return pointer to the semi-attached volume pointer
3066 * @retval NULL an error occurred (check value of *ec)
3067 * @retval vp volume successfully attaching
3069 * @pre no locks held
3071 * @post VOL_LOCK held
3074 attach2(Error * ec, VolId volumeId, char *path, struct DiskPartition64 *partp,
3075 Volume * vp, int isbusy, int mode, int *acheckedOut)
3077 /* have we read in the header successfully? */
3078 int read_header = 0;
3080 #ifdef AFS_DEMAND_ATTACH_FS
3081 /* should we FreeVolume(vp) instead of VCheckFree(vp) in the error
3085 /* in the case of an error, to what state should the volume be
3087 VolState error_state = VOL_STATE_ERROR;
3088 #endif /* AFS_DEMAND_ATTACH_FS */
3092 vp->vnodeIndex[vLarge].handle = NULL;
3093 vp->vnodeIndex[vSmall].handle = NULL;
3094 vp->diskDataHandle = NULL;
3095 vp->linkHandle = NULL;
3099 #ifdef AFS_DEMAND_ATTACH_FS
3100 attach_check_vop(ec, volumeId, partp, vp, acheckedOut);
3102 attach_volume_header(ec, vp, partp, mode, 0, acheckedOut);
3105 attach_volume_header(ec, vp, partp, mode, 0, acheckedOut);
3106 #endif /* !AFS_DEMAND_ATTACH_FS */
3108 if (*ec == VNOVOL) {
3109 /* if the volume doesn't exist, skip straight to 'error' so we don't
3110 * request a salvage */
3111 goto unlocked_error;
3117 vp->specialStatus = (byte) (isbusy ? VBUSY : 0);
3118 vp->shuttingDown = 0;
3119 vp->goingOffline = 0;
3121 #ifdef AFS_DEMAND_ATTACH_FS
3122 vp->stats.last_attach = FT_ApproxTime();
3123 vp->stats.attaches++;
3127 IncUInt64(&VStats.attaches);
3128 vp->cacheCheck = ++VolumeCacheCheck;
3129 /* just in case this ever rolls over */
3130 if (!vp->cacheCheck)
3131 vp->cacheCheck = ++VolumeCacheCheck;
3134 #ifdef AFS_DEMAND_ATTACH_FS
3135 V_attachFlags(vp) |= VOL_HDR_LOADED;
3136 vp->stats.last_hdr_load = vp->stats.last_attach;
3137 #endif /* AFS_DEMAND_ATTACH_FS */
3141 struct IndexFileHeader iHead;
3143 #if OPENAFS_VOL_STATS
3145 * We just read in the diskstuff part of the header. If the detailed
3146 * volume stats area has not yet been initialized, we should bzero the
3147 * area and mark it as initialized.
3149 if (!(V_stat_initialized(vp))) {
3150 memset((V_stat_area(vp)), 0, VOL_STATS_BYTES);
3151 V_stat_initialized(vp) = 1;
3153 #endif /* OPENAFS_VOL_STATS */
3155 (void)ReadHeader(ec, vp->vnodeIndex[vSmall].handle,
3156 (char *)&iHead, sizeof(iHead),
3157 SMALLINDEXMAGIC, SMALLINDEXVERSION);
3160 Log("VAttachVolume: Error reading smallVnode vol header %s; error=%u\n", path, *ec);
3165 struct IndexFileHeader iHead;
3167 (void)ReadHeader(ec, vp->vnodeIndex[vLarge].handle,
3168 (char *)&iHead, sizeof(iHead),
3169 LARGEINDEXMAGIC, LARGEINDEXVERSION);
3172 Log("VAttachVolume: Error reading largeVnode vol header %s; error=%u\n", path, *ec);
3176 #ifdef AFS_NAMEI_ENV
3178 struct versionStamp stamp;
3180 (void)ReadHeader(ec, V_linkHandle(vp), (char *)&stamp,
3181 sizeof(stamp), LINKTABLEMAGIC, LINKTABLEVERSION);
3184 Log("VAttachVolume: Error reading namei vol header %s; error=%u\n", path, *ec);
3187 #endif /* AFS_NAMEI_ENV */
3189 #if defined(AFS_DEMAND_ATTACH_FS)
3190 if (*ec && ((*ec != VOFFLINE) || (V_attachState(vp) != VOL_STATE_UNATTACHED))) {
3192 if (!VCanScheduleSalvage()) {
3193 Log("VAttachVolume: Error attaching volume %s; volume needs salvage; error=%u\n", path, *ec);
3195 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER |
3196 VOL_SALVAGE_NO_OFFLINE);
3201 /* volume operation in progress */
3202 goto unlocked_error;
3204 #else /* AFS_DEMAND_ATTACH_FS */
3206 Log("VAttachVolume: Error attaching volume %s; volume needs salvage; error=%u\n", path, *ec);
3207 goto unlocked_error;
3209 #endif /* AFS_DEMAND_ATTACH_FS */
3211 if (V_needsSalvaged(vp)) {
3212 if (vp->specialStatus)
3213 vp->specialStatus = 0;
3215 #if defined(AFS_DEMAND_ATTACH_FS)
3216 if (!VCanScheduleSalvage()) {
3217 Log("VAttachVolume: volume salvage flag is ON for %s; volume needs salvage\n", path);
3219 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_INVALIDATE_HEADER |
3220 VOL_SALVAGE_NO_OFFLINE);
3223 #else /* AFS_DEMAND_ATTACH_FS */
3225 #endif /* AFS_DEMAND_ATTACH_FS */
3231 vp->nextVnodeUnique = V_uniquifier(vp);
3233 if (VShouldCheckInUse(mode) && V_inUse(vp) && VolumeWriteable(vp)) {
3234 if (!V_needsSalvaged(vp)) {
3235 V_needsSalvaged(vp) = 1;
3236 VUpdateVolume_r(ec, vp, 0);
3238 #if defined(AFS_DEMAND_ATTACH_FS)
3239 if (!VCanScheduleSalvage()) {
3240 Log("VAttachVolume: volume %s needs to be salvaged; not attached.\n", path);
3242 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_INVALIDATE_HEADER |
3243 VOL_SALVAGE_NO_OFFLINE);
3246 #else /* AFS_DEMAND_ATTACH_FS */
3247 Log("VAttachVolume: volume %s needs to be salvaged; not attached.\n", path);
3249 #endif /* AFS_DEMAND_ATTACH_FS */
3254 if (programType == fileServer && V_destroyMe(vp) == DESTROY_ME) {
3255 /* Only check destroyMe if we are the fileserver, since the
3256 * volserver et al sometimes need to work with volumes with
3257 * destroyMe set. Examples are 'temporary' volumes the
3258 * volserver creates, and when we create a volume (destroyMe
3259 * is set on creation; sometimes a separate volserver
3260 * transaction is created to clear destroyMe).
3263 #if defined(AFS_DEMAND_ATTACH_FS)
3264 /* schedule a salvage so the volume goes away on disk */
3265 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER |
3266 VOL_SALVAGE_NO_OFFLINE);
3267 VChangeState_r(vp, VOL_STATE_ERROR);
3270 #endif /* AFS_DEMAND_ATTACH_FS */
3271 Log("VAttachVolume: volume %s is junk; it should be destroyed at next salvage\n", path);
3276 vp->vnodeIndex[vSmall].bitmap = vp->vnodeIndex[vLarge].bitmap = NULL;
3277 #ifndef BITMAP_LATER
3278 if (programType == fileServer && VolumeWriteable(vp)) {
3280 for (i = 0; i < nVNODECLASSES; i++) {
3281 VGetBitmap_r(ec, vp, i);
3283 #ifdef AFS_DEMAND_ATTACH_FS
3284 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER |
3285 VOL_SALVAGE_NO_OFFLINE);
3287 #endif /* AFS_DEMAND_ATTACH_FS */
3288 Log("VAttachVolume: error getting bitmap for volume (%s)\n",
3294 #endif /* BITMAP_LATER */
3296 if (VInit >= 2 && V_needsCallback(vp)) {
3297 if (V_BreakVolumeCallbacks) {
3298 Log("VAttachVolume: Volume %lu was changed externally; breaking callbacks\n",
3299 afs_printable_uint32_lu(V_id(vp)));
3300 V_needsCallback(vp) = 0;
3302 (*V_BreakVolumeCallbacks) (V_id(vp));
3305 VUpdateVolume_r(ec, vp, 0);
3307 #ifdef FSSYNC_BUILD_CLIENT
3308 else if (VCanUseFSSYNC()) {
3309 afs_int32 fsync_code;
3311 V_needsCallback(vp) = 0;
3313 fsync_code = FSYNC_VolOp(V_id(vp), NULL, FSYNC_VOL_BREAKCBKS, FSYNC_WHATEVER, NULL);
3317 V_needsCallback(vp) = 1;
3318 Log("Error trying to tell the fileserver to break callbacks for "
3319 "changed volume %lu; error code %ld\n",
3320 afs_printable_uint32_lu(V_id(vp)),
3321 afs_printable_int32_ld(fsync_code));
3323 VUpdateVolume_r(ec, vp, 0);
3326 #endif /* FSSYNC_BUILD_CLIENT */
3329 Log("VAttachVolume: error %d clearing needsCallback on volume "
3330 "%lu; needs salvage\n", (int)*ec,
3331 afs_printable_uint32_lu(V_id(vp)));
3332 #ifdef AFS_DEMAND_ATTACH_FS
3333 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER |
3334 VOL_SALVAGE_NO_OFFLINE);
3336 #else /* !AFS_DEMAND_ATTACH_FS */
3338 #endif /* !AFS_DEMAND_ATTACh_FS */
3343 if (programType == fileServer) {
3344 if (vp->specialStatus)
3345 vp->specialStatus = 0;
3346 if (V_blessed(vp) && V_inService(vp) && !V_needsSalvaged(vp)) {
3347 V_inUse(vp) = fileServer;
3348 V_offlineMessage(vp)[0] = '\0';
3352 #ifdef AFS_DEMAND_ATTACH_FS
3353 /* Put the vol into PREATTACHED state, so if someone tries to
3354 * access it again, we try to attach, see that we're not blessed,
3355 * and give a VNOVOL error again. Putting it into UNATTACHED state
3356 * would result in a VOFFLINE error instead. */
3357 error_state = VOL_STATE_PREATTACHED;
3358 #endif /* AFS_DEMAND_ATTACH_FS */
3360 /* mimic e.g. GetVolume errors */
3361 if (!V_blessed(vp)) {
3362 Log("Volume %lu offline: not blessed\n", afs_printable_uint32_lu(V_id(vp)));
3363 FreeVolumeHeader(vp);
3364 } else if (!V_inService(vp)) {
3365 Log("Volume %lu offline: not in service\n", afs_printable_uint32_lu(V_id(vp)));
3366 FreeVolumeHeader(vp);
3368 Log("Volume %lu offline: needs salvage\n", afs_printable_uint32_lu(V_id(vp)));
3370 #ifdef AFS_DEMAND_ATTACH_FS
3371 error_state = VOL_STATE_ERROR;
3372 /* see if we can recover */
3373 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_INVALIDATE_HEADER);
3376 #ifdef AFS_DEMAND_ATTACH_FS
3382 #ifdef AFS_DEMAND_ATTACH_FS
3383 if ((mode != V_PEEK) && (mode != V_SECRETLY))
3384 V_inUse(vp) = programType;
3385 #endif /* AFS_DEMAND_ATTACH_FS */
3386 V_checkoutMode(vp) = mode;
3389 AddVolumeToHashTable(vp, V_id(vp));
3390 #ifdef AFS_DEMAND_ATTACH_FS
3391 if (VCanUnlockAttached() && (V_attachFlags(vp) & VOL_LOCKED)) {
3394 if ((programType != fileServer) ||
3395 (V_inUse(vp) == fileServer)) {
3396 AddVolumeToVByPList_r(vp);
3398 VChangeState_r(vp, VOL_STATE_ATTACHED);
3400 VChangeState_r(vp, VOL_STATE_UNATTACHED);
3409 #ifdef AFS_DEMAND_ATTACH_FS
3410 if (!VIsErrorState(V_attachState(vp))) {
3411 VChangeState_r(vp, error_state);
3413 #endif /* AFS_DEMAND_ATTACH_FS */
3416 VReleaseVolumeHandles_r(vp);
3419 #ifdef AFS_DEMAND_ATTACH_FS
3426 #else /* !AFS_DEMAND_ATTACH_FS */
3428 #endif /* !AFS_DEMAND_ATTACH_FS */
3432 /* Attach an existing volume.
3433 The volume also normally goes online at this time.
3434 An offline volume must be reattached to make it go online.
3438 VAttachVolume(Error * ec, VolumeId volumeId, int mode)
3442 retVal = VAttachVolume_r(ec, volumeId, mode);
3448 VAttachVolume_r(Error * ec, VolumeId volumeId, int mode)
3451 VGetVolumePath(ec, volumeId, &part, &name);
3455 vp = VGetVolume_r(&error, volumeId);
3457 osi_Assert(V_inUse(vp) == 0);
3458 VDetachVolume_r(ec, vp);
3462 return VAttachVolumeByName_r(ec, part, name, mode);
3465 /* Increment a reference count to a volume, sans context swaps. Requires
3466 * possibly reading the volume header in from the disk, since there's
3467 * an invariant in the volume package that nUsers>0 ==> vp->header is valid.
3469 * N.B. This call can fail if we can't read in the header!! In this case
3470 * we still guarantee we won't context swap, but the ref count won't be
3471 * incremented (otherwise we'd violate the invariant).
3473 /* NOTE: with the demand attach fileserver extensions, the global lock
3474 * is dropped within VHold */
3475 #ifdef AFS_DEMAND_ATTACH_FS
3477 VHold_r(Volume * vp)
3481 VCreateReservation_r(vp);
3482 VWaitExclusiveState_r(vp);
3484 LoadVolumeHeader(&error, vp);
3486 VCancelReservation_r(vp);
3490 VCancelReservation_r(vp);
3493 #else /* AFS_DEMAND_ATTACH_FS */
3495 VHold_r(Volume * vp)
3499 LoadVolumeHeader(&error, vp);
3505 #endif /* AFS_DEMAND_ATTACH_FS */
3507 /**** volume timeout-related stuff ****/
3509 #ifdef AFS_PTHREAD_ENV
3511 static struct timespec *shutdown_timeout;
3512 static pthread_once_t shutdown_timeout_once = PTHREAD_ONCE_INIT;
3515 VTimedOut(const struct timespec *ts)
3520 if (ts->tv_sec == 0) {
3521 /* short-circuit; this will have always timed out */
3525 code = gettimeofday(&tv, NULL);
3527 Log("Error %d from gettimeofday, assuming we have not timed out\n", errno);
3528 /* assume no timeout; failure mode is we just wait longer than normal
3529 * instead of returning errors when we shouldn't */
3533 if (tv.tv_sec < ts->tv_sec ||
3534 (tv.tv_sec == ts->tv_sec && tv.tv_usec*1000 < ts->tv_nsec)) {
3543 * Calculate an absolute timeout.
3545 * @param[out] ts A timeout that is "timeout" seconds from now, if we return
3546 * NULL, the memory is not touched
3547 * @param[in] timeout How long the timeout should be from now
3549 * @return timeout to use
3550 * @retval NULL no timeout; wait forever
3551 * @retval non-NULL the given value for "ts"
3555 static struct timespec *
3556 VCalcTimeout(struct timespec *ts, afs_int32 timeout)
3566 ts->tv_sec = ts->tv_nsec = 0;
3570 code = gettimeofday(&now, NULL);
3572 Log("Error %d from gettimeofday, falling back to 'forever' timeout\n", errno);
3576 ts->tv_sec = now.tv_sec + timeout;
3577 ts->tv_nsec = now.tv_usec * 1000;
3583 * Initialize the shutdown_timeout global.
3586 VShutdownTimeoutInit(void)
3588 struct timespec *ts;
3590 ts = malloc(sizeof(*ts));
3592 shutdown_timeout = VCalcTimeout(ts, vol_opts.offline_shutdown_timeout);
3594 if (!shutdown_timeout) {
3600 * Figure out the timeout that should be used for waiting for offline volumes.
3602 * @param[out] ats Storage space for a local timeout value if needed
3604 * @return The timeout value that should be used
3605 * @retval NULL No timeout; wait forever for offlining volumes
3606 * @retval non-NULL A pointer to the absolute time that should be used as
3607 * the deadline for waiting for offlining volumes.
3609 * @note If we return non-NULL, the pointer we return may or may not be the
3612 static const struct timespec *
3613 VOfflineTimeout(struct timespec *ats)
3615 if (vol_shutting_down) {
3616 osi_Assert(pthread_once(&shutdown_timeout_once, VShutdownTimeoutInit) == 0);
3617 return shutdown_timeout;
3619 return VCalcTimeout(ats, vol_opts.offline_timeout);
3623 #else /* AFS_PTHREAD_ENV */
3625 /* Waiting a certain amount of time for offlining volumes is not supported
3626 * for LWP due to a lack of primitives. So, we never time out */
3627 # define VTimedOut(x) (0)
3628 # define VOfflineTimeout(x) (NULL)
3630 #endif /* !AFS_PTHREAD_ENV */
3638 retVal = VHold_r(vp);
3645 VIsGoingOffline_r(struct Volume *vp)
3649 if (vp->goingOffline) {
3650 if (vp->specialStatus) {
3651 code = vp->specialStatus;
3652 } else if (V_inService(vp) == 0 || V_blessed(vp) == 0) {
3663 * Tell the caller if a volume is waiting to go offline.
3665 * @param[in] vp The volume we want to know about
3667 * @return volume status
3668 * @retval 0 volume is not waiting to go offline, go ahead and use it
3669 * @retval nonzero volume is waiting to offline, and give the returned code
3670 * as an error to anyone accessing the volume
3672 * @pre VOL_LOCK is NOT held
3673 * @pre caller holds a heavyweight reference on vp
3676 VIsGoingOffline(struct Volume *vp)
3681 code = VIsGoingOffline_r(vp);
3688 * Register an RX call with a volume.
3690 * @param[inout] ec Error code; if unset when passed in, may be set if
3691 * the volume starts going offline
3692 * @param[out] client_ec @see GetVolume
3693 * @param[in] vp Volume struct
3694 * @param[in] cbv VCallByVol struct containing the RX call to register
3696 * @pre VOL_LOCK held
3697 * @pre caller holds heavy ref on vp
3702 VRegisterCall_r(Error *ec, Error *client_ec, Volume *vp, struct VCallByVol *cbv)
3705 #ifdef AFS_DEMAND_ATTACH_FS
3707 /* just in case the volume started going offline after we got the
3708 * reference to it... otherwise, if the volume started going
3709 * offline right at the end of GetVolume(), we might race with the
3710 * RX call scanner, and return success and add our cbv to the
3711 * rx_call_list _after_ the scanner has scanned the list. */
3712 *ec = VIsGoingOffline_r(vp);
3718 while (V_attachState(vp) == VOL_STATE_SCANNING_RXCALLS) {
3719 VWaitStateChange_r(vp);
3721 #endif /* AFS_DEMAND_ATTACH_FS */
3723 queue_Prepend(&vp->rx_call_list, cbv);
3728 * Deregister an RX call with a volume.
3730 * @param[in] vp Volume struct
3731 * @param[in] cbv VCallByVol struct containing the RX call to deregister
3733 * @pre VOL_LOCK held
3734 * @pre caller holds heavy ref on vp
3739 VDeregisterCall_r(Volume *vp, struct VCallByVol *cbv)
3741 if (cbv && queue_IsOnQueue(cbv)) {
3742 #ifdef AFS_DEMAND_ATTACH_FS
3743 while (V_attachState(vp) == VOL_STATE_SCANNING_RXCALLS) {
3744 VWaitStateChange_r(vp);
3746 #endif /* AFS_DEMAND_ATTACH_FS */
3752 /***************************************************/
3753 /* get and put volume routines */
3754 /***************************************************/
3757 * put back a heavyweight reference to a volume object.
3759 * @param[in] vp volume object pointer
3761 * @pre VOL_LOCK held
3763 * @post heavyweight volume reference put back.
3764 * depending on state, volume may have been taken offline,
3765 * detached, salvaged, freed, etc.
3767 * @internal volume package internal use only
3770 VPutVolume_r(Volume * vp)
3772 osi_Assert(--vp->nUsers >= 0);
3773 if (vp->nUsers == 0) {
3775 ReleaseVolumeHeader(vp->header);
3776 #ifdef AFS_DEMAND_ATTACH_FS
3777 if (!VCheckDetach(vp)) {
3781 #else /* AFS_DEMAND_ATTACH_FS */
3783 #endif /* AFS_DEMAND_ATTACH_FS */
3788 VPutVolume(Volume * vp)
3796 * Puts a volume reference obtained with VGetVolumeWithCall.
3798 * @param[in] vp Volume struct
3799 * @param[in] cbv VCallByVol struct given to VGetVolumeWithCall, or NULL if none
3801 * @pre VOL_LOCK is NOT held
3804 VPutVolumeWithCall(Volume *vp, struct VCallByVol *cbv)
3807 VDeregisterCall_r(vp, cbv);
3812 /* Get a pointer to an attached volume. The pointer is returned regardless
3813 of whether or not the volume is in service or on/off line. An error
3814 code, however, is returned with an indication of the volume's status */
3816 VGetVolume(Error * ec, Error * client_ec, VolId volumeId)
3820 retVal = GetVolume(ec, client_ec, volumeId, NULL, 0);
3826 * Get a volume reference associated with an RX call.
3828 * @param[out] ec @see GetVolume
3829 * @param[out] client_ec @see GetVolume
3830 * @param[in] volumeId @see GetVolume
3831 * @param[in] ts How long to wait for going-offline volumes (absolute time).
3832 * If NULL, wait forever. If ts->tv_sec == 0, return immediately
3833 * with an error if the volume is going offline.
3834 * @param[in] cbv Contains an RX call to be associated with this volume
3835 * reference. This call may be interrupted if the volume is
3836 * requested to go offline while we hold a ref on it. Give NULL
3837 * to not associate an RX call with this reference.
3839 * @return @see GetVolume
3841 * @note for LWP builds, ts must be NULL
3843 * @note A reference obtained with this function MUST be put back with
3844 * VPutVolumeWithCall
3847 VGetVolumeWithCall(Error * ec, Error * client_ec, VolId volumeId,
3848 const struct timespec *ts, struct VCallByVol *cbv)
3852 retVal = GetVolume(ec, client_ec, volumeId, NULL, ts);
3853 VRegisterCall_r(ec, client_ec, retVal, cbv);
3859 VGetVolume_r(Error * ec, VolId volumeId)
3861 return GetVolume(ec, NULL, volumeId, NULL, NULL);
3864 /* try to get a volume we've previously looked up */
3865 /* for demand attach fs, caller MUST NOT hold a ref count on vp */
3867 VGetVolumeByVp_r(Error * ec, Volume * vp)
3869 return GetVolume(ec, NULL, vp->hashid, vp, NULL);
3873 * private interface for getting a volume handle
3875 * @param[out] ec error code (0 if no error)
3876 * @param[out] client_ec wire error code to be given to clients
3877 * @param[in] volumeId ID of the volume we want
3878 * @param[in] hint optional hint for hash lookups, or NULL
3879 * @param[in] timeout absolute deadline for waiting for the volume to go
3880 * offline, if it is going offline. NULL to wait forever.
3882 * @return a volume handle for the specified volume
3883 * @retval NULL an error occurred, or the volume is in such a state that
3884 * we cannot load a header or return any volume struct
3886 * @note for DAFS, caller must NOT hold a ref count on 'hint'
3888 * @note 'timeout' is only checked if the volume is actually going offline; so
3889 * if you pass timeout->tv_sec = 0, this will exhibit typical
3890 * nonblocking behavior.
3892 * @note for LWP builds, 'timeout' must be NULL
3895 GetVolume(Error * ec, Error * client_ec, VolId volumeId, Volume * hint,
3896 const struct timespec *timeout)
3899 /* pull this profiling/debugging code out of regular builds */
3901 #define VGET_CTR_INC(x) x++
3902 unsigned short V0 = 0, V1 = 0, V2 = 0, V3 = 0, V5 = 0, V6 =
3903 0, V7 = 0, V8 = 0, V9 = 0;
3904 unsigned short V10 = 0, V11 = 0, V12 = 0, V13 = 0, V14 = 0, V15 = 0;
3906 #define VGET_CTR_INC(x)
3908 #ifdef AFS_DEMAND_ATTACH_FS
3909 Volume *avp, * rvp = hint;
3913 * if VInit is zero, the volume package dynamic
3914 * data structures have not been initialized yet,
3915 * and we must immediately return an error
3921 *client_ec = VOFFLINE;
3926 #ifdef AFS_DEMAND_ATTACH_FS
3928 VCreateReservation_r(rvp);
3930 #endif /* AFS_DEMAND_ATTACH_FS */
3938 vp = VLookupVolume_r(ec, volumeId, vp);
3944 #ifdef AFS_DEMAND_ATTACH_FS
3945 if (rvp && (rvp != vp)) {
3946 /* break reservation on old vp */
3947 VCancelReservation_r(rvp);
3950 #endif /* AFS_DEMAND_ATTACH_FS */
3956 /* Until we have reached an initialization level of 2
3957 * we don't know whether this volume exists or not.
3958 * We can't sleep and retry later because before a volume
3959 * is attached, the caller tries to get it first. Just
3960 * return VOFFLINE and the caller can choose whether to
3961 * retry the command or not. */
3971 IncUInt64(&VStats.hdr_gets);
3973 #ifdef AFS_DEMAND_ATTACH_FS
3974 /* block if someone else is performing an exclusive op on this volume */
3977 VCreateReservation_r(rvp);
3979 VWaitExclusiveState_r(vp);
3981 /* short circuit with VNOVOL in the following circumstances:
3984 * - VOL_STATE_SHUTTING_DOWN
3986 if ((V_attachState(vp) == VOL_STATE_ERROR) ||
3987 (V_attachState(vp) == VOL_STATE_SHUTTING_DOWN) ||
3988 (V_attachState(vp) == VOL_STATE_GOING_OFFLINE)) {
3995 * short circuit with VOFFLINE for VOL_STATE_UNATTACHED and
3996 * VNOVOL for VOL_STATE_DELETED
3998 if ((V_attachState(vp) == VOL_STATE_UNATTACHED) ||
3999 (V_attachState(vp) == VOL_STATE_DELETED)) {
4000 if (vp->specialStatus) {
4001 *ec = vp->specialStatus;
4002 } else if (V_attachState(vp) == VOL_STATE_DELETED) {
4011 /* allowable states:
4018 if (vp->salvage.requested) {
4019 VUpdateSalvagePriority_r(vp);
4022 if (V_attachState(vp) == VOL_STATE_PREATTACHED) {
4023 avp = VAttachVolumeByVp_r(ec, vp, 0);
4026 /* VAttachVolumeByVp_r can return a pointer
4027 * != the vp passed to it under certain
4028 * conditions; make sure we don't leak
4029 * reservations if that happens */
4031 VCancelReservation_r(rvp);
4033 VCreateReservation_r(rvp);
4043 if (!vp->pending_vol_op) {
4058 if (VIsSalvaging(vp) || (*ec == VSALVAGING)) {
4060 /* see CheckVnode() in afsfileprocs.c for an explanation
4061 * of this error code logic */
4062 afs_uint32 now = FT_ApproxTime();
4063 if ((vp->stats.last_salvage + (10 * 60)) >= now) {
4066 *client_ec = VRESTARTING;
4075 #ifdef AFS_DEMAND_ATTACH_FS
4077 * this test MUST happen after VAttachVolymeByVp, so vol_op_state is
4078 * not VolOpRunningUnknown (attach2 would have converted it to Online
4082 /* only valid before/during demand attachment */
4083 osi_Assert(!vp->pending_vol_op || vp->pending_vol_op->vol_op_state != FSSYNC_VolOpRunningUnknown);
4085 /* deny getvolume due to running mutually exclusive vol op */
4086 if (vp->pending_vol_op && vp->pending_vol_op->vol_op_state==FSSYNC_VolOpRunningOffline) {
4088 * volume cannot remain online during this volume operation.
4091 if (vp->specialStatus) {
4093 * special status codes outrank normal VOFFLINE code
4095 *ec = vp->specialStatus;
4097 *client_ec = vp->specialStatus;
4101 /* see CheckVnode() in afsfileprocs.c for an explanation
4102 * of this error code logic */
4103 afs_uint32 now = FT_ApproxTime();
4104 if ((vp->stats.last_vol_op + (10 * 60)) >= now) {
4107 *client_ec = VRESTARTING;
4112 VChangeState_r(vp, VOL_STATE_UNATTACHED);
4113 FreeVolumeHeader(vp);
4117 #endif /* AFS_DEMAND_ATTACH_FS */
4119 LoadVolumeHeader(ec, vp);
4122 /* Only log the error if it was a totally unexpected error. Simply
4123 * a missing inode is likely to be caused by the volume being deleted */
4124 if (errno != ENXIO || LogLevel)
4125 Log("Volume %u: couldn't reread volume header\n",
4127 #ifdef AFS_DEMAND_ATTACH_FS
4128 if (VCanScheduleSalvage()) {
4129 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
4134 #else /* AFS_DEMAND_ATTACH_FS */
4137 #endif /* AFS_DEMAND_ATTACH_FS */
4142 if (vp->shuttingDown) {
4149 if (programType == fileServer) {
4151 if (vp->goingOffline) {
4152 if (timeout && VTimedOut(timeout)) {
4153 /* we've timed out; don't wait for the vol */
4156 #ifdef AFS_DEMAND_ATTACH_FS
4157 /* wait for the volume to go offline */
4158 if (V_attachState(vp) == VOL_STATE_GOING_OFFLINE) {
4159 VTimedWaitStateChange_r(vp, timeout, NULL);
4161 #elif defined(AFS_PTHREAD_ENV)
4162 VOL_CV_TIMEDWAIT(&vol_put_volume_cond, timeout, NULL);
4163 #else /* AFS_PTHREAD_ENV */
4164 /* LWP has no timed wait, so the caller better not be
4166 osi_Assert(!timeout);
4167 LWP_WaitProcess(VPutVolume);
4168 #endif /* AFS_PTHREAD_ENV */
4172 if (vp->specialStatus) {
4174 *ec = vp->specialStatus;
4175 } else if (V_inService(vp) == 0 || V_blessed(vp) == 0) {
4178 } else if (V_inUse(vp) == 0 || vp->goingOffline) {
4189 #ifdef AFS_DEMAND_ATTACH_FS
4190 /* if no error, bump nUsers */
4193 VLRU_UpdateAccess_r(vp);
4196 VCancelReservation_r(rvp);
4199 if (client_ec && !*client_ec) {
4202 #else /* AFS_DEMAND_ATTACH_FS */
4203 /* if no error, bump nUsers */
4210 #endif /* AFS_DEMAND_ATTACH_FS */
4213 osi_Assert(vp || *ec);
4218 /***************************************************/
4219 /* Volume offline/detach routines */
4220 /***************************************************/
4222 /* caller MUST hold a heavyweight ref on vp */
4223 #ifdef AFS_DEMAND_ATTACH_FS
4225 VTakeOffline_r(Volume * vp)
4229 osi_Assert(vp->nUsers > 0);
4230 osi_Assert(programType == fileServer);
4232 VCreateReservation_r(vp);
4233 VWaitExclusiveState_r(vp);
4235 vp->goingOffline = 1;
4236 V_needsSalvaged(vp) = 1;
4238 VRequestSalvage_r(&error, vp, SALVSYNC_ERROR, 0);
4239 VCancelReservation_r(vp);
4241 #else /* AFS_DEMAND_ATTACH_FS */
4243 VTakeOffline_r(Volume * vp)
4245 osi_Assert(vp->nUsers > 0);
4246 osi_Assert(programType == fileServer);
4248 vp->goingOffline = 1;
4249 V_needsSalvaged(vp) = 1;
4251 #endif /* AFS_DEMAND_ATTACH_FS */
4254 VTakeOffline(Volume * vp)
4262 * force a volume offline.
4264 * @param[in] vp volume object pointer
4265 * @param[in] flags flags (see note below)
4267 * @note the flag VOL_FORCEOFF_NOUPDATE is a recursion control flag
4268 * used when VUpdateVolume_r needs to call VForceOffline_r
4269 * (which in turn would normally call VUpdateVolume_r)
4271 * @see VUpdateVolume_r
4273 * @pre VOL_LOCK must be held.
4274 * for DAFS, caller must hold ref.
4276 * @note for DAFS, it _is safe_ to call this function from an
4279 * @post needsSalvaged flag is set.
4280 * for DAFS, salvage is requested.
4281 * no further references to the volume through the volume
4282 * package will be honored.
4283 * all file descriptor and vnode caches are invalidated.
4285 * @warning this is a heavy-handed interface. it results in
4286 * a volume going offline regardless of the current
4287 * reference count state.
4289 * @internal volume package internal use only
4292 VForceOffline_r(Volume * vp, int flags)
4296 #ifdef AFS_DEMAND_ATTACH_FS
4297 VChangeState_r(vp, VOL_STATE_ERROR);
4302 strcpy(V_offlineMessage(vp),
4303 "Forced offline due to internal error: volume needs to be salvaged");
4304 Log("Volume %u forced offline: it needs salvaging!\n", V_id(vp));
4307 vp->goingOffline = 0;
4308 V_needsSalvaged(vp) = 1;
4309 if (!(flags & VOL_FORCEOFF_NOUPDATE)) {
4310 VUpdateVolume_r(&error, vp, VOL_UPDATE_NOFORCEOFF);
4313 #ifdef AFS_DEMAND_ATTACH_FS
4314 VRequestSalvage_r(&error, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
4315 #endif /* AFS_DEMAND_ATTACH_FS */
4317 #ifdef AFS_PTHREAD_ENV
4318 CV_BROADCAST(&vol_put_volume_cond);
4319 #else /* AFS_PTHREAD_ENV */
4320 LWP_NoYieldSignal(VPutVolume);
4321 #endif /* AFS_PTHREAD_ENV */
4323 VReleaseVolumeHandles_r(vp);
4327 * force a volume offline.
4329 * @param[in] vp volume object pointer
4331 * @see VForceOffline_r
4334 VForceOffline(Volume * vp)
4337 VForceOffline_r(vp, 0);
4342 * Iterate over the RX calls associated with a volume, and interrupt them.
4344 * @param[in] vp The volume whose RX calls we want to scan
4346 * @pre VOL_LOCK held
4349 VScanCalls_r(struct Volume *vp)
4351 struct VCallByVol *cbv, *ncbv;
4353 #ifdef AFS_DEMAND_ATTACH_FS
4354 VolState state_save;
4357 if (queue_IsEmpty(&vp->rx_call_list))
4358 return; /* no calls to interrupt */
4359 if (!vol_opts.interrupt_rxcall)
4360 return; /* we have no function with which to interrupt calls */
4361 err = VIsGoingOffline_r(vp);
4363 return; /* we're not going offline anymore */
4365 #ifdef AFS_DEMAND_ATTACH_FS
4366 VWaitExclusiveState_r(vp);
4367 state_save = VChangeState_r(vp, VOL_STATE_SCANNING_RXCALLS);
4369 #endif /* AFS_DEMAND_ATTACH_FS */
4371 for(queue_Scan(&vp->rx_call_list, cbv, ncbv, VCallByVol)) {
4373 struct rx_peer *peer;
4375 peer = rx_PeerOf(rx_ConnectionOf(cbv->call));
4377 Log("Offlining volume %lu while client %s:%u is trying to read "
4378 "from it; kicking client off with error %ld\n",
4379 (long unsigned) vp->hashid,
4380 afs_inet_ntoa_r(rx_HostOf(peer), hoststr),
4381 (unsigned) ntohs(rx_PortOf(peer)),
4384 (*vol_opts.interrupt_rxcall) (cbv->call, err);
4387 #ifdef AFS_DEMAND_ATTACH_FS
4389 VChangeState_r(vp, state_save);
4390 #endif /* AFS_DEMAND_ATTACH_FS */
4393 #ifdef AFS_DEMAND_ATTACH_FS
4395 * Wait for a vp to go offline.
4397 * @param[out] ec 1 if a salvage on the volume has been requested and
4398 * salvok == 0, 0 otherwise
4399 * @param[in] vp The volume to wait for
4400 * @param[in] salvok If 0, we return immediately with *ec = 1 if the volume
4401 * has been requested to salvage. Otherwise we keep waiting
4402 * until the volume has gone offline.
4404 * @pre VOL_LOCK held
4405 * @pre caller holds a lightweight ref on vp
4410 VWaitForOfflineByVp_r(Error *ec, struct Volume *vp, int salvok)
4412 struct timespec timeout_ts;
4413 const struct timespec *ts;
4416 ts = VOfflineTimeout(&timeout_ts);
4420 while (!VIsOfflineState(V_attachState(vp)) && !timedout) {
4421 if (!salvok && vp->salvage.requested) {
4425 VTimedWaitStateChange_r(vp, ts, &timedout);
4428 /* we didn't time out, so the volume must be offline, so we're done */
4432 /* If we got here, we timed out waiting for the volume to go offline.
4433 * Kick off the accessing RX calls and wait again */
4437 while (!VIsOfflineState(V_attachState(vp))) {
4438 if (!salvok && vp->salvage.requested) {
4443 VWaitStateChange_r(vp);
4447 #else /* AFS_DEMAND_ATTACH_FS */
4450 * Wait for a volume to go offline.
4452 * @pre VOL_LOCK held
4454 * @note non-DAFS only (for DAFS, use @see WaitForOfflineByVp_r)
4457 VWaitForOffline_r(Error *ec, VolumeId volid)
4460 const struct timespec *ts;
4461 #ifdef AFS_PTHREAD_ENV
4462 struct timespec timeout_ts;
4465 ts = VOfflineTimeout(&timeout_ts);
4467 vp = GetVolume(ec, NULL, volid, NULL, ts);
4469 /* error occurred so bad that we can't even get a vp; we have no
4470 * information on the vol so we don't know whether to wait, so just
4474 if (!VIsGoingOffline_r(vp)) {
4475 /* volume is no longer going offline, so we're done */
4480 /* If we got here, we timed out waiting for the volume to go offline.
4481 * Kick off the accessing RX calls and wait again */
4487 vp = VGetVolume_r(ec, volid);
4489 /* In case it was reattached... */
4493 #endif /* !AFS_DEMAND_ATTACH_FS */
4495 /* The opposite of VAttachVolume. The volume header is written to disk, with
4496 the inUse bit turned off. A copy of the header is maintained in memory,
4497 however (which is why this is VOffline, not VDetach).
4500 VOffline_r(Volume * vp, char *message)
4503 #ifndef AFS_DEMAND_ATTACH_FS
4504 VolumeId vid = V_id(vp);
4507 osi_Assert(programType != volumeUtility && programType != volumeServer);
4512 if (V_offlineMessage(vp)[0] == '\0')
4513 strncpy(V_offlineMessage(vp), message, sizeof(V_offlineMessage(vp)));
4514 V_offlineMessage(vp)[sizeof(V_offlineMessage(vp)) - 1] = '\0';
4516 vp->goingOffline = 1;
4517 #ifdef AFS_DEMAND_ATTACH_FS
4518 VChangeState_r(vp, VOL_STATE_GOING_OFFLINE);
4519 VCreateReservation_r(vp);
4521 VWaitForOfflineByVp_r(&error, vp, 1);
4522 VCancelReservation_r(vp);
4523 #else /* AFS_DEMAND_ATTACH_FS */
4525 VWaitForOffline_r(&error, vid);
4526 #endif /* AFS_DEMAND_ATTACH_FS */
4529 #ifdef AFS_DEMAND_ATTACH_FS
4531 * Take a volume offline in order to perform a volume operation.
4533 * @param[inout] ec address in which to store error code
4534 * @param[in] vp volume object pointer
4535 * @param[in] message volume offline status message
4538 * - VOL_LOCK is held
4539 * - caller MUST hold a heavyweight ref on vp
4542 * - volume is taken offline
4543 * - if possible, volume operation is promoted to running state
4544 * - on failure, *ec is set to nonzero
4546 * @note Although this function does not return any value, it may
4547 * still fail to promote our pending volume operation to
4548 * a running state. Any caller MUST check the value of *ec,
4549 * and MUST NOT blindly assume success.
4551 * @warning if the caller does not hold a lightweight ref on vp,
4552 * then it MUST NOT reference vp after this function
4553 * returns to the caller.
4555 * @internal volume package internal use only
4558 VOfflineForVolOp_r(Error *ec, Volume *vp, char *message)
4561 osi_Assert(vp->pending_vol_op);
4567 if (V_offlineMessage(vp)[0] == '\0')
4568 strncpy(V_offlineMessage(vp), message, sizeof(V_offlineMessage(vp)));
4569 V_offlineMessage(vp)[sizeof(V_offlineMessage(vp)) - 1] = '\0';
4571 vp->goingOffline = 1;
4572 VChangeState_r(vp, VOL_STATE_GOING_OFFLINE);
4573 VCreateReservation_r(vp);
4576 if (vp->pending_vol_op->com.programType != salvageServer) {
4577 /* do not give corrupted volumes to the volserver */
4582 VWaitForOfflineByVp_r(ec, vp, salvok);
4584 VCancelReservation_r(vp);
4586 #endif /* AFS_DEMAND_ATTACH_FS */
4589 VOffline(Volume * vp, char *message)
4592 VOffline_r(vp, message);
4596 /* This gets used for the most part by utility routines that don't want
4597 * to keep all the volume headers around. Generally, the file server won't
4598 * call this routine, because then the offline message in the volume header
4599 * (or other information) won't be available to clients. For NAMEI, also
4600 * close the file handles. However, the fileserver does call this during
4601 * an attach following a volume operation.
4604 VDetachVolume_r(Error * ec, Volume * vp)
4606 #ifdef FSSYNC_BUILD_CLIENT
4608 struct DiskPartition64 *tpartp;
4609 int notifyServer = 0;
4610 int useDone = FSYNC_VOL_ON;
4612 if (VCanUseFSSYNC()) {
4613 notifyServer = vp->needsPutBack;
4614 if (V_destroyMe(vp) == DESTROY_ME)
4615 useDone = FSYNC_VOL_LEAVE_OFF;
4616 #ifdef AFS_DEMAND_ATTACH_FS
4617 else if (!V_blessed(vp) || !V_inService(vp))
4618 useDone = FSYNC_VOL_LEAVE_OFF;
4621 tpartp = vp->partition;
4623 #endif /* FSSYNC_BUILD_CLIENT */
4625 *ec = 0; /* always "succeeds" */
4626 DeleteVolumeFromHashTable(vp);
4627 vp->shuttingDown = 1;
4628 #ifdef AFS_DEMAND_ATTACH_FS
4629 DeleteVolumeFromVByPList_r(vp);
4631 VChangeState_r(vp, VOL_STATE_SHUTTING_DOWN);
4633 if (programType != fileServer)
4635 #endif /* AFS_DEMAND_ATTACH_FS */
4637 /* Will be detached sometime in the future--this is OK since volume is offline */
4639 /* XXX the following code should really be moved to VCheckDetach() since the volume
4640 * is not technically detached until the refcounts reach zero
4642 #ifdef FSSYNC_BUILD_CLIENT
4643 if (VCanUseFSSYNC() && notifyServer) {
4644 if (notifyServer == VOL_PUTBACK_DELETE) {
4645 /* Only send FSYNC_VOL_DONE if the volume was actually deleted.
4646 * volserver code will set needsPutBack to VOL_PUTBACK_DELETE
4647 * to signify a deleted volume. */
4648 useDone = FSYNC_VOL_DONE;
4651 * Note: The server is not notified in the case of a bogus volume
4652 * explicitly to make it possible to create a volume, do a partial
4653 * restore, then abort the operation without ever putting the volume
4654 * online. This is essential in the case of a volume move operation
4655 * between two partitions on the same server. In that case, there
4656 * would be two instances of the same volume, one of them bogus,
4657 * which the file server would attempt to put on line
4659 FSYNC_VolOp(volume, tpartp->name, useDone, 0, NULL);
4660 /* XXX this code path is only hit by volume utilities, thus
4661 * V_BreakVolumeCallbacks will always be NULL. if we really
4662 * want to break callbacks in this path we need to use FSYNC_VolOp() */
4664 /* Dettaching it so break all callbacks on it */
4665 if (V_BreakVolumeCallbacks) {
4666 Log("volume %u detached; breaking all call backs\n", volume);
4667 (*V_BreakVolumeCallbacks) (volume);
4671 #endif /* FSSYNC_BUILD_CLIENT */
4675 VDetachVolume(Error * ec, Volume * vp)
4678 VDetachVolume_r(ec, vp);
4683 /***************************************************/
4684 /* Volume fd/inode handle closing routines */
4685 /***************************************************/
4687 /* For VDetachVolume, we close all cached file descriptors, but keep
4688 * the Inode handles in case we need to read from a busy volume.
4690 /* for demand attach, caller MUST hold ref count on vp */
4692 VCloseVolumeHandles_r(Volume * vp)
4694 #ifdef AFS_DEMAND_ATTACH_FS
4695 VolState state_save;
4697 state_save = VChangeState_r(vp, VOL_STATE_OFFLINING);
4702 DFlushVolume(vp->hashid);
4704 #ifdef AFS_DEMAND_ATTACH_FS
4708 /* DAFS: VCloseVnodeFiles_r drops the glock internally */
4709 VCloseVnodeFiles_r(vp);
4711 #ifdef AFS_DEMAND_ATTACH_FS
4715 /* Too time consuming and unnecessary for the volserver */
4716 if (programType == fileServer) {
4717 IH_CONDSYNC(vp->vnodeIndex[vLarge].handle);
4718 IH_CONDSYNC(vp->vnodeIndex[vSmall].handle);
4719 IH_CONDSYNC(vp->diskDataHandle);
4721 IH_CONDSYNC(vp->linkHandle);
4722 #endif /* AFS_NT40_ENV */
4725 IH_REALLYCLOSE(vp->vnodeIndex[vLarge].handle);
4726 IH_REALLYCLOSE(vp->vnodeIndex[vSmall].handle);
4727 IH_REALLYCLOSE(vp->diskDataHandle);
4728 IH_REALLYCLOSE(vp->linkHandle);
4730 #ifdef AFS_DEMAND_ATTACH_FS
4731 if ((V_attachFlags(vp) & VOL_LOCKED)) {
4736 VChangeState_r(vp, state_save);
4740 /* For both VForceOffline and VOffline, we close all relevant handles.
4741 * For VOffline, if we re-attach the volume, the files may possible be
4742 * different than before.
4744 /* for demand attach, caller MUST hold a ref count on vp */
4746 VReleaseVolumeHandles_r(Volume * vp)
4748 #ifdef AFS_DEMAND_ATTACH_FS
4749 VolState state_save;
4751 state_save = VChangeState_r(vp, VOL_STATE_DETACHING);
4756 DFlushVolume(vp->hashid);
4758 #ifdef AFS_DEMAND_ATTACH_FS
4762 VReleaseVnodeFiles_r(vp); /* DAFS: releases the glock internally */
4764 #ifdef AFS_DEMAND_ATTACH_FS
4768 /* Too time consuming and unnecessary for the volserver */
4769 if (programType == fileServer) {
4770 IH_CONDSYNC(vp->vnodeIndex[vLarge].handle);
4771 IH_CONDSYNC(vp->vnodeIndex[vSmall].handle);
4772 IH_CONDSYNC(vp->diskDataHandle);
4774 IH_CONDSYNC(vp->linkHandle);
4775 #endif /* AFS_NT40_ENV */
4778 IH_RELEASE(vp->vnodeIndex[vLarge].handle);
4779 IH_RELEASE(vp->vnodeIndex[vSmall].handle);
4780 IH_RELEASE(vp->diskDataHandle);
4781 IH_RELEASE(vp->linkHandle);
4783 #ifdef AFS_DEMAND_ATTACH_FS
4784 if ((V_attachFlags(vp) & VOL_LOCKED)) {
4789 VChangeState_r(vp, state_save);
4794 /***************************************************/
4795 /* Volume write and fsync routines */
4796 /***************************************************/
4799 VUpdateVolume_r(Error * ec, Volume * vp, int flags)
4801 #ifdef AFS_DEMAND_ATTACH_FS
4802 VolState state_save;
4804 if (flags & VOL_UPDATE_WAIT) {
4805 VCreateReservation_r(vp);
4806 VWaitExclusiveState_r(vp);
4811 if (programType == fileServer)
4813 (V_inUse(vp) ? V_nextVnodeUnique(vp) +
4814 200 : V_nextVnodeUnique(vp));
4816 #ifdef AFS_DEMAND_ATTACH_FS
4817 state_save = VChangeState_r(vp, VOL_STATE_UPDATING);
4821 WriteVolumeHeader_r(ec, vp);
4823 #ifdef AFS_DEMAND_ATTACH_FS
4825 VChangeState_r(vp, state_save);
4826 if (flags & VOL_UPDATE_WAIT) {
4827 VCancelReservation_r(vp);
4832 Log("VUpdateVolume: error updating volume header, volume %u (%s)\n",
4833 V_id(vp), V_name(vp));
4834 /* try to update on-disk header,
4835 * while preventing infinite recursion */
4836 if (!(flags & VOL_UPDATE_NOFORCEOFF)) {
4837 VForceOffline_r(vp, VOL_FORCEOFF_NOUPDATE);
4843 VUpdateVolume(Error * ec, Volume * vp)
4846 VUpdateVolume_r(ec, vp, VOL_UPDATE_WAIT);
4851 VSyncVolume_r(Error * ec, Volume * vp, int flags)
4855 #ifdef AFS_DEMAND_ATTACH_FS
4856 VolState state_save;
4859 if (flags & VOL_SYNC_WAIT) {
4860 VUpdateVolume_r(ec, vp, VOL_UPDATE_WAIT);
4862 VUpdateVolume_r(ec, vp, 0);
4865 #ifdef AFS_DEMAND_ATTACH_FS
4866 state_save = VChangeState_r(vp, VOL_STATE_UPDATING);
4869 fdP = IH_OPEN(V_diskDataHandle(vp));
4870 osi_Assert(fdP != NULL);
4871 code = FDH_SYNC(fdP);
4872 osi_Assert(code == 0);
4874 #ifdef AFS_DEMAND_ATTACH_FS
4876 VChangeState_r(vp, state_save);
4882 VSyncVolume(Error * ec, Volume * vp)
4885 VSyncVolume_r(ec, vp, VOL_SYNC_WAIT);
4890 /***************************************************/
4891 /* Volume dealloaction routines */
4892 /***************************************************/
4894 #ifdef AFS_DEMAND_ATTACH_FS
4896 FreeVolume(Volume * vp)
4898 /* free the heap space, iff it's safe.
4899 * otherwise, pull it out of the hash table, so it
4900 * will get deallocated when all refs to it go away */
4901 if (!VCheckFree(vp)) {
4902 DeleteVolumeFromHashTable(vp);
4903 DeleteVolumeFromVByPList_r(vp);
4905 /* make sure we invalidate the header cache entry */
4906 FreeVolumeHeader(vp);
4909 #endif /* AFS_DEMAND_ATTACH_FS */
4912 ReallyFreeVolume(Volume * vp)
4917 #ifdef AFS_DEMAND_ATTACH_FS
4919 VChangeState_r(vp, VOL_STATE_FREED);
4920 if (vp->pending_vol_op)
4921 free(vp->pending_vol_op);
4922 #endif /* AFS_DEMAND_ATTACH_FS */
4923 for (i = 0; i < nVNODECLASSES; i++)
4924 if (vp->vnodeIndex[i].bitmap)
4925 free(vp->vnodeIndex[i].bitmap);
4926 FreeVolumeHeader(vp);
4927 #ifndef AFS_DEMAND_ATTACH_FS
4928 DeleteVolumeFromHashTable(vp);
4929 #endif /* AFS_DEMAND_ATTACH_FS */
4933 /* check to see if we should shutdown this volume
4934 * returns 1 if volume was freed, 0 otherwise */
4935 #ifdef AFS_DEMAND_ATTACH_FS
4937 VCheckDetach(Volume * vp)
4942 if (vp->nUsers || vp->nWaiters)
4945 if (vp->shuttingDown) {
4947 if ((programType != fileServer) &&
4948 (V_inUse(vp) == programType) &&
4949 ((V_checkoutMode(vp) == V_VOLUPD) ||
4950 (V_checkoutMode(vp) == V_SECRETLY) ||
4951 ((V_checkoutMode(vp) == V_CLONE) &&
4952 (VolumeWriteable(vp))))) {
4954 VUpdateVolume_r(&ec, vp, VOL_UPDATE_NOFORCEOFF);
4956 Log("VCheckDetach: volume header update for volume %u "
4957 "failed with errno %d\n", vp->hashid, errno);
4960 VReleaseVolumeHandles_r(vp);
4962 ReallyFreeVolume(vp);
4963 if (programType == fileServer) {
4964 CV_BROADCAST(&vol_put_volume_cond);
4969 #else /* AFS_DEMAND_ATTACH_FS */
4971 VCheckDetach(Volume * vp)
4979 if (vp->shuttingDown) {
4981 if ((programType != fileServer) &&
4982 (V_inUse(vp) == programType) &&
4983 ((V_checkoutMode(vp) == V_VOLUPD) ||
4984 (V_checkoutMode(vp) == V_SECRETLY) ||
4985 ((V_checkoutMode(vp) == V_CLONE) &&
4986 (VolumeWriteable(vp))))) {
4988 VUpdateVolume_r(&ec, vp, VOL_UPDATE_NOFORCEOFF);
4990 Log("VCheckDetach: volume header update for volume %u failed with errno %d\n",
4994 VReleaseVolumeHandles_r(vp);
4995 ReallyFreeVolume(vp);
4996 if (programType == fileServer) {
4997 #if defined(AFS_PTHREAD_ENV)
4998 CV_BROADCAST(&vol_put_volume_cond);
4999 #else /* AFS_PTHREAD_ENV */
5000 LWP_NoYieldSignal(VPutVolume);
5001 #endif /* AFS_PTHREAD_ENV */
5006 #endif /* AFS_DEMAND_ATTACH_FS */
5008 /* check to see if we should offline this volume
5009 * return 1 if volume went offline, 0 otherwise */
5010 #ifdef AFS_DEMAND_ATTACH_FS
5012 VCheckOffline(Volume * vp)
5016 if (vp->goingOffline && !vp->nUsers) {
5018 osi_Assert(programType == fileServer);
5019 osi_Assert((V_attachState(vp) != VOL_STATE_ATTACHED) &&
5020 (V_attachState(vp) != VOL_STATE_FREED) &&
5021 (V_attachState(vp) != VOL_STATE_PREATTACHED) &&
5022 (V_attachState(vp) != VOL_STATE_UNATTACHED) &&
5023 (V_attachState(vp) != VOL_STATE_DELETED));
5027 * VOL_STATE_GOING_OFFLINE
5028 * VOL_STATE_SHUTTING_DOWN
5029 * VIsErrorState(V_attachState(vp))
5030 * VIsExclusiveState(V_attachState(vp))
5033 VCreateReservation_r(vp);
5034 VChangeState_r(vp, VOL_STATE_OFFLINING);
5037 /* must clear the goingOffline flag before we drop the glock */
5038 vp->goingOffline = 0;
5043 /* perform async operations */
5044 VUpdateVolume_r(&error, vp, 0);
5045 VCloseVolumeHandles_r(vp);
5048 if (V_offlineMessage(vp)[0]) {
5049 Log("VOffline: Volume %lu (%s) is now offline (%s)\n",
5050 afs_printable_uint32_lu(V_id(vp)), V_name(vp),
5051 V_offlineMessage(vp));
5053 Log("VOffline: Volume %lu (%s) is now offline\n",
5054 afs_printable_uint32_lu(V_id(vp)), V_name(vp));
5058 /* invalidate the volume header cache entry */
5059 FreeVolumeHeader(vp);
5061 /* if nothing changed state to error or salvaging,
5062 * drop state to unattached */
5063 if (!VIsErrorState(V_attachState(vp))) {
5064 VChangeState_r(vp, VOL_STATE_UNATTACHED);
5066 VCancelReservation_r(vp);
5067 /* no usage of vp is safe beyond this point */
5071 #else /* AFS_DEMAND_ATTACH_FS */
5073 VCheckOffline(Volume * vp)
5077 if (vp->goingOffline && !vp->nUsers) {
5079 osi_Assert(programType == fileServer);
5082 vp->goingOffline = 0;
5084 VUpdateVolume_r(&error, vp, 0);
5085 VCloseVolumeHandles_r(vp);
5087 if (V_offlineMessage(vp)[0]) {
5088 Log("VOffline: Volume %lu (%s) is now offline (%s)\n",
5089 afs_printable_uint32_lu(V_id(vp)), V_name(vp),
5090 V_offlineMessage(vp));
5092 Log("VOffline: Volume %lu (%s) is now offline\n",
5093 afs_printable_uint32_lu(V_id(vp)), V_name(vp));
5096 FreeVolumeHeader(vp);
5097 #ifdef AFS_PTHREAD_ENV
5098 CV_BROADCAST(&vol_put_volume_cond);
5099 #else /* AFS_PTHREAD_ENV */
5100 LWP_NoYieldSignal(VPutVolume);
5101 #endif /* AFS_PTHREAD_ENV */
5105 #endif /* AFS_DEMAND_ATTACH_FS */
5107 /***************************************************/
5108 /* demand attach fs ref counting routines */
5109 /***************************************************/
5111 #ifdef AFS_DEMAND_ATTACH_FS
5112 /* the following two functions handle reference counting for
5113 * asynchronous operations on volume structs.
5115 * their purpose is to prevent a VDetachVolume or VShutdown
5116 * from free()ing the Volume struct during an async i/o op */
5118 /* register with the async volume op ref counter */
5119 /* VCreateReservation_r moved into inline code header because it
5120 * is now needed in vnode.c -- tkeiser 11/20/2007
5124 * decrement volume-package internal refcount.
5126 * @param vp volume object pointer
5128 * @internal volume package internal use only
5131 * @arg VOL_LOCK is held
5132 * @arg lightweight refcount held
5134 * @post volume waiters refcount is decremented; volume may
5135 * have been deallocated/shutdown/offlined/salvaged/
5136 * whatever during the process
5138 * @warning once you have tossed your last reference (you can acquire
5139 * lightweight refs recursively) it is NOT SAFE to reference
5140 * a volume object pointer ever again
5142 * @see VCreateReservation_r
5144 * @note DEMAND_ATTACH_FS only
5147 VCancelReservation_r(Volume * vp)
5149 osi_Assert(--vp->nWaiters >= 0);
5150 if (vp->nWaiters == 0) {
5152 if (!VCheckDetach(vp)) {
5159 /* check to see if we should free this volume now
5160 * return 1 if volume was freed, 0 otherwise */
5162 VCheckFree(Volume * vp)
5165 if ((vp->nUsers == 0) &&
5166 (vp->nWaiters == 0) &&
5167 !(V_attachFlags(vp) & (VOL_IN_HASH |
5171 ReallyFreeVolume(vp);
5176 #endif /* AFS_DEMAND_ATTACH_FS */
5179 /***************************************************/
5180 /* online volume operations routines */
5181 /***************************************************/
5183 #ifdef AFS_DEMAND_ATTACH_FS
5185 * register a volume operation on a given volume.
5187 * @param[in] vp volume object
5188 * @param[in] vopinfo volume operation info object
5190 * @pre VOL_LOCK is held
5192 * @post volume operation info object attached to volume object.
5193 * volume operation statistics updated.
5195 * @note by "attached" we mean a copy of the passed in object is made
5197 * @internal volume package internal use only
5200 VRegisterVolOp_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
5202 FSSYNC_VolOp_info * info;
5204 /* attach a vol op info node to the volume struct */
5205 info = (FSSYNC_VolOp_info *) malloc(sizeof(FSSYNC_VolOp_info));
5206 osi_Assert(info != NULL);
5207 memcpy(info, vopinfo, sizeof(FSSYNC_VolOp_info));
5208 vp->pending_vol_op = info;
5211 vp->stats.last_vol_op = FT_ApproxTime();
5212 vp->stats.vol_ops++;
5213 IncUInt64(&VStats.vol_ops);
5219 * deregister the volume operation attached to this volume.
5221 * @param[in] vp volume object pointer
5223 * @pre VOL_LOCK is held
5225 * @post the volume operation info object is detached from the volume object
5227 * @internal volume package internal use only
5230 VDeregisterVolOp_r(Volume * vp)
5232 if (vp->pending_vol_op) {
5233 free(vp->pending_vol_op);
5234 vp->pending_vol_op = NULL;
5238 #endif /* AFS_DEMAND_ATTACH_FS */
5241 * determine whether it is safe to leave a volume online during
5242 * the volume operation described by the vopinfo object.
5244 * @param[in] vp volume object
5245 * @param[in] vopinfo volume operation info object
5247 * @return whether it is safe to leave volume online
5248 * @retval 0 it is NOT SAFE to leave the volume online
5249 * @retval 1 it is safe to leave the volume online during the operation
5252 * @arg VOL_LOCK is held
5253 * @arg disk header attached to vp (heavyweight ref on vp will guarantee
5254 * this condition is met)
5256 * @internal volume package internal use only
5259 VVolOpLeaveOnline_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
5261 return (vopinfo->vol_op_state == FSSYNC_VolOpRunningOnline ||
5262 (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
5263 (vopinfo->com.reason == V_READONLY ||
5264 (!VolumeWriteable(vp) &&
5265 (vopinfo->com.reason == V_CLONE ||
5266 vopinfo->com.reason == V_DUMP)))));
5270 * same as VVolOpLeaveOnline_r, but does not require a volume with an attached
5273 * @param[in] vp volume object
5274 * @param[in] vopinfo volume operation info object
5276 * @return whether it is safe to leave volume online
5277 * @retval 0 it is NOT SAFE to leave the volume online
5278 * @retval 1 it is safe to leave the volume online during the operation
5279 * @retval -1 unsure; volume header is required in order to know whether or
5280 * not is is safe to leave the volume online
5282 * @pre VOL_LOCK is held
5284 * @internal volume package internal use only
5287 VVolOpLeaveOnlineNoHeader_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
5289 /* follow the logic in VVolOpLeaveOnline_r; this is the same, except
5290 * assume that we don't know VolumeWriteable; return -1 if the answer
5291 * depends on VolumeWriteable */
5293 if (vopinfo->vol_op_state == FSSYNC_VolOpRunningOnline) {
5296 if (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
5297 vopinfo->com.reason == V_READONLY) {
5301 if (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
5302 (vopinfo->com.reason == V_CLONE ||
5303 vopinfo->com.reason == V_DUMP)) {
5305 /* must know VolumeWriteable */
5312 * determine whether VBUSY should be set during this volume operation.
5314 * @param[in] vp volume object
5315 * @param[in] vopinfo volume operation info object
5317 * @return whether VBUSY should be set
5318 * @retval 0 VBUSY does NOT need to be set
5319 * @retval 1 VBUSY SHOULD be set
5321 * @pre VOL_LOCK is held
5323 * @internal volume package internal use only
5326 VVolOpSetVBusy_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
5328 return ((vopinfo->com.command == FSYNC_VOL_OFF &&
5329 vopinfo->com.reason == FSYNC_SALVAGE) ||
5330 (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
5331 (vopinfo->com.reason == V_CLONE ||
5332 vopinfo->com.reason == V_DUMP)));
5336 /***************************************************/
5337 /* online salvager routines */
5338 /***************************************************/
5339 #if defined(AFS_DEMAND_ATTACH_FS)
5342 * offline a volume to let it be salvaged.
5344 * @param[in] vp Volume to offline
5346 * @return whether we offlined the volume successfully
5347 * @retval 0 volume was not offlined
5348 * @retval 1 volume is now offline
5350 * @note This is similar to VCheckOffline, but slightly different. We do not
5351 * deal with vp->goingOffline, and we try to avoid touching the volume
5352 * header except just to set needsSalvaged
5354 * @pre VOL_LOCK held
5355 * @pre vp->nUsers == 0
5356 * @pre V_attachState(vp) == VOL_STATE_SALVAGE_REQ
5359 VOfflineForSalvage_r(struct Volume *vp)
5363 VCreateReservation_r(vp);
5364 VWaitExclusiveState_r(vp);
5366 if (vp->nUsers || V_attachState(vp) == VOL_STATE_SALVAGING) {
5367 /* Someone's using the volume, or someone got to scheduling the salvage
5368 * before us. I don't think either of these should be possible, as we
5369 * should gain no new heavyweight references while we're trying to
5370 * salvage, but just to be sure... */
5371 VCancelReservation_r(vp);
5375 VChangeState_r(vp, VOL_STATE_OFFLINING);
5379 V_needsSalvaged(vp) = 1;
5380 /* ignore error; updating needsSalvaged is just best effort */
5381 VUpdateVolume_r(&error, vp, VOL_UPDATE_NOFORCEOFF);
5383 VCloseVolumeHandles_r(vp);
5385 FreeVolumeHeader(vp);
5387 /* volume has been effectively offlined; we can mark it in the SALVAGING
5388 * state now, which lets FSSYNC give it away */
5389 VChangeState_r(vp, VOL_STATE_SALVAGING);
5391 VCancelReservation_r(vp);
5397 * check whether a salvage needs to be performed on this volume.
5399 * @param[in] vp pointer to volume object
5401 * @return status code
5402 * @retval 0 no salvage scheduled
5403 * @retval 1 a salvage has been scheduled with the salvageserver
5405 * @pre VOL_LOCK is held
5407 * @post if salvage request flag is set and nUsers and nWaiters are zero,
5408 * then a salvage will be requested
5410 * @note this is one of the event handlers called by VCancelReservation_r
5412 * @note the caller must check if the volume needs to be freed after calling
5413 * this; the volume may not have any references or be on any lists after
5414 * we return, and we do not free it
5416 * @see VCancelReservation_r
5418 * @internal volume package internal use only.
5421 VCheckSalvage(Volume * vp)
5424 #if defined(SALVSYNC_BUILD_CLIENT) || defined(FSSYNC_BUILD_CLIENT)
5427 if (!vp->salvage.requested) {
5431 /* prevent recursion; some of the code below creates and removes
5432 * lightweight refs, which can call VCheckSalvage */
5433 if (vp->salvage.scheduling) {
5436 vp->salvage.scheduling = 1;
5438 if (V_attachState(vp) == VOL_STATE_SALVAGE_REQ) {
5439 if (!VOfflineForSalvage_r(vp)) {
5440 vp->salvage.scheduling = 0;
5445 if (vp->salvage.requested) {
5446 VScheduleSalvage_r(vp);
5449 vp->salvage.scheduling = 0;
5450 #endif /* SALVSYNC_BUILD_CLIENT || FSSYNC_BUILD_CLIENT */
5455 * request volume salvage.
5457 * @param[out] ec computed client error code
5458 * @param[in] vp volume object pointer
5459 * @param[in] reason reason code (passed to salvageserver via SALVSYNC)
5460 * @param[in] flags see flags note below
5463 * VOL_SALVAGE_INVALIDATE_HEADER causes volume header cache entry
5464 * to be invalidated.
5466 * @pre VOL_LOCK is held.
5468 * @post volume state is changed.
5469 * for fileserver, salvage will be requested once refcount reaches zero.
5471 * @return operation status code
5472 * @retval 0 volume salvage will occur
5473 * @retval 1 volume salvage could not be scheduled
5477 * @note in the fileserver, this call does not synchronously schedule a volume
5478 * salvage. rather, it sets volume state so that when volume refcounts
5479 * reach zero, a volume salvage will occur. by "refcounts", we mean both
5480 * nUsers and nWaiters must be zero.
5482 * @internal volume package internal use only.
5485 VRequestSalvage_r(Error * ec, Volume * vp, int reason, int flags)
5489 * for DAFS volume utilities that are not supposed to schedule salvages,
5490 * just transition to error state instead
5492 if (!VCanScheduleSalvage()) {
5493 VChangeState_r(vp, VOL_STATE_ERROR);
5498 if (programType != fileServer && !VCanUseFSSYNC()) {
5499 VChangeState_r(vp, VOL_STATE_ERROR);
5504 if (!vp->salvage.requested) {
5505 vp->salvage.requested = 1;
5506 vp->salvage.reason = reason;
5507 vp->stats.last_salvage = FT_ApproxTime();
5509 /* Note that it is not possible for us to reach this point if a
5510 * salvage is already running on this volume (even if the fileserver
5511 * was restarted during the salvage). If a salvage were running, the
5512 * salvager would have write-locked the volume header file, so when
5513 * we tried to lock the volume header, the lock would have failed,
5514 * and we would have failed during attachment prior to calling
5515 * VRequestSalvage. So we know that we can schedule salvages without
5516 * fear of a salvage already running for this volume. */
5518 if (vp->stats.salvages < SALVAGE_COUNT_MAX) {
5520 /* if we don't need to offline the volume, we can go directly
5521 * to SALVAGING. SALVAGING says the volume is offline and is
5522 * either salvaging or ready to be handed to the salvager.
5523 * SALVAGE_REQ says that we want to salvage the volume, but we
5524 * are waiting for it to go offline first. */
5525 if (flags & VOL_SALVAGE_NO_OFFLINE) {
5526 VChangeState_r(vp, VOL_STATE_SALVAGING);
5528 VChangeState_r(vp, VOL_STATE_SALVAGE_REQ);
5529 if (vp->nUsers == 0) {
5530 /* normally VOfflineForSalvage_r would be called from
5531 * PutVolume et al when nUsers reaches 0, but if
5532 * it's already 0, just do it ourselves, since PutVolume
5533 * isn't going to get called */
5534 VOfflineForSalvage_r(vp);
5537 /* If we are non-fileserver, we're telling the fileserver to
5538 * salvage the vol, so we don't need to give it back separately. */
5539 vp->needsPutBack = 0;
5543 Log("VRequestSalvage: volume %u online salvaged too many times; forced offline.\n", vp->hashid);
5545 /* make sure neither VScheduleSalvage_r nor
5546 * VUpdateSalvagePriority_r try to schedule another salvage */
5547 vp->salvage.requested = vp->salvage.scheduled = 0;
5549 VChangeState_r(vp, VOL_STATE_ERROR);
5553 if (flags & VOL_SALVAGE_INVALIDATE_HEADER) {
5554 /* Instead of ReleaseVolumeHeader, we do FreeVolumeHeader()
5555 so that the the next VAttachVolumeByVp_r() invocation
5556 of attach2() will pull in a cached header
5557 entry and fail, then load a fresh one from disk and attach
5560 FreeVolumeHeader(vp);
5567 * update salvageserver scheduling priority for a volume.
5569 * @param[in] vp pointer to volume object
5571 * @return operation status
5573 * @retval 1 request denied, or SALVSYNC communications failure
5575 * @pre VOL_LOCK is held.
5577 * @post in-core salvage priority counter is incremented. if at least
5578 * SALVAGE_PRIO_UPDATE_INTERVAL seconds have elapsed since the
5579 * last SALVSYNC_RAISEPRIO request, we contact the salvageserver
5580 * to update its priority queue. if no salvage is scheduled,
5581 * this function is a no-op.
5583 * @note DAFS fileserver only
5585 * @note this should be called whenever a VGetVolume fails due to a
5586 * pending salvage request
5588 * @todo should set exclusive state and drop glock around salvsync call
5590 * @internal volume package internal use only.
5593 VUpdateSalvagePriority_r(Volume * vp)
5597 #ifdef SALVSYNC_BUILD_CLIENT
5602 now = FT_ApproxTime();
5604 /* update the salvageserver priority queue occasionally so that
5605 * frequently requested volumes get moved to the head of the queue
5607 if ((vp->salvage.scheduled) &&
5608 (vp->stats.last_salvage_req < (now-SALVAGE_PRIO_UPDATE_INTERVAL))) {
5609 code = SALVSYNC_SalvageVolume(vp->hashid,
5610 VPartitionPath(vp->partition),
5615 vp->stats.last_salvage_req = now;
5616 if (code != SYNC_OK) {
5620 #endif /* SALVSYNC_BUILD_CLIENT */
5625 #if defined(SALVSYNC_BUILD_CLIENT) || defined(FSSYNC_BUILD_CLIENT)
5627 /* A couple of little helper functions. These return true if we tried to
5628 * use this mechanism to schedule a salvage, false if we haven't tried.
5629 * If we did try a salvage then the results are contained in code.
5633 try_SALVSYNC(Volume *vp, char *partName, int *code) {
5634 #ifdef SALVSYNC_BUILD_CLIENT
5635 if (VCanUseSALVSYNC()) {
5636 Log("Scheduling salvage for volume %lu on part %s over SALVSYNC\n",
5637 afs_printable_uint32_lu(vp->hashid), partName);
5639 /* can't use V_id() since there's no guarantee
5640 * we have the disk data header at this point */
5641 *code = SALVSYNC_SalvageVolume(vp->hashid,
5654 try_FSSYNC(Volume *vp, char *partName, int *code) {
5655 #ifdef FSSYNC_BUILD_CLIENT
5656 if (VCanUseFSSYNC()) {
5657 Log("Scheduling salvage for volume %lu on part %s over FSSYNC\n",
5658 afs_printable_uint32_lu(vp->hashid), partName);
5661 * If we aren't the fileserver, tell the fileserver the volume
5662 * needs to be salvaged. We could directly tell the
5663 * salvageserver, but the fileserver keeps track of some stats
5664 * related to salvages, and handles some other salvage-related
5665 * complications for us.
5667 *code = FSYNC_VolOp(vp->hashid, partName,
5668 FSYNC_VOL_FORCE_ERROR, FSYNC_SALVAGE, NULL);
5671 #endif /* FSSYNC_BUILD_CLIENT */
5676 * schedule a salvage with the salvage server or fileserver.
5678 * @param[in] vp pointer to volume object
5680 * @return operation status
5681 * @retval 0 salvage scheduled successfully
5682 * @retval 1 salvage not scheduled, or SALVSYNC/FSSYNC com error
5685 * @arg VOL_LOCK is held.
5686 * @arg nUsers and nWaiters should be zero.
5688 * @post salvageserver or fileserver is sent a salvage request
5690 * @note If we are the fileserver, the request will be sent to the salvage
5691 * server over SALVSYNC. If we are not the fileserver, the request will be
5692 * sent to the fileserver over FSSYNC (FSYNC_VOL_FORCE_ERROR/FSYNC_SALVAGE).
5694 * @note the caller must check if the volume needs to be freed after calling
5695 * this; the volume may not have any references or be on any lists after
5696 * we return, and we do not free it
5700 * @internal volume package internal use only.
5703 VScheduleSalvage_r(Volume * vp)
5707 VolState state_save;
5708 VThreadOptions_t * thread_opts;
5711 osi_Assert(VCanUseSALVSYNC() || VCanUseFSSYNC());
5713 if (vp->nWaiters || vp->nUsers) {
5717 /* prevent endless salvage,attach,salvage,attach,... loops */
5718 if (vp->stats.salvages >= SALVAGE_COUNT_MAX)
5722 * don't perform salvsync ops on certain threads
5724 thread_opts = pthread_getspecific(VThread_key);
5725 if (thread_opts == NULL) {
5726 thread_opts = &VThread_defaults;
5728 if (thread_opts->disallow_salvsync || vol_disallow_salvsync) {
5732 if (vp->salvage.scheduled) {
5736 VCreateReservation_r(vp);
5737 VWaitExclusiveState_r(vp);
5740 * XXX the scheduling process should really be done asynchronously
5741 * to avoid fssync deadlocks
5743 if (!vp->salvage.scheduled) {
5744 /* if we haven't previously scheduled a salvage, do so now
5746 * set the volume to an exclusive state and drop the lock
5747 * around the SALVSYNC call
5749 strlcpy(partName, VPartitionPath(vp->partition), sizeof(partName));
5750 state_save = VChangeState_r(vp, VOL_STATE_SALVSYNC_REQ);
5753 osi_Assert(try_SALVSYNC(vp, partName, &code) ||
5754 try_FSSYNC(vp, partName, &code));
5757 VChangeState_r(vp, state_save);
5759 if (code == SYNC_OK) {
5760 vp->salvage.scheduled = 1;
5761 vp->stats.last_salvage_req = FT_ApproxTime();
5762 if (VCanUseSALVSYNC()) {
5763 /* don't record these stats for non-fileservers; let the
5764 * fileserver take care of these */
5765 vp->stats.salvages++;
5766 IncUInt64(&VStats.salvages);
5771 case SYNC_BAD_COMMAND:
5772 case SYNC_COM_ERROR:
5775 Log("VScheduleSalvage_r: Salvage request for volume %lu "
5776 "denied\n", afs_printable_uint32_lu(vp->hashid));
5779 Log("VScheduleSalvage_r: Salvage request for volume %lu "
5780 "received unknown protocol error %d\n",
5781 afs_printable_uint32_lu(vp->hashid), code);
5785 if (VCanUseFSSYNC()) {
5786 VChangeState_r(vp, VOL_STATE_ERROR);
5791 /* NB: this is cancelling the reservation we obtained above, but we do
5792 * not call VCancelReservation_r, since that may trigger the vp dtor,
5793 * possibly free'ing the vp. We need to keep the vp around after
5794 * this, as the caller may reference vp without any refs. Instead, it
5795 * is the duty of the caller to inspect 'vp' after we return to see if
5796 * needs to be freed. */
5797 osi_Assert(--vp->nWaiters >= 0);
5800 #endif /* SALVSYNC_BUILD_CLIENT || FSSYNC_BUILD_CLIENT */
5802 #ifdef SALVSYNC_BUILD_CLIENT
5805 * connect to the salvageserver SYNC service.
5807 * @return operation status
5811 * @post connection to salvageserver SYNC service established
5813 * @see VConnectSALV_r
5814 * @see VDisconnectSALV
5815 * @see VReconnectSALV
5822 retVal = VConnectSALV_r();
5828 * connect to the salvageserver SYNC service.
5830 * @return operation status
5834 * @pre VOL_LOCK is held.
5836 * @post connection to salvageserver SYNC service established
5839 * @see VDisconnectSALV_r
5840 * @see VReconnectSALV_r
5841 * @see SALVSYNC_clientInit
5843 * @internal volume package internal use only.
5846 VConnectSALV_r(void)
5848 return SALVSYNC_clientInit();
5852 * disconnect from the salvageserver SYNC service.
5854 * @return operation status
5857 * @pre client should have a live connection to the salvageserver
5859 * @post connection to salvageserver SYNC service destroyed
5861 * @see VDisconnectSALV_r
5863 * @see VReconnectSALV
5866 VDisconnectSALV(void)
5869 VDisconnectSALV_r();
5875 * disconnect from the salvageserver SYNC service.
5877 * @return operation status
5881 * @arg VOL_LOCK is held.
5882 * @arg client should have a live connection to the salvageserver.
5884 * @post connection to salvageserver SYNC service destroyed
5886 * @see VDisconnectSALV
5887 * @see VConnectSALV_r
5888 * @see VReconnectSALV_r
5889 * @see SALVSYNC_clientFinis
5891 * @internal volume package internal use only.
5894 VDisconnectSALV_r(void)
5896 return SALVSYNC_clientFinis();
5900 * disconnect and then re-connect to the salvageserver SYNC service.
5902 * @return operation status
5906 * @pre client should have a live connection to the salvageserver
5908 * @post old connection is dropped, and a new one is established
5911 * @see VDisconnectSALV
5912 * @see VReconnectSALV_r
5915 VReconnectSALV(void)
5919 retVal = VReconnectSALV_r();
5925 * disconnect and then re-connect to the salvageserver SYNC service.
5927 * @return operation status
5932 * @arg VOL_LOCK is held.
5933 * @arg client should have a live connection to the salvageserver.
5935 * @post old connection is dropped, and a new one is established
5937 * @see VConnectSALV_r
5938 * @see VDisconnectSALV
5939 * @see VReconnectSALV
5940 * @see SALVSYNC_clientReconnect
5942 * @internal volume package internal use only.
5945 VReconnectSALV_r(void)
5947 return SALVSYNC_clientReconnect();
5949 #endif /* SALVSYNC_BUILD_CLIENT */
5950 #endif /* AFS_DEMAND_ATTACH_FS */
5953 /***************************************************/
5954 /* FSSYNC routines */
5955 /***************************************************/
5957 /* This must be called by any volume utility which needs to run while the
5958 file server is also running. This is separated from VInitVolumePackage2 so
5959 that a utility can fork--and each of the children can independently
5960 initialize communication with the file server */
5961 #ifdef FSSYNC_BUILD_CLIENT
5963 * connect to the fileserver SYNC service.
5965 * @return operation status
5970 * @arg VInit must equal 2.
5971 * @arg Program Type must not be fileserver or salvager.
5973 * @post connection to fileserver SYNC service established
5976 * @see VDisconnectFS
5977 * @see VChildProcReconnectFS
5984 retVal = VConnectFS_r();
5990 * connect to the fileserver SYNC service.
5992 * @return operation status
5997 * @arg VInit must equal 2.
5998 * @arg Program Type must not be fileserver or salvager.
5999 * @arg VOL_LOCK is held.
6001 * @post connection to fileserver SYNC service established
6004 * @see VDisconnectFS_r
6005 * @see VChildProcReconnectFS_r
6007 * @internal volume package internal use only.
6013 osi_Assert((VInit == 2) &&
6014 (programType != fileServer) &&
6015 (programType != salvager));
6016 rc = FSYNC_clientInit();
6024 * disconnect from the fileserver SYNC service.
6027 * @arg client should have a live connection to the fileserver.
6028 * @arg VOL_LOCK is held.
6029 * @arg Program Type must not be fileserver or salvager.
6031 * @post connection to fileserver SYNC service destroyed
6033 * @see VDisconnectFS
6035 * @see VChildProcReconnectFS_r
6037 * @internal volume package internal use only.
6040 VDisconnectFS_r(void)
6042 osi_Assert((programType != fileServer) &&
6043 (programType != salvager));
6044 FSYNC_clientFinis();
6049 * disconnect from the fileserver SYNC service.
6052 * @arg client should have a live connection to the fileserver.
6053 * @arg Program Type must not be fileserver or salvager.
6055 * @post connection to fileserver SYNC service destroyed
6057 * @see VDisconnectFS_r
6059 * @see VChildProcReconnectFS
6070 * connect to the fileserver SYNC service from a child process following a fork.
6072 * @return operation status
6077 * @arg VOL_LOCK is held.
6078 * @arg current FSYNC handle is shared with a parent process
6080 * @post current FSYNC handle is discarded and a new connection to the
6081 * fileserver SYNC service is established
6083 * @see VChildProcReconnectFS
6085 * @see VDisconnectFS_r
6087 * @internal volume package internal use only.
6090 VChildProcReconnectFS_r(void)
6092 return FSYNC_clientChildProcReconnect();
6096 * connect to the fileserver SYNC service from a child process following a fork.
6098 * @return operation status
6102 * @pre current FSYNC handle is shared with a parent process
6104 * @post current FSYNC handle is discarded and a new connection to the
6105 * fileserver SYNC service is established
6107 * @see VChildProcReconnectFS_r
6109 * @see VDisconnectFS
6112 VChildProcReconnectFS(void)
6116 ret = VChildProcReconnectFS_r();
6120 #endif /* FSSYNC_BUILD_CLIENT */
6123 /***************************************************/
6124 /* volume bitmap routines */
6125 /***************************************************/
6128 * allocate a vnode bitmap number for the vnode
6130 * @param[out] ec error code
6131 * @param[in] vp volume object pointer
6132 * @param[in] index vnode index number for the vnode
6133 * @param[in] flags flag values described in note
6135 * @note for DAFS, flags parameter controls locking behavior.
6136 * If (flags & VOL_ALLOC_BITMAP_WAIT) is set, then this function
6137 * will create a reservation and block on any other exclusive
6138 * operations. Otherwise, this function assumes the caller
6139 * already has exclusive access to vp, and we just change the
6142 * @pre VOL_LOCK held
6144 * @return bit number allocated
6150 VAllocBitmapEntry_r(Error * ec, Volume * vp,
6151 struct vnodeIndex *index, int flags)
6155 #ifdef AFS_DEMAND_ATTACH_FS
6156 VolState state_save;
6157 #endif /* AFS_DEMAND_ATTACH_FS */
6161 /* This test is probably redundant */
6162 if (!VolumeWriteable(vp)) {
6163 *ec = (bit32) VREADONLY;
6167 #ifdef AFS_DEMAND_ATTACH_FS
6168 if (flags & VOL_ALLOC_BITMAP_WAIT) {
6169 VCreateReservation_r(vp);
6170 VWaitExclusiveState_r(vp);
6172 state_save = VChangeState_r(vp, VOL_STATE_GET_BITMAP);
6173 #endif /* AFS_DEMAND_ATTACH_FS */
6176 if ((programType == fileServer) && !index->bitmap) {
6178 #ifndef AFS_DEMAND_ATTACH_FS
6179 /* demand attach fs uses the volume state to avoid races.
6180 * specialStatus field is not used at all */
6182 if (vp->specialStatus == VBUSY) {
6183 if (vp->goingOffline) { /* vos dump waiting for the volume to
6184 * go offline. We probably come here
6185 * from AddNewReadableResidency */
6188 while (vp->specialStatus == VBUSY) {
6189 #ifdef AFS_PTHREAD_ENV
6193 #else /* !AFS_PTHREAD_ENV */
6195 #endif /* !AFS_PTHREAD_ENV */
6199 #endif /* !AFS_DEMAND_ATTACH_FS */
6201 if (!index->bitmap) {
6202 #ifndef AFS_DEMAND_ATTACH_FS
6203 vp->specialStatus = VBUSY; /* Stop anyone else from using it. */
6204 #endif /* AFS_DEMAND_ATTACH_FS */
6205 for (i = 0; i < nVNODECLASSES; i++) {
6206 VGetBitmap_r(ec, vp, i);
6208 #ifdef AFS_DEMAND_ATTACH_FS
6209 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
6210 #else /* AFS_DEMAND_ATTACH_FS */
6211 DeleteVolumeFromHashTable(vp);
6212 vp->shuttingDown = 1; /* Let who has it free it. */
6213 vp->specialStatus = 0;
6214 #endif /* AFS_DEMAND_ATTACH_FS */
6218 #ifndef AFS_DEMAND_ATTACH_FS
6220 vp->specialStatus = 0; /* Allow others to have access. */
6221 #endif /* AFS_DEMAND_ATTACH_FS */
6224 #endif /* BITMAP_LATER */
6226 #ifdef AFS_DEMAND_ATTACH_FS
6228 #endif /* AFS_DEMAND_ATTACH_FS */
6229 bp = index->bitmap + index->bitmapOffset;
6230 ep = index->bitmap + index->bitmapSize;
6232 if ((*(bit32 *) bp) != (bit32) 0xffffffff) {
6234 index->bitmapOffset = (afs_uint32) (bp - index->bitmap);
6237 o = ffs(~*bp) - 1; /* ffs is documented in BSTRING(3) */
6239 ret = ((bp - index->bitmap) * 8 + o);
6240 #ifdef AFS_DEMAND_ATTACH_FS
6242 #endif /* AFS_DEMAND_ATTACH_FS */
6245 bp += sizeof(bit32) /* i.e. 4 */ ;
6247 /* No bit map entry--must grow bitmap */
6249 realloc(index->bitmap, index->bitmapSize + VOLUME_BITMAP_GROWSIZE);
6250 osi_Assert(bp != NULL);
6252 bp += index->bitmapSize;
6253 memset(bp, 0, VOLUME_BITMAP_GROWSIZE);
6254 index->bitmapOffset = index->bitmapSize;
6255 index->bitmapSize += VOLUME_BITMAP_GROWSIZE;
6257 ret = index->bitmapOffset * 8;
6258 #ifdef AFS_DEMAND_ATTACH_FS
6260 #endif /* AFS_DEMAND_ATTACH_FS */
6263 #ifdef AFS_DEMAND_ATTACH_FS
6264 VChangeState_r(vp, state_save);
6265 if (flags & VOL_ALLOC_BITMAP_WAIT) {
6266 VCancelReservation_r(vp);
6268 #endif /* AFS_DEMAND_ATTACH_FS */
6273 VAllocBitmapEntry(Error * ec, Volume * vp, struct vnodeIndex * index)
6277 retVal = VAllocBitmapEntry_r(ec, vp, index, VOL_ALLOC_BITMAP_WAIT);
6283 VFreeBitMapEntry_r(Error * ec, Volume *vp, struct vnodeIndex *index,
6284 unsigned bitNumber, int flags)
6286 unsigned int offset;
6290 #ifdef AFS_DEMAND_ATTACH_FS
6291 if (flags & VOL_FREE_BITMAP_WAIT) {
6292 /* VAllocBitmapEntry_r allocs bitmap entries under an exclusive volume
6293 * state, so ensure we're not in an exclusive volume state when we update
6295 VCreateReservation_r(vp);
6296 VWaitExclusiveState_r(vp);
6303 #endif /* BITMAP_LATER */
6305 offset = bitNumber >> 3;
6306 if (offset >= index->bitmapSize) {
6310 if (offset < index->bitmapOffset)
6311 index->bitmapOffset = offset & ~3; /* Truncate to nearest bit32 */
6312 *(index->bitmap + offset) &= ~(1 << (bitNumber & 0x7));
6315 #ifdef AFS_DEMAND_ATTACH_FS
6316 VCancelReservation_r(vp);
6318 return; /* make the compiler happy for non-DAFS */
6322 VFreeBitMapEntry(Error * ec, Volume *vp, struct vnodeIndex *index,
6326 VFreeBitMapEntry_r(ec, vp, index, bitNumber, VOL_FREE_BITMAP_WAIT);
6330 /* this function will drop the glock internally.
6331 * for old pthread fileservers, this is safe thanks to vbusy.
6333 * for demand attach fs, caller must have already called
6334 * VCreateReservation_r and VWaitExclusiveState_r */
6336 VGetBitmap_r(Error * ec, Volume * vp, VnodeClass class)
6338 StreamHandle_t *file;
6339 afs_sfsize_t nVnodes, size;
6340 struct VnodeClassInfo *vcp = &VnodeClassInfo[class];
6341 struct vnodeIndex *vip = &vp->vnodeIndex[class];
6342 struct VnodeDiskObject *vnode;
6343 unsigned int unique = 0;
6347 #endif /* BITMAP_LATER */
6348 #ifdef AFS_DEMAND_ATTACH_FS
6349 VolState state_save;
6350 #endif /* AFS_DEMAND_ATTACH_FS */
6354 #ifdef AFS_DEMAND_ATTACH_FS
6355 state_save = VChangeState_r(vp, VOL_STATE_GET_BITMAP);
6356 #endif /* AFS_DEMAND_ATTACH_FS */
6359 fdP = IH_OPEN(vip->handle);
6360 osi_Assert(fdP != NULL);
6361 file = FDH_FDOPEN(fdP, "r");
6362 osi_Assert(file != NULL);
6363 vnode = (VnodeDiskObject *) malloc(vcp->diskSize);
6364 osi_Assert(vnode != NULL);
6365 size = OS_SIZE(fdP->fd_fd);
6366 osi_Assert(size != -1);
6367 nVnodes = (size <= vcp->diskSize ? 0 : size - vcp->diskSize)
6369 vip->bitmapSize = ((nVnodes / 8) + 10) / 4 * 4; /* The 10 is a little extra so
6370 * a few files can be created in this volume,
6371 * the whole thing is rounded up to nearest 4
6372 * bytes, because the bit map allocator likes
6375 BitMap = (byte *) calloc(1, vip->bitmapSize);
6376 osi_Assert(BitMap != NULL);
6377 #else /* BITMAP_LATER */
6378 vip->bitmap = (byte *) calloc(1, vip->bitmapSize);
6379 osi_Assert(vip->bitmap != NULL);
6380 vip->bitmapOffset = 0;
6381 #endif /* BITMAP_LATER */
6382 if (STREAM_ASEEK(file, vcp->diskSize) != -1) {
6384 for (bitNumber = 0; bitNumber < nVnodes + 100; bitNumber++) {
6385 if (STREAM_READ(vnode, vcp->diskSize, 1, file) != 1)
6387 if (vnode->type != vNull) {
6388 if (vnode->vnodeMagic != vcp->magic) {
6389 Log("GetBitmap: addled vnode index in volume %s; volume needs salvage\n", V_name(vp));
6394 *(BitMap + (bitNumber >> 3)) |= (1 << (bitNumber & 0x7));
6395 #else /* BITMAP_LATER */
6396 *(vip->bitmap + (bitNumber >> 3)) |= (1 << (bitNumber & 0x7));
6397 #endif /* BITMAP_LATER */
6398 if (unique <= vnode->uniquifier)
6399 unique = vnode->uniquifier + 1;
6401 #ifndef AFS_PTHREAD_ENV
6402 if ((bitNumber & 0x00ff) == 0x0ff) { /* every 256 iterations */
6405 #endif /* !AFS_PTHREAD_ENV */
6408 if (vp->nextVnodeUnique < unique) {
6409 Log("GetBitmap: bad volume uniquifier for volume %s; volume needs salvage\n", V_name(vp));
6412 /* Paranoia, partly justified--I think fclose after fdopen
6413 * doesn't seem to close fd. In any event, the documentation
6414 * doesn't specify, so it's safer to close it twice.
6422 /* There may have been a racing condition with some other thread, both
6423 * creating the bitmaps for this volume. If the other thread was faster
6424 * the pointer to bitmap should already be filled and we can free ours.
6426 if (vip->bitmap == NULL) {
6427 vip->bitmap = BitMap;
6428 vip->bitmapOffset = 0;
6430 free((byte *) BitMap);
6431 #endif /* BITMAP_LATER */
6432 #ifdef AFS_DEMAND_ATTACH_FS
6433 VChangeState_r(vp, state_save);
6434 #endif /* AFS_DEMAND_ATTACH_FS */
6438 /***************************************************/
6439 /* Volume Path and Volume Number utility routines */
6440 /***************************************************/
6443 * find the first occurrence of a volume header file and return the path.
6445 * @param[out] ec outbound error code
6446 * @param[in] volumeId volume id to find
6447 * @param[out] partitionp pointer to disk partition path string
6448 * @param[out] namep pointer to volume header file name string
6450 * @post path to first occurrence of volume header is returned in partitionp
6451 * and namep, or ec is set accordingly.
6453 * @warning this function is NOT re-entrant -- partitionp and namep point to
6454 * static data segments
6456 * @note if a volume utility inadvertently leaves behind a stale volume header
6457 * on a vice partition, it is possible for callers to get the wrong one,
6458 * depending on the order of the disk partition linked list.
6462 VGetVolumePath(Error * ec, VolId volumeId, char **partitionp, char **namep)
6464 static char partition[VMAXPATHLEN], name[VMAXPATHLEN];
6465 char path[VMAXPATHLEN];
6467 struct DiskPartition64 *dp;
6470 name[0] = OS_DIRSEPC;
6471 snprintf(&name[1], (sizeof name) - 1, VFORMAT,
6472 afs_printable_uint32_lu(volumeId));
6473 for (dp = DiskPartitionList; dp; dp = dp->next) {
6474 struct afs_stat_st status;
6475 strcpy(path, VPartitionPath(dp));
6477 if (afs_stat(path, &status) == 0) {
6478 strcpy(partition, dp->name);
6485 *partitionp = *namep = NULL;
6487 *partitionp = partition;
6493 * extract a volume number from a volume header filename string.
6495 * @param[in] name volume header filename string
6497 * @return volume number
6499 * @note the string must be of the form VFORMAT. the only permissible
6500 * deviation is a leading OS_DIRSEPC character.
6505 VolumeNumber(char *name)
6507 if (*name == OS_DIRSEPC)
6509 return strtoul(name + 1, NULL, 10);
6513 * compute the volume header filename.
6515 * @param[in] volumeId
6517 * @return volume header filename
6519 * @post volume header filename string is constructed
6521 * @warning this function is NOT re-entrant -- the returned string is
6522 * stored in a static char array. see VolumeExternalName_r
6523 * for a re-entrant equivalent.
6525 * @see VolumeExternalName_r
6527 * @deprecated due to the above re-entrancy warning, this interface should
6528 * be considered deprecated. Please use VolumeExternalName_r
6532 VolumeExternalName(VolumeId volumeId)
6534 static char name[VMAXPATHLEN];
6535 snprintf(name, sizeof name, VFORMAT, afs_printable_uint32_lu(volumeId));
6540 * compute the volume header filename.
6542 * @param[in] volumeId
6543 * @param[inout] name array in which to store filename
6544 * @param[in] len length of name array
6546 * @return result code from afs_snprintf
6548 * @see VolumeExternalName
6551 * @note re-entrant equivalent of VolumeExternalName
6554 VolumeExternalName_r(VolumeId volumeId, char * name, size_t len)
6556 return snprintf(name, len, VFORMAT, afs_printable_uint32_lu(volumeId));
6560 /***************************************************/
6561 /* Volume Usage Statistics routines */
6562 /***************************************************/
6564 #if OPENAFS_VOL_STATS
6565 #define OneDay (86400) /* 24 hours' worth of seconds */
6567 #define OneDay (24*60*60) /* 24 hours */
6568 #endif /* OPENAFS_VOL_STATS */
6571 Midnight(time_t t) {
6572 struct tm local, *l;
6575 #if defined(AFS_PTHREAD_ENV) && !defined(AFS_NT40_ENV)
6576 l = localtime_r(&t, &local);
6582 /* the following is strictly speaking problematic on the
6583 switching day to daylight saving time, after the switch,
6584 as tm_isdst does not match. Similarly, on the looong day when
6585 switching back the OneDay check will not do what naively expected!
6586 The effects are minor, though, and more a matter of interpreting
6588 #ifndef AFS_PTHREAD_ENV
6591 local.tm_hour = local.tm_min=local.tm_sec = 0;
6592 midnight = mktime(&local);
6593 if (midnight != (time_t) -1) return(midnight);
6595 return( (t/OneDay)*OneDay );
6599 /*------------------------------------------------------------------------
6600 * [export] VAdjustVolumeStatistics
6603 * If we've passed midnight, we need to update all the day use
6604 * statistics as well as zeroing the detailed volume statistics
6605 * (if we are implementing them).
6608 * vp : Pointer to the volume structure describing the lucky
6609 * volume being considered for update.
6615 * Nothing interesting.
6619 *------------------------------------------------------------------------*/
6622 VAdjustVolumeStatistics_r(Volume * vp)
6624 unsigned int now = FT_ApproxTime();
6626 if (now - V_dayUseDate(vp) > OneDay) {
6629 ndays = (now - V_dayUseDate(vp)) / OneDay;
6630 for (i = 6; i > ndays - 1; i--)
6631 V_weekUse(vp)[i] = V_weekUse(vp)[i - ndays];
6632 for (i = 0; i < ndays - 1 && i < 7; i++)
6633 V_weekUse(vp)[i] = 0;
6635 V_weekUse(vp)[ndays - 1] = V_dayUse(vp);
6637 V_dayUseDate(vp) = Midnight(now);
6639 #if OPENAFS_VOL_STATS
6641 * All we need to do is bzero the entire VOL_STATS_BYTES of
6642 * the detailed volume statistics area.
6644 memset((V_stat_area(vp)), 0, VOL_STATS_BYTES);
6645 #endif /* OPENAFS_VOL_STATS */
6648 /*It's been more than a day of collection */
6650 * Always return happily.
6653 } /*VAdjustVolumeStatistics */
6656 VAdjustVolumeStatistics(Volume * vp)
6660 retVal = VAdjustVolumeStatistics_r(vp);
6666 VBumpVolumeUsage_r(Volume * vp)
6668 unsigned int now = FT_ApproxTime();
6669 V_accessDate(vp) = now;
6670 if (now - V_dayUseDate(vp) > OneDay)
6671 VAdjustVolumeStatistics_r(vp);
6673 * Save the volume header image to disk after every 128 bumps to dayUse.
6675 if ((V_dayUse(vp)++ & 127) == 0) {
6677 VUpdateVolume_r(&error, vp, VOL_UPDATE_WAIT);
6682 VBumpVolumeUsage(Volume * vp)
6685 VBumpVolumeUsage_r(vp);
6690 VSetDiskUsage_r(void)
6692 #ifndef AFS_DEMAND_ATTACH_FS
6693 static int FifteenMinuteCounter = 0;
6697 /* NOTE: Don't attempt to access the partitions list until the
6698 * initialization level indicates that all volumes are attached,
6699 * which implies that all partitions are initialized. */
6700 #ifdef AFS_PTHREAD_ENV
6701 VOL_CV_WAIT(&vol_vinit_cond);
6702 #else /* AFS_PTHREAD_ENV */
6704 #endif /* AFS_PTHREAD_ENV */
6707 VResetDiskUsage_r();
6709 #ifndef AFS_DEMAND_ATTACH_FS
6710 if (++FifteenMinuteCounter == 3) {
6711 FifteenMinuteCounter = 0;
6714 #endif /* !AFS_DEMAND_ATTACH_FS */
6726 /***************************************************/
6727 /* Volume Update List routines */
6728 /***************************************************/
6730 /* The number of minutes that a volume hasn't been updated before the
6731 * "Dont salvage" flag in the volume header will be turned on */
6732 #define SALVAGE_INTERVAL (10*60)
6737 * volume update list functionality has been moved into the VLRU
6738 * the DONT_SALVAGE flag is now set during VLRU demotion
6741 #ifndef AFS_DEMAND_ATTACH_FS
6742 static VolumeId *UpdateList = NULL; /* Pointer to array of Volume ID's */
6743 static int nUpdatedVolumes = 0; /* Updated with entry in UpdateList, salvage after crash flag on */
6744 static int updateSize = 0; /* number of entries possible */
6745 #define UPDATE_LIST_SIZE 128 /* initial size increment (must be a power of 2!) */
6746 #endif /* !AFS_DEMAND_ATTACH_FS */
6749 VAddToVolumeUpdateList_r(Error * ec, Volume * vp)
6752 vp->updateTime = FT_ApproxTime();
6753 if (V_dontSalvage(vp) == 0)
6755 V_dontSalvage(vp) = 0;
6756 VSyncVolume_r(ec, vp, 0);
6757 #ifdef AFS_DEMAND_ATTACH_FS
6758 V_attachFlags(vp) &= ~(VOL_HDR_DONTSALV);
6759 #else /* !AFS_DEMAND_ATTACH_FS */
6762 if (UpdateList == NULL) {
6763 updateSize = UPDATE_LIST_SIZE;
6764 UpdateList = (VolumeId *) malloc(sizeof(VolumeId) * updateSize);
6766 if (nUpdatedVolumes == updateSize) {
6768 if (updateSize > 524288) {
6769 Log("warning: there is likely a bug in the volume update scanner\n");
6773 (VolumeId *) realloc(UpdateList,
6774 sizeof(VolumeId) * updateSize);
6777 osi_Assert(UpdateList != NULL);
6778 UpdateList[nUpdatedVolumes++] = V_id(vp);
6779 #endif /* !AFS_DEMAND_ATTACH_FS */
6782 #ifndef AFS_DEMAND_ATTACH_FS
6784 VScanUpdateList(void)
6789 afs_uint32 now = FT_ApproxTime();
6790 /* Be careful with this code, since it works with interleaved calls to AddToVolumeUpdateList */
6791 for (i = gap = 0; i < nUpdatedVolumes; i++) {
6793 UpdateList[i - gap] = UpdateList[i];
6795 /* XXX this routine needlessly messes up the Volume LRU by
6796 * breaking the LRU temporal-locality assumptions.....
6797 * we should use a special volume header allocator here */
6798 vp = VGetVolume_r(&error, UpdateList[i - gap] = UpdateList[i]);
6801 } else if (vp->nUsers == 1 && now - vp->updateTime > SALVAGE_INTERVAL) {
6802 V_dontSalvage(vp) = DONT_SALVAGE;
6803 VUpdateVolume_r(&error, vp, 0); /* No need to fsync--not critical */
6811 #ifndef AFS_PTHREAD_ENV
6813 #endif /* !AFS_PTHREAD_ENV */
6815 nUpdatedVolumes -= gap;
6817 #endif /* !AFS_DEMAND_ATTACH_FS */
6820 /***************************************************/
6821 /* Volume LRU routines */
6822 /***************************************************/
6827 * with demand attach fs, we attempt to soft detach(1)
6828 * volumes which have not been accessed in a long time
6829 * in order to speed up fileserver shutdown
6831 * (1) by soft detach we mean a process very similar
6832 * to VOffline, except the final state of the
6833 * Volume will be VOL_STATE_PREATTACHED, instead
6834 * of the usual VOL_STATE_UNATTACHED
6836 #ifdef AFS_DEMAND_ATTACH_FS
6838 /* implementation is reminiscent of a generational GC
6840 * queue 0 is newly attached volumes. this queue is
6841 * sorted by attach timestamp
6843 * queue 1 is volumes that have been around a bit
6844 * longer than queue 0. this queue is sorted by
6847 * queue 2 is volumes tha have been around the longest.
6848 * this queue is unsorted
6850 * queue 3 is volumes that have been marked as
6851 * candidates for soft detachment. this queue is
6854 #define VLRU_GENERATIONS 3 /**< number of generations in VLRU */
6855 #define VLRU_QUEUES 5 /**< total number of VLRU queues */
6858 * definition of a VLRU queue.
6861 volatile struct rx_queue q;
6868 * main VLRU data structure.
6871 struct VLRU_q q[VLRU_QUEUES]; /**< VLRU queues */
6874 /** time interval (in seconds) between promotion passes for
6875 * each young generation queue. */
6876 afs_uint32 promotion_interval[VLRU_GENERATIONS-1];
6878 /** time interval (in seconds) between soft detach candidate
6879 * scans for each generation queue.
6881 * scan_interval[VLRU_QUEUE_CANDIDATE] defines how frequently
6882 * we perform a soft detach pass. */
6883 afs_uint32 scan_interval[VLRU_GENERATIONS+1];
6885 /* scheduler state */
6886 int next_idx; /**< next queue to receive attention */
6887 afs_uint32 last_promotion[VLRU_GENERATIONS-1]; /**< timestamp of last promotion scan */
6888 afs_uint32 last_scan[VLRU_GENERATIONS+1]; /**< timestamp of last detach scan */
6890 int scanner_state; /**< state of scanner thread */
6891 pthread_cond_t cv; /**< state transition CV */
6894 /** global VLRU state */
6895 static struct VLRU volume_LRU;
6898 * defined states for VLRU scanner thread.
6901 VLRU_SCANNER_STATE_OFFLINE = 0, /**< vlru scanner thread is offline */
6902 VLRU_SCANNER_STATE_ONLINE = 1, /**< vlru scanner thread is online */
6903 VLRU_SCANNER_STATE_SHUTTING_DOWN = 2, /**< vlru scanner thread is shutting down */
6904 VLRU_SCANNER_STATE_PAUSING = 3, /**< vlru scanner thread is getting ready to pause */
6905 VLRU_SCANNER_STATE_PAUSED = 4 /**< vlru scanner thread is paused */
6906 } vlru_thread_state_t;
6908 /* vlru disk data header stuff */
6909 #define VLRU_DISK_MAGIC 0x7a8b9cad /**< vlru disk entry magic number */
6910 #define VLRU_DISK_VERSION 1 /**< vlru disk entry version number */
6912 /** vlru default expiration time (for eventual fs state serialization of vlru data) */
6913 #define VLRU_DUMP_EXPIRATION_TIME (60*60*24*7) /* expire vlru data after 1 week */
6916 /** minimum volume inactivity (in seconds) before a volume becomes eligible for
6917 * soft detachment. */
6918 static afs_uint32 VLRU_offline_thresh = VLRU_DEFAULT_OFFLINE_THRESH;
6920 /** time interval (in seconds) between VLRU scanner thread soft detach passes. */
6921 static afs_uint32 VLRU_offline_interval = VLRU_DEFAULT_OFFLINE_INTERVAL;
6923 /** maximum number of volumes to soft detach in a VLRU soft detach pass. */
6924 static afs_uint32 VLRU_offline_max = VLRU_DEFAULT_OFFLINE_MAX;
6926 /** VLRU control flag. non-zero value implies VLRU subsystem is activated. */
6927 static afs_uint32 VLRU_enabled = 1;
6929 /* queue synchronization routines */
6930 static void VLRU_BeginExclusive_r(struct VLRU_q * q);
6931 static void VLRU_EndExclusive_r(struct VLRU_q * q);
6932 static void VLRU_Wait_r(struct VLRU_q * q);
6935 * set VLRU subsystem tunable parameters.
6937 * @param[in] option tunable option to modify
6938 * @param[in] val new value for tunable parameter
6940 * @pre @c VInitVolumePackage2 has not yet been called.
6942 * @post tunable parameter is modified
6946 * @note valid option parameters are:
6947 * @arg @c VLRU_SET_THRESH
6948 * set the period of inactivity after which
6949 * volumes are eligible for soft detachment
6950 * @arg @c VLRU_SET_INTERVAL
6951 * set the time interval between calls
6952 * to the volume LRU "garbage collector"
6953 * @arg @c VLRU_SET_MAX
6954 * set the max number of volumes to deallocate
6958 VLRU_SetOptions(int option, afs_uint32 val)
6960 if (option == VLRU_SET_THRESH) {
6961 VLRU_offline_thresh = val;
6962 } else if (option == VLRU_SET_INTERVAL) {
6963 VLRU_offline_interval = val;
6964 } else if (option == VLRU_SET_MAX) {
6965 VLRU_offline_max = val;
6966 } else if (option == VLRU_SET_ENABLED) {
6969 VLRU_ComputeConstants();
6973 * compute VLRU internal timing parameters.
6975 * @post VLRU scanner thread internal timing parameters are computed
6977 * @note computes internal timing parameters based upon user-modifiable
6978 * tunable parameters.
6982 * @internal volume package internal use only.
6985 VLRU_ComputeConstants(void)
6987 afs_uint32 factor = VLRU_offline_thresh / VLRU_offline_interval;
6989 /* compute the candidate scan interval */
6990 volume_LRU.scan_interval[VLRU_QUEUE_CANDIDATE] = VLRU_offline_interval;
6992 /* compute the promotion intervals */
6993 volume_LRU.promotion_interval[VLRU_QUEUE_NEW] = VLRU_offline_thresh * 2;
6994 volume_LRU.promotion_interval[VLRU_QUEUE_MID] = VLRU_offline_thresh * 4;
6997 /* compute the gen 0 scan interval */
6998 volume_LRU.scan_interval[VLRU_QUEUE_NEW] = VLRU_offline_thresh / 8;
7000 /* compute the gen 0 scan interval */
7001 volume_LRU.scan_interval[VLRU_QUEUE_NEW] = VLRU_offline_interval * 2;
7006 * initialize VLRU subsystem.
7008 * @pre this function has not yet been called
7010 * @post VLRU subsystem is initialized and VLRU scanner thread is starting
7014 * @internal volume package internal use only.
7020 pthread_attr_t attrs;
7023 if (!VLRU_enabled) {
7024 Log("VLRU: disabled\n");
7028 /* initialize each of the VLRU queues */
7029 for (i = 0; i < VLRU_QUEUES; i++) {
7030 queue_Init(&volume_LRU.q[i]);
7031 volume_LRU.q[i].len = 0;
7032 volume_LRU.q[i].busy = 0;
7033 CV_INIT(&volume_LRU.q[i].cv, "vol lru", CV_DEFAULT, 0);
7036 /* setup the timing constants */
7037 VLRU_ComputeConstants();
7039 /* XXX put inside LogLevel check? */
7040 Log("VLRU: starting scanner with the following configuration parameters:\n");
7041 Log("VLRU: offlining volumes after minimum of %d seconds of inactivity\n", VLRU_offline_thresh);
7042 Log("VLRU: running VLRU soft detach pass every %d seconds\n", VLRU_offline_interval);
7043 Log("VLRU: taking up to %d volumes offline per pass\n", VLRU_offline_max);
7044 Log("VLRU: scanning generation 0 for inactive volumes every %d seconds\n", volume_LRU.scan_interval[0]);
7045 Log("VLRU: scanning for promotion/demotion between generations 0 and 1 every %d seconds\n", volume_LRU.promotion_interval[0]);
7046 Log("VLRU: scanning for promotion/demotion between generations 1 and 2 every %d seconds\n", volume_LRU.promotion_interval[1]);
7048 /* start up the VLRU scanner */
7049 volume_LRU.scanner_state = VLRU_SCANNER_STATE_OFFLINE;
7050 if (programType == fileServer) {
7051 CV_INIT(&volume_LRU.cv, "vol lru", CV_DEFAULT, 0);
7052 osi_Assert(pthread_attr_init(&attrs) == 0);
7053 osi_Assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
7054 osi_Assert(pthread_create(&tid, &attrs, &VLRU_ScannerThread, NULL) == 0);
7059 * initialize the VLRU-related fields of a newly allocated volume object.
7061 * @param[in] vp pointer to volume object
7064 * @arg @c VOL_LOCK is held.
7065 * @arg volume object is not on a VLRU queue.
7067 * @post VLRU fields are initialized to indicate that volume object is not
7068 * currently registered with the VLRU subsystem
7072 * @internal volume package interal use only.
7075 VLRU_Init_Node_r(Volume * vp)
7080 osi_Assert(queue_IsNotOnQueue(&vp->vlru));
7081 vp->vlru.idx = VLRU_QUEUE_INVALID;
7085 * add a volume object to a VLRU queue.
7087 * @param[in] vp pointer to volume object
7090 * @arg @c VOL_LOCK is held.
7091 * @arg caller MUST hold a lightweight ref on @p vp.
7092 * @arg caller MUST NOT hold exclusive ownership of the VLRU queue.
7094 * @post the volume object is added to the appropriate VLRU queue
7096 * @note if @c vp->vlru.idx contains the index of a valid VLRU queue,
7097 * then the volume is added to that queue. Otherwise, the value
7098 * @c VLRU_QUEUE_NEW is stored into @c vp->vlru.idx and the
7099 * volume is added to the NEW generation queue.
7101 * @note @c VOL_LOCK may be dropped internally
7103 * @note Volume state is temporarily set to @c VOL_STATE_VLRU_ADD
7104 * during the add operation, and is restored to the previous
7105 * state prior to return.
7109 * @internal volume package internal use only.
7112 VLRU_Add_r(Volume * vp)
7115 VolState state_save;
7120 if (queue_IsOnQueue(&vp->vlru))
7123 state_save = VChangeState_r(vp, VOL_STATE_VLRU_ADD);
7126 if ((idx < 0) || (idx >= VLRU_QUEUE_INVALID)) {
7127 idx = VLRU_QUEUE_NEW;
7130 VLRU_Wait_r(&volume_LRU.q[idx]);
7132 /* repeat check since VLRU_Wait_r may have dropped
7134 if (queue_IsNotOnQueue(&vp->vlru)) {
7136 queue_Prepend(&volume_LRU.q[idx], &vp->vlru);
7137 volume_LRU.q[idx].len++;
7138 V_attachFlags(vp) |= VOL_ON_VLRU;
7139 vp->stats.last_promote = FT_ApproxTime();
7142 VChangeState_r(vp, state_save);
7146 * delete a volume object from a VLRU queue.
7148 * @param[in] vp pointer to volume object
7151 * @arg @c VOL_LOCK is held.
7152 * @arg caller MUST hold a lightweight ref on @p vp.
7153 * @arg caller MUST NOT hold exclusive ownership of the VLRU queue.
7155 * @post volume object is removed from the VLRU queue
7157 * @note @c VOL_LOCK may be dropped internally
7161 * @todo We should probably set volume state to something exlcusive
7162 * (as @c VLRU_Add_r does) prior to dropping @c VOL_LOCK.
7164 * @internal volume package internal use only.
7167 VLRU_Delete_r(Volume * vp)
7174 if (queue_IsNotOnQueue(&vp->vlru))
7180 if (idx == VLRU_QUEUE_INVALID)
7182 VLRU_Wait_r(&volume_LRU.q[idx]);
7183 } while (idx != vp->vlru.idx);
7185 /* now remove from the VLRU and update
7186 * the appropriate counter */
7187 queue_Remove(&vp->vlru);
7188 volume_LRU.q[idx].len--;
7189 vp->vlru.idx = VLRU_QUEUE_INVALID;
7190 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
7194 * tell the VLRU subsystem that a volume was just accessed.
7196 * @param[in] vp pointer to volume object
7199 * @arg @c VOL_LOCK is held
7200 * @arg caller MUST hold a lightweight ref on @p vp
7201 * @arg caller MUST NOT hold exclusive ownership of any VLRU queue
7203 * @post volume VLRU access statistics are updated. If the volume was on
7204 * the VLRU soft detach candidate queue, it is moved to the NEW
7207 * @note @c VOL_LOCK may be dropped internally
7211 * @internal volume package internal use only.
7214 VLRU_UpdateAccess_r(Volume * vp)
7216 Volume * rvp = NULL;
7221 if (queue_IsNotOnQueue(&vp->vlru))
7224 osi_Assert(V_attachFlags(vp) & VOL_ON_VLRU);
7226 /* update the access timestamp */
7227 vp->stats.last_get = FT_ApproxTime();
7230 * if the volume is on the soft detach candidate
7231 * list, we need to safely move it back to a
7232 * regular generation. this has to be done
7233 * carefully so we don't race against the scanner
7237 /* if this volume is on the soft detach candidate queue,
7238 * then grab exclusive access to the necessary queues */
7239 if (vp->vlru.idx == VLRU_QUEUE_CANDIDATE) {
7241 VCreateReservation_r(rvp);
7243 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
7244 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
7245 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7246 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7249 /* make sure multiple threads don't race to update */
7250 if (vp->vlru.idx == VLRU_QUEUE_CANDIDATE) {
7251 VLRU_SwitchQueues(vp, VLRU_QUEUE_NEW, 1);
7255 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7256 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
7257 VCancelReservation_r(rvp);
7262 * switch a volume between two VLRU queues.
7264 * @param[in] vp pointer to volume object
7265 * @param[in] new_idx index of VLRU queue onto which the volume will be moved
7266 * @param[in] append controls whether the volume will be appended or
7267 * prepended to the queue. A nonzero value means it will
7268 * be appended; zero means it will be prepended.
7270 * @pre The new (and old, if applicable) queue(s) must either be owned
7271 * exclusively by the calling thread for asynchronous manipulation,
7272 * or the queue(s) must be quiescent and VOL_LOCK must be held.
7273 * Please see VLRU_BeginExclusive_r, VLRU_EndExclusive_r and VLRU_Wait_r
7274 * for further details of the queue asynchronous processing mechanism.
7276 * @post If the volume object was already on a VLRU queue, it is
7277 * removed from the queue. Depending on the value of the append
7278 * parameter, the volume object is either appended or prepended
7279 * to the VLRU queue referenced by the new_idx parameter.
7283 * @see VLRU_BeginExclusive_r
7284 * @see VLRU_EndExclusive_r
7287 * @internal volume package internal use only.
7290 VLRU_SwitchQueues(Volume * vp, int new_idx, int append)
7292 if (queue_IsNotOnQueue(&vp->vlru))
7295 queue_Remove(&vp->vlru);
7296 volume_LRU.q[vp->vlru.idx].len--;
7298 /* put the volume back on the correct generational queue */
7300 queue_Append(&volume_LRU.q[new_idx], &vp->vlru);
7302 queue_Prepend(&volume_LRU.q[new_idx], &vp->vlru);
7305 volume_LRU.q[new_idx].len++;
7306 vp->vlru.idx = new_idx;
7310 * VLRU background thread.
7312 * The VLRU Scanner Thread is responsible for periodically scanning through
7313 * each VLRU queue looking for volumes which should be moved to another
7314 * queue, or soft detached.
7316 * @param[in] args unused thread arguments parameter
7318 * @return unused thread return value
7319 * @retval NULL always
7321 * @internal volume package internal use only.
7324 VLRU_ScannerThread(void * args)
7326 afs_uint32 now, min_delay, delay;
7327 int i, min_idx, min_op, overdue, state;
7329 /* set t=0 for promotion cycle to be
7330 * fileserver startup */
7331 now = FT_ApproxTime();
7332 for (i=0; i < VLRU_GENERATIONS-1; i++) {
7333 volume_LRU.last_promotion[i] = now;
7336 /* don't start the scanner until VLRU_offline_thresh
7337 * plus a small delay for VInitVolumePackage2 to finish
7340 sleep(VLRU_offline_thresh + 60);
7342 /* set t=0 for scan cycle to be now */
7343 now = FT_ApproxTime();
7344 for (i=0; i < VLRU_GENERATIONS+1; i++) {
7345 volume_LRU.last_scan[i] = now;
7349 if (volume_LRU.scanner_state == VLRU_SCANNER_STATE_OFFLINE) {
7350 volume_LRU.scanner_state = VLRU_SCANNER_STATE_ONLINE;
7353 while ((state = volume_LRU.scanner_state) != VLRU_SCANNER_STATE_SHUTTING_DOWN) {
7354 /* check to see if we've been asked to pause */
7355 if (volume_LRU.scanner_state == VLRU_SCANNER_STATE_PAUSING) {
7356 volume_LRU.scanner_state = VLRU_SCANNER_STATE_PAUSED;
7357 CV_BROADCAST(&volume_LRU.cv);
7359 VOL_CV_WAIT(&volume_LRU.cv);
7360 } while (volume_LRU.scanner_state == VLRU_SCANNER_STATE_PAUSED);
7363 /* scheduling can happen outside the glock */
7366 /* figure out what is next on the schedule */
7368 /* figure out a potential schedule for the new generation first */
7370 min_delay = volume_LRU.scan_interval[0] + volume_LRU.last_scan[0] - now;
7373 if (min_delay > volume_LRU.scan_interval[0]) {
7374 /* unsigned overflow -- we're overdue to run this scan */
7379 /* if we're not overdue for gen 0, figure out schedule for candidate gen */
7381 i = VLRU_QUEUE_CANDIDATE;
7382 delay = volume_LRU.scan_interval[i] + volume_LRU.last_scan[i] - now;
7383 if (delay < min_delay) {
7387 if (delay > volume_LRU.scan_interval[i]) {
7388 /* unsigned overflow -- we're overdue to run this scan */
7395 /* if we're still not overdue for something, figure out schedules for promotions */
7396 for (i=0; !overdue && i < VLRU_GENERATIONS-1; i++) {
7397 delay = volume_LRU.promotion_interval[i] + volume_LRU.last_promotion[i] - now;
7398 if (delay < min_delay) {
7403 if (delay > volume_LRU.promotion_interval[i]) {
7404 /* unsigned overflow -- we're overdue to run this promotion */
7413 /* sleep as needed */
7418 /* do whatever is next */
7421 VLRU_Promote_r(min_idx);
7422 VLRU_Demote_r(min_idx+1);
7424 VLRU_Scan_r(min_idx);
7426 now = FT_ApproxTime();
7429 Log("VLRU scanner asked to go offline (scanner_state=%d)\n", state);
7431 /* signal that scanner is down */
7432 volume_LRU.scanner_state = VLRU_SCANNER_STATE_OFFLINE;
7433 CV_BROADCAST(&volume_LRU.cv);
7439 * promote volumes from one VLRU generation to the next.
7441 * This routine scans a VLRU generation looking for volumes which are
7442 * eligible to be promoted to the next generation. All volumes which
7443 * meet the eligibility requirement are promoted.
7445 * Promotion eligibility is based upon meeting both of the following
7448 * @arg The volume has been accessed since the last promotion:
7449 * @c (vp->stats.last_get >= vp->stats.last_promote)
7450 * @arg The last promotion occurred at least
7451 * @c volume_LRU.promotion_interval[idx] seconds ago
7453 * As a performance optimization, promotions are "globbed". In other
7454 * words, we promote arbitrarily large contiguous sublists of elements
7457 * @param[in] idx VLRU queue index to scan
7461 * @internal VLRU internal use only.
7464 VLRU_Promote_r(int idx)
7466 int len, chaining, promote;
7467 afs_uint32 now, thresh;
7468 struct rx_queue *qp, *nqp;
7469 Volume * vp, *start = NULL, *end = NULL;
7471 /* get exclusive access to two chains, and drop the glock */
7472 VLRU_Wait_r(&volume_LRU.q[idx]);
7473 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
7474 VLRU_Wait_r(&volume_LRU.q[idx+1]);
7475 VLRU_BeginExclusive_r(&volume_LRU.q[idx+1]);
7478 thresh = volume_LRU.promotion_interval[idx];
7479 now = FT_ApproxTime();
7482 for (queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
7483 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7484 promote = (((vp->stats.last_promote + thresh) <= now) &&
7485 (vp->stats.last_get >= vp->stats.last_promote));
7493 /* promote and prepend chain */
7494 queue_MoveChainAfter(&volume_LRU.q[idx+1], &start->vlru, &end->vlru);
7508 /* promote and prepend */
7509 queue_MoveChainAfter(&volume_LRU.q[idx+1], &start->vlru, &end->vlru);
7513 volume_LRU.q[idx].len -= len;
7514 volume_LRU.q[idx+1].len += len;
7517 /* release exclusive access to the two chains */
7519 volume_LRU.last_promotion[idx] = now;
7520 VLRU_EndExclusive_r(&volume_LRU.q[idx+1]);
7521 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
7524 /* run the demotions */
7526 VLRU_Demote_r(int idx)
7529 int len, chaining, demote;
7530 afs_uint32 now, thresh;
7531 struct rx_queue *qp, *nqp;
7532 Volume * vp, *start = NULL, *end = NULL;
7533 Volume ** salv_flag_vec = NULL;
7534 int salv_vec_offset = 0;
7536 osi_Assert(idx == VLRU_QUEUE_MID || idx == VLRU_QUEUE_OLD);
7538 /* get exclusive access to two chains, and drop the glock */
7539 VLRU_Wait_r(&volume_LRU.q[idx-1]);
7540 VLRU_BeginExclusive_r(&volume_LRU.q[idx-1]);
7541 VLRU_Wait_r(&volume_LRU.q[idx]);
7542 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
7545 /* no big deal if this allocation fails */
7546 if (volume_LRU.q[idx].len) {
7547 salv_flag_vec = (Volume **) malloc(volume_LRU.q[idx].len * sizeof(Volume *));
7550 now = FT_ApproxTime();
7551 thresh = volume_LRU.promotion_interval[idx-1];
7554 for (queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
7555 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7556 demote = (((vp->stats.last_promote + thresh) <= now) &&
7557 (vp->stats.last_get < (now - thresh)));
7559 /* we now do volume update list DONT_SALVAGE flag setting during
7560 * demotion passes */
7561 if (salv_flag_vec &&
7562 !(V_attachFlags(vp) & VOL_HDR_DONTSALV) &&
7564 (vp->updateTime < (now - SALVAGE_INTERVAL)) &&
7565 (V_attachState(vp) == VOL_STATE_ATTACHED)) {
7566 salv_flag_vec[salv_vec_offset++] = vp;
7567 VCreateReservation_r(vp);
7576 /* demote and append chain */
7577 queue_MoveChainBefore(&volume_LRU.q[idx-1], &start->vlru, &end->vlru);
7591 queue_MoveChainBefore(&volume_LRU.q[idx-1], &start->vlru, &end->vlru);
7595 volume_LRU.q[idx].len -= len;
7596 volume_LRU.q[idx-1].len += len;
7599 /* release exclusive access to the two chains */
7601 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
7602 VLRU_EndExclusive_r(&volume_LRU.q[idx-1]);
7604 /* now go back and set the DONT_SALVAGE flags as appropriate */
7605 if (salv_flag_vec) {
7607 for (i = 0; i < salv_vec_offset; i++) {
7608 vp = salv_flag_vec[i];
7609 if (!(V_attachFlags(vp) & VOL_HDR_DONTSALV) &&
7610 (vp->updateTime < (now - SALVAGE_INTERVAL)) &&
7611 (V_attachState(vp) == VOL_STATE_ATTACHED)) {
7614 V_attachFlags(vp) |= VOL_HDR_DONTSALV;
7615 V_dontSalvage(vp) = DONT_SALVAGE;
7616 VUpdateVolume_r(&ec, vp, 0);
7620 VCancelReservation_r(vp);
7622 free(salv_flag_vec);
7626 /* run a pass of the VLRU GC scanner */
7628 VLRU_Scan_r(int idx)
7630 afs_uint32 now, thresh;
7631 struct rx_queue *qp, *nqp;
7635 osi_Assert(idx == VLRU_QUEUE_NEW || idx == VLRU_QUEUE_CANDIDATE);
7637 /* gain exclusive access to the idx VLRU */
7638 VLRU_Wait_r(&volume_LRU.q[idx]);
7639 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
7641 if (idx != VLRU_QUEUE_CANDIDATE) {
7642 /* gain exclusive access to the candidate VLRU */
7643 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7644 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7647 now = FT_ApproxTime();
7648 thresh = now - VLRU_offline_thresh;
7650 /* perform candidate selection and soft detaching */
7651 if (idx == VLRU_QUEUE_CANDIDATE) {
7652 /* soft detach some volumes from the candidate pool */
7656 for (i=0,queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
7657 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7658 if (i >= VLRU_offline_max) {
7661 /* check timestamp to see if it's a candidate for soft detaching */
7662 if (vp->stats.last_get <= thresh) {
7664 if (VCheckSoftDetach(vp, thresh))
7670 /* scan for volumes to become soft detach candidates */
7671 for (i=1,queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue),i++) {
7672 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7674 /* check timestamp to see if it's a candidate for soft detaching */
7675 if (vp->stats.last_get <= thresh) {
7676 VCheckSoftDetachCandidate(vp, thresh);
7679 if (!(i&0x7f)) { /* lock coarsening optimization */
7687 /* relinquish exclusive access to the VLRU chains */
7691 volume_LRU.last_scan[idx] = now;
7692 if (idx != VLRU_QUEUE_CANDIDATE) {
7693 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7695 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
7698 /* check whether volume is safe to soft detach
7699 * caller MUST NOT hold a ref count on vp */
7701 VCheckSoftDetach(Volume * vp, afs_uint32 thresh)
7705 if (vp->nUsers || vp->nWaiters)
7708 if (vp->stats.last_get <= thresh) {
7709 ret = VSoftDetachVolume_r(vp, thresh);
7715 /* check whether volume should be made a
7716 * soft detach candidate */
7718 VCheckSoftDetachCandidate(Volume * vp, afs_uint32 thresh)
7721 if (vp->nUsers || vp->nWaiters)
7726 osi_Assert(idx == VLRU_QUEUE_NEW);
7728 if (vp->stats.last_get <= thresh) {
7729 /* move to candidate pool */
7730 queue_Remove(&vp->vlru);
7731 volume_LRU.q[VLRU_QUEUE_NEW].len--;
7732 queue_Prepend(&volume_LRU.q[VLRU_QUEUE_CANDIDATE], &vp->vlru);
7733 vp->vlru.idx = VLRU_QUEUE_CANDIDATE;
7734 volume_LRU.q[VLRU_QUEUE_CANDIDATE].len++;
7742 /* begin exclusive access on VLRU */
7744 VLRU_BeginExclusive_r(struct VLRU_q * q)
7746 osi_Assert(q->busy == 0);
7750 /* end exclusive access on VLRU */
7752 VLRU_EndExclusive_r(struct VLRU_q * q)
7754 osi_Assert(q->busy);
7756 CV_BROADCAST(&q->cv);
7759 /* wait for another thread to end exclusive access on VLRU */
7761 VLRU_Wait_r(struct VLRU_q * q)
7764 VOL_CV_WAIT(&q->cv);
7769 * volume soft detach
7771 * caller MUST NOT hold a ref count on vp */
7773 VSoftDetachVolume_r(Volume * vp, afs_uint32 thresh)
7778 osi_Assert(vp->vlru.idx == VLRU_QUEUE_CANDIDATE);
7780 ts_save = vp->stats.last_get;
7781 if (ts_save > thresh)
7784 if (vp->nUsers || vp->nWaiters)
7787 if (VIsExclusiveState(V_attachState(vp))) {
7791 switch (V_attachState(vp)) {
7792 case VOL_STATE_UNATTACHED:
7793 case VOL_STATE_PREATTACHED:
7794 case VOL_STATE_ERROR:
7795 case VOL_STATE_GOING_OFFLINE:
7796 case VOL_STATE_SHUTTING_DOWN:
7797 case VOL_STATE_SALVAGING:
7798 case VOL_STATE_DELETED:
7799 volume_LRU.q[vp->vlru.idx].len--;
7801 /* create and cancel a reservation to
7802 * give the volume an opportunity to
7804 VCreateReservation_r(vp);
7805 queue_Remove(&vp->vlru);
7806 vp->vlru.idx = VLRU_QUEUE_INVALID;
7807 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
7808 VCancelReservation_r(vp);
7814 /* hold the volume and take it offline.
7815 * no need for reservations, as VHold_r
7816 * takes care of that internally. */
7817 if (VHold_r(vp) == 0) {
7818 /* vhold drops the glock, so now we should
7819 * check to make sure we aren't racing against
7820 * other threads. if we are racing, offlining vp
7821 * would be wasteful, and block the scanner for a while
7825 (vp->shuttingDown) ||
7826 (vp->goingOffline) ||
7827 (vp->stats.last_get != ts_save)) {
7828 /* looks like we're racing someone else. bail */
7832 /* pull it off the VLRU */
7833 osi_Assert(vp->vlru.idx == VLRU_QUEUE_CANDIDATE);
7834 volume_LRU.q[VLRU_QUEUE_CANDIDATE].len--;
7835 queue_Remove(&vp->vlru);
7836 vp->vlru.idx = VLRU_QUEUE_INVALID;
7837 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
7839 /* take if offline */
7840 VOffline_r(vp, "volume has been soft detached");
7842 /* invalidate the volume header cache */
7843 FreeVolumeHeader(vp);
7846 IncUInt64(&VStats.soft_detaches);
7847 vp->stats.soft_detaches++;
7849 /* put in pre-attached state so demand
7850 * attacher can work on it */
7851 VChangeState_r(vp, VOL_STATE_PREATTACHED);
7857 #endif /* AFS_DEMAND_ATTACH_FS */
7860 /***************************************************/
7861 /* Volume Header Cache routines */
7862 /***************************************************/
7865 * volume header cache.
7867 struct volume_hdr_LRU_t volume_hdr_LRU;
7870 * initialize the volume header cache.
7872 * @param[in] howMany number of header cache entries to preallocate
7874 * @pre VOL_LOCK held. Function has never been called before.
7876 * @post howMany cache entries are allocated, initialized, and added
7877 * to the LRU list. Header cache statistics are initialized.
7879 * @note only applicable to fileServer program type. Should only be
7880 * called once during volume package initialization.
7882 * @internal volume package internal use only.
7885 VInitVolumeHeaderCache(afs_uint32 howMany)
7887 struct volHeader *hp;
7888 if (programType != fileServer)
7890 queue_Init(&volume_hdr_LRU);
7891 volume_hdr_LRU.stats.free = 0;
7892 volume_hdr_LRU.stats.used = howMany;
7893 volume_hdr_LRU.stats.attached = 0;
7894 hp = (struct volHeader *)(calloc(howMany, sizeof(struct volHeader)));
7895 osi_Assert(hp != NULL);
7898 /* We are using ReleaseVolumeHeader to initialize the values on the header list
7899 * to ensure they have the right values
7901 ReleaseVolumeHeader(hp++);
7905 * get a volume header and attach it to the volume object.
7907 * @param[in] vp pointer to volume object
7909 * @return cache entry status
7910 * @retval 0 volume header was newly attached; cache data is invalid
7911 * @retval 1 volume header was previously attached; cache data is valid
7913 * @pre VOL_LOCK held. For DAFS, lightweight ref must be held on volume object.
7915 * @post volume header attached to volume object. if necessary, header cache
7916 * entry on LRU is synchronized to disk. Header is removed from LRU list.
7918 * @note VOL_LOCK may be dropped
7920 * @warning this interface does not load header data from disk. it merely
7921 * attaches a header object to the volume object, and may sync the old
7922 * header cache data out to disk in the process.
7924 * @internal volume package internal use only.
7927 GetVolumeHeader(Volume * vp)
7930 struct volHeader *hd;
7932 static int everLogged = 0;
7934 #ifdef AFS_DEMAND_ATTACH_FS
7935 VolState vp_save = 0, back_save = 0;
7937 /* XXX debug 9/19/05 we've apparently got
7938 * a ref counting bug somewhere that's
7939 * breaking the nUsers == 0 => header on LRU
7941 if (vp->header && queue_IsNotOnQueue(vp->header)) {
7942 Log("nUsers == 0, but header not on LRU\n");
7947 old = (vp->header != NULL); /* old == volume already has a header */
7949 if (programType != fileServer) {
7950 /* for volume utilities, we allocate volHeaders as needed */
7952 hd = (struct volHeader *)calloc(1, sizeof(*vp->header));
7953 osi_Assert(hd != NULL);
7956 #ifdef AFS_DEMAND_ATTACH_FS
7957 V_attachFlags(vp) |= VOL_HDR_ATTACHED;
7961 /* for the fileserver, we keep a volume header cache */
7963 /* the header we previously dropped in the lru is
7964 * still available. pull it off the lru and return */
7967 osi_Assert(hd->back == vp);
7968 #ifdef AFS_DEMAND_ATTACH_FS
7969 V_attachFlags(vp) &= ~(VOL_HDR_IN_LRU);
7972 /* we need to grab a new element off the LRU */
7973 if (queue_IsNotEmpty(&volume_hdr_LRU)) {
7974 /* grab an element and pull off of LRU */
7975 hd = queue_First(&volume_hdr_LRU, volHeader);
7978 /* LRU is empty, so allocate a new volHeader
7979 * this is probably indicative of a leak, so let the user know */
7980 hd = (struct volHeader *)calloc(1, sizeof(struct volHeader));
7981 osi_Assert(hd != NULL);
7983 Log("****Allocated more volume headers, probably leak****\n");
7986 volume_hdr_LRU.stats.free++;
7989 /* this header used to belong to someone else.
7990 * we'll need to check if the header needs to
7991 * be sync'd out to disk */
7993 #ifdef AFS_DEMAND_ATTACH_FS
7994 /* if hd->back were in an exclusive state, then
7995 * its volHeader would not be on the LRU... */
7996 osi_Assert(!VIsExclusiveState(V_attachState(hd->back)));
7999 if (hd->diskstuff.inUse) {
8000 /* volume was in use, so we'll need to sync
8001 * its header to disk */
8003 #ifdef AFS_DEMAND_ATTACH_FS
8004 back_save = VChangeState_r(hd->back, VOL_STATE_UPDATING);
8005 vp_save = VChangeState_r(vp, VOL_STATE_HDR_ATTACHING);
8006 VCreateReservation_r(hd->back);
8010 WriteVolumeHeader_r(&error, hd->back);
8011 /* Ignore errors; catch them later */
8013 #ifdef AFS_DEMAND_ATTACH_FS
8018 hd->back->header = NULL;
8019 #ifdef AFS_DEMAND_ATTACH_FS
8020 V_attachFlags(hd->back) &= ~(VOL_HDR_ATTACHED | VOL_HDR_LOADED | VOL_HDR_IN_LRU);
8022 if (hd->diskstuff.inUse) {
8023 VChangeState_r(hd->back, back_save);
8024 VCancelReservation_r(hd->back);
8025 VChangeState_r(vp, vp_save);
8029 volume_hdr_LRU.stats.attached++;
8033 #ifdef AFS_DEMAND_ATTACH_FS
8034 V_attachFlags(vp) |= VOL_HDR_ATTACHED;
8037 volume_hdr_LRU.stats.free--;
8038 volume_hdr_LRU.stats.used++;
8040 IncUInt64(&VStats.hdr_gets);
8041 #ifdef AFS_DEMAND_ATTACH_FS
8042 IncUInt64(&vp->stats.hdr_gets);
8043 vp->stats.last_hdr_get = FT_ApproxTime();
8050 * make sure volume header is attached and contains valid cache data.
8052 * @param[out] ec outbound error code
8053 * @param[in] vp pointer to volume object
8055 * @pre VOL_LOCK held. For DAFS, lightweight ref held on vp.
8057 * @post header cache entry attached, and loaded with valid data, or
8058 * *ec is nonzero, and the header is released back into the LRU.
8060 * @internal volume package internal use only.
8063 LoadVolumeHeader(Error * ec, Volume * vp)
8065 #ifdef AFS_DEMAND_ATTACH_FS
8066 VolState state_save;
8070 if (vp->nUsers == 0 && !GetVolumeHeader(vp)) {
8071 IncUInt64(&VStats.hdr_loads);
8072 state_save = VChangeState_r(vp, VOL_STATE_HDR_LOADING);
8075 ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
8076 sizeof(V_disk(vp)), VOLUMEINFOMAGIC,
8078 IncUInt64(&vp->stats.hdr_loads);
8079 now = FT_ApproxTime();
8083 V_attachFlags(vp) |= VOL_HDR_LOADED;
8084 vp->stats.last_hdr_load = now;
8086 VChangeState_r(vp, state_save);
8088 #else /* AFS_DEMAND_ATTACH_FS */
8090 if (vp->nUsers == 0 && !GetVolumeHeader(vp)) {
8091 IncUInt64(&VStats.hdr_loads);
8093 ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
8094 sizeof(V_disk(vp)), VOLUMEINFOMAGIC,
8097 #endif /* AFS_DEMAND_ATTACH_FS */
8099 /* maintain (nUsers==0) => header in LRU invariant */
8100 FreeVolumeHeader(vp);
8105 * release a header cache entry back into the LRU list.
8107 * @param[in] hd pointer to volume header cache object
8109 * @pre VOL_LOCK held.
8111 * @post header cache object appended onto end of LRU list.
8113 * @note only applicable to fileServer program type.
8115 * @note used to place a header cache entry back into the
8116 * LRU pool without invalidating it as a cache entry.
8118 * @internal volume package internal use only.
8121 ReleaseVolumeHeader(struct volHeader *hd)
8123 if (programType != fileServer)
8125 if (!hd || queue_IsOnQueue(hd)) /* no header, or header already released */
8127 queue_Append(&volume_hdr_LRU, hd);
8128 #ifdef AFS_DEMAND_ATTACH_FS
8130 V_attachFlags(hd->back) |= VOL_HDR_IN_LRU;
8133 volume_hdr_LRU.stats.free++;
8134 volume_hdr_LRU.stats.used--;
8138 * free/invalidate a volume header cache entry.
8140 * @param[in] vp pointer to volume object
8142 * @pre VOL_LOCK is held.
8144 * @post For fileserver, header cache entry is returned to LRU, and it is
8145 * invalidated as a cache entry. For volume utilities, the header
8146 * cache entry is freed.
8148 * @note For fileserver, this should be utilized instead of ReleaseVolumeHeader
8149 * whenever it is necessary to invalidate the header cache entry.
8151 * @see ReleaseVolumeHeader
8153 * @internal volume package internal use only.
8156 FreeVolumeHeader(Volume * vp)
8158 struct volHeader *hd = vp->header;
8161 if (programType == fileServer) {
8162 ReleaseVolumeHeader(hd);
8167 #ifdef AFS_DEMAND_ATTACH_FS
8168 V_attachFlags(vp) &= ~(VOL_HDR_ATTACHED | VOL_HDR_IN_LRU | VOL_HDR_LOADED);
8170 volume_hdr_LRU.stats.attached--;
8175 /***************************************************/
8176 /* Volume Hash Table routines */
8177 /***************************************************/
8180 * set size of volume object hash table.
8182 * @param[in] logsize log(2) of desired hash table size
8184 * @return operation status
8186 * @retval -1 failure
8188 * @pre MUST be called prior to VInitVolumePackage2
8190 * @post Volume Hash Table will have 2^logsize buckets
8193 VSetVolHashSize(int logsize)
8195 /* 64 to 268435456 hash buckets seems like a reasonable range */
8196 if ((logsize < 6 ) || (logsize > 28)) {
8201 VolumeHashTable.Size = 1 << logsize;
8202 VolumeHashTable.Mask = VolumeHashTable.Size - 1;
8204 /* we can't yet support runtime modification of this
8205 * parameter. we'll need a configuration rwlock to
8206 * make runtime modification feasible.... */
8213 * initialize dynamic data structures for volume hash table.
8215 * @post hash table is allocated, and fields are initialized.
8217 * @internal volume package internal use only.
8220 VInitVolumeHash(void)
8224 VolumeHashTable.Table = (VolumeHashChainHead *) calloc(VolumeHashTable.Size,
8225 sizeof(VolumeHashChainHead));
8226 osi_Assert(VolumeHashTable.Table != NULL);
8228 for (i=0; i < VolumeHashTable.Size; i++) {
8229 queue_Init(&VolumeHashTable.Table[i]);
8230 #ifdef AFS_DEMAND_ATTACH_FS
8231 CV_INIT(&VolumeHashTable.Table[i].chain_busy_cv, "vhash busy", CV_DEFAULT, 0);
8232 #endif /* AFS_DEMAND_ATTACH_FS */
8237 * add a volume object to the hash table.
8239 * @param[in] vp pointer to volume object
8240 * @param[in] hashid hash of volume id
8242 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
8245 * @post volume is added to hash chain.
8247 * @internal volume package internal use only.
8249 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
8250 * asynchronous hash chain reordering to finish.
8253 AddVolumeToHashTable(Volume * vp, int hashid)
8255 VolumeHashChainHead * head;
8257 if (queue_IsOnQueue(vp))
8260 head = &VolumeHashTable.Table[VOLUME_HASH(hashid)];
8262 #ifdef AFS_DEMAND_ATTACH_FS
8263 /* wait for the hash chain to become available */
8266 V_attachFlags(vp) |= VOL_IN_HASH;
8267 vp->chainCacheCheck = ++head->cacheCheck;
8268 #endif /* AFS_DEMAND_ATTACH_FS */
8271 vp->hashid = hashid;
8272 queue_Append(head, vp);
8273 vp->vnodeHashOffset = VolumeHashOffset_r();
8277 * delete a volume object from the hash table.
8279 * @param[in] vp pointer to volume object
8281 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
8284 * @post volume is removed from hash chain.
8286 * @internal volume package internal use only.
8288 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
8289 * asynchronous hash chain reordering to finish.
8292 DeleteVolumeFromHashTable(Volume * vp)
8294 VolumeHashChainHead * head;
8296 if (!queue_IsOnQueue(vp))
8299 head = &VolumeHashTable.Table[VOLUME_HASH(vp->hashid)];
8301 #ifdef AFS_DEMAND_ATTACH_FS
8302 /* wait for the hash chain to become available */
8305 V_attachFlags(vp) &= ~(VOL_IN_HASH);
8307 #endif /* AFS_DEMAND_ATTACH_FS */
8311 /* do NOT reset hashid to zero, as the online
8312 * salvager package may need to know the volume id
8313 * after the volume is removed from the hash */
8317 * lookup a volume object in the hash table given a volume id.
8319 * @param[out] ec error code return
8320 * @param[in] volumeId volume id
8321 * @param[in] hint volume object which we believe could be the correct
8324 * @return volume object pointer
8325 * @retval NULL no such volume id is registered with the hash table.
8327 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
8330 * @post volume object with the given id is returned. volume object and
8331 * hash chain access statistics are updated. hash chain may have
8334 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
8335 * asynchronous hash chain reordering operation to finish, or
8336 * in order for us to perform an asynchronous chain reordering.
8338 * @note Hash chain reorderings occur when the access count for the
8339 * volume object being looked up exceeds the sum of the previous
8340 * node's (the node ahead of it in the hash chain linked list)
8341 * access count plus the constant VOLUME_HASH_REORDER_THRESHOLD.
8343 * @note For DAFS, the hint parameter allows us to short-circuit if the
8344 * cacheCheck fields match between the hash chain head and the
8345 * hint volume object.
8348 VLookupVolume_r(Error * ec, VolId volumeId, Volume * hint)
8352 #ifdef AFS_DEMAND_ATTACH_FS
8355 VolumeHashChainHead * head;
8358 head = &VolumeHashTable.Table[VOLUME_HASH(volumeId)];
8360 #ifdef AFS_DEMAND_ATTACH_FS
8361 /* wait for the hash chain to become available */
8364 /* check to see if we can short circuit without walking the hash chain */
8365 if (hint && (hint->chainCacheCheck == head->cacheCheck)) {
8366 IncUInt64(&hint->stats.hash_short_circuits);
8369 #endif /* AFS_DEMAND_ATTACH_FS */
8371 /* someday we need to either do per-chain locks, RWlocks,
8372 * or both for volhash access.
8373 * (and move to a data structure with better cache locality) */
8375 /* search the chain for this volume id */
8376 for(queue_Scan(head, vp, np, Volume)) {
8378 if ((vp->hashid == volumeId)) {
8383 if (queue_IsEnd(head, vp)) {
8387 #ifdef AFS_DEMAND_ATTACH_FS
8388 /* update hash chain statistics */
8391 FillInt64(lks, 0, looks);
8392 AddUInt64(head->looks, lks, &head->looks);
8393 AddUInt64(VStats.hash_looks, lks, &VStats.hash_looks);
8394 IncUInt64(&head->gets);
8399 IncUInt64(&vp->stats.hash_lookups);
8401 /* for demand attach fileserver, we permit occasional hash chain reordering
8402 * so that frequently looked up volumes move towards the head of the chain */
8403 pp = queue_Prev(vp, Volume);
8404 if (!queue_IsEnd(head, pp)) {
8405 FillInt64(thresh, 0, VOLUME_HASH_REORDER_THRESHOLD);
8406 AddUInt64(thresh, pp->stats.hash_lookups, &thresh);
8407 if (GEInt64(vp->stats.hash_lookups, thresh)) {
8408 VReorderHash_r(head, pp, vp);
8412 /* update the short-circuit cache check */
8413 vp->chainCacheCheck = head->cacheCheck;
8415 #endif /* AFS_DEMAND_ATTACH_FS */
8420 #ifdef AFS_DEMAND_ATTACH_FS
8421 /* perform volume hash chain reordering.
8423 * advance a subchain beginning at vp ahead of
8424 * the adjacent subchain ending at pp */
8426 VReorderHash_r(VolumeHashChainHead * head, Volume * pp, Volume * vp)
8428 Volume *tp, *np, *lp;
8429 afs_uint64 move_thresh;
8431 /* this should never be called if the chain is already busy, so
8432 * no need to wait for other exclusive chain ops to finish */
8434 /* this is a rather heavy set of operations,
8435 * so let's set the chain busy flag and drop
8437 VHashBeginExclusive_r(head);
8440 /* scan forward in the chain from vp looking for the last element
8441 * in the chain we want to advance */
8442 FillInt64(move_thresh, 0, VOLUME_HASH_REORDER_CHAIN_THRESH);
8443 AddUInt64(move_thresh, pp->stats.hash_lookups, &move_thresh);
8444 for(queue_ScanFrom(head, vp, tp, np, Volume)) {
8445 if (LTInt64(tp->stats.hash_lookups, move_thresh)) {
8449 lp = queue_Prev(tp, Volume);
8451 /* scan backwards from pp to determine where to splice and
8452 * insert the subchain we're advancing */
8453 for(queue_ScanBackwardsFrom(head, pp, tp, np, Volume)) {
8454 if (GTInt64(tp->stats.hash_lookups, move_thresh)) {
8458 tp = queue_Next(tp, Volume);
8460 /* rebalance chain(vp,...,lp) ahead of chain(tp,...,pp) */
8461 queue_MoveChainBefore(tp,vp,lp);
8464 IncUInt64(&VStats.hash_reorders);
8466 IncUInt64(&head->reorders);
8468 /* wake up any threads waiting for the hash chain */
8469 VHashEndExclusive_r(head);
8473 /* demand-attach fs volume hash
8474 * asynchronous exclusive operations */
8477 * begin an asynchronous exclusive operation on a volume hash chain.
8479 * @param[in] head pointer to volume hash chain head object
8481 * @pre VOL_LOCK held. hash chain is quiescent.
8483 * @post hash chain marked busy.
8485 * @note this interface is used in conjunction with VHashEndExclusive_r and
8486 * VHashWait_r to perform asynchronous (wrt VOL_LOCK) operations on a
8487 * volume hash chain. Its main use case is hash chain reordering, which
8488 * has the potential to be a highly latent operation.
8490 * @see VHashEndExclusive_r
8495 * @internal volume package internal use only.
8498 VHashBeginExclusive_r(VolumeHashChainHead * head)
8500 osi_Assert(head->busy == 0);
8505 * relinquish exclusive ownership of a volume hash chain.
8507 * @param[in] head pointer to volume hash chain head object
8509 * @pre VOL_LOCK held. thread owns the hash chain exclusively.
8511 * @post hash chain is marked quiescent. threads awaiting use of
8512 * chain are awakened.
8514 * @see VHashBeginExclusive_r
8519 * @internal volume package internal use only.
8522 VHashEndExclusive_r(VolumeHashChainHead * head)
8524 osi_Assert(head->busy);
8526 CV_BROADCAST(&head->chain_busy_cv);
8530 * wait for all asynchronous operations on a hash chain to complete.
8532 * @param[in] head pointer to volume hash chain head object
8534 * @pre VOL_LOCK held.
8536 * @post hash chain object is quiescent.
8538 * @see VHashBeginExclusive_r
8539 * @see VHashEndExclusive_r
8543 * @note This interface should be called before any attempt to
8544 * traverse the hash chain. It is permissible for a thread
8545 * to gain exclusive access to the chain, and then perform
8546 * latent operations on the chain asynchronously wrt the
8549 * @warning if waiting is necessary, VOL_LOCK is dropped
8551 * @internal volume package internal use only.
8554 VHashWait_r(VolumeHashChainHead * head)
8556 while (head->busy) {
8557 VOL_CV_WAIT(&head->chain_busy_cv);
8560 #endif /* AFS_DEMAND_ATTACH_FS */
8563 /***************************************************/
8564 /* Volume by Partition List routines */
8565 /***************************************************/
8568 * demand attach fileserver adds a
8569 * linked list of volumes to each
8570 * partition object, thus allowing
8571 * for quick enumeration of all
8572 * volumes on a partition
8575 #ifdef AFS_DEMAND_ATTACH_FS
8577 * add a volume to its disk partition VByPList.
8579 * @param[in] vp pointer to volume object
8581 * @pre either the disk partition VByPList is owned exclusively
8582 * by the calling thread, or the list is quiescent and
8585 * @post volume is added to disk partition VByPList
8589 * @warning it is the caller's responsibility to ensure list
8592 * @see VVByPListWait_r
8593 * @see VVByPListBeginExclusive_r
8594 * @see VVByPListEndExclusive_r
8596 * @internal volume package internal use only.
8599 AddVolumeToVByPList_r(Volume * vp)
8601 if (queue_IsNotOnQueue(&vp->vol_list)) {
8602 queue_Append(&vp->partition->vol_list, &vp->vol_list);
8603 V_attachFlags(vp) |= VOL_ON_VBYP_LIST;
8604 vp->partition->vol_list.len++;
8609 * delete a volume from its disk partition VByPList.
8611 * @param[in] vp pointer to volume object
8613 * @pre either the disk partition VByPList is owned exclusively
8614 * by the calling thread, or the list is quiescent and
8617 * @post volume is removed from the disk partition VByPList
8621 * @warning it is the caller's responsibility to ensure list
8624 * @see VVByPListWait_r
8625 * @see VVByPListBeginExclusive_r
8626 * @see VVByPListEndExclusive_r
8628 * @internal volume package internal use only.
8631 DeleteVolumeFromVByPList_r(Volume * vp)
8633 if (queue_IsOnQueue(&vp->vol_list)) {
8634 queue_Remove(&vp->vol_list);
8635 V_attachFlags(vp) &= ~(VOL_ON_VBYP_LIST);
8636 vp->partition->vol_list.len--;
8641 * begin an asynchronous exclusive operation on a VByPList.
8643 * @param[in] dp pointer to disk partition object
8645 * @pre VOL_LOCK held. VByPList is quiescent.
8647 * @post VByPList marked busy.
8649 * @note this interface is used in conjunction with VVByPListEndExclusive_r and
8650 * VVByPListWait_r to perform asynchronous (wrt VOL_LOCK) operations on a
8653 * @see VVByPListEndExclusive_r
8654 * @see VVByPListWait_r
8658 * @internal volume package internal use only.
8660 /* take exclusive control over the list */
8662 VVByPListBeginExclusive_r(struct DiskPartition64 * dp)
8664 osi_Assert(dp->vol_list.busy == 0);
8665 dp->vol_list.busy = 1;
8669 * relinquish exclusive ownership of a VByPList.
8671 * @param[in] dp pointer to disk partition object
8673 * @pre VOL_LOCK held. thread owns the VByPList exclusively.
8675 * @post VByPList is marked quiescent. threads awaiting use of
8676 * the list are awakened.
8678 * @see VVByPListBeginExclusive_r
8679 * @see VVByPListWait_r
8683 * @internal volume package internal use only.
8686 VVByPListEndExclusive_r(struct DiskPartition64 * dp)
8688 osi_Assert(dp->vol_list.busy);
8689 dp->vol_list.busy = 0;
8690 CV_BROADCAST(&dp->vol_list.cv);
8694 * wait for all asynchronous operations on a VByPList to complete.
8696 * @param[in] dp pointer to disk partition object
8698 * @pre VOL_LOCK is held.
8700 * @post disk partition's VByP list is quiescent
8704 * @note This interface should be called before any attempt to
8705 * traverse the VByPList. It is permissible for a thread
8706 * to gain exclusive access to the list, and then perform
8707 * latent operations on the list asynchronously wrt the
8710 * @warning if waiting is necessary, VOL_LOCK is dropped
8712 * @see VVByPListEndExclusive_r
8713 * @see VVByPListBeginExclusive_r
8715 * @internal volume package internal use only.
8718 VVByPListWait_r(struct DiskPartition64 * dp)
8720 while (dp->vol_list.busy) {
8721 VOL_CV_WAIT(&dp->vol_list.cv);
8724 #endif /* AFS_DEMAND_ATTACH_FS */
8726 /***************************************************/
8727 /* Volume Cache Statistics routines */
8728 /***************************************************/
8731 VPrintCacheStats_r(void)
8733 afs_uint32 get_hi, get_lo, load_hi, load_lo;
8734 struct VnodeClassInfo *vcp;
8735 vcp = &VnodeClassInfo[vLarge];
8736 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);
8737 vcp = &VnodeClassInfo[vSmall];
8738 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);
8739 SplitInt64(VStats.hdr_gets, get_hi, get_lo);
8740 SplitInt64(VStats.hdr_loads, load_hi, load_lo);
8741 Log("Volume header cache, %d entries, %d gets, %d replacements\n",
8742 VStats.hdr_cache_size, get_lo, load_lo);
8746 VPrintCacheStats(void)
8749 VPrintCacheStats_r();
8753 #ifdef AFS_DEMAND_ATTACH_FS
8755 UInt64ToDouble(afs_uint64 * x)
8757 static double c32 = 4.0 * 1.073741824 * 1000000000.0;
8759 SplitInt64(*x, h, l);
8760 return (((double)h) * c32) + ((double) l);
8764 DoubleToPrintable(double x, char * buf, int len)
8766 static double billion = 1000000000.0;
8769 y[0] = (afs_uint32) (x / (billion * billion));
8770 y[1] = (afs_uint32) ((x - (((double)y[0]) * billion * billion)) / billion);
8771 y[2] = (afs_uint32) (x - ((((double)y[0]) * billion * billion) + (((double)y[1]) * billion)));
8774 snprintf(buf, len, "%d%09d%09d", y[0], y[1], y[2]);
8776 snprintf(buf, len, "%d%09d", y[1], y[2]);
8778 snprintf(buf, len, "%d", y[2]);
8784 struct VLRUExtStatsEntry {
8788 struct VLRUExtStats {
8794 } queue_info[VLRU_QUEUE_INVALID];
8795 struct VLRUExtStatsEntry * vec;
8799 * add a 256-entry fudge factor onto the vector in case state changes
8800 * out from under us.
8802 #define VLRU_EXT_STATS_VEC_LEN_FUDGE 256
8805 * collect extended statistics for the VLRU subsystem.
8807 * @param[out] stats pointer to stats structure to be populated
8808 * @param[in] nvols number of volumes currently known to exist
8810 * @pre VOL_LOCK held
8812 * @post stats->vec allocated and populated
8814 * @return operation status
8819 VVLRUExtStats_r(struct VLRUExtStats * stats, afs_uint32 nvols)
8821 afs_uint32 cur, idx, len;
8822 struct rx_queue * qp, * nqp;
8824 struct VLRUExtStatsEntry * vec;
8826 len = nvols + VLRU_EXT_STATS_VEC_LEN_FUDGE;
8827 vec = stats->vec = calloc(len,
8828 sizeof(struct VLRUExtStatsEntry));
8834 for (idx = VLRU_QUEUE_NEW; idx < VLRU_QUEUE_INVALID; idx++) {
8835 VLRU_Wait_r(&volume_LRU.q[idx]);
8836 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
8839 stats->queue_info[idx].start = cur;
8841 for (queue_Scan(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
8843 /* out of space in vec */
8846 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
8847 vec[cur].volid = vp->hashid;
8851 stats->queue_info[idx].len = cur - stats->queue_info[idx].start;
8854 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
8862 #define ENUMTOSTRING(en) #en
8863 #define ENUMCASE(en) \
8864 case en: return ENUMTOSTRING(en)
8867 vlru_idx_to_string(int idx)
8870 ENUMCASE(VLRU_QUEUE_NEW);
8871 ENUMCASE(VLRU_QUEUE_MID);
8872 ENUMCASE(VLRU_QUEUE_OLD);
8873 ENUMCASE(VLRU_QUEUE_CANDIDATE);
8874 ENUMCASE(VLRU_QUEUE_HELD);
8875 ENUMCASE(VLRU_QUEUE_INVALID);
8877 return "**UNKNOWN**";
8882 VPrintExtendedCacheStats_r(int flags)
8885 afs_uint32 vol_sum = 0;
8892 struct stats looks, gets, reorders, len;
8893 struct stats ch_looks, ch_gets, ch_reorders;
8895 VolumeHashChainHead *head;
8897 struct VLRUExtStats vlru_stats;
8899 /* zero out stats */
8900 memset(&looks, 0, sizeof(struct stats));
8901 memset(&gets, 0, sizeof(struct stats));
8902 memset(&reorders, 0, sizeof(struct stats));
8903 memset(&len, 0, sizeof(struct stats));
8904 memset(&ch_looks, 0, sizeof(struct stats));
8905 memset(&ch_gets, 0, sizeof(struct stats));
8906 memset(&ch_reorders, 0, sizeof(struct stats));
8908 for (i = 0; i < VolumeHashTable.Size; i++) {
8909 head = &VolumeHashTable.Table[i];
8912 VHashBeginExclusive_r(head);
8915 ch_looks.sum = UInt64ToDouble(&head->looks);
8916 ch_gets.sum = UInt64ToDouble(&head->gets);
8917 ch_reorders.sum = UInt64ToDouble(&head->reorders);
8919 /* update global statistics */
8921 looks.sum += ch_looks.sum;
8922 gets.sum += ch_gets.sum;
8923 reorders.sum += ch_reorders.sum;
8924 len.sum += (double)head->len;
8925 vol_sum += head->len;
8928 len.min = (double) head->len;
8929 len.max = (double) head->len;
8930 looks.min = ch_looks.sum;
8931 looks.max = ch_looks.sum;
8932 gets.min = ch_gets.sum;
8933 gets.max = ch_gets.sum;
8934 reorders.min = ch_reorders.sum;
8935 reorders.max = ch_reorders.sum;
8937 if (((double)head->len) < len.min)
8938 len.min = (double) head->len;
8939 if (((double)head->len) > len.max)
8940 len.max = (double) head->len;
8941 if (ch_looks.sum < looks.min)
8942 looks.min = ch_looks.sum;
8943 else if (ch_looks.sum > looks.max)
8944 looks.max = ch_looks.sum;
8945 if (ch_gets.sum < gets.min)
8946 gets.min = ch_gets.sum;
8947 else if (ch_gets.sum > gets.max)
8948 gets.max = ch_gets.sum;
8949 if (ch_reorders.sum < reorders.min)
8950 reorders.min = ch_reorders.sum;
8951 else if (ch_reorders.sum > reorders.max)
8952 reorders.max = ch_reorders.sum;
8956 if ((flags & VOL_STATS_PER_CHAIN2) && queue_IsNotEmpty(head)) {
8957 /* compute detailed per-chain stats */
8958 struct stats hdr_loads, hdr_gets;
8959 double v_looks, v_loads, v_gets;
8961 /* initialize stats with data from first element in chain */
8962 vp = queue_First(head, Volume);
8963 v_looks = UInt64ToDouble(&vp->stats.hash_lookups);
8964 v_loads = UInt64ToDouble(&vp->stats.hdr_loads);
8965 v_gets = UInt64ToDouble(&vp->stats.hdr_gets);
8966 ch_gets.min = ch_gets.max = v_looks;
8967 hdr_loads.min = hdr_loads.max = v_loads;
8968 hdr_gets.min = hdr_gets.max = v_gets;
8969 hdr_loads.sum = hdr_gets.sum = 0;
8971 vp = queue_Next(vp, Volume);
8973 /* pull in stats from remaining elements in chain */
8974 for (queue_ScanFrom(head, vp, vp, np, Volume)) {
8975 v_looks = UInt64ToDouble(&vp->stats.hash_lookups);
8976 v_loads = UInt64ToDouble(&vp->stats.hdr_loads);
8977 v_gets = UInt64ToDouble(&vp->stats.hdr_gets);
8979 hdr_loads.sum += v_loads;
8980 hdr_gets.sum += v_gets;
8982 if (v_looks < ch_gets.min)
8983 ch_gets.min = v_looks;
8984 else if (v_looks > ch_gets.max)
8985 ch_gets.max = v_looks;
8987 if (v_loads < hdr_loads.min)
8988 hdr_loads.min = v_loads;
8989 else if (v_loads > hdr_loads.max)
8990 hdr_loads.max = v_loads;
8992 if (v_gets < hdr_gets.min)
8993 hdr_gets.min = v_gets;
8994 else if (v_gets > hdr_gets.max)
8995 hdr_gets.max = v_gets;
8998 /* compute per-chain averages */
8999 ch_gets.avg = ch_gets.sum / ((double)head->len);
9000 hdr_loads.avg = hdr_loads.sum / ((double)head->len);
9001 hdr_gets.avg = hdr_gets.sum / ((double)head->len);
9003 /* dump per-chain stats */
9004 Log("Volume hash chain %d : len=%d, looks=%s, reorders=%s\n",
9006 DoubleToPrintable(ch_looks.sum, pr_buf[0], sizeof(pr_buf[0])),
9007 DoubleToPrintable(ch_reorders.sum, pr_buf[1], sizeof(pr_buf[1])));
9008 Log("\tVolume gets : min=%s, max=%s, avg=%s, total=%s\n",
9009 DoubleToPrintable(ch_gets.min, pr_buf[0], sizeof(pr_buf[0])),
9010 DoubleToPrintable(ch_gets.max, pr_buf[1], sizeof(pr_buf[1])),
9011 DoubleToPrintable(ch_gets.avg, pr_buf[2], sizeof(pr_buf[2])),
9012 DoubleToPrintable(ch_gets.sum, pr_buf[3], sizeof(pr_buf[3])));
9013 Log("\tHDR gets : min=%s, max=%s, avg=%s, total=%s\n",
9014 DoubleToPrintable(hdr_gets.min, pr_buf[0], sizeof(pr_buf[0])),
9015 DoubleToPrintable(hdr_gets.max, pr_buf[1], sizeof(pr_buf[1])),
9016 DoubleToPrintable(hdr_gets.avg, pr_buf[2], sizeof(pr_buf[2])),
9017 DoubleToPrintable(hdr_gets.sum, pr_buf[3], sizeof(pr_buf[3])));
9018 Log("\tHDR loads : min=%s, max=%s, avg=%s, total=%s\n",
9019 DoubleToPrintable(hdr_loads.min, pr_buf[0], sizeof(pr_buf[0])),
9020 DoubleToPrintable(hdr_loads.max, pr_buf[1], sizeof(pr_buf[1])),
9021 DoubleToPrintable(hdr_loads.avg, pr_buf[2], sizeof(pr_buf[2])),
9022 DoubleToPrintable(hdr_loads.sum, pr_buf[3], sizeof(pr_buf[3])));
9023 } else if (flags & VOL_STATS_PER_CHAIN) {
9024 /* dump simple per-chain stats */
9025 Log("Volume hash chain %d : len=%d, looks=%s, gets=%s, reorders=%s\n",
9027 DoubleToPrintable(ch_looks.sum, pr_buf[0], sizeof(pr_buf[0])),
9028 DoubleToPrintable(ch_gets.sum, pr_buf[1], sizeof(pr_buf[1])),
9029 DoubleToPrintable(ch_reorders.sum, pr_buf[2], sizeof(pr_buf[2])));
9033 VHashEndExclusive_r(head);
9038 /* compute global averages */
9039 len.avg = len.sum / ((double)VolumeHashTable.Size);
9040 looks.avg = looks.sum / ((double)VolumeHashTable.Size);
9041 gets.avg = gets.sum / ((double)VolumeHashTable.Size);
9042 reorders.avg = reorders.sum / ((double)VolumeHashTable.Size);
9044 /* dump global stats */
9045 Log("Volume hash summary: %d buckets\n", VolumeHashTable.Size);
9046 Log(" chain length : min=%s, max=%s, avg=%s, total=%s\n",
9047 DoubleToPrintable(len.min, pr_buf[0], sizeof(pr_buf[0])),
9048 DoubleToPrintable(len.max, pr_buf[1], sizeof(pr_buf[1])),
9049 DoubleToPrintable(len.avg, pr_buf[2], sizeof(pr_buf[2])),
9050 DoubleToPrintable(len.sum, pr_buf[3], sizeof(pr_buf[3])));
9051 Log(" looks : min=%s, max=%s, avg=%s, total=%s\n",
9052 DoubleToPrintable(looks.min, pr_buf[0], sizeof(pr_buf[0])),
9053 DoubleToPrintable(looks.max, pr_buf[1], sizeof(pr_buf[1])),
9054 DoubleToPrintable(looks.avg, pr_buf[2], sizeof(pr_buf[2])),
9055 DoubleToPrintable(looks.sum, pr_buf[3], sizeof(pr_buf[3])));
9056 Log(" gets : min=%s, max=%s, avg=%s, total=%s\n",
9057 DoubleToPrintable(gets.min, pr_buf[0], sizeof(pr_buf[0])),
9058 DoubleToPrintable(gets.max, pr_buf[1], sizeof(pr_buf[1])),
9059 DoubleToPrintable(gets.avg, pr_buf[2], sizeof(pr_buf[2])),
9060 DoubleToPrintable(gets.sum, pr_buf[3], sizeof(pr_buf[3])));
9061 Log(" reorders : min=%s, max=%s, avg=%s, total=%s\n",
9062 DoubleToPrintable(reorders.min, pr_buf[0], sizeof(pr_buf[0])),
9063 DoubleToPrintable(reorders.max, pr_buf[1], sizeof(pr_buf[1])),
9064 DoubleToPrintable(reorders.avg, pr_buf[2], sizeof(pr_buf[2])),
9065 DoubleToPrintable(reorders.sum, pr_buf[3], sizeof(pr_buf[3])));
9067 /* print extended disk related statistics */
9069 struct DiskPartition64 * diskP;
9070 afs_uint32 vol_count[VOLMAXPARTS+1];
9071 byte part_exists[VOLMAXPARTS+1];
9075 memset(vol_count, 0, sizeof(vol_count));
9076 memset(part_exists, 0, sizeof(part_exists));
9080 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
9082 vol_count[id] = diskP->vol_list.len;
9083 part_exists[id] = 1;
9087 for (i = 0; i <= VOLMAXPARTS; i++) {
9088 if (part_exists[i]) {
9089 /* XXX while this is currently safe, it is a violation
9090 * of the VGetPartitionById_r interface contract. */
9091 diskP = VGetPartitionById_r(i, 0);
9093 Log("Partition %s has %d online volumes\n",
9094 VPartitionPath(diskP), diskP->vol_list.len);
9101 /* print extended VLRU statistics */
9102 if (VVLRUExtStats_r(&vlru_stats, vol_sum) == 0) {
9103 afs_uint32 idx, cur, lpos;
9108 Log("VLRU State Dump:\n\n");
9110 for (idx = VLRU_QUEUE_NEW; idx < VLRU_QUEUE_INVALID; idx++) {
9111 Log("\t%s:\n", vlru_idx_to_string(idx));
9114 for (cur = vlru_stats.queue_info[idx].start;
9115 cur < vlru_stats.queue_info[idx].len;
9117 line[lpos++] = vlru_stats.vec[cur].volid;
9119 Log("\t\t%u, %u, %u, %u, %u,\n",
9120 line[0], line[1], line[2], line[3], line[4]);
9129 Log("\t\t%u, %u, %u, %u, %u\n",
9130 line[0], line[1], line[2], line[3], line[4]);
9135 free(vlru_stats.vec);
9142 VPrintExtendedCacheStats(int flags)
9145 VPrintExtendedCacheStats_r(flags);
9148 #endif /* AFS_DEMAND_ATTACH_FS */
9151 VCanScheduleSalvage(void)
9153 return vol_opts.canScheduleSalvage;
9159 return vol_opts.canUseFSSYNC;
9163 VCanUseSALVSYNC(void)
9165 return vol_opts.canUseSALVSYNC;
9169 VCanUnsafeAttach(void)
9171 return vol_opts.unsafe_attach;