2 * Copyright 2000, International Business Machines Corporation and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
9 * Portions Copyright (c) 2005-2008 Sine Nomine Associates
12 /* 1/1/89: NB: this stuff is all going to be replaced. Don't take it too seriously */
17 Institution: The Information Technology Center, Carnegie-Mellon University
21 #include <afsconfig.h>
22 #include <afs/param.h>
30 #ifdef HAVE_SYS_FILE_H
34 #ifdef AFS_PTHREAD_ENV
35 # include <opr/lock.h>
37 # include <opr/lockstub.h>
40 #include <afs/afsint.h>
42 #include <rx/rx_queue.h>
45 #if !defined(AFS_SGI_ENV)
48 #else /* AFS_OSF_ENV */
49 #ifdef AFS_VFSINCL_ENV
52 #include <sys/fs/ufs_fs.h>
54 #if defined(AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV)
55 #include <ufs/ufs/dinode.h>
56 #include <ufs/ffs/fs.h>
61 #else /* AFS_VFSINCL_ENV */
62 #if !defined(AFS_AIX_ENV) && !defined(AFS_LINUX20_ENV) && !defined(AFS_XBSD_ENV) && !defined(AFS_DARWIN_ENV)
65 #endif /* AFS_VFSINCL_ENV */
66 #endif /* AFS_OSF_ENV */
67 #endif /* AFS_SGI_ENV */
68 #endif /* !AFS_NT40_ENV */
76 #if defined(AFS_SUN_ENV) || defined(AFS_SUN5_ENV)
78 #include <sys/mnttab.h>
79 #include <sys/mntent.h>
85 #if defined(AFS_SGI_ENV)
88 #ifndef AFS_LINUX20_ENV
89 #include <fstab.h> /* Need to find in libc 5, present in libc 6 */
92 #endif /* AFS_SGI_ENV */
94 #endif /* AFS_HPUX_ENV */
98 #include <afs/errors.h>
101 #include <afs/afssyscalls.h>
103 #include <afs/afsutil.h>
104 #include "daemon_com.h"
106 #include "salvsync.h"
109 #include "partition.h"
110 #include "volume_inline.h"
115 #ifdef AFS_PTHREAD_ENV
116 pthread_mutex_t vol_glock_mutex;
117 pthread_mutex_t vol_trans_mutex;
118 pthread_cond_t vol_put_volume_cond;
119 pthread_cond_t vol_sleep_cond;
120 pthread_cond_t vol_init_attach_cond;
121 pthread_cond_t vol_vinit_cond;
122 int vol_attach_threads = 1;
123 #endif /* AFS_PTHREAD_ENV */
125 /* start-time configurable I/O parameters */
126 ih_init_params vol_io_params;
128 #ifdef AFS_DEMAND_ATTACH_FS
129 pthread_mutex_t vol_salvsync_mutex;
132 * Set this to 1 to disallow SALVSYNC communication in all threads; used
133 * during shutdown, since the salvageserver may have gone away.
135 static volatile sig_atomic_t vol_disallow_salvsync = 0;
136 #endif /* AFS_DEMAND_ATTACH_FS */
139 * has VShutdown_r been called / is VShutdown_r running?
141 static int vol_shutting_down = 0;
144 extern void *calloc(), *realloc();
147 /* Forward declarations */
148 static Volume *attach2(Error * ec, VolId volumeId, char *path,
149 struct DiskPartition64 *partp, Volume * vp,
150 int isbusy, int mode, int *acheckedOut);
151 static void ReallyFreeVolume(Volume * vp);
152 #ifdef AFS_DEMAND_ATTACH_FS
153 static void FreeVolume(Volume * vp);
154 #else /* !AFS_DEMAND_ATTACH_FS */
155 #define FreeVolume(vp) ReallyFreeVolume(vp)
156 static void VScanUpdateList(void);
157 #endif /* !AFS_DEMAND_ATTACH_FS */
158 static void VInitVolumeHeaderCache(afs_uint32 howMany);
159 static int GetVolumeHeader(Volume * vp);
160 static void ReleaseVolumeHeader(struct volHeader *hd);
161 static void FreeVolumeHeader(Volume * vp);
162 static void AddVolumeToHashTable(Volume * vp, int hashid);
163 static void DeleteVolumeFromHashTable(Volume * vp);
165 static int VHold(Volume * vp);
167 static int VHold_r(Volume * vp);
168 static void VGetBitmap_r(Error * ec, Volume * vp, VnodeClass class);
169 static void VReleaseVolumeHandles_r(Volume * vp);
170 static void VCloseVolumeHandles_r(Volume * vp);
171 static void LoadVolumeHeader(Error * ec, Volume * vp);
172 static int VCheckOffline(Volume * vp);
173 static int VCheckDetach(Volume * vp);
174 static Volume * GetVolume(Error * ec, Error * client_ec, VolId volumeId,
175 Volume * hint, const struct timespec *ts);
177 int LogLevel; /* Vice loglevel--not defined as extern so that it will be
178 * defined when not linked with vice, XXXX */
179 ProgramType programType; /* The type of program using the package */
180 static VolumePackageOptions vol_opts;
182 /* extended volume package statistics */
185 #ifdef VOL_LOCK_DEBUG
186 pthread_t vol_glock_holder = 0;
190 /* this parameter needs to be tunable at runtime.
191 * 128 was really inadequate for largish servers -- at 16384 volumes this
192 * puts average chain length at 128, thus an average 65 deref's to find a volptr.
193 * talk about bad spatial locality...
195 * an AVL or splay tree might work a lot better, but we'll just increase
196 * the default hash table size for now
198 #define DEFAULT_VOLUME_HASH_SIZE 256 /* Must be a power of 2!! */
199 #define DEFAULT_VOLUME_HASH_MASK (DEFAULT_VOLUME_HASH_SIZE-1)
200 #define VOLUME_HASH(volumeId) (volumeId&(VolumeHashTable.Mask))
203 * turn volume hash chains into partially ordered lists.
204 * when the threshold is exceeded between two adjacent elements,
205 * perform a chain rebalancing operation.
207 * keep the threshold high in order to keep cache line invalidates
208 * low "enough" on SMPs
210 #define VOLUME_HASH_REORDER_THRESHOLD 200
213 * when possible, don't just reorder single elements, but reorder
214 * entire chains of elements at once. a chain of elements that
215 * exceed the element previous to the pivot by at least CHAIN_THRESH
216 * accesses are moved in front of the chain whose elements have at
217 * least CHAIN_THRESH less accesses than the pivot element
219 #define VOLUME_HASH_REORDER_CHAIN_THRESH (VOLUME_HASH_REORDER_THRESHOLD / 2)
221 #include "rx/rx_queue.h"
224 VolumeHashTable_t VolumeHashTable = {
225 DEFAULT_VOLUME_HASH_SIZE,
226 DEFAULT_VOLUME_HASH_MASK,
231 static void VInitVolumeHash(void);
235 /* This macro is used where an ffs() call does not exist. Was in util/ffs.c */
239 afs_int32 ffs_tmp = x;
243 for (ffs_i = 1;; ffs_i++) {
250 #endif /* !AFS_HAVE_FFS */
252 #ifdef AFS_PTHREAD_ENV
254 * disk partition queue element
256 typedef struct diskpartition_queue_t {
257 struct rx_queue queue; /**< queue header */
258 struct DiskPartition64 *diskP; /**< disk partition table entry */
259 } diskpartition_queue_t;
261 #ifndef AFS_DEMAND_ATTACH_FS
263 typedef struct vinitvolumepackage_thread_t {
264 struct rx_queue queue;
265 pthread_cond_t thread_done_cv;
266 int n_threads_complete;
267 } vinitvolumepackage_thread_t;
268 static void * VInitVolumePackageThread(void * args);
270 #else /* !AFS_DEMAND_ATTTACH_FS */
271 #define VINIT_BATCH_MAX_SIZE 512
274 * disk partition work queue
276 struct partition_queue {
277 struct rx_queue head; /**< diskpartition_queue_t queue */
278 pthread_mutex_t mutex;
283 * volumes parameters for preattach
285 struct volume_init_batch {
286 struct rx_queue queue; /**< queue header */
287 int thread; /**< posting worker thread */
288 int last; /**< indicates thread is done */
289 int size; /**< number of volume ids in batch */
290 Volume *batch[VINIT_BATCH_MAX_SIZE]; /**< volumes ids to preattach */
294 * volume parameters work queue
296 struct volume_init_queue {
297 struct rx_queue head; /**< volume_init_batch queue */
298 pthread_mutex_t mutex;
303 * volume init worker thread parameters
305 struct vinitvolumepackage_thread_param {
306 int nthreads; /**< total number of worker threads */
307 int thread; /**< thread number for this worker thread */
308 struct partition_queue *pq; /**< queue partitions to scan */
309 struct volume_init_queue *vq; /**< queue of volume to preattach */
312 static void *VInitVolumePackageThread(void *args);
313 static struct DiskPartition64 *VInitNextPartition(struct partition_queue *pq);
314 static VolId VInitNextVolumeId(DIR *dirp);
315 static int VInitPreAttachVolumes(int nthreads, struct volume_init_queue *vq);
317 #endif /* !AFS_DEMAND_ATTACH_FS */
318 #endif /* AFS_PTHREAD_ENV */
320 #ifndef AFS_DEMAND_ATTACH_FS
321 static int VAttachVolumesByPartition(struct DiskPartition64 *diskP,
322 int * nAttached, int * nUnattached);
323 #endif /* AFS_DEMAND_ATTACH_FS */
326 #ifdef AFS_DEMAND_ATTACH_FS
327 /* demand attach fileserver extensions */
330 * in the future we will support serialization of VLRU state into the fs_state
333 * these structures are the beginning of that effort
335 struct VLRU_DiskHeader {
336 struct versionStamp stamp; /* magic and structure version number */
337 afs_uint32 mtime; /* time of dump to disk */
338 afs_uint32 num_records; /* number of VLRU_DiskEntry records */
341 struct VLRU_DiskEntry {
342 afs_uint32 vid; /* volume ID */
343 afs_uint32 idx; /* generation */
344 afs_uint32 last_get; /* timestamp of last get */
347 struct VLRU_StartupQueue {
348 struct VLRU_DiskEntry * entry;
353 typedef struct vshutdown_thread_t {
355 pthread_mutex_t lock;
357 pthread_cond_t master_cv;
359 int n_threads_complete;
361 int schedule_version;
364 byte n_parts_done_pass;
365 byte part_thread_target[VOLMAXPARTS+1];
366 byte part_done_pass[VOLMAXPARTS+1];
367 struct rx_queue * part_pass_head[VOLMAXPARTS+1];
368 int stats[4][VOLMAXPARTS+1];
369 } vshutdown_thread_t;
370 static void * VShutdownThread(void * args);
373 static Volume * VAttachVolumeByVp_r(Error * ec, Volume * vp, int mode);
374 static int VCheckFree(Volume * vp);
377 static void AddVolumeToVByPList_r(Volume * vp);
378 static void DeleteVolumeFromVByPList_r(Volume * vp);
379 static void VVByPListBeginExclusive_r(struct DiskPartition64 * dp);
380 static void VVByPListEndExclusive_r(struct DiskPartition64 * dp);
381 static void VVByPListWait_r(struct DiskPartition64 * dp);
383 /* online salvager */
385 VCHECK_SALVAGE_OK = 0, /**< no pending salvage */
386 VCHECK_SALVAGE_SCHEDULED = 1, /**< salvage has been scheduled */
387 VCHECK_SALVAGE_ASYNC = 2, /**< salvage being scheduled */
388 VCHECK_SALVAGE_DENIED = 3, /**< salvage not scheduled; denied */
389 VCHECK_SALVAGE_FAIL = 4 /**< salvage not scheduled; failed */
391 static int VCheckSalvage(Volume * vp);
392 #if defined(SALVSYNC_BUILD_CLIENT) || defined(FSSYNC_BUILD_CLIENT)
393 static int VScheduleSalvage_r(Volume * vp);
396 /* Volume hash table */
397 static void VReorderHash_r(VolumeHashChainHead * head, Volume * pp, Volume * vp);
398 static void VHashBeginExclusive_r(VolumeHashChainHead * head);
399 static void VHashEndExclusive_r(VolumeHashChainHead * head);
400 static void VHashWait_r(VolumeHashChainHead * head);
403 static int ShutdownVByPForPass_r(struct DiskPartition64 * dp, int pass);
404 static int ShutdownVolumeWalk_r(struct DiskPartition64 * dp, int pass,
405 struct rx_queue ** idx);
406 static void ShutdownController(vshutdown_thread_t * params);
407 static void ShutdownCreateSchedule(vshutdown_thread_t * params);
410 static void VLRU_ComputeConstants(void);
411 static void VInitVLRU(void);
412 static void VLRU_Init_Node_r(Volume * vp);
413 static void VLRU_Add_r(Volume * vp);
414 static void VLRU_Delete_r(Volume * vp);
415 static void VLRU_UpdateAccess_r(Volume * vp);
416 static void * VLRU_ScannerThread(void * args);
417 static void VLRU_Scan_r(int idx);
418 static void VLRU_Promote_r(int idx);
419 static void VLRU_Demote_r(int idx);
420 static void VLRU_SwitchQueues(Volume * vp, int new_idx, int append);
423 static int VCheckSoftDetach(Volume * vp, afs_uint32 thresh);
424 static int VCheckSoftDetachCandidate(Volume * vp, afs_uint32 thresh);
425 static int VSoftDetachVolume_r(Volume * vp, afs_uint32 thresh);
428 pthread_key_t VThread_key;
429 VThreadOptions_t VThread_defaults = {
430 0 /**< allow salvsync */
432 #endif /* AFS_DEMAND_ATTACH_FS */
435 struct Lock vol_listLock; /* Lock obtained when listing volumes:
436 * prevents a volume from being missed
437 * if the volume is attached during a
441 /* Common message used when the volume goes off line */
442 char *VSalvageMessage =
443 "Files in this volume are currently unavailable; call operations";
445 int VInit; /* 0 - uninitialized,
446 * 1 - initialized but not all volumes have been attached,
447 * 2 - initialized and all volumes have been attached,
448 * 3 - initialized, all volumes have been attached, and
449 * VConnectFS() has completed. */
451 static int vinit_attach_abort = 0;
453 bit32 VolumeCacheCheck; /* Incremented everytime a volume goes on line--
454 * used to stamp volume headers and in-core
455 * vnodes. When the volume goes on-line the
456 * vnode will be invalidated
457 * access only with VOL_LOCK held */
462 /***************************************************/
463 /* Startup routines */
464 /***************************************************/
466 #if defined(FAST_RESTART) && defined(AFS_DEMAND_ATTACH_FS)
467 # error FAST_RESTART and DAFS are incompatible. For the DAFS equivalent \
468 of FAST_RESTART, use the -unsafe-nosalvage fileserver argument
472 * assign default values to a VolumePackageOptions struct.
474 * Always call this on a VolumePackageOptions struct first, then set any
475 * specific options you want, then call VInitVolumePackage2.
477 * @param[in] pt caller's program type
478 * @param[out] opts volume package options
481 VOptDefaults(ProgramType pt, VolumePackageOptions *opts)
483 opts->nLargeVnodes = opts->nSmallVnodes = 5;
486 opts->canScheduleSalvage = 0;
487 opts->canUseFSSYNC = 0;
488 opts->canUseSALVSYNC = 0;
490 opts->interrupt_rxcall = NULL;
491 opts->offline_timeout = -1;
492 opts->offline_shutdown_timeout = -1;
493 opts->usage_threshold = 128;
494 opts->usage_rate_limit = 5;
497 opts->unsafe_attach = 1;
498 #else /* !FAST_RESTART */
499 opts->unsafe_attach = 0;
500 #endif /* !FAST_RESTART */
504 opts->canScheduleSalvage = 1;
505 opts->canUseSALVSYNC = 1;
509 opts->canUseFSSYNC = 1;
513 opts->nLargeVnodes = 0;
514 opts->nSmallVnodes = 0;
516 opts->canScheduleSalvage = 1;
517 opts->canUseFSSYNC = 1;
527 * Set VInit to a certain value, and signal waiters.
529 * @param[in] value the value to set VInit to
534 VSetVInit_r(int value)
537 opr_cv_broadcast(&vol_vinit_cond);
541 VLogOfflineTimeout(const char *type, afs_int32 timeout)
547 Log("VInitVolumePackage: Interrupting clients accessing %s "
548 "immediately\n", type);
550 Log("VInitVolumePackage: Interrupting clients accessing %s "
551 "after %ld second%s\n", type, (long)timeout, timeout==1?"":"s");
556 VInitVolumePackage2(ProgramType pt, VolumePackageOptions * opts)
558 int errors = 0; /* Number of errors while finding vice partitions. */
563 #ifndef AFS_PTHREAD_ENV
564 if (opts->offline_timeout != -1 || opts->offline_shutdown_timeout != -1) {
565 Log("VInitVolumePackage: offline_timeout and/or "
566 "offline_shutdown_timeout was specified, but the volume package "
567 "does not support these for LWP builds\n");
571 VLogOfflineTimeout("volumes going offline", opts->offline_timeout);
572 VLogOfflineTimeout("volumes going offline during shutdown",
573 opts->offline_shutdown_timeout);
575 memset(&VStats, 0, sizeof(VStats));
576 VStats.hdr_cache_size = 200;
578 VInitPartitionPackage();
580 #ifdef AFS_DEMAND_ATTACH_FS
581 if (programType == fileServer) {
584 VLRU_SetOptions(VLRU_SET_ENABLED, 0);
586 opr_Verify(pthread_key_create(&VThread_key, NULL) == 0);
589 opr_mutex_init(&vol_glock_mutex);
590 opr_mutex_init(&vol_trans_mutex);
591 opr_cv_init(&vol_put_volume_cond);
592 opr_cv_init(&vol_sleep_cond);
593 opr_cv_init(&vol_init_attach_cond);
594 opr_cv_init(&vol_vinit_cond);
595 #ifndef AFS_PTHREAD_ENV
597 #endif /* AFS_PTHREAD_ENV */
598 Lock_Init(&vol_listLock);
600 srandom(time(0)); /* For VGetVolumeInfo */
602 #ifdef AFS_DEMAND_ATTACH_FS
603 opr_mutex_init(&vol_salvsync_mutex);
604 #endif /* AFS_DEMAND_ATTACH_FS */
606 /* Ok, we have done enough initialization that fileserver can
607 * start accepting calls, even though the volumes may not be
608 * available just yet.
612 #if defined(AFS_DEMAND_ATTACH_FS) && defined(SALVSYNC_BUILD_SERVER)
613 if (programType == salvageServer) {
616 #endif /* AFS_DEMAND_ATTACH_FS */
617 #ifdef FSSYNC_BUILD_SERVER
618 if (programType == fileServer) {
622 #if defined(AFS_DEMAND_ATTACH_FS) && defined(SALVSYNC_BUILD_CLIENT)
623 if (VCanUseSALVSYNC()) {
624 /* establish a connection to the salvager at this point */
625 opr_Verify(VConnectSALV() != 0);
627 #endif /* AFS_DEMAND_ATTACH_FS */
629 if (opts->volcache > VStats.hdr_cache_size)
630 VStats.hdr_cache_size = opts->volcache;
631 VInitVolumeHeaderCache(VStats.hdr_cache_size);
633 VInitVnodes(vLarge, opts->nLargeVnodes);
634 VInitVnodes(vSmall, opts->nSmallVnodes);
637 errors = VAttachPartitions();
641 if (programType != fileServer) {
642 errors = VInitAttachVolumes(programType);
648 #ifdef FSSYNC_BUILD_CLIENT
649 if (VCanUseFSSYNC()) {
651 #ifdef AFS_DEMAND_ATTACH_FS
652 if (programType == salvageServer) {
653 Log("Unable to connect to file server; aborted\n");
656 #endif /* AFS_DEMAND_ATTACH_FS */
657 Log("Unable to connect to file server; will retry at need\n");
660 #endif /* FSSYNC_BUILD_CLIENT */
665 #if !defined(AFS_PTHREAD_ENV)
667 * Attach volumes in vice partitions
669 * @param[in] pt calling program type
672 * @note This is the original, non-threaded version of attach parititions.
674 * @post VInit state is 2
677 VInitAttachVolumes(ProgramType pt)
679 opr_Assert(VInit==1);
680 if (pt == fileServer) {
681 struct DiskPartition64 *diskP;
682 /* Attach all the volumes in this partition */
683 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
684 int nAttached = 0, nUnattached = 0;
685 opr_Verify(VAttachVolumesByPartition(diskP,
686 &nAttached, &nUnattached)
691 VSetVInit_r(2); /* Initialized, and all volumes have been attached */
692 LWP_NoYieldSignal(VInitAttachVolumes);
696 #endif /* !AFS_PTHREAD_ENV */
698 #if defined(AFS_PTHREAD_ENV) && !defined(AFS_DEMAND_ATTACH_FS)
700 * Attach volumes in vice partitions
702 * @param[in] pt calling program type
705 * @note Threaded version of attach parititions.
707 * @post VInit state is 2
710 VInitAttachVolumes(ProgramType pt)
712 opr_Assert(VInit==1);
713 if (pt == fileServer) {
714 struct DiskPartition64 *diskP;
715 struct vinitvolumepackage_thread_t params;
716 struct diskpartition_queue_t * dpq;
717 int i, threads, parts;
719 pthread_attr_t attrs;
721 opr_cv_init(¶ms.thread_done_cv);
723 params.n_threads_complete = 0;
725 /* create partition work queue */
726 for (parts=0, diskP = DiskPartitionList; diskP; diskP = diskP->next, parts++) {
727 dpq = malloc(sizeof(struct diskpartition_queue_t));
728 opr_Assert(dpq != NULL);
730 queue_Append(¶ms,dpq);
733 threads = min(parts, vol_attach_threads);
736 /* spawn off a bunch of initialization threads */
737 opr_Verify(pthread_attr_init(&attrs) == 0);
738 opr_Verify(pthread_attr_setdetachstate(&attrs,
739 PTHREAD_CREATE_DETACHED)
742 Log("VInitVolumePackage: beginning parallel fileserver startup\n");
743 Log("VInitVolumePackage: using %d threads to attach volumes on %d partitions\n",
747 for (i=0; i < threads; i++) {
750 opr_Verify(pthread_create(&tid, &attrs,
751 &VInitVolumePackageThread,
753 AFS_SIGSET_RESTORE();
756 while(params.n_threads_complete < threads) {
757 VOL_CV_WAIT(¶ms.thread_done_cv);
761 opr_Verify(pthread_attr_destroy(&attrs) == 0);
763 /* if we're only going to run one init thread, don't bother creating
765 Log("VInitVolumePackage: beginning single-threaded fileserver startup\n");
766 Log("VInitVolumePackage: using 1 thread to attach volumes on %d partition(s)\n",
769 VInitVolumePackageThread(¶ms);
772 opr_cv_destroy(¶ms.thread_done_cv);
775 VSetVInit_r(2); /* Initialized, and all volumes have been attached */
776 opr_cv_broadcast(&vol_init_attach_cond);
782 VInitVolumePackageThread(void * args) {
784 struct DiskPartition64 *diskP;
785 struct vinitvolumepackage_thread_t * params;
786 struct diskpartition_queue_t * dpq;
788 params = (vinitvolumepackage_thread_t *) args;
792 /* Attach all the volumes in this partition */
793 while (queue_IsNotEmpty(params)) {
794 int nAttached = 0, nUnattached = 0;
796 if (vinit_attach_abort) {
797 Log("Aborting initialization\n");
801 dpq = queue_First(params,diskpartition_queue_t);
807 opr_Verify(VAttachVolumesByPartition(diskP, &nAttached,
814 params->n_threads_complete++;
815 opr_cv_signal(¶ms->thread_done_cv);
819 #endif /* AFS_PTHREAD_ENV && !AFS_DEMAND_ATTACH_FS */
821 #if defined(AFS_DEMAND_ATTACH_FS)
823 * Attach volumes in vice partitions
825 * @param[in] pt calling program type
828 * @note Threaded version of attach partitions.
830 * @post VInit state is 2
833 VInitAttachVolumes(ProgramType pt)
835 opr_Assert(VInit==1);
836 if (pt == fileServer) {
838 struct DiskPartition64 *diskP;
839 struct partition_queue pq;
840 struct volume_init_queue vq;
842 int i, threads, parts;
844 pthread_attr_t attrs;
846 /* create partition work queue */
849 opr_mutex_init(&pq.mutex);
850 for (parts = 0, diskP = DiskPartitionList; diskP; diskP = diskP->next, parts++) {
851 struct diskpartition_queue_t *dp;
852 dp = malloc(sizeof(struct diskpartition_queue_t));
853 opr_Assert(dp != NULL);
855 queue_Append(&pq, dp);
858 /* number of worker threads; at least one, not to exceed the number of partitions */
859 threads = min(parts, vol_attach_threads);
861 /* create volume work queue */
864 opr_mutex_init(&vq.mutex);
866 opr_Verify(pthread_attr_init(&attrs) == 0);
867 opr_Verify(pthread_attr_setdetachstate(&attrs,
868 PTHREAD_CREATE_DETACHED) == 0);
870 Log("VInitVolumePackage: beginning parallel fileserver startup\n");
871 Log("VInitVolumePackage: using %d threads to pre-attach volumes on %d partitions\n",
874 /* create threads to scan disk partitions. */
875 for (i=0; i < threads; i++) {
876 struct vinitvolumepackage_thread_param *params;
879 params = malloc(sizeof(struct vinitvolumepackage_thread_param));
883 params->nthreads = threads;
884 params->thread = i+1;
887 opr_Verify(pthread_create(&tid, &attrs,
888 &VInitVolumePackageThread,
889 (void*)params) == 0);
890 AFS_SIGSET_RESTORE();
893 VInitPreAttachVolumes(threads, &vq);
895 opr_Verify(pthread_attr_destroy(&attrs) == 0);
896 opr_cv_destroy(&pq.cv);
897 opr_mutex_destroy(&pq.mutex);
898 opr_cv_destroy(&vq.cv);
899 opr_mutex_destroy(&vq.mutex);
903 VSetVInit_r(2); /* Initialized, and all volumes have been attached */
904 opr_cv_broadcast(&vol_init_attach_cond);
911 * Volume package initialization worker thread. Scan partitions for volume
912 * header files. Gather batches of volume ids and dispatch them to
913 * the main thread to be preattached. The volume preattachement is done
914 * in the main thread to avoid global volume lock contention.
917 VInitVolumePackageThread(void *args)
919 struct vinitvolumepackage_thread_param *params;
920 struct DiskPartition64 *partition;
921 struct partition_queue *pq;
922 struct volume_init_queue *vq;
923 struct volume_init_batch *vb;
926 params = (struct vinitvolumepackage_thread_param *)args;
932 vb = malloc(sizeof(struct volume_init_batch));
934 vb->thread = params->thread;
938 Log("Scanning partitions on thread %d of %d\n", params->thread, params->nthreads);
939 while((partition = VInitNextPartition(pq))) {
943 Log("Partition %s: pre-attaching volumes\n", partition->name);
944 dirp = opendir(VPartitionPath(partition));
946 Log("opendir on Partition %s failed, errno=%d!\n", partition->name, errno);
949 while ((vid = VInitNextVolumeId(dirp))) {
950 Volume *vp = calloc(1, sizeof(Volume));
952 vp->device = partition->device;
953 vp->partition = partition;
955 queue_Init(&vp->vnode_list);
956 queue_Init(&vp->rx_call_list);
957 opr_cv_init(&V_attachCV(vp));
959 vb->batch[vb->size++] = vp;
960 if (vb->size == VINIT_BATCH_MAX_SIZE) {
961 opr_mutex_enter(&vq->mutex);
962 queue_Append(vq, vb);
963 opr_cv_broadcast(&vq->cv);
964 opr_mutex_exit(&vq->mutex);
966 vb = malloc(sizeof(struct volume_init_batch));
968 vb->thread = params->thread;
977 opr_mutex_enter(&vq->mutex);
978 queue_Append(vq, vb);
979 opr_cv_broadcast(&vq->cv);
980 opr_mutex_exit(&vq->mutex);
982 Log("Partition scan thread %d of %d ended\n", params->thread, params->nthreads);
988 * Read next element from the pre-populated partition list.
990 static struct DiskPartition64*
991 VInitNextPartition(struct partition_queue *pq)
993 struct DiskPartition64 *partition;
994 struct diskpartition_queue_t *dp; /* queue element */
996 if (vinit_attach_abort) {
997 Log("Aborting volume preattach thread.\n");
1001 /* get next partition to scan */
1002 opr_mutex_enter(&pq->mutex);
1003 if (queue_IsEmpty(pq)) {
1004 opr_mutex_exit(&pq->mutex);
1007 dp = queue_First(pq, diskpartition_queue_t);
1009 opr_mutex_exit(&pq->mutex);
1012 opr_Assert(dp->diskP);
1014 partition = dp->diskP;
1020 * Find next volume id on the partition.
1023 VInitNextVolumeId(DIR *dirp)
1029 while((d = readdir(dirp))) {
1030 if (vinit_attach_abort) {
1031 Log("Aborting volume preattach thread.\n");
1034 ext = strrchr(d->d_name, '.');
1035 if (d->d_name[0] == 'V' && ext && strcmp(ext, VHDREXT) == 0) {
1036 vid = VolumeNumber(d->d_name);
1040 Log("Warning: bogus volume header file: %s\n", d->d_name);
1047 * Preattach volumes in batches to avoid lock contention.
1050 VInitPreAttachVolumes(int nthreads, struct volume_init_queue *vq)
1052 struct volume_init_batch *vb;
1056 /* dequeue next volume */
1057 opr_mutex_enter(&vq->mutex);
1058 if (queue_IsEmpty(vq)) {
1059 opr_cv_wait(&vq->cv, &vq->mutex);
1061 vb = queue_First(vq, volume_init_batch);
1063 opr_mutex_exit(&vq->mutex);
1067 for (i = 0; i<vb->size; i++) {
1073 dup = VLookupVolume_r(&ec, vp->hashid, NULL);
1075 Log("Error looking up volume, code=%d\n", ec);
1078 Log("Warning: Duplicate volume id %d detected.\n", vp->hashid);
1081 /* put pre-attached volume onto the hash table
1082 * and bring it up to the pre-attached state */
1083 AddVolumeToHashTable(vp, vp->hashid);
1084 AddVolumeToVByPList_r(vp);
1085 VLRU_Init_Node_r(vp);
1086 VChangeState_r(vp, VOL_STATE_PREATTACHED);
1099 #endif /* AFS_DEMAND_ATTACH_FS */
1101 #if !defined(AFS_DEMAND_ATTACH_FS)
1103 * attach all volumes on a given disk partition
1106 VAttachVolumesByPartition(struct DiskPartition64 *diskP, int * nAttached, int * nUnattached)
1112 Log("Partition %s: attaching volumes\n", diskP->name);
1113 dirp = opendir(VPartitionPath(diskP));
1115 Log("opendir on Partition %s failed!\n", diskP->name);
1119 while ((dp = readdir(dirp))) {
1121 p = strrchr(dp->d_name, '.');
1123 if (vinit_attach_abort) {
1124 Log("Partition %s: abort attach volumes\n", diskP->name);
1128 if (p != NULL && strcmp(p, VHDREXT) == 0) {
1131 vp = VAttachVolumeByName(&error, diskP->name, dp->d_name,
1133 (*(vp ? nAttached : nUnattached))++;
1134 if (error == VOFFLINE)
1135 Log("Volume %d stays offline (/vice/offline/%s exists)\n", VolumeNumber(dp->d_name), dp->d_name);
1136 else if (LogLevel >= 5) {
1137 Log("Partition %s: attached volume %d (%s)\n",
1138 diskP->name, VolumeNumber(dp->d_name),
1147 Log("Partition %s: attached %d volumes; %d volumes not attached\n", diskP->name, *nAttached, *nUnattached);
1152 #endif /* !AFS_DEMAND_ATTACH_FS */
1154 /***************************************************/
1155 /* Shutdown routines */
1156 /***************************************************/
1160 * highly multithreaded volume package shutdown
1162 * with the demand attach fileserver extensions,
1163 * VShutdown has been modified to be multithreaded.
1164 * In order to achieve optimal use of many threads,
1165 * the shutdown code involves one control thread and
1166 * n shutdown worker threads. The control thread
1167 * periodically examines the number of volumes available
1168 * for shutdown on each partition, and produces a worker
1169 * thread allocation schedule. The idea is to eliminate
1170 * redundant scheduling computation on the workers by
1171 * having a single master scheduler.
1173 * The scheduler's objectives are:
1175 * each partition with volumes remaining gets allocated
1176 * at least 1 thread (assuming sufficient threads)
1178 * threads are allocated proportional to the number of
1179 * volumes remaining to be offlined. This ensures that
1180 * the OS I/O scheduler has many requests to elevator
1181 * seek on partitions that will (presumably) take the
1182 * longest amount of time (from now) to finish shutdown
1183 * (3) keep threads busy
1184 * when there are extra threads, they are assigned to
1185 * partitions using a simple round-robin algorithm
1187 * In the future, we may wish to add the ability to adapt
1188 * to the relative performance patterns of each disk
1193 * multi-step shutdown process
1195 * demand attach shutdown is a four-step process. Each
1196 * shutdown "pass" shuts down increasingly more difficult
1197 * volumes. The main purpose is to achieve better cache
1198 * utilization during shutdown.
1201 * shutdown volumes in the unattached, pre-attached
1204 * shutdown attached volumes with cached volume headers
1206 * shutdown all volumes in non-exclusive states
1208 * shutdown all remaining volumes
1211 #ifdef AFS_DEMAND_ATTACH_FS
1217 struct DiskPartition64 * diskP;
1218 struct diskpartition_queue_t * dpq;
1219 vshutdown_thread_t params;
1221 pthread_attr_t attrs;
1223 memset(¶ms, 0, sizeof(vshutdown_thread_t));
1226 Log("VShutdown: aborting attach volumes\n");
1227 vinit_attach_abort = 1;
1228 VOL_CV_WAIT(&vol_init_attach_cond);
1231 for (params.n_parts=0, diskP = DiskPartitionList;
1232 diskP; diskP = diskP->next, params.n_parts++);
1234 Log("VShutdown: shutting down on-line volumes on %d partition%s...\n",
1235 params.n_parts, params.n_parts > 1 ? "s" : "");
1237 vol_shutting_down = 1;
1239 if (vol_attach_threads > 1) {
1240 /* prepare for parallel shutdown */
1241 params.n_threads = vol_attach_threads;
1242 opr_mutex_init(¶ms.lock);
1243 opr_cv_init(¶ms.cv);
1244 opr_cv_init(¶ms.master_cv);
1245 opr_Verify(pthread_attr_init(&attrs) == 0);
1246 opr_Verify(pthread_attr_setdetachstate(&attrs,
1247 PTHREAD_CREATE_DETACHED) == 0);
1248 queue_Init(¶ms);
1250 /* setup the basic partition information structures for
1251 * parallel shutdown */
1252 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1254 struct rx_queue * qp, * nqp;
1258 VVByPListWait_r(diskP);
1259 VVByPListBeginExclusive_r(diskP);
1262 for (queue_Scan(&diskP->vol_list, qp, nqp, rx_queue)) {
1263 vp = (Volume *)((char *)qp - offsetof(Volume, vol_list));
1267 Log("VShutdown: partition %s has %d volumes with attached headers\n",
1268 VPartitionPath(diskP), count);
1271 /* build up the pass 0 shutdown work queue */
1272 dpq = malloc(sizeof(struct diskpartition_queue_t));
1273 opr_Assert(dpq != NULL);
1275 queue_Prepend(¶ms, dpq);
1277 params.part_pass_head[diskP->index] = queue_First(&diskP->vol_list, rx_queue);
1280 Log("VShutdown: beginning parallel fileserver shutdown\n");
1281 Log("VShutdown: using %d threads to offline volumes on %d partition%s\n",
1282 vol_attach_threads, params.n_parts, params.n_parts > 1 ? "s" : "" );
1284 /* do pass 0 shutdown */
1285 opr_mutex_enter(¶ms.lock);
1286 for (i=0; i < params.n_threads; i++) {
1287 opr_Verify(pthread_create(&tid, &attrs, &VShutdownThread,
1291 /* wait for all the pass 0 shutdowns to complete */
1292 while (params.n_threads_complete < params.n_threads) {
1293 CV_WAIT(¶ms.master_cv, ¶ms.lock);
1295 params.n_threads_complete = 0;
1297 opr_cv_broadcast(¶ms.cv);
1298 opr_mutex_exit(¶ms.lock);
1300 Log("VShutdown: pass 0 completed using the 1 thread per partition algorithm\n");
1301 Log("VShutdown: starting passes 1 through 3 using finely-granular mp-fast algorithm\n");
1303 /* run the parallel shutdown scheduler. it will drop the glock internally */
1304 ShutdownController(¶ms);
1306 /* wait for all the workers to finish pass 3 and terminate */
1307 while (params.pass < 4) {
1308 VOL_CV_WAIT(¶ms.cv);
1311 opr_Verify(pthread_attr_destroy(&attrs) == 0);
1312 opr_cv_destroy(¶ms.cv);
1313 opr_cv_destroy(¶ms.master_cv);
1314 opr_mutex_destroy(¶ms.lock);
1316 /* drop the VByPList exclusive reservations */
1317 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1318 VVByPListEndExclusive_r(diskP);
1319 Log("VShutdown: %s stats : (pass[0]=%d, pass[1]=%d, pass[2]=%d, pass[3]=%d)\n",
1320 VPartitionPath(diskP),
1321 params.stats[0][diskP->index],
1322 params.stats[1][diskP->index],
1323 params.stats[2][diskP->index],
1324 params.stats[3][diskP->index]);
1327 Log("VShutdown: shutdown finished using %d threads\n", params.n_threads);
1329 /* if we're only going to run one shutdown thread, don't bother creating
1331 Log("VShutdown: beginning single-threaded fileserver shutdown\n");
1333 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1334 VShutdownByPartition_r(diskP);
1338 Log("VShutdown: complete.\n");
1341 #else /* AFS_DEMAND_ATTACH_FS */
1351 Log("VShutdown: aborting attach volumes\n");
1352 vinit_attach_abort = 1;
1353 #ifdef AFS_PTHREAD_ENV
1354 VOL_CV_WAIT(&vol_init_attach_cond);
1356 LWP_WaitProcess(VInitAttachVolumes);
1357 #endif /* AFS_PTHREAD_ENV */
1360 Log("VShutdown: shutting down on-line volumes...\n");
1361 vol_shutting_down = 1;
1362 for (i = 0; i < VolumeHashTable.Size; i++) {
1363 /* try to hold first volume in the hash table */
1364 for (queue_Scan(&VolumeHashTable.Table[i],vp,np,Volume)) {
1368 Log("VShutdown: Attempting to take volume %u offline.\n",
1371 /* next, take the volume offline (drops reference count) */
1372 VOffline_r(vp, "File server was shut down");
1376 Log("VShutdown: complete.\n");
1378 #endif /* AFS_DEMAND_ATTACH_FS */
1384 opr_Assert(VInit>0);
1391 * stop new activity (e.g. SALVSYNC) from occurring
1393 * Use this to make the volume package less busy; for example, during
1394 * shutdown. This doesn't actually shutdown/detach anything in the
1395 * volume package, but prevents certain processes from ocurring. For
1396 * example, preventing new SALVSYNC communication in DAFS. In theory, we
1397 * could also use this to prevent new volume attachment, or prevent
1398 * other programs from checking out volumes, etc.
1403 #ifdef AFS_DEMAND_ATTACH_FS
1404 /* make sure we don't try to contact the salvageserver, since it may
1405 * not be around anymore */
1406 vol_disallow_salvsync = 1;
1410 #ifdef AFS_DEMAND_ATTACH_FS
1413 * shutdown control thread
1416 ShutdownController(vshutdown_thread_t * params)
1419 struct DiskPartition64 * diskP;
1421 vshutdown_thread_t shadow;
1423 ShutdownCreateSchedule(params);
1425 while ((params->pass < 4) &&
1426 (params->n_threads_complete < params->n_threads)) {
1427 /* recompute schedule once per second */
1429 memcpy(&shadow, params, sizeof(vshutdown_thread_t));
1433 Log("ShutdownController: schedule version=%d, vol_remaining=%d, pass=%d\n",
1434 shadow.schedule_version, shadow.vol_remaining, shadow.pass);
1435 Log("ShutdownController: n_threads_complete=%d, n_parts_done_pass=%d\n",
1436 shadow.n_threads_complete, shadow.n_parts_done_pass);
1437 for (diskP = DiskPartitionList; diskP; diskP=diskP->next) {
1439 Log("ShutdownController: part[%d] : (len=%d, thread_target=%d, done_pass=%d, pass_head=%p)\n",
1441 diskP->vol_list.len,
1442 shadow.part_thread_target[id],
1443 shadow.part_done_pass[id],
1444 shadow.part_pass_head[id]);
1450 ShutdownCreateSchedule(params);
1454 /* create the shutdown thread work schedule.
1455 * this scheduler tries to implement fairness
1456 * by allocating at least 1 thread to each
1457 * partition with volumes to be shutdown,
1458 * and then it attempts to allocate remaining
1459 * threads based upon the amount of work left
1462 ShutdownCreateSchedule(vshutdown_thread_t * params)
1464 struct DiskPartition64 * diskP;
1465 int sum, thr_workload, thr_left;
1466 int part_residue[VOLMAXPARTS+1];
1469 /* compute the total number of outstanding volumes */
1471 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1472 sum += diskP->vol_list.len;
1475 params->schedule_version++;
1476 params->vol_remaining = sum;
1481 /* compute average per-thread workload */
1482 thr_workload = sum / params->n_threads;
1483 if (sum % params->n_threads)
1486 thr_left = params->n_threads;
1487 memset(&part_residue, 0, sizeof(part_residue));
1489 /* for fairness, give every partition with volumes remaining
1490 * at least one thread */
1491 for (diskP = DiskPartitionList; diskP && thr_left; diskP = diskP->next) {
1493 if (diskP->vol_list.len) {
1494 params->part_thread_target[id] = 1;
1497 params->part_thread_target[id] = 0;
1501 if (thr_left && thr_workload) {
1502 /* compute length-weighted workloads */
1505 for (diskP = DiskPartitionList; diskP && thr_left; diskP = diskP->next) {
1507 delta = (diskP->vol_list.len / thr_workload) -
1508 params->part_thread_target[id];
1512 if (delta < thr_left) {
1513 params->part_thread_target[id] += delta;
1516 params->part_thread_target[id] += thr_left;
1524 /* try to assign any leftover threads to partitions that
1525 * had volume lengths closer to needing thread_target+1 */
1526 int max_residue, max_id = 0;
1528 /* compute the residues */
1529 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1531 part_residue[id] = diskP->vol_list.len -
1532 (params->part_thread_target[id] * thr_workload);
1535 /* now try to allocate remaining threads to partitions with the
1536 * highest residues */
1539 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1541 if (part_residue[id] > max_residue) {
1542 max_residue = part_residue[id];
1551 params->part_thread_target[max_id]++;
1553 part_residue[max_id] = 0;
1558 /* punt and give any remaining threads equally to each partition */
1560 if (thr_left >= params->n_parts) {
1561 alloc = thr_left / params->n_parts;
1562 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1564 params->part_thread_target[id] += alloc;
1569 /* finish off the last of the threads */
1570 for (diskP = DiskPartitionList; thr_left && diskP; diskP = diskP->next) {
1572 params->part_thread_target[id]++;
1578 /* worker thread for parallel shutdown */
1580 VShutdownThread(void * args)
1582 vshutdown_thread_t * params;
1583 int found, pass, schedule_version_save, count;
1584 struct DiskPartition64 *diskP;
1585 struct diskpartition_queue_t * dpq;
1588 params = (vshutdown_thread_t *) args;
1590 /* acquire the shutdown pass 0 lock */
1591 opr_mutex_enter(¶ms->lock);
1593 /* if there's still pass 0 work to be done,
1594 * get a work entry, and do a pass 0 shutdown */
1595 if (queue_IsNotEmpty(params)) {
1596 dpq = queue_First(params, diskpartition_queue_t);
1598 opr_mutex_exit(¶ms->lock);
1604 while (ShutdownVolumeWalk_r(diskP, 0, ¶ms->part_pass_head[id]))
1606 params->stats[0][diskP->index] = count;
1607 opr_mutex_enter(¶ms->lock);
1610 params->n_threads_complete++;
1611 if (params->n_threads_complete == params->n_threads) {
1612 /* notify control thread that all workers have completed pass 0 */
1613 opr_cv_signal(¶ms->master_cv);
1615 while (params->pass == 0) {
1616 opr_cv_wait(¶ms->cv, ¶ms->lock);
1620 opr_mutex_exit(¶ms->lock);
1623 pass = params->pass;
1624 opr_Assert(pass > 0);
1626 /* now escalate through the more complicated shutdowns */
1628 schedule_version_save = params->schedule_version;
1630 /* find a disk partition to work on */
1631 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1633 if (params->part_thread_target[id] && !params->part_done_pass[id]) {
1634 params->part_thread_target[id]--;
1641 /* hmm. for some reason the controller thread couldn't find anything for
1642 * us to do. let's see if there's anything we can do */
1643 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1645 if (diskP->vol_list.len && !params->part_done_pass[id]) {
1648 } else if (!params->part_done_pass[id]) {
1649 params->part_done_pass[id] = 1;
1650 params->n_parts_done_pass++;
1652 Log("VShutdown: done shutting down volumes on partition %s.\n",
1653 VPartitionPath(diskP));
1659 /* do work on this partition until either the controller
1660 * creates a new schedule, or we run out of things to do
1661 * on this partition */
1664 while (!params->part_done_pass[id] &&
1665 (schedule_version_save == params->schedule_version)) {
1666 /* ShutdownVolumeWalk_r will drop the glock internally */
1667 if (!ShutdownVolumeWalk_r(diskP, pass, ¶ms->part_pass_head[id])) {
1668 if (!params->part_done_pass[id]) {
1669 params->part_done_pass[id] = 1;
1670 params->n_parts_done_pass++;
1672 Log("VShutdown: done shutting down volumes on partition %s.\n",
1673 VPartitionPath(diskP));
1681 params->stats[pass][id] += count;
1683 /* ok, everyone is done this pass, proceed */
1686 params->n_threads_complete++;
1687 while (params->pass == pass) {
1688 if (params->n_threads_complete == params->n_threads) {
1689 /* we are the last thread to complete, so we will
1690 * reinitialize worker pool state for the next pass */
1691 params->n_threads_complete = 0;
1692 params->n_parts_done_pass = 0;
1694 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1696 params->part_done_pass[id] = 0;
1697 params->part_pass_head[id] = queue_First(&diskP->vol_list, rx_queue);
1700 /* compute a new thread schedule before releasing all the workers */
1701 ShutdownCreateSchedule(params);
1703 /* wake up all the workers */
1704 opr_cv_broadcast(¶ms->cv);
1707 Log("VShutdown: pass %d completed using %d threads on %d partitions\n",
1708 pass, params->n_threads, params->n_parts);
1711 VOL_CV_WAIT(¶ms->cv);
1714 pass = params->pass;
1728 /* shut down all volumes on a given disk partition
1730 * note that this function will not allow mp-fast
1731 * shutdown of a partition */
1733 VShutdownByPartition_r(struct DiskPartition64 * dp)
1739 /* wait for other exclusive ops to finish */
1740 VVByPListWait_r(dp);
1742 /* begin exclusive access */
1743 VVByPListBeginExclusive_r(dp);
1745 /* pick the low-hanging fruit first,
1746 * then do the complicated ones last
1747 * (has the advantage of keeping
1748 * in-use volumes up until the bitter end) */
1749 for (pass = 0, total=0; pass < 4; pass++) {
1750 pass_stats[pass] = ShutdownVByPForPass_r(dp, pass);
1751 total += pass_stats[pass];
1754 /* end exclusive access */
1755 VVByPListEndExclusive_r(dp);
1757 Log("VShutdownByPartition: shut down %d volumes on %s (pass[0]=%d, pass[1]=%d, pass[2]=%d, pass[3]=%d)\n",
1758 total, VPartitionPath(dp), pass_stats[0], pass_stats[1], pass_stats[2], pass_stats[3]);
1763 /* internal shutdown functionality
1765 * for multi-pass shutdown:
1766 * 0 to only "shutdown" {pre,un}attached and error state volumes
1767 * 1 to also shutdown attached volumes w/ volume header loaded
1768 * 2 to also shutdown attached volumes w/o volume header loaded
1769 * 3 to also shutdown exclusive state volumes
1771 * caller MUST hold exclusive access on the hash chain
1772 * because we drop vol_glock_mutex internally
1774 * this function is reentrant for passes 1--3
1775 * (e.g. multiple threads can cooperate to
1776 * shutdown a partition mp-fast)
1778 * pass 0 is not scaleable because the volume state data is
1779 * synchronized by vol_glock mutex, and the locking overhead
1780 * is too high to drop the lock long enough to do linked list
1784 ShutdownVByPForPass_r(struct DiskPartition64 * dp, int pass)
1786 struct rx_queue * q = queue_First(&dp->vol_list, rx_queue);
1789 while (ShutdownVolumeWalk_r(dp, pass, &q))
1795 /* conditionally shutdown one volume on partition dp
1796 * returns 1 if a volume was shutdown in this pass,
1799 ShutdownVolumeWalk_r(struct DiskPartition64 * dp, int pass,
1800 struct rx_queue ** idx)
1802 struct rx_queue *qp, *nqp;
1807 for (queue_ScanFrom(&dp->vol_list, qp, qp, nqp, rx_queue)) {
1808 vp = (Volume *) (((char *)qp) - offsetof(Volume, vol_list));
1812 if ((V_attachState(vp) != VOL_STATE_UNATTACHED) &&
1813 (V_attachState(vp) != VOL_STATE_ERROR) &&
1814 (V_attachState(vp) != VOL_STATE_DELETED) &&
1815 (V_attachState(vp) != VOL_STATE_PREATTACHED)) {
1819 if ((V_attachState(vp) == VOL_STATE_ATTACHED) &&
1820 (vp->header == NULL)) {
1824 if (VIsExclusiveState(V_attachState(vp))) {
1829 DeleteVolumeFromVByPList_r(vp);
1830 VShutdownVolume_r(vp);
1840 * shutdown a specific volume
1842 /* caller MUST NOT hold a heavyweight ref on vp */
1844 VShutdownVolume_r(Volume * vp)
1848 VCreateReservation_r(vp);
1850 if (LogLevel >= 5) {
1851 Log("VShutdownVolume_r: vid=%u, device=%d, state=%hu\n",
1852 vp->hashid, vp->partition->device, V_attachState(vp));
1855 /* wait for other blocking ops to finish */
1856 VWaitExclusiveState_r(vp);
1858 opr_Assert(VIsValidState(V_attachState(vp)));
1860 switch(V_attachState(vp)) {
1861 case VOL_STATE_SALVAGING:
1862 /* Leave salvaging volumes alone. Any in-progress salvages will
1863 * continue working after viced shuts down. This is intentional.
1866 case VOL_STATE_PREATTACHED:
1867 case VOL_STATE_ERROR:
1868 VChangeState_r(vp, VOL_STATE_UNATTACHED);
1869 case VOL_STATE_UNATTACHED:
1870 case VOL_STATE_DELETED:
1872 case VOL_STATE_GOING_OFFLINE:
1873 case VOL_STATE_SHUTTING_DOWN:
1874 case VOL_STATE_ATTACHED:
1878 Log("VShutdown: Attempting to take volume %u offline.\n",
1881 /* take the volume offline (drops reference count) */
1882 VOffline_r(vp, "File server was shut down");
1889 VCancelReservation_r(vp);
1893 #endif /* AFS_DEMAND_ATTACH_FS */
1896 /***************************************************/
1897 /* Header I/O routines */
1898 /***************************************************/
1901 HeaderName(bit32 magic)
1904 case VOLUMEINFOMAGIC:
1905 return "volume info";
1906 case SMALLINDEXMAGIC:
1907 return "small index";
1908 case LARGEINDEXMAGIC:
1909 return "large index";
1910 case LINKTABLEMAGIC:
1911 return "link table";
1916 /* open a descriptor for the inode (h),
1917 * read in an on-disk structure into buffer (to) of size (size),
1918 * verify versionstamp in structure has magic (magic) and
1919 * optionally verify version (version) if (version) is nonzero
1922 ReadHeader(Error * ec, IHandle_t * h, char *to, int size, bit32 magic,
1925 struct versionStamp *vsn;
1927 afs_sfsize_t nbytes;
1932 Log("ReadHeader: Null inode handle argument for %s header file.\n",
1940 Log("ReadHeader: Failed to open %s header file "
1941 "(volume=%u, inode=%s); errno=%d\n", HeaderName(magic), h->ih_vid,
1942 PrintInode(stmp, h->ih_ino), errno);
1947 vsn = (struct versionStamp *)to;
1948 nbytes = FDH_PREAD(fdP, to, size, 0);
1950 Log("ReadHeader: Failed to read %s header file "
1951 "(volume=%u, inode=%s); errno=%d\n", HeaderName(magic), h->ih_vid,
1952 PrintInode(stmp, h->ih_ino), errno);
1954 FDH_REALLYCLOSE(fdP);
1957 if (nbytes != size) {
1958 Log("ReadHeader: Incorrect number of bytes read from %s header file "
1959 "(volume=%u, inode=%s); expected=%d, read=%d\n",
1960 HeaderName(magic), h->ih_vid, PrintInode(stmp, h->ih_ino), size,
1963 FDH_REALLYCLOSE(fdP);
1966 if (vsn->magic != magic) {
1967 Log("ReadHeader: Incorrect magic for %s header file "
1968 "(volume=%u, inode=%s); expected=0x%x, read=0x%x\n",
1969 HeaderName(magic), h->ih_vid, PrintInode(stmp, h->ih_ino), magic,
1972 FDH_REALLYCLOSE(fdP);
1978 /* Check is conditional, in case caller wants to inspect version himself */
1979 if (version && vsn->version != version) {
1980 Log("ReadHeader: Incorrect version for %s header file "
1981 "(volume=%u, inode=%s); expected=%x, read=%x\n",
1982 HeaderName(magic), h->ih_vid, PrintInode(stmp, h->ih_ino),
1983 version, vsn->version);
1989 WriteVolumeHeader_r(Error * ec, Volume * vp)
1991 IHandle_t *h = V_diskDataHandle(vp);
2001 if (FDH_PWRITE(fdP, (char *)&V_disk(vp), sizeof(V_disk(vp)), 0)
2002 != sizeof(V_disk(vp))) {
2004 FDH_REALLYCLOSE(fdP);
2010 /* VolumeHeaderToDisk
2011 * Allows for storing 64 bit inode numbers in on-disk volume header
2014 /* convert in-memory representation of a volume header to the
2015 * on-disk representation of a volume header */
2017 VolumeHeaderToDisk(VolumeDiskHeader_t * dh, VolumeHeader_t * h)
2020 memset(dh, 0, sizeof(VolumeDiskHeader_t));
2021 dh->stamp = h->stamp;
2023 dh->parent = h->parent;
2025 #ifdef AFS_64BIT_IOPS_ENV
2026 dh->volumeInfo_lo = (afs_int32) h->volumeInfo & 0xffffffff;
2027 dh->volumeInfo_hi = (afs_int32) (h->volumeInfo >> 32) & 0xffffffff;
2028 dh->smallVnodeIndex_lo = (afs_int32) h->smallVnodeIndex & 0xffffffff;
2029 dh->smallVnodeIndex_hi =
2030 (afs_int32) (h->smallVnodeIndex >> 32) & 0xffffffff;
2031 dh->largeVnodeIndex_lo = (afs_int32) h->largeVnodeIndex & 0xffffffff;
2032 dh->largeVnodeIndex_hi =
2033 (afs_int32) (h->largeVnodeIndex >> 32) & 0xffffffff;
2034 dh->linkTable_lo = (afs_int32) h->linkTable & 0xffffffff;
2035 dh->linkTable_hi = (afs_int32) (h->linkTable >> 32) & 0xffffffff;
2037 dh->volumeInfo_lo = h->volumeInfo;
2038 dh->smallVnodeIndex_lo = h->smallVnodeIndex;
2039 dh->largeVnodeIndex_lo = h->largeVnodeIndex;
2040 dh->linkTable_lo = h->linkTable;
2044 /* DiskToVolumeHeader
2045 * Converts an on-disk representation of a volume header to
2046 * the in-memory representation of a volume header.
2048 * Makes the assumption that AFS has *always*
2049 * zero'd the volume header file so that high parts of inode
2050 * numbers are 0 in older (SGI EFS) volume header files.
2053 DiskToVolumeHeader(VolumeHeader_t * h, VolumeDiskHeader_t * dh)
2055 memset(h, 0, sizeof(VolumeHeader_t));
2056 h->stamp = dh->stamp;
2058 h->parent = dh->parent;
2060 #ifdef AFS_64BIT_IOPS_ENV
2062 (Inode) dh->volumeInfo_lo | ((Inode) dh->volumeInfo_hi << 32);
2064 h->smallVnodeIndex =
2065 (Inode) dh->smallVnodeIndex_lo | ((Inode) dh->
2066 smallVnodeIndex_hi << 32);
2068 h->largeVnodeIndex =
2069 (Inode) dh->largeVnodeIndex_lo | ((Inode) dh->
2070 largeVnodeIndex_hi << 32);
2072 (Inode) dh->linkTable_lo | ((Inode) dh->linkTable_hi << 32);
2074 h->volumeInfo = dh->volumeInfo_lo;
2075 h->smallVnodeIndex = dh->smallVnodeIndex_lo;
2076 h->largeVnodeIndex = dh->largeVnodeIndex_lo;
2077 h->linkTable = dh->linkTable_lo;
2082 /***************************************************/
2083 /* Volume Attachment routines */
2084 /***************************************************/
2086 #ifdef AFS_DEMAND_ATTACH_FS
2088 * pre-attach a volume given its path.
2090 * @param[out] ec outbound error code
2091 * @param[in] partition partition path string
2092 * @param[in] name volume id string
2094 * @return volume object pointer
2096 * @note A pre-attached volume will only have its partition
2097 * and hashid fields initialized. At first call to
2098 * VGetVolume, the volume will be fully attached.
2102 VPreAttachVolumeByName(Error * ec, char *partition, char *name)
2106 vp = VPreAttachVolumeByName_r(ec, partition, name);
2112 * pre-attach a volume given its path.
2114 * @param[out] ec outbound error code
2115 * @param[in] partition path to vice partition
2116 * @param[in] name volume id string
2118 * @return volume object pointer
2120 * @pre VOL_LOCK held
2122 * @internal volume package internal use only.
2125 VPreAttachVolumeByName_r(Error * ec, char *partition, char *name)
2127 return VPreAttachVolumeById_r(ec,
2129 VolumeNumber(name));
2133 * pre-attach a volume given its path and numeric volume id.
2135 * @param[out] ec error code return
2136 * @param[in] partition path to vice partition
2137 * @param[in] volumeId numeric volume id
2139 * @return volume object pointer
2141 * @pre VOL_LOCK held
2143 * @internal volume package internal use only.
2146 VPreAttachVolumeById_r(Error * ec,
2151 struct DiskPartition64 *partp;
2155 opr_Assert(programType == fileServer);
2157 if (!(partp = VGetPartition_r(partition, 0))) {
2159 Log("VPreAttachVolumeById_r: Error getting partition (%s)\n", partition);
2163 vp = VLookupVolume_r(ec, volumeId, NULL);
2168 return VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2172 * preattach a volume.
2174 * @param[out] ec outbound error code
2175 * @param[in] partp pointer to partition object
2176 * @param[in] vp pointer to volume object
2177 * @param[in] vid volume id
2179 * @return volume object pointer
2181 * @pre VOL_LOCK is held.
2183 * @warning Returned volume object pointer does not have to
2184 * equal the pointer passed in as argument vp. There
2185 * are potential race conditions which can result in
2186 * the pointers having different values. It is up to
2187 * the caller to make sure that references are handled
2188 * properly in this case.
2190 * @note If there is already a volume object registered with
2191 * the same volume id, its pointer MUST be passed as
2192 * argument vp. Failure to do so will result in a silent
2193 * failure to preattach.
2195 * @internal volume package internal use only.
2198 VPreAttachVolumeByVp_r(Error * ec,
2199 struct DiskPartition64 * partp,
2207 /* check to see if pre-attach already happened */
2209 (V_attachState(vp) != VOL_STATE_UNATTACHED) &&
2210 (V_attachState(vp) != VOL_STATE_DELETED) &&
2211 (V_attachState(vp) != VOL_STATE_PREATTACHED) &&
2212 !VIsErrorState(V_attachState(vp))) {
2214 * pre-attach is a no-op in all but the following cases:
2216 * - volume is unattached
2217 * - volume is in an error state
2218 * - volume is pre-attached
2220 Log("VPreattachVolumeByVp_r: volume %u not in quiescent state (state %u flags 0x%x)\n",
2221 vid, V_attachState(vp), V_attachFlags(vp));
2224 /* we're re-attaching a volume; clear out some old state */
2225 memset(&vp->salvage, 0, sizeof(struct VolumeOnlineSalvage));
2227 if (V_partition(vp) != partp) {
2228 /* XXX potential race */
2229 DeleteVolumeFromVByPList_r(vp);
2232 /* if we need to allocate a new Volume struct,
2233 * go ahead and drop the vol glock, otherwise
2234 * do the basic setup synchronised, as it's
2235 * probably not worth dropping the lock */
2238 /* allocate the volume structure */
2239 vp = nvp = calloc(1, sizeof(Volume));
2240 opr_Assert(vp != NULL);
2241 queue_Init(&vp->vnode_list);
2242 queue_Init(&vp->rx_call_list);
2243 opr_cv_init(&V_attachCV(vp));
2246 /* link the volume with its associated vice partition */
2247 vp->device = partp->device;
2248 vp->partition = partp;
2251 vp->specialStatus = 0;
2253 /* if we dropped the lock, reacquire the lock,
2254 * check for pre-attach races, and then add
2255 * the volume to the hash table */
2258 nvp = VLookupVolume_r(ec, vid, NULL);
2263 } else if (nvp) { /* race detected */
2268 /* hack to make up for VChangeState_r() decrementing
2269 * the old state counter */
2270 VStats.state_levels[0]++;
2274 /* put pre-attached volume onto the hash table
2275 * and bring it up to the pre-attached state */
2276 AddVolumeToHashTable(vp, vp->hashid);
2277 AddVolumeToVByPList_r(vp);
2278 VLRU_Init_Node_r(vp);
2279 VChangeState_r(vp, VOL_STATE_PREATTACHED);
2282 Log("VPreAttachVolumeByVp_r: volume %u pre-attached\n", vp->hashid);
2290 #endif /* AFS_DEMAND_ATTACH_FS */
2292 /* Attach an existing volume, given its pathname, and return a
2293 pointer to the volume header information. The volume also
2294 normally goes online at this time. An offline volume
2295 must be reattached to make it go online */
2297 VAttachVolumeByName(Error * ec, char *partition, char *name, int mode)
2301 retVal = VAttachVolumeByName_r(ec, partition, name, mode);
2307 VAttachVolumeByName_r(Error * ec, char *partition, char *name, int mode)
2310 struct DiskPartition64 *partp;
2315 #ifdef AFS_DEMAND_ATTACH_FS
2316 VolumeStats stats_save;
2318 #endif /* AFS_DEMAND_ATTACH_FS */
2322 volumeId = VolumeNumber(name);
2324 if (!(partp = VGetPartition_r(partition, 0))) {
2326 Log("VAttachVolume: Error getting partition (%s)\n", partition);
2330 if (VRequiresPartLock()) {
2331 opr_Assert(VInit == 3);
2332 VLockPartition_r(partition);
2333 } else if (programType == fileServer) {
2334 #ifdef AFS_DEMAND_ATTACH_FS
2335 /* lookup the volume in the hash table */
2336 vp = VLookupVolume_r(ec, volumeId, NULL);
2342 /* save any counters that are supposed to
2343 * be monotonically increasing over the
2344 * lifetime of the fileserver */
2345 memcpy(&stats_save, &vp->stats, sizeof(VolumeStats));
2347 memset(&stats_save, 0, sizeof(VolumeStats));
2350 /* if there's something in the hash table, and it's not
2351 * in the pre-attach state, then we may need to detach
2352 * it before proceeding */
2353 if (vp && (V_attachState(vp) != VOL_STATE_PREATTACHED)) {
2354 VCreateReservation_r(vp);
2355 VWaitExclusiveState_r(vp);
2357 /* at this point state must be one of:
2367 if (vp->specialStatus == VBUSY)
2370 /* if it's already attached, see if we can return it */
2371 if (V_attachState(vp) == VOL_STATE_ATTACHED) {
2372 VGetVolumeByVp_r(ec, vp);
2373 if (V_inUse(vp) == fileServer) {
2374 VCancelReservation_r(vp);
2378 /* otherwise, we need to detach, and attempt to re-attach */
2379 VDetachVolume_r(ec, vp);
2381 Log("VAttachVolume: Error detaching old volume instance (%s)\n", name);
2384 /* if it isn't fully attached, delete from the hash tables,
2385 and let the refcounter handle the rest */
2386 DeleteVolumeFromHashTable(vp);
2387 DeleteVolumeFromVByPList_r(vp);
2390 VCancelReservation_r(vp);
2394 /* pre-attach volume if it hasn't been done yet */
2396 (V_attachState(vp) == VOL_STATE_UNATTACHED) ||
2397 (V_attachState(vp) == VOL_STATE_DELETED) ||
2398 (V_attachState(vp) == VOL_STATE_ERROR)) {
2400 vp = VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2406 opr_Assert(vp != NULL);
2408 /* handle pre-attach races
2410 * multiple threads can race to pre-attach a volume,
2411 * but we can't let them race beyond that
2413 * our solution is to let the first thread to bring
2414 * the volume into an exclusive state win; the other
2415 * threads just wait until it finishes bringing the
2416 * volume online, and then they do a vgetvolumebyvp
2418 if (svp && (svp != vp)) {
2419 /* wait for other exclusive ops to finish */
2420 VCreateReservation_r(vp);
2421 VWaitExclusiveState_r(vp);
2423 /* get a heavyweight ref, kill the lightweight ref, and return */
2424 VGetVolumeByVp_r(ec, vp);
2425 VCancelReservation_r(vp);
2429 /* at this point, we are chosen as the thread to do
2430 * demand attachment for this volume. all other threads
2431 * doing a getvolume on vp->hashid will block until we finish */
2433 /* make sure any old header cache entries are invalidated
2434 * before proceeding */
2435 FreeVolumeHeader(vp);
2437 VChangeState_r(vp, VOL_STATE_ATTACHING);
2439 /* restore any saved counters */
2440 memcpy(&vp->stats, &stats_save, sizeof(VolumeStats));
2441 #else /* AFS_DEMAND_ATTACH_FS */
2442 vp = VGetVolume_r(ec, volumeId);
2444 if (V_inUse(vp) == fileServer)
2446 if (vp->specialStatus == VBUSY)
2448 VDetachVolume_r(ec, vp);
2450 Log("VAttachVolume: Error detaching volume (%s)\n", name);
2454 #endif /* AFS_DEMAND_ATTACH_FS */
2458 strcpy(path, VPartitionPath(partp));
2462 strcat(path, OS_DIRSEP);
2466 vp = (Volume *) calloc(1, sizeof(Volume));
2467 opr_Assert(vp != NULL);
2468 vp->hashid = volumeId;
2469 vp->device = partp->device;
2470 vp->partition = partp;
2471 queue_Init(&vp->vnode_list);
2472 queue_Init(&vp->rx_call_list);
2473 #ifdef AFS_DEMAND_ATTACH_FS
2474 opr_cv_init(&V_attachCV(vp));
2475 #endif /* AFS_DEMAND_ATTACH_FS */
2478 /* attach2 is entered without any locks, and returns
2479 * with vol_glock_mutex held */
2480 vp = attach2(ec, volumeId, path, partp, vp, isbusy, mode, &checkedOut);
2482 if (VCanUseFSSYNC() && vp) {
2483 #ifdef AFS_DEMAND_ATTACH_FS
2484 if ((mode == V_VOLUPD) || (VolumeWriteable(vp) && (mode == V_CLONE))) {
2485 /* mark volume header as in use so that volser crashes lead to a
2486 * salvage attempt */
2487 VUpdateVolume_r(ec, vp, 0);
2489 /* for dafs, we should tell the fileserver, except for V_PEEK
2490 * where we know it is not necessary */
2491 if (mode == V_PEEK) {
2492 vp->needsPutBack = 0;
2494 vp->needsPutBack = VOL_PUTBACK;
2496 #else /* !AFS_DEMAND_ATTACH_FS */
2497 /* duplicate computation in fssync.c about whether the server
2498 * takes the volume offline or not. If the volume isn't
2499 * offline, we must not return it when we detach the volume,
2500 * or the server will abort */
2501 if (mode == V_READONLY || mode == V_PEEK
2502 || (!VolumeWriteable(vp) && (mode == V_CLONE || mode == V_DUMP)))
2503 vp->needsPutBack = 0;
2505 vp->needsPutBack = VOL_PUTBACK;
2506 #endif /* !AFS_DEMAND_ATTACH_FS */
2508 #ifdef FSSYNC_BUILD_CLIENT
2509 /* Only give back the vol to the fileserver if we checked it out; attach2
2510 * will set checkedOut only if we successfully checked it out from the
2512 if (VCanUseFSSYNC() && vp == NULL && checkedOut) {
2514 #ifdef AFS_DEMAND_ATTACH_FS
2515 /* If we couldn't attach but we scheduled a salvage, we already
2516 * notified the fileserver; don't online it now */
2517 if (*ec != VSALVAGING)
2518 #endif /* AFS_DEMAND_ATTACH_FS */
2519 FSYNC_VolOp(volumeId, partition, FSYNC_VOL_ON, 0, NULL);
2522 if (programType == fileServer && vp) {
2523 #ifdef AFS_DEMAND_ATTACH_FS
2525 * we can get here in cases where we don't "own"
2526 * the volume (e.g. volume owned by a utility).
2527 * short circuit around potential disk header races.
2529 if (V_attachState(vp) != VOL_STATE_ATTACHED) {
2533 VUpdateVolume_r(ec, vp, 0);
2535 Log("VAttachVolume: Error updating volume\n");
2540 if (VolumeWriteable(vp) && V_dontSalvage(vp) == 0) {
2541 #ifndef AFS_DEMAND_ATTACH_FS
2542 /* This is a hack: by temporarily setting the incore
2543 * dontSalvage flag ON, the volume will be put back on the
2544 * Update list (with dontSalvage OFF again). It will then
2545 * come back in N minutes with DONT_SALVAGE eventually
2546 * set. This is the way that volumes that have never had
2547 * it set get it set; or that volumes that have been
2548 * offline without DONT SALVAGE having been set also
2549 * eventually get it set */
2550 V_dontSalvage(vp) = DONT_SALVAGE;
2551 #endif /* !AFS_DEMAND_ATTACH_FS */
2552 VAddToVolumeUpdateList_r(ec, vp);
2554 Log("VAttachVolume: Error adding volume to update list\n");
2561 Log("VOnline: volume %u (%s) attached and online\n", V_id(vp),
2566 if (VRequiresPartLock()) {
2567 VUnlockPartition_r(partition);
2570 #ifdef AFS_DEMAND_ATTACH_FS
2571 /* attach failed; make sure we're in error state */
2572 if (vp && !VIsErrorState(V_attachState(vp))) {
2573 VChangeState_r(vp, VOL_STATE_ERROR);
2575 #endif /* AFS_DEMAND_ATTACH_FS */
2582 #ifdef AFS_DEMAND_ATTACH_FS
2583 /* VAttachVolumeByVp_r
2585 * finish attaching a volume that is
2586 * in a less than fully attached state
2588 /* caller MUST hold a ref count on vp */
2590 VAttachVolumeByVp_r(Error * ec, Volume * vp, int mode)
2592 char name[VMAXPATHLEN];
2594 struct DiskPartition64 *partp;
2598 Volume * nvp = NULL;
2599 VolumeStats stats_save;
2603 /* volume utility should never call AttachByVp */
2604 opr_Assert(programType == fileServer);
2606 volumeId = vp->hashid;
2607 partp = vp->partition;
2608 VolumeExternalName_r(volumeId, name, sizeof(name));
2611 /* if another thread is performing a blocking op, wait */
2612 VWaitExclusiveState_r(vp);
2614 memcpy(&stats_save, &vp->stats, sizeof(VolumeStats));
2616 /* if it's already attached, see if we can return it */
2617 if (V_attachState(vp) == VOL_STATE_ATTACHED) {
2618 VGetVolumeByVp_r(ec, vp);
2619 if (V_inUse(vp) == fileServer) {
2622 if (vp->specialStatus == VBUSY)
2624 VDetachVolume_r(ec, vp);
2626 Log("VAttachVolume: Error detaching volume (%s)\n", name);
2632 /* pre-attach volume if it hasn't been done yet */
2634 (V_attachState(vp) == VOL_STATE_UNATTACHED) ||
2635 (V_attachState(vp) == VOL_STATE_DELETED) ||
2636 (V_attachState(vp) == VOL_STATE_ERROR)) {
2637 nvp = VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2643 VCreateReservation_r(nvp);
2648 opr_Assert(vp != NULL);
2649 VChangeState_r(vp, VOL_STATE_ATTACHING);
2651 /* restore monotonically increasing stats */
2652 memcpy(&vp->stats, &stats_save, sizeof(VolumeStats));
2656 /* compute path to disk header */
2657 strcpy(path, VPartitionPath(partp));
2661 strcat(path, OS_DIRSEP);
2666 * NOTE: attach2 is entered without any locks, and returns
2667 * with vol_glock_mutex held */
2668 vp = attach2(ec, volumeId, path, partp, vp, isbusy, mode, &checkedOut);
2671 * the event that an error was encountered, or
2672 * the volume was not brought to an attached state
2673 * for any reason, skip to the end. We cannot
2674 * safely call VUpdateVolume unless we "own" it.
2678 (V_attachState(vp) != VOL_STATE_ATTACHED)) {
2682 VUpdateVolume_r(ec, vp, 0);
2684 Log("VAttachVolume: Error updating volume %u\n", vp->hashid);
2688 if (VolumeWriteable(vp) && V_dontSalvage(vp) == 0) {
2689 #ifndef AFS_DEMAND_ATTACH_FS
2690 /* This is a hack: by temporarily setting the incore
2691 * dontSalvage flag ON, the volume will be put back on the
2692 * Update list (with dontSalvage OFF again). It will then
2693 * come back in N minutes with DONT_SALVAGE eventually
2694 * set. This is the way that volumes that have never had
2695 * it set get it set; or that volumes that have been
2696 * offline without DONT SALVAGE having been set also
2697 * eventually get it set */
2698 V_dontSalvage(vp) = DONT_SALVAGE;
2699 #endif /* !AFS_DEMAND_ATTACH_FS */
2700 VAddToVolumeUpdateList_r(ec, vp);
2702 Log("VAttachVolume: Error adding volume %u to update list\n", vp->hashid);
2709 Log("VOnline: volume %u (%s) attached and online\n", V_id(vp),
2713 VCancelReservation_r(nvp);
2716 if (*ec && (*ec != VOFFLINE) && (*ec != VSALVAGE)) {
2717 if (vp && !VIsErrorState(V_attachState(vp))) {
2718 VChangeState_r(vp, VOL_STATE_ERROR);
2727 * lock a volume on disk (non-blocking).
2729 * @param[in] vp The volume to lock
2730 * @param[in] locktype READ_LOCK or WRITE_LOCK
2732 * @return operation status
2733 * @retval 0 success, lock was obtained
2734 * @retval EBUSY a conflicting lock was held by another process
2735 * @retval EIO error acquiring lock
2737 * @pre If we're in the fileserver, vp is in an exclusive state
2739 * @pre vp is not already locked
2742 VLockVolumeNB(Volume *vp, int locktype)
2746 opr_Assert(programType != fileServer
2747 || VIsExclusiveState(V_attachState(vp)));
2748 opr_Assert(!(V_attachFlags(vp) & VOL_LOCKED));
2750 code = VLockVolumeByIdNB(vp->hashid, vp->partition, locktype);
2752 V_attachFlags(vp) |= VOL_LOCKED;
2759 * unlock a volume on disk that was locked with VLockVolumeNB.
2761 * @param[in] vp volume to unlock
2763 * @pre If we're in the fileserver, vp is in an exclusive state
2765 * @pre vp has already been locked
2768 VUnlockVolume(Volume *vp)
2770 opr_Assert(programType != fileServer
2771 || VIsExclusiveState(V_attachState(vp)));
2772 opr_Assert((V_attachFlags(vp) & VOL_LOCKED));
2774 VUnlockVolumeById(vp->hashid, vp->partition);
2776 V_attachFlags(vp) &= ~VOL_LOCKED;
2778 #endif /* AFS_DEMAND_ATTACH_FS */
2781 * read in a vol header, possibly lock the vol header, and possibly check out
2782 * the vol header from the fileserver, as part of volume attachment.
2784 * @param[out] ec error code
2785 * @param[in] vp volume pointer object
2786 * @param[in] partp disk partition object of the attaching partition
2787 * @param[in] mode attachment mode such as V_VOLUPD, V_DUMP, etc (see
2789 * @param[in] peek 1 to just try to read in the volume header and make sure
2790 * we don't try to lock the vol, or check it out from
2791 * FSSYNC or anything like that; 0 otherwise, for 'normal'
2793 * @param[out] acheckedOut If we successfully checked-out the volume from
2794 * the fileserver (if we needed to), this is set
2795 * to 1, otherwise it is untouched.
2797 * @note As part of DAFS volume attachment, the volume header may be either
2798 * read- or write-locked to ensure mutual exclusion of certain volume
2799 * operations. In some cases in order to determine whether we need to
2800 * read- or write-lock the header, we need to read in the header to see
2801 * if the volume is RW or not. So, if we read in the header under a
2802 * read-lock and determine that we actually need a write-lock on the
2803 * volume header, this function will drop the read lock, acquire a write
2804 * lock, and read the header in again.
2807 attach_volume_header(Error *ec, Volume *vp, struct DiskPartition64 *partp,
2808 int mode, int peek, int *acheckedOut)
2810 struct VolumeDiskHeader diskHeader;
2811 struct VolumeHeader header;
2814 int lock_tries = 0, checkout_tries = 0;
2816 VolumeId volid = vp->hashid;
2817 #ifdef FSSYNC_BUILD_CLIENT
2818 int checkout, done_checkout = 0;
2819 #endif /* FSSYNC_BUILD_CLIENT */
2820 #ifdef AFS_DEMAND_ATTACH_FS
2821 int locktype = 0, use_locktype = -1;
2822 #endif /* AFS_DEMAND_ATTACH_FS */
2828 if (lock_tries > VOL_MAX_CHECKOUT_RETRIES) {
2829 Log("VAttachVolume: retried too many times trying to lock header for "
2830 "vol %lu part %s; giving up\n", afs_printable_uint32_lu(volid),
2831 VPartitionPath(partp));
2835 if (checkout_tries > VOL_MAX_CHECKOUT_RETRIES) {
2836 Log("VAttachVolume: retried too many times trying to checkout "
2837 "vol %lu part %s; giving up\n", afs_printable_uint32_lu(volid),
2838 VPartitionPath(partp));
2843 if (VReadVolumeDiskHeader(volid, partp, NULL)) {
2844 /* short-circuit the 'volume does not exist' case */
2849 #ifdef FSSYNC_BUILD_CLIENT
2850 checkout = !done_checkout;
2852 if (!peek && checkout && VMustCheckoutVolume(mode)) {
2854 memset(&res, 0, sizeof(res));
2856 if (FSYNC_VolOp(volid, partp->name, FSYNC_VOL_NEEDVOLUME, mode, &res)
2859 if (res.hdr.reason == FSYNC_SALVAGE) {
2860 Log("VAttachVolume: file server says volume %lu is salvaging\n",
2861 afs_printable_uint32_lu(volid));
2864 Log("VAttachVolume: attach of volume %lu apparently denied by file server\n",
2865 afs_printable_uint32_lu(volid));
2866 *ec = VNOVOL; /* XXXX */
2874 #ifdef AFS_DEMAND_ATTACH_FS
2875 if (use_locktype < 0) {
2876 /* don't know whether vol is RO or RW; assume it's RO and we can retry
2877 * if it turns out to be RW */
2878 locktype = VVolLockType(mode, 0);
2881 /* a previous try says we should use use_locktype to lock the volume,
2883 locktype = use_locktype;
2886 if (!peek && locktype) {
2887 code = VLockVolumeNB(vp, locktype);
2889 if (code == EBUSY) {
2890 Log("VAttachVolume: another program has vol %lu locked\n",
2891 afs_printable_uint32_lu(volid));
2893 Log("VAttachVolume: error %d trying to lock vol %lu\n",
2894 code, afs_printable_uint32_lu(volid));
2901 #endif /* AFS_DEMAND_ATTACH_FS */
2903 code = VReadVolumeDiskHeader(volid, partp, &diskHeader);
2913 DiskToVolumeHeader(&header, &diskHeader);
2915 IH_INIT(vp->vnodeIndex[vLarge].handle, partp->device, header.parent,
2916 header.largeVnodeIndex);
2917 IH_INIT(vp->vnodeIndex[vSmall].handle, partp->device, header.parent,
2918 header.smallVnodeIndex);
2919 IH_INIT(vp->diskDataHandle, partp->device, header.parent,
2921 IH_INIT(vp->linkHandle, partp->device, header.parent, header.linkTable);
2924 /* only need to do this once */
2926 GetVolumeHeader(vp);
2930 #if defined(AFS_DEMAND_ATTACH_FS) && defined(FSSYNC_BUILD_CLIENT)
2931 /* demand attach changes the V_PEEK mechanism
2933 * we can now suck the current disk data structure over
2934 * the fssync interface without going to disk
2936 * (technically, we don't need to restrict this feature
2937 * to demand attach fileservers. However, I'm trying
2938 * to limit the number of common code changes)
2940 if (VCanUseFSSYNC() && (mode == V_PEEK || peek)) {
2942 res.payload.len = sizeof(VolumeDiskData);
2943 res.payload.buf = &vp->header->diskstuff;
2945 if (FSYNC_VolOp(vp->hashid,
2947 FSYNC_VOL_QUERY_HDR,
2950 goto disk_header_loaded;
2953 #endif /* AFS_DEMAND_ATTACH_FS && FSSYNC_BUILD_CLIENT */
2954 (void)ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
2955 sizeof(V_disk(vp)), VOLUMEINFOMAGIC, VOLUMEINFOVERSION);
2957 #ifdef AFS_DEMAND_ATTACH_FS
2960 IncUInt64(&VStats.hdr_loads);
2961 IncUInt64(&vp->stats.hdr_loads);
2963 #endif /* AFS_DEMAND_ATTACH_FS */
2966 Log("VAttachVolume: Error reading diskDataHandle header for vol %lu; "
2967 "error=%u\n", afs_printable_uint32_lu(volid), *ec);
2971 #ifdef AFS_DEMAND_ATTACH_FS
2972 # ifdef FSSYNC_BUILD_CLIENT
2974 # endif /* FSSYNC_BUILD_CLIENT */
2976 /* if the lock type we actually used to lock the volume is different than
2977 * the lock type we should have used, retry with the lock type we should
2979 use_locktype = VVolLockType(mode, VolumeWriteable(vp));
2980 if (locktype != use_locktype) {
2984 #endif /* AFS_DEMAND_ATTACH_FS */
2989 #if defined(AFS_DEMAND_ATTACH_FS) && defined(FSSYNC_BUILD_CLIENT)
2990 if (!peek && *ec == 0 && retry == 0 && VMustCheckoutVolume(mode)) {
2992 code = FSYNC_VerifyCheckout(volid, partp->name, FSYNC_VOL_NEEDVOLUME, mode);
2994 if (code == SYNC_DENIED) {
2995 /* must retry checkout; fileserver no longer thinks we have
3001 } else if (code != SYNC_OK) {
3005 #endif /* AFS_DEMAND_ATTACH_FS && FSSYNC_BUILD_CLIENT */
3008 /* either we are going to be called again for a second pass, or we
3009 * encountered an error; clean up in either case */
3011 #ifdef AFS_DEMAND_ATTACH_FS
3012 if ((V_attachFlags(vp) & VOL_LOCKED)) {
3015 #endif /* AFS_DEMAND_ATTACH_FS */
3016 if (vp->linkHandle) {
3017 IH_RELEASE(vp->vnodeIndex[vLarge].handle);
3018 IH_RELEASE(vp->vnodeIndex[vSmall].handle);
3019 IH_RELEASE(vp->diskDataHandle);
3020 IH_RELEASE(vp->linkHandle);
3026 FreeVolumeHeader(vp);
3036 #ifdef AFS_DEMAND_ATTACH_FS
3038 attach_check_vop(Error *ec, VolumeId volid, struct DiskPartition64 *partp,
3039 Volume *vp, int *acheckedOut)
3043 if (vp->pending_vol_op) {
3047 if (vp->pending_vol_op->vol_op_state == FSSYNC_VolOpRunningUnknown) {
3049 code = VVolOpLeaveOnlineNoHeader_r(vp, vp->pending_vol_op);
3051 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOnline;
3052 } else if (code == 0) {
3053 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOffline;
3056 /* we need the vol header to determine if the volume can be
3057 * left online for the vop, so... get the header */
3061 /* attach header with peek=1 to avoid checking out the volume
3062 * or locking it; we just want the header info, we're not
3063 * messing with the volume itself at all */
3064 attach_volume_header(ec, vp, partp, V_PEEK, 1, acheckedOut);
3071 if (VVolOpLeaveOnline_r(vp, vp->pending_vol_op)) {
3072 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOnline;
3074 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOffline;
3077 /* make sure we grab a new vol header and re-open stuff on
3078 * actual attachment; we can't keep the data we grabbed, since
3079 * it was not done under a lock and thus not safe */
3080 FreeVolumeHeader(vp);
3081 VReleaseVolumeHandles_r(vp);
3084 /* see if the pending volume op requires exclusive access */
3085 switch (vp->pending_vol_op->vol_op_state) {
3086 case FSSYNC_VolOpPending:
3087 /* this should never happen */
3088 opr_Assert(vp->pending_vol_op->vol_op_state
3089 != FSSYNC_VolOpPending);
3092 case FSSYNC_VolOpRunningUnknown:
3093 /* this should never happen; we resolved 'unknown' above */
3094 opr_Assert(vp->pending_vol_op->vol_op_state
3095 != FSSYNC_VolOpRunningUnknown);
3098 case FSSYNC_VolOpRunningOffline:
3099 /* mark the volume down */
3101 VChangeState_r(vp, VOL_STATE_UNATTACHED);
3103 /* do not set V_offlineMessage here; we don't have ownership of
3104 * the volume (and probably do not have the header loaded), so we
3105 * can't alter the disk header */
3107 /* check to see if we should set the specialStatus flag */
3108 if (VVolOpSetVBusy_r(vp, vp->pending_vol_op)) {
3109 /* don't overwrite specialStatus if it was already set to
3110 * something else (e.g. VMOVED) */
3111 if (!vp->specialStatus) {
3112 vp->specialStatus = VBUSY;
3124 #endif /* AFS_DEMAND_ATTACH_FS */
3127 * volume attachment helper function.
3129 * @param[out] ec error code
3130 * @param[in] volumeId volume ID of the attaching volume
3131 * @param[in] path full path to the volume header .vol file
3132 * @param[in] partp disk partition object for the attaching partition
3133 * @param[in] vp volume object; vp->hashid, vp->device, vp->partition,
3134 * vp->vnode_list, vp->rx_call_list, and V_attachCV (for
3135 * DAFS) should already be initialized
3136 * @param[in] isbusy 1 if vp->specialStatus should be set to VBUSY; that is,
3137 * if there is a volume operation running for this volume
3138 * that should set the volume to VBUSY during its run. 0
3139 * otherwise. (see VVolOpSetVBusy_r)
3140 * @param[in] mode attachment mode such as V_VOLUPD, V_DUMP, etc (see
3142 * @param[out] acheckedOut If we successfully checked-out the volume from
3143 * the fileserver (if we needed to), this is set
3144 * to 1, otherwise it is 0.
3146 * @return pointer to the semi-attached volume pointer
3147 * @retval NULL an error occurred (check value of *ec)
3148 * @retval vp volume successfully attaching
3150 * @pre no locks held
3152 * @post VOL_LOCK held
3155 attach2(Error * ec, VolId volumeId, char *path, struct DiskPartition64 *partp,
3156 Volume * vp, int isbusy, int mode, int *acheckedOut)
3158 /* have we read in the header successfully? */
3159 int read_header = 0;
3161 #ifdef AFS_DEMAND_ATTACH_FS
3162 /* should we FreeVolume(vp) instead of VCheckFree(vp) in the error
3166 /* in the case of an error, to what state should the volume be
3168 VolState error_state = VOL_STATE_ERROR;
3169 #endif /* AFS_DEMAND_ATTACH_FS */
3173 vp->vnodeIndex[vLarge].handle = NULL;
3174 vp->vnodeIndex[vSmall].handle = NULL;
3175 vp->diskDataHandle = NULL;
3176 vp->linkHandle = NULL;
3180 #ifdef AFS_DEMAND_ATTACH_FS
3181 attach_check_vop(ec, volumeId, partp, vp, acheckedOut);
3183 attach_volume_header(ec, vp, partp, mode, 0, acheckedOut);
3186 attach_volume_header(ec, vp, partp, mode, 0, acheckedOut);
3187 #endif /* !AFS_DEMAND_ATTACH_FS */
3189 if (*ec == VNOVOL) {
3190 /* if the volume doesn't exist, skip straight to 'error' so we don't
3191 * request a salvage */
3192 goto unlocked_error;
3198 /* ensure that we don't override specialStatus if it was set to
3199 * something else (e.g. VMOVED) */
3200 if (isbusy && !vp->specialStatus) {
3201 vp->specialStatus = VBUSY;
3203 vp->shuttingDown = 0;
3204 vp->goingOffline = 0;
3206 #ifdef AFS_DEMAND_ATTACH_FS
3207 vp->stats.last_attach = FT_ApproxTime();
3208 vp->stats.attaches++;
3212 IncUInt64(&VStats.attaches);
3213 vp->cacheCheck = ++VolumeCacheCheck;
3214 /* just in case this ever rolls over */
3215 if (!vp->cacheCheck)
3216 vp->cacheCheck = ++VolumeCacheCheck;
3219 #ifdef AFS_DEMAND_ATTACH_FS
3220 V_attachFlags(vp) |= VOL_HDR_LOADED;
3221 vp->stats.last_hdr_load = vp->stats.last_attach;
3222 #endif /* AFS_DEMAND_ATTACH_FS */
3226 struct IndexFileHeader iHead;
3229 * We just read in the diskstuff part of the header. If the detailed
3230 * volume stats area has not yet been initialized, we should bzero the
3231 * area and mark it as initialized.
3233 if (!(V_stat_initialized(vp))) {
3234 memset((V_stat_area(vp)), 0, VOL_STATS_BYTES);
3235 V_stat_initialized(vp) = 1;
3238 (void)ReadHeader(ec, vp->vnodeIndex[vSmall].handle,
3239 (char *)&iHead, sizeof(iHead),
3240 SMALLINDEXMAGIC, SMALLINDEXVERSION);
3243 Log("VAttachVolume: Error reading smallVnode vol header %s; error=%u\n", path, *ec);
3248 struct IndexFileHeader iHead;
3250 (void)ReadHeader(ec, vp->vnodeIndex[vLarge].handle,
3251 (char *)&iHead, sizeof(iHead),
3252 LARGEINDEXMAGIC, LARGEINDEXVERSION);
3255 Log("VAttachVolume: Error reading largeVnode vol header %s; error=%u\n", path, *ec);
3259 #ifdef AFS_NAMEI_ENV
3261 struct versionStamp stamp;
3263 (void)ReadHeader(ec, V_linkHandle(vp), (char *)&stamp,
3264 sizeof(stamp), LINKTABLEMAGIC, LINKTABLEVERSION);
3267 Log("VAttachVolume: Error reading namei vol header %s; error=%u\n", path, *ec);
3270 #endif /* AFS_NAMEI_ENV */
3272 #if defined(AFS_DEMAND_ATTACH_FS)
3273 if (*ec && ((*ec != VOFFLINE) || (V_attachState(vp) != VOL_STATE_UNATTACHED))) {
3275 if (!VCanScheduleSalvage()) {
3276 Log("VAttachVolume: Error attaching volume %s; volume needs salvage; error=%u\n", path, *ec);
3278 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_NO_OFFLINE);
3283 /* volume operation in progress */
3285 /* we have already transitioned the vp away from ATTACHING state, so we
3286 * can go right to the end of attach2, and we do not need to transition
3288 goto error_notbroken;
3290 #else /* AFS_DEMAND_ATTACH_FS */
3292 Log("VAttachVolume: Error attaching volume %s; volume needs salvage; error=%u\n", path, *ec);
3293 goto unlocked_error;
3295 #endif /* AFS_DEMAND_ATTACH_FS */
3297 if (V_needsSalvaged(vp)) {
3298 if (vp->specialStatus)
3299 vp->specialStatus = 0;
3301 #if defined(AFS_DEMAND_ATTACH_FS)
3302 if (!VCanScheduleSalvage()) {
3303 Log("VAttachVolume: volume salvage flag is ON for %s; volume needs salvage\n", path);
3305 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_NO_OFFLINE);
3308 #else /* AFS_DEMAND_ATTACH_FS */
3310 #endif /* AFS_DEMAND_ATTACH_FS */
3316 vp->nextVnodeUnique = V_uniquifier(vp);
3318 if (VShouldCheckInUse(mode) && V_inUse(vp) && VolumeWriteable(vp)) {
3319 if (!V_needsSalvaged(vp)) {
3320 V_needsSalvaged(vp) = 1;
3321 VUpdateVolume_r(ec, vp, 0);
3323 #if defined(AFS_DEMAND_ATTACH_FS)
3324 if (!VCanScheduleSalvage()) {
3325 Log("VAttachVolume: volume %s needs to be salvaged; not attached.\n", path);
3327 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_NO_OFFLINE);
3330 #else /* AFS_DEMAND_ATTACH_FS */
3331 Log("VAttachVolume: volume %s needs to be salvaged; not attached.\n", path);
3333 #endif /* AFS_DEMAND_ATTACH_FS */
3338 if (programType == fileServer && V_destroyMe(vp) == DESTROY_ME) {
3339 /* Only check destroyMe if we are the fileserver, since the
3340 * volserver et al sometimes need to work with volumes with
3341 * destroyMe set. Examples are 'temporary' volumes the
3342 * volserver creates, and when we create a volume (destroyMe
3343 * is set on creation; sometimes a separate volserver
3344 * transaction is created to clear destroyMe).
3347 #if defined(AFS_DEMAND_ATTACH_FS)
3348 /* schedule a salvage so the volume goes away on disk */
3349 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_NO_OFFLINE);
3350 VChangeState_r(vp, VOL_STATE_ERROR);
3353 #endif /* AFS_DEMAND_ATTACH_FS */
3354 Log("VAttachVolume: volume %s is junk; it should be destroyed at next salvage\n", path);
3359 vp->vnodeIndex[vSmall].bitmap = vp->vnodeIndex[vLarge].bitmap = NULL;
3360 #ifndef BITMAP_LATER
3361 if (programType == fileServer && VolumeWriteable(vp)) {
3363 for (i = 0; i < nVNODECLASSES; i++) {
3364 VGetBitmap_r(ec, vp, i);
3366 #ifdef AFS_DEMAND_ATTACH_FS
3367 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_NO_OFFLINE);
3369 #endif /* AFS_DEMAND_ATTACH_FS */
3370 Log("VAttachVolume: error getting bitmap for volume (%s)\n",
3376 #endif /* BITMAP_LATER */
3378 if (VInit >= 2 && V_needsCallback(vp)) {
3379 if (V_BreakVolumeCallbacks) {
3380 Log("VAttachVolume: Volume %lu was changed externally; breaking callbacks\n",
3381 afs_printable_uint32_lu(V_id(vp)));
3382 V_needsCallback(vp) = 0;
3384 (*V_BreakVolumeCallbacks) (V_id(vp));
3387 VUpdateVolume_r(ec, vp, 0);
3389 #ifdef FSSYNC_BUILD_CLIENT
3390 else if (VCanUseFSSYNC()) {
3391 afs_int32 fsync_code;
3393 V_needsCallback(vp) = 0;
3395 fsync_code = FSYNC_VolOp(V_id(vp), NULL, FSYNC_VOL_BREAKCBKS, FSYNC_WHATEVER, NULL);
3399 V_needsCallback(vp) = 1;
3400 Log("Error trying to tell the fileserver to break callbacks for "
3401 "changed volume %lu; error code %ld\n",
3402 afs_printable_uint32_lu(V_id(vp)),
3403 afs_printable_int32_ld(fsync_code));
3405 VUpdateVolume_r(ec, vp, 0);
3408 #endif /* FSSYNC_BUILD_CLIENT */
3411 Log("VAttachVolume: error %d clearing needsCallback on volume "
3412 "%lu; needs salvage\n", (int)*ec,
3413 afs_printable_uint32_lu(V_id(vp)));
3414 #ifdef AFS_DEMAND_ATTACH_FS
3415 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_NO_OFFLINE);
3417 #else /* !AFS_DEMAND_ATTACH_FS */
3419 #endif /* !AFS_DEMAND_ATTACh_FS */
3424 if (programType == fileServer) {
3425 if (vp->specialStatus)
3426 vp->specialStatus = 0;
3427 if (V_blessed(vp) && V_inService(vp) && !V_needsSalvaged(vp)) {
3428 V_inUse(vp) = fileServer;
3429 V_offlineMessage(vp)[0] = '\0';
3433 #ifdef AFS_DEMAND_ATTACH_FS
3434 /* Put the vol into PREATTACHED state, so if someone tries to
3435 * access it again, we try to attach, see that we're not blessed,
3436 * and give a VNOVOL error again. Putting it into UNATTACHED state
3437 * would result in a VOFFLINE error instead. */
3438 error_state = VOL_STATE_PREATTACHED;
3439 #endif /* AFS_DEMAND_ATTACH_FS */
3441 /* mimic e.g. GetVolume errors */
3442 if (!V_blessed(vp)) {
3443 Log("Volume %lu offline: not blessed\n", afs_printable_uint32_lu(V_id(vp)));
3444 FreeVolumeHeader(vp);
3445 } else if (!V_inService(vp)) {
3446 Log("Volume %lu offline: not in service\n", afs_printable_uint32_lu(V_id(vp)));
3447 FreeVolumeHeader(vp);
3449 Log("Volume %lu offline: needs salvage\n", afs_printable_uint32_lu(V_id(vp)));
3451 #ifdef AFS_DEMAND_ATTACH_FS
3452 error_state = VOL_STATE_ERROR;
3453 /* see if we can recover */
3454 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_NO_OFFLINE);
3457 #ifdef AFS_DEMAND_ATTACH_FS
3463 #ifdef AFS_DEMAND_ATTACH_FS
3464 if ((mode != V_PEEK) && (mode != V_SECRETLY) && (mode != V_READONLY))
3465 V_inUse(vp) = programType;
3466 #endif /* AFS_DEMAND_ATTACH_FS */
3467 V_checkoutMode(vp) = mode;
3470 AddVolumeToHashTable(vp, V_id(vp));
3471 #ifdef AFS_DEMAND_ATTACH_FS
3472 if (VCanUnlockAttached() && (V_attachFlags(vp) & VOL_LOCKED)) {
3475 if ((programType != fileServer) ||
3476 (V_inUse(vp) == fileServer)) {
3477 AddVolumeToVByPList_r(vp);
3479 VChangeState_r(vp, VOL_STATE_ATTACHED);
3481 VChangeState_r(vp, VOL_STATE_UNATTACHED);
3490 #ifdef AFS_DEMAND_ATTACH_FS
3491 if (!VIsErrorState(V_attachState(vp))) {
3492 if (VIsErrorState(error_state)) {
3493 Log("attach2: forcing vol %u to error state (state %u flags 0x%x ec %d)\n",
3494 vp->hashid, V_attachState(vp), V_attachFlags(vp), *ec);
3496 VChangeState_r(vp, error_state);
3498 #endif /* AFS_DEMAND_ATTACH_FS */
3501 VReleaseVolumeHandles_r(vp);
3504 #ifdef AFS_DEMAND_ATTACH_FS
3506 if (VCheckSalvage(vp) == VCHECK_SALVAGE_FAIL) {
3507 /* The salvage could not be scheduled with the salvage server
3508 * due to a hard error. Reset the error code to prevent retry loops by
3510 if (*ec == VSALVAGING) {
3519 #else /* !AFS_DEMAND_ATTACH_FS */
3521 #endif /* !AFS_DEMAND_ATTACH_FS */
3525 /* Attach an existing volume.
3526 The volume also normally goes online at this time.
3527 An offline volume must be reattached to make it go online.
3531 VAttachVolume(Error * ec, VolumeId volumeId, int mode)
3535 retVal = VAttachVolume_r(ec, volumeId, mode);
3541 VAttachVolume_r(Error * ec, VolumeId volumeId, int mode)
3544 VGetVolumePath(ec, volumeId, &part, &name);
3548 vp = VGetVolume_r(&error, volumeId);
3550 opr_Assert(V_inUse(vp) == 0);
3551 VDetachVolume_r(ec, vp);
3555 return VAttachVolumeByName_r(ec, part, name, mode);
3558 /* Increment a reference count to a volume, sans context swaps. Requires
3559 * possibly reading the volume header in from the disk, since there's
3560 * an invariant in the volume package that nUsers>0 ==> vp->header is valid.
3562 * N.B. This call can fail if we can't read in the header!! In this case
3563 * we still guarantee we won't context swap, but the ref count won't be
3564 * incremented (otherwise we'd violate the invariant).
3566 /* NOTE: with the demand attach fileserver extensions, the global lock
3567 * is dropped within VHold */
3568 #ifdef AFS_DEMAND_ATTACH_FS
3570 VHold_r(Volume * vp)
3574 VCreateReservation_r(vp);
3575 VWaitExclusiveState_r(vp);
3577 LoadVolumeHeader(&error, vp);
3579 VCancelReservation_r(vp);
3583 VCancelReservation_r(vp);
3586 #else /* AFS_DEMAND_ATTACH_FS */
3588 VHold_r(Volume * vp)
3592 LoadVolumeHeader(&error, vp);
3598 #endif /* AFS_DEMAND_ATTACH_FS */
3600 /**** volume timeout-related stuff ****/
3602 #ifdef AFS_PTHREAD_ENV
3604 static struct timespec *shutdown_timeout;
3605 static pthread_once_t shutdown_timeout_once = PTHREAD_ONCE_INIT;
3608 VTimedOut(const struct timespec *ts)
3613 if (ts->tv_sec == 0) {
3614 /* short-circuit; this will have always timed out */
3618 code = gettimeofday(&tv, NULL);
3620 Log("Error %d from gettimeofday, assuming we have not timed out\n", errno);
3621 /* assume no timeout; failure mode is we just wait longer than normal
3622 * instead of returning errors when we shouldn't */
3626 if (tv.tv_sec < ts->tv_sec ||
3627 (tv.tv_sec == ts->tv_sec && tv.tv_usec*1000 < ts->tv_nsec)) {
3636 * Calculate an absolute timeout.
3638 * @param[out] ts A timeout that is "timeout" seconds from now, if we return
3639 * NULL, the memory is not touched
3640 * @param[in] timeout How long the timeout should be from now
3642 * @return timeout to use
3643 * @retval NULL no timeout; wait forever
3644 * @retval non-NULL the given value for "ts"
3648 static struct timespec *
3649 VCalcTimeout(struct timespec *ts, afs_int32 timeout)
3659 ts->tv_sec = ts->tv_nsec = 0;
3663 code = gettimeofday(&now, NULL);
3665 Log("Error %d from gettimeofday, falling back to 'forever' timeout\n", errno);
3669 ts->tv_sec = now.tv_sec + timeout;
3670 ts->tv_nsec = now.tv_usec * 1000;
3676 * Initialize the shutdown_timeout global.
3679 VShutdownTimeoutInit(void)
3681 struct timespec *ts;
3683 ts = malloc(sizeof(*ts));
3685 shutdown_timeout = VCalcTimeout(ts, vol_opts.offline_shutdown_timeout);
3687 if (!shutdown_timeout) {
3693 * Figure out the timeout that should be used for waiting for offline volumes.
3695 * @param[out] ats Storage space for a local timeout value if needed
3697 * @return The timeout value that should be used
3698 * @retval NULL No timeout; wait forever for offlining volumes
3699 * @retval non-NULL A pointer to the absolute time that should be used as
3700 * the deadline for waiting for offlining volumes.
3702 * @note If we return non-NULL, the pointer we return may or may not be the
3705 static const struct timespec *
3706 VOfflineTimeout(struct timespec *ats)
3708 if (vol_shutting_down) {
3709 opr_Verify(pthread_once(&shutdown_timeout_once,
3710 VShutdownTimeoutInit) == 0);
3711 return shutdown_timeout;
3713 return VCalcTimeout(ats, vol_opts.offline_timeout);
3717 #else /* AFS_PTHREAD_ENV */
3719 /* Waiting a certain amount of time for offlining volumes is not supported
3720 * for LWP due to a lack of primitives. So, we never time out */
3721 # define VTimedOut(x) (0)
3722 # define VOfflineTimeout(x) (NULL)
3724 #endif /* !AFS_PTHREAD_ENV */
3732 retVal = VHold_r(vp);
3739 VIsGoingOffline_r(struct Volume *vp)
3743 if (vp->goingOffline) {
3744 if (vp->specialStatus) {
3745 code = vp->specialStatus;
3746 } else if (V_inService(vp) == 0 || V_blessed(vp) == 0) {
3757 * Tell the caller if a volume is waiting to go offline.
3759 * @param[in] vp The volume we want to know about
3761 * @return volume status
3762 * @retval 0 volume is not waiting to go offline, go ahead and use it
3763 * @retval nonzero volume is waiting to offline, and give the returned code
3764 * as an error to anyone accessing the volume
3766 * @pre VOL_LOCK is NOT held
3767 * @pre caller holds a heavyweight reference on vp
3770 VIsGoingOffline(struct Volume *vp)
3775 code = VIsGoingOffline_r(vp);
3782 * Register an RX call with a volume.
3784 * @param[inout] ec Error code; if unset when passed in, may be set if
3785 * the volume starts going offline
3786 * @param[out] client_ec @see GetVolume
3787 * @param[in] vp Volume struct
3788 * @param[in] cbv VCallByVol struct containing the RX call to register
3790 * @pre VOL_LOCK held
3791 * @pre caller holds heavy ref on vp
3796 VRegisterCall_r(Error *ec, Error *client_ec, Volume *vp, struct VCallByVol *cbv)
3799 #ifdef AFS_DEMAND_ATTACH_FS
3801 /* just in case the volume started going offline after we got the
3802 * reference to it... otherwise, if the volume started going