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
13 Institution: The Information Technology Center, Carnegie-Mellon University
17 #include <afsconfig.h>
18 #include <afs/param.h>
22 #include <sys/types.h>
34 #ifdef AFS_PTHREAD_ENV
36 #else /* AFS_PTHREAD_ENV */
37 #include <afs/assert.h>
38 #endif /* AFS_PTHREAD_ENV */
41 #include <afs/afsint.h>
43 #include <afs/errors.h>
46 #include <afs/afssyscalls.h>
48 #include <afs/afsutil.h>
55 #include "volume_inline.h"
56 #include "partition.h"
57 #include "viceinode.h"
59 #include "volinodes.h"
60 #include "vol_prototypes.h"
64 #include <sys/lockf.h>
66 #if defined(AFS_SUN5_ENV) || defined(AFS_NT40_ENV) || defined(AFS_LINUX20_ENV)
73 #define afs_open open64
74 #else /* !O_LARGEFILE */
76 #endif /* !O_LARGEFILE */
78 #define nFILES (sizeof (stuff)/sizeof(struct stuff))
80 /* Note: the volume creation functions herein leave the destroyMe flag in the
81 volume header ON: this means that the volumes will not be attached by the
82 file server and WILL BE DESTROYED the next time a system salvage is performed */
84 #ifdef FSSYNC_BUILD_CLIENT
86 RemoveInodes(Device dev, VolumeId vid)
91 /* This relies on the fact that IDEC only needs the device and NT only
92 * needs the dev and vid to decrement volume special files.
94 IH_INIT(handle, dev, vid, -1);
95 for (i = 0; i < nFILES; i++) {
96 Inode inode = *stuff[i].inode;
98 IH_DEC(handle, inode, vid);
104 VCreateVolume(Error * ec, char *partname, VolId volumeId, VolId parentId)
105 { /* Should be the same as volumeId if there is
109 retVal = VCreateVolume_r(ec, partname, volumeId, parentId);
115 VCreateVolume_r(Error * ec, char *partname, VolId volumeId, VolId parentId)
116 { /* Should be the same as volumeId if there is
120 char headerName[VMAXPATHLEN], volumePath[VMAXPATHLEN];
122 struct DiskPartition64 *partition;
123 struct VolumeDiskHeader diskHeader;
129 # ifdef AFS_DEMAND_ATTACH_FS
131 # endif /* AFS_DEMAND_ATTACH_FS */
134 memset(&vol, 0, sizeof(vol));
136 vol.parentId = parentId;
137 vol.copyDate = time(0); /* The only date which really means when this
138 * @i(instance) of this volume was created.
139 * Creation date does not mean this */
141 /* Initialize handle for error case below. */
144 /* Verify that the parition is valid before writing to it. */
145 if (!(partition = VGetPartition_r(partname, 0))) {
146 Log("VCreateVolume: partition %s is not in service.\n", partname);
150 #if defined(NEARINODE_HINT)
151 nearInodeHash(volumeId, nearInode);
152 nearInode %= partition->f_files;
154 VGetVolumePath(ec, vol.id, &part, &name);
155 if (*ec == VNOVOL || !strcmp(partition->name, part)) {
156 /* this case is ok */
158 /* return EXDEV if it's a clone to an alternate partition
159 * otherwise assume it's a move */
160 if (vol.parentId != vol.id) {
167 # ifdef AFS_DEMAND_ATTACH_FS
168 /* volume doesn't exist yet, but we must lock it to try to prevent something
169 * else from reading it when we're e.g. half way through creating it (or
170 * something tries to create the same volume at the same time) */
171 locktype = VVolLockType(V_VOLUPD, 1);
172 rc = VLockVolumeByIdNB(volumeId, partition, locktype);
174 Log("VCreateVolume: vol %lu already locked by someone else\n",
175 afs_printable_uint32_lu(volumeId));
179 # else /* AFS_DEMAND_ATTACH_FS */
180 VLockPartition_r(partname);
181 # endif /* !AFS_DEMAND_ATTACH_FS */
183 memset(&tempHeader, 0, sizeof(tempHeader));
184 tempHeader.stamp.magic = VOLUMEHEADERMAGIC;
185 tempHeader.stamp.version = VOLUMEHEADERVERSION;
186 tempHeader.id = vol.id;
187 tempHeader.parent = vol.parentId;
188 vol.stamp.magic = VOLUMEINFOMAGIC;
189 vol.stamp.version = VOLUMEINFOVERSION;
190 vol.destroyMe = DESTROY_ME;
191 (void)afs_snprintf(headerName, sizeof headerName, VFORMAT, afs_printable_uint32_lu(vol.id));
192 (void)afs_snprintf(volumePath, sizeof volumePath, "%s/%s",
193 VPartitionPath(partition), headerName);
194 rc = stat(volumePath, &st);
195 if (rc == 0 || errno != ENOENT) {
197 Log("VCreateVolume: Header file %s already exists!\n",
201 Log("VCreateVolume: Error %d trying to stat header file %s\n",
207 device = partition->device;
209 for (i = 0; i < nFILES; i++) {
210 register struct stuff *p = &stuff[i];
215 IH_CREATE(NULL, device, VPartitionPath(partition), nearInode,
216 (p->inodeType == VI_LINKTABLE) ? vol.parentId : vol.id,
217 INODESPECIAL, p->inodeType, vol.parentId);
218 if (!(VALID_INO(*(p->inode)))) {
219 if (errno == EEXIST && (p->inodeType == VI_LINKTABLE)) {
220 /* Increment the reference count instead. */
225 *(p->inode) = nt_MakeSpecIno(VI_LINKTABLE);
227 *(p->inode) = namei_MakeSpecIno(vol.parentId, VI_LINKTABLE);
229 IH_INIT(lh, device, parentId, *(p->inode));
235 code = IH_INC(lh, *(p->inode), parentId);
236 FDH_REALLYCLOSE(fdP);
245 IH_CREATE(NULL, device, VPartitionPath(partition), nearInode,
246 vol.id, INODESPECIAL, p->inodeType, vol.parentId);
249 if (!VALID_INO(*(p->inode))) {
250 Log("VCreateVolume: Problem creating %s file associated with volume header %s\n", p->description, volumePath);
254 RemoveInodes(device, vol.id);
258 VDestroyVolumeDiskHeader(partition, volumeId, parentId);
260 # ifdef AFS_DEMAND_ATTACH_FS
262 VUnlockVolumeById(volumeId, partition);
264 # endif /* AFS_DEMAND_ATTACH_FS */
267 IH_INIT(handle, device, vol.parentId, *(p->inode));
268 fdP = IH_OPEN(handle);
270 Log("VCreateVolume: Problem iopen inode %s (err=%d)\n",
271 PrintInode(NULL, *(p->inode)), errno);
274 if (FDH_SEEK(fdP, 0, SEEK_SET) < 0) {
275 Log("VCreateVolume: Problem lseek inode %s (err=%d)\n",
276 PrintInode(NULL, *(p->inode)), errno);
277 FDH_REALLYCLOSE(fdP);
280 if (FDH_WRITE(fdP, (char *)&p->stamp, sizeof(p->stamp)) !=
282 Log("VCreateVolume: Problem writing to inode %s (err=%d)\n",
283 PrintInode(NULL, *(p->inode)), errno);
284 FDH_REALLYCLOSE(fdP);
287 FDH_REALLYCLOSE(fdP);
289 nearInode = *(p->inode);
292 IH_INIT(handle, device, vol.parentId, tempHeader.volumeInfo);
293 fdP = IH_OPEN(handle);
295 Log("VCreateVolume: Problem iopen inode %s (err=%d)\n",
296 PrintInode(NULL, tempHeader.volumeInfo), errno);
299 if (FDH_SEEK(fdP, 0, SEEK_SET) < 0) {
300 Log("VCreateVolume: Problem lseek inode %s (err=%d)\n",
301 PrintInode(NULL, tempHeader.volumeInfo), errno);
302 FDH_REALLYCLOSE(fdP);
305 if (FDH_WRITE(fdP, (char *)&vol, sizeof(vol)) != sizeof(vol)) {
306 Log("VCreateVolume: Problem writing to inode %s (err=%d)\n",
307 PrintInode(NULL, tempHeader.volumeInfo), errno);
308 FDH_REALLYCLOSE(fdP);
314 VolumeHeaderToDisk(&diskHeader, &tempHeader);
315 rc = VCreateVolumeDiskHeader(&diskHeader, partition);
317 Log("VCreateVolume: Error %d trying to write volume header for "
318 "volume %u on partition %s; volume not created\n", rc,
319 vol.id, VPartitionPath(partition));
326 # ifdef AFS_DEMAND_ATTACH_FS
328 VUnlockVolumeById(volumeId, partition);
330 # endif /* AFS_DEMAND_ATTACH_FS */
331 return (VAttachVolumeByName_r(ec, partname, headerName, V_SECRETLY));
333 #endif /* FSSYNC_BUILD_CLIENT */
337 AssignVolumeName(register VolumeDiskData * vol, char *name, char *ext)
340 AssignVolumeName_r(vol, name, ext);
345 AssignVolumeName_r(register VolumeDiskData * vol, char *name, char *ext)
348 strncpy(vol->name, name, VNAMESIZE - 1);
349 vol->name[VNAMESIZE - 1] = '\0';
350 dot = strrchr(vol->name, '.');
351 if (dot && (strcmp(dot, ".backup") == 0 || strcmp(dot, ".readonly") == 0))
354 strncat(vol->name, ext, VNAMESIZE - 1 - strlen(vol->name));
358 CopyVolumeHeader_r(VolumeDiskData * from, VolumeDiskData * to)
360 /* The id and parentId fields are not copied; these are inviolate--the to volume
361 * is assumed to have already been created. The id's cannot be changed once
362 * creation has taken place, since they are embedded in the various inodes associated
363 * with the volume. The copydate is also inviolate--it always reflects the time
364 * this volume was created (compare with the creation date--the creation date of
365 * a backup volume is the creation date of the original parent, because the backup
366 * is used to backup the parent volume). */
370 parent = to->parentId;
371 copydate = to->copyDate;
372 memcpy(to, from, sizeof(*from));
374 to->parentId = parent;
375 to->copyDate = copydate;
376 to->destroyMe = DESTROY_ME; /* Caller must always clear this!!! */
377 to->stamp.magic = VOLUMEINFOMAGIC;
378 to->stamp.version = VOLUMEINFOVERSION;
383 CopyVolumeHeader(VolumeDiskData * from, VolumeDiskData * to)
388 code = CopyVolumeHeader_r(from, to);
394 ClearVolumeStats(register VolumeDiskData * vol)
397 ClearVolumeStats_r(vol);
402 ClearVolumeStats_r(register VolumeDiskData * vol)
404 memset(vol->weekUse, 0, sizeof(vol->weekUse));
410 * read an existing volume disk header.
412 * @param[in] volid volume id
413 * @param[in] dp disk partition object
414 * @param[out] hdr volume disk header or NULL
416 * @note if hdr is NULL, this is essentially an existence test for the vol
419 * @return operation status
421 * @retval -1 volume header doesn't exist
422 * @retval EIO failed to read volume header
427 VReadVolumeDiskHeader(VolumeId volid,
428 struct DiskPartition64 * dp,
429 VolumeDiskHeader_t * hdr)
433 char path[MAXPATHLEN];
435 (void)afs_snprintf(path, sizeof(path),
437 VPartitionPath(dp), afs_printable_uint32_lu(volid));
438 fd = open(path, O_RDONLY);
440 Log("VReadVolumeDiskHeader: Couldn't open header for volume %lu (errno %d).\n",
441 afs_printable_uint32_lu(volid), errno);
444 } else if (hdr && read(fd, hdr, sizeof(*hdr)) != sizeof(*hdr)) {
445 Log("VReadVolumeDiskHeader: Couldn't read header for volume %lu.\n",
446 afs_printable_uint32_lu(volid));
456 #ifdef FSSYNC_BUILD_CLIENT
458 * write an existing volume disk header.
460 * @param[in] hdr volume disk header
461 * @param[in] dp disk partition object
462 * @param[in] cr assert if O_CREAT | O_EXCL should be passed to open()
464 * @return operation status
466 * @retval -1 volume header doesn't exist
467 * @retval EIO failed to write volume header
472 _VWriteVolumeDiskHeader(VolumeDiskHeader_t * hdr,
473 struct DiskPartition64 * dp,
478 char path[MAXPATHLEN];
480 #ifdef AFS_DEMAND_ATTACH_FS
481 /* prevent racing with VGC scanners reading the vol header while we are
483 code = VPartHeaderLock(dp, READ_LOCK);
487 #endif /* AFS_DEMAND_ATTACH_FS */
491 (void)afs_snprintf(path, sizeof(path),
493 VPartitionPath(dp), afs_printable_uint32_lu(hdr->id));
494 fd = open(path, flags, 0644);
497 Log("_VWriteVolumeDiskHeader: Couldn't open header for volume %lu, "
498 "error = %d\n", afs_printable_uint32_lu(hdr->id), errno);
499 } else if (write(fd, hdr, sizeof(*hdr)) != sizeof(*hdr)) {
500 Log("_VWriteVolumeDiskHeader: Couldn't write header for volume %lu, "
501 "error = %d\n", afs_printable_uint32_lu(hdr->id), errno);
506 if (close(fd) != 0) {
507 Log("_VWriteVolumeDiskHeader: Error closing header for volume "
508 "%lu, errno %d\n", afs_printable_uint32_lu(hdr->id), errno);
512 #ifdef AFS_DEMAND_ATTACH_FS
513 VPartHeaderUnlock(dp, READ_LOCK);
514 #endif /* AFS_DEMAND_ATTACH_FS */
520 * write an existing volume disk header.
522 * @param[in] hdr volume disk header
523 * @param[in] dp disk partition object
525 * @return operation status
527 * @retval ENOENT volume header doesn't exist
528 * @retval EIO failed to write volume header
531 VWriteVolumeDiskHeader(VolumeDiskHeader_t * hdr,
532 struct DiskPartition64 * dp)
536 #ifdef AFS_DEMAND_ATTACH_FS
537 VolumeDiskHeader_t oldhdr;
538 int delvgc = 0, addvgc = 0;
541 /* first, see if anything with the volume IDs have changed; if so, we
542 * need to update the VGC */
544 code = VReadVolumeDiskHeader(hdr->id, dp, &oldhdr);
545 if (code == 0 && (oldhdr.id != hdr->id || oldhdr.parent != hdr->parent)) {
546 /* the vol id or parent vol id changed; need to delete the VGC entry
547 * for the old vol id/parent, and add the new one */
552 /* couldn't get the old header info; add the new header info to the
553 * VGC in case it hasn't been added yet */
557 #endif /* AFS_DEMAND_ATTACH_FS */
559 code = _VWriteVolumeDiskHeader(hdr, dp, 0);
564 #ifdef AFS_DEMAND_ATTACH_FS
566 memset(&res, 0, sizeof(res));
567 code = FSYNC_VGCDel(dp->name, oldhdr.parent, oldhdr.id, FSYNC_WHATEVER, &res);
569 /* unknown vol id is okay; it just further suggests the old header
570 * data was bogus, which is fine since we're trying to fix it */
571 if (code && res.hdr.reason != FSYNC_UNKNOWN_VOLID) {
572 Log("VWriteVolumeDiskHeader: FSYNC_VGCDel(%s, %lu, %lu) "
573 "failed with code %ld reason %ld\n", dp->name,
574 afs_printable_uint32_lu(oldhdr.parent),
575 afs_printable_uint32_lu(oldhdr.id),
576 afs_printable_int32_ld(code),
577 afs_printable_int32_ld(res.hdr.reason));
582 memset(&res, 0, sizeof(res));
583 code = FSYNC_VGCAdd(dp->name, hdr->parent, hdr->id, FSYNC_WHATEVER, &res);
585 Log("VWriteVolumeDiskHeader: FSYNC_VGCAdd(%s, %lu, %lu) "
586 "failed with code %ld reason %ld\n", dp->name,
587 afs_printable_uint32_lu(hdr->parent),
588 afs_printable_uint32_lu(hdr->id),
589 afs_printable_int32_ld(code),
590 afs_printable_int32_ld(res.hdr.reason));
594 #endif /* AFS_DEMAND_ATTACH_FS */
601 * create and write a volume disk header to disk.
603 * @param[in] hdr volume disk header
604 * @param[in] dp disk partition object
606 * @return operation status
608 * @retval EEXIST volume header already exists
609 * @retval EIO failed to write volume header
614 VCreateVolumeDiskHeader(VolumeDiskHeader_t * hdr,
615 struct DiskPartition64 * dp)
618 #ifdef AFS_DEMAND_ATTACH_FS
620 #endif /* AFS_DEMAND_ATTACH_FS */
622 code = _VWriteVolumeDiskHeader(hdr, dp, O_CREAT | O_EXCL);
627 #ifdef AFS_DEMAND_ATTACH_FS
628 memset(&res, 0, sizeof(res));
629 code = FSYNC_VGCAdd(dp->name, hdr->parent, hdr->id, FSYNC_WHATEVER, &res);
631 Log("VCreateVolumeDiskHeader: FSYNC_VGCAdd(%s, %lu, %lu) failed "
632 "with code %ld reason %ld\n", dp->name,
633 afs_printable_uint32_lu(hdr->parent),
634 afs_printable_uint32_lu(hdr->id),
635 afs_printable_int32_ld(code),
636 afs_printable_int32_ld(res.hdr.reason));
638 #endif /* AFS_DEMAND_ATTACH_FS */
646 * destroy a volume disk header.
648 * @param[in] dp disk partition object
649 * @param[in] volid volume id
650 * @param[in] parent parent's volume id, 0 if unknown
652 * @return operation status
655 * @note if parent is 0, the parent volume ID will be looked up from the
658 * @note for non-DAFS, parent is currently ignored
661 VDestroyVolumeDiskHeader(struct DiskPartition64 * dp,
666 char path[MAXPATHLEN];
667 #ifdef AFS_DEMAND_ATTACH_FS
669 #endif /* AFS_DEMAND_ATTACH_FS */
671 (void)afs_snprintf(path, sizeof(path),
673 VPartitionPath(dp), afs_printable_uint32_lu(volid));
676 Log("VDestroyVolumeDiskHeader: Couldn't unlink disk header, error = %d\n", errno);
680 #ifdef AFS_DEMAND_ATTACH_FS
681 memset(&res, 0, sizeof(res));
683 FSSYNC_VGQry_response_t q_res;
685 code = FSYNC_VGCQuery(dp->name, volid, &q_res, &res);
687 Log("VDestroyVolumeDiskHeader: FSYNC_VGCQuery(%s, %lu) failed "
688 "with code %ld, reason %ld\n", dp->name,
689 afs_printable_uint32_lu(volid), afs_printable_int32_ld(code),
690 afs_printable_int32_ld(res.hdr.reason));
697 code = FSYNC_VGCDel(dp->name, parent, volid, FSYNC_WHATEVER, &res);
699 Log("VDestroyVolumeDiskHeader: FSYNC_VGCDel(%s, %lu, %lu) failed "
700 "with code %ld reason %ld\n", dp->name,
701 afs_printable_uint32_lu(parent),
702 afs_printable_uint32_lu(volid),
703 afs_printable_int32_ld(code),
704 afs_printable_int32_ld(res.hdr.reason));
706 #endif /* AFS_DEMAND_ATTACH_FS */
711 #endif /* FSSYNC_BUILD_CLIENT */
714 * handle a single vol header as part of VWalkVolumeHeaders.
716 * @param[in] dp disk partition
717 * @param[in] volfunc function to call when a vol header is successfully read
718 * @param[in] name full path name to the .vol header
719 * @param[out] hdr header data read in from the .vol header
720 * @param[in] locked 1 if the partition headers are locked, 0 otherwise
721 * @param[in] rock the rock to pass to volfunc
723 * @return operation status
725 * @retval -1 fatal error, stop scanning
726 * @retval 1 failed to read header
727 * @retval 2 volfunc callback indicated error after header read
730 _VHandleVolumeHeader(struct DiskPartition64 *dp, VWalkVolFunc volfunc,
731 const char *name, struct VolumeDiskHeader *hdr,
732 int locked, void *rock)
737 if ((fd = afs_open(name, O_RDONLY)) == -1
738 || read(fd, hdr, sizeof(*hdr))
740 || hdr->stamp.magic != VOLUMEHEADERMAGIC) {
748 #ifdef AFSFS_DEMAND_ATTACH_FS
750 VPartHeaderUnlock(dp);
752 #endif /* AFS_DEMAND_ATTACH_FS */
754 if (!error && volfunc) {
755 /* the volume header seems fine; call the caller-supplied
756 * 'we-found-a-volume-header' function */
759 #ifdef AFS_DEMAND_ATTACH_FS
763 #endif /* AFS_DEMAND_ATTACH_FS */
765 error = (*volfunc) (dp, name, hdr, last, rock);
774 #ifdef AFS_DEMAND_ATTACH_FS
775 if (error && !locked) {
777 /* retry reading the volume header under the partition
778 * header lock, just to be safe and ensure we're not
779 * racing something rewriting the vol header */
780 code = VPartHeaderLock(dp, WRITE_LOCK);
782 Log("Error acquiring partition write lock when "
783 "looking at header %s\n", name);
787 return _VHandleVolumeHeader(dp, volfunc, name, hdr, 1, rock);
789 #endif /* AFS_DEMAND_ATTACH_FS */
795 * walk through the list of volume headers on a partition.
797 * This function looks through all of the .vol headers on a partition, reads in
798 * each header, and calls the supplied volfunc function on each one. If the
799 * header cannot be read (or volfunc returns a positive error code), DAFS will
800 * VPartHeaderExLock() and retry. If that fails, or if we are non-DAFS, errfunc
801 * will be called (which typically will unlink the problem volume header).
803 * If volfunc returns a negative error code, walking the partition will stop
804 * and we will return an error immediately.
806 * @param[in] dp partition to walk
807 * @param[in] partpath the path opendir()
808 * @param[in] volfunc the function to call when a header is encountered, or
809 * NULL to just skip over valid headers
810 * @param[in] errfunc the function to call when a problematic header is
811 * encountered, or NULL to just skip over bad headers
812 * @param[in] rock rock for volfunc and errfunc
817 * @return operation status
819 * @retval negative fatal error, walk did not finish
822 VWalkVolumeHeaders(struct DiskPartition64 *dp, const char *partpath,
823 VWalkVolFunc volfunc, VWalkErrFunc errfunc, void *rock)
826 struct dirent *dentry = NULL;
828 struct VolumeDiskHeader diskHeader;
830 dirp = opendir(partpath);
832 Log("VWalkVolumeHeaders: cannot open directory %s\n", partpath);
837 while ((dentry = readdir(dirp))) {
838 char *p = dentry->d_name;
839 p = strrchr(dentry->d_name, '.');
840 if (p != NULL && strcmp(p, VHDREXT) == 0) {
841 char name[VMAXPATHLEN];
843 sprintf(name, "%s/%s", partpath, dentry->d_name);
845 code = _VHandleVolumeHeader(dp, volfunc, name, &diskHeader, -1, rock);
847 /* fatal error, stop walking */
850 if (code && errfunc) {
851 /* error with header; call the caller-supplied vol error
854 struct VolumeDiskHeader *hdr = &diskHeader;
856 /* we failed to read the header at all, so don't pass in
860 (*errfunc) (dp, name, hdr, rock);
874 #ifdef AFS_PTHREAD_ENV
875 # define AFS_LF_LOCK(lf) assert(pthread_mutex_lock(&((lf)->mutex)) == 0)
876 # define AFS_LF_UNLOCK(lf) assert(pthread_mutex_unlock(&((lf)->mutex)) == 0)
878 # define AFS_LF_LOCK(lf)
879 # define AFS_LF_UNLOCK(lf)
880 #endif /* AFS_PTHREAD_ENV */
883 * initialize a struct VLockFile.
885 * @param[in] lf struct VLockFile to initialize
886 * @param[in] path Full path to the file to use for locks. The string contents
890 VLockFileInit(struct VLockFile *lf, const char *path)
892 memset(lf, 0, sizeof(*lf));
893 lf->path = strdup(path);
895 #ifdef AFS_PTHREAD_ENV
896 assert(pthread_mutex_init(&lf->mutex, NULL) == 0);
897 #endif /* AFS_PTHREAD_ENV */
902 _VOpenPath(const char *path)
906 handle = CreateFile(path,
907 GENERIC_READ | GENERIC_WRITE,
908 FILE_SHARE_READ | FILE_SHARE_WRITE,
911 FILE_ATTRIBUTE_HIDDEN,
913 if (handle == INVALID_HANDLE_VALUE) {
921 _VLockFd(FD_t handle, afs_uint32 offset, int locktype, int nonblock)
926 if (locktype == WRITE_LOCK) {
927 flags |= LOCKFILE_EXCLUSIVE_LOCK;
930 flags |= LOCKFILE_FAIL_IMMEDIATELY;
933 memset(&lap, 0, sizeof(lap));
936 if (!LockFileEx(handle, flags, 0, 1, 0, &lap)) {
937 if (GetLastError() == ERROR_LOCK_VIOLATION) {
947 _VUnlockFd(struct VLockFile *lf, afs_uint32 offset)
951 memset(&lap, 0, sizeof(lap));
954 UnlockFileEx(lf->fd, 0, 1, 0, &lap);
958 _VCloseFd(struct VLockFile *lf)
963 #else /* !AFS_NT40_ENV */
966 * open a file on the local filesystem suitable for locking
968 * @param[in] path abs path of the file to open
970 * @return file descriptor
971 * @retval INVALID_FD failure opening file
974 _VOpenPath(const char *path)
978 fd = open(path, O_RDWR | O_CREAT, 0660);
986 * lock an offset in a file descriptor.
988 * @param[in] fd file descriptor to lock
989 * @param[in] offset offset in file to lock
990 * @param[in] locktype READ_LOCK or WRITE_LOCK
991 * @param[in] nonblock 1 to fail immediately, 0 to wait to acquire lock
993 * @return operation status
995 * @retval EBUSY someone else is holding a conflicting lock and nonblock=1 was
997 * @retval EIO error acquiring file lock
1000 _VLockFd(int fd, afs_uint32 offset, int locktype, int nonblock)
1002 int l_type = F_WRLCK;
1006 if (locktype == READ_LOCK) {
1013 sf.l_start = offset;
1016 sf.l_whence = SEEK_SET;
1018 if (fcntl(fd, cmd, &sf)) {
1019 if (nonblock && (errno == EACCES || errno == EAGAIN)) {
1020 /* We asked for a nonblocking lock, and it was already locked */
1022 if (fcntl(fd, F_GETLK, &sf) != 0 || sf.l_pid == 0) {
1023 Log("_VLockFd: fcntl failed with error %d when trying to "
1024 "query the conflicting lock for fd %d (locktype=%d, "
1025 "offset=%lu)\n", errno, fd, locktype,
1026 afs_printable_uint32_lu(offset));
1028 Log("_VLockFd: conflicting lock held on fd %d, offset %lu by "
1029 "pid %ld (locktype=%d)\n", fd,
1030 afs_printable_uint32_lu(offset), (long int)sf.l_pid,
1035 Log("_VLockFd: fcntl failed with error %d when trying to lock "
1036 "fd %d (locktype=%d, offset=%lu)\n", errno, fd, locktype,
1037 afs_printable_uint32_lu(offset));
1045 * close a file descriptor used for file locking.
1047 * @param[in] fd file descriptor to close
1053 Log("_VCloseFd: error %d closing fd %d\n",
1059 * unlock a file offset in a file descriptor.
1061 * @param[in] fd file descriptor to unlock
1062 * @param[in] offset offset to unlock
1065 _VUnlockFd(int fd, afs_uint32 offset)
1069 sf.l_start = offset;
1071 sf.l_type = F_UNLCK;
1072 sf.l_whence = SEEK_SET;
1074 if (fcntl(fd, F_SETLK, &sf)) {
1075 Log("_VUnlockFd: fcntl failed with error %d when trying to unlock "
1076 "fd %d\n", errno, fd);
1079 #endif /* !AFS_NT40_ENV */
1082 * reinitialize a struct VLockFile.
1084 * Use this to close the lock file (unlocking any locks in it), and effectively
1085 * restore lf to the state it was in when it was initialized. This is the same
1086 * as unlocking all of the locks on the file, without having to remember what
1087 * all of the locks were. Do not unlock previously held locks after calling
1090 * @param[in] lf struct VLockFile to reinit
1092 * @pre nobody is waiting for a lock on this lockfile or otherwise using
1093 * this lockfile at all
1096 VLockFileReinit(struct VLockFile *lf)
1098 #ifdef AFS_PTHREAD_ENV
1099 assert(pthread_mutex_lock(&lf->mutex) == 0);
1100 #endif /* AFS_PTHREAD_ENV */
1102 if (lf->fd != INVALID_FD) {
1104 lf->fd = INVALID_FD;
1109 #ifdef AFS_PTHREAD_ENV
1110 assert(pthread_mutex_unlock(&lf->mutex) == 0);
1111 #endif /* AFS_PTHREAD_ENV */
1115 * lock a file on disk for the process.
1117 * @param[in] lf the struct VLockFile representing the file to lock
1118 * @param[in] offset the offset in the file to lock
1119 * @param[in] locktype READ_LOCK or WRITE_LOCK
1120 * @param[in] nonblock 0 to wait for conflicting locks to clear before
1121 * obtaining the lock; 1 to fail immediately if a
1122 * conflicting lock is held by someone else
1124 * @return operation status
1126 * @retval EBUSY someone else is holding a conflicting lock and nonblock=1 was
1128 * @retval EIO error acquiring file lock
1132 * @note do not try to lock/unlock the same offset in the same file from
1133 * different threads; use VGetDiskLock to protect threads from each other in
1134 * addition to other processes
1137 VLockFileLock(struct VLockFile *lf, afs_uint32 offset, int locktype, int nonblock)
1141 assert(locktype == READ_LOCK || locktype == WRITE_LOCK);
1145 if (lf->fd == INVALID_FD) {
1146 lf->fd = _VOpenPath(lf->path);
1147 if (lf->fd == INVALID_FD) {
1157 code = _VLockFd(lf->fd, offset, locktype, nonblock);
1161 if (--lf->refcount < 1) {
1163 lf->fd = INVALID_FD;
1172 VLockFileUnlock(struct VLockFile *lf, afs_uint32 offset)
1176 assert(lf->fd != INVALID_FD);
1178 if (--lf->refcount < 1) {
1180 lf->fd = INVALID_FD;
1182 _VUnlockFd(lf->fd, offset);
1188 #ifdef AFS_DEMAND_ATTACH_FS
1191 * initialize a struct VDiskLock.
1193 * @param[in] dl struct VDiskLock to initialize
1194 * @param[in] lf the struct VLockFile to associate with this disk lock
1197 VDiskLockInit(struct VDiskLock *dl, struct VLockFile *lf, afs_uint32 offset)
1200 memset(dl, 0, sizeof(*dl));
1201 Lock_Init(&dl->rwlock);
1202 assert(pthread_mutex_init(&dl->mutex, NULL) == 0);
1203 assert(pthread_cond_init(&dl->cv, NULL) == 0);
1205 dl->offset = offset;
1209 * acquire a lock on a file on local disk.
1211 * @param[in] dl the VDiskLock structure corresponding to the file on disk
1212 * @param[in] locktype READ_LOCK if you want a read lock, or WRITE_LOCK if
1213 * you want a write lock
1214 * @param[in] nonblock 0 to wait for conflicting locks to clear before
1215 * obtaining the lock; 1 to fail immediately if a
1216 * conflicting lock is held by someone else
1218 * @return operation status
1220 * @retval EBUSY someone else is holding a conflicting lock and nonblock=1 was
1222 * @retval EIO error acquiring file lock
1226 * @note while normal fcntl-y locks on Unix systems generally only work per-
1227 * process, this interface also deals with locks between threads in the
1228 * process in addition to different processes acquiring the lock
1231 VGetDiskLock(struct VDiskLock *dl, int locktype, int nonblock)
1234 assert(locktype == READ_LOCK || locktype == WRITE_LOCK);
1237 if (locktype == READ_LOCK) {
1238 ObtainReadLockNoBlock(&dl->rwlock, code);
1240 ObtainWriteLockNoBlock(&dl->rwlock, code);
1247 } else if (locktype == READ_LOCK) {
1248 ObtainReadLock(&dl->rwlock);
1250 ObtainWriteLock(&dl->rwlock);
1253 assert(pthread_mutex_lock(&dl->mutex) == 0);
1255 if ((dl->flags & VDISKLOCK_ACQUIRING)) {
1256 /* Some other thread is waiting to acquire an fs lock. If nonblock=1,
1257 * we can return immediately, since we know we'll need to wait to
1258 * acquire. Otherwise, wait for the other thread to finish acquiring
1263 while ((dl->flags & VDISKLOCK_ACQUIRING)) {
1264 assert(pthread_cond_wait(&dl->cv, &dl->mutex) == 0);
1269 if (code == 0 && !(dl->flags & VDISKLOCK_ACQUIRED)) {
1270 /* no other thread holds the lock on the actual file; so grab one */
1272 /* first try, don't block on the lock to see if we can get it without
1274 code = VLockFileLock(dl->lockfile, dl->offset, locktype, 1);
1276 if (code == EBUSY && !nonblock) {
1278 /* mark that we are waiting on the fs lock */
1279 dl->flags |= VDISKLOCK_ACQUIRING;
1281 assert(pthread_mutex_unlock(&dl->mutex) == 0);
1282 code = VLockFileLock(dl->lockfile, dl->offset, locktype, nonblock);
1283 assert(pthread_mutex_lock(&dl->mutex) == 0);
1285 dl->flags &= ~VDISKLOCK_ACQUIRING;
1288 dl->flags |= VDISKLOCK_ACQUIRED;
1291 assert(pthread_cond_broadcast(&dl->cv) == 0);
1296 if (locktype == READ_LOCK) {
1297 ReleaseReadLock(&dl->rwlock);
1299 ReleaseWriteLock(&dl->rwlock);
1302 /* successfully got the lock, so inc the number of unlocks we need
1303 * to do before we can unlock the actual file */
1307 assert(pthread_mutex_unlock(&dl->mutex) == 0);
1313 * release a lock on a file on local disk.
1315 * @param[in] dl the struct VDiskLock to release
1316 * @param[in] locktype READ_LOCK if you are unlocking a read lock, or
1317 * WRITE_LOCK if you are unlocking a write lock
1319 * @return operation status
1323 VReleaseDiskLock(struct VDiskLock *dl, int locktype)
1325 assert(locktype == READ_LOCK || locktype == WRITE_LOCK);
1327 assert(pthread_mutex_lock(&dl->mutex) == 0);
1328 assert(dl->lockers > 0);
1330 if (--dl->lockers < 1) {
1331 /* no threads are holding this lock anymore, so we can release the
1332 * actual disk lock */
1333 VLockFileUnlock(dl->lockfile, dl->offset);
1334 dl->flags &= ~VDISKLOCK_ACQUIRED;
1337 assert(pthread_mutex_unlock(&dl->mutex) == 0);
1339 if (locktype == READ_LOCK) {
1340 ReleaseReadLock(&dl->rwlock);
1342 ReleaseWriteLock(&dl->rwlock);
1346 #endif /* AFS_DEMAND_ATTACH_FS */