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>
23 #include <sys/types.h>
35 #include <afs/afs_assert.h>
38 #include <afs/afsint.h>
40 #include <afs/errors.h>
43 #include <afs/afssyscalls.h>
45 #include <afs/afsutil.h>
52 #include "volume_inline.h"
53 #include "partition.h"
54 #include "viceinode.h"
56 #include "volinodes.h"
57 #include "vol_prototypes.h"
61 #include <sys/lockf.h>
63 #if defined(AFS_SUN5_ENV) || defined(AFS_NT40_ENV) || defined(AFS_LINUX20_ENV)
69 /* Note: the volume creation functions herein leave the destroyMe flag in the
70 volume header ON: this means that the volumes will not be attached by the
71 file server and WILL BE DESTROYED the next time a system salvage is performed */
73 #ifdef FSSYNC_BUILD_CLIENT
75 RemoveInodes(struct afs_inode_info *stuff, Device dev, VolumeId vid)
80 /* This relies on the fact that IDEC only needs the device and NT only
81 * needs the dev and vid to decrement volume special files.
83 IH_INIT(handle, dev, vid, -1);
84 for (i = 0; i < MAXINODETYPE; i++) {
85 Inode inode = *stuff[i].inode;
87 IH_DEC(handle, inode, vid);
93 VCreateVolume(Error * ec, char *partname, VolId volumeId, VolId parentId)
94 { /* Should be the same as volumeId if there is
98 retVal = VCreateVolume_r(ec, partname, volumeId, parentId);
104 VCreateVolume_r(Error * ec, char *partname, VolId volumeId, VolId parentId)
105 { /* Should be the same as volumeId if there is
109 char headerName[VMAXPATHLEN], volumePath[VMAXPATHLEN];
111 struct DiskPartition64 *partition;
112 struct VolumeDiskHeader diskHeader;
118 struct VolumeHeader tempHeader;
119 struct afs_inode_info stuff[MAXINODETYPE];
121 # ifdef AFS_DEMAND_ATTACH_FS
123 # endif /* AFS_DEMAND_ATTACH_FS */
125 init_inode_info(&tempHeader, stuff);
128 memset(&vol, 0, sizeof(vol));
130 vol.parentId = parentId;
131 vol.copyDate = time(0); /* The only date which really means when this
132 * @i(instance) of this volume was created.
133 * Creation date does not mean this */
135 /* Initialize handle for error case below. */
138 /* Verify that the parition is valid before writing to it. */
139 if (!(partition = VGetPartition_r(partname, 0))) {
140 Log("VCreateVolume: partition %s is not in service.\n", partname);
144 #if defined(NEARINODE_HINT)
145 nearInodeHash(volumeId, nearInode);
146 nearInode %= partition->f_files;
148 VGetVolumePath(ec, vol.id, &part, &name);
149 if (*ec == VNOVOL || !strcmp(partition->name, part)) {
150 /* this case is ok */
152 /* return EXDEV if it's a clone to an alternate partition
153 * otherwise assume it's a move */
154 if (vol.parentId != vol.id) {
161 # ifdef AFS_DEMAND_ATTACH_FS
162 /* volume doesn't exist yet, but we must lock it to try to prevent something
163 * else from reading it when we're e.g. half way through creating it (or
164 * something tries to create the same volume at the same time) */
165 locktype = VVolLockType(V_VOLUPD, 1);
166 rc = VLockVolumeByIdNB(volumeId, partition, locktype);
168 Log("VCreateVolume: vol %lu already locked by someone else\n",
169 afs_printable_uint32_lu(volumeId));
173 # else /* AFS_DEMAND_ATTACH_FS */
174 VLockPartition_r(partname);
175 # endif /* !AFS_DEMAND_ATTACH_FS */
177 memset(&tempHeader, 0, sizeof(tempHeader));
178 tempHeader.stamp.magic = VOLUMEHEADERMAGIC;
179 tempHeader.stamp.version = VOLUMEHEADERVERSION;
180 tempHeader.id = vol.id;
181 tempHeader.parent = vol.parentId;
182 vol.stamp.magic = VOLUMEINFOMAGIC;
183 vol.stamp.version = VOLUMEINFOVERSION;
184 vol.destroyMe = DESTROY_ME;
185 (void)afs_snprintf(headerName, sizeof headerName, VFORMAT, afs_printable_uint32_lu(vol.id));
186 (void)afs_snprintf(volumePath, sizeof volumePath, "%s" OS_DIRSEP "%s",
187 VPartitionPath(partition), headerName);
188 rc = stat(volumePath, &st);
189 if (rc == 0 || errno != ENOENT) {
191 Log("VCreateVolume: Header file %s already exists!\n",
195 Log("VCreateVolume: Error %d trying to stat header file %s\n",
201 device = partition->device;
203 for (i = 0; i < MAXINODETYPE; i++) {
204 struct afs_inode_info *p = &stuff[i];
209 IH_CREATE(NULL, device, VPartitionPath(partition), nearInode,
210 (p->inodeType == VI_LINKTABLE) ? vol.parentId : vol.id,
211 INODESPECIAL, p->inodeType, vol.parentId);
212 if (!(VALID_INO(*(p->inode)))) {
213 if (errno == EEXIST && (p->inodeType == VI_LINKTABLE)) {
214 /* Increment the reference count instead. */
218 *(p->inode) = namei_MakeSpecIno(vol.parentId, VI_LINKTABLE);
219 IH_INIT(lh, device, parentId, *(p->inode));
225 code = IH_INC(lh, *(p->inode), parentId);
226 FDH_REALLYCLOSE(fdP);
235 IH_CREATE(NULL, device, VPartitionPath(partition), nearInode,
236 vol.id, INODESPECIAL, p->inodeType, vol.parentId);
239 if (!VALID_INO(*(p->inode))) {
240 Log("VCreateVolume: Problem creating %s file associated with volume header %s\n", p->description, volumePath);
244 RemoveInodes(stuff, device, vol.id);
248 VDestroyVolumeDiskHeader(partition, volumeId, parentId);
250 # ifdef AFS_DEMAND_ATTACH_FS
252 VUnlockVolumeById(volumeId, partition);
254 # endif /* AFS_DEMAND_ATTACH_FS */
257 IH_INIT(handle, device, vol.parentId, *(p->inode));
258 fdP = IH_OPEN(handle);
260 Log("VCreateVolume: Problem iopen inode %s (err=%d)\n",
261 PrintInode(stmp, *(p->inode)), errno);
264 if (FDH_PWRITE(fdP, (char *)&p->stamp, sizeof(p->stamp), 0) !=
266 Log("VCreateVolume: Problem writing to inode %s (err=%d)\n",
267 PrintInode(stmp, *(p->inode)), errno);
268 FDH_REALLYCLOSE(fdP);
271 FDH_REALLYCLOSE(fdP);
273 nearInode = *(p->inode);
276 IH_INIT(handle, device, vol.parentId, tempHeader.volumeInfo);
277 fdP = IH_OPEN(handle);
279 Log("VCreateVolume: Problem iopen inode %s (err=%d)\n",
280 PrintInode(stmp, tempHeader.volumeInfo), errno);
283 if (FDH_PWRITE(fdP, (char *)&vol, sizeof(vol), 0) != sizeof(vol)) {
284 Log("VCreateVolume: Problem writing to inode %s (err=%d)\n",
285 PrintInode(stmp, tempHeader.volumeInfo), errno);
286 FDH_REALLYCLOSE(fdP);
292 VolumeHeaderToDisk(&diskHeader, &tempHeader);
293 rc = VCreateVolumeDiskHeader(&diskHeader, partition);
295 Log("VCreateVolume: Error %d trying to write volume header for "
296 "volume %u on partition %s; volume not created\n", rc,
297 vol.id, VPartitionPath(partition));
304 # ifdef AFS_DEMAND_ATTACH_FS
306 VUnlockVolumeById(volumeId, partition);
308 # endif /* AFS_DEMAND_ATTACH_FS */
309 return (VAttachVolumeByName_r(ec, partname, headerName, V_SECRETLY));
311 #endif /* FSSYNC_BUILD_CLIENT */
315 AssignVolumeName(VolumeDiskData * vol, char *name, char *ext)
318 AssignVolumeName_r(vol, name, ext);
323 AssignVolumeName_r(VolumeDiskData * vol, char *name, char *ext)
326 strncpy(vol->name, name, VNAMESIZE - 1);
327 vol->name[VNAMESIZE - 1] = '\0';
328 dot = strrchr(vol->name, '.');
329 if (dot && (strcmp(dot, ".backup") == 0 || strcmp(dot, ".readonly") == 0))
332 strncat(vol->name, ext, VNAMESIZE - 1 - strlen(vol->name));
336 CopyVolumeHeader_r(VolumeDiskData * from, VolumeDiskData * to)
338 /* The id and parentId fields are not copied; these are inviolate--the to volume
339 * is assumed to have already been created. The id's cannot be changed once
340 * creation has taken place, since they are embedded in the various inodes associated
341 * with the volume. The copydate is also inviolate--it always reflects the time
342 * this volume was created (compare with the creation date--the creation date of
343 * a backup volume is the creation date of the original parent, because the backup
344 * is used to backup the parent volume). */
348 parent = to->parentId;
349 copydate = to->copyDate;
350 memcpy(to, from, sizeof(*from));
352 to->parentId = parent;
353 to->copyDate = copydate;
354 to->destroyMe = DESTROY_ME; /* Caller must always clear this!!! */
355 to->stamp.magic = VOLUMEINFOMAGIC;
356 to->stamp.version = VOLUMEINFOVERSION;
361 CopyVolumeHeader(VolumeDiskData * from, VolumeDiskData * to)
366 code = CopyVolumeHeader_r(from, to);
372 ClearVolumeStats(VolumeDiskData * vol)
375 ClearVolumeStats_r(vol);
380 ClearVolumeStats_r(VolumeDiskData * vol)
382 memset(vol->weekUse, 0, sizeof(vol->weekUse));
388 * read an existing volume disk header.
390 * @param[in] volid volume id
391 * @param[in] dp disk partition object
392 * @param[out] hdr volume disk header or NULL
394 * @note if hdr is NULL, this is essentially an existence test for the vol
397 * @return operation status
399 * @retval -1 volume header doesn't exist
400 * @retval EIO failed to read volume header
405 VReadVolumeDiskHeader(VolumeId volid,
406 struct DiskPartition64 * dp,
407 VolumeDiskHeader_t * hdr)
411 char path[MAXPATHLEN];
413 (void)afs_snprintf(path, sizeof(path),
414 "%s" OS_DIRSEP VFORMAT,
415 VPartitionPath(dp), afs_printable_uint32_lu(volid));
416 fd = open(path, O_RDONLY);
418 Log("VReadVolumeDiskHeader: Couldn't open header for volume %lu (errno %d).\n",
419 afs_printable_uint32_lu(volid), errno);
422 } else if (hdr && read(fd, hdr, sizeof(*hdr)) != sizeof(*hdr)) {
423 Log("VReadVolumeDiskHeader: Couldn't read header for volume %lu.\n",
424 afs_printable_uint32_lu(volid));
434 #ifdef FSSYNC_BUILD_CLIENT
436 * write an existing volume disk header.
438 * @param[in] hdr volume disk header
439 * @param[in] dp disk partition object
440 * @param[in] cr assert if O_CREAT | O_EXCL should be passed to open()
442 * @return operation status
444 * @retval -1 volume header doesn't exist
445 * @retval EIO failed to write volume header
450 _VWriteVolumeDiskHeader(VolumeDiskHeader_t * hdr,
451 struct DiskPartition64 * dp,
456 char path[MAXPATHLEN];
458 #ifdef AFS_DEMAND_ATTACH_FS
459 /* prevent racing with VGC scanners reading the vol header while we are
461 code = VPartHeaderLock(dp, READ_LOCK);
465 #endif /* AFS_DEMAND_ATTACH_FS */
469 (void)afs_snprintf(path, sizeof(path),
470 "%s" OS_DIRSEP VFORMAT,
471 VPartitionPath(dp), afs_printable_uint32_lu(hdr->id));
472 fd = open(path, flags, 0644);
475 Log("_VWriteVolumeDiskHeader: Couldn't open header for volume %lu, "
476 "error = %d\n", afs_printable_uint32_lu(hdr->id), errno);
477 } else if (write(fd, hdr, sizeof(*hdr)) != sizeof(*hdr)) {
478 Log("_VWriteVolumeDiskHeader: Couldn't write header for volume %lu, "
479 "error = %d\n", afs_printable_uint32_lu(hdr->id), errno);
484 if (close(fd) != 0) {
485 Log("_VWriteVolumeDiskHeader: Error closing header for volume "
486 "%lu, errno %d\n", afs_printable_uint32_lu(hdr->id), errno);
490 #ifdef AFS_DEMAND_ATTACH_FS
491 VPartHeaderUnlock(dp, READ_LOCK);
492 #endif /* AFS_DEMAND_ATTACH_FS */
498 * write an existing volume disk header.
500 * @param[in] hdr volume disk header
501 * @param[in] dp disk partition object
503 * @return operation status
505 * @retval ENOENT volume header doesn't exist
506 * @retval EIO failed to write volume header
509 VWriteVolumeDiskHeader(VolumeDiskHeader_t * hdr,
510 struct DiskPartition64 * dp)
514 #ifdef AFS_DEMAND_ATTACH_FS
515 VolumeDiskHeader_t oldhdr;
516 int delvgc = 0, addvgc = 0;
519 /* first, see if anything with the volume IDs have changed; if so, we
520 * need to update the VGC */
522 code = VReadVolumeDiskHeader(hdr->id, dp, &oldhdr);
523 if (code == 0 && (oldhdr.id != hdr->id || oldhdr.parent != hdr->parent)) {
524 /* the vol id or parent vol id changed; need to delete the VGC entry
525 * for the old vol id/parent, and add the new one */
530 /* couldn't get the old header info; add the new header info to the
531 * VGC in case it hasn't been added yet */
535 #endif /* AFS_DEMAND_ATTACH_FS */
537 code = _VWriteVolumeDiskHeader(hdr, dp, 0);
542 #ifdef AFS_DEMAND_ATTACH_FS
544 memset(&res, 0, sizeof(res));
545 code = FSYNC_VGCDel(dp->name, oldhdr.parent, oldhdr.id, FSYNC_WHATEVER, &res);
547 /* unknown vol id is okay; it just further suggests the old header
548 * data was bogus, which is fine since we're trying to fix it */
549 if (code && res.hdr.reason != FSYNC_UNKNOWN_VOLID) {
550 Log("VWriteVolumeDiskHeader: FSYNC_VGCDel(%s, %lu, %lu) "
551 "failed with code %ld reason %ld\n", dp->name,
552 afs_printable_uint32_lu(oldhdr.parent),
553 afs_printable_uint32_lu(oldhdr.id),
554 afs_printable_int32_ld(code),
555 afs_printable_int32_ld(res.hdr.reason));
560 memset(&res, 0, sizeof(res));
561 code = FSYNC_VGCAdd(dp->name, hdr->parent, hdr->id, FSYNC_WHATEVER, &res);
563 Log("VWriteVolumeDiskHeader: FSYNC_VGCAdd(%s, %lu, %lu) "
564 "failed with code %ld reason %ld\n", dp->name,
565 afs_printable_uint32_lu(hdr->parent),
566 afs_printable_uint32_lu(hdr->id),
567 afs_printable_int32_ld(code),
568 afs_printable_int32_ld(res.hdr.reason));
572 #endif /* AFS_DEMAND_ATTACH_FS */
579 * create and write a volume disk header to disk.
581 * @param[in] hdr volume disk header
582 * @param[in] dp disk partition object
584 * @return operation status
586 * @retval EEXIST volume header already exists
587 * @retval EIO failed to write volume header
592 VCreateVolumeDiskHeader(VolumeDiskHeader_t * hdr,
593 struct DiskPartition64 * dp)
596 #ifdef AFS_DEMAND_ATTACH_FS
598 #endif /* AFS_DEMAND_ATTACH_FS */
600 code = _VWriteVolumeDiskHeader(hdr, dp, O_CREAT | O_EXCL);
605 #ifdef AFS_DEMAND_ATTACH_FS
606 memset(&res, 0, sizeof(res));
607 code = FSYNC_VGCAdd(dp->name, hdr->parent, hdr->id, FSYNC_WHATEVER, &res);
609 Log("VCreateVolumeDiskHeader: FSYNC_VGCAdd(%s, %lu, %lu) failed "
610 "with code %ld reason %ld\n", dp->name,
611 afs_printable_uint32_lu(hdr->parent),
612 afs_printable_uint32_lu(hdr->id),
613 afs_printable_int32_ld(code),
614 afs_printable_int32_ld(res.hdr.reason));
616 #endif /* AFS_DEMAND_ATTACH_FS */
624 * destroy a volume disk header.
626 * @param[in] dp disk partition object
627 * @param[in] volid volume id
628 * @param[in] parent parent's volume id, 0 if unknown
630 * @return operation status
633 * @note if parent is 0, the parent volume ID will be looked up from the
636 * @note for non-DAFS, parent is currently ignored
639 VDestroyVolumeDiskHeader(struct DiskPartition64 * dp,
644 char path[MAXPATHLEN];
645 #ifdef AFS_DEMAND_ATTACH_FS
647 #endif /* AFS_DEMAND_ATTACH_FS */
649 (void)afs_snprintf(path, sizeof(path),
650 "%s" OS_DIRSEP VFORMAT,
651 VPartitionPath(dp), afs_printable_uint32_lu(volid));
654 Log("VDestroyVolumeDiskHeader: Couldn't unlink disk header, error = %d\n", errno);
658 #ifdef AFS_DEMAND_ATTACH_FS
659 memset(&res, 0, sizeof(res));
661 FSSYNC_VGQry_response_t q_res;
663 code = FSYNC_VGCQuery(dp->name, volid, &q_res, &res);
665 Log("VDestroyVolumeDiskHeader: FSYNC_VGCQuery(%s, %lu) failed "
666 "with code %ld, reason %ld\n", dp->name,
667 afs_printable_uint32_lu(volid), afs_printable_int32_ld(code),
668 afs_printable_int32_ld(res.hdr.reason));
675 code = FSYNC_VGCDel(dp->name, parent, volid, FSYNC_WHATEVER, &res);
677 Log("VDestroyVolumeDiskHeader: FSYNC_VGCDel(%s, %lu, %lu) failed "
678 "with code %ld reason %ld\n", dp->name,
679 afs_printable_uint32_lu(parent),
680 afs_printable_uint32_lu(volid),
681 afs_printable_int32_ld(code),
682 afs_printable_int32_ld(res.hdr.reason));
684 #endif /* AFS_DEMAND_ATTACH_FS */
689 #endif /* FSSYNC_BUILD_CLIENT */
692 * handle a single vol header as part of VWalkVolumeHeaders.
694 * @param[in] dp disk partition
695 * @param[in] volfunc function to call when a vol header is successfully read
696 * @param[in] name full path name to the .vol header
697 * @param[out] hdr header data read in from the .vol header
698 * @param[in] locked 1 if the partition headers are locked, 0 otherwise
699 * @param[in] rock the rock to pass to volfunc
701 * @return operation status
703 * @retval -1 fatal error, stop scanning
704 * @retval 1 failed to read header
705 * @retval 2 volfunc callback indicated error after header read
708 _VHandleVolumeHeader(struct DiskPartition64 *dp, VWalkVolFunc volfunc,
709 const char *name, struct VolumeDiskHeader *hdr,
710 int locked, void *rock)
715 if ((fd = OS_OPEN(name, O_RDONLY, 0)) == INVALID_FD
716 || OS_READ(fd, hdr, sizeof(*hdr))
718 || hdr->stamp.magic != VOLUMEHEADERMAGIC) {
722 if (fd != INVALID_FD) {
726 #ifdef AFSFS_DEMAND_ATTACH_FS
728 VPartHeaderUnlock(dp);
730 #endif /* AFS_DEMAND_ATTACH_FS */
732 if (!error && volfunc) {
733 /* the volume header seems fine; call the caller-supplied
734 * 'we-found-a-volume-header' function */
737 #ifdef AFS_DEMAND_ATTACH_FS
741 #endif /* AFS_DEMAND_ATTACH_FS */
743 error = (*volfunc) (dp, name, hdr, last, rock);
752 #ifdef AFS_DEMAND_ATTACH_FS
753 if (error && !locked) {
755 /* retry reading the volume header under the partition
756 * header lock, just to be safe and ensure we're not
757 * racing something rewriting the vol header */
758 code = VPartHeaderLock(dp, WRITE_LOCK);
760 Log("Error acquiring partition write lock when "
761 "looking at header %s\n", name);
765 return _VHandleVolumeHeader(dp, volfunc, name, hdr, 1, rock);
767 #endif /* AFS_DEMAND_ATTACH_FS */
773 * walk through the list of volume headers on a partition.
775 * This function looks through all of the .vol headers on a partition, reads in
776 * each header, and calls the supplied volfunc function on each one. If the
777 * header cannot be read (or volfunc returns a positive error code), DAFS will
778 * VPartHeaderExLock() and retry. If that fails, or if we are non-DAFS, errfunc
779 * will be called (which typically will unlink the problem volume header).
781 * If volfunc returns a negative error code, walking the partition will stop
782 * and we will return an error immediately.
784 * @param[in] dp partition to walk
785 * @param[in] partpath the path opendir()
786 * @param[in] volfunc the function to call when a header is encountered, or
787 * NULL to just skip over valid headers
788 * @param[in] errfunc the function to call when a problematic header is
789 * encountered, or NULL to just skip over bad headers
790 * @param[in] rock rock for volfunc and errfunc
795 * @return operation status
797 * @retval negative fatal error, walk did not finish
800 VWalkVolumeHeaders(struct DiskPartition64 *dp, const char *partpath,
801 VWalkVolFunc volfunc, VWalkErrFunc errfunc, void *rock)
804 struct dirent *dentry = NULL;
806 struct VolumeDiskHeader diskHeader;
808 dirp = opendir(partpath);
810 Log("VWalkVolumeHeaders: cannot open directory %s\n", partpath);
815 while ((dentry = readdir(dirp))) {
816 char *p = dentry->d_name;
817 p = strrchr(dentry->d_name, '.');
818 if (p != NULL && strcmp(p, VHDREXT) == 0) {
819 char name[VMAXPATHLEN];
821 sprintf(name, "%s" OS_DIRSEP "%s", partpath, dentry->d_name);
823 code = _VHandleVolumeHeader(dp, volfunc, name, &diskHeader, -1, rock);
825 /* fatal error, stop walking */
828 if (code && errfunc) {
829 /* error with header; call the caller-supplied vol error
832 struct VolumeDiskHeader *hdr = &diskHeader;
834 /* we failed to read the header at all, so don't pass in
838 (*errfunc) (dp, name, hdr, rock);
853 * initialize a struct VLockFile.
855 * @param[in] lf struct VLockFile to initialize
856 * @param[in] path Full path to the file to use for locks. The string contents
860 VLockFileInit(struct VLockFile *lf, const char *path)
862 memset(lf, 0, sizeof(*lf));
863 lf->path = strdup(path);
865 MUTEX_INIT(&lf->mutex, "vlockfile", MUTEX_DEFAULT, 0);
870 _VOpenPath(const char *path)
874 handle = CreateFile(path,
875 GENERIC_READ | GENERIC_WRITE,
876 FILE_SHARE_READ | FILE_SHARE_WRITE,
879 FILE_ATTRIBUTE_HIDDEN,
881 if (handle == INVALID_HANDLE_VALUE) {
889 _VLockFd(FD_t handle, afs_uint32 offset, int locktype, int nonblock)
894 if (locktype == WRITE_LOCK) {
895 flags |= LOCKFILE_EXCLUSIVE_LOCK;
898 flags |= LOCKFILE_FAIL_IMMEDIATELY;
901 memset(&lap, 0, sizeof(lap));
904 if (!LockFileEx(handle, flags, 0, 1, 0, &lap)) {
905 if (GetLastError() == ERROR_LOCK_VIOLATION) {
915 _VUnlockFd(struct VLockFile *lf, afs_uint32 offset)
919 memset(&lap, 0, sizeof(lap));
922 UnlockFileEx(lf->fd, 0, 1, 0, &lap);
926 _VCloseFd(struct VLockFile *lf)
931 #else /* !AFS_NT40_ENV */
934 * open a file on the local filesystem suitable for locking
936 * @param[in] path abs path of the file to open
938 * @return file descriptor
939 * @retval INVALID_FD failure opening file
942 _VOpenPath(const char *path)
946 fd = open(path, O_RDWR | O_CREAT, 0660);
954 * lock an offset in a file descriptor.
956 * @param[in] fd file descriptor to lock
957 * @param[in] offset offset in file to lock
958 * @param[in] locktype READ_LOCK or WRITE_LOCK
959 * @param[in] nonblock 1 to fail immediately, 0 to wait to acquire lock
961 * @return operation status
963 * @retval EBUSY someone else is holding a conflicting lock and nonblock=1 was
965 * @retval EIO error acquiring file lock
968 _VLockFd(int fd, afs_uint32 offset, int locktype, int nonblock)
970 int l_type = F_WRLCK;
974 if (locktype == READ_LOCK) {
984 sf.l_whence = SEEK_SET;
986 if (fcntl(fd, cmd, &sf)) {
987 if (nonblock && (errno == EACCES || errno == EAGAIN)) {
988 /* We asked for a nonblocking lock, and it was already locked */
990 if (fcntl(fd, F_GETLK, &sf) != 0 || sf.l_pid == 0) {
991 Log("_VLockFd: fcntl failed with error %d when trying to "
992 "query the conflicting lock for fd %d (locktype=%d, "
993 "offset=%lu)\n", errno, fd, locktype,
994 afs_printable_uint32_lu(offset));
996 Log("_VLockFd: conflicting lock held on fd %d, offset %lu by "
997 "pid %ld (locktype=%d)\n", fd,
998 afs_printable_uint32_lu(offset), (long int)sf.l_pid,
1003 Log("_VLockFd: fcntl failed with error %d when trying to lock "
1004 "fd %d (locktype=%d, offset=%lu)\n", errno, fd, locktype,
1005 afs_printable_uint32_lu(offset));
1013 * close a file descriptor used for file locking.
1015 * @param[in] fd file descriptor to close
1021 Log("_VCloseFd: error %d closing fd %d\n",
1027 * unlock a file offset in a file descriptor.
1029 * @param[in] fd file descriptor to unlock
1030 * @param[in] offset offset to unlock
1033 _VUnlockFd(int fd, afs_uint32 offset)
1037 sf.l_start = offset;
1039 sf.l_type = F_UNLCK;
1040 sf.l_whence = SEEK_SET;
1042 if (fcntl(fd, F_SETLK, &sf)) {
1043 Log("_VUnlockFd: fcntl failed with error %d when trying to unlock "
1044 "fd %d\n", errno, fd);
1047 #endif /* !AFS_NT40_ENV */
1050 * reinitialize a struct VLockFile.
1052 * Use this to close the lock file (unlocking any locks in it), and effectively
1053 * restore lf to the state it was in when it was initialized. This is the same
1054 * as unlocking all of the locks on the file, without having to remember what
1055 * all of the locks were. Do not unlock previously held locks after calling
1058 * @param[in] lf struct VLockFile to reinit
1060 * @pre nobody is waiting for a lock on this lockfile or otherwise using
1061 * this lockfile at all
1064 VLockFileReinit(struct VLockFile *lf)
1066 MUTEX_ENTER(&lf->mutex);
1068 if (lf->fd != INVALID_FD) {
1070 lf->fd = INVALID_FD;
1075 MUTEX_EXIT(&lf->mutex);
1079 * lock a file on disk for the process.
1081 * @param[in] lf the struct VLockFile representing the file to lock
1082 * @param[in] offset the offset in the file to lock
1083 * @param[in] locktype READ_LOCK or WRITE_LOCK
1084 * @param[in] nonblock 0 to wait for conflicting locks to clear before
1085 * obtaining the lock; 1 to fail immediately if a
1086 * conflicting lock is held by someone else
1088 * @return operation status
1090 * @retval EBUSY someone else is holding a conflicting lock and nonblock=1 was
1092 * @retval EIO error acquiring file lock
1096 * @note do not try to lock/unlock the same offset in the same file from
1097 * different threads; use VGetDiskLock to protect threads from each other in
1098 * addition to other processes
1101 VLockFileLock(struct VLockFile *lf, afs_uint32 offset, int locktype, int nonblock)
1105 osi_Assert(locktype == READ_LOCK || locktype == WRITE_LOCK);
1107 MUTEX_ENTER(&lf->mutex);
1109 if (lf->fd == INVALID_FD) {
1110 lf->fd = _VOpenPath(lf->path);
1111 if (lf->fd == INVALID_FD) {
1112 MUTEX_EXIT(&lf->mutex);
1119 MUTEX_EXIT(&lf->mutex);
1121 code = _VLockFd(lf->fd, offset, locktype, nonblock);
1124 MUTEX_ENTER(&lf->mutex);
1125 if (--lf->refcount < 1) {
1127 lf->fd = INVALID_FD;
1129 MUTEX_EXIT(&lf->mutex);
1136 VLockFileUnlock(struct VLockFile *lf, afs_uint32 offset)
1138 MUTEX_ENTER(&lf->mutex);
1140 osi_Assert(lf->fd != INVALID_FD);
1142 if (--lf->refcount < 1) {
1144 lf->fd = INVALID_FD;
1146 _VUnlockFd(lf->fd, offset);
1149 MUTEX_EXIT(&lf->mutex);
1152 #ifdef AFS_DEMAND_ATTACH_FS
1155 * initialize a struct VDiskLock.
1157 * @param[in] dl struct VDiskLock to initialize
1158 * @param[in] lf the struct VLockFile to associate with this disk lock
1161 VDiskLockInit(struct VDiskLock *dl, struct VLockFile *lf, afs_uint32 offset)
1164 memset(dl, 0, sizeof(*dl));
1165 Lock_Init(&dl->rwlock);
1166 MUTEX_INIT(&dl->mutex, "disklock", MUTEX_DEFAULT, 0);
1167 CV_INIT(&dl->cv, "disklock cv", CV_DEFAULT, 0);
1169 dl->offset = offset;
1173 * acquire a lock on a file on local disk.
1175 * @param[in] dl the VDiskLock structure corresponding to the file on disk
1176 * @param[in] locktype READ_LOCK if you want a read lock, or WRITE_LOCK if
1177 * you want a write lock
1178 * @param[in] nonblock 0 to wait for conflicting locks to clear before
1179 * obtaining the lock; 1 to fail immediately if a
1180 * conflicting lock is held by someone else
1182 * @return operation status
1184 * @retval EBUSY someone else is holding a conflicting lock and nonblock=1 was
1186 * @retval EIO error acquiring file lock
1190 * @note while normal fcntl-y locks on Unix systems generally only work per-
1191 * process, this interface also deals with locks between threads in the
1192 * process in addition to different processes acquiring the lock
1195 VGetDiskLock(struct VDiskLock *dl, int locktype, int nonblock)
1198 osi_Assert(locktype == READ_LOCK || locktype == WRITE_LOCK);
1201 if (locktype == READ_LOCK) {
1202 ObtainReadLockNoBlock(&dl->rwlock, code);
1204 ObtainWriteLockNoBlock(&dl->rwlock, code);
1211 } else if (locktype == READ_LOCK) {
1212 ObtainReadLock(&dl->rwlock);
1214 ObtainWriteLock(&dl->rwlock);
1217 MUTEX_ENTER(&dl->mutex);
1219 if ((dl->flags & VDISKLOCK_ACQUIRING)) {
1220 /* Some other thread is waiting to acquire an fs lock. If nonblock=1,
1221 * we can return immediately, since we know we'll need to wait to
1222 * acquire. Otherwise, wait for the other thread to finish acquiring
1227 while ((dl->flags & VDISKLOCK_ACQUIRING)) {
1228 CV_WAIT(&dl->cv, &dl->mutex);
1233 if (code == 0 && !(dl->flags & VDISKLOCK_ACQUIRED)) {
1234 /* no other thread holds the lock on the actual file; so grab one */
1236 /* first try, don't block on the lock to see if we can get it without
1238 code = VLockFileLock(dl->lockfile, dl->offset, locktype, 1);
1240 if (code == EBUSY && !nonblock) {
1242 /* mark that we are waiting on the fs lock */
1243 dl->flags |= VDISKLOCK_ACQUIRING;
1245 MUTEX_EXIT(&dl->mutex);
1246 code = VLockFileLock(dl->lockfile, dl->offset, locktype, nonblock);
1247 MUTEX_ENTER(&dl->mutex);
1249 dl->flags &= ~VDISKLOCK_ACQUIRING;
1252 dl->flags |= VDISKLOCK_ACQUIRED;
1255 CV_BROADCAST(&dl->cv);
1260 if (locktype == READ_LOCK) {
1261 ReleaseReadLock(&dl->rwlock);
1263 ReleaseWriteLock(&dl->rwlock);
1266 /* successfully got the lock, so inc the number of unlocks we need
1267 * to do before we can unlock the actual file */
1271 MUTEX_EXIT(&dl->mutex);
1277 * release a lock on a file on local disk.
1279 * @param[in] dl the struct VDiskLock to release
1280 * @param[in] locktype READ_LOCK if you are unlocking a read lock, or
1281 * WRITE_LOCK if you are unlocking a write lock
1283 * @return operation status
1287 VReleaseDiskLock(struct VDiskLock *dl, int locktype)
1289 osi_Assert(locktype == READ_LOCK || locktype == WRITE_LOCK);
1291 MUTEX_ENTER(&dl->mutex);
1292 osi_Assert(dl->lockers > 0);
1294 if (--dl->lockers < 1) {
1295 /* no threads are holding this lock anymore, so we can release the
1296 * actual disk lock */
1297 VLockFileUnlock(dl->lockfile, dl->offset);
1298 dl->flags &= ~VDISKLOCK_ACQUIRED;
1301 MUTEX_EXIT(&dl->mutex);
1303 if (locktype == READ_LOCK) {
1304 ReleaseReadLock(&dl->rwlock);
1306 ReleaseWriteLock(&dl->rwlock);
1310 #endif /* AFS_DEMAND_ATTACH_FS */