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 #ifdef HAVE_SYS_FILE_H
26 #ifdef HAVE_SYS_LOCKF_H
27 #include <sys/lockf.h>
31 #include <afs/afsint.h>
33 #include <afs/errors.h>
36 #include <afs/afssyscalls.h>
38 #include <afs/afsutil.h>
44 #include "volume_inline.h"
45 #include "partition.h"
46 #include "viceinode.h"
48 #include "volinodes.h"
49 #include "vol_prototypes.h"
54 # define AFS_SETLKW F_SETLKW64
55 # define AFS_SETLK F_SETLK64
56 # define afs_st_flock flock64
58 # define AFS_SETLKW F_SETLKW
59 # define AFS_SETLK F_SETLK
60 # define afs_st_flock flock
64 /* Note: the volume creation functions herein leave the destroyMe flag in the
65 volume header ON: this means that the volumes will not be attached by the
66 file server and WILL BE DESTROYED the next time a system salvage is performed */
68 #ifdef FSSYNC_BUILD_CLIENT
70 RemoveInodes(struct afs_inode_info *stuff, Device dev, VolumeId parent,
76 /* This relies on the fact that IDEC only needs the device and NT only
77 * needs the dev and vid to decrement volume special files.
79 IH_INIT(handle, dev, parent, -1);
80 for (i = 0; i < MAXINODETYPE; i++) {
81 Inode inode = *stuff[i].inode;
82 if (VALID_INO(inode)) {
83 if (stuff[i].inodeType == VI_LINKTABLE) {
84 IH_DEC(handle, inode, parent);
86 IH_DEC(handle, inode, vid);
94 VCreateVolume(Error * ec, char *partname, VolId volumeId, VolId parentId)
95 { /* Should be the same as volumeId if there is
99 retVal = VCreateVolume_r(ec, partname, volumeId, parentId);
105 VCreateVolume_r(Error * ec, char *partname, VolId volumeId, VolId parentId)
106 { /* Should be the same as volumeId if there is
110 char headerName[VMAXPATHLEN], volumePath[VMAXPATHLEN];
112 struct DiskPartition64 *partition;
113 struct VolumeDiskHeader diskHeader;
116 Inode nearInode AFS_UNUSED = 0;
119 struct VolumeHeader tempHeader;
120 struct afs_inode_info stuff[MAXINODETYPE];
122 # ifdef AFS_DEMAND_ATTACH_FS
124 # endif /* AFS_DEMAND_ATTACH_FS */
126 init_inode_info(&tempHeader, stuff);
129 memset(&vol, 0, sizeof(vol));
131 vol.parentId = parentId;
132 vol.copyDate = time(0); /* The only date which really means when this
133 * @i(instance) of this volume was created.
134 * Creation date does not mean this */
136 /* Initialize handle for error case below. */
139 /* Verify that the parition is valid before writing to it. */
140 if (!(partition = VGetPartition_r(partname, 0))) {
141 Log("VCreateVolume: partition %s is not in service.\n", partname);
145 #if defined(NEARINODE_HINT)
146 nearInodeHash(volumeId, nearInode);
147 nearInode %= partition->f_files;
149 VGetVolumePath(ec, vol.id, &part, &name);
150 if (*ec == VNOVOL || !strcmp(partition->name, part)) {
151 /* this case is ok */
153 /* return EXDEV if it's a clone to an alternate partition
154 * otherwise assume it's a move */
155 if (vol.parentId != vol.id) {
162 # ifdef AFS_DEMAND_ATTACH_FS
163 /* volume doesn't exist yet, but we must lock it to try to prevent something
164 * else from reading it when we're e.g. half way through creating it (or
165 * something tries to create the same volume at the same time) */
166 locktype = VVolLockType(V_VOLUPD, 1);
167 rc = VLockVolumeByIdNB(volumeId, partition, locktype);
169 Log("VCreateVolume: vol %lu already locked by someone else\n",
170 afs_printable_uint32_lu(volumeId));
174 # else /* AFS_DEMAND_ATTACH_FS */
175 VLockPartition_r(partname);
176 # endif /* !AFS_DEMAND_ATTACH_FS */
178 memset(&tempHeader, 0, sizeof(tempHeader));
179 tempHeader.stamp.magic = VOLUMEHEADERMAGIC;
180 tempHeader.stamp.version = VOLUMEHEADERVERSION;
181 tempHeader.id = vol.id;
182 tempHeader.parent = vol.parentId;
183 vol.stamp.magic = VOLUMEINFOMAGIC;
184 vol.stamp.version = VOLUMEINFOVERSION;
185 vol.destroyMe = DESTROY_ME;
186 snprintf(headerName, sizeof headerName, VFORMAT,
187 afs_printable_uint32_lu(vol.id));
188 snprintf(volumePath, sizeof volumePath, "%s" OS_DIRSEP "%s",
189 VPartitionPath(partition), headerName);
190 rc = stat(volumePath, &st);
191 if (rc == 0 || errno != ENOENT) {
193 Log("VCreateVolume: Header file %s already exists!\n",
197 Log("VCreateVolume: Error %d trying to stat header file %s\n",
203 device = partition->device;
205 for (i = 0; i < MAXINODETYPE; i++) {
206 struct afs_inode_info *p = &stuff[i];
211 IH_CREATE(NULL, device, VPartitionPath(partition), nearInode,
212 (p->inodeType == VI_LINKTABLE) ? vol.parentId : vol.id,
213 INODESPECIAL, p->inodeType, vol.parentId);
214 if (!(VALID_INO(*(p->inode)))) {
215 if (errno == EEXIST && (p->inodeType == VI_LINKTABLE)) {
216 /* Increment the reference count instead. */
220 *(p->inode) = namei_MakeSpecIno(vol.parentId, VI_LINKTABLE);
221 IH_INIT(lh, device, parentId, *(p->inode));
227 code = IH_INC(lh, *(p->inode), parentId);
228 FDH_REALLYCLOSE(fdP);
237 IH_CREATE(NULL, device, VPartitionPath(partition), nearInode,
238 vol.id, INODESPECIAL, p->inodeType, vol.parentId);
241 if (!VALID_INO(*(p->inode))) {
242 Log("VCreateVolume: Problem creating %s file associated with volume header %s\n", p->description, volumePath);
246 RemoveInodes(stuff, device, vol.parentId, vol.id);
250 VDestroyVolumeDiskHeader(partition, volumeId, parentId);
252 # ifdef AFS_DEMAND_ATTACH_FS
254 VUnlockVolumeById(volumeId, partition);
256 # endif /* AFS_DEMAND_ATTACH_FS */
259 IH_INIT(handle, device, vol.parentId, *(p->inode));
260 fdP = IH_OPEN(handle);
262 Log("VCreateVolume: Problem iopen inode %s (err=%d)\n",
263 PrintInode(stmp, *(p->inode)), errno);
266 if (FDH_PWRITE(fdP, (char *)&p->stamp, sizeof(p->stamp), 0) !=
268 Log("VCreateVolume: Problem writing to inode %s (err=%d)\n",
269 PrintInode(stmp, *(p->inode)), errno);
270 FDH_REALLYCLOSE(fdP);
273 FDH_REALLYCLOSE(fdP);
275 nearInode = *(p->inode);
278 IH_INIT(handle, device, vol.parentId, tempHeader.volumeInfo);
279 fdP = IH_OPEN(handle);
281 Log("VCreateVolume: Problem iopen inode %s (err=%d)\n",
282 PrintInode(stmp, tempHeader.volumeInfo), errno);
285 if (FDH_PWRITE(fdP, (char *)&vol, sizeof(vol), 0) != sizeof(vol)) {
286 Log("VCreateVolume: Problem writing to inode %s (err=%d)\n",
287 PrintInode(stmp, tempHeader.volumeInfo), errno);
288 FDH_REALLYCLOSE(fdP);
294 VolumeHeaderToDisk(&diskHeader, &tempHeader);
295 rc = VCreateVolumeDiskHeader(&diskHeader, partition);
297 Log("VCreateVolume: Error %d trying to write volume header for "
298 "volume %u on partition %s; volume not created\n", rc,
299 vol.id, VPartitionPath(partition));
306 # ifdef AFS_DEMAND_ATTACH_FS
308 VUnlockVolumeById(volumeId, partition);
310 # endif /* AFS_DEMAND_ATTACH_FS */
311 return (VAttachVolumeByName_r(ec, partname, headerName, V_SECRETLY));
313 #endif /* FSSYNC_BUILD_CLIENT */
317 AssignVolumeName(VolumeDiskData * vol, char *name, char *ext)
320 AssignVolumeName_r(vol, name, ext);
325 AssignVolumeName_r(VolumeDiskData * vol, char *name, char *ext)
328 strncpy(vol->name, name, VNAMESIZE - 1);
329 vol->name[VNAMESIZE - 1] = '\0';
330 dot = strrchr(vol->name, '.');
331 if (dot && (strcmp(dot, ".backup") == 0 || strcmp(dot, ".readonly") == 0))
334 strncat(vol->name, ext, VNAMESIZE - 1 - strlen(vol->name));
338 CopyVolumeHeader_r(VolumeDiskData * from, VolumeDiskData * to)
340 /* The id and parentId fields are not copied; these are inviolate--the to volume
341 * is assumed to have already been created. The id's cannot be changed once
342 * creation has taken place, since they are embedded in the various inodes associated
343 * with the volume. The copydate is also inviolate--it always reflects the time
344 * this volume was created (compare with the creation date--the creation date of
345 * a backup volume is the creation date of the original parent, because the backup
346 * is used to backup the parent volume). */
350 parent = to->parentId;
351 copydate = to->copyDate;
352 memcpy(to, from, sizeof(*from));
354 to->parentId = parent;
355 to->copyDate = copydate;
356 to->destroyMe = DESTROY_ME; /* Caller must always clear this!!! */
357 to->stamp.magic = VOLUMEINFOMAGIC;
358 to->stamp.version = VOLUMEINFOVERSION;
363 CopyVolumeHeader(VolumeDiskData * from, VolumeDiskData * to)
368 code = CopyVolumeHeader_r(from, to);
374 ClearVolumeStats(VolumeDiskData * vol)
377 ClearVolumeStats_r(vol);
382 ClearVolumeStats_r(VolumeDiskData * vol)
384 memset(vol->weekUse, 0, sizeof(vol->weekUse));
390 CopyVolumeStats_r(VolumeDiskData * from, VolumeDiskData * to)
392 memcpy(to->weekUse, from->weekUse, sizeof(to->weekUse));
393 to->dayUse = from->dayUse;
394 to->dayUseDate = from->dayUseDate;
395 if (from->stat_initialized) {
396 memcpy(to->stat_reads, from->stat_reads, sizeof(to->stat_reads));
397 memcpy(to->stat_writes, from->stat_writes, sizeof(to->stat_writes));
398 memcpy(to->stat_fileSameAuthor, from->stat_fileSameAuthor,
399 sizeof(to->stat_fileSameAuthor));
400 memcpy(to->stat_fileDiffAuthor, from->stat_fileDiffAuthor,
401 sizeof(to->stat_fileDiffAuthor));
402 memcpy(to->stat_dirSameAuthor, from->stat_dirSameAuthor,
403 sizeof(to->stat_dirSameAuthor));
404 memcpy(to->stat_dirDiffAuthor, from->stat_dirDiffAuthor,
405 sizeof(to->stat_dirDiffAuthor));
410 CopyVolumeStats(VolumeDiskData * from, VolumeDiskData * to)
413 CopyVolumeStats_r(from, to);
418 * read an existing volume disk header.
420 * @param[in] volid volume id
421 * @param[in] dp disk partition object
422 * @param[out] hdr volume disk header or NULL
424 * @note if hdr is NULL, this is essentially an existence test for the vol
427 * @return operation status
429 * @retval -1 volume header doesn't exist
430 * @retval EIO failed to read volume header
435 VReadVolumeDiskHeader(VolumeId volid,
436 struct DiskPartition64 * dp,
437 VolumeDiskHeader_t * hdr)
441 char path[MAXPATHLEN];
443 snprintf(path, sizeof(path), "%s" OS_DIRSEP VFORMAT,
444 VPartitionPath(dp), afs_printable_uint32_lu(volid));
445 fd = open(path, O_RDONLY);
447 Log("VReadVolumeDiskHeader: Couldn't open header for volume %lu (errno %d).\n",
448 afs_printable_uint32_lu(volid), errno);
451 } else if (hdr && read(fd, hdr, sizeof(*hdr)) != sizeof(*hdr)) {
452 Log("VReadVolumeDiskHeader: Couldn't read header for volume %lu.\n",
453 afs_printable_uint32_lu(volid));
463 #ifdef FSSYNC_BUILD_CLIENT
465 * write an existing volume disk header.
467 * @param[in] hdr volume disk header
468 * @param[in] dp disk partition object
469 * @param[in] cr assert if O_CREAT | O_EXCL should be passed to open()
471 * @return operation status
473 * @retval -1 volume header doesn't exist
474 * @retval EIO failed to write volume header
479 _VWriteVolumeDiskHeader(VolumeDiskHeader_t * hdr,
480 struct DiskPartition64 * dp,
485 char path[MAXPATHLEN];
487 #ifdef AFS_DEMAND_ATTACH_FS
488 /* prevent racing with VGC scanners reading the vol header while we are
490 code = VPartHeaderLock(dp, READ_LOCK);
494 #endif /* AFS_DEMAND_ATTACH_FS */
498 snprintf(path, sizeof(path), "%s" OS_DIRSEP VFORMAT,
499 VPartitionPath(dp), afs_printable_uint32_lu(hdr->id));
500 fd = open(path, flags, 0644);
503 Log("_VWriteVolumeDiskHeader: Couldn't open header for volume %lu, "
504 "error = %d\n", afs_printable_uint32_lu(hdr->id), errno);
505 } else if (write(fd, hdr, sizeof(*hdr)) != sizeof(*hdr)) {
506 Log("_VWriteVolumeDiskHeader: Couldn't write header for volume %lu, "
507 "error = %d\n", afs_printable_uint32_lu(hdr->id), errno);
512 if (close(fd) != 0) {
513 Log("_VWriteVolumeDiskHeader: Error closing header for volume "
514 "%lu, errno %d\n", afs_printable_uint32_lu(hdr->id), errno);
518 #ifdef AFS_DEMAND_ATTACH_FS
519 VPartHeaderUnlock(dp, READ_LOCK);
520 #endif /* AFS_DEMAND_ATTACH_FS */
526 * write an existing volume disk header.
528 * @param[in] hdr volume disk header
529 * @param[in] dp disk partition object
531 * @return operation status
533 * @retval ENOENT volume header doesn't exist
534 * @retval EIO failed to write volume header
537 VWriteVolumeDiskHeader(VolumeDiskHeader_t * hdr,
538 struct DiskPartition64 * dp)
542 #ifdef AFS_DEMAND_ATTACH_FS
543 VolumeDiskHeader_t oldhdr;
544 int delvgc = 0, addvgc = 0;
547 /* first, see if anything with the volume IDs have changed; if so, we
548 * need to update the VGC */
550 code = VReadVolumeDiskHeader(hdr->id, dp, &oldhdr);
551 if (code == 0 && (oldhdr.id != hdr->id || oldhdr.parent != hdr->parent)) {
552 /* the vol id or parent vol id changed; need to delete the VGC entry
553 * for the old vol id/parent, and add the new one */
558 /* couldn't get the old header info; add the new header info to the
559 * VGC in case it hasn't been added yet */
563 #endif /* AFS_DEMAND_ATTACH_FS */
565 code = _VWriteVolumeDiskHeader(hdr, dp, 0);
570 #ifdef AFS_DEMAND_ATTACH_FS
572 memset(&res, 0, sizeof(res));
573 code = FSYNC_VGCDel(dp->name, oldhdr.parent, oldhdr.id, FSYNC_WHATEVER, &res);
575 /* unknown vol id is okay; it just further suggests the old header
576 * data was bogus, which is fine since we're trying to fix it */
577 if (code && res.hdr.reason != FSYNC_UNKNOWN_VOLID) {
578 Log("VWriteVolumeDiskHeader: FSYNC_VGCDel(%s, %lu, %lu) "
579 "failed with code %ld reason %ld\n", dp->name,
580 afs_printable_uint32_lu(oldhdr.parent),
581 afs_printable_uint32_lu(oldhdr.id),
582 afs_printable_int32_ld(code),
583 afs_printable_int32_ld(res.hdr.reason));
588 memset(&res, 0, sizeof(res));
589 code = FSYNC_VGCAdd(dp->name, hdr->parent, hdr->id, FSYNC_WHATEVER, &res);
591 Log("VWriteVolumeDiskHeader: FSYNC_VGCAdd(%s, %lu, %lu) "
592 "failed with code %ld reason %ld\n", dp->name,
593 afs_printable_uint32_lu(hdr->parent),
594 afs_printable_uint32_lu(hdr->id),
595 afs_printable_int32_ld(code),
596 afs_printable_int32_ld(res.hdr.reason));
600 #endif /* AFS_DEMAND_ATTACH_FS */
607 * create and write a volume disk header to disk.
609 * @param[in] hdr volume disk header
610 * @param[in] dp disk partition object
612 * @return operation status
614 * @retval EEXIST volume header already exists
615 * @retval EIO failed to write volume header
620 VCreateVolumeDiskHeader(VolumeDiskHeader_t * hdr,
621 struct DiskPartition64 * dp)
624 #ifdef AFS_DEMAND_ATTACH_FS
626 #endif /* AFS_DEMAND_ATTACH_FS */
628 code = _VWriteVolumeDiskHeader(hdr, dp, O_CREAT | O_EXCL);
633 #ifdef AFS_DEMAND_ATTACH_FS
634 memset(&res, 0, sizeof(res));
635 code = FSYNC_VGCAdd(dp->name, hdr->parent, hdr->id, FSYNC_WHATEVER, &res);
637 Log("VCreateVolumeDiskHeader: FSYNC_VGCAdd(%s, %lu, %lu) failed "
638 "with code %ld reason %ld\n", dp->name,
639 afs_printable_uint32_lu(hdr->parent),
640 afs_printable_uint32_lu(hdr->id),
641 afs_printable_int32_ld(code),
642 afs_printable_int32_ld(res.hdr.reason));
644 #endif /* AFS_DEMAND_ATTACH_FS */
652 * destroy a volume disk header.
654 * @param[in] dp disk partition object
655 * @param[in] volid volume id
656 * @param[in] parent parent's volume id, 0 if unknown
658 * @return operation status
661 * @note if parent is 0, the parent volume ID will be looked up from the
664 * @note for non-DAFS, parent is currently ignored
667 VDestroyVolumeDiskHeader(struct DiskPartition64 * dp,
672 char path[MAXPATHLEN];
673 #ifdef AFS_DEMAND_ATTACH_FS
675 #endif /* AFS_DEMAND_ATTACH_FS */
677 snprintf(path, sizeof(path), "%s" OS_DIRSEP VFORMAT,
678 VPartitionPath(dp), afs_printable_uint32_lu(volid));
681 Log("VDestroyVolumeDiskHeader: Couldn't unlink disk header, error = %d\n", errno);
685 #ifdef AFS_DEMAND_ATTACH_FS
686 memset(&res, 0, sizeof(res));
688 FSSYNC_VGQry_response_t q_res;
690 code = FSYNC_VGCQuery(dp->name, volid, &q_res, &res);
692 Log("VDestroyVolumeDiskHeader: FSYNC_VGCQuery(%s, %lu) failed "
693 "with code %ld, reason %ld\n", dp->name,
694 afs_printable_uint32_lu(volid), afs_printable_int32_ld(code),
695 afs_printable_int32_ld(res.hdr.reason));
702 code = FSYNC_VGCDel(dp->name, parent, volid, FSYNC_WHATEVER, &res);
704 Log("VDestroyVolumeDiskHeader: FSYNC_VGCDel(%s, %lu, %lu) failed "
705 "with code %ld reason %ld\n", dp->name,
706 afs_printable_uint32_lu(parent),
707 afs_printable_uint32_lu(volid),
708 afs_printable_int32_ld(code),
709 afs_printable_int32_ld(res.hdr.reason));
711 #endif /* AFS_DEMAND_ATTACH_FS */
716 #endif /* FSSYNC_BUILD_CLIENT */
719 * handle a single vol header as part of VWalkVolumeHeaders.
721 * @param[in] dp disk partition
722 * @param[in] volfunc function to call when a vol header is successfully read
723 * @param[in] name full path name to the .vol header
724 * @param[out] hdr header data read in from the .vol header
725 * @param[in] locked 1 if the partition headers are locked, 0 otherwise
726 * @param[in] rock the rock to pass to volfunc
728 * @return operation status
730 * @retval -1 fatal error, stop scanning
731 * @retval 1 failed to read header
732 * @retval 2 volfunc callback indicated error after header read
735 _VHandleVolumeHeader(struct DiskPartition64 *dp, VWalkVolFunc volfunc,
736 const char *name, struct VolumeDiskHeader *hdr,
737 int locked, void *rock)
742 if ((fd = OS_OPEN(name, O_RDONLY, 0)) == INVALID_FD
743 || OS_READ(fd, hdr, sizeof(*hdr))
745 || hdr->stamp.magic != VOLUMEHEADERMAGIC) {
749 if (fd != INVALID_FD) {
753 #ifdef AFSFS_DEMAND_ATTACH_FS
755 VPartHeaderUnlock(dp);
757 #endif /* AFS_DEMAND_ATTACH_FS */
759 if (!error && volfunc) {
760 /* the volume header seems fine; call the caller-supplied
761 * 'we-found-a-volume-header' function */
764 #ifdef AFS_DEMAND_ATTACH_FS
768 #endif /* AFS_DEMAND_ATTACH_FS */
770 error = (*volfunc) (dp, name, hdr, last, rock);
779 #ifdef AFS_DEMAND_ATTACH_FS
780 if (error && !locked) {
782 /* retry reading the volume header under the partition
783 * header lock, just to be safe and ensure we're not
784 * racing something rewriting the vol header */
785 code = VPartHeaderLock(dp, WRITE_LOCK);
787 Log("Error acquiring partition write lock when "
788 "looking at header %s\n", name);
792 return _VHandleVolumeHeader(dp, volfunc, name, hdr, 1, rock);
794 #endif /* AFS_DEMAND_ATTACH_FS */
800 * walk through the list of volume headers on a partition.
802 * This function looks through all of the .vol headers on a partition, reads in
803 * each header, and calls the supplied volfunc function on each one. If the
804 * header cannot be read (or volfunc returns a positive error code), DAFS will
805 * VPartHeaderExLock() and retry. If that fails, or if we are non-DAFS, errfunc
806 * will be called (which typically will unlink the problem volume header).
808 * If volfunc returns a negative error code, walking the partition will stop
809 * and we will return an error immediately.
811 * @param[in] dp partition to walk
812 * @param[in] partpath the path opendir()
813 * @param[in] volfunc the function to call when a header is encountered, or
814 * NULL to just skip over valid headers
815 * @param[in] errfunc the function to call when a problematic header is
816 * encountered, or NULL to just skip over bad headers
817 * @param[in] rock rock for volfunc and errfunc
822 * @return operation status
824 * @retval negative fatal error, walk did not finish
827 VWalkVolumeHeaders(struct DiskPartition64 *dp, const char *partpath,
828 VWalkVolFunc volfunc, VWalkErrFunc errfunc, void *rock)
831 struct dirent *dentry;
833 struct VolumeDiskHeader diskHeader;
835 dirp = opendir(partpath);
837 Log("VWalkVolumeHeaders: cannot open directory %s\n", partpath);
842 while ((dentry = readdir(dirp)) != NULL) {
844 p = strrchr(dentry->d_name, '.');
845 if (p != NULL && strcmp(p, VHDREXT) == 0) {
846 char name[VMAXPATHLEN];
848 sprintf(name, "%s" OS_DIRSEP "%s", partpath, dentry->d_name);
850 code = _VHandleVolumeHeader(dp, volfunc, name, &diskHeader, -1, rock);
852 /* fatal error, stop walking */
855 if (code && errfunc) {
856 /* error with header; call the caller-supplied vol error
859 struct VolumeDiskHeader *hdr = &diskHeader;
861 /* we failed to read the header at all, so don't pass in
865 (*errfunc) (dp, name, hdr, rock);
880 * initialize a struct VLockFile.
882 * @param[in] lf struct VLockFile to initialize
883 * @param[in] path Full path to the file to use for locks. The string contents
887 VLockFileInit(struct VLockFile *lf, const char *path)
889 memset(lf, 0, sizeof(*lf));
890 lf->path = strdup(path);
892 MUTEX_INIT(&lf->mutex, "vlockfile", MUTEX_DEFAULT, 0);
897 _VOpenPath(const char *path)
901 handle = CreateFile(path,
902 GENERIC_READ | GENERIC_WRITE,
903 FILE_SHARE_READ | FILE_SHARE_WRITE,
906 FILE_ATTRIBUTE_HIDDEN,
908 if (handle == INVALID_HANDLE_VALUE) {
916 _VLockFd(FD_t handle, afs_uint32 offset, int locktype, int nonblock)
921 if (locktype == WRITE_LOCK) {
922 flags |= LOCKFILE_EXCLUSIVE_LOCK;
925 flags |= LOCKFILE_FAIL_IMMEDIATELY;
928 memset(&lap, 0, sizeof(lap));
931 if (!LockFileEx(handle, flags, 0, 1, 0, &lap)) {
932 if (GetLastError() == ERROR_LOCK_VIOLATION) {
942 _VUnlockFd(FD_t handle, afs_uint32 offset)
946 memset(&lap, 0, sizeof(lap));
949 UnlockFileEx(handle, 0, 1, 0, &lap);
953 _VCloseFd(FD_t handle)
958 #else /* !AFS_NT40_ENV */
961 * open a file on the local filesystem suitable for locking
963 * @param[in] path abs path of the file to open
965 * @return file descriptor
966 * @retval INVALID_FD failure opening file
969 _VOpenPath(const char *path)
973 fd = open(path, O_RDWR | O_CREAT, 0660);
981 * lock an offset in a file descriptor.
983 * @param[in] fd file descriptor to lock
984 * @param[in] offset offset in file to lock
985 * @param[in] locktype READ_LOCK or WRITE_LOCK
986 * @param[in] nonblock 1 to fail immediately, 0 to wait to acquire lock
988 * @return operation status
990 * @retval EBUSY someone else is holding a conflicting lock and nonblock=1 was
992 * @retval EIO error acquiring file lock
995 _VLockFd(FD_t fd, afs_uint32 offset, int locktype, int nonblock)
997 int l_type = F_WRLCK;
998 int cmd = AFS_SETLKW;
999 struct afs_st_flock sf;
1001 if (locktype == READ_LOCK) {
1008 sf.l_start = offset;
1011 sf.l_whence = SEEK_SET;
1013 if (fcntl(fd, cmd, &sf)) {
1014 if (nonblock && (errno == EACCES || errno == EAGAIN)) {
1015 /* We asked for a nonblocking lock, and it was already locked */
1017 if (fcntl(fd, F_GETLK, &sf) != 0 || sf.l_pid == 0) {
1018 Log("_VLockFd: fcntl failed with error %d when trying to "
1019 "query the conflicting lock for fd %d (locktype=%d, "
1020 "offset=%lu)\n", errno, fd, locktype,
1021 afs_printable_uint32_lu(offset));
1023 Log("_VLockFd: conflicting lock held on fd %d, offset %lu by "
1024 "pid %ld (locktype=%d)\n", fd,
1025 afs_printable_uint32_lu(offset), (long int)sf.l_pid,
1030 Log("_VLockFd: fcntl failed with error %d when trying to lock "
1031 "fd %d (locktype=%d, offset=%lu)\n", errno, fd, locktype,
1032 afs_printable_uint32_lu(offset));
1040 * close a file descriptor used for file locking.
1042 * @param[in] fd file descriptor to close
1048 Log("_VCloseFd: error %d closing fd %d\n",
1054 * unlock a file offset in a file descriptor.
1056 * @param[in] fd file descriptor to unlock
1057 * @param[in] offset offset to unlock
1060 _VUnlockFd(FD_t fd, afs_uint32 offset)
1062 struct afs_st_flock sf;
1064 sf.l_start = offset;
1066 sf.l_type = F_UNLCK;
1067 sf.l_whence = SEEK_SET;
1069 if (fcntl(fd, AFS_SETLK, &sf)) {
1070 Log("_VUnlockFd: fcntl failed with error %d when trying to unlock "
1071 "fd %d\n", errno, fd);
1074 #endif /* !AFS_NT40_ENV */
1077 * reinitialize a struct VLockFile.
1079 * Use this to close the lock file (unlocking any locks in it), and effectively
1080 * restore lf to the state it was in when it was initialized. This is the same
1081 * as unlocking all of the locks on the file, without having to remember what
1082 * all of the locks were. Do not unlock previously held locks after calling
1085 * @param[in] lf struct VLockFile to reinit
1087 * @pre nobody is waiting for a lock on this lockfile or otherwise using
1088 * this lockfile at all
1091 VLockFileReinit(struct VLockFile *lf)
1093 MUTEX_ENTER(&lf->mutex);
1095 if (lf->fd != INVALID_FD) {
1097 lf->fd = INVALID_FD;
1102 MUTEX_EXIT(&lf->mutex);
1106 * lock a file on disk for the process.
1108 * @param[in] lf the struct VLockFile representing the file to lock
1109 * @param[in] offset the offset in the file to lock
1110 * @param[in] locktype READ_LOCK or WRITE_LOCK
1111 * @param[in] nonblock 0 to wait for conflicting locks to clear before
1112 * obtaining the lock; 1 to fail immediately if a
1113 * conflicting lock is held by someone else
1115 * @return operation status
1117 * @retval EBUSY someone else is holding a conflicting lock and nonblock=1 was
1119 * @retval EIO error acquiring file lock
1123 * @note do not try to lock/unlock the same offset in the same file from
1124 * different threads; use VGetDiskLock to protect threads from each other in
1125 * addition to other processes
1128 VLockFileLock(struct VLockFile *lf, afs_uint32 offset, int locktype, int nonblock)
1132 osi_Assert(locktype == READ_LOCK || locktype == WRITE_LOCK);
1134 MUTEX_ENTER(&lf->mutex);
1136 if (lf->fd == INVALID_FD) {
1137 lf->fd = _VOpenPath(lf->path);
1138 if (lf->fd == INVALID_FD) {
1139 MUTEX_EXIT(&lf->mutex);
1146 MUTEX_EXIT(&lf->mutex);
1148 code = _VLockFd(lf->fd, offset, locktype, nonblock);
1151 MUTEX_ENTER(&lf->mutex);
1152 if (--lf->refcount < 1) {
1154 lf->fd = INVALID_FD;
1156 MUTEX_EXIT(&lf->mutex);
1163 VLockFileUnlock(struct VLockFile *lf, afs_uint32 offset)
1165 MUTEX_ENTER(&lf->mutex);
1167 osi_Assert(lf->fd != INVALID_FD);
1169 if (--lf->refcount < 1) {
1171 lf->fd = INVALID_FD;
1173 _VUnlockFd(lf->fd, offset);
1176 MUTEX_EXIT(&lf->mutex);
1179 #ifdef AFS_DEMAND_ATTACH_FS
1182 * initialize a struct VDiskLock.
1184 * @param[in] dl struct VDiskLock to initialize
1185 * @param[in] lf the struct VLockFile to associate with this disk lock
1188 VDiskLockInit(struct VDiskLock *dl, struct VLockFile *lf, afs_uint32 offset)
1191 memset(dl, 0, sizeof(*dl));
1192 Lock_Init(&dl->rwlock);
1193 MUTEX_INIT(&dl->mutex, "disklock", MUTEX_DEFAULT, 0);
1194 CV_INIT(&dl->cv, "disklock cv", CV_DEFAULT, 0);
1196 dl->offset = offset;
1200 * acquire a lock on a file on local disk.
1202 * @param[in] dl the VDiskLock structure corresponding to the file on disk
1203 * @param[in] locktype READ_LOCK if you want a read lock, or WRITE_LOCK if
1204 * you want a write lock
1205 * @param[in] nonblock 0 to wait for conflicting locks to clear before
1206 * obtaining the lock; 1 to fail immediately if a
1207 * conflicting lock is held by someone else
1209 * @return operation status
1211 * @retval EBUSY someone else is holding a conflicting lock and nonblock=1 was
1213 * @retval EIO error acquiring file lock
1217 * @note while normal fcntl-y locks on Unix systems generally only work per-
1218 * process, this interface also deals with locks between threads in the
1219 * process in addition to different processes acquiring the lock
1222 VGetDiskLock(struct VDiskLock *dl, int locktype, int nonblock)
1225 osi_Assert(locktype == READ_LOCK || locktype == WRITE_LOCK);
1228 if (locktype == READ_LOCK) {
1229 ObtainReadLockNoBlock(&dl->rwlock, code);
1231 ObtainWriteLockNoBlock(&dl->rwlock, code);
1238 } else if (locktype == READ_LOCK) {
1239 ObtainReadLock(&dl->rwlock);
1241 ObtainWriteLock(&dl->rwlock);
1244 MUTEX_ENTER(&dl->mutex);
1246 if ((dl->flags & VDISKLOCK_ACQUIRING)) {
1247 /* Some other thread is waiting to acquire an fs lock. If nonblock=1,
1248 * we can return immediately, since we know we'll need to wait to
1249 * acquire. Otherwise, wait for the other thread to finish acquiring
1254 while ((dl->flags & VDISKLOCK_ACQUIRING)) {
1255 CV_WAIT(&dl->cv, &dl->mutex);
1260 if (code == 0 && !(dl->flags & VDISKLOCK_ACQUIRED)) {
1261 /* no other thread holds the lock on the actual file; so grab one */
1263 /* first try, don't block on the lock to see if we can get it without
1265 code = VLockFileLock(dl->lockfile, dl->offset, locktype, 1);
1267 if (code == EBUSY && !nonblock) {
1269 /* mark that we are waiting on the fs lock */
1270 dl->flags |= VDISKLOCK_ACQUIRING;
1272 MUTEX_EXIT(&dl->mutex);
1273 code = VLockFileLock(dl->lockfile, dl->offset, locktype, nonblock);
1274 MUTEX_ENTER(&dl->mutex);
1276 dl->flags &= ~VDISKLOCK_ACQUIRING;
1279 dl->flags |= VDISKLOCK_ACQUIRED;
1282 CV_BROADCAST(&dl->cv);
1287 if (locktype == READ_LOCK) {
1288 ReleaseReadLock(&dl->rwlock);
1290 ReleaseWriteLock(&dl->rwlock);
1293 /* successfully got the lock, so inc the number of unlocks we need
1294 * to do before we can unlock the actual file */
1298 MUTEX_EXIT(&dl->mutex);
1304 * release a lock on a file on local disk.
1306 * @param[in] dl the struct VDiskLock to release
1307 * @param[in] locktype READ_LOCK if you are unlocking a read lock, or
1308 * WRITE_LOCK if you are unlocking a write lock
1310 * @return operation status
1314 VReleaseDiskLock(struct VDiskLock *dl, int locktype)
1316 osi_Assert(locktype == READ_LOCK || locktype == WRITE_LOCK);
1318 MUTEX_ENTER(&dl->mutex);
1319 osi_Assert(dl->lockers > 0);
1321 if (--dl->lockers < 1) {
1322 /* no threads are holding this lock anymore, so we can release the
1323 * actual disk lock */
1324 VLockFileUnlock(dl->lockfile, dl->offset);
1325 dl->flags &= ~VDISKLOCK_ACQUIRED;
1328 MUTEX_EXIT(&dl->mutex);
1330 if (locktype == READ_LOCK) {
1331 ReleaseReadLock(&dl->rwlock);
1333 ReleaseWriteLock(&dl->rwlock);
1337 #endif /* AFS_DEMAND_ATTACH_FS */