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 * read an existing volume disk header.
392 * @param[in] volid volume id
393 * @param[in] dp disk partition object
394 * @param[out] hdr volume disk header or NULL
396 * @note if hdr is NULL, this is essentially an existence test for the vol
399 * @return operation status
401 * @retval -1 volume header doesn't exist
402 * @retval EIO failed to read volume header
407 VReadVolumeDiskHeader(VolumeId volid,
408 struct DiskPartition64 * dp,
409 VolumeDiskHeader_t * hdr)
413 char path[MAXPATHLEN];
415 snprintf(path, sizeof(path), "%s" OS_DIRSEP VFORMAT,
416 VPartitionPath(dp), afs_printable_uint32_lu(volid));
417 fd = open(path, O_RDONLY);
419 Log("VReadVolumeDiskHeader: Couldn't open header for volume %lu (errno %d).\n",
420 afs_printable_uint32_lu(volid), errno);
423 } else if (hdr && read(fd, hdr, sizeof(*hdr)) != sizeof(*hdr)) {
424 Log("VReadVolumeDiskHeader: Couldn't read header for volume %lu.\n",
425 afs_printable_uint32_lu(volid));
435 #ifdef FSSYNC_BUILD_CLIENT
437 * write an existing volume disk header.
439 * @param[in] hdr volume disk header
440 * @param[in] dp disk partition object
441 * @param[in] cr assert if O_CREAT | O_EXCL should be passed to open()
443 * @return operation status
445 * @retval -1 volume header doesn't exist
446 * @retval EIO failed to write volume header
451 _VWriteVolumeDiskHeader(VolumeDiskHeader_t * hdr,
452 struct DiskPartition64 * dp,
457 char path[MAXPATHLEN];
459 #ifdef AFS_DEMAND_ATTACH_FS
460 /* prevent racing with VGC scanners reading the vol header while we are
462 code = VPartHeaderLock(dp, READ_LOCK);
466 #endif /* AFS_DEMAND_ATTACH_FS */
470 snprintf(path, sizeof(path), "%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 snprintf(path, sizeof(path), "%s" OS_DIRSEP VFORMAT,
650 VPartitionPath(dp), afs_printable_uint32_lu(volid));
653 Log("VDestroyVolumeDiskHeader: Couldn't unlink disk header, error = %d\n", errno);
657 #ifdef AFS_DEMAND_ATTACH_FS
658 memset(&res, 0, sizeof(res));
660 FSSYNC_VGQry_response_t q_res;
662 code = FSYNC_VGCQuery(dp->name, volid, &q_res, &res);
664 Log("VDestroyVolumeDiskHeader: FSYNC_VGCQuery(%s, %lu) failed "
665 "with code %ld, reason %ld\n", dp->name,
666 afs_printable_uint32_lu(volid), afs_printable_int32_ld(code),
667 afs_printable_int32_ld(res.hdr.reason));
674 code = FSYNC_VGCDel(dp->name, parent, volid, FSYNC_WHATEVER, &res);
676 Log("VDestroyVolumeDiskHeader: FSYNC_VGCDel(%s, %lu, %lu) failed "
677 "with code %ld reason %ld\n", dp->name,
678 afs_printable_uint32_lu(parent),
679 afs_printable_uint32_lu(volid),
680 afs_printable_int32_ld(code),
681 afs_printable_int32_ld(res.hdr.reason));
683 #endif /* AFS_DEMAND_ATTACH_FS */
688 #endif /* FSSYNC_BUILD_CLIENT */
691 * handle a single vol header as part of VWalkVolumeHeaders.
693 * @param[in] dp disk partition
694 * @param[in] volfunc function to call when a vol header is successfully read
695 * @param[in] name full path name to the .vol header
696 * @param[out] hdr header data read in from the .vol header
697 * @param[in] locked 1 if the partition headers are locked, 0 otherwise
698 * @param[in] rock the rock to pass to volfunc
700 * @return operation status
702 * @retval -1 fatal error, stop scanning
703 * @retval 1 failed to read header
704 * @retval 2 volfunc callback indicated error after header read
707 _VHandleVolumeHeader(struct DiskPartition64 *dp, VWalkVolFunc volfunc,
708 const char *name, struct VolumeDiskHeader *hdr,
709 int locked, void *rock)
714 if ((fd = OS_OPEN(name, O_RDONLY, 0)) == INVALID_FD
715 || OS_READ(fd, hdr, sizeof(*hdr))
717 || hdr->stamp.magic != VOLUMEHEADERMAGIC) {
721 if (fd != INVALID_FD) {
725 #ifdef AFSFS_DEMAND_ATTACH_FS
727 VPartHeaderUnlock(dp);
729 #endif /* AFS_DEMAND_ATTACH_FS */
731 if (!error && volfunc) {
732 /* the volume header seems fine; call the caller-supplied
733 * 'we-found-a-volume-header' function */
736 #ifdef AFS_DEMAND_ATTACH_FS
740 #endif /* AFS_DEMAND_ATTACH_FS */
742 error = (*volfunc) (dp, name, hdr, last, rock);
751 #ifdef AFS_DEMAND_ATTACH_FS
752 if (error && !locked) {
754 /* retry reading the volume header under the partition
755 * header lock, just to be safe and ensure we're not
756 * racing something rewriting the vol header */
757 code = VPartHeaderLock(dp, WRITE_LOCK);
759 Log("Error acquiring partition write lock when "
760 "looking at header %s\n", name);
764 return _VHandleVolumeHeader(dp, volfunc, name, hdr, 1, rock);
766 #endif /* AFS_DEMAND_ATTACH_FS */
772 * walk through the list of volume headers on a partition.
774 * This function looks through all of the .vol headers on a partition, reads in
775 * each header, and calls the supplied volfunc function on each one. If the
776 * header cannot be read (or volfunc returns a positive error code), DAFS will
777 * VPartHeaderExLock() and retry. If that fails, or if we are non-DAFS, errfunc
778 * will be called (which typically will unlink the problem volume header).
780 * If volfunc returns a negative error code, walking the partition will stop
781 * and we will return an error immediately.
783 * @param[in] dp partition to walk
784 * @param[in] partpath the path opendir()
785 * @param[in] volfunc the function to call when a header is encountered, or
786 * NULL to just skip over valid headers
787 * @param[in] errfunc the function to call when a problematic header is
788 * encountered, or NULL to just skip over bad headers
789 * @param[in] rock rock for volfunc and errfunc
794 * @return operation status
796 * @retval negative fatal error, walk did not finish
799 VWalkVolumeHeaders(struct DiskPartition64 *dp, const char *partpath,
800 VWalkVolFunc volfunc, VWalkErrFunc errfunc, void *rock)
803 struct dirent *dentry = NULL;
805 struct VolumeDiskHeader diskHeader;
807 dirp = opendir(partpath);
809 Log("VWalkVolumeHeaders: cannot open directory %s\n", partpath);
814 while ((dentry = readdir(dirp))) {
815 char *p = dentry->d_name;
816 p = strrchr(dentry->d_name, '.');
817 if (p != NULL && strcmp(p, VHDREXT) == 0) {
818 char name[VMAXPATHLEN];
820 sprintf(name, "%s" OS_DIRSEP "%s", partpath, dentry->d_name);
822 code = _VHandleVolumeHeader(dp, volfunc, name, &diskHeader, -1, rock);
824 /* fatal error, stop walking */
827 if (code && errfunc) {
828 /* error with header; call the caller-supplied vol error
831 struct VolumeDiskHeader *hdr = &diskHeader;
833 /* we failed to read the header at all, so don't pass in
837 (*errfunc) (dp, name, hdr, rock);
852 * initialize a struct VLockFile.
854 * @param[in] lf struct VLockFile to initialize
855 * @param[in] path Full path to the file to use for locks. The string contents
859 VLockFileInit(struct VLockFile *lf, const char *path)
861 memset(lf, 0, sizeof(*lf));
862 lf->path = strdup(path);
864 MUTEX_INIT(&lf->mutex, "vlockfile", MUTEX_DEFAULT, 0);
869 _VOpenPath(const char *path)
873 handle = CreateFile(path,
874 GENERIC_READ | GENERIC_WRITE,
875 FILE_SHARE_READ | FILE_SHARE_WRITE,
878 FILE_ATTRIBUTE_HIDDEN,
880 if (handle == INVALID_HANDLE_VALUE) {
888 _VLockFd(FD_t handle, afs_uint32 offset, int locktype, int nonblock)
893 if (locktype == WRITE_LOCK) {
894 flags |= LOCKFILE_EXCLUSIVE_LOCK;
897 flags |= LOCKFILE_FAIL_IMMEDIATELY;
900 memset(&lap, 0, sizeof(lap));
903 if (!LockFileEx(handle, flags, 0, 1, 0, &lap)) {
904 if (GetLastError() == ERROR_LOCK_VIOLATION) {
914 _VUnlockFd(FD_t handle, afs_uint32 offset)
918 memset(&lap, 0, sizeof(lap));
921 UnlockFileEx(handle, 0, 1, 0, &lap);
925 _VCloseFd(FD_t handle)
930 #else /* !AFS_NT40_ENV */
933 * open a file on the local filesystem suitable for locking
935 * @param[in] path abs path of the file to open
937 * @return file descriptor
938 * @retval INVALID_FD failure opening file
941 _VOpenPath(const char *path)
945 fd = open(path, O_RDWR | O_CREAT, 0660);
953 * lock an offset in a file descriptor.
955 * @param[in] fd file descriptor to lock
956 * @param[in] offset offset in file to lock
957 * @param[in] locktype READ_LOCK or WRITE_LOCK
958 * @param[in] nonblock 1 to fail immediately, 0 to wait to acquire lock
960 * @return operation status
962 * @retval EBUSY someone else is holding a conflicting lock and nonblock=1 was
964 * @retval EIO error acquiring file lock
967 _VLockFd(FD_t fd, afs_uint32 offset, int locktype, int nonblock)
969 int l_type = F_WRLCK;
970 int cmd = AFS_SETLKW;
971 struct afs_st_flock sf;
973 if (locktype == READ_LOCK) {
983 sf.l_whence = SEEK_SET;
985 if (fcntl(fd, cmd, &sf)) {
986 if (nonblock && (errno == EACCES || errno == EAGAIN)) {
987 /* We asked for a nonblocking lock, and it was already locked */
989 if (fcntl(fd, F_GETLK, &sf) != 0 || sf.l_pid == 0) {
990 Log("_VLockFd: fcntl failed with error %d when trying to "
991 "query the conflicting lock for fd %d (locktype=%d, "
992 "offset=%lu)\n", errno, fd, locktype,
993 afs_printable_uint32_lu(offset));
995 Log("_VLockFd: conflicting lock held on fd %d, offset %lu by "
996 "pid %ld (locktype=%d)\n", fd,
997 afs_printable_uint32_lu(offset), (long int)sf.l_pid,
1002 Log("_VLockFd: fcntl failed with error %d when trying to lock "
1003 "fd %d (locktype=%d, offset=%lu)\n", errno, fd, locktype,
1004 afs_printable_uint32_lu(offset));
1012 * close a file descriptor used for file locking.
1014 * @param[in] fd file descriptor to close
1020 Log("_VCloseFd: error %d closing fd %d\n",
1026 * unlock a file offset in a file descriptor.
1028 * @param[in] fd file descriptor to unlock
1029 * @param[in] offset offset to unlock
1032 _VUnlockFd(FD_t fd, afs_uint32 offset)
1034 struct afs_st_flock sf;
1036 sf.l_start = offset;
1038 sf.l_type = F_UNLCK;
1039 sf.l_whence = SEEK_SET;
1041 if (fcntl(fd, AFS_SETLK, &sf)) {
1042 Log("_VUnlockFd: fcntl failed with error %d when trying to unlock "
1043 "fd %d\n", errno, fd);
1046 #endif /* !AFS_NT40_ENV */
1049 * reinitialize a struct VLockFile.
1051 * Use this to close the lock file (unlocking any locks in it), and effectively
1052 * restore lf to the state it was in when it was initialized. This is the same
1053 * as unlocking all of the locks on the file, without having to remember what
1054 * all of the locks were. Do not unlock previously held locks after calling
1057 * @param[in] lf struct VLockFile to reinit
1059 * @pre nobody is waiting for a lock on this lockfile or otherwise using
1060 * this lockfile at all
1063 VLockFileReinit(struct VLockFile *lf)
1065 MUTEX_ENTER(&lf->mutex);
1067 if (lf->fd != INVALID_FD) {
1069 lf->fd = INVALID_FD;
1074 MUTEX_EXIT(&lf->mutex);
1078 * lock a file on disk for the process.
1080 * @param[in] lf the struct VLockFile representing the file to lock
1081 * @param[in] offset the offset in the file to lock
1082 * @param[in] locktype READ_LOCK or WRITE_LOCK
1083 * @param[in] nonblock 0 to wait for conflicting locks to clear before
1084 * obtaining the lock; 1 to fail immediately if a
1085 * conflicting lock is held by someone else
1087 * @return operation status
1089 * @retval EBUSY someone else is holding a conflicting lock and nonblock=1 was
1091 * @retval EIO error acquiring file lock
1095 * @note do not try to lock/unlock the same offset in the same file from
1096 * different threads; use VGetDiskLock to protect threads from each other in
1097 * addition to other processes
1100 VLockFileLock(struct VLockFile *lf, afs_uint32 offset, int locktype, int nonblock)
1104 osi_Assert(locktype == READ_LOCK || locktype == WRITE_LOCK);
1106 MUTEX_ENTER(&lf->mutex);
1108 if (lf->fd == INVALID_FD) {
1109 lf->fd = _VOpenPath(lf->path);
1110 if (lf->fd == INVALID_FD) {
1111 MUTEX_EXIT(&lf->mutex);
1118 MUTEX_EXIT(&lf->mutex);
1120 code = _VLockFd(lf->fd, offset, locktype, nonblock);
1123 MUTEX_ENTER(&lf->mutex);
1124 if (--lf->refcount < 1) {
1126 lf->fd = INVALID_FD;
1128 MUTEX_EXIT(&lf->mutex);
1135 VLockFileUnlock(struct VLockFile *lf, afs_uint32 offset)
1137 MUTEX_ENTER(&lf->mutex);
1139 osi_Assert(lf->fd != INVALID_FD);
1141 if (--lf->refcount < 1) {
1143 lf->fd = INVALID_FD;
1145 _VUnlockFd(lf->fd, offset);
1148 MUTEX_EXIT(&lf->mutex);
1151 #ifdef AFS_DEMAND_ATTACH_FS
1154 * initialize a struct VDiskLock.
1156 * @param[in] dl struct VDiskLock to initialize
1157 * @param[in] lf the struct VLockFile to associate with this disk lock
1160 VDiskLockInit(struct VDiskLock *dl, struct VLockFile *lf, afs_uint32 offset)
1163 memset(dl, 0, sizeof(*dl));
1164 Lock_Init(&dl->rwlock);
1165 MUTEX_INIT(&dl->mutex, "disklock", MUTEX_DEFAULT, 0);
1166 CV_INIT(&dl->cv, "disklock cv", CV_DEFAULT, 0);
1168 dl->offset = offset;
1172 * acquire a lock on a file on local disk.
1174 * @param[in] dl the VDiskLock structure corresponding to the file on disk
1175 * @param[in] locktype READ_LOCK if you want a read lock, or WRITE_LOCK if
1176 * you want a write lock
1177 * @param[in] nonblock 0 to wait for conflicting locks to clear before
1178 * obtaining the lock; 1 to fail immediately if a
1179 * conflicting lock is held by someone else
1181 * @return operation status
1183 * @retval EBUSY someone else is holding a conflicting lock and nonblock=1 was
1185 * @retval EIO error acquiring file lock
1189 * @note while normal fcntl-y locks on Unix systems generally only work per-
1190 * process, this interface also deals with locks between threads in the
1191 * process in addition to different processes acquiring the lock
1194 VGetDiskLock(struct VDiskLock *dl, int locktype, int nonblock)
1197 osi_Assert(locktype == READ_LOCK || locktype == WRITE_LOCK);
1200 if (locktype == READ_LOCK) {
1201 ObtainReadLockNoBlock(&dl->rwlock, code);
1203 ObtainWriteLockNoBlock(&dl->rwlock, code);
1210 } else if (locktype == READ_LOCK) {
1211 ObtainReadLock(&dl->rwlock);
1213 ObtainWriteLock(&dl->rwlock);
1216 MUTEX_ENTER(&dl->mutex);
1218 if ((dl->flags & VDISKLOCK_ACQUIRING)) {
1219 /* Some other thread is waiting to acquire an fs lock. If nonblock=1,
1220 * we can return immediately, since we know we'll need to wait to
1221 * acquire. Otherwise, wait for the other thread to finish acquiring
1226 while ((dl->flags & VDISKLOCK_ACQUIRING)) {
1227 CV_WAIT(&dl->cv, &dl->mutex);
1232 if (code == 0 && !(dl->flags & VDISKLOCK_ACQUIRED)) {
1233 /* no other thread holds the lock on the actual file; so grab one */
1235 /* first try, don't block on the lock to see if we can get it without
1237 code = VLockFileLock(dl->lockfile, dl->offset, locktype, 1);
1239 if (code == EBUSY && !nonblock) {
1241 /* mark that we are waiting on the fs lock */
1242 dl->flags |= VDISKLOCK_ACQUIRING;
1244 MUTEX_EXIT(&dl->mutex);
1245 code = VLockFileLock(dl->lockfile, dl->offset, locktype, nonblock);
1246 MUTEX_ENTER(&dl->mutex);
1248 dl->flags &= ~VDISKLOCK_ACQUIRING;
1251 dl->flags |= VDISKLOCK_ACQUIRED;
1254 CV_BROADCAST(&dl->cv);
1259 if (locktype == READ_LOCK) {
1260 ReleaseReadLock(&dl->rwlock);
1262 ReleaseWriteLock(&dl->rwlock);
1265 /* successfully got the lock, so inc the number of unlocks we need
1266 * to do before we can unlock the actual file */
1270 MUTEX_EXIT(&dl->mutex);
1276 * release a lock on a file on local disk.
1278 * @param[in] dl the struct VDiskLock to release
1279 * @param[in] locktype READ_LOCK if you are unlocking a read lock, or
1280 * WRITE_LOCK if you are unlocking a write lock
1282 * @return operation status
1286 VReleaseDiskLock(struct VDiskLock *dl, int locktype)
1288 osi_Assert(locktype == READ_LOCK || locktype == WRITE_LOCK);
1290 MUTEX_ENTER(&dl->mutex);
1291 osi_Assert(dl->lockers > 0);
1293 if (--dl->lockers < 1) {
1294 /* no threads are holding this lock anymore, so we can release the
1295 * actual disk lock */
1296 VLockFileUnlock(dl->lockfile, dl->offset);
1297 dl->flags &= ~VDISKLOCK_ACQUIRED;
1300 MUTEX_EXIT(&dl->mutex);
1302 if (locktype == READ_LOCK) {
1303 ReleaseReadLock(&dl->rwlock);
1305 ReleaseWriteLock(&dl->rwlock);
1309 #endif /* AFS_DEMAND_ATTACH_FS */