/*
* Copyright 2000, International Business Machines Corporation and others.
* All Rights Reserved.
- *
+ *
* This software has been released under the terms of the IBM Public
* License. For details, see the LICENSE file in the top-level source
* directory or online at http://www.openafs.org/dl/license10.html
#include <afsconfig.h>
#include <afs/param.h>
+#include <roken.h>
+#include <afs/opr.h>
-#include <stdio.h>
-#include <sys/types.h>
-#include <errno.h>
-#ifdef AFS_NT40_ENV
-#include <time.h>
-#include <fcntl.h>
-#else
-#include <sys/time.h>
+#ifdef HAVE_SYS_FILE_H
#include <sys/file.h>
-#include <unistd.h>
#endif
-#include <sys/stat.h>
+
+#ifdef HAVE_SYS_LOCKF_H
+#include <sys/lockf.h>
+#endif
+
#ifdef AFS_PTHREAD_ENV
-#include <assert.h>
-#else /* AFS_PTHREAD_ENV */
-#include <afs/assert.h>
-#endif /* AFS_PTHREAD_ENV */
+# include <opr/lock.h>
+#else
+# include <opr/lockstub.h>
+#endif
+#include <rx/rx_queue.h>
#include <rx/xdr.h>
#include <afs/afsint.h>
#include "nfs.h"
#include <afs/afsutil.h>
#ifdef AFS_NT40_ENV
#include "ntops.h"
-#include <io.h>
#endif
#include "vnode.h"
#include "volume.h"
+#include "volume_inline.h"
#include "partition.h"
#include "viceinode.h"
#include "volinodes.h"
#include "vol_prototypes.h"
+#include "common.h"
-#ifdef AFS_AIX_ENV
-#include <sys/lockf.h>
-#endif
-#if defined(AFS_SUN5_ENV) || defined(AFS_NT40_ENV) || defined(AFS_LINUX20_ENV)
-#include <string.h>
-#else
-#include <strings.h>
+#ifndef AFS_NT40_ENV
+# ifdef O_LARGEFILE
+# define AFS_SETLKW F_SETLKW64
+# define AFS_SETLK F_SETLK64
+# define afs_st_flock flock64
+# else
+# define AFS_SETLKW F_SETLKW
+# define AFS_SETLK F_SETLK
+# define afs_st_flock flock
+# endif
#endif
-#ifdef O_LARGEFILE
-#define afs_open open64
-#else /* !O_LARGEFILE */
-#define afs_open open
-#endif /* !O_LARGEFILE */
-
-/*@printflike@*/ extern void Log(const char *format, ...);
-
-#define nFILES (sizeof (stuff)/sizeof(struct stuff))
-
/* Note: the volume creation functions herein leave the destroyMe flag in the
volume header ON: this means that the volumes will not be attached by the
file server and WILL BE DESTROYED the next time a system salvage is performed */
+#ifdef FSSYNC_BUILD_CLIENT
static void
-RemoveInodes(Device dev, VolumeId vid)
+RemoveInodes(struct afs_inode_info *stuff, Device dev, VolumeId parent,
+ VolumeId vid)
{
- register int i;
+ int i;
IHandle_t *handle;
/* This relies on the fact that IDEC only needs the device and NT only
* needs the dev and vid to decrement volume special files.
*/
- IH_INIT(handle, dev, vid, -1);
- for (i = 0; i < nFILES; i++) {
+ IH_INIT(handle, dev, parent, -1);
+ for (i = 0; i < MAXINODETYPE; i++) {
Inode inode = *stuff[i].inode;
- if (VALID_INO(inode))
- IH_DEC(handle, inode, vid);
+ if (VALID_INO(inode)) {
+ if (stuff[i].inodeType == VI_LINKTABLE) {
+ IH_DEC(handle, inode, parent);
+ } else {
+ IH_DEC(handle, inode, vid);
+ }
+ }
}
IH_RELEASE(handle);
}
Volume *
-VCreateVolume(Error * ec, char *partname, VolId volumeId, VolId parentId)
+VCreateVolume(Error * ec, char *partname, VolumeId volumeId, VolumeId parentId)
{ /* Should be the same as volumeId if there is
* no parent */
Volume *retVal;
}
Volume *
-VCreateVolume_r(Error * ec, char *partname, VolId volumeId, VolId parentId)
+VCreateVolume_r(Error * ec, char *partname, VolumeId volumeId, VolumeId parentId)
{ /* Should be the same as volumeId if there is
* no parent */
VolumeDiskData vol;
- int fd, i;
- char headerName[32], volumePath[64];
+ int i, rc;
+ char headerName[VMAXPATHLEN], volumePath[VMAXPATHLEN + 1];
Device device;
struct DiskPartition64 *partition;
struct VolumeDiskHeader diskHeader;
IHandle_t *handle;
FdHandle_t *fdP;
- Inode nearInode = 0;
+ Inode nearInode AFS_UNUSED = 0;
char *part, *name;
+ struct stat st;
+ struct VolumeHeader tempHeader;
+ struct afs_inode_info stuff[MAXINODETYPE];
+ afs_ino_str_t stmp;
+# ifdef AFS_DEMAND_ATTACH_FS
+ int locktype = 0;
+# endif /* AFS_DEMAND_ATTACH_FS */
+
+ init_inode_info(&tempHeader, stuff);
*ec = 0;
memset(&vol, 0, sizeof(vol));
if (*ec == VNOVOL || !strcmp(partition->name, part)) {
/* this case is ok */
} else {
- /* return EXDEV if it's a clone to an alternate partition
+ /* return EXDEV if it's a clone or read-only to an alternate partition
* otherwise assume it's a move */
if (vol.parentId != vol.id) {
+ Log("VCreateVolume: volume %" AFS_VOLID_FMT " for parent %" AFS_VOLID_FMT
+ " found on %s; unable to create volume on %s.\n",
+ afs_printable_VolumeId_lu(vol.id),
+ afs_printable_VolumeId_lu(vol.parentId), part, partition->name);
*ec = EXDEV;
return NULL;
}
}
+ *ec = 0;
+
+# ifdef AFS_DEMAND_ATTACH_FS
+ /* volume doesn't exist yet, but we must lock it to try to prevent something
+ * else from reading it when we're e.g. half way through creating it (or
+ * something tries to create the same volume at the same time) */
+ locktype = VVolLockType(V_VOLUPD, 1);
+ rc = VLockVolumeByIdNB(volumeId, partition, locktype);
+ if (rc) {
+ Log("VCreateVolume: vol %lu already locked by someone else\n",
+ afs_printable_uint32_lu(volumeId));
+ *ec = VNOVOL;
+ return NULL;
+ }
+# else /* AFS_DEMAND_ATTACH_FS */
VLockPartition_r(partname);
+# endif /* !AFS_DEMAND_ATTACH_FS */
+
memset(&tempHeader, 0, sizeof(tempHeader));
tempHeader.stamp.magic = VOLUMEHEADERMAGIC;
tempHeader.stamp.version = VOLUMEHEADERVERSION;
vol.stamp.magic = VOLUMEINFOMAGIC;
vol.stamp.version = VOLUMEINFOVERSION;
vol.destroyMe = DESTROY_ME;
- (void)afs_snprintf(headerName, sizeof headerName, VFORMAT, afs_printable_uint32_lu(vol.id));
- (void)afs_snprintf(volumePath, sizeof volumePath, "%s/%s",
- VPartitionPath(partition), headerName);
- fd = afs_open(volumePath, O_CREAT | O_EXCL | O_WRONLY, 0600);
- if (fd == -1) {
- if (errno == EEXIST) {
+ snprintf(headerName, sizeof headerName, VFORMAT,
+ afs_printable_VolumeId_lu(vol.id));
+ snprintf(volumePath, sizeof volumePath, "%s" OS_DIRSEP "%s",
+ VPartitionPath(partition), headerName);
+ rc = stat(volumePath, &st);
+ if (rc == 0 || errno != ENOENT) {
+ if (rc == 0) {
Log("VCreateVolume: Header file %s already exists!\n",
volumePath);
*ec = VVOLEXISTS;
} else {
- Log("VCreateVolume: Couldn't create header file %s for volume %u\n", volumePath, vol.id);
+ Log("VCreateVolume: Error %d trying to stat header file %s\n",
+ errno, volumePath);
*ec = VNOVOL;
}
- return NULL;
+ goto bad_noheader;
}
device = partition->device;
- for (i = 0; i < nFILES; i++) {
- register struct stuff *p = &stuff[i];
+ for (i = 0; i < MAXINODETYPE; i++) {
+ struct afs_inode_info *p = &stuff[i];
if (p->obsolete)
continue;
#ifdef AFS_NAMEI_ENV
(p->inodeType == VI_LINKTABLE) ? vol.parentId : vol.id,
INODESPECIAL, p->inodeType, vol.parentId);
if (!(VALID_INO(*(p->inode)))) {
- if (errno == EEXIST) {
+ if (errno == EEXIST && (p->inodeType == VI_LINKTABLE)) {
/* Increment the reference count instead. */
IHandle_t *lh;
int code;
-#ifdef AFS_NT40_ENV
- *(p->inode) = nt_MakeSpecIno(VI_LINKTABLE);
-#else
*(p->inode) = namei_MakeSpecIno(vol.parentId, VI_LINKTABLE);
-#endif
IH_INIT(lh, device, parentId, *(p->inode));
fdP = IH_OPEN(lh);
if (fdP == NULL) {
bad:
if (handle)
IH_RELEASE(handle);
- RemoveInodes(device, vol.id);
- *ec = VNOVOL;
- close(fd);
+ RemoveInodes(stuff, device, vol.parentId, vol.id);
+ if (!*ec) {
+ *ec = VNOVOL;
+ }
+ VDestroyVolumeDiskHeader(partition, volumeId, parentId);
+ bad_noheader:
+# ifdef AFS_DEMAND_ATTACH_FS
+ if (locktype) {
+ VUnlockVolumeById(volumeId, partition);
+ }
+# endif /* AFS_DEMAND_ATTACH_FS */
return NULL;
}
IH_INIT(handle, device, vol.parentId, *(p->inode));
fdP = IH_OPEN(handle);
if (fdP == NULL) {
Log("VCreateVolume: Problem iopen inode %s (err=%d)\n",
- PrintInode(NULL, *(p->inode)), errno);
+ PrintInode(stmp, *(p->inode)), errno);
goto bad;
}
- if (FDH_SEEK(fdP, 0, SEEK_SET) < 0) {
- Log("VCreateVolume: Problem lseek inode %s (err=%d)\n",
- PrintInode(NULL, *(p->inode)), errno);
- FDH_REALLYCLOSE(fdP);
- goto bad;
- }
- if (FDH_WRITE(fdP, (char *)&p->stamp, sizeof(p->stamp)) !=
+ if (FDH_PWRITE(fdP, (char *)&p->stamp, sizeof(p->stamp), 0) !=
sizeof(p->stamp)) {
- Log("VCreateVolume: Problem writing to inode %s (err=%d)\n",
- PrintInode(NULL, *(p->inode)), errno);
+ Log("VCreateVolume: Problem writing to inode %s (err=%d)\n",
+ PrintInode(stmp, *(p->inode)), errno);
FDH_REALLYCLOSE(fdP);
goto bad;
}
fdP = IH_OPEN(handle);
if (fdP == NULL) {
Log("VCreateVolume: Problem iopen inode %s (err=%d)\n",
- PrintInode(NULL, tempHeader.volumeInfo), errno);
- unlink(volumePath);
- goto bad;
- }
- if (FDH_SEEK(fdP, 0, SEEK_SET) < 0) {
- Log("VCreateVolume: Problem lseek inode %s (err=%d)\n",
- PrintInode(NULL, tempHeader.volumeInfo), errno);
- FDH_REALLYCLOSE(fdP);
- unlink(volumePath);
+ PrintInode(stmp, tempHeader.volumeInfo), errno);
goto bad;
}
- if (FDH_WRITE(fdP, (char *)&vol, sizeof(vol)) != sizeof(vol)) {
+ if (FDH_PWRITE(fdP, (char *)&vol, sizeof(vol), 0) != sizeof(vol)) {
Log("VCreateVolume: Problem writing to inode %s (err=%d)\n",
- PrintInode(NULL, tempHeader.volumeInfo), errno);
+ PrintInode(stmp, tempHeader.volumeInfo), errno);
FDH_REALLYCLOSE(fdP);
- unlink(volumePath);
goto bad;
}
FDH_CLOSE(fdP);
IH_RELEASE(handle);
VolumeHeaderToDisk(&diskHeader, &tempHeader);
- if (write(fd, &diskHeader, sizeof(diskHeader)) != sizeof(diskHeader)) {
- Log("VCreateVolume: Unable to write volume header %s; volume %u not created\n", volumePath, vol.id);
- unlink(volumePath);
+ rc = VCreateVolumeDiskHeader(&diskHeader, partition);
+ if (rc) {
+ Log("VCreateVolume: Error %d trying to write volume header for "
+ "volume %" AFS_VOLID_FMT " on partition %s; volume not created\n", rc,
+ afs_printable_VolumeId_lu(vol.id), VPartitionPath(partition));
+ if (rc == EEXIST) {
+ *ec = VVOLEXISTS;
+ }
goto bad;
}
- fsync(fd);
- close(fd);
+
+# ifdef AFS_DEMAND_ATTACH_FS
+ if (locktype) {
+ VUnlockVolumeById(volumeId, partition);
+ }
+# endif /* AFS_DEMAND_ATTACH_FS */
return (VAttachVolumeByName_r(ec, partname, headerName, V_SECRETLY));
}
+#endif /* FSSYNC_BUILD_CLIENT */
void
-AssignVolumeName(register VolumeDiskData * vol, char *name, char *ext)
+AssignVolumeName(VolumeDiskData * vol, char *name, char *ext)
{
VOL_LOCK;
AssignVolumeName_r(vol, name, ext);
}
void
-AssignVolumeName_r(register VolumeDiskData * vol, char *name, char *ext)
+AssignVolumeName_r(VolumeDiskData * vol, char *name, char *ext)
{
- register char *dot;
+ char *dot;
strncpy(vol->name, name, VNAMESIZE - 1);
vol->name[VNAMESIZE - 1] = '\0';
dot = strrchr(vol->name, '.');
}
void
-ClearVolumeStats(register VolumeDiskData * vol)
+ClearVolumeStats(VolumeDiskData * vol)
{
VOL_LOCK;
ClearVolumeStats_r(vol);
}
void
-ClearVolumeStats_r(register VolumeDiskData * vol)
+ClearVolumeStats_r(VolumeDiskData * vol)
{
memset(vol->weekUse, 0, sizeof(vol->weekUse));
vol->dayUse = 0;
vol->dayUseDate = 0;
}
+
+void
+CopyVolumeStats_r(VolumeDiskData * from, VolumeDiskData * to)
+{
+ memcpy(to->weekUse, from->weekUse, sizeof(to->weekUse));
+ to->dayUse = from->dayUse;
+ to->dayUseDate = from->dayUseDate;
+ if (from->stat_initialized) {
+ memcpy(to->stat_reads, from->stat_reads, sizeof(to->stat_reads));
+ memcpy(to->stat_writes, from->stat_writes, sizeof(to->stat_writes));
+ memcpy(to->stat_fileSameAuthor, from->stat_fileSameAuthor,
+ sizeof(to->stat_fileSameAuthor));
+ memcpy(to->stat_fileDiffAuthor, from->stat_fileDiffAuthor,
+ sizeof(to->stat_fileDiffAuthor));
+ memcpy(to->stat_dirSameAuthor, from->stat_dirSameAuthor,
+ sizeof(to->stat_dirSameAuthor));
+ memcpy(to->stat_dirDiffAuthor, from->stat_dirDiffAuthor,
+ sizeof(to->stat_dirDiffAuthor));
+ }
+}
+
+void
+CopyVolumeStats(VolumeDiskData * from, VolumeDiskData * to)
+{
+ VOL_LOCK;
+ CopyVolumeStats_r(from, to);
+ VOL_UNLOCK;
+}
+
+/**
+ * read an existing volume disk header.
+ *
+ * @param[in] volid volume id
+ * @param[in] dp disk partition object
+ * @param[out] hdr volume disk header or NULL
+ *
+ * @note if hdr is NULL, this is essentially an existence test for the vol
+ * header
+ *
+ * @return operation status
+ * @retval 0 success
+ * @retval -1 volume header doesn't exist
+ * @retval EIO failed to read volume header
+ *
+ * @internal
+ */
+afs_int32
+VReadVolumeDiskHeader(VolumeId volid,
+ struct DiskPartition64 * dp,
+ VolumeDiskHeader_t * hdr)
+{
+ afs_int32 code = 0;
+ int fd;
+ char path[MAXPATHLEN];
+
+ snprintf(path, sizeof(path), "%s" OS_DIRSEP VFORMAT,
+ VPartitionPath(dp), afs_printable_VolumeId_lu(volid));
+ fd = open(path, O_RDONLY);
+ if (fd < 0) {
+ Log("VReadVolumeDiskHeader: Couldn't open header for volume %" AFS_VOLID_FMT " (errno %d).\n",
+ afs_printable_VolumeId_lu(volid), errno);
+ code = -1;
+
+ } else if (hdr && read(fd, hdr, sizeof(*hdr)) != sizeof(*hdr)) {
+ Log("VReadVolumeDiskHeader: Couldn't read header for volume %" AFS_VOLID_FMT ".\n",
+ afs_printable_VolumeId_lu(volid));
+ code = EIO;
+ }
+
+ if (fd >= 0) {
+ close(fd);
+ }
+ return code;
+}
+
+#ifdef FSSYNC_BUILD_CLIENT
+/**
+ * write an existing volume disk header.
+ *
+ * @param[in] hdr volume disk header
+ * @param[in] dp disk partition object
+ * @param[in] cr assert if O_CREAT | O_EXCL should be passed to open()
+ *
+ * @return operation status
+ * @retval 0 success
+ * @retval -1 volume header doesn't exist
+ * @retval EIO failed to write volume header
+ *
+ * @internal
+ */
+static afs_int32
+_VWriteVolumeDiskHeader(VolumeDiskHeader_t * hdr,
+ struct DiskPartition64 * dp,
+ int flags)
+{
+ afs_int32 code = 0;
+ int fd;
+ char path[MAXPATHLEN];
+
+#ifdef AFS_DEMAND_ATTACH_FS
+ /* prevent racing with VGC scanners reading the vol header while we are
+ * writing it */
+ code = VPartHeaderLock(dp, READ_LOCK);
+ if (code) {
+ return EIO;
+ }
+#endif /* AFS_DEMAND_ATTACH_FS */
+
+ flags |= O_RDWR;
+
+ snprintf(path, sizeof(path), "%s" OS_DIRSEP VFORMAT,
+ VPartitionPath(dp), afs_printable_VolumeId_lu(hdr->id));
+ fd = open(path, flags, 0644);
+ if (fd < 0) {
+ code = errno;
+ Log("_VWriteVolumeDiskHeader: Couldn't open header for volume %lu, "
+ "error = %d\n", afs_printable_uint32_lu(hdr->id), errno);
+ } else if (write(fd, hdr, sizeof(*hdr)) != sizeof(*hdr)) {
+ Log("_VWriteVolumeDiskHeader: Couldn't write header for volume %lu, "
+ "error = %d\n", afs_printable_uint32_lu(hdr->id), errno);
+ code = EIO;
+ }
+
+ if (fd >= 0) {
+ if (close(fd) != 0) {
+ Log("_VWriteVolumeDiskHeader: Error closing header for volume "
+ "%lu, errno %d\n", afs_printable_uint32_lu(hdr->id), errno);
+ }
+ }
+
+#ifdef AFS_DEMAND_ATTACH_FS
+ VPartHeaderUnlock(dp, READ_LOCK);
+#endif /* AFS_DEMAND_ATTACH_FS */
+
+ return code;
+}
+
+/**
+ * write an existing volume disk header.
+ *
+ * @param[in] hdr volume disk header
+ * @param[in] dp disk partition object
+ *
+ * @return operation status
+ * @retval 0 success
+ * @retval ENOENT volume header doesn't exist
+ * @retval EIO failed to write volume header
+ */
+afs_int32
+VWriteVolumeDiskHeader(VolumeDiskHeader_t * hdr,
+ struct DiskPartition64 * dp)
+{
+ afs_int32 code;
+
+#ifdef AFS_DEMAND_ATTACH_FS
+ VolumeDiskHeader_t oldhdr;
+ int delvgc = 0, addvgc = 0;
+ SYNC_response res;
+
+ /* first, see if anything with the volume IDs have changed; if so, we
+ * need to update the VGC */
+
+ code = VReadVolumeDiskHeader(hdr->id, dp, &oldhdr);
+ if (code == 0 && (oldhdr.id != hdr->id || oldhdr.parent != hdr->parent)) {
+ /* the vol id or parent vol id changed; need to delete the VGC entry
+ * for the old vol id/parent, and add the new one */
+ delvgc = 1;
+ addvgc = 1;
+
+ } else if (code) {
+ /* couldn't get the old header info; add the new header info to the
+ * VGC in case it hasn't been added yet */
+ addvgc = 1;
+ }
+
+#endif /* AFS_DEMAND_ATTACH_FS */
+
+ code = _VWriteVolumeDiskHeader(hdr, dp, 0);
+ if (code) {
+ goto done;
+ }
+
+#ifdef AFS_DEMAND_ATTACH_FS
+ if (delvgc) {
+ memset(&res, 0, sizeof(res));
+ code = FSYNC_VGCDel(dp->name, oldhdr.parent, oldhdr.id, FSYNC_WHATEVER, &res);
+
+ /* unknown vol id is okay; it just further suggests the old header
+ * data was bogus, which is fine since we're trying to fix it */
+ if (code && res.hdr.reason != FSYNC_UNKNOWN_VOLID) {
+ Log("VWriteVolumeDiskHeader: FSYNC_VGCDel(%s, %lu, %lu) "
+ "failed with code %ld reason %ld\n", dp->name,
+ afs_printable_uint32_lu(oldhdr.parent),
+ afs_printable_uint32_lu(oldhdr.id),
+ afs_printable_int32_ld(code),
+ afs_printable_int32_ld(res.hdr.reason));
+ }
+
+ }
+ if (addvgc) {
+ memset(&res, 0, sizeof(res));
+ code = FSYNC_VGCAdd(dp->name, hdr->parent, hdr->id, FSYNC_WHATEVER, &res);
+ if (code) {
+ Log("VWriteVolumeDiskHeader: FSYNC_VGCAdd(%s, %lu, %lu) "
+ "failed with code %ld reason %ld\n", dp->name,
+ afs_printable_uint32_lu(hdr->parent),
+ afs_printable_uint32_lu(hdr->id),
+ afs_printable_int32_ld(code),
+ afs_printable_int32_ld(res.hdr.reason));
+ }
+ }
+
+#endif /* AFS_DEMAND_ATTACH_FS */
+
+ done:
+ return code;
+}
+
+/**
+ * create and write a volume disk header to disk.
+ *
+ * @param[in] hdr volume disk header
+ * @param[in] dp disk partition object
+ *
+ * @return operation status
+ * @retval 0 success
+ * @retval EEXIST volume header already exists
+ * @retval EIO failed to write volume header
+ *
+ * @internal
+ */
+afs_int32
+VCreateVolumeDiskHeader(VolumeDiskHeader_t * hdr,
+ struct DiskPartition64 * dp)
+{
+ afs_int32 code = 0;
+#ifdef AFS_DEMAND_ATTACH_FS
+ SYNC_response res;
+#endif /* AFS_DEMAND_ATTACH_FS */
+
+ code = _VWriteVolumeDiskHeader(hdr, dp, O_CREAT | O_EXCL);
+ if (code) {
+ goto done;
+ }
+
+#ifdef AFS_DEMAND_ATTACH_FS
+ memset(&res, 0, sizeof(res));
+ code = FSYNC_VGCAdd(dp->name, hdr->parent, hdr->id, FSYNC_WHATEVER, &res);
+ if (code) {
+ Log("VCreateVolumeDiskHeader: FSYNC_VGCAdd(%s, %lu, %lu) failed "
+ "with code %ld reason %ld\n", dp->name,
+ afs_printable_uint32_lu(hdr->parent),
+ afs_printable_uint32_lu(hdr->id),
+ afs_printable_int32_ld(code),
+ afs_printable_int32_ld(res.hdr.reason));
+ }
+#endif /* AFS_DEMAND_ATTACH_FS */
+
+ done:
+ return code;
+}
+
+
+/**
+ * destroy a volume disk header.
+ *
+ * @param[in] dp disk partition object
+ * @param[in] volid volume id
+ * @param[in] parent parent's volume id, 0 if unknown
+ *
+ * @return operation status
+ * @retval 0 success
+ *
+ * @note for non-DAFS, parent is currently ignored
+ */
+afs_int32
+VDestroyVolumeDiskHeader(struct DiskPartition64 * dp,
+ VolumeId volid,
+ VolumeId parent)
+{
+ afs_int32 code = 0;
+ char path[MAXPATHLEN];
+#ifdef AFS_DEMAND_ATTACH_FS
+ SYNC_response res;
+#endif /* AFS_DEMAND_ATTACH_FS */
+
+ snprintf(path, sizeof(path), "%s" OS_DIRSEP VFORMAT,
+ VPartitionPath(dp), afs_printable_VolumeId_lu(volid));
+ code = unlink(path);
+ if (code) {
+ Log("VDestroyVolumeDiskHeader: Couldn't unlink disk header, error = %d\n", errno);
+ goto done;
+ }
+
+#ifdef AFS_DEMAND_ATTACH_FS
+ /* Remove the volume entry from the fileserver's volume group cache, if found. */
+ memset(&res, 0, sizeof(res));
+ code = FSYNC_VGCDel(dp->name, parent, volid, FSYNC_WHATEVER, &res);
+ if (code) {
+ Log("VDestroyVolumeDiskHeader: FSYNC_VGCDel(%s, %" AFS_VOLID_FMT ", %" AFS_VOLID_FMT ") failed "
+ "with code %ld reason %ld\n", dp->name,
+ afs_printable_VolumeId_lu(parent),
+ afs_printable_VolumeId_lu(volid),
+ afs_printable_int32_ld(code),
+ afs_printable_int32_ld(res.hdr.reason));
+ }
+#endif /* AFS_DEMAND_ATTACH_FS */
+
+ done:
+ return code;
+}
+#endif /* FSSYNC_BUILD_CLIENT */
+
+/**
+ * handle a single vol header as part of VWalkVolumeHeaders.
+ *
+ * @param[in] dp disk partition
+ * @param[in] volfunc function to call when a vol header is successfully read
+ * @param[in] name full path name to the .vol header
+ * @param[out] hdr header data read in from the .vol header
+ * @param[in] locked 1 if the partition headers are locked, 0 otherwise
+ * @param[in] rock the rock to pass to volfunc
+ *
+ * @return operation status
+ * @retval 0 success
+ * @retval -1 fatal error, stop scanning
+ * @retval 1 failed to read header
+ * @retval 2 volfunc callback indicated error after header read
+ */
+static int
+_VHandleVolumeHeader(struct DiskPartition64 *dp, VWalkVolFunc volfunc,
+ const char *name, struct VolumeDiskHeader *hdr,
+ int locked, void *rock)
+{
+ int error = 0;
+ FD_t fd;
+
+ if ((fd = OS_OPEN(name, O_RDONLY, 0)) == INVALID_FD
+ || OS_READ(fd, hdr, sizeof(*hdr))
+ != sizeof(*hdr)
+ || hdr->stamp.magic != VOLUMEHEADERMAGIC) {
+ error = 1;
+ }
+
+ if (fd != INVALID_FD) {
+ OS_CLOSE(fd);
+ }
+
+#ifdef AFSFS_DEMAND_ATTACH_FS
+ if (locked) {
+ VPartHeaderUnlock(dp);
+ }
+#endif /* AFS_DEMAND_ATTACH_FS */
+
+ if (!error && volfunc) {
+ /* the volume header seems fine; call the caller-supplied
+ * 'we-found-a-volume-header' function */
+ int last = 1;
+
+#ifdef AFS_DEMAND_ATTACH_FS
+ if (!locked) {
+ last = 0;
+ }
+#endif /* AFS_DEMAND_ATTACH_FS */
+
+ error = (*volfunc) (dp, name, hdr, last, rock);
+ if (error < 0) {
+ return -1;
+ }
+ if (error) {
+ error = 2;
+ }
+ }
+
+#ifdef AFS_DEMAND_ATTACH_FS
+ if (error && !locked) {
+ int code;
+ /* retry reading the volume header under the partition
+ * header lock, just to be safe and ensure we're not
+ * racing something rewriting the vol header */
+ code = VPartHeaderLock(dp, WRITE_LOCK);
+ if (code) {
+ Log("Error acquiring partition write lock when "
+ "looking at header %s\n", name);
+ return -1;
+ }
+
+ return _VHandleVolumeHeader(dp, volfunc, name, hdr, 1, rock);
+ }
+#endif /* AFS_DEMAND_ATTACH_FS */
+
+ return error;
+}
+
+/**
+ * walk through the list of volume headers on a partition.
+ *
+ * This function looks through all of the .vol headers on a partition, reads in
+ * each header, and calls the supplied volfunc function on each one. If the
+ * header cannot be read (or volfunc returns a positive error code), DAFS will
+ * VPartHeaderExLock() and retry. If that fails, or if we are non-DAFS, errfunc
+ * will be called (which typically will unlink the problem volume header).
+ *
+ * If volfunc returns a negative error code, walking the partition will stop
+ * and we will return an error immediately.
+ *
+ * @param[in] dp partition to walk
+ * @param[in] partpath the path opendir()
+ * @param[in] volfunc the function to call when a header is encountered, or
+ * NULL to just skip over valid headers
+ * @param[in] errfunc the function to call when a problematic header is
+ * encountered, or NULL to just skip over bad headers
+ * @param[in] rock rock for volfunc and errfunc
+ *
+ * @see VWalkVolFunc
+ * @see VWalkErrFunc
+ *
+ * @return operation status
+ * @retval 0 success
+ * @retval negative fatal error, walk did not finish
+ */
+int
+VWalkVolumeHeaders(struct DiskPartition64 *dp, const char *partpath,
+ VWalkVolFunc volfunc, VWalkErrFunc errfunc, void *rock)
+{
+ DIR *dirp;
+ struct dirent *dentry;
+ int code = 0;
+ struct VolumeDiskHeader diskHeader;
+
+ dirp = opendir(partpath);
+ if (!dirp) {
+ Log("VWalkVolumeHeaders: cannot open directory %s\n", partpath);
+ code = -1;
+ goto done;
+ }
+
+ while ((dentry = readdir(dirp)) != NULL) {
+ char *p;
+ p = strrchr(dentry->d_name, '.');
+ if (p != NULL && strcmp(p, VHDREXT) == 0) {
+ char name[VMAXPATHLEN];
+
+ snprintf(name, VMAXPATHLEN, "%s" OS_DIRSEP "%s", partpath, dentry->d_name);
+
+ code = _VHandleVolumeHeader(dp, volfunc, name, &diskHeader, -1, rock);
+ if (code < 0) {
+ /* fatal error, stop walking */
+ goto done;
+ }
+ if (code && errfunc) {
+ /* error with header; call the caller-supplied vol error
+ * function */
+
+ struct VolumeDiskHeader *hdr = &diskHeader;
+ if (code == 1) {
+ /* we failed to read the header at all, so don't pass in
+ * the header ptr */
+ hdr = NULL;
+ }
+ (*errfunc) (dp, name, hdr, rock);
+ }
+ code = 0;
+ }
+ }
+ done:
+ if (dirp) {
+ closedir(dirp);
+ dirp = NULL;
+ }
+
+ return code;
+}
+
+/**
+ * initialize a struct VLockFile.
+ *
+ * @param[in] lf struct VLockFile to initialize
+ * @param[in] path Full path to the file to use for locks. The string contents
+ * are copied.
+ */
+void
+VLockFileInit(struct VLockFile *lf, const char *path)
+{
+ memset(lf, 0, sizeof(*lf));
+ lf->path = strdup(path);
+ lf->fd = INVALID_FD;
+ opr_mutex_init(&lf->mutex);
+}
+
+#ifdef AFS_NT40_ENV
+static_inline FD_t
+_VOpenPath(const char *path)
+{
+ HANDLE handle;
+
+ handle = CreateFile(path,
+ GENERIC_READ | GENERIC_WRITE,
+ FILE_SHARE_READ | FILE_SHARE_WRITE,
+ NULL,
+ OPEN_ALWAYS,
+ FILE_ATTRIBUTE_HIDDEN,
+ NULL);
+ if (handle == INVALID_HANDLE_VALUE) {
+ return INVALID_FD;
+ }
+
+ return handle;
+}
+
+static_inline int
+_VLockFd(FD_t handle, afs_uint32 offset, int locktype, int nonblock)
+{
+ DWORD flags = 0;
+ OVERLAPPED lap;
+
+ if (locktype == WRITE_LOCK) {
+ flags |= LOCKFILE_EXCLUSIVE_LOCK;
+ }
+ if (nonblock) {
+ flags |= LOCKFILE_FAIL_IMMEDIATELY;
+ }
+
+ memset(&lap, 0, sizeof(lap));
+ lap.Offset = offset;
+
+ if (!LockFileEx(handle, flags, 0, 1, 0, &lap)) {
+ if (GetLastError() == ERROR_LOCK_VIOLATION) {
+ return EBUSY;
+ }
+ return EIO;
+ }
+
+ return 0;
+}
+
+static_inline void
+_VUnlockFd(FD_t handle, afs_uint32 offset)
+{
+ OVERLAPPED lap;
+
+ memset(&lap, 0, sizeof(lap));
+ lap.Offset = offset;
+
+ UnlockFileEx(handle, 0, 1, 0, &lap);
+}
+
+static_inline void
+_VCloseFd(FD_t handle)
+{
+ CloseHandle(handle);
+}
+
+#else /* !AFS_NT40_ENV */
+
+/**
+ * open a file on the local filesystem suitable for locking
+ *
+ * @param[in] path abs path of the file to open
+ *
+ * @return file descriptor
+ * @retval INVALID_FD failure opening file
+ */
+static_inline FD_t
+_VOpenPath(const char *path)
+{
+ int fd;
+
+ fd = open(path, O_RDWR | O_CREAT, 0660);
+ if (fd < 0) {
+ return INVALID_FD;
+ }
+ return fd;
+}
+
+/**
+ * lock an offset in a file descriptor.
+ *
+ * @param[in] fd file descriptor to lock
+ * @param[in] offset offset in file to lock
+ * @param[in] locktype READ_LOCK or WRITE_LOCK
+ * @param[in] nonblock 1 to fail immediately, 0 to wait to acquire lock
+ *
+ * @return operation status
+ * @retval 0 success
+ * @retval EBUSY someone else is holding a conflicting lock and nonblock=1 was
+ * specified
+ * @retval EIO error acquiring file lock
+ */
+static_inline int
+_VLockFd(FD_t fd, afs_uint32 offset, int locktype, int nonblock)
+{
+ int l_type = F_WRLCK;
+ int cmd = AFS_SETLKW;
+ struct afs_st_flock sf;
+
+ opr_Assert(fd >= 0);
+
+ if (locktype == READ_LOCK) {
+ l_type = F_RDLCK;
+ }
+ if (nonblock) {
+ cmd = AFS_SETLK;
+ }
+
+ sf.l_start = offset;
+ sf.l_len = 1;
+ sf.l_type = l_type;
+ sf.l_whence = SEEK_SET;
+
+ if (fcntl(fd, cmd, &sf)) {
+ if (nonblock && (errno == EACCES || errno == EAGAIN)) {
+ /* We asked for a nonblocking lock, and it was already locked */
+ sf.l_pid = 0;
+ if (fcntl(fd, F_GETLK, &sf) != 0 || sf.l_pid == 0) {
+ Log("_VLockFd: fcntl failed with error %d when trying to "
+ "query the conflicting lock for fd %d (locktype=%d, "
+ "offset=%lu)\n", errno, fd, locktype,
+ afs_printable_uint32_lu(offset));
+ } else {
+ Log("_VLockFd: conflicting lock held on fd %d, offset %lu by "
+ "pid %ld (locktype=%d)\n", fd,
+ afs_printable_uint32_lu(offset), (long int)sf.l_pid,
+ locktype);
+ }
+ return EBUSY;
+ }
+ Log("_VLockFd: fcntl failed with error %d when trying to lock "
+ "fd %d (locktype=%d, offset=%lu)\n", errno, fd, locktype,
+ afs_printable_uint32_lu(offset));
+ return EIO;
+ }
+
+ return 0;
+}
+
+/**
+ * close a file descriptor used for file locking.
+ *
+ * @param[in] fd file descriptor to close
+ */
+static_inline void
+_VCloseFd(FD_t fd)
+{
+ if (close(fd)) {
+ Log("_VCloseFd: error %d closing fd %d\n",
+ errno, fd);
+ }
+}
+
+/**
+ * unlock a file offset in a file descriptor.
+ *
+ * @param[in] fd file descriptor to unlock
+ * @param[in] offset offset to unlock
+ */
+static_inline void
+_VUnlockFd(FD_t fd, afs_uint32 offset)
+{
+ struct afs_st_flock sf;
+
+ sf.l_start = offset;
+ sf.l_len = 1;
+ sf.l_type = F_UNLCK;
+ sf.l_whence = SEEK_SET;
+
+ if (fcntl(fd, AFS_SETLK, &sf)) {
+ Log("_VUnlockFd: fcntl failed with error %d when trying to unlock "
+ "fd %d\n", errno, fd);
+ }
+}
+#endif /* !AFS_NT40_ENV */
+
+/**
+ * reinitialize a struct VLockFile.
+ *
+ * Use this to close the lock file (unlocking any locks in it), and effectively
+ * restore lf to the state it was in when it was initialized. This is the same
+ * as unlocking all of the locks on the file, without having to remember what
+ * all of the locks were. Do not unlock previously held locks after calling
+ * this.
+ *
+ * @param[in] lf struct VLockFile to reinit
+ *
+ * @pre nobody is waiting for a lock on this lockfile or otherwise using
+ * this lockfile at all
+ */
+void
+VLockFileReinit(struct VLockFile *lf)
+{
+ opr_mutex_enter(&lf->mutex);
+
+ if (lf->fd != INVALID_FD) {
+ _VCloseFd(lf->fd);
+ lf->fd = INVALID_FD;
+ }
+
+ lf->refcount = 0;
+
+ opr_mutex_exit(&lf->mutex);
+}
+
+/**
+ * lock a file on disk for the process.
+ *
+ * @param[in] lf the struct VLockFile representing the file to lock
+ * @param[in] offset the offset in the file to lock
+ * @param[in] locktype READ_LOCK or WRITE_LOCK
+ * @param[in] nonblock 0 to wait for conflicting locks to clear before
+ * obtaining the lock; 1 to fail immediately if a
+ * conflicting lock is held by someone else
+ *
+ * @return operation status
+ * @retval 0 success
+ * @retval EBUSY someone else is holding a conflicting lock and nonblock=1 was
+ * specified
+ * @retval EIO error acquiring file lock
+ *
+ * @note DAFS only
+ *
+ * @note do not try to lock/unlock the same offset in the same file from
+ * different threads; use VGetDiskLock to protect threads from each other in
+ * addition to other processes
+ */
+int
+VLockFileLock(struct VLockFile *lf, afs_uint32 offset, int locktype, int nonblock)
+{
+ int code;
+
+ opr_Assert(locktype == READ_LOCK || locktype == WRITE_LOCK);
+
+ opr_mutex_enter(&lf->mutex);
+
+ if (lf->fd == INVALID_FD) {
+ opr_Assert(lf->refcount == 0);
+ lf->fd = _VOpenPath(lf->path);
+ if (lf->fd == INVALID_FD) {
+ opr_mutex_exit(&lf->mutex);
+ return EIO;
+ }
+ }
+
+ lf->refcount++;
+
+ opr_Assert(lf->refcount > 0);
+
+ opr_mutex_exit(&lf->mutex);
+
+ code = _VLockFd(lf->fd, offset, locktype, nonblock);
+
+ if (code) {
+ opr_mutex_enter(&lf->mutex);
+ opr_Assert(lf->refcount > 0);
+ if (--lf->refcount < 1) {
+ _VCloseFd(lf->fd);
+ lf->fd = INVALID_FD;
+ }
+ opr_mutex_exit(&lf->mutex);
+ }
+
+ return code;
+}
+
+void
+VLockFileUnlock(struct VLockFile *lf, afs_uint32 offset)
+{
+ opr_mutex_enter(&lf->mutex);
+
+ opr_Assert(lf->fd != INVALID_FD);
+ opr_Assert(lf->refcount > 0);
+
+ if (--lf->refcount < 1) {
+ _VCloseFd(lf->fd);
+ lf->fd = INVALID_FD;
+ } else {
+ _VUnlockFd(lf->fd, offset);
+ }
+
+ opr_mutex_exit(&lf->mutex);
+}
+
+#ifdef AFS_DEMAND_ATTACH_FS
+
+/**
+ * initialize a struct VDiskLock.
+ *
+ * @param[in] dl struct VDiskLock to initialize
+ * @param[in] lf the struct VLockFile to associate with this disk lock
+ */
+void
+VDiskLockInit(struct VDiskLock *dl, struct VLockFile *lf, afs_uint32 offset)
+{
+ opr_Assert(lf);
+ memset(dl, 0, sizeof(*dl));
+ Lock_Init(&dl->rwlock);
+ opr_mutex_init(&dl->mutex);
+ opr_cv_init(&dl->cv);
+ dl->lockfile = lf;
+ dl->offset = offset;
+}
+
+/**
+ * acquire a lock on a file on local disk.
+ *
+ * @param[in] dl the VDiskLock structure corresponding to the file on disk
+ * @param[in] locktype READ_LOCK if you want a read lock, or WRITE_LOCK if
+ * you want a write lock
+ * @param[in] nonblock 0 to wait for conflicting locks to clear before
+ * obtaining the lock; 1 to fail immediately if a
+ * conflicting lock is held by someone else
+ *
+ * @return operation status
+ * @retval 0 success
+ * @retval EBUSY someone else is holding a conflicting lock and nonblock=1 was
+ * specified
+ * @retval EIO error acquiring file lock
+ *
+ * @note DAFS only
+ *
+ * @note while normal fcntl-y locks on Unix systems generally only work per-
+ * process, this interface also deals with locks between threads in the
+ * process in addition to different processes acquiring the lock
+ */
+int
+VGetDiskLock(struct VDiskLock *dl, int locktype, int nonblock)
+{
+ int code = 0;
+ opr_Assert(locktype == READ_LOCK || locktype == WRITE_LOCK);
+
+ if (nonblock) {
+ if (locktype == READ_LOCK) {
+ ObtainReadLockNoBlock(&dl->rwlock, code);
+ } else {
+ ObtainWriteLockNoBlock(&dl->rwlock, code);
+ }
+
+ if (code) {
+ return EBUSY;
+ }
+
+ } else if (locktype == READ_LOCK) {
+ ObtainReadLock(&dl->rwlock);
+ } else {
+ ObtainWriteLock(&dl->rwlock);
+ }
+
+ opr_mutex_enter(&dl->mutex);
+
+ if ((dl->flags & VDISKLOCK_ACQUIRING)) {
+ /* Some other thread is waiting to acquire an fs lock. If nonblock=1,
+ * we can return immediately, since we know we'll need to wait to
+ * acquire. Otherwise, wait for the other thread to finish acquiring
+ * the fs lock */
+ if (nonblock) {
+ code = EBUSY;
+ } else {
+ while ((dl->flags & VDISKLOCK_ACQUIRING)) {
+ opr_cv_wait(&dl->cv, &dl->mutex);
+ }
+ }
+ }
+
+ if (code == 0 && !(dl->flags & VDISKLOCK_ACQUIRED)) {
+ /* no other thread holds the lock on the actual file; so grab one */
+
+ /* first try, don't block on the lock to see if we can get it without
+ * waiting */
+ code = VLockFileLock(dl->lockfile, dl->offset, locktype, 1);
+
+ if (code == EBUSY && !nonblock) {
+
+ /* mark that we are waiting on the fs lock */
+ dl->flags |= VDISKLOCK_ACQUIRING;
+
+ opr_mutex_exit(&dl->mutex);
+ code = VLockFileLock(dl->lockfile, dl->offset, locktype, nonblock);
+ opr_mutex_enter(&dl->mutex);
+
+ dl->flags &= ~VDISKLOCK_ACQUIRING;
+
+ if (code == 0) {
+ dl->flags |= VDISKLOCK_ACQUIRED;
+ }
+
+ opr_cv_broadcast(&dl->cv);
+ }
+ }
+
+ if (code) {
+ if (locktype == READ_LOCK) {
+ ReleaseReadLock(&dl->rwlock);
+ } else {
+ ReleaseWriteLock(&dl->rwlock);
+ }
+ } else {
+ /* successfully got the lock, so inc the number of unlocks we need
+ * to do before we can unlock the actual file */
+ ++dl->lockers;
+ }
+
+ opr_mutex_exit(&dl->mutex);
+
+ return code;
+}
+
+/**
+ * release a lock on a file on local disk.
+ *
+ * @param[in] dl the struct VDiskLock to release
+ * @param[in] locktype READ_LOCK if you are unlocking a read lock, or
+ * WRITE_LOCK if you are unlocking a write lock
+ *
+ * @return operation status
+ * @retval 0 success
+ */
+void
+VReleaseDiskLock(struct VDiskLock *dl, int locktype)
+{
+ opr_Assert(locktype == READ_LOCK || locktype == WRITE_LOCK);
+
+ opr_mutex_enter(&dl->mutex);
+ opr_Assert(dl->lockers > 0);
+
+ if (--dl->lockers < 1) {
+ /* no threads are holding this lock anymore, so we can release the
+ * actual disk lock */
+ VLockFileUnlock(dl->lockfile, dl->offset);
+ dl->flags &= ~VDISKLOCK_ACQUIRED;
+ }
+
+ opr_mutex_exit(&dl->mutex);
+
+ if (locktype == READ_LOCK) {
+ ReleaseReadLock(&dl->rwlock);
+ } else {
+ ReleaseWriteLock(&dl->rwlock);
+ }
+}
+
+#endif /* AFS_DEMAND_ATTACH_FS */
git://git.openafs.org / openafs.git / blobdiff