/*
* 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
Logs missing inode numbers.
2.3 features:
- Increments directory version number by 200 (rather than by 1) when it is salvaged, in order to prevent problems due to the fact that a version number can be promised to a workstation before it is written to disk. If the server crashes, it may have an older version. Salvaging it could bring the version number up to the same version the workstation believed it already had a call back on.
+ Increments directory version number by 200 (rather than by 1) when it is salvaged, in order to prevent problems due to the fact that a version number can be promised to a workstation before it is written to disk. If the server crashes, it may have an older version. Salvaging it could bring the version number up to the same version the workstation believed it already had a call back on.
2.4 features:
Locks the file /vice/vol/salvage.lock before starting. Aborts if it can't acquire the lock.
#include <afsconfig.h>
#include <afs/param.h>
+#include <afs/procmgmt.h>
+#include <roken.h>
-#ifndef AFS_NT40_ENV
-#include <sys/param.h>
-#include <sys/file.h>
-#ifndef ITIMER_REAL
-#include <sys/time.h>
-#endif /* ITIMER_REAL */
+#ifdef HAVE_SYS_FILE_H
+# include <sys/file.h>
#endif
-#include <stdlib.h>
-#include <stdio.h>
-#include <string.h>
-#include <dirent.h>
-#include <sys/stat.h>
-#include <time.h>
-#include <errno.h>
+
#ifdef AFS_NT40_ENV
-#include <io.h>
#include <WINNT/afsevent.h>
#endif
#ifndef WCOREDUMP
#endif
#include <rx/xdr.h>
#include <afs/afsint.h>
-#include <afs/assert.h>
+#include <afs/afs_assert.h>
#if !defined(AFS_SGI_ENV) && !defined(AFS_NT40_ENV)
#if defined(AFS_VFSINCL_ENV)
#include <sys/vnode.h>
#ifdef AFS_OSF_ENV
#include <ufs/inode.h>
#else /* AFS_OSF_ENV */
-#if !defined(AFS_LINUX20_ENV) && !defined(AFS_XBSD_ENV)
+#if !defined(AFS_LINUX20_ENV) && !defined(AFS_XBSD_ENV) && !defined(AFS_DARWIN_ENV)
#include <sys/inode.h>
#endif
#endif
#include <sys/lockf.h>
#else
#ifdef AFS_HPUX_ENV
-#include <unistd.h>
#include <checklist.h>
#else
#if defined(AFS_SGI_ENV)
-#include <unistd.h>
-#include <fcntl.h>
#include <mntent.h>
#else
#if defined(AFS_SUN_ENV) || defined(AFS_SUN5_ENV)
#ifdef AFS_SUN5_ENV
-#include <unistd.h>
#include <sys/mnttab.h>
#include <sys/mntent.h>
#else
#endif /* AFS_HPUX_ENV */
#endif
#endif
-#include <fcntl.h>
#ifndef AFS_NT40_ENV
#include <afs/osi_inode.h>
#endif
#include <afs/dir.h>
#include <afs/afsutil.h>
#include <afs/fileutil.h>
-#include <afs/procmgmt.h> /* signal(), kill(), wait(), etc. */
-#ifndef AFS_NT40_ENV
-#include <syslog.h>
-#endif
#include "nfs.h"
#include "lwp.h"
#include "volume.h"
#include "partition.h"
#include "daemon_com.h"
+#include "daemon_com_inline.h"
#include "fssync.h"
+#include "volume_inline.h"
#include "salvsync.h"
#include "viceinode.h"
#include "salvage.h"
#include "volinodes.h" /* header magic number, etc. stuff */
#include "vol-salvage.h"
+#include "common.h"
#include "vol_internal.h"
+#include <afs/acl.h>
+#include <afs/prs_fs.h>
#ifdef FSSYNC_BUILD_CLIENT
#include "vg_cache.h"
#include <pthread.h>
#endif
-/*@+fcnmacros +macrofcndecl@*/
-#ifdef O_LARGEFILE
-#ifdef S_SPLINT_S
-extern off64_t afs_lseek(int FD, off64_t O, int F);
-#endif /*S_SPLINT_S */
-#define afs_lseek(FD, O, F) lseek64(FD, (off64_t) (O), F)
-#define afs_stat stat64
-#define afs_fstat fstat64
-#define afs_open open64
-#define afs_fopen fopen64
-#else /* !O_LARGEFILE */
-#ifdef S_SPLINT_S
-extern off_t afs_lseek(int FD, off_t O, int F);
-#endif /*S_SPLINT_S */
-#define afs_lseek(FD, O, F) lseek(FD, (off_t) (O), F)
-#define afs_stat stat
-#define afs_fstat fstat
-#define afs_open open
-#define afs_fopen fopen
-#endif /* !O_LARGEFILE */
-/*@=fcnmacros =macrofcndecl@*/
-
#ifdef AFS_OSF_ENV
extern void *calloc();
#endif
#define ROOTINODE 2 /* Root inode of a 4.2 Unix file system
* partition */
-Device fileSysDevice; /* The device number of the current
- * partition being salvaged */
-#ifdef AFS_NT40_ENV
-char fileSysPath[8];
-#else
-char *fileSysPath; /* The path of the mounted partition currently
- * being salvaged, i.e. the directory
- * containing the volume headers */
-#endif
-char *fileSysPathName; /* NT needs this to make name pretty in log. */
-IHandle_t *VGLinkH; /* Link handle for current volume group. */
-int VGLinkH_cnt; /* # of references to lnk handle. */
-struct DiskPartition64 *fileSysPartition; /* Partition being salvaged */
+/**
+ * information that is 'global' to a particular salvage job.
+ */
+struct SalvInfo {
+ Device fileSysDevice; /**< The device number of the current partition
+ * being salvaged */
+ char fileSysPath[8]; /**< The path of the mounted partition currently
+ * being salvaged, i.e. the directory containing
+ * the volume headers */
+ char *fileSysPathName; /**< NT needs this to make name pretty log. */
+ IHandle_t *VGLinkH; /**< Link handle for current volume group. */
+ int VGLinkH_cnt; /**< # of references to lnk handle. */
+ struct DiskPartition64 *fileSysPartition; /**< Partition being salvaged */
+
#ifndef AFS_NT40_ENV
-char *fileSysDeviceName; /* The block device where the file system
- * being salvaged was mounted */
-char *filesysfulldev;
+ char *fileSysDeviceName; /**< The block device where the file system being
+ * salvaged was mounted */
+ char *filesysfulldev;
#endif
-int VolumeChanged; /* Set by any routine which would change the volume in
- * a way which would require callback is to be broken if the
- * volume was put back on line by an active file server */
-
-VolumeDiskData VolInfo; /* A copy of the last good or salvaged volume header dealt with */
-
-int nVolumesInInodeFile; /* Number of read-write volumes summarized */
-int inodeFd; /* File descriptor for inode file */
-
-
-struct VnodeInfo vnodeInfo[nVNODECLASSES];
-
-
-struct VolumeSummary *volumeSummaryp = NULL; /* Holds all the volumes in a part */
-int nVolumes; /* Number of volumes (read-write and read-only)
- * in volume summary */
+ int VolumeChanged; /**< Set by any routine which would change the
+ * volume in a way which would require callbacks
+ * to be broken if the volume was put back on
+ * on line by an active file server */
+
+ VolumeDiskData VolInfo; /**< A copy of the last good or salvaged volume
+ * header dealt with */
+
+ int nVolumesInInodeFile; /**< Number of read-write volumes summarized */
+ FD_t inodeFd; /**< File descriptor for inode file */
+
+ struct VolumeSummary *volumeSummaryp; /**< Holds all the volumes in a part */
+ int nVolumes; /**< Number of volumes (read-write and read-only)
+ * in volume summary */
+ struct InodeSummary *inodeSummary; /**< contains info on all the relevant
+ * inodes */
+
+ struct VnodeInfo vnodeInfo[nVNODECLASSES]; /**< contains info on all of the
+ * vnodes in the volume that
+ * we are currently looking
+ * at */
+ int useFSYNC; /**< 0 if the fileserver is unavailable; 1 if we should try
+ * to contact the fileserver over FSYNC */
+};
char *tmpdir = NULL;
/* Forward declarations */
-/*@printflike@*/ void Log(const char *format, ...);
-/*@printflike@*/ void Abort(const char *format, ...);
-static int IsVnodeOrphaned(VnodeId vnode);
-static int AskVolumeSummary(VolumeId singleVolumeNumber);
+static int IsVnodeOrphaned(struct SalvInfo *salvinfo, VnodeId vnode);
+static int AskVolumeSummary(struct SalvInfo *salvinfo,
+ VolumeId singleVolumeNumber);
+static void MaybeAskOnline(struct SalvInfo *salvinfo, VolumeId volumeId);
+static void AskError(struct SalvInfo *salvinfo, VolumeId volumeId);
+
+#if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
+static int LockVolume(struct SalvInfo *salvinfo, VolumeId volumeId);
+#endif /* AFS_DEMAND_ATTACH_FS || AFS_DEMAND_ATTACH_UTIL */
/* Uniquifier stored in the Inode */
static Unique
}
static int
-BadError(register int aerror)
+BadError(int aerror)
{
if (aerror == EPERM || aerror == ENXIO || aerror == ENOENT)
return 1;
childJob_t myjob = { SALVAGER_MAGIC, NOT_CHILD, "" };
#endif
-/* Get the salvage lock if not already held. Hold until process exits. */
-void
-ObtainSalvageLock(void)
+/**
+ * Get the salvage lock if not already held. Hold until process exits.
+ *
+ * @param[in] locktype READ_LOCK or WRITE_LOCK
+ */
+static void
+_ObtainSalvageLock(int locktype)
{
- FD_t salvageLock;
+ struct VLockFile salvageLock;
+ int offset = 0;
+ int nonblock = 1;
+ int code;
-#ifdef AFS_NT40_ENV
- salvageLock =
- (FD_t)CreateFile(AFSDIR_SERVER_SLVGLOCK_FILEPATH, 0, 0, NULL,
- OPEN_ALWAYS, FILE_ATTRIBUTE_NORMAL, NULL);
- if (salvageLock == INVALID_FD) {
+ VLockFileInit(&salvageLock, AFSDIR_SERVER_SLVGLOCK_FILEPATH);
+
+ code = VLockFileLock(&salvageLock, offset, locktype, nonblock);
+ if (code == EBUSY) {
fprintf(stderr,
- "salvager: There appears to be another salvager running! Aborted.\n");
+ "salvager: There appears to be another salvager running! "
+ "Aborted.\n");
Exit(1);
- }
-#else
- salvageLock =
- afs_open(AFSDIR_SERVER_SLVGLOCK_FILEPATH, O_CREAT | O_RDWR, 0666);
- if (salvageLock < 0) {
+ } else if (code) {
fprintf(stderr,
- "salvager: can't open salvage lock file %s, aborting\n",
- AFSDIR_SERVER_SLVGLOCK_FILEPATH);
+ "salvager: Error %d trying to acquire salvage lock! "
+ "Aborted.\n", code);
Exit(1);
}
-#ifdef AFS_DARWIN_ENV
- if (flock(salvageLock, LOCK_EX) == -1) {
-#else
- if (lockf(salvageLock, F_LOCK, 0) == -1) {
-#endif
- fprintf(stderr,
- "salvager: There appears to be another salvager running! Aborted.\n");
- Exit(1);
- }
-#endif
+}
+void
+ObtainSalvageLock(void)
+{
+ _ObtainSalvageLock(WRITE_LOCK);
+}
+void
+ObtainSharedSalvageLock(void)
+{
+ _ObtainSalvageLock(READ_LOCK);
}
FILE *mntfp;
struct mntent *mntent;
- assert(mntfp = setmntent(MOUNTED, "r"));
+ osi_Assert(mntfp = setmntent(MOUNTED, "r"));
while (mntent = getmntent(mntfp)) {
if (!strcmp(part, mntent->mnt_dir))
break;
/* Check if the given inode is the root of the filesystem. */
#ifndef AFS_SGI_XFS_IOPS_ENV
int
-IsRootInode(struct afs_stat *status)
+IsRootInode(struct afs_stat_st *status)
{
/*
* The root inode is not a fixed value in XFS partitions. So we need to
- * see if the partition is in the list of mounted partitions. This only
+ * see if the partition is in the list of mounted partitions. This only
* affects the SalvageFileSys path, so we check there.
*/
return (status->st_ino == ROOTINODE);
SameDisk(struct DiskPartition64 *p1, struct DiskPartition64 *p2)
{
#define RES_LEN 256
- char res[RES_LEN];
- int d1, d2;
+ char res1[RES_LEN];
+ char res2[RES_LEN];
+
static int dowarn = 1;
- if (!QueryDosDevice(p1->devName, res, RES_LEN - 1))
+ if (!QueryDosDevice(p1->devName, res1, RES_LEN - 1))
return 1;
- if (strncmp(res, HDSTR, HDLEN)) {
+ if (strncmp(res1, HDSTR, HDLEN)) {
if (dowarn) {
dowarn = 0;
Log("WARNING: QueryDosDevice is returning %s, not %s for %s\n",
- res, HDSTR, p1->devName);
+ res1, HDSTR, p1->devName);
}
- return 1;
}
- d1 = atoi(&res[HDLEN]);
-
- if (!QueryDosDevice(p2->devName, res, RES_LEN - 1))
+ if (!QueryDosDevice(p2->devName, res2, RES_LEN - 1))
return 1;
- if (strncmp(res, HDSTR, HDLEN)) {
+ if (strncmp(res2, HDSTR, HDLEN)) {
if (dowarn) {
dowarn = 0;
Log("WARNING: QueryDosDevice is returning %s, not %s for %s\n",
- res, HDSTR, p2->devName);
+ res2, HDSTR, p2->devName);
}
- return 1;
}
- d2 = atoi(&res[HDLEN]);
- return d1 == d2;
+ return (0 == _strnicmp(res1, res2, RES_LEN - 1));
}
#else
#define SameDisk(P1, P2) ((P1)->device/PartsPerDisk == (P2)->device/PartsPerDisk)
}
if (thisjob) {
- /* Check to see if thisjob is for a disk that we are already
+ /* Check to see if thisjob is for a disk that we are already
* salvaging. If it is, link it in as the next job to do. The
- * jobs array has 1 entry per disk being salvages. numjobs is
- * the total number of disks currently being salvaged. In
+ * jobs array has 1 entry per disk being salvages. numjobs is
+ * the total number of disks currently being salvaged. In
* order to keep thejobs array compact, when a disk is
- * completed, the hightest element in the jobs array is moved
+ * completed, the hightest element in the jobs array is moved
* down to now open slot.
*/
for (j = 0; j < numjobs; j++) {
* job to finish. When it's done, clean up after it.
*/
pid = wait(&wstatus);
- assert(pid != -1);
+ osi_Assert(pid != -1);
for (j = 0; j < numjobs; j++) { /* Find which job it is */
if (pid == jobs[j]->pid)
break;
}
- assert(j < numjobs);
+ osi_Assert(j < numjobs);
if (WCOREDUMP(wstatus)) { /* Say if the job core dumped */
Log("Salvage of %s core dumped!\n", jobs[j]->partP->name);
}
jobs[j] = jobs[j]->nextjob; /* Step to next part on same disk */
free(oldjob); /* free the old job */
- /* If there is another partition on the disk to salvage, then
+ /* If there is another partition on the disk to salvage, then
* say we will start it (startjob). If not, then put thisjob there
* and say we will start it.
*/
ShowLog = 0;
for (fd = 0; fd < 16; fd++)
close(fd);
- open("/", 0);
+ open(OS_DIRSEP, 0);
dup2(0, 1);
dup2(0, 2);
#ifndef AFS_NT40_ENV
} else
#endif
{
- (void)afs_snprintf(logFileName, sizeof logFileName,
- "%s.%d",
- AFSDIR_SERVER_SLVGLOG_FILEPATH,
- jobs[startjob]->jobnumb);
+ snprintf(logFileName, sizeof logFileName, "%s.%d",
+ AFSDIR_SERVER_SLVGLOG_FILEPATH,
+ jobs[startjob]->jobnumb);
logFile = afs_fopen(logFileName, "w");
}
if (!logFile)
#endif
if (!partP) {
for (i = 0; i < jobcount; i++) {
- (void)afs_snprintf(logFileName, sizeof logFileName, "%s.%d",
- AFSDIR_SERVER_SLVGLOG_FILEPATH, i);
+ snprintf(logFileName, sizeof logFileName, "%s.%d",
+ AFSDIR_SERVER_SLVGLOG_FILEPATH, i);
if ((passLog = afs_fopen(logFileName, "r"))) {
while (fgets(buf, sizeof(buf), passLog)) {
fputs(buf, logFile);
{
char pbuf[128], *ptr;
strcpy(pbuf, pbuffer);
- ptr = (char *)strrchr(pbuf, '/');
+ ptr = (char *)strrchr(pbuf, OS_DIRSEPC);
if (ptr) {
*ptr = '\0';
strcpy(wpath, pbuf);
} else
return NULL;
- ptr = (char *)strrchr(pbuffer, '/');
+ ptr = (char *)strrchr(pbuffer, OS_DIRSEPC);
if (ptr) {
strcpy(pbuffer, ptr + 1);
return pbuffer;
{
char *name, *tdir;
char inodeListPath[256];
- FILE *inodeFile;
+ FD_t inodeFile = INVALID_FD;
static char tmpDevName[100];
static char wpath[100];
struct VolumeSummary *vsp, *esp;
int i, j;
int code;
+ int tries = 0;
+ struct SalvInfo l_salvinfo;
+ struct SalvInfo *salvinfo = &l_salvinfo;
+
+ retry:
+ memset(salvinfo, 0, sizeof(*salvinfo));
+
+ tries++;
+ if (inodeFile != INVALID_FD) {
+ OS_CLOSE(inodeFile);
+ inodeFile = INVALID_FD;
+ }
+ if (tries > VOL_MAX_CHECKOUT_RETRIES) {
+ Abort("Raced too many times with fileserver restarts while trying to "
+ "checkout/lock volumes; Aborted\n");
+ }
+#if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
+ if (tries > 1) {
+ /* unlock all previous volume locks, since we're about to lock them
+ * again */
+ VLockFileReinit(&partP->volLockFile);
+ }
+#endif /* AFS_DEMAND_ATTACH_FS || AFS_DEMAND_ATTACH_UTIL */
- fileSysPartition = partP;
- fileSysDevice = fileSysPartition->device;
- fileSysPathName = VPartitionPath(fileSysPartition);
+ salvinfo->fileSysPartition = partP;
+ salvinfo->fileSysDevice = salvinfo->fileSysPartition->device;
+ salvinfo->fileSysPathName = VPartitionPath(salvinfo->fileSysPartition);
#ifdef AFS_NT40_ENV
/* Opendir can fail on "C:" but not on "C:\" if C is empty! */
- (void)sprintf(fileSysPath, "%s\\", fileSysPathName);
+ (void)sprintf(salvinfo->fileSysPath, "%s" OS_DIRSEP, salvinfo->fileSysPathName);
name = partP->devName;
#else
- fileSysPath = fileSysPathName;
+ strlcpy(salvinfo->fileSysPath, salvinfo->fileSysPathName, sizeof(salvinfo->fileSysPath));
strcpy(tmpDevName, partP->devName);
name = get_DevName(tmpDevName, wpath);
- fileSysDeviceName = name;
- filesysfulldev = wpath;
+ salvinfo->fileSysDeviceName = name;
+ salvinfo->filesysfulldev = wpath;
#endif
- VLockPartition(partP->name);
- if (singleVolumeNumber || ForceSalvage)
+ if (singleVolumeNumber) {
+#if !(defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL))
+ /* only non-DAFS locks the partition when salvaging a single volume;
+ * DAFS will lock the individual volumes in the VG */
+ VLockPartition(partP->name);
+#endif /* !(AFS_DEMAND_ATTACH_FS || AFS_DEMAND_ATTACH_UTIL) */
+
ForceSalvage = 1;
- else
- ForceSalvage = UseTheForceLuke(fileSysPath);
- if (singleVolumeNumber) {
/* salvageserver already setup fssync conn for us */
if ((programType != salvageServer) && !VConnectFS()) {
Abort("Couldn't connect to file server\n");
}
- AskOffline(singleVolumeNumber, partP->name);
+
+ salvinfo->useFSYNC = 1;
+ AskOffline(salvinfo, singleVolumeNumber);
+#if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
+ if (LockVolume(salvinfo, singleVolumeNumber)) {
+ goto retry;
+ }
+#endif /* AFS_DEMAND_ATTACH_FS || AFS_DEMAND_ATTACH_UTIL */
+
} else {
+ salvinfo->useFSYNC = 0;
+ VLockPartition(partP->name);
+ if (ForceSalvage) {
+ ForceSalvage = 1;
+ } else {
+ ForceSalvage = UseTheForceLuke(salvinfo->fileSysPath);
+ }
if (!Showmode)
Log("SALVAGING FILE SYSTEM PARTITION %s (device=%s%s)\n",
partP->name, name, (Testing ? "(READONLY mode)" : ""));
/*
- * Remove any leftover /vicepa/salvage.inodes.* or /vicepa/salvage.temp.*
- * files
+ * Remove any leftover /vicepa/salvage.inodes.* or /vicepa/salvage.temp.*
+ * files
*/
{
DIR *dirp;
struct dirent *dp;
- assert((dirp = opendir(fileSysPath)) != NULL);
+ osi_Assert((dirp = opendir(salvinfo->fileSysPath)) != NULL);
while ((dp = readdir(dirp))) {
if (!strncmp(dp->d_name, "salvage.inodes.", 15)
|| !strncmp(dp->d_name, "salvage.temp.", 13)) {
char npath[1024];
Log("Removing old salvager temp files %s\n", dp->d_name);
- strcpy(npath, fileSysPath);
- strcat(npath, "/");
+ strcpy(npath, salvinfo->fileSysPath);
+ strcat(npath, OS_DIRSEP);
strcat(npath, dp->d_name);
- unlink(npath);
+ OS_UNLINK(npath);
}
}
closedir(dirp);
}
- tdir = (tmpdir ? tmpdir : fileSysPath);
+ tdir = (tmpdir ? tmpdir : salvinfo->fileSysPath);
#ifdef AFS_NT40_ENV
(void)_putenv("TMP="); /* If "TMP" is set, then that overrides tdir. */
(void)strncpy(inodeListPath, _tempnam(tdir, "salvage.inodes."), 255);
#else
- snprintf(inodeListPath, 255, "%s/salvage.inodes.%s.%d", tdir, name,
+ snprintf(inodeListPath, 255, "%s" OS_DIRSEP "salvage.inodes.%s.%d", tdir, name,
getpid());
#endif
- inodeFile = fopen(inodeListPath, "w+b");
- if (!inodeFile) {
+ inodeFile = OS_OPEN(inodeListPath, O_RDWR|O_TRUNC|O_CREAT, 0666);
+ if (inodeFile == INVALID_FD) {
Abort("Error %d when creating inode description file %s; not salvaged\n", errno, inodeListPath);
}
#ifdef AFS_NT40_ENV
* semantics of unlink. In most places in the salvager, we really do
* mean to unlink the file at that point. Those places have been
* modified to actually do that so that the NT crt can be used there.
+ *
+ * jaltman - On NT delete on close cannot be applied to a file while the
+ * process has an open file handle that does not have DELETE file
+ * access and FILE_SHARE_DELETE. fopen() calls CreateFile() without
+ * delete privileges. As a result the nt_unlink() call will always
+ * fail.
*/
code = nt_unlink(inodeListPath);
#else
Log("Error %d when trying to unlink %s\n", errno, inodeListPath);
}
- if (GetInodeSummary(inodeFile, singleVolumeNumber) < 0) {
- fclose(inodeFile);
+ if (GetInodeSummary(salvinfo, inodeFile, singleVolumeNumber) < 0) {
+ OS_CLOSE(inodeFile);
return;
}
- inodeFd = fileno(inodeFile);
- if (inodeFd == -1)
+ salvinfo->inodeFd = inodeFile;
+ if (salvinfo->inodeFd == INVALID_FD)
Abort("Temporary file %s is missing...\n", inodeListPath);
- afs_lseek(inodeFd, 0L, SEEK_SET);
+ OS_SEEK(salvinfo->inodeFd, 0L, SEEK_SET);
if (ListInodeOption) {
- PrintInodeList();
+ PrintInodeList(salvinfo);
+ if (singleVolumeNumber) {
+ /* We've checked out the volume from the fileserver, and we need
+ * to give it back. We don't know if the volume exists or not,
+ * so we don't know whether to AskOnline or not. Try to determine
+ * if the volume exists by trying to read the volume header, and
+ * AskOnline if it is readable. */
+ MaybeAskOnline(salvinfo, singleVolumeNumber);
+ }
return;
}
/* enumerate volumes in the partition.
* Fix up inodes on last volume in set (whether it is read-write
* or read-only).
*/
- GetVolumeSummary(singleVolumeNumber);
+ if (GetVolumeSummary(salvinfo, singleVolumeNumber)) {
+ goto retry;
+ }
- for (i = j = 0, vsp = volumeSummaryp, esp = vsp + nVolumes;
- i < nVolumesInInodeFile; i = j) {
- VolumeId rwvid = inodeSummary[i].RWvolumeId;
+ for (i = j = 0, vsp = salvinfo->volumeSummaryp, esp = vsp + salvinfo->nVolumes;
+ i < salvinfo->nVolumesInInodeFile; i = j) {
+ VolumeId rwvid = salvinfo->inodeSummary[i].RWvolumeId;
for (j = i;
- j < nVolumesInInodeFile && inodeSummary[j].RWvolumeId == rwvid;
+ j < salvinfo->nVolumesInInodeFile && salvinfo->inodeSummary[j].RWvolumeId == rwvid;
j++) {
- VolumeId vid = inodeSummary[j].volumeId;
+ VolumeId vid = salvinfo->inodeSummary[j].volumeId;
struct VolumeSummary *tsp;
/* Scan volume list (from partition root directory) looking for the
* current rw volume number in the volume list from the inode scan.
* delete it now. */
for (; vsp < esp && (vsp->header.parent < rwvid); vsp++) {
if (vsp->fileName)
- DeleteExtraVolumeHeaderFile(vsp);
+ DeleteExtraVolumeHeaderFile(salvinfo, vsp);
}
/* Now match up the volume summary info from the root directory with the
* entry in the volume list obtained from scanning inodes */
- inodeSummary[j].volSummary = NULL;
+ salvinfo->inodeSummary[j].volSummary = NULL;
for (tsp = vsp; tsp < esp && (tsp->header.parent == rwvid); tsp++) {
if (tsp->header.id == vid) {
- inodeSummary[j].volSummary = tsp;
+ salvinfo->inodeSummary[j].volSummary = tsp;
tsp->fileName = 0;
break;
}
/* Salvage the group of volumes (several read-only + 1 read/write)
* starting with the current read-only volume we're looking at.
*/
- SalvageVolumeGroup(&inodeSummary[i], j - i);
+#ifdef AFS_NT40_ENV
+ nt_SalvageVolumeGroup(salvinfo, &salvinfo->inodeSummary[i], j - i);
+#else
+ DoSalvageVolumeGroup(salvinfo, &salvinfo->inodeSummary[i], j - i);
+#endif /* AFS_NT40_ENV */
+
}
/* Delete any additional volumes that were listed in the partition but which didn't have any corresponding inodes */
for (; vsp < esp; vsp++) {
if (vsp->fileName)
- DeleteExtraVolumeHeaderFile(vsp);
+ DeleteExtraVolumeHeaderFile(salvinfo, vsp);
}
if (!singleVolumeNumber) /* Remove the FORCESALVAGE file */
- RemoveTheForce(fileSysPath);
+ RemoveTheForce(salvinfo->fileSysPath);
if (!Testing && singleVolumeNumber) {
- AskOnline(singleVolumeNumber, fileSysPartition->name);
+ int foundSVN = 0;
+#if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
+ /* unlock vol headers so the fs can attach them when we AskOnline */
+ VLockFileReinit(&salvinfo->fileSysPartition->volLockFile);
+#endif /* AFS_DEMAND_ATTACH_FS || AFS_DEMAND_ATTACH_UTIL */
/* Step through the volumeSummary list and set all volumes on-line.
- * The volumes were taken off-line in GetVolumeSummary.
+ * Most volumes were taken off-line in GetVolumeSummary.
+ * If a volume was deleted, don't tell the fileserver anything, since
+ * we already told the fileserver the volume was deleted back when we
+ * we destroyed the volume header.
+ * Also, make sure we bring the singleVolumeNumber back online first.
*/
- for (j = 0; j < nVolumes; j++) {
- AskOnline(volumeSummaryp[j].header.id, fileSysPartition->name);
+
+ for (j = 0; j < salvinfo->nVolumes; j++) {
+ if (salvinfo->volumeSummaryp[j].header.id == singleVolumeNumber) {
+ foundSVN = 1;
+ if (!salvinfo->volumeSummaryp[j].deleted) {
+ AskOnline(salvinfo, singleVolumeNumber);
+ }
+ }
+ }
+
+ if (!foundSVN) {
+ /* If singleVolumeNumber is not in our volumeSummary, it means that
+ * at least one other volume in the VG is on the partition, but the
+ * RW volume is not. We've already AskOffline'd it by now, though,
+ * so make sure we don't still have the volume checked out. */
+ AskDelete(salvinfo, singleVolumeNumber);
+ }
+
+ for (j = 0; j < salvinfo->nVolumes; j++) {
+ if (salvinfo->volumeSummaryp[j].header.id != singleVolumeNumber) {
+ if (!salvinfo->volumeSummaryp[j].deleted) {
+ AskOnline(salvinfo, salvinfo->volumeSummaryp[j].header.id);
+ }
+ }
}
} else {
if (!Showmode)
Log("SALVAGING OF PARTITION %s%s COMPLETED\n",
- fileSysPartition->name, (Testing ? " (READONLY mode)" : ""));
+ salvinfo->fileSysPartition->name, (Testing ? " (READONLY mode)" : ""));
}
- fclose(inodeFile); /* SalvageVolumeGroup was the last which needed it. */
+ OS_CLOSE(inodeFile); /* SalvageVolumeGroup was the last which needed it. */
}
void
-DeleteExtraVolumeHeaderFile(register struct VolumeSummary *vsp)
+DeleteExtraVolumeHeaderFile(struct SalvInfo *salvinfo, struct VolumeSummary *vsp)
{
char path[64];
- sprintf(path, "%s/%s", fileSysPath, vsp->fileName);
+ sprintf(path, "%s" OS_DIRSEP "%s", salvinfo->fileSysPath, vsp->fileName);
if (!Showmode)
Log("The volume header file %s is not associated with any actual data (%sdeleted)\n", path, (Testing ? "would have been " : ""));
if (!Testing) {
afs_int32 code;
- code = VDestroyVolumeDiskHeader(fileSysPartition, vsp->header.id, vsp->header.parent);
+ code = VDestroyVolumeDiskHeader(salvinfo->fileSysPartition, vsp->header.id, vsp->header.parent);
if (code) {
Log("Error %ld destroying volume disk header for volume %lu\n",
afs_printable_int32_ld(code),
if (unlink(path) && errno != ENOENT) {
Log("Unable to unlink %s (errno = %d)\n", path, errno);
}
+ if (salvinfo->useFSYNC) {
+ AskDelete(salvinfo, vsp->header.id);
+ }
+ vsp->deleted = 1;
}
vsp->fileName = 0;
}
int
CompareInodes(const void *_p1, const void *_p2)
{
- register const struct ViceInodeInfo *p1 = _p1;
- register const struct ViceInodeInfo *p2 = _p2;
+ const struct ViceInodeInfo *p1 = _p1;
+ const struct ViceInodeInfo *p2 = _p2;
if (p1->u.vnode.vnodeNumber == INODESPECIAL
|| p2->u.vnode.vnodeNumber == INODESPECIAL) {
VolumeId p1rwid, p2rwid;
}
void
-CountVolumeInodes(register struct ViceInodeInfo *ip, int maxInodes,
- register struct InodeSummary *summary)
+CountVolumeInodes(struct ViceInodeInfo *ip, int maxInodes,
+ struct InodeSummary *summary)
{
VolumeId volume = ip->u.vnode.volumeId;
VolumeId rwvolume = volume;
- register int n, nSpecial;
- register Unique maxunique;
+ int n, nSpecial;
+ Unique maxunique;
n = nSpecial = 0;
maxunique = 0;
while (maxInodes-- && volume == ip->u.vnode.volumeId) {
* be unlinked by the caller.
*/
int
-GetInodeSummary(FILE *inodeFile, VolumeId singleVolumeNumber)
+GetInodeSummary(struct SalvInfo *salvinfo, FD_t inodeFile, VolumeId singleVolumeNumber)
{
- struct afs_stat status;
int forceSal, err;
int code;
- struct ViceInodeInfo *ip;
+ struct ViceInodeInfo *ip, *ip_save;
struct InodeSummary summary;
char summaryFileName[50];
- FILE *summaryFile;
+ FD_t summaryFile = INVALID_FD;
#ifdef AFS_NT40_ENV
- char *dev = fileSysPath;
- char *wpath = fileSysPath;
+ char *dev = salvinfo->fileSysPath;
+ char *wpath = salvinfo->fileSysPath;
#else
- char *dev = fileSysDeviceName;
- char *wpath = filesysfulldev;
+ char *dev = salvinfo->fileSysDeviceName;
+ char *wpath = salvinfo->filesysfulldev;
#endif
- char *part = fileSysPath;
+ char *part = salvinfo->fileSysPath;
char *tdir;
+ int i;
+ int retcode = 0;
+ int deleted = 0;
+ afs_sfsize_t st_size;
/* This file used to come from vfsck; cobble it up ourselves now... */
if ((err =
- ListViceInodes(dev, fileSysPath, inodeFile,
+ ListViceInodes(dev, salvinfo->fileSysPath, inodeFile,
singleVolumeNumber ? OnlyOneVolume : 0,
singleVolumeNumber, &forceSal, forceR, wpath, NULL)) < 0) {
if (err == -2) {
Log("*** I/O error %d when writing a tmp inode file; Not salvaged %s ***\nIncrease space on partition or use '-tmpdir'\n", errno, dev);
- return -1;
+ retcode = -1;
+ goto error;
}
Abort("Unable to get inodes for \"%s\"; not salvaged\n", dev);
}
Log("***Forced salvage of all volumes on this partition***\n");
ForceSalvage = 1;
}
- fseek(inodeFile, 0L, SEEK_SET);
- inodeFd = fileno(inodeFile);
- if (inodeFd == -1 || afs_fstat(inodeFd, &status) == -1) {
+ OS_SEEK(inodeFile, 0L, SEEK_SET);
+ salvinfo->inodeFd = inodeFile;
+ if (salvinfo->inodeFd == INVALID_FD ||
+ (st_size = OS_SIZE(salvinfo->inodeFd)) == -1) {
Abort("No inode description file for \"%s\"; not salvaged\n", dev);
}
tdir = (tmpdir ? tmpdir : part);
#ifdef AFS_NT40_ENV
(void)_putenv("TMP="); /* If "TMP" is set, then that overrides tdir. */
- (void)strcpy(summaryFileName, _tempnam(tdir, "salvage.temp"));
+ (void)strcpy(summaryFileName, _tempnam(tdir, "salvage.temp."));
#else
- (void)afs_snprintf(summaryFileName, sizeof summaryFileName,
- "%s/salvage.temp.%d", tdir, getpid());
+ snprintf(summaryFileName, sizeof summaryFileName,
+ "%s" OS_DIRSEP "salvage.temp.%d", tdir, getpid());
#endif
- summaryFile = afs_fopen(summaryFileName, "a+");
- if (summaryFile == NULL) {
+ summaryFile = OS_OPEN(summaryFileName, O_RDWR|O_APPEND|O_CREAT, 0666);
+ if (summaryFile == INVALID_FD) {
Abort("Unable to create inode summary file\n");
}
* semantics of unlink. In most places in the salvager, we really do
* mean to unlink the file at that point. Those places have been
* modified to actually do that so that the NT crt can be used there.
+ *
+ * jaltman - As commented elsewhere, this cannot work because fopen()
+ * does not open files with DELETE and FILE_SHARE_DELETE.
*/
code = nt_unlink(summaryFileName);
#else
}
if (!canfork || debug || Fork() == 0) {
- int nInodes;
- unsigned long st_size=(unsigned long) status.st_size;
- nInodes = st_size / sizeof(struct ViceInodeInfo);
+ int nInodes = st_size / sizeof(struct ViceInodeInfo);
if (nInodes == 0) {
- fclose(summaryFile);
+ OS_CLOSE(summaryFile);
if (!singleVolumeNumber) /* Remove the FORCESALVAGE file */
- RemoveTheForce(fileSysPath);
+ RemoveTheForce(salvinfo->fileSysPath);
else {
struct VolumeSummary *vsp;
int i;
+ int foundSVN = 0;
- GetVolumeSummary(singleVolumeNumber);
+ GetVolumeSummary(salvinfo, singleVolumeNumber);
- for (i = 0, vsp = volumeSummaryp; i < nVolumes; i++) {
- if (vsp->fileName)
- DeleteExtraVolumeHeaderFile(vsp);
+ for (i = 0, vsp = salvinfo->volumeSummaryp; i < salvinfo->nVolumes; i++) {
+ if (vsp->fileName) {
+ if (vsp->header.id == singleVolumeNumber) {
+ foundSVN = 1;
+ }
+ DeleteExtraVolumeHeaderFile(salvinfo, vsp);
+ }
+ }
+
+ if (!foundSVN) {
+ if (Testing) {
+ MaybeAskOnline(salvinfo, singleVolumeNumber);
+ } else {
+ /* make sure we get rid of stray .vol headers, even if
+ * they're not in our volume summary (might happen if
+ * e.g. something else created them and they're not in the
+ * fileserver VGC) */
+ VDestroyVolumeDiskHeader(salvinfo->fileSysPartition,
+ singleVolumeNumber, 0 /*parent*/);
+ AskDelete(salvinfo, singleVolumeNumber);
+ }
}
}
Log("%s vice inodes on %s; not salvaged\n",
singleVolumeNumber ? "No applicable" : "No", dev);
- return -1;
+ retcode = -1;
+ deleted = 1;
+ goto error;
}
ip = (struct ViceInodeInfo *)malloc(nInodes*sizeof(struct ViceInodeInfo));
if (ip == NULL) {
- fclose(summaryFile);
+ OS_CLOSE(summaryFile);
Abort
("Unable to allocate enough space to read inode table; %s not salvaged\n",
dev);
}
- if (read(inodeFd, ip, st_size) != st_size) {
- fclose(summaryFile);
+ if (OS_READ(salvinfo->inodeFd, ip, st_size) != st_size) {
+ OS_CLOSE(summaryFile);
Abort("Unable to read inode table; %s not salvaged\n", dev);
}
qsort(ip, nInodes, sizeof(struct ViceInodeInfo), CompareInodes);
- if (afs_lseek(inodeFd, 0, SEEK_SET) == -1
- || write(inodeFd, ip, st_size) != st_size) {
- fclose(summaryFile);
+ if (OS_SEEK(salvinfo->inodeFd, 0, SEEK_SET) == -1
+ || OS_WRITE(salvinfo->inodeFd, ip, st_size) != st_size) {
+ OS_CLOSE(summaryFile);
Abort("Unable to rewrite inode table; %s not salvaged\n", dev);
}
summary.index = 0;
+ ip_save = ip;
while (nInodes) {
CountVolumeInodes(ip, nInodes, &summary);
- if (fwrite(&summary, sizeof(summary), 1, summaryFile) != 1) {
+ if (OS_WRITE(summaryFile, &summary, sizeof(summary)) != sizeof(summary)) {
Log("Difficulty writing summary file (errno = %d); %s not salvaged\n", errno, dev);
- fclose(summaryFile);
- return -1;
+ OS_CLOSE(summaryFile);
+ retcode = -1;
+ goto error;
}
summary.index += (summary.nInodes);
nInodes -= summary.nInodes;
ip += summary.nInodes;
}
+ free(ip_save);
+ ip = ip_save = NULL;
/* Following fflush is not fclose, because if it was debug mode would not work */
- if (fflush(summaryFile) == EOF || fsync(fileno(summaryFile)) == -1) {
+ if (OS_SYNC(summaryFile) == -1) {
Log("Unable to write summary file (errno = %d); %s not salvaged\n", errno, dev);
- fclose(summaryFile);
- return -1;
+ OS_CLOSE(summaryFile);
+ retcode = -1;
+ goto error;
}
if (canfork && !debug) {
ShowLog = 0;
}
} else {
if (Wait("Inode summary") == -1) {
- fclose(summaryFile);
+ OS_CLOSE(summaryFile);
Exit(1); /* salvage of this partition aborted */
}
}
- assert(afs_fstat(fileno(summaryFile), &status) != -1);
- if (status.st_size != 0) {
+
+ st_size = OS_SIZE(summaryFile);
+ osi_Assert(st_size >= 0);
+ if (st_size != 0) {
int ret;
- unsigned long st_status=(unsigned long)status.st_size;
- inodeSummary = (struct InodeSummary *)malloc(st_status);
- assert(inodeSummary != NULL);
+ salvinfo->inodeSummary = (struct InodeSummary *)malloc(st_size);
+ osi_Assert(salvinfo->inodeSummary != NULL);
/* For GNU we need to do lseek to get the file pointer moved. */
- assert(afs_lseek(fileno(summaryFile), 0, SEEK_SET) == 0);
- ret = read(fileno(summaryFile), inodeSummary, st_status);
- assert(ret == st_status);
+ osi_Assert(OS_SEEK(summaryFile, 0, SEEK_SET) == 0);
+ ret = OS_READ(summaryFile, salvinfo->inodeSummary, st_size);
+ osi_Assert(ret == st_size);
}
- nVolumesInInodeFile =(unsigned long)(status.st_size) / sizeof(struct InodeSummary);
- Log("%d nVolumesInInodeFile %d \n",nVolumesInInodeFile,(unsigned long)(status.st_size));
- fclose(summaryFile);
- return 0;
+ salvinfo->nVolumesInInodeFile = st_size / sizeof(struct InodeSummary);
+ for (i = 0; i < salvinfo->nVolumesInInodeFile; i++) {
+ salvinfo->inodeSummary[i].volSummary = NULL;
+ }
+ Log("%d nVolumesInInodeFile %lu \n",salvinfo->nVolumesInInodeFile,(unsigned long)st_size);
+ OS_CLOSE(summaryFile);
+
+ error:
+ if (retcode && singleVolumeNumber && !deleted) {
+ AskError(salvinfo, singleVolumeNumber);
+ }
+
+ return retcode;
}
/* Comparison routine for volume sort.
int
CompareVolumes(const void *_p1, const void *_p2)
{
- register const struct VolumeSummary *p1 = _p1;
- register const struct VolumeSummary *p2 = _p2;
+ const struct VolumeSummary *p1 = _p1;
+ const struct VolumeSummary *p2 = _p2;
if (p1->header.parent != p2->header.parent)
return p1->header.parent < p2->header.parent ? -1 : 1;
if (p1->header.id == p1->header.parent) /* p1 is rw volume */
* salvaging a whole partition
*
* @return whether we obtained the volume summary information or not
- * @retval 0 success; we obtained the volume summary information
- * @retval nonzero we did not get the volume summary information; either the
- * fileserver responded with an error, or we are not supposed to
- * ask the fileserver for the information (e.g. we are salvaging
- * the entire partition or we are not the salvageserver)
+ * @retval 0 success; we obtained the volume summary information
+ * @retval -1 we raced with a fileserver restart; volume locks and checkout
+ * must be retried
+ * @retval 1 we did not get the volume summary information; either the
+ * fileserver responded with an error, or we are not supposed to
+ * ask the fileserver for the information (e.g. we are salvaging
+ * the entire partition or we are not the salvageserver)
*
* @note for non-DAFS, always returns 1
*/
static int
-AskVolumeSummary(VolumeId singleVolumeNumber)
+AskVolumeSummary(struct SalvInfo *salvinfo, VolumeId singleVolumeNumber)
{
afs_int32 code = 1;
-#ifdef FSSYNC_BUILD_CLIENT
+#if defined(FSSYNC_BUILD_CLIENT) && defined(AFS_DEMAND_ATTACH_FS)
if (programType == salvageServer) {
if (singleVolumeNumber) {
FSSYNC_VGQry_response_t q_res;
memset(&res, 0, sizeof(res));
- code = FSYNC_VGCQuery(fileSysPartition->name, singleVolumeNumber, &q_res, &res);
+ code = FSYNC_VGCQuery(salvinfo->fileSysPartition->name, singleVolumeNumber, &q_res, &res);
/*
* We must wait for the partition to finish scanning before
*/
if (code == SYNC_FAILED && res.hdr.reason == FSYNC_PART_SCANNING) {
Log("waiting for fileserver to finish scanning partition %s...\n",
- fileSysPartition->name);
+ salvinfo->fileSysPartition->name);
for (i = 1; code == SYNC_FAILED && res.hdr.reason == FSYNC_PART_SCANNING; i++) {
/* linearly ramp up from 1 to 10 seconds; nothing fancy,
* polled only once every ten seconds. */
sleep((i > 10) ? (i = 10) : i);
- code = FSYNC_VGCQuery(fileSysPartition->name, singleVolumeNumber, &q_res, &res);
+ code = FSYNC_VGCQuery(salvinfo->fileSysPartition->name, singleVolumeNumber, &q_res, &res);
}
}
Log("fileserver refused VGCQuery request for volume %lu on "
"partition %s, code %ld reason %ld\n",
afs_printable_uint32_lu(singleVolumeNumber),
- fileSysPartition->name,
+ salvinfo->fileSysPartition->name,
afs_printable_int32_ld(code),
afs_printable_int32_ld(res.hdr.reason));
goto done;
#ifdef SALVSYNC_BUILD_CLIENT
if (SALVSYNC_LinkVolume(q_res.rw,
singleVolumeNumber,
- fileSysPartition->name,
+ salvinfo->fileSysPartition->name,
NULL) != SYNC_OK) {
Log("schedule request failed\n");
}
Exit(SALSRV_EXIT_VOLGROUP_LINK);
}
- volumeSummaryp = malloc(VOL_VG_MAX_VOLS * sizeof(struct VolumeSummary));
- assert(volumeSummaryp != NULL);
+ salvinfo->volumeSummaryp = calloc(VOL_VG_MAX_VOLS, sizeof(struct VolumeSummary));
+ osi_Assert(salvinfo->volumeSummaryp != NULL);
- nVolumes = 0;
- vsp = volumeSummaryp;
+ salvinfo->nVolumes = 0;
+ vsp = salvinfo->volumeSummaryp;
for (i = 0; i < VOL_VG_MAX_VOLS; i++) {
char name[VMAXPATHLEN];
continue;
}
+ /* AskOffline for singleVolumeNumber was called much earlier */
if (q_res.children[i] != singleVolumeNumber) {
- AskOffline(q_res.children[i], fileSysPartition->name);
+ AskOffline(salvinfo, q_res.children[i]);
+ if (LockVolume(salvinfo, q_res.children[i])) {
+ /* need to retry */
+ return -1;
+ }
}
- code = VReadVolumeDiskHeader(q_res.children[i], fileSysPartition, &diskHdr);
+
+ code = VReadVolumeDiskHeader(q_res.children[i], salvinfo->fileSysPartition, &diskHdr);
if (code) {
Log("Cannot read header for %lu; trying to salvage group anyway\n",
afs_printable_uint32_lu(q_res.children[i]));
DiskToVolumeHeader(&vsp->header, &diskHdr);
VolumeExternalName_r(q_res.children[i], name, sizeof(name));
vsp->fileName = ToString(name);
- nVolumes++;
+ salvinfo->nVolumes++;
vsp++;
}
- qsort(volumeSummaryp, nVolumes, sizeof(struct VolumeSummary),
+ qsort(salvinfo->volumeSummaryp, salvinfo->nVolumes, sizeof(struct VolumeSummary),
CompareVolumes);
}
done:
"entire partition\n");
}
}
-#endif /* FSSYNC_BUILD_CLIENT */
+#endif /* FSSYNC_BUILD_CLIENT && AFS_DEMAND_ATTACH_FS */
return code;
}
afs_int32 nVolumes; /**< # of vols we've encountered */
afs_int32 totalVolumes; /**< max # of vols we should encounter (the
* # of vols we've alloc'd memory for) */
+ int retry; /**< do we need to retry vol lock/checkout? */
+ struct SalvInfo *salvinfo; /**< salvage job info */
};
/**
* information needed to record the volume summary data
*
* @return operation status
- * @retval 0 success
- * @retval 1 volume header is mis-named and should be deleted
+ * @retval 0 success
+ * @retval -1 volume locking raced with fileserver restart; checking out
+ * and locking volumes needs to be retried
+ * @retval 1 volume header is mis-named and should be deleted
*/
static int
RecordHeader(struct DiskPartition64 *dp, const char *name,
struct SalvageScanParams *params;
struct VolumeSummary summary;
VolumeId singleVolumeNumber;
+ struct SalvInfo *salvinfo;
params = (struct SalvageScanParams *)rock;
+ memset(&summary, 0, sizeof(summary));
+
singleVolumeNumber = params->singleVolumeNumber;
+ salvinfo = params->salvinfo;
DiskToVolumeHeader(&summary.header, hdr);
/* check if the header file is incorrectly named */
int badname = 0;
- const char *base = strrchr(name, '/');
+ const char *base = strrchr(name, OS_DIRSEPC);
if (base) {
base++;
} else {
base = name;
}
- (void)afs_snprintf(nameShouldBe, sizeof nameShouldBe,
- VFORMAT, afs_printable_uint32_lu(summary.header.id));
+ snprintf(nameShouldBe, sizeof nameShouldBe,
+ VFORMAT, afs_printable_uint32_lu(summary.header.id));
if (strcmp(nameShouldBe, base)) {
* the last try looking at it; avoid AskOffline'ing the same vol
* multiple times */
- if (singleVolumeNumber
+ if (singleVolumeNumber
&& summary.header.id != singleVolumeNumber) {
/* don't offline singleVolumeNumber; we already did that
* earlier */
- AskOffline(summary.header.id, fileSysPartition->name);
+ AskOffline(salvinfo, summary.header.id);
+
+#if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
+ if (!badname) {
+ /* don't lock the volume if the header is bad, since we're
+ * about to delete it anyway. */
+ if (LockVolume(salvinfo, summary.header.id)) {
+ params->retry = 1;
+ return -1;
+ }
+ }
+#endif /* AFS_DEMAND_ATTACH_FS || AFS_DEMAND_ATTACH_UTIL */
}
}
if (badname) {
}
}
-void
-GetVolumeSummary(VolumeId singleVolumeNumber)
+/**
+ * Populates salvinfo->volumeSummaryp with volume summary information, either by asking
+ * the fileserver for VG information, or by scanning the /vicepX partition.
+ *
+ * @param[in] singleVolumeNumber the volume ID of the single volume group we
+ * are salvaging, or 0 if this is a partition
+ * salvage
+ *
+ * @return operation status
+ * @retval 0 success
+ * @retval -1 we raced with a fileserver restart; checking out and locking
+ * volumes must be retried
+ */
+int
+GetVolumeSummary(struct SalvInfo *salvinfo, VolumeId singleVolumeNumber)
{
afs_int32 nvols = 0;
struct SalvageScanParams params;
int code;
- if (AskVolumeSummary(singleVolumeNumber) == 0) {
+ code = AskVolumeSummary(salvinfo, singleVolumeNumber);
+ if (code == 0) {
/* we successfully got the vol information from the fileserver; no
* need to scan the partition */
- return;
+ return 0;
+ }
+ if (code < 0) {
+ /* we need to retry volume checkout */
+ return code;
}
if (!singleVolumeNumber) {
/* Count how many volumes we have in /vicepX */
- code = VWalkVolumeHeaders(fileSysPartition, fileSysPath, CountHeader,
+ code = VWalkVolumeHeaders(salvinfo->fileSysPartition, salvinfo->fileSysPath, CountHeader,
NULL, &nvols);
if (code < 0) {
- Abort("Can't read directory %s; not salvaged\n", fileSysPath);
+ Abort("Can't read directory %s; not salvaged\n", salvinfo->fileSysPath);
}
if (!nvols)
nvols = 1;
nvols = VOL_VG_MAX_VOLS;
}
- volumeSummaryp = malloc(nvols * sizeof(struct VolumeSummary));
- assert(volumeSummaryp != NULL);
+ salvinfo->volumeSummaryp = calloc(nvols, sizeof(struct VolumeSummary));
+ osi_Assert(salvinfo->volumeSummaryp != NULL);
params.singleVolumeNumber = singleVolumeNumber;
- params.vsp = volumeSummaryp;
+ params.vsp = salvinfo->volumeSummaryp;
params.nVolumes = 0;
params.totalVolumes = nvols;
+ params.retry = 0;
+ params.salvinfo = salvinfo;
/* walk the partition directory of volume headers and record the info
* about them; unlinking invalid headers */
- code = VWalkVolumeHeaders(fileSysPartition, fileSysPath, RecordHeader,
+ code = VWalkVolumeHeaders(salvinfo->fileSysPartition, salvinfo->fileSysPath, RecordHeader,
UnlinkHeader, ¶ms);
+ if (params.retry) {
+ /* we apparently need to retry checking-out/locking volumes */
+ return -1;
+ }
if (code < 0) {
Abort("Failed to get volume header summary\n");
}
- nVolumes = params.nVolumes;
+ salvinfo->nVolumes = params.nVolumes;
- qsort(volumeSummaryp, nVolumes, sizeof(struct VolumeSummary),
+ qsort(salvinfo->volumeSummaryp, salvinfo->nVolumes, sizeof(struct VolumeSummary),
CompareVolumes);
+
+ return 0;
}
/* Find the link table. This should be associated with the RW volume or, if
* a RO only site, then the RO volume. For now, be cautious and hunt carefully.
*/
Inode
-FindLinkHandle(register struct InodeSummary *isp, int nVols,
+FindLinkHandle(struct InodeSummary *isp, int nVols,
struct ViceInodeInfo *allInodes)
{
int i, j;
}
int
-CreateLinkTable(register struct InodeSummary *isp, Inode ino)
+CreateLinkTable(struct SalvInfo *salvinfo, struct InodeSummary *isp, Inode ino)
{
struct versionStamp version;
FdHandle_t *fdP;
if (!VALID_INO(ino))
ino =
- IH_CREATE(NULL, fileSysDevice, fileSysPath, 0, isp->volumeId,
+ IH_CREATE(NULL, salvinfo->fileSysDevice, salvinfo->fileSysPath, 0, isp->volumeId,
INODESPECIAL, VI_LINKTABLE, isp->RWvolumeId);
if (!VALID_INO(ino))
Abort
("Unable to allocate link table inode for volume %u (error = %d)\n",
isp->RWvolumeId, errno);
- IH_INIT(VGLinkH, fileSysDevice, isp->RWvolumeId, ino);
- fdP = IH_OPEN(VGLinkH);
+ IH_INIT(salvinfo->VGLinkH, salvinfo->fileSysDevice, isp->RWvolumeId, ino);
+ fdP = IH_OPEN(salvinfo->VGLinkH);
if (fdP == NULL)
Abort("Can't open link table for volume %u (error = %d)\n",
isp->RWvolumeId, errno);
version.magic = LINKTABLEMAGIC;
version.version = LINKTABLEVERSION;
- if (FDH_WRITE(fdP, (char *)&version, sizeof(version))
+ if (FDH_PWRITE(fdP, (char *)&version, sizeof(version), 0)
!= sizeof(version))
Abort("Can't truncate link table for volume %u (error = %d)\n",
isp->RWvolumeId, errno);
nt_SVG(void *arg)
{
SVGParms_t *parms = (SVGParms_t *) arg;
- DoSalvageVolumeGroup(parms->svgp_inodeSummaryp, parms->svgp_count);
+ DoSalvageVolumeGroup(parms->svgp_salvinfo, parms->svgp_inodeSummaryp, parms->svgp_count);
return NULL;
}
void
-SalvageVolumeGroup(register struct InodeSummary *isp, int nVols)
+nt_SalvageVolumeGroup(struct SalvInfo *salvinfo, struct InodeSummary *isp, int nVols)
{
pthread_t tid;
pthread_attr_t tattr;
SVGParms_t parms;
/* Initialize per volume global variables, even if later code does so */
- VolumeChanged = 0;
- VGLinkH = NULL;
- VGLinkH_cnt = 0;
- memset(&VolInfo, 0, sizeof(VolInfo));
+ salvinfo->VolumeChanged = 0;
+ salvinfo->VGLinkH = NULL;
+ salvinfo->VGLinkH_cnt = 0;
+ memset(&salvinfo->VolInfo, 0, sizeof(salvinfo->VolInfo));
parms.svgp_inodeSummaryp = isp;
parms.svgp_count = nVols;
+ parms.svgp_salvinfo = salvinfo;
code = pthread_attr_init(&tattr);
if (code) {
Log("Failed to salvage volume group %u: pthread_attr_init()\n",
#endif /* AFS_NT40_ENV */
void
-DoSalvageVolumeGroup(register struct InodeSummary *isp, int nVols)
+DoSalvageVolumeGroup(struct SalvInfo *salvinfo, struct InodeSummary *isp, int nVols)
{
struct ViceInodeInfo *inodes, *allInodes, *ip;
int i, totalInodes, size, salvageTo;
int VGLinkH_p1 =0;
FdHandle_t *fdP = NULL;
- VGLinkH_cnt = 0;
+ salvinfo->VGLinkH_cnt = 0;
haveRWvolume = (isp->volumeId == isp->RWvolumeId
&& isp->nSpecialInodes > 0);
if ((!ShowMounts) || (ShowMounts && !haveRWvolume)) {
- if (!ForceSalvage && QuickCheck(isp, nVols))
+ if (!ForceSalvage && QuickCheck(salvinfo, isp, nVols))
return;
}
if (ShowMounts && !haveRWvolume)
allInodes = inodes - isp->index; /* this would the base of all the inodes
* for the partition, if all the inodes
* had been read into memory */
- assert(afs_lseek
- (inodeFd, isp->index * sizeof(struct ViceInodeInfo),
+ osi_Assert(OS_SEEK
+ (salvinfo->inodeFd, isp->index * sizeof(struct ViceInodeInfo),
SEEK_SET) != -1);
- assert(read(inodeFd, inodes, size) == size);
+ osi_Assert(OS_READ(salvinfo->inodeFd, inodes, size) == size);
/* Don't try to salvage a read write volume if there isn't one on this
* partition */
#ifdef AFS_NAMEI_ENV
ino = FindLinkHandle(isp, nVols, allInodes);
if (VALID_INO(ino)) {
- IH_INIT(VGLinkH, fileSysDevice, isp->RWvolumeId, ino);
- fdP = IH_OPEN(VGLinkH);
+ IH_INIT(salvinfo->VGLinkH, salvinfo->fileSysDevice, isp->RWvolumeId, ino);
+ fdP = IH_OPEN(salvinfo->VGLinkH);
+ }
+ if (VALID_INO(ino) && fdP != NULL) {
+ struct versionStamp header;
+ afs_sfsize_t nBytes;
+
+ nBytes = FDH_PREAD(fdP, (char *)&header, sizeof(struct versionStamp), 0);
+ if (nBytes != sizeof(struct versionStamp)
+ || header.magic != LINKTABLEMAGIC) {
+ Log("Bad linktable header for volume %u.\n", isp->RWvolumeId);
+ FDH_REALLYCLOSE(fdP);
+ fdP = NULL;
+ }
}
if (!VALID_INO(ino) || fdP == NULL) {
Log("%s link table for volume %u.\n",
Testing ? "Would have recreated" : "Recreating", isp->RWvolumeId);
if (Testing) {
- IH_INIT(VGLinkH, fileSysDevice, -1, -1);
+ IH_INIT(salvinfo->VGLinkH, salvinfo->fileSysDevice, -1, -1);
} else {
int i, j;
struct ViceInodeInfo *ip;
- CreateLinkTable(isp, ino);
- fdP = IH_OPEN(VGLinkH);
+ CreateLinkTable(salvinfo, isp, ino);
+ fdP = IH_OPEN(salvinfo->VGLinkH);
/* Sync fake 1 link counts to the link table, now that it exists */
if (fdP) {
for (i = 0; i < nVols; i++) {
ip = allInodes + isp[i].index;
for (j = isp[i].nSpecialInodes; j < isp[i].nInodes; j++) {
-#ifdef AFS_NT40_ENV
- nt_SetLinkCount(fdP, ip[j].inodeNumber, 1, 1);
-#else
namei_SetLinkCount(fdP, ip[j].inodeNumber, 1, 1);
-#endif
}
}
}
if (fdP)
FDH_REALLYCLOSE(fdP);
#else
- IH_INIT(VGLinkH, fileSysDevice, -1, -1);
+ IH_INIT(salvinfo->VGLinkH, salvinfo->fileSysDevice, -1, -1);
#endif
/* Salvage in reverse order--read/write volume last; this way any
* way the whole RO volume can be deleted, below, if anything goes wrong */
for (check = 1; check >= 0; check--) {
int deleteMe;
- if (SalvageVolumeHeaderFile(lisp, allInodes, rw, check, &deleteMe)
+ if (SalvageVolumeHeaderFile(salvinfo, lisp, allInodes, rw, check, &deleteMe)
== -1) {
- MaybeZapVolume(lisp, "Volume header", deleteMe, check);
+ MaybeZapVolume(salvinfo, lisp, "Volume header", deleteMe, check);
if (rw && deleteMe) {
haveRWvolume = 0; /* This will cause its inodes to be deleted--since salvage
* volume won't be called */
}
if (rw && check == 1)
continue;
- if (SalvageVnodes(isp, lisp, allInodes, check) == -1) {
- MaybeZapVolume(lisp, "Vnode index", 0, check);
+ if (SalvageVnodes(salvinfo, isp, lisp, allInodes, check) == -1) {
+ MaybeZapVolume(salvinfo, lisp, "Vnode index", 0, check);
break;
}
}
/* Fix actual inode counts */
if (!Showmode) {
+ afs_ino_str_t stmp;
Log("totalInodes %d\n",totalInodes);
for (ip = inodes; totalInodes; ip++, totalInodes--) {
static int TraceBadLinkCounts = 0;
#ifdef AFS_NAMEI_ENV
- if (VGLinkH->ih_ino == ip->inodeNumber) {
- dec_VGLinkH = ip->linkCount - VGLinkH_cnt;
+ if (salvinfo->VGLinkH->ih_ino == ip->inodeNumber) {
+ dec_VGLinkH = ip->linkCount - salvinfo->VGLinkH_cnt;
VGLinkH_p1 = ip->u.param[0];
continue; /* Deal with this last. */
}
#endif
if (ip->linkCount != 0 && TraceBadLinkCounts) {
TraceBadLinkCounts--; /* Limit reports, per volume */
- Log("#### DEBUG #### Link count incorrect by %d; inode %s, size %llu, p=(%u,%u,%u,%u)\n", ip->linkCount, PrintInode(NULL, ip->inodeNumber), (afs_uintmax_t) ip->byteCount, ip->u.param[0], ip->u.param[1], ip->u.param[2], ip->u.param[3]);
+ Log("#### DEBUG #### Link count incorrect by %d; inode %s, size %llu, p=(%u,%u,%u,%u)\n", ip->linkCount, PrintInode(stmp, ip->inodeNumber), (afs_uintmax_t) ip->byteCount, ip->u.param[0], ip->u.param[1], ip->u.param[2], ip->u.param[3]);
}
while (ip->linkCount > 0) {
/* below used to assert, not break */
if (!Testing) {
- if (IH_DEC(VGLinkH, ip->inodeNumber, ip->u.param[0])) {
+ if (IH_DEC(salvinfo->VGLinkH, ip->inodeNumber, ip->u.param[0])) {
Log("idec failed. inode %s errno %d\n",
- PrintInode(NULL, ip->inodeNumber), errno);
+ PrintInode(stmp, ip->inodeNumber), errno);
break;
}
}
while (ip->linkCount < 0) {
/* these used to be asserts */
if (!Testing) {
- if (IH_INC(VGLinkH, ip->inodeNumber, ip->u.param[0])) {
+ if (IH_INC(salvinfo->VGLinkH, ip->inodeNumber, ip->u.param[0])) {
Log("iinc failed. inode %s errno %d\n",
- PrintInode(NULL, ip->inodeNumber), errno);
+ PrintInode(stmp, ip->inodeNumber), errno);
break;
}
}
}
#ifdef AFS_NAMEI_ENV
while (dec_VGLinkH > 0) {
- if (IH_DEC(VGLinkH, VGLinkH->ih_ino, VGLinkH_p1) < 0) {
+ if (IH_DEC(salvinfo->VGLinkH, salvinfo->VGLinkH->ih_ino, VGLinkH_p1) < 0) {
Log("idec failed on link table, errno = %d\n", errno);
}
dec_VGLinkH--;
}
while (dec_VGLinkH < 0) {
- if (IH_INC(VGLinkH, VGLinkH->ih_ino, VGLinkH_p1) < 0) {
+ if (IH_INC(salvinfo->VGLinkH, salvinfo->VGLinkH->ih_ino, VGLinkH_p1) < 0) {
Log("iinc failed on link table, errno = %d\n", errno);
}
dec_VGLinkH++;
free(inodes);
/* Directory consistency checks on the rw volume */
if (haveRWvolume)
- SalvageVolume(isp, VGLinkH);
- IH_RELEASE(VGLinkH);
+ SalvageVolume(salvinfo, isp, salvinfo->VGLinkH);
+ IH_RELEASE(salvinfo->VGLinkH);
if (canfork && !debug) {
ShowLog = 0;
}
int
-QuickCheck(register struct InodeSummary *isp, int nVols)
+QuickCheck(struct SalvInfo *salvinfo, struct InodeSummary *isp, int nVols)
{
/* Check headers BEFORE forking */
- register int i;
+ int i;
IHandle_t *h;
for (i = 0; i < nVols; i++) {
continue;
return 0;
}
- IH_INIT(h, fileSysDevice, vs->header.parent, vs->header.volumeInfo);
+ IH_INIT(h, salvinfo->fileSysDevice, vs->header.parent, vs->header.volumeInfo);
if (IH_IREAD(h, 0, (char *)&volHeader, sizeof(volHeader))
== sizeof(volHeader)
&& volHeader.stamp.magic == VOLUMEINFOMAGIC
*/
int
-SalvageVolumeHeaderFile(register struct InodeSummary *isp,
- register struct ViceInodeInfo *inodes, int RW,
+SalvageVolumeHeaderFile(struct SalvInfo *salvinfo, struct InodeSummary *isp,
+ struct ViceInodeInfo *inodes, int RW,
int check, int *deleteMe)
{
int i;
- register struct ViceInodeInfo *ip;
+ struct ViceInodeInfo *ip;
int allinodesobsolete = 1;
struct VolumeDiskHeader diskHeader;
afs_int32 (*writefunc)(VolumeDiskHeader_t *, struct DiskPartition64 *) = NULL;
int *skip;
+ struct VolumeHeader tempHeader;
+ struct afs_inode_info stuff[MAXINODETYPE];
/* keeps track of special inodes that are probably 'good'; they are
* referenced in the vol header, and are included in the given inodes
* if we detect duplicate special inodes */
}
+ init_inode_info(&tempHeader, stuff);
+
/*
* First, look at the special inodes and see if any are referenced by
* the existing volume header. If we find duplicate special inodes, we
}
}
for (i = 0; i < isp->nSpecialInodes; i++) {
+ afs_ino_str_t stmp;
ip = &inodes[isp->index + i];
if (ip->u.special.type <= 0 || ip->u.special.type > MAXINODETYPE) {
if (check) {
Log("Rubbish header inode %s of type %d\n",
- PrintInode(NULL, ip->inodeNumber),
+ PrintInode(stmp, ip->inodeNumber),
ip->u.special.type);
if (skip) {
free(skip);
return -1;
}
Log("Rubbish header inode %s of type %d; deleted\n",
- PrintInode(NULL, ip->inodeNumber),
+ PrintInode(stmp, ip->inodeNumber),
ip->u.special.type);
} else if (!stuff[ip->u.special.type - 1].obsolete) {
if (skip && skip[i]) {
if (orphans == ORPH_REMOVE) {
Log("Removing orphan special inode %s of type %d\n",
- PrintInode(NULL, ip->inodeNumber), ip->u.special.type);
+ PrintInode(stmp, ip->inodeNumber), ip->u.special.type);
continue;
} else {
Log("Ignoring orphan special inode %s of type %d\n",
- PrintInode(NULL, ip->inodeNumber), ip->u.special.type);
+ PrintInode(stmp, ip->inodeNumber), ip->u.special.type);
/* fall through to the ip->linkCount--; line below */
}
} else {
}
if (!check)
- VGLinkH_cnt++; /* one for every header. */
+ salvinfo->VGLinkH_cnt++; /* one for every header. */
if (!RW && !check && isp->volSummary) {
ClearROInUseBit(isp->volSummary);
if (stuff[i].inodeType == VI_LINKTABLE) {
/* Gross hack: SalvageHeader does a bcmp on the volume header.
* And we may have recreated the link table earlier, so set the
- * RW header as well.
+ * RW header as well. The header magic was already checked.
*/
- if (VALID_INO(VGLinkH->ih_ino)) {
- *stuff[i].inode = VGLinkH->ih_ino;
+ if (VALID_INO(salvinfo->VGLinkH->ih_ino)) {
+ *stuff[i].inode = salvinfo->VGLinkH->ih_ino;
}
continue;
}
- if (SalvageHeader(&stuff[i], isp, check, deleteMe) == -1 && check)
+ if (SalvageHeader(salvinfo, &stuff[i], isp, check, deleteMe) == -1 && check)
return -1;
}
if (isp->volSummary == NULL) {
char path[64];
char headerName[64];
- (void)afs_snprintf(headerName, sizeof headerName, VFORMAT, afs_printable_uint32_lu(isp->volumeId));
- (void)afs_snprintf(path, sizeof path, "%s/%s", fileSysPath, headerName);
+ snprintf(headerName, sizeof headerName, VFORMAT,
+ afs_printable_uint32_lu(isp->volumeId));
+ snprintf(path, sizeof path, "%s" OS_DIRSEP "%s",
+ salvinfo->fileSysPath, headerName);
if (check) {
Log("No header file for volume %u\n", isp->volumeId);
return -1;
Log("No header file for volume %u; %screating %s\n",
isp->volumeId, (Testing ? "it would have been " : ""),
path);
- isp->volSummary = (struct VolumeSummary *)
- malloc(sizeof(struct VolumeSummary));
+ isp->volSummary = calloc(1, sizeof(struct VolumeSummary));
isp->volSummary->fileName = ToString(headerName);
writefunc = VCreateVolumeDiskHeader;
if (isp->volSummary->fileName) {
strcpy(headerName, isp->volSummary->fileName);
} else {
- (void)afs_snprintf(headerName, sizeof headerName, VFORMAT, afs_printable_uint32_lu(isp->volumeId));
+ snprintf(headerName, sizeof headerName, VFORMAT,
+ afs_printable_uint32_lu(isp->volumeId));
isp->volSummary->fileName = ToString(headerName);
}
- (void)afs_snprintf(path, sizeof path, "%s/%s", fileSysPath, headerName);
+ snprintf(path, sizeof path, "%s" OS_DIRSEP "%s",
+ salvinfo->fileSysPath, headerName);
Log("Header file %s is damaged or no longer valid%s\n", path,
(check ? "" : "; repairing"));
} else {
afs_int32 code;
VolumeHeaderToDisk(&diskHeader, &tempHeader);
- code = (*writefunc)(&diskHeader, fileSysPartition);
+ code = (*writefunc)(&diskHeader, salvinfo->fileSysPartition);
if (code) {
Log("Error %ld writing volume header file for volume %lu\n",
afs_printable_int32_ld(code),
}
}
}
- IH_INIT(isp->volSummary->volumeInfoHandle, fileSysDevice, isp->RWvolumeId,
+ IH_INIT(isp->volSummary->volumeInfoHandle, salvinfo->fileSysDevice, isp->RWvolumeId,
isp->volSummary->header.volumeInfo);
return 0;
}
int
-SalvageHeader(register struct stuff *sp, struct InodeSummary *isp, int check,
- int *deleteMe)
+SalvageHeader(struct SalvInfo *salvinfo, struct afs_inode_info *sp,
+ struct InodeSummary *isp, int check, int *deleteMe)
{
union {
VolumeDiskData volumeInfo;
} header;
IHandle_t *specH;
int recreate = 0;
- afs_int32 code;
+ ssize_t nBytes;
FdHandle_t *fdP;
if (sp->obsolete)
return 0;
#ifndef AFS_NAMEI_ENV
if (sp->inodeType == VI_LINKTABLE)
- return 0;
+ return 0; /* header magic was already checked */
#endif
if (*(sp->inode) == 0) {
if (check) {
(Testing ? "it would have recreated it" : "recreating"));
if (!Testing) {
*(sp->inode) =
- IH_CREATE(NULL, fileSysDevice, fileSysPath, 0, isp->volumeId,
+ IH_CREATE(NULL, salvinfo->fileSysDevice, salvinfo->fileSysPath, 0, isp->volumeId,
INODESPECIAL, sp->inodeType, isp->RWvolumeId);
if (!VALID_INO(*(sp->inode)))
Abort
recreate = 1;
}
- IH_INIT(specH, fileSysDevice, isp->RWvolumeId, *(sp->inode));
+ IH_INIT(specH, salvinfo->fileSysDevice, isp->RWvolumeId, *(sp->inode));
fdP = IH_OPEN(specH);
if (OKToZap && (fdP == NULL) && BadError(errno)) {
/* bail out early and destroy the volume */
sp->description, errno);
if (!recreate
- && (FDH_READ(fdP, (char *)&header, sp->size) != sp->size
+ && (FDH_PREAD(fdP, (char *)&header, sp->size, 0) != sp->size
|| header.fileHeader.magic != sp->stamp.magic)) {
if (check) {
Log("Part of the header (%s) is corrupted\n", sp->description);
Log("Part of the header (%s) is corrupted; recreating\n",
sp->description);
recreate = 1;
+ /* header can be garbage; make sure we don't read garbage data from
+ * it below */
+ memset(&header, 0, sizeof(header));
}
if (sp->inodeType == VI_VOLINFO
&& header.volumeInfo.destroyMe == DESTROY_ME) {
Abort
("Internal error: recreating volume header (%s) in check mode\n",
sp->description);
- code = FDH_TRUNC(fdP, 0);
- if (code == -1)
+ nBytes = FDH_TRUNC(fdP, 0);
+ if (nBytes == -1)
Abort("Unable to truncate volume header file (%s) (error = %d)\n",
sp->description, errno);
header.volumeInfo.uniquifier = (isp->maxUniquifier + 1) + 1000;
header.volumeInfo.type = (isp->volumeId == isp->RWvolumeId ? readwriteVolume : readonlyVolume); /* XXXX */
header.volumeInfo.needsCallback = 0;
- gettimeofday(&tp, 0);
+ gettimeofday(&tp, NULL);
header.volumeInfo.creationDate = tp.tv_sec;
- if (FDH_SEEK(fdP, 0, SEEK_SET) < 0) {
- Abort
- ("Unable to seek to beginning of volume header file (%s) (errno = %d)\n",
- sp->description, errno);
- }
- code =
- FDH_WRITE(fdP, (char *)&header.volumeInfo,
- sizeof(header.volumeInfo));
- if (code != sizeof(header.volumeInfo)) {
- if (code < 0)
+ nBytes =
+ FDH_PWRITE(fdP, (char *)&header.volumeInfo,
+ sizeof(header.volumeInfo), 0);
+ if (nBytes != sizeof(header.volumeInfo)) {
+ if (nBytes < 0)
Abort
("Unable to write volume header file (%s) (errno = %d)\n",
sp->description, errno);
sp->description);
}
} else {
- if (FDH_SEEK(fdP, 0, SEEK_SET) < 0) {
- Abort
- ("Unable to seek to beginning of volume header file (%s) (errno = %d)\n",
- sp->description, errno);
- }
- code = FDH_WRITE(fdP, (char *)&sp->stamp, sizeof(sp->stamp));
- if (code != sizeof(sp->stamp)) {
- if (code < 0)
+ nBytes = FDH_PWRITE(fdP, (char *)&sp->stamp, sizeof(sp->stamp), 0);
+ if (nBytes != sizeof(sp->stamp)) {
+ if (nBytes < 0)
Abort
("Unable to write version stamp in volume header file (%s) (errno = %d)\n",
sp->description, errno);
FDH_REALLYCLOSE(fdP);
IH_RELEASE(specH);
if (sp->inodeType == VI_VOLINFO) {
- VolInfo = header.volumeInfo;
+ salvinfo->VolInfo = header.volumeInfo;
if (check) {
char update[25];
- if (VolInfo.updateDate) {
- strcpy(update, TimeStamp(VolInfo.updateDate, 0));
+
+ if (salvinfo->VolInfo.updateDate) {
+ strcpy(update, TimeStamp(salvinfo->VolInfo.updateDate, 0));
if (!Showmode)
- Log("%s (%u) %supdated %s\n", VolInfo.name, VolInfo.id,
+ Log("%s (%u) %supdated %s\n", salvinfo->VolInfo.name,
+ salvinfo->VolInfo.id,
(Testing ? "it would have been " : ""), update);
} else {
- strcpy(update, TimeStamp(VolInfo.creationDate, 0));
+ strcpy(update, TimeStamp(salvinfo->VolInfo.creationDate, 0));
if (!Showmode)
- Log("%s (%u) not updated (created %s)\n", VolInfo.name,
- VolInfo.id, update);
+ Log("%s (%u) not updated (created %s)\n",
+ salvinfo->VolInfo.name, salvinfo->VolInfo.id, update);
}
}
}
int
-SalvageVnodes(register struct InodeSummary *rwIsp,
- register struct InodeSummary *thisIsp,
- register struct ViceInodeInfo *inodes, int check)
+SalvageVnodes(struct SalvInfo *salvinfo,
+ struct InodeSummary *rwIsp,
+ struct InodeSummary *thisIsp,
+ struct ViceInodeInfo *inodes, int check)
{
int ilarge, ismall, ioffset, RW, nInodes;
ioffset = rwIsp->index + rwIsp->nSpecialInodes; /* first inode */
RW = (rwIsp == thisIsp);
nInodes = (rwIsp->nInodes - rwIsp->nSpecialInodes);
ismall =
- SalvageIndex(thisIsp->volSummary->header.smallVnodeIndex, vSmall, RW,
+ SalvageIndex(salvinfo, thisIsp->volSummary->header.smallVnodeIndex, vSmall, RW,
&inodes[ioffset], nInodes, thisIsp->volSummary, check);
if (check && ismall == -1)
return -1;
ilarge =
- SalvageIndex(thisIsp->volSummary->header.largeVnodeIndex, vLarge, RW,
+ SalvageIndex(salvinfo, thisIsp->volSummary->header.largeVnodeIndex, vLarge, RW,
&inodes[ioffset], nInodes, thisIsp->volSummary, check);
return (ilarge == 0 && ismall == 0 ? 0 : -1);
}
int
-SalvageIndex(Inode ino, VnodeClass class, int RW,
- register struct ViceInodeInfo *ip, int nInodes,
- struct VolumeSummary *volSummary, int check)
+SalvageIndex(struct SalvInfo *salvinfo, Inode ino, VnodeClass class, int RW,
+ struct ViceInodeInfo *ip, int nInodes,
+ struct VolumeSummary *volSummary, int check)
{
- VolumeId volumeNumber;
char buf[SIZEOF_LARGEDISKVNODE];
struct VnodeDiskObject *vnode = (struct VnodeDiskObject *)buf;
int err = 0;
StreamHandle_t *file;
struct VnodeClassInfo *vcp;
afs_sfsize_t size;
+ afs_sfsize_t nVnodes;
afs_fsize_t vnodeLength;
- int vnodeIndex, nVnodes;
+ int vnodeIndex;
afs_ino_str_t stmp1, stmp2;
IHandle_t *handle;
FdHandle_t *fdP;
- volumeNumber = volSummary->header.id;
- IH_INIT(handle, fileSysDevice, volSummary->header.parent, ino);
+ IH_INIT(handle, salvinfo->fileSysDevice, volSummary->header.parent, ino);
fdP = IH_OPEN(handle);
- assert(fdP != NULL);
+ osi_Assert(fdP != NULL);
file = FDH_FDOPEN(fdP, "r+");
- assert(file != NULL);
+ osi_Assert(file != NULL);
vcp = &VnodeClassInfo[class];
size = OS_SIZE(fdP->fd_fd);
- assert(size != -1);
+ osi_Assert(size != -1);
nVnodes = (size / vcp->diskSize) - 1;
if (nVnodes > 0) {
- assert((nVnodes + 1) * vcp->diskSize == size);
- assert(STREAM_SEEK(file, vcp->diskSize, 0) == 0);
+ osi_Assert((nVnodes + 1) * vcp->diskSize == size);
+ osi_Assert(STREAM_ASEEK(file, vcp->diskSize) == 0);
} else {
nVnodes = 0;
}
if (vnode->type != vNull) {
int vnodeChanged = 0;
int vnodeNumber = bitNumberToVnodeNumber(vnodeIndex, class);
- /* Log programs that belong to root (potentially suid root);
- * don't bother for read-only or backup volumes */
-#ifdef notdef /* This is done elsewhere */
- if (ShowRootFiles && RW && vnode->owner == 0 && vnodeNumber != 1)
- Log("OWNER IS ROOT %s %u dir %u vnode %u author %u owner %u mode %o\n", VolInfo.name, volumeNumber, vnode->parent, vnodeNumber, vnode->author, vnode->owner, vnode->modeBits);
-#endif
if (VNDISK_GET_INO(vnode) == 0) {
if (RW) {
/* Log("### DEBUG ### Deleted Vnode with 0 inode (vnode %d)\n", vnodeNumber); */
} else {
if (vcp->magic != vnode->vnodeMagic) {
/* bad magic #, probably partially created vnode */
+ if (check) {
+ Log("Partially allocated vnode %d: bad magic (is %lx should be %lx)\n",
+ vnodeNumber, afs_printable_uint32_lu(vnode->vnodeMagic),
+ afs_printable_uint32_lu(vcp->magic));
+ memset(vnode, 0, vcp->diskSize);
+ err = -1;
+ goto zooks;
+ }
Log("Partially allocated vnode %d deleted.\n",
vnodeNumber);
memset(vnode, 0, vcp->diskSize);
/* For RW volume, look for vnode with matching inode number;
* if no such match, take the first determined by our sort
* order */
- register struct ViceInodeInfo *lip = ip;
- register int lnInodes = nInodes;
+ struct ViceInodeInfo *lip = ip;
+ int lnInodes = nInodes;
while (lnInodes
&& lip->u.vnode.vnodeNumber == vnodeNumber) {
if (VNDISK_GET_INO(vnode) == lip->inodeNumber) {
/*
* Because of the possibility of the uniquifier overflows (> 4M)
* we compare them modulo the low 22-bits; we shouldn't worry
- * about mismatching since they shouldn't to many old
+ * about mismatching since they shouldn't to many old
* uniquifiers of the same vnode...
*/
if (IUnique(vu) != IUnique(iu)) {
ip++;
nInodes--;
} else { /* no matching inode */
+ afs_ino_str_t stmp;
if (VNDISK_GET_INO(vnode) != 0
|| vnode->type == vDirectory) {
/* No matching inode--get rid of the vnode */
if (check) {
if (VNDISK_GET_INO(vnode)) {
if (!Showmode) {
- Log("Vnode %d (unique %u): corresponding inode %s is missing\n", vnodeNumber, vnode->uniquifier, PrintInode(NULL, VNDISK_GET_INO(vnode)));
+ Log("Vnode %d (unique %u): corresponding inode %s is missing\n", vnodeNumber, vnode->uniquifier, PrintInode(stmp, VNDISK_GET_INO(vnode)));
}
} else {
if (!Showmode)
if (VNDISK_GET_INO(vnode)) {
if (!Showmode) {
time_t serverModifyTime = vnode->serverModifyTime;
- Log("Vnode %d (unique %u): corresponding inode %s is missing; vnode deleted, vnode mod time=%s", vnodeNumber, vnode->uniquifier, PrintInode(NULL, VNDISK_GET_INO(vnode)), ctime(&serverModifyTime));
+ Log("Vnode %d (unique %u): corresponding inode %s is missing; vnode deleted, vnode mod time=%s", vnodeNumber, vnode->uniquifier, PrintInode(stmp, VNDISK_GET_INO(vnode)), ctime(&serverModifyTime));
}
} else {
if (!Showmode) {
memset(vnode, 0, vcp->diskSize);
vnodeChanged = 1;
} else {
- /* Should not reach here becuase we checked for
+ /* Should not reach here becuase we checked for
* (inodeNumber == 0) above. And where we zero the vnode,
* we also goto vnodeDone.
*/
}
} /* VNDISK_GET_INO(vnode) != 0 */
vnodeDone:
- assert(!(vnodeChanged && check));
+ osi_Assert(!(vnodeChanged && check));
if (vnodeChanged && !Testing) {
- assert(IH_IWRITE
+ osi_Assert(IH_IWRITE
(handle, vnodeIndexOffset(vcp, vnodeNumber),
(char *)vnode, vcp->diskSize)
== vcp->diskSize);
- VolumeChanged = 1; /* For break call back */
+ salvinfo->VolumeChanged = 1; /* For break call back */
}
}
}
}
struct VnodeEssence *
-CheckVnodeNumber(VnodeId vnodeNumber)
+CheckVnodeNumber(struct SalvInfo *salvinfo, VnodeId vnodeNumber)
{
VnodeClass class;
struct VnodeInfo *vip;
int offset;
class = vnodeIdToClass(vnodeNumber);
- vip = &vnodeInfo[class];
+ vip = &salvinfo->vnodeInfo[class];
offset = vnodeIdToBitNumber(vnodeNumber);
return (offset >= vip->nVnodes ? NULL : &vip->vnodes[offset]);
}
void
-CopyOnWrite(register struct DirSummary *dir)
+CopyOnWrite(struct SalvInfo *salvinfo, struct DirSummary *dir)
{
/* Copy the directory unconditionally if we are going to change it:
* not just if was cloned.
DFlush(); /* Well justified paranoia... */
code =
- IH_IREAD(vnodeInfo[vLarge].handle,
+ IH_IREAD(salvinfo->vnodeInfo[vLarge].handle,
vnodeIndexOffset(vcp, dir->vnodeNumber), (char *)&vnode,
sizeof(vnode));
- assert(code == sizeof(vnode));
+ osi_Assert(code == sizeof(vnode));
oldinode = VNDISK_GET_INO(&vnode);
/* Increment the version number by a whole lot to avoid problems with
* clients that were promised new version numbers--but the file server
* crashed before the versions were written to disk.
*/
newinode =
- IH_CREATE(dir->ds_linkH, fileSysDevice, fileSysPath, 0, dir->rwVid,
+ IH_CREATE(dir->ds_linkH, salvinfo->fileSysDevice, salvinfo->fileSysPath, 0, dir->rwVid,
dir->vnodeNumber, vnode.uniquifier, vnode.dataVersion +=
200);
- assert(VALID_INO(newinode));
- assert(CopyInode(fileSysDevice, oldinode, newinode, dir->rwVid) == 0);
+ osi_Assert(VALID_INO(newinode));
+ osi_Assert(CopyInode(salvinfo->fileSysDevice, oldinode, newinode, dir->rwVid) == 0);
vnode.cloned = 0;
VNDISK_SET_INO(&vnode, newinode);
code =
- IH_IWRITE(vnodeInfo[vLarge].handle,
+ IH_IWRITE(salvinfo->vnodeInfo[vLarge].handle,
vnodeIndexOffset(vcp, dir->vnodeNumber), (char *)&vnode,
sizeof(vnode));
- assert(code == sizeof(vnode));
+ osi_Assert(code == sizeof(vnode));
SetSalvageDirHandle(&dir->dirHandle, dir->dirHandle.dirh_handle->ih_vid,
- fileSysDevice, newinode);
+ salvinfo->fileSysDevice, newinode,
+ &salvinfo->VolumeChanged);
/* Don't delete the original inode right away, because the directory is
* still being scanned.
*/
}
/*
- * This function should either successfully create a new dir, or give up
- * and leave things the way they were. In particular, if it fails to write
- * the new dir properly, it should return w/o changing the reference to the
+ * This function should either successfully create a new dir, or give up
+ * and leave things the way they were. In particular, if it fails to write
+ * the new dir properly, it should return w/o changing the reference to the
* old dir.
*/
void
-CopyAndSalvage(register struct DirSummary *dir)
+CopyAndSalvage(struct SalvInfo *salvinfo, struct DirSummary *dir)
{
struct VnodeDiskObject vnode;
struct VnodeClassInfo *vcp = &VnodeClassInfo[vLarge];
return;
Log("Salvaging directory %u...\n", dir->vnodeNumber);
lcode =
- IH_IREAD(vnodeInfo[vLarge].handle,
+ IH_IREAD(salvinfo->vnodeInfo[vLarge].handle,
vnodeIndexOffset(vcp, dir->vnodeNumber), (char *)&vnode,
sizeof(vnode));
- assert(lcode == sizeof(vnode));
+ osi_Assert(lcode == sizeof(vnode));
oldinode = VNDISK_GET_INO(&vnode);
/* Increment the version number by a whole lot to avoid problems with
* clients that were promised new version numbers--but the file server
* crashed before the versions were written to disk.
*/
newinode =
- IH_CREATE(dir->ds_linkH, fileSysDevice, fileSysPath, 0, dir->rwVid,
+ IH_CREATE(dir->ds_linkH, salvinfo->fileSysDevice, salvinfo->fileSysPath, 0, dir->rwVid,
dir->vnodeNumber, vnode.uniquifier, vnode.dataVersion +=
200);
- assert(VALID_INO(newinode));
- SetSalvageDirHandle(&newdir, dir->rwVid, fileSysDevice, newinode);
+ osi_Assert(VALID_INO(newinode));
+ SetSalvageDirHandle(&newdir, dir->rwVid, salvinfo->fileSysDevice, newinode,
+ &salvinfo->VolumeChanged);
- /* Assign . and .. vnode numbers from dir and vnode.parent.
+ /* Assign . and .. vnode numbers from dir and vnode.parent.
* The uniquifier for . is in the vnode.
- * The uniquifier for .. might be set to a bogus value of 1 and
+ * The uniquifier for .. might be set to a bogus value of 1 and
* the salvager will later clean it up.
*/
- if (vnode.parent && (vnodeEssence = CheckVnodeNumber(vnode.parent))) {
+ if (vnode.parent && (vnodeEssence = CheckVnodeNumber(salvinfo, vnode.parent))) {
parentUnique = (vnodeEssence->unique ? vnodeEssence->unique : 1);
}
code =
if (code) {
Log("also failed to decrement link count on new inode");
}
- assert(1 == 2);
+ osi_Assert(1 == 2);
}
Log("Checking the results of the directory salvage...\n");
if (!DirOK(&newdir)) {
Log("Directory salvage failed!!!; restoring old version of the directory.\n");
code = IH_DEC(dir->ds_linkH, newinode, dir->rwVid);
- assert(code == 0);
- assert(1 == 2);
+ osi_Assert(code == 0);
+ osi_Assert(1 == 2);
}
vnode.cloned = 0;
VNDISK_SET_INO(&vnode, newinode);
- length = Length(&newdir);
+ length = afs_dir_Length(&newdir);
VNDISK_SET_LEN(&vnode, length);
lcode =
- IH_IWRITE(vnodeInfo[vLarge].handle,
+ IH_IWRITE(salvinfo->vnodeInfo[vLarge].handle,
vnodeIndexOffset(vcp, dir->vnodeNumber), (char *)&vnode,
sizeof(vnode));
- assert(lcode == sizeof(vnode));
+ osi_Assert(lcode == sizeof(vnode));
#if 0
#ifdef AFS_NT40_ENV
- nt_sync(fileSysDevice);
+ nt_sync(salvinfo->fileSysDevice);
#else
sync(); /* this is slow, but hopefully rarely called. We don't have
* an open FD on the file itself to fsync.
*/
#endif
#else
- vnodeInfo[vLarge].handle->ih_synced = 1;
+ salvinfo->vnodeInfo[vLarge].handle->ih_synced = 1;
#endif
/* make sure old directory file is really closed */
fdP = IH_OPEN(dir->dirHandle.dirh_handle);
FDH_REALLYCLOSE(fdP);
-
+
code = IH_DEC(dir->ds_linkH, oldinode, dir->rwVid);
- assert(code == 0);
+ osi_Assert(code == 0);
dir->dirHandle = newdir;
}
+/**
+ * arguments for JudgeEntry.
+ */
+struct judgeEntry_params {
+ struct DirSummary *dir; /**< directory we're examining entries in */
+ struct SalvInfo *salvinfo; /**< SalvInfo for the current salvage job */
+};
+
int
-JudgeEntry(void *dirVal, char *name, afs_int32 vnodeNumber,
+JudgeEntry(void *arock, char *name, afs_int32 vnodeNumber,
afs_int32 unique)
{
- struct DirSummary *dir = (struct DirSummary *)dirVal;
+ struct judgeEntry_params *params = arock;
+ struct DirSummary *dir = params->dir;
+ struct SalvInfo *salvinfo = params->salvinfo;
struct VnodeEssence *vnodeEssence;
afs_int32 dirOrphaned, todelete;
- dirOrphaned = IsVnodeOrphaned(dir->vnodeNumber);
+ dirOrphaned = IsVnodeOrphaned(salvinfo, dir->vnodeNumber);
- vnodeEssence = CheckVnodeNumber(vnodeNumber);
+ vnodeEssence = CheckVnodeNumber(salvinfo, vnodeNumber);
if (vnodeEssence == NULL) {
if (!Showmode) {
- Log("dir vnode %u: invalid entry deleted: %s/%s (vnode %u, unique %u)\n", dir->vnodeNumber, dir->name ? dir->name : "??", name, vnodeNumber, unique);
+ Log("dir vnode %u: invalid entry deleted: %s" OS_DIRSEP "%s (vnode %u, unique %u)\n", dir->vnodeNumber, dir->name ? dir->name : "??", name, vnodeNumber, unique);
}
if (!Testing) {
- CopyOnWrite(dir);
- assert(Delete(&dir->dirHandle, name) == 0);
+ CopyOnWrite(salvinfo, dir);
+ osi_Assert(afs_dir_Delete(&dir->dirHandle, name) == 0);
}
return 0;
}
#ifdef AFS_AIX_ENV
#ifndef AFS_NAMEI_ENV
- /* On AIX machines, don't allow entries to point to inode 0. That is a special
+ /* On AIX machines, don't allow entries to point to inode 0. That is a special
* mount inode for the partition. If this inode were deleted, it would crash
* the machine.
*/
if (vnodeEssence->InodeNumber == 0) {
- Log("dir vnode %d: invalid entry: %s/%s has no inode (vnode %d, unique %d)%s\n", dir->vnodeNumber, (dir->name ? dir->name : "??"), name, vnodeNumber, unique, (Testing ? "-- would have deleted" : " -- deleted"));
+ Log("dir vnode %d: invalid entry: %s" OS_DIRSEP "%s has no inode (vnode %d, unique %d)%s\n", dir->vnodeNumber, (dir->name ? dir->name : "??"), name, vnodeNumber, unique, (Testing ? "-- would have deleted" : " -- deleted"));
if (!Testing) {
- CopyOnWrite(dir);
- assert(Delete(&dir->dirHandle, name) == 0);
+ CopyOnWrite(salvinfo, dir);
+ osi_Assert(Delete(&dir->dirHandle, name) == 0);
}
return 0;
}
if (!(vnodeNumber & 1) && !Showmode
&& !(vnodeEssence->count || vnodeEssence->unique
|| vnodeEssence->modeBits)) {
- Log("dir vnode %u: invalid entry: %s/%s (vnode %u, unique %u)%s\n",
+ Log("dir vnode %u: invalid entry: %s" OS_DIRSEP "%s (vnode %u, unique %u)%s\n",
dir->vnodeNumber, (dir->name ? dir->name : "??"), name,
vnodeNumber, unique,
((!unique) ? (Testing ? "-- would have deleted" : " -- deleted") :
""));
if (!unique) {
if (!Testing) {
- CopyOnWrite(dir);
- assert(Delete(&dir->dirHandle, name) == 0);
+ CopyOnWrite(salvinfo, dir);
+ osi_Assert(afs_dir_Delete(&dir->dirHandle, name) == 0);
}
return 0;
}
if (!vnodeEssence->unique
&& ((strcmp(name, "..") == 0) || (strcmp(name, ".") == 0))) {
/* This is an orphaned directory. Don't delete the . or ..
- * entry. Otherwise, it will get created in the next
+ * entry. Otherwise, it will get created in the next
* salvage and deleted again here. So Just skip it.
*/
return 0;
todelete = ((!vnodeEssence->unique || dirOrphaned) ? 1 : 0);
if (!Showmode) {
- Log("dir vnode %u: %s/%s (vnode %u): unique changed from %u to %u %s\n", dir->vnodeNumber, (dir->name ? dir->name : "??"), name, vnodeNumber, unique, vnodeEssence->unique, (!todelete ? "" : (Testing ? "-- would have deleted" : "-- deleted")));
+ Log("dir vnode %u: %s" OS_DIRSEP "%s (vnode %u): unique changed from %u to %u %s\n", dir->vnodeNumber, (dir->name ? dir->name : "??"), name, vnodeNumber, unique, vnodeEssence->unique, (!todelete ? "" : (Testing ? "-- would have deleted" : "-- deleted")));
}
if (!Testing) {
AFSFid fid;
fid.Vnode = vnodeNumber;
fid.Unique = vnodeEssence->unique;
- CopyOnWrite(dir);
- assert(Delete(&dir->dirHandle, name) == 0);
+ CopyOnWrite(salvinfo, dir);
+ osi_Assert(afs_dir_Delete(&dir->dirHandle, name) == 0);
if (!todelete)
- assert(Create(&dir->dirHandle, name, &fid) == 0);
+ osi_Assert(afs_dir_Create(&dir->dirHandle, name, &fid) == 0);
}
if (todelete)
return 0; /* no need to continue */
if (!Showmode)
Log("directory vnode %u.%u: bad '.' entry (was %u.%u); fixed\n", dir->vnodeNumber, dir->unique, vnodeNumber, unique);
if (!Testing) {
- CopyOnWrite(dir);
- assert(Delete(&dir->dirHandle, ".") == 0);
+ CopyOnWrite(salvinfo, dir);
+ osi_Assert(afs_dir_Delete(&dir->dirHandle, ".") == 0);
fid.Vnode = dir->vnodeNumber;
fid.Unique = dir->unique;
- assert(Create(&dir->dirHandle, ".", &fid) == 0);
+ osi_Assert(afs_dir_Create(&dir->dirHandle, ".", &fid) == 0);
}
vnodeNumber = fid.Vnode; /* Get the new Essence */
unique = fid.Unique;
- vnodeEssence = CheckVnodeNumber(vnodeNumber);
+ vnodeEssence = CheckVnodeNumber(salvinfo, vnodeNumber);
}
dir->haveDot = 1;
} else if (strcmp(name, "..") == 0) {
if (dir->parent) {
struct VnodeEssence *dotdot;
pa.Vnode = dir->parent;
- dotdot = CheckVnodeNumber(pa.Vnode);
- assert(dotdot != NULL); /* XXX Should not be assert */
+ dotdot = CheckVnodeNumber(salvinfo, pa.Vnode);
+ osi_Assert(dotdot != NULL); /* XXX Should not be assert */
pa.Unique = dotdot->unique;
} else {
pa.Vnode = dir->vnodeNumber;
if (!Showmode)
Log("directory vnode %u.%u: bad '..' entry (was %u.%u); fixed\n", dir->vnodeNumber, dir->unique, vnodeNumber, unique);
if (!Testing) {
- CopyOnWrite(dir);
- assert(Delete(&dir->dirHandle, "..") == 0);
- assert(Create(&dir->dirHandle, "..", &pa) == 0);
+ CopyOnWrite(salvinfo, dir);
+ osi_Assert(afs_dir_Delete(&dir->dirHandle, "..") == 0);
+ osi_Assert(afs_dir_Create(&dir->dirHandle, "..", &pa) == 0);
}
vnodeNumber = pa.Vnode; /* Get the new Essence */
unique = pa.Unique;
- vnodeEssence = CheckVnodeNumber(vnodeNumber);
+ vnodeEssence = CheckVnodeNumber(salvinfo, vnodeNumber);
}
dir->haveDotDot = 1;
} else if (strncmp(name, ".__afs", 6) == 0) {
Log("dir vnode %u: special old unlink-while-referenced file %s %s deleted (vnode %u)\n", dir->vnodeNumber, name, (Testing ? "would have been" : "is"), vnodeNumber);
}
if (!Testing) {
- CopyOnWrite(dir);
- assert(Delete(&dir->dirHandle, name) == 0);
+ CopyOnWrite(salvinfo, dir);
+ osi_Assert(afs_dir_Delete(&dir->dirHandle, name) == 0);
}
vnodeEssence->claimed = 0; /* Not claimed: Orphaned */
vnodeEssence->todelete = 1; /* Will later delete vnode and decr inode */
return 0;
} else {
if (ShowSuid && (vnodeEssence->modeBits & 06000))
- Log("FOUND suid/sgid file: %s/%s (%u.%u %05o) author %u (vnode %u dir %u)\n", dir->name ? dir->name : "??", name, vnodeEssence->owner, vnodeEssence->group, vnodeEssence->modeBits, vnodeEssence->author, vnodeNumber, dir->vnodeNumber);
+ Log("FOUND suid/sgid file: %s" OS_DIRSEP "%s (%u.%u %05o) author %u (vnode %u dir %u)\n", dir->name ? dir->name : "??", name, vnodeEssence->owner, vnodeEssence->group, vnodeEssence->modeBits, vnodeEssence->author, vnodeNumber, dir->vnodeNumber);
if (/* ShowMounts && */ (vnodeEssence->type == vSymlink)
&& !(vnodeEssence->modeBits & 0111)) {
- int code, size;
+ afs_sfsize_t nBytes;
+ afs_sfsize_t size;
char buf[1025];
IHandle_t *ihP;
FdHandle_t *fdP;
- IH_INIT(ihP, fileSysDevice, dir->dirHandle.dirh_handle->ih_vid,
+ IH_INIT(ihP, salvinfo->fileSysDevice, dir->dirHandle.dirh_handle->ih_vid,
vnodeEssence->InodeNumber);
fdP = IH_OPEN(ihP);
if (fdP == NULL) {
}
size = FDH_SIZE(fdP);
if (size < 0) {
- Log("ERROR %s mount point has invalid size %d, vnode %u\n", dir->vname, size, vnodeNumber);
+ Log("ERROR %s mount point has invalid size %d, vnode %u\n", dir->vname, (int)size, vnodeNumber);
FDH_REALLYCLOSE(fdP);
IH_RELEASE(ihP);
return 0;
}
-
+
if (size > 1024)
size = 1024;
- code = FDH_READ(fdP, buf, size);
- if (code == size) {
+ nBytes = FDH_PREAD(fdP, buf, size, 0);
+ if (nBytes == size) {
buf[size] = '\0';
if ( (*buf != '#' && *buf != '%') || buf[strlen(buf)-1] != '.' ) {
- Log("Volume %u (%s) mount point %s/%s to '%s' invalid, %s to symbolic link\n",
+ Log("Volume %u (%s) mount point %s" OS_DIRSEP "%s to '%s' invalid, %s to symbolic link\n",
dir->dirHandle.dirh_handle->ih_vid, dir->vname, dir->name ? dir->name : "??", name, buf,
Testing ? "would convert" : "converted");
vnodeEssence->modeBits |= 0111;
vnodeEssence->changed = 1;
- } else if (ShowMounts) Log("In volume %u (%s) found mountpoint %s/%s to '%s'\n",
+ } else if (ShowMounts) Log("In volume %u (%s) found mountpoint %s" OS_DIRSEP "%s to '%s'\n",
dir->dirHandle.dirh_handle->ih_vid, dir->vname,
dir->name ? dir->name : "??", name, buf);
} else {
Log("Volume %s cound not read mount point vnode %u size %d code %d\n",
- dir->vname, vnodeNumber, size, code);
+ dir->vname, vnodeNumber, (int)size, (int)nBytes);
}
FDH_REALLYCLOSE(fdP);
IH_RELEASE(ihP);
}
if (ShowRootFiles && vnodeEssence->owner == 0 && vnodeNumber != 1)
- Log("FOUND root file: %s/%s (%u.%u %05o) author %u (vnode %u dir %u)\n", dir->name ? dir->name : "??", name, vnodeEssence->owner, vnodeEssence->group, vnodeEssence->modeBits, vnodeEssence->author, vnodeNumber, dir->vnodeNumber);
+ Log("FOUND root file: %s" OS_DIRSEP "%s (%u.%u %05o) author %u (vnode %u dir %u)\n", dir->name ? dir->name : "??", name, vnodeEssence->owner, vnodeEssence->group, vnodeEssence->modeBits, vnodeEssence->author, vnodeNumber, dir->vnodeNumber);
if (vnodeIdToClass(vnodeNumber) == vLarge
&& vnodeEssence->name == NULL) {
char *n;
}
/* The directory entry points to the vnode. Check to see if the
- * vnode points back to the directory. If not, then let the
- * directory claim it (else it might end up orphaned). Vnodes
+ * vnode points back to the directory. If not, then let the
+ * directory claim it (else it might end up orphaned). Vnodes
* already claimed by another directory are deleted from this
* directory: hardlinks to the same vnode are not allowed
* from different directories.
* another non-orphaned dir).
*/
if (!Showmode) {
- Log("dir vnode %u: %s/%s (vnode %u, unique %u) -- parent vnode %schanged from %u to %u\n", dir->vnodeNumber, (dir->name ? dir->name : "??"), name, vnodeNumber, unique, (Testing ? "would have been " : ""), vnodeEssence->parent, dir->vnodeNumber);
+ Log("dir vnode %u: %s" OS_DIRSEP "%s (vnode %u, unique %u) -- parent vnode %schanged from %u to %u\n", dir->vnodeNumber, (dir->name ? dir->name : "??"), name, vnodeNumber, unique, (Testing ? "would have been " : ""), vnodeEssence->parent, dir->vnodeNumber);
}
vnodeEssence->parent = dir->vnodeNumber;
vnodeEssence->changed = 1;
/* Vnode was claimed by another directory */
if (!Showmode) {
if (dirOrphaned) {
- Log("dir vnode %u: %s/%s parent vnode is %u (vnode %u, unique %u) -- %sdeleted\n", dir->vnodeNumber, (dir->name ? dir->name : "??"), name, vnodeEssence->parent, vnodeNumber, unique, (Testing ? "would have been " : ""));
+ Log("dir vnode %u: %s" OS_DIRSEP "%s parent vnode is %u (vnode %u, unique %u) -- %sdeleted\n", dir->vnodeNumber, (dir->name ? dir->name : "??"), name, vnodeEssence->parent, vnodeNumber, unique, (Testing ? "would have been " : ""));
} else if (vnodeNumber == 1) {
- Log("dir vnode %d: %s/%s is invalid (vnode %d, unique %d) -- %sdeleted\n", dir->vnodeNumber, (dir->name ? dir->name : "??"), name, vnodeNumber, unique, (Testing ? "would have been " : ""));
+ Log("dir vnode %d: %s" OS_DIRSEP "%s is invalid (vnode %d, unique %d) -- %sdeleted\n", dir->vnodeNumber, (dir->name ? dir->name : "??"), name, vnodeNumber, unique, (Testing ? "would have been " : ""));
} else {
- Log("dir vnode %u: %s/%s already claimed by directory vnode %u (vnode %u, unique %u) -- %sdeleted\n", dir->vnodeNumber, (dir->name ? dir->name : "??"), name, vnodeEssence->parent, vnodeNumber, unique, (Testing ? "would have been " : ""));
+ Log("dir vnode %u: %s" OS_DIRSEP "%s already claimed by directory vnode %u (vnode %u, unique %u) -- %sdeleted\n", dir->vnodeNumber, (dir->name ? dir->name : "??"), name, vnodeEssence->parent, vnodeNumber, unique, (Testing ? "would have been " : ""));
}
}
if (!Testing) {
- CopyOnWrite(dir);
- assert(Delete(&dir->dirHandle, name) == 0);
+ CopyOnWrite(salvinfo, dir);
+ osi_Assert(afs_dir_Delete(&dir->dirHandle, name) == 0);
}
return 0;
}
}
void
-DistilVnodeEssence(VolumeId rwVId, VnodeClass class, Inode ino, Unique * maxu)
+DistilVnodeEssence(struct SalvInfo *salvinfo, VolumeId rwVId,
+ VnodeClass class, Inode ino, Unique * maxu)
{
- register struct VnodeInfo *vip = &vnodeInfo[class];
+ struct VnodeInfo *vip = &salvinfo->vnodeInfo[class];
struct VnodeClassInfo *vcp = &VnodeClassInfo[class];
char buf[SIZEOF_LARGEDISKVNODE];
struct VnodeDiskObject *vnode = (struct VnodeDiskObject *)buf;
int nVnodes;
FdHandle_t *fdP;
- IH_INIT(vip->handle, fileSysDevice, rwVId, ino);
+ IH_INIT(vip->handle, salvinfo->fileSysDevice, rwVId, ino);
fdP = IH_OPEN(vip->handle);
- assert(fdP != NULL);
+ osi_Assert(fdP != NULL);
file = FDH_FDOPEN(fdP, "r+");
- assert(file != NULL);
+ osi_Assert(file != NULL);
size = OS_SIZE(fdP->fd_fd);
- assert(size != -1);
+ osi_Assert(size != -1);
vip->nVnodes = (size / vcp->diskSize) - 1;
if (vip->nVnodes > 0) {
- assert((vip->nVnodes + 1) * vcp->diskSize == size);
- assert(STREAM_SEEK(file, vcp->diskSize, 0) == 0);
- assert((vip->vnodes = (struct VnodeEssence *)
+ osi_Assert((vip->nVnodes + 1) * vcp->diskSize == size);
+ osi_Assert(STREAM_ASEEK(file, vcp->diskSize) == 0);
+ osi_Assert((vip->vnodes = (struct VnodeEssence *)
calloc(vip->nVnodes, sizeof(struct VnodeEssence))) != NULL);
if (class == vLarge) {
- assert((vip->inodes = (Inode *)
+ osi_Assert((vip->inodes = (Inode *)
calloc(vip->nVnodes, sizeof(Inode))) != NULL);
} else {
vip->inodes = NULL;
nVnodes && STREAM_READ(vnode, vcp->diskSize, 1, file) == 1;
nVnodes--, vnodeIndex++) {
if (vnode->type != vNull) {
- register struct VnodeEssence *vep = &vip->vnodes[vnodeIndex];
+ struct VnodeEssence *vep = &vip->vnodes[vnodeIndex];
afs_fsize_t vnodeLength;
vip->nAllocatedVnodes++;
vep->count = vnode->linkCount;
VnodeId vnodeNumber = bitNumberToVnodeNumber(vnodeIndex, class);
vip->nAllocatedVnodes--;
memset(vnode, 0, sizeof(vnode));
- IH_IWRITE(vnodeInfo[vSmall].handle,
+ IH_IWRITE(salvinfo->vnodeInfo[vSmall].handle,
vnodeIndexOffset(vcp, vnodeNumber),
(char *)&vnode, sizeof(vnode));
- VolumeChanged = 1;
+ salvinfo->VolumeChanged = 1;
} else
vip->inodes[vnodeIndex] = VNDISK_GET_INO(vnode);
}
}
static char *
-GetDirName(VnodeId vnode, struct VnodeEssence *vp, char *path)
+GetDirName(struct SalvInfo *salvinfo, VnodeId vnode, struct VnodeEssence *vp,
+ char *path)
{
struct VnodeEssence *parentvp;
strcpy(path, ".");
return path;
}
- if (vp->parent && vp->name && (parentvp = CheckVnodeNumber(vp->parent))
- && GetDirName(vp->parent, parentvp, path)) {
- strcat(path, "/");
+ if (vp->parent && vp->name && (parentvp = CheckVnodeNumber(salvinfo, vp->parent))
+ && GetDirName(salvinfo, vp->parent, parentvp, path)) {
+ strcat(path, OS_DIRSEP);
strcat(path, vp->name);
return path;
}
* vnodes must be "claimed". The vep->claimed flag is set in JudgeEntry().
*/
static int
-IsVnodeOrphaned(VnodeId vnode)
+IsVnodeOrphaned(struct SalvInfo *salvinfo, VnodeId vnode)
{
struct VnodeEssence *vep;
return (1); /* Vnode zero does not exist */
if (vnode == 1)
return (0); /* The root dir vnode is always claimed */
- vep = CheckVnodeNumber(vnode); /* Get the vnode essence */
+ vep = CheckVnodeNumber(salvinfo, vnode); /* Get the vnode essence */
if (!vep || !vep->claimed)
return (1); /* Vnode is not claimed - it is orphaned */
- return (IsVnodeOrphaned(vep->parent));
+ return (IsVnodeOrphaned(salvinfo, vep->parent));
}
void
-SalvageDir(char *name, VolumeId rwVid, struct VnodeInfo *dirVnodeInfo,
- IHandle_t * alinkH, int i, struct DirSummary *rootdir,
- int *rootdirfound)
+SalvageDir(struct SalvInfo *salvinfo, char *name, VolumeId rwVid,
+ struct VnodeInfo *dirVnodeInfo, IHandle_t * alinkH, int i,
+ struct DirSummary *rootdir, int *rootdirfound)
{
static struct DirSummary dir;
static struct DirHandle dirHandle;
dirVnodeInfo->vnodes[i].changed = 1;
}
} else {
- parent = CheckVnodeNumber(dirVnodeInfo->vnodes[i].parent);
+ parent = CheckVnodeNumber(salvinfo, dirVnodeInfo->vnodes[i].parent);
if (parent && parent->salvaged == 0)
- SalvageDir(name, rwVid, dirVnodeInfo, alinkH,
+ SalvageDir(salvinfo, name, rwVid, dirVnodeInfo, alinkH,
vnodeIdToBitNumber(dirVnodeInfo->vnodes[i].parent),
rootdir, rootdirfound);
}
dir.parent = dirVnodeInfo->vnodes[i].parent;
dir.haveDot = dir.haveDotDot = 0;
dir.ds_linkH = alinkH;
- SetSalvageDirHandle(&dir.dirHandle, dir.rwVid, fileSysDevice,
- dirVnodeInfo->inodes[i]);
+ SetSalvageDirHandle(&dir.dirHandle, dir.rwVid, salvinfo->fileSysDevice,
+ dirVnodeInfo->inodes[i], &salvinfo->VolumeChanged);
dirok = ((RebuildDirs && !Testing) ? 0 : DirOK(&dir.dirHandle));
if (!dirok) {
(Testing ? "skipping" : "salvaging"));
}
if (!Testing) {
- CopyAndSalvage(&dir);
+ CopyAndSalvage(salvinfo, &dir);
dirok = 1;
+ dirVnodeInfo->inodes[i] = dir.dirHandle.dirh_inode;
}
}
dirHandle = dir.dirHandle;
dir.name =
- GetDirName(bitNumberToVnodeNumber(i, vLarge),
+ GetDirName(salvinfo, bitNumberToVnodeNumber(i, vLarge),
&dirVnodeInfo->vnodes[i], path);
if (dirok) {
/* If enumeration failed for random reasons, we will probably delete
* too much stuff, so we guard against this instead.
*/
- assert(EnumerateDir(&dirHandle, JudgeEntry, &dir) == 0);
+ struct judgeEntry_params judge_params;
+ judge_params.salvinfo = salvinfo;
+ judge_params.dir = &dir;
+
+ osi_Assert(afs_dir_EnumerateDir(&dirHandle, JudgeEntry,
+ &judge_params) == 0);
}
/* Delete the old directory if it was copied in order to salvage.
DFlush();
if (dir.copied && !Testing) {
code = IH_DEC(dir.ds_linkH, dirHandle.dirh_handle->ih_ino, rwVid);
- assert(code == 0);
+ osi_Assert(code == 0);
dirVnodeInfo->inodes[i] = dir.dirHandle.dirh_inode;
}
return;
}
+/**
+ * Get a new FID that can be used to create a new file.
+ *
+ * @param[in] volHeader vol header for the volume
+ * @param[in] class what type of vnode we'll be creating (vLarge or vSmall)
+ * @param[out] afid the FID that we can use (only Vnode and Unique are set)
+ * @param[inout] maxunique max uniquifier for all vnodes in the volume;
+ * updated to the new max unique if we create a new
+ * vnode
+ */
+static void
+GetNewFID(struct SalvInfo *salvinfo, VolumeDiskData *volHeader,
+ VnodeClass class, AFSFid *afid, Unique *maxunique)
+{
+ int i;
+ for (i = 0; i < salvinfo->vnodeInfo[class].nVnodes; i++) {
+ if (salvinfo->vnodeInfo[class].vnodes[i].type == vNull) {
+ break;
+ }
+ }
+ if (i == salvinfo->vnodeInfo[class].nVnodes) {
+ /* no free vnodes; make a new one */
+ salvinfo->vnodeInfo[class].nVnodes++;
+ salvinfo->vnodeInfo[class].vnodes =
+ realloc(salvinfo->vnodeInfo[class].vnodes,
+ sizeof(struct VnodeEssence) * (i+1));
+
+ salvinfo->vnodeInfo[class].vnodes[i].type = vNull;
+ }
+
+ afid->Vnode = bitNumberToVnodeNumber(i, class);
+
+ if (volHeader->uniquifier < (*maxunique + 1)) {
+ /* header uniq is bad; it will get bumped by 2000 later */
+ afid->Unique = *maxunique + 1 + 2000;
+ (*maxunique)++;
+ } else {
+ /* header uniq seems okay; just use that */
+ afid->Unique = *maxunique = volHeader->uniquifier++;
+ }
+}
+
+/**
+ * Create a vnode for a README file explaining not to use a recreated-root vol.
+ *
+ * @param[in] volHeader vol header for the volume
+ * @param[in] alinkH ihandle for i/o for the volume
+ * @param[in] vid volume id
+ * @param[inout] maxunique max uniquifier for all vnodes in the volume;
+ * updated to the new max unique if we create a new
+ * vnode
+ * @param[out] afid FID for the new readme vnode
+ * @param[out] ainode the inode for the new readme file
+ *
+ * @return operation status
+ * @retval 0 success
+ * @retval -1 error
+ */
+static int
+CreateReadme(struct SalvInfo *salvinfo, VolumeDiskData *volHeader,
+ IHandle_t *alinkH, VolumeId vid, Unique *maxunique, AFSFid *afid,
+ Inode *ainode)
+{
+ Inode readmeinode;
+ struct VnodeDiskObject *rvnode = NULL;
+ afs_sfsize_t bytes;
+ IHandle_t *readmeH = NULL;
+ struct VnodeEssence *vep;
+ afs_fsize_t length;
+ time_t now = time(NULL);
+
+ /* Try to make the note brief, but informative. Only administrators should
+ * be able to read this file at first, so we can hopefully assume they
+ * know what AFS is, what a volume is, etc. */
+ char readme[] =
+"This volume has been salvaged, but has lost its original root directory.\n"
+"The root directory that exists now has been recreated from orphan files\n"
+"from the rest of the volume. This recreated root directory may interfere\n"
+"with old cached data on clients, and there is no way the salvager can\n"
+"reasonably prevent that. So, it is recommended that you do not continue to\n"
+"use this volume, but only copy the salvaged data to a new volume.\n"
+"Continuing to use this volume as it exists now may cause some clients to\n"
+"behave oddly when accessing this volume.\n"
+"\n\t -- Your friendly neighborhood OpenAFS salvager\n";
+ /* ^ the person reading this probably just lost some data, so they could
+ * use some cheering up. */
+
+ /* -1 for the trailing NUL */
+ length = sizeof(readme) - 1;
+
+ GetNewFID(salvinfo, volHeader, vSmall, afid, maxunique);
+
+ vep = &salvinfo->vnodeInfo[vSmall].vnodes[vnodeIdToBitNumber(afid->Vnode)];
+
+ /* create the inode and write the contents */
+ readmeinode = IH_CREATE(alinkH, salvinfo->fileSysDevice,
+ salvinfo->fileSysPath, 0, vid,
+ afid->Vnode, afid->Unique, 1);
+ if (!VALID_INO(readmeinode)) {
+ Log("CreateReadme: readme IH_CREATE failed\n");
+ goto error;
+ }
+
+ IH_INIT(readmeH, salvinfo->fileSysDevice, vid, readmeinode);
+ bytes = IH_IWRITE(readmeH, 0, readme, length);
+ IH_RELEASE(readmeH);
+
+ if (bytes != length) {
+ Log("CreateReadme: IWRITE failed (%d/%d)\n", (int)bytes,
+ (int)sizeof(readme));
+ goto error;
+ }
+
+ /* create the vnode and write it out */
+ rvnode = calloc(1, SIZEOF_SMALLDISKVNODE);
+ if (!rvnode) {
+ Log("CreateRootDir: error alloc'ing memory\n");
+ goto error;
+ }
+
+ rvnode->type = vFile;
+ rvnode->cloned = 0;
+ rvnode->modeBits = 0777;
+ rvnode->linkCount = 1;
+ VNDISK_SET_LEN(rvnode, length);
+ rvnode->uniquifier = afid->Unique;
+ rvnode->dataVersion = 1;
+ VNDISK_SET_INO(rvnode, readmeinode);
+ rvnode->unixModifyTime = rvnode->serverModifyTime = now;
+ rvnode->author = 0;
+ rvnode->owner = 0;
+ rvnode->parent = 1;
+ rvnode->group = 0;
+ rvnode->vnodeMagic = VnodeClassInfo[vSmall].magic;
+
+ bytes = IH_IWRITE(salvinfo->vnodeInfo[vSmall].handle,
+ vnodeIndexOffset(&VnodeClassInfo[vSmall], afid->Vnode),
+ (char*)rvnode, SIZEOF_SMALLDISKVNODE);
+
+ if (bytes != SIZEOF_SMALLDISKVNODE) {
+ Log("CreateReadme: IH_IWRITE failed (%d/%d)\n", (int)bytes,
+ (int)SIZEOF_SMALLDISKVNODE);
+ goto error;
+ }
+
+ /* update VnodeEssence for new readme vnode */
+ salvinfo->vnodeInfo[vSmall].nAllocatedVnodes++;
+ vep->count = 0;
+ vep->blockCount = nBlocks(length);
+ salvinfo->vnodeInfo[vSmall].volumeBlockCount += vep->blockCount;
+ vep->parent = rvnode->parent;
+ vep->unique = rvnode->uniquifier;
+ vep->modeBits = rvnode->modeBits;
+ vep->InodeNumber = VNDISK_GET_INO(rvnode);
+ vep->type = rvnode->type;
+ vep->author = rvnode->author;
+ vep->owner = rvnode->owner;
+ vep->group = rvnode->group;
+
+ free(rvnode);
+ rvnode = NULL;
+
+ vep->claimed = 1;
+ vep->changed = 0;
+ vep->salvaged = 1;
+ vep->todelete = 0;
+
+ *ainode = readmeinode;
+
+ return 0;
+
+ error:
+ if (IH_DEC(alinkH, readmeinode, vid)) {
+ Log("CreateReadme (recovery): IH_DEC failed\n");
+ }
+
+ if (rvnode) {
+ free(rvnode);
+ rvnode = NULL;
+ }
+
+ return -1;
+}
+
+/**
+ * create a root dir for a volume that lacks one.
+ *
+ * @param[in] volHeader vol header for the volume
+ * @param[in] alinkH ihandle for disk access for this volume group
+ * @param[in] vid volume id we're dealing with
+ * @param[out] rootdir populated with info about the new root dir
+ * @param[inout] maxunique max uniquifier for all vnodes in the volume;
+ * updated to the new max unique if we create a new
+ * vnode
+ *
+ * @return operation status
+ * @retval 0 success
+ * @retval -1 error
+ */
+static int
+CreateRootDir(struct SalvInfo *salvinfo, VolumeDiskData *volHeader,
+ IHandle_t *alinkH, VolumeId vid, struct DirSummary *rootdir,
+ Unique *maxunique)
+{
+ FileVersion dv;
+ int decroot = 0, decreadme = 0;
+ AFSFid did, readmeid;
+ afs_fsize_t length;
+ Inode rootinode;
+ struct VnodeDiskObject *rootvnode = NULL;
+ struct acl_accessList *ACL;
+ Inode *ip;
+ afs_sfsize_t bytes;
+ struct VnodeEssence *vep;
+ Inode readmeinode;
+ time_t now = time(NULL);
+
+ if (!salvinfo->vnodeInfo[vLarge].vnodes && !salvinfo->vnodeInfo[vSmall].vnodes) {
+ Log("Not creating new root dir; volume appears to lack any vnodes\n");
+ goto error;
+ }
+
+ if (!salvinfo->vnodeInfo[vLarge].vnodes) {
+ /* We don't have any large vnodes in the volume; allocate room
+ * for one so we can recreate the root dir */
+ salvinfo->vnodeInfo[vLarge].nVnodes = 1;
+ salvinfo->vnodeInfo[vLarge].vnodes = calloc(1, sizeof(struct VnodeEssence));
+ salvinfo->vnodeInfo[vLarge].inodes = calloc(1, sizeof(Inode));
+
+ osi_Assert(salvinfo->vnodeInfo[vLarge].vnodes);
+ osi_Assert(salvinfo->vnodeInfo[vLarge].inodes);
+ }
+
+ vep = &salvinfo->vnodeInfo[vLarge].vnodes[vnodeIdToBitNumber(1)];
+ ip = &salvinfo->vnodeInfo[vLarge].inodes[vnodeIdToBitNumber(1)];
+ if (vep->type != vNull) {
+ Log("Not creating new root dir; existing vnode 1 is non-null\n");
+ goto error;
+ }
+
+ if (CreateReadme(salvinfo, volHeader, alinkH, vid, maxunique, &readmeid,
+ &readmeinode) != 0) {
+ goto error;
+ }
+ decreadme = 1;
+
+ /* set the DV to a very high number, so it is unlikely that we collide
+ * with a cached DV */
+ dv = 1 << 30;
+
+ rootinode = IH_CREATE(alinkH, salvinfo->fileSysDevice, salvinfo->fileSysPath,
+ 0, vid, 1, 1, dv);
+ if (!VALID_INO(rootinode)) {
+ Log("CreateRootDir: IH_CREATE failed\n");
+ goto error;
+ }
+ decroot = 1;
+
+ SetSalvageDirHandle(&rootdir->dirHandle, vid, salvinfo->fileSysDevice,
+ rootinode, &salvinfo->VolumeChanged);
+ did.Volume = vid;
+ did.Vnode = 1;
+ did.Unique = 1;
+ if (afs_dir_MakeDir(&rootdir->dirHandle, (afs_int32*)&did, (afs_int32*)&did)) {
+ Log("CreateRootDir: MakeDir failed\n");
+ goto error;
+ }
+ if (afs_dir_Create(&rootdir->dirHandle, "README.ROOTDIR", &readmeid)) {
+ Log("CreateRootDir: Create failed\n");
+ goto error;
+ }
+ DFlush();
+ length = afs_dir_Length(&rootdir->dirHandle);
+ DZap(&rootdir->dirHandle);
+
+ /* create the new root dir vnode */
+ rootvnode = calloc(1, SIZEOF_LARGEDISKVNODE);
+ if (!rootvnode) {
+ Log("CreateRootDir: malloc failed\n");
+ goto error;
+ }
+
+ /* only give 'rl' permissions to 'system:administrators'. We do this to
+ * try to catch the attention of an administrator, that they should not
+ * be writing to this directory or continue to use it. */
+ ACL = VVnodeDiskACL(rootvnode);
+ ACL->size = sizeof(struct acl_accessList);
+ ACL->version = ACL_ACLVERSION;
+ ACL->total = 1;
+ ACL->positive = 1;
+ ACL->negative = 0;
+ ACL->entries[0].id = -204; /* system:administrators */
+ ACL->entries[0].rights = PRSFS_READ | PRSFS_LOOKUP;
+
+ rootvnode->type = vDirectory;
+ rootvnode->cloned = 0;
+ rootvnode->modeBits = 0777;
+ rootvnode->linkCount = 2;
+ VNDISK_SET_LEN(rootvnode, length);
+ rootvnode->uniquifier = 1;
+ rootvnode->dataVersion = dv;
+ VNDISK_SET_INO(rootvnode, rootinode);
+ rootvnode->unixModifyTime = rootvnode->serverModifyTime = now;
+ rootvnode->author = 0;
+ rootvnode->owner = 0;
+ rootvnode->parent = 0;
+ rootvnode->group = 0;
+ rootvnode->vnodeMagic = VnodeClassInfo[vLarge].magic;
+
+ /* write it out to disk */
+ bytes = IH_IWRITE(salvinfo->vnodeInfo[vLarge].handle,
+ vnodeIndexOffset(&VnodeClassInfo[vLarge], 1),
+ (char*)rootvnode, SIZEOF_LARGEDISKVNODE);
+
+ if (bytes != SIZEOF_LARGEDISKVNODE) {
+ /* just cast to int and don't worry about printing real 64-bit ints;
+ * a large disk vnode isn't anywhere near the 32-bit limit */
+ Log("CreateRootDir: IH_IWRITE failed (%d/%d)\n", (int)bytes,
+ (int)SIZEOF_LARGEDISKVNODE);
+ goto error;
+ }
+
+ /* update VnodeEssence for the new root vnode */
+ salvinfo->vnodeInfo[vLarge].nAllocatedVnodes++;
+ vep->count = 0;
+ vep->blockCount = nBlocks(length);
+ salvinfo->vnodeInfo[vLarge].volumeBlockCount += vep->blockCount;
+ vep->parent = rootvnode->parent;
+ vep->unique = rootvnode->uniquifier;
+ vep->modeBits = rootvnode->modeBits;
+ vep->InodeNumber = VNDISK_GET_INO(rootvnode);
+ vep->type = rootvnode->type;
+ vep->author = rootvnode->author;
+ vep->owner = rootvnode->owner;
+ vep->group = rootvnode->group;
+
+ free(rootvnode);
+ rootvnode = NULL;
+
+ vep->claimed = 0;
+ vep->changed = 0;
+ vep->salvaged = 1;
+ vep->todelete = 0;
+
+ /* update DirSummary for the new root vnode */
+ rootdir->vnodeNumber = 1;
+ rootdir->unique = 1;
+ rootdir->haveDot = 1;
+ rootdir->haveDotDot = 1;
+ rootdir->rwVid = vid;
+ rootdir->copied = 0;
+ rootdir->parent = 0;
+ rootdir->name = strdup(".");
+ rootdir->vname = volHeader->name;
+ rootdir->ds_linkH = alinkH;
+
+ *ip = rootinode;
+
+ return 0;
+
+ error:
+ if (decroot && IH_DEC(alinkH, rootinode, vid)) {
+ Log("CreateRootDir (recovery): IH_DEC (root) failed\n");
+ }
+ if (decreadme && IH_DEC(alinkH, readmeinode, vid)) {
+ Log("CreateRootDir (recovery): IH_DEC (readme) failed\n");
+ }
+ if (rootvnode) {
+ free(rootvnode);
+ rootvnode = NULL;
+ }
+ return -1;
+}
+
+/**
+ * salvage a volume group.
+ *
+ * @param[in] salvinfo information for the curent salvage job
+ * @param[in] rwIsp inode summary for rw volume
+ * @param[in] alinkH link table inode handle
+ *
+ * @return operation status
+ * @retval 0 success
+ */
int
-SalvageVolume(register struct InodeSummary *rwIsp, IHandle_t * alinkH)
+SalvageVolume(struct SalvInfo *salvinfo, struct InodeSummary *rwIsp, IHandle_t * alinkH)
{
/* This routine, for now, will only be called for read-write volumes */
int i, j, code;
int BlocksInVolume = 0, FilesInVolume = 0;
- register VnodeClass class;
+ VnodeClass class;
struct DirSummary rootdir, oldrootdir;
struct VnodeInfo *dirVnodeInfo;
struct VnodeDiskObject vnode;
VnodeId LFVnode, ThisVnode;
Unique LFUnique, ThisUnique;
char npath[128];
+ int newrootdir = 0;
vid = rwIsp->volSummary->header.id;
- IH_INIT(h, fileSysDevice, vid, rwIsp->volSummary->header.volumeInfo);
+ IH_INIT(h, salvinfo->fileSysDevice, vid, rwIsp->volSummary->header.volumeInfo);
nBytes = IH_IREAD(h, 0, (char *)&volHeader, sizeof(volHeader));
- assert(nBytes == sizeof(volHeader));
- assert(volHeader.stamp.magic == VOLUMEINFOMAGIC);
- assert(volHeader.destroyMe != DESTROY_ME);
+ osi_Assert(nBytes == sizeof(volHeader));
+ osi_Assert(volHeader.stamp.magic == VOLUMEINFOMAGIC);
+ osi_Assert(volHeader.destroyMe != DESTROY_ME);
/* (should not have gotten this far with DESTROY_ME flag still set!) */
- DistilVnodeEssence(vid, vLarge, rwIsp->volSummary->header.largeVnodeIndex,
- &maxunique);
- DistilVnodeEssence(vid, vSmall, rwIsp->volSummary->header.smallVnodeIndex,
- &maxunique);
+ DistilVnodeEssence(salvinfo, vid, vLarge,
+ rwIsp->volSummary->header.largeVnodeIndex, &maxunique);
+ DistilVnodeEssence(salvinfo, vid, vSmall,
+ rwIsp->volSummary->header.smallVnodeIndex, &maxunique);
- dirVnodeInfo = &vnodeInfo[vLarge];
+ dirVnodeInfo = &salvinfo->vnodeInfo[vLarge];
for (i = 0; i < dirVnodeInfo->nVnodes; i++) {
- SalvageDir(volHeader.name, vid, dirVnodeInfo, alinkH, i, &rootdir,
- &rootdirfound);
+ SalvageDir(salvinfo, volHeader.name, vid, dirVnodeInfo, alinkH, i,
+ &rootdir, &rootdirfound);
}
#ifdef AFS_NT40_ENV
- nt_sync(fileSysDevice);
+ nt_sync(salvinfo->fileSysDevice);
#else
sync(); /* This used to be done lower level, for every dir */
#endif
return 0;
}
+ if (!rootdirfound && (orphans == ORPH_ATTACH) && !Testing) {
+
+ Log("Cannot find root directory for volume %lu; attempting to create "
+ "a new one\n", afs_printable_uint32_lu(vid));
+
+ code = CreateRootDir(salvinfo, &volHeader, alinkH, vid, &rootdir,
+ &maxunique);
+ if (code == 0) {
+ rootdirfound = 1;
+ newrootdir = 1;
+ salvinfo->VolumeChanged = 1;
+ }
+ }
+
/* Parse each vnode looking for orphaned vnodes and
* connect them to the tree as orphaned (if requested).
*/
oldrootdir = rootdir;
for (class = 0; class < nVNODECLASSES; class++) {
- for (v = 0; v < vnodeInfo[class].nVnodes; v++) {
- vep = &(vnodeInfo[class].vnodes[v]);
+ for (v = 0; v < salvinfo->vnodeInfo[class].nVnodes; v++) {
+ vep = &(salvinfo->vnodeInfo[class].vnodes[v]);
ThisVnode = bitNumberToVnodeNumber(v, class);
ThisUnique = vep->unique;
*/
if (class == vLarge) { /* directory vnode */
pv = vnodeIdToBitNumber(vep->parent);
- if (vnodeInfo[vLarge].vnodes[pv].unique != 0)
- vnodeInfo[vLarge].vnodes[pv].count++;
+ if (salvinfo->vnodeInfo[vLarge].vnodes[pv].unique != 0) {
+ if (vep->parent == 1 && newrootdir) {
+ /* this vnode's parent was the volume root, and
+ * we just created the volume root. So, the parent
+ * dir didn't exist during JudgeEntry, so the link
+ * count was not inc'd there, so don't dec it here.
+ */
+
+ /* noop */
+
+ } else {
+ salvinfo->vnodeInfo[vLarge].vnodes[pv].count++;
+ }
+ }
}
if (!rootdirfound)
LFVnode = rootdir.vnodeNumber; /* Lost+Found vnode number */
LFUnique = rootdir.unique; /* Lost+Found uniquifier */
- /* Update this orphaned vnode's info. Its parent info and
+ /* Update this orphaned vnode's info. Its parent info and
* link count (do for orphaned directories and files).
*/
vep->parent = LFVnode; /* Parent is the root dir */
vep->claimed = 1;
vep->count--; /* Inc link count (root dir will pt to it) */
- /* If this orphaned vnode is a directory, change '..'.
+ /* If this orphaned vnode is a directory, change '..'.
* The name of the orphaned dir/file is unknown, so we
* build a unique name. No need to CopyOnWrite the directory
* since it is not connected to tree in BK or RO volume and
DirHandle dh;
/* Remove and recreate the ".." entry in this orphaned directory */
- SetSalvageDirHandle(&dh, vid, fileSysDevice,
- vnodeInfo[class].inodes[v]);
+ SetSalvageDirHandle(&dh, vid, salvinfo->fileSysDevice,
+ salvinfo->vnodeInfo[class].inodes[v],
+ &salvinfo->VolumeChanged);
pa.Vnode = LFVnode;
pa.Unique = LFUnique;
- assert(Delete(&dh, "..") == 0);
- assert(Create(&dh, "..", &pa) == 0);
+ osi_Assert(afs_dir_Delete(&dh, "..") == 0);
+ osi_Assert(afs_dir_Create(&dh, "..", &pa) == 0);
/* The original parent's link count was decremented above.
* Here we increment the new parent's link count.
*/
pv = vnodeIdToBitNumber(LFVnode);
- vnodeInfo[vLarge].vnodes[pv].count--;
+ salvinfo->vnodeInfo[vLarge].vnodes[pv].count--;
}
pa.Vnode = ThisVnode;
pa.Unique = ThisUnique;
- (void)afs_snprintf(npath, sizeof npath, "%s.%u.%u",
- ((class ==
- vLarge) ? "__ORPHANDIR__" :
- "__ORPHANFILE__"), ThisVnode,
- ThisUnique);
+ snprintf(npath, sizeof npath, "%s.%u.%u",
+ ((class == vLarge) ? "__ORPHANDIR__"
+ : "__ORPHANFILE__"),
+ ThisVnode, ThisUnique);
- CopyOnWrite(&rootdir);
- code = Create(&rootdir.dirHandle, npath, &pa);
+ CopyOnWrite(salvinfo, &rootdir);
+ code = afs_dir_Create(&rootdir.dirHandle, npath, &pa);
if (!code)
break;
ThisUnique += 50; /* Try creating a different file */
}
- assert(code == 0);
+ osi_Assert(code == 0);
Log("Attaching orphaned %s to volume's root dir as %s\n",
((class == vLarge) ? "directory" : "file"), npath);
}
/* Delete the old rootinode directory if the rootdir was CopyOnWrite */
DFlush();
- if (!oldrootdir.copied && rootdir.copied) {
+ if (rootdirfound && !oldrootdir.copied && rootdir.copied) {
code =
IH_DEC(oldrootdir.ds_linkH, oldrootdir.dirHandle.dirh_inode,
oldrootdir.rwVid);
- assert(code == 0);
+ osi_Assert(code == 0);
/* dirVnodeInfo->inodes[?] is not updated with new inode number */
}
* will get removed here also (if requested).
*/
for (class = 0; class < nVNODECLASSES; class++) {
- int nVnodes = vnodeInfo[class].nVnodes;
+ afs_sfsize_t nVnodes = salvinfo->vnodeInfo[class].nVnodes;
struct VnodeClassInfo *vcp = &VnodeClassInfo[class];
- struct VnodeEssence *vnodes = vnodeInfo[class].vnodes;
- FilesInVolume += vnodeInfo[class].nAllocatedVnodes;
- BlocksInVolume += vnodeInfo[class].volumeBlockCount;
+ struct VnodeEssence *vnodes = salvinfo->vnodeInfo[class].vnodes;
+ FilesInVolume += salvinfo->vnodeInfo[class].nAllocatedVnodes;
+ BlocksInVolume += salvinfo->vnodeInfo[class].volumeBlockCount;
for (i = 0; i < nVnodes; i++) {
- register struct VnodeEssence *vnp = &vnodes[i];
+ struct VnodeEssence *vnp = &vnodes[i];
VnodeId vnodeNumber = bitNumberToVnodeNumber(i, class);
/* If the vnode is good but is unclaimed (not listed in
* any directory entries), then it is orphaned.
*/
orphaned = -1;
- if ((vnp->type != 0) && (orphaned = IsVnodeOrphaned(vnodeNumber))) {
+ if ((vnp->type != 0) && (orphaned = IsVnodeOrphaned(salvinfo, vnodeNumber))) {
vnp->claimed = 0; /* Makes IsVnodeOrphaned calls faster */
vnp->changed = 1;
}
if (vnp->changed || vnp->count) {
int oldCount;
- int code;
nBytes =
- IH_IREAD(vnodeInfo[class].handle,
+ IH_IREAD(salvinfo->vnodeInfo[class].handle,
vnodeIndexOffset(vcp, vnodeNumber),
(char *)&vnode, sizeof(vnode));
- assert(nBytes == sizeof(vnode));
+ osi_Assert(nBytes == sizeof(vnode));
vnode.parent = vnp->parent;
oldCount = vnode.linkCount;
vnode.linkCount = vnode.linkCount - vnp->count;
if (orphaned == -1)
- orphaned = IsVnodeOrphaned(vnodeNumber);
+ orphaned = IsVnodeOrphaned(salvinfo, vnodeNumber);
if (orphaned) {
if (!vnp->todelete) {
/* Orphans should have already been attached (if requested) */
- assert(orphans != ORPH_ATTACH);
+ osi_Assert(orphans != ORPH_ATTACH);
oblocks += vnp->blockCount;
ofiles++;
}
if (VNDISK_GET_INO(&vnode)) {
code =
IH_DEC(alinkH, VNDISK_GET_INO(&vnode), vid);
- assert(code == 0);
+ osi_Assert(code == 0);
}
memset(&vnode, 0, sizeof(vnode));
}
vnode.dataVersion++;
if (!Testing) {
nBytes =
- IH_IWRITE(vnodeInfo[class].handle,
+ IH_IWRITE(salvinfo->vnodeInfo[class].handle,
vnodeIndexOffset(vcp, vnodeNumber),
(char *)&vnode, sizeof(vnode));
- assert(nBytes == sizeof(vnode));
+ osi_Assert(nBytes == sizeof(vnode));
}
- VolumeChanged = 1;
+ salvinfo->VolumeChanged = 1;
}
}
}
}
for (class = 0; class < nVNODECLASSES; class++) {
- register struct VnodeInfo *vip = &vnodeInfo[class];
+ struct VnodeInfo *vip = &salvinfo->vnodeInfo[class];
for (i = 0; i < vip->nVnodes; i++)
if (vip->vnodes[i].name)
free(vip->vnodes[i].name);
volHeader.uniquifier = (maxunique + 1 + 2000);
}
+ if (newrootdir) {
+ Log("*** WARNING: Root directory recreated, but volume is fragile! "
+ "Only use this salvaged volume to copy data to another volume; "
+ "do not continue to use this volume (%lu) as-is.\n",
+ afs_printable_uint32_lu(vid));
+ }
+
+ if (!Testing && salvinfo->VolumeChanged) {
+#ifdef FSSYNC_BUILD_CLIENT
+ if (salvinfo->useFSYNC) {
+ afs_int32 fsync_code;
+
+ fsync_code = FSYNC_VolOp(vid, NULL, FSYNC_VOL_BREAKCBKS, FSYNC_SALVAGE, NULL);
+ if (fsync_code) {
+ Log("Error trying to tell the fileserver to break callbacks for "
+ "changed volume %lu; error code %ld\n",
+ afs_printable_uint32_lu(vid),
+ afs_printable_int32_ld(fsync_code));
+ } else {
+ salvinfo->VolumeChanged = 0;
+ }
+ }
+#endif /* FSSYNC_BUILD_CLIENT */
+
+#if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
+ if (!salvinfo->useFSYNC) {
+ /* A volume's contents have changed, but the fileserver will not
+ * break callbacks on the volume until it tries to load the vol
+ * header. So, to reduce the amount of time a client could have
+ * stale data, remove fsstate.dat, so the fileserver will init
+ * callback state with all clients. This is a very coarse hammer,
+ * and in the future we should just record which volumes have
+ * changed. */
+ code = unlink(AFSDIR_SERVER_FSSTATE_FILEPATH);
+ if (code && errno != ENOENT) {
+ Log("Error %d when trying to unlink FS state file %s\n", errno,
+ AFSDIR_SERVER_FSSTATE_FILEPATH);
+ }
+ }
+#endif
+ }
+
/* Turn off the inUse bit; the volume's been salvaged! */
volHeader.inUse = 0; /* clear flag indicating inUse@last crash */
volHeader.needsSalvaged = 0; /* clear 'damaged' flag */
volHeader.inService = 1; /* allow service again */
- volHeader.needsCallback = (VolumeChanged != 0);
+ volHeader.needsCallback = (salvinfo->VolumeChanged != 0);
volHeader.dontSalvage = DONT_SALVAGE;
- VolumeChanged = 0;
+ salvinfo->VolumeChanged = 0;
if (!Testing) {
nBytes = IH_IWRITE(h, 0, (char *)&volHeader, sizeof(volHeader));
- assert(nBytes == sizeof(volHeader));
+ osi_Assert(nBytes == sizeof(volHeader));
}
if (!Showmode) {
Log("%sSalvaged %s (%u): %d files, %d blocks\n",
(Testing ? "It would have " : ""), volHeader.name, volHeader.id,
FilesInVolume, BlocksInVolume);
}
- IH_RELEASE(vnodeInfo[vSmall].handle);
- IH_RELEASE(vnodeInfo[vLarge].handle);
+
+ IH_RELEASE(salvinfo->vnodeInfo[vSmall].handle);
+ IH_RELEASE(salvinfo->vnodeInfo[vLarge].handle);
IH_RELEASE(h);
return 0;
}
VolumeDiskData volHeader;
nBytes = IH_IREAD(h, 0, (char *)&volHeader, sizeof(volHeader));
- assert(nBytes == sizeof(volHeader));
- assert(volHeader.stamp.magic == VOLUMEINFOMAGIC);
+ osi_Assert(nBytes == sizeof(volHeader));
+ osi_Assert(volHeader.stamp.magic == VOLUMEINFOMAGIC);
volHeader.inUse = 0;
volHeader.needsSalvaged = 0;
volHeader.inService = 1;
volHeader.dontSalvage = DONT_SALVAGE;
if (!Testing) {
nBytes = IH_IWRITE(h, 0, (char *)&volHeader, sizeof(volHeader));
- assert(nBytes == sizeof(volHeader));
+ osi_Assert(nBytes == sizeof(volHeader));
}
}
* deleteMe - Always do so, only a partial volume.
*/
void
-MaybeZapVolume(register struct InodeSummary *isp, char *message, int deleteMe,
- int check)
+MaybeZapVolume(struct SalvInfo *salvinfo, struct InodeSummary *isp,
+ char *message, int deleteMe, int check)
{
if (readOnly(isp) || deleteMe) {
if (isp->volSummary && isp->volSummary->fileName) {
if (!Testing) {
afs_int32 code;
char path[64];
- sprintf(path, "%s/%s", fileSysPath, isp->volSummary->fileName);
+ sprintf(path, "%s" OS_DIRSEP "%s", salvinfo->fileSysPath, isp->volSummary->fileName);
- code = VDestroyVolumeDiskHeader(fileSysPartition, isp->volumeId, isp->RWvolumeId);
+ code = VDestroyVolumeDiskHeader(salvinfo->fileSysPartition, isp->volumeId, isp->RWvolumeId);
if (code) {
Log("Error %ld destroying volume disk header for volume %lu\n",
afs_printable_int32_ld(code),
if (unlink(path) && errno != ENOENT) {
Log("Unable to unlink %s (errno = %d)\n", path, errno);
}
+ if (salvinfo->useFSYNC) {
+ AskDelete(salvinfo, isp->volumeId);
+ }
+ isp->volSummary->deleted = 1;
}
}
} else if (!check) {
}
}
+#if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
+/**
+ * Locks a volume on disk for salvaging.
+ *
+ * @param[in] volumeId volume ID to lock
+ *
+ * @return operation status
+ * @retval 0 success
+ * @retval -1 volume lock raced with a fileserver restart; all volumes must
+ * checked out and locked again
+ *
+ * @note DAFS only
+ */
+static int
+LockVolume(struct SalvInfo *salvinfo, VolumeId volumeId)
+{
+ afs_int32 code;
+ int locktype;
+
+ /* should always be WRITE_LOCK, but keep the lock-type logic all
+ * in one place, in VVolLockType. Params will be ignored, but
+ * try to provide what we're logically doing. */
+ locktype = VVolLockType(V_VOLUPD, 1);
+
+ code = VLockVolumeByIdNB(volumeId, salvinfo->fileSysPartition, locktype);
+ if (code) {
+ if (code == EBUSY) {
+ Abort("Someone else appears to be using volume %lu; Aborted\n",
+ afs_printable_uint32_lu(volumeId));
+ }
+ Abort("Error %ld trying to lock volume %lu; Aborted\n",
+ afs_printable_int32_ld(code),
+ afs_printable_uint32_lu(volumeId));
+ }
+
+ code = FSYNC_VerifyCheckout(volumeId, salvinfo->fileSysPathName, FSYNC_VOL_OFF, FSYNC_SALVAGE);
+ if (code == SYNC_DENIED) {
+ /* need to retry checking out volumes */
+ return -1;
+ }
+ if (code != SYNC_OK) {
+ Abort("FSYNC_VerifyCheckout failed for volume %lu with code %ld\n",
+ afs_printable_uint32_lu(volumeId), afs_printable_int32_ld(code));
+ }
+
+ /* set inUse = programType in the volume header to ensure that nobody
+ * tries to use this volume again without salvaging, if we somehow crash
+ * or otherwise exit before finishing the salvage.
+ */
+ if (!Testing) {
+ IHandle_t *h;
+ struct VolumeHeader header;
+ struct VolumeDiskHeader diskHeader;
+ struct VolumeDiskData volHeader;
+
+ code = VReadVolumeDiskHeader(volumeId, salvinfo->fileSysPartition, &diskHeader);
+ if (code) {
+ return 0;
+ }
+
+ DiskToVolumeHeader(&header, &diskHeader);
+
+ IH_INIT(h, salvinfo->fileSysDevice, header.parent, header.volumeInfo);
+ if (IH_IREAD(h, 0, (char*)&volHeader, sizeof(volHeader)) != sizeof(volHeader) ||
+ volHeader.stamp.magic != VOLUMEINFOMAGIC) {
+
+ IH_RELEASE(h);
+ return 0;
+ }
+
+ volHeader.inUse = programType;
+
+ /* If we can't re-write the header, bail out and error. We don't
+ * assert when reading the header, since it's possible the
+ * header isn't really there (when there's no data associated
+ * with the volume; we just delete the vol header file in that
+ * case). But if it's there enough that we can read it, but
+ * somehow we cannot write to it to signify we're salvaging it,
+ * we've got a big problem and we cannot continue. */
+ osi_Assert(IH_IWRITE(h, 0, (char*)&volHeader, sizeof(volHeader)) == sizeof(volHeader));
+
+ IH_RELEASE(h);
+ }
+
+ return 0;
+}
+#endif /* AFS_DEMAND_ATTACH_FS || AFS_DEMAND_ATTACH_UTIL */
+
+static void
+AskError(struct SalvInfo *salvinfo, VolumeId volumeId)
+{
+#if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
+ afs_int32 code;
+ code = FSYNC_VolOp(volumeId, salvinfo->fileSysPartition->name,
+ FSYNC_VOL_FORCE_ERROR, FSYNC_WHATEVER, NULL);
+ if (code != SYNC_OK) {
+ Log("AskError: failed to force volume %lu into error state; "
+ "SYNC error code %ld (%s)\n", (long unsigned)volumeId,
+ (long)code, SYNC_res2string(code));
+ }
+#endif /* AFS_DEMAND_ATTACH_FS || AFS_DEMAND_ATTACH_UTIL */
+}
void
-AskOffline(VolumeId volumeId, char * partition)
+AskOffline(struct SalvInfo *salvinfo, VolumeId volumeId)
{
afs_int32 code, i;
SYNC_response res;
memset(&res, 0, sizeof(res));
for (i = 0; i < 3; i++) {
- code = FSYNC_VolOp(volumeId, partition, FSYNC_VOL_OFF, FSYNC_SALVAGE, &res);
+ code = FSYNC_VolOp(volumeId, salvinfo->fileSysPartition->name,
+ FSYNC_VOL_OFF, FSYNC_SALVAGE, &res);
if (code == SYNC_OK) {
break;
} else if (code == SYNC_DENIED) {
-#ifdef DEMAND_ATTACH_ENABLE
- Log("AskOffline: file server denied offline request; a general salvage may be required.\n");
-#else
- Log("AskOffline: file server denied offline request; a general salvage is required.\n");
-#endif
+ if (AskDAFS())
+ Log("AskOffline: file server denied offline request; a general salvage may be required.\n");
+ else
+ Log("AskOffline: file server denied offline request; a general salvage is required.\n");
Abort("Salvage aborted\n");
} else if (code == SYNC_BAD_COMMAND) {
Log("AskOffline: fssync protocol mismatch (bad command word '%d'); salvage aborting.\n",
FSYNC_VOL_OFF);
-#ifdef DEMAND_ATTACH_ENABLE
- Log("AskOffline: please make sure fileserver, volserver, salvageserver and salvager binaries are same version.\n");
+ if (AskDAFS()) {
+#if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
+ Log("AskOffline: please make sure dafileserver, davolserver, salvageserver and dasalvager binaries are same version.\n");
+#else
+ Log("AskOffline: fileserver is DAFS but we are not.\n");
+#endif
+ } else {
+#if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
+ Log("AskOffline: fileserver is not DAFS but we are.\n");
#else
- Log("AskOffline: please make sure fileserver, volserver and salvager binaries are same version.\n");
+ Log("AskOffline: please make sure fileserver, volserver and salvager binaries are same version.\n");
#endif
+ }
Abort("Salvage aborted\n");
} else if (i < 2) {
/* try it again */
Log("AskOffline: request for fileserver to take volume offline failed; salvage aborting.\n");
Abort("Salvage aborted\n");
}
+}
-#ifdef AFS_DEMAND_ATTACH_FS
- /* set inUse = programType in the volume header. We do this in case
- * the fileserver restarts/crashes while we are salvaging.
- * Otherwise, the fileserver could attach the volume again on
- * startup while we are salvaging, which would be very bad, or
- * schedule another salvage while we are salvaging, which would be
- * annoying. */
- if (!Testing) {
- IHandle_t *h;
- struct VolumeHeader header;
- struct VolumeDiskHeader diskHeader;
- struct VolumeDiskData volHeader;
+/* don't want to pass around state; remember it here */
+static int isDAFS = -1;
+int
+AskDAFS(void)
+{
+ afs_int32 code, i, ret = 0;
+ SYNC_response res;
- code = VReadVolumeDiskHeader(volumeId, fileSysPartition, &diskHeader);
- if (code) {
- return;
- }
+ /* we don't care if we race. the answer shouldn't change */
+ if (isDAFS != -1)
+ return isDAFS;
- DiskToVolumeHeader(&header, &diskHeader);
+ memset(&res, 0, sizeof(res));
- IH_INIT(h, fileSysDevice, header.parent, header.volumeInfo);
- if (IH_IREAD(h, 0, (char*)&volHeader, sizeof(volHeader)) != sizeof(volHeader) ||
- volHeader.stamp.magic != VOLUMEINFOMAGIC) {
+ for (i = 0; i < 3; i++) {
+ code = FSYNC_VolOp(1, NULL,
+ FSYNC_VOL_QUERY_VOP, FSYNC_SALVAGE, &res);
- IH_RELEASE(h);
- return;
+ if (code == SYNC_OK) {
+ ret = 1;
+ break;
+ } else if (code == SYNC_DENIED) {
+ ret = 1;
+ break;
+ } else if (code == SYNC_BAD_COMMAND) {
+ ret = 0;
+ break;
+ } else if (code == SYNC_FAILED) {
+ if (res.hdr.reason == FSYNC_UNKNOWN_VOLID)
+ ret = 1;
+ else
+ ret = 0;
+ break;
+ } else if (i < 2) {
+ /* try it again */
+ Log("AskDAFS: request to query fileserver failed; trying again...\n");
+ FSYNC_clientFinis();
+ FSYNC_clientInit();
}
+ }
- volHeader.inUse = programType;
-
- /* If we can't re-write the header, bail out and error. We don't
- * assert when reading the header, since it's possible the
- * header isn't really there (when there's no data associated
- * with the volume; we just delete the vol header file in that
- * case). But if it's there enough that we can read it, but
- * somehow we cannot write to it to signify we're salvaging it,
- * we've got a big problem and we cannot continue. */
- assert(IH_IWRITE(h, 0, (char*)&volHeader, sizeof(volHeader)) == sizeof(volHeader));
+ isDAFS = ret;
+ return ret;
+}
- IH_RELEASE(h);
+static void
+MaybeAskOnline(struct SalvInfo *salvinfo, VolumeId volumeId)
+{
+ struct VolumeDiskHeader diskHdr;
+ int code;
+ code = VReadVolumeDiskHeader(volumeId, salvinfo->fileSysPartition, &diskHdr);
+ if (code) {
+ /* volume probably does not exist; no need to bring back online */
+ return;
}
-#endif /* AFS_DEMAND_ATTACH_FS */
+ AskOnline(salvinfo, volumeId);
}
void
-AskOnline(VolumeId volumeId, char *partition)
+AskOnline(struct SalvInfo *salvinfo, VolumeId volumeId)
{
afs_int32 code, i;
for (i = 0; i < 3; i++) {
- code = FSYNC_VolOp(volumeId, partition, FSYNC_VOL_ON, FSYNC_WHATEVER, NULL);
+ code = FSYNC_VolOp(volumeId, salvinfo->fileSysPartition->name,
+ FSYNC_VOL_ON, FSYNC_WHATEVER, NULL);
if (code == SYNC_OK) {
break;
} else if (code == SYNC_DENIED) {
- Log("AskOnline: file server denied online request to volume %u partition %s; trying again...\n", volumeId, partition);
+ Log("AskOnline: file server denied online request to volume %u partition %s; trying again...\n", volumeId, salvinfo->fileSysPartition->name);
} else if (code == SYNC_BAD_COMMAND) {
Log("AskOnline: fssync protocol mismatch (bad command word '%d')\n",
FSYNC_VOL_ON);
-#ifdef DEMAND_ATTACH_ENABLE
- Log("AskOnline: please make sure fileserver, volserver, salvageserver and salvager binaries are same version.\n");
+ Log("AskOnline: please make sure file server binaries are same version.\n");
+ break;
+ } else if (i < 2) {
+ /* try it again */
+ Log("AskOnline: request for fileserver to put volume online failed; trying again...\n");
+ FSYNC_clientFinis();
+ FSYNC_clientInit();
+ }
+ }
+}
+
+void
+AskDelete(struct SalvInfo *salvinfo, VolumeId volumeId)
+{
+ afs_int32 code, i;
+ SYNC_response res;
+
+ for (i = 0; i < 3; i++) {
+ memset(&res, 0, sizeof(res));
+ code = FSYNC_VolOp(volumeId, salvinfo->fileSysPartition->name,
+ FSYNC_VOL_DONE, FSYNC_SALVAGE, &res);
+
+ if (code == SYNC_OK) {
+ break;
+ } else if (code == SYNC_DENIED) {
+ Log("AskOnline: file server denied DONE request to volume %u partition %s; trying again...\n", volumeId, salvinfo->fileSysPartition->name);
+ } else if (code == SYNC_BAD_COMMAND) {
+ Log("AskOnline: fssync protocol mismatch (bad command word '%d')\n",
+ FSYNC_VOL_DONE);
+ if (AskDAFS()) {
+#if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
+ Log("AskOnline: please make sure dafileserver, davolserver, salvageserver and dasalvager binaries are same version.\n");
#else
- Log("AskOnline: please make sure fileserver, volserver and salvager binaries are same version.\n");
+ Log("AskOnline: fileserver is DAFS but we are not.\n");
#endif
+ } else {
+#if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
+ Log("AskOnline: fileserver is not DAFS but we are.\n");
+#else
+ Log("AskOnline: please make sure fileserver, volserver and salvager binaries are same version.\n");
+#endif
+ }
+ break;
+ } else if (code == SYNC_FAILED &&
+ (res.hdr.reason == FSYNC_UNKNOWN_VOLID ||
+ res.hdr.reason == FSYNC_WRONG_PART)) {
+ /* volume is already effectively 'deleted' */
break;
} else if (i < 2) {
/* try it again */
- Log("AskOnline: request for fileserver to take volume offline failed; trying again...\n");
+ Log("AskOnline: request for fileserver to delete volume failed; trying again...\n");
FSYNC_clientFinis();
FSYNC_clientInit();
}
char buf[4096];
IHandle_t *srcH, *destH;
FdHandle_t *srcFdP, *destFdP;
- register int n = 0;
+ ssize_t nBytes = 0;
+ afs_foff_t size = 0;
IH_INIT(srcH, device, rwvolume, inode1);
srcFdP = IH_OPEN(srcH);
- assert(srcFdP != NULL);
+ osi_Assert(srcFdP != NULL);
IH_INIT(destH, device, rwvolume, inode2);
destFdP = IH_OPEN(destH);
- assert(n != -1);
- while ((n = FDH_READ(srcFdP, buf, sizeof(buf))) > 0)
- assert(FDH_WRITE(destFdP, buf, n) == n);
- assert(n == 0);
+ while ((nBytes = FDH_PREAD(srcFdP, buf, sizeof(buf), size)) > 0) {
+ osi_Assert(FDH_PWRITE(destFdP, buf, nBytes, size) == nBytes);
+ size += nBytes;
+ }
+ osi_Assert(nBytes == 0);
FDH_REALLYCLOSE(srcFdP);
FDH_REALLYCLOSE(destFdP);
IH_RELEASE(srcH);
}
void
-PrintInodeList(void)
+PrintInodeList(struct SalvInfo *salvinfo)
{
- register struct ViceInodeInfo *ip;
+ struct ViceInodeInfo *ip;
struct ViceInodeInfo *buf;
- struct afs_stat status;
- register int nInodes;
-
- assert(afs_fstat(inodeFd, &status) == 0);
- buf = (struct ViceInodeInfo *)malloc(status.st_size);
- assert(buf != NULL);
- nInodes = status.st_size / sizeof(struct ViceInodeInfo);
- assert(read(inodeFd, buf, status.st_size) == status.st_size);
+ int nInodes;
+ afs_ino_str_t stmp;
+ afs_sfsize_t st_size;
+
+ st_size = OS_SIZE(salvinfo->inodeFd);
+ osi_Assert(st_size >= 0);
+ buf = (struct ViceInodeInfo *)malloc(st_size);
+ osi_Assert(buf != NULL);
+ nInodes = st_size / sizeof(struct ViceInodeInfo);
+ osi_Assert(OS_READ(salvinfo->inodeFd, buf, st_size) == st_size);
for (ip = buf; nInodes--; ip++) {
Log("Inode:%s, linkCount=%d, size=%#llx, p=(%u,%u,%u,%u)\n",
- PrintInode(NULL, ip->inodeNumber), ip->linkCount,
+ PrintInode(stmp, ip->inodeNumber), ip->linkCount,
(afs_uintmax_t) ip->byteCount, ip->u.param[0], ip->u.param[1],
ip->u.param[2], ip->u.param[3]);
}
}
void
-PrintInodeSummary(void)
+PrintInodeSummary(struct SalvInfo *salvinfo)
{
int i;
struct InodeSummary *isp;
- for (i = 0; i < nVolumesInInodeFile; i++) {
- isp = &inodeSummary[i];
+ for (i = 0; i < salvinfo->nVolumesInInodeFile; i++) {
+ isp = &salvinfo->inodeSummary[i];
Log("VID:%u, RW:%u, index:%d, nInodes:%d, nSpecialInodes:%d, maxUniquifier:%u, volSummary\n", isp->volumeId, isp->RWvolumeId, isp->index, isp->nInodes, isp->nSpecialInodes, isp->maxUniquifier);
}
}
void
-PrintVolumeSummary(void)
+PrintVolumeSummary(struct SalvInfo *salvinfo)
{
int i;
struct VolumeSummary *vsp;
- for (i = 0, vsp = volumeSummaryp; i < nVolumes; vsp++, i++) {
+ for (i = 0, vsp = salvinfo->volumeSummaryp; i < salvinfo->nVolumes; vsp++, i++) {
Log("fileName:%s, header, wouldNeedCallback\n", vsp->fileName);
}
}
int f;
#ifdef AFS_NT40_ENV
f = 0;
- assert(0); /* Fork is never executed in the NT code path */
+ osi_Assert(0); /* Fork is never executed in the NT code path */
#else
f = fork();
- assert(f >= 0);
+ osi_Assert(f >= 0);
#ifdef AFS_DEMAND_ATTACH_FS
if ((f == 0) && (programType == salvageServer)) {
/* we are a salvageserver child */
int status;
int pid;
pid = wait(&status);
- assert(pid != -1);
+ osi_Assert(pid != -1);
if (WCOREDUMP(status))
Log("\"%s\" core dumped!\n", prog);
if (WIFSIGNALED(status) != 0 || WEXITSTATUS(status) != 0)
now = time(0);
lt = localtime(&now);
- (void)afs_snprintf(stampSlvgLog, sizeof stampSlvgLog,
- "%s.%04d-%02d-%02d.%02d:%02d:%02d",
- log_path, lt->tm_year + 1900,
- lt->tm_mon + 1, lt->tm_mday, lt->tm_hour, lt->tm_min,
- lt->tm_sec);
+ snprintf(stampSlvgLog, sizeof stampSlvgLog,
+ "%s.%04d-%02d-%02d.%02d:%02d:%02d", log_path,
+ lt->tm_year + 1900, lt->tm_mon + 1, lt->tm_mday, lt->tm_hour,
+ lt->tm_min, lt->tm_sec);
/* try to link the logfile to a timestamped filename */
/* if it fails, oh well, nothing we can do */
va_list args;
va_start(args, format);
- (void)afs_vsnprintf(tmp, sizeof tmp, format, args);
+ vsnprintf(tmp, sizeof tmp, format, args);
va_end(args);
#ifndef AFS_NT40_ENV
if (useSyslog) {
} else
#endif
if (logFile) {
- gettimeofday(&now, 0);
+ gettimeofday(&now, NULL);
fprintf(logFile, "%s %s", TimeStamp(now.tv_sec, 1), tmp);
fflush(logFile);
}
char tmp[1024];
va_start(args, format);
- (void)afs_vsnprintf(tmp, sizeof tmp, format, args);
+ vsnprintf(tmp, sizeof tmp, format, args);
va_end(args);
#ifndef AFS_NT40_ENV
if (useSyslog) {
char *
ToString(const char *s)
{
- register char *p;
+ char *p;
p = (char *)malloc(strlen(s) + 1);
- assert(p != NULL);
+ osi_Assert(p != NULL);
strcpy(p, s);
return p;
}
RemoveTheForce(char *path)
{
char target[1024];
- struct afs_stat force; /* so we can use afs_stat to find it */
+ struct afs_stat_st force; /* so we can use afs_stat to find it */
strcpy(target,path);
strcat(target,"/FORCESALVAGE");
if (!Testing && ForceSalvage) {
int
UseTheForceLuke(char *path)
{
- struct afs_stat force;
+ struct afs_stat_st force;
char target[1024];
strcpy(target,path);
strcat(target,"/FORCESALVAGE");
git://git.openafs.org / openafs.git / blobdiff