#include <io.h>
#include <WINNT/afsevent.h>
#endif
-#if defined(AFS_AIX_ENV) || defined(AFS_SUN4_ENV)
-#define WCOREDUMP(x) (x & 0200)
+#ifndef WCOREDUMP
+#define WCOREDUMP(x) ((x) & 0200)
#endif
#include <rx/xdr.h>
#include <afs/afsint.h>
#include "salvage.h"
#include "volinodes.h" /* header magic number, etc. stuff */
#include "vol-salvage.h"
+#include "vol_internal.h"
+
+#ifdef FSSYNC_BUILD_CLIENT
+#include "vg_cache.h"
+#endif
+
#ifdef AFS_NT40_ENV
#include <pthread.h>
#endif
struct VnodeInfo vnodeInfo[nVNODECLASSES];
-struct VolumeSummary *volumeSummaryp; /* Holds all the volumes in a part */
+struct VolumeSummary *volumeSummaryp = NULL; /* Holds all the volumes in a part */
int nVolumes; /* Number of volumes (read-write and read-only)
* in volume summary */
/*@printflike@*/ void Log(const char *format, ...);
/*@printflike@*/ void Abort(const char *format, ...);
static int IsVnodeOrphaned(VnodeId vnode);
+static int AskVolumeSummary(VolumeId singleVolumeNumber);
/* Uniquifier stored in the Inode */
static Unique
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);
}
{
char *name, *tdir;
char inodeListPath[256];
+ FILE *inodeFile;
static char tmpDevName[100];
static char wpath[100];
struct VolumeSummary *vsp, *esp;
int i, j;
+ int code;
fileSysPartition = partP;
fileSysDevice = fileSysPartition->device;
snprintf(inodeListPath, 255, "%s/salvage.inodes.%s.%d", tdir, name,
getpid());
#endif
- if (GetInodeSummary(inodeListPath, singleVolumeNumber) < 0) {
- unlink(inodeListPath);
- return;
+
+ inodeFile = fopen(inodeListPath, "w+b");
+ if (!inodeFile) {
+ Abort("Error %d when creating inode description file %s; not salvaged\n", errno, inodeListPath);
}
#ifdef AFS_NT40_ENV
/* Using nt_unlink here since we're really using the delete on close
* 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.
*/
- inodeFd =
- _open_osfhandle((intptr_t)nt_open(inodeListPath, O_RDWR, 0), O_RDWR);
- nt_unlink(inodeListPath); /* NT's crt unlink won't if file is open. */
+ code = nt_unlink(inodeListPath);
#else
- inodeFd = afs_open(inodeListPath, O_RDONLY);
- unlink(inodeListPath);
+ code = unlink(inodeListPath);
#endif
+ if (code < 0) {
+ Log("Error %d when trying to unlink %s\n", errno, inodeListPath);
+ }
+
+ if (GetInodeSummary(inodeFile, singleVolumeNumber) < 0) {
+ fclose(inodeFile);
+ return;
+ }
+ inodeFd = fileno(inodeFile);
if (inodeFd == -1)
Abort("Temporary file %s is missing...\n", inodeListPath);
+ afs_lseek(inodeFd, 0L, SEEK_SET);
if (ListInodeOption) {
PrintInodeList();
return;
fileSysPartition->name, (Testing ? " (READONLY mode)" : ""));
}
- close(inodeFd); /* SalvageVolumeGroup was the last which needed it. */
+ fclose(inodeFile); /* SalvageVolumeGroup was the last which needed it. */
}
void
DeleteExtraVolumeHeaderFile(register struct VolumeSummary *vsp)
{
+ char path[64];
+ sprintf(path, "%s/%s", fileSysPath, vsp->fileName);
+
if (!Showmode)
- Log("The volume header file %s is not associated with any actual data (%sdeleted)\n", vsp->fileName, (Testing ? "would have been " : ""));
- if (!Testing)
- unlink(vsp->fileName);
+ 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);
+ if (code) {
+ Log("Error %ld destroying volume disk header for volume %lu\n",
+ afs_printable_int32_ld(code),
+ afs_printable_uint32_lu(vsp->header.id));
+ }
+
+ /* make sure we actually delete the fileName file; ENOENT
+ * is fine, since VDestroyVolumeDiskHeader probably already
+ * unlinked it */
+ if (unlink(path) && errno != ENOENT) {
+ Log("Unable to unlink %s (errno = %d)\n", path, errno);
+ }
+ }
vsp->fileName = 0;
}
* be unlinked by the caller.
*/
int
-GetInodeSummary(char *path, VolumeId singleVolumeNumber)
+GetInodeSummary(FILE *inodeFile, VolumeId singleVolumeNumber)
{
struct afs_stat status;
int forceSal, err;
+ int code;
struct ViceInodeInfo *ip;
struct InodeSummary summary;
char summaryFileName[50];
/* This file used to come from vfsck; cobble it up ourselves now... */
if ((err =
- ListViceInodes(dev, fileSysPath, path,
+ ListViceInodes(dev, 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 %s; Not salvaged %s ***\nIncrease space on partition or use '-tmpdir'\n", errno, path, dev);
+ 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;
}
- unlink(path);
Abort("Unable to get inodes for \"%s\"; not salvaged\n", dev);
}
if (forceSal && !ForceSalvage) {
Log("***Forced salvage of all volumes on this partition***\n");
ForceSalvage = 1;
}
- inodeFd = afs_open(path, O_RDWR);
+ fseek(inodeFile, 0L, SEEK_SET);
+ inodeFd = fileno(inodeFile);
if (inodeFd == -1 || afs_fstat(inodeFd, &status) == -1) {
- unlink(path);
Abort("No inode description file for \"%s\"; not salvaged\n", dev);
}
tdir = (tmpdir ? tmpdir : part);
#endif
summaryFile = afs_fopen(summaryFileName, "a+");
if (summaryFile == NULL) {
- close(inodeFd);
- unlink(path);
Abort("Unable to create inode summary file\n");
}
+
+#ifdef AFS_NT40_ENV
+ /* Using nt_unlink here since we're really using the delete on close
+ * 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.
+ */
+ code = nt_unlink(summaryFileName);
+#else
+ code = unlink(summaryFileName);
+#endif
+ if (code < 0) {
+ Log("Error %d when trying to unlink %s\n", errno, summaryFileName);
+ }
+
if (!canfork || debug || Fork() == 0) {
int nInodes;
unsigned long st_size=(unsigned long) status.st_size;
nInodes = st_size / sizeof(struct ViceInodeInfo);
if (nInodes == 0) {
fclose(summaryFile);
- close(inodeFd);
- unlink(summaryFileName);
if (!singleVolumeNumber) /* Remove the FORCESALVAGE file */
RemoveTheForce(fileSysPath);
else {
ip = (struct ViceInodeInfo *)malloc(nInodes*sizeof(struct ViceInodeInfo));
if (ip == NULL) {
fclose(summaryFile);
- close(inodeFd);
- unlink(path);
- unlink(summaryFileName);
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);
- close(inodeFd);
- unlink(path);
- unlink(summaryFileName);
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);
- close(inodeFd);
- unlink(path);
- unlink(summaryFileName);
Abort("Unable to rewrite inode table; %s not salvaged\n", dev);
}
summary.index = 0;
if (fwrite(&summary, sizeof(summary), 1, summaryFile) != 1) {
Log("Difficulty writing summary file (errno = %d); %s not salvaged\n", errno, dev);
fclose(summaryFile);
- close(inodeFd);
return -1;
}
summary.index += (summary.nInodes);
if (fflush(summaryFile) == EOF || fsync(fileno(summaryFile)) == -1) {
Log("Unable to write summary file (errno = %d); %s not salvaged\n", errno, dev);
fclose(summaryFile);
- close(inodeFd);
return -1;
}
if (canfork && !debug) {
} else {
if (Wait("Inode summary") == -1) {
fclose(summaryFile);
- close(inodeFd);
- unlink(path);
- unlink(summaryFileName);
Exit(1); /* salvage of this partition aborted */
}
}
nVolumesInInodeFile =(unsigned long)(status.st_size) / sizeof(struct InodeSummary);
Log("%d nVolumesInInodeFile %d \n",nVolumesInInodeFile,(unsigned long)(status.st_size));
fclose(summaryFile);
- close(inodeFd);
- unlink(summaryFileName);
return 0;
}
return p1->header.id < p2->header.id ? -1 : 1; /* Both read-only */
}
+/**
+ * Gleans volumeSummary information by asking the fileserver
+ *
+ * @param[in] singleVolumeNumber the volume we're salvaging. 0 if we're
+ * 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)
+ *
+ * @note for non-DAFS, always returns 1
+ */
+static int
+AskVolumeSummary(VolumeId singleVolumeNumber)
+{
+ afs_int32 code = 1;
+#ifdef FSSYNC_BUILD_CLIENT
+ if (programType == salvageServer) {
+ if (singleVolumeNumber) {
+ FSSYNC_VGQry_response_t q_res;
+ SYNC_response res;
+ struct VolumeSummary *vsp;
+ int i;
+ struct VolumeDiskHeader diskHdr;
+
+ memset(&res, 0, sizeof(res));
+
+ code = FSYNC_VGCQuery(fileSysPartition->name, singleVolumeNumber, &q_res, &res);
+
+ /*
+ * We must wait for the partition to finish scanning before
+ * can continue, since we will not know if we got the entire
+ * VG membership unless the partition is fully scanned.
+ * We could, in theory, just scan the partition ourselves if
+ * the VG cache is not ready, but we would be doing the exact
+ * same scan the fileserver is doing; it will almost always
+ * be faster to wait for the fileserver. The only exceptions
+ * are if the partition does not take very long to scan, and
+ * in that case it's fast either way, so who cares?
+ */
+ if (code == SYNC_FAILED && res.hdr.reason == FSYNC_PART_SCANNING) {
+ Log("waiting for fileserver to finish scanning partition %s...\n",
+ 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,
+ * just so small partitions don't need to wait over 10
+ * seconds every time, and large partitions are generally
+ * polled only once every ten seconds. */
+ sleep((i > 10) ? (i = 10) : i);
+
+ code = FSYNC_VGCQuery(fileSysPartition->name, singleVolumeNumber, &q_res, &res);
+ }
+ }
+
+ if (code == SYNC_FAILED && res.hdr.reason == FSYNC_UNKNOWN_VOLID) {
+ /* This can happen if there's no header for the volume
+ * we're salvaging, or no headers exist for the VG (if
+ * we're salvaging an RW). Act as if we got a response
+ * with no VG members. The headers may be created during
+ * salvaging, if there are inodes in this VG. */
+ code = 0;
+ memset(&q_res, 0, sizeof(q_res));
+ q_res.rw = singleVolumeNumber;
+ }
+
+ if (code) {
+ Log("fileserver refused VGCQuery request for volume %lu on "
+ "partition %s, code %ld reason %ld\n",
+ afs_printable_uint32_lu(singleVolumeNumber),
+ fileSysPartition->name,
+ afs_printable_int32_ld(code),
+ afs_printable_int32_ld(res.hdr.reason));
+ goto done;
+ }
+
+ if (q_res.rw != singleVolumeNumber) {
+ Log("fileserver requested salvage of clone %lu; scheduling salvage of volume group %lu...\n",
+ afs_printable_uint32_lu(singleVolumeNumber),
+ afs_printable_uint32_lu(q_res.rw));
+#ifdef SALVSYNC_BUILD_CLIENT
+ if (SALVSYNC_LinkVolume(q_res.rw,
+ singleVolumeNumber,
+ fileSysPartition->name,
+ NULL) != SYNC_OK) {
+ Log("schedule request failed\n");
+ }
+#endif /* SALVSYNC_BUILD_CLIENT */
+ Exit(SALSRV_EXIT_VOLGROUP_LINK);
+ }
+
+ volumeSummaryp = malloc(VOL_VG_MAX_VOLS * sizeof(struct VolumeSummary));
+ assert(volumeSummaryp != NULL);
+
+ nVolumes = 0;
+ vsp = volumeSummaryp;
+
+ for (i = 0; i < VOL_VG_MAX_VOLS; i++) {
+ char name[VMAXPATHLEN];
+
+ if (!q_res.children[i]) {
+ continue;
+ }
+
+ if (q_res.children[i] != singleVolumeNumber) {
+ AskOffline(q_res.children[i], fileSysPartition->name);
+ }
+ code = VReadVolumeDiskHeader(q_res.children[i], fileSysPartition, &diskHdr);
+ if (code) {
+ Log("Cannot read header for %lu; trying to salvage group anyway\n",
+ afs_printable_uint32_lu(q_res.children[i]));
+ code = 0;
+ continue;
+ }
+
+ DiskToVolumeHeader(&vsp->header, &diskHdr);
+ VolumeExternalName_r(q_res.children[i], name, sizeof(name));
+ vsp->fileName = ToString(name);
+ nVolumes++;
+ vsp++;
+ }
+
+ qsort(volumeSummaryp, nVolumes, sizeof(struct VolumeSummary),
+ CompareVolumes);
+ }
+ done:
+ if (code) {
+ Log("Cannot get volume summary from fileserver; falling back to scanning "
+ "entire partition\n");
+ }
+ }
+#endif /* FSSYNC_BUILD_CLIENT */
+ return code;
+}
+
+/**
+ * count how many volume headers are found by VWalkVolumeHeaders.
+ *
+ * @param[in] dp the disk partition (unused)
+ * @param[in] name full path to the .vol header (unused)
+ * @param[in] hdr the header data (unused)
+ * @param[in] last whether this is the last try or not (unused)
+ * @param[in] rock actually an afs_int32*; the running count of how many
+ * volumes we have found
+ *
+ * @retval 0 always
+ */
+static int
+CountHeader(struct DiskPartition64 *dp, const char *name,
+ struct VolumeDiskHeader *hdr, int last, void *rock)
+{
+ afs_int32 *nvols = (afs_int32 *)rock;
+ (*nvols)++;
+ return 0;
+}
+
+/**
+ * parameters to pass to the VWalkVolumeHeaders callbacks when recording volume
+ * data.
+ */
+struct SalvageScanParams {
+ VolumeId singleVolumeNumber; /**< 0 for a partition-salvage, otherwise the
+ * vol id of the VG we're salvaging */
+ struct VolumeSummary *vsp; /**< ptr to the current volume summary object
+ * we're filling in */
+ 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) */
+};
+
+/**
+ * records volume summary info found from VWalkVolumeHeaders.
+ *
+ * Found volumes are also taken offline if they are in the specific volume
+ * group we are looking for.
+ *
+ * @param[in] dp the disk partition
+ * @param[in] name full path to the .vol header
+ * @param[in] hdr the header data
+ * @param[in] last 1 if this is the last try to read the header, 0 otherwise
+ * @param[in] rock actually a struct SalvageScanParams*, containing the
+ * 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
+ */
+static int
+RecordHeader(struct DiskPartition64 *dp, const char *name,
+ struct VolumeDiskHeader *hdr, int last, void *rock)
+{
+ char nameShouldBe[64];
+ struct SalvageScanParams *params;
+ struct VolumeSummary summary;
+ VolumeId singleVolumeNumber;
+
+ params = (struct SalvageScanParams *)rock;
+
+ singleVolumeNumber = params->singleVolumeNumber;
+
+ DiskToVolumeHeader(&summary.header, hdr);
+
+ if (singleVolumeNumber && summary.header.id == singleVolumeNumber
+ && summary.header.parent != singleVolumeNumber) {
+
+ if (programType == salvageServer) {
+#ifdef SALVSYNC_BUILD_CLIENT
+ Log("fileserver requested salvage of clone %u; scheduling salvage of volume group %u...\n",
+ summary.header.id, summary.header.parent);
+ if (SALVSYNC_LinkVolume(summary.header.parent,
+ summary.header.id,
+ dp->name,
+ NULL) != SYNC_OK) {
+ Log("schedule request failed\n");
+ }
+#endif
+ Exit(SALSRV_EXIT_VOLGROUP_LINK);
+
+ } else {
+ Log("%u is a read-only volume; not salvaged\n",
+ singleVolumeNumber);
+ Exit(1);
+ }
+ }
+
+ if (!singleVolumeNumber || summary.header.id == singleVolumeNumber
+ || summary.header.parent == singleVolumeNumber) {
+
+ /* check if the header file is incorrectly named */
+ int badname = 0;
+ const char *base = strrchr(name, '/');
+ if (base) {
+ base++;
+ } else {
+ base = name;
+ }
+
+ (void)afs_snprintf(nameShouldBe, sizeof nameShouldBe,
+ VFORMAT, afs_printable_uint32_lu(summary.header.id));
+
+
+ if (strcmp(nameShouldBe, base)) {
+ /* .vol file has wrong name; retry/delete */
+ badname = 1;
+ }
+
+ if (!badname || last) {
+ /* only offline the volume if the header is good, or if this is
+ * the last try looking at it; avoid AskOffline'ing the same vol
+ * multiple times */
+
+ if (singleVolumeNumber
+ && summary.header.id != singleVolumeNumber) {
+ /* don't offline singleVolumeNumber; we already did that
+ * earlier */
+
+ AskOffline(summary.header.id, fileSysPartition->name);
+ }
+ }
+ if (badname) {
+ if (last && !Showmode) {
+ Log("Volume header file %s is incorrectly named (should be %s "
+ "not %s); %sdeleted (it will be recreated later, if "
+ "necessary)\n", name, nameShouldBe, base,
+ (Testing ? "it would have been " : ""));
+ }
+ return 1;
+ }
+
+ summary.fileName = ToString(base);
+ params->nVolumes++;
+
+ if (params->nVolumes > params->totalVolumes) {
+ /* We found more volumes than we found on the first partition walk;
+ * apparently something created a volume while we were
+ * partition-salvaging, or we found more than 20 vols when salvaging a
+ * particular volume. Abort if we detect this, since other programs
+ * supposed to not touch the partition while it is partition-salvaging,
+ * and we shouldn't find more than 20 vols in a VG.
+ */
+ Abort("Found %ld vol headers, but should have found at most %ld! "
+ "Make sure the volserver/fileserver are not running at the "
+ "same time as a partition salvage\n",
+ afs_printable_int32_ld(params->nVolumes),
+ afs_printable_int32_ld(params->totalVolumes));
+ }
+
+ memcpy(params->vsp, &summary, sizeof(summary));
+ params->vsp++;
+ }
+
+ return 0;
+}
+
+/**
+ * possibly unlinks bad volume headers found from VWalkVolumeHeaders.
+ *
+ * If the header could not be read in at all, the header is always unlinked.
+ * If instead RecordHeader said the header was bad (that is, the header file
+ * is mis-named), we only unlink if we are doing a partition salvage, as
+ * opposed to salvaging a specific volume group.
+ *
+ * @param[in] dp the disk partition
+ * @param[in] name full path to the .vol header
+ * @param[in] hdr header data, or NULL if the header could not be read
+ * @param[in] rock actually a struct SalvageScanParams*, with some information
+ * about the scan
+ */
+static void
+UnlinkHeader(struct DiskPartition64 *dp, const char *name,
+ struct VolumeDiskHeader *hdr, void *rock)
+{
+ struct SalvageScanParams *params;
+ int dounlink = 0;
+
+ params = (struct SalvageScanParams *)rock;
+
+ if (!hdr) {
+ /* no header; header is too bogus to read in at all */
+ if (!Showmode) {
+ Log("%s is not a legitimate volume header file; %sdeleted\n", name, (Testing ? "it would have been " : ""));
+ }
+ if (!Testing) {
+ dounlink = 1;
+ }
+
+ } else if (!params->singleVolumeNumber) {
+ /* We were able to read in a header, but RecordHeader said something
+ * was wrong with it. We only unlink those if we are doing a partition
+ * salvage. */
+ if (!Testing) {
+ dounlink = 1;
+ }
+ }
+
+ if (dounlink && unlink(name)) {
+ Log("Error %d while trying to unlink %s\n", errno, name);
+ }
+}
+
void
GetVolumeSummary(VolumeId singleVolumeNumber)
{
- DIR *dirp = NULL;
afs_int32 nvols = 0;
- struct VolumeSummary *vsp, vs;
- struct VolumeDiskHeader diskHeader;
- struct dirent *dp;
+ struct SalvageScanParams params;
+ int code;
+
+ if (AskVolumeSummary(singleVolumeNumber) == 0) {
+ /* we successfully got the vol information from the fileserver; no
+ * need to scan the partition */
+ return;
+ }
- /* Get headers from volume directory */
- dirp = opendir(fileSysPath);
- if (dirp == NULL)
- Abort("Can't read directory %s; not salvaged\n", fileSysPath);
if (!singleVolumeNumber) {
- while ((dp = readdir(dirp))) {
- char *p = dp->d_name;
- p = strrchr(dp->d_name, '.');
- if (p != NULL && strcmp(p, VHDREXT) == 0) {
- int fd;
- char name[64];
- sprintf(name, "%s/%s", fileSysPath, dp->d_name);
- if ((fd = afs_open(name, O_RDONLY)) != -1
- && read(fd, (char *)&diskHeader, sizeof(diskHeader))
- == sizeof(diskHeader)
- && diskHeader.stamp.magic == VOLUMEHEADERMAGIC) {
- DiskToVolumeHeader(&vs.header, &diskHeader);
- nvols++;
- }
- close(fd);
- }
+ /* Count how many volumes we have in /vicepX */
+ code = VWalkVolumeHeaders(fileSysPartition, fileSysPath, CountHeader,
+ NULL, &nvols);
+ if (code < 0) {
+ Abort("Can't read directory %s; not salvaged\n", fileSysPath);
}
-#ifdef AFS_NT40_ENV
- closedir(dirp);
- dirp = opendir("."); /* No rewinddir for NT */
-#else
- rewinddir(dirp);
-#endif
if (!nvols)
nvols = 1;
- volumeSummaryp =
- (struct VolumeSummary *)malloc(nvols *
- sizeof(struct VolumeSummary));
- } else
- volumeSummaryp =
- (struct VolumeSummary *)malloc(20 * sizeof(struct VolumeSummary));
+ } else {
+ nvols = VOL_VG_MAX_VOLS;
+ }
+
+ volumeSummaryp = malloc(nvols * sizeof(struct VolumeSummary));
assert(volumeSummaryp != NULL);
- nVolumes = 0;
- vsp = volumeSummaryp;
- while ((dp = readdir(dirp))) {
- char *p = dp->d_name;
- p = strrchr(dp->d_name, '.');
- if (p != NULL && strcmp(p, VHDREXT) == 0) {
- int error = 0;
- int fd;
- char name[64];
- sprintf(name, "%s/%s", fileSysPath, dp->d_name);
- if ((fd = afs_open(name, O_RDONLY)) == -1
- || read(fd, &diskHeader, sizeof(diskHeader))
- != sizeof(diskHeader)
- || diskHeader.stamp.magic != VOLUMEHEADERMAGIC) {
- error = 1;
- }
- close(fd);
- if (error) {
- if (!singleVolumeNumber) {
- if (!Showmode)
- Log("%s/%s is not a legitimate volume header file; %sdeleted\n", fileSysPathName, dp->d_name, (Testing ? "it would have been " : ""));
- if (!Testing)
- unlink(dp->d_name);
- }
- } else {
- char nameShouldBe[64];
- DiskToVolumeHeader(&vsp->header, &diskHeader);
- if (singleVolumeNumber && vsp->header.id == singleVolumeNumber
- && vsp->header.parent != singleVolumeNumber) {
- if (programType == salvageServer) {
-#ifdef SALVSYNC_BUILD_CLIENT
- Log("fileserver requested salvage of clone %u; scheduling salvage of volume group %u...\n",
- vsp->header.id, vsp->header.parent);
- if (SALVSYNC_LinkVolume(vsp->header.parent,
- vsp->header.id,
- fileSysPartition->name,
- NULL) != SYNC_OK) {
- Log("schedule request failed\n");
- }
-#endif
- Exit(SALSRV_EXIT_VOLGROUP_LINK);
- } else {
- Log("%u is a read-only volume; not salvaged\n",
- singleVolumeNumber);
- Exit(1);
- }
- }
- if (!singleVolumeNumber
- || (vsp->header.id == singleVolumeNumber
- || vsp->header.parent == singleVolumeNumber)) {
- (void)afs_snprintf(nameShouldBe, sizeof nameShouldBe,
- VFORMAT, afs_printable_uint32_lu(vsp->header.id));
- if (singleVolumeNumber
- && vsp->header.id != singleVolumeNumber)
- AskOffline(vsp->header.id, fileSysPartition->name);
- if (strcmp(nameShouldBe, dp->d_name)) {
- if (!Showmode)
- Log("Volume header file %s is incorrectly named; %sdeleted (it will be recreated later, if necessary)\n", dp->d_name, (Testing ? "it would have been " : ""));
- if (!Testing)
- unlink(dp->d_name);
- } else {
- vsp->fileName = ToString(dp->d_name);
- nVolumes++;
- vsp++;
- }
- }
- }
- close(fd);
- }
+ params.singleVolumeNumber = singleVolumeNumber;
+ params.vsp = volumeSummaryp;
+ params.nVolumes = 0;
+ params.totalVolumes = nvols;
+
+ /* walk the partition directory of volume headers and record the info
+ * about them; unlinking invalid headers */
+ code = VWalkVolumeHeaders(fileSysPartition, fileSysPath, RecordHeader,
+ UnlinkHeader, ¶ms);
+ if (code < 0) {
+ Abort("Failed to get volume header summary\n");
}
- closedir(dirp);
+ nVolumes = params.nVolumes;
+
qsort(volumeSummaryp, nVolumes, sizeof(struct VolumeSummary),
CompareVolumes);
}
register struct ViceInodeInfo *inodes, int RW,
int check, int *deleteMe)
{
- int headerFd = 0;
int i;
register struct ViceInodeInfo *ip;
int allinodesobsolete = 1;
struct VolumeDiskHeader diskHeader;
+ afs_int32 (*writefunc)(VolumeDiskHeader_t *, struct DiskPartition64 *) = NULL;
+ int *skip;
+
+ /* keeps track of special inodes that are probably 'good'; they are
+ * referenced in the vol header, and are included in the given inodes
+ * array */
+ struct {
+ int valid;
+ Inode inode;
+ } goodspecial[MAXINODETYPE];
if (deleteMe)
*deleteMe = 0;
+
+ memset(goodspecial, 0, sizeof(goodspecial));
+
+ skip = malloc(isp->nSpecialInodes * sizeof(*skip));
+ if (skip) {
+ memset(skip, 0, isp->nSpecialInodes * sizeof(*skip));
+ } else {
+ Log("cannot allocate memory for inode skip array when salvaging "
+ "volume %lu; not performing duplicate special inode recovery\n",
+ afs_printable_uint32_lu(isp->volumeId));
+ /* still try to perform the salvage; the skip array only does anything
+ * if we detect duplicate special inodes */
+ }
+
+ /*
+ * First, look at the special inodes and see if any are referenced by
+ * the existing volume header. If we find duplicate special inodes, we
+ * can use this information to use the referenced inode (it's more
+ * likely to be the 'good' one), and throw away the duplicates.
+ */
+ if (isp->volSummary && skip) {
+ /* use tempHeader, so we can use the stuff[] array to easily index
+ * into the isp->volSummary special inodes */
+ memcpy(&tempHeader, &isp->volSummary->header, sizeof(struct VolumeHeader));
+
+ for (i = 0; i < isp->nSpecialInodes; i++) {
+ ip = &inodes[isp->index + i];
+ if (ip->u.special.type <= 0 || ip->u.special.type > MAXINODETYPE) {
+ /* will get taken care of in a later loop */
+ continue;
+ }
+ if (ip->inodeNumber == *(stuff[ip->u.special.type - 1].inode)) {
+ goodspecial[ip->u.special.type-1].valid = 1;
+ goodspecial[ip->u.special.type-1].inode = ip->inodeNumber;
+ }
+ }
+ }
+
memset(&tempHeader, 0, sizeof(tempHeader));
tempHeader.stamp.magic = VOLUMEHEADERMAGIC;
tempHeader.stamp.version = VOLUMEHEADERVERSION;
tempHeader.id = isp->volumeId;
tempHeader.parent = isp->RWvolumeId;
+
/* Check for duplicates (inodes are sorted by type field) */
for (i = 0; i < isp->nSpecialInodes - 1; i++) {
ip = &inodes[isp->index + i];
if (ip->u.special.type == (ip + 1)->u.special.type) {
- if (!Showmode)
- Log("Duplicate special inodes in volume header; salvage of volume %u aborted\n", isp->volumeId);
- return -1;
+ afs_ino_str_t stmp1, stmp2;
+
+ if (ip->u.special.type <= 0 || ip->u.special.type > MAXINODETYPE) {
+ /* Will be caught in the loop below */
+ continue;
+ }
+ if (!Showmode) {
+ Log("Duplicate special %d inodes for volume %u found (%s, %s);\n",
+ ip->u.special.type, isp->volumeId,
+ PrintInode(stmp1, ip->inodeNumber),
+ PrintInode(stmp2, (ip+1)->inodeNumber));
+ }
+ if (skip && goodspecial[ip->u.special.type-1].valid) {
+ Inode gi = goodspecial[ip->u.special.type-1].inode;
+
+ if (!Showmode) {
+ Log("using special inode referenced by vol header (%s)\n",
+ PrintInode(stmp1, gi));
+ }
+
+ /* the volume header references some special inode of
+ * this type in the inodes array; are we it? */
+ if (ip->inodeNumber != gi) {
+ skip[i] = 1;
+ } else if ((ip+1)->inodeNumber != gi) {
+ /* in case this is the last iteration; we need to
+ * make sure we check ip+1, too */
+ skip[i+1] = 1;
+ }
+ } else {
+ if (!Showmode)
+ Log("cannot determine which is correct; salvage of volume %u aborted\n", isp->volumeId);
+ if (skip) {
+ free(skip);
+ }
+ return -1;
+ }
}
}
for (i = 0; i < isp->nSpecialInodes; i++) {
ip = &inodes[isp->index + i];
if (ip->u.special.type <= 0 || ip->u.special.type > MAXINODETYPE) {
if (check) {
- Log("Rubbish header inode\n");
+ Log("Rubbish header inode %s of type %d\n",
+ PrintInode(NULL, ip->inodeNumber),
+ ip->u.special.type);
+ if (skip) {
+ free(skip);
+ }
return -1;
}
- Log("Rubbish header inode; deleted\n");
+ Log("Rubbish header inode %s of type %d; deleted\n",
+ PrintInode(NULL, ip->inodeNumber),
+ ip->u.special.type);
} else if (!stuff[ip->u.special.type - 1].obsolete) {
- *(stuff[ip->u.special.type - 1].inode) = ip->inodeNumber;
+ 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);
+ continue;
+ } else {
+ Log("Ignoring orphan special inode %s of type %d\n",
+ PrintInode(NULL, ip->inodeNumber), ip->u.special.type);
+ /* fall through to the ip->linkCount--; line below */
+ }
+ } else {
+ *(stuff[ip->u.special.type - 1].inode) = ip->inodeNumber;
+ allinodesobsolete = 0;
+ }
if (!check && ip->u.special.type != VI_LINKTABLE)
ip->linkCount--; /* Keep the inode around */
- allinodesobsolete = 0;
}
}
+ if (skip) {
+ free(skip);
+ }
+ skip = NULL;
if (allinodesobsolete) {
if (deleteMe)
Log("No header file for volume %u; %screating %s\n",
isp->volumeId, (Testing ? "it would have been " : ""),
path);
- headerFd = afs_open(path, O_RDWR | O_CREAT | O_TRUNC, 0644);
- assert(headerFd != -1);
isp->volSummary = (struct VolumeSummary *)
malloc(sizeof(struct VolumeSummary));
isp->volSummary->fileName = ToString(headerName);
+
+ writefunc = VCreateVolumeDiskHeader;
} else {
char path[64];
char headerName[64];
if (check)
return -1;
- headerFd = afs_open(path, O_RDWR | O_TRUNC, 0644);
- assert(headerFd != -1);
+ writefunc = VWriteVolumeDiskHeader;
}
}
- if (headerFd) {
+ if (writefunc) {
memcpy(&isp->volSummary->header, &tempHeader,
sizeof(struct VolumeHeader));
if (Testing) {
Log("It would have written a new header file for volume %u\n",
isp->volumeId);
} else {
+ afs_int32 code;
VolumeHeaderToDisk(&diskHeader, &tempHeader);
- if (write(headerFd, &diskHeader, sizeof(struct VolumeDiskHeader))
- != sizeof(struct VolumeDiskHeader)) {
- Log("Couldn't rewrite volume header file!\n");
- close(headerFd);
+ code = (*writefunc)(&diskHeader, fileSysPartition);
+ if (code) {
+ Log("Error %ld writing volume header file for volume %lu\n",
+ afs_printable_int32_ld(code),
+ afs_printable_uint32_lu(diskHeader.id));
return -1;
}
}
- close(headerFd);
}
IH_INIT(isp->volSummary->volumeInfoHandle, fileSysDevice, isp->RWvolumeId,
isp->volSummary->header.volumeInfo);
}
int
-JudgeEntry(struct DirSummary *dir, char *name, VnodeId vnodeNumber,
- Unique unique)
+JudgeEntry(void *dirVal, char *name, afs_int32 vnodeNumber,
+ afs_int32 unique)
{
+ struct DirSummary *dir = (struct DirSummary *)dirVal;
struct VnodeEssence *vnodeEssence;
afs_int32 dirOrphaned, todelete;
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")));
}
if (!Testing) {
- ViceFid fid;
+ AFSFid fid;
fid.Vnode = vnodeNumber;
fid.Unique = vnodeEssence->unique;
CopyOnWrite(dir);
if (strcmp(name, ".") == 0) {
if (dir->vnodeNumber != vnodeNumber || (dir->unique != unique)) {
- ViceFid fid;
+ AFSFid fid;
if (!Showmode)
Log("directory vnode %u.%u: bad '.' entry (was %u.%u); fixed\n", dir->vnodeNumber, dir->unique, vnodeNumber, unique);
if (!Testing) {
}
dir->haveDot = 1;
} else if (strcmp(name, "..") == 0) {
- ViceFid pa;
+ AFSFid pa;
if (dir->parent) {
struct VnodeEssence *dotdot;
pa.Vnode = dir->parent;
vep->owner = vnode->owner;
vep->group = vnode->group;
if (vnode->type == vDirectory) {
- assert(class == vLarge);
- vip->inodes[vnodeIndex] = VNDISK_GET_INO(vnode);
+ if (class != vLarge) {
+ VnodeId vnodeNumber = bitNumberToVnodeNumber(vnodeIndex, class);
+ vip->nAllocatedVnodes--;
+ memset(vnode, 0, sizeof(vnode));
+ IH_IWRITE(vnodeInfo[vSmall].handle,
+ vnodeIndexOffset(vcp, vnodeNumber),
+ (char *)&vnode, sizeof(vnode));
+ VolumeChanged = 1;
+ } else
+ vip->inodes[vnodeIndex] = VNDISK_GET_INO(vnode);
}
}
}
afs_int32 v, pv;
IHandle_t *h;
afs_sfsize_t nBytes;
- ViceFid pa;
+ AFSFid pa;
VnodeId LFVnode, ThisVnode;
Unique LFUnique, ThisUnique;
char npath[128];
* won't be visible there.
*/
if (class == vLarge) {
- ViceFid pa;
+ AFSFid pa;
DirHandle dh;
/* Remove and recreate the ".." entry in this orphaned directory */
if (!Showmode)
Log("it will be deleted instead. It should be recloned.\n");
}
- if (!Testing)
- unlink(isp->volSummary->fileName);
+ if (!Testing) {
+ afs_int32 code;
+ char path[64];
+ sprintf(path, "%s/%s", fileSysPath, isp->volSummary->fileName);
+
+ code = VDestroyVolumeDiskHeader(fileSysPartition, isp->volumeId, isp->RWvolumeId);
+ if (code) {
+ Log("Error %ld destroying volume disk header for volume %lu\n",
+ afs_printable_int32_ld(code),
+ afs_printable_uint32_lu(isp->volumeId));
+ }
+
+ /* make sure we actually delete the fileName file; ENOENT
+ * is fine, since VDestroyVolumeDiskHeader probably already
+ * unlinked it */
+ if (unlink(path) && errno != ENOENT) {
+ Log("Unable to unlink %s (errno = %d)\n", path, errno);
+ }
+ }
}
} else if (!check) {
Log("%s salvage was unsuccessful: read-write volume %u\n", message,
AskOffline(VolumeId volumeId, char * partition)
{
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, NULL);
+ code = FSYNC_VolOp(volumeId, partition, FSYNC_VOL_OFF, FSYNC_SALVAGE, &res);
if (code == SYNC_OK) {
break;
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;
+
+ code = VReadVolumeDiskHeader(volumeId, fileSysPartition, &diskHeader);
+ if (code) {
+ return;
+ }
+
+ DiskToVolumeHeader(&header, &diskHeader);
+
+ IH_INIT(h, fileSysDevice, header.parent, header.volumeInfo);
+ if (IH_IREAD(h, 0, (char*)&volHeader, sizeof(volHeader)) != sizeof(volHeader) ||
+ volHeader.stamp.magic != VOLUMEINFOMAGIC) {
+
+ IH_RELEASE(h);
+ return;
+ }
+
+ 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));
+
+ IH_RELEASE(h);
+ }
+#endif /* AFS_DEMAND_ATTACH_FS */
}
void
}
#endif
- rewind(logFile);
- fclose(logFile);
+ if (logFile) {
+ rewind(logFile);
+ fclose(logFile);
+ }
logFile = afs_fopen(AFSDIR_SERVER_SLVGLOG_FILEPATH, "r");
}
char *
-ToString(char *s)
+ToString(const char *s)
{
register char *p;
p = (char *)malloc(strlen(s) + 1);