2 * Copyright 2000, International Business Machines Corporation and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
12 * Module: vol-salvage.c
13 * Institution: The Information Technology Center, Carnegie-Mellon University
17 Correct handling of bad "." and ".." entries.
18 Message if volume has "destroyMe" flag set--but doesn't delete yet.
19 Link count bug fixed--bug was that vnodeEssence link count was unsigned
20 14 bits. Needs to be signed.
23 Change to DirHandle stuff to make sure that cache entries are reused at the
24 right time (this parallels the file server change, but is not identical).
26 Added calls to directory salvager routines; doesn't salvage dir unless debug=1.
29 Fixed bug which was causing inode link counts to go bad (thus leaking
31 Vnodes with 0 inode pointers in RW volumes are now deleted.
32 An inode with a matching inode number to the vnode is preferred to an
33 inode with a higer data version.
34 Bug is probably fixed that was causing data version to remain wrong,
35 despite assurances from the salvager to the contrary.
38 Added limited salvaging: unless ForceSalvage is on, then the volume will
39 not be salvaged if the dontSalvage flag is set in the Volume Header.
40 The ForceSalvage flag is turned on if an individual volume is salvaged or
41 if the file FORCESALVAGE exists in the partition header of the file system
42 being salvaged. This isn't used for anything but could be set by vfsck.
43 A -f flag was also added to force salvage.
46 It now deletes obsolete volume inodes without complaining
49 Repairs rw volume headers (again).
52 Correlates volume headers & inodes correctly, thus preventing occasional deletion
53 of read-only volumes...
54 No longer forces a directory salvage for volume 144 (which may be a good volume
56 Some of the messages are cleaned up or made more explicit. One or two added.
58 A bug was fixed which forced salvage of read-only volumes without a corresponding
62 When a volume header is recreated, the new name will be "bogus.volume#"
65 Directory salvaging turned on!!!
68 Prints warning messages for setuid programs.
71 Logs missing inode numbers.
74 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.
77 Locks the file /vice/vol/salvage.lock before starting. Aborts if it can't acquire the lock.
78 Time stamps on log entries.
79 Fcntl on stdout to cause all entries to be appended.
80 Problems writing to temporary files are now all detected.
81 Inode summary files are now dynamically named (so that multiple salvagers wouldn't conflict).
82 Some cleanup of error messages.
86 #include <afsconfig.h>
87 #include <afs/param.h>
89 #include <afs/procmgmt.h>
92 #ifdef HAVE_SYS_FILE_H
93 # include <sys/file.h>
97 #include <WINNT/afsevent.h>
100 #define WCOREDUMP(x) ((x) & 0200)
103 #include <afs/afsint.h>
104 #if !defined(AFS_SGI_ENV) && !defined(AFS_NT40_ENV)
105 #if defined(AFS_VFSINCL_ENV)
106 #include <sys/vnode.h>
108 #include <sys/fs/ufs_inode.h>
110 #if defined(AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV)
111 #include <ufs/ufs/dinode.h>
112 #include <ufs/ffs/fs.h>
114 #include <ufs/inode.h>
117 #else /* AFS_VFSINCL_ENV */
119 #include <ufs/inode.h>
120 #else /* AFS_OSF_ENV */
121 #if !defined(AFS_LINUX20_ENV) && !defined(AFS_XBSD_ENV) && !defined(AFS_DARWIN_ENV)
122 #include <sys/inode.h>
125 #endif /* AFS_VFSINCL_ENV */
126 #endif /* AFS_SGI_ENV */
129 #include <sys/lockf.h>
132 #include <checklist.h>
134 #if defined(AFS_SGI_ENV)
137 #if defined(AFS_SUN_ENV) || defined(AFS_SUN5_ENV)
139 #include <sys/mnttab.h>
140 #include <sys/mntent.h>
145 #endif /* AFS_SGI_ENV */
146 #endif /* AFS_HPUX_ENV */
150 #include <afs/osi_inode.h>
154 #include <afs/afsutil.h>
155 #include <afs/fileutil.h>
160 #include <afs/afssyscalls.h>
164 #include "partition.h"
165 #include "daemon_com.h"
166 #include "daemon_com_inline.h"
168 #include "fssync_inline.h"
169 #include "volume_inline.h"
170 #include "salvsync.h"
171 #include "viceinode.h"
173 #include "volinodes.h" /* header magic number, etc. stuff */
174 #include "vol-salvage.h"
176 #include "vol_internal.h"
178 #include <afs/prs_fs.h>
180 #ifdef FSSYNC_BUILD_CLIENT
181 #include "vg_cache.h"
189 extern void *calloc();
191 static char *TimeStamp(time_t clock, int precision);
194 int debug; /* -d flag */
195 extern int Testing; /* -n flag */
196 int ListInodeOption; /* -i flag */
197 int ShowRootFiles; /* -r flag */
198 int RebuildDirs; /* -sal flag */
199 int Parallel = 4; /* -para X flag */
200 int PartsPerDisk = 8; /* Salvage up to 8 partitions on same disk sequentially */
201 int forceR = 0; /* -b flag */
202 int ShowLog = 0; /* -showlog flag */
203 int ShowSuid = 0; /* -showsuid flag */
204 int ShowMounts = 0; /* -showmounts flag */
205 int orphans = ORPH_IGNORE; /* -orphans option */
210 int useSyslog = 0; /* -syslog flag */
211 int useSyslogFacility = LOG_DAEMON; /* -syslogfacility option */
220 #define MAXPARALLEL 32
222 int OKToZap; /* -o flag */
223 int ForceSalvage; /* If salvage should occur despite the DONT_SALVAGE flag
224 * in the volume header */
226 FILE *logFile = 0; /* one of {/usr/afs/logs,/vice/file}/SalvageLog */
228 #define ROOTINODE 2 /* Root inode of a 4.2 Unix file system
231 * information that is 'global' to a particular salvage job.
234 Device fileSysDevice; /**< The device number of the current partition
236 char fileSysPath[9]; /**< The path of the mounted partition currently
237 * being salvaged, i.e. the directory containing
238 * the volume headers */
239 char *fileSysPathName; /**< NT needs this to make name pretty log. */
240 IHandle_t *VGLinkH; /**< Link handle for current volume group. */
241 int VGLinkH_cnt; /**< # of references to lnk handle. */
242 struct DiskPartition64 *fileSysPartition; /**< Partition being salvaged */
245 char *fileSysDeviceName; /**< The block device where the file system being
246 * salvaged was mounted */
247 char *filesysfulldev;
249 int VolumeChanged; /**< Set by any routine which would change the
250 * volume in a way which would require callbacks
251 * to be broken if the volume was put back on
252 * on line by an active file server */
254 VolumeDiskData VolInfo; /**< A copy of the last good or salvaged volume
255 * header dealt with */
257 int nVolumesInInodeFile; /**< Number of read-write volumes summarized */
258 FD_t inodeFd; /**< File descriptor for inode file */
260 struct VolumeSummary *volumeSummaryp; /**< Holds all the volumes in a part */
261 int nVolumes; /**< Number of volumes (read-write and read-only)
262 * in volume summary */
263 struct InodeSummary *inodeSummary; /**< contains info on all the relevant
266 struct VnodeInfo vnodeInfo[nVNODECLASSES]; /**< contains info on all of the
267 * vnodes in the volume that
268 * we are currently looking
270 int useFSYNC; /**< 0 if the fileserver is unavailable; 1 if we should try
271 * to contact the fileserver over FSYNC */
278 /* Forward declarations */
279 static int IsVnodeOrphaned(struct SalvInfo *salvinfo, VnodeId vnode);
280 static int AskVolumeSummary(struct SalvInfo *salvinfo,
281 VolumeId singleVolumeNumber);
282 static void MaybeAskOnline(struct SalvInfo *salvinfo, VolumeId volumeId);
283 static void AskError(struct SalvInfo *salvinfo, VolumeId volumeId);
285 #if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
286 static int LockVolume(struct SalvInfo *salvinfo, VolumeId volumeId);
287 #endif /* AFS_DEMAND_ATTACH_FS || AFS_DEMAND_ATTACH_UTIL */
289 /* Uniquifier stored in the Inode */
294 return (u & 0x3fffff);
296 #if defined(AFS_SGI_EXMAG)
297 return (u & SGI_UNIQMASK);
300 #endif /* AFS_SGI_EXMAG */
307 if (aerror == EPERM || aerror == ENXIO || aerror == ENOENT)
309 return 0; /* otherwise may be transient, e.g. EMFILE */
314 char *save_args[MAX_ARGS];
316 extern pthread_t main_thread;
317 childJob_t myjob = { SALVAGER_MAGIC, NOT_CHILD, "" };
321 * Get the salvage lock if not already held. Hold until process exits.
323 * @param[in] locktype READ_LOCK or WRITE_LOCK
326 _ObtainSalvageLock(int locktype)
328 struct VLockFile salvageLock;
333 VLockFileInit(&salvageLock, AFSDIR_SERVER_SLVGLOCK_FILEPATH);
335 code = VLockFileLock(&salvageLock, offset, locktype, nonblock);
338 "salvager: There appears to be another salvager running! "
343 "salvager: Error %d trying to acquire salvage lock! "
349 ObtainSalvageLock(void)
351 _ObtainSalvageLock(WRITE_LOCK);
354 ObtainSharedSalvageLock(void)
356 _ObtainSalvageLock(READ_LOCK);
360 #ifdef AFS_SGI_XFS_IOPS_ENV
361 /* Check if the given partition is mounted. For XFS, the root inode is not a
362 * constant. So we check the hard way.
365 IsPartitionMounted(char *part)
368 struct mntent *mntent;
370 osi_Assert(mntfp = setmntent(MOUNTED, "r"));
371 while (mntent = getmntent(mntfp)) {
372 if (!strcmp(part, mntent->mnt_dir))
377 return mntent ? 1 : 1;
380 /* Check if the given inode is the root of the filesystem. */
381 #ifndef AFS_SGI_XFS_IOPS_ENV
383 IsRootInode(struct afs_stat_st *status)
386 * The root inode is not a fixed value in XFS partitions. So we need to
387 * see if the partition is in the list of mounted partitions. This only
388 * affects the SalvageFileSys path, so we check there.
390 return (status->st_ino == ROOTINODE);
395 #ifndef AFS_NAMEI_ENV
396 /* We don't want to salvage big files filesystems, since we can't put volumes on
400 CheckIfBigFilesFS(char *mountPoint, char *devName)
402 struct superblock fs;
405 if (strncmp(devName, "/dev/", 5)) {
406 (void)sprintf(name, "/dev/%s", devName);
408 (void)strcpy(name, devName);
411 if (ReadSuper(&fs, name) < 0) {
412 Log("Unable to read superblock. Not salvaging partition %s.\n",
416 if (IsBigFilesFileSystem(&fs)) {
417 Log("Partition %s is a big files filesystem, not salvaging.\n",
427 #define HDSTR "\\Device\\Harddisk"
428 #define HDLEN (sizeof(HDSTR)-1) /* Length of "\Device\Harddisk" */
430 SameDisk(struct DiskPartition64 *p1, struct DiskPartition64 *p2)
436 static int dowarn = 1;
438 if (!QueryDosDevice(p1->devName, res1, RES_LEN - 1))
440 if (strncmp(res1, HDSTR, HDLEN)) {
443 Log("WARNING: QueryDosDevice is returning %s, not %s for %s\n",
444 res1, HDSTR, p1->devName);
447 if (!QueryDosDevice(p2->devName, res2, RES_LEN - 1))
449 if (strncmp(res2, HDSTR, HDLEN)) {
452 Log("WARNING: QueryDosDevice is returning %s, not %s for %s\n",
453 res2, HDSTR, p2->devName);
457 return (0 == _strnicmp(res1, res2, RES_LEN - 1));
460 #define SameDisk(P1, P2) ((P1)->device/PartsPerDisk == (P2)->device/PartsPerDisk)
463 /* This assumes that two partitions with the same device number divided by
464 * PartsPerDisk are on the same disk.
467 SalvageFileSysParallel(struct DiskPartition64 *partP)
470 struct DiskPartition64 *partP;
471 int pid; /* Pid for this job */
472 int jobnumb; /* Log file job number */
473 struct job *nextjob; /* Next partition on disk to salvage */
475 static struct job *jobs[MAXPARALLEL] = { 0 }; /* Need to zero this */
476 struct job *thisjob = 0;
477 static int numjobs = 0;
478 static int jobcount = 0;
484 char logFileName[256];
488 /* We have a partition to salvage. Copy it into thisjob */
489 thisjob = (struct job *)malloc(sizeof(struct job));
491 Log("Can't salvage '%s'. Not enough memory\n", partP->name);
494 memset(thisjob, 0, sizeof(struct job));
495 thisjob->partP = partP;
496 thisjob->jobnumb = jobcount;
498 } else if (jobcount == 0) {
499 /* We are asking to wait for all jobs (partp == 0), yet we never
502 Log("No file system partitions named %s* found; not salvaged\n",
503 VICE_PARTITION_PREFIX);
507 if (debug || Parallel == 1) {
509 SalvageFileSys(thisjob->partP, 0);
516 /* Check to see if thisjob is for a disk that we are already
517 * salvaging. If it is, link it in as the next job to do. The
518 * jobs array has 1 entry per disk being salvages. numjobs is
519 * the total number of disks currently being salvaged. In
520 * order to keep thejobs array compact, when a disk is
521 * completed, the hightest element in the jobs array is moved
522 * down to now open slot.
524 for (j = 0; j < numjobs; j++) {
525 if (SameDisk(jobs[j]->partP, thisjob->partP)) {
526 /* On same disk, add it to this list and return */
527 thisjob->nextjob = jobs[j]->nextjob;
528 jobs[j]->nextjob = thisjob;
535 /* Loop until we start thisjob or until all existing jobs are finished */
536 while (thisjob || (!partP && (numjobs > 0))) {
537 startjob = -1; /* No new job to start */
539 if ((numjobs >= Parallel) || (!partP && (numjobs > 0))) {
540 /* Either the max jobs are running or we have to wait for all
541 * the jobs to finish. In either case, we wait for at least one
542 * job to finish. When it's done, clean up after it.
544 pid = wait(&wstatus);
545 osi_Assert(pid != -1);
546 for (j = 0; j < numjobs; j++) { /* Find which job it is */
547 if (pid == jobs[j]->pid)
550 osi_Assert(j < numjobs);
551 if (WCOREDUMP(wstatus)) { /* Say if the job core dumped */
552 Log("Salvage of %s core dumped!\n", jobs[j]->partP->name);
555 numjobs--; /* job no longer running */
556 oldjob = jobs[j]; /* remember */
557 jobs[j] = jobs[j]->nextjob; /* Step to next part on same disk */
558 free(oldjob); /* free the old job */
560 /* If there is another partition on the disk to salvage, then
561 * say we will start it (startjob). If not, then put thisjob there
562 * and say we will start it.
564 if (jobs[j]) { /* Another partitions to salvage */
565 startjob = j; /* Will start it */
566 } else { /* There is not another partition to salvage */
568 jobs[j] = thisjob; /* Add thisjob */
570 startjob = j; /* Will start it */
572 jobs[j] = jobs[numjobs]; /* Move last job up to this slot */
573 startjob = -1; /* Don't start it - already running */
577 /* We don't have to wait for a job to complete */
579 jobs[numjobs] = thisjob; /* Add this job */
581 startjob = numjobs; /* Will start it */
585 /* Start up a new salvage job on a partition in job slot "startjob" */
586 if (startjob != -1) {
588 Log("Starting salvage of file system partition %s\n",
589 jobs[startjob]->partP->name);
591 /* For NT, we not only fork, but re-exec the salvager. Pass in the
592 * commands and pass the child job number via the data path.
595 nt_SalvagePartition(jobs[startjob]->partP->name,
596 jobs[startjob]->jobnumb);
597 jobs[startjob]->pid = pid;
602 jobs[startjob]->pid = pid;
608 for (fd = 0; fd < 16; fd++)
615 openlog("salvager", LOG_PID, useSyslogFacility);
619 snprintf(logFileName, sizeof logFileName, "%s.%d",
620 AFSDIR_SERVER_SLVGLOG_FILEPATH,
621 jobs[startjob]->jobnumb);
622 logFile = afs_fopen(logFileName, "w");
627 SalvageFileSys1(jobs[startjob]->partP, 0);
632 } /* while ( thisjob || (!partP && numjobs > 0) ) */
634 /* If waited for all jobs to complete, now collect log files and return */
636 if (!useSyslog) /* if syslogging - no need to collect */
639 for (i = 0; i < jobcount; i++) {
640 snprintf(logFileName, sizeof logFileName, "%s.%d",
641 AFSDIR_SERVER_SLVGLOG_FILEPATH, i);
642 if ((passLog = afs_fopen(logFileName, "r"))) {
643 while (fgets(buf, sizeof(buf), passLog)) {
648 (void)unlink(logFileName);
657 SalvageFileSys(struct DiskPartition64 *partP, VolumeId singleVolumeNumber)
659 if (!canfork || debug || Fork() == 0) {
660 SalvageFileSys1(partP, singleVolumeNumber);
661 if (canfork && !debug) {
666 Wait("SalvageFileSys");
670 get_DevName(char *pbuffer, char *wpath)
672 char pbuf[128], *ptr;
673 strcpy(pbuf, pbuffer);
674 ptr = (char *)strrchr(pbuf, OS_DIRSEPC);
680 ptr = (char *)strrchr(pbuffer, OS_DIRSEPC);
682 strcpy(pbuffer, ptr + 1);
689 SalvageFileSys1(struct DiskPartition64 *partP, VolumeId singleVolumeNumber)
692 char inodeListPath[256];
693 FD_t inodeFile = INVALID_FD;
694 static char tmpDevName[100];
695 static char wpath[100];
696 struct VolumeSummary *vsp, *esp;
700 struct SalvInfo l_salvinfo;
701 struct SalvInfo *salvinfo = &l_salvinfo;
704 memset(salvinfo, 0, sizeof(*salvinfo));
707 if (inodeFile != INVALID_FD) {
709 inodeFile = INVALID_FD;
711 if (tries > VOL_MAX_CHECKOUT_RETRIES) {
712 Abort("Raced too many times with fileserver restarts while trying to "
713 "checkout/lock volumes; Aborted\n");
715 #if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
717 /* unlock all previous volume locks, since we're about to lock them
719 VLockFileReinit(&partP->volLockFile);
721 #endif /* AFS_DEMAND_ATTACH_FS || AFS_DEMAND_ATTACH_UTIL */
723 salvinfo->fileSysPartition = partP;
724 salvinfo->fileSysDevice = salvinfo->fileSysPartition->device;
725 salvinfo->fileSysPathName = VPartitionPath(salvinfo->fileSysPartition);
728 /* Opendir can fail on "C:" but not on "C:\" if C is empty! */
729 (void)sprintf(salvinfo->fileSysPath, "%s" OS_DIRSEP, salvinfo->fileSysPathName);
730 name = partP->devName;
732 strlcpy(salvinfo->fileSysPath, salvinfo->fileSysPathName, sizeof(salvinfo->fileSysPath));
733 strcpy(tmpDevName, partP->devName);
734 name = get_DevName(tmpDevName, wpath);
735 salvinfo->fileSysDeviceName = name;
736 salvinfo->filesysfulldev = wpath;
739 if (singleVolumeNumber) {
740 #if !(defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL))
741 /* only non-DAFS locks the partition when salvaging a single volume;
742 * DAFS will lock the individual volumes in the VG */
743 VLockPartition(partP->name);
744 #endif /* !(AFS_DEMAND_ATTACH_FS || AFS_DEMAND_ATTACH_UTIL) */
748 /* salvageserver already setup fssync conn for us */
749 if ((programType != salvageServer) && !VConnectFS()) {
750 Abort("Couldn't connect to file server\n");
753 salvinfo->useFSYNC = 1;
754 AskOffline(salvinfo, singleVolumeNumber);
755 #if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
756 if (LockVolume(salvinfo, singleVolumeNumber)) {
759 #endif /* AFS_DEMAND_ATTACH_FS || AFS_DEMAND_ATTACH_UTIL */
762 salvinfo->useFSYNC = 0;
763 VLockPartition(partP->name);
767 ForceSalvage = UseTheForceLuke(salvinfo->fileSysPath);
770 Log("SALVAGING FILE SYSTEM PARTITION %s (device=%s%s)\n",
771 partP->name, name, (Testing ? "(READONLY mode)" : ""));
773 Log("***Forced salvage of all volumes on this partition***\n");
778 * Remove any leftover /vicepa/salvage.inodes.* or /vicepa/salvage.temp.*
785 osi_Assert((dirp = opendir(salvinfo->fileSysPath)) != NULL);
786 while ((dp = readdir(dirp))) {
787 if (!strncmp(dp->d_name, "salvage.inodes.", 15)
788 || !strncmp(dp->d_name, "salvage.temp.", 13)) {
790 Log("Removing old salvager temp files %s\n", dp->d_name);
791 strcpy(npath, salvinfo->fileSysPath);
792 strcat(npath, OS_DIRSEP);
793 strcat(npath, dp->d_name);
799 tdir = (tmpdir ? tmpdir : salvinfo->fileSysPath);
801 (void)_putenv("TMP="); /* If "TMP" is set, then that overrides tdir. */
802 (void)strncpy(inodeListPath, _tempnam(tdir, "salvage.inodes."), 255);
804 snprintf(inodeListPath, 255, "%s" OS_DIRSEP "salvage.inodes.%s.%d", tdir, name,
808 inodeFile = OS_OPEN(inodeListPath, O_RDWR|O_TRUNC|O_CREAT, 0666);
809 if (inodeFile == INVALID_FD) {
810 Abort("Error %d when creating inode description file %s; not salvaged\n", errno, inodeListPath);
813 /* Using nt_unlink here since we're really using the delete on close
814 * semantics of unlink. In most places in the salvager, we really do
815 * mean to unlink the file at that point. Those places have been
816 * modified to actually do that so that the NT crt can be used there.
818 * jaltman - On NT delete on close cannot be applied to a file while the
819 * process has an open file handle that does not have DELETE file
820 * access and FILE_SHARE_DELETE. fopen() calls CreateFile() without
821 * delete privileges. As a result the nt_unlink() call will always
824 code = nt_unlink(inodeListPath);
826 code = unlink(inodeListPath);
829 Log("Error %d when trying to unlink %s\n", errno, inodeListPath);
832 if (GetInodeSummary(salvinfo, inodeFile, singleVolumeNumber) < 0) {
836 salvinfo->inodeFd = inodeFile;
837 if (salvinfo->inodeFd == INVALID_FD)
838 Abort("Temporary file %s is missing...\n", inodeListPath);
839 OS_SEEK(salvinfo->inodeFd, 0L, SEEK_SET);
840 if (ListInodeOption) {
841 PrintInodeList(salvinfo);
842 if (singleVolumeNumber) {
843 /* We've checked out the volume from the fileserver, and we need
844 * to give it back. We don't know if the volume exists or not,
845 * so we don't know whether to AskOnline or not. Try to determine
846 * if the volume exists by trying to read the volume header, and
847 * AskOnline if it is readable. */
848 MaybeAskOnline(salvinfo, singleVolumeNumber);
852 /* enumerate volumes in the partition.
853 * figure out sets of read-only + rw volumes.
854 * salvage each set, read-only volumes first, then read-write.
855 * Fix up inodes on last volume in set (whether it is read-write
858 if (GetVolumeSummary(salvinfo, singleVolumeNumber)) {
862 if (singleVolumeNumber) {
863 /* If we delete a volume during the salvage, we indicate as such by
864 * setting the volsummary->deleted field. We need to know if we
865 * deleted a volume or not in order to know which volumes to bring
866 * back online after the salvage. If we fork, we will lose this
867 * information, since volsummary->deleted will not get set in the
868 * parent. So, don't fork. */
872 for (i = j = 0, vsp = salvinfo->volumeSummaryp, esp = vsp + salvinfo->nVolumes;
873 i < salvinfo->nVolumesInInodeFile; i = j) {
874 VolumeId rwvid = salvinfo->inodeSummary[i].RWvolumeId;
876 j < salvinfo->nVolumesInInodeFile && salvinfo->inodeSummary[j].RWvolumeId == rwvid;
878 VolumeId vid = salvinfo->inodeSummary[j].volumeId;
879 struct VolumeSummary *tsp;
880 /* Scan volume list (from partition root directory) looking for the
881 * current rw volume number in the volume list from the inode scan.
882 * If there is one here that is not in the inode volume list,
884 for (; vsp < esp && (vsp->header.parent < rwvid); vsp++) {
886 DeleteExtraVolumeHeaderFile(salvinfo, vsp);
888 /* Now match up the volume summary info from the root directory with the
889 * entry in the volume list obtained from scanning inodes */
890 salvinfo->inodeSummary[j].volSummary = NULL;
891 for (tsp = vsp; tsp < esp && (tsp->header.parent == rwvid); tsp++) {
892 if (tsp->header.id == vid) {
893 salvinfo->inodeSummary[j].volSummary = tsp;
899 /* Salvage the group of volumes (several read-only + 1 read/write)
900 * starting with the current read-only volume we're looking at.
903 nt_SalvageVolumeGroup(salvinfo, &salvinfo->inodeSummary[i], j - i);
905 DoSalvageVolumeGroup(salvinfo, &salvinfo->inodeSummary[i], j - i);
906 #endif /* AFS_NT40_ENV */
910 /* Delete any additional volumes that were listed in the partition but which didn't have any corresponding inodes */
911 for (; vsp < esp; vsp++) {
913 DeleteExtraVolumeHeaderFile(salvinfo, vsp);
916 if (!singleVolumeNumber) /* Remove the FORCESALVAGE file */
917 RemoveTheForce(salvinfo->fileSysPath);
919 if (!Testing && singleVolumeNumber) {
921 #if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
922 /* unlock vol headers so the fs can attach them when we AskOnline */
923 VLockFileReinit(&salvinfo->fileSysPartition->volLockFile);
924 #endif /* AFS_DEMAND_ATTACH_FS || AFS_DEMAND_ATTACH_UTIL */
926 /* Step through the volumeSummary list and set all volumes on-line.
927 * Most volumes were taken off-line in GetVolumeSummary.
928 * If a volume was deleted, don't tell the fileserver anything, since
929 * we already told the fileserver the volume was deleted back when we
930 * we destroyed the volume header.
931 * Also, make sure we bring the singleVolumeNumber back online first.
934 for (j = 0; j < salvinfo->nVolumes; j++) {
935 if (salvinfo->volumeSummaryp[j].header.id == singleVolumeNumber) {
937 if (!salvinfo->volumeSummaryp[j].deleted) {
938 AskOnline(salvinfo, singleVolumeNumber);
944 /* If singleVolumeNumber is not in our volumeSummary, it means that
945 * at least one other volume in the VG is on the partition, but the
946 * RW volume is not. We've already AskOffline'd it by now, though,
947 * so make sure we don't still have the volume checked out. */
948 AskDelete(salvinfo, singleVolumeNumber);
951 for (j = 0; j < salvinfo->nVolumes; j++) {
952 if (salvinfo->volumeSummaryp[j].header.id != singleVolumeNumber) {
953 if (!salvinfo->volumeSummaryp[j].deleted) {
954 AskOnline(salvinfo, salvinfo->volumeSummaryp[j].header.id);
960 Log("SALVAGING OF PARTITION %s%s COMPLETED\n",
961 salvinfo->fileSysPartition->name, (Testing ? " (READONLY mode)" : ""));
964 OS_CLOSE(inodeFile); /* SalvageVolumeGroup was the last which needed it. */
968 DeleteExtraVolumeHeaderFile(struct SalvInfo *salvinfo, struct VolumeSummary *vsp)
971 char filename[VMAXPATHLEN];
977 VolumeExternalName_r(vsp->header.id, filename, sizeof(filename));
978 sprintf(path, "%s" OS_DIRSEP "%s", salvinfo->fileSysPath, filename);
981 Log("The volume header file %s is not associated with any actual data (%sdeleted)\n", path, (Testing ? "would have been " : ""));
984 code = VDestroyVolumeDiskHeader(salvinfo->fileSysPartition, vsp->header.id, vsp->header.parent);
986 Log("Error %ld destroying volume disk header for volume %lu\n",
987 afs_printable_int32_ld(code),
988 afs_printable_uint32_lu(vsp->header.id));
991 /* make sure we actually delete the header file; ENOENT
992 * is fine, since VDestroyVolumeDiskHeader probably already
994 if (unlink(path) && errno != ENOENT) {
995 Log("Unable to unlink %s (errno = %d)\n", path, errno);
997 if (salvinfo->useFSYNC) {
998 AskDelete(salvinfo, vsp->header.id);
1005 CompareInodes(const void *_p1, const void *_p2)
1007 const struct ViceInodeInfo *p1 = _p1;
1008 const struct ViceInodeInfo *p2 = _p2;
1009 if (p1->u.vnode.vnodeNumber == INODESPECIAL
1010 || p2->u.vnode.vnodeNumber == INODESPECIAL) {
1011 VolumeId p1rwid, p2rwid;
1013 (p1->u.vnode.vnodeNumber ==
1014 INODESPECIAL ? p1->u.special.parentId : p1->u.vnode.volumeId);
1016 (p2->u.vnode.vnodeNumber ==
1017 INODESPECIAL ? p2->u.special.parentId : p2->u.vnode.volumeId);
1018 if (p1rwid < p2rwid)
1020 if (p1rwid > p2rwid)
1022 if (p1->u.vnode.vnodeNumber == INODESPECIAL
1023 && p2->u.vnode.vnodeNumber == INODESPECIAL) {
1024 if (p1->u.vnode.volumeId == p2->u.vnode.volumeId)
1025 return (p1->u.special.type < p2->u.special.type ? -1 : 1);
1026 if (p1->u.vnode.volumeId == p1rwid)
1028 if (p2->u.vnode.volumeId == p2rwid)
1030 return (p1->u.vnode.volumeId < p2->u.vnode.volumeId ? -1 : 1);
1032 if (p1->u.vnode.vnodeNumber != INODESPECIAL)
1033 return (p2->u.vnode.volumeId == p2rwid ? 1 : -1);
1034 return (p1->u.vnode.volumeId == p1rwid ? -1 : 1);
1036 if (p1->u.vnode.volumeId < p2->u.vnode.volumeId)
1038 if (p1->u.vnode.volumeId > p2->u.vnode.volumeId)
1040 if (p1->u.vnode.vnodeNumber < p2->u.vnode.vnodeNumber)
1042 if (p1->u.vnode.vnodeNumber > p2->u.vnode.vnodeNumber)
1044 /* The following tests are reversed, so that the most desirable
1045 * of several similar inodes comes first */
1046 if (p1->u.vnode.vnodeUniquifier > p2->u.vnode.vnodeUniquifier) {
1047 #ifdef AFS_3DISPARES
1048 if (p1->u.vnode.vnodeUniquifier > 3775414 /* 90% of 4.2M */ &&
1049 p2->u.vnode.vnodeUniquifier < 419490 /* 10% of 4.2M */ )
1052 #ifdef AFS_SGI_EXMAG
1053 if (p1->u.vnode.vnodeUniquifier > 15099494 /* 90% of 16M */ &&
1054 p2->u.vnode.vnodeUniquifier < 1677721 /* 10% of 16M */ )
1059 if (p1->u.vnode.vnodeUniquifier < p2->u.vnode.vnodeUniquifier) {
1060 #ifdef AFS_3DISPARES
1061 if (p2->u.vnode.vnodeUniquifier > 3775414 /* 90% of 4.2M */ &&
1062 p1->u.vnode.vnodeUniquifier < 419490 /* 10% of 4.2M */ )
1065 #ifdef AFS_SGI_EXMAG
1066 if (p2->u.vnode.vnodeUniquifier > 15099494 /* 90% of 16M */ &&
1067 p1->u.vnode.vnodeUniquifier < 1677721 /* 10% of 16M */ )
1072 if (p1->u.vnode.inodeDataVersion > p2->u.vnode.inodeDataVersion) {
1073 #ifdef AFS_3DISPARES
1074 if (p1->u.vnode.inodeDataVersion > 1887437 /* 90% of 2.1M */ &&
1075 p2->u.vnode.inodeDataVersion < 209716 /* 10% of 2.1M */ )
1078 #ifdef AFS_SGI_EXMAG
1079 if (p1->u.vnode.inodeDataVersion > 15099494 /* 90% of 16M */ &&
1080 p2->u.vnode.inodeDataVersion < 1677721 /* 10% of 16M */ )
1085 if (p1->u.vnode.inodeDataVersion < p2->u.vnode.inodeDataVersion) {
1086 #ifdef AFS_3DISPARES
1087 if (p2->u.vnode.inodeDataVersion > 1887437 /* 90% of 2.1M */ &&
1088 p1->u.vnode.inodeDataVersion < 209716 /* 10% of 2.1M */ )
1091 #ifdef AFS_SGI_EXMAG
1092 if (p2->u.vnode.inodeDataVersion > 15099494 /* 90% of 16M */ &&
1093 p1->u.vnode.inodeDataVersion < 1677721 /* 10% of 16M */ )
1102 CountVolumeInodes(struct ViceInodeInfo *ip, int maxInodes,
1103 struct InodeSummary *summary)
1105 VolumeId volume = ip->u.vnode.volumeId;
1106 VolumeId rwvolume = volume;
1111 while (maxInodes-- && volume == ip->u.vnode.volumeId) {
1113 if (ip->u.vnode.vnodeNumber == INODESPECIAL) {
1115 rwvolume = ip->u.special.parentId;
1116 /* This isn't quite right, as there could (in error) be different
1117 * parent inodes in different special vnodes */
1119 if (maxunique < ip->u.vnode.vnodeUniquifier)
1120 maxunique = ip->u.vnode.vnodeUniquifier;
1124 summary->volumeId = volume;
1125 summary->RWvolumeId = rwvolume;
1126 summary->nInodes = n;
1127 summary->nSpecialInodes = nSpecial;
1128 summary->maxUniquifier = maxunique;
1132 OnlyOneVolume(struct ViceInodeInfo *inodeinfo, afs_uint32 singleVolumeNumber, void *rock)
1134 if (inodeinfo->u.vnode.vnodeNumber == INODESPECIAL)
1135 return (inodeinfo->u.special.parentId == singleVolumeNumber);
1136 return (inodeinfo->u.vnode.volumeId == singleVolumeNumber);
1141 * Collect list of inodes in file named by path. If a truly fatal error,
1142 * unlink the file and abort. For lessor errors, return -1. The file will
1143 * be unlinked by the caller.
1146 GetInodeSummary(struct SalvInfo *salvinfo, FD_t inodeFile, VolumeId singleVolumeNumber)
1150 struct ViceInodeInfo *ip, *ip_save;
1151 struct InodeSummary summary;
1152 char summaryFileName[50];
1153 FD_t summaryFile = INVALID_FD;
1155 char *dev = salvinfo->fileSysPath;
1156 char *wpath = salvinfo->fileSysPath;
1158 char *dev = salvinfo->fileSysDeviceName;
1159 char *wpath = salvinfo->filesysfulldev;
1161 char *part = salvinfo->fileSysPath;
1166 afs_sfsize_t st_size;
1168 /* This file used to come from vfsck; cobble it up ourselves now... */
1170 ListViceInodes(dev, salvinfo->fileSysPath, inodeFile,
1171 singleVolumeNumber ? OnlyOneVolume : 0,
1172 singleVolumeNumber, &forceSal, forceR, wpath, NULL)) < 0) {
1174 Log("*** I/O error %d when writing a tmp inode file; Not salvaged %s ***\nIncrease space on partition or use '-tmpdir'\n", errno, dev);
1178 Abort("Unable to get inodes for \"%s\"; not salvaged\n", dev);
1180 if (forceSal && !ForceSalvage) {
1181 Log("***Forced salvage of all volumes on this partition***\n");
1184 OS_SEEK(inodeFile, 0L, SEEK_SET);
1185 salvinfo->inodeFd = inodeFile;
1186 if (salvinfo->inodeFd == INVALID_FD ||
1187 (st_size = OS_SIZE(salvinfo->inodeFd)) == -1) {
1188 Abort("No inode description file for \"%s\"; not salvaged\n", dev);
1190 tdir = (tmpdir ? tmpdir : part);
1192 (void)_putenv("TMP="); /* If "TMP" is set, then that overrides tdir. */
1193 (void)strcpy(summaryFileName, _tempnam(tdir, "salvage.temp."));
1195 snprintf(summaryFileName, sizeof summaryFileName,
1196 "%s" OS_DIRSEP "salvage.temp.%d", tdir, getpid());
1198 summaryFile = OS_OPEN(summaryFileName, O_RDWR|O_APPEND|O_CREAT, 0666);
1199 if (summaryFile == INVALID_FD) {
1200 Abort("Unable to create inode summary file\n");
1204 /* Using nt_unlink here since we're really using the delete on close
1205 * semantics of unlink. In most places in the salvager, we really do
1206 * mean to unlink the file at that point. Those places have been
1207 * modified to actually do that so that the NT crt can be used there.
1209 * jaltman - As commented elsewhere, this cannot work because fopen()
1210 * does not open files with DELETE and FILE_SHARE_DELETE.
1212 code = nt_unlink(summaryFileName);
1214 code = unlink(summaryFileName);
1217 Log("Error %d when trying to unlink %s\n", errno, summaryFileName);
1220 if (!canfork || debug || Fork() == 0) {
1221 int nInodes = st_size / sizeof(struct ViceInodeInfo);
1223 OS_CLOSE(summaryFile);
1224 if (!singleVolumeNumber) /* Remove the FORCESALVAGE file */
1225 RemoveTheForce(salvinfo->fileSysPath);
1227 struct VolumeSummary *vsp;
1231 GetVolumeSummary(salvinfo, singleVolumeNumber);
1233 for (i = 0, vsp = salvinfo->volumeSummaryp; i < salvinfo->nVolumes; i++) {
1235 if (vsp->header.id == singleVolumeNumber) {
1238 DeleteExtraVolumeHeaderFile(salvinfo, vsp);
1244 MaybeAskOnline(salvinfo, singleVolumeNumber);
1246 /* make sure we get rid of stray .vol headers, even if
1247 * they're not in our volume summary (might happen if
1248 * e.g. something else created them and they're not in the
1249 * fileserver VGC) */
1250 VDestroyVolumeDiskHeader(salvinfo->fileSysPartition,
1251 singleVolumeNumber, 0 /*parent*/);
1252 AskDelete(salvinfo, singleVolumeNumber);
1256 Log("%s vice inodes on %s; not salvaged\n",
1257 singleVolumeNumber ? "No applicable" : "No", dev);
1262 ip = (struct ViceInodeInfo *)malloc(nInodes*sizeof(struct ViceInodeInfo));
1264 OS_CLOSE(summaryFile);
1266 ("Unable to allocate enough space to read inode table; %s not salvaged\n",
1269 if (OS_READ(salvinfo->inodeFd, ip, st_size) != st_size) {
1270 OS_CLOSE(summaryFile);
1271 Abort("Unable to read inode table; %s not salvaged\n", dev);
1273 qsort(ip, nInodes, sizeof(struct ViceInodeInfo), CompareInodes);
1274 if (OS_SEEK(salvinfo->inodeFd, 0, SEEK_SET) == -1
1275 || OS_WRITE(salvinfo->inodeFd, ip, st_size) != st_size) {
1276 OS_CLOSE(summaryFile);
1277 Abort("Unable to rewrite inode table; %s not salvaged\n", dev);
1282 CountVolumeInodes(ip, nInodes, &summary);
1283 if (OS_WRITE(summaryFile, &summary, sizeof(summary)) != sizeof(summary)) {
1284 Log("Difficulty writing summary file (errno = %d); %s not salvaged\n", errno, dev);
1285 OS_CLOSE(summaryFile);
1289 summary.index += (summary.nInodes);
1290 nInodes -= summary.nInodes;
1291 ip += summary.nInodes;
1294 ip = ip_save = NULL;
1295 /* Following fflush is not fclose, because if it was debug mode would not work */
1296 if (OS_SYNC(summaryFile) == -1) {
1297 Log("Unable to write summary file (errno = %d); %s not salvaged\n", errno, dev);
1298 OS_CLOSE(summaryFile);
1302 if (canfork && !debug) {
1307 if (Wait("Inode summary") == -1) {
1308 OS_CLOSE(summaryFile);
1309 Exit(1); /* salvage of this partition aborted */
1313 st_size = OS_SIZE(summaryFile);
1314 osi_Assert(st_size >= 0);
1317 salvinfo->inodeSummary = (struct InodeSummary *)malloc(st_size);
1318 osi_Assert(salvinfo->inodeSummary != NULL);
1319 /* For GNU we need to do lseek to get the file pointer moved. */
1320 osi_Assert(OS_SEEK(summaryFile, 0, SEEK_SET) == 0);
1321 ret = OS_READ(summaryFile, salvinfo->inodeSummary, st_size);
1322 osi_Assert(ret == st_size);
1324 salvinfo->nVolumesInInodeFile = st_size / sizeof(struct InodeSummary);
1325 for (i = 0; i < salvinfo->nVolumesInInodeFile; i++) {
1326 salvinfo->inodeSummary[i].volSummary = NULL;
1328 Log("%d nVolumesInInodeFile %lu \n",salvinfo->nVolumesInInodeFile,(unsigned long)st_size);
1329 OS_CLOSE(summaryFile);
1332 if (retcode && singleVolumeNumber && !deleted) {
1333 AskError(salvinfo, singleVolumeNumber);
1339 /* Comparison routine for volume sort.
1340 This is setup so that a read-write volume comes immediately before
1341 any read-only clones of that volume */
1343 CompareVolumes(const void *_p1, const void *_p2)
1345 const struct VolumeSummary *p1 = _p1;
1346 const struct VolumeSummary *p2 = _p2;
1347 if (p1->header.parent != p2->header.parent)
1348 return p1->header.parent < p2->header.parent ? -1 : 1;
1349 if (p1->header.id == p1->header.parent) /* p1 is rw volume */
1351 if (p2->header.id == p2->header.parent) /* p2 is rw volume */
1353 return p1->header.id < p2->header.id ? -1 : 1; /* Both read-only */
1357 * Gleans volumeSummary information by asking the fileserver
1359 * @param[in] singleVolumeNumber the volume we're salvaging. 0 if we're
1360 * salvaging a whole partition
1362 * @return whether we obtained the volume summary information or not
1363 * @retval 0 success; we obtained the volume summary information
1364 * @retval -1 we raced with a fileserver restart; volume locks and checkout
1366 * @retval 1 we did not get the volume summary information; either the
1367 * fileserver responded with an error, or we are not supposed to
1368 * ask the fileserver for the information (e.g. we are salvaging
1369 * the entire partition or we are not the salvageserver)
1371 * @note for non-DAFS, always returns 1
1374 AskVolumeSummary(struct SalvInfo *salvinfo, VolumeId singleVolumeNumber)
1377 #if defined(FSSYNC_BUILD_CLIENT) && defined(AFS_DEMAND_ATTACH_FS)
1378 if (programType == salvageServer) {
1379 if (singleVolumeNumber) {
1380 FSSYNC_VGQry_response_t q_res;
1382 struct VolumeSummary *vsp;
1384 struct VolumeDiskHeader diskHdr;
1386 memset(&res, 0, sizeof(res));
1388 code = FSYNC_VGCQuery(salvinfo->fileSysPartition->name, singleVolumeNumber, &q_res, &res);
1391 * We must wait for the partition to finish scanning before
1392 * can continue, since we will not know if we got the entire
1393 * VG membership unless the partition is fully scanned.
1394 * We could, in theory, just scan the partition ourselves if
1395 * the VG cache is not ready, but we would be doing the exact
1396 * same scan the fileserver is doing; it will almost always
1397 * be faster to wait for the fileserver. The only exceptions
1398 * are if the partition does not take very long to scan, and
1399 * in that case it's fast either way, so who cares?
1401 if (code == SYNC_FAILED && res.hdr.reason == FSYNC_PART_SCANNING) {
1402 Log("waiting for fileserver to finish scanning partition %s...\n",
1403 salvinfo->fileSysPartition->name);
1405 for (i = 1; code == SYNC_FAILED && res.hdr.reason == FSYNC_PART_SCANNING; i++) {
1406 /* linearly ramp up from 1 to 10 seconds; nothing fancy,
1407 * just so small partitions don't need to wait over 10
1408 * seconds every time, and large partitions are generally
1409 * polled only once every ten seconds. */
1410 sleep((i > 10) ? (i = 10) : i);
1412 code = FSYNC_VGCQuery(salvinfo->fileSysPartition->name, singleVolumeNumber, &q_res, &res);
1416 if (code == SYNC_FAILED && res.hdr.reason == FSYNC_UNKNOWN_VOLID) {
1417 /* This can happen if there's no header for the volume
1418 * we're salvaging, or no headers exist for the VG (if
1419 * we're salvaging an RW). Act as if we got a response
1420 * with no VG members. The headers may be created during
1421 * salvaging, if there are inodes in this VG. */
1423 memset(&q_res, 0, sizeof(q_res));
1424 q_res.rw = singleVolumeNumber;
1428 Log("fileserver refused VGCQuery request for volume %lu on "
1429 "partition %s, code %ld reason %ld\n",
1430 afs_printable_uint32_lu(singleVolumeNumber),
1431 salvinfo->fileSysPartition->name,
1432 afs_printable_int32_ld(code),
1433 afs_printable_int32_ld(res.hdr.reason));
1437 if (q_res.rw != singleVolumeNumber) {
1438 Log("fileserver requested salvage of clone %lu; scheduling salvage of volume group %lu...\n",
1439 afs_printable_uint32_lu(singleVolumeNumber),
1440 afs_printable_uint32_lu(q_res.rw));
1441 #ifdef SALVSYNC_BUILD_CLIENT
1442 if (SALVSYNC_LinkVolume(q_res.rw,
1444 salvinfo->fileSysPartition->name,
1446 Log("schedule request failed\n");
1448 #endif /* SALVSYNC_BUILD_CLIENT */
1449 Exit(SALSRV_EXIT_VOLGROUP_LINK);
1452 salvinfo->volumeSummaryp = calloc(VOL_VG_MAX_VOLS, sizeof(struct VolumeSummary));
1453 osi_Assert(salvinfo->volumeSummaryp != NULL);
1455 salvinfo->nVolumes = 0;
1456 vsp = salvinfo->volumeSummaryp;
1458 for (i = 0; i < VOL_VG_MAX_VOLS; i++) {
1459 char name[VMAXPATHLEN];
1461 if (!q_res.children[i]) {
1465 /* AskOffline for singleVolumeNumber was called much earlier */
1466 if (q_res.children[i] != singleVolumeNumber) {
1467 AskOffline(salvinfo, q_res.children[i]);
1468 if (LockVolume(salvinfo, q_res.children[i])) {
1474 code = VReadVolumeDiskHeader(q_res.children[i], salvinfo->fileSysPartition, &diskHdr);
1476 Log("Cannot read header for %lu; trying to salvage group anyway\n",
1477 afs_printable_uint32_lu(q_res.children[i]));
1482 DiskToVolumeHeader(&vsp->header, &diskHdr);
1483 VolumeExternalName_r(q_res.children[i], name, sizeof(name));
1485 salvinfo->nVolumes++;
1489 qsort(salvinfo->volumeSummaryp, salvinfo->nVolumes, sizeof(struct VolumeSummary),
1494 Log("Cannot get volume summary from fileserver; falling back to scanning "
1495 "entire partition\n");
1498 #endif /* FSSYNC_BUILD_CLIENT && AFS_DEMAND_ATTACH_FS */
1503 * count how many volume headers are found by VWalkVolumeHeaders.
1505 * @param[in] dp the disk partition (unused)
1506 * @param[in] name full path to the .vol header (unused)
1507 * @param[in] hdr the header data (unused)
1508 * @param[in] last whether this is the last try or not (unused)
1509 * @param[in] rock actually an afs_int32*; the running count of how many
1510 * volumes we have found
1515 CountHeader(struct DiskPartition64 *dp, const char *name,
1516 struct VolumeDiskHeader *hdr, int last, void *rock)
1518 afs_int32 *nvols = (afs_int32 *)rock;
1524 * parameters to pass to the VWalkVolumeHeaders callbacks when recording volume
1527 struct SalvageScanParams {
1528 VolumeId singleVolumeNumber; /**< 0 for a partition-salvage, otherwise the
1529 * vol id of the VG we're salvaging */
1530 struct VolumeSummary *vsp; /**< ptr to the current volume summary object
1531 * we're filling in */
1532 afs_int32 nVolumes; /**< # of vols we've encountered */
1533 afs_int32 totalVolumes; /**< max # of vols we should encounter (the
1534 * # of vols we've alloc'd memory for) */
1535 int retry; /**< do we need to retry vol lock/checkout? */
1536 struct SalvInfo *salvinfo; /**< salvage job info */
1540 * records volume summary info found from VWalkVolumeHeaders.
1542 * Found volumes are also taken offline if they are in the specific volume
1543 * group we are looking for.
1545 * @param[in] dp the disk partition
1546 * @param[in] name full path to the .vol header
1547 * @param[in] hdr the header data
1548 * @param[in] last 1 if this is the last try to read the header, 0 otherwise
1549 * @param[in] rock actually a struct SalvageScanParams*, containing the
1550 * information needed to record the volume summary data
1552 * @return operation status
1554 * @retval -1 volume locking raced with fileserver restart; checking out
1555 * and locking volumes needs to be retried
1556 * @retval 1 volume header is mis-named and should be deleted
1559 RecordHeader(struct DiskPartition64 *dp, const char *name,
1560 struct VolumeDiskHeader *hdr, int last, void *rock)
1562 char nameShouldBe[64];
1563 struct SalvageScanParams *params;
1564 struct VolumeSummary summary;
1565 VolumeId singleVolumeNumber;
1566 struct SalvInfo *salvinfo;
1568 params = (struct SalvageScanParams *)rock;
1570 memset(&summary, 0, sizeof(summary));
1572 singleVolumeNumber = params->singleVolumeNumber;
1573 salvinfo = params->salvinfo;
1575 DiskToVolumeHeader(&summary.header, hdr);
1577 if (singleVolumeNumber && summary.header.id == singleVolumeNumber
1578 && summary.header.parent != singleVolumeNumber) {
1580 if (programType == salvageServer) {
1581 #ifdef SALVSYNC_BUILD_CLIENT
1582 Log("fileserver requested salvage of clone %u; scheduling salvage of volume group %u...\n",
1583 summary.header.id, summary.header.parent);
1584 if (SALVSYNC_LinkVolume(summary.header.parent,
1588 Log("schedule request failed\n");
1591 Exit(SALSRV_EXIT_VOLGROUP_LINK);
1594 Log("%u is a read-only volume; not salvaged\n",
1595 singleVolumeNumber);
1600 if (!singleVolumeNumber || summary.header.id == singleVolumeNumber
1601 || summary.header.parent == singleVolumeNumber) {
1603 /* check if the header file is incorrectly named */
1605 const char *base = strrchr(name, OS_DIRSEPC);
1612 snprintf(nameShouldBe, sizeof nameShouldBe,
1613 VFORMAT, afs_printable_uint32_lu(summary.header.id));
1616 if (strcmp(nameShouldBe, base)) {
1617 /* .vol file has wrong name; retry/delete */
1621 if (!badname || last) {
1622 /* only offline the volume if the header is good, or if this is
1623 * the last try looking at it; avoid AskOffline'ing the same vol
1626 if (singleVolumeNumber
1627 && summary.header.id != singleVolumeNumber) {
1628 /* don't offline singleVolumeNumber; we already did that
1631 AskOffline(salvinfo, summary.header.id);
1633 #if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
1635 /* don't lock the volume if the header is bad, since we're
1636 * about to delete it anyway. */
1637 if (LockVolume(salvinfo, summary.header.id)) {
1642 #endif /* AFS_DEMAND_ATTACH_FS || AFS_DEMAND_ATTACH_UTIL */
1646 if (last && !Showmode) {
1647 Log("Volume header file %s is incorrectly named (should be %s "
1648 "not %s); %sdeleted (it will be recreated later, if "
1649 "necessary)\n", name, nameShouldBe, base,
1650 (Testing ? "it would have been " : ""));
1658 if (params->nVolumes > params->totalVolumes) {
1659 /* We found more volumes than we found on the first partition walk;
1660 * apparently something created a volume while we were
1661 * partition-salvaging, or we found more than 20 vols when salvaging a
1662 * particular volume. Abort if we detect this, since other programs
1663 * supposed to not touch the partition while it is partition-salvaging,
1664 * and we shouldn't find more than 20 vols in a VG.
1666 Abort("Found %ld vol headers, but should have found at most %ld! "
1667 "Make sure the volserver/fileserver are not running at the "
1668 "same time as a partition salvage\n",
1669 afs_printable_int32_ld(params->nVolumes),
1670 afs_printable_int32_ld(params->totalVolumes));
1673 memcpy(params->vsp, &summary, sizeof(summary));
1681 * possibly unlinks bad volume headers found from VWalkVolumeHeaders.
1683 * If the header could not be read in at all, the header is always unlinked.
1684 * If instead RecordHeader said the header was bad (that is, the header file
1685 * is mis-named), we only unlink if we are doing a partition salvage, as
1686 * opposed to salvaging a specific volume group.
1688 * @param[in] dp the disk partition
1689 * @param[in] name full path to the .vol header
1690 * @param[in] hdr header data, or NULL if the header could not be read
1691 * @param[in] rock actually a struct SalvageScanParams*, with some information
1695 UnlinkHeader(struct DiskPartition64 *dp, const char *name,
1696 struct VolumeDiskHeader *hdr, void *rock)
1698 struct SalvageScanParams *params;
1701 params = (struct SalvageScanParams *)rock;
1704 /* no header; header is too bogus to read in at all */
1706 Log("%s is not a legitimate volume header file; %sdeleted\n", name, (Testing ? "it would have been " : ""));
1712 } else if (!params->singleVolumeNumber) {
1713 /* We were able to read in a header, but RecordHeader said something
1714 * was wrong with it. We only unlink those if we are doing a partition
1721 if (dounlink && unlink(name)) {
1722 Log("Error %d while trying to unlink %s\n", errno, name);
1727 * Populates salvinfo->volumeSummaryp with volume summary information, either by asking
1728 * the fileserver for VG information, or by scanning the /vicepX partition.
1730 * @param[in] singleVolumeNumber the volume ID of the single volume group we
1731 * are salvaging, or 0 if this is a partition
1734 * @return operation status
1736 * @retval -1 we raced with a fileserver restart; checking out and locking
1737 * volumes must be retried
1740 GetVolumeSummary(struct SalvInfo *salvinfo, VolumeId singleVolumeNumber)
1742 afs_int32 nvols = 0;
1743 struct SalvageScanParams params;
1746 code = AskVolumeSummary(salvinfo, singleVolumeNumber);
1748 /* we successfully got the vol information from the fileserver; no
1749 * need to scan the partition */
1753 /* we need to retry volume checkout */
1757 if (!singleVolumeNumber) {
1758 /* Count how many volumes we have in /vicepX */
1759 code = VWalkVolumeHeaders(salvinfo->fileSysPartition, salvinfo->fileSysPath, CountHeader,
1762 Abort("Can't read directory %s; not salvaged\n", salvinfo->fileSysPath);
1767 nvols = VOL_VG_MAX_VOLS;
1770 salvinfo->volumeSummaryp = calloc(nvols, sizeof(struct VolumeSummary));
1771 osi_Assert(salvinfo->volumeSummaryp != NULL);
1773 params.singleVolumeNumber = singleVolumeNumber;
1774 params.vsp = salvinfo->volumeSummaryp;
1775 params.nVolumes = 0;
1776 params.totalVolumes = nvols;
1778 params.salvinfo = salvinfo;
1780 /* walk the partition directory of volume headers and record the info
1781 * about them; unlinking invalid headers */
1782 code = VWalkVolumeHeaders(salvinfo->fileSysPartition, salvinfo->fileSysPath, RecordHeader,
1783 UnlinkHeader, ¶ms);
1785 /* we apparently need to retry checking-out/locking volumes */
1789 Abort("Failed to get volume header summary\n");
1791 salvinfo->nVolumes = params.nVolumes;
1793 qsort(salvinfo->volumeSummaryp, salvinfo->nVolumes, sizeof(struct VolumeSummary),
1799 /* Find the link table. This should be associated with the RW volume or, if
1800 * a RO only site, then the RO volume. For now, be cautious and hunt carefully.
1803 FindLinkHandle(struct InodeSummary *isp, int nVols,
1804 struct ViceInodeInfo *allInodes)
1807 struct ViceInodeInfo *ip;
1809 for (i = 0; i < nVols; i++) {
1810 ip = allInodes + isp[i].index;
1811 for (j = 0; j < isp[i].nSpecialInodes; j++) {
1812 if (ip[j].u.special.type == VI_LINKTABLE)
1813 return ip[j].inodeNumber;
1820 CreateLinkTable(struct SalvInfo *salvinfo, struct InodeSummary *isp, Inode ino)
1822 struct versionStamp version;
1825 if (!VALID_INO(ino))
1827 IH_CREATE(NULL, salvinfo->fileSysDevice, salvinfo->fileSysPath, 0, isp->RWvolumeId,
1828 INODESPECIAL, VI_LINKTABLE, isp->RWvolumeId);
1829 if (!VALID_INO(ino))
1831 ("Unable to allocate link table inode for volume %u (error = %d)\n",
1832 isp->RWvolumeId, errno);
1833 IH_INIT(salvinfo->VGLinkH, salvinfo->fileSysDevice, isp->RWvolumeId, ino);
1834 fdP = IH_OPEN(salvinfo->VGLinkH);
1836 Abort("Can't open link table for volume %u (error = %d)\n",
1837 isp->RWvolumeId, errno);
1839 if (FDH_TRUNC(fdP, sizeof(version) + sizeof(short)) < 0)
1840 Abort("Can't truncate link table for volume %u (error = %d)\n",
1841 isp->RWvolumeId, errno);
1843 version.magic = LINKTABLEMAGIC;
1844 version.version = LINKTABLEVERSION;
1846 if (FDH_PWRITE(fdP, (char *)&version, sizeof(version), 0)
1848 Abort("Can't truncate link table for volume %u (error = %d)\n",
1849 isp->RWvolumeId, errno);
1851 FDH_REALLYCLOSE(fdP);
1853 /* If the volume summary exits (i.e., the V*.vol header file exists),
1854 * then set this inode there as well.
1856 if (isp->volSummary)
1857 isp->volSummary->header.linkTable = ino;
1866 SVGParms_t *parms = (SVGParms_t *) arg;
1867 DoSalvageVolumeGroup(parms->svgp_salvinfo, parms->svgp_inodeSummaryp, parms->svgp_count);
1872 nt_SalvageVolumeGroup(struct SalvInfo *salvinfo, struct InodeSummary *isp, int nVols)
1875 pthread_attr_t tattr;
1879 /* Initialize per volume global variables, even if later code does so */
1880 salvinfo->VolumeChanged = 0;
1881 salvinfo->VGLinkH = NULL;
1882 salvinfo->VGLinkH_cnt = 0;
1883 memset(&salvinfo->VolInfo, 0, sizeof(salvinfo->VolInfo));
1885 parms.svgp_inodeSummaryp = isp;
1886 parms.svgp_count = nVols;
1887 parms.svgp_salvinfo = salvinfo;
1888 code = pthread_attr_init(&tattr);
1890 Log("Failed to salvage volume group %u: pthread_attr_init()\n",
1894 code = pthread_attr_setdetachstate(&tattr, PTHREAD_CREATE_JOINABLE);
1896 Log("Failed to salvage volume group %u: pthread_attr_setdetachstate()\n", isp->RWvolumeId);
1899 code = pthread_create(&tid, &tattr, nt_SVG, &parms);
1901 Log("Failed to create thread to salvage volume group %u\n",
1905 (void)pthread_join(tid, NULL);
1907 #endif /* AFS_NT40_ENV */
1910 DoSalvageVolumeGroup(struct SalvInfo *salvinfo, struct InodeSummary *isp, int nVols)
1912 struct ViceInodeInfo *inodes, *allInodes, *ip;
1913 int i, totalInodes, size, salvageTo;
1917 int dec_VGLinkH = 0;
1919 FdHandle_t *fdP = NULL;
1921 salvinfo->VGLinkH_cnt = 0;
1922 haveRWvolume = (isp->volumeId == isp->RWvolumeId
1923 && isp->nSpecialInodes > 0);
1924 if ((!ShowMounts) || (ShowMounts && !haveRWvolume)) {
1925 if (!ForceSalvage && QuickCheck(salvinfo, isp, nVols))
1928 if (ShowMounts && !haveRWvolume)
1930 if (canfork && !debug && Fork() != 0) {
1931 (void)Wait("Salvage volume group");
1934 for (i = 0, totalInodes = 0; i < nVols; i++)
1935 totalInodes += isp[i].nInodes;
1936 size = totalInodes * sizeof(struct ViceInodeInfo);
1937 inodes = (struct ViceInodeInfo *)malloc(size);
1938 allInodes = inodes - isp->index; /* this would the base of all the inodes
1939 * for the partition, if all the inodes
1940 * had been read into memory */
1942 (salvinfo->inodeFd, isp->index * sizeof(struct ViceInodeInfo),
1944 osi_Assert(OS_READ(salvinfo->inodeFd, inodes, size) == size);
1946 /* Don't try to salvage a read write volume if there isn't one on this
1948 salvageTo = haveRWvolume ? 0 : 1;
1950 #ifdef AFS_NAMEI_ENV
1951 ino = FindLinkHandle(isp, nVols, allInodes);
1952 if (VALID_INO(ino)) {
1953 IH_INIT(salvinfo->VGLinkH, salvinfo->fileSysDevice, isp->RWvolumeId, ino);
1954 fdP = IH_OPEN(salvinfo->VGLinkH);
1956 if (VALID_INO(ino) && fdP != NULL) {
1957 struct versionStamp header;
1958 afs_sfsize_t nBytes;
1960 nBytes = FDH_PREAD(fdP, (char *)&header, sizeof(struct versionStamp), 0);
1961 if (nBytes != sizeof(struct versionStamp)
1962 || header.magic != LINKTABLEMAGIC) {
1963 Log("Bad linktable header for volume %u.\n", isp->RWvolumeId);
1964 FDH_REALLYCLOSE(fdP);
1968 if (!VALID_INO(ino) || fdP == NULL) {
1969 Log("%s link table for volume %u.\n",
1970 Testing ? "Would have recreated" : "Recreating", isp->RWvolumeId);
1972 IH_INIT(salvinfo->VGLinkH, salvinfo->fileSysDevice, -1, -1);
1975 struct ViceInodeInfo *ip;
1976 CreateLinkTable(salvinfo, isp, ino);
1977 fdP = IH_OPEN(salvinfo->VGLinkH);
1978 /* Sync fake 1 link counts to the link table, now that it exists */
1980 for (i = 0; i < nVols; i++) {
1981 ip = allInodes + isp[i].index;
1982 for (j = isp[i].nSpecialInodes; j < isp[i].nInodes; j++) {
1983 namei_SetLinkCount(fdP, ip[j].inodeNumber, 1, 1);
1990 FDH_REALLYCLOSE(fdP);
1992 IH_INIT(salvinfo->VGLinkH, salvinfo->fileSysDevice, -1, -1);
1995 /* Salvage in reverse order--read/write volume last; this way any
1996 * Inodes not referenced by the time we salvage the read/write volume
1997 * can be picked up by the read/write volume */
1998 /* ACTUALLY, that's not done right now--the inodes just vanish */
1999 for (i = nVols - 1; i >= salvageTo; i--) {
2001 struct InodeSummary *lisp = &isp[i];
2002 #ifdef AFS_NAMEI_ENV
2003 /* If only the RO is present on this partition, the link table
2004 * shows up as a RW volume special file. Need to make sure the
2005 * salvager doesn't try to salvage the non-existent RW.
2007 if (rw && nVols > 1 && isp[i].nSpecialInodes == 1) {
2008 /* If this only special inode is the link table, continue */
2009 if (inodes->u.special.type == VI_LINKTABLE) {
2016 Log("%s VOLUME %u%s.\n", rw ? "SALVAGING" : "CHECKING CLONED",
2017 lisp->volumeId, (Testing ? "(READONLY mode)" : ""));
2018 /* Check inodes twice. The second time do things seriously. This
2019 * way the whole RO volume can be deleted, below, if anything goes wrong */
2020 for (check = 1; check >= 0; check--) {
2022 if (SalvageVolumeHeaderFile(salvinfo, lisp, allInodes, rw, check, &deleteMe)
2024 MaybeZapVolume(salvinfo, lisp, "Volume header", deleteMe, check);
2025 if (rw && deleteMe) {
2026 haveRWvolume = 0; /* This will cause its inodes to be deleted--since salvage
2027 * volume won't be called */
2033 if (rw && check == 1)
2035 if (SalvageVnodes(salvinfo, isp, lisp, allInodes, check) == -1) {
2036 MaybeZapVolume(salvinfo, lisp, "Vnode index", 0, check);
2042 /* Fix actual inode counts */
2045 Log("totalInodes %d\n",totalInodes);
2046 for (ip = inodes; totalInodes; ip++, totalInodes--) {
2047 static int TraceBadLinkCounts = 0;
2048 #ifdef AFS_NAMEI_ENV
2049 if (salvinfo->VGLinkH->ih_ino == ip->inodeNumber) {
2050 dec_VGLinkH = ip->linkCount - salvinfo->VGLinkH_cnt;
2051 VGLinkH_p1 = ip->u.param[0];
2052 continue; /* Deal with this last. */
2055 if (ip->linkCount != 0 && TraceBadLinkCounts) {
2056 TraceBadLinkCounts--; /* Limit reports, per volume */
2057 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]);
2059 while (ip->linkCount > 0) {
2060 /* below used to assert, not break */
2062 if (IH_DEC(salvinfo->VGLinkH, ip->inodeNumber, ip->u.param[0])) {
2063 Log("idec failed. inode %s errno %d\n",
2064 PrintInode(stmp, ip->inodeNumber), errno);
2070 while (ip->linkCount < 0) {
2071 /* these used to be asserts */
2073 if (IH_INC(salvinfo->VGLinkH, ip->inodeNumber, ip->u.param[0])) {
2074 Log("iinc failed. inode %s errno %d\n",
2075 PrintInode(stmp, ip->inodeNumber), errno);
2082 #ifdef AFS_NAMEI_ENV
2083 while (dec_VGLinkH > 0) {
2084 if (IH_DEC(salvinfo->VGLinkH, salvinfo->VGLinkH->ih_ino, VGLinkH_p1) < 0) {
2085 Log("idec failed on link table, errno = %d\n", errno);
2089 while (dec_VGLinkH < 0) {
2090 if (IH_INC(salvinfo->VGLinkH, salvinfo->VGLinkH->ih_ino, VGLinkH_p1) < 0) {
2091 Log("iinc failed on link table, errno = %d\n", errno);
2098 /* Directory consistency checks on the rw volume */
2100 SalvageVolume(salvinfo, isp, salvinfo->VGLinkH);
2101 IH_RELEASE(salvinfo->VGLinkH);
2103 if (canfork && !debug) {
2110 QuickCheck(struct SalvInfo *salvinfo, struct InodeSummary *isp, int nVols)
2112 /* Check headers BEFORE forking */
2116 for (i = 0; i < nVols; i++) {
2117 struct VolumeSummary *vs = isp[i].volSummary;
2118 VolumeDiskData volHeader;
2120 /* Don't salvage just because phantom rw volume is there... */
2121 /* (If a read-only volume exists, read/write inodes must also exist) */
2122 if (i == 0 && isp->nSpecialInodes == 0 && nVols > 1)
2126 IH_INIT(h, salvinfo->fileSysDevice, vs->header.parent, vs->header.volumeInfo);
2127 if (IH_IREAD(h, 0, (char *)&volHeader, sizeof(volHeader))
2128 == sizeof(volHeader)
2129 && volHeader.stamp.magic == VOLUMEINFOMAGIC
2130 && volHeader.dontSalvage == DONT_SALVAGE
2131 && volHeader.needsSalvaged == 0 && volHeader.destroyMe == 0) {
2132 if (volHeader.inUse != 0) {
2133 volHeader.inUse = 0;
2134 volHeader.inService = 1;
2136 if (IH_IWRITE(h, 0, (char *)&volHeader, sizeof(volHeader))
2137 != sizeof(volHeader)) {
2153 /* SalvageVolumeHeaderFile
2155 * Salvage the top level V*.vol header file. Make sure the special files
2156 * exist and that there are no duplicates.
2158 * Calls SalvageHeader for each possible type of volume special file.
2162 SalvageVolumeHeaderFile(struct SalvInfo *salvinfo, struct InodeSummary *isp,
2163 struct ViceInodeInfo *inodes, int RW,
2164 int check, int *deleteMe)
2167 struct ViceInodeInfo *ip;
2168 int allinodesobsolete = 1;
2169 struct VolumeDiskHeader diskHeader;
2170 afs_int32 (*writefunc)(VolumeDiskHeader_t *, struct DiskPartition64 *) = NULL;
2172 struct VolumeHeader tempHeader;
2173 struct afs_inode_info stuff[MAXINODETYPE];
2175 /* keeps track of special inodes that are probably 'good'; they are
2176 * referenced in the vol header, and are included in the given inodes
2181 } goodspecial[MAXINODETYPE];
2186 memset(goodspecial, 0, sizeof(goodspecial));
2188 skip = malloc(isp->nSpecialInodes * sizeof(*skip));
2190 memset(skip, 0, isp->nSpecialInodes * sizeof(*skip));
2192 Log("cannot allocate memory for inode skip array when salvaging "
2193 "volume %lu; not performing duplicate special inode recovery\n",
2194 afs_printable_uint32_lu(isp->volumeId));
2195 /* still try to perform the salvage; the skip array only does anything
2196 * if we detect duplicate special inodes */
2199 init_inode_info(&tempHeader, stuff);
2202 * First, look at the special inodes and see if any are referenced by
2203 * the existing volume header. If we find duplicate special inodes, we
2204 * can use this information to use the referenced inode (it's more
2205 * likely to be the 'good' one), and throw away the duplicates.
2207 if (isp->volSummary && skip) {
2208 /* use tempHeader, so we can use the stuff[] array to easily index
2209 * into the isp->volSummary special inodes */
2210 memcpy(&tempHeader, &isp->volSummary->header, sizeof(struct VolumeHeader));
2212 for (i = 0; i < isp->nSpecialInodes; i++) {
2213 ip = &inodes[isp->index + i];
2214 if (ip->u.special.type <= 0 || ip->u.special.type > MAXINODETYPE) {
2215 /* will get taken care of in a later loop */
2218 if (ip->inodeNumber == *(stuff[ip->u.special.type - 1].inode)) {
2219 goodspecial[ip->u.special.type-1].valid = 1;
2220 goodspecial[ip->u.special.type-1].inode = ip->inodeNumber;
2225 memset(&tempHeader, 0, sizeof(tempHeader));
2226 tempHeader.stamp.magic = VOLUMEHEADERMAGIC;
2227 tempHeader.stamp.version = VOLUMEHEADERVERSION;
2228 tempHeader.id = isp->volumeId;
2229 tempHeader.parent = isp->RWvolumeId;
2231 /* Check for duplicates (inodes are sorted by type field) */
2232 for (i = 0; i < isp->nSpecialInodes - 1; i++) {
2233 ip = &inodes[isp->index + i];
2234 if (ip->u.special.type == (ip + 1)->u.special.type) {
2235 afs_ino_str_t stmp1, stmp2;
2237 if (ip->u.special.type <= 0 || ip->u.special.type > MAXINODETYPE) {
2238 /* Will be caught in the loop below */
2242 Log("Duplicate special %d inodes for volume %u found (%s, %s);\n",
2243 ip->u.special.type, isp->volumeId,
2244 PrintInode(stmp1, ip->inodeNumber),
2245 PrintInode(stmp2, (ip+1)->inodeNumber));
2247 if (skip && goodspecial[ip->u.special.type-1].valid) {
2248 Inode gi = goodspecial[ip->u.special.type-1].inode;
2251 Log("using special inode referenced by vol header (%s)\n",
2252 PrintInode(stmp1, gi));
2255 /* the volume header references some special inode of
2256 * this type in the inodes array; are we it? */
2257 if (ip->inodeNumber != gi) {
2259 } else if ((ip+1)->inodeNumber != gi) {
2260 /* in case this is the last iteration; we need to
2261 * make sure we check ip+1, too */
2266 Log("cannot determine which is correct; salvage of volume %u aborted\n", isp->volumeId);
2274 for (i = 0; i < isp->nSpecialInodes; i++) {
2276 ip = &inodes[isp->index + i];
2277 if (ip->u.special.type <= 0 || ip->u.special.type > MAXINODETYPE) {
2279 Log("Rubbish header inode %s of type %d\n",
2280 PrintInode(stmp, ip->inodeNumber),
2281 ip->u.special.type);
2287 Log("Rubbish header inode %s of type %d; deleted\n",
2288 PrintInode(stmp, ip->inodeNumber),
2289 ip->u.special.type);
2290 } else if (!stuff[ip->u.special.type - 1].obsolete) {
2291 if (skip && skip[i]) {
2292 if (orphans == ORPH_REMOVE) {
2293 Log("Removing orphan special inode %s of type %d\n",
2294 PrintInode(stmp, ip->inodeNumber), ip->u.special.type);
2297 Log("Ignoring orphan special inode %s of type %d\n",
2298 PrintInode(stmp, ip->inodeNumber), ip->u.special.type);
2299 /* fall through to the ip->linkCount--; line below */
2302 *(stuff[ip->u.special.type - 1].inode) = ip->inodeNumber;
2303 allinodesobsolete = 0;
2305 if (!check && ip->u.special.type != VI_LINKTABLE)
2306 ip->linkCount--; /* Keep the inode around */
2314 if (allinodesobsolete) {
2321 salvinfo->VGLinkH_cnt++; /* one for every header. */
2323 if (!RW && !check && isp->volSummary) {
2324 ClearROInUseBit(isp->volSummary);
2328 for (i = 0; i < MAXINODETYPE; i++) {
2329 if (stuff[i].inodeType == VI_LINKTABLE) {
2330 /* Gross hack: SalvageHeader does a bcmp on the volume header.
2331 * And we may have recreated the link table earlier, so set the
2332 * RW header as well. The header magic was already checked.
2334 if (VALID_INO(salvinfo->VGLinkH->ih_ino)) {
2335 *stuff[i].inode = salvinfo->VGLinkH->ih_ino;
2339 if (SalvageHeader(salvinfo, &stuff[i], isp, check, deleteMe) == -1 && check)
2343 if (isp->volSummary == NULL) {
2345 char headerName[64];
2346 snprintf(headerName, sizeof headerName, VFORMAT,
2347 afs_printable_uint32_lu(isp->volumeId));
2348 snprintf(path, sizeof path, "%s" OS_DIRSEP "%s",
2349 salvinfo->fileSysPath, headerName);
2351 Log("No header file for volume %u\n", isp->volumeId);
2355 Log("No header file for volume %u; %screating %s\n",
2356 isp->volumeId, (Testing ? "it would have been " : ""),
2358 isp->volSummary = calloc(1, sizeof(struct VolumeSummary));
2360 writefunc = VCreateVolumeDiskHeader;
2363 char headerName[64];
2364 /* hack: these two fields are obsolete... */
2365 isp->volSummary->header.volumeAcl = 0;
2366 isp->volSummary->header.volumeMountTable = 0;
2369 (&isp->volSummary->header, &tempHeader,
2370 sizeof(struct VolumeHeader))) {
2371 VolumeExternalName_r(isp->volumeId, headerName, sizeof(headerName));
2372 snprintf(path, sizeof path, "%s" OS_DIRSEP "%s",
2373 salvinfo->fileSysPath, headerName);
2375 Log("Header file %s is damaged or no longer valid%s\n", path,
2376 (check ? "" : "; repairing"));
2380 writefunc = VWriteVolumeDiskHeader;
2384 memcpy(&isp->volSummary->header, &tempHeader,
2385 sizeof(struct VolumeHeader));
2388 Log("It would have written a new header file for volume %u\n",
2392 VolumeHeaderToDisk(&diskHeader, &tempHeader);
2393 code = (*writefunc)(&diskHeader, salvinfo->fileSysPartition);
2395 Log("Error %ld writing volume header file for volume %lu\n",
2396 afs_printable_int32_ld(code),
2397 afs_printable_uint32_lu(diskHeader.id));
2402 IH_INIT(isp->volSummary->volumeInfoHandle, salvinfo->fileSysDevice, isp->RWvolumeId,
2403 isp->volSummary->header.volumeInfo);
2408 SalvageHeader(struct SalvInfo *salvinfo, struct afs_inode_info *sp,
2409 struct InodeSummary *isp, int check, int *deleteMe)
2412 VolumeDiskData volumeInfo;
2413 struct versionStamp fileHeader;
2422 #ifndef AFS_NAMEI_ENV
2423 if (sp->inodeType == VI_LINKTABLE)
2424 return 0; /* header magic was already checked */
2426 if (*(sp->inode) == 0) {
2428 Log("Missing inode in volume header (%s)\n", sp->description);
2432 Log("Missing inode in volume header (%s); %s\n", sp->description,
2433 (Testing ? "it would have recreated it" : "recreating"));
2436 IH_CREATE(NULL, salvinfo->fileSysDevice, salvinfo->fileSysPath, 0, isp->volumeId,
2437 INODESPECIAL, sp->inodeType, isp->RWvolumeId);
2438 if (!VALID_INO(*(sp->inode)))
2440 ("Unable to allocate inode (%s) for volume header (error = %d)\n",
2441 sp->description, errno);
2446 IH_INIT(specH, salvinfo->fileSysDevice, isp->RWvolumeId, *(sp->inode));
2447 fdP = IH_OPEN(specH);
2448 if (OKToZap && (fdP == NULL) && BadError(errno)) {
2449 /* bail out early and destroy the volume */
2451 Log("Still can't open volume header inode (%s), destroying volume\n", sp->description);
2458 Abort("Unable to open inode (%s) of volume header (error = %d)\n",
2459 sp->description, errno);
2462 && (FDH_PREAD(fdP, (char *)&header, sp->size, 0) != sp->size
2463 || header.fileHeader.magic != sp->stamp.magic)) {
2465 Log("Part of the header (%s) is corrupted\n", sp->description);
2466 FDH_REALLYCLOSE(fdP);
2470 Log("Part of the header (%s) is corrupted; recreating\n",
2473 /* header can be garbage; make sure we don't read garbage data from
2475 memset(&header, 0, sizeof(header));
2477 if (sp->inodeType == VI_VOLINFO
2478 && header.volumeInfo.destroyMe == DESTROY_ME) {
2481 FDH_REALLYCLOSE(fdP);
2485 if (recreate && !Testing) {
2488 ("Internal error: recreating volume header (%s) in check mode\n",
2490 nBytes = FDH_TRUNC(fdP, 0);
2492 Abort("Unable to truncate volume header file (%s) (error = %d)\n",
2493 sp->description, errno);
2495 /* The following code should be moved into vutil.c */
2496 if (sp->inodeType == VI_VOLINFO) {
2498 memset(&header.volumeInfo, 0, sizeof(header.volumeInfo));
2499 header.volumeInfo.stamp = sp->stamp;
2500 header.volumeInfo.id = isp->volumeId;
2501 header.volumeInfo.parentId = isp->RWvolumeId;
2502 sprintf(header.volumeInfo.name, "bogus.%u", isp->volumeId);
2503 Log("Warning: the name of volume %u is now \"bogus.%u\"\n",
2504 isp->volumeId, isp->volumeId);
2505 header.volumeInfo.inService = 0;
2506 header.volumeInfo.blessed = 0;
2507 /* The + 1000 is a hack in case there are any files out in venus caches */
2508 header.volumeInfo.uniquifier = (isp->maxUniquifier + 1) + 1000;
2509 header.volumeInfo.type = (isp->volumeId == isp->RWvolumeId ? readwriteVolume : readonlyVolume); /* XXXX */
2510 header.volumeInfo.needsCallback = 0;
2511 gettimeofday(&tp, NULL);
2512 header.volumeInfo.creationDate = tp.tv_sec;
2514 FDH_PWRITE(fdP, (char *)&header.volumeInfo,
2515 sizeof(header.volumeInfo), 0);
2516 if (nBytes != sizeof(header.volumeInfo)) {
2519 ("Unable to write volume header file (%s) (errno = %d)\n",
2520 sp->description, errno);
2521 Abort("Unable to write entire volume header file (%s)\n",
2525 nBytes = FDH_PWRITE(fdP, (char *)&sp->stamp, sizeof(sp->stamp), 0);
2526 if (nBytes != sizeof(sp->stamp)) {
2529 ("Unable to write version stamp in volume header file (%s) (errno = %d)\n",
2530 sp->description, errno);
2532 ("Unable to write entire version stamp in volume header file (%s)\n",
2537 FDH_REALLYCLOSE(fdP);
2539 if (sp->inodeType == VI_VOLINFO) {
2540 salvinfo->VolInfo = header.volumeInfo;
2544 if (salvinfo->VolInfo.updateDate) {
2545 strcpy(update, TimeStamp(salvinfo->VolInfo.updateDate, 0));
2547 Log("%s (%u) %supdated %s\n", salvinfo->VolInfo.name,
2548 salvinfo->VolInfo.id,
2549 (Testing ? "it would have been " : ""), update);
2551 strcpy(update, TimeStamp(salvinfo->VolInfo.creationDate, 0));
2553 Log("%s (%u) not updated (created %s)\n",
2554 salvinfo->VolInfo.name, salvinfo->VolInfo.id, update);
2564 SalvageVnodes(struct SalvInfo *salvinfo,
2565 struct InodeSummary *rwIsp,
2566 struct InodeSummary *thisIsp,
2567 struct ViceInodeInfo *inodes, int check)
2569 int ilarge, ismall, ioffset, RW, nInodes;
2570 ioffset = rwIsp->index + rwIsp->nSpecialInodes; /* first inode */
2573 RW = (rwIsp == thisIsp);
2574 nInodes = (rwIsp->nInodes - rwIsp->nSpecialInodes);
2576 SalvageIndex(salvinfo, thisIsp->volSummary->header.smallVnodeIndex, vSmall, RW,
2577 &inodes[ioffset], nInodes, thisIsp->volSummary, check);
2578 if (check && ismall == -1)
2581 SalvageIndex(salvinfo, thisIsp->volSummary->header.largeVnodeIndex, vLarge, RW,
2582 &inodes[ioffset], nInodes, thisIsp->volSummary, check);
2583 return (ilarge == 0 && ismall == 0 ? 0 : -1);
2587 SalvageIndex(struct SalvInfo *salvinfo, Inode ino, VnodeClass class, int RW,
2588 struct ViceInodeInfo *ip, int nInodes,
2589 struct VolumeSummary *volSummary, int check)
2591 char buf[SIZEOF_LARGEDISKVNODE];
2592 struct VnodeDiskObject *vnode = (struct VnodeDiskObject *)buf;
2594 StreamHandle_t *file;
2595 struct VnodeClassInfo *vcp;
2597 afs_sfsize_t nVnodes;
2598 afs_fsize_t vnodeLength;
2600 afs_ino_str_t stmp1, stmp2;
2604 IH_INIT(handle, salvinfo->fileSysDevice, volSummary->header.parent, ino);
2605 fdP = IH_OPEN(handle);
2606 osi_Assert(fdP != NULL);
2607 file = FDH_FDOPEN(fdP, "r+");
2608 osi_Assert(file != NULL);
2609 vcp = &VnodeClassInfo[class];
2610 size = OS_SIZE(fdP->fd_fd);
2611 osi_Assert(size != -1);
2612 nVnodes = (size / vcp->diskSize) - 1;
2614 osi_Assert((nVnodes + 1) * vcp->diskSize == size);
2615 osi_Assert(STREAM_ASEEK(file, vcp->diskSize) == 0);
2619 for (vnodeIndex = 0;
2620 nVnodes && STREAM_READ(vnode, vcp->diskSize, 1, file) == 1;
2621 nVnodes--, vnodeIndex++) {
2622 if (vnode->type != vNull) {
2623 int vnodeChanged = 0;
2624 int vnodeNumber = bitNumberToVnodeNumber(vnodeIndex, class);
2625 if (VNDISK_GET_INO(vnode) == 0) {
2627 /* Log("### DEBUG ### Deleted Vnode with 0 inode (vnode %d)\n", vnodeNumber); */
2628 memset(vnode, 0, vcp->diskSize);
2632 if (vcp->magic != vnode->vnodeMagic) {
2633 /* bad magic #, probably partially created vnode */
2635 Log("Partially allocated vnode %d: bad magic (is %lx should be %lx)\n",
2636 vnodeNumber, afs_printable_uint32_lu(vnode->vnodeMagic),
2637 afs_printable_uint32_lu(vcp->magic));
2638 memset(vnode, 0, vcp->diskSize);
2642 Log("Partially allocated vnode %d deleted.\n",
2644 memset(vnode, 0, vcp->diskSize);
2648 /* ****** Should do a bit more salvage here: e.g. make sure
2649 * vnode type matches what it should be given the index */
2650 while (nInodes && ip->u.vnode.vnodeNumber < vnodeNumber) {
2651 /* if (vnodeIdToClass(ip->u.vnode.vnodeNumber) == class && RW) {
2652 * Log("Inode %d: says it belongs to non-existing vnode %d\n",
2653 * ip->inodeNumber, ip->u.vnode.vnodeNumber);
2660 while (nInodes && ip->u.vnode.vnodeNumber == vnodeNumber) {
2661 /* The following doesn't work, because the version number
2662 * is not maintained correctly by the file server */
2663 /*if (vnode->uniquifier == ip->u.vnode.vnodeUniquifier &&
2664 * vnode->dataVersion == ip->u.vnode.inodeDataVersion)
2666 if (VNDISK_GET_INO(vnode) == ip->inodeNumber)
2672 /* For RW volume, look for vnode with matching inode number;
2673 * if no such match, take the first determined by our sort
2675 struct ViceInodeInfo *lip = ip;
2676 int lnInodes = nInodes;
2678 && lip->u.vnode.vnodeNumber == vnodeNumber) {
2679 if (VNDISK_GET_INO(vnode) == lip->inodeNumber) {
2688 if (nInodes && ip->u.vnode.vnodeNumber == vnodeNumber) {
2689 /* "Matching" inode */
2693 vu = vnode->uniquifier;
2694 iu = ip->u.vnode.vnodeUniquifier;
2695 vd = vnode->dataVersion;
2696 id = ip->u.vnode.inodeDataVersion;
2698 * Because of the possibility of the uniquifier overflows (> 4M)
2699 * we compare them modulo the low 22-bits; we shouldn't worry
2700 * about mismatching since they shouldn't to many old
2701 * uniquifiers of the same vnode...
2703 if (IUnique(vu) != IUnique(iu)) {
2705 Log("Vnode %u: vnode.unique, %u, does not match inode unique, %u; fixed, but status will be wrong\n", vnodeNumber, IUnique(vu), IUnique(iu));
2708 vnode->uniquifier = iu;
2709 #ifdef AFS_3DISPARES
2710 vnode->dataVersion = (id >= vd ?
2713 1887437 ? vd : id) :
2716 1887437 ? id : vd));
2718 #if defined(AFS_SGI_EXMAG)
2719 vnode->dataVersion = (id >= vd ?
2722 15099494 ? vd : id) :
2725 15099494 ? id : vd));
2727 vnode->dataVersion = (id > vd ? id : vd);
2728 #endif /* AFS_SGI_EXMAG */
2729 #endif /* AFS_3DISPARES */
2732 /* don't bother checking for vd > id any more, since
2733 * partial file transfers always result in this state,
2734 * and you can't do much else anyway (you've already
2735 * found the best data you can) */
2736 #ifdef AFS_3DISPARES
2737 if (!vnodeIsDirectory(vnodeNumber)
2738 && ((vd < id && (id - vd) < 1887437)
2739 || ((vd > id && (vd - id) > 1887437)))) {
2741 #if defined(AFS_SGI_EXMAG)
2742 if (!vnodeIsDirectory(vnodeNumber)
2743 && ((vd < id && (id - vd) < 15099494)
2744 || ((vd > id && (vd - id) > 15099494)))) {
2746 if (!vnodeIsDirectory(vnodeNumber) && vd < id) {
2747 #endif /* AFS_SGI_EXMAG */
2750 Log("Vnode %d: version < inode version; fixed (old status)\n", vnodeNumber);
2751 vnode->dataVersion = id;
2756 if (ip->inodeNumber != VNDISK_GET_INO(vnode)) {
2759 Log("Vnode %d: inode number incorrect (is %s should be %s). FileSize=%llu\n", vnodeNumber, PrintInode(stmp1, VNDISK_GET_INO(vnode)), PrintInode(stmp2, ip->inodeNumber), (afs_uintmax_t) ip->byteCount);
2761 VNDISK_SET_INO(vnode, ip->inodeNumber);
2766 Log("Vnode %d: inode number incorrect; changed from %s to %s. FileSize=%llu\n", vnodeNumber, PrintInode(stmp1, VNDISK_GET_INO(vnode)), PrintInode(stmp2, ip->inodeNumber), (afs_uintmax_t) ip->byteCount);
2768 VNDISK_SET_INO(vnode, ip->inodeNumber);
2771 VNDISK_GET_LEN(vnodeLength, vnode);
2772 if (ip->byteCount != vnodeLength) {
2775 Log("Vnode %d: length incorrect; (is %llu should be %llu)\n", vnodeNumber, (afs_uintmax_t) vnodeLength, (afs_uintmax_t) ip->byteCount);
2780 Log("Vnode %d: length incorrect; changed from %llu to %llu\n", vnodeNumber, (afs_uintmax_t) vnodeLength, (afs_uintmax_t) ip->byteCount);
2781 VNDISK_SET_LEN(vnode, ip->byteCount);
2785 ip->linkCount--; /* Keep the inode around */
2788 } else { /* no matching inode */
2790 if (VNDISK_GET_INO(vnode) != 0
2791 || vnode->type == vDirectory) {
2792 /* No matching inode--get rid of the vnode */
2794 if (VNDISK_GET_INO(vnode)) {
2796 Log("Vnode %d (unique %u): corresponding inode %s is missing\n", vnodeNumber, vnode->uniquifier, PrintInode(stmp, VNDISK_GET_INO(vnode)));
2800 Log("Vnode %d (unique %u): bad directory vnode (no inode number listed)\n", vnodeNumber, vnode->uniquifier);
2805 if (VNDISK_GET_INO(vnode)) {
2807 time_t serverModifyTime = vnode->serverModifyTime;
2808 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));
2812 time_t serverModifyTime = vnode->serverModifyTime;
2813 Log("Vnode %d (unique %u): bad directory vnode (no inode number listed); vnode deleted, vnode mod time=%s", vnodeNumber, vnode->uniquifier, ctime(&serverModifyTime));
2816 memset(vnode, 0, vcp->diskSize);
2819 /* Should not reach here becuase we checked for
2820 * (inodeNumber == 0) above. And where we zero the vnode,
2821 * we also goto vnodeDone.
2825 while (nInodes && ip->u.vnode.vnodeNumber == vnodeNumber) {
2829 } /* VNDISK_GET_INO(vnode) != 0 */
2831 osi_Assert(!(vnodeChanged && check));
2832 if (vnodeChanged && !Testing) {
2833 osi_Assert(IH_IWRITE
2834 (handle, vnodeIndexOffset(vcp, vnodeNumber),
2835 (char *)vnode, vcp->diskSize)
2837 salvinfo->VolumeChanged = 1; /* For break call back */
2848 struct VnodeEssence *
2849 CheckVnodeNumber(struct SalvInfo *salvinfo, VnodeId vnodeNumber)
2852 struct VnodeInfo *vip;
2855 class = vnodeIdToClass(vnodeNumber);
2856 vip = &salvinfo->vnodeInfo[class];
2857 offset = vnodeIdToBitNumber(vnodeNumber);
2858 return (offset >= vip->nVnodes ? NULL : &vip->vnodes[offset]);
2862 CopyOnWrite(struct SalvInfo *salvinfo, struct DirSummary *dir)
2864 /* Copy the directory unconditionally if we are going to change it:
2865 * not just if was cloned.
2867 struct VnodeDiskObject vnode;
2868 struct VnodeClassInfo *vcp = &VnodeClassInfo[vLarge];
2869 Inode oldinode, newinode;
2872 if (dir->copied || Testing)
2874 DFlush(); /* Well justified paranoia... */
2877 IH_IREAD(salvinfo->vnodeInfo[vLarge].handle,
2878 vnodeIndexOffset(vcp, dir->vnodeNumber), (char *)&vnode,
2880 osi_Assert(code == sizeof(vnode));
2881 oldinode = VNDISK_GET_INO(&vnode);
2882 /* Increment the version number by a whole lot to avoid problems with
2883 * clients that were promised new version numbers--but the file server
2884 * crashed before the versions were written to disk.
2887 IH_CREATE(dir->ds_linkH, salvinfo->fileSysDevice, salvinfo->fileSysPath, 0, dir->rwVid,
2888 dir->vnodeNumber, vnode.uniquifier, vnode.dataVersion +=
2890 osi_Assert(VALID_INO(newinode));
2891 osi_Assert(CopyInode(salvinfo->fileSysDevice, oldinode, newinode, dir->rwVid) == 0);
2893 VNDISK_SET_INO(&vnode, newinode);
2895 IH_IWRITE(salvinfo->vnodeInfo[vLarge].handle,
2896 vnodeIndexOffset(vcp, dir->vnodeNumber), (char *)&vnode,
2898 osi_Assert(code == sizeof(vnode));
2900 SetSalvageDirHandle(&dir->dirHandle, dir->dirHandle.dirh_handle->ih_vid,
2901 salvinfo->fileSysDevice, newinode,
2902 &salvinfo->VolumeChanged);
2903 /* Don't delete the original inode right away, because the directory is
2904 * still being scanned.
2910 * This function should either successfully create a new dir, or give up
2911 * and leave things the way they were. In particular, if it fails to write
2912 * the new dir properly, it should return w/o changing the reference to the
2916 CopyAndSalvage(struct SalvInfo *salvinfo, struct DirSummary *dir)
2918 struct VnodeDiskObject vnode;
2919 struct VnodeClassInfo *vcp = &VnodeClassInfo[vLarge];
2920 Inode oldinode, newinode;
2925 afs_int32 parentUnique = 1;
2926 struct VnodeEssence *vnodeEssence;
2931 Log("Salvaging directory %u...\n", dir->vnodeNumber);
2933 IH_IREAD(salvinfo->vnodeInfo[vLarge].handle,
2934 vnodeIndexOffset(vcp, dir->vnodeNumber), (char *)&vnode,
2936 osi_Assert(lcode == sizeof(vnode));
2937 oldinode = VNDISK_GET_INO(&vnode);
2938 /* Increment the version number by a whole lot to avoid problems with
2939 * clients that were promised new version numbers--but the file server
2940 * crashed before the versions were written to disk.
2943 IH_CREATE(dir->ds_linkH, salvinfo->fileSysDevice, salvinfo->fileSysPath, 0, dir->rwVid,
2944 dir->vnodeNumber, vnode.uniquifier, vnode.dataVersion +=
2946 osi_Assert(VALID_INO(newinode));
2947 SetSalvageDirHandle(&newdir, dir->rwVid, salvinfo->fileSysDevice, newinode,
2948 &salvinfo->VolumeChanged);
2950 /* Assign . and .. vnode numbers from dir and vnode.parent.
2951 * The uniquifier for . is in the vnode.
2952 * The uniquifier for .. might be set to a bogus value of 1 and
2953 * the salvager will later clean it up.
2955 if (vnode.parent && (vnodeEssence = CheckVnodeNumber(salvinfo, vnode.parent))) {
2956 parentUnique = (vnodeEssence->unique ? vnodeEssence->unique : 1);
2959 DirSalvage(&dir->dirHandle, &newdir, dir->vnodeNumber,
2961 (vnode.parent ? vnode.parent : dir->vnodeNumber),
2966 /* didn't really build the new directory properly, let's just give up. */
2967 code = IH_DEC(dir->ds_linkH, newinode, dir->rwVid);
2968 Log("Directory salvage returned code %d, continuing.\n", code);
2970 Log("also failed to decrement link count on new inode");
2974 Log("Checking the results of the directory salvage...\n");
2975 if (!DirOK(&newdir)) {
2976 Log("Directory salvage failed!!!; restoring old version of the directory.\n");
2977 code = IH_DEC(dir->ds_linkH, newinode, dir->rwVid);
2978 osi_Assert(code == 0);
2982 VNDISK_SET_INO(&vnode, newinode);
2983 length = afs_dir_Length(&newdir);
2984 VNDISK_SET_LEN(&vnode, length);
2986 IH_IWRITE(salvinfo->vnodeInfo[vLarge].handle,
2987 vnodeIndexOffset(vcp, dir->vnodeNumber), (char *)&vnode,
2989 osi_Assert(lcode == sizeof(vnode));
2992 nt_sync(salvinfo->fileSysDevice);
2994 sync(); /* this is slow, but hopefully rarely called. We don't have
2995 * an open FD on the file itself to fsync.
2999 salvinfo->vnodeInfo[vLarge].handle->ih_synced = 1;
3001 /* make sure old directory file is really closed */
3002 fdP = IH_OPEN(dir->dirHandle.dirh_handle);
3003 FDH_REALLYCLOSE(fdP);
3005 code = IH_DEC(dir->ds_linkH, oldinode, dir->rwVid);
3006 osi_Assert(code == 0);
3007 dir->dirHandle = newdir;
3011 * arguments for JudgeEntry.
3013 struct judgeEntry_params {
3014 struct DirSummary *dir; /**< directory we're examining entries in */
3015 struct SalvInfo *salvinfo; /**< SalvInfo for the current salvage job */
3019 JudgeEntry(void *arock, char *name, afs_int32 vnodeNumber,
3022 struct judgeEntry_params *params = arock;
3023 struct DirSummary *dir = params->dir;
3024 struct SalvInfo *salvinfo = params->salvinfo;
3025 struct VnodeEssence *vnodeEssence;
3026 afs_int32 dirOrphaned, todelete;
3028 dirOrphaned = IsVnodeOrphaned(salvinfo, dir->vnodeNumber);
3030 vnodeEssence = CheckVnodeNumber(salvinfo, vnodeNumber);
3031 if (vnodeEssence == NULL) {
3033 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);
3036 CopyOnWrite(salvinfo, dir);
3037 osi_Assert(afs_dir_Delete(&dir->dirHandle, name) == 0);
3042 #ifndef AFS_NAMEI_ENV
3043 /* On AIX machines, don't allow entries to point to inode 0. That is a special
3044 * mount inode for the partition. If this inode were deleted, it would crash
3047 if (vnodeEssence->InodeNumber == 0) {
3048 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"));
3050 CopyOnWrite(salvinfo, dir);
3051 osi_Assert(Delete(&dir->dirHandle, name) == 0);
3058 if (!(vnodeNumber & 1) && !Showmode
3059 && !(vnodeEssence->count || vnodeEssence->unique
3060 || vnodeEssence->modeBits)) {
3061 Log("dir vnode %u: invalid entry: %s" OS_DIRSEP "%s (vnode %u, unique %u)%s\n",
3062 dir->vnodeNumber, (dir->name ? dir->name : "??"), name,
3063 vnodeNumber, unique,
3064 ((!unique) ? (Testing ? "-- would have deleted" : " -- deleted") :
3068 CopyOnWrite(salvinfo, dir);
3069 osi_Assert(afs_dir_Delete(&dir->dirHandle, name) == 0);
3075 /* Check if the Uniquifiers match. If not, change the directory entry
3076 * so its unique matches the vnode unique. Delete if the unique is zero
3077 * or if the directory is orphaned.
3079 if (!vnodeEssence->unique || (vnodeEssence->unique) != unique) {
3080 if (!vnodeEssence->unique
3081 && ((strcmp(name, "..") == 0) || (strcmp(name, ".") == 0))) {
3082 /* This is an orphaned directory. Don't delete the . or ..
3083 * entry. Otherwise, it will get created in the next
3084 * salvage and deleted again here. So Just skip it.
3089 todelete = ((!vnodeEssence->unique || dirOrphaned) ? 1 : 0);
3092 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")));
3096 fid.Vnode = vnodeNumber;
3097 fid.Unique = vnodeEssence->unique;
3098 CopyOnWrite(salvinfo, dir);
3099 osi_Assert(afs_dir_Delete(&dir->dirHandle, name) == 0);
3101 osi_Assert(afs_dir_Create(&dir->dirHandle, name, &fid) == 0);
3104 return 0; /* no need to continue */
3107 if (strcmp(name, ".") == 0) {
3108 if (dir->vnodeNumber != vnodeNumber || (dir->unique != unique)) {
3111 Log("directory vnode %u.%u: bad '.' entry (was %u.%u); fixed\n", dir->vnodeNumber, dir->unique, vnodeNumber, unique);
3113 CopyOnWrite(salvinfo, dir);
3114 osi_Assert(afs_dir_Delete(&dir->dirHandle, ".") == 0);
3115 fid.Vnode = dir->vnodeNumber;
3116 fid.Unique = dir->unique;
3117 osi_Assert(afs_dir_Create(&dir->dirHandle, ".", &fid) == 0);
3120 vnodeNumber = fid.Vnode; /* Get the new Essence */
3121 unique = fid.Unique;
3122 vnodeEssence = CheckVnodeNumber(salvinfo, vnodeNumber);
3125 } else if (strcmp(name, "..") == 0) {
3128 struct VnodeEssence *dotdot;
3129 pa.Vnode = dir->parent;
3130 dotdot = CheckVnodeNumber(salvinfo, pa.Vnode);
3131 osi_Assert(dotdot != NULL); /* XXX Should not be assert */
3132 pa.Unique = dotdot->unique;
3134 pa.Vnode = dir->vnodeNumber;
3135 pa.Unique = dir->unique;
3137 if ((pa.Vnode != vnodeNumber) || (pa.Unique != unique)) {
3139 Log("directory vnode %u.%u: bad '..' entry (was %u.%u); fixed\n", dir->vnodeNumber, dir->unique, vnodeNumber, unique);
3141 CopyOnWrite(salvinfo, dir);
3142 osi_Assert(afs_dir_Delete(&dir->dirHandle, "..") == 0);
3143 osi_Assert(afs_dir_Create(&dir->dirHandle, "..", &pa) == 0);
3146 vnodeNumber = pa.Vnode; /* Get the new Essence */
3148 vnodeEssence = CheckVnodeNumber(salvinfo, vnodeNumber);
3150 dir->haveDotDot = 1;
3151 } else if (strncmp(name, ".__afs", 6) == 0) {
3153 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);
3156 CopyOnWrite(salvinfo, dir);
3157 osi_Assert(afs_dir_Delete(&dir->dirHandle, name) == 0);
3159 vnodeEssence->claimed = 0; /* Not claimed: Orphaned */
3160 vnodeEssence->todelete = 1; /* Will later delete vnode and decr inode */
3163 if (ShowSuid && (vnodeEssence->modeBits & 06000))
3164 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);
3165 if (/* ShowMounts && */ (vnodeEssence->type == vSymlink)
3166 && !(vnodeEssence->modeBits & 0111)) {
3167 afs_sfsize_t nBytes;
3173 IH_INIT(ihP, salvinfo->fileSysDevice, dir->dirHandle.dirh_handle->ih_vid,
3174 vnodeEssence->InodeNumber);
3177 Log("ERROR %s could not open mount point vnode %u\n", dir->vname, vnodeNumber);
3181 size = FDH_SIZE(fdP);
3183 Log("ERROR %s mount point has invalid size %d, vnode %u\n", dir->vname, (int)size, vnodeNumber);
3184 FDH_REALLYCLOSE(fdP);
3191 nBytes = FDH_PREAD(fdP, buf, size, 0);
3192 if (nBytes == size) {
3194 if ( (*buf != '#' && *buf != '%') || buf[strlen(buf)-1] != '.' ) {
3195 Log("Volume %u (%s) mount point %s" OS_DIRSEP "%s to '%s' invalid, %s to symbolic link\n",
3196 dir->dirHandle.dirh_handle->ih_vid, dir->vname, dir->name ? dir->name : "??", name, buf,
3197 Testing ? "would convert" : "converted");
3198 vnodeEssence->modeBits |= 0111;
3199 vnodeEssence->changed = 1;
3200 } else if (ShowMounts) Log("In volume %u (%s) found mountpoint %s" OS_DIRSEP "%s to '%s'\n",
3201 dir->dirHandle.dirh_handle->ih_vid, dir->vname,
3202 dir->name ? dir->name : "??", name, buf);
3204 Log("Volume %s cound not read mount point vnode %u size %d code %d\n",
3205 dir->vname, vnodeNumber, (int)size, (int)nBytes);
3207 FDH_REALLYCLOSE(fdP);
3210 if (ShowRootFiles && vnodeEssence->owner == 0 && vnodeNumber != 1)
3211 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);
3212 if (vnodeIdToClass(vnodeNumber) == vLarge
3213 && vnodeEssence->name == NULL) {
3215 if ((n = (char *)malloc(strlen(name) + 1)))
3217 vnodeEssence->name = n;
3220 /* The directory entry points to the vnode. Check to see if the
3221 * vnode points back to the directory. If not, then let the
3222 * directory claim it (else it might end up orphaned). Vnodes
3223 * already claimed by another directory are deleted from this
3224 * directory: hardlinks to the same vnode are not allowed
3225 * from different directories.
3227 if (vnodeEssence->parent != dir->vnodeNumber) {
3228 if (!vnodeEssence->claimed && !dirOrphaned && vnodeNumber != 1) {
3229 /* Vnode does not point back to this directory.
3230 * Orphaned dirs cannot claim a file (it may belong to
3231 * another non-orphaned dir).
3234 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);
3236 vnodeEssence->parent = dir->vnodeNumber;
3237 vnodeEssence->changed = 1;
3239 /* Vnode was claimed by another directory */
3242 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 " : ""));
3243 } else if (vnodeNumber == 1) {
3244 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 " : ""));
3246 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 " : ""));
3250 CopyOnWrite(salvinfo, dir);
3251 osi_Assert(afs_dir_Delete(&dir->dirHandle, name) == 0);
3256 /* This directory claims the vnode */
3257 vnodeEssence->claimed = 1;
3259 vnodeEssence->count--;
3264 DistilVnodeEssence(struct SalvInfo *salvinfo, VolumeId rwVId,
3265 VnodeClass class, Inode ino, Unique * maxu)
3267 struct VnodeInfo *vip = &salvinfo->vnodeInfo[class];
3268 struct VnodeClassInfo *vcp = &VnodeClassInfo[class];
3269 char buf[SIZEOF_LARGEDISKVNODE];
3270 struct VnodeDiskObject *vnode = (struct VnodeDiskObject *)buf;
3272 StreamHandle_t *file;
3277 IH_INIT(vip->handle, salvinfo->fileSysDevice, rwVId, ino);
3278 fdP = IH_OPEN(vip->handle);
3279 osi_Assert(fdP != NULL);
3280 file = FDH_FDOPEN(fdP, "r+");
3281 osi_Assert(file != NULL);
3282 size = OS_SIZE(fdP->fd_fd);
3283 osi_Assert(size != -1);
3284 vip->nVnodes = (size / vcp->diskSize) - 1;
3285 if (vip->nVnodes > 0) {
3286 osi_Assert((vip->nVnodes + 1) * vcp->diskSize == size);
3287 osi_Assert(STREAM_ASEEK(file, vcp->diskSize) == 0);
3288 osi_Assert((vip->vnodes = (struct VnodeEssence *)
3289 calloc(vip->nVnodes, sizeof(struct VnodeEssence))) != NULL);
3290 if (class == vLarge) {
3291 osi_Assert((vip->inodes = (Inode *)
3292 calloc(vip->nVnodes, sizeof(Inode))) != NULL);
3301 vip->volumeBlockCount = vip->nAllocatedVnodes = 0;
3302 for (vnodeIndex = 0, nVnodes = vip->nVnodes;
3303 nVnodes && STREAM_READ(vnode, vcp->diskSize, 1, file) == 1;
3304 nVnodes--, vnodeIndex++) {
3305 if (vnode->type != vNull) {
3306 struct VnodeEssence *vep = &vip->vnodes[vnodeIndex];
3307 afs_fsize_t vnodeLength;
3308 vip->nAllocatedVnodes++;
3309 vep->count = vnode->linkCount;
3310 VNDISK_GET_LEN(vnodeLength, vnode);
3311 vep->blockCount = nBlocks(vnodeLength);
3312 vip->volumeBlockCount += vep->blockCount;
3313 vep->parent = vnode->parent;
3314 vep->unique = vnode->uniquifier;
3315 if (*maxu < vnode->uniquifier)
3316 *maxu = vnode->uniquifier;
3317 vep->modeBits = vnode->modeBits;
3318 vep->InodeNumber = VNDISK_GET_INO(vnode);
3319 vep->type = vnode->type;
3320 vep->author = vnode->author;
3321 vep->owner = vnode->owner;
3322 vep->group = vnode->group;
3323 if (vnode->type == vDirectory) {
3324 if (class != vLarge) {
3325 VnodeId vnodeNumber = bitNumberToVnodeNumber(vnodeIndex, class);
3326 vip->nAllocatedVnodes--;
3327 memset(vnode, 0, sizeof(vnode));
3328 IH_IWRITE(salvinfo->vnodeInfo[vSmall].handle,
3329 vnodeIndexOffset(vcp, vnodeNumber),
3330 (char *)&vnode, sizeof(vnode));
3331 salvinfo->VolumeChanged = 1;
3333 vip->inodes[vnodeIndex] = VNDISK_GET_INO(vnode);
3342 GetDirName(struct SalvInfo *salvinfo, VnodeId vnode, struct VnodeEssence *vp,
3345 struct VnodeEssence *parentvp;
3351 if (vp->parent && vp->name && (parentvp = CheckVnodeNumber(salvinfo, vp->parent))
3352 && GetDirName(salvinfo, vp->parent, parentvp, path)) {
3353 strcat(path, OS_DIRSEP);
3354 strcat(path, vp->name);
3360 /* To determine if a vnode is orhpaned or not, the vnode and all its parent
3361 * vnodes must be "claimed". The vep->claimed flag is set in JudgeEntry().
3364 IsVnodeOrphaned(struct SalvInfo *salvinfo, VnodeId vnode)
3366 struct VnodeEssence *vep;
3369 return (1); /* Vnode zero does not exist */
3371 return (0); /* The root dir vnode is always claimed */
3372 vep = CheckVnodeNumber(salvinfo, vnode); /* Get the vnode essence */
3373 if (!vep || !vep->claimed)
3374 return (1); /* Vnode is not claimed - it is orphaned */
3376 return (IsVnodeOrphaned(salvinfo, vep->parent));
3380 SalvageDir(struct SalvInfo *salvinfo, char *name, VolumeId rwVid,
3381 struct VnodeInfo *dirVnodeInfo, IHandle_t * alinkH, int i,
3382 struct DirSummary *rootdir, int *rootdirfound)
3384 static struct DirSummary dir;
3385 static struct DirHandle dirHandle;
3386 struct VnodeEssence *parent;
3387 static char path[MAXPATHLEN];
3390 if (dirVnodeInfo->vnodes[i].salvaged)
3391 return; /* already salvaged */
3394 dirVnodeInfo->vnodes[i].salvaged = 1;
3396 if (dirVnodeInfo->inodes[i] == 0)
3397 return; /* Not allocated to a directory */
3399 if (bitNumberToVnodeNumber(i, vLarge) == 1) {
3400 if (dirVnodeInfo->vnodes[i].parent) {
3401 Log("Bad parent, vnode 1; %s...\n",
3402 (Testing ? "skipping" : "salvaging"));
3403 dirVnodeInfo->vnodes[i].parent = 0;
3404 dirVnodeInfo->vnodes[i].changed = 1;
3407 parent = CheckVnodeNumber(salvinfo, dirVnodeInfo->vnodes[i].parent);
3408 if (parent && parent->salvaged == 0)
3409 SalvageDir(salvinfo, name, rwVid, dirVnodeInfo, alinkH,
3410 vnodeIdToBitNumber(dirVnodeInfo->vnodes[i].parent),
3411 rootdir, rootdirfound);
3414 dir.vnodeNumber = bitNumberToVnodeNumber(i, vLarge);
3415 dir.unique = dirVnodeInfo->vnodes[i].unique;
3418 dir.parent = dirVnodeInfo->vnodes[i].parent;
3419 dir.haveDot = dir.haveDotDot = 0;
3420 dir.ds_linkH = alinkH;
3421 SetSalvageDirHandle(&dir.dirHandle, dir.rwVid, salvinfo->fileSysDevice,
3422 dirVnodeInfo->inodes[i], &salvinfo->VolumeChanged);
3424 dirok = ((RebuildDirs && !Testing) ? 0 : DirOK(&dir.dirHandle));
3427 Log("Directory bad, vnode %u; %s...\n", dir.vnodeNumber,
3428 (Testing ? "skipping" : "salvaging"));
3431 CopyAndSalvage(salvinfo, &dir);
3433 dirVnodeInfo->inodes[i] = dir.dirHandle.dirh_inode;
3436 dirHandle = dir.dirHandle;
3439 GetDirName(salvinfo, bitNumberToVnodeNumber(i, vLarge),
3440 &dirVnodeInfo->vnodes[i], path);
3443 /* If enumeration failed for random reasons, we will probably delete
3444 * too much stuff, so we guard against this instead.
3446 struct judgeEntry_params judge_params;
3447 judge_params.salvinfo = salvinfo;
3448 judge_params.dir = &dir;
3450 osi_Assert(afs_dir_EnumerateDir(&dirHandle, JudgeEntry,
3451 &judge_params) == 0);
3454 /* Delete the old directory if it was copied in order to salvage.
3455 * CopyOnWrite has written the new inode # to the disk, but we still
3456 * have the old one in our local structure here. Thus, we idec the
3460 if (dir.copied && !Testing) {
3461 code = IH_DEC(dir.ds_linkH, dirHandle.dirh_handle->ih_ino, rwVid);
3462 osi_Assert(code == 0);
3463 dirVnodeInfo->inodes[i] = dir.dirHandle.dirh_inode;
3466 /* Remember rootdir DirSummary _after_ it has been judged */
3467 if (dir.vnodeNumber == 1 && dir.unique == 1) {
3468 memcpy(rootdir, &dir, sizeof(struct DirSummary));
3476 * Get a new FID that can be used to create a new file.
3478 * @param[in] volHeader vol header for the volume
3479 * @param[in] class what type of vnode we'll be creating (vLarge or vSmall)
3480 * @param[out] afid the FID that we can use (only Vnode and Unique are set)
3481 * @param[inout] maxunique max uniquifier for all vnodes in the volume;
3482 * updated to the new max unique if we create a new
3486 GetNewFID(struct SalvInfo *salvinfo, VolumeDiskData *volHeader,
3487 VnodeClass class, AFSFid *afid, Unique *maxunique)
3490 for (i = 0; i < salvinfo->vnodeInfo[class].nVnodes; i++) {
3491 if (salvinfo->vnodeInfo[class].vnodes[i].type == vNull) {
3495 if (i == salvinfo->vnodeInfo[class].nVnodes) {
3496 /* no free vnodes; make a new one */
3497 salvinfo->vnodeInfo[class].nVnodes++;
3498 salvinfo->vnodeInfo[class].vnodes =
3499 realloc(salvinfo->vnodeInfo[class].vnodes,
3500 sizeof(struct VnodeEssence) * (i+1));
3502 salvinfo->vnodeInfo[class].vnodes[i].type = vNull;
3505 afid->Vnode = bitNumberToVnodeNumber(i, class);
3507 if (volHeader->uniquifier < (*maxunique + 1)) {
3508 /* header uniq is bad; it will get bumped by 2000 later */
3509 afid->Unique = *maxunique + 1 + 2000;
3512 /* header uniq seems okay; just use that */
3513 afid->Unique = *maxunique = volHeader->uniquifier++;
3518 * Create a vnode for a README file explaining not to use a recreated-root vol.
3520 * @param[in] volHeader vol header for the volume
3521 * @param[in] alinkH ihandle for i/o for the volume
3522 * @param[in] vid volume id
3523 * @param[inout] maxunique max uniquifier for all vnodes in the volume;
3524 * updated to the new max unique if we create a new
3526 * @param[out] afid FID for the new readme vnode
3527 * @param[out] ainode the inode for the new readme file
3529 * @return operation status
3534 CreateReadme(struct SalvInfo *salvinfo, VolumeDiskData *volHeader,
3535 IHandle_t *alinkH, VolumeId vid, Unique *maxunique, AFSFid *afid,
3539 struct VnodeDiskObject *rvnode = NULL;
3541 IHandle_t *readmeH = NULL;
3542 struct VnodeEssence *vep;
3544 time_t now = time(NULL);
3546 /* Try to make the note brief, but informative. Only administrators should
3547 * be able to read this file at first, so we can hopefully assume they
3548 * know what AFS is, what a volume is, etc. */
3550 "This volume has been salvaged, but has lost its original root directory.\n"
3551 "The root directory that exists now has been recreated from orphan files\n"
3552 "from the rest of the volume. This recreated root directory may interfere\n"
3553 "with old cached data on clients, and there is no way the salvager can\n"
3554 "reasonably prevent that. So, it is recommended that you do not continue to\n"
3555 "use this volume, but only copy the salvaged data to a new volume.\n"
3556 "Continuing to use this volume as it exists now may cause some clients to\n"
3557 "behave oddly when accessing this volume.\n"
3558 "\n\t -- Your friendly neighborhood OpenAFS salvager\n";
3559 /* ^ the person reading this probably just lost some data, so they could
3560 * use some cheering up. */
3562 /* -1 for the trailing NUL */
3563 length = sizeof(readme) - 1;
3565 GetNewFID(salvinfo, volHeader, vSmall, afid, maxunique);
3567 vep = &salvinfo->vnodeInfo[vSmall].vnodes[vnodeIdToBitNumber(afid->Vnode)];
3569 /* create the inode and write the contents */
3570 readmeinode = IH_CREATE(alinkH, salvinfo->fileSysDevice,
3571 salvinfo->fileSysPath, 0, vid,
3572 afid->Vnode, afid->Unique, 1);
3573 if (!VALID_INO(readmeinode)) {
3574 Log("CreateReadme: readme IH_CREATE failed\n");
3578 IH_INIT(readmeH, salvinfo->fileSysDevice, vid, readmeinode);
3579 bytes = IH_IWRITE(readmeH, 0, readme, length);
3580 IH_RELEASE(readmeH);
3582 if (bytes != length) {
3583 Log("CreateReadme: IWRITE failed (%d/%d)\n", (int)bytes,
3584 (int)sizeof(readme));
3588 /* create the vnode and write it out */
3589 rvnode = calloc(1, SIZEOF_SMALLDISKVNODE);
3591 Log("CreateRootDir: error alloc'ing memory\n");
3595 rvnode->type = vFile;
3597 rvnode->modeBits = 0777;
3598 rvnode->linkCount = 1;
3599 VNDISK_SET_LEN(rvnode, length);
3600 rvnode->uniquifier = afid->Unique;
3601 rvnode->dataVersion = 1;
3602 VNDISK_SET_INO(rvnode, readmeinode);
3603 rvnode->unixModifyTime = rvnode->serverModifyTime = now;
3608 rvnode->vnodeMagic = VnodeClassInfo[vSmall].magic;
3610 bytes = IH_IWRITE(salvinfo->vnodeInfo[vSmall].handle,
3611 vnodeIndexOffset(&VnodeClassInfo[vSmall], afid->Vnode),
3612 (char*)rvnode, SIZEOF_SMALLDISKVNODE);
3614 if (bytes != SIZEOF_SMALLDISKVNODE) {
3615 Log("CreateReadme: IH_IWRITE failed (%d/%d)\n", (int)bytes,
3616 (int)SIZEOF_SMALLDISKVNODE);
3620 /* update VnodeEssence for new readme vnode */
3621 salvinfo->vnodeInfo[vSmall].nAllocatedVnodes++;
3623 vep->blockCount = nBlocks(length);
3624 salvinfo->vnodeInfo[vSmall].volumeBlockCount += vep->blockCount;
3625 vep->parent = rvnode->parent;
3626 vep->unique = rvnode->uniquifier;
3627 vep->modeBits = rvnode->modeBits;
3628 vep->InodeNumber = VNDISK_GET_INO(rvnode);
3629 vep->type = rvnode->type;
3630 vep->author = rvnode->author;
3631 vep->owner = rvnode->owner;
3632 vep->group = rvnode->group;
3642 *ainode = readmeinode;
3647 if (IH_DEC(alinkH, readmeinode, vid)) {
3648 Log("CreateReadme (recovery): IH_DEC failed\n");
3660 * create a root dir for a volume that lacks one.
3662 * @param[in] volHeader vol header for the volume
3663 * @param[in] alinkH ihandle for disk access for this volume group
3664 * @param[in] vid volume id we're dealing with
3665 * @param[out] rootdir populated with info about the new root dir
3666 * @param[inout] maxunique max uniquifier for all vnodes in the volume;
3667 * updated to the new max unique if we create a new
3670 * @return operation status
3675 CreateRootDir(struct SalvInfo *salvinfo, VolumeDiskData *volHeader,
3676 IHandle_t *alinkH, VolumeId vid, struct DirSummary *rootdir,
3680 int decroot = 0, decreadme = 0;
3681 AFSFid did, readmeid;
3684 struct VnodeDiskObject *rootvnode = NULL;
3685 struct acl_accessList *ACL;
3688 struct VnodeEssence *vep;
3689 Inode readmeinode = 0;
3690 time_t now = time(NULL);
3692 if (!salvinfo->vnodeInfo[vLarge].vnodes && !salvinfo->vnodeInfo[vSmall].vnodes) {
3693 Log("Not creating new root dir; volume appears to lack any vnodes\n");
3697 if (!salvinfo->vnodeInfo[vLarge].vnodes) {
3698 /* We don't have any large vnodes in the volume; allocate room
3699 * for one so we can recreate the root dir */
3700 salvinfo->vnodeInfo[vLarge].nVnodes = 1;
3701 salvinfo->vnodeInfo[vLarge].vnodes = calloc(1, sizeof(struct VnodeEssence));
3702 salvinfo->vnodeInfo[vLarge].inodes = calloc(1, sizeof(Inode));
3704 osi_Assert(salvinfo->vnodeInfo[vLarge].vnodes);
3705 osi_Assert(salvinfo->vnodeInfo[vLarge].inodes);
3708 vep = &salvinfo->vnodeInfo[vLarge].vnodes[vnodeIdToBitNumber(1)];
3709 ip = &salvinfo->vnodeInfo[vLarge].inodes[vnodeIdToBitNumber(1)];
3710 if (vep->type != vNull) {
3711 Log("Not creating new root dir; existing vnode 1 is non-null\n");
3715 if (CreateReadme(salvinfo, volHeader, alinkH, vid, maxunique, &readmeid,
3716 &readmeinode) != 0) {
3721 /* set the DV to a very high number, so it is unlikely that we collide
3722 * with a cached DV */
3725 rootinode = IH_CREATE(alinkH, salvinfo->fileSysDevice, salvinfo->fileSysPath,
3727 if (!VALID_INO(rootinode)) {
3728 Log("CreateRootDir: IH_CREATE failed\n");
3733 SetSalvageDirHandle(&rootdir->dirHandle, vid, salvinfo->fileSysDevice,
3734 rootinode, &salvinfo->VolumeChanged);
3738 if (afs_dir_MakeDir(&rootdir->dirHandle, (afs_int32*)&did, (afs_int32*)&did)) {
3739 Log("CreateRootDir: MakeDir failed\n");
3742 if (afs_dir_Create(&rootdir->dirHandle, "README.ROOTDIR", &readmeid)) {
3743 Log("CreateRootDir: Create failed\n");
3747 length = afs_dir_Length(&rootdir->dirHandle);
3748 DZap(&rootdir->dirHandle);
3750 /* create the new root dir vnode */
3751 rootvnode = calloc(1, SIZEOF_LARGEDISKVNODE);
3753 Log("CreateRootDir: malloc failed\n");
3757 /* only give 'rl' permissions to 'system:administrators'. We do this to
3758 * try to catch the attention of an administrator, that they should not
3759 * be writing to this directory or continue to use it. */
3760 ACL = VVnodeDiskACL(rootvnode);
3761 ACL->size = sizeof(struct acl_accessList);
3762 ACL->version = ACL_ACLVERSION;
3766 ACL->entries[0].id = -204; /* system:administrators */
3767 ACL->entries[0].rights = PRSFS_READ | PRSFS_LOOKUP;
3769 rootvnode->type = vDirectory;
3770 rootvnode->cloned = 0;
3771 rootvnode->modeBits = 0777;
3772 rootvnode->linkCount = 2;
3773 VNDISK_SET_LEN(rootvnode, length);
3774 rootvnode->uniquifier = 1;
3775 rootvnode->dataVersion = dv;
3776 VNDISK_SET_INO(rootvnode, rootinode);
3777 rootvnode->unixModifyTime = rootvnode->serverModifyTime = now;
3778 rootvnode->author = 0;
3779 rootvnode->owner = 0;
3780 rootvnode->parent = 0;
3781 rootvnode->group = 0;
3782 rootvnode->vnodeMagic = VnodeClassInfo[vLarge].magic;
3784 /* write it out to disk */
3785 bytes = IH_IWRITE(salvinfo->vnodeInfo[vLarge].handle,
3786 vnodeIndexOffset(&VnodeClassInfo[vLarge], 1),
3787 (char*)rootvnode, SIZEOF_LARGEDISKVNODE);
3789 if (bytes != SIZEOF_LARGEDISKVNODE) {
3790 /* just cast to int and don't worry about printing real 64-bit ints;
3791 * a large disk vnode isn't anywhere near the 32-bit limit */
3792 Log("CreateRootDir: IH_IWRITE failed (%d/%d)\n", (int)bytes,
3793 (int)SIZEOF_LARGEDISKVNODE);
3797 /* update VnodeEssence for the new root vnode */
3798 salvinfo->vnodeInfo[vLarge].nAllocatedVnodes++;
3800 vep->blockCount = nBlocks(length);
3801 salvinfo->vnodeInfo[vLarge].volumeBlockCount += vep->blockCount;
3802 vep->parent = rootvnode->parent;
3803 vep->unique = rootvnode->uniquifier;
3804 vep->modeBits = rootvnode->modeBits;
3805 vep->InodeNumber = VNDISK_GET_INO(rootvnode);
3806 vep->type = rootvnode->type;
3807 vep->author = rootvnode->author;
3808 vep->owner = rootvnode->owner;
3809 vep->group = rootvnode->group;
3819 /* update DirSummary for the new root vnode */
3820 rootdir->vnodeNumber = 1;
3821 rootdir->unique = 1;
3822 rootdir->haveDot = 1;
3823 rootdir->haveDotDot = 1;
3824 rootdir->rwVid = vid;
3825 rootdir->copied = 0;
3826 rootdir->parent = 0;
3827 rootdir->name = strdup(".");
3828 rootdir->vname = volHeader->name;
3829 rootdir->ds_linkH = alinkH;
3836 if (decroot && IH_DEC(alinkH, rootinode, vid)) {
3837 Log("CreateRootDir (recovery): IH_DEC (root) failed\n");
3839 if (decreadme && IH_DEC(alinkH, readmeinode, vid)) {
3840 Log("CreateRootDir (recovery): IH_DEC (readme) failed\n");
3850 * salvage a volume group.
3852 * @param[in] salvinfo information for the curent salvage job
3853 * @param[in] rwIsp inode summary for rw volume
3854 * @param[in] alinkH link table inode handle
3856 * @return operation status
3860 SalvageVolume(struct SalvInfo *salvinfo, struct InodeSummary *rwIsp, IHandle_t * alinkH)
3862 /* This routine, for now, will only be called for read-write volumes */
3864 int BlocksInVolume = 0, FilesInVolume = 0;
3866 struct DirSummary rootdir, oldrootdir;
3867 struct VnodeInfo *dirVnodeInfo;
3868 struct VnodeDiskObject vnode;
3869 VolumeDiskData volHeader;
3871 int orphaned, rootdirfound = 0;
3872 Unique maxunique = 0; /* the maxUniquifier from the vnodes */
3873 afs_int32 ofiles = 0, oblocks = 0; /* Number of orphaned files/blocks */
3874 struct VnodeEssence *vep;
3877 afs_sfsize_t nBytes;
3879 VnodeId LFVnode, ThisVnode;
3880 Unique LFUnique, ThisUnique;
3884 vid = rwIsp->volSummary->header.id;
3885 IH_INIT(h, salvinfo->fileSysDevice, vid, rwIsp->volSummary->header.volumeInfo);
3886 nBytes = IH_IREAD(h, 0, (char *)&volHeader, sizeof(volHeader));
3887 osi_Assert(nBytes == sizeof(volHeader));
3888 osi_Assert(volHeader.stamp.magic == VOLUMEINFOMAGIC);
3889 osi_Assert(volHeader.destroyMe != DESTROY_ME);
3890 /* (should not have gotten this far with DESTROY_ME flag still set!) */
3892 DistilVnodeEssence(salvinfo, vid, vLarge,
3893 rwIsp->volSummary->header.largeVnodeIndex, &maxunique);
3894 DistilVnodeEssence(salvinfo, vid, vSmall,
3895 rwIsp->volSummary->header.smallVnodeIndex, &maxunique);
3897 dirVnodeInfo = &salvinfo->vnodeInfo[vLarge];
3898 for (i = 0; i < dirVnodeInfo->nVnodes; i++) {
3899 SalvageDir(salvinfo, volHeader.name, vid, dirVnodeInfo, alinkH, i,
3900 &rootdir, &rootdirfound);
3903 nt_sync(salvinfo->fileSysDevice);
3905 sync(); /* This used to be done lower level, for every dir */
3912 if (!rootdirfound && (orphans == ORPH_ATTACH) && !Testing) {
3914 Log("Cannot find root directory for volume %lu; attempting to create "
3915 "a new one\n", afs_printable_uint32_lu(vid));
3917 code = CreateRootDir(salvinfo, &volHeader, alinkH, vid, &rootdir,
3922 salvinfo->VolumeChanged = 1;
3926 /* Parse each vnode looking for orphaned vnodes and
3927 * connect them to the tree as orphaned (if requested).
3929 oldrootdir = rootdir;
3930 for (class = 0; class < nVNODECLASSES; class++) {
3931 for (v = 0; v < salvinfo->vnodeInfo[class].nVnodes; v++) {
3932 vep = &(salvinfo->vnodeInfo[class].vnodes[v]);
3933 ThisVnode = bitNumberToVnodeNumber(v, class);
3934 ThisUnique = vep->unique;
3936 if ((vep->type == 0) || vep->claimed || ThisVnode == 1)
3937 continue; /* Ignore unused, claimed, and root vnodes */
3939 /* This vnode is orphaned. If it is a directory vnode, then the '..'
3940 * entry in this vnode had incremented the parent link count (In
3941 * JudgeEntry()). We need to go to the parent and decrement that
3942 * link count. But if the parent's unique is zero, then the parent
3943 * link count was not incremented in JudgeEntry().
3945 if (class == vLarge) { /* directory vnode */
3946 pv = vnodeIdToBitNumber(vep->parent);
3947 if (salvinfo->vnodeInfo[vLarge].vnodes[pv].unique != 0) {
3948 if (vep->parent == 1 && newrootdir) {
3949 /* this vnode's parent was the volume root, and
3950 * we just created the volume root. So, the parent
3951 * dir didn't exist during JudgeEntry, so the link
3952 * count was not inc'd there, so don't dec it here.
3958 salvinfo->vnodeInfo[vLarge].vnodes[pv].count++;
3964 continue; /* If no rootdir, can't attach orphaned files */
3966 /* Here we attach orphaned files and directories into the
3967 * root directory, LVVnode, making sure link counts stay correct.
3969 if ((orphans == ORPH_ATTACH) && !vep->todelete && !Testing) {
3970 LFVnode = rootdir.vnodeNumber; /* Lost+Found vnode number */
3971 LFUnique = rootdir.unique; /* Lost+Found uniquifier */
3973 /* Update this orphaned vnode's info. Its parent info and
3974 * link count (do for orphaned directories and files).
3976 vep->parent = LFVnode; /* Parent is the root dir */
3977 vep->unique = LFUnique;
3980 vep->count--; /* Inc link count (root dir will pt to it) */
3982 /* If this orphaned vnode is a directory, change '..'.
3983 * The name of the orphaned dir/file is unknown, so we
3984 * build a unique name. No need to CopyOnWrite the directory
3985 * since it is not connected to tree in BK or RO volume and
3986 * won't be visible there.
3988 if (class == vLarge) {
3992 /* Remove and recreate the ".." entry in this orphaned directory */
3993 SetSalvageDirHandle(&dh, vid, salvinfo->fileSysDevice,
3994 salvinfo->vnodeInfo[class].inodes[v],
3995 &salvinfo->VolumeChanged);
3997 pa.Unique = LFUnique;
3998 osi_Assert(afs_dir_Delete(&dh, "..") == 0);
3999 osi_Assert(afs_dir_Create(&dh, "..", &pa) == 0);
4001 /* The original parent's link count was decremented above.
4002 * Here we increment the new parent's link count.
4004 pv = vnodeIdToBitNumber(LFVnode);
4005 salvinfo->vnodeInfo[vLarge].vnodes[pv].count--;
4009 /* Go to the root dir and add this entry. The link count of the
4010 * root dir was incremented when ".." was created. Try 10 times.
4012 for (j = 0; j < 10; j++) {
4013 pa.Vnode = ThisVnode;
4014 pa.Unique = ThisUnique;
4016 snprintf(npath, sizeof npath, "%s.%u.%u",
4017 ((class == vLarge) ? "__ORPHANDIR__"
4018 : "__ORPHANFILE__"),
4019 ThisVnode, ThisUnique);
4021 CopyOnWrite(salvinfo, &rootdir);
4022 code = afs_dir_Create(&rootdir.dirHandle, npath, &pa);
4026 ThisUnique += 50; /* Try creating a different file */
4028 osi_Assert(code == 0);
4029 Log("Attaching orphaned %s to volume's root dir as %s\n",
4030 ((class == vLarge) ? "directory" : "file"), npath);
4032 } /* for each vnode in the class */
4033 } /* for each class of vnode */
4035 /* Delete the old rootinode directory if the rootdir was CopyOnWrite */
4037 if (rootdirfound && !oldrootdir.copied && rootdir.copied) {
4039 IH_DEC(oldrootdir.ds_linkH, oldrootdir.dirHandle.dirh_inode,
4041 osi_Assert(code == 0);
4042 /* dirVnodeInfo->inodes[?] is not updated with new inode number */
4045 DFlush(); /* Flush the changes */
4046 if (!rootdirfound && (orphans == ORPH_ATTACH)) {
4047 Log("Cannot attach orphaned files and directories: Root directory not found\n");
4048 orphans = ORPH_IGNORE;
4051 /* Write out all changed vnodes. Orphaned files and directories
4052 * will get removed here also (if requested).
4054 for (class = 0; class < nVNODECLASSES; class++) {
4055 afs_sfsize_t nVnodes = salvinfo->vnodeInfo[class].nVnodes;
4056 struct VnodeClassInfo *vcp = &VnodeClassInfo[class];
4057 struct VnodeEssence *vnodes = salvinfo->vnodeInfo[class].vnodes;
4058 FilesInVolume += salvinfo->vnodeInfo[class].nAllocatedVnodes;
4059 BlocksInVolume += salvinfo->vnodeInfo[class].volumeBlockCount;
4060 for (i = 0; i < nVnodes; i++) {
4061 struct VnodeEssence *vnp = &vnodes[i];
4062 VnodeId vnodeNumber = bitNumberToVnodeNumber(i, class);
4064 /* If the vnode is good but is unclaimed (not listed in
4065 * any directory entries), then it is orphaned.
4068 if ((vnp->type != 0) && (orphaned = IsVnodeOrphaned(salvinfo, vnodeNumber))) {
4069 vnp->claimed = 0; /* Makes IsVnodeOrphaned calls faster */
4073 if (vnp->changed || vnp->count) {
4076 IH_IREAD(salvinfo->vnodeInfo[class].handle,
4077 vnodeIndexOffset(vcp, vnodeNumber),
4078 (char *)&vnode, sizeof(vnode));
4079 osi_Assert(nBytes == sizeof(vnode));
4081 vnode.parent = vnp->parent;
4082 oldCount = vnode.linkCount;
4083 vnode.linkCount = vnode.linkCount - vnp->count;
4086 orphaned = IsVnodeOrphaned(salvinfo, vnodeNumber);
4088 if (!vnp->todelete) {
4089 /* Orphans should have already been attached (if requested) */
4090 osi_Assert(orphans != ORPH_ATTACH);
4091 oblocks += vnp->blockCount;
4094 if (((orphans == ORPH_REMOVE) || vnp->todelete)
4096 BlocksInVolume -= vnp->blockCount;
4098 if (VNDISK_GET_INO(&vnode)) {
4100 IH_DEC(alinkH, VNDISK_GET_INO(&vnode), vid);
4101 osi_Assert(code == 0);
4103 memset(&vnode, 0, sizeof(vnode));
4105 } else if (vnp->count) {
4107 Log("Vnode %u: link count incorrect (was %d, %s %d)\n", vnodeNumber, oldCount, (Testing ? "would have changed to" : "now"), vnode.linkCount);
4110 vnode.modeBits = vnp->modeBits;
4113 vnode.dataVersion++;
4116 IH_IWRITE(salvinfo->vnodeInfo[class].handle,
4117 vnodeIndexOffset(vcp, vnodeNumber),
4118 (char *)&vnode, sizeof(vnode));
4119 osi_Assert(nBytes == sizeof(vnode));
4121 salvinfo->VolumeChanged = 1;
4125 if (!Showmode && ofiles) {
4126 Log("%s %d orphaned files and directories (approx. %u KB)\n",
4128 && (orphans == ORPH_REMOVE)) ? "Removed" : "Found", ofiles,
4132 for (class = 0; class < nVNODECLASSES; class++) {
4133 struct VnodeInfo *vip = &salvinfo->vnodeInfo[class];
4134 for (i = 0; i < vip->nVnodes; i++)
4135 if (vip->vnodes[i].name)
4136 free(vip->vnodes[i].name);
4143 /* Set correct resource utilization statistics */
4144 volHeader.filecount = FilesInVolume;
4145 volHeader.diskused = BlocksInVolume;
4147 /* Make sure the uniquifer is big enough: maxunique is the real maxUniquifier */
4148 if (volHeader.uniquifier < (maxunique + 1)) {
4150 Log("Volume uniquifier is too low; fixed\n");
4151 /* Plus 2,000 in case there are workstations out there with
4152 * cached vnodes that have since been deleted
4154 volHeader.uniquifier = (maxunique + 1 + 2000);
4158 Log("*** WARNING: Root directory recreated, but volume is fragile! "
4159 "Only use this salvaged volume to copy data to another volume; "
4160 "do not continue to use this volume (%lu) as-is.\n",
4161 afs_printable_uint32_lu(vid));
4164 if (!Testing && salvinfo->VolumeChanged) {
4165 #ifdef FSSYNC_BUILD_CLIENT
4166 if (salvinfo->useFSYNC) {
4167 afs_int32 fsync_code;
4169 fsync_code = FSYNC_VolOp(vid, NULL, FSYNC_VOL_BREAKCBKS, FSYNC_SALVAGE, NULL);
4171 Log("Error trying to tell the fileserver to break callbacks for "
4172 "changed volume %lu; error code %ld\n",
4173 afs_printable_uint32_lu(vid),
4174 afs_printable_int32_ld(fsync_code));
4176 salvinfo->VolumeChanged = 0;
4179 #endif /* FSSYNC_BUILD_CLIENT */
4181 #if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
4182 if (!salvinfo->useFSYNC) {
4183 /* A volume's contents have changed, but the fileserver will not
4184 * break callbacks on the volume until it tries to load the vol
4185 * header. So, to reduce the amount of time a client could have
4186 * stale data, remove fsstate.dat, so the fileserver will init
4187 * callback state with all clients. This is a very coarse hammer,
4188 * and in the future we should just record which volumes have
4190 code = unlink(AFSDIR_SERVER_FSSTATE_FILEPATH);
4191 if (code && errno != ENOENT) {
4192 Log("Error %d when trying to unlink FS state file %s\n", errno,
4193 AFSDIR_SERVER_FSSTATE_FILEPATH);
4199 /* Turn off the inUse bit; the volume's been salvaged! */
4200 volHeader.inUse = 0; /* clear flag indicating inUse@last crash */
4201 volHeader.needsSalvaged = 0; /* clear 'damaged' flag */
4202 volHeader.inService = 1; /* allow service again */
4203 volHeader.needsCallback = (salvinfo->VolumeChanged != 0);
4204 volHeader.dontSalvage = DONT_SALVAGE;
4205 salvinfo->VolumeChanged = 0;
4207 nBytes = IH_IWRITE(h, 0, (char *)&volHeader, sizeof(volHeader));
4208 osi_Assert(nBytes == sizeof(volHeader));
4211 Log("%sSalvaged %s (%u): %d files, %d blocks\n",
4212 (Testing ? "It would have " : ""), volHeader.name, volHeader.id,
4213 FilesInVolume, BlocksInVolume);
4216 IH_RELEASE(salvinfo->vnodeInfo[vSmall].handle);
4217 IH_RELEASE(salvinfo->vnodeInfo[vLarge].handle);
4223 ClearROInUseBit(struct VolumeSummary *summary)
4225 IHandle_t *h = summary->volumeInfoHandle;
4226 afs_sfsize_t nBytes;
4228 VolumeDiskData volHeader;
4230 nBytes = IH_IREAD(h, 0, (char *)&volHeader, sizeof(volHeader));
4231 osi_Assert(nBytes == sizeof(volHeader));
4232 osi_Assert(volHeader.stamp.magic == VOLUMEINFOMAGIC);
4233 volHeader.inUse = 0;
4234 volHeader.needsSalvaged = 0;
4235 volHeader.inService = 1;
4236 volHeader.dontSalvage = DONT_SALVAGE;
4238 nBytes = IH_IWRITE(h, 0, (char *)&volHeader, sizeof(volHeader));
4239 osi_Assert(nBytes == sizeof(volHeader));
4244 * Possible delete the volume.
4246 * deleteMe - Always do so, only a partial volume.
4249 MaybeZapVolume(struct SalvInfo *salvinfo, struct InodeSummary *isp,
4250 char *message, int deleteMe, int check)
4252 if (readOnly(isp) || deleteMe) {
4253 if (isp->volSummary && !isp->volSummary->deleted) {
4256 Log("Volume %u (is only a partial volume--probably an attempt was made to move/restore it when a machine crash occured.\n", isp->volumeId);
4258 Log("It will be deleted on this server (you may find it elsewhere)\n");
4261 Log("Volume %u needs to be salvaged. Since it is read-only, however,\n", isp->volumeId);
4263 Log("it will be deleted instead. It should be recloned.\n");
4268 char filename[VMAXPATHLEN];
4269 VolumeExternalName_r(isp->volumeId, filename, sizeof(filename));
4270 sprintf(path, "%s" OS_DIRSEP "%s", salvinfo->fileSysPath, filename);
4272 code = VDestroyVolumeDiskHeader(salvinfo->fileSysPartition, isp->volumeId, isp->RWvolumeId);
4274 Log("Error %ld destroying volume disk header for volume %lu\n",
4275 afs_printable_int32_ld(code),
4276 afs_printable_uint32_lu(isp->volumeId));
4279 /* make sure we actually delete the header file; ENOENT
4280 * is fine, since VDestroyVolumeDiskHeader probably already
4282 if (unlink(path) && errno != ENOENT) {
4283 Log("Unable to unlink %s (errno = %d)\n", path, errno);
4285 if (salvinfo->useFSYNC) {
4286 AskDelete(salvinfo, isp->volumeId);
4288 isp->volSummary->deleted = 1;
4291 } else if (!check) {
4292 Log("%s salvage was unsuccessful: read-write volume %u\n", message,
4294 Abort("Salvage of volume %u aborted\n", isp->volumeId);
4298 #if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
4300 * Locks a volume on disk for salvaging.
4302 * @param[in] volumeId volume ID to lock
4304 * @return operation status
4306 * @retval -1 volume lock raced with a fileserver restart; all volumes must
4307 * checked out and locked again
4312 LockVolume(struct SalvInfo *salvinfo, VolumeId volumeId)
4317 /* should always be WRITE_LOCK, but keep the lock-type logic all
4318 * in one place, in VVolLockType. Params will be ignored, but
4319 * try to provide what we're logically doing. */
4320 locktype = VVolLockType(V_VOLUPD, 1);
4322 code = VLockVolumeByIdNB(volumeId, salvinfo->fileSysPartition, locktype);
4324 if (code == EBUSY) {
4325 Abort("Someone else appears to be using volume %lu; Aborted\n",
4326 afs_printable_uint32_lu(volumeId));
4328 Abort("Error %ld trying to lock volume %lu; Aborted\n",
4329 afs_printable_int32_ld(code),
4330 afs_printable_uint32_lu(volumeId));
4333 code = FSYNC_VerifyCheckout(volumeId, salvinfo->fileSysPartition->name, FSYNC_VOL_OFF, FSYNC_SALVAGE);
4334 if (code == SYNC_DENIED) {
4335 /* need to retry checking out volumes */
4338 if (code != SYNC_OK) {
4339 Abort("FSYNC_VerifyCheckout failed for volume %lu with code %ld\n",
4340 afs_printable_uint32_lu(volumeId), afs_printable_int32_ld(code));
4343 /* set inUse = programType in the volume header to ensure that nobody
4344 * tries to use this volume again without salvaging, if we somehow crash
4345 * or otherwise exit before finishing the salvage.
4349 struct VolumeHeader header;
4350 struct VolumeDiskHeader diskHeader;
4351 struct VolumeDiskData volHeader;
4353 code = VReadVolumeDiskHeader(volumeId, salvinfo->fileSysPartition, &diskHeader);
4358 DiskToVolumeHeader(&header, &diskHeader);
4360 IH_INIT(h, salvinfo->fileSysDevice, header.parent, header.volumeInfo);
4361 if (IH_IREAD(h, 0, (char*)&volHeader, sizeof(volHeader)) != sizeof(volHeader) ||
4362 volHeader.stamp.magic != VOLUMEINFOMAGIC) {
4368 volHeader.inUse = programType;
4370 /* If we can't re-write the header, bail out and error. We don't
4371 * assert when reading the header, since it's possible the
4372 * header isn't really there (when there's no data associated
4373 * with the volume; we just delete the vol header file in that
4374 * case). But if it's there enough that we can read it, but
4375 * somehow we cannot write to it to signify we're salvaging it,
4376 * we've got a big problem and we cannot continue. */
4377 osi_Assert(IH_IWRITE(h, 0, (char*)&volHeader, sizeof(volHeader)) == sizeof(volHeader));
4384 #endif /* AFS_DEMAND_ATTACH_FS || AFS_DEMAND_ATTACH_UTIL */
4387 AskError(struct SalvInfo *salvinfo, VolumeId volumeId)
4389 #if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
4391 code = FSYNC_VolOp(volumeId, salvinfo->fileSysPartition->name,
4392 FSYNC_VOL_FORCE_ERROR, FSYNC_WHATEVER, NULL);
4393 if (code != SYNC_OK) {
4394 Log("AskError: failed to force volume %lu into error state; "
4395 "SYNC error code %ld (%s)\n", (long unsigned)volumeId,
4396 (long)code, SYNC_res2string(code));
4398 #endif /* AFS_DEMAND_ATTACH_FS || AFS_DEMAND_ATTACH_UTIL */
4402 AskOffline(struct SalvInfo *salvinfo, VolumeId volumeId)
4407 memset(&res, 0, sizeof(res));
4409 for (i = 0; i < 3; i++) {
4410 code = FSYNC_VolOp(volumeId, salvinfo->fileSysPartition->name,
4411 FSYNC_VOL_OFF, FSYNC_SALVAGE, &res);
4413 if (code == SYNC_OK) {
4415 } else if (code == SYNC_DENIED) {
4417 Log("AskOffline: file server denied offline request; a general salvage may be required.\n");
4419 Log("AskOffline: file server denied offline request; a general salvage is required.\n");
4420 Abort("Salvage aborted\n");
4421 } else if (code == SYNC_BAD_COMMAND) {
4422 Log("AskOffline: fssync protocol mismatch (bad command word '%d'); salvage aborting.\n",
4425 #if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
4426 Log("AskOffline: please make sure dafileserver, davolserver, salvageserver and dasalvager binaries are same version.\n");
4428 Log("AskOffline: fileserver is DAFS but we are not.\n");
4431 #if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
4432 Log("AskOffline: fileserver is not DAFS but we are.\n");
4434 Log("AskOffline: please make sure fileserver, volserver and salvager binaries are same version.\n");
4437 Abort("Salvage aborted\n");
4440 Log("AskOffline: request for fileserver to take volume offline failed; trying again...\n");
4441 FSYNC_clientFinis();
4445 if (code != SYNC_OK) {
4446 Log("AskOffline: request for fileserver to take volume offline failed; salvage aborting.\n");
4447 Abort("Salvage aborted\n");
4451 /* don't want to pass around state; remember it here */
4452 static int isDAFS = -1;
4457 afs_int32 code = 1, i;
4459 /* we don't care if we race. the answer shouldn't change */
4463 memset(&res, 0, sizeof(res));
4465 for (i = 0; code && i < 3; i++) {
4466 code = FSYNC_VolOp(0, NULL, FSYNC_VOL_LISTVOLUMES, FSYNC_SALVAGE, &res);
4468 Log("AskDAFS: FSYNC_VOL_LISTVOLUMES failed with code %ld reason "
4469 "%ld (%s); trying again...\n", (long)code, (long)res.hdr.reason,
4470 FSYNC_reason2string(res.hdr.reason));
4471 FSYNC_clientFinis();
4477 Log("AskDAFS: could not determine DAFS-ness, assuming not DAFS\n");
4481 if ((res.hdr.flags & SYNC_FLAG_DAFS_EXTENSIONS)) {
4491 MaybeAskOnline(struct SalvInfo *salvinfo, VolumeId volumeId)
4493 struct VolumeDiskHeader diskHdr;
4495 code = VReadVolumeDiskHeader(volumeId, salvinfo->fileSysPartition, &diskHdr);
4497 /* volume probably does not exist; no need to bring back online */
4500 AskOnline(salvinfo, volumeId);
4504 AskOnline(struct SalvInfo *salvinfo, VolumeId volumeId)
4508 for (i = 0; i < 3; i++) {
4509 code = FSYNC_VolOp(volumeId, salvinfo->fileSysPartition->name,
4510 FSYNC_VOL_ON, FSYNC_WHATEVER, NULL);
4512 if (code == SYNC_OK) {
4514 } else if (code == SYNC_DENIED) {
4515 Log("AskOnline: file server denied online request to volume %u partition %s; trying again...\n", volumeId, salvinfo->fileSysPartition->name);
4516 } else if (code == SYNC_BAD_COMMAND) {
4517 Log("AskOnline: fssync protocol mismatch (bad command word '%d')\n",
4519 Log("AskOnline: please make sure file server binaries are same version.\n");
4523 Log("AskOnline: request for fileserver to put volume online failed; trying again...\n");
4524 FSYNC_clientFinis();
4531 AskDelete(struct SalvInfo *salvinfo, VolumeId volumeId)
4536 for (i = 0; i < 3; i++) {
4537 memset(&res, 0, sizeof(res));
4538 code = FSYNC_VolOp(volumeId, salvinfo->fileSysPartition->name,
4539 FSYNC_VOL_DONE, FSYNC_SALVAGE, &res);
4541 if (code == SYNC_OK) {
4543 } else if (code == SYNC_DENIED) {
4544 Log("AskOnline: file server denied DONE request to volume %u partition %s; trying again...\n", volumeId, salvinfo->fileSysPartition->name);
4545 } else if (code == SYNC_BAD_COMMAND) {
4546 Log("AskOnline: fssync protocol mismatch (bad command word '%d')\n",
4549 #if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
4550 Log("AskOnline: please make sure dafileserver, davolserver, salvageserver and dasalvager binaries are same version.\n");
4552 Log("AskOnline: fileserver is DAFS but we are not.\n");
4555 #if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
4556 Log("AskOnline: fileserver is not DAFS but we are.\n");
4558 Log("AskOnline: please make sure fileserver, volserver and salvager binaries are same version.\n");
4562 } else if (code == SYNC_FAILED &&
4563 (res.hdr.reason == FSYNC_UNKNOWN_VOLID ||
4564 res.hdr.reason == FSYNC_WRONG_PART)) {
4565 /* volume is already effectively 'deleted' */
4569 Log("AskOnline: request for fileserver to delete volume failed; trying again...\n");
4570 FSYNC_clientFinis();
4577 CopyInode(Device device, Inode inode1, Inode inode2, int rwvolume)
4579 /* Volume parameter is passed in case iopen is upgraded in future to
4580 * require a volume Id to be passed
4583 IHandle_t *srcH, *destH;
4584 FdHandle_t *srcFdP, *destFdP;
4586 afs_foff_t size = 0;
4588 IH_INIT(srcH, device, rwvolume, inode1);
4589 srcFdP = IH_OPEN(srcH);
4590 osi_Assert(srcFdP != NULL);
4591 IH_INIT(destH, device, rwvolume, inode2);
4592 destFdP = IH_OPEN(destH);
4593 while ((nBytes = FDH_PREAD(srcFdP, buf, sizeof(buf), size)) > 0) {
4594 osi_Assert(FDH_PWRITE(destFdP, buf, nBytes, size) == nBytes);
4597 osi_Assert(nBytes == 0);
4598 FDH_REALLYCLOSE(srcFdP);
4599 FDH_REALLYCLOSE(destFdP);
4606 PrintInodeList(struct SalvInfo *salvinfo)
4608 struct ViceInodeInfo *ip;
4609 struct ViceInodeInfo *buf;
4612 afs_sfsize_t st_size;
4614 st_size = OS_SIZE(salvinfo->inodeFd);
4615 osi_Assert(st_size >= 0);
4616 buf = (struct ViceInodeInfo *)malloc(st_size);
4617 osi_Assert(buf != NULL);
4618 nInodes = st_size / sizeof(struct ViceInodeInfo);
4619 osi_Assert(OS_READ(salvinfo->inodeFd, buf, st_size) == st_size);
4620 for (ip = buf; nInodes--; ip++) {
4621 Log("Inode:%s, linkCount=%d, size=%#llx, p=(%u,%u,%u,%u)\n",
4622 PrintInode(stmp, ip->inodeNumber), ip->linkCount,
4623 (afs_uintmax_t) ip->byteCount, ip->u.param[0], ip->u.param[1],
4624 ip->u.param[2], ip->u.param[3]);
4630 PrintInodeSummary(struct SalvInfo *salvinfo)
4633 struct InodeSummary *isp;
4635 for (i = 0; i < salvinfo->nVolumesInInodeFile; i++) {
4636 isp = &salvinfo->inodeSummary[i];
4637 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);
4647 osi_Assert(0); /* Fork is never executed in the NT code path */
4651 #ifdef AFS_DEMAND_ATTACH_FS
4652 if ((f == 0) && (programType == salvageServer)) {
4653 /* we are a salvageserver child */
4654 #ifdef FSSYNC_BUILD_CLIENT
4655 VChildProcReconnectFS_r();
4657 #ifdef SALVSYNC_BUILD_CLIENT
4661 #endif /* AFS_DEMAND_ATTACH_FS */
4662 #endif /* !AFS_NT40_ENV */
4672 #ifdef AFS_DEMAND_ATTACH_FS
4673 if (programType == salvageServer) {
4674 #ifdef SALVSYNC_BUILD_CLIENT
4677 #ifdef FSSYNC_BUILD_CLIENT
4681 #endif /* AFS_DEMAND_ATTACH_FS */
4684 if (main_thread != pthread_self())
4685 pthread_exit((void *)code);
4698 pid = wait(&status);
4699 osi_Assert(pid != -1);
4700 if (WCOREDUMP(status))
4701 Log("\"%s\" core dumped!\n", prog);
4702 if (WIFSIGNALED(status) != 0 || WEXITSTATUS(status) != 0)
4708 TimeStamp(time_t clock, int precision)
4711 static char timestamp[20];
4712 lt = localtime(&clock);
4714 (void)strftime(timestamp, 20, "%m/%d/%Y %H:%M:%S", lt);
4716 (void)strftime(timestamp, 20, "%m/%d/%Y %H:%M", lt);
4721 CheckLogFile(char * log_path)
4723 char oldSlvgLog[AFSDIR_PATH_MAX];
4725 #ifndef AFS_NT40_ENV
4732 strcpy(oldSlvgLog, log_path);
4733 strcat(oldSlvgLog, ".old");
4735 renamefile(log_path, oldSlvgLog);
4736 logFile = afs_fopen(log_path, "a");
4738 if (!logFile) { /* still nothing, use stdout */
4742 #ifndef AFS_NAMEI_ENV
4743 AFS_DEBUG_IOPS_LOG(logFile);
4748 #ifndef AFS_NT40_ENV
4750 TimeStampLogFile(char * log_path)
4752 char stampSlvgLog[AFSDIR_PATH_MAX];
4757 lt = localtime(&now);
4758 snprintf(stampSlvgLog, sizeof stampSlvgLog,
4759 "%s.%04d-%02d-%02d.%02d:%02d:%02d", log_path,
4760 lt->tm_year + 1900, lt->tm_mon + 1, lt->tm_mday, lt->tm_hour,
4761 lt->tm_min, lt->tm_sec);
4763 /* try to link the logfile to a timestamped filename */
4764 /* if it fails, oh well, nothing we can do */
4765 link(log_path, stampSlvgLog);
4774 #ifndef AFS_NT40_ENV
4776 printf("Can't show log since using syslog.\n");
4787 logFile = afs_fopen(AFSDIR_SERVER_SLVGLOG_FILEPATH, "r");
4790 printf("Can't read %s, exiting\n", AFSDIR_SERVER_SLVGLOG_FILEPATH);
4793 while (fgets(line, sizeof(line), logFile))
4800 Log(const char *format, ...)
4806 va_start(args, format);
4807 vsnprintf(tmp, sizeof tmp, format, args);
4809 #ifndef AFS_NT40_ENV
4811 syslog(LOG_INFO, "%s", tmp);
4815 gettimeofday(&now, NULL);
4816 fprintf(logFile, "%s %s", TimeStamp(now.tv_sec, 1), tmp);
4822 Abort(const char *format, ...)
4827 va_start(args, format);
4828 vsnprintf(tmp, sizeof tmp, format, args);
4830 #ifndef AFS_NT40_ENV
4832 syslog(LOG_INFO, "%s", tmp);
4836 fprintf(logFile, "%s", tmp);
4848 ToString(const char *s)
4851 p = (char *)malloc(strlen(s) + 1);
4852 osi_Assert(p != NULL);
4857 /* Remove the FORCESALVAGE file */
4859 RemoveTheForce(char *path)
4862 struct afs_stat_st force; /* so we can use afs_stat to find it */
4863 strcpy(target,path);
4864 strcat(target,"/FORCESALVAGE");
4865 if (!Testing && ForceSalvage) {
4866 if (afs_stat(target,&force) == 0) unlink(target);
4870 #ifndef AFS_AIX32_ENV
4872 * UseTheForceLuke - see if we can use the force
4875 UseTheForceLuke(char *path)
4877 struct afs_stat_st force;
4879 strcpy(target,path);
4880 strcat(target,"/FORCESALVAGE");
4882 return (afs_stat(target, &force) == 0);
4886 * UseTheForceLuke - see if we can use the force
4889 * The VRMIX fsck will not muck with the filesystem it is supposedly
4890 * fixing and create a "FORCESALVAGE" file (by design). Instead, we
4891 * muck directly with the root inode, which is within the normal
4893 * ListViceInodes() has a side effect of setting ForceSalvage if
4894 * it detects a need, based on root inode examination.
4897 UseTheForceLuke(char *path)
4900 return 0; /* sorry OB1 */
4905 /* NT support routines */
4907 static char execpathname[MAX_PATH];
4909 nt_SalvagePartition(char *partName, int jobn)
4914 if (!*execpathname) {
4915 n = GetModuleFileName(NULL, execpathname, MAX_PATH - 1);
4916 if (!n || n == 1023)
4919 job.cj_magic = SALVAGER_MAGIC;
4920 job.cj_number = jobn;
4921 (void)strcpy(job.cj_part, partName);
4922 pid = (int)spawnprocveb(execpathname, save_args, NULL, &job, sizeof(job));
4927 nt_SetupPartitionSalvage(void *datap, int len)
4929 childJob_t *jobp = (childJob_t *) datap;
4930 char logname[AFSDIR_PATH_MAX];
4932 if (len != sizeof(childJob_t))
4934 if (jobp->cj_magic != SALVAGER_MAGIC)
4939 (void)sprintf(logname, "%s.%d", AFSDIR_SERVER_SLVGLOG_FILEPATH,
4941 logFile = afs_fopen(logname, "w");
4949 #endif /* AFS_NT40_ENV */