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 #ifdef AFS_PTHREAD_ENV
104 # include <opr/lock.h>
107 #include <afs/afsint.h>
108 #if !defined(AFS_SGI_ENV) && !defined(AFS_NT40_ENV)
109 #if defined(AFS_VFSINCL_ENV)
110 #include <sys/vnode.h>
112 #include <sys/fs/ufs_inode.h>
114 #if defined(AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV)
115 #include <ufs/ufs/dinode.h>
116 #include <ufs/ffs/fs.h>
118 #include <ufs/inode.h>
121 #else /* AFS_VFSINCL_ENV */
122 #if !defined(AFS_LINUX20_ENV) && !defined(AFS_XBSD_ENV) && !defined(AFS_DARWIN_ENV)
123 #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>
156 #include <rx/rx_queue.h>
161 #include <afs/afssyscalls.h>
165 #include "partition.h"
166 #include "daemon_com.h"
167 #include "daemon_com_inline.h"
169 #include "fssync_inline.h"
170 #include "volume_inline.h"
171 #include "salvsync.h"
172 #include "viceinode.h"
174 #include "volinodes.h" /* header magic number, etc. stuff */
175 #include "vol-salvage.h"
177 #include "vol_internal.h"
179 #include <afs/prs_fs.h>
181 #ifdef FSSYNC_BUILD_CLIENT
182 #include "vg_cache.h"
189 #define SALV_BUFFER_SIZE 1024
191 static char *TimeStamp(char *buffer, size_t size, 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 char *ShowLogFilename = NULL; /* log file name for -showlog */
204 int ShowSuid = 0; /* -showsuid flag */
205 int ShowMounts = 0; /* -showmounts flag */
206 int orphans = ORPH_IGNORE; /* -orphans option */
208 int ClientMode = 0; /* running as salvager server client */
216 #define MAXPARALLEL 32
218 int OKToZap; /* -o flag */
219 int ForceSalvage; /* If salvage should occur despite the DONT_SALVAGE flag
220 * in the volume header */
222 #define ROOTINODE 2 /* Root inode of a 4.2 Unix file system
225 * information that is 'global' to a particular salvage job.
228 Device fileSysDevice; /**< The device number of the current partition
230 char fileSysPath[9]; /**< The path of the mounted partition currently
231 * being salvaged, i.e. the directory containing
232 * the volume headers */
233 char *fileSysPathName; /**< NT needs this to make name pretty log. */
234 IHandle_t *VGLinkH; /**< Link handle for current volume group. */
235 int VGLinkH_cnt; /**< # of references to lnk handle. */
236 struct DiskPartition64 *fileSysPartition; /**< Partition being salvaged */
239 char *fileSysDeviceName; /**< The block device where the file system being
240 * salvaged was mounted */
241 char *filesysfulldev;
243 int VolumeChanged; /**< Set by any routine which would change the
244 * volume in a way which would require callbacks
245 * to be broken if the volume was put back on
246 * on line by an active file server */
248 VolumeDiskData VolInfo; /**< A copy of the last good or salvaged volume
249 * header dealt with */
251 int nVolumesInInodeFile; /**< Number of read-write volumes summarized */
252 FD_t inodeFd; /**< File descriptor for inode file */
254 struct VolumeSummary *volumeSummaryp; /**< Holds all the volumes in a part */
255 int nVolumes; /**< Number of volumes (read-write and read-only)
256 * in volume summary */
257 struct InodeSummary *inodeSummary; /**< contains info on all the relevant
260 struct VnodeInfo vnodeInfo[nVNODECLASSES]; /**< contains info on all of the
261 * vnodes in the volume that
262 * we are currently looking
264 int useFSYNC; /**< 0 if the fileserver is unavailable; 1 if we should try
265 * to contact the fileserver over FSYNC */
272 /* Forward declarations */
273 static void QuietExit(int) AFS_NORETURN;
274 static void SalvageShowLog(void);
275 static int IsVnodeOrphaned(struct SalvInfo *salvinfo, VnodeId vnode);
276 static int AskVolumeSummary(struct SalvInfo *salvinfo,
277 VolumeId singleVolumeNumber);
278 static void MaybeAskOnline(struct SalvInfo *salvinfo, VolumeId volumeId);
279 static void AskError(struct SalvInfo *salvinfo, VolumeId volumeId);
281 #ifdef AFS_DEMAND_ATTACH_FS
282 static int LockVolume(struct SalvInfo *salvinfo, VolumeId volumeId);
283 #endif /* AFS_DEMAND_ATTACH_FS */
285 /* Uniquifier stored in the Inode */
290 return (u & 0x3fffff);
292 #if defined(AFS_SGI_EXMAG)
293 return (u & SGI_UNIQMASK);
296 #endif /* AFS_SGI_EXMAG */
303 if (aerror == EPERM || aerror == ENXIO || aerror == ENOENT)
305 return 0; /* otherwise may be transient, e.g. EMFILE */
310 char *save_args[MAX_ARGS];
312 extern pthread_t main_thread;
313 childJob_t myjob = { SALVAGER_MAGIC, NOT_CHILD, "" };
317 * Get the salvage lock if not already held. Hold until process exits.
319 * @param[in] locktype READ_LOCK or WRITE_LOCK
322 _ObtainSalvageLock(int locktype)
324 struct VLockFile salvageLock;
329 VLockFileInit(&salvageLock, AFSDIR_SERVER_SLVGLOCK_FILEPATH);
331 code = VLockFileLock(&salvageLock, offset, locktype, nonblock);
334 "salvager: There appears to be another salvager running! "
339 "salvager: Error %d trying to acquire salvage lock! "
345 ObtainSalvageLock(void)
347 _ObtainSalvageLock(WRITE_LOCK);
350 ObtainSharedSalvageLock(void)
352 _ObtainSalvageLock(READ_LOCK);
356 #ifdef AFS_SGI_XFS_IOPS_ENV
357 /* Check if the given partition is mounted. For XFS, the root inode is not a
358 * constant. So we check the hard way.
361 IsPartitionMounted(char *part)
364 struct mntent *mntent;
366 opr_Verify(mntfp = setmntent(MOUNTED, "r"));
367 while (mntent = getmntent(mntfp)) {
368 if (!strcmp(part, mntent->mnt_dir))
373 return mntent ? 1 : 1;
376 /* Check if the given inode is the root of the filesystem. */
377 #ifndef AFS_SGI_XFS_IOPS_ENV
379 IsRootInode(struct afs_stat_st *status)
382 * The root inode is not a fixed value in XFS partitions. So we need to
383 * see if the partition is in the list of mounted partitions. This only
384 * affects the SalvageFileSys path, so we check there.
386 return (status->st_ino == ROOTINODE);
391 #ifndef AFS_NAMEI_ENV
392 /* We don't want to salvage big files filesystems, since we can't put volumes on
396 CheckIfBigFilesFS(char *mountPoint, char *devName)
398 struct superblock fs;
401 if (strncmp(devName, "/dev/", 5)) {
402 (void)sprintf(name, "/dev/%s", devName);
404 (void)strcpy(name, devName);
407 if (ReadSuper(&fs, name) < 0) {
408 Log("Unable to read superblock. Not salvaging partition %s.\n",
412 if (IsBigFilesFileSystem(&fs)) {
413 Log("Partition %s is a big files filesystem, not salvaging.\n",
423 #define HDSTR "\\Device\\Harddisk"
424 #define HDLEN (sizeof(HDSTR)-1) /* Length of "\Device\Harddisk" */
426 SameDisk(struct DiskPartition64 *p1, struct DiskPartition64 *p2)
432 static int dowarn = 1;
434 if (!QueryDosDevice(p1->devName, res1, RES_LEN - 1))
436 if (strncmp(res1, HDSTR, HDLEN)) {
439 Log("WARNING: QueryDosDevice is returning %s, not %s for %s\n",
440 res1, HDSTR, p1->devName);
443 if (!QueryDosDevice(p2->devName, res2, RES_LEN - 1))
445 if (strncmp(res2, HDSTR, HDLEN)) {
448 Log("WARNING: QueryDosDevice is returning %s, not %s for %s\n",
449 res2, HDSTR, p2->devName);
453 return (0 == _strnicmp(res1, res2, RES_LEN - 1));
456 #define SameDisk(P1, P2) ((P1)->device/PartsPerDisk == (P2)->device/PartsPerDisk)
459 /* This assumes that two partitions with the same device number divided by
460 * PartsPerDisk are on the same disk.
463 SalvageFileSysParallel(struct DiskPartition64 *partP)
466 struct DiskPartition64 *partP;
467 int pid; /* Pid for this job */
468 int jobnumb; /* Log file job number */
469 struct job *nextjob; /* Next partition on disk to salvage */
471 static struct job *jobs[MAXPARALLEL] = { 0 }; /* Need to zero this */
472 struct job *thisjob = 0;
473 static int numjobs = 0;
474 static int jobcount = 0;
482 /* We have a partition to salvage. Copy it into thisjob */
483 thisjob = calloc(1, sizeof(struct job));
485 Log("Can't salvage '%s'. Not enough memory\n", partP->name);
488 thisjob->partP = partP;
489 thisjob->jobnumb = jobcount;
491 } else if (jobcount == 0) {
492 /* We are asking to wait for all jobs (partp == 0), yet we never
495 Log("No file system partitions named %s* found; not salvaged\n",
496 VICE_PARTITION_PREFIX);
500 if (debug || Parallel == 1) {
502 SalvageFileSys(thisjob->partP, 0);
509 /* Check to see if thisjob is for a disk that we are already
510 * salvaging. If it is, link it in as the next job to do. The
511 * jobs array has 1 entry per disk being salvages. numjobs is
512 * the total number of disks currently being salvaged. In
513 * order to keep thejobs array compact, when a disk is
514 * completed, the hightest element in the jobs array is moved
515 * down to now open slot.
517 for (j = 0; j < numjobs; j++) {
518 if (SameDisk(jobs[j]->partP, thisjob->partP)) {
519 /* On same disk, add it to this list and return */
520 thisjob->nextjob = jobs[j]->nextjob;
521 jobs[j]->nextjob = thisjob;
528 /* Loop until we start thisjob or until all existing jobs are finished */
529 while (thisjob || (!partP && (numjobs > 0))) {
530 startjob = -1; /* No new job to start */
532 if ((numjobs >= Parallel) || (!partP && (numjobs > 0))) {
533 /* Either the max jobs are running or we have to wait for all
534 * the jobs to finish. In either case, we wait for at least one
535 * job to finish. When it's done, clean up after it.
537 pid = wait(&wstatus);
538 opr_Assert(pid != -1);
539 for (j = 0; j < numjobs; j++) { /* Find which job it is */
540 if (pid == jobs[j]->pid)
543 opr_Assert(j < numjobs);
544 if (WCOREDUMP(wstatus)) { /* Say if the job core dumped */
545 Log("Salvage of %s core dumped!\n", jobs[j]->partP->name);
548 numjobs--; /* job no longer running */
549 oldjob = jobs[j]; /* remember */
550 jobs[j] = jobs[j]->nextjob; /* Step to next part on same disk */
551 free(oldjob); /* free the old job */
553 /* If there is another partition on the disk to salvage, then
554 * say we will start it (startjob). If not, then put thisjob there
555 * and say we will start it.
557 if (jobs[j]) { /* Another partitions to salvage */
558 startjob = j; /* Will start it */
559 } else { /* There is not another partition to salvage */
561 jobs[j] = thisjob; /* Add thisjob */
563 startjob = j; /* Will start it */
565 jobs[j] = jobs[numjobs]; /* Move last job up to this slot */
566 startjob = -1; /* Don't start it - already running */
570 /* We don't have to wait for a job to complete */
572 jobs[numjobs] = thisjob; /* Add this job */
574 startjob = numjobs; /* Will start it */
578 /* Start up a new salvage job on a partition in job slot "startjob" */
579 if (startjob != -1) {
581 Log("Starting salvage of file system partition %s\n",
582 jobs[startjob]->partP->name);
584 /* For NT, we not only fork, but re-exec the salvager. Pass in the
585 * commands and pass the child job number via the data path.
588 nt_SalvagePartition(jobs[startjob]->partP->name,
589 jobs[startjob]->jobnumb);
590 jobs[startjob]->pid = pid;
595 jobs[startjob]->pid = pid;
600 struct logOptions logopts;
602 memset(&logopts, 0, sizeof(logopts));
603 logopts.lopt_dest = logDest_file;
605 for (fd = 0; fd < 16; fd++)
611 ShowLog = 0; /* Child processes do not display. */
612 if (asprintf(&filename, "%s.%d",
613 AFSDIR_SERVER_SLVGLOG_FILEPATH,
614 jobs[startjob]->jobnumb) >= 0) {
615 logopts.lopt_filename = filename;
620 SalvageFileSys1(jobs[startjob]->partP, 0);
625 } /* while ( thisjob || (!partP && numjobs > 0) ) */
628 * If waited for all jobs to complete, now collect log files and return.
629 * No files can be collected when logging to the system log (syslog).
631 if (GetLogDest() == logDest_file) {
633 char *buf = calloc(1, SALV_BUFFER_SIZE);
637 Log("out of memory");
639 for (i = 0; i < jobcount; i++) {
640 if (asprintf(&logFileName, "%s.%d",
641 AFSDIR_SERVER_SLVGLOG_FILEPATH, i) < 0) {
642 Log("out of memory");
645 if ((passLog = afs_fopen(logFileName, "r"))) {
646 while (fgets(buf, SALV_BUFFER_SIZE, passLog)) {
647 WriteLogBuffer(buf, strlen(buf));
651 (void)unlink(logFileName);
663 SalvageFileSys(struct DiskPartition64 *partP, VolumeId singleVolumeNumber)
665 if (!canfork || debug || Fork() == 0) {
666 SalvageFileSys1(partP, singleVolumeNumber);
667 if (canfork && !debug) {
671 Wait("SalvageFileSys");
675 get_DevName(char *pbuffer, char *wpath)
677 char pbuf[128], *ptr;
678 strcpy(pbuf, pbuffer);
679 ptr = (char *)strrchr(pbuf, OS_DIRSEPC);
685 ptr = (char *)strrchr(pbuffer, OS_DIRSEPC);
687 strcpy(pbuffer, ptr + 1);
694 SalvageFileSys1(struct DiskPartition64 *partP, VolumeId singleVolumeNumber)
697 char inodeListPath[256];
698 FD_t inodeFile = INVALID_FD;
699 static char tmpDevName[100];
700 static char wpath[100];
701 struct VolumeSummary *vsp, *esp;
705 struct SalvInfo l_salvinfo;
706 struct SalvInfo *salvinfo = &l_salvinfo;
709 memset(salvinfo, 0, sizeof(*salvinfo));
712 if (inodeFile != INVALID_FD) {
714 inodeFile = INVALID_FD;
716 if (tries > VOL_MAX_CHECKOUT_RETRIES) {
717 Abort("Raced too many times with fileserver restarts while trying to "
718 "checkout/lock volumes; Aborted\n");
720 #ifdef AFS_DEMAND_ATTACH_FS
722 /* unlock all previous volume locks, since we're about to lock them
724 VLockFileReinit(&partP->volLockFile);
726 #endif /* AFS_DEMAND_ATTACH_FS */
728 salvinfo->fileSysPartition = partP;
729 salvinfo->fileSysDevice = salvinfo->fileSysPartition->device;
730 salvinfo->fileSysPathName = VPartitionPath(salvinfo->fileSysPartition);
733 /* Opendir can fail on "C:" but not on "C:\" if C is empty! */
734 (void)sprintf(salvinfo->fileSysPath, "%s" OS_DIRSEP, salvinfo->fileSysPathName);
735 name = partP->devName;
737 strlcpy(salvinfo->fileSysPath, salvinfo->fileSysPathName, sizeof(salvinfo->fileSysPath));
738 strcpy(tmpDevName, partP->devName);
739 name = get_DevName(tmpDevName, wpath);
740 salvinfo->fileSysDeviceName = name;
741 salvinfo->filesysfulldev = wpath;
744 if (singleVolumeNumber) {
745 #ifndef AFS_DEMAND_ATTACH_FS
746 /* only non-DAFS locks the partition when salvaging a single volume;
747 * DAFS will lock the individual volumes in the VG */
748 VLockPartition(partP->name);
749 #endif /* !AFS_DEMAND_ATTACH_FS */
753 /* salvageserver already setup fssync conn for us */
754 if ((programType != salvageServer) && !VConnectFS()) {
755 Abort("Couldn't connect to file server\n");
758 salvinfo->useFSYNC = 1;
759 AskOffline(salvinfo, singleVolumeNumber);
760 #ifdef AFS_DEMAND_ATTACH_FS
761 if (LockVolume(salvinfo, singleVolumeNumber)) {
764 #endif /* AFS_DEMAND_ATTACH_FS */
767 salvinfo->useFSYNC = 0;
768 VLockPartition(partP->name);
772 ForceSalvage = UseTheForceLuke(salvinfo->fileSysPath);
775 Log("SALVAGING FILE SYSTEM PARTITION %s (device=%s%s)\n",
776 partP->name, name, (Testing ? "(READONLY mode)" : ""));
778 Log("***Forced salvage of all volumes on this partition***\n");
783 * Remove any leftover /vicepa/salvage.inodes.* or /vicepa/salvage.temp.*
790 opr_Verify((dirp = opendir(salvinfo->fileSysPath)) != NULL);
791 while ((dp = readdir(dirp))) {
792 if (!strncmp(dp->d_name, "salvage.inodes.", 15)
793 || !strncmp(dp->d_name, "salvage.temp.", 13)) {
795 Log("Removing old salvager temp files %s\n", dp->d_name);
796 strcpy(npath, salvinfo->fileSysPath);
797 strcat(npath, OS_DIRSEP);
798 strcat(npath, dp->d_name);
804 tdir = (tmpdir ? tmpdir : salvinfo->fileSysPath);
806 (void)_putenv("TMP="); /* If "TMP" is set, then that overrides tdir. */
807 (void)strncpy(inodeListPath, _tempnam(tdir, "salvage.inodes."), 255);
809 snprintf(inodeListPath, 255, "%s" OS_DIRSEP "salvage.inodes.%s.%d", tdir, name,
813 inodeFile = OS_OPEN(inodeListPath, O_RDWR|O_TRUNC|O_CREAT, 0666);
814 if (inodeFile == INVALID_FD) {
815 Abort("Error %d when creating inode description file %s; not salvaged\n", errno, inodeListPath);
818 /* Using nt_unlink here since we're really using the delete on close
819 * semantics of unlink. In most places in the salvager, we really do
820 * mean to unlink the file at that point. Those places have been
821 * modified to actually do that so that the NT crt can be used there.
823 * jaltman - On NT delete on close cannot be applied to a file while the
824 * process has an open file handle that does not have DELETE file
825 * access and FILE_SHARE_DELETE. fopen() calls CreateFile() without
826 * delete privileges. As a result the nt_unlink() call will always
829 code = nt_unlink(inodeListPath);
831 code = unlink(inodeListPath);
834 Log("Error %d when trying to unlink %s\n", errno, inodeListPath);
837 if (GetInodeSummary(salvinfo, inodeFile, singleVolumeNumber) < 0) {
841 salvinfo->inodeFd = inodeFile;
842 if (salvinfo->inodeFd == INVALID_FD)
843 Abort("Temporary file %s is missing...\n", inodeListPath);
844 OS_SEEK(salvinfo->inodeFd, 0L, SEEK_SET);
845 if (ListInodeOption) {
846 PrintInodeList(salvinfo);
847 if (singleVolumeNumber) {
848 /* We've checked out the volume from the fileserver, and we need
849 * to give it back. We don't know if the volume exists or not,
850 * so we don't know whether to AskOnline or not. Try to determine
851 * if the volume exists by trying to read the volume header, and
852 * AskOnline if it is readable. */
853 MaybeAskOnline(salvinfo, singleVolumeNumber);
857 /* enumerate volumes in the partition.
858 * figure out sets of read-only + rw volumes.
859 * salvage each set, read-only volumes first, then read-write.
860 * Fix up inodes on last volume in set (whether it is read-write
863 if (GetVolumeSummary(salvinfo, singleVolumeNumber)) {
867 if (singleVolumeNumber) {
868 /* If we delete a volume during the salvage, we indicate as such by
869 * setting the volsummary->deleted field. We need to know if we
870 * deleted a volume or not in order to know which volumes to bring
871 * back online after the salvage. If we fork, we will lose this
872 * information, since volsummary->deleted will not get set in the
873 * parent. So, don't fork. */
877 for (i = j = 0, vsp = salvinfo->volumeSummaryp, esp = vsp + salvinfo->nVolumes;
878 i < salvinfo->nVolumesInInodeFile; i = j) {
879 VolumeId rwvid = salvinfo->inodeSummary[i].RWvolumeId;
881 j < salvinfo->nVolumesInInodeFile && salvinfo->inodeSummary[j].RWvolumeId == rwvid;
883 VolumeId vid = salvinfo->inodeSummary[j].volumeId;
884 struct VolumeSummary *tsp;
885 /* Scan volume list (from partition root directory) looking for the
886 * current rw volume number in the volume list from the inode scan.
887 * If there is one here that is not in the inode volume list,
889 for (; vsp < esp && (vsp->header.parent < rwvid); vsp++) {
891 DeleteExtraVolumeHeaderFile(salvinfo, vsp);
893 /* Now match up the volume summary info from the root directory with the
894 * entry in the volume list obtained from scanning inodes */
895 salvinfo->inodeSummary[j].volSummary = NULL;
896 for (tsp = vsp; tsp < esp && (tsp->header.parent == rwvid); tsp++) {
897 if (tsp->header.id == vid) {
898 salvinfo->inodeSummary[j].volSummary = tsp;
904 /* Salvage the group of volumes (several read-only + 1 read/write)
905 * starting with the current read-only volume we're looking at.
908 nt_SalvageVolumeGroup(salvinfo, &salvinfo->inodeSummary[i], j - i);
910 DoSalvageVolumeGroup(salvinfo, &salvinfo->inodeSummary[i], j - i);
911 #endif /* AFS_NT40_ENV */
915 /* Delete any additional volumes that were listed in the partition but which didn't have any corresponding inodes */
916 for (; vsp < esp; vsp++) {
918 DeleteExtraVolumeHeaderFile(salvinfo, vsp);
921 if (!singleVolumeNumber) /* Remove the FORCESALVAGE file */
922 RemoveTheForce(salvinfo->fileSysPath);
924 if (!Testing && singleVolumeNumber) {
926 #ifdef AFS_DEMAND_ATTACH_FS
927 /* unlock vol headers so the fs can attach them when we AskOnline */
928 VLockFileReinit(&salvinfo->fileSysPartition->volLockFile);
929 #endif /* AFS_DEMAND_ATTACH_FS */
931 /* Step through the volumeSummary list and set all volumes on-line.
932 * Most volumes were taken off-line in GetVolumeSummary.
933 * If a volume was deleted, don't tell the fileserver anything, since
934 * we already told the fileserver the volume was deleted back when we
935 * we destroyed the volume header.
936 * Also, make sure we bring the singleVolumeNumber back online first.
939 for (j = 0; j < salvinfo->nVolumes; j++) {
940 if (salvinfo->volumeSummaryp[j].header.id == singleVolumeNumber) {
942 if (!salvinfo->volumeSummaryp[j].deleted) {
943 AskOnline(salvinfo, singleVolumeNumber);
949 /* If singleVolumeNumber is not in our volumeSummary, it means that
950 * at least one other volume in the VG is on the partition, but the
951 * RW volume is not. We've already AskOffline'd it by now, though,
952 * so make sure we don't still have the volume checked out. */
953 AskDelete(salvinfo, singleVolumeNumber);
956 for (j = 0; j < salvinfo->nVolumes; j++) {
957 if (salvinfo->volumeSummaryp[j].header.id != singleVolumeNumber) {
958 if (!salvinfo->volumeSummaryp[j].deleted) {
959 AskOnline(salvinfo, salvinfo->volumeSummaryp[j].header.id);
965 Log("SALVAGING OF PARTITION %s%s COMPLETED\n",
966 salvinfo->fileSysPartition->name, (Testing ? " (READONLY mode)" : ""));
969 OS_CLOSE(inodeFile); /* SalvageVolumeGroup was the last which needed it. */
973 DeleteExtraVolumeHeaderFile(struct SalvInfo *salvinfo, struct VolumeSummary *vsp)
976 char filename[VMAXPATHLEN];
982 VolumeExternalName_r(vsp->header.id, filename, sizeof(filename));
983 sprintf(path, "%s" OS_DIRSEP "%s", salvinfo->fileSysPath, filename);
986 Log("The volume header file %s is not associated with any actual data (%sdeleted)\n", path, (Testing ? "would have been " : ""));
989 code = VDestroyVolumeDiskHeader(salvinfo->fileSysPartition, vsp->header.id, vsp->header.parent);
991 Log("Error %ld destroying volume disk header for volume %" AFS_VOLID_FMT "\n",
992 afs_printable_int32_ld(code),
993 afs_printable_VolumeId_lu(vsp->header.id));
996 /* make sure we actually delete the header file; ENOENT
997 * is fine, since VDestroyVolumeDiskHeader probably already
999 if (unlink(path) && errno != ENOENT) {
1000 Log("Unable to unlink %s (errno = %d)\n", path, errno);
1002 if (salvinfo->useFSYNC) {
1003 AskDelete(salvinfo, vsp->header.id);
1010 CompareInodes(const void *_p1, const void *_p2)
1012 const struct ViceInodeInfo *p1 = _p1;
1013 const struct ViceInodeInfo *p2 = _p2;
1014 if (p1->u.vnode.vnodeNumber == INODESPECIAL
1015 || p2->u.vnode.vnodeNumber == INODESPECIAL) {
1016 VolumeId p1rwid, p2rwid;
1018 (p1->u.vnode.vnodeNumber ==
1019 INODESPECIAL ? p1->u.special.parentId : p1->u.vnode.volumeId);
1021 (p2->u.vnode.vnodeNumber ==
1022 INODESPECIAL ? p2->u.special.parentId : p2->u.vnode.volumeId);
1023 if (p1rwid < p2rwid)
1025 if (p1rwid > p2rwid)
1027 if (p1->u.vnode.vnodeNumber == INODESPECIAL
1028 && p2->u.vnode.vnodeNumber == INODESPECIAL) {
1029 if (p1->u.vnode.volumeId == p2->u.vnode.volumeId)
1030 return (p1->u.special.type < p2->u.special.type ? -1 : 1);
1031 if (p1->u.vnode.volumeId == p1rwid)
1033 if (p2->u.vnode.volumeId == p2rwid)
1035 return (p1->u.vnode.volumeId < p2->u.vnode.volumeId ? -1 : 1);
1037 if (p1->u.vnode.vnodeNumber != INODESPECIAL)
1038 return (p2->u.vnode.volumeId == p2rwid ? 1 : -1);
1039 return (p1->u.vnode.volumeId == p1rwid ? -1 : 1);
1041 if (p1->u.vnode.volumeId < p2->u.vnode.volumeId)
1043 if (p1->u.vnode.volumeId > p2->u.vnode.volumeId)
1045 if (p1->u.vnode.vnodeNumber < p2->u.vnode.vnodeNumber)
1047 if (p1->u.vnode.vnodeNumber > p2->u.vnode.vnodeNumber)
1049 /* The following tests are reversed, so that the most desirable
1050 * of several similar inodes comes first */
1051 if (p1->u.vnode.vnodeUniquifier > p2->u.vnode.vnodeUniquifier) {
1052 #ifdef AFS_3DISPARES
1053 if (p1->u.vnode.vnodeUniquifier > 3775414 /* 90% of 4.2M */ &&
1054 p2->u.vnode.vnodeUniquifier < 419490 /* 10% of 4.2M */ )
1057 #ifdef AFS_SGI_EXMAG
1058 if (p1->u.vnode.vnodeUniquifier > 15099494 /* 90% of 16M */ &&
1059 p2->u.vnode.vnodeUniquifier < 1677721 /* 10% of 16M */ )
1064 if (p1->u.vnode.vnodeUniquifier < p2->u.vnode.vnodeUniquifier) {
1065 #ifdef AFS_3DISPARES
1066 if (p2->u.vnode.vnodeUniquifier > 3775414 /* 90% of 4.2M */ &&
1067 p1->u.vnode.vnodeUniquifier < 419490 /* 10% of 4.2M */ )
1070 #ifdef AFS_SGI_EXMAG
1071 if (p2->u.vnode.vnodeUniquifier > 15099494 /* 90% of 16M */ &&
1072 p1->u.vnode.vnodeUniquifier < 1677721 /* 10% of 16M */ )
1077 if (p1->u.vnode.inodeDataVersion > p2->u.vnode.inodeDataVersion) {
1078 #ifdef AFS_3DISPARES
1079 if (p1->u.vnode.inodeDataVersion > 1887437 /* 90% of 2.1M */ &&
1080 p2->u.vnode.inodeDataVersion < 209716 /* 10% of 2.1M */ )
1083 #ifdef AFS_SGI_EXMAG
1084 if (p1->u.vnode.inodeDataVersion > 15099494 /* 90% of 16M */ &&
1085 p2->u.vnode.inodeDataVersion < 1677721 /* 10% of 16M */ )
1090 if (p1->u.vnode.inodeDataVersion < p2->u.vnode.inodeDataVersion) {
1091 #ifdef AFS_3DISPARES
1092 if (p2->u.vnode.inodeDataVersion > 1887437 /* 90% of 2.1M */ &&
1093 p1->u.vnode.inodeDataVersion < 209716 /* 10% of 2.1M */ )
1096 #ifdef AFS_SGI_EXMAG
1097 if (p2->u.vnode.inodeDataVersion > 15099494 /* 90% of 16M */ &&
1098 p1->u.vnode.inodeDataVersion < 1677721 /* 10% of 16M */ )
1107 CountVolumeInodes(struct ViceInodeInfo *ip, int maxInodes,
1108 struct InodeSummary *summary)
1110 VolumeId volume = ip->u.vnode.volumeId;
1111 VolumeId rwvolume = volume;
1116 while (maxInodes-- && volume == ip->u.vnode.volumeId) {
1118 if (ip->u.vnode.vnodeNumber == INODESPECIAL) {
1120 rwvolume = ip->u.special.parentId;
1121 /* This isn't quite right, as there could (in error) be different
1122 * parent inodes in different special vnodes */
1124 if (maxunique < ip->u.vnode.vnodeUniquifier)
1125 maxunique = ip->u.vnode.vnodeUniquifier;
1129 summary->volumeId = volume;
1130 summary->RWvolumeId = rwvolume;
1131 summary->nInodes = n;
1132 summary->nSpecialInodes = nSpecial;
1133 summary->maxUniquifier = maxunique;
1137 OnlyOneVolume(struct ViceInodeInfo *inodeinfo, VolumeId singleVolumeNumber, void *rock)
1139 if (inodeinfo->u.vnode.vnodeNumber == INODESPECIAL)
1140 return (inodeinfo->u.special.parentId == singleVolumeNumber);
1141 return (inodeinfo->u.vnode.volumeId == singleVolumeNumber);
1146 * Collect list of inodes in file named by path. If a truly fatal error,
1147 * unlink the file and abort. For lessor errors, return -1. The file will
1148 * be unlinked by the caller.
1151 GetInodeSummary(struct SalvInfo *salvinfo, FD_t inodeFile, VolumeId singleVolumeNumber)
1155 struct ViceInodeInfo *ip, *ip_save;
1156 struct InodeSummary summary;
1157 char summaryFileName[50];
1158 FD_t summaryFile = INVALID_FD;
1160 char *dev = salvinfo->fileSysPath;
1161 char *wpath = salvinfo->fileSysPath;
1163 char *dev = salvinfo->fileSysDeviceName;
1164 char *wpath = salvinfo->filesysfulldev;
1166 char *part = salvinfo->fileSysPath;
1171 afs_sfsize_t st_size;
1173 /* This file used to come from vfsck; cobble it up ourselves now... */
1175 ListViceInodes(dev, salvinfo->fileSysPath, inodeFile,
1176 singleVolumeNumber ? OnlyOneVolume : 0,
1177 singleVolumeNumber, &forceSal, forceR, wpath, NULL)) < 0) {
1179 Log("*** I/O error %d when writing a tmp inode file; Not salvaged %s ***\nIncrease space on partition or use '-tmpdir'\n", errno, dev);
1183 Abort("Unable to get inodes for \"%s\"; not salvaged\n", dev);
1185 if (forceSal && !ForceSalvage) {
1186 Log("***Forced salvage of all volumes on this partition***\n");
1189 OS_SEEK(inodeFile, 0L, SEEK_SET);
1190 salvinfo->inodeFd = inodeFile;
1191 if (salvinfo->inodeFd == INVALID_FD ||
1192 (st_size = OS_SIZE(salvinfo->inodeFd)) == -1) {
1193 Abort("No inode description file for \"%s\"; not salvaged\n", dev);
1195 tdir = (tmpdir ? tmpdir : part);
1197 (void)_putenv("TMP="); /* If "TMP" is set, then that overrides tdir. */
1198 (void)strcpy(summaryFileName, _tempnam(tdir, "salvage.temp."));
1200 snprintf(summaryFileName, sizeof summaryFileName,
1201 "%s" OS_DIRSEP "salvage.temp.%d", tdir, getpid());
1203 summaryFile = OS_OPEN(summaryFileName, O_RDWR|O_APPEND|O_CREAT, 0666);
1204 if (summaryFile == INVALID_FD) {
1205 Abort("Unable to create inode summary file\n");
1209 /* Using nt_unlink here since we're really using the delete on close
1210 * semantics of unlink. In most places in the salvager, we really do
1211 * mean to unlink the file at that point. Those places have been
1212 * modified to actually do that so that the NT crt can be used there.
1214 * jaltman - As commented elsewhere, this cannot work because fopen()
1215 * does not open files with DELETE and FILE_SHARE_DELETE.
1217 code = nt_unlink(summaryFileName);
1219 code = unlink(summaryFileName);
1222 Log("Error %d when trying to unlink %s\n", errno, summaryFileName);
1225 if (!canfork || debug || Fork() == 0) {
1226 int nInodes = st_size / sizeof(struct ViceInodeInfo);
1228 OS_CLOSE(summaryFile);
1229 if (!singleVolumeNumber) /* Remove the FORCESALVAGE file */
1230 RemoveTheForce(salvinfo->fileSysPath);
1232 struct VolumeSummary *vsp;
1236 GetVolumeSummary(salvinfo, singleVolumeNumber);
1238 for (i = 0, vsp = salvinfo->volumeSummaryp; i < salvinfo->nVolumes; i++) {
1240 if (vsp->header.id == singleVolumeNumber) {
1243 DeleteExtraVolumeHeaderFile(salvinfo, vsp);
1249 MaybeAskOnline(salvinfo, singleVolumeNumber);
1251 /* make sure we get rid of stray .vol headers, even if
1252 * they're not in our volume summary (might happen if
1253 * e.g. something else created them and they're not in the
1254 * fileserver VGC) */
1255 VDestroyVolumeDiskHeader(salvinfo->fileSysPartition,
1256 singleVolumeNumber, 0 /*parent*/);
1257 AskDelete(salvinfo, singleVolumeNumber);
1261 Log("%s vice inodes on %s; not salvaged\n",
1262 singleVolumeNumber ? "No applicable" : "No", dev);
1267 ip = malloc(nInodes*sizeof(struct ViceInodeInfo));
1269 OS_CLOSE(summaryFile);
1271 ("Unable to allocate enough space to read inode table; %s not salvaged\n",
1274 if (OS_READ(salvinfo->inodeFd, ip, st_size) != st_size) {
1275 OS_CLOSE(summaryFile);
1276 Abort("Unable to read inode table; %s not salvaged\n", dev);
1278 qsort(ip, nInodes, sizeof(struct ViceInodeInfo), CompareInodes);
1279 if (OS_SEEK(salvinfo->inodeFd, 0, SEEK_SET) == -1
1280 || OS_WRITE(salvinfo->inodeFd, ip, st_size) != st_size) {
1281 OS_CLOSE(summaryFile);
1282 Abort("Unable to rewrite inode table; %s not salvaged\n", dev);
1287 CountVolumeInodes(ip, nInodes, &summary);
1288 if (OS_WRITE(summaryFile, &summary, sizeof(summary)) != sizeof(summary)) {
1289 Log("Difficulty writing summary file (errno = %d); %s not salvaged\n", errno, dev);
1290 OS_CLOSE(summaryFile);
1294 summary.index += (summary.nInodes);
1295 nInodes -= summary.nInodes;
1296 ip += summary.nInodes;
1299 ip = ip_save = NULL;
1300 /* Following fflush is not fclose, because if it was debug mode would not work */
1301 if (OS_SYNC(summaryFile) == -1) {
1302 Log("Unable to write summary file (errno = %d); %s not salvaged\n", errno, dev);
1303 OS_CLOSE(summaryFile);
1307 if (canfork && !debug) {
1311 if (Wait("Inode summary") == -1) {
1312 OS_CLOSE(summaryFile);
1313 Exit(1); /* salvage of this partition aborted */
1317 st_size = OS_SIZE(summaryFile);
1318 opr_Assert(st_size >= 0);
1321 salvinfo->inodeSummary = malloc(st_size);
1322 opr_Assert(salvinfo->inodeSummary != NULL);
1323 /* For GNU we need to do lseek to get the file pointer moved. */
1324 opr_Assert(OS_SEEK(summaryFile, 0, SEEK_SET) == 0);
1325 ret = OS_READ(summaryFile, salvinfo->inodeSummary, st_size);
1326 opr_Assert(ret == st_size);
1328 salvinfo->nVolumesInInodeFile = st_size / sizeof(struct InodeSummary);
1329 for (i = 0; i < salvinfo->nVolumesInInodeFile; i++) {
1330 salvinfo->inodeSummary[i].volSummary = NULL;
1332 Log("%d nVolumesInInodeFile %lu \n",salvinfo->nVolumesInInodeFile,(unsigned long)st_size);
1333 OS_CLOSE(summaryFile);
1336 if (retcode && singleVolumeNumber && !deleted) {
1337 AskError(salvinfo, singleVolumeNumber);
1343 /* Comparison routine for volume sort.
1344 This is setup so that a read-write volume comes immediately before
1345 any read-only clones of that volume */
1347 CompareVolumes(const void *_p1, const void *_p2)
1349 const struct VolumeSummary *p1 = _p1;
1350 const struct VolumeSummary *p2 = _p2;
1351 if (p1->header.parent != p2->header.parent)
1352 return p1->header.parent < p2->header.parent ? -1 : 1;
1353 if (p1->header.id == p1->header.parent) /* p1 is rw volume */
1355 if (p2->header.id == p2->header.parent) /* p2 is rw volume */
1357 return p1->header.id < p2->header.id ? -1 : 1; /* Both read-only */
1361 * Gleans volumeSummary information by asking the fileserver
1363 * @param[in] singleVolumeNumber the volume we're salvaging. 0 if we're
1364 * salvaging a whole partition
1366 * @return whether we obtained the volume summary information or not
1367 * @retval 0 success; we obtained the volume summary information
1368 * @retval -1 we raced with a fileserver restart; volume locks and checkout
1370 * @retval 1 we did not get the volume summary information; either the
1371 * fileserver responded with an error, or we are not supposed to
1372 * ask the fileserver for the information (e.g. we are salvaging
1373 * the entire partition or we are not the salvageserver)
1375 * @note for non-DAFS, always returns 1
1378 AskVolumeSummary(struct SalvInfo *salvinfo, VolumeId singleVolumeNumber)
1381 #if defined(FSSYNC_BUILD_CLIENT) && defined(AFS_DEMAND_ATTACH_FS)
1382 if (programType == salvageServer) {
1383 if (singleVolumeNumber) {
1384 FSSYNC_VGQry_response_t q_res;
1386 struct VolumeSummary *vsp;
1388 struct VolumeDiskHeader diskHdr;
1390 memset(&res, 0, sizeof(res));
1392 code = FSYNC_VGCQuery(salvinfo->fileSysPartition->name, singleVolumeNumber, &q_res, &res);
1395 * We must wait for the partition to finish scanning before
1396 * can continue, since we will not know if we got the entire
1397 * VG membership unless the partition is fully scanned.
1398 * We could, in theory, just scan the partition ourselves if
1399 * the VG cache is not ready, but we would be doing the exact
1400 * same scan the fileserver is doing; it will almost always
1401 * be faster to wait for the fileserver. The only exceptions
1402 * are if the partition does not take very long to scan, and
1403 * in that case it's fast either way, so who cares?
1405 if (code == SYNC_FAILED && res.hdr.reason == FSYNC_PART_SCANNING) {
1406 Log("waiting for fileserver to finish scanning partition %s...\n",
1407 salvinfo->fileSysPartition->name);
1409 for (i = 1; code == SYNC_FAILED && res.hdr.reason == FSYNC_PART_SCANNING; i++) {
1410 /* linearly ramp up from 1 to 10 seconds; nothing fancy,
1411 * just so small partitions don't need to wait over 10
1412 * seconds every time, and large partitions are generally
1413 * polled only once every ten seconds. */
1414 sleep((i > 10) ? (i = 10) : i);
1416 code = FSYNC_VGCQuery(salvinfo->fileSysPartition->name, singleVolumeNumber, &q_res, &res);
1420 if (code == SYNC_FAILED && res.hdr.reason == FSYNC_UNKNOWN_VOLID) {
1421 /* This can happen if there's no header for the volume
1422 * we're salvaging, or no headers exist for the VG (if
1423 * we're salvaging an RW). Act as if we got a response
1424 * with no VG members. The headers may be created during
1425 * salvaging, if there are inodes in this VG. */
1427 memset(&q_res, 0, sizeof(q_res));
1428 q_res.rw = singleVolumeNumber;
1432 Log("fileserver refused VGCQuery request for volume %" AFS_VOLID_FMT " on "
1433 "partition %s, code %ld reason %ld\n",
1434 afs_printable_VolumeId_lu(singleVolumeNumber),
1435 salvinfo->fileSysPartition->name,
1436 afs_printable_int32_ld(code),
1437 afs_printable_int32_ld(res.hdr.reason));
1441 if (q_res.rw != singleVolumeNumber) {
1442 Log("fileserver requested salvage of clone %" AFS_VOLID_FMT "; scheduling salvage of volume group %" AFS_VOLID_FMT "...\n",
1443 afs_printable_VolumeId_lu(singleVolumeNumber),
1444 afs_printable_VolumeId_lu(q_res.rw));
1445 #ifdef SALVSYNC_BUILD_CLIENT
1446 if (SALVSYNC_LinkVolume(q_res.rw,
1448 salvinfo->fileSysPartition->name,
1450 Log("schedule request failed\n");
1452 #endif /* SALVSYNC_BUILD_CLIENT */
1453 Exit(SALSRV_EXIT_VOLGROUP_LINK);
1456 salvinfo->volumeSummaryp = calloc(VOL_VG_MAX_VOLS, sizeof(struct VolumeSummary));
1457 opr_Assert(salvinfo->volumeSummaryp != NULL);
1459 salvinfo->nVolumes = 0;
1460 vsp = salvinfo->volumeSummaryp;
1462 for (i = 0; i < VOL_VG_MAX_VOLS; i++) {
1463 char name[VMAXPATHLEN];
1465 if (!q_res.children[i]) {
1469 /* AskOffline for singleVolumeNumber was called much earlier */
1470 if (q_res.children[i] != singleVolumeNumber) {
1471 AskOffline(salvinfo, q_res.children[i]);
1472 if (LockVolume(salvinfo, q_res.children[i])) {
1478 code = VReadVolumeDiskHeader(q_res.children[i], salvinfo->fileSysPartition, &diskHdr);
1480 Log("Cannot read header for %lu; trying to salvage group anyway\n",
1481 afs_printable_uint32_lu(q_res.children[i]));
1486 DiskToVolumeHeader(&vsp->header, &diskHdr);
1487 VolumeExternalName_r(q_res.children[i], name, sizeof(name));
1489 salvinfo->nVolumes++;
1493 qsort(salvinfo->volumeSummaryp, salvinfo->nVolumes, sizeof(struct VolumeSummary),
1498 Log("Cannot get volume summary from fileserver; falling back to scanning "
1499 "entire partition\n");
1502 #endif /* FSSYNC_BUILD_CLIENT && AFS_DEMAND_ATTACH_FS */
1507 * count how many volume headers are found by VWalkVolumeHeaders.
1509 * @param[in] dp the disk partition (unused)
1510 * @param[in] name full path to the .vol header (unused)
1511 * @param[in] hdr the header data (unused)
1512 * @param[in] last whether this is the last try or not (unused)
1513 * @param[in] rock actually an afs_int32*; the running count of how many
1514 * volumes we have found
1519 CountHeader(struct DiskPartition64 *dp, const char *name,
1520 struct VolumeDiskHeader *hdr, int last, void *rock)
1522 afs_int32 *nvols = (afs_int32 *)rock;
1528 * parameters to pass to the VWalkVolumeHeaders callbacks when recording volume
1531 struct SalvageScanParams {
1532 VolumeId singleVolumeNumber; /**< 0 for a partition-salvage, otherwise the
1533 * vol id of the VG we're salvaging */
1534 struct VolumeSummary *vsp; /**< ptr to the current volume summary object
1535 * we're filling in */
1536 afs_int32 nVolumes; /**< # of vols we've encountered */
1537 afs_int32 totalVolumes; /**< max # of vols we should encounter (the
1538 * # of vols we've alloc'd memory for) */
1539 int retry; /**< do we need to retry vol lock/checkout? */
1540 struct SalvInfo *salvinfo; /**< salvage job info */
1544 * records volume summary info found from VWalkVolumeHeaders.
1546 * Found volumes are also taken offline if they are in the specific volume
1547 * group we are looking for.
1549 * @param[in] dp the disk partition
1550 * @param[in] name full path to the .vol header
1551 * @param[in] hdr the header data
1552 * @param[in] last 1 if this is the last try to read the header, 0 otherwise
1553 * @param[in] rock actually a struct SalvageScanParams*, containing the
1554 * information needed to record the volume summary data
1556 * @return operation status
1558 * @retval -1 volume locking raced with fileserver restart; checking out
1559 * and locking volumes needs to be retried
1560 * @retval 1 volume header is mis-named and should be deleted
1563 RecordHeader(struct DiskPartition64 *dp, const char *name,
1564 struct VolumeDiskHeader *hdr, int last, void *rock)
1566 char nameShouldBe[64];
1567 struct SalvageScanParams *params;
1568 struct VolumeSummary summary;
1569 VolumeId singleVolumeNumber;
1570 struct SalvInfo *salvinfo;
1572 params = (struct SalvageScanParams *)rock;
1574 memset(&summary, 0, sizeof(summary));
1576 singleVolumeNumber = params->singleVolumeNumber;
1577 salvinfo = params->salvinfo;
1579 DiskToVolumeHeader(&summary.header, hdr);
1581 if (singleVolumeNumber && summary.header.id == singleVolumeNumber
1582 && summary.header.parent != singleVolumeNumber) {
1584 if (programType == salvageServer) {
1585 #ifdef SALVSYNC_BUILD_CLIENT
1586 Log("fileserver requested salvage of clone %" AFS_VOLID_FMT "; scheduling salvage of volume group %" AFS_VOLID_FMT "...\n",
1587 afs_printable_VolumeId_lu(summary.header.id),
1588 afs_printable_VolumeId_lu(summary.header.parent));
1589 if (SALVSYNC_LinkVolume(summary.header.parent,
1593 Log("schedule request failed\n");
1596 Exit(SALSRV_EXIT_VOLGROUP_LINK);
1599 Log("%" AFS_VOLID_FMT " is a read-only volume; not salvaged\n",
1600 afs_printable_VolumeId_lu(singleVolumeNumber));
1605 if (!singleVolumeNumber || summary.header.id == singleVolumeNumber
1606 || summary.header.parent == singleVolumeNumber) {
1608 /* check if the header file is incorrectly named */
1610 const char *base = strrchr(name, OS_DIRSEPC);
1617 snprintf(nameShouldBe, sizeof nameShouldBe,
1618 VFORMAT, afs_printable_VolumeId_lu(summary.header.id));
1621 if (strcmp(nameShouldBe, base)) {
1622 /* .vol file has wrong name; retry/delete */
1626 if (!badname || last) {
1627 /* only offline the volume if the header is good, or if this is
1628 * the last try looking at it; avoid AskOffline'ing the same vol
1631 if (singleVolumeNumber
1632 && summary.header.id != singleVolumeNumber) {
1633 /* don't offline singleVolumeNumber; we already did that
1636 AskOffline(salvinfo, summary.header.id);
1638 #ifdef AFS_DEMAND_ATTACH_FS
1640 /* don't lock the volume if the header is bad, since we're
1641 * about to delete it anyway. */
1642 if (LockVolume(salvinfo, summary.header.id)) {
1647 #endif /* AFS_DEMAND_ATTACH_FS */
1651 if (last && !Showmode) {
1652 Log("Volume header file %s is incorrectly named (should be %s "
1653 "not %s); %sdeleted (it will be recreated later, if "
1654 "necessary)\n", name, nameShouldBe, base,
1655 (Testing ? "it would have been " : ""));
1663 if (params->nVolumes > params->totalVolumes) {
1664 /* We found more volumes than we found on the first partition walk;
1665 * apparently something created a volume while we were
1666 * partition-salvaging, or we found more than 20 vols when salvaging a
1667 * particular volume. Abort if we detect this, since other programs
1668 * supposed to not touch the partition while it is partition-salvaging,
1669 * and we shouldn't find more than 20 vols in a VG.
1671 Abort("Found %ld vol headers, but should have found at most %ld! "
1672 "Make sure the volserver/fileserver are not running at the "
1673 "same time as a partition salvage\n",
1674 afs_printable_int32_ld(params->nVolumes),
1675 afs_printable_int32_ld(params->totalVolumes));
1678 memcpy(params->vsp, &summary, sizeof(summary));
1686 * possibly unlinks bad volume headers found from VWalkVolumeHeaders.
1688 * If the header could not be read in at all, the header is always unlinked.
1689 * If instead RecordHeader said the header was bad (that is, the header file
1690 * is mis-named), we only unlink if we are doing a partition salvage, as
1691 * opposed to salvaging a specific volume group.
1693 * @param[in] dp the disk partition
1694 * @param[in] name full path to the .vol header
1695 * @param[in] hdr header data, or NULL if the header could not be read
1696 * @param[in] rock actually a struct SalvageScanParams*, with some information
1700 UnlinkHeader(struct DiskPartition64 *dp, const char *name,
1701 struct VolumeDiskHeader *hdr, void *rock)
1703 struct SalvageScanParams *params;
1706 params = (struct SalvageScanParams *)rock;
1709 /* no header; header is too bogus to read in at all */
1711 Log("%s is not a legitimate volume header file; %sdeleted\n", name, (Testing ? "it would have been " : ""));
1717 } else if (!params->singleVolumeNumber) {
1718 /* We were able to read in a header, but RecordHeader said something
1719 * was wrong with it. We only unlink those if we are doing a partition
1726 if (dounlink && unlink(name)) {
1727 Log("Error %d while trying to unlink %s\n", errno, name);
1732 * Populates salvinfo->volumeSummaryp with volume summary information, either by asking
1733 * the fileserver for VG information, or by scanning the /vicepX partition.
1735 * @param[in] singleVolumeNumber the volume ID of the single volume group we
1736 * are salvaging, or 0 if this is a partition
1739 * @return operation status
1741 * @retval -1 we raced with a fileserver restart; checking out and locking
1742 * volumes must be retried
1745 GetVolumeSummary(struct SalvInfo *salvinfo, VolumeId singleVolumeNumber)
1747 afs_int32 nvols = 0;
1748 struct SalvageScanParams params;
1751 code = AskVolumeSummary(salvinfo, singleVolumeNumber);
1753 /* we successfully got the vol information from the fileserver; no
1754 * need to scan the partition */
1758 /* we need to retry volume checkout */
1762 if (!singleVolumeNumber) {
1763 /* Count how many volumes we have in /vicepX */
1764 code = VWalkVolumeHeaders(salvinfo->fileSysPartition, salvinfo->fileSysPath, CountHeader,
1767 Abort("Can't read directory %s; not salvaged\n", salvinfo->fileSysPath);
1772 nvols = VOL_VG_MAX_VOLS;
1775 salvinfo->volumeSummaryp = calloc(nvols, sizeof(struct VolumeSummary));
1776 opr_Assert(salvinfo->volumeSummaryp != NULL);
1778 params.singleVolumeNumber = singleVolumeNumber;
1779 params.vsp = salvinfo->volumeSummaryp;
1780 params.nVolumes = 0;
1781 params.totalVolumes = nvols;
1783 params.salvinfo = salvinfo;
1785 /* walk the partition directory of volume headers and record the info
1786 * about them; unlinking invalid headers */
1787 code = VWalkVolumeHeaders(salvinfo->fileSysPartition, salvinfo->fileSysPath, RecordHeader,
1788 UnlinkHeader, ¶ms);
1790 /* we apparently need to retry checking-out/locking volumes */
1794 Abort("Failed to get volume header summary\n");
1796 salvinfo->nVolumes = params.nVolumes;
1798 qsort(salvinfo->volumeSummaryp, salvinfo->nVolumes, sizeof(struct VolumeSummary),
1804 #ifdef AFS_NAMEI_ENV
1805 /* Find the link table. This should be associated with the RW volume, even
1806 * if there is only an RO volume at this site.
1809 FindLinkHandle(struct InodeSummary *isp, int nVols,
1810 struct ViceInodeInfo *allInodes)
1813 struct ViceInodeInfo *ip;
1815 for (i = 0; i < nVols; i++) {
1816 ip = allInodes + isp[i].index;
1817 for (j = 0; j < isp[i].nSpecialInodes; j++) {
1818 if (ip[j].u.special.volumeId == isp->RWvolumeId &&
1819 ip[j].u.special.parentId == isp->RWvolumeId &&
1820 ip[j].u.special.type == VI_LINKTABLE) {
1821 return ip[j].inodeNumber;
1829 CheckDupLinktable(struct SalvInfo *salvinfo, struct InodeSummary *isp, struct ViceInodeInfo *ip)
1832 if (ip->u.vnode.vnodeNumber != INODESPECIAL) {
1833 /* not a linktable; process as a normal file */
1836 if (ip->u.special.type != VI_LINKTABLE) {
1837 /* not a linktable; process as a normal file */
1841 /* make sure nothing inc/decs it */
1844 if (ip->u.special.volumeId == ip->u.special.parentId) {
1845 /* This is a little weird, but shouldn't break anything, and there is
1846 * no known way that this can happen; just do nothing, in case deleting
1847 * it would screw something up. */
1848 Log("Inode %s appears to be a valid linktable for id (%u), but it's not\n",
1849 PrintInode(stmp, ip->inodeNumber), ip->u.special.parentId);
1850 Log("the linktable for our volume group (%u). This is unusual, since\n",
1852 Log("there should only be one linktable per volume group. I'm leaving\n");
1853 Log("it alone, just to be safe.\n");
1857 Log("Linktable %s appears to be invalid (parentid/volumeid mismatch: %u != %u)\n",
1858 PrintInode(stmp, ip->inodeNumber), ip->u.special.parentId, ip->u.special.volumeId);
1860 Log("Would have deleted linktable inode %s\n", PrintInode(stmp, ip->inodeNumber));
1865 Log("Deleting linktable inode %s\n", PrintInode(stmp, ip->inodeNumber));
1866 IH_INIT(tmpH, salvinfo->fileSysDevice, isp->RWvolumeId, ip->inodeNumber);
1867 namei_HandleToName(&ufs_name, tmpH);
1868 if (unlink(ufs_name.n_path) < 0) {
1869 Log("Error %d unlinking path %s\n", errno, ufs_name.n_path);
1878 CreateLinkTable(struct SalvInfo *salvinfo, struct InodeSummary *isp, Inode ino)
1880 struct versionStamp version;
1883 if (!VALID_INO(ino))
1885 IH_CREATE(NULL, salvinfo->fileSysDevice, salvinfo->fileSysPath, 0, isp->RWvolumeId,
1886 INODESPECIAL, VI_LINKTABLE, isp->RWvolumeId);
1887 if (!VALID_INO(ino))
1889 ("Unable to allocate link table inode for volume %" AFS_VOLID_FMT " (error = %d)\n",
1890 afs_printable_VolumeId_lu(isp->RWvolumeId), errno);
1891 IH_INIT(salvinfo->VGLinkH, salvinfo->fileSysDevice, isp->RWvolumeId, ino);
1892 fdP = IH_OPEN(salvinfo->VGLinkH);
1894 Abort("Can't open link table for volume %" AFS_VOLID_FMT " (error = %d)\n",
1895 afs_printable_VolumeId_lu(isp->RWvolumeId), errno);
1897 if (FDH_TRUNC(fdP, sizeof(version) + sizeof(short)) < 0)
1898 Abort("Can't truncate link table for volume %" AFS_VOLID_FMT " (error = %d)\n",
1899 afs_printable_VolumeId_lu(isp->RWvolumeId), errno);
1901 version.magic = LINKTABLEMAGIC;
1902 version.version = LINKTABLEVERSION;
1904 if (FDH_PWRITE(fdP, (char *)&version, sizeof(version), 0)
1906 Abort("Can't truncate link table for volume %" AFS_VOLID_FMT " (error = %d)\n",
1907 afs_printable_VolumeId_lu(isp->RWvolumeId), errno);
1909 FDH_REALLYCLOSE(fdP);
1911 /* If the volume summary exits (i.e., the V*.vol header file exists),
1912 * then set this inode there as well.
1914 if (isp->volSummary)
1915 isp->volSummary->header.linkTable = ino;
1924 SVGParms_t *parms = (SVGParms_t *) arg;
1925 DoSalvageVolumeGroup(parms->svgp_salvinfo, parms->svgp_inodeSummaryp, parms->svgp_count);
1930 nt_SalvageVolumeGroup(struct SalvInfo *salvinfo, struct InodeSummary *isp, int nVols)
1933 pthread_attr_t tattr;
1937 /* Initialize per volume global variables, even if later code does so */
1938 salvinfo->VolumeChanged = 0;
1939 salvinfo->VGLinkH = NULL;
1940 salvinfo->VGLinkH_cnt = 0;
1941 memset(&salvinfo->VolInfo, 0, sizeof(salvinfo->VolInfo));
1943 parms.svgp_inodeSummaryp = isp;
1944 parms.svgp_count = nVols;
1945 parms.svgp_salvinfo = salvinfo;
1946 code = pthread_attr_init(&tattr);
1948 Log("Failed to salvage volume group %u: pthread_attr_init()\n",
1952 code = pthread_attr_setdetachstate(&tattr, PTHREAD_CREATE_JOINABLE);
1954 Log("Failed to salvage volume group %u: pthread_attr_setdetachstate()\n", isp->RWvolumeId);
1957 code = pthread_create(&tid, &tattr, nt_SVG, &parms);
1959 Log("Failed to create thread to salvage volume group %u\n",
1963 (void)pthread_join(tid, NULL);
1965 #endif /* AFS_NT40_ENV */
1968 DoSalvageVolumeGroup(struct SalvInfo *salvinfo, struct InodeSummary *isp, int nVols)
1970 struct ViceInodeInfo *inodes, *allInodes, *ip;
1971 int i, totalInodes, size, salvageTo;
1975 int dec_VGLinkH = 0;
1977 FdHandle_t *fdP = NULL;
1979 salvinfo->VGLinkH_cnt = 0;
1980 haveRWvolume = (isp->volumeId == isp->RWvolumeId
1981 && isp->nSpecialInodes > 0);
1982 if ((!ShowMounts) || (ShowMounts && !haveRWvolume)) {
1983 if (!ForceSalvage && QuickCheck(salvinfo, isp, nVols))
1986 if (ShowMounts && !haveRWvolume)
1988 if (canfork && !debug && Fork() != 0) {
1989 (void)Wait("Salvage volume group");
1992 for (i = 0, totalInodes = 0; i < nVols; i++)
1993 totalInodes += isp[i].nInodes;
1994 size = totalInodes * sizeof(struct ViceInodeInfo);
1995 inodes = malloc(size);
1996 allInodes = inodes - isp->index; /* this would the base of all the inodes
1997 * for the partition, if all the inodes
1998 * had been read into memory */
2000 (salvinfo->inodeFd, isp->index * sizeof(struct ViceInodeInfo),
2002 opr_Verify(OS_READ(salvinfo->inodeFd, inodes, size) == size);
2004 /* Don't try to salvage a read write volume if there isn't one on this
2006 salvageTo = haveRWvolume ? 0 : 1;
2008 #ifdef AFS_NAMEI_ENV
2009 ino = FindLinkHandle(isp, nVols, allInodes);
2010 if (VALID_INO(ino)) {
2011 IH_INIT(salvinfo->VGLinkH, salvinfo->fileSysDevice, isp->RWvolumeId, ino);
2012 fdP = IH_OPEN(salvinfo->VGLinkH);
2014 if (VALID_INO(ino) && fdP != NULL) {
2015 struct versionStamp header;
2016 afs_sfsize_t nBytes;
2018 nBytes = FDH_PREAD(fdP, (char *)&header, sizeof(struct versionStamp), 0);
2019 if (nBytes != sizeof(struct versionStamp)
2020 || header.magic != LINKTABLEMAGIC) {
2021 Log("Bad linktable header for volume %" AFS_VOLID_FMT ".\n", afs_printable_VolumeId_lu(isp->RWvolumeId));
2022 FDH_REALLYCLOSE(fdP);
2026 if (!VALID_INO(ino) || fdP == NULL) {
2027 Log("%s link table for volume %" AFS_VOLID_FMT ".\n",
2028 Testing ? "Would have recreated" : "Recreating", afs_printable_VolumeId_lu(isp->RWvolumeId));
2030 IH_INIT(salvinfo->VGLinkH, salvinfo->fileSysDevice, -1, -1);
2033 struct ViceInodeInfo *ip;
2034 CreateLinkTable(salvinfo, isp, ino);
2035 fdP = IH_OPEN(salvinfo->VGLinkH);
2036 /* Sync fake 1 link counts to the link table, now that it exists */
2038 for (i = 0; i < nVols; i++) {
2039 ip = allInodes + isp[i].index;
2040 for (j = isp[i].nSpecialInodes; j < isp[i].nInodes; j++) {
2041 namei_SetLinkCount(fdP, ip[j].inodeNumber, 1, 0);
2042 ip[j].linkCount = 1;
2049 FDH_REALLYCLOSE(fdP);
2051 IH_INIT(salvinfo->VGLinkH, salvinfo->fileSysDevice, -1, -1);
2054 /* Salvage in reverse order--read/write volume last; this way any
2055 * Inodes not referenced by the time we salvage the read/write volume
2056 * can be picked up by the read/write volume */
2057 /* ACTUALLY, that's not done right now--the inodes just vanish */
2058 for (i = nVols - 1; i >= salvageTo; i--) {
2060 struct InodeSummary *lisp = &isp[i];
2061 #ifdef AFS_NAMEI_ENV
2062 if (rw && (nVols > 1 || isp[i].nSpecialInodes == isp[i].nInodes)) {
2063 /* If nVols > 1, we have more than one vol in this volgroup, so
2064 * the RW inodes we detected may just be for the linktable, and
2065 * there is no actual RW volume.
2067 * Additionally, if we only have linktable inodes (no other
2068 * special inodes, no data inodes), there is also no actual RW
2069 * volume to salvage; this is just cruft left behind by something
2070 * else. In that case nVols will only be 1, though, so also
2071 * perform this linktables-only check if we don't have any
2072 * non-special inodes. */
2074 int all_linktables = 1;
2075 for (inode_i = 0; inode_i < isp[i].nSpecialInodes; inode_i++) {
2076 if (inodes[inode_i].u.special.type != VI_LINKTABLE) {
2081 if (all_linktables) {
2082 /* All we have are linktable special inodes, so skip salvaging
2083 * the RW; there was never an RW volume here. If we don't do
2084 * this, we risk creating a new "phantom" RW that the VLDB
2085 * doesn't know about, which is confusing and can cause
2093 Log("%s VOLUME %" AFS_VOLID_FMT "%s.\n", rw ? "SALVAGING" : "CHECKING CLONED",
2094 afs_printable_VolumeId_lu(lisp->volumeId), (Testing ? "(READONLY mode)" : ""));
2095 /* Check inodes twice. The second time do things seriously. This
2096 * way the whole RO volume can be deleted, below, if anything goes wrong */
2097 for (check = 1; check >= 0; check--) {
2099 if (SalvageVolumeHeaderFile(salvinfo, lisp, allInodes, rw, check, &deleteMe)
2101 MaybeZapVolume(salvinfo, lisp, "Volume header", deleteMe, check);
2102 if (rw && deleteMe) {
2103 haveRWvolume = 0; /* This will cause its inodes to be deleted--since salvage
2104 * volume won't be called */
2110 if (rw && check == 1)
2112 if (SalvageVnodes(salvinfo, isp, lisp, allInodes, check) == -1) {
2113 MaybeZapVolume(salvinfo, lisp, "Vnode index", 0, check);
2119 /* Fix actual inode counts */
2122 Log("totalInodes %d\n",totalInodes);
2123 for (ip = inodes; totalInodes; ip++, totalInodes--) {
2124 static int TraceBadLinkCounts = 0;
2125 #ifdef AFS_NAMEI_ENV
2126 if (salvinfo->VGLinkH->ih_ino == ip->inodeNumber) {
2127 dec_VGLinkH = ip->linkCount - salvinfo->VGLinkH_cnt;
2128 VGLinkH_p1 = ip->u.param[0];
2129 continue; /* Deal with this last. */
2130 } else if (CheckDupLinktable(salvinfo, isp, ip)) {
2131 /* Don't touch this inode; CheckDupLinktable has handled it */
2135 if (ip->linkCount != 0 && TraceBadLinkCounts) {
2136 TraceBadLinkCounts--; /* Limit reports, per volume */
2137 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]); /* VolumeId in param */
2140 /* If ip->linkCount is non-zero at this point, then the linkcount
2141 * for the inode on disk is wrong. Initially linkCount is set to
2142 * the actual link count of the inode on disk, and then we (the
2143 * salvager) decrement it for every reference to that inode that we
2144 * find. So if linkCount is still positive by this point, it means
2145 * that the linkcount on disk is too high, so we should DEC the
2146 * inode. If linkCount is negative, it means the linkcount is too
2147 * low, so we should INC the inode.
2149 * If we get an error while INC'ing or DEC'ing, that's a little
2150 * odd and indicates a bug, but try to continue anyway, so the
2151 * volume may still be made accessible. */
2152 while (ip->linkCount > 0) {
2154 if (IH_DEC(salvinfo->VGLinkH, ip->inodeNumber, ip->u.param[0])) {
2155 Log("idec failed. inode %s errno %d\n",
2156 PrintInode(stmp, ip->inodeNumber), errno);
2162 while (ip->linkCount < 0) {
2164 if (IH_INC(salvinfo->VGLinkH, ip->inodeNumber, ip->u.param[0])) {
2165 Log("iinc failed. inode %s errno %d\n",
2166 PrintInode(stmp, ip->inodeNumber), errno);
2173 #ifdef AFS_NAMEI_ENV
2174 while (dec_VGLinkH > 0) {
2175 if (IH_DEC(salvinfo->VGLinkH, salvinfo->VGLinkH->ih_ino, VGLinkH_p1) < 0) {
2176 Log("idec failed on link table, errno = %d\n", errno);
2180 while (dec_VGLinkH < 0) {
2181 if (IH_INC(salvinfo->VGLinkH, salvinfo->VGLinkH->ih_ino, VGLinkH_p1) < 0) {
2182 Log("iinc failed on link table, errno = %d\n", errno);
2189 /* Directory consistency checks on the rw volume */
2191 SalvageVolume(salvinfo, isp, salvinfo->VGLinkH);
2192 IH_RELEASE(salvinfo->VGLinkH);
2194 if (canfork && !debug) {
2200 QuickCheck(struct SalvInfo *salvinfo, struct InodeSummary *isp, int nVols)
2202 /* Check headers BEFORE forking */
2206 for (i = 0; i < nVols; i++) {
2207 struct VolumeSummary *vs = isp[i].volSummary;
2208 VolumeDiskData volHeader;
2210 /* Don't salvage just because phantom rw volume is there... */
2211 /* (If a read-only volume exists, read/write inodes must also exist) */
2212 if (i == 0 && isp->nSpecialInodes == 0 && nVols > 1)
2216 IH_INIT(h, salvinfo->fileSysDevice, vs->header.parent, vs->header.volumeInfo);
2217 if (IH_IREAD(h, 0, (char *)&volHeader, sizeof(volHeader))
2218 == sizeof(volHeader)
2219 && volHeader.stamp.magic == VOLUMEINFOMAGIC
2220 && volHeader.dontSalvage == DONT_SALVAGE
2221 && volHeader.needsSalvaged == 0 && volHeader.destroyMe == 0) {
2222 if (volHeader.inUse != 0) {
2223 volHeader.inUse = 0;
2224 volHeader.inService = 1;
2226 if (IH_IWRITE(h, 0, (char *)&volHeader, sizeof(volHeader))
2227 != sizeof(volHeader)) {
2243 /* SalvageVolumeHeaderFile
2245 * Salvage the top level V*.vol header file. Make sure the special files
2246 * exist and that there are no duplicates.
2248 * Calls SalvageHeader for each possible type of volume special file.
2252 SalvageVolumeHeaderFile(struct SalvInfo *salvinfo, struct InodeSummary *isp,
2253 struct ViceInodeInfo *inodes, int RW,
2254 int check, int *deleteMe)
2257 struct ViceInodeInfo *ip;
2258 int allinodesobsolete = 1;
2259 struct VolumeDiskHeader diskHeader;
2260 afs_int32 (*writefunc)(VolumeDiskHeader_t *, struct DiskPartition64 *) = NULL;
2262 struct VolumeHeader tempHeader;
2263 struct afs_inode_info stuff[MAXINODETYPE];
2265 /* keeps track of special inodes that are probably 'good'; they are
2266 * referenced in the vol header, and are included in the given inodes
2271 } goodspecial[MAXINODETYPE];
2276 memset(goodspecial, 0, sizeof(goodspecial));
2278 skip = calloc(isp->nSpecialInodes, sizeof(*skip));
2280 Log("cannot allocate memory for inode skip array when salvaging "
2281 "volume %lu; not performing duplicate special inode recovery\n",
2282 afs_printable_uint32_lu(isp->volumeId));
2283 /* still try to perform the salvage; the skip array only does anything
2284 * if we detect duplicate special inodes */
2287 init_inode_info(&tempHeader, stuff);
2290 * First, look at the special inodes and see if any are referenced by
2291 * the existing volume header. If we find duplicate special inodes, we
2292 * can use this information to use the referenced inode (it's more
2293 * likely to be the 'good' one), and throw away the duplicates.
2295 if (isp->volSummary && skip) {
2296 /* use tempHeader, so we can use the stuff[] array to easily index
2297 * into the isp->volSummary special inodes */
2298 memcpy(&tempHeader, &isp->volSummary->header, sizeof(struct VolumeHeader));
2300 for (i = 0; i < isp->nSpecialInodes; i++) {
2301 ip = &inodes[isp->index + i];
2302 if (ip->u.special.type <= 0 || ip->u.special.type > MAXINODETYPE) {
2303 /* will get taken care of in a later loop */
2306 if (ip->inodeNumber == *(stuff[ip->u.special.type - 1].inode)) {
2307 goodspecial[ip->u.special.type-1].valid = 1;
2308 goodspecial[ip->u.special.type-1].inode = ip->inodeNumber;
2313 memset(&tempHeader, 0, sizeof(tempHeader));
2314 tempHeader.stamp.magic = VOLUMEHEADERMAGIC;
2315 tempHeader.stamp.version = VOLUMEHEADERVERSION;
2316 tempHeader.id = isp->volumeId;
2317 tempHeader.parent = isp->RWvolumeId;
2319 /* Check for duplicates (inodes are sorted by type field) */
2320 for (i = 0; i < isp->nSpecialInodes - 1; i++) {
2321 ip = &inodes[isp->index + i];
2322 if (ip->u.special.type == (ip + 1)->u.special.type) {
2323 afs_ino_str_t stmp1, stmp2;
2325 if (ip->u.special.type <= 0 || ip->u.special.type > MAXINODETYPE) {
2326 /* Will be caught in the loop below */
2330 Log("Duplicate special %d inodes for volume %" AFS_VOLID_FMT " found (%s, %s);\n",
2331 ip->u.special.type, afs_printable_VolumeId_lu(isp->volumeId),
2332 PrintInode(stmp1, ip->inodeNumber),
2333 PrintInode(stmp2, (ip+1)->inodeNumber));
2335 if (skip && goodspecial[ip->u.special.type-1].valid) {
2336 Inode gi = goodspecial[ip->u.special.type-1].inode;
2339 Log("using special inode referenced by vol header (%s)\n",
2340 PrintInode(stmp1, gi));
2343 /* the volume header references some special inode of
2344 * this type in the inodes array; are we it? */
2345 if (ip->inodeNumber != gi) {
2347 } else if ((ip+1)->inodeNumber != gi) {
2348 /* in case this is the last iteration; we need to
2349 * make sure we check ip+1, too */
2354 Log("cannot determine which is correct; salvage of volume %" AFS_VOLID_FMT " aborted\n", afs_printable_VolumeId_lu(isp->volumeId));
2362 for (i = 0; i < isp->nSpecialInodes; i++) {
2364 ip = &inodes[isp->index + i];
2365 if (ip->u.special.type <= 0 || ip->u.special.type > MAXINODETYPE) {
2367 Log("Rubbish header inode %s of type %d\n",
2368 PrintInode(stmp, ip->inodeNumber),
2369 ip->u.special.type);
2375 Log("Rubbish header inode %s of type %d; deleted\n",
2376 PrintInode(stmp, ip->inodeNumber),
2377 ip->u.special.type);
2378 } else if (!stuff[ip->u.special.type - 1].obsolete) {
2379 if (skip && skip[i]) {
2380 if (orphans == ORPH_REMOVE) {
2381 Log("Removing orphan special inode %s of type %d\n",
2382 PrintInode(stmp, ip->inodeNumber), ip->u.special.type);
2385 Log("Ignoring orphan special inode %s of type %d\n",
2386 PrintInode(stmp, ip->inodeNumber), ip->u.special.type);
2387 /* fall through to the ip->linkCount--; line below */
2390 *(stuff[ip->u.special.type - 1].inode) = ip->inodeNumber;
2391 allinodesobsolete = 0;
2393 if (!check && ip->u.special.type != VI_LINKTABLE)
2394 ip->linkCount--; /* Keep the inode around */
2402 if (allinodesobsolete) {
2409 salvinfo->VGLinkH_cnt++; /* one for every header. */
2411 if (!RW && !check && isp->volSummary) {
2412 ClearROInUseBit(isp->volSummary);
2416 for (i = 0; i < MAXINODETYPE; i++) {
2417 if (stuff[i].inodeType == VI_LINKTABLE) {
2418 /* Gross hack: SalvageHeader does a bcmp on the volume header.
2419 * And we may have recreated the link table earlier, so set the
2420 * RW header as well. The header magic was already checked.
2422 if (VALID_INO(salvinfo->VGLinkH->ih_ino)) {
2423 *stuff[i].inode = salvinfo->VGLinkH->ih_ino;
2427 if (SalvageHeader(salvinfo, &stuff[i], isp, check, deleteMe) == -1 && check)
2431 if (isp->volSummary == NULL) {
2433 char headerName[64];
2434 snprintf(headerName, sizeof headerName, VFORMAT,
2435 afs_printable_VolumeId_lu(isp->volumeId));
2436 snprintf(path, sizeof path, "%s" OS_DIRSEP "%s",
2437 salvinfo->fileSysPath, headerName);
2439 Log("No header file for volume %" AFS_VOLID_FMT "\n", afs_printable_VolumeId_lu(isp->volumeId));
2443 Log("No header file for volume %" AFS_VOLID_FMT "; %screating %s\n",
2444 afs_printable_VolumeId_lu(isp->volumeId), (Testing ? "it would have been " : ""),
2446 isp->volSummary = calloc(1, sizeof(struct VolumeSummary));
2448 writefunc = VCreateVolumeDiskHeader;
2451 char headerName[64];
2452 /* hack: these two fields are obsolete... */
2453 isp->volSummary->header.volumeAcl = 0;
2454 isp->volSummary->header.volumeMountTable = 0;
2457 (&isp->volSummary->header, &tempHeader,
2458 sizeof(struct VolumeHeader))) {
2459 VolumeExternalName_r(isp->volumeId, headerName, sizeof(headerName));
2460 snprintf(path, sizeof path, "%s" OS_DIRSEP "%s",
2461 salvinfo->fileSysPath, headerName);
2463 Log("Header file %s is damaged or no longer valid%s\n", path,
2464 (check ? "" : "; repairing"));
2468 writefunc = VWriteVolumeDiskHeader;
2472 memcpy(&isp->volSummary->header, &tempHeader,
2473 sizeof(struct VolumeHeader));
2476 Log("It would have written a new header file for volume %" AFS_VOLID_FMT "\n",
2477 afs_printable_VolumeId_lu(isp->volumeId));
2480 VolumeHeaderToDisk(&diskHeader, &tempHeader);
2481 code = (*writefunc)(&diskHeader, salvinfo->fileSysPartition);
2483 Log("Error %ld writing volume header file for volume %" AFS_VOLID_FMT "\n",
2484 afs_printable_int32_ld(code),
2485 afs_printable_VolumeId_lu(diskHeader.id));
2490 IH_INIT(isp->volSummary->volumeInfoHandle, salvinfo->fileSysDevice, isp->RWvolumeId,
2491 isp->volSummary->header.volumeInfo);
2496 SalvageHeader(struct SalvInfo *salvinfo, struct afs_inode_info *sp,
2497 struct InodeSummary *isp, int check, int *deleteMe)
2500 VolumeDiskData volumeInfo;
2501 struct versionStamp fileHeader;
2510 #ifndef AFS_NAMEI_ENV
2511 if (sp->inodeType == VI_LINKTABLE)
2512 return 0; /* header magic was already checked */
2514 if (*(sp->inode) == 0) {
2516 Log("Missing inode in volume header (%s)\n", sp->description);
2520 Log("Missing inode in volume header (%s); %s\n", sp->description,
2521 (Testing ? "it would have recreated it" : "recreating"));
2524 IH_CREATE(NULL, salvinfo->fileSysDevice, salvinfo->fileSysPath, 0, isp->volumeId,
2525 INODESPECIAL, sp->inodeType, isp->RWvolumeId);
2526 if (!VALID_INO(*(sp->inode)))
2528 ("Unable to allocate inode (%s) for volume header (error = %d)\n",
2529 sp->description, errno);
2534 IH_INIT(specH, salvinfo->fileSysDevice, isp->RWvolumeId, *(sp->inode));
2535 fdP = IH_OPEN(specH);
2536 if (OKToZap && (fdP == NULL) && BadError(errno)) {
2537 /* bail out early and destroy the volume */
2539 Log("Still can't open volume header inode (%s), destroying volume\n", sp->description);
2546 Abort("Unable to open inode (%s) of volume header (error = %d)\n",
2547 sp->description, errno);
2550 && (FDH_PREAD(fdP, (char *)&header, sp->size, 0) != sp->size
2551 || header.fileHeader.magic != sp->stamp.magic)) {
2553 Log("Part of the header (%s) is corrupted\n", sp->description);
2554 FDH_REALLYCLOSE(fdP);
2558 Log("Part of the header (%s) is corrupted; recreating\n",
2561 /* header can be garbage; make sure we don't read garbage data from
2563 memset(&header, 0, sizeof(header));
2565 #ifdef AFS_NAMEI_ENV
2566 if (namei_FixSpecialOGM(fdP, check)) {
2567 Log("Error with namei header OGM data (%s)\n", sp->description);
2568 FDH_REALLYCLOSE(fdP);
2573 if (sp->inodeType == VI_VOLINFO
2574 && header.volumeInfo.destroyMe == DESTROY_ME) {
2577 FDH_REALLYCLOSE(fdP);
2581 if (recreate && !Testing) {
2584 ("Internal error: recreating volume header (%s) in check mode\n",
2586 nBytes = FDH_TRUNC(fdP, 0);
2588 Abort("Unable to truncate volume header file (%s) (error = %d)\n",
2589 sp->description, errno);
2591 /* The following code should be moved into vutil.c */
2592 if (sp->inodeType == VI_VOLINFO) {
2594 memset(&header.volumeInfo, 0, sizeof(header.volumeInfo));
2595 header.volumeInfo.stamp = sp->stamp;
2596 header.volumeInfo.id = isp->volumeId;
2597 header.volumeInfo.parentId = isp->RWvolumeId;
2598 sprintf(header.volumeInfo.name, "bogus.%" AFS_VOLID_FMT, afs_printable_VolumeId_lu(isp->volumeId));
2599 Log("Warning: the name of volume %" AFS_VOLID_FMT " is now \"bogus.%" AFS_VOLID_FMT "\"\n",
2600 afs_printable_VolumeId_lu(isp->volumeId), afs_printable_VolumeId_lu(isp->volumeId));
2601 header.volumeInfo.inService = 0;
2602 header.volumeInfo.blessed = 0;
2603 /* The + 1000 is a hack in case there are any files out in venus caches */
2604 header.volumeInfo.uniquifier = (isp->maxUniquifier + 1) + 1000;
2605 header.volumeInfo.type = (isp->volumeId == isp->RWvolumeId ? readwriteVolume : readonlyVolume); /* XXXX */
2606 header.volumeInfo.needsCallback = 0;
2607 gettimeofday(&tp, NULL);
2608 header.volumeInfo.creationDate = tp.tv_sec;
2610 FDH_PWRITE(fdP, (char *)&header.volumeInfo,
2611 sizeof(header.volumeInfo), 0);
2612 if (nBytes != sizeof(header.volumeInfo)) {
2615 ("Unable to write volume header file (%s) (errno = %d)\n",
2616 sp->description, errno);
2617 Abort("Unable to write entire volume header file (%s)\n",
2621 nBytes = FDH_PWRITE(fdP, (char *)&sp->stamp, sizeof(sp->stamp), 0);
2622 if (nBytes != sizeof(sp->stamp)) {
2625 ("Unable to write version stamp in volume header file (%s) (errno = %d)\n",
2626 sp->description, errno);
2628 ("Unable to write entire version stamp in volume header file (%s)\n",
2633 FDH_REALLYCLOSE(fdP);
2635 if (sp->inodeType == VI_VOLINFO) {
2636 salvinfo->VolInfo = header.volumeInfo;
2641 if (salvinfo->VolInfo.updateDate) {
2642 strcpy(update, TimeStamp(buffer, sizeof(buffer), salvinfo->VolInfo.updateDate, 0));
2644 Log("%s (%" AFS_VOLID_FMT ") %supdated %s\n", salvinfo->VolInfo.name,
2645 afs_printable_VolumeId_lu(salvinfo->VolInfo.id),
2646 (Testing ? "it would have been " : ""), update);
2648 strcpy(update, TimeStamp(buffer, sizeof(buffer), salvinfo->VolInfo.creationDate, 0));
2650 Log("%s (%" AFS_VOLID_FMT ") not updated (created %s)\n",
2651 salvinfo->VolInfo.name, afs_printable_VolumeId_lu(salvinfo->VolInfo.id), update);
2661 SalvageVnodes(struct SalvInfo *salvinfo,
2662 struct InodeSummary *rwIsp,
2663 struct InodeSummary *thisIsp,
2664 struct ViceInodeInfo *inodes, int check)
2666 int ilarge, ismall, ioffset, RW, nInodes;
2667 ioffset = rwIsp->index + rwIsp->nSpecialInodes; /* first inode */
2670 RW = (rwIsp == thisIsp);
2671 nInodes = (rwIsp->nInodes - rwIsp->nSpecialInodes);
2673 SalvageIndex(salvinfo, thisIsp->volSummary->header.smallVnodeIndex, vSmall, RW,
2674 &inodes[ioffset], nInodes, thisIsp->volSummary, check);
2675 if (check && ismall == -1)
2678 SalvageIndex(salvinfo, thisIsp->volSummary->header.largeVnodeIndex, vLarge, RW,
2679 &inodes[ioffset], nInodes, thisIsp->volSummary, check);
2680 return (ilarge == 0 && ismall == 0 ? 0 : -1);
2684 SalvageIndex(struct SalvInfo *salvinfo, Inode ino, VnodeClass class, int RW,
2685 struct ViceInodeInfo *ip, int nInodes,
2686 struct VolumeSummary *volSummary, int check)
2688 char buf[SIZEOF_LARGEDISKVNODE];
2689 struct VnodeDiskObject *vnode = (struct VnodeDiskObject *)buf;
2691 StreamHandle_t *file;
2692 struct VnodeClassInfo *vcp;
2694 afs_sfsize_t nVnodes;
2695 afs_fsize_t vnodeLength;
2697 afs_ino_str_t stmp1, stmp2;
2701 IH_INIT(handle, salvinfo->fileSysDevice, volSummary->header.parent, ino);
2702 fdP = IH_OPEN(handle);
2703 opr_Assert(fdP != NULL);
2704 file = FDH_FDOPEN(fdP, "r+");
2705 opr_Assert(file != NULL);
2706 vcp = &VnodeClassInfo[class];
2707 size = OS_SIZE(fdP->fd_fd);
2708 opr_Assert(size != -1);
2709 nVnodes = (size / vcp->diskSize) - 1;
2711 opr_Assert((nVnodes + 1) * vcp->diskSize == size);
2712 opr_Verify(STREAM_ASEEK(file, vcp->diskSize) == 0);
2716 for (vnodeIndex = 0;
2717 nVnodes && STREAM_READ(vnode, vcp->diskSize, 1, file) == 1;
2718 nVnodes--, vnodeIndex++) {
2719 if (vnode->type != vNull) {
2720 int vnodeChanged = 0;
2721 int vnodeNumber = bitNumberToVnodeNumber(vnodeIndex, class);
2722 if (VNDISK_GET_INO(vnode) == 0) {
2724 /* Log("### DEBUG ### Deleted Vnode with 0 inode (vnode %d)\n", vnodeNumber); */
2725 memset(vnode, 0, vcp->diskSize);
2729 if (vcp->magic != vnode->vnodeMagic) {
2730 /* bad magic #, probably partially created vnode */
2732 Log("Partially allocated vnode %d: bad magic (is %lx should be %lx)\n",
2733 vnodeNumber, afs_printable_uint32_lu(vnode->vnodeMagic),
2734 afs_printable_uint32_lu(vcp->magic));
2735 memset(vnode, 0, vcp->diskSize);
2739 Log("Partially allocated vnode %d deleted.\n",
2741 memset(vnode, 0, vcp->diskSize);
2745 /* ****** Should do a bit more salvage here: e.g. make sure
2746 * vnode type matches what it should be given the index */
2747 while (nInodes && ip->u.vnode.vnodeNumber < vnodeNumber) {
2748 /* if (vnodeIdToClass(ip->u.vnode.vnodeNumber) == class && RW) {
2749 * Log("Inode %d: says it belongs to non-existing vnode %d\n",
2750 * ip->inodeNumber, ip->u.vnode.vnodeNumber);
2757 while (nInodes && ip->u.vnode.vnodeNumber == vnodeNumber) {
2758 /* The following doesn't work, because the version number
2759 * is not maintained correctly by the file server */
2760 /*if (vnode->uniquifier == ip->u.vnode.vnodeUniquifier &&
2761 * vnode->dataVersion == ip->u.vnode.inodeDataVersion)
2763 if (VNDISK_GET_INO(vnode) == ip->inodeNumber)
2769 /* For RW volume, look for vnode with matching inode number;
2770 * if no such match, take the first determined by our sort
2772 struct ViceInodeInfo *lip = ip;
2773 int lnInodes = nInodes;
2775 && lip->u.vnode.vnodeNumber == vnodeNumber) {
2776 if (VNDISK_GET_INO(vnode) == lip->inodeNumber) {
2785 if (nInodes && ip->u.vnode.vnodeNumber == vnodeNumber) {
2786 /* "Matching" inode */
2790 vu = vnode->uniquifier;
2791 iu = ip->u.vnode.vnodeUniquifier;
2792 vd = vnode->dataVersion;
2793 id = ip->u.vnode.inodeDataVersion;
2795 * Because of the possibility of the uniquifier overflows (> 4M)
2796 * we compare them modulo the low 22-bits; we shouldn't worry
2797 * about mismatching since they shouldn't to many old
2798 * uniquifiers of the same vnode...
2800 if (IUnique(vu) != IUnique(iu)) {
2802 Log("Vnode %u: vnode.unique, %u, does not match inode unique, %u; fixed, but status will be wrong\n", vnodeNumber, IUnique(vu), IUnique(iu));
2805 vnode->uniquifier = iu;
2806 #ifdef AFS_3DISPARES
2807 vnode->dataVersion = (id >= vd ?
2810 1887437 ? vd : id) :
2813 1887437 ? id : vd));
2815 #if defined(AFS_SGI_EXMAG)
2816 vnode->dataVersion = (id >= vd ?
2819 15099494 ? vd : id) :
2822 15099494 ? id : vd));
2824 vnode->dataVersion = (id > vd ? id : vd);
2825 #endif /* AFS_SGI_EXMAG */
2826 #endif /* AFS_3DISPARES */
2829 /* don't bother checking for vd > id any more, since
2830 * partial file transfers always result in this state,
2831 * and you can't do much else anyway (you've already
2832 * found the best data you can) */
2833 #ifdef AFS_3DISPARES
2834 if (!vnodeIsDirectory(vnodeNumber)
2835 && ((vd < id && (id - vd) < 1887437)
2836 || ((vd > id && (vd - id) > 1887437)))) {
2838 #if defined(AFS_SGI_EXMAG)
2839 if (!vnodeIsDirectory(vnodeNumber)
2840 && ((vd < id && (id - vd) < 15099494)
2841 || ((vd > id && (vd - id) > 15099494)))) {
2843 if (!vnodeIsDirectory(vnodeNumber) && vd < id) {
2844 #endif /* AFS_SGI_EXMAG */
2847 Log("Vnode %d: version < inode version; fixed (old status)\n", vnodeNumber);
2848 vnode->dataVersion = id;
2853 if (ip->inodeNumber != VNDISK_GET_INO(vnode)) {
2856 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);
2858 VNDISK_SET_INO(vnode, ip->inodeNumber);
2863 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);
2865 VNDISK_SET_INO(vnode, ip->inodeNumber);
2868 VNDISK_GET_LEN(vnodeLength, vnode);
2869 if (ip->byteCount != vnodeLength) {
2872 Log("Vnode %d: length incorrect; (is %llu should be %llu)\n", vnodeNumber, (afs_uintmax_t) vnodeLength, (afs_uintmax_t) ip->byteCount);
2877 Log("Vnode %d: length incorrect; changed from %llu to %llu\n", vnodeNumber, (afs_uintmax_t) vnodeLength, (afs_uintmax_t) ip->byteCount);
2878 VNDISK_SET_LEN(vnode, ip->byteCount);
2882 ip->linkCount--; /* Keep the inode around */
2885 } else { /* no matching inode */
2887 if (VNDISK_GET_INO(vnode) != 0
2888 || vnode->type == vDirectory) {
2889 /* No matching inode--get rid of the vnode */
2891 if (VNDISK_GET_INO(vnode)) {
2893 Log("Vnode %d (unique %u): corresponding inode %s is missing\n", vnodeNumber, vnode->uniquifier, PrintInode(stmp, VNDISK_GET_INO(vnode)));
2897 Log("Vnode %d (unique %u): bad directory vnode (no inode number listed)\n", vnodeNumber, vnode->uniquifier);
2902 if (VNDISK_GET_INO(vnode)) {
2904 time_t serverModifyTime = vnode->serverModifyTime;
2905 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));
2909 time_t serverModifyTime = vnode->serverModifyTime;
2910 Log("Vnode %d (unique %u): bad directory vnode (no inode number listed); vnode deleted, vnode mod time=%s", vnodeNumber, vnode->uniquifier, ctime(&serverModifyTime));
2913 memset(vnode, 0, vcp->diskSize);
2916 /* Should not reach here becuase we checked for
2917 * (inodeNumber == 0) above. And where we zero the vnode,
2918 * we also goto vnodeDone.
2922 while (nInodes && ip->u.vnode.vnodeNumber == vnodeNumber) {
2926 } /* VNDISK_GET_INO(vnode) != 0 */
2928 opr_Assert(!(vnodeChanged && check));
2929 if (vnodeChanged && !Testing) {
2930 opr_Verify(IH_IWRITE(handle,
2931 vnodeIndexOffset(vcp, vnodeNumber),
2932 (char *)vnode, vcp->diskSize)
2934 salvinfo->VolumeChanged = 1; /* For break call back */
2945 struct VnodeEssence *
2946 CheckVnodeNumber(struct SalvInfo *salvinfo, VnodeId vnodeNumber)
2949 struct VnodeInfo *vip;
2952 class = vnodeIdToClass(vnodeNumber);
2953 vip = &salvinfo->vnodeInfo[class];
2954 offset = vnodeIdToBitNumber(vnodeNumber);
2955 return (offset >= vip->nVnodes ? NULL : &vip->vnodes[offset]);
2959 CopyOnWrite(struct SalvInfo *salvinfo, struct DirSummary *dir)
2961 /* Copy the directory unconditionally if we are going to change it:
2962 * not just if was cloned.
2964 struct VnodeDiskObject vnode;
2965 struct VnodeClassInfo *vcp = &VnodeClassInfo[vLarge];
2966 Inode oldinode, newinode;
2969 if (dir->copied || Testing)
2971 DFlush(); /* Well justified paranoia... */
2974 IH_IREAD(salvinfo->vnodeInfo[vLarge].handle,
2975 vnodeIndexOffset(vcp, dir->vnodeNumber), (char *)&vnode,
2977 opr_Assert(code == sizeof(vnode));
2978 oldinode = VNDISK_GET_INO(&vnode);
2979 /* Increment the version number by a whole lot to avoid problems with
2980 * clients that were promised new version numbers--but the file server
2981 * crashed before the versions were written to disk.
2984 IH_CREATE(dir->ds_linkH, salvinfo->fileSysDevice, salvinfo->fileSysPath, 0, dir->rwVid,
2985 dir->vnodeNumber, vnode.uniquifier, vnode.dataVersion +=
2987 opr_Assert(VALID_INO(newinode));
2988 opr_Verify(CopyInode(salvinfo->fileSysDevice, oldinode, newinode,
2991 VNDISK_SET_INO(&vnode, newinode);
2993 IH_IWRITE(salvinfo->vnodeInfo[vLarge].handle,
2994 vnodeIndexOffset(vcp, dir->vnodeNumber), (char *)&vnode,
2996 opr_Assert(code == sizeof(vnode));
2998 SetSalvageDirHandle(&dir->dirHandle, dir->dirHandle.dirh_handle->ih_vid,
2999 salvinfo->fileSysDevice, newinode,
3000 &salvinfo->VolumeChanged);
3001 /* Don't delete the original inode right away, because the directory is
3002 * still being scanned.
3008 * This function should either successfully create a new dir, or give up
3009 * and leave things the way they were. In particular, if it fails to write
3010 * the new dir properly, it should return w/o changing the reference to the
3014 CopyAndSalvage(struct SalvInfo *salvinfo, struct DirSummary *dir)
3016 struct VnodeDiskObject vnode;
3017 struct VnodeClassInfo *vcp = &VnodeClassInfo[vLarge];
3018 Inode oldinode, newinode;
3023 afs_int32 parentUnique = 1;
3024 struct VnodeEssence *vnodeEssence;
3029 Log("Salvaging directory %u...\n", dir->vnodeNumber);
3031 IH_IREAD(salvinfo->vnodeInfo[vLarge].handle,
3032 vnodeIndexOffset(vcp, dir->vnodeNumber), (char *)&vnode,
3034 opr_Assert(lcode == sizeof(vnode));
3035 oldinode = VNDISK_GET_INO(&vnode);
3036 /* Increment the version number by a whole lot to avoid problems with
3037 * clients that were promised new version numbers--but the file server
3038 * crashed before the versions were written to disk.
3041 IH_CREATE(dir->ds_linkH, salvinfo->fileSysDevice, salvinfo->fileSysPath, 0, dir->rwVid,
3042 dir->vnodeNumber, vnode.uniquifier, vnode.dataVersion +=
3044 opr_Assert(VALID_INO(newinode));
3045 SetSalvageDirHandle(&newdir, dir->rwVid, salvinfo->fileSysDevice, newinode,
3046 &salvinfo->VolumeChanged);
3048 /* Assign . and .. vnode numbers from dir and vnode.parent.
3049 * The uniquifier for . is in the vnode.
3050 * The uniquifier for .. might be set to a bogus value of 1 and
3051 * the salvager will later clean it up.
3053 if (vnode.parent && (vnodeEssence = CheckVnodeNumber(salvinfo, vnode.parent))) {
3054 parentUnique = (vnodeEssence->unique ? vnodeEssence->unique : 1);
3057 DirSalvage(&dir->dirHandle, &newdir, dir->vnodeNumber,
3059 (vnode.parent ? vnode.parent : dir->vnodeNumber),
3064 /* didn't really build the new directory properly, let's just give up. */
3065 code = IH_DEC(dir->ds_linkH, newinode, dir->rwVid);
3066 Log("Directory salvage returned code %d, continuing.\n", code);
3068 Log("also failed to decrement link count on new inode");
3072 Log("Checking the results of the directory salvage...\n");
3073 if (!DirOK(&newdir)) {
3074 Log("Directory salvage failed!!!; restoring old version of the directory.\n");
3075 code = IH_DEC(dir->ds_linkH, newinode, dir->rwVid);
3076 opr_Assert(code == 0);
3080 VNDISK_SET_INO(&vnode, newinode);
3081 length = afs_dir_Length(&newdir);
3082 VNDISK_SET_LEN(&vnode, length);
3084 IH_IWRITE(salvinfo->vnodeInfo[vLarge].handle,
3085 vnodeIndexOffset(vcp, dir->vnodeNumber), (char *)&vnode,
3087 opr_Assert(lcode == sizeof(vnode));
3088 IH_CONDSYNC(salvinfo->vnodeInfo[vLarge].handle);
3090 /* make sure old directory file is really closed */
3091 fdP = IH_OPEN(dir->dirHandle.dirh_handle);
3092 FDH_REALLYCLOSE(fdP);
3094 code = IH_DEC(dir->ds_linkH, oldinode, dir->rwVid);
3095 opr_Assert(code == 0);
3096 dir->dirHandle = newdir;
3100 * arguments for JudgeEntry.
3102 struct judgeEntry_params {
3103 struct DirSummary *dir; /**< directory we're examining entries in */
3104 struct SalvInfo *salvinfo; /**< SalvInfo for the current salvage job */
3108 JudgeEntry(void *arock, char *name, afs_int32 vnodeNumber,
3111 struct judgeEntry_params *params = arock;
3112 struct DirSummary *dir = params->dir;
3113 struct SalvInfo *salvinfo = params->salvinfo;
3114 struct VnodeEssence *vnodeEssence;
3115 afs_int32 dirOrphaned, todelete;
3117 dirOrphaned = IsVnodeOrphaned(salvinfo, dir->vnodeNumber);
3119 vnodeEssence = CheckVnodeNumber(salvinfo, vnodeNumber);
3120 if (vnodeEssence == NULL) {
3122 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);
3125 CopyOnWrite(salvinfo, dir);
3126 opr_Verify(afs_dir_Delete(&dir->dirHandle, name) == 0);
3131 #ifndef AFS_NAMEI_ENV
3132 /* On AIX machines, don't allow entries to point to inode 0. That is a special
3133 * mount inode for the partition. If this inode were deleted, it would crash
3136 if (vnodeEssence->InodeNumber == 0) {
3137 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"));
3139 CopyOnWrite(salvinfo, dir);
3140 opr_Verify(afs_dir_Delete(&dir->dirHandle, name) == 0);
3147 if (!(vnodeNumber & 1) && !Showmode
3148 && !(vnodeEssence->count || vnodeEssence->unique
3149 || vnodeEssence->modeBits)) {
3150 Log("dir vnode %u: invalid entry: %s" OS_DIRSEP "%s (vnode %u, unique %u)%s\n",
3151 dir->vnodeNumber, (dir->name ? dir->name : "??"), name,
3152 vnodeNumber, unique,
3153 ((!unique) ? (Testing ? "-- would have deleted" : " -- deleted") :
3157 CopyOnWrite(salvinfo, dir);
3158 opr_Verify(afs_dir_Delete(&dir->dirHandle, name) == 0);
3164 /* Check if the Uniquifiers match. If not, change the directory entry
3165 * so its unique matches the vnode unique. Delete if the unique is zero
3166 * or if the directory is orphaned.
3168 if (!vnodeEssence->unique || (vnodeEssence->unique) != unique) {
3169 todelete = ((!vnodeEssence->unique || dirOrphaned) ? 1 : 0);
3172 && ((strcmp(name, "..") == 0) || (strcmp(name, ".") == 0))) {
3174 /* This is an orphaned directory. Don't delete the . or ..
3175 * entry. Otherwise, it will get created in the next
3176 * salvage and deleted again here. So Just skip it.
3180 /* (vnodeEssence->unique == 0 && ('.' || '..'));
3181 * Entries arriving here should be deleted, but the directory
3182 * is not orphaned. Therefore, the entry must be pointing at
3183 * the wrong vnode. Skip the 'else' clause and fall through;
3184 * the code below will repair the entry so it correctly points
3185 * at the vnode of the current directory (if '.') or the parent
3186 * directory (if '..'). */
3189 Log("dir vnode %u: %s" OS_DIRSEP "%s (vnode %u): unique changed from %u to %u %s\n",
3190 dir->vnodeNumber, (dir->name ? dir->name : "??"), name, vnodeNumber, unique,
3191 vnodeEssence->unique, (!todelete ? "" : (Testing ? "-- would have deleted" : "-- deleted")));
3195 fid.Vnode = vnodeNumber;
3196 fid.Unique = vnodeEssence->unique;
3197 CopyOnWrite(salvinfo, dir);
3198 opr_Verify(afs_dir_Delete(&dir->dirHandle, name) == 0);
3200 opr_Verify(afs_dir_Create(&dir->dirHandle, name, &fid) == 0);
3203 return 0; /* no need to continue */
3207 if (strcmp(name, ".") == 0) {
3208 if (dir->vnodeNumber != vnodeNumber || (dir->unique != unique)) {
3210 Log("directory vnode %u.%u: bad '.' entry (was %u.%u); fixed\n", dir->vnodeNumber, dir->unique, vnodeNumber, unique);
3213 CopyOnWrite(salvinfo, dir);
3214 opr_Verify(afs_dir_Delete(&dir->dirHandle, ".") == 0);
3215 fid.Vnode = dir->vnodeNumber;
3216 fid.Unique = dir->unique;
3217 opr_Verify(afs_dir_Create(&dir->dirHandle, ".", &fid) == 0);
3218 vnodeNumber = fid.Vnode; /* Get the new Essence */
3219 unique = fid.Unique;
3220 vnodeEssence = CheckVnodeNumber(salvinfo, vnodeNumber);
3224 } else if (strcmp(name, "..") == 0) {
3227 struct VnodeEssence *dotdot;
3228 pa.Vnode = dir->parent;
3229 dotdot = CheckVnodeNumber(salvinfo, pa.Vnode);
3230 opr_Assert(dotdot != NULL); /* XXX Should not be assert */
3231 pa.Unique = dotdot->unique;
3233 pa.Vnode = dir->vnodeNumber;
3234 pa.Unique = dir->unique;
3236 if ((pa.Vnode != vnodeNumber) || (pa.Unique != unique)) {
3238 Log("directory vnode %u.%u: bad '..' entry (was %u.%u); fixed\n", dir->vnodeNumber, dir->unique, vnodeNumber, unique);
3240 CopyOnWrite(salvinfo, dir);
3241 opr_Verify(afs_dir_Delete(&dir->dirHandle, "..") == 0);
3242 opr_Verify(afs_dir_Create(&dir->dirHandle, "..", &pa) == 0);
3245 vnodeNumber = pa.Vnode; /* Get the new Essence */
3247 vnodeEssence = CheckVnodeNumber(salvinfo, vnodeNumber);
3249 dir->haveDotDot = 1;
3250 } else if (strncmp(name, ".__afs", 6) == 0) {
3252 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);
3255 CopyOnWrite(salvinfo, dir);
3256 opr_Verify(afs_dir_Delete(&dir->dirHandle, name) == 0);
3258 vnodeEssence->claimed = 0; /* Not claimed: Orphaned */
3259 vnodeEssence->todelete = 1; /* Will later delete vnode and decr inode */
3262 if (ShowSuid && (vnodeEssence->modeBits & 06000))
3263 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);
3264 if (/* ShowMounts && */ (vnodeEssence->type == vSymlink)
3265 && !(vnodeEssence->modeBits & 0111)) {
3266 afs_sfsize_t nBytes;
3272 IH_INIT(ihP, salvinfo->fileSysDevice, dir->dirHandle.dirh_handle->ih_vid,
3273 vnodeEssence->InodeNumber);
3276 Log("ERROR %s could not open mount point vnode %u\n", dir->vname, vnodeNumber);
3280 size = FDH_SIZE(fdP);
3282 Log("ERROR %s mount point has invalid size %d, vnode %u\n", dir->vname, (int)size, vnodeNumber);
3283 FDH_REALLYCLOSE(fdP);
3290 nBytes = FDH_PREAD(fdP, buf, size, 0);
3291 if (nBytes == size) {
3293 if ( (*buf != '#' && *buf != '%') || buf[strlen(buf)-1] != '.' ) {
3294 Log("Volume %" AFS_VOLID_FMT " (%s) mount point %s" OS_DIRSEP "%s to '%s' invalid, %s to symbolic link\n",
3295 afs_printable_VolumeId_lu(dir->dirHandle.dirh_handle->ih_vid), dir->vname, dir->name ? dir->name : "??", name, buf,
3296 Testing ? "would convert" : "converted");
3297 vnodeEssence->modeBits |= 0111;
3298 vnodeEssence->changed = 1;
3299 } else if (ShowMounts)
3300 Log("In volume %" AFS_VOLID_FMT " (%s) found mountpoint %s" OS_DIRSEP "%s to '%s'\n",
3301 afs_printable_VolumeId_lu(dir->dirHandle.dirh_handle->ih_vid),
3302 dir->vname, dir->name ? dir->name : "??", name, buf);
3304 Log("Volume %s cound not read mount point vnode %u size %d code %d\n",
3305 dir->vname, vnodeNumber, (int)size, (int)nBytes);
3307 FDH_REALLYCLOSE(fdP);
3310 if (ShowRootFiles && vnodeEssence->owner == 0 && vnodeNumber != 1)
3311 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);
3312 if (vnodeIdToClass(vnodeNumber) == vLarge
3313 && vnodeEssence->name == NULL) {
3314 vnodeEssence->name = strdup(name);
3317 /* The directory entry points to the vnode. Check to see if the
3318 * vnode points back to the directory. If not, then let the
3319 * directory claim it (else it might end up orphaned). Vnodes
3320 * already claimed by another directory are deleted from this
3321 * directory: hardlinks to the same vnode are not allowed
3322 * from different directories.
3324 if (vnodeEssence->parent != dir->vnodeNumber) {
3325 if (!vnodeEssence->claimed && !dirOrphaned && vnodeNumber != 1) {
3326 /* Vnode does not point back to this directory.
3327 * Orphaned dirs cannot claim a file (it may belong to
3328 * another non-orphaned dir).
3331 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);
3333 vnodeEssence->parent = dir->vnodeNumber;
3334 vnodeEssence->changed = 1;
3336 /* Vnode was claimed by another directory */
3339 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 " : ""));
3340 } else if (vnodeNumber == 1) {
3341 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 " : ""));
3343 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 " : ""));
3347 CopyOnWrite(salvinfo, dir);
3348 opr_Verify(afs_dir_Delete(&dir->dirHandle, name) == 0);
3353 /* This directory claims the vnode */
3354 vnodeEssence->claimed = 1;
3356 vnodeEssence->count--;
3361 DistilVnodeEssence(struct SalvInfo *salvinfo, VolumeId rwVId,
3362 VnodeClass class, Inode ino, Unique * maxu)
3364 struct VnodeInfo *vip = &salvinfo->vnodeInfo[class];
3365 struct VnodeClassInfo *vcp = &VnodeClassInfo[class];
3366 char buf[SIZEOF_LARGEDISKVNODE];
3367 struct VnodeDiskObject *vnode = (struct VnodeDiskObject *)buf;
3369 StreamHandle_t *file;
3374 IH_INIT(vip->handle, salvinfo->fileSysDevice, rwVId, ino);
3375 fdP = IH_OPEN(vip->handle);
3376 opr_Assert(fdP != NULL);
3377 file = FDH_FDOPEN(fdP, "r+");
3378 opr_Assert(file != NULL);
3379 size = OS_SIZE(fdP->fd_fd);
3380 opr_Assert(size != -1);
3381 vip->nVnodes = (size / vcp->diskSize) - 1;
3382 if (vip->nVnodes > 0) {
3383 opr_Assert((vip->nVnodes + 1) * vcp->diskSize == size);
3384 opr_Verify(STREAM_ASEEK(file, vcp->diskSize) == 0);
3385 opr_Verify((vip->vnodes = calloc(vip->nVnodes,
3386 sizeof(struct VnodeEssence)))
3388 if (class == vLarge) {
3389 opr_Verify((vip->inodes = calloc(vip->nVnodes, sizeof(Inode)))
3399 vip->volumeBlockCount = vip->nAllocatedVnodes = 0;
3400 for (vnodeIndex = 0, nVnodes = vip->nVnodes;
3401 nVnodes && STREAM_READ(vnode, vcp->diskSize, 1, file) == 1;
3402 nVnodes--, vnodeIndex++) {
3403 if (vnode->type != vNull) {
3404 struct VnodeEssence *vep = &vip->vnodes[vnodeIndex];
3405 afs_fsize_t vnodeLength;
3406 vip->nAllocatedVnodes++;
3407 vep->count = vnode->linkCount;
3408 VNDISK_GET_LEN(vnodeLength, vnode);
3409 vep->blockCount = nBlocks(vnodeLength);
3410 vip->volumeBlockCount += vep->blockCount;
3411 vep->parent = vnode->parent;
3412 vep->unique = vnode->uniquifier;
3413 if (*maxu < vnode->uniquifier)
3414 *maxu = vnode->uniquifier;
3415 vep->modeBits = vnode->modeBits;
3416 vep->InodeNumber = VNDISK_GET_INO(vnode);
3417 vep->type = vnode->type;
3418 vep->author = vnode->author;
3419 vep->owner = vnode->owner;
3420 vep->group = vnode->group;
3421 if (vnode->type == vDirectory) {
3422 if (class != vLarge) {
3423 VnodeId vnodeNumber = bitNumberToVnodeNumber(vnodeIndex, class);
3424 vip->nAllocatedVnodes--;
3425 memset(vnode, 0, sizeof(*vnode));
3426 IH_IWRITE(salvinfo->vnodeInfo[vSmall].handle,
3427 vnodeIndexOffset(vcp, vnodeNumber),
3428 (char *)&vnode, sizeof(vnode));
3429 salvinfo->VolumeChanged = 1;
3431 vip->inodes[vnodeIndex] = VNDISK_GET_INO(vnode);
3440 GetDirName(struct SalvInfo *salvinfo, VnodeId vnode, struct VnodeEssence *vp,
3443 struct VnodeEssence *parentvp;
3449 if (vp->parent && vp->name && (parentvp = CheckVnodeNumber(salvinfo, vp->parent))
3450 && GetDirName(salvinfo, vp->parent, parentvp, path)) {
3451 strcat(path, OS_DIRSEP);
3452 strcat(path, vp->name);
3458 /* To determine if a vnode is orhpaned or not, the vnode and all its parent
3459 * vnodes must be "claimed". The vep->claimed flag is set in JudgeEntry().
3462 IsVnodeOrphaned(struct SalvInfo *salvinfo, VnodeId vnode)
3464 struct VnodeEssence *vep;
3467 return (1); /* Vnode zero does not exist */
3469 return (0); /* The root dir vnode is always claimed */
3470 vep = CheckVnodeNumber(salvinfo, vnode); /* Get the vnode essence */
3471 if (!vep || !vep->claimed)
3472 return (1); /* Vnode is not claimed - it is orphaned */
3474 return (IsVnodeOrphaned(salvinfo, vep->parent));
3478 SalvageDir(struct SalvInfo *salvinfo, char *name, VolumeId rwVid,
3479 struct VnodeInfo *dirVnodeInfo, IHandle_t * alinkH, int i,
3480 struct DirSummary *rootdir, int *rootdirfound)
3482 static struct DirSummary dir;
3483 static struct DirHandle dirHandle;
3484 struct VnodeEssence *parent;
3485 static char path[MAXPATHLEN];
3488 if (dirVnodeInfo->vnodes[i].salvaged)
3489 return; /* already salvaged */
3492 dirVnodeInfo->vnodes[i].salvaged = 1;
3494 if (dirVnodeInfo->inodes[i] == 0)
3495 return; /* Not allocated to a directory */
3497 if (bitNumberToVnodeNumber(i, vLarge) == 1) {
3498 if (dirVnodeInfo->vnodes[i].parent) {
3499 Log("Bad parent, vnode 1; %s...\n",
3500 (Testing ? "skipping" : "salvaging"));
3501 dirVnodeInfo->vnodes[i].parent = 0;
3502 dirVnodeInfo->vnodes[i].changed = 1;
3505 parent = CheckVnodeNumber(salvinfo, dirVnodeInfo->vnodes[i].parent);
3506 if (parent && parent->salvaged == 0)
3507 SalvageDir(salvinfo, name, rwVid, dirVnodeInfo, alinkH,
3508 vnodeIdToBitNumber(dirVnodeInfo->vnodes[i].parent),
3509 rootdir, rootdirfound);
3512 dir.vnodeNumber = bitNumberToVnodeNumber(i, vLarge);
3513 dir.unique = dirVnodeInfo->vnodes[i].unique;
3516 dir.parent = dirVnodeInfo->vnodes[i].parent;
3517 dir.haveDot = dir.haveDotDot = 0;
3518 dir.ds_linkH = alinkH;
3519 SetSalvageDirHandle(&dir.dirHandle, dir.rwVid, salvinfo->fileSysDevice,
3520 dirVnodeInfo->inodes[i], &salvinfo->VolumeChanged);
3522 dirok = ((RebuildDirs && !Testing) ? 0 : DirOK(&dir.dirHandle));
3525 Log("Directory bad, vnode %u; %s...\n", dir.vnodeNumber,
3526 (Testing ? "skipping" : "salvaging"));
3529 CopyAndSalvage(salvinfo, &dir);
3531 dirVnodeInfo->inodes[i] = dir.dirHandle.dirh_inode;
3534 dirHandle = dir.dirHandle;
3537 GetDirName(salvinfo, bitNumberToVnodeNumber(i, vLarge),
3538 &dirVnodeInfo->vnodes[i], path);
3541 /* If enumeration failed for random reasons, we will probably delete
3542 * too much stuff, so we guard against this instead.
3544 struct judgeEntry_params judge_params;
3545 judge_params.salvinfo = salvinfo;
3546 judge_params.dir = &dir;
3548 opr_Verify(afs_dir_EnumerateDir(&dirHandle, JudgeEntry,
3549 &judge_params) == 0);
3552 /* Delete the old directory if it was copied in order to salvage.
3553 * CopyOnWrite has written the new inode # to the disk, but we still
3554 * have the old one in our local structure here. Thus, we idec the
3558 if (dir.copied && !Testing) {
3559 code = IH_DEC(dir.ds_linkH, dirHandle.dirh_handle->ih_ino, rwVid);
3560 opr_Assert(code == 0);
3561 dirVnodeInfo->inodes[i] = dir.dirHandle.dirh_inode;
3564 /* Remember rootdir DirSummary _after_ it has been judged */
3565 if (dir.vnodeNumber == 1 && dir.unique == 1) {
3566 memcpy(rootdir, &dir, sizeof(struct DirSummary));
3574 * Get a new FID that can be used to create a new file.
3576 * @param[in] volHeader vol header for the volume
3577 * @param[in] class what type of vnode we'll be creating (vLarge or vSmall)
3578 * @param[out] afid the FID that we can use (only Vnode and Unique are set)
3579 * @param[inout] maxunique max uniquifier for all vnodes in the volume;
3580 * updated to the new max unique if we create a new
3584 GetNewFID(struct SalvInfo *salvinfo, VolumeDiskData *volHeader,
3585 VnodeClass class, AFSFid *afid, Unique *maxunique)
3588 for (i = 0; i < salvinfo->vnodeInfo[class].nVnodes; i++) {
3589 if (salvinfo->vnodeInfo[class].vnodes[i].type == vNull) {
3593 if (i == salvinfo->vnodeInfo[class].nVnodes) {
3594 /* no free vnodes; make a new one */
3595 salvinfo->vnodeInfo[class].nVnodes++;
3596 salvinfo->vnodeInfo[class].vnodes =
3597 realloc(salvinfo->vnodeInfo[class].vnodes,
3598 sizeof(struct VnodeEssence) * (i+1));
3600 salvinfo->vnodeInfo[class].vnodes[i].type = vNull;
3603 afid->Vnode = bitNumberToVnodeNumber(i, class);
3605 if (volHeader->uniquifier < (*maxunique + 1)) {
3606 /* header uniq is bad; it will get bumped by 2000 later */
3607 afid->Unique = *maxunique + 1 + 2000;
3610 /* header uniq seems okay; just use that */
3611 afid->Unique = *maxunique = volHeader->uniquifier++;
3616 * Create a vnode for a README file explaining not to use a recreated-root vol.
3618 * @param[in] volHeader vol header for the volume
3619 * @param[in] alinkH ihandle for i/o for the volume
3620 * @param[in] vid volume id
3621 * @param[inout] maxunique max uniquifier for all vnodes in the volume;
3622 * updated to the new max unique if we create a new
3624 * @param[out] afid FID for the new readme vnode
3625 * @param[out] ainode the inode for the new readme file
3627 * @return operation status
3632 CreateReadme(struct SalvInfo *salvinfo, VolumeDiskData *volHeader,
3633 IHandle_t *alinkH, VolumeId vid, Unique *maxunique, AFSFid *afid,
3637 struct VnodeDiskObject *rvnode = NULL;
3639 IHandle_t *readmeH = NULL;
3640 struct VnodeEssence *vep;
3642 time_t now = time(NULL);
3644 /* Try to make the note brief, but informative. Only administrators should
3645 * be able to read this file at first, so we can hopefully assume they
3646 * know what AFS is, what a volume is, etc. */
3648 "This volume has been salvaged, but has lost its original root directory.\n"
3649 "The root directory that exists now has been recreated from orphan files\n"
3650 "from the rest of the volume. This recreated root directory may interfere\n"
3651 "with old cached data on clients, and there is no way the salvager can\n"
3652 "reasonably prevent that. So, it is recommended that you do not continue to\n"
3653 "use this volume, but only copy the salvaged data to a new volume.\n"
3654 "Continuing to use this volume as it exists now may cause some clients to\n"
3655 "behave oddly when accessing this volume.\n"
3656 "\n\t -- Your friendly neighborhood OpenAFS salvager\n";
3657 /* ^ the person reading this probably just lost some data, so they could
3658 * use some cheering up. */
3660 /* -1 for the trailing NUL */
3661 length = sizeof(readme) - 1;
3663 GetNewFID(salvinfo, volHeader, vSmall, afid, maxunique);
3665 vep = &salvinfo->vnodeInfo[vSmall].vnodes[vnodeIdToBitNumber(afid->Vnode)];
3667 /* create the inode and write the contents */
3668 readmeinode = IH_CREATE(alinkH, salvinfo->fileSysDevice,
3669 salvinfo->fileSysPath, 0, vid,
3670 afid->Vnode, afid->Unique, 1);
3671 if (!VALID_INO(readmeinode)) {
3672 Log("CreateReadme: readme IH_CREATE failed\n");
3676 IH_INIT(readmeH, salvinfo->fileSysDevice, vid, readmeinode);
3677 bytes = IH_IWRITE(readmeH, 0, readme, length);
3678 IH_RELEASE(readmeH);
3680 if (bytes != length) {
3681 Log("CreateReadme: IWRITE failed (%d/%d)\n", (int)bytes,
3682 (int)sizeof(readme));
3686 /* create the vnode and write it out */
3687 rvnode = calloc(1, SIZEOF_SMALLDISKVNODE);
3689 Log("CreateRootDir: error alloc'ing memory\n");
3693 rvnode->type = vFile;
3695 rvnode->modeBits = 0777;
3696 rvnode->linkCount = 1;
3697 VNDISK_SET_LEN(rvnode, length);
3698 rvnode->uniquifier = afid->Unique;
3699 rvnode->dataVersion = 1;
3700 VNDISK_SET_INO(rvnode, readmeinode);
3701 rvnode->unixModifyTime = rvnode->serverModifyTime = now;
3706 rvnode->vnodeMagic = VnodeClassInfo[vSmall].magic;
3708 bytes = IH_IWRITE(salvinfo->vnodeInfo[vSmall].handle,
3709 vnodeIndexOffset(&VnodeClassInfo[vSmall], afid->Vnode),
3710 (char*)rvnode, SIZEOF_SMALLDISKVNODE);
3712 if (bytes != SIZEOF_SMALLDISKVNODE) {
3713 Log("CreateReadme: IH_IWRITE failed (%d/%d)\n", (int)bytes,
3714 (int)SIZEOF_SMALLDISKVNODE);
3718 /* update VnodeEssence for new readme vnode */
3719 salvinfo->vnodeInfo[vSmall].nAllocatedVnodes++;
3721 vep->blockCount = nBlocks(length);
3722 salvinfo->vnodeInfo[vSmall].volumeBlockCount += vep->blockCount;
3723 vep->parent = rvnode->parent;
3724 vep->unique = rvnode->uniquifier;
3725 vep->modeBits = rvnode->modeBits;
3726 vep->InodeNumber = VNDISK_GET_INO(rvnode);
3727 vep->type = rvnode->type;
3728 vep->author = rvnode->author;
3729 vep->owner = rvnode->owner;
3730 vep->group = rvnode->group;
3740 *ainode = readmeinode;
3745 if (IH_DEC(alinkH, readmeinode, vid)) {
3746 Log("CreateReadme (recovery): IH_DEC failed\n");
3758 * create a root dir for a volume that lacks one.
3760 * @param[in] volHeader vol header for the volume
3761 * @param[in] alinkH ihandle for disk access for this volume group
3762 * @param[in] vid volume id we're dealing with
3763 * @param[out] rootdir populated with info about the new root dir
3764 * @param[inout] maxunique max uniquifier for all vnodes in the volume;
3765 * updated to the new max unique if we create a new
3768 * @return operation status
3773 CreateRootDir(struct SalvInfo *salvinfo, VolumeDiskData *volHeader,
3774 IHandle_t *alinkH, VolumeId vid, struct DirSummary *rootdir,
3778 int decroot = 0, decreadme = 0;
3779 AFSFid did, readmeid;
3782 struct VnodeDiskObject *rootvnode = NULL;
3783 struct acl_accessList *ACL;
3786 struct VnodeEssence *vep;
3787 Inode readmeinode = 0;
3788 time_t now = time(NULL);
3790 if (!salvinfo->vnodeInfo[vLarge].vnodes && !salvinfo->vnodeInfo[vSmall].vnodes) {
3791 Log("Not creating new root dir; volume appears to lack any vnodes\n");
3795 if (!salvinfo->vnodeInfo[vLarge].vnodes) {
3796 /* We don't have any large vnodes in the volume; allocate room
3797 * for one so we can recreate the root dir */
3798 salvinfo->vnodeInfo[vLarge].nVnodes = 1;
3799 salvinfo->vnodeInfo[vLarge].vnodes = calloc(1, sizeof(struct VnodeEssence));
3800 salvinfo->vnodeInfo[vLarge].inodes = calloc(1, sizeof(Inode));
3802 opr_Assert(salvinfo->vnodeInfo[vLarge].vnodes);
3803 opr_Assert(salvinfo->vnodeInfo[vLarge].inodes);
3806 vep = &salvinfo->vnodeInfo[vLarge].vnodes[vnodeIdToBitNumber(1)];
3807 ip = &salvinfo->vnodeInfo[vLarge].inodes[vnodeIdToBitNumber(1)];
3808 if (vep->type != vNull) {
3809 Log("Not creating new root dir; existing vnode 1 is non-null\n");
3813 if (CreateReadme(salvinfo, volHeader, alinkH, vid, maxunique, &readmeid,
3814 &readmeinode) != 0) {
3819 /* set the DV to a very high number, so it is unlikely that we collide
3820 * with a cached DV */
3823 rootinode = IH_CREATE(alinkH, salvinfo->fileSysDevice, salvinfo->fileSysPath,
3825 if (!VALID_INO(rootinode)) {
3826 Log("CreateRootDir: IH_CREATE failed\n");
3831 SetSalvageDirHandle(&rootdir->dirHandle, vid, salvinfo->fileSysDevice,
3832 rootinode, &salvinfo->VolumeChanged);
3836 if (afs_dir_MakeDir(&rootdir->dirHandle, (afs_int32*)&did, (afs_int32*)&did)) {
3837 Log("CreateRootDir: MakeDir failed\n");
3840 if (afs_dir_Create(&rootdir->dirHandle, "README.ROOTDIR", &readmeid)) {
3841 Log("CreateRootDir: Create failed\n");
3845 length = afs_dir_Length(&rootdir->dirHandle);
3846 DZap(&rootdir->dirHandle);
3848 /* create the new root dir vnode */
3849 rootvnode = calloc(1, SIZEOF_LARGEDISKVNODE);
3851 Log("CreateRootDir: malloc failed\n");
3855 /* only give 'rl' permissions to 'system:administrators'. We do this to
3856 * try to catch the attention of an administrator, that they should not
3857 * be writing to this directory or continue to use it. */
3858 ACL = VVnodeDiskACL(rootvnode);
3859 ACL->size = sizeof(struct acl_accessList);
3860 ACL->version = ACL_ACLVERSION;
3864 ACL->entries[0].id = -204; /* system:administrators */
3865 ACL->entries[0].rights = PRSFS_READ | PRSFS_LOOKUP;
3867 rootvnode->type = vDirectory;
3868 rootvnode->cloned = 0;
3869 rootvnode->modeBits = 0777;
3870 rootvnode->linkCount = 2;
3871 VNDISK_SET_LEN(rootvnode, length);
3872 rootvnode->uniquifier = 1;
3873 rootvnode->dataVersion = dv;
3874 VNDISK_SET_INO(rootvnode, rootinode);
3875 rootvnode->unixModifyTime = rootvnode->serverModifyTime = now;
3876 rootvnode->author = 0;
3877 rootvnode->owner = 0;
3878 rootvnode->parent = 0;
3879 rootvnode->group = 0;
3880 rootvnode->vnodeMagic = VnodeClassInfo[vLarge].magic;
3882 /* write it out to disk */
3883 bytes = IH_IWRITE(salvinfo->vnodeInfo[vLarge].handle,
3884 vnodeIndexOffset(&VnodeClassInfo[vLarge], 1),
3885 (char*)rootvnode, SIZEOF_LARGEDISKVNODE);
3887 if (bytes != SIZEOF_LARGEDISKVNODE) {
3888 /* just cast to int and don't worry about printing real 64-bit ints;
3889 * a large disk vnode isn't anywhere near the 32-bit limit */
3890 Log("CreateRootDir: IH_IWRITE failed (%d/%d)\n", (int)bytes,
3891 (int)SIZEOF_LARGEDISKVNODE);
3895 /* update VnodeEssence for the new root vnode */
3896 salvinfo->vnodeInfo[vLarge].nAllocatedVnodes++;
3898 vep->blockCount = nBlocks(length);
3899 salvinfo->vnodeInfo[vLarge].volumeBlockCount += vep->blockCount;
3900 vep->parent = rootvnode->parent;
3901 vep->unique = rootvnode->uniquifier;
3902 vep->modeBits = rootvnode->modeBits;
3903 vep->InodeNumber = VNDISK_GET_INO(rootvnode);
3904 vep->type = rootvnode->type;
3905 vep->author = rootvnode->author;
3906 vep->owner = rootvnode->owner;
3907 vep->group = rootvnode->group;
3917 /* update DirSummary for the new root vnode */
3918 rootdir->vnodeNumber = 1;
3919 rootdir->unique = 1;
3920 rootdir->haveDot = 1;
3921 rootdir->haveDotDot = 1;
3922 rootdir->rwVid = vid;
3923 rootdir->copied = 0;
3924 rootdir->parent = 0;
3925 rootdir->name = strdup(".");
3926 rootdir->vname = volHeader->name;
3927 rootdir->ds_linkH = alinkH;
3934 if (decroot && IH_DEC(alinkH, rootinode, vid)) {
3935 Log("CreateRootDir (recovery): IH_DEC (root) failed\n");
3937 if (decreadme && IH_DEC(alinkH, readmeinode, vid)) {
3938 Log("CreateRootDir (recovery): IH_DEC (readme) failed\n");
3948 * salvage a volume group.
3950 * @param[in] salvinfo information for the curent salvage job
3951 * @param[in] rwIsp inode summary for rw volume
3952 * @param[in] alinkH link table inode handle
3954 * @return operation status
3958 SalvageVolume(struct SalvInfo *salvinfo, struct InodeSummary *rwIsp, IHandle_t * alinkH)
3960 /* This routine, for now, will only be called for read-write volumes */
3962 int BlocksInVolume = 0, FilesInVolume = 0;
3964 struct DirSummary rootdir, oldrootdir;
3965 struct VnodeInfo *dirVnodeInfo;
3966 struct VnodeDiskObject vnode;
3967 VolumeDiskData volHeader;
3969 int orphaned, rootdirfound = 0;
3970 Unique maxunique = 0; /* the maxUniquifier from the vnodes */
3971 afs_int32 ofiles = 0, oblocks = 0; /* Number of orphaned files/blocks */
3972 struct VnodeEssence *vep;
3975 afs_sfsize_t nBytes;
3977 VnodeId LFVnode, ThisVnode;
3978 Unique LFUnique, ThisUnique;
3982 vid = rwIsp->volSummary->header.id;
3983 IH_INIT(h, salvinfo->fileSysDevice, vid, rwIsp->volSummary->header.volumeInfo);
3984 nBytes = IH_IREAD(h, 0, (char *)&volHeader, sizeof(volHeader));
3985 opr_Assert(nBytes == sizeof(volHeader));
3986 opr_Assert(volHeader.stamp.magic == VOLUMEINFOMAGIC);
3987 opr_Assert(volHeader.destroyMe != DESTROY_ME);
3988 /* (should not have gotten this far with DESTROY_ME flag still set!) */
3990 DistilVnodeEssence(salvinfo, vid, vLarge,
3991 rwIsp->volSummary->header.largeVnodeIndex, &maxunique);
3992 DistilVnodeEssence(salvinfo, vid, vSmall,
3993 rwIsp->volSummary->header.smallVnodeIndex, &maxunique);
3995 dirVnodeInfo = &salvinfo->vnodeInfo[vLarge];
3996 for (i = 0; i < dirVnodeInfo->nVnodes; i++) {
3997 SalvageDir(salvinfo, volHeader.name, vid, dirVnodeInfo, alinkH, i,
3998 &rootdir, &rootdirfound);
4001 nt_sync(salvinfo->fileSysDevice);
4003 sync(); /* This used to be done lower level, for every dir */
4010 if (!rootdirfound && (orphans == ORPH_ATTACH) && !Testing) {
4012 Log("Cannot find root directory for volume %lu; attempting to create "
4013 "a new one\n", afs_printable_uint32_lu(vid));
4015 code = CreateRootDir(salvinfo, &volHeader, alinkH, vid, &rootdir,
4020 salvinfo->VolumeChanged = 1;
4024 /* Parse each vnode looking for orphaned vnodes and
4025 * connect them to the tree as orphaned (if requested).
4027 oldrootdir = rootdir;
4028 for (class = 0; class < nVNODECLASSES; class++) {
4029 for (v = 0; v < salvinfo->vnodeInfo[class].nVnodes; v++) {
4030 vep = &(salvinfo->vnodeInfo[class].vnodes[v]);
4031 ThisVnode = bitNumberToVnodeNumber(v, class);
4032 ThisUnique = vep->unique;
4034 if ((vep->type == 0) || vep->claimed || ThisVnode == 1)
4035 continue; /* Ignore unused, claimed, and root vnodes */
4037 /* This vnode is orphaned. If it is a directory vnode, then the '..'
4038 * entry in this vnode had incremented the parent link count (In
4039 * JudgeEntry()). We need to go to the parent and decrement that
4040 * link count. But if the parent's unique is zero, then the parent
4041 * link count was not incremented in JudgeEntry().
4043 if (class == vLarge) { /* directory vnode */
4044 pv = vnodeIdToBitNumber(vep->parent);
4045 if (salvinfo->vnodeInfo[vLarge].vnodes[pv].unique != 0) {
4046 if (vep->parent == 1 && newrootdir) {
4047 /* this vnode's parent was the volume root, and
4048 * we just created the volume root. So, the parent
4049 * dir didn't exist during JudgeEntry, so the link
4050 * count was not inc'd there, so don't dec it here.
4056 salvinfo->vnodeInfo[vLarge].vnodes[pv].count++;
4062 continue; /* If no rootdir, can't attach orphaned files */
4064 /* Here we attach orphaned files and directories into the
4065 * root directory, LVVnode, making sure link counts stay correct.
4067 if ((orphans == ORPH_ATTACH) && !vep->todelete && !Testing) {
4068 LFVnode = rootdir.vnodeNumber; /* Lost+Found vnode number */
4069 LFUnique = rootdir.unique; /* Lost+Found uniquifier */
4071 /* Update this orphaned vnode's info. Its parent info and
4072 * link count (do for orphaned directories and files).
4074 vep->parent = LFVnode; /* Parent is the root dir */
4075 vep->unique = LFUnique;
4078 vep->count--; /* Inc link count (root dir will pt to it) */
4080 /* If this orphaned vnode is a directory, change '..'.
4081 * The name of the orphaned dir/file is unknown, so we
4082 * build a unique name. No need to CopyOnWrite the directory
4083 * since it is not connected to tree in BK or RO volume and
4084 * won't be visible there.
4086 if (class == vLarge) {
4090 /* Remove and recreate the ".." entry in this orphaned directory */
4091 SetSalvageDirHandle(&dh, vid, salvinfo->fileSysDevice,
4092 salvinfo->vnodeInfo[class].inodes[v],
4093 &salvinfo->VolumeChanged);
4095 pa.Unique = LFUnique;
4096 opr_Verify(afs_dir_Delete(&dh, "..") == 0);
4097 opr_Verify(afs_dir_Create(&dh, "..", &pa) == 0);
4099 /* The original parent's link count was decremented above.
4100 * Here we increment the new parent's link count.
4102 pv = vnodeIdToBitNumber(LFVnode);
4103 salvinfo->vnodeInfo[vLarge].vnodes[pv].count--;
4107 /* Go to the root dir and add this entry. The link count of the
4108 * root dir was incremented when ".." was created. Try 10 times.
4110 for (j = 0; j < 10; j++) {
4111 pa.Vnode = ThisVnode;
4112 pa.Unique = ThisUnique;
4114 snprintf(npath, sizeof npath, "%s.%u.%u",
4115 ((class == vLarge) ? "__ORPHANDIR__"
4116 : "__ORPHANFILE__"),
4117 ThisVnode, ThisUnique);
4119 CopyOnWrite(salvinfo, &rootdir);
4120 code = afs_dir_Create(&rootdir.dirHandle, npath, &pa);
4124 ThisUnique += 50; /* Try creating a different file */
4126 opr_Assert(code == 0);
4127 Log("Attaching orphaned %s to volume's root dir as %s\n",
4128 ((class == vLarge) ? "directory" : "file"), npath);
4130 } /* for each vnode in the class */
4131 } /* for each class of vnode */
4133 /* Delete the old rootinode directory if the rootdir was CopyOnWrite */
4135 if (rootdirfound && !oldrootdir.copied && rootdir.copied) {
4137 IH_DEC(oldrootdir.ds_linkH, oldrootdir.dirHandle.dirh_inode,
4139 opr_Assert(code == 0);
4140 /* dirVnodeInfo->inodes[?] is not updated with new inode number */
4143 DFlush(); /* Flush the changes */
4144 if (!rootdirfound && (orphans == ORPH_ATTACH)) {
4145 Log("Cannot attach orphaned files and directories: Root directory not found\n");
4146 orphans = ORPH_IGNORE;
4149 /* Write out all changed vnodes. Orphaned files and directories
4150 * will get removed here also (if requested).
4152 for (class = 0; class < nVNODECLASSES; class++) {
4153 afs_sfsize_t nVnodes = salvinfo->vnodeInfo[class].nVnodes;
4154 struct VnodeClassInfo *vcp = &VnodeClassInfo[class];
4155 struct VnodeEssence *vnodes = salvinfo->vnodeInfo[class].vnodes;
4156 FilesInVolume += salvinfo->vnodeInfo[class].nAllocatedVnodes;
4157 BlocksInVolume += salvinfo->vnodeInfo[class].volumeBlockCount;
4158 for (i = 0; i < nVnodes; i++) {
4159 struct VnodeEssence *vnp = &vnodes[i];
4160 VnodeId vnodeNumber = bitNumberToVnodeNumber(i, class);
4162 /* If the vnode is good but is unclaimed (not listed in
4163 * any directory entries), then it is orphaned.
4166 if ((vnp->type != 0) && (orphaned = IsVnodeOrphaned(salvinfo, vnodeNumber))) {
4167 vnp->claimed = 0; /* Makes IsVnodeOrphaned calls faster */
4171 if (vnp->changed || vnp->count) {
4174 IH_IREAD(salvinfo->vnodeInfo[class].handle,
4175 vnodeIndexOffset(vcp, vnodeNumber),
4176 (char *)&vnode, sizeof(vnode));
4177 opr_Assert(nBytes == sizeof(vnode));
4179 vnode.parent = vnp->parent;
4180 oldCount = vnode.linkCount;
4181 vnode.linkCount = vnode.linkCount - vnp->count;
4184 orphaned = IsVnodeOrphaned(salvinfo, vnodeNumber);
4186 if (!vnp->todelete) {
4187 /* Orphans should have already been attached (if requested) */
4188 opr_Assert(orphans != ORPH_ATTACH);
4189 oblocks += vnp->blockCount;
4192 if (((orphans == ORPH_REMOVE) || vnp->todelete)
4194 BlocksInVolume -= vnp->blockCount;
4196 if (VNDISK_GET_INO(&vnode)) {
4198 IH_DEC(alinkH, VNDISK_GET_INO(&vnode), vid);
4199 opr_Assert(code == 0);
4201 memset(&vnode, 0, sizeof(vnode));
4203 } else if (vnp->count) {
4205 Log("Vnode %u: link count incorrect (was %d, %s %d)\n", vnodeNumber, oldCount, (Testing ? "would have changed to" : "now"), vnode.linkCount);
4208 vnode.modeBits = vnp->modeBits;
4211 vnode.dataVersion++;
4214 IH_IWRITE(salvinfo->vnodeInfo[class].handle,
4215 vnodeIndexOffset(vcp, vnodeNumber),
4216 (char *)&vnode, sizeof(vnode));
4217 opr_Assert(nBytes == sizeof(vnode));
4219 salvinfo->VolumeChanged = 1;
4223 if (!Showmode && ofiles) {
4224 Log("%s %d orphaned files and directories (approx. %u KB)\n",
4226 && (orphans == ORPH_REMOVE)) ? "Removed" : "Found", ofiles,
4230 for (class = 0; class < nVNODECLASSES; class++) {
4231 struct VnodeInfo *vip = &salvinfo->vnodeInfo[class];
4232 for (i = 0; i < vip->nVnodes; i++)
4233 if (vip->vnodes[i].name)
4234 free(vip->vnodes[i].name);
4241 /* Set correct resource utilization statistics */
4242 volHeader.filecount = FilesInVolume;
4243 volHeader.diskused = BlocksInVolume;
4245 /* Make sure the uniquifer is big enough: maxunique is the real maxUniquifier */
4246 if (volHeader.uniquifier < (maxunique + 1)) {
4248 Log("Volume uniquifier %u is too low (max uniq %u); fixed\n", volHeader.uniquifier, maxunique);
4249 /* Plus 2,000 in case there are workstations out there with
4250 * cached vnodes that have since been deleted
4252 volHeader.uniquifier = (maxunique + 1 + 2000);
4256 Log("*** WARNING: Root directory recreated, but volume is fragile! "
4257 "Only use this salvaged volume to copy data to another volume; "
4258 "do not continue to use this volume (%lu) as-is.\n",
4259 afs_printable_uint32_lu(vid));
4262 if (!Testing && salvinfo->VolumeChanged) {
4263 #ifdef FSSYNC_BUILD_CLIENT
4264 if (salvinfo->useFSYNC) {
4265 afs_int32 fsync_code;
4267 fsync_code = FSYNC_VolOp(vid, NULL, FSYNC_VOL_BREAKCBKS, FSYNC_SALVAGE, NULL);
4269 Log("Error trying to tell the fileserver to break callbacks for "
4270 "changed volume %lu; error code %ld\n",
4271 afs_printable_uint32_lu(vid),
4272 afs_printable_int32_ld(fsync_code));
4274 salvinfo->VolumeChanged = 0;
4277 #endif /* FSSYNC_BUILD_CLIENT */
4279 #ifdef AFS_DEMAND_ATTACH_FS
4280 if (!salvinfo->useFSYNC) {
4281 /* A volume's contents have changed, but the fileserver will not
4282 * break callbacks on the volume until it tries to load the vol
4283 * header. So, to reduce the amount of time a client could have
4284 * stale data, remove fsstate.dat, so the fileserver will init
4285 * callback state with all clients. This is a very coarse hammer,
4286 * and in the future we should just record which volumes have
4288 code = unlink(AFSDIR_SERVER_FSSTATE_FILEPATH);
4289 if (code && errno != ENOENT) {
4290 Log("Error %d when trying to unlink FS state file %s\n", errno,
4291 AFSDIR_SERVER_FSSTATE_FILEPATH);
4297 /* Turn off the inUse bit; the volume's been salvaged! */
4298 volHeader.inUse = 0; /* clear flag indicating inUse@last crash */
4299 volHeader.needsSalvaged = 0; /* clear 'damaged' flag */
4300 volHeader.inService = 1; /* allow service again */
4301 if (salvinfo->VolumeChanged) {
4302 volHeader.needsCallback = 1;
4303 volHeader.updateDate = time(NULL);
4305 volHeader.needsCallback = 0;
4307 volHeader.dontSalvage = DONT_SALVAGE;
4308 salvinfo->VolumeChanged = 0;
4310 nBytes = IH_IWRITE(h, 0, (char *)&volHeader, sizeof(volHeader));
4311 opr_Assert(nBytes == sizeof(volHeader));
4314 Log("%sSalvaged %s (%" AFS_VOLID_FMT "): %d files, %d blocks\n",
4315 (Testing ? "It would have " : ""), volHeader.name, afs_printable_VolumeId_lu(volHeader.id),
4316 FilesInVolume, BlocksInVolume);
4319 IH_RELEASE(salvinfo->vnodeInfo[vSmall].handle);
4320 IH_RELEASE(salvinfo->vnodeInfo[vLarge].handle);
4326 ClearROInUseBit(struct VolumeSummary *summary)
4328 IHandle_t *h = summary->volumeInfoHandle;
4329 afs_sfsize_t nBytes;
4331 VolumeDiskData volHeader;
4333 nBytes = IH_IREAD(h, 0, (char *)&volHeader, sizeof(volHeader));
4334 opr_Assert(nBytes == sizeof(volHeader));
4335 opr_Assert(volHeader.stamp.magic == VOLUMEINFOMAGIC);
4336 volHeader.inUse = 0;
4337 volHeader.needsSalvaged = 0;
4338 volHeader.inService = 1;
4339 volHeader.dontSalvage = DONT_SALVAGE;
4341 nBytes = IH_IWRITE(h, 0, (char *)&volHeader, sizeof(volHeader));
4342 opr_Assert(nBytes == sizeof(volHeader));
4347 * Possible delete the volume.
4349 * deleteMe - Always do so, only a partial volume.
4352 MaybeZapVolume(struct SalvInfo *salvinfo, struct InodeSummary *isp,
4353 char *message, int deleteMe, int check)
4355 if (readOnly(isp) || deleteMe) {
4356 if (isp->volSummary && !isp->volSummary->deleted) {
4359 Log("Volume %" AFS_VOLID_FMT " (is only a partial volume--probably an attempt was made to move/restore it when a machine crash occured.\n", afs_printable_VolumeId_lu(isp->volumeId));
4361 Log("It will be deleted on this server (you may find it elsewhere)\n");
4364 Log("Volume %" AFS_VOLID_FMT " needs to be salvaged. Since it is read-only, however,\n", afs_printable_VolumeId_lu(isp->volumeId));
4366 Log("it will be deleted instead. It should be recloned.\n");
4371 char filename[VMAXPATHLEN];
4372 VolumeExternalName_r(isp->volumeId, filename, sizeof(filename));
4373 sprintf(path, "%s" OS_DIRSEP "%s", salvinfo->fileSysPath, filename);
4375 code = VDestroyVolumeDiskHeader(salvinfo->fileSysPartition, isp->volumeId, isp->RWvolumeId);
4377 Log("Error %ld destroying volume disk header for volume %" AFS_VOLID_FMT "\n",
4378 afs_printable_int32_ld(code),
4379 afs_printable_VolumeId_lu(isp->volumeId));
4382 /* make sure we actually delete the header file; ENOENT
4383 * is fine, since VDestroyVolumeDiskHeader probably already
4385 if (unlink(path) && errno != ENOENT) {
4386 Log("Unable to unlink %s (errno = %d)\n", path, errno);
4388 if (salvinfo->useFSYNC) {
4389 AskDelete(salvinfo, isp->volumeId);
4391 isp->volSummary->deleted = 1;
4394 } else if (!check) {
4395 Log("%s salvage was unsuccessful: read-write volume %" AFS_VOLID_FMT "\n", message,
4396 afs_printable_VolumeId_lu(isp->volumeId));
4397 Abort("Salvage of volume %" AFS_VOLID_FMT " aborted\n", afs_printable_VolumeId_lu(isp->volumeId));
4401 #ifdef AFS_DEMAND_ATTACH_FS
4403 * Locks a volume on disk for salvaging.
4405 * @param[in] volumeId volume ID to lock
4407 * @return operation status
4409 * @retval -1 volume lock raced with a fileserver restart; all volumes must
4410 * checked out and locked again
4415 LockVolume(struct SalvInfo *salvinfo, VolumeId volumeId)
4420 /* should always be WRITE_LOCK, but keep the lock-type logic all
4421 * in one place, in VVolLockType. Params will be ignored, but
4422 * try to provide what we're logically doing. */
4423 locktype = VVolLockType(V_VOLUPD, 1);
4425 code = VLockVolumeByIdNB(volumeId, salvinfo->fileSysPartition, locktype);
4427 if (code == EBUSY) {
4428 Abort("Someone else appears to be using volume %lu; Aborted\n",
4429 afs_printable_uint32_lu(volumeId));
4431 Abort("Error %ld trying to lock volume %lu; Aborted\n",
4432 afs_printable_int32_ld(code),
4433 afs_printable_uint32_lu(volumeId));
4436 code = FSYNC_VerifyCheckout(volumeId, salvinfo->fileSysPartition->name, FSYNC_VOL_OFF, FSYNC_SALVAGE);
4437 if (code == SYNC_DENIED) {
4438 /* need to retry checking out volumes */
4441 if (code != SYNC_OK) {
4442 Abort("FSYNC_VerifyCheckout failed for volume %lu with code %ld\n",
4443 afs_printable_uint32_lu(volumeId), afs_printable_int32_ld(code));
4446 /* set inUse = programType in the volume header to ensure that nobody
4447 * tries to use this volume again without salvaging, if we somehow crash
4448 * or otherwise exit before finishing the salvage.
4452 struct VolumeHeader header;
4453 struct VolumeDiskHeader diskHeader;
4454 struct VolumeDiskData volHeader;
4456 code = VReadVolumeDiskHeader(volumeId, salvinfo->fileSysPartition, &diskHeader);
4461 DiskToVolumeHeader(&header, &diskHeader);
4463 IH_INIT(h, salvinfo->fileSysDevice, header.parent, header.volumeInfo);
4464 if (IH_IREAD(h, 0, (char*)&volHeader, sizeof(volHeader)) != sizeof(volHeader) ||
4465 volHeader.stamp.magic != VOLUMEINFOMAGIC) {
4471 volHeader.inUse = programType;
4473 /* If we can't re-write the header, bail out and error. We don't
4474 * assert when reading the header, since it's possible the
4475 * header isn't really there (when there's no data associated
4476 * with the volume; we just delete the vol header file in that
4477 * case). But if it's there enough that we can read it, but
4478 * somehow we cannot write to it to signify we're salvaging it,
4479 * we've got a big problem and we cannot continue. */
4480 opr_Verify(IH_IWRITE(h, 0, (char*)&volHeader, sizeof(volHeader))
4481 == sizeof(volHeader));
4488 #endif /* AFS_DEMAND_ATTACH_FS */
4491 AskError(struct SalvInfo *salvinfo, VolumeId volumeId)
4493 #if defined(AFS_DEMAND_ATTACH_FS) || defined(AFS_DEMAND_ATTACH_UTIL)
4495 code = FSYNC_VolOp(volumeId, salvinfo->fileSysPartition->name,
4496 FSYNC_VOL_FORCE_ERROR, FSYNC_WHATEVER, NULL);
4497 if (code != SYNC_OK) {
4498 Log("AskError: failed to force volume %lu into error state; "
4499 "SYNC error code %ld (%s)\n", (long unsigned)volumeId,
4500 (long)code, SYNC_res2string(code));
4502 #endif /* AFS_DEMAND_ATTACH_FS || AFS_DEMAND_ATTACH_UTIL */
4506 AskOffline(struct SalvInfo *salvinfo, VolumeId volumeId)
4511 memset(&res, 0, sizeof(res));
4513 for (i = 0; i < 3; i++) {
4514 code = FSYNC_VolOp(volumeId, salvinfo->fileSysPartition->name,
4515 FSYNC_VOL_OFF, FSYNC_SALVAGE, &res);
4517 if (code == SYNC_OK) {
4519 } else if (code == SYNC_DENIED) {
4521 Log("AskOffline: file server denied offline request; a general salvage may be required.\n");
4523 Log("AskOffline: file server denied offline request; a general salvage is required.\n");
4524 Abort("Salvage aborted\n");
4525 } else if (code == SYNC_BAD_COMMAND) {
4526 Log("AskOffline: fssync protocol mismatch (bad command word '%d'); salvage aborting.\n",
4529 #ifdef AFS_DEMAND_ATTACH_FS
4530 Log("AskOffline: please make sure dafileserver, davolserver, salvageserver and dasalvager binaries are same version.\n");
4532 Log("AskOffline: fileserver is DAFS but we are not.\n");
4535 #ifdef AFS_DEMAND_ATTACH_FS
4536 Log("AskOffline: fileserver is not DAFS but we are.\n");
4538 Log("AskOffline: please make sure fileserver, volserver and salvager binaries are same version.\n");
4541 Abort("Salvage aborted\n");
4544 Log("AskOffline: request for fileserver to take volume offline failed; trying again...\n");
4545 FSYNC_clientFinis();
4549 if (code != SYNC_OK) {
4550 Log("AskOffline: request for fileserver to take volume offline failed; salvage aborting.\n");
4551 Abort("Salvage aborted\n");
4555 /* don't want to pass around state; remember it here */
4556 static int isDAFS = -1;
4561 afs_int32 code = 1, i;
4563 /* we don't care if we race. the answer shouldn't change */
4567 memset(&res, 0, sizeof(res));
4569 for (i = 0; code && i < 3; i++) {
4570 code = FSYNC_VolOp(0, NULL, FSYNC_VOL_LISTVOLUMES, FSYNC_SALVAGE, &res);
4572 Log("AskDAFS: FSYNC_VOL_LISTVOLUMES failed with code %ld reason "
4573 "%ld (%s); trying again...\n", (long)code, (long)res.hdr.reason,
4574 FSYNC_reason2string(res.hdr.reason));
4575 FSYNC_clientFinis();
4581 Log("AskDAFS: could not determine DAFS-ness, assuming not DAFS\n");
4585 if ((res.hdr.flags & SYNC_FLAG_DAFS_EXTENSIONS)) {
4595 MaybeAskOnline(struct SalvInfo *salvinfo, VolumeId volumeId)
4597 struct VolumeDiskHeader diskHdr;
4599 code = VReadVolumeDiskHeader(volumeId, salvinfo->fileSysPartition, &diskHdr);
4601 /* volume probably does not exist; no need to bring back online */
4604 AskOnline(salvinfo, volumeId);
4608 AskOnline(struct SalvInfo *salvinfo, VolumeId volumeId)
4612 for (i = 0; i < 3; i++) {
4613 code = FSYNC_VolOp(volumeId, salvinfo->fileSysPartition->name,
4614 FSYNC_VOL_ON, FSYNC_WHATEVER, NULL);
4616 if (code == SYNC_OK) {
4618 } else if (code == SYNC_DENIED) {
4619 Log("AskOnline: file server denied online request to volume %" AFS_VOLID_FMT " partition %s; trying again...\n", afs_printable_VolumeId_lu(volumeId), salvinfo->fileSysPartition->name);
4620 } else if (code == SYNC_BAD_COMMAND) {
4621 Log("AskOnline: fssync protocol mismatch (bad command word '%d')\n",
4623 Log("AskOnline: please make sure file server binaries are same version.\n");
4627 Log("AskOnline: request for fileserver to put volume online failed; trying again...\n");
4628 FSYNC_clientFinis();
4635 AskDelete(struct SalvInfo *salvinfo, VolumeId volumeId)
4640 for (i = 0; i < 3; i++) {
4641 memset(&res, 0, sizeof(res));
4642 code = FSYNC_VolOp(volumeId, salvinfo->fileSysPartition->name,
4643 FSYNC_VOL_DONE, FSYNC_SALVAGE, &res);
4645 if (code == SYNC_OK) {
4647 } else if (code == SYNC_DENIED) {
4648 Log("AskOnline: file server denied DONE request to volume %" AFS_VOLID_FMT " partition %s; trying again...\n", afs_printable_VolumeId_lu(volumeId), salvinfo->fileSysPartition->name);
4649 } else if (code == SYNC_BAD_COMMAND) {
4650 Log("AskOnline: fssync protocol mismatch (bad command word '%d')\n",
4653 #ifdef AFS_DEMAND_ATTACH_FS
4654 Log("AskOnline: please make sure dafileserver, davolserver, salvageserver and dasalvager binaries are same version.\n");
4656 Log("AskOnline: fileserver is DAFS but we are not.\n");
4659 #ifdef AFS_DEMAND_ATTACH_FS
4660 Log("AskOnline: fileserver is not DAFS but we are.\n");
4662 Log("AskOnline: please make sure fileserver, volserver and salvager binaries are same version.\n");
4666 } else if (code == SYNC_FAILED &&
4667 (res.hdr.reason == FSYNC_UNKNOWN_VOLID ||
4668 res.hdr.reason == FSYNC_WRONG_PART)) {
4669 /* volume is already effectively 'deleted' */
4673 Log("AskOnline: request for fileserver to delete volume failed; trying again...\n");
4674 FSYNC_clientFinis();
4681 CopyInode(Device device, Inode inode1, Inode inode2, int rwvolume)
4683 /* Volume parameter is passed in case iopen is upgraded in future to
4684 * require a volume Id to be passed
4687 IHandle_t *srcH, *destH;
4688 FdHandle_t *srcFdP, *destFdP;
4690 afs_foff_t size = 0;
4692 IH_INIT(srcH, device, rwvolume, inode1);
4693 srcFdP = IH_OPEN(srcH);
4694 opr_Assert(srcFdP != NULL);
4695 IH_INIT(destH, device, rwvolume, inode2);
4696 destFdP = IH_OPEN(destH);
4697 while ((nBytes = FDH_PREAD(srcFdP, buf, sizeof(buf), size)) > 0) {
4698 opr_Verify(FDH_PWRITE(destFdP, buf, nBytes, size) == nBytes);
4701 opr_Assert(nBytes == 0);
4702 FDH_REALLYCLOSE(srcFdP);
4703 FDH_REALLYCLOSE(destFdP);
4710 PrintInodeList(struct SalvInfo *salvinfo)
4712 struct ViceInodeInfo *ip;
4713 struct ViceInodeInfo *buf;
4716 afs_sfsize_t st_size;
4718 st_size = OS_SIZE(salvinfo->inodeFd);
4719 opr_Assert(st_size >= 0);
4720 buf = malloc(st_size);
4721 opr_Assert(buf != NULL);
4722 nInodes = st_size / sizeof(struct ViceInodeInfo);
4723 opr_Verify(OS_READ(salvinfo->inodeFd, buf, st_size) == st_size);
4724 for (ip = buf; nInodes--; ip++) {
4725 Log("Inode:%s, linkCount=%d, size=%#llx, p=(%" AFS_VOLID_FMT ",%u,%u,%u)\n", /* VolumeId in param */
4726 PrintInode(stmp, ip->inodeNumber), ip->linkCount,
4727 (afs_uintmax_t) ip->byteCount,
4728 afs_printable_VolumeId_lu(ip->u.param[0]), ip->u.param[1],
4729 ip->u.param[2], ip->u.param[3]);
4735 PrintInodeSummary(struct SalvInfo *salvinfo)
4738 struct InodeSummary *isp;
4740 for (i = 0; i < salvinfo->nVolumesInInodeFile; i++) {
4741 isp = &salvinfo->inodeSummary[i];
4742 Log("VID:%" AFS_VOLID_FMT ", RW:%" AFS_VOLID_FMT ", index:%d, nInodes:%d, nSpecialInodes:%d, maxUniquifier:%u, volSummary\n", afs_printable_VolumeId_lu(isp->volumeId), afs_printable_VolumeId_lu(isp->RWvolumeId), isp->index, isp->nInodes, isp->nSpecialInodes, isp->maxUniquifier);
4752 opr_Assert(0); /* Fork is never executed in the NT code path */
4756 #ifdef AFS_DEMAND_ATTACH_FS
4757 if ((f == 0) && (programType == salvageServer)) {
4758 /* we are a salvageserver child */
4759 #ifdef FSSYNC_BUILD_CLIENT
4760 VChildProcReconnectFS_r();
4762 #ifdef SALVSYNC_BUILD_CLIENT
4766 #endif /* AFS_DEMAND_ATTACH_FS */
4767 #endif /* !AFS_NT40_ENV */
4774 #ifdef AFS_DEMAND_ATTACH_FS
4775 if (programType == salvageServer) {
4776 /* release all volume locks before closing down our SYNC channels.
4777 * the fileserver may try to online volumes we have checked out when
4778 * we close down FSSYNC, so we should make sure we don't have those
4779 * volumes locked when it does */
4780 struct DiskPartition64 *dp;
4782 for (i = 0; i <= VOLMAXPARTS; i++) {
4783 dp = VGetPartitionById(i, 0);
4785 VLockFileReinit(&dp->volLockFile);
4788 # ifdef SALVSYNC_BUILD_CLIENT
4791 # ifdef FSSYNC_BUILD_CLIENT
4795 #endif /* AFS_DEMAND_ATTACH_FS */
4798 if (main_thread != pthread_self())
4799 pthread_exit((void *)code);
4820 pid = wait(&status);
4821 opr_Assert(pid != -1);
4822 if (WCOREDUMP(status))
4823 Log("\"%s\" core dumped!\n", prog);
4824 if (WIFSIGNALED(status) != 0 || WEXITSTATUS(status) != 0)
4830 TimeStamp(char *buffer, size_t size, time_t clock, int precision)
4835 lt = localtime(&clock);
4837 nbytes = strftime(buffer, size, "%m/%d/%Y %H:%M:%S", lt);
4839 nbytes = strftime(buffer, size, "%m/%d/%Y %H:%M", lt);
4841 memset(buffer, 0, size);
4846 SalvageShowLog(void)
4852 if (ShowLog == 0 || ClientMode) {
4853 return; /* nothing to do */
4855 filename = strdup(GetLogFilename());
4856 opr_Assert(filename != NULL);
4859 logFile = afs_fopen(filename, "r");
4861 printf("Can't read %s, exiting\n", ShowLogFilename);
4863 while (fgets(line, sizeof(line), logFile))
4871 vLog(const char *format, va_list args)
4874 vFSLog(format, args);
4879 gettimeofday(&now, NULL);
4880 fprintf(stderr, "%s ", TimeStamp(buffer, sizeof(buffer), now.tv_sec, 1));
4881 vfprintf(stderr, format, args);
4887 Log(const char *format, ...)
4891 va_start(args, format);
4897 Abort(const char *format, ...)
4901 va_start(args, format);
4911 ToString(const char *s)
4915 opr_Assert(p != NULL);
4919 /* Remove the FORCESALVAGE file */
4921 RemoveTheForce(char *path)
4924 struct afs_stat_st force; /* so we can use afs_stat to find it */
4925 strcpy(target,path);
4926 strcat(target,"/FORCESALVAGE");
4927 if (!Testing && ForceSalvage) {
4928 if (afs_stat(target,&force) == 0) unlink(target);
4932 #ifndef AFS_AIX32_ENV
4934 * UseTheForceLuke - see if we can use the force
4937 UseTheForceLuke(char *path)
4939 struct afs_stat_st force;
4941 strcpy(target,path);
4942 strcat(target,"/FORCESALVAGE");
4944 return (afs_stat(target, &force) == 0);
4948 * UseTheForceLuke - see if we can use the force
4951 * The VRMIX fsck will not muck with the filesystem it is supposedly
4952 * fixing and create a "FORCESALVAGE" file (by design). Instead, we
4953 * muck directly with the root inode, which is within the normal
4955 * ListViceInodes() has a side effect of setting ForceSalvage if
4956 * it detects a need, based on root inode examination.
4959 UseTheForceLuke(char *path)
4962 return 0; /* sorry OB1 */
4967 /* NT support routines */
4969 static char execpathname[MAX_PATH];
4971 nt_SalvagePartition(char *partName, int jobn)
4976 if (!*execpathname) {
4977 n = GetModuleFileName(NULL, execpathname, MAX_PATH - 1);
4978 if (!n || n == 1023)
4981 job.cj_magic = SALVAGER_MAGIC;
4982 job.cj_number = jobn;
4983 (void)strcpy(job.cj_part, partName);
4984 pid = (int)spawnprocveb(execpathname, save_args, NULL, &job, sizeof(job));
4989 nt_SetupPartitionSalvage(void *datap, int len)
4991 childJob_t *jobp = (childJob_t *) datap;
4994 if (len != sizeof(childJob_t))
4996 if (jobp->cj_magic != SALVAGER_MAGIC)
5001 if (asprintf(&logname, "%s.%d", AFSDIR_SERVER_SLVGLOG_FILEPATH,
5002 myjob.cj_number) < 0)
5011 #endif /* AFS_NT40_ENV */