2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1980, 1986, 1993
5 * The Regents of the University of California. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 static const char sccsid[] = "@(#)utilities.c 8.6 (Berkeley) 5/19/95";
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
40 #include <sys/param.h>
42 #include <sys/types.h>
43 #include <sys/sysctl.h>
45 #include <sys/disklabel.h>
46 #include <sys/ioctl.h>
49 #include <ufs/ufs/dinode.h>
50 #include <ufs/ufs/dir.h>
51 #include <ufs/ffs/fs.h>
69 static struct bufarea *allocbuf(const char *);
70 static void cg_write(struct bufarea *);
71 static void slowio_start(void);
72 static void slowio_end(void);
73 static void printIOstats(void);
74 static void prtbuf(const char *, struct bufarea *);
76 static long diskreads, totaldiskreads, totalreads; /* Disk cache statistics */
77 static struct timespec startpass, finishpass;
78 struct timeval slowio_starttime;
79 int slowio_delay_usec = 10000; /* Initial IO delay for background fsck */
81 static struct bufarea cgblk; /* backup buffer for cylinder group blocks */
82 static TAILQ_HEAD(bufqueue, bufarea) bufqueuehd; /* head of buffer cache LRU */
83 static LIST_HEAD(bufhash, bufarea) bufhashhd[HASHSIZE]; /* buffer hash list */
84 static int numbufs; /* size of buffer cache */
85 static int cachelookups; /* number of cache lookups */
86 static int cachereads; /* number of cache reads */
87 static int flushtries; /* number of tries to reclaim memory */
89 char *buftype[BT_NUMBUFTYPES] = BT_NAMES;
94 diskreads = totaldiskreads = totalreads = 0;
95 bzero(&startpass, sizeof(struct timespec));
96 bzero(&finishpass, sizeof(struct timespec));
97 bzero(&slowio_starttime, sizeof(struct timeval));
98 slowio_delay_usec = 10000;
104 ftypeok(union dinode *dp)
106 switch (DIP(dp, di_mode) & IFMT) {
119 printf("bad file type 0%o\n", DIP(dp, di_mode));
125 reply(const char *question)
131 pfatal("INTERNAL ERROR: GOT TO reply()");
132 persevere = !strcmp(question, "CONTINUE");
134 if (!persevere && (nflag || (fswritefd < 0 && bkgrdflag == 0))) {
135 printf("%s? no\n\n", question);
139 if (yflag || (persevere && nflag)) {
140 printf("%s? yes\n\n", question);
144 printf("%s? [yn] ", question);
145 (void) fflush(stdout);
147 while (c != '\n' && getc(stdin) != '\n') {
153 } while (c != 'y' && c != 'Y' && c != 'n' && c != 'N');
155 if (c == 'y' || c == 'Y')
162 * Look up state information for an inode.
167 static struct inostat unallocated = { USTATE, 0, 0 };
168 struct inostatlist *ilp;
172 errx(EEXIT, "inoinfo: inumber %ju out of range",
174 ilp = &inostathead[inum / sblock.fs_ipg];
175 iloff = inum % sblock.fs_ipg;
176 if (iloff >= ilp->il_numalloced)
177 return (&unallocated);
178 return (&ilp->il_stat[iloff]);
182 * Malloc buffers and set up cache.
189 if ((cgblk.b_un.b_buf = Malloc((unsigned int)sblock.fs_bsize)) == NULL)
190 errx(EEXIT, "Initial malloc(%d) failed", sblock.fs_bsize);
191 initbarea(&cgblk, BT_CYLGRP);
192 numbufs = cachelookups = cachereads = 0;
193 TAILQ_INIT(&bufqueuehd);
194 for (i = 0; i < HASHSIZE; i++)
195 LIST_INIT(&bufhashhd[i]);
196 for (i = 0; i < BT_NUMBUFTYPES; i++) {
197 readtime[i].tv_sec = totalreadtime[i].tv_sec = 0;
198 readtime[i].tv_nsec = totalreadtime[i].tv_nsec = 0;
199 readcnt[i] = totalreadcnt[i] = 0;
203 static struct bufarea *
204 allocbuf(const char *failreason)
209 bp = (struct bufarea *)Malloc(sizeof(struct bufarea));
210 bufp = Malloc((unsigned int)sblock.fs_bsize);
211 if (bp == NULL || bufp == NULL) {
212 errx(EEXIT, "%s", failreason);
216 bp->b_un.b_buf = bufp;
217 TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
218 initbarea(bp, BT_UNKNOWN);
223 * Manage cylinder group buffers.
225 * Use getblk() here rather than cgget() because the cylinder group
226 * may be corrupted but we want it anyway so we can fix it.
228 static struct bufarea *cgbufs; /* header for cylinder group cache */
229 static int flushtries; /* number of tries to reclaim memory */
234 struct bufarea *cgbp;
237 if ((unsigned) cg >= sblock.fs_ncg)
238 errx(EEXIT, "cglookup: out of range cylinder group %d", cg);
239 if (cgbufs == NULL) {
240 cgbufs = calloc(sblock.fs_ncg, sizeof(struct bufarea));
242 errx(EEXIT, "Cannot allocate cylinder group buffers");
245 if (cgbp->b_un.b_cg != NULL)
249 cgp = Malloc((unsigned int)sblock.fs_cgsize);
252 errx(EEXIT,"Ran out of memory during journal recovery");
253 flush(fswritefd, &cgblk);
254 getblk(&cgblk, cgtod(&sblock, cg), sblock.fs_cgsize);
257 cgbp->b_un.b_cg = cgp;
258 initbarea(cgbp, BT_CYLGRP);
259 getblk(cgbp, cgtod(&sblock, cg), sblock.fs_cgsize);
264 * Mark a cylinder group buffer as dirty.
265 * Update its check-hash if they are enabled.
268 cgdirty(struct bufarea *cgbp)
272 cg = cgbp->b_un.b_cg;
273 if ((sblock.fs_metackhash & CK_CYLGRP) != 0) {
276 calculate_crc32c(~0L, (void *)cg, sblock.fs_cgsize);
282 * Attempt to flush a cylinder group cache entry.
283 * Return whether the flush was successful.
288 struct bufarea *cgbp;
290 if (sujrecovery || flushtries == sblock.fs_ncg || cgbufs == NULL)
292 cgbp = &cgbufs[flushtries++];
293 if (cgbp->b_un.b_cg == NULL)
295 flush(fswritefd, cgbp);
296 free(cgbp->b_un.b_buf);
297 cgbp->b_un.b_buf = NULL;
302 * Manage a cache of directory blocks.
305 getdatablk(ufs2_daddr_t blkno, long size, int type)
308 struct bufhash *bhdp;
311 /* If out of range, return empty buffer with b_err == -1 */
312 if (type != BT_INODES && chkrange(blkno, size / sblock.fs_fsize)) {
316 bhdp = &bufhashhd[HASH(blkno)];
317 LIST_FOREACH(bp, bhdp, b_hash)
318 if (bp->b_bno == fsbtodb(&sblock, blkno)) {
319 if (debug && bp->b_size != size) {
320 prtbuf("getdatablk: size mismatch", bp);
321 pfatal("getdatablk: b_size %d != size %ld\n",
327 * Move long-term busy buffer back to the front of the LRU so we
328 * do not endless inspect them for recycling.
330 bp = TAILQ_LAST(&bufqueuehd, bufqueue);
331 if (bp != NULL && bp->b_refcnt != 0) {
332 TAILQ_REMOVE(&bufqueuehd, bp, b_list);
333 TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
336 * Allocate up to the minimum number of buffers before
337 * considering recycling any of them.
339 if (size > sblock.fs_bsize)
340 errx(EEXIT, "Excessive buffer size %ld > %d\n", size,
342 if (numbufs < MINBUFS) {
343 bp = allocbuf("cannot create minimal buffer pool");
344 } else if (sujrecovery) {
346 * SUJ recovery does not want anything written until it
347 * has successfully completed (so it can fail back to
348 * full fsck). Thus, we can only recycle clean buffers.
350 TAILQ_FOREACH_REVERSE(bp, &bufqueuehd, bufqueue, b_list)
351 if ((bp->b_flags & B_DIRTY) == 0 && bp->b_refcnt == 0)
354 bp = allocbuf("Ran out of memory during "
357 LIST_REMOVE(bp, b_hash);
360 * Recycle oldest non-busy buffer.
362 TAILQ_FOREACH_REVERSE(bp, &bufqueuehd, bufqueue, b_list)
363 if (bp->b_refcnt == 0)
366 bp = allocbuf("Ran out of memory for buffers");
368 LIST_REMOVE(bp, b_hash);
370 flush(fswritefd, bp);
372 LIST_INSERT_HEAD(bhdp, bp, b_hash);
373 getblk(bp, blkno, size);
377 if (debug && bp->b_type != type) {
378 printf("getdatablk: buffer type changed to %s",
382 TAILQ_REMOVE(&bufqueuehd, bp, b_list);
383 TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
390 getblk(struct bufarea *bp, ufs2_daddr_t blk, long size)
393 struct timespec start, finish;
395 dblk = fsbtodb(&sblock, blk);
396 if (bp->b_bno == dblk) {
400 readcnt[bp->b_type]++;
401 clock_gettime(CLOCK_REALTIME_PRECISE, &start);
403 if (bp->b_type != BT_EMPTY)
405 blread(fsreadfd, bp->b_un.b_buf, dblk, size);
409 clock_gettime(CLOCK_REALTIME_PRECISE, &finish);
410 timespecsub(&finish, &start, &finish);
411 timespecadd(&readtime[bp->b_type], &finish,
412 &readtime[bp->b_type]);
420 brelse(struct bufarea *bp)
423 if (bp->b_refcnt <= 0)
424 prtbuf("brelse: buffer with negative reference count", bp);
429 flush(int fd, struct bufarea *bp)
433 if ((bp->b_flags & B_DIRTY) == 0)
435 bp->b_flags &= ~B_DIRTY;
437 pfatal("WRITING IN READ_ONLY MODE.\n");
441 pfatal("WRITING %sZERO'ED BLOCK %lld TO DISK\n",
442 (bp->b_errs == bp->b_size / dev_bsize) ? "" : "PARTIALLY ",
443 (long long)bp->b_bno);
446 * Write using the appropriate function.
448 switch (bp->b_type) {
451 pfatal("BUFFER %p DOES NOT MATCH SBLK %p\n",
453 if (sbput(fd, bp->b_un.b_fs, 0) == 0)
459 if (cgput(fswritefd, &sblock, bp->b_un.b_cg) == 0)
463 if (debug && sblock.fs_magic == FS_UFS2_MAGIC) {
464 struct ufs2_dinode *dp = bp->b_un.b_dinode2;
467 for (i = 0; i < INOPB(&sblock); dp++, i++) {
468 if (ffs_verify_dinode_ckhash(&sblock, dp) == 0)
470 pwarn("flush: INODE CHECK-HASH FAILED");
472 ip.i_dp = (union dinode *)dp;
473 ip.i_number = bp->b_index + i;
475 if (preen || reply("FIX") != 0) {
477 printf(" (FIXED)\n");
478 ffs_update_dinode_ckhash(&sblock, dp);
485 blwrite(fd, bp->b_un.b_buf, bp->b_bno, bp->b_size);
491 * Journaled soft updates does not maintain cylinder group summary
492 * information during cleanup, so this routine recalculates the summary
493 * information and updates the superblock summary in preparation for
494 * writing out the cylinder group.
497 cg_write(struct bufarea *bp)
499 ufs1_daddr_t fragno, cgbno, maxbno;
506 * Fix the frag and cluster summary.
509 cgp->cg_cs.cs_nbfree = 0;
510 cgp->cg_cs.cs_nffree = 0;
511 bzero(&cgp->cg_frsum, sizeof(cgp->cg_frsum));
512 maxbno = fragstoblks(&sblock, sblock.fs_fpg);
513 if (sblock.fs_contigsumsize > 0) {
514 for (i = 1; i <= sblock.fs_contigsumsize; i++)
515 cg_clustersum(cgp)[i] = 0;
516 bzero(cg_clustersfree(cgp), howmany(maxbno, CHAR_BIT));
518 blksfree = cg_blksfree(cgp);
519 for (cgbno = 0; cgbno < maxbno; cgbno++) {
520 if (ffs_isfreeblock(&sblock, blksfree, cgbno))
522 if (ffs_isblock(&sblock, blksfree, cgbno)) {
523 ffs_clusteracct(&sblock, cgp, cgbno, 1);
524 cgp->cg_cs.cs_nbfree++;
527 fragno = blkstofrags(&sblock, cgbno);
528 blk = blkmap(&sblock, blksfree, fragno);
529 ffs_fragacct(&sblock, blk, cgp->cg_frsum, 1);
530 for (i = 0; i < sblock.fs_frag; i++)
531 if (isset(blksfree, fragno + i))
532 cgp->cg_cs.cs_nffree++;
535 * Update the superblock cg summary from our now correct values
536 * before writing the block.
538 sblock.fs_cs(&sblock, cgp->cg_cgx) = cgp->cg_cs;
542 rwerror(const char *mesg, ufs2_daddr_t blk)
549 pfatal("CANNOT %s: %ld", mesg, (long)blk);
550 if (reply("CONTINUE") == 0)
555 ckfini(int markclean)
557 struct bufarea *bp, *nbp;
558 struct inoinfo *inp, *ninp;
559 int ofsmodified, cnt, cg, i;
563 if ((!(sblock.fs_flags & FS_UNCLEAN)) != markclean) {
564 cmd.value = FS_UNCLEAN;
565 cmd.size = markclean ? -1 : 1;
566 if (sysctlbyname("vfs.ffs.setflags", 0, 0,
567 &cmd, sizeof cmd) == -1)
568 pwarn("CANNOT SET FILE SYSTEM DIRTY FLAG\n");
570 printf("\n***** FILE SYSTEM MARKED %s *****\n",
571 markclean ? "CLEAN" : "DIRTY");
575 } else if (!preen && !markclean) {
576 printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
581 if (debug && cachelookups > 0)
582 printf("cache with %d buffers missed %d of %d (%d%%)\n",
583 numbufs, cachereads, cachelookups,
584 (int)(cachereads * 100 / cachelookups));
586 (void)close(fsreadfd);
590 * To remain idempotent with partial truncations the buffers
591 * must be flushed in this order:
592 * 1) cylinder groups (bitmaps)
593 * 2) indirect, directory, external attribute, and data blocks
596 * This ordering preserves access to the modified pointers
597 * until they are freed.
599 /* Step 1: cylinder groups */
601 printf("Flush Cylinder groups\n");
602 if (cgbufs != NULL) {
603 for (cnt = 0; cnt < sblock.fs_ncg; cnt++) {
604 if (cgbufs[cnt].b_un.b_cg == NULL)
606 flush(fswritefd, &cgbufs[cnt]);
607 free(cgbufs[cnt].b_un.b_cg);
612 flush(fswritefd, &cgblk);
613 free(cgblk.b_un.b_buf);
614 cgblk.b_un.b_buf = NULL;
616 /* Step 2: indirect, directory, external attribute, and data blocks */
618 printf("Flush indirect, directory, external attribute, "
619 "and data blocks\n");
620 if (pdirbp != NULL) {
624 TAILQ_FOREACH_REVERSE_SAFE(bp, &bufqueuehd, bufqueue, b_list, nbp) {
625 switch (bp->b_type) {
626 /* These should not be in the buffer cache list */
631 prtbuf("ckfini: improper buffer type on cache list",bp);
633 /* These are the ones to flush in this step */
636 pfatal("Unused BT_EMPTY buffer for block %jd\n",
637 (intmax_t)bp->b_bno);
646 /* These are the ones to flush in the next step */
650 if (debug && bp->b_refcnt != 0) {
651 prtbuf("ckfini: clearing in-use buffer", bp);
652 pfatal("ckfini: clearing in-use buffer\n");
654 TAILQ_REMOVE(&bufqueuehd, bp, b_list);
656 flush(fswritefd, bp);
657 free(bp->b_un.b_buf);
660 /* Step 3: inode blocks */
662 printf("Flush inode blocks\n");
663 if (icachebp != NULL) {
667 TAILQ_FOREACH_REVERSE_SAFE(bp, &bufqueuehd, bufqueue, b_list, nbp) {
668 if (debug && bp->b_refcnt != 0) {
669 prtbuf("ckfini: clearing in-use buffer", bp);
670 pfatal("ckfini: clearing in-use buffer\n");
672 TAILQ_REMOVE(&bufqueuehd, bp, b_list);
674 flush(fswritefd, bp);
675 free(bp->b_un.b_buf);
679 errx(EEXIT, "panic: lost %d buffers", numbufs - cnt);
680 /* Step 4: superblock */
682 printf("Flush the superblock\n");
683 flush(fswritefd, &sblk);
684 if (havesb && cursnapshot == 0 && sblock.fs_magic == FS_UFS2_MAGIC &&
685 sblk.b_bno != sblock.fs_sblockloc / dev_bsize &&
686 !preen && reply("UPDATE STANDARD SUPERBLOCK")) {
687 /* Change the write destination to standard superblock */
688 sblock.fs_sblockactualloc = sblock.fs_sblockloc;
689 sblk.b_bno = sblock.fs_sblockloc / dev_bsize;
691 flush(fswritefd, &sblk);
693 if (cursnapshot == 0 && sblock.fs_clean != markclean) {
694 if ((sblock.fs_clean = markclean) != 0) {
695 sblock.fs_flags &= ~(FS_UNCLEAN | FS_NEEDSFSCK);
696 sblock.fs_pendingblocks = 0;
697 sblock.fs_pendinginodes = 0;
700 ofsmodified = fsmodified;
701 flush(fswritefd, &sblk);
702 fsmodified = ofsmodified;
704 printf("\n***** FILE SYSTEM MARKED %s *****\n",
705 markclean ? "CLEAN" : "DIRTY");
711 printf("\n***** FILE SYSTEM IS CLEAN *****\n");
713 printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
718 * Free allocated tracking structures.
720 if (blockmap != NULL)
723 if (inostathead != NULL) {
724 for (cg = 0; cg < sblock.fs_ncg; cg++)
725 if (inostathead[cg].il_stat != NULL)
726 free((char *)inostathead[cg].il_stat);
733 if (inphead != NULL) {
734 for (i = 0; i < dirhash; i++) {
735 for (inp = inphead[i]; inp != NULL; inp = ninp) {
736 ninp = inp->i_nexthash;
744 (void)close(fsreadfd);
745 (void)close(fswritefd);
749 * Print out I/O statistics.
758 if (diskreads == 0) {
759 printf("%s: no I/O\n\n", what);
762 if (startpass.tv_sec == 0)
763 startpass = startprog;
764 printf("%s: I/O statistics\n", what);
766 totaldiskreads += diskreads;
768 for (i = 0; i < BT_NUMBUFTYPES; i++) {
769 timespecadd(&totalreadtime[i], &readtime[i], &totalreadtime[i]);
770 totalreadcnt[i] += readcnt[i];
771 readtime[i].tv_sec = readtime[i].tv_nsec = 0;
774 clock_gettime(CLOCK_REALTIME_PRECISE, &startpass);
784 printf("Final I/O statistics\n");
785 totaldiskreads += diskreads;
786 diskreads = totaldiskreads;
787 startpass = startprog;
788 for (i = 0; i < BT_NUMBUFTYPES; i++) {
789 timespecadd(&totalreadtime[i], &readtime[i], &totalreadtime[i]);
790 totalreadcnt[i] += readcnt[i];
791 readtime[i] = totalreadtime[i];
792 readcnt[i] = totalreadcnt[i];
797 static void printIOstats(void)
799 long long msec, totalmsec;
802 clock_gettime(CLOCK_REALTIME_PRECISE, &finishpass);
803 timespecsub(&finishpass, &startpass, &finishpass);
804 printf("Running time: %jd.%03ld sec\n",
805 (intmax_t)finishpass.tv_sec, finishpass.tv_nsec / 1000000);
806 printf("buffer reads by type:\n");
807 for (totalmsec = 0, i = 0; i < BT_NUMBUFTYPES; i++)
808 totalmsec += readtime[i].tv_sec * 1000 +
809 readtime[i].tv_nsec / 1000000;
812 for (i = 0; i < BT_NUMBUFTYPES; i++) {
816 readtime[i].tv_sec * 1000 + readtime[i].tv_nsec / 1000000;
817 printf("%21s:%8ld %2ld.%ld%% %4jd.%03ld sec %2lld.%lld%%\n",
818 buftype[i], readcnt[i], readcnt[i] * 100 / diskreads,
819 (readcnt[i] * 1000 / diskreads) % 10,
820 (intmax_t)readtime[i].tv_sec, readtime[i].tv_nsec / 1000000,
821 msec * 100 / totalmsec, (msec * 1000 / totalmsec) % 10);
827 blread(int fd, char *buf, ufs2_daddr_t blk, long size)
839 if (pread(fd, buf, (int)size, offset) == size) {
846 * This is handled specially here instead of in rwerror because
847 * rwerror is used for all sorts of errors, not just true read/write
848 * errors. It should be refactored and fixed.
851 pfatal("CANNOT READ_BLK: %ld", (long)blk);
852 errx(EEXIT, "ABORTING DUE TO READ ERRORS");
854 rwerror("READ BLK", blk);
857 memset(buf, 0, (size_t)size);
858 printf("THE FOLLOWING DISK SECTORS COULD NOT BE READ:");
859 for (cp = buf, i = 0; i < size; i += secsize, cp += secsize) {
860 if (pread(fd, cp, (int)secsize, offset + i) != secsize) {
861 if (secsize != dev_bsize && dev_bsize != 1)
862 printf(" %jd (%jd),",
863 (intmax_t)(blk * dev_bsize + i) / secsize,
864 (intmax_t)blk + i / dev_bsize);
866 printf(" %jd,", (intmax_t)blk + i / dev_bsize);
877 blwrite(int fd, char *buf, ufs2_daddr_t blk, ssize_t size)
887 if (pwrite(fd, buf, size, offset) == size) {
892 rwerror("WRITE BLK", blk);
893 printf("THE FOLLOWING SECTORS COULD NOT BE WRITTEN:");
894 for (cp = buf, i = 0; i < size; i += dev_bsize, cp += dev_bsize)
895 if (pwrite(fd, cp, dev_bsize, offset + i) != dev_bsize)
896 printf(" %jd,", (intmax_t)blk + i / dev_bsize);
902 blerase(int fd, ufs2_daddr_t blk, long size)
908 ioarg[0] = blk * dev_bsize;
910 ioctl(fd, DIOCGDELETE, ioarg);
911 /* we don't really care if we succeed or not */
916 * Fill a contiguous region with all-zeroes. Note ZEROBUFSIZE is by
917 * definition a multiple of dev_bsize.
920 blzero(int fd, ufs2_daddr_t blk, long size)
928 zero = calloc(ZEROBUFSIZE, 1);
930 errx(EEXIT, "cannot allocate buffer pool");
932 offset = blk * dev_bsize;
933 if (lseek(fd, offset, 0) < 0)
934 rwerror("SEEK BLK", blk);
936 len = MIN(ZEROBUFSIZE, size);
937 if (write(fd, zero, len) != len)
938 rwerror("WRITE BLK", blk);
939 blk += len / dev_bsize;
945 * Verify cylinder group's magic number and other parameters. If the
946 * test fails, offer an option to rebuild the whole cylinder group.
949 check_cgmagic(int cg, struct bufarea *cgbp, int request_rebuild)
951 struct cg *cgp = cgbp->b_un.b_cg;
952 uint32_t cghash, calchash;
953 static int prevfailcg = -1;
956 * Extended cylinder group checks.
958 calchash = cgp->cg_ckhash;
959 if ((sblock.fs_metackhash & CK_CYLGRP) != 0 &&
960 (ckhashadd & CK_CYLGRP) == 0) {
961 cghash = cgp->cg_ckhash;
963 calchash = calculate_crc32c(~0L, (void *)cgp, sblock.fs_cgsize);
964 cgp->cg_ckhash = cghash;
966 if (cgp->cg_ckhash == calchash &&
969 ((sblock.fs_magic == FS_UFS1_MAGIC &&
970 cgp->cg_old_niblk == sblock.fs_ipg &&
971 cgp->cg_ndblk <= sblock.fs_fpg &&
972 cgp->cg_old_ncyl <= sblock.fs_old_cpg) ||
973 (sblock.fs_magic == FS_UFS2_MAGIC &&
974 cgp->cg_niblk == sblock.fs_ipg &&
975 cgp->cg_ndblk <= sblock.fs_fpg &&
976 cgp->cg_initediblk <= sblock.fs_ipg))) {
979 if (prevfailcg == cg)
982 pfatal("CYLINDER GROUP %d: INTEGRITY CHECK FAILED", cg);
983 if (!request_rebuild) {
987 if (!reply("REBUILD CYLINDER GROUP")) {
988 printf("YOU WILL NEED TO RERUN FSCK.\n");
993 * Zero out the cylinder group and then initialize critical fields.
994 * Bit maps and summaries will be recalculated by later passes.
996 memset(cgp, 0, (size_t)sblock.fs_cgsize);
997 cgp->cg_magic = CG_MAGIC;
999 cgp->cg_niblk = sblock.fs_ipg;
1000 cgp->cg_initediblk = MIN(sblock.fs_ipg, 2 * INOPB(&sblock));
1001 if (cgbase(&sblock, cg) + sblock.fs_fpg < sblock.fs_size)
1002 cgp->cg_ndblk = sblock.fs_fpg;
1004 cgp->cg_ndblk = sblock.fs_size - cgbase(&sblock, cg);
1005 cgp->cg_iusedoff = &cgp->cg_space[0] - (u_char *)(&cgp->cg_firstfield);
1006 if (sblock.fs_magic == FS_UFS1_MAGIC) {
1008 cgp->cg_initediblk = 0;
1009 cgp->cg_old_ncyl = sblock.fs_old_cpg;
1010 cgp->cg_old_niblk = sblock.fs_ipg;
1011 cgp->cg_old_btotoff = cgp->cg_iusedoff;
1012 cgp->cg_old_boff = cgp->cg_old_btotoff +
1013 sblock.fs_old_cpg * sizeof(int32_t);
1014 cgp->cg_iusedoff = cgp->cg_old_boff +
1015 sblock.fs_old_cpg * sizeof(u_int16_t);
1017 cgp->cg_freeoff = cgp->cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
1018 cgp->cg_nextfreeoff = cgp->cg_freeoff + howmany(sblock.fs_fpg,CHAR_BIT);
1019 if (sblock.fs_contigsumsize > 0) {
1020 cgp->cg_nclusterblks = cgp->cg_ndblk / sblock.fs_frag;
1021 cgp->cg_clustersumoff =
1022 roundup(cgp->cg_nextfreeoff, sizeof(u_int32_t));
1023 cgp->cg_clustersumoff -= sizeof(u_int32_t);
1024 cgp->cg_clusteroff = cgp->cg_clustersumoff +
1025 (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
1026 cgp->cg_nextfreeoff = cgp->cg_clusteroff +
1027 howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
1034 * allocate a data block with the specified number of fragments
1037 allocblk(long frags)
1039 int i, j, k, cg, baseblk;
1040 struct bufarea *cgbp;
1043 if (frags <= 0 || frags > sblock.fs_frag)
1045 for (i = 0; i < maxfsblock - sblock.fs_frag; i += sblock.fs_frag) {
1046 for (j = 0; j <= sblock.fs_frag - frags; j++) {
1047 if (testbmap(i + j))
1049 for (k = 1; k < frags; k++)
1050 if (testbmap(i + j + k))
1056 cg = dtog(&sblock, i + j);
1057 cgbp = cglookup(cg);
1058 cgp = cgbp->b_un.b_cg;
1059 if (!check_cgmagic(cg, cgbp, 0)) {
1060 i = (cg + 1) * sblock.fs_fpg - sblock.fs_frag;
1063 baseblk = dtogd(&sblock, i + j);
1064 for (k = 0; k < frags; k++) {
1066 clrbit(cg_blksfree(cgp), baseblk + k);
1069 if (frags == sblock.fs_frag)
1070 cgp->cg_cs.cs_nbfree--;
1072 cgp->cg_cs.cs_nffree -= frags;
1081 * Slow down IO so as to leave some disk bandwidth for other processes
1087 /* Delay one in every 8 operations */
1088 slowio_pollcnt = (slowio_pollcnt + 1) & 7;
1089 if (slowio_pollcnt == 0) {
1090 gettimeofday(&slowio_starttime, NULL);
1100 if (slowio_pollcnt != 0)
1103 /* Update the slowdown interval. */
1104 gettimeofday(&tv, NULL);
1105 delay_usec = (tv.tv_sec - slowio_starttime.tv_sec) * 1000000 +
1106 (tv.tv_usec - slowio_starttime.tv_usec);
1107 if (delay_usec < 64)
1109 if (delay_usec > 2500000)
1110 delay_usec = 2500000;
1111 slowio_delay_usec = (slowio_delay_usec * 63 + delay_usec) >> 6;
1112 /* delay by 8 times the average IO delay */
1113 if (slowio_delay_usec > 64)
1114 usleep(slowio_delay_usec * 8);
1121 getpathname(char *namebuf, ino_t curdir, ino_t ino)
1126 struct inodesc idesc;
1127 static int busy = 0;
1129 if (curdir == ino && ino == UFS_ROOTINO) {
1130 (void)strcpy(namebuf, "/");
1133 if (busy || !INO_IS_DVALID(curdir)) {
1134 (void)strcpy(namebuf, "?");
1138 memset(&idesc, 0, sizeof(struct inodesc));
1139 idesc.id_type = DATA;
1140 idesc.id_fix = IGNORE;
1141 cp = &namebuf[MAXPATHLEN - 1];
1143 if (curdir != ino) {
1144 idesc.id_parent = curdir;
1147 while (ino != UFS_ROOTINO) {
1148 idesc.id_number = ino;
1149 idesc.id_func = findino;
1150 idesc.id_name = strdup("..");
1152 if ((ckinode(ip.i_dp, &idesc) & FOUND) == 0) {
1158 idesc.id_number = idesc.id_parent;
1159 idesc.id_parent = ino;
1160 idesc.id_func = findname;
1161 idesc.id_name = namebuf;
1162 ginode(idesc.id_number, &ip);
1163 if ((ckinode(ip.i_dp, &idesc) & FOUND) == 0) {
1168 len = strlen(namebuf);
1170 memmove(cp, namebuf, (size_t)len);
1172 if (cp < &namebuf[UFS_MAXNAMLEN])
1174 ino = idesc.id_number;
1177 if (ino != UFS_ROOTINO)
1179 memmove(namebuf, cp, (size_t)(&namebuf[MAXPATHLEN] - cp));
1183 catch(int sig __unused)
1191 * When preening, allow a single quit to signal
1192 * a special exit after file system checks complete
1193 * so that reboot sequence may be interrupted.
1196 catchquit(int sig __unused)
1198 printf("returning to single-user after file system check\n");
1200 (void)signal(SIGQUIT, SIG_DFL);
1204 * determine whether an inode should be fixed.
1207 dofix(struct inodesc *idesc, const char *msg)
1210 switch (idesc->id_fix) {
1213 if (idesc->id_type == DATA)
1214 direrror(idesc->id_number, msg);
1218 printf(" (SALVAGED)\n");
1219 idesc->id_fix = FIX;
1222 if (reply("SALVAGE") == 0) {
1223 idesc->id_fix = NOFIX;
1226 idesc->id_fix = FIX;
1237 errx(EEXIT, "UNKNOWN INODESC FIX MODE %d", idesc->id_fix);
1246 * Print details about a buffer.
1249 prtbuf(const char *msg, struct bufarea *bp)
1252 printf("%s: bp %p, type %s, bno %jd, size %d, refcnt %d, flags %s, "
1253 "index %jd\n", msg, bp, BT_BUFTYPE(bp->b_type), (intmax_t) bp->b_bno,
1254 bp->b_size, bp->b_refcnt, bp->b_flags & B_DIRTY ? "dirty" : "clean",
1255 (intmax_t) bp->b_index);
1259 * An unexpected inconsistency occurred.
1260 * Die if preening or file system is running with soft dependency protocol,
1261 * otherwise just print message and continue.
1264 pfatal(const char *fmt, ...)
1269 (void)vfprintf(stdout, fmt, ap);
1272 (void)fprintf(stdout,
1273 "\nUNEXPECTED SOFT UPDATE INCONSISTENCY\n");
1275 * Force foreground fsck to clean up inconsistency.
1278 cmd.value = FS_NEEDSFSCK;
1280 if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1281 &cmd, sizeof cmd) == -1)
1282 pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1283 fprintf(stdout, "CANNOT RUN IN BACKGROUND\n");
1289 if (cdevname == NULL)
1290 cdevname = strdup("fsck");
1291 (void)fprintf(stdout, "%s: ", cdevname);
1292 (void)vfprintf(stdout, fmt, ap);
1293 (void)fprintf(stdout,
1294 "\n%s: UNEXPECTED%sINCONSISTENCY; RUN fsck MANUALLY.\n",
1295 cdevname, usedsoftdep ? " SOFT UPDATE " : " ");
1297 * Force foreground fsck to clean up inconsistency.
1300 cmd.value = FS_NEEDSFSCK;
1302 if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1303 &cmd, sizeof cmd) == -1)
1304 pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1311 * Pwarn just prints a message when not preening or running soft dependency
1312 * protocol, or a warning (preceded by filename) when preening.
1315 pwarn(const char *fmt, ...)
1320 (void)fprintf(stdout, "%s: ", cdevname);
1321 (void)vfprintf(stdout, fmt, ap);
1326 * Stub for routines from kernel.
1329 panic(const char *fmt, ...)
1333 pfatal("INTERNAL INCONSISTENCY:");
1334 (void)vfprintf(stdout, fmt, ap);