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 struct bufarea *cgbufs; /* header for cylinder group cache */
88 static int flushtries; /* number of tries to reclaim memory */
90 char *buftype[BT_NUMBUFTYPES] = BT_NAMES;
95 diskreads = totaldiskreads = totalreads = 0;
96 bzero(&startpass, sizeof(struct timespec));
97 bzero(&finishpass, sizeof(struct timespec));
98 bzero(&slowio_starttime, sizeof(struct timeval));
99 slowio_delay_usec = 10000;
105 ftypeok(union dinode *dp)
107 switch (DIP(dp, di_mode) & IFMT) {
120 printf("bad file type 0%o\n", DIP(dp, di_mode));
126 reply(const char *question)
132 pfatal("INTERNAL ERROR: GOT TO reply()");
133 persevere = !strcmp(question, "CONTINUE");
135 if (!persevere && (nflag || (fswritefd < 0 && bkgrdflag == 0))) {
136 printf("%s? no\n\n", question);
140 if (yflag || (persevere && nflag)) {
141 printf("%s? yes\n\n", question);
145 printf("%s? [yn] ", question);
146 (void) fflush(stdout);
148 while (c != '\n' && getc(stdin) != '\n') {
154 } while (c != 'y' && c != 'Y' && c != 'n' && c != 'N');
156 if (c == 'y' || c == 'Y')
163 * Look up state information for an inode.
168 static struct inostat unallocated = { USTATE, 0, 0 };
169 struct inostatlist *ilp;
173 errx(EEXIT, "inoinfo: inumber %ju out of range",
175 ilp = &inostathead[inum / sblock.fs_ipg];
176 iloff = inum % sblock.fs_ipg;
177 if (iloff >= ilp->il_numalloced)
178 return (&unallocated);
179 return (&ilp->il_stat[iloff]);
183 * Malloc buffers and set up cache.
190 pdirbp = (struct bufarea *)0;
191 bzero(&cgblk, sizeof(struct bufarea));
192 cgblk.b_un.b_buf = Malloc((unsigned int)sblock.fs_bsize);
193 if (cgblk.b_un.b_buf == NULL)
194 errx(EEXIT, "Initial malloc(%d) failed", sblock.fs_bsize);
195 initbarea(&cgblk, BT_CYLGRP);
197 numbufs = cachelookups = cachereads = 0;
198 TAILQ_INIT(&bufqueuehd);
199 for (i = 0; i < HASHSIZE; i++)
200 LIST_INIT(&bufhashhd[i]);
201 for (i = 0; i < BT_NUMBUFTYPES; i++) {
202 readtime[i].tv_sec = totalreadtime[i].tv_sec = 0;
203 readtime[i].tv_nsec = totalreadtime[i].tv_nsec = 0;
204 readcnt[i] = totalreadcnt[i] = 0;
208 static struct bufarea *
209 allocbuf(const char *failreason)
214 bp = (struct bufarea *)Malloc(sizeof(struct bufarea));
215 bufp = Malloc((unsigned int)sblock.fs_bsize);
216 if (bp == NULL || bufp == NULL) {
217 errx(EEXIT, "%s", failreason);
221 bp->b_un.b_buf = bufp;
222 TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
223 initbarea(bp, BT_UNKNOWN);
228 * Manage cylinder group buffers.
230 * Use getblk() here rather than cgget() because the cylinder group
231 * may be corrupted but we want it anyway so we can fix it.
233 static struct bufarea *cgbufs; /* header for cylinder group cache */
234 static int flushtries; /* number of tries to reclaim memory */
239 struct bufarea *cgbp;
242 if ((unsigned) cg >= sblock.fs_ncg)
243 errx(EEXIT, "cglookup: out of range cylinder group %d", cg);
244 if (cgbufs == NULL) {
245 cgbufs = calloc(sblock.fs_ncg, sizeof(struct bufarea));
247 errx(EEXIT, "Cannot allocate cylinder group buffers");
250 if (cgbp->b_un.b_cg != NULL)
254 cgp = Malloc((unsigned int)sblock.fs_cgsize);
257 errx(EEXIT,"Ran out of memory during journal recovery");
258 getblk(&cgblk, cgtod(&sblock, cg), sblock.fs_cgsize);
261 cgbp->b_un.b_cg = cgp;
262 initbarea(cgbp, BT_CYLGRP);
263 getblk(cgbp, cgtod(&sblock, cg), sblock.fs_cgsize);
268 * Mark a cylinder group buffer as dirty.
269 * Update its check-hash if they are enabled.
272 cgdirty(struct bufarea *cgbp)
276 cg = cgbp->b_un.b_cg;
277 if ((sblock.fs_metackhash & CK_CYLGRP) != 0) {
280 calculate_crc32c(~0L, (void *)cg, sblock.fs_cgsize);
286 * Attempt to flush a cylinder group cache entry.
287 * Return whether the flush was successful.
292 struct bufarea *cgbp;
294 if (sujrecovery || flushtries == sblock.fs_ncg || cgbufs == NULL)
296 cgbp = &cgbufs[flushtries++];
297 if (cgbp->b_un.b_cg == NULL)
299 flush(fswritefd, cgbp);
300 free(cgbp->b_un.b_buf);
301 cgbp->b_un.b_buf = NULL;
306 * Manage a cache of directory blocks.
309 getdatablk(ufs2_daddr_t blkno, long size, int type)
312 struct bufhash *bhdp;
315 /* If out of range, return empty buffer with b_err == -1 */
316 if (type != BT_INODES && chkrange(blkno, size / sblock.fs_fsize)) {
320 bhdp = &bufhashhd[HASH(blkno)];
321 LIST_FOREACH(bp, bhdp, b_hash)
322 if (bp->b_bno == fsbtodb(&sblock, blkno)) {
323 if (debug && bp->b_size != size) {
324 prtbuf("getdatablk: size mismatch", bp);
325 pfatal("getdatablk: b_size %d != size %ld\n",
331 * Move long-term busy buffer back to the front of the LRU so we
332 * do not endless inspect them for recycling.
334 bp = TAILQ_LAST(&bufqueuehd, bufqueue);
335 if (bp != NULL && bp->b_refcnt != 0) {
336 TAILQ_REMOVE(&bufqueuehd, bp, b_list);
337 TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
340 * Allocate up to the minimum number of buffers before
341 * considering recycling any of them.
343 if (size > sblock.fs_bsize)
344 errx(EEXIT, "Excessive buffer size %ld > %d\n", size,
346 if (numbufs < MINBUFS) {
347 bp = allocbuf("cannot create minimal buffer pool");
348 } else if (sujrecovery) {
350 * SUJ recovery does not want anything written until it
351 * has successfully completed (so it can fail back to
352 * full fsck). Thus, we can only recycle clean buffers.
354 TAILQ_FOREACH_REVERSE(bp, &bufqueuehd, bufqueue, b_list)
355 if ((bp->b_flags & B_DIRTY) == 0 && bp->b_refcnt == 0)
358 bp = allocbuf("Ran out of memory during "
361 LIST_REMOVE(bp, b_hash);
364 * Recycle oldest non-busy buffer.
366 TAILQ_FOREACH_REVERSE(bp, &bufqueuehd, bufqueue, b_list)
367 if (bp->b_refcnt == 0)
370 bp = allocbuf("Ran out of memory for buffers");
372 LIST_REMOVE(bp, b_hash);
374 flush(fswritefd, bp);
376 LIST_INSERT_HEAD(bhdp, bp, b_hash);
377 getblk(bp, blkno, size);
381 if (debug && bp->b_type != type) {
382 printf("getdatablk: buffer type changed to %s",
386 TAILQ_REMOVE(&bufqueuehd, bp, b_list);
387 TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
394 getblk(struct bufarea *bp, ufs2_daddr_t blk, long size)
397 struct timespec start, finish;
399 dblk = fsbtodb(&sblock, blk);
400 if (bp->b_bno == dblk) {
404 readcnt[bp->b_type]++;
405 clock_gettime(CLOCK_REALTIME_PRECISE, &start);
407 if (bp->b_type != BT_EMPTY)
409 blread(fsreadfd, bp->b_un.b_buf, dblk, size);
413 clock_gettime(CLOCK_REALTIME_PRECISE, &finish);
414 timespecsub(&finish, &start, &finish);
415 timespecadd(&readtime[bp->b_type], &finish,
416 &readtime[bp->b_type]);
424 brelse(struct bufarea *bp)
427 if (bp->b_refcnt <= 0)
428 prtbuf("brelse: buffer with negative reference count", bp);
433 flush(int fd, struct bufarea *bp)
437 if ((bp->b_flags & B_DIRTY) == 0)
439 bp->b_flags &= ~B_DIRTY;
441 pfatal("WRITING IN READ_ONLY MODE.\n");
445 pfatal("WRITING %sZERO'ED BLOCK %lld TO DISK\n",
446 (bp->b_errs == bp->b_size / dev_bsize) ? "" : "PARTIALLY ",
447 (long long)bp->b_bno);
450 * Write using the appropriate function.
452 switch (bp->b_type) {
455 pfatal("BUFFER %p DOES NOT MATCH SBLK %p\n",
457 if (sbput(fd, bp->b_un.b_fs, 0) == 0)
463 if (cgput(fswritefd, &sblock, bp->b_un.b_cg) == 0)
467 if (debug && sblock.fs_magic == FS_UFS2_MAGIC) {
468 struct ufs2_dinode *dp = bp->b_un.b_dinode2;
471 for (i = 0; i < INOPB(&sblock); dp++, i++) {
472 if (ffs_verify_dinode_ckhash(&sblock, dp) == 0)
474 pwarn("flush: INODE CHECK-HASH FAILED");
476 ip.i_dp = (union dinode *)dp;
477 ip.i_number = bp->b_index + i;
479 if (preen || reply("FIX") != 0) {
481 printf(" (FIXED)\n");
482 ffs_update_dinode_ckhash(&sblock, dp);
489 blwrite(fd, bp->b_un.b_buf, bp->b_bno, bp->b_size);
495 * Journaled soft updates does not maintain cylinder group summary
496 * information during cleanup, so this routine recalculates the summary
497 * information and updates the superblock summary in preparation for
498 * writing out the cylinder group.
501 cg_write(struct bufarea *bp)
503 ufs1_daddr_t fragno, cgbno, maxbno;
510 * Fix the frag and cluster summary.
513 cgp->cg_cs.cs_nbfree = 0;
514 cgp->cg_cs.cs_nffree = 0;
515 bzero(&cgp->cg_frsum, sizeof(cgp->cg_frsum));
516 maxbno = fragstoblks(&sblock, sblock.fs_fpg);
517 if (sblock.fs_contigsumsize > 0) {
518 for (i = 1; i <= sblock.fs_contigsumsize; i++)
519 cg_clustersum(cgp)[i] = 0;
520 bzero(cg_clustersfree(cgp), howmany(maxbno, CHAR_BIT));
522 blksfree = cg_blksfree(cgp);
523 for (cgbno = 0; cgbno < maxbno; cgbno++) {
524 if (ffs_isfreeblock(&sblock, blksfree, cgbno))
526 if (ffs_isblock(&sblock, blksfree, cgbno)) {
527 ffs_clusteracct(&sblock, cgp, cgbno, 1);
528 cgp->cg_cs.cs_nbfree++;
531 fragno = blkstofrags(&sblock, cgbno);
532 blk = blkmap(&sblock, blksfree, fragno);
533 ffs_fragacct(&sblock, blk, cgp->cg_frsum, 1);
534 for (i = 0; i < sblock.fs_frag; i++)
535 if (isset(blksfree, fragno + i))
536 cgp->cg_cs.cs_nffree++;
539 * Update the superblock cg summary from our now correct values
540 * before writing the block.
542 sblock.fs_cs(&sblock, cgp->cg_cgx) = cgp->cg_cs;
546 rwerror(const char *mesg, ufs2_daddr_t blk)
553 pfatal("CANNOT %s: %ld", mesg, (long)blk);
554 if (reply("CONTINUE") == 0)
559 ckfini(int markclean)
561 struct bufarea *bp, *nbp;
562 int ofsmodified, cnt;
566 if ((!(sblock.fs_flags & FS_UNCLEAN)) != markclean) {
567 cmd.value = FS_UNCLEAN;
568 cmd.size = markclean ? -1 : 1;
569 if (sysctlbyname("vfs.ffs.setflags", 0, 0,
570 &cmd, sizeof cmd) == -1)
571 rwerror("SET FILE SYSTEM FLAGS", FS_UNCLEAN);
573 printf("\n***** FILE SYSTEM MARKED %s *****\n",
574 markclean ? "CLEAN" : "DIRTY");
578 } else if (!preen && !markclean) {
579 printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
583 if (debug && totalreads > 0)
584 printf("cache with %d buffers missed %d of %d (%d%%)\n",
585 numbufs, cachereads, cachelookups,
586 (int)(cachereads * 100 / cachelookups));
588 (void)close(fsreadfd);
592 * To remain idempotent with partial truncations the buffers
593 * must be flushed in this order:
594 * 1) cylinder groups (bitmaps)
595 * 2) indirect, directory, external attribute, and data blocks
598 * This ordering preserves access to the modified pointers
599 * until they are freed.
601 /* Step 1: cylinder groups */
603 printf("Flush Cylinder groups\n");
604 if (cgbufs != NULL) {
605 for (cnt = 0; cnt < sblock.fs_ncg; cnt++) {
606 if (cgbufs[cnt].b_un.b_cg == NULL)
608 flush(fswritefd, &cgbufs[cnt]);
609 free(cgbufs[cnt].b_un.b_cg);
613 flush(fswritefd, &cgblk);
614 free(cgblk.b_un.b_buf);
616 /* Step 2: indirect, directory, external attribute, and data blocks */
618 printf("Flush indirect, directory, external attribute, "
619 "and data blocks\n");
622 TAILQ_FOREACH_REVERSE_SAFE(bp, &bufqueuehd, bufqueue, b_list, nbp) {
623 switch (bp->b_type) {
624 /* These should not be in the buffer cache list */
629 prtbuf("ckfini: improper buffer type on cache list",bp);
631 /* These are the ones to flush in this step */
634 pfatal("Unused BT_EMPTY buffer for block %jd\n",
635 (intmax_t)bp->b_bno);
644 /* These are the ones to flush in the next step */
648 if (debug && bp->b_refcnt != 0) {
649 prtbuf("ckfini: clearing in-use buffer", bp);
650 pfatal("ckfini: clearing in-use buffer\n");
652 TAILQ_REMOVE(&bufqueuehd, bp, b_list);
654 flush(fswritefd, bp);
655 free(bp->b_un.b_buf);
658 /* Step 3: inode blocks */
660 printf("Flush inode blocks\n");
661 if (icachebp != NULL)
663 TAILQ_FOREACH_REVERSE_SAFE(bp, &bufqueuehd, bufqueue, b_list, nbp) {
664 if (debug && bp->b_refcnt != 0) {
665 prtbuf("ckfini: clearing in-use buffer", bp);
666 pfatal("ckfini: clearing in-use buffer\n");
668 TAILQ_REMOVE(&bufqueuehd, bp, b_list);
670 flush(fswritefd, bp);
671 free(bp->b_un.b_buf);
675 errx(EEXIT, "panic: lost %d buffers", numbufs - cnt);
676 /* Step 4: superblock */
678 printf("Flush the superblock\n");
679 flush(fswritefd, &sblk);
680 if (havesb && cursnapshot == 0 && sblock.fs_magic == FS_UFS2_MAGIC &&
681 sblk.b_bno != sblock.fs_sblockloc / dev_bsize &&
682 !preen && reply("UPDATE STANDARD SUPERBLOCK")) {
683 /* Change the write destination to standard superblock */
684 sblock.fs_sblockactualloc = sblock.fs_sblockloc;
685 sblk.b_bno = sblock.fs_sblockloc / dev_bsize;
687 flush(fswritefd, &sblk);
689 pdirbp = (struct bufarea *)0;
690 if (cursnapshot == 0 && sblock.fs_clean != markclean) {
691 if ((sblock.fs_clean = markclean) != 0) {
692 sblock.fs_flags &= ~(FS_UNCLEAN | FS_NEEDSFSCK);
693 sblock.fs_pendingblocks = 0;
694 sblock.fs_pendinginodes = 0;
697 ofsmodified = fsmodified;
698 flush(fswritefd, &sblk);
699 fsmodified = ofsmodified;
701 printf("\n***** FILE SYSTEM MARKED %s *****\n",
702 markclean ? "CLEAN" : "DIRTY");
708 printf("\n***** FILE SYSTEM IS CLEAN *****\n");
710 printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
715 (void)close(fsreadfd);
716 (void)close(fswritefd);
720 * Print out I/O statistics.
729 if (diskreads == 0) {
730 printf("%s: no I/O\n\n", what);
733 if (startpass.tv_sec == 0)
734 startpass = startprog;
735 printf("%s: I/O statistics\n", what);
737 totaldiskreads += diskreads;
739 for (i = 0; i < BT_NUMBUFTYPES; i++) {
740 timespecadd(&totalreadtime[i], &readtime[i], &totalreadtime[i]);
741 totalreadcnt[i] += readcnt[i];
742 readtime[i].tv_sec = readtime[i].tv_nsec = 0;
745 clock_gettime(CLOCK_REALTIME_PRECISE, &startpass);
755 printf("Final I/O statistics\n");
756 totaldiskreads += diskreads;
757 diskreads = totaldiskreads;
758 startpass = startprog;
759 for (i = 0; i < BT_NUMBUFTYPES; i++) {
760 timespecadd(&totalreadtime[i], &readtime[i], &totalreadtime[i]);
761 totalreadcnt[i] += readcnt[i];
762 readtime[i] = totalreadtime[i];
763 readcnt[i] = totalreadcnt[i];
768 static void printIOstats(void)
770 long long msec, totalmsec;
773 clock_gettime(CLOCK_REALTIME_PRECISE, &finishpass);
774 timespecsub(&finishpass, &startpass, &finishpass);
775 printf("Running time: %jd.%03ld sec\n",
776 (intmax_t)finishpass.tv_sec, finishpass.tv_nsec / 1000000);
777 printf("buffer reads by type:\n");
778 for (totalmsec = 0, i = 0; i < BT_NUMBUFTYPES; i++)
779 totalmsec += readtime[i].tv_sec * 1000 +
780 readtime[i].tv_nsec / 1000000;
783 for (i = 0; i < BT_NUMBUFTYPES; i++) {
787 readtime[i].tv_sec * 1000 + readtime[i].tv_nsec / 1000000;
788 printf("%21s:%8ld %2ld.%ld%% %4jd.%03ld sec %2lld.%lld%%\n",
789 buftype[i], readcnt[i], readcnt[i] * 100 / diskreads,
790 (readcnt[i] * 1000 / diskreads) % 10,
791 (intmax_t)readtime[i].tv_sec, readtime[i].tv_nsec / 1000000,
792 msec * 100 / totalmsec, (msec * 1000 / totalmsec) % 10);
798 blread(int fd, char *buf, ufs2_daddr_t blk, long size)
810 if (pread(fd, buf, (int)size, offset) == size) {
817 * This is handled specially here instead of in rwerror because
818 * rwerror is used for all sorts of errors, not just true read/write
819 * errors. It should be refactored and fixed.
822 pfatal("CANNOT READ_BLK: %ld", (long)blk);
823 errx(EEXIT, "ABORTING DUE TO READ ERRORS");
825 rwerror("READ BLK", blk);
828 memset(buf, 0, (size_t)size);
829 printf("THE FOLLOWING DISK SECTORS COULD NOT BE READ:");
830 for (cp = buf, i = 0; i < size; i += secsize, cp += secsize) {
831 if (pread(fd, cp, (int)secsize, offset + i) != secsize) {
832 if (secsize != dev_bsize && dev_bsize != 1)
833 printf(" %jd (%jd),",
834 (intmax_t)(blk * dev_bsize + i) / secsize,
835 (intmax_t)blk + i / dev_bsize);
837 printf(" %jd,", (intmax_t)blk + i / dev_bsize);
848 blwrite(int fd, char *buf, ufs2_daddr_t blk, ssize_t size)
858 if (pwrite(fd, buf, size, offset) == size) {
863 rwerror("WRITE BLK", blk);
864 printf("THE FOLLOWING SECTORS COULD NOT BE WRITTEN:");
865 for (cp = buf, i = 0; i < size; i += dev_bsize, cp += dev_bsize)
866 if (pwrite(fd, cp, dev_bsize, offset + i) != dev_bsize)
867 printf(" %jd,", (intmax_t)blk + i / dev_bsize);
873 blerase(int fd, ufs2_daddr_t blk, long size)
879 ioarg[0] = blk * dev_bsize;
881 ioctl(fd, DIOCGDELETE, ioarg);
882 /* we don't really care if we succeed or not */
887 * Fill a contiguous region with all-zeroes. Note ZEROBUFSIZE is by
888 * definition a multiple of dev_bsize.
891 blzero(int fd, ufs2_daddr_t blk, long size)
899 zero = calloc(ZEROBUFSIZE, 1);
901 errx(EEXIT, "cannot allocate buffer pool");
903 offset = blk * dev_bsize;
904 if (lseek(fd, offset, 0) < 0)
905 rwerror("SEEK BLK", blk);
907 len = MIN(ZEROBUFSIZE, size);
908 if (write(fd, zero, len) != len)
909 rwerror("WRITE BLK", blk);
910 blk += len / dev_bsize;
916 * Verify cylinder group's magic number and other parameters. If the
917 * test fails, offer an option to rebuild the whole cylinder group.
920 check_cgmagic(int cg, struct bufarea *cgbp, int request_rebuild)
922 struct cg *cgp = cgbp->b_un.b_cg;
923 uint32_t cghash, calchash;
926 * Extended cylinder group checks.
928 calchash = cgp->cg_ckhash;
929 if ((sblock.fs_metackhash & CK_CYLGRP) != 0) {
930 cghash = cgp->cg_ckhash;
932 calchash = calculate_crc32c(~0L, (void *)cgp, sblock.fs_cgsize);
933 cgp->cg_ckhash = cghash;
935 if (cgp->cg_ckhash == calchash &&
938 ((sblock.fs_magic == FS_UFS1_MAGIC &&
939 cgp->cg_old_niblk == sblock.fs_ipg &&
940 cgp->cg_ndblk <= sblock.fs_fpg &&
941 cgp->cg_old_ncyl <= sblock.fs_old_cpg) ||
942 (sblock.fs_magic == FS_UFS2_MAGIC &&
943 cgp->cg_niblk == sblock.fs_ipg &&
944 cgp->cg_ndblk <= sblock.fs_fpg &&
945 cgp->cg_initediblk <= sblock.fs_ipg))) {
948 pfatal("CYLINDER GROUP %d: INTEGRITY CHECK FAILED", cg);
949 if (!request_rebuild)
951 if (!reply("REBUILD CYLINDER GROUP")) {
952 printf("YOU WILL NEED TO RERUN FSCK.\n");
957 * Zero out the cylinder group and then initialize critical fields.
958 * Bit maps and summaries will be recalculated by later passes.
960 memset(cgp, 0, (size_t)sblock.fs_cgsize);
961 cgp->cg_magic = CG_MAGIC;
963 cgp->cg_niblk = sblock.fs_ipg;
964 cgp->cg_initediblk = MIN(sblock.fs_ipg, 2 * INOPB(&sblock));
965 if (cgbase(&sblock, cg) + sblock.fs_fpg < sblock.fs_size)
966 cgp->cg_ndblk = sblock.fs_fpg;
968 cgp->cg_ndblk = sblock.fs_size - cgbase(&sblock, cg);
969 cgp->cg_iusedoff = &cgp->cg_space[0] - (u_char *)(&cgp->cg_firstfield);
970 if (sblock.fs_magic == FS_UFS1_MAGIC) {
972 cgp->cg_initediblk = 0;
973 cgp->cg_old_ncyl = sblock.fs_old_cpg;
974 cgp->cg_old_niblk = sblock.fs_ipg;
975 cgp->cg_old_btotoff = cgp->cg_iusedoff;
976 cgp->cg_old_boff = cgp->cg_old_btotoff +
977 sblock.fs_old_cpg * sizeof(int32_t);
978 cgp->cg_iusedoff = cgp->cg_old_boff +
979 sblock.fs_old_cpg * sizeof(u_int16_t);
981 cgp->cg_freeoff = cgp->cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
982 cgp->cg_nextfreeoff = cgp->cg_freeoff + howmany(sblock.fs_fpg,CHAR_BIT);
983 if (sblock.fs_contigsumsize > 0) {
984 cgp->cg_nclusterblks = cgp->cg_ndblk / sblock.fs_frag;
985 cgp->cg_clustersumoff =
986 roundup(cgp->cg_nextfreeoff, sizeof(u_int32_t));
987 cgp->cg_clustersumoff -= sizeof(u_int32_t);
988 cgp->cg_clusteroff = cgp->cg_clustersumoff +
989 (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
990 cgp->cg_nextfreeoff = cgp->cg_clusteroff +
991 howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
998 * allocate a data block with the specified number of fragments
1001 allocblk(long frags)
1003 int i, j, k, cg, baseblk;
1004 struct bufarea *cgbp;
1007 if (frags <= 0 || frags > sblock.fs_frag)
1009 for (i = 0; i < maxfsblock - sblock.fs_frag; i += sblock.fs_frag) {
1010 for (j = 0; j <= sblock.fs_frag - frags; j++) {
1011 if (testbmap(i + j))
1013 for (k = 1; k < frags; k++)
1014 if (testbmap(i + j + k))
1020 cg = dtog(&sblock, i + j);
1021 cgbp = cglookup(cg);
1022 cgp = cgbp->b_un.b_cg;
1023 if (!check_cgmagic(cg, cgbp, 0))
1025 baseblk = dtogd(&sblock, i + j);
1026 for (k = 0; k < frags; k++) {
1028 clrbit(cg_blksfree(cgp), baseblk + k);
1031 if (frags == sblock.fs_frag)
1032 cgp->cg_cs.cs_nbfree--;
1034 cgp->cg_cs.cs_nffree -= frags;
1043 * Slow down IO so as to leave some disk bandwidth for other processes
1049 /* Delay one in every 8 operations */
1050 slowio_pollcnt = (slowio_pollcnt + 1) & 7;
1051 if (slowio_pollcnt == 0) {
1052 gettimeofday(&slowio_starttime, NULL);
1062 if (slowio_pollcnt != 0)
1065 /* Update the slowdown interval. */
1066 gettimeofday(&tv, NULL);
1067 delay_usec = (tv.tv_sec - slowio_starttime.tv_sec) * 1000000 +
1068 (tv.tv_usec - slowio_starttime.tv_usec);
1069 if (delay_usec < 64)
1071 if (delay_usec > 2500000)
1072 delay_usec = 2500000;
1073 slowio_delay_usec = (slowio_delay_usec * 63 + delay_usec) >> 6;
1074 /* delay by 8 times the average IO delay */
1075 if (slowio_delay_usec > 64)
1076 usleep(slowio_delay_usec * 8);
1083 getpathname(char *namebuf, ino_t curdir, ino_t ino)
1088 struct inodesc idesc;
1089 static int busy = 0;
1091 if (curdir == ino && ino == UFS_ROOTINO) {
1092 (void)strcpy(namebuf, "/");
1095 if (busy || !INO_IS_DVALID(curdir)) {
1096 (void)strcpy(namebuf, "?");
1100 memset(&idesc, 0, sizeof(struct inodesc));
1101 idesc.id_type = DATA;
1102 idesc.id_fix = IGNORE;
1103 cp = &namebuf[MAXPATHLEN - 1];
1105 if (curdir != ino) {
1106 idesc.id_parent = curdir;
1109 while (ino != UFS_ROOTINO) {
1110 idesc.id_number = ino;
1111 idesc.id_func = findino;
1112 idesc.id_name = strdup("..");
1114 if ((ckinode(ip.i_dp, &idesc) & FOUND) == 0) {
1120 idesc.id_number = idesc.id_parent;
1121 idesc.id_parent = ino;
1122 idesc.id_func = findname;
1123 idesc.id_name = namebuf;
1124 ginode(idesc.id_number, &ip);
1125 if ((ckinode(ip.i_dp, &idesc) & FOUND) == 0) {
1130 len = strlen(namebuf);
1132 memmove(cp, namebuf, (size_t)len);
1134 if (cp < &namebuf[UFS_MAXNAMLEN])
1136 ino = idesc.id_number;
1139 if (ino != UFS_ROOTINO)
1141 memmove(namebuf, cp, (size_t)(&namebuf[MAXPATHLEN] - cp));
1145 catch(int sig __unused)
1153 * When preening, allow a single quit to signal
1154 * a special exit after file system checks complete
1155 * so that reboot sequence may be interrupted.
1158 catchquit(int sig __unused)
1160 printf("returning to single-user after file system check\n");
1162 (void)signal(SIGQUIT, SIG_DFL);
1166 * determine whether an inode should be fixed.
1169 dofix(struct inodesc *idesc, const char *msg)
1172 switch (idesc->id_fix) {
1175 if (idesc->id_type == DATA)
1176 direrror(idesc->id_number, msg);
1180 printf(" (SALVAGED)\n");
1181 idesc->id_fix = FIX;
1184 if (reply("SALVAGE") == 0) {
1185 idesc->id_fix = NOFIX;
1188 idesc->id_fix = FIX;
1199 errx(EEXIT, "UNKNOWN INODESC FIX MODE %d", idesc->id_fix);
1208 * Print details about a buffer.
1211 prtbuf(const char *msg, struct bufarea *bp)
1214 printf("%s: bp %p, type %s, bno %jd, size %d, refcnt %d, flags %s, "
1215 "index %jd\n", msg, bp, BT_BUFTYPE(bp->b_type), (intmax_t) bp->b_bno,
1216 bp->b_size, bp->b_refcnt, bp->b_flags & B_DIRTY ? "dirty" : "clean",
1217 (intmax_t) bp->b_index);
1221 * An unexpected inconsistency occurred.
1222 * Die if preening or file system is running with soft dependency protocol,
1223 * otherwise just print message and continue.
1226 pfatal(const char *fmt, ...)
1231 (void)vfprintf(stdout, fmt, ap);
1234 (void)fprintf(stdout,
1235 "\nUNEXPECTED SOFT UPDATE INCONSISTENCY\n");
1237 * Force foreground fsck to clean up inconsistency.
1240 cmd.value = FS_NEEDSFSCK;
1242 if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1243 &cmd, sizeof cmd) == -1)
1244 pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1245 fprintf(stdout, "CANNOT RUN IN BACKGROUND\n");
1251 if (cdevname == NULL)
1252 cdevname = strdup("fsck");
1253 (void)fprintf(stdout, "%s: ", cdevname);
1254 (void)vfprintf(stdout, fmt, ap);
1255 (void)fprintf(stdout,
1256 "\n%s: UNEXPECTED%sINCONSISTENCY; RUN fsck MANUALLY.\n",
1257 cdevname, usedsoftdep ? " SOFT UPDATE " : " ");
1259 * Force foreground fsck to clean up inconsistency.
1262 cmd.value = FS_NEEDSFSCK;
1264 if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1265 &cmd, sizeof cmd) == -1)
1266 pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1273 * Pwarn just prints a message when not preening or running soft dependency
1274 * protocol, or a warning (preceded by filename) when preening.
1277 pwarn(const char *fmt, ...)
1282 (void)fprintf(stdout, "%s: ", cdevname);
1283 (void)vfprintf(stdout, fmt, ap);
1288 * Stub for routines from kernel.
1291 panic(const char *fmt, ...)
1295 pfatal("INTERNAL INCONSISTENCY:");
1296 (void)vfprintf(stdout, fmt, ap);