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 getblk(&cgblk, cgtod(&sblock, cg), sblock.fs_cgsize);
256 cgbp->b_un.b_cg = cgp;
257 initbarea(cgbp, BT_CYLGRP);
258 getblk(cgbp, cgtod(&sblock, cg), sblock.fs_cgsize);
263 * Mark a cylinder group buffer as dirty.
264 * Update its check-hash if they are enabled.
267 cgdirty(struct bufarea *cgbp)
271 cg = cgbp->b_un.b_cg;
272 if ((sblock.fs_metackhash & CK_CYLGRP) != 0) {
275 calculate_crc32c(~0L, (void *)cg, sblock.fs_cgsize);
281 * Attempt to flush a cylinder group cache entry.
282 * Return whether the flush was successful.
287 struct bufarea *cgbp;
289 if (sujrecovery || flushtries == sblock.fs_ncg || cgbufs == NULL)
291 cgbp = &cgbufs[flushtries++];
292 if (cgbp->b_un.b_cg == NULL)
294 flush(fswritefd, cgbp);
295 free(cgbp->b_un.b_buf);
296 cgbp->b_un.b_buf = NULL;
301 * Manage a cache of directory blocks.
304 getdatablk(ufs2_daddr_t blkno, long size, int type)
307 struct bufhash *bhdp;
310 /* If out of range, return empty buffer with b_err == -1 */
311 if (type != BT_INODES && chkrange(blkno, size / sblock.fs_fsize)) {
315 bhdp = &bufhashhd[HASH(blkno)];
316 LIST_FOREACH(bp, bhdp, b_hash)
317 if (bp->b_bno == fsbtodb(&sblock, blkno)) {
318 if (debug && bp->b_size != size) {
319 prtbuf("getdatablk: size mismatch", bp);
320 pfatal("getdatablk: b_size %d != size %ld\n",
326 * Move long-term busy buffer back to the front of the LRU so we
327 * do not endless inspect them for recycling.
329 bp = TAILQ_LAST(&bufqueuehd, bufqueue);
330 if (bp != NULL && bp->b_refcnt != 0) {
331 TAILQ_REMOVE(&bufqueuehd, bp, b_list);
332 TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
335 * Allocate up to the minimum number of buffers before
336 * considering recycling any of them.
338 if (size > sblock.fs_bsize)
339 errx(EEXIT, "Excessive buffer size %ld > %d\n", size,
341 if (numbufs < MINBUFS) {
342 bp = allocbuf("cannot create minimal buffer pool");
343 } else if (sujrecovery) {
345 * SUJ recovery does not want anything written until it
346 * has successfully completed (so it can fail back to
347 * full fsck). Thus, we can only recycle clean buffers.
349 TAILQ_FOREACH_REVERSE(bp, &bufqueuehd, bufqueue, b_list)
350 if ((bp->b_flags & B_DIRTY) == 0 && bp->b_refcnt == 0)
353 bp = allocbuf("Ran out of memory during "
356 LIST_REMOVE(bp, b_hash);
359 * Recycle oldest non-busy buffer.
361 TAILQ_FOREACH_REVERSE(bp, &bufqueuehd, bufqueue, b_list)
362 if (bp->b_refcnt == 0)
365 bp = allocbuf("Ran out of memory for buffers");
367 LIST_REMOVE(bp, b_hash);
369 flush(fswritefd, bp);
371 LIST_INSERT_HEAD(bhdp, bp, b_hash);
372 getblk(bp, blkno, size);
376 if (debug && bp->b_type != type) {
377 printf("getdatablk: buffer type changed to %s",
381 TAILQ_REMOVE(&bufqueuehd, bp, b_list);
382 TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
389 getblk(struct bufarea *bp, ufs2_daddr_t blk, long size)
392 struct timespec start, finish;
394 dblk = fsbtodb(&sblock, blk);
395 if (bp->b_bno == dblk) {
399 readcnt[bp->b_type]++;
400 clock_gettime(CLOCK_REALTIME_PRECISE, &start);
402 if (bp->b_type != BT_EMPTY)
404 blread(fsreadfd, bp->b_un.b_buf, dblk, size);
408 clock_gettime(CLOCK_REALTIME_PRECISE, &finish);
409 timespecsub(&finish, &start, &finish);
410 timespecadd(&readtime[bp->b_type], &finish,
411 &readtime[bp->b_type]);
419 brelse(struct bufarea *bp)
422 if (bp->b_refcnt <= 0)
423 prtbuf("brelse: buffer with negative reference count", bp);
428 flush(int fd, struct bufarea *bp)
432 if ((bp->b_flags & B_DIRTY) == 0)
434 bp->b_flags &= ~B_DIRTY;
436 pfatal("WRITING IN READ_ONLY MODE.\n");
440 pfatal("WRITING %sZERO'ED BLOCK %lld TO DISK\n",
441 (bp->b_errs == bp->b_size / dev_bsize) ? "" : "PARTIALLY ",
442 (long long)bp->b_bno);
445 * Write using the appropriate function.
447 switch (bp->b_type) {
450 pfatal("BUFFER %p DOES NOT MATCH SBLK %p\n",
452 if (sbput(fd, bp->b_un.b_fs, 0) == 0)
458 if (cgput(fswritefd, &sblock, bp->b_un.b_cg) == 0)
462 if (debug && sblock.fs_magic == FS_UFS2_MAGIC) {
463 struct ufs2_dinode *dp = bp->b_un.b_dinode2;
466 for (i = 0; i < INOPB(&sblock); dp++, i++) {
467 if (ffs_verify_dinode_ckhash(&sblock, dp) == 0)
469 pwarn("flush: INODE CHECK-HASH FAILED");
471 ip.i_dp = (union dinode *)dp;
472 ip.i_number = bp->b_index + i;
474 if (preen || reply("FIX") != 0) {
476 printf(" (FIXED)\n");
477 ffs_update_dinode_ckhash(&sblock, dp);
484 blwrite(fd, bp->b_un.b_buf, bp->b_bno, bp->b_size);
490 * Journaled soft updates does not maintain cylinder group summary
491 * information during cleanup, so this routine recalculates the summary
492 * information and updates the superblock summary in preparation for
493 * writing out the cylinder group.
496 cg_write(struct bufarea *bp)
498 ufs1_daddr_t fragno, cgbno, maxbno;
505 * Fix the frag and cluster summary.
508 cgp->cg_cs.cs_nbfree = 0;
509 cgp->cg_cs.cs_nffree = 0;
510 bzero(&cgp->cg_frsum, sizeof(cgp->cg_frsum));
511 maxbno = fragstoblks(&sblock, sblock.fs_fpg);
512 if (sblock.fs_contigsumsize > 0) {
513 for (i = 1; i <= sblock.fs_contigsumsize; i++)
514 cg_clustersum(cgp)[i] = 0;
515 bzero(cg_clustersfree(cgp), howmany(maxbno, CHAR_BIT));
517 blksfree = cg_blksfree(cgp);
518 for (cgbno = 0; cgbno < maxbno; cgbno++) {
519 if (ffs_isfreeblock(&sblock, blksfree, cgbno))
521 if (ffs_isblock(&sblock, blksfree, cgbno)) {
522 ffs_clusteracct(&sblock, cgp, cgbno, 1);
523 cgp->cg_cs.cs_nbfree++;
526 fragno = blkstofrags(&sblock, cgbno);
527 blk = blkmap(&sblock, blksfree, fragno);
528 ffs_fragacct(&sblock, blk, cgp->cg_frsum, 1);
529 for (i = 0; i < sblock.fs_frag; i++)
530 if (isset(blksfree, fragno + i))
531 cgp->cg_cs.cs_nffree++;
534 * Update the superblock cg summary from our now correct values
535 * before writing the block.
537 sblock.fs_cs(&sblock, cgp->cg_cgx) = cgp->cg_cs;
541 rwerror(const char *mesg, ufs2_daddr_t blk)
548 pfatal("CANNOT %s: %ld", mesg, (long)blk);
549 if (reply("CONTINUE") == 0)
554 ckfini(int markclean)
556 struct bufarea *bp, *nbp;
557 struct inoinfo *inp, *ninp;
558 int ofsmodified, cnt, cg, i;
562 if ((!(sblock.fs_flags & FS_UNCLEAN)) != markclean) {
563 cmd.value = FS_UNCLEAN;
564 cmd.size = markclean ? -1 : 1;
565 if (sysctlbyname("vfs.ffs.setflags", 0, 0,
566 &cmd, sizeof cmd) == -1)
567 rwerror("SET FILE SYSTEM FLAGS", FS_UNCLEAN);
569 printf("\n***** FILE SYSTEM MARKED %s *****\n",
570 markclean ? "CLEAN" : "DIRTY");
574 } else if (!preen && !markclean) {
575 printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
579 if (debug && cachelookups > 0)
580 printf("cache with %d buffers missed %d of %d (%d%%)\n",
581 numbufs, cachereads, cachelookups,
582 (int)(cachereads * 100 / cachelookups));
584 (void)close(fsreadfd);
588 * To remain idempotent with partial truncations the buffers
589 * must be flushed in this order:
590 * 1) cylinder groups (bitmaps)
591 * 2) indirect, directory, external attribute, and data blocks
594 * This ordering preserves access to the modified pointers
595 * until they are freed.
597 /* Step 1: cylinder groups */
599 printf("Flush Cylinder groups\n");
600 if (cgbufs != NULL) {
601 for (cnt = 0; cnt < sblock.fs_ncg; cnt++) {
602 if (cgbufs[cnt].b_un.b_cg == NULL)
604 flush(fswritefd, &cgbufs[cnt]);
605 free(cgbufs[cnt].b_un.b_cg);
610 flush(fswritefd, &cgblk);
611 free(cgblk.b_un.b_buf);
612 cgblk.b_un.b_buf = NULL;
614 /* Step 2: indirect, directory, external attribute, and data blocks */
616 printf("Flush indirect, directory, external attribute, "
617 "and data blocks\n");
618 if (pdirbp != NULL) {
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) {
665 TAILQ_FOREACH_REVERSE_SAFE(bp, &bufqueuehd, bufqueue, b_list, nbp) {
666 if (debug && bp->b_refcnt != 0) {
667 prtbuf("ckfini: clearing in-use buffer", bp);
668 pfatal("ckfini: clearing in-use buffer\n");
670 TAILQ_REMOVE(&bufqueuehd, bp, b_list);
672 flush(fswritefd, bp);
673 free(bp->b_un.b_buf);
677 errx(EEXIT, "panic: lost %d buffers", numbufs - cnt);
678 /* Step 4: superblock */
680 printf("Flush the superblock\n");
681 flush(fswritefd, &sblk);
682 if (havesb && cursnapshot == 0 && sblock.fs_magic == FS_UFS2_MAGIC &&
683 sblk.b_bno != sblock.fs_sblockloc / dev_bsize &&
684 !preen && reply("UPDATE STANDARD SUPERBLOCK")) {
685 /* Change the write destination to standard superblock */
686 sblock.fs_sblockactualloc = sblock.fs_sblockloc;
687 sblk.b_bno = sblock.fs_sblockloc / dev_bsize;
689 flush(fswritefd, &sblk);
691 if (cursnapshot == 0 && sblock.fs_clean != markclean) {
692 if ((sblock.fs_clean = markclean) != 0) {
693 sblock.fs_flags &= ~(FS_UNCLEAN | FS_NEEDSFSCK);
694 sblock.fs_pendingblocks = 0;
695 sblock.fs_pendinginodes = 0;
698 ofsmodified = fsmodified;
699 flush(fswritefd, &sblk);
700 fsmodified = ofsmodified;
702 printf("\n***** FILE SYSTEM MARKED %s *****\n",
703 markclean ? "CLEAN" : "DIRTY");
709 printf("\n***** FILE SYSTEM IS CLEAN *****\n");
711 printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
716 * Free allocated tracking structures.
718 if (blockmap != NULL)
721 if (inostathead != NULL) {
722 for (cg = 0; cg < sblock.fs_ncg; cg++)
723 if (inostathead[cg].il_stat != NULL)
724 free((char *)inostathead[cg].il_stat);
731 if (inphead != NULL) {
732 for (i = 0; i < dirhash; i++) {
733 for (inp = inphead[i]; inp != NULL; inp = ninp) {
734 ninp = inp->i_nexthash;
742 (void)close(fsreadfd);
743 (void)close(fswritefd);
747 * Print out I/O statistics.
756 if (diskreads == 0) {
757 printf("%s: no I/O\n\n", what);
760 if (startpass.tv_sec == 0)
761 startpass = startprog;
762 printf("%s: I/O statistics\n", what);
764 totaldiskreads += diskreads;
766 for (i = 0; i < BT_NUMBUFTYPES; i++) {
767 timespecadd(&totalreadtime[i], &readtime[i], &totalreadtime[i]);
768 totalreadcnt[i] += readcnt[i];
769 readtime[i].tv_sec = readtime[i].tv_nsec = 0;
772 clock_gettime(CLOCK_REALTIME_PRECISE, &startpass);
782 printf("Final I/O statistics\n");
783 totaldiskreads += diskreads;
784 diskreads = totaldiskreads;
785 startpass = startprog;
786 for (i = 0; i < BT_NUMBUFTYPES; i++) {
787 timespecadd(&totalreadtime[i], &readtime[i], &totalreadtime[i]);
788 totalreadcnt[i] += readcnt[i];
789 readtime[i] = totalreadtime[i];
790 readcnt[i] = totalreadcnt[i];
795 static void printIOstats(void)
797 long long msec, totalmsec;
800 clock_gettime(CLOCK_REALTIME_PRECISE, &finishpass);
801 timespecsub(&finishpass, &startpass, &finishpass);
802 printf("Running time: %jd.%03ld sec\n",
803 (intmax_t)finishpass.tv_sec, finishpass.tv_nsec / 1000000);
804 printf("buffer reads by type:\n");
805 for (totalmsec = 0, i = 0; i < BT_NUMBUFTYPES; i++)
806 totalmsec += readtime[i].tv_sec * 1000 +
807 readtime[i].tv_nsec / 1000000;
810 for (i = 0; i < BT_NUMBUFTYPES; i++) {
814 readtime[i].tv_sec * 1000 + readtime[i].tv_nsec / 1000000;
815 printf("%21s:%8ld %2ld.%ld%% %4jd.%03ld sec %2lld.%lld%%\n",
816 buftype[i], readcnt[i], readcnt[i] * 100 / diskreads,
817 (readcnt[i] * 1000 / diskreads) % 10,
818 (intmax_t)readtime[i].tv_sec, readtime[i].tv_nsec / 1000000,
819 msec * 100 / totalmsec, (msec * 1000 / totalmsec) % 10);
825 blread(int fd, char *buf, ufs2_daddr_t blk, long size)
837 if (pread(fd, buf, (int)size, offset) == size) {
844 * This is handled specially here instead of in rwerror because
845 * rwerror is used for all sorts of errors, not just true read/write
846 * errors. It should be refactored and fixed.
849 pfatal("CANNOT READ_BLK: %ld", (long)blk);
850 errx(EEXIT, "ABORTING DUE TO READ ERRORS");
852 rwerror("READ BLK", blk);
855 memset(buf, 0, (size_t)size);
856 printf("THE FOLLOWING DISK SECTORS COULD NOT BE READ:");
857 for (cp = buf, i = 0; i < size; i += secsize, cp += secsize) {
858 if (pread(fd, cp, (int)secsize, offset + i) != secsize) {
859 if (secsize != dev_bsize && dev_bsize != 1)
860 printf(" %jd (%jd),",
861 (intmax_t)(blk * dev_bsize + i) / secsize,
862 (intmax_t)blk + i / dev_bsize);
864 printf(" %jd,", (intmax_t)blk + i / dev_bsize);
875 blwrite(int fd, char *buf, ufs2_daddr_t blk, ssize_t size)
885 if (pwrite(fd, buf, size, offset) == size) {
890 rwerror("WRITE BLK", blk);
891 printf("THE FOLLOWING SECTORS COULD NOT BE WRITTEN:");
892 for (cp = buf, i = 0; i < size; i += dev_bsize, cp += dev_bsize)
893 if (pwrite(fd, cp, dev_bsize, offset + i) != dev_bsize)
894 printf(" %jd,", (intmax_t)blk + i / dev_bsize);
900 blerase(int fd, ufs2_daddr_t blk, long size)
906 ioarg[0] = blk * dev_bsize;
908 ioctl(fd, DIOCGDELETE, ioarg);
909 /* we don't really care if we succeed or not */
914 * Fill a contiguous region with all-zeroes. Note ZEROBUFSIZE is by
915 * definition a multiple of dev_bsize.
918 blzero(int fd, ufs2_daddr_t blk, long size)
926 zero = calloc(ZEROBUFSIZE, 1);
928 errx(EEXIT, "cannot allocate buffer pool");
930 offset = blk * dev_bsize;
931 if (lseek(fd, offset, 0) < 0)
932 rwerror("SEEK BLK", blk);
934 len = MIN(ZEROBUFSIZE, size);
935 if (write(fd, zero, len) != len)
936 rwerror("WRITE BLK", blk);
937 blk += len / dev_bsize;
943 * Verify cylinder group's magic number and other parameters. If the
944 * test fails, offer an option to rebuild the whole cylinder group.
947 check_cgmagic(int cg, struct bufarea *cgbp, int request_rebuild)
949 struct cg *cgp = cgbp->b_un.b_cg;
950 uint32_t cghash, calchash;
951 static int prevfailcg = -1;
954 * Extended cylinder group checks.
956 calchash = cgp->cg_ckhash;
957 if ((sblock.fs_metackhash & CK_CYLGRP) != 0 &&
958 (ckhashadd & CK_CYLGRP) == 0) {
959 cghash = cgp->cg_ckhash;
961 calchash = calculate_crc32c(~0L, (void *)cgp, sblock.fs_cgsize);
962 cgp->cg_ckhash = cghash;
964 if (cgp->cg_ckhash == calchash &&
967 ((sblock.fs_magic == FS_UFS1_MAGIC &&
968 cgp->cg_old_niblk == sblock.fs_ipg &&
969 cgp->cg_ndblk <= sblock.fs_fpg &&
970 cgp->cg_old_ncyl <= sblock.fs_old_cpg) ||
971 (sblock.fs_magic == FS_UFS2_MAGIC &&
972 cgp->cg_niblk == sblock.fs_ipg &&
973 cgp->cg_ndblk <= sblock.fs_fpg &&
974 cgp->cg_initediblk <= sblock.fs_ipg))) {
977 if (prevfailcg == cg)
980 pfatal("CYLINDER GROUP %d: INTEGRITY CHECK FAILED", cg);
981 if (!request_rebuild) {
985 if (!reply("REBUILD CYLINDER GROUP")) {
986 printf("YOU WILL NEED TO RERUN FSCK.\n");
991 * Zero out the cylinder group and then initialize critical fields.
992 * Bit maps and summaries will be recalculated by later passes.
994 memset(cgp, 0, (size_t)sblock.fs_cgsize);
995 cgp->cg_magic = CG_MAGIC;
997 cgp->cg_niblk = sblock.fs_ipg;
998 cgp->cg_initediblk = MIN(sblock.fs_ipg, 2 * INOPB(&sblock));
999 if (cgbase(&sblock, cg) + sblock.fs_fpg < sblock.fs_size)
1000 cgp->cg_ndblk = sblock.fs_fpg;
1002 cgp->cg_ndblk = sblock.fs_size - cgbase(&sblock, cg);
1003 cgp->cg_iusedoff = &cgp->cg_space[0] - (u_char *)(&cgp->cg_firstfield);
1004 if (sblock.fs_magic == FS_UFS1_MAGIC) {
1006 cgp->cg_initediblk = 0;
1007 cgp->cg_old_ncyl = sblock.fs_old_cpg;
1008 cgp->cg_old_niblk = sblock.fs_ipg;
1009 cgp->cg_old_btotoff = cgp->cg_iusedoff;
1010 cgp->cg_old_boff = cgp->cg_old_btotoff +
1011 sblock.fs_old_cpg * sizeof(int32_t);
1012 cgp->cg_iusedoff = cgp->cg_old_boff +
1013 sblock.fs_old_cpg * sizeof(u_int16_t);
1015 cgp->cg_freeoff = cgp->cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
1016 cgp->cg_nextfreeoff = cgp->cg_freeoff + howmany(sblock.fs_fpg,CHAR_BIT);
1017 if (sblock.fs_contigsumsize > 0) {
1018 cgp->cg_nclusterblks = cgp->cg_ndblk / sblock.fs_frag;
1019 cgp->cg_clustersumoff =
1020 roundup(cgp->cg_nextfreeoff, sizeof(u_int32_t));
1021 cgp->cg_clustersumoff -= sizeof(u_int32_t);
1022 cgp->cg_clusteroff = cgp->cg_clustersumoff +
1023 (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
1024 cgp->cg_nextfreeoff = cgp->cg_clusteroff +
1025 howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
1032 * allocate a data block with the specified number of fragments
1035 allocblk(long frags)
1037 int i, j, k, cg, baseblk;
1038 struct bufarea *cgbp;
1041 if (frags <= 0 || frags > sblock.fs_frag)
1043 for (i = 0; i < maxfsblock - sblock.fs_frag; i += sblock.fs_frag) {
1044 for (j = 0; j <= sblock.fs_frag - frags; j++) {
1045 if (testbmap(i + j))
1047 for (k = 1; k < frags; k++)
1048 if (testbmap(i + j + k))
1054 cg = dtog(&sblock, i + j);
1055 cgbp = cglookup(cg);
1056 cgp = cgbp->b_un.b_cg;
1057 if (!check_cgmagic(cg, cgbp, 0)) {
1058 i = (cg + 1) * sblock.fs_fpg - sblock.fs_frag;
1061 baseblk = dtogd(&sblock, i + j);
1062 for (k = 0; k < frags; k++) {
1064 clrbit(cg_blksfree(cgp), baseblk + k);
1067 if (frags == sblock.fs_frag)
1068 cgp->cg_cs.cs_nbfree--;
1070 cgp->cg_cs.cs_nffree -= frags;
1079 * Slow down IO so as to leave some disk bandwidth for other processes
1085 /* Delay one in every 8 operations */
1086 slowio_pollcnt = (slowio_pollcnt + 1) & 7;
1087 if (slowio_pollcnt == 0) {
1088 gettimeofday(&slowio_starttime, NULL);
1098 if (slowio_pollcnt != 0)
1101 /* Update the slowdown interval. */
1102 gettimeofday(&tv, NULL);
1103 delay_usec = (tv.tv_sec - slowio_starttime.tv_sec) * 1000000 +
1104 (tv.tv_usec - slowio_starttime.tv_usec);
1105 if (delay_usec < 64)
1107 if (delay_usec > 2500000)
1108 delay_usec = 2500000;
1109 slowio_delay_usec = (slowio_delay_usec * 63 + delay_usec) >> 6;
1110 /* delay by 8 times the average IO delay */
1111 if (slowio_delay_usec > 64)
1112 usleep(slowio_delay_usec * 8);
1119 getpathname(char *namebuf, ino_t curdir, ino_t ino)
1124 struct inodesc idesc;
1125 static int busy = 0;
1127 if (curdir == ino && ino == UFS_ROOTINO) {
1128 (void)strcpy(namebuf, "/");
1131 if (busy || !INO_IS_DVALID(curdir)) {
1132 (void)strcpy(namebuf, "?");
1136 memset(&idesc, 0, sizeof(struct inodesc));
1137 idesc.id_type = DATA;
1138 idesc.id_fix = IGNORE;
1139 cp = &namebuf[MAXPATHLEN - 1];
1141 if (curdir != ino) {
1142 idesc.id_parent = curdir;
1145 while (ino != UFS_ROOTINO) {
1146 idesc.id_number = ino;
1147 idesc.id_func = findino;
1148 idesc.id_name = strdup("..");
1150 if ((ckinode(ip.i_dp, &idesc) & FOUND) == 0) {
1156 idesc.id_number = idesc.id_parent;
1157 idesc.id_parent = ino;
1158 idesc.id_func = findname;
1159 idesc.id_name = namebuf;
1160 ginode(idesc.id_number, &ip);
1161 if ((ckinode(ip.i_dp, &idesc) & FOUND) == 0) {
1166 len = strlen(namebuf);
1168 memmove(cp, namebuf, (size_t)len);
1170 if (cp < &namebuf[UFS_MAXNAMLEN])
1172 ino = idesc.id_number;
1175 if (ino != UFS_ROOTINO)
1177 memmove(namebuf, cp, (size_t)(&namebuf[MAXPATHLEN] - cp));
1181 catch(int sig __unused)
1189 * When preening, allow a single quit to signal
1190 * a special exit after file system checks complete
1191 * so that reboot sequence may be interrupted.
1194 catchquit(int sig __unused)
1196 printf("returning to single-user after file system check\n");
1198 (void)signal(SIGQUIT, SIG_DFL);
1202 * determine whether an inode should be fixed.
1205 dofix(struct inodesc *idesc, const char *msg)
1208 switch (idesc->id_fix) {
1211 if (idesc->id_type == DATA)
1212 direrror(idesc->id_number, msg);
1216 printf(" (SALVAGED)\n");
1217 idesc->id_fix = FIX;
1220 if (reply("SALVAGE") == 0) {
1221 idesc->id_fix = NOFIX;
1224 idesc->id_fix = FIX;
1235 errx(EEXIT, "UNKNOWN INODESC FIX MODE %d", idesc->id_fix);
1244 * Print details about a buffer.
1247 prtbuf(const char *msg, struct bufarea *bp)
1250 printf("%s: bp %p, type %s, bno %jd, size %d, refcnt %d, flags %s, "
1251 "index %jd\n", msg, bp, BT_BUFTYPE(bp->b_type), (intmax_t) bp->b_bno,
1252 bp->b_size, bp->b_refcnt, bp->b_flags & B_DIRTY ? "dirty" : "clean",
1253 (intmax_t) bp->b_index);
1257 * An unexpected inconsistency occurred.
1258 * Die if preening or file system is running with soft dependency protocol,
1259 * otherwise just print message and continue.
1262 pfatal(const char *fmt, ...)
1267 (void)vfprintf(stdout, fmt, ap);
1270 (void)fprintf(stdout,
1271 "\nUNEXPECTED SOFT UPDATE INCONSISTENCY\n");
1273 * Force foreground fsck to clean up inconsistency.
1276 cmd.value = FS_NEEDSFSCK;
1278 if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1279 &cmd, sizeof cmd) == -1)
1280 pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1281 fprintf(stdout, "CANNOT RUN IN BACKGROUND\n");
1287 if (cdevname == NULL)
1288 cdevname = strdup("fsck");
1289 (void)fprintf(stdout, "%s: ", cdevname);
1290 (void)vfprintf(stdout, fmt, ap);
1291 (void)fprintf(stdout,
1292 "\n%s: UNEXPECTED%sINCONSISTENCY; RUN fsck MANUALLY.\n",
1293 cdevname, usedsoftdep ? " SOFT UPDATE " : " ");
1295 * Force foreground fsck to clean up inconsistency.
1298 cmd.value = FS_NEEDSFSCK;
1300 if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1301 &cmd, sizeof cmd) == -1)
1302 pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1309 * Pwarn just prints a message when not preening or running soft dependency
1310 * protocol, or a warning (preceded by filename) when preening.
1313 pwarn(const char *fmt, ...)
1318 (void)fprintf(stdout, "%s: ", cdevname);
1319 (void)vfprintf(stdout, fmt, ap);
1324 * Stub for routines from kernel.
1327 panic(const char *fmt, ...)
1331 pfatal("INTERNAL INCONSISTENCY:");
1332 (void)vfprintf(stdout, fmt, ap);