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") == 0 ||
133 strcmp(question, "LOOK FOR ALTERNATE SUPERBLOCKS") == 0;
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 if ((cgblk.b_un.b_buf = Malloc((unsigned int)sblock.fs_bsize)) == NULL)
191 errx(EEXIT, "Initial malloc(%d) failed", sblock.fs_bsize);
192 initbarea(&cgblk, BT_CYLGRP);
193 numbufs = cachelookups = cachereads = 0;
194 TAILQ_INIT(&bufqueuehd);
195 for (i = 0; i < HASHSIZE; i++)
196 LIST_INIT(&bufhashhd[i]);
197 for (i = 0; i < BT_NUMBUFTYPES; i++) {
198 readtime[i].tv_sec = totalreadtime[i].tv_sec = 0;
199 readtime[i].tv_nsec = totalreadtime[i].tv_nsec = 0;
200 readcnt[i] = totalreadcnt[i] = 0;
204 static struct bufarea *
205 allocbuf(const char *failreason)
210 bp = (struct bufarea *)Malloc(sizeof(struct bufarea));
211 bufp = Malloc((unsigned int)sblock.fs_bsize);
212 if (bp == NULL || bufp == NULL) {
213 errx(EEXIT, "%s", failreason);
217 bp->b_un.b_buf = bufp;
218 TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
219 initbarea(bp, BT_UNKNOWN);
224 * Manage cylinder group buffers.
226 * Use getblk() here rather than cgget() because the cylinder group
227 * may be corrupted but we want it anyway so we can fix it.
229 static struct bufarea *cgbufs; /* header for cylinder group cache */
230 static int flushtries; /* number of tries to reclaim memory */
235 struct bufarea *cgbp;
238 if ((unsigned) cg >= sblock.fs_ncg)
239 errx(EEXIT, "cglookup: out of range cylinder group %d", cg);
240 if (cgbufs == NULL) {
241 cgbufs = calloc(sblock.fs_ncg, sizeof(struct bufarea));
243 errx(EEXIT, "Cannot allocate cylinder group buffers");
246 if (cgbp->b_un.b_cg != NULL)
250 cgp = Malloc((unsigned int)sblock.fs_cgsize);
253 errx(EEXIT,"Ran out of memory during journal recovery");
254 flush(fswritefd, &cgblk);
255 getblk(&cgblk, cgtod(&sblock, cg), sblock.fs_cgsize);
258 cgbp->b_un.b_cg = cgp;
259 initbarea(cgbp, BT_CYLGRP);
260 getblk(cgbp, cgtod(&sblock, cg), sblock.fs_cgsize);
265 * Mark a cylinder group buffer as dirty.
266 * Update its check-hash if they are enabled.
269 cgdirty(struct bufarea *cgbp)
273 cg = cgbp->b_un.b_cg;
274 if ((sblock.fs_metackhash & CK_CYLGRP) != 0) {
277 calculate_crc32c(~0L, (void *)cg, sblock.fs_cgsize);
283 * Attempt to flush a cylinder group cache entry.
284 * Return whether the flush was successful.
289 struct bufarea *cgbp;
291 if (sujrecovery || flushtries == sblock.fs_ncg || cgbufs == NULL)
293 cgbp = &cgbufs[flushtries++];
294 if (cgbp->b_un.b_cg == NULL)
296 flush(fswritefd, cgbp);
297 free(cgbp->b_un.b_buf);
298 cgbp->b_un.b_buf = NULL;
303 * Manage a cache of directory blocks.
306 getdatablk(ufs2_daddr_t blkno, long size, int type)
309 struct bufhash *bhdp;
312 /* If out of range, return empty buffer with b_err == -1 */
313 if (type != BT_INODES && chkrange(blkno, size / sblock.fs_fsize)) {
317 bhdp = &bufhashhd[HASH(blkno)];
318 LIST_FOREACH(bp, bhdp, b_hash)
319 if (bp->b_bno == fsbtodb(&sblock, blkno)) {
320 if (debug && bp->b_size != size) {
321 prtbuf("getdatablk: size mismatch", bp);
322 pfatal("getdatablk: b_size %d != size %ld\n",
328 * Move long-term busy buffer back to the front of the LRU so we
329 * do not endless inspect them for recycling.
331 bp = TAILQ_LAST(&bufqueuehd, bufqueue);
332 if (bp != NULL && bp->b_refcnt != 0) {
333 TAILQ_REMOVE(&bufqueuehd, bp, b_list);
334 TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
337 * Allocate up to the minimum number of buffers before
338 * considering recycling any of them.
340 if (size > sblock.fs_bsize)
341 errx(EEXIT, "Excessive buffer size %ld > %d\n", size,
343 if (numbufs < MINBUFS) {
344 bp = allocbuf("cannot create minimal buffer pool");
345 } else if (sujrecovery) {
347 * SUJ recovery does not want anything written until it
348 * has successfully completed (so it can fail back to
349 * full fsck). Thus, we can only recycle clean buffers.
351 TAILQ_FOREACH_REVERSE(bp, &bufqueuehd, bufqueue, b_list)
352 if ((bp->b_flags & B_DIRTY) == 0 && bp->b_refcnt == 0)
355 bp = allocbuf("Ran out of memory during "
358 LIST_REMOVE(bp, b_hash);
361 * Recycle oldest non-busy buffer.
363 TAILQ_FOREACH_REVERSE(bp, &bufqueuehd, bufqueue, b_list)
364 if (bp->b_refcnt == 0)
367 bp = allocbuf("Ran out of memory for buffers");
369 LIST_REMOVE(bp, b_hash);
371 flush(fswritefd, bp);
373 LIST_INSERT_HEAD(bhdp, bp, b_hash);
374 getblk(bp, blkno, size);
378 if (debug && bp->b_type != type) {
379 printf("getdatablk: buffer type changed to %s",
383 TAILQ_REMOVE(&bufqueuehd, bp, b_list);
384 TAILQ_INSERT_HEAD(&bufqueuehd, bp, b_list);
391 getblk(struct bufarea *bp, ufs2_daddr_t blk, long size)
394 struct timespec start, finish;
396 dblk = fsbtodb(&sblock, blk);
397 if (bp->b_bno == dblk) {
401 readcnt[bp->b_type]++;
402 clock_gettime(CLOCK_REALTIME_PRECISE, &start);
404 if (bp->b_type != BT_EMPTY)
406 blread(fsreadfd, bp->b_un.b_buf, dblk, size);
410 clock_gettime(CLOCK_REALTIME_PRECISE, &finish);
411 timespecsub(&finish, &start, &finish);
412 timespecadd(&readtime[bp->b_type], &finish,
413 &readtime[bp->b_type]);
421 brelse(struct bufarea *bp)
424 if (bp->b_refcnt <= 0)
425 prtbuf("brelse: buffer with negative reference count", bp);
430 flush(int fd, struct bufarea *bp)
434 if ((bp->b_flags & B_DIRTY) == 0)
436 bp->b_flags &= ~B_DIRTY;
438 pfatal("WRITING IN READ_ONLY MODE.\n");
442 pfatal("WRITING %sZERO'ED BLOCK %lld TO DISK\n",
443 (bp->b_errs == bp->b_size / dev_bsize) ? "" : "PARTIALLY ",
444 (long long)bp->b_bno);
447 * Write using the appropriate function.
449 switch (bp->b_type) {
452 pfatal("BUFFER %p DOES NOT MATCH SBLK %p\n",
454 if (sbput(fd, bp->b_un.b_fs, 0) == 0)
460 if (cgput(fswritefd, &sblock, bp->b_un.b_cg) == 0)
464 if (debug && sblock.fs_magic == FS_UFS2_MAGIC) {
465 struct ufs2_dinode *dp = bp->b_un.b_dinode2;
468 for (i = 0; i < bp->b_size; dp++, i += sizeof(*dp)) {
469 if (ffs_verify_dinode_ckhash(&sblock, dp) == 0)
471 pwarn("flush: INODE CHECK-HASH FAILED");
473 ip.i_dp = (union dinode *)dp;
474 ip.i_number = bp->b_index + i;
476 if (preen || reply("FIX") != 0) {
478 printf(" (FIXED)\n");
479 ffs_update_dinode_ckhash(&sblock, dp);
486 blwrite(fd, bp->b_un.b_buf, bp->b_bno, bp->b_size);
492 * Journaled soft updates does not maintain cylinder group summary
493 * information during cleanup, so this routine recalculates the summary
494 * information and updates the superblock summary in preparation for
495 * writing out the cylinder group.
498 cg_write(struct bufarea *bp)
500 ufs1_daddr_t fragno, cgbno, maxbno;
507 * Fix the frag and cluster summary.
510 cgp->cg_cs.cs_nbfree = 0;
511 cgp->cg_cs.cs_nffree = 0;
512 bzero(&cgp->cg_frsum, sizeof(cgp->cg_frsum));
513 maxbno = fragstoblks(&sblock, sblock.fs_fpg);
514 if (sblock.fs_contigsumsize > 0) {
515 for (i = 1; i <= sblock.fs_contigsumsize; i++)
516 cg_clustersum(cgp)[i] = 0;
517 bzero(cg_clustersfree(cgp), howmany(maxbno, CHAR_BIT));
519 blksfree = cg_blksfree(cgp);
520 for (cgbno = 0; cgbno < maxbno; cgbno++) {
521 if (ffs_isfreeblock(&sblock, blksfree, cgbno))
523 if (ffs_isblock(&sblock, blksfree, cgbno)) {
524 ffs_clusteracct(&sblock, cgp, cgbno, 1);
525 cgp->cg_cs.cs_nbfree++;
528 fragno = blkstofrags(&sblock, cgbno);
529 blk = blkmap(&sblock, blksfree, fragno);
530 ffs_fragacct(&sblock, blk, cgp->cg_frsum, 1);
531 for (i = 0; i < sblock.fs_frag; i++)
532 if (isset(blksfree, fragno + i))
533 cgp->cg_cs.cs_nffree++;
536 * Update the superblock cg summary from our now correct values
537 * before writing the block.
539 sblock.fs_cs(&sblock, cgp->cg_cgx) = cgp->cg_cs;
543 rwerror(const char *mesg, ufs2_daddr_t blk)
550 pfatal("CANNOT %s: %ld", mesg, (long)blk);
551 if (reply("CONTINUE") == 0)
556 ckfini(int markclean)
558 struct bufarea *bp, *nbp;
559 struct inoinfo *inp, *ninp;
560 int ofsmodified, cnt, cg, i;
564 if ((!(sblock.fs_flags & FS_UNCLEAN)) != markclean) {
565 cmd.value = FS_UNCLEAN;
566 cmd.size = markclean ? -1 : 1;
567 if (sysctlbyname("vfs.ffs.setflags", 0, 0,
568 &cmd, sizeof cmd) == -1)
569 pwarn("CANNOT SET FILE SYSTEM DIRTY FLAG\n");
571 printf("\n***** FILE SYSTEM MARKED %s *****\n",
572 markclean ? "CLEAN" : "DIRTY");
576 } else if (!preen && !markclean) {
577 printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
582 if (debug && cachelookups > 0)
583 printf("cache with %d buffers missed %d of %d (%d%%)\n",
584 numbufs, cachereads, cachelookups,
585 (int)(cachereads * 100 / cachelookups));
587 (void)close(fsreadfd);
591 * To remain idempotent with partial truncations the buffers
592 * must be flushed in this order:
593 * 1) cylinder groups (bitmaps)
594 * 2) indirect, directory, external attribute, and data blocks
597 * This ordering preserves access to the modified pointers
598 * until they are freed.
600 /* Step 1: cylinder groups */
602 printf("Flush Cylinder groups\n");
603 if (cgbufs != NULL) {
604 for (cnt = 0; cnt < sblock.fs_ncg; cnt++) {
605 if (cgbufs[cnt].b_un.b_cg == NULL)
607 flush(fswritefd, &cgbufs[cnt]);
608 free(cgbufs[cnt].b_un.b_cg);
613 flush(fswritefd, &cgblk);
614 free(cgblk.b_un.b_buf);
615 cgblk.b_un.b_buf = NULL;
617 /* Step 2: indirect, directory, external attribute, and data blocks */
619 printf("Flush indirect, directory, external attribute, "
620 "and data blocks\n");
621 if (pdirbp != NULL) {
625 TAILQ_FOREACH_REVERSE_SAFE(bp, &bufqueuehd, bufqueue, b_list, nbp) {
626 switch (bp->b_type) {
627 /* These should not be in the buffer cache list */
632 prtbuf("ckfini: improper buffer type on cache list",bp);
634 /* These are the ones to flush in this step */
637 pfatal("Unused BT_EMPTY buffer for block %jd\n",
638 (intmax_t)bp->b_bno);
647 /* These are the ones to flush in the next step */
651 if (debug && bp->b_refcnt != 0) {
652 prtbuf("ckfini: clearing in-use buffer", bp);
653 pfatal("ckfini: clearing in-use buffer\n");
655 TAILQ_REMOVE(&bufqueuehd, bp, b_list);
657 flush(fswritefd, bp);
658 free(bp->b_un.b_buf);
661 /* Step 3: inode blocks */
663 printf("Flush inode blocks\n");
664 if (icachebp != NULL) {
668 TAILQ_FOREACH_REVERSE_SAFE(bp, &bufqueuehd, bufqueue, b_list, nbp) {
669 if (debug && bp->b_refcnt != 0) {
670 prtbuf("ckfini: clearing in-use buffer", bp);
671 pfatal("ckfini: clearing in-use buffer\n");
673 TAILQ_REMOVE(&bufqueuehd, bp, b_list);
675 flush(fswritefd, bp);
676 free(bp->b_un.b_buf);
680 errx(EEXIT, "panic: lost %d buffers", numbufs - cnt);
681 /* Step 4: superblock */
683 printf("Flush the superblock\n");
684 flush(fswritefd, &sblk);
685 if (havesb && cursnapshot == 0 &&
686 sblk.b_bno != sblock.fs_sblockloc / dev_bsize) {
687 if (preen || reply("UPDATE STANDARD SUPERBLOCK")) {
688 /* Change write destination to standard superblock */
689 sblock.fs_sblockactualloc = sblock.fs_sblockloc;
690 sblk.b_bno = sblock.fs_sblockloc / dev_bsize;
692 flush(fswritefd, &sblk);
697 if (cursnapshot == 0 && sblock.fs_clean != markclean) {
698 if ((sblock.fs_clean = markclean) != 0) {
699 sblock.fs_flags &= ~(FS_UNCLEAN | FS_NEEDSFSCK);
700 sblock.fs_pendingblocks = 0;
701 sblock.fs_pendinginodes = 0;
704 ofsmodified = fsmodified;
705 flush(fswritefd, &sblk);
706 fsmodified = ofsmodified;
708 printf("\n***** FILE SYSTEM MARKED %s *****\n",
709 markclean ? "CLEAN" : "DIRTY");
715 printf("\n***** FILE SYSTEM IS CLEAN *****\n");
717 printf("\n***** FILE SYSTEM STILL DIRTY *****\n");
722 * Free allocated tracking structures.
724 if (blockmap != NULL)
727 if (inostathead != NULL) {
728 for (cg = 0; cg < sblock.fs_ncg; cg++)
729 if (inostathead[cg].il_stat != NULL)
730 free((char *)inostathead[cg].il_stat);
737 if (inphead != NULL) {
738 for (i = 0; i < dirhash; i++) {
739 for (inp = inphead[i]; inp != NULL; inp = ninp) {
740 ninp = inp->i_nexthash;
748 (void)close(fsreadfd);
749 (void)close(fswritefd);
753 * Print out I/O statistics.
762 if (diskreads == 0) {
763 printf("%s: no I/O\n\n", what);
766 if (startpass.tv_sec == 0)
767 startpass = startprog;
768 printf("%s: I/O statistics\n", what);
770 totaldiskreads += diskreads;
772 for (i = 0; i < BT_NUMBUFTYPES; i++) {
773 timespecadd(&totalreadtime[i], &readtime[i], &totalreadtime[i]);
774 totalreadcnt[i] += readcnt[i];
775 readtime[i].tv_sec = readtime[i].tv_nsec = 0;
778 clock_gettime(CLOCK_REALTIME_PRECISE, &startpass);
788 printf("Final I/O statistics\n");
789 totaldiskreads += diskreads;
790 diskreads = totaldiskreads;
791 startpass = startprog;
792 for (i = 0; i < BT_NUMBUFTYPES; i++) {
793 timespecadd(&totalreadtime[i], &readtime[i], &totalreadtime[i]);
794 totalreadcnt[i] += readcnt[i];
795 readtime[i] = totalreadtime[i];
796 readcnt[i] = totalreadcnt[i];
801 static void printIOstats(void)
803 long long msec, totalmsec;
806 clock_gettime(CLOCK_REALTIME_PRECISE, &finishpass);
807 timespecsub(&finishpass, &startpass, &finishpass);
808 printf("Running time: %jd.%03ld sec\n",
809 (intmax_t)finishpass.tv_sec, finishpass.tv_nsec / 1000000);
810 printf("buffer reads by type:\n");
811 for (totalmsec = 0, i = 0; i < BT_NUMBUFTYPES; i++)
812 totalmsec += readtime[i].tv_sec * 1000 +
813 readtime[i].tv_nsec / 1000000;
816 for (i = 0; i < BT_NUMBUFTYPES; i++) {
820 readtime[i].tv_sec * 1000 + readtime[i].tv_nsec / 1000000;
821 printf("%21s:%8ld %2ld.%ld%% %4jd.%03ld sec %2lld.%lld%%\n",
822 buftype[i], readcnt[i], readcnt[i] * 100 / diskreads,
823 (readcnt[i] * 1000 / diskreads) % 10,
824 (intmax_t)readtime[i].tv_sec, readtime[i].tv_nsec / 1000000,
825 msec * 100 / totalmsec, (msec * 1000 / totalmsec) % 10);
831 blread(int fd, char *buf, ufs2_daddr_t blk, long size)
843 if (pread(fd, buf, (int)size, offset) == size) {
850 * This is handled specially here instead of in rwerror because
851 * rwerror is used for all sorts of errors, not just true read/write
852 * errors. It should be refactored and fixed.
855 pfatal("CANNOT READ_BLK: %ld", (long)blk);
856 errx(EEXIT, "ABORTING DUE TO READ ERRORS");
858 rwerror("READ BLK", blk);
861 memset(buf, 0, (size_t)size);
862 printf("THE FOLLOWING DISK SECTORS COULD NOT BE READ:");
863 for (cp = buf, i = 0; i < size; i += secsize, cp += secsize) {
864 if (pread(fd, cp, (int)secsize, offset + i) != secsize) {
865 if (secsize != dev_bsize && dev_bsize != 1)
866 printf(" %jd (%jd),",
867 (intmax_t)(blk * dev_bsize + i) / secsize,
868 (intmax_t)blk + i / dev_bsize);
870 printf(" %jd,", (intmax_t)blk + i / dev_bsize);
881 blwrite(int fd, char *buf, ufs2_daddr_t blk, ssize_t size)
891 if (pwrite(fd, buf, size, offset) == size) {
896 rwerror("WRITE BLK", blk);
897 printf("THE FOLLOWING SECTORS COULD NOT BE WRITTEN:");
898 for (cp = buf, i = 0; i < size; i += dev_bsize, cp += dev_bsize)
899 if (pwrite(fd, cp, dev_bsize, offset + i) != dev_bsize)
900 printf(" %jd,", (intmax_t)blk + i / dev_bsize);
906 blerase(int fd, ufs2_daddr_t blk, long size)
912 ioarg[0] = blk * dev_bsize;
914 ioctl(fd, DIOCGDELETE, ioarg);
915 /* we don't really care if we succeed or not */
920 * Fill a contiguous region with all-zeroes. Note ZEROBUFSIZE is by
921 * definition a multiple of dev_bsize.
924 blzero(int fd, ufs2_daddr_t blk, long size)
932 zero = calloc(ZEROBUFSIZE, 1);
934 errx(EEXIT, "cannot allocate buffer pool");
936 offset = blk * dev_bsize;
937 if (lseek(fd, offset, 0) < 0)
938 rwerror("SEEK BLK", blk);
940 len = MIN(ZEROBUFSIZE, size);
941 if (write(fd, zero, len) != len)
942 rwerror("WRITE BLK", blk);
943 blk += len / dev_bsize;
949 * Verify cylinder group's magic number and other parameters. If the
950 * test fails, offer an option to rebuild the whole cylinder group.
953 #define CHK(lhs, op, rhs, fmt) \
955 pwarn("UFS%d cylinder group %d failed: " \
956 "%s (" #fmt ") %s %s (" #fmt ")\n", \
957 sblock.fs_magic == FS_UFS1_MAGIC ? 1 : 2, cg, \
958 #lhs, (intmax_t)lhs, #op, #rhs, (intmax_t)rhs); \
962 check_cgmagic(int cg, struct bufarea *cgbp, int request_rebuild)
964 struct cg *cgp = cgbp->b_un.b_cg;
965 uint32_t cghash, calchash;
966 static int prevfailcg = -1;
970 * Extended cylinder group checks.
972 calchash = cgp->cg_ckhash;
973 if ((sblock.fs_metackhash & CK_CYLGRP) != 0 &&
974 (ckhashadd & CK_CYLGRP) == 0) {
975 cghash = cgp->cg_ckhash;
977 calchash = calculate_crc32c(~0L, (void *)cgp, sblock.fs_cgsize);
978 cgp->cg_ckhash = cghash;
981 CHK(cgp->cg_ckhash, !=, calchash, "%jd");
982 CHK(cg_chkmagic(cgp), ==, 0, "%jd");
983 CHK(cgp->cg_cgx, !=, cg, "%jd");
984 CHK(cgp->cg_ndblk, >, sblock.fs_fpg, "%jd");
985 if (sblock.fs_magic == FS_UFS1_MAGIC) {
986 CHK(cgp->cg_old_niblk, !=, sblock.fs_ipg, "%jd");
987 CHK(cgp->cg_old_ncyl, >, sblock.fs_old_cpg, "%jd");
988 } else if (sblock.fs_magic == FS_UFS2_MAGIC) {
989 CHK(cgp->cg_niblk, !=, sblock.fs_ipg, "%jd");
990 CHK(cgp->cg_initediblk, >, sblock.fs_ipg, "%jd");
994 if (prevfailcg == cg)
997 pfatal("CYLINDER GROUP %d: INTEGRITY CHECK FAILED", cg);
998 if (!request_rebuild) {
1002 if (!reply("REBUILD CYLINDER GROUP")) {
1003 printf("YOU WILL NEED TO RERUN FSCK.\n");
1008 * Zero out the cylinder group and then initialize critical fields.
1009 * Bit maps and summaries will be recalculated by later passes.
1011 memset(cgp, 0, (size_t)sblock.fs_cgsize);
1012 cgp->cg_magic = CG_MAGIC;
1014 cgp->cg_niblk = sblock.fs_ipg;
1015 cgp->cg_initediblk = MIN(sblock.fs_ipg, 2 * INOPB(&sblock));
1016 if (cgbase(&sblock, cg) + sblock.fs_fpg < sblock.fs_size)
1017 cgp->cg_ndblk = sblock.fs_fpg;
1019 cgp->cg_ndblk = sblock.fs_size - cgbase(&sblock, cg);
1020 cgp->cg_iusedoff = &cgp->cg_space[0] - (u_char *)(&cgp->cg_firstfield);
1021 if (sblock.fs_magic == FS_UFS1_MAGIC) {
1023 cgp->cg_initediblk = 0;
1024 cgp->cg_old_ncyl = sblock.fs_old_cpg;
1025 cgp->cg_old_niblk = sblock.fs_ipg;
1026 cgp->cg_old_btotoff = cgp->cg_iusedoff;
1027 cgp->cg_old_boff = cgp->cg_old_btotoff +
1028 sblock.fs_old_cpg * sizeof(int32_t);
1029 cgp->cg_iusedoff = cgp->cg_old_boff +
1030 sblock.fs_old_cpg * sizeof(u_int16_t);
1032 cgp->cg_freeoff = cgp->cg_iusedoff + howmany(sblock.fs_ipg, CHAR_BIT);
1033 cgp->cg_nextfreeoff = cgp->cg_freeoff + howmany(sblock.fs_fpg,CHAR_BIT);
1034 if (sblock.fs_contigsumsize > 0) {
1035 cgp->cg_nclusterblks = cgp->cg_ndblk / sblock.fs_frag;
1036 cgp->cg_clustersumoff =
1037 roundup(cgp->cg_nextfreeoff, sizeof(u_int32_t));
1038 cgp->cg_clustersumoff -= sizeof(u_int32_t);
1039 cgp->cg_clusteroff = cgp->cg_clustersumoff +
1040 (sblock.fs_contigsumsize + 1) * sizeof(u_int32_t);
1041 cgp->cg_nextfreeoff = cgp->cg_clusteroff +
1042 howmany(fragstoblks(&sblock, sblock.fs_fpg), CHAR_BIT);
1044 cgp->cg_ckhash = calculate_crc32c(~0L, (void *)cgp, sblock.fs_cgsize);
1050 * allocate a data block with the specified number of fragments
1053 allocblk(long frags)
1055 int i, j, k, cg, baseblk;
1056 struct bufarea *cgbp;
1059 if (frags <= 0 || frags > sblock.fs_frag)
1061 for (i = 0; i < maxfsblock - sblock.fs_frag; i += sblock.fs_frag) {
1062 for (j = 0; j <= sblock.fs_frag - frags; j++) {
1063 if (testbmap(i + j))
1065 for (k = 1; k < frags; k++)
1066 if (testbmap(i + j + k))
1072 cg = dtog(&sblock, i + j);
1073 cgbp = cglookup(cg);
1074 cgp = cgbp->b_un.b_cg;
1075 if (!check_cgmagic(cg, cgbp, 0)) {
1076 i = (cg + 1) * sblock.fs_fpg - sblock.fs_frag;
1079 baseblk = dtogd(&sblock, i + j);
1080 for (k = 0; k < frags; k++) {
1082 clrbit(cg_blksfree(cgp), baseblk + k);
1085 if (frags == sblock.fs_frag)
1086 cgp->cg_cs.cs_nbfree--;
1088 cgp->cg_cs.cs_nffree -= frags;
1097 * Slow down IO so as to leave some disk bandwidth for other processes
1103 /* Delay one in every 8 operations */
1104 slowio_pollcnt = (slowio_pollcnt + 1) & 7;
1105 if (slowio_pollcnt == 0) {
1106 gettimeofday(&slowio_starttime, NULL);
1116 if (slowio_pollcnt != 0)
1119 /* Update the slowdown interval. */
1120 gettimeofday(&tv, NULL);
1121 delay_usec = (tv.tv_sec - slowio_starttime.tv_sec) * 1000000 +
1122 (tv.tv_usec - slowio_starttime.tv_usec);
1123 if (delay_usec < 64)
1125 if (delay_usec > 2500000)
1126 delay_usec = 2500000;
1127 slowio_delay_usec = (slowio_delay_usec * 63 + delay_usec) >> 6;
1128 /* delay by 8 times the average IO delay */
1129 if (slowio_delay_usec > 64)
1130 usleep(slowio_delay_usec * 8);
1137 getpathname(char *namebuf, ino_t curdir, ino_t ino)
1142 struct inodesc idesc;
1143 static int busy = 0;
1145 if (curdir == ino && ino == UFS_ROOTINO) {
1146 (void)strcpy(namebuf, "/");
1149 if (busy || !INO_IS_DVALID(curdir)) {
1150 (void)strcpy(namebuf, "?");
1154 memset(&idesc, 0, sizeof(struct inodesc));
1155 idesc.id_type = DATA;
1156 idesc.id_fix = IGNORE;
1157 cp = &namebuf[MAXPATHLEN - 1];
1159 if (curdir != ino) {
1160 idesc.id_parent = curdir;
1163 while (ino != UFS_ROOTINO) {
1164 idesc.id_number = ino;
1165 idesc.id_func = findino;
1166 idesc.id_name = strdup("..");
1168 if ((ckinode(ip.i_dp, &idesc) & FOUND) == 0) {
1174 idesc.id_number = idesc.id_parent;
1175 idesc.id_parent = ino;
1176 idesc.id_func = findname;
1177 idesc.id_name = namebuf;
1178 ginode(idesc.id_number, &ip);
1179 if ((ckinode(ip.i_dp, &idesc) & FOUND) == 0) {
1184 len = strlen(namebuf);
1186 memmove(cp, namebuf, (size_t)len);
1188 if (cp < &namebuf[UFS_MAXNAMLEN])
1190 ino = idesc.id_number;
1193 if (ino != UFS_ROOTINO)
1195 memmove(namebuf, cp, (size_t)(&namebuf[MAXPATHLEN] - cp));
1199 catch(int sig __unused)
1207 * When preening, allow a single quit to signal
1208 * a special exit after file system checks complete
1209 * so that reboot sequence may be interrupted.
1212 catchquit(int sig __unused)
1214 printf("returning to single-user after file system check\n");
1216 (void)signal(SIGQUIT, SIG_DFL);
1220 * determine whether an inode should be fixed.
1223 dofix(struct inodesc *idesc, const char *msg)
1226 switch (idesc->id_fix) {
1229 if (idesc->id_type == DATA)
1230 direrror(idesc->id_number, msg);
1234 printf(" (SALVAGED)\n");
1235 idesc->id_fix = FIX;
1238 if (reply("SALVAGE") == 0) {
1239 idesc->id_fix = NOFIX;
1242 idesc->id_fix = FIX;
1253 errx(EEXIT, "UNKNOWN INODESC FIX MODE %d", idesc->id_fix);
1262 * Print details about a buffer.
1265 prtbuf(const char *msg, struct bufarea *bp)
1268 printf("%s: bp %p, type %s, bno %jd, size %d, refcnt %d, flags %s, "
1269 "index %jd\n", msg, bp, BT_BUFTYPE(bp->b_type), (intmax_t) bp->b_bno,
1270 bp->b_size, bp->b_refcnt, bp->b_flags & B_DIRTY ? "dirty" : "clean",
1271 (intmax_t) bp->b_index);
1275 * An unexpected inconsistency occurred.
1276 * Die if preening or file system is running with soft dependency protocol,
1277 * otherwise just print message and continue.
1280 pfatal(const char *fmt, ...)
1285 (void)vfprintf(stdout, fmt, ap);
1288 (void)fprintf(stdout,
1289 "\nUNEXPECTED SOFT UPDATE INCONSISTENCY\n");
1291 * Force foreground fsck to clean up inconsistency.
1294 cmd.value = FS_NEEDSFSCK;
1296 if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1297 &cmd, sizeof cmd) == -1)
1298 pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1299 fprintf(stdout, "CANNOT RUN IN BACKGROUND\n");
1305 if (cdevname == NULL)
1306 cdevname = strdup("fsck");
1307 (void)fprintf(stdout, "%s: ", cdevname);
1308 (void)vfprintf(stdout, fmt, ap);
1309 (void)fprintf(stdout,
1310 "\n%s: UNEXPECTED%sINCONSISTENCY; RUN fsck MANUALLY.\n",
1311 cdevname, usedsoftdep ? " SOFT UPDATE " : " ");
1313 * Force foreground fsck to clean up inconsistency.
1316 cmd.value = FS_NEEDSFSCK;
1318 if (sysctlbyname("vfs.ffs.setflags", 0, 0,
1319 &cmd, sizeof cmd) == -1)
1320 pwarn("CANNOT SET FS_NEEDSFSCK FLAG\n");
1327 * Pwarn just prints a message when not preening or running soft dependency
1328 * protocol, or a warning (preceded by filename) when preening.
1331 pwarn(const char *fmt, ...)
1336 (void)fprintf(stdout, "%s: ", cdevname);
1337 (void)vfprintf(stdout, fmt, ap);
1342 * Stub for routines from kernel.
1345 panic(const char *fmt, ...)
1349 pfatal("INTERNAL INCONSISTENCY:");
1350 (void)vfprintf(stdout, fmt, ap);