xref: /titanic_44/usr/src/cmd/fs.d/ufs/fsck/pass5.c (revision 28167c24ba5be8b7c1d05e02d053f4a55cd21cc9)
1 /*
2  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
3  * Use is subject to license terms.
4  */
5 
6 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
7 /*	  All Rights Reserved  	*/
8 
9 /*
10  * Copyright (c) 1980, 1986, 1990 The Regents of the University of California.
11  * All rights reserved.
12  *
13  * Redistribution and use in source and binary forms are permitted
14  * provided that: (1) source distributions retain this entire copyright
15  * notice and comment, and (2) distributions including binaries display
16  * the following acknowledgement:  ``This product includes software
17  * developed by the University of California, Berkeley and its contributors''
18  * in the documentation or other materials provided with the distribution
19  * and in all advertising materials mentioning features or use of this
20  * software. Neither the name of the University nor the names of its
21  * contributors may be used to endorse or promote products derived
22  * from this software without specific prior written permission.
23  * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
24  * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
25  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
26  */
27 
28 #pragma ident	"%Z%%M%	%I%	%E% SMI"
29 
30 #include <stdio.h>
31 #include <stdlib.h>
32 #include <unistd.h>
33 #include <string.h>
34 #include <sys/param.h>
35 #include <sys/mntent.h>
36 #include <sys/fs/ufs_fs.h>
37 #include <sys/vnode.h>
38 #include <sys/fs/ufs_inode.h>
39 #include "fsck.h"
40 
41 static int check_maps(uchar_t *, uchar_t *, int, int, char *, int, int);
42 
43 void
44 pass5(void)
45 {
46 	caddr_t err;
47 	int32_t c, blk, frags;
48 	int cg_fatal;
49 	size_t	basesize, sumsize, mapsize;
50 	int excessdirs;
51 	int inomapsize, blkmapsize;
52 	int update_csums, bad_csum, update_bitmaps;
53 	struct fs *fs = &sblock;
54 	struct cg *cg = &cgrp;
55 	diskaddr_t dbase, dmax;
56 	diskaddr_t d;
57 	uint64_t i, j;
58 	struct csum *cs;
59 	struct csum backup_cs;
60 	time_t now;
61 	struct csum cstotal;
62 	struct inodesc idesc;
63 	union {				/* keep lint happy about alignment */
64 		struct cg cg;		/* the rest of buf has the bitmaps */
65 		char buf[MAXBSIZE];
66 	} u;
67 	caddr_t buf = u.buf;
68 	struct cg *newcg = &u.cg;
69 
70 	(void) memset((void *)buf, 0, sizeof (u.buf));
71 	newcg->cg_niblk = fs->fs_ipg;
72 
73 	if (fs->fs_postblformat != FS_DYNAMICPOSTBLFMT) {
74 		pfatal("UNSUPPORTED ROTATIONAL TABLE FORMAT %d\n",
75 			fs->fs_postblformat);
76 		errexit("Program terminated.");
77 		/* NOTREACHED */
78 	}
79 
80 	/* LINTED this subtraction can't overflow and is int32-aligned */
81 	basesize = &newcg->cg_space[0] - (uchar_t *)newcg;
82 
83 	/*
84 	 * We reserve the space for the old rotation summary
85 	 * tables for the benefit of old kernels, but do not
86 	 * maintain them in modern kernels. In time, they could
87 	 * theoretically go away, if we wanted to deal with
88 	 * changing the on-disk format.
89 	 */
90 
91 	/*
92 	 * Note that we don't use any of the cg_*() macros until
93 	 * after cg_sanity() has approved of what we've got.
94 	 */
95 	newcg->cg_btotoff = basesize;
96 	newcg->cg_boff = newcg->cg_btotoff + fs->fs_cpg * sizeof (daddr32_t);
97 	newcg->cg_iusedoff = newcg->cg_boff +
98 		fs->fs_cpg * fs->fs_nrpos * sizeof (uint16_t);
99 	(void) memset(&newcg->cg_space[0], 0, newcg->cg_iusedoff - basesize);
100 
101 	inomapsize = howmany(fs->fs_ipg, NBBY);
102 	newcg->cg_freeoff = newcg->cg_iusedoff + inomapsize;
103 	blkmapsize = howmany(fs->fs_fpg, NBBY);
104 	newcg->cg_nextfreeoff = newcg->cg_freeoff + blkmapsize;
105 	newcg->cg_magic = CG_MAGIC;
106 
107 	sumsize = newcg->cg_iusedoff - newcg->cg_btotoff;
108 	mapsize = newcg->cg_nextfreeoff - newcg->cg_iusedoff;
109 
110 	init_inodesc(&idesc);
111 	idesc.id_type = ADDR;
112 	(void) memset((void *)&cstotal, 0, sizeof (struct csum));
113 	now = time(NULL);
114 
115 	/*
116 	 * If the last fragments in the file system don't make up a
117 	 * full file system block, mark the bits in the blockmap
118 	 * that correspond to those missing fragments as "allocated",
119 	 * so that the last block doesn't get counted as a free block
120 	 * and those missing fragments don't get counted as free frags.
121 	 */
122 	j = blknum(fs, (uint64_t)fs->fs_size + fs->fs_frag - 1);
123 	for (i = fs->fs_size; i < j; i++)
124 		setbmap(i);
125 
126 	/*
127 	 * The cg summaries are not always updated when using
128 	 * logging.  Since we're really concerned with getting a
129 	 * sane filesystem, rather than in trying to debug UFS
130 	 * corner cases, logically we would just always recompute
131 	 * them.  However, it is disconcerting to users to be asked
132 	 * about updating the summaries when, from their point of
133 	 * view, there's been no indication of a problem up to this
134 	 * point.  So, only do it if we find a discrepancy.
135 	 */
136 	update_csums = -1;
137 	update_bitmaps = 0;
138 	for (c = 0; c < fs->fs_ncg; c++) {
139 		backup_cs = cstotal;
140 
141 		/*
142 		 * cg_sanity() will catch i/o errors for us.
143 		 */
144 		(void) getblk(&cgblk, (diskaddr_t)cgtod(fs, c),
145 		    (size_t)fs->fs_cgsize);
146 		err = cg_sanity(cg, c, &cg_fatal);
147 		if (err != NULL) {
148 			pfatal("CG %d: %s\n", c, err);
149 			free((void *)err);
150 			if (cg_fatal)
151 				errexit(
152 	    "Irreparable cylinder group header problem.  Program terminated.");
153 			if (reply("REPAIR") == 0)
154 				errexit("Program terminated.");
155 			fix_cg(cg, c);
156 		}
157 		/*
158 		 * If the on-disk timestamp is in the future, then it
159 		 * by definition is wrong.  Otherwise, if it's in
160 		 * the past, then use that value so that we don't
161 		 * declare a spurious mismatch.
162 		 */
163 		if (now > cg->cg_time)
164 			newcg->cg_time = cg->cg_time;
165 		else
166 			newcg->cg_time = now;
167 		newcg->cg_cgx = c;
168 		dbase = cgbase(fs, c);
169 		dmax = dbase + fs->fs_fpg;
170 		if (dmax > fs->fs_size)
171 			dmax = fs->fs_size;
172 		newcg->cg_ndblk = dmax - dbase;
173 		if (c == fs->fs_ncg - 1)
174 			newcg->cg_ncyl = fs->fs_ncyl % fs->fs_cpg;
175 		else
176 			newcg->cg_ncyl = fs->fs_cpg;
177 		newcg->cg_niblk = sblock.fs_ipg;
178 		newcg->cg_cs.cs_ndir = 0;
179 		newcg->cg_cs.cs_nffree = 0;
180 		newcg->cg_cs.cs_nbfree = 0;
181 		newcg->cg_cs.cs_nifree = fs->fs_ipg;
182 		if ((cg->cg_rotor >= 0) && (cg->cg_rotor < newcg->cg_ndblk))
183 			newcg->cg_rotor = cg->cg_rotor;
184 		else
185 			newcg->cg_rotor = 0;
186 		if ((cg->cg_frotor >= 0) && (cg->cg_frotor < newcg->cg_ndblk))
187 			newcg->cg_frotor = cg->cg_frotor;
188 		else
189 			newcg->cg_frotor = 0;
190 		if ((cg->cg_irotor >= 0) && (cg->cg_irotor < newcg->cg_niblk))
191 			newcg->cg_irotor = cg->cg_irotor;
192 		else
193 			newcg->cg_irotor = 0;
194 		(void) memset((void *)&newcg->cg_frsum[0], 0,
195 		    sizeof (newcg->cg_frsum));
196 		(void) memset((void *)cg_inosused(newcg), 0, (size_t)mapsize);
197 		/* LINTED macro is int32-aligned per newcg->cg_btotoff above */
198 		(void) memset((void *)&cg_blktot(newcg)[0], 0,
199 		    sumsize + mapsize);
200 		j = fs->fs_ipg * c;
201 		for (i = 0; i < fs->fs_ipg; j++, i++) {
202 			switch (statemap[j] & ~(INORPHAN | INDELAYD)) {
203 
204 			case USTATE:
205 				break;
206 
207 			case DSTATE:
208 			case DCLEAR:
209 			case DFOUND:
210 			case DZLINK:
211 				newcg->cg_cs.cs_ndir++;
212 				/* FALLTHROUGH */
213 
214 			case FSTATE:
215 			case FCLEAR:
216 			case FZLINK:
217 			case SSTATE:
218 			case SCLEAR:
219 				newcg->cg_cs.cs_nifree--;
220 				setbit(cg_inosused(newcg), i);
221 				break;
222 
223 			default:
224 				if (j < UFSROOTINO)
225 					break;
226 				errexit("BAD STATE 0x%x FOR INODE I=%d",
227 				    statemap[j], (int)j);
228 			}
229 		}
230 		if (c == 0) {
231 			for (i = 0; i < UFSROOTINO; i++) {
232 				setbit(cg_inosused(newcg), i);
233 				newcg->cg_cs.cs_nifree--;
234 			}
235 		}
236 		/*
237 		 * Count up what fragments and blocks are free, and
238 		 * reflect the relevant section of blockmap[] into
239 		 * newcg's map.
240 		 */
241 		for (i = 0, d = dbase;
242 		    d < dmax;
243 		    d += fs->fs_frag, i += fs->fs_frag) {
244 			frags = 0;
245 			for (j = 0; j < fs->fs_frag; j++) {
246 				if (testbmap(d + j))
247 					continue;
248 				setbit(cg_blksfree(newcg), i + j);
249 				frags++;
250 			}
251 			if (frags == fs->fs_frag) {
252 				newcg->cg_cs.cs_nbfree++;
253 				j = cbtocylno(fs, i);
254 				/* LINTED macro is int32-aligned per above */
255 				cg_blktot(newcg)[j]++;
256 				/* LINTED cg_blks(newcg) is aligned */
257 				cg_blks(fs, newcg, j)[cbtorpos(fs, i)]++;
258 			} else if (frags > 0) {
259 				newcg->cg_cs.cs_nffree += frags;
260 				blk = blkmap(fs, cg_blksfree(newcg), i);
261 				fragacct(fs, blk, newcg->cg_frsum, 1);
262 			}
263 		}
264 		cstotal.cs_nffree += newcg->cg_cs.cs_nffree;
265 		cstotal.cs_nbfree += newcg->cg_cs.cs_nbfree;
266 		cstotal.cs_nifree += newcg->cg_cs.cs_nifree;
267 		cstotal.cs_ndir += newcg->cg_cs.cs_ndir;
268 
269 		/*
270 		 * Note that, just like the kernel, we dynamically
271 		 * allocated an array to hold the csums and stuffed
272 		 * the pointer into the in-core superblock's fs_u.fs_csp
273 		 * field.  This means that the fs_u field contains a
274 		 * random value when the disk version is examined, but
275 		 * fs_cs() gives us a valid pointer nonetheless.
276 		 */
277 		cs = &fs->fs_cs(fs, c);
278 		bad_csum = (memcmp((void *)cs, (void *)&newcg->cg_cs,
279 		    sizeof (*cs)) != 0);
280 
281 		/*
282 		 * Has the user told us what to do yet?  If not, find out.
283 		 */
284 		if (bad_csum && (update_csums == -1)) {
285 			if (preen) {
286 				update_csums = 1;
287 				(void) printf("CORRECTING BAD CG SUMMARIES\n");
288 			} else if (update_csums == -1) {
289 				update_csums = (reply(
290 				    "CORRECT BAD CG SUMMARIES") == 1);
291 			}
292 		}
293 
294 		if (bad_csum && (update_csums == 1)) {
295 			(void) memmove((void *)cs, (void *)&newcg->cg_cs,
296 			    sizeof (*cs));
297 			sbdirty();
298 
299 			(void) memmove((void *)cg, (void *)newcg,
300 			    (size_t)basesize);
301 			/* LINTED per cg_sanity() */
302 			(void) memmove((void *)&cg_blktot(cg)[0],
303 			    /* LINTED macro aligned as above */
304 			    (void *)&cg_blktot(newcg)[0], sumsize);
305 			cgdirty();
306 			(void) printf("CORRECTED SUMMARY FOR CG %d\n", c);
307 		}
308 
309 		excessdirs = cg->cg_cs.cs_ndir - newcg->cg_cs.cs_ndir;
310 		if (excessdirs < 0) {
311 			pfatal("LOST %d DIRECTORIES IN CG %d\n",
312 			    -excessdirs, c);
313 			excessdirs = 0;
314 		}
315 		if (excessdirs > 0) {
316 			if (check_maps((uchar_t *)cg_inosused(newcg),
317 			    (uchar_t *)cg_inosused(cg), inomapsize,
318 			    cg->cg_cgx * fs->fs_ipg, "DIR", 0, excessdirs)) {
319 				if (!verbose)
320 					(void) printf("DIR BITMAP WRONG ");
321 				if (preen || update_bitmaps ||
322 				    reply("FIX") == 1) {
323 					(void) memmove((void *)cg_inosused(cg),
324 					    (void *)cg_inosused(newcg),
325 					    inomapsize);
326 					cgdirty();
327 					if (preen ||
328 					    (!verbose && update_bitmaps))
329 						(void) printf("(CORRECTED)\n");
330 					update_bitmaps = 1;
331 				}
332 			}
333 		}
334 
335 		if (check_maps((uchar_t *)cg_inosused(newcg),
336 		    (uchar_t *)cg_inosused(cg), inomapsize,
337 		    cg->cg_cgx * fs->fs_ipg, "FILE", excessdirs, fs->fs_ipg)) {
338 			if (!verbose)
339 				(void) printf("FILE BITMAP WRONG ");
340 			if (preen || update_bitmaps || reply("FIX") == 1) {
341 				(void) memmove((void *)cg_inosused(cg),
342 				    (void *)cg_inosused(newcg), inomapsize);
343 				cgdirty();
344 				if (preen ||
345 				    (!verbose && update_bitmaps))
346 					(void) printf("(CORRECTED)\n");
347 				update_bitmaps = 1;
348 			}
349 		}
350 
351 		if (check_maps((uchar_t *)cg_blksfree(cg),
352 		    (uchar_t *)cg_blksfree(newcg), blkmapsize,
353 		    cg->cg_cgx * fs->fs_fpg, "FRAG", 0, fs->fs_fpg)) {
354 			if (!verbose)
355 				(void) printf("FRAG BITMAP WRONG ");
356 			if (preen || update_bitmaps || reply("FIX") == 1) {
357 				(void) memmove((void *)cg_blksfree(cg),
358 				    (void *)cg_blksfree(newcg), blkmapsize);
359 				cgdirty();
360 				if (preen ||
361 				    (!verbose && update_bitmaps))
362 					(void) printf("(CORRECTED)\n");
363 				update_bitmaps = 1;
364 			}
365 		}
366 
367 		/*
368 		 * Fixing one set of problems often shows up more in the
369 		 * same cg.  Just to make sure, go back and check it
370 		 * again if we found something this time through.
371 		 */
372 		if (cgisdirty()) {
373 			cgflush();
374 			cstotal = backup_cs;
375 			c--;
376 		}
377 	}
378 
379 	if ((fflag || !(islog && islogok)) &&
380 	    (memcmp((void *)&cstotal, (void *)&fs->fs_cstotal,
381 	    sizeof (struct csum)) != 0)) {
382 		if (dofix(&idesc, "CORRECT GLOBAL SUMMARY")) {
383 			(void) memmove((void *)&fs->fs_cstotal,
384 			    (void *)&cstotal, sizeof (struct csum));
385 			fs->fs_ronly = 0;
386 			fs->fs_fmod = 0;
387 			sbdirty();
388 		} else {
389 			iscorrupt = 1;
390 		}
391 	}
392 }
393 
394 /*
395  * Compare two allocation bitmaps, reporting any discrepancies.
396  *
397  * If a mismatch is found, if the bit is set in map1, it's considered
398  * to be an indication that the corresponding resource is supposed
399  * to be free, but isn't.  Otherwise, it's considered marked as allocated
400  * but not found to be so.  In other words, if the two maps being compared
401  * use a set bit to indicate something is free, pass the on-disk map
402  * first.  Otherwise, pass the calculated map first.
403  */
404 static int
405 check_maps(
406 	uchar_t *map1,	/* map of claimed allocations */
407 	uchar_t *map2,	/* map of determined allocations */
408 	int mapsize,	/* size of above two maps */
409 	int startvalue,	/* resource value for first element in map */
410 	char *name,	/* name of resource found in maps */
411 	int skip,	/* number of entries to skip before starting to free */
412 	int limit)	/* limit on number of entries to free */
413 {
414 	long i, j, k, l, m, n, size;
415 	int astart, aend, ustart, uend;
416 	int mismatch;
417 
418 	mismatch = 0;
419 	astart = ustart = aend = uend = -1;
420 	for (i = 0; i < mapsize; i++) {
421 		j = *map1++;
422 		k = *map2++;
423 		if (j == k)
424 			continue;
425 		for (m = 0, l = 1; m < NBBY; m++, l <<= 1) {
426 			if ((j & l) == (k & l))
427 				continue;
428 			n = startvalue + i * NBBY + m;
429 			if ((j & l) != 0) {
430 				if (astart == -1) {
431 					astart = aend = n;
432 					continue;
433 				}
434 				if (aend + 1 == n) {
435 					aend = n;
436 					continue;
437 				}
438 				if (verbose) {
439 					if (astart == aend)
440 						pwarn(
441 			    "ALLOCATED %s %d WAS MARKED FREE ON DISK\n",
442 						    name, astart);
443 					else
444 						pwarn(
445 			    "ALLOCATED %sS %d-%d WERE MARKED FREE ON DISK\n",
446 						    name, astart, aend);
447 				}
448 				mismatch = 1;
449 				astart = aend = n;
450 			} else {
451 				if (ustart == -1) {
452 					ustart = uend = n;
453 					continue;
454 				}
455 				if (uend + 1 == n) {
456 					uend = n;
457 					continue;
458 				}
459 				size = uend - ustart + 1;
460 				if (size <= skip) {
461 					skip -= size;
462 					ustart = uend = n;
463 					continue;
464 				}
465 				if (skip > 0) {
466 					ustart += skip;
467 					size -= skip;
468 					skip = 0;
469 				}
470 				if (size > limit)
471 					size = limit;
472 				if (verbose) {
473 					if (size == 1)
474 						pwarn(
475 			    "UNALLOCATED %s %d WAS MARKED USED ON DISK\n",
476 						    name, ustart);
477 					else
478 						pwarn(
479 			    "UNALLOCATED %sS %d-%ld WERE MARKED USED ON DISK\n",
480 						    name, ustart,
481 						    ustart + size - 1);
482 				}
483 				mismatch = 1;
484 				limit -= size;
485 				if (limit <= 0)
486 					return (mismatch);
487 				ustart = uend = n;
488 			}
489 		}
490 	}
491 	if (astart != -1) {
492 		if (verbose) {
493 			if (astart == aend)
494 				pwarn(
495 			    "ALLOCATED %s %d WAS MARKED FREE ON DISK\n",
496 				    name, astart);
497 			else
498 				pwarn(
499 			    "ALLOCATED %sS %d-%d WERE MARKED FREE ON DISK\n",
500 				    name, astart, aend);
501 		}
502 		mismatch = 1;
503 	}
504 	if (ustart != -1) {
505 		size = uend - ustart + 1;
506 		if (size <= skip)
507 			return (mismatch);
508 		if (skip > 0) {
509 			ustart += skip;
510 			size -= skip;
511 		}
512 		if (size > limit)
513 			size = limit;
514 		if (verbose) {
515 			if (size == 1)
516 				pwarn(
517 			    "UNALLOCATED %s %d WAS MARKED USED ON DISK\n",
518 				    name, ustart);
519 			else
520 				pwarn(
521 		    "UNALLOCATED %sS %d-%ld WERE MARKED USED ON DISK\n",
522 				    name, ustart, ustart + size - 1);
523 		}
524 		mismatch = 1;
525 	}
526 	return (mismatch);
527 }
528