xref: /dragonfly/sbin/dump/traverse.c (revision 6e285212)
1 /*-
2  * Copyright (c) 1980, 1988, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  * @(#)traverse.c	8.7 (Berkeley) 6/15/95
34  * $FreeBSD: src/sbin/dump/traverse.c,v 1.10.2.6 2003/04/14 20:10:35 johan Exp $
35  * $DragonFly: src/sbin/dump/traverse.c,v 1.2 2003/06/17 04:27:32 dillon Exp $
36  */
37 
38 #include <sys/param.h>
39 #include <sys/stat.h>
40 #ifdef sunos
41 #include <sys/vnode.h>
42 
43 #include <ufs/fs.h>
44 #include <ufs/fsdir.h>
45 #include <ufs/inode.h>
46 #else
47 #include <ufs/ufs/dir.h>
48 #include <ufs/ufs/dinode.h>
49 #include <ufs/ffs/fs.h>
50 #endif
51 
52 #include <protocols/dumprestore.h>
53 
54 #include <ctype.h>
55 #include <stdio.h>
56 #ifdef __STDC__
57 #include <errno.h>
58 #include <string.h>
59 #include <unistd.h>
60 #endif
61 
62 #include "dump.h"
63 
64 #define	HASDUMPEDFILE	0x1
65 #define	HASSUBDIRS	0x2
66 
67 #ifdef	FS_44INODEFMT
68 typedef	quad_t fsizeT;
69 #else
70 typedef	long fsizeT;
71 #endif
72 
73 static	int dirindir __P((ino_t ino, daddr_t blkno, int level, long *size,
74     long *tapesize, int nodump));
75 static	void dmpindir __P((ino_t ino, daddr_t blk, int level, fsizeT *size));
76 static	int searchdir __P((ino_t ino, daddr_t blkno, long size, long filesize,
77     long *tapesize, int nodump));
78 
79 /*
80  * This is an estimation of the number of TP_BSIZE blocks in the file.
81  * It estimates the number of blocks in files with holes by assuming
82  * that all of the blocks accounted for by di_blocks are data blocks
83  * (when some of the blocks are usually used for indirect pointers);
84  * hence the estimate may be high.
85  */
86 long
87 blockest(dp)
88 	register struct dinode *dp;
89 {
90 	long blkest, sizeest;
91 
92 	/*
93 	 * dp->di_size is the size of the file in bytes.
94 	 * dp->di_blocks stores the number of sectors actually in the file.
95 	 * If there are more sectors than the size would indicate, this just
96 	 *	means that there are indirect blocks in the file or unused
97 	 *	sectors in the last file block; we can safely ignore these
98 	 *	(blkest = sizeest below).
99 	 * If the file is bigger than the number of sectors would indicate,
100 	 *	then the file has holes in it.	In this case we must use the
101 	 *	block count to estimate the number of data blocks used, but
102 	 *	we use the actual size for estimating the number of indirect
103 	 *	dump blocks (sizeest vs. blkest in the indirect block
104 	 *	calculation).
105 	 */
106 	blkest = howmany(dbtob(dp->di_blocks), TP_BSIZE);
107 	sizeest = howmany(dp->di_size, TP_BSIZE);
108 	if (blkest > sizeest)
109 		blkest = sizeest;
110 	if (dp->di_size > sblock->fs_bsize * NDADDR) {
111 		/* calculate the number of indirect blocks on the dump tape */
112 		blkest +=
113 			howmany(sizeest - NDADDR * sblock->fs_bsize / TP_BSIZE,
114 			TP_NINDIR);
115 	}
116 	return (blkest + 1);
117 }
118 
119 /* Auxiliary macro to pick up files changed since previous dump. */
120 #define	CHANGEDSINCE(dp, t) \
121 	((dp)->di_mtime >= (t) || (dp)->di_ctime >= (t))
122 
123 /* The WANTTODUMP macro decides whether a file should be dumped. */
124 #ifdef UF_NODUMP
125 #define	WANTTODUMP(dp) \
126 	(CHANGEDSINCE(dp, spcl.c_ddate) && \
127 	 (nonodump || ((dp)->di_flags & UF_NODUMP) != UF_NODUMP))
128 #else
129 #define	WANTTODUMP(dp) CHANGEDSINCE(dp, spcl.c_ddate)
130 #endif
131 
132 /*
133  * Dump pass 1.
134  *
135  * Walk the inode list for a filesystem to find all allocated inodes
136  * that have been modified since the previous dump time. Also, find all
137  * the directories in the filesystem.
138  */
139 int
140 mapfiles(maxino, tapesize)
141 	ino_t maxino;
142 	long *tapesize;
143 {
144 	register int mode;
145 	register ino_t ino;
146 	register struct dinode *dp;
147 	int anydirskipped = 0;
148 
149 	for (ino = ROOTINO; ino < maxino; ino++) {
150 		dp = getino(ino);
151 		if ((mode = (dp->di_mode & IFMT)) == 0)
152 			continue;
153 		/*
154 		 * Everything must go in usedinomap so that a check
155 		 * for "in dumpdirmap but not in usedinomap" to detect
156 		 * dirs with nodump set has a chance of succeeding
157 		 * (this is used in mapdirs()).
158 		 */
159 		SETINO(ino, usedinomap);
160 		if (mode == IFDIR)
161 			SETINO(ino, dumpdirmap);
162 		if (WANTTODUMP(dp)) {
163 			SETINO(ino, dumpinomap);
164 			if (mode != IFREG && mode != IFDIR && mode != IFLNK)
165 				*tapesize += 1;
166 			else
167 				*tapesize += blockest(dp);
168 			continue;
169 		}
170 		if (mode == IFDIR) {
171 			if (!nonodump && (dp->di_flags & UF_NODUMP))
172 				CLRINO(ino, usedinomap);
173 			anydirskipped = 1;
174 		}
175 	}
176 	/*
177 	 * Restore gets very upset if the root is not dumped,
178 	 * so ensure that it always is dumped.
179 	 */
180 	SETINO(ROOTINO, dumpinomap);
181 	return (anydirskipped);
182 }
183 
184 /*
185  * Dump pass 2.
186  *
187  * Scan each directory on the filesystem to see if it has any modified
188  * files in it. If it does, and has not already been added to the dump
189  * list (because it was itself modified), then add it. If a directory
190  * has not been modified itself, contains no modified files and has no
191  * subdirectories, then it can be deleted from the dump list and from
192  * the list of directories. By deleting it from the list of directories,
193  * its parent may now qualify for the same treatment on this or a later
194  * pass using this algorithm.
195  */
196 int
197 mapdirs(maxino, tapesize)
198 	ino_t maxino;
199 	long *tapesize;
200 {
201 	register struct	dinode *dp;
202 	register int i, isdir, nodump;
203 	register char *map;
204 	register ino_t ino;
205 	struct dinode di;
206 	long filesize;
207 	int ret, change = 0;
208 
209 	isdir = 0;		/* XXX just to get gcc to shut up */
210 	for (map = dumpdirmap, ino = 1; ino < maxino; ino++) {
211 		if (((ino - 1) % NBBY) == 0)	/* map is offset by 1 */
212 			isdir = *map++;
213 		else
214 			isdir >>= 1;
215 		/*
216 		 * If a directory has been removed from usedinomap, it
217 		 * either has the nodump flag set, or has inherited
218 		 * it.  Although a directory can't be in dumpinomap if
219 		 * it isn't in usedinomap, we have to go through it to
220 		 * propagate the nodump flag.
221 		 */
222 		nodump = !nonodump && (TSTINO(ino, usedinomap) == 0);
223 		if ((isdir & 1) == 0 || (TSTINO(ino, dumpinomap) && !nodump))
224 			continue;
225 		dp = getino(ino);
226 		di = *dp;	/* inode buf may change in searchdir(). */
227 		filesize = di.di_size;
228 		for (ret = 0, i = 0; filesize > 0 && i < NDADDR; i++) {
229 			if (di.di_db[i] != 0)
230 				ret |= searchdir(ino, di.di_db[i],
231 					(long)dblksize(sblock, dp, i),
232 					filesize, tapesize, nodump);
233 			if (ret & HASDUMPEDFILE)
234 				filesize = 0;
235 			else
236 				filesize -= sblock->fs_bsize;
237 		}
238 		for (i = 0; filesize > 0 && i < NIADDR; i++) {
239 			if (di.di_ib[i] == 0)
240 				continue;
241 			ret |= dirindir(ino, di.di_ib[i], i, &filesize,
242 			    tapesize, nodump);
243 		}
244 		if (ret & HASDUMPEDFILE) {
245 			SETINO(ino, dumpinomap);
246 			*tapesize += blockest(dp);
247 			change = 1;
248 			continue;
249 		}
250 		if (nodump) {
251 			if (ret & HASSUBDIRS)
252 				change = 1;	/* subdirs inherit nodump */
253 			CLRINO(ino, dumpdirmap);
254 		} else if ((ret & HASSUBDIRS) == 0)
255 			if (!TSTINO(ino, dumpinomap)) {
256 				CLRINO(ino, dumpdirmap);
257 				change = 1;
258 			}
259 	}
260 	return (change);
261 }
262 
263 /*
264  * Read indirect blocks, and pass the data blocks to be searched
265  * as directories. Quit as soon as any entry is found that will
266  * require the directory to be dumped.
267  */
268 static int
269 dirindir(ino, blkno, ind_level, filesize, tapesize, nodump)
270 	ino_t ino;
271 	daddr_t blkno;
272 	int ind_level;
273 	long *filesize;
274 	long *tapesize;
275 	int nodump;
276 {
277 	int ret = 0;
278 	register int i;
279 	daddr_t	idblk[MAXNINDIR];
280 
281 	bread(fsbtodb(sblock, blkno), (char *)idblk, (int)sblock->fs_bsize);
282 	if (ind_level <= 0) {
283 		for (i = 0; *filesize > 0 && i < NINDIR(sblock); i++) {
284 			blkno = idblk[i];
285 			if (blkno != 0)
286 				ret |= searchdir(ino, blkno, sblock->fs_bsize,
287 					*filesize, tapesize, nodump);
288 			if (ret & HASDUMPEDFILE)
289 				*filesize = 0;
290 			else
291 				*filesize -= sblock->fs_bsize;
292 		}
293 		return (ret);
294 	}
295 	ind_level--;
296 	for (i = 0; *filesize > 0 && i < NINDIR(sblock); i++) {
297 		blkno = idblk[i];
298 		if (blkno != 0)
299 			ret |= dirindir(ino, blkno, ind_level, filesize,
300 			    tapesize, nodump);
301 	}
302 	return (ret);
303 }
304 
305 /*
306  * Scan a disk block containing directory information looking to see if
307  * any of the entries are on the dump list and to see if the directory
308  * contains any subdirectories.
309  */
310 static int
311 searchdir(ino, blkno, size, filesize, tapesize, nodump)
312 	ino_t ino;
313 	daddr_t blkno;
314 	register long size;
315 	long filesize;
316 	long *tapesize;
317 	int nodump;
318 {
319 	register struct direct *dp;
320 	register struct dinode *ip;
321 	register long loc, ret = 0;
322 	char dblk[MAXBSIZE];
323 
324 	bread(fsbtodb(sblock, blkno), dblk, (int)size);
325 	if (filesize < size)
326 		size = filesize;
327 	for (loc = 0; loc < size; ) {
328 		dp = (struct direct *)(dblk + loc);
329 		if (dp->d_reclen == 0) {
330 			msg("corrupted directory, inumber %d\n", ino);
331 			break;
332 		}
333 		loc += dp->d_reclen;
334 		if (dp->d_ino == 0)
335 			continue;
336 		if (dp->d_name[0] == '.') {
337 			if (dp->d_name[1] == '\0')
338 				continue;
339 			if (dp->d_name[1] == '.' && dp->d_name[2] == '\0')
340 				continue;
341 		}
342 		if (nodump) {
343 			ip = getino(dp->d_ino);
344 			if (TSTINO(dp->d_ino, dumpinomap)) {
345 				CLRINO(dp->d_ino, dumpinomap);
346 				CLRINO(dp->d_ino, usedinomap);
347 				*tapesize -= blockest(ip);
348 			}
349 			/* Add back to dumpdirmap to propagate nodump. */
350 			if ((ip->di_mode & IFMT) == IFDIR) {
351 				SETINO(dp->d_ino, dumpdirmap);
352 				ret |= HASSUBDIRS;
353 			}
354 		} else {
355 			if (TSTINO(dp->d_ino, dumpinomap)) {
356 				ret |= HASDUMPEDFILE;
357 				if (ret & HASSUBDIRS)
358 					break;
359 			}
360 			if (TSTINO(dp->d_ino, dumpdirmap)) {
361 				ret |= HASSUBDIRS;
362 				if (ret & HASDUMPEDFILE)
363 					break;
364 			}
365 		}
366 	}
367 	return (ret);
368 }
369 
370 /*
371  * Dump passes 3 and 4.
372  *
373  * Dump the contents of an inode to tape.
374  */
375 void
376 dumpino(dp, ino)
377 	register struct dinode *dp;
378 	ino_t ino;
379 {
380 	int ind_level, cnt;
381 	fsizeT size;
382 	char buf[TP_BSIZE];
383 
384 	if (newtape) {
385 		newtape = 0;
386 		dumpmap(dumpinomap, TS_BITS, ino);
387 	}
388 	CLRINO(ino, dumpinomap);
389 	spcl.c_dinode = *dp;
390 	spcl.c_type = TS_INODE;
391 	spcl.c_count = 0;
392 	switch (dp->di_mode & S_IFMT) {
393 
394 	case 0:
395 		/*
396 		 * Freed inode.
397 		 */
398 		return;
399 
400 	case S_IFLNK:
401 		/*
402 		 * Check for short symbolic link.
403 		 */
404 #ifdef FS_44INODEFMT
405 		if (dp->di_size > 0 &&
406 		    dp->di_size < sblock->fs_maxsymlinklen) {
407 			spcl.c_addr[0] = 1;
408 			spcl.c_count = 1;
409 			writeheader(ino);
410 			memmove(buf, dp->di_shortlink, (u_long)dp->di_size);
411 			buf[dp->di_size] = '\0';
412 			writerec(buf, 0);
413 			return;
414 		}
415 #endif
416 		/* fall through */
417 
418 	case S_IFDIR:
419 	case S_IFREG:
420 		if (dp->di_size > 0)
421 			break;
422 		/* fall through */
423 
424 	case S_IFIFO:
425 	case S_IFSOCK:
426 	case S_IFCHR:
427 	case S_IFBLK:
428 		writeheader(ino);
429 		return;
430 
431 	default:
432 		msg("Warning: undefined file type 0%o\n", dp->di_mode & IFMT);
433 		return;
434 	}
435 	if (dp->di_size > NDADDR * sblock->fs_bsize)
436 		cnt = NDADDR * sblock->fs_frag;
437 	else
438 		cnt = howmany(dp->di_size, sblock->fs_fsize);
439 	blksout(&dp->di_db[0], cnt, ino);
440 	if ((size = dp->di_size - NDADDR * sblock->fs_bsize) <= 0)
441 		return;
442 	for (ind_level = 0; ind_level < NIADDR; ind_level++) {
443 		dmpindir(ino, dp->di_ib[ind_level], ind_level, &size);
444 		if (size <= 0)
445 			return;
446 	}
447 }
448 
449 /*
450  * Read indirect blocks, and pass the data blocks to be dumped.
451  */
452 static void
453 dmpindir(ino, blk, ind_level, size)
454 	ino_t ino;
455 	daddr_t blk;
456 	int ind_level;
457 	fsizeT *size;
458 {
459 	int i, cnt;
460 	daddr_t idblk[MAXNINDIR];
461 
462 	if (blk != 0)
463 		bread(fsbtodb(sblock, blk), (char *)idblk, (int) sblock->fs_bsize);
464 	else
465 		memset(idblk, 0, (int)sblock->fs_bsize);
466 	if (ind_level <= 0) {
467 		if (*size < NINDIR(sblock) * sblock->fs_bsize)
468 			cnt = howmany(*size, sblock->fs_fsize);
469 		else
470 			cnt = NINDIR(sblock) * sblock->fs_frag;
471 		*size -= NINDIR(sblock) * sblock->fs_bsize;
472 		blksout(&idblk[0], cnt, ino);
473 		return;
474 	}
475 	ind_level--;
476 	for (i = 0; i < NINDIR(sblock); i++) {
477 		dmpindir(ino, idblk[i], ind_level, size);
478 		if (*size <= 0)
479 			return;
480 	}
481 }
482 
483 /*
484  * Collect up the data into tape record sized buffers and output them.
485  */
486 void
487 blksout(blkp, frags, ino)
488 	daddr_t *blkp;
489 	int frags;
490 	ino_t ino;
491 {
492 	register daddr_t *bp;
493 	int i, j, count, blks, tbperdb;
494 
495 	blks = howmany(frags * sblock->fs_fsize, TP_BSIZE);
496 	tbperdb = sblock->fs_bsize >> tp_bshift;
497 	for (i = 0; i < blks; i += TP_NINDIR) {
498 		if (i + TP_NINDIR > blks)
499 			count = blks;
500 		else
501 			count = i + TP_NINDIR;
502 		for (j = i; j < count; j++)
503 			if (blkp[j / tbperdb] != 0)
504 				spcl.c_addr[j - i] = 1;
505 			else
506 				spcl.c_addr[j - i] = 0;
507 		spcl.c_count = count - i;
508 		writeheader(ino);
509 		bp = &blkp[i / tbperdb];
510 		for (j = i; j < count; j += tbperdb, bp++)
511 			if (*bp != 0) {
512 				if (j + tbperdb <= count)
513 					dumpblock(*bp, (int)sblock->fs_bsize);
514 				else
515 					dumpblock(*bp, (count - j) * TP_BSIZE);
516 			}
517 		spcl.c_type = TS_ADDR;
518 	}
519 }
520 
521 /*
522  * Dump a map to the tape.
523  */
524 void
525 dumpmap(map, type, ino)
526 	char *map;
527 	int type;
528 	ino_t ino;
529 {
530 	register int i;
531 	char *cp;
532 
533 	spcl.c_type = type;
534 	spcl.c_count = howmany(mapsize * sizeof(char), TP_BSIZE);
535 	writeheader(ino);
536 	for (i = 0, cp = map; i < spcl.c_count; i++, cp += TP_BSIZE)
537 		writerec(cp, 0);
538 }
539 
540 /*
541  * Write a header record to the dump tape.
542  */
543 void
544 writeheader(ino)
545 	ino_t ino;
546 {
547 	register int32_t sum, cnt, *lp;
548 
549 	spcl.c_inumber = ino;
550 	spcl.c_magic = NFS_MAGIC;
551 	spcl.c_checksum = 0;
552 	lp = (int32_t *)&spcl;
553 	sum = 0;
554 	cnt = sizeof(union u_spcl) / (4 * sizeof(int32_t));
555 	while (--cnt >= 0) {
556 		sum += *lp++;
557 		sum += *lp++;
558 		sum += *lp++;
559 		sum += *lp++;
560 	}
561 	spcl.c_checksum = CHECKSUM - sum;
562 	writerec((char *)&spcl, 1);
563 }
564 
565 struct dinode *
566 getino(inum)
567 	ino_t inum;
568 {
569 	static daddr_t minino, maxino;
570 	static struct dinode inoblock[MAXINOPB];
571 
572 	curino = inum;
573 	if (inum >= minino && inum < maxino)
574 		return (&inoblock[inum - minino]);
575 	bread(fsbtodb(sblock, ino_to_fsba(sblock, inum)), (char *)inoblock,
576 	    (int)sblock->fs_bsize);
577 	minino = inum - (inum % INOPB(sblock));
578 	maxino = minino + INOPB(sblock);
579 	return (&inoblock[inum - minino]);
580 }
581 
582 /*
583  * Read a chunk of data from the disk.
584  * Try to recover from hard errors by reading in sector sized pieces.
585  * Error recovery is attempted at most BREADEMAX times before seeking
586  * consent from the operator to continue.
587  */
588 int	breaderrors = 0;
589 #define	BREADEMAX 32
590 
591 void
592 bread(blkno, buf, size)
593 	daddr_t blkno;
594 	char *buf;
595 	int size;
596 {
597 	int cnt, i;
598 
599 loop:
600 	cnt = cread(diskfd, buf, size, ((off_t)blkno << dev_bshift));
601 	if (cnt == size)
602 		return;
603 	if (blkno + (size / dev_bsize) > fsbtodb(sblock, sblock->fs_size)) {
604 		/*
605 		 * Trying to read the final fragment.
606 		 *
607 		 * NB - dump only works in TP_BSIZE blocks, hence
608 		 * rounds `dev_bsize' fragments up to TP_BSIZE pieces.
609 		 * It should be smarter about not actually trying to
610 		 * read more than it can get, but for the time being
611 		 * we punt and scale back the read only when it gets
612 		 * us into trouble. (mkm 9/25/83)
613 		 */
614 		size -= dev_bsize;
615 		goto loop;
616 	}
617 	if (cnt == -1)
618 		msg("read error from %s: %s: [block %d]: count=%d\n",
619 			disk, strerror(errno), blkno, size);
620 	else
621 		msg("short read error from %s: [block %d]: count=%d, got=%d\n",
622 			disk, blkno, size, cnt);
623 	if (++breaderrors > BREADEMAX) {
624 		msg("More than %d block read errors from %s\n",
625 			BREADEMAX, disk);
626 		broadcast("DUMP IS AILING!\n");
627 		msg("This is an unrecoverable error.\n");
628 		if (!query("Do you want to attempt to continue?")){
629 			dumpabort(0);
630 			/*NOTREACHED*/
631 		} else
632 			breaderrors = 0;
633 	}
634 	/*
635 	 * Zero buffer, then try to read each sector of buffer separately,
636 	 * and bypass the cache.
637 	 */
638 	memset(buf, 0, size);
639 	for (i = 0; i < size; i += dev_bsize, buf += dev_bsize, blkno++) {
640 		if ((cnt = pread(diskfd, buf, (int)dev_bsize,
641 		    ((off_t)blkno << dev_bshift))) == dev_bsize)
642 			continue;
643 		if (cnt == -1) {
644 			msg("read error from %s: %s: [sector %d]: count=%d\n",
645 				disk, strerror(errno), blkno, dev_bsize);
646 			continue;
647 		}
648 		msg("short read error from %s: [sector %d]: count=%d, got=%d\n",
649 			disk, blkno, dev_bsize, cnt);
650 	}
651 }
652