xref: /freebsd/sbin/dump/traverse.c (revision 39beb93c3f8bdbf72a61fda42300b5ebed7390c8)
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  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 #ifndef lint
31 #if 0
32 static char sccsid[] = "@(#)traverse.c	8.7 (Berkeley) 6/15/95";
33 #endif
34 static const char rcsid[] =
35   "$FreeBSD$";
36 #endif /* not lint */
37 
38 #include <sys/param.h>
39 #include <sys/stat.h>
40 
41 #include <ufs/ufs/dir.h>
42 #include <ufs/ufs/dinode.h>
43 #include <ufs/ffs/fs.h>
44 
45 #include <protocols/dumprestore.h>
46 
47 #include <ctype.h>
48 #include <errno.h>
49 #include <inttypes.h>
50 #include <limits.h>
51 #include <stdio.h>
52 #include <stdlib.h>
53 #include <string.h>
54 #include <timeconv.h>
55 #include <unistd.h>
56 
57 #include "dump.h"
58 
59 union dinode {
60 	struct ufs1_dinode dp1;
61 	struct ufs2_dinode dp2;
62 };
63 #define	DIP(dp, field) \
64 	((sblock->fs_magic == FS_UFS1_MAGIC) ? \
65 	(dp)->dp1.field : (dp)->dp2.field)
66 #define DIP_SET(dp, field, val) do {\
67 	if (sblock->fs_magic == FS_UFS1_MAGIC) \
68 		(dp)->dp1.field = (val); \
69 	else \
70 		(dp)->dp2.field = (val); \
71 	} while (0)
72 
73 #define	HASDUMPEDFILE	0x1
74 #define	HASSUBDIRS	0x2
75 
76 static	int dirindir(ino_t ino, ufs2_daddr_t blkno, int level, long *size,
77     long *tapesize, int nodump, ino_t maxino);
78 static	void dmpindir(union dinode *dp, ino_t ino, ufs2_daddr_t blk, int level,
79     off_t *size);
80 static	void ufs1_blksout(ufs1_daddr_t *blkp, int frags, ino_t ino);
81 static	void ufs2_blksout(union dinode *dp, ufs2_daddr_t *blkp, int frags,
82     ino_t ino, int last);
83 static	int appendextdata(union dinode *dp);
84 static	void writeextdata(union dinode *dp, ino_t ino, int added);
85 static	int searchdir(ino_t ino, ufs2_daddr_t blkno, long size, long filesize,
86     long *tapesize, int nodump, ino_t maxino);
87 static	long blockest(union dinode *dp);
88 
89 /*
90  * This is an estimation of the number of TP_BSIZE blocks in the file.
91  * It estimates the number of blocks in files with holes by assuming
92  * that all of the blocks accounted for by di_blocks are data blocks
93  * (when some of the blocks are usually used for indirect pointers);
94  * hence the estimate may be high.
95  */
96 static long
97 blockest(union dinode *dp)
98 {
99 	long blkest, sizeest;
100 
101 	/*
102 	 * dp->di_size is the size of the file in bytes.
103 	 * dp->di_blocks stores the number of sectors actually in the file.
104 	 * If there are more sectors than the size would indicate, this just
105 	 *	means that there are indirect blocks in the file or unused
106 	 *	sectors in the last file block; we can safely ignore these
107 	 *	(blkest = sizeest below).
108 	 * If the file is bigger than the number of sectors would indicate,
109 	 *	then the file has holes in it.	In this case we must use the
110 	 *	block count to estimate the number of data blocks used, but
111 	 *	we use the actual size for estimating the number of indirect
112 	 *	dump blocks (sizeest vs. blkest in the indirect block
113 	 *	calculation).
114 	 */
115 	if ((DIP(dp, di_flags) & SF_SNAPSHOT) != 0)
116 		return (1);
117 	blkest = howmany(dbtob(DIP(dp, di_blocks)), TP_BSIZE);
118 	sizeest = howmany(DIP(dp, di_size), TP_BSIZE);
119 	if (blkest > sizeest)
120 		blkest = sizeest;
121 	if (DIP(dp, di_size) > sblock->fs_bsize * NDADDR) {
122 		/* calculate the number of indirect blocks on the dump tape */
123 		blkest +=
124 			howmany(sizeest - NDADDR * sblock->fs_bsize / TP_BSIZE,
125 			TP_NINDIR);
126 	}
127 	return (blkest + 1);
128 }
129 
130 /* Auxiliary macro to pick up files changed since previous dump. */
131 #define	CHANGEDSINCE(dp, t) \
132 	(DIP(dp, di_mtime) >= (t) || DIP(dp, di_ctime) >= (t))
133 
134 /* The WANTTODUMP macro decides whether a file should be dumped. */
135 #ifdef UF_NODUMP
136 #define	WANTTODUMP(dp) \
137 	(CHANGEDSINCE(dp, spcl.c_ddate) && \
138 	 (nonodump || (DIP(dp, di_flags) & UF_NODUMP) != UF_NODUMP))
139 #else
140 #define	WANTTODUMP(dp) CHANGEDSINCE(dp, spcl.c_ddate)
141 #endif
142 
143 /*
144  * Dump pass 1.
145  *
146  * Walk the inode list for a file system to find all allocated inodes
147  * that have been modified since the previous dump time. Also, find all
148  * the directories in the file system.
149  */
150 int
151 mapfiles(ino_t maxino, long *tapesize)
152 {
153 	int i, cg, mode, inosused;
154 	int anydirskipped = 0;
155 	union dinode *dp;
156 	struct cg *cgp;
157 	ino_t ino;
158 	u_char *cp;
159 
160 	if ((cgp = malloc(sblock->fs_cgsize)) == NULL)
161 		quit("mapfiles: cannot allocate memory.\n");
162 	for (cg = 0; cg < sblock->fs_ncg; cg++) {
163 		ino = cg * sblock->fs_ipg;
164 		bread(fsbtodb(sblock, cgtod(sblock, cg)), (char *)cgp,
165 		    sblock->fs_cgsize);
166 		if (sblock->fs_magic == FS_UFS2_MAGIC)
167 			inosused = cgp->cg_initediblk;
168 		else
169 			inosused = sblock->fs_ipg;
170 		/*
171 		 * If we are using soft updates, then we can trust the
172 		 * cylinder group inode allocation maps to tell us which
173 		 * inodes are allocated. We will scan the used inode map
174 		 * to find the inodes that are really in use, and then
175 		 * read only those inodes in from disk.
176 		 */
177 		if (sblock->fs_flags & FS_DOSOFTDEP) {
178 			if (!cg_chkmagic(cgp))
179 				quit("mapfiles: cg %d: bad magic number\n", cg);
180 			cp = &cg_inosused(cgp)[(inosused - 1) / CHAR_BIT];
181 			for ( ; inosused > 0; inosused -= CHAR_BIT, cp--) {
182 				if (*cp == 0)
183 					continue;
184 				for (i = 1 << (CHAR_BIT - 1); i > 0; i >>= 1) {
185 					if (*cp & i)
186 						break;
187 					inosused--;
188 				}
189 				break;
190 			}
191 			if (inosused <= 0)
192 				continue;
193 		}
194 		for (i = 0; i < inosused; i++, ino++) {
195 			if (ino < ROOTINO ||
196 			    (dp = getino(ino, &mode)) == NULL ||
197 			    (mode & IFMT) == 0)
198 				continue;
199 			if (ino >= maxino) {
200 				msg("Skipping inode %d >= maxino %d\n",
201 				    ino, maxino);
202 				continue;
203 			}
204 			/*
205 			 * Everything must go in usedinomap so that a check
206 			 * for "in dumpdirmap but not in usedinomap" to detect
207 			 * dirs with nodump set has a chance of succeeding
208 			 * (this is used in mapdirs()).
209 			 */
210 			SETINO(ino, usedinomap);
211 			if (mode == IFDIR)
212 				SETINO(ino, dumpdirmap);
213 			if (WANTTODUMP(dp)) {
214 				SETINO(ino, dumpinomap);
215 				if (mode != IFREG &&
216 				    mode != IFDIR &&
217 				    mode != IFLNK)
218 					*tapesize += 1;
219 				else
220 					*tapesize += blockest(dp);
221 				continue;
222 			}
223 			if (mode == IFDIR) {
224 				if (!nonodump &&
225 				    (DIP(dp, di_flags) & UF_NODUMP))
226 					CLRINO(ino, usedinomap);
227 				anydirskipped = 1;
228 			}
229 		}
230 	}
231 	/*
232 	 * Restore gets very upset if the root is not dumped,
233 	 * so ensure that it always is dumped.
234 	 */
235 	SETINO(ROOTINO, dumpinomap);
236 	return (anydirskipped);
237 }
238 
239 /*
240  * Dump pass 2.
241  *
242  * Scan each directory on the file system to see if it has any modified
243  * files in it. If it does, and has not already been added to the dump
244  * list (because it was itself modified), then add it. If a directory
245  * has not been modified itself, contains no modified files and has no
246  * subdirectories, then it can be deleted from the dump list and from
247  * the list of directories. By deleting it from the list of directories,
248  * its parent may now qualify for the same treatment on this or a later
249  * pass using this algorithm.
250  */
251 int
252 mapdirs(ino_t maxino, long *tapesize)
253 {
254 	union dinode *dp;
255 	int i, isdir, nodump;
256 	char *map;
257 	ino_t ino;
258 	union dinode di;
259 	long filesize;
260 	int ret, change = 0;
261 
262 	isdir = 0;		/* XXX just to get gcc to shut up */
263 	for (map = dumpdirmap, ino = 1; ino < maxino; ino++) {
264 		if (((ino - 1) % CHAR_BIT) == 0)	/* map is offset by 1 */
265 			isdir = *map++;
266 		else
267 			isdir >>= 1;
268 		/*
269 		 * If a directory has been removed from usedinomap, it
270 		 * either has the nodump flag set, or has inherited
271 		 * it.  Although a directory can't be in dumpinomap if
272 		 * it isn't in usedinomap, we have to go through it to
273 		 * propagate the nodump flag.
274 		 */
275 		nodump = !nonodump && (TSTINO(ino, usedinomap) == 0);
276 		if ((isdir & 1) == 0 || (TSTINO(ino, dumpinomap) && !nodump))
277 			continue;
278 		dp = getino(ino, &i);
279 		/*
280 		 * inode buf may change in searchdir().
281 		 */
282 		if (sblock->fs_magic == FS_UFS1_MAGIC)
283 			di.dp1 = dp->dp1;
284 		else
285 			di.dp2 = dp->dp2;
286 		filesize = DIP(&di, di_size);
287 		for (ret = 0, i = 0; filesize > 0 && i < NDADDR; i++) {
288 			if (DIP(&di, di_db[i]) != 0)
289 				ret |= searchdir(ino, DIP(&di, di_db[i]),
290 				    (long)sblksize(sblock, DIP(&di, di_size),
291 				    i), filesize, tapesize, nodump, maxino);
292 			if (ret & HASDUMPEDFILE)
293 				filesize = 0;
294 			else
295 				filesize -= sblock->fs_bsize;
296 		}
297 		for (i = 0; filesize > 0 && i < NIADDR; i++) {
298 			if (DIP(&di, di_ib[i]) == 0)
299 				continue;
300 			ret |= dirindir(ino, DIP(&di, di_ib[i]), i, &filesize,
301 			    tapesize, nodump, maxino);
302 		}
303 		if (ret & HASDUMPEDFILE) {
304 			SETINO(ino, dumpinomap);
305 			*tapesize += blockest(&di);
306 			change = 1;
307 			continue;
308 		}
309 		if (nodump) {
310 			if (ret & HASSUBDIRS)
311 				change = 1;	/* subdirs inherit nodump */
312 			CLRINO(ino, dumpdirmap);
313 		} else if ((ret & HASSUBDIRS) == 0)
314 			if (!TSTINO(ino, dumpinomap)) {
315 				CLRINO(ino, dumpdirmap);
316 				change = 1;
317 			}
318 	}
319 	return (change);
320 }
321 
322 /*
323  * Read indirect blocks, and pass the data blocks to be searched
324  * as directories. Quit as soon as any entry is found that will
325  * require the directory to be dumped.
326  */
327 static int
328 dirindir(
329 	ino_t ino,
330 	ufs2_daddr_t blkno,
331 	int ind_level,
332 	long *filesize,
333 	long *tapesize,
334 	int nodump,
335 	ino_t maxino)
336 {
337 	union {
338 		ufs1_daddr_t ufs1[MAXBSIZE / sizeof(ufs1_daddr_t)];
339 		ufs2_daddr_t ufs2[MAXBSIZE / sizeof(ufs2_daddr_t)];
340 	} idblk;
341 	int ret = 0;
342 	int i;
343 
344 	bread(fsbtodb(sblock, blkno), (char *)&idblk, (int)sblock->fs_bsize);
345 	if (ind_level <= 0) {
346 		for (i = 0; *filesize > 0 && i < NINDIR(sblock); i++) {
347 			if (sblock->fs_magic == FS_UFS1_MAGIC)
348 				blkno = idblk.ufs1[i];
349 			else
350 				blkno = idblk.ufs2[i];
351 			if (blkno != 0)
352 				ret |= searchdir(ino, blkno, sblock->fs_bsize,
353 					*filesize, tapesize, nodump, maxino);
354 			if (ret & HASDUMPEDFILE)
355 				*filesize = 0;
356 			else
357 				*filesize -= sblock->fs_bsize;
358 		}
359 		return (ret);
360 	}
361 	ind_level--;
362 	for (i = 0; *filesize > 0 && i < NINDIR(sblock); i++) {
363 		if (sblock->fs_magic == FS_UFS1_MAGIC)
364 			blkno = idblk.ufs1[i];
365 		else
366 			blkno = idblk.ufs2[i];
367 		if (blkno != 0)
368 			ret |= dirindir(ino, blkno, ind_level, filesize,
369 			    tapesize, nodump, maxino);
370 	}
371 	return (ret);
372 }
373 
374 /*
375  * Scan a disk block containing directory information looking to see if
376  * any of the entries are on the dump list and to see if the directory
377  * contains any subdirectories.
378  */
379 static int
380 searchdir(
381 	ino_t ino,
382 	ufs2_daddr_t blkno,
383 	long size,
384 	long filesize,
385 	long *tapesize,
386 	int nodump,
387 	ino_t maxino)
388 {
389 	int mode;
390 	struct direct *dp;
391 	union dinode *ip;
392 	long loc, ret = 0;
393 	static caddr_t dblk;
394 
395 	if (dblk == NULL && (dblk = malloc(sblock->fs_bsize)) == NULL)
396 		quit("searchdir: cannot allocate indirect memory.\n");
397 	bread(fsbtodb(sblock, blkno), dblk, (int)size);
398 	if (filesize < size)
399 		size = filesize;
400 	for (loc = 0; loc < size; ) {
401 		dp = (struct direct *)(dblk + loc);
402 		if (dp->d_reclen == 0) {
403 			msg("corrupted directory, inumber %d\n", ino);
404 			break;
405 		}
406 		loc += dp->d_reclen;
407 		if (dp->d_ino == 0)
408 			continue;
409 		if (dp->d_ino >= maxino) {
410 			msg("corrupted directory entry, d_ino %d >= %d\n",
411 			    dp->d_ino, maxino);
412 			break;
413 		}
414 		if (dp->d_name[0] == '.') {
415 			if (dp->d_name[1] == '\0')
416 				continue;
417 			if (dp->d_name[1] == '.' && dp->d_name[2] == '\0')
418 				continue;
419 		}
420 		if (nodump) {
421 			ip = getino(dp->d_ino, &mode);
422 			if (TSTINO(dp->d_ino, dumpinomap)) {
423 				CLRINO(dp->d_ino, dumpinomap);
424 				*tapesize -= blockest(ip);
425 			}
426 			/*
427 			 * Add back to dumpdirmap and remove from usedinomap
428 			 * to propagate nodump.
429 			 */
430 			if (mode == IFDIR) {
431 				SETINO(dp->d_ino, dumpdirmap);
432 				CLRINO(dp->d_ino, usedinomap);
433 				ret |= HASSUBDIRS;
434 			}
435 		} else {
436 			if (TSTINO(dp->d_ino, dumpinomap)) {
437 				ret |= HASDUMPEDFILE;
438 				if (ret & HASSUBDIRS)
439 					break;
440 			}
441 			if (TSTINO(dp->d_ino, dumpdirmap)) {
442 				ret |= HASSUBDIRS;
443 				if (ret & HASDUMPEDFILE)
444 					break;
445 			}
446 		}
447 	}
448 	return (ret);
449 }
450 
451 /*
452  * Dump passes 3 and 4.
453  *
454  * Dump the contents of an inode to tape.
455  */
456 void
457 dumpino(union dinode *dp, ino_t ino)
458 {
459 	int ind_level, cnt, last, added;
460 	off_t size;
461 	char buf[TP_BSIZE];
462 
463 	if (newtape) {
464 		newtape = 0;
465 		dumpmap(dumpinomap, TS_BITS, ino);
466 	}
467 	CLRINO(ino, dumpinomap);
468 	/*
469 	 * Zero out the size of a snapshot so that it will be dumped
470 	 * as a zero length file.
471 	 */
472 	if ((DIP(dp, di_flags) & SF_SNAPSHOT) != 0) {
473 		DIP_SET(dp, di_size, 0);
474 		DIP_SET(dp, di_flags, DIP(dp, di_flags) & ~SF_SNAPSHOT);
475 	}
476 	if (sblock->fs_magic == FS_UFS1_MAGIC) {
477 		spcl.c_mode = dp->dp1.di_mode;
478 		spcl.c_size = dp->dp1.di_size;
479 		spcl.c_extsize = 0;
480 		spcl.c_atime = _time32_to_time(dp->dp1.di_atime);
481 		spcl.c_atimensec = dp->dp1.di_atimensec;
482 		spcl.c_mtime = _time32_to_time(dp->dp1.di_mtime);
483 		spcl.c_mtimensec = dp->dp1.di_mtimensec;
484 		spcl.c_birthtime = 0;
485 		spcl.c_birthtimensec = 0;
486 		spcl.c_rdev = dp->dp1.di_rdev;
487 		spcl.c_file_flags = dp->dp1.di_flags;
488 		spcl.c_uid = dp->dp1.di_uid;
489 		spcl.c_gid = dp->dp1.di_gid;
490 	} else {
491 		spcl.c_mode = dp->dp2.di_mode;
492 		spcl.c_size = dp->dp2.di_size;
493 		spcl.c_extsize = dp->dp2.di_extsize;
494 		spcl.c_atime = _time64_to_time(dp->dp2.di_atime);
495 		spcl.c_atimensec = dp->dp2.di_atimensec;
496 		spcl.c_mtime = _time64_to_time(dp->dp2.di_mtime);
497 		spcl.c_mtimensec = dp->dp2.di_mtimensec;
498 		spcl.c_birthtime = _time64_to_time(dp->dp2.di_birthtime);
499 		spcl.c_birthtimensec = dp->dp2.di_birthnsec;
500 		spcl.c_rdev = dp->dp2.di_rdev;
501 		spcl.c_file_flags = dp->dp2.di_flags;
502 		spcl.c_uid = dp->dp2.di_uid;
503 		spcl.c_gid = dp->dp2.di_gid;
504 	}
505 	spcl.c_type = TS_INODE;
506 	spcl.c_count = 0;
507 	switch (DIP(dp, di_mode) & S_IFMT) {
508 
509 	case 0:
510 		/*
511 		 * Freed inode.
512 		 */
513 		return;
514 
515 	case S_IFLNK:
516 		/*
517 		 * Check for short symbolic link.
518 		 */
519 		if (DIP(dp, di_size) > 0 &&
520 		    DIP(dp, di_size) < sblock->fs_maxsymlinklen) {
521 			spcl.c_addr[0] = 1;
522 			spcl.c_count = 1;
523 			added = appendextdata(dp);
524 			writeheader(ino);
525 			if (sblock->fs_magic == FS_UFS1_MAGIC)
526 				memmove(buf, (caddr_t)dp->dp1.di_db,
527 				    (u_long)DIP(dp, di_size));
528 			else
529 				memmove(buf, (caddr_t)dp->dp2.di_db,
530 				    (u_long)DIP(dp, di_size));
531 			buf[DIP(dp, di_size)] = '\0';
532 			writerec(buf, 0);
533 			writeextdata(dp, ino, added);
534 			return;
535 		}
536 		/* FALLTHROUGH */
537 
538 	case S_IFDIR:
539 	case S_IFREG:
540 		if (DIP(dp, di_size) > 0)
541 			break;
542 		/* FALLTHROUGH */
543 
544 	case S_IFIFO:
545 	case S_IFSOCK:
546 	case S_IFCHR:
547 	case S_IFBLK:
548 		added = appendextdata(dp);
549 		writeheader(ino);
550 		writeextdata(dp, ino, added);
551 		return;
552 
553 	default:
554 		msg("Warning: undefined file type 0%o\n",
555 		    DIP(dp, di_mode) & IFMT);
556 		return;
557 	}
558 	if (DIP(dp, di_size) > NDADDR * sblock->fs_bsize) {
559 		cnt = NDADDR * sblock->fs_frag;
560 		last = 0;
561 	} else {
562 		cnt = howmany(DIP(dp, di_size), sblock->fs_fsize);
563 		last = 1;
564 	}
565 	if (sblock->fs_magic == FS_UFS1_MAGIC)
566 		ufs1_blksout(&dp->dp1.di_db[0], cnt, ino);
567 	else
568 		ufs2_blksout(dp, &dp->dp2.di_db[0], cnt, ino, last);
569 	if ((size = DIP(dp, di_size) - NDADDR * sblock->fs_bsize) <= 0)
570 		return;
571 	for (ind_level = 0; ind_level < NIADDR; ind_level++) {
572 		dmpindir(dp, ino, DIP(dp, di_ib[ind_level]), ind_level, &size);
573 		if (size <= 0)
574 			return;
575 	}
576 }
577 
578 /*
579  * Read indirect blocks, and pass the data blocks to be dumped.
580  */
581 static void
582 dmpindir(union dinode *dp, ino_t ino, ufs2_daddr_t blk, int ind_level,
583 	off_t *size)
584 {
585 	union {
586 		ufs1_daddr_t ufs1[MAXBSIZE / sizeof(ufs1_daddr_t)];
587 		ufs2_daddr_t ufs2[MAXBSIZE / sizeof(ufs2_daddr_t)];
588 	} idblk;
589 	int i, cnt, last;
590 
591 	if (blk != 0)
592 		bread(fsbtodb(sblock, blk), (char *)&idblk,
593 		    (int)sblock->fs_bsize);
594 	else
595 		memset(&idblk, 0, sblock->fs_bsize);
596 	if (ind_level <= 0) {
597 		if (*size > NINDIR(sblock) * sblock->fs_bsize) {
598 			cnt = NINDIR(sblock) * sblock->fs_frag;
599 			last = 0;
600 		} else {
601 			cnt = howmany(*size, sblock->fs_fsize);
602 			last = 1;
603 		}
604 		*size -= NINDIR(sblock) * sblock->fs_bsize;
605 		if (sblock->fs_magic == FS_UFS1_MAGIC)
606 			ufs1_blksout(idblk.ufs1, cnt, ino);
607 		else
608 			ufs2_blksout(dp, idblk.ufs2, cnt, ino, last);
609 		return;
610 	}
611 	ind_level--;
612 	for (i = 0; i < NINDIR(sblock); i++) {
613 		if (sblock->fs_magic == FS_UFS1_MAGIC)
614 			dmpindir(dp, ino, idblk.ufs1[i], ind_level, size);
615 		else
616 			dmpindir(dp, ino, idblk.ufs2[i], ind_level, size);
617 		if (*size <= 0)
618 			return;
619 	}
620 }
621 
622 /*
623  * Collect up the data into tape record sized buffers and output them.
624  */
625 static void
626 ufs1_blksout(ufs1_daddr_t *blkp, int frags, ino_t ino)
627 {
628 	ufs1_daddr_t *bp;
629 	int i, j, count, blks, tbperdb;
630 
631 	blks = howmany(frags * sblock->fs_fsize, TP_BSIZE);
632 	tbperdb = sblock->fs_bsize >> tp_bshift;
633 	for (i = 0; i < blks; i += TP_NINDIR) {
634 		if (i + TP_NINDIR > blks)
635 			count = blks;
636 		else
637 			count = i + TP_NINDIR;
638 		for (j = i; j < count; j++)
639 			if (blkp[j / tbperdb] != 0)
640 				spcl.c_addr[j - i] = 1;
641 			else
642 				spcl.c_addr[j - i] = 0;
643 		spcl.c_count = count - i;
644 		writeheader(ino);
645 		bp = &blkp[i / tbperdb];
646 		for (j = i; j < count; j += tbperdb, bp++)
647 			if (*bp != 0) {
648 				if (j + tbperdb <= count)
649 					dumpblock(*bp, (int)sblock->fs_bsize);
650 				else
651 					dumpblock(*bp, (count - j) * TP_BSIZE);
652 			}
653 		spcl.c_type = TS_ADDR;
654 	}
655 }
656 
657 /*
658  * Collect up the data into tape record sized buffers and output them.
659  */
660 static void
661 ufs2_blksout(union dinode *dp, ufs2_daddr_t *blkp, int frags, ino_t ino,
662 	int last)
663 {
664 	ufs2_daddr_t *bp;
665 	int i, j, count, resid, blks, tbperdb, added;
666 	static int writingextdata = 0;
667 
668 	/*
669 	 * Calculate the number of TP_BSIZE blocks to be dumped.
670 	 * For filesystems with a fragment size bigger than TP_BSIZE,
671 	 * only part of the final fragment may need to be dumped.
672 	 */
673 	blks = howmany(frags * sblock->fs_fsize, TP_BSIZE);
674 	if (last) {
675 		resid = howmany(fragoff(sblock, dp->dp2.di_size), TP_BSIZE);
676 		if (resid > 0)
677 			blks -= howmany(sblock->fs_fsize, TP_BSIZE) - resid;
678 	}
679 	tbperdb = sblock->fs_bsize >> tp_bshift;
680 	for (i = 0; i < blks; i += TP_NINDIR) {
681 		if (i + TP_NINDIR > blks)
682 			count = blks;
683 		else
684 			count = i + TP_NINDIR;
685 		for (j = i; j < count; j++)
686 			if (blkp[j / tbperdb] != 0)
687 				spcl.c_addr[j - i] = 1;
688 			else
689 				spcl.c_addr[j - i] = 0;
690 		spcl.c_count = count - i;
691 		if (last && count == blks && !writingextdata)
692 			added = appendextdata(dp);
693 		writeheader(ino);
694 		bp = &blkp[i / tbperdb];
695 		for (j = i; j < count; j += tbperdb, bp++)
696 			if (*bp != 0) {
697 				if (j + tbperdb <= count)
698 					dumpblock(*bp, (int)sblock->fs_bsize);
699 				else
700 					dumpblock(*bp, (count - j) * TP_BSIZE);
701 			}
702 		spcl.c_type = TS_ADDR;
703 		spcl.c_count = 0;
704 		if (last && count == blks && !writingextdata) {
705 			writingextdata = 1;
706 			writeextdata(dp, ino, added);
707 			writingextdata = 0;
708 		}
709 	}
710 }
711 
712 /*
713  * If there is room in the current block for the extended attributes
714  * as well as the file data, update the header to reflect the added
715  * attribute data at the end. Attributes are placed at the end so that
716  * old versions of restore will correctly restore the file and simply
717  * discard the extra data at the end that it does not understand.
718  * The attribute data is dumped following the file data by the
719  * writeextdata() function (below).
720  */
721 static int
722 appendextdata(union dinode *dp)
723 {
724 	int i, blks, tbperdb;
725 
726 	/*
727 	 * If no extended attributes, there is nothing to do.
728 	 */
729 	if (spcl.c_extsize == 0)
730 		return (0);
731 	/*
732 	 * If there is not enough room at the end of this block
733 	 * to add the extended attributes, then rather than putting
734 	 * part of them here, we simply push them entirely into a
735 	 * new block rather than putting some here and some later.
736 	 */
737 	if (spcl.c_extsize > NXADDR * sblock->fs_bsize)
738 		blks = howmany(NXADDR * sblock->fs_bsize, TP_BSIZE);
739 	else
740 		blks = howmany(spcl.c_extsize, TP_BSIZE);
741 	if (spcl.c_count + blks > TP_NINDIR)
742 		return (0);
743 	/*
744 	 * Update the block map in the header to indicate the added
745 	 * extended attribute. They will be appended after the file
746 	 * data by the writeextdata() routine.
747 	 */
748 	tbperdb = sblock->fs_bsize >> tp_bshift;
749 	for (i = 0; i < blks; i++)
750 		if (&dp->dp2.di_extb[i / tbperdb] != 0)
751 				spcl.c_addr[spcl.c_count + i] = 1;
752 			else
753 				spcl.c_addr[spcl.c_count + i] = 0;
754 	spcl.c_count += blks;
755 	return (blks);
756 }
757 
758 /*
759  * Dump the extended attribute data. If there was room in the file
760  * header, then all we need to do is output the data blocks. If there
761  * was not room in the file header, then an additional TS_ADDR header
762  * is created to hold the attribute data.
763  */
764 static void
765 writeextdata(union dinode *dp, ino_t ino, int added)
766 {
767 	int i, frags, blks, tbperdb, last;
768 	ufs2_daddr_t *bp;
769 	off_t size;
770 
771 	/*
772 	 * If no extended attributes, there is nothing to do.
773 	 */
774 	if (spcl.c_extsize == 0)
775 		return;
776 	/*
777 	 * If there was no room in the file block for the attributes,
778 	 * dump them out in a new block, otherwise just dump the data.
779 	 */
780 	if (added == 0) {
781 		if (spcl.c_extsize > NXADDR * sblock->fs_bsize) {
782 			frags = NXADDR * sblock->fs_frag;
783 			last = 0;
784 		} else {
785 			frags = howmany(spcl.c_extsize, sblock->fs_fsize);
786 			last = 1;
787 		}
788 		ufs2_blksout(dp, &dp->dp2.di_extb[0], frags, ino, last);
789 	} else {
790 		if (spcl.c_extsize > NXADDR * sblock->fs_bsize)
791 			blks = howmany(NXADDR * sblock->fs_bsize, TP_BSIZE);
792 		else
793 			blks = howmany(spcl.c_extsize, TP_BSIZE);
794 		tbperdb = sblock->fs_bsize >> tp_bshift;
795 		for (i = 0; i < blks; i += tbperdb) {
796 			bp = &dp->dp2.di_extb[i / tbperdb];
797 			if (*bp != 0) {
798 				if (i + tbperdb <= blks)
799 					dumpblock(*bp, (int)sblock->fs_bsize);
800 				else
801 					dumpblock(*bp, (blks - i) * TP_BSIZE);
802 			}
803 		}
804 
805 	}
806 	/*
807 	 * If an indirect block is added for extended attributes, then
808 	 * di_exti below should be changed to the structure element
809 	 * that references the extended attribute indirect block. This
810 	 * definition is here only to make it compile without complaint.
811 	 */
812 #define di_exti di_spare[0]
813 	/*
814 	 * If the extended attributes fall into an indirect block,
815 	 * dump it as well.
816 	 */
817 	if ((size = spcl.c_extsize - NXADDR * sblock->fs_bsize) > 0)
818 		dmpindir(dp, ino, dp->dp2.di_exti, 0, &size);
819 }
820 
821 /*
822  * Dump a map to the tape.
823  */
824 void
825 dumpmap(char *map, int type, ino_t ino)
826 {
827 	int i;
828 	char *cp;
829 
830 	spcl.c_type = type;
831 	spcl.c_count = howmany(mapsize * sizeof(char), TP_BSIZE);
832 	writeheader(ino);
833 	for (i = 0, cp = map; i < spcl.c_count; i++, cp += TP_BSIZE)
834 		writerec(cp, 0);
835 }
836 
837 /*
838  * Write a header record to the dump tape.
839  */
840 void
841 writeheader(ino_t ino)
842 {
843 	int32_t sum, cnt, *lp;
844 
845 	if (rsync_friendly >= 2) {
846 		/* don't track changes to access time */
847 		spcl.c_atime = spcl.c_mtime;
848 		spcl.c_atimensec = spcl.c_mtimensec;
849 	}
850 	spcl.c_inumber = ino;
851 	spcl.c_magic = FS_UFS2_MAGIC;
852 	spcl.c_checksum = 0;
853 	lp = (int32_t *)&spcl;
854 	sum = 0;
855 	cnt = sizeof(union u_spcl) / (4 * sizeof(int32_t));
856 	while (--cnt >= 0) {
857 		sum += *lp++;
858 		sum += *lp++;
859 		sum += *lp++;
860 		sum += *lp++;
861 	}
862 	spcl.c_checksum = CHECKSUM - sum;
863 	writerec((char *)&spcl, 1);
864 }
865 
866 union dinode *
867 getino(ino_t inum, int *modep)
868 {
869 	static ino_t minino, maxino;
870 	static caddr_t inoblock;
871 	struct ufs1_dinode *dp1;
872 	struct ufs2_dinode *dp2;
873 
874 	if (inoblock == NULL && (inoblock = malloc(sblock->fs_bsize)) == NULL)
875 		quit("cannot allocate inode memory.\n");
876 	curino = inum;
877 	if (inum >= minino && inum < maxino)
878 		goto gotit;
879 	bread(fsbtodb(sblock, ino_to_fsba(sblock, inum)), inoblock,
880 	    (int)sblock->fs_bsize);
881 	minino = inum - (inum % INOPB(sblock));
882 	maxino = minino + INOPB(sblock);
883 gotit:
884 	if (sblock->fs_magic == FS_UFS1_MAGIC) {
885 		dp1 = &((struct ufs1_dinode *)inoblock)[inum - minino];
886 		*modep = (dp1->di_mode & IFMT);
887 		return ((union dinode *)dp1);
888 	}
889 	dp2 = &((struct ufs2_dinode *)inoblock)[inum - minino];
890 	*modep = (dp2->di_mode & IFMT);
891 	return ((union dinode *)dp2);
892 }
893 
894 /*
895  * Read a chunk of data from the disk.
896  * Try to recover from hard errors by reading in sector sized pieces.
897  * Error recovery is attempted at most BREADEMAX times before seeking
898  * consent from the operator to continue.
899  */
900 int	breaderrors = 0;
901 #define	BREADEMAX 32
902 
903 void
904 bread(ufs2_daddr_t blkno, char *buf, int size)
905 {
906 	int secsize, bytes, resid, xfer, base, cnt, i;
907 	static char *tmpbuf;
908 	off_t offset;
909 
910 loop:
911 	offset = blkno << dev_bshift;
912 	secsize = sblock->fs_fsize;
913 	base = offset % secsize;
914 	resid = size % secsize;
915 	/*
916 	 * If the transfer request starts or ends on a non-sector
917 	 * boundary, we must read the entire sector and copy out
918 	 * just the part that we need.
919 	 */
920 	if (base == 0 && resid == 0) {
921 		cnt = cread(diskfd, buf, size, offset);
922 		if (cnt == size)
923 			return;
924 	} else {
925 		if (tmpbuf == NULL && (tmpbuf = malloc(secsize)) == 0)
926 			quit("buffer malloc failed\n");
927 		xfer = 0;
928 		bytes = size;
929 		if (base != 0) {
930 			cnt = cread(diskfd, tmpbuf, secsize, offset - base);
931 			if (cnt != secsize)
932 				goto bad;
933 			xfer = MIN(secsize - base, size);
934 			offset += xfer;
935 			bytes -= xfer;
936 			resid = bytes % secsize;
937 			memcpy(buf, &tmpbuf[base], xfer);
938 		}
939 		if (bytes >= secsize) {
940 			cnt = cread(diskfd, &buf[xfer], bytes - resid, offset);
941 			if (cnt != bytes - resid)
942 				goto bad;
943 			xfer += cnt;
944 			offset += cnt;
945 		}
946 		if (resid == 0)
947 			return;
948 		cnt = cread(diskfd, tmpbuf, secsize, offset);
949 		if (cnt == secsize) {
950 			memcpy(&buf[xfer], tmpbuf, resid);
951 			return;
952 		}
953 	}
954 bad:
955 	if (blkno + (size / dev_bsize) > fsbtodb(sblock, sblock->fs_size)) {
956 		/*
957 		 * Trying to read the final fragment.
958 		 *
959 		 * NB - dump only works in TP_BSIZE blocks, hence
960 		 * rounds `dev_bsize' fragments up to TP_BSIZE pieces.
961 		 * It should be smarter about not actually trying to
962 		 * read more than it can get, but for the time being
963 		 * we punt and scale back the read only when it gets
964 		 * us into trouble. (mkm 9/25/83)
965 		 */
966 		size -= dev_bsize;
967 		goto loop;
968 	}
969 	if (cnt == -1)
970 		msg("read error from %s: %s: [block %jd]: count=%d\n",
971 			disk, strerror(errno), (intmax_t)blkno, size);
972 	else
973 		msg("short read error from %s: [block %jd]: count=%d, got=%d\n",
974 			disk, (intmax_t)blkno, size, cnt);
975 	if (++breaderrors > BREADEMAX) {
976 		msg("More than %d block read errors from %s\n",
977 			BREADEMAX, disk);
978 		broadcast("DUMP IS AILING!\n");
979 		msg("This is an unrecoverable error.\n");
980 		if (!query("Do you want to attempt to continue?")){
981 			dumpabort(0);
982 			/*NOTREACHED*/
983 		} else
984 			breaderrors = 0;
985 	}
986 	/*
987 	 * Zero buffer, then try to read each sector of buffer separately,
988 	 * and bypass the cache.
989 	 */
990 	memset(buf, 0, size);
991 	for (i = 0; i < size; i += dev_bsize, buf += dev_bsize, blkno++) {
992 		if ((cnt = pread(diskfd, buf, (int)dev_bsize,
993 		    ((off_t)blkno << dev_bshift))) == dev_bsize)
994 			continue;
995 		if (cnt == -1) {
996 			msg("read error from %s: %s: [sector %jd]: count=%ld\n",
997 			    disk, strerror(errno), (intmax_t)blkno, dev_bsize);
998 			continue;
999 		}
1000 		msg("short read from %s: [sector %jd]: count=%ld, got=%d\n",
1001 		    disk, (intmax_t)blkno, dev_bsize, cnt);
1002 	}
1003 }
1004