xref: /freebsd/sbin/dump/traverse.c (revision 5e3190f700637fcfc1a52daeaa4a031fdd2557c7)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1980, 1988, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
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.
18  *
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
29  * SUCH DAMAGE.
30  */
31 
32 #ifndef lint
33 #if 0
34 static char sccsid[] = "@(#)traverse.c	8.7 (Berkeley) 6/15/95";
35 #endif
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 <assert.h>
48 #include <ctype.h>
49 #include <errno.h>
50 #include <inttypes.h>
51 #include <limits.h>
52 #include <stdio.h>
53 #include <stdlib.h>
54 #include <string.h>
55 #include <timeconv.h>
56 #include <unistd.h>
57 
58 #include "dump.h"
59 
60 union dinode {
61 	struct ufs1_dinode dp1;
62 	struct ufs2_dinode dp2;
63 };
64 #define	DIP(dp, field) \
65 	((sblock->fs_magic == FS_UFS1_MAGIC) ? \
66 	(dp)->dp1.field : (dp)->dp2.field)
67 #define DIP_SET(dp, field, val) do {\
68 	if (sblock->fs_magic == FS_UFS1_MAGIC) \
69 		(dp)->dp1.field = (val); \
70 	else \
71 		(dp)->dp2.field = (val); \
72 	} while (0)
73 
74 #define	HASDUMPEDFILE	0x1
75 #define	HASSUBDIRS	0x2
76 
77 static	int dirindir(ino_t ino, ufs2_daddr_t blkno, int level, long *size,
78     long *tapesize, int nodump, ino_t maxino);
79 static	void dmpindir(union dinode *dp, ino_t ino, ufs2_daddr_t blk, int level,
80     off_t *size);
81 static	void ufs1_blksout(ufs1_daddr_t *blkp, int frags, ino_t ino);
82 static	void ufs2_blksout(union dinode *dp, ufs2_daddr_t *blkp, int frags,
83     ino_t ino, int last);
84 static	int appendextdata(union dinode *dp);
85 static	void writeextdata(union dinode *dp, ino_t ino, int added);
86 static	int searchdir(ino_t ino, ufs2_daddr_t blkno, long size, long filesize,
87     long *tapesize, int nodump, ino_t maxino);
88 static	long blockest(union dinode *dp);
89 
90 /*
91  * This is an estimation of the number of TP_BSIZE blocks in the file.
92  * It estimates the number of blocks in files with holes by assuming
93  * that all of the blocks accounted for by di_blocks are data blocks
94  * (when some of the blocks are usually used for indirect pointers);
95  * hence the estimate may be high.
96  */
97 static long
98 blockest(union dinode *dp)
99 {
100 	long blkest, sizeest;
101 
102 	/*
103 	 * dp->di_size is the size of the file in bytes.
104 	 * dp->di_blocks stores the number of sectors actually in the file.
105 	 * If there are more sectors than the size would indicate, this just
106 	 *	means that there are indirect blocks in the file or unused
107 	 *	sectors in the last file block; we can safely ignore these
108 	 *	(blkest = sizeest below).
109 	 * If the file is bigger than the number of sectors would indicate,
110 	 *	then the file has holes in it.	In this case we must use the
111 	 *	block count to estimate the number of data blocks used, but
112 	 *	we use the actual size for estimating the number of indirect
113 	 *	dump blocks (sizeest vs. blkest in the indirect block
114 	 *	calculation).
115 	 */
116 	if ((DIP(dp, di_flags) & SF_SNAPSHOT) != 0)
117 		return (1);
118 	blkest = howmany(dbtob(DIP(dp, di_blocks)), TP_BSIZE);
119 	sizeest = howmany(DIP(dp, di_size), TP_BSIZE);
120 	if (blkest > sizeest)
121 		blkest = sizeest;
122 	if (DIP(dp, di_size) > sblock->fs_bsize * UFS_NDADDR) {
123 		/* calculate the number of indirect blocks on the dump tape */
124 		blkest += howmany(sizeest -
125 		    UFS_NDADDR * sblock->fs_bsize / TP_BSIZE, 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 		blkread(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 < UFS_ROOTINO ||
196 			    (dp = getino(ino, &mode)) == NULL ||
197 			    (mode & IFMT) == 0)
198 				continue;
199 			if (ino >= maxino) {
200 				msg("Skipping inode %ju >= maxino %ju\n",
201 				    (uintmax_t)ino, (uintmax_t)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(UFS_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 < UFS_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 < UFS_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 	blkread(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 	blkread(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 %ju\n",
404 			    (uintmax_t)ino);
405 			break;
406 		}
407 		loc += dp->d_reclen;
408 		if (dp->d_ino == 0)
409 			continue;
410 		if (dp->d_ino >= maxino) {
411 			msg("corrupted directory entry, d_ino %ju >= %ju\n",
412 			    (uintmax_t)dp->d_ino, (uintmax_t)maxino);
413 			break;
414 		}
415 		if (dp->d_name[0] == '.') {
416 			if (dp->d_name[1] == '\0')
417 				continue;
418 			if (dp->d_name[1] == '.' && dp->d_name[2] == '\0')
419 				continue;
420 		}
421 		if (nodump) {
422 			ip = getino(dp->d_ino, &mode);
423 			if (TSTINO(dp->d_ino, dumpinomap)) {
424 				CLRINO(dp->d_ino, dumpinomap);
425 				*tapesize -= blockest(ip);
426 			}
427 			/*
428 			 * Add back to dumpdirmap and remove from usedinomap
429 			 * to propagate nodump.
430 			 */
431 			if (mode == IFDIR) {
432 				SETINO(dp->d_ino, dumpdirmap);
433 				CLRINO(dp->d_ino, usedinomap);
434 				ret |= HASSUBDIRS;
435 			}
436 		} else {
437 			if (TSTINO(dp->d_ino, dumpinomap)) {
438 				ret |= HASDUMPEDFILE;
439 				if (ret & HASSUBDIRS)
440 					break;
441 			}
442 			if (TSTINO(dp->d_ino, dumpdirmap)) {
443 				ret |= HASSUBDIRS;
444 				if (ret & HASDUMPEDFILE)
445 					break;
446 			}
447 		}
448 	}
449 	return (ret);
450 }
451 
452 /*
453  * Dump passes 3 and 4.
454  *
455  * Dump the contents of an inode to tape.
456  */
457 void
458 dumpino(union dinode *dp, ino_t ino)
459 {
460 	int ind_level, cnt, last, added;
461 	off_t size;
462 	char buf[TP_BSIZE];
463 
464 	if (newtape) {
465 		newtape = 0;
466 		dumpmap(dumpinomap, TS_BITS, ino);
467 	}
468 	CLRINO(ino, dumpinomap);
469 	/*
470 	 * Zero out the size of a snapshot so that it will be dumped
471 	 * as a zero length file.
472 	 */
473 	if ((DIP(dp, di_flags) & SF_SNAPSHOT) != 0) {
474 		DIP_SET(dp, di_size, 0);
475 		DIP_SET(dp, di_flags, DIP(dp, di_flags) & ~SF_SNAPSHOT);
476 	}
477 	if (sblock->fs_magic == FS_UFS1_MAGIC) {
478 		spcl.c_mode = dp->dp1.di_mode;
479 		spcl.c_size = dp->dp1.di_size;
480 		spcl.c_extsize = 0;
481 		spcl.c_atime = _time32_to_time(dp->dp1.di_atime);
482 		spcl.c_atimensec = dp->dp1.di_atimensec;
483 		spcl.c_mtime = _time32_to_time(dp->dp1.di_mtime);
484 		spcl.c_mtimensec = dp->dp1.di_mtimensec;
485 		spcl.c_birthtime = 0;
486 		spcl.c_birthtimensec = 0;
487 		spcl.c_rdev = dp->dp1.di_rdev;
488 		spcl.c_file_flags = dp->dp1.di_flags;
489 		spcl.c_uid = dp->dp1.di_uid;
490 		spcl.c_gid = dp->dp1.di_gid;
491 	} else {
492 		spcl.c_mode = dp->dp2.di_mode;
493 		spcl.c_size = dp->dp2.di_size;
494 		spcl.c_extsize = dp->dp2.di_extsize;
495 		spcl.c_atime = _time64_to_time(dp->dp2.di_atime);
496 		spcl.c_atimensec = dp->dp2.di_atimensec;
497 		spcl.c_mtime = _time64_to_time(dp->dp2.di_mtime);
498 		spcl.c_mtimensec = dp->dp2.di_mtimensec;
499 		spcl.c_birthtime = _time64_to_time(dp->dp2.di_birthtime);
500 		spcl.c_birthtimensec = dp->dp2.di_birthnsec;
501 		spcl.c_rdev = dp->dp2.di_rdev;
502 		spcl.c_file_flags = dp->dp2.di_flags;
503 		spcl.c_uid = dp->dp2.di_uid;
504 		spcl.c_gid = dp->dp2.di_gid;
505 	}
506 	spcl.c_type = TS_INODE;
507 	spcl.c_count = 0;
508 	switch (DIP(dp, di_mode) & S_IFMT) {
509 
510 	case 0:
511 		/*
512 		 * Freed inode.
513 		 */
514 		return;
515 
516 	case S_IFLNK:
517 		/*
518 		 * Check for short symbolic link.
519 		 */
520 		if (DIP(dp, di_size) > 0 &&
521 		    DIP(dp, di_size) < sblock->fs_maxsymlinklen) {
522 			spcl.c_addr[0] = 1;
523 			spcl.c_count = 1;
524 			added = appendextdata(dp);
525 			writeheader(ino);
526 			memmove(buf, DIP(dp, di_shortlink),
527 			    (u_long)DIP(dp, di_size));
528 			buf[DIP(dp, di_size)] = '\0';
529 			writerec(buf, 0);
530 			writeextdata(dp, ino, added);
531 			return;
532 		}
533 		/* FALLTHROUGH */
534 
535 	case S_IFDIR:
536 	case S_IFREG:
537 		if (DIP(dp, di_size) > 0)
538 			break;
539 		/* FALLTHROUGH */
540 
541 	case S_IFIFO:
542 	case S_IFSOCK:
543 	case S_IFCHR:
544 	case S_IFBLK:
545 		added = appendextdata(dp);
546 		writeheader(ino);
547 		writeextdata(dp, ino, added);
548 		return;
549 
550 	default:
551 		msg("Warning: undefined file type 0%o\n",
552 		    DIP(dp, di_mode) & IFMT);
553 		return;
554 	}
555 	if (DIP(dp, di_size) > UFS_NDADDR * sblock->fs_bsize) {
556 		cnt = UFS_NDADDR * sblock->fs_frag;
557 		last = 0;
558 	} else {
559 		cnt = howmany(DIP(dp, di_size), sblock->fs_fsize);
560 		last = 1;
561 	}
562 	if (sblock->fs_magic == FS_UFS1_MAGIC)
563 		ufs1_blksout(&dp->dp1.di_db[0], cnt, ino);
564 	else
565 		ufs2_blksout(dp, &dp->dp2.di_db[0], cnt, ino, last);
566 	if ((size = DIP(dp, di_size) - UFS_NDADDR * sblock->fs_bsize) <= 0)
567 		return;
568 	for (ind_level = 0; ind_level < UFS_NIADDR; ind_level++) {
569 		dmpindir(dp, ino, DIP(dp, di_ib[ind_level]), ind_level, &size);
570 		if (size <= 0)
571 			return;
572 	}
573 }
574 
575 /*
576  * Read indirect blocks, and pass the data blocks to be dumped.
577  */
578 static void
579 dmpindir(union dinode *dp, ino_t ino, ufs2_daddr_t blk, int ind_level,
580 	off_t *size)
581 {
582 	union {
583 		ufs1_daddr_t ufs1[MAXBSIZE / sizeof(ufs1_daddr_t)];
584 		ufs2_daddr_t ufs2[MAXBSIZE / sizeof(ufs2_daddr_t)];
585 	} idblk;
586 	int i, cnt, last;
587 
588 	if (blk != 0)
589 		blkread(fsbtodb(sblock, blk), (char *)&idblk,
590 		    (int)sblock->fs_bsize);
591 	else
592 		memset(&idblk, 0, sblock->fs_bsize);
593 	if (ind_level <= 0) {
594 		if (*size > NINDIR(sblock) * sblock->fs_bsize) {
595 			cnt = NINDIR(sblock) * sblock->fs_frag;
596 			last = 0;
597 		} else {
598 			cnt = howmany(*size, sblock->fs_fsize);
599 			last = 1;
600 		}
601 		*size -= NINDIR(sblock) * sblock->fs_bsize;
602 		if (sblock->fs_magic == FS_UFS1_MAGIC)
603 			ufs1_blksout(idblk.ufs1, cnt, ino);
604 		else
605 			ufs2_blksout(dp, idblk.ufs2, cnt, ino, last);
606 		return;
607 	}
608 	ind_level--;
609 	for (i = 0; i < NINDIR(sblock); i++) {
610 		if (sblock->fs_magic == FS_UFS1_MAGIC)
611 			dmpindir(dp, ino, idblk.ufs1[i], ind_level, size);
612 		else
613 			dmpindir(dp, ino, idblk.ufs2[i], ind_level, size);
614 		if (*size <= 0)
615 			return;
616 	}
617 }
618 
619 /*
620  * Collect up the data into tape record sized buffers and output them.
621  */
622 static void
623 ufs1_blksout(ufs1_daddr_t *blkp, int frags, ino_t ino)
624 {
625 	ufs1_daddr_t *bp;
626 	int i, j, count, blks, tbperdb;
627 
628 	blks = howmany(frags * sblock->fs_fsize, TP_BSIZE);
629 	tbperdb = sblock->fs_bsize >> tp_bshift;
630 	for (i = 0; i < blks; i += TP_NINDIR) {
631 		if (i + TP_NINDIR > blks)
632 			count = blks;
633 		else
634 			count = i + TP_NINDIR;
635 		assert(count <= TP_NINDIR + i);
636 		for (j = i; j < count; j++)
637 			if (blkp[j / tbperdb] != 0)
638 				spcl.c_addr[j - i] = 1;
639 			else
640 				spcl.c_addr[j - i] = 0;
641 		spcl.c_count = count - i;
642 		writeheader(ino);
643 		bp = &blkp[i / tbperdb];
644 		for (j = i; j < count; j += tbperdb, bp++)
645 			if (*bp != 0) {
646 				if (j + tbperdb <= count)
647 					dumpblock(*bp, (int)sblock->fs_bsize);
648 				else
649 					dumpblock(*bp, (count - j) * TP_BSIZE);
650 			}
651 		spcl.c_type = TS_ADDR;
652 	}
653 }
654 
655 /*
656  * Collect up the data into tape record sized buffers and output them.
657  */
658 static void
659 ufs2_blksout(union dinode *dp, ufs2_daddr_t *blkp, int frags, ino_t ino,
660 	int last)
661 {
662 	ufs2_daddr_t *bp;
663 	int i, j, count, resid, blks, tbperdb, added;
664 	static int writingextdata = 0;
665 
666 	/*
667 	 * Calculate the number of TP_BSIZE blocks to be dumped.
668 	 * For filesystems with a fragment size bigger than TP_BSIZE,
669 	 * only part of the final fragment may need to be dumped.
670 	 */
671 	blks = howmany(frags * sblock->fs_fsize, TP_BSIZE);
672 	if (last) {
673 		if (writingextdata)
674 			resid = howmany(fragoff(sblock, spcl.c_extsize),
675 			    TP_BSIZE);
676 		else
677 			resid = howmany(fragoff(sblock, dp->dp2.di_size),
678 			    TP_BSIZE);
679 		if (resid > 0)
680 			blks -= howmany(sblock->fs_fsize, TP_BSIZE) - resid;
681 	}
682 	tbperdb = sblock->fs_bsize >> tp_bshift;
683 	for (i = 0; i < blks; i += TP_NINDIR) {
684 		if (i + TP_NINDIR > blks)
685 			count = blks;
686 		else
687 			count = i + TP_NINDIR;
688 		assert(count <= TP_NINDIR + i);
689 		for (j = i; j < count; j++)
690 			if (blkp[j / tbperdb] != 0)
691 				spcl.c_addr[j - i] = 1;
692 			else
693 				spcl.c_addr[j - i] = 0;
694 		spcl.c_count = count - i;
695 		if (last && count == blks && !writingextdata)
696 			added = appendextdata(dp);
697 		writeheader(ino);
698 		bp = &blkp[i / tbperdb];
699 		for (j = i; j < count; j += tbperdb, bp++)
700 			if (*bp != 0) {
701 				if (j + tbperdb <= count)
702 					dumpblock(*bp, (int)sblock->fs_bsize);
703 				else
704 					dumpblock(*bp, (count - j) * TP_BSIZE);
705 			}
706 		spcl.c_type = TS_ADDR;
707 		spcl.c_count = 0;
708 		if (last && count == blks && !writingextdata) {
709 			writingextdata = 1;
710 			writeextdata(dp, ino, added);
711 			writingextdata = 0;
712 		}
713 	}
714 }
715 
716 /*
717  * If there is room in the current block for the extended attributes
718  * as well as the file data, update the header to reflect the added
719  * attribute data at the end. Attributes are placed at the end so that
720  * old versions of restore will correctly restore the file and simply
721  * discard the extra data at the end that it does not understand.
722  * The attribute data is dumped following the file data by the
723  * writeextdata() function (below).
724  */
725 static int
726 appendextdata(union dinode *dp)
727 {
728 	int i, blks, tbperdb;
729 
730 	/*
731 	 * If no extended attributes, there is nothing to do.
732 	 */
733 	if (spcl.c_extsize == 0)
734 		return (0);
735 	/*
736 	 * If there is not enough room at the end of this block
737 	 * to add the extended attributes, then rather than putting
738 	 * part of them here, we simply push them entirely into a
739 	 * new block rather than putting some here and some later.
740 	 */
741 	if (spcl.c_extsize > UFS_NXADDR * sblock->fs_bsize)
742 		blks = howmany(UFS_NXADDR * sblock->fs_bsize, TP_BSIZE);
743 	else
744 		blks = howmany(spcl.c_extsize, TP_BSIZE);
745 	if (spcl.c_count + blks > TP_NINDIR)
746 		return (0);
747 	/*
748 	 * Update the block map in the header to indicate the added
749 	 * extended attribute. They will be appended after the file
750 	 * data by the writeextdata() routine.
751 	 */
752 	tbperdb = sblock->fs_bsize >> tp_bshift;
753 	assert(spcl.c_count + blks <= TP_NINDIR);
754 	for (i = 0; i < blks; i++)
755 		if (&dp->dp2.di_extb[i / tbperdb] != 0)
756 				spcl.c_addr[spcl.c_count + i] = 1;
757 			else
758 				spcl.c_addr[spcl.c_count + i] = 0;
759 	spcl.c_count += blks;
760 	return (blks);
761 }
762 
763 /*
764  * Dump the extended attribute data. If there was room in the file
765  * header, then all we need to do is output the data blocks. If there
766  * was not room in the file header, then an additional TS_ADDR header
767  * is created to hold the attribute data.
768  */
769 static void
770 writeextdata(union dinode *dp, ino_t ino, int added)
771 {
772 	int i, frags, blks, tbperdb, last;
773 	ufs2_daddr_t *bp;
774 	off_t size;
775 
776 	/*
777 	 * If no extended attributes, there is nothing to do.
778 	 */
779 	if (spcl.c_extsize == 0)
780 		return;
781 	/*
782 	 * If there was no room in the file block for the attributes,
783 	 * dump them out in a new block, otherwise just dump the data.
784 	 */
785 	if (added == 0) {
786 		if (spcl.c_extsize > UFS_NXADDR * sblock->fs_bsize) {
787 			frags = UFS_NXADDR * sblock->fs_frag;
788 			last = 0;
789 		} else {
790 			frags = howmany(spcl.c_extsize, sblock->fs_fsize);
791 			last = 1;
792 		}
793 		ufs2_blksout(dp, &dp->dp2.di_extb[0], frags, ino, last);
794 	} else {
795 		if (spcl.c_extsize > UFS_NXADDR * sblock->fs_bsize)
796 			blks = howmany(UFS_NXADDR * sblock->fs_bsize, TP_BSIZE);
797 		else
798 			blks = howmany(spcl.c_extsize, TP_BSIZE);
799 		tbperdb = sblock->fs_bsize >> tp_bshift;
800 		for (i = 0; i < blks; i += tbperdb) {
801 			bp = &dp->dp2.di_extb[i / tbperdb];
802 			if (*bp != 0) {
803 				if (i + tbperdb <= blks)
804 					dumpblock(*bp, (int)sblock->fs_bsize);
805 				else
806 					dumpblock(*bp, (blks - i) * TP_BSIZE);
807 			}
808 		}
809 
810 	}
811 	/*
812 	 * If an indirect block is added for extended attributes, then
813 	 * di_exti below should be changed to the structure element
814 	 * that references the extended attribute indirect block. This
815 	 * definition is here only to make it compile without complaint.
816 	 */
817 #define di_exti di_spare[0]
818 	/*
819 	 * If the extended attributes fall into an indirect block,
820 	 * dump it as well.
821 	 */
822 	if ((size = spcl.c_extsize - UFS_NXADDR * sblock->fs_bsize) > 0)
823 		dmpindir(dp, ino, dp->dp2.di_exti, 0, &size);
824 }
825 
826 /*
827  * Dump a map to the tape.
828  */
829 void
830 dumpmap(char *map, int type, ino_t ino)
831 {
832 	int i;
833 	char *cp;
834 
835 	spcl.c_type = type;
836 	spcl.c_count = howmany(mapsize * sizeof(char), TP_BSIZE);
837 	writeheader(ino);
838 	for (i = 0, cp = map; i < spcl.c_count; i++, cp += TP_BSIZE)
839 		writerec(cp, 0);
840 }
841 
842 /*
843  * Write a header record to the dump tape.
844  */
845 void
846 writeheader(ino_t ino)
847 {
848 	int32_t sum, cnt, *lp;
849 
850 	if (rsync_friendly >= 2) {
851 		/* don't track changes to access time */
852 		spcl.c_atime = spcl.c_mtime;
853 		spcl.c_atimensec = spcl.c_mtimensec;
854 	}
855 	spcl.c_inumber = ino;
856 	spcl.c_magic = FS_UFS2_MAGIC;
857 	spcl.c_checksum = 0;
858 	lp = (int32_t *)&spcl;
859 	sum = 0;
860 	cnt = sizeof(union u_spcl) / (4 * sizeof(int32_t));
861 	while (--cnt >= 0) {
862 		sum += *lp++;
863 		sum += *lp++;
864 		sum += *lp++;
865 		sum += *lp++;
866 	}
867 	spcl.c_checksum = CHECKSUM - sum;
868 	writerec((char *)&spcl, 1);
869 }
870 
871 union dinode *
872 getino(ino_t inum, int *modep)
873 {
874 	static ino_t minino, maxino;
875 	static caddr_t inoblock;
876 	struct ufs1_dinode *dp1;
877 	struct ufs2_dinode *dp2;
878 
879 	if (inoblock == NULL && (inoblock = malloc(sblock->fs_bsize)) == NULL)
880 		quit("cannot allocate inode memory.\n");
881 	curino = inum;
882 	if (inum >= minino && inum < maxino)
883 		goto gotit;
884 	blkread(fsbtodb(sblock, ino_to_fsba(sblock, inum)), inoblock,
885 	    (int)sblock->fs_bsize);
886 	minino = inum - (inum % INOPB(sblock));
887 	maxino = minino + INOPB(sblock);
888 gotit:
889 	if (sblock->fs_magic == FS_UFS1_MAGIC) {
890 		dp1 = &((struct ufs1_dinode *)inoblock)[inum - minino];
891 		*modep = (dp1->di_mode & IFMT);
892 		return ((union dinode *)dp1);
893 	}
894 	dp2 = &((struct ufs2_dinode *)inoblock)[inum - minino];
895 	*modep = (dp2->di_mode & IFMT);
896 	return ((union dinode *)dp2);
897 }
898 
899 /*
900  * Read a chunk of data from the disk.
901  * Try to recover from hard errors by reading in sector sized pieces.
902  * Error recovery is attempted at most BREADEMAX times before seeking
903  * consent from the operator to continue.
904  */
905 int	breaderrors = 0;
906 #define	BREADEMAX 32
907 
908 void
909 blkread(ufs2_daddr_t blkno, char *buf, int size)
910 {
911 	int secsize, bytes, resid, xfer, base, cnt, i;
912 	static char *tmpbuf;
913 	off_t offset;
914 
915 loop:
916 	offset = blkno << dev_bshift;
917 	secsize = sblock->fs_fsize;
918 	base = offset % secsize;
919 	resid = size % secsize;
920 	/*
921 	 * If the transfer request starts or ends on a non-sector
922 	 * boundary, we must read the entire sector and copy out
923 	 * just the part that we need.
924 	 */
925 	if (base == 0 && resid == 0) {
926 		cnt = cread(diskfd, buf, size, offset);
927 		if (cnt == size)
928 			return;
929 	} else {
930 		if (tmpbuf == NULL && (tmpbuf = malloc(secsize)) == NULL)
931 			quit("buffer malloc failed\n");
932 		xfer = 0;
933 		bytes = size;
934 		if (base != 0) {
935 			cnt = cread(diskfd, tmpbuf, secsize, offset - base);
936 			if (cnt != secsize)
937 				goto bad;
938 			xfer = MIN(secsize - base, size);
939 			offset += xfer;
940 			bytes -= xfer;
941 			resid = bytes % secsize;
942 			memcpy(buf, &tmpbuf[base], xfer);
943 		}
944 		if (bytes >= secsize) {
945 			cnt = cread(diskfd, &buf[xfer], bytes - resid, offset);
946 			if (cnt != bytes - resid)
947 				goto bad;
948 			xfer += cnt;
949 			offset += cnt;
950 		}
951 		if (resid == 0)
952 			return;
953 		cnt = cread(diskfd, tmpbuf, secsize, offset);
954 		if (cnt == secsize) {
955 			memcpy(&buf[xfer], tmpbuf, resid);
956 			return;
957 		}
958 	}
959 bad:
960 	if (blkno + (size / dev_bsize) > fsbtodb(sblock, sblock->fs_size)) {
961 		/*
962 		 * Trying to read the final fragment.
963 		 *
964 		 * NB - dump only works in TP_BSIZE blocks, hence
965 		 * rounds `dev_bsize' fragments up to TP_BSIZE pieces.
966 		 * It should be smarter about not actually trying to
967 		 * read more than it can get, but for the time being
968 		 * we punt and scale back the read only when it gets
969 		 * us into trouble. (mkm 9/25/83)
970 		 */
971 		size -= dev_bsize;
972 		goto loop;
973 	}
974 	if (cnt == -1)
975 		msg("read error from %s: %s: [block %jd]: count=%d\n",
976 			disk, strerror(errno), (intmax_t)blkno, size);
977 	else
978 		msg("short read error from %s: [block %jd]: count=%d, got=%d\n",
979 			disk, (intmax_t)blkno, size, cnt);
980 	if (++breaderrors > BREADEMAX) {
981 		msg("More than %d block read errors from %s\n",
982 			BREADEMAX, disk);
983 		broadcast("DUMP IS AILING!\n");
984 		msg("This is an unrecoverable error.\n");
985 		if (!query("Do you want to attempt to continue?")){
986 			dumpabort(0);
987 			/*NOTREACHED*/
988 		} else
989 			breaderrors = 0;
990 	}
991 	/*
992 	 * Zero buffer, then try to read each sector of buffer separately,
993 	 * and bypass the cache.
994 	 */
995 	memset(buf, 0, size);
996 	for (i = 0; i < size; i += dev_bsize, buf += dev_bsize, blkno++) {
997 		if ((cnt = pread(diskfd, buf, (int)dev_bsize,
998 		    ((off_t)blkno << dev_bshift))) == dev_bsize)
999 			continue;
1000 		if (cnt == -1) {
1001 			msg("read error from %s: %s: [sector %jd]: count=%ld\n",
1002 			    disk, strerror(errno), (intmax_t)blkno, dev_bsize);
1003 			continue;
1004 		}
1005 		msg("short read from %s: [sector %jd]: count=%ld, got=%d\n",
1006 		    disk, (intmax_t)blkno, dev_bsize, cnt);
1007 	}
1008 }
1009