traverse.c revision 330897
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 * 4. 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
34static char sccsid[] = "@(#)traverse.c	8.7 (Berkeley) 6/15/95";
35#endif
36static const char rcsid[] =
37  "$FreeBSD: stable/11/sbin/dump/traverse.c 330897 2018-03-14 03:19:51Z eadler $";
38#endif /* not lint */
39
40#include <sys/param.h>
41#include <sys/stat.h>
42
43#include <ufs/ufs/dir.h>
44#include <ufs/ufs/dinode.h>
45#include <ufs/ffs/fs.h>
46
47#include <protocols/dumprestore.h>
48
49#include <ctype.h>
50#include <errno.h>
51#include <inttypes.h>
52#include <limits.h>
53#include <stdio.h>
54#include <stdlib.h>
55#include <string.h>
56#include <timeconv.h>
57#include <unistd.h>
58
59#include "dump.h"
60
61union dinode {
62	struct ufs1_dinode dp1;
63	struct ufs2_dinode dp2;
64};
65#define	DIP(dp, field) \
66	((sblock->fs_magic == FS_UFS1_MAGIC) ? \
67	(dp)->dp1.field : (dp)->dp2.field)
68#define DIP_SET(dp, field, val) do {\
69	if (sblock->fs_magic == FS_UFS1_MAGIC) \
70		(dp)->dp1.field = (val); \
71	else \
72		(dp)->dp2.field = (val); \
73	} while (0)
74
75#define	HASDUMPEDFILE	0x1
76#define	HASSUBDIRS	0x2
77
78static	int dirindir(ino_t ino, ufs2_daddr_t blkno, int level, long *size,
79    long *tapesize, int nodump, ino_t maxino);
80static	void dmpindir(union dinode *dp, ino_t ino, ufs2_daddr_t blk, int level,
81    off_t *size);
82static	void ufs1_blksout(ufs1_daddr_t *blkp, int frags, ino_t ino);
83static	void ufs2_blksout(union dinode *dp, ufs2_daddr_t *blkp, int frags,
84    ino_t ino, int last);
85static	int appendextdata(union dinode *dp);
86static	void writeextdata(union dinode *dp, ino_t ino, int added);
87static	int searchdir(ino_t ino, ufs2_daddr_t blkno, long size, long filesize,
88    long *tapesize, int nodump, ino_t maxino);
89static	long blockest(union dinode *dp);
90
91/*
92 * This is an estimation of the number of TP_BSIZE blocks in the file.
93 * It estimates the number of blocks in files with holes by assuming
94 * that all of the blocks accounted for by di_blocks are data blocks
95 * (when some of the blocks are usually used for indirect pointers);
96 * hence the estimate may be high.
97 */
98static long
99blockest(union dinode *dp)
100{
101	long blkest, sizeest;
102
103	/*
104	 * dp->di_size is the size of the file in bytes.
105	 * dp->di_blocks stores the number of sectors actually in the file.
106	 * If there are more sectors than the size would indicate, this just
107	 *	means that there are indirect blocks in the file or unused
108	 *	sectors in the last file block; we can safely ignore these
109	 *	(blkest = sizeest below).
110	 * If the file is bigger than the number of sectors would indicate,
111	 *	then the file has holes in it.	In this case we must use the
112	 *	block count to estimate the number of data blocks used, but
113	 *	we use the actual size for estimating the number of indirect
114	 *	dump blocks (sizeest vs. blkest in the indirect block
115	 *	calculation).
116	 */
117	if ((DIP(dp, di_flags) & SF_SNAPSHOT) != 0)
118		return (1);
119	blkest = howmany(dbtob(DIP(dp, di_blocks)), TP_BSIZE);
120	sizeest = howmany(DIP(dp, di_size), TP_BSIZE);
121	if (blkest > sizeest)
122		blkest = sizeest;
123	if (DIP(dp, di_size) > sblock->fs_bsize * NDADDR) {
124		/* calculate the number of indirect blocks on the dump tape */
125		blkest +=
126			howmany(sizeest - NDADDR * sblock->fs_bsize / TP_BSIZE,
127			TP_NINDIR);
128	}
129	return (blkest + 1);
130}
131
132/* Auxiliary macro to pick up files changed since previous dump. */
133#define	CHANGEDSINCE(dp, t) \
134	(DIP(dp, di_mtime) >= (t) || DIP(dp, di_ctime) >= (t))
135
136/* The WANTTODUMP macro decides whether a file should be dumped. */
137#ifdef UF_NODUMP
138#define	WANTTODUMP(dp) \
139	(CHANGEDSINCE(dp, spcl.c_ddate) && \
140	 (nonodump || (DIP(dp, di_flags) & UF_NODUMP) != UF_NODUMP))
141#else
142#define	WANTTODUMP(dp) CHANGEDSINCE(dp, spcl.c_ddate)
143#endif
144
145/*
146 * Dump pass 1.
147 *
148 * Walk the inode list for a file system to find all allocated inodes
149 * that have been modified since the previous dump time. Also, find all
150 * the directories in the file system.
151 */
152int
153mapfiles(ino_t maxino, long *tapesize)
154{
155	int i, cg, mode, inosused;
156	int anydirskipped = 0;
157	union dinode *dp;
158	struct cg *cgp;
159	ino_t ino;
160	u_char *cp;
161
162	if ((cgp = malloc(sblock->fs_cgsize)) == NULL)
163		quit("mapfiles: cannot allocate memory.\n");
164	for (cg = 0; cg < sblock->fs_ncg; cg++) {
165		ino = cg * sblock->fs_ipg;
166		bread(fsbtodb(sblock, cgtod(sblock, cg)), (char *)cgp,
167		    sblock->fs_cgsize);
168		if (sblock->fs_magic == FS_UFS2_MAGIC)
169			inosused = cgp->cg_initediblk;
170		else
171			inosused = sblock->fs_ipg;
172		/*
173		 * If we are using soft updates, then we can trust the
174		 * cylinder group inode allocation maps to tell us which
175		 * inodes are allocated. We will scan the used inode map
176		 * to find the inodes that are really in use, and then
177		 * read only those inodes in from disk.
178		 */
179		if (sblock->fs_flags & FS_DOSOFTDEP) {
180			if (!cg_chkmagic(cgp))
181				quit("mapfiles: cg %d: bad magic number\n", cg);
182			cp = &cg_inosused(cgp)[(inosused - 1) / CHAR_BIT];
183			for ( ; inosused > 0; inosused -= CHAR_BIT, cp--) {
184				if (*cp == 0)
185					continue;
186				for (i = 1 << (CHAR_BIT - 1); i > 0; i >>= 1) {
187					if (*cp & i)
188						break;
189					inosused--;
190				}
191				break;
192			}
193			if (inosused <= 0)
194				continue;
195		}
196		for (i = 0; i < inosused; i++, ino++) {
197			if (ino < ROOTINO ||
198			    (dp = getino(ino, &mode)) == NULL ||
199			    (mode & IFMT) == 0)
200				continue;
201			if (ino >= maxino) {
202				msg("Skipping inode %ju >= maxino %ju\n",
203				    (uintmax_t)ino, (uintmax_t)maxino);
204				continue;
205			}
206			/*
207			 * Everything must go in usedinomap so that a check
208			 * for "in dumpdirmap but not in usedinomap" to detect
209			 * dirs with nodump set has a chance of succeeding
210			 * (this is used in mapdirs()).
211			 */
212			SETINO(ino, usedinomap);
213			if (mode == IFDIR)
214				SETINO(ino, dumpdirmap);
215			if (WANTTODUMP(dp)) {
216				SETINO(ino, dumpinomap);
217				if (mode != IFREG &&
218				    mode != IFDIR &&
219				    mode != IFLNK)
220					*tapesize += 1;
221				else
222					*tapesize += blockest(dp);
223				continue;
224			}
225			if (mode == IFDIR) {
226				if (!nonodump &&
227				    (DIP(dp, di_flags) & UF_NODUMP))
228					CLRINO(ino, usedinomap);
229				anydirskipped = 1;
230			}
231		}
232	}
233	/*
234	 * Restore gets very upset if the root is not dumped,
235	 * so ensure that it always is dumped.
236	 */
237	SETINO(ROOTINO, dumpinomap);
238	return (anydirskipped);
239}
240
241/*
242 * Dump pass 2.
243 *
244 * Scan each directory on the file system to see if it has any modified
245 * files in it. If it does, and has not already been added to the dump
246 * list (because it was itself modified), then add it. If a directory
247 * has not been modified itself, contains no modified files and has no
248 * subdirectories, then it can be deleted from the dump list and from
249 * the list of directories. By deleting it from the list of directories,
250 * its parent may now qualify for the same treatment on this or a later
251 * pass using this algorithm.
252 */
253int
254mapdirs(ino_t maxino, long *tapesize)
255{
256	union dinode *dp;
257	int i, isdir, nodump;
258	char *map;
259	ino_t ino;
260	union dinode di;
261	long filesize;
262	int ret, change = 0;
263
264	isdir = 0;		/* XXX just to get gcc to shut up */
265	for (map = dumpdirmap, ino = 1; ino < maxino; ino++) {
266		if (((ino - 1) % CHAR_BIT) == 0)	/* map is offset by 1 */
267			isdir = *map++;
268		else
269			isdir >>= 1;
270		/*
271		 * If a directory has been removed from usedinomap, it
272		 * either has the nodump flag set, or has inherited
273		 * it.  Although a directory can't be in dumpinomap if
274		 * it isn't in usedinomap, we have to go through it to
275		 * propagate the nodump flag.
276		 */
277		nodump = !nonodump && (TSTINO(ino, usedinomap) == 0);
278		if ((isdir & 1) == 0 || (TSTINO(ino, dumpinomap) && !nodump))
279			continue;
280		dp = getino(ino, &i);
281		/*
282		 * inode buf may change in searchdir().
283		 */
284		if (sblock->fs_magic == FS_UFS1_MAGIC)
285			di.dp1 = dp->dp1;
286		else
287			di.dp2 = dp->dp2;
288		filesize = DIP(&di, di_size);
289		for (ret = 0, i = 0; filesize > 0 && i < NDADDR; i++) {
290			if (DIP(&di, di_db[i]) != 0)
291				ret |= searchdir(ino, DIP(&di, di_db[i]),
292				    (long)sblksize(sblock, DIP(&di, di_size),
293				    i), filesize, tapesize, nodump, maxino);
294			if (ret & HASDUMPEDFILE)
295				filesize = 0;
296			else
297				filesize -= sblock->fs_bsize;
298		}
299		for (i = 0; filesize > 0 && i < NIADDR; i++) {
300			if (DIP(&di, di_ib[i]) == 0)
301				continue;
302			ret |= dirindir(ino, DIP(&di, di_ib[i]), i, &filesize,
303			    tapesize, nodump, maxino);
304		}
305		if (ret & HASDUMPEDFILE) {
306			SETINO(ino, dumpinomap);
307			*tapesize += blockest(&di);
308			change = 1;
309			continue;
310		}
311		if (nodump) {
312			if (ret & HASSUBDIRS)
313				change = 1;	/* subdirs inherit nodump */
314			CLRINO(ino, dumpdirmap);
315		} else if ((ret & HASSUBDIRS) == 0)
316			if (!TSTINO(ino, dumpinomap)) {
317				CLRINO(ino, dumpdirmap);
318				change = 1;
319			}
320	}
321	return (change);
322}
323
324/*
325 * Read indirect blocks, and pass the data blocks to be searched
326 * as directories. Quit as soon as any entry is found that will
327 * require the directory to be dumped.
328 */
329static int
330dirindir(
331	ino_t ino,
332	ufs2_daddr_t blkno,
333	int ind_level,
334	long *filesize,
335	long *tapesize,
336	int nodump,
337	ino_t maxino)
338{
339	union {
340		ufs1_daddr_t ufs1[MAXBSIZE / sizeof(ufs1_daddr_t)];
341		ufs2_daddr_t ufs2[MAXBSIZE / sizeof(ufs2_daddr_t)];
342	} idblk;
343	int ret = 0;
344	int i;
345
346	bread(fsbtodb(sblock, blkno), (char *)&idblk, (int)sblock->fs_bsize);
347	if (ind_level <= 0) {
348		for (i = 0; *filesize > 0 && i < NINDIR(sblock); i++) {
349			if (sblock->fs_magic == FS_UFS1_MAGIC)
350				blkno = idblk.ufs1[i];
351			else
352				blkno = idblk.ufs2[i];
353			if (blkno != 0)
354				ret |= searchdir(ino, blkno, sblock->fs_bsize,
355					*filesize, tapesize, nodump, maxino);
356			if (ret & HASDUMPEDFILE)
357				*filesize = 0;
358			else
359				*filesize -= sblock->fs_bsize;
360		}
361		return (ret);
362	}
363	ind_level--;
364	for (i = 0; *filesize > 0 && i < NINDIR(sblock); i++) {
365		if (sblock->fs_magic == FS_UFS1_MAGIC)
366			blkno = idblk.ufs1[i];
367		else
368			blkno = idblk.ufs2[i];
369		if (blkno != 0)
370			ret |= dirindir(ino, blkno, ind_level, filesize,
371			    tapesize, nodump, maxino);
372	}
373	return (ret);
374}
375
376/*
377 * Scan a disk block containing directory information looking to see if
378 * any of the entries are on the dump list and to see if the directory
379 * contains any subdirectories.
380 */
381static int
382searchdir(
383	ino_t ino,
384	ufs2_daddr_t blkno,
385	long size,
386	long filesize,
387	long *tapesize,
388	int nodump,
389	ino_t maxino)
390{
391	int mode;
392	struct direct *dp;
393	union dinode *ip;
394	long loc, ret = 0;
395	static caddr_t dblk;
396
397	if (dblk == NULL && (dblk = malloc(sblock->fs_bsize)) == NULL)
398		quit("searchdir: cannot allocate indirect memory.\n");
399	bread(fsbtodb(sblock, blkno), dblk, (int)size);
400	if (filesize < size)
401		size = filesize;
402	for (loc = 0; loc < size; ) {
403		dp = (struct direct *)(dblk + loc);
404		if (dp->d_reclen == 0) {
405			msg("corrupted directory, inumber %ju\n",
406			    (uintmax_t)ino);
407			break;
408		}
409		loc += dp->d_reclen;
410		if (dp->d_ino == 0)
411			continue;
412		if (dp->d_ino >= maxino) {
413			msg("corrupted directory entry, d_ino %ju >= %ju\n",
414			    (uintmax_t)dp->d_ino, (uintmax_t)maxino);
415			break;
416		}
417		if (dp->d_name[0] == '.') {
418			if (dp->d_name[1] == '\0')
419				continue;
420			if (dp->d_name[1] == '.' && dp->d_name[2] == '\0')
421				continue;
422		}
423		if (nodump) {
424			ip = getino(dp->d_ino, &mode);
425			if (TSTINO(dp->d_ino, dumpinomap)) {
426				CLRINO(dp->d_ino, dumpinomap);
427				*tapesize -= blockest(ip);
428			}
429			/*
430			 * Add back to dumpdirmap and remove from usedinomap
431			 * to propagate nodump.
432			 */
433			if (mode == IFDIR) {
434				SETINO(dp->d_ino, dumpdirmap);
435				CLRINO(dp->d_ino, usedinomap);
436				ret |= HASSUBDIRS;
437			}
438		} else {
439			if (TSTINO(dp->d_ino, dumpinomap)) {
440				ret |= HASDUMPEDFILE;
441				if (ret & HASSUBDIRS)
442					break;
443			}
444			if (TSTINO(dp->d_ino, dumpdirmap)) {
445				ret |= HASSUBDIRS;
446				if (ret & HASDUMPEDFILE)
447					break;
448			}
449		}
450	}
451	return (ret);
452}
453
454/*
455 * Dump passes 3 and 4.
456 *
457 * Dump the contents of an inode to tape.
458 */
459void
460dumpino(union dinode *dp, ino_t ino)
461{
462	int ind_level, cnt, last, added;
463	off_t size;
464	char buf[TP_BSIZE];
465
466	if (newtape) {
467		newtape = 0;
468		dumpmap(dumpinomap, TS_BITS, ino);
469	}
470	CLRINO(ino, dumpinomap);
471	/*
472	 * Zero out the size of a snapshot so that it will be dumped
473	 * as a zero length file.
474	 */
475	if ((DIP(dp, di_flags) & SF_SNAPSHOT) != 0) {
476		DIP_SET(dp, di_size, 0);
477		DIP_SET(dp, di_flags, DIP(dp, di_flags) & ~SF_SNAPSHOT);
478	}
479	if (sblock->fs_magic == FS_UFS1_MAGIC) {
480		spcl.c_mode = dp->dp1.di_mode;
481		spcl.c_size = dp->dp1.di_size;
482		spcl.c_extsize = 0;
483		spcl.c_atime = _time32_to_time(dp->dp1.di_atime);
484		spcl.c_atimensec = dp->dp1.di_atimensec;
485		spcl.c_mtime = _time32_to_time(dp->dp1.di_mtime);
486		spcl.c_mtimensec = dp->dp1.di_mtimensec;
487		spcl.c_birthtime = 0;
488		spcl.c_birthtimensec = 0;
489		spcl.c_rdev = dp->dp1.di_rdev;
490		spcl.c_file_flags = dp->dp1.di_flags;
491		spcl.c_uid = dp->dp1.di_uid;
492		spcl.c_gid = dp->dp1.di_gid;
493	} else {
494		spcl.c_mode = dp->dp2.di_mode;
495		spcl.c_size = dp->dp2.di_size;
496		spcl.c_extsize = dp->dp2.di_extsize;
497		spcl.c_atime = _time64_to_time(dp->dp2.di_atime);
498		spcl.c_atimensec = dp->dp2.di_atimensec;
499		spcl.c_mtime = _time64_to_time(dp->dp2.di_mtime);
500		spcl.c_mtimensec = dp->dp2.di_mtimensec;
501		spcl.c_birthtime = _time64_to_time(dp->dp2.di_birthtime);
502		spcl.c_birthtimensec = dp->dp2.di_birthnsec;
503		spcl.c_rdev = dp->dp2.di_rdev;
504		spcl.c_file_flags = dp->dp2.di_flags;
505		spcl.c_uid = dp->dp2.di_uid;
506		spcl.c_gid = dp->dp2.di_gid;
507	}
508	spcl.c_type = TS_INODE;
509	spcl.c_count = 0;
510	switch (DIP(dp, di_mode) & S_IFMT) {
511
512	case 0:
513		/*
514		 * Freed inode.
515		 */
516		return;
517
518	case S_IFLNK:
519		/*
520		 * Check for short symbolic link.
521		 */
522		if (DIP(dp, di_size) > 0 &&
523		    DIP(dp, di_size) < sblock->fs_maxsymlinklen) {
524			spcl.c_addr[0] = 1;
525			spcl.c_count = 1;
526			added = appendextdata(dp);
527			writeheader(ino);
528			if (sblock->fs_magic == FS_UFS1_MAGIC)
529				memmove(buf, (caddr_t)dp->dp1.di_db,
530				    (u_long)DIP(dp, di_size));
531			else
532				memmove(buf, (caddr_t)dp->dp2.di_db,
533				    (u_long)DIP(dp, di_size));
534			buf[DIP(dp, di_size)] = '\0';
535			writerec(buf, 0);
536			writeextdata(dp, ino, added);
537			return;
538		}
539		/* FALLTHROUGH */
540
541	case S_IFDIR:
542	case S_IFREG:
543		if (DIP(dp, di_size) > 0)
544			break;
545		/* FALLTHROUGH */
546
547	case S_IFIFO:
548	case S_IFSOCK:
549	case S_IFCHR:
550	case S_IFBLK:
551		added = appendextdata(dp);
552		writeheader(ino);
553		writeextdata(dp, ino, added);
554		return;
555
556	default:
557		msg("Warning: undefined file type 0%o\n",
558		    DIP(dp, di_mode) & IFMT);
559		return;
560	}
561	if (DIP(dp, di_size) > NDADDR * sblock->fs_bsize) {
562		cnt = NDADDR * sblock->fs_frag;
563		last = 0;
564	} else {
565		cnt = howmany(DIP(dp, di_size), sblock->fs_fsize);
566		last = 1;
567	}
568	if (sblock->fs_magic == FS_UFS1_MAGIC)
569		ufs1_blksout(&dp->dp1.di_db[0], cnt, ino);
570	else
571		ufs2_blksout(dp, &dp->dp2.di_db[0], cnt, ino, last);
572	if ((size = DIP(dp, di_size) - NDADDR * sblock->fs_bsize) <= 0)
573		return;
574	for (ind_level = 0; ind_level < NIADDR; ind_level++) {
575		dmpindir(dp, ino, DIP(dp, di_ib[ind_level]), ind_level, &size);
576		if (size <= 0)
577			return;
578	}
579}
580
581/*
582 * Read indirect blocks, and pass the data blocks to be dumped.
583 */
584static void
585dmpindir(union dinode *dp, ino_t ino, ufs2_daddr_t blk, int ind_level,
586	off_t *size)
587{
588	union {
589		ufs1_daddr_t ufs1[MAXBSIZE / sizeof(ufs1_daddr_t)];
590		ufs2_daddr_t ufs2[MAXBSIZE / sizeof(ufs2_daddr_t)];
591	} idblk;
592	int i, cnt, last;
593
594	if (blk != 0)
595		bread(fsbtodb(sblock, blk), (char *)&idblk,
596		    (int)sblock->fs_bsize);
597	else
598		memset(&idblk, 0, sblock->fs_bsize);
599	if (ind_level <= 0) {
600		if (*size > NINDIR(sblock) * sblock->fs_bsize) {
601			cnt = NINDIR(sblock) * sblock->fs_frag;
602			last = 0;
603		} else {
604			cnt = howmany(*size, sblock->fs_fsize);
605			last = 1;
606		}
607		*size -= NINDIR(sblock) * sblock->fs_bsize;
608		if (sblock->fs_magic == FS_UFS1_MAGIC)
609			ufs1_blksout(idblk.ufs1, cnt, ino);
610		else
611			ufs2_blksout(dp, idblk.ufs2, cnt, ino, last);
612		return;
613	}
614	ind_level--;
615	for (i = 0; i < NINDIR(sblock); i++) {
616		if (sblock->fs_magic == FS_UFS1_MAGIC)
617			dmpindir(dp, ino, idblk.ufs1[i], ind_level, size);
618		else
619			dmpindir(dp, ino, idblk.ufs2[i], ind_level, size);
620		if (*size <= 0)
621			return;
622	}
623}
624
625/*
626 * Collect up the data into tape record sized buffers and output them.
627 */
628static void
629ufs1_blksout(ufs1_daddr_t *blkp, int frags, ino_t ino)
630{
631	ufs1_daddr_t *bp;
632	int i, j, count, blks, tbperdb;
633
634	blks = howmany(frags * sblock->fs_fsize, TP_BSIZE);
635	tbperdb = sblock->fs_bsize >> tp_bshift;
636	for (i = 0; i < blks; i += TP_NINDIR) {
637		if (i + TP_NINDIR > blks)
638			count = blks;
639		else
640			count = i + TP_NINDIR;
641		for (j = i; j < count; j++)
642			if (blkp[j / tbperdb] != 0)
643				spcl.c_addr[j - i] = 1;
644			else
645				spcl.c_addr[j - i] = 0;
646		spcl.c_count = count - i;
647		writeheader(ino);
648		bp = &blkp[i / tbperdb];
649		for (j = i; j < count; j += tbperdb, bp++)
650			if (*bp != 0) {
651				if (j + tbperdb <= count)
652					dumpblock(*bp, (int)sblock->fs_bsize);
653				else
654					dumpblock(*bp, (count - j) * TP_BSIZE);
655			}
656		spcl.c_type = TS_ADDR;
657	}
658}
659
660/*
661 * Collect up the data into tape record sized buffers and output them.
662 */
663static void
664ufs2_blksout(union dinode *dp, ufs2_daddr_t *blkp, int frags, ino_t ino,
665	int last)
666{
667	ufs2_daddr_t *bp;
668	int i, j, count, resid, blks, tbperdb, added;
669	static int writingextdata = 0;
670
671	/*
672	 * Calculate the number of TP_BSIZE blocks to be dumped.
673	 * For filesystems with a fragment size bigger than TP_BSIZE,
674	 * only part of the final fragment may need to be dumped.
675	 */
676	blks = howmany(frags * sblock->fs_fsize, TP_BSIZE);
677	if (last) {
678		if (writingextdata)
679			resid = howmany(fragoff(sblock, spcl.c_extsize),
680			    TP_BSIZE);
681		else
682			resid = howmany(fragoff(sblock, dp->dp2.di_size),
683			    TP_BSIZE);
684		if (resid > 0)
685			blks -= howmany(sblock->fs_fsize, TP_BSIZE) - resid;
686	}
687	tbperdb = sblock->fs_bsize >> tp_bshift;
688	for (i = 0; i < blks; i += TP_NINDIR) {
689		if (i + TP_NINDIR > blks)
690			count = blks;
691		else
692			count = i + TP_NINDIR;
693		for (j = i; j < count; j++)
694			if (blkp[j / tbperdb] != 0)
695				spcl.c_addr[j - i] = 1;
696			else
697				spcl.c_addr[j - i] = 0;
698		spcl.c_count = count - i;
699		if (last && count == blks && !writingextdata)
700			added = appendextdata(dp);
701		writeheader(ino);
702		bp = &blkp[i / tbperdb];
703		for (j = i; j < count; j += tbperdb, bp++)
704			if (*bp != 0) {
705				if (j + tbperdb <= count)
706					dumpblock(*bp, (int)sblock->fs_bsize);
707				else
708					dumpblock(*bp, (count - j) * TP_BSIZE);
709			}
710		spcl.c_type = TS_ADDR;
711		spcl.c_count = 0;
712		if (last && count == blks && !writingextdata) {
713			writingextdata = 1;
714			writeextdata(dp, ino, added);
715			writingextdata = 0;
716		}
717	}
718}
719
720/*
721 * If there is room in the current block for the extended attributes
722 * as well as the file data, update the header to reflect the added
723 * attribute data at the end. Attributes are placed at the end so that
724 * old versions of restore will correctly restore the file and simply
725 * discard the extra data at the end that it does not understand.
726 * The attribute data is dumped following the file data by the
727 * writeextdata() function (below).
728 */
729static int
730appendextdata(union dinode *dp)
731{
732	int i, blks, tbperdb;
733
734	/*
735	 * If no extended attributes, there is nothing to do.
736	 */
737	if (spcl.c_extsize == 0)
738		return (0);
739	/*
740	 * If there is not enough room at the end of this block
741	 * to add the extended attributes, then rather than putting
742	 * part of them here, we simply push them entirely into a
743	 * new block rather than putting some here and some later.
744	 */
745	if (spcl.c_extsize > NXADDR * sblock->fs_bsize)
746		blks = howmany(NXADDR * sblock->fs_bsize, TP_BSIZE);
747	else
748		blks = howmany(spcl.c_extsize, TP_BSIZE);
749	if (spcl.c_count + blks > TP_NINDIR)
750		return (0);
751	/*
752	 * Update the block map in the header to indicate the added
753	 * extended attribute. They will be appended after the file
754	 * data by the writeextdata() routine.
755	 */
756	tbperdb = sblock->fs_bsize >> tp_bshift;
757	for (i = 0; i < blks; i++)
758		if (&dp->dp2.di_extb[i / tbperdb] != 0)
759				spcl.c_addr[spcl.c_count + i] = 1;
760			else
761				spcl.c_addr[spcl.c_count + i] = 0;
762	spcl.c_count += blks;
763	return (blks);
764}
765
766/*
767 * Dump the extended attribute data. If there was room in the file
768 * header, then all we need to do is output the data blocks. If there
769 * was not room in the file header, then an additional TS_ADDR header
770 * is created to hold the attribute data.
771 */
772static void
773writeextdata(union dinode *dp, ino_t ino, int added)
774{
775	int i, frags, blks, tbperdb, last;
776	ufs2_daddr_t *bp;
777	off_t size;
778
779	/*
780	 * If no extended attributes, there is nothing to do.
781	 */
782	if (spcl.c_extsize == 0)
783		return;
784	/*
785	 * If there was no room in the file block for the attributes,
786	 * dump them out in a new block, otherwise just dump the data.
787	 */
788	if (added == 0) {
789		if (spcl.c_extsize > NXADDR * sblock->fs_bsize) {
790			frags = NXADDR * sblock->fs_frag;
791			last = 0;
792		} else {
793			frags = howmany(spcl.c_extsize, sblock->fs_fsize);
794			last = 1;
795		}
796		ufs2_blksout(dp, &dp->dp2.di_extb[0], frags, ino, last);
797	} else {
798		if (spcl.c_extsize > NXADDR * sblock->fs_bsize)
799			blks = howmany(NXADDR * sblock->fs_bsize, TP_BSIZE);
800		else
801			blks = howmany(spcl.c_extsize, TP_BSIZE);
802		tbperdb = sblock->fs_bsize >> tp_bshift;
803		for (i = 0; i < blks; i += tbperdb) {
804			bp = &dp->dp2.di_extb[i / tbperdb];
805			if (*bp != 0) {
806				if (i + tbperdb <= blks)
807					dumpblock(*bp, (int)sblock->fs_bsize);
808				else
809					dumpblock(*bp, (blks - i) * TP_BSIZE);
810			}
811		}
812
813	}
814	/*
815	 * If an indirect block is added for extended attributes, then
816	 * di_exti below should be changed to the structure element
817	 * that references the extended attribute indirect block. This
818	 * definition is here only to make it compile without complaint.
819	 */
820#define di_exti di_spare[0]
821	/*
822	 * If the extended attributes fall into an indirect block,
823	 * dump it as well.
824	 */
825	if ((size = spcl.c_extsize - NXADDR * sblock->fs_bsize) > 0)
826		dmpindir(dp, ino, dp->dp2.di_exti, 0, &size);
827}
828
829/*
830 * Dump a map to the tape.
831 */
832void
833dumpmap(char *map, int type, ino_t ino)
834{
835	int i;
836	char *cp;
837
838	spcl.c_type = type;
839	spcl.c_count = howmany(mapsize * sizeof(char), TP_BSIZE);
840	writeheader(ino);
841	for (i = 0, cp = map; i < spcl.c_count; i++, cp += TP_BSIZE)
842		writerec(cp, 0);
843}
844
845/*
846 * Write a header record to the dump tape.
847 */
848void
849writeheader(ino_t ino)
850{
851	int32_t sum, cnt, *lp;
852
853	if (rsync_friendly >= 2) {
854		/* don't track changes to access time */
855		spcl.c_atime = spcl.c_mtime;
856		spcl.c_atimensec = spcl.c_mtimensec;
857	}
858	spcl.c_inumber = ino;
859	spcl.c_magic = FS_UFS2_MAGIC;
860	spcl.c_checksum = 0;
861	lp = (int32_t *)&spcl;
862	sum = 0;
863	cnt = sizeof(union u_spcl) / (4 * sizeof(int32_t));
864	while (--cnt >= 0) {
865		sum += *lp++;
866		sum += *lp++;
867		sum += *lp++;
868		sum += *lp++;
869	}
870	spcl.c_checksum = CHECKSUM - sum;
871	writerec((char *)&spcl, 1);
872}
873
874union dinode *
875getino(ino_t inum, int *modep)
876{
877	static ino_t minino, maxino;
878	static caddr_t inoblock;
879	struct ufs1_dinode *dp1;
880	struct ufs2_dinode *dp2;
881
882	if (inoblock == NULL && (inoblock = malloc(sblock->fs_bsize)) == NULL)
883		quit("cannot allocate inode memory.\n");
884	curino = inum;
885	if (inum >= minino && inum < maxino)
886		goto gotit;
887	bread(fsbtodb(sblock, ino_to_fsba(sblock, inum)), inoblock,
888	    (int)sblock->fs_bsize);
889	minino = inum - (inum % INOPB(sblock));
890	maxino = minino + INOPB(sblock);
891gotit:
892	if (sblock->fs_magic == FS_UFS1_MAGIC) {
893		dp1 = &((struct ufs1_dinode *)inoblock)[inum - minino];
894		*modep = (dp1->di_mode & IFMT);
895		return ((union dinode *)dp1);
896	}
897	dp2 = &((struct ufs2_dinode *)inoblock)[inum - minino];
898	*modep = (dp2->di_mode & IFMT);
899	return ((union dinode *)dp2);
900}
901
902/*
903 * Read a chunk of data from the disk.
904 * Try to recover from hard errors by reading in sector sized pieces.
905 * Error recovery is attempted at most BREADEMAX times before seeking
906 * consent from the operator to continue.
907 */
908int	breaderrors = 0;
909#define	BREADEMAX 32
910
911void
912bread(ufs2_daddr_t blkno, char *buf, int size)
913{
914	int secsize, bytes, resid, xfer, base, cnt, i;
915	static char *tmpbuf;
916	off_t offset;
917
918loop:
919	offset = blkno << dev_bshift;
920	secsize = sblock->fs_fsize;
921	base = offset % secsize;
922	resid = size % secsize;
923	/*
924	 * If the transfer request starts or ends on a non-sector
925	 * boundary, we must read the entire sector and copy out
926	 * just the part that we need.
927	 */
928	if (base == 0 && resid == 0) {
929		cnt = cread(diskfd, buf, size, offset);
930		if (cnt == size)
931			return;
932	} else {
933		if (tmpbuf == NULL && (tmpbuf = malloc(secsize)) == NULL)
934			quit("buffer malloc failed\n");
935		xfer = 0;
936		bytes = size;
937		if (base != 0) {
938			cnt = cread(diskfd, tmpbuf, secsize, offset - base);
939			if (cnt != secsize)
940				goto bad;
941			xfer = MIN(secsize - base, size);
942			offset += xfer;
943			bytes -= xfer;
944			resid = bytes % secsize;
945			memcpy(buf, &tmpbuf[base], xfer);
946		}
947		if (bytes >= secsize) {
948			cnt = cread(diskfd, &buf[xfer], bytes - resid, offset);
949			if (cnt != bytes - resid)
950				goto bad;
951			xfer += cnt;
952			offset += cnt;
953		}
954		if (resid == 0)
955			return;
956		cnt = cread(diskfd, tmpbuf, secsize, offset);
957		if (cnt == secsize) {
958			memcpy(&buf[xfer], tmpbuf, resid);
959			return;
960		}
961	}
962bad:
963	if (blkno + (size / dev_bsize) > fsbtodb(sblock, sblock->fs_size)) {
964		/*
965		 * Trying to read the final fragment.
966		 *
967		 * NB - dump only works in TP_BSIZE blocks, hence
968		 * rounds `dev_bsize' fragments up to TP_BSIZE pieces.
969		 * It should be smarter about not actually trying to
970		 * read more than it can get, but for the time being
971		 * we punt and scale back the read only when it gets
972		 * us into trouble. (mkm 9/25/83)
973		 */
974		size -= dev_bsize;
975		goto loop;
976	}
977	if (cnt == -1)
978		msg("read error from %s: %s: [block %jd]: count=%d\n",
979			disk, strerror(errno), (intmax_t)blkno, size);
980	else
981		msg("short read error from %s: [block %jd]: count=%d, got=%d\n",
982			disk, (intmax_t)blkno, size, cnt);
983	if (++breaderrors > BREADEMAX) {
984		msg("More than %d block read errors from %s\n",
985			BREADEMAX, disk);
986		broadcast("DUMP IS AILING!\n");
987		msg("This is an unrecoverable error.\n");
988		if (!query("Do you want to attempt to continue?")){
989			dumpabort(0);
990			/*NOTREACHED*/
991		} else
992			breaderrors = 0;
993	}
994	/*
995	 * Zero buffer, then try to read each sector of buffer separately,
996	 * and bypass the cache.
997	 */
998	memset(buf, 0, size);
999	for (i = 0; i < size; i += dev_bsize, buf += dev_bsize, blkno++) {
1000		if ((cnt = pread(diskfd, buf, (int)dev_bsize,
1001		    ((off_t)blkno << dev_bshift))) == dev_bsize)
1002			continue;
1003		if (cnt == -1) {
1004			msg("read error from %s: %s: [sector %jd]: count=%ld\n",
1005			    disk, strerror(errno), (intmax_t)blkno, dev_bsize);
1006			continue;
1007		}
1008		msg("short read from %s: [sector %jd]: count=%ld, got=%d\n",
1009		    disk, (intmax_t)blkno, dev_bsize, cnt);
1010	}
1011}
1012