xref: /freebsd/bin/pax/tables.h (revision cbd30a72ca196976c1c700400ecd424baa1b9c16)
1 /*-
2  * Copyright (c) 1992 Keith Muller.
3  * Copyright (c) 1992, 1993
4  *	The Regents of the University of California.  All rights reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * Keith Muller of the University of California, San Diego.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	@(#)tables.h	8.1 (Berkeley) 5/31/93
34  * $FreeBSD$
35  */
36 
37 /*
38  * data structures and constants used by the different databases kept by pax
39  */
40 
41 /*
42  * Hash Table Sizes MUST BE PRIME, if set too small performance suffers.
43  * Probably safe to expect 500000 inodes per tape. Assuming good key
44  * distribution (inodes) chains of under 50 long (worse case) is ok.
45  */
46 #define L_TAB_SZ	2503		/* hard link hash table size */
47 #define F_TAB_SZ	50503		/* file time hash table size */
48 #define N_TAB_SZ	541		/* interactive rename hash table */
49 #define D_TAB_SZ	317		/* unique device mapping table */
50 #define A_TAB_SZ	317		/* ftree dir access time reset table */
51 #define MAXKEYLEN	64		/* max number of chars for hash */
52 
53 /*
54  * file hard link structure (hashed by dev/ino and chained) used to find the
55  * hard links in a file system or with some archive formats (cpio)
56  */
57 typedef struct hrdlnk {
58 	char		*name;	/* name of first file seen with this ino/dev */
59 	dev_t		dev;	/* files device number */
60 	ino_t		ino;	/* files inode number */
61 	u_long		nlink;	/* expected link count */
62 	struct hrdlnk	*fow;
63 } HRDLNK;
64 
65 /*
66  * Archive write update file time table (the -u, -C flag), hashed by filename.
67  * Filenames are stored in a scratch file at seek offset into the file. The
68  * file time (mod time) and the file name length (for a quick check) are
69  * stored in a hash table node. We were forced to use a scratch file because
70  * with -u, the mtime for every node in the archive must always be available
71  * to compare against (and this data can get REALLY large with big archives).
72  * By being careful to read only when we have a good chance of a match, the
73  * performance loss is not measurable (and the size of the archive we can
74  * handle is greatly increased).
75  */
76 typedef struct ftm {
77 	int		namelen;	/* file name length */
78 	time_t		mtime;		/* files last modification time */
79 	off_t		seek;		/* location in scratch file */
80 	struct ftm	*fow;
81 } FTM;
82 
83 /*
84  * Interactive rename table (-i flag), hashed by orig filename.
85  * We assume this will not be a large table as this mapping data can only be
86  * obtained through interactive input by the user. Nobody is going to type in
87  * changes for 500000 files? We use chaining to resolve collisions.
88  */
89 
90 typedef struct namt {
91 	char		*oname;		/* old name */
92 	char		*nname;		/* new name typed in by the user */
93 	struct namt	*fow;
94 } NAMT;
95 
96 /*
97  * Unique device mapping tables. Some protocols (e.g. cpio) require that the
98  * <c_dev,c_ino> pair will uniquely identify a file in an archive unless they
99  * are links to the same file. Appending to archives can break this. For those
100  * protocols that have this requirement we map c_dev to a unique value not seen
101  * in the archive when we append. We also try to handle inode truncation with
102  * this table. (When the inode field in the archive header are too small, we
103  * remap the dev on writes to remove accidental collisions).
104  *
105  * The list is hashed by device number using chain collision resolution. Off of
106  * each DEVT are linked the various remaps for this device based on those bits
107  * in the inode which were truncated. For example if we are just remapping to
108  * avoid a device number during an update append, off the DEVT we would have
109  * only a single DLIST that has a truncation id of 0 (no inode bits were
110  * stripped for this device so far). When we spot inode truncation we create
111  * a new mapping based on the set of bits in the inode which were stripped off.
112  * so if the top four bits of the inode are stripped and they have a pattern of
113  * 0110...... (where . are those bits not truncated) we would have a mapping
114  * assigned for all inodes that has the same 0110.... pattern (with this dev
115  * number of course). This keeps the mapping sparse and should be able to store
116  * close to the limit of files which can be represented by the optimal
117  * combination of dev and inode bits, and without creating a fouled up archive.
118  * Note we also remap truncated devs in the same way (an exercise for the
119  * dedicated reader; always wanted to say that...:)
120  */
121 
122 typedef struct devt {
123 	dev_t		dev;	/* the orig device number we now have to map */
124 	struct devt	*fow;	/* new device map list */
125 	struct dlist	*list;	/* map list based on inode truncation bits */
126 } DEVT;
127 
128 typedef struct dlist {
129 	ino_t trunc_bits;	/* truncation pattern for a specific map */
130 	dev_t dev;		/* the new device id we use */
131 	struct dlist *fow;
132 } DLIST;
133 
134 /*
135  * ftree directory access time reset table. When we are done with with a
136  * subtree we reset the access and mod time of the directory when the tflag is
137  * set. Not really explicitly specified in the pax spec, but easy and fast to
138  * do (and this may have even been intended in the spec, it is not clear).
139  * table is hashed by inode with chaining.
140  */
141 
142 typedef struct atdir {
143 	char *name;	/* name of directory to reset */
144 	dev_t dev;	/* dev and inode for fast lookup */
145 	ino_t ino;
146 	time_t mtime;	/* access and mod time to reset to */
147 	time_t atime;
148 	struct atdir *fow;
149 } ATDIR;
150 
151 /*
152  * created directory time and mode storage entry. After pax is finished during
153  * extraction or copy, we must reset directory access modes and times that
154  * may have been modified after creation (they no longer have the specified
155  * times and/or modes). We must reset time in the reverse order of creation,
156  * because entries are added  from the top of the file tree to the bottom.
157  * We MUST reset times from leaf to root (it will not work the other
158  * direction).  Entries are recorded into a spool file to make reverse
159  * reading faster.
160  */
161 
162 typedef struct dirdata {
163 	int nlen;	/* length of the directory name (includes \0) */
164 	off_t npos;	/* position in file where this dir name starts */
165 	mode_t mode;	/* file mode to restore */
166 	time_t mtime;	/* mtime to set */
167 	time_t atime;	/* atime to set */
168 	int frc_mode;	/* do we force mode settings? */
169 } DIRDATA;
170