xref: /original-bsd/bin/pax/tables.h (revision 3631e65b)
1d684a8f3Smuller /*-
2d684a8f3Smuller  * Copyright (c) 1992 Keith Muller.
3*3631e65bSbostic  * Copyright (c) 1992, 1993
4*3631e65bSbostic  *	The Regents of the University of California.  All rights reserved.
5d684a8f3Smuller  *
6d684a8f3Smuller  * This code is derived from software contributed to Berkeley by
7d684a8f3Smuller  * Keith Muller of the University of California, San Diego.
8d684a8f3Smuller  *
9d684a8f3Smuller  * %sccs.include.redist.c%
10d684a8f3Smuller  *
11*3631e65bSbostic  *	@(#)tables.h	8.1 (Berkeley) 05/31/93
12d684a8f3Smuller  */
13d684a8f3Smuller 
14d684a8f3Smuller /*
15d684a8f3Smuller  * data structures and constants used by the different databases kept by pax
16d684a8f3Smuller  */
17d684a8f3Smuller 
18d684a8f3Smuller /*
19d684a8f3Smuller  * Hash Table Sizes MUST BE PRIME, if set too small performance suffers.
20d684a8f3Smuller  * Probably safe to expect 500000 inodes per tape. Assuming good key
21d684a8f3Smuller  * distribution (inodes) chains of under 50 long (worse case) is ok.
22d684a8f3Smuller  */
23d684a8f3Smuller #define L_TAB_SZ	2503		/* hard link hash table size */
24d684a8f3Smuller #define F_TAB_SZ	50503		/* file time hash table size */
25d684a8f3Smuller #define N_TAB_SZ	541		/* interactive rename hash table */
26d684a8f3Smuller #define D_TAB_SZ	317		/* unique device mapping table */
27d684a8f3Smuller #define A_TAB_SZ	317		/* ftree dir access time reset table */
28d684a8f3Smuller #define MAXKEYLEN	64		/* max number of chars for hash */
29d684a8f3Smuller 
30d684a8f3Smuller /*
31d684a8f3Smuller  * file hard link structure (hashed by dev/ino and chained) used to find the
32d684a8f3Smuller  * hard links in a file system or with some archive formats (cpio)
33d684a8f3Smuller  */
34d684a8f3Smuller typedef struct hrdlnk {
35d684a8f3Smuller 	char		*name;	/* name of first file seen with this ino/dev */
36d684a8f3Smuller 	dev_t		dev;	/* files device number */
37d684a8f3Smuller 	ino_t		ino;	/* files inode number */
38d684a8f3Smuller 	u_long		nlink;	/* expected link count */
39d684a8f3Smuller 	struct hrdlnk	*fow;
40d684a8f3Smuller } HRDLNK;
41d684a8f3Smuller 
42d684a8f3Smuller /*
43d684a8f3Smuller  * Archive write update file time table (the -u, -C flag), hashed by filename.
44d684a8f3Smuller  * Filenames are stored in a scratch file at seek offset into the file. The
45d684a8f3Smuller  * file time (mod time) and the file name length (for a quick check) are
46d684a8f3Smuller  * stored in a hash table node. We were forced to use a scratch file because
47d684a8f3Smuller  * with -u, the mtime for every node in the archive must always be available
48d684a8f3Smuller  * to compare against (and this data can get REALLY large with big archives).
49d684a8f3Smuller  * By being careful to read only when we have a good chance of a match, the
50d684a8f3Smuller  * performance loss is not measurable (and the size of the archive we can
51d684a8f3Smuller  * handle is greatly increased).
52d684a8f3Smuller  */
53d684a8f3Smuller typedef struct ftm {
54d684a8f3Smuller 	int		namelen;	/* file name length */
55d684a8f3Smuller 	time_t		mtime;		/* files last modification time */
56d684a8f3Smuller 	off_t		seek;		/* loacation in scratch file */
57d684a8f3Smuller 	struct ftm	*fow;
58d684a8f3Smuller } FTM;
59d684a8f3Smuller 
60d684a8f3Smuller /*
61d684a8f3Smuller  * Interactive rename table (-i flag), hashed by orig filename.
62d684a8f3Smuller  * We assume this will not be a large table as this mapping data can only be
63d684a8f3Smuller  * obtained through interactive input by the user. Nobody is going to type in
64d684a8f3Smuller  * changes for 500000 files? We use chaining to resolve collisions.
65d684a8f3Smuller  */
66d684a8f3Smuller 
67d684a8f3Smuller typedef struct namt {
68d684a8f3Smuller 	char		*oname;		/* old name */
69d684a8f3Smuller 	char		*nname;		/* new name typed in by the user */
70d684a8f3Smuller 	struct namt	*fow;
71d684a8f3Smuller } NAMT;
72d684a8f3Smuller 
73d684a8f3Smuller /*
74d684a8f3Smuller  * Unique device mapping tables. Some protocols (e.g. cpio) require that the
75d684a8f3Smuller  * <c_dev,c_ino> pair will uniquely identify a file in an archive unless they
76d684a8f3Smuller  * are links to the same file. Appending to archives can break this. For those
77d684a8f3Smuller  * protocols that have this requirement we map c_dev to a unique value not seen
78d684a8f3Smuller  * in the archive when we append. We also try to handle inode truncation with
79d684a8f3Smuller  * this table. (When the inode field in the archive header are too small, we
80d684a8f3Smuller  * remap the dev on writes to remove accidental collisions).
81d684a8f3Smuller  *
82d684a8f3Smuller  * The list is hashed by device number using chain collision resolution. Off of
83d684a8f3Smuller  * each DEVT are linked the various remaps for this device based on those bits
84d684a8f3Smuller  * in the inode which were truncated. For example if we are just remapping to
85d684a8f3Smuller  * avoid a device number during an update append, off the DEVT we would have
86d684a8f3Smuller  * only a single DLIST that has a truncation id of 0 (no inode bits were
87d684a8f3Smuller  * stripped for this device so far). When we spot inode truncation we create
88d684a8f3Smuller  * a new mapping based on the set of bits in the inode which were stripped off.
89d684a8f3Smuller  * so if the top four bits of the inode are stripped and they have a pattern of
90d684a8f3Smuller  * 0110...... (where . are those bits not truncated) we would have a mapping
91d684a8f3Smuller  * assigned for all inodes that has the same 0110.... pattern (with this dev
92d684a8f3Smuller  * number of course). This keeps the mapping sparse and should be able to store
93d684a8f3Smuller  * close to the limit of files which can be represented by the optimal
94d684a8f3Smuller  * combination of dev and inode bits, and without creating a fouled up archive.
95d684a8f3Smuller  * Note we also remap truncated devs in the same way (an exercise for the
96d684a8f3Smuller  * dedicated reader; always wanted to say that...:)
97d684a8f3Smuller  */
98d684a8f3Smuller 
99d684a8f3Smuller typedef struct devt {
100d684a8f3Smuller 	dev_t		dev;	/* the orig device number we now have to map */
101d684a8f3Smuller 	struct devt	*fow;	/* new device map list */
102d684a8f3Smuller 	struct dlist	*list;	/* map list based on inode truncation bits */
103d684a8f3Smuller } DEVT;
104d684a8f3Smuller 
105d684a8f3Smuller typedef struct dlist {
106d684a8f3Smuller 	ino_t trunc_bits;	/* truncation pattern for a specific map */
107d684a8f3Smuller 	dev_t dev;		/* the new device id we use */
108d684a8f3Smuller 	struct dlist *fow;
109d684a8f3Smuller } DLIST;
110d684a8f3Smuller 
111d684a8f3Smuller /*
112d684a8f3Smuller  * ftree directory access time reset table. When we are done with with a
113d684a8f3Smuller  * subtree we reset the access and mod time of the directory when the tflag is
114d684a8f3Smuller  * set. Not really explicitly specified in the pax spec, but easy and fast to
115d684a8f3Smuller  * do (and this may have even been intended in the spec, it is not clear).
116d684a8f3Smuller  * table is hashed by inode with chaining.
117d684a8f3Smuller  */
118d684a8f3Smuller 
119d684a8f3Smuller typedef struct atdir {
120d684a8f3Smuller 	char *name;	/* name of directory to reset */
121d684a8f3Smuller 	dev_t dev;	/* dev and inode for fast lookup */
122d684a8f3Smuller 	ino_t ino;
123d684a8f3Smuller 	time_t mtime;	/* access and mod time to reset to */
124d684a8f3Smuller 	time_t atime;
125d684a8f3Smuller 	struct atdir *fow;
126d684a8f3Smuller } ATDIR;
127d684a8f3Smuller 
128d684a8f3Smuller /*
129d684a8f3Smuller  * created directory time and mode storage entry. After pax is finished during
130d684a8f3Smuller  * extraction or copy, we must reset directory access modes and times that
131d684a8f3Smuller  * may have been modified after creation (they no longer have the specified
132d684a8f3Smuller  * times and/or modes). We must reset time in the reverse order of creation,
133d684a8f3Smuller  * because entries are added  from the top of the file tree to the bottom.
134d684a8f3Smuller  * We MUST reset times from leaf to root (it will not work the other
135d684a8f3Smuller  * direction).  Entries are recorded into a spool file to make reverse
136d684a8f3Smuller  * reading faster.
137d684a8f3Smuller  */
138d684a8f3Smuller 
139d684a8f3Smuller typedef struct dirdata {
140d684a8f3Smuller 	int nlen;	/* length of the directory name (includes \0) */
141d684a8f3Smuller 	off_t npos;	/* position in file where this dir name starts */
142d684a8f3Smuller 	mode_t mode;	/* file mode to restore */
143d684a8f3Smuller 	time_t mtime;	/* mtime to set */
144d684a8f3Smuller 	time_t atime;	/* atime to set */
145d684a8f3Smuller 	int frc_mode;	/* do we force mode settings? */
146d684a8f3Smuller } DIRDATA;
147