xref: /original-bsd/bin/pax/tables.h (revision d684a8f3)
1*d684a8f3Smuller /*-
2*d684a8f3Smuller  * Copyright (c) 1992 Keith Muller.
3*d684a8f3Smuller  * Copyright (c) 1992 The Regents of the University of California.
4*d684a8f3Smuller  * All rights reserved.
5*d684a8f3Smuller  *
6*d684a8f3Smuller  * This code is derived from software contributed to Berkeley by
7*d684a8f3Smuller  * Keith Muller of the University of California, San Diego.
8*d684a8f3Smuller  *
9*d684a8f3Smuller  * %sccs.include.redist.c%
10*d684a8f3Smuller  *
11*d684a8f3Smuller  *	@(#)tables.h	1.1 (Berkeley) 12/13/92
12*d684a8f3Smuller  */
13*d684a8f3Smuller 
14*d684a8f3Smuller /*
15*d684a8f3Smuller  * data structures and constants used by the different databases kept by pax
16*d684a8f3Smuller  */
17*d684a8f3Smuller 
18*d684a8f3Smuller /*
19*d684a8f3Smuller  * Hash Table Sizes MUST BE PRIME, if set too small performance suffers.
20*d684a8f3Smuller  * Probably safe to expect 500000 inodes per tape. Assuming good key
21*d684a8f3Smuller  * distribution (inodes) chains of under 50 long (worse case) is ok.
22*d684a8f3Smuller  */
23*d684a8f3Smuller #define L_TAB_SZ	2503		/* hard link hash table size */
24*d684a8f3Smuller #define F_TAB_SZ	50503		/* file time hash table size */
25*d684a8f3Smuller #define N_TAB_SZ	541		/* interactive rename hash table */
26*d684a8f3Smuller #define D_TAB_SZ	317		/* unique device mapping table */
27*d684a8f3Smuller #define A_TAB_SZ	317		/* ftree dir access time reset table */
28*d684a8f3Smuller #define MAXKEYLEN	64		/* max number of chars for hash */
29*d684a8f3Smuller 
30*d684a8f3Smuller /*
31*d684a8f3Smuller  * file hard link structure (hashed by dev/ino and chained) used to find the
32*d684a8f3Smuller  * hard links in a file system or with some archive formats (cpio)
33*d684a8f3Smuller  */
34*d684a8f3Smuller typedef struct hrdlnk {
35*d684a8f3Smuller 	char		*name;	/* name of first file seen with this ino/dev */
36*d684a8f3Smuller 	dev_t		dev;	/* files device number */
37*d684a8f3Smuller 	ino_t		ino;	/* files inode number */
38*d684a8f3Smuller 	u_long		nlink;	/* expected link count */
39*d684a8f3Smuller 	struct hrdlnk	*fow;
40*d684a8f3Smuller } HRDLNK;
41*d684a8f3Smuller 
42*d684a8f3Smuller /*
43*d684a8f3Smuller  * Archive write update file time table (the -u, -C flag), hashed by filename.
44*d684a8f3Smuller  * Filenames are stored in a scratch file at seek offset into the file. The
45*d684a8f3Smuller  * file time (mod time) and the file name length (for a quick check) are
46*d684a8f3Smuller  * stored in a hash table node. We were forced to use a scratch file because
47*d684a8f3Smuller  * with -u, the mtime for every node in the archive must always be available
48*d684a8f3Smuller  * to compare against (and this data can get REALLY large with big archives).
49*d684a8f3Smuller  * By being careful to read only when we have a good chance of a match, the
50*d684a8f3Smuller  * performance loss is not measurable (and the size of the archive we can
51*d684a8f3Smuller  * handle is greatly increased).
52*d684a8f3Smuller  */
53*d684a8f3Smuller typedef struct ftm {
54*d684a8f3Smuller 	int		namelen;	/* file name length */
55*d684a8f3Smuller 	time_t		mtime;		/* files last modification time */
56*d684a8f3Smuller 	off_t		seek;		/* loacation in scratch file */
57*d684a8f3Smuller 	struct ftm	*fow;
58*d684a8f3Smuller } FTM;
59*d684a8f3Smuller 
60*d684a8f3Smuller /*
61*d684a8f3Smuller  * Interactive rename table (-i flag), hashed by orig filename.
62*d684a8f3Smuller  * We assume this will not be a large table as this mapping data can only be
63*d684a8f3Smuller  * obtained through interactive input by the user. Nobody is going to type in
64*d684a8f3Smuller  * changes for 500000 files? We use chaining to resolve collisions.
65*d684a8f3Smuller  */
66*d684a8f3Smuller 
67*d684a8f3Smuller typedef struct namt {
68*d684a8f3Smuller 	char		*oname;		/* old name */
69*d684a8f3Smuller 	char		*nname;		/* new name typed in by the user */
70*d684a8f3Smuller 	struct namt	*fow;
71*d684a8f3Smuller } NAMT;
72*d684a8f3Smuller 
73*d684a8f3Smuller /*
74*d684a8f3Smuller  * Unique device mapping tables. Some protocols (e.g. cpio) require that the
75*d684a8f3Smuller  * <c_dev,c_ino> pair will uniquely identify a file in an archive unless they
76*d684a8f3Smuller  * are links to the same file. Appending to archives can break this. For those
77*d684a8f3Smuller  * protocols that have this requirement we map c_dev to a unique value not seen
78*d684a8f3Smuller  * in the archive when we append. We also try to handle inode truncation with
79*d684a8f3Smuller  * this table. (When the inode field in the archive header are too small, we
80*d684a8f3Smuller  * remap the dev on writes to remove accidental collisions).
81*d684a8f3Smuller  *
82*d684a8f3Smuller  * The list is hashed by device number using chain collision resolution. Off of
83*d684a8f3Smuller  * each DEVT are linked the various remaps for this device based on those bits
84*d684a8f3Smuller  * in the inode which were truncated. For example if we are just remapping to
85*d684a8f3Smuller  * avoid a device number during an update append, off the DEVT we would have
86*d684a8f3Smuller  * only a single DLIST that has a truncation id of 0 (no inode bits were
87*d684a8f3Smuller  * stripped for this device so far). When we spot inode truncation we create
88*d684a8f3Smuller  * a new mapping based on the set of bits in the inode which were stripped off.
89*d684a8f3Smuller  * so if the top four bits of the inode are stripped and they have a pattern of
90*d684a8f3Smuller  * 0110...... (where . are those bits not truncated) we would have a mapping
91*d684a8f3Smuller  * assigned for all inodes that has the same 0110.... pattern (with this dev
92*d684a8f3Smuller  * number of course). This keeps the mapping sparse and should be able to store
93*d684a8f3Smuller  * close to the limit of files which can be represented by the optimal
94*d684a8f3Smuller  * combination of dev and inode bits, and without creating a fouled up archive.
95*d684a8f3Smuller  * Note we also remap truncated devs in the same way (an exercise for the
96*d684a8f3Smuller  * dedicated reader; always wanted to say that...:)
97*d684a8f3Smuller  */
98*d684a8f3Smuller 
99*d684a8f3Smuller typedef struct devt {
100*d684a8f3Smuller 	dev_t		dev;	/* the orig device number we now have to map */
101*d684a8f3Smuller 	struct devt	*fow;	/* new device map list */
102*d684a8f3Smuller 	struct dlist	*list;	/* map list based on inode truncation bits */
103*d684a8f3Smuller } DEVT;
104*d684a8f3Smuller 
105*d684a8f3Smuller typedef struct dlist {
106*d684a8f3Smuller 	ino_t trunc_bits;	/* truncation pattern for a specific map */
107*d684a8f3Smuller 	dev_t dev;		/* the new device id we use */
108*d684a8f3Smuller 	struct dlist *fow;
109*d684a8f3Smuller } DLIST;
110*d684a8f3Smuller 
111*d684a8f3Smuller /*
112*d684a8f3Smuller  * ftree directory access time reset table. When we are done with with a
113*d684a8f3Smuller  * subtree we reset the access and mod time of the directory when the tflag is
114*d684a8f3Smuller  * set. Not really explicitly specified in the pax spec, but easy and fast to
115*d684a8f3Smuller  * do (and this may have even been intended in the spec, it is not clear).
116*d684a8f3Smuller  * table is hashed by inode with chaining.
117*d684a8f3Smuller  */
118*d684a8f3Smuller 
119*d684a8f3Smuller typedef struct atdir {
120*d684a8f3Smuller 	char *name;	/* name of directory to reset */
121*d684a8f3Smuller 	dev_t dev;	/* dev and inode for fast lookup */
122*d684a8f3Smuller 	ino_t ino;
123*d684a8f3Smuller 	time_t mtime;	/* access and mod time to reset to */
124*d684a8f3Smuller 	time_t atime;
125*d684a8f3Smuller 	struct atdir *fow;
126*d684a8f3Smuller } ATDIR;
127*d684a8f3Smuller 
128*d684a8f3Smuller /*
129*d684a8f3Smuller  * created directory time and mode storage entry. After pax is finished during
130*d684a8f3Smuller  * extraction or copy, we must reset directory access modes and times that
131*d684a8f3Smuller  * may have been modified after creation (they no longer have the specified
132*d684a8f3Smuller  * times and/or modes). We must reset time in the reverse order of creation,
133*d684a8f3Smuller  * because entries are added  from the top of the file tree to the bottom.
134*d684a8f3Smuller  * We MUST reset times from leaf to root (it will not work the other
135*d684a8f3Smuller  * direction).  Entries are recorded into a spool file to make reverse
136*d684a8f3Smuller  * reading faster.
137*d684a8f3Smuller  */
138*d684a8f3Smuller 
139*d684a8f3Smuller typedef struct dirdata {
140*d684a8f3Smuller 	int nlen;	/* length of the directory name (includes \0) */
141*d684a8f3Smuller 	off_t npos;	/* position in file where this dir name starts */
142*d684a8f3Smuller 	mode_t mode;	/* file mode to restore */
143*d684a8f3Smuller 	time_t mtime;	/* mtime to set */
144*d684a8f3Smuller 	time_t atime;	/* atime to set */
145*d684a8f3Smuller 	int frc_mode;	/* do we force mode settings? */
146*d684a8f3Smuller } DIRDATA;
147