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