1*57121Smuller /*- 2*57121Smuller * Copyright (c) 1992 Keith Muller. 3*57121Smuller * Copyright (c) 1992 The Regents of the University of California. 4*57121Smuller * All rights reserved. 5*57121Smuller * 6*57121Smuller * This code is derived from software contributed to Berkeley by 7*57121Smuller * Keith Muller of the University of California, San Diego. 8*57121Smuller * 9*57121Smuller * %sccs.include.redist.c% 10*57121Smuller * 11*57121Smuller * @(#)tables.h 1.1 (Berkeley) 12/13/92 12*57121Smuller */ 13*57121Smuller 14*57121Smuller /* 15*57121Smuller * data structures and constants used by the different databases kept by pax 16*57121Smuller */ 17*57121Smuller 18*57121Smuller /* 19*57121Smuller * Hash Table Sizes MUST BE PRIME, if set too small performance suffers. 20*57121Smuller * Probably safe to expect 500000 inodes per tape. Assuming good key 21*57121Smuller * distribution (inodes) chains of under 50 long (worse case) is ok. 22*57121Smuller */ 23*57121Smuller #define L_TAB_SZ 2503 /* hard link hash table size */ 24*57121Smuller #define F_TAB_SZ 50503 /* file time hash table size */ 25*57121Smuller #define N_TAB_SZ 541 /* interactive rename hash table */ 26*57121Smuller #define D_TAB_SZ 317 /* unique device mapping table */ 27*57121Smuller #define A_TAB_SZ 317 /* ftree dir access time reset table */ 28*57121Smuller #define MAXKEYLEN 64 /* max number of chars for hash */ 29*57121Smuller 30*57121Smuller /* 31*57121Smuller * file hard link structure (hashed by dev/ino and chained) used to find the 32*57121Smuller * hard links in a file system or with some archive formats (cpio) 33*57121Smuller */ 34*57121Smuller typedef struct hrdlnk { 35*57121Smuller char *name; /* name of first file seen with this ino/dev */ 36*57121Smuller dev_t dev; /* files device number */ 37*57121Smuller ino_t ino; /* files inode number */ 38*57121Smuller u_long nlink; /* expected link count */ 39*57121Smuller struct hrdlnk *fow; 40*57121Smuller } HRDLNK; 41*57121Smuller 42*57121Smuller /* 43*57121Smuller * Archive write update file time table (the -u, -C flag), hashed by filename. 44*57121Smuller * Filenames are stored in a scratch file at seek offset into the file. The 45*57121Smuller * file time (mod time) and the file name length (for a quick check) are 46*57121Smuller * stored in a hash table node. We were forced to use a scratch file because 47*57121Smuller * with -u, the mtime for every node in the archive must always be available 48*57121Smuller * to compare against (and this data can get REALLY large with big archives). 49*57121Smuller * By being careful to read only when we have a good chance of a match, the 50*57121Smuller * performance loss is not measurable (and the size of the archive we can 51*57121Smuller * handle is greatly increased). 52*57121Smuller */ 53*57121Smuller typedef struct ftm { 54*57121Smuller int namelen; /* file name length */ 55*57121Smuller time_t mtime; /* files last modification time */ 56*57121Smuller off_t seek; /* loacation in scratch file */ 57*57121Smuller struct ftm *fow; 58*57121Smuller } FTM; 59*57121Smuller 60*57121Smuller /* 61*57121Smuller * Interactive rename table (-i flag), hashed by orig filename. 62*57121Smuller * We assume this will not be a large table as this mapping data can only be 63*57121Smuller * obtained through interactive input by the user. Nobody is going to type in 64*57121Smuller * changes for 500000 files? We use chaining to resolve collisions. 65*57121Smuller */ 66*57121Smuller 67*57121Smuller typedef struct namt { 68*57121Smuller char *oname; /* old name */ 69*57121Smuller char *nname; /* new name typed in by the user */ 70*57121Smuller struct namt *fow; 71*57121Smuller } NAMT; 72*57121Smuller 73*57121Smuller /* 74*57121Smuller * Unique device mapping tables. Some protocols (e.g. cpio) require that the 75*57121Smuller * <c_dev,c_ino> pair will uniquely identify a file in an archive unless they 76*57121Smuller * are links to the same file. Appending to archives can break this. For those 77*57121Smuller * protocols that have this requirement we map c_dev to a unique value not seen 78*57121Smuller * in the archive when we append. We also try to handle inode truncation with 79*57121Smuller * this table. (When the inode field in the archive header are too small, we 80*57121Smuller * remap the dev on writes to remove accidental collisions). 81*57121Smuller * 82*57121Smuller * The list is hashed by device number using chain collision resolution. Off of 83*57121Smuller * each DEVT are linked the various remaps for this device based on those bits 84*57121Smuller * in the inode which were truncated. For example if we are just remapping to 85*57121Smuller * avoid a device number during an update append, off the DEVT we would have 86*57121Smuller * only a single DLIST that has a truncation id of 0 (no inode bits were 87*57121Smuller * stripped for this device so far). When we spot inode truncation we create 88*57121Smuller * a new mapping based on the set of bits in the inode which were stripped off. 89*57121Smuller * so if the top four bits of the inode are stripped and they have a pattern of 90*57121Smuller * 0110...... (where . are those bits not truncated) we would have a mapping 91*57121Smuller * assigned for all inodes that has the same 0110.... pattern (with this dev 92*57121Smuller * number of course). This keeps the mapping sparse and should be able to store 93*57121Smuller * close to the limit of files which can be represented by the optimal 94*57121Smuller * combination of dev and inode bits, and without creating a fouled up archive. 95*57121Smuller * Note we also remap truncated devs in the same way (an exercise for the 96*57121Smuller * dedicated reader; always wanted to say that...:) 97*57121Smuller */ 98*57121Smuller 99*57121Smuller typedef struct devt { 100*57121Smuller dev_t dev; /* the orig device number we now have to map */ 101*57121Smuller struct devt *fow; /* new device map list */ 102*57121Smuller struct dlist *list; /* map list based on inode truncation bits */ 103*57121Smuller } DEVT; 104*57121Smuller 105*57121Smuller typedef struct dlist { 106*57121Smuller ino_t trunc_bits; /* truncation pattern for a specific map */ 107*57121Smuller dev_t dev; /* the new device id we use */ 108*57121Smuller struct dlist *fow; 109*57121Smuller } DLIST; 110*57121Smuller 111*57121Smuller /* 112*57121Smuller * ftree directory access time reset table. When we are done with with a 113*57121Smuller * subtree we reset the access and mod time of the directory when the tflag is 114*57121Smuller * set. Not really explicitly specified in the pax spec, but easy and fast to 115*57121Smuller * do (and this may have even been intended in the spec, it is not clear). 116*57121Smuller * table is hashed by inode with chaining. 117*57121Smuller */ 118*57121Smuller 119*57121Smuller typedef struct atdir { 120*57121Smuller char *name; /* name of directory to reset */ 121*57121Smuller dev_t dev; /* dev and inode for fast lookup */ 122*57121Smuller ino_t ino; 123*57121Smuller time_t mtime; /* access and mod time to reset to */ 124*57121Smuller time_t atime; 125*57121Smuller struct atdir *fow; 126*57121Smuller } ATDIR; 127*57121Smuller 128*57121Smuller /* 129*57121Smuller * created directory time and mode storage entry. After pax is finished during 130*57121Smuller * extraction or copy, we must reset directory access modes and times that 131*57121Smuller * may have been modified after creation (they no longer have the specified 132*57121Smuller * times and/or modes). We must reset time in the reverse order of creation, 133*57121Smuller * because entries are added from the top of the file tree to the bottom. 134*57121Smuller * We MUST reset times from leaf to root (it will not work the other 135*57121Smuller * direction). Entries are recorded into a spool file to make reverse 136*57121Smuller * reading faster. 137*57121Smuller */ 138*57121Smuller 139*57121Smuller typedef struct dirdata { 140*57121Smuller int nlen; /* length of the directory name (includes \0) */ 141*57121Smuller off_t npos; /* position in file where this dir name starts */ 142*57121Smuller mode_t mode; /* file mode to restore */ 143*57121Smuller time_t mtime; /* mtime to set */ 144*57121Smuller time_t atime; /* atime to set */ 145*57121Smuller int frc_mode; /* do we force mode settings? */ 146*57121Smuller } DIRDATA; 147