157121Smuller /*- 257121Smuller * Copyright (c) 1992 Keith Muller. 3*60676Sbostic * Copyright (c) 1992, 1993 4*60676Sbostic * The Regents of the University of California. All rights reserved. 557121Smuller * 657121Smuller * This code is derived from software contributed to Berkeley by 757121Smuller * Keith Muller of the University of California, San Diego. 857121Smuller * 957121Smuller * %sccs.include.redist.c% 1057121Smuller * 11*60676Sbostic * @(#)tables.h 8.1 (Berkeley) 05/31/93 1257121Smuller */ 1357121Smuller 1457121Smuller /* 1557121Smuller * data structures and constants used by the different databases kept by pax 1657121Smuller */ 1757121Smuller 1857121Smuller /* 1957121Smuller * Hash Table Sizes MUST BE PRIME, if set too small performance suffers. 2057121Smuller * Probably safe to expect 500000 inodes per tape. Assuming good key 2157121Smuller * distribution (inodes) chains of under 50 long (worse case) is ok. 2257121Smuller */ 2357121Smuller #define L_TAB_SZ 2503 /* hard link hash table size */ 2457121Smuller #define F_TAB_SZ 50503 /* file time hash table size */ 2557121Smuller #define N_TAB_SZ 541 /* interactive rename hash table */ 2657121Smuller #define D_TAB_SZ 317 /* unique device mapping table */ 2757121Smuller #define A_TAB_SZ 317 /* ftree dir access time reset table */ 2857121Smuller #define MAXKEYLEN 64 /* max number of chars for hash */ 2957121Smuller 3057121Smuller /* 3157121Smuller * file hard link structure (hashed by dev/ino and chained) used to find the 3257121Smuller * hard links in a file system or with some archive formats (cpio) 3357121Smuller */ 3457121Smuller typedef struct hrdlnk { 3557121Smuller char *name; /* name of first file seen with this ino/dev */ 3657121Smuller dev_t dev; /* files device number */ 3757121Smuller ino_t ino; /* files inode number */ 3857121Smuller u_long nlink; /* expected link count */ 3957121Smuller struct hrdlnk *fow; 4057121Smuller } HRDLNK; 4157121Smuller 4257121Smuller /* 4357121Smuller * Archive write update file time table (the -u, -C flag), hashed by filename. 4457121Smuller * Filenames are stored in a scratch file at seek offset into the file. The 4557121Smuller * file time (mod time) and the file name length (for a quick check) are 4657121Smuller * stored in a hash table node. We were forced to use a scratch file because 4757121Smuller * with -u, the mtime for every node in the archive must always be available 4857121Smuller * to compare against (and this data can get REALLY large with big archives). 4957121Smuller * By being careful to read only when we have a good chance of a match, the 5057121Smuller * performance loss is not measurable (and the size of the archive we can 5157121Smuller * handle is greatly increased). 5257121Smuller */ 5357121Smuller typedef struct ftm { 5457121Smuller int namelen; /* file name length */ 5557121Smuller time_t mtime; /* files last modification time */ 5657121Smuller off_t seek; /* loacation in scratch file */ 5757121Smuller struct ftm *fow; 5857121Smuller } FTM; 5957121Smuller 6057121Smuller /* 6157121Smuller * Interactive rename table (-i flag), hashed by orig filename. 6257121Smuller * We assume this will not be a large table as this mapping data can only be 6357121Smuller * obtained through interactive input by the user. Nobody is going to type in 6457121Smuller * changes for 500000 files? We use chaining to resolve collisions. 6557121Smuller */ 6657121Smuller 6757121Smuller typedef struct namt { 6857121Smuller char *oname; /* old name */ 6957121Smuller char *nname; /* new name typed in by the user */ 7057121Smuller struct namt *fow; 7157121Smuller } NAMT; 7257121Smuller 7357121Smuller /* 7457121Smuller * Unique device mapping tables. Some protocols (e.g. cpio) require that the 7557121Smuller * <c_dev,c_ino> pair will uniquely identify a file in an archive unless they 7657121Smuller * are links to the same file. Appending to archives can break this. For those 7757121Smuller * protocols that have this requirement we map c_dev to a unique value not seen 7857121Smuller * in the archive when we append. We also try to handle inode truncation with 7957121Smuller * this table. (When the inode field in the archive header are too small, we 8057121Smuller * remap the dev on writes to remove accidental collisions). 8157121Smuller * 8257121Smuller * The list is hashed by device number using chain collision resolution. Off of 8357121Smuller * each DEVT are linked the various remaps for this device based on those bits 8457121Smuller * in the inode which were truncated. For example if we are just remapping to 8557121Smuller * avoid a device number during an update append, off the DEVT we would have 8657121Smuller * only a single DLIST that has a truncation id of 0 (no inode bits were 8757121Smuller * stripped for this device so far). When we spot inode truncation we create 8857121Smuller * a new mapping based on the set of bits in the inode which were stripped off. 8957121Smuller * so if the top four bits of the inode are stripped and they have a pattern of 9057121Smuller * 0110...... (where . are those bits not truncated) we would have a mapping 9157121Smuller * assigned for all inodes that has the same 0110.... pattern (with this dev 9257121Smuller * number of course). This keeps the mapping sparse and should be able to store 9357121Smuller * close to the limit of files which can be represented by the optimal 9457121Smuller * combination of dev and inode bits, and without creating a fouled up archive. 9557121Smuller * Note we also remap truncated devs in the same way (an exercise for the 9657121Smuller * dedicated reader; always wanted to say that...:) 9757121Smuller */ 9857121Smuller 9957121Smuller typedef struct devt { 10057121Smuller dev_t dev; /* the orig device number we now have to map */ 10157121Smuller struct devt *fow; /* new device map list */ 10257121Smuller struct dlist *list; /* map list based on inode truncation bits */ 10357121Smuller } DEVT; 10457121Smuller 10557121Smuller typedef struct dlist { 10657121Smuller ino_t trunc_bits; /* truncation pattern for a specific map */ 10757121Smuller dev_t dev; /* the new device id we use */ 10857121Smuller struct dlist *fow; 10957121Smuller } DLIST; 11057121Smuller 11157121Smuller /* 11257121Smuller * ftree directory access time reset table. When we are done with with a 11357121Smuller * subtree we reset the access and mod time of the directory when the tflag is 11457121Smuller * set. Not really explicitly specified in the pax spec, but easy and fast to 11557121Smuller * do (and this may have even been intended in the spec, it is not clear). 11657121Smuller * table is hashed by inode with chaining. 11757121Smuller */ 11857121Smuller 11957121Smuller typedef struct atdir { 12057121Smuller char *name; /* name of directory to reset */ 12157121Smuller dev_t dev; /* dev and inode for fast lookup */ 12257121Smuller ino_t ino; 12357121Smuller time_t mtime; /* access and mod time to reset to */ 12457121Smuller time_t atime; 12557121Smuller struct atdir *fow; 12657121Smuller } ATDIR; 12757121Smuller 12857121Smuller /* 12957121Smuller * created directory time and mode storage entry. After pax is finished during 13057121Smuller * extraction or copy, we must reset directory access modes and times that 13157121Smuller * may have been modified after creation (they no longer have the specified 13257121Smuller * times and/or modes). We must reset time in the reverse order of creation, 13357121Smuller * because entries are added from the top of the file tree to the bottom. 13457121Smuller * We MUST reset times from leaf to root (it will not work the other 13557121Smuller * direction). Entries are recorded into a spool file to make reverse 13657121Smuller * reading faster. 13757121Smuller */ 13857121Smuller 13957121Smuller typedef struct dirdata { 14057121Smuller int nlen; /* length of the directory name (includes \0) */ 14157121Smuller off_t npos; /* position in file where this dir name starts */ 14257121Smuller mode_t mode; /* file mode to restore */ 14357121Smuller time_t mtime; /* mtime to set */ 14457121Smuller time_t atime; /* atime to set */ 14557121Smuller int frc_mode; /* do we force mode settings? */ 14657121Smuller } DIRDATA; 147