* Copyright (c) 1992 Keith Muller.
* Copyright (c) 1992, 1993
* The Regents of the University of California. All rights reserved.
* This code is derived from software contributed to Berkeley by
* Keith Muller of the University of California, San Diego.
* %sccs.include.redist.c%
static char sccsid
[] = "@(#)tables.c 8.1 (Berkeley) %G%";
* Routines for controlling the contents of all the different databases pax
* keeps. Tables are dynamically created only when they are needed. The
* goal was speed and the ability to work with HUGE archives. The databases
* were kept simple, but do have complex rules for when the contents change.
* As of this writing, the posix library functions were more complex than
* needed for this application (pax databases have very short lifetimes and
* do not survive after pax is finished). Pax is required to handle very
* large archives. These database routines carefully combine memory usage and
* temporary file storage in ways which will not significantly impact runtime
* performance while allowing the largest possible archives to be handled.
* Trying to force the fit to the posix databases routines was not considered
static HRDLNK
**ltab
= NULL
; /* hard link table for detecting hard links */
static FTM
**ftab
= NULL
; /* file time table for updating arch */
static NAMT
**ntab
= NULL
; /* interactive rename storage table */
static DEVT
**dtab
= NULL
; /* device/inode mapping tables */
static ATDIR
**atab
= NULL
; /* file tree directory time reset table */
static int dirfd
= -1; /* storage for setting created dir time/mode */
static u_long dircnt
; /* entries in dir time/mode storage */
static int ffd
= -1; /* tmp file for file time table name storage */
static DEVT
*chk_dev
__P((dev_t
, int));
* hard link table routines
* The hard link table tries to detect hard links to files using the device and
* inode values. We do this when writing an archive, so we can tell the format
* write routine that this file is a hard link to another file. The format
* write routine then can store this file in whatever way it wants (as a hard
* link if the format supports that like tar, or ignore this info like cpio).
* (Actually a field in the format driver table tells us if the format wants
* hard link info. if not, we do not waste time looking for them). We also use
* the same table when reading an archive. In that situation, this table is
* used by the format read routine to detect hard links from stored dev and
* inode numbers (like cpio). This will allow pax to create a link when one
* can be detected by the archive format.
* Creates the hard link table.
* 0 if created, -1 if failure
if ((ltab
= (HRDLNK
**)calloc(L_TAB_SZ
, sizeof(HRDLNK
*))) == NULL
) {
warn(1, "Cannot allocate memory for hard link table");
* Looks up entry in hard link hash table. If found, it copies the name
* of the file it is linked to (we already saw that file) into ln_name.
* lnkcnt is decremented and if goes to 1 the node is deleted from the
* database. (We have seen all the links to this file). If not found,
* we add the file to the database if it has the potential for having
* hard links to other files we may process (it has a link count > 1)
* if found returns 1; if not found returns 0; -1 on error
chk_lnk(register ARCHD
*arcn
)
* ignore those nodes that cannot have hard links
if ((arcn
->type
== PAX_DIR
) || (arcn
->sb
.st_nlink
<= 1))
* hash inode number and look for this file
indx
= ((unsigned)arcn
->sb
.st_ino
) % L_TAB_SZ
;
if ((pt
= ltab
[indx
]) != NULL
) {
* it's hash chain in not empty, walk down looking for it
if ((pt
->ino
== arcn
->sb
.st_ino
) &&
(pt
->dev
== arcn
->sb
.st_dev
))
* found a link. set the node type and copy in the
* name of the file it is to link to. we need to
* handle hardlinks to regular files differently than
arcn
->ln_nlen
= l_strncpy(arcn
->ln_name
, pt
->name
,
if (arcn
->type
== PAX_REG
)
* if we have found all the links to this file, remove
(void)free((char *)pt
->name
);
* we never saw this file before. It has links so we add it to the
* front of this hash chain
if ((pt
= (HRDLNK
*)malloc(sizeof(HRDLNK
))) != NULL
) {
if ((pt
->name
= strdup(arcn
->name
)) != NULL
) {
pt
->dev
= arcn
->sb
.st_dev
;
pt
->ino
= arcn
->sb
.st_ino
;
pt
->nlink
= arcn
->sb
.st_nlink
;
warn(1, "Hard link table out of memory");
* remove reference for a file that we may have added to the data base as
* a potential source for hard links. We ended up not using the file, so
* we do not want to accidently point another file at it later on.
purg_lnk(register ARCHD
*arcn
)
* do not bother to look if it could not be in the database
if ((arcn
->sb
.st_nlink
<= 1) || (arcn
->type
== PAX_DIR
) ||
(arcn
->type
== PAX_HLK
) || (arcn
->type
== PAX_HRG
))
* find the hash chain for this inode value, if empty return
indx
= ((unsigned)arcn
->sb
.st_ino
) % L_TAB_SZ
;
if ((pt
= ltab
[indx
]) == NULL
)
* walk down the list looking for the inode/dev pair, unlink and
if ((pt
->ino
== arcn
->sb
.st_ino
) &&
(pt
->dev
== arcn
->sb
.st_dev
))
(void)free((char *)pt
->name
);
* pull apart a existing link table so we can reuse it. We do this between
* read and write phases of append with update. (The format may have
* used the link table, and we need to start with a fresh table for the
for (i
= 0; i
< L_TAB_SZ
; ++i
) {
* free up each entry on this chain
(void)free((char *)ppt
->name
);
* modification time table routines
* The modification time table keeps track of last modification times for all
* files stored in an archive during a write phase when -u is set. We only
* add a file to the archive if it is newer than a file with the same name
* already stored on the archive (if there is no other file with the same
* name on the archive it is added). This applies to writes and appends.
* An append with an -u must read the archive and store the modification time
* for every file on that archive before starting the write phase. It is clear
* that this is one HUGE database. To save memory space, the actual file names
* are stored in a scatch file and indexed by an in memory hash table. The
* hash table is indexed by hashing the file path. The nodes in the table store
* the length of the filename and the lseek offset within the scratch file
* where the actual name is stored. Since there are never any deletions to this
* table, fragmentation of the scratch file is never a issue. Lookups seem to
* not exhibit any locality at all (files in the database are rarely
* looked up more than once...). So caching is just a waste of memory. The
* only limitation is the amount of scatch file space available to store the
* create the file time hash table and open for read/write the scratch
* file. (after created it is unlinked, so when we exit we leave
* 0 if the table and file was created ok, -1 otherwise
if ((ftab
= (FTM
**)calloc(F_TAB_SZ
, sizeof(FTM
*))) == NULL
) {
warn(1, "Cannot allocate memory for file time table");
* get random name and create temporary scratch file, unlink name
* so it will get removed on exit
if ((pt
= tempnam((char *)NULL
, (char *)NULL
)) == NULL
)
if ((ffd
= open(pt
, O_RDWR
| O_CREAT
, S_IRWXU
)) < 0) {
syswarn(1, errno
, "Unable to open temporary file: %s", pt
);
* looks up entry in file time hash table. If not found, the file is
* added to the hash table and the file named stored in the scratch file.
* If a file with the same name is found, the file times are compared and
* the most recent file time is retained. If the new file was younger (or
* was not in the database) the new file is selected for storage.
* 0 if file should be added to the archive, 1 if it should be skipped,
chk_ftime(register ARCHD
*arcn
)
char ckname
[PAXPATHLEN
+1];
* no info, go ahead and add to archive
* hash the pathname and look up in table
indx
= st_hash(arcn
->name
, namelen
, F_TAB_SZ
);
if ((pt
= ftab
[indx
]) != NULL
) {
* the hash chain is not empty, walk down looking for match
* only read up the path names if the lengths match, speeds
if (pt
->namelen
== namelen
) {
* potential match, have to read the name
if (lseek(ffd
,pt
->seek
,SEEK_SET
) != pt
->seek
) {
"Failed ftime table seek");
if (read(ffd
, ckname
, namelen
) != namelen
) {
"Failed ftime table read");
* if the names match, we are done
if (!strncmp(ckname
, arcn
->name
, namelen
))
* try the next entry on the chain
* found the file, compare the times, save the newer
if (arcn
->sb
.st_mtime
> pt
->mtime
) {
pt
->mtime
= arcn
->sb
.st_mtime
;
if ((pt
= (FTM
*)malloc(sizeof(FTM
))) != NULL
) {
* add the name at the end of the scratch file, saving the
* offset. add the file to the head of the hash chain
if ((pt
->seek
= lseek(ffd
, (off_t
)0, SEEK_END
)) >= 0) {
if (write(ffd
, arcn
->name
, namelen
) == namelen
) {
pt
->mtime
= arcn
->sb
.st_mtime
;
syswarn(1, errno
, "Failed write to file time table");
syswarn(1, errno
, "Failed seek on file time table");
warn(1, "File time table ran out of memory");
* Interactive rename table routines
* The interactive rename table keeps track of the new names that the user
* assignes to files from tty input. Since this map is unique for each file
* we must store it in case there is a reference to the file later in archive
* (a link). Otherwise we will be unable to find the file we know was
* extracted. The remapping of these files is stored in a memory based hash
* table (it is assumed since input must come from /dev/tty, it is unlikely to
* be a very large table).
* create the interactive rename table
* 0 if successful, -1 otherwise
if ((ntab
= (NAMT
**)calloc(N_TAB_SZ
, sizeof(NAMT
*))) == NULL
) {
warn(1, "Cannot allocate memory for interactive rename table");
* add the new name to old name mapping just created by the user.
* If an old name mapping is found (there may be duplicate names on an
* archive) only the most recent is kept.
* 0 if added, -1 otherwise
add_name(register char *oname
, int onamelen
, char *nname
)
add_name(oname
, onamelen
, nname
)
warn(0, "No interactive rename table, links may fail\n");
* look to see if we have already mapped this file, if so we
indx
= st_hash(oname
, onamelen
, N_TAB_SZ
);
if ((pt
= ntab
[indx
]) != NULL
) {
* look down the has chain for the file
while ((pt
!= NULL
) && (strcmp(oname
, pt
->oname
) != 0))
* found an old mapping, replace it with the new one
* the user just input (if it is different)
if (strcmp(nname
, pt
->nname
) == 0)
(void)free((char *)pt
->nname
);
if ((pt
->nname
= strdup(nname
)) == NULL
) {
warn(1, "Cannot update rename table");
* this is a new mapping, add it to the table
if ((pt
= (NAMT
*)malloc(sizeof(NAMT
))) != NULL
) {
if ((pt
->oname
= strdup(oname
)) != NULL
) {
if ((pt
->nname
= strdup(nname
)) != NULL
) {
(void)free((char *)pt
->oname
);
warn(1, "Interactive rename table out of memory");
* look up a link name to see if it points at a file that has been
* remapped by the user. If found, the link is adjusted to contain the
* new name (oname is the link to name)
sub_name(register char *oname
, int *onamelen
)
sub_name(oname
, onamelen
)
* look the name up in the hash table
indx
= st_hash(oname
, *onamelen
, N_TAB_SZ
);
if ((pt
= ntab
[indx
]) == NULL
)
* walk down the hash cahin looking for a match
if (strcmp(oname
, pt
->oname
) == 0) {
* found it, replace it with the new name
* and return (we know that oname has enough space)
*onamelen
= l_strncpy(oname
, pt
->nname
, PAXPATHLEN
+1);
* device/inode mapping table routines
* (used with formats that store device and inodes fields)
* device/inode mapping tables remap the device field in a archive header. The
* device/inode fields are used to determine when files are hard links to each
* other. However these values have very little meaning outside of that. This
* database is used to solve one of two different problems.
* 1) when files are appended to an archive, while the new files may have hard
* links to each other, you cannot determine if they have hard links to any
* file already stored on the archive from a prior run of pax. We must assume
* that these inode/device pairs are unique only within a SINGLE run of pax
* (which adds a set of files to an archive). So we have to make sure the
* inode/dev pairs we add each time are always unique. We do this by observing
* while the inode field is very dense, the use of the dev field is fairly
* sparse. Within each run of pax, we remap any device number of a new archive
* member that has a device number used in a prior run and already stored in a
* file on the archive. During the read phase of the append, we store the
* device numbers used and mark them to not be used by any file during the
* write phase. If during write we go to use one of those old device numbers,
* we remap it to a new value.
* 2) Often the fields in the archive header used to store these values are
* too small to store the entire value. The result is an inode or device value
* which can be truncated. This really can foul up an archive. With truncation
* we end up creating links between files that are really not links (after
* truncation the inodes are the same value). We address that by detecting
* truncation and forcing a remap of the device field to split truncated
* inodes away from each other. Each truncation creates a pattern of bits that
* are removed. We use this pattern of truncated bits to partition the inodes
* on a single device to many different devices (each one represented by the
* truncated bit pattern). All inodes on the same device that have the same
* truncation pattern are mapped to the same new device. Two inodes that
* truncate to the same value clearly will always have different truncation
* bit patterns, so they will be split from away each other. When we spot
* device truncation we remap the device number to a non truncated value.
* (for more info see table.h for the data structures involved).
* create the device mapping table
* 0 if successful, -1 otherwise
if ((dtab
= (DEVT
**)calloc(D_TAB_SZ
, sizeof(DEVT
*))) == NULL
) {
warn(1, "Cannot allocate memory for device mapping table");
* add a device number to the table. this will force the device to be
* remapped to a new value if it be used during a write phase. This
* function is called during the read phase of an append to prohibit the
* use of any device number already in the archive.
* 0 if added ok, -1 otherwise
add_dev(register ARCHD
*arcn
)
if (chk_dev(arcn
->sb
.st_dev
, 1) == NULL
)
* check for a device value in the device table. If not found and the add
* flag is set, it is added. This does NOT assign any mapping values, just
* adds the device number as one that need to be remapped. If this device
* is alread mapped, just return with a pointer to that entry.
* pointer to the entry for this device in the device map table. Null
* if the add flag is not set and the device is not in the table (it is
* not been seen yet). If add is set and the device cannot be added, null
* is returned (indicates an error).
chk_dev(dev_t dev
, int add
)
* look to see if this device is already in the table
indx
= ((unsigned)dev
) % D_TAB_SZ
;
if ((pt
= dtab
[indx
]) != NULL
) {
while ((pt
!= NULL
) && (pt
->dev
!= dev
))
* found it, return a pointer to it
* not in table, we add it only if told to as this may just be a check
* to see if a device number is being used.
* allocate a node for this device and add it to the front of the hash
* chain. Note we do not assign remaps values here, so the pt->list
if ((pt
= (DEVT
*)malloc(sizeof(DEVT
))) == NULL
) {
warn(1, "Device map table out of memory");
* given an inode and device storage mask (the mask has a 1 for each bit
* the archive format is able to store in a header), we check for inode
* and device truncation and remap the device as required. Device mapping
* can also occur when during the read phase of append a device number was
* seen (and was marked as do not use during the write phase). WE ASSUME
* that unsigned longs are the same size or bigger than the fields used
* for ino_t and dev_t. If not the types will have to be changed.
* 0 if all ok, -1 otherwise.
map_dev(register ARCHD
*arcn
, u_long dev_mask
, u_long ino_mask
)
map_dev(arcn
, dev_mask
, ino_mask
)
static dev_t lastdev
= 0; /* next device number to try */
* check for device and inode truncation, and extract the truncated
if ((arcn
->sb
.st_dev
& (dev_t
)dev_mask
) != arcn
->sb
.st_dev
)
if ((nino
= arcn
->sb
.st_ino
& (ino_t
)ino_mask
) != arcn
->sb
.st_ino
) {
trunc_bits
= arcn
->sb
.st_ino
& (ino_t
)(~ino_mask
);
* see if this device is already being mapped, look up the device
* then find the truncation bit pattern which applies
if ((pt
= chk_dev(arcn
->sb
.st_dev
, 0)) != NULL
) {
* this device is already marked to be remapped
for (dpt
= pt
->list
; dpt
!= NULL
; dpt
= dpt
->fow
)
if (dpt
->trunc_bits
== trunc_bits
)
* we are being remapped for this device and pattern
* change the device number to be stored and return
arcn
->sb
.st_dev
= dpt
->dev
;
* this device is not being remapped YET. if we do not have any
* form of truncation, we do not need a remap
if (!trc_ino
&& !trc_dev
)
* we have truncation, have to add this as a device to remap
if ((pt
= chk_dev(arcn
->sb
.st_dev
, 1)) == NULL
)
* if we just have a truncated inode, we have to make sure that
* all future inodes that do not truncate (they have the
* truncation pattern of all 0's) continue to map to the same
* device number. We probably have already written inodes with
* this device number to the archive with the truncation
* pattern of all 0's. So we add the mapping for all 0's to the
if (!trc_dev
&& (trunc_bits
!= 0)) {
if ((dpt
= (DLIST
*)malloc(sizeof(DLIST
))) == NULL
)
dpt
->dev
= arcn
->sb
.st_dev
;
* look for a device number not being used. We must watch for wrap
* around on lastdev (so we do not get stuck looking forever!)
if (chk_dev(lastdev
, 0) != NULL
)
* found an unused value. If we have reached truncation point
* for this format we are hosed, so we give up. Otherwise we
if (((lastdev
& ((dev_t
)dev_mask
)) != lastdev
) ||
(chk_dev(lastdev
, 1) == NULL
))
if ((lastdev
<= 0) || ((dpt
= (DLIST
*)malloc(sizeof(DLIST
))) == NULL
))
* got a new device number, store it under this truncation pattern.
* change the device number this file is being stored with.
dpt
->trunc_bits
= trunc_bits
;
arcn
->sb
.st_dev
= lastdev
;
warn(1, "Unable to fix truncated inode/device field when storing %s",
warn(0, "Archive may create improper hard links when extracted");
* directory access/mod time reset table routines (for directories READ by pax)
* The pax -t flag requires that access times of archive files to be the same
* before being read by pax. For regular files, access time is restored after
* the file has been copied. This database provides the same functionality for
* directories read during file tree traversal. Restoring directory access time
* is more complex than files since directories may be read several times until
* all the descendants in their subtree are visited by fts. Directory access
* and modification times are stored during the fts pre-order visit (done
* before any descendants in the subtree is visited) and restored after the
* fts post-order visit (after all the descendants have been visited). In the
* case of premature exit from a subtree (like from the effects of -n), any
* directory entries left in this database are reset during final cleanup
* operations of pax. Entries are hashed by inode number for fast lookup.
* create the directory access time database for directories READ by pax.
* 0 is created ok, -1 otherwise.
if ((atab
= (ATDIR
**)calloc(A_TAB_SZ
, sizeof(ATDIR
*))) == NULL
) {
warn(1,"Cannot allocate space for directory access time table");
* walk through the directory access time table and reset the access time
* of any directory who still has an entry left in the database. These
* entries are for directories READ by pax
* for each non-empty hash table entry reset all the directories
for (i
= 0; i
< A_TAB_SZ
; ++i
) {
if ((pt
= atab
[i
]) == NULL
)
* remember to force the times, set_ftime() looks at pmtime
* and patime, which only applies to things CREATED by pax,
* not read by pax. Read time reset is controlled by -t.
for (; pt
!= NULL
; pt
= pt
->fow
)
set_ftime(pt
->name
, pt
->mtime
, pt
->atime
, 1);
* add a directory to the directory access time table. Table is hashed
* and chained by inode number. This is for directories READ by pax
add_atdir(char *fname
, dev_t dev
, ino_t ino
, time_t mtime
, time_t atime
)
add_atdir(fname
, dev
, ino
, mtime
, atime
)
* make sure this directory is not already in the table, if so just
* return (the older entry always has the correct time). The only
* way this will happen is when the same subtree can be traversed by
* different args to pax and the -n option is aborting fts out of a
* subtree before all the post-order visits have been made).
indx
= ((unsigned)ino
) % A_TAB_SZ
;
if ((pt
= atab
[indx
]) != NULL
) {
if ((pt
->ino
== ino
) && (pt
->dev
== dev
))
* oops, already there. Leave it alone.
* add it to the front of the hash chain
if ((pt
= (ATDIR
*)malloc(sizeof(ATDIR
))) != NULL
) {
if ((pt
->name
= strdup(fname
)) != NULL
) {
warn(1, "Directory access time reset table ran out of memory");
* look up a directory by inode and device number to obtain the access
* and modification time you want to set to. If found, the modification
* and access time parameters are set and the entry is removed from the
* table (as it is no longer needed). These are for directories READ by
* 0 if found, -1 if not found.
get_atdir(dev_t dev
, ino_t ino
, time_t *mtime
, time_t *atime
)
get_atdir(dev
, ino
, mtime
, atime
)
* hash by inode and search the chain for an inode and device match
indx
= ((unsigned)ino
) % A_TAB_SZ
;
if ((pt
= atab
[indx
]) == NULL
)
if ((pt
->ino
== ino
) && (pt
->dev
== dev
))
* no match, go to next one
* return if we did not find it.
* found it. return the times and remove the entry from the table.
(void)free((char *)pt
->name
);
* directory access mode and time storage routines (for directories CREATED
* Pax requires that extracted directories, by default, have their access/mod
* times and permissions set to the values specified in the archive. During the
* actions of extracting (and creating the destination subtree during -rw copy)
* directories extracted may be modified after being created. Even worse is
* that these directories may have been created with file permissions which
* prohibits any descendants of these directories from being extracted. When
* directories are created by pax, access rights may be added to permit the
* creation of files in their subtree. Every time pax creates a directory, the
* times and file permissions specified by the archive are stored. After all
* files have been extracted (or copied), these directories have their times
* and file modes reset to the stored values. The directory info is restored in
* reverse order as entries were added to the data file from root to leaf. To
* restore atime properly, we must go backwards. The data file consists of
* records with two parts, the file name followed by a DIRDATA trailer. The
* fixed sized trailer contains the size of the name plus the off_t location in
* the file. To restore we work backwards through the file reading the trailer
* set up the directory time and file mode storage for directories CREATED
if ((pt
= tempnam((char *)NULL
, (char *)NULL
)) == NULL
)
* unlink the file so it goes away at termination by itself
if ((dirfd
= open(pt
, O_RDWR
|O_CREAT
, 0600)) >= 0) {
warn(1, "Unable to create temporary file for directory times: %s", pt
);
* add the mode and times for a newly CREATED directory
* name is name of the directory, psb the stat buffer with the data in it,
* frc_mode is a flag that says whether to force the setting of the mode
* (ignoring the user set values for preserving file mode). Frc_mode is
* for the case where we created a file and found that the resulting
* directory was not writeable and the user asked for file modes to NOT
* be preserved. (we have to preserve what was created by default, so we
* have to force the setting at the end. this is stated explicitly in the
add_dir(char *name
, int nlen
, struct stat
*psb
, int frc_mode
)
add_dir(name
, nlen
, psb
, frc_mode
)
* get current position (where file name will start) so we can store it
if ((dblk
.npos
= lseek(dirfd
, 0L, SEEK_CUR
)) < 0) {
warn(1,"Unable to store mode and times for directory: %s",name
);
* write the file name followed by the trailer
dblk
.mode
= psb
->st_mode
& 0xffff;
dblk
.mtime
= psb
->st_mtime
;
dblk
.atime
= psb
->st_atime
;
dblk
.frc_mode
= frc_mode
;
if ((write(dirfd
, name
, dblk
.nlen
) == dblk
.nlen
) &&
(write(dirfd
, (char *)&dblk
, sizeof(dblk
)) == sizeof(dblk
))) {
warn(1,"Unable to store mode and times for created directory: %s",name
);
* process all file modes and times stored for directories CREATED
* read backwards through the file and process each directory
for (cnt
= 0; cnt
< dircnt
; ++cnt
) {
* read the trailer, then the file name, if this fails
if (lseek(dirfd
, -((off_t
)sizeof(dblk
)), SEEK_CUR
) < 0)
if (read(dirfd
,(char *)&dblk
, sizeof(dblk
)) != sizeof(dblk
))
if (lseek(dirfd
, dblk
.npos
, SEEK_SET
) < 0)
if (read(dirfd
, name
, dblk
.nlen
) != dblk
.nlen
)
if (lseek(dirfd
, dblk
.npos
, SEEK_SET
) < 0)
* frc_mode set, make sure we set the file modes even if
* the user didn't ask for it (see file_subs.c for more info)
if (pmode
|| dblk
.frc_mode
)
set_pmode(name
, dblk
.mode
);
set_ftime(name
, dblk
.mtime
, dblk
.atime
, 0);
warn(1,"Unable to set mode and times for created directories");
* database independent routines
* hashes filenames to a u_int for hashing into a table. Looks at the tail
* end of file, as this provides far better distribution than any other
* part of the name. For performance reasons we only care about the last
* MAXKEYLEN chars (should be at LEAST large enough to pick off the file
* name). Was tested on 500,000 name file tree traversal from the root
* and gave almost a perfectly uniform distribution of keys when used with
* prime sized tables (MAXKEYLEN was 128 in test). Hashes (sizeof int)
* chars at a time and pads with 0 for last addition.
* the hash value of the string MOD (%) the table size.
st_hash(char *name
, int len
, int tabsz
)
st_hash(name
, len
, tabsz
)
* only look at the tail up to MAXKEYLEN, we do not need to waste
* time here (remember these are pathnames, the tail is what will
pt
= &(name
[len
- MAXKEYLEN
]);
* calculate the number of u_int size steps in the string and if
* there is a runt to deal with
steps
= len
/sizeof(u_int
);
res
= len
% sizeof(u_int
);
* add up the value of the string in unsigned integer sized pieces
* too bad we cannot have unsigned int aligned strings, then we
* could avoid the expensive copy.
for (i
= 0; i
< steps
; ++i
) {
end
= pt
+ sizeof(u_int
);
* add in the runt padded with zero to the right
* return the result mod the table size