2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1992 Keith Muller.
5 * Copyright (c) 1992, 1993
6 * The Regents of the University of California. All rights reserved.
8 * This code is derived from software contributed to Berkeley by
9 * Keith Muller of the University of California, San Diego.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. Neither the name of the University nor the names of its contributors
20 * may be used to endorse or promote products derived from this software
21 * without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 static char sccsid[] = "@(#)tables.c 8.1 (Berkeley) 5/31/93";
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
44 #include <sys/types.h>
47 #include <sys/fcntl.h>
58 * Routines for controlling the contents of all the different databases pax
59 * keeps. Tables are dynamically created only when they are needed. The
60 * goal was speed and the ability to work with HUGE archives. The databases
61 * were kept simple, but do have complex rules for when the contents change.
62 * As of this writing, the POSIX library functions were more complex than
63 * needed for this application (pax databases have very short lifetimes and
64 * do not survive after pax is finished). Pax is required to handle very
65 * large archives. These database routines carefully combine memory usage and
66 * temporary file storage in ways which will not significantly impact runtime
67 * performance while allowing the largest possible archives to be handled.
68 * Trying to force the fit to the POSIX databases routines was not considered
72 static HRDLNK **ltab = NULL; /* hard link table for detecting hard links */
73 static FTM **ftab = NULL; /* file time table for updating arch */
74 static NAMT **ntab = NULL; /* interactive rename storage table */
75 static DEVT **dtab = NULL; /* device/inode mapping tables */
76 static ATDIR **atab = NULL; /* file tree directory time reset table */
77 static int dirfd = -1; /* storage for setting created dir time/mode */
78 static u_long dircnt; /* entries in dir time/mode storage */
79 static int ffd = -1; /* tmp file for file time table name storage */
81 static DEVT *chk_dev(dev_t, int);
84 * hard link table routines
86 * The hard link table tries to detect hard links to files using the device and
87 * inode values. We do this when writing an archive, so we can tell the format
88 * write routine that this file is a hard link to another file. The format
89 * write routine then can store this file in whatever way it wants (as a hard
90 * link if the format supports that like tar, or ignore this info like cpio).
91 * (Actually a field in the format driver table tells us if the format wants
92 * hard link info. if not, we do not waste time looking for them). We also use
93 * the same table when reading an archive. In that situation, this table is
94 * used by the format read routine to detect hard links from stored dev and
95 * inode numbers (like cpio). This will allow pax to create a link when one
96 * can be detected by the archive format.
101 * Creates the hard link table.
103 * 0 if created, -1 if failure
111 if ((ltab = (HRDLNK **)calloc(L_TAB_SZ, sizeof(HRDLNK *))) == NULL) {
112 paxwarn(1, "Cannot allocate memory for hard link table");
120 * Looks up entry in hard link hash table. If found, it copies the name
121 * of the file it is linked to (we already saw that file) into ln_name.
122 * lnkcnt is decremented and if goes to 1 the node is deleted from the
123 * database. (We have seen all the links to this file). If not found,
124 * we add the file to the database if it has the potential for having
125 * hard links to other files we may process (it has a link count > 1)
127 * if found returns 1; if not found returns 0; -1 on error
140 * ignore those nodes that cannot have hard links
142 if ((arcn->type == PAX_DIR) || (arcn->sb.st_nlink <= 1))
146 * hash inode number and look for this file
148 indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ;
149 if ((pt = ltab[indx]) != NULL) {
151 * it's hash chain in not empty, walk down looking for it
155 if ((pt->ino == arcn->sb.st_ino) &&
156 (pt->dev == arcn->sb.st_dev))
164 * found a link. set the node type and copy in the
165 * name of the file it is to link to. we need to
166 * handle hardlinks to regular files differently than
169 arcn->ln_nlen = l_strncpy(arcn->ln_name, pt->name,
170 sizeof(arcn->ln_name) - 1);
171 arcn->ln_name[arcn->ln_nlen] = '\0';
172 if (arcn->type == PAX_REG)
173 arcn->type = PAX_HRG;
175 arcn->type = PAX_HLK;
178 * if we have found all the links to this file, remove
179 * it from the database
181 if (--pt->nlink <= 1) {
191 * we never saw this file before. It has links so we add it to the
192 * front of this hash chain
194 if ((pt = (HRDLNK *)malloc(sizeof(HRDLNK))) != NULL) {
195 if ((pt->name = strdup(arcn->name)) != NULL) {
196 pt->dev = arcn->sb.st_dev;
197 pt->ino = arcn->sb.st_ino;
198 pt->nlink = arcn->sb.st_nlink;
199 pt->fow = ltab[indx];
206 paxwarn(1, "Hard link table out of memory");
212 * remove reference for a file that we may have added to the data base as
213 * a potential source for hard links. We ended up not using the file, so
214 * we do not want to accidentally point another file at it later on.
218 purg_lnk(ARCHD *arcn)
227 * do not bother to look if it could not be in the database
229 if ((arcn->sb.st_nlink <= 1) || (arcn->type == PAX_DIR) ||
230 (arcn->type == PAX_HLK) || (arcn->type == PAX_HRG))
234 * find the hash chain for this inode value, if empty return
236 indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ;
237 if ((pt = ltab[indx]) == NULL)
241 * walk down the list looking for the inode/dev pair, unlink and
246 if ((pt->ino == arcn->sb.st_ino) &&
247 (pt->dev == arcn->sb.st_dev))
265 * Pull apart an existing link table so we can reuse it. We do this between
266 * read and write phases of append with update. (The format may have
267 * used the link table, and we need to start with a fresh table for the
281 for (i = 0; i < L_TAB_SZ; ++i) {
288 * free up each entry on this chain
301 * modification time table routines
303 * The modification time table keeps track of last modification times for all
304 * files stored in an archive during a write phase when -u is set. We only
305 * add a file to the archive if it is newer than a file with the same name
306 * already stored on the archive (if there is no other file with the same
307 * name on the archive it is added). This applies to writes and appends.
308 * An append with an -u must read the archive and store the modification time
309 * for every file on that archive before starting the write phase. It is clear
310 * that this is one HUGE database. To save memory space, the actual file names
311 * are stored in a scratch file and indexed by an in memory hash table. The
312 * hash table is indexed by hashing the file path. The nodes in the table store
313 * the length of the filename and the lseek offset within the scratch file
314 * where the actual name is stored. Since there are never any deletions to this
315 * table, fragmentation of the scratch file is never an issue. Lookups seem to
316 * not exhibit any locality at all (files in the database are rarely
317 * looked up more than once...). So caching is just a waste of memory. The
318 * only limitation is the amount of scratch file space available to store the
324 * create the file time hash table and open for read/write the scratch
325 * file. (after created it is unlinked, so when we exit we leave
328 * 0 if the table and file was created ok, -1 otherwise
337 if ((ftab = (FTM **)calloc(F_TAB_SZ, sizeof(FTM *))) == NULL) {
338 paxwarn(1, "Cannot allocate memory for file time table");
343 * get random name and create temporary scratch file, unlink name
344 * so it will get removed on exit
346 memcpy(tempbase, _TFILE_BASE, sizeof(_TFILE_BASE));
347 if ((ffd = mkstemp(tempfile)) < 0) {
348 syswarn(1, errno, "Unable to create temporary file: %s",
352 (void)unlink(tempfile);
359 * looks up entry in file time hash table. If not found, the file is
360 * added to the hash table and the file named stored in the scratch file.
361 * If a file with the same name is found, the file times are compared and
362 * the most recent file time is retained. If the new file was younger (or
363 * was not in the database) the new file is selected for storage.
365 * 0 if file should be added to the archive, 1 if it should be skipped,
370 chk_ftime(ARCHD *arcn)
375 char ckname[PAXPATHLEN+1];
378 * no info, go ahead and add to archive
384 * hash the pathname and look up in table
386 namelen = arcn->nlen;
387 indx = st_hash(arcn->name, namelen, F_TAB_SZ);
388 if ((pt = ftab[indx]) != NULL) {
390 * the hash chain is not empty, walk down looking for match
391 * only read up the path names if the lengths match, speeds
392 * up the search a lot
395 if (pt->namelen == namelen) {
397 * potential match, have to read the name
398 * from the scratch file.
400 if (lseek(ffd,pt->seek,SEEK_SET) != pt->seek) {
402 "Failed ftime table seek");
405 if (read(ffd, ckname, namelen) != namelen) {
407 "Failed ftime table read");
412 * if the names match, we are done
414 if (!strncmp(ckname, arcn->name, namelen))
419 * try the next entry on the chain
426 * found the file, compare the times, save the newer
428 if (arcn->sb.st_mtime > pt->mtime) {
432 pt->mtime = arcn->sb.st_mtime;
443 * not in table, add it
445 if ((pt = (FTM *)malloc(sizeof(FTM))) != NULL) {
447 * add the name at the end of the scratch file, saving the
448 * offset. add the file to the head of the hash chain
450 if ((pt->seek = lseek(ffd, (off_t)0, SEEK_END)) >= 0) {
451 if (write(ffd, arcn->name, namelen) == namelen) {
452 pt->mtime = arcn->sb.st_mtime;
453 pt->namelen = namelen;
454 pt->fow = ftab[indx];
458 syswarn(1, errno, "Failed write to file time table");
460 syswarn(1, errno, "Failed seek on file time table");
462 paxwarn(1, "File time table ran out of memory");
470 * Interactive rename table routines
472 * The interactive rename table keeps track of the new names that the user
473 * assigns to files from tty input. Since this map is unique for each file
474 * we must store it in case there is a reference to the file later in archive
475 * (a link). Otherwise we will be unable to find the file we know was
476 * extracted. The remapping of these files is stored in a memory based hash
477 * table (it is assumed since input must come from /dev/tty, it is unlikely to
478 * be a very large table).
483 * create the interactive rename table
485 * 0 if successful, -1 otherwise
493 if ((ntab = (NAMT **)calloc(N_TAB_SZ, sizeof(NAMT *))) == NULL) {
494 paxwarn(1, "Cannot allocate memory for interactive rename table");
502 * add the new name to old name mapping just created by the user.
503 * If an old name mapping is found (there may be duplicate names on an
504 * archive) only the most recent is kept.
506 * 0 if added, -1 otherwise
510 add_name(char *oname, int onamelen, char *nname)
517 * should never happen
519 paxwarn(0, "No interactive rename table, links may fail\n");
524 * look to see if we have already mapped this file, if so we
527 indx = st_hash(oname, onamelen, N_TAB_SZ);
528 if ((pt = ntab[indx]) != NULL) {
530 * look down the has chain for the file
532 while ((pt != NULL) && (strcmp(oname, pt->oname) != 0))
537 * found an old mapping, replace it with the new one
538 * the user just input (if it is different)
540 if (strcmp(nname, pt->nname) == 0)
544 if ((pt->nname = strdup(nname)) == NULL) {
545 paxwarn(1, "Cannot update rename table");
553 * this is a new mapping, add it to the table
555 if ((pt = (NAMT *)malloc(sizeof(NAMT))) != NULL) {
556 if ((pt->oname = strdup(oname)) != NULL) {
557 if ((pt->nname = strdup(nname)) != NULL) {
558 pt->fow = ntab[indx];
566 paxwarn(1, "Interactive rename table out of memory");
572 * look up a link name to see if it points at a file that has been
573 * remapped by the user. If found, the link is adjusted to contain the
574 * new name (oname is the link to name)
578 sub_name(char *oname, int *onamelen, size_t onamesize)
586 * look the name up in the hash table
588 indx = st_hash(oname, *onamelen, N_TAB_SZ);
589 if ((pt = ntab[indx]) == NULL)
594 * walk down the hash chain looking for a match
596 if (strcmp(oname, pt->oname) == 0) {
598 * found it, replace it with the new name
599 * and return (we know that oname has enough space)
601 *onamelen = l_strncpy(oname, pt->nname, onamesize - 1);
602 oname[*onamelen] = '\0';
609 * no match, just return
615 * device/inode mapping table routines
616 * (used with formats that store device and inodes fields)
618 * device/inode mapping tables remap the device field in an archive header. The
619 * device/inode fields are used to determine when files are hard links to each
620 * other. However these values have very little meaning outside of that. This
621 * database is used to solve one of two different problems.
623 * 1) when files are appended to an archive, while the new files may have hard
624 * links to each other, you cannot determine if they have hard links to any
625 * file already stored on the archive from a prior run of pax. We must assume
626 * that these inode/device pairs are unique only within a SINGLE run of pax
627 * (which adds a set of files to an archive). So we have to make sure the
628 * inode/dev pairs we add each time are always unique. We do this by observing
629 * while the inode field is very dense, the use of the dev field is fairly
630 * sparse. Within each run of pax, we remap any device number of a new archive
631 * member that has a device number used in a prior run and already stored in a
632 * file on the archive. During the read phase of the append, we store the
633 * device numbers used and mark them to not be used by any file during the
634 * write phase. If during write we go to use one of those old device numbers,
635 * we remap it to a new value.
637 * 2) Often the fields in the archive header used to store these values are
638 * too small to store the entire value. The result is an inode or device value
639 * which can be truncated. This really can foul up an archive. With truncation
640 * we end up creating links between files that are really not links (after
641 * truncation the inodes are the same value). We address that by detecting
642 * truncation and forcing a remap of the device field to split truncated
643 * inodes away from each other. Each truncation creates a pattern of bits that
644 * are removed. We use this pattern of truncated bits to partition the inodes
645 * on a single device to many different devices (each one represented by the
646 * truncated bit pattern). All inodes on the same device that have the same
647 * truncation pattern are mapped to the same new device. Two inodes that
648 * truncate to the same value clearly will always have different truncation
649 * bit patterns, so they will be split from away each other. When we spot
650 * device truncation we remap the device number to a non truncated value.
651 * (for more info see table.h for the data structures involved).
656 * create the device mapping table
658 * 0 if successful, -1 otherwise
666 if ((dtab = (DEVT **)calloc(D_TAB_SZ, sizeof(DEVT *))) == NULL) {
667 paxwarn(1, "Cannot allocate memory for device mapping table");
675 * add a device number to the table. this will force the device to be
676 * remapped to a new value if it be used during a write phase. This
677 * function is called during the read phase of an append to prohibit the
678 * use of any device number already in the archive.
680 * 0 if added ok, -1 otherwise
686 if (chk_dev(arcn->sb.st_dev, 1) == NULL)
693 * check for a device value in the device table. If not found and the add
694 * flag is set, it is added. This does NOT assign any mapping values, just
695 * adds the device number as one that need to be remapped. If this device
696 * is already mapped, just return with a pointer to that entry.
698 * pointer to the entry for this device in the device map table. Null
699 * if the add flag is not set and the device is not in the table (it is
700 * not been seen yet). If add is set and the device cannot be added, null
701 * is returned (indicates an error).
705 chk_dev(dev_t dev, int add)
713 * look to see if this device is already in the table
715 indx = ((unsigned)dev) % D_TAB_SZ;
716 if ((pt = dtab[indx]) != NULL) {
717 while ((pt != NULL) && (pt->dev != dev))
721 * found it, return a pointer to it
728 * not in table, we add it only if told to as this may just be a check
729 * to see if a device number is being used.
735 * allocate a node for this device and add it to the front of the hash
736 * chain. Note we do not assign remaps values here, so the pt->list
739 if ((pt = (DEVT *)malloc(sizeof(DEVT))) == NULL) {
740 paxwarn(1, "Device map table out of memory");
745 pt->fow = dtab[indx];
751 * given an inode and device storage mask (the mask has a 1 for each bit
752 * the archive format is able to store in a header), we check for inode
753 * and device truncation and remap the device as required. Device mapping
754 * can also occur when during the read phase of append a device number was
755 * seen (and was marked as do not use during the write phase). WE ASSUME
756 * that unsigned longs are the same size or bigger than the fields used
757 * for ino_t and dev_t. If not the types will have to be changed.
759 * 0 if all ok, -1 otherwise.
763 map_dev(ARCHD *arcn, u_long dev_mask, u_long ino_mask)
767 static dev_t lastdev = 0; /* next device number to try */
770 ino_t trunc_bits = 0;
776 * check for device and inode truncation, and extract the truncated
779 if ((arcn->sb.st_dev & (dev_t)dev_mask) != arcn->sb.st_dev)
781 if ((nino = arcn->sb.st_ino & (ino_t)ino_mask) != arcn->sb.st_ino) {
783 trunc_bits = arcn->sb.st_ino & (ino_t)(~ino_mask);
787 * see if this device is already being mapped, look up the device
788 * then find the truncation bit pattern which applies
790 if ((pt = chk_dev(arcn->sb.st_dev, 0)) != NULL) {
792 * this device is already marked to be remapped
794 for (dpt = pt->list; dpt != NULL; dpt = dpt->fow)
795 if (dpt->trunc_bits == trunc_bits)
800 * we are being remapped for this device and pattern
801 * change the device number to be stored and return
803 arcn->sb.st_dev = dpt->dev;
804 arcn->sb.st_ino = nino;
809 * this device is not being remapped YET. if we do not have any
810 * form of truncation, we do not need a remap
812 if (!trc_ino && !trc_dev)
816 * we have truncation, have to add this as a device to remap
818 if ((pt = chk_dev(arcn->sb.st_dev, 1)) == NULL)
822 * if we just have a truncated inode, we have to make sure that
823 * all future inodes that do not truncate (they have the
824 * truncation pattern of all 0's) continue to map to the same
825 * device number. We probably have already written inodes with
826 * this device number to the archive with the truncation
827 * pattern of all 0's. So we add the mapping for all 0's to the
828 * same device number.
830 if (!trc_dev && (trunc_bits != 0)) {
831 if ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL)
834 dpt->dev = arcn->sb.st_dev;
841 * look for a device number not being used. We must watch for wrap
842 * around on lastdev (so we do not get stuck looking forever!)
844 while (++lastdev > 0) {
845 if (chk_dev(lastdev, 0) != NULL)
848 * found an unused value. If we have reached truncation point
849 * for this format we are hosed, so we give up. Otherwise we
850 * mark it as being used.
852 if (((lastdev & ((dev_t)dev_mask)) != lastdev) ||
853 (chk_dev(lastdev, 1) == NULL))
858 if ((lastdev <= 0) || ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL))
862 * got a new device number, store it under this truncation pattern.
863 * change the device number this file is being stored with.
865 dpt->trunc_bits = trunc_bits;
869 arcn->sb.st_dev = lastdev;
870 arcn->sb.st_ino = nino;
874 paxwarn(1, "Unable to fix truncated inode/device field when storing %s",
876 paxwarn(0, "Archive may create improper hard links when extracted");
881 * directory access/mod time reset table routines (for directories READ by pax)
883 * The pax -t flag requires that access times of archive files to be the same
884 * before being read by pax. For regular files, access time is restored after
885 * the file has been copied. This database provides the same functionality for
886 * directories read during file tree traversal. Restoring directory access time
887 * is more complex than files since directories may be read several times until
888 * all the descendants in their subtree are visited by fts. Directory access
889 * and modification times are stored during the fts pre-order visit (done
890 * before any descendants in the subtree is visited) and restored after the
891 * fts post-order visit (after all the descendants have been visited). In the
892 * case of premature exit from a subtree (like from the effects of -n), any
893 * directory entries left in this database are reset during final cleanup
894 * operations of pax. Entries are hashed by inode number for fast lookup.
899 * create the directory access time database for directories READ by pax.
901 * 0 is created ok, -1 otherwise.
909 if ((atab = (ATDIR **)calloc(A_TAB_SZ, sizeof(ATDIR *))) == NULL) {
910 paxwarn(1,"Cannot allocate space for directory access time table");
919 * walk through the directory access time table and reset the access time
920 * of any directory who still has an entry left in the database. These
921 * entries are for directories READ by pax
933 * for each non-empty hash table entry reset all the directories
936 for (i = 0; i < A_TAB_SZ; ++i) {
937 if ((pt = atab[i]) == NULL)
940 * remember to force the times, set_ftime() looks at pmtime
941 * and patime, which only applies to things CREATED by pax,
942 * not read by pax. Read time reset is controlled by -t.
944 for (; pt != NULL; pt = pt->fow)
945 set_ftime(pt->name, pt->mtime, pt->atime, 1);
951 * add a directory to the directory access time table. Table is hashed
952 * and chained by inode number. This is for directories READ by pax
956 add_atdir(char *fname, dev_t dev, ino_t ino, time_t mtime, time_t atime)
965 * make sure this directory is not already in the table, if so just
966 * return (the older entry always has the correct time). The only
967 * way this will happen is when the same subtree can be traversed by
968 * different args to pax and the -n option is aborting fts out of a
969 * subtree before all the post-order visits have been made).
971 indx = ((unsigned)ino) % A_TAB_SZ;
972 if ((pt = atab[indx]) != NULL) {
974 if ((pt->ino == ino) && (pt->dev == dev))
980 * oops, already there. Leave it alone.
987 * add it to the front of the hash chain
989 if ((pt = (ATDIR *)malloc(sizeof(ATDIR))) != NULL) {
990 if ((pt->name = strdup(fname)) != NULL) {
995 pt->fow = atab[indx];
1002 paxwarn(1, "Directory access time reset table ran out of memory");
1008 * look up a directory by inode and device number to obtain the access
1009 * and modification time you want to set to. If found, the modification
1010 * and access time parameters are set and the entry is removed from the
1011 * table (as it is no longer needed). These are for directories READ by
1014 * 0 if found, -1 if not found.
1018 get_atdir(dev_t dev, ino_t ino, time_t *mtime, time_t *atime)
1027 * hash by inode and search the chain for an inode and device match
1029 indx = ((unsigned)ino) % A_TAB_SZ;
1030 if ((pt = atab[indx]) == NULL)
1033 ppt = &(atab[indx]);
1034 while (pt != NULL) {
1035 if ((pt->ino == ino) && (pt->dev == dev))
1038 * no match, go to next one
1045 * return if we did not find it.
1051 * found it. return the times and remove the entry from the table.
1062 * directory access mode and time storage routines (for directories CREATED
1065 * Pax requires that extracted directories, by default, have their access/mod
1066 * times and permissions set to the values specified in the archive. During the
1067 * actions of extracting (and creating the destination subtree during -rw copy)
1068 * directories extracted may be modified after being created. Even worse is
1069 * that these directories may have been created with file permissions which
1070 * prohibits any descendants of these directories from being extracted. When
1071 * directories are created by pax, access rights may be added to permit the
1072 * creation of files in their subtree. Every time pax creates a directory, the
1073 * times and file permissions specified by the archive are stored. After all
1074 * files have been extracted (or copied), these directories have their times
1075 * and file modes reset to the stored values. The directory info is restored in
1076 * reverse order as entries were added to the data file from root to leaf. To
1077 * restore atime properly, we must go backwards. The data file consists of
1078 * records with two parts, the file name followed by a DIRDATA trailer. The
1079 * fixed sized trailer contains the size of the name plus the off_t location in
1080 * the file. To restore we work backwards through the file reading the trailer
1081 * then the file name.
1086 * set up the directory time and file mode storage for directories CREATED
1089 * 0 if ok, -1 otherwise
1100 * unlink the file so it goes away at termination by itself
1102 memcpy(tempbase, _TFILE_BASE, sizeof(_TFILE_BASE));
1103 if ((dirfd = mkstemp(tempfile)) >= 0) {
1104 (void)unlink(tempfile);
1107 paxwarn(1, "Unable to create temporary file for directory times: %s",
1114 * add the mode and times for a newly CREATED directory
1115 * name is name of the directory, psb the stat buffer with the data in it,
1116 * frc_mode is a flag that says whether to force the setting of the mode
1117 * (ignoring the user set values for preserving file mode). Frc_mode is
1118 * for the case where we created a file and found that the resulting
1119 * directory was not writeable and the user asked for file modes to NOT
1120 * be preserved. (we have to preserve what was created by default, so we
1121 * have to force the setting at the end. this is stated explicitly in the
1126 add_dir(char *name, int nlen, struct stat *psb, int frc_mode)
1134 * get current position (where file name will start) so we can store it
1137 if ((dblk.npos = lseek(dirfd, 0L, SEEK_CUR)) < 0) {
1138 paxwarn(1,"Unable to store mode and times for directory: %s",name);
1143 * write the file name followed by the trailer
1145 dblk.nlen = nlen + 1;
1146 dblk.mode = psb->st_mode & 0xffff;
1147 dblk.mtime = psb->st_mtime;
1148 dblk.atime = psb->st_atime;
1149 dblk.frc_mode = frc_mode;
1150 if ((write(dirfd, name, dblk.nlen) == dblk.nlen) &&
1151 (write(dirfd, (char *)&dblk, sizeof(dblk)) == sizeof(dblk))) {
1156 paxwarn(1,"Unable to store mode and times for created directory: %s",name);
1162 * process all file modes and times stored for directories CREATED
1169 char name[PAXPATHLEN+1];
1176 * read backwards through the file and process each directory
1178 for (cnt = 0; cnt < dircnt; ++cnt) {
1180 * read the trailer, then the file name, if this fails
1183 if (lseek(dirfd, -((off_t)sizeof(dblk)), SEEK_CUR) < 0)
1185 if (read(dirfd,(char *)&dblk, sizeof(dblk)) != sizeof(dblk))
1187 if (lseek(dirfd, dblk.npos, SEEK_SET) < 0)
1189 if (read(dirfd, name, dblk.nlen) != dblk.nlen)
1191 if (lseek(dirfd, dblk.npos, SEEK_SET) < 0)
1195 * frc_mode set, make sure we set the file modes even if
1196 * the user didn't ask for it (see file_subs.c for more info)
1198 if (pmode || dblk.frc_mode)
1199 set_pmode(name, dblk.mode);
1200 if (patime || pmtime)
1201 set_ftime(name, dblk.mtime, dblk.atime, 0);
1207 paxwarn(1,"Unable to set mode and times for created directories");
1212 * database independent routines
1217 * hashes filenames to a u_int for hashing into a table. Looks at the tail
1218 * end of file, as this provides far better distribution than any other
1219 * part of the name. For performance reasons we only care about the last
1220 * MAXKEYLEN chars (should be at LEAST large enough to pick off the file
1221 * name). Was tested on 500,000 name file tree traversal from the root
1222 * and gave almost a perfectly uniform distribution of keys when used with
1223 * prime sized tables (MAXKEYLEN was 128 in test). Hashes (sizeof int)
1224 * chars at a time and pads with 0 for last addition.
1226 * the hash value of the string MOD (%) the table size.
1230 st_hash(char *name, int len, int tabsz)
1242 * only look at the tail up to MAXKEYLEN, we do not need to waste
1243 * time here (remember these are pathnames, the tail is what will
1244 * spread out the keys)
1246 if (len > MAXKEYLEN) {
1247 pt = &(name[len - MAXKEYLEN]);
1253 * calculate the number of u_int size steps in the string and if
1254 * there is a runt to deal with
1256 steps = len/sizeof(u_int);
1257 res = len % sizeof(u_int);
1260 * add up the value of the string in unsigned integer sized pieces
1261 * too bad we cannot have unsigned int aligned strings, then we
1262 * could avoid the expensive copy.
1264 for (i = 0; i < steps; ++i) {
1265 end = pt + sizeof(u_int);
1266 dest = (char *)&val;
1273 * add in the runt padded with zero to the right
1278 dest = (char *)&val;
1285 * return the result mod the table size
1287 return(key % tabsz);