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
36 #include <sys/types.h>
39 #include <sys/fcntl.h>
50 * Routines for controlling the contents of all the different databases pax
51 * keeps. Tables are dynamically created only when they are needed. The
52 * goal was speed and the ability to work with HUGE archives. The databases
53 * were kept simple, but do have complex rules for when the contents change.
54 * As of this writing, the POSIX library functions were more complex than
55 * needed for this application (pax databases have very short lifetimes and
56 * do not survive after pax is finished). Pax is required to handle very
57 * large archives. These database routines carefully combine memory usage and
58 * temporary file storage in ways which will not significantly impact runtime
59 * performance while allowing the largest possible archives to be handled.
60 * Trying to force the fit to the POSIX database routines was not considered
64 static HRDLNK **ltab = NULL; /* hard link table for detecting hard links */
65 static FTM **ftab = NULL; /* file time table for updating arch */
66 static NAMT **ntab = NULL; /* interactive rename storage table */
67 static DEVT **dtab = NULL; /* device/inode mapping tables */
68 static ATDIR **atab = NULL; /* file tree directory time reset table */
69 static int dirfd = -1; /* storage for setting created dir time/mode */
70 static u_long dircnt; /* entries in dir time/mode storage */
71 static int ffd = -1; /* tmp file for file time table name storage */
73 static DEVT *chk_dev(dev_t, int);
76 * hard link table routines
78 * The hard link table tries to detect hard links to files using the device and
79 * inode values. We do this when writing an archive, so we can tell the format
80 * write routine that this file is a hard link to another file. The format
81 * write routine then can store this file in whatever way it wants (as a hard
82 * link if the format supports that like tar, or ignore this info like cpio).
83 * (Actually a field in the format driver table tells us if the format wants
84 * hard link info. if not, we do not waste time looking for them). We also use
85 * the same table when reading an archive. In that situation, this table is
86 * used by the format read routine to detect hard links from stored dev and
87 * inode numbers (like cpio). This will allow pax to create a link when one
88 * can be detected by the archive format.
93 * Creates the hard link table.
95 * 0 if created, -1 if failure
103 if ((ltab = (HRDLNK **)calloc(L_TAB_SZ, sizeof(HRDLNK *))) == NULL) {
104 paxwarn(1, "Cannot allocate memory for hard link table");
112 * Looks up entry in hard link hash table. If found, it copies the name
113 * of the file it is linked to (we already saw that file) into ln_name.
114 * lnkcnt is decremented and if goes to 1 the node is deleted from the
115 * database. (We have seen all the links to this file). If not found,
116 * we add the file to the database if it has the potential for having
117 * hard links to other files we may process (it has a link count > 1)
119 * if found returns 1; if not found returns 0; -1 on error
132 * ignore those nodes that cannot have hard links
134 if ((arcn->type == PAX_DIR) || (arcn->sb.st_nlink <= 1))
138 * hash inode number and look for this file
140 indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ;
141 if ((pt = ltab[indx]) != NULL) {
143 * it's hash chain in not empty, walk down looking for it
147 if ((pt->ino == arcn->sb.st_ino) &&
148 (pt->dev == arcn->sb.st_dev))
156 * found a link. set the node type and copy in the
157 * name of the file it is to link to. we need to
158 * handle hardlinks to regular files differently than
161 arcn->ln_nlen = l_strncpy(arcn->ln_name, pt->name,
162 sizeof(arcn->ln_name) - 1);
163 arcn->ln_name[arcn->ln_nlen] = '\0';
164 if (arcn->type == PAX_REG)
165 arcn->type = PAX_HRG;
167 arcn->type = PAX_HLK;
170 * if we have found all the links to this file, remove
171 * it from the database
173 if (--pt->nlink <= 1) {
183 * we never saw this file before. It has links so we add it to the
184 * front of this hash chain
186 if ((pt = (HRDLNK *)malloc(sizeof(HRDLNK))) != NULL) {
187 if ((pt->name = strdup(arcn->name)) != NULL) {
188 pt->dev = arcn->sb.st_dev;
189 pt->ino = arcn->sb.st_ino;
190 pt->nlink = arcn->sb.st_nlink;
191 pt->fow = ltab[indx];
198 paxwarn(1, "Hard link table out of memory");
204 * remove reference for a file that we may have added to the data base as
205 * a potential source for hard links. We ended up not using the file, so
206 * we do not want to accidentally point another file at it later on.
210 purg_lnk(ARCHD *arcn)
219 * do not bother to look if it could not be in the database
221 if ((arcn->sb.st_nlink <= 1) || (arcn->type == PAX_DIR) ||
222 (arcn->type == PAX_HLK) || (arcn->type == PAX_HRG))
226 * find the hash chain for this inode value, if empty return
228 indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ;
229 if ((pt = ltab[indx]) == NULL)
233 * walk down the list looking for the inode/dev pair, unlink and
238 if ((pt->ino == arcn->sb.st_ino) &&
239 (pt->dev == arcn->sb.st_dev))
257 * Pull apart an existing link table so we can reuse it. We do this between
258 * read and write phases of append with update. (The format may have
259 * used the link table, and we need to start with a fresh table for the
273 for (i = 0; i < L_TAB_SZ; ++i) {
280 * free up each entry on this chain
293 * modification time table routines
295 * The modification time table keeps track of last modification times for all
296 * files stored in an archive during a write phase when -u is set. We only
297 * add a file to the archive if it is newer than a file with the same name
298 * already stored on the archive (if there is no other file with the same
299 * name on the archive it is added). This applies to writes and appends.
300 * An append with an -u must read the archive and store the modification time
301 * for every file on that archive before starting the write phase. It is clear
302 * that this is one HUGE database. To save memory space, the actual file names
303 * are stored in a scratch file and indexed by an in memory hash table. The
304 * hash table is indexed by hashing the file path. The nodes in the table store
305 * the length of the filename and the lseek offset within the scratch file
306 * where the actual name is stored. Since there are never any deletions from
307 * this table, fragmentation of the scratch file is never an issue. Lookups
308 * seem to not exhibit any locality at all (files in the database are rarely
309 * looked up more than once...). So caching is just a waste of memory. The
310 * only limitation is the amount of scratch file space available to store the
316 * create the file time hash table and open for read/write the scratch
317 * file. (after created it is unlinked, so when we exit we leave
320 * 0 if the table and file was created ok, -1 otherwise
329 if ((ftab = (FTM **)calloc(F_TAB_SZ, sizeof(FTM *))) == NULL) {
330 paxwarn(1, "Cannot allocate memory for file time table");
335 * get random name and create temporary scratch file, unlink name
336 * so it will get removed on exit
338 memcpy(tempbase, _TFILE_BASE, sizeof(_TFILE_BASE));
339 if ((ffd = mkstemp(tempfile)) < 0) {
340 syswarn(1, errno, "Unable to create temporary file: %s",
344 (void)unlink(tempfile);
351 * looks up entry in file time hash table. If not found, the file is
352 * added to the hash table and the file named stored in the scratch file.
353 * If a file with the same name is found, the file times are compared and
354 * the most recent file time is retained. If the new file was younger (or
355 * was not in the database) the new file is selected for storage.
357 * 0 if file should be added to the archive, 1 if it should be skipped,
362 chk_ftime(ARCHD *arcn)
367 char ckname[PAXPATHLEN+1];
370 * no info, go ahead and add to archive
376 * hash the pathname and look up in table
378 namelen = arcn->nlen;
379 indx = st_hash(arcn->name, namelen, F_TAB_SZ);
380 if ((pt = ftab[indx]) != NULL) {
382 * the hash chain is not empty, walk down looking for match
383 * only read up the path names if the lengths match, speeds
384 * up the search a lot
387 if (pt->namelen == namelen) {
389 * potential match, have to read the name
390 * from the scratch file.
392 if (lseek(ffd,pt->seek,SEEK_SET) != pt->seek) {
394 "Failed ftime table seek");
397 if (read(ffd, ckname, namelen) != namelen) {
399 "Failed ftime table read");
404 * if the names match, we are done
406 if (!strncmp(ckname, arcn->name, namelen))
411 * try the next entry on the chain
418 * found the file, compare the times, save the newer
420 if (arcn->sb.st_mtime > pt->mtime) {
424 pt->mtime = arcn->sb.st_mtime;
435 * not in table, add it
437 if ((pt = (FTM *)malloc(sizeof(FTM))) != NULL) {
439 * add the name at the end of the scratch file, saving the
440 * offset. add the file to the head of the hash chain
442 if ((pt->seek = lseek(ffd, (off_t)0, SEEK_END)) >= 0) {
443 if (write(ffd, arcn->name, namelen) == namelen) {
444 pt->mtime = arcn->sb.st_mtime;
445 pt->namelen = namelen;
446 pt->fow = ftab[indx];
450 syswarn(1, errno, "Failed write to file time table");
452 syswarn(1, errno, "Failed seek on file time table");
454 paxwarn(1, "File time table ran out of memory");
462 * Interactive rename table routines
464 * The interactive rename table keeps track of the new names that the user
465 * assigns to files from tty input. Since this map is unique for each file
466 * we must store it in case there is a reference to the file later in archive
467 * (a link). Otherwise we will be unable to find the file we know was
468 * extracted. The remapping of these files is stored in a memory based hash
469 * table (it is assumed since input must come from /dev/tty, it is unlikely to
470 * be a very large table).
475 * create the interactive rename table
477 * 0 if successful, -1 otherwise
485 if ((ntab = (NAMT **)calloc(N_TAB_SZ, sizeof(NAMT *))) == NULL) {
486 paxwarn(1, "Cannot allocate memory for interactive rename table");
494 * add the new name to old name mapping just created by the user.
495 * If an old name mapping is found (there may be duplicate names on an
496 * archive) only the most recent is kept.
498 * 0 if added, -1 otherwise
502 add_name(char *oname, int onamelen, char *nname)
509 * should never happen
511 paxwarn(0, "No interactive rename table, links may fail\n");
516 * look to see if we have already mapped this file, if so we
519 indx = st_hash(oname, onamelen, N_TAB_SZ);
520 if ((pt = ntab[indx]) != NULL) {
522 * look down the has chain for the file
524 while ((pt != NULL) && (strcmp(oname, pt->oname) != 0))
529 * found an old mapping, replace it with the new one
530 * the user just input (if it is different)
532 if (strcmp(nname, pt->nname) == 0)
536 if ((pt->nname = strdup(nname)) == NULL) {
537 paxwarn(1, "Cannot update rename table");
545 * this is a new mapping, add it to the table
547 if ((pt = (NAMT *)malloc(sizeof(NAMT))) != NULL) {
548 if ((pt->oname = strdup(oname)) != NULL) {
549 if ((pt->nname = strdup(nname)) != NULL) {
550 pt->fow = ntab[indx];
558 paxwarn(1, "Interactive rename table out of memory");
564 * look up a link name to see if it points at a file that has been
565 * remapped by the user. If found, the link is adjusted to contain the
566 * new name (oname is the link to name)
570 sub_name(char *oname, int *onamelen, size_t onamesize)
578 * look the name up in the hash table
580 indx = st_hash(oname, *onamelen, N_TAB_SZ);
581 if ((pt = ntab[indx]) == NULL)
586 * walk down the hash chain looking for a match
588 if (strcmp(oname, pt->oname) == 0) {
590 * found it, replace it with the new name
591 * and return (we know that oname has enough space)
593 *onamelen = l_strncpy(oname, pt->nname, onamesize - 1);
594 oname[*onamelen] = '\0';
601 * no match, just return
607 * device/inode mapping table routines
608 * (used with formats that store device and inodes fields)
610 * device/inode mapping tables remap the device field in an archive header. The
611 * device/inode fields are used to determine when files are hard links to each
612 * other. However these values have very little meaning outside of that. This
613 * database is used to solve one of two different problems.
615 * 1) when files are appended to an archive, while the new files may have hard
616 * links to each other, you cannot determine if they have hard links to any
617 * file already stored on the archive from a prior run of pax. We must assume
618 * that these inode/device pairs are unique only within a SINGLE run of pax
619 * (which adds a set of files to an archive). So we have to make sure the
620 * inode/dev pairs we add each time are always unique. We do this by observing
621 * while the inode field is very dense, the use of the dev field is fairly
622 * sparse. Within each run of pax, we remap any device number of a new archive
623 * member that has a device number used in a prior run and already stored in a
624 * file on the archive. During the read phase of the append, we store the
625 * device numbers used and mark them to not be used by any file during the
626 * write phase. If during write we go to use one of those old device numbers,
627 * we remap it to a new value.
629 * 2) Often the fields in the archive header used to store these values are
630 * too small to store the entire value. The result is an inode or device value
631 * which can be truncated. This really can foul up an archive. With truncation
632 * we end up creating links between files that are really not links (after
633 * truncation the inodes are the same value). We address that by detecting
634 * truncation and forcing a remap of the device field to split truncated
635 * inodes away from each other. Each truncation creates a pattern of bits that
636 * are removed. We use this pattern of truncated bits to partition the inodes
637 * on a single device to many different devices (each one represented by the
638 * truncated bit pattern). All inodes on the same device that have the same
639 * truncation pattern are mapped to the same new device. Two inodes that
640 * truncate to the same value clearly will always have different truncation
641 * bit patterns, so they will be split from away each other. When we spot
642 * device truncation we remap the device number to a non truncated value.
643 * (for more info see table.h for the data structures involved).
648 * create the device mapping table
650 * 0 if successful, -1 otherwise
658 if ((dtab = (DEVT **)calloc(D_TAB_SZ, sizeof(DEVT *))) == NULL) {
659 paxwarn(1, "Cannot allocate memory for device mapping table");
667 * add a device number to the table. this will force the device to be
668 * remapped to a new value if it be used during a write phase. This
669 * function is called during the read phase of an append to prohibit the
670 * use of any device number already in the archive.
672 * 0 if added ok, -1 otherwise
678 if (chk_dev(arcn->sb.st_dev, 1) == NULL)
685 * check for a device value in the device table. If not found and the add
686 * flag is set, it is added. This does NOT assign any mapping values, just
687 * adds the device number as one that need to be remapped. If this device
688 * is already mapped, just return with a pointer to that entry.
690 * pointer to the entry for this device in the device map table. Null
691 * if the add flag is not set and the device is not in the table (it is
692 * not been seen yet). If add is set and the device cannot be added, null
693 * is returned (indicates an error).
697 chk_dev(dev_t dev, int add)
705 * look to see if this device is already in the table
707 indx = ((unsigned)dev) % D_TAB_SZ;
708 if ((pt = dtab[indx]) != NULL) {
709 while ((pt != NULL) && (pt->dev != dev))
713 * found it, return a pointer to it
720 * not in table, we add it only if told to as this may just be a check
721 * to see if a device number is being used.
727 * allocate a node for this device and add it to the front of the hash
728 * chain. Note we do not assign remaps values here, so the pt->list
731 if ((pt = (DEVT *)malloc(sizeof(DEVT))) == NULL) {
732 paxwarn(1, "Device map table out of memory");
737 pt->fow = dtab[indx];
743 * given an inode and device storage mask (the mask has a 1 for each bit
744 * the archive format is able to store in a header), we check for inode
745 * and device truncation and remap the device as required. Device mapping
746 * can also occur when during the read phase of append a device number was
747 * seen (and was marked as do not use during the write phase). WE ASSUME
748 * that unsigned longs are the same size or bigger than the fields used
749 * for ino_t and dev_t. If not the types will have to be changed.
751 * 0 if all ok, -1 otherwise.
755 map_dev(ARCHD *arcn, u_long dev_mask, u_long ino_mask)
759 static dev_t lastdev = 0; /* next device number to try */
762 ino_t trunc_bits = 0;
768 * check for device and inode truncation, and extract the truncated
771 if ((arcn->sb.st_dev & (dev_t)dev_mask) != arcn->sb.st_dev)
773 if ((nino = arcn->sb.st_ino & (ino_t)ino_mask) != arcn->sb.st_ino) {
775 trunc_bits = arcn->sb.st_ino & (ino_t)(~ino_mask);
779 * see if this device is already being mapped, look up the device
780 * then find the truncation bit pattern which applies
782 if ((pt = chk_dev(arcn->sb.st_dev, 0)) != NULL) {
784 * this device is already marked to be remapped
786 for (dpt = pt->list; dpt != NULL; dpt = dpt->fow)
787 if (dpt->trunc_bits == trunc_bits)
792 * we are being remapped for this device and pattern
793 * change the device number to be stored and return
795 arcn->sb.st_dev = dpt->dev;
796 arcn->sb.st_ino = nino;
801 * this device is not being remapped YET. if we do not have any
802 * form of truncation, we do not need a remap
804 if (!trc_ino && !trc_dev)
808 * we have truncation, have to add this as a device to remap
810 if ((pt = chk_dev(arcn->sb.st_dev, 1)) == NULL)
814 * if we just have a truncated inode, we have to make sure that
815 * all future inodes that do not truncate (they have the
816 * truncation pattern of all 0's) continue to map to the same
817 * device number. We probably have already written inodes with
818 * this device number to the archive with the truncation
819 * pattern of all 0's. So we add the mapping for all 0's to the
820 * same device number.
822 if (!trc_dev && (trunc_bits != 0)) {
823 if ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL)
826 dpt->dev = arcn->sb.st_dev;
833 * look for a device number not being used. We must watch for wrap
834 * around on lastdev (so we do not get stuck looking forever!)
836 while (++lastdev > 0) {
837 if (chk_dev(lastdev, 0) != NULL)
840 * found an unused value. If we have reached truncation point
841 * for this format we are hosed, so we give up. Otherwise we
842 * mark it as being used.
844 if (((lastdev & ((dev_t)dev_mask)) != lastdev) ||
845 (chk_dev(lastdev, 1) == NULL))
850 if ((lastdev <= 0) || ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL))
854 * got a new device number, store it under this truncation pattern.
855 * change the device number this file is being stored with.
857 dpt->trunc_bits = trunc_bits;
861 arcn->sb.st_dev = lastdev;
862 arcn->sb.st_ino = nino;
866 paxwarn(1, "Unable to fix truncated inode/device field when storing %s",
868 paxwarn(0, "Archive may create improper hard links when extracted");
873 * directory access/mod time reset table routines (for directories READ by pax)
875 * The pax -t flag requires that access times of archive files be the same
876 * before being read by pax. For regular files, access time is restored after
877 * the file has been copied. This database provides the same functionality for
878 * directories read during file tree traversal. Restoring directory access time
879 * is more complex than files since directories may be read several times until
880 * all the descendants in their subtree are visited by fts. Directory access
881 * and modification times are stored during the fts pre-order visit (done
882 * before any descendants in the subtree are visited) and restored after the
883 * fts post-order visit (after all the descendants have been visited). In the
884 * case of premature exit from a subtree (like from the effects of -n), any
885 * directory entries left in this database are reset during final cleanup
886 * operations of pax. Entries are hashed by inode number for fast lookup.
891 * create the directory access time database for directories READ by pax.
893 * 0 is created ok, -1 otherwise.
901 if ((atab = (ATDIR **)calloc(A_TAB_SZ, sizeof(ATDIR *))) == NULL) {
902 paxwarn(1,"Cannot allocate space for directory access time table");
911 * walk through the directory access time table and reset the access time
912 * of any directory who still has an entry left in the database. These
913 * entries are for directories READ by pax
925 * for each non-empty hash table entry reset all the directories
928 for (i = 0; i < A_TAB_SZ; ++i) {
929 if ((pt = atab[i]) == NULL)
932 * remember to force the times, set_ftime() looks at pmtime
933 * and patime, which only applies to things CREATED by pax,
934 * not read by pax. Read time reset is controlled by -t.
936 for (; pt != NULL; pt = pt->fow)
937 set_ftime(pt->name, pt->mtime, pt->atime, 1);
943 * add a directory to the directory access time table. Table is hashed
944 * and chained by inode number. This is for directories READ by pax
948 add_atdir(char *fname, dev_t dev, ino_t ino, time_t mtime, time_t atime)
957 * make sure this directory is not already in the table, if so just
958 * return (the older entry always has the correct time). The only
959 * way this will happen is when the same subtree can be traversed by
960 * different args to pax and the -n option is aborting fts out of a
961 * subtree before all the post-order visits have been made.
963 indx = ((unsigned)ino) % A_TAB_SZ;
964 if ((pt = atab[indx]) != NULL) {
966 if ((pt->ino == ino) && (pt->dev == dev))
972 * oops, already there. Leave it alone.
979 * add it to the front of the hash chain
981 if ((pt = (ATDIR *)malloc(sizeof(ATDIR))) != NULL) {
982 if ((pt->name = strdup(fname)) != NULL) {
987 pt->fow = atab[indx];
994 paxwarn(1, "Directory access time reset table ran out of memory");
1000 * look up a directory by inode and device number to obtain the access
1001 * and modification time you want to set to. If found, the modification
1002 * and access time parameters are set and the entry is removed from the
1003 * table (as it is no longer needed). These are for directories READ by
1006 * 0 if found, -1 if not found.
1010 get_atdir(dev_t dev, ino_t ino, time_t *mtime, time_t *atime)
1019 * hash by inode and search the chain for an inode and device match
1021 indx = ((unsigned)ino) % A_TAB_SZ;
1022 if ((pt = atab[indx]) == NULL)
1025 ppt = &(atab[indx]);
1026 while (pt != NULL) {
1027 if ((pt->ino == ino) && (pt->dev == dev))
1030 * no match, go to next one
1037 * return if we did not find it.
1043 * found it. return the times and remove the entry from the table.
1054 * directory access mode and time storage routines (for directories CREATED
1057 * Pax requires that extracted directories, by default, have their access/mod
1058 * times and permissions set to the values specified in the archive. During the
1059 * actions of extracting (and creating the destination subtree during -rw copy)
1060 * directories extracted may be modified after being created. Even worse is
1061 * that these directories may have been created with file permissions which
1062 * prohibits any descendants of these directories from being extracted. When
1063 * directories are created by pax, access rights may be added to permit the
1064 * creation of files in their subtree. Every time pax creates a directory, the
1065 * times and file permissions specified by the archive are stored. After all
1066 * files have been extracted (or copied), these directories have their times
1067 * and file modes reset to the stored values. The directory info is restored in
1068 * reverse order as entries were added to the data file from root to leaf. To
1069 * restore atime properly, we must go backwards. The data file consists of
1070 * records with two parts, the file name followed by a DIRDATA trailer. The
1071 * fixed sized trailer contains the size of the name plus the off_t location in
1072 * the file. To restore we work backwards through the file reading the trailer
1073 * then the file name.
1078 * set up the directory time and file mode storage for directories CREATED
1081 * 0 if ok, -1 otherwise
1092 * unlink the file so it goes away at termination by itself
1094 memcpy(tempbase, _TFILE_BASE, sizeof(_TFILE_BASE));
1095 if ((dirfd = mkstemp(tempfile)) >= 0) {
1096 (void)unlink(tempfile);
1099 paxwarn(1, "Unable to create temporary file for directory times: %s",
1106 * add the mode and times for a newly CREATED directory
1107 * name is name of the directory, psb the stat buffer with the data in it,
1108 * frc_mode is a flag that says whether to force the setting of the mode
1109 * (ignoring the user set values for preserving file mode). Frc_mode is
1110 * for the case where we created a file and found that the resulting
1111 * directory was not writeable and the user asked for file modes to NOT
1112 * be preserved. (we have to preserve what was created by default, so we
1113 * have to force the setting at the end. this is stated explicitly in the
1118 add_dir(char *name, int nlen, struct stat *psb, int frc_mode)
1126 * get current position (where file name will start) so we can store it
1129 if ((dblk.npos = lseek(dirfd, 0L, SEEK_CUR)) < 0) {
1130 paxwarn(1,"Unable to store mode and times for directory: %s",name);
1135 * write the file name followed by the trailer
1137 dblk.nlen = nlen + 1;
1138 dblk.mode = psb->st_mode & 0xffff;
1139 dblk.mtime = psb->st_mtime;
1140 dblk.atime = psb->st_atime;
1141 dblk.frc_mode = frc_mode;
1142 if ((write(dirfd, name, dblk.nlen) == dblk.nlen) &&
1143 (write(dirfd, (char *)&dblk, sizeof(dblk)) == sizeof(dblk))) {
1148 paxwarn(1,"Unable to store mode and times for created directory: %s",name);
1154 * process all file modes and times stored for directories CREATED
1161 char name[PAXPATHLEN+1];
1168 * read backwards through the file and process each directory
1170 for (cnt = 0; cnt < dircnt; ++cnt) {
1172 * read the trailer, then the file name, if this fails
1175 if (lseek(dirfd, -((off_t)sizeof(dblk)), SEEK_CUR) < 0)
1177 if (read(dirfd,(char *)&dblk, sizeof(dblk)) != sizeof(dblk))
1179 if (lseek(dirfd, dblk.npos, SEEK_SET) < 0)
1181 if (read(dirfd, name, dblk.nlen) != dblk.nlen)
1183 if (lseek(dirfd, dblk.npos, SEEK_SET) < 0)
1187 * frc_mode set, make sure we set the file modes even if
1188 * the user didn't ask for it (see file_subs.c for more info)
1190 if (pmode || dblk.frc_mode)
1191 set_pmode(name, dblk.mode);
1192 if (patime || pmtime)
1193 set_ftime(name, dblk.mtime, dblk.atime, 0);
1199 paxwarn(1,"Unable to set mode and times for created directories");
1204 * database independent routines
1209 * hashes filenames to a u_int for hashing into a table. Looks at the tail
1210 * end of file, as this provides far better distribution than any other
1211 * part of the name. For performance reasons we only care about the last
1212 * MAXKEYLEN chars (should be at LEAST large enough to pick off the file
1213 * name). Was tested on 500,000 name file tree traversal from the root
1214 * and gave almost a perfectly uniform distribution of keys when used with
1215 * prime sized tables (MAXKEYLEN was 128 in test). Hashes (sizeof int)
1216 * chars at a time and pads with 0 for last addition.
1218 * the hash value of the string MOD (%) the table size.
1222 st_hash(char *name, int len, int tabsz)
1234 * only look at the tail up to MAXKEYLEN, we do not need to waste
1235 * time here (remember these are pathnames, the tail is what will
1236 * spread out the keys)
1238 if (len > MAXKEYLEN) {
1239 pt = &(name[len - MAXKEYLEN]);
1245 * calculate the number of u_int size steps in the string and if
1246 * there is a runt to deal with
1248 steps = len/sizeof(u_int);
1249 res = len % sizeof(u_int);
1252 * add up the value of the string in unsigned integer sized pieces
1253 * too bad we cannot have unsigned int aligned strings, then we
1254 * could avoid the expensive copy.
1256 for (i = 0; i < steps; ++i) {
1257 end = pt + sizeof(u_int);
1258 dest = (char *)&val;
1265 * add in the runt padded with zero to the right
1270 dest = (char *)&val;
1277 * return the result mod the table size
1279 return(key % tabsz);