Add the DragonFly cvs id and perform general cleanups on cvs/rcs/sccs ids. Most
[dragonfly.git] / bin / pax / tables.c
CommitLineData
984263bc
MD
1/*-
2 * Copyright (c) 1992 Keith Muller.
3 * Copyright (c) 1992, 1993
4 * The Regents of the University of California. All rights reserved.
5 *
6 * This code is derived from software contributed to Berkeley by
7 * Keith Muller of the University of California, San Diego.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. All advertising materials mentioning features or use of this software
18 * must display the following acknowledgement:
19 * This product includes software developed by the University of
20 * California, Berkeley and its contributors.
21 * 4. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
24 *
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * SUCH DAMAGE.
1de703da
MD
36 *
37 * @(#)tables.c 8.1 (Berkeley) 5/31/93
38 * $FreeBSD: src/bin/pax/tables.c,v 1.13.2.1 2001/08/01 05:03:12 obrien Exp $
39 * $DragonFly: src/bin/pax/tables.c,v 1.2 2003/06/17 04:22:50 dillon Exp $
984263bc
MD
40 */
41
984263bc
MD
42#include <sys/types.h>
43#include <sys/time.h>
44#include <sys/stat.h>
45#include <sys/fcntl.h>
46#include <errno.h>
47#include <stdio.h>
48#include <stdlib.h>
49#include <string.h>
50#include <unistd.h>
51#include "pax.h"
52#include "tables.h"
53#include "extern.h"
54
55/*
56 * Routines for controlling the contents of all the different databases pax
57 * keeps. Tables are dynamically created only when they are needed. The
58 * goal was speed and the ability to work with HUGE archives. The databases
59 * were kept simple, but do have complex rules for when the contents change.
60 * As of this writing, the POSIX library functions were more complex than
61 * needed for this application (pax databases have very short lifetimes and
62 * do not survive after pax is finished). Pax is required to handle very
63 * large archives. These database routines carefully combine memory usage and
64 * temporary file storage in ways which will not significantly impact runtime
65 * performance while allowing the largest possible archives to be handled.
66 * Trying to force the fit to the POSIX databases routines was not considered
67 * time well spent.
68 */
69
70static HRDLNK **ltab = NULL; /* hard link table for detecting hard links */
71static FTM **ftab = NULL; /* file time table for updating arch */
72static NAMT **ntab = NULL; /* interactive rename storage table */
73static DEVT **dtab = NULL; /* device/inode mapping tables */
74static ATDIR **atab = NULL; /* file tree directory time reset table */
75static int dirfd = -1; /* storage for setting created dir time/mode */
76static u_long dircnt; /* entries in dir time/mode storage */
77static int ffd = -1; /* tmp file for file time table name storage */
78
79static DEVT *chk_dev __P((dev_t, int));
80
81/*
82 * hard link table routines
83 *
84 * The hard link table tries to detect hard links to files using the device and
85 * inode values. We do this when writing an archive, so we can tell the format
86 * write routine that this file is a hard link to another file. The format
87 * write routine then can store this file in whatever way it wants (as a hard
88 * link if the format supports that like tar, or ignore this info like cpio).
89 * (Actually a field in the format driver table tells us if the format wants
90 * hard link info. if not, we do not waste time looking for them). We also use
91 * the same table when reading an archive. In that situation, this table is
92 * used by the format read routine to detect hard links from stored dev and
93 * inode numbers (like cpio). This will allow pax to create a link when one
94 * can be detected by the archive format.
95 */
96
97/*
98 * lnk_start
99 * Creates the hard link table.
100 * Return:
101 * 0 if created, -1 if failure
102 */
103
104#ifdef __STDC__
105int
106lnk_start(void)
107#else
108int
109lnk_start()
110#endif
111{
112 if (ltab != NULL)
113 return(0);
114 if ((ltab = (HRDLNK **)calloc(L_TAB_SZ, sizeof(HRDLNK *))) == NULL) {
115 paxwarn(1, "Cannot allocate memory for hard link table");
116 return(-1);
117 }
118 return(0);
119}
120
121/*
122 * chk_lnk()
123 * Looks up entry in hard link hash table. If found, it copies the name
124 * of the file it is linked to (we already saw that file) into ln_name.
125 * lnkcnt is decremented and if goes to 1 the node is deleted from the
126 * database. (We have seen all the links to this file). If not found,
127 * we add the file to the database if it has the potential for having
128 * hard links to other files we may process (it has a link count > 1)
129 * Return:
130 * if found returns 1; if not found returns 0; -1 on error
131 */
132
133#ifdef __STDC__
134int
135chk_lnk(register ARCHD *arcn)
136#else
137int
138chk_lnk(arcn)
139 register ARCHD *arcn;
140#endif
141{
142 register HRDLNK *pt;
143 register HRDLNK **ppt;
144 register u_int indx;
145
146 if (ltab == NULL)
147 return(-1);
148 /*
149 * ignore those nodes that cannot have hard links
150 */
151 if ((arcn->type == PAX_DIR) || (arcn->sb.st_nlink <= 1))
152 return(0);
153
154 /*
155 * hash inode number and look for this file
156 */
157 indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ;
158 if ((pt = ltab[indx]) != NULL) {
159 /*
160 * it's hash chain in not empty, walk down looking for it
161 */
162 ppt = &(ltab[indx]);
163 while (pt != NULL) {
164 if ((pt->ino == arcn->sb.st_ino) &&
165 (pt->dev == arcn->sb.st_dev))
166 break;
167 ppt = &(pt->fow);
168 pt = pt->fow;
169 }
170
171 if (pt != NULL) {
172 /*
173 * found a link. set the node type and copy in the
174 * name of the file it is to link to. we need to
175 * handle hardlinks to regular files differently than
176 * other links.
177 */
178 arcn->ln_nlen = l_strncpy(arcn->ln_name, pt->name,
179 sizeof(arcn->ln_name) - 1);
180 arcn->ln_name[arcn->ln_nlen] = '\0';
181 if (arcn->type == PAX_REG)
182 arcn->type = PAX_HRG;
183 else
184 arcn->type = PAX_HLK;
185
186 /*
187 * if we have found all the links to this file, remove
188 * it from the database
189 */
190 if (--pt->nlink <= 1) {
191 *ppt = pt->fow;
192 (void)free((char *)pt->name);
193 (void)free((char *)pt);
194 }
195 return(1);
196 }
197 }
198
199 /*
200 * we never saw this file before. It has links so we add it to the
201 * front of this hash chain
202 */
203 if ((pt = (HRDLNK *)malloc(sizeof(HRDLNK))) != NULL) {
204 if ((pt->name = strdup(arcn->name)) != NULL) {
205 pt->dev = arcn->sb.st_dev;
206 pt->ino = arcn->sb.st_ino;
207 pt->nlink = arcn->sb.st_nlink;
208 pt->fow = ltab[indx];
209 ltab[indx] = pt;
210 return(0);
211 }
212 (void)free((char *)pt);
213 }
214
215 paxwarn(1, "Hard link table out of memory");
216 return(-1);
217}
218
219/*
220 * purg_lnk
221 * remove reference for a file that we may have added to the data base as
222 * a potential source for hard links. We ended up not using the file, so
223 * we do not want to accidently point another file at it later on.
224 */
225
226#ifdef __STDC__
227void
228purg_lnk(register ARCHD *arcn)
229#else
230void
231purg_lnk(arcn)
232 register ARCHD *arcn;
233#endif
234{
235 register HRDLNK *pt;
236 register HRDLNK **ppt;
237 register u_int indx;
238
239 if (ltab == NULL)
240 return;
241 /*
242 * do not bother to look if it could not be in the database
243 */
244 if ((arcn->sb.st_nlink <= 1) || (arcn->type == PAX_DIR) ||
245 (arcn->type == PAX_HLK) || (arcn->type == PAX_HRG))
246 return;
247
248 /*
249 * find the hash chain for this inode value, if empty return
250 */
251 indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ;
252 if ((pt = ltab[indx]) == NULL)
253 return;
254
255 /*
256 * walk down the list looking for the inode/dev pair, unlink and
257 * free if found
258 */
259 ppt = &(ltab[indx]);
260 while (pt != NULL) {
261 if ((pt->ino == arcn->sb.st_ino) &&
262 (pt->dev == arcn->sb.st_dev))
263 break;
264 ppt = &(pt->fow);
265 pt = pt->fow;
266 }
267 if (pt == NULL)
268 return;
269
270 /*
271 * remove and free it
272 */
273 *ppt = pt->fow;
274 (void)free((char *)pt->name);
275 (void)free((char *)pt);
276}
277
278/*
279 * lnk_end()
280 * pull apart a existing link table so we can reuse it. We do this between
281 * read and write phases of append with update. (The format may have
282 * used the link table, and we need to start with a fresh table for the
283 * write phase
284 */
285
286#ifdef __STDC__
287void
288lnk_end(void)
289#else
290void
291lnk_end()
292#endif
293{
294 register int i;
295 register HRDLNK *pt;
296 register HRDLNK *ppt;
297
298 if (ltab == NULL)
299 return;
300
301 for (i = 0; i < L_TAB_SZ; ++i) {
302 if (ltab[i] == NULL)
303 continue;
304 pt = ltab[i];
305 ltab[i] = NULL;
306
307 /*
308 * free up each entry on this chain
309 */
310 while (pt != NULL) {
311 ppt = pt;
312 pt = ppt->fow;
313 (void)free((char *)ppt->name);
314 (void)free((char *)ppt);
315 }
316 }
317 return;
318}
319
320/*
321 * modification time table routines
322 *
323 * The modification time table keeps track of last modification times for all
324 * files stored in an archive during a write phase when -u is set. We only
325 * add a file to the archive if it is newer than a file with the same name
326 * already stored on the archive (if there is no other file with the same
327 * name on the archive it is added). This applies to writes and appends.
328 * An append with an -u must read the archive and store the modification time
329 * for every file on that archive before starting the write phase. It is clear
330 * that this is one HUGE database. To save memory space, the actual file names
331 * are stored in a scatch file and indexed by an in memory hash table. The
332 * hash table is indexed by hashing the file path. The nodes in the table store
333 * the length of the filename and the lseek offset within the scratch file
334 * where the actual name is stored. Since there are never any deletions to this
335 * table, fragmentation of the scratch file is never a issue. Lookups seem to
336 * not exhibit any locality at all (files in the database are rarely
337 * looked up more than once...). So caching is just a waste of memory. The
338 * only limitation is the amount of scatch file space available to store the
339 * path names.
340 */
341
342/*
343 * ftime_start()
344 * create the file time hash table and open for read/write the scratch
345 * file. (after created it is unlinked, so when we exit we leave
346 * no witnesses).
347 * Return:
348 * 0 if the table and file was created ok, -1 otherwise
349 */
350
351#ifdef __STDC__
352int
353ftime_start(void)
354#else
355int
356ftime_start()
357#endif
358{
359
360 if (ftab != NULL)
361 return(0);
362 if ((ftab = (FTM **)calloc(F_TAB_SZ, sizeof(FTM *))) == NULL) {
363 paxwarn(1, "Cannot allocate memory for file time table");
364 return(-1);
365 }
366
367 /*
368 * get random name and create temporary scratch file, unlink name
369 * so it will get removed on exit
370 */
371 memcpy(tempbase, _TFILE_BASE, sizeof(_TFILE_BASE));
372 if ((ffd = mkstemp(tempfile)) < 0) {
373 syswarn(1, errno, "Unable to create temporary file: %s",
374 tempfile);
375 return(-1);
376 }
377 (void)unlink(tempfile);
378
379 return(0);
380}
381
382/*
383 * chk_ftime()
384 * looks up entry in file time hash table. If not found, the file is
385 * added to the hash table and the file named stored in the scratch file.
386 * If a file with the same name is found, the file times are compared and
387 * the most recent file time is retained. If the new file was younger (or
388 * was not in the database) the new file is selected for storage.
389 * Return:
390 * 0 if file should be added to the archive, 1 if it should be skipped,
391 * -1 on error
392 */
393
394#ifdef __STDC__
395int
396chk_ftime(register ARCHD *arcn)
397#else
398int
399chk_ftime(arcn)
400 register ARCHD *arcn;
401#endif
402{
403 register FTM *pt;
404 register int namelen;
405 register u_int indx;
406 char ckname[PAXPATHLEN+1];
407
408 /*
409 * no info, go ahead and add to archive
410 */
411 if (ftab == NULL)
412 return(0);
413
414 /*
415 * hash the pathname and look up in table
416 */
417 namelen = arcn->nlen;
418 indx = st_hash(arcn->name, namelen, F_TAB_SZ);
419 if ((pt = ftab[indx]) != NULL) {
420 /*
421 * the hash chain is not empty, walk down looking for match
422 * only read up the path names if the lengths match, speeds
423 * up the search a lot
424 */
425 while (pt != NULL) {
426 if (pt->namelen == namelen) {
427 /*
428 * potential match, have to read the name
429 * from the scratch file.
430 */
431 if (lseek(ffd,pt->seek,SEEK_SET) != pt->seek) {
432 syswarn(1, errno,
433 "Failed ftime table seek");
434 return(-1);
435 }
436 if (read(ffd, ckname, namelen) != namelen) {
437 syswarn(1, errno,
438 "Failed ftime table read");
439 return(-1);
440 }
441
442 /*
443 * if the names match, we are done
444 */
445 if (!strncmp(ckname, arcn->name, namelen))
446 break;
447 }
448
449 /*
450 * try the next entry on the chain
451 */
452 pt = pt->fow;
453 }
454
455 if (pt != NULL) {
456 /*
457 * found the file, compare the times, save the newer
458 */
459 if (arcn->sb.st_mtime > pt->mtime) {
460 /*
461 * file is newer
462 */
463 pt->mtime = arcn->sb.st_mtime;
464 return(0);
465 }
466 /*
467 * file is older
468 */
469 return(1);
470 }
471 }
472
473 /*
474 * not in table, add it
475 */
476 if ((pt = (FTM *)malloc(sizeof(FTM))) != NULL) {
477 /*
478 * add the name at the end of the scratch file, saving the
479 * offset. add the file to the head of the hash chain
480 */
481 if ((pt->seek = lseek(ffd, (off_t)0, SEEK_END)) >= 0) {
482 if (write(ffd, arcn->name, namelen) == namelen) {
483 pt->mtime = arcn->sb.st_mtime;
484 pt->namelen = namelen;
485 pt->fow = ftab[indx];
486 ftab[indx] = pt;
487 return(0);
488 }
489 syswarn(1, errno, "Failed write to file time table");
490 } else
491 syswarn(1, errno, "Failed seek on file time table");
492 } else
493 paxwarn(1, "File time table ran out of memory");
494
495 if (pt != NULL)
496 (void)free((char *)pt);
497 return(-1);
498}
499
500/*
501 * Interactive rename table routines
502 *
503 * The interactive rename table keeps track of the new names that the user
504 * assigns to files from tty input. Since this map is unique for each file
505 * we must store it in case there is a reference to the file later in archive
506 * (a link). Otherwise we will be unable to find the file we know was
507 * extracted. The remapping of these files is stored in a memory based hash
508 * table (it is assumed since input must come from /dev/tty, it is unlikely to
509 * be a very large table).
510 */
511
512/*
513 * name_start()
514 * create the interactive rename table
515 * Return:
516 * 0 if successful, -1 otherwise
517 */
518
519#ifdef __STDC__
520int
521name_start(void)
522#else
523int
524name_start()
525#endif
526{
527 if (ntab != NULL)
528 return(0);
529 if ((ntab = (NAMT **)calloc(N_TAB_SZ, sizeof(NAMT *))) == NULL) {
530 paxwarn(1, "Cannot allocate memory for interactive rename table");
531 return(-1);
532 }
533 return(0);
534}
535
536/*
537 * add_name()
538 * add the new name to old name mapping just created by the user.
539 * If an old name mapping is found (there may be duplicate names on an
540 * archive) only the most recent is kept.
541 * Return:
542 * 0 if added, -1 otherwise
543 */
544
545#ifdef __STDC__
546int
547add_name(register char *oname, int onamelen, char *nname)
548#else
549int
550add_name(oname, onamelen, nname)
551 register char *oname;
552 int onamelen;
553 char *nname;
554#endif
555{
556 register NAMT *pt;
557 register u_int indx;
558
559 if (ntab == NULL) {
560 /*
561 * should never happen
562 */
563 paxwarn(0, "No interactive rename table, links may fail\n");
564 return(0);
565 }
566
567 /*
568 * look to see if we have already mapped this file, if so we
569 * will update it
570 */
571 indx = st_hash(oname, onamelen, N_TAB_SZ);
572 if ((pt = ntab[indx]) != NULL) {
573 /*
574 * look down the has chain for the file
575 */
576 while ((pt != NULL) && (strcmp(oname, pt->oname) != 0))
577 pt = pt->fow;
578
579 if (pt != NULL) {
580 /*
581 * found an old mapping, replace it with the new one
582 * the user just input (if it is different)
583 */
584 if (strcmp(nname, pt->nname) == 0)
585 return(0);
586
587 (void)free((char *)pt->nname);
588 if ((pt->nname = strdup(nname)) == NULL) {
589 paxwarn(1, "Cannot update rename table");
590 return(-1);
591 }
592 return(0);
593 }
594 }
595
596 /*
597 * this is a new mapping, add it to the table
598 */
599 if ((pt = (NAMT *)malloc(sizeof(NAMT))) != NULL) {
600 if ((pt->oname = strdup(oname)) != NULL) {
601 if ((pt->nname = strdup(nname)) != NULL) {
602 pt->fow = ntab[indx];
603 ntab[indx] = pt;
604 return(0);
605 }
606 (void)free((char *)pt->oname);
607 }
608 (void)free((char *)pt);
609 }
610 paxwarn(1, "Interactive rename table out of memory");
611 return(-1);
612}
613
614/*
615 * sub_name()
616 * look up a link name to see if it points at a file that has been
617 * remapped by the user. If found, the link is adjusted to contain the
618 * new name (oname is the link to name)
619 */
620
621#ifdef __STDC__
622void
623sub_name(register char *oname, int *onamelen, size_t onamesize)
624#else
625void
626sub_name(oname, onamelen, onamesize)
627 register char *oname;
628 int *onamelen;
629 size_t onamesize;
630#endif
631{
632 register NAMT *pt;
633 register u_int indx;
634
635 if (ntab == NULL)
636 return;
637 /*
638 * look the name up in the hash table
639 */
640 indx = st_hash(oname, *onamelen, N_TAB_SZ);
641 if ((pt = ntab[indx]) == NULL)
642 return;
643
644 while (pt != NULL) {
645 /*
646 * walk down the hash chain looking for a match
647 */
648 if (strcmp(oname, pt->oname) == 0) {
649 /*
650 * found it, replace it with the new name
651 * and return (we know that oname has enough space)
652 */
653 *onamelen = l_strncpy(oname, pt->nname, onamesize - 1);
654 oname[*onamelen] = '\0';
655 return;
656 }
657 pt = pt->fow;
658 }
659
660 /*
661 * no match, just return
662 */
663 return;
664}
665
666/*
667 * device/inode mapping table routines
668 * (used with formats that store device and inodes fields)
669 *
670 * device/inode mapping tables remap the device field in a archive header. The
671 * device/inode fields are used to determine when files are hard links to each
672 * other. However these values have very little meaning outside of that. This
673 * database is used to solve one of two different problems.
674 *
675 * 1) when files are appended to an archive, while the new files may have hard
676 * links to each other, you cannot determine if they have hard links to any
677 * file already stored on the archive from a prior run of pax. We must assume
678 * that these inode/device pairs are unique only within a SINGLE run of pax
679 * (which adds a set of files to an archive). So we have to make sure the
680 * inode/dev pairs we add each time are always unique. We do this by observing
681 * while the inode field is very dense, the use of the dev field is fairly
682 * sparse. Within each run of pax, we remap any device number of a new archive
683 * member that has a device number used in a prior run and already stored in a
684 * file on the archive. During the read phase of the append, we store the
685 * device numbers used and mark them to not be used by any file during the
686 * write phase. If during write we go to use one of those old device numbers,
687 * we remap it to a new value.
688 *
689 * 2) Often the fields in the archive header used to store these values are
690 * too small to store the entire value. The result is an inode or device value
691 * which can be truncated. This really can foul up an archive. With truncation
692 * we end up creating links between files that are really not links (after
693 * truncation the inodes are the same value). We address that by detecting
694 * truncation and forcing a remap of the device field to split truncated
695 * inodes away from each other. Each truncation creates a pattern of bits that
696 * are removed. We use this pattern of truncated bits to partition the inodes
697 * on a single device to many different devices (each one represented by the
698 * truncated bit pattern). All inodes on the same device that have the same
699 * truncation pattern are mapped to the same new device. Two inodes that
700 * truncate to the same value clearly will always have different truncation
701 * bit patterns, so they will be split from away each other. When we spot
702 * device truncation we remap the device number to a non truncated value.
703 * (for more info see table.h for the data structures involved).
704 */
705
706/*
707 * dev_start()
708 * create the device mapping table
709 * Return:
710 * 0 if successful, -1 otherwise
711 */
712
713#ifdef __STDC__
714int
715dev_start(void)
716#else
717int
718dev_start()
719#endif
720{
721 if (dtab != NULL)
722 return(0);
723 if ((dtab = (DEVT **)calloc(D_TAB_SZ, sizeof(DEVT *))) == NULL) {
724 paxwarn(1, "Cannot allocate memory for device mapping table");
725 return(-1);
726 }
727 return(0);
728}
729
730/*
731 * add_dev()
732 * add a device number to the table. this will force the device to be
733 * remapped to a new value if it be used during a write phase. This
734 * function is called during the read phase of an append to prohibit the
735 * use of any device number already in the archive.
736 * Return:
737 * 0 if added ok, -1 otherwise
738 */
739
740#ifdef __STDC__
741int
742add_dev(register ARCHD *arcn)
743#else
744int
745add_dev(arcn)
746 register ARCHD *arcn;
747#endif
748{
749 if (chk_dev(arcn->sb.st_dev, 1) == NULL)
750 return(-1);
751 return(0);
752}
753
754/*
755 * chk_dev()
756 * check for a device value in the device table. If not found and the add
757 * flag is set, it is added. This does NOT assign any mapping values, just
758 * adds the device number as one that need to be remapped. If this device
759 * is already mapped, just return with a pointer to that entry.
760 * Return:
761 * pointer to the entry for this device in the device map table. Null
762 * if the add flag is not set and the device is not in the table (it is
763 * not been seen yet). If add is set and the device cannot be added, null
764 * is returned (indicates an error).
765 */
766
767#ifdef __STDC__
768static DEVT *
769chk_dev(dev_t dev, int add)
770#else
771static DEVT *
772chk_dev(dev, add)
773 dev_t dev;
774 int add;
775#endif
776{
777 register DEVT *pt;
778 register u_int indx;
779
780 if (dtab == NULL)
781 return(NULL);
782 /*
783 * look to see if this device is already in the table
784 */
785 indx = ((unsigned)dev) % D_TAB_SZ;
786 if ((pt = dtab[indx]) != NULL) {
787 while ((pt != NULL) && (pt->dev != dev))
788 pt = pt->fow;
789
790 /*
791 * found it, return a pointer to it
792 */
793 if (pt != NULL)
794 return(pt);
795 }
796
797 /*
798 * not in table, we add it only if told to as this may just be a check
799 * to see if a device number is being used.
800 */
801 if (add == 0)
802 return(NULL);
803
804 /*
805 * allocate a node for this device and add it to the front of the hash
806 * chain. Note we do not assign remaps values here, so the pt->list
807 * list must be NULL.
808 */
809 if ((pt = (DEVT *)malloc(sizeof(DEVT))) == NULL) {
810 paxwarn(1, "Device map table out of memory");
811 return(NULL);
812 }
813 pt->dev = dev;
814 pt->list = NULL;
815 pt->fow = dtab[indx];
816 dtab[indx] = pt;
817 return(pt);
818}
819/*
820 * map_dev()
821 * given an inode and device storage mask (the mask has a 1 for each bit
822 * the archive format is able to store in a header), we check for inode
823 * and device truncation and remap the device as required. Device mapping
824 * can also occur when during the read phase of append a device number was
825 * seen (and was marked as do not use during the write phase). WE ASSUME
826 * that unsigned longs are the same size or bigger than the fields used
827 * for ino_t and dev_t. If not the types will have to be changed.
828 * Return:
829 * 0 if all ok, -1 otherwise.
830 */
831
832#ifdef __STDC__
833int
834map_dev(register ARCHD *arcn, u_long dev_mask, u_long ino_mask)
835#else
836int
837map_dev(arcn, dev_mask, ino_mask)
838 register ARCHD *arcn;
839 u_long dev_mask;
840 u_long ino_mask;
841#endif
842{
843 register DEVT *pt;
844 register DLIST *dpt;
845 static dev_t lastdev = 0; /* next device number to try */
846 int trc_ino = 0;
847 int trc_dev = 0;
848 ino_t trunc_bits = 0;
849 ino_t nino;
850
851 if (dtab == NULL)
852 return(0);
853 /*
854 * check for device and inode truncation, and extract the truncated
855 * bit pattern.
856 */
857 if ((arcn->sb.st_dev & (dev_t)dev_mask) != arcn->sb.st_dev)
858 ++trc_dev;
859 if ((nino = arcn->sb.st_ino & (ino_t)ino_mask) != arcn->sb.st_ino) {
860 ++trc_ino;
861 trunc_bits = arcn->sb.st_ino & (ino_t)(~ino_mask);
862 }
863
864 /*
865 * see if this device is already being mapped, look up the device
866 * then find the truncation bit pattern which applies
867 */
868 if ((pt = chk_dev(arcn->sb.st_dev, 0)) != NULL) {
869 /*
870 * this device is already marked to be remapped
871 */
872 for (dpt = pt->list; dpt != NULL; dpt = dpt->fow)
873 if (dpt->trunc_bits == trunc_bits)
874 break;
875
876 if (dpt != NULL) {
877 /*
878 * we are being remapped for this device and pattern
879 * change the device number to be stored and return
880 */
881 arcn->sb.st_dev = dpt->dev;
882 arcn->sb.st_ino = nino;
883 return(0);
884 }
885 } else {
886 /*
887 * this device is not being remapped YET. if we do not have any
888 * form of truncation, we do not need a remap
889 */
890 if (!trc_ino && !trc_dev)
891 return(0);
892
893 /*
894 * we have truncation, have to add this as a device to remap
895 */
896 if ((pt = chk_dev(arcn->sb.st_dev, 1)) == NULL)
897 goto bad;
898
899 /*
900 * if we just have a truncated inode, we have to make sure that
901 * all future inodes that do not truncate (they have the
902 * truncation pattern of all 0's) continue to map to the same
903 * device number. We probably have already written inodes with
904 * this device number to the archive with the truncation
905 * pattern of all 0's. So we add the mapping for all 0's to the
906 * same device number.
907 */
908 if (!trc_dev && (trunc_bits != 0)) {
909 if ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL)
910 goto bad;
911 dpt->trunc_bits = 0;
912 dpt->dev = arcn->sb.st_dev;
913 dpt->fow = pt->list;
914 pt->list = dpt;
915 }
916 }
917
918 /*
919 * look for a device number not being used. We must watch for wrap
920 * around on lastdev (so we do not get stuck looking forever!)
921 */
922 while (++lastdev > 0) {
923 if (chk_dev(lastdev, 0) != NULL)
924 continue;
925 /*
926 * found an unused value. If we have reached truncation point
927 * for this format we are hosed, so we give up. Otherwise we
928 * mark it as being used.
929 */
930 if (((lastdev & ((dev_t)dev_mask)) != lastdev) ||
931 (chk_dev(lastdev, 1) == NULL))
932 goto bad;
933 break;
934 }
935
936 if ((lastdev <= 0) || ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL))
937 goto bad;
938
939 /*
940 * got a new device number, store it under this truncation pattern.
941 * change the device number this file is being stored with.
942 */
943 dpt->trunc_bits = trunc_bits;
944 dpt->dev = lastdev;
945 dpt->fow = pt->list;
946 pt->list = dpt;
947 arcn->sb.st_dev = lastdev;
948 arcn->sb.st_ino = nino;
949 return(0);
950
951 bad:
952 paxwarn(1, "Unable to fix truncated inode/device field when storing %s",
953 arcn->name);
954 paxwarn(0, "Archive may create improper hard links when extracted");
955 return(0);
956}
957
958/*
959 * directory access/mod time reset table routines (for directories READ by pax)
960 *
961 * The pax -t flag requires that access times of archive files to be the same
962 * before being read by pax. For regular files, access time is restored after
963 * the file has been copied. This database provides the same functionality for
964 * directories read during file tree traversal. Restoring directory access time
965 * is more complex than files since directories may be read several times until
966 * all the descendants in their subtree are visited by fts. Directory access
967 * and modification times are stored during the fts pre-order visit (done
968 * before any descendants in the subtree is visited) and restored after the
969 * fts post-order visit (after all the descendants have been visited). In the
970 * case of premature exit from a subtree (like from the effects of -n), any
971 * directory entries left in this database are reset during final cleanup
972 * operations of pax. Entries are hashed by inode number for fast lookup.
973 */
974
975/*
976 * atdir_start()
977 * create the directory access time database for directories READ by pax.
978 * Return:
979 * 0 is created ok, -1 otherwise.
980 */
981
982#ifdef __STDC__
983int
984atdir_start(void)
985#else
986int
987atdir_start()
988#endif
989{
990 if (atab != NULL)
991 return(0);
992 if ((atab = (ATDIR **)calloc(A_TAB_SZ, sizeof(ATDIR *))) == NULL) {
993 paxwarn(1,"Cannot allocate space for directory access time table");
994 return(-1);
995 }
996 return(0);
997}
998
999
1000/*
1001 * atdir_end()
1002 * walk through the directory access time table and reset the access time
1003 * of any directory who still has an entry left in the database. These
1004 * entries are for directories READ by pax
1005 */
1006
1007#ifdef __STDC__
1008void
1009atdir_end(void)
1010#else
1011void
1012atdir_end()
1013#endif
1014{
1015 register ATDIR *pt;
1016 register int i;
1017
1018 if (atab == NULL)
1019 return;
1020 /*
1021 * for each non-empty hash table entry reset all the directories
1022 * chained there.
1023 */
1024 for (i = 0; i < A_TAB_SZ; ++i) {
1025 if ((pt = atab[i]) == NULL)
1026 continue;
1027 /*
1028 * remember to force the times, set_ftime() looks at pmtime
1029 * and patime, which only applies to things CREATED by pax,
1030 * not read by pax. Read time reset is controlled by -t.
1031 */
1032 for (; pt != NULL; pt = pt->fow)
1033 set_ftime(pt->name, pt->mtime, pt->atime, 1);
1034 }
1035}
1036
1037/*
1038 * add_atdir()
1039 * add a directory to the directory access time table. Table is hashed
1040 * and chained by inode number. This is for directories READ by pax
1041 */
1042
1043#ifdef __STDC__
1044void
1045add_atdir(char *fname, dev_t dev, ino_t ino, time_t mtime, time_t atime)
1046#else
1047void
1048add_atdir(fname, dev, ino, mtime, atime)
1049 char *fname;
1050 dev_t dev;
1051 ino_t ino;
1052 time_t mtime;
1053 time_t atime;
1054#endif
1055{
1056 register ATDIR *pt;
1057 register u_int indx;
1058
1059 if (atab == NULL)
1060 return;
1061
1062 /*
1063 * make sure this directory is not already in the table, if so just
1064 * return (the older entry always has the correct time). The only
1065 * way this will happen is when the same subtree can be traversed by
1066 * different args to pax and the -n option is aborting fts out of a
1067 * subtree before all the post-order visits have been made).
1068 */
1069 indx = ((unsigned)ino) % A_TAB_SZ;
1070 if ((pt = atab[indx]) != NULL) {
1071 while (pt != NULL) {
1072 if ((pt->ino == ino) && (pt->dev == dev))
1073 break;
1074 pt = pt->fow;
1075 }
1076
1077 /*
1078 * oops, already there. Leave it alone.
1079 */
1080 if (pt != NULL)
1081 return;
1082 }
1083
1084 /*
1085 * add it to the front of the hash chain
1086 */
1087 if ((pt = (ATDIR *)malloc(sizeof(ATDIR))) != NULL) {
1088 if ((pt->name = strdup(fname)) != NULL) {
1089 pt->dev = dev;
1090 pt->ino = ino;
1091 pt->mtime = mtime;
1092 pt->atime = atime;
1093 pt->fow = atab[indx];
1094 atab[indx] = pt;
1095 return;
1096 }
1097 (void)free((char *)pt);
1098 }
1099
1100 paxwarn(1, "Directory access time reset table ran out of memory");
1101 return;
1102}
1103
1104/*
1105 * get_atdir()
1106 * look up a directory by inode and device number to obtain the access
1107 * and modification time you want to set to. If found, the modification
1108 * and access time parameters are set and the entry is removed from the
1109 * table (as it is no longer needed). These are for directories READ by
1110 * pax
1111 * Return:
1112 * 0 if found, -1 if not found.
1113 */
1114
1115#ifdef __STDC__
1116int
1117get_atdir(dev_t dev, ino_t ino, time_t *mtime, time_t *atime)
1118#else
1119int
1120get_atdir(dev, ino, mtime, atime)
1121 dev_t dev;
1122 ino_t ino;
1123 time_t *mtime;
1124 time_t *atime;
1125#endif
1126{
1127 register ATDIR *pt;
1128 register ATDIR **ppt;
1129 register u_int indx;
1130
1131 if (atab == NULL)
1132 return(-1);
1133 /*
1134 * hash by inode and search the chain for an inode and device match
1135 */
1136 indx = ((unsigned)ino) % A_TAB_SZ;
1137 if ((pt = atab[indx]) == NULL)
1138 return(-1);
1139
1140 ppt = &(atab[indx]);
1141 while (pt != NULL) {
1142 if ((pt->ino == ino) && (pt->dev == dev))
1143 break;
1144 /*
1145 * no match, go to next one
1146 */
1147 ppt = &(pt->fow);
1148 pt = pt->fow;
1149 }
1150
1151 /*
1152 * return if we did not find it.
1153 */
1154 if (pt == NULL)
1155 return(-1);
1156
1157 /*
1158 * found it. return the times and remove the entry from the table.
1159 */
1160 *ppt = pt->fow;
1161 *mtime = pt->mtime;
1162 *atime = pt->atime;
1163 (void)free((char *)pt->name);
1164 (void)free((char *)pt);
1165 return(0);
1166}
1167
1168/*
1169 * directory access mode and time storage routines (for directories CREATED
1170 * by pax).
1171 *
1172 * Pax requires that extracted directories, by default, have their access/mod
1173 * times and permissions set to the values specified in the archive. During the
1174 * actions of extracting (and creating the destination subtree during -rw copy)
1175 * directories extracted may be modified after being created. Even worse is
1176 * that these directories may have been created with file permissions which
1177 * prohibits any descendants of these directories from being extracted. When
1178 * directories are created by pax, access rights may be added to permit the
1179 * creation of files in their subtree. Every time pax creates a directory, the
1180 * times and file permissions specified by the archive are stored. After all
1181 * files have been extracted (or copied), these directories have their times
1182 * and file modes reset to the stored values. The directory info is restored in
1183 * reverse order as entries were added to the data file from root to leaf. To
1184 * restore atime properly, we must go backwards. The data file consists of
1185 * records with two parts, the file name followed by a DIRDATA trailer. The
1186 * fixed sized trailer contains the size of the name plus the off_t location in
1187 * the file. To restore we work backwards through the file reading the trailer
1188 * then the file name.
1189 */
1190
1191/*
1192 * dir_start()
1193 * set up the directory time and file mode storage for directories CREATED
1194 * by pax.
1195 * Return:
1196 * 0 if ok, -1 otherwise
1197 */
1198
1199#ifdef __STDC__
1200int
1201dir_start(void)
1202#else
1203int
1204dir_start()
1205#endif
1206{
1207
1208 if (dirfd != -1)
1209 return(0);
1210
1211 /*
1212 * unlink the file so it goes away at termination by itself
1213 */
1214 memcpy(tempbase, _TFILE_BASE, sizeof(_TFILE_BASE));
1215 if ((dirfd = mkstemp(tempfile)) >= 0) {
1216 (void)unlink(tempfile);
1217 return(0);
1218 }
1219 paxwarn(1, "Unable to create temporary file for directory times: %s",
1220 tempfile);
1221 return(-1);
1222}
1223
1224/*
1225 * add_dir()
1226 * add the mode and times for a newly CREATED directory
1227 * name is name of the directory, psb the stat buffer with the data in it,
1228 * frc_mode is a flag that says whether to force the setting of the mode
1229 * (ignoring the user set values for preserving file mode). Frc_mode is
1230 * for the case where we created a file and found that the resulting
1231 * directory was not writeable and the user asked for file modes to NOT
1232 * be preserved. (we have to preserve what was created by default, so we
1233 * have to force the setting at the end. this is stated explicitly in the
1234 * pax spec)
1235 */
1236
1237#ifdef __STDC__
1238void
1239add_dir(char *name, int nlen, struct stat *psb, int frc_mode)
1240#else
1241void
1242add_dir(name, nlen, psb, frc_mode)
1243 char *name;
1244 int nlen;
1245 struct stat *psb;
1246 int frc_mode;
1247#endif
1248{
1249 DIRDATA dblk;
1250
1251 if (dirfd < 0)
1252 return;
1253
1254 /*
1255 * get current position (where file name will start) so we can store it
1256 * in the trailer
1257 */
1258 if ((dblk.npos = lseek(dirfd, 0L, SEEK_CUR)) < 0) {
1259 paxwarn(1,"Unable to store mode and times for directory: %s",name);
1260 return;
1261 }
1262
1263 /*
1264 * write the file name followed by the trailer
1265 */
1266 dblk.nlen = nlen + 1;
1267 dblk.mode = psb->st_mode & 0xffff;
1268 dblk.mtime = psb->st_mtime;
1269 dblk.atime = psb->st_atime;
1270 dblk.frc_mode = frc_mode;
1271 if ((write(dirfd, name, dblk.nlen) == dblk.nlen) &&
1272 (write(dirfd, (char *)&dblk, sizeof(dblk)) == sizeof(dblk))) {
1273 ++dircnt;
1274 return;
1275 }
1276
1277 paxwarn(1,"Unable to store mode and times for created directory: %s",name);
1278 return;
1279}
1280
1281/*
1282 * proc_dir()
1283 * process all file modes and times stored for directories CREATED
1284 * by pax
1285 */
1286
1287#ifdef __STDC__
1288void
1289proc_dir(void)
1290#else
1291void
1292proc_dir()
1293#endif
1294{
1295 char name[PAXPATHLEN+1];
1296 DIRDATA dblk;
1297 u_long cnt;
1298
1299 if (dirfd < 0)
1300 return;
1301 /*
1302 * read backwards through the file and process each directory
1303 */
1304 for (cnt = 0; cnt < dircnt; ++cnt) {
1305 /*
1306 * read the trailer, then the file name, if this fails
1307 * just give up.
1308 */
1309 if (lseek(dirfd, -((off_t)sizeof(dblk)), SEEK_CUR) < 0)
1310 break;
1311 if (read(dirfd,(char *)&dblk, sizeof(dblk)) != sizeof(dblk))
1312 break;
1313 if (lseek(dirfd, dblk.npos, SEEK_SET) < 0)
1314 break;
1315 if (read(dirfd, name, dblk.nlen) != dblk.nlen)
1316 break;
1317 if (lseek(dirfd, dblk.npos, SEEK_SET) < 0)
1318 break;
1319
1320 /*
1321 * frc_mode set, make sure we set the file modes even if
1322 * the user didn't ask for it (see file_subs.c for more info)
1323 */
1324 if (pmode || dblk.frc_mode)
1325 set_pmode(name, dblk.mode);
1326 if (patime || pmtime)
1327 set_ftime(name, dblk.mtime, dblk.atime, 0);
1328 }
1329
1330 (void)close(dirfd);
1331 dirfd = -1;
1332 if (cnt != dircnt)
1333 paxwarn(1,"Unable to set mode and times for created directories");
1334 return;
1335}
1336
1337/*
1338 * database independent routines
1339 */
1340
1341/*
1342 * st_hash()
1343 * hashes filenames to a u_int for hashing into a table. Looks at the tail
1344 * end of file, as this provides far better distribution than any other
1345 * part of the name. For performance reasons we only care about the last
1346 * MAXKEYLEN chars (should be at LEAST large enough to pick off the file
1347 * name). Was tested on 500,000 name file tree traversal from the root
1348 * and gave almost a perfectly uniform distribution of keys when used with
1349 * prime sized tables (MAXKEYLEN was 128 in test). Hashes (sizeof int)
1350 * chars at a time and pads with 0 for last addition.
1351 * Return:
1352 * the hash value of the string MOD (%) the table size.
1353 */
1354
1355#ifdef __STDC__
1356u_int
1357st_hash(char *name, int len, int tabsz)
1358#else
1359u_int
1360st_hash(name, len, tabsz)
1361 char *name;
1362 int len;
1363 int tabsz;
1364#endif
1365{
1366 register char *pt;
1367 register char *dest;
1368 register char *end;
1369 register int i;
1370 register u_int key = 0;
1371 register int steps;
1372 register int res;
1373 u_int val;
1374
1375 /*
1376 * only look at the tail up to MAXKEYLEN, we do not need to waste
1377 * time here (remember these are pathnames, the tail is what will
1378 * spread out the keys)
1379 */
1380 if (len > MAXKEYLEN) {
1381 pt = &(name[len - MAXKEYLEN]);
1382 len = MAXKEYLEN;
1383 } else
1384 pt = name;
1385
1386 /*
1387 * calculate the number of u_int size steps in the string and if
1388 * there is a runt to deal with
1389 */
1390 steps = len/sizeof(u_int);
1391 res = len % sizeof(u_int);
1392
1393 /*
1394 * add up the value of the string in unsigned integer sized pieces
1395 * too bad we cannot have unsigned int aligned strings, then we
1396 * could avoid the expensive copy.
1397 */
1398 for (i = 0; i < steps; ++i) {
1399 end = pt + sizeof(u_int);
1400 dest = (char *)&val;
1401 while (pt < end)
1402 *dest++ = *pt++;
1403 key += val;
1404 }
1405
1406 /*
1407 * add in the runt padded with zero to the right
1408 */
1409 if (res) {
1410 val = 0;
1411 end = pt + res;
1412 dest = (char *)&val;
1413 while (pt < end)
1414 *dest++ = *pt++;
1415 key += val;
1416 }
1417
1418 /*
1419 * return the result mod the table size
1420 */
1421 return(key % tabsz);
1422}