tables.c revision 1.5 1 1.5 tls /* $NetBSD: tables.c,v 1.5 1997/01/11 02:06:44 tls Exp $ */
2 1.4 cgd
3 1.1 jtc /*-
4 1.1 jtc * Copyright (c) 1992 Keith Muller.
5 1.1 jtc * Copyright (c) 1992, 1993
6 1.1 jtc * The Regents of the University of California. All rights reserved.
7 1.1 jtc *
8 1.1 jtc * This code is derived from software contributed to Berkeley by
9 1.1 jtc * Keith Muller of the University of California, San Diego.
10 1.1 jtc *
11 1.1 jtc * Redistribution and use in source and binary forms, with or without
12 1.1 jtc * modification, are permitted provided that the following conditions
13 1.1 jtc * are met:
14 1.1 jtc * 1. Redistributions of source code must retain the above copyright
15 1.1 jtc * notice, this list of conditions and the following disclaimer.
16 1.1 jtc * 2. Redistributions in binary form must reproduce the above copyright
17 1.1 jtc * notice, this list of conditions and the following disclaimer in the
18 1.1 jtc * documentation and/or other materials provided with the distribution.
19 1.1 jtc * 3. All advertising materials mentioning features or use of this software
20 1.1 jtc * must display the following acknowledgement:
21 1.1 jtc * This product includes software developed by the University of
22 1.1 jtc * California, Berkeley and its contributors.
23 1.1 jtc * 4. Neither the name of the University nor the names of its contributors
24 1.1 jtc * may be used to endorse or promote products derived from this software
25 1.1 jtc * without specific prior written permission.
26 1.1 jtc *
27 1.1 jtc * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 1.1 jtc * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 1.1 jtc * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 1.1 jtc * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 1.1 jtc * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 1.1 jtc * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 1.1 jtc * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 1.1 jtc * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 1.1 jtc * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 1.1 jtc * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 1.1 jtc * SUCH DAMAGE.
38 1.1 jtc */
39 1.1 jtc
40 1.1 jtc #ifndef lint
41 1.4 cgd #if 0
42 1.4 cgd static char sccsid[] = "@(#)tables.c 8.1 (Berkeley) 5/31/93";
43 1.4 cgd #else
44 1.5 tls static char rcsid[] = "$NetBSD: tables.c,v 1.5 1997/01/11 02:06:44 tls Exp $";
45 1.4 cgd #endif
46 1.1 jtc #endif /* not lint */
47 1.1 jtc
48 1.1 jtc #include <sys/types.h>
49 1.1 jtc #include <sys/time.h>
50 1.1 jtc #include <sys/stat.h>
51 1.1 jtc #include <sys/param.h>
52 1.1 jtc #include <sys/fcntl.h>
53 1.1 jtc #include <stdio.h>
54 1.1 jtc #include <ctype.h>
55 1.1 jtc #include <string.h>
56 1.1 jtc #include <unistd.h>
57 1.1 jtc #include <errno.h>
58 1.1 jtc #include <stdlib.h>
59 1.1 jtc #include "pax.h"
60 1.1 jtc #include "tables.h"
61 1.1 jtc #include "extern.h"
62 1.1 jtc
63 1.1 jtc /*
64 1.1 jtc * Routines for controlling the contents of all the different databases pax
65 1.1 jtc * keeps. Tables are dynamically created only when they are needed. The
66 1.1 jtc * goal was speed and the ability to work with HUGE archives. The databases
67 1.1 jtc * were kept simple, but do have complex rules for when the contents change.
68 1.1 jtc * As of this writing, the posix library functions were more complex than
69 1.1 jtc * needed for this application (pax databases have very short lifetimes and
70 1.1 jtc * do not survive after pax is finished). Pax is required to handle very
71 1.1 jtc * large archives. These database routines carefully combine memory usage and
72 1.1 jtc * temporary file storage in ways which will not significantly impact runtime
73 1.1 jtc * performance while allowing the largest possible archives to be handled.
74 1.1 jtc * Trying to force the fit to the posix databases routines was not considered
75 1.1 jtc * time well spent.
76 1.1 jtc */
77 1.1 jtc
78 1.1 jtc static HRDLNK **ltab = NULL; /* hard link table for detecting hard links */
79 1.1 jtc static FTM **ftab = NULL; /* file time table for updating arch */
80 1.1 jtc static NAMT **ntab = NULL; /* interactive rename storage table */
81 1.1 jtc static DEVT **dtab = NULL; /* device/inode mapping tables */
82 1.1 jtc static ATDIR **atab = NULL; /* file tree directory time reset table */
83 1.1 jtc static int dirfd = -1; /* storage for setting created dir time/mode */
84 1.1 jtc static u_long dircnt; /* entries in dir time/mode storage */
85 1.1 jtc static int ffd = -1; /* tmp file for file time table name storage */
86 1.1 jtc
87 1.1 jtc static DEVT *chk_dev __P((dev_t, int));
88 1.1 jtc
89 1.1 jtc /*
90 1.1 jtc * hard link table routines
91 1.1 jtc *
92 1.1 jtc * The hard link table tries to detect hard links to files using the device and
93 1.1 jtc * inode values. We do this when writing an archive, so we can tell the format
94 1.1 jtc * write routine that this file is a hard link to another file. The format
95 1.1 jtc * write routine then can store this file in whatever way it wants (as a hard
96 1.1 jtc * link if the format supports that like tar, or ignore this info like cpio).
97 1.1 jtc * (Actually a field in the format driver table tells us if the format wants
98 1.1 jtc * hard link info. if not, we do not waste time looking for them). We also use
99 1.1 jtc * the same table when reading an archive. In that situation, this table is
100 1.1 jtc * used by the format read routine to detect hard links from stored dev and
101 1.1 jtc * inode numbers (like cpio). This will allow pax to create a link when one
102 1.1 jtc * can be detected by the archive format.
103 1.1 jtc */
104 1.1 jtc
105 1.1 jtc /*
106 1.1 jtc * lnk_start
107 1.1 jtc * Creates the hard link table.
108 1.1 jtc * Return:
109 1.1 jtc * 0 if created, -1 if failure
110 1.1 jtc */
111 1.1 jtc
112 1.1 jtc #if __STDC__
113 1.1 jtc int
114 1.1 jtc lnk_start(void)
115 1.1 jtc #else
116 1.1 jtc int
117 1.1 jtc lnk_start()
118 1.1 jtc #endif
119 1.1 jtc {
120 1.1 jtc if (ltab != NULL)
121 1.1 jtc return(0);
122 1.1 jtc if ((ltab = (HRDLNK **)calloc(L_TAB_SZ, sizeof(HRDLNK *))) == NULL) {
123 1.1 jtc warn(1, "Cannot allocate memory for hard link table");
124 1.1 jtc return(-1);
125 1.1 jtc }
126 1.1 jtc return(0);
127 1.1 jtc }
128 1.1 jtc
129 1.1 jtc /*
130 1.1 jtc * chk_lnk()
131 1.1 jtc * Looks up entry in hard link hash table. If found, it copies the name
132 1.1 jtc * of the file it is linked to (we already saw that file) into ln_name.
133 1.1 jtc * lnkcnt is decremented and if goes to 1 the node is deleted from the
134 1.1 jtc * database. (We have seen all the links to this file). If not found,
135 1.1 jtc * we add the file to the database if it has the potential for having
136 1.1 jtc * hard links to other files we may process (it has a link count > 1)
137 1.1 jtc * Return:
138 1.1 jtc * if found returns 1; if not found returns 0; -1 on error
139 1.1 jtc */
140 1.1 jtc
141 1.1 jtc #if __STDC__
142 1.1 jtc int
143 1.5 tls chk_lnk(ARCHD *arcn)
144 1.1 jtc #else
145 1.1 jtc int
146 1.1 jtc chk_lnk(arcn)
147 1.5 tls ARCHD *arcn;
148 1.1 jtc #endif
149 1.1 jtc {
150 1.5 tls HRDLNK *pt;
151 1.5 tls HRDLNK **ppt;
152 1.5 tls u_int indx;
153 1.1 jtc
154 1.1 jtc if (ltab == NULL)
155 1.1 jtc return(-1);
156 1.1 jtc /*
157 1.1 jtc * ignore those nodes that cannot have hard links
158 1.1 jtc */
159 1.1 jtc if ((arcn->type == PAX_DIR) || (arcn->sb.st_nlink <= 1))
160 1.1 jtc return(0);
161 1.1 jtc
162 1.1 jtc /*
163 1.1 jtc * hash inode number and look for this file
164 1.1 jtc */
165 1.1 jtc indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ;
166 1.1 jtc if ((pt = ltab[indx]) != NULL) {
167 1.1 jtc /*
168 1.1 jtc * it's hash chain in not empty, walk down looking for it
169 1.1 jtc */
170 1.1 jtc ppt = &(ltab[indx]);
171 1.1 jtc while (pt != NULL) {
172 1.1 jtc if ((pt->ino == arcn->sb.st_ino) &&
173 1.1 jtc (pt->dev == arcn->sb.st_dev))
174 1.1 jtc break;
175 1.1 jtc ppt = &(pt->fow);
176 1.1 jtc pt = pt->fow;
177 1.1 jtc }
178 1.1 jtc
179 1.1 jtc if (pt != NULL) {
180 1.1 jtc /*
181 1.1 jtc * found a link. set the node type and copy in the
182 1.1 jtc * name of the file it is to link to. we need to
183 1.1 jtc * handle hardlinks to regular files differently than
184 1.1 jtc * other links.
185 1.1 jtc */
186 1.1 jtc arcn->ln_nlen = l_strncpy(arcn->ln_name, pt->name,
187 1.1 jtc PAXPATHLEN+1);
188 1.1 jtc if (arcn->type == PAX_REG)
189 1.1 jtc arcn->type = PAX_HRG;
190 1.1 jtc else
191 1.1 jtc arcn->type = PAX_HLK;
192 1.1 jtc
193 1.1 jtc /*
194 1.1 jtc * if we have found all the links to this file, remove
195 1.1 jtc * it from the database
196 1.1 jtc */
197 1.1 jtc if (--pt->nlink <= 1) {
198 1.1 jtc *ppt = pt->fow;
199 1.1 jtc (void)free((char *)pt->name);
200 1.1 jtc (void)free((char *)pt);
201 1.1 jtc }
202 1.1 jtc return(1);
203 1.1 jtc }
204 1.1 jtc }
205 1.1 jtc
206 1.1 jtc /*
207 1.1 jtc * we never saw this file before. It has links so we add it to the
208 1.1 jtc * front of this hash chain
209 1.1 jtc */
210 1.1 jtc if ((pt = (HRDLNK *)malloc(sizeof(HRDLNK))) != NULL) {
211 1.1 jtc if ((pt->name = strdup(arcn->name)) != NULL) {
212 1.1 jtc pt->dev = arcn->sb.st_dev;
213 1.1 jtc pt->ino = arcn->sb.st_ino;
214 1.1 jtc pt->nlink = arcn->sb.st_nlink;
215 1.1 jtc pt->fow = ltab[indx];
216 1.1 jtc ltab[indx] = pt;
217 1.1 jtc return(0);
218 1.1 jtc }
219 1.1 jtc (void)free((char *)pt);
220 1.1 jtc }
221 1.1 jtc
222 1.1 jtc warn(1, "Hard link table out of memory");
223 1.1 jtc return(-1);
224 1.1 jtc }
225 1.1 jtc
226 1.1 jtc /*
227 1.1 jtc * purg_lnk
228 1.1 jtc * remove reference for a file that we may have added to the data base as
229 1.1 jtc * a potential source for hard links. We ended up not using the file, so
230 1.1 jtc * we do not want to accidently point another file at it later on.
231 1.1 jtc */
232 1.1 jtc
233 1.1 jtc #if __STDC__
234 1.1 jtc void
235 1.5 tls purg_lnk(ARCHD *arcn)
236 1.1 jtc #else
237 1.1 jtc void
238 1.1 jtc purg_lnk(arcn)
239 1.5 tls ARCHD *arcn;
240 1.1 jtc #endif
241 1.1 jtc {
242 1.5 tls HRDLNK *pt;
243 1.5 tls HRDLNK **ppt;
244 1.5 tls u_int indx;
245 1.1 jtc
246 1.1 jtc if (ltab == NULL)
247 1.1 jtc return;
248 1.1 jtc /*
249 1.1 jtc * do not bother to look if it could not be in the database
250 1.1 jtc */
251 1.1 jtc if ((arcn->sb.st_nlink <= 1) || (arcn->type == PAX_DIR) ||
252 1.1 jtc (arcn->type == PAX_HLK) || (arcn->type == PAX_HRG))
253 1.1 jtc return;
254 1.1 jtc
255 1.1 jtc /*
256 1.1 jtc * find the hash chain for this inode value, if empty return
257 1.1 jtc */
258 1.1 jtc indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ;
259 1.1 jtc if ((pt = ltab[indx]) == NULL)
260 1.1 jtc return;
261 1.1 jtc
262 1.1 jtc /*
263 1.1 jtc * walk down the list looking for the inode/dev pair, unlink and
264 1.1 jtc * free if found
265 1.1 jtc */
266 1.1 jtc ppt = &(ltab[indx]);
267 1.1 jtc while (pt != NULL) {
268 1.1 jtc if ((pt->ino == arcn->sb.st_ino) &&
269 1.1 jtc (pt->dev == arcn->sb.st_dev))
270 1.1 jtc break;
271 1.1 jtc ppt = &(pt->fow);
272 1.1 jtc pt = pt->fow;
273 1.1 jtc }
274 1.1 jtc if (pt == NULL)
275 1.1 jtc return;
276 1.1 jtc
277 1.1 jtc /*
278 1.1 jtc * remove and free it
279 1.1 jtc */
280 1.1 jtc *ppt = pt->fow;
281 1.1 jtc (void)free((char *)pt->name);
282 1.1 jtc (void)free((char *)pt);
283 1.1 jtc }
284 1.1 jtc
285 1.1 jtc /*
286 1.1 jtc * lnk_end()
287 1.1 jtc * pull apart a existing link table so we can reuse it. We do this between
288 1.1 jtc * read and write phases of append with update. (The format may have
289 1.1 jtc * used the link table, and we need to start with a fresh table for the
290 1.1 jtc * write phase
291 1.1 jtc */
292 1.1 jtc
293 1.1 jtc #if __STDC__
294 1.1 jtc void
295 1.1 jtc lnk_end(void)
296 1.1 jtc #else
297 1.1 jtc void
298 1.1 jtc lnk_end()
299 1.1 jtc #endif
300 1.1 jtc {
301 1.5 tls int i;
302 1.5 tls HRDLNK *pt;
303 1.5 tls HRDLNK *ppt;
304 1.1 jtc
305 1.1 jtc if (ltab == NULL)
306 1.1 jtc return;
307 1.1 jtc
308 1.1 jtc for (i = 0; i < L_TAB_SZ; ++i) {
309 1.1 jtc if (ltab[i] == NULL)
310 1.1 jtc continue;
311 1.1 jtc pt = ltab[i];
312 1.1 jtc ltab[i] = NULL;
313 1.1 jtc
314 1.1 jtc /*
315 1.1 jtc * free up each entry on this chain
316 1.1 jtc */
317 1.1 jtc while (pt != NULL) {
318 1.1 jtc ppt = pt;
319 1.1 jtc pt = ppt->fow;
320 1.1 jtc (void)free((char *)ppt->name);
321 1.1 jtc (void)free((char *)ppt);
322 1.1 jtc }
323 1.1 jtc }
324 1.1 jtc return;
325 1.1 jtc }
326 1.1 jtc
327 1.1 jtc /*
328 1.1 jtc * modification time table routines
329 1.1 jtc *
330 1.1 jtc * The modification time table keeps track of last modification times for all
331 1.1 jtc * files stored in an archive during a write phase when -u is set. We only
332 1.1 jtc * add a file to the archive if it is newer than a file with the same name
333 1.1 jtc * already stored on the archive (if there is no other file with the same
334 1.1 jtc * name on the archive it is added). This applies to writes and appends.
335 1.1 jtc * An append with an -u must read the archive and store the modification time
336 1.1 jtc * for every file on that archive before starting the write phase. It is clear
337 1.1 jtc * that this is one HUGE database. To save memory space, the actual file names
338 1.1 jtc * are stored in a scatch file and indexed by an in memory hash table. The
339 1.1 jtc * hash table is indexed by hashing the file path. The nodes in the table store
340 1.1 jtc * the length of the filename and the lseek offset within the scratch file
341 1.1 jtc * where the actual name is stored. Since there are never any deletions to this
342 1.1 jtc * table, fragmentation of the scratch file is never a issue. Lookups seem to
343 1.1 jtc * not exhibit any locality at all (files in the database are rarely
344 1.1 jtc * looked up more than once...). So caching is just a waste of memory. The
345 1.1 jtc * only limitation is the amount of scatch file space available to store the
346 1.1 jtc * path names.
347 1.1 jtc */
348 1.1 jtc
349 1.1 jtc /*
350 1.1 jtc * ftime_start()
351 1.1 jtc * create the file time hash table and open for read/write the scratch
352 1.1 jtc * file. (after created it is unlinked, so when we exit we leave
353 1.1 jtc * no witnesses).
354 1.1 jtc * Return:
355 1.1 jtc * 0 if the table and file was created ok, -1 otherwise
356 1.1 jtc */
357 1.1 jtc
358 1.1 jtc #if __STDC__
359 1.1 jtc int
360 1.1 jtc ftime_start(void)
361 1.1 jtc #else
362 1.1 jtc int
363 1.1 jtc ftime_start()
364 1.1 jtc #endif
365 1.1 jtc {
366 1.1 jtc char *pt;
367 1.1 jtc
368 1.1 jtc if (ftab != NULL)
369 1.1 jtc return(0);
370 1.1 jtc if ((ftab = (FTM **)calloc(F_TAB_SZ, sizeof(FTM *))) == NULL) {
371 1.1 jtc warn(1, "Cannot allocate memory for file time table");
372 1.1 jtc return(-1);
373 1.1 jtc }
374 1.1 jtc
375 1.1 jtc /*
376 1.1 jtc * get random name and create temporary scratch file, unlink name
377 1.1 jtc * so it will get removed on exit
378 1.1 jtc */
379 1.1 jtc if ((pt = tempnam((char *)NULL, (char *)NULL)) == NULL)
380 1.1 jtc return(-1);
381 1.1 jtc (void)unlink(pt);
382 1.1 jtc
383 1.1 jtc if ((ffd = open(pt, O_RDWR | O_CREAT, S_IRWXU)) < 0) {
384 1.1 jtc syswarn(1, errno, "Unable to open temporary file: %s", pt);
385 1.1 jtc return(-1);
386 1.1 jtc }
387 1.1 jtc
388 1.1 jtc (void)unlink(pt);
389 1.1 jtc return(0);
390 1.1 jtc }
391 1.1 jtc
392 1.1 jtc /*
393 1.1 jtc * chk_ftime()
394 1.1 jtc * looks up entry in file time hash table. If not found, the file is
395 1.1 jtc * added to the hash table and the file named stored in the scratch file.
396 1.1 jtc * If a file with the same name is found, the file times are compared and
397 1.1 jtc * the most recent file time is retained. If the new file was younger (or
398 1.1 jtc * was not in the database) the new file is selected for storage.
399 1.1 jtc * Return:
400 1.1 jtc * 0 if file should be added to the archive, 1 if it should be skipped,
401 1.1 jtc * -1 on error
402 1.1 jtc */
403 1.1 jtc
404 1.1 jtc #if __STDC__
405 1.1 jtc int
406 1.5 tls chk_ftime(ARCHD *arcn)
407 1.1 jtc #else
408 1.1 jtc int
409 1.1 jtc chk_ftime(arcn)
410 1.5 tls ARCHD *arcn;
411 1.1 jtc #endif
412 1.1 jtc {
413 1.5 tls FTM *pt;
414 1.5 tls int namelen;
415 1.5 tls u_int indx;
416 1.1 jtc char ckname[PAXPATHLEN+1];
417 1.1 jtc
418 1.1 jtc /*
419 1.1 jtc * no info, go ahead and add to archive
420 1.1 jtc */
421 1.1 jtc if (ftab == NULL)
422 1.1 jtc return(0);
423 1.1 jtc
424 1.1 jtc /*
425 1.1 jtc * hash the pathname and look up in table
426 1.1 jtc */
427 1.1 jtc namelen = arcn->nlen;
428 1.1 jtc indx = st_hash(arcn->name, namelen, F_TAB_SZ);
429 1.1 jtc if ((pt = ftab[indx]) != NULL) {
430 1.1 jtc /*
431 1.1 jtc * the hash chain is not empty, walk down looking for match
432 1.1 jtc * only read up the path names if the lengths match, speeds
433 1.1 jtc * up the search a lot
434 1.1 jtc */
435 1.1 jtc while (pt != NULL) {
436 1.1 jtc if (pt->namelen == namelen) {
437 1.1 jtc /*
438 1.1 jtc * potential match, have to read the name
439 1.1 jtc * from the scratch file.
440 1.1 jtc */
441 1.1 jtc if (lseek(ffd,pt->seek,SEEK_SET) != pt->seek) {
442 1.1 jtc syswarn(1, errno,
443 1.1 jtc "Failed ftime table seek");
444 1.1 jtc return(-1);
445 1.1 jtc }
446 1.1 jtc if (read(ffd, ckname, namelen) != namelen) {
447 1.1 jtc syswarn(1, errno,
448 1.1 jtc "Failed ftime table read");
449 1.1 jtc return(-1);
450 1.1 jtc }
451 1.1 jtc
452 1.1 jtc /*
453 1.1 jtc * if the names match, we are done
454 1.1 jtc */
455 1.1 jtc if (!strncmp(ckname, arcn->name, namelen))
456 1.1 jtc break;
457 1.1 jtc }
458 1.1 jtc
459 1.1 jtc /*
460 1.1 jtc * try the next entry on the chain
461 1.1 jtc */
462 1.1 jtc pt = pt->fow;
463 1.1 jtc }
464 1.1 jtc
465 1.1 jtc if (pt != NULL) {
466 1.1 jtc /*
467 1.1 jtc * found the file, compare the times, save the newer
468 1.1 jtc */
469 1.1 jtc if (arcn->sb.st_mtime > pt->mtime) {
470 1.1 jtc /*
471 1.1 jtc * file is newer
472 1.1 jtc */
473 1.1 jtc pt->mtime = arcn->sb.st_mtime;
474 1.1 jtc return(0);
475 1.1 jtc }
476 1.1 jtc /*
477 1.1 jtc * file is older
478 1.1 jtc */
479 1.1 jtc return(1);
480 1.1 jtc }
481 1.1 jtc }
482 1.1 jtc
483 1.1 jtc /*
484 1.1 jtc * not in table, add it
485 1.1 jtc */
486 1.1 jtc if ((pt = (FTM *)malloc(sizeof(FTM))) != NULL) {
487 1.1 jtc /*
488 1.1 jtc * add the name at the end of the scratch file, saving the
489 1.1 jtc * offset. add the file to the head of the hash chain
490 1.1 jtc */
491 1.1 jtc if ((pt->seek = lseek(ffd, (off_t)0, SEEK_END)) >= 0) {
492 1.1 jtc if (write(ffd, arcn->name, namelen) == namelen) {
493 1.1 jtc pt->mtime = arcn->sb.st_mtime;
494 1.1 jtc pt->namelen = namelen;
495 1.1 jtc pt->fow = ftab[indx];
496 1.1 jtc ftab[indx] = pt;
497 1.1 jtc return(0);
498 1.1 jtc }
499 1.1 jtc syswarn(1, errno, "Failed write to file time table");
500 1.1 jtc } else
501 1.1 jtc syswarn(1, errno, "Failed seek on file time table");
502 1.1 jtc } else
503 1.1 jtc warn(1, "File time table ran out of memory");
504 1.1 jtc
505 1.1 jtc if (pt != NULL)
506 1.1 jtc (void)free((char *)pt);
507 1.1 jtc return(-1);
508 1.1 jtc }
509 1.1 jtc
510 1.1 jtc /*
511 1.1 jtc * Interactive rename table routines
512 1.1 jtc *
513 1.1 jtc * The interactive rename table keeps track of the new names that the user
514 1.1 jtc * assignes to files from tty input. Since this map is unique for each file
515 1.1 jtc * we must store it in case there is a reference to the file later in archive
516 1.1 jtc * (a link). Otherwise we will be unable to find the file we know was
517 1.1 jtc * extracted. The remapping of these files is stored in a memory based hash
518 1.1 jtc * table (it is assumed since input must come from /dev/tty, it is unlikely to
519 1.1 jtc * be a very large table).
520 1.1 jtc */
521 1.1 jtc
522 1.1 jtc /*
523 1.1 jtc * name_start()
524 1.1 jtc * create the interactive rename table
525 1.1 jtc * Return:
526 1.1 jtc * 0 if successful, -1 otherwise
527 1.1 jtc */
528 1.1 jtc
529 1.1 jtc #if __STDC__
530 1.1 jtc int
531 1.1 jtc name_start(void)
532 1.1 jtc #else
533 1.1 jtc int
534 1.1 jtc name_start()
535 1.1 jtc #endif
536 1.1 jtc {
537 1.1 jtc if (ntab != NULL)
538 1.1 jtc return(0);
539 1.1 jtc if ((ntab = (NAMT **)calloc(N_TAB_SZ, sizeof(NAMT *))) == NULL) {
540 1.1 jtc warn(1, "Cannot allocate memory for interactive rename table");
541 1.1 jtc return(-1);
542 1.1 jtc }
543 1.1 jtc return(0);
544 1.1 jtc }
545 1.1 jtc
546 1.1 jtc /*
547 1.1 jtc * add_name()
548 1.1 jtc * add the new name to old name mapping just created by the user.
549 1.1 jtc * If an old name mapping is found (there may be duplicate names on an
550 1.1 jtc * archive) only the most recent is kept.
551 1.1 jtc * Return:
552 1.1 jtc * 0 if added, -1 otherwise
553 1.1 jtc */
554 1.1 jtc
555 1.1 jtc #if __STDC__
556 1.1 jtc int
557 1.5 tls add_name(char *oname, int onamelen, char *nname)
558 1.1 jtc #else
559 1.1 jtc int
560 1.1 jtc add_name(oname, onamelen, nname)
561 1.5 tls char *oname;
562 1.1 jtc int onamelen;
563 1.1 jtc char *nname;
564 1.1 jtc #endif
565 1.1 jtc {
566 1.5 tls NAMT *pt;
567 1.5 tls u_int indx;
568 1.1 jtc
569 1.1 jtc if (ntab == NULL) {
570 1.1 jtc /*
571 1.1 jtc * should never happen
572 1.1 jtc */
573 1.1 jtc warn(0, "No interactive rename table, links may fail\n");
574 1.1 jtc return(0);
575 1.1 jtc }
576 1.1 jtc
577 1.1 jtc /*
578 1.1 jtc * look to see if we have already mapped this file, if so we
579 1.1 jtc * will update it
580 1.1 jtc */
581 1.1 jtc indx = st_hash(oname, onamelen, N_TAB_SZ);
582 1.1 jtc if ((pt = ntab[indx]) != NULL) {
583 1.1 jtc /*
584 1.1 jtc * look down the has chain for the file
585 1.1 jtc */
586 1.1 jtc while ((pt != NULL) && (strcmp(oname, pt->oname) != 0))
587 1.1 jtc pt = pt->fow;
588 1.1 jtc
589 1.1 jtc if (pt != NULL) {
590 1.1 jtc /*
591 1.1 jtc * found an old mapping, replace it with the new one
592 1.1 jtc * the user just input (if it is different)
593 1.1 jtc */
594 1.1 jtc if (strcmp(nname, pt->nname) == 0)
595 1.1 jtc return(0);
596 1.1 jtc
597 1.1 jtc (void)free((char *)pt->nname);
598 1.1 jtc if ((pt->nname = strdup(nname)) == NULL) {
599 1.1 jtc warn(1, "Cannot update rename table");
600 1.1 jtc return(-1);
601 1.1 jtc }
602 1.1 jtc return(0);
603 1.1 jtc }
604 1.1 jtc }
605 1.1 jtc
606 1.1 jtc /*
607 1.1 jtc * this is a new mapping, add it to the table
608 1.1 jtc */
609 1.1 jtc if ((pt = (NAMT *)malloc(sizeof(NAMT))) != NULL) {
610 1.1 jtc if ((pt->oname = strdup(oname)) != NULL) {
611 1.1 jtc if ((pt->nname = strdup(nname)) != NULL) {
612 1.1 jtc pt->fow = ntab[indx];
613 1.1 jtc ntab[indx] = pt;
614 1.1 jtc return(0);
615 1.1 jtc }
616 1.1 jtc (void)free((char *)pt->oname);
617 1.1 jtc }
618 1.1 jtc (void)free((char *)pt);
619 1.1 jtc }
620 1.1 jtc warn(1, "Interactive rename table out of memory");
621 1.1 jtc return(-1);
622 1.1 jtc }
623 1.1 jtc
624 1.1 jtc /*
625 1.1 jtc * sub_name()
626 1.1 jtc * look up a link name to see if it points at a file that has been
627 1.1 jtc * remapped by the user. If found, the link is adjusted to contain the
628 1.1 jtc * new name (oname is the link to name)
629 1.1 jtc */
630 1.1 jtc
631 1.1 jtc #if __STDC__
632 1.1 jtc void
633 1.5 tls sub_name(char *oname, int *onamelen)
634 1.1 jtc #else
635 1.1 jtc void
636 1.1 jtc sub_name(oname, onamelen)
637 1.5 tls char *oname;
638 1.1 jtc int *onamelen;
639 1.1 jtc #endif
640 1.1 jtc {
641 1.5 tls NAMT *pt;
642 1.5 tls u_int indx;
643 1.1 jtc
644 1.1 jtc if (ntab == NULL)
645 1.1 jtc return;
646 1.1 jtc /*
647 1.1 jtc * look the name up in the hash table
648 1.1 jtc */
649 1.1 jtc indx = st_hash(oname, *onamelen, N_TAB_SZ);
650 1.1 jtc if ((pt = ntab[indx]) == NULL)
651 1.1 jtc return;
652 1.1 jtc
653 1.1 jtc while (pt != NULL) {
654 1.1 jtc /*
655 1.1 jtc * walk down the hash cahin looking for a match
656 1.1 jtc */
657 1.1 jtc if (strcmp(oname, pt->oname) == 0) {
658 1.1 jtc /*
659 1.1 jtc * found it, replace it with the new name
660 1.1 jtc * and return (we know that oname has enough space)
661 1.1 jtc */
662 1.1 jtc *onamelen = l_strncpy(oname, pt->nname, PAXPATHLEN+1);
663 1.1 jtc return;
664 1.1 jtc }
665 1.1 jtc pt = pt->fow;
666 1.1 jtc }
667 1.1 jtc
668 1.1 jtc /*
669 1.1 jtc * no match, just return
670 1.1 jtc */
671 1.1 jtc return;
672 1.1 jtc }
673 1.1 jtc
674 1.1 jtc /*
675 1.1 jtc * device/inode mapping table routines
676 1.1 jtc * (used with formats that store device and inodes fields)
677 1.1 jtc *
678 1.1 jtc * device/inode mapping tables remap the device field in a archive header. The
679 1.1 jtc * device/inode fields are used to determine when files are hard links to each
680 1.1 jtc * other. However these values have very little meaning outside of that. This
681 1.1 jtc * database is used to solve one of two different problems.
682 1.1 jtc *
683 1.1 jtc * 1) when files are appended to an archive, while the new files may have hard
684 1.1 jtc * links to each other, you cannot determine if they have hard links to any
685 1.1 jtc * file already stored on the archive from a prior run of pax. We must assume
686 1.1 jtc * that these inode/device pairs are unique only within a SINGLE run of pax
687 1.1 jtc * (which adds a set of files to an archive). So we have to make sure the
688 1.1 jtc * inode/dev pairs we add each time are always unique. We do this by observing
689 1.1 jtc * while the inode field is very dense, the use of the dev field is fairly
690 1.1 jtc * sparse. Within each run of pax, we remap any device number of a new archive
691 1.1 jtc * member that has a device number used in a prior run and already stored in a
692 1.1 jtc * file on the archive. During the read phase of the append, we store the
693 1.1 jtc * device numbers used and mark them to not be used by any file during the
694 1.1 jtc * write phase. If during write we go to use one of those old device numbers,
695 1.1 jtc * we remap it to a new value.
696 1.1 jtc *
697 1.1 jtc * 2) Often the fields in the archive header used to store these values are
698 1.1 jtc * too small to store the entire value. The result is an inode or device value
699 1.1 jtc * which can be truncated. This really can foul up an archive. With truncation
700 1.1 jtc * we end up creating links between files that are really not links (after
701 1.1 jtc * truncation the inodes are the same value). We address that by detecting
702 1.1 jtc * truncation and forcing a remap of the device field to split truncated
703 1.1 jtc * inodes away from each other. Each truncation creates a pattern of bits that
704 1.1 jtc * are removed. We use this pattern of truncated bits to partition the inodes
705 1.1 jtc * on a single device to many different devices (each one represented by the
706 1.1 jtc * truncated bit pattern). All inodes on the same device that have the same
707 1.1 jtc * truncation pattern are mapped to the same new device. Two inodes that
708 1.1 jtc * truncate to the same value clearly will always have different truncation
709 1.1 jtc * bit patterns, so they will be split from away each other. When we spot
710 1.1 jtc * device truncation we remap the device number to a non truncated value.
711 1.1 jtc * (for more info see table.h for the data structures involved).
712 1.1 jtc */
713 1.1 jtc
714 1.1 jtc /*
715 1.1 jtc * dev_start()
716 1.1 jtc * create the device mapping table
717 1.1 jtc * Return:
718 1.1 jtc * 0 if successful, -1 otherwise
719 1.1 jtc */
720 1.1 jtc
721 1.1 jtc #if __STDC__
722 1.1 jtc int
723 1.1 jtc dev_start(void)
724 1.1 jtc #else
725 1.1 jtc int
726 1.1 jtc dev_start()
727 1.1 jtc #endif
728 1.1 jtc {
729 1.1 jtc if (dtab != NULL)
730 1.1 jtc return(0);
731 1.1 jtc if ((dtab = (DEVT **)calloc(D_TAB_SZ, sizeof(DEVT *))) == NULL) {
732 1.1 jtc warn(1, "Cannot allocate memory for device mapping table");
733 1.1 jtc return(-1);
734 1.1 jtc }
735 1.1 jtc return(0);
736 1.1 jtc }
737 1.1 jtc
738 1.1 jtc /*
739 1.1 jtc * add_dev()
740 1.1 jtc * add a device number to the table. this will force the device to be
741 1.1 jtc * remapped to a new value if it be used during a write phase. This
742 1.1 jtc * function is called during the read phase of an append to prohibit the
743 1.1 jtc * use of any device number already in the archive.
744 1.1 jtc * Return:
745 1.1 jtc * 0 if added ok, -1 otherwise
746 1.1 jtc */
747 1.1 jtc
748 1.1 jtc #if __STDC__
749 1.1 jtc int
750 1.5 tls add_dev(ARCHD *arcn)
751 1.1 jtc #else
752 1.1 jtc int
753 1.1 jtc add_dev(arcn)
754 1.5 tls ARCHD *arcn;
755 1.1 jtc #endif
756 1.1 jtc {
757 1.1 jtc if (chk_dev(arcn->sb.st_dev, 1) == NULL)
758 1.1 jtc return(-1);
759 1.1 jtc return(0);
760 1.1 jtc }
761 1.1 jtc
762 1.1 jtc /*
763 1.1 jtc * chk_dev()
764 1.1 jtc * check for a device value in the device table. If not found and the add
765 1.1 jtc * flag is set, it is added. This does NOT assign any mapping values, just
766 1.1 jtc * adds the device number as one that need to be remapped. If this device
767 1.1 jtc * is alread mapped, just return with a pointer to that entry.
768 1.1 jtc * Return:
769 1.1 jtc * pointer to the entry for this device in the device map table. Null
770 1.1 jtc * if the add flag is not set and the device is not in the table (it is
771 1.1 jtc * not been seen yet). If add is set and the device cannot be added, null
772 1.1 jtc * is returned (indicates an error).
773 1.1 jtc */
774 1.1 jtc
775 1.1 jtc #if __STDC__
776 1.1 jtc static DEVT *
777 1.1 jtc chk_dev(dev_t dev, int add)
778 1.1 jtc #else
779 1.1 jtc static DEVT *
780 1.1 jtc chk_dev(dev, add)
781 1.1 jtc dev_t dev;
782 1.1 jtc int add;
783 1.1 jtc #endif
784 1.1 jtc {
785 1.5 tls DEVT *pt;
786 1.5 tls u_int indx;
787 1.1 jtc
788 1.1 jtc if (dtab == NULL)
789 1.1 jtc return(NULL);
790 1.1 jtc /*
791 1.1 jtc * look to see if this device is already in the table
792 1.1 jtc */
793 1.1 jtc indx = ((unsigned)dev) % D_TAB_SZ;
794 1.1 jtc if ((pt = dtab[indx]) != NULL) {
795 1.1 jtc while ((pt != NULL) && (pt->dev != dev))
796 1.1 jtc pt = pt->fow;
797 1.1 jtc
798 1.1 jtc /*
799 1.1 jtc * found it, return a pointer to it
800 1.1 jtc */
801 1.1 jtc if (pt != NULL)
802 1.1 jtc return(pt);
803 1.1 jtc }
804 1.1 jtc
805 1.1 jtc /*
806 1.1 jtc * not in table, we add it only if told to as this may just be a check
807 1.1 jtc * to see if a device number is being used.
808 1.1 jtc */
809 1.1 jtc if (add == 0)
810 1.1 jtc return(NULL);
811 1.1 jtc
812 1.1 jtc /*
813 1.1 jtc * allocate a node for this device and add it to the front of the hash
814 1.1 jtc * chain. Note we do not assign remaps values here, so the pt->list
815 1.1 jtc * list must be NULL.
816 1.1 jtc */
817 1.1 jtc if ((pt = (DEVT *)malloc(sizeof(DEVT))) == NULL) {
818 1.1 jtc warn(1, "Device map table out of memory");
819 1.1 jtc return(NULL);
820 1.1 jtc }
821 1.1 jtc pt->dev = dev;
822 1.1 jtc pt->list = NULL;
823 1.1 jtc pt->fow = dtab[indx];
824 1.1 jtc dtab[indx] = pt;
825 1.1 jtc return(pt);
826 1.1 jtc }
827 1.1 jtc /*
828 1.1 jtc * map_dev()
829 1.1 jtc * given an inode and device storage mask (the mask has a 1 for each bit
830 1.1 jtc * the archive format is able to store in a header), we check for inode
831 1.1 jtc * and device truncation and remap the device as required. Device mapping
832 1.1 jtc * can also occur when during the read phase of append a device number was
833 1.1 jtc * seen (and was marked as do not use during the write phase). WE ASSUME
834 1.1 jtc * that unsigned longs are the same size or bigger than the fields used
835 1.1 jtc * for ino_t and dev_t. If not the types will have to be changed.
836 1.1 jtc * Return:
837 1.1 jtc * 0 if all ok, -1 otherwise.
838 1.1 jtc */
839 1.1 jtc
840 1.1 jtc #if __STDC__
841 1.1 jtc int
842 1.5 tls map_dev(ARCHD *arcn, u_long dev_mask, u_long ino_mask)
843 1.1 jtc #else
844 1.1 jtc int
845 1.1 jtc map_dev(arcn, dev_mask, ino_mask)
846 1.5 tls ARCHD *arcn;
847 1.1 jtc u_long dev_mask;
848 1.1 jtc u_long ino_mask;
849 1.1 jtc #endif
850 1.1 jtc {
851 1.5 tls DEVT *pt;
852 1.5 tls DLIST *dpt;
853 1.1 jtc static dev_t lastdev = 0; /* next device number to try */
854 1.1 jtc int trc_ino = 0;
855 1.1 jtc int trc_dev = 0;
856 1.1 jtc ino_t trunc_bits = 0;
857 1.1 jtc ino_t nino;
858 1.1 jtc
859 1.1 jtc if (dtab == NULL)
860 1.1 jtc return(0);
861 1.1 jtc /*
862 1.1 jtc * check for device and inode truncation, and extract the truncated
863 1.1 jtc * bit pattern.
864 1.1 jtc */
865 1.1 jtc if ((arcn->sb.st_dev & (dev_t)dev_mask) != arcn->sb.st_dev)
866 1.1 jtc ++trc_dev;
867 1.1 jtc if ((nino = arcn->sb.st_ino & (ino_t)ino_mask) != arcn->sb.st_ino) {
868 1.1 jtc ++trc_ino;
869 1.1 jtc trunc_bits = arcn->sb.st_ino & (ino_t)(~ino_mask);
870 1.1 jtc }
871 1.1 jtc
872 1.1 jtc /*
873 1.1 jtc * see if this device is already being mapped, look up the device
874 1.1 jtc * then find the truncation bit pattern which applies
875 1.1 jtc */
876 1.1 jtc if ((pt = chk_dev(arcn->sb.st_dev, 0)) != NULL) {
877 1.1 jtc /*
878 1.1 jtc * this device is already marked to be remapped
879 1.1 jtc */
880 1.1 jtc for (dpt = pt->list; dpt != NULL; dpt = dpt->fow)
881 1.1 jtc if (dpt->trunc_bits == trunc_bits)
882 1.1 jtc break;
883 1.1 jtc
884 1.1 jtc if (dpt != NULL) {
885 1.1 jtc /*
886 1.1 jtc * we are being remapped for this device and pattern
887 1.1 jtc * change the device number to be stored and return
888 1.1 jtc */
889 1.1 jtc arcn->sb.st_dev = dpt->dev;
890 1.1 jtc arcn->sb.st_ino = nino;
891 1.1 jtc return(0);
892 1.1 jtc }
893 1.1 jtc } else {
894 1.1 jtc /*
895 1.1 jtc * this device is not being remapped YET. if we do not have any
896 1.1 jtc * form of truncation, we do not need a remap
897 1.1 jtc */
898 1.1 jtc if (!trc_ino && !trc_dev)
899 1.1 jtc return(0);
900 1.1 jtc
901 1.1 jtc /*
902 1.1 jtc * we have truncation, have to add this as a device to remap
903 1.1 jtc */
904 1.1 jtc if ((pt = chk_dev(arcn->sb.st_dev, 1)) == NULL)
905 1.1 jtc goto bad;
906 1.1 jtc
907 1.1 jtc /*
908 1.1 jtc * if we just have a truncated inode, we have to make sure that
909 1.1 jtc * all future inodes that do not truncate (they have the
910 1.1 jtc * truncation pattern of all 0's) continue to map to the same
911 1.1 jtc * device number. We probably have already written inodes with
912 1.1 jtc * this device number to the archive with the truncation
913 1.1 jtc * pattern of all 0's. So we add the mapping for all 0's to the
914 1.1 jtc * same device number.
915 1.1 jtc */
916 1.1 jtc if (!trc_dev && (trunc_bits != 0)) {
917 1.1 jtc if ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL)
918 1.1 jtc goto bad;
919 1.1 jtc dpt->trunc_bits = 0;
920 1.1 jtc dpt->dev = arcn->sb.st_dev;
921 1.1 jtc dpt->fow = pt->list;
922 1.1 jtc pt->list = dpt;
923 1.1 jtc }
924 1.1 jtc }
925 1.1 jtc
926 1.1 jtc /*
927 1.1 jtc * look for a device number not being used. We must watch for wrap
928 1.1 jtc * around on lastdev (so we do not get stuck looking forever!)
929 1.1 jtc */
930 1.1 jtc while (++lastdev > 0) {
931 1.1 jtc if (chk_dev(lastdev, 0) != NULL)
932 1.1 jtc continue;
933 1.1 jtc /*
934 1.1 jtc * found an unused value. If we have reached truncation point
935 1.1 jtc * for this format we are hosed, so we give up. Otherwise we
936 1.1 jtc * mark it as being used.
937 1.1 jtc */
938 1.1 jtc if (((lastdev & ((dev_t)dev_mask)) != lastdev) ||
939 1.1 jtc (chk_dev(lastdev, 1) == NULL))
940 1.1 jtc goto bad;
941 1.1 jtc break;
942 1.1 jtc }
943 1.1 jtc
944 1.1 jtc if ((lastdev <= 0) || ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL))
945 1.1 jtc goto bad;
946 1.1 jtc
947 1.1 jtc /*
948 1.1 jtc * got a new device number, store it under this truncation pattern.
949 1.1 jtc * change the device number this file is being stored with.
950 1.1 jtc */
951 1.1 jtc dpt->trunc_bits = trunc_bits;
952 1.1 jtc dpt->dev = lastdev;
953 1.1 jtc dpt->fow = pt->list;
954 1.1 jtc pt->list = dpt;
955 1.1 jtc arcn->sb.st_dev = lastdev;
956 1.1 jtc arcn->sb.st_ino = nino;
957 1.1 jtc return(0);
958 1.1 jtc
959 1.1 jtc bad:
960 1.1 jtc warn(1, "Unable to fix truncated inode/device field when storing %s",
961 1.1 jtc arcn->name);
962 1.1 jtc warn(0, "Archive may create improper hard links when extracted");
963 1.1 jtc return(0);
964 1.1 jtc }
965 1.1 jtc
966 1.1 jtc /*
967 1.1 jtc * directory access/mod time reset table routines (for directories READ by pax)
968 1.1 jtc *
969 1.1 jtc * The pax -t flag requires that access times of archive files to be the same
970 1.1 jtc * before being read by pax. For regular files, access time is restored after
971 1.1 jtc * the file has been copied. This database provides the same functionality for
972 1.1 jtc * directories read during file tree traversal. Restoring directory access time
973 1.1 jtc * is more complex than files since directories may be read several times until
974 1.1 jtc * all the descendants in their subtree are visited by fts. Directory access
975 1.1 jtc * and modification times are stored during the fts pre-order visit (done
976 1.1 jtc * before any descendants in the subtree is visited) and restored after the
977 1.1 jtc * fts post-order visit (after all the descendants have been visited). In the
978 1.1 jtc * case of premature exit from a subtree (like from the effects of -n), any
979 1.1 jtc * directory entries left in this database are reset during final cleanup
980 1.1 jtc * operations of pax. Entries are hashed by inode number for fast lookup.
981 1.1 jtc */
982 1.1 jtc
983 1.1 jtc /*
984 1.1 jtc * atdir_start()
985 1.1 jtc * create the directory access time database for directories READ by pax.
986 1.1 jtc * Return:
987 1.1 jtc * 0 is created ok, -1 otherwise.
988 1.1 jtc */
989 1.1 jtc
990 1.1 jtc #if __STDC__
991 1.1 jtc int
992 1.1 jtc atdir_start(void)
993 1.1 jtc #else
994 1.1 jtc int
995 1.1 jtc atdir_start()
996 1.1 jtc #endif
997 1.1 jtc {
998 1.1 jtc if (atab != NULL)
999 1.1 jtc return(0);
1000 1.1 jtc if ((atab = (ATDIR **)calloc(A_TAB_SZ, sizeof(ATDIR *))) == NULL) {
1001 1.1 jtc warn(1,"Cannot allocate space for directory access time table");
1002 1.1 jtc return(-1);
1003 1.1 jtc }
1004 1.1 jtc return(0);
1005 1.1 jtc }
1006 1.1 jtc
1007 1.1 jtc
1008 1.1 jtc /*
1009 1.1 jtc * atdir_end()
1010 1.1 jtc * walk through the directory access time table and reset the access time
1011 1.1 jtc * of any directory who still has an entry left in the database. These
1012 1.1 jtc * entries are for directories READ by pax
1013 1.1 jtc */
1014 1.1 jtc
1015 1.1 jtc #if __STDC__
1016 1.1 jtc void
1017 1.1 jtc atdir_end(void)
1018 1.1 jtc #else
1019 1.1 jtc void
1020 1.1 jtc atdir_end()
1021 1.1 jtc #endif
1022 1.1 jtc {
1023 1.5 tls ATDIR *pt;
1024 1.5 tls int i;
1025 1.1 jtc
1026 1.1 jtc if (atab == NULL)
1027 1.1 jtc return;
1028 1.1 jtc /*
1029 1.1 jtc * for each non-empty hash table entry reset all the directories
1030 1.1 jtc * chained there.
1031 1.1 jtc */
1032 1.1 jtc for (i = 0; i < A_TAB_SZ; ++i) {
1033 1.1 jtc if ((pt = atab[i]) == NULL)
1034 1.1 jtc continue;
1035 1.1 jtc /*
1036 1.1 jtc * remember to force the times, set_ftime() looks at pmtime
1037 1.1 jtc * and patime, which only applies to things CREATED by pax,
1038 1.1 jtc * not read by pax. Read time reset is controlled by -t.
1039 1.1 jtc */
1040 1.1 jtc for (; pt != NULL; pt = pt->fow)
1041 1.1 jtc set_ftime(pt->name, pt->mtime, pt->atime, 1);
1042 1.1 jtc }
1043 1.1 jtc }
1044 1.1 jtc
1045 1.1 jtc /*
1046 1.1 jtc * add_atdir()
1047 1.1 jtc * add a directory to the directory access time table. Table is hashed
1048 1.1 jtc * and chained by inode number. This is for directories READ by pax
1049 1.1 jtc */
1050 1.1 jtc
1051 1.1 jtc #if __STDC__
1052 1.1 jtc void
1053 1.1 jtc add_atdir(char *fname, dev_t dev, ino_t ino, time_t mtime, time_t atime)
1054 1.1 jtc #else
1055 1.1 jtc void
1056 1.1 jtc add_atdir(fname, dev, ino, mtime, atime)
1057 1.1 jtc char *fname;
1058 1.1 jtc dev_t dev;
1059 1.1 jtc ino_t ino;
1060 1.1 jtc time_t mtime;
1061 1.1 jtc time_t atime;
1062 1.1 jtc #endif
1063 1.1 jtc {
1064 1.5 tls ATDIR *pt;
1065 1.5 tls u_int indx;
1066 1.1 jtc
1067 1.1 jtc if (atab == NULL)
1068 1.1 jtc return;
1069 1.1 jtc
1070 1.1 jtc /*
1071 1.1 jtc * make sure this directory is not already in the table, if so just
1072 1.1 jtc * return (the older entry always has the correct time). The only
1073 1.1 jtc * way this will happen is when the same subtree can be traversed by
1074 1.1 jtc * different args to pax and the -n option is aborting fts out of a
1075 1.1 jtc * subtree before all the post-order visits have been made).
1076 1.1 jtc */
1077 1.1 jtc indx = ((unsigned)ino) % A_TAB_SZ;
1078 1.1 jtc if ((pt = atab[indx]) != NULL) {
1079 1.1 jtc while (pt != NULL) {
1080 1.1 jtc if ((pt->ino == ino) && (pt->dev == dev))
1081 1.1 jtc break;
1082 1.1 jtc pt = pt->fow;
1083 1.1 jtc }
1084 1.1 jtc
1085 1.1 jtc /*
1086 1.1 jtc * oops, already there. Leave it alone.
1087 1.1 jtc */
1088 1.1 jtc if (pt != NULL)
1089 1.1 jtc return;
1090 1.1 jtc }
1091 1.1 jtc
1092 1.1 jtc /*
1093 1.1 jtc * add it to the front of the hash chain
1094 1.1 jtc */
1095 1.1 jtc if ((pt = (ATDIR *)malloc(sizeof(ATDIR))) != NULL) {
1096 1.1 jtc if ((pt->name = strdup(fname)) != NULL) {
1097 1.1 jtc pt->dev = dev;
1098 1.1 jtc pt->ino = ino;
1099 1.1 jtc pt->mtime = mtime;
1100 1.1 jtc pt->atime = atime;
1101 1.1 jtc pt->fow = atab[indx];
1102 1.1 jtc atab[indx] = pt;
1103 1.1 jtc return;
1104 1.1 jtc }
1105 1.1 jtc (void)free((char *)pt);
1106 1.1 jtc }
1107 1.1 jtc
1108 1.1 jtc warn(1, "Directory access time reset table ran out of memory");
1109 1.1 jtc return;
1110 1.1 jtc }
1111 1.1 jtc
1112 1.1 jtc /*
1113 1.1 jtc * get_atdir()
1114 1.1 jtc * look up a directory by inode and device number to obtain the access
1115 1.1 jtc * and modification time you want to set to. If found, the modification
1116 1.1 jtc * and access time parameters are set and the entry is removed from the
1117 1.1 jtc * table (as it is no longer needed). These are for directories READ by
1118 1.1 jtc * pax
1119 1.1 jtc * Return:
1120 1.1 jtc * 0 if found, -1 if not found.
1121 1.1 jtc */
1122 1.1 jtc
1123 1.1 jtc #if __STDC__
1124 1.1 jtc int
1125 1.1 jtc get_atdir(dev_t dev, ino_t ino, time_t *mtime, time_t *atime)
1126 1.1 jtc #else
1127 1.1 jtc int
1128 1.1 jtc get_atdir(dev, ino, mtime, atime)
1129 1.1 jtc dev_t dev;
1130 1.1 jtc ino_t ino;
1131 1.1 jtc time_t *mtime;
1132 1.1 jtc time_t *atime;
1133 1.1 jtc #endif
1134 1.1 jtc {
1135 1.5 tls ATDIR *pt;
1136 1.5 tls ATDIR **ppt;
1137 1.5 tls u_int indx;
1138 1.1 jtc
1139 1.1 jtc if (atab == NULL)
1140 1.1 jtc return(-1);
1141 1.1 jtc /*
1142 1.1 jtc * hash by inode and search the chain for an inode and device match
1143 1.1 jtc */
1144 1.1 jtc indx = ((unsigned)ino) % A_TAB_SZ;
1145 1.1 jtc if ((pt = atab[indx]) == NULL)
1146 1.1 jtc return(-1);
1147 1.1 jtc
1148 1.1 jtc ppt = &(atab[indx]);
1149 1.1 jtc while (pt != NULL) {
1150 1.1 jtc if ((pt->ino == ino) && (pt->dev == dev))
1151 1.1 jtc break;
1152 1.1 jtc /*
1153 1.1 jtc * no match, go to next one
1154 1.1 jtc */
1155 1.1 jtc ppt = &(pt->fow);
1156 1.1 jtc pt = pt->fow;
1157 1.1 jtc }
1158 1.1 jtc
1159 1.1 jtc /*
1160 1.1 jtc * return if we did not find it.
1161 1.1 jtc */
1162 1.1 jtc if (pt == NULL)
1163 1.1 jtc return(-1);
1164 1.1 jtc
1165 1.1 jtc /*
1166 1.1 jtc * found it. return the times and remove the entry from the table.
1167 1.1 jtc */
1168 1.1 jtc *ppt = pt->fow;
1169 1.1 jtc *mtime = pt->mtime;
1170 1.1 jtc *atime = pt->atime;
1171 1.1 jtc (void)free((char *)pt->name);
1172 1.1 jtc (void)free((char *)pt);
1173 1.1 jtc return(0);
1174 1.1 jtc }
1175 1.1 jtc
1176 1.1 jtc /*
1177 1.1 jtc * directory access mode and time storage routines (for directories CREATED
1178 1.1 jtc * by pax).
1179 1.1 jtc *
1180 1.1 jtc * Pax requires that extracted directories, by default, have their access/mod
1181 1.1 jtc * times and permissions set to the values specified in the archive. During the
1182 1.1 jtc * actions of extracting (and creating the destination subtree during -rw copy)
1183 1.1 jtc * directories extracted may be modified after being created. Even worse is
1184 1.1 jtc * that these directories may have been created with file permissions which
1185 1.1 jtc * prohibits any descendants of these directories from being extracted. When
1186 1.1 jtc * directories are created by pax, access rights may be added to permit the
1187 1.1 jtc * creation of files in their subtree. Every time pax creates a directory, the
1188 1.1 jtc * times and file permissions specified by the archive are stored. After all
1189 1.1 jtc * files have been extracted (or copied), these directories have their times
1190 1.1 jtc * and file modes reset to the stored values. The directory info is restored in
1191 1.1 jtc * reverse order as entries were added to the data file from root to leaf. To
1192 1.1 jtc * restore atime properly, we must go backwards. The data file consists of
1193 1.1 jtc * records with two parts, the file name followed by a DIRDATA trailer. The
1194 1.1 jtc * fixed sized trailer contains the size of the name plus the off_t location in
1195 1.1 jtc * the file. To restore we work backwards through the file reading the trailer
1196 1.1 jtc * then the file name.
1197 1.1 jtc */
1198 1.1 jtc
1199 1.1 jtc /*
1200 1.1 jtc * dir_start()
1201 1.1 jtc * set up the directory time and file mode storage for directories CREATED
1202 1.1 jtc * by pax.
1203 1.1 jtc * Return:
1204 1.1 jtc * 0 if ok, -1 otherwise
1205 1.1 jtc */
1206 1.1 jtc
1207 1.1 jtc #if __STDC__
1208 1.1 jtc int
1209 1.1 jtc dir_start(void)
1210 1.1 jtc #else
1211 1.1 jtc int
1212 1.1 jtc dir_start()
1213 1.1 jtc #endif
1214 1.1 jtc {
1215 1.1 jtc char *pt;
1216 1.1 jtc
1217 1.1 jtc if (dirfd != -1)
1218 1.1 jtc return(0);
1219 1.1 jtc if ((pt = tempnam((char *)NULL, (char *)NULL)) == NULL)
1220 1.1 jtc return(-1);
1221 1.1 jtc
1222 1.1 jtc /*
1223 1.1 jtc * unlink the file so it goes away at termination by itself
1224 1.1 jtc */
1225 1.1 jtc (void)unlink(pt);
1226 1.1 jtc if ((dirfd = open(pt, O_RDWR|O_CREAT, 0600)) >= 0) {
1227 1.1 jtc (void)unlink(pt);
1228 1.1 jtc return(0);
1229 1.1 jtc }
1230 1.1 jtc warn(1, "Unable to create temporary file for directory times: %s", pt);
1231 1.1 jtc return(-1);
1232 1.1 jtc }
1233 1.1 jtc
1234 1.1 jtc /*
1235 1.1 jtc * add_dir()
1236 1.1 jtc * add the mode and times for a newly CREATED directory
1237 1.1 jtc * name is name of the directory, psb the stat buffer with the data in it,
1238 1.1 jtc * frc_mode is a flag that says whether to force the setting of the mode
1239 1.1 jtc * (ignoring the user set values for preserving file mode). Frc_mode is
1240 1.1 jtc * for the case where we created a file and found that the resulting
1241 1.1 jtc * directory was not writeable and the user asked for file modes to NOT
1242 1.1 jtc * be preserved. (we have to preserve what was created by default, so we
1243 1.1 jtc * have to force the setting at the end. this is stated explicitly in the
1244 1.1 jtc * pax spec)
1245 1.1 jtc */
1246 1.1 jtc
1247 1.1 jtc #if __STDC__
1248 1.1 jtc void
1249 1.1 jtc add_dir(char *name, int nlen, struct stat *psb, int frc_mode)
1250 1.1 jtc #else
1251 1.1 jtc void
1252 1.1 jtc add_dir(name, nlen, psb, frc_mode)
1253 1.1 jtc char *name;
1254 1.1 jtc int nlen;
1255 1.1 jtc struct stat *psb;
1256 1.1 jtc int frc_mode;
1257 1.1 jtc #endif
1258 1.1 jtc {
1259 1.1 jtc DIRDATA dblk;
1260 1.1 jtc
1261 1.1 jtc if (dirfd < 0)
1262 1.1 jtc return;
1263 1.1 jtc
1264 1.1 jtc /*
1265 1.1 jtc * get current position (where file name will start) so we can store it
1266 1.1 jtc * in the trailer
1267 1.1 jtc */
1268 1.1 jtc if ((dblk.npos = lseek(dirfd, 0L, SEEK_CUR)) < 0) {
1269 1.1 jtc warn(1,"Unable to store mode and times for directory: %s",name);
1270 1.1 jtc return;
1271 1.1 jtc }
1272 1.1 jtc
1273 1.1 jtc /*
1274 1.1 jtc * write the file name followed by the trailer
1275 1.1 jtc */
1276 1.1 jtc dblk.nlen = nlen + 1;
1277 1.1 jtc dblk.mode = psb->st_mode & 0xffff;
1278 1.1 jtc dblk.mtime = psb->st_mtime;
1279 1.1 jtc dblk.atime = psb->st_atime;
1280 1.1 jtc dblk.frc_mode = frc_mode;
1281 1.1 jtc if ((write(dirfd, name, dblk.nlen) == dblk.nlen) &&
1282 1.1 jtc (write(dirfd, (char *)&dblk, sizeof(dblk)) == sizeof(dblk))) {
1283 1.1 jtc ++dircnt;
1284 1.1 jtc return;
1285 1.1 jtc }
1286 1.1 jtc
1287 1.1 jtc warn(1,"Unable to store mode and times for created directory: %s",name);
1288 1.1 jtc return;
1289 1.1 jtc }
1290 1.1 jtc
1291 1.1 jtc /*
1292 1.1 jtc * proc_dir()
1293 1.1 jtc * process all file modes and times stored for directories CREATED
1294 1.1 jtc * by pax
1295 1.1 jtc */
1296 1.1 jtc
1297 1.1 jtc #if __STDC__
1298 1.1 jtc void
1299 1.1 jtc proc_dir(void)
1300 1.1 jtc #else
1301 1.1 jtc void
1302 1.1 jtc proc_dir()
1303 1.1 jtc #endif
1304 1.1 jtc {
1305 1.1 jtc char name[PAXPATHLEN+1];
1306 1.1 jtc DIRDATA dblk;
1307 1.1 jtc u_long cnt;
1308 1.1 jtc
1309 1.1 jtc if (dirfd < 0)
1310 1.1 jtc return;
1311 1.1 jtc /*
1312 1.1 jtc * read backwards through the file and process each directory
1313 1.1 jtc */
1314 1.1 jtc for (cnt = 0; cnt < dircnt; ++cnt) {
1315 1.1 jtc /*
1316 1.1 jtc * read the trailer, then the file name, if this fails
1317 1.1 jtc * just give up.
1318 1.1 jtc */
1319 1.1 jtc if (lseek(dirfd, -((off_t)sizeof(dblk)), SEEK_CUR) < 0)
1320 1.1 jtc break;
1321 1.1 jtc if (read(dirfd,(char *)&dblk, sizeof(dblk)) != sizeof(dblk))
1322 1.1 jtc break;
1323 1.1 jtc if (lseek(dirfd, dblk.npos, SEEK_SET) < 0)
1324 1.1 jtc break;
1325 1.1 jtc if (read(dirfd, name, dblk.nlen) != dblk.nlen)
1326 1.1 jtc break;
1327 1.1 jtc if (lseek(dirfd, dblk.npos, SEEK_SET) < 0)
1328 1.1 jtc break;
1329 1.1 jtc
1330 1.1 jtc /*
1331 1.1 jtc * frc_mode set, make sure we set the file modes even if
1332 1.1 jtc * the user didn't ask for it (see file_subs.c for more info)
1333 1.1 jtc */
1334 1.1 jtc if (pmode || dblk.frc_mode)
1335 1.1 jtc set_pmode(name, dblk.mode);
1336 1.1 jtc if (patime || pmtime)
1337 1.1 jtc set_ftime(name, dblk.mtime, dblk.atime, 0);
1338 1.1 jtc }
1339 1.1 jtc
1340 1.1 jtc (void)close(dirfd);
1341 1.1 jtc dirfd = -1;
1342 1.1 jtc if (cnt != dircnt)
1343 1.1 jtc warn(1,"Unable to set mode and times for created directories");
1344 1.1 jtc return;
1345 1.1 jtc }
1346 1.1 jtc
1347 1.1 jtc /*
1348 1.1 jtc * database independent routines
1349 1.1 jtc */
1350 1.1 jtc
1351 1.1 jtc /*
1352 1.1 jtc * st_hash()
1353 1.1 jtc * hashes filenames to a u_int for hashing into a table. Looks at the tail
1354 1.1 jtc * end of file, as this provides far better distribution than any other
1355 1.1 jtc * part of the name. For performance reasons we only care about the last
1356 1.1 jtc * MAXKEYLEN chars (should be at LEAST large enough to pick off the file
1357 1.1 jtc * name). Was tested on 500,000 name file tree traversal from the root
1358 1.1 jtc * and gave almost a perfectly uniform distribution of keys when used with
1359 1.1 jtc * prime sized tables (MAXKEYLEN was 128 in test). Hashes (sizeof int)
1360 1.1 jtc * chars at a time and pads with 0 for last addition.
1361 1.1 jtc * Return:
1362 1.1 jtc * the hash value of the string MOD (%) the table size.
1363 1.1 jtc */
1364 1.1 jtc
1365 1.1 jtc #if __STDC__
1366 1.1 jtc u_int
1367 1.1 jtc st_hash(char *name, int len, int tabsz)
1368 1.1 jtc #else
1369 1.1 jtc u_int
1370 1.1 jtc st_hash(name, len, tabsz)
1371 1.1 jtc char *name;
1372 1.1 jtc int len;
1373 1.1 jtc int tabsz;
1374 1.1 jtc #endif
1375 1.1 jtc {
1376 1.5 tls char *pt;
1377 1.5 tls char *dest;
1378 1.5 tls char *end;
1379 1.5 tls int i;
1380 1.5 tls u_int key = 0;
1381 1.5 tls int steps;
1382 1.5 tls int res;
1383 1.1 jtc u_int val;
1384 1.1 jtc
1385 1.1 jtc /*
1386 1.1 jtc * only look at the tail up to MAXKEYLEN, we do not need to waste
1387 1.1 jtc * time here (remember these are pathnames, the tail is what will
1388 1.1 jtc * spread out the keys)
1389 1.1 jtc */
1390 1.1 jtc if (len > MAXKEYLEN) {
1391 1.1 jtc pt = &(name[len - MAXKEYLEN]);
1392 1.1 jtc len = MAXKEYLEN;
1393 1.1 jtc } else
1394 1.1 jtc pt = name;
1395 1.1 jtc
1396 1.1 jtc /*
1397 1.1 jtc * calculate the number of u_int size steps in the string and if
1398 1.1 jtc * there is a runt to deal with
1399 1.1 jtc */
1400 1.1 jtc steps = len/sizeof(u_int);
1401 1.1 jtc res = len % sizeof(u_int);
1402 1.1 jtc
1403 1.1 jtc /*
1404 1.1 jtc * add up the value of the string in unsigned integer sized pieces
1405 1.1 jtc * too bad we cannot have unsigned int aligned strings, then we
1406 1.1 jtc * could avoid the expensive copy.
1407 1.1 jtc */
1408 1.1 jtc for (i = 0; i < steps; ++i) {
1409 1.1 jtc end = pt + sizeof(u_int);
1410 1.1 jtc dest = (char *)&val;
1411 1.1 jtc while (pt < end)
1412 1.1 jtc *dest++ = *pt++;
1413 1.1 jtc key += val;
1414 1.1 jtc }
1415 1.1 jtc
1416 1.1 jtc /*
1417 1.1 jtc * add in the runt padded with zero to the right
1418 1.1 jtc */
1419 1.1 jtc if (res) {
1420 1.1 jtc val = 0;
1421 1.1 jtc end = pt + res;
1422 1.1 jtc dest = (char *)&val;
1423 1.1 jtc while (pt < end)
1424 1.1 jtc *dest++ = *pt++;
1425 1.1 jtc key += val;
1426 1.1 jtc }
1427 1.1 jtc
1428 1.1 jtc /*
1429 1.1 jtc * return the result mod the table size
1430 1.1 jtc */
1431 1.1 jtc return(key % tabsz);
1432 1.1 jtc }
1433