tables.c revision 1.1.1.1 1 1.1 jtc /*-
2 1.1 jtc * Copyright (c) 1992 Keith Muller.
3 1.1 jtc * Copyright (c) 1992, 1993
4 1.1 jtc * The Regents of the University of California. All rights reserved.
5 1.1 jtc *
6 1.1 jtc * This code is derived from software contributed to Berkeley by
7 1.1 jtc * Keith Muller of the University of California, San Diego.
8 1.1 jtc *
9 1.1 jtc * Redistribution and use in source and binary forms, with or without
10 1.1 jtc * modification, are permitted provided that the following conditions
11 1.1 jtc * are met:
12 1.1 jtc * 1. Redistributions of source code must retain the above copyright
13 1.1 jtc * notice, this list of conditions and the following disclaimer.
14 1.1 jtc * 2. Redistributions in binary form must reproduce the above copyright
15 1.1 jtc * notice, this list of conditions and the following disclaimer in the
16 1.1 jtc * documentation and/or other materials provided with the distribution.
17 1.1 jtc * 3. All advertising materials mentioning features or use of this software
18 1.1 jtc * must display the following acknowledgement:
19 1.1 jtc * This product includes software developed by the University of
20 1.1 jtc * California, Berkeley and its contributors.
21 1.1 jtc * 4. Neither the name of the University nor the names of its contributors
22 1.1 jtc * may be used to endorse or promote products derived from this software
23 1.1 jtc * without specific prior written permission.
24 1.1 jtc *
25 1.1 jtc * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 1.1 jtc * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 1.1 jtc * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 1.1 jtc * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 1.1 jtc * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 1.1 jtc * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 1.1 jtc * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 1.1 jtc * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 1.1 jtc * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 1.1 jtc * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 1.1 jtc * SUCH DAMAGE.
36 1.1 jtc */
37 1.1 jtc
38 1.1 jtc #ifndef lint
39 1.1 jtc static char sccsid[] = "@(#)tables.c 8.1 (Berkeley) 5/31/93";
40 1.1 jtc #endif /* not lint */
41 1.1 jtc
42 1.1 jtc #include <sys/types.h>
43 1.1 jtc #include <sys/time.h>
44 1.1 jtc #include <sys/stat.h>
45 1.1 jtc #include <sys/param.h>
46 1.1 jtc #include <sys/fcntl.h>
47 1.1 jtc #include <stdio.h>
48 1.1 jtc #include <ctype.h>
49 1.1 jtc #include <string.h>
50 1.1 jtc #include <unistd.h>
51 1.1 jtc #include <errno.h>
52 1.1 jtc #include <stdlib.h>
53 1.1 jtc #include "pax.h"
54 1.1 jtc #include "tables.h"
55 1.1 jtc #include "extern.h"
56 1.1 jtc
57 1.1 jtc /*
58 1.1 jtc * Routines for controlling the contents of all the different databases pax
59 1.1 jtc * keeps. Tables are dynamically created only when they are needed. The
60 1.1 jtc * goal was speed and the ability to work with HUGE archives. The databases
61 1.1 jtc * were kept simple, but do have complex rules for when the contents change.
62 1.1 jtc * As of this writing, the posix library functions were more complex than
63 1.1 jtc * needed for this application (pax databases have very short lifetimes and
64 1.1 jtc * do not survive after pax is finished). Pax is required to handle very
65 1.1 jtc * large archives. These database routines carefully combine memory usage and
66 1.1 jtc * temporary file storage in ways which will not significantly impact runtime
67 1.1 jtc * performance while allowing the largest possible archives to be handled.
68 1.1 jtc * Trying to force the fit to the posix databases routines was not considered
69 1.1 jtc * time well spent.
70 1.1 jtc */
71 1.1 jtc
72 1.1 jtc static HRDLNK **ltab = NULL; /* hard link table for detecting hard links */
73 1.1 jtc static FTM **ftab = NULL; /* file time table for updating arch */
74 1.1 jtc static NAMT **ntab = NULL; /* interactive rename storage table */
75 1.1 jtc static DEVT **dtab = NULL; /* device/inode mapping tables */
76 1.1 jtc static ATDIR **atab = NULL; /* file tree directory time reset table */
77 1.1 jtc static int dirfd = -1; /* storage for setting created dir time/mode */
78 1.1 jtc static u_long dircnt; /* entries in dir time/mode storage */
79 1.1 jtc static int ffd = -1; /* tmp file for file time table name storage */
80 1.1 jtc
81 1.1 jtc static DEVT *chk_dev __P((dev_t, int));
82 1.1 jtc
83 1.1 jtc /*
84 1.1 jtc * hard link table routines
85 1.1 jtc *
86 1.1 jtc * The hard link table tries to detect hard links to files using the device and
87 1.1 jtc * inode values. We do this when writing an archive, so we can tell the format
88 1.1 jtc * write routine that this file is a hard link to another file. The format
89 1.1 jtc * write routine then can store this file in whatever way it wants (as a hard
90 1.1 jtc * link if the format supports that like tar, or ignore this info like cpio).
91 1.1 jtc * (Actually a field in the format driver table tells us if the format wants
92 1.1 jtc * hard link info. if not, we do not waste time looking for them). We also use
93 1.1 jtc * the same table when reading an archive. In that situation, this table is
94 1.1 jtc * used by the format read routine to detect hard links from stored dev and
95 1.1 jtc * inode numbers (like cpio). This will allow pax to create a link when one
96 1.1 jtc * can be detected by the archive format.
97 1.1 jtc */
98 1.1 jtc
99 1.1 jtc /*
100 1.1 jtc * lnk_start
101 1.1 jtc * Creates the hard link table.
102 1.1 jtc * Return:
103 1.1 jtc * 0 if created, -1 if failure
104 1.1 jtc */
105 1.1 jtc
106 1.1 jtc #if __STDC__
107 1.1 jtc int
108 1.1 jtc lnk_start(void)
109 1.1 jtc #else
110 1.1 jtc int
111 1.1 jtc lnk_start()
112 1.1 jtc #endif
113 1.1 jtc {
114 1.1 jtc if (ltab != NULL)
115 1.1 jtc return(0);
116 1.1 jtc if ((ltab = (HRDLNK **)calloc(L_TAB_SZ, sizeof(HRDLNK *))) == NULL) {
117 1.1 jtc warn(1, "Cannot allocate memory for hard link table");
118 1.1 jtc return(-1);
119 1.1 jtc }
120 1.1 jtc return(0);
121 1.1 jtc }
122 1.1 jtc
123 1.1 jtc /*
124 1.1 jtc * chk_lnk()
125 1.1 jtc * Looks up entry in hard link hash table. If found, it copies the name
126 1.1 jtc * of the file it is linked to (we already saw that file) into ln_name.
127 1.1 jtc * lnkcnt is decremented and if goes to 1 the node is deleted from the
128 1.1 jtc * database. (We have seen all the links to this file). If not found,
129 1.1 jtc * we add the file to the database if it has the potential for having
130 1.1 jtc * hard links to other files we may process (it has a link count > 1)
131 1.1 jtc * Return:
132 1.1 jtc * if found returns 1; if not found returns 0; -1 on error
133 1.1 jtc */
134 1.1 jtc
135 1.1 jtc #if __STDC__
136 1.1 jtc int
137 1.1 jtc chk_lnk(register ARCHD *arcn)
138 1.1 jtc #else
139 1.1 jtc int
140 1.1 jtc chk_lnk(arcn)
141 1.1 jtc register ARCHD *arcn;
142 1.1 jtc #endif
143 1.1 jtc {
144 1.1 jtc register HRDLNK *pt;
145 1.1 jtc register HRDLNK **ppt;
146 1.1 jtc register u_int indx;
147 1.1 jtc
148 1.1 jtc if (ltab == NULL)
149 1.1 jtc return(-1);
150 1.1 jtc /*
151 1.1 jtc * ignore those nodes that cannot have hard links
152 1.1 jtc */
153 1.1 jtc if ((arcn->type == PAX_DIR) || (arcn->sb.st_nlink <= 1))
154 1.1 jtc return(0);
155 1.1 jtc
156 1.1 jtc /*
157 1.1 jtc * hash inode number and look for this file
158 1.1 jtc */
159 1.1 jtc indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ;
160 1.1 jtc if ((pt = ltab[indx]) != NULL) {
161 1.1 jtc /*
162 1.1 jtc * it's hash chain in not empty, walk down looking for it
163 1.1 jtc */
164 1.1 jtc ppt = &(ltab[indx]);
165 1.1 jtc while (pt != NULL) {
166 1.1 jtc if ((pt->ino == arcn->sb.st_ino) &&
167 1.1 jtc (pt->dev == arcn->sb.st_dev))
168 1.1 jtc break;
169 1.1 jtc ppt = &(pt->fow);
170 1.1 jtc pt = pt->fow;
171 1.1 jtc }
172 1.1 jtc
173 1.1 jtc if (pt != NULL) {
174 1.1 jtc /*
175 1.1 jtc * found a link. set the node type and copy in the
176 1.1 jtc * name of the file it is to link to. we need to
177 1.1 jtc * handle hardlinks to regular files differently than
178 1.1 jtc * other links.
179 1.1 jtc */
180 1.1 jtc arcn->ln_nlen = l_strncpy(arcn->ln_name, pt->name,
181 1.1 jtc PAXPATHLEN+1);
182 1.1 jtc if (arcn->type == PAX_REG)
183 1.1 jtc arcn->type = PAX_HRG;
184 1.1 jtc else
185 1.1 jtc arcn->type = PAX_HLK;
186 1.1 jtc
187 1.1 jtc /*
188 1.1 jtc * if we have found all the links to this file, remove
189 1.1 jtc * it from the database
190 1.1 jtc */
191 1.1 jtc if (--pt->nlink <= 1) {
192 1.1 jtc *ppt = pt->fow;
193 1.1 jtc (void)free((char *)pt->name);
194 1.1 jtc (void)free((char *)pt);
195 1.1 jtc }
196 1.1 jtc return(1);
197 1.1 jtc }
198 1.1 jtc }
199 1.1 jtc
200 1.1 jtc /*
201 1.1 jtc * we never saw this file before. It has links so we add it to the
202 1.1 jtc * front of this hash chain
203 1.1 jtc */
204 1.1 jtc if ((pt = (HRDLNK *)malloc(sizeof(HRDLNK))) != NULL) {
205 1.1 jtc if ((pt->name = strdup(arcn->name)) != NULL) {
206 1.1 jtc pt->dev = arcn->sb.st_dev;
207 1.1 jtc pt->ino = arcn->sb.st_ino;
208 1.1 jtc pt->nlink = arcn->sb.st_nlink;
209 1.1 jtc pt->fow = ltab[indx];
210 1.1 jtc ltab[indx] = pt;
211 1.1 jtc return(0);
212 1.1 jtc }
213 1.1 jtc (void)free((char *)pt);
214 1.1 jtc }
215 1.1 jtc
216 1.1 jtc warn(1, "Hard link table out of memory");
217 1.1 jtc return(-1);
218 1.1 jtc }
219 1.1 jtc
220 1.1 jtc /*
221 1.1 jtc * purg_lnk
222 1.1 jtc * remove reference for a file that we may have added to the data base as
223 1.1 jtc * a potential source for hard links. We ended up not using the file, so
224 1.1 jtc * we do not want to accidently point another file at it later on.
225 1.1 jtc */
226 1.1 jtc
227 1.1 jtc #if __STDC__
228 1.1 jtc void
229 1.1 jtc purg_lnk(register ARCHD *arcn)
230 1.1 jtc #else
231 1.1 jtc void
232 1.1 jtc purg_lnk(arcn)
233 1.1 jtc register ARCHD *arcn;
234 1.1 jtc #endif
235 1.1 jtc {
236 1.1 jtc register HRDLNK *pt;
237 1.1 jtc register HRDLNK **ppt;
238 1.1 jtc register u_int indx;
239 1.1 jtc
240 1.1 jtc if (ltab == NULL)
241 1.1 jtc return;
242 1.1 jtc /*
243 1.1 jtc * do not bother to look if it could not be in the database
244 1.1 jtc */
245 1.1 jtc if ((arcn->sb.st_nlink <= 1) || (arcn->type == PAX_DIR) ||
246 1.1 jtc (arcn->type == PAX_HLK) || (arcn->type == PAX_HRG))
247 1.1 jtc return;
248 1.1 jtc
249 1.1 jtc /*
250 1.1 jtc * find the hash chain for this inode value, if empty return
251 1.1 jtc */
252 1.1 jtc indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ;
253 1.1 jtc if ((pt = ltab[indx]) == NULL)
254 1.1 jtc return;
255 1.1 jtc
256 1.1 jtc /*
257 1.1 jtc * walk down the list looking for the inode/dev pair, unlink and
258 1.1 jtc * free if found
259 1.1 jtc */
260 1.1 jtc ppt = &(ltab[indx]);
261 1.1 jtc while (pt != NULL) {
262 1.1 jtc if ((pt->ino == arcn->sb.st_ino) &&
263 1.1 jtc (pt->dev == arcn->sb.st_dev))
264 1.1 jtc break;
265 1.1 jtc ppt = &(pt->fow);
266 1.1 jtc pt = pt->fow;
267 1.1 jtc }
268 1.1 jtc if (pt == NULL)
269 1.1 jtc return;
270 1.1 jtc
271 1.1 jtc /*
272 1.1 jtc * remove and free it
273 1.1 jtc */
274 1.1 jtc *ppt = pt->fow;
275 1.1 jtc (void)free((char *)pt->name);
276 1.1 jtc (void)free((char *)pt);
277 1.1 jtc }
278 1.1 jtc
279 1.1 jtc /*
280 1.1 jtc * lnk_end()
281 1.1 jtc * pull apart a existing link table so we can reuse it. We do this between
282 1.1 jtc * read and write phases of append with update. (The format may have
283 1.1 jtc * used the link table, and we need to start with a fresh table for the
284 1.1 jtc * write phase
285 1.1 jtc */
286 1.1 jtc
287 1.1 jtc #if __STDC__
288 1.1 jtc void
289 1.1 jtc lnk_end(void)
290 1.1 jtc #else
291 1.1 jtc void
292 1.1 jtc lnk_end()
293 1.1 jtc #endif
294 1.1 jtc {
295 1.1 jtc register int i;
296 1.1 jtc register HRDLNK *pt;
297 1.1 jtc register HRDLNK *ppt;
298 1.1 jtc
299 1.1 jtc if (ltab == NULL)
300 1.1 jtc return;
301 1.1 jtc
302 1.1 jtc for (i = 0; i < L_TAB_SZ; ++i) {
303 1.1 jtc if (ltab[i] == NULL)
304 1.1 jtc continue;
305 1.1 jtc pt = ltab[i];
306 1.1 jtc ltab[i] = NULL;
307 1.1 jtc
308 1.1 jtc /*
309 1.1 jtc * free up each entry on this chain
310 1.1 jtc */
311 1.1 jtc while (pt != NULL) {
312 1.1 jtc ppt = pt;
313 1.1 jtc pt = ppt->fow;
314 1.1 jtc (void)free((char *)ppt->name);
315 1.1 jtc (void)free((char *)ppt);
316 1.1 jtc }
317 1.1 jtc }
318 1.1 jtc return;
319 1.1 jtc }
320 1.1 jtc
321 1.1 jtc /*
322 1.1 jtc * modification time table routines
323 1.1 jtc *
324 1.1 jtc * The modification time table keeps track of last modification times for all
325 1.1 jtc * files stored in an archive during a write phase when -u is set. We only
326 1.1 jtc * add a file to the archive if it is newer than a file with the same name
327 1.1 jtc * already stored on the archive (if there is no other file with the same
328 1.1 jtc * name on the archive it is added). This applies to writes and appends.
329 1.1 jtc * An append with an -u must read the archive and store the modification time
330 1.1 jtc * for every file on that archive before starting the write phase. It is clear
331 1.1 jtc * that this is one HUGE database. To save memory space, the actual file names
332 1.1 jtc * are stored in a scatch file and indexed by an in memory hash table. The
333 1.1 jtc * hash table is indexed by hashing the file path. The nodes in the table store
334 1.1 jtc * the length of the filename and the lseek offset within the scratch file
335 1.1 jtc * where the actual name is stored. Since there are never any deletions to this
336 1.1 jtc * table, fragmentation of the scratch file is never a issue. Lookups seem to
337 1.1 jtc * not exhibit any locality at all (files in the database are rarely
338 1.1 jtc * looked up more than once...). So caching is just a waste of memory. The
339 1.1 jtc * only limitation is the amount of scatch file space available to store the
340 1.1 jtc * path names.
341 1.1 jtc */
342 1.1 jtc
343 1.1 jtc /*
344 1.1 jtc * ftime_start()
345 1.1 jtc * create the file time hash table and open for read/write the scratch
346 1.1 jtc * file. (after created it is unlinked, so when we exit we leave
347 1.1 jtc * no witnesses).
348 1.1 jtc * Return:
349 1.1 jtc * 0 if the table and file was created ok, -1 otherwise
350 1.1 jtc */
351 1.1 jtc
352 1.1 jtc #if __STDC__
353 1.1 jtc int
354 1.1 jtc ftime_start(void)
355 1.1 jtc #else
356 1.1 jtc int
357 1.1 jtc ftime_start()
358 1.1 jtc #endif
359 1.1 jtc {
360 1.1 jtc char *pt;
361 1.1 jtc
362 1.1 jtc if (ftab != NULL)
363 1.1 jtc return(0);
364 1.1 jtc if ((ftab = (FTM **)calloc(F_TAB_SZ, sizeof(FTM *))) == NULL) {
365 1.1 jtc warn(1, "Cannot allocate memory for file time table");
366 1.1 jtc return(-1);
367 1.1 jtc }
368 1.1 jtc
369 1.1 jtc /*
370 1.1 jtc * get random name and create temporary scratch file, unlink name
371 1.1 jtc * so it will get removed on exit
372 1.1 jtc */
373 1.1 jtc if ((pt = tempnam((char *)NULL, (char *)NULL)) == NULL)
374 1.1 jtc return(-1);
375 1.1 jtc (void)unlink(pt);
376 1.1 jtc
377 1.1 jtc if ((ffd = open(pt, O_RDWR | O_CREAT, S_IRWXU)) < 0) {
378 1.1 jtc syswarn(1, errno, "Unable to open temporary file: %s", pt);
379 1.1 jtc return(-1);
380 1.1 jtc }
381 1.1 jtc
382 1.1 jtc (void)unlink(pt);
383 1.1 jtc return(0);
384 1.1 jtc }
385 1.1 jtc
386 1.1 jtc /*
387 1.1 jtc * chk_ftime()
388 1.1 jtc * looks up entry in file time hash table. If not found, the file is
389 1.1 jtc * added to the hash table and the file named stored in the scratch file.
390 1.1 jtc * If a file with the same name is found, the file times are compared and
391 1.1 jtc * the most recent file time is retained. If the new file was younger (or
392 1.1 jtc * was not in the database) the new file is selected for storage.
393 1.1 jtc * Return:
394 1.1 jtc * 0 if file should be added to the archive, 1 if it should be skipped,
395 1.1 jtc * -1 on error
396 1.1 jtc */
397 1.1 jtc
398 1.1 jtc #if __STDC__
399 1.1 jtc int
400 1.1 jtc chk_ftime(register ARCHD *arcn)
401 1.1 jtc #else
402 1.1 jtc int
403 1.1 jtc chk_ftime(arcn)
404 1.1 jtc register ARCHD *arcn;
405 1.1 jtc #endif
406 1.1 jtc {
407 1.1 jtc register FTM *pt;
408 1.1 jtc register int namelen;
409 1.1 jtc register u_int indx;
410 1.1 jtc char ckname[PAXPATHLEN+1];
411 1.1 jtc
412 1.1 jtc /*
413 1.1 jtc * no info, go ahead and add to archive
414 1.1 jtc */
415 1.1 jtc if (ftab == NULL)
416 1.1 jtc return(0);
417 1.1 jtc
418 1.1 jtc /*
419 1.1 jtc * hash the pathname and look up in table
420 1.1 jtc */
421 1.1 jtc namelen = arcn->nlen;
422 1.1 jtc indx = st_hash(arcn->name, namelen, F_TAB_SZ);
423 1.1 jtc if ((pt = ftab[indx]) != NULL) {
424 1.1 jtc /*
425 1.1 jtc * the hash chain is not empty, walk down looking for match
426 1.1 jtc * only read up the path names if the lengths match, speeds
427 1.1 jtc * up the search a lot
428 1.1 jtc */
429 1.1 jtc while (pt != NULL) {
430 1.1 jtc if (pt->namelen == namelen) {
431 1.1 jtc /*
432 1.1 jtc * potential match, have to read the name
433 1.1 jtc * from the scratch file.
434 1.1 jtc */
435 1.1 jtc if (lseek(ffd,pt->seek,SEEK_SET) != pt->seek) {
436 1.1 jtc syswarn(1, errno,
437 1.1 jtc "Failed ftime table seek");
438 1.1 jtc return(-1);
439 1.1 jtc }
440 1.1 jtc if (read(ffd, ckname, namelen) != namelen) {
441 1.1 jtc syswarn(1, errno,
442 1.1 jtc "Failed ftime table read");
443 1.1 jtc return(-1);
444 1.1 jtc }
445 1.1 jtc
446 1.1 jtc /*
447 1.1 jtc * if the names match, we are done
448 1.1 jtc */
449 1.1 jtc if (!strncmp(ckname, arcn->name, namelen))
450 1.1 jtc break;
451 1.1 jtc }
452 1.1 jtc
453 1.1 jtc /*
454 1.1 jtc * try the next entry on the chain
455 1.1 jtc */
456 1.1 jtc pt = pt->fow;
457 1.1 jtc }
458 1.1 jtc
459 1.1 jtc if (pt != NULL) {
460 1.1 jtc /*
461 1.1 jtc * found the file, compare the times, save the newer
462 1.1 jtc */
463 1.1 jtc if (arcn->sb.st_mtime > pt->mtime) {
464 1.1 jtc /*
465 1.1 jtc * file is newer
466 1.1 jtc */
467 1.1 jtc pt->mtime = arcn->sb.st_mtime;
468 1.1 jtc return(0);
469 1.1 jtc }
470 1.1 jtc /*
471 1.1 jtc * file is older
472 1.1 jtc */
473 1.1 jtc return(1);
474 1.1 jtc }
475 1.1 jtc }
476 1.1 jtc
477 1.1 jtc /*
478 1.1 jtc * not in table, add it
479 1.1 jtc */
480 1.1 jtc if ((pt = (FTM *)malloc(sizeof(FTM))) != NULL) {
481 1.1 jtc /*
482 1.1 jtc * add the name at the end of the scratch file, saving the
483 1.1 jtc * offset. add the file to the head of the hash chain
484 1.1 jtc */
485 1.1 jtc if ((pt->seek = lseek(ffd, (off_t)0, SEEK_END)) >= 0) {
486 1.1 jtc if (write(ffd, arcn->name, namelen) == namelen) {
487 1.1 jtc pt->mtime = arcn->sb.st_mtime;
488 1.1 jtc pt->namelen = namelen;
489 1.1 jtc pt->fow = ftab[indx];
490 1.1 jtc ftab[indx] = pt;
491 1.1 jtc return(0);
492 1.1 jtc }
493 1.1 jtc syswarn(1, errno, "Failed write to file time table");
494 1.1 jtc } else
495 1.1 jtc syswarn(1, errno, "Failed seek on file time table");
496 1.1 jtc } else
497 1.1 jtc warn(1, "File time table ran out of memory");
498 1.1 jtc
499 1.1 jtc if (pt != NULL)
500 1.1 jtc (void)free((char *)pt);
501 1.1 jtc return(-1);
502 1.1 jtc }
503 1.1 jtc
504 1.1 jtc /*
505 1.1 jtc * Interactive rename table routines
506 1.1 jtc *
507 1.1 jtc * The interactive rename table keeps track of the new names that the user
508 1.1 jtc * assignes to files from tty input. Since this map is unique for each file
509 1.1 jtc * we must store it in case there is a reference to the file later in archive
510 1.1 jtc * (a link). Otherwise we will be unable to find the file we know was
511 1.1 jtc * extracted. The remapping of these files is stored in a memory based hash
512 1.1 jtc * table (it is assumed since input must come from /dev/tty, it is unlikely to
513 1.1 jtc * be a very large table).
514 1.1 jtc */
515 1.1 jtc
516 1.1 jtc /*
517 1.1 jtc * name_start()
518 1.1 jtc * create the interactive rename table
519 1.1 jtc * Return:
520 1.1 jtc * 0 if successful, -1 otherwise
521 1.1 jtc */
522 1.1 jtc
523 1.1 jtc #if __STDC__
524 1.1 jtc int
525 1.1 jtc name_start(void)
526 1.1 jtc #else
527 1.1 jtc int
528 1.1 jtc name_start()
529 1.1 jtc #endif
530 1.1 jtc {
531 1.1 jtc if (ntab != NULL)
532 1.1 jtc return(0);
533 1.1 jtc if ((ntab = (NAMT **)calloc(N_TAB_SZ, sizeof(NAMT *))) == NULL) {
534 1.1 jtc warn(1, "Cannot allocate memory for interactive rename table");
535 1.1 jtc return(-1);
536 1.1 jtc }
537 1.1 jtc return(0);
538 1.1 jtc }
539 1.1 jtc
540 1.1 jtc /*
541 1.1 jtc * add_name()
542 1.1 jtc * add the new name to old name mapping just created by the user.
543 1.1 jtc * If an old name mapping is found (there may be duplicate names on an
544 1.1 jtc * archive) only the most recent is kept.
545 1.1 jtc * Return:
546 1.1 jtc * 0 if added, -1 otherwise
547 1.1 jtc */
548 1.1 jtc
549 1.1 jtc #if __STDC__
550 1.1 jtc int
551 1.1 jtc add_name(register char *oname, int onamelen, char *nname)
552 1.1 jtc #else
553 1.1 jtc int
554 1.1 jtc add_name(oname, onamelen, nname)
555 1.1 jtc register char *oname;
556 1.1 jtc int onamelen;
557 1.1 jtc char *nname;
558 1.1 jtc #endif
559 1.1 jtc {
560 1.1 jtc register NAMT *pt;
561 1.1 jtc register u_int indx;
562 1.1 jtc
563 1.1 jtc if (ntab == NULL) {
564 1.1 jtc /*
565 1.1 jtc * should never happen
566 1.1 jtc */
567 1.1 jtc warn(0, "No interactive rename table, links may fail\n");
568 1.1 jtc return(0);
569 1.1 jtc }
570 1.1 jtc
571 1.1 jtc /*
572 1.1 jtc * look to see if we have already mapped this file, if so we
573 1.1 jtc * will update it
574 1.1 jtc */
575 1.1 jtc indx = st_hash(oname, onamelen, N_TAB_SZ);
576 1.1 jtc if ((pt = ntab[indx]) != NULL) {
577 1.1 jtc /*
578 1.1 jtc * look down the has chain for the file
579 1.1 jtc */
580 1.1 jtc while ((pt != NULL) && (strcmp(oname, pt->oname) != 0))
581 1.1 jtc pt = pt->fow;
582 1.1 jtc
583 1.1 jtc if (pt != NULL) {
584 1.1 jtc /*
585 1.1 jtc * found an old mapping, replace it with the new one
586 1.1 jtc * the user just input (if it is different)
587 1.1 jtc */
588 1.1 jtc if (strcmp(nname, pt->nname) == 0)
589 1.1 jtc return(0);
590 1.1 jtc
591 1.1 jtc (void)free((char *)pt->nname);
592 1.1 jtc if ((pt->nname = strdup(nname)) == NULL) {
593 1.1 jtc warn(1, "Cannot update rename table");
594 1.1 jtc return(-1);
595 1.1 jtc }
596 1.1 jtc return(0);
597 1.1 jtc }
598 1.1 jtc }
599 1.1 jtc
600 1.1 jtc /*
601 1.1 jtc * this is a new mapping, add it to the table
602 1.1 jtc */
603 1.1 jtc if ((pt = (NAMT *)malloc(sizeof(NAMT))) != NULL) {
604 1.1 jtc if ((pt->oname = strdup(oname)) != NULL) {
605 1.1 jtc if ((pt->nname = strdup(nname)) != NULL) {
606 1.1 jtc pt->fow = ntab[indx];
607 1.1 jtc ntab[indx] = pt;
608 1.1 jtc return(0);
609 1.1 jtc }
610 1.1 jtc (void)free((char *)pt->oname);
611 1.1 jtc }
612 1.1 jtc (void)free((char *)pt);
613 1.1 jtc }
614 1.1 jtc warn(1, "Interactive rename table out of memory");
615 1.1 jtc return(-1);
616 1.1 jtc }
617 1.1 jtc
618 1.1 jtc /*
619 1.1 jtc * sub_name()
620 1.1 jtc * look up a link name to see if it points at a file that has been
621 1.1 jtc * remapped by the user. If found, the link is adjusted to contain the
622 1.1 jtc * new name (oname is the link to name)
623 1.1 jtc */
624 1.1 jtc
625 1.1 jtc #if __STDC__
626 1.1 jtc void
627 1.1 jtc sub_name(register char *oname, int *onamelen)
628 1.1 jtc #else
629 1.1 jtc void
630 1.1 jtc sub_name(oname, onamelen)
631 1.1 jtc register char *oname;
632 1.1 jtc int *onamelen;
633 1.1 jtc #endif
634 1.1 jtc {
635 1.1 jtc register NAMT *pt;
636 1.1 jtc register u_int indx;
637 1.1 jtc
638 1.1 jtc if (ntab == NULL)
639 1.1 jtc return;
640 1.1 jtc /*
641 1.1 jtc * look the name up in the hash table
642 1.1 jtc */
643 1.1 jtc indx = st_hash(oname, *onamelen, N_TAB_SZ);
644 1.1 jtc if ((pt = ntab[indx]) == NULL)
645 1.1 jtc return;
646 1.1 jtc
647 1.1 jtc while (pt != NULL) {
648 1.1 jtc /*
649 1.1 jtc * walk down the hash cahin looking for a match
650 1.1 jtc */
651 1.1 jtc if (strcmp(oname, pt->oname) == 0) {
652 1.1 jtc /*
653 1.1 jtc * found it, replace it with the new name
654 1.1 jtc * and return (we know that oname has enough space)
655 1.1 jtc */
656 1.1 jtc *onamelen = l_strncpy(oname, pt->nname, PAXPATHLEN+1);
657 1.1 jtc return;
658 1.1 jtc }
659 1.1 jtc pt = pt->fow;
660 1.1 jtc }
661 1.1 jtc
662 1.1 jtc /*
663 1.1 jtc * no match, just return
664 1.1 jtc */
665 1.1 jtc return;
666 1.1 jtc }
667 1.1 jtc
668 1.1 jtc /*
669 1.1 jtc * device/inode mapping table routines
670 1.1 jtc * (used with formats that store device and inodes fields)
671 1.1 jtc *
672 1.1 jtc * device/inode mapping tables remap the device field in a archive header. The
673 1.1 jtc * device/inode fields are used to determine when files are hard links to each
674 1.1 jtc * other. However these values have very little meaning outside of that. This
675 1.1 jtc * database is used to solve one of two different problems.
676 1.1 jtc *
677 1.1 jtc * 1) when files are appended to an archive, while the new files may have hard
678 1.1 jtc * links to each other, you cannot determine if they have hard links to any
679 1.1 jtc * file already stored on the archive from a prior run of pax. We must assume
680 1.1 jtc * that these inode/device pairs are unique only within a SINGLE run of pax
681 1.1 jtc * (which adds a set of files to an archive). So we have to make sure the
682 1.1 jtc * inode/dev pairs we add each time are always unique. We do this by observing
683 1.1 jtc * while the inode field is very dense, the use of the dev field is fairly
684 1.1 jtc * sparse. Within each run of pax, we remap any device number of a new archive
685 1.1 jtc * member that has a device number used in a prior run and already stored in a
686 1.1 jtc * file on the archive. During the read phase of the append, we store the
687 1.1 jtc * device numbers used and mark them to not be used by any file during the
688 1.1 jtc * write phase. If during write we go to use one of those old device numbers,
689 1.1 jtc * we remap it to a new value.
690 1.1 jtc *
691 1.1 jtc * 2) Often the fields in the archive header used to store these values are
692 1.1 jtc * too small to store the entire value. The result is an inode or device value
693 1.1 jtc * which can be truncated. This really can foul up an archive. With truncation
694 1.1 jtc * we end up creating links between files that are really not links (after
695 1.1 jtc * truncation the inodes are the same value). We address that by detecting
696 1.1 jtc * truncation and forcing a remap of the device field to split truncated
697 1.1 jtc * inodes away from each other. Each truncation creates a pattern of bits that
698 1.1 jtc * are removed. We use this pattern of truncated bits to partition the inodes
699 1.1 jtc * on a single device to many different devices (each one represented by the
700 1.1 jtc * truncated bit pattern). All inodes on the same device that have the same
701 1.1 jtc * truncation pattern are mapped to the same new device. Two inodes that
702 1.1 jtc * truncate to the same value clearly will always have different truncation
703 1.1 jtc * bit patterns, so they will be split from away each other. When we spot
704 1.1 jtc * device truncation we remap the device number to a non truncated value.
705 1.1 jtc * (for more info see table.h for the data structures involved).
706 1.1 jtc */
707 1.1 jtc
708 1.1 jtc /*
709 1.1 jtc * dev_start()
710 1.1 jtc * create the device mapping table
711 1.1 jtc * Return:
712 1.1 jtc * 0 if successful, -1 otherwise
713 1.1 jtc */
714 1.1 jtc
715 1.1 jtc #if __STDC__
716 1.1 jtc int
717 1.1 jtc dev_start(void)
718 1.1 jtc #else
719 1.1 jtc int
720 1.1 jtc dev_start()
721 1.1 jtc #endif
722 1.1 jtc {
723 1.1 jtc if (dtab != NULL)
724 1.1 jtc return(0);
725 1.1 jtc if ((dtab = (DEVT **)calloc(D_TAB_SZ, sizeof(DEVT *))) == NULL) {
726 1.1 jtc warn(1, "Cannot allocate memory for device mapping table");
727 1.1 jtc return(-1);
728 1.1 jtc }
729 1.1 jtc return(0);
730 1.1 jtc }
731 1.1 jtc
732 1.1 jtc /*
733 1.1 jtc * add_dev()
734 1.1 jtc * add a device number to the table. this will force the device to be
735 1.1 jtc * remapped to a new value if it be used during a write phase. This
736 1.1 jtc * function is called during the read phase of an append to prohibit the
737 1.1 jtc * use of any device number already in the archive.
738 1.1 jtc * Return:
739 1.1 jtc * 0 if added ok, -1 otherwise
740 1.1 jtc */
741 1.1 jtc
742 1.1 jtc #if __STDC__
743 1.1 jtc int
744 1.1 jtc add_dev(register ARCHD *arcn)
745 1.1 jtc #else
746 1.1 jtc int
747 1.1 jtc add_dev(arcn)
748 1.1 jtc register ARCHD *arcn;
749 1.1 jtc #endif
750 1.1 jtc {
751 1.1 jtc if (chk_dev(arcn->sb.st_dev, 1) == NULL)
752 1.1 jtc return(-1);
753 1.1 jtc return(0);
754 1.1 jtc }
755 1.1 jtc
756 1.1 jtc /*
757 1.1 jtc * chk_dev()
758 1.1 jtc * check for a device value in the device table. If not found and the add
759 1.1 jtc * flag is set, it is added. This does NOT assign any mapping values, just
760 1.1 jtc * adds the device number as one that need to be remapped. If this device
761 1.1 jtc * is alread mapped, just return with a pointer to that entry.
762 1.1 jtc * Return:
763 1.1 jtc * pointer to the entry for this device in the device map table. Null
764 1.1 jtc * if the add flag is not set and the device is not in the table (it is
765 1.1 jtc * not been seen yet). If add is set and the device cannot be added, null
766 1.1 jtc * is returned (indicates an error).
767 1.1 jtc */
768 1.1 jtc
769 1.1 jtc #if __STDC__
770 1.1 jtc static DEVT *
771 1.1 jtc chk_dev(dev_t dev, int add)
772 1.1 jtc #else
773 1.1 jtc static DEVT *
774 1.1 jtc chk_dev(dev, add)
775 1.1 jtc dev_t dev;
776 1.1 jtc int add;
777 1.1 jtc #endif
778 1.1 jtc {
779 1.1 jtc register DEVT *pt;
780 1.1 jtc register u_int indx;
781 1.1 jtc
782 1.1 jtc if (dtab == NULL)
783 1.1 jtc return(NULL);
784 1.1 jtc /*
785 1.1 jtc * look to see if this device is already in the table
786 1.1 jtc */
787 1.1 jtc indx = ((unsigned)dev) % D_TAB_SZ;
788 1.1 jtc if ((pt = dtab[indx]) != NULL) {
789 1.1 jtc while ((pt != NULL) && (pt->dev != dev))
790 1.1 jtc pt = pt->fow;
791 1.1 jtc
792 1.1 jtc /*
793 1.1 jtc * found it, return a pointer to it
794 1.1 jtc */
795 1.1 jtc if (pt != NULL)
796 1.1 jtc return(pt);
797 1.1 jtc }
798 1.1 jtc
799 1.1 jtc /*
800 1.1 jtc * not in table, we add it only if told to as this may just be a check
801 1.1 jtc * to see if a device number is being used.
802 1.1 jtc */
803 1.1 jtc if (add == 0)
804 1.1 jtc return(NULL);
805 1.1 jtc
806 1.1 jtc /*
807 1.1 jtc * allocate a node for this device and add it to the front of the hash
808 1.1 jtc * chain. Note we do not assign remaps values here, so the pt->list
809 1.1 jtc * list must be NULL.
810 1.1 jtc */
811 1.1 jtc if ((pt = (DEVT *)malloc(sizeof(DEVT))) == NULL) {
812 1.1 jtc warn(1, "Device map table out of memory");
813 1.1 jtc return(NULL);
814 1.1 jtc }
815 1.1 jtc pt->dev = dev;
816 1.1 jtc pt->list = NULL;
817 1.1 jtc pt->fow = dtab[indx];
818 1.1 jtc dtab[indx] = pt;
819 1.1 jtc return(pt);
820 1.1 jtc }
821 1.1 jtc /*
822 1.1 jtc * map_dev()
823 1.1 jtc * given an inode and device storage mask (the mask has a 1 for each bit
824 1.1 jtc * the archive format is able to store in a header), we check for inode
825 1.1 jtc * and device truncation and remap the device as required. Device mapping
826 1.1 jtc * can also occur when during the read phase of append a device number was
827 1.1 jtc * seen (and was marked as do not use during the write phase). WE ASSUME
828 1.1 jtc * that unsigned longs are the same size or bigger than the fields used
829 1.1 jtc * for ino_t and dev_t. If not the types will have to be changed.
830 1.1 jtc * Return:
831 1.1 jtc * 0 if all ok, -1 otherwise.
832 1.1 jtc */
833 1.1 jtc
834 1.1 jtc #if __STDC__
835 1.1 jtc int
836 1.1 jtc map_dev(register ARCHD *arcn, u_long dev_mask, u_long ino_mask)
837 1.1 jtc #else
838 1.1 jtc int
839 1.1 jtc map_dev(arcn, dev_mask, ino_mask)
840 1.1 jtc register ARCHD *arcn;
841 1.1 jtc u_long dev_mask;
842 1.1 jtc u_long ino_mask;
843 1.1 jtc #endif
844 1.1 jtc {
845 1.1 jtc register DEVT *pt;
846 1.1 jtc register DLIST *dpt;
847 1.1 jtc static dev_t lastdev = 0; /* next device number to try */
848 1.1 jtc int trc_ino = 0;
849 1.1 jtc int trc_dev = 0;
850 1.1 jtc ino_t trunc_bits = 0;
851 1.1 jtc ino_t nino;
852 1.1 jtc
853 1.1 jtc if (dtab == NULL)
854 1.1 jtc return(0);
855 1.1 jtc /*
856 1.1 jtc * check for device and inode truncation, and extract the truncated
857 1.1 jtc * bit pattern.
858 1.1 jtc */
859 1.1 jtc if ((arcn->sb.st_dev & (dev_t)dev_mask) != arcn->sb.st_dev)
860 1.1 jtc ++trc_dev;
861 1.1 jtc if ((nino = arcn->sb.st_ino & (ino_t)ino_mask) != arcn->sb.st_ino) {
862 1.1 jtc ++trc_ino;
863 1.1 jtc trunc_bits = arcn->sb.st_ino & (ino_t)(~ino_mask);
864 1.1 jtc }
865 1.1 jtc
866 1.1 jtc /*
867 1.1 jtc * see if this device is already being mapped, look up the device
868 1.1 jtc * then find the truncation bit pattern which applies
869 1.1 jtc */
870 1.1 jtc if ((pt = chk_dev(arcn->sb.st_dev, 0)) != NULL) {
871 1.1 jtc /*
872 1.1 jtc * this device is already marked to be remapped
873 1.1 jtc */
874 1.1 jtc for (dpt = pt->list; dpt != NULL; dpt = dpt->fow)
875 1.1 jtc if (dpt->trunc_bits == trunc_bits)
876 1.1 jtc break;
877 1.1 jtc
878 1.1 jtc if (dpt != NULL) {
879 1.1 jtc /*
880 1.1 jtc * we are being remapped for this device and pattern
881 1.1 jtc * change the device number to be stored and return
882 1.1 jtc */
883 1.1 jtc arcn->sb.st_dev = dpt->dev;
884 1.1 jtc arcn->sb.st_ino = nino;
885 1.1 jtc return(0);
886 1.1 jtc }
887 1.1 jtc } else {
888 1.1 jtc /*
889 1.1 jtc * this device is not being remapped YET. if we do not have any
890 1.1 jtc * form of truncation, we do not need a remap
891 1.1 jtc */
892 1.1 jtc if (!trc_ino && !trc_dev)
893 1.1 jtc return(0);
894 1.1 jtc
895 1.1 jtc /*
896 1.1 jtc * we have truncation, have to add this as a device to remap
897 1.1 jtc */
898 1.1 jtc if ((pt = chk_dev(arcn->sb.st_dev, 1)) == NULL)
899 1.1 jtc goto bad;
900 1.1 jtc
901 1.1 jtc /*
902 1.1 jtc * if we just have a truncated inode, we have to make sure that
903 1.1 jtc * all future inodes that do not truncate (they have the
904 1.1 jtc * truncation pattern of all 0's) continue to map to the same
905 1.1 jtc * device number. We probably have already written inodes with
906 1.1 jtc * this device number to the archive with the truncation
907 1.1 jtc * pattern of all 0's. So we add the mapping for all 0's to the
908 1.1 jtc * same device number.
909 1.1 jtc */
910 1.1 jtc if (!trc_dev && (trunc_bits != 0)) {
911 1.1 jtc if ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL)
912 1.1 jtc goto bad;
913 1.1 jtc dpt->trunc_bits = 0;
914 1.1 jtc dpt->dev = arcn->sb.st_dev;
915 1.1 jtc dpt->fow = pt->list;
916 1.1 jtc pt->list = dpt;
917 1.1 jtc }
918 1.1 jtc }
919 1.1 jtc
920 1.1 jtc /*
921 1.1 jtc * look for a device number not being used. We must watch for wrap
922 1.1 jtc * around on lastdev (so we do not get stuck looking forever!)
923 1.1 jtc */
924 1.1 jtc while (++lastdev > 0) {
925 1.1 jtc if (chk_dev(lastdev, 0) != NULL)
926 1.1 jtc continue;
927 1.1 jtc /*
928 1.1 jtc * found an unused value. If we have reached truncation point
929 1.1 jtc * for this format we are hosed, so we give up. Otherwise we
930 1.1 jtc * mark it as being used.
931 1.1 jtc */
932 1.1 jtc if (((lastdev & ((dev_t)dev_mask)) != lastdev) ||
933 1.1 jtc (chk_dev(lastdev, 1) == NULL))
934 1.1 jtc goto bad;
935 1.1 jtc break;
936 1.1 jtc }
937 1.1 jtc
938 1.1 jtc if ((lastdev <= 0) || ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL))
939 1.1 jtc goto bad;
940 1.1 jtc
941 1.1 jtc /*
942 1.1 jtc * got a new device number, store it under this truncation pattern.
943 1.1 jtc * change the device number this file is being stored with.
944 1.1 jtc */
945 1.1 jtc dpt->trunc_bits = trunc_bits;
946 1.1 jtc dpt->dev = lastdev;
947 1.1 jtc dpt->fow = pt->list;
948 1.1 jtc pt->list = dpt;
949 1.1 jtc arcn->sb.st_dev = lastdev;
950 1.1 jtc arcn->sb.st_ino = nino;
951 1.1 jtc return(0);
952 1.1 jtc
953 1.1 jtc bad:
954 1.1 jtc warn(1, "Unable to fix truncated inode/device field when storing %s",
955 1.1 jtc arcn->name);
956 1.1 jtc warn(0, "Archive may create improper hard links when extracted");
957 1.1 jtc return(0);
958 1.1 jtc }
959 1.1 jtc
960 1.1 jtc /*
961 1.1 jtc * directory access/mod time reset table routines (for directories READ by pax)
962 1.1 jtc *
963 1.1 jtc * The pax -t flag requires that access times of archive files to be the same
964 1.1 jtc * before being read by pax. For regular files, access time is restored after
965 1.1 jtc * the file has been copied. This database provides the same functionality for
966 1.1 jtc * directories read during file tree traversal. Restoring directory access time
967 1.1 jtc * is more complex than files since directories may be read several times until
968 1.1 jtc * all the descendants in their subtree are visited by fts. Directory access
969 1.1 jtc * and modification times are stored during the fts pre-order visit (done
970 1.1 jtc * before any descendants in the subtree is visited) and restored after the
971 1.1 jtc * fts post-order visit (after all the descendants have been visited). In the
972 1.1 jtc * case of premature exit from a subtree (like from the effects of -n), any
973 1.1 jtc * directory entries left in this database are reset during final cleanup
974 1.1 jtc * operations of pax. Entries are hashed by inode number for fast lookup.
975 1.1 jtc */
976 1.1 jtc
977 1.1 jtc /*
978 1.1 jtc * atdir_start()
979 1.1 jtc * create the directory access time database for directories READ by pax.
980 1.1 jtc * Return:
981 1.1 jtc * 0 is created ok, -1 otherwise.
982 1.1 jtc */
983 1.1 jtc
984 1.1 jtc #if __STDC__
985 1.1 jtc int
986 1.1 jtc atdir_start(void)
987 1.1 jtc #else
988 1.1 jtc int
989 1.1 jtc atdir_start()
990 1.1 jtc #endif
991 1.1 jtc {
992 1.1 jtc if (atab != NULL)
993 1.1 jtc return(0);
994 1.1 jtc if ((atab = (ATDIR **)calloc(A_TAB_SZ, sizeof(ATDIR *))) == NULL) {
995 1.1 jtc warn(1,"Cannot allocate space for directory access time table");
996 1.1 jtc return(-1);
997 1.1 jtc }
998 1.1 jtc return(0);
999 1.1 jtc }
1000 1.1 jtc
1001 1.1 jtc
1002 1.1 jtc /*
1003 1.1 jtc * atdir_end()
1004 1.1 jtc * walk through the directory access time table and reset the access time
1005 1.1 jtc * of any directory who still has an entry left in the database. These
1006 1.1 jtc * entries are for directories READ by pax
1007 1.1 jtc */
1008 1.1 jtc
1009 1.1 jtc #if __STDC__
1010 1.1 jtc void
1011 1.1 jtc atdir_end(void)
1012 1.1 jtc #else
1013 1.1 jtc void
1014 1.1 jtc atdir_end()
1015 1.1 jtc #endif
1016 1.1 jtc {
1017 1.1 jtc register ATDIR *pt;
1018 1.1 jtc register int i;
1019 1.1 jtc
1020 1.1 jtc if (atab == NULL)
1021 1.1 jtc return;
1022 1.1 jtc /*
1023 1.1 jtc * for each non-empty hash table entry reset all the directories
1024 1.1 jtc * chained there.
1025 1.1 jtc */
1026 1.1 jtc for (i = 0; i < A_TAB_SZ; ++i) {
1027 1.1 jtc if ((pt = atab[i]) == NULL)
1028 1.1 jtc continue;
1029 1.1 jtc /*
1030 1.1 jtc * remember to force the times, set_ftime() looks at pmtime
1031 1.1 jtc * and patime, which only applies to things CREATED by pax,
1032 1.1 jtc * not read by pax. Read time reset is controlled by -t.
1033 1.1 jtc */
1034 1.1 jtc for (; pt != NULL; pt = pt->fow)
1035 1.1 jtc set_ftime(pt->name, pt->mtime, pt->atime, 1);
1036 1.1 jtc }
1037 1.1 jtc }
1038 1.1 jtc
1039 1.1 jtc /*
1040 1.1 jtc * add_atdir()
1041 1.1 jtc * add a directory to the directory access time table. Table is hashed
1042 1.1 jtc * and chained by inode number. This is for directories READ by pax
1043 1.1 jtc */
1044 1.1 jtc
1045 1.1 jtc #if __STDC__
1046 1.1 jtc void
1047 1.1 jtc add_atdir(char *fname, dev_t dev, ino_t ino, time_t mtime, time_t atime)
1048 1.1 jtc #else
1049 1.1 jtc void
1050 1.1 jtc add_atdir(fname, dev, ino, mtime, atime)
1051 1.1 jtc char *fname;
1052 1.1 jtc dev_t dev;
1053 1.1 jtc ino_t ino;
1054 1.1 jtc time_t mtime;
1055 1.1 jtc time_t atime;
1056 1.1 jtc #endif
1057 1.1 jtc {
1058 1.1 jtc register ATDIR *pt;
1059 1.1 jtc register u_int indx;
1060 1.1 jtc
1061 1.1 jtc if (atab == NULL)
1062 1.1 jtc return;
1063 1.1 jtc
1064 1.1 jtc /*
1065 1.1 jtc * make sure this directory is not already in the table, if so just
1066 1.1 jtc * return (the older entry always has the correct time). The only
1067 1.1 jtc * way this will happen is when the same subtree can be traversed by
1068 1.1 jtc * different args to pax and the -n option is aborting fts out of a
1069 1.1 jtc * subtree before all the post-order visits have been made).
1070 1.1 jtc */
1071 1.1 jtc indx = ((unsigned)ino) % A_TAB_SZ;
1072 1.1 jtc if ((pt = atab[indx]) != NULL) {
1073 1.1 jtc while (pt != NULL) {
1074 1.1 jtc if ((pt->ino == ino) && (pt->dev == dev))
1075 1.1 jtc break;
1076 1.1 jtc pt = pt->fow;
1077 1.1 jtc }
1078 1.1 jtc
1079 1.1 jtc /*
1080 1.1 jtc * oops, already there. Leave it alone.
1081 1.1 jtc */
1082 1.1 jtc if (pt != NULL)
1083 1.1 jtc return;
1084 1.1 jtc }
1085 1.1 jtc
1086 1.1 jtc /*
1087 1.1 jtc * add it to the front of the hash chain
1088 1.1 jtc */
1089 1.1 jtc if ((pt = (ATDIR *)malloc(sizeof(ATDIR))) != NULL) {
1090 1.1 jtc if ((pt->name = strdup(fname)) != NULL) {
1091 1.1 jtc pt->dev = dev;
1092 1.1 jtc pt->ino = ino;
1093 1.1 jtc pt->mtime = mtime;
1094 1.1 jtc pt->atime = atime;
1095 1.1 jtc pt->fow = atab[indx];
1096 1.1 jtc atab[indx] = pt;
1097 1.1 jtc return;
1098 1.1 jtc }
1099 1.1 jtc (void)free((char *)pt);
1100 1.1 jtc }
1101 1.1 jtc
1102 1.1 jtc warn(1, "Directory access time reset table ran out of memory");
1103 1.1 jtc return;
1104 1.1 jtc }
1105 1.1 jtc
1106 1.1 jtc /*
1107 1.1 jtc * get_atdir()
1108 1.1 jtc * look up a directory by inode and device number to obtain the access
1109 1.1 jtc * and modification time you want to set to. If found, the modification
1110 1.1 jtc * and access time parameters are set and the entry is removed from the
1111 1.1 jtc * table (as it is no longer needed). These are for directories READ by
1112 1.1 jtc * pax
1113 1.1 jtc * Return:
1114 1.1 jtc * 0 if found, -1 if not found.
1115 1.1 jtc */
1116 1.1 jtc
1117 1.1 jtc #if __STDC__
1118 1.1 jtc int
1119 1.1 jtc get_atdir(dev_t dev, ino_t ino, time_t *mtime, time_t *atime)
1120 1.1 jtc #else
1121 1.1 jtc int
1122 1.1 jtc get_atdir(dev, ino, mtime, atime)
1123 1.1 jtc dev_t dev;
1124 1.1 jtc ino_t ino;
1125 1.1 jtc time_t *mtime;
1126 1.1 jtc time_t *atime;
1127 1.1 jtc #endif
1128 1.1 jtc {
1129 1.1 jtc register ATDIR *pt;
1130 1.1 jtc register ATDIR **ppt;
1131 1.1 jtc register u_int indx;
1132 1.1 jtc
1133 1.1 jtc if (atab == NULL)
1134 1.1 jtc return(-1);
1135 1.1 jtc /*
1136 1.1 jtc * hash by inode and search the chain for an inode and device match
1137 1.1 jtc */
1138 1.1 jtc indx = ((unsigned)ino) % A_TAB_SZ;
1139 1.1 jtc if ((pt = atab[indx]) == NULL)
1140 1.1 jtc return(-1);
1141 1.1 jtc
1142 1.1 jtc ppt = &(atab[indx]);
1143 1.1 jtc while (pt != NULL) {
1144 1.1 jtc if ((pt->ino == ino) && (pt->dev == dev))
1145 1.1 jtc break;
1146 1.1 jtc /*
1147 1.1 jtc * no match, go to next one
1148 1.1 jtc */
1149 1.1 jtc ppt = &(pt->fow);
1150 1.1 jtc pt = pt->fow;
1151 1.1 jtc }
1152 1.1 jtc
1153 1.1 jtc /*
1154 1.1 jtc * return if we did not find it.
1155 1.1 jtc */
1156 1.1 jtc if (pt == NULL)
1157 1.1 jtc return(-1);
1158 1.1 jtc
1159 1.1 jtc /*
1160 1.1 jtc * found it. return the times and remove the entry from the table.
1161 1.1 jtc */
1162 1.1 jtc *ppt = pt->fow;
1163 1.1 jtc *mtime = pt->mtime;
1164 1.1 jtc *atime = pt->atime;
1165 1.1 jtc (void)free((char *)pt->name);
1166 1.1 jtc (void)free((char *)pt);
1167 1.1 jtc return(0);
1168 1.1 jtc }
1169 1.1 jtc
1170 1.1 jtc /*
1171 1.1 jtc * directory access mode and time storage routines (for directories CREATED
1172 1.1 jtc * by pax).
1173 1.1 jtc *
1174 1.1 jtc * Pax requires that extracted directories, by default, have their access/mod
1175 1.1 jtc * times and permissions set to the values specified in the archive. During the
1176 1.1 jtc * actions of extracting (and creating the destination subtree during -rw copy)
1177 1.1 jtc * directories extracted may be modified after being created. Even worse is
1178 1.1 jtc * that these directories may have been created with file permissions which
1179 1.1 jtc * prohibits any descendants of these directories from being extracted. When
1180 1.1 jtc * directories are created by pax, access rights may be added to permit the
1181 1.1 jtc * creation of files in their subtree. Every time pax creates a directory, the
1182 1.1 jtc * times and file permissions specified by the archive are stored. After all
1183 1.1 jtc * files have been extracted (or copied), these directories have their times
1184 1.1 jtc * and file modes reset to the stored values. The directory info is restored in
1185 1.1 jtc * reverse order as entries were added to the data file from root to leaf. To
1186 1.1 jtc * restore atime properly, we must go backwards. The data file consists of
1187 1.1 jtc * records with two parts, the file name followed by a DIRDATA trailer. The
1188 1.1 jtc * fixed sized trailer contains the size of the name plus the off_t location in
1189 1.1 jtc * the file. To restore we work backwards through the file reading the trailer
1190 1.1 jtc * then the file name.
1191 1.1 jtc */
1192 1.1 jtc
1193 1.1 jtc /*
1194 1.1 jtc * dir_start()
1195 1.1 jtc * set up the directory time and file mode storage for directories CREATED
1196 1.1 jtc * by pax.
1197 1.1 jtc * Return:
1198 1.1 jtc * 0 if ok, -1 otherwise
1199 1.1 jtc */
1200 1.1 jtc
1201 1.1 jtc #if __STDC__
1202 1.1 jtc int
1203 1.1 jtc dir_start(void)
1204 1.1 jtc #else
1205 1.1 jtc int
1206 1.1 jtc dir_start()
1207 1.1 jtc #endif
1208 1.1 jtc {
1209 1.1 jtc char *pt;
1210 1.1 jtc
1211 1.1 jtc if (dirfd != -1)
1212 1.1 jtc return(0);
1213 1.1 jtc if ((pt = tempnam((char *)NULL, (char *)NULL)) == NULL)
1214 1.1 jtc return(-1);
1215 1.1 jtc
1216 1.1 jtc /*
1217 1.1 jtc * unlink the file so it goes away at termination by itself
1218 1.1 jtc */
1219 1.1 jtc (void)unlink(pt);
1220 1.1 jtc if ((dirfd = open(pt, O_RDWR|O_CREAT, 0600)) >= 0) {
1221 1.1 jtc (void)unlink(pt);
1222 1.1 jtc return(0);
1223 1.1 jtc }
1224 1.1 jtc warn(1, "Unable to create temporary file for directory times: %s", pt);
1225 1.1 jtc return(-1);
1226 1.1 jtc }
1227 1.1 jtc
1228 1.1 jtc /*
1229 1.1 jtc * add_dir()
1230 1.1 jtc * add the mode and times for a newly CREATED directory
1231 1.1 jtc * name is name of the directory, psb the stat buffer with the data in it,
1232 1.1 jtc * frc_mode is a flag that says whether to force the setting of the mode
1233 1.1 jtc * (ignoring the user set values for preserving file mode). Frc_mode is
1234 1.1 jtc * for the case where we created a file and found that the resulting
1235 1.1 jtc * directory was not writeable and the user asked for file modes to NOT
1236 1.1 jtc * be preserved. (we have to preserve what was created by default, so we
1237 1.1 jtc * have to force the setting at the end. this is stated explicitly in the
1238 1.1 jtc * pax spec)
1239 1.1 jtc */
1240 1.1 jtc
1241 1.1 jtc #if __STDC__
1242 1.1 jtc void
1243 1.1 jtc add_dir(char *name, int nlen, struct stat *psb, int frc_mode)
1244 1.1 jtc #else
1245 1.1 jtc void
1246 1.1 jtc add_dir(name, nlen, psb, frc_mode)
1247 1.1 jtc char *name;
1248 1.1 jtc int nlen;
1249 1.1 jtc struct stat *psb;
1250 1.1 jtc int frc_mode;
1251 1.1 jtc #endif
1252 1.1 jtc {
1253 1.1 jtc DIRDATA dblk;
1254 1.1 jtc
1255 1.1 jtc if (dirfd < 0)
1256 1.1 jtc return;
1257 1.1 jtc
1258 1.1 jtc /*
1259 1.1 jtc * get current position (where file name will start) so we can store it
1260 1.1 jtc * in the trailer
1261 1.1 jtc */
1262 1.1 jtc if ((dblk.npos = lseek(dirfd, 0L, SEEK_CUR)) < 0) {
1263 1.1 jtc warn(1,"Unable to store mode and times for directory: %s",name);
1264 1.1 jtc return;
1265 1.1 jtc }
1266 1.1 jtc
1267 1.1 jtc /*
1268 1.1 jtc * write the file name followed by the trailer
1269 1.1 jtc */
1270 1.1 jtc dblk.nlen = nlen + 1;
1271 1.1 jtc dblk.mode = psb->st_mode & 0xffff;
1272 1.1 jtc dblk.mtime = psb->st_mtime;
1273 1.1 jtc dblk.atime = psb->st_atime;
1274 1.1 jtc dblk.frc_mode = frc_mode;
1275 1.1 jtc if ((write(dirfd, name, dblk.nlen) == dblk.nlen) &&
1276 1.1 jtc (write(dirfd, (char *)&dblk, sizeof(dblk)) == sizeof(dblk))) {
1277 1.1 jtc ++dircnt;
1278 1.1 jtc return;
1279 1.1 jtc }
1280 1.1 jtc
1281 1.1 jtc warn(1,"Unable to store mode and times for created directory: %s",name);
1282 1.1 jtc return;
1283 1.1 jtc }
1284 1.1 jtc
1285 1.1 jtc /*
1286 1.1 jtc * proc_dir()
1287 1.1 jtc * process all file modes and times stored for directories CREATED
1288 1.1 jtc * by pax
1289 1.1 jtc */
1290 1.1 jtc
1291 1.1 jtc #if __STDC__
1292 1.1 jtc void
1293 1.1 jtc proc_dir(void)
1294 1.1 jtc #else
1295 1.1 jtc void
1296 1.1 jtc proc_dir()
1297 1.1 jtc #endif
1298 1.1 jtc {
1299 1.1 jtc char name[PAXPATHLEN+1];
1300 1.1 jtc DIRDATA dblk;
1301 1.1 jtc u_long cnt;
1302 1.1 jtc
1303 1.1 jtc if (dirfd < 0)
1304 1.1 jtc return;
1305 1.1 jtc /*
1306 1.1 jtc * read backwards through the file and process each directory
1307 1.1 jtc */
1308 1.1 jtc for (cnt = 0; cnt < dircnt; ++cnt) {
1309 1.1 jtc /*
1310 1.1 jtc * read the trailer, then the file name, if this fails
1311 1.1 jtc * just give up.
1312 1.1 jtc */
1313 1.1 jtc if (lseek(dirfd, -((off_t)sizeof(dblk)), SEEK_CUR) < 0)
1314 1.1 jtc break;
1315 1.1 jtc if (read(dirfd,(char *)&dblk, sizeof(dblk)) != sizeof(dblk))
1316 1.1 jtc break;
1317 1.1 jtc if (lseek(dirfd, dblk.npos, SEEK_SET) < 0)
1318 1.1 jtc break;
1319 1.1 jtc if (read(dirfd, name, dblk.nlen) != dblk.nlen)
1320 1.1 jtc break;
1321 1.1 jtc if (lseek(dirfd, dblk.npos, SEEK_SET) < 0)
1322 1.1 jtc break;
1323 1.1 jtc
1324 1.1 jtc /*
1325 1.1 jtc * frc_mode set, make sure we set the file modes even if
1326 1.1 jtc * the user didn't ask for it (see file_subs.c for more info)
1327 1.1 jtc */
1328 1.1 jtc if (pmode || dblk.frc_mode)
1329 1.1 jtc set_pmode(name, dblk.mode);
1330 1.1 jtc if (patime || pmtime)
1331 1.1 jtc set_ftime(name, dblk.mtime, dblk.atime, 0);
1332 1.1 jtc }
1333 1.1 jtc
1334 1.1 jtc (void)close(dirfd);
1335 1.1 jtc dirfd = -1;
1336 1.1 jtc if (cnt != dircnt)
1337 1.1 jtc warn(1,"Unable to set mode and times for created directories");
1338 1.1 jtc return;
1339 1.1 jtc }
1340 1.1 jtc
1341 1.1 jtc /*
1342 1.1 jtc * database independent routines
1343 1.1 jtc */
1344 1.1 jtc
1345 1.1 jtc /*
1346 1.1 jtc * st_hash()
1347 1.1 jtc * hashes filenames to a u_int for hashing into a table. Looks at the tail
1348 1.1 jtc * end of file, as this provides far better distribution than any other
1349 1.1 jtc * part of the name. For performance reasons we only care about the last
1350 1.1 jtc * MAXKEYLEN chars (should be at LEAST large enough to pick off the file
1351 1.1 jtc * name). Was tested on 500,000 name file tree traversal from the root
1352 1.1 jtc * and gave almost a perfectly uniform distribution of keys when used with
1353 1.1 jtc * prime sized tables (MAXKEYLEN was 128 in test). Hashes (sizeof int)
1354 1.1 jtc * chars at a time and pads with 0 for last addition.
1355 1.1 jtc * Return:
1356 1.1 jtc * the hash value of the string MOD (%) the table size.
1357 1.1 jtc */
1358 1.1 jtc
1359 1.1 jtc #if __STDC__
1360 1.1 jtc u_int
1361 1.1 jtc st_hash(char *name, int len, int tabsz)
1362 1.1 jtc #else
1363 1.1 jtc u_int
1364 1.1 jtc st_hash(name, len, tabsz)
1365 1.1 jtc char *name;
1366 1.1 jtc int len;
1367 1.1 jtc int tabsz;
1368 1.1 jtc #endif
1369 1.1 jtc {
1370 1.1 jtc register char *pt;
1371 1.1 jtc register char *dest;
1372 1.1 jtc register char *end;
1373 1.1 jtc register int i;
1374 1.1 jtc register u_int key = 0;
1375 1.1 jtc register int steps;
1376 1.1 jtc register int res;
1377 1.1 jtc u_int val;
1378 1.1 jtc
1379 1.1 jtc /*
1380 1.1 jtc * only look at the tail up to MAXKEYLEN, we do not need to waste
1381 1.1 jtc * time here (remember these are pathnames, the tail is what will
1382 1.1 jtc * spread out the keys)
1383 1.1 jtc */
1384 1.1 jtc if (len > MAXKEYLEN) {
1385 1.1 jtc pt = &(name[len - MAXKEYLEN]);
1386 1.1 jtc len = MAXKEYLEN;
1387 1.1 jtc } else
1388 1.1 jtc pt = name;
1389 1.1 jtc
1390 1.1 jtc /*
1391 1.1 jtc * calculate the number of u_int size steps in the string and if
1392 1.1 jtc * there is a runt to deal with
1393 1.1 jtc */
1394 1.1 jtc steps = len/sizeof(u_int);
1395 1.1 jtc res = len % sizeof(u_int);
1396 1.1 jtc
1397 1.1 jtc /*
1398 1.1 jtc * add up the value of the string in unsigned integer sized pieces
1399 1.1 jtc * too bad we cannot have unsigned int aligned strings, then we
1400 1.1 jtc * could avoid the expensive copy.
1401 1.1 jtc */
1402 1.1 jtc for (i = 0; i < steps; ++i) {
1403 1.1 jtc end = pt + sizeof(u_int);
1404 1.1 jtc dest = (char *)&val;
1405 1.1 jtc while (pt < end)
1406 1.1 jtc *dest++ = *pt++;
1407 1.1 jtc key += val;
1408 1.1 jtc }
1409 1.1 jtc
1410 1.1 jtc /*
1411 1.1 jtc * add in the runt padded with zero to the right
1412 1.1 jtc */
1413 1.1 jtc if (res) {
1414 1.1 jtc val = 0;
1415 1.1 jtc end = pt + res;
1416 1.1 jtc dest = (char *)&val;
1417 1.1 jtc while (pt < end)
1418 1.1 jtc *dest++ = *pt++;
1419 1.1 jtc key += val;
1420 1.1 jtc }
1421 1.1 jtc
1422 1.1 jtc /*
1423 1.1 jtc * return the result mod the table size
1424 1.1 jtc */
1425 1.1 jtc return(key % tabsz);
1426 1.1 jtc }
1427