1 /*
2 * dirent_uri.c: a library to manipulate URIs and directory entries.
3 *
4 * ====================================================================
5 * Licensed to the Apache Software Foundation (ASF) under one
6 * or more contributor license agreements. See the NOTICE file
7 * distributed with this work for additional information
8 * regarding copyright ownership. The ASF licenses this file
9 * to you under the Apache License, Version 2.0 (the
10 * "License"); you may not use this file except in compliance
11 * with the License. You may obtain a copy of the License at
12 *
13 * http://www.apache.org/licenses/LICENSE-2.0
14 *
15 * Unless required by applicable law or agreed to in writing,
16 * software distributed under the License is distributed on an
17 * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
18 * KIND, either express or implied. See the License for the
19 * specific language governing permissions and limitations
20 * under the License.
21 * ====================================================================
22 */
23
24
25
26 #include <string.h>
27 #include <assert.h>
28 #include <ctype.h>
29
30 #include <apr_uri.h>
31 #include <apr_lib.h>
32
33 #include "svn_private_config.h"
34 #include "svn_string.h"
35 #include "svn_dirent_uri.h"
36 #include "svn_path.h"
37 #include "svn_ctype.h"
38
39 #include "dirent_uri.h"
40 #include "private/svn_fspath.h"
41 #include "private/svn_cert.h"
42
43 /* The canonical empty path. Can this be changed? Well, change the empty
44 test below and the path library will work, not so sure about the fs/wc
45 libraries. */
46 #define SVN_EMPTY_PATH ""
47
48 /* TRUE if s is the canonical empty path, FALSE otherwise */
49 #define SVN_PATH_IS_EMPTY(s) ((s)[0] == '\0')
50
51 /* TRUE if s,n is the platform's empty path ("."), FALSE otherwise. Can
52 this be changed? Well, the path library will work, not so sure about
53 the OS! */
54 #define SVN_PATH_IS_PLATFORM_EMPTY(s,n) ((n) == 1 && (s)[0] == '.')
55
56 /* This check must match the check on top of dirent_uri-tests.c and
57 path-tests.c */
58 #if defined(WIN32) || defined(__CYGWIN__) || defined(__OS2__)
59 #define SVN_USE_DOS_PATHS
60 #endif
61
62 /* Path type definition. Used only by internal functions. */
63 typedef enum path_type_t {
64 type_uri,
65 type_dirent,
66 type_relpath
67 } path_type_t;
68
69
70 /**** Forward declarations *****/
71
72 static svn_boolean_t
73 relpath_is_canonical(const char *relpath);
74
75
76 /**** Internal implementation functions *****/
77
78 /* Return an internal-style new path based on PATH, allocated in POOL.
79 *
80 * "Internal-style" means that separators are all '/'.
81 */
82 static const char *
internal_style(const char * path,apr_pool_t * pool)83 internal_style(const char *path, apr_pool_t *pool)
84 {
85 #if '/' != SVN_PATH_LOCAL_SEPARATOR
86 {
87 char *p = apr_pstrdup(pool, path);
88 path = p;
89
90 /* Convert all local-style separators to the canonical ones. */
91 for (; *p != '\0'; ++p)
92 if (*p == SVN_PATH_LOCAL_SEPARATOR)
93 *p = '/';
94 }
95 #endif
96
97 return path;
98 }
99
100 /* Locale insensitive tolower() for converting parts of dirents and urls
101 while canonicalizing */
102 static char
canonicalize_to_lower(char c)103 canonicalize_to_lower(char c)
104 {
105 if (c < 'A' || c > 'Z')
106 return c;
107 else
108 return (char)(c - 'A' + 'a');
109 }
110
111 /* Locale insensitive toupper() for converting parts of dirents and urls
112 while canonicalizing */
113 static char
canonicalize_to_upper(char c)114 canonicalize_to_upper(char c)
115 {
116 if (c < 'a' || c > 'z')
117 return c;
118 else
119 return (char)(c - 'a' + 'A');
120 }
121
122 /* Calculates the length of the dirent absolute or non absolute root in
123 DIRENT, return 0 if dirent is not rooted */
124 static apr_size_t
dirent_root_length(const char * dirent,apr_size_t len)125 dirent_root_length(const char *dirent, apr_size_t len)
126 {
127 #ifdef SVN_USE_DOS_PATHS
128 if (len >= 2 && dirent[1] == ':' &&
129 ((dirent[0] >= 'A' && dirent[0] <= 'Z') ||
130 (dirent[0] >= 'a' && dirent[0] <= 'z')))
131 {
132 return (len > 2 && dirent[2] == '/') ? 3 : 2;
133 }
134
135 if (len > 2 && dirent[0] == '/' && dirent[1] == '/')
136 {
137 apr_size_t i = 2;
138
139 while (i < len && dirent[i] != '/')
140 i++;
141
142 if (i == len)
143 return len; /* Cygwin drive alias, invalid path on WIN32 */
144
145 i++; /* Skip '/' */
146
147 while (i < len && dirent[i] != '/')
148 i++;
149
150 return i;
151 }
152 #endif /* SVN_USE_DOS_PATHS */
153 if (len >= 1 && dirent[0] == '/')
154 return 1;
155
156 return 0;
157 }
158
159
160 /* Return the length of substring necessary to encompass the entire
161 * previous dirent segment in DIRENT, which should be a LEN byte string.
162 *
163 * A trailing slash will not be included in the returned length except
164 * in the case in which DIRENT is absolute and there are no more
165 * previous segments.
166 */
167 static apr_size_t
dirent_previous_segment(const char * dirent,apr_size_t len)168 dirent_previous_segment(const char *dirent,
169 apr_size_t len)
170 {
171 if (len == 0)
172 return 0;
173
174 --len;
175 while (len > 0 && dirent[len] != '/'
176 #ifdef SVN_USE_DOS_PATHS
177 && (dirent[len] != ':' || len != 1)
178 #endif /* SVN_USE_DOS_PATHS */
179 )
180 --len;
181
182 /* check if the remaining segment including trailing '/' is a root dirent */
183 if (dirent_root_length(dirent, len+1) == len + 1)
184 return len + 1;
185 else
186 return len;
187 }
188
189 /* Calculates the length occupied by the schema defined root of URI */
190 static apr_size_t
uri_schema_root_length(const char * uri,apr_size_t len)191 uri_schema_root_length(const char *uri, apr_size_t len)
192 {
193 apr_size_t i;
194
195 for (i = 0; i < len; i++)
196 {
197 if (uri[i] == '/')
198 {
199 if (i > 0 && uri[i-1] == ':' && i < len-1 && uri[i+1] == '/')
200 {
201 /* We have an absolute uri */
202 if (i == 5 && strncmp("file", uri, 4) == 0)
203 return 7; /* file:// */
204 else
205 {
206 for (i += 2; i < len; i++)
207 if (uri[i] == '/')
208 return i;
209
210 return len; /* Only a hostname is found */
211 }
212 }
213 else
214 return 0;
215 }
216 }
217
218 return 0;
219 }
220
221 /* Returns TRUE if svn_dirent_is_absolute(dirent) or when dirent has
222 a non absolute root. (E.g. '/' or 'F:' on Windows) */
223 static svn_boolean_t
dirent_is_rooted(const char * dirent)224 dirent_is_rooted(const char *dirent)
225 {
226 if (! dirent)
227 return FALSE;
228
229 /* Root on all systems */
230 if (dirent[0] == '/')
231 return TRUE;
232
233 /* On Windows, dirent is also absolute when it starts with 'H:' or 'H:/'
234 where 'H' is any letter. */
235 #ifdef SVN_USE_DOS_PATHS
236 if (((dirent[0] >= 'A' && dirent[0] <= 'Z') ||
237 (dirent[0] >= 'a' && dirent[0] <= 'z')) &&
238 (dirent[1] == ':'))
239 return TRUE;
240 #endif /* SVN_USE_DOS_PATHS */
241
242 return FALSE;
243 }
244
245 /* Return the length of substring necessary to encompass the entire
246 * previous relpath segment in RELPATH, which should be a LEN byte string.
247 *
248 * A trailing slash will not be included in the returned length.
249 */
250 static apr_size_t
relpath_previous_segment(const char * relpath,apr_size_t len)251 relpath_previous_segment(const char *relpath,
252 apr_size_t len)
253 {
254 if (len == 0)
255 return 0;
256
257 --len;
258 while (len > 0 && relpath[len] != '/')
259 --len;
260
261 return len;
262 }
263
264 /* Return the length of substring necessary to encompass the entire
265 * previous uri segment in URI, which should be a LEN byte string.
266 *
267 * A trailing slash will not be included in the returned length except
268 * in the case in which URI is absolute and there are no more
269 * previous segments.
270 */
271 static apr_size_t
uri_previous_segment(const char * uri,apr_size_t len)272 uri_previous_segment(const char *uri,
273 apr_size_t len)
274 {
275 apr_size_t root_length;
276 apr_size_t i = len;
277 if (len == 0)
278 return 0;
279
280 root_length = uri_schema_root_length(uri, len);
281
282 --i;
283 while (len > root_length && uri[i] != '/')
284 --i;
285
286 if (i == 0 && len > 1 && *uri == '/')
287 return 1;
288
289 return i;
290 }
291
292 /* Return the canonicalized version of PATH, of type TYPE, allocated in
293 * POOL.
294 */
295 static const char *
canonicalize(path_type_t type,const char * path,apr_pool_t * pool)296 canonicalize(path_type_t type, const char *path, apr_pool_t *pool)
297 {
298 char *canon, *dst;
299 const char *src;
300 apr_size_t seglen;
301 apr_size_t schemelen = 0;
302 apr_size_t canon_segments = 0;
303 svn_boolean_t url = FALSE;
304 char *schema_data = NULL;
305
306 /* "" is already canonical, so just return it; note that later code
307 depends on path not being zero-length. */
308 if (SVN_PATH_IS_EMPTY(path))
309 {
310 assert(type != type_uri);
311 return "";
312 }
313
314 dst = canon = apr_pcalloc(pool, strlen(path) + 1);
315
316 /* If this is supposed to be an URI, it should start with
317 "scheme://". We'll copy the scheme, host name, etc. to DST and
318 set URL = TRUE. */
319 src = path;
320 if (type == type_uri)
321 {
322 assert(*src != '/');
323
324 while (*src && (*src != '/') && (*src != ':'))
325 src++;
326
327 if (*src == ':' && *(src+1) == '/' && *(src+2) == '/')
328 {
329 const char *seg;
330
331 url = TRUE;
332
333 /* Found a scheme, convert to lowercase and copy to dst. */
334 src = path;
335 while (*src != ':')
336 {
337 *(dst++) = canonicalize_to_lower((*src++));
338 schemelen++;
339 }
340 *(dst++) = ':';
341 *(dst++) = '/';
342 *(dst++) = '/';
343 src += 3;
344 schemelen += 3;
345
346 /* This might be the hostname */
347 seg = src;
348 while (*src && (*src != '/') && (*src != '@'))
349 src++;
350
351 if (*src == '@')
352 {
353 /* Copy the username & password. */
354 seglen = src - seg + 1;
355 memcpy(dst, seg, seglen);
356 dst += seglen;
357 src++;
358 }
359 else
360 src = seg;
361
362 /* Found a hostname, convert to lowercase and copy to dst. */
363 if (*src == '[')
364 {
365 *(dst++) = *(src++); /* Copy '[' */
366
367 while (*src == ':'
368 || (*src >= '0' && (*src <= '9'))
369 || (*src >= 'a' && (*src <= 'f'))
370 || (*src >= 'A' && (*src <= 'F')))
371 {
372 *(dst++) = canonicalize_to_lower((*src++));
373 }
374
375 if (*src == ']')
376 *(dst++) = *(src++); /* Copy ']' */
377 }
378 else
379 while (*src && (*src != '/') && (*src != ':'))
380 *(dst++) = canonicalize_to_lower((*src++));
381
382 if (*src == ':')
383 {
384 /* We probably have a port number: Is it a default portnumber
385 which doesn't belong in a canonical url? */
386 if (src[1] == '8' && src[2] == '0'
387 && (src[3]== '/'|| !src[3])
388 && !strncmp(canon, "http:", 5))
389 {
390 src += 3;
391 }
392 else if (src[1] == '4' && src[2] == '4' && src[3] == '3'
393 && (src[4]== '/'|| !src[4])
394 && !strncmp(canon, "https:", 6))
395 {
396 src += 4;
397 }
398 else if (src[1] == '3' && src[2] == '6'
399 && src[3] == '9' && src[4] == '0'
400 && (src[5]== '/'|| !src[5])
401 && !strncmp(canon, "svn:", 4))
402 {
403 src += 5;
404 }
405 else if (src[1] == '/' || !src[1])
406 {
407 src += 1;
408 }
409
410 while (*src && (*src != '/'))
411 *(dst++) = canonicalize_to_lower((*src++));
412 }
413
414 /* Copy trailing slash, or null-terminator. */
415 *(dst) = *(src);
416
417 /* Move src and dst forward only if we are not
418 * at null-terminator yet. */
419 if (*src)
420 {
421 src++;
422 dst++;
423 schema_data = dst;
424 }
425
426 canon_segments = 1;
427 }
428 }
429
430 /* Copy to DST any separator or drive letter that must come before the
431 first regular path segment. */
432 if (! url && type != type_relpath)
433 {
434 src = path;
435 /* If this is an absolute path, then just copy over the initial
436 separator character. */
437 if (*src == '/')
438 {
439 *(dst++) = *(src++);
440
441 #ifdef SVN_USE_DOS_PATHS
442 /* On Windows permit two leading separator characters which means an
443 * UNC path. */
444 if ((type == type_dirent) && *src == '/')
445 *(dst++) = *(src++);
446 #endif /* SVN_USE_DOS_PATHS */
447 }
448 #ifdef SVN_USE_DOS_PATHS
449 /* On Windows the first segment can be a drive letter, which we normalize
450 to upper case. */
451 else if (type == type_dirent &&
452 ((*src >= 'a' && *src <= 'z') ||
453 (*src >= 'A' && *src <= 'Z')) &&
454 (src[1] == ':'))
455 {
456 *(dst++) = canonicalize_to_upper(*(src++));
457 /* Leave the ':' to be processed as (or as part of) a path segment
458 by the following code block, so we need not care whether it has
459 a slash after it. */
460 }
461 #endif /* SVN_USE_DOS_PATHS */
462 }
463
464 while (*src)
465 {
466 /* Parse each segment, finding the closing '/' (which might look
467 like '%2F' for URIs). */
468 const char *next = src;
469 apr_size_t slash_len = 0;
470
471 while (*next
472 && (next[0] != '/')
473 && (! (type == type_uri && next[0] == '%' && next[1] == '2' &&
474 canonicalize_to_upper(next[2]) == 'F')))
475 {
476 ++next;
477 }
478
479 /* Record how long our "slash" is. */
480 if (next[0] == '/')
481 slash_len = 1;
482 else if (type == type_uri && next[0] == '%')
483 slash_len = 3;
484
485 seglen = next - src;
486
487 if (seglen == 0
488 || (seglen == 1 && src[0] == '.')
489 || (type == type_uri && seglen == 3 && src[0] == '%' && src[1] == '2'
490 && canonicalize_to_upper(src[2]) == 'E'))
491 {
492 /* Empty or noop segment, so do nothing. (For URIs, '%2E'
493 is equivalent to '.'). */
494 }
495 #ifdef SVN_USE_DOS_PATHS
496 /* If this is the first path segment of a file:// URI and it contains a
497 windows drive letter, convert the drive letter to upper case. */
498 else if (url && canon_segments == 1 && seglen == 2 &&
499 (strncmp(canon, "file:", 5) == 0) &&
500 src[0] >= 'a' && src[0] <= 'z' && src[1] == ':')
501 {
502 *(dst++) = canonicalize_to_upper(src[0]);
503 *(dst++) = ':';
504 if (*next)
505 *(dst++) = *next;
506 canon_segments++;
507 }
508 #endif /* SVN_USE_DOS_PATHS */
509 else
510 {
511 /* An actual segment, append it to the destination path */
512 memcpy(dst, src, seglen);
513 dst += seglen;
514 if (slash_len)
515 *(dst++) = '/';
516 canon_segments++;
517 }
518
519 /* Skip over trailing slash to the next segment. */
520 src = next + slash_len;
521 }
522
523 /* Remove the trailing slash if there was at least one
524 * canonical segment and the last segment ends with a slash.
525 *
526 * But keep in mind that, for URLs, the scheme counts as a
527 * canonical segment -- so if path is ONLY a scheme (such
528 * as "https://") we should NOT remove the trailing slash. */
529 if ((canon_segments > 0 && *(dst - 1) == '/')
530 && ! (url && path[schemelen] == '\0'))
531 {
532 dst --;
533 }
534
535 *dst = '\0';
536
537 #ifdef SVN_USE_DOS_PATHS
538 /* Skip leading double slashes when there are less than 2
539 * canon segments. UNC paths *MUST* have two segments. */
540 if ((type == type_dirent) && canon[0] == '/' && canon[1] == '/')
541 {
542 if (canon_segments < 2)
543 return canon + 1;
544 else
545 {
546 /* Now we're sure this is a valid UNC path, convert the server name
547 (the first path segment) to lowercase as Windows treats it as case
548 insensitive.
549 Note: normally the share name is treated as case insensitive too,
550 but it seems to be possible to configure Samba to treat those as
551 case sensitive, so better leave that alone. */
552 for (dst = canon + 2; *dst && *dst != '/'; dst++)
553 *dst = canonicalize_to_lower(*dst);
554 }
555 }
556 #endif /* SVN_USE_DOS_PATHS */
557
558 /* Check the normalization of characters in a uri */
559 if (schema_data)
560 {
561 int need_extra = 0;
562 src = schema_data;
563
564 while (*src)
565 {
566 switch (*src)
567 {
568 case '/':
569 break;
570 case '%':
571 if (!svn_ctype_isxdigit(*(src+1)) ||
572 !svn_ctype_isxdigit(*(src+2)))
573 need_extra += 2;
574 else
575 src += 2;
576 break;
577 default:
578 if (!svn_uri__char_validity[(unsigned char)*src])
579 need_extra += 2;
580 break;
581 }
582 src++;
583 }
584
585 if (need_extra > 0)
586 {
587 apr_size_t pre_schema_size = (apr_size_t)(schema_data - canon);
588
589 dst = apr_palloc(pool, (apr_size_t)(src - canon) + need_extra + 1);
590 memcpy(dst, canon, pre_schema_size);
591 canon = dst;
592
593 dst += pre_schema_size;
594 }
595 else
596 dst = schema_data;
597
598 src = schema_data;
599
600 while (*src)
601 {
602 switch (*src)
603 {
604 case '/':
605 *(dst++) = '/';
606 break;
607 case '%':
608 if (!svn_ctype_isxdigit(*(src+1)) ||
609 !svn_ctype_isxdigit(*(src+2)))
610 {
611 *(dst++) = '%';
612 *(dst++) = '2';
613 *(dst++) = '5';
614 }
615 else
616 {
617 char digitz[3];
618 int val;
619
620 digitz[0] = *(++src);
621 digitz[1] = *(++src);
622 digitz[2] = 0;
623
624 val = (int)strtol(digitz, NULL, 16);
625
626 if (svn_uri__char_validity[(unsigned char)val])
627 *(dst++) = (char)val;
628 else
629 {
630 *(dst++) = '%';
631 *(dst++) = canonicalize_to_upper(digitz[0]);
632 *(dst++) = canonicalize_to_upper(digitz[1]);
633 }
634 }
635 break;
636 default:
637 if (!svn_uri__char_validity[(unsigned char)*src])
638 {
639 apr_snprintf(dst, 4, "%%%02X", (unsigned char)*src);
640 dst += 3;
641 }
642 else
643 *(dst++) = *src;
644 break;
645 }
646 src++;
647 }
648 *dst = '\0';
649 }
650
651 return canon;
652 }
653
654 /* Return the string length of the longest common ancestor of PATH1 and PATH2.
655 * Pass type_uri for TYPE if PATH1 and PATH2 are URIs, and type_dirent if
656 * PATH1 and PATH2 are regular paths.
657 *
658 * If the two paths do not share a common ancestor, return 0.
659 *
660 * New strings are allocated in POOL.
661 */
662 static apr_size_t
get_longest_ancestor_length(path_type_t types,const char * path1,const char * path2,apr_pool_t * pool)663 get_longest_ancestor_length(path_type_t types,
664 const char *path1,
665 const char *path2,
666 apr_pool_t *pool)
667 {
668 apr_size_t path1_len, path2_len;
669 apr_size_t i = 0;
670 apr_size_t last_dirsep = 0;
671 #ifdef SVN_USE_DOS_PATHS
672 svn_boolean_t unc = FALSE;
673 #endif
674
675 path1_len = strlen(path1);
676 path2_len = strlen(path2);
677
678 if (SVN_PATH_IS_EMPTY(path1) || SVN_PATH_IS_EMPTY(path2))
679 return 0;
680
681 while (path1[i] == path2[i])
682 {
683 /* Keep track of the last directory separator we hit. */
684 if (path1[i] == '/')
685 last_dirsep = i;
686
687 i++;
688
689 /* If we get to the end of either path, break out. */
690 if ((i == path1_len) || (i == path2_len))
691 break;
692 }
693
694 /* two special cases:
695 1. '/' is the longest common ancestor of '/' and '/foo' */
696 if (i == 1 && path1[0] == '/' && path2[0] == '/')
697 return 1;
698 /* 2. '' is the longest common ancestor of any non-matching
699 * strings 'foo' and 'bar' */
700 if (types == type_dirent && i == 0)
701 return 0;
702
703 /* Handle some windows specific cases */
704 #ifdef SVN_USE_DOS_PATHS
705 if (types == type_dirent)
706 {
707 /* don't count the '//' from UNC paths */
708 if (last_dirsep == 1 && path1[0] == '/' && path1[1] == '/')
709 {
710 last_dirsep = 0;
711 unc = TRUE;
712 }
713
714 /* X:/ and X:/foo */
715 if (i == 3 && path1[2] == '/' && path1[1] == ':')
716 return i;
717
718 /* Cannot use SVN_ERR_ASSERT here, so we'll have to crash, sorry.
719 * Note that this assertion triggers only if the code above has
720 * been broken. The code below relies on this assertion, because
721 * it uses [i - 1] as index. */
722 assert(i > 0);
723
724 /* X: and X:/ */
725 if ((path1[i - 1] == ':' && path2[i] == '/') ||
726 (path2[i - 1] == ':' && path1[i] == '/'))
727 return 0;
728 /* X: and X:foo */
729 if (path1[i - 1] == ':' || path2[i - 1] == ':')
730 return i;
731 }
732 #endif /* SVN_USE_DOS_PATHS */
733
734 /* last_dirsep is now the offset of the last directory separator we
735 crossed before reaching a non-matching byte. i is the offset of
736 that non-matching byte, and is guaranteed to be <= the length of
737 whichever path is shorter.
738 If one of the paths is the common part return that. */
739 if (((i == path1_len) && (path2[i] == '/'))
740 || ((i == path2_len) && (path1[i] == '/'))
741 || ((i == path1_len) && (i == path2_len)))
742 return i;
743 else
744 {
745 /* Nothing in common but the root folder '/' or 'X:/' for Windows
746 dirents. */
747 #ifdef SVN_USE_DOS_PATHS
748 if (! unc)
749 {
750 /* X:/foo and X:/bar returns X:/ */
751 if ((types == type_dirent) &&
752 last_dirsep == 2 && path1[1] == ':' && path1[2] == '/'
753 && path2[1] == ':' && path2[2] == '/')
754 return 3;
755 #endif /* SVN_USE_DOS_PATHS */
756 if (last_dirsep == 0 && path1[0] == '/' && path2[0] == '/')
757 return 1;
758 #ifdef SVN_USE_DOS_PATHS
759 }
760 #endif
761 }
762
763 return last_dirsep;
764 }
765
766 /* Determine whether PATH2 is a child of PATH1.
767 *
768 * PATH2 is a child of PATH1 if
769 * 1) PATH1 is empty, and PATH2 is not empty and not an absolute path.
770 * or
771 * 2) PATH2 is has n components, PATH1 has x < n components,
772 * and PATH1 matches PATH2 in all its x components.
773 * Components are separated by a slash, '/'.
774 *
775 * Pass type_uri for TYPE if PATH1 and PATH2 are URIs, and type_dirent if
776 * PATH1 and PATH2 are regular paths.
777 *
778 * If PATH2 is not a child of PATH1, return NULL.
779 *
780 * If PATH2 is a child of PATH1, and POOL is not NULL, allocate a copy
781 * of the child part of PATH2 in POOL and return a pointer to the
782 * newly allocated child part.
783 *
784 * If PATH2 is a child of PATH1, and POOL is NULL, return a pointer
785 * pointing to the child part of PATH2.
786 * */
787 static const char *
is_child(path_type_t type,const char * path1,const char * path2,apr_pool_t * pool)788 is_child(path_type_t type, const char *path1, const char *path2,
789 apr_pool_t *pool)
790 {
791 apr_size_t i;
792
793 /* Allow "" and "foo" or "H:foo" to be parent/child */
794 if (SVN_PATH_IS_EMPTY(path1)) /* "" is the parent */
795 {
796 if (SVN_PATH_IS_EMPTY(path2)) /* "" not a child */
797 return NULL;
798
799 /* check if this is an absolute path */
800 if ((type == type_uri) ||
801 (type == type_dirent && dirent_is_rooted(path2)))
802 return NULL;
803 else
804 /* everything else is child */
805 return pool ? apr_pstrdup(pool, path2) : path2;
806 }
807
808 /* Reach the end of at least one of the paths. How should we handle
809 things like path1:"foo///bar" and path2:"foo/bar/baz"? It doesn't
810 appear to arise in the current Subversion code, it's not clear to me
811 if they should be parent/child or not. */
812 /* Hmmm... aren't paths assumed to be canonical in this function?
813 * How can "foo///bar" even happen if the paths are canonical? */
814 for (i = 0; path1[i] && path2[i]; i++)
815 if (path1[i] != path2[i])
816 return NULL;
817
818 /* FIXME: This comment does not really match
819 * the checks made in the code it refers to: */
820 /* There are two cases that are parent/child
821 ... path1[i] == '\0'
822 .../foo path2[i] == '/'
823 or
824 / path1[i] == '\0'
825 /foo path2[i] != '/'
826
827 Other root paths (like X:/) fall under the former case:
828 X:/ path1[i] == '\0'
829 X:/foo path2[i] != '/'
830
831 Check for '//' to avoid matching '/' and '//srv'.
832 */
833 if (path1[i] == '\0' && path2[i])
834 {
835 if (path1[i - 1] == '/'
836 #ifdef SVN_USE_DOS_PATHS
837 || ((type == type_dirent) && path1[i - 1] == ':')
838 #endif
839 )
840 {
841 if (path2[i] == '/')
842 /* .../
843 * ..../
844 * i */
845 return NULL;
846 else
847 /* .../
848 * .../foo
849 * i */
850 return pool ? apr_pstrdup(pool, path2 + i) : path2 + i;
851 }
852 else if (path2[i] == '/')
853 {
854 if (path2[i + 1])
855 /* ...
856 * .../foo
857 * i */
858 return pool ? apr_pstrdup(pool, path2 + i + 1) : path2 + i + 1;
859 else
860 /* ...
861 * .../
862 * i */
863 return NULL;
864 }
865 }
866
867 /* Otherwise, path2 isn't a child. */
868 return NULL;
869 }
870
871
872 /**** Public API functions ****/
873
874 const char *
svn_dirent_internal_style(const char * dirent,apr_pool_t * pool)875 svn_dirent_internal_style(const char *dirent, apr_pool_t *pool)
876 {
877 return svn_dirent_canonicalize(internal_style(dirent, pool), pool);
878 }
879
880 const char *
svn_dirent_local_style(const char * dirent,apr_pool_t * pool)881 svn_dirent_local_style(const char *dirent, apr_pool_t *pool)
882 {
883 /* Internally, Subversion represents the current directory with the
884 empty string. But users like to see "." . */
885 if (SVN_PATH_IS_EMPTY(dirent))
886 return ".";
887
888 #if '/' != SVN_PATH_LOCAL_SEPARATOR
889 {
890 char *p = apr_pstrdup(pool, dirent);
891 dirent = p;
892
893 /* Convert all canonical separators to the local-style ones. */
894 for (; *p != '\0'; ++p)
895 if (*p == '/')
896 *p = SVN_PATH_LOCAL_SEPARATOR;
897 }
898 #endif
899
900 return dirent;
901 }
902
903 const char *
svn_relpath__internal_style(const char * relpath,apr_pool_t * pool)904 svn_relpath__internal_style(const char *relpath,
905 apr_pool_t *pool)
906 {
907 return svn_relpath_canonicalize(internal_style(relpath, pool), pool);
908 }
909
910
911 /* We decided against using apr_filepath_root here because of the negative
912 performance impact (creating a pool and converting strings ). */
913 svn_boolean_t
svn_dirent_is_root(const char * dirent,apr_size_t len)914 svn_dirent_is_root(const char *dirent, apr_size_t len)
915 {
916 #ifdef SVN_USE_DOS_PATHS
917 /* On Windows and Cygwin, 'H:' or 'H:/' (where 'H' is any letter)
918 are also root directories */
919 if ((len == 2 || ((len == 3) && (dirent[2] == '/'))) &&
920 (dirent[1] == ':') &&
921 ((dirent[0] >= 'A' && dirent[0] <= 'Z') ||
922 (dirent[0] >= 'a' && dirent[0] <= 'z')))
923 return TRUE;
924
925 /* On Windows and Cygwin //server/share is a root directory,
926 and on Cygwin //drive is a drive alias */
927 if (len >= 2 && dirent[0] == '/' && dirent[1] == '/'
928 && dirent[len - 1] != '/')
929 {
930 int segments = 0;
931 apr_size_t i;
932 for (i = len; i >= 2; i--)
933 {
934 if (dirent[i] == '/')
935 {
936 segments ++;
937 if (segments > 1)
938 return FALSE;
939 }
940 }
941 #ifdef __CYGWIN__
942 return (segments <= 1);
943 #else
944 return (segments == 1); /* //drive is invalid on plain Windows */
945 #endif
946 }
947 #endif
948
949 /* directory is root if it's equal to '/' */
950 if (len == 1 && dirent[0] == '/')
951 return TRUE;
952
953 return FALSE;
954 }
955
956 svn_boolean_t
svn_uri_is_root(const char * uri,apr_size_t len)957 svn_uri_is_root(const char *uri, apr_size_t len)
958 {
959 assert(svn_uri_is_canonical(uri, NULL));
960 return (len == uri_schema_root_length(uri, len));
961 }
962
svn_dirent_join(const char * base,const char * component,apr_pool_t * pool)963 char *svn_dirent_join(const char *base,
964 const char *component,
965 apr_pool_t *pool)
966 {
967 apr_size_t blen = strlen(base);
968 apr_size_t clen = strlen(component);
969 char *dirent;
970 int add_separator;
971
972 assert(svn_dirent_is_canonical(base, pool));
973 assert(svn_dirent_is_canonical(component, pool));
974
975 /* If the component is absolute, then return it. */
976 if (svn_dirent_is_absolute(component))
977 return apr_pmemdup(pool, component, clen + 1);
978
979 /* If either is empty return the other */
980 if (SVN_PATH_IS_EMPTY(base))
981 return apr_pmemdup(pool, component, clen + 1);
982 if (SVN_PATH_IS_EMPTY(component))
983 return apr_pmemdup(pool, base, blen + 1);
984
985 #ifdef SVN_USE_DOS_PATHS
986 if (component[0] == '/')
987 {
988 /* '/' is drive relative on Windows, not absolute like on Posix */
989 if (dirent_is_rooted(base))
990 {
991 /* Join component without '/' to root-of(base) */
992 blen = dirent_root_length(base, blen);
993 component++;
994 clen--;
995
996 if (blen == 2 && base[1] == ':') /* "C:" case */
997 {
998 char *root = apr_pmemdup(pool, base, 3);
999 root[2] = '/'; /* We don't need the final '\0' */
1000
1001 base = root;
1002 blen = 3;
1003 }
1004
1005 if (clen == 0)
1006 return apr_pstrndup(pool, base, blen);
1007 }
1008 else
1009 return apr_pmemdup(pool, component, clen + 1);
1010 }
1011 else if (dirent_is_rooted(component))
1012 return apr_pmemdup(pool, component, clen + 1);
1013 #endif /* SVN_USE_DOS_PATHS */
1014
1015 /* if last character of base is already a separator, don't add a '/' */
1016 add_separator = 1;
1017 if (base[blen - 1] == '/'
1018 #ifdef SVN_USE_DOS_PATHS
1019 || base[blen - 1] == ':'
1020 #endif
1021 )
1022 add_separator = 0;
1023
1024 /* Construct the new, combined dirent. */
1025 dirent = apr_palloc(pool, blen + add_separator + clen + 1);
1026 memcpy(dirent, base, blen);
1027 if (add_separator)
1028 dirent[blen] = '/';
1029 memcpy(dirent + blen + add_separator, component, clen + 1);
1030
1031 return dirent;
1032 }
1033
svn_dirent_join_many(apr_pool_t * pool,const char * base,...)1034 char *svn_dirent_join_many(apr_pool_t *pool, const char *base, ...)
1035 {
1036 #define MAX_SAVED_LENGTHS 10
1037 apr_size_t saved_lengths[MAX_SAVED_LENGTHS];
1038 apr_size_t total_len;
1039 int nargs;
1040 va_list va;
1041 const char *s;
1042 apr_size_t len;
1043 char *dirent;
1044 char *p;
1045 int add_separator;
1046 int base_arg = 0;
1047
1048 total_len = strlen(base);
1049
1050 assert(svn_dirent_is_canonical(base, pool));
1051
1052 /* if last character of base is already a separator, don't add a '/' */
1053 add_separator = 1;
1054 if (total_len == 0
1055 || base[total_len - 1] == '/'
1056 #ifdef SVN_USE_DOS_PATHS
1057 || base[total_len - 1] == ':'
1058 #endif
1059 )
1060 add_separator = 0;
1061
1062 saved_lengths[0] = total_len;
1063
1064 /* Compute the length of the resulting string. */
1065
1066 nargs = 0;
1067 va_start(va, base);
1068 while ((s = va_arg(va, const char *)) != NULL)
1069 {
1070 len = strlen(s);
1071
1072 assert(svn_dirent_is_canonical(s, pool));
1073
1074 if (SVN_PATH_IS_EMPTY(s))
1075 continue;
1076
1077 if (nargs++ < MAX_SAVED_LENGTHS)
1078 saved_lengths[nargs] = len;
1079
1080 if (dirent_is_rooted(s))
1081 {
1082 total_len = len;
1083 base_arg = nargs;
1084
1085 #ifdef SVN_USE_DOS_PATHS
1086 if (!svn_dirent_is_absolute(s)) /* Handle non absolute roots */
1087 {
1088 /* Set new base and skip the current argument */
1089 base = s = svn_dirent_join(base, s, pool);
1090 base_arg++;
1091 saved_lengths[0] = total_len = len = strlen(s);
1092 }
1093 else
1094 #endif /* SVN_USE_DOS_PATHS */
1095 {
1096 base = ""; /* Don't add base */
1097 saved_lengths[0] = 0;
1098 }
1099
1100 add_separator = 1;
1101 if (s[len - 1] == '/'
1102 #ifdef SVN_USE_DOS_PATHS
1103 || s[len - 1] == ':'
1104 #endif
1105 )
1106 add_separator = 0;
1107 }
1108 else if (nargs <= base_arg + 1)
1109 {
1110 total_len += add_separator + len;
1111 }
1112 else
1113 {
1114 total_len += 1 + len;
1115 }
1116 }
1117 va_end(va);
1118
1119 /* base == "/" and no further components. just return that. */
1120 if (add_separator == 0 && total_len == 1)
1121 return apr_pmemdup(pool, "/", 2);
1122
1123 /* we got the total size. allocate it, with room for a NULL character. */
1124 dirent = p = apr_palloc(pool, total_len + 1);
1125
1126 /* if we aren't supposed to skip forward to an absolute component, and if
1127 this is not an empty base that we are skipping, then copy the base
1128 into the output. */
1129 if (! SVN_PATH_IS_EMPTY(base))
1130 {
1131 memcpy(p, base, len = saved_lengths[0]);
1132 p += len;
1133 }
1134
1135 nargs = 0;
1136 va_start(va, base);
1137 while ((s = va_arg(va, const char *)) != NULL)
1138 {
1139 if (SVN_PATH_IS_EMPTY(s))
1140 continue;
1141
1142 if (++nargs < base_arg)
1143 continue;
1144
1145 if (nargs < MAX_SAVED_LENGTHS)
1146 len = saved_lengths[nargs];
1147 else
1148 len = strlen(s);
1149
1150 /* insert a separator if we aren't copying in the first component
1151 (which can happen when base_arg is set). also, don't put in a slash
1152 if the prior character is a slash (occurs when prior component
1153 is "/"). */
1154 if (p != dirent &&
1155 ( ! (nargs - 1 <= base_arg) || add_separator))
1156 *p++ = '/';
1157
1158 /* copy the new component and advance the pointer */
1159 memcpy(p, s, len);
1160 p += len;
1161 }
1162 va_end(va);
1163
1164 *p = '\0';
1165 assert((apr_size_t)(p - dirent) == total_len);
1166
1167 return dirent;
1168 }
1169
1170 char *
svn_relpath_join(const char * base,const char * component,apr_pool_t * pool)1171 svn_relpath_join(const char *base,
1172 const char *component,
1173 apr_pool_t *pool)
1174 {
1175 apr_size_t blen = strlen(base);
1176 apr_size_t clen = strlen(component);
1177 char *path;
1178
1179 assert(relpath_is_canonical(base));
1180 assert(relpath_is_canonical(component));
1181
1182 /* If either is empty return the other */
1183 if (blen == 0)
1184 return apr_pmemdup(pool, component, clen + 1);
1185 if (clen == 0)
1186 return apr_pmemdup(pool, base, blen + 1);
1187
1188 path = apr_palloc(pool, blen + 1 + clen + 1);
1189 memcpy(path, base, blen);
1190 path[blen] = '/';
1191 memcpy(path + blen + 1, component, clen + 1);
1192
1193 return path;
1194 }
1195
1196 char *
svn_dirent_dirname(const char * dirent,apr_pool_t * pool)1197 svn_dirent_dirname(const char *dirent, apr_pool_t *pool)
1198 {
1199 apr_size_t len = strlen(dirent);
1200
1201 assert(svn_dirent_is_canonical(dirent, pool));
1202
1203 if (len == dirent_root_length(dirent, len))
1204 return apr_pstrmemdup(pool, dirent, len);
1205 else
1206 return apr_pstrmemdup(pool, dirent, dirent_previous_segment(dirent, len));
1207 }
1208
1209 const char *
svn_dirent_basename(const char * dirent,apr_pool_t * pool)1210 svn_dirent_basename(const char *dirent, apr_pool_t *pool)
1211 {
1212 apr_size_t len = strlen(dirent);
1213 apr_size_t start;
1214
1215 assert(!pool || svn_dirent_is_canonical(dirent, pool));
1216
1217 if (svn_dirent_is_root(dirent, len))
1218 return "";
1219 else
1220 {
1221 start = len;
1222 while (start > 0 && dirent[start - 1] != '/'
1223 #ifdef SVN_USE_DOS_PATHS
1224 && dirent[start - 1] != ':'
1225 #endif
1226 )
1227 --start;
1228 }
1229
1230 if (pool)
1231 return apr_pstrmemdup(pool, dirent + start, len - start);
1232 else
1233 return dirent + start;
1234 }
1235
1236 void
svn_dirent_split(const char ** dirpath,const char ** base_name,const char * dirent,apr_pool_t * pool)1237 svn_dirent_split(const char **dirpath,
1238 const char **base_name,
1239 const char *dirent,
1240 apr_pool_t *pool)
1241 {
1242 assert(dirpath != base_name);
1243
1244 if (dirpath)
1245 *dirpath = svn_dirent_dirname(dirent, pool);
1246
1247 if (base_name)
1248 *base_name = svn_dirent_basename(dirent, pool);
1249 }
1250
1251 char *
svn_relpath_dirname(const char * relpath,apr_pool_t * pool)1252 svn_relpath_dirname(const char *relpath,
1253 apr_pool_t *pool)
1254 {
1255 apr_size_t len = strlen(relpath);
1256
1257 assert(relpath_is_canonical(relpath));
1258
1259 return apr_pstrmemdup(pool, relpath,
1260 relpath_previous_segment(relpath, len));
1261 }
1262
1263 const char *
svn_relpath_basename(const char * relpath,apr_pool_t * pool)1264 svn_relpath_basename(const char *relpath,
1265 apr_pool_t *pool)
1266 {
1267 apr_size_t len = strlen(relpath);
1268 apr_size_t start;
1269
1270 assert(relpath_is_canonical(relpath));
1271
1272 start = len;
1273 while (start > 0 && relpath[start - 1] != '/')
1274 --start;
1275
1276 if (pool)
1277 return apr_pstrmemdup(pool, relpath + start, len - start);
1278 else
1279 return relpath + start;
1280 }
1281
1282 void
svn_relpath_split(const char ** dirpath,const char ** base_name,const char * relpath,apr_pool_t * pool)1283 svn_relpath_split(const char **dirpath,
1284 const char **base_name,
1285 const char *relpath,
1286 apr_pool_t *pool)
1287 {
1288 assert(dirpath != base_name);
1289
1290 if (dirpath)
1291 *dirpath = svn_relpath_dirname(relpath, pool);
1292
1293 if (base_name)
1294 *base_name = svn_relpath_basename(relpath, pool);
1295 }
1296
1297 char *
svn_uri_dirname(const char * uri,apr_pool_t * pool)1298 svn_uri_dirname(const char *uri, apr_pool_t *pool)
1299 {
1300 apr_size_t len = strlen(uri);
1301
1302 assert(svn_uri_is_canonical(uri, pool));
1303
1304 if (svn_uri_is_root(uri, len))
1305 return apr_pstrmemdup(pool, uri, len);
1306 else
1307 return apr_pstrmemdup(pool, uri, uri_previous_segment(uri, len));
1308 }
1309
1310 const char *
svn_uri_basename(const char * uri,apr_pool_t * pool)1311 svn_uri_basename(const char *uri, apr_pool_t *pool)
1312 {
1313 apr_size_t len = strlen(uri);
1314 apr_size_t start;
1315
1316 assert(svn_uri_is_canonical(uri, NULL));
1317
1318 if (svn_uri_is_root(uri, len))
1319 return "";
1320
1321 start = len;
1322 while (start > 0 && uri[start - 1] != '/')
1323 --start;
1324
1325 return svn_path_uri_decode(uri + start, pool);
1326 }
1327
1328 void
svn_uri_split(const char ** dirpath,const char ** base_name,const char * uri,apr_pool_t * pool)1329 svn_uri_split(const char **dirpath,
1330 const char **base_name,
1331 const char *uri,
1332 apr_pool_t *pool)
1333 {
1334 assert(dirpath != base_name);
1335
1336 if (dirpath)
1337 *dirpath = svn_uri_dirname(uri, pool);
1338
1339 if (base_name)
1340 *base_name = svn_uri_basename(uri, pool);
1341 }
1342
1343 char *
svn_dirent_get_longest_ancestor(const char * dirent1,const char * dirent2,apr_pool_t * pool)1344 svn_dirent_get_longest_ancestor(const char *dirent1,
1345 const char *dirent2,
1346 apr_pool_t *pool)
1347 {
1348 return apr_pstrndup(pool, dirent1,
1349 get_longest_ancestor_length(type_dirent, dirent1,
1350 dirent2, pool));
1351 }
1352
1353 char *
svn_relpath_get_longest_ancestor(const char * relpath1,const char * relpath2,apr_pool_t * pool)1354 svn_relpath_get_longest_ancestor(const char *relpath1,
1355 const char *relpath2,
1356 apr_pool_t *pool)
1357 {
1358 assert(relpath_is_canonical(relpath1));
1359 assert(relpath_is_canonical(relpath2));
1360
1361 return apr_pstrndup(pool, relpath1,
1362 get_longest_ancestor_length(type_relpath, relpath1,
1363 relpath2, pool));
1364 }
1365
1366 char *
svn_uri_get_longest_ancestor(const char * uri1,const char * uri2,apr_pool_t * pool)1367 svn_uri_get_longest_ancestor(const char *uri1,
1368 const char *uri2,
1369 apr_pool_t *pool)
1370 {
1371 apr_size_t uri_ancestor_len;
1372 apr_size_t i = 0;
1373
1374 assert(svn_uri_is_canonical(uri1, NULL));
1375 assert(svn_uri_is_canonical(uri2, NULL));
1376
1377 /* Find ':' */
1378 while (1)
1379 {
1380 /* No shared protocol => no common prefix */
1381 if (uri1[i] != uri2[i])
1382 return apr_pmemdup(pool, SVN_EMPTY_PATH,
1383 sizeof(SVN_EMPTY_PATH));
1384
1385 if (uri1[i] == ':')
1386 break;
1387
1388 /* They're both URLs, so EOS can't come before ':' */
1389 assert((uri1[i] != '\0') && (uri2[i] != '\0'));
1390
1391 i++;
1392 }
1393
1394 i += 3; /* Advance past '://' */
1395
1396 uri_ancestor_len = get_longest_ancestor_length(type_uri, uri1 + i,
1397 uri2 + i, pool);
1398
1399 if (uri_ancestor_len == 0 ||
1400 (uri_ancestor_len == 1 && (uri1 + i)[0] == '/'))
1401 return apr_pmemdup(pool, SVN_EMPTY_PATH, sizeof(SVN_EMPTY_PATH));
1402 else
1403 return apr_pstrndup(pool, uri1, uri_ancestor_len + i);
1404 }
1405
1406 const char *
svn_dirent_is_child(const char * parent_dirent,const char * child_dirent,apr_pool_t * pool)1407 svn_dirent_is_child(const char *parent_dirent,
1408 const char *child_dirent,
1409 apr_pool_t *pool)
1410 {
1411 return is_child(type_dirent, parent_dirent, child_dirent, pool);
1412 }
1413
1414 const char *
svn_dirent_skip_ancestor(const char * parent_dirent,const char * child_dirent)1415 svn_dirent_skip_ancestor(const char *parent_dirent,
1416 const char *child_dirent)
1417 {
1418 apr_size_t len = strlen(parent_dirent);
1419 apr_size_t root_len;
1420
1421 if (0 != strncmp(parent_dirent, child_dirent, len))
1422 return NULL; /* parent_dirent is no ancestor of child_dirent */
1423
1424 if (child_dirent[len] == 0)
1425 return ""; /* parent_dirent == child_dirent */
1426
1427 /* Child == parent + more-characters */
1428
1429 root_len = dirent_root_length(child_dirent, strlen(child_dirent));
1430 if (root_len > len)
1431 /* Different root, e.g. ("" "/...") or ("//z" "//z/share") */
1432 return NULL;
1433
1434 /* Now, child == [root-of-parent] + [rest-of-parent] + more-characters.
1435 * It must be one of the following forms.
1436 *
1437 * rlen parent child bad? rlen=len? c[len]=/?
1438 * 0 "" "foo" *
1439 * 0 "b" "bad" !
1440 * 0 "b" "b/foo" *
1441 * 1 "/" "/foo" *
1442 * 1 "/b" "/bad" !
1443 * 1 "/b" "/b/foo" *
1444 * 2 "a:" "a:foo" *
1445 * 2 "a:b" "a:bad" !
1446 * 2 "a:b" "a:b/foo" *
1447 * 3 "a:/" "a:/foo" *
1448 * 3 "a:/b" "a:/bad" !
1449 * 3 "a:/b" "a:/b/foo" *
1450 * 5 "//s/s" "//s/s/foo" * *
1451 * 5 "//s/s/b" "//s/s/bad" !
1452 * 5 "//s/s/b" "//s/s/b/foo" *
1453 */
1454
1455 if (child_dirent[len] == '/')
1456 /* "parent|child" is one of:
1457 * "[a:]b|/foo" "[a:]/b|/foo" "//s/s|/foo" "//s/s/b|/foo" */
1458 return child_dirent + len + 1;
1459
1460 if (root_len == len)
1461 /* "parent|child" is "|foo" "/|foo" "a:|foo" "a:/|foo" "//s/s|/foo" */
1462 return child_dirent + len;
1463
1464 return NULL;
1465 }
1466
1467 const char *
svn_relpath_skip_ancestor(const char * parent_relpath,const char * child_relpath)1468 svn_relpath_skip_ancestor(const char *parent_relpath,
1469 const char *child_relpath)
1470 {
1471 apr_size_t len = strlen(parent_relpath);
1472
1473 assert(relpath_is_canonical(parent_relpath));
1474 assert(relpath_is_canonical(child_relpath));
1475
1476 if (len == 0)
1477 return child_relpath;
1478
1479 if (0 != strncmp(parent_relpath, child_relpath, len))
1480 return NULL; /* parent_relpath is no ancestor of child_relpath */
1481
1482 if (child_relpath[len] == 0)
1483 return ""; /* parent_relpath == child_relpath */
1484
1485 if (child_relpath[len] == '/')
1486 return child_relpath + len + 1;
1487
1488 return NULL;
1489 }
1490
1491
1492 /* */
1493 static const char *
uri_skip_ancestor(const char * parent_uri,const char * child_uri)1494 uri_skip_ancestor(const char *parent_uri,
1495 const char *child_uri)
1496 {
1497 apr_size_t len = strlen(parent_uri);
1498
1499 assert(svn_uri_is_canonical(parent_uri, NULL));
1500 assert(svn_uri_is_canonical(child_uri, NULL));
1501
1502 if (0 != strncmp(parent_uri, child_uri, len))
1503 return NULL; /* parent_uri is no ancestor of child_uri */
1504
1505 if (child_uri[len] == 0)
1506 return ""; /* parent_uri == child_uri */
1507
1508 if (child_uri[len] == '/')
1509 return child_uri + len + 1;
1510
1511 return NULL;
1512 }
1513
1514 const char *
svn_uri_skip_ancestor(const char * parent_uri,const char * child_uri,apr_pool_t * result_pool)1515 svn_uri_skip_ancestor(const char *parent_uri,
1516 const char *child_uri,
1517 apr_pool_t *result_pool)
1518 {
1519 const char *result = uri_skip_ancestor(parent_uri, child_uri);
1520
1521 return result ? svn_path_uri_decode(result, result_pool) : NULL;
1522 }
1523
1524 svn_boolean_t
svn_dirent_is_ancestor(const char * parent_dirent,const char * child_dirent)1525 svn_dirent_is_ancestor(const char *parent_dirent, const char *child_dirent)
1526 {
1527 return svn_dirent_skip_ancestor(parent_dirent, child_dirent) != NULL;
1528 }
1529
1530 svn_boolean_t
svn_uri__is_ancestor(const char * parent_uri,const char * child_uri)1531 svn_uri__is_ancestor(const char *parent_uri, const char *child_uri)
1532 {
1533 return uri_skip_ancestor(parent_uri, child_uri) != NULL;
1534 }
1535
1536
1537 svn_boolean_t
svn_dirent_is_absolute(const char * dirent)1538 svn_dirent_is_absolute(const char *dirent)
1539 {
1540 if (! dirent)
1541 return FALSE;
1542
1543 /* dirent is absolute if it starts with '/' on non-Windows platforms
1544 or with '//' on Windows platforms */
1545 if (dirent[0] == '/'
1546 #ifdef SVN_USE_DOS_PATHS
1547 && dirent[1] == '/' /* Single '/' depends on current drive */
1548 #endif
1549 )
1550 return TRUE;
1551
1552 /* On Windows, dirent is also absolute when it starts with 'H:/'
1553 where 'H' is any letter. */
1554 #ifdef SVN_USE_DOS_PATHS
1555 if (((dirent[0] >= 'A' && dirent[0] <= 'Z')) &&
1556 (dirent[1] == ':') && (dirent[2] == '/'))
1557 return TRUE;
1558 #endif /* SVN_USE_DOS_PATHS */
1559
1560 return FALSE;
1561 }
1562
1563 svn_error_t *
svn_dirent_get_absolute(const char ** pabsolute,const char * relative,apr_pool_t * pool)1564 svn_dirent_get_absolute(const char **pabsolute,
1565 const char *relative,
1566 apr_pool_t *pool)
1567 {
1568 char *buffer;
1569 apr_status_t apr_err;
1570 const char *path_apr;
1571
1572 SVN_ERR_ASSERT(! svn_path_is_url(relative));
1573
1574 /* Merge the current working directory with the relative dirent. */
1575 SVN_ERR(svn_path_cstring_from_utf8(&path_apr, relative, pool));
1576
1577 apr_err = apr_filepath_merge(&buffer, NULL,
1578 path_apr,
1579 APR_FILEPATH_NOTRELATIVE,
1580 pool);
1581 if (apr_err)
1582 {
1583 /* In some cases when the passed path or its ancestor(s) do not exist
1584 or no longer exist apr returns an error.
1585
1586 In many of these cases we would like to return a path anyway, when the
1587 passed path was already a safe absolute path. So check for that now to
1588 avoid an error.
1589
1590 svn_dirent_is_absolute() doesn't perform the necessary checks to see
1591 if the path doesn't need post processing to be in the canonical absolute
1592 format.
1593 */
1594
1595 if (svn_dirent_is_absolute(relative)
1596 && svn_dirent_is_canonical(relative, pool)
1597 && !svn_path_is_backpath_present(relative))
1598 {
1599 *pabsolute = apr_pstrdup(pool, relative);
1600 return SVN_NO_ERROR;
1601 }
1602
1603 return svn_error_createf(SVN_ERR_BAD_FILENAME,
1604 svn_error_create(apr_err, NULL, NULL),
1605 _("Couldn't determine absolute path of '%s'"),
1606 svn_dirent_local_style(relative, pool));
1607 }
1608
1609 SVN_ERR(svn_path_cstring_to_utf8(pabsolute, buffer, pool));
1610 *pabsolute = svn_dirent_canonicalize(*pabsolute, pool);
1611 return SVN_NO_ERROR;
1612 }
1613
1614 const char *
svn_uri_canonicalize(const char * uri,apr_pool_t * pool)1615 svn_uri_canonicalize(const char *uri, apr_pool_t *pool)
1616 {
1617 return canonicalize(type_uri, uri, pool);
1618 }
1619
1620 const char *
svn_relpath_canonicalize(const char * relpath,apr_pool_t * pool)1621 svn_relpath_canonicalize(const char *relpath, apr_pool_t *pool)
1622 {
1623 return canonicalize(type_relpath, relpath, pool);
1624 }
1625
1626 const char *
svn_dirent_canonicalize(const char * dirent,apr_pool_t * pool)1627 svn_dirent_canonicalize(const char *dirent, apr_pool_t *pool)
1628 {
1629 const char *dst = canonicalize(type_dirent, dirent, pool);
1630
1631 #ifdef SVN_USE_DOS_PATHS
1632 /* Handle a specific case on Windows where path == "X:/". Here we have to
1633 append the final '/', as svn_path_canonicalize will chop this of. */
1634 if (((dirent[0] >= 'A' && dirent[0] <= 'Z') ||
1635 (dirent[0] >= 'a' && dirent[0] <= 'z')) &&
1636 dirent[1] == ':' && dirent[2] == '/' &&
1637 dst[3] == '\0')
1638 {
1639 char *dst_slash = apr_pcalloc(pool, 4);
1640 dst_slash[0] = canonicalize_to_upper(dirent[0]);
1641 dst_slash[1] = ':';
1642 dst_slash[2] = '/';
1643 dst_slash[3] = '\0';
1644
1645 return dst_slash;
1646 }
1647 #endif /* SVN_USE_DOS_PATHS */
1648
1649 return dst;
1650 }
1651
1652 svn_boolean_t
svn_dirent_is_canonical(const char * dirent,apr_pool_t * scratch_pool)1653 svn_dirent_is_canonical(const char *dirent, apr_pool_t *scratch_pool)
1654 {
1655 const char *ptr = dirent;
1656 if (*ptr == '/')
1657 {
1658 ptr++;
1659 #ifdef SVN_USE_DOS_PATHS
1660 /* Check for UNC paths */
1661 if (*ptr == '/')
1662 {
1663 /* TODO: Scan hostname and sharename and fall back to part code */
1664
1665 /* ### Fall back to old implementation */
1666 return (strcmp(dirent, svn_dirent_canonicalize(dirent, scratch_pool))
1667 == 0);
1668 }
1669 #endif /* SVN_USE_DOS_PATHS */
1670 }
1671 #ifdef SVN_USE_DOS_PATHS
1672 else if (((*ptr >= 'a' && *ptr <= 'z') || (*ptr >= 'A' && *ptr <= 'Z')) &&
1673 (ptr[1] == ':'))
1674 {
1675 /* The only canonical drive names are "A:"..."Z:", no lower case */
1676 if (*ptr < 'A' || *ptr > 'Z')
1677 return FALSE;
1678
1679 ptr += 2;
1680
1681 if (*ptr == '/')
1682 ptr++;
1683 }
1684 #endif /* SVN_USE_DOS_PATHS */
1685
1686 return relpath_is_canonical(ptr);
1687 }
1688
1689 static svn_boolean_t
relpath_is_canonical(const char * relpath)1690 relpath_is_canonical(const char *relpath)
1691 {
1692 const char *ptr = relpath, *seg = relpath;
1693
1694 /* RELPATH is canonical if it has:
1695 * - no '.' segments
1696 * - no start and closing '/'
1697 * - no '//'
1698 */
1699
1700 if (*relpath == '\0')
1701 return TRUE;
1702
1703 if (*ptr == '/')
1704 return FALSE;
1705
1706 /* Now validate the rest of the path. */
1707 while(1)
1708 {
1709 apr_size_t seglen = ptr - seg;
1710
1711 if (seglen == 1 && *seg == '.')
1712 return FALSE; /* /./ */
1713
1714 if (*ptr == '/' && *(ptr+1) == '/')
1715 return FALSE; /* // */
1716
1717 if (! *ptr && *(ptr - 1) == '/')
1718 return FALSE; /* foo/ */
1719
1720 if (! *ptr)
1721 break;
1722
1723 if (*ptr == '/')
1724 ptr++;
1725 seg = ptr;
1726
1727 while (*ptr && (*ptr != '/'))
1728 ptr++;
1729 }
1730
1731 return TRUE;
1732 }
1733
1734 svn_boolean_t
svn_relpath_is_canonical(const char * relpath)1735 svn_relpath_is_canonical(const char *relpath)
1736 {
1737 return relpath_is_canonical(relpath);
1738 }
1739
1740 svn_boolean_t
svn_uri_is_canonical(const char * uri,apr_pool_t * scratch_pool)1741 svn_uri_is_canonical(const char *uri, apr_pool_t *scratch_pool)
1742 {
1743 const char *ptr = uri, *seg = uri;
1744 const char *schema_data = NULL;
1745
1746 /* URI is canonical if it has:
1747 * - lowercase URL scheme
1748 * - lowercase URL hostname
1749 * - no '.' segments
1750 * - no closing '/'
1751 * - no '//'
1752 * - uppercase hex-encoded pair digits ("%AB", not "%ab")
1753 */
1754
1755 if (*uri == '\0')
1756 return FALSE;
1757
1758 if (! svn_path_is_url(uri))
1759 return FALSE;
1760
1761 /* Skip the scheme. */
1762 while (*ptr && (*ptr != '/') && (*ptr != ':'))
1763 ptr++;
1764
1765 /* No scheme? No good. */
1766 if (! (*ptr == ':' && *(ptr+1) == '/' && *(ptr+2) == '/'))
1767 return FALSE;
1768
1769 /* Found a scheme, check that it's all lowercase. */
1770 ptr = uri;
1771 while (*ptr != ':')
1772 {
1773 if (*ptr >= 'A' && *ptr <= 'Z')
1774 return FALSE;
1775 ptr++;
1776 }
1777 /* Skip :// */
1778 ptr += 3;
1779
1780 /* Scheme only? That works. */
1781 if (! *ptr)
1782 return TRUE;
1783
1784 /* This might be the hostname */
1785 seg = ptr;
1786 while (*ptr && (*ptr != '/') && (*ptr != '@'))
1787 ptr++;
1788
1789 if (*ptr == '@')
1790 seg = ptr + 1;
1791
1792 /* Found a hostname, check that it's all lowercase. */
1793 ptr = seg;
1794
1795 if (*ptr == '[')
1796 {
1797 ptr++;
1798 while (*ptr == ':'
1799 || (*ptr >= '0' && *ptr <= '9')
1800 || (*ptr >= 'a' && *ptr <= 'f'))
1801 {
1802 ptr++;
1803 }
1804
1805 if (*ptr != ']')
1806 return FALSE;
1807 ptr++;
1808 }
1809 else
1810 while (*ptr && *ptr != '/' && *ptr != ':')
1811 {
1812 if (*ptr >= 'A' && *ptr <= 'Z')
1813 return FALSE;
1814 ptr++;
1815 }
1816
1817 /* Found a portnumber */
1818 if (*ptr == ':')
1819 {
1820 apr_int64_t port = 0;
1821
1822 ptr++;
1823 schema_data = ptr;
1824
1825 while (*ptr >= '0' && *ptr <= '9')
1826 {
1827 port = 10 * port + (*ptr - '0');
1828 ptr++;
1829 }
1830
1831 if (ptr == schema_data)
1832 return FALSE; /* Fail on "http://host:" */
1833
1834 if (*ptr && *ptr != '/')
1835 return FALSE; /* Not a port number */
1836
1837 if (port == 80 && strncmp(uri, "http:", 5) == 0)
1838 return FALSE;
1839 else if (port == 443 && strncmp(uri, "https:", 6) == 0)
1840 return FALSE;
1841 else if (port == 3690 && strncmp(uri, "svn:", 4) == 0)
1842 return FALSE;
1843 }
1844
1845 schema_data = ptr;
1846
1847 #ifdef SVN_USE_DOS_PATHS
1848 if (schema_data && *ptr == '/')
1849 {
1850 /* If this is a file url, ptr now points to the third '/' in
1851 file:///C:/path. Check that if we have such a URL the drive
1852 letter is in uppercase. */
1853 if (strncmp(uri, "file:", 5) == 0 &&
1854 ! (*(ptr+1) >= 'A' && *(ptr+1) <= 'Z') &&
1855 *(ptr+2) == ':')
1856 return FALSE;
1857 }
1858 #endif /* SVN_USE_DOS_PATHS */
1859
1860 /* Now validate the rest of the URI. */
1861 seg = ptr;
1862 while (*ptr && (*ptr != '/'))
1863 ptr++;
1864 while(1)
1865 {
1866 apr_size_t seglen = ptr - seg;
1867
1868 if (seglen == 1 && *seg == '.')
1869 return FALSE; /* /./ */
1870
1871 if (*ptr == '/' && *(ptr+1) == '/')
1872 return FALSE; /* // */
1873
1874 if (! *ptr && *(ptr - 1) == '/' && ptr - 1 != uri)
1875 return FALSE; /* foo/ */
1876
1877 if (! *ptr)
1878 break;
1879
1880 if (*ptr == '/')
1881 ptr++;
1882
1883 seg = ptr;
1884 while (*ptr && (*ptr != '/'))
1885 ptr++;
1886 }
1887
1888 ptr = schema_data;
1889
1890 while (*ptr)
1891 {
1892 if (*ptr == '%')
1893 {
1894 char digitz[3];
1895 int val;
1896
1897 /* Can't usesvn_ctype_isxdigit() because lower case letters are
1898 not in our canonical format */
1899 if (((*(ptr+1) < '0' || *(ptr+1) > '9'))
1900 && (*(ptr+1) < 'A' || *(ptr+1) > 'F'))
1901 return FALSE;
1902 else if (((*(ptr+2) < '0' || *(ptr+2) > '9'))
1903 && (*(ptr+2) < 'A' || *(ptr+2) > 'F'))
1904 return FALSE;
1905
1906 digitz[0] = *(++ptr);
1907 digitz[1] = *(++ptr);
1908 digitz[2] = '\0';
1909 val = (int)strtol(digitz, NULL, 16);
1910
1911 if (svn_uri__char_validity[val])
1912 return FALSE; /* Should not have been escaped */
1913 }
1914 else if (*ptr != '/' && !svn_uri__char_validity[(unsigned char)*ptr])
1915 return FALSE; /* Character should have been escaped */
1916 ptr++;
1917 }
1918
1919 return TRUE;
1920 }
1921
1922 svn_error_t *
svn_dirent_condense_targets(const char ** pcommon,apr_array_header_t ** pcondensed_targets,const apr_array_header_t * targets,svn_boolean_t remove_redundancies,apr_pool_t * result_pool,apr_pool_t * scratch_pool)1923 svn_dirent_condense_targets(const char **pcommon,
1924 apr_array_header_t **pcondensed_targets,
1925 const apr_array_header_t *targets,
1926 svn_boolean_t remove_redundancies,
1927 apr_pool_t *result_pool,
1928 apr_pool_t *scratch_pool)
1929 {
1930 int i, num_condensed = targets->nelts;
1931 svn_boolean_t *removed;
1932 apr_array_header_t *abs_targets;
1933
1934 /* Early exit when there's no data to work on. */
1935 if (targets->nelts <= 0)
1936 {
1937 *pcommon = NULL;
1938 if (pcondensed_targets)
1939 *pcondensed_targets = NULL;
1940 return SVN_NO_ERROR;
1941 }
1942
1943 /* Get the absolute path of the first target. */
1944 SVN_ERR(svn_dirent_get_absolute(pcommon,
1945 APR_ARRAY_IDX(targets, 0, const char *),
1946 scratch_pool));
1947
1948 /* Early exit when there's only one dirent to work on. */
1949 if (targets->nelts == 1)
1950 {
1951 *pcommon = apr_pstrdup(result_pool, *pcommon);
1952 if (pcondensed_targets)
1953 *pcondensed_targets = apr_array_make(result_pool, 0,
1954 sizeof(const char *));
1955 return SVN_NO_ERROR;
1956 }
1957
1958 /* Copy the targets array, but with absolute dirents instead of
1959 relative. Also, find the pcommon argument by finding what is
1960 common in all of the absolute dirents. NOTE: This is not as
1961 efficient as it could be. The calculation of the basedir could
1962 be done in the loop below, which would save some calls to
1963 svn_dirent_get_longest_ancestor. I decided to do it this way
1964 because I thought it would be simpler, since this way, we don't
1965 even do the loop if we don't need to condense the targets. */
1966
1967 removed = apr_pcalloc(scratch_pool, (targets->nelts *
1968 sizeof(svn_boolean_t)));
1969 abs_targets = apr_array_make(scratch_pool, targets->nelts,
1970 sizeof(const char *));
1971
1972 APR_ARRAY_PUSH(abs_targets, const char *) = *pcommon;
1973
1974 for (i = 1; i < targets->nelts; ++i)
1975 {
1976 const char *rel = APR_ARRAY_IDX(targets, i, const char *);
1977 const char *absolute;
1978 SVN_ERR(svn_dirent_get_absolute(&absolute, rel, scratch_pool));
1979 APR_ARRAY_PUSH(abs_targets, const char *) = absolute;
1980 *pcommon = svn_dirent_get_longest_ancestor(*pcommon, absolute,
1981 scratch_pool);
1982 }
1983
1984 *pcommon = apr_pstrdup(result_pool, *pcommon);
1985
1986 if (pcondensed_targets != NULL)
1987 {
1988 size_t basedir_len;
1989
1990 if (remove_redundancies)
1991 {
1992 /* Find the common part of each pair of targets. If
1993 common part is equal to one of the dirents, the other
1994 is a child of it, and can be removed. If a target is
1995 equal to *pcommon, it can also be removed. */
1996
1997 /* First pass: when one non-removed target is a child of
1998 another non-removed target, remove the child. */
1999 for (i = 0; i < abs_targets->nelts; ++i)
2000 {
2001 int j;
2002
2003 if (removed[i])
2004 continue;
2005
2006 for (j = i + 1; j < abs_targets->nelts; ++j)
2007 {
2008 const char *abs_targets_i;
2009 const char *abs_targets_j;
2010 const char *ancestor;
2011
2012 if (removed[j])
2013 continue;
2014
2015 abs_targets_i = APR_ARRAY_IDX(abs_targets, i, const char *);
2016 abs_targets_j = APR_ARRAY_IDX(abs_targets, j, const char *);
2017
2018 ancestor = svn_dirent_get_longest_ancestor
2019 (abs_targets_i, abs_targets_j, scratch_pool);
2020
2021 if (*ancestor == '\0')
2022 continue;
2023
2024 if (strcmp(ancestor, abs_targets_i) == 0)
2025 {
2026 removed[j] = TRUE;
2027 num_condensed--;
2028 }
2029 else if (strcmp(ancestor, abs_targets_j) == 0)
2030 {
2031 removed[i] = TRUE;
2032 num_condensed--;
2033 }
2034 }
2035 }
2036
2037 /* Second pass: when a target is the same as *pcommon,
2038 remove the target. */
2039 for (i = 0; i < abs_targets->nelts; ++i)
2040 {
2041 const char *abs_targets_i = APR_ARRAY_IDX(abs_targets, i,
2042 const char *);
2043
2044 if ((strcmp(abs_targets_i, *pcommon) == 0) && (! removed[i]))
2045 {
2046 removed[i] = TRUE;
2047 num_condensed--;
2048 }
2049 }
2050 }
2051
2052 /* Now create the return array, and copy the non-removed items */
2053 basedir_len = strlen(*pcommon);
2054 *pcondensed_targets = apr_array_make(result_pool, num_condensed,
2055 sizeof(const char *));
2056
2057 for (i = 0; i < abs_targets->nelts; ++i)
2058 {
2059 const char *rel_item = APR_ARRAY_IDX(abs_targets, i, const char *);
2060
2061 /* Skip this if it's been removed. */
2062 if (removed[i])
2063 continue;
2064
2065 /* If a common prefix was found, condensed_targets are given
2066 relative to that prefix. */
2067 if (basedir_len > 0)
2068 {
2069 /* Only advance our pointer past a dirent separator if
2070 REL_ITEM isn't the same as *PCOMMON.
2071
2072 If *PCOMMON is a root dirent, basedir_len will already
2073 include the closing '/', so never advance the pointer
2074 here.
2075 */
2076 rel_item += basedir_len;
2077 if (rel_item[0] &&
2078 ! svn_dirent_is_root(*pcommon, basedir_len))
2079 rel_item++;
2080 }
2081
2082 APR_ARRAY_PUSH(*pcondensed_targets, const char *)
2083 = apr_pstrdup(result_pool, rel_item);
2084 }
2085 }
2086
2087 return SVN_NO_ERROR;
2088 }
2089
2090 svn_error_t *
svn_uri_condense_targets(const char ** pcommon,apr_array_header_t ** pcondensed_targets,const apr_array_header_t * targets,svn_boolean_t remove_redundancies,apr_pool_t * result_pool,apr_pool_t * scratch_pool)2091 svn_uri_condense_targets(const char **pcommon,
2092 apr_array_header_t **pcondensed_targets,
2093 const apr_array_header_t *targets,
2094 svn_boolean_t remove_redundancies,
2095 apr_pool_t *result_pool,
2096 apr_pool_t *scratch_pool)
2097 {
2098 int i, num_condensed = targets->nelts;
2099 apr_array_header_t *uri_targets;
2100 svn_boolean_t *removed;
2101
2102 /* Early exit when there's no data to work on. */
2103 if (targets->nelts <= 0)
2104 {
2105 *pcommon = NULL;
2106 if (pcondensed_targets)
2107 *pcondensed_targets = NULL;
2108 return SVN_NO_ERROR;
2109 }
2110
2111 *pcommon = svn_uri_canonicalize(APR_ARRAY_IDX(targets, 0, const char *),
2112 scratch_pool);
2113
2114 /* Early exit when there's only one uri to work on. */
2115 if (targets->nelts == 1)
2116 {
2117 *pcommon = apr_pstrdup(result_pool, *pcommon);
2118 if (pcondensed_targets)
2119 *pcondensed_targets = apr_array_make(result_pool, 0,
2120 sizeof(const char *));
2121 return SVN_NO_ERROR;
2122 }
2123
2124 /* Find the pcommon argument by finding what is common in all of the
2125 uris. NOTE: This is not as efficient as it could be. The calculation
2126 of the basedir could be done in the loop below, which would
2127 save some calls to svn_uri_get_longest_ancestor. I decided to do it
2128 this way because I thought it would be simpler, since this way, we don't
2129 even do the loop if we don't need to condense the targets. */
2130
2131 removed = apr_pcalloc(scratch_pool, (targets->nelts *
2132 sizeof(svn_boolean_t)));
2133 uri_targets = apr_array_make(scratch_pool, targets->nelts,
2134 sizeof(const char *));
2135
2136 APR_ARRAY_PUSH(uri_targets, const char *) = *pcommon;
2137
2138 for (i = 1; i < targets->nelts; ++i)
2139 {
2140 const char *uri = svn_uri_canonicalize(
2141 APR_ARRAY_IDX(targets, i, const char *),
2142 scratch_pool);
2143 APR_ARRAY_PUSH(uri_targets, const char *) = uri;
2144
2145 /* If the commonmost ancestor so far is empty, there's no point
2146 in continuing to search for a common ancestor at all. But
2147 we'll keep looping for the sake of canonicalizing the
2148 targets, I suppose. */
2149 if (**pcommon != '\0')
2150 *pcommon = svn_uri_get_longest_ancestor(*pcommon, uri,
2151 scratch_pool);
2152 }
2153
2154 *pcommon = apr_pstrdup(result_pool, *pcommon);
2155
2156 if (pcondensed_targets != NULL)
2157 {
2158 size_t basedir_len;
2159
2160 if (remove_redundancies)
2161 {
2162 /* Find the common part of each pair of targets. If
2163 common part is equal to one of the dirents, the other
2164 is a child of it, and can be removed. If a target is
2165 equal to *pcommon, it can also be removed. */
2166
2167 /* First pass: when one non-removed target is a child of
2168 another non-removed target, remove the child. */
2169 for (i = 0; i < uri_targets->nelts; ++i)
2170 {
2171 int j;
2172
2173 if (removed[i])
2174 continue;
2175
2176 for (j = i + 1; j < uri_targets->nelts; ++j)
2177 {
2178 const char *uri_i;
2179 const char *uri_j;
2180 const char *ancestor;
2181
2182 if (removed[j])
2183 continue;
2184
2185 uri_i = APR_ARRAY_IDX(uri_targets, i, const char *);
2186 uri_j = APR_ARRAY_IDX(uri_targets, j, const char *);
2187
2188 ancestor = svn_uri_get_longest_ancestor(uri_i,
2189 uri_j,
2190 scratch_pool);
2191
2192 if (*ancestor == '\0')
2193 continue;
2194
2195 if (strcmp(ancestor, uri_i) == 0)
2196 {
2197 removed[j] = TRUE;
2198 num_condensed--;
2199 }
2200 else if (strcmp(ancestor, uri_j) == 0)
2201 {
2202 removed[i] = TRUE;
2203 num_condensed--;
2204 }
2205 }
2206 }
2207
2208 /* Second pass: when a target is the same as *pcommon,
2209 remove the target. */
2210 for (i = 0; i < uri_targets->nelts; ++i)
2211 {
2212 const char *uri_targets_i = APR_ARRAY_IDX(uri_targets, i,
2213 const char *);
2214
2215 if ((strcmp(uri_targets_i, *pcommon) == 0) && (! removed[i]))
2216 {
2217 removed[i] = TRUE;
2218 num_condensed--;
2219 }
2220 }
2221 }
2222
2223 /* Now create the return array, and copy the non-removed items */
2224 basedir_len = strlen(*pcommon);
2225 *pcondensed_targets = apr_array_make(result_pool, num_condensed,
2226 sizeof(const char *));
2227
2228 for (i = 0; i < uri_targets->nelts; ++i)
2229 {
2230 const char *rel_item = APR_ARRAY_IDX(uri_targets, i, const char *);
2231
2232 /* Skip this if it's been removed. */
2233 if (removed[i])
2234 continue;
2235
2236 /* If a common prefix was found, condensed_targets are given
2237 relative to that prefix. */
2238 if (basedir_len > 0)
2239 {
2240 /* Only advance our pointer past a dirent separator if
2241 REL_ITEM isn't the same as *PCOMMON.
2242
2243 If *PCOMMON is a root dirent, basedir_len will already
2244 include the closing '/', so never advance the pointer
2245 here.
2246 */
2247 rel_item += basedir_len;
2248 if ((rel_item[0] == '/') ||
2249 (rel_item[0] && !svn_uri_is_root(*pcommon, basedir_len)))
2250 {
2251 rel_item++;
2252 }
2253 }
2254
2255 APR_ARRAY_PUSH(*pcondensed_targets, const char *)
2256 = svn_path_uri_decode(rel_item, result_pool);
2257 }
2258 }
2259
2260 return SVN_NO_ERROR;
2261 }
2262
2263 svn_error_t *
svn_dirent_is_under_root(svn_boolean_t * under_root,const char ** result_path,const char * base_path,const char * path,apr_pool_t * result_pool)2264 svn_dirent_is_under_root(svn_boolean_t *under_root,
2265 const char **result_path,
2266 const char *base_path,
2267 const char *path,
2268 apr_pool_t *result_pool)
2269 {
2270 apr_status_t status;
2271 char *full_path;
2272
2273 *under_root = FALSE;
2274 if (result_path)
2275 *result_path = NULL;
2276
2277 status = apr_filepath_merge(&full_path,
2278 base_path,
2279 path,
2280 APR_FILEPATH_NOTABOVEROOT
2281 | APR_FILEPATH_SECUREROOTTEST,
2282 result_pool);
2283
2284 if (status == APR_SUCCESS)
2285 {
2286 if (result_path)
2287 *result_path = svn_dirent_canonicalize(full_path, result_pool);
2288 *under_root = TRUE;
2289 return SVN_NO_ERROR;
2290 }
2291 else if (status == APR_EABOVEROOT)
2292 {
2293 *under_root = FALSE;
2294 return SVN_NO_ERROR;
2295 }
2296
2297 return svn_error_wrap_apr(status, NULL);
2298 }
2299
2300 svn_error_t *
svn_uri_get_dirent_from_file_url(const char ** dirent,const char * url,apr_pool_t * pool)2301 svn_uri_get_dirent_from_file_url(const char **dirent,
2302 const char *url,
2303 apr_pool_t *pool)
2304 {
2305 const char *hostname, *path;
2306
2307 SVN_ERR_ASSERT(svn_uri_is_canonical(url, pool));
2308
2309 /* Verify that the URL is well-formed (loosely) */
2310
2311 /* First, check for the "file://" prefix. */
2312 if (strncmp(url, "file://", 7) != 0)
2313 return svn_error_createf(SVN_ERR_RA_ILLEGAL_URL, NULL,
2314 _("Local URL '%s' does not contain 'file://' "
2315 "prefix"), url);
2316
2317 /* Find the HOSTNAME portion and the PATH portion of the URL. The host
2318 name is between the "file://" prefix and the next occurence of '/'. We
2319 are considering everything from that '/' until the end of the URL to be
2320 the absolute path portion of the URL.
2321 If we got just "file://", treat it the same as "file:///". */
2322 hostname = url + 7;
2323 path = strchr(hostname, '/');
2324 if (path)
2325 hostname = apr_pstrmemdup(pool, hostname, path - hostname);
2326 else
2327 path = "/";
2328
2329 /* URI-decode HOSTNAME, and set it to NULL if it is "" or "localhost". */
2330 if (*hostname == '\0')
2331 hostname = NULL;
2332 else
2333 {
2334 hostname = svn_path_uri_decode(hostname, pool);
2335 if (strcmp(hostname, "localhost") == 0)
2336 hostname = NULL;
2337 }
2338
2339 /* Duplicate the URL, starting at the top of the path.
2340 At the same time, we URI-decode the path. */
2341 #ifdef SVN_USE_DOS_PATHS
2342 /* On Windows, we'll typically have to skip the leading / if the
2343 path starts with a drive letter. Like most Web browsers, We
2344 support two variants of this scheme:
2345
2346 file:///X:/path and
2347 file:///X|/path
2348
2349 Note that, at least on WinNT and above, file:////./X:/path will
2350 also work, so we must make sure the transformation doesn't break
2351 that, and file:///path (that looks within the current drive
2352 only) should also keep working.
2353 If we got a non-empty hostname other than localhost, we convert this
2354 into an UNC path. In this case, we obviously don't strip the slash
2355 even if the path looks like it starts with a drive letter.
2356 */
2357 {
2358 static const char valid_drive_letters[] =
2359 "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz";
2360 /* Casting away const! */
2361 char *dup_path = (char *)svn_path_uri_decode(path, pool);
2362
2363 /* This check assumes ':' and '|' are already decoded! */
2364 if (!hostname && dup_path[1] && strchr(valid_drive_letters, dup_path[1])
2365 && (dup_path[2] == ':' || dup_path[2] == '|'))
2366 {
2367 /* Skip the leading slash. */
2368 ++dup_path;
2369
2370 if (dup_path[1] == '|')
2371 dup_path[1] = ':';
2372
2373 if (dup_path[2] == '/' || dup_path[2] == '\0')
2374 {
2375 if (dup_path[2] == '\0')
2376 {
2377 /* A valid dirent for the driveroot must be like "C:/" instead of
2378 just "C:" or svn_dirent_join() will use the current directory
2379 on the drive instead */
2380 char *new_path = apr_pcalloc(pool, 4);
2381 new_path[0] = dup_path[0];
2382 new_path[1] = ':';
2383 new_path[2] = '/';
2384 new_path[3] = '\0';
2385 dup_path = new_path;
2386 }
2387 }
2388 }
2389 if (hostname)
2390 {
2391 if (dup_path[0] == '/' && dup_path[1] == '\0')
2392 return svn_error_createf(SVN_ERR_RA_ILLEGAL_URL, NULL,
2393 _("Local URL '%s' contains only a hostname, "
2394 "no path"), url);
2395
2396 /* We still know that the path starts with a slash. */
2397 *dirent = apr_pstrcat(pool, "//", hostname, dup_path, NULL);
2398 }
2399 else
2400 *dirent = dup_path;
2401 }
2402 #else /* !SVN_USE_DOS_PATHS */
2403 /* Currently, the only hostnames we are allowing on non-Win32 platforms
2404 are the empty string and 'localhost'. */
2405 if (hostname)
2406 return svn_error_createf(SVN_ERR_RA_ILLEGAL_URL, NULL,
2407 _("Local URL '%s' contains unsupported hostname"),
2408 url);
2409
2410 *dirent = svn_path_uri_decode(path, pool);
2411 #endif /* SVN_USE_DOS_PATHS */
2412 return SVN_NO_ERROR;
2413 }
2414
2415 svn_error_t *
svn_uri_get_file_url_from_dirent(const char ** url,const char * dirent,apr_pool_t * pool)2416 svn_uri_get_file_url_from_dirent(const char **url,
2417 const char *dirent,
2418 apr_pool_t *pool)
2419 {
2420 assert(svn_dirent_is_canonical(dirent, pool));
2421
2422 SVN_ERR(svn_dirent_get_absolute(&dirent, dirent, pool));
2423
2424 dirent = svn_path_uri_encode(dirent, pool);
2425
2426 #ifndef SVN_USE_DOS_PATHS
2427 if (dirent[0] == '/' && dirent[1] == '\0')
2428 dirent = NULL; /* "file://" is the canonical form of "file:///" */
2429
2430 *url = apr_pstrcat(pool, "file://", dirent, (char *)NULL);
2431 #else
2432 if (dirent[0] == '/')
2433 {
2434 /* Handle UNC paths //server/share -> file://server/share */
2435 assert(dirent[1] == '/'); /* Expect UNC, not non-absolute */
2436
2437 *url = apr_pstrcat(pool, "file:", dirent, NULL);
2438 }
2439 else
2440 {
2441 char *uri = apr_pstrcat(pool, "file:///", dirent, NULL);
2442 apr_size_t len = 8 /* strlen("file:///") */ + strlen(dirent);
2443
2444 /* "C:/" is a canonical dirent on Windows,
2445 but "file:///C:/" is not a canonical uri */
2446 if (uri[len-1] == '/')
2447 uri[len-1] = '\0';
2448
2449 *url = uri;
2450 }
2451 #endif
2452
2453 return SVN_NO_ERROR;
2454 }
2455
2456
2457
2458 /* -------------- The fspath API (see private/svn_fspath.h) -------------- */
2459
2460 svn_boolean_t
svn_fspath__is_canonical(const char * fspath)2461 svn_fspath__is_canonical(const char *fspath)
2462 {
2463 return fspath[0] == '/' && relpath_is_canonical(fspath + 1);
2464 }
2465
2466
2467 const char *
svn_fspath__canonicalize(const char * fspath,apr_pool_t * pool)2468 svn_fspath__canonicalize(const char *fspath,
2469 apr_pool_t *pool)
2470 {
2471 if ((fspath[0] == '/') && (fspath[1] == '\0'))
2472 return "/";
2473
2474 return apr_pstrcat(pool, "/", svn_relpath_canonicalize(fspath, pool),
2475 (char *)NULL);
2476 }
2477
2478
2479 svn_boolean_t
svn_fspath__is_root(const char * fspath,apr_size_t len)2480 svn_fspath__is_root(const char *fspath, apr_size_t len)
2481 {
2482 /* directory is root if it's equal to '/' */
2483 return (len == 1 && fspath[0] == '/');
2484 }
2485
2486
2487 const char *
svn_fspath__skip_ancestor(const char * parent_fspath,const char * child_fspath)2488 svn_fspath__skip_ancestor(const char *parent_fspath,
2489 const char *child_fspath)
2490 {
2491 assert(svn_fspath__is_canonical(parent_fspath));
2492 assert(svn_fspath__is_canonical(child_fspath));
2493
2494 return svn_relpath_skip_ancestor(parent_fspath + 1, child_fspath + 1);
2495 }
2496
2497
2498 const char *
svn_fspath__dirname(const char * fspath,apr_pool_t * pool)2499 svn_fspath__dirname(const char *fspath,
2500 apr_pool_t *pool)
2501 {
2502 assert(svn_fspath__is_canonical(fspath));
2503
2504 if (fspath[0] == '/' && fspath[1] == '\0')
2505 return apr_pstrdup(pool, fspath);
2506 else
2507 return apr_pstrcat(pool, "/", svn_relpath_dirname(fspath + 1, pool),
2508 (char *)NULL);
2509 }
2510
2511
2512 const char *
svn_fspath__basename(const char * fspath,apr_pool_t * pool)2513 svn_fspath__basename(const char *fspath,
2514 apr_pool_t *pool)
2515 {
2516 const char *result;
2517 assert(svn_fspath__is_canonical(fspath));
2518
2519 result = svn_relpath_basename(fspath + 1, pool);
2520
2521 assert(strchr(result, '/') == NULL);
2522 return result;
2523 }
2524
2525 void
svn_fspath__split(const char ** dirpath,const char ** base_name,const char * fspath,apr_pool_t * result_pool)2526 svn_fspath__split(const char **dirpath,
2527 const char **base_name,
2528 const char *fspath,
2529 apr_pool_t *result_pool)
2530 {
2531 assert(dirpath != base_name);
2532
2533 if (dirpath)
2534 *dirpath = svn_fspath__dirname(fspath, result_pool);
2535
2536 if (base_name)
2537 *base_name = svn_fspath__basename(fspath, result_pool);
2538 }
2539
2540 char *
svn_fspath__join(const char * fspath,const char * relpath,apr_pool_t * result_pool)2541 svn_fspath__join(const char *fspath,
2542 const char *relpath,
2543 apr_pool_t *result_pool)
2544 {
2545 char *result;
2546 assert(svn_fspath__is_canonical(fspath));
2547 assert(svn_relpath_is_canonical(relpath));
2548
2549 if (relpath[0] == '\0')
2550 result = apr_pstrdup(result_pool, fspath);
2551 else if (fspath[1] == '\0')
2552 result = apr_pstrcat(result_pool, "/", relpath, (char *)NULL);
2553 else
2554 result = apr_pstrcat(result_pool, fspath, "/", relpath, (char *)NULL);
2555
2556 assert(svn_fspath__is_canonical(result));
2557 return result;
2558 }
2559
2560 char *
svn_fspath__get_longest_ancestor(const char * fspath1,const char * fspath2,apr_pool_t * result_pool)2561 svn_fspath__get_longest_ancestor(const char *fspath1,
2562 const char *fspath2,
2563 apr_pool_t *result_pool)
2564 {
2565 char *result;
2566 assert(svn_fspath__is_canonical(fspath1));
2567 assert(svn_fspath__is_canonical(fspath2));
2568
2569 result = apr_pstrcat(result_pool, "/",
2570 svn_relpath_get_longest_ancestor(fspath1 + 1,
2571 fspath2 + 1,
2572 result_pool),
2573 (char *)NULL);
2574
2575 assert(svn_fspath__is_canonical(result));
2576 return result;
2577 }
2578
2579
2580
2581
2582 /* -------------- The urlpath API (see private/svn_fspath.h) ------------- */
2583
2584 const char *
svn_urlpath__canonicalize(const char * uri,apr_pool_t * pool)2585 svn_urlpath__canonicalize(const char *uri,
2586 apr_pool_t *pool)
2587 {
2588 if (svn_path_is_url(uri))
2589 {
2590 uri = svn_uri_canonicalize(uri, pool);
2591 }
2592 else
2593 {
2594 uri = svn_fspath__canonicalize(uri, pool);
2595 /* Do a little dance to normalize hex encoding. */
2596 uri = svn_path_uri_decode(uri, pool);
2597 uri = svn_path_uri_encode(uri, pool);
2598 }
2599 return uri;
2600 }
2601
2602
2603 /* -------------- The cert API (see private/svn_cert.h) ------------- */
2604
2605 svn_boolean_t
svn_cert__match_dns_identity(svn_string_t * pattern,svn_string_t * hostname)2606 svn_cert__match_dns_identity(svn_string_t *pattern, svn_string_t *hostname)
2607 {
2608 apr_size_t pattern_pos = 0, hostname_pos = 0;
2609
2610 /* support leading wildcards that composed of the only character in the
2611 * left-most label. */
2612 if (pattern->len >= 2 &&
2613 pattern->data[pattern_pos] == '*' &&
2614 pattern->data[pattern_pos + 1] == '.')
2615 {
2616 while (hostname_pos < hostname->len &&
2617 hostname->data[hostname_pos] != '.')
2618 {
2619 hostname_pos++;
2620 }
2621 /* Assume that the wildcard must match something. Rule 2 says
2622 * that *.example.com should not match example.com. If the wildcard
2623 * ends up not matching anything then it matches .example.com which
2624 * seems to be essentially the same as just example.com */
2625 if (hostname_pos == 0)
2626 return FALSE;
2627
2628 pattern_pos++;
2629 }
2630
2631 while (pattern_pos < pattern->len && hostname_pos < hostname->len)
2632 {
2633 char pattern_c = pattern->data[pattern_pos];
2634 char hostname_c = hostname->data[hostname_pos];
2635
2636 /* fold case as described in RFC 4343.
2637 * Note: We actually convert to lowercase, since our URI
2638 * canonicalization code converts to lowercase and generally
2639 * most certs are issued with lowercase DNS names, meaning
2640 * this avoids the fold operation in most cases. The RFC
2641 * suggests the opposite transformation, but doesn't require
2642 * any specific implementation in any case. It is critical
2643 * that this folding be locale independent so you can't use
2644 * tolower(). */
2645 pattern_c = canonicalize_to_lower(pattern_c);
2646 hostname_c = canonicalize_to_lower(hostname_c);
2647
2648 if (pattern_c != hostname_c)
2649 {
2650 /* doesn't match */
2651 return FALSE;
2652 }
2653 else
2654 {
2655 /* characters match so skip both */
2656 pattern_pos++;
2657 hostname_pos++;
2658 }
2659 }
2660
2661 /* ignore a trailing period on the hostname since this has no effect on the
2662 * security of the matching. See the following for the long explanation as
2663 * to why:
2664 * https://bugzilla.mozilla.org/show_bug.cgi?id=134402#c28
2665 */
2666 if (pattern_pos == pattern->len &&
2667 hostname_pos == hostname->len - 1 &&
2668 hostname->data[hostname_pos] == '.')
2669 hostname_pos++;
2670
2671 if (pattern_pos != pattern->len || hostname_pos != hostname->len)
2672 {
2673 /* end didn't match */
2674 return FALSE;
2675 }
2676
2677 return TRUE;
2678 }
2679