2 Unix SMB/CIFS implementation.
3 new hash based name mangling implementation
4 Copyright (C) Andrew Tridgell 2002
6 This program is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 This program is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with this program; if not, write to the Free Software
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
22 this mangling scheme uses the following format
26 where nnnnn is a base 36 hash, and A represents characters from the original string
28 The hash is taken of the leading part of the long filename, in uppercase
30 for simplicity, we only allow ascii characters in 8.3 names
35 ===============================================================================
38 This file deliberately uses non-multibyte string functions in many places. This
39 is *not* a mistake. This code is multi-byte safe, but it gets this property
40 through some very subtle knowledge of the way multi-byte strings are encoded
41 and the fact that this mangling algorithm only supports ascii characters in
44 please don't convert this file to use the *_m() functions!!
45 ===============================================================================
52 #define M_DEBUG(level, x) DEBUG(level, x)
54 #define M_DEBUG(level, x)
57 /* these flags are used to mark characters in as having particular
59 #define FLAG_BASECHAR 1
61 #define FLAG_ILLEGAL 4
62 #define FLAG_WILDCARD 8
64 /* the "possible" flags are used as a fast way to find possible DOS
66 #define FLAG_POSSIBLE1 16
67 #define FLAG_POSSIBLE2 32
68 #define FLAG_POSSIBLE3 64
69 #define FLAG_POSSIBLE4 128
71 /* by default have a max of 4096 entries in the cache. */
72 #ifndef MANGLE_CACHE_SIZE
73 #define MANGLE_CACHE_SIZE 4096
76 /* these tables are used to provide fast tests for characters */
77 static unsigned char char_flags[256];
79 #define FLAG_CHECK(c, flag) (char_flags[(unsigned char)(c)] & (flag))
81 /* we will use a very simple direct mapped prefix cache. The big
82 advantage of this cache structure is speed and low memory usage
84 The cache is indexed by the low-order bits of the hash, and confirmed by
85 hashing the resulting cache entry to match the known hash
87 static char **prefix_cache;
88 static u32 *prefix_cache_hashes;
90 /* these are the characters we use in the 8.3 hash. Must be 36 chars long */
91 const char *basechars = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ";
92 static unsigned char base_reverse[256];
93 #define base_forward(v) basechars[v]
95 /* the list of reserved dos names - all of these are illegal */
96 const char *reserved_names[] = { "AUX", "LOCK$", "CON", "COM1", "COM2", "COM3", "COM4",
97 "LPT1", "LPT2", "LPT3", "NUL", "PRN", NULL };
100 hash a string of the specified length. The string does not need to be
103 this hash needs to be fast with a low collision rate (what hash doesn't?)
105 static u32 mangle_hash(const char *key, unsigned length)
111 /* we have to uppercase here to ensure that the mangled name
112 doesn't depend on the case of the long name. Note that this
113 is the only place where we need to use a multi-byte string
115 strncpy(str, key, length);
119 /* the length of a multi-byte string can change after a strupper_m */
120 length = strlen(str);
122 /* Set the initial value from the key size. */
123 for (value = 0x238F13AF * length, i=0; i < length; i++) {
124 value = (value + (((unsigned char)str[i]) << (i*5 % 24)));
127 /* note that we force it to a 31 bit hash, to keep within the limits
128 of the 36^6 mangle space */
129 return (1103515243 * value + 12345) & ~0x80000000;
133 initialise (ie. allocate) the prefix cache
135 static BOOL cache_init(void)
137 if (prefix_cache) return True;
139 prefix_cache = malloc(sizeof(char *) * MANGLE_CACHE_SIZE);
140 if (!prefix_cache) return False;
142 prefix_cache_hashes = malloc(sizeof(u32) * MANGLE_CACHE_SIZE);
143 if (!prefix_cache_hashes) return False;
145 memset(prefix_cache, 0, sizeof(char *) * MANGLE_CACHE_SIZE);
146 memset(prefix_cache_hashes, 0, sizeof(char *) * MANGLE_CACHE_SIZE);
151 insert an entry into the prefix cache. The string might not be null
153 static void cache_insert(const char *prefix, int length, u32 hash)
155 int i = hash % MANGLE_CACHE_SIZE;
157 if (prefix_cache[i]) {
158 free(prefix_cache[i]);
161 prefix_cache[i] = strndup(prefix, length);
162 prefix_cache_hashes[i] = hash;
166 lookup an entry in the prefix cache. Return NULL if not found.
168 static const char *cache_lookup(u32 hash)
170 int i = hash % MANGLE_CACHE_SIZE;
172 if (!prefix_cache[i] || hash != prefix_cache_hashes[i]) {
176 /* yep, it matched */
177 return prefix_cache[i];
182 determine if a string is possibly in a mangled format, ignoring
185 In this algorithm, mangled names use only pure ascii characters (no
186 multi-byte) so we can avoid doing a UCS2 conversion
188 static BOOL is_mangled(const char *name)
192 M_DEBUG(0,("is_mangled %s ?\n", name));
194 /* the best distinguishing characteristic is the ~ */
195 if (name[6] != '~') return False;
197 /* check the length */
199 if (len > 12 || len < 8) return False;
201 /* check extension */
203 if (name[8] != '.') return False;
204 for (i=9; name[i]; i++) {
205 if (! FLAG_CHECK(name[i], FLAG_ASCII)) {
211 /* check first character */
212 if (! FLAG_CHECK(name[0], FLAG_ASCII)) {
216 /* check rest of hash */
217 if (! FLAG_CHECK(name[7], FLAG_BASECHAR)) {
221 if (! FLAG_CHECK(name[i], FLAG_BASECHAR)) {
226 M_DEBUG(0,("is_mangled %s -> yes\n", name));
233 see if a filename is an allowable 8.3 name.
235 we are only going to allow ascii characters in 8.3 names, as this
236 simplifies things greatly (it means that we know the string won't
237 get larger when converted from UNIX to DOS formats)
239 static BOOL is_8_3(const char *name, BOOL check_case)
244 /* as a special case, the names '.' and '..' are allowable 8.3 names */
245 if (name[0] == '.') {
246 if (!name[1] || (name[1] == '.' && !name[2])) {
251 /* the simplest test is on the overall length of the
252 filename. Note that we deliberately use the ascii string
253 length (not the multi-byte one) as it is faster, and gives us
254 the result we need in this case. Using strlen_m would not
255 only be slower, it would be incorrect */
257 if (len > 12) return False;
259 /* find the '.'. Note that once again we use the non-multibyte
261 dot_p = strchr(name, '.');
264 /* if the name doesn't contain a '.' then its length
265 must be less than 8 */
270 int prefix_len, suffix_len;
272 /* if it does contain a dot then the prefix must be <=
273 8 and the suffix <= 3 in length */
274 prefix_len = PTR_DIFF(dot_p, name);
275 suffix_len = len - (prefix_len+1);
277 if (prefix_len > 8 || suffix_len > 3) {
281 /* a 8.3 name cannot contain more than 1 '.' */
282 if (strchr(dot_p+1, '.')) {
287 /* the length are all OK. Now check to see if the characters themselves are OK */
288 for (i=0; name[i]; i++) {
289 /* note that we allow wildcard petterns! */
290 if (!FLAG_CHECK(name[i], FLAG_ASCII|FLAG_WILDCARD) && name[i] != '.') {
295 /* it is a good 8.3 name */
301 reset the mangling cache on a smb.conf reload. This only really makes sense for
302 mangling backends that have parameters in smb.conf, and as this backend doesn't
303 this is a NULL operation
305 static void mangle_reset(void)
312 try to find a 8.3 name in the cache, and if found then
313 replace the string with the original long name.
315 The filename must be able to hold at least sizeof(fstring)
317 static BOOL check_cache(char *name)
319 u32 hash, multiplier;
324 /* make sure that this is a mangled name from this cache */
325 if (!is_mangled(name)) {
326 M_DEBUG(0,("check_cache: %s -> not mangled\n", name));
330 /* we need to extract the hash from the 8.3 name */
331 hash = base_reverse[(unsigned char)name[7]];
332 for (multiplier=36, i=5;i>=1;i--) {
333 u32 v = base_reverse[(unsigned char)name[i]];
334 hash += multiplier * v;
338 /* now look in the prefix cache for that hash */
339 prefix = cache_lookup(hash);
341 M_DEBUG(0,("check_cache: %s -> %08X -> not found\n", name, hash));
345 /* we found it - construct the full name */
346 strncpy(extension, name+9, 3);
350 M_DEBUG(0,("check_cache: %s -> %s.%s\n", name, prefix, extension));
351 slprintf(name, sizeof(fstring), "%s.%s", prefix, extension);
353 M_DEBUG(0,("check_cache: %s -> %s\n", name, prefix));
354 fstrcpy(name, prefix);
362 look for a DOS reserved name
364 static BOOL is_reserved_name(const char *name)
366 if (FLAG_CHECK(name[0], FLAG_POSSIBLE1) &&
367 FLAG_CHECK(name[1], FLAG_POSSIBLE2) &&
368 FLAG_CHECK(name[2], FLAG_POSSIBLE3) &&
369 FLAG_CHECK(name[3], FLAG_POSSIBLE4)) {
370 /* a likely match, scan the lot */
372 for (i=0; reserved_names[i]; i++) {
373 int len = strlen(reserved_names[i]);
374 /* note that we match on COM1 as well as COM1.foo */
375 if (strncasecmp(name, reserved_names[i], len) == 0 &&
376 (name[len] == '.' || name[len] == 0)) {
386 see if a filename is a legal long filename
388 static BOOL is_legal_name(const char *name)
391 if (FLAG_CHECK(name[0], FLAG_ILLEGAL)) {
401 the main forward mapping function, which converts a long filename to
404 if need83 is not set then we only do the mangling if the name is illegal
407 if cache83 is not set then we don't cache the result
409 the name parameter must be able to hold 13 bytes
411 static BOOL name_map(char *name, BOOL need83, BOOL cache83)
416 int extension_length, i;
421 /* reserved names are handled specially */
422 if (!is_reserved_name(name)) {
423 /* if the name is already a valid 8.3 name then we don't need to
425 if (is_8_3(name, False)) {
429 /* if the caller doesn't strictly need 8.3 then just check for illegal
431 if (!need83 && is_legal_name(name)) {
436 /* find the '.' if any */
437 dot_p = strrchr(name, '.');
440 /* if the extension contains any illegal characters or
441 is too long or zero length then we treat it as part
443 for (i=0; i<4 && dot_p[i+1]; i++) {
444 if (! FLAG_CHECK(dot_p[i+1], FLAG_ASCII)) {
449 if (i == 0 || i == 4) dot_p = NULL;
452 /* the leading character in the mangled name is taken from
453 the first character of the name, if it is ascii
454 otherwise '_' is used
457 if (! FLAG_CHECK(lead_char, FLAG_ASCII)) {
460 lead_char = toupper(lead_char);
462 /* the prefix is anything up to the first dot */
464 prefix_len = PTR_DIFF(dot_p, name);
466 prefix_len = strlen(name);
469 /* the extension of the mangled name is taken from the first 3
470 ascii chars after the dot */
471 extension_length = 0;
473 for (i=1; extension_length < 3 && dot_p[i]; i++) {
475 if (FLAG_CHECK(c, FLAG_ASCII)) {
476 extension[extension_length++] = toupper(c);
481 /* find the hash for this prefix */
482 v = hash = mangle_hash(name, prefix_len);
484 /* now form the mangled name. */
485 new_name[0] = lead_char;
486 new_name[7] = base_forward(v % 36);
488 for (i=5; i>=1; i--) {
490 new_name[i] = base_forward(v % 36);
493 /* add the extension */
494 if (extension_length) {
496 memcpy(&new_name[9], extension, extension_length);
497 new_name[9+extension_length] = 0;
503 /* put it in the cache */
504 cache_insert(name, prefix_len, hash);
507 M_DEBUG(0,("name_map: %s -> %08X -> %s (cache=%d)\n",
508 name, hash, new_name, cache83));
510 /* and overwrite the old name */
511 fstrcpy(name, new_name);
513 /* all done, we've managed to mangle it */
518 /* initialise the flags table
520 we allow only a very restricted set of characters as 'ascii' in this
521 mangling backend. This isn't a significant problem as modern clients
522 use the 'long' filenames anyway, and those don't have these
525 static void init_tables(void)
529 memset(char_flags, 0, sizeof(char_flags));
531 for (i=0;i<128;i++) {
532 if ((i >= '0' && i <= '9') ||
533 (i >= 'a' && i <= 'z') ||
534 (i >= 'A' && i <= 'Z')) {
535 char_flags[i] |= (FLAG_ASCII | FLAG_BASECHAR);
537 if (strchr("_-$~", i)) {
538 char_flags[i] |= FLAG_ASCII;
541 if (strchr("*\\/?<>|\":", i)) {
542 char_flags[i] |= FLAG_ILLEGAL;
545 if (strchr("*?\"<>", i)) {
546 char_flags[i] |= FLAG_WILDCARD;
550 memset(base_reverse, 0, sizeof(base_reverse));
552 base_reverse[(unsigned char)base_forward(i)] = i;
555 /* fill in the reserved names flags. These are used as a very
556 fast filter for finding possible DOS reserved filenames */
557 for (i=0; reserved_names[i]; i++) {
558 unsigned char c1, c2, c3, c4;
560 c1 = (unsigned char)reserved_names[i][0];
561 c2 = (unsigned char)reserved_names[i][1];
562 c3 = (unsigned char)reserved_names[i][2];
563 c4 = (unsigned char)reserved_names[i][3];
565 char_flags[c1] |= FLAG_POSSIBLE1;
566 char_flags[c2] |= FLAG_POSSIBLE2;
567 char_flags[c3] |= FLAG_POSSIBLE3;
568 char_flags[c4] |= FLAG_POSSIBLE4;
569 char_flags[tolower(c1)] |= FLAG_POSSIBLE1;
570 char_flags[tolower(c2)] |= FLAG_POSSIBLE2;
571 char_flags[tolower(c3)] |= FLAG_POSSIBLE3;
572 char_flags[tolower(c4)] |= FLAG_POSSIBLE4;
574 char_flags[(unsigned char)'.'] |= FLAG_POSSIBLE4;
580 the following provides the abstraction layer to make it easier
581 to drop in an alternative mangling implementation */
582 static struct mangle_fns mangle_fns = {
590 /* return the methods for this mangling implementation */
591 struct mangle_fns *mangle_hash2_init(void)