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 FLAG_BASECHAR 1
54 #define FLAG_ILLEGAL 4
55 #define FLAG_POSSIBLE1 8
56 #define FLAG_POSSIBLE2 16
57 #define FLAG_POSSIBLE3 32
59 #define CACHE_SIZE 8192
61 /* these tables are used to provide fast tests for characters */
62 static unsigned char char_flags[256];
64 /* we will use a very simple direct mapped prefix cache. The big
65 advantage of this cache structure is speed and low memory usage
67 The cache is indexed by the low-order bits of the hash, and confirmed by
68 hashing the resulting cache entry to match the known hash
70 static char **prefix_cache;
72 /* these are the characters we use in the 8.3 hash. Must be 36 chars long */
73 const char *basechars = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ";
74 static unsigned char base_reverse[256];
75 #define base_forward(v) basechars[v]
77 /* the list of reserved dos names - all of these are illegal */
78 const char *reserved_names[] = { "AUX", "LOCK$", "CON", "COM1", "COM2", "COM3", "COM4",
79 "LPT1", "LPT2", "LPT3", "NUL", "PRN", NULL };
82 hash a string of the specified length. The string does not need to be
85 this hash needs to be fast with a low collision rate (what hash doesn't?)
87 static u32 mangle_hash(const char *key, unsigned length)
93 strncpy(str, key, length);
97 /* the length of a multi-byte string can change after a strupper */
100 /* Set the initial value from the key size. */
101 for (value = 0x238F13AF * length, i=0; i < length; i++) {
102 value = (value + (((unsigned char)str[i]) << (i*5 % 24)));
105 return (1103515243 * value + 12345);
109 initialise the prefix cache
111 static BOOL cache_init(void)
113 if (prefix_cache) return True;
115 prefix_cache = malloc(sizeof(char *) * CACHE_SIZE);
116 if (!prefix_cache) return False;
118 memset(prefix_cache, 0, sizeof(char *) * CACHE_SIZE);
123 insert an entry into the prefix cache. The string may not be null terminated
125 static void cache_insert(const char *prefix, int length, u32 hash)
127 int i = hash % CACHE_SIZE;
129 if (prefix_cache[i]) {
130 free(prefix_cache[i]);
133 prefix_cache[i] = strndup(prefix, length);
137 lookup an entry in the prefix cache. Return NULL if not found.
139 static const char *cache_lookup(u32 hash)
141 int i = hash % CACHE_SIZE;
143 if (!prefix_cache[i]) return NULL;
145 /* we have a possible match - compute the hash to confirm */
146 if (hash != mangle_hash(prefix_cache[i], strlen(prefix_cache[i]))) {
150 /* yep, it matched */
151 return prefix_cache[i];
156 determine if a string is possibly in a mangled format, ignoring
159 In this algorithm, mangled names use only pure ascii characters (no
160 multi-byte) so we can avoid doing a UCS2 conversion
162 static BOOL is_mangled(const char *name)
164 /* the best distinguishing characteristic is the ~ */
165 if (name[6] != '~') return False;
167 /* check extension */
168 /* check first character */
169 /* check rest of hash */
176 see if a filename is an allowable 8.3 name.
178 we are only going to allow ascii characters in 8.3 names, as this
179 simplifies things greatly (it means that we know the string won't
180 get larger when converted from UNIX to DOS formats)
182 static BOOL is_8_3(const char *name, BOOL check_case)
187 /* as a special case, the names '.' and '..' are allowable 8.3 names */
188 if (name[0] == '.') {
189 if (!name[1] || (name[1] == '.' && !name[2])) {
194 /* the simplest test is on the overall length of the
195 filename. Note that we deliberately use the ascii string
196 length (not the multi-byte one) as it is faster, and gives us
197 the result we need in this case. Using strlen_m would not
198 only be slower, it would be incorrect */
200 if (len > 12) return False;
202 /* find the '.'. Note that once again we use the non-multibyte
204 dot_p = strchr(name, '.');
207 /* if the name doesn't contain a '.' then its length
208 must be less than 8 */
213 int prefix_len, suffix_len;
215 /* if it does contain a dot then the prefix must be <=
216 8 and the suffix <= 3 in length */
217 prefix_len = PTR_DIFF(dot_p, name);
218 suffix_len = len - (prefix_len+1);
220 if (prefix_len > 8 || suffix_len > 3) {
224 /* a 8.3 name cannot contain more than 1 '.' */
225 if (strchr(dot_p+1, '.')) {
230 /* the length are all OK. Now check to see if the characters themselves are OK */
231 for (i=0; name[i]; i++) {
232 if (!(char_flags[(unsigned)(name[i])] & FLAG_ASCII)) {
237 /* it is a good 8.3 name */
243 reset the mangling cache on a smb.conf reload. This only really makes sense for
244 mangling backends that have parameters in smb.conf, and as this backend doesn't
245 this is a NULL operation
247 static void mangle_reset(void)
254 try to find a 8.3 name in the cache, and if found then
255 replace the string with the original long name.
257 The filename must be able to hold at least sizeof(fstring)
259 static BOOL check_cache(char *name)
261 u32 hash, multiplier;
266 /* make sure that this is a mangled name from this cache */
267 if (!is_mangled(name)) {
271 /* we need to extract the hash from the 8.3 name */
272 hash = base_reverse[(unsigned char)name[7]];
273 for (multiplier=36, i=5;i>=1;i--) {
274 u32 v = base_reverse[(unsigned char)name[i]];
275 hash += multiplier * v;
279 /* now look in the prefix cache for that hash */
280 prefix = cache_lookup(hash);
285 /* we found it - construct the full name */
286 strncpy(extension, name+9, 3);
289 slprintf(name, sizeof(fstring), "%s.%s", prefix, extension);
291 fstrcpy(name, prefix);
298 look for a DOS reserved name
300 static BOOL is_reserved_name(const char *name)
302 if ((char_flags[(unsigned char)name[0]] & FLAG_POSSIBLE1) &&
303 (char_flags[(unsigned char)name[1]] & FLAG_POSSIBLE2) &&
304 (char_flags[(unsigned char)name[2]] & FLAG_POSSIBLE3)) {
305 /* a likely match, scan the lot */
307 for (i=0; reserved_names[i]; i++) {
308 int len = strlen(reserved_names[i]);
309 /* note that we match on COM1 as well as COM1.foo */
310 if (strncasecmp(name, reserved_names[i], len) == 0 &&
311 (name[len] == '.' || name[len] == 0)) {
321 see if a filename is a legal long filename
323 static BOOL is_legal_name(const char *name)
326 if (char_flags[(unsigned char)*name] & FLAG_ILLEGAL) {
336 the main forward mapping function, which converts a long filename to
339 if need83 is not set then we only do the mangling if the name is illegal
342 if cache83 is not set then we don't cache the result
344 the name parameter must be able to hold 13 bytes
346 static BOOL name_map(char *name, BOOL need83, BOOL cache83)
351 int extension_length, i;
356 /* reserved names are handled specially */
357 if (!is_reserved_name(name)) {
358 /* if the name is already a valid 8.3 name then we don't need to
360 if (is_8_3(name, False)) {
364 /* if the caller doesn't strictly need 8.3 then just check for illegal
366 if (!need83 && is_legal_name(name)) {
371 /* find the '.' if any */
372 dot_p = strchr(name, '.');
374 /* the leading character in the mangled name is taken from
375 the first character of the name, if it is ascii
376 otherwise '_' is used
379 if (! (char_flags[(unsigned char)lead_char] & FLAG_ASCII)) {
383 /* the prefix is anything up to the first dot */
385 prefix_len = PTR_DIFF(dot_p, name);
387 prefix_len = strlen(name);
390 /* the extension of the mangled name is taken from the first 3
391 ascii chars after the dot */
392 extension_length = 0;
394 for (i=1; extension_length < 3 && dot_p[i]; i++) {
396 if (char_flags[(unsigned char)c] & FLAG_ASCII) {
397 extension[extension_length++] = c;
402 /* find the hash for this prefix */
403 v = hash = mangle_hash(name, prefix_len);
405 /* now form the mangled name */
406 new_name[0] = lead_char;
407 new_name[7] = base_forward(v % 36);
409 for (i=5; i>=1; i--) {
411 new_name[i] = base_forward(v % 36);
414 /* add the extension */
415 if (extension_length) {
417 memcpy(&new_name[9], extension, extension_length);
418 new_name[9+extension_length] = 0;
424 /* put it in the cache */
425 cache_insert(name, prefix_len, hash);
428 /* and overwrite the old name */
429 fstrcpy(name, new_name);
431 /* all done, we've managed to mangle it */
436 /* initialise the flags table
438 we allow only a very restricted set of characters as 'ascii' in this
439 mangling backend. This isn't a significant problem as modern clients
440 use the 'long' filenames anyway, and those don't have these
443 static void init_tables(void)
447 memset(char_flags, 0, sizeof(char_flags));
449 for (i=0;i<128;i++) {
450 if ((i >= '0' && i <= '9') ||
451 (i >= 'a' && i <= 'z') ||
452 (i >= 'A' && i <= 'Z')) {
453 char_flags[i] |= (FLAG_ASCII | FLAG_BASECHAR);
455 if (strchr("._-$~", i)) {
456 char_flags[i] |= FLAG_ASCII;
459 if (strchr("*\\/?<>|\":", i)) {
460 char_flags[i] |= FLAG_ILLEGAL;
464 memset(base_reverse, 0, sizeof(base_reverse));
466 base_reverse[(unsigned char)base_forward(i)] = i;
469 /* fill in the reserved names flags. These are used as a very
470 fast filter for finding possible DOS reserved filenames */
471 for (i=0; reserved_names[i]; i++) {
472 unsigned char c1, c2, c3;
474 c1 = (unsigned char)reserved_names[i][0];
475 c2 = (unsigned char)reserved_names[i][1];
476 c3 = (unsigned char)reserved_names[i][2];
478 char_flags[c1] |= FLAG_POSSIBLE1;
479 char_flags[c2] |= FLAG_POSSIBLE2;
480 char_flags[c3] |= FLAG_POSSIBLE3;
481 char_flags[tolower(c1)] |= FLAG_POSSIBLE1;
482 char_flags[tolower(c2)] |= FLAG_POSSIBLE2;
483 char_flags[tolower(c3)] |= FLAG_POSSIBLE3;
489 the following provides the abstraction layer to make it easier
490 to drop in an alternative mangling implementation */
491 static struct mangle_fns mangle_fns = {
499 /* return the methods for this mangling implementation */
500 struct mangle_fns *mangle_hash2_init(void)