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 static 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 static const char *reserved_names[] =
97 { "AUX", "LOCK$", "CON", "COM1", "COM2", "COM3", "COM4",
98 "LPT1", "LPT2", "LPT3", "NUL", "PRN", NULL };
101 hash a string of the specified length. The string does not need to be
104 this hash needs to be fast with a low collision rate (what hash doesn't?)
106 static u32 mangle_hash(const char *key, unsigned length)
112 /* we have to uppercase here to ensure that the mangled name
113 doesn't depend on the case of the long name. Note that this
114 is the only place where we need to use a multi-byte string
116 strncpy(str, key, length);
120 /* the length of a multi-byte string can change after a strupper_m */
121 length = strlen(str);
123 /* Set the initial value from the key size. */
124 for (value = 0x238F13AF * length, i=0; i < length; i++) {
125 value = (value + (((unsigned char)str[i]) << (i*5 % 24)));
128 /* note that we force it to a 31 bit hash, to keep within the limits
129 of the 36^6 mangle space */
130 return (1103515243 * value + 12345) & ~0x80000000;
134 initialise (ie. allocate) the prefix cache
136 static BOOL cache_init(void)
138 if (prefix_cache) return True;
140 prefix_cache = malloc(sizeof(char *) * MANGLE_CACHE_SIZE);
141 if (!prefix_cache) return False;
143 prefix_cache_hashes = malloc(sizeof(u32) * MANGLE_CACHE_SIZE);
144 if (!prefix_cache_hashes) return False;
146 memset(prefix_cache, 0, sizeof(char *) * MANGLE_CACHE_SIZE);
147 memset(prefix_cache_hashes, 0, sizeof(char *) * MANGLE_CACHE_SIZE);
152 insert an entry into the prefix cache. The string might not be null
154 static void cache_insert(const char *prefix, int length, u32 hash)
156 int i = hash % MANGLE_CACHE_SIZE;
158 if (prefix_cache[i]) {
159 free(prefix_cache[i]);
162 prefix_cache[i] = strndup(prefix, length);
163 prefix_cache_hashes[i] = hash;
167 lookup an entry in the prefix cache. Return NULL if not found.
169 static const char *cache_lookup(u32 hash)
171 int i = hash % MANGLE_CACHE_SIZE;
173 if (!prefix_cache[i] || hash != prefix_cache_hashes[i]) {
177 /* yep, it matched */
178 return prefix_cache[i];
183 determine if a string is possibly in a mangled format, ignoring
186 In this algorithm, mangled names use only pure ascii characters (no
187 multi-byte) so we can avoid doing a UCS2 conversion
189 static BOOL is_mangled_component(const char *name)
193 M_DEBUG(0,("is_mangled_component %s ?\n", name));
195 /* the best distinguishing characteristic is the ~ */
196 if (name[6] != '~') return False;
198 /* check the length */
200 if (len > 12 || len < 8) return False;
202 /* check extension */
204 if (name[8] != '.') return False;
205 for (i=9; name[i]; i++) {
206 if (! FLAG_CHECK(name[i], FLAG_ASCII)) {
212 /* check first character */
213 if (! FLAG_CHECK(name[0], FLAG_ASCII)) {
217 /* check rest of hash */
218 if (! FLAG_CHECK(name[7], FLAG_BASECHAR)) {
222 if (! FLAG_CHECK(name[i], FLAG_BASECHAR)) {
227 M_DEBUG(0,("is_mangled %s -> yes\n", name));
235 determine if a string is possibly in a mangled format, ignoring
238 In this algorithm, mangled names use only pure ascii characters (no
239 multi-byte) so we can avoid doing a UCS2 conversion
241 NOTE! This interface must be able to handle a path with unix
242 directory separators. It should return true if any component is
245 static BOOL is_mangled(const char *name)
250 M_DEBUG(0,("is_mangled %s ?\n", name));
252 for (s=name; (p=strchr(s, '/')); s=p+1) {
253 char *component = strndup(s, PTR_DIFF(p, s));
254 if (is_mangled_component(component)) {
261 /* and the last part ... */
262 return is_mangled_component(s);
267 see if a filename is an allowable 8.3 name.
269 we are only going to allow ascii characters in 8.3 names, as this
270 simplifies things greatly (it means that we know the string won't
271 get larger when converted from UNIX to DOS formats)
273 static BOOL is_8_3(const char *name, BOOL check_case)
278 /* as a special case, the names '.' and '..' are allowable 8.3 names */
279 if (name[0] == '.') {
280 if (!name[1] || (name[1] == '.' && !name[2])) {
285 /* the simplest test is on the overall length of the
286 filename. Note that we deliberately use the ascii string
287 length (not the multi-byte one) as it is faster, and gives us
288 the result we need in this case. Using strlen_m would not
289 only be slower, it would be incorrect */
291 if (len > 12) return False;
293 /* find the '.'. Note that once again we use the non-multibyte
295 dot_p = strchr(name, '.');
298 /* if the name doesn't contain a '.' then its length
299 must be less than 8 */
304 int prefix_len, suffix_len;
306 /* if it does contain a dot then the prefix must be <=
307 8 and the suffix <= 3 in length */
308 prefix_len = PTR_DIFF(dot_p, name);
309 suffix_len = len - (prefix_len+1);
311 if (prefix_len > 8 || suffix_len > 3) {
315 /* a 8.3 name cannot contain more than 1 '.' */
316 if (strchr(dot_p+1, '.')) {
321 /* the length are all OK. Now check to see if the characters themselves are OK */
322 for (i=0; name[i]; i++) {
323 /* note that we allow wildcard petterns! */
324 if (!FLAG_CHECK(name[i], FLAG_ASCII|FLAG_WILDCARD) && name[i] != '.') {
329 /* it is a good 8.3 name */
335 reset the mangling cache on a smb.conf reload. This only really makes sense for
336 mangling backends that have parameters in smb.conf, and as this backend doesn't
337 this is a NULL operation
339 static void mangle_reset(void)
346 try to find a 8.3 name in the cache, and if found then
347 replace the string with the original long name.
349 The filename must be able to hold at least sizeof(fstring)
351 static BOOL check_cache(char *name)
353 u32 hash, multiplier;
358 /* make sure that this is a mangled name from this cache */
359 if (!is_mangled(name)) {
360 M_DEBUG(0,("check_cache: %s -> not mangled\n", name));
364 /* we need to extract the hash from the 8.3 name */
365 hash = base_reverse[(unsigned char)name[7]];
366 for (multiplier=36, i=5;i>=1;i--) {
367 u32 v = base_reverse[(unsigned char)name[i]];
368 hash += multiplier * v;
372 /* now look in the prefix cache for that hash */
373 prefix = cache_lookup(hash);
375 M_DEBUG(0,("check_cache: %s -> %08X -> not found\n", name, hash));
379 /* we found it - construct the full name */
380 if (name[8] == '.') {
381 strncpy(extension, name+9, 3);
388 M_DEBUG(0,("check_cache: %s -> %s.%s\n", name, prefix, extension));
389 slprintf(name, sizeof(fstring), "%s.%s", prefix, extension);
391 M_DEBUG(0,("check_cache: %s -> %s\n", name, prefix));
392 fstrcpy(name, prefix);
400 look for a DOS reserved name
402 static BOOL is_reserved_name(const char *name)
404 if (FLAG_CHECK(name[0], FLAG_POSSIBLE1) &&
405 FLAG_CHECK(name[1], FLAG_POSSIBLE2) &&
406 FLAG_CHECK(name[2], FLAG_POSSIBLE3) &&
407 FLAG_CHECK(name[3], FLAG_POSSIBLE4)) {
408 /* a likely match, scan the lot */
410 for (i=0; reserved_names[i]; i++) {
411 int len = strlen(reserved_names[i]);
412 /* note that we match on COM1 as well as COM1.foo */
413 if (strncasecmp(name, reserved_names[i], len) == 0 &&
414 (name[len] == '.' || name[len] == 0)) {
424 see if a filename is a legal long filename
426 static BOOL is_legal_name(const char *name)
429 if (FLAG_CHECK(name[0], FLAG_ILLEGAL)) {
439 the main forward mapping function, which converts a long filename to
442 if need83 is not set then we only do the mangling if the name is illegal
445 if cache83 is not set then we don't cache the result
447 the name parameter must be able to hold 13 bytes
449 static BOOL name_map(char *name, BOOL need83, BOOL cache83)
454 int extension_length, i;
459 /* reserved names are handled specially */
460 if (!is_reserved_name(name)) {
461 /* if the name is already a valid 8.3 name then we don't need to
463 if (is_8_3(name, False)) {
467 /* if the caller doesn't strictly need 8.3 then just check for illegal
469 if (!need83 && is_legal_name(name)) {
474 /* find the '.' if any */
475 dot_p = strrchr(name, '.');
478 /* if the extension contains any illegal characters or
479 is too long or zero length then we treat it as part
481 for (i=0; i<4 && dot_p[i+1]; i++) {
482 if (! FLAG_CHECK(dot_p[i+1], FLAG_ASCII)) {
487 if (i == 0 || i == 4) dot_p = NULL;
490 /* the leading character in the mangled name is taken from
491 the first character of the name, if it is ascii
492 otherwise '_' is used
495 if (! FLAG_CHECK(lead_char, FLAG_ASCII)) {
498 lead_char = toupper(lead_char);
500 /* the prefix is anything up to the first dot */
502 prefix_len = PTR_DIFF(dot_p, name);
504 prefix_len = strlen(name);
507 /* the extension of the mangled name is taken from the first 3
508 ascii chars after the dot */
509 extension_length = 0;
511 for (i=1; extension_length < 3 && dot_p[i]; i++) {
513 if (FLAG_CHECK(c, FLAG_ASCII)) {
514 extension[extension_length++] = toupper(c);
519 /* find the hash for this prefix */
520 v = hash = mangle_hash(name, prefix_len);
522 /* now form the mangled name. */
523 new_name[0] = lead_char;
524 new_name[7] = base_forward(v % 36);
526 for (i=5; i>=1; i--) {
528 new_name[i] = base_forward(v % 36);
531 /* add the extension */
532 if (extension_length) {
534 memcpy(&new_name[9], extension, extension_length);
535 new_name[9+extension_length] = 0;
541 /* put it in the cache */
542 cache_insert(name, prefix_len, hash);
545 M_DEBUG(0,("name_map: %s -> %08X -> %s (cache=%d)\n",
546 name, hash, new_name, cache83));
548 /* and overwrite the old name */
549 fstrcpy(name, new_name);
551 /* all done, we've managed to mangle it */
556 /* initialise the flags table
558 we allow only a very restricted set of characters as 'ascii' in this
559 mangling backend. This isn't a significant problem as modern clients
560 use the 'long' filenames anyway, and those don't have these
563 static void init_tables(void)
567 memset(char_flags, 0, sizeof(char_flags));
569 for (i=0;i<128;i++) {
570 if ((i >= '0' && i <= '9') ||
571 (i >= 'a' && i <= 'z') ||
572 (i >= 'A' && i <= 'Z')) {
573 char_flags[i] |= (FLAG_ASCII | FLAG_BASECHAR);
575 if (strchr("_-$~", i)) {
576 char_flags[i] |= FLAG_ASCII;
579 if (strchr("*\\/?<>|\":", i)) {
580 char_flags[i] |= FLAG_ILLEGAL;
583 if (strchr("*?\"<>", i)) {
584 char_flags[i] |= FLAG_WILDCARD;
588 memset(base_reverse, 0, sizeof(base_reverse));
590 base_reverse[(unsigned char)base_forward(i)] = i;
593 /* fill in the reserved names flags. These are used as a very
594 fast filter for finding possible DOS reserved filenames */
595 for (i=0; reserved_names[i]; i++) {
596 unsigned char c1, c2, c3, c4;
598 c1 = (unsigned char)reserved_names[i][0];
599 c2 = (unsigned char)reserved_names[i][1];
600 c3 = (unsigned char)reserved_names[i][2];
601 c4 = (unsigned char)reserved_names[i][3];
603 char_flags[c1] |= FLAG_POSSIBLE1;
604 char_flags[c2] |= FLAG_POSSIBLE2;
605 char_flags[c3] |= FLAG_POSSIBLE3;
606 char_flags[c4] |= FLAG_POSSIBLE4;
607 char_flags[tolower(c1)] |= FLAG_POSSIBLE1;
608 char_flags[tolower(c2)] |= FLAG_POSSIBLE2;
609 char_flags[tolower(c3)] |= FLAG_POSSIBLE3;
610 char_flags[tolower(c4)] |= FLAG_POSSIBLE4;
612 char_flags[(unsigned char)'.'] |= FLAG_POSSIBLE4;
618 the following provides the abstraction layer to make it easier
619 to drop in an alternative mangling implementation */
620 static struct mangle_fns mangle_fns = {
628 /* return the methods for this mangling implementation */
629 struct mangle_fns *mangle_hash2_init(void)