2 Unix SMB/CIFS implementation.
3 new hash based name mangling implementation
4 Copyright (C) Andrew Tridgell 2002
5 Copyright (C) Simo Sorce 2002
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2 of the License, or
10 (at your option) any later version.
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program; if not, write to the Free Software
19 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 this mangling scheme uses the following format
27 where nnnnn is a base 36 hash, and A represents characters from the original string
29 The hash is taken of the leading part of the long filename, in uppercase
31 for simplicity, we only allow ascii characters in 8.3 names
34 /* hash alghorithm changed to FNV1 by idra@samba.org (Simo Sorce).
35 * see http://www.isthe.com/chongo/tech/comp/fnv/index.html for a
36 * discussion on Fowler / Noll / Vo (FNV) Hash by one of it's authors
40 ===============================================================================
43 This file deliberately uses non-multibyte string functions in many places. This
44 is *not* a mistake. This code is multi-byte safe, but it gets this property
45 through some very subtle knowledge of the way multi-byte strings are encoded
46 and the fact that this mangling algorithm only supports ascii characters in
49 please don't convert this file to use the *_m() functions!!
50 ===============================================================================
57 #define M_DEBUG(level, x) DEBUG(level, x)
59 #define M_DEBUG(level, x)
62 /* these flags are used to mark characters in as having particular
64 #define FLAG_BASECHAR 1
66 #define FLAG_ILLEGAL 4
67 #define FLAG_WILDCARD 8
69 /* the "possible" flags are used as a fast way to find possible DOS
71 #define FLAG_POSSIBLE1 16
72 #define FLAG_POSSIBLE2 32
73 #define FLAG_POSSIBLE3 64
74 #define FLAG_POSSIBLE4 128
76 /* by default have a max of 4096 entries in the cache. */
77 #ifndef MANGLE_CACHE_SIZE
78 #define MANGLE_CACHE_SIZE 4096
81 #define FNV1_PRIME 0x01000193
82 /*the following number is a fnv1 of the string: idra@samba.org 2002 */
83 #define FNV1_INIT 0xa6b93095
85 /* these tables are used to provide fast tests for characters */
86 static unsigned char char_flags[256];
88 #define FLAG_CHECK(c, flag) (char_flags[(unsigned char)(c)] & (flag))
91 this determines how many characters are used from the original filename
92 in the 8.3 mangled name. A larger value leads to a weaker hash and more collisions.
93 The largest possible value is 6.
95 static unsigned mangle_prefix;
97 /* we will use a very simple direct mapped prefix cache. The big
98 advantage of this cache structure is speed and low memory usage
100 The cache is indexed by the low-order bits of the hash, and confirmed by
101 hashing the resulting cache entry to match the known hash
103 static char **prefix_cache;
104 static u32 *prefix_cache_hashes;
106 /* these are the characters we use in the 8.3 hash. Must be 36 chars long */
107 static const char *basechars = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ";
108 static unsigned char base_reverse[256];
109 #define base_forward(v) basechars[v]
111 /* the list of reserved dos names - all of these are illegal */
112 static const char *reserved_names[] =
113 { "AUX", "LOCK$", "CON", "COM1", "COM2", "COM3", "COM4",
114 "LPT1", "LPT2", "LPT3", "NUL", "PRN", NULL };
117 hash a string of the specified length. The string does not need to be
120 this hash needs to be fast with a low collision rate (what hash doesn't?)
122 static u32 mangle_hash(const char *key, unsigned int length)
128 /* we have to uppercase here to ensure that the mangled name
129 doesn't depend on the case of the long name. Note that this
130 is the only place where we need to use a multi-byte string
132 length = MIN(length,sizeof(fstring)-1);
133 strncpy(str, key, length);
137 /* the length of a multi-byte string can change after a strupper_m */
138 length = strlen(str);
140 /* Set the initial value from the key size. */
141 for (value = FNV1_INIT, i=0; i < length; i++) {
142 value *= (u32)FNV1_PRIME;
143 value ^= (u32)(str[i]);
146 /* note that we force it to a 31 bit hash, to keep within the limits
147 of the 36^6 mangle space */
148 return value & ~0x80000000;
152 initialise (ie. allocate) the prefix cache
154 static BOOL cache_init(void)
160 prefix_cache = SMB_CALLOC_ARRAY(char *,MANGLE_CACHE_SIZE);
165 prefix_cache_hashes = SMB_CALLOC_ARRAY(u32, MANGLE_CACHE_SIZE);
166 if (!prefix_cache_hashes) {
174 insert an entry into the prefix cache. The string might not be null
176 static void cache_insert(const char *prefix, int length, u32 hash)
178 int i = hash % MANGLE_CACHE_SIZE;
180 if (prefix_cache[i]) {
181 free(prefix_cache[i]);
184 prefix_cache[i] = SMB_STRNDUP(prefix, length);
185 prefix_cache_hashes[i] = hash;
189 lookup an entry in the prefix cache. Return NULL if not found.
191 static const char *cache_lookup(u32 hash)
193 int i = hash % MANGLE_CACHE_SIZE;
195 if (!prefix_cache[i] || hash != prefix_cache_hashes[i]) {
199 /* yep, it matched */
200 return prefix_cache[i];
205 determine if a string is possibly in a mangled format, ignoring
208 In this algorithm, mangled names use only pure ascii characters (no
209 multi-byte) so we can avoid doing a UCS2 conversion
211 static BOOL is_mangled_component(const char *name, size_t len)
215 M_DEBUG(10,("is_mangled_component %s (len %u) ?\n", name, (unsigned int)len));
217 /* check the length */
218 if (len > 12 || len < 8)
221 /* the best distinguishing characteristic is the ~ */
225 /* check extension */
229 for (i=9; name[i] && i < len; i++) {
230 if (! FLAG_CHECK(name[i], FLAG_ASCII)) {
236 /* check lead characters */
237 for (i=0;i<mangle_prefix;i++) {
238 if (! FLAG_CHECK(name[i], FLAG_ASCII)) {
243 /* check rest of hash */
244 if (! FLAG_CHECK(name[7], FLAG_BASECHAR)) {
247 for (i=mangle_prefix;i<6;i++) {
248 if (! FLAG_CHECK(name[i], FLAG_BASECHAR)) {
253 M_DEBUG(10,("is_mangled_component %s (len %u) -> yes\n", name, (unsigned int)len));
261 determine if a string is possibly in a mangled format, ignoring
264 In this algorithm, mangled names use only pure ascii characters (no
265 multi-byte) so we can avoid doing a UCS2 conversion
267 NOTE! This interface must be able to handle a path with unix
268 directory separators. It should return true if any component is
271 static BOOL is_mangled(const char *name)
276 M_DEBUG(10,("is_mangled %s ?\n", name));
278 for (s=name; (p=strchr(s, '/')); s=p+1) {
279 if (is_mangled_component(s, PTR_DIFF(p, s))) {
284 /* and the last part ... */
285 return is_mangled_component(s,strlen(s));
290 see if a filename is an allowable 8.3 name.
292 we are only going to allow ascii characters in 8.3 names, as this
293 simplifies things greatly (it means that we know the string won't
294 get larger when converted from UNIX to DOS formats)
296 static BOOL is_8_3(const char *name, BOOL check_case, BOOL allow_wildcards)
301 /* as a special case, the names '.' and '..' are allowable 8.3 names */
302 if (name[0] == '.') {
303 if (!name[1] || (name[1] == '.' && !name[2])) {
308 /* the simplest test is on the overall length of the
309 filename. Note that we deliberately use the ascii string
310 length (not the multi-byte one) as it is faster, and gives us
311 the result we need in this case. Using strlen_m would not
312 only be slower, it would be incorrect */
317 /* find the '.'. Note that once again we use the non-multibyte
319 dot_p = strchr(name, '.');
322 /* if the name doesn't contain a '.' then its length
323 must be less than 8 */
328 int prefix_len, suffix_len;
330 /* if it does contain a dot then the prefix must be <=
331 8 and the suffix <= 3 in length */
332 prefix_len = PTR_DIFF(dot_p, name);
333 suffix_len = len - (prefix_len+1);
335 if (prefix_len > 8 || suffix_len > 3 || suffix_len == 0) {
339 /* a 8.3 name cannot contain more than 1 '.' */
340 if (strchr(dot_p+1, '.')) {
345 /* the length are all OK. Now check to see if the characters themselves are OK */
346 for (i=0; name[i]; i++) {
347 /* note that we may allow wildcard petterns! */
348 if (!FLAG_CHECK(name[i], FLAG_ASCII|(allow_wildcards ? FLAG_WILDCARD : 0)) && name[i] != '.') {
353 /* it is a good 8.3 name */
359 reset the mangling cache on a smb.conf reload. This only really makes sense for
360 mangling backends that have parameters in smb.conf, and as this backend doesn't
361 this is a NULL operation
363 static void mangle_reset(void)
370 try to find a 8.3 name in the cache, and if found then
371 replace the string with the original long name.
373 static BOOL check_cache(char *name, size_t maxlen)
375 u32 hash, multiplier;
380 /* make sure that this is a mangled name from this cache */
381 if (!is_mangled(name)) {
382 M_DEBUG(10,("check_cache: %s -> not mangled\n", name));
386 /* we need to extract the hash from the 8.3 name */
387 hash = base_reverse[(unsigned char)name[7]];
388 for (multiplier=36, i=5;i>=mangle_prefix;i--) {
389 u32 v = base_reverse[(unsigned char)name[i]];
390 hash += multiplier * v;
394 /* now look in the prefix cache for that hash */
395 prefix = cache_lookup(hash);
397 M_DEBUG(10,("check_cache: %s -> %08X -> not found\n", name, hash));
401 /* we found it - construct the full name */
402 if (name[8] == '.') {
403 strncpy(extension, name+9, 3);
410 M_DEBUG(10,("check_cache: %s -> %s.%s\n", name, prefix, extension));
411 slprintf(name, maxlen, "%s.%s", prefix, extension);
413 M_DEBUG(10,("check_cache: %s -> %s\n", name, prefix));
414 safe_strcpy(name, prefix, maxlen);
422 look for a DOS reserved name
424 static BOOL is_reserved_name(const char *name)
426 if (FLAG_CHECK(name[0], FLAG_POSSIBLE1) &&
427 FLAG_CHECK(name[1], FLAG_POSSIBLE2) &&
428 FLAG_CHECK(name[2], FLAG_POSSIBLE3) &&
429 FLAG_CHECK(name[3], FLAG_POSSIBLE4)) {
430 /* a likely match, scan the lot */
432 for (i=0; reserved_names[i]; i++) {
433 int len = strlen(reserved_names[i]);
434 /* note that we match on COM1 as well as COM1.foo */
435 if (strnequal(name, reserved_names[i], len) &&
436 (name[len] == '.' || name[len] == 0)) {
446 See if a filename is a legal long filename.
447 A filename ending in a '.' is not legal unless it's "." or "..". JRA.
450 static BOOL is_legal_name(const char *name)
452 const char *dot_pos = NULL;
457 if (((unsigned int)name[0]) > 128 && (name[1] != 0)) {
458 /* Possible start of mb character. */
461 * Note that if CH_UNIX is utf8 a string may be 3
462 * bytes, but this is ok as mb utf8 characters don't
463 * contain embedded ascii bytes. We are really checking
464 * for mb UNIX asian characters like Japanese (SJIS) here.
467 if (convert_string(CH_UNIX, CH_UCS2, name, 2, mbc, 2, False) == 2) {
468 /* Was a good mb string. */
474 if (FLAG_CHECK(name[0], FLAG_ILLEGAL)) {
477 if (name[0] == '.') {
487 if (alldots && (numdots == 1 || numdots == 2))
488 return True; /* . or .. is a valid name */
490 /* A valid long name cannot end in '.' */
491 if (dot_pos[1] == '\0')
499 the main forward mapping function, which converts a long filename to
502 if need83 is not set then we only do the mangling if the name is illegal
505 if cache83 is not set then we don't cache the result
507 the name parameter must be able to hold 13 bytes
509 static void name_map(fstring name, BOOL need83, BOOL cache83, int default_case)
514 unsigned int extension_length, i;
515 unsigned int prefix_len;
519 /* reserved names are handled specially */
520 if (!is_reserved_name(name)) {
521 /* if the name is already a valid 8.3 name then we don't need to
523 if (is_8_3(name, False, False)) {
527 /* if the caller doesn't strictly need 8.3 then just check for illegal
529 if (!need83 && is_legal_name(name)) {
534 /* find the '.' if any */
535 dot_p = strrchr(name, '.');
538 /* if the extension contains any illegal characters or
539 is too long or zero length then we treat it as part
541 for (i=0; i<4 && dot_p[i+1]; i++) {
542 if (! FLAG_CHECK(dot_p[i+1], FLAG_ASCII)) {
547 if (i == 0 || i == 4) dot_p = NULL;
550 /* the leading characters in the mangled name is taken from
551 the first characters of the name, if they are ascii otherwise
554 for (i=0;i<mangle_prefix && name[i];i++) {
555 lead_chars[i] = name[i];
556 if (! FLAG_CHECK(lead_chars[i], FLAG_ASCII)) {
559 lead_chars[i] = toupper(lead_chars[i]);
561 for (;i<mangle_prefix;i++) {
565 /* the prefix is anything up to the first dot */
567 prefix_len = PTR_DIFF(dot_p, name);
569 prefix_len = strlen(name);
572 /* the extension of the mangled name is taken from the first 3
573 ascii chars after the dot */
574 extension_length = 0;
576 for (i=1; extension_length < 3 && dot_p[i]; i++) {
578 if (FLAG_CHECK(c, FLAG_ASCII)) {
579 extension[extension_length++] = toupper(c);
584 /* find the hash for this prefix */
585 v = hash = mangle_hash(name, prefix_len);
587 /* now form the mangled name. */
588 for (i=0;i<mangle_prefix;i++) {
589 new_name[i] = lead_chars[i];
591 new_name[7] = base_forward(v % 36);
593 for (i=5; i>=mangle_prefix; i--) {
595 new_name[i] = base_forward(v % 36);
598 /* add the extension */
599 if (extension_length) {
601 memcpy(&new_name[9], extension, extension_length);
602 new_name[9+extension_length] = 0;
608 /* put it in the cache */
609 cache_insert(name, prefix_len, hash);
612 M_DEBUG(10,("name_map: %s -> %08X -> %s (cache=%d)\n",
613 name, hash, new_name, cache83));
615 /* and overwrite the old name */
616 fstrcpy(name, new_name);
618 /* all done, we've managed to mangle it */
622 /* initialise the flags table
624 we allow only a very restricted set of characters as 'ascii' in this
625 mangling backend. This isn't a significant problem as modern clients
626 use the 'long' filenames anyway, and those don't have these
629 static void init_tables(void)
633 memset(char_flags, 0, sizeof(char_flags));
635 for (i=1;i<128;i++) {
636 if ((i >= '0' && i <= '9') ||
637 (i >= 'a' && i <= 'z') ||
638 (i >= 'A' && i <= 'Z')) {
639 char_flags[i] |= (FLAG_ASCII | FLAG_BASECHAR);
641 if (strchr("_-$~", i)) {
642 char_flags[i] |= FLAG_ASCII;
645 if (strchr("*\\/?<>|\":", i)) {
646 char_flags[i] |= FLAG_ILLEGAL;
649 if (strchr("*?\"<>", i)) {
650 char_flags[i] |= FLAG_WILDCARD;
654 memset(base_reverse, 0, sizeof(base_reverse));
656 base_reverse[(unsigned char)base_forward(i)] = i;
659 /* fill in the reserved names flags. These are used as a very
660 fast filter for finding possible DOS reserved filenames */
661 for (i=0; reserved_names[i]; i++) {
662 unsigned char c1, c2, c3, c4;
664 c1 = (unsigned char)reserved_names[i][0];
665 c2 = (unsigned char)reserved_names[i][1];
666 c3 = (unsigned char)reserved_names[i][2];
667 c4 = (unsigned char)reserved_names[i][3];
669 char_flags[c1] |= FLAG_POSSIBLE1;
670 char_flags[c2] |= FLAG_POSSIBLE2;
671 char_flags[c3] |= FLAG_POSSIBLE3;
672 char_flags[c4] |= FLAG_POSSIBLE4;
673 char_flags[tolower(c1)] |= FLAG_POSSIBLE1;
674 char_flags[tolower(c2)] |= FLAG_POSSIBLE2;
675 char_flags[tolower(c3)] |= FLAG_POSSIBLE3;
676 char_flags[tolower(c4)] |= FLAG_POSSIBLE4;
678 char_flags[(unsigned char)'.'] |= FLAG_POSSIBLE4;
684 the following provides the abstraction layer to make it easier
685 to drop in an alternative mangling implementation */
686 static struct mangle_fns mangle_fns = {
694 /* return the methods for this mangling implementation */
695 struct mangle_fns *mangle_hash2_init(void)
697 /* the mangle prefix can only be in the mange 1 to 6 */
698 mangle_prefix = lp_mangle_prefix();
699 if (mangle_prefix > 6) {
702 if (mangle_prefix < 1) {