xref: /openbmc/u-boot/lib/hashtable.c (revision 9b107e6138e719ea5a0b924862a9b109c020c7ac)
1 /*
2  * This implementation is based on code from uClibc-0.9.30.3 but was
3  * modified and extended for use within U-Boot.
4  *
5  * Copyright (C) 2010 Wolfgang Denk <wd@denx.de>
6  *
7  * Original license header:
8  *
9  * Copyright (C) 1993, 1995, 1996, 1997, 2002 Free Software Foundation, Inc.
10  * This file is part of the GNU C Library.
11  * Contributed by Ulrich Drepper <drepper@gnu.ai.mit.edu>, 1993.
12  *
13  * The GNU C Library is free software; you can redistribute it and/or
14  * modify it under the terms of the GNU Lesser General Public
15  * License as published by the Free Software Foundation; either
16  * version 2.1 of the License, or (at your option) any later version.
17  *
18  * The GNU C Library is distributed in the hope that it will be useful,
19  * but WITHOUT ANY WARRANTY; without even the implied warranty of
20  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
21  * Lesser General Public License for more details.
22  *
23  * You should have received a copy of the GNU Lesser General Public
24  * License along with the GNU C Library; if not, write to the Free
25  * Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
26  * 02111-1307 USA.
27  */
28 
29 #include <errno.h>
30 #include <malloc.h>
31 
32 #ifdef USE_HOSTCC		/* HOST build */
33 # include <string.h>
34 # include <assert.h>
35 
36 # ifndef debug
37 #  ifdef DEBUG
38 #   define debug(fmt,args...)	printf(fmt ,##args)
39 #  else
40 #   define debug(fmt,args...)
41 #  endif
42 # endif
43 #else				/* U-Boot build */
44 # include <common.h>
45 # include <linux/string.h>
46 #endif
47 
48 #ifndef	CONFIG_ENV_MIN_ENTRIES	/* minimum number of entries */
49 #define	CONFIG_ENV_MIN_ENTRIES 64
50 #endif
51 #ifndef	CONFIG_ENV_MAX_ENTRIES	/* maximum number of entries */
52 #define	CONFIG_ENV_MAX_ENTRIES 512
53 #endif
54 
55 #include "search.h"
56 
57 /*
58  * [Aho,Sethi,Ullman] Compilers: Principles, Techniques and Tools, 1986
59  * [Knuth]	      The Art of Computer Programming, part 3 (6.4)
60  */
61 
62 /*
63  * The non-reentrant version use a global space for storing the hash table.
64  */
65 static struct hsearch_data htab;
66 
67 /*
68  * The reentrant version has no static variables to maintain the state.
69  * Instead the interface of all functions is extended to take an argument
70  * which describes the current status.
71  */
72 typedef struct _ENTRY {
73 	unsigned int used;
74 	ENTRY entry;
75 } _ENTRY;
76 
77 
78 /*
79  * hcreate()
80  */
81 
82 /*
83  * For the used double hash method the table size has to be a prime. To
84  * correct the user given table size we need a prime test.  This trivial
85  * algorithm is adequate because
86  * a)  the code is (most probably) called a few times per program run and
87  * b)  the number is small because the table must fit in the core
88  * */
89 static int isprime(unsigned int number)
90 {
91 	/* no even number will be passed */
92 	unsigned int div = 3;
93 
94 	while (div * div < number && number % div != 0)
95 		div += 2;
96 
97 	return number % div != 0;
98 }
99 
100 int hcreate(size_t nel)
101 {
102 	return hcreate_r(nel, &htab);
103 }
104 
105 /*
106  * Before using the hash table we must allocate memory for it.
107  * Test for an existing table are done. We allocate one element
108  * more as the found prime number says. This is done for more effective
109  * indexing as explained in the comment for the hsearch function.
110  * The contents of the table is zeroed, especially the field used
111  * becomes zero.
112  */
113 int hcreate_r(size_t nel, struct hsearch_data *htab)
114 {
115 	/* Test for correct arguments.  */
116 	if (htab == NULL) {
117 		__set_errno(EINVAL);
118 		return 0;
119 	}
120 
121 	/* There is still another table active. Return with error. */
122 	if (htab->table != NULL)
123 		return 0;
124 
125 	/* Change nel to the first prime number not smaller as nel. */
126 	nel |= 1;		/* make odd */
127 	while (!isprime(nel))
128 		nel += 2;
129 
130 	htab->size = nel;
131 	htab->filled = 0;
132 
133 	/* allocate memory and zero out */
134 	htab->table = (_ENTRY *) calloc(htab->size + 1, sizeof(_ENTRY));
135 	if (htab->table == NULL)
136 		return 0;
137 
138 	/* everything went alright */
139 	return 1;
140 }
141 
142 
143 /*
144  * hdestroy()
145  */
146 void hdestroy(void)
147 {
148 	hdestroy_r(&htab);
149 }
150 
151 /*
152  * After using the hash table it has to be destroyed. The used memory can
153  * be freed and the local static variable can be marked as not used.
154  */
155 void hdestroy_r(struct hsearch_data *htab)
156 {
157 	int i;
158 
159 	/* Test for correct arguments.  */
160 	if (htab == NULL) {
161 		__set_errno(EINVAL);
162 		return;
163 	}
164 
165 	/* free used memory */
166 	for (i = 1; i <= htab->size; ++i) {
167 		if (htab->table[i].used) {
168 			ENTRY *ep = &htab->table[i].entry;
169 
170 			free(ep->key);
171 			free(ep->data);
172 		}
173 	}
174 	free(htab->table);
175 
176 	/* the sign for an existing table is an value != NULL in htable */
177 	htab->table = NULL;
178 }
179 
180 /*
181  * hsearch()
182  */
183 
184 /*
185  * This is the search function. It uses double hashing with open addressing.
186  * The argument item.key has to be a pointer to an zero terminated, most
187  * probably strings of chars. The function for generating a number of the
188  * strings is simple but fast. It can be replaced by a more complex function
189  * like ajw (see [Aho,Sethi,Ullman]) if the needs are shown.
190  *
191  * We use an trick to speed up the lookup. The table is created by hcreate
192  * with one more element available. This enables us to use the index zero
193  * special. This index will never be used because we store the first hash
194  * index in the field used where zero means not used. Every other value
195  * means used. The used field can be used as a first fast comparison for
196  * equality of the stored and the parameter value. This helps to prevent
197  * unnecessary expensive calls of strcmp.
198  *
199  * This implementation differs from the standard library version of
200  * this function in a number of ways:
201  *
202  * - While the standard version does not make any assumptions about
203  *   the type of the stored data objects at all, this implementation
204  *   works with NUL terminated strings only.
205  * - Instead of storing just pointers to the original objects, we
206  *   create local copies so the caller does not need to care about the
207  *   data any more.
208  * - The standard implementation does not provide a way to update an
209  *   existing entry.  This version will create a new entry or update an
210  *   existing one when both "action == ENTER" and "item.data != NULL".
211  * - Instead of returning 1 on success, we return the index into the
212  *   internal hash table, which is also guaranteed to be positive.
213  *   This allows us direct access to the found hash table slot for
214  *   example for functions like hdelete().
215  */
216 
217 ENTRY *hsearch(ENTRY item, ACTION action)
218 {
219 	ENTRY *result;
220 
221 	(void) hsearch_r(item, action, &result, &htab);
222 
223 	return result;
224 }
225 
226 int hsearch_r(ENTRY item, ACTION action, ENTRY ** retval,
227 	      struct hsearch_data *htab)
228 {
229 	unsigned int hval;
230 	unsigned int count;
231 	unsigned int len = strlen(item.key);
232 	unsigned int idx;
233 
234 	/* Compute an value for the given string. Perhaps use a better method. */
235 	hval = len;
236 	count = len;
237 	while (count-- > 0) {
238 		hval <<= 4;
239 		hval += item.key[count];
240 	}
241 
242 	/*
243 	 * First hash function:
244 	 * simply take the modul but prevent zero.
245 	 */
246 	hval %= htab->size;
247 	if (hval == 0)
248 		++hval;
249 
250 	/* The first index tried. */
251 	idx = hval;
252 
253 	if (htab->table[idx].used) {
254 		/*
255 		 * Further action might be required according to the
256 		 * action value.
257 		 */
258 		unsigned hval2;
259 
260 		if (htab->table[idx].used == hval
261 		    && strcmp(item.key, htab->table[idx].entry.key) == 0) {
262 			/* Overwrite existing value? */
263 			if ((action == ENTER) && (item.data != NULL)) {
264 				free(htab->table[idx].entry.data);
265 				htab->table[idx].entry.data =
266 					strdup(item.data);
267 				if (!htab->table[idx].entry.data) {
268 					__set_errno(ENOMEM);
269 					*retval = NULL;
270 					return 0;
271 				}
272 			}
273 			/* return found entry */
274 			*retval = &htab->table[idx].entry;
275 			return idx;
276 		}
277 
278 		/*
279 		 * Second hash function:
280 		 * as suggested in [Knuth]
281 		 */
282 		hval2 = 1 + hval % (htab->size - 2);
283 
284 		do {
285 			/*
286 			 * Because SIZE is prime this guarantees to
287 			 * step through all available indices.
288 			 */
289 			if (idx <= hval2)
290 				idx = htab->size + idx - hval2;
291 			else
292 				idx -= hval2;
293 
294 			/*
295 			 * If we visited all entries leave the loop
296 			 * unsuccessfully.
297 			 */
298 			if (idx == hval)
299 				break;
300 
301 			/* If entry is found use it. */
302 			if ((htab->table[idx].used == hval)
303 			    && strcmp(item.key, htab->table[idx].entry.key) == 0) {
304 				/* Overwrite existing value? */
305 				if ((action == ENTER) && (item.data != NULL)) {
306 					free(htab->table[idx].entry.data);
307 					htab->table[idx].entry.data =
308 						strdup(item.data);
309 					if (!htab->table[idx].entry.data) {
310 						__set_errno(ENOMEM);
311 						*retval = NULL;
312 						return 0;
313 					}
314 				}
315 				/* return found entry */
316 				*retval = &htab->table[idx].entry;
317 				return idx;
318 			}
319 		}
320 		while (htab->table[idx].used);
321 	}
322 
323 	/* An empty bucket has been found. */
324 	if (action == ENTER) {
325 		/*
326 		 * If table is full and another entry should be
327 		 * entered return with error.
328 		 */
329 		if (htab->filled == htab->size) {
330 			__set_errno(ENOMEM);
331 			*retval = NULL;
332 			return 0;
333 		}
334 
335 		/*
336 		 * Create new entry;
337 		 * create copies of item.key and item.data
338 		 */
339 		htab->table[idx].used = hval;
340 		htab->table[idx].entry.key = strdup(item.key);
341 		htab->table[idx].entry.data = strdup(item.data);
342 		if (!htab->table[idx].entry.key ||
343 		    !htab->table[idx].entry.data) {
344 			__set_errno(ENOMEM);
345 			*retval = NULL;
346 			return 0;
347 		}
348 
349 		++htab->filled;
350 
351 		/* return new entry */
352 		*retval = &htab->table[idx].entry;
353 		return 1;
354 	}
355 
356 	__set_errno(ESRCH);
357 	*retval = NULL;
358 	return 0;
359 }
360 
361 
362 /*
363  * hdelete()
364  */
365 
366 /*
367  * The standard implementation of hsearch(3) does not provide any way
368  * to delete any entries from the hash table.  We extend the code to
369  * do that.
370  */
371 
372 int hdelete(const char *key)
373 {
374 	return hdelete_r(key, &htab);
375 }
376 
377 int hdelete_r(const char *key, struct hsearch_data *htab)
378 {
379 	ENTRY e, *ep;
380 	int idx;
381 
382 	debug("hdelete: DELETE key \"%s\"\n", key);
383 
384 	e.key = (char *)key;
385 
386 	if ((idx = hsearch_r(e, FIND, &ep, htab)) == 0) {
387 		__set_errno(ESRCH);
388 		return 0;	/* not found */
389 	}
390 
391 	/* free used ENTRY */
392 	debug("hdelete: DELETING key \"%s\"\n", key);
393 
394 	free(ep->key);
395 	free(ep->data);
396 	htab->table[idx].used = 0;
397 
398 	--htab->filled;
399 
400 	return 1;
401 }
402 
403 /*
404  * hexport()
405  */
406 
407 /*
408  * Export the data stored in the hash table in linearized form.
409  *
410  * Entries are exported as "name=value" strings, separated by an
411  * arbitrary (non-NUL, of course) separator character. This allows to
412  * use this function both when formatting the U-Boot environment for
413  * external storage (using '\0' as separator), but also when using it
414  * for the "printenv" command to print all variables, simply by using
415  * as '\n" as separator. This can also be used for new features like
416  * exporting the environment data as text file, including the option
417  * for later re-import.
418  *
419  * The entries in the result list will be sorted by ascending key
420  * values.
421  *
422  * If the separator character is different from NUL, then any
423  * separator characters and backslash characters in the values will
424  * be escaped by a preceeding backslash in output. This is needed for
425  * example to enable multi-line values, especially when the output
426  * shall later be parsed (for example, for re-import).
427  *
428  * There are several options how the result buffer is handled:
429  *
430  * *resp  size
431  * -----------
432  *  NULL    0	A string of sufficient length will be allocated.
433  *  NULL   >0	A string of the size given will be
434  *		allocated. An error will be returned if the size is
435  *		not sufficient.  Any unused bytes in the string will
436  *		be '\0'-padded.
437  * !NULL    0	The user-supplied buffer will be used. No length
438  *		checking will be performed, i. e. it is assumed that
439  *		the buffer size will always be big enough. DANGEROUS.
440  * !NULL   >0	The user-supplied buffer will be used. An error will
441  *		be returned if the size is not sufficient.  Any unused
442  *		bytes in the string will be '\0'-padded.
443  */
444 
445 ssize_t hexport(const char sep, char **resp, size_t size)
446 {
447 	return hexport_r(&htab, sep, resp, size);
448 }
449 
450 static int cmpkey(const void *p1, const void *p2)
451 {
452 	ENTRY *e1 = *(ENTRY **) p1;
453 	ENTRY *e2 = *(ENTRY **) p2;
454 
455 	return (strcmp(e1->key, e2->key));
456 }
457 
458 ssize_t hexport_r(struct hsearch_data *htab, const char sep,
459 		 char **resp, size_t size)
460 {
461 	ENTRY *list[htab->size];
462 	char *res, *p;
463 	size_t totlen;
464 	int i, n;
465 
466 	/* Test for correct arguments.  */
467 	if ((resp == NULL) || (htab == NULL)) {
468 		__set_errno(EINVAL);
469 		return (-1);
470 	}
471 
472 	debug("EXPORT  table = %p, htab.size = %d, htab.filled = %d, size = %d\n",
473 		htab, htab->size, htab->filled, size);
474 	/*
475 	 * Pass 1:
476 	 * search used entries,
477 	 * save addresses and compute total length
478 	 */
479 	for (i = 1, n = 0, totlen = 0; i <= htab->size; ++i) {
480 
481 		if (htab->table[i].used) {
482 			ENTRY *ep = &htab->table[i].entry;
483 
484 			list[n++] = ep;
485 
486 			totlen += strlen(ep->key) + 2;
487 
488 			if (sep == '\0') {
489 				totlen += strlen(ep->data);
490 			} else {	/* check if escapes are needed */
491 				char *s = ep->data;
492 
493 				while (*s) {
494 					++totlen;
495 					/* add room for needed escape chars */
496 					if ((*s == sep) || (*s == '\\'))
497 						++totlen;
498 					++s;
499 				}
500 			}
501 			totlen += 2;	/* for '=' and 'sep' char */
502 		}
503 	}
504 
505 #ifdef DEBUG
506 	/* Pass 1a: print unsorted list */
507 	printf("Unsorted: n=%d\n", n);
508 	for (i = 0; i < n; ++i) {
509 		printf("\t%3d: %p ==> %-10s => %s\n",
510 		       i, list[i], list[i]->key, list[i]->data);
511 	}
512 #endif
513 
514 	/* Sort list by keys */
515 	qsort(list, n, sizeof(ENTRY *), cmpkey);
516 
517 	/* Check if the user supplied buffer size is sufficient */
518 	if (size) {
519 		if (size < totlen + 1) {	/* provided buffer too small */
520 			debug("### buffer too small: %d, but need %d\n",
521 				size, totlen + 1);
522 			__set_errno(ENOMEM);
523 			return (-1);
524 		}
525 	} else {
526 		size = totlen + 1;
527 	}
528 
529 	/* Check if the user provided a buffer */
530 	if (*resp) {
531 		/* yes; clear it */
532 		res = *resp;
533 		memset(res, '\0', size);
534 	} else {
535 		/* no, allocate and clear one */
536 		*resp = res = calloc(1, size);
537 		if (res == NULL) {
538 			__set_errno(ENOMEM);
539 			return (-1);
540 		}
541 	}
542 	/*
543 	 * Pass 2:
544 	 * export sorted list of result data
545 	 */
546 	for (i = 0, p = res; i < n; ++i) {
547 		char *s;
548 
549 		s = list[i]->key;
550 		while (*s)
551 			*p++ = *s++;
552 		*p++ = '=';
553 
554 		s = list[i]->data;
555 
556 		while (*s) {
557 			if ((*s == sep) || (*s == '\\'))
558 				*p++ = '\\';	/* escape */
559 			*p++ = *s++;
560 		}
561 		*p++ = sep;
562 	}
563 	*p = '\0';		/* terminate result */
564 
565 	return size;
566 }
567 
568 
569 /*
570  * himport()
571  */
572 
573 /*
574  * Import linearized data into hash table.
575  *
576  * This is the inverse function to hexport(): it takes a linear list
577  * of "name=value" pairs and creates hash table entries from it.
578  *
579  * Entries without "value", i. e. consisting of only "name" or
580  * "name=", will cause this entry to be deleted from the hash table.
581  *
582  * The "flag" argument can be used to control the behaviour: when the
583  * H_NOCLEAR bit is set, then an existing hash table will kept, i. e.
584  * new data will be added to an existing hash table; otherwise, old
585  * data will be discarded and a new hash table will be created.
586  *
587  * The separator character for the "name=value" pairs can be selected,
588  * so we both support importing from externally stored environment
589  * data (separated by NUL characters) and from plain text files
590  * (entries separated by newline characters).
591  *
592  * To allow for nicely formatted text input, leading white space
593  * (sequences of SPACE and TAB chars) is ignored, and entries starting
594  * (after removal of any leading white space) with a '#' character are
595  * considered comments and ignored.
596  *
597  * [NOTE: this means that a variable name cannot start with a '#'
598  * character.]
599  *
600  * When using a non-NUL separator character, backslash is used as
601  * escape character in the value part, allowing for example for
602  * multi-line values.
603  *
604  * In theory, arbitrary separator characters can be used, but only
605  * '\0' and '\n' have really been tested.
606  */
607 
608 int himport(const char *env, size_t size, const char sep, int flag)
609 {
610 	return himport_r(&htab, env, size, sep, flag);
611 }
612 
613 int himport_r(struct hsearch_data *htab,
614 	      const char *env, size_t size, const char sep, int flag)
615 {
616 	char *data, *sp, *dp, *name, *value;
617 
618 	/* Test for correct arguments.  */
619 	if (htab == NULL) {
620 		__set_errno(EINVAL);
621 		return 0;
622 	}
623 
624 	/* we allocate new space to make sure we can write to the array */
625 	if ((data = malloc(size)) == NULL) {
626 		debug("himport_r: can't malloc %d bytes\n", size);
627 		__set_errno(ENOMEM);
628 		return 0;
629 	}
630 	memcpy(data, env, size);
631 	dp = data;
632 
633 	if ((flag & H_NOCLEAR) == 0) {
634 		/* Destroy old hash table if one exists */
635 		debug("Destroy Hash Table: %p table = %p\n", htab,
636 		       htab->table);
637 		if (htab->table)
638 			hdestroy_r(htab);
639 	}
640 
641 	/*
642 	 * Create new hash table (if needed).  The computation of the hash
643 	 * table size is based on heuristics: in a sample of some 70+
644 	 * existing systems we found an average size of 39+ bytes per entry
645 	 * in the environment (for the whole key=value pair). Assuming a
646 	 * size of 8 per entry (= safety factor of ~5) should provide enough
647 	 * safety margin for any existing environment definitions and still
648 	 * allow for more than enough dynamic additions. Note that the
649 	 * "size" argument is supposed to give the maximum enviroment size
650 	 * (CONFIG_ENV_SIZE).  This heuristics will result in
651 	 * unreasonably large numbers (and thus memory footprint) for
652 	 * big flash environments (>8,000 entries for 64 KB
653 	 * envrionment size), so we clip it to a reasonable value.
654 	 * On the other hand we need to add some more entries for free
655 	 * space when importing very small buffers. Both boundaries can
656 	 * be overwritten in the board config file if needed.
657 	 */
658 
659 	if (!htab->table) {
660 		int nent = CONFIG_ENV_MIN_ENTRIES + size / 8;
661 
662 		if (nent > CONFIG_ENV_MAX_ENTRIES)
663 			nent = CONFIG_ENV_MAX_ENTRIES;
664 
665 		debug("Create Hash Table: N=%d\n", nent);
666 
667 		if (hcreate_r(nent, htab) == 0) {
668 			free(data);
669 			return 0;
670 		}
671 	}
672 
673 	/* Parse environment; allow for '\0' and 'sep' as separators */
674 	do {
675 		ENTRY e, *rv;
676 
677 		/* skip leading white space */
678 		while ((*dp == ' ') || (*dp == '\t'))
679 			++dp;
680 
681 		/* skip comment lines */
682 		if (*dp == '#') {
683 			while (*dp && (*dp != sep))
684 				++dp;
685 			++dp;
686 			continue;
687 		}
688 
689 		/* parse name */
690 		for (name = dp; *dp != '=' && *dp && *dp != sep; ++dp)
691 			;
692 
693 		/* deal with "name" and "name=" entries (delete var) */
694 		if (*dp == '\0' || *(dp + 1) == '\0' ||
695 		    *dp == sep || *(dp + 1) == sep) {
696 			if (*dp == '=')
697 				*dp++ = '\0';
698 			*dp++ = '\0';	/* terminate name */
699 
700 			debug("DELETE CANDIDATE: \"%s\"\n", name);
701 
702 			if (hdelete_r(name, htab) == 0)
703 				debug("DELETE ERROR ##############################\n");
704 
705 			continue;
706 		}
707 		*dp++ = '\0';	/* terminate name */
708 
709 		/* parse value; deal with escapes */
710 		for (value = sp = dp; *dp && (*dp != sep); ++dp) {
711 			if ((*dp == '\\') && *(dp + 1))
712 				++dp;
713 			*sp++ = *dp;
714 		}
715 		*sp++ = '\0';	/* terminate value */
716 		++dp;
717 
718 		/* enter into hash table */
719 		e.key = name;
720 		e.data = value;
721 
722 		hsearch_r(e, ENTER, &rv, htab);
723 		if (rv == NULL) {
724 			printf("himport_r: can't insert \"%s=%s\" into hash table\n",
725 				name, value);
726 			return 0;
727 		}
728 
729 		debug("INSERT: table %p, filled %d/%d rv %p ==> name=\"%s\" value=\"%s\"\n",
730 			htab, htab->filled, htab->size,
731 			rv, name, value);
732 	} while ((dp < data + size) && *dp);	/* size check needed for text */
733 						/* without '\0' termination */
734 	debug("INSERT: free(data = %p)\n", data);
735 	free(data);
736 
737 	debug("INSERT: done\n");
738 	return 1;		/* everything OK */
739 }
740