xref: /dragonfly/libexec/rtld-elf/malloc.c (revision 650094e1)
1 /*-
2  * Copyright (c) 1983 Regents of the University of California.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33 
34 /*
35  * malloc.c (Caltech) 2/21/82
36  * Chris Kingsley, kingsley@cit-20.
37  *
38  * This is a very fast storage allocator.  It allocates blocks of a small
39  * number of different sizes, and keeps free lists of each size.  Blocks that
40  * don't exactly fit are passed up to the next larger size.  In this
41  * implementation, the available sizes are 2^n-4 (or 2^n-10) bytes long.
42  * This is designed for use in a virtual memory environment.
43  */
44 
45 #include <sys/types.h>
46 #include <paths.h>
47 #include <stdarg.h>
48 #include <stdio.h>
49 #include <stdlib.h>
50 #include <string.h>
51 #include <unistd.h>
52 #include <sys/param.h>
53 #include <sys/mman.h>
54 #include "rtld_printf.h"
55 #ifndef BSD
56 #define MAP_COPY	MAP_PRIVATE
57 #define MAP_FILE	0
58 #define MAP_ANON	0
59 #endif
60 
61 #ifndef BSD		/* Need do better than this */
62 #define NEED_DEV_ZERO	1
63 #endif
64 
65 static void morecore();
66 static int findbucket();
67 
68 /*
69  * Pre-allocate mmap'ed pages
70  */
71 #define	NPOOLPAGES	(32*1024/pagesz)
72 static caddr_t		pagepool_start, pagepool_end;
73 static int		morepages();
74 
75 /*
76  * The overhead on a block is at least 4 bytes.  When free, this space
77  * contains a pointer to the next free block, and the bottom two bits must
78  * be zero.  When in use, the first byte is set to MAGIC, and the second
79  * byte is the size index.  The remaining bytes are for alignment.
80  * If range checking is enabled then a second word holds the size of the
81  * requested block, less 1, rounded up to a multiple of sizeof(RMAGIC).
82  * The order of elements is critical: ov_magic must overlay the low order
83  * bits of ov_next, and ov_magic can not be a valid ov_next bit pattern.
84  */
85 union	overhead {
86 	union	overhead *ov_next;	/* when free */
87 	struct {
88 		u_char	ovu_magic;	/* magic number */
89 		u_char	ovu_index;	/* bucket # */
90 #ifdef RCHECK
91 		u_short	ovu_rmagic;	/* range magic number */
92 		u_int	ovu_size;	/* actual block size */
93 #endif
94 	} ovu;
95 #define	ov_magic	ovu.ovu_magic
96 #define	ov_index	ovu.ovu_index
97 #define	ov_rmagic	ovu.ovu_rmagic
98 #define	ov_size		ovu.ovu_size
99 };
100 
101 #define	MAGIC		0xef		/* magic # on accounting info */
102 #define RMAGIC		0x5555		/* magic # on range info */
103 
104 #ifdef RCHECK
105 #define	RSLOP		sizeof (u_short)
106 #else
107 #define	RSLOP		0
108 #endif
109 
110 /*
111  * nextf[i] is the pointer to the next free block of size 2^(i+3).  The
112  * smallest allocatable block is 8 bytes.  The overhead information
113  * precedes the data area returned to the user.
114  */
115 #define	NBUCKETS 30
116 static	union overhead *nextf[NBUCKETS];
117 
118 static	int pagesz;			/* page size */
119 static	int pagebucket;			/* page size bucket */
120 
121 #ifdef MSTATS
122 /*
123  * nmalloc[i] is the difference between the number of mallocs and frees
124  * for a given block size.
125  */
126 static	u_int nmalloc[NBUCKETS];
127 #include <stdio.h>
128 #endif
129 
130 #if defined(MALLOC_DEBUG) || defined(RCHECK)
131 #define	ASSERT(p)   if (!(p)) botch("p")
132 #include <stdio.h>
133 static void
134 botch(s)
135 	char *s;
136 {
137 	fprintf(stderr, "\r\nassertion botched: %s\r\n", s);
138  	(void) fflush(stderr);		/* just in case user buffered it */
139 	abort();
140 }
141 #else
142 #define	ASSERT(p)
143 #endif
144 
145 /* Debugging stuff */
146 #define TRACE()	rtld_printf("TRACE %s:%d\n", __FILE__, __LINE__)
147 
148 void *
149 malloc(nbytes)
150 	size_t nbytes;
151 {
152   	register union overhead *op;
153   	register int bucket;
154 	register long n;
155 	register unsigned amt;
156 
157 	/*
158 	 * First time malloc is called, setup page size and
159 	 * align break pointer so all data will be page aligned.
160 	 */
161 	if (pagesz == 0) {
162 		pagesz = n = getpagesize();
163 		if (morepages(NPOOLPAGES) == 0)
164 			return NULL;
165 		op = (union overhead *)(pagepool_start);
166   		n = n - sizeof (*op) - ((long)op & (n - 1));
167 		if (n < 0)
168 			n += pagesz;
169   		if (n) {
170 			pagepool_start += n;
171 		}
172 		bucket = 0;
173 		amt = 8;
174 		while ((unsigned)pagesz > amt) {
175 			amt <<= 1;
176 			bucket++;
177 		}
178 		pagebucket = bucket;
179 	}
180 	/*
181 	 * Convert amount of memory requested into closest block size
182 	 * stored in hash buckets which satisfies request.
183 	 * Account for space used per block for accounting.
184 	 */
185 	if (nbytes <= (unsigned long)(n = pagesz - sizeof (*op) - RSLOP)) {
186 #ifndef RCHECK
187 		amt = 8;	/* size of first bucket */
188 		bucket = 0;
189 #else
190 		amt = 16;	/* size of first bucket */
191 		bucket = 1;
192 #endif
193 		n = -(sizeof (*op) + RSLOP);
194 	} else {
195 		amt = pagesz;
196 		bucket = pagebucket;
197 	}
198 	while (nbytes > amt + n) {
199 		amt <<= 1;
200 		if (amt == 0)
201 			return (NULL);
202 		bucket++;
203 	}
204 	/*
205 	 * If nothing in hash bucket right now,
206 	 * request more memory from the system.
207 	 */
208   	if ((op = nextf[bucket]) == NULL) {
209   		morecore(bucket);
210   		if ((op = nextf[bucket]) == NULL)
211   			return (NULL);
212 	}
213 	/* remove from linked list */
214   	nextf[bucket] = op->ov_next;
215 	op->ov_magic = MAGIC;
216 	op->ov_index = bucket;
217 #ifdef MSTATS
218   	nmalloc[bucket]++;
219 #endif
220 #ifdef RCHECK
221 	/*
222 	 * Record allocated size of block and
223 	 * bound space with magic numbers.
224 	 */
225 	op->ov_size = (nbytes + RSLOP - 1) & ~(RSLOP - 1);
226 	op->ov_rmagic = RMAGIC;
227   	*(u_short *)((caddr_t)(op + 1) + op->ov_size) = RMAGIC;
228 #endif
229   	return ((char *)(op + 1));
230 }
231 
232 /*
233  * Used by rtld.c, if we don't override it here the calloc from
234  * libc may try to pull in the malloc/realloc/free from libc too.
235  */
236 void *
237 calloc(size_t num, size_t size)
238 {
239 	void *p;
240 
241 	size *= num;
242 	if ((p = malloc(size)) != NULL)
243 		bzero(p, size);
244 	return(p);
245 }
246 
247 /*
248  * Allocate more memory to the indicated bucket.
249  */
250 static void
251 morecore(bucket)
252 	int bucket;
253 {
254   	register union overhead *op;
255 	register int sz;		/* size of desired block */
256   	int amt;			/* amount to allocate */
257   	int nblks;			/* how many blocks we get */
258 
259 	/*
260 	 * sbrk_size <= 0 only for big, FLUFFY, requests (about
261 	 * 2^30 bytes on a VAX, I think) or for a negative arg.
262 	 */
263 	sz = 1 << (bucket + 3);
264 #ifdef MALLOC_DEBUG
265 	ASSERT(sz > 0);
266 #else
267 	if (sz <= 0)
268 		return;
269 #endif
270 	if (sz < pagesz) {
271 		amt = pagesz;
272   		nblks = amt / sz;
273 	} else {
274 		amt = sz + pagesz;
275 		nblks = 1;
276 	}
277 	if (amt > pagepool_end - pagepool_start)
278 		if (morepages(amt/pagesz + NPOOLPAGES) == 0)
279 			return;
280 	op = (union overhead *)pagepool_start;
281 	pagepool_start += amt;
282 
283 	/*
284 	 * Add new memory allocated to that on
285 	 * free list for this hash bucket.
286 	 */
287   	nextf[bucket] = op;
288   	while (--nblks > 0) {
289 		op->ov_next = (union overhead *)((caddr_t)op + sz);
290 		op = (union overhead *)((caddr_t)op + sz);
291   	}
292 }
293 
294 void
295 free(cp)
296 	void *cp;
297 {
298   	register int size;
299 	register union overhead *op;
300 
301   	if (cp == NULL)
302   		return;
303 	op = (union overhead *)((caddr_t)cp - sizeof (union overhead));
304 #ifdef MALLOC_DEBUG
305   	ASSERT(op->ov_magic == MAGIC);		/* make sure it was in use */
306 #else
307 	if (op->ov_magic != MAGIC)
308 		return;				/* sanity */
309 #endif
310 #ifdef RCHECK
311   	ASSERT(op->ov_rmagic == RMAGIC);
312 	ASSERT(*(u_short *)((caddr_t)(op + 1) + op->ov_size) == RMAGIC);
313 #endif
314   	size = op->ov_index;
315   	ASSERT(size < NBUCKETS);
316 	op->ov_next = nextf[size];	/* also clobbers ov_magic */
317   	nextf[size] = op;
318 #ifdef MSTATS
319   	nmalloc[size]--;
320 #endif
321 }
322 
323 /*
324  * When a program attempts "storage compaction" as mentioned in the
325  * old malloc man page, it realloc's an already freed block.  Usually
326  * this is the last block it freed; occasionally it might be farther
327  * back.  We have to search all the free lists for the block in order
328  * to determine its bucket: 1st we make one pass thru the lists
329  * checking only the first block in each; if that fails we search
330  * ``realloc_srchlen'' blocks in each list for a match (the variable
331  * is extern so the caller can modify it).  If that fails we just copy
332  * however many bytes was given to realloc() and hope it's not huge.
333  */
334 int realloc_srchlen = 4;	/* 4 should be plenty, -1 =>'s whole list */
335 
336 void *
337 realloc(cp, nbytes)
338 	void *cp;
339 	size_t nbytes;
340 {
341   	register u_int onb;
342 	register int i;
343 	union overhead *op;
344   	char *res;
345 	int was_alloced = 0;
346 
347   	if (cp == NULL)
348   		return (malloc(nbytes));
349 	op = (union overhead *)((caddr_t)cp - sizeof (union overhead));
350 	if (op->ov_magic == MAGIC) {
351 		was_alloced++;
352 		i = op->ov_index;
353 	} else {
354 		/*
355 		 * Already free, doing "compaction".
356 		 *
357 		 * Search for the old block of memory on the
358 		 * free list.  First, check the most common
359 		 * case (last element free'd), then (this failing)
360 		 * the last ``realloc_srchlen'' items free'd.
361 		 * If all lookups fail, then assume the size of
362 		 * the memory block being realloc'd is the
363 		 * largest possible (so that all "nbytes" of new
364 		 * memory are copied into).  Note that this could cause
365 		 * a memory fault if the old area was tiny, and the moon
366 		 * is gibbous.  However, that is very unlikely.
367 		 */
368 		if ((i = findbucket(op, 1)) < 0 &&
369 		    (i = findbucket(op, realloc_srchlen)) < 0)
370 			i = NBUCKETS;
371 	}
372 	onb = 1 << (i + 3);
373 	if (onb < (u_int)pagesz)
374 		onb -= sizeof (*op) + RSLOP;
375 	else
376 		onb += pagesz - sizeof (*op) - RSLOP;
377 	/* avoid the copy if same size block */
378 	if (was_alloced) {
379 		if (i) {
380 			i = 1 << (i + 2);
381 			if (i < pagesz)
382 				i -= sizeof (*op) + RSLOP;
383 			else
384 				i += pagesz - sizeof (*op) - RSLOP;
385 		}
386 		if (nbytes <= onb && nbytes > (size_t)i) {
387 #ifdef RCHECK
388 			op->ov_size = (nbytes + RSLOP - 1) & ~(RSLOP - 1);
389 			*(u_short *)((caddr_t)(op + 1) + op->ov_size) = RMAGIC;
390 #endif
391 			return(cp);
392 		} else
393 			free(cp);
394 	}
395   	if ((res = malloc(nbytes)) == NULL)
396   		return (NULL);
397   	if (cp != res)		/* common optimization if "compacting" */
398 		bcopy(cp, res, (nbytes < onb) ? nbytes : onb);
399   	return (res);
400 }
401 
402 /*
403  * Search ``srchlen'' elements of each free list for a block whose
404  * header starts at ``freep''.  If srchlen is -1 search the whole list.
405  * Return bucket number, or -1 if not found.
406  */
407 static int
408 findbucket(freep, srchlen)
409 	union overhead *freep;
410 	int srchlen;
411 {
412 	register union overhead *p;
413 	register int i, j;
414 
415 	for (i = 0; i < NBUCKETS; i++) {
416 		j = 0;
417 		for (p = nextf[i]; p && j != srchlen; p = p->ov_next) {
418 			if (p == freep)
419 				return (i);
420 			j++;
421 		}
422 	}
423 	return (-1);
424 }
425 
426 #ifdef MSTATS
427 /*
428  * mstats - print out statistics about malloc
429  *
430  * Prints two lines of numbers, one showing the length of the free list
431  * for each size category, the second showing the number of mallocs -
432  * frees for each size category.
433  */
434 mstats(s)
435 	char *s;
436 {
437   	register int i, j;
438   	register union overhead *p;
439   	int totfree = 0,
440   	totused = 0;
441 
442   	fprintf(stderr, "Memory allocation statistics %s\nfree:\t", s);
443   	for (i = 0; i < NBUCKETS; i++) {
444   		for (j = 0, p = nextf[i]; p; p = p->ov_next, j++)
445   			;
446   		fprintf(stderr, " %d", j);
447   		totfree += j * (1 << (i + 3));
448   	}
449   	fprintf(stderr, "\nused:\t");
450   	for (i = 0; i < NBUCKETS; i++) {
451   		fprintf(stderr, " %d", nmalloc[i]);
452   		totused += nmalloc[i] * (1 << (i + 3));
453   	}
454   	fprintf(stderr, "\n\tTotal in use: %d, total free: %d\n",
455 	    totused, totfree);
456 }
457 #endif
458 
459 
460 static int
461 morepages(n)
462 int	n;
463 {
464 	int	fd = -1;
465 	int	offset;
466 
467 #ifdef NEED_DEV_ZERO
468 	fd = open(_PATH_DEVZERO, O_RDWR, 0);
469 	if (fd == -1)
470 		perror(_PATH_DEVZERO);
471 #endif
472 
473 	if (pagepool_end - pagepool_start > pagesz) {
474 		caddr_t	addr = (caddr_t)
475 			(((long)pagepool_start + pagesz - 1) & ~(pagesz - 1));
476 		if (munmap(addr, pagepool_end - addr) != 0)
477 			rtld_fdprintf(STDERR_FILENO, "morepages: munmap %p",
478 			    addr);
479 	}
480 
481 	offset = (long)pagepool_start - ((long)pagepool_start & ~(pagesz - 1));
482 
483 	if ((pagepool_start = mmap(0, n * pagesz,
484 			PROT_READ|PROT_WRITE,
485 			MAP_ANON|MAP_COPY, fd, 0)) == (caddr_t)-1) {
486 		rtld_printf("Cannot map anonymous memory\n");
487 		return 0;
488 	}
489 	pagepool_end = pagepool_start + n * pagesz;
490 	pagepool_start += offset;
491 
492 #ifdef NEED_DEV_ZERO
493 	close(fd);
494 #endif
495 	return n;
496 }
497