xref: /dragonfly/libexec/rtld-elf/malloc.c (revision 0212bfce)
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. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 /*
31  * malloc.c (Caltech) 2/21/82
32  * Chris Kingsley, kingsley@cit-20.
33  *
34  * This is a very fast storage allocator.  It allocates blocks of a small
35  * number of different sizes, and keeps free lists of each size.  Blocks that
36  * don't exactly fit are passed up to the next larger size.  In this
37  * implementation, the available sizes are 2^n-4 (or 2^n-10) bytes long.
38  * This is designed for use in a virtual memory environment.
39  */
40 
41 #include <sys/types.h>
42 #include <paths.h>
43 #include <stdarg.h>
44 #include <stdio.h>
45 #include <stdlib.h>
46 #include <string.h>
47 #include <unistd.h>
48 #include <sys/param.h>
49 #include <sys/mman.h>
50 #include "rtld_printf.h"
51 
52 static void morecore();
53 static int findbucket();
54 
55 /*
56  * Pre-allocate mmap'ed pages
57  */
58 #define	NPOOLPAGES	(32*1024/pagesz)
59 static caddr_t		pagepool_start, pagepool_end;
60 static int		morepages();
61 
62 /*
63  * The overhead on a block is at least 4 bytes.  When free, this space
64  * contains a pointer to the next free block, and the bottom two bits must
65  * be zero.  When in use, the first byte is set to MAGIC, and the second
66  * byte is the size index.  The remaining bytes are for alignment.
67  * If range checking is enabled then a second word holds the size of the
68  * requested block, less 1, rounded up to a multiple of sizeof(RMAGIC).
69  * The order of elements is critical: ov_magic must overlay the low order
70  * bits of ov_next, and ov_magic can not be a valid ov_next bit pattern.
71  */
72 union	overhead {
73 	union	overhead *ov_next;	/* when free */
74 	struct {
75 		u_char	ovu_magic;	/* magic number */
76 		u_char	ovu_index;	/* bucket # */
77 #ifdef RCHECK
78 		u_short	ovu_rmagic;	/* range magic number */
79 		u_int	ovu_size;	/* actual block size */
80 #endif
81 	} ovu;
82 #define	ov_magic	ovu.ovu_magic
83 #define	ov_index	ovu.ovu_index
84 #define	ov_rmagic	ovu.ovu_rmagic
85 #define	ov_size		ovu.ovu_size
86 };
87 
88 #define	MAGIC		0xef		/* magic # on accounting info */
89 #define RMAGIC		0x5555		/* magic # on range info */
90 
91 #ifdef RCHECK
92 #define	RSLOP		sizeof (u_short)
93 #else
94 #define	RSLOP		0
95 #endif
96 
97 /*
98  * nextf[i] is the pointer to the next free block of size 2^(i+3).  The
99  * smallest allocatable block is 8 bytes.  The overhead information
100  * precedes the data area returned to the user.
101  */
102 #define	NBUCKETS 30
103 static	union overhead *nextf[NBUCKETS];
104 
105 static	int pagesz;			/* page size */
106 static	int pagebucket;			/* page size bucket */
107 
108 #ifdef MSTATS
109 /*
110  * nmalloc[i] is the difference between the number of mallocs and frees
111  * for a given block size.
112  */
113 static	u_int nmalloc[NBUCKETS];
114 #include <stdio.h>
115 #endif
116 
117 #if defined(MALLOC_DEBUG) || defined(RCHECK)
118 #define	ASSERT(p)   if (!(p)) botch("p")
119 #include <stdio.h>
120 static void
121 botch(char *s)
122 {
123 	fprintf(stderr, "\r\nassertion botched: %s\r\n", s);
124 	(void) fflush(stderr);		/* just in case user buffered it */
125 	abort();
126 }
127 #else
128 #define	ASSERT(p)
129 #endif
130 
131 /* Debugging stuff */
132 #define TRACE()	rtld_printf("TRACE %s:%d\n", __FILE__, __LINE__)
133 
134 void *
135 malloc(size_t nbytes)
136 {
137 	union overhead *op;
138 	int bucket;
139 	long n;
140 	size_t amt;
141 
142 	/*
143 	 * First time malloc is called, setup page size and
144 	 * align break pointer so all data will be page aligned.
145 	 */
146 	if (pagesz == 0) {
147 		pagesz = n = getpagesize();
148 		if (morepages(NPOOLPAGES) == 0)
149 			return NULL;
150 		op = (union overhead *)(pagepool_start);
151 		n = n - sizeof (*op) - ((long)op & (n - 1));
152 		if (n < 0)
153 			n += pagesz;
154 		if (n) {
155 			pagepool_start += n;
156 		}
157 		bucket = 0;
158 		amt = 8;
159 		while ((size_t)pagesz > amt) {
160 			amt <<= 1;
161 			bucket++;
162 		}
163 		pagebucket = bucket;
164 	}
165 	/*
166 	 * Convert amount of memory requested into closest block size
167 	 * stored in hash buckets which satisfies request.
168 	 * Account for space used per block for accounting.
169 	 */
170 	if (nbytes <= (unsigned long)(n = pagesz - sizeof (*op) - RSLOP)) {
171 #ifndef RCHECK
172 		amt = 8;	/* size of first bucket */
173 		bucket = 0;
174 #else
175 		amt = 16;	/* size of first bucket */
176 		bucket = 1;
177 #endif
178 		n = -(sizeof (*op) + RSLOP);
179 	} else {
180 		amt = (size_t)pagesz;
181 		bucket = pagebucket;
182 	}
183 	while (nbytes > amt + n) {
184 		amt <<= 1;
185 		if (amt == 0)
186 			return (NULL);
187 		bucket++;
188 	}
189 	/*
190 	 * If nothing in hash bucket right now,
191 	 * request more memory from the system.
192 	 */
193 	if ((op = nextf[bucket]) == NULL) {
194 		morecore(bucket);
195 		if ((op = nextf[bucket]) == NULL)
196 			return (NULL);
197 	}
198 	/* remove from linked list */
199 	nextf[bucket] = op->ov_next;
200 	op->ov_magic = MAGIC;
201 	op->ov_index = bucket;
202 #ifdef MSTATS
203 	nmalloc[bucket]++;
204 #endif
205 #ifdef RCHECK
206 	/*
207 	 * Record allocated size of block and
208 	 * bound space with magic numbers.
209 	 */
210 	op->ov_size = roundup2(nbytes, RSLOP);
211 	op->ov_rmagic = RMAGIC;
212 	*(u_short *)((caddr_t)(op + 1) + op->ov_size) = RMAGIC;
213 #endif
214 	return ((char *)(op + 1));
215 }
216 
217 /*
218  * Used by rtld.c, if we don't override it here the calloc from
219  * libc may try to pull in the malloc/realloc/free from libc too.
220  */
221 void *
222 calloc(size_t num, size_t size)
223 {
224 	void *p;
225 
226 	size *= num;
227 	if ((p = malloc(size)) != NULL)
228 		bzero(p, size);
229 	return(p);
230 }
231 
232 /*
233  * Allocate more memory to the indicated bucket.
234  */
235 static void
236 morecore(int bucket)
237 {
238 	union overhead *op;
239 	size_t sz;		/* size of desired block */
240 	size_t amt;		/* amount to allocate */
241 	int nblks;		/* how many blocks we get */
242 
243 	/*
244 	 * sbrk_size <= 0 only for big, FLUFFY, requests (about
245 	 * 2^30 bytes on a VAX, I think) or for a negative arg.
246 	 */
247 	sz = 1LU << (bucket + 3);
248 #ifdef MALLOC_DEBUG
249 	ASSERT(sz > 0);
250 #else
251 	if (sz == 0)
252 		return;
253 #endif
254 	if (sz < pagesz) {
255 		amt = pagesz;
256 		nblks = amt / sz;
257 	} else {
258 		amt = sz + pagesz;
259 		nblks = 1;
260 	}
261 	if (amt > pagepool_end - pagepool_start)
262 		if (morepages(amt/pagesz + NPOOLPAGES) == 0)
263 			return;
264 	op = (union overhead *)pagepool_start;
265 	pagepool_start += amt;
266 
267 	/*
268 	 * Add new memory allocated to that on
269 	 * free list for this hash bucket.
270 	 */
271 	nextf[bucket] = op;
272 	while (--nblks > 0) {
273 		op->ov_next = (union overhead *)((caddr_t)op + sz);
274 		op = (union overhead *)((caddr_t)op + sz);
275 	}
276 }
277 
278 void
279 free(void *cp)
280 {
281 	int size;
282 	union overhead *op;
283 
284 	if (cp == NULL)
285 		return;
286 	op = (union overhead *)((caddr_t)cp - sizeof (union overhead));
287 #ifdef MALLOC_DEBUG
288 	ASSERT(op->ov_magic == MAGIC);		/* make sure it was in use */
289 #else
290 	if (op->ov_magic != MAGIC)
291 		return;				/* sanity */
292 #endif
293 #ifdef RCHECK
294 	ASSERT(op->ov_rmagic == RMAGIC);
295 	ASSERT(*(u_short *)((caddr_t)(op + 1) + op->ov_size) == RMAGIC);
296 #endif
297 	size = op->ov_index;
298 	ASSERT(size < NBUCKETS);
299 	op->ov_next = nextf[size];	/* also clobbers ov_magic */
300 	nextf[size] = op;
301 #ifdef MSTATS
302 	nmalloc[size]--;
303 #endif
304 }
305 
306 /*
307  * When a program attempts "storage compaction" as mentioned in the
308  * old malloc man page, it realloc's an already freed block.  Usually
309  * this is the last block it freed; occasionally it might be farther
310  * back.  We have to search all the free lists for the block in order
311  * to determine its bucket: 1st we make one pass thru the lists
312  * checking only the first block in each; if that fails we search
313  * ``realloc_srchlen'' blocks in each list for a match (the variable
314  * is extern so the caller can modify it).  If that fails we just copy
315  * however many bytes was given to realloc() and hope it's not huge.
316  */
317 int realloc_srchlen = 4;	/* 4 should be plenty, -1 =>'s whole list */
318 
319 void *
320 realloc(void *cp, size_t nbytes)
321 {
322 	size_t onb;
323 	int i;
324 	union overhead *op;
325 	char *res;
326 	int was_alloced = 0;
327 
328 	if (cp == NULL)
329 		return (malloc(nbytes));
330 	op = (union overhead *)((caddr_t)cp - sizeof (union overhead));
331 	if (op->ov_magic == MAGIC) {
332 		was_alloced++;
333 		i = op->ov_index;
334 	} else {
335 		/*
336 		 * Already free, doing "compaction".
337 		 *
338 		 * Search for the old block of memory on the
339 		 * free list.  First, check the most common
340 		 * case (last element free'd), then (this failing)
341 		 * the last ``realloc_srchlen'' items free'd.
342 		 * If all lookups fail, then assume the size of
343 		 * the memory block being realloc'd is the
344 		 * largest possible (so that all "nbytes" of new
345 		 * memory are copied into).  Note that this could cause
346 		 * a memory fault if the old area was tiny, and the moon
347 		 * is gibbous.  However, that is very unlikely.
348 		 */
349 		if ((i = findbucket(op, 1)) < 0 &&
350 		    (i = findbucket(op, realloc_srchlen)) < 0)
351 			i = NBUCKETS;
352 	}
353 	onb = 1LU << (i + 3);
354 	if (onb < (u_int)pagesz)
355 		onb -= sizeof (*op) + RSLOP;
356 	else
357 		onb += pagesz - sizeof (*op) - RSLOP;
358 	/* avoid the copy if same size block */
359 	if (was_alloced) {
360 		if (i) {
361 			i = 1LU << (i + 2);
362 			if (i < pagesz)
363 				i -= sizeof (*op) + RSLOP;
364 			else
365 				i += pagesz - sizeof (*op) - RSLOP;
366 		}
367 		if (nbytes <= onb && nbytes > (size_t)i) {
368 #ifdef RCHECK
369 			op->ov_size = roundup2(nbytes, RSLOP);
370 			*(u_short *)((caddr_t)(op + 1) + op->ov_size) = RMAGIC;
371 #endif
372 			return(cp);
373 		} else
374 			free(cp);
375 	}
376 	if ((res = malloc(nbytes)) == NULL)
377 		return (NULL);
378 	if (cp != res)		/* common optimization if "compacting" */
379 		bcopy(cp, res, (nbytes < onb) ? nbytes : onb);
380 	return (res);
381 }
382 
383 /*
384  * Search ``srchlen'' elements of each free list for a block whose
385  * header starts at ``freep''.  If srchlen is -1 search the whole list.
386  * Return bucket number, or -1 if not found.
387  */
388 static int
389 findbucket(union overhead *freep, int srchlen)
390 {
391 	union overhead *p;
392 	int i, j;
393 
394 	for (i = 0; i < NBUCKETS; i++) {
395 		j = 0;
396 		for (p = nextf[i]; p && j != srchlen; p = p->ov_next) {
397 			if (p == freep)
398 				return (i);
399 			j++;
400 		}
401 	}
402 	return (-1);
403 }
404 
405 #ifdef MSTATS
406 /*
407  * mstats - print out statistics about malloc
408  *
409  * Prints two lines of numbers, one showing the length of the free list
410  * for each size category, the second showing the number of mallocs -
411  * frees for each size category.
412  */
413 void
414 mstats(char *s)
415 {
416 	int i, j;
417 	union overhead *p;
418 	size_t totfree, totused;
419 
420 	totfree = 0;
421 	totused = 0;
422 
423 	fprintf(stderr, "Memory allocation statistics %s\nfree:\t", s);
424 	for (i = 0; i < NBUCKETS; i++) {
425 		for (j = 0, p = nextf[i]; p; p = p->ov_next, j++)
426 			;
427 		fprintf(stderr, " %d", j);
428 		totfree += j * (1LU << (i + 3));
429 	}
430 	fprintf(stderr, "\nused:\t");
431 	for (i = 0; i < NBUCKETS; i++) {
432 		fprintf(stderr, " %d", nmalloc[i]);
433 		totused += nmalloc[i] * (1LU << (i + 3));
434 	}
435 	fprintf(stderr, "\n\tTotal in use: %zd, total free: %zd\n",
436 		totused, totfree);
437 }
438 #endif
439 
440 
441 static int
442 morepages(int n)
443 {
444 	int	fd = -1;
445 	int	offset;
446 
447 	if (pagepool_end - pagepool_start > pagesz) {
448 		caddr_t	addr = (caddr_t)
449 			roundup2((long)pagepool_start, pagesz);
450 		if (munmap(addr, pagepool_end - addr) != 0)
451 			rtld_fdprintf(STDERR_FILENO, "morepages: munmap %p",
452 			    addr);
453 	}
454 
455 	offset = (long)pagepool_start - rounddown2((long)pagepool_start, pagesz);
456 
457 	if ((pagepool_start = mmap(0, n * pagesz,
458 			PROT_READ|PROT_WRITE,
459 			MAP_ANON|MAP_COPY, fd, 0)) == (caddr_t)-1) {
460 		rtld_printf("Cannot map anonymous memory\n");
461 		return 0;
462 	}
463 	pagepool_end = pagepool_start + n * pagesz;
464 	pagepool_start += offset;
465 
466 	return n;
467 }
468