xref: /freebsd/libexec/rtld-elf/rtld_malloc.c (revision daceb336172a6b0572de864b97e70b28451ca636)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1983 Regents of the University of California.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 #if defined(LIBC_SCCS) && !defined(lint)
33 /*static char *sccsid = "from: @(#)malloc.c	5.11 (Berkeley) 2/23/91";*/
34 static char *rcsid = "$FreeBSD$";
35 #endif /* LIBC_SCCS and not lint */
36 
37 /*
38  * malloc.c (Caltech) 2/21/82
39  * Chris Kingsley, kingsley@cit-20.
40  *
41  * This is a very fast storage allocator.  It allocates blocks of a small
42  * number of different sizes, and keeps free lists of each size.  Blocks that
43  * don't exactly fit are passed up to the next larger size.  In this
44  * implementation, the available sizes are 2^n-4 (or 2^n-10) bytes long.
45  * This is designed for use in a virtual memory environment.
46  */
47 
48 #include <sys/param.h>
49 #include <sys/sysctl.h>
50 #include <sys/mman.h>
51 #include <errno.h>
52 #include <stdarg.h>
53 #include <stddef.h>
54 #include <string.h>
55 #include <unistd.h>
56 #include "rtld.h"
57 #include "rtld_printf.h"
58 #include "paths.h"
59 
60 /*
61  * Pre-allocate mmap'ed pages
62  */
63 #define	NPOOLPAGES	(128*1024/pagesz)
64 static caddr_t		pagepool_start, pagepool_end;
65 
66 /*
67  * The overhead on a block is at least 4 bytes.  When free, this space
68  * contains a pointer to the next free block, and the bottom two bits must
69  * be zero.  When in use, the first byte is set to MAGIC, and the second
70  * byte is the size index.  The remaining bytes are for alignment.
71  * If range checking is enabled then a second word holds the size of the
72  * requested block, less 1, rounded up to a multiple of sizeof(RMAGIC).
73  * The order of elements is critical: ov_magic must overlay the low order
74  * bits of ov_next, and ov_magic can not be a valid ov_next bit pattern.
75  */
76 union	overhead {
77 	union	overhead *ov_next;	/* when free */
78 	struct {
79 		u_char	ovu_magic;	/* magic number */
80 		u_char	ovu_index;	/* bucket # */
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 static void morecore(int bucket);
89 static int morepages(int n);
90 static int findbucket(union overhead *freep, int srchlen);
91 
92 
93 #define	MAGIC		0xef		/* magic # on accounting info */
94 #define RMAGIC		0x5555		/* magic # on range info */
95 
96 /*
97  * nextf[i] is the pointer to the next free block of size 2^(i+3).  The
98  * smallest allocatable block is 8 bytes.  The overhead information
99  * precedes the data area returned to the user.
100  */
101 #define	NBUCKETS 30
102 static	union overhead *nextf[NBUCKETS];
103 
104 static	int pagesz;			/* page size */
105 static	int pagebucket;			/* page size bucket */
106 
107 /*
108  * The array of supported page sizes is provided by the user, i.e., the
109  * program that calls this storage allocator.  That program must initialize
110  * the array before making its first call to allocate storage.  The array
111  * must contain at least one page size.  The page sizes must be stored in
112  * increasing order.
113  */
114 
115 void *
116 __crt_malloc(size_t nbytes)
117 {
118 	union overhead *op;
119 	int bucket;
120 	ssize_t n;
121 	size_t amt;
122 
123 	/*
124 	 * First time malloc is called, setup page size and
125 	 * align break pointer so all data will be page aligned.
126 	 */
127 	if (pagesz == 0) {
128 		pagesz = n = pagesizes[0];
129 		if (morepages(NPOOLPAGES) == 0)
130 			return NULL;
131 		op = (union overhead *)(pagepool_start);
132   		n = n - sizeof (*op) - ((long)op & (n - 1));
133 		if (n < 0)
134 			n += pagesz;
135   		if (n) {
136 			pagepool_start += n;
137 		}
138 		bucket = 0;
139 		amt = 8;
140 		while ((unsigned)pagesz > amt) {
141 			amt <<= 1;
142 			bucket++;
143 		}
144 		pagebucket = bucket;
145 	}
146 	/*
147 	 * Convert amount of memory requested into closest block size
148 	 * stored in hash buckets which satisfies request.
149 	 * Account for space used per block for accounting.
150 	 */
151 	if (nbytes <= (unsigned long)(n = pagesz - sizeof(*op))) {
152 		amt = 8;	/* size of first bucket */
153 		bucket = 0;
154 		n = -sizeof(*op);
155 	} else {
156 		amt = pagesz;
157 		bucket = pagebucket;
158 	}
159 	while (nbytes > amt + n) {
160 		amt <<= 1;
161 		if (amt == 0)
162 			return (NULL);
163 		bucket++;
164 	}
165 	/*
166 	 * If nothing in hash bucket right now,
167 	 * request more memory from the system.
168 	 */
169   	if ((op = nextf[bucket]) == NULL) {
170   		morecore(bucket);
171   		if ((op = nextf[bucket]) == NULL)
172   			return (NULL);
173 	}
174 	/* remove from linked list */
175   	nextf[bucket] = op->ov_next;
176 	op->ov_magic = MAGIC;
177 	op->ov_index = bucket;
178   	return ((char *)(op + 1));
179 }
180 
181 void *
182 __crt_calloc(size_t num, size_t size)
183 {
184 	void *ret;
185 
186 	if (size != 0 && (num * size) / size != num) {
187 		/* size_t overflow. */
188 		return (NULL);
189 	}
190 
191 	if ((ret = __crt_malloc(num * size)) != NULL)
192 		memset(ret, 0, num * size);
193 
194 	return (ret);
195 }
196 
197 /*
198  * Allocate more memory to the indicated bucket.
199  */
200 static void
201 morecore(int bucket)
202 {
203 	union overhead *op;
204 	int sz;		/* size of desired block */
205   	int amt;			/* amount to allocate */
206   	int nblks;			/* how many blocks we get */
207 
208 	/*
209 	 * sbrk_size <= 0 only for big, FLUFFY, requests (about
210 	 * 2^30 bytes on a VAX, I think) or for a negative arg.
211 	 */
212 	if ((unsigned)bucket >= NBBY * sizeof(int) - 4)
213 		return;
214 	sz = 1 << (bucket + 3);
215 	if (sz < pagesz) {
216 		amt = pagesz;
217   		nblks = amt / sz;
218 	} else {
219 		amt = sz + pagesz;
220 		nblks = 1;
221 	}
222 	if (amt > pagepool_end - pagepool_start)
223 		if (morepages(amt/pagesz + NPOOLPAGES) == 0)
224 			return;
225 	op = (union overhead *)pagepool_start;
226 	pagepool_start += amt;
227 
228 	/*
229 	 * Add new memory allocated to that on
230 	 * free list for this hash bucket.
231 	 */
232   	nextf[bucket] = op;
233   	while (--nblks > 0) {
234 		op->ov_next = (union overhead *)((caddr_t)op + sz);
235 		op = (union overhead *)((caddr_t)op + sz);
236   	}
237 }
238 
239 void
240 __crt_free(void *cp)
241 {
242 	int size;
243 	union overhead *op;
244 
245   	if (cp == NULL)
246   		return;
247 	op = (union overhead *)((caddr_t)cp - sizeof (union overhead));
248 	if (op->ov_magic != MAGIC)
249 		return;				/* sanity */
250   	size = op->ov_index;
251 	op->ov_next = nextf[size];	/* also clobbers ov_magic */
252   	nextf[size] = op;
253 }
254 
255 /*
256  * When a program attempts "storage compaction" as mentioned in the
257  * old malloc man page, it realloc's an already freed block.  Usually
258  * this is the last block it freed; occasionally it might be farther
259  * back.  We have to search all the free lists for the block in order
260  * to determine its bucket: 1st we make one pass through the lists
261  * checking only the first block in each; if that fails we search
262  * ``realloc_srchlen'' blocks in each list for a match (the variable
263  * is extern so the caller can modify it).  If that fails we just copy
264  * however many bytes was given to realloc() and hope it's not huge.
265  */
266 static int realloc_srchlen = 4;	/* 4 should be plenty, -1 =>'s whole list */
267 
268 void *
269 __crt_realloc(void *cp, size_t nbytes)
270 {
271 	u_int onb;
272 	int i;
273 	union overhead *op;
274   	char *res;
275 	int was_alloced = 0;
276 
277   	if (cp == NULL)
278 		return (__crt_malloc(nbytes));
279 	op = (union overhead *)((caddr_t)cp - sizeof (union overhead));
280 	if (op->ov_magic == MAGIC) {
281 		was_alloced++;
282 		i = op->ov_index;
283 	} else {
284 		/*
285 		 * Already free, doing "compaction".
286 		 *
287 		 * Search for the old block of memory on the
288 		 * free list.  First, check the most common
289 		 * case (last element free'd), then (this failing)
290 		 * the last ``realloc_srchlen'' items free'd.
291 		 * If all lookups fail, then assume the size of
292 		 * the memory block being realloc'd is the
293 		 * largest possible (so that all "nbytes" of new
294 		 * memory are copied into).  Note that this could cause
295 		 * a memory fault if the old area was tiny, and the moon
296 		 * is gibbous.  However, that is very unlikely.
297 		 */
298 		if ((i = findbucket(op, 1)) < 0 &&
299 		    (i = findbucket(op, realloc_srchlen)) < 0)
300 			i = NBUCKETS;
301 	}
302 	onb = 1 << (i + 3);
303 	if (onb < (u_int)pagesz)
304 		onb -= sizeof(*op);
305 	else
306 		onb += pagesz - sizeof(*op);
307 	/* avoid the copy if same size block */
308 	if (was_alloced) {
309 		if (i) {
310 			i = 1 << (i + 2);
311 			if (i < pagesz)
312 				i -= sizeof(*op);
313 			else
314 				i += pagesz - sizeof(*op);
315 		}
316 		if (nbytes <= onb && nbytes > (size_t)i)
317 			return (cp);
318 		__crt_free(cp);
319 	}
320   	if ((res = __crt_malloc(nbytes)) == NULL)
321 		return (NULL);
322   	if (cp != res)		/* common optimization if "compacting" */
323 		bcopy(cp, res, (nbytes < onb) ? nbytes : onb);
324   	return (res);
325 }
326 
327 /*
328  * Search ``srchlen'' elements of each free list for a block whose
329  * header starts at ``freep''.  If srchlen is -1 search the whole list.
330  * Return bucket number, or -1 if not found.
331  */
332 static int
333 findbucket(union overhead *freep, int srchlen)
334 {
335 	union overhead *p;
336 	int i, j;
337 
338 	for (i = 0; i < NBUCKETS; i++) {
339 		j = 0;
340 		for (p = nextf[i]; p && j != srchlen; p = p->ov_next) {
341 			if (p == freep)
342 				return (i);
343 			j++;
344 		}
345 	}
346 	return (-1);
347 }
348 
349 static int
350 morepages(int n)
351 {
352 	int	fd = -1;
353 	int	offset;
354 
355 	if (pagepool_end - pagepool_start > pagesz) {
356 		caddr_t	addr = (caddr_t)
357 			(((long)pagepool_start + pagesz - 1) & ~(pagesz - 1));
358 		if (munmap(addr, pagepool_end - addr) != 0) {
359 #ifdef IN_RTLD
360 			rtld_fdprintf(STDERR_FILENO, _BASENAME_RTLD ": "
361 			    "morepages: cannot munmap %p: %s\n",
362 			    addr, rtld_strerror(errno));
363 #endif
364 		}
365 	}
366 
367 	offset = (long)pagepool_start - ((long)pagepool_start & ~(pagesz - 1));
368 
369 	if ((pagepool_start = mmap(0, n * pagesz,
370 			PROT_READ|PROT_WRITE,
371 			MAP_ANON|MAP_PRIVATE, fd, 0)) == (caddr_t)-1) {
372 #ifdef IN_RTLD
373 		rtld_fdprintf(STDERR_FILENO, _BASENAME_RTLD ": morepages: "
374 		    "cannot mmap anonymous memory: %s\n",
375 		    rtld_strerror(errno));
376 #endif
377 		return 0;
378 	}
379 	pagepool_end = pagepool_start + n * pagesz;
380 	pagepool_start += offset;
381 
382 	return n;
383 }
384