2 * Copyright (c) 1983 Regents of the University of California.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
35 * malloc.c (Caltech) 2/21/82
36 * Chris Kingsley, kingsley@cit-20.
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.
45 #include <sys/types.h>
52 #include <sys/param.h>
54 #include "rtld_printf.h"
56 static void morecore();
57 static int findbucket();
60 * Pre-allocate mmap'ed pages
62 #define NPOOLPAGES (32*1024/pagesz)
63 static caddr_t pagepool_start, pagepool_end;
64 static int morepages();
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.
77 union overhead *ov_next; /* when free */
79 u_char ovu_magic; /* magic number */
80 u_char ovu_index; /* bucket # */
82 u_short ovu_rmagic; /* range magic number */
83 u_int ovu_size; /* actual block size */
86 #define ov_magic ovu.ovu_magic
87 #define ov_index ovu.ovu_index
88 #define ov_rmagic ovu.ovu_rmagic
89 #define ov_size ovu.ovu_size
92 #define MAGIC 0xef /* magic # on accounting info */
93 #define RMAGIC 0x5555 /* magic # on range info */
96 #define RSLOP sizeof (u_short)
102 * nextf[i] is the pointer to the next free block of size 2^(i+3). The
103 * smallest allocatable block is 8 bytes. The overhead information
104 * precedes the data area returned to the user.
107 static union overhead *nextf[NBUCKETS];
109 static int pagesz; /* page size */
110 static int pagebucket; /* page size bucket */
114 * nmalloc[i] is the difference between the number of mallocs and frees
115 * for a given block size.
117 static u_int nmalloc[NBUCKETS];
121 #if defined(MALLOC_DEBUG) || defined(RCHECK)
122 #define ASSERT(p) if (!(p)) botch("p")
127 fprintf(stderr, "\r\nassertion botched: %s\r\n", s);
128 (void) fflush(stderr); /* just in case user buffered it */
135 /* Debugging stuff */
136 #define TRACE() rtld_printf("TRACE %s:%d\n", __FILE__, __LINE__)
139 malloc(size_t nbytes)
141 register union overhead *op;
144 register unsigned amt;
147 * First time malloc is called, setup page size and
148 * align break pointer so all data will be page aligned.
151 pagesz = n = getpagesize();
152 if (morepages(NPOOLPAGES) == 0)
154 op = (union overhead *)(pagepool_start);
155 n = n - sizeof (*op) - ((long)op & (n - 1));
163 while ((unsigned)pagesz > amt) {
170 * Convert amount of memory requested into closest block size
171 * stored in hash buckets which satisfies request.
172 * Account for space used per block for accounting.
174 if (nbytes <= (unsigned long)(n = pagesz - sizeof (*op) - RSLOP)) {
176 amt = 8; /* size of first bucket */
179 amt = 16; /* size of first bucket */
182 n = -(sizeof (*op) + RSLOP);
187 while (nbytes > amt + n) {
194 * If nothing in hash bucket right now,
195 * request more memory from the system.
197 if ((op = nextf[bucket]) == NULL) {
199 if ((op = nextf[bucket]) == NULL)
202 /* remove from linked list */
203 nextf[bucket] = op->ov_next;
204 op->ov_magic = MAGIC;
205 op->ov_index = bucket;
211 * Record allocated size of block and
212 * bound space with magic numbers.
214 op->ov_size = (nbytes + RSLOP - 1) & ~(RSLOP - 1);
215 op->ov_rmagic = RMAGIC;
216 *(u_short *)((caddr_t)(op + 1) + op->ov_size) = RMAGIC;
218 return ((char *)(op + 1));
222 * Used by rtld.c, if we don't override it here the calloc from
223 * libc may try to pull in the malloc/realloc/free from libc too.
226 calloc(size_t num, size_t size)
231 if ((p = malloc(size)) != NULL)
237 * Allocate more memory to the indicated bucket.
242 register union overhead *op;
243 register int sz; /* size of desired block */
244 int amt; /* amount to allocate */
245 int nblks; /* how many blocks we get */
248 * sbrk_size <= 0 only for big, FLUFFY, requests (about
249 * 2^30 bytes on a VAX, I think) or for a negative arg.
251 sz = 1 << (bucket + 3);
265 if (amt > pagepool_end - pagepool_start)
266 if (morepages(amt/pagesz + NPOOLPAGES) == 0)
268 op = (union overhead *)pagepool_start;
269 pagepool_start += amt;
272 * Add new memory allocated to that on
273 * free list for this hash bucket.
276 while (--nblks > 0) {
277 op->ov_next = (union overhead *)((caddr_t)op + sz);
278 op = (union overhead *)((caddr_t)op + sz);
286 register union overhead *op;
290 op = (union overhead *)((caddr_t)cp - sizeof (union overhead));
292 ASSERT(op->ov_magic == MAGIC); /* make sure it was in use */
294 if (op->ov_magic != MAGIC)
298 ASSERT(op->ov_rmagic == RMAGIC);
299 ASSERT(*(u_short *)((caddr_t)(op + 1) + op->ov_size) == RMAGIC);
302 ASSERT(size < NBUCKETS);
303 op->ov_next = nextf[size]; /* also clobbers ov_magic */
311 * When a program attempts "storage compaction" as mentioned in the
312 * old malloc man page, it realloc's an already freed block. Usually
313 * this is the last block it freed; occasionally it might be farther
314 * back. We have to search all the free lists for the block in order
315 * to determine its bucket: 1st we make one pass thru the lists
316 * checking only the first block in each; if that fails we search
317 * ``realloc_srchlen'' blocks in each list for a match (the variable
318 * is extern so the caller can modify it). If that fails we just copy
319 * however many bytes was given to realloc() and hope it's not huge.
321 int realloc_srchlen = 4; /* 4 should be plenty, -1 =>'s whole list */
324 realloc(void *cp, size_t nbytes)
333 return (malloc(nbytes));
334 op = (union overhead *)((caddr_t)cp - sizeof (union overhead));
335 if (op->ov_magic == MAGIC) {
340 * Already free, doing "compaction".
342 * Search for the old block of memory on the
343 * free list. First, check the most common
344 * case (last element free'd), then (this failing)
345 * the last ``realloc_srchlen'' items free'd.
346 * If all lookups fail, then assume the size of
347 * the memory block being realloc'd is the
348 * largest possible (so that all "nbytes" of new
349 * memory are copied into). Note that this could cause
350 * a memory fault if the old area was tiny, and the moon
351 * is gibbous. However, that is very unlikely.
353 if ((i = findbucket(op, 1)) < 0 &&
354 (i = findbucket(op, realloc_srchlen)) < 0)
358 if (onb < (u_int)pagesz)
359 onb -= sizeof (*op) + RSLOP;
361 onb += pagesz - sizeof (*op) - RSLOP;
362 /* avoid the copy if same size block */
367 i -= sizeof (*op) + RSLOP;
369 i += pagesz - sizeof (*op) - RSLOP;
371 if (nbytes <= onb && nbytes > (size_t)i) {
373 op->ov_size = (nbytes + RSLOP - 1) & ~(RSLOP - 1);
374 *(u_short *)((caddr_t)(op + 1) + op->ov_size) = RMAGIC;
380 if ((res = malloc(nbytes)) == NULL)
382 if (cp != res) /* common optimization if "compacting" */
383 bcopy(cp, res, (nbytes < onb) ? nbytes : onb);
388 * Search ``srchlen'' elements of each free list for a block whose
389 * header starts at ``freep''. If srchlen is -1 search the whole list.
390 * Return bucket number, or -1 if not found.
393 findbucket(union overhead *freep, int srchlen)
395 register union overhead *p;
398 for (i = 0; i < NBUCKETS; i++) {
400 for (p = nextf[i]; p && j != srchlen; p = p->ov_next) {
411 * mstats - print out statistics about malloc
413 * Prints two lines of numbers, one showing the length of the free list
414 * for each size category, the second showing the number of mallocs -
415 * frees for each size category.
421 register union overhead *p;
425 fprintf(stderr, "Memory allocation statistics %s\nfree:\t", s);
426 for (i = 0; i < NBUCKETS; i++) {
427 for (j = 0, p = nextf[i]; p; p = p->ov_next, j++)
429 fprintf(stderr, " %d", j);
430 totfree += j * (1 << (i + 3));
432 fprintf(stderr, "\nused:\t");
433 for (i = 0; i < NBUCKETS; i++) {
434 fprintf(stderr, " %d", nmalloc[i]);
435 totused += nmalloc[i] * (1 << (i + 3));
437 fprintf(stderr, "\n\tTotal in use: %d, total free: %d\n",
449 if (pagepool_end - pagepool_start > pagesz) {
450 caddr_t addr = (caddr_t)
451 (((long)pagepool_start + pagesz - 1) & ~(pagesz - 1));
452 if (munmap(addr, pagepool_end - addr) != 0)
453 rtld_fdprintf(STDERR_FILENO, "morepages: munmap %p",
457 offset = (long)pagepool_start - ((long)pagepool_start & ~(pagesz - 1));
459 if ((pagepool_start = mmap(0, n * pagesz,
460 PROT_READ|PROT_WRITE,
461 MAP_ANON|MAP_COPY, fd, 0)) == (caddr_t)-1) {
462 rtld_printf("Cannot map anonymous memory\n");
465 pagepool_end = pagepool_start + n * pagesz;
466 pagepool_start += offset;