xref: /freebsd/sys/kern/kern_malloc.c (revision 1aaed33edb24c98a09130cd66667d6a795b6b2a8)
1 /*-
2  * Copyright (c) 1987, 1991, 1993
3  *	The Regents of the University of California.
4  * Copyright (c) 2005-2009 Robert N. M. Watson
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  * 4. 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  *	@(#)kern_malloc.c	8.3 (Berkeley) 1/4/94
32  */
33 
34 /*
35  * Kernel malloc(9) implementation -- general purpose kernel memory allocator
36  * based on memory types.  Back end is implemented using the UMA(9) zone
37  * allocator.  A set of fixed-size buckets are used for smaller allocations,
38  * and a special UMA allocation interface is used for larger allocations.
39  * Callers declare memory types, and statistics are maintained independently
40  * for each memory type.  Statistics are maintained per-CPU for performance
41  * reasons.  See malloc(9) and comments in malloc.h for a detailed
42  * description.
43  */
44 
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47 
48 #include "opt_ddb.h"
49 #include "opt_kdtrace.h"
50 #include "opt_vm.h"
51 
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/kdb.h>
55 #include <sys/kernel.h>
56 #include <sys/lock.h>
57 #include <sys/malloc.h>
58 #include <sys/mbuf.h>
59 #include <sys/mutex.h>
60 #include <sys/vmmeter.h>
61 #include <sys/proc.h>
62 #include <sys/sbuf.h>
63 #include <sys/sysctl.h>
64 #include <sys/time.h>
65 
66 #include <vm/vm.h>
67 #include <vm/pmap.h>
68 #include <vm/vm_param.h>
69 #include <vm/vm_kern.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vm_map.h>
72 #include <vm/vm_page.h>
73 #include <vm/uma.h>
74 #include <vm/uma_int.h>
75 #include <vm/uma_dbg.h>
76 
77 #ifdef DEBUG_MEMGUARD
78 #include <vm/memguard.h>
79 #endif
80 #ifdef DEBUG_REDZONE
81 #include <vm/redzone.h>
82 #endif
83 
84 #if defined(INVARIANTS) && defined(__i386__)
85 #include <machine/cpu.h>
86 #endif
87 
88 #include <ddb/ddb.h>
89 
90 #ifdef KDTRACE_HOOKS
91 #include <sys/dtrace_bsd.h>
92 
93 dtrace_malloc_probe_func_t	dtrace_malloc_probe;
94 #endif
95 
96 /*
97  * When realloc() is called, if the new size is sufficiently smaller than
98  * the old size, realloc() will allocate a new, smaller block to avoid
99  * wasting memory. 'Sufficiently smaller' is defined as: newsize <=
100  * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'.
101  */
102 #ifndef REALLOC_FRACTION
103 #define	REALLOC_FRACTION	1	/* new block if <= half the size */
104 #endif
105 
106 /*
107  * Centrally define some common malloc types.
108  */
109 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches");
110 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
111 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers");
112 
113 MALLOC_DEFINE(M_IP6OPT, "ip6opt", "IPv6 options");
114 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
115 
116 static void kmeminit(void *);
117 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_FIRST, kmeminit, NULL);
118 
119 static MALLOC_DEFINE(M_FREE, "free", "should be on free list");
120 
121 static struct malloc_type *kmemstatistics;
122 static vm_offset_t kmembase;
123 static vm_offset_t kmemlimit;
124 static int kmemcount;
125 
126 #define KMEM_ZSHIFT	4
127 #define KMEM_ZBASE	16
128 #define KMEM_ZMASK	(KMEM_ZBASE - 1)
129 
130 #define KMEM_ZMAX	PAGE_SIZE
131 #define KMEM_ZSIZE	(KMEM_ZMAX >> KMEM_ZSHIFT)
132 static uint8_t kmemsize[KMEM_ZSIZE + 1];
133 
134 #ifndef MALLOC_DEBUG_MAXZONES
135 #define	MALLOC_DEBUG_MAXZONES	1
136 #endif
137 static int numzones = MALLOC_DEBUG_MAXZONES;
138 
139 /*
140  * Small malloc(9) memory allocations are allocated from a set of UMA buckets
141  * of various sizes.
142  *
143  * XXX: The comment here used to read "These won't be powers of two for
144  * long."  It's possible that a significant amount of wasted memory could be
145  * recovered by tuning the sizes of these buckets.
146  */
147 struct {
148 	int kz_size;
149 	char *kz_name;
150 	uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES];
151 } kmemzones[] = {
152 	{16, "16", },
153 	{32, "32", },
154 	{64, "64", },
155 	{128, "128", },
156 	{256, "256", },
157 	{512, "512", },
158 	{1024, "1024", },
159 	{2048, "2048", },
160 	{4096, "4096", },
161 #if PAGE_SIZE > 4096
162 	{8192, "8192", },
163 #if PAGE_SIZE > 8192
164 	{16384, "16384", },
165 #if PAGE_SIZE > 16384
166 	{32768, "32768", },
167 #if PAGE_SIZE > 32768
168 	{65536, "65536", },
169 #if PAGE_SIZE > 65536
170 #error	"Unsupported PAGE_SIZE"
171 #endif	/* 65536 */
172 #endif	/* 32768 */
173 #endif	/* 16384 */
174 #endif	/* 8192 */
175 #endif	/* 4096 */
176 	{0, NULL},
177 };
178 
179 /*
180  * Zone to allocate malloc type descriptions from.  For ABI reasons, memory
181  * types are described by a data structure passed by the declaring code, but
182  * the malloc(9) implementation has its own data structure describing the
183  * type and statistics.  This permits the malloc(9)-internal data structures
184  * to be modified without breaking binary-compiled kernel modules that
185  * declare malloc types.
186  */
187 static uma_zone_t mt_zone;
188 
189 u_long vm_kmem_size;
190 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0,
191     "Size of kernel memory");
192 
193 static u_long vm_kmem_size_min;
194 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0,
195     "Minimum size of kernel memory");
196 
197 static u_long vm_kmem_size_max;
198 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0,
199     "Maximum size of kernel memory");
200 
201 static u_int vm_kmem_size_scale;
202 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0,
203     "Scale factor for kernel memory size");
204 
205 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS);
206 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size,
207     CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
208     sysctl_kmem_map_size, "LU", "Current kmem_map allocation size");
209 
210 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS);
211 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free,
212     CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
213     sysctl_kmem_map_free, "LU", "Largest contiguous free range in kmem_map");
214 
215 /*
216  * The malloc_mtx protects the kmemstatistics linked list.
217  */
218 struct mtx malloc_mtx;
219 
220 #ifdef MALLOC_PROFILE
221 uint64_t krequests[KMEM_ZSIZE + 1];
222 
223 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS);
224 #endif
225 
226 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS);
227 
228 /*
229  * time_uptime of the last malloc(9) failure (induced or real).
230  */
231 static time_t t_malloc_fail;
232 
233 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1)
234 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0,
235     "Kernel malloc debugging options");
236 #endif
237 
238 /*
239  * malloc(9) fault injection -- cause malloc failures every (n) mallocs when
240  * the caller specifies M_NOWAIT.  If set to 0, no failures are caused.
241  */
242 #ifdef MALLOC_MAKE_FAILURES
243 static int malloc_failure_rate;
244 static int malloc_nowait_count;
245 static int malloc_failure_count;
246 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RW,
247     &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail");
248 TUNABLE_INT("debug.malloc.failure_rate", &malloc_failure_rate);
249 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD,
250     &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures");
251 #endif
252 
253 static int
254 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS)
255 {
256 	u_long size;
257 
258 	size = kmem_map->size;
259 	return (sysctl_handle_long(oidp, &size, 0, req));
260 }
261 
262 static int
263 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS)
264 {
265 	u_long size;
266 
267 	vm_map_lock_read(kmem_map);
268 	size = kmem_map->root != NULL ? kmem_map->root->max_free :
269 	    kmem_map->max_offset - kmem_map->min_offset;
270 	vm_map_unlock_read(kmem_map);
271 	return (sysctl_handle_long(oidp, &size, 0, req));
272 }
273 
274 /*
275  * malloc(9) uma zone separation -- sub-page buffer overruns in one
276  * malloc type will affect only a subset of other malloc types.
277  */
278 #if MALLOC_DEBUG_MAXZONES > 1
279 static void
280 tunable_set_numzones(void)
281 {
282 
283 	TUNABLE_INT_FETCH("debug.malloc.numzones",
284 	    &numzones);
285 
286 	/* Sanity check the number of malloc uma zones. */
287 	if (numzones <= 0)
288 		numzones = 1;
289 	if (numzones > MALLOC_DEBUG_MAXZONES)
290 		numzones = MALLOC_DEBUG_MAXZONES;
291 }
292 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL);
293 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN,
294     &numzones, 0, "Number of malloc uma subzones");
295 
296 /*
297  * Any number that changes regularly is an okay choice for the
298  * offset.  Build numbers are pretty good of you have them.
299  */
300 static u_int zone_offset = __FreeBSD_version;
301 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset);
302 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN,
303     &zone_offset, 0, "Separate malloc types by examining the "
304     "Nth character in the malloc type short description.");
305 
306 static u_int
307 mtp_get_subzone(const char *desc)
308 {
309 	size_t len;
310 	u_int val;
311 
312 	if (desc == NULL || (len = strlen(desc)) == 0)
313 		return (0);
314 	val = desc[zone_offset % len];
315 	return (val % numzones);
316 }
317 #elif MALLOC_DEBUG_MAXZONES == 0
318 #error "MALLOC_DEBUG_MAXZONES must be positive."
319 #else
320 static inline u_int
321 mtp_get_subzone(const char *desc)
322 {
323 
324 	return (0);
325 }
326 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
327 
328 int
329 malloc_last_fail(void)
330 {
331 
332 	return (time_uptime - t_malloc_fail);
333 }
334 
335 /*
336  * An allocation has succeeded -- update malloc type statistics for the
337  * amount of bucket size.  Occurs within a critical section so that the
338  * thread isn't preempted and doesn't migrate while updating per-PCU
339  * statistics.
340  */
341 static void
342 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size,
343     int zindx)
344 {
345 	struct malloc_type_internal *mtip;
346 	struct malloc_type_stats *mtsp;
347 
348 	critical_enter();
349 	mtip = mtp->ks_handle;
350 	mtsp = &mtip->mti_stats[curcpu];
351 	if (size > 0) {
352 		mtsp->mts_memalloced += size;
353 		mtsp->mts_numallocs++;
354 	}
355 	if (zindx != -1)
356 		mtsp->mts_size |= 1 << zindx;
357 
358 #ifdef KDTRACE_HOOKS
359 	if (dtrace_malloc_probe != NULL) {
360 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC];
361 		if (probe_id != 0)
362 			(dtrace_malloc_probe)(probe_id,
363 			    (uintptr_t) mtp, (uintptr_t) mtip,
364 			    (uintptr_t) mtsp, size, zindx);
365 	}
366 #endif
367 
368 	critical_exit();
369 }
370 
371 void
372 malloc_type_allocated(struct malloc_type *mtp, unsigned long size)
373 {
374 
375 	if (size > 0)
376 		malloc_type_zone_allocated(mtp, size, -1);
377 }
378 
379 /*
380  * A free operation has occurred -- update malloc type statistics for the
381  * amount of the bucket size.  Occurs within a critical section so that the
382  * thread isn't preempted and doesn't migrate while updating per-CPU
383  * statistics.
384  */
385 void
386 malloc_type_freed(struct malloc_type *mtp, unsigned long size)
387 {
388 	struct malloc_type_internal *mtip;
389 	struct malloc_type_stats *mtsp;
390 
391 	critical_enter();
392 	mtip = mtp->ks_handle;
393 	mtsp = &mtip->mti_stats[curcpu];
394 	mtsp->mts_memfreed += size;
395 	mtsp->mts_numfrees++;
396 
397 #ifdef KDTRACE_HOOKS
398 	if (dtrace_malloc_probe != NULL) {
399 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE];
400 		if (probe_id != 0)
401 			(dtrace_malloc_probe)(probe_id,
402 			    (uintptr_t) mtp, (uintptr_t) mtip,
403 			    (uintptr_t) mtsp, size, 0);
404 	}
405 #endif
406 
407 	critical_exit();
408 }
409 
410 /*
411  *	contigmalloc:
412  *
413  *	Allocate a block of physically contiguous memory.
414  *
415  *	If M_NOWAIT is set, this routine will not block and return NULL if
416  *	the allocation fails.
417  */
418 void *
419 contigmalloc(unsigned long size, struct malloc_type *type, int flags,
420     vm_paddr_t low, vm_paddr_t high, unsigned long alignment,
421     vm_paddr_t boundary)
422 {
423 	void *ret;
424 
425 	ret = (void *)kmem_alloc_contig(kernel_map, size, flags, low, high,
426 	    alignment, boundary, VM_MEMATTR_DEFAULT);
427 	if (ret != NULL)
428 		malloc_type_allocated(type, round_page(size));
429 	return (ret);
430 }
431 
432 /*
433  *	contigfree:
434  *
435  *	Free a block of memory allocated by contigmalloc.
436  *
437  *	This routine may not block.
438  */
439 void
440 contigfree(void *addr, unsigned long size, struct malloc_type *type)
441 {
442 
443 	kmem_free(kernel_map, (vm_offset_t)addr, size);
444 	malloc_type_freed(type, round_page(size));
445 }
446 
447 /*
448  *	malloc:
449  *
450  *	Allocate a block of memory.
451  *
452  *	If M_NOWAIT is set, this routine will not block and return NULL if
453  *	the allocation fails.
454  */
455 void *
456 malloc(unsigned long size, struct malloc_type *mtp, int flags)
457 {
458 	int indx;
459 	struct malloc_type_internal *mtip;
460 	caddr_t va;
461 	uma_zone_t zone;
462 #if defined(DIAGNOSTIC) || defined(DEBUG_REDZONE)
463 	unsigned long osize = size;
464 #endif
465 
466 #ifdef INVARIANTS
467 	KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic"));
468 	/*
469 	 * Check that exactly one of M_WAITOK or M_NOWAIT is specified.
470 	 */
471 	indx = flags & (M_WAITOK | M_NOWAIT);
472 	if (indx != M_NOWAIT && indx != M_WAITOK) {
473 		static	struct timeval lasterr;
474 		static	int curerr, once;
475 		if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) {
476 			printf("Bad malloc flags: %x\n", indx);
477 			kdb_backtrace();
478 			flags |= M_WAITOK;
479 			once++;
480 		}
481 	}
482 #endif
483 #ifdef MALLOC_MAKE_FAILURES
484 	if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
485 		atomic_add_int(&malloc_nowait_count, 1);
486 		if ((malloc_nowait_count % malloc_failure_rate) == 0) {
487 			atomic_add_int(&malloc_failure_count, 1);
488 			t_malloc_fail = time_uptime;
489 			return (NULL);
490 		}
491 	}
492 #endif
493 	if (flags & M_WAITOK)
494 		KASSERT(curthread->td_intr_nesting_level == 0,
495 		   ("malloc(M_WAITOK) in interrupt context"));
496 
497 #ifdef DEBUG_MEMGUARD
498 	if (memguard_cmp_mtp(mtp, size)) {
499 		va = memguard_alloc(size, flags);
500 		if (va != NULL)
501 			return (va);
502 		/* This is unfortunate but should not be fatal. */
503 	}
504 #endif
505 
506 #ifdef DEBUG_REDZONE
507 	size = redzone_size_ntor(size);
508 #endif
509 
510 	if (size <= KMEM_ZMAX) {
511 		mtip = mtp->ks_handle;
512 		if (size & KMEM_ZMASK)
513 			size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
514 		indx = kmemsize[size >> KMEM_ZSHIFT];
515 		KASSERT(mtip->mti_zone < numzones,
516 		    ("mti_zone %u out of range %d",
517 		    mtip->mti_zone, numzones));
518 		zone = kmemzones[indx].kz_zone[mtip->mti_zone];
519 #ifdef MALLOC_PROFILE
520 		krequests[size >> KMEM_ZSHIFT]++;
521 #endif
522 		va = uma_zalloc(zone, flags);
523 		if (va != NULL)
524 			size = zone->uz_size;
525 		malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
526 	} else {
527 		size = roundup(size, PAGE_SIZE);
528 		zone = NULL;
529 		va = uma_large_malloc(size, flags);
530 		malloc_type_allocated(mtp, va == NULL ? 0 : size);
531 	}
532 	if (flags & M_WAITOK)
533 		KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL"));
534 	else if (va == NULL)
535 		t_malloc_fail = time_uptime;
536 #ifdef DIAGNOSTIC
537 	if (va != NULL && !(flags & M_ZERO)) {
538 		memset(va, 0x70, osize);
539 	}
540 #endif
541 #ifdef DEBUG_REDZONE
542 	if (va != NULL)
543 		va = redzone_setup(va, osize);
544 #endif
545 	return ((void *) va);
546 }
547 
548 /*
549  *	free:
550  *
551  *	Free a block of memory allocated by malloc.
552  *
553  *	This routine may not block.
554  */
555 void
556 free(void *addr, struct malloc_type *mtp)
557 {
558 	uma_slab_t slab;
559 	u_long size;
560 
561 	KASSERT(mtp->ks_magic == M_MAGIC, ("free: bad malloc type magic"));
562 
563 	/* free(NULL, ...) does nothing */
564 	if (addr == NULL)
565 		return;
566 
567 #ifdef DEBUG_MEMGUARD
568 	if (is_memguard_addr(addr)) {
569 		memguard_free(addr);
570 		return;
571 	}
572 #endif
573 
574 #ifdef DEBUG_REDZONE
575 	redzone_check(addr);
576 	addr = redzone_addr_ntor(addr);
577 #endif
578 
579 	slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK));
580 
581 	if (slab == NULL)
582 		panic("free: address %p(%p) has not been allocated.\n",
583 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
584 
585 
586 	if (!(slab->us_flags & UMA_SLAB_MALLOC)) {
587 #ifdef INVARIANTS
588 		struct malloc_type **mtpp = addr;
589 #endif
590 		size = slab->us_keg->uk_size;
591 #ifdef INVARIANTS
592 		/*
593 		 * Cache a pointer to the malloc_type that most recently freed
594 		 * this memory here.  This way we know who is most likely to
595 		 * have stepped on it later.
596 		 *
597 		 * This code assumes that size is a multiple of 8 bytes for
598 		 * 64 bit machines
599 		 */
600 		mtpp = (struct malloc_type **)
601 		    ((unsigned long)mtpp & ~UMA_ALIGN_PTR);
602 		mtpp += (size - sizeof(struct malloc_type *)) /
603 		    sizeof(struct malloc_type *);
604 		*mtpp = mtp;
605 #endif
606 		uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab);
607 	} else {
608 		size = slab->us_size;
609 		uma_large_free(slab);
610 	}
611 	malloc_type_freed(mtp, size);
612 }
613 
614 /*
615  *	realloc: change the size of a memory block
616  */
617 void *
618 realloc(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
619 {
620 	uma_slab_t slab;
621 	unsigned long alloc;
622 	void *newaddr;
623 
624 	KASSERT(mtp->ks_magic == M_MAGIC,
625 	    ("realloc: bad malloc type magic"));
626 
627 	/* realloc(NULL, ...) is equivalent to malloc(...) */
628 	if (addr == NULL)
629 		return (malloc(size, mtp, flags));
630 
631 	/*
632 	 * XXX: Should report free of old memory and alloc of new memory to
633 	 * per-CPU stats.
634 	 */
635 
636 #ifdef DEBUG_MEMGUARD
637 	if (is_memguard_addr(addr))
638 		return (memguard_realloc(addr, size, mtp, flags));
639 #endif
640 
641 #ifdef DEBUG_REDZONE
642 	slab = NULL;
643 	alloc = redzone_get_size(addr);
644 #else
645 	slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK));
646 
647 	/* Sanity check */
648 	KASSERT(slab != NULL,
649 	    ("realloc: address %p out of range", (void *)addr));
650 
651 	/* Get the size of the original block */
652 	if (!(slab->us_flags & UMA_SLAB_MALLOC))
653 		alloc = slab->us_keg->uk_size;
654 	else
655 		alloc = slab->us_size;
656 
657 	/* Reuse the original block if appropriate */
658 	if (size <= alloc
659 	    && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE))
660 		return (addr);
661 #endif /* !DEBUG_REDZONE */
662 
663 	/* Allocate a new, bigger (or smaller) block */
664 	if ((newaddr = malloc(size, mtp, flags)) == NULL)
665 		return (NULL);
666 
667 	/* Copy over original contents */
668 	bcopy(addr, newaddr, min(size, alloc));
669 	free(addr, mtp);
670 	return (newaddr);
671 }
672 
673 /*
674  *	reallocf: same as realloc() but free memory on failure.
675  */
676 void *
677 reallocf(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
678 {
679 	void *mem;
680 
681 	if ((mem = realloc(addr, size, mtp, flags)) == NULL)
682 		free(addr, mtp);
683 	return (mem);
684 }
685 
686 /*
687  * Initialize the kernel memory allocator
688  */
689 /* ARGSUSED*/
690 static void
691 kmeminit(void *dummy)
692 {
693 	uint8_t indx;
694 	u_long mem_size, tmp;
695 	int i;
696 
697 	mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
698 
699 	/*
700 	 * Try to auto-tune the kernel memory size, so that it is
701 	 * more applicable for a wider range of machine sizes.  The
702 	 * VM_KMEM_SIZE_MAX is dependent on the maximum KVA space
703 	 * available.
704 	 *
705 	 * Note that the kmem_map is also used by the zone allocator,
706 	 * so make sure that there is enough space.
707 	 */
708 	vm_kmem_size = VM_KMEM_SIZE + nmbclusters * PAGE_SIZE;
709 	mem_size = cnt.v_page_count;
710 
711 #if defined(VM_KMEM_SIZE_SCALE)
712 	vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
713 #endif
714 	TUNABLE_INT_FETCH("vm.kmem_size_scale", &vm_kmem_size_scale);
715 	if (vm_kmem_size_scale > 0 &&
716 	    (mem_size / vm_kmem_size_scale) > (vm_kmem_size / PAGE_SIZE))
717 		vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE;
718 
719 #if defined(VM_KMEM_SIZE_MIN)
720 	vm_kmem_size_min = VM_KMEM_SIZE_MIN;
721 #endif
722 	TUNABLE_ULONG_FETCH("vm.kmem_size_min", &vm_kmem_size_min);
723 	if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min) {
724 		vm_kmem_size = vm_kmem_size_min;
725 	}
726 
727 #if defined(VM_KMEM_SIZE_MAX)
728 	vm_kmem_size_max = VM_KMEM_SIZE_MAX;
729 #endif
730 	TUNABLE_ULONG_FETCH("vm.kmem_size_max", &vm_kmem_size_max);
731 	if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
732 		vm_kmem_size = vm_kmem_size_max;
733 
734 	/* Allow final override from the kernel environment */
735 	TUNABLE_ULONG_FETCH("vm.kmem_size", &vm_kmem_size);
736 
737 	/*
738 	 * Limit kmem virtual size to twice the physical memory.
739 	 * This allows for kmem map sparseness, but limits the size
740 	 * to something sane.  Be careful to not overflow the 32bit
741 	 * ints while doing the check or the adjustment.
742 	 */
743 	if (vm_kmem_size / 2 / PAGE_SIZE > mem_size)
744 		vm_kmem_size = 2 * mem_size * PAGE_SIZE;
745 
746 #ifdef DEBUG_MEMGUARD
747 	tmp = memguard_fudge(vm_kmem_size, vm_kmem_size_max);
748 #else
749 	tmp = vm_kmem_size;
750 #endif
751 	kmem_map = kmem_suballoc(kernel_map, &kmembase, &kmemlimit,
752 	    tmp, TRUE);
753 	kmem_map->system_map = 1;
754 
755 #ifdef DEBUG_MEMGUARD
756 	/*
757 	 * Initialize MemGuard if support compiled in.  MemGuard is a
758 	 * replacement allocator used for detecting tamper-after-free
759 	 * scenarios as they occur.  It is only used for debugging.
760 	 */
761 	memguard_init(kmem_map);
762 #endif
763 
764 	uma_startup2();
765 
766 	mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal),
767 #ifdef INVARIANTS
768 	    mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
769 #else
770 	    NULL, NULL, NULL, NULL,
771 #endif
772 	    UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
773 	for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
774 		int size = kmemzones[indx].kz_size;
775 		char *name = kmemzones[indx].kz_name;
776 		int subzone;
777 
778 		for (subzone = 0; subzone < numzones; subzone++) {
779 			kmemzones[indx].kz_zone[subzone] =
780 			    uma_zcreate(name, size,
781 #ifdef INVARIANTS
782 			    mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
783 #else
784 			    NULL, NULL, NULL, NULL,
785 #endif
786 			    UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
787 		}
788 		for (;i <= size; i+= KMEM_ZBASE)
789 			kmemsize[i >> KMEM_ZSHIFT] = indx;
790 
791 	}
792 }
793 
794 void
795 malloc_init(void *data)
796 {
797 	struct malloc_type_internal *mtip;
798 	struct malloc_type *mtp;
799 
800 	KASSERT(cnt.v_page_count != 0, ("malloc_register before vm_init"));
801 
802 	mtp = data;
803 	if (mtp->ks_magic != M_MAGIC)
804 		panic("malloc_init: bad malloc type magic");
805 
806 	mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO);
807 	mtp->ks_handle = mtip;
808 	mtip->mti_zone = mtp_get_subzone(mtp->ks_shortdesc);
809 
810 	mtx_lock(&malloc_mtx);
811 	mtp->ks_next = kmemstatistics;
812 	kmemstatistics = mtp;
813 	kmemcount++;
814 	mtx_unlock(&malloc_mtx);
815 }
816 
817 void
818 malloc_uninit(void *data)
819 {
820 	struct malloc_type_internal *mtip;
821 	struct malloc_type_stats *mtsp;
822 	struct malloc_type *mtp, *temp;
823 	uma_slab_t slab;
824 	long temp_allocs, temp_bytes;
825 	int i;
826 
827 	mtp = data;
828 	KASSERT(mtp->ks_magic == M_MAGIC,
829 	    ("malloc_uninit: bad malloc type magic"));
830 	KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL"));
831 
832 	mtx_lock(&malloc_mtx);
833 	mtip = mtp->ks_handle;
834 	mtp->ks_handle = NULL;
835 	if (mtp != kmemstatistics) {
836 		for (temp = kmemstatistics; temp != NULL;
837 		    temp = temp->ks_next) {
838 			if (temp->ks_next == mtp) {
839 				temp->ks_next = mtp->ks_next;
840 				break;
841 			}
842 		}
843 		KASSERT(temp,
844 		    ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc));
845 	} else
846 		kmemstatistics = mtp->ks_next;
847 	kmemcount--;
848 	mtx_unlock(&malloc_mtx);
849 
850 	/*
851 	 * Look for memory leaks.
852 	 */
853 	temp_allocs = temp_bytes = 0;
854 	for (i = 0; i < MAXCPU; i++) {
855 		mtsp = &mtip->mti_stats[i];
856 		temp_allocs += mtsp->mts_numallocs;
857 		temp_allocs -= mtsp->mts_numfrees;
858 		temp_bytes += mtsp->mts_memalloced;
859 		temp_bytes -= mtsp->mts_memfreed;
860 	}
861 	if (temp_allocs > 0 || temp_bytes > 0) {
862 		printf("Warning: memory type %s leaked memory on destroy "
863 		    "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc,
864 		    temp_allocs, temp_bytes);
865 	}
866 
867 	slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK));
868 	uma_zfree_arg(mt_zone, mtip, slab);
869 }
870 
871 struct malloc_type *
872 malloc_desc2type(const char *desc)
873 {
874 	struct malloc_type *mtp;
875 
876 	mtx_assert(&malloc_mtx, MA_OWNED);
877 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
878 		if (strcmp(mtp->ks_shortdesc, desc) == 0)
879 			return (mtp);
880 	}
881 	return (NULL);
882 }
883 
884 static int
885 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
886 {
887 	struct malloc_type_stream_header mtsh;
888 	struct malloc_type_internal *mtip;
889 	struct malloc_type_header mth;
890 	struct malloc_type *mtp;
891 	int error, i;
892 	struct sbuf sbuf;
893 
894 	error = sysctl_wire_old_buffer(req, 0);
895 	if (error != 0)
896 		return (error);
897 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
898 	mtx_lock(&malloc_mtx);
899 
900 	/*
901 	 * Insert stream header.
902 	 */
903 	bzero(&mtsh, sizeof(mtsh));
904 	mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
905 	mtsh.mtsh_maxcpus = MAXCPU;
906 	mtsh.mtsh_count = kmemcount;
907 	(void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh));
908 
909 	/*
910 	 * Insert alternating sequence of type headers and type statistics.
911 	 */
912 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
913 		mtip = (struct malloc_type_internal *)mtp->ks_handle;
914 
915 		/*
916 		 * Insert type header.
917 		 */
918 		bzero(&mth, sizeof(mth));
919 		strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
920 		(void)sbuf_bcat(&sbuf, &mth, sizeof(mth));
921 
922 		/*
923 		 * Insert type statistics for each CPU.
924 		 */
925 		for (i = 0; i < MAXCPU; i++) {
926 			(void)sbuf_bcat(&sbuf, &mtip->mti_stats[i],
927 			    sizeof(mtip->mti_stats[i]));
928 		}
929 	}
930 	mtx_unlock(&malloc_mtx);
931 	error = sbuf_finish(&sbuf);
932 	sbuf_delete(&sbuf);
933 	return (error);
934 }
935 
936 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT,
937     0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats",
938     "Return malloc types");
939 
940 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
941     "Count of kernel malloc types");
942 
943 void
944 malloc_type_list(malloc_type_list_func_t *func, void *arg)
945 {
946 	struct malloc_type *mtp, **bufmtp;
947 	int count, i;
948 	size_t buflen;
949 
950 	mtx_lock(&malloc_mtx);
951 restart:
952 	mtx_assert(&malloc_mtx, MA_OWNED);
953 	count = kmemcount;
954 	mtx_unlock(&malloc_mtx);
955 
956 	buflen = sizeof(struct malloc_type *) * count;
957 	bufmtp = malloc(buflen, M_TEMP, M_WAITOK);
958 
959 	mtx_lock(&malloc_mtx);
960 
961 	if (count < kmemcount) {
962 		free(bufmtp, M_TEMP);
963 		goto restart;
964 	}
965 
966 	for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++)
967 		bufmtp[i] = mtp;
968 
969 	mtx_unlock(&malloc_mtx);
970 
971 	for (i = 0; i < count; i++)
972 		(func)(bufmtp[i], arg);
973 
974 	free(bufmtp, M_TEMP);
975 }
976 
977 #ifdef DDB
978 DB_SHOW_COMMAND(malloc, db_show_malloc)
979 {
980 	struct malloc_type_internal *mtip;
981 	struct malloc_type *mtp;
982 	uint64_t allocs, frees;
983 	uint64_t alloced, freed;
984 	int i;
985 
986 	db_printf("%18s %12s  %12s %12s\n", "Type", "InUse", "MemUse",
987 	    "Requests");
988 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
989 		mtip = (struct malloc_type_internal *)mtp->ks_handle;
990 		allocs = 0;
991 		frees = 0;
992 		alloced = 0;
993 		freed = 0;
994 		for (i = 0; i < MAXCPU; i++) {
995 			allocs += mtip->mti_stats[i].mts_numallocs;
996 			frees += mtip->mti_stats[i].mts_numfrees;
997 			alloced += mtip->mti_stats[i].mts_memalloced;
998 			freed += mtip->mti_stats[i].mts_memfreed;
999 		}
1000 		db_printf("%18s %12ju %12juK %12ju\n",
1001 		    mtp->ks_shortdesc, allocs - frees,
1002 		    (alloced - freed + 1023) / 1024, allocs);
1003 	}
1004 }
1005 
1006 #if MALLOC_DEBUG_MAXZONES > 1
1007 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches)
1008 {
1009 	struct malloc_type_internal *mtip;
1010 	struct malloc_type *mtp;
1011 	u_int subzone;
1012 
1013 	if (!have_addr) {
1014 		db_printf("Usage: show multizone_matches <malloc type/addr>\n");
1015 		return;
1016 	}
1017 	mtp = (void *)addr;
1018 	if (mtp->ks_magic != M_MAGIC) {
1019 		db_printf("Magic %lx does not match expected %x\n",
1020 		    mtp->ks_magic, M_MAGIC);
1021 		return;
1022 	}
1023 
1024 	mtip = mtp->ks_handle;
1025 	subzone = mtip->mti_zone;
1026 
1027 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1028 		mtip = mtp->ks_handle;
1029 		if (mtip->mti_zone != subzone)
1030 			continue;
1031 		db_printf("%s\n", mtp->ks_shortdesc);
1032 	}
1033 }
1034 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
1035 #endif /* DDB */
1036 
1037 #ifdef MALLOC_PROFILE
1038 
1039 static int
1040 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS)
1041 {
1042 	struct sbuf sbuf;
1043 	uint64_t count;
1044 	uint64_t waste;
1045 	uint64_t mem;
1046 	int error;
1047 	int rsize;
1048 	int size;
1049 	int i;
1050 
1051 	waste = 0;
1052 	mem = 0;
1053 
1054 	error = sysctl_wire_old_buffer(req, 0);
1055 	if (error != 0)
1056 		return (error);
1057 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
1058 	sbuf_printf(&sbuf,
1059 	    "\n  Size                    Requests  Real Size\n");
1060 	for (i = 0; i < KMEM_ZSIZE; i++) {
1061 		size = i << KMEM_ZSHIFT;
1062 		rsize = kmemzones[kmemsize[i]].kz_size;
1063 		count = (long long unsigned)krequests[i];
1064 
1065 		sbuf_printf(&sbuf, "%6d%28llu%11d\n", size,
1066 		    (unsigned long long)count, rsize);
1067 
1068 		if ((rsize * count) > (size * count))
1069 			waste += (rsize * count) - (size * count);
1070 		mem += (rsize * count);
1071 	}
1072 	sbuf_printf(&sbuf,
1073 	    "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n",
1074 	    (unsigned long long)mem, (unsigned long long)waste);
1075 	error = sbuf_finish(&sbuf);
1076 	sbuf_delete(&sbuf);
1077 	return (error);
1078 }
1079 
1080 SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD,
1081     NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling");
1082 #endif /* MALLOC_PROFILE */
1083