xref: /freebsd/sys/kern/kern_malloc.c (revision c6ec7d31830ab1c80edae95ad5e4b9dba10c47ac)
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 static vm_offset_t vm_min_kernel_address = VM_MIN_KERNEL_ADDRESS;
190 SYSCTL_ULONG(_vm, OID_AUTO, min_kernel_address, CTLFLAG_RD,
191     &vm_min_kernel_address, 0, "Min kernel address");
192 
193 #ifndef __sparc64__
194 static vm_offset_t vm_max_kernel_address = VM_MAX_KERNEL_ADDRESS;
195 #endif
196 SYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLAG_RD,
197     &vm_max_kernel_address, 0, "Max kernel address");
198 
199 u_long vm_kmem_size;
200 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0,
201     "Size of kernel memory");
202 
203 static u_long vm_kmem_size_min;
204 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0,
205     "Minimum size of kernel memory");
206 
207 static u_long vm_kmem_size_max;
208 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0,
209     "Maximum size of kernel memory");
210 
211 static u_int vm_kmem_size_scale;
212 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0,
213     "Scale factor for kernel memory size");
214 
215 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS);
216 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size,
217     CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
218     sysctl_kmem_map_size, "LU", "Current kmem_map allocation size");
219 
220 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS);
221 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free,
222     CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
223     sysctl_kmem_map_free, "LU", "Largest contiguous free range in kmem_map");
224 
225 /*
226  * The malloc_mtx protects the kmemstatistics linked list.
227  */
228 struct mtx malloc_mtx;
229 
230 #ifdef MALLOC_PROFILE
231 uint64_t krequests[KMEM_ZSIZE + 1];
232 
233 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS);
234 #endif
235 
236 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS);
237 
238 /*
239  * time_uptime of the last malloc(9) failure (induced or real).
240  */
241 static time_t t_malloc_fail;
242 
243 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1)
244 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0,
245     "Kernel malloc debugging options");
246 #endif
247 
248 /*
249  * malloc(9) fault injection -- cause malloc failures every (n) mallocs when
250  * the caller specifies M_NOWAIT.  If set to 0, no failures are caused.
251  */
252 #ifdef MALLOC_MAKE_FAILURES
253 static int malloc_failure_rate;
254 static int malloc_nowait_count;
255 static int malloc_failure_count;
256 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RW,
257     &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail");
258 TUNABLE_INT("debug.malloc.failure_rate", &malloc_failure_rate);
259 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD,
260     &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures");
261 #endif
262 
263 static int
264 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS)
265 {
266 	u_long size;
267 
268 	size = kmem_map->size;
269 	return (sysctl_handle_long(oidp, &size, 0, req));
270 }
271 
272 static int
273 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS)
274 {
275 	u_long size;
276 
277 	vm_map_lock_read(kmem_map);
278 	size = kmem_map->root != NULL ? kmem_map->root->max_free :
279 	    kmem_map->max_offset - kmem_map->min_offset;
280 	vm_map_unlock_read(kmem_map);
281 	return (sysctl_handle_long(oidp, &size, 0, req));
282 }
283 
284 /*
285  * malloc(9) uma zone separation -- sub-page buffer overruns in one
286  * malloc type will affect only a subset of other malloc types.
287  */
288 #if MALLOC_DEBUG_MAXZONES > 1
289 static void
290 tunable_set_numzones(void)
291 {
292 
293 	TUNABLE_INT_FETCH("debug.malloc.numzones",
294 	    &numzones);
295 
296 	/* Sanity check the number of malloc uma zones. */
297 	if (numzones <= 0)
298 		numzones = 1;
299 	if (numzones > MALLOC_DEBUG_MAXZONES)
300 		numzones = MALLOC_DEBUG_MAXZONES;
301 }
302 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL);
303 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN,
304     &numzones, 0, "Number of malloc uma subzones");
305 
306 /*
307  * Any number that changes regularly is an okay choice for the
308  * offset.  Build numbers are pretty good of you have them.
309  */
310 static u_int zone_offset = __FreeBSD_version;
311 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset);
312 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN,
313     &zone_offset, 0, "Separate malloc types by examining the "
314     "Nth character in the malloc type short description.");
315 
316 static u_int
317 mtp_get_subzone(const char *desc)
318 {
319 	size_t len;
320 	u_int val;
321 
322 	if (desc == NULL || (len = strlen(desc)) == 0)
323 		return (0);
324 	val = desc[zone_offset % len];
325 	return (val % numzones);
326 }
327 #elif MALLOC_DEBUG_MAXZONES == 0
328 #error "MALLOC_DEBUG_MAXZONES must be positive."
329 #else
330 static inline u_int
331 mtp_get_subzone(const char *desc)
332 {
333 
334 	return (0);
335 }
336 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
337 
338 int
339 malloc_last_fail(void)
340 {
341 
342 	return (time_uptime - t_malloc_fail);
343 }
344 
345 /*
346  * An allocation has succeeded -- update malloc type statistics for the
347  * amount of bucket size.  Occurs within a critical section so that the
348  * thread isn't preempted and doesn't migrate while updating per-PCU
349  * statistics.
350  */
351 static void
352 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size,
353     int zindx)
354 {
355 	struct malloc_type_internal *mtip;
356 	struct malloc_type_stats *mtsp;
357 
358 	critical_enter();
359 	mtip = mtp->ks_handle;
360 	mtsp = &mtip->mti_stats[curcpu];
361 	if (size > 0) {
362 		mtsp->mts_memalloced += size;
363 		mtsp->mts_numallocs++;
364 	}
365 	if (zindx != -1)
366 		mtsp->mts_size |= 1 << zindx;
367 
368 #ifdef KDTRACE_HOOKS
369 	if (dtrace_malloc_probe != NULL) {
370 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC];
371 		if (probe_id != 0)
372 			(dtrace_malloc_probe)(probe_id,
373 			    (uintptr_t) mtp, (uintptr_t) mtip,
374 			    (uintptr_t) mtsp, size, zindx);
375 	}
376 #endif
377 
378 	critical_exit();
379 }
380 
381 void
382 malloc_type_allocated(struct malloc_type *mtp, unsigned long size)
383 {
384 
385 	if (size > 0)
386 		malloc_type_zone_allocated(mtp, size, -1);
387 }
388 
389 /*
390  * A free operation has occurred -- update malloc type statistics for the
391  * amount of the bucket size.  Occurs within a critical section so that the
392  * thread isn't preempted and doesn't migrate while updating per-CPU
393  * statistics.
394  */
395 void
396 malloc_type_freed(struct malloc_type *mtp, unsigned long size)
397 {
398 	struct malloc_type_internal *mtip;
399 	struct malloc_type_stats *mtsp;
400 
401 	critical_enter();
402 	mtip = mtp->ks_handle;
403 	mtsp = &mtip->mti_stats[curcpu];
404 	mtsp->mts_memfreed += size;
405 	mtsp->mts_numfrees++;
406 
407 #ifdef KDTRACE_HOOKS
408 	if (dtrace_malloc_probe != NULL) {
409 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE];
410 		if (probe_id != 0)
411 			(dtrace_malloc_probe)(probe_id,
412 			    (uintptr_t) mtp, (uintptr_t) mtip,
413 			    (uintptr_t) mtsp, size, 0);
414 	}
415 #endif
416 
417 	critical_exit();
418 }
419 
420 /*
421  *	contigmalloc:
422  *
423  *	Allocate a block of physically contiguous memory.
424  *
425  *	If M_NOWAIT is set, this routine will not block and return NULL if
426  *	the allocation fails.
427  */
428 void *
429 contigmalloc(unsigned long size, struct malloc_type *type, int flags,
430     vm_paddr_t low, vm_paddr_t high, unsigned long alignment,
431     vm_paddr_t boundary)
432 {
433 	void *ret;
434 
435 	ret = (void *)kmem_alloc_contig(kernel_map, size, flags, low, high,
436 	    alignment, boundary, VM_MEMATTR_DEFAULT);
437 	if (ret != NULL)
438 		malloc_type_allocated(type, round_page(size));
439 	return (ret);
440 }
441 
442 /*
443  *	contigfree:
444  *
445  *	Free a block of memory allocated by contigmalloc.
446  *
447  *	This routine may not block.
448  */
449 void
450 contigfree(void *addr, unsigned long size, struct malloc_type *type)
451 {
452 
453 	kmem_free(kernel_map, (vm_offset_t)addr, size);
454 	malloc_type_freed(type, round_page(size));
455 }
456 
457 /*
458  *	malloc:
459  *
460  *	Allocate a block of memory.
461  *
462  *	If M_NOWAIT is set, this routine will not block and return NULL if
463  *	the allocation fails.
464  */
465 void *
466 malloc(unsigned long size, struct malloc_type *mtp, int flags)
467 {
468 	int indx;
469 	struct malloc_type_internal *mtip;
470 	caddr_t va;
471 	uma_zone_t zone;
472 #if defined(DIAGNOSTIC) || defined(DEBUG_REDZONE)
473 	unsigned long osize = size;
474 #endif
475 
476 #ifdef INVARIANTS
477 	KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic"));
478 	/*
479 	 * Check that exactly one of M_WAITOK or M_NOWAIT is specified.
480 	 */
481 	indx = flags & (M_WAITOK | M_NOWAIT);
482 	if (indx != M_NOWAIT && indx != M_WAITOK) {
483 		static	struct timeval lasterr;
484 		static	int curerr, once;
485 		if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) {
486 			printf("Bad malloc flags: %x\n", indx);
487 			kdb_backtrace();
488 			flags |= M_WAITOK;
489 			once++;
490 		}
491 	}
492 #endif
493 #ifdef MALLOC_MAKE_FAILURES
494 	if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
495 		atomic_add_int(&malloc_nowait_count, 1);
496 		if ((malloc_nowait_count % malloc_failure_rate) == 0) {
497 			atomic_add_int(&malloc_failure_count, 1);
498 			t_malloc_fail = time_uptime;
499 			return (NULL);
500 		}
501 	}
502 #endif
503 	if (flags & M_WAITOK)
504 		KASSERT(curthread->td_intr_nesting_level == 0,
505 		   ("malloc(M_WAITOK) in interrupt context"));
506 
507 #ifdef DEBUG_MEMGUARD
508 	if (memguard_cmp_mtp(mtp, size)) {
509 		va = memguard_alloc(size, flags);
510 		if (va != NULL)
511 			return (va);
512 		/* This is unfortunate but should not be fatal. */
513 	}
514 #endif
515 
516 #ifdef DEBUG_REDZONE
517 	size = redzone_size_ntor(size);
518 #endif
519 
520 	if (size <= KMEM_ZMAX) {
521 		mtip = mtp->ks_handle;
522 		if (size & KMEM_ZMASK)
523 			size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
524 		indx = kmemsize[size >> KMEM_ZSHIFT];
525 		KASSERT(mtip->mti_zone < numzones,
526 		    ("mti_zone %u out of range %d",
527 		    mtip->mti_zone, numzones));
528 		zone = kmemzones[indx].kz_zone[mtip->mti_zone];
529 #ifdef MALLOC_PROFILE
530 		krequests[size >> KMEM_ZSHIFT]++;
531 #endif
532 		va = uma_zalloc(zone, flags);
533 		if (va != NULL)
534 			size = zone->uz_size;
535 		malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
536 	} else {
537 		size = roundup(size, PAGE_SIZE);
538 		zone = NULL;
539 		va = uma_large_malloc(size, flags);
540 		malloc_type_allocated(mtp, va == NULL ? 0 : size);
541 	}
542 	if (flags & M_WAITOK)
543 		KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL"));
544 	else if (va == NULL)
545 		t_malloc_fail = time_uptime;
546 #ifdef DIAGNOSTIC
547 	if (va != NULL && !(flags & M_ZERO)) {
548 		memset(va, 0x70, osize);
549 	}
550 #endif
551 #ifdef DEBUG_REDZONE
552 	if (va != NULL)
553 		va = redzone_setup(va, osize);
554 #endif
555 	return ((void *) va);
556 }
557 
558 /*
559  *	free:
560  *
561  *	Free a block of memory allocated by malloc.
562  *
563  *	This routine may not block.
564  */
565 void
566 free(void *addr, struct malloc_type *mtp)
567 {
568 	uma_slab_t slab;
569 	u_long size;
570 
571 	KASSERT(mtp->ks_magic == M_MAGIC, ("free: bad malloc type magic"));
572 
573 	/* free(NULL, ...) does nothing */
574 	if (addr == NULL)
575 		return;
576 
577 #ifdef DEBUG_MEMGUARD
578 	if (is_memguard_addr(addr)) {
579 		memguard_free(addr);
580 		return;
581 	}
582 #endif
583 
584 #ifdef DEBUG_REDZONE
585 	redzone_check(addr);
586 	addr = redzone_addr_ntor(addr);
587 #endif
588 
589 	slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK));
590 
591 	if (slab == NULL)
592 		panic("free: address %p(%p) has not been allocated.\n",
593 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
594 
595 
596 	if (!(slab->us_flags & UMA_SLAB_MALLOC)) {
597 #ifdef INVARIANTS
598 		struct malloc_type **mtpp = addr;
599 #endif
600 		size = slab->us_keg->uk_size;
601 #ifdef INVARIANTS
602 		/*
603 		 * Cache a pointer to the malloc_type that most recently freed
604 		 * this memory here.  This way we know who is most likely to
605 		 * have stepped on it later.
606 		 *
607 		 * This code assumes that size is a multiple of 8 bytes for
608 		 * 64 bit machines
609 		 */
610 		mtpp = (struct malloc_type **)
611 		    ((unsigned long)mtpp & ~UMA_ALIGN_PTR);
612 		mtpp += (size - sizeof(struct malloc_type *)) /
613 		    sizeof(struct malloc_type *);
614 		*mtpp = mtp;
615 #endif
616 		uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab);
617 	} else {
618 		size = slab->us_size;
619 		uma_large_free(slab);
620 	}
621 	malloc_type_freed(mtp, size);
622 }
623 
624 /*
625  *	realloc: change the size of a memory block
626  */
627 void *
628 realloc(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
629 {
630 	uma_slab_t slab;
631 	unsigned long alloc;
632 	void *newaddr;
633 
634 	KASSERT(mtp->ks_magic == M_MAGIC,
635 	    ("realloc: bad malloc type magic"));
636 
637 	/* realloc(NULL, ...) is equivalent to malloc(...) */
638 	if (addr == NULL)
639 		return (malloc(size, mtp, flags));
640 
641 	/*
642 	 * XXX: Should report free of old memory and alloc of new memory to
643 	 * per-CPU stats.
644 	 */
645 
646 #ifdef DEBUG_MEMGUARD
647 	if (is_memguard_addr(addr))
648 		return (memguard_realloc(addr, size, mtp, flags));
649 #endif
650 
651 #ifdef DEBUG_REDZONE
652 	slab = NULL;
653 	alloc = redzone_get_size(addr);
654 #else
655 	slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK));
656 
657 	/* Sanity check */
658 	KASSERT(slab != NULL,
659 	    ("realloc: address %p out of range", (void *)addr));
660 
661 	/* Get the size of the original block */
662 	if (!(slab->us_flags & UMA_SLAB_MALLOC))
663 		alloc = slab->us_keg->uk_size;
664 	else
665 		alloc = slab->us_size;
666 
667 	/* Reuse the original block if appropriate */
668 	if (size <= alloc
669 	    && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE))
670 		return (addr);
671 #endif /* !DEBUG_REDZONE */
672 
673 	/* Allocate a new, bigger (or smaller) block */
674 	if ((newaddr = malloc(size, mtp, flags)) == NULL)
675 		return (NULL);
676 
677 	/* Copy over original contents */
678 	bcopy(addr, newaddr, min(size, alloc));
679 	free(addr, mtp);
680 	return (newaddr);
681 }
682 
683 /*
684  *	reallocf: same as realloc() but free memory on failure.
685  */
686 void *
687 reallocf(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
688 {
689 	void *mem;
690 
691 	if ((mem = realloc(addr, size, mtp, flags)) == NULL)
692 		free(addr, mtp);
693 	return (mem);
694 }
695 
696 /*
697  * Initialize the kernel memory allocator
698  */
699 /* ARGSUSED*/
700 static void
701 kmeminit(void *dummy)
702 {
703 	uint8_t indx;
704 	u_long mem_size, tmp;
705 	int i;
706 
707 	mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
708 
709 	/*
710 	 * Try to auto-tune the kernel memory size, so that it is
711 	 * more applicable for a wider range of machine sizes.  The
712 	 * VM_KMEM_SIZE_MAX is dependent on the maximum KVA space
713 	 * available.
714 	 *
715 	 * Note that the kmem_map is also used by the zone allocator,
716 	 * so make sure that there is enough space.
717 	 */
718 	vm_kmem_size = VM_KMEM_SIZE + nmbclusters * PAGE_SIZE;
719 	mem_size = cnt.v_page_count;
720 
721 #if defined(VM_KMEM_SIZE_SCALE)
722 	vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
723 #endif
724 	TUNABLE_INT_FETCH("vm.kmem_size_scale", &vm_kmem_size_scale);
725 	if (vm_kmem_size_scale > 0 &&
726 	    (mem_size / vm_kmem_size_scale) > (vm_kmem_size / PAGE_SIZE))
727 		vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE;
728 
729 #if defined(VM_KMEM_SIZE_MIN)
730 	vm_kmem_size_min = VM_KMEM_SIZE_MIN;
731 #endif
732 	TUNABLE_ULONG_FETCH("vm.kmem_size_min", &vm_kmem_size_min);
733 	if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min) {
734 		vm_kmem_size = vm_kmem_size_min;
735 	}
736 
737 #if defined(VM_KMEM_SIZE_MAX)
738 	vm_kmem_size_max = VM_KMEM_SIZE_MAX;
739 #endif
740 	TUNABLE_ULONG_FETCH("vm.kmem_size_max", &vm_kmem_size_max);
741 	if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
742 		vm_kmem_size = vm_kmem_size_max;
743 
744 	/* Allow final override from the kernel environment */
745 	TUNABLE_ULONG_FETCH("vm.kmem_size", &vm_kmem_size);
746 
747 	/*
748 	 * Limit kmem virtual size to twice the physical memory.
749 	 * This allows for kmem map sparseness, but limits the size
750 	 * to something sane.  Be careful to not overflow the 32bit
751 	 * ints while doing the check or the adjustment.
752 	 */
753 	if (vm_kmem_size / 2 / PAGE_SIZE > mem_size)
754 		vm_kmem_size = 2 * mem_size * PAGE_SIZE;
755 
756 #ifdef DEBUG_MEMGUARD
757 	tmp = memguard_fudge(vm_kmem_size, kernel_map);
758 #else
759 	tmp = vm_kmem_size;
760 #endif
761 	kmem_map = kmem_suballoc(kernel_map, &kmembase, &kmemlimit,
762 	    tmp, TRUE);
763 	kmem_map->system_map = 1;
764 
765 #ifdef DEBUG_MEMGUARD
766 	/*
767 	 * Initialize MemGuard if support compiled in.  MemGuard is a
768 	 * replacement allocator used for detecting tamper-after-free
769 	 * scenarios as they occur.  It is only used for debugging.
770 	 */
771 	memguard_init(kmem_map);
772 #endif
773 
774 	uma_startup2();
775 
776 	mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal),
777 #ifdef INVARIANTS
778 	    mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
779 #else
780 	    NULL, NULL, NULL, NULL,
781 #endif
782 	    UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
783 	for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
784 		int size = kmemzones[indx].kz_size;
785 		char *name = kmemzones[indx].kz_name;
786 		int subzone;
787 
788 		for (subzone = 0; subzone < numzones; subzone++) {
789 			kmemzones[indx].kz_zone[subzone] =
790 			    uma_zcreate(name, size,
791 #ifdef INVARIANTS
792 			    mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
793 #else
794 			    NULL, NULL, NULL, NULL,
795 #endif
796 			    UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
797 		}
798 		for (;i <= size; i+= KMEM_ZBASE)
799 			kmemsize[i >> KMEM_ZSHIFT] = indx;
800 
801 	}
802 }
803 
804 void
805 malloc_init(void *data)
806 {
807 	struct malloc_type_internal *mtip;
808 	struct malloc_type *mtp;
809 
810 	KASSERT(cnt.v_page_count != 0, ("malloc_register before vm_init"));
811 
812 	mtp = data;
813 	if (mtp->ks_magic != M_MAGIC)
814 		panic("malloc_init: bad malloc type magic");
815 
816 	mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO);
817 	mtp->ks_handle = mtip;
818 	mtip->mti_zone = mtp_get_subzone(mtp->ks_shortdesc);
819 
820 	mtx_lock(&malloc_mtx);
821 	mtp->ks_next = kmemstatistics;
822 	kmemstatistics = mtp;
823 	kmemcount++;
824 	mtx_unlock(&malloc_mtx);
825 }
826 
827 void
828 malloc_uninit(void *data)
829 {
830 	struct malloc_type_internal *mtip;
831 	struct malloc_type_stats *mtsp;
832 	struct malloc_type *mtp, *temp;
833 	uma_slab_t slab;
834 	long temp_allocs, temp_bytes;
835 	int i;
836 
837 	mtp = data;
838 	KASSERT(mtp->ks_magic == M_MAGIC,
839 	    ("malloc_uninit: bad malloc type magic"));
840 	KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL"));
841 
842 	mtx_lock(&malloc_mtx);
843 	mtip = mtp->ks_handle;
844 	mtp->ks_handle = NULL;
845 	if (mtp != kmemstatistics) {
846 		for (temp = kmemstatistics; temp != NULL;
847 		    temp = temp->ks_next) {
848 			if (temp->ks_next == mtp) {
849 				temp->ks_next = mtp->ks_next;
850 				break;
851 			}
852 		}
853 		KASSERT(temp,
854 		    ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc));
855 	} else
856 		kmemstatistics = mtp->ks_next;
857 	kmemcount--;
858 	mtx_unlock(&malloc_mtx);
859 
860 	/*
861 	 * Look for memory leaks.
862 	 */
863 	temp_allocs = temp_bytes = 0;
864 	for (i = 0; i < MAXCPU; i++) {
865 		mtsp = &mtip->mti_stats[i];
866 		temp_allocs += mtsp->mts_numallocs;
867 		temp_allocs -= mtsp->mts_numfrees;
868 		temp_bytes += mtsp->mts_memalloced;
869 		temp_bytes -= mtsp->mts_memfreed;
870 	}
871 	if (temp_allocs > 0 || temp_bytes > 0) {
872 		printf("Warning: memory type %s leaked memory on destroy "
873 		    "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc,
874 		    temp_allocs, temp_bytes);
875 	}
876 
877 	slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK));
878 	uma_zfree_arg(mt_zone, mtip, slab);
879 }
880 
881 struct malloc_type *
882 malloc_desc2type(const char *desc)
883 {
884 	struct malloc_type *mtp;
885 
886 	mtx_assert(&malloc_mtx, MA_OWNED);
887 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
888 		if (strcmp(mtp->ks_shortdesc, desc) == 0)
889 			return (mtp);
890 	}
891 	return (NULL);
892 }
893 
894 static int
895 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
896 {
897 	struct malloc_type_stream_header mtsh;
898 	struct malloc_type_internal *mtip;
899 	struct malloc_type_header mth;
900 	struct malloc_type *mtp;
901 	int error, i;
902 	struct sbuf sbuf;
903 
904 	error = sysctl_wire_old_buffer(req, 0);
905 	if (error != 0)
906 		return (error);
907 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
908 	mtx_lock(&malloc_mtx);
909 
910 	/*
911 	 * Insert stream header.
912 	 */
913 	bzero(&mtsh, sizeof(mtsh));
914 	mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
915 	mtsh.mtsh_maxcpus = MAXCPU;
916 	mtsh.mtsh_count = kmemcount;
917 	(void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh));
918 
919 	/*
920 	 * Insert alternating sequence of type headers and type statistics.
921 	 */
922 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
923 		mtip = (struct malloc_type_internal *)mtp->ks_handle;
924 
925 		/*
926 		 * Insert type header.
927 		 */
928 		bzero(&mth, sizeof(mth));
929 		strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
930 		(void)sbuf_bcat(&sbuf, &mth, sizeof(mth));
931 
932 		/*
933 		 * Insert type statistics for each CPU.
934 		 */
935 		for (i = 0; i < MAXCPU; i++) {
936 			(void)sbuf_bcat(&sbuf, &mtip->mti_stats[i],
937 			    sizeof(mtip->mti_stats[i]));
938 		}
939 	}
940 	mtx_unlock(&malloc_mtx);
941 	error = sbuf_finish(&sbuf);
942 	sbuf_delete(&sbuf);
943 	return (error);
944 }
945 
946 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT,
947     0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats",
948     "Return malloc types");
949 
950 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
951     "Count of kernel malloc types");
952 
953 void
954 malloc_type_list(malloc_type_list_func_t *func, void *arg)
955 {
956 	struct malloc_type *mtp, **bufmtp;
957 	int count, i;
958 	size_t buflen;
959 
960 	mtx_lock(&malloc_mtx);
961 restart:
962 	mtx_assert(&malloc_mtx, MA_OWNED);
963 	count = kmemcount;
964 	mtx_unlock(&malloc_mtx);
965 
966 	buflen = sizeof(struct malloc_type *) * count;
967 	bufmtp = malloc(buflen, M_TEMP, M_WAITOK);
968 
969 	mtx_lock(&malloc_mtx);
970 
971 	if (count < kmemcount) {
972 		free(bufmtp, M_TEMP);
973 		goto restart;
974 	}
975 
976 	for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++)
977 		bufmtp[i] = mtp;
978 
979 	mtx_unlock(&malloc_mtx);
980 
981 	for (i = 0; i < count; i++)
982 		(func)(bufmtp[i], arg);
983 
984 	free(bufmtp, M_TEMP);
985 }
986 
987 #ifdef DDB
988 DB_SHOW_COMMAND(malloc, db_show_malloc)
989 {
990 	struct malloc_type_internal *mtip;
991 	struct malloc_type *mtp;
992 	uint64_t allocs, frees;
993 	uint64_t alloced, freed;
994 	int i;
995 
996 	db_printf("%18s %12s  %12s %12s\n", "Type", "InUse", "MemUse",
997 	    "Requests");
998 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
999 		mtip = (struct malloc_type_internal *)mtp->ks_handle;
1000 		allocs = 0;
1001 		frees = 0;
1002 		alloced = 0;
1003 		freed = 0;
1004 		for (i = 0; i < MAXCPU; i++) {
1005 			allocs += mtip->mti_stats[i].mts_numallocs;
1006 			frees += mtip->mti_stats[i].mts_numfrees;
1007 			alloced += mtip->mti_stats[i].mts_memalloced;
1008 			freed += mtip->mti_stats[i].mts_memfreed;
1009 		}
1010 		db_printf("%18s %12ju %12juK %12ju\n",
1011 		    mtp->ks_shortdesc, allocs - frees,
1012 		    (alloced - freed + 1023) / 1024, allocs);
1013 		if (db_pager_quit)
1014 			break;
1015 	}
1016 }
1017 
1018 #if MALLOC_DEBUG_MAXZONES > 1
1019 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches)
1020 {
1021 	struct malloc_type_internal *mtip;
1022 	struct malloc_type *mtp;
1023 	u_int subzone;
1024 
1025 	if (!have_addr) {
1026 		db_printf("Usage: show multizone_matches <malloc type/addr>\n");
1027 		return;
1028 	}
1029 	mtp = (void *)addr;
1030 	if (mtp->ks_magic != M_MAGIC) {
1031 		db_printf("Magic %lx does not match expected %x\n",
1032 		    mtp->ks_magic, M_MAGIC);
1033 		return;
1034 	}
1035 
1036 	mtip = mtp->ks_handle;
1037 	subzone = mtip->mti_zone;
1038 
1039 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1040 		mtip = mtp->ks_handle;
1041 		if (mtip->mti_zone != subzone)
1042 			continue;
1043 		db_printf("%s\n", mtp->ks_shortdesc);
1044 		if (db_pager_quit)
1045 			break;
1046 	}
1047 }
1048 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
1049 #endif /* DDB */
1050 
1051 #ifdef MALLOC_PROFILE
1052 
1053 static int
1054 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS)
1055 {
1056 	struct sbuf sbuf;
1057 	uint64_t count;
1058 	uint64_t waste;
1059 	uint64_t mem;
1060 	int error;
1061 	int rsize;
1062 	int size;
1063 	int i;
1064 
1065 	waste = 0;
1066 	mem = 0;
1067 
1068 	error = sysctl_wire_old_buffer(req, 0);
1069 	if (error != 0)
1070 		return (error);
1071 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
1072 	sbuf_printf(&sbuf,
1073 	    "\n  Size                    Requests  Real Size\n");
1074 	for (i = 0; i < KMEM_ZSIZE; i++) {
1075 		size = i << KMEM_ZSHIFT;
1076 		rsize = kmemzones[kmemsize[i]].kz_size;
1077 		count = (long long unsigned)krequests[i];
1078 
1079 		sbuf_printf(&sbuf, "%6d%28llu%11d\n", size,
1080 		    (unsigned long long)count, rsize);
1081 
1082 		if ((rsize * count) > (size * count))
1083 			waste += (rsize * count) - (size * count);
1084 		mem += (rsize * count);
1085 	}
1086 	sbuf_printf(&sbuf,
1087 	    "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n",
1088 	    (unsigned long long)mem, (unsigned long long)waste);
1089 	error = sbuf_finish(&sbuf);
1090 	sbuf_delete(&sbuf);
1091 	return (error);
1092 }
1093 
1094 SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD,
1095     NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling");
1096 #endif /* MALLOC_PROFILE */
1097