xref: /freebsd/sys/kern/kern_malloc.c (revision 3642298923e528d795e3a30ec165d2b469e28b40)
1 /*-
2  * Copyright (c) 1987, 1991, 1993
3  *	The Regents of the University of California.
4  * Copyright (c) 2005 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 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_vm.h"
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/kdb.h>
42 #include <sys/kernel.h>
43 #include <sys/lock.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/mutex.h>
47 #include <sys/vmmeter.h>
48 #include <sys/proc.h>
49 #include <sys/sbuf.h>
50 #include <sys/sysctl.h>
51 #include <sys/time.h>
52 
53 #include <vm/vm.h>
54 #include <vm/pmap.h>
55 #include <vm/vm_param.h>
56 #include <vm/vm_kern.h>
57 #include <vm/vm_extern.h>
58 #include <vm/vm_map.h>
59 #include <vm/vm_page.h>
60 #include <vm/uma.h>
61 #include <vm/uma_int.h>
62 #include <vm/uma_dbg.h>
63 
64 #ifdef DEBUG_MEMGUARD
65 #include <vm/memguard.h>
66 #endif
67 
68 #if defined(INVARIANTS) && defined(__i386__)
69 #include <machine/cpu.h>
70 #endif
71 
72 /*
73  * When realloc() is called, if the new size is sufficiently smaller than
74  * the old size, realloc() will allocate a new, smaller block to avoid
75  * wasting memory. 'Sufficiently smaller' is defined as: newsize <=
76  * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'.
77  */
78 #ifndef REALLOC_FRACTION
79 #define	REALLOC_FRACTION	1	/* new block if <= half the size */
80 #endif
81 
82 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches");
83 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
84 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers");
85 
86 MALLOC_DEFINE(M_IP6OPT, "ip6opt", "IPv6 options");
87 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
88 
89 static void kmeminit(void *);
90 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_FIRST, kmeminit, NULL)
91 
92 static MALLOC_DEFINE(M_FREE, "free", "should be on free list");
93 
94 static struct malloc_type *kmemstatistics;
95 static char *kmembase;
96 static char *kmemlimit;
97 static int kmemcount;
98 
99 #define KMEM_ZSHIFT	4
100 #define KMEM_ZBASE	16
101 #define KMEM_ZMASK	(KMEM_ZBASE - 1)
102 
103 #define KMEM_ZMAX	PAGE_SIZE
104 #define KMEM_ZSIZE	(KMEM_ZMAX >> KMEM_ZSHIFT)
105 static u_int8_t kmemsize[KMEM_ZSIZE + 1];
106 
107 /* These won't be powers of two for long */
108 struct {
109 	int kz_size;
110 	char *kz_name;
111 	uma_zone_t kz_zone;
112 } kmemzones[] = {
113 	{16, "16", NULL},
114 	{32, "32", NULL},
115 	{64, "64", NULL},
116 	{128, "128", NULL},
117 	{256, "256", NULL},
118 	{512, "512", NULL},
119 	{1024, "1024", NULL},
120 	{2048, "2048", NULL},
121 	{4096, "4096", NULL},
122 #if PAGE_SIZE > 4096
123 	{8192, "8192", NULL},
124 #if PAGE_SIZE > 8192
125 	{16384, "16384", NULL},
126 #if PAGE_SIZE > 16384
127 	{32768, "32768", NULL},
128 #if PAGE_SIZE > 32768
129 	{65536, "65536", NULL},
130 #if PAGE_SIZE > 65536
131 #error	"Unsupported PAGE_SIZE"
132 #endif	/* 65536 */
133 #endif	/* 32768 */
134 #endif	/* 16384 */
135 #endif	/* 8192 */
136 #endif	/* 4096 */
137 	{0, NULL},
138 };
139 
140 static uma_zone_t mt_zone;
141 
142 #ifdef DEBUG_MEMGUARD
143 u_int vm_memguard_divisor;
144 SYSCTL_UINT(_vm, OID_AUTO, memguard_divisor, CTLFLAG_RD, &vm_memguard_divisor,
145     0, "(kmem_size/memguard_divisor) == memguard submap size");
146 #endif
147 
148 u_int vm_kmem_size;
149 SYSCTL_UINT(_vm, OID_AUTO, kmem_size, CTLFLAG_RD, &vm_kmem_size, 0,
150     "Size of kernel memory");
151 
152 u_int vm_kmem_size_max;
153 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RD, &vm_kmem_size_max, 0,
154     "Maximum size of kernel memory");
155 
156 u_int vm_kmem_size_scale;
157 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RD, &vm_kmem_size_scale, 0,
158     "Scale factor for kernel memory size");
159 
160 /*
161  * The malloc_mtx protects the kmemstatistics linked list.
162  */
163 
164 struct mtx malloc_mtx;
165 
166 #ifdef MALLOC_PROFILE
167 uint64_t krequests[KMEM_ZSIZE + 1];
168 
169 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS);
170 #endif
171 
172 static int sysctl_kern_malloc(SYSCTL_HANDLER_ARGS);
173 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS);
174 
175 /* time_uptime of last malloc(9) failure */
176 static time_t t_malloc_fail;
177 
178 #ifdef MALLOC_MAKE_FAILURES
179 /*
180  * Causes malloc failures every (n) mallocs with M_NOWAIT.  If set to 0,
181  * doesn't cause failures.
182  */
183 SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0,
184     "Kernel malloc debugging options");
185 
186 static int malloc_failure_rate;
187 static int malloc_nowait_count;
188 static int malloc_failure_count;
189 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RW,
190     &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail");
191 TUNABLE_INT("debug.malloc.failure_rate", &malloc_failure_rate);
192 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD,
193     &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures");
194 #endif
195 
196 int
197 malloc_last_fail(void)
198 {
199 
200 	return (time_uptime - t_malloc_fail);
201 }
202 
203 /*
204  * Add this to the informational malloc_type bucket.
205  */
206 static void
207 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size,
208     int zindx)
209 {
210 	struct malloc_type_internal *mtip;
211 	struct malloc_type_stats *mtsp;
212 
213 	critical_enter();
214 	mtip = mtp->ks_handle;
215 	mtsp = &mtip->mti_stats[curcpu];
216 	if (size > 0) {
217 		mtsp->mts_memalloced += size;
218 		mtsp->mts_numallocs++;
219 	}
220 	if (zindx != -1)
221 		mtsp->mts_size |= 1 << zindx;
222 	critical_exit();
223 }
224 
225 void
226 malloc_type_allocated(struct malloc_type *mtp, unsigned long size)
227 {
228 
229 	if (size > 0)
230 		malloc_type_zone_allocated(mtp, size, -1);
231 }
232 
233 /*
234  * Remove this allocation from the informational malloc_type bucket.
235  */
236 void
237 malloc_type_freed(struct malloc_type *mtp, unsigned long size)
238 {
239 	struct malloc_type_internal *mtip;
240 	struct malloc_type_stats *mtsp;
241 
242 	critical_enter();
243 	mtip = mtp->ks_handle;
244 	mtsp = &mtip->mti_stats[curcpu];
245 	mtsp->mts_memfreed += size;
246 	mtsp->mts_numfrees++;
247 	critical_exit();
248 }
249 
250 /*
251  *	malloc:
252  *
253  *	Allocate a block of memory.
254  *
255  *	If M_NOWAIT is set, this routine will not block and return NULL if
256  *	the allocation fails.
257  */
258 void *
259 malloc(unsigned long size, struct malloc_type *mtp, int flags)
260 {
261 	int indx;
262 	caddr_t va;
263 	uma_zone_t zone;
264 	uma_keg_t keg;
265 #ifdef DIAGNOSTIC
266 	unsigned long osize = size;
267 #endif
268 
269 #ifdef INVARIANTS
270 	/*
271 	 * Check that exactly one of M_WAITOK or M_NOWAIT is specified.
272 	 */
273 	indx = flags & (M_WAITOK | M_NOWAIT);
274 	if (indx != M_NOWAIT && indx != M_WAITOK) {
275 		static	struct timeval lasterr;
276 		static	int curerr, once;
277 		if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) {
278 			printf("Bad malloc flags: %x\n", indx);
279 			kdb_backtrace();
280 			flags |= M_WAITOK;
281 			once++;
282 		}
283 	}
284 #endif
285 #if 0
286 	if (size == 0)
287 		kdb_enter("zero size malloc");
288 #endif
289 #ifdef MALLOC_MAKE_FAILURES
290 	if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
291 		atomic_add_int(&malloc_nowait_count, 1);
292 		if ((malloc_nowait_count % malloc_failure_rate) == 0) {
293 			atomic_add_int(&malloc_failure_count, 1);
294 			t_malloc_fail = time_uptime;
295 			return (NULL);
296 		}
297 	}
298 #endif
299 	if (flags & M_WAITOK)
300 		KASSERT(curthread->td_intr_nesting_level == 0,
301 		   ("malloc(M_WAITOK) in interrupt context"));
302 
303 #ifdef DEBUG_MEMGUARD
304 	/* XXX CHANGEME! */
305 	if (mtp == M_SUBPROC)
306 		return memguard_alloc(size, flags);
307 #endif
308 
309 	if (size <= KMEM_ZMAX) {
310 		if (size & KMEM_ZMASK)
311 			size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
312 		indx = kmemsize[size >> KMEM_ZSHIFT];
313 		zone = kmemzones[indx].kz_zone;
314 		keg = zone->uz_keg;
315 #ifdef MALLOC_PROFILE
316 		krequests[size >> KMEM_ZSHIFT]++;
317 #endif
318 		va = uma_zalloc(zone, flags);
319 		if (va != NULL)
320 			size = keg->uk_size;
321 		malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
322 	} else {
323 		size = roundup(size, PAGE_SIZE);
324 		zone = NULL;
325 		keg = NULL;
326 		va = uma_large_malloc(size, flags);
327 		malloc_type_allocated(mtp, va == NULL ? 0 : size);
328 	}
329 	if (flags & M_WAITOK)
330 		KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL"));
331 	else if (va == NULL)
332 		t_malloc_fail = time_uptime;
333 #ifdef DIAGNOSTIC
334 	if (va != NULL && !(flags & M_ZERO)) {
335 		memset(va, 0x70, osize);
336 	}
337 #endif
338 	return ((void *) va);
339 }
340 
341 /*
342  *	free:
343  *
344  *	Free a block of memory allocated by malloc.
345  *
346  *	This routine may not block.
347  */
348 void
349 free(void *addr, struct malloc_type *mtp)
350 {
351 	uma_slab_t slab;
352 	u_long size;
353 
354 	/* free(NULL, ...) does nothing */
355 	if (addr == NULL)
356 		return;
357 
358 #ifdef DEBUG_MEMGUARD
359 	/* XXX CHANGEME! */
360 	if (mtp == M_SUBPROC) {
361 		memguard_free(addr);
362 		return;
363 	}
364 #endif
365 
366 	size = 0;
367 
368 	slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK));
369 
370 	if (slab == NULL)
371 		panic("free: address %p(%p) has not been allocated.\n",
372 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
373 
374 
375 	if (!(slab->us_flags & UMA_SLAB_MALLOC)) {
376 #ifdef INVARIANTS
377 		struct malloc_type **mtpp = addr;
378 #endif
379 		size = slab->us_keg->uk_size;
380 #ifdef INVARIANTS
381 		/*
382 		 * Cache a pointer to the malloc_type that most recently freed
383 		 * this memory here.  This way we know who is most likely to
384 		 * have stepped on it later.
385 		 *
386 		 * This code assumes that size is a multiple of 8 bytes for
387 		 * 64 bit machines
388 		 */
389 		mtpp = (struct malloc_type **)
390 		    ((unsigned long)mtpp & ~UMA_ALIGN_PTR);
391 		mtpp += (size - sizeof(struct malloc_type *)) /
392 		    sizeof(struct malloc_type *);
393 		*mtpp = mtp;
394 #endif
395 		uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab);
396 	} else {
397 		size = slab->us_size;
398 		uma_large_free(slab);
399 	}
400 	malloc_type_freed(mtp, size);
401 }
402 
403 /*
404  *	realloc: change the size of a memory block
405  */
406 void *
407 realloc(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
408 {
409 	uma_slab_t slab;
410 	unsigned long alloc;
411 	void *newaddr;
412 
413 	/* realloc(NULL, ...) is equivalent to malloc(...) */
414 	if (addr == NULL)
415 		return (malloc(size, mtp, flags));
416 
417 	/*
418 	 * XXX: Should report free of old memory and alloc of new memory to
419 	 * per-CPU stats.
420 	 */
421 
422 #ifdef DEBUG_MEMGUARD
423 /* XXX: CHANGEME! */
424 if (mtp == M_SUBPROC) {
425 	slab = NULL;
426 	alloc = size;
427 } else {
428 #endif
429 
430 	slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK));
431 
432 	/* Sanity check */
433 	KASSERT(slab != NULL,
434 	    ("realloc: address %p out of range", (void *)addr));
435 
436 	/* Get the size of the original block */
437 	if (slab->us_keg)
438 		alloc = slab->us_keg->uk_size;
439 	else
440 		alloc = slab->us_size;
441 
442 	/* Reuse the original block if appropriate */
443 	if (size <= alloc
444 	    && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE))
445 		return (addr);
446 
447 #ifdef DEBUG_MEMGUARD
448 }
449 #endif
450 
451 	/* Allocate a new, bigger (or smaller) block */
452 	if ((newaddr = malloc(size, mtp, flags)) == NULL)
453 		return (NULL);
454 
455 	/* Copy over original contents */
456 	bcopy(addr, newaddr, min(size, alloc));
457 	free(addr, mtp);
458 	return (newaddr);
459 }
460 
461 /*
462  *	reallocf: same as realloc() but free memory on failure.
463  */
464 void *
465 reallocf(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
466 {
467 	void *mem;
468 
469 	if ((mem = realloc(addr, size, mtp, flags)) == NULL)
470 		free(addr, mtp);
471 	return (mem);
472 }
473 
474 /*
475  * Initialize the kernel memory allocator
476  */
477 /* ARGSUSED*/
478 static void
479 kmeminit(void *dummy)
480 {
481 	u_int8_t indx;
482 	u_long mem_size;
483 	int i;
484 
485 	mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
486 
487 	/*
488 	 * Try to auto-tune the kernel memory size, so that it is
489 	 * more applicable for a wider range of machine sizes.
490 	 * On an X86, a VM_KMEM_SIZE_SCALE value of 4 is good, while
491 	 * a VM_KMEM_SIZE of 12MB is a fair compromise.  The
492 	 * VM_KMEM_SIZE_MAX is dependent on the maximum KVA space
493 	 * available, and on an X86 with a total KVA space of 256MB,
494 	 * try to keep VM_KMEM_SIZE_MAX at 80MB or below.
495 	 *
496 	 * Note that the kmem_map is also used by the zone allocator,
497 	 * so make sure that there is enough space.
498 	 */
499 	vm_kmem_size = VM_KMEM_SIZE + nmbclusters * PAGE_SIZE;
500 	mem_size = cnt.v_page_count;
501 
502 #if defined(VM_KMEM_SIZE_SCALE)
503 	vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
504 #endif
505 	TUNABLE_INT_FETCH("vm.kmem_size_scale", &vm_kmem_size_scale);
506 	if (vm_kmem_size_scale > 0 &&
507 	    (mem_size / vm_kmem_size_scale) > (vm_kmem_size / PAGE_SIZE))
508 		vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE;
509 
510 #if defined(VM_KMEM_SIZE_MAX)
511 	vm_kmem_size_max = VM_KMEM_SIZE_MAX;
512 #endif
513 	TUNABLE_INT_FETCH("vm.kmem_size_max", &vm_kmem_size_max);
514 	if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
515 		vm_kmem_size = vm_kmem_size_max;
516 
517 	/* Allow final override from the kernel environment */
518 #ifndef BURN_BRIDGES
519 	if (TUNABLE_INT_FETCH("kern.vm.kmem.size", &vm_kmem_size) != 0)
520 		printf("kern.vm.kmem.size is now called vm.kmem_size!\n");
521 #endif
522 	TUNABLE_INT_FETCH("vm.kmem_size", &vm_kmem_size);
523 
524 	/*
525 	 * Limit kmem virtual size to twice the physical memory.
526 	 * This allows for kmem map sparseness, but limits the size
527 	 * to something sane. Be careful to not overflow the 32bit
528 	 * ints while doing the check.
529 	 */
530 	if (((vm_kmem_size / 2) / PAGE_SIZE) > cnt.v_page_count)
531 		vm_kmem_size = 2 * cnt.v_page_count * PAGE_SIZE;
532 
533 	/*
534 	 * Tune settings based on the kernel map's size at this time.
535 	 */
536 	init_param3(vm_kmem_size / PAGE_SIZE);
537 
538 	kmem_map = kmem_suballoc(kernel_map, (vm_offset_t *)&kmembase,
539 		(vm_offset_t *)&kmemlimit, vm_kmem_size);
540 	kmem_map->system_map = 1;
541 
542 #ifdef DEBUG_MEMGUARD
543 	/*
544 	 * Initialize MemGuard if support compiled in.  MemGuard is a
545 	 * replacement allocator used for detecting tamper-after-free
546 	 * scenarios as they occur.  It is only used for debugging.
547 	 */
548 	vm_memguard_divisor = 10;
549 	TUNABLE_INT_FETCH("vm.memguard_divisor", &vm_memguard_divisor);
550 
551 	/* Pick a conservative value if provided value sucks. */
552 	if ((vm_memguard_divisor <= 0) ||
553 	    ((vm_kmem_size / vm_memguard_divisor) == 0))
554 		vm_memguard_divisor = 10;
555 	memguard_init(kmem_map, vm_kmem_size / vm_memguard_divisor);
556 #endif
557 
558 	uma_startup2();
559 
560 	mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal),
561 #ifdef INVARIANTS
562 	    mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
563 #else
564 	    NULL, NULL, NULL, NULL,
565 #endif
566 	    UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
567 	for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
568 		int size = kmemzones[indx].kz_size;
569 		char *name = kmemzones[indx].kz_name;
570 
571 		kmemzones[indx].kz_zone = uma_zcreate(name, size,
572 #ifdef INVARIANTS
573 		    mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
574 #else
575 		    NULL, NULL, NULL, NULL,
576 #endif
577 		    UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
578 
579 		for (;i <= size; i+= KMEM_ZBASE)
580 			kmemsize[i >> KMEM_ZSHIFT] = indx;
581 
582 	}
583 }
584 
585 void
586 malloc_init(void *data)
587 {
588 	struct malloc_type_internal *mtip;
589 	struct malloc_type *mtp;
590 
591 	KASSERT(cnt.v_page_count != 0, ("malloc_register before vm_init"));
592 
593 	mtp = data;
594 	mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO);
595 	mtp->ks_handle = mtip;
596 
597 	mtx_lock(&malloc_mtx);
598 	mtp->ks_next = kmemstatistics;
599 	kmemstatistics = mtp;
600 	kmemcount++;
601 	mtx_unlock(&malloc_mtx);
602 }
603 
604 void
605 malloc_uninit(void *data)
606 {
607 	struct malloc_type_internal *mtip;
608 	struct malloc_type *mtp, *temp;
609 
610 	mtp = data;
611 	KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL"));
612 	mtx_lock(&malloc_mtx);
613 	mtip = mtp->ks_handle;
614 	mtp->ks_handle = NULL;
615 	if (mtp != kmemstatistics) {
616 		for (temp = kmemstatistics; temp != NULL;
617 		    temp = temp->ks_next) {
618 			if (temp->ks_next == mtp)
619 				temp->ks_next = mtp->ks_next;
620 		}
621 	} else
622 		kmemstatistics = mtp->ks_next;
623 	kmemcount--;
624 	mtx_unlock(&malloc_mtx);
625 	uma_zfree(mt_zone, mtip);
626 }
627 
628 static int
629 sysctl_kern_malloc(SYSCTL_HANDLER_ARGS)
630 {
631 	struct malloc_type_stats mts_local, *mtsp;
632 	struct malloc_type_internal *mtip;
633 	struct malloc_type *mtp;
634 	struct sbuf sbuf;
635 	long temp_allocs, temp_bytes;
636 	int linesize = 128;
637 	int bufsize;
638 	int first;
639 	int error;
640 	char *buf;
641 	int cnt;
642 	int i;
643 
644 	cnt = 0;
645 
646 	/* Guess at how much room is needed. */
647 	mtx_lock(&malloc_mtx);
648 	cnt = kmemcount;
649 	mtx_unlock(&malloc_mtx);
650 
651 	bufsize = linesize * (cnt + 1);
652 	buf = malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO);
653 	sbuf_new(&sbuf, buf, bufsize, SBUF_FIXEDLEN);
654 
655 	mtx_lock(&malloc_mtx);
656 	sbuf_printf(&sbuf,
657 	    "\n        Type  InUse MemUse HighUse Requests  Size(s)\n");
658 	for (mtp = kmemstatistics; cnt != 0 && mtp != NULL;
659 	    mtp = mtp->ks_next, cnt--) {
660 		mtip = mtp->ks_handle;
661 		bzero(&mts_local, sizeof(mts_local));
662 		for (i = 0; i < MAXCPU; i++) {
663 			mtsp = &mtip->mti_stats[i];
664 			mts_local.mts_memalloced += mtsp->mts_memalloced;
665 			mts_local.mts_memfreed += mtsp->mts_memfreed;
666 			mts_local.mts_numallocs += mtsp->mts_numallocs;
667 			mts_local.mts_numfrees += mtsp->mts_numfrees;
668 			mts_local.mts_size |= mtsp->mts_size;
669 		}
670 		if (mts_local.mts_numallocs == 0)
671 			continue;
672 
673 		/*
674 		 * Due to races in per-CPU statistics gather, it's possible to
675 		 * get a slightly negative number here.  If we do, approximate
676 		 * with 0.
677 		 */
678 		if (mts_local.mts_numallocs > mts_local.mts_numfrees)
679 			temp_allocs = mts_local.mts_numallocs -
680 			    mts_local.mts_numfrees;
681 		else
682 			temp_allocs = 0;
683 
684 		/*
685 		 * Ditto for bytes allocated.
686 		 */
687 		if (mts_local.mts_memalloced > mts_local.mts_memfreed)
688 			temp_bytes = mts_local.mts_memalloced -
689 			    mts_local.mts_memfreed;
690 		else
691 			temp_bytes = 0;
692 
693 		/*
694 		 * High-waterwark is no longer easily available, so we just
695 		 * print '-' for that column.
696 		 */
697 		sbuf_printf(&sbuf, "%13s%6lu%6luK       -%9llu",
698 		    mtp->ks_shortdesc,
699 		    temp_allocs,
700 		    (temp_bytes + 1023) / 1024,
701 		    (unsigned long long)mts_local.mts_numallocs);
702 
703 		first = 1;
704 		for (i = 0; i < sizeof(kmemzones) / sizeof(kmemzones[0]) - 1;
705 		    i++) {
706 			if (mts_local.mts_size & (1 << i)) {
707 				if (first)
708 					sbuf_printf(&sbuf, "  ");
709 				else
710 					sbuf_printf(&sbuf, ",");
711 				sbuf_printf(&sbuf, "%s",
712 				    kmemzones[i].kz_name);
713 				first = 0;
714 			}
715 		}
716 		sbuf_printf(&sbuf, "\n");
717 	}
718 	sbuf_finish(&sbuf);
719 	mtx_unlock(&malloc_mtx);
720 
721 	error = SYSCTL_OUT(req, sbuf_data(&sbuf), sbuf_len(&sbuf));
722 
723 	sbuf_delete(&sbuf);
724 	free(buf, M_TEMP);
725 	return (error);
726 }
727 
728 SYSCTL_OID(_kern, OID_AUTO, malloc, CTLTYPE_STRING|CTLFLAG_RD,
729     NULL, 0, sysctl_kern_malloc, "A", "Malloc Stats");
730 
731 static int
732 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
733 {
734 	struct malloc_type_stream_header mtsh;
735 	struct malloc_type_internal *mtip;
736 	struct malloc_type_header mth;
737 	struct malloc_type *mtp;
738 	int buflen, count, error, i;
739 	struct sbuf sbuf;
740 	char *buffer;
741 
742 	mtx_lock(&malloc_mtx);
743 restart:
744 	mtx_assert(&malloc_mtx, MA_OWNED);
745 	count = kmemcount;
746 	mtx_unlock(&malloc_mtx);
747 	buflen = sizeof(mtsh) + count * (sizeof(mth) +
748 	    sizeof(struct malloc_type_stats) * MAXCPU) + 1;
749 	buffer = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO);
750 	mtx_lock(&malloc_mtx);
751 	if (count < kmemcount) {
752 		free(buffer, M_TEMP);
753 		goto restart;
754 	}
755 
756 	sbuf_new(&sbuf, buffer, buflen, SBUF_FIXEDLEN);
757 
758 	/*
759 	 * Insert stream header.
760 	 */
761 	bzero(&mtsh, sizeof(mtsh));
762 	mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
763 	mtsh.mtsh_maxcpus = MAXCPU;
764 	mtsh.mtsh_count = kmemcount;
765 	if (sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh)) < 0) {
766 		mtx_unlock(&malloc_mtx);
767 		error = ENOMEM;
768 		goto out;
769 	}
770 
771 	/*
772 	 * Insert alternating sequence of type headers and type statistics.
773 	 */
774 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
775 		mtip = (struct malloc_type_internal *)mtp->ks_handle;
776 
777 		/*
778 		 * Insert type header.
779 		 */
780 		bzero(&mth, sizeof(mth));
781 		strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
782 		if (sbuf_bcat(&sbuf, &mth, sizeof(mth)) < 0) {
783 			mtx_unlock(&malloc_mtx);
784 			error = ENOMEM;
785 			goto out;
786 		}
787 
788 		/*
789 		 * Insert type statistics for each CPU.
790 		 */
791 		for (i = 0; i < MAXCPU; i++) {
792 			if (sbuf_bcat(&sbuf, &mtip->mti_stats[i],
793 			    sizeof(mtip->mti_stats[i])) < 0) {
794 				mtx_unlock(&malloc_mtx);
795 				error = ENOMEM;
796 				goto out;
797 			}
798 		}
799 	}
800 	mtx_unlock(&malloc_mtx);
801 	sbuf_finish(&sbuf);
802 	error = SYSCTL_OUT(req, sbuf_data(&sbuf), sbuf_len(&sbuf));
803 out:
804 	sbuf_delete(&sbuf);
805 	free(buffer, M_TEMP);
806 	return (error);
807 }
808 
809 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT,
810     0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats",
811     "Return malloc types");
812 
813 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
814     "Count of kernel malloc types");
815 
816 #ifdef MALLOC_PROFILE
817 
818 static int
819 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS)
820 {
821 	int linesize = 64;
822 	struct sbuf sbuf;
823 	uint64_t count;
824 	uint64_t waste;
825 	uint64_t mem;
826 	int bufsize;
827 	int error;
828 	char *buf;
829 	int rsize;
830 	int size;
831 	int i;
832 
833 	bufsize = linesize * (KMEM_ZSIZE + 1);
834 	bufsize += 128; 	/* For the stats line */
835 	bufsize += 128; 	/* For the banner line */
836 	waste = 0;
837 	mem = 0;
838 
839 	buf = malloc(bufsize, M_TEMP, M_WAITOK|M_ZERO);
840 	sbuf_new(&sbuf, buf, bufsize, SBUF_FIXEDLEN);
841 	sbuf_printf(&sbuf,
842 	    "\n  Size                    Requests  Real Size\n");
843 	for (i = 0; i < KMEM_ZSIZE; i++) {
844 		size = i << KMEM_ZSHIFT;
845 		rsize = kmemzones[kmemsize[i]].kz_size;
846 		count = (long long unsigned)krequests[i];
847 
848 		sbuf_printf(&sbuf, "%6d%28llu%11d\n", size,
849 		    (unsigned long long)count, rsize);
850 
851 		if ((rsize * count) > (size * count))
852 			waste += (rsize * count) - (size * count);
853 		mem += (rsize * count);
854 	}
855 	sbuf_printf(&sbuf,
856 	    "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n",
857 	    (unsigned long long)mem, (unsigned long long)waste);
858 	sbuf_finish(&sbuf);
859 
860 	error = SYSCTL_OUT(req, sbuf_data(&sbuf), sbuf_len(&sbuf));
861 
862 	sbuf_delete(&sbuf);
863 	free(buf, M_TEMP);
864 	return (error);
865 }
866 
867 SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD,
868     NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling");
869 #endif /* MALLOC_PROFILE */
870