xref: /freebsd/sys/kern/kern_malloc.c (revision 006afc9b5c6a74f22ce7bbea4c307a940c52f069)
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 struct malloc_type *kmemstatistics;
120 static vm_offset_t kmembase;
121 static vm_offset_t kmemlimit;
122 static int kmemcount;
123 
124 #define KMEM_ZSHIFT	4
125 #define KMEM_ZBASE	16
126 #define KMEM_ZMASK	(KMEM_ZBASE - 1)
127 
128 #define KMEM_ZMAX	PAGE_SIZE
129 #define KMEM_ZSIZE	(KMEM_ZMAX >> KMEM_ZSHIFT)
130 static uint8_t kmemsize[KMEM_ZSIZE + 1];
131 
132 #ifndef MALLOC_DEBUG_MAXZONES
133 #define	MALLOC_DEBUG_MAXZONES	1
134 #endif
135 static int numzones = MALLOC_DEBUG_MAXZONES;
136 
137 /*
138  * Small malloc(9) memory allocations are allocated from a set of UMA buckets
139  * of various sizes.
140  *
141  * XXX: The comment here used to read "These won't be powers of two for
142  * long."  It's possible that a significant amount of wasted memory could be
143  * recovered by tuning the sizes of these buckets.
144  */
145 struct {
146 	int kz_size;
147 	char *kz_name;
148 	uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES];
149 } kmemzones[] = {
150 	{16, "16", },
151 	{32, "32", },
152 	{64, "64", },
153 	{128, "128", },
154 	{256, "256", },
155 	{512, "512", },
156 	{1024, "1024", },
157 	{2048, "2048", },
158 	{4096, "4096", },
159 #if PAGE_SIZE > 4096
160 	{8192, "8192", },
161 #if PAGE_SIZE > 8192
162 	{16384, "16384", },
163 #if PAGE_SIZE > 16384
164 	{32768, "32768", },
165 #if PAGE_SIZE > 32768
166 	{65536, "65536", },
167 #if PAGE_SIZE > 65536
168 #error	"Unsupported PAGE_SIZE"
169 #endif	/* 65536 */
170 #endif	/* 32768 */
171 #endif	/* 16384 */
172 #endif	/* 8192 */
173 #endif	/* 4096 */
174 	{0, NULL},
175 };
176 
177 /*
178  * Zone to allocate malloc type descriptions from.  For ABI reasons, memory
179  * types are described by a data structure passed by the declaring code, but
180  * the malloc(9) implementation has its own data structure describing the
181  * type and statistics.  This permits the malloc(9)-internal data structures
182  * to be modified without breaking binary-compiled kernel modules that
183  * declare malloc types.
184  */
185 static uma_zone_t mt_zone;
186 
187 u_long vm_kmem_size;
188 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0,
189     "Size of kernel memory");
190 
191 static u_long vm_kmem_size_min;
192 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0,
193     "Minimum size of kernel memory");
194 
195 static u_long vm_kmem_size_max;
196 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0,
197     "Maximum size of kernel memory");
198 
199 static u_int vm_kmem_size_scale;
200 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0,
201     "Scale factor for kernel memory size");
202 
203 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS);
204 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size,
205     CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
206     sysctl_kmem_map_size, "LU", "Current kmem_map allocation size");
207 
208 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS);
209 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free,
210     CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
211     sysctl_kmem_map_free, "LU", "Largest contiguous free range in kmem_map");
212 
213 /*
214  * The malloc_mtx protects the kmemstatistics linked list.
215  */
216 struct mtx malloc_mtx;
217 
218 #ifdef MALLOC_PROFILE
219 uint64_t krequests[KMEM_ZSIZE + 1];
220 
221 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS);
222 #endif
223 
224 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS);
225 
226 /*
227  * time_uptime of the last malloc(9) failure (induced or real).
228  */
229 static time_t t_malloc_fail;
230 
231 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1)
232 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0,
233     "Kernel malloc debugging options");
234 #endif
235 
236 /*
237  * malloc(9) fault injection -- cause malloc failures every (n) mallocs when
238  * the caller specifies M_NOWAIT.  If set to 0, no failures are caused.
239  */
240 #ifdef MALLOC_MAKE_FAILURES
241 static int malloc_failure_rate;
242 static int malloc_nowait_count;
243 static int malloc_failure_count;
244 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RW,
245     &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail");
246 TUNABLE_INT("debug.malloc.failure_rate", &malloc_failure_rate);
247 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD,
248     &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures");
249 #endif
250 
251 static int
252 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS)
253 {
254 	u_long size;
255 
256 	size = kmem_map->size;
257 	return (sysctl_handle_long(oidp, &size, 0, req));
258 }
259 
260 static int
261 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS)
262 {
263 	u_long size;
264 
265 	vm_map_lock_read(kmem_map);
266 	size = kmem_map->root != NULL ? kmem_map->root->max_free :
267 	    kmem_map->max_offset - kmem_map->min_offset;
268 	vm_map_unlock_read(kmem_map);
269 	return (sysctl_handle_long(oidp, &size, 0, req));
270 }
271 
272 /*
273  * malloc(9) uma zone separation -- sub-page buffer overruns in one
274  * malloc type will affect only a subset of other malloc types.
275  */
276 #if MALLOC_DEBUG_MAXZONES > 1
277 static void
278 tunable_set_numzones(void)
279 {
280 
281 	TUNABLE_INT_FETCH("debug.malloc.numzones",
282 	    &numzones);
283 
284 	/* Sanity check the number of malloc uma zones. */
285 	if (numzones <= 0)
286 		numzones = 1;
287 	if (numzones > MALLOC_DEBUG_MAXZONES)
288 		numzones = MALLOC_DEBUG_MAXZONES;
289 }
290 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL);
291 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN,
292     &numzones, 0, "Number of malloc uma subzones");
293 
294 /*
295  * Any number that changes regularly is an okay choice for the
296  * offset.  Build numbers are pretty good of you have them.
297  */
298 static u_int zone_offset = __FreeBSD_version;
299 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset);
300 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN,
301     &zone_offset, 0, "Separate malloc types by examining the "
302     "Nth character in the malloc type short description.");
303 
304 static u_int
305 mtp_get_subzone(const char *desc)
306 {
307 	size_t len;
308 	u_int val;
309 
310 	if (desc == NULL || (len = strlen(desc)) == 0)
311 		return (0);
312 	val = desc[zone_offset % len];
313 	return (val % numzones);
314 }
315 #elif MALLOC_DEBUG_MAXZONES == 0
316 #error "MALLOC_DEBUG_MAXZONES must be positive."
317 #else
318 static inline u_int
319 mtp_get_subzone(const char *desc)
320 {
321 
322 	return (0);
323 }
324 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
325 
326 int
327 malloc_last_fail(void)
328 {
329 
330 	return (time_uptime - t_malloc_fail);
331 }
332 
333 /*
334  * An allocation has succeeded -- update malloc type statistics for the
335  * amount of bucket size.  Occurs within a critical section so that the
336  * thread isn't preempted and doesn't migrate while updating per-PCU
337  * statistics.
338  */
339 static void
340 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size,
341     int zindx)
342 {
343 	struct malloc_type_internal *mtip;
344 	struct malloc_type_stats *mtsp;
345 
346 	critical_enter();
347 	mtip = mtp->ks_handle;
348 	mtsp = &mtip->mti_stats[curcpu];
349 	if (size > 0) {
350 		mtsp->mts_memalloced += size;
351 		mtsp->mts_numallocs++;
352 	}
353 	if (zindx != -1)
354 		mtsp->mts_size |= 1 << zindx;
355 
356 #ifdef KDTRACE_HOOKS
357 	if (dtrace_malloc_probe != NULL) {
358 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC];
359 		if (probe_id != 0)
360 			(dtrace_malloc_probe)(probe_id,
361 			    (uintptr_t) mtp, (uintptr_t) mtip,
362 			    (uintptr_t) mtsp, size, zindx);
363 	}
364 #endif
365 
366 	critical_exit();
367 }
368 
369 void
370 malloc_type_allocated(struct malloc_type *mtp, unsigned long size)
371 {
372 
373 	if (size > 0)
374 		malloc_type_zone_allocated(mtp, size, -1);
375 }
376 
377 /*
378  * A free operation has occurred -- update malloc type statistics for the
379  * amount of the bucket size.  Occurs within a critical section so that the
380  * thread isn't preempted and doesn't migrate while updating per-CPU
381  * statistics.
382  */
383 void
384 malloc_type_freed(struct malloc_type *mtp, unsigned long size)
385 {
386 	struct malloc_type_internal *mtip;
387 	struct malloc_type_stats *mtsp;
388 
389 	critical_enter();
390 	mtip = mtp->ks_handle;
391 	mtsp = &mtip->mti_stats[curcpu];
392 	mtsp->mts_memfreed += size;
393 	mtsp->mts_numfrees++;
394 
395 #ifdef KDTRACE_HOOKS
396 	if (dtrace_malloc_probe != NULL) {
397 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE];
398 		if (probe_id != 0)
399 			(dtrace_malloc_probe)(probe_id,
400 			    (uintptr_t) mtp, (uintptr_t) mtip,
401 			    (uintptr_t) mtsp, size, 0);
402 	}
403 #endif
404 
405 	critical_exit();
406 }
407 
408 /*
409  *	contigmalloc:
410  *
411  *	Allocate a block of physically contiguous memory.
412  *
413  *	If M_NOWAIT is set, this routine will not block and return NULL if
414  *	the allocation fails.
415  */
416 void *
417 contigmalloc(unsigned long size, struct malloc_type *type, int flags,
418     vm_paddr_t low, vm_paddr_t high, unsigned long alignment,
419     vm_paddr_t boundary)
420 {
421 	void *ret;
422 
423 	ret = (void *)kmem_alloc_contig(kernel_map, size, flags, low, high,
424 	    alignment, boundary, VM_MEMATTR_DEFAULT);
425 	if (ret != NULL)
426 		malloc_type_allocated(type, round_page(size));
427 	return (ret);
428 }
429 
430 /*
431  *	contigfree:
432  *
433  *	Free a block of memory allocated by contigmalloc.
434  *
435  *	This routine may not block.
436  */
437 void
438 contigfree(void *addr, unsigned long size, struct malloc_type *type)
439 {
440 
441 	kmem_free(kernel_map, (vm_offset_t)addr, size);
442 	malloc_type_freed(type, round_page(size));
443 }
444 
445 /*
446  *	malloc:
447  *
448  *	Allocate a block of memory.
449  *
450  *	If M_NOWAIT is set, this routine will not block and return NULL if
451  *	the allocation fails.
452  */
453 void *
454 malloc(unsigned long size, struct malloc_type *mtp, int flags)
455 {
456 	int indx;
457 	struct malloc_type_internal *mtip;
458 	caddr_t va;
459 	uma_zone_t zone;
460 #if defined(DIAGNOSTIC) || defined(DEBUG_REDZONE)
461 	unsigned long osize = size;
462 #endif
463 
464 #ifdef INVARIANTS
465 	KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic"));
466 	/*
467 	 * Check that exactly one of M_WAITOK or M_NOWAIT is specified.
468 	 */
469 	indx = flags & (M_WAITOK | M_NOWAIT);
470 	if (indx != M_NOWAIT && indx != M_WAITOK) {
471 		static	struct timeval lasterr;
472 		static	int curerr, once;
473 		if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) {
474 			printf("Bad malloc flags: %x\n", indx);
475 			kdb_backtrace();
476 			flags |= M_WAITOK;
477 			once++;
478 		}
479 	}
480 #endif
481 #ifdef MALLOC_MAKE_FAILURES
482 	if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
483 		atomic_add_int(&malloc_nowait_count, 1);
484 		if ((malloc_nowait_count % malloc_failure_rate) == 0) {
485 			atomic_add_int(&malloc_failure_count, 1);
486 			t_malloc_fail = time_uptime;
487 			return (NULL);
488 		}
489 	}
490 #endif
491 	if (flags & M_WAITOK)
492 		KASSERT(curthread->td_intr_nesting_level == 0,
493 		   ("malloc(M_WAITOK) in interrupt context"));
494 
495 #ifdef DEBUG_MEMGUARD
496 	if (memguard_cmp_mtp(mtp, size)) {
497 		va = memguard_alloc(size, flags);
498 		if (va != NULL)
499 			return (va);
500 		/* This is unfortunate but should not be fatal. */
501 	}
502 #endif
503 
504 #ifdef DEBUG_REDZONE
505 	size = redzone_size_ntor(size);
506 #endif
507 
508 	if (size <= KMEM_ZMAX) {
509 		mtip = mtp->ks_handle;
510 		if (size & KMEM_ZMASK)
511 			size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
512 		indx = kmemsize[size >> KMEM_ZSHIFT];
513 		KASSERT(mtip->mti_zone < numzones,
514 		    ("mti_zone %u out of range %d",
515 		    mtip->mti_zone, numzones));
516 		zone = kmemzones[indx].kz_zone[mtip->mti_zone];
517 #ifdef MALLOC_PROFILE
518 		krequests[size >> KMEM_ZSHIFT]++;
519 #endif
520 		va = uma_zalloc(zone, flags);
521 		if (va != NULL)
522 			size = zone->uz_size;
523 		malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
524 	} else {
525 		size = roundup(size, PAGE_SIZE);
526 		zone = NULL;
527 		va = uma_large_malloc(size, flags);
528 		malloc_type_allocated(mtp, va == NULL ? 0 : size);
529 	}
530 	if (flags & M_WAITOK)
531 		KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL"));
532 	else if (va == NULL)
533 		t_malloc_fail = time_uptime;
534 #ifdef DIAGNOSTIC
535 	if (va != NULL && !(flags & M_ZERO)) {
536 		memset(va, 0x70, osize);
537 	}
538 #endif
539 #ifdef DEBUG_REDZONE
540 	if (va != NULL)
541 		va = redzone_setup(va, osize);
542 #endif
543 	return ((void *) va);
544 }
545 
546 /*
547  *	free:
548  *
549  *	Free a block of memory allocated by malloc.
550  *
551  *	This routine may not block.
552  */
553 void
554 free(void *addr, struct malloc_type *mtp)
555 {
556 	uma_slab_t slab;
557 	u_long size;
558 
559 	KASSERT(mtp->ks_magic == M_MAGIC, ("free: bad malloc type magic"));
560 
561 	/* free(NULL, ...) does nothing */
562 	if (addr == NULL)
563 		return;
564 
565 #ifdef DEBUG_MEMGUARD
566 	if (is_memguard_addr(addr)) {
567 		memguard_free(addr);
568 		return;
569 	}
570 #endif
571 
572 #ifdef DEBUG_REDZONE
573 	redzone_check(addr);
574 	addr = redzone_addr_ntor(addr);
575 #endif
576 
577 	slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK));
578 
579 	if (slab == NULL)
580 		panic("free: address %p(%p) has not been allocated.\n",
581 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
582 
583 	if (!(slab->us_flags & UMA_SLAB_MALLOC)) {
584 #ifdef INVARIANTS
585 		struct malloc_type **mtpp = addr;
586 #endif
587 		size = slab->us_keg->uk_size;
588 #ifdef INVARIANTS
589 		/*
590 		 * Cache a pointer to the malloc_type that most recently freed
591 		 * this memory here.  This way we know who is most likely to
592 		 * have stepped on it later.
593 		 *
594 		 * This code assumes that size is a multiple of 8 bytes for
595 		 * 64 bit machines
596 		 */
597 		mtpp = (struct malloc_type **)
598 		    ((unsigned long)mtpp & ~UMA_ALIGN_PTR);
599 		mtpp += (size - sizeof(struct malloc_type *)) /
600 		    sizeof(struct malloc_type *);
601 		*mtpp = mtp;
602 #endif
603 		uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab);
604 	} else {
605 		size = slab->us_size;
606 		uma_large_free(slab);
607 	}
608 	malloc_type_freed(mtp, size);
609 }
610 
611 /*
612  *	realloc: change the size of a memory block
613  */
614 void *
615 realloc(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
616 {
617 	uma_slab_t slab;
618 	unsigned long alloc;
619 	void *newaddr;
620 
621 	KASSERT(mtp->ks_magic == M_MAGIC,
622 	    ("realloc: bad malloc type magic"));
623 
624 	/* realloc(NULL, ...) is equivalent to malloc(...) */
625 	if (addr == NULL)
626 		return (malloc(size, mtp, flags));
627 
628 	/*
629 	 * XXX: Should report free of old memory and alloc of new memory to
630 	 * per-CPU stats.
631 	 */
632 
633 #ifdef DEBUG_MEMGUARD
634 	if (is_memguard_addr(addr))
635 		return (memguard_realloc(addr, size, mtp, flags));
636 #endif
637 
638 #ifdef DEBUG_REDZONE
639 	slab = NULL;
640 	alloc = redzone_get_size(addr);
641 #else
642 	slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK));
643 
644 	/* Sanity check */
645 	KASSERT(slab != NULL,
646 	    ("realloc: address %p out of range", (void *)addr));
647 
648 	/* Get the size of the original block */
649 	if (!(slab->us_flags & UMA_SLAB_MALLOC))
650 		alloc = slab->us_keg->uk_size;
651 	else
652 		alloc = slab->us_size;
653 
654 	/* Reuse the original block if appropriate */
655 	if (size <= alloc
656 	    && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE))
657 		return (addr);
658 #endif /* !DEBUG_REDZONE */
659 
660 	/* Allocate a new, bigger (or smaller) block */
661 	if ((newaddr = malloc(size, mtp, flags)) == NULL)
662 		return (NULL);
663 
664 	/* Copy over original contents */
665 	bcopy(addr, newaddr, min(size, alloc));
666 	free(addr, mtp);
667 	return (newaddr);
668 }
669 
670 /*
671  *	reallocf: same as realloc() but free memory on failure.
672  */
673 void *
674 reallocf(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
675 {
676 	void *mem;
677 
678 	if ((mem = realloc(addr, size, mtp, flags)) == NULL)
679 		free(addr, mtp);
680 	return (mem);
681 }
682 
683 /*
684  * Initialize the kernel memory allocator
685  */
686 /* ARGSUSED*/
687 static void
688 kmeminit(void *dummy)
689 {
690 	uint8_t indx;
691 	u_long mem_size, tmp;
692 	int i;
693 
694 	mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
695 
696 	/*
697 	 * Try to auto-tune the kernel memory size, so that it is
698 	 * more applicable for a wider range of machine sizes.  The
699 	 * VM_KMEM_SIZE_MAX is dependent on the maximum KVA space
700 	 * available.
701 	 *
702 	 * Note that the kmem_map is also used by the zone allocator,
703 	 * so make sure that there is enough space.
704 	 */
705 	vm_kmem_size = VM_KMEM_SIZE + nmbclusters * PAGE_SIZE;
706 	mem_size = cnt.v_page_count;
707 
708 #if defined(VM_KMEM_SIZE_SCALE)
709 	vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
710 #endif
711 	TUNABLE_INT_FETCH("vm.kmem_size_scale", &vm_kmem_size_scale);
712 	if (vm_kmem_size_scale > 0 &&
713 	    (mem_size / vm_kmem_size_scale) > (vm_kmem_size / PAGE_SIZE))
714 		vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE;
715 
716 #if defined(VM_KMEM_SIZE_MIN)
717 	vm_kmem_size_min = VM_KMEM_SIZE_MIN;
718 #endif
719 	TUNABLE_ULONG_FETCH("vm.kmem_size_min", &vm_kmem_size_min);
720 	if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min) {
721 		vm_kmem_size = vm_kmem_size_min;
722 	}
723 
724 #if defined(VM_KMEM_SIZE_MAX)
725 	vm_kmem_size_max = VM_KMEM_SIZE_MAX;
726 #endif
727 	TUNABLE_ULONG_FETCH("vm.kmem_size_max", &vm_kmem_size_max);
728 	if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
729 		vm_kmem_size = vm_kmem_size_max;
730 
731 	/* Allow final override from the kernel environment */
732 	TUNABLE_ULONG_FETCH("vm.kmem_size", &vm_kmem_size);
733 
734 	/*
735 	 * Limit kmem virtual size to twice the physical memory.
736 	 * This allows for kmem map sparseness, but limits the size
737 	 * to something sane.  Be careful to not overflow the 32bit
738 	 * ints while doing the check or the adjustment.
739 	 */
740 	if (vm_kmem_size / 2 / PAGE_SIZE > mem_size)
741 		vm_kmem_size = 2 * mem_size * PAGE_SIZE;
742 
743 #ifdef DEBUG_MEMGUARD
744 	tmp = memguard_fudge(vm_kmem_size, kernel_map);
745 #else
746 	tmp = vm_kmem_size;
747 #endif
748 	kmem_map = kmem_suballoc(kernel_map, &kmembase, &kmemlimit,
749 	    tmp, TRUE);
750 	kmem_map->system_map = 1;
751 
752 #ifdef DEBUG_MEMGUARD
753 	/*
754 	 * Initialize MemGuard if support compiled in.  MemGuard is a
755 	 * replacement allocator used for detecting tamper-after-free
756 	 * scenarios as they occur.  It is only used for debugging.
757 	 */
758 	memguard_init(kmem_map);
759 #endif
760 
761 	uma_startup2();
762 
763 	mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal),
764 #ifdef INVARIANTS
765 	    mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
766 #else
767 	    NULL, NULL, NULL, NULL,
768 #endif
769 	    UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
770 	for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
771 		int size = kmemzones[indx].kz_size;
772 		char *name = kmemzones[indx].kz_name;
773 		int subzone;
774 
775 		for (subzone = 0; subzone < numzones; subzone++) {
776 			kmemzones[indx].kz_zone[subzone] =
777 			    uma_zcreate(name, size,
778 #ifdef INVARIANTS
779 			    mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
780 #else
781 			    NULL, NULL, NULL, NULL,
782 #endif
783 			    UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
784 		}
785 		for (;i <= size; i+= KMEM_ZBASE)
786 			kmemsize[i >> KMEM_ZSHIFT] = indx;
787 
788 	}
789 }
790 
791 void
792 malloc_init(void *data)
793 {
794 	struct malloc_type_internal *mtip;
795 	struct malloc_type *mtp;
796 
797 	KASSERT(cnt.v_page_count != 0, ("malloc_register before vm_init"));
798 
799 	mtp = data;
800 	if (mtp->ks_magic != M_MAGIC)
801 		panic("malloc_init: bad malloc type magic");
802 
803 	mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO);
804 	mtp->ks_handle = mtip;
805 	mtip->mti_zone = mtp_get_subzone(mtp->ks_shortdesc);
806 
807 	mtx_lock(&malloc_mtx);
808 	mtp->ks_next = kmemstatistics;
809 	kmemstatistics = mtp;
810 	kmemcount++;
811 	mtx_unlock(&malloc_mtx);
812 }
813 
814 void
815 malloc_uninit(void *data)
816 {
817 	struct malloc_type_internal *mtip;
818 	struct malloc_type_stats *mtsp;
819 	struct malloc_type *mtp, *temp;
820 	uma_slab_t slab;
821 	long temp_allocs, temp_bytes;
822 	int i;
823 
824 	mtp = data;
825 	KASSERT(mtp->ks_magic == M_MAGIC,
826 	    ("malloc_uninit: bad malloc type magic"));
827 	KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL"));
828 
829 	mtx_lock(&malloc_mtx);
830 	mtip = mtp->ks_handle;
831 	mtp->ks_handle = NULL;
832 	if (mtp != kmemstatistics) {
833 		for (temp = kmemstatistics; temp != NULL;
834 		    temp = temp->ks_next) {
835 			if (temp->ks_next == mtp) {
836 				temp->ks_next = mtp->ks_next;
837 				break;
838 			}
839 		}
840 		KASSERT(temp,
841 		    ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc));
842 	} else
843 		kmemstatistics = mtp->ks_next;
844 	kmemcount--;
845 	mtx_unlock(&malloc_mtx);
846 
847 	/*
848 	 * Look for memory leaks.
849 	 */
850 	temp_allocs = temp_bytes = 0;
851 	for (i = 0; i < MAXCPU; i++) {
852 		mtsp = &mtip->mti_stats[i];
853 		temp_allocs += mtsp->mts_numallocs;
854 		temp_allocs -= mtsp->mts_numfrees;
855 		temp_bytes += mtsp->mts_memalloced;
856 		temp_bytes -= mtsp->mts_memfreed;
857 	}
858 	if (temp_allocs > 0 || temp_bytes > 0) {
859 		printf("Warning: memory type %s leaked memory on destroy "
860 		    "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc,
861 		    temp_allocs, temp_bytes);
862 	}
863 
864 	slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK));
865 	uma_zfree_arg(mt_zone, mtip, slab);
866 }
867 
868 struct malloc_type *
869 malloc_desc2type(const char *desc)
870 {
871 	struct malloc_type *mtp;
872 
873 	mtx_assert(&malloc_mtx, MA_OWNED);
874 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
875 		if (strcmp(mtp->ks_shortdesc, desc) == 0)
876 			return (mtp);
877 	}
878 	return (NULL);
879 }
880 
881 static int
882 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
883 {
884 	struct malloc_type_stream_header mtsh;
885 	struct malloc_type_internal *mtip;
886 	struct malloc_type_header mth;
887 	struct malloc_type *mtp;
888 	int error, i;
889 	struct sbuf sbuf;
890 
891 	error = sysctl_wire_old_buffer(req, 0);
892 	if (error != 0)
893 		return (error);
894 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
895 	mtx_lock(&malloc_mtx);
896 
897 	/*
898 	 * Insert stream header.
899 	 */
900 	bzero(&mtsh, sizeof(mtsh));
901 	mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
902 	mtsh.mtsh_maxcpus = MAXCPU;
903 	mtsh.mtsh_count = kmemcount;
904 	(void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh));
905 
906 	/*
907 	 * Insert alternating sequence of type headers and type statistics.
908 	 */
909 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
910 		mtip = (struct malloc_type_internal *)mtp->ks_handle;
911 
912 		/*
913 		 * Insert type header.
914 		 */
915 		bzero(&mth, sizeof(mth));
916 		strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
917 		(void)sbuf_bcat(&sbuf, &mth, sizeof(mth));
918 
919 		/*
920 		 * Insert type statistics for each CPU.
921 		 */
922 		for (i = 0; i < MAXCPU; i++) {
923 			(void)sbuf_bcat(&sbuf, &mtip->mti_stats[i],
924 			    sizeof(mtip->mti_stats[i]));
925 		}
926 	}
927 	mtx_unlock(&malloc_mtx);
928 	error = sbuf_finish(&sbuf);
929 	sbuf_delete(&sbuf);
930 	return (error);
931 }
932 
933 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT,
934     0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats",
935     "Return malloc types");
936 
937 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
938     "Count of kernel malloc types");
939 
940 void
941 malloc_type_list(malloc_type_list_func_t *func, void *arg)
942 {
943 	struct malloc_type *mtp, **bufmtp;
944 	int count, i;
945 	size_t buflen;
946 
947 	mtx_lock(&malloc_mtx);
948 restart:
949 	mtx_assert(&malloc_mtx, MA_OWNED);
950 	count = kmemcount;
951 	mtx_unlock(&malloc_mtx);
952 
953 	buflen = sizeof(struct malloc_type *) * count;
954 	bufmtp = malloc(buflen, M_TEMP, M_WAITOK);
955 
956 	mtx_lock(&malloc_mtx);
957 
958 	if (count < kmemcount) {
959 		free(bufmtp, M_TEMP);
960 		goto restart;
961 	}
962 
963 	for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++)
964 		bufmtp[i] = mtp;
965 
966 	mtx_unlock(&malloc_mtx);
967 
968 	for (i = 0; i < count; i++)
969 		(func)(bufmtp[i], arg);
970 
971 	free(bufmtp, M_TEMP);
972 }
973 
974 #ifdef DDB
975 DB_SHOW_COMMAND(malloc, db_show_malloc)
976 {
977 	struct malloc_type_internal *mtip;
978 	struct malloc_type *mtp;
979 	uint64_t allocs, frees;
980 	uint64_t alloced, freed;
981 	int i;
982 
983 	db_printf("%18s %12s  %12s %12s\n", "Type", "InUse", "MemUse",
984 	    "Requests");
985 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
986 		mtip = (struct malloc_type_internal *)mtp->ks_handle;
987 		allocs = 0;
988 		frees = 0;
989 		alloced = 0;
990 		freed = 0;
991 		for (i = 0; i < MAXCPU; i++) {
992 			allocs += mtip->mti_stats[i].mts_numallocs;
993 			frees += mtip->mti_stats[i].mts_numfrees;
994 			alloced += mtip->mti_stats[i].mts_memalloced;
995 			freed += mtip->mti_stats[i].mts_memfreed;
996 		}
997 		db_printf("%18s %12ju %12juK %12ju\n",
998 		    mtp->ks_shortdesc, allocs - frees,
999 		    (alloced - freed + 1023) / 1024, allocs);
1000 		if (db_pager_quit)
1001 			break;
1002 	}
1003 }
1004 
1005 #if MALLOC_DEBUG_MAXZONES > 1
1006 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches)
1007 {
1008 	struct malloc_type_internal *mtip;
1009 	struct malloc_type *mtp;
1010 	u_int subzone;
1011 
1012 	if (!have_addr) {
1013 		db_printf("Usage: show multizone_matches <malloc type/addr>\n");
1014 		return;
1015 	}
1016 	mtp = (void *)addr;
1017 	if (mtp->ks_magic != M_MAGIC) {
1018 		db_printf("Magic %lx does not match expected %x\n",
1019 		    mtp->ks_magic, M_MAGIC);
1020 		return;
1021 	}
1022 
1023 	mtip = mtp->ks_handle;
1024 	subzone = mtip->mti_zone;
1025 
1026 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1027 		mtip = mtp->ks_handle;
1028 		if (mtip->mti_zone != subzone)
1029 			continue;
1030 		db_printf("%s\n", mtp->ks_shortdesc);
1031 		if (db_pager_quit)
1032 			break;
1033 	}
1034 }
1035 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
1036 #endif /* DDB */
1037 
1038 #ifdef MALLOC_PROFILE
1039 
1040 static int
1041 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS)
1042 {
1043 	struct sbuf sbuf;
1044 	uint64_t count;
1045 	uint64_t waste;
1046 	uint64_t mem;
1047 	int error;
1048 	int rsize;
1049 	int size;
1050 	int i;
1051 
1052 	waste = 0;
1053 	mem = 0;
1054 
1055 	error = sysctl_wire_old_buffer(req, 0);
1056 	if (error != 0)
1057 		return (error);
1058 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
1059 	sbuf_printf(&sbuf,
1060 	    "\n  Size                    Requests  Real Size\n");
1061 	for (i = 0; i < KMEM_ZSIZE; i++) {
1062 		size = i << KMEM_ZSHIFT;
1063 		rsize = kmemzones[kmemsize[i]].kz_size;
1064 		count = (long long unsigned)krequests[i];
1065 
1066 		sbuf_printf(&sbuf, "%6d%28llu%11d\n", size,
1067 		    (unsigned long long)count, rsize);
1068 
1069 		if ((rsize * count) > (size * count))
1070 			waste += (rsize * count) - (size * count);
1071 		mem += (rsize * count);
1072 	}
1073 	sbuf_printf(&sbuf,
1074 	    "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n",
1075 	    (unsigned long long)mem, (unsigned long long)waste);
1076 	error = sbuf_finish(&sbuf);
1077 	sbuf_delete(&sbuf);
1078 	return (error);
1079 }
1080 
1081 SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD,
1082     NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling");
1083 #endif /* MALLOC_PROFILE */
1084