xref: /freebsd/sys/kern/kern_malloc.c (revision 4543ef516683042d46f3bd3bb8a4f3f746e00499)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1987, 1991, 1993
5  *	The Regents of the University of California.
6  * Copyright (c) 2005-2009 Robert N. M. Watson
7  * Copyright (c) 2008 Otto Moerbeek <otto@drijf.net> (mallocarray)
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
35 /*
36  * Kernel malloc(9) implementation -- general purpose kernel memory allocator
37  * based on memory types.  Back end is implemented using the UMA(9) zone
38  * allocator.  A set of fixed-size buckets are used for smaller allocations,
39  * and a special UMA allocation interface is used for larger allocations.
40  * Callers declare memory types, and statistics are maintained independently
41  * for each memory type.  Statistics are maintained per-CPU for performance
42  * reasons.  See malloc(9) and comments in malloc.h for a detailed
43  * description.
44  */
45 
46 #include <sys/cdefs.h>
47 #include "opt_ddb.h"
48 #include "opt_vm.h"
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/asan.h>
53 #include <sys/kdb.h>
54 #include <sys/kernel.h>
55 #include <sys/lock.h>
56 #include <sys/malloc.h>
57 #include <sys/msan.h>
58 #include <sys/mutex.h>
59 #include <sys/vmmeter.h>
60 #include <sys/proc.h>
61 #include <sys/queue.h>
62 #include <sys/sbuf.h>
63 #include <sys/smp.h>
64 #include <sys/sysctl.h>
65 #include <sys/time.h>
66 #include <sys/vmem.h>
67 #ifdef EPOCH_TRACE
68 #include <sys/epoch.h>
69 #endif
70 
71 #include <vm/vm.h>
72 #include <vm/pmap.h>
73 #include <vm/vm_domainset.h>
74 #include <vm/vm_pageout.h>
75 #include <vm/vm_param.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_phys.h>
81 #include <vm/vm_pagequeue.h>
82 #include <vm/uma.h>
83 #include <vm/uma_int.h>
84 #include <vm/uma_dbg.h>
85 
86 #ifdef DEBUG_MEMGUARD
87 #include <vm/memguard.h>
88 #endif
89 #ifdef DEBUG_REDZONE
90 #include <vm/redzone.h>
91 #endif
92 
93 #if defined(INVARIANTS) && defined(__i386__)
94 #include <machine/cpu.h>
95 #endif
96 
97 #include <ddb/ddb.h>
98 
99 #ifdef KDTRACE_HOOKS
100 #include <sys/dtrace_bsd.h>
101 
102 bool	__read_frequently			dtrace_malloc_enabled;
103 dtrace_malloc_probe_func_t __read_mostly	dtrace_malloc_probe;
104 #endif
105 
106 #if defined(INVARIANTS) || defined(MALLOC_MAKE_FAILURES) ||		\
107     defined(DEBUG_MEMGUARD) || defined(DEBUG_REDZONE)
108 #define	MALLOC_DEBUG	1
109 #endif
110 
111 #if defined(KASAN) || defined(DEBUG_REDZONE)
112 #define	DEBUG_REDZONE_ARG_DEF	, unsigned long osize
113 #define	DEBUG_REDZONE_ARG	, osize
114 #else
115 #define	DEBUG_REDZONE_ARG_DEF
116 #define	DEBUG_REDZONE_ARG
117 #endif
118 
119 /*
120  * When realloc() is called, if the new size is sufficiently smaller than
121  * the old size, realloc() will allocate a new, smaller block to avoid
122  * wasting memory. 'Sufficiently smaller' is defined as: newsize <=
123  * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'.
124  */
125 #ifndef REALLOC_FRACTION
126 #define	REALLOC_FRACTION	1	/* new block if <= half the size */
127 #endif
128 
129 /*
130  * Centrally define some common malloc types.
131  */
132 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches");
133 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
134 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers");
135 
136 static struct malloc_type *kmemstatistics;
137 static int kmemcount;
138 
139 #define KMEM_ZSHIFT	4
140 #define KMEM_ZBASE	16
141 #define KMEM_ZMASK	(KMEM_ZBASE - 1)
142 
143 #define KMEM_ZMAX	65536
144 #define KMEM_ZSIZE	(KMEM_ZMAX >> KMEM_ZSHIFT)
145 static uint8_t kmemsize[KMEM_ZSIZE + 1];
146 
147 #ifndef MALLOC_DEBUG_MAXZONES
148 #define	MALLOC_DEBUG_MAXZONES	1
149 #endif
150 static int numzones = MALLOC_DEBUG_MAXZONES;
151 
152 /*
153  * Small malloc(9) memory allocations are allocated from a set of UMA buckets
154  * of various sizes.
155  *
156  * Warning: the layout of the struct is duplicated in libmemstat for KVM support.
157  *
158  * XXX: The comment here used to read "These won't be powers of two for
159  * long."  It's possible that a significant amount of wasted memory could be
160  * recovered by tuning the sizes of these buckets.
161  */
162 struct {
163 	int kz_size;
164 	const char *kz_name;
165 	uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES];
166 } kmemzones[] = {
167 	{16, "malloc-16", },
168 	{32, "malloc-32", },
169 	{64, "malloc-64", },
170 	{128, "malloc-128", },
171 	{256, "malloc-256", },
172 	{384, "malloc-384", },
173 	{512, "malloc-512", },
174 	{1024, "malloc-1024", },
175 	{2048, "malloc-2048", },
176 	{4096, "malloc-4096", },
177 	{8192, "malloc-8192", },
178 	{16384, "malloc-16384", },
179 	{32768, "malloc-32768", },
180 	{65536, "malloc-65536", },
181 	{0, NULL},
182 };
183 
184 u_long vm_kmem_size;
185 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0,
186     "Size of kernel memory");
187 
188 static u_long kmem_zmax = KMEM_ZMAX;
189 SYSCTL_ULONG(_vm, OID_AUTO, kmem_zmax, CTLFLAG_RDTUN, &kmem_zmax, 0,
190     "Maximum allocation size that malloc(9) would use UMA as backend");
191 
192 static u_long vm_kmem_size_min;
193 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0,
194     "Minimum size of kernel memory");
195 
196 static u_long vm_kmem_size_max;
197 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0,
198     "Maximum size of kernel memory");
199 
200 static u_int vm_kmem_size_scale;
201 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0,
202     "Scale factor for kernel memory size");
203 
204 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS);
205 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size,
206     CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
207     sysctl_kmem_map_size, "LU", "Current kmem allocation size");
208 
209 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS);
210 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free,
211     CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
212     sysctl_kmem_map_free, "LU", "Free space in kmem");
213 
214 static SYSCTL_NODE(_vm, OID_AUTO, malloc, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
215     "Malloc information");
216 
217 static u_int vm_malloc_zone_count = nitems(kmemzones);
218 SYSCTL_UINT(_vm_malloc, OID_AUTO, zone_count,
219     CTLFLAG_RD, &vm_malloc_zone_count, 0,
220     "Number of malloc zones");
221 
222 static int sysctl_vm_malloc_zone_sizes(SYSCTL_HANDLER_ARGS);
223 SYSCTL_PROC(_vm_malloc, OID_AUTO, zone_sizes,
224     CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, NULL, 0,
225     sysctl_vm_malloc_zone_sizes, "S", "Zone sizes used by malloc");
226 
227 /*
228  * The malloc_mtx protects the kmemstatistics linked list.
229  */
230 struct mtx malloc_mtx;
231 
232 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS);
233 
234 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1)
235 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
236     "Kernel malloc debugging options");
237 #endif
238 
239 /*
240  * malloc(9) fault injection -- cause malloc failures every (n) mallocs when
241  * the caller specifies M_NOWAIT.  If set to 0, no failures are caused.
242  */
243 #ifdef MALLOC_MAKE_FAILURES
244 static int malloc_failure_rate;
245 static int malloc_nowait_count;
246 static int malloc_failure_count;
247 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RWTUN,
248     &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail");
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 = uma_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, limit;
266 
267 	/* The sysctl is unsigned, implement as a saturation value. */
268 	size = uma_size();
269 	limit = uma_limit();
270 	if (size > limit)
271 		size = 0;
272 	else
273 		size = limit - size;
274 	return (sysctl_handle_long(oidp, &size, 0, req));
275 }
276 
277 static int
278 sysctl_vm_malloc_zone_sizes(SYSCTL_HANDLER_ARGS)
279 {
280 	int sizes[nitems(kmemzones)];
281 	int i;
282 
283 	for (i = 0; i < nitems(kmemzones); i++) {
284 		sizes[i] = kmemzones[i].kz_size;
285 	}
286 
287 	return (SYSCTL_OUT(req, &sizes, sizeof(sizes)));
288 }
289 
290 /*
291  * malloc(9) uma zone separation -- sub-page buffer overruns in one
292  * malloc type will affect only a subset of other malloc types.
293  */
294 #if MALLOC_DEBUG_MAXZONES > 1
295 static void
296 tunable_set_numzones(void)
297 {
298 
299 	TUNABLE_INT_FETCH("debug.malloc.numzones",
300 	    &numzones);
301 
302 	/* Sanity check the number of malloc uma zones. */
303 	if (numzones <= 0)
304 		numzones = 1;
305 	if (numzones > MALLOC_DEBUG_MAXZONES)
306 		numzones = MALLOC_DEBUG_MAXZONES;
307 }
308 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL);
309 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
310     &numzones, 0, "Number of malloc uma subzones");
311 
312 /*
313  * Any number that changes regularly is an okay choice for the
314  * offset.  Build numbers are pretty good of you have them.
315  */
316 static u_int zone_offset = __FreeBSD_version;
317 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset);
318 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN,
319     &zone_offset, 0, "Separate malloc types by examining the "
320     "Nth character in the malloc type short description.");
321 
322 static void
323 mtp_set_subzone(struct malloc_type *mtp)
324 {
325 	struct malloc_type_internal *mtip;
326 	const char *desc;
327 	size_t len;
328 	u_int val;
329 
330 	mtip = &mtp->ks_mti;
331 	desc = mtp->ks_shortdesc;
332 	if (desc == NULL || (len = strlen(desc)) == 0)
333 		val = 0;
334 	else
335 		val = desc[zone_offset % len];
336 	mtip->mti_zone = (val % numzones);
337 }
338 
339 static inline u_int
340 mtp_get_subzone(struct malloc_type *mtp)
341 {
342 	struct malloc_type_internal *mtip;
343 
344 	mtip = &mtp->ks_mti;
345 
346 	KASSERT(mtip->mti_zone < numzones,
347 	    ("mti_zone %u out of range %d",
348 	    mtip->mti_zone, numzones));
349 	return (mtip->mti_zone);
350 }
351 #elif MALLOC_DEBUG_MAXZONES == 0
352 #error "MALLOC_DEBUG_MAXZONES must be positive."
353 #else
354 static void
355 mtp_set_subzone(struct malloc_type *mtp)
356 {
357 	struct malloc_type_internal *mtip;
358 
359 	mtip = &mtp->ks_mti;
360 	mtip->mti_zone = 0;
361 }
362 
363 static inline u_int
364 mtp_get_subzone(struct malloc_type *mtp)
365 {
366 
367 	return (0);
368 }
369 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
370 
371 /*
372  * An allocation has succeeded -- update malloc type statistics for the
373  * amount of bucket size.  Occurs within a critical section so that the
374  * thread isn't preempted and doesn't migrate while updating per-PCU
375  * statistics.
376  */
377 static void
378 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size,
379     int zindx)
380 {
381 	struct malloc_type_internal *mtip;
382 	struct malloc_type_stats *mtsp;
383 
384 	critical_enter();
385 	mtip = &mtp->ks_mti;
386 	mtsp = zpcpu_get(mtip->mti_stats);
387 	if (size > 0) {
388 		mtsp->mts_memalloced += size;
389 		mtsp->mts_numallocs++;
390 	}
391 	if (zindx != -1)
392 		mtsp->mts_size |= 1 << zindx;
393 
394 #ifdef KDTRACE_HOOKS
395 	if (__predict_false(dtrace_malloc_enabled)) {
396 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC];
397 		if (probe_id != 0)
398 			(dtrace_malloc_probe)(probe_id,
399 			    (uintptr_t) mtp, (uintptr_t) mtip,
400 			    (uintptr_t) mtsp, size, zindx);
401 	}
402 #endif
403 
404 	critical_exit();
405 }
406 
407 void
408 malloc_type_allocated(struct malloc_type *mtp, unsigned long size)
409 {
410 
411 	if (size > 0)
412 		malloc_type_zone_allocated(mtp, size, -1);
413 }
414 
415 /*
416  * A free operation has occurred -- update malloc type statistics for the
417  * amount of the bucket size.  Occurs within a critical section so that the
418  * thread isn't preempted and doesn't migrate while updating per-CPU
419  * statistics.
420  */
421 void
422 malloc_type_freed(struct malloc_type *mtp, unsigned long size)
423 {
424 	struct malloc_type_internal *mtip;
425 	struct malloc_type_stats *mtsp;
426 
427 	critical_enter();
428 	mtip = &mtp->ks_mti;
429 	mtsp = zpcpu_get(mtip->mti_stats);
430 	mtsp->mts_memfreed += size;
431 	mtsp->mts_numfrees++;
432 
433 #ifdef KDTRACE_HOOKS
434 	if (__predict_false(dtrace_malloc_enabled)) {
435 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE];
436 		if (probe_id != 0)
437 			(dtrace_malloc_probe)(probe_id,
438 			    (uintptr_t) mtp, (uintptr_t) mtip,
439 			    (uintptr_t) mtsp, size, 0);
440 	}
441 #endif
442 
443 	critical_exit();
444 }
445 
446 /*
447  *	contigmalloc:
448  *
449  *	Allocate a block of physically contiguous memory.
450  *
451  *	If M_NOWAIT is set, this routine will not block and return NULL if
452  *	the allocation fails.
453  */
454 void *
455 contigmalloc(unsigned long size, struct malloc_type *type, int flags,
456     vm_paddr_t low, vm_paddr_t high, unsigned long alignment,
457     vm_paddr_t boundary)
458 {
459 	void *ret;
460 
461 	ret = (void *)kmem_alloc_contig(size, flags, low, high, alignment,
462 	    boundary, VM_MEMATTR_DEFAULT);
463 	if (ret != NULL)
464 		malloc_type_allocated(type, round_page(size));
465 	return (ret);
466 }
467 
468 void *
469 contigmalloc_domainset(unsigned long size, struct malloc_type *type,
470     struct domainset *ds, int flags, vm_paddr_t low, vm_paddr_t high,
471     unsigned long alignment, vm_paddr_t boundary)
472 {
473 	void *ret;
474 
475 	ret = (void *)kmem_alloc_contig_domainset(ds, size, flags, low, high,
476 	    alignment, boundary, VM_MEMATTR_DEFAULT);
477 	if (ret != NULL)
478 		malloc_type_allocated(type, round_page(size));
479 	return (ret);
480 }
481 
482 /*
483  *	contigfree:
484  *
485  *	Free a block of memory allocated by contigmalloc.
486  *
487  *	This routine may not block.
488  */
489 void
490 contigfree(void *addr, unsigned long size, struct malloc_type *type)
491 {
492 
493 	kmem_free(addr, size);
494 	malloc_type_freed(type, round_page(size));
495 }
496 
497 #ifdef MALLOC_DEBUG
498 static int
499 malloc_dbg(caddr_t *vap, size_t *sizep, struct malloc_type *mtp,
500     int flags)
501 {
502 #ifdef INVARIANTS
503 	int indx;
504 
505 	KASSERT(mtp->ks_version == M_VERSION, ("malloc: bad malloc type version"));
506 	/*
507 	 * Check that exactly one of M_WAITOK or M_NOWAIT is specified.
508 	 */
509 	indx = flags & (M_WAITOK | M_NOWAIT);
510 	if (indx != M_NOWAIT && indx != M_WAITOK) {
511 		static	struct timeval lasterr;
512 		static	int curerr, once;
513 		if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) {
514 			printf("Bad malloc flags: %x\n", indx);
515 			kdb_backtrace();
516 			flags |= M_WAITOK;
517 			once++;
518 		}
519 	}
520 #endif
521 #ifdef MALLOC_MAKE_FAILURES
522 	if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
523 		atomic_add_int(&malloc_nowait_count, 1);
524 		if ((malloc_nowait_count % malloc_failure_rate) == 0) {
525 			atomic_add_int(&malloc_failure_count, 1);
526 			*vap = NULL;
527 			return (EJUSTRETURN);
528 		}
529 	}
530 #endif
531 	if (flags & M_WAITOK) {
532 		KASSERT(curthread->td_intr_nesting_level == 0,
533 		   ("malloc(M_WAITOK) in interrupt context"));
534 		if (__predict_false(!THREAD_CAN_SLEEP())) {
535 #ifdef EPOCH_TRACE
536 			epoch_trace_list(curthread);
537 #endif
538 			KASSERT(0,
539 			    ("malloc(M_WAITOK) with sleeping prohibited"));
540 		}
541 	}
542 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
543 	    ("malloc: called with spinlock or critical section held"));
544 
545 #ifdef DEBUG_MEMGUARD
546 	if (memguard_cmp_mtp(mtp, *sizep)) {
547 		*vap = memguard_alloc(*sizep, flags);
548 		if (*vap != NULL)
549 			return (EJUSTRETURN);
550 		/* This is unfortunate but should not be fatal. */
551 	}
552 #endif
553 
554 #ifdef DEBUG_REDZONE
555 	*sizep = redzone_size_ntor(*sizep);
556 #endif
557 
558 	return (0);
559 }
560 #endif
561 
562 /*
563  * Handle large allocations and frees by using kmem_malloc directly.
564  */
565 static inline bool
566 malloc_large_slab(uma_slab_t slab)
567 {
568 	uintptr_t va;
569 
570 	va = (uintptr_t)slab;
571 	return ((va & 1) != 0);
572 }
573 
574 static inline size_t
575 malloc_large_size(uma_slab_t slab)
576 {
577 	uintptr_t va;
578 
579 	va = (uintptr_t)slab;
580 	return (va >> 1);
581 }
582 
583 static caddr_t __noinline
584 malloc_large(size_t size, struct malloc_type *mtp, struct domainset *policy,
585     int flags DEBUG_REDZONE_ARG_DEF)
586 {
587 	void *va;
588 
589 	size = roundup(size, PAGE_SIZE);
590 	va = kmem_malloc_domainset(policy, size, flags);
591 	if (va != NULL) {
592 		/* The low bit is unused for slab pointers. */
593 		vsetzoneslab((uintptr_t)va, NULL, (void *)((size << 1) | 1));
594 		uma_total_inc(size);
595 	}
596 	malloc_type_allocated(mtp, va == NULL ? 0 : size);
597 	if (__predict_false(va == NULL)) {
598 		KASSERT((flags & M_WAITOK) == 0,
599 		    ("malloc(M_WAITOK) returned NULL"));
600 	} else {
601 #ifdef DEBUG_REDZONE
602 		va = redzone_setup(va, osize);
603 #endif
604 		kasan_mark(va, osize, size, KASAN_MALLOC_REDZONE);
605 	}
606 	return (va);
607 }
608 
609 static void
610 free_large(void *addr, size_t size)
611 {
612 
613 	kmem_free(addr, size);
614 	uma_total_dec(size);
615 }
616 
617 /*
618  *	malloc:
619  *
620  *	Allocate a block of memory.
621  *
622  *	If M_NOWAIT is set, this routine will not block and return NULL if
623  *	the allocation fails.
624  */
625 void *
626 (malloc)(size_t size, struct malloc_type *mtp, int flags)
627 {
628 	int indx;
629 	caddr_t va;
630 	uma_zone_t zone;
631 #if defined(DEBUG_REDZONE) || defined(KASAN)
632 	unsigned long osize = size;
633 #endif
634 
635 	MPASS((flags & M_EXEC) == 0);
636 
637 #ifdef MALLOC_DEBUG
638 	va = NULL;
639 	if (malloc_dbg(&va, &size, mtp, flags) != 0)
640 		return (va);
641 #endif
642 
643 	if (__predict_false(size > kmem_zmax))
644 		return (malloc_large(size, mtp, DOMAINSET_RR(), flags
645 		    DEBUG_REDZONE_ARG));
646 
647 	if (size & KMEM_ZMASK)
648 		size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
649 	indx = kmemsize[size >> KMEM_ZSHIFT];
650 	zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)];
651 	va = uma_zalloc_arg(zone, zone, flags);
652 	if (va != NULL) {
653 		size = zone->uz_size;
654 		if ((flags & M_ZERO) == 0) {
655 			kmsan_mark(va, size, KMSAN_STATE_UNINIT);
656 			kmsan_orig(va, size, KMSAN_TYPE_MALLOC, KMSAN_RET_ADDR);
657 		}
658 	}
659 	malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
660 	if (__predict_false(va == NULL)) {
661 		KASSERT((flags & M_WAITOK) == 0,
662 		    ("malloc(M_WAITOK) returned NULL"));
663 	}
664 #ifdef DEBUG_REDZONE
665 	if (va != NULL)
666 		va = redzone_setup(va, osize);
667 #endif
668 #ifdef KASAN
669 	if (va != NULL)
670 		kasan_mark((void *)va, osize, size, KASAN_MALLOC_REDZONE);
671 #endif
672 	return ((void *) va);
673 }
674 
675 static void *
676 malloc_domain(size_t *sizep, int *indxp, struct malloc_type *mtp, int domain,
677     int flags)
678 {
679 	uma_zone_t zone;
680 	caddr_t va;
681 	size_t size;
682 	int indx;
683 
684 	size = *sizep;
685 	KASSERT(size <= kmem_zmax && (flags & M_EXEC) == 0,
686 	    ("malloc_domain: Called with bad flag / size combination."));
687 	if (size & KMEM_ZMASK)
688 		size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
689 	indx = kmemsize[size >> KMEM_ZSHIFT];
690 	zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)];
691 	va = uma_zalloc_domain(zone, zone, domain, flags);
692 	if (va != NULL)
693 		*sizep = zone->uz_size;
694 	*indxp = indx;
695 	return ((void *)va);
696 }
697 
698 void *
699 malloc_domainset(size_t size, struct malloc_type *mtp, struct domainset *ds,
700     int flags)
701 {
702 	struct vm_domainset_iter di;
703 	caddr_t va;
704 	int domain;
705 	int indx;
706 #if defined(KASAN) || defined(DEBUG_REDZONE)
707 	unsigned long osize = size;
708 #endif
709 
710 	MPASS((flags & M_EXEC) == 0);
711 
712 #ifdef MALLOC_DEBUG
713 	va = NULL;
714 	if (malloc_dbg(&va, &size, mtp, flags) != 0)
715 		return (va);
716 #endif
717 
718 	if (__predict_false(size > kmem_zmax))
719 		return (malloc_large(size, mtp, DOMAINSET_RR(), flags
720 		    DEBUG_REDZONE_ARG));
721 
722 	vm_domainset_iter_policy_init(&di, ds, &domain, &flags);
723 	do {
724 		va = malloc_domain(&size, &indx, mtp, domain, flags);
725 	} while (va == NULL && vm_domainset_iter_policy(&di, &domain) == 0);
726 	malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
727 	if (__predict_false(va == NULL)) {
728 		KASSERT((flags & M_WAITOK) == 0,
729 		    ("malloc(M_WAITOK) returned NULL"));
730 	}
731 #ifdef DEBUG_REDZONE
732 	if (va != NULL)
733 		va = redzone_setup(va, osize);
734 #endif
735 #ifdef KASAN
736 	if (va != NULL)
737 		kasan_mark((void *)va, osize, size, KASAN_MALLOC_REDZONE);
738 #endif
739 #ifdef KMSAN
740 	if ((flags & M_ZERO) == 0) {
741 		kmsan_mark(va, size, KMSAN_STATE_UNINIT);
742 		kmsan_orig(va, size, KMSAN_TYPE_MALLOC, KMSAN_RET_ADDR);
743 	}
744 #endif
745 	return (va);
746 }
747 
748 /*
749  * Allocate an executable area.
750  */
751 void *
752 malloc_exec(size_t size, struct malloc_type *mtp, int flags)
753 {
754 
755 	return (malloc_domainset_exec(size, mtp, DOMAINSET_RR(), flags));
756 }
757 
758 void *
759 malloc_domainset_exec(size_t size, struct malloc_type *mtp, struct domainset *ds,
760     int flags)
761 {
762 #if defined(DEBUG_REDZONE) || defined(KASAN)
763 	unsigned long osize = size;
764 #endif
765 #ifdef MALLOC_DEBUG
766 	caddr_t va;
767 #endif
768 
769 	flags |= M_EXEC;
770 
771 #ifdef MALLOC_DEBUG
772 	va = NULL;
773 	if (malloc_dbg(&va, &size, mtp, flags) != 0)
774 		return (va);
775 #endif
776 
777 	return (malloc_large(size, mtp, ds, flags DEBUG_REDZONE_ARG));
778 }
779 
780 void *
781 malloc_aligned(size_t size, size_t align, struct malloc_type *type, int flags)
782 {
783 	return (malloc_domainset_aligned(size, align, type, DOMAINSET_RR(),
784 	    flags));
785 }
786 
787 void *
788 malloc_domainset_aligned(size_t size, size_t align,
789     struct malloc_type *mtp, struct domainset *ds, int flags)
790 {
791 	void *res;
792 	size_t asize;
793 
794 	KASSERT(powerof2(align),
795 	    ("malloc_domainset_aligned: wrong align %#zx size %#zx",
796 	    align, size));
797 	KASSERT(align <= PAGE_SIZE,
798 	    ("malloc_domainset_aligned: align %#zx (size %#zx) too large",
799 	    align, size));
800 
801 	/*
802 	 * Round the allocation size up to the next power of 2,
803 	 * because we can only guarantee alignment for
804 	 * power-of-2-sized allocations.  Further increase the
805 	 * allocation size to align if the rounded size is less than
806 	 * align, since malloc zones provide alignment equal to their
807 	 * size.
808 	 */
809 	if (size == 0)
810 		size = 1;
811 	asize = size <= align ? align : 1UL << flsl(size - 1);
812 
813 	res = malloc_domainset(asize, mtp, ds, flags);
814 	KASSERT(res == NULL || ((uintptr_t)res & (align - 1)) == 0,
815 	    ("malloc_domainset_aligned: result not aligned %p size %#zx "
816 	    "allocsize %#zx align %#zx", res, size, asize, align));
817 	return (res);
818 }
819 
820 void *
821 mallocarray(size_t nmemb, size_t size, struct malloc_type *type, int flags)
822 {
823 
824 	if (WOULD_OVERFLOW(nmemb, size))
825 		panic("mallocarray: %zu * %zu overflowed", nmemb, size);
826 
827 	return (malloc(size * nmemb, type, flags));
828 }
829 
830 void *
831 mallocarray_domainset(size_t nmemb, size_t size, struct malloc_type *type,
832     struct domainset *ds, int flags)
833 {
834 
835 	if (WOULD_OVERFLOW(nmemb, size))
836 		panic("mallocarray_domainset: %zu * %zu overflowed", nmemb, size);
837 
838 	return (malloc_domainset(size * nmemb, type, ds, flags));
839 }
840 
841 #if defined(INVARIANTS) && !defined(KASAN)
842 static void
843 free_save_type(void *addr, struct malloc_type *mtp, u_long size)
844 {
845 	struct malloc_type **mtpp = addr;
846 
847 	/*
848 	 * Cache a pointer to the malloc_type that most recently freed
849 	 * this memory here.  This way we know who is most likely to
850 	 * have stepped on it later.
851 	 *
852 	 * This code assumes that size is a multiple of 8 bytes for
853 	 * 64 bit machines
854 	 */
855 	mtpp = (struct malloc_type **) ((unsigned long)mtpp & ~UMA_ALIGN_PTR);
856 	mtpp += (size - sizeof(struct malloc_type *)) /
857 	    sizeof(struct malloc_type *);
858 	*mtpp = mtp;
859 }
860 #endif
861 
862 #ifdef MALLOC_DEBUG
863 static int
864 free_dbg(void **addrp, struct malloc_type *mtp)
865 {
866 	void *addr;
867 
868 	addr = *addrp;
869 	KASSERT(mtp->ks_version == M_VERSION, ("free: bad malloc type version"));
870 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
871 	    ("free: called with spinlock or critical section held"));
872 
873 	/* free(NULL, ...) does nothing */
874 	if (addr == NULL)
875 		return (EJUSTRETURN);
876 
877 #ifdef DEBUG_MEMGUARD
878 	if (is_memguard_addr(addr)) {
879 		memguard_free(addr);
880 		return (EJUSTRETURN);
881 	}
882 #endif
883 
884 #ifdef DEBUG_REDZONE
885 	redzone_check(addr);
886 	*addrp = redzone_addr_ntor(addr);
887 #endif
888 
889 	return (0);
890 }
891 #endif
892 
893 /*
894  *	free:
895  *
896  *	Free a block of memory allocated by malloc.
897  *
898  *	This routine may not block.
899  */
900 void
901 free(void *addr, struct malloc_type *mtp)
902 {
903 	uma_zone_t zone;
904 	uma_slab_t slab;
905 	u_long size;
906 
907 #ifdef MALLOC_DEBUG
908 	if (free_dbg(&addr, mtp) != 0)
909 		return;
910 #endif
911 	/* free(NULL, ...) does nothing */
912 	if (addr == NULL)
913 		return;
914 
915 	vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
916 	if (slab == NULL)
917 		panic("free: address %p(%p) has not been allocated.\n",
918 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
919 
920 	if (__predict_true(!malloc_large_slab(slab))) {
921 		size = zone->uz_size;
922 #if defined(INVARIANTS) && !defined(KASAN)
923 		free_save_type(addr, mtp, size);
924 #endif
925 		uma_zfree_arg(zone, addr, slab);
926 	} else {
927 		size = malloc_large_size(slab);
928 		free_large(addr, size);
929 	}
930 	malloc_type_freed(mtp, size);
931 }
932 
933 /*
934  *	zfree:
935  *
936  *	Zero then free a block of memory allocated by malloc.
937  *
938  *	This routine may not block.
939  */
940 void
941 zfree(void *addr, struct malloc_type *mtp)
942 {
943 	uma_zone_t zone;
944 	uma_slab_t slab;
945 	u_long size;
946 
947 #ifdef MALLOC_DEBUG
948 	if (free_dbg(&addr, mtp) != 0)
949 		return;
950 #endif
951 	/* free(NULL, ...) does nothing */
952 	if (addr == NULL)
953 		return;
954 
955 	vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
956 	if (slab == NULL)
957 		panic("free: address %p(%p) has not been allocated.\n",
958 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
959 
960 	if (__predict_true(!malloc_large_slab(slab))) {
961 		size = zone->uz_size;
962 #if defined(INVARIANTS) && !defined(KASAN)
963 		free_save_type(addr, mtp, size);
964 #endif
965 		kasan_mark(addr, size, size, 0);
966 		explicit_bzero(addr, size);
967 		uma_zfree_arg(zone, addr, slab);
968 	} else {
969 		size = malloc_large_size(slab);
970 		kasan_mark(addr, size, size, 0);
971 		explicit_bzero(addr, size);
972 		free_large(addr, size);
973 	}
974 	malloc_type_freed(mtp, size);
975 }
976 
977 /*
978  *	realloc: change the size of a memory block
979  */
980 void *
981 realloc(void *addr, size_t size, struct malloc_type *mtp, int flags)
982 {
983 #ifndef DEBUG_REDZONE
984 	uma_zone_t zone;
985 	uma_slab_t slab;
986 #endif
987 	unsigned long alloc;
988 	void *newaddr;
989 
990 	KASSERT(mtp->ks_version == M_VERSION,
991 	    ("realloc: bad malloc type version"));
992 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
993 	    ("realloc: called with spinlock or critical section held"));
994 
995 	/* realloc(NULL, ...) is equivalent to malloc(...) */
996 	if (addr == NULL)
997 		return (malloc(size, mtp, flags));
998 
999 	/*
1000 	 * XXX: Should report free of old memory and alloc of new memory to
1001 	 * per-CPU stats.
1002 	 */
1003 
1004 #ifdef DEBUG_MEMGUARD
1005 	if (is_memguard_addr(addr))
1006 		return (memguard_realloc(addr, size, mtp, flags));
1007 #endif
1008 
1009 #ifdef DEBUG_REDZONE
1010 	alloc = redzone_get_size(addr);
1011 #else
1012 	vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
1013 
1014 	/* Sanity check */
1015 	KASSERT(slab != NULL,
1016 	    ("realloc: address %p out of range", (void *)addr));
1017 
1018 	/* Get the size of the original block */
1019 	if (!malloc_large_slab(slab))
1020 		alloc = zone->uz_size;
1021 	else
1022 		alloc = malloc_large_size(slab);
1023 
1024 	/* Reuse the original block if appropriate */
1025 	if (size <= alloc &&
1026 	    (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE)) {
1027 		kasan_mark((void *)addr, size, alloc, KASAN_MALLOC_REDZONE);
1028 		return (addr);
1029 	}
1030 #endif /* !DEBUG_REDZONE */
1031 
1032 	/* Allocate a new, bigger (or smaller) block */
1033 	if ((newaddr = malloc(size, mtp, flags)) == NULL)
1034 		return (NULL);
1035 
1036 	/*
1037 	 * Copy over original contents.  For KASAN, the redzone must be marked
1038 	 * valid before performing the copy.
1039 	 */
1040 	kasan_mark(addr, alloc, alloc, 0);
1041 	bcopy(addr, newaddr, min(size, alloc));
1042 	free(addr, mtp);
1043 	return (newaddr);
1044 }
1045 
1046 /*
1047  *	reallocf: same as realloc() but free memory on failure.
1048  */
1049 void *
1050 reallocf(void *addr, size_t size, struct malloc_type *mtp, int flags)
1051 {
1052 	void *mem;
1053 
1054 	if ((mem = realloc(addr, size, mtp, flags)) == NULL)
1055 		free(addr, mtp);
1056 	return (mem);
1057 }
1058 
1059 /*
1060  * 	malloc_size: returns the number of bytes allocated for a request of the
1061  * 		     specified size
1062  */
1063 size_t
1064 malloc_size(size_t size)
1065 {
1066 	int indx;
1067 
1068 	if (size > kmem_zmax)
1069 		return (0);
1070 	if (size & KMEM_ZMASK)
1071 		size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
1072 	indx = kmemsize[size >> KMEM_ZSHIFT];
1073 	return (kmemzones[indx].kz_size);
1074 }
1075 
1076 /*
1077  *	malloc_usable_size: returns the usable size of the allocation.
1078  */
1079 size_t
1080 malloc_usable_size(const void *addr)
1081 {
1082 #ifndef DEBUG_REDZONE
1083 	uma_zone_t zone;
1084 	uma_slab_t slab;
1085 #endif
1086 	u_long size;
1087 
1088 	if (addr == NULL)
1089 		return (0);
1090 
1091 #ifdef DEBUG_MEMGUARD
1092 	if (is_memguard_addr(__DECONST(void *, addr)))
1093 		return (memguard_get_req_size(addr));
1094 #endif
1095 
1096 #ifdef DEBUG_REDZONE
1097 	size = redzone_get_size(__DECONST(void *, addr));
1098 #else
1099 	vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
1100 	if (slab == NULL)
1101 		panic("malloc_usable_size: address %p(%p) is not allocated.\n",
1102 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
1103 
1104 	if (!malloc_large_slab(slab))
1105 		size = zone->uz_size;
1106 	else
1107 		size = malloc_large_size(slab);
1108 #endif
1109 
1110 	/*
1111 	 * Unmark the redzone to avoid reports from consumers who are
1112 	 * (presumably) about to use the full allocation size.
1113 	 */
1114 	kasan_mark(addr, size, size, 0);
1115 
1116 	return (size);
1117 }
1118 
1119 CTASSERT(VM_KMEM_SIZE_SCALE >= 1);
1120 
1121 /*
1122  * Initialize the kernel memory (kmem) arena.
1123  */
1124 void
1125 kmeminit(void)
1126 {
1127 	u_long mem_size;
1128 	u_long tmp;
1129 
1130 #ifdef VM_KMEM_SIZE
1131 	if (vm_kmem_size == 0)
1132 		vm_kmem_size = VM_KMEM_SIZE;
1133 #endif
1134 #ifdef VM_KMEM_SIZE_MIN
1135 	if (vm_kmem_size_min == 0)
1136 		vm_kmem_size_min = VM_KMEM_SIZE_MIN;
1137 #endif
1138 #ifdef VM_KMEM_SIZE_MAX
1139 	if (vm_kmem_size_max == 0)
1140 		vm_kmem_size_max = VM_KMEM_SIZE_MAX;
1141 #endif
1142 	/*
1143 	 * Calculate the amount of kernel virtual address (KVA) space that is
1144 	 * preallocated to the kmem arena.  In order to support a wide range
1145 	 * of machines, it is a function of the physical memory size,
1146 	 * specifically,
1147 	 *
1148 	 *	min(max(physical memory size / VM_KMEM_SIZE_SCALE,
1149 	 *	    VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX)
1150 	 *
1151 	 * Every architecture must define an integral value for
1152 	 * VM_KMEM_SIZE_SCALE.  However, the definitions of VM_KMEM_SIZE_MIN
1153 	 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and
1154 	 * ceiling on this preallocation, are optional.  Typically,
1155 	 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on
1156 	 * a given architecture.
1157 	 */
1158 	mem_size = vm_cnt.v_page_count;
1159 	if (mem_size <= 32768) /* delphij XXX 128MB */
1160 		kmem_zmax = PAGE_SIZE;
1161 
1162 	if (vm_kmem_size_scale < 1)
1163 		vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
1164 
1165 	/*
1166 	 * Check if we should use defaults for the "vm_kmem_size"
1167 	 * variable:
1168 	 */
1169 	if (vm_kmem_size == 0) {
1170 		vm_kmem_size = mem_size / vm_kmem_size_scale;
1171 		vm_kmem_size = vm_kmem_size * PAGE_SIZE < vm_kmem_size ?
1172 		    vm_kmem_size_max : vm_kmem_size * PAGE_SIZE;
1173 		if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min)
1174 			vm_kmem_size = vm_kmem_size_min;
1175 		if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
1176 			vm_kmem_size = vm_kmem_size_max;
1177 	}
1178 	if (vm_kmem_size == 0)
1179 		panic("Tune VM_KMEM_SIZE_* for the platform");
1180 
1181 	/*
1182 	 * The amount of KVA space that is preallocated to the
1183 	 * kmem arena can be set statically at compile-time or manually
1184 	 * through the kernel environment.  However, it is still limited to
1185 	 * twice the physical memory size, which has been sufficient to handle
1186 	 * the most severe cases of external fragmentation in the kmem arena.
1187 	 */
1188 	if (vm_kmem_size / 2 / PAGE_SIZE > mem_size)
1189 		vm_kmem_size = 2 * mem_size * PAGE_SIZE;
1190 
1191 	vm_kmem_size = round_page(vm_kmem_size);
1192 
1193 	/*
1194 	 * With KASAN or KMSAN enabled, dynamically allocated kernel memory is
1195 	 * shadowed.  Account for this when setting the UMA limit.
1196 	 */
1197 #if defined(KASAN)
1198 	vm_kmem_size = (vm_kmem_size * KASAN_SHADOW_SCALE) /
1199 	    (KASAN_SHADOW_SCALE + 1);
1200 #elif defined(KMSAN)
1201 	vm_kmem_size /= 3;
1202 #endif
1203 
1204 #ifdef DEBUG_MEMGUARD
1205 	tmp = memguard_fudge(vm_kmem_size, kernel_map);
1206 #else
1207 	tmp = vm_kmem_size;
1208 #endif
1209 	uma_set_limit(tmp);
1210 
1211 #ifdef DEBUG_MEMGUARD
1212 	/*
1213 	 * Initialize MemGuard if support compiled in.  MemGuard is a
1214 	 * replacement allocator used for detecting tamper-after-free
1215 	 * scenarios as they occur.  It is only used for debugging.
1216 	 */
1217 	memguard_init(kernel_arena);
1218 #endif
1219 }
1220 
1221 /*
1222  * Initialize the kernel memory allocator
1223  */
1224 /* ARGSUSED*/
1225 static void
1226 mallocinit(void *dummy)
1227 {
1228 	int i;
1229 	uint8_t indx;
1230 
1231 	mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
1232 
1233 	kmeminit();
1234 
1235 	if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX)
1236 		kmem_zmax = KMEM_ZMAX;
1237 
1238 	for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
1239 		int size = kmemzones[indx].kz_size;
1240 		const char *name = kmemzones[indx].kz_name;
1241 		size_t align;
1242 		int subzone;
1243 
1244 		align = UMA_ALIGN_PTR;
1245 		if (powerof2(size) && size > sizeof(void *))
1246 			align = MIN(size, PAGE_SIZE) - 1;
1247 		for (subzone = 0; subzone < numzones; subzone++) {
1248 			kmemzones[indx].kz_zone[subzone] =
1249 			    uma_zcreate(name, size,
1250 #if defined(INVARIANTS) && !defined(KASAN) && !defined(KMSAN)
1251 			    mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
1252 #else
1253 			    NULL, NULL, NULL, NULL,
1254 #endif
1255 			    align, UMA_ZONE_MALLOC);
1256 		}
1257 		for (;i <= size; i+= KMEM_ZBASE)
1258 			kmemsize[i >> KMEM_ZSHIFT] = indx;
1259 	}
1260 }
1261 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_SECOND, mallocinit, NULL);
1262 
1263 void
1264 malloc_init(void *data)
1265 {
1266 	struct malloc_type_internal *mtip;
1267 	struct malloc_type *mtp;
1268 
1269 	KASSERT(vm_cnt.v_page_count != 0, ("malloc_register before vm_init"));
1270 
1271 	mtp = data;
1272 	if (mtp->ks_version != M_VERSION)
1273 		panic("malloc_init: type %s with unsupported version %lu",
1274 		    mtp->ks_shortdesc, mtp->ks_version);
1275 
1276 	mtip = &mtp->ks_mti;
1277 	mtip->mti_stats = uma_zalloc_pcpu(pcpu_zone_64, M_WAITOK | M_ZERO);
1278 	mtp_set_subzone(mtp);
1279 
1280 	mtx_lock(&malloc_mtx);
1281 	mtp->ks_next = kmemstatistics;
1282 	kmemstatistics = mtp;
1283 	kmemcount++;
1284 	mtx_unlock(&malloc_mtx);
1285 }
1286 
1287 void
1288 malloc_uninit(void *data)
1289 {
1290 	struct malloc_type_internal *mtip;
1291 	struct malloc_type_stats *mtsp;
1292 	struct malloc_type *mtp, *temp;
1293 	long temp_allocs, temp_bytes;
1294 	int i;
1295 
1296 	mtp = data;
1297 	KASSERT(mtp->ks_version == M_VERSION,
1298 	    ("malloc_uninit: bad malloc type version"));
1299 
1300 	mtx_lock(&malloc_mtx);
1301 	mtip = &mtp->ks_mti;
1302 	if (mtp != kmemstatistics) {
1303 		for (temp = kmemstatistics; temp != NULL;
1304 		    temp = temp->ks_next) {
1305 			if (temp->ks_next == mtp) {
1306 				temp->ks_next = mtp->ks_next;
1307 				break;
1308 			}
1309 		}
1310 		KASSERT(temp,
1311 		    ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc));
1312 	} else
1313 		kmemstatistics = mtp->ks_next;
1314 	kmemcount--;
1315 	mtx_unlock(&malloc_mtx);
1316 
1317 	/*
1318 	 * Look for memory leaks.
1319 	 */
1320 	temp_allocs = temp_bytes = 0;
1321 	for (i = 0; i <= mp_maxid; i++) {
1322 		mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1323 		temp_allocs += mtsp->mts_numallocs;
1324 		temp_allocs -= mtsp->mts_numfrees;
1325 		temp_bytes += mtsp->mts_memalloced;
1326 		temp_bytes -= mtsp->mts_memfreed;
1327 	}
1328 	if (temp_allocs > 0 || temp_bytes > 0) {
1329 		printf("Warning: memory type %s leaked memory on destroy "
1330 		    "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc,
1331 		    temp_allocs, temp_bytes);
1332 	}
1333 
1334 	uma_zfree_pcpu(pcpu_zone_64, mtip->mti_stats);
1335 }
1336 
1337 struct malloc_type *
1338 malloc_desc2type(const char *desc)
1339 {
1340 	struct malloc_type *mtp;
1341 
1342 	mtx_assert(&malloc_mtx, MA_OWNED);
1343 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1344 		if (strcmp(mtp->ks_shortdesc, desc) == 0)
1345 			return (mtp);
1346 	}
1347 	return (NULL);
1348 }
1349 
1350 static int
1351 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
1352 {
1353 	struct malloc_type_stream_header mtsh;
1354 	struct malloc_type_internal *mtip;
1355 	struct malloc_type_stats *mtsp, zeromts;
1356 	struct malloc_type_header mth;
1357 	struct malloc_type *mtp;
1358 	int error, i;
1359 	struct sbuf sbuf;
1360 
1361 	error = sysctl_wire_old_buffer(req, 0);
1362 	if (error != 0)
1363 		return (error);
1364 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
1365 	sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL);
1366 	mtx_lock(&malloc_mtx);
1367 
1368 	bzero(&zeromts, sizeof(zeromts));
1369 
1370 	/*
1371 	 * Insert stream header.
1372 	 */
1373 	bzero(&mtsh, sizeof(mtsh));
1374 	mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
1375 	mtsh.mtsh_maxcpus = MAXCPU;
1376 	mtsh.mtsh_count = kmemcount;
1377 	(void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh));
1378 
1379 	/*
1380 	 * Insert alternating sequence of type headers and type statistics.
1381 	 */
1382 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1383 		mtip = &mtp->ks_mti;
1384 
1385 		/*
1386 		 * Insert type header.
1387 		 */
1388 		bzero(&mth, sizeof(mth));
1389 		strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
1390 		(void)sbuf_bcat(&sbuf, &mth, sizeof(mth));
1391 
1392 		/*
1393 		 * Insert type statistics for each CPU.
1394 		 */
1395 		for (i = 0; i <= mp_maxid; i++) {
1396 			mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1397 			(void)sbuf_bcat(&sbuf, mtsp, sizeof(*mtsp));
1398 		}
1399 		/*
1400 		 * Fill in the missing CPUs.
1401 		 */
1402 		for (; i < MAXCPU; i++) {
1403 			(void)sbuf_bcat(&sbuf, &zeromts, sizeof(zeromts));
1404 		}
1405 	}
1406 	mtx_unlock(&malloc_mtx);
1407 	error = sbuf_finish(&sbuf);
1408 	sbuf_delete(&sbuf);
1409 	return (error);
1410 }
1411 
1412 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats,
1413     CTLFLAG_RD | CTLTYPE_STRUCT | CTLFLAG_MPSAFE, 0, 0,
1414     sysctl_kern_malloc_stats, "s,malloc_type_ustats",
1415     "Return malloc types");
1416 
1417 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
1418     "Count of kernel malloc types");
1419 
1420 void
1421 malloc_type_list(malloc_type_list_func_t *func, void *arg)
1422 {
1423 	struct malloc_type *mtp, **bufmtp;
1424 	int count, i;
1425 	size_t buflen;
1426 
1427 	mtx_lock(&malloc_mtx);
1428 restart:
1429 	mtx_assert(&malloc_mtx, MA_OWNED);
1430 	count = kmemcount;
1431 	mtx_unlock(&malloc_mtx);
1432 
1433 	buflen = sizeof(struct malloc_type *) * count;
1434 	bufmtp = malloc(buflen, M_TEMP, M_WAITOK);
1435 
1436 	mtx_lock(&malloc_mtx);
1437 
1438 	if (count < kmemcount) {
1439 		free(bufmtp, M_TEMP);
1440 		goto restart;
1441 	}
1442 
1443 	for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++)
1444 		bufmtp[i] = mtp;
1445 
1446 	mtx_unlock(&malloc_mtx);
1447 
1448 	for (i = 0; i < count; i++)
1449 		(func)(bufmtp[i], arg);
1450 
1451 	free(bufmtp, M_TEMP);
1452 }
1453 
1454 #ifdef DDB
1455 static int64_t
1456 get_malloc_stats(const struct malloc_type_internal *mtip, uint64_t *allocs,
1457     uint64_t *inuse)
1458 {
1459 	const struct malloc_type_stats *mtsp;
1460 	uint64_t frees, alloced, freed;
1461 	int i;
1462 
1463 	*allocs = 0;
1464 	frees = 0;
1465 	alloced = 0;
1466 	freed = 0;
1467 	for (i = 0; i <= mp_maxid; i++) {
1468 		mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1469 
1470 		*allocs += mtsp->mts_numallocs;
1471 		frees += mtsp->mts_numfrees;
1472 		alloced += mtsp->mts_memalloced;
1473 		freed += mtsp->mts_memfreed;
1474 	}
1475 	*inuse = *allocs - frees;
1476 	return (alloced - freed);
1477 }
1478 
1479 DB_SHOW_COMMAND_FLAGS(malloc, db_show_malloc, DB_CMD_MEMSAFE)
1480 {
1481 	const char *fmt_hdr, *fmt_entry;
1482 	struct malloc_type *mtp;
1483 	uint64_t allocs, inuse;
1484 	int64_t size;
1485 	/* variables for sorting */
1486 	struct malloc_type *last_mtype, *cur_mtype;
1487 	int64_t cur_size, last_size;
1488 	int ties;
1489 
1490 	if (modif[0] == 'i') {
1491 		fmt_hdr = "%s,%s,%s,%s\n";
1492 		fmt_entry = "\"%s\",%ju,%jdK,%ju\n";
1493 	} else {
1494 		fmt_hdr = "%18s %12s  %12s %12s\n";
1495 		fmt_entry = "%18s %12ju %12jdK %12ju\n";
1496 	}
1497 
1498 	db_printf(fmt_hdr, "Type", "InUse", "MemUse", "Requests");
1499 
1500 	/* Select sort, largest size first. */
1501 	last_mtype = NULL;
1502 	last_size = INT64_MAX;
1503 	for (;;) {
1504 		cur_mtype = NULL;
1505 		cur_size = -1;
1506 		ties = 0;
1507 
1508 		for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1509 			/*
1510 			 * In the case of size ties, print out mtypes
1511 			 * in the order they are encountered.  That is,
1512 			 * when we encounter the most recently output
1513 			 * mtype, we have already printed all preceding
1514 			 * ties, and we must print all following ties.
1515 			 */
1516 			if (mtp == last_mtype) {
1517 				ties = 1;
1518 				continue;
1519 			}
1520 			size = get_malloc_stats(&mtp->ks_mti, &allocs,
1521 			    &inuse);
1522 			if (size > cur_size && size < last_size + ties) {
1523 				cur_size = size;
1524 				cur_mtype = mtp;
1525 			}
1526 		}
1527 		if (cur_mtype == NULL)
1528 			break;
1529 
1530 		size = get_malloc_stats(&cur_mtype->ks_mti, &allocs, &inuse);
1531 		db_printf(fmt_entry, cur_mtype->ks_shortdesc, inuse,
1532 		    howmany(size, 1024), allocs);
1533 
1534 		if (db_pager_quit)
1535 			break;
1536 
1537 		last_mtype = cur_mtype;
1538 		last_size = cur_size;
1539 	}
1540 }
1541 
1542 #if MALLOC_DEBUG_MAXZONES > 1
1543 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches)
1544 {
1545 	struct malloc_type_internal *mtip;
1546 	struct malloc_type *mtp;
1547 	u_int subzone;
1548 
1549 	if (!have_addr) {
1550 		db_printf("Usage: show multizone_matches <malloc type/addr>\n");
1551 		return;
1552 	}
1553 	mtp = (void *)addr;
1554 	if (mtp->ks_version != M_VERSION) {
1555 		db_printf("Version %lx does not match expected %x\n",
1556 		    mtp->ks_version, M_VERSION);
1557 		return;
1558 	}
1559 
1560 	mtip = &mtp->ks_mti;
1561 	subzone = mtip->mti_zone;
1562 
1563 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1564 		mtip = &mtp->ks_mti;
1565 		if (mtip->mti_zone != subzone)
1566 			continue;
1567 		db_printf("%s\n", mtp->ks_shortdesc);
1568 		if (db_pager_quit)
1569 			break;
1570 	}
1571 }
1572 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
1573 #endif /* DDB */
1574