xref: /freebsd/sys/kern/kern_malloc.c (revision 22cf89c938886d14f5796fc49f9f020c23ea8eaf)
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  *	@(#)kern_malloc.c	8.3 (Berkeley) 1/4/94
35  */
36 
37 /*
38  * Kernel malloc(9) implementation -- general purpose kernel memory allocator
39  * based on memory types.  Back end is implemented using the UMA(9) zone
40  * allocator.  A set of fixed-size buckets are used for smaller allocations,
41  * and a special UMA allocation interface is used for larger allocations.
42  * Callers declare memory types, and statistics are maintained independently
43  * for each memory type.  Statistics are maintained per-CPU for performance
44  * reasons.  See malloc(9) and comments in malloc.h for a detailed
45  * description.
46  */
47 
48 #include <sys/cdefs.h>
49 #include "opt_ddb.h"
50 #include "opt_vm.h"
51 
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/asan.h>
55 #include <sys/kdb.h>
56 #include <sys/kernel.h>
57 #include <sys/lock.h>
58 #include <sys/malloc.h>
59 #include <sys/msan.h>
60 #include <sys/mutex.h>
61 #include <sys/vmmeter.h>
62 #include <sys/proc.h>
63 #include <sys/queue.h>
64 #include <sys/sbuf.h>
65 #include <sys/smp.h>
66 #include <sys/sysctl.h>
67 #include <sys/time.h>
68 #include <sys/vmem.h>
69 #ifdef EPOCH_TRACE
70 #include <sys/epoch.h>
71 #endif
72 
73 #include <vm/vm.h>
74 #include <vm/pmap.h>
75 #include <vm/vm_domainset.h>
76 #include <vm/vm_pageout.h>
77 #include <vm/vm_param.h>
78 #include <vm/vm_kern.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_page.h>
82 #include <vm/vm_phys.h>
83 #include <vm/vm_pagequeue.h>
84 #include <vm/uma.h>
85 #include <vm/uma_int.h>
86 #include <vm/uma_dbg.h>
87 
88 #ifdef DEBUG_MEMGUARD
89 #include <vm/memguard.h>
90 #endif
91 #ifdef DEBUG_REDZONE
92 #include <vm/redzone.h>
93 #endif
94 
95 #if defined(INVARIANTS) && defined(__i386__)
96 #include <machine/cpu.h>
97 #endif
98 
99 #include <ddb/ddb.h>
100 
101 #ifdef KDTRACE_HOOKS
102 #include <sys/dtrace_bsd.h>
103 
104 bool	__read_frequently			dtrace_malloc_enabled;
105 dtrace_malloc_probe_func_t __read_mostly	dtrace_malloc_probe;
106 #endif
107 
108 #if defined(INVARIANTS) || defined(MALLOC_MAKE_FAILURES) ||		\
109     defined(DEBUG_MEMGUARD) || defined(DEBUG_REDZONE)
110 #define	MALLOC_DEBUG	1
111 #endif
112 
113 #if defined(KASAN) || defined(DEBUG_REDZONE)
114 #define	DEBUG_REDZONE_ARG_DEF	, unsigned long osize
115 #define	DEBUG_REDZONE_ARG	, osize
116 #else
117 #define	DEBUG_REDZONE_ARG_DEF
118 #define	DEBUG_REDZONE_ARG
119 #endif
120 
121 /*
122  * When realloc() is called, if the new size is sufficiently smaller than
123  * the old size, realloc() will allocate a new, smaller block to avoid
124  * wasting memory. 'Sufficiently smaller' is defined as: newsize <=
125  * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'.
126  */
127 #ifndef REALLOC_FRACTION
128 #define	REALLOC_FRACTION	1	/* new block if <= half the size */
129 #endif
130 
131 /*
132  * Centrally define some common malloc types.
133  */
134 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches");
135 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
136 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers");
137 
138 static struct malloc_type *kmemstatistics;
139 static int kmemcount;
140 
141 #define KMEM_ZSHIFT	4
142 #define KMEM_ZBASE	16
143 #define KMEM_ZMASK	(KMEM_ZBASE - 1)
144 
145 #define KMEM_ZMAX	65536
146 #define KMEM_ZSIZE	(KMEM_ZMAX >> KMEM_ZSHIFT)
147 static uint8_t kmemsize[KMEM_ZSIZE + 1];
148 
149 #ifndef MALLOC_DEBUG_MAXZONES
150 #define	MALLOC_DEBUG_MAXZONES	1
151 #endif
152 static int numzones = MALLOC_DEBUG_MAXZONES;
153 
154 /*
155  * Small malloc(9) memory allocations are allocated from a set of UMA buckets
156  * of various sizes.
157  *
158  * Warning: the layout of the struct is duplicated in libmemstat for KVM support.
159  *
160  * XXX: The comment here used to read "These won't be powers of two for
161  * long."  It's possible that a significant amount of wasted memory could be
162  * recovered by tuning the sizes of these buckets.
163  */
164 struct {
165 	int kz_size;
166 	const char *kz_name;
167 	uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES];
168 } kmemzones[] = {
169 	{16, "malloc-16", },
170 	{32, "malloc-32", },
171 	{64, "malloc-64", },
172 	{128, "malloc-128", },
173 	{256, "malloc-256", },
174 	{384, "malloc-384", },
175 	{512, "malloc-512", },
176 	{1024, "malloc-1024", },
177 	{2048, "malloc-2048", },
178 	{4096, "malloc-4096", },
179 	{8192, "malloc-8192", },
180 	{16384, "malloc-16384", },
181 	{32768, "malloc-32768", },
182 	{65536, "malloc-65536", },
183 	{0, NULL},
184 };
185 
186 u_long vm_kmem_size;
187 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0,
188     "Size of kernel memory");
189 
190 static u_long kmem_zmax = KMEM_ZMAX;
191 SYSCTL_ULONG(_vm, OID_AUTO, kmem_zmax, CTLFLAG_RDTUN, &kmem_zmax, 0,
192     "Maximum allocation size that malloc(9) would use UMA as backend");
193 
194 static u_long vm_kmem_size_min;
195 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0,
196     "Minimum size of kernel memory");
197 
198 static u_long vm_kmem_size_max;
199 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0,
200     "Maximum size of kernel memory");
201 
202 static u_int vm_kmem_size_scale;
203 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0,
204     "Scale factor for kernel memory size");
205 
206 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS);
207 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size,
208     CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
209     sysctl_kmem_map_size, "LU", "Current kmem allocation size");
210 
211 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS);
212 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free,
213     CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
214     sysctl_kmem_map_free, "LU", "Free space in kmem");
215 
216 static SYSCTL_NODE(_vm, OID_AUTO, malloc, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
217     "Malloc information");
218 
219 static u_int vm_malloc_zone_count = nitems(kmemzones);
220 SYSCTL_UINT(_vm_malloc, OID_AUTO, zone_count,
221     CTLFLAG_RD, &vm_malloc_zone_count, 0,
222     "Number of malloc zones");
223 
224 static int sysctl_vm_malloc_zone_sizes(SYSCTL_HANDLER_ARGS);
225 SYSCTL_PROC(_vm_malloc, OID_AUTO, zone_sizes,
226     CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, NULL, 0,
227     sysctl_vm_malloc_zone_sizes, "S", "Zone sizes used by malloc");
228 
229 /*
230  * The malloc_mtx protects the kmemstatistics linked list.
231  */
232 struct mtx malloc_mtx;
233 
234 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS);
235 
236 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1)
237 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
238     "Kernel malloc debugging options");
239 #endif
240 
241 /*
242  * malloc(9) fault injection -- cause malloc failures every (n) mallocs when
243  * the caller specifies M_NOWAIT.  If set to 0, no failures are caused.
244  */
245 #ifdef MALLOC_MAKE_FAILURES
246 static int malloc_failure_rate;
247 static int malloc_nowait_count;
248 static int malloc_failure_count;
249 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RWTUN,
250     &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail");
251 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD,
252     &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures");
253 #endif
254 
255 static int
256 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS)
257 {
258 	u_long size;
259 
260 	size = uma_size();
261 	return (sysctl_handle_long(oidp, &size, 0, req));
262 }
263 
264 static int
265 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS)
266 {
267 	u_long size, limit;
268 
269 	/* The sysctl is unsigned, implement as a saturation value. */
270 	size = uma_size();
271 	limit = uma_limit();
272 	if (size > limit)
273 		size = 0;
274 	else
275 		size = limit - size;
276 	return (sysctl_handle_long(oidp, &size, 0, req));
277 }
278 
279 static int
280 sysctl_vm_malloc_zone_sizes(SYSCTL_HANDLER_ARGS)
281 {
282 	int sizes[nitems(kmemzones)];
283 	int i;
284 
285 	for (i = 0; i < nitems(kmemzones); i++) {
286 		sizes[i] = kmemzones[i].kz_size;
287 	}
288 
289 	return (SYSCTL_OUT(req, &sizes, sizeof(sizes)));
290 }
291 
292 /*
293  * malloc(9) uma zone separation -- sub-page buffer overruns in one
294  * malloc type will affect only a subset of other malloc types.
295  */
296 #if MALLOC_DEBUG_MAXZONES > 1
297 static void
298 tunable_set_numzones(void)
299 {
300 
301 	TUNABLE_INT_FETCH("debug.malloc.numzones",
302 	    &numzones);
303 
304 	/* Sanity check the number of malloc uma zones. */
305 	if (numzones <= 0)
306 		numzones = 1;
307 	if (numzones > MALLOC_DEBUG_MAXZONES)
308 		numzones = MALLOC_DEBUG_MAXZONES;
309 }
310 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL);
311 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
312     &numzones, 0, "Number of malloc uma subzones");
313 
314 /*
315  * Any number that changes regularly is an okay choice for the
316  * offset.  Build numbers are pretty good of you have them.
317  */
318 static u_int zone_offset = __FreeBSD_version;
319 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset);
320 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN,
321     &zone_offset, 0, "Separate malloc types by examining the "
322     "Nth character in the malloc type short description.");
323 
324 static void
325 mtp_set_subzone(struct malloc_type *mtp)
326 {
327 	struct malloc_type_internal *mtip;
328 	const char *desc;
329 	size_t len;
330 	u_int val;
331 
332 	mtip = &mtp->ks_mti;
333 	desc = mtp->ks_shortdesc;
334 	if (desc == NULL || (len = strlen(desc)) == 0)
335 		val = 0;
336 	else
337 		val = desc[zone_offset % len];
338 	mtip->mti_zone = (val % numzones);
339 }
340 
341 static inline u_int
342 mtp_get_subzone(struct malloc_type *mtp)
343 {
344 	struct malloc_type_internal *mtip;
345 
346 	mtip = &mtp->ks_mti;
347 
348 	KASSERT(mtip->mti_zone < numzones,
349 	    ("mti_zone %u out of range %d",
350 	    mtip->mti_zone, numzones));
351 	return (mtip->mti_zone);
352 }
353 #elif MALLOC_DEBUG_MAXZONES == 0
354 #error "MALLOC_DEBUG_MAXZONES must be positive."
355 #else
356 static void
357 mtp_set_subzone(struct malloc_type *mtp)
358 {
359 	struct malloc_type_internal *mtip;
360 
361 	mtip = &mtp->ks_mti;
362 	mtip->mti_zone = 0;
363 }
364 
365 static inline u_int
366 mtp_get_subzone(struct malloc_type *mtp)
367 {
368 
369 	return (0);
370 }
371 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
372 
373 /*
374  * An allocation has succeeded -- update malloc type statistics for the
375  * amount of bucket size.  Occurs within a critical section so that the
376  * thread isn't preempted and doesn't migrate while updating per-PCU
377  * statistics.
378  */
379 static void
380 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size,
381     int zindx)
382 {
383 	struct malloc_type_internal *mtip;
384 	struct malloc_type_stats *mtsp;
385 
386 	critical_enter();
387 	mtip = &mtp->ks_mti;
388 	mtsp = zpcpu_get(mtip->mti_stats);
389 	if (size > 0) {
390 		mtsp->mts_memalloced += size;
391 		mtsp->mts_numallocs++;
392 	}
393 	if (zindx != -1)
394 		mtsp->mts_size |= 1 << zindx;
395 
396 #ifdef KDTRACE_HOOKS
397 	if (__predict_false(dtrace_malloc_enabled)) {
398 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC];
399 		if (probe_id != 0)
400 			(dtrace_malloc_probe)(probe_id,
401 			    (uintptr_t) mtp, (uintptr_t) mtip,
402 			    (uintptr_t) mtsp, size, zindx);
403 	}
404 #endif
405 
406 	critical_exit();
407 }
408 
409 void
410 malloc_type_allocated(struct malloc_type *mtp, unsigned long size)
411 {
412 
413 	if (size > 0)
414 		malloc_type_zone_allocated(mtp, size, -1);
415 }
416 
417 /*
418  * A free operation has occurred -- update malloc type statistics for the
419  * amount of the bucket size.  Occurs within a critical section so that the
420  * thread isn't preempted and doesn't migrate while updating per-CPU
421  * statistics.
422  */
423 void
424 malloc_type_freed(struct malloc_type *mtp, unsigned long size)
425 {
426 	struct malloc_type_internal *mtip;
427 	struct malloc_type_stats *mtsp;
428 
429 	critical_enter();
430 	mtip = &mtp->ks_mti;
431 	mtsp = zpcpu_get(mtip->mti_stats);
432 	mtsp->mts_memfreed += size;
433 	mtsp->mts_numfrees++;
434 
435 #ifdef KDTRACE_HOOKS
436 	if (__predict_false(dtrace_malloc_enabled)) {
437 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE];
438 		if (probe_id != 0)
439 			(dtrace_malloc_probe)(probe_id,
440 			    (uintptr_t) mtp, (uintptr_t) mtip,
441 			    (uintptr_t) mtsp, size, 0);
442 	}
443 #endif
444 
445 	critical_exit();
446 }
447 
448 /*
449  *	contigmalloc:
450  *
451  *	Allocate a block of physically contiguous memory.
452  *
453  *	If M_NOWAIT is set, this routine will not block and return NULL if
454  *	the allocation fails.
455  */
456 void *
457 contigmalloc(unsigned long size, struct malloc_type *type, int flags,
458     vm_paddr_t low, vm_paddr_t high, unsigned long alignment,
459     vm_paddr_t boundary)
460 {
461 	void *ret;
462 
463 	ret = (void *)kmem_alloc_contig(size, flags, low, high, alignment,
464 	    boundary, VM_MEMATTR_DEFAULT);
465 	if (ret != NULL)
466 		malloc_type_allocated(type, round_page(size));
467 	return (ret);
468 }
469 
470 void *
471 contigmalloc_domainset(unsigned long size, struct malloc_type *type,
472     struct domainset *ds, int flags, vm_paddr_t low, vm_paddr_t high,
473     unsigned long alignment, vm_paddr_t boundary)
474 {
475 	void *ret;
476 
477 	ret = (void *)kmem_alloc_contig_domainset(ds, size, flags, low, high,
478 	    alignment, boundary, VM_MEMATTR_DEFAULT);
479 	if (ret != NULL)
480 		malloc_type_allocated(type, round_page(size));
481 	return (ret);
482 }
483 
484 /*
485  *	contigfree:
486  *
487  *	Free a block of memory allocated by contigmalloc.
488  *
489  *	This routine may not block.
490  */
491 void
492 contigfree(void *addr, unsigned long size, struct malloc_type *type)
493 {
494 
495 	kmem_free(addr, size);
496 	malloc_type_freed(type, round_page(size));
497 }
498 
499 #ifdef MALLOC_DEBUG
500 static int
501 malloc_dbg(caddr_t *vap, size_t *sizep, struct malloc_type *mtp,
502     int flags)
503 {
504 #ifdef INVARIANTS
505 	int indx;
506 
507 	KASSERT(mtp->ks_version == M_VERSION, ("malloc: bad malloc type version"));
508 	/*
509 	 * Check that exactly one of M_WAITOK or M_NOWAIT is specified.
510 	 */
511 	indx = flags & (M_WAITOK | M_NOWAIT);
512 	if (indx != M_NOWAIT && indx != M_WAITOK) {
513 		static	struct timeval lasterr;
514 		static	int curerr, once;
515 		if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) {
516 			printf("Bad malloc flags: %x\n", indx);
517 			kdb_backtrace();
518 			flags |= M_WAITOK;
519 			once++;
520 		}
521 	}
522 #endif
523 #ifdef MALLOC_MAKE_FAILURES
524 	if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
525 		atomic_add_int(&malloc_nowait_count, 1);
526 		if ((malloc_nowait_count % malloc_failure_rate) == 0) {
527 			atomic_add_int(&malloc_failure_count, 1);
528 			*vap = NULL;
529 			return (EJUSTRETURN);
530 		}
531 	}
532 #endif
533 	if (flags & M_WAITOK) {
534 		KASSERT(curthread->td_intr_nesting_level == 0,
535 		   ("malloc(M_WAITOK) in interrupt context"));
536 		if (__predict_false(!THREAD_CAN_SLEEP())) {
537 #ifdef EPOCH_TRACE
538 			epoch_trace_list(curthread);
539 #endif
540 			KASSERT(0,
541 			    ("malloc(M_WAITOK) with sleeping prohibited"));
542 		}
543 	}
544 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
545 	    ("malloc: called with spinlock or critical section held"));
546 
547 #ifdef DEBUG_MEMGUARD
548 	if (memguard_cmp_mtp(mtp, *sizep)) {
549 		*vap = memguard_alloc(*sizep, flags);
550 		if (*vap != NULL)
551 			return (EJUSTRETURN);
552 		/* This is unfortunate but should not be fatal. */
553 	}
554 #endif
555 
556 #ifdef DEBUG_REDZONE
557 	*sizep = redzone_size_ntor(*sizep);
558 #endif
559 
560 	return (0);
561 }
562 #endif
563 
564 /*
565  * Handle large allocations and frees by using kmem_malloc directly.
566  */
567 static inline bool
568 malloc_large_slab(uma_slab_t slab)
569 {
570 	uintptr_t va;
571 
572 	va = (uintptr_t)slab;
573 	return ((va & 1) != 0);
574 }
575 
576 static inline size_t
577 malloc_large_size(uma_slab_t slab)
578 {
579 	uintptr_t va;
580 
581 	va = (uintptr_t)slab;
582 	return (va >> 1);
583 }
584 
585 static caddr_t __noinline
586 malloc_large(size_t size, struct malloc_type *mtp, struct domainset *policy,
587     int flags DEBUG_REDZONE_ARG_DEF)
588 {
589 	void *va;
590 
591 	size = roundup(size, PAGE_SIZE);
592 	va = kmem_malloc_domainset(policy, size, flags);
593 	if (va != NULL) {
594 		/* The low bit is unused for slab pointers. */
595 		vsetzoneslab((uintptr_t)va, NULL, (void *)((size << 1) | 1));
596 		uma_total_inc(size);
597 	}
598 	malloc_type_allocated(mtp, va == NULL ? 0 : size);
599 	if (__predict_false(va == NULL)) {
600 		KASSERT((flags & M_WAITOK) == 0,
601 		    ("malloc(M_WAITOK) returned NULL"));
602 	} else {
603 #ifdef DEBUG_REDZONE
604 		va = redzone_setup(va, osize);
605 #endif
606 		kasan_mark(va, osize, size, KASAN_MALLOC_REDZONE);
607 	}
608 	return (va);
609 }
610 
611 static void
612 free_large(void *addr, size_t size)
613 {
614 
615 	kmem_free(addr, size);
616 	uma_total_dec(size);
617 }
618 
619 /*
620  *	malloc:
621  *
622  *	Allocate a block of memory.
623  *
624  *	If M_NOWAIT is set, this routine will not block and return NULL if
625  *	the allocation fails.
626  */
627 void *
628 (malloc)(size_t size, struct malloc_type *mtp, int flags)
629 {
630 	int indx;
631 	caddr_t va;
632 	uma_zone_t zone;
633 #if defined(DEBUG_REDZONE) || defined(KASAN)
634 	unsigned long osize = size;
635 #endif
636 
637 	MPASS((flags & M_EXEC) == 0);
638 
639 #ifdef MALLOC_DEBUG
640 	va = NULL;
641 	if (malloc_dbg(&va, &size, mtp, flags) != 0)
642 		return (va);
643 #endif
644 
645 	if (__predict_false(size > kmem_zmax))
646 		return (malloc_large(size, mtp, DOMAINSET_RR(), flags
647 		    DEBUG_REDZONE_ARG));
648 
649 	if (size & KMEM_ZMASK)
650 		size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
651 	indx = kmemsize[size >> KMEM_ZSHIFT];
652 	zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)];
653 	va = uma_zalloc(zone, flags);
654 	if (va != NULL) {
655 		size = zone->uz_size;
656 		if ((flags & M_ZERO) == 0) {
657 			kmsan_mark(va, size, KMSAN_STATE_UNINIT);
658 			kmsan_orig(va, size, KMSAN_TYPE_MALLOC, KMSAN_RET_ADDR);
659 		}
660 	}
661 	malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
662 	if (__predict_false(va == NULL)) {
663 		KASSERT((flags & M_WAITOK) == 0,
664 		    ("malloc(M_WAITOK) returned NULL"));
665 	}
666 #ifdef DEBUG_REDZONE
667 	if (va != NULL)
668 		va = redzone_setup(va, osize);
669 #endif
670 #ifdef KASAN
671 	if (va != NULL)
672 		kasan_mark((void *)va, osize, size, KASAN_MALLOC_REDZONE);
673 #endif
674 	return ((void *) va);
675 }
676 
677 static void *
678 malloc_domain(size_t *sizep, int *indxp, struct malloc_type *mtp, int domain,
679     int flags)
680 {
681 	uma_zone_t zone;
682 	caddr_t va;
683 	size_t size;
684 	int indx;
685 
686 	size = *sizep;
687 	KASSERT(size <= kmem_zmax && (flags & M_EXEC) == 0,
688 	    ("malloc_domain: Called with bad flag / size combination."));
689 	if (size & KMEM_ZMASK)
690 		size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
691 	indx = kmemsize[size >> KMEM_ZSHIFT];
692 	zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)];
693 	va = uma_zalloc_domain(zone, NULL, domain, flags);
694 	if (va != NULL)
695 		*sizep = zone->uz_size;
696 	*indxp = indx;
697 	return ((void *)va);
698 }
699 
700 void *
701 malloc_domainset(size_t size, struct malloc_type *mtp, struct domainset *ds,
702     int flags)
703 {
704 	struct vm_domainset_iter di;
705 	caddr_t va;
706 	int domain;
707 	int indx;
708 #if defined(KASAN) || defined(DEBUG_REDZONE)
709 	unsigned long osize = size;
710 #endif
711 
712 	MPASS((flags & M_EXEC) == 0);
713 
714 #ifdef MALLOC_DEBUG
715 	va = NULL;
716 	if (malloc_dbg(&va, &size, mtp, flags) != 0)
717 		return (va);
718 #endif
719 
720 	if (__predict_false(size > kmem_zmax))
721 		return (malloc_large(size, mtp, DOMAINSET_RR(), flags
722 		    DEBUG_REDZONE_ARG));
723 
724 	vm_domainset_iter_policy_init(&di, ds, &domain, &flags);
725 	do {
726 		va = malloc_domain(&size, &indx, mtp, domain, flags);
727 	} while (va == NULL && vm_domainset_iter_policy(&di, &domain) == 0);
728 	malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
729 	if (__predict_false(va == NULL)) {
730 		KASSERT((flags & M_WAITOK) == 0,
731 		    ("malloc(M_WAITOK) returned NULL"));
732 	}
733 #ifdef DEBUG_REDZONE
734 	if (va != NULL)
735 		va = redzone_setup(va, osize);
736 #endif
737 #ifdef KASAN
738 	if (va != NULL)
739 		kasan_mark((void *)va, osize, size, KASAN_MALLOC_REDZONE);
740 #endif
741 #ifdef KMSAN
742 	if ((flags & M_ZERO) == 0) {
743 		kmsan_mark(va, size, KMSAN_STATE_UNINIT);
744 		kmsan_orig(va, size, KMSAN_TYPE_MALLOC, KMSAN_RET_ADDR);
745 	}
746 #endif
747 	return (va);
748 }
749 
750 /*
751  * Allocate an executable area.
752  */
753 void *
754 malloc_exec(size_t size, struct malloc_type *mtp, int flags)
755 {
756 
757 	return (malloc_domainset_exec(size, mtp, DOMAINSET_RR(), flags));
758 }
759 
760 void *
761 malloc_domainset_exec(size_t size, struct malloc_type *mtp, struct domainset *ds,
762     int flags)
763 {
764 #if defined(DEBUG_REDZONE) || defined(KASAN)
765 	unsigned long osize = size;
766 #endif
767 #ifdef MALLOC_DEBUG
768 	caddr_t va;
769 #endif
770 
771 	flags |= M_EXEC;
772 
773 #ifdef MALLOC_DEBUG
774 	va = NULL;
775 	if (malloc_dbg(&va, &size, mtp, flags) != 0)
776 		return (va);
777 #endif
778 
779 	return (malloc_large(size, mtp, ds, flags DEBUG_REDZONE_ARG));
780 }
781 
782 void *
783 malloc_aligned(size_t size, size_t align, struct malloc_type *type, int flags)
784 {
785 	return (malloc_domainset_aligned(size, align, type, DOMAINSET_RR(),
786 	    flags));
787 }
788 
789 void *
790 malloc_domainset_aligned(size_t size, size_t align,
791     struct malloc_type *mtp, struct domainset *ds, int flags)
792 {
793 	void *res;
794 	size_t asize;
795 
796 	KASSERT(powerof2(align),
797 	    ("malloc_domainset_aligned: wrong align %#zx size %#zx",
798 	    align, size));
799 	KASSERT(align <= PAGE_SIZE,
800 	    ("malloc_domainset_aligned: align %#zx (size %#zx) too large",
801 	    align, size));
802 
803 	/*
804 	 * Round the allocation size up to the next power of 2,
805 	 * because we can only guarantee alignment for
806 	 * power-of-2-sized allocations.  Further increase the
807 	 * allocation size to align if the rounded size is less than
808 	 * align, since malloc zones provide alignment equal to their
809 	 * size.
810 	 */
811 	if (size == 0)
812 		size = 1;
813 	asize = size <= align ? align : 1UL << flsl(size - 1);
814 
815 	res = malloc_domainset(asize, mtp, ds, flags);
816 	KASSERT(res == NULL || ((uintptr_t)res & (align - 1)) == 0,
817 	    ("malloc_domainset_aligned: result not aligned %p size %#zx "
818 	    "allocsize %#zx align %#zx", res, size, asize, align));
819 	return (res);
820 }
821 
822 void *
823 mallocarray(size_t nmemb, size_t size, struct malloc_type *type, int flags)
824 {
825 
826 	if (WOULD_OVERFLOW(nmemb, size))
827 		panic("mallocarray: %zu * %zu overflowed", nmemb, size);
828 
829 	return (malloc(size * nmemb, type, flags));
830 }
831 
832 void *
833 mallocarray_domainset(size_t nmemb, size_t size, struct malloc_type *type,
834     struct domainset *ds, int flags)
835 {
836 
837 	if (WOULD_OVERFLOW(nmemb, size))
838 		panic("mallocarray_domainset: %zu * %zu overflowed", nmemb, size);
839 
840 	return (malloc_domainset(size * nmemb, type, ds, flags));
841 }
842 
843 #if defined(INVARIANTS) && !defined(KASAN)
844 static void
845 free_save_type(void *addr, struct malloc_type *mtp, u_long size)
846 {
847 	struct malloc_type **mtpp = addr;
848 
849 	/*
850 	 * Cache a pointer to the malloc_type that most recently freed
851 	 * this memory here.  This way we know who is most likely to
852 	 * have stepped on it later.
853 	 *
854 	 * This code assumes that size is a multiple of 8 bytes for
855 	 * 64 bit machines
856 	 */
857 	mtpp = (struct malloc_type **) ((unsigned long)mtpp & ~UMA_ALIGN_PTR);
858 	mtpp += (size - sizeof(struct malloc_type *)) /
859 	    sizeof(struct malloc_type *);
860 	*mtpp = mtp;
861 }
862 #endif
863 
864 #ifdef MALLOC_DEBUG
865 static int
866 free_dbg(void **addrp, struct malloc_type *mtp)
867 {
868 	void *addr;
869 
870 	addr = *addrp;
871 	KASSERT(mtp->ks_version == M_VERSION, ("free: bad malloc type version"));
872 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
873 	    ("free: called with spinlock or critical section held"));
874 
875 	/* free(NULL, ...) does nothing */
876 	if (addr == NULL)
877 		return (EJUSTRETURN);
878 
879 #ifdef DEBUG_MEMGUARD
880 	if (is_memguard_addr(addr)) {
881 		memguard_free(addr);
882 		return (EJUSTRETURN);
883 	}
884 #endif
885 
886 #ifdef DEBUG_REDZONE
887 	redzone_check(addr);
888 	*addrp = redzone_addr_ntor(addr);
889 #endif
890 
891 	return (0);
892 }
893 #endif
894 
895 /*
896  *	free:
897  *
898  *	Free a block of memory allocated by malloc.
899  *
900  *	This routine may not block.
901  */
902 void
903 free(void *addr, struct malloc_type *mtp)
904 {
905 	uma_zone_t zone;
906 	uma_slab_t slab;
907 	u_long size;
908 
909 #ifdef MALLOC_DEBUG
910 	if (free_dbg(&addr, mtp) != 0)
911 		return;
912 #endif
913 	/* free(NULL, ...) does nothing */
914 	if (addr == NULL)
915 		return;
916 
917 	vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
918 	if (slab == NULL)
919 		panic("free: address %p(%p) has not been allocated.\n",
920 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
921 
922 	if (__predict_true(!malloc_large_slab(slab))) {
923 		size = zone->uz_size;
924 #if defined(INVARIANTS) && !defined(KASAN)
925 		free_save_type(addr, mtp, size);
926 #endif
927 		uma_zfree_arg(zone, addr, slab);
928 	} else {
929 		size = malloc_large_size(slab);
930 		free_large(addr, size);
931 	}
932 	malloc_type_freed(mtp, size);
933 }
934 
935 /*
936  *	zfree:
937  *
938  *	Zero then free a block of memory allocated by malloc.
939  *
940  *	This routine may not block.
941  */
942 void
943 zfree(void *addr, struct malloc_type *mtp)
944 {
945 	uma_zone_t zone;
946 	uma_slab_t slab;
947 	u_long size;
948 
949 #ifdef MALLOC_DEBUG
950 	if (free_dbg(&addr, mtp) != 0)
951 		return;
952 #endif
953 	/* free(NULL, ...) does nothing */
954 	if (addr == NULL)
955 		return;
956 
957 	vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
958 	if (slab == NULL)
959 		panic("free: address %p(%p) has not been allocated.\n",
960 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
961 
962 	if (__predict_true(!malloc_large_slab(slab))) {
963 		size = zone->uz_size;
964 #if defined(INVARIANTS) && !defined(KASAN)
965 		free_save_type(addr, mtp, size);
966 #endif
967 		kasan_mark(addr, size, size, 0);
968 		explicit_bzero(addr, size);
969 		uma_zfree_arg(zone, addr, slab);
970 	} else {
971 		size = malloc_large_size(slab);
972 		kasan_mark(addr, size, size, 0);
973 		explicit_bzero(addr, size);
974 		free_large(addr, size);
975 	}
976 	malloc_type_freed(mtp, size);
977 }
978 
979 /*
980  *	realloc: change the size of a memory block
981  */
982 void *
983 realloc(void *addr, size_t size, struct malloc_type *mtp, int flags)
984 {
985 #ifndef DEBUG_REDZONE
986 	uma_zone_t zone;
987 	uma_slab_t slab;
988 #endif
989 	unsigned long alloc;
990 	void *newaddr;
991 
992 	KASSERT(mtp->ks_version == M_VERSION,
993 	    ("realloc: bad malloc type version"));
994 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
995 	    ("realloc: called with spinlock or critical section held"));
996 
997 	/* realloc(NULL, ...) is equivalent to malloc(...) */
998 	if (addr == NULL)
999 		return (malloc(size, mtp, flags));
1000 
1001 	/*
1002 	 * XXX: Should report free of old memory and alloc of new memory to
1003 	 * per-CPU stats.
1004 	 */
1005 
1006 #ifdef DEBUG_MEMGUARD
1007 	if (is_memguard_addr(addr))
1008 		return (memguard_realloc(addr, size, mtp, flags));
1009 #endif
1010 
1011 #ifdef DEBUG_REDZONE
1012 	alloc = redzone_get_size(addr);
1013 #else
1014 	vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
1015 
1016 	/* Sanity check */
1017 	KASSERT(slab != NULL,
1018 	    ("realloc: address %p out of range", (void *)addr));
1019 
1020 	/* Get the size of the original block */
1021 	if (!malloc_large_slab(slab))
1022 		alloc = zone->uz_size;
1023 	else
1024 		alloc = malloc_large_size(slab);
1025 
1026 	/* Reuse the original block if appropriate */
1027 	if (size <= alloc &&
1028 	    (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE)) {
1029 		kasan_mark((void *)addr, size, alloc, KASAN_MALLOC_REDZONE);
1030 		return (addr);
1031 	}
1032 #endif /* !DEBUG_REDZONE */
1033 
1034 	/* Allocate a new, bigger (or smaller) block */
1035 	if ((newaddr = malloc(size, mtp, flags)) == NULL)
1036 		return (NULL);
1037 
1038 	/*
1039 	 * Copy over original contents.  For KASAN, the redzone must be marked
1040 	 * valid before performing the copy.
1041 	 */
1042 	kasan_mark(addr, alloc, alloc, 0);
1043 	bcopy(addr, newaddr, min(size, alloc));
1044 	free(addr, mtp);
1045 	return (newaddr);
1046 }
1047 
1048 /*
1049  *	reallocf: same as realloc() but free memory on failure.
1050  */
1051 void *
1052 reallocf(void *addr, size_t size, struct malloc_type *mtp, int flags)
1053 {
1054 	void *mem;
1055 
1056 	if ((mem = realloc(addr, size, mtp, flags)) == NULL)
1057 		free(addr, mtp);
1058 	return (mem);
1059 }
1060 
1061 /*
1062  * 	malloc_size: returns the number of bytes allocated for a request of the
1063  * 		     specified size
1064  */
1065 size_t
1066 malloc_size(size_t size)
1067 {
1068 	int indx;
1069 
1070 	if (size > kmem_zmax)
1071 		return (0);
1072 	if (size & KMEM_ZMASK)
1073 		size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
1074 	indx = kmemsize[size >> KMEM_ZSHIFT];
1075 	return (kmemzones[indx].kz_size);
1076 }
1077 
1078 /*
1079  *	malloc_usable_size: returns the usable size of the allocation.
1080  */
1081 size_t
1082 malloc_usable_size(const void *addr)
1083 {
1084 #ifndef DEBUG_REDZONE
1085 	uma_zone_t zone;
1086 	uma_slab_t slab;
1087 #endif
1088 	u_long size;
1089 
1090 	if (addr == NULL)
1091 		return (0);
1092 
1093 #ifdef DEBUG_MEMGUARD
1094 	if (is_memguard_addr(__DECONST(void *, addr)))
1095 		return (memguard_get_req_size(addr));
1096 #endif
1097 
1098 #ifdef DEBUG_REDZONE
1099 	size = redzone_get_size(__DECONST(void *, addr));
1100 #else
1101 	vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
1102 	if (slab == NULL)
1103 		panic("malloc_usable_size: address %p(%p) is not allocated.\n",
1104 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
1105 
1106 	if (!malloc_large_slab(slab))
1107 		size = zone->uz_size;
1108 	else
1109 		size = malloc_large_size(slab);
1110 #endif
1111 
1112 	/*
1113 	 * Unmark the redzone to avoid reports from consumers who are
1114 	 * (presumably) about to use the full allocation size.
1115 	 */
1116 	kasan_mark(addr, size, size, 0);
1117 
1118 	return (size);
1119 }
1120 
1121 CTASSERT(VM_KMEM_SIZE_SCALE >= 1);
1122 
1123 /*
1124  * Initialize the kernel memory (kmem) arena.
1125  */
1126 void
1127 kmeminit(void)
1128 {
1129 	u_long mem_size;
1130 	u_long tmp;
1131 
1132 #ifdef VM_KMEM_SIZE
1133 	if (vm_kmem_size == 0)
1134 		vm_kmem_size = VM_KMEM_SIZE;
1135 #endif
1136 #ifdef VM_KMEM_SIZE_MIN
1137 	if (vm_kmem_size_min == 0)
1138 		vm_kmem_size_min = VM_KMEM_SIZE_MIN;
1139 #endif
1140 #ifdef VM_KMEM_SIZE_MAX
1141 	if (vm_kmem_size_max == 0)
1142 		vm_kmem_size_max = VM_KMEM_SIZE_MAX;
1143 #endif
1144 	/*
1145 	 * Calculate the amount of kernel virtual address (KVA) space that is
1146 	 * preallocated to the kmem arena.  In order to support a wide range
1147 	 * of machines, it is a function of the physical memory size,
1148 	 * specifically,
1149 	 *
1150 	 *	min(max(physical memory size / VM_KMEM_SIZE_SCALE,
1151 	 *	    VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX)
1152 	 *
1153 	 * Every architecture must define an integral value for
1154 	 * VM_KMEM_SIZE_SCALE.  However, the definitions of VM_KMEM_SIZE_MIN
1155 	 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and
1156 	 * ceiling on this preallocation, are optional.  Typically,
1157 	 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on
1158 	 * a given architecture.
1159 	 */
1160 	mem_size = vm_cnt.v_page_count;
1161 	if (mem_size <= 32768) /* delphij XXX 128MB */
1162 		kmem_zmax = PAGE_SIZE;
1163 
1164 	if (vm_kmem_size_scale < 1)
1165 		vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
1166 
1167 	/*
1168 	 * Check if we should use defaults for the "vm_kmem_size"
1169 	 * variable:
1170 	 */
1171 	if (vm_kmem_size == 0) {
1172 		vm_kmem_size = mem_size / vm_kmem_size_scale;
1173 		vm_kmem_size = vm_kmem_size * PAGE_SIZE < vm_kmem_size ?
1174 		    vm_kmem_size_max : vm_kmem_size * PAGE_SIZE;
1175 		if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min)
1176 			vm_kmem_size = vm_kmem_size_min;
1177 		if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
1178 			vm_kmem_size = vm_kmem_size_max;
1179 	}
1180 	if (vm_kmem_size == 0)
1181 		panic("Tune VM_KMEM_SIZE_* for the platform");
1182 
1183 	/*
1184 	 * The amount of KVA space that is preallocated to the
1185 	 * kmem arena can be set statically at compile-time or manually
1186 	 * through the kernel environment.  However, it is still limited to
1187 	 * twice the physical memory size, which has been sufficient to handle
1188 	 * the most severe cases of external fragmentation in the kmem arena.
1189 	 */
1190 	if (vm_kmem_size / 2 / PAGE_SIZE > mem_size)
1191 		vm_kmem_size = 2 * mem_size * PAGE_SIZE;
1192 
1193 	vm_kmem_size = round_page(vm_kmem_size);
1194 
1195 	/*
1196 	 * With KASAN or KMSAN enabled, dynamically allocated kernel memory is
1197 	 * shadowed.  Account for this when setting the UMA limit.
1198 	 */
1199 #if defined(KASAN)
1200 	vm_kmem_size = (vm_kmem_size * KASAN_SHADOW_SCALE) /
1201 	    (KASAN_SHADOW_SCALE + 1);
1202 #elif defined(KMSAN)
1203 	vm_kmem_size /= 3;
1204 #endif
1205 
1206 #ifdef DEBUG_MEMGUARD
1207 	tmp = memguard_fudge(vm_kmem_size, kernel_map);
1208 #else
1209 	tmp = vm_kmem_size;
1210 #endif
1211 	uma_set_limit(tmp);
1212 
1213 #ifdef DEBUG_MEMGUARD
1214 	/*
1215 	 * Initialize MemGuard if support compiled in.  MemGuard is a
1216 	 * replacement allocator used for detecting tamper-after-free
1217 	 * scenarios as they occur.  It is only used for debugging.
1218 	 */
1219 	memguard_init(kernel_arena);
1220 #endif
1221 }
1222 
1223 /*
1224  * Initialize the kernel memory allocator
1225  */
1226 /* ARGSUSED*/
1227 static void
1228 mallocinit(void *dummy)
1229 {
1230 	int i;
1231 	uint8_t indx;
1232 
1233 	mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
1234 
1235 	kmeminit();
1236 
1237 	if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX)
1238 		kmem_zmax = KMEM_ZMAX;
1239 
1240 	for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
1241 		int size = kmemzones[indx].kz_size;
1242 		const char *name = kmemzones[indx].kz_name;
1243 		size_t align;
1244 		int subzone;
1245 
1246 		align = UMA_ALIGN_PTR;
1247 		if (powerof2(size) && size > sizeof(void *))
1248 			align = MIN(size, PAGE_SIZE) - 1;
1249 		for (subzone = 0; subzone < numzones; subzone++) {
1250 			kmemzones[indx].kz_zone[subzone] =
1251 			    uma_zcreate(name, size,
1252 #if defined(INVARIANTS) && !defined(KASAN) && !defined(KMSAN)
1253 			    mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
1254 #else
1255 			    NULL, NULL, NULL, NULL,
1256 #endif
1257 			    align, UMA_ZONE_MALLOC);
1258 		}
1259 		for (;i <= size; i+= KMEM_ZBASE)
1260 			kmemsize[i >> KMEM_ZSHIFT] = indx;
1261 	}
1262 }
1263 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_SECOND, mallocinit, NULL);
1264 
1265 void
1266 malloc_init(void *data)
1267 {
1268 	struct malloc_type_internal *mtip;
1269 	struct malloc_type *mtp;
1270 
1271 	KASSERT(vm_cnt.v_page_count != 0, ("malloc_register before vm_init"));
1272 
1273 	mtp = data;
1274 	if (mtp->ks_version != M_VERSION)
1275 		panic("malloc_init: type %s with unsupported version %lu",
1276 		    mtp->ks_shortdesc, mtp->ks_version);
1277 
1278 	mtip = &mtp->ks_mti;
1279 	mtip->mti_stats = uma_zalloc_pcpu(pcpu_zone_64, M_WAITOK | M_ZERO);
1280 	mtp_set_subzone(mtp);
1281 
1282 	mtx_lock(&malloc_mtx);
1283 	mtp->ks_next = kmemstatistics;
1284 	kmemstatistics = mtp;
1285 	kmemcount++;
1286 	mtx_unlock(&malloc_mtx);
1287 }
1288 
1289 void
1290 malloc_uninit(void *data)
1291 {
1292 	struct malloc_type_internal *mtip;
1293 	struct malloc_type_stats *mtsp;
1294 	struct malloc_type *mtp, *temp;
1295 	long temp_allocs, temp_bytes;
1296 	int i;
1297 
1298 	mtp = data;
1299 	KASSERT(mtp->ks_version == M_VERSION,
1300 	    ("malloc_uninit: bad malloc type version"));
1301 
1302 	mtx_lock(&malloc_mtx);
1303 	mtip = &mtp->ks_mti;
1304 	if (mtp != kmemstatistics) {
1305 		for (temp = kmemstatistics; temp != NULL;
1306 		    temp = temp->ks_next) {
1307 			if (temp->ks_next == mtp) {
1308 				temp->ks_next = mtp->ks_next;
1309 				break;
1310 			}
1311 		}
1312 		KASSERT(temp,
1313 		    ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc));
1314 	} else
1315 		kmemstatistics = mtp->ks_next;
1316 	kmemcount--;
1317 	mtx_unlock(&malloc_mtx);
1318 
1319 	/*
1320 	 * Look for memory leaks.
1321 	 */
1322 	temp_allocs = temp_bytes = 0;
1323 	for (i = 0; i <= mp_maxid; i++) {
1324 		mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1325 		temp_allocs += mtsp->mts_numallocs;
1326 		temp_allocs -= mtsp->mts_numfrees;
1327 		temp_bytes += mtsp->mts_memalloced;
1328 		temp_bytes -= mtsp->mts_memfreed;
1329 	}
1330 	if (temp_allocs > 0 || temp_bytes > 0) {
1331 		printf("Warning: memory type %s leaked memory on destroy "
1332 		    "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc,
1333 		    temp_allocs, temp_bytes);
1334 	}
1335 
1336 	uma_zfree_pcpu(pcpu_zone_64, mtip->mti_stats);
1337 }
1338 
1339 struct malloc_type *
1340 malloc_desc2type(const char *desc)
1341 {
1342 	struct malloc_type *mtp;
1343 
1344 	mtx_assert(&malloc_mtx, MA_OWNED);
1345 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1346 		if (strcmp(mtp->ks_shortdesc, desc) == 0)
1347 			return (mtp);
1348 	}
1349 	return (NULL);
1350 }
1351 
1352 static int
1353 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
1354 {
1355 	struct malloc_type_stream_header mtsh;
1356 	struct malloc_type_internal *mtip;
1357 	struct malloc_type_stats *mtsp, zeromts;
1358 	struct malloc_type_header mth;
1359 	struct malloc_type *mtp;
1360 	int error, i;
1361 	struct sbuf sbuf;
1362 
1363 	error = sysctl_wire_old_buffer(req, 0);
1364 	if (error != 0)
1365 		return (error);
1366 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
1367 	sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL);
1368 	mtx_lock(&malloc_mtx);
1369 
1370 	bzero(&zeromts, sizeof(zeromts));
1371 
1372 	/*
1373 	 * Insert stream header.
1374 	 */
1375 	bzero(&mtsh, sizeof(mtsh));
1376 	mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
1377 	mtsh.mtsh_maxcpus = MAXCPU;
1378 	mtsh.mtsh_count = kmemcount;
1379 	(void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh));
1380 
1381 	/*
1382 	 * Insert alternating sequence of type headers and type statistics.
1383 	 */
1384 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1385 		mtip = &mtp->ks_mti;
1386 
1387 		/*
1388 		 * Insert type header.
1389 		 */
1390 		bzero(&mth, sizeof(mth));
1391 		strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
1392 		(void)sbuf_bcat(&sbuf, &mth, sizeof(mth));
1393 
1394 		/*
1395 		 * Insert type statistics for each CPU.
1396 		 */
1397 		for (i = 0; i <= mp_maxid; i++) {
1398 			mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1399 			(void)sbuf_bcat(&sbuf, mtsp, sizeof(*mtsp));
1400 		}
1401 		/*
1402 		 * Fill in the missing CPUs.
1403 		 */
1404 		for (; i < MAXCPU; i++) {
1405 			(void)sbuf_bcat(&sbuf, &zeromts, sizeof(zeromts));
1406 		}
1407 	}
1408 	mtx_unlock(&malloc_mtx);
1409 	error = sbuf_finish(&sbuf);
1410 	sbuf_delete(&sbuf);
1411 	return (error);
1412 }
1413 
1414 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats,
1415     CTLFLAG_RD | CTLTYPE_STRUCT | CTLFLAG_MPSAFE, 0, 0,
1416     sysctl_kern_malloc_stats, "s,malloc_type_ustats",
1417     "Return malloc types");
1418 
1419 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
1420     "Count of kernel malloc types");
1421 
1422 void
1423 malloc_type_list(malloc_type_list_func_t *func, void *arg)
1424 {
1425 	struct malloc_type *mtp, **bufmtp;
1426 	int count, i;
1427 	size_t buflen;
1428 
1429 	mtx_lock(&malloc_mtx);
1430 restart:
1431 	mtx_assert(&malloc_mtx, MA_OWNED);
1432 	count = kmemcount;
1433 	mtx_unlock(&malloc_mtx);
1434 
1435 	buflen = sizeof(struct malloc_type *) * count;
1436 	bufmtp = malloc(buflen, M_TEMP, M_WAITOK);
1437 
1438 	mtx_lock(&malloc_mtx);
1439 
1440 	if (count < kmemcount) {
1441 		free(bufmtp, M_TEMP);
1442 		goto restart;
1443 	}
1444 
1445 	for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++)
1446 		bufmtp[i] = mtp;
1447 
1448 	mtx_unlock(&malloc_mtx);
1449 
1450 	for (i = 0; i < count; i++)
1451 		(func)(bufmtp[i], arg);
1452 
1453 	free(bufmtp, M_TEMP);
1454 }
1455 
1456 #ifdef DDB
1457 static int64_t
1458 get_malloc_stats(const struct malloc_type_internal *mtip, uint64_t *allocs,
1459     uint64_t *inuse)
1460 {
1461 	const struct malloc_type_stats *mtsp;
1462 	uint64_t frees, alloced, freed;
1463 	int i;
1464 
1465 	*allocs = 0;
1466 	frees = 0;
1467 	alloced = 0;
1468 	freed = 0;
1469 	for (i = 0; i <= mp_maxid; i++) {
1470 		mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1471 
1472 		*allocs += mtsp->mts_numallocs;
1473 		frees += mtsp->mts_numfrees;
1474 		alloced += mtsp->mts_memalloced;
1475 		freed += mtsp->mts_memfreed;
1476 	}
1477 	*inuse = *allocs - frees;
1478 	return (alloced - freed);
1479 }
1480 
1481 DB_SHOW_COMMAND_FLAGS(malloc, db_show_malloc, DB_CMD_MEMSAFE)
1482 {
1483 	const char *fmt_hdr, *fmt_entry;
1484 	struct malloc_type *mtp;
1485 	uint64_t allocs, inuse;
1486 	int64_t size;
1487 	/* variables for sorting */
1488 	struct malloc_type *last_mtype, *cur_mtype;
1489 	int64_t cur_size, last_size;
1490 	int ties;
1491 
1492 	if (modif[0] == 'i') {
1493 		fmt_hdr = "%s,%s,%s,%s\n";
1494 		fmt_entry = "\"%s\",%ju,%jdK,%ju\n";
1495 	} else {
1496 		fmt_hdr = "%18s %12s  %12s %12s\n";
1497 		fmt_entry = "%18s %12ju %12jdK %12ju\n";
1498 	}
1499 
1500 	db_printf(fmt_hdr, "Type", "InUse", "MemUse", "Requests");
1501 
1502 	/* Select sort, largest size first. */
1503 	last_mtype = NULL;
1504 	last_size = INT64_MAX;
1505 	for (;;) {
1506 		cur_mtype = NULL;
1507 		cur_size = -1;
1508 		ties = 0;
1509 
1510 		for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1511 			/*
1512 			 * In the case of size ties, print out mtypes
1513 			 * in the order they are encountered.  That is,
1514 			 * when we encounter the most recently output
1515 			 * mtype, we have already printed all preceding
1516 			 * ties, and we must print all following ties.
1517 			 */
1518 			if (mtp == last_mtype) {
1519 				ties = 1;
1520 				continue;
1521 			}
1522 			size = get_malloc_stats(&mtp->ks_mti, &allocs,
1523 			    &inuse);
1524 			if (size > cur_size && size < last_size + ties) {
1525 				cur_size = size;
1526 				cur_mtype = mtp;
1527 			}
1528 		}
1529 		if (cur_mtype == NULL)
1530 			break;
1531 
1532 		size = get_malloc_stats(&cur_mtype->ks_mti, &allocs, &inuse);
1533 		db_printf(fmt_entry, cur_mtype->ks_shortdesc, inuse,
1534 		    howmany(size, 1024), allocs);
1535 
1536 		if (db_pager_quit)
1537 			break;
1538 
1539 		last_mtype = cur_mtype;
1540 		last_size = cur_size;
1541 	}
1542 }
1543 
1544 #if MALLOC_DEBUG_MAXZONES > 1
1545 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches)
1546 {
1547 	struct malloc_type_internal *mtip;
1548 	struct malloc_type *mtp;
1549 	u_int subzone;
1550 
1551 	if (!have_addr) {
1552 		db_printf("Usage: show multizone_matches <malloc type/addr>\n");
1553 		return;
1554 	}
1555 	mtp = (void *)addr;
1556 	if (mtp->ks_version != M_VERSION) {
1557 		db_printf("Version %lx does not match expected %x\n",
1558 		    mtp->ks_version, M_VERSION);
1559 		return;
1560 	}
1561 
1562 	mtip = &mtp->ks_mti;
1563 	subzone = mtip->mti_zone;
1564 
1565 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1566 		mtip = &mtp->ks_mti;
1567 		if (mtip->mti_zone != subzone)
1568 			continue;
1569 		db_printf("%s\n", mtp->ks_shortdesc);
1570 		if (db_pager_quit)
1571 			break;
1572 	}
1573 }
1574 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
1575 #endif /* DDB */
1576