xref: /freebsd/sys/kern/kern_malloc.c (revision d0f9b0bd19bf5dc4031164fd26c5bbc4bab1cc74)
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 typedef	enum {
120 	SLAB_COOKIE_SLAB_PTR		= 0x0,
121 	SLAB_COOKIE_MALLOC_LARGE	= 0x1,
122 	SLAB_COOKIE_CONTIG_MALLOC	= 0x2,
123 } slab_cookie_t;
124 #define	SLAB_COOKIE_MASK		0x3
125 #define	SLAB_COOKIE_SHIFT		2
126 #define	GET_SLAB_COOKIE(_slab)						\
127     ((slab_cookie_t)(uintptr_t)(_slab) & SLAB_COOKIE_MASK)
128 
129 /*
130  * When realloc() is called, if the new size is sufficiently smaller than
131  * the old size, realloc() will allocate a new, smaller block to avoid
132  * wasting memory. 'Sufficiently smaller' is defined as: newsize <=
133  * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'.
134  */
135 #ifndef REALLOC_FRACTION
136 #define	REALLOC_FRACTION	1	/* new block if <= half the size */
137 #endif
138 
139 /*
140  * Centrally define some common malloc types.
141  */
142 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches");
143 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
144 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers");
145 
146 static struct malloc_type *kmemstatistics;
147 static int kmemcount;
148 
149 #define KMEM_ZSHIFT	4
150 #define KMEM_ZBASE	16
151 #define KMEM_ZMASK	(KMEM_ZBASE - 1)
152 
153 #define KMEM_ZMAX	65536
154 #define KMEM_ZSIZE	(KMEM_ZMAX >> KMEM_ZSHIFT)
155 static uint8_t kmemsize[KMEM_ZSIZE + 1];
156 
157 #ifndef MALLOC_DEBUG_MAXZONES
158 #define	MALLOC_DEBUG_MAXZONES	1
159 #endif
160 static int numzones = MALLOC_DEBUG_MAXZONES;
161 
162 /*
163  * Small malloc(9) memory allocations are allocated from a set of UMA buckets
164  * of various sizes.
165  *
166  * Warning: the layout of the struct is duplicated in libmemstat for KVM support.
167  *
168  * XXX: The comment here used to read "These won't be powers of two for
169  * long."  It's possible that a significant amount of wasted memory could be
170  * recovered by tuning the sizes of these buckets.
171  */
172 struct {
173 	int kz_size;
174 	const char *kz_name;
175 	uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES];
176 } kmemzones[] = {
177 	{16, "malloc-16", },
178 	{32, "malloc-32", },
179 	{64, "malloc-64", },
180 	{128, "malloc-128", },
181 	{256, "malloc-256", },
182 	{384, "malloc-384", },
183 	{512, "malloc-512", },
184 	{1024, "malloc-1024", },
185 	{2048, "malloc-2048", },
186 	{4096, "malloc-4096", },
187 	{8192, "malloc-8192", },
188 	{16384, "malloc-16384", },
189 	{32768, "malloc-32768", },
190 	{65536, "malloc-65536", },
191 	{0, NULL},
192 };
193 
194 u_long vm_kmem_size;
195 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0,
196     "Size of kernel memory");
197 
198 static u_long kmem_zmax = KMEM_ZMAX;
199 SYSCTL_ULONG(_vm, OID_AUTO, kmem_zmax, CTLFLAG_RDTUN, &kmem_zmax, 0,
200     "Maximum allocation size that malloc(9) would use UMA as backend");
201 
202 static u_long vm_kmem_size_min;
203 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0,
204     "Minimum size of kernel memory");
205 
206 static u_long vm_kmem_size_max;
207 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0,
208     "Maximum size of kernel memory");
209 
210 static u_int vm_kmem_size_scale;
211 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0,
212     "Scale factor for kernel memory size");
213 
214 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS);
215 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size,
216     CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
217     sysctl_kmem_map_size, "LU", "Current kmem allocation size");
218 
219 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS);
220 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free,
221     CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
222     sysctl_kmem_map_free, "LU", "Free space in kmem");
223 
224 static SYSCTL_NODE(_vm, OID_AUTO, malloc, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
225     "Malloc information");
226 
227 static u_int vm_malloc_zone_count = nitems(kmemzones);
228 SYSCTL_UINT(_vm_malloc, OID_AUTO, zone_count,
229     CTLFLAG_RD, &vm_malloc_zone_count, 0,
230     "Number of malloc zones");
231 
232 static int sysctl_vm_malloc_zone_sizes(SYSCTL_HANDLER_ARGS);
233 SYSCTL_PROC(_vm_malloc, OID_AUTO, zone_sizes,
234     CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, NULL, 0,
235     sysctl_vm_malloc_zone_sizes, "S", "Zone sizes used by malloc");
236 
237 /*
238  * The malloc_mtx protects the kmemstatistics linked list.
239  */
240 struct mtx malloc_mtx;
241 
242 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS);
243 
244 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1)
245 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
246     "Kernel malloc debugging options");
247 #endif
248 
249 /*
250  * malloc(9) fault injection -- cause malloc failures every (n) mallocs when
251  * the caller specifies M_NOWAIT.  If set to 0, no failures are caused.
252  */
253 #ifdef MALLOC_MAKE_FAILURES
254 static int malloc_failure_rate;
255 static int malloc_nowait_count;
256 static int malloc_failure_count;
257 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RWTUN,
258     &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail");
259 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD,
260     &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures");
261 #endif
262 
263 static int
sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS)264 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS)
265 {
266 	u_long size;
267 
268 	size = uma_size();
269 	return (sysctl_handle_long(oidp, &size, 0, req));
270 }
271 
272 static int
sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS)273 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS)
274 {
275 	u_long size, limit;
276 
277 	/* The sysctl is unsigned, implement as a saturation value. */
278 	size = uma_size();
279 	limit = uma_limit();
280 	if (size > limit)
281 		size = 0;
282 	else
283 		size = limit - size;
284 	return (sysctl_handle_long(oidp, &size, 0, req));
285 }
286 
287 static int
sysctl_vm_malloc_zone_sizes(SYSCTL_HANDLER_ARGS)288 sysctl_vm_malloc_zone_sizes(SYSCTL_HANDLER_ARGS)
289 {
290 	int sizes[nitems(kmemzones)];
291 	int i;
292 
293 	for (i = 0; i < nitems(kmemzones); i++) {
294 		sizes[i] = kmemzones[i].kz_size;
295 	}
296 
297 	return (SYSCTL_OUT(req, &sizes, sizeof(sizes)));
298 }
299 
300 /*
301  * malloc(9) uma zone separation -- sub-page buffer overruns in one
302  * malloc type will affect only a subset of other malloc types.
303  */
304 #if MALLOC_DEBUG_MAXZONES > 1
305 static void
tunable_set_numzones(void)306 tunable_set_numzones(void)
307 {
308 
309 	TUNABLE_INT_FETCH("debug.malloc.numzones",
310 	    &numzones);
311 
312 	/* Sanity check the number of malloc uma zones. */
313 	if (numzones <= 0)
314 		numzones = 1;
315 	if (numzones > MALLOC_DEBUG_MAXZONES)
316 		numzones = MALLOC_DEBUG_MAXZONES;
317 }
318 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL);
319 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
320     &numzones, 0, "Number of malloc uma subzones");
321 
322 /*
323  * Any number that changes regularly is an okay choice for the
324  * offset.  Build numbers are pretty good of you have them.
325  */
326 static u_int zone_offset = __FreeBSD_version;
327 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset);
328 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN,
329     &zone_offset, 0, "Separate malloc types by examining the "
330     "Nth character in the malloc type short description.");
331 
332 static void
mtp_set_subzone(struct malloc_type * mtp)333 mtp_set_subzone(struct malloc_type *mtp)
334 {
335 	struct malloc_type_internal *mtip;
336 	const char *desc;
337 	size_t len;
338 	u_int val;
339 
340 	mtip = &mtp->ks_mti;
341 	desc = mtp->ks_shortdesc;
342 	if (desc == NULL || (len = strlen(desc)) == 0)
343 		val = 0;
344 	else
345 		val = desc[zone_offset % len];
346 	mtip->mti_zone = (val % numzones);
347 }
348 
349 static inline u_int
mtp_get_subzone(struct malloc_type * mtp)350 mtp_get_subzone(struct malloc_type *mtp)
351 {
352 	struct malloc_type_internal *mtip;
353 
354 	mtip = &mtp->ks_mti;
355 
356 	KASSERT(mtip->mti_zone < numzones,
357 	    ("mti_zone %u out of range %d",
358 	    mtip->mti_zone, numzones));
359 	return (mtip->mti_zone);
360 }
361 #elif MALLOC_DEBUG_MAXZONES == 0
362 #error "MALLOC_DEBUG_MAXZONES must be positive."
363 #else
364 static void
mtp_set_subzone(struct malloc_type * mtp)365 mtp_set_subzone(struct malloc_type *mtp)
366 {
367 	struct malloc_type_internal *mtip;
368 
369 	mtip = &mtp->ks_mti;
370 	mtip->mti_zone = 0;
371 }
372 
373 static inline u_int
mtp_get_subzone(struct malloc_type * mtp)374 mtp_get_subzone(struct malloc_type *mtp)
375 {
376 
377 	return (0);
378 }
379 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
380 
381 /*
382  * An allocation has succeeded -- update malloc type statistics for the
383  * amount of bucket size.  Occurs within a critical section so that the
384  * thread isn't preempted and doesn't migrate while updating per-PCU
385  * statistics.
386  */
387 static void
malloc_type_zone_allocated(struct malloc_type * mtp,unsigned long size,int zindx)388 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size,
389     int zindx)
390 {
391 	struct malloc_type_internal *mtip;
392 	struct malloc_type_stats *mtsp;
393 
394 	critical_enter();
395 	mtip = &mtp->ks_mti;
396 	mtsp = zpcpu_get(mtip->mti_stats);
397 	if (size > 0) {
398 		mtsp->mts_memalloced += size;
399 		mtsp->mts_numallocs++;
400 	}
401 	if (zindx != -1)
402 		mtsp->mts_size |= 1 << zindx;
403 
404 #ifdef KDTRACE_HOOKS
405 	if (__predict_false(dtrace_malloc_enabled)) {
406 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC];
407 		if (probe_id != 0)
408 			(dtrace_malloc_probe)(probe_id,
409 			    (uintptr_t) mtp, (uintptr_t) mtip,
410 			    (uintptr_t) mtsp, size, zindx);
411 	}
412 #endif
413 
414 	critical_exit();
415 }
416 
417 void
malloc_type_allocated(struct malloc_type * mtp,unsigned long size)418 malloc_type_allocated(struct malloc_type *mtp, unsigned long size)
419 {
420 
421 	if (size > 0)
422 		malloc_type_zone_allocated(mtp, size, -1);
423 }
424 
425 /*
426  * A free operation has occurred -- update malloc type statistics for the
427  * amount of the bucket size.  Occurs within a critical section so that the
428  * thread isn't preempted and doesn't migrate while updating per-CPU
429  * statistics.
430  */
431 void
malloc_type_freed(struct malloc_type * mtp,unsigned long size)432 malloc_type_freed(struct malloc_type *mtp, unsigned long size)
433 {
434 	struct malloc_type_internal *mtip;
435 	struct malloc_type_stats *mtsp;
436 
437 	critical_enter();
438 	mtip = &mtp->ks_mti;
439 	mtsp = zpcpu_get(mtip->mti_stats);
440 	mtsp->mts_memfreed += size;
441 	mtsp->mts_numfrees++;
442 
443 #ifdef KDTRACE_HOOKS
444 	if (__predict_false(dtrace_malloc_enabled)) {
445 		uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE];
446 		if (probe_id != 0)
447 			(dtrace_malloc_probe)(probe_id,
448 			    (uintptr_t) mtp, (uintptr_t) mtip,
449 			    (uintptr_t) mtsp, size, 0);
450 	}
451 #endif
452 
453 	critical_exit();
454 }
455 
456 /*
457  *	contigmalloc:
458  *
459  *	Allocate a block of physically contiguous memory.
460  *
461  *	If M_NOWAIT is set, this routine will not block and return NULL if
462  *	the allocation fails.
463  */
464 #define	IS_CONTIG_MALLOC(_slab)						\
465     (GET_SLAB_COOKIE(_slab) == SLAB_COOKIE_CONTIG_MALLOC)
466 #define	CONTIG_MALLOC_SLAB(_size)					\
467     ((void *)(((_size) << SLAB_COOKIE_SHIFT) | SLAB_COOKIE_CONTIG_MALLOC))
468 static inline size_t
contigmalloc_size(uma_slab_t slab)469 contigmalloc_size(uma_slab_t slab)
470 {
471 	uintptr_t va;
472 
473 	KASSERT(IS_CONTIG_MALLOC(slab),
474 	    ("%s: called on non-contigmalloc allocation: %p", __func__, slab));
475 	va = (uintptr_t)slab;
476 	return (va >> SLAB_COOKIE_SHIFT);
477 }
478 
479 void *
contigmalloc(unsigned long size,struct malloc_type * type,int flags,vm_paddr_t low,vm_paddr_t high,unsigned long alignment,vm_paddr_t boundary)480 contigmalloc(unsigned long size, struct malloc_type *type, int flags,
481     vm_paddr_t low, vm_paddr_t high, unsigned long alignment,
482     vm_paddr_t boundary)
483 {
484 	void *ret;
485 
486 	ret = (void *)kmem_alloc_contig(size, flags, low, high, alignment,
487 	    boundary, VM_MEMATTR_DEFAULT);
488 	if (ret != NULL) {
489 		/* Use low bits unused for slab pointers. */
490 		vsetzoneslab((uintptr_t)ret, NULL, CONTIG_MALLOC_SLAB(size));
491 		malloc_type_allocated(type, round_page(size));
492 	}
493 	return (ret);
494 }
495 
496 void *
contigmalloc_domainset(unsigned long size,struct malloc_type * type,struct domainset * ds,int flags,vm_paddr_t low,vm_paddr_t high,unsigned long alignment,vm_paddr_t boundary)497 contigmalloc_domainset(unsigned long size, struct malloc_type *type,
498     struct domainset *ds, int flags, vm_paddr_t low, vm_paddr_t high,
499     unsigned long alignment, vm_paddr_t boundary)
500 {
501 	void *ret;
502 
503 	ret = (void *)kmem_alloc_contig_domainset(ds, size, flags, low, high,
504 	    alignment, boundary, VM_MEMATTR_DEFAULT);
505 	if (ret != NULL) {
506 		/* Use low bits unused for slab pointers. */
507 		vsetzoneslab((uintptr_t)ret, NULL, CONTIG_MALLOC_SLAB(size));
508 		malloc_type_allocated(type, round_page(size));
509 	}
510 	return (ret);
511 }
512 #undef	IS_CONTIG_MALLOC
513 #undef	CONTIG_MALLOC_SLAB
514 
515 /* contigfree(9) is deprecated. */
516 void
contigfree(void * addr,unsigned long size __unused,struct malloc_type * type)517 contigfree(void *addr, unsigned long size __unused, struct malloc_type *type)
518 {
519 	free(addr, type);
520 }
521 
522 #ifdef MALLOC_DEBUG
523 static int
malloc_dbg(caddr_t * vap,size_t * sizep,struct malloc_type * mtp,int flags)524 malloc_dbg(caddr_t *vap, size_t *sizep, struct malloc_type *mtp,
525     int flags)
526 {
527 	KASSERT(mtp->ks_version == M_VERSION, ("malloc: bad malloc type version"));
528 	KASSERT((flags & (M_WAITOK | M_NOWAIT)) != 0,
529 	    ("malloc: flags must include either M_WAITOK or M_NOWAIT"));
530 	KASSERT((flags & (M_WAITOK | M_NOWAIT)) != (M_WAITOK | M_NOWAIT),
531 	    ("malloc: flags may not include both M_WAITOK and M_NOWAIT"));
532 	KASSERT((flags & M_NEVERFREED) == 0,
533 	    ("malloc: M_NEVERFREED is for internal use only"));
534 #ifdef MALLOC_MAKE_FAILURES
535 	if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
536 		atomic_add_int(&malloc_nowait_count, 1);
537 		if ((malloc_nowait_count % malloc_failure_rate) == 0) {
538 			atomic_add_int(&malloc_failure_count, 1);
539 			*vap = NULL;
540 			return (EJUSTRETURN);
541 		}
542 	}
543 #endif
544 	if (flags & M_WAITOK) {
545 		KASSERT(curthread->td_intr_nesting_level == 0,
546 		   ("malloc(M_WAITOK) in interrupt context"));
547 		if (__predict_false(!THREAD_CAN_SLEEP())) {
548 #ifdef EPOCH_TRACE
549 			epoch_trace_list(curthread);
550 #endif
551 			KASSERT(0,
552 			    ("malloc(M_WAITOK) with sleeping prohibited"));
553 		}
554 	}
555 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
556 	    ("malloc: called with spinlock or critical section held"));
557 
558 #ifdef DEBUG_MEMGUARD
559 	if (memguard_cmp_mtp(mtp, *sizep)) {
560 		*vap = memguard_alloc(*sizep, flags);
561 		if (*vap != NULL)
562 			return (EJUSTRETURN);
563 		/* This is unfortunate but should not be fatal. */
564 	}
565 #endif
566 
567 #ifdef DEBUG_REDZONE
568 	*sizep = redzone_size_ntor(*sizep);
569 #endif
570 
571 	return (0);
572 }
573 #endif
574 
575 /*
576  * Handle large allocations and frees by using kmem_malloc directly.
577  */
578 #define	IS_MALLOC_LARGE(_slab)						\
579     (GET_SLAB_COOKIE(_slab) == SLAB_COOKIE_MALLOC_LARGE)
580 #define	MALLOC_LARGE_SLAB(_size)					\
581     ((void *)(((_size) << SLAB_COOKIE_SHIFT) | SLAB_COOKIE_MALLOC_LARGE))
582 static inline size_t
malloc_large_size(uma_slab_t slab)583 malloc_large_size(uma_slab_t slab)
584 {
585 	uintptr_t va;
586 
587 	va = (uintptr_t)slab;
588 	KASSERT(IS_MALLOC_LARGE(slab),
589 	    ("%s: called on non-malloc_large allocation: %p", __func__, slab));
590 	return (va >> SLAB_COOKIE_SHIFT);
591 }
592 
593 static caddr_t __noinline
malloc_large(size_t size,struct malloc_type * mtp,struct domainset * policy,int flags DEBUG_REDZONE_ARG_DEF)594 malloc_large(size_t size, struct malloc_type *mtp, struct domainset *policy,
595     int flags DEBUG_REDZONE_ARG_DEF)
596 {
597 	void *va;
598 
599 	size = roundup(size, PAGE_SIZE);
600 	va = kmem_malloc_domainset(policy, size, flags);
601 	if (va != NULL) {
602 		/* Use low bits unused for slab pointers. */
603 		vsetzoneslab((uintptr_t)va, NULL, MALLOC_LARGE_SLAB(size));
604 		uma_total_inc(size);
605 	}
606 	malloc_type_allocated(mtp, va == NULL ? 0 : size);
607 	if (__predict_false(va == NULL)) {
608 		KASSERT((flags & M_WAITOK) == 0,
609 		    ("malloc(M_WAITOK) returned NULL"));
610 	} else {
611 #ifdef DEBUG_REDZONE
612 		va = redzone_setup(va, osize);
613 #endif
614 		kasan_mark(va, osize, size, KASAN_MALLOC_REDZONE);
615 	}
616 	return (va);
617 }
618 
619 static void
free_large(void * addr,size_t size)620 free_large(void *addr, size_t size)
621 {
622 
623 	kmem_free(addr, size);
624 	uma_total_dec(size);
625 }
626 #undef	IS_MALLOC_LARGE
627 #undef	MALLOC_LARGE_SLAB
628 
629 /*
630  *	malloc:
631  *
632  *	Allocate a block of memory.
633  *
634  *	If M_NOWAIT is set, this routine will not block and return NULL if
635  *	the allocation fails.
636  */
637 void *
638 (malloc)(size_t size, struct malloc_type *mtp, int flags)
639 {
640 	int indx;
641 	caddr_t va;
642 	uma_zone_t zone;
643 #if defined(DEBUG_REDZONE) || defined(KASAN)
644 	unsigned long osize = size;
645 #endif
646 
647 	MPASS((flags & M_EXEC) == 0);
648 
649 #ifdef MALLOC_DEBUG
650 	va = NULL;
651 	if (malloc_dbg(&va, &size, mtp, flags) != 0)
652 		return (va);
653 #endif
654 
655 	if (__predict_false(size > kmem_zmax))
656 		return (malloc_large(size, mtp, DOMAINSET_RR(), flags
657 		    DEBUG_REDZONE_ARG));
658 
659 	if (size & KMEM_ZMASK)
660 		size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
661 	indx = kmemsize[size >> KMEM_ZSHIFT];
662 	zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)];
663 	va = uma_zalloc_arg(zone, zone, flags);
664 	if (va != NULL) {
665 		size = zone->uz_size;
666 		if ((flags & M_ZERO) == 0) {
667 			kmsan_mark(va, size, KMSAN_STATE_UNINIT);
668 			kmsan_orig(va, size, KMSAN_TYPE_MALLOC, KMSAN_RET_ADDR);
669 		}
670 	}
671 	malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
672 	if (__predict_false(va == NULL)) {
673 		KASSERT((flags & M_WAITOK) == 0,
674 		    ("malloc(M_WAITOK) returned NULL"));
675 	}
676 #ifdef DEBUG_REDZONE
677 	if (va != NULL)
678 		va = redzone_setup(va, osize);
679 #endif
680 #ifdef KASAN
681 	if (va != NULL)
682 		kasan_mark((void *)va, osize, size, KASAN_MALLOC_REDZONE);
683 #endif
684 	return ((void *) va);
685 }
686 
687 static void *
malloc_domain(size_t * sizep,int * indxp,struct malloc_type * mtp,int domain,int flags)688 malloc_domain(size_t *sizep, int *indxp, struct malloc_type *mtp, int domain,
689     int flags)
690 {
691 	uma_zone_t zone;
692 	caddr_t va;
693 	size_t size;
694 	int indx;
695 
696 	size = *sizep;
697 	KASSERT(size <= kmem_zmax && (flags & M_EXEC) == 0,
698 	    ("malloc_domain: Called with bad flag / size combination"));
699 	if (size & KMEM_ZMASK)
700 		size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
701 	indx = kmemsize[size >> KMEM_ZSHIFT];
702 	zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)];
703 	va = uma_zalloc_domain(zone, zone, domain, flags);
704 	if (va != NULL)
705 		*sizep = zone->uz_size;
706 	*indxp = indx;
707 	return ((void *)va);
708 }
709 
710 void *
malloc_domainset(size_t size,struct malloc_type * mtp,struct domainset * ds,int flags)711 malloc_domainset(size_t size, struct malloc_type *mtp, struct domainset *ds,
712     int flags)
713 {
714 	struct vm_domainset_iter di;
715 	caddr_t va;
716 	int domain;
717 	int indx;
718 #if defined(KASAN) || defined(DEBUG_REDZONE)
719 	unsigned long osize = size;
720 #endif
721 
722 	MPASS((flags & M_EXEC) == 0);
723 
724 #ifdef MALLOC_DEBUG
725 	va = NULL;
726 	if (malloc_dbg(&va, &size, mtp, flags) != 0)
727 		return (va);
728 #endif
729 
730 	if (__predict_false(size > kmem_zmax))
731 		return (malloc_large(size, mtp, DOMAINSET_RR(), flags
732 		    DEBUG_REDZONE_ARG));
733 
734 	vm_domainset_iter_policy_init(&di, ds, &domain, &flags);
735 	do {
736 		va = malloc_domain(&size, &indx, mtp, domain, flags);
737 	} while (va == NULL && vm_domainset_iter_policy(&di, &domain) == 0);
738 	malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
739 	if (__predict_false(va == NULL)) {
740 		KASSERT((flags & M_WAITOK) == 0,
741 		    ("malloc(M_WAITOK) returned NULL"));
742 	}
743 #ifdef DEBUG_REDZONE
744 	if (va != NULL)
745 		va = redzone_setup(va, osize);
746 #endif
747 #ifdef KASAN
748 	if (va != NULL)
749 		kasan_mark((void *)va, osize, size, KASAN_MALLOC_REDZONE);
750 #endif
751 #ifdef KMSAN
752 	if ((flags & M_ZERO) == 0) {
753 		kmsan_mark(va, size, KMSAN_STATE_UNINIT);
754 		kmsan_orig(va, size, KMSAN_TYPE_MALLOC, KMSAN_RET_ADDR);
755 	}
756 #endif
757 	return (va);
758 }
759 
760 /*
761  * Allocate an executable area.
762  */
763 void *
malloc_exec(size_t size,struct malloc_type * mtp,int flags)764 malloc_exec(size_t size, struct malloc_type *mtp, int flags)
765 {
766 
767 	return (malloc_domainset_exec(size, mtp, DOMAINSET_RR(), flags));
768 }
769 
770 void *
malloc_domainset_exec(size_t size,struct malloc_type * mtp,struct domainset * ds,int flags)771 malloc_domainset_exec(size_t size, struct malloc_type *mtp, struct domainset *ds,
772     int flags)
773 {
774 #if defined(DEBUG_REDZONE) || defined(KASAN)
775 	unsigned long osize = size;
776 #endif
777 #ifdef MALLOC_DEBUG
778 	caddr_t va;
779 #endif
780 
781 	flags |= M_EXEC;
782 
783 #ifdef MALLOC_DEBUG
784 	va = NULL;
785 	if (malloc_dbg(&va, &size, mtp, flags) != 0)
786 		return (va);
787 #endif
788 
789 	return (malloc_large(size, mtp, ds, flags DEBUG_REDZONE_ARG));
790 }
791 
792 void *
malloc_aligned(size_t size,size_t align,struct malloc_type * type,int flags)793 malloc_aligned(size_t size, size_t align, struct malloc_type *type, int flags)
794 {
795 	return (malloc_domainset_aligned(size, align, type, DOMAINSET_RR(),
796 	    flags));
797 }
798 
799 void *
malloc_domainset_aligned(size_t size,size_t align,struct malloc_type * mtp,struct domainset * ds,int flags)800 malloc_domainset_aligned(size_t size, size_t align,
801     struct malloc_type *mtp, struct domainset *ds, int flags)
802 {
803 	void *res;
804 	size_t asize;
805 
806 	KASSERT(powerof2(align),
807 	    ("malloc_domainset_aligned: wrong align %#zx size %#zx",
808 	    align, size));
809 	KASSERT(align <= PAGE_SIZE,
810 	    ("malloc_domainset_aligned: align %#zx (size %#zx) too large",
811 	    align, size));
812 
813 	/*
814 	 * Round the allocation size up to the next power of 2,
815 	 * because we can only guarantee alignment for
816 	 * power-of-2-sized allocations.  Further increase the
817 	 * allocation size to align if the rounded size is less than
818 	 * align, since malloc zones provide alignment equal to their
819 	 * size.
820 	 */
821 	if (size == 0)
822 		size = 1;
823 	asize = size <= align ? align : 1UL << flsl(size - 1);
824 
825 	res = malloc_domainset(asize, mtp, ds, flags);
826 	KASSERT(res == NULL || ((uintptr_t)res & (align - 1)) == 0,
827 	    ("malloc_domainset_aligned: result not aligned %p size %#zx "
828 	    "allocsize %#zx align %#zx", res, size, asize, align));
829 	return (res);
830 }
831 
832 void *
mallocarray(size_t nmemb,size_t size,struct malloc_type * type,int flags)833 mallocarray(size_t nmemb, size_t size, struct malloc_type *type, int flags)
834 {
835 
836 	if (WOULD_OVERFLOW(nmemb, size))
837 		panic("mallocarray: %zu * %zu overflowed", nmemb, size);
838 
839 	return (malloc(size * nmemb, type, flags));
840 }
841 
842 void *
mallocarray_domainset(size_t nmemb,size_t size,struct malloc_type * type,struct domainset * ds,int flags)843 mallocarray_domainset(size_t nmemb, size_t size, struct malloc_type *type,
844     struct domainset *ds, int flags)
845 {
846 
847 	if (WOULD_OVERFLOW(nmemb, size))
848 		panic("mallocarray_domainset: %zu * %zu overflowed", nmemb, size);
849 
850 	return (malloc_domainset(size * nmemb, type, ds, flags));
851 }
852 
853 #if defined(INVARIANTS) && !defined(KASAN)
854 static void
free_save_type(void * addr,struct malloc_type * mtp,u_long size)855 free_save_type(void *addr, struct malloc_type *mtp, u_long size)
856 {
857 	struct malloc_type **mtpp = addr;
858 
859 	/*
860 	 * Cache a pointer to the malloc_type that most recently freed
861 	 * this memory here.  This way we know who is most likely to
862 	 * have stepped on it later.
863 	 *
864 	 * This code assumes that size is a multiple of 8 bytes for
865 	 * 64 bit machines
866 	 */
867 	mtpp = (struct malloc_type **) ((unsigned long)mtpp & ~UMA_ALIGN_PTR);
868 	mtpp += (size - sizeof(struct malloc_type *)) /
869 	    sizeof(struct malloc_type *);
870 	*mtpp = mtp;
871 }
872 #endif
873 
874 #ifdef MALLOC_DEBUG
875 static int
free_dbg(void ** addrp,struct malloc_type * mtp)876 free_dbg(void **addrp, struct malloc_type *mtp)
877 {
878 	void *addr;
879 
880 	addr = *addrp;
881 	KASSERT(mtp->ks_version == M_VERSION, ("free: bad malloc type version"));
882 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
883 	    ("free: called with spinlock or critical section held"));
884 
885 	/* free(NULL, ...) does nothing */
886 	if (addr == NULL)
887 		return (EJUSTRETURN);
888 
889 #ifdef DEBUG_MEMGUARD
890 	if (is_memguard_addr(addr)) {
891 		memguard_free(addr);
892 		return (EJUSTRETURN);
893 	}
894 #endif
895 
896 #ifdef DEBUG_REDZONE
897 	redzone_check(addr);
898 	*addrp = redzone_addr_ntor(addr);
899 #endif
900 
901 	return (0);
902 }
903 #endif
904 
905 static __always_inline void
_free(void * addr,struct malloc_type * mtp,bool dozero)906 _free(void *addr, struct malloc_type *mtp, bool dozero)
907 {
908 	uma_zone_t zone;
909 	uma_slab_t slab;
910 	u_long size;
911 
912 #ifdef MALLOC_DEBUG
913 	if (free_dbg(&addr, mtp) != 0)
914 		return;
915 #endif
916 	/* free(NULL, ...) does nothing */
917 	if (addr == NULL)
918 		return;
919 
920 	vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
921 	if (slab == NULL)
922 		panic("%s(%d): address %p(%p) has not been allocated", __func__,
923 		    dozero, addr, (void *)((uintptr_t)addr & (~UMA_SLAB_MASK)));
924 
925 	switch (GET_SLAB_COOKIE(slab)) {
926 	case __predict_true(SLAB_COOKIE_SLAB_PTR):
927 		size = zone->uz_size;
928 #if defined(INVARIANTS) && !defined(KASAN)
929 		free_save_type(addr, mtp, size);
930 #endif
931 		if (dozero) {
932 			kasan_mark(addr, size, size, 0);
933 			explicit_bzero(addr, size);
934 		}
935 		uma_zfree_arg(zone, addr, slab);
936 		break;
937 	case SLAB_COOKIE_MALLOC_LARGE:
938 		size = malloc_large_size(slab);
939 		if (dozero) {
940 			kasan_mark(addr, size, size, 0);
941 			explicit_bzero(addr, size);
942 		}
943 		free_large(addr, size);
944 		break;
945 	case SLAB_COOKIE_CONTIG_MALLOC:
946 		size = round_page(contigmalloc_size(slab));
947 		if (dozero)
948 			explicit_bzero(addr, size);
949 		kmem_free(addr, size);
950 		break;
951 	default:
952 		panic("%s(%d): addr %p slab %p with unknown cookie %d",
953 		    __func__, dozero, addr, slab, GET_SLAB_COOKIE(slab));
954 		/* NOTREACHED */
955 	}
956 	malloc_type_freed(mtp, size);
957 }
958 
959 /*
960  * free:
961  *	Free a block of memory allocated by malloc/contigmalloc.
962  *	This routine may not block.
963  */
964 void
free(void * addr,struct malloc_type * mtp)965 free(void *addr, struct malloc_type *mtp)
966 {
967 	_free(addr, mtp, false);
968 }
969 
970 /*
971  * zfree:
972  *	Zero then free a block of memory allocated by malloc/contigmalloc.
973  *	This routine may not block.
974  */
975 void
zfree(void * addr,struct malloc_type * mtp)976 zfree(void *addr, struct malloc_type *mtp)
977 {
978 	_free(addr, mtp, true);
979 }
980 
981 /*
982  *	realloc: change the size of a memory block
983  */
984 void *
realloc(void * addr,size_t size,struct malloc_type * mtp,int flags)985 realloc(void *addr, size_t size, struct malloc_type *mtp, int flags)
986 {
987 #ifndef DEBUG_REDZONE
988 	uma_zone_t zone;
989 	uma_slab_t slab;
990 #endif
991 	unsigned long alloc;
992 	void *newaddr;
993 
994 	KASSERT(mtp->ks_version == M_VERSION,
995 	    ("realloc: bad malloc type version"));
996 	KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
997 	    ("realloc: called with spinlock or critical section held"));
998 
999 	/* realloc(NULL, ...) is equivalent to malloc(...) */
1000 	if (addr == NULL)
1001 		return (malloc(size, mtp, flags));
1002 
1003 	/*
1004 	 * XXX: Should report free of old memory and alloc of new memory to
1005 	 * per-CPU stats.
1006 	 */
1007 
1008 #ifdef DEBUG_MEMGUARD
1009 	if (is_memguard_addr(addr))
1010 		return (memguard_realloc(addr, size, mtp, flags));
1011 #endif
1012 
1013 #ifdef DEBUG_REDZONE
1014 	alloc = redzone_get_size(addr);
1015 #else
1016 	vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
1017 
1018 	/* Sanity check */
1019 	KASSERT(slab != NULL,
1020 	    ("realloc: address %p out of range", (void *)addr));
1021 
1022 	/* Get the size of the original block */
1023 	switch (GET_SLAB_COOKIE(slab)) {
1024 	case __predict_true(SLAB_COOKIE_SLAB_PTR):
1025 		alloc = zone->uz_size;
1026 		break;
1027 	case SLAB_COOKIE_MALLOC_LARGE:
1028 		alloc = malloc_large_size(slab);
1029 		break;
1030 	default:
1031 #ifdef INVARIANTS
1032 		panic("%s: called for addr %p of unsupported allocation type; "
1033 		    "slab %p cookie %d", __func__, addr, slab, GET_SLAB_COOKIE(slab));
1034 #endif
1035 		return (NULL);
1036 	}
1037 
1038 	/* Reuse the original block if appropriate */
1039 	if (size <= alloc &&
1040 	    (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE)) {
1041 		kasan_mark((void *)addr, size, alloc, KASAN_MALLOC_REDZONE);
1042 		return (addr);
1043 	}
1044 #endif /* !DEBUG_REDZONE */
1045 
1046 	/* Allocate a new, bigger (or smaller) block */
1047 	if ((newaddr = malloc(size, mtp, flags)) == NULL)
1048 		return (NULL);
1049 
1050 	/*
1051 	 * Copy over original contents.  For KASAN, the redzone must be marked
1052 	 * valid before performing the copy.
1053 	 */
1054 	kasan_mark(addr, alloc, alloc, 0);
1055 	bcopy(addr, newaddr, min(size, alloc));
1056 	free(addr, mtp);
1057 	return (newaddr);
1058 }
1059 
1060 /*
1061  *	reallocf: same as realloc() but free memory on failure.
1062  */
1063 void *
reallocf(void * addr,size_t size,struct malloc_type * mtp,int flags)1064 reallocf(void *addr, size_t size, struct malloc_type *mtp, int flags)
1065 {
1066 	void *mem;
1067 
1068 	if ((mem = realloc(addr, size, mtp, flags)) == NULL)
1069 		free(addr, mtp);
1070 	return (mem);
1071 }
1072 
1073 /*
1074  * 	malloc_size: returns the number of bytes allocated for a request of the
1075  * 		     specified size
1076  */
1077 size_t
malloc_size(size_t size)1078 malloc_size(size_t size)
1079 {
1080 	int indx;
1081 
1082 	if (size > kmem_zmax)
1083 		return (round_page(size));
1084 	if (size & KMEM_ZMASK)
1085 		size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
1086 	indx = kmemsize[size >> KMEM_ZSHIFT];
1087 	return (kmemzones[indx].kz_size);
1088 }
1089 
1090 /*
1091  *	malloc_usable_size: returns the usable size of the allocation.
1092  */
1093 size_t
malloc_usable_size(const void * addr)1094 malloc_usable_size(const void *addr)
1095 {
1096 #ifndef DEBUG_REDZONE
1097 	uma_zone_t zone;
1098 	uma_slab_t slab;
1099 #endif
1100 	u_long size;
1101 
1102 	if (addr == NULL)
1103 		return (0);
1104 
1105 #ifdef DEBUG_MEMGUARD
1106 	if (is_memguard_addr(__DECONST(void *, addr)))
1107 		return (memguard_get_req_size(addr));
1108 #endif
1109 
1110 #ifdef DEBUG_REDZONE
1111 	size = redzone_get_size(__DECONST(void *, addr));
1112 #else
1113 	vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
1114 	if (slab == NULL)
1115 		panic("malloc_usable_size: address %p(%p) is not allocated",
1116 		    addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
1117 
1118 	switch (GET_SLAB_COOKIE(slab)) {
1119 	case __predict_true(SLAB_COOKIE_SLAB_PTR):
1120 		size = zone->uz_size;
1121 		break;
1122 	case SLAB_COOKIE_MALLOC_LARGE:
1123 		size = malloc_large_size(slab);
1124 		break;
1125 	default:
1126 		__assert_unreachable();
1127 		size = 0;
1128 		break;
1129 	}
1130 #endif
1131 
1132 	/*
1133 	 * Unmark the redzone to avoid reports from consumers who are
1134 	 * (presumably) about to use the full allocation size.
1135 	 */
1136 	kasan_mark(addr, size, size, 0);
1137 
1138 	return (size);
1139 }
1140 
1141 CTASSERT(VM_KMEM_SIZE_SCALE >= 1);
1142 
1143 /*
1144  * Initialize the kernel memory (kmem) arena.
1145  */
1146 void
kmeminit(void)1147 kmeminit(void)
1148 {
1149 	u_long mem_size;
1150 	u_long tmp;
1151 
1152 #ifdef VM_KMEM_SIZE
1153 	if (vm_kmem_size == 0)
1154 		vm_kmem_size = VM_KMEM_SIZE;
1155 #endif
1156 #ifdef VM_KMEM_SIZE_MIN
1157 	if (vm_kmem_size_min == 0)
1158 		vm_kmem_size_min = VM_KMEM_SIZE_MIN;
1159 #endif
1160 #ifdef VM_KMEM_SIZE_MAX
1161 	if (vm_kmem_size_max == 0)
1162 		vm_kmem_size_max = VM_KMEM_SIZE_MAX;
1163 #endif
1164 	/*
1165 	 * Calculate the amount of kernel virtual address (KVA) space that is
1166 	 * preallocated to the kmem arena.  In order to support a wide range
1167 	 * of machines, it is a function of the physical memory size,
1168 	 * specifically,
1169 	 *
1170 	 *	min(max(physical memory size / VM_KMEM_SIZE_SCALE,
1171 	 *	    VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX)
1172 	 *
1173 	 * Every architecture must define an integral value for
1174 	 * VM_KMEM_SIZE_SCALE.  However, the definitions of VM_KMEM_SIZE_MIN
1175 	 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and
1176 	 * ceiling on this preallocation, are optional.  Typically,
1177 	 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on
1178 	 * a given architecture.
1179 	 */
1180 	mem_size = vm_cnt.v_page_count;
1181 	if (mem_size <= 32768) /* delphij XXX 128MB */
1182 		kmem_zmax = PAGE_SIZE;
1183 
1184 	if (vm_kmem_size_scale < 1)
1185 		vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
1186 
1187 	/*
1188 	 * Check if we should use defaults for the "vm_kmem_size"
1189 	 * variable:
1190 	 */
1191 	if (vm_kmem_size == 0) {
1192 		vm_kmem_size = mem_size / vm_kmem_size_scale;
1193 		vm_kmem_size = vm_kmem_size * PAGE_SIZE < vm_kmem_size ?
1194 		    vm_kmem_size_max : vm_kmem_size * PAGE_SIZE;
1195 		if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min)
1196 			vm_kmem_size = vm_kmem_size_min;
1197 		if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
1198 			vm_kmem_size = vm_kmem_size_max;
1199 	}
1200 	if (vm_kmem_size == 0)
1201 		panic("Tune VM_KMEM_SIZE_* for the platform");
1202 
1203 	/*
1204 	 * The amount of KVA space that is preallocated to the
1205 	 * kmem arena can be set statically at compile-time or manually
1206 	 * through the kernel environment.  However, it is still limited to
1207 	 * twice the physical memory size, which has been sufficient to handle
1208 	 * the most severe cases of external fragmentation in the kmem arena.
1209 	 */
1210 	if (vm_kmem_size / 2 / PAGE_SIZE > mem_size)
1211 		vm_kmem_size = 2 * mem_size * PAGE_SIZE;
1212 
1213 	vm_kmem_size = round_page(vm_kmem_size);
1214 
1215 	/*
1216 	 * With KASAN or KMSAN enabled, dynamically allocated kernel memory is
1217 	 * shadowed.  Account for this when setting the UMA limit.
1218 	 */
1219 #if defined(KASAN)
1220 	vm_kmem_size = (vm_kmem_size * KASAN_SHADOW_SCALE) /
1221 	    (KASAN_SHADOW_SCALE + 1);
1222 #elif defined(KMSAN)
1223 	vm_kmem_size /= 3;
1224 #endif
1225 
1226 #ifdef DEBUG_MEMGUARD
1227 	tmp = memguard_fudge(vm_kmem_size, kernel_map);
1228 #else
1229 	tmp = vm_kmem_size;
1230 #endif
1231 	uma_set_limit(tmp);
1232 
1233 #ifdef DEBUG_MEMGUARD
1234 	/*
1235 	 * Initialize MemGuard if support compiled in.  MemGuard is a
1236 	 * replacement allocator used for detecting tamper-after-free
1237 	 * scenarios as they occur.  It is only used for debugging.
1238 	 */
1239 	memguard_init(kernel_arena);
1240 #endif
1241 }
1242 
1243 /*
1244  * Initialize the kernel memory allocator
1245  */
1246 /* ARGSUSED*/
1247 static void
mallocinit(void * dummy)1248 mallocinit(void *dummy)
1249 {
1250 	int i;
1251 	uint8_t indx;
1252 
1253 	mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
1254 
1255 	kmeminit();
1256 
1257 	if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX)
1258 		kmem_zmax = KMEM_ZMAX;
1259 
1260 	for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
1261 		int size = kmemzones[indx].kz_size;
1262 		const char *name = kmemzones[indx].kz_name;
1263 		size_t align;
1264 		int subzone;
1265 
1266 		align = UMA_ALIGN_PTR;
1267 		if (powerof2(size) && size > sizeof(void *))
1268 			align = MIN(size, PAGE_SIZE) - 1;
1269 		for (subzone = 0; subzone < numzones; subzone++) {
1270 			kmemzones[indx].kz_zone[subzone] =
1271 			    uma_zcreate(name, size,
1272 #if defined(INVARIANTS) && !defined(KASAN) && !defined(KMSAN)
1273 			    mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
1274 #else
1275 			    NULL, NULL, NULL, NULL,
1276 #endif
1277 			    align, UMA_ZONE_MALLOC);
1278 		}
1279 		for (;i <= size; i+= KMEM_ZBASE)
1280 			kmemsize[i >> KMEM_ZSHIFT] = indx;
1281 	}
1282 }
1283 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_SECOND, mallocinit, NULL);
1284 
1285 void
malloc_init(void * data)1286 malloc_init(void *data)
1287 {
1288 	struct malloc_type_internal *mtip;
1289 	struct malloc_type *mtp;
1290 
1291 	KASSERT(vm_cnt.v_page_count != 0,
1292 	    ("malloc_init() called before vm_mem_init()"));
1293 
1294 	mtp = data;
1295 	if (mtp->ks_version != M_VERSION)
1296 		panic("malloc_init: type %s with unsupported version %lu",
1297 		    mtp->ks_shortdesc, mtp->ks_version);
1298 
1299 	mtip = &mtp->ks_mti;
1300 	mtip->mti_stats = uma_zalloc_pcpu(pcpu_zone_64, M_WAITOK | M_ZERO);
1301 	mtp_set_subzone(mtp);
1302 
1303 	mtx_lock(&malloc_mtx);
1304 	mtp->ks_next = kmemstatistics;
1305 	kmemstatistics = mtp;
1306 	kmemcount++;
1307 	mtx_unlock(&malloc_mtx);
1308 }
1309 
1310 void
malloc_uninit(void * data)1311 malloc_uninit(void *data)
1312 {
1313 	struct malloc_type_internal *mtip;
1314 	struct malloc_type_stats *mtsp;
1315 	struct malloc_type *mtp, *temp;
1316 	long temp_allocs, temp_bytes;
1317 	int i;
1318 
1319 	mtp = data;
1320 	KASSERT(mtp->ks_version == M_VERSION,
1321 	    ("malloc_uninit: bad malloc type version"));
1322 
1323 	mtx_lock(&malloc_mtx);
1324 	mtip = &mtp->ks_mti;
1325 	if (mtp != kmemstatistics) {
1326 		for (temp = kmemstatistics; temp != NULL;
1327 		    temp = temp->ks_next) {
1328 			if (temp->ks_next == mtp) {
1329 				temp->ks_next = mtp->ks_next;
1330 				break;
1331 			}
1332 		}
1333 		KASSERT(temp,
1334 		    ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc));
1335 	} else
1336 		kmemstatistics = mtp->ks_next;
1337 	kmemcount--;
1338 	mtx_unlock(&malloc_mtx);
1339 
1340 	/*
1341 	 * Look for memory leaks.
1342 	 */
1343 	temp_allocs = temp_bytes = 0;
1344 	for (i = 0; i <= mp_maxid; i++) {
1345 		mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1346 		temp_allocs += mtsp->mts_numallocs;
1347 		temp_allocs -= mtsp->mts_numfrees;
1348 		temp_bytes += mtsp->mts_memalloced;
1349 		temp_bytes -= mtsp->mts_memfreed;
1350 	}
1351 	if (temp_allocs > 0 || temp_bytes > 0) {
1352 		printf("Warning: memory type %s leaked memory on destroy "
1353 		    "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc,
1354 		    temp_allocs, temp_bytes);
1355 	}
1356 
1357 	uma_zfree_pcpu(pcpu_zone_64, mtip->mti_stats);
1358 }
1359 
1360 struct malloc_type *
malloc_desc2type(const char * desc)1361 malloc_desc2type(const char *desc)
1362 {
1363 	struct malloc_type *mtp;
1364 
1365 	mtx_assert(&malloc_mtx, MA_OWNED);
1366 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1367 		if (strcmp(mtp->ks_shortdesc, desc) == 0)
1368 			return (mtp);
1369 	}
1370 	return (NULL);
1371 }
1372 
1373 static int
sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)1374 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
1375 {
1376 	struct malloc_type_stream_header mtsh;
1377 	struct malloc_type_internal *mtip;
1378 	struct malloc_type_stats *mtsp, zeromts;
1379 	struct malloc_type_header mth;
1380 	struct malloc_type *mtp;
1381 	int error, i;
1382 	struct sbuf sbuf;
1383 
1384 	error = sysctl_wire_old_buffer(req, 0);
1385 	if (error != 0)
1386 		return (error);
1387 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
1388 	sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL);
1389 	mtx_lock(&malloc_mtx);
1390 
1391 	bzero(&zeromts, sizeof(zeromts));
1392 
1393 	/*
1394 	 * Insert stream header.
1395 	 */
1396 	bzero(&mtsh, sizeof(mtsh));
1397 	mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
1398 	mtsh.mtsh_maxcpus = MAXCPU;
1399 	mtsh.mtsh_count = kmemcount;
1400 	(void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh));
1401 
1402 	/*
1403 	 * Insert alternating sequence of type headers and type statistics.
1404 	 */
1405 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1406 		mtip = &mtp->ks_mti;
1407 
1408 		/*
1409 		 * Insert type header.
1410 		 */
1411 		bzero(&mth, sizeof(mth));
1412 		strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
1413 		(void)sbuf_bcat(&sbuf, &mth, sizeof(mth));
1414 
1415 		/*
1416 		 * Insert type statistics for each CPU.
1417 		 */
1418 		for (i = 0; i <= mp_maxid; i++) {
1419 			mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1420 			(void)sbuf_bcat(&sbuf, mtsp, sizeof(*mtsp));
1421 		}
1422 		/*
1423 		 * Fill in the missing CPUs.
1424 		 */
1425 		for (; i < MAXCPU; i++) {
1426 			(void)sbuf_bcat(&sbuf, &zeromts, sizeof(zeromts));
1427 		}
1428 	}
1429 	mtx_unlock(&malloc_mtx);
1430 	error = sbuf_finish(&sbuf);
1431 	sbuf_delete(&sbuf);
1432 	return (error);
1433 }
1434 
1435 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats,
1436     CTLFLAG_RD | CTLTYPE_STRUCT | CTLFLAG_MPSAFE, 0, 0,
1437     sysctl_kern_malloc_stats, "s,malloc_type_ustats",
1438     "Return malloc types");
1439 
1440 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
1441     "Count of kernel malloc types");
1442 
1443 void
malloc_type_list(malloc_type_list_func_t * func,void * arg)1444 malloc_type_list(malloc_type_list_func_t *func, void *arg)
1445 {
1446 	struct malloc_type *mtp, **bufmtp;
1447 	int count, i;
1448 	size_t buflen;
1449 
1450 	mtx_lock(&malloc_mtx);
1451 restart:
1452 	mtx_assert(&malloc_mtx, MA_OWNED);
1453 	count = kmemcount;
1454 	mtx_unlock(&malloc_mtx);
1455 
1456 	buflen = sizeof(struct malloc_type *) * count;
1457 	bufmtp = malloc(buflen, M_TEMP, M_WAITOK);
1458 
1459 	mtx_lock(&malloc_mtx);
1460 
1461 	if (count < kmemcount) {
1462 		free(bufmtp, M_TEMP);
1463 		goto restart;
1464 	}
1465 
1466 	for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++)
1467 		bufmtp[i] = mtp;
1468 
1469 	mtx_unlock(&malloc_mtx);
1470 
1471 	for (i = 0; i < count; i++)
1472 		(func)(bufmtp[i], arg);
1473 
1474 	free(bufmtp, M_TEMP);
1475 }
1476 
1477 #ifdef DDB
1478 static int64_t
get_malloc_stats(const struct malloc_type_internal * mtip,uint64_t * allocs,uint64_t * inuse)1479 get_malloc_stats(const struct malloc_type_internal *mtip, uint64_t *allocs,
1480     uint64_t *inuse)
1481 {
1482 	const struct malloc_type_stats *mtsp;
1483 	uint64_t frees, alloced, freed;
1484 	int i;
1485 
1486 	*allocs = 0;
1487 	frees = 0;
1488 	alloced = 0;
1489 	freed = 0;
1490 	for (i = 0; i <= mp_maxid; i++) {
1491 		mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1492 
1493 		*allocs += mtsp->mts_numallocs;
1494 		frees += mtsp->mts_numfrees;
1495 		alloced += mtsp->mts_memalloced;
1496 		freed += mtsp->mts_memfreed;
1497 	}
1498 	*inuse = *allocs - frees;
1499 	return (alloced - freed);
1500 }
1501 
DB_SHOW_COMMAND_FLAGS(malloc,db_show_malloc,DB_CMD_MEMSAFE)1502 DB_SHOW_COMMAND_FLAGS(malloc, db_show_malloc, DB_CMD_MEMSAFE)
1503 {
1504 	const char *fmt_hdr, *fmt_entry;
1505 	struct malloc_type *mtp;
1506 	uint64_t allocs, inuse;
1507 	int64_t size;
1508 	/* variables for sorting */
1509 	struct malloc_type *last_mtype, *cur_mtype;
1510 	int64_t cur_size, last_size;
1511 	int ties;
1512 
1513 	if (modif[0] == 'i') {
1514 		fmt_hdr = "%s,%s,%s,%s\n";
1515 		fmt_entry = "\"%s\",%ju,%jdK,%ju\n";
1516 	} else {
1517 		fmt_hdr = "%18s %12s  %12s %12s\n";
1518 		fmt_entry = "%18s %12ju %12jdK %12ju\n";
1519 	}
1520 
1521 	db_printf(fmt_hdr, "Type", "InUse", "MemUse", "Requests");
1522 
1523 	/* Select sort, largest size first. */
1524 	last_mtype = NULL;
1525 	last_size = INT64_MAX;
1526 	for (;;) {
1527 		cur_mtype = NULL;
1528 		cur_size = -1;
1529 		ties = 0;
1530 
1531 		for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1532 			/*
1533 			 * In the case of size ties, print out mtypes
1534 			 * in the order they are encountered.  That is,
1535 			 * when we encounter the most recently output
1536 			 * mtype, we have already printed all preceding
1537 			 * ties, and we must print all following ties.
1538 			 */
1539 			if (mtp == last_mtype) {
1540 				ties = 1;
1541 				continue;
1542 			}
1543 			size = get_malloc_stats(&mtp->ks_mti, &allocs,
1544 			    &inuse);
1545 			if (size > cur_size && size < last_size + ties) {
1546 				cur_size = size;
1547 				cur_mtype = mtp;
1548 			}
1549 		}
1550 		if (cur_mtype == NULL)
1551 			break;
1552 
1553 		size = get_malloc_stats(&cur_mtype->ks_mti, &allocs, &inuse);
1554 		db_printf(fmt_entry, cur_mtype->ks_shortdesc, inuse,
1555 		    howmany(size, 1024), allocs);
1556 
1557 		if (db_pager_quit)
1558 			break;
1559 
1560 		last_mtype = cur_mtype;
1561 		last_size = cur_size;
1562 	}
1563 }
1564 
1565 #if MALLOC_DEBUG_MAXZONES > 1
DB_SHOW_COMMAND(multizone_matches,db_show_multizone_matches)1566 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches)
1567 {
1568 	struct malloc_type_internal *mtip;
1569 	struct malloc_type *mtp;
1570 	u_int subzone;
1571 
1572 	if (!have_addr) {
1573 		db_printf("Usage: show multizone_matches <malloc type/addr>\n");
1574 		return;
1575 	}
1576 	mtp = (void *)addr;
1577 	if (mtp->ks_version != M_VERSION) {
1578 		db_printf("Version %lx does not match expected %x\n",
1579 		    mtp->ks_version, M_VERSION);
1580 		return;
1581 	}
1582 
1583 	mtip = &mtp->ks_mti;
1584 	subzone = mtip->mti_zone;
1585 
1586 	for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1587 		mtip = &mtp->ks_mti;
1588 		if (mtip->mti_zone != subzone)
1589 			continue;
1590 		db_printf("%s\n", mtp->ks_shortdesc);
1591 		if (db_pager_quit)
1592 			break;
1593 	}
1594 }
1595 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
1596 #endif /* DDB */
1597