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 osize,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 osize, 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 unsigned long size;
486
487 #ifdef DEBUG_REDZONE
488 size = redzone_size_ntor(osize);
489 #else
490 size = osize;
491 #endif
492
493 ret = (void *)kmem_alloc_contig(size, flags, low, high, alignment,
494 boundary, VM_MEMATTR_DEFAULT);
495 if (ret != NULL) {
496 /* Use low bits unused for slab pointers. */
497 vsetzoneslab((uintptr_t)ret, NULL, CONTIG_MALLOC_SLAB(size));
498 malloc_type_allocated(type, round_page(size));
499 #ifdef DEBUG_REDZONE
500 ret = redzone_setup(ret, osize);
501 #endif
502 }
503 return (ret);
504 }
505
506 void *
contigmalloc_domainset(unsigned long osize,struct malloc_type * type,struct domainset * ds,int flags,vm_paddr_t low,vm_paddr_t high,unsigned long alignment,vm_paddr_t boundary)507 contigmalloc_domainset(unsigned long osize, struct malloc_type *type,
508 struct domainset *ds, int flags, vm_paddr_t low, vm_paddr_t high,
509 unsigned long alignment, vm_paddr_t boundary)
510 {
511 void *ret;
512 unsigned long size;
513
514 #ifdef DEBUG_REDZONE
515 size = redzone_size_ntor(osize);
516 #else
517 size = osize;
518 #endif
519
520 ret = (void *)kmem_alloc_contig_domainset(ds, size, flags, low, high,
521 alignment, boundary, VM_MEMATTR_DEFAULT);
522 if (ret != NULL) {
523 /* Use low bits unused for slab pointers. */
524 vsetzoneslab((uintptr_t)ret, NULL, CONTIG_MALLOC_SLAB(size));
525 malloc_type_allocated(type, round_page(size));
526 #ifdef DEBUG_REDZONE
527 ret = redzone_setup(ret, osize);
528 #endif
529 }
530 return (ret);
531 }
532 #undef IS_CONTIG_MALLOC
533 #undef CONTIG_MALLOC_SLAB
534
535 /* contigfree(9) is deprecated. */
536 void
contigfree(void * addr,unsigned long size __unused,struct malloc_type * type)537 contigfree(void *addr, unsigned long size __unused, struct malloc_type *type)
538 {
539 free(addr, type);
540 }
541
542 #ifdef MALLOC_DEBUG
543 static int
malloc_dbg(caddr_t * vap,size_t * sizep,struct malloc_type * mtp,int flags)544 malloc_dbg(caddr_t *vap, size_t *sizep, struct malloc_type *mtp,
545 int flags)
546 {
547 KASSERT(mtp->ks_version == M_VERSION, ("malloc: bad malloc type version"));
548 KASSERT((flags & (M_WAITOK | M_NOWAIT)) != 0,
549 ("malloc: flags must include either M_WAITOK or M_NOWAIT"));
550 KASSERT((flags & (M_WAITOK | M_NOWAIT)) != (M_WAITOK | M_NOWAIT),
551 ("malloc: flags may not include both M_WAITOK and M_NOWAIT"));
552 KASSERT((flags & M_NEVERFREED) == 0,
553 ("malloc: M_NEVERFREED is for internal use only"));
554 #ifdef MALLOC_MAKE_FAILURES
555 if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
556 atomic_add_int(&malloc_nowait_count, 1);
557 if ((malloc_nowait_count % malloc_failure_rate) == 0) {
558 atomic_add_int(&malloc_failure_count, 1);
559 *vap = NULL;
560 return (EJUSTRETURN);
561 }
562 }
563 #endif
564 if (flags & M_WAITOK) {
565 KASSERT(curthread->td_intr_nesting_level == 0,
566 ("malloc(M_WAITOK) in interrupt context"));
567 if (__predict_false(!THREAD_CAN_SLEEP())) {
568 #ifdef EPOCH_TRACE
569 epoch_trace_list(curthread);
570 #endif
571 KASSERT(0,
572 ("malloc(M_WAITOK) with sleeping prohibited"));
573 }
574 }
575 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
576 ("malloc: called with spinlock or critical section held"));
577
578 #ifdef DEBUG_MEMGUARD
579 if (memguard_cmp_mtp(mtp, *sizep)) {
580 *vap = memguard_alloc(*sizep, flags);
581 if (*vap != NULL)
582 return (EJUSTRETURN);
583 /* This is unfortunate but should not be fatal. */
584 }
585 #endif
586
587 #ifdef DEBUG_REDZONE
588 *sizep = redzone_size_ntor(*sizep);
589 #endif
590
591 return (0);
592 }
593 #endif
594
595 /*
596 * Handle large allocations and frees by using kmem_malloc directly.
597 */
598 #define IS_MALLOC_LARGE(_slab) \
599 (GET_SLAB_COOKIE(_slab) == SLAB_COOKIE_MALLOC_LARGE)
600 #define MALLOC_LARGE_SLAB(_size) \
601 ((void *)(((_size) << SLAB_COOKIE_SHIFT) | SLAB_COOKIE_MALLOC_LARGE))
602 static inline size_t
malloc_large_size(uma_slab_t slab)603 malloc_large_size(uma_slab_t slab)
604 {
605 uintptr_t va;
606
607 va = (uintptr_t)slab;
608 KASSERT(IS_MALLOC_LARGE(slab),
609 ("%s: called on non-malloc_large allocation: %p", __func__, slab));
610 return (va >> SLAB_COOKIE_SHIFT);
611 }
612
613 static caddr_t __noinline
malloc_large(size_t size,struct malloc_type * mtp,struct domainset * policy,int flags DEBUG_REDZONE_ARG_DEF)614 malloc_large(size_t size, struct malloc_type *mtp, struct domainset *policy,
615 int flags DEBUG_REDZONE_ARG_DEF)
616 {
617 void *va;
618
619 size = roundup(size, PAGE_SIZE);
620 va = kmem_malloc_domainset(policy, size, flags);
621 if (va != NULL) {
622 /* Use low bits unused for slab pointers. */
623 vsetzoneslab((uintptr_t)va, NULL, MALLOC_LARGE_SLAB(size));
624 uma_total_inc(size);
625 }
626 malloc_type_allocated(mtp, va == NULL ? 0 : size);
627 if (__predict_false(va == NULL)) {
628 KASSERT((flags & M_WAITOK) == 0,
629 ("malloc(M_WAITOK) returned NULL"));
630 } else {
631 #ifdef DEBUG_REDZONE
632 va = redzone_setup(va, osize);
633 #endif
634 kasan_mark(va, osize, size, KASAN_MALLOC_REDZONE);
635 }
636 return (va);
637 }
638
639 static void
free_large(void * addr,size_t size)640 free_large(void *addr, size_t size)
641 {
642
643 kmem_free(addr, size);
644 uma_total_dec(size);
645 }
646 #undef IS_MALLOC_LARGE
647 #undef MALLOC_LARGE_SLAB
648
649 /*
650 * malloc:
651 *
652 * Allocate a block of memory.
653 *
654 * If M_NOWAIT is set, this routine will not block and return NULL if
655 * the allocation fails.
656 */
657 void *
658 (malloc)(size_t size, struct malloc_type *mtp, int flags)
659 {
660 int indx;
661 caddr_t va;
662 uma_zone_t zone;
663 #if defined(DEBUG_REDZONE) || defined(KASAN)
664 unsigned long osize = size;
665 #endif
666
667 MPASS((flags & M_EXEC) == 0);
668
669 #ifdef MALLOC_DEBUG
670 va = NULL;
671 if (malloc_dbg(&va, &size, mtp, flags) != 0)
672 return (va);
673 #endif
674
675 if (__predict_false(size > kmem_zmax))
676 return (malloc_large(size, mtp, DOMAINSET_RR(), flags
677 DEBUG_REDZONE_ARG));
678
679 if (size & KMEM_ZMASK)
680 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
681 indx = kmemsize[size >> KMEM_ZSHIFT];
682 zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)];
683 va = uma_zalloc_arg(zone, zone, flags);
684 if (va != NULL) {
685 size = zone->uz_size;
686 if ((flags & M_ZERO) == 0) {
687 kmsan_mark(va, size, KMSAN_STATE_UNINIT);
688 kmsan_orig(va, size, KMSAN_TYPE_MALLOC, KMSAN_RET_ADDR);
689 }
690 }
691 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
692 if (__predict_false(va == NULL)) {
693 KASSERT((flags & M_WAITOK) == 0,
694 ("malloc(M_WAITOK) returned NULL"));
695 }
696 #ifdef DEBUG_REDZONE
697 if (va != NULL)
698 va = redzone_setup(va, osize);
699 #endif
700 #ifdef KASAN
701 if (va != NULL)
702 kasan_mark((void *)va, osize, size, KASAN_MALLOC_REDZONE);
703 #endif
704 return ((void *) va);
705 }
706
707 static void *
malloc_domain(size_t * sizep,int * indxp,struct malloc_type * mtp,int domain,int flags)708 malloc_domain(size_t *sizep, int *indxp, struct malloc_type *mtp, int domain,
709 int flags)
710 {
711 uma_zone_t zone;
712 caddr_t va;
713 size_t size;
714 int indx;
715
716 size = *sizep;
717 KASSERT(size <= kmem_zmax && (flags & M_EXEC) == 0,
718 ("malloc_domain: Called with bad flag / size combination"));
719 if (size & KMEM_ZMASK)
720 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
721 indx = kmemsize[size >> KMEM_ZSHIFT];
722 zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)];
723 va = uma_zalloc_domain(zone, zone, domain, flags);
724 if (va != NULL)
725 *sizep = zone->uz_size;
726 *indxp = indx;
727 return ((void *)va);
728 }
729
730 void *
malloc_domainset(size_t size,struct malloc_type * mtp,struct domainset * ds,int flags)731 malloc_domainset(size_t size, struct malloc_type *mtp, struct domainset *ds,
732 int flags)
733 {
734 struct vm_domainset_iter di;
735 caddr_t va;
736 int domain;
737 int indx;
738 #if defined(KASAN) || defined(DEBUG_REDZONE)
739 unsigned long osize = size;
740 #endif
741
742 MPASS((flags & M_EXEC) == 0);
743
744 #ifdef MALLOC_DEBUG
745 va = NULL;
746 if (malloc_dbg(&va, &size, mtp, flags) != 0)
747 return (va);
748 #endif
749
750 if (__predict_false(size > kmem_zmax))
751 return (malloc_large(size, mtp, DOMAINSET_RR(), flags
752 DEBUG_REDZONE_ARG));
753
754 vm_domainset_iter_policy_init(&di, ds, &domain, &flags);
755 do {
756 va = malloc_domain(&size, &indx, mtp, domain, flags);
757 } while (va == NULL && vm_domainset_iter_policy(&di, &domain) == 0);
758 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
759 if (__predict_false(va == NULL)) {
760 KASSERT((flags & M_WAITOK) == 0,
761 ("malloc(M_WAITOK) returned NULL"));
762 }
763 #ifdef DEBUG_REDZONE
764 if (va != NULL)
765 va = redzone_setup(va, osize);
766 #endif
767 #ifdef KASAN
768 if (va != NULL)
769 kasan_mark((void *)va, osize, size, KASAN_MALLOC_REDZONE);
770 #endif
771 #ifdef KMSAN
772 if ((flags & M_ZERO) == 0) {
773 kmsan_mark(va, size, KMSAN_STATE_UNINIT);
774 kmsan_orig(va, size, KMSAN_TYPE_MALLOC, KMSAN_RET_ADDR);
775 }
776 #endif
777 return (va);
778 }
779
780 /*
781 * Allocate an executable area.
782 */
783 void *
malloc_exec(size_t size,struct malloc_type * mtp,int flags)784 malloc_exec(size_t size, struct malloc_type *mtp, int flags)
785 {
786
787 return (malloc_domainset_exec(size, mtp, DOMAINSET_RR(), flags));
788 }
789
790 void *
malloc_domainset_exec(size_t size,struct malloc_type * mtp,struct domainset * ds,int flags)791 malloc_domainset_exec(size_t size, struct malloc_type *mtp, struct domainset *ds,
792 int flags)
793 {
794 #if defined(DEBUG_REDZONE) || defined(KASAN)
795 unsigned long osize = size;
796 #endif
797 #ifdef MALLOC_DEBUG
798 caddr_t va;
799 #endif
800
801 flags |= M_EXEC;
802
803 #ifdef MALLOC_DEBUG
804 va = NULL;
805 if (malloc_dbg(&va, &size, mtp, flags) != 0)
806 return (va);
807 #endif
808
809 return (malloc_large(size, mtp, ds, flags DEBUG_REDZONE_ARG));
810 }
811
812 void *
malloc_aligned(size_t size,size_t align,struct malloc_type * type,int flags)813 malloc_aligned(size_t size, size_t align, struct malloc_type *type, int flags)
814 {
815 return (malloc_domainset_aligned(size, align, type, DOMAINSET_RR(),
816 flags));
817 }
818
819 void *
malloc_domainset_aligned(size_t size,size_t align,struct malloc_type * mtp,struct domainset * ds,int flags)820 malloc_domainset_aligned(size_t size, size_t align,
821 struct malloc_type *mtp, struct domainset *ds, int flags)
822 {
823 void *res;
824 size_t asize;
825
826 KASSERT(powerof2(align),
827 ("malloc_domainset_aligned: wrong align %#zx size %#zx",
828 align, size));
829 KASSERT(align <= PAGE_SIZE,
830 ("malloc_domainset_aligned: align %#zx (size %#zx) too large",
831 align, size));
832
833 /*
834 * Round the allocation size up to the next power of 2,
835 * because we can only guarantee alignment for
836 * power-of-2-sized allocations. Further increase the
837 * allocation size to align if the rounded size is less than
838 * align, since malloc zones provide alignment equal to their
839 * size.
840 */
841 if (size == 0)
842 size = 1;
843 asize = size <= align ? align : 1UL << flsl(size - 1);
844
845 res = malloc_domainset(asize, mtp, ds, flags);
846 KASSERT(res == NULL || ((uintptr_t)res & (align - 1)) == 0,
847 ("malloc_domainset_aligned: result not aligned %p size %#zx "
848 "allocsize %#zx align %#zx", res, size, asize, align));
849 return (res);
850 }
851
852 void *
mallocarray(size_t nmemb,size_t size,struct malloc_type * type,int flags)853 mallocarray(size_t nmemb, size_t size, struct malloc_type *type, int flags)
854 {
855
856 if (WOULD_OVERFLOW(nmemb, size))
857 panic("mallocarray: %zu * %zu overflowed", nmemb, size);
858
859 return (malloc(size * nmemb, type, flags));
860 }
861
862 void *
mallocarray_domainset(size_t nmemb,size_t size,struct malloc_type * type,struct domainset * ds,int flags)863 mallocarray_domainset(size_t nmemb, size_t size, struct malloc_type *type,
864 struct domainset *ds, int flags)
865 {
866
867 if (WOULD_OVERFLOW(nmemb, size))
868 panic("mallocarray_domainset: %zu * %zu overflowed", nmemb, size);
869
870 return (malloc_domainset(size * nmemb, type, ds, flags));
871 }
872
873 #if defined(INVARIANTS) && !defined(KASAN)
874 static void
free_save_type(void * addr,struct malloc_type * mtp,u_long size)875 free_save_type(void *addr, struct malloc_type *mtp, u_long size)
876 {
877 struct malloc_type **mtpp = addr;
878
879 /*
880 * Cache a pointer to the malloc_type that most recently freed
881 * this memory here. This way we know who is most likely to
882 * have stepped on it later.
883 *
884 * This code assumes that size is a multiple of 8 bytes for
885 * 64 bit machines
886 */
887 mtpp = (struct malloc_type **) ((unsigned long)mtpp & ~UMA_ALIGN_PTR);
888 mtpp += (size - sizeof(struct malloc_type *)) /
889 sizeof(struct malloc_type *);
890 *mtpp = mtp;
891 }
892 #endif
893
894 #ifdef MALLOC_DEBUG
895 static int
free_dbg(void ** addrp,struct malloc_type * mtp)896 free_dbg(void **addrp, struct malloc_type *mtp)
897 {
898 void *addr;
899
900 addr = *addrp;
901 KASSERT(mtp->ks_version == M_VERSION, ("free: bad malloc type version"));
902 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
903 ("free: called with spinlock or critical section held"));
904
905 /* free(NULL, ...) does nothing */
906 if (addr == NULL)
907 return (EJUSTRETURN);
908
909 #ifdef DEBUG_MEMGUARD
910 if (is_memguard_addr(addr)) {
911 memguard_free(addr);
912 return (EJUSTRETURN);
913 }
914 #endif
915
916 #ifdef DEBUG_REDZONE
917 redzone_check(addr);
918 *addrp = redzone_addr_ntor(addr);
919 #endif
920
921 return (0);
922 }
923 #endif
924
925 static __always_inline void
_free(void * addr,struct malloc_type * mtp,bool dozero)926 _free(void *addr, struct malloc_type *mtp, bool dozero)
927 {
928 uma_zone_t zone;
929 uma_slab_t slab;
930 u_long size;
931
932 #ifdef MALLOC_DEBUG
933 if (free_dbg(&addr, mtp) != 0)
934 return;
935 #endif
936 /* free(NULL, ...) does nothing */
937 if (addr == NULL)
938 return;
939
940 vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
941 if (slab == NULL)
942 panic("%s(%d): address %p(%p) has not been allocated", __func__,
943 dozero, addr, (void *)((uintptr_t)addr & (~UMA_SLAB_MASK)));
944
945 switch (GET_SLAB_COOKIE(slab)) {
946 case __predict_true(SLAB_COOKIE_SLAB_PTR):
947 size = zone->uz_size;
948 #if defined(INVARIANTS) && !defined(KASAN)
949 free_save_type(addr, mtp, size);
950 #endif
951 if (dozero) {
952 kasan_mark(addr, size, size, 0);
953 explicit_bzero(addr, size);
954 }
955 uma_zfree_arg(zone, addr, slab);
956 break;
957 case SLAB_COOKIE_MALLOC_LARGE:
958 size = malloc_large_size(slab);
959 if (dozero) {
960 kasan_mark(addr, size, size, 0);
961 explicit_bzero(addr, size);
962 }
963 free_large(addr, size);
964 break;
965 case SLAB_COOKIE_CONTIG_MALLOC:
966 size = round_page(contigmalloc_size(slab));
967 if (dozero)
968 explicit_bzero(addr, size);
969 kmem_free(addr, size);
970 break;
971 default:
972 panic("%s(%d): addr %p slab %p with unknown cookie %d",
973 __func__, dozero, addr, slab, GET_SLAB_COOKIE(slab));
974 /* NOTREACHED */
975 }
976 malloc_type_freed(mtp, size);
977 }
978
979 /*
980 * free:
981 * Free a block of memory allocated by malloc/contigmalloc.
982 * This routine may not block.
983 */
984 void
free(void * addr,struct malloc_type * mtp)985 free(void *addr, struct malloc_type *mtp)
986 {
987 _free(addr, mtp, false);
988 }
989
990 /*
991 * zfree:
992 * Zero then free a block of memory allocated by malloc/contigmalloc.
993 * This routine may not block.
994 */
995 void
zfree(void * addr,struct malloc_type * mtp)996 zfree(void *addr, struct malloc_type *mtp)
997 {
998 _free(addr, mtp, true);
999 }
1000
1001 /*
1002 * realloc: change the size of a memory block
1003 */
1004 void *
realloc(void * addr,size_t size,struct malloc_type * mtp,int flags)1005 realloc(void *addr, size_t size, struct malloc_type *mtp, int flags)
1006 {
1007 #ifndef DEBUG_REDZONE
1008 uma_zone_t zone;
1009 uma_slab_t slab;
1010 #endif
1011 unsigned long alloc;
1012 void *newaddr;
1013
1014 KASSERT(mtp->ks_version == M_VERSION,
1015 ("realloc: bad malloc type version"));
1016 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
1017 ("realloc: called with spinlock or critical section held"));
1018
1019 /* realloc(NULL, ...) is equivalent to malloc(...) */
1020 if (addr == NULL)
1021 return (malloc(size, mtp, flags));
1022
1023 /*
1024 * XXX: Should report free of old memory and alloc of new memory to
1025 * per-CPU stats.
1026 */
1027
1028 #ifdef DEBUG_MEMGUARD
1029 if (is_memguard_addr(addr))
1030 return (memguard_realloc(addr, size, mtp, flags));
1031 #endif
1032
1033 #ifdef DEBUG_REDZONE
1034 alloc = redzone_get_size(addr);
1035 #else
1036 vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
1037
1038 /* Sanity check */
1039 KASSERT(slab != NULL,
1040 ("realloc: address %p out of range", (void *)addr));
1041
1042 /* Get the size of the original block */
1043 switch (GET_SLAB_COOKIE(slab)) {
1044 case __predict_true(SLAB_COOKIE_SLAB_PTR):
1045 alloc = zone->uz_size;
1046 break;
1047 case SLAB_COOKIE_MALLOC_LARGE:
1048 alloc = malloc_large_size(slab);
1049 break;
1050 default:
1051 #ifdef INVARIANTS
1052 panic("%s: called for addr %p of unsupported allocation type; "
1053 "slab %p cookie %d", __func__, addr, slab, GET_SLAB_COOKIE(slab));
1054 #endif
1055 return (NULL);
1056 }
1057
1058 /* Reuse the original block if appropriate */
1059 if (size <= alloc &&
1060 (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE)) {
1061 kasan_mark((void *)addr, size, alloc, KASAN_MALLOC_REDZONE);
1062 return (addr);
1063 }
1064 #endif /* !DEBUG_REDZONE */
1065
1066 /* Allocate a new, bigger (or smaller) block */
1067 if ((newaddr = malloc(size, mtp, flags)) == NULL)
1068 return (NULL);
1069
1070 /*
1071 * Copy over original contents. For KASAN, the redzone must be marked
1072 * valid before performing the copy.
1073 */
1074 kasan_mark(addr, alloc, alloc, 0);
1075 bcopy(addr, newaddr, min(size, alloc));
1076 free(addr, mtp);
1077 return (newaddr);
1078 }
1079
1080 /*
1081 * reallocf: same as realloc() but free memory on failure.
1082 */
1083 void *
reallocf(void * addr,size_t size,struct malloc_type * mtp,int flags)1084 reallocf(void *addr, size_t size, struct malloc_type *mtp, int flags)
1085 {
1086 void *mem;
1087
1088 if ((mem = realloc(addr, size, mtp, flags)) == NULL)
1089 free(addr, mtp);
1090 return (mem);
1091 }
1092
1093 /*
1094 * malloc_size: returns the number of bytes allocated for a request of the
1095 * specified size
1096 */
1097 size_t
malloc_size(size_t size)1098 malloc_size(size_t size)
1099 {
1100 int indx;
1101
1102 if (size > kmem_zmax)
1103 return (round_page(size));
1104 if (size & KMEM_ZMASK)
1105 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
1106 indx = kmemsize[size >> KMEM_ZSHIFT];
1107 return (kmemzones[indx].kz_size);
1108 }
1109
1110 /*
1111 * malloc_usable_size: returns the usable size of the allocation.
1112 */
1113 size_t
malloc_usable_size(const void * addr)1114 malloc_usable_size(const void *addr)
1115 {
1116 #ifndef DEBUG_REDZONE
1117 uma_zone_t zone;
1118 uma_slab_t slab;
1119 #endif
1120 u_long size;
1121
1122 if (addr == NULL)
1123 return (0);
1124
1125 #ifdef DEBUG_MEMGUARD
1126 if (is_memguard_addr(__DECONST(void *, addr)))
1127 return (memguard_get_req_size(addr));
1128 #endif
1129
1130 #ifdef DEBUG_REDZONE
1131 size = redzone_get_size(__DECONST(void *, addr));
1132 #else
1133 vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab);
1134 if (slab == NULL)
1135 panic("malloc_usable_size: address %p(%p) is not allocated",
1136 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
1137
1138 switch (GET_SLAB_COOKIE(slab)) {
1139 case __predict_true(SLAB_COOKIE_SLAB_PTR):
1140 size = zone->uz_size;
1141 break;
1142 case SLAB_COOKIE_MALLOC_LARGE:
1143 size = malloc_large_size(slab);
1144 break;
1145 default:
1146 __assert_unreachable();
1147 size = 0;
1148 break;
1149 }
1150 #endif
1151
1152 /*
1153 * Unmark the redzone to avoid reports from consumers who are
1154 * (presumably) about to use the full allocation size.
1155 */
1156 kasan_mark(addr, size, size, 0);
1157
1158 return (size);
1159 }
1160
1161 CTASSERT(VM_KMEM_SIZE_SCALE >= 1);
1162
1163 /*
1164 * Initialize the kernel memory (kmem) arena.
1165 */
1166 void
kmeminit(void)1167 kmeminit(void)
1168 {
1169 u_long mem_size;
1170 u_long tmp;
1171
1172 #ifdef VM_KMEM_SIZE
1173 if (vm_kmem_size == 0)
1174 vm_kmem_size = VM_KMEM_SIZE;
1175 #endif
1176 #ifdef VM_KMEM_SIZE_MIN
1177 if (vm_kmem_size_min == 0)
1178 vm_kmem_size_min = VM_KMEM_SIZE_MIN;
1179 #endif
1180 #ifdef VM_KMEM_SIZE_MAX
1181 if (vm_kmem_size_max == 0)
1182 vm_kmem_size_max = VM_KMEM_SIZE_MAX;
1183 #endif
1184 /*
1185 * Calculate the amount of kernel virtual address (KVA) space that is
1186 * preallocated to the kmem arena. In order to support a wide range
1187 * of machines, it is a function of the physical memory size,
1188 * specifically,
1189 *
1190 * min(max(physical memory size / VM_KMEM_SIZE_SCALE,
1191 * VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX)
1192 *
1193 * Every architecture must define an integral value for
1194 * VM_KMEM_SIZE_SCALE. However, the definitions of VM_KMEM_SIZE_MIN
1195 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and
1196 * ceiling on this preallocation, are optional. Typically,
1197 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on
1198 * a given architecture.
1199 */
1200 mem_size = vm_cnt.v_page_count;
1201 if (mem_size <= 32768) /* delphij XXX 128MB */
1202 kmem_zmax = PAGE_SIZE;
1203
1204 if (vm_kmem_size_scale < 1)
1205 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
1206
1207 /*
1208 * Check if we should use defaults for the "vm_kmem_size"
1209 * variable:
1210 */
1211 if (vm_kmem_size == 0) {
1212 vm_kmem_size = mem_size / vm_kmem_size_scale;
1213 vm_kmem_size = vm_kmem_size * PAGE_SIZE < vm_kmem_size ?
1214 vm_kmem_size_max : vm_kmem_size * PAGE_SIZE;
1215 if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min)
1216 vm_kmem_size = vm_kmem_size_min;
1217 if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
1218 vm_kmem_size = vm_kmem_size_max;
1219 }
1220 if (vm_kmem_size == 0)
1221 panic("Tune VM_KMEM_SIZE_* for the platform");
1222
1223 /*
1224 * The amount of KVA space that is preallocated to the
1225 * kmem arena can be set statically at compile-time or manually
1226 * through the kernel environment. However, it is still limited to
1227 * twice the physical memory size, which has been sufficient to handle
1228 * the most severe cases of external fragmentation in the kmem arena.
1229 */
1230 if (vm_kmem_size / 2 / PAGE_SIZE > mem_size)
1231 vm_kmem_size = 2 * mem_size * PAGE_SIZE;
1232
1233 vm_kmem_size = round_page(vm_kmem_size);
1234
1235 /*
1236 * With KASAN or KMSAN enabled, dynamically allocated kernel memory is
1237 * shadowed. Account for this when setting the UMA limit.
1238 */
1239 #if defined(KASAN)
1240 vm_kmem_size = (vm_kmem_size * KASAN_SHADOW_SCALE) /
1241 (KASAN_SHADOW_SCALE + 1);
1242 #elif defined(KMSAN)
1243 vm_kmem_size /= 3;
1244 #endif
1245
1246 #ifdef DEBUG_MEMGUARD
1247 tmp = memguard_fudge(vm_kmem_size, kernel_map);
1248 #else
1249 tmp = vm_kmem_size;
1250 #endif
1251 uma_set_limit(tmp);
1252
1253 #ifdef DEBUG_MEMGUARD
1254 /*
1255 * Initialize MemGuard if support compiled in. MemGuard is a
1256 * replacement allocator used for detecting tamper-after-free
1257 * scenarios as they occur. It is only used for debugging.
1258 */
1259 memguard_init(kernel_arena);
1260 #endif
1261 }
1262
1263 /*
1264 * Initialize the kernel memory allocator
1265 */
1266 /* ARGSUSED*/
1267 static void
mallocinit(void * dummy)1268 mallocinit(void *dummy)
1269 {
1270 int i;
1271 uint8_t indx;
1272
1273 mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
1274
1275 kmeminit();
1276
1277 if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX)
1278 kmem_zmax = KMEM_ZMAX;
1279
1280 for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
1281 int size = kmemzones[indx].kz_size;
1282 const char *name = kmemzones[indx].kz_name;
1283 size_t align;
1284 int subzone;
1285
1286 align = UMA_ALIGN_PTR;
1287 if (powerof2(size) && size > sizeof(void *))
1288 align = MIN(size, PAGE_SIZE) - 1;
1289 for (subzone = 0; subzone < numzones; subzone++) {
1290 kmemzones[indx].kz_zone[subzone] =
1291 uma_zcreate(name, size,
1292 #if defined(INVARIANTS) && !defined(KASAN) && !defined(KMSAN)
1293 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
1294 #else
1295 NULL, NULL, NULL, NULL,
1296 #endif
1297 align, UMA_ZONE_MALLOC);
1298 }
1299 for (;i <= size; i+= KMEM_ZBASE)
1300 kmemsize[i >> KMEM_ZSHIFT] = indx;
1301 }
1302 }
1303 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_SECOND, mallocinit, NULL);
1304
1305 void
malloc_init(void * data)1306 malloc_init(void *data)
1307 {
1308 struct malloc_type_internal *mtip;
1309 struct malloc_type *mtp;
1310
1311 KASSERT(vm_cnt.v_page_count != 0,
1312 ("malloc_init() called before vm_mem_init()"));
1313
1314 mtp = data;
1315 if (mtp->ks_version != M_VERSION)
1316 panic("malloc_init: type %s with unsupported version %lu",
1317 mtp->ks_shortdesc, mtp->ks_version);
1318
1319 mtip = &mtp->ks_mti;
1320 mtip->mti_stats = uma_zalloc_pcpu(pcpu_zone_64, M_WAITOK | M_ZERO);
1321 mtp_set_subzone(mtp);
1322
1323 mtx_lock(&malloc_mtx);
1324 mtp->ks_next = kmemstatistics;
1325 kmemstatistics = mtp;
1326 kmemcount++;
1327 mtx_unlock(&malloc_mtx);
1328 }
1329
1330 void
malloc_uninit(void * data)1331 malloc_uninit(void *data)
1332 {
1333 struct malloc_type_internal *mtip;
1334 struct malloc_type_stats *mtsp;
1335 struct malloc_type *mtp, *temp;
1336 long temp_allocs, temp_bytes;
1337 int i;
1338
1339 mtp = data;
1340 KASSERT(mtp->ks_version == M_VERSION,
1341 ("malloc_uninit: bad malloc type version"));
1342
1343 mtx_lock(&malloc_mtx);
1344 mtip = &mtp->ks_mti;
1345 if (mtp != kmemstatistics) {
1346 for (temp = kmemstatistics; temp != NULL;
1347 temp = temp->ks_next) {
1348 if (temp->ks_next == mtp) {
1349 temp->ks_next = mtp->ks_next;
1350 break;
1351 }
1352 }
1353 KASSERT(temp,
1354 ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc));
1355 } else
1356 kmemstatistics = mtp->ks_next;
1357 kmemcount--;
1358 mtx_unlock(&malloc_mtx);
1359
1360 /*
1361 * Look for memory leaks.
1362 */
1363 temp_allocs = temp_bytes = 0;
1364 for (i = 0; i <= mp_maxid; i++) {
1365 mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1366 temp_allocs += mtsp->mts_numallocs;
1367 temp_allocs -= mtsp->mts_numfrees;
1368 temp_bytes += mtsp->mts_memalloced;
1369 temp_bytes -= mtsp->mts_memfreed;
1370 }
1371 if (temp_allocs > 0 || temp_bytes > 0) {
1372 printf("Warning: memory type %s leaked memory on destroy "
1373 "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc,
1374 temp_allocs, temp_bytes);
1375 }
1376
1377 uma_zfree_pcpu(pcpu_zone_64, mtip->mti_stats);
1378 }
1379
1380 struct malloc_type *
malloc_desc2type(const char * desc)1381 malloc_desc2type(const char *desc)
1382 {
1383 struct malloc_type *mtp;
1384
1385 mtx_assert(&malloc_mtx, MA_OWNED);
1386 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1387 if (strcmp(mtp->ks_shortdesc, desc) == 0)
1388 return (mtp);
1389 }
1390 return (NULL);
1391 }
1392
1393 static int
sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)1394 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
1395 {
1396 struct malloc_type_stream_header mtsh;
1397 struct malloc_type_internal *mtip;
1398 struct malloc_type_stats *mtsp, zeromts;
1399 struct malloc_type_header mth;
1400 struct malloc_type *mtp;
1401 int error, i;
1402 struct sbuf sbuf;
1403
1404 error = sysctl_wire_old_buffer(req, 0);
1405 if (error != 0)
1406 return (error);
1407 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
1408 sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL);
1409 mtx_lock(&malloc_mtx);
1410
1411 bzero(&zeromts, sizeof(zeromts));
1412
1413 /*
1414 * Insert stream header.
1415 */
1416 bzero(&mtsh, sizeof(mtsh));
1417 mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
1418 mtsh.mtsh_maxcpus = MAXCPU;
1419 mtsh.mtsh_count = kmemcount;
1420 (void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh));
1421
1422 /*
1423 * Insert alternating sequence of type headers and type statistics.
1424 */
1425 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1426 mtip = &mtp->ks_mti;
1427
1428 /*
1429 * Insert type header.
1430 */
1431 bzero(&mth, sizeof(mth));
1432 strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
1433 (void)sbuf_bcat(&sbuf, &mth, sizeof(mth));
1434
1435 /*
1436 * Insert type statistics for each CPU.
1437 */
1438 for (i = 0; i <= mp_maxid; i++) {
1439 mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1440 (void)sbuf_bcat(&sbuf, mtsp, sizeof(*mtsp));
1441 }
1442 /*
1443 * Fill in the missing CPUs.
1444 */
1445 for (; i < MAXCPU; i++) {
1446 (void)sbuf_bcat(&sbuf, &zeromts, sizeof(zeromts));
1447 }
1448 }
1449 mtx_unlock(&malloc_mtx);
1450 error = sbuf_finish(&sbuf);
1451 sbuf_delete(&sbuf);
1452 return (error);
1453 }
1454
1455 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats,
1456 CTLFLAG_RD | CTLTYPE_STRUCT | CTLFLAG_MPSAFE, 0, 0,
1457 sysctl_kern_malloc_stats, "s,malloc_type_ustats",
1458 "Return malloc types");
1459
1460 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
1461 "Count of kernel malloc types");
1462
1463 void
malloc_type_list(malloc_type_list_func_t * func,void * arg)1464 malloc_type_list(malloc_type_list_func_t *func, void *arg)
1465 {
1466 struct malloc_type *mtp, **bufmtp;
1467 int count, i;
1468 size_t buflen;
1469
1470 mtx_lock(&malloc_mtx);
1471 restart:
1472 mtx_assert(&malloc_mtx, MA_OWNED);
1473 count = kmemcount;
1474 mtx_unlock(&malloc_mtx);
1475
1476 buflen = sizeof(struct malloc_type *) * count;
1477 bufmtp = malloc(buflen, M_TEMP, M_WAITOK);
1478
1479 mtx_lock(&malloc_mtx);
1480
1481 if (count < kmemcount) {
1482 free(bufmtp, M_TEMP);
1483 goto restart;
1484 }
1485
1486 for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++)
1487 bufmtp[i] = mtp;
1488
1489 mtx_unlock(&malloc_mtx);
1490
1491 for (i = 0; i < count; i++)
1492 (func)(bufmtp[i], arg);
1493
1494 free(bufmtp, M_TEMP);
1495 }
1496
1497 #ifdef DDB
1498 static int64_t
get_malloc_stats(const struct malloc_type_internal * mtip,uint64_t * allocs,uint64_t * inuse)1499 get_malloc_stats(const struct malloc_type_internal *mtip, uint64_t *allocs,
1500 uint64_t *inuse)
1501 {
1502 const struct malloc_type_stats *mtsp;
1503 uint64_t frees, alloced, freed;
1504 int i;
1505
1506 *allocs = 0;
1507 frees = 0;
1508 alloced = 0;
1509 freed = 0;
1510 for (i = 0; i <= mp_maxid; i++) {
1511 mtsp = zpcpu_get_cpu(mtip->mti_stats, i);
1512
1513 *allocs += mtsp->mts_numallocs;
1514 frees += mtsp->mts_numfrees;
1515 alloced += mtsp->mts_memalloced;
1516 freed += mtsp->mts_memfreed;
1517 }
1518 *inuse = *allocs - frees;
1519 return (alloced - freed);
1520 }
1521
DB_SHOW_COMMAND_FLAGS(malloc,db_show_malloc,DB_CMD_MEMSAFE)1522 DB_SHOW_COMMAND_FLAGS(malloc, db_show_malloc, DB_CMD_MEMSAFE)
1523 {
1524 const char *fmt_hdr, *fmt_entry;
1525 struct malloc_type *mtp;
1526 uint64_t allocs, inuse;
1527 int64_t size;
1528 /* variables for sorting */
1529 struct malloc_type *last_mtype, *cur_mtype;
1530 int64_t cur_size, last_size;
1531 int ties;
1532
1533 if (modif[0] == 'i') {
1534 fmt_hdr = "%s,%s,%s,%s\n";
1535 fmt_entry = "\"%s\",%ju,%jdK,%ju\n";
1536 } else {
1537 fmt_hdr = "%18s %12s %12s %12s\n";
1538 fmt_entry = "%18s %12ju %12jdK %12ju\n";
1539 }
1540
1541 db_printf(fmt_hdr, "Type", "InUse", "MemUse", "Requests");
1542
1543 /* Select sort, largest size first. */
1544 last_mtype = NULL;
1545 last_size = INT64_MAX;
1546 for (;;) {
1547 cur_mtype = NULL;
1548 cur_size = -1;
1549 ties = 0;
1550
1551 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1552 /*
1553 * In the case of size ties, print out mtypes
1554 * in the order they are encountered. That is,
1555 * when we encounter the most recently output
1556 * mtype, we have already printed all preceding
1557 * ties, and we must print all following ties.
1558 */
1559 if (mtp == last_mtype) {
1560 ties = 1;
1561 continue;
1562 }
1563 size = get_malloc_stats(&mtp->ks_mti, &allocs,
1564 &inuse);
1565 if (size > cur_size && size < last_size + ties) {
1566 cur_size = size;
1567 cur_mtype = mtp;
1568 }
1569 }
1570 if (cur_mtype == NULL)
1571 break;
1572
1573 size = get_malloc_stats(&cur_mtype->ks_mti, &allocs, &inuse);
1574 db_printf(fmt_entry, cur_mtype->ks_shortdesc, inuse,
1575 howmany(size, 1024), allocs);
1576
1577 if (db_pager_quit)
1578 break;
1579
1580 last_mtype = cur_mtype;
1581 last_size = cur_size;
1582 }
1583 }
1584
1585 #if MALLOC_DEBUG_MAXZONES > 1
DB_SHOW_COMMAND(multizone_matches,db_show_multizone_matches)1586 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches)
1587 {
1588 struct malloc_type_internal *mtip;
1589 struct malloc_type *mtp;
1590 u_int subzone;
1591
1592 if (!have_addr) {
1593 db_printf("Usage: show multizone_matches <malloc type/addr>\n");
1594 return;
1595 }
1596 mtp = (void *)addr;
1597 if (mtp->ks_version != M_VERSION) {
1598 db_printf("Version %lx does not match expected %x\n",
1599 mtp->ks_version, M_VERSION);
1600 return;
1601 }
1602
1603 mtip = &mtp->ks_mti;
1604 subzone = mtip->mti_zone;
1605
1606 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1607 mtip = &mtp->ks_mti;
1608 if (mtip->mti_zone != subzone)
1609 continue;
1610 db_printf("%s\n", mtp->ks_shortdesc);
1611 if (db_pager_quit)
1612 break;
1613 }
1614 }
1615 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
1616 #endif /* DDB */
1617