1 /*-
2 * Copyright (c) 1987, 1991, 1993
3 * The Regents of the University of California.
4 * Copyright (c) 2005-2009 Robert N. M. Watson
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 4. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 *
31 * @(#)kern_malloc.c 8.3 (Berkeley) 1/4/94
32 */
33
34 /*
35 * Kernel malloc(9) implementation -- general purpose kernel memory allocator
36 * based on memory types. Back end is implemented using the UMA(9) zone
37 * allocator. A set of fixed-size buckets are used for smaller allocations,
38 * and a special UMA allocation interface is used for larger allocations.
39 * Callers declare memory types, and statistics are maintained independently
40 * for each memory type. Statistics are maintained per-CPU for performance
41 * reasons. See malloc(9) and comments in malloc.h for a detailed
42 * description.
43 */
44
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47
48 #include "opt_ddb.h"
49 #include "opt_kdtrace.h"
50 #include "opt_vm.h"
51
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/kdb.h>
55 #include <sys/kernel.h>
56 #include <sys/lock.h>
57 #include <sys/malloc.h>
58 #include <sys/mutex.h>
59 #include <sys/vmmeter.h>
60 #include <sys/proc.h>
61 #include <sys/sbuf.h>
62 #include <sys/sysctl.h>
63 #include <sys/time.h>
64 #include <sys/vmem.h>
65
66 #include <vm/vm.h>
67 #include <vm/pmap.h>
68 #include <vm/vm_pageout.h>
69 #include <vm/vm_param.h>
70 #include <vm/vm_kern.h>
71 #include <vm/vm_extern.h>
72 #include <vm/vm_map.h>
73 #include <vm/vm_page.h>
74 #include <vm/uma.h>
75 #include <vm/uma_int.h>
76 #include <vm/uma_dbg.h>
77
78 #ifdef DEBUG_MEMGUARD
79 #include <vm/memguard.h>
80 #endif
81 #ifdef DEBUG_REDZONE
82 #include <vm/redzone.h>
83 #endif
84
85 #if defined(INVARIANTS) && defined(__i386__)
86 #include <machine/cpu.h>
87 #endif
88
89 #include <ddb/ddb.h>
90
91 #ifdef KDTRACE_HOOKS
92 #include <sys/dtrace_bsd.h>
93
94 dtrace_malloc_probe_func_t dtrace_malloc_probe;
95 #endif
96
97 /*
98 * When realloc() is called, if the new size is sufficiently smaller than
99 * the old size, realloc() will allocate a new, smaller block to avoid
100 * wasting memory. 'Sufficiently smaller' is defined as: newsize <=
101 * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'.
102 */
103 #ifndef REALLOC_FRACTION
104 #define REALLOC_FRACTION 1 /* new block if <= half the size */
105 #endif
106
107 /*
108 * Centrally define some common malloc types.
109 */
110 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches");
111 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
112 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers");
113
114 MALLOC_DEFINE(M_IP6OPT, "ip6opt", "IPv6 options");
115 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
116
117 static struct malloc_type *kmemstatistics;
118 static int kmemcount;
119
120 #define KMEM_ZSHIFT 4
121 #define KMEM_ZBASE 16
122 #define KMEM_ZMASK (KMEM_ZBASE - 1)
123
124 #define KMEM_ZMAX 65536
125 #define KMEM_ZSIZE (KMEM_ZMAX >> KMEM_ZSHIFT)
126 static uint8_t kmemsize[KMEM_ZSIZE + 1];
127
128 #ifndef MALLOC_DEBUG_MAXZONES
129 #define MALLOC_DEBUG_MAXZONES 1
130 #endif
131 static int numzones = MALLOC_DEBUG_MAXZONES;
132
133 /*
134 * Small malloc(9) memory allocations are allocated from a set of UMA buckets
135 * of various sizes.
136 *
137 * XXX: The comment here used to read "These won't be powers of two for
138 * long." It's possible that a significant amount of wasted memory could be
139 * recovered by tuning the sizes of these buckets.
140 */
141 struct {
142 int kz_size;
143 char *kz_name;
144 uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES];
145 } kmemzones[] = {
146 {16, "16", },
147 {32, "32", },
148 {64, "64", },
149 {128, "128", },
150 {256, "256", },
151 {512, "512", },
152 {1024, "1024", },
153 {2048, "2048", },
154 {4096, "4096", },
155 {8192, "8192", },
156 {16384, "16384", },
157 {32768, "32768", },
158 {65536, "65536", },
159 {0, NULL},
160 };
161
162 /*
163 * Zone to allocate malloc type descriptions from. For ABI reasons, memory
164 * types are described by a data structure passed by the declaring code, but
165 * the malloc(9) implementation has its own data structure describing the
166 * type and statistics. This permits the malloc(9)-internal data structures
167 * to be modified without breaking binary-compiled kernel modules that
168 * declare malloc types.
169 */
170 static uma_zone_t mt_zone;
171
172 u_long vm_kmem_size;
173 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0,
174 "Size of kernel memory");
175
176 static u_long kmem_zmax = KMEM_ZMAX;
177 SYSCTL_ULONG(_vm, OID_AUTO, kmem_zmax, CTLFLAG_RDTUN, &kmem_zmax, 0,
178 "Maximum allocation size that malloc(9) would use UMA as backend");
179
180 static u_long vm_kmem_size_min;
181 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0,
182 "Minimum size of kernel memory");
183
184 static u_long vm_kmem_size_max;
185 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0,
186 "Maximum size of kernel memory");
187
188 static u_int vm_kmem_size_scale;
189 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0,
190 "Scale factor for kernel memory size");
191
192 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS);
193 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size,
194 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
195 sysctl_kmem_map_size, "LU", "Current kmem allocation size");
196
197 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS);
198 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free,
199 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
200 sysctl_kmem_map_free, "LU", "Free space in kmem");
201
202 /*
203 * The malloc_mtx protects the kmemstatistics linked list.
204 */
205 struct mtx malloc_mtx;
206
207 #ifdef MALLOC_PROFILE
208 uint64_t krequests[KMEM_ZSIZE + 1];
209
210 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS);
211 #endif
212
213 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS);
214
215 /*
216 * time_uptime of the last malloc(9) failure (induced or real).
217 */
218 static time_t t_malloc_fail;
219
220 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1)
221 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0,
222 "Kernel malloc debugging options");
223 #endif
224
225 /*
226 * malloc(9) fault injection -- cause malloc failures every (n) mallocs when
227 * the caller specifies M_NOWAIT. If set to 0, no failures are caused.
228 */
229 #ifdef MALLOC_MAKE_FAILURES
230 static int malloc_failure_rate;
231 static int malloc_nowait_count;
232 static int malloc_failure_count;
233 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RW,
234 &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail");
235 TUNABLE_INT("debug.malloc.failure_rate", &malloc_failure_rate);
236 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD,
237 &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures");
238 #endif
239
240 static int
sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS)241 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS)
242 {
243 u_long size;
244
245 size = vmem_size(kmem_arena, VMEM_ALLOC);
246 return (sysctl_handle_long(oidp, &size, 0, req));
247 }
248
249 static int
sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS)250 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS)
251 {
252 u_long size;
253
254 size = vmem_size(kmem_arena, VMEM_FREE);
255 return (sysctl_handle_long(oidp, &size, 0, req));
256 }
257
258 /*
259 * malloc(9) uma zone separation -- sub-page buffer overruns in one
260 * malloc type will affect only a subset of other malloc types.
261 */
262 #if MALLOC_DEBUG_MAXZONES > 1
263 static void
tunable_set_numzones(void)264 tunable_set_numzones(void)
265 {
266
267 TUNABLE_INT_FETCH("debug.malloc.numzones",
268 &numzones);
269
270 /* Sanity check the number of malloc uma zones. */
271 if (numzones <= 0)
272 numzones = 1;
273 if (numzones > MALLOC_DEBUG_MAXZONES)
274 numzones = MALLOC_DEBUG_MAXZONES;
275 }
276 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL);
277 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN,
278 &numzones, 0, "Number of malloc uma subzones");
279
280 /*
281 * Any number that changes regularly is an okay choice for the
282 * offset. Build numbers are pretty good of you have them.
283 */
284 static u_int zone_offset = __FreeBSD_version;
285 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset);
286 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN,
287 &zone_offset, 0, "Separate malloc types by examining the "
288 "Nth character in the malloc type short description.");
289
290 static u_int
mtp_get_subzone(const char * desc)291 mtp_get_subzone(const char *desc)
292 {
293 size_t len;
294 u_int val;
295
296 if (desc == NULL || (len = strlen(desc)) == 0)
297 return (0);
298 val = desc[zone_offset % len];
299 return (val % numzones);
300 }
301 #elif MALLOC_DEBUG_MAXZONES == 0
302 #error "MALLOC_DEBUG_MAXZONES must be positive."
303 #else
304 static inline u_int
mtp_get_subzone(const char * desc)305 mtp_get_subzone(const char *desc)
306 {
307
308 return (0);
309 }
310 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
311
312 int
malloc_last_fail(void)313 malloc_last_fail(void)
314 {
315
316 return (time_uptime - t_malloc_fail);
317 }
318
319 /*
320 * An allocation has succeeded -- update malloc type statistics for the
321 * amount of bucket size. Occurs within a critical section so that the
322 * thread isn't preempted and doesn't migrate while updating per-PCU
323 * statistics.
324 */
325 static void
malloc_type_zone_allocated(struct malloc_type * mtp,unsigned long size,int zindx)326 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size,
327 int zindx)
328 {
329 struct malloc_type_internal *mtip;
330 struct malloc_type_stats *mtsp;
331
332 critical_enter();
333 mtip = mtp->ks_handle;
334 mtsp = &mtip->mti_stats[curcpu];
335 if (size > 0) {
336 mtsp->mts_memalloced += size;
337 mtsp->mts_numallocs++;
338 }
339 if (zindx != -1)
340 mtsp->mts_size |= 1 << zindx;
341
342 #ifdef KDTRACE_HOOKS
343 if (dtrace_malloc_probe != NULL) {
344 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC];
345 if (probe_id != 0)
346 (dtrace_malloc_probe)(probe_id,
347 (uintptr_t) mtp, (uintptr_t) mtip,
348 (uintptr_t) mtsp, size, zindx);
349 }
350 #endif
351
352 critical_exit();
353 }
354
355 void
malloc_type_allocated(struct malloc_type * mtp,unsigned long size)356 malloc_type_allocated(struct malloc_type *mtp, unsigned long size)
357 {
358
359 if (size > 0)
360 malloc_type_zone_allocated(mtp, size, -1);
361 }
362
363 /*
364 * A free operation has occurred -- update malloc type statistics for the
365 * amount of the bucket size. Occurs within a critical section so that the
366 * thread isn't preempted and doesn't migrate while updating per-CPU
367 * statistics.
368 */
369 void
malloc_type_freed(struct malloc_type * mtp,unsigned long size)370 malloc_type_freed(struct malloc_type *mtp, unsigned long size)
371 {
372 struct malloc_type_internal *mtip;
373 struct malloc_type_stats *mtsp;
374
375 critical_enter();
376 mtip = mtp->ks_handle;
377 mtsp = &mtip->mti_stats[curcpu];
378 mtsp->mts_memfreed += size;
379 mtsp->mts_numfrees++;
380
381 #ifdef KDTRACE_HOOKS
382 if (dtrace_malloc_probe != NULL) {
383 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE];
384 if (probe_id != 0)
385 (dtrace_malloc_probe)(probe_id,
386 (uintptr_t) mtp, (uintptr_t) mtip,
387 (uintptr_t) mtsp, size, 0);
388 }
389 #endif
390
391 critical_exit();
392 }
393
394 /*
395 * contigmalloc:
396 *
397 * Allocate a block of physically contiguous memory.
398 *
399 * If M_NOWAIT is set, this routine will not block and return NULL if
400 * the allocation fails.
401 */
402 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)403 contigmalloc(unsigned long size, struct malloc_type *type, int flags,
404 vm_paddr_t low, vm_paddr_t high, unsigned long alignment,
405 vm_paddr_t boundary)
406 {
407 void *ret;
408
409 ret = (void *)kmem_alloc_contig(kernel_arena, size, flags, low, high,
410 alignment, boundary, VM_MEMATTR_DEFAULT);
411 if (ret != NULL)
412 malloc_type_allocated(type, round_page(size));
413 return (ret);
414 }
415
416 /*
417 * contigfree:
418 *
419 * Free a block of memory allocated by contigmalloc.
420 *
421 * This routine may not block.
422 */
423 void
contigfree(void * addr,unsigned long size,struct malloc_type * type)424 contigfree(void *addr, unsigned long size, struct malloc_type *type)
425 {
426
427 kmem_free(kernel_arena, (vm_offset_t)addr, size);
428 malloc_type_freed(type, round_page(size));
429 }
430
431 /*
432 * malloc:
433 *
434 * Allocate a block of memory.
435 *
436 * If M_NOWAIT is set, this routine will not block and return NULL if
437 * the allocation fails.
438 */
439 void *
malloc(unsigned long size,struct malloc_type * mtp,int flags)440 malloc(unsigned long size, struct malloc_type *mtp, int flags)
441 {
442 int indx;
443 struct malloc_type_internal *mtip;
444 caddr_t va;
445 uma_zone_t zone;
446 #if defined(DIAGNOSTIC) || defined(DEBUG_REDZONE)
447 unsigned long osize = size;
448 #endif
449
450 #ifdef INVARIANTS
451 KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic"));
452 /*
453 * Check that exactly one of M_WAITOK or M_NOWAIT is specified.
454 */
455 indx = flags & (M_WAITOK | M_NOWAIT);
456 if (indx != M_NOWAIT && indx != M_WAITOK) {
457 static struct timeval lasterr;
458 static int curerr, once;
459 if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) {
460 printf("Bad malloc flags: %x\n", indx);
461 kdb_backtrace();
462 flags |= M_WAITOK;
463 once++;
464 }
465 }
466 #endif
467 #ifdef MALLOC_MAKE_FAILURES
468 if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
469 atomic_add_int(&malloc_nowait_count, 1);
470 if ((malloc_nowait_count % malloc_failure_rate) == 0) {
471 atomic_add_int(&malloc_failure_count, 1);
472 t_malloc_fail = time_uptime;
473 return (NULL);
474 }
475 }
476 #endif
477 if (flags & M_WAITOK)
478 KASSERT(curthread->td_intr_nesting_level == 0,
479 ("malloc(M_WAITOK) in interrupt context"));
480
481 #ifdef DEBUG_MEMGUARD
482 if (memguard_cmp_mtp(mtp, size)) {
483 va = memguard_alloc(size, flags);
484 if (va != NULL)
485 return (va);
486 /* This is unfortunate but should not be fatal. */
487 }
488 #endif
489
490 #ifdef DEBUG_REDZONE
491 size = redzone_size_ntor(size);
492 #endif
493
494 if (size <= kmem_zmax) {
495 mtip = mtp->ks_handle;
496 if (size & KMEM_ZMASK)
497 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
498 indx = kmemsize[size >> KMEM_ZSHIFT];
499 KASSERT(mtip->mti_zone < numzones,
500 ("mti_zone %u out of range %d",
501 mtip->mti_zone, numzones));
502 zone = kmemzones[indx].kz_zone[mtip->mti_zone];
503 #ifdef MALLOC_PROFILE
504 krequests[size >> KMEM_ZSHIFT]++;
505 #endif
506 va = uma_zalloc(zone, flags);
507 if (va != NULL)
508 size = zone->uz_size;
509 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
510 } else {
511 size = roundup(size, PAGE_SIZE);
512 zone = NULL;
513 va = uma_large_malloc(size, flags);
514 malloc_type_allocated(mtp, va == NULL ? 0 : size);
515 }
516 if (flags & M_WAITOK)
517 KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL"));
518 else if (va == NULL)
519 t_malloc_fail = time_uptime;
520 #ifdef DIAGNOSTIC
521 if (va != NULL && !(flags & M_ZERO)) {
522 memset(va, 0x70, osize);
523 }
524 #endif
525 #ifdef DEBUG_REDZONE
526 if (va != NULL)
527 va = redzone_setup(va, osize);
528 #endif
529 return ((void *) va);
530 }
531
532 /*
533 * free:
534 *
535 * Free a block of memory allocated by malloc.
536 *
537 * This routine may not block.
538 */
539 void
free(void * addr,struct malloc_type * mtp)540 free(void *addr, struct malloc_type *mtp)
541 {
542 uma_slab_t slab;
543 u_long size;
544
545 KASSERT(mtp->ks_magic == M_MAGIC, ("free: bad malloc type magic"));
546
547 /* free(NULL, ...) does nothing */
548 if (addr == NULL)
549 return;
550
551 #ifdef DEBUG_MEMGUARD
552 if (is_memguard_addr(addr)) {
553 memguard_free(addr);
554 return;
555 }
556 #endif
557
558 #ifdef DEBUG_REDZONE
559 redzone_check(addr);
560 addr = redzone_addr_ntor(addr);
561 #endif
562
563 slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK));
564
565 if (slab == NULL)
566 panic("free: address %p(%p) has not been allocated.\n",
567 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
568
569 if (!(slab->us_flags & UMA_SLAB_MALLOC)) {
570 #ifdef INVARIANTS
571 struct malloc_type **mtpp = addr;
572 #endif
573 size = slab->us_keg->uk_size;
574 #ifdef INVARIANTS
575 /*
576 * Cache a pointer to the malloc_type that most recently freed
577 * this memory here. This way we know who is most likely to
578 * have stepped on it later.
579 *
580 * This code assumes that size is a multiple of 8 bytes for
581 * 64 bit machines
582 */
583 mtpp = (struct malloc_type **)
584 ((unsigned long)mtpp & ~UMA_ALIGN_PTR);
585 mtpp += (size - sizeof(struct malloc_type *)) /
586 sizeof(struct malloc_type *);
587 *mtpp = mtp;
588 #endif
589 uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab);
590 } else {
591 size = slab->us_size;
592 uma_large_free(slab);
593 }
594 malloc_type_freed(mtp, size);
595 }
596
597 /*
598 * realloc: change the size of a memory block
599 */
600 void *
realloc(void * addr,unsigned long size,struct malloc_type * mtp,int flags)601 realloc(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
602 {
603 uma_slab_t slab;
604 unsigned long alloc;
605 void *newaddr;
606
607 KASSERT(mtp->ks_magic == M_MAGIC,
608 ("realloc: bad malloc type magic"));
609
610 /* realloc(NULL, ...) is equivalent to malloc(...) */
611 if (addr == NULL)
612 return (malloc(size, mtp, flags));
613
614 /*
615 * XXX: Should report free of old memory and alloc of new memory to
616 * per-CPU stats.
617 */
618
619 #ifdef DEBUG_MEMGUARD
620 if (is_memguard_addr(addr))
621 return (memguard_realloc(addr, size, mtp, flags));
622 #endif
623
624 #ifdef DEBUG_REDZONE
625 slab = NULL;
626 alloc = redzone_get_size(addr);
627 #else
628 slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK));
629
630 /* Sanity check */
631 KASSERT(slab != NULL,
632 ("realloc: address %p out of range", (void *)addr));
633
634 /* Get the size of the original block */
635 if (!(slab->us_flags & UMA_SLAB_MALLOC))
636 alloc = slab->us_keg->uk_size;
637 else
638 alloc = slab->us_size;
639
640 /* Reuse the original block if appropriate */
641 if (size <= alloc
642 && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE))
643 return (addr);
644 #endif /* !DEBUG_REDZONE */
645
646 /* Allocate a new, bigger (or smaller) block */
647 if ((newaddr = malloc(size, mtp, flags)) == NULL)
648 return (NULL);
649
650 /* Copy over original contents */
651 bcopy(addr, newaddr, min(size, alloc));
652 free(addr, mtp);
653 return (newaddr);
654 }
655
656 /*
657 * reallocf: same as realloc() but free memory on failure.
658 */
659 void *
reallocf(void * addr,unsigned long size,struct malloc_type * mtp,int flags)660 reallocf(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
661 {
662 void *mem;
663
664 if ((mem = realloc(addr, size, mtp, flags)) == NULL)
665 free(addr, mtp);
666 return (mem);
667 }
668
669 /*
670 * Wake the uma reclamation pagedaemon thread when we exhaust KVA. It
671 * will call the lowmem handler and uma_reclaim() callbacks in a
672 * context that is safe.
673 */
674 static void
kmem_reclaim(vmem_t * vm,int flags)675 kmem_reclaim(vmem_t *vm, int flags)
676 {
677
678 uma_reclaim_wakeup();
679 pagedaemon_wakeup();
680 }
681
682 CTASSERT(VM_KMEM_SIZE_SCALE >= 1);
683
684 /*
685 * Initialize the kernel memory (kmem) arena.
686 */
687 void
kmeminit(void)688 kmeminit(void)
689 {
690 u_long mem_size, tmp;
691
692 /*
693 * Calculate the amount of kernel virtual address (KVA) space that is
694 * preallocated to the kmem arena. In order to support a wide range
695 * of machines, it is a function of the physical memory size,
696 * specifically,
697 *
698 * min(max(physical memory size / VM_KMEM_SIZE_SCALE,
699 * VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX)
700 *
701 * Every architecture must define an integral value for
702 * VM_KMEM_SIZE_SCALE. However, the definitions of VM_KMEM_SIZE_MIN
703 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and
704 * ceiling on this preallocation, are optional. Typically,
705 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on
706 * a given architecture.
707 */
708 mem_size = cnt.v_page_count;
709
710 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
711 TUNABLE_INT_FETCH("vm.kmem_size_scale", &vm_kmem_size_scale);
712 if (vm_kmem_size_scale < 1)
713 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
714
715 vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE;
716
717 #if defined(VM_KMEM_SIZE_MIN)
718 vm_kmem_size_min = VM_KMEM_SIZE_MIN;
719 #endif
720 TUNABLE_ULONG_FETCH("vm.kmem_size_min", &vm_kmem_size_min);
721 if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min)
722 vm_kmem_size = vm_kmem_size_min;
723
724 #if defined(VM_KMEM_SIZE_MAX)
725 vm_kmem_size_max = VM_KMEM_SIZE_MAX;
726 #endif
727 TUNABLE_ULONG_FETCH("vm.kmem_size_max", &vm_kmem_size_max);
728 if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
729 vm_kmem_size = vm_kmem_size_max;
730
731 /*
732 * Alternatively, the amount of KVA space that is preallocated to the
733 * kmem arena can be set statically at compile-time or manually
734 * through the kernel environment. However, it is still limited to
735 * twice the physical memory size, which has been sufficient to handle
736 * the most severe cases of external fragmentation in the kmem arena.
737 */
738 #if defined(VM_KMEM_SIZE)
739 vm_kmem_size = VM_KMEM_SIZE;
740 #endif
741 TUNABLE_ULONG_FETCH("vm.kmem_size", &vm_kmem_size);
742 if (vm_kmem_size / 2 / PAGE_SIZE > mem_size)
743 vm_kmem_size = 2 * mem_size * PAGE_SIZE;
744
745 vm_kmem_size = round_page(vm_kmem_size);
746 #ifdef DEBUG_MEMGUARD
747 tmp = memguard_fudge(vm_kmem_size, kernel_map);
748 #else
749 tmp = vm_kmem_size;
750 #endif
751 vmem_init(kmem_arena, "kmem arena", kva_alloc(tmp), tmp, PAGE_SIZE,
752 0, 0);
753 vmem_set_reclaim(kmem_arena, kmem_reclaim);
754
755 #ifdef DEBUG_MEMGUARD
756 /*
757 * Initialize MemGuard if support compiled in. MemGuard is a
758 * replacement allocator used for detecting tamper-after-free
759 * scenarios as they occur. It is only used for debugging.
760 */
761 memguard_init(kmem_arena);
762 #endif
763 }
764
765 /*
766 * Initialize the kernel memory allocator
767 */
768 /* ARGSUSED*/
769 static void
mallocinit(void * dummy)770 mallocinit(void *dummy)
771 {
772 int i;
773 uint8_t indx;
774
775 mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
776
777 kmeminit();
778
779 uma_startup2();
780
781 if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX)
782 kmem_zmax = KMEM_ZMAX;
783
784 mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal),
785 #ifdef INVARIANTS
786 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
787 #else
788 NULL, NULL, NULL, NULL,
789 #endif
790 UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
791 for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
792 int size = kmemzones[indx].kz_size;
793 char *name = kmemzones[indx].kz_name;
794 int subzone;
795
796 for (subzone = 0; subzone < numzones; subzone++) {
797 kmemzones[indx].kz_zone[subzone] =
798 uma_zcreate(name, size,
799 #ifdef INVARIANTS
800 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
801 #else
802 NULL, NULL, NULL, NULL,
803 #endif
804 UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
805 }
806 for (;i <= size; i+= KMEM_ZBASE)
807 kmemsize[i >> KMEM_ZSHIFT] = indx;
808
809 }
810 }
811 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_FIRST, mallocinit, NULL);
812
813 void
malloc_init(void * data)814 malloc_init(void *data)
815 {
816 struct malloc_type_internal *mtip;
817 struct malloc_type *mtp;
818
819 KASSERT(cnt.v_page_count != 0, ("malloc_register before vm_init"));
820
821 mtp = data;
822 if (mtp->ks_magic != M_MAGIC)
823 panic("malloc_init: bad malloc type magic");
824
825 mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO);
826 mtp->ks_handle = mtip;
827 mtip->mti_zone = mtp_get_subzone(mtp->ks_shortdesc);
828
829 mtx_lock(&malloc_mtx);
830 mtp->ks_next = kmemstatistics;
831 kmemstatistics = mtp;
832 kmemcount++;
833 mtx_unlock(&malloc_mtx);
834 }
835
836 void
malloc_uninit(void * data)837 malloc_uninit(void *data)
838 {
839 struct malloc_type_internal *mtip;
840 struct malloc_type_stats *mtsp;
841 struct malloc_type *mtp, *temp;
842 uma_slab_t slab;
843 long temp_allocs, temp_bytes;
844 int i;
845
846 mtp = data;
847 KASSERT(mtp->ks_magic == M_MAGIC,
848 ("malloc_uninit: bad malloc type magic"));
849 KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL"));
850
851 mtx_lock(&malloc_mtx);
852 mtip = mtp->ks_handle;
853 mtp->ks_handle = NULL;
854 if (mtp != kmemstatistics) {
855 for (temp = kmemstatistics; temp != NULL;
856 temp = temp->ks_next) {
857 if (temp->ks_next == mtp) {
858 temp->ks_next = mtp->ks_next;
859 break;
860 }
861 }
862 KASSERT(temp,
863 ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc));
864 } else
865 kmemstatistics = mtp->ks_next;
866 kmemcount--;
867 mtx_unlock(&malloc_mtx);
868
869 /*
870 * Look for memory leaks.
871 */
872 temp_allocs = temp_bytes = 0;
873 for (i = 0; i < MAXCPU; i++) {
874 mtsp = &mtip->mti_stats[i];
875 temp_allocs += mtsp->mts_numallocs;
876 temp_allocs -= mtsp->mts_numfrees;
877 temp_bytes += mtsp->mts_memalloced;
878 temp_bytes -= mtsp->mts_memfreed;
879 }
880 if (temp_allocs > 0 || temp_bytes > 0) {
881 printf("Warning: memory type %s leaked memory on destroy "
882 "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc,
883 temp_allocs, temp_bytes);
884 }
885
886 slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK));
887 uma_zfree_arg(mt_zone, mtip, slab);
888 }
889
890 struct malloc_type *
malloc_desc2type(const char * desc)891 malloc_desc2type(const char *desc)
892 {
893 struct malloc_type *mtp;
894
895 mtx_assert(&malloc_mtx, MA_OWNED);
896 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
897 if (strcmp(mtp->ks_shortdesc, desc) == 0)
898 return (mtp);
899 }
900 return (NULL);
901 }
902
903 static int
sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)904 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
905 {
906 struct malloc_type_stream_header mtsh;
907 struct malloc_type_internal *mtip;
908 struct malloc_type_header mth;
909 struct malloc_type *mtp;
910 int error, i;
911 struct sbuf sbuf;
912
913 error = sysctl_wire_old_buffer(req, 0);
914 if (error != 0)
915 return (error);
916 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
917 mtx_lock(&malloc_mtx);
918
919 /*
920 * Insert stream header.
921 */
922 bzero(&mtsh, sizeof(mtsh));
923 mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
924 mtsh.mtsh_maxcpus = MAXCPU;
925 mtsh.mtsh_count = kmemcount;
926 (void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh));
927
928 /*
929 * Insert alternating sequence of type headers and type statistics.
930 */
931 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
932 mtip = (struct malloc_type_internal *)mtp->ks_handle;
933
934 /*
935 * Insert type header.
936 */
937 bzero(&mth, sizeof(mth));
938 strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
939 (void)sbuf_bcat(&sbuf, &mth, sizeof(mth));
940
941 /*
942 * Insert type statistics for each CPU.
943 */
944 for (i = 0; i < MAXCPU; i++) {
945 (void)sbuf_bcat(&sbuf, &mtip->mti_stats[i],
946 sizeof(mtip->mti_stats[i]));
947 }
948 }
949 mtx_unlock(&malloc_mtx);
950 error = sbuf_finish(&sbuf);
951 sbuf_delete(&sbuf);
952 return (error);
953 }
954
955 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT,
956 0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats",
957 "Return malloc types");
958
959 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
960 "Count of kernel malloc types");
961
962 void
malloc_type_list(malloc_type_list_func_t * func,void * arg)963 malloc_type_list(malloc_type_list_func_t *func, void *arg)
964 {
965 struct malloc_type *mtp, **bufmtp;
966 int count, i;
967 size_t buflen;
968
969 mtx_lock(&malloc_mtx);
970 restart:
971 mtx_assert(&malloc_mtx, MA_OWNED);
972 count = kmemcount;
973 mtx_unlock(&malloc_mtx);
974
975 buflen = sizeof(struct malloc_type *) * count;
976 bufmtp = malloc(buflen, M_TEMP, M_WAITOK);
977
978 mtx_lock(&malloc_mtx);
979
980 if (count < kmemcount) {
981 free(bufmtp, M_TEMP);
982 goto restart;
983 }
984
985 for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++)
986 bufmtp[i] = mtp;
987
988 mtx_unlock(&malloc_mtx);
989
990 for (i = 0; i < count; i++)
991 (func)(bufmtp[i], arg);
992
993 free(bufmtp, M_TEMP);
994 }
995
996 #ifdef DDB
DB_SHOW_COMMAND(malloc,db_show_malloc)997 DB_SHOW_COMMAND(malloc, db_show_malloc)
998 {
999 struct malloc_type_internal *mtip;
1000 struct malloc_type *mtp;
1001 uint64_t allocs, frees;
1002 uint64_t alloced, freed;
1003 int i;
1004
1005 db_printf("%18s %12s %12s %12s\n", "Type", "InUse", "MemUse",
1006 "Requests");
1007 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1008 mtip = (struct malloc_type_internal *)mtp->ks_handle;
1009 allocs = 0;
1010 frees = 0;
1011 alloced = 0;
1012 freed = 0;
1013 for (i = 0; i < MAXCPU; i++) {
1014 allocs += mtip->mti_stats[i].mts_numallocs;
1015 frees += mtip->mti_stats[i].mts_numfrees;
1016 alloced += mtip->mti_stats[i].mts_memalloced;
1017 freed += mtip->mti_stats[i].mts_memfreed;
1018 }
1019 db_printf("%18s %12ju %12juK %12ju\n",
1020 mtp->ks_shortdesc, allocs - frees,
1021 (alloced - freed + 1023) / 1024, allocs);
1022 if (db_pager_quit)
1023 break;
1024 }
1025 }
1026
1027 #if MALLOC_DEBUG_MAXZONES > 1
DB_SHOW_COMMAND(multizone_matches,db_show_multizone_matches)1028 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches)
1029 {
1030 struct malloc_type_internal *mtip;
1031 struct malloc_type *mtp;
1032 u_int subzone;
1033
1034 if (!have_addr) {
1035 db_printf("Usage: show multizone_matches <malloc type/addr>\n");
1036 return;
1037 }
1038 mtp = (void *)addr;
1039 if (mtp->ks_magic != M_MAGIC) {
1040 db_printf("Magic %lx does not match expected %x\n",
1041 mtp->ks_magic, M_MAGIC);
1042 return;
1043 }
1044
1045 mtip = mtp->ks_handle;
1046 subzone = mtip->mti_zone;
1047
1048 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1049 mtip = mtp->ks_handle;
1050 if (mtip->mti_zone != subzone)
1051 continue;
1052 db_printf("%s\n", mtp->ks_shortdesc);
1053 if (db_pager_quit)
1054 break;
1055 }
1056 }
1057 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
1058 #endif /* DDB */
1059
1060 #ifdef MALLOC_PROFILE
1061
1062 static int
sysctl_kern_mprof(SYSCTL_HANDLER_ARGS)1063 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS)
1064 {
1065 struct sbuf sbuf;
1066 uint64_t count;
1067 uint64_t waste;
1068 uint64_t mem;
1069 int error;
1070 int rsize;
1071 int size;
1072 int i;
1073
1074 waste = 0;
1075 mem = 0;
1076
1077 error = sysctl_wire_old_buffer(req, 0);
1078 if (error != 0)
1079 return (error);
1080 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
1081 sbuf_printf(&sbuf,
1082 "\n Size Requests Real Size\n");
1083 for (i = 0; i < KMEM_ZSIZE; i++) {
1084 size = i << KMEM_ZSHIFT;
1085 rsize = kmemzones[kmemsize[i]].kz_size;
1086 count = (long long unsigned)krequests[i];
1087
1088 sbuf_printf(&sbuf, "%6d%28llu%11d\n", size,
1089 (unsigned long long)count, rsize);
1090
1091 if ((rsize * count) > (size * count))
1092 waste += (rsize * count) - (size * count);
1093 mem += (rsize * count);
1094 }
1095 sbuf_printf(&sbuf,
1096 "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n",
1097 (unsigned long long)mem, (unsigned long long)waste);
1098 error = sbuf_finish(&sbuf);
1099 sbuf_delete(&sbuf);
1100 return (error);
1101 }
1102
1103 SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD,
1104 NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling");
1105 #endif /* MALLOC_PROFILE */
1106