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: stable/9/sys/kern/kern_malloc.c 262863 2014-03-06 18:50:35Z dumbbell $");
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/mbuf.h>
59 #include <sys/mutex.h>
60 #include <sys/vmmeter.h>
61 #include <sys/proc.h>
62 #include <sys/sbuf.h>
63 #include <sys/sysctl.h>
64 #include <sys/time.h>
65
66 #include <vm/vm.h>
67 #include <vm/pmap.h>
68 #include <vm/vm_param.h>
69 #include <vm/vm_kern.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vm_map.h>
72 #include <vm/vm_page.h>
73 #include <vm/uma.h>
74 #include <vm/uma_int.h>
75 #include <vm/uma_dbg.h>
76
77 #ifdef DEBUG_MEMGUARD
78 #include <vm/memguard.h>
79 #endif
80 #ifdef DEBUG_REDZONE
81 #include <vm/redzone.h>
82 #endif
83
84 #if defined(INVARIANTS) && defined(__i386__)
85 #include <machine/cpu.h>
86 #endif
87
88 #include <ddb/ddb.h>
89
90 #ifdef KDTRACE_HOOKS
91 #include <sys/dtrace_bsd.h>
92
93 dtrace_malloc_probe_func_t dtrace_malloc_probe;
94 #endif
95
96 /*
97 * When realloc() is called, if the new size is sufficiently smaller than
98 * the old size, realloc() will allocate a new, smaller block to avoid
99 * wasting memory. 'Sufficiently smaller' is defined as: newsize <=
100 * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'.
101 */
102 #ifndef REALLOC_FRACTION
103 #define REALLOC_FRACTION 1 /* new block if <= half the size */
104 #endif
105
106 /*
107 * Centrally define some common malloc types.
108 */
109 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches");
110 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory");
111 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers");
112
113 MALLOC_DEFINE(M_IP6OPT, "ip6opt", "IPv6 options");
114 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
115
116 static void kmeminit(void *);
117 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_FIRST, kmeminit, NULL);
118
119 static struct malloc_type *kmemstatistics;
120 static vm_offset_t kmembase;
121 static vm_offset_t kmemlimit;
122 static int kmemcount;
123
124 #define KMEM_ZSHIFT 4
125 #define KMEM_ZBASE 16
126 #define KMEM_ZMASK (KMEM_ZBASE - 1)
127
128 #define KMEM_ZMAX PAGE_SIZE
129 #define KMEM_ZSIZE (KMEM_ZMAX >> KMEM_ZSHIFT)
130 static uint8_t kmemsize[KMEM_ZSIZE + 1];
131
132 #ifndef MALLOC_DEBUG_MAXZONES
133 #define MALLOC_DEBUG_MAXZONES 1
134 #endif
135 static int numzones = MALLOC_DEBUG_MAXZONES;
136
137 /*
138 * Small malloc(9) memory allocations are allocated from a set of UMA buckets
139 * of various sizes.
140 *
141 * XXX: The comment here used to read "These won't be powers of two for
142 * long." It's possible that a significant amount of wasted memory could be
143 * recovered by tuning the sizes of these buckets.
144 */
145 struct {
146 int kz_size;
147 char *kz_name;
148 uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES];
149 } kmemzones[] = {
150 {16, "16", },
151 {32, "32", },
152 {64, "64", },
153 {128, "128", },
154 {256, "256", },
155 {512, "512", },
156 {1024, "1024", },
157 {2048, "2048", },
158 {4096, "4096", },
159 #if PAGE_SIZE > 4096
160 {8192, "8192", },
161 #if PAGE_SIZE > 8192
162 {16384, "16384", },
163 #if PAGE_SIZE > 16384
164 {32768, "32768", },
165 #if PAGE_SIZE > 32768
166 {65536, "65536", },
167 #if PAGE_SIZE > 65536
168 #error "Unsupported PAGE_SIZE"
169 #endif /* 65536 */
170 #endif /* 32768 */
171 #endif /* 16384 */
172 #endif /* 8192 */
173 #endif /* 4096 */
174 {0, NULL},
175 };
176
177 /*
178 * Zone to allocate malloc type descriptions from. For ABI reasons, memory
179 * types are described by a data structure passed by the declaring code, but
180 * the malloc(9) implementation has its own data structure describing the
181 * type and statistics. This permits the malloc(9)-internal data structures
182 * to be modified without breaking binary-compiled kernel modules that
183 * declare malloc types.
184 */
185 static uma_zone_t mt_zone;
186
187 u_long vm_kmem_size;
188 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0,
189 "Size of kernel memory");
190
191 static u_long vm_kmem_size_min;
192 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0,
193 "Minimum size of kernel memory");
194
195 static u_long vm_kmem_size_max;
196 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0,
197 "Maximum size of kernel memory");
198
199 static u_int vm_kmem_size_scale;
200 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0,
201 "Scale factor for kernel memory size");
202
203 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS);
204 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size,
205 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
206 sysctl_kmem_map_size, "LU", "Current kmem_map allocation size");
207
208 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS);
209 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free,
210 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0,
211 sysctl_kmem_map_free, "LU", "Largest contiguous free range in kmem_map");
212
213 /*
214 * The malloc_mtx protects the kmemstatistics linked list.
215 */
216 struct mtx malloc_mtx;
217
218 #ifdef MALLOC_PROFILE
219 uint64_t krequests[KMEM_ZSIZE + 1];
220
221 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS);
222 #endif
223
224 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS);
225
226 /*
227 * time_uptime of the last malloc(9) failure (induced or real).
228 */
229 static time_t t_malloc_fail;
230
231 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1)
232 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0,
233 "Kernel malloc debugging options");
234 #endif
235
236 /*
237 * malloc(9) fault injection -- cause malloc failures every (n) mallocs when
238 * the caller specifies M_NOWAIT. If set to 0, no failures are caused.
239 */
240 #ifdef MALLOC_MAKE_FAILURES
241 static int malloc_failure_rate;
242 static int malloc_nowait_count;
243 static int malloc_failure_count;
244 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RW,
245 &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail");
246 TUNABLE_INT("debug.malloc.failure_rate", &malloc_failure_rate);
247 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD,
248 &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures");
249 #endif
250
251 static int
sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS)252 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS)
253 {
254 u_long size;
255
256 size = kmem_map->size;
257 return (sysctl_handle_long(oidp, &size, 0, req));
258 }
259
260 static int
sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS)261 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS)
262 {
263 u_long size;
264
265 vm_map_lock_read(kmem_map);
266 size = kmem_map->root != NULL ? kmem_map->root->max_free :
267 kmem_map->max_offset - kmem_map->min_offset;
268 vm_map_unlock_read(kmem_map);
269 return (sysctl_handle_long(oidp, &size, 0, req));
270 }
271
272 /*
273 * malloc(9) uma zone separation -- sub-page buffer overruns in one
274 * malloc type will affect only a subset of other malloc types.
275 */
276 #if MALLOC_DEBUG_MAXZONES > 1
277 static void
tunable_set_numzones(void)278 tunable_set_numzones(void)
279 {
280
281 TUNABLE_INT_FETCH("debug.malloc.numzones",
282 &numzones);
283
284 /* Sanity check the number of malloc uma zones. */
285 if (numzones <= 0)
286 numzones = 1;
287 if (numzones > MALLOC_DEBUG_MAXZONES)
288 numzones = MALLOC_DEBUG_MAXZONES;
289 }
290 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL);
291 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN,
292 &numzones, 0, "Number of malloc uma subzones");
293
294 /*
295 * Any number that changes regularly is an okay choice for the
296 * offset. Build numbers are pretty good of you have them.
297 */
298 static u_int zone_offset = __FreeBSD_version;
299 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset);
300 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN,
301 &zone_offset, 0, "Separate malloc types by examining the "
302 "Nth character in the malloc type short description.");
303
304 static u_int
mtp_get_subzone(const char * desc)305 mtp_get_subzone(const char *desc)
306 {
307 size_t len;
308 u_int val;
309
310 if (desc == NULL || (len = strlen(desc)) == 0)
311 return (0);
312 val = desc[zone_offset % len];
313 return (val % numzones);
314 }
315 #elif MALLOC_DEBUG_MAXZONES == 0
316 #error "MALLOC_DEBUG_MAXZONES must be positive."
317 #else
318 static inline u_int
mtp_get_subzone(const char * desc)319 mtp_get_subzone(const char *desc)
320 {
321
322 return (0);
323 }
324 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
325
326 int
malloc_last_fail(void)327 malloc_last_fail(void)
328 {
329
330 return (time_uptime - t_malloc_fail);
331 }
332
333 /*
334 * An allocation has succeeded -- update malloc type statistics for the
335 * amount of bucket size. Occurs within a critical section so that the
336 * thread isn't preempted and doesn't migrate while updating per-PCU
337 * statistics.
338 */
339 static void
malloc_type_zone_allocated(struct malloc_type * mtp,unsigned long size,int zindx)340 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size,
341 int zindx)
342 {
343 struct malloc_type_internal *mtip;
344 struct malloc_type_stats *mtsp;
345
346 critical_enter();
347 mtip = mtp->ks_handle;
348 mtsp = &mtip->mti_stats[curcpu];
349 if (size > 0) {
350 mtsp->mts_memalloced += size;
351 mtsp->mts_numallocs++;
352 }
353 if (zindx != -1)
354 mtsp->mts_size |= 1 << zindx;
355
356 #ifdef KDTRACE_HOOKS
357 if (dtrace_malloc_probe != NULL) {
358 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC];
359 if (probe_id != 0)
360 (dtrace_malloc_probe)(probe_id,
361 (uintptr_t) mtp, (uintptr_t) mtip,
362 (uintptr_t) mtsp, size, zindx);
363 }
364 #endif
365
366 critical_exit();
367 }
368
369 void
malloc_type_allocated(struct malloc_type * mtp,unsigned long size)370 malloc_type_allocated(struct malloc_type *mtp, unsigned long size)
371 {
372
373 if (size > 0)
374 malloc_type_zone_allocated(mtp, size, -1);
375 }
376
377 /*
378 * A free operation has occurred -- update malloc type statistics for the
379 * amount of the bucket size. Occurs within a critical section so that the
380 * thread isn't preempted and doesn't migrate while updating per-CPU
381 * statistics.
382 */
383 void
malloc_type_freed(struct malloc_type * mtp,unsigned long size)384 malloc_type_freed(struct malloc_type *mtp, unsigned long size)
385 {
386 struct malloc_type_internal *mtip;
387 struct malloc_type_stats *mtsp;
388
389 critical_enter();
390 mtip = mtp->ks_handle;
391 mtsp = &mtip->mti_stats[curcpu];
392 mtsp->mts_memfreed += size;
393 mtsp->mts_numfrees++;
394
395 #ifdef KDTRACE_HOOKS
396 if (dtrace_malloc_probe != NULL) {
397 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE];
398 if (probe_id != 0)
399 (dtrace_malloc_probe)(probe_id,
400 (uintptr_t) mtp, (uintptr_t) mtip,
401 (uintptr_t) mtsp, size, 0);
402 }
403 #endif
404
405 critical_exit();
406 }
407
408 /*
409 * contigmalloc:
410 *
411 * Allocate a block of physically contiguous memory.
412 *
413 * If M_NOWAIT is set, this routine will not block and return NULL if
414 * the allocation fails.
415 */
416 void *
contigmalloc(unsigned long size,struct malloc_type * type,int flags,vm_paddr_t low,vm_paddr_t high,unsigned long alignment,unsigned long boundary)417 contigmalloc(unsigned long size, struct malloc_type *type, int flags,
418 vm_paddr_t low, vm_paddr_t high, unsigned long alignment,
419 unsigned long boundary)
420 {
421 void *ret;
422
423 ret = (void *)kmem_alloc_contig(kernel_map, size, flags, low, high,
424 alignment, boundary, VM_MEMATTR_DEFAULT);
425 if (ret != NULL)
426 malloc_type_allocated(type, round_page(size));
427 return (ret);
428 }
429
430 /*
431 * contigfree:
432 *
433 * Free a block of memory allocated by contigmalloc.
434 *
435 * This routine may not block.
436 */
437 void
contigfree(void * addr,unsigned long size,struct malloc_type * type)438 contigfree(void *addr, unsigned long size, struct malloc_type *type)
439 {
440
441 kmem_free(kernel_map, (vm_offset_t)addr, size);
442 malloc_type_freed(type, round_page(size));
443 }
444
445 /*
446 * malloc:
447 *
448 * Allocate a block of memory.
449 *
450 * If M_NOWAIT is set, this routine will not block and return NULL if
451 * the allocation fails.
452 */
453 void *
malloc(unsigned long size,struct malloc_type * mtp,int flags)454 malloc(unsigned long size, struct malloc_type *mtp, int flags)
455 {
456 int indx;
457 struct malloc_type_internal *mtip;
458 caddr_t va;
459 uma_zone_t zone;
460 #if defined(DIAGNOSTIC) || defined(DEBUG_REDZONE)
461 unsigned long osize = size;
462 #endif
463
464 #ifdef INVARIANTS
465 KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic"));
466 /*
467 * Check that exactly one of M_WAITOK or M_NOWAIT is specified.
468 */
469 indx = flags & (M_WAITOK | M_NOWAIT);
470 if (indx != M_NOWAIT && indx != M_WAITOK) {
471 static struct timeval lasterr;
472 static int curerr, once;
473 if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) {
474 printf("Bad malloc flags: %x\n", indx);
475 kdb_backtrace();
476 flags |= M_WAITOK;
477 once++;
478 }
479 }
480 #endif
481 #ifdef MALLOC_MAKE_FAILURES
482 if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
483 atomic_add_int(&malloc_nowait_count, 1);
484 if ((malloc_nowait_count % malloc_failure_rate) == 0) {
485 atomic_add_int(&malloc_failure_count, 1);
486 t_malloc_fail = time_uptime;
487 return (NULL);
488 }
489 }
490 #endif
491 if (flags & M_WAITOK)
492 KASSERT(curthread->td_intr_nesting_level == 0,
493 ("malloc(M_WAITOK) in interrupt context"));
494
495 #ifdef DEBUG_MEMGUARD
496 if (memguard_cmp(mtp, size)) {
497 va = memguard_alloc(size, flags);
498 if (va != NULL)
499 return (va);
500 /* This is unfortunate but should not be fatal. */
501 }
502 #endif
503
504 #ifdef DEBUG_REDZONE
505 size = redzone_size_ntor(size);
506 #endif
507
508 if (size <= KMEM_ZMAX) {
509 mtip = mtp->ks_handle;
510 if (size & KMEM_ZMASK)
511 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
512 indx = kmemsize[size >> KMEM_ZSHIFT];
513 KASSERT(mtip->mti_zone < numzones,
514 ("mti_zone %u out of range %d",
515 mtip->mti_zone, numzones));
516 zone = kmemzones[indx].kz_zone[mtip->mti_zone];
517 #ifdef MALLOC_PROFILE
518 krequests[size >> KMEM_ZSHIFT]++;
519 #endif
520 va = uma_zalloc(zone, flags);
521 if (va != NULL)
522 size = zone->uz_size;
523 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
524 } else {
525 size = roundup(size, PAGE_SIZE);
526 zone = NULL;
527 va = uma_large_malloc(size, flags);
528 malloc_type_allocated(mtp, va == NULL ? 0 : size);
529 }
530 if (flags & M_WAITOK)
531 KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL"));
532 else if (va == NULL)
533 t_malloc_fail = time_uptime;
534 #ifdef DIAGNOSTIC
535 if (va != NULL && !(flags & M_ZERO)) {
536 memset(va, 0x70, osize);
537 }
538 #endif
539 #ifdef DEBUG_REDZONE
540 if (va != NULL)
541 va = redzone_setup(va, osize);
542 #endif
543 return ((void *) va);
544 }
545
546 /*
547 * free:
548 *
549 * Free a block of memory allocated by malloc.
550 *
551 * This routine may not block.
552 */
553 void
free(void * addr,struct malloc_type * mtp)554 free(void *addr, struct malloc_type *mtp)
555 {
556 uma_slab_t slab;
557 u_long size;
558
559 KASSERT(mtp->ks_magic == M_MAGIC, ("free: bad malloc type magic"));
560
561 /* free(NULL, ...) does nothing */
562 if (addr == NULL)
563 return;
564
565 #ifdef DEBUG_MEMGUARD
566 if (is_memguard_addr(addr)) {
567 memguard_free(addr);
568 return;
569 }
570 #endif
571
572 #ifdef DEBUG_REDZONE
573 redzone_check(addr);
574 addr = redzone_addr_ntor(addr);
575 #endif
576
577 slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK));
578
579 if (slab == NULL)
580 panic("free: address %p(%p) has not been allocated.\n",
581 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK)));
582
583
584 if (!(slab->us_flags & UMA_SLAB_MALLOC)) {
585 #ifdef INVARIANTS
586 struct malloc_type **mtpp = addr;
587 #endif
588 size = slab->us_keg->uk_size;
589 #ifdef INVARIANTS
590 /*
591 * Cache a pointer to the malloc_type that most recently freed
592 * this memory here. This way we know who is most likely to
593 * have stepped on it later.
594 *
595 * This code assumes that size is a multiple of 8 bytes for
596 * 64 bit machines
597 */
598 mtpp = (struct malloc_type **)
599 ((unsigned long)mtpp & ~UMA_ALIGN_PTR);
600 mtpp += (size - sizeof(struct malloc_type *)) /
601 sizeof(struct malloc_type *);
602 *mtpp = mtp;
603 #endif
604 uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab);
605 } else {
606 size = slab->us_size;
607 uma_large_free(slab);
608 }
609 malloc_type_freed(mtp, size);
610 }
611
612 /*
613 * realloc: change the size of a memory block
614 */
615 void *
realloc(void * addr,unsigned long size,struct malloc_type * mtp,int flags)616 realloc(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
617 {
618 uma_slab_t slab;
619 unsigned long alloc;
620 void *newaddr;
621
622 KASSERT(mtp->ks_magic == M_MAGIC,
623 ("realloc: bad malloc type magic"));
624
625 /* realloc(NULL, ...) is equivalent to malloc(...) */
626 if (addr == NULL)
627 return (malloc(size, mtp, flags));
628
629 /*
630 * XXX: Should report free of old memory and alloc of new memory to
631 * per-CPU stats.
632 */
633
634 #ifdef DEBUG_MEMGUARD
635 if (is_memguard_addr(addr))
636 return (memguard_realloc(addr, size, mtp, flags));
637 #endif
638
639 #ifdef DEBUG_REDZONE
640 slab = NULL;
641 alloc = redzone_get_size(addr);
642 #else
643 slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK));
644
645 /* Sanity check */
646 KASSERT(slab != NULL,
647 ("realloc: address %p out of range", (void *)addr));
648
649 /* Get the size of the original block */
650 if (!(slab->us_flags & UMA_SLAB_MALLOC))
651 alloc = slab->us_keg->uk_size;
652 else
653 alloc = slab->us_size;
654
655 /* Reuse the original block if appropriate */
656 if (size <= alloc
657 && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE))
658 return (addr);
659 #endif /* !DEBUG_REDZONE */
660
661 /* Allocate a new, bigger (or smaller) block */
662 if ((newaddr = malloc(size, mtp, flags)) == NULL)
663 return (NULL);
664
665 /* Copy over original contents */
666 bcopy(addr, newaddr, min(size, alloc));
667 free(addr, mtp);
668 return (newaddr);
669 }
670
671 /*
672 * reallocf: same as realloc() but free memory on failure.
673 */
674 void *
reallocf(void * addr,unsigned long size,struct malloc_type * mtp,int flags)675 reallocf(void *addr, unsigned long size, struct malloc_type *mtp, int flags)
676 {
677 void *mem;
678
679 if ((mem = realloc(addr, size, mtp, flags)) == NULL)
680 free(addr, mtp);
681 return (mem);
682 }
683
684 /*
685 * Initialize the kernel memory allocator
686 */
687 /* ARGSUSED*/
688 static void
kmeminit(void * dummy)689 kmeminit(void *dummy)
690 {
691 uint8_t indx;
692 u_long mem_size, tmp;
693 int i;
694
695 mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF);
696
697 /*
698 * Try to auto-tune the kernel memory size, so that it is
699 * more applicable for a wider range of machine sizes. The
700 * VM_KMEM_SIZE_MAX is dependent on the maximum KVA space
701 * available.
702 *
703 * Note that the kmem_map is also used by the zone allocator,
704 * so make sure that there is enough space.
705 */
706 vm_kmem_size = VM_KMEM_SIZE + nmbclusters * PAGE_SIZE;
707 mem_size = cnt.v_page_count;
708
709 #if defined(VM_KMEM_SIZE_SCALE)
710 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE;
711 #endif
712 TUNABLE_INT_FETCH("vm.kmem_size_scale", &vm_kmem_size_scale);
713 if (vm_kmem_size_scale > 0 &&
714 (mem_size / vm_kmem_size_scale) > (vm_kmem_size / PAGE_SIZE))
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
725 #if defined(VM_KMEM_SIZE_MAX)
726 vm_kmem_size_max = VM_KMEM_SIZE_MAX;
727 #endif
728 TUNABLE_ULONG_FETCH("vm.kmem_size_max", &vm_kmem_size_max);
729 if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max)
730 vm_kmem_size = vm_kmem_size_max;
731
732 /* Allow final override from the kernel environment */
733 TUNABLE_ULONG_FETCH("vm.kmem_size", &vm_kmem_size);
734
735 /*
736 * Limit kmem virtual size to twice the physical memory.
737 * This allows for kmem map sparseness, but limits the size
738 * to something sane. Be careful to not overflow the 32bit
739 * ints while doing the check or the adjustment.
740 */
741 if (vm_kmem_size / 2 / PAGE_SIZE > mem_size)
742 vm_kmem_size = 2 * mem_size * PAGE_SIZE;
743
744 #ifdef DEBUG_MEMGUARD
745 tmp = memguard_fudge(vm_kmem_size, kernel_map);
746 #else
747 tmp = vm_kmem_size;
748 #endif
749 kmem_map = kmem_suballoc(kernel_map, &kmembase, &kmemlimit,
750 tmp, TRUE);
751 kmem_map->system_map = 1;
752
753 #ifdef DEBUG_MEMGUARD
754 /*
755 * Initialize MemGuard if support compiled in. MemGuard is a
756 * replacement allocator used for detecting tamper-after-free
757 * scenarios as they occur. It is only used for debugging.
758 */
759 memguard_init(kmem_map);
760 #endif
761
762 uma_startup2();
763
764 mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal),
765 #ifdef INVARIANTS
766 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
767 #else
768 NULL, NULL, NULL, NULL,
769 #endif
770 UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
771 for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) {
772 int size = kmemzones[indx].kz_size;
773 char *name = kmemzones[indx].kz_name;
774 int subzone;
775
776 for (subzone = 0; subzone < numzones; subzone++) {
777 kmemzones[indx].kz_zone[subzone] =
778 uma_zcreate(name, size,
779 #ifdef INVARIANTS
780 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini,
781 #else
782 NULL, NULL, NULL, NULL,
783 #endif
784 UMA_ALIGN_PTR, UMA_ZONE_MALLOC);
785 }
786 for (;i <= size; i+= KMEM_ZBASE)
787 kmemsize[i >> KMEM_ZSHIFT] = indx;
788
789 }
790 }
791
792 void
malloc_init(void * data)793 malloc_init(void *data)
794 {
795 struct malloc_type_internal *mtip;
796 struct malloc_type *mtp;
797
798 KASSERT(cnt.v_page_count != 0, ("malloc_register before vm_init"));
799
800 mtp = data;
801 if (mtp->ks_magic != M_MAGIC)
802 panic("malloc_init: bad malloc type magic");
803
804 mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO);
805 mtp->ks_handle = mtip;
806 mtip->mti_zone = mtp_get_subzone(mtp->ks_shortdesc);
807
808 mtx_lock(&malloc_mtx);
809 mtp->ks_next = kmemstatistics;
810 kmemstatistics = mtp;
811 kmemcount++;
812 mtx_unlock(&malloc_mtx);
813 }
814
815 void
malloc_uninit(void * data)816 malloc_uninit(void *data)
817 {
818 struct malloc_type_internal *mtip;
819 struct malloc_type_stats *mtsp;
820 struct malloc_type *mtp, *temp;
821 uma_slab_t slab;
822 long temp_allocs, temp_bytes;
823 int i;
824
825 mtp = data;
826 KASSERT(mtp->ks_magic == M_MAGIC,
827 ("malloc_uninit: bad malloc type magic"));
828 KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL"));
829
830 mtx_lock(&malloc_mtx);
831 mtip = mtp->ks_handle;
832 mtp->ks_handle = NULL;
833 if (mtp != kmemstatistics) {
834 for (temp = kmemstatistics; temp != NULL;
835 temp = temp->ks_next) {
836 if (temp->ks_next == mtp) {
837 temp->ks_next = mtp->ks_next;
838 break;
839 }
840 }
841 KASSERT(temp,
842 ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc));
843 } else
844 kmemstatistics = mtp->ks_next;
845 kmemcount--;
846 mtx_unlock(&malloc_mtx);
847
848 /*
849 * Look for memory leaks.
850 */
851 temp_allocs = temp_bytes = 0;
852 for (i = 0; i < MAXCPU; i++) {
853 mtsp = &mtip->mti_stats[i];
854 temp_allocs += mtsp->mts_numallocs;
855 temp_allocs -= mtsp->mts_numfrees;
856 temp_bytes += mtsp->mts_memalloced;
857 temp_bytes -= mtsp->mts_memfreed;
858 }
859 if (temp_allocs > 0 || temp_bytes > 0) {
860 printf("Warning: memory type %s leaked memory on destroy "
861 "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc,
862 temp_allocs, temp_bytes);
863 }
864
865 slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK));
866 uma_zfree_arg(mt_zone, mtip, slab);
867 }
868
869 struct malloc_type *
malloc_desc2type(const char * desc)870 malloc_desc2type(const char *desc)
871 {
872 struct malloc_type *mtp;
873
874 mtx_assert(&malloc_mtx, MA_OWNED);
875 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
876 if (strcmp(mtp->ks_shortdesc, desc) == 0)
877 return (mtp);
878 }
879 return (NULL);
880 }
881
882 static int
sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)883 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS)
884 {
885 struct malloc_type_stream_header mtsh;
886 struct malloc_type_internal *mtip;
887 struct malloc_type_header mth;
888 struct malloc_type *mtp;
889 int error, i;
890 struct sbuf sbuf;
891
892 error = sysctl_wire_old_buffer(req, 0);
893 if (error != 0)
894 return (error);
895 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
896 mtx_lock(&malloc_mtx);
897
898 /*
899 * Insert stream header.
900 */
901 bzero(&mtsh, sizeof(mtsh));
902 mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION;
903 mtsh.mtsh_maxcpus = MAXCPU;
904 mtsh.mtsh_count = kmemcount;
905 (void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh));
906
907 /*
908 * Insert alternating sequence of type headers and type statistics.
909 */
910 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
911 mtip = (struct malloc_type_internal *)mtp->ks_handle;
912
913 /*
914 * Insert type header.
915 */
916 bzero(&mth, sizeof(mth));
917 strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME);
918 (void)sbuf_bcat(&sbuf, &mth, sizeof(mth));
919
920 /*
921 * Insert type statistics for each CPU.
922 */
923 for (i = 0; i < MAXCPU; i++) {
924 (void)sbuf_bcat(&sbuf, &mtip->mti_stats[i],
925 sizeof(mtip->mti_stats[i]));
926 }
927 }
928 mtx_unlock(&malloc_mtx);
929 error = sbuf_finish(&sbuf);
930 sbuf_delete(&sbuf);
931 return (error);
932 }
933
934 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT,
935 0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats",
936 "Return malloc types");
937
938 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0,
939 "Count of kernel malloc types");
940
941 void
malloc_type_list(malloc_type_list_func_t * func,void * arg)942 malloc_type_list(malloc_type_list_func_t *func, void *arg)
943 {
944 struct malloc_type *mtp, **bufmtp;
945 int count, i;
946 size_t buflen;
947
948 mtx_lock(&malloc_mtx);
949 restart:
950 mtx_assert(&malloc_mtx, MA_OWNED);
951 count = kmemcount;
952 mtx_unlock(&malloc_mtx);
953
954 buflen = sizeof(struct malloc_type *) * count;
955 bufmtp = malloc(buflen, M_TEMP, M_WAITOK);
956
957 mtx_lock(&malloc_mtx);
958
959 if (count < kmemcount) {
960 free(bufmtp, M_TEMP);
961 goto restart;
962 }
963
964 for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++)
965 bufmtp[i] = mtp;
966
967 mtx_unlock(&malloc_mtx);
968
969 for (i = 0; i < count; i++)
970 (func)(bufmtp[i], arg);
971
972 free(bufmtp, M_TEMP);
973 }
974
975 #ifdef DDB
DB_SHOW_COMMAND(malloc,db_show_malloc)976 DB_SHOW_COMMAND(malloc, db_show_malloc)
977 {
978 struct malloc_type_internal *mtip;
979 struct malloc_type *mtp;
980 uint64_t allocs, frees;
981 uint64_t alloced, freed;
982 int i;
983
984 db_printf("%18s %12s %12s %12s\n", "Type", "InUse", "MemUse",
985 "Requests");
986 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
987 mtip = (struct malloc_type_internal *)mtp->ks_handle;
988 allocs = 0;
989 frees = 0;
990 alloced = 0;
991 freed = 0;
992 for (i = 0; i < MAXCPU; i++) {
993 allocs += mtip->mti_stats[i].mts_numallocs;
994 frees += mtip->mti_stats[i].mts_numfrees;
995 alloced += mtip->mti_stats[i].mts_memalloced;
996 freed += mtip->mti_stats[i].mts_memfreed;
997 }
998 db_printf("%18s %12ju %12juK %12ju\n",
999 mtp->ks_shortdesc, allocs - frees,
1000 (alloced - freed + 1023) / 1024, allocs);
1001 if (db_pager_quit)
1002 break;
1003 }
1004 }
1005
1006 #if MALLOC_DEBUG_MAXZONES > 1
DB_SHOW_COMMAND(multizone_matches,db_show_multizone_matches)1007 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches)
1008 {
1009 struct malloc_type_internal *mtip;
1010 struct malloc_type *mtp;
1011 u_int subzone;
1012
1013 if (!have_addr) {
1014 db_printf("Usage: show multizone_matches <malloc type/addr>\n");
1015 return;
1016 }
1017 mtp = (void *)addr;
1018 if (mtp->ks_magic != M_MAGIC) {
1019 db_printf("Magic %lx does not match expected %x\n",
1020 mtp->ks_magic, M_MAGIC);
1021 return;
1022 }
1023
1024 mtip = mtp->ks_handle;
1025 subzone = mtip->mti_zone;
1026
1027 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) {
1028 mtip = mtp->ks_handle;
1029 if (mtip->mti_zone != subzone)
1030 continue;
1031 db_printf("%s\n", mtp->ks_shortdesc);
1032 if (db_pager_quit)
1033 break;
1034 }
1035 }
1036 #endif /* MALLOC_DEBUG_MAXZONES > 1 */
1037 #endif /* DDB */
1038
1039 #ifdef MALLOC_PROFILE
1040
1041 static int
sysctl_kern_mprof(SYSCTL_HANDLER_ARGS)1042 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS)
1043 {
1044 struct sbuf sbuf;
1045 uint64_t count;
1046 uint64_t waste;
1047 uint64_t mem;
1048 int error;
1049 int rsize;
1050 int size;
1051 int i;
1052
1053 waste = 0;
1054 mem = 0;
1055
1056 error = sysctl_wire_old_buffer(req, 0);
1057 if (error != 0)
1058 return (error);
1059 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
1060 sbuf_printf(&sbuf,
1061 "\n Size Requests Real Size\n");
1062 for (i = 0; i < KMEM_ZSIZE; i++) {
1063 size = i << KMEM_ZSHIFT;
1064 rsize = kmemzones[kmemsize[i]].kz_size;
1065 count = (long long unsigned)krequests[i];
1066
1067 sbuf_printf(&sbuf, "%6d%28llu%11d\n", size,
1068 (unsigned long long)count, rsize);
1069
1070 if ((rsize * count) > (size * count))
1071 waste += (rsize * count) - (size * count);
1072 mem += (rsize * count);
1073 }
1074 sbuf_printf(&sbuf,
1075 "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n",
1076 (unsigned long long)mem, (unsigned long long)waste);
1077 error = sbuf_finish(&sbuf);
1078 sbuf_delete(&sbuf);
1079 return (error);
1080 }
1081
1082 SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD,
1083 NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling");
1084 #endif /* MALLOC_PROFILE */
1085