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 * Copyright (c) 2008 Otto Moerbeek <otto@drijf.net> (mallocarray) 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 4. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * @(#)kern_malloc.c 8.3 (Berkeley) 1/4/94 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__FBSDID("$FreeBSD: stable/10/sys/kern/kern_malloc.c 328276 2018-01-23 04:37:31Z kp $"); 48 49#include "opt_ddb.h" 50#include "opt_kdtrace.h" 51#include "opt_vm.h" 52 53#include <sys/param.h> 54#include <sys/systm.h> 55#include <sys/kdb.h> 56#include <sys/kernel.h> 57#include <sys/lock.h> 58#include <sys/malloc.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#include <sys/vmem.h> 66 67#include <vm/vm.h> 68#include <vm/pmap.h> 69#include <vm/vm_pageout.h> 70#include <vm/vm_param.h> 71#include <vm/vm_kern.h> 72#include <vm/vm_extern.h> 73#include <vm/vm_map.h> 74#include <vm/vm_page.h> 75#include <vm/uma.h> 76#include <vm/uma_int.h> 77#include <vm/uma_dbg.h> 78 79#ifdef DEBUG_MEMGUARD 80#include <vm/memguard.h> 81#endif 82#ifdef DEBUG_REDZONE 83#include <vm/redzone.h> 84#endif 85 86#if defined(INVARIANTS) && defined(__i386__) 87#include <machine/cpu.h> 88#endif 89 90#include <ddb/ddb.h> 91 92#ifdef KDTRACE_HOOKS 93#include <sys/dtrace_bsd.h> 94 95dtrace_malloc_probe_func_t dtrace_malloc_probe; 96#endif 97 98/* 99 * When realloc() is called, if the new size is sufficiently smaller than 100 * the old size, realloc() will allocate a new, smaller block to avoid 101 * wasting memory. 'Sufficiently smaller' is defined as: newsize <= 102 * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'. 103 */ 104#ifndef REALLOC_FRACTION 105#define REALLOC_FRACTION 1 /* new block if <= half the size */ 106#endif 107 108/* 109 * Centrally define some common malloc types. 110 */ 111MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches"); 112MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory"); 113MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers"); 114 115MALLOC_DEFINE(M_IP6OPT, "ip6opt", "IPv6 options"); 116MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery"); 117 118static struct malloc_type *kmemstatistics; 119static int kmemcount; 120 121#define KMEM_ZSHIFT 4 122#define KMEM_ZBASE 16 123#define KMEM_ZMASK (KMEM_ZBASE - 1) 124 125#define KMEM_ZMAX 65536 126#define KMEM_ZSIZE (KMEM_ZMAX >> KMEM_ZSHIFT) 127static uint8_t kmemsize[KMEM_ZSIZE + 1]; 128 129#ifndef MALLOC_DEBUG_MAXZONES 130#define MALLOC_DEBUG_MAXZONES 1 131#endif 132static int numzones = MALLOC_DEBUG_MAXZONES; 133 134/* 135 * Small malloc(9) memory allocations are allocated from a set of UMA buckets 136 * of various sizes. 137 * 138 * XXX: The comment here used to read "These won't be powers of two for 139 * long." It's possible that a significant amount of wasted memory could be 140 * recovered by tuning the sizes of these buckets. 141 */ 142struct { 143 int kz_size; 144 char *kz_name; 145 uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES]; 146} kmemzones[] = { 147 {16, "16", }, 148 {32, "32", }, 149 {64, "64", }, 150 {128, "128", }, 151 {256, "256", }, 152 {512, "512", }, 153 {1024, "1024", }, 154 {2048, "2048", }, 155 {4096, "4096", }, 156 {8192, "8192", }, 157 {16384, "16384", }, 158 {32768, "32768", }, 159 {65536, "65536", }, 160 {0, NULL}, 161}; 162 163/* 164 * Zone to allocate malloc type descriptions from. For ABI reasons, memory 165 * types are described by a data structure passed by the declaring code, but 166 * the malloc(9) implementation has its own data structure describing the 167 * type and statistics. This permits the malloc(9)-internal data structures 168 * to be modified without breaking binary-compiled kernel modules that 169 * declare malloc types. 170 */ 171static uma_zone_t mt_zone; 172 173u_long vm_kmem_size; 174SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0, 175 "Size of kernel memory"); 176 177static u_long kmem_zmax = KMEM_ZMAX; 178SYSCTL_ULONG(_vm, OID_AUTO, kmem_zmax, CTLFLAG_RDTUN, &kmem_zmax, 0, 179 "Maximum allocation size that malloc(9) would use UMA as backend"); 180 181static u_long vm_kmem_size_min; 182SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0, 183 "Minimum size of kernel memory"); 184 185static u_long vm_kmem_size_max; 186SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0, 187 "Maximum size of kernel memory"); 188 189static u_int vm_kmem_size_scale; 190SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0, 191 "Scale factor for kernel memory size"); 192 193static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS); 194SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size, 195 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, 196 sysctl_kmem_map_size, "LU", "Current kmem allocation size"); 197 198static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS); 199SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free, 200 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, 201 sysctl_kmem_map_free, "LU", "Free space in kmem"); 202 203/* 204 * The malloc_mtx protects the kmemstatistics linked list. 205 */ 206struct mtx malloc_mtx; 207 208#ifdef MALLOC_PROFILE 209uint64_t krequests[KMEM_ZSIZE + 1]; 210 211static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS); 212#endif 213 214static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS); 215 216/* 217 * time_uptime of the last malloc(9) failure (induced or real). 218 */ 219static time_t t_malloc_fail; 220 221#if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1) 222static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0, 223 "Kernel malloc debugging options"); 224#endif 225 226/* 227 * malloc(9) fault injection -- cause malloc failures every (n) mallocs when 228 * the caller specifies M_NOWAIT. If set to 0, no failures are caused. 229 */ 230#ifdef MALLOC_MAKE_FAILURES 231static int malloc_failure_rate; 232static int malloc_nowait_count; 233static int malloc_failure_count; 234SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RW, 235 &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail"); 236TUNABLE_INT("debug.malloc.failure_rate", &malloc_failure_rate); 237SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD, 238 &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures"); 239#endif 240 241static int 242sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS) 243{ 244 u_long size; 245 246 size = vmem_size(kmem_arena, VMEM_ALLOC); 247 return (sysctl_handle_long(oidp, &size, 0, req)); 248} 249 250static int 251sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS) 252{ 253 u_long size; 254 255 size = vmem_size(kmem_arena, VMEM_FREE); 256 return (sysctl_handle_long(oidp, &size, 0, req)); 257} 258 259/* 260 * malloc(9) uma zone separation -- sub-page buffer overruns in one 261 * malloc type will affect only a subset of other malloc types. 262 */ 263#if MALLOC_DEBUG_MAXZONES > 1 264static void 265tunable_set_numzones(void) 266{ 267 268 TUNABLE_INT_FETCH("debug.malloc.numzones", 269 &numzones); 270 271 /* Sanity check the number of malloc uma zones. */ 272 if (numzones <= 0) 273 numzones = 1; 274 if (numzones > MALLOC_DEBUG_MAXZONES) 275 numzones = MALLOC_DEBUG_MAXZONES; 276} 277SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL); 278SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN, 279 &numzones, 0, "Number of malloc uma subzones"); 280 281/* 282 * Any number that changes regularly is an okay choice for the 283 * offset. Build numbers are pretty good of you have them. 284 */ 285static u_int zone_offset = __FreeBSD_version; 286TUNABLE_INT("debug.malloc.zone_offset", &zone_offset); 287SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN, 288 &zone_offset, 0, "Separate malloc types by examining the " 289 "Nth character in the malloc type short description."); 290 291static u_int 292mtp_get_subzone(const char *desc) 293{ 294 size_t len; 295 u_int val; 296 297 if (desc == NULL || (len = strlen(desc)) == 0) 298 return (0); 299 val = desc[zone_offset % len]; 300 return (val % numzones); 301} 302#elif MALLOC_DEBUG_MAXZONES == 0 303#error "MALLOC_DEBUG_MAXZONES must be positive." 304#else 305static inline u_int 306mtp_get_subzone(const char *desc) 307{ 308 309 return (0); 310} 311#endif /* MALLOC_DEBUG_MAXZONES > 1 */ 312 313int 314malloc_last_fail(void) 315{ 316 317 return (time_uptime - t_malloc_fail); 318} 319 320/* 321 * An allocation has succeeded -- update malloc type statistics for the 322 * amount of bucket size. Occurs within a critical section so that the 323 * thread isn't preempted and doesn't migrate while updating per-PCU 324 * statistics. 325 */ 326static void 327malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size, 328 int zindx) 329{ 330 struct malloc_type_internal *mtip; 331 struct malloc_type_stats *mtsp; 332 333 critical_enter(); 334 mtip = mtp->ks_handle; 335 mtsp = &mtip->mti_stats[curcpu]; 336 if (size > 0) { 337 mtsp->mts_memalloced += size; 338 mtsp->mts_numallocs++; 339 } 340 if (zindx != -1) 341 mtsp->mts_size |= 1 << zindx; 342 343#ifdef KDTRACE_HOOKS 344 if (dtrace_malloc_probe != NULL) { 345 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC]; 346 if (probe_id != 0) 347 (dtrace_malloc_probe)(probe_id, 348 (uintptr_t) mtp, (uintptr_t) mtip, 349 (uintptr_t) mtsp, size, zindx); 350 } 351#endif 352 353 critical_exit(); 354} 355 356void 357malloc_type_allocated(struct malloc_type *mtp, unsigned long size) 358{ 359 360 if (size > 0) 361 malloc_type_zone_allocated(mtp, size, -1); 362} 363 364/* 365 * A free operation has occurred -- update malloc type statistics for the 366 * amount of the bucket size. Occurs within a critical section so that the 367 * thread isn't preempted and doesn't migrate while updating per-CPU 368 * statistics. 369 */ 370void 371malloc_type_freed(struct malloc_type *mtp, unsigned long size) 372{ 373 struct malloc_type_internal *mtip; 374 struct malloc_type_stats *mtsp; 375 376 critical_enter(); 377 mtip = mtp->ks_handle; 378 mtsp = &mtip->mti_stats[curcpu]; 379 mtsp->mts_memfreed += size; 380 mtsp->mts_numfrees++; 381 382#ifdef KDTRACE_HOOKS 383 if (dtrace_malloc_probe != NULL) { 384 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE]; 385 if (probe_id != 0) 386 (dtrace_malloc_probe)(probe_id, 387 (uintptr_t) mtp, (uintptr_t) mtip, 388 (uintptr_t) mtsp, size, 0); 389 } 390#endif 391 392 critical_exit(); 393} 394 395/* 396 * contigmalloc: 397 * 398 * Allocate a block of physically contiguous memory. 399 * 400 * If M_NOWAIT is set, this routine will not block and return NULL if 401 * the allocation fails. 402 */ 403void * 404contigmalloc(unsigned long size, struct malloc_type *type, int flags, 405 vm_paddr_t low, vm_paddr_t high, unsigned long alignment, 406 vm_paddr_t boundary) 407{ 408 void *ret; 409 410 ret = (void *)kmem_alloc_contig(kernel_arena, size, flags, low, high, 411 alignment, boundary, VM_MEMATTR_DEFAULT); 412 if (ret != NULL) 413 malloc_type_allocated(type, round_page(size)); 414 return (ret); 415} 416 417/* 418 * contigfree: 419 * 420 * Free a block of memory allocated by contigmalloc. 421 * 422 * This routine may not block. 423 */ 424void 425contigfree(void *addr, unsigned long size, struct malloc_type *type) 426{ 427 428 kmem_free(kernel_arena, (vm_offset_t)addr, size); 429 malloc_type_freed(type, round_page(size)); 430} 431 432/* 433 * malloc: 434 * 435 * Allocate a block of memory. 436 * 437 * If M_NOWAIT is set, this routine will not block and return NULL if 438 * the allocation fails. 439 */ 440void * 441malloc(unsigned long size, struct malloc_type *mtp, int flags) 442{ 443 int indx; 444 struct malloc_type_internal *mtip; 445 caddr_t va; 446 uma_zone_t zone; 447#if defined(DIAGNOSTIC) || defined(DEBUG_REDZONE) 448 unsigned long osize = size; 449#endif 450 451#ifdef INVARIANTS 452 KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic")); 453 /* 454 * Check that exactly one of M_WAITOK or M_NOWAIT is specified. 455 */ 456 indx = flags & (M_WAITOK | M_NOWAIT); 457 if (indx != M_NOWAIT && indx != M_WAITOK) { 458 static struct timeval lasterr; 459 static int curerr, once; 460 if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) { 461 printf("Bad malloc flags: %x\n", indx); 462 kdb_backtrace(); 463 flags |= M_WAITOK; 464 once++; 465 } 466 } 467#endif 468#ifdef MALLOC_MAKE_FAILURES 469 if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) { 470 atomic_add_int(&malloc_nowait_count, 1); 471 if ((malloc_nowait_count % malloc_failure_rate) == 0) { 472 atomic_add_int(&malloc_failure_count, 1); 473 t_malloc_fail = time_uptime; 474 return (NULL); 475 } 476 } 477#endif 478 if (flags & M_WAITOK) 479 KASSERT(curthread->td_intr_nesting_level == 0, 480 ("malloc(M_WAITOK) in interrupt context")); 481 482#ifdef DEBUG_MEMGUARD 483 if (memguard_cmp_mtp(mtp, size)) { 484 va = memguard_alloc(size, flags); 485 if (va != NULL) 486 return (va); 487 /* This is unfortunate but should not be fatal. */ 488 } 489#endif 490 491#ifdef DEBUG_REDZONE 492 size = redzone_size_ntor(size); 493#endif 494 495 if (size <= kmem_zmax) { 496 mtip = mtp->ks_handle; 497 if (size & KMEM_ZMASK) 498 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE; 499 indx = kmemsize[size >> KMEM_ZSHIFT]; 500 KASSERT(mtip->mti_zone < numzones, 501 ("mti_zone %u out of range %d", 502 mtip->mti_zone, numzones)); 503 zone = kmemzones[indx].kz_zone[mtip->mti_zone]; 504#ifdef MALLOC_PROFILE 505 krequests[size >> KMEM_ZSHIFT]++; 506#endif 507 va = uma_zalloc(zone, flags); 508 if (va != NULL) 509 size = zone->uz_size; 510 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx); 511 } else { 512 size = roundup(size, PAGE_SIZE); 513 zone = NULL; 514 va = uma_large_malloc(size, flags); 515 malloc_type_allocated(mtp, va == NULL ? 0 : size); 516 } 517 if (flags & M_WAITOK) 518 KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL")); 519 else if (va == NULL) 520 t_malloc_fail = time_uptime; 521#ifdef DIAGNOSTIC 522 if (va != NULL && !(flags & M_ZERO)) { 523 memset(va, 0x70, osize); 524 } 525#endif 526#ifdef DEBUG_REDZONE 527 if (va != NULL) 528 va = redzone_setup(va, osize); 529#endif 530 return ((void *) va); 531} 532 533void * 534mallocarray(size_t nmemb, size_t size, struct malloc_type *type, int flags) 535{ 536 537 if (WOULD_OVERFLOW(nmemb, size)) 538 panic("mallocarray: %zu * %zu overflowed", nmemb, size); 539 540 return (malloc(size * nmemb, type, flags)); 541} 542 543/* 544 * free: 545 * 546 * Free a block of memory allocated by malloc. 547 * 548 * This routine may not block. 549 */ 550void 551free(void *addr, struct malloc_type *mtp) 552{ 553 uma_slab_t slab; 554 u_long size; 555 556 KASSERT(mtp->ks_magic == M_MAGIC, ("free: bad malloc type magic")); 557 558 /* free(NULL, ...) does nothing */ 559 if (addr == NULL) 560 return; 561 562#ifdef DEBUG_MEMGUARD 563 if (is_memguard_addr(addr)) { 564 memguard_free(addr); 565 return; 566 } 567#endif 568 569#ifdef DEBUG_REDZONE 570 redzone_check(addr); 571 addr = redzone_addr_ntor(addr); 572#endif 573 574 slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK)); 575 576 if (slab == NULL) 577 panic("free: address %p(%p) has not been allocated.\n", 578 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK))); 579 580 if (!(slab->us_flags & UMA_SLAB_MALLOC)) { 581#ifdef INVARIANTS 582 struct malloc_type **mtpp = addr; 583#endif 584 size = slab->us_keg->uk_size; 585#ifdef INVARIANTS 586 /* 587 * Cache a pointer to the malloc_type that most recently freed 588 * this memory here. This way we know who is most likely to 589 * have stepped on it later. 590 * 591 * This code assumes that size is a multiple of 8 bytes for 592 * 64 bit machines 593 */ 594 mtpp = (struct malloc_type **) 595 ((unsigned long)mtpp & ~UMA_ALIGN_PTR); 596 mtpp += (size - sizeof(struct malloc_type *)) / 597 sizeof(struct malloc_type *); 598 *mtpp = mtp; 599#endif 600 uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab); 601 } else { 602 size = slab->us_size; 603 uma_large_free(slab); 604 } 605 malloc_type_freed(mtp, size); 606} 607 608/* 609 * realloc: change the size of a memory block 610 */ 611void * 612realloc(void *addr, unsigned long size, struct malloc_type *mtp, int flags) 613{ 614 uma_slab_t slab; 615 unsigned long alloc; 616 void *newaddr; 617 618 KASSERT(mtp->ks_magic == M_MAGIC, 619 ("realloc: bad malloc type magic")); 620 621 /* realloc(NULL, ...) is equivalent to malloc(...) */ 622 if (addr == NULL) 623 return (malloc(size, mtp, flags)); 624 625 /* 626 * XXX: Should report free of old memory and alloc of new memory to 627 * per-CPU stats. 628 */ 629 630#ifdef DEBUG_MEMGUARD 631 if (is_memguard_addr(addr)) 632 return (memguard_realloc(addr, size, mtp, flags)); 633#endif 634 635#ifdef DEBUG_REDZONE 636 slab = NULL; 637 alloc = redzone_get_size(addr); 638#else 639 slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK)); 640 641 /* Sanity check */ 642 KASSERT(slab != NULL, 643 ("realloc: address %p out of range", (void *)addr)); 644 645 /* Get the size of the original block */ 646 if (!(slab->us_flags & UMA_SLAB_MALLOC)) 647 alloc = slab->us_keg->uk_size; 648 else 649 alloc = slab->us_size; 650 651 /* Reuse the original block if appropriate */ 652 if (size <= alloc 653 && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE)) 654 return (addr); 655#endif /* !DEBUG_REDZONE */ 656 657 /* Allocate a new, bigger (or smaller) block */ 658 if ((newaddr = malloc(size, mtp, flags)) == NULL) 659 return (NULL); 660 661 /* Copy over original contents */ 662 bcopy(addr, newaddr, min(size, alloc)); 663 free(addr, mtp); 664 return (newaddr); 665} 666 667/* 668 * reallocf: same as realloc() but free memory on failure. 669 */ 670void * 671reallocf(void *addr, unsigned long size, struct malloc_type *mtp, int flags) 672{ 673 void *mem; 674 675 if ((mem = realloc(addr, size, mtp, flags)) == NULL) 676 free(addr, mtp); 677 return (mem); 678} 679 680/* 681 * Wake the uma reclamation pagedaemon thread when we exhaust KVA. It 682 * will call the lowmem handler and uma_reclaim() callbacks in a 683 * context that is safe. 684 */ 685static void 686kmem_reclaim(vmem_t *vm, int flags) 687{ 688 689 uma_reclaim_wakeup(); 690 pagedaemon_wakeup(); 691} 692 693CTASSERT(VM_KMEM_SIZE_SCALE >= 1); 694 695/* 696 * Initialize the kernel memory (kmem) arena. 697 */ 698void 699kmeminit(void) 700{ 701 u_long mem_size, tmp; 702 703 /* 704 * Calculate the amount of kernel virtual address (KVA) space that is 705 * preallocated to the kmem arena. In order to support a wide range 706 * of machines, it is a function of the physical memory size, 707 * specifically, 708 * 709 * min(max(physical memory size / VM_KMEM_SIZE_SCALE, 710 * VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX) 711 * 712 * Every architecture must define an integral value for 713 * VM_KMEM_SIZE_SCALE. However, the definitions of VM_KMEM_SIZE_MIN 714 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and 715 * ceiling on this preallocation, are optional. Typically, 716 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on 717 * a given architecture. 718 */ 719 mem_size = cnt.v_page_count; 720 721 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE; 722 TUNABLE_INT_FETCH("vm.kmem_size_scale", &vm_kmem_size_scale); 723 if (vm_kmem_size_scale < 1) 724 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE; 725 726 vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE; 727 728#if defined(VM_KMEM_SIZE_MIN) 729 vm_kmem_size_min = VM_KMEM_SIZE_MIN; 730#endif 731 TUNABLE_ULONG_FETCH("vm.kmem_size_min", &vm_kmem_size_min); 732 if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min) 733 vm_kmem_size = vm_kmem_size_min; 734 735#if defined(VM_KMEM_SIZE_MAX) 736 vm_kmem_size_max = VM_KMEM_SIZE_MAX; 737#endif 738 TUNABLE_ULONG_FETCH("vm.kmem_size_max", &vm_kmem_size_max); 739 if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max) 740 vm_kmem_size = vm_kmem_size_max; 741 742 /* 743 * Alternatively, the amount of KVA space that is preallocated to the 744 * kmem arena can be set statically at compile-time or manually 745 * through the kernel environment. However, it is still limited to 746 * twice the physical memory size, which has been sufficient to handle 747 * the most severe cases of external fragmentation in the kmem arena. 748 */ 749#if defined(VM_KMEM_SIZE) 750 vm_kmem_size = VM_KMEM_SIZE; 751#endif 752 TUNABLE_ULONG_FETCH("vm.kmem_size", &vm_kmem_size); 753 if (vm_kmem_size / 2 / PAGE_SIZE > mem_size) 754 vm_kmem_size = 2 * mem_size * PAGE_SIZE; 755 756 vm_kmem_size = round_page(vm_kmem_size); 757#ifdef DEBUG_MEMGUARD 758 tmp = memguard_fudge(vm_kmem_size, kernel_map); 759#else 760 tmp = vm_kmem_size; 761#endif 762 vmem_init(kmem_arena, "kmem arena", kva_alloc(tmp), tmp, PAGE_SIZE, 763 0, 0); 764 vmem_set_reclaim(kmem_arena, kmem_reclaim); 765 766#ifdef DEBUG_MEMGUARD 767 /* 768 * Initialize MemGuard if support compiled in. MemGuard is a 769 * replacement allocator used for detecting tamper-after-free 770 * scenarios as they occur. It is only used for debugging. 771 */ 772 memguard_init(kmem_arena); 773#endif 774} 775 776/* 777 * Initialize the kernel memory allocator 778 */ 779/* ARGSUSED*/ 780static void 781mallocinit(void *dummy) 782{ 783 int i; 784 uint8_t indx; 785 786 mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF); 787 788 kmeminit(); 789 790 uma_startup2(); 791 792 if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX) 793 kmem_zmax = KMEM_ZMAX; 794 795 mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal), 796#ifdef INVARIANTS 797 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 798#else 799 NULL, NULL, NULL, NULL, 800#endif 801 UMA_ALIGN_PTR, UMA_ZONE_MALLOC); 802 for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) { 803 int size = kmemzones[indx].kz_size; 804 char *name = kmemzones[indx].kz_name; 805 int subzone; 806 807 for (subzone = 0; subzone < numzones; subzone++) { 808 kmemzones[indx].kz_zone[subzone] = 809 uma_zcreate(name, size, 810#ifdef INVARIANTS 811 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 812#else 813 NULL, NULL, NULL, NULL, 814#endif 815 UMA_ALIGN_PTR, UMA_ZONE_MALLOC); 816 } 817 for (;i <= size; i+= KMEM_ZBASE) 818 kmemsize[i >> KMEM_ZSHIFT] = indx; 819 820 } 821} 822SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_FIRST, mallocinit, NULL); 823 824void 825malloc_init(void *data) 826{ 827 struct malloc_type_internal *mtip; 828 struct malloc_type *mtp; 829 830 KASSERT(cnt.v_page_count != 0, ("malloc_register before vm_init")); 831 832 mtp = data; 833 if (mtp->ks_magic != M_MAGIC) 834 panic("malloc_init: bad malloc type magic"); 835 836 mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO); 837 mtp->ks_handle = mtip; 838 mtip->mti_zone = mtp_get_subzone(mtp->ks_shortdesc); 839 840 mtx_lock(&malloc_mtx); 841 mtp->ks_next = kmemstatistics; 842 kmemstatistics = mtp; 843 kmemcount++; 844 mtx_unlock(&malloc_mtx); 845} 846 847void 848malloc_uninit(void *data) 849{ 850 struct malloc_type_internal *mtip; 851 struct malloc_type_stats *mtsp; 852 struct malloc_type *mtp, *temp; 853 uma_slab_t slab; 854 long temp_allocs, temp_bytes; 855 int i; 856 857 mtp = data; 858 KASSERT(mtp->ks_magic == M_MAGIC, 859 ("malloc_uninit: bad malloc type magic")); 860 KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL")); 861 862 mtx_lock(&malloc_mtx); 863 mtip = mtp->ks_handle; 864 mtp->ks_handle = NULL; 865 if (mtp != kmemstatistics) { 866 for (temp = kmemstatistics; temp != NULL; 867 temp = temp->ks_next) { 868 if (temp->ks_next == mtp) { 869 temp->ks_next = mtp->ks_next; 870 break; 871 } 872 } 873 KASSERT(temp, 874 ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc)); 875 } else 876 kmemstatistics = mtp->ks_next; 877 kmemcount--; 878 mtx_unlock(&malloc_mtx); 879 880 /* 881 * Look for memory leaks. 882 */ 883 temp_allocs = temp_bytes = 0; 884 for (i = 0; i < MAXCPU; i++) { 885 mtsp = &mtip->mti_stats[i]; 886 temp_allocs += mtsp->mts_numallocs; 887 temp_allocs -= mtsp->mts_numfrees; 888 temp_bytes += mtsp->mts_memalloced; 889 temp_bytes -= mtsp->mts_memfreed; 890 } 891 if (temp_allocs > 0 || temp_bytes > 0) { 892 printf("Warning: memory type %s leaked memory on destroy " 893 "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc, 894 temp_allocs, temp_bytes); 895 } 896 897 slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK)); 898 uma_zfree_arg(mt_zone, mtip, slab); 899} 900 901struct malloc_type * 902malloc_desc2type(const char *desc) 903{ 904 struct malloc_type *mtp; 905 906 mtx_assert(&malloc_mtx, MA_OWNED); 907 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 908 if (strcmp(mtp->ks_shortdesc, desc) == 0) 909 return (mtp); 910 } 911 return (NULL); 912} 913 914static int 915sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS) 916{ 917 struct malloc_type_stream_header mtsh; 918 struct malloc_type_internal *mtip; 919 struct malloc_type_header mth; 920 struct malloc_type *mtp; 921 int error, i; 922 struct sbuf sbuf; 923 924 error = sysctl_wire_old_buffer(req, 0); 925 if (error != 0) 926 return (error); 927 sbuf_new_for_sysctl(&sbuf, NULL, 128, req); 928 mtx_lock(&malloc_mtx); 929 930 /* 931 * Insert stream header. 932 */ 933 bzero(&mtsh, sizeof(mtsh)); 934 mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION; 935 mtsh.mtsh_maxcpus = MAXCPU; 936 mtsh.mtsh_count = kmemcount; 937 (void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh)); 938 939 /* 940 * Insert alternating sequence of type headers and type statistics. 941 */ 942 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 943 mtip = (struct malloc_type_internal *)mtp->ks_handle; 944 945 /* 946 * Insert type header. 947 */ 948 bzero(&mth, sizeof(mth)); 949 strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME); 950 (void)sbuf_bcat(&sbuf, &mth, sizeof(mth)); 951 952 /* 953 * Insert type statistics for each CPU. 954 */ 955 for (i = 0; i < MAXCPU; i++) { 956 (void)sbuf_bcat(&sbuf, &mtip->mti_stats[i], 957 sizeof(mtip->mti_stats[i])); 958 } 959 } 960 mtx_unlock(&malloc_mtx); 961 error = sbuf_finish(&sbuf); 962 sbuf_delete(&sbuf); 963 return (error); 964} 965 966SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT, 967 0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats", 968 "Return malloc types"); 969 970SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0, 971 "Count of kernel malloc types"); 972 973void 974malloc_type_list(malloc_type_list_func_t *func, void *arg) 975{ 976 struct malloc_type *mtp, **bufmtp; 977 int count, i; 978 size_t buflen; 979 980 mtx_lock(&malloc_mtx); 981restart: 982 mtx_assert(&malloc_mtx, MA_OWNED); 983 count = kmemcount; 984 mtx_unlock(&malloc_mtx); 985 986 buflen = sizeof(struct malloc_type *) * count; 987 bufmtp = malloc(buflen, M_TEMP, M_WAITOK); 988 989 mtx_lock(&malloc_mtx); 990 991 if (count < kmemcount) { 992 free(bufmtp, M_TEMP); 993 goto restart; 994 } 995 996 for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++) 997 bufmtp[i] = mtp; 998 999 mtx_unlock(&malloc_mtx); 1000 1001 for (i = 0; i < count; i++) 1002 (func)(bufmtp[i], arg); 1003 1004 free(bufmtp, M_TEMP); 1005} 1006 1007#ifdef DDB 1008DB_SHOW_COMMAND(malloc, db_show_malloc) 1009{ 1010 struct malloc_type_internal *mtip; 1011 struct malloc_type *mtp; 1012 uint64_t allocs, frees; 1013 uint64_t alloced, freed; 1014 int i; 1015 1016 db_printf("%18s %12s %12s %12s\n", "Type", "InUse", "MemUse", 1017 "Requests"); 1018 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1019 mtip = (struct malloc_type_internal *)mtp->ks_handle; 1020 allocs = 0; 1021 frees = 0; 1022 alloced = 0; 1023 freed = 0; 1024 for (i = 0; i < MAXCPU; i++) { 1025 allocs += mtip->mti_stats[i].mts_numallocs; 1026 frees += mtip->mti_stats[i].mts_numfrees; 1027 alloced += mtip->mti_stats[i].mts_memalloced; 1028 freed += mtip->mti_stats[i].mts_memfreed; 1029 } 1030 db_printf("%18s %12ju %12juK %12ju\n", 1031 mtp->ks_shortdesc, allocs - frees, 1032 (alloced - freed + 1023) / 1024, allocs); 1033 if (db_pager_quit) 1034 break; 1035 } 1036} 1037 1038#if MALLOC_DEBUG_MAXZONES > 1 1039DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches) 1040{ 1041 struct malloc_type_internal *mtip; 1042 struct malloc_type *mtp; 1043 u_int subzone; 1044 1045 if (!have_addr) { 1046 db_printf("Usage: show multizone_matches <malloc type/addr>\n"); 1047 return; 1048 } 1049 mtp = (void *)addr; 1050 if (mtp->ks_magic != M_MAGIC) { 1051 db_printf("Magic %lx does not match expected %x\n", 1052 mtp->ks_magic, M_MAGIC); 1053 return; 1054 } 1055 1056 mtip = mtp->ks_handle; 1057 subzone = mtip->mti_zone; 1058 1059 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1060 mtip = mtp->ks_handle; 1061 if (mtip->mti_zone != subzone) 1062 continue; 1063 db_printf("%s\n", mtp->ks_shortdesc); 1064 if (db_pager_quit) 1065 break; 1066 } 1067} 1068#endif /* MALLOC_DEBUG_MAXZONES > 1 */ 1069#endif /* DDB */ 1070 1071#ifdef MALLOC_PROFILE 1072 1073static int 1074sysctl_kern_mprof(SYSCTL_HANDLER_ARGS) 1075{ 1076 struct sbuf sbuf; 1077 uint64_t count; 1078 uint64_t waste; 1079 uint64_t mem; 1080 int error; 1081 int rsize; 1082 int size; 1083 int i; 1084 1085 waste = 0; 1086 mem = 0; 1087 1088 error = sysctl_wire_old_buffer(req, 0); 1089 if (error != 0) 1090 return (error); 1091 sbuf_new_for_sysctl(&sbuf, NULL, 128, req); 1092 sbuf_printf(&sbuf, 1093 "\n Size Requests Real Size\n"); 1094 for (i = 0; i < KMEM_ZSIZE; i++) { 1095 size = i << KMEM_ZSHIFT; 1096 rsize = kmemzones[kmemsize[i]].kz_size; 1097 count = (long long unsigned)krequests[i]; 1098 1099 sbuf_printf(&sbuf, "%6d%28llu%11d\n", size, 1100 (unsigned long long)count, rsize); 1101 1102 if ((rsize * count) > (size * count)) 1103 waste += (rsize * count) - (size * count); 1104 mem += (rsize * count); 1105 } 1106 sbuf_printf(&sbuf, 1107 "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n", 1108 (unsigned long long)mem, (unsigned long long)waste); 1109 error = sbuf_finish(&sbuf); 1110 sbuf_delete(&sbuf); 1111 return (error); 1112} 1113 1114SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD, 1115 NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling"); 1116#endif /* MALLOC_PROFILE */ 1117