1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1987, 1991, 1993 5 * The Regents of the University of California. 6 * Copyright (c) 2005-2009 Robert N. M. Watson 7 * Copyright (c) 2008 Otto Moerbeek <otto@drijf.net> (mallocarray) 8 * All rights reserved. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)kern_malloc.c 8.3 (Berkeley) 1/4/94 35 */ 36 37 /* 38 * Kernel malloc(9) implementation -- general purpose kernel memory allocator 39 * based on memory types. Back end is implemented using the UMA(9) zone 40 * allocator. A set of fixed-size buckets are used for smaller allocations, 41 * and a special UMA allocation interface is used for larger allocations. 42 * Callers declare memory types, and statistics are maintained independently 43 * for each memory type. Statistics are maintained per-CPU for performance 44 * reasons. See malloc(9) and comments in malloc.h for a detailed 45 * description. 46 */ 47 48 #include <sys/cdefs.h> 49 __FBSDID("$FreeBSD$"); 50 51 #include "opt_ddb.h" 52 #include "opt_vm.h" 53 54 #include <sys/param.h> 55 #include <sys/systm.h> 56 #include <sys/kdb.h> 57 #include <sys/kernel.h> 58 #include <sys/lock.h> 59 #include <sys/malloc.h> 60 #include <sys/mutex.h> 61 #include <sys/vmmeter.h> 62 #include <sys/proc.h> 63 #include <sys/sbuf.h> 64 #include <sys/sysctl.h> 65 #include <sys/time.h> 66 #include <sys/vmem.h> 67 68 #include <vm/vm.h> 69 #include <vm/pmap.h> 70 #include <vm/vm_pageout.h> 71 #include <vm/vm_param.h> 72 #include <vm/vm_kern.h> 73 #include <vm/vm_extern.h> 74 #include <vm/vm_map.h> 75 #include <vm/vm_page.h> 76 #include <vm/uma.h> 77 #include <vm/uma_int.h> 78 #include <vm/uma_dbg.h> 79 80 #ifdef DEBUG_MEMGUARD 81 #include <vm/memguard.h> 82 #endif 83 #ifdef DEBUG_REDZONE 84 #include <vm/redzone.h> 85 #endif 86 87 #if defined(INVARIANTS) && defined(__i386__) 88 #include <machine/cpu.h> 89 #endif 90 91 #include <ddb/ddb.h> 92 93 #ifdef KDTRACE_HOOKS 94 #include <sys/dtrace_bsd.h> 95 96 dtrace_malloc_probe_func_t dtrace_malloc_probe; 97 #endif 98 99 #if defined(INVARIANTS) || defined(MALLOC_MAKE_FAILURES) || \ 100 defined(DEBUG_MEMGUARD) || defined(DEBUG_REDZONE) 101 #define MALLOC_DEBUG 1 102 #endif 103 104 /* 105 * When realloc() is called, if the new size is sufficiently smaller than 106 * the old size, realloc() will allocate a new, smaller block to avoid 107 * wasting memory. 'Sufficiently smaller' is defined as: newsize <= 108 * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'. 109 */ 110 #ifndef REALLOC_FRACTION 111 #define REALLOC_FRACTION 1 /* new block if <= half the size */ 112 #endif 113 114 /* 115 * Centrally define some common malloc types. 116 */ 117 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches"); 118 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory"); 119 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers"); 120 121 static struct malloc_type *kmemstatistics; 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 65536 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 {8192, "8192", }, 160 {16384, "16384", }, 161 {32768, "32768", }, 162 {65536, "65536", }, 163 {0, NULL}, 164 }; 165 166 /* 167 * Zone to allocate malloc type descriptions from. For ABI reasons, memory 168 * types are described by a data structure passed by the declaring code, but 169 * the malloc(9) implementation has its own data structure describing the 170 * type and statistics. This permits the malloc(9)-internal data structures 171 * to be modified without breaking binary-compiled kernel modules that 172 * declare malloc types. 173 */ 174 static uma_zone_t mt_zone; 175 176 u_long vm_kmem_size; 177 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0, 178 "Size of kernel memory"); 179 180 static u_long kmem_zmax = KMEM_ZMAX; 181 SYSCTL_ULONG(_vm, OID_AUTO, kmem_zmax, CTLFLAG_RDTUN, &kmem_zmax, 0, 182 "Maximum allocation size that malloc(9) would use UMA as backend"); 183 184 static u_long vm_kmem_size_min; 185 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0, 186 "Minimum size of kernel memory"); 187 188 static u_long vm_kmem_size_max; 189 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0, 190 "Maximum size of kernel memory"); 191 192 static u_int vm_kmem_size_scale; 193 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0, 194 "Scale factor for kernel memory size"); 195 196 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS); 197 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size, 198 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, 199 sysctl_kmem_map_size, "LU", "Current kmem allocation size"); 200 201 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS); 202 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free, 203 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, 204 sysctl_kmem_map_free, "LU", "Free space in kmem"); 205 206 /* 207 * The malloc_mtx protects the kmemstatistics linked list. 208 */ 209 struct mtx malloc_mtx; 210 211 #ifdef MALLOC_PROFILE 212 uint64_t krequests[KMEM_ZSIZE + 1]; 213 214 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS); 215 #endif 216 217 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS); 218 219 /* 220 * time_uptime of the last malloc(9) failure (induced or real). 221 */ 222 static time_t t_malloc_fail; 223 224 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1) 225 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0, 226 "Kernel malloc debugging options"); 227 #endif 228 229 /* 230 * malloc(9) fault injection -- cause malloc failures every (n) mallocs when 231 * the caller specifies M_NOWAIT. If set to 0, no failures are caused. 232 */ 233 #ifdef MALLOC_MAKE_FAILURES 234 static int malloc_failure_rate; 235 static int malloc_nowait_count; 236 static int malloc_failure_count; 237 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RWTUN, 238 &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail"); 239 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD, 240 &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures"); 241 #endif 242 243 static int 244 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS) 245 { 246 u_long size; 247 248 size = uma_size(); 249 return (sysctl_handle_long(oidp, &size, 0, req)); 250 } 251 252 static int 253 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS) 254 { 255 u_long size, limit; 256 257 /* The sysctl is unsigned, implement as a saturation value. */ 258 size = uma_size(); 259 limit = uma_limit(); 260 if (size > limit) 261 size = 0; 262 else 263 size = limit - size; 264 return (sysctl_handle_long(oidp, &size, 0, req)); 265 } 266 267 /* 268 * malloc(9) uma zone separation -- sub-page buffer overruns in one 269 * malloc type will affect only a subset of other malloc types. 270 */ 271 #if MALLOC_DEBUG_MAXZONES > 1 272 static void 273 tunable_set_numzones(void) 274 { 275 276 TUNABLE_INT_FETCH("debug.malloc.numzones", 277 &numzones); 278 279 /* Sanity check the number of malloc uma zones. */ 280 if (numzones <= 0) 281 numzones = 1; 282 if (numzones > MALLOC_DEBUG_MAXZONES) 283 numzones = MALLOC_DEBUG_MAXZONES; 284 } 285 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL); 286 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 287 &numzones, 0, "Number of malloc uma subzones"); 288 289 /* 290 * Any number that changes regularly is an okay choice for the 291 * offset. Build numbers are pretty good of you have them. 292 */ 293 static u_int zone_offset = __FreeBSD_version; 294 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset); 295 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN, 296 &zone_offset, 0, "Separate malloc types by examining the " 297 "Nth character in the malloc type short description."); 298 299 static u_int 300 mtp_get_subzone(const char *desc) 301 { 302 size_t len; 303 u_int val; 304 305 if (desc == NULL || (len = strlen(desc)) == 0) 306 return (0); 307 val = desc[zone_offset % len]; 308 return (val % numzones); 309 } 310 #elif MALLOC_DEBUG_MAXZONES == 0 311 #error "MALLOC_DEBUG_MAXZONES must be positive." 312 #else 313 static inline u_int 314 mtp_get_subzone(const char *desc) 315 { 316 317 return (0); 318 } 319 #endif /* MALLOC_DEBUG_MAXZONES > 1 */ 320 321 int 322 malloc_last_fail(void) 323 { 324 325 return (time_uptime - t_malloc_fail); 326 } 327 328 /* 329 * An allocation has succeeded -- update malloc type statistics for the 330 * amount of bucket size. Occurs within a critical section so that the 331 * thread isn't preempted and doesn't migrate while updating per-PCU 332 * statistics. 333 */ 334 static void 335 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size, 336 int zindx) 337 { 338 struct malloc_type_internal *mtip; 339 struct malloc_type_stats *mtsp; 340 341 critical_enter(); 342 mtip = mtp->ks_handle; 343 mtsp = &mtip->mti_stats[curcpu]; 344 if (size > 0) { 345 mtsp->mts_memalloced += size; 346 mtsp->mts_numallocs++; 347 } 348 if (zindx != -1) 349 mtsp->mts_size |= 1 << zindx; 350 351 #ifdef KDTRACE_HOOKS 352 if (dtrace_malloc_probe != NULL) { 353 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC]; 354 if (probe_id != 0) 355 (dtrace_malloc_probe)(probe_id, 356 (uintptr_t) mtp, (uintptr_t) mtip, 357 (uintptr_t) mtsp, size, zindx); 358 } 359 #endif 360 361 critical_exit(); 362 } 363 364 void 365 malloc_type_allocated(struct malloc_type *mtp, unsigned long size) 366 { 367 368 if (size > 0) 369 malloc_type_zone_allocated(mtp, size, -1); 370 } 371 372 /* 373 * A free operation has occurred -- update malloc type statistics for the 374 * amount of the bucket size. Occurs within a critical section so that the 375 * thread isn't preempted and doesn't migrate while updating per-CPU 376 * statistics. 377 */ 378 void 379 malloc_type_freed(struct malloc_type *mtp, unsigned long size) 380 { 381 struct malloc_type_internal *mtip; 382 struct malloc_type_stats *mtsp; 383 384 critical_enter(); 385 mtip = mtp->ks_handle; 386 mtsp = &mtip->mti_stats[curcpu]; 387 mtsp->mts_memfreed += size; 388 mtsp->mts_numfrees++; 389 390 #ifdef KDTRACE_HOOKS 391 if (dtrace_malloc_probe != NULL) { 392 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE]; 393 if (probe_id != 0) 394 (dtrace_malloc_probe)(probe_id, 395 (uintptr_t) mtp, (uintptr_t) mtip, 396 (uintptr_t) mtsp, size, 0); 397 } 398 #endif 399 400 critical_exit(); 401 } 402 403 /* 404 * contigmalloc: 405 * 406 * Allocate a block of physically contiguous memory. 407 * 408 * If M_NOWAIT is set, this routine will not block and return NULL if 409 * the allocation fails. 410 */ 411 void * 412 contigmalloc(unsigned long size, struct malloc_type *type, int flags, 413 vm_paddr_t low, vm_paddr_t high, unsigned long alignment, 414 vm_paddr_t boundary) 415 { 416 void *ret; 417 418 ret = (void *)kmem_alloc_contig(kernel_arena, size, flags, low, high, 419 alignment, boundary, VM_MEMATTR_DEFAULT); 420 if (ret != NULL) 421 malloc_type_allocated(type, round_page(size)); 422 return (ret); 423 } 424 425 void * 426 contigmalloc_domain(unsigned long size, struct malloc_type *type, 427 int domain, int flags, vm_paddr_t low, vm_paddr_t high, 428 unsigned long alignment, vm_paddr_t boundary) 429 { 430 void *ret; 431 432 ret = (void *)kmem_alloc_contig_domain(domain, size, flags, low, high, 433 alignment, boundary, VM_MEMATTR_DEFAULT); 434 if (ret != NULL) 435 malloc_type_allocated(type, round_page(size)); 436 return (ret); 437 } 438 439 /* 440 * contigfree: 441 * 442 * Free a block of memory allocated by contigmalloc. 443 * 444 * This routine may not block. 445 */ 446 void 447 contigfree(void *addr, unsigned long size, struct malloc_type *type) 448 { 449 450 kmem_free(kernel_arena, (vm_offset_t)addr, size); 451 malloc_type_freed(type, round_page(size)); 452 } 453 454 #ifdef MALLOC_DEBUG 455 static int 456 malloc_dbg(caddr_t *vap, size_t *sizep, struct malloc_type *mtp, 457 int flags) 458 { 459 #ifdef INVARIANTS 460 int indx; 461 462 KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic")); 463 /* 464 * Check that exactly one of M_WAITOK or M_NOWAIT is specified. 465 */ 466 indx = flags & (M_WAITOK | M_NOWAIT); 467 if (indx != M_NOWAIT && indx != M_WAITOK) { 468 static struct timeval lasterr; 469 static int curerr, once; 470 if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) { 471 printf("Bad malloc flags: %x\n", indx); 472 kdb_backtrace(); 473 flags |= M_WAITOK; 474 once++; 475 } 476 } 477 #endif 478 #ifdef MALLOC_MAKE_FAILURES 479 if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) { 480 atomic_add_int(&malloc_nowait_count, 1); 481 if ((malloc_nowait_count % malloc_failure_rate) == 0) { 482 atomic_add_int(&malloc_failure_count, 1); 483 t_malloc_fail = time_uptime; 484 *vap = NULL; 485 return (EJUSTRETURN); 486 } 487 } 488 #endif 489 if (flags & M_WAITOK) 490 KASSERT(curthread->td_intr_nesting_level == 0, 491 ("malloc(M_WAITOK) in interrupt context")); 492 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), 493 ("malloc: called with spinlock or critical section held")); 494 495 #ifdef DEBUG_MEMGUARD 496 if (memguard_cmp_mtp(mtp, *sizep)) { 497 *vap = memguard_alloc(*sizep, flags); 498 if (*vap != NULL) 499 return (EJUSTRETURN); 500 /* This is unfortunate but should not be fatal. */ 501 } 502 #endif 503 504 #ifdef DEBUG_REDZONE 505 *sizep = redzone_size_ntor(*sizep); 506 #endif 507 508 return (0); 509 } 510 #endif 511 512 /* 513 * malloc: 514 * 515 * Allocate a block of memory. 516 * 517 * If M_NOWAIT is set, this routine will not block and return NULL if 518 * the allocation fails. 519 */ 520 void * 521 malloc(size_t size, struct malloc_type *mtp, int flags) 522 { 523 int indx; 524 struct malloc_type_internal *mtip; 525 caddr_t va; 526 uma_zone_t zone; 527 #if defined(DEBUG_REDZONE) 528 unsigned long osize = size; 529 #endif 530 531 #ifdef MALLOC_DEBUG 532 if (malloc_dbg(&va, &size, mtp, flags) != 0) 533 return (va); 534 #endif 535 536 if (size <= kmem_zmax) { 537 mtip = mtp->ks_handle; 538 if (size & KMEM_ZMASK) 539 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE; 540 indx = kmemsize[size >> KMEM_ZSHIFT]; 541 KASSERT(mtip->mti_zone < numzones, 542 ("mti_zone %u out of range %d", 543 mtip->mti_zone, numzones)); 544 zone = kmemzones[indx].kz_zone[mtip->mti_zone]; 545 #ifdef MALLOC_PROFILE 546 krequests[size >> KMEM_ZSHIFT]++; 547 #endif 548 va = uma_zalloc(zone, flags); 549 if (va != NULL) 550 size = zone->uz_size; 551 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx); 552 } else { 553 size = roundup(size, PAGE_SIZE); 554 zone = NULL; 555 va = uma_large_malloc(size, flags); 556 malloc_type_allocated(mtp, va == NULL ? 0 : size); 557 } 558 if (flags & M_WAITOK) 559 KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL")); 560 else if (va == NULL) 561 t_malloc_fail = time_uptime; 562 #ifdef DEBUG_REDZONE 563 if (va != NULL) 564 va = redzone_setup(va, osize); 565 #endif 566 return ((void *) va); 567 } 568 569 void * 570 malloc_domain(size_t size, struct malloc_type *mtp, int domain, 571 int flags) 572 { 573 int indx; 574 struct malloc_type_internal *mtip; 575 caddr_t va; 576 uma_zone_t zone; 577 #if defined(DEBUG_REDZONE) 578 unsigned long osize = size; 579 #endif 580 581 #ifdef MALLOC_DEBUG 582 if (malloc_dbg(&va, &size, mtp, flags) != 0) 583 return (va); 584 #endif 585 if (size <= kmem_zmax) { 586 mtip = mtp->ks_handle; 587 if (size & KMEM_ZMASK) 588 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE; 589 indx = kmemsize[size >> KMEM_ZSHIFT]; 590 KASSERT(mtip->mti_zone < numzones, 591 ("mti_zone %u out of range %d", 592 mtip->mti_zone, numzones)); 593 zone = kmemzones[indx].kz_zone[mtip->mti_zone]; 594 #ifdef MALLOC_PROFILE 595 krequests[size >> KMEM_ZSHIFT]++; 596 #endif 597 va = uma_zalloc_domain(zone, NULL, domain, flags); 598 if (va != NULL) 599 size = zone->uz_size; 600 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx); 601 } else { 602 size = roundup(size, PAGE_SIZE); 603 zone = NULL; 604 va = uma_large_malloc_domain(size, domain, flags); 605 malloc_type_allocated(mtp, va == NULL ? 0 : size); 606 } 607 if (flags & M_WAITOK) 608 KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL")); 609 else if (va == NULL) 610 t_malloc_fail = time_uptime; 611 #ifdef DEBUG_REDZONE 612 if (va != NULL) 613 va = redzone_setup(va, osize); 614 #endif 615 return ((void *) va); 616 } 617 618 void * 619 mallocarray(size_t nmemb, size_t size, struct malloc_type *type, int flags) 620 { 621 622 if (WOULD_OVERFLOW(nmemb, size)) 623 panic("mallocarray: %zu * %zu overflowed", nmemb, size); 624 625 return (malloc(size * nmemb, type, flags)); 626 } 627 628 #ifdef INVARIANTS 629 static void 630 free_save_type(void *addr, struct malloc_type *mtp, u_long size) 631 { 632 struct malloc_type **mtpp = addr; 633 634 /* 635 * Cache a pointer to the malloc_type that most recently freed 636 * this memory here. This way we know who is most likely to 637 * have stepped on it later. 638 * 639 * This code assumes that size is a multiple of 8 bytes for 640 * 64 bit machines 641 */ 642 mtpp = (struct malloc_type **) ((unsigned long)mtpp & ~UMA_ALIGN_PTR); 643 mtpp += (size - sizeof(struct malloc_type *)) / 644 sizeof(struct malloc_type *); 645 *mtpp = mtp; 646 } 647 #endif 648 649 #ifdef MALLOC_DEBUG 650 static int 651 free_dbg(void **addrp, struct malloc_type *mtp) 652 { 653 void *addr; 654 655 addr = *addrp; 656 KASSERT(mtp->ks_magic == M_MAGIC, ("free: bad malloc type magic")); 657 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), 658 ("free: called with spinlock or critical section held")); 659 660 /* free(NULL, ...) does nothing */ 661 if (addr == NULL) 662 return (EJUSTRETURN); 663 664 #ifdef DEBUG_MEMGUARD 665 if (is_memguard_addr(addr)) { 666 memguard_free(addr); 667 return (EJUSTRETURN); 668 } 669 #endif 670 671 #ifdef DEBUG_REDZONE 672 redzone_check(addr); 673 *addrp = redzone_addr_ntor(addr); 674 #endif 675 676 return (0); 677 } 678 #endif 679 680 /* 681 * free: 682 * 683 * Free a block of memory allocated by malloc. 684 * 685 * This routine may not block. 686 */ 687 void 688 free(void *addr, struct malloc_type *mtp) 689 { 690 uma_slab_t slab; 691 u_long size; 692 693 #ifdef MALLOC_DEBUG 694 if (free_dbg(&addr, mtp) != 0) 695 return; 696 #endif 697 /* free(NULL, ...) does nothing */ 698 if (addr == NULL) 699 return; 700 701 slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK)); 702 if (slab == NULL) 703 panic("free: address %p(%p) has not been allocated.\n", 704 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK))); 705 706 if (!(slab->us_flags & UMA_SLAB_MALLOC)) { 707 size = slab->us_keg->uk_size; 708 #ifdef INVARIANTS 709 free_save_type(addr, mtp, size); 710 #endif 711 uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab); 712 } else { 713 size = slab->us_size; 714 uma_large_free(slab); 715 } 716 malloc_type_freed(mtp, size); 717 } 718 719 void 720 free_domain(void *addr, struct malloc_type *mtp) 721 { 722 uma_slab_t slab; 723 u_long size; 724 725 #ifdef MALLOC_DEBUG 726 if (free_dbg(&addr, mtp) != 0) 727 return; 728 #endif 729 730 /* free(NULL, ...) does nothing */ 731 if (addr == NULL) 732 return; 733 734 slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK)); 735 if (slab == NULL) 736 panic("free_domain: address %p(%p) has not been allocated.\n", 737 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK))); 738 739 if (!(slab->us_flags & UMA_SLAB_MALLOC)) { 740 size = slab->us_keg->uk_size; 741 #ifdef INVARIANTS 742 free_save_type(addr, mtp, size); 743 #endif 744 uma_zfree_domain(LIST_FIRST(&slab->us_keg->uk_zones), 745 addr, slab); 746 } else { 747 size = slab->us_size; 748 uma_large_free(slab); 749 } 750 malloc_type_freed(mtp, size); 751 } 752 753 /* 754 * realloc: change the size of a memory block 755 */ 756 void * 757 realloc(void *addr, size_t size, struct malloc_type *mtp, int flags) 758 { 759 uma_slab_t slab; 760 unsigned long alloc; 761 void *newaddr; 762 763 KASSERT(mtp->ks_magic == M_MAGIC, 764 ("realloc: bad malloc type magic")); 765 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), 766 ("realloc: called with spinlock or critical section held")); 767 768 /* realloc(NULL, ...) is equivalent to malloc(...) */ 769 if (addr == NULL) 770 return (malloc(size, mtp, flags)); 771 772 /* 773 * XXX: Should report free of old memory and alloc of new memory to 774 * per-CPU stats. 775 */ 776 777 #ifdef DEBUG_MEMGUARD 778 if (is_memguard_addr(addr)) 779 return (memguard_realloc(addr, size, mtp, flags)); 780 #endif 781 782 #ifdef DEBUG_REDZONE 783 slab = NULL; 784 alloc = redzone_get_size(addr); 785 #else 786 slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK)); 787 788 /* Sanity check */ 789 KASSERT(slab != NULL, 790 ("realloc: address %p out of range", (void *)addr)); 791 792 /* Get the size of the original block */ 793 if (!(slab->us_flags & UMA_SLAB_MALLOC)) 794 alloc = slab->us_keg->uk_size; 795 else 796 alloc = slab->us_size; 797 798 /* Reuse the original block if appropriate */ 799 if (size <= alloc 800 && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE)) 801 return (addr); 802 #endif /* !DEBUG_REDZONE */ 803 804 /* Allocate a new, bigger (or smaller) block */ 805 if ((newaddr = malloc(size, mtp, flags)) == NULL) 806 return (NULL); 807 808 /* Copy over original contents */ 809 bcopy(addr, newaddr, min(size, alloc)); 810 free(addr, mtp); 811 return (newaddr); 812 } 813 814 /* 815 * reallocf: same as realloc() but free memory on failure. 816 */ 817 void * 818 reallocf(void *addr, size_t size, struct malloc_type *mtp, int flags) 819 { 820 void *mem; 821 822 if ((mem = realloc(addr, size, mtp, flags)) == NULL) 823 free(addr, mtp); 824 return (mem); 825 } 826 827 #ifndef __sparc64__ 828 CTASSERT(VM_KMEM_SIZE_SCALE >= 1); 829 #endif 830 831 /* 832 * Initialize the kernel memory (kmem) arena. 833 */ 834 void 835 kmeminit(void) 836 { 837 u_long mem_size; 838 u_long tmp; 839 840 #ifdef VM_KMEM_SIZE 841 if (vm_kmem_size == 0) 842 vm_kmem_size = VM_KMEM_SIZE; 843 #endif 844 #ifdef VM_KMEM_SIZE_MIN 845 if (vm_kmem_size_min == 0) 846 vm_kmem_size_min = VM_KMEM_SIZE_MIN; 847 #endif 848 #ifdef VM_KMEM_SIZE_MAX 849 if (vm_kmem_size_max == 0) 850 vm_kmem_size_max = VM_KMEM_SIZE_MAX; 851 #endif 852 /* 853 * Calculate the amount of kernel virtual address (KVA) space that is 854 * preallocated to the kmem arena. In order to support a wide range 855 * of machines, it is a function of the physical memory size, 856 * specifically, 857 * 858 * min(max(physical memory size / VM_KMEM_SIZE_SCALE, 859 * VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX) 860 * 861 * Every architecture must define an integral value for 862 * VM_KMEM_SIZE_SCALE. However, the definitions of VM_KMEM_SIZE_MIN 863 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and 864 * ceiling on this preallocation, are optional. Typically, 865 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on 866 * a given architecture. 867 */ 868 mem_size = vm_cnt.v_page_count; 869 if (mem_size <= 32768) /* delphij XXX 128MB */ 870 kmem_zmax = PAGE_SIZE; 871 872 if (vm_kmem_size_scale < 1) 873 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE; 874 875 /* 876 * Check if we should use defaults for the "vm_kmem_size" 877 * variable: 878 */ 879 if (vm_kmem_size == 0) { 880 vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE; 881 882 if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min) 883 vm_kmem_size = vm_kmem_size_min; 884 if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max) 885 vm_kmem_size = vm_kmem_size_max; 886 } 887 888 /* 889 * The amount of KVA space that is preallocated to the 890 * kmem arena can be set statically at compile-time or manually 891 * through the kernel environment. However, it is still limited to 892 * twice the physical memory size, which has been sufficient to handle 893 * the most severe cases of external fragmentation in the kmem arena. 894 */ 895 if (vm_kmem_size / 2 / PAGE_SIZE > mem_size) 896 vm_kmem_size = 2 * mem_size * PAGE_SIZE; 897 898 vm_kmem_size = round_page(vm_kmem_size); 899 #ifdef DEBUG_MEMGUARD 900 tmp = memguard_fudge(vm_kmem_size, kernel_map); 901 #else 902 tmp = vm_kmem_size; 903 #endif 904 uma_set_limit(tmp); 905 906 #ifdef DEBUG_MEMGUARD 907 /* 908 * Initialize MemGuard if support compiled in. MemGuard is a 909 * replacement allocator used for detecting tamper-after-free 910 * scenarios as they occur. It is only used for debugging. 911 */ 912 memguard_init(kernel_arena); 913 #endif 914 } 915 916 /* 917 * Initialize the kernel memory allocator 918 */ 919 /* ARGSUSED*/ 920 static void 921 mallocinit(void *dummy) 922 { 923 int i; 924 uint8_t indx; 925 926 mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF); 927 928 kmeminit(); 929 930 if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX) 931 kmem_zmax = KMEM_ZMAX; 932 933 mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal), 934 #ifdef INVARIANTS 935 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 936 #else 937 NULL, NULL, NULL, NULL, 938 #endif 939 UMA_ALIGN_PTR, UMA_ZONE_MALLOC); 940 for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) { 941 int size = kmemzones[indx].kz_size; 942 char *name = kmemzones[indx].kz_name; 943 int subzone; 944 945 for (subzone = 0; subzone < numzones; subzone++) { 946 kmemzones[indx].kz_zone[subzone] = 947 uma_zcreate(name, size, 948 #ifdef INVARIANTS 949 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 950 #else 951 NULL, NULL, NULL, NULL, 952 #endif 953 UMA_ALIGN_PTR, UMA_ZONE_MALLOC); 954 } 955 for (;i <= size; i+= KMEM_ZBASE) 956 kmemsize[i >> KMEM_ZSHIFT] = indx; 957 958 } 959 } 960 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_SECOND, mallocinit, NULL); 961 962 void 963 malloc_init(void *data) 964 { 965 struct malloc_type_internal *mtip; 966 struct malloc_type *mtp; 967 968 KASSERT(vm_cnt.v_page_count != 0, ("malloc_register before vm_init")); 969 970 mtp = data; 971 if (mtp->ks_magic != M_MAGIC) 972 panic("malloc_init: bad malloc type magic"); 973 974 mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO); 975 mtp->ks_handle = mtip; 976 mtip->mti_zone = mtp_get_subzone(mtp->ks_shortdesc); 977 978 mtx_lock(&malloc_mtx); 979 mtp->ks_next = kmemstatistics; 980 kmemstatistics = mtp; 981 kmemcount++; 982 mtx_unlock(&malloc_mtx); 983 } 984 985 void 986 malloc_uninit(void *data) 987 { 988 struct malloc_type_internal *mtip; 989 struct malloc_type_stats *mtsp; 990 struct malloc_type *mtp, *temp; 991 uma_slab_t slab; 992 long temp_allocs, temp_bytes; 993 int i; 994 995 mtp = data; 996 KASSERT(mtp->ks_magic == M_MAGIC, 997 ("malloc_uninit: bad malloc type magic")); 998 KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL")); 999 1000 mtx_lock(&malloc_mtx); 1001 mtip = mtp->ks_handle; 1002 mtp->ks_handle = NULL; 1003 if (mtp != kmemstatistics) { 1004 for (temp = kmemstatistics; temp != NULL; 1005 temp = temp->ks_next) { 1006 if (temp->ks_next == mtp) { 1007 temp->ks_next = mtp->ks_next; 1008 break; 1009 } 1010 } 1011 KASSERT(temp, 1012 ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc)); 1013 } else 1014 kmemstatistics = mtp->ks_next; 1015 kmemcount--; 1016 mtx_unlock(&malloc_mtx); 1017 1018 /* 1019 * Look for memory leaks. 1020 */ 1021 temp_allocs = temp_bytes = 0; 1022 for (i = 0; i < MAXCPU; i++) { 1023 mtsp = &mtip->mti_stats[i]; 1024 temp_allocs += mtsp->mts_numallocs; 1025 temp_allocs -= mtsp->mts_numfrees; 1026 temp_bytes += mtsp->mts_memalloced; 1027 temp_bytes -= mtsp->mts_memfreed; 1028 } 1029 if (temp_allocs > 0 || temp_bytes > 0) { 1030 printf("Warning: memory type %s leaked memory on destroy " 1031 "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc, 1032 temp_allocs, temp_bytes); 1033 } 1034 1035 slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK)); 1036 uma_zfree_arg(mt_zone, mtip, slab); 1037 } 1038 1039 struct malloc_type * 1040 malloc_desc2type(const char *desc) 1041 { 1042 struct malloc_type *mtp; 1043 1044 mtx_assert(&malloc_mtx, MA_OWNED); 1045 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1046 if (strcmp(mtp->ks_shortdesc, desc) == 0) 1047 return (mtp); 1048 } 1049 return (NULL); 1050 } 1051 1052 static int 1053 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS) 1054 { 1055 struct malloc_type_stream_header mtsh; 1056 struct malloc_type_internal *mtip; 1057 struct malloc_type_header mth; 1058 struct malloc_type *mtp; 1059 int error, i; 1060 struct sbuf sbuf; 1061 1062 error = sysctl_wire_old_buffer(req, 0); 1063 if (error != 0) 1064 return (error); 1065 sbuf_new_for_sysctl(&sbuf, NULL, 128, req); 1066 sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL); 1067 mtx_lock(&malloc_mtx); 1068 1069 /* 1070 * Insert stream header. 1071 */ 1072 bzero(&mtsh, sizeof(mtsh)); 1073 mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION; 1074 mtsh.mtsh_maxcpus = MAXCPU; 1075 mtsh.mtsh_count = kmemcount; 1076 (void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh)); 1077 1078 /* 1079 * Insert alternating sequence of type headers and type statistics. 1080 */ 1081 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1082 mtip = (struct malloc_type_internal *)mtp->ks_handle; 1083 1084 /* 1085 * Insert type header. 1086 */ 1087 bzero(&mth, sizeof(mth)); 1088 strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME); 1089 (void)sbuf_bcat(&sbuf, &mth, sizeof(mth)); 1090 1091 /* 1092 * Insert type statistics for each CPU. 1093 */ 1094 for (i = 0; i < MAXCPU; i++) { 1095 (void)sbuf_bcat(&sbuf, &mtip->mti_stats[i], 1096 sizeof(mtip->mti_stats[i])); 1097 } 1098 } 1099 mtx_unlock(&malloc_mtx); 1100 error = sbuf_finish(&sbuf); 1101 sbuf_delete(&sbuf); 1102 return (error); 1103 } 1104 1105 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT, 1106 0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats", 1107 "Return malloc types"); 1108 1109 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0, 1110 "Count of kernel malloc types"); 1111 1112 void 1113 malloc_type_list(malloc_type_list_func_t *func, void *arg) 1114 { 1115 struct malloc_type *mtp, **bufmtp; 1116 int count, i; 1117 size_t buflen; 1118 1119 mtx_lock(&malloc_mtx); 1120 restart: 1121 mtx_assert(&malloc_mtx, MA_OWNED); 1122 count = kmemcount; 1123 mtx_unlock(&malloc_mtx); 1124 1125 buflen = sizeof(struct malloc_type *) * count; 1126 bufmtp = malloc(buflen, M_TEMP, M_WAITOK); 1127 1128 mtx_lock(&malloc_mtx); 1129 1130 if (count < kmemcount) { 1131 free(bufmtp, M_TEMP); 1132 goto restart; 1133 } 1134 1135 for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++) 1136 bufmtp[i] = mtp; 1137 1138 mtx_unlock(&malloc_mtx); 1139 1140 for (i = 0; i < count; i++) 1141 (func)(bufmtp[i], arg); 1142 1143 free(bufmtp, M_TEMP); 1144 } 1145 1146 #ifdef DDB 1147 DB_SHOW_COMMAND(malloc, db_show_malloc) 1148 { 1149 struct malloc_type_internal *mtip; 1150 struct malloc_type *mtp; 1151 uint64_t allocs, frees; 1152 uint64_t alloced, freed; 1153 int i; 1154 1155 db_printf("%18s %12s %12s %12s\n", "Type", "InUse", "MemUse", 1156 "Requests"); 1157 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1158 mtip = (struct malloc_type_internal *)mtp->ks_handle; 1159 allocs = 0; 1160 frees = 0; 1161 alloced = 0; 1162 freed = 0; 1163 for (i = 0; i < MAXCPU; i++) { 1164 allocs += mtip->mti_stats[i].mts_numallocs; 1165 frees += mtip->mti_stats[i].mts_numfrees; 1166 alloced += mtip->mti_stats[i].mts_memalloced; 1167 freed += mtip->mti_stats[i].mts_memfreed; 1168 } 1169 db_printf("%18s %12ju %12juK %12ju\n", 1170 mtp->ks_shortdesc, allocs - frees, 1171 (alloced - freed + 1023) / 1024, allocs); 1172 if (db_pager_quit) 1173 break; 1174 } 1175 } 1176 1177 #if MALLOC_DEBUG_MAXZONES > 1 1178 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches) 1179 { 1180 struct malloc_type_internal *mtip; 1181 struct malloc_type *mtp; 1182 u_int subzone; 1183 1184 if (!have_addr) { 1185 db_printf("Usage: show multizone_matches <malloc type/addr>\n"); 1186 return; 1187 } 1188 mtp = (void *)addr; 1189 if (mtp->ks_magic != M_MAGIC) { 1190 db_printf("Magic %lx does not match expected %x\n", 1191 mtp->ks_magic, M_MAGIC); 1192 return; 1193 } 1194 1195 mtip = mtp->ks_handle; 1196 subzone = mtip->mti_zone; 1197 1198 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1199 mtip = mtp->ks_handle; 1200 if (mtip->mti_zone != subzone) 1201 continue; 1202 db_printf("%s\n", mtp->ks_shortdesc); 1203 if (db_pager_quit) 1204 break; 1205 } 1206 } 1207 #endif /* MALLOC_DEBUG_MAXZONES > 1 */ 1208 #endif /* DDB */ 1209 1210 #ifdef MALLOC_PROFILE 1211 1212 static int 1213 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS) 1214 { 1215 struct sbuf sbuf; 1216 uint64_t count; 1217 uint64_t waste; 1218 uint64_t mem; 1219 int error; 1220 int rsize; 1221 int size; 1222 int i; 1223 1224 waste = 0; 1225 mem = 0; 1226 1227 error = sysctl_wire_old_buffer(req, 0); 1228 if (error != 0) 1229 return (error); 1230 sbuf_new_for_sysctl(&sbuf, NULL, 128, req); 1231 sbuf_printf(&sbuf, 1232 "\n Size Requests Real Size\n"); 1233 for (i = 0; i < KMEM_ZSIZE; i++) { 1234 size = i << KMEM_ZSHIFT; 1235 rsize = kmemzones[kmemsize[i]].kz_size; 1236 count = (long long unsigned)krequests[i]; 1237 1238 sbuf_printf(&sbuf, "%6d%28llu%11d\n", size, 1239 (unsigned long long)count, rsize); 1240 1241 if ((rsize * count) > (size * count)) 1242 waste += (rsize * count) - (size * count); 1243 mem += (rsize * count); 1244 } 1245 sbuf_printf(&sbuf, 1246 "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n", 1247 (unsigned long long)mem, (unsigned long long)waste); 1248 error = sbuf_finish(&sbuf); 1249 sbuf_delete(&sbuf); 1250 return (error); 1251 } 1252 1253 SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD, 1254 NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling"); 1255 #endif /* MALLOC_PROFILE */ 1256