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 <[email protected]> (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 extern void uma_startup2(void); 100 101 #if defined(INVARIANTS) || defined(MALLOC_MAKE_FAILURES) || \ 102 defined(DEBUG_MEMGUARD) || defined(DEBUG_REDZONE) 103 #define MALLOC_DEBUG 1 104 #endif 105 106 /* 107 * When realloc() is called, if the new size is sufficiently smaller than 108 * the old size, realloc() will allocate a new, smaller block to avoid 109 * wasting memory. 'Sufficiently smaller' is defined as: newsize <= 110 * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'. 111 */ 112 #ifndef REALLOC_FRACTION 113 #define REALLOC_FRACTION 1 /* new block if <= half the size */ 114 #endif 115 116 /* 117 * Centrally define some common malloc types. 118 */ 119 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches"); 120 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory"); 121 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers"); 122 123 static struct malloc_type *kmemstatistics; 124 static int kmemcount; 125 126 #define KMEM_ZSHIFT 4 127 #define KMEM_ZBASE 16 128 #define KMEM_ZMASK (KMEM_ZBASE - 1) 129 130 #define KMEM_ZMAX 65536 131 #define KMEM_ZSIZE (KMEM_ZMAX >> KMEM_ZSHIFT) 132 static uint8_t kmemsize[KMEM_ZSIZE + 1]; 133 134 #ifndef MALLOC_DEBUG_MAXZONES 135 #define MALLOC_DEBUG_MAXZONES 1 136 #endif 137 static int numzones = MALLOC_DEBUG_MAXZONES; 138 139 /* 140 * Small malloc(9) memory allocations are allocated from a set of UMA buckets 141 * of various sizes. 142 * 143 * XXX: The comment here used to read "These won't be powers of two for 144 * long." It's possible that a significant amount of wasted memory could be 145 * recovered by tuning the sizes of these buckets. 146 */ 147 struct { 148 int kz_size; 149 char *kz_name; 150 uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES]; 151 } kmemzones[] = { 152 {16, "16", }, 153 {32, "32", }, 154 {64, "64", }, 155 {128, "128", }, 156 {256, "256", }, 157 {512, "512", }, 158 {1024, "1024", }, 159 {2048, "2048", }, 160 {4096, "4096", }, 161 {8192, "8192", }, 162 {16384, "16384", }, 163 {32768, "32768", }, 164 {65536, "65536", }, 165 {0, NULL}, 166 }; 167 168 /* 169 * Zone to allocate malloc type descriptions from. For ABI reasons, memory 170 * types are described by a data structure passed by the declaring code, but 171 * the malloc(9) implementation has its own data structure describing the 172 * type and statistics. This permits the malloc(9)-internal data structures 173 * to be modified without breaking binary-compiled kernel modules that 174 * declare malloc types. 175 */ 176 static uma_zone_t mt_zone; 177 178 u_long vm_kmem_size; 179 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0, 180 "Size of kernel memory"); 181 182 static u_long kmem_zmax = KMEM_ZMAX; 183 SYSCTL_ULONG(_vm, OID_AUTO, kmem_zmax, CTLFLAG_RDTUN, &kmem_zmax, 0, 184 "Maximum allocation size that malloc(9) would use UMA as backend"); 185 186 static u_long vm_kmem_size_min; 187 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0, 188 "Minimum size of kernel memory"); 189 190 static u_long vm_kmem_size_max; 191 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0, 192 "Maximum size of kernel memory"); 193 194 static u_int vm_kmem_size_scale; 195 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0, 196 "Scale factor for kernel memory size"); 197 198 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS); 199 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size, 200 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, 201 sysctl_kmem_map_size, "LU", "Current kmem allocation size"); 202 203 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS); 204 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free, 205 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, 206 sysctl_kmem_map_free, "LU", "Free space in kmem"); 207 208 /* 209 * The malloc_mtx protects the kmemstatistics linked list. 210 */ 211 struct mtx malloc_mtx; 212 213 #ifdef MALLOC_PROFILE 214 uint64_t krequests[KMEM_ZSIZE + 1]; 215 216 static int sysctl_kern_mprof(SYSCTL_HANDLER_ARGS); 217 #endif 218 219 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS); 220 221 /* 222 * time_uptime of the last malloc(9) failure (induced or real). 223 */ 224 static time_t t_malloc_fail; 225 226 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1) 227 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD, 0, 228 "Kernel malloc debugging options"); 229 #endif 230 231 /* 232 * malloc(9) fault injection -- cause malloc failures every (n) mallocs when 233 * the caller specifies M_NOWAIT. If set to 0, no failures are caused. 234 */ 235 #ifdef MALLOC_MAKE_FAILURES 236 static int malloc_failure_rate; 237 static int malloc_nowait_count; 238 static int malloc_failure_count; 239 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RWTUN, 240 &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail"); 241 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD, 242 &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures"); 243 #endif 244 245 static int 246 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS) 247 { 248 u_long size; 249 250 size = uma_size(); 251 return (sysctl_handle_long(oidp, &size, 0, req)); 252 } 253 254 static int 255 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS) 256 { 257 u_long size, limit; 258 259 /* The sysctl is unsigned, implement as a saturation value. */ 260 size = uma_size(); 261 limit = uma_limit(); 262 if (size > limit) 263 size = 0; 264 else 265 size = limit - size; 266 return (sysctl_handle_long(oidp, &size, 0, req)); 267 } 268 269 /* 270 * malloc(9) uma zone separation -- sub-page buffer overruns in one 271 * malloc type will affect only a subset of other malloc types. 272 */ 273 #if MALLOC_DEBUG_MAXZONES > 1 274 static void 275 tunable_set_numzones(void) 276 { 277 278 TUNABLE_INT_FETCH("debug.malloc.numzones", 279 &numzones); 280 281 /* Sanity check the number of malloc uma zones. */ 282 if (numzones <= 0) 283 numzones = 1; 284 if (numzones > MALLOC_DEBUG_MAXZONES) 285 numzones = MALLOC_DEBUG_MAXZONES; 286 } 287 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL); 288 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 289 &numzones, 0, "Number of malloc uma subzones"); 290 291 /* 292 * Any number that changes regularly is an okay choice for the 293 * offset. Build numbers are pretty good of you have them. 294 */ 295 static u_int zone_offset = __FreeBSD_version; 296 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset); 297 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN, 298 &zone_offset, 0, "Separate malloc types by examining the " 299 "Nth character in the malloc type short description."); 300 301 static u_int 302 mtp_get_subzone(const char *desc) 303 { 304 size_t len; 305 u_int val; 306 307 if (desc == NULL || (len = strlen(desc)) == 0) 308 return (0); 309 val = desc[zone_offset % len]; 310 return (val % numzones); 311 } 312 #elif MALLOC_DEBUG_MAXZONES == 0 313 #error "MALLOC_DEBUG_MAXZONES must be positive." 314 #else 315 static inline u_int 316 mtp_get_subzone(const char *desc) 317 { 318 319 return (0); 320 } 321 #endif /* MALLOC_DEBUG_MAXZONES > 1 */ 322 323 int 324 malloc_last_fail(void) 325 { 326 327 return (time_uptime - t_malloc_fail); 328 } 329 330 /* 331 * An allocation has succeeded -- update malloc type statistics for the 332 * amount of bucket size. Occurs within a critical section so that the 333 * thread isn't preempted and doesn't migrate while updating per-PCU 334 * statistics. 335 */ 336 static void 337 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size, 338 int zindx) 339 { 340 struct malloc_type_internal *mtip; 341 struct malloc_type_stats *mtsp; 342 343 critical_enter(); 344 mtip = mtp->ks_handle; 345 mtsp = &mtip->mti_stats[curcpu]; 346 if (size > 0) { 347 mtsp->mts_memalloced += size; 348 mtsp->mts_numallocs++; 349 } 350 if (zindx != -1) 351 mtsp->mts_size |= 1 << zindx; 352 353 #ifdef KDTRACE_HOOKS 354 if (dtrace_malloc_probe != NULL) { 355 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC]; 356 if (probe_id != 0) 357 (dtrace_malloc_probe)(probe_id, 358 (uintptr_t) mtp, (uintptr_t) mtip, 359 (uintptr_t) mtsp, size, zindx); 360 } 361 #endif 362 363 critical_exit(); 364 } 365 366 void 367 malloc_type_allocated(struct malloc_type *mtp, unsigned long size) 368 { 369 370 if (size > 0) 371 malloc_type_zone_allocated(mtp, size, -1); 372 } 373 374 /* 375 * A free operation has occurred -- update malloc type statistics for the 376 * amount of the bucket size. Occurs within a critical section so that the 377 * thread isn't preempted and doesn't migrate while updating per-CPU 378 * statistics. 379 */ 380 void 381 malloc_type_freed(struct malloc_type *mtp, unsigned long size) 382 { 383 struct malloc_type_internal *mtip; 384 struct malloc_type_stats *mtsp; 385 386 critical_enter(); 387 mtip = mtp->ks_handle; 388 mtsp = &mtip->mti_stats[curcpu]; 389 mtsp->mts_memfreed += size; 390 mtsp->mts_numfrees++; 391 392 #ifdef KDTRACE_HOOKS 393 if (dtrace_malloc_probe != NULL) { 394 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE]; 395 if (probe_id != 0) 396 (dtrace_malloc_probe)(probe_id, 397 (uintptr_t) mtp, (uintptr_t) mtip, 398 (uintptr_t) mtsp, size, 0); 399 } 400 #endif 401 402 critical_exit(); 403 } 404 405 /* 406 * contigmalloc: 407 * 408 * Allocate a block of physically contiguous memory. 409 * 410 * If M_NOWAIT is set, this routine will not block and return NULL if 411 * the allocation fails. 412 */ 413 void * 414 contigmalloc(unsigned long size, struct malloc_type *type, int flags, 415 vm_paddr_t low, vm_paddr_t high, unsigned long alignment, 416 vm_paddr_t boundary) 417 { 418 void *ret; 419 420 ret = (void *)kmem_alloc_contig(kernel_arena, size, flags, low, high, 421 alignment, boundary, VM_MEMATTR_DEFAULT); 422 if (ret != NULL) 423 malloc_type_allocated(type, round_page(size)); 424 return (ret); 425 } 426 427 void * 428 contigmalloc_domain(unsigned long size, struct malloc_type *type, 429 int domain, int flags, vm_paddr_t low, vm_paddr_t high, 430 unsigned long alignment, vm_paddr_t boundary) 431 { 432 void *ret; 433 434 ret = (void *)kmem_alloc_contig_domain(domain, size, flags, low, high, 435 alignment, boundary, VM_MEMATTR_DEFAULT); 436 if (ret != NULL) 437 malloc_type_allocated(type, round_page(size)); 438 return (ret); 439 } 440 441 /* 442 * contigfree: 443 * 444 * Free a block of memory allocated by contigmalloc. 445 * 446 * This routine may not block. 447 */ 448 void 449 contigfree(void *addr, unsigned long size, struct malloc_type *type) 450 { 451 452 kmem_free(kernel_arena, (vm_offset_t)addr, size); 453 malloc_type_freed(type, round_page(size)); 454 } 455 456 #ifdef MALLOC_DEBUG 457 static int 458 malloc_dbg(caddr_t *vap, size_t *sizep, struct malloc_type *mtp, 459 int flags) 460 { 461 #ifdef INVARIANTS 462 int indx; 463 464 KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic")); 465 /* 466 * Check that exactly one of M_WAITOK or M_NOWAIT is specified. 467 */ 468 indx = flags & (M_WAITOK | M_NOWAIT); 469 if (indx != M_NOWAIT && indx != M_WAITOK) { 470 static struct timeval lasterr; 471 static int curerr, once; 472 if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) { 473 printf("Bad malloc flags: %x\n", indx); 474 kdb_backtrace(); 475 flags |= M_WAITOK; 476 once++; 477 } 478 } 479 #endif 480 #ifdef MALLOC_MAKE_FAILURES 481 if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) { 482 atomic_add_int(&malloc_nowait_count, 1); 483 if ((malloc_nowait_count % malloc_failure_rate) == 0) { 484 atomic_add_int(&malloc_failure_count, 1); 485 t_malloc_fail = time_uptime; 486 *vap = NULL; 487 return (EJUSTRETURN); 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 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), 495 ("malloc: called with spinlock or critical section held")); 496 497 #ifdef DEBUG_MEMGUARD 498 if (memguard_cmp_mtp(mtp, *sizep)) { 499 *vap = memguard_alloc(*sizep, flags); 500 if (*vap != NULL) 501 return (EJUSTRETURN); 502 /* This is unfortunate but should not be fatal. */ 503 } 504 #endif 505 506 #ifdef DEBUG_REDZONE 507 *sizep = redzone_size_ntor(*sizep); 508 #endif 509 510 return (0); 511 } 512 #endif 513 514 /* 515 * malloc: 516 * 517 * Allocate a block of memory. 518 * 519 * If M_NOWAIT is set, this routine will not block and return NULL if 520 * the allocation fails. 521 */ 522 void * 523 malloc(size_t size, struct malloc_type *mtp, int flags) 524 { 525 int indx; 526 struct malloc_type_internal *mtip; 527 caddr_t va; 528 uma_zone_t zone; 529 #if defined(DEBUG_REDZONE) 530 unsigned long osize = size; 531 #endif 532 533 #ifdef MALLOC_DEBUG 534 if (malloc_dbg(&va, &size, mtp, flags) != 0) 535 return (va); 536 #endif 537 538 if (size <= kmem_zmax) { 539 mtip = mtp->ks_handle; 540 if (size & KMEM_ZMASK) 541 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE; 542 indx = kmemsize[size >> KMEM_ZSHIFT]; 543 KASSERT(mtip->mti_zone < numzones, 544 ("mti_zone %u out of range %d", 545 mtip->mti_zone, numzones)); 546 zone = kmemzones[indx].kz_zone[mtip->mti_zone]; 547 #ifdef MALLOC_PROFILE 548 krequests[size >> KMEM_ZSHIFT]++; 549 #endif 550 va = uma_zalloc(zone, flags); 551 if (va != NULL) 552 size = zone->uz_size; 553 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx); 554 } else { 555 size = roundup(size, PAGE_SIZE); 556 zone = NULL; 557 va = uma_large_malloc(size, flags); 558 malloc_type_allocated(mtp, va == NULL ? 0 : size); 559 } 560 if (flags & M_WAITOK) 561 KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL")); 562 else if (va == NULL) 563 t_malloc_fail = time_uptime; 564 #ifdef DEBUG_REDZONE 565 if (va != NULL) 566 va = redzone_setup(va, osize); 567 #endif 568 return ((void *) va); 569 } 570 571 void * 572 malloc_domain(size_t size, struct malloc_type *mtp, int domain, 573 int flags) 574 { 575 int indx; 576 struct malloc_type_internal *mtip; 577 caddr_t va; 578 uma_zone_t zone; 579 #if defined(DEBUG_REDZONE) 580 unsigned long osize = size; 581 #endif 582 583 #ifdef MALLOC_DEBUG 584 if (malloc_dbg(&va, &size, mtp, flags) != 0) 585 return (va); 586 #endif 587 if (size <= kmem_zmax) { 588 mtip = mtp->ks_handle; 589 if (size & KMEM_ZMASK) 590 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE; 591 indx = kmemsize[size >> KMEM_ZSHIFT]; 592 KASSERT(mtip->mti_zone < numzones, 593 ("mti_zone %u out of range %d", 594 mtip->mti_zone, numzones)); 595 zone = kmemzones[indx].kz_zone[mtip->mti_zone]; 596 #ifdef MALLOC_PROFILE 597 krequests[size >> KMEM_ZSHIFT]++; 598 #endif 599 va = uma_zalloc_domain(zone, NULL, domain, flags); 600 if (va != NULL) 601 size = zone->uz_size; 602 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx); 603 } else { 604 size = roundup(size, PAGE_SIZE); 605 zone = NULL; 606 va = uma_large_malloc_domain(size, domain, flags); 607 malloc_type_allocated(mtp, va == NULL ? 0 : size); 608 } 609 if (flags & M_WAITOK) 610 KASSERT(va != NULL, ("malloc(M_WAITOK) returned NULL")); 611 else if (va == NULL) 612 t_malloc_fail = time_uptime; 613 #ifdef DEBUG_REDZONE 614 if (va != NULL) 615 va = redzone_setup(va, osize); 616 #endif 617 return ((void *) va); 618 } 619 620 void * 621 mallocarray(size_t nmemb, size_t size, struct malloc_type *type, int flags) 622 { 623 624 if (WOULD_OVERFLOW(nmemb, size)) 625 panic("mallocarray: %zu * %zu overflowed", nmemb, size); 626 627 return (malloc(size * nmemb, type, flags)); 628 } 629 630 #ifdef INVARIANTS 631 static void 632 free_save_type(void *addr, struct malloc_type *mtp, u_long size) 633 { 634 struct malloc_type **mtpp = addr; 635 636 /* 637 * Cache a pointer to the malloc_type that most recently freed 638 * this memory here. This way we know who is most likely to 639 * have stepped on it later. 640 * 641 * This code assumes that size is a multiple of 8 bytes for 642 * 64 bit machines 643 */ 644 mtpp = (struct malloc_type **) ((unsigned long)mtpp & ~UMA_ALIGN_PTR); 645 mtpp += (size - sizeof(struct malloc_type *)) / 646 sizeof(struct malloc_type *); 647 *mtpp = mtp; 648 } 649 #endif 650 651 #ifdef MALLOC_DEBUG 652 static int 653 free_dbg(void **addrp, struct malloc_type *mtp) 654 { 655 void *addr; 656 657 addr = *addrp; 658 KASSERT(mtp->ks_magic == M_MAGIC, ("free: bad malloc type magic")); 659 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), 660 ("free: called with spinlock or critical section held")); 661 662 /* free(NULL, ...) does nothing */ 663 if (addr == NULL) 664 return (EJUSTRETURN); 665 666 #ifdef DEBUG_MEMGUARD 667 if (is_memguard_addr(addr)) { 668 memguard_free(addr); 669 return (EJUSTRETURN); 670 } 671 #endif 672 673 #ifdef DEBUG_REDZONE 674 redzone_check(addr); 675 *addrp = redzone_addr_ntor(addr); 676 #endif 677 678 return (0); 679 } 680 #endif 681 682 /* 683 * free: 684 * 685 * Free a block of memory allocated by malloc. 686 * 687 * This routine may not block. 688 */ 689 void 690 free(void *addr, struct malloc_type *mtp) 691 { 692 uma_slab_t slab; 693 u_long size; 694 695 #ifdef MALLOC_DEBUG 696 if (free_dbg(&addr, mtp) != 0) 697 return; 698 #endif 699 /* free(NULL, ...) does nothing */ 700 if (addr == NULL) 701 return; 702 703 slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK)); 704 if (slab == NULL) 705 panic("free: address %p(%p) has not been allocated.\n", 706 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK))); 707 708 if (!(slab->us_flags & UMA_SLAB_MALLOC)) { 709 size = slab->us_keg->uk_size; 710 #ifdef INVARIANTS 711 free_save_type(addr, mtp, size); 712 #endif 713 uma_zfree_arg(LIST_FIRST(&slab->us_keg->uk_zones), addr, slab); 714 } else { 715 size = slab->us_size; 716 uma_large_free(slab); 717 } 718 malloc_type_freed(mtp, size); 719 } 720 721 void 722 free_domain(void *addr, struct malloc_type *mtp) 723 { 724 uma_slab_t slab; 725 u_long size; 726 727 #ifdef MALLOC_DEBUG 728 if (free_dbg(&addr, mtp) != 0) 729 return; 730 #endif 731 732 /* free(NULL, ...) does nothing */ 733 if (addr == NULL) 734 return; 735 736 slab = vtoslab((vm_offset_t)addr & (~UMA_SLAB_MASK)); 737 if (slab == NULL) 738 panic("free_domain: address %p(%p) has not been allocated.\n", 739 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK))); 740 741 if (!(slab->us_flags & UMA_SLAB_MALLOC)) { 742 size = slab->us_keg->uk_size; 743 #ifdef INVARIANTS 744 free_save_type(addr, mtp, size); 745 #endif 746 uma_zfree_domain(LIST_FIRST(&slab->us_keg->uk_zones), 747 addr, slab); 748 } else { 749 size = slab->us_size; 750 uma_large_free(slab); 751 } 752 malloc_type_freed(mtp, size); 753 } 754 755 /* 756 * realloc: change the size of a memory block 757 */ 758 void * 759 realloc(void *addr, size_t size, struct malloc_type *mtp, int flags) 760 { 761 uma_slab_t slab; 762 unsigned long alloc; 763 void *newaddr; 764 765 KASSERT(mtp->ks_magic == M_MAGIC, 766 ("realloc: bad malloc type magic")); 767 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), 768 ("realloc: called with spinlock or critical section held")); 769 770 /* realloc(NULL, ...) is equivalent to malloc(...) */ 771 if (addr == NULL) 772 return (malloc(size, mtp, flags)); 773 774 /* 775 * XXX: Should report free of old memory and alloc of new memory to 776 * per-CPU stats. 777 */ 778 779 #ifdef DEBUG_MEMGUARD 780 if (is_memguard_addr(addr)) 781 return (memguard_realloc(addr, size, mtp, flags)); 782 #endif 783 784 #ifdef DEBUG_REDZONE 785 slab = NULL; 786 alloc = redzone_get_size(addr); 787 #else 788 slab = vtoslab((vm_offset_t)addr & ~(UMA_SLAB_MASK)); 789 790 /* Sanity check */ 791 KASSERT(slab != NULL, 792 ("realloc: address %p out of range", (void *)addr)); 793 794 /* Get the size of the original block */ 795 if (!(slab->us_flags & UMA_SLAB_MALLOC)) 796 alloc = slab->us_keg->uk_size; 797 else 798 alloc = slab->us_size; 799 800 /* Reuse the original block if appropriate */ 801 if (size <= alloc 802 && (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE)) 803 return (addr); 804 #endif /* !DEBUG_REDZONE */ 805 806 /* Allocate a new, bigger (or smaller) block */ 807 if ((newaddr = malloc(size, mtp, flags)) == NULL) 808 return (NULL); 809 810 /* Copy over original contents */ 811 bcopy(addr, newaddr, min(size, alloc)); 812 free(addr, mtp); 813 return (newaddr); 814 } 815 816 /* 817 * reallocf: same as realloc() but free memory on failure. 818 */ 819 void * 820 reallocf(void *addr, size_t size, struct malloc_type *mtp, int flags) 821 { 822 void *mem; 823 824 if ((mem = realloc(addr, size, mtp, flags)) == NULL) 825 free(addr, mtp); 826 return (mem); 827 } 828 829 #ifndef __sparc64__ 830 CTASSERT(VM_KMEM_SIZE_SCALE >= 1); 831 #endif 832 833 /* 834 * Initialize the kernel memory (kmem) arena. 835 */ 836 void 837 kmeminit(void) 838 { 839 u_long mem_size; 840 u_long tmp; 841 842 #ifdef VM_KMEM_SIZE 843 if (vm_kmem_size == 0) 844 vm_kmem_size = VM_KMEM_SIZE; 845 #endif 846 #ifdef VM_KMEM_SIZE_MIN 847 if (vm_kmem_size_min == 0) 848 vm_kmem_size_min = VM_KMEM_SIZE_MIN; 849 #endif 850 #ifdef VM_KMEM_SIZE_MAX 851 if (vm_kmem_size_max == 0) 852 vm_kmem_size_max = VM_KMEM_SIZE_MAX; 853 #endif 854 /* 855 * Calculate the amount of kernel virtual address (KVA) space that is 856 * preallocated to the kmem arena. In order to support a wide range 857 * of machines, it is a function of the physical memory size, 858 * specifically, 859 * 860 * min(max(physical memory size / VM_KMEM_SIZE_SCALE, 861 * VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX) 862 * 863 * Every architecture must define an integral value for 864 * VM_KMEM_SIZE_SCALE. However, the definitions of VM_KMEM_SIZE_MIN 865 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and 866 * ceiling on this preallocation, are optional. Typically, 867 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on 868 * a given architecture. 869 */ 870 mem_size = vm_cnt.v_page_count; 871 if (mem_size <= 32768) /* delphij XXX 128MB */ 872 kmem_zmax = PAGE_SIZE; 873 874 if (vm_kmem_size_scale < 1) 875 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE; 876 877 /* 878 * Check if we should use defaults for the "vm_kmem_size" 879 * variable: 880 */ 881 if (vm_kmem_size == 0) { 882 vm_kmem_size = (mem_size / vm_kmem_size_scale) * PAGE_SIZE; 883 884 if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min) 885 vm_kmem_size = vm_kmem_size_min; 886 if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max) 887 vm_kmem_size = vm_kmem_size_max; 888 } 889 890 /* 891 * The amount of KVA space that is preallocated to the 892 * kmem arena can be set statically at compile-time or manually 893 * through the kernel environment. However, it is still limited to 894 * twice the physical memory size, which has been sufficient to handle 895 * the most severe cases of external fragmentation in the kmem arena. 896 */ 897 if (vm_kmem_size / 2 / PAGE_SIZE > mem_size) 898 vm_kmem_size = 2 * mem_size * PAGE_SIZE; 899 900 vm_kmem_size = round_page(vm_kmem_size); 901 #ifdef DEBUG_MEMGUARD 902 tmp = memguard_fudge(vm_kmem_size, kernel_map); 903 #else 904 tmp = vm_kmem_size; 905 #endif 906 uma_set_limit(tmp); 907 908 #ifdef DEBUG_MEMGUARD 909 /* 910 * Initialize MemGuard if support compiled in. MemGuard is a 911 * replacement allocator used for detecting tamper-after-free 912 * scenarios as they occur. It is only used for debugging. 913 */ 914 memguard_init(kernel_arena); 915 #endif 916 } 917 918 /* 919 * Initialize the kernel memory allocator 920 */ 921 /* ARGSUSED*/ 922 static void 923 mallocinit(void *dummy) 924 { 925 int i; 926 uint8_t indx; 927 928 mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF); 929 930 kmeminit(); 931 932 uma_startup2(); 933 934 if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX) 935 kmem_zmax = KMEM_ZMAX; 936 937 mt_zone = uma_zcreate("mt_zone", sizeof(struct malloc_type_internal), 938 #ifdef INVARIANTS 939 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 940 #else 941 NULL, NULL, NULL, NULL, 942 #endif 943 UMA_ALIGN_PTR, UMA_ZONE_MALLOC); 944 for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) { 945 int size = kmemzones[indx].kz_size; 946 char *name = kmemzones[indx].kz_name; 947 int subzone; 948 949 for (subzone = 0; subzone < numzones; subzone++) { 950 kmemzones[indx].kz_zone[subzone] = 951 uma_zcreate(name, size, 952 #ifdef INVARIANTS 953 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 954 #else 955 NULL, NULL, NULL, NULL, 956 #endif 957 UMA_ALIGN_PTR, UMA_ZONE_MALLOC); 958 } 959 for (;i <= size; i+= KMEM_ZBASE) 960 kmemsize[i >> KMEM_ZSHIFT] = indx; 961 962 } 963 } 964 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_SECOND, mallocinit, NULL); 965 966 void 967 malloc_init(void *data) 968 { 969 struct malloc_type_internal *mtip; 970 struct malloc_type *mtp; 971 972 KASSERT(vm_cnt.v_page_count != 0, ("malloc_register before vm_init")); 973 974 mtp = data; 975 if (mtp->ks_magic != M_MAGIC) 976 panic("malloc_init: bad malloc type magic"); 977 978 mtip = uma_zalloc(mt_zone, M_WAITOK | M_ZERO); 979 mtp->ks_handle = mtip; 980 mtip->mti_zone = mtp_get_subzone(mtp->ks_shortdesc); 981 982 mtx_lock(&malloc_mtx); 983 mtp->ks_next = kmemstatistics; 984 kmemstatistics = mtp; 985 kmemcount++; 986 mtx_unlock(&malloc_mtx); 987 } 988 989 void 990 malloc_uninit(void *data) 991 { 992 struct malloc_type_internal *mtip; 993 struct malloc_type_stats *mtsp; 994 struct malloc_type *mtp, *temp; 995 uma_slab_t slab; 996 long temp_allocs, temp_bytes; 997 int i; 998 999 mtp = data; 1000 KASSERT(mtp->ks_magic == M_MAGIC, 1001 ("malloc_uninit: bad malloc type magic")); 1002 KASSERT(mtp->ks_handle != NULL, ("malloc_deregister: cookie NULL")); 1003 1004 mtx_lock(&malloc_mtx); 1005 mtip = mtp->ks_handle; 1006 mtp->ks_handle = NULL; 1007 if (mtp != kmemstatistics) { 1008 for (temp = kmemstatistics; temp != NULL; 1009 temp = temp->ks_next) { 1010 if (temp->ks_next == mtp) { 1011 temp->ks_next = mtp->ks_next; 1012 break; 1013 } 1014 } 1015 KASSERT(temp, 1016 ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc)); 1017 } else 1018 kmemstatistics = mtp->ks_next; 1019 kmemcount--; 1020 mtx_unlock(&malloc_mtx); 1021 1022 /* 1023 * Look for memory leaks. 1024 */ 1025 temp_allocs = temp_bytes = 0; 1026 for (i = 0; i < MAXCPU; i++) { 1027 mtsp = &mtip->mti_stats[i]; 1028 temp_allocs += mtsp->mts_numallocs; 1029 temp_allocs -= mtsp->mts_numfrees; 1030 temp_bytes += mtsp->mts_memalloced; 1031 temp_bytes -= mtsp->mts_memfreed; 1032 } 1033 if (temp_allocs > 0 || temp_bytes > 0) { 1034 printf("Warning: memory type %s leaked memory on destroy " 1035 "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc, 1036 temp_allocs, temp_bytes); 1037 } 1038 1039 slab = vtoslab((vm_offset_t) mtip & (~UMA_SLAB_MASK)); 1040 uma_zfree_arg(mt_zone, mtip, slab); 1041 } 1042 1043 struct malloc_type * 1044 malloc_desc2type(const char *desc) 1045 { 1046 struct malloc_type *mtp; 1047 1048 mtx_assert(&malloc_mtx, MA_OWNED); 1049 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1050 if (strcmp(mtp->ks_shortdesc, desc) == 0) 1051 return (mtp); 1052 } 1053 return (NULL); 1054 } 1055 1056 static int 1057 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS) 1058 { 1059 struct malloc_type_stream_header mtsh; 1060 struct malloc_type_internal *mtip; 1061 struct malloc_type_header mth; 1062 struct malloc_type *mtp; 1063 int error, i; 1064 struct sbuf sbuf; 1065 1066 error = sysctl_wire_old_buffer(req, 0); 1067 if (error != 0) 1068 return (error); 1069 sbuf_new_for_sysctl(&sbuf, NULL, 128, req); 1070 sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL); 1071 mtx_lock(&malloc_mtx); 1072 1073 /* 1074 * Insert stream header. 1075 */ 1076 bzero(&mtsh, sizeof(mtsh)); 1077 mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION; 1078 mtsh.mtsh_maxcpus = MAXCPU; 1079 mtsh.mtsh_count = kmemcount; 1080 (void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh)); 1081 1082 /* 1083 * Insert alternating sequence of type headers and type statistics. 1084 */ 1085 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1086 mtip = (struct malloc_type_internal *)mtp->ks_handle; 1087 1088 /* 1089 * Insert type header. 1090 */ 1091 bzero(&mth, sizeof(mth)); 1092 strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME); 1093 (void)sbuf_bcat(&sbuf, &mth, sizeof(mth)); 1094 1095 /* 1096 * Insert type statistics for each CPU. 1097 */ 1098 for (i = 0; i < MAXCPU; i++) { 1099 (void)sbuf_bcat(&sbuf, &mtip->mti_stats[i], 1100 sizeof(mtip->mti_stats[i])); 1101 } 1102 } 1103 mtx_unlock(&malloc_mtx); 1104 error = sbuf_finish(&sbuf); 1105 sbuf_delete(&sbuf); 1106 return (error); 1107 } 1108 1109 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT, 1110 0, 0, sysctl_kern_malloc_stats, "s,malloc_type_ustats", 1111 "Return malloc types"); 1112 1113 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0, 1114 "Count of kernel malloc types"); 1115 1116 void 1117 malloc_type_list(malloc_type_list_func_t *func, void *arg) 1118 { 1119 struct malloc_type *mtp, **bufmtp; 1120 int count, i; 1121 size_t buflen; 1122 1123 mtx_lock(&malloc_mtx); 1124 restart: 1125 mtx_assert(&malloc_mtx, MA_OWNED); 1126 count = kmemcount; 1127 mtx_unlock(&malloc_mtx); 1128 1129 buflen = sizeof(struct malloc_type *) * count; 1130 bufmtp = malloc(buflen, M_TEMP, M_WAITOK); 1131 1132 mtx_lock(&malloc_mtx); 1133 1134 if (count < kmemcount) { 1135 free(bufmtp, M_TEMP); 1136 goto restart; 1137 } 1138 1139 for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++) 1140 bufmtp[i] = mtp; 1141 1142 mtx_unlock(&malloc_mtx); 1143 1144 for (i = 0; i < count; i++) 1145 (func)(bufmtp[i], arg); 1146 1147 free(bufmtp, M_TEMP); 1148 } 1149 1150 #ifdef DDB 1151 DB_SHOW_COMMAND(malloc, db_show_malloc) 1152 { 1153 struct malloc_type_internal *mtip; 1154 struct malloc_type *mtp; 1155 uint64_t allocs, frees; 1156 uint64_t alloced, freed; 1157 int i; 1158 1159 db_printf("%18s %12s %12s %12s\n", "Type", "InUse", "MemUse", 1160 "Requests"); 1161 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1162 mtip = (struct malloc_type_internal *)mtp->ks_handle; 1163 allocs = 0; 1164 frees = 0; 1165 alloced = 0; 1166 freed = 0; 1167 for (i = 0; i < MAXCPU; i++) { 1168 allocs += mtip->mti_stats[i].mts_numallocs; 1169 frees += mtip->mti_stats[i].mts_numfrees; 1170 alloced += mtip->mti_stats[i].mts_memalloced; 1171 freed += mtip->mti_stats[i].mts_memfreed; 1172 } 1173 db_printf("%18s %12ju %12juK %12ju\n", 1174 mtp->ks_shortdesc, allocs - frees, 1175 (alloced - freed + 1023) / 1024, allocs); 1176 if (db_pager_quit) 1177 break; 1178 } 1179 } 1180 1181 #if MALLOC_DEBUG_MAXZONES > 1 1182 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches) 1183 { 1184 struct malloc_type_internal *mtip; 1185 struct malloc_type *mtp; 1186 u_int subzone; 1187 1188 if (!have_addr) { 1189 db_printf("Usage: show multizone_matches <malloc type/addr>\n"); 1190 return; 1191 } 1192 mtp = (void *)addr; 1193 if (mtp->ks_magic != M_MAGIC) { 1194 db_printf("Magic %lx does not match expected %x\n", 1195 mtp->ks_magic, M_MAGIC); 1196 return; 1197 } 1198 1199 mtip = mtp->ks_handle; 1200 subzone = mtip->mti_zone; 1201 1202 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1203 mtip = mtp->ks_handle; 1204 if (mtip->mti_zone != subzone) 1205 continue; 1206 db_printf("%s\n", mtp->ks_shortdesc); 1207 if (db_pager_quit) 1208 break; 1209 } 1210 } 1211 #endif /* MALLOC_DEBUG_MAXZONES > 1 */ 1212 #endif /* DDB */ 1213 1214 #ifdef MALLOC_PROFILE 1215 1216 static int 1217 sysctl_kern_mprof(SYSCTL_HANDLER_ARGS) 1218 { 1219 struct sbuf sbuf; 1220 uint64_t count; 1221 uint64_t waste; 1222 uint64_t mem; 1223 int error; 1224 int rsize; 1225 int size; 1226 int i; 1227 1228 waste = 0; 1229 mem = 0; 1230 1231 error = sysctl_wire_old_buffer(req, 0); 1232 if (error != 0) 1233 return (error); 1234 sbuf_new_for_sysctl(&sbuf, NULL, 128, req); 1235 sbuf_printf(&sbuf, 1236 "\n Size Requests Real Size\n"); 1237 for (i = 0; i < KMEM_ZSIZE; i++) { 1238 size = i << KMEM_ZSHIFT; 1239 rsize = kmemzones[kmemsize[i]].kz_size; 1240 count = (long long unsigned)krequests[i]; 1241 1242 sbuf_printf(&sbuf, "%6d%28llu%11d\n", size, 1243 (unsigned long long)count, rsize); 1244 1245 if ((rsize * count) > (size * count)) 1246 waste += (rsize * count) - (size * count); 1247 mem += (rsize * count); 1248 } 1249 sbuf_printf(&sbuf, 1250 "\nTotal memory used:\t%30llu\nTotal Memory wasted:\t%30llu\n", 1251 (unsigned long long)mem, (unsigned long long)waste); 1252 error = sbuf_finish(&sbuf); 1253 sbuf_delete(&sbuf); 1254 return (error); 1255 } 1256 1257 SYSCTL_OID(_kern, OID_AUTO, mprof, CTLTYPE_STRING|CTLFLAG_RD, 1258 NULL, 0, sysctl_kern_mprof, "A", "Malloc Profiling"); 1259 #endif /* MALLOC_PROFILE */ 1260