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/asan.h> 57 #include <sys/kdb.h> 58 #include <sys/kernel.h> 59 #include <sys/lock.h> 60 #include <sys/malloc.h> 61 #include <sys/msan.h> 62 #include <sys/mutex.h> 63 #include <sys/vmmeter.h> 64 #include <sys/proc.h> 65 #include <sys/queue.h> 66 #include <sys/sbuf.h> 67 #include <sys/smp.h> 68 #include <sys/sysctl.h> 69 #include <sys/time.h> 70 #include <sys/vmem.h> 71 #ifdef EPOCH_TRACE 72 #include <sys/epoch.h> 73 #endif 74 75 #include <vm/vm.h> 76 #include <vm/pmap.h> 77 #include <vm/vm_domainset.h> 78 #include <vm/vm_pageout.h> 79 #include <vm/vm_param.h> 80 #include <vm/vm_kern.h> 81 #include <vm/vm_extern.h> 82 #include <vm/vm_map.h> 83 #include <vm/vm_page.h> 84 #include <vm/vm_phys.h> 85 #include <vm/vm_pagequeue.h> 86 #include <vm/uma.h> 87 #include <vm/uma_int.h> 88 #include <vm/uma_dbg.h> 89 90 #ifdef DEBUG_MEMGUARD 91 #include <vm/memguard.h> 92 #endif 93 #ifdef DEBUG_REDZONE 94 #include <vm/redzone.h> 95 #endif 96 97 #if defined(INVARIANTS) && defined(__i386__) 98 #include <machine/cpu.h> 99 #endif 100 101 #include <ddb/ddb.h> 102 103 #ifdef KDTRACE_HOOKS 104 #include <sys/dtrace_bsd.h> 105 106 bool __read_frequently dtrace_malloc_enabled; 107 dtrace_malloc_probe_func_t __read_mostly dtrace_malloc_probe; 108 #endif 109 110 #if defined(INVARIANTS) || defined(MALLOC_MAKE_FAILURES) || \ 111 defined(DEBUG_MEMGUARD) || defined(DEBUG_REDZONE) 112 #define MALLOC_DEBUG 1 113 #endif 114 115 #if defined(KASAN) || defined(DEBUG_REDZONE) 116 #define DEBUG_REDZONE_ARG_DEF , unsigned long osize 117 #define DEBUG_REDZONE_ARG , osize 118 #else 119 #define DEBUG_REDZONE_ARG_DEF 120 #define DEBUG_REDZONE_ARG 121 #endif 122 123 /* 124 * When realloc() is called, if the new size is sufficiently smaller than 125 * the old size, realloc() will allocate a new, smaller block to avoid 126 * wasting memory. 'Sufficiently smaller' is defined as: newsize <= 127 * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'. 128 */ 129 #ifndef REALLOC_FRACTION 130 #define REALLOC_FRACTION 1 /* new block if <= half the size */ 131 #endif 132 133 /* 134 * Centrally define some common malloc types. 135 */ 136 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches"); 137 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory"); 138 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers"); 139 140 static struct malloc_type *kmemstatistics; 141 static int kmemcount; 142 143 #define KMEM_ZSHIFT 4 144 #define KMEM_ZBASE 16 145 #define KMEM_ZMASK (KMEM_ZBASE - 1) 146 147 #define KMEM_ZMAX 65536 148 #define KMEM_ZSIZE (KMEM_ZMAX >> KMEM_ZSHIFT) 149 static uint8_t kmemsize[KMEM_ZSIZE + 1]; 150 151 #ifndef MALLOC_DEBUG_MAXZONES 152 #define MALLOC_DEBUG_MAXZONES 1 153 #endif 154 static int numzones = MALLOC_DEBUG_MAXZONES; 155 156 /* 157 * Small malloc(9) memory allocations are allocated from a set of UMA buckets 158 * of various sizes. 159 * 160 * Warning: the layout of the struct is duplicated in libmemstat for KVM support. 161 * 162 * XXX: The comment here used to read "These won't be powers of two for 163 * long." It's possible that a significant amount of wasted memory could be 164 * recovered by tuning the sizes of these buckets. 165 */ 166 struct { 167 int kz_size; 168 const char *kz_name; 169 uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES]; 170 } kmemzones[] = { 171 {16, "malloc-16", }, 172 {32, "malloc-32", }, 173 {64, "malloc-64", }, 174 {128, "malloc-128", }, 175 {256, "malloc-256", }, 176 {384, "malloc-384", }, 177 {512, "malloc-512", }, 178 {1024, "malloc-1024", }, 179 {2048, "malloc-2048", }, 180 {4096, "malloc-4096", }, 181 {8192, "malloc-8192", }, 182 {16384, "malloc-16384", }, 183 {32768, "malloc-32768", }, 184 {65536, "malloc-65536", }, 185 {0, NULL}, 186 }; 187 188 u_long vm_kmem_size; 189 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0, 190 "Size of kernel memory"); 191 192 static u_long kmem_zmax = KMEM_ZMAX; 193 SYSCTL_ULONG(_vm, OID_AUTO, kmem_zmax, CTLFLAG_RDTUN, &kmem_zmax, 0, 194 "Maximum allocation size that malloc(9) would use UMA as backend"); 195 196 static u_long vm_kmem_size_min; 197 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0, 198 "Minimum size of kernel memory"); 199 200 static u_long vm_kmem_size_max; 201 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0, 202 "Maximum size of kernel memory"); 203 204 static u_int vm_kmem_size_scale; 205 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0, 206 "Scale factor for kernel memory size"); 207 208 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS); 209 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size, 210 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, 211 sysctl_kmem_map_size, "LU", "Current kmem allocation size"); 212 213 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS); 214 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free, 215 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, 216 sysctl_kmem_map_free, "LU", "Free space in kmem"); 217 218 static SYSCTL_NODE(_vm, OID_AUTO, malloc, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 219 "Malloc information"); 220 221 static u_int vm_malloc_zone_count = nitems(kmemzones); 222 SYSCTL_UINT(_vm_malloc, OID_AUTO, zone_count, 223 CTLFLAG_RD, &vm_malloc_zone_count, 0, 224 "Number of malloc zones"); 225 226 static int sysctl_vm_malloc_zone_sizes(SYSCTL_HANDLER_ARGS); 227 SYSCTL_PROC(_vm_malloc, OID_AUTO, zone_sizes, 228 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, NULL, 0, 229 sysctl_vm_malloc_zone_sizes, "S", "Zone sizes used by malloc"); 230 231 /* 232 * The malloc_mtx protects the kmemstatistics linked list. 233 */ 234 struct mtx malloc_mtx; 235 236 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS); 237 238 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1) 239 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 240 "Kernel malloc debugging options"); 241 #endif 242 243 /* 244 * malloc(9) fault injection -- cause malloc failures every (n) mallocs when 245 * the caller specifies M_NOWAIT. If set to 0, no failures are caused. 246 */ 247 #ifdef MALLOC_MAKE_FAILURES 248 static int malloc_failure_rate; 249 static int malloc_nowait_count; 250 static int malloc_failure_count; 251 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RWTUN, 252 &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail"); 253 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD, 254 &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures"); 255 #endif 256 257 static int 258 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS) 259 { 260 u_long size; 261 262 size = uma_size(); 263 return (sysctl_handle_long(oidp, &size, 0, req)); 264 } 265 266 static int 267 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS) 268 { 269 u_long size, limit; 270 271 /* The sysctl is unsigned, implement as a saturation value. */ 272 size = uma_size(); 273 limit = uma_limit(); 274 if (size > limit) 275 size = 0; 276 else 277 size = limit - size; 278 return (sysctl_handle_long(oidp, &size, 0, req)); 279 } 280 281 static int 282 sysctl_vm_malloc_zone_sizes(SYSCTL_HANDLER_ARGS) 283 { 284 int sizes[nitems(kmemzones)]; 285 int i; 286 287 for (i = 0; i < nitems(kmemzones); i++) { 288 sizes[i] = kmemzones[i].kz_size; 289 } 290 291 return (SYSCTL_OUT(req, &sizes, sizeof(sizes))); 292 } 293 294 /* 295 * malloc(9) uma zone separation -- sub-page buffer overruns in one 296 * malloc type will affect only a subset of other malloc types. 297 */ 298 #if MALLOC_DEBUG_MAXZONES > 1 299 static void 300 tunable_set_numzones(void) 301 { 302 303 TUNABLE_INT_FETCH("debug.malloc.numzones", 304 &numzones); 305 306 /* Sanity check the number of malloc uma zones. */ 307 if (numzones <= 0) 308 numzones = 1; 309 if (numzones > MALLOC_DEBUG_MAXZONES) 310 numzones = MALLOC_DEBUG_MAXZONES; 311 } 312 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL); 313 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 314 &numzones, 0, "Number of malloc uma subzones"); 315 316 /* 317 * Any number that changes regularly is an okay choice for the 318 * offset. Build numbers are pretty good of you have them. 319 */ 320 static u_int zone_offset = __FreeBSD_version; 321 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset); 322 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN, 323 &zone_offset, 0, "Separate malloc types by examining the " 324 "Nth character in the malloc type short description."); 325 326 static void 327 mtp_set_subzone(struct malloc_type *mtp) 328 { 329 struct malloc_type_internal *mtip; 330 const char *desc; 331 size_t len; 332 u_int val; 333 334 mtip = &mtp->ks_mti; 335 desc = mtp->ks_shortdesc; 336 if (desc == NULL || (len = strlen(desc)) == 0) 337 val = 0; 338 else 339 val = desc[zone_offset % len]; 340 mtip->mti_zone = (val % numzones); 341 } 342 343 static inline u_int 344 mtp_get_subzone(struct malloc_type *mtp) 345 { 346 struct malloc_type_internal *mtip; 347 348 mtip = &mtp->ks_mti; 349 350 KASSERT(mtip->mti_zone < numzones, 351 ("mti_zone %u out of range %d", 352 mtip->mti_zone, numzones)); 353 return (mtip->mti_zone); 354 } 355 #elif MALLOC_DEBUG_MAXZONES == 0 356 #error "MALLOC_DEBUG_MAXZONES must be positive." 357 #else 358 static void 359 mtp_set_subzone(struct malloc_type *mtp) 360 { 361 struct malloc_type_internal *mtip; 362 363 mtip = &mtp->ks_mti; 364 mtip->mti_zone = 0; 365 } 366 367 static inline u_int 368 mtp_get_subzone(struct malloc_type *mtp) 369 { 370 371 return (0); 372 } 373 #endif /* MALLOC_DEBUG_MAXZONES > 1 */ 374 375 /* 376 * An allocation has succeeded -- update malloc type statistics for the 377 * amount of bucket size. Occurs within a critical section so that the 378 * thread isn't preempted and doesn't migrate while updating per-PCU 379 * statistics. 380 */ 381 static void 382 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size, 383 int zindx) 384 { 385 struct malloc_type_internal *mtip; 386 struct malloc_type_stats *mtsp; 387 388 critical_enter(); 389 mtip = &mtp->ks_mti; 390 mtsp = zpcpu_get(mtip->mti_stats); 391 if (size > 0) { 392 mtsp->mts_memalloced += size; 393 mtsp->mts_numallocs++; 394 } 395 if (zindx != -1) 396 mtsp->mts_size |= 1 << zindx; 397 398 #ifdef KDTRACE_HOOKS 399 if (__predict_false(dtrace_malloc_enabled)) { 400 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC]; 401 if (probe_id != 0) 402 (dtrace_malloc_probe)(probe_id, 403 (uintptr_t) mtp, (uintptr_t) mtip, 404 (uintptr_t) mtsp, size, zindx); 405 } 406 #endif 407 408 critical_exit(); 409 } 410 411 void 412 malloc_type_allocated(struct malloc_type *mtp, unsigned long size) 413 { 414 415 if (size > 0) 416 malloc_type_zone_allocated(mtp, size, -1); 417 } 418 419 /* 420 * A free operation has occurred -- update malloc type statistics for the 421 * amount of the bucket size. Occurs within a critical section so that the 422 * thread isn't preempted and doesn't migrate while updating per-CPU 423 * statistics. 424 */ 425 void 426 malloc_type_freed(struct malloc_type *mtp, unsigned long size) 427 { 428 struct malloc_type_internal *mtip; 429 struct malloc_type_stats *mtsp; 430 431 critical_enter(); 432 mtip = &mtp->ks_mti; 433 mtsp = zpcpu_get(mtip->mti_stats); 434 mtsp->mts_memfreed += size; 435 mtsp->mts_numfrees++; 436 437 #ifdef KDTRACE_HOOKS 438 if (__predict_false(dtrace_malloc_enabled)) { 439 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE]; 440 if (probe_id != 0) 441 (dtrace_malloc_probe)(probe_id, 442 (uintptr_t) mtp, (uintptr_t) mtip, 443 (uintptr_t) mtsp, size, 0); 444 } 445 #endif 446 447 critical_exit(); 448 } 449 450 /* 451 * contigmalloc: 452 * 453 * Allocate a block of physically contiguous memory. 454 * 455 * If M_NOWAIT is set, this routine will not block and return NULL if 456 * the allocation fails. 457 */ 458 void * 459 contigmalloc(unsigned long size, struct malloc_type *type, int flags, 460 vm_paddr_t low, vm_paddr_t high, unsigned long alignment, 461 vm_paddr_t boundary) 462 { 463 void *ret; 464 465 ret = (void *)kmem_alloc_contig(size, flags, low, high, alignment, 466 boundary, VM_MEMATTR_DEFAULT); 467 if (ret != NULL) 468 malloc_type_allocated(type, round_page(size)); 469 return (ret); 470 } 471 472 void * 473 contigmalloc_domainset(unsigned long size, struct malloc_type *type, 474 struct domainset *ds, int flags, vm_paddr_t low, vm_paddr_t high, 475 unsigned long alignment, vm_paddr_t boundary) 476 { 477 void *ret; 478 479 ret = (void *)kmem_alloc_contig_domainset(ds, size, flags, low, high, 480 alignment, boundary, VM_MEMATTR_DEFAULT); 481 if (ret != NULL) 482 malloc_type_allocated(type, round_page(size)); 483 return (ret); 484 } 485 486 /* 487 * contigfree: 488 * 489 * Free a block of memory allocated by contigmalloc. 490 * 491 * This routine may not block. 492 */ 493 void 494 contigfree(void *addr, unsigned long size, struct malloc_type *type) 495 { 496 497 kmem_free((vm_offset_t)addr, size); 498 malloc_type_freed(type, round_page(size)); 499 } 500 501 #ifdef MALLOC_DEBUG 502 static int 503 malloc_dbg(caddr_t *vap, size_t *sizep, struct malloc_type *mtp, 504 int flags) 505 { 506 #ifdef INVARIANTS 507 int indx; 508 509 KASSERT(mtp->ks_version == M_VERSION, ("malloc: bad malloc type version")); 510 /* 511 * Check that exactly one of M_WAITOK or M_NOWAIT is specified. 512 */ 513 indx = flags & (M_WAITOK | M_NOWAIT); 514 if (indx != M_NOWAIT && indx != M_WAITOK) { 515 static struct timeval lasterr; 516 static int curerr, once; 517 if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) { 518 printf("Bad malloc flags: %x\n", indx); 519 kdb_backtrace(); 520 flags |= M_WAITOK; 521 once++; 522 } 523 } 524 #endif 525 #ifdef MALLOC_MAKE_FAILURES 526 if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) { 527 atomic_add_int(&malloc_nowait_count, 1); 528 if ((malloc_nowait_count % malloc_failure_rate) == 0) { 529 atomic_add_int(&malloc_failure_count, 1); 530 *vap = NULL; 531 return (EJUSTRETURN); 532 } 533 } 534 #endif 535 if (flags & M_WAITOK) { 536 KASSERT(curthread->td_intr_nesting_level == 0, 537 ("malloc(M_WAITOK) in interrupt context")); 538 if (__predict_false(!THREAD_CAN_SLEEP())) { 539 #ifdef EPOCH_TRACE 540 epoch_trace_list(curthread); 541 #endif 542 KASSERT(0, 543 ("malloc(M_WAITOK) with sleeping prohibited")); 544 } 545 } 546 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), 547 ("malloc: called with spinlock or critical section held")); 548 549 #ifdef DEBUG_MEMGUARD 550 if (memguard_cmp_mtp(mtp, *sizep)) { 551 *vap = memguard_alloc(*sizep, flags); 552 if (*vap != NULL) 553 return (EJUSTRETURN); 554 /* This is unfortunate but should not be fatal. */ 555 } 556 #endif 557 558 #ifdef DEBUG_REDZONE 559 *sizep = redzone_size_ntor(*sizep); 560 #endif 561 562 return (0); 563 } 564 #endif 565 566 /* 567 * Handle large allocations and frees by using kmem_malloc directly. 568 */ 569 static inline bool 570 malloc_large_slab(uma_slab_t slab) 571 { 572 uintptr_t va; 573 574 va = (uintptr_t)slab; 575 return ((va & 1) != 0); 576 } 577 578 static inline size_t 579 malloc_large_size(uma_slab_t slab) 580 { 581 uintptr_t va; 582 583 va = (uintptr_t)slab; 584 return (va >> 1); 585 } 586 587 static caddr_t __noinline 588 malloc_large(size_t size, struct malloc_type *mtp, struct domainset *policy, 589 int flags DEBUG_REDZONE_ARG_DEF) 590 { 591 vm_offset_t kva; 592 caddr_t va; 593 594 size = roundup(size, PAGE_SIZE); 595 kva = kmem_malloc_domainset(policy, size, flags); 596 if (kva != 0) { 597 /* The low bit is unused for slab pointers. */ 598 vsetzoneslab(kva, NULL, (void *)((size << 1) | 1)); 599 uma_total_inc(size); 600 } 601 va = (caddr_t)kva; 602 malloc_type_allocated(mtp, va == NULL ? 0 : size); 603 if (__predict_false(va == NULL)) { 604 KASSERT((flags & M_WAITOK) == 0, 605 ("malloc(M_WAITOK) returned NULL")); 606 } else { 607 #ifdef DEBUG_REDZONE 608 va = redzone_setup(va, osize); 609 #endif 610 kasan_mark((void *)va, osize, size, KASAN_MALLOC_REDZONE); 611 } 612 return (va); 613 } 614 615 static void 616 free_large(void *addr, size_t size) 617 { 618 619 kmem_free((vm_offset_t)addr, size); 620 uma_total_dec(size); 621 } 622 623 /* 624 * malloc: 625 * 626 * Allocate a block of memory. 627 * 628 * If M_NOWAIT is set, this routine will not block and return NULL if 629 * the allocation fails. 630 */ 631 void * 632 (malloc)(size_t size, struct malloc_type *mtp, int flags) 633 { 634 int indx; 635 caddr_t va; 636 uma_zone_t zone; 637 #if defined(DEBUG_REDZONE) || defined(KASAN) 638 unsigned long osize = size; 639 #endif 640 641 MPASS((flags & M_EXEC) == 0); 642 643 #ifdef MALLOC_DEBUG 644 va = NULL; 645 if (malloc_dbg(&va, &size, mtp, flags) != 0) 646 return (va); 647 #endif 648 649 if (__predict_false(size > kmem_zmax)) 650 return (malloc_large(size, mtp, DOMAINSET_RR(), flags 651 DEBUG_REDZONE_ARG)); 652 653 if (size & KMEM_ZMASK) 654 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE; 655 indx = kmemsize[size >> KMEM_ZSHIFT]; 656 zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)]; 657 va = uma_zalloc(zone, flags); 658 if (va != NULL) { 659 size = zone->uz_size; 660 if ((flags & M_ZERO) == 0) { 661 kmsan_mark(va, size, KMSAN_STATE_UNINIT); 662 kmsan_orig(va, size, KMSAN_TYPE_MALLOC, KMSAN_RET_ADDR); 663 } 664 } 665 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx); 666 if (__predict_false(va == NULL)) { 667 KASSERT((flags & M_WAITOK) == 0, 668 ("malloc(M_WAITOK) returned NULL")); 669 } 670 #ifdef DEBUG_REDZONE 671 if (va != NULL) 672 va = redzone_setup(va, osize); 673 #endif 674 #ifdef KASAN 675 if (va != NULL) 676 kasan_mark((void *)va, osize, size, KASAN_MALLOC_REDZONE); 677 #endif 678 return ((void *) va); 679 } 680 681 static void * 682 malloc_domain(size_t *sizep, int *indxp, struct malloc_type *mtp, int domain, 683 int flags) 684 { 685 uma_zone_t zone; 686 caddr_t va; 687 size_t size; 688 int indx; 689 690 size = *sizep; 691 KASSERT(size <= kmem_zmax && (flags & M_EXEC) == 0, 692 ("malloc_domain: Called with bad flag / size combination.")); 693 if (size & KMEM_ZMASK) 694 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE; 695 indx = kmemsize[size >> KMEM_ZSHIFT]; 696 zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)]; 697 va = uma_zalloc_domain(zone, NULL, domain, flags); 698 if (va != NULL) 699 *sizep = zone->uz_size; 700 *indxp = indx; 701 return ((void *)va); 702 } 703 704 void * 705 malloc_domainset(size_t size, struct malloc_type *mtp, struct domainset *ds, 706 int flags) 707 { 708 struct vm_domainset_iter di; 709 caddr_t va; 710 int domain; 711 int indx; 712 #if defined(KASAN) || defined(DEBUG_REDZONE) 713 unsigned long osize = size; 714 #endif 715 716 MPASS((flags & M_EXEC) == 0); 717 718 #ifdef MALLOC_DEBUG 719 va = NULL; 720 if (malloc_dbg(&va, &size, mtp, flags) != 0) 721 return (va); 722 #endif 723 724 if (__predict_false(size > kmem_zmax)) 725 return (malloc_large(size, mtp, DOMAINSET_RR(), flags 726 DEBUG_REDZONE_ARG)); 727 728 vm_domainset_iter_policy_init(&di, ds, &domain, &flags); 729 do { 730 va = malloc_domain(&size, &indx, mtp, domain, flags); 731 } while (va == NULL && vm_domainset_iter_policy(&di, &domain) == 0); 732 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx); 733 if (__predict_false(va == NULL)) { 734 KASSERT((flags & M_WAITOK) == 0, 735 ("malloc(M_WAITOK) returned NULL")); 736 } 737 #ifdef DEBUG_REDZONE 738 if (va != NULL) 739 va = redzone_setup(va, osize); 740 #endif 741 #ifdef KASAN 742 if (va != NULL) 743 kasan_mark((void *)va, osize, size, KASAN_MALLOC_REDZONE); 744 #endif 745 #ifdef KMSAN 746 if ((flags & M_ZERO) == 0) { 747 kmsan_mark(va, size, KMSAN_STATE_UNINIT); 748 kmsan_orig(va, size, KMSAN_TYPE_MALLOC, KMSAN_RET_ADDR); 749 } 750 #endif 751 return (va); 752 } 753 754 /* 755 * Allocate an executable area. 756 */ 757 void * 758 malloc_exec(size_t size, struct malloc_type *mtp, int flags) 759 { 760 761 return (malloc_domainset_exec(size, mtp, DOMAINSET_RR(), flags)); 762 } 763 764 void * 765 malloc_domainset_exec(size_t size, struct malloc_type *mtp, struct domainset *ds, 766 int flags) 767 { 768 #if defined(DEBUG_REDZONE) || defined(KASAN) 769 unsigned long osize = size; 770 #endif 771 #ifdef MALLOC_DEBUG 772 caddr_t va; 773 #endif 774 775 flags |= M_EXEC; 776 777 #ifdef MALLOC_DEBUG 778 va = NULL; 779 if (malloc_dbg(&va, &size, mtp, flags) != 0) 780 return (va); 781 #endif 782 783 return (malloc_large(size, mtp, ds, flags DEBUG_REDZONE_ARG)); 784 } 785 786 void * 787 malloc_aligned(size_t size, size_t align, struct malloc_type *type, int flags) 788 { 789 return (malloc_domainset_aligned(size, align, type, DOMAINSET_RR(), 790 flags)); 791 } 792 793 void * 794 malloc_domainset_aligned(size_t size, size_t align, 795 struct malloc_type *mtp, struct domainset *ds, int flags) 796 { 797 void *res; 798 size_t asize; 799 800 KASSERT(align != 0 && powerof2(align), 801 ("malloc_domainset_aligned: wrong align %#zx size %#zx", 802 align, size)); 803 KASSERT(align <= PAGE_SIZE, 804 ("malloc_domainset_aligned: align %#zx (size %#zx) too large", 805 align, size)); 806 807 /* 808 * Round the allocation size up to the next power of 2, 809 * because we can only guarantee alignment for 810 * power-of-2-sized allocations. Further increase the 811 * allocation size to align if the rounded size is less than 812 * align, since malloc zones provide alignment equal to their 813 * size. 814 */ 815 asize = size <= align ? align : 1UL << flsl(size - 1); 816 817 res = malloc_domainset(asize, mtp, ds, flags); 818 KASSERT(res == NULL || ((uintptr_t)res & (align - 1)) == 0, 819 ("malloc_domainset_aligned: result not aligned %p size %#zx " 820 "allocsize %#zx align %#zx", res, size, asize, align)); 821 return (res); 822 } 823 824 void * 825 mallocarray(size_t nmemb, size_t size, struct malloc_type *type, int flags) 826 { 827 828 if (WOULD_OVERFLOW(nmemb, size)) 829 panic("mallocarray: %zu * %zu overflowed", nmemb, size); 830 831 return (malloc(size * nmemb, type, flags)); 832 } 833 834 void * 835 mallocarray_domainset(size_t nmemb, size_t size, struct malloc_type *type, 836 struct domainset *ds, int flags) 837 { 838 839 if (WOULD_OVERFLOW(nmemb, size)) 840 panic("mallocarray_domainset: %zu * %zu overflowed", nmemb, size); 841 842 return (malloc_domainset(size * nmemb, type, ds, flags)); 843 } 844 845 #if defined(INVARIANTS) && !defined(KASAN) 846 static void 847 free_save_type(void *addr, struct malloc_type *mtp, u_long size) 848 { 849 struct malloc_type **mtpp = addr; 850 851 /* 852 * Cache a pointer to the malloc_type that most recently freed 853 * this memory here. This way we know who is most likely to 854 * have stepped on it later. 855 * 856 * This code assumes that size is a multiple of 8 bytes for 857 * 64 bit machines 858 */ 859 mtpp = (struct malloc_type **) ((unsigned long)mtpp & ~UMA_ALIGN_PTR); 860 mtpp += (size - sizeof(struct malloc_type *)) / 861 sizeof(struct malloc_type *); 862 *mtpp = mtp; 863 } 864 #endif 865 866 #ifdef MALLOC_DEBUG 867 static int 868 free_dbg(void **addrp, struct malloc_type *mtp) 869 { 870 void *addr; 871 872 addr = *addrp; 873 KASSERT(mtp->ks_version == M_VERSION, ("free: bad malloc type version")); 874 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), 875 ("free: called with spinlock or critical section held")); 876 877 /* free(NULL, ...) does nothing */ 878 if (addr == NULL) 879 return (EJUSTRETURN); 880 881 #ifdef DEBUG_MEMGUARD 882 if (is_memguard_addr(addr)) { 883 memguard_free(addr); 884 return (EJUSTRETURN); 885 } 886 #endif 887 888 #ifdef DEBUG_REDZONE 889 redzone_check(addr); 890 *addrp = redzone_addr_ntor(addr); 891 #endif 892 893 return (0); 894 } 895 #endif 896 897 /* 898 * free: 899 * 900 * Free a block of memory allocated by malloc. 901 * 902 * This routine may not block. 903 */ 904 void 905 free(void *addr, struct malloc_type *mtp) 906 { 907 uma_zone_t zone; 908 uma_slab_t slab; 909 u_long size; 910 911 #ifdef MALLOC_DEBUG 912 if (free_dbg(&addr, mtp) != 0) 913 return; 914 #endif 915 /* free(NULL, ...) does nothing */ 916 if (addr == NULL) 917 return; 918 919 vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab); 920 if (slab == NULL) 921 panic("free: address %p(%p) has not been allocated.\n", 922 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK))); 923 924 if (__predict_true(!malloc_large_slab(slab))) { 925 size = zone->uz_size; 926 #if defined(INVARIANTS) && !defined(KASAN) 927 free_save_type(addr, mtp, size); 928 #endif 929 uma_zfree_arg(zone, addr, slab); 930 } else { 931 size = malloc_large_size(slab); 932 free_large(addr, size); 933 } 934 malloc_type_freed(mtp, size); 935 } 936 937 /* 938 * zfree: 939 * 940 * Zero then free a block of memory allocated by malloc. 941 * 942 * This routine may not block. 943 */ 944 void 945 zfree(void *addr, struct malloc_type *mtp) 946 { 947 uma_zone_t zone; 948 uma_slab_t slab; 949 u_long size; 950 951 #ifdef MALLOC_DEBUG 952 if (free_dbg(&addr, mtp) != 0) 953 return; 954 #endif 955 /* free(NULL, ...) does nothing */ 956 if (addr == NULL) 957 return; 958 959 vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab); 960 if (slab == NULL) 961 panic("free: address %p(%p) has not been allocated.\n", 962 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK))); 963 964 if (__predict_true(!malloc_large_slab(slab))) { 965 size = zone->uz_size; 966 #if defined(INVARIANTS) && !defined(KASAN) 967 free_save_type(addr, mtp, size); 968 #endif 969 kasan_mark(addr, size, size, 0); 970 explicit_bzero(addr, size); 971 uma_zfree_arg(zone, addr, slab); 972 } else { 973 size = malloc_large_size(slab); 974 kasan_mark(addr, size, size, 0); 975 explicit_bzero(addr, size); 976 free_large(addr, size); 977 } 978 malloc_type_freed(mtp, size); 979 } 980 981 /* 982 * realloc: change the size of a memory block 983 */ 984 void * 985 realloc(void *addr, size_t size, struct malloc_type *mtp, int flags) 986 { 987 uma_zone_t zone; 988 uma_slab_t slab; 989 unsigned long alloc; 990 void *newaddr; 991 992 KASSERT(mtp->ks_version == M_VERSION, 993 ("realloc: bad malloc type version")); 994 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), 995 ("realloc: called with spinlock or critical section held")); 996 997 /* realloc(NULL, ...) is equivalent to malloc(...) */ 998 if (addr == NULL) 999 return (malloc(size, mtp, flags)); 1000 1001 /* 1002 * XXX: Should report free of old memory and alloc of new memory to 1003 * per-CPU stats. 1004 */ 1005 1006 #ifdef DEBUG_MEMGUARD 1007 if (is_memguard_addr(addr)) 1008 return (memguard_realloc(addr, size, mtp, flags)); 1009 #endif 1010 1011 #ifdef DEBUG_REDZONE 1012 slab = NULL; 1013 zone = NULL; 1014 alloc = redzone_get_size(addr); 1015 #else 1016 vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab); 1017 1018 /* Sanity check */ 1019 KASSERT(slab != NULL, 1020 ("realloc: address %p out of range", (void *)addr)); 1021 1022 /* Get the size of the original block */ 1023 if (!malloc_large_slab(slab)) 1024 alloc = zone->uz_size; 1025 else 1026 alloc = malloc_large_size(slab); 1027 1028 /* Reuse the original block if appropriate */ 1029 if (size <= alloc && 1030 (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE)) { 1031 kasan_mark((void *)addr, size, alloc, KASAN_MALLOC_REDZONE); 1032 return (addr); 1033 } 1034 #endif /* !DEBUG_REDZONE */ 1035 1036 /* Allocate a new, bigger (or smaller) block */ 1037 if ((newaddr = malloc(size, mtp, flags)) == NULL) 1038 return (NULL); 1039 1040 /* 1041 * Copy over original contents. For KASAN, the redzone must be marked 1042 * valid before performing the copy. 1043 */ 1044 kasan_mark(addr, alloc, alloc, 0); 1045 bcopy(addr, newaddr, min(size, alloc)); 1046 free(addr, mtp); 1047 return (newaddr); 1048 } 1049 1050 /* 1051 * reallocf: same as realloc() but free memory on failure. 1052 */ 1053 void * 1054 reallocf(void *addr, size_t size, struct malloc_type *mtp, int flags) 1055 { 1056 void *mem; 1057 1058 if ((mem = realloc(addr, size, mtp, flags)) == NULL) 1059 free(addr, mtp); 1060 return (mem); 1061 } 1062 1063 /* 1064 * malloc_size: returns the number of bytes allocated for a request of the 1065 * specified size 1066 */ 1067 size_t 1068 malloc_size(size_t size) 1069 { 1070 int indx; 1071 1072 if (size > kmem_zmax) 1073 return (0); 1074 if (size & KMEM_ZMASK) 1075 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE; 1076 indx = kmemsize[size >> KMEM_ZSHIFT]; 1077 return (kmemzones[indx].kz_size); 1078 } 1079 1080 /* 1081 * malloc_usable_size: returns the usable size of the allocation. 1082 */ 1083 size_t 1084 malloc_usable_size(const void *addr) 1085 { 1086 #ifndef DEBUG_REDZONE 1087 uma_zone_t zone; 1088 uma_slab_t slab; 1089 #endif 1090 u_long size; 1091 1092 if (addr == NULL) 1093 return (0); 1094 1095 #ifdef DEBUG_MEMGUARD 1096 if (is_memguard_addr(__DECONST(void *, addr))) 1097 return (memguard_get_req_size(addr)); 1098 #endif 1099 1100 #ifdef DEBUG_REDZONE 1101 size = redzone_get_size(__DECONST(void *, addr)); 1102 #else 1103 vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab); 1104 if (slab == NULL) 1105 panic("malloc_usable_size: address %p(%p) is not allocated.\n", 1106 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK))); 1107 1108 if (!malloc_large_slab(slab)) 1109 size = zone->uz_size; 1110 else 1111 size = malloc_large_size(slab); 1112 #endif 1113 return (size); 1114 } 1115 1116 CTASSERT(VM_KMEM_SIZE_SCALE >= 1); 1117 1118 /* 1119 * Initialize the kernel memory (kmem) arena. 1120 */ 1121 void 1122 kmeminit(void) 1123 { 1124 u_long mem_size; 1125 u_long tmp; 1126 1127 #ifdef VM_KMEM_SIZE 1128 if (vm_kmem_size == 0) 1129 vm_kmem_size = VM_KMEM_SIZE; 1130 #endif 1131 #ifdef VM_KMEM_SIZE_MIN 1132 if (vm_kmem_size_min == 0) 1133 vm_kmem_size_min = VM_KMEM_SIZE_MIN; 1134 #endif 1135 #ifdef VM_KMEM_SIZE_MAX 1136 if (vm_kmem_size_max == 0) 1137 vm_kmem_size_max = VM_KMEM_SIZE_MAX; 1138 #endif 1139 /* 1140 * Calculate the amount of kernel virtual address (KVA) space that is 1141 * preallocated to the kmem arena. In order to support a wide range 1142 * of machines, it is a function of the physical memory size, 1143 * specifically, 1144 * 1145 * min(max(physical memory size / VM_KMEM_SIZE_SCALE, 1146 * VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX) 1147 * 1148 * Every architecture must define an integral value for 1149 * VM_KMEM_SIZE_SCALE. However, the definitions of VM_KMEM_SIZE_MIN 1150 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and 1151 * ceiling on this preallocation, are optional. Typically, 1152 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on 1153 * a given architecture. 1154 */ 1155 mem_size = vm_cnt.v_page_count; 1156 if (mem_size <= 32768) /* delphij XXX 128MB */ 1157 kmem_zmax = PAGE_SIZE; 1158 1159 if (vm_kmem_size_scale < 1) 1160 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE; 1161 1162 /* 1163 * Check if we should use defaults for the "vm_kmem_size" 1164 * variable: 1165 */ 1166 if (vm_kmem_size == 0) { 1167 vm_kmem_size = mem_size / vm_kmem_size_scale; 1168 vm_kmem_size = vm_kmem_size * PAGE_SIZE < vm_kmem_size ? 1169 vm_kmem_size_max : vm_kmem_size * PAGE_SIZE; 1170 if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min) 1171 vm_kmem_size = vm_kmem_size_min; 1172 if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max) 1173 vm_kmem_size = vm_kmem_size_max; 1174 } 1175 if (vm_kmem_size == 0) 1176 panic("Tune VM_KMEM_SIZE_* for the platform"); 1177 1178 /* 1179 * The amount of KVA space that is preallocated to the 1180 * kmem arena can be set statically at compile-time or manually 1181 * through the kernel environment. However, it is still limited to 1182 * twice the physical memory size, which has been sufficient to handle 1183 * the most severe cases of external fragmentation in the kmem arena. 1184 */ 1185 if (vm_kmem_size / 2 / PAGE_SIZE > mem_size) 1186 vm_kmem_size = 2 * mem_size * PAGE_SIZE; 1187 1188 vm_kmem_size = round_page(vm_kmem_size); 1189 1190 /* 1191 * With KASAN or KMSAN enabled, dynamically allocated kernel memory is 1192 * shadowed. Account for this when setting the UMA limit. 1193 */ 1194 #if defined(KASAN) 1195 vm_kmem_size = (vm_kmem_size * KASAN_SHADOW_SCALE) / 1196 (KASAN_SHADOW_SCALE + 1); 1197 #elif defined(KMSAN) 1198 vm_kmem_size /= 3; 1199 #endif 1200 1201 #ifdef DEBUG_MEMGUARD 1202 tmp = memguard_fudge(vm_kmem_size, kernel_map); 1203 #else 1204 tmp = vm_kmem_size; 1205 #endif 1206 uma_set_limit(tmp); 1207 1208 #ifdef DEBUG_MEMGUARD 1209 /* 1210 * Initialize MemGuard if support compiled in. MemGuard is a 1211 * replacement allocator used for detecting tamper-after-free 1212 * scenarios as they occur. It is only used for debugging. 1213 */ 1214 memguard_init(kernel_arena); 1215 #endif 1216 } 1217 1218 /* 1219 * Initialize the kernel memory allocator 1220 */ 1221 /* ARGSUSED*/ 1222 static void 1223 mallocinit(void *dummy) 1224 { 1225 int i; 1226 uint8_t indx; 1227 1228 mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF); 1229 1230 kmeminit(); 1231 1232 if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX) 1233 kmem_zmax = KMEM_ZMAX; 1234 1235 for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) { 1236 int size = kmemzones[indx].kz_size; 1237 const char *name = kmemzones[indx].kz_name; 1238 size_t align; 1239 int subzone; 1240 1241 align = UMA_ALIGN_PTR; 1242 if (powerof2(size) && size > sizeof(void *)) 1243 align = MIN(size, PAGE_SIZE) - 1; 1244 for (subzone = 0; subzone < numzones; subzone++) { 1245 kmemzones[indx].kz_zone[subzone] = 1246 uma_zcreate(name, size, 1247 #if defined(INVARIANTS) && !defined(KASAN) && !defined(KMSAN) 1248 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 1249 #else 1250 NULL, NULL, NULL, NULL, 1251 #endif 1252 align, UMA_ZONE_MALLOC); 1253 } 1254 for (;i <= size; i+= KMEM_ZBASE) 1255 kmemsize[i >> KMEM_ZSHIFT] = indx; 1256 } 1257 } 1258 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_SECOND, mallocinit, NULL); 1259 1260 void 1261 malloc_init(void *data) 1262 { 1263 struct malloc_type_internal *mtip; 1264 struct malloc_type *mtp; 1265 1266 KASSERT(vm_cnt.v_page_count != 0, ("malloc_register before vm_init")); 1267 1268 mtp = data; 1269 if (mtp->ks_version != M_VERSION) 1270 panic("malloc_init: type %s with unsupported version %lu", 1271 mtp->ks_shortdesc, mtp->ks_version); 1272 1273 mtip = &mtp->ks_mti; 1274 mtip->mti_stats = uma_zalloc_pcpu(pcpu_zone_64, M_WAITOK | M_ZERO); 1275 mtp_set_subzone(mtp); 1276 1277 mtx_lock(&malloc_mtx); 1278 mtp->ks_next = kmemstatistics; 1279 kmemstatistics = mtp; 1280 kmemcount++; 1281 mtx_unlock(&malloc_mtx); 1282 } 1283 1284 void 1285 malloc_uninit(void *data) 1286 { 1287 struct malloc_type_internal *mtip; 1288 struct malloc_type_stats *mtsp; 1289 struct malloc_type *mtp, *temp; 1290 long temp_allocs, temp_bytes; 1291 int i; 1292 1293 mtp = data; 1294 KASSERT(mtp->ks_version == M_VERSION, 1295 ("malloc_uninit: bad malloc type version")); 1296 1297 mtx_lock(&malloc_mtx); 1298 mtip = &mtp->ks_mti; 1299 if (mtp != kmemstatistics) { 1300 for (temp = kmemstatistics; temp != NULL; 1301 temp = temp->ks_next) { 1302 if (temp->ks_next == mtp) { 1303 temp->ks_next = mtp->ks_next; 1304 break; 1305 } 1306 } 1307 KASSERT(temp, 1308 ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc)); 1309 } else 1310 kmemstatistics = mtp->ks_next; 1311 kmemcount--; 1312 mtx_unlock(&malloc_mtx); 1313 1314 /* 1315 * Look for memory leaks. 1316 */ 1317 temp_allocs = temp_bytes = 0; 1318 for (i = 0; i <= mp_maxid; i++) { 1319 mtsp = zpcpu_get_cpu(mtip->mti_stats, i); 1320 temp_allocs += mtsp->mts_numallocs; 1321 temp_allocs -= mtsp->mts_numfrees; 1322 temp_bytes += mtsp->mts_memalloced; 1323 temp_bytes -= mtsp->mts_memfreed; 1324 } 1325 if (temp_allocs > 0 || temp_bytes > 0) { 1326 printf("Warning: memory type %s leaked memory on destroy " 1327 "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc, 1328 temp_allocs, temp_bytes); 1329 } 1330 1331 uma_zfree_pcpu(pcpu_zone_64, mtip->mti_stats); 1332 } 1333 1334 struct malloc_type * 1335 malloc_desc2type(const char *desc) 1336 { 1337 struct malloc_type *mtp; 1338 1339 mtx_assert(&malloc_mtx, MA_OWNED); 1340 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1341 if (strcmp(mtp->ks_shortdesc, desc) == 0) 1342 return (mtp); 1343 } 1344 return (NULL); 1345 } 1346 1347 static int 1348 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS) 1349 { 1350 struct malloc_type_stream_header mtsh; 1351 struct malloc_type_internal *mtip; 1352 struct malloc_type_stats *mtsp, zeromts; 1353 struct malloc_type_header mth; 1354 struct malloc_type *mtp; 1355 int error, i; 1356 struct sbuf sbuf; 1357 1358 error = sysctl_wire_old_buffer(req, 0); 1359 if (error != 0) 1360 return (error); 1361 sbuf_new_for_sysctl(&sbuf, NULL, 128, req); 1362 sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL); 1363 mtx_lock(&malloc_mtx); 1364 1365 bzero(&zeromts, sizeof(zeromts)); 1366 1367 /* 1368 * Insert stream header. 1369 */ 1370 bzero(&mtsh, sizeof(mtsh)); 1371 mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION; 1372 mtsh.mtsh_maxcpus = MAXCPU; 1373 mtsh.mtsh_count = kmemcount; 1374 (void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh)); 1375 1376 /* 1377 * Insert alternating sequence of type headers and type statistics. 1378 */ 1379 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1380 mtip = &mtp->ks_mti; 1381 1382 /* 1383 * Insert type header. 1384 */ 1385 bzero(&mth, sizeof(mth)); 1386 strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME); 1387 (void)sbuf_bcat(&sbuf, &mth, sizeof(mth)); 1388 1389 /* 1390 * Insert type statistics for each CPU. 1391 */ 1392 for (i = 0; i <= mp_maxid; i++) { 1393 mtsp = zpcpu_get_cpu(mtip->mti_stats, i); 1394 (void)sbuf_bcat(&sbuf, mtsp, sizeof(*mtsp)); 1395 } 1396 /* 1397 * Fill in the missing CPUs. 1398 */ 1399 for (; i < MAXCPU; i++) { 1400 (void)sbuf_bcat(&sbuf, &zeromts, sizeof(zeromts)); 1401 } 1402 } 1403 mtx_unlock(&malloc_mtx); 1404 error = sbuf_finish(&sbuf); 1405 sbuf_delete(&sbuf); 1406 return (error); 1407 } 1408 1409 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, 1410 CTLFLAG_RD | CTLTYPE_STRUCT | CTLFLAG_MPSAFE, 0, 0, 1411 sysctl_kern_malloc_stats, "s,malloc_type_ustats", 1412 "Return malloc types"); 1413 1414 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0, 1415 "Count of kernel malloc types"); 1416 1417 void 1418 malloc_type_list(malloc_type_list_func_t *func, void *arg) 1419 { 1420 struct malloc_type *mtp, **bufmtp; 1421 int count, i; 1422 size_t buflen; 1423 1424 mtx_lock(&malloc_mtx); 1425 restart: 1426 mtx_assert(&malloc_mtx, MA_OWNED); 1427 count = kmemcount; 1428 mtx_unlock(&malloc_mtx); 1429 1430 buflen = sizeof(struct malloc_type *) * count; 1431 bufmtp = malloc(buflen, M_TEMP, M_WAITOK); 1432 1433 mtx_lock(&malloc_mtx); 1434 1435 if (count < kmemcount) { 1436 free(bufmtp, M_TEMP); 1437 goto restart; 1438 } 1439 1440 for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++) 1441 bufmtp[i] = mtp; 1442 1443 mtx_unlock(&malloc_mtx); 1444 1445 for (i = 0; i < count; i++) 1446 (func)(bufmtp[i], arg); 1447 1448 free(bufmtp, M_TEMP); 1449 } 1450 1451 #ifdef DDB 1452 static int64_t 1453 get_malloc_stats(const struct malloc_type_internal *mtip, uint64_t *allocs, 1454 uint64_t *inuse) 1455 { 1456 const struct malloc_type_stats *mtsp; 1457 uint64_t frees, alloced, freed; 1458 int i; 1459 1460 *allocs = 0; 1461 frees = 0; 1462 alloced = 0; 1463 freed = 0; 1464 for (i = 0; i <= mp_maxid; i++) { 1465 mtsp = zpcpu_get_cpu(mtip->mti_stats, i); 1466 1467 *allocs += mtsp->mts_numallocs; 1468 frees += mtsp->mts_numfrees; 1469 alloced += mtsp->mts_memalloced; 1470 freed += mtsp->mts_memfreed; 1471 } 1472 *inuse = *allocs - frees; 1473 return (alloced - freed); 1474 } 1475 1476 DB_SHOW_COMMAND(malloc, db_show_malloc) 1477 { 1478 const char *fmt_hdr, *fmt_entry; 1479 struct malloc_type *mtp; 1480 uint64_t allocs, inuse; 1481 int64_t size; 1482 /* variables for sorting */ 1483 struct malloc_type *last_mtype, *cur_mtype; 1484 int64_t cur_size, last_size; 1485 int ties; 1486 1487 if (modif[0] == 'i') { 1488 fmt_hdr = "%s,%s,%s,%s\n"; 1489 fmt_entry = "\"%s\",%ju,%jdK,%ju\n"; 1490 } else { 1491 fmt_hdr = "%18s %12s %12s %12s\n"; 1492 fmt_entry = "%18s %12ju %12jdK %12ju\n"; 1493 } 1494 1495 db_printf(fmt_hdr, "Type", "InUse", "MemUse", "Requests"); 1496 1497 /* Select sort, largest size first. */ 1498 last_mtype = NULL; 1499 last_size = INT64_MAX; 1500 for (;;) { 1501 cur_mtype = NULL; 1502 cur_size = -1; 1503 ties = 0; 1504 1505 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1506 /* 1507 * In the case of size ties, print out mtypes 1508 * in the order they are encountered. That is, 1509 * when we encounter the most recently output 1510 * mtype, we have already printed all preceding 1511 * ties, and we must print all following ties. 1512 */ 1513 if (mtp == last_mtype) { 1514 ties = 1; 1515 continue; 1516 } 1517 size = get_malloc_stats(&mtp->ks_mti, &allocs, 1518 &inuse); 1519 if (size > cur_size && size < last_size + ties) { 1520 cur_size = size; 1521 cur_mtype = mtp; 1522 } 1523 } 1524 if (cur_mtype == NULL) 1525 break; 1526 1527 size = get_malloc_stats(&cur_mtype->ks_mti, &allocs, &inuse); 1528 db_printf(fmt_entry, cur_mtype->ks_shortdesc, inuse, 1529 howmany(size, 1024), allocs); 1530 1531 if (db_pager_quit) 1532 break; 1533 1534 last_mtype = cur_mtype; 1535 last_size = cur_size; 1536 } 1537 } 1538 1539 #if MALLOC_DEBUG_MAXZONES > 1 1540 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches) 1541 { 1542 struct malloc_type_internal *mtip; 1543 struct malloc_type *mtp; 1544 u_int subzone; 1545 1546 if (!have_addr) { 1547 db_printf("Usage: show multizone_matches <malloc type/addr>\n"); 1548 return; 1549 } 1550 mtp = (void *)addr; 1551 if (mtp->ks_version != M_VERSION) { 1552 db_printf("Version %lx does not match expected %x\n", 1553 mtp->ks_version, M_VERSION); 1554 return; 1555 } 1556 1557 mtip = &mtp->ks_mti; 1558 subzone = mtip->mti_zone; 1559 1560 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1561 mtip = &mtp->ks_mti; 1562 if (mtip->mti_zone != subzone) 1563 continue; 1564 db_printf("%s\n", mtp->ks_shortdesc); 1565 if (db_pager_quit) 1566 break; 1567 } 1568 } 1569 #endif /* MALLOC_DEBUG_MAXZONES > 1 */ 1570 #endif /* DDB */ 1571