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 35 /* 36 * Kernel malloc(9) implementation -- general purpose kernel memory allocator 37 * based on memory types. Back end is implemented using the UMA(9) zone 38 * allocator. A set of fixed-size buckets are used for smaller allocations, 39 * and a special UMA allocation interface is used for larger allocations. 40 * Callers declare memory types, and statistics are maintained independently 41 * for each memory type. Statistics are maintained per-CPU for performance 42 * reasons. See malloc(9) and comments in malloc.h for a detailed 43 * description. 44 */ 45 46 #include <sys/cdefs.h> 47 #include "opt_ddb.h" 48 #include "opt_vm.h" 49 50 #include <sys/param.h> 51 #include <sys/systm.h> 52 #include <sys/asan.h> 53 #include <sys/kdb.h> 54 #include <sys/kernel.h> 55 #include <sys/lock.h> 56 #include <sys/malloc.h> 57 #include <sys/msan.h> 58 #include <sys/mutex.h> 59 #include <sys/vmmeter.h> 60 #include <sys/proc.h> 61 #include <sys/queue.h> 62 #include <sys/sbuf.h> 63 #include <sys/smp.h> 64 #include <sys/sysctl.h> 65 #include <sys/time.h> 66 #include <sys/vmem.h> 67 #ifdef EPOCH_TRACE 68 #include <sys/epoch.h> 69 #endif 70 71 #include <vm/vm.h> 72 #include <vm/pmap.h> 73 #include <vm/vm_domainset.h> 74 #include <vm/vm_pageout.h> 75 #include <vm/vm_param.h> 76 #include <vm/vm_kern.h> 77 #include <vm/vm_extern.h> 78 #include <vm/vm_map.h> 79 #include <vm/vm_page.h> 80 #include <vm/vm_phys.h> 81 #include <vm/vm_pagequeue.h> 82 #include <vm/uma.h> 83 #include <vm/uma_int.h> 84 #include <vm/uma_dbg.h> 85 86 #ifdef DEBUG_MEMGUARD 87 #include <vm/memguard.h> 88 #endif 89 #ifdef DEBUG_REDZONE 90 #include <vm/redzone.h> 91 #endif 92 93 #if defined(INVARIANTS) && defined(__i386__) 94 #include <machine/cpu.h> 95 #endif 96 97 #include <ddb/ddb.h> 98 99 #ifdef KDTRACE_HOOKS 100 #include <sys/dtrace_bsd.h> 101 102 bool __read_frequently dtrace_malloc_enabled; 103 dtrace_malloc_probe_func_t __read_mostly dtrace_malloc_probe; 104 #endif 105 106 #if defined(INVARIANTS) || defined(MALLOC_MAKE_FAILURES) || \ 107 defined(DEBUG_MEMGUARD) || defined(DEBUG_REDZONE) 108 #define MALLOC_DEBUG 1 109 #endif 110 111 #if defined(KASAN) || defined(DEBUG_REDZONE) 112 #define DEBUG_REDZONE_ARG_DEF , unsigned long osize 113 #define DEBUG_REDZONE_ARG , osize 114 #else 115 #define DEBUG_REDZONE_ARG_DEF 116 #define DEBUG_REDZONE_ARG 117 #endif 118 119 typedef enum { 120 SLAB_COOKIE_SLAB_PTR = 0x0, 121 SLAB_COOKIE_MALLOC_LARGE = 0x1, 122 SLAB_COOKIE_CONTIG_MALLOC = 0x2, 123 } slab_cookie_t; 124 #define SLAB_COOKIE_MASK 0x3 125 #define SLAB_COOKIE_SHIFT 2 126 #define GET_SLAB_COOKIE(_slab) \ 127 ((slab_cookie_t)(uintptr_t)(_slab) & SLAB_COOKIE_MASK) 128 129 /* 130 * When realloc() is called, if the new size is sufficiently smaller than 131 * the old size, realloc() will allocate a new, smaller block to avoid 132 * wasting memory. 'Sufficiently smaller' is defined as: newsize <= 133 * oldsize / 2^n, where REALLOC_FRACTION defines the value of 'n'. 134 */ 135 #ifndef REALLOC_FRACTION 136 #define REALLOC_FRACTION 1 /* new block if <= half the size */ 137 #endif 138 139 /* 140 * Centrally define some common malloc types. 141 */ 142 MALLOC_DEFINE(M_CACHE, "cache", "Various Dynamically allocated caches"); 143 MALLOC_DEFINE(M_DEVBUF, "devbuf", "device driver memory"); 144 MALLOC_DEFINE(M_TEMP, "temp", "misc temporary data buffers"); 145 146 static struct malloc_type *kmemstatistics; 147 static int kmemcount; 148 149 #define KMEM_ZSHIFT 4 150 #define KMEM_ZBASE 16 151 #define KMEM_ZMASK (KMEM_ZBASE - 1) 152 153 #define KMEM_ZMAX 65536 154 #define KMEM_ZSIZE (KMEM_ZMAX >> KMEM_ZSHIFT) 155 static uint8_t kmemsize[KMEM_ZSIZE + 1]; 156 157 #ifndef MALLOC_DEBUG_MAXZONES 158 #define MALLOC_DEBUG_MAXZONES 1 159 #endif 160 static int numzones = MALLOC_DEBUG_MAXZONES; 161 162 /* 163 * Small malloc(9) memory allocations are allocated from a set of UMA buckets 164 * of various sizes. 165 * 166 * Warning: the layout of the struct is duplicated in libmemstat for KVM support. 167 * 168 * XXX: The comment here used to read "These won't be powers of two for 169 * long." It's possible that a significant amount of wasted memory could be 170 * recovered by tuning the sizes of these buckets. 171 */ 172 struct { 173 int kz_size; 174 const char *kz_name; 175 uma_zone_t kz_zone[MALLOC_DEBUG_MAXZONES]; 176 } kmemzones[] = { 177 {16, "malloc-16", }, 178 {32, "malloc-32", }, 179 {64, "malloc-64", }, 180 {128, "malloc-128", }, 181 {256, "malloc-256", }, 182 {384, "malloc-384", }, 183 {512, "malloc-512", }, 184 {1024, "malloc-1024", }, 185 {2048, "malloc-2048", }, 186 {4096, "malloc-4096", }, 187 {8192, "malloc-8192", }, 188 {16384, "malloc-16384", }, 189 {32768, "malloc-32768", }, 190 {65536, "malloc-65536", }, 191 {0, NULL}, 192 }; 193 194 u_long vm_kmem_size; 195 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size, CTLFLAG_RDTUN, &vm_kmem_size, 0, 196 "Size of kernel memory"); 197 198 static u_long kmem_zmax = KMEM_ZMAX; 199 SYSCTL_ULONG(_vm, OID_AUTO, kmem_zmax, CTLFLAG_RDTUN, &kmem_zmax, 0, 200 "Maximum allocation size that malloc(9) would use UMA as backend"); 201 202 static u_long vm_kmem_size_min; 203 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_min, CTLFLAG_RDTUN, &vm_kmem_size_min, 0, 204 "Minimum size of kernel memory"); 205 206 static u_long vm_kmem_size_max; 207 SYSCTL_ULONG(_vm, OID_AUTO, kmem_size_max, CTLFLAG_RDTUN, &vm_kmem_size_max, 0, 208 "Maximum size of kernel memory"); 209 210 static u_int vm_kmem_size_scale; 211 SYSCTL_UINT(_vm, OID_AUTO, kmem_size_scale, CTLFLAG_RDTUN, &vm_kmem_size_scale, 0, 212 "Scale factor for kernel memory size"); 213 214 static int sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS); 215 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_size, 216 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, 217 sysctl_kmem_map_size, "LU", "Current kmem allocation size"); 218 219 static int sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS); 220 SYSCTL_PROC(_vm, OID_AUTO, kmem_map_free, 221 CTLFLAG_RD | CTLTYPE_ULONG | CTLFLAG_MPSAFE, NULL, 0, 222 sysctl_kmem_map_free, "LU", "Free space in kmem"); 223 224 static SYSCTL_NODE(_vm, OID_AUTO, malloc, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 225 "Malloc information"); 226 227 static u_int vm_malloc_zone_count = nitems(kmemzones); 228 SYSCTL_UINT(_vm_malloc, OID_AUTO, zone_count, 229 CTLFLAG_RD, &vm_malloc_zone_count, 0, 230 "Number of malloc zones"); 231 232 static int sysctl_vm_malloc_zone_sizes(SYSCTL_HANDLER_ARGS); 233 SYSCTL_PROC(_vm_malloc, OID_AUTO, zone_sizes, 234 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, NULL, 0, 235 sysctl_vm_malloc_zone_sizes, "S", "Zone sizes used by malloc"); 236 237 /* 238 * The malloc_mtx protects the kmemstatistics linked list. 239 */ 240 struct mtx malloc_mtx; 241 242 static int sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS); 243 244 #if defined(MALLOC_MAKE_FAILURES) || (MALLOC_DEBUG_MAXZONES > 1) 245 static SYSCTL_NODE(_debug, OID_AUTO, malloc, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 246 "Kernel malloc debugging options"); 247 #endif 248 249 /* 250 * malloc(9) fault injection -- cause malloc failures every (n) mallocs when 251 * the caller specifies M_NOWAIT. If set to 0, no failures are caused. 252 */ 253 #ifdef MALLOC_MAKE_FAILURES 254 static int malloc_failure_rate; 255 static int malloc_nowait_count; 256 static int malloc_failure_count; 257 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_rate, CTLFLAG_RWTUN, 258 &malloc_failure_rate, 0, "Every (n) mallocs with M_NOWAIT will fail"); 259 SYSCTL_INT(_debug_malloc, OID_AUTO, failure_count, CTLFLAG_RD, 260 &malloc_failure_count, 0, "Number of imposed M_NOWAIT malloc failures"); 261 #endif 262 263 static int 264 sysctl_kmem_map_size(SYSCTL_HANDLER_ARGS) 265 { 266 u_long size; 267 268 size = uma_size(); 269 return (sysctl_handle_long(oidp, &size, 0, req)); 270 } 271 272 static int 273 sysctl_kmem_map_free(SYSCTL_HANDLER_ARGS) 274 { 275 u_long size, limit; 276 277 /* The sysctl is unsigned, implement as a saturation value. */ 278 size = uma_size(); 279 limit = uma_limit(); 280 if (size > limit) 281 size = 0; 282 else 283 size = limit - size; 284 return (sysctl_handle_long(oidp, &size, 0, req)); 285 } 286 287 static int 288 sysctl_vm_malloc_zone_sizes(SYSCTL_HANDLER_ARGS) 289 { 290 int sizes[nitems(kmemzones)]; 291 int i; 292 293 for (i = 0; i < nitems(kmemzones); i++) { 294 sizes[i] = kmemzones[i].kz_size; 295 } 296 297 return (SYSCTL_OUT(req, &sizes, sizeof(sizes))); 298 } 299 300 /* 301 * malloc(9) uma zone separation -- sub-page buffer overruns in one 302 * malloc type will affect only a subset of other malloc types. 303 */ 304 #if MALLOC_DEBUG_MAXZONES > 1 305 static void 306 tunable_set_numzones(void) 307 { 308 309 TUNABLE_INT_FETCH("debug.malloc.numzones", 310 &numzones); 311 312 /* Sanity check the number of malloc uma zones. */ 313 if (numzones <= 0) 314 numzones = 1; 315 if (numzones > MALLOC_DEBUG_MAXZONES) 316 numzones = MALLOC_DEBUG_MAXZONES; 317 } 318 SYSINIT(numzones, SI_SUB_TUNABLES, SI_ORDER_ANY, tunable_set_numzones, NULL); 319 SYSCTL_INT(_debug_malloc, OID_AUTO, numzones, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 320 &numzones, 0, "Number of malloc uma subzones"); 321 322 /* 323 * Any number that changes regularly is an okay choice for the 324 * offset. Build numbers are pretty good of you have them. 325 */ 326 static u_int zone_offset = __FreeBSD_version; 327 TUNABLE_INT("debug.malloc.zone_offset", &zone_offset); 328 SYSCTL_UINT(_debug_malloc, OID_AUTO, zone_offset, CTLFLAG_RDTUN, 329 &zone_offset, 0, "Separate malloc types by examining the " 330 "Nth character in the malloc type short description."); 331 332 static void 333 mtp_set_subzone(struct malloc_type *mtp) 334 { 335 struct malloc_type_internal *mtip; 336 const char *desc; 337 size_t len; 338 u_int val; 339 340 mtip = &mtp->ks_mti; 341 desc = mtp->ks_shortdesc; 342 if (desc == NULL || (len = strlen(desc)) == 0) 343 val = 0; 344 else 345 val = desc[zone_offset % len]; 346 mtip->mti_zone = (val % numzones); 347 } 348 349 static inline u_int 350 mtp_get_subzone(struct malloc_type *mtp) 351 { 352 struct malloc_type_internal *mtip; 353 354 mtip = &mtp->ks_mti; 355 356 KASSERT(mtip->mti_zone < numzones, 357 ("mti_zone %u out of range %d", 358 mtip->mti_zone, numzones)); 359 return (mtip->mti_zone); 360 } 361 #elif MALLOC_DEBUG_MAXZONES == 0 362 #error "MALLOC_DEBUG_MAXZONES must be positive." 363 #else 364 static void 365 mtp_set_subzone(struct malloc_type *mtp) 366 { 367 struct malloc_type_internal *mtip; 368 369 mtip = &mtp->ks_mti; 370 mtip->mti_zone = 0; 371 } 372 373 static inline u_int 374 mtp_get_subzone(struct malloc_type *mtp) 375 { 376 377 return (0); 378 } 379 #endif /* MALLOC_DEBUG_MAXZONES > 1 */ 380 381 /* 382 * An allocation has succeeded -- update malloc type statistics for the 383 * amount of bucket size. Occurs within a critical section so that the 384 * thread isn't preempted and doesn't migrate while updating per-PCU 385 * statistics. 386 */ 387 static void 388 malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size, 389 int zindx) 390 { 391 struct malloc_type_internal *mtip; 392 struct malloc_type_stats *mtsp; 393 394 critical_enter(); 395 mtip = &mtp->ks_mti; 396 mtsp = zpcpu_get(mtip->mti_stats); 397 if (size > 0) { 398 mtsp->mts_memalloced += size; 399 mtsp->mts_numallocs++; 400 } 401 if (zindx != -1) 402 mtsp->mts_size |= 1 << zindx; 403 404 #ifdef KDTRACE_HOOKS 405 if (__predict_false(dtrace_malloc_enabled)) { 406 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC]; 407 if (probe_id != 0) 408 (dtrace_malloc_probe)(probe_id, 409 (uintptr_t) mtp, (uintptr_t) mtip, 410 (uintptr_t) mtsp, size, zindx); 411 } 412 #endif 413 414 critical_exit(); 415 } 416 417 void 418 malloc_type_allocated(struct malloc_type *mtp, unsigned long size) 419 { 420 421 if (size > 0) 422 malloc_type_zone_allocated(mtp, size, -1); 423 } 424 425 /* 426 * A free operation has occurred -- update malloc type statistics for the 427 * amount of the bucket size. Occurs within a critical section so that the 428 * thread isn't preempted and doesn't migrate while updating per-CPU 429 * statistics. 430 */ 431 void 432 malloc_type_freed(struct malloc_type *mtp, unsigned long size) 433 { 434 struct malloc_type_internal *mtip; 435 struct malloc_type_stats *mtsp; 436 437 critical_enter(); 438 mtip = &mtp->ks_mti; 439 mtsp = zpcpu_get(mtip->mti_stats); 440 mtsp->mts_memfreed += size; 441 mtsp->mts_numfrees++; 442 443 #ifdef KDTRACE_HOOKS 444 if (__predict_false(dtrace_malloc_enabled)) { 445 uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE]; 446 if (probe_id != 0) 447 (dtrace_malloc_probe)(probe_id, 448 (uintptr_t) mtp, (uintptr_t) mtip, 449 (uintptr_t) mtsp, size, 0); 450 } 451 #endif 452 453 critical_exit(); 454 } 455 456 /* 457 * contigmalloc: 458 * 459 * Allocate a block of physically contiguous memory. 460 * 461 * If M_NOWAIT is set, this routine will not block and return NULL if 462 * the allocation fails. 463 */ 464 #define IS_CONTIG_MALLOC(_slab) \ 465 (GET_SLAB_COOKIE(_slab) == SLAB_COOKIE_CONTIG_MALLOC) 466 #define CONTIG_MALLOC_SLAB(_size) \ 467 ((void *)(((_size) << SLAB_COOKIE_SHIFT) | SLAB_COOKIE_CONTIG_MALLOC)) 468 static inline size_t 469 contigmalloc_size(uma_slab_t slab) 470 { 471 uintptr_t va; 472 473 KASSERT(IS_CONTIG_MALLOC(slab), 474 ("%s: called on non-contigmalloc allocation: %p", __func__, slab)); 475 va = (uintptr_t)slab; 476 return (va >> SLAB_COOKIE_SHIFT); 477 } 478 479 void * 480 contigmalloc(unsigned long size, struct malloc_type *type, int flags, 481 vm_paddr_t low, vm_paddr_t high, unsigned long alignment, 482 vm_paddr_t boundary) 483 { 484 void *ret; 485 486 ret = (void *)kmem_alloc_contig(size, flags, low, high, alignment, 487 boundary, VM_MEMATTR_DEFAULT); 488 if (ret != NULL) { 489 /* Use low bits unused for slab pointers. */ 490 vsetzoneslab((uintptr_t)ret, NULL, CONTIG_MALLOC_SLAB(size)); 491 malloc_type_allocated(type, round_page(size)); 492 } 493 return (ret); 494 } 495 496 void * 497 contigmalloc_domainset(unsigned long size, struct malloc_type *type, 498 struct domainset *ds, int flags, vm_paddr_t low, vm_paddr_t high, 499 unsigned long alignment, vm_paddr_t boundary) 500 { 501 void *ret; 502 503 ret = (void *)kmem_alloc_contig_domainset(ds, size, flags, low, high, 504 alignment, boundary, VM_MEMATTR_DEFAULT); 505 if (ret != NULL) { 506 /* Use low bits unused for slab pointers. */ 507 vsetzoneslab((uintptr_t)ret, NULL, CONTIG_MALLOC_SLAB(size)); 508 malloc_type_allocated(type, round_page(size)); 509 } 510 return (ret); 511 } 512 #undef IS_CONTIG_MALLOC 513 #undef CONTIG_MALLOC_SLAB 514 515 /* contigfree(9) is deprecated. */ 516 void 517 contigfree(void *addr, unsigned long size __unused, struct malloc_type *type) 518 { 519 free(addr, type); 520 } 521 522 #ifdef MALLOC_DEBUG 523 static int 524 malloc_dbg(caddr_t *vap, size_t *sizep, struct malloc_type *mtp, 525 int flags) 526 { 527 #ifdef INVARIANTS 528 int indx; 529 530 KASSERT(mtp->ks_version == M_VERSION, ("malloc: bad malloc type version")); 531 /* 532 * Check that exactly one of M_WAITOK or M_NOWAIT is specified. 533 */ 534 indx = flags & (M_WAITOK | M_NOWAIT); 535 if (indx != M_NOWAIT && indx != M_WAITOK) { 536 static struct timeval lasterr; 537 static int curerr, once; 538 if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) { 539 printf("Bad malloc flags: %x\n", indx); 540 kdb_backtrace(); 541 flags |= M_WAITOK; 542 once++; 543 } 544 } 545 KASSERT((flags & M_NEVERFREED) == 0, 546 ("malloc: M_NEVERFREED is for internal use only")); 547 #endif 548 #ifdef MALLOC_MAKE_FAILURES 549 if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) { 550 atomic_add_int(&malloc_nowait_count, 1); 551 if ((malloc_nowait_count % malloc_failure_rate) == 0) { 552 atomic_add_int(&malloc_failure_count, 1); 553 *vap = NULL; 554 return (EJUSTRETURN); 555 } 556 } 557 #endif 558 if (flags & M_WAITOK) { 559 KASSERT(curthread->td_intr_nesting_level == 0, 560 ("malloc(M_WAITOK) in interrupt context")); 561 if (__predict_false(!THREAD_CAN_SLEEP())) { 562 #ifdef EPOCH_TRACE 563 epoch_trace_list(curthread); 564 #endif 565 KASSERT(0, 566 ("malloc(M_WAITOK) with sleeping prohibited")); 567 } 568 } 569 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), 570 ("malloc: called with spinlock or critical section held")); 571 572 #ifdef DEBUG_MEMGUARD 573 if (memguard_cmp_mtp(mtp, *sizep)) { 574 *vap = memguard_alloc(*sizep, flags); 575 if (*vap != NULL) 576 return (EJUSTRETURN); 577 /* This is unfortunate but should not be fatal. */ 578 } 579 #endif 580 581 #ifdef DEBUG_REDZONE 582 *sizep = redzone_size_ntor(*sizep); 583 #endif 584 585 return (0); 586 } 587 #endif 588 589 /* 590 * Handle large allocations and frees by using kmem_malloc directly. 591 */ 592 #define IS_MALLOC_LARGE(_slab) \ 593 (GET_SLAB_COOKIE(_slab) == SLAB_COOKIE_MALLOC_LARGE) 594 #define MALLOC_LARGE_SLAB(_size) \ 595 ((void *)(((_size) << SLAB_COOKIE_SHIFT) | SLAB_COOKIE_MALLOC_LARGE)) 596 static inline size_t 597 malloc_large_size(uma_slab_t slab) 598 { 599 uintptr_t va; 600 601 va = (uintptr_t)slab; 602 KASSERT(IS_MALLOC_LARGE(slab), 603 ("%s: called on non-malloc_large allocation: %p", __func__, slab)); 604 return (va >> SLAB_COOKIE_SHIFT); 605 } 606 607 static caddr_t __noinline 608 malloc_large(size_t size, struct malloc_type *mtp, struct domainset *policy, 609 int flags DEBUG_REDZONE_ARG_DEF) 610 { 611 void *va; 612 613 size = roundup(size, PAGE_SIZE); 614 va = kmem_malloc_domainset(policy, size, flags); 615 if (va != NULL) { 616 /* Use low bits unused for slab pointers. */ 617 vsetzoneslab((uintptr_t)va, NULL, MALLOC_LARGE_SLAB(size)); 618 uma_total_inc(size); 619 } 620 malloc_type_allocated(mtp, va == NULL ? 0 : size); 621 if (__predict_false(va == NULL)) { 622 KASSERT((flags & M_WAITOK) == 0, 623 ("malloc(M_WAITOK) returned NULL")); 624 } else { 625 #ifdef DEBUG_REDZONE 626 va = redzone_setup(va, osize); 627 #endif 628 kasan_mark(va, osize, size, KASAN_MALLOC_REDZONE); 629 } 630 return (va); 631 } 632 633 static void 634 free_large(void *addr, size_t size) 635 { 636 637 kmem_free(addr, size); 638 uma_total_dec(size); 639 } 640 #undef IS_MALLOC_LARGE 641 #undef MALLOC_LARGE_SLAB 642 643 /* 644 * malloc: 645 * 646 * Allocate a block of memory. 647 * 648 * If M_NOWAIT is set, this routine will not block and return NULL if 649 * the allocation fails. 650 */ 651 void * 652 (malloc)(size_t size, struct malloc_type *mtp, int flags) 653 { 654 int indx; 655 caddr_t va; 656 uma_zone_t zone; 657 #if defined(DEBUG_REDZONE) || defined(KASAN) 658 unsigned long osize = size; 659 #endif 660 661 MPASS((flags & M_EXEC) == 0); 662 663 #ifdef MALLOC_DEBUG 664 va = NULL; 665 if (malloc_dbg(&va, &size, mtp, flags) != 0) 666 return (va); 667 #endif 668 669 if (__predict_false(size > kmem_zmax)) 670 return (malloc_large(size, mtp, DOMAINSET_RR(), flags 671 DEBUG_REDZONE_ARG)); 672 673 if (size & KMEM_ZMASK) 674 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE; 675 indx = kmemsize[size >> KMEM_ZSHIFT]; 676 zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)]; 677 va = uma_zalloc_arg(zone, zone, flags); 678 if (va != NULL) { 679 size = zone->uz_size; 680 if ((flags & M_ZERO) == 0) { 681 kmsan_mark(va, size, KMSAN_STATE_UNINIT); 682 kmsan_orig(va, size, KMSAN_TYPE_MALLOC, KMSAN_RET_ADDR); 683 } 684 } 685 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx); 686 if (__predict_false(va == NULL)) { 687 KASSERT((flags & M_WAITOK) == 0, 688 ("malloc(M_WAITOK) returned NULL")); 689 } 690 #ifdef DEBUG_REDZONE 691 if (va != NULL) 692 va = redzone_setup(va, osize); 693 #endif 694 #ifdef KASAN 695 if (va != NULL) 696 kasan_mark((void *)va, osize, size, KASAN_MALLOC_REDZONE); 697 #endif 698 return ((void *) va); 699 } 700 701 static void * 702 malloc_domain(size_t *sizep, int *indxp, struct malloc_type *mtp, int domain, 703 int flags) 704 { 705 uma_zone_t zone; 706 caddr_t va; 707 size_t size; 708 int indx; 709 710 size = *sizep; 711 KASSERT(size <= kmem_zmax && (flags & M_EXEC) == 0, 712 ("malloc_domain: Called with bad flag / size combination")); 713 if (size & KMEM_ZMASK) 714 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE; 715 indx = kmemsize[size >> KMEM_ZSHIFT]; 716 zone = kmemzones[indx].kz_zone[mtp_get_subzone(mtp)]; 717 va = uma_zalloc_domain(zone, zone, domain, flags); 718 if (va != NULL) 719 *sizep = zone->uz_size; 720 *indxp = indx; 721 return ((void *)va); 722 } 723 724 void * 725 malloc_domainset(size_t size, struct malloc_type *mtp, struct domainset *ds, 726 int flags) 727 { 728 struct vm_domainset_iter di; 729 caddr_t va; 730 int domain; 731 int indx; 732 #if defined(KASAN) || defined(DEBUG_REDZONE) 733 unsigned long osize = size; 734 #endif 735 736 MPASS((flags & M_EXEC) == 0); 737 738 #ifdef MALLOC_DEBUG 739 va = NULL; 740 if (malloc_dbg(&va, &size, mtp, flags) != 0) 741 return (va); 742 #endif 743 744 if (__predict_false(size > kmem_zmax)) 745 return (malloc_large(size, mtp, DOMAINSET_RR(), flags 746 DEBUG_REDZONE_ARG)); 747 748 vm_domainset_iter_policy_init(&di, ds, &domain, &flags); 749 do { 750 va = malloc_domain(&size, &indx, mtp, domain, flags); 751 } while (va == NULL && vm_domainset_iter_policy(&di, &domain) == 0); 752 malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx); 753 if (__predict_false(va == NULL)) { 754 KASSERT((flags & M_WAITOK) == 0, 755 ("malloc(M_WAITOK) returned NULL")); 756 } 757 #ifdef DEBUG_REDZONE 758 if (va != NULL) 759 va = redzone_setup(va, osize); 760 #endif 761 #ifdef KASAN 762 if (va != NULL) 763 kasan_mark((void *)va, osize, size, KASAN_MALLOC_REDZONE); 764 #endif 765 #ifdef KMSAN 766 if ((flags & M_ZERO) == 0) { 767 kmsan_mark(va, size, KMSAN_STATE_UNINIT); 768 kmsan_orig(va, size, KMSAN_TYPE_MALLOC, KMSAN_RET_ADDR); 769 } 770 #endif 771 return (va); 772 } 773 774 /* 775 * Allocate an executable area. 776 */ 777 void * 778 malloc_exec(size_t size, struct malloc_type *mtp, int flags) 779 { 780 781 return (malloc_domainset_exec(size, mtp, DOMAINSET_RR(), flags)); 782 } 783 784 void * 785 malloc_domainset_exec(size_t size, struct malloc_type *mtp, struct domainset *ds, 786 int flags) 787 { 788 #if defined(DEBUG_REDZONE) || defined(KASAN) 789 unsigned long osize = size; 790 #endif 791 #ifdef MALLOC_DEBUG 792 caddr_t va; 793 #endif 794 795 flags |= M_EXEC; 796 797 #ifdef MALLOC_DEBUG 798 va = NULL; 799 if (malloc_dbg(&va, &size, mtp, flags) != 0) 800 return (va); 801 #endif 802 803 return (malloc_large(size, mtp, ds, flags DEBUG_REDZONE_ARG)); 804 } 805 806 void * 807 malloc_aligned(size_t size, size_t align, struct malloc_type *type, int flags) 808 { 809 return (malloc_domainset_aligned(size, align, type, DOMAINSET_RR(), 810 flags)); 811 } 812 813 void * 814 malloc_domainset_aligned(size_t size, size_t align, 815 struct malloc_type *mtp, struct domainset *ds, int flags) 816 { 817 void *res; 818 size_t asize; 819 820 KASSERT(powerof2(align), 821 ("malloc_domainset_aligned: wrong align %#zx size %#zx", 822 align, size)); 823 KASSERT(align <= PAGE_SIZE, 824 ("malloc_domainset_aligned: align %#zx (size %#zx) too large", 825 align, size)); 826 827 /* 828 * Round the allocation size up to the next power of 2, 829 * because we can only guarantee alignment for 830 * power-of-2-sized allocations. Further increase the 831 * allocation size to align if the rounded size is less than 832 * align, since malloc zones provide alignment equal to their 833 * size. 834 */ 835 if (size == 0) 836 size = 1; 837 asize = size <= align ? align : 1UL << flsl(size - 1); 838 839 res = malloc_domainset(asize, mtp, ds, flags); 840 KASSERT(res == NULL || ((uintptr_t)res & (align - 1)) == 0, 841 ("malloc_domainset_aligned: result not aligned %p size %#zx " 842 "allocsize %#zx align %#zx", res, size, asize, align)); 843 return (res); 844 } 845 846 void * 847 mallocarray(size_t nmemb, size_t size, struct malloc_type *type, int flags) 848 { 849 850 if (WOULD_OVERFLOW(nmemb, size)) 851 panic("mallocarray: %zu * %zu overflowed", nmemb, size); 852 853 return (malloc(size * nmemb, type, flags)); 854 } 855 856 void * 857 mallocarray_domainset(size_t nmemb, size_t size, struct malloc_type *type, 858 struct domainset *ds, int flags) 859 { 860 861 if (WOULD_OVERFLOW(nmemb, size)) 862 panic("mallocarray_domainset: %zu * %zu overflowed", nmemb, size); 863 864 return (malloc_domainset(size * nmemb, type, ds, flags)); 865 } 866 867 #if defined(INVARIANTS) && !defined(KASAN) 868 static void 869 free_save_type(void *addr, struct malloc_type *mtp, u_long size) 870 { 871 struct malloc_type **mtpp = addr; 872 873 /* 874 * Cache a pointer to the malloc_type that most recently freed 875 * this memory here. This way we know who is most likely to 876 * have stepped on it later. 877 * 878 * This code assumes that size is a multiple of 8 bytes for 879 * 64 bit machines 880 */ 881 mtpp = (struct malloc_type **) ((unsigned long)mtpp & ~UMA_ALIGN_PTR); 882 mtpp += (size - sizeof(struct malloc_type *)) / 883 sizeof(struct malloc_type *); 884 *mtpp = mtp; 885 } 886 #endif 887 888 #ifdef MALLOC_DEBUG 889 static int 890 free_dbg(void **addrp, struct malloc_type *mtp) 891 { 892 void *addr; 893 894 addr = *addrp; 895 KASSERT(mtp->ks_version == M_VERSION, ("free: bad malloc type version")); 896 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), 897 ("free: called with spinlock or critical section held")); 898 899 /* free(NULL, ...) does nothing */ 900 if (addr == NULL) 901 return (EJUSTRETURN); 902 903 #ifdef DEBUG_MEMGUARD 904 if (is_memguard_addr(addr)) { 905 memguard_free(addr); 906 return (EJUSTRETURN); 907 } 908 #endif 909 910 #ifdef DEBUG_REDZONE 911 redzone_check(addr); 912 *addrp = redzone_addr_ntor(addr); 913 #endif 914 915 return (0); 916 } 917 #endif 918 919 static __always_inline void 920 _free(void *addr, struct malloc_type *mtp, bool dozero) 921 { 922 uma_zone_t zone; 923 uma_slab_t slab; 924 u_long size; 925 926 #ifdef MALLOC_DEBUG 927 if (free_dbg(&addr, mtp) != 0) 928 return; 929 #endif 930 /* free(NULL, ...) does nothing */ 931 if (addr == NULL) 932 return; 933 934 vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab); 935 if (slab == NULL) 936 panic("%s(%d): address %p(%p) has not been allocated", __func__, 937 dozero, addr, (void *)((uintptr_t)addr & (~UMA_SLAB_MASK))); 938 939 switch (GET_SLAB_COOKIE(slab)) { 940 case __predict_true(SLAB_COOKIE_SLAB_PTR): 941 size = zone->uz_size; 942 #if defined(INVARIANTS) && !defined(KASAN) 943 free_save_type(addr, mtp, size); 944 #endif 945 if (dozero) { 946 kasan_mark(addr, size, size, 0); 947 explicit_bzero(addr, size); 948 } 949 uma_zfree_arg(zone, addr, slab); 950 break; 951 case SLAB_COOKIE_MALLOC_LARGE: 952 size = malloc_large_size(slab); 953 if (dozero) { 954 kasan_mark(addr, size, size, 0); 955 explicit_bzero(addr, size); 956 } 957 free_large(addr, size); 958 break; 959 case SLAB_COOKIE_CONTIG_MALLOC: 960 size = round_page(contigmalloc_size(slab)); 961 if (dozero) 962 explicit_bzero(addr, size); 963 kmem_free(addr, size); 964 break; 965 default: 966 panic("%s(%d): addr %p slab %p with unknown cookie %d", 967 __func__, dozero, addr, slab, GET_SLAB_COOKIE(slab)); 968 /* NOTREACHED */ 969 } 970 malloc_type_freed(mtp, size); 971 } 972 973 /* 974 * free: 975 * Free a block of memory allocated by malloc/contigmalloc. 976 * This routine may not block. 977 */ 978 void 979 free(void *addr, struct malloc_type *mtp) 980 { 981 _free(addr, mtp, false); 982 } 983 984 /* 985 * zfree: 986 * Zero then free a block of memory allocated by malloc/contigmalloc. 987 * This routine may not block. 988 */ 989 void 990 zfree(void *addr, struct malloc_type *mtp) 991 { 992 _free(addr, mtp, true); 993 } 994 995 /* 996 * realloc: change the size of a memory block 997 */ 998 void * 999 realloc(void *addr, size_t size, struct malloc_type *mtp, int flags) 1000 { 1001 #ifndef DEBUG_REDZONE 1002 uma_zone_t zone; 1003 uma_slab_t slab; 1004 #endif 1005 unsigned long alloc; 1006 void *newaddr; 1007 1008 KASSERT(mtp->ks_version == M_VERSION, 1009 ("realloc: bad malloc type version")); 1010 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), 1011 ("realloc: called with spinlock or critical section held")); 1012 1013 /* realloc(NULL, ...) is equivalent to malloc(...) */ 1014 if (addr == NULL) 1015 return (malloc(size, mtp, flags)); 1016 1017 /* 1018 * XXX: Should report free of old memory and alloc of new memory to 1019 * per-CPU stats. 1020 */ 1021 1022 #ifdef DEBUG_MEMGUARD 1023 if (is_memguard_addr(addr)) 1024 return (memguard_realloc(addr, size, mtp, flags)); 1025 #endif 1026 1027 #ifdef DEBUG_REDZONE 1028 alloc = redzone_get_size(addr); 1029 #else 1030 vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab); 1031 1032 /* Sanity check */ 1033 KASSERT(slab != NULL, 1034 ("realloc: address %p out of range", (void *)addr)); 1035 1036 /* Get the size of the original block */ 1037 switch (GET_SLAB_COOKIE(slab)) { 1038 case __predict_true(SLAB_COOKIE_SLAB_PTR): 1039 alloc = zone->uz_size; 1040 break; 1041 case SLAB_COOKIE_MALLOC_LARGE: 1042 alloc = malloc_large_size(slab); 1043 break; 1044 default: 1045 #ifdef INVARIANTS 1046 panic("%s: called for addr %p of unsupported allocation type; " 1047 "slab %p cookie %d", __func__, addr, slab, GET_SLAB_COOKIE(slab)); 1048 #endif 1049 return (NULL); 1050 } 1051 1052 /* Reuse the original block if appropriate */ 1053 if (size <= alloc && 1054 (size > (alloc >> REALLOC_FRACTION) || alloc == MINALLOCSIZE)) { 1055 kasan_mark((void *)addr, size, alloc, KASAN_MALLOC_REDZONE); 1056 return (addr); 1057 } 1058 #endif /* !DEBUG_REDZONE */ 1059 1060 /* Allocate a new, bigger (or smaller) block */ 1061 if ((newaddr = malloc(size, mtp, flags)) == NULL) 1062 return (NULL); 1063 1064 /* 1065 * Copy over original contents. For KASAN, the redzone must be marked 1066 * valid before performing the copy. 1067 */ 1068 kasan_mark(addr, alloc, alloc, 0); 1069 bcopy(addr, newaddr, min(size, alloc)); 1070 free(addr, mtp); 1071 return (newaddr); 1072 } 1073 1074 /* 1075 * reallocf: same as realloc() but free memory on failure. 1076 */ 1077 void * 1078 reallocf(void *addr, size_t size, struct malloc_type *mtp, int flags) 1079 { 1080 void *mem; 1081 1082 if ((mem = realloc(addr, size, mtp, flags)) == NULL) 1083 free(addr, mtp); 1084 return (mem); 1085 } 1086 1087 /* 1088 * malloc_size: returns the number of bytes allocated for a request of the 1089 * specified size 1090 */ 1091 size_t 1092 malloc_size(size_t size) 1093 { 1094 int indx; 1095 1096 if (size > kmem_zmax) 1097 return (round_page(size)); 1098 if (size & KMEM_ZMASK) 1099 size = (size & ~KMEM_ZMASK) + KMEM_ZBASE; 1100 indx = kmemsize[size >> KMEM_ZSHIFT]; 1101 return (kmemzones[indx].kz_size); 1102 } 1103 1104 /* 1105 * malloc_usable_size: returns the usable size of the allocation. 1106 */ 1107 size_t 1108 malloc_usable_size(const void *addr) 1109 { 1110 #ifndef DEBUG_REDZONE 1111 uma_zone_t zone; 1112 uma_slab_t slab; 1113 #endif 1114 u_long size; 1115 1116 if (addr == NULL) 1117 return (0); 1118 1119 #ifdef DEBUG_MEMGUARD 1120 if (is_memguard_addr(__DECONST(void *, addr))) 1121 return (memguard_get_req_size(addr)); 1122 #endif 1123 1124 #ifdef DEBUG_REDZONE 1125 size = redzone_get_size(__DECONST(void *, addr)); 1126 #else 1127 vtozoneslab((vm_offset_t)addr & (~UMA_SLAB_MASK), &zone, &slab); 1128 if (slab == NULL) 1129 panic("malloc_usable_size: address %p(%p) is not allocated", 1130 addr, (void *)((u_long)addr & (~UMA_SLAB_MASK))); 1131 1132 switch (GET_SLAB_COOKIE(slab)) { 1133 case __predict_true(SLAB_COOKIE_SLAB_PTR): 1134 size = zone->uz_size; 1135 break; 1136 case SLAB_COOKIE_MALLOC_LARGE: 1137 size = malloc_large_size(slab); 1138 break; 1139 default: 1140 __assert_unreachable(); 1141 size = 0; 1142 break; 1143 } 1144 #endif 1145 1146 /* 1147 * Unmark the redzone to avoid reports from consumers who are 1148 * (presumably) about to use the full allocation size. 1149 */ 1150 kasan_mark(addr, size, size, 0); 1151 1152 return (size); 1153 } 1154 1155 CTASSERT(VM_KMEM_SIZE_SCALE >= 1); 1156 1157 /* 1158 * Initialize the kernel memory (kmem) arena. 1159 */ 1160 void 1161 kmeminit(void) 1162 { 1163 u_long mem_size; 1164 u_long tmp; 1165 1166 #ifdef VM_KMEM_SIZE 1167 if (vm_kmem_size == 0) 1168 vm_kmem_size = VM_KMEM_SIZE; 1169 #endif 1170 #ifdef VM_KMEM_SIZE_MIN 1171 if (vm_kmem_size_min == 0) 1172 vm_kmem_size_min = VM_KMEM_SIZE_MIN; 1173 #endif 1174 #ifdef VM_KMEM_SIZE_MAX 1175 if (vm_kmem_size_max == 0) 1176 vm_kmem_size_max = VM_KMEM_SIZE_MAX; 1177 #endif 1178 /* 1179 * Calculate the amount of kernel virtual address (KVA) space that is 1180 * preallocated to the kmem arena. In order to support a wide range 1181 * of machines, it is a function of the physical memory size, 1182 * specifically, 1183 * 1184 * min(max(physical memory size / VM_KMEM_SIZE_SCALE, 1185 * VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX) 1186 * 1187 * Every architecture must define an integral value for 1188 * VM_KMEM_SIZE_SCALE. However, the definitions of VM_KMEM_SIZE_MIN 1189 * and VM_KMEM_SIZE_MAX, which represent respectively the floor and 1190 * ceiling on this preallocation, are optional. Typically, 1191 * VM_KMEM_SIZE_MAX is itself a function of the available KVA space on 1192 * a given architecture. 1193 */ 1194 mem_size = vm_cnt.v_page_count; 1195 if (mem_size <= 32768) /* delphij XXX 128MB */ 1196 kmem_zmax = PAGE_SIZE; 1197 1198 if (vm_kmem_size_scale < 1) 1199 vm_kmem_size_scale = VM_KMEM_SIZE_SCALE; 1200 1201 /* 1202 * Check if we should use defaults for the "vm_kmem_size" 1203 * variable: 1204 */ 1205 if (vm_kmem_size == 0) { 1206 vm_kmem_size = mem_size / vm_kmem_size_scale; 1207 vm_kmem_size = vm_kmem_size * PAGE_SIZE < vm_kmem_size ? 1208 vm_kmem_size_max : vm_kmem_size * PAGE_SIZE; 1209 if (vm_kmem_size_min > 0 && vm_kmem_size < vm_kmem_size_min) 1210 vm_kmem_size = vm_kmem_size_min; 1211 if (vm_kmem_size_max > 0 && vm_kmem_size >= vm_kmem_size_max) 1212 vm_kmem_size = vm_kmem_size_max; 1213 } 1214 if (vm_kmem_size == 0) 1215 panic("Tune VM_KMEM_SIZE_* for the platform"); 1216 1217 /* 1218 * The amount of KVA space that is preallocated to the 1219 * kmem arena can be set statically at compile-time or manually 1220 * through the kernel environment. However, it is still limited to 1221 * twice the physical memory size, which has been sufficient to handle 1222 * the most severe cases of external fragmentation in the kmem arena. 1223 */ 1224 if (vm_kmem_size / 2 / PAGE_SIZE > mem_size) 1225 vm_kmem_size = 2 * mem_size * PAGE_SIZE; 1226 1227 vm_kmem_size = round_page(vm_kmem_size); 1228 1229 /* 1230 * With KASAN or KMSAN enabled, dynamically allocated kernel memory is 1231 * shadowed. Account for this when setting the UMA limit. 1232 */ 1233 #if defined(KASAN) 1234 vm_kmem_size = (vm_kmem_size * KASAN_SHADOW_SCALE) / 1235 (KASAN_SHADOW_SCALE + 1); 1236 #elif defined(KMSAN) 1237 vm_kmem_size /= 3; 1238 #endif 1239 1240 #ifdef DEBUG_MEMGUARD 1241 tmp = memguard_fudge(vm_kmem_size, kernel_map); 1242 #else 1243 tmp = vm_kmem_size; 1244 #endif 1245 uma_set_limit(tmp); 1246 1247 #ifdef DEBUG_MEMGUARD 1248 /* 1249 * Initialize MemGuard if support compiled in. MemGuard is a 1250 * replacement allocator used for detecting tamper-after-free 1251 * scenarios as they occur. It is only used for debugging. 1252 */ 1253 memguard_init(kernel_arena); 1254 #endif 1255 } 1256 1257 /* 1258 * Initialize the kernel memory allocator 1259 */ 1260 /* ARGSUSED*/ 1261 static void 1262 mallocinit(void *dummy) 1263 { 1264 int i; 1265 uint8_t indx; 1266 1267 mtx_init(&malloc_mtx, "malloc", NULL, MTX_DEF); 1268 1269 kmeminit(); 1270 1271 if (kmem_zmax < PAGE_SIZE || kmem_zmax > KMEM_ZMAX) 1272 kmem_zmax = KMEM_ZMAX; 1273 1274 for (i = 0, indx = 0; kmemzones[indx].kz_size != 0; indx++) { 1275 int size = kmemzones[indx].kz_size; 1276 const char *name = kmemzones[indx].kz_name; 1277 size_t align; 1278 int subzone; 1279 1280 align = UMA_ALIGN_PTR; 1281 if (powerof2(size) && size > sizeof(void *)) 1282 align = MIN(size, PAGE_SIZE) - 1; 1283 for (subzone = 0; subzone < numzones; subzone++) { 1284 kmemzones[indx].kz_zone[subzone] = 1285 uma_zcreate(name, size, 1286 #if defined(INVARIANTS) && !defined(KASAN) && !defined(KMSAN) 1287 mtrash_ctor, mtrash_dtor, mtrash_init, mtrash_fini, 1288 #else 1289 NULL, NULL, NULL, NULL, 1290 #endif 1291 align, UMA_ZONE_MALLOC); 1292 } 1293 for (;i <= size; i+= KMEM_ZBASE) 1294 kmemsize[i >> KMEM_ZSHIFT] = indx; 1295 } 1296 } 1297 SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_SECOND, mallocinit, NULL); 1298 1299 void 1300 malloc_init(void *data) 1301 { 1302 struct malloc_type_internal *mtip; 1303 struct malloc_type *mtp; 1304 1305 KASSERT(vm_cnt.v_page_count != 0, 1306 ("malloc_init() called before vm_mem_init()")); 1307 1308 mtp = data; 1309 if (mtp->ks_version != M_VERSION) 1310 panic("malloc_init: type %s with unsupported version %lu", 1311 mtp->ks_shortdesc, mtp->ks_version); 1312 1313 mtip = &mtp->ks_mti; 1314 mtip->mti_stats = uma_zalloc_pcpu(pcpu_zone_64, M_WAITOK | M_ZERO); 1315 mtp_set_subzone(mtp); 1316 1317 mtx_lock(&malloc_mtx); 1318 mtp->ks_next = kmemstatistics; 1319 kmemstatistics = mtp; 1320 kmemcount++; 1321 mtx_unlock(&malloc_mtx); 1322 } 1323 1324 void 1325 malloc_uninit(void *data) 1326 { 1327 struct malloc_type_internal *mtip; 1328 struct malloc_type_stats *mtsp; 1329 struct malloc_type *mtp, *temp; 1330 long temp_allocs, temp_bytes; 1331 int i; 1332 1333 mtp = data; 1334 KASSERT(mtp->ks_version == M_VERSION, 1335 ("malloc_uninit: bad malloc type version")); 1336 1337 mtx_lock(&malloc_mtx); 1338 mtip = &mtp->ks_mti; 1339 if (mtp != kmemstatistics) { 1340 for (temp = kmemstatistics; temp != NULL; 1341 temp = temp->ks_next) { 1342 if (temp->ks_next == mtp) { 1343 temp->ks_next = mtp->ks_next; 1344 break; 1345 } 1346 } 1347 KASSERT(temp, 1348 ("malloc_uninit: type '%s' not found", mtp->ks_shortdesc)); 1349 } else 1350 kmemstatistics = mtp->ks_next; 1351 kmemcount--; 1352 mtx_unlock(&malloc_mtx); 1353 1354 /* 1355 * Look for memory leaks. 1356 */ 1357 temp_allocs = temp_bytes = 0; 1358 for (i = 0; i <= mp_maxid; i++) { 1359 mtsp = zpcpu_get_cpu(mtip->mti_stats, i); 1360 temp_allocs += mtsp->mts_numallocs; 1361 temp_allocs -= mtsp->mts_numfrees; 1362 temp_bytes += mtsp->mts_memalloced; 1363 temp_bytes -= mtsp->mts_memfreed; 1364 } 1365 if (temp_allocs > 0 || temp_bytes > 0) { 1366 printf("Warning: memory type %s leaked memory on destroy " 1367 "(%ld allocations, %ld bytes leaked).\n", mtp->ks_shortdesc, 1368 temp_allocs, temp_bytes); 1369 } 1370 1371 uma_zfree_pcpu(pcpu_zone_64, mtip->mti_stats); 1372 } 1373 1374 struct malloc_type * 1375 malloc_desc2type(const char *desc) 1376 { 1377 struct malloc_type *mtp; 1378 1379 mtx_assert(&malloc_mtx, MA_OWNED); 1380 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1381 if (strcmp(mtp->ks_shortdesc, desc) == 0) 1382 return (mtp); 1383 } 1384 return (NULL); 1385 } 1386 1387 static int 1388 sysctl_kern_malloc_stats(SYSCTL_HANDLER_ARGS) 1389 { 1390 struct malloc_type_stream_header mtsh; 1391 struct malloc_type_internal *mtip; 1392 struct malloc_type_stats *mtsp, zeromts; 1393 struct malloc_type_header mth; 1394 struct malloc_type *mtp; 1395 int error, i; 1396 struct sbuf sbuf; 1397 1398 error = sysctl_wire_old_buffer(req, 0); 1399 if (error != 0) 1400 return (error); 1401 sbuf_new_for_sysctl(&sbuf, NULL, 128, req); 1402 sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL); 1403 mtx_lock(&malloc_mtx); 1404 1405 bzero(&zeromts, sizeof(zeromts)); 1406 1407 /* 1408 * Insert stream header. 1409 */ 1410 bzero(&mtsh, sizeof(mtsh)); 1411 mtsh.mtsh_version = MALLOC_TYPE_STREAM_VERSION; 1412 mtsh.mtsh_maxcpus = MAXCPU; 1413 mtsh.mtsh_count = kmemcount; 1414 (void)sbuf_bcat(&sbuf, &mtsh, sizeof(mtsh)); 1415 1416 /* 1417 * Insert alternating sequence of type headers and type statistics. 1418 */ 1419 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1420 mtip = &mtp->ks_mti; 1421 1422 /* 1423 * Insert type header. 1424 */ 1425 bzero(&mth, sizeof(mth)); 1426 strlcpy(mth.mth_name, mtp->ks_shortdesc, MALLOC_MAX_NAME); 1427 (void)sbuf_bcat(&sbuf, &mth, sizeof(mth)); 1428 1429 /* 1430 * Insert type statistics for each CPU. 1431 */ 1432 for (i = 0; i <= mp_maxid; i++) { 1433 mtsp = zpcpu_get_cpu(mtip->mti_stats, i); 1434 (void)sbuf_bcat(&sbuf, mtsp, sizeof(*mtsp)); 1435 } 1436 /* 1437 * Fill in the missing CPUs. 1438 */ 1439 for (; i < MAXCPU; i++) { 1440 (void)sbuf_bcat(&sbuf, &zeromts, sizeof(zeromts)); 1441 } 1442 } 1443 mtx_unlock(&malloc_mtx); 1444 error = sbuf_finish(&sbuf); 1445 sbuf_delete(&sbuf); 1446 return (error); 1447 } 1448 1449 SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, 1450 CTLFLAG_RD | CTLTYPE_STRUCT | CTLFLAG_MPSAFE, 0, 0, 1451 sysctl_kern_malloc_stats, "s,malloc_type_ustats", 1452 "Return malloc types"); 1453 1454 SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0, 1455 "Count of kernel malloc types"); 1456 1457 void 1458 malloc_type_list(malloc_type_list_func_t *func, void *arg) 1459 { 1460 struct malloc_type *mtp, **bufmtp; 1461 int count, i; 1462 size_t buflen; 1463 1464 mtx_lock(&malloc_mtx); 1465 restart: 1466 mtx_assert(&malloc_mtx, MA_OWNED); 1467 count = kmemcount; 1468 mtx_unlock(&malloc_mtx); 1469 1470 buflen = sizeof(struct malloc_type *) * count; 1471 bufmtp = malloc(buflen, M_TEMP, M_WAITOK); 1472 1473 mtx_lock(&malloc_mtx); 1474 1475 if (count < kmemcount) { 1476 free(bufmtp, M_TEMP); 1477 goto restart; 1478 } 1479 1480 for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++) 1481 bufmtp[i] = mtp; 1482 1483 mtx_unlock(&malloc_mtx); 1484 1485 for (i = 0; i < count; i++) 1486 (func)(bufmtp[i], arg); 1487 1488 free(bufmtp, M_TEMP); 1489 } 1490 1491 #ifdef DDB 1492 static int64_t 1493 get_malloc_stats(const struct malloc_type_internal *mtip, uint64_t *allocs, 1494 uint64_t *inuse) 1495 { 1496 const struct malloc_type_stats *mtsp; 1497 uint64_t frees, alloced, freed; 1498 int i; 1499 1500 *allocs = 0; 1501 frees = 0; 1502 alloced = 0; 1503 freed = 0; 1504 for (i = 0; i <= mp_maxid; i++) { 1505 mtsp = zpcpu_get_cpu(mtip->mti_stats, i); 1506 1507 *allocs += mtsp->mts_numallocs; 1508 frees += mtsp->mts_numfrees; 1509 alloced += mtsp->mts_memalloced; 1510 freed += mtsp->mts_memfreed; 1511 } 1512 *inuse = *allocs - frees; 1513 return (alloced - freed); 1514 } 1515 1516 DB_SHOW_COMMAND_FLAGS(malloc, db_show_malloc, DB_CMD_MEMSAFE) 1517 { 1518 const char *fmt_hdr, *fmt_entry; 1519 struct malloc_type *mtp; 1520 uint64_t allocs, inuse; 1521 int64_t size; 1522 /* variables for sorting */ 1523 struct malloc_type *last_mtype, *cur_mtype; 1524 int64_t cur_size, last_size; 1525 int ties; 1526 1527 if (modif[0] == 'i') { 1528 fmt_hdr = "%s,%s,%s,%s\n"; 1529 fmt_entry = "\"%s\",%ju,%jdK,%ju\n"; 1530 } else { 1531 fmt_hdr = "%18s %12s %12s %12s\n"; 1532 fmt_entry = "%18s %12ju %12jdK %12ju\n"; 1533 } 1534 1535 db_printf(fmt_hdr, "Type", "InUse", "MemUse", "Requests"); 1536 1537 /* Select sort, largest size first. */ 1538 last_mtype = NULL; 1539 last_size = INT64_MAX; 1540 for (;;) { 1541 cur_mtype = NULL; 1542 cur_size = -1; 1543 ties = 0; 1544 1545 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1546 /* 1547 * In the case of size ties, print out mtypes 1548 * in the order they are encountered. That is, 1549 * when we encounter the most recently output 1550 * mtype, we have already printed all preceding 1551 * ties, and we must print all following ties. 1552 */ 1553 if (mtp == last_mtype) { 1554 ties = 1; 1555 continue; 1556 } 1557 size = get_malloc_stats(&mtp->ks_mti, &allocs, 1558 &inuse); 1559 if (size > cur_size && size < last_size + ties) { 1560 cur_size = size; 1561 cur_mtype = mtp; 1562 } 1563 } 1564 if (cur_mtype == NULL) 1565 break; 1566 1567 size = get_malloc_stats(&cur_mtype->ks_mti, &allocs, &inuse); 1568 db_printf(fmt_entry, cur_mtype->ks_shortdesc, inuse, 1569 howmany(size, 1024), allocs); 1570 1571 if (db_pager_quit) 1572 break; 1573 1574 last_mtype = cur_mtype; 1575 last_size = cur_size; 1576 } 1577 } 1578 1579 #if MALLOC_DEBUG_MAXZONES > 1 1580 DB_SHOW_COMMAND(multizone_matches, db_show_multizone_matches) 1581 { 1582 struct malloc_type_internal *mtip; 1583 struct malloc_type *mtp; 1584 u_int subzone; 1585 1586 if (!have_addr) { 1587 db_printf("Usage: show multizone_matches <malloc type/addr>\n"); 1588 return; 1589 } 1590 mtp = (void *)addr; 1591 if (mtp->ks_version != M_VERSION) { 1592 db_printf("Version %lx does not match expected %x\n", 1593 mtp->ks_version, M_VERSION); 1594 return; 1595 } 1596 1597 mtip = &mtp->ks_mti; 1598 subzone = mtip->mti_zone; 1599 1600 for (mtp = kmemstatistics; mtp != NULL; mtp = mtp->ks_next) { 1601 mtip = &mtp->ks_mti; 1602 if (mtip->mti_zone != subzone) 1603 continue; 1604 db_printf("%s\n", mtp->ks_shortdesc); 1605 if (db_pager_quit) 1606 break; 1607 } 1608 } 1609 #endif /* MALLOC_DEBUG_MAXZONES > 1 */ 1610 #endif /* DDB */ 1611