1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2005, Bosko Milekic <bmilekic@FreeBSD.org>. 5 * Copyright (c) 2010 Isilon Systems, Inc. (http://www.isilon.com/) 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice unmodified, this list of conditions, and the following 13 * disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 19 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 20 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 21 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 22 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 23 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 27 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 28 */ 29 30 #include <sys/cdefs.h> 31 __FBSDID("$FreeBSD$"); 32 33 /* 34 * MemGuard is a simple replacement allocator for debugging only 35 * which provides ElectricFence-style memory barrier protection on 36 * objects being allocated, and is used to detect tampering-after-free 37 * scenarios. 38 * 39 * See the memguard(9) man page for more information on using MemGuard. 40 */ 41 42 #include "opt_vm.h" 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/kernel.h> 47 #include <sys/types.h> 48 #include <sys/queue.h> 49 #include <sys/lock.h> 50 #include <sys/mutex.h> 51 #include <sys/malloc.h> 52 #include <sys/sysctl.h> 53 #include <sys/vmem.h> 54 #include <sys/vmmeter.h> 55 56 #include <vm/vm.h> 57 #include <vm/uma.h> 58 #include <vm/vm_param.h> 59 #include <vm/vm_page.h> 60 #include <vm/vm_map.h> 61 #include <vm/vm_object.h> 62 #include <vm/vm_kern.h> 63 #include <vm/vm_extern.h> 64 #include <vm/uma_int.h> 65 #include <vm/memguard.h> 66 67 static SYSCTL_NODE(_vm, OID_AUTO, memguard, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 68 "MemGuard data"); 69 /* 70 * The vm_memguard_divisor variable controls how much of kernel_arena should be 71 * reserved for MemGuard. 72 */ 73 static u_int vm_memguard_divisor; 74 SYSCTL_UINT(_vm_memguard, OID_AUTO, divisor, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 75 &vm_memguard_divisor, 76 0, "(kmem_size/memguard_divisor) == memguard submap size"); 77 78 /* 79 * Short description (ks_shortdesc) of memory type to monitor. 80 */ 81 static char vm_memguard_desc[128] = ""; 82 static struct malloc_type *vm_memguard_mtype = NULL; 83 TUNABLE_STR("vm.memguard.desc", vm_memguard_desc, sizeof(vm_memguard_desc)); 84 static int 85 memguard_sysctl_desc(SYSCTL_HANDLER_ARGS) 86 { 87 char desc[sizeof(vm_memguard_desc)]; 88 int error; 89 90 strlcpy(desc, vm_memguard_desc, sizeof(desc)); 91 error = sysctl_handle_string(oidp, desc, sizeof(desc), req); 92 if (error != 0 || req->newptr == NULL) 93 return (error); 94 95 mtx_lock(&malloc_mtx); 96 /* If mtp is NULL, it will be initialized in memguard_cmp() */ 97 vm_memguard_mtype = malloc_desc2type(desc); 98 strlcpy(vm_memguard_desc, desc, sizeof(vm_memguard_desc)); 99 mtx_unlock(&malloc_mtx); 100 return (error); 101 } 102 SYSCTL_PROC(_vm_memguard, OID_AUTO, desc, 103 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0, 104 memguard_sysctl_desc, "A", "Short description of memory type to monitor"); 105 106 static int 107 memguard_sysctl_mapused(SYSCTL_HANDLER_ARGS) 108 { 109 vmem_size_t size; 110 111 size = vmem_size(memguard_arena, VMEM_ALLOC); 112 return (sysctl_handle_long(oidp, &size, sizeof(size), req)); 113 } 114 115 static vm_offset_t memguard_base; 116 static vm_size_t memguard_mapsize; 117 static vm_size_t memguard_physlimit; 118 static u_long memguard_wasted; 119 static u_long memguard_succ; 120 static u_long memguard_fail_kva; 121 static u_long memguard_fail_pgs; 122 123 SYSCTL_ULONG(_vm_memguard, OID_AUTO, mapsize, CTLFLAG_RD, 124 &memguard_mapsize, 0, "MemGuard private arena size"); 125 SYSCTL_ULONG(_vm_memguard, OID_AUTO, phys_limit, CTLFLAG_RD, 126 &memguard_physlimit, 0, "Limit on MemGuard memory consumption"); 127 SYSCTL_ULONG(_vm_memguard, OID_AUTO, wasted, CTLFLAG_RD, 128 &memguard_wasted, 0, "Excess memory used through page promotion"); 129 SYSCTL_ULONG(_vm_memguard, OID_AUTO, numalloc, CTLFLAG_RD, 130 &memguard_succ, 0, "Count of successful MemGuard allocations"); 131 SYSCTL_ULONG(_vm_memguard, OID_AUTO, fail_kva, CTLFLAG_RD, 132 &memguard_fail_kva, 0, "MemGuard failures due to lack of KVA"); 133 SYSCTL_ULONG(_vm_memguard, OID_AUTO, fail_pgs, CTLFLAG_RD, 134 &memguard_fail_pgs, 0, "MemGuard failures due to lack of pages"); 135 136 #define MG_GUARD_AROUND 0x001 137 #define MG_GUARD_ALLLARGE 0x002 138 #define MG_GUARD_NOFREE 0x004 139 static int memguard_options = MG_GUARD_AROUND; 140 SYSCTL_INT(_vm_memguard, OID_AUTO, options, CTLFLAG_RWTUN, 141 &memguard_options, 0, 142 "MemGuard options:\n" 143 "\t0x001 - add guard pages around each allocation\n" 144 "\t0x002 - always use MemGuard for allocations over a page\n" 145 "\t0x004 - guard uma(9) zones with UMA_ZONE_NOFREE flag"); 146 147 static u_int memguard_minsize; 148 static u_long memguard_minsize_reject; 149 SYSCTL_UINT(_vm_memguard, OID_AUTO, minsize, CTLFLAG_RW, 150 &memguard_minsize, 0, "Minimum size for page promotion"); 151 SYSCTL_ULONG(_vm_memguard, OID_AUTO, minsize_reject, CTLFLAG_RD, 152 &memguard_minsize_reject, 0, "# times rejected for size"); 153 154 static u_int memguard_frequency; 155 static u_long memguard_frequency_hits; 156 SYSCTL_UINT(_vm_memguard, OID_AUTO, frequency, CTLFLAG_RWTUN, 157 &memguard_frequency, 0, "Times in 100000 that MemGuard will randomly run"); 158 SYSCTL_ULONG(_vm_memguard, OID_AUTO, frequency_hits, CTLFLAG_RD, 159 &memguard_frequency_hits, 0, "# times MemGuard randomly chose"); 160 161 162 /* 163 * Return a fudged value to be used for vm_kmem_size for allocating 164 * the kernel_arena. 165 */ 166 unsigned long 167 memguard_fudge(unsigned long km_size, const struct vm_map *parent_map) 168 { 169 u_long mem_pgs, parent_size; 170 171 vm_memguard_divisor = 10; 172 /* CTFLAG_RDTUN doesn't work during the early boot process. */ 173 TUNABLE_INT_FETCH("vm.memguard.divisor", &vm_memguard_divisor); 174 175 parent_size = vm_map_max(parent_map) - vm_map_min(parent_map) + 176 PAGE_SIZE; 177 /* Pick a conservative value if provided value sucks. */ 178 if ((vm_memguard_divisor <= 0) || 179 ((parent_size / vm_memguard_divisor) == 0)) 180 vm_memguard_divisor = 10; 181 /* 182 * Limit consumption of physical pages to 183 * 1/vm_memguard_divisor of system memory. If the KVA is 184 * smaller than this then the KVA limit comes into play first. 185 * This prevents memguard's page promotions from completely 186 * using up memory, since most malloc(9) calls are sub-page. 187 */ 188 mem_pgs = vm_cnt.v_page_count; 189 memguard_physlimit = (mem_pgs / vm_memguard_divisor) * PAGE_SIZE; 190 /* 191 * We want as much KVA as we can take safely. Use at most our 192 * allotted fraction of the parent map's size. Limit this to 193 * twice the physical memory to avoid using too much memory as 194 * pagetable pages (size must be multiple of PAGE_SIZE). 195 */ 196 memguard_mapsize = round_page(parent_size / vm_memguard_divisor); 197 if (memguard_mapsize / (2 * PAGE_SIZE) > mem_pgs) 198 memguard_mapsize = mem_pgs * 2 * PAGE_SIZE; 199 if (km_size + memguard_mapsize > parent_size) 200 memguard_mapsize = 0; 201 return (km_size + memguard_mapsize); 202 } 203 204 /* 205 * Initialize the MemGuard mock allocator. All objects from MemGuard come 206 * out of a single contiguous chunk of kernel address space that is managed 207 * by a vmem arena. 208 */ 209 void 210 memguard_init(vmem_t *parent) 211 { 212 vm_offset_t base; 213 214 vmem_alloc(parent, memguard_mapsize, M_BESTFIT | M_WAITOK, &base); 215 vmem_init(memguard_arena, "memguard arena", base, memguard_mapsize, 216 PAGE_SIZE, 0, M_WAITOK); 217 memguard_base = base; 218 219 printf("MEMGUARD DEBUGGING ALLOCATOR INITIALIZED:\n"); 220 printf("\tMEMGUARD map base: 0x%lx\n", (u_long)base); 221 printf("\tMEMGUARD map size: %jd KBytes\n", 222 (uintmax_t)memguard_mapsize >> 10); 223 } 224 225 /* 226 * Run things that can't be done as early as memguard_init(). 227 */ 228 static void 229 memguard_sysinit(void) 230 { 231 struct sysctl_oid_list *parent; 232 233 parent = SYSCTL_STATIC_CHILDREN(_vm_memguard); 234 SYSCTL_ADD_UAUTO(NULL, parent, OID_AUTO, "mapstart", 235 CTLFLAG_RD, &memguard_base, 236 "MemGuard KVA base"); 237 SYSCTL_ADD_UAUTO(NULL, parent, OID_AUTO, "maplimit", 238 CTLFLAG_RD, &memguard_mapsize, 239 "MemGuard KVA size"); 240 SYSCTL_ADD_PROC(NULL, parent, OID_AUTO, "mapused", 241 CTLFLAG_RD | CTLFLAG_MPSAFE | CTLTYPE_ULONG, NULL, 0, memguard_sysctl_mapused, "LU", 242 "MemGuard KVA used"); 243 } 244 SYSINIT(memguard, SI_SUB_KLD, SI_ORDER_ANY, memguard_sysinit, NULL); 245 246 /* 247 * v2sizep() converts a virtual address of the first page allocated for 248 * an item to a pointer to u_long recording the size of the original 249 * allocation request. 250 * 251 * This routine is very similar to those defined by UMA in uma_int.h. 252 * The difference is that this routine stores the originally allocated 253 * size in one of the page's fields that is unused when the page is 254 * wired rather than the object field, which is used. 255 */ 256 static u_long * 257 v2sizep(vm_offset_t va) 258 { 259 vm_paddr_t pa; 260 struct vm_page *p; 261 262 pa = pmap_kextract(va); 263 if (pa == 0) 264 panic("MemGuard detected double-free of %p", (void *)va); 265 p = PHYS_TO_VM_PAGE(pa); 266 KASSERT(vm_page_wired(p) && p->a.queue == PQ_NONE, 267 ("MEMGUARD: Expected wired page %p in vtomgfifo!", p)); 268 return (&p->plinks.memguard.p); 269 } 270 271 static u_long * 272 v2sizev(vm_offset_t va) 273 { 274 vm_paddr_t pa; 275 struct vm_page *p; 276 277 pa = pmap_kextract(va); 278 if (pa == 0) 279 panic("MemGuard detected double-free of %p", (void *)va); 280 p = PHYS_TO_VM_PAGE(pa); 281 KASSERT(vm_page_wired(p) && p->a.queue == PQ_NONE, 282 ("MEMGUARD: Expected wired page %p in vtomgfifo!", p)); 283 return (&p->plinks.memguard.v); 284 } 285 286 /* 287 * Allocate a single object of specified size with specified flags 288 * (either M_WAITOK or M_NOWAIT). 289 */ 290 void * 291 memguard_alloc(unsigned long req_size, int flags) 292 { 293 vm_offset_t addr, origaddr; 294 u_long size_p, size_v; 295 int do_guard, error, rv; 296 297 size_p = round_page(req_size); 298 if (size_p == 0) 299 return (NULL); 300 301 /* 302 * To ensure there are holes on both sides of the allocation, 303 * request 2 extra pages of KVA. Save the value of memguard_options 304 * so that we use a consistent value throughout this function. 305 */ 306 size_v = size_p; 307 do_guard = (memguard_options & MG_GUARD_AROUND) != 0; 308 if (do_guard) 309 size_v += 2 * PAGE_SIZE; 310 311 /* 312 * When we pass our memory limit, reject sub-page allocations. 313 * Page-size and larger allocations will use the same amount 314 * of physical memory whether we allocate or hand off to 315 * malloc_large(), so keep those. 316 */ 317 if (vmem_size(memguard_arena, VMEM_ALLOC) >= memguard_physlimit && 318 req_size < PAGE_SIZE) { 319 addr = (vm_offset_t)NULL; 320 memguard_fail_pgs++; 321 goto out; 322 } 323 324 /* 325 * Attempt to avoid address reuse for as long as possible, to increase 326 * the likelihood of catching a use-after-free. 327 */ 328 error = vmem_alloc(memguard_arena, size_v, M_NEXTFIT | M_NOWAIT, 329 &origaddr); 330 if (error != 0) { 331 memguard_fail_kva++; 332 addr = (vm_offset_t)NULL; 333 goto out; 334 } 335 addr = origaddr; 336 if (do_guard) 337 addr += PAGE_SIZE; 338 rv = kmem_back(kernel_object, addr, size_p, flags); 339 if (rv != KERN_SUCCESS) { 340 vmem_xfree(memguard_arena, origaddr, size_v); 341 memguard_fail_pgs++; 342 addr = (vm_offset_t)NULL; 343 goto out; 344 } 345 *v2sizep(trunc_page(addr)) = req_size; 346 *v2sizev(trunc_page(addr)) = size_v; 347 memguard_succ++; 348 if (req_size < PAGE_SIZE) { 349 memguard_wasted += (PAGE_SIZE - req_size); 350 if (do_guard) { 351 /* 352 * Align the request to 16 bytes, and return 353 * an address near the end of the page, to 354 * better detect array overrun. 355 */ 356 req_size = roundup2(req_size, 16); 357 addr += (PAGE_SIZE - req_size); 358 } 359 } 360 out: 361 return ((void *)addr); 362 } 363 364 int 365 is_memguard_addr(void *addr) 366 { 367 vm_offset_t a = (vm_offset_t)(uintptr_t)addr; 368 369 return (a >= memguard_base && a < memguard_base + memguard_mapsize); 370 } 371 372 /* 373 * Free specified single object. 374 */ 375 void 376 memguard_free(void *ptr) 377 { 378 vm_offset_t addr; 379 u_long req_size, size, sizev; 380 char *temp; 381 int i; 382 383 addr = trunc_page((uintptr_t)ptr); 384 req_size = *v2sizep(addr); 385 sizev = *v2sizev(addr); 386 size = round_page(req_size); 387 388 /* 389 * Page should not be guarded right now, so force a write. 390 * The purpose of this is to increase the likelihood of 391 * catching a double-free, but not necessarily a 392 * tamper-after-free (the second thread freeing might not 393 * write before freeing, so this forces it to and, 394 * subsequently, trigger a fault). 395 */ 396 temp = ptr; 397 for (i = 0; i < size; i += PAGE_SIZE) 398 temp[i] = 'M'; 399 400 /* 401 * This requires carnal knowledge of the implementation of 402 * kmem_free(), but since we've already replaced kmem_malloc() 403 * above, it's not really any worse. We want to use the 404 * vm_map lock to serialize updates to memguard_wasted, since 405 * we had the lock at increment. 406 */ 407 kmem_unback(kernel_object, addr, size); 408 if (sizev > size) 409 addr -= PAGE_SIZE; 410 vmem_xfree(memguard_arena, addr, sizev); 411 if (req_size < PAGE_SIZE) 412 memguard_wasted -= (PAGE_SIZE - req_size); 413 } 414 415 /* 416 * Re-allocate an allocation that was originally guarded. 417 */ 418 void * 419 memguard_realloc(void *addr, unsigned long size, struct malloc_type *mtp, 420 int flags) 421 { 422 void *newaddr; 423 u_long old_size; 424 425 /* 426 * Allocate the new block. Force the allocation to be guarded 427 * as the original may have been guarded through random 428 * chance, and that should be preserved. 429 */ 430 if ((newaddr = memguard_alloc(size, flags)) == NULL) 431 return (NULL); 432 433 /* Copy over original contents. */ 434 old_size = *v2sizep(trunc_page((uintptr_t)addr)); 435 bcopy(addr, newaddr, min(size, old_size)); 436 memguard_free(addr); 437 return (newaddr); 438 } 439 440 static int 441 memguard_cmp(unsigned long size) 442 { 443 444 if (size < memguard_minsize) { 445 memguard_minsize_reject++; 446 return (0); 447 } 448 if ((memguard_options & MG_GUARD_ALLLARGE) != 0 && size >= PAGE_SIZE) 449 return (1); 450 if (memguard_frequency > 0 && 451 (random() % 100000) < memguard_frequency) { 452 memguard_frequency_hits++; 453 return (1); 454 } 455 456 return (0); 457 } 458 459 int 460 memguard_cmp_mtp(struct malloc_type *mtp, unsigned long size) 461 { 462 463 if (memguard_cmp(size)) 464 return(1); 465 466 #if 1 467 /* 468 * The safest way of comparsion is to always compare short description 469 * string of memory type, but it is also the slowest way. 470 */ 471 return (strcmp(mtp->ks_shortdesc, vm_memguard_desc) == 0); 472 #else 473 /* 474 * If we compare pointers, there are two possible problems: 475 * 1. Memory type was unloaded and new memory type was allocated at the 476 * same address. 477 * 2. Memory type was unloaded and loaded again, but allocated at a 478 * different address. 479 */ 480 if (vm_memguard_mtype != NULL) 481 return (mtp == vm_memguard_mtype); 482 if (strcmp(mtp->ks_shortdesc, vm_memguard_desc) == 0) { 483 vm_memguard_mtype = mtp; 484 return (1); 485 } 486 return (0); 487 #endif 488 } 489 490 int 491 memguard_cmp_zone(uma_zone_t zone) 492 { 493 494 if ((memguard_options & MG_GUARD_NOFREE) == 0 && 495 zone->uz_flags & UMA_ZONE_NOFREE) 496 return (0); 497 498 if (memguard_cmp(zone->uz_size)) 499 return (1); 500 501 /* 502 * The safest way of comparsion is to always compare zone name, 503 * but it is also the slowest way. 504 */ 505 return (strcmp(zone->uz_name, vm_memguard_desc) == 0); 506 } 507