1 /*- 2 * Copyright (c) 2005, Bosko Milekic <bmilekic@FreeBSD.org>. 3 * Copyright (c) 2010 Isilon Systems, Inc. (http://www.isilon.com/) 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice unmodified, this list of conditions, and the following 11 * disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 18 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 19 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 21 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 22 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 23 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 24 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 25 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 26 */ 27 28 #include <sys/cdefs.h> 29 __FBSDID("$FreeBSD$"); 30 31 /* 32 * MemGuard is a simple replacement allocator for debugging only 33 * which provides ElectricFence-style memory barrier protection on 34 * objects being allocated, and is used to detect tampering-after-free 35 * scenarios. 36 * 37 * See the memguard(9) man page for more information on using MemGuard. 38 */ 39 40 #include "opt_vm.h" 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/kernel.h> 45 #include <sys/types.h> 46 #include <sys/queue.h> 47 #include <sys/lock.h> 48 #include <sys/mutex.h> 49 #include <sys/malloc.h> 50 #include <sys/sysctl.h> 51 #include <sys/vmem.h> 52 53 #include <vm/vm.h> 54 #include <vm/uma.h> 55 #include <vm/vm_param.h> 56 #include <vm/vm_page.h> 57 #include <vm/vm_map.h> 58 #include <vm/vm_object.h> 59 #include <vm/vm_kern.h> 60 #include <vm/vm_extern.h> 61 #include <vm/uma_int.h> 62 #include <vm/memguard.h> 63 64 static SYSCTL_NODE(_vm, OID_AUTO, memguard, CTLFLAG_RW, NULL, "MemGuard data"); 65 /* 66 * The vm_memguard_divisor variable controls how much of kmem_map should be 67 * reserved for MemGuard. 68 */ 69 static u_int vm_memguard_divisor; 70 SYSCTL_UINT(_vm_memguard, OID_AUTO, divisor, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 71 &vm_memguard_divisor, 72 0, "(kmem_size/memguard_divisor) == memguard submap size"); 73 74 /* 75 * Short description (ks_shortdesc) of memory type to monitor. 76 */ 77 static char vm_memguard_desc[128] = ""; 78 static struct malloc_type *vm_memguard_mtype = NULL; 79 TUNABLE_STR("vm.memguard.desc", vm_memguard_desc, sizeof(vm_memguard_desc)); 80 static int 81 memguard_sysctl_desc(SYSCTL_HANDLER_ARGS) 82 { 83 char desc[sizeof(vm_memguard_desc)]; 84 int error; 85 86 strlcpy(desc, vm_memguard_desc, sizeof(desc)); 87 error = sysctl_handle_string(oidp, desc, sizeof(desc), req); 88 if (error != 0 || req->newptr == NULL) 89 return (error); 90 91 mtx_lock(&malloc_mtx); 92 /* If mtp is NULL, it will be initialized in memguard_cmp() */ 93 vm_memguard_mtype = malloc_desc2type(desc); 94 strlcpy(vm_memguard_desc, desc, sizeof(vm_memguard_desc)); 95 mtx_unlock(&malloc_mtx); 96 return (error); 97 } 98 SYSCTL_PROC(_vm_memguard, OID_AUTO, desc, 99 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0, 100 memguard_sysctl_desc, "A", "Short description of memory type to monitor"); 101 102 static vm_offset_t memguard_cursor; 103 static vm_offset_t memguard_base; 104 static vm_size_t memguard_mapsize; 105 static vm_size_t memguard_physlimit; 106 static u_long memguard_wasted; 107 static u_long memguard_wrap; 108 static u_long memguard_succ; 109 static u_long memguard_fail_kva; 110 static u_long memguard_fail_pgs; 111 112 SYSCTL_ULONG(_vm_memguard, OID_AUTO, cursor, CTLFLAG_RD, 113 &memguard_cursor, 0, "MemGuard cursor"); 114 SYSCTL_ULONG(_vm_memguard, OID_AUTO, mapsize, CTLFLAG_RD, 115 &memguard_mapsize, 0, "MemGuard private arena size"); 116 SYSCTL_ULONG(_vm_memguard, OID_AUTO, phys_limit, CTLFLAG_RD, 117 &memguard_physlimit, 0, "Limit on MemGuard memory consumption"); 118 SYSCTL_ULONG(_vm_memguard, OID_AUTO, wasted, CTLFLAG_RD, 119 &memguard_wasted, 0, "Excess memory used through page promotion"); 120 SYSCTL_ULONG(_vm_memguard, OID_AUTO, wrapcnt, CTLFLAG_RD, 121 &memguard_wrap, 0, "MemGuard cursor wrap count"); 122 SYSCTL_ULONG(_vm_memguard, OID_AUTO, numalloc, CTLFLAG_RD, 123 &memguard_succ, 0, "Count of successful MemGuard allocations"); 124 SYSCTL_ULONG(_vm_memguard, OID_AUTO, fail_kva, CTLFLAG_RD, 125 &memguard_fail_kva, 0, "MemGuard failures due to lack of KVA"); 126 SYSCTL_ULONG(_vm_memguard, OID_AUTO, fail_pgs, CTLFLAG_RD, 127 &memguard_fail_pgs, 0, "MemGuard failures due to lack of pages"); 128 129 #define MG_GUARD_AROUND 0x001 130 #define MG_GUARD_ALLLARGE 0x002 131 #define MG_GUARD_NOFREE 0x004 132 static int memguard_options = MG_GUARD_AROUND; 133 SYSCTL_INT(_vm_memguard, OID_AUTO, options, CTLFLAG_RWTUN, 134 &memguard_options, 0, 135 "MemGuard options:\n" 136 "\t0x001 - add guard pages around each allocation\n" 137 "\t0x002 - always use MemGuard for allocations over a page\n" 138 "\t0x004 - guard uma(9) zones with UMA_ZONE_NOFREE flag"); 139 140 static u_int memguard_minsize; 141 static u_long memguard_minsize_reject; 142 SYSCTL_UINT(_vm_memguard, OID_AUTO, minsize, CTLFLAG_RW, 143 &memguard_minsize, 0, "Minimum size for page promotion"); 144 SYSCTL_ULONG(_vm_memguard, OID_AUTO, minsize_reject, CTLFLAG_RD, 145 &memguard_minsize_reject, 0, "# times rejected for size"); 146 147 static u_int memguard_frequency; 148 static u_long memguard_frequency_hits; 149 SYSCTL_UINT(_vm_memguard, OID_AUTO, frequency, CTLFLAG_RWTUN, 150 &memguard_frequency, 0, "Times in 100000 that MemGuard will randomly run"); 151 SYSCTL_ULONG(_vm_memguard, OID_AUTO, frequency_hits, CTLFLAG_RD, 152 &memguard_frequency_hits, 0, "# times MemGuard randomly chose"); 153 154 155 /* 156 * Return a fudged value to be used for vm_kmem_size for allocating 157 * the kmem_map. The memguard memory will be a submap. 158 */ 159 unsigned long 160 memguard_fudge(unsigned long km_size, const struct vm_map *parent_map) 161 { 162 u_long mem_pgs, parent_size; 163 164 vm_memguard_divisor = 10; 165 /* CTFLAG_RDTUN doesn't work during the early boot process. */ 166 TUNABLE_INT_FETCH("vm.memguard.divisor", &vm_memguard_divisor); 167 168 parent_size = vm_map_max(parent_map) - vm_map_min(parent_map) + 169 PAGE_SIZE; 170 /* Pick a conservative value if provided value sucks. */ 171 if ((vm_memguard_divisor <= 0) || 172 ((parent_size / vm_memguard_divisor) == 0)) 173 vm_memguard_divisor = 10; 174 /* 175 * Limit consumption of physical pages to 176 * 1/vm_memguard_divisor of system memory. If the KVA is 177 * smaller than this then the KVA limit comes into play first. 178 * This prevents memguard's page promotions from completely 179 * using up memory, since most malloc(9) calls are sub-page. 180 */ 181 mem_pgs = vm_cnt.v_page_count; 182 memguard_physlimit = (mem_pgs / vm_memguard_divisor) * PAGE_SIZE; 183 /* 184 * We want as much KVA as we can take safely. Use at most our 185 * allotted fraction of the parent map's size. Limit this to 186 * twice the physical memory to avoid using too much memory as 187 * pagetable pages (size must be multiple of PAGE_SIZE). 188 */ 189 memguard_mapsize = round_page(parent_size / vm_memguard_divisor); 190 if (memguard_mapsize / (2 * PAGE_SIZE) > mem_pgs) 191 memguard_mapsize = mem_pgs * 2 * PAGE_SIZE; 192 if (km_size + memguard_mapsize > parent_size) 193 memguard_mapsize = 0; 194 return (km_size + memguard_mapsize); 195 } 196 197 /* 198 * Initialize the MemGuard mock allocator. All objects from MemGuard come 199 * out of a single VM map (contiguous chunk of address space). 200 */ 201 void 202 memguard_init(vmem_t *parent) 203 { 204 vm_offset_t base; 205 206 vmem_alloc(parent, memguard_mapsize, M_BESTFIT | M_WAITOK, &base); 207 vmem_init(memguard_arena, "memguard arena", base, memguard_mapsize, 208 PAGE_SIZE, 0, M_WAITOK); 209 memguard_cursor = base; 210 memguard_base = base; 211 212 printf("MEMGUARD DEBUGGING ALLOCATOR INITIALIZED:\n"); 213 printf("\tMEMGUARD map base: 0x%lx\n", (u_long)base); 214 printf("\tMEMGUARD map size: %jd KBytes\n", 215 (uintmax_t)memguard_mapsize >> 10); 216 } 217 218 /* 219 * Run things that can't be done as early as memguard_init(). 220 */ 221 static void 222 memguard_sysinit(void) 223 { 224 struct sysctl_oid_list *parent; 225 226 parent = SYSCTL_STATIC_CHILDREN(_vm_memguard); 227 228 SYSCTL_ADD_UAUTO(NULL, parent, OID_AUTO, "mapstart", CTLFLAG_RD, 229 &memguard_base, "MemGuard KVA base"); 230 SYSCTL_ADD_UAUTO(NULL, parent, OID_AUTO, "maplimit", CTLFLAG_RD, 231 &memguard_mapsize, "MemGuard KVA size"); 232 #if 0 233 SYSCTL_ADD_ULONG(NULL, parent, OID_AUTO, "mapused", CTLFLAG_RD, 234 &memguard_map->size, "MemGuard KVA used"); 235 #endif 236 } 237 SYSINIT(memguard, SI_SUB_KLD, SI_ORDER_ANY, memguard_sysinit, NULL); 238 239 /* 240 * v2sizep() converts a virtual address of the first page allocated for 241 * an item to a pointer to u_long recording the size of the original 242 * allocation request. 243 * 244 * This routine is very similar to those defined by UMA in uma_int.h. 245 * The difference is that this routine stores the originally allocated 246 * size in one of the page's fields that is unused when the page is 247 * wired rather than the object field, which is used. 248 */ 249 static u_long * 250 v2sizep(vm_offset_t va) 251 { 252 vm_paddr_t pa; 253 struct vm_page *p; 254 255 pa = pmap_kextract(va); 256 if (pa == 0) 257 panic("MemGuard detected double-free of %p", (void *)va); 258 p = PHYS_TO_VM_PAGE(pa); 259 KASSERT(p->wire_count != 0 && p->queue == PQ_NONE, 260 ("MEMGUARD: Expected wired page %p in vtomgfifo!", p)); 261 return (&p->plinks.memguard.p); 262 } 263 264 static u_long * 265 v2sizev(vm_offset_t va) 266 { 267 vm_paddr_t pa; 268 struct vm_page *p; 269 270 pa = pmap_kextract(va); 271 if (pa == 0) 272 panic("MemGuard detected double-free of %p", (void *)va); 273 p = PHYS_TO_VM_PAGE(pa); 274 KASSERT(p->wire_count != 0 && p->queue == PQ_NONE, 275 ("MEMGUARD: Expected wired page %p in vtomgfifo!", p)); 276 return (&p->plinks.memguard.v); 277 } 278 279 /* 280 * Allocate a single object of specified size with specified flags 281 * (either M_WAITOK or M_NOWAIT). 282 */ 283 void * 284 memguard_alloc(unsigned long req_size, int flags) 285 { 286 vm_offset_t addr; 287 u_long size_p, size_v; 288 int do_guard, rv; 289 290 size_p = round_page(req_size); 291 if (size_p == 0) 292 return (NULL); 293 /* 294 * To ensure there are holes on both sides of the allocation, 295 * request 2 extra pages of KVA. We will only actually add a 296 * vm_map_entry and get pages for the original request. Save 297 * the value of memguard_options so we have a consistent 298 * value. 299 */ 300 size_v = size_p; 301 do_guard = (memguard_options & MG_GUARD_AROUND) != 0; 302 if (do_guard) 303 size_v += 2 * PAGE_SIZE; 304 305 /* 306 * When we pass our memory limit, reject sub-page allocations. 307 * Page-size and larger allocations will use the same amount 308 * of physical memory whether we allocate or hand off to 309 * uma_large_alloc(), so keep those. 310 */ 311 if (vmem_size(memguard_arena, VMEM_ALLOC) >= memguard_physlimit && 312 req_size < PAGE_SIZE) { 313 addr = (vm_offset_t)NULL; 314 memguard_fail_pgs++; 315 goto out; 316 } 317 /* 318 * Keep a moving cursor so we don't recycle KVA as long as 319 * possible. It's not perfect, since we don't know in what 320 * order previous allocations will be free'd, but it's simple 321 * and fast, and requires O(1) additional storage if guard 322 * pages are not used. 323 * 324 * XXX This scheme will lead to greater fragmentation of the 325 * map, unless vm_map_findspace() is tweaked. 326 */ 327 for (;;) { 328 if (vmem_xalloc(memguard_arena, size_v, 0, 0, 0, 329 memguard_cursor, VMEM_ADDR_MAX, 330 M_BESTFIT | M_NOWAIT, &addr) == 0) 331 break; 332 /* 333 * The map has no space. This may be due to 334 * fragmentation, or because the cursor is near the 335 * end of the map. 336 */ 337 if (memguard_cursor == memguard_base) { 338 memguard_fail_kva++; 339 addr = (vm_offset_t)NULL; 340 goto out; 341 } 342 memguard_wrap++; 343 memguard_cursor = memguard_base; 344 } 345 if (do_guard) 346 addr += PAGE_SIZE; 347 rv = kmem_back(kmem_object, addr, size_p, flags); 348 if (rv != KERN_SUCCESS) { 349 vmem_xfree(memguard_arena, addr, size_v); 350 memguard_fail_pgs++; 351 addr = (vm_offset_t)NULL; 352 goto out; 353 } 354 memguard_cursor = addr + size_v; 355 *v2sizep(trunc_page(addr)) = req_size; 356 *v2sizev(trunc_page(addr)) = size_v; 357 memguard_succ++; 358 if (req_size < PAGE_SIZE) { 359 memguard_wasted += (PAGE_SIZE - req_size); 360 if (do_guard) { 361 /* 362 * Align the request to 16 bytes, and return 363 * an address near the end of the page, to 364 * better detect array overrun. 365 */ 366 req_size = roundup2(req_size, 16); 367 addr += (PAGE_SIZE - req_size); 368 } 369 } 370 out: 371 return ((void *)addr); 372 } 373 374 int 375 is_memguard_addr(void *addr) 376 { 377 vm_offset_t a = (vm_offset_t)(uintptr_t)addr; 378 379 return (a >= memguard_base && a < memguard_base + memguard_mapsize); 380 } 381 382 /* 383 * Free specified single object. 384 */ 385 void 386 memguard_free(void *ptr) 387 { 388 vm_offset_t addr; 389 u_long req_size, size, sizev; 390 char *temp; 391 int i; 392 393 addr = trunc_page((uintptr_t)ptr); 394 req_size = *v2sizep(addr); 395 sizev = *v2sizev(addr); 396 size = round_page(req_size); 397 398 /* 399 * Page should not be guarded right now, so force a write. 400 * The purpose of this is to increase the likelihood of 401 * catching a double-free, but not necessarily a 402 * tamper-after-free (the second thread freeing might not 403 * write before freeing, so this forces it to and, 404 * subsequently, trigger a fault). 405 */ 406 temp = ptr; 407 for (i = 0; i < size; i += PAGE_SIZE) 408 temp[i] = 'M'; 409 410 /* 411 * This requires carnal knowledge of the implementation of 412 * kmem_free(), but since we've already replaced kmem_malloc() 413 * above, it's not really any worse. We want to use the 414 * vm_map lock to serialize updates to memguard_wasted, since 415 * we had the lock at increment. 416 */ 417 kmem_unback(kmem_object, addr, size); 418 if (sizev > size) 419 addr -= PAGE_SIZE; 420 vmem_xfree(memguard_arena, addr, sizev); 421 if (req_size < PAGE_SIZE) 422 memguard_wasted -= (PAGE_SIZE - req_size); 423 } 424 425 /* 426 * Re-allocate an allocation that was originally guarded. 427 */ 428 void * 429 memguard_realloc(void *addr, unsigned long size, struct malloc_type *mtp, 430 int flags) 431 { 432 void *newaddr; 433 u_long old_size; 434 435 /* 436 * Allocate the new block. Force the allocation to be guarded 437 * as the original may have been guarded through random 438 * chance, and that should be preserved. 439 */ 440 if ((newaddr = memguard_alloc(size, flags)) == NULL) 441 return (NULL); 442 443 /* Copy over original contents. */ 444 old_size = *v2sizep(trunc_page((uintptr_t)addr)); 445 bcopy(addr, newaddr, min(size, old_size)); 446 memguard_free(addr); 447 return (newaddr); 448 } 449 450 static int 451 memguard_cmp(unsigned long size) 452 { 453 454 if (size < memguard_minsize) { 455 memguard_minsize_reject++; 456 return (0); 457 } 458 if ((memguard_options & MG_GUARD_ALLLARGE) != 0 && size >= PAGE_SIZE) 459 return (1); 460 if (memguard_frequency > 0 && 461 (random() % 100000) < memguard_frequency) { 462 memguard_frequency_hits++; 463 return (1); 464 } 465 466 return (0); 467 } 468 469 int 470 memguard_cmp_mtp(struct malloc_type *mtp, unsigned long size) 471 { 472 473 if (memguard_cmp(size)) 474 return(1); 475 476 #if 1 477 /* 478 * The safest way of comparsion is to always compare short description 479 * string of memory type, but it is also the slowest way. 480 */ 481 return (strcmp(mtp->ks_shortdesc, vm_memguard_desc) == 0); 482 #else 483 /* 484 * If we compare pointers, there are two possible problems: 485 * 1. Memory type was unloaded and new memory type was allocated at the 486 * same address. 487 * 2. Memory type was unloaded and loaded again, but allocated at a 488 * different address. 489 */ 490 if (vm_memguard_mtype != NULL) 491 return (mtp == vm_memguard_mtype); 492 if (strcmp(mtp->ks_shortdesc, vm_memguard_desc) == 0) { 493 vm_memguard_mtype = mtp; 494 return (1); 495 } 496 return (0); 497 #endif 498 } 499 500 int 501 memguard_cmp_zone(uma_zone_t zone) 502 { 503 504 if ((memguard_options & MG_GUARD_NOFREE) == 0 && 505 zone->uz_flags & UMA_ZONE_NOFREE) 506 return (0); 507 508 if (memguard_cmp(zone->uz_size)) 509 return (1); 510 511 /* 512 * The safest way of comparsion is to always compare zone name, 513 * but it is also the slowest way. 514 */ 515 return (strcmp(zone->uz_name, vm_memguard_desc) == 0); 516 } 517