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_extern.h> 60 #include <vm/uma_int.h> 61 #include <vm/memguard.h> 62 63 static SYSCTL_NODE(_vm, OID_AUTO, memguard, CTLFLAG_RW, NULL, "MemGuard data"); 64 /* 65 * The vm_memguard_divisor variable controls how much of kmem_map should be 66 * reserved for MemGuard. 67 */ 68 static u_int vm_memguard_divisor; 69 SYSCTL_UINT(_vm_memguard, OID_AUTO, divisor, CTLFLAG_RDTUN, 70 &vm_memguard_divisor, 71 0, "(kmem_size/memguard_divisor) == memguard submap size"); 72 73 /* 74 * Short description (ks_shortdesc) of memory type to monitor. 75 */ 76 static char vm_memguard_desc[128] = ""; 77 static struct malloc_type *vm_memguard_mtype = NULL; 78 TUNABLE_STR("vm.memguard.desc", vm_memguard_desc, sizeof(vm_memguard_desc)); 79 static int 80 memguard_sysctl_desc(SYSCTL_HANDLER_ARGS) 81 { 82 char desc[sizeof(vm_memguard_desc)]; 83 int error; 84 85 strlcpy(desc, vm_memguard_desc, sizeof(desc)); 86 error = sysctl_handle_string(oidp, desc, sizeof(desc), req); 87 if (error != 0 || req->newptr == NULL) 88 return (error); 89 90 mtx_lock(&malloc_mtx); 91 /* 92 * If mtp is NULL, it will be initialized in memguard_cmp(). 93 */ 94 vm_memguard_mtype = malloc_desc2type(desc); 95 strlcpy(vm_memguard_desc, desc, sizeof(vm_memguard_desc)); 96 mtx_unlock(&malloc_mtx); 97 return (error); 98 } 99 SYSCTL_PROC(_vm_memguard, OID_AUTO, desc, 100 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0, 101 memguard_sysctl_desc, "A", "Short description of memory type to monitor"); 102 103 static vmem_t *memguard_map = NULL; 104 static vm_offset_t memguard_cursor; 105 static vm_offset_t memguard_base; 106 static vm_size_t memguard_mapsize; 107 static vm_size_t memguard_physlimit; 108 static u_long memguard_wasted; 109 static u_long memguard_wrap; 110 static u_long memguard_succ; 111 static u_long memguard_fail_kva; 112 static u_long memguard_fail_pgs; 113 114 SYSCTL_ULONG(_vm_memguard, OID_AUTO, cursor, CTLFLAG_RD, 115 &memguard_cursor, 0, "MemGuard cursor"); 116 SYSCTL_ULONG(_vm_memguard, OID_AUTO, mapsize, CTLFLAG_RD, 117 &memguard_mapsize, 0, "MemGuard private arena size"); 118 SYSCTL_ULONG(_vm_memguard, OID_AUTO, phys_limit, CTLFLAG_RD, 119 &memguard_physlimit, 0, "Limit on MemGuard memory consumption"); 120 SYSCTL_ULONG(_vm_memguard, OID_AUTO, wasted, CTLFLAG_RD, 121 &memguard_wasted, 0, "Excess memory used through page promotion"); 122 SYSCTL_ULONG(_vm_memguard, OID_AUTO, wrapcnt, CTLFLAG_RD, 123 &memguard_wrap, 0, "MemGuard cursor wrap count"); 124 SYSCTL_ULONG(_vm_memguard, OID_AUTO, numalloc, CTLFLAG_RD, 125 &memguard_succ, 0, "Count of successful MemGuard allocations"); 126 SYSCTL_ULONG(_vm_memguard, OID_AUTO, fail_kva, CTLFLAG_RD, 127 &memguard_fail_kva, 0, "MemGuard failures due to lack of KVA"); 128 SYSCTL_ULONG(_vm_memguard, OID_AUTO, fail_pgs, CTLFLAG_RD, 129 &memguard_fail_pgs, 0, "MemGuard failures due to lack of pages"); 130 131 #define MG_GUARD_AROUND 0x001 132 #define MG_GUARD_ALLLARGE 0x002 133 #define MG_GUARD_NOFREE 0x004 134 static int memguard_options = MG_GUARD_AROUND; 135 TUNABLE_INT("vm.memguard.options", &memguard_options); 136 SYSCTL_INT(_vm_memguard, OID_AUTO, options, CTLFLAG_RW, 137 &memguard_options, 0, 138 "MemGuard options:\n" 139 "\t0x001 - add guard pages around each allocation\n" 140 "\t0x002 - always use MemGuard for allocations over a page\n" 141 "\t0x004 - guard uma(9) zones with UMA_ZONE_NOFREE flag"); 142 143 static u_int memguard_minsize; 144 static u_long memguard_minsize_reject; 145 SYSCTL_UINT(_vm_memguard, OID_AUTO, minsize, CTLFLAG_RW, 146 &memguard_minsize, 0, "Minimum size for page promotion"); 147 SYSCTL_ULONG(_vm_memguard, OID_AUTO, minsize_reject, CTLFLAG_RD, 148 &memguard_minsize_reject, 0, "# times rejected for size"); 149 150 static u_int memguard_frequency; 151 static u_long memguard_frequency_hits; 152 TUNABLE_INT("vm.memguard.frequency", &memguard_frequency); 153 SYSCTL_UINT(_vm_memguard, OID_AUTO, frequency, CTLFLAG_RW, 154 &memguard_frequency, 0, "Times in 100000 that MemGuard will randomly run"); 155 SYSCTL_ULONG(_vm_memguard, OID_AUTO, frequency_hits, CTLFLAG_RD, 156 &memguard_frequency_hits, 0, "# times MemGuard randomly chose"); 157 158 159 /* 160 * Return a fudged value to be used for vm_kmem_size for allocating 161 * the kmem_map. The memguard memory will be a submap. 162 */ 163 unsigned long 164 memguard_fudge(unsigned long km_size, const struct vm_map *parent_map) 165 { 166 u_long mem_pgs, parent_size; 167 168 vm_memguard_divisor = 10; 169 TUNABLE_INT_FETCH("vm.memguard.divisor", &vm_memguard_divisor); 170 171 parent_size = vm_map_max(parent_map) - vm_map_min(parent_map) + 172 PAGE_SIZE; 173 /* Pick a conservative value if provided value sucks. */ 174 if ((vm_memguard_divisor <= 0) || 175 ((parent_size / vm_memguard_divisor) == 0)) 176 vm_memguard_divisor = 10; 177 /* 178 * Limit consumption of physical pages to 179 * 1/vm_memguard_divisor of system memory. If the KVA is 180 * smaller than this then the KVA limit comes into play first. 181 * This prevents memguard's page promotions from completely 182 * using up memory, since most malloc(9) calls are sub-page. 183 */ 184 mem_pgs = cnt.v_page_count; 185 memguard_physlimit = (mem_pgs / vm_memguard_divisor) * PAGE_SIZE; 186 /* 187 * We want as much KVA as we can take safely. Use at most our 188 * allotted fraction of the parent map's size. Limit this to 189 * twice the physical memory to avoid using too much memory as 190 * pagetable pages (size must be multiple of PAGE_SIZE). 191 */ 192 memguard_mapsize = round_page(parent_size / vm_memguard_divisor); 193 if (memguard_mapsize / (2 * PAGE_SIZE) > mem_pgs) 194 memguard_mapsize = mem_pgs * 2 * PAGE_SIZE; 195 if (km_size + memguard_mapsize > parent_size) 196 memguard_mapsize = 0; 197 return (km_size + memguard_mapsize); 198 } 199 200 /* 201 * Initialize the MemGuard mock allocator. All objects from MemGuard come 202 * out of a single VM map (contiguous chunk of address space). 203 */ 204 void 205 memguard_init(vmem_t *parent) 206 { 207 vm_offset_t base; 208 209 vmem_alloc(parent, memguard_mapsize, M_WAITOK, &base); 210 memguard_map = vmem_create("memguard arena", base, memguard_mapsize, 211 PAGE_SIZE, 0, M_WAITOK); 212 memguard_cursor = base; 213 memguard_base = base; 214 215 printf("MEMGUARD DEBUGGING ALLOCATOR INITIALIZED:\n"); 216 printf("\tMEMGUARD map base: 0x%lx\n", (u_long)base); 217 printf("\tMEMGUARD map size: %jd KBytes\n", 218 (uintmax_t)memguard_mapsize >> 10); 219 } 220 221 /* 222 * Run things that can't be done as early as memguard_init(). 223 */ 224 static void 225 memguard_sysinit(void) 226 { 227 struct sysctl_oid_list *parent; 228 229 parent = SYSCTL_STATIC_CHILDREN(_vm_memguard); 230 231 SYSCTL_ADD_ULONG(NULL, parent, OID_AUTO, "mapstart", CTLFLAG_RD, 232 &memguard_base, "MemGuard KVA base"); 233 SYSCTL_ADD_ULONG(NULL, parent, OID_AUTO, "maplimit", CTLFLAG_RD, 234 &memguard_mapsize, "MemGuard KVA size"); 235 #if 0 236 SYSCTL_ADD_ULONG(NULL, parent, OID_AUTO, "mapused", CTLFLAG_RD, 237 &memguard_map->size, "MemGuard KVA used"); 238 #endif 239 } 240 SYSINIT(memguard, SI_SUB_KLD, SI_ORDER_ANY, memguard_sysinit, NULL); 241 242 /* 243 * v2sizep() converts a virtual address of the first page allocated for 244 * an item to a pointer to u_long recording the size of the original 245 * allocation request. 246 * 247 * This routine is very similar to those defined by UMA in uma_int.h. 248 * The difference is that this routine stores the originally allocated 249 * size in one of the page's fields that is unused when the page is 250 * wired rather than the object field, which is used. 251 */ 252 static u_long * 253 v2sizep(vm_offset_t va) 254 { 255 vm_paddr_t pa; 256 struct vm_page *p; 257 258 pa = pmap_kextract(va); 259 if (pa == 0) 260 panic("MemGuard detected double-free of %p", (void *)va); 261 p = PHYS_TO_VM_PAGE(pa); 262 KASSERT(p->wire_count != 0 && p->queue == PQ_NONE, 263 ("MEMGUARD: Expected wired page %p in vtomgfifo!", p)); 264 return (&p->plinks.memguard.p); 265 } 266 267 static u_long * 268 v2sizev(vm_offset_t va) 269 { 270 vm_paddr_t pa; 271 struct vm_page *p; 272 273 pa = pmap_kextract(va); 274 if (pa == 0) 275 panic("MemGuard detected double-free of %p", (void *)va); 276 p = PHYS_TO_VM_PAGE(pa); 277 KASSERT(p->wire_count != 0 && p->queue == PQ_NONE, 278 ("MEMGUARD: Expected wired page %p in vtomgfifo!", p)); 279 return (&p->plinks.memguard.v); 280 } 281 282 /* 283 * Allocate a single object of specified size with specified flags 284 * (either M_WAITOK or M_NOWAIT). 285 */ 286 void * 287 memguard_alloc(unsigned long req_size, int flags) 288 { 289 vm_offset_t addr; 290 u_long size_p, size_v; 291 int do_guard, rv; 292 293 size_p = round_page(req_size); 294 if (size_p == 0) 295 return (NULL); 296 /* 297 * To ensure there are holes on both sides of the allocation, 298 * request 2 extra pages of KVA. We will only actually add a 299 * vm_map_entry and get pages for the original request. Save 300 * the value of memguard_options so we have a consistent 301 * value. 302 */ 303 size_v = size_p; 304 do_guard = (memguard_options & MG_GUARD_AROUND) != 0; 305 if (do_guard) 306 size_v += 2 * PAGE_SIZE; 307 308 /* 309 * When we pass our memory limit, reject sub-page allocations. 310 * Page-size and larger allocations will use the same amount 311 * of physical memory whether we allocate or hand off to 312 * uma_large_alloc(), so keep those. 313 */ 314 if (vmem_size(memguard_map, VMEM_ALLOC) >= memguard_physlimit && 315 req_size < PAGE_SIZE) { 316 addr = (vm_offset_t)NULL; 317 memguard_fail_pgs++; 318 goto out; 319 } 320 /* 321 * Keep a moving cursor so we don't recycle KVA as long as 322 * possible. It's not perfect, since we don't know in what 323 * order previous allocations will be free'd, but it's simple 324 * and fast, and requires O(1) additional storage if guard 325 * pages are not used. 326 * 327 * XXX This scheme will lead to greater fragmentation of the 328 * map, unless vm_map_findspace() is tweaked. 329 */ 330 for (;;) { 331 if (vmem_xalloc(memguard_map, size_v, 0, 0, 0, memguard_cursor, 332 VMEM_ADDR_MAX, M_BESTFIT | M_NOWAIT, &addr) == 0) 333 break; 334 /* 335 * The map has no space. This may be due to 336 * fragmentation, or because the cursor is near the 337 * end of the map. 338 */ 339 if (memguard_cursor == memguard_base) { 340 memguard_fail_kva++; 341 addr = (vm_offset_t)NULL; 342 goto out; 343 } 344 memguard_wrap++; 345 memguard_cursor = memguard_base; 346 } 347 if (do_guard) 348 addr += PAGE_SIZE; 349 rv = kmem_back(kmem_object, addr, size_p, flags); 350 if (rv != KERN_SUCCESS) { 351 vmem_xfree(memguard_map, addr, size_v); 352 memguard_fail_pgs++; 353 addr = (vm_offset_t)NULL; 354 goto out; 355 } 356 memguard_cursor = addr + size_v; 357 *v2sizep(trunc_page(addr)) = req_size; 358 *v2sizev(trunc_page(addr)) = size_v; 359 memguard_succ++; 360 if (req_size < PAGE_SIZE) { 361 memguard_wasted += (PAGE_SIZE - req_size); 362 if (do_guard) { 363 /* 364 * Align the request to 16 bytes, and return 365 * an address near the end of the page, to 366 * better detect array overrun. 367 */ 368 req_size = roundup2(req_size, 16); 369 addr += (PAGE_SIZE - req_size); 370 } 371 } 372 out: 373 return ((void *)addr); 374 } 375 376 int 377 is_memguard_addr(void *addr) 378 { 379 vm_offset_t a = (vm_offset_t)(uintptr_t)addr; 380 381 return (a >= memguard_base && a < memguard_base + memguard_mapsize); 382 } 383 384 /* 385 * Free specified single object. 386 */ 387 void 388 memguard_free(void *ptr) 389 { 390 vm_offset_t addr; 391 u_long req_size, size, sizev; 392 char *temp; 393 int i; 394 395 addr = trunc_page((uintptr_t)ptr); 396 req_size = *v2sizep(addr); 397 sizev = *v2sizev(addr); 398 size = round_page(req_size); 399 400 /* 401 * Page should not be guarded right now, so force a write. 402 * The purpose of this is to increase the likelihood of 403 * catching a double-free, but not necessarily a 404 * tamper-after-free (the second thread freeing might not 405 * write before freeing, so this forces it to and, 406 * subsequently, trigger a fault). 407 */ 408 temp = ptr; 409 for (i = 0; i < size; i += PAGE_SIZE) 410 temp[i] = 'M'; 411 412 /* 413 * This requires carnal knowledge of the implementation of 414 * kmem_free(), but since we've already replaced kmem_malloc() 415 * above, it's not really any worse. We want to use the 416 * vm_map lock to serialize updates to memguard_wasted, since 417 * we had the lock at increment. 418 */ 419 kmem_unback(kmem_object, addr, size); 420 if (sizev > size) 421 addr -= PAGE_SIZE; 422 vmem_xfree(memguard_map, addr, sizev); 423 if (req_size < PAGE_SIZE) 424 memguard_wasted -= (PAGE_SIZE - req_size); 425 } 426 427 /* 428 * Re-allocate an allocation that was originally guarded. 429 */ 430 void * 431 memguard_realloc(void *addr, unsigned long size, struct malloc_type *mtp, 432 int flags) 433 { 434 void *newaddr; 435 u_long old_size; 436 437 /* 438 * Allocate the new block. Force the allocation to be guarded 439 * as the original may have been guarded through random 440 * chance, and that should be preserved. 441 */ 442 if ((newaddr = memguard_alloc(size, flags)) == NULL) 443 return (NULL); 444 445 /* Copy over original contents. */ 446 old_size = *v2sizep(trunc_page((uintptr_t)addr)); 447 bcopy(addr, newaddr, min(size, old_size)); 448 memguard_free(addr); 449 return (newaddr); 450 } 451 452 static int 453 memguard_cmp(unsigned long size) 454 { 455 456 if (size < memguard_minsize) { 457 memguard_minsize_reject++; 458 return (0); 459 } 460 if ((memguard_options & MG_GUARD_ALLLARGE) != 0 && size >= PAGE_SIZE) 461 return (1); 462 if (memguard_frequency > 0 && 463 (random() % 100000) < memguard_frequency) { 464 memguard_frequency_hits++; 465 return (1); 466 } 467 468 return (0); 469 } 470 471 int 472 memguard_cmp_mtp(struct malloc_type *mtp, unsigned long size) 473 { 474 475 if (memguard_cmp(size)) 476 return(1); 477 478 #if 1 479 /* 480 * The safest way of comparsion is to always compare short description 481 * string of memory type, but it is also the slowest way. 482 */ 483 return (strcmp(mtp->ks_shortdesc, vm_memguard_desc) == 0); 484 #else 485 /* 486 * If we compare pointers, there are two possible problems: 487 * 1. Memory type was unloaded and new memory type was allocated at the 488 * same address. 489 * 2. Memory type was unloaded and loaded again, but allocated at a 490 * different address. 491 */ 492 if (vm_memguard_mtype != NULL) 493 return (mtp == vm_memguard_mtype); 494 if (strcmp(mtp->ks_shortdesc, vm_memguard_desc) == 0) { 495 vm_memguard_mtype = mtp; 496 return (1); 497 } 498 return (0); 499 #endif 500 } 501 502 int 503 memguard_cmp_zone(uma_zone_t zone) 504 { 505 506 if ((memguard_options & MG_GUARD_NOFREE) == 0 && 507 zone->uz_flags & UMA_ZONE_NOFREE) 508 return (0); 509 510 if (memguard_cmp(zone->uz_size)) 511 return (1); 512 513 /* 514 * The safest way of comparsion is to always compare zone name, 515 * but it is also the slowest way. 516 */ 517 return (strcmp(zone->uz_name, vm_memguard_desc) == 0); 518 } 519