2 * Copyright (c) 2005, Bosko Milekic <bmilekic@FreeBSD.org>.
3 * Copyright (c) 2010 Isilon Systems, Inc. (http://www.isilon.com/)
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice unmodified, this list of conditions, and the following
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.
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.
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
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
37 * See the memguard(9) man page for more information on using MemGuard.
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>
48 #include <sys/mutex.h>
49 #include <sys/malloc.h>
50 #include <sys/sysctl.h>
54 #include <vm/vm_param.h>
55 #include <vm/vm_page.h>
56 #include <vm/vm_map.h>
57 #include <vm/vm_object.h>
58 #include <vm/vm_extern.h>
59 #include <vm/memguard.h>
61 static SYSCTL_NODE(_vm, OID_AUTO, memguard, CTLFLAG_RW, NULL, "MemGuard data");
63 * The vm_memguard_divisor variable controls how much of kmem_map should be
64 * reserved for MemGuard.
66 static u_int vm_memguard_divisor;
67 SYSCTL_UINT(_vm_memguard, OID_AUTO, divisor, CTLFLAG_RDTUN,
69 0, "(kmem_size/memguard_divisor) == memguard submap size");
72 * Short description (ks_shortdesc) of memory type to monitor.
74 static char vm_memguard_desc[128] = "";
75 static struct malloc_type *vm_memguard_mtype = NULL;
76 TUNABLE_STR("vm.memguard.desc", vm_memguard_desc, sizeof(vm_memguard_desc));
78 memguard_sysctl_desc(SYSCTL_HANDLER_ARGS)
80 char desc[sizeof(vm_memguard_desc)];
83 strlcpy(desc, vm_memguard_desc, sizeof(desc));
84 error = sysctl_handle_string(oidp, desc, sizeof(desc), req);
85 if (error != 0 || req->newptr == NULL)
88 mtx_lock(&malloc_mtx);
90 * If mtp is NULL, it will be initialized in memguard_cmp().
92 vm_memguard_mtype = malloc_desc2type(desc);
93 strlcpy(vm_memguard_desc, desc, sizeof(vm_memguard_desc));
94 mtx_unlock(&malloc_mtx);
97 SYSCTL_PROC(_vm_memguard, OID_AUTO, desc,
98 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0,
99 memguard_sysctl_desc, "A", "Short description of memory type to monitor");
101 static vm_map_t memguard_map = NULL;
102 static vm_offset_t memguard_cursor;
103 static vm_size_t memguard_mapsize;
104 static vm_size_t memguard_physlimit;
105 static u_long memguard_wasted;
106 static u_long memguard_wrap;
107 static u_long memguard_succ;
108 static u_long memguard_fail_kva;
109 static u_long memguard_fail_pgs;
111 SYSCTL_ULONG(_vm_memguard, OID_AUTO, cursor, CTLFLAG_RD,
112 &memguard_cursor, 0, "MemGuard cursor");
113 SYSCTL_ULONG(_vm_memguard, OID_AUTO, mapsize, CTLFLAG_RD,
114 &memguard_mapsize, 0, "MemGuard private vm_map size");
115 SYSCTL_ULONG(_vm_memguard, OID_AUTO, phys_limit, CTLFLAG_RD,
116 &memguard_physlimit, 0, "Limit on MemGuard memory consumption");
117 SYSCTL_ULONG(_vm_memguard, OID_AUTO, wasted, CTLFLAG_RD,
118 &memguard_wasted, 0, "Excess memory used through page promotion");
119 SYSCTL_ULONG(_vm_memguard, OID_AUTO, wrapcnt, CTLFLAG_RD,
120 &memguard_wrap, 0, "MemGuard cursor wrap count");
121 SYSCTL_ULONG(_vm_memguard, OID_AUTO, numalloc, CTLFLAG_RD,
122 &memguard_succ, 0, "Count of successful MemGuard allocations");
123 SYSCTL_ULONG(_vm_memguard, OID_AUTO, fail_kva, CTLFLAG_RD,
124 &memguard_fail_kva, 0, "MemGuard failures due to lack of KVA");
125 SYSCTL_ULONG(_vm_memguard, OID_AUTO, fail_pgs, CTLFLAG_RD,
126 &memguard_fail_pgs, 0, "MemGuard failures due to lack of pages");
128 #define MG_GUARD 0x001
129 #define MG_ALLLARGE 0x002
130 static int memguard_options = MG_GUARD;
131 TUNABLE_INT("vm.memguard.options", &memguard_options);
132 SYSCTL_INT(_vm_memguard, OID_AUTO, options, CTLFLAG_RW,
133 &memguard_options, 0,
134 "MemGuard options:\n"
135 "\t0x001 - add guard pages around each allocation\n"
136 "\t0x002 - always use MemGuard for allocations over a page");
138 static u_int memguard_minsize;
139 static u_long memguard_minsize_reject;
140 SYSCTL_UINT(_vm_memguard, OID_AUTO, minsize, CTLFLAG_RW,
141 &memguard_minsize, 0, "Minimum size for page promotion");
142 SYSCTL_ULONG(_vm_memguard, OID_AUTO, minsize_reject, CTLFLAG_RD,
143 &memguard_minsize_reject, 0, "# times rejected for size");
145 static u_int memguard_frequency;
146 static u_long memguard_frequency_hits;
147 TUNABLE_INT("vm.memguard.frequency", &memguard_frequency);
148 SYSCTL_UINT(_vm_memguard, OID_AUTO, frequency, CTLFLAG_RW,
149 &memguard_frequency, 0, "Times in 100000 that MemGuard will randomly run");
150 SYSCTL_ULONG(_vm_memguard, OID_AUTO, frequency_hits, CTLFLAG_RD,
151 &memguard_frequency_hits, 0, "# times MemGuard randomly chose");
155 * Return a fudged value to be used for vm_kmem_size for allocating
156 * the kmem_map. The memguard memory will be a submap.
159 memguard_fudge(unsigned long km_size, const struct vm_map *parent_map)
161 u_long mem_pgs, parent_size;
163 vm_memguard_divisor = 10;
164 TUNABLE_INT_FETCH("vm.memguard.divisor", &vm_memguard_divisor);
166 parent_size = vm_map_max(parent_map) - vm_map_min(parent_map) +
168 /* Pick a conservative value if provided value sucks. */
169 if ((vm_memguard_divisor <= 0) ||
170 ((parent_size / vm_memguard_divisor) == 0))
171 vm_memguard_divisor = 10;
173 * Limit consumption of physical pages to
174 * 1/vm_memguard_divisor of system memory. If the KVA is
175 * smaller than this then the KVA limit comes into play first.
176 * This prevents memguard's page promotions from completely
177 * using up memory, since most malloc(9) calls are sub-page.
179 mem_pgs = cnt.v_page_count;
180 memguard_physlimit = (mem_pgs / vm_memguard_divisor) * PAGE_SIZE;
182 * We want as much KVA as we can take safely. Use at most our
183 * allotted fraction of the parent map's size. Limit this to
184 * twice the physical memory to avoid using too much memory as
185 * pagetable pages (size must be multiple of PAGE_SIZE).
187 memguard_mapsize = round_page(parent_size / vm_memguard_divisor);
188 if (memguard_mapsize / (2 * PAGE_SIZE) > mem_pgs)
189 memguard_mapsize = mem_pgs * 2 * PAGE_SIZE;
190 if (km_size + memguard_mapsize > parent_size)
191 memguard_mapsize = 0;
192 return (km_size + memguard_mapsize);
196 * Initialize the MemGuard mock allocator. All objects from MemGuard come
197 * out of a single VM map (contiguous chunk of address space).
200 memguard_init(vm_map_t parent_map)
202 vm_offset_t base, limit;
204 memguard_map = kmem_suballoc(parent_map, &base, &limit,
205 memguard_mapsize, FALSE);
206 memguard_map->system_map = 1;
207 KASSERT(memguard_mapsize == limit - base,
208 ("Expected %lu, got %lu", (u_long)memguard_mapsize,
209 (u_long)(limit - base)));
210 memguard_cursor = base;
212 printf("MEMGUARD DEBUGGING ALLOCATOR INITIALIZED:\n");
213 printf("\tMEMGUARD map base: 0x%lx\n", (u_long)base);
214 printf("\tMEMGUARD map limit: 0x%lx\n", (u_long)limit);
215 printf("\tMEMGUARD map size: %jd KBytes\n",
216 (uintmax_t)memguard_mapsize >> 10);
220 * Run things that can't be done as early as memguard_init().
223 memguard_sysinit(void)
225 struct sysctl_oid_list *parent;
227 parent = SYSCTL_STATIC_CHILDREN(_vm_memguard);
229 SYSCTL_ADD_UAUTO(NULL, parent, OID_AUTO, "mapstart", CTLFLAG_RD,
230 &memguard_map->min_offset, "MemGuard KVA base");
231 SYSCTL_ADD_UAUTO(NULL, parent, OID_AUTO, "maplimit", CTLFLAG_RD,
232 &memguard_map->max_offset, "MemGuard KVA end");
233 SYSCTL_ADD_UAUTO(NULL, parent, OID_AUTO, "mapused", CTLFLAG_RD,
234 &memguard_map->size, "MemGuard KVA used");
236 SYSINIT(memguard, SI_SUB_KLD, SI_ORDER_ANY, memguard_sysinit, NULL);
239 * v2sizep() converts a virtual address of the first page allocated for
240 * an item to a pointer to u_long recording the size of the original
241 * allocation request.
243 * This routine is very similar to those defined by UMA in uma_int.h.
244 * The difference is that this routine stores the originally allocated
245 * size in one of the page's fields that is unused when the page is
246 * wired rather than the object field, which is used.
249 v2sizep(vm_offset_t va)
254 pa = pmap_kextract(va);
256 panic("MemGuard detected double-free of %p", (void *)va);
257 p = PHYS_TO_VM_PAGE(pa);
258 KASSERT(p->wire_count != 0 && p->queue == PQ_NONE,
259 ("MEMGUARD: Expected wired page %p in vtomgfifo!", p));
260 return ((u_long *)&p->pageq.tqe_next);
264 * Allocate a single object of specified size with specified flags
265 * (either M_WAITOK or M_NOWAIT).
268 memguard_alloc(unsigned long req_size, int flags)
271 u_long size_p, size_v;
274 size_p = round_page(req_size);
278 * To ensure there are holes on both sides of the allocation,
279 * request 2 extra pages of KVA. We will only actually add a
280 * vm_map_entry and get pages for the original request. Save
281 * the value of memguard_options so we have a consistent
285 do_guard = (memguard_options & MG_GUARD) != 0;
287 size_v += 2 * PAGE_SIZE;
289 vm_map_lock(memguard_map);
291 * When we pass our memory limit, reject sub-page allocations.
292 * Page-size and larger allocations will use the same amount
293 * of physical memory whether we allocate or hand off to
294 * uma_large_alloc(), so keep those.
296 if (memguard_map->size >= memguard_physlimit &&
297 req_size < PAGE_SIZE) {
298 addr = (vm_offset_t)NULL;
303 * Keep a moving cursor so we don't recycle KVA as long as
304 * possible. It's not perfect, since we don't know in what
305 * order previous allocations will be free'd, but it's simple
306 * and fast, and requires O(1) additional storage if guard
307 * pages are not used.
309 * XXX This scheme will lead to greater fragmentation of the
310 * map, unless vm_map_findspace() is tweaked.
313 rv = vm_map_findspace(memguard_map, memguard_cursor,
315 if (rv == KERN_SUCCESS)
318 * The map has no space. This may be due to
319 * fragmentation, or because the cursor is near the
322 if (memguard_cursor == vm_map_min(memguard_map)) {
324 addr = (vm_offset_t)NULL;
328 memguard_cursor = vm_map_min(memguard_map);
332 rv = kmem_back(memguard_map, addr, size_p, flags);
333 if (rv != KERN_SUCCESS) {
335 addr = (vm_offset_t)NULL;
338 memguard_cursor = addr + size_p;
339 *v2sizep(trunc_page(addr)) = req_size;
341 if (req_size < PAGE_SIZE) {
342 memguard_wasted += (PAGE_SIZE - req_size);
345 * Align the request to 16 bytes, and return
346 * an address near the end of the page, to
347 * better detect array overrun.
349 req_size = roundup2(req_size, 16);
350 addr += (PAGE_SIZE - req_size);
354 vm_map_unlock(memguard_map);
355 return ((void *)addr);
359 is_memguard_addr(void *addr)
361 vm_offset_t a = (vm_offset_t)(uintptr_t)addr;
363 return (a >= memguard_map->min_offset && a < memguard_map->max_offset);
367 * Free specified single object.
370 memguard_free(void *ptr)
373 u_long req_size, size;
377 addr = trunc_page((uintptr_t)ptr);
378 req_size = *v2sizep(addr);
379 size = round_page(req_size);
382 * Page should not be guarded right now, so force a write.
383 * The purpose of this is to increase the likelihood of
384 * catching a double-free, but not necessarily a
385 * tamper-after-free (the second thread freeing might not
386 * write before freeing, so this forces it to and,
387 * subsequently, trigger a fault).
390 for (i = 0; i < size; i += PAGE_SIZE)
394 * This requires carnal knowledge of the implementation of
395 * kmem_free(), but since we've already replaced kmem_malloc()
396 * above, it's not really any worse. We want to use the
397 * vm_map lock to serialize updates to memguard_wasted, since
398 * we had the lock at increment.
400 vm_map_lock(memguard_map);
401 if (req_size < PAGE_SIZE)
402 memguard_wasted -= (PAGE_SIZE - req_size);
403 (void)vm_map_delete(memguard_map, addr, addr + size);
404 vm_map_unlock(memguard_map);
408 * Re-allocate an allocation that was originally guarded.
411 memguard_realloc(void *addr, unsigned long size, struct malloc_type *mtp,
418 * Allocate the new block. Force the allocation to be guarded
419 * as the original may have been guarded through random
420 * chance, and that should be preserved.
422 if ((newaddr = memguard_alloc(size, flags)) == NULL)
425 /* Copy over original contents. */
426 old_size = *v2sizep(trunc_page((uintptr_t)addr));
427 bcopy(addr, newaddr, min(size, old_size));
433 memguard_cmp(struct malloc_type *mtp, unsigned long size)
436 if (size < memguard_minsize) {
437 memguard_minsize_reject++;
440 if ((memguard_options & MG_ALLLARGE) != 0 && size >= PAGE_SIZE)
442 if (memguard_frequency > 0 &&
443 (random() % 100000) < memguard_frequency) {
444 memguard_frequency_hits++;
449 * The safest way of comparsion is to always compare short description
450 * string of memory type, but it is also the slowest way.
452 return (strcmp(mtp->ks_shortdesc, vm_memguard_desc) == 0);
455 * If we compare pointers, there are two possible problems:
456 * 1. Memory type was unloaded and new memory type was allocated at the
458 * 2. Memory type was unloaded and loaded again, but allocated at a
461 if (vm_memguard_mtype != NULL)
462 return (mtp == vm_memguard_mtype);
463 if (strcmp(mtp->ks_shortdesc, vm_memguard_desc) == 0) {
464 vm_memguard_mtype = mtp;