Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(459)

Side by Side Diff: third_party/WebKit/Source/wtf/allocator/PartitionAlloc.h

Issue 2518253002: Move Partition Allocator into Chromium base. (Closed)
Patch Set: Move OOM_CRASH into its own, more specific header. Fixes Windows build. Created 4 years ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View unified diff | Download patch
OLDNEW
(Empty)
1 /*
2 * Copyright (C) 2013 Google Inc. All rights reserved.
3 *
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions are
6 * met:
7 *
8 * * Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * * Redistributions in binary form must reproduce the above
11 * copyright notice, this list of conditions and the following disclaimer
12 * in the documentation and/or other materials provided with the
13 * distribution.
14 * * Neither the name of Google Inc. nor the names of its
15 * contributors may be used to endorse or promote products derived from
16 * this software without specific prior written permission.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
19 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
20 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
21 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
22 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
23 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
24 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
28 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29 */
30
31 #ifndef WTF_PartitionAlloc_h
32 #define WTF_PartitionAlloc_h
33
34 // DESCRIPTION
35 // partitionAlloc() / partitionAllocGeneric() and partitionFree() /
36 // partitionFreeGeneric() are approximately analagous to malloc() and free().
37 //
38 // The main difference is that a PartitionRoot / PartitionRootGeneric object
39 // must be supplied to these functions, representing a specific "heap partition"
40 // that will be used to satisfy the allocation. Different partitions are
41 // guaranteed to exist in separate address spaces, including being separate from
42 // the main system heap. If the contained objects are all freed, physical memory
43 // is returned to the system but the address space remains reserved.
44 // See PartitionAlloc.md for other security properties PartitionAlloc provides.
45 //
46 // THE ONLY LEGITIMATE WAY TO OBTAIN A PartitionRoot IS THROUGH THE
47 // SizeSpecificPartitionAllocator / PartitionAllocatorGeneric classes. To
48 // minimize the instruction count to the fullest extent possible, the
49 // PartitionRoot is really just a header adjacent to other data areas provided
50 // by the allocator class.
51 //
52 // The partitionAlloc() variant of the API has the following caveats:
53 // - Allocations and frees against a single partition must be single threaded.
54 // - Allocations must not exceed a max size, chosen at compile-time via a
55 // templated parameter to PartitionAllocator.
56 // - Allocation sizes must be aligned to the system pointer size.
57 // - Allocations are bucketed exactly according to size.
58 //
59 // And for partitionAllocGeneric():
60 // - Multi-threaded use against a single partition is ok; locking is handled.
61 // - Allocations of any arbitrary size can be handled (subject to a limit of
62 // INT_MAX bytes for security reasons).
63 // - Bucketing is by approximate size, for example an allocation of 4000 bytes
64 // might be placed into a 4096-byte bucket. Bucket sizes are chosen to try and
65 // keep worst-case waste to ~10%.
66 //
67 // The allocators are designed to be extremely fast, thanks to the following
68 // properties and design:
69 // - Just two single (reasonably predicatable) branches in the hot / fast path
70 // for both allocating and (significantly) freeing.
71 // - A minimal number of operations in the hot / fast path, with the slow paths
72 // in separate functions, leading to the possibility of inlining.
73 // - Each partition page (which is usually multiple physical pages) has a
74 // metadata structure which allows fast mapping of free() address to an
75 // underlying bucket.
76 // - Supports a lock-free API for fast performance in single-threaded cases.
77 // - The freelist for a given bucket is split across a number of partition
78 // pages, enabling various simple tricks to try and minimize fragmentation.
79 // - Fine-grained bucket sizes leading to less waste and better packing.
80 //
81 // The following security properties could be investigated in the future:
82 // - Per-object bucketing (instead of per-size) is mostly available at the API,
83 // but not used yet.
84 // - No randomness of freelist entries or bucket position.
85 // - Better checking for wild pointers in free().
86 // - Better freelist masking function to guarantee fault on 32-bit.
87
88 #include "wtf/Assertions.h"
89 #include "wtf/BitwiseOperations.h"
90 #include "wtf/ByteSwap.h"
91 #include "wtf/CPU.h"
92 #include "wtf/SpinLock.h"
93 #include "wtf/TypeTraits.h"
94 #include "wtf/allocator/PageAllocator.h"
95
96 #include <limits.h>
97
98 #if defined(MEMORY_TOOL_REPLACES_ALLOCATOR)
99 #include <stdlib.h>
100 #endif
101
102 #if ENABLE(ASSERT)
103 #include <string.h>
104 #endif
105
106 namespace WTF {
107
108 // Allocation granularity of sizeof(void*) bytes.
109 static const size_t kAllocationGranularity = sizeof(void*);
110 static const size_t kAllocationGranularityMask = kAllocationGranularity - 1;
111 static const size_t kBucketShift = (kAllocationGranularity == 8) ? 3 : 2;
112
113 // Underlying partition storage pages are a power-of-two size. It is typical
114 // for a partition page to be based on multiple system pages. Most references to
115 // "page" refer to partition pages.
116 // We also have the concept of "super pages" -- these are the underlying system
117 // allocations we make. Super pages contain multiple partition pages inside them
118 // and include space for a small amount of metadata per partition page.
119 // Inside super pages, we store "slot spans". A slot span is a continguous range
120 // of one or more partition pages that stores allocations of the same size.
121 // Slot span sizes are adjusted depending on the allocation size, to make sure
122 // the packing does not lead to unused (wasted) space at the end of the last
123 // system page of the span. For our current max slot span size of 64k and other
124 // constant values, we pack _all_ partitionAllocGeneric() sizes perfectly up
125 // against the end of a system page.
126 static const size_t kPartitionPageShift = 14; // 16KB
127 static const size_t kPartitionPageSize = 1 << kPartitionPageShift;
128 static const size_t kPartitionPageOffsetMask = kPartitionPageSize - 1;
129 static const size_t kPartitionPageBaseMask = ~kPartitionPageOffsetMask;
130 static const size_t kMaxPartitionPagesPerSlotSpan = 4;
131
132 // To avoid fragmentation via never-used freelist entries, we hand out partition
133 // freelist sections gradually, in units of the dominant system page size.
134 // What we're actually doing is avoiding filling the full partition page (16 KB)
135 // with freelist pointers right away. Writing freelist pointers will fault and
136 // dirty a private page, which is very wasteful if we never actually store
137 // objects there.
138 static const size_t kNumSystemPagesPerPartitionPage =
139 kPartitionPageSize / kSystemPageSize;
140 static const size_t kMaxSystemPagesPerSlotSpan =
141 kNumSystemPagesPerPartitionPage * kMaxPartitionPagesPerSlotSpan;
142
143 // We reserve virtual address space in 2MB chunks (aligned to 2MB as well).
144 // These chunks are called "super pages". We do this so that we can store
145 // metadata in the first few pages of each 2MB aligned section. This leads to
146 // a very fast free(). We specifically choose 2MB because this virtual address
147 // block represents a full but single PTE allocation on ARM, ia32 and x64.
148 //
149 // The layout of the super page is as follows. The sizes below are the same
150 // for 32 bit and 64 bit.
151 //
152 // | Guard page (4KB) |
153 // | Metadata page (4KB) |
154 // | Guard pages (8KB) |
155 // | Slot span |
156 // | Slot span |
157 // | ... |
158 // | Slot span |
159 // | Guard page (4KB) |
160 //
161 // - Each slot span is a contiguous range of one or more PartitionPages.
162 // - The metadata page has the following format. Note that the PartitionPage
163 // that is not at the head of a slot span is "unused". In other words,
164 // the metadata for the slot span is stored only in the first PartitionPage
165 // of the slot span. Metadata accesses to other PartitionPages are
166 // redirected to the first PartitionPage.
167 //
168 // | SuperPageExtentEntry (32B) |
169 // | PartitionPage of slot span 1 (32B, used) |
170 // | PartitionPage of slot span 1 (32B, unused) |
171 // | PartitionPage of slot span 1 (32B, unused) |
172 // | PartitionPage of slot span 2 (32B, used) |
173 // | PartitionPage of slot span 3 (32B, used) |
174 // | ... |
175 // | PartitionPage of slot span N (32B, unused) |
176 //
177 // A direct mapped page has a similar layout to fake it looking like a super
178 // page:
179 //
180 // | Guard page (4KB) |
181 // | Metadata page (4KB) |
182 // | Guard pages (8KB) |
183 // | Direct mapped object |
184 // | Guard page (4KB) |
185 //
186 // - The metadata page has the following layout:
187 //
188 // | SuperPageExtentEntry (32B) |
189 // | PartitionPage (32B) |
190 // | PartitionBucket (32B) |
191 // | PartitionDirectMapExtent (8B) |
192 static const size_t kSuperPageShift = 21; // 2MB
193 static const size_t kSuperPageSize = 1 << kSuperPageShift;
194 static const size_t kSuperPageOffsetMask = kSuperPageSize - 1;
195 static const size_t kSuperPageBaseMask = ~kSuperPageOffsetMask;
196 static const size_t kNumPartitionPagesPerSuperPage =
197 kSuperPageSize / kPartitionPageSize;
198
199 static const size_t kPageMetadataShift = 5; // 32 bytes per partition page.
200 static const size_t kPageMetadataSize = 1 << kPageMetadataShift;
201
202 // The following kGeneric* constants apply to the generic variants of the API.
203 // The "order" of an allocation is closely related to the power-of-two size of
204 // the allocation. More precisely, the order is the bit index of the
205 // most-significant-bit in the allocation size, where the bit numbers starts
206 // at index 1 for the least-significant-bit.
207 // In terms of allocation sizes, order 0 covers 0, order 1 covers 1, order 2
208 // covers 2->3, order 3 covers 4->7, order 4 covers 8->15.
209 static const size_t kGenericMinBucketedOrder = 4; // 8 bytes.
210 static const size_t kGenericMaxBucketedOrder =
211 20; // Largest bucketed order is 1<<(20-1) (storing 512KB -> almost 1MB)
212 static const size_t kGenericNumBucketedOrders =
213 (kGenericMaxBucketedOrder - kGenericMinBucketedOrder) + 1;
214 // Eight buckets per order (for the higher orders), e.g. order 8 is 128, 144,
215 // 160, ..., 240:
216 static const size_t kGenericNumBucketsPerOrderBits = 3;
217 static const size_t kGenericNumBucketsPerOrder =
218 1 << kGenericNumBucketsPerOrderBits;
219 static const size_t kGenericNumBuckets =
220 kGenericNumBucketedOrders * kGenericNumBucketsPerOrder;
221 static const size_t kGenericSmallestBucket = 1
222 << (kGenericMinBucketedOrder - 1);
223 static const size_t kGenericMaxBucketSpacing =
224 1 << ((kGenericMaxBucketedOrder - 1) - kGenericNumBucketsPerOrderBits);
225 static const size_t kGenericMaxBucketed =
226 (1 << (kGenericMaxBucketedOrder - 1)) +
227 ((kGenericNumBucketsPerOrder - 1) * kGenericMaxBucketSpacing);
228 static const size_t kGenericMinDirectMappedDownsize =
229 kGenericMaxBucketed +
230 1; // Limit when downsizing a direct mapping using realloc().
231 static const size_t kGenericMaxDirectMapped = INT_MAX - kSystemPageSize;
232 static const size_t kBitsPerSizet = sizeof(void*) * CHAR_BIT;
233
234 // Constants for the memory reclaim logic.
235 static const size_t kMaxFreeableSpans = 16;
236
237 // If the total size in bytes of allocated but not committed pages exceeds this
238 // value (probably it is a "out of virtual address space" crash),
239 // a special crash stack trace is generated at |partitionOutOfMemory|.
240 // This is to distinguish "out of virtual address space" from
241 // "out of physical memory" in crash reports.
242 static const size_t kReasonableSizeOfUnusedPages = 1024 * 1024 * 1024; // 1GiB
243
244 #if ENABLE(ASSERT)
245 // These two byte values match tcmalloc.
246 static const unsigned char kUninitializedByte = 0xAB;
247 static const unsigned char kFreedByte = 0xCD;
248 static const size_t kCookieSize =
249 16; // Handles alignment up to XMM instructions on Intel.
250 static const unsigned char kCookieValue[kCookieSize] = {
251 0xDE, 0xAD, 0xBE, 0xEF, 0xCA, 0xFE, 0xD0, 0x0D,
252 0x13, 0x37, 0xF0, 0x05, 0xBA, 0x11, 0xAB, 0x1E};
253 #endif
254
255 struct PartitionBucket;
256 struct PartitionRootBase;
257
258 struct PartitionFreelistEntry {
259 PartitionFreelistEntry* next;
260 };
261
262 // Some notes on page states. A page can be in one of four major states:
263 // 1) Active.
264 // 2) Full.
265 // 3) Empty.
266 // 4) Decommitted.
267 // An active page has available free slots. A full page has no free slots. An
268 // empty page has no free slots, and a decommitted page is an empty page that
269 // had its backing memory released back to the system.
270 // There are two linked lists tracking the pages. The "active page" list is an
271 // approximation of a list of active pages. It is an approximation because
272 // full, empty and decommitted pages may briefly be present in the list until
273 // we next do a scan over it.
274 // The "empty page" list is an accurate list of pages which are either empty
275 // or decommitted.
276 //
277 // The significant page transitions are:
278 // - free() will detect when a full page has a slot free()'d and immediately
279 // return the page to the head of the active list.
280 // - free() will detect when a page is fully emptied. It _may_ add it to the
281 // empty list or it _may_ leave it on the active list until a future list scan.
282 // - malloc() _may_ scan the active page list in order to fulfil the request.
283 // If it does this, full, empty and decommitted pages encountered will be
284 // booted out of the active list. If there are no suitable active pages found,
285 // an empty or decommitted page (if one exists) will be pulled from the empty
286 // list on to the active list.
287 struct PartitionPage {
288 PartitionFreelistEntry* freelistHead;
289 PartitionPage* nextPage;
290 PartitionBucket* bucket;
291 // Deliberately signed, 0 for empty or decommitted page, -n for full pages:
292 int16_t numAllocatedSlots;
293 uint16_t numUnprovisionedSlots;
294 uint16_t pageOffset;
295 int16_t emptyCacheIndex; // -1 if not in the empty cache.
296 };
297
298 struct PartitionBucket {
299 PartitionPage* activePagesHead; // Accessed most in hot path => goes first.
300 PartitionPage* emptyPagesHead;
301 PartitionPage* decommittedPagesHead;
302 uint32_t slotSize;
303 unsigned numSystemPagesPerSlotSpan : 8;
304 unsigned numFullPages : 24;
305 };
306
307 // An "extent" is a span of consecutive superpages. We link to the partition's
308 // next extent (if there is one) at the very start of a superpage's metadata
309 // area.
310 struct PartitionSuperPageExtentEntry {
311 PartitionRootBase* root;
312 char* superPageBase;
313 char* superPagesEnd;
314 PartitionSuperPageExtentEntry* next;
315 };
316
317 struct PartitionDirectMapExtent {
318 PartitionDirectMapExtent* nextExtent;
319 PartitionDirectMapExtent* prevExtent;
320 PartitionBucket* bucket;
321 size_t mapSize; // Mapped size, not including guard pages and meta-data.
322 };
323
324 struct WTF_EXPORT PartitionRootBase {
325 size_t totalSizeOfCommittedPages;
326 size_t totalSizeOfSuperPages;
327 size_t totalSizeOfDirectMappedPages;
328 // Invariant: totalSizeOfCommittedPages <=
329 // totalSizeOfSuperPages + totalSizeOfDirectMappedPages.
330 unsigned numBuckets;
331 unsigned maxAllocation;
332 bool initialized;
333 char* nextSuperPage;
334 char* nextPartitionPage;
335 char* nextPartitionPageEnd;
336 PartitionSuperPageExtentEntry* currentExtent;
337 PartitionSuperPageExtentEntry* firstExtent;
338 PartitionDirectMapExtent* directMapList;
339 PartitionPage* globalEmptyPageRing[kMaxFreeableSpans];
340 int16_t globalEmptyPageRingIndex;
341 uintptr_t invertedSelf;
342
343 static SpinLock gInitializedLock;
344 static bool gInitialized;
345 // gSeedPage is used as a sentinel to indicate that there is no page
346 // in the active page list. We can use nullptr, but in that case we need
347 // to add a null-check branch to the hot allocation path. We want to avoid
348 // that.
349 static PartitionPage gSeedPage;
350 static PartitionBucket gPagedBucket;
351 // gOomHandlingFunction is invoked when ParitionAlloc hits OutOfMemory.
352 static void (*gOomHandlingFunction)();
353 };
354
355 // Never instantiate a PartitionRoot directly, instead use PartitionAlloc.
356 struct PartitionRoot : public PartitionRootBase {
357 // The PartitionAlloc templated class ensures the following is correct.
358 ALWAYS_INLINE PartitionBucket* buckets() {
359 return reinterpret_cast<PartitionBucket*>(this + 1);
360 }
361 ALWAYS_INLINE const PartitionBucket* buckets() const {
362 return reinterpret_cast<const PartitionBucket*>(this + 1);
363 }
364 };
365
366 // Never instantiate a PartitionRootGeneric directly, instead use
367 // PartitionAllocatorGeneric.
368 struct PartitionRootGeneric : public PartitionRootBase {
369 SpinLock lock;
370 // Some pre-computed constants.
371 size_t orderIndexShifts[kBitsPerSizet + 1];
372 size_t orderSubIndexMasks[kBitsPerSizet + 1];
373 // The bucket lookup table lets us map a size_t to a bucket quickly.
374 // The trailing +1 caters for the overflow case for very large allocation
375 // sizes. It is one flat array instead of a 2D array because in the 2D
376 // world, we'd need to index array[blah][max+1] which risks undefined
377 // behavior.
378 PartitionBucket*
379 bucketLookups[((kBitsPerSizet + 1) * kGenericNumBucketsPerOrder) + 1];
380 PartitionBucket buckets[kGenericNumBuckets];
381 };
382
383 // Flags for partitionAllocGenericFlags.
384 enum PartitionAllocFlags {
385 PartitionAllocReturnNull = 1 << 0,
386 };
387
388 // Struct used to retrieve total memory usage of a partition. Used by
389 // PartitionStatsDumper implementation.
390 struct PartitionMemoryStats {
391 size_t totalMmappedBytes; // Total bytes mmaped from the system.
392 size_t totalCommittedBytes; // Total size of commmitted pages.
393 size_t totalResidentBytes; // Total bytes provisioned by the partition.
394 size_t totalActiveBytes; // Total active bytes in the partition.
395 size_t totalDecommittableBytes; // Total bytes that could be decommitted.
396 size_t totalDiscardableBytes; // Total bytes that could be discarded.
397 };
398
399 // Struct used to retrieve memory statistics about a partition bucket. Used by
400 // PartitionStatsDumper implementation.
401 struct PartitionBucketMemoryStats {
402 bool isValid; // Used to check if the stats is valid.
403 bool isDirectMap; // True if this is a direct mapping; size will not be
404 // unique.
405 uint32_t bucketSlotSize; // The size of the slot in bytes.
406 uint32_t allocatedPageSize; // Total size the partition page allocated from
407 // the system.
408 uint32_t activeBytes; // Total active bytes used in the bucket.
409 uint32_t residentBytes; // Total bytes provisioned in the bucket.
410 uint32_t decommittableBytes; // Total bytes that could be decommitted.
411 uint32_t discardableBytes; // Total bytes that could be discarded.
412 uint32_t numFullPages; // Number of pages with all slots allocated.
413 uint32_t numActivePages; // Number of pages that have at least one
414 // provisioned slot.
415 uint32_t numEmptyPages; // Number of pages that are empty
416 // but not decommitted.
417 uint32_t numDecommittedPages; // Number of pages that are empty
418 // and decommitted.
419 };
420
421 // Interface that is passed to partitionDumpStats and
422 // partitionDumpStatsGeneric for using the memory statistics.
423 class WTF_EXPORT PartitionStatsDumper {
424 public:
425 // Called to dump total memory used by partition, once per partition.
426 virtual void partitionDumpTotals(const char* partitionName,
427 const PartitionMemoryStats*) = 0;
428
429 // Called to dump stats about buckets, for each bucket.
430 virtual void partitionsDumpBucketStats(const char* partitionName,
431 const PartitionBucketMemoryStats*) = 0;
432 };
433
434 WTF_EXPORT void partitionAllocGlobalInit(void (*oomHandlingFunction)());
435 WTF_EXPORT void partitionAllocInit(PartitionRoot*,
436 size_t numBuckets,
437 size_t maxAllocation);
438 WTF_EXPORT bool partitionAllocShutdown(PartitionRoot*);
439 WTF_EXPORT void partitionAllocGenericInit(PartitionRootGeneric*);
440 WTF_EXPORT bool partitionAllocGenericShutdown(PartitionRootGeneric*);
441
442 enum PartitionPurgeFlags {
443 // Decommitting the ring list of empty pages is reasonably fast.
444 PartitionPurgeDecommitEmptyPages = 1 << 0,
445 // Discarding unused system pages is slower, because it involves walking all
446 // freelists in all active partition pages of all buckets >= system page
447 // size. It often frees a similar amount of memory to decommitting the empty
448 // pages, though.
449 PartitionPurgeDiscardUnusedSystemPages = 1 << 1,
450 };
451
452 WTF_EXPORT void partitionPurgeMemory(PartitionRoot*, int);
453 WTF_EXPORT void partitionPurgeMemoryGeneric(PartitionRootGeneric*, int);
454
455 WTF_EXPORT NEVER_INLINE void* partitionAllocSlowPath(PartitionRootBase*,
456 int,
457 size_t,
458 PartitionBucket*);
459 WTF_EXPORT NEVER_INLINE void partitionFreeSlowPath(PartitionPage*);
460 WTF_EXPORT NEVER_INLINE void* partitionReallocGeneric(PartitionRootGeneric*,
461 void*,
462 size_t,
463 const char* typeName);
464
465 WTF_EXPORT void partitionDumpStats(PartitionRoot*,
466 const char* partitionName,
467 bool isLightDump,
468 PartitionStatsDumper*);
469 WTF_EXPORT void partitionDumpStatsGeneric(PartitionRootGeneric*,
470 const char* partitionName,
471 bool isLightDump,
472 PartitionStatsDumper*);
473
474 class WTF_EXPORT PartitionAllocHooks {
475 public:
476 typedef void AllocationHook(void* address, size_t, const char* typeName);
477 typedef void FreeHook(void* address);
478
479 static void setAllocationHook(AllocationHook* hook) {
480 m_allocationHook = hook;
481 }
482 static void setFreeHook(FreeHook* hook) { m_freeHook = hook; }
483
484 static void allocationHookIfEnabled(void* address,
485 size_t size,
486 const char* typeName) {
487 AllocationHook* allocationHook = m_allocationHook;
488 if (UNLIKELY(allocationHook != nullptr))
489 allocationHook(address, size, typeName);
490 }
491
492 static void freeHookIfEnabled(void* address) {
493 FreeHook* freeHook = m_freeHook;
494 if (UNLIKELY(freeHook != nullptr))
495 freeHook(address);
496 }
497
498 static void reallocHookIfEnabled(void* oldAddress,
499 void* newAddress,
500 size_t size,
501 const char* typeName) {
502 // Report a reallocation as a free followed by an allocation.
503 AllocationHook* allocationHook = m_allocationHook;
504 FreeHook* freeHook = m_freeHook;
505 if (UNLIKELY(allocationHook && freeHook)) {
506 freeHook(oldAddress);
507 allocationHook(newAddress, size, typeName);
508 }
509 }
510
511 private:
512 // Pointers to hook functions that PartitionAlloc will call on allocation and
513 // free if the pointers are non-null.
514 static AllocationHook* m_allocationHook;
515 static FreeHook* m_freeHook;
516 };
517
518 // In official builds, do not include type info string literals to avoid
519 // bloating the binary.
520 #if defined(OFFICIAL_BUILD)
521 #define WTF_HEAP_PROFILER_TYPE_NAME(T) nullptr
522 #else
523 #define WTF_HEAP_PROFILER_TYPE_NAME(T) ::WTF::getStringWithTypeName<T>()
524 #endif
525
526 ALWAYS_INLINE PartitionFreelistEntry* partitionFreelistMask(
527 PartitionFreelistEntry* ptr) {
528 // We use bswap on little endian as a fast mask for two reasons:
529 // 1) If an object is freed and its vtable used where the attacker doesn't
530 // get the chance to run allocations between the free and use, the vtable
531 // dereference is likely to fault.
532 // 2) If the attacker has a linear buffer overflow and elects to try and
533 // corrupt a freelist pointer, partial pointer overwrite attacks are
534 // thwarted.
535 // For big endian, similar guarantees are arrived at with a negation.
536 #if CPU(BIG_ENDIAN)
537 uintptr_t masked = ~reinterpret_cast<uintptr_t>(ptr);
538 #else
539 uintptr_t masked = bswapuintptrt(reinterpret_cast<uintptr_t>(ptr));
540 #endif
541 return reinterpret_cast<PartitionFreelistEntry*>(masked);
542 }
543
544 ALWAYS_INLINE size_t partitionCookieSizeAdjustAdd(size_t size) {
545 #if ENABLE(ASSERT)
546 // Add space for cookies, checking for integer overflow.
547 ASSERT(size + (2 * kCookieSize) > size);
548 size += 2 * kCookieSize;
549 #endif
550 return size;
551 }
552
553 ALWAYS_INLINE size_t partitionCookieSizeAdjustSubtract(size_t size) {
554 #if ENABLE(ASSERT)
555 // Remove space for cookies.
556 ASSERT(size >= 2 * kCookieSize);
557 size -= 2 * kCookieSize;
558 #endif
559 return size;
560 }
561
562 ALWAYS_INLINE void* partitionCookieFreePointerAdjust(void* ptr) {
563 #if ENABLE(ASSERT)
564 // The value given to the application is actually just after the cookie.
565 ptr = static_cast<char*>(ptr) - kCookieSize;
566 #endif
567 return ptr;
568 }
569
570 ALWAYS_INLINE void partitionCookieWriteValue(void* ptr) {
571 #if ENABLE(ASSERT)
572 unsigned char* cookiePtr = reinterpret_cast<unsigned char*>(ptr);
573 for (size_t i = 0; i < kCookieSize; ++i, ++cookiePtr)
574 *cookiePtr = kCookieValue[i];
575 #endif
576 }
577
578 ALWAYS_INLINE void partitionCookieCheckValue(void* ptr) {
579 #if ENABLE(ASSERT)
580 unsigned char* cookiePtr = reinterpret_cast<unsigned char*>(ptr);
581 for (size_t i = 0; i < kCookieSize; ++i, ++cookiePtr)
582 ASSERT(*cookiePtr == kCookieValue[i]);
583 #endif
584 }
585
586 ALWAYS_INLINE char* partitionSuperPageToMetadataArea(char* ptr) {
587 uintptr_t pointerAsUint = reinterpret_cast<uintptr_t>(ptr);
588 ASSERT(!(pointerAsUint & kSuperPageOffsetMask));
589 // The metadata area is exactly one system page (the guard page) into the
590 // super page.
591 return reinterpret_cast<char*>(pointerAsUint + kSystemPageSize);
592 }
593
594 ALWAYS_INLINE PartitionPage* partitionPointerToPageNoAlignmentCheck(void* ptr) {
595 uintptr_t pointerAsUint = reinterpret_cast<uintptr_t>(ptr);
596 char* superPagePtr =
597 reinterpret_cast<char*>(pointerAsUint & kSuperPageBaseMask);
598 uintptr_t partitionPageIndex =
599 (pointerAsUint & kSuperPageOffsetMask) >> kPartitionPageShift;
600 // Index 0 is invalid because it is the metadata and guard area and
601 // the last index is invalid because it is a guard page.
602 ASSERT(partitionPageIndex);
603 ASSERT(partitionPageIndex < kNumPartitionPagesPerSuperPage - 1);
604 PartitionPage* page = reinterpret_cast<PartitionPage*>(
605 partitionSuperPageToMetadataArea(superPagePtr) +
606 (partitionPageIndex << kPageMetadataShift));
607 // Partition pages in the same slot span can share the same page object.
608 // Adjust for that.
609 size_t delta = page->pageOffset << kPageMetadataShift;
610 page =
611 reinterpret_cast<PartitionPage*>(reinterpret_cast<char*>(page) - delta);
612 return page;
613 }
614
615 ALWAYS_INLINE void* partitionPageToPointer(const PartitionPage* page) {
616 uintptr_t pointerAsUint = reinterpret_cast<uintptr_t>(page);
617 uintptr_t superPageOffset = (pointerAsUint & kSuperPageOffsetMask);
618 ASSERT(superPageOffset > kSystemPageSize);
619 ASSERT(superPageOffset < kSystemPageSize + (kNumPartitionPagesPerSuperPage *
620 kPageMetadataSize));
621 uintptr_t partitionPageIndex =
622 (superPageOffset - kSystemPageSize) >> kPageMetadataShift;
623 // Index 0 is invalid because it is the metadata area and the last index is
624 // invalid because it is a guard page.
625 ASSERT(partitionPageIndex);
626 ASSERT(partitionPageIndex < kNumPartitionPagesPerSuperPage - 1);
627 uintptr_t superPageBase = (pointerAsUint & kSuperPageBaseMask);
628 void* ret = reinterpret_cast<void*>(
629 superPageBase + (partitionPageIndex << kPartitionPageShift));
630 return ret;
631 }
632
633 ALWAYS_INLINE PartitionPage* partitionPointerToPage(void* ptr) {
634 PartitionPage* page = partitionPointerToPageNoAlignmentCheck(ptr);
635 // Checks that the pointer is a multiple of bucket size.
636 ASSERT(!((reinterpret_cast<uintptr_t>(ptr) -
637 reinterpret_cast<uintptr_t>(partitionPageToPointer(page))) %
638 page->bucket->slotSize));
639 return page;
640 }
641
642 ALWAYS_INLINE bool partitionBucketIsDirectMapped(
643 const PartitionBucket* bucket) {
644 return !bucket->numSystemPagesPerSlotSpan;
645 }
646
647 ALWAYS_INLINE size_t partitionBucketBytes(const PartitionBucket* bucket) {
648 return bucket->numSystemPagesPerSlotSpan * kSystemPageSize;
649 }
650
651 ALWAYS_INLINE uint16_t partitionBucketSlots(const PartitionBucket* bucket) {
652 return static_cast<uint16_t>(partitionBucketBytes(bucket) / bucket->slotSize);
653 }
654
655 ALWAYS_INLINE size_t* partitionPageGetRawSizePtr(PartitionPage* page) {
656 // For single-slot buckets which span more than one partition page, we
657 // have some spare metadata space to store the raw allocation size. We
658 // can use this to report better statistics.
659 PartitionBucket* bucket = page->bucket;
660 if (bucket->slotSize <= kMaxSystemPagesPerSlotSpan * kSystemPageSize)
661 return nullptr;
662
663 ASSERT((bucket->slotSize % kSystemPageSize) == 0);
664 ASSERT(partitionBucketIsDirectMapped(bucket) ||
665 partitionBucketSlots(bucket) == 1);
666 page++;
667 return reinterpret_cast<size_t*>(&page->freelistHead);
668 }
669
670 ALWAYS_INLINE size_t partitionPageGetRawSize(PartitionPage* page) {
671 size_t* rawSizePtr = partitionPageGetRawSizePtr(page);
672 if (UNLIKELY(rawSizePtr != nullptr))
673 return *rawSizePtr;
674 return 0;
675 }
676
677 ALWAYS_INLINE PartitionRootBase* partitionPageToRoot(PartitionPage* page) {
678 PartitionSuperPageExtentEntry* extentEntry =
679 reinterpret_cast<PartitionSuperPageExtentEntry*>(
680 reinterpret_cast<uintptr_t>(page) & kSystemPageBaseMask);
681 return extentEntry->root;
682 }
683
684 ALWAYS_INLINE bool partitionPointerIsValid(void* ptr) {
685 PartitionPage* page = partitionPointerToPage(ptr);
686 PartitionRootBase* root = partitionPageToRoot(page);
687 return root->invertedSelf == ~reinterpret_cast<uintptr_t>(root);
688 }
689
690 ALWAYS_INLINE void* partitionBucketAlloc(PartitionRootBase* root,
691 int flags,
692 size_t size,
693 PartitionBucket* bucket) {
694 PartitionPage* page = bucket->activePagesHead;
695 // Check that this page is neither full nor freed.
696 ASSERT(page->numAllocatedSlots >= 0);
697 void* ret = page->freelistHead;
698 if (LIKELY(ret != 0)) {
699 // If these asserts fire, you probably corrupted memory.
700 ASSERT(partitionPointerIsValid(ret));
701 // All large allocations must go through the slow path to correctly
702 // update the size metadata.
703 ASSERT(partitionPageGetRawSize(page) == 0);
704 PartitionFreelistEntry* newHead =
705 partitionFreelistMask(static_cast<PartitionFreelistEntry*>(ret)->next);
706 page->freelistHead = newHead;
707 page->numAllocatedSlots++;
708 } else {
709 ret = partitionAllocSlowPath(root, flags, size, bucket);
710 ASSERT(!ret || partitionPointerIsValid(ret));
711 }
712 #if ENABLE(ASSERT)
713 if (!ret)
714 return 0;
715 // Fill the uninitialized pattern, and write the cookies.
716 page = partitionPointerToPage(ret);
717 size_t slotSize = page->bucket->slotSize;
718 size_t rawSize = partitionPageGetRawSize(page);
719 if (rawSize) {
720 ASSERT(rawSize == size);
721 slotSize = rawSize;
722 }
723 size_t noCookieSize = partitionCookieSizeAdjustSubtract(slotSize);
724 char* charRet = static_cast<char*>(ret);
725 // The value given to the application is actually just after the cookie.
726 ret = charRet + kCookieSize;
727 memset(ret, kUninitializedByte, noCookieSize);
728 partitionCookieWriteValue(charRet);
729 partitionCookieWriteValue(charRet + kCookieSize + noCookieSize);
730 #endif
731 return ret;
732 }
733
734 ALWAYS_INLINE void* partitionAlloc(PartitionRoot* root,
735 size_t size,
736 const char* typeName) {
737 #if defined(MEMORY_TOOL_REPLACES_ALLOCATOR)
738 void* result = malloc(size);
739 RELEASE_ASSERT(result);
740 return result;
741 #else
742 size_t requestedSize = size;
743 size = partitionCookieSizeAdjustAdd(size);
744 ASSERT(root->initialized);
745 size_t index = size >> kBucketShift;
746 ASSERT(index < root->numBuckets);
747 ASSERT(size == index << kBucketShift);
748 PartitionBucket* bucket = &root->buckets()[index];
749 void* result = partitionBucketAlloc(root, 0, size, bucket);
750 PartitionAllocHooks::allocationHookIfEnabled(result, requestedSize, typeName);
751 return result;
752 #endif // defined(MEMORY_TOOL_REPLACES_ALLOCATOR)
753 }
754
755 ALWAYS_INLINE void partitionFreeWithPage(void* ptr, PartitionPage* page) {
756 // If these asserts fire, you probably corrupted memory.
757 #if ENABLE(ASSERT)
758 size_t slotSize = page->bucket->slotSize;
759 size_t rawSize = partitionPageGetRawSize(page);
760 if (rawSize)
761 slotSize = rawSize;
762 partitionCookieCheckValue(ptr);
763 partitionCookieCheckValue(reinterpret_cast<char*>(ptr) + slotSize -
764 kCookieSize);
765 memset(ptr, kFreedByte, slotSize);
766 #endif
767 ASSERT(page->numAllocatedSlots);
768 PartitionFreelistEntry* freelistHead = page->freelistHead;
769 ASSERT(!freelistHead || partitionPointerIsValid(freelistHead));
770 SECURITY_CHECK(ptr != freelistHead); // Catches an immediate double free.
771 // Look for double free one level deeper in debug.
772 SECURITY_DCHECK(!freelistHead ||
773 ptr != partitionFreelistMask(freelistHead->next));
774 PartitionFreelistEntry* entry = static_cast<PartitionFreelistEntry*>(ptr);
775 entry->next = partitionFreelistMask(freelistHead);
776 page->freelistHead = entry;
777 --page->numAllocatedSlots;
778 if (UNLIKELY(page->numAllocatedSlots <= 0)) {
779 partitionFreeSlowPath(page);
780 } else {
781 // All single-slot allocations must go through the slow path to
782 // correctly update the size metadata.
783 ASSERT(partitionPageGetRawSize(page) == 0);
784 }
785 }
786
787 ALWAYS_INLINE void partitionFree(void* ptr) {
788 #if defined(MEMORY_TOOL_REPLACES_ALLOCATOR)
789 free(ptr);
790 #else
791 PartitionAllocHooks::freeHookIfEnabled(ptr);
792 ptr = partitionCookieFreePointerAdjust(ptr);
793 ASSERT(partitionPointerIsValid(ptr));
794 PartitionPage* page = partitionPointerToPage(ptr);
795 partitionFreeWithPage(ptr, page);
796 #endif
797 }
798
799 ALWAYS_INLINE PartitionBucket* partitionGenericSizeToBucket(
800 PartitionRootGeneric* root,
801 size_t size) {
802 size_t order = kBitsPerSizet - CountLeadingZeroBitsSizeT(size);
803 // The order index is simply the next few bits after the most significant bit.
804 size_t orderIndex = (size >> root->orderIndexShifts[order]) &
805 (kGenericNumBucketsPerOrder - 1);
806 // And if the remaining bits are non-zero we must bump the bucket up.
807 size_t subOrderIndex = size & root->orderSubIndexMasks[order];
808 PartitionBucket* bucket =
809 root->bucketLookups[(order << kGenericNumBucketsPerOrderBits) +
810 orderIndex + !!subOrderIndex];
811 ASSERT(!bucket->slotSize || bucket->slotSize >= size);
812 ASSERT(!(bucket->slotSize % kGenericSmallestBucket));
813 return bucket;
814 }
815
816 ALWAYS_INLINE void* partitionAllocGenericFlags(PartitionRootGeneric* root,
817 int flags,
818 size_t size,
819 const char* typeName) {
820 #if defined(MEMORY_TOOL_REPLACES_ALLOCATOR)
821 void* result = malloc(size);
822 RELEASE_ASSERT(result || flags & PartitionAllocReturnNull);
823 return result;
824 #else
825 ASSERT(root->initialized);
826 size_t requestedSize = size;
827 size = partitionCookieSizeAdjustAdd(size);
828 PartitionBucket* bucket = partitionGenericSizeToBucket(root, size);
829 void* ret = nullptr;
830 {
831 SpinLock::Guard guard(root->lock);
832 ret = partitionBucketAlloc(root, flags, size, bucket);
833 }
834 PartitionAllocHooks::allocationHookIfEnabled(ret, requestedSize, typeName);
835 return ret;
836 #endif
837 }
838
839 ALWAYS_INLINE void* partitionAllocGeneric(PartitionRootGeneric* root,
840 size_t size,
841 const char* typeName) {
842 return partitionAllocGenericFlags(root, 0, size, typeName);
843 }
844
845 ALWAYS_INLINE void partitionFreeGeneric(PartitionRootGeneric* root, void* ptr) {
846 #if defined(MEMORY_TOOL_REPLACES_ALLOCATOR)
847 free(ptr);
848 #else
849 ASSERT(root->initialized);
850
851 if (UNLIKELY(!ptr))
852 return;
853
854 PartitionAllocHooks::freeHookIfEnabled(ptr);
855 ptr = partitionCookieFreePointerAdjust(ptr);
856 ASSERT(partitionPointerIsValid(ptr));
857 PartitionPage* page = partitionPointerToPage(ptr);
858 {
859 SpinLock::Guard guard(root->lock);
860 partitionFreeWithPage(ptr, page);
861 }
862 #endif
863 }
864
865 ALWAYS_INLINE size_t partitionDirectMapSize(size_t size) {
866 // Caller must check that the size is not above the kGenericMaxDirectMapped
867 // limit before calling. This also guards against integer overflow in the
868 // calculation here.
869 ASSERT(size <= kGenericMaxDirectMapped);
870 return (size + kSystemPageOffsetMask) & kSystemPageBaseMask;
871 }
872
873 ALWAYS_INLINE size_t partitionAllocActualSize(PartitionRootGeneric* root,
874 size_t size) {
875 #if defined(MEMORY_TOOL_REPLACES_ALLOCATOR)
876 return size;
877 #else
878 ASSERT(root->initialized);
879 size = partitionCookieSizeAdjustAdd(size);
880 PartitionBucket* bucket = partitionGenericSizeToBucket(root, size);
881 if (LIKELY(!partitionBucketIsDirectMapped(bucket))) {
882 size = bucket->slotSize;
883 } else if (size > kGenericMaxDirectMapped) {
884 // Too large to allocate => return the size unchanged.
885 } else {
886 ASSERT(bucket == &PartitionRootBase::gPagedBucket);
887 size = partitionDirectMapSize(size);
888 }
889 return partitionCookieSizeAdjustSubtract(size);
890 #endif
891 }
892
893 ALWAYS_INLINE bool partitionAllocSupportsGetSize() {
894 #if defined(MEMORY_TOOL_REPLACES_ALLOCATOR)
895 return false;
896 #else
897 return true;
898 #endif
899 }
900
901 ALWAYS_INLINE size_t partitionAllocGetSize(void* ptr) {
902 // No need to lock here. Only 'ptr' being freed by another thread could
903 // cause trouble, and the caller is responsible for that not happening.
904 ASSERT(partitionAllocSupportsGetSize());
905 ptr = partitionCookieFreePointerAdjust(ptr);
906 ASSERT(partitionPointerIsValid(ptr));
907 PartitionPage* page = partitionPointerToPage(ptr);
908 size_t size = page->bucket->slotSize;
909 return partitionCookieSizeAdjustSubtract(size);
910 }
911
912 // N (or more accurately, N - sizeof(void*)) represents the largest size in
913 // bytes that will be handled by a SizeSpecificPartitionAllocator.
914 // Attempts to partitionAlloc() more than this amount will fail.
915 template <size_t N>
916 class SizeSpecificPartitionAllocator {
917 public:
918 static const size_t kMaxAllocation = N - kAllocationGranularity;
919 static const size_t kNumBuckets = N / kAllocationGranularity;
920 void init() {
921 partitionAllocInit(&m_partitionRoot, kNumBuckets, kMaxAllocation);
922 }
923 bool shutdown() { return partitionAllocShutdown(&m_partitionRoot); }
924 ALWAYS_INLINE PartitionRoot* root() { return &m_partitionRoot; }
925
926 private:
927 PartitionRoot m_partitionRoot;
928 PartitionBucket m_actualBuckets[kNumBuckets];
929 };
930
931 class PartitionAllocatorGeneric {
932 public:
933 void init() { partitionAllocGenericInit(&m_partitionRoot); }
934 bool shutdown() { return partitionAllocGenericShutdown(&m_partitionRoot); }
935 ALWAYS_INLINE PartitionRootGeneric* root() { return &m_partitionRoot; }
936
937 private:
938 PartitionRootGeneric m_partitionRoot;
939 };
940
941 } // namespace WTF
942
943 using WTF::SizeSpecificPartitionAllocator;
944 using WTF::PartitionAllocatorGeneric;
945 using WTF::PartitionRoot;
946 using WTF::partitionAllocInit;
947 using WTF::partitionAllocShutdown;
948 using WTF::partitionAlloc;
949 using WTF::partitionFree;
950 using WTF::partitionAllocGeneric;
951 using WTF::partitionFreeGeneric;
952 using WTF::partitionReallocGeneric;
953 using WTF::partitionAllocActualSize;
954 using WTF::partitionAllocSupportsGetSize;
955 using WTF::partitionAllocGetSize;
956
957 #endif // WTF_PartitionAlloc_h
OLDNEW

Powered by Google App Engine
This is Rietveld 408576698