1 ========================
2 Scudo Hardened Allocator
3 ========================
11 The Scudo Hardened Allocator is a user-mode allocator based on LLVM Sanitizer's
12 CombinedAllocator, which aims at providing additional mitigations against heap
13 based vulnerabilities, while maintaining good performance.
15 The name "Scudo" has been retained from the initial implementation (Escudo
16 meaning Shield in Spanish and Portuguese).
22 Every chunk of heap memory will be preceded by a chunk header. This has two
23 purposes, the first one being to store various information about the chunk,
24 the second one being to detect potential heap overflows. In order to achieve
25 this, the header will be checksumed, involving the pointer to the chunk itself
26 and a global secret. Any corruption of the header will be detected when said
27 header is accessed, and the process terminated.
29 The following information is stored in the header:
31 - the 16-bit checksum;
32 - the user requested size for that chunk, which is necessary for reallocation
34 - the state of the chunk (available, allocated or quarantined);
35 - the allocation type (malloc, new, new[] or memalign), to detect potential
36 mismatches in the allocation APIs used;
37 - whether or not the chunk is offseted (ie: if the chunk beginning is different
38 than the backend allocation beginning, which is most often the case with some
40 - the associated offset;
43 On x64, which is currently the only architecture supported, the header fits
44 within 16-bytes, which works nicely with the minimum alignment requirements.
46 The checksum is computed as a CRC32 (requiring the SSE 4.2 instruction set)
47 of the global secret, the chunk pointer itself, and the 16 bytes of header with
48 the checksum field zeroed out.
50 The header is atomically loaded and stored to prevent races (this requires
51 platform support such as the cmpxchg16b instruction). This is important as two
52 consecutive chunks could belong to different threads. We also want to avoid
53 any type of double fetches of information located in the header, and use local
54 copies of the header for this purpose.
58 A delayed freelist allows us to not return a chunk directly to the backend, but
59 to keep it aside for a while. Once a criterion is met, the delayed freelist is
60 emptied, and the quarantined chunks are returned to the backend. This helps
61 mitigate use-after-free vulnerabilities by reducing the determinism of the
62 allocation and deallocation patterns.
64 This feature is using the Sanitizer's Quarantine as its base, and the amount of
65 memory that it can hold is configurable by the user (see the Options section
70 It is important for the allocator to not make use of fixed addresses. We use
71 the dynamic base option for the SizeClassAllocator, allowing us to benefit
72 from the randomness of mmap.
79 The allocator static library can be built from the LLVM build tree thanks to
80 the "scudo" CMake rule. The associated tests can be exercised thanks to the
81 "check-scudo" CMake rule.
83 Linking the static library to your project can require the use of the
84 "whole-archive" linker flag (or equivalent), depending on your linker.
85 Additional flags might also be necessary.
87 Your linked binary should now make use of the Scudo allocation and deallocation
92 Several aspects of the allocator can be configured through environment options,
93 following the usual ASan options syntax, through the variable SCUDO_OPTIONS.
95 For example: SCUDO_OPTIONS="DeleteSizeMismatch=1:QuarantineSizeMb=16".
97 The following options are available:
99 - QuarantineSizeMb (integer, defaults to 64): the size (in Mb) of quarantine
100 used to delay the actual deallocation of chunks. Lower value may reduce
101 memory usage but decrease the effectiveness of the mitigation; a negative
102 value will fallback to a default of 64Mb;
104 - ThreadLocalQuarantineSizeKb (integer, default to 1024): the size (in Kb) of
105 per-thread cache used to offload the global quarantine. Lower value may
106 reduce memory usage but might increase the contention on the global
109 - DeallocationTypeMismatch (boolean, defaults to true): whether or not we report
110 errors on malloc/delete, new/free, new/delete[], etc;
112 - DeleteSizeMismatch (boolean, defaults to true): whether or not we report
113 errors on mismatch between size of new and delete;
115 - ZeroContents (boolean, defaults to false): whether or not we zero chunk
116 contents on allocation and deallocation.