1========================
2Scudo Hardened Allocator
3========================
4
5.. contents::
6   :local:
7   :depth: 1
8
9Introduction
10============
11
12The Scudo Hardened Allocator is a user-mode allocator based on LLVM Sanitizer's
13CombinedAllocator, which aims at providing additional mitigations against heap
14based vulnerabilities, while maintaining good performance.
15
16Currently, the allocator supports (was tested on) the following architectures:
17
18- i386 (& i686) (32-bit);
19- x86_64 (64-bit);
20- armhf (32-bit);
21- AArch64 (64-bit);
22- MIPS (32-bit & 64-bit).
23
24The name "Scudo" has been retained from the initial implementation (Escudo
25meaning Shield in Spanish and Portuguese).
26
27Design
28======
29
30Allocator
31---------
32Scudo can be considered a Frontend to the Sanitizers' common allocator (later
33referenced as the Backend). It is split between a Primary allocator, fast and
34efficient, that services smaller allocation sizes, and a Secondary allocator
35that services larger allocation sizes and is backed by the operating system
36memory mapping primitives.
37
38Scudo was designed with security in mind, but aims at striking a good balance
39between security and performance. It is highly tunable and configurable.
40
41Chunk Header
42------------
43Every chunk of heap memory will be preceded by a chunk header. This has two
44purposes, the first one being to store various information about the chunk,
45the second one being to detect potential heap overflows. In order to achieve
46this, the header will be checksummed, involving the pointer to the chunk itself
47and a global secret. Any corruption of the header will be detected when said
48header is accessed, and the process terminated.
49
50The following information is stored in the header:
51
52- the 16-bit checksum;
53- the class ID for that chunk, which is the "bucket" where the chunk resides
54  for Primary backed allocations, or 0 for Secondary backed allocations;
55- the size (Primary) or unused bytes amount (Secondary) for that chunk, which is
56  necessary for computing the size of the chunk;
57- the state of the chunk (available, allocated or quarantined);
58- the allocation type (malloc, new, new[] or memalign), to detect potential
59  mismatches in the allocation APIs used;
60- the offset of the chunk, which is the distance in bytes from the beginning of
61  the returned chunk to the beginning of the Backend allocation;
62
63This header fits within 8 bytes, on all platforms supported.
64
65The checksum is computed as a CRC32 (made faster with hardware support)
66of the global secret, the chunk pointer itself, and the 8 bytes of header with
67the checksum field zeroed out. It is not intended to be cryptographically
68strong.
69
70The header is atomically loaded and stored to prevent races. This is important
71as two consecutive chunks could belong to different threads. We also want to
72avoid any type of double fetches of information located in the header, and use
73local copies of the header for this purpose.
74
75Delayed Freelist
76-----------------
77A delayed freelist allows us to not return a chunk directly to the Backend, but
78to keep it aside for a while. Once a criterion is met, the delayed freelist is
79emptied, and the quarantined chunks are returned to the Backend. This helps
80mitigate use-after-free vulnerabilities by reducing the determinism of the
81allocation and deallocation patterns.
82
83This feature is using the Sanitizer's Quarantine as its base, and the amount of
84memory that it can hold is configurable by the user (see the Options section
85below).
86
87Randomness
88----------
89It is important for the allocator to not make use of fixed addresses. We use
90the dynamic base option for the SizeClassAllocator, allowing us to benefit
91from the randomness of the system memory mapping functions.
92
93Usage
94=====
95
96Library
97-------
98The allocator static library can be built from the LLVM build tree thanks to
99the ``scudo`` CMake rule. The associated tests can be exercised thanks to the
100``check-scudo`` CMake rule.
101
102Linking the static library to your project can require the use of the
103``whole-archive`` linker flag (or equivalent), depending on your linker.
104Additional flags might also be necessary.
105
106Your linked binary should now make use of the Scudo allocation and deallocation
107functions.
108
109You may also build Scudo like this:
110
111.. code:: console
112
113  cd $LLVM/projects/compiler-rt/lib
114  clang++ -fPIC -std=c++11 -msse4.2 -O2 -I. scudo/*.cpp \
115    $(\ls sanitizer_common/*.{cc,S} | grep -v "sanitizer_termination\|sanitizer_common_nolibc\|sancov_\|sanitizer_unwind\|sanitizer_symbol") \
116    -shared -o libscudo.so -pthread
117
118and then use it with existing binaries as follows:
119
120.. code:: console
121
122  LD_PRELOAD=`pwd`/libscudo.so ./a.out
123
124Clang
125-----
126With a recent version of Clang (post rL317337), the allocator can be linked with
127a binary at compilation using the ``-fsanitize=scudo`` command-line argument, if
128the target platform is supported. Currently, the only other Sanitizer Scudo is
129compatible with is UBSan (eg: ``-fsanitize=scudo,undefined``). Compiling with
130Scudo will also enforce PIE for the output binary.
131
132Options
133-------
134Several aspects of the allocator can be configured on a per process basis
135through the following ways:
136
137- at compile time, by defining ``SCUDO_DEFAULT_OPTIONS`` to the options string
138  you want set by default;
139
140- by defining a ``__scudo_default_options`` function in one's program that
141  returns the options string to be parsed. Said function must have the following
142  prototype: ``extern "C" const char* __scudo_default_options(void)``, with a
143  default visibility. This will override the compile time define;
144
145- through the environment variable SCUDO_OPTIONS, containing the options string
146  to be parsed. Options defined this way will override any definition made
147  through ``__scudo_default_options``.
148
149The options string follows a syntax similar to ASan, where distinct options
150can be assigned in the same string, separated by colons.
151
152For example, using the environment variable:
153
154.. code:: console
155
156  SCUDO_OPTIONS="DeleteSizeMismatch=1:QuarantineSizeKb=64" ./a.out
157
158Or using the function:
159
160.. code:: cpp
161
162  extern "C" const char *__scudo_default_options() {
163    return "DeleteSizeMismatch=1:QuarantineSizeKb=64";
164  }
165
166
167The following options are available:
168
169+-----------------------------+----------------+----------------+------------------------------------------------+
170| Option                      | 64-bit default | 32-bit default | Description                                    |
171+-----------------------------+----------------+----------------+------------------------------------------------+
172| QuarantineSizeKb            | 256            | 64             | The size (in Kb) of quarantine used to delay   |
173|                             |                |                | the actual deallocation of chunks. Lower value |
174|                             |                |                | may reduce memory usage but decrease the       |
175|                             |                |                | effectiveness of the mitigation; a negative    |
176|                             |                |                | value will fallback to the defaults. Setting   |
177|                             |                |                | *both* this and ThreadLocalQuarantineSizeKb to |
178|                             |                |                | zero will disable the quarantine entirely.     |
179+-----------------------------+----------------+----------------+------------------------------------------------+
180| QuarantineChunksUpToSize    | 2048           | 512            | Size (in bytes) up to which chunks can be      |
181|                             |                |                | quarantined.                                   |
182+-----------------------------+----------------+----------------+------------------------------------------------+
183| ThreadLocalQuarantineSizeKb | 1024           | 256            | The size (in Kb) of per-thread cache use to    |
184|                             |                |                | offload the global quarantine. Lower value may |
185|                             |                |                | reduce memory usage but might increase         |
186|                             |                |                | contention on the global quarantine. Setting   |
187|                             |                |                | *both* this and QuarantineSizeKb to zero will  |
188|                             |                |                | disable the quarantine entirely.               |
189+-----------------------------+----------------+----------------+------------------------------------------------+
190| DeallocationTypeMismatch    | true           | true           | Whether or not we report errors on             |
191|                             |                |                | malloc/delete, new/free, new/delete[], etc.    |
192+-----------------------------+----------------+----------------+------------------------------------------------+
193| DeleteSizeMismatch          | true           | true           | Whether or not we report errors on mismatch    |
194|                             |                |                | between sizes of new and delete.               |
195+-----------------------------+----------------+----------------+------------------------------------------------+
196| ZeroContents                | false          | false          | Whether or not we zero chunk contents on       |
197|                             |                |                | allocation and deallocation.                   |
198+-----------------------------+----------------+----------------+------------------------------------------------+
199
200Allocator related common Sanitizer options can also be passed through Scudo
201options, such as ``allocator_may_return_null`` or ``abort_on_error``. A detailed
202list including those can be found here:
203https://github.com/google/sanitizers/wiki/SanitizerCommonFlags.
204
205Error Types
206===========
207
208The allocator will output an error message, and potentially terminate the
209process, when an unexpected behavior is detected. The output usually starts with
210``"Scudo ERROR:"`` followed by a short summary of the problem that occurred as
211well as the pointer(s) involved. Once again, Scudo is meant to be a mitigation,
212and might not be the most useful of tools to help you root-cause the issue,
213please consider `ASan <https://github.com/google/sanitizers/wiki/AddressSanitizer>`_
214for this purpose.
215
216Here is a list of the current error messages and their potential cause:
217
218- ``"corrupted chunk header"``: the checksum verification of the chunk header
219  has failed. This is likely due to one of two things: the header was
220  overwritten (partially or totally), or the pointer passed to the function is
221  not a chunk at all;
222
223- ``"race on chunk header"``: two different threads are attempting to manipulate
224  the same header at the same time. This is usually symptomatic of a
225  race-condition or general lack of locking when performing operations on that
226  chunk;
227
228- ``"invalid chunk state"``: the chunk is not in the expected state for a given
229  operation, eg: it is not allocated when trying to free it, or it's not
230  quarantined when trying to recycle it, etc. A double-free is the typical
231  reason this error would occur;
232
233- ``"misaligned pointer"``: we strongly enforce basic alignment requirements, 8
234  bytes on 32-bit platforms, 16 bytes on 64-bit platforms. If a pointer passed
235  to our functions does not fit those, something is definitely wrong.
236
237- ``"allocation type mismatch"``: when the optional deallocation type mismatch
238  check is enabled, a deallocation function called on a chunk has to match the
239  type of function that was called to allocate it. Security implications of such
240  a mismatch are not necessarily obvious but situational at best;
241
242- ``"invalid sized delete"``: when the C++14 sized delete operator is used, and
243  the optional check enabled, this indicates that the size passed when
244  deallocating a chunk is not congruent with the one requested when allocating
245  it. This is likely to be a `compiler issue <https://software.intel.com/en-us/forums/intel-c-compiler/topic/783942>`_,
246  as was the case with Intel C++ Compiler, or some type confusion on the object
247  being deallocated;
248
249- ``"RSS limit exhausted"``: the maximum RSS optionally specified has been
250  exceeded;
251
252Several other error messages relate to parameter checking on the libc allocation
253APIs and are fairly straightforward to understand.
254