README.md
1This document describes how malloc / new calls are routed in the various Chrome
2platforms.
3
4Bear in mind that the chromium codebase does not always just use `malloc()`.
5Some examples:
6 - Large parts of the renderer (Blink) use two home-brewed allocators,
7 PartitionAlloc and BlinkGC (Oilpan).
8 - Some subsystems, such as the V8 JavaScript engine, handle memory management
9 autonomously.
10 - Various parts of the codebase use abstractions such as `SharedMemory` or
11 `DiscardableMemory` which, similarly to the above, have their own page-level
12 memory management.
13
14Background
15----------
16The `allocator` target defines at compile-time the platform-specific choice of
17the allocator and extra-hooks which services calls to malloc/new. The relevant
18build-time flags involved are `use_allocator_shim` and
19`use_partition_alloc_as_malloc`.
20
21By default, these are true on all platforms except iOS (not yet supported) and
22NaCl (no plan to support).
23Furthermore, when building with a sanitizer (e.g. `asan`, `msan`, ...) both the
24allocator and the shim layer are disabled.
25
26
27Layering and build deps
28-----------------------
29The `allocator` target provides the linker flags required for the Windows shim
30layer. The `base` target is (almost) the only one depending on `allocator`. No
31other targets should depend on it, with the exception of the very few
32executables / dynamic libraries that don't depend, either directly or
33indirectly, on `base` within the scope of a linker unit.
34
35More importantly, **no other place outside of `/base` should depend on the
36specific allocator**.
37If such a functional dependency is required that should be achieved using
38abstractions in `base` (see `/base/memory/`)
39
40**Why `base` depends on `allocator`?**
41Because it needs to provide services that depend on the actual allocator
42implementation. In the past `base` used to pretend to be allocator-agnostic
43and get the dependencies injected by other layers. This ended up being an
44inconsistent mess.
45See the [allocator cleanup doc][url-allocator-cleanup] for more context.
46
47Linker unit targets (executables and shared libraries) that depend in some way
48on `base` (most of the targets in the codebase) automatically get the correct
49set of linker flags to pull in the Windows shim-layer (if needed).
50
51
52Source code
53-----------
54This directory contains just the allocator (i.e. shim) layer that switches
55between the different underlying memory allocation implementations.
56
57
58Unified allocator shim
59----------------------
60On most platforms, Chrome overrides the malloc / operator new symbols (and
61corresponding free / delete and other variants). This is to enforce security
62checks and lately to enable the
63[memory-infra heap profiler][url-memory-infra-heap-profiler].
64Historically each platform had its special logic for defining the allocator
65symbols in different places of the codebase. The unified allocator shim is
66a project aimed to unify the symbol definition and allocator routing logic in
67a central place.
68
69 - Full documentation: [Allocator shim design doc][url-allocator-shim].
70 - Current state: Available and enabled by default on Android, CrOS, Linux,
71 Mac OS and Windows.
72 - Tracking bug: [crbug.com/550886](https://crbug.com/550886).
73 - Build-time flag: `use_allocator_shim`.
74
75**Overview of the unified allocator shim**
76The allocator shim consists of three stages:
77```
78+-------------------------+ +-----------------------+ +----------------+
79| malloc & friends | -> | shim layer | -> | Routing to |
80| symbols definition | | implementation | | allocator |
81+-------------------------+ +-----------------------+ +----------------+
82| - libc symbols (malloc, | | - Security checks | | - glibc |
83| calloc, free, ...) | | - Chain of dispatchers| | - Android |
84| - C++ symbols (operator | | that can intercept | | bionic |
85| new, delete, ...) | | and override | | - WinHeap |
86| - glibc weak symbols | | allocations | | - Partition |
87| (__libc_malloc, ...) | +-----------------------+ | Alloc |
88+-------------------------+ +----------------+
89```
90
91**1. malloc symbols definition**
92This stage takes care of overriding the symbols `malloc`, `free`,
93`operator new`, `operator delete` and friends and routing those calls inside the
94allocator shim (next point).
95This is taken care of by the headers in `allocator_shim_override_*`.
96
97*On Windows*: Windows' UCRT (Universal C Runtime) exports weak symbols, that we
98can override in `allocator_shim_override_ucrt_symbols_win.h`.
99
100*On Linux/CrOS*: the allocator symbols are defined as exported global symbols
101in `allocator_shim_override_libc_symbols.h` (for `malloc`, `free` and friends)
102and in `allocator_shim_override_cpp_symbols.h` (for `operator new`,
103`operator delete` and friends).
104This enables proper interposition of malloc symbols referenced by the main
105executable and any third party libraries. Symbol resolution on Linux is a breadth first search that starts from the root link unit, that is the executable
106(see EXECUTABLE AND LINKABLE FORMAT (ELF) - Portable Formats Specification).
107The Linux/CrOS shim was introduced by
108[crrev.com/1675143004](https://crrev.com/1675143004).
109
110*On Android*: load-time symbol interposition (unlike the Linux/CrOS case) is not
111possible. This is because Android processes are `fork()`-ed from the Android
112zygote, which pre-loads libc.so and only later native code gets loaded via
113`dlopen()` (symbols from `dlopen()`-ed libraries get a different resolution
114scope).
115In this case, the approach instead of wrapping symbol resolution at link time
116(i.e. during the build), via the `--Wl,-wrap,malloc` linker flag.
117The use of this wrapping flag causes:
118 - All references to allocator symbols in the Chrome codebase to be rewritten as
119 references to `__wrap_malloc` and friends. The `__wrap_malloc` symbols are
120 defined in the `allocator_shim_override_linker_wrapped_symbols.h` and
121 route allocator calls inside the shim layer.
122 - The reference to the original `malloc` symbols (which typically is defined by
123 the system's libc.so) are accessible via the special `__real_malloc` and
124 friends symbols (which will be relocated, at load time, against `malloc`).
125
126In summary, this approach is transparent to the dynamic loader, which still sees
127undefined symbol references to malloc symbols.
128These symbols will be resolved against libc.so as usual.
129More details in [crrev.com/1719433002](https://crrev.com/1719433002).
130
131**2. Shim layer implementation**
132This stage contains the actual shim implementation. This consists of:
133- A singly linked list of dispatchers (structs with function pointers to `malloc`-like functions). Dispatchers can be dynamically inserted at runtime
134(using the `InsertAllocatorDispatch` API). They can intercept and override
135allocator calls.
136- The security checks (suicide on malloc-failure via `std::new_handler`, etc).
137This happens inside `allocator_shim.cc`
138
139**3. Final allocator routing**
140The final element of the aforementioned dispatcher chain is statically defined
141at build time and ultimately routes the allocator calls to the actual allocator
142(as described in the *Background* section above). This is taken care of by the
143headers in `allocator_shim_default_dispatch_to_*` files.
144
145
146Related links
147-------------
148- [Unified allocator shim doc - Feb 2016][url-allocator-shim]
149- [Allocator cleanup doc - Jan 2016][url-allocator-cleanup]
150- [Proposal to use PartitionAlloc as default allocator](https://crbug.com/339604)
151- [Memory-Infra: Tools to profile memory usage in Chrome](/docs/memory-infra/README.md)
152
153[url-allocator-cleanup]: https://docs.google.com/document/d/1V77Kgp_4tfaaWPEZVxNevoD02wXiatnAv7Ssgr0hmjg/edit?usp=sharing
154[url-memory-infra-heap-profiler]: /docs/memory-infra/heap_profiler.md
155[url-allocator-shim]: https://docs.google.com/document/d/1yKlO1AO4XjpDad9rjcBOI15EKdAGsuGO_IeZy0g0kxo/edit?usp=sharing
156