FireDBG for Rust

Sub-systems

For more information on the architecture of FireDBG, please read Architecture of FireDBG.

codelldb

lldb binding; forked from codelldb

command

FireDBG command line interface

debugger

Debugging engine of FireDBG

indexer

Transforms .firedbg.ss into .sqlite file

library

FireDBG Support Library

parser

Parses source files in workspace

protocol

Event stream protocol of .firedbg.ss

FireDBG Command Line Interface

firedbg-cli is a CLI to invoke all FireDBG operations.

Cargo Workspace

The firedbg command can only act on Cargo Workspace. If you have a simple dependency free rust file, you still need to put it under a cargo workspace for firedbg to work properly.

There are two ways to tell firedbg where is the root directory of a cargo workspace:

1. By default, the current directory will be the root directory of a cargo workspace
2. Or, overriding it with --workspace-root option, i.e. firedbg --workspace-root <WORKSPACE-ROOT>

Common Subcommands

• cache: Parse all .rs source files in the current workspace
• clean: Cleanup the firedbg/ folder
• list-target: List all runnable targets
• run: Run a binary target with debugging enabled
• example: Run an example with debugging enabled
• test: Run an integrated test with debugging enabled
• unit-test: Run a unit test with debugging enabled
• index: Run indexer on the latest run and save it as a .sqlite db file
• list-run: List all firedbg runs
• open: Open debugger view in VS Code
• help: Print help message or the help of the given subcommand(s)

You can get the help messages by appending the --help flag.

The firedbg.toml Config File

By default FireDBG will only trace the function calls of the debugging package. If you want to trace other packages in your local workspace, you will need to create a firedbg.toml config file on your workspace root.

[workspace.members]
quicksort = { trace = "full" }
# Syntax: <PACKAGE> = { trace = "<full | none>" }

FireDBG Source Parser for Rust

Based on syn.

firedbg-rust-parser is a Rust source code parser. It can parse a Rust source file, walk the abstract syntax tree of Rust, then produce a list of breakpoints for the debugger to pause the program at the beginning of every function call.

Walking the AST

We will walk the Rust AST, syn::Item, and collect all forms of function / method:

1. Free standalone function, syn::Item::Fn
2. Impl function, syn::Item::Impl
3. Trait default function, syn::Item::Trait
4. Impl trait function, syn::Item::Impl
5. Nested function, walking the syn::Item recursively
6. Function defined inside inline module, syn::Item::Mod

Breakable Span

A span is a region of source code, denoted by a ranged line and column number tuple, along with macro expansion information. It allows the debugger to set a breakpoint at the correct location. The debugger will set the breakpoint either at the start or the end of the breakable span.

fn func() -> i32 {
/*                ^-- Start of Breakable Span: (Line 1, Column 19)  */
    let mut i = 0;
/*  ^-- End of Breakable Span: (Line 3, Column 5)  */
    for _ in (1..10) {
        i += 1;
    }
    i
}

Ideas

The current implementation is rudimentary, but we get exactly what we need. We considered embedding Rust Analyzer, for a few advantages: 1) to get the fully-qualified type names 2) to traverse the static call graph. The problem is resource usage: we'd end up running the compiler frontend thrice (by cargo, by language server, by firedbg).

FireDBG Event Stream Protocol

The FireDBG Event Stream is serialized according to the SeaStreamer File Format, which by convention has the .ss extension. The Protocol defines the different streams and formats of the messages on top of the file format, and thus they have the .firedbg.ss extension. The file format is not tightly-coupled with the stream protocol, as it is possible to stream to/from a different backend, e.g. Redis.

There are currently 4 streams:

| Stream Key | Format | Description | |:----------:|:------:|:-----------:| | info | Json | DebuggerInfo: debugger version, debug target, arguments and exit code, etc | | file | Json | SourceFile: relative path to the source file | | breakpoint | Json | Breakpoint: breakpoints created and the source location | | event | Binary | Event: function call, function return, etc |

FireDBG Event Indexer

firedbg-stream-indexer is a streaming indexer. It can stream events from .firedbg.ss files, index them in real-time, and write updates to .sqlite incrementally.

There are 4 event types:

| Event Code | Event Type | Description | |:----------:|:----:|:-----------:| | B | Breakpoint | e.g. a breakpoint hit by fire::dbg! | P | Panic | Program panic | | F | Function Call | - | | R | Function Return | - |

The indexer reconstructs the call stack for each thread from the event stream, and write a parent_frame_id for each F event.

The indexer also deserializes the value blobs and translates them into JSON. The JSON is then transformed into pretty-printed Rust-like value strings:

Value Blob -> RValue -> Lifted RValue -> Pretty Print

The database schema can be found under indexer/src/entity/, which is defined by a set of SeaORM entities.

Highly recommend you to install a SQLite extension for VS Code. You can find some sample indexes in the Testbench.

FireDBG Debugger Engine for Rust

Based on lldb.

This library is semver exempt. The version number is intended to track rustc's version.

Debugger Config

Configuration can be set via the FIREDBG_RUST_CONFIG environment variable. e.g. FIREDBG_RUST_CONFIG=MAX_ARRAY_SIZE=100;DONT_TRACE_ALLOCATION

| Key | Type | Description | |:---:|:----:|:-----------:| | MAX_ARRAY_SIZE | usize | Maximum number of items in array, string and other containers | | RECURSIVE_DEREF_LIMIT | usize | Recursive limit; i.e. this limits the depth of a binary tree | | KEEP_HASH_ORDER | bool | If set, don't sort hash maps by hash key | | DONT_TRACE_ALLOCATION | bool | If set, don't trace heap allocations |

Instruction Set

Currently supports x64 (aka amd64) and arm64 (aka aarch64). There are quite some assembly parsing and register fiddling catered to each platform. There are some assumptions to pointers being 8 bytes in the codebase. It requires considerable effort to support a new architecture, but we are open to commercial collaboration.

Operating Systems

There is no OS specific code for now. lldb is used on both Linux and macOS. But on macOS we'd connect to the host's lldb-server. If we decided to support Windows natively, we'd need to make a Windows backend.

The debugger binary must be compiled for the same platform as the target binary. In addition, we assume they both use the same rustc (not necessarily exactly the same, but ABI compatible with each other).

Standard Library Types

We intend to build-in all handling of standard library types. For HashMap, frozen-hashbrown is used. In the future, we want to open scripting interface (maybe via rhai) to handle vendor library types (e.g. DateTime, Decimal), in a similar sense to Natvis.

Binary Value Format

The format for serializing Rust values can be best understood by reading SourceReader::read_values() in reader.rs. It should be pretty straight-forward, the only tricky part is ReaderContext which is for resolving memory references.

Return Value Capture

This is highly architecture specific. We try to capture the return value at the moment the function returns, i.e. at the ret instruction. Not everything is on the stack, sometimes the return value will be passed through registers.

According to Rust's ABI convention, it means if the return value is:

1. One or two primitives, each no bigger than 64 bits. This includes (i64, i64) and struct { a: i64, b: i64 }.

2. One i128 / u128; will be split into rax / rdx

3. One or two f32 / f64; will be put in xmm0 / xmm1

4. Option<T> and Result<T, E>; where T & E is no bigger than 64 bits

   The enum discriminant is in rax, where:

   | Type | T | E | |:----:|:-:|:-:| | Option | Some = 1 | None = 0 | | Result | Ok = 0 | Err = 1 |

   and the T / E will be in rdx.

Unfortunately it is much more complicated than the above summary. Right now the implementation is all mess, and completely based on heuristics. If we got the chance to do this properly, may be we can generate rust code and inspect the assembly (systematically). Say our return type is (T, F):

fn probe() -> (T, F) { todo!() }
fn extractor() {
    let res = probe();
    std::hint::black_box(&res);
}

If we disassemble the binary, we should see:

extractor:
call probe
mov ..
call black_box

So betw