Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/ashvardanian/libsee
Link to this library and it will log all the LibC functions you are calling and how much time you are spending in them!
https://github.com/ashvardanian/libsee
assembly benchmark benchmarking fuzzing glibc ld-preload ld-preload-hack libc
Last synced: 3 months ago
JSON representation
Link to this library and it will log all the LibC functions you are calling and how much time you are spending in them!
- Host: GitHub
- URL: https://github.com/ashvardanian/libsee
- Owner: ashvardanian
- License: apache-2.0
- Created: 2024-03-05T02:10:11.000Z (11 months ago)
- Default Branch: main
- Last Pushed: 2024-08-24T23:35:11.000Z (5 months ago)
- Last Synced: 2024-10-15T01:25:07.360Z (3 months ago)
- Topics: assembly, benchmark, benchmarking, fuzzing, glibc, ld-preload, ld-preload-hack, libc
- Language: C
- Homepage:
- Size: 140 KB
- Stars: 11
- Watchers: 3
- Forks: 1
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- Contributing: CONTRIBUTING.md
- License: LICENSE
Awesome Lists containing this project
README
> _See where you use LibC the most._
> _Trace calls failing tests. Then - roast!_
__LibSee__ is a single-file library for profiling LibC calls and 🔜 fuzzy testing.
To download and compile the script and run your favorite query:
```bash
gcc -g -O2 -fno-builtin -fPIC -nostdlib -nostartfiles -shared -o libsee.so libsee.c
```
LibSee overrides LibC symbols using `LD_PRELOAD`, profiling the most commonly used functions, and, optionally, fuzzing their behaviour for testing.
The library yields a few binaries when compiled:
```bash
libsee.so # Profiles LibC calls
libsee_and_knee.so # Correct LibC behaviour, but fuzzed!
```
## Tricks Used
There are several things worth knowing, that came handy implementing this.
- One way to implement this library would be to override the `_start` symbols, but implementing correct loading sequence for a binary is tricky, so I use conventional `dlsym` to lookup the symbols on first function invocation.
- On `x86_64` architecture, the `rdtscp` instruction yields both the CPU cycle and also the unique identifier of the core. Very handy if you are profiling a multi-threaded application.
- Once the unloading sequence reaches `libsee.so`, the `STDOUT` is already closed. So if you want to print to the console, you may want to reopen the `/dev/tty` device before printing usage stats.
- On MacOS the `sprintf`, `vsprintf`, `snprintf`, `vsnprintf` are macros. You have to `#undef` them.
- On `Release` builds compilers love replacing your code with `memset` and `memcpy` calls. As the symbol can't be found from inside LibSee, it will `SEGFAULT` so don't forget to disable such optimizations for built-ins `-fno-builtin`.
- Aarch64 doesn't seem to have an `open` system call, but it [has the generalized `openat`](https://github.com/torvalds/linux/blob/bf3a69c6861ff4dc7892d895c87074af7bc1c400/include/uapi/asm-generic/unistd.h#L158-L159) number 56.
## Coverage
LibC standard is surprisingly long, so not all of the functions are covered.
Feel free to suggest PRs covering the rest:
- [x] [memory management](https://en.cppreference.com/w/c/memory)
- [x] [byte-strings](https://en.cppreference.com/w/c/string/byte)
- [x] [algorithms](https://en.cppreference.com/w/c/algorithm)
- [x] [date and time](https://en.cppreference.com/w/c/chrono)
- [x] [input/output](https://en.cppreference.com/w/c/io)
- [ ] [concurrency and atomics](https://en.cppreference.com/w/c/thread)
- [ ] retrieving error numbers
- [ ] [numerics](https://en.cppreference.com/w/c/numeric)
- [ ] [wide-character strings](https://en.cppreference.com/w/c/string/wide)
- [ ] [multibyte strings](https://en.cppreference.com/w/c/string/multibyte)
- [ ] [wide-character IO](https://en.cppreference.com/w/c/io)
- [ ] [localization](https://en.cppreference.com/w/c/locale)
- [ ] anything newer than C 11
[Program support](https://en.cppreference.com/w/c/program) utilities aren't intended.