code-structure.md

Code structure for module and BPF implementation

This is the common code structure for the module’s tcp_powertcp.c and the BPF implementation’s powertcp.bpf.c. It uses direct #includes of source files (instead of using multiple compilation units with headers) to enable full inlining and other optimizations for both the module and BPF implementation.

The parts must appear in this order:

1. General #includes, including the required linux/types.h (module) or vmlinux.h (BPF)

2. An #include defining INT-related values and types in this order:

   1. Provide max_n_hops as an enumerator

   2. #include "powertcp_int.c" which provides struct powertcp_int (requires max_n_hops) and other structs

   3. Provide a typedef for powertcp_int_impl_t, which can alias e.g. a struct or a pointer, and constants max_ts and max_tx_bytes

   This include should be named tcp_powertcp_METHOD_head.c resp. powertcp_METHOD_head.bpf.c, e.g. tcp_powertcp_foobar_head.c or powertcp_tcp-int_head.bpf.c.

   powertcp_no-int_head.c shows the required content of this file.

3. #defines for various POWERTCP_* macros as needed; their default values are defined in powertcp_head.c

4. #include "powertcp_head.c" providing the core struct powertcp (requires powertcp_int_impl_t), the variables for the algorithm parameters, and default #defines for the still undefined POWERTCP_* macros

5. Additional (algorithm) parameter variables, other static/constant variables

6. Definitions of the required, module- or BPF-specific functions

7. An #include defining the INT-related functions

   This include should be named tcp_powertcp_METHOD.c resp. powertcp_METHOD.bpf.c, e.g. tcp_powertcp_foobar.c or powertcp_tcp-int.bpf.c.

   powertcp_no-int.c shows the required content of this file.

8. #include "powertcp.c" of the algorithm implementation

9. Additional definitions of functions requiring the PowerTCP tcp_congestion_ops instances, e.g. module_init and module_exit