Re: git: fabf705f4b5a - main - pf: fix pf divert-to loop

On 10/19/23, Gleb Smirnoff <glebius@freebsd.org> wrote:
> On Thu, Oct 19, 2023 at 12:37:39PM +0000, Kristof Provost wrote:
> K> +++ b/sys/netinet/ip_var.h
> ...
> K> +/* pf specific mtag for divert(4) support */
> K> +enum { PF_DIVERT_MTAG_DIR_IN=1, PF_DIVERT_MTAG_DIR_OUT=2 };
> K> +struct pf_divert_mtag {
> K> +	uint16_t idir;	// initial pkt direction
> K> +	uint16_t ndir;	// a) divert(4) port upon initial diversion
> K> +			// b) new direction upon pkt re-enter
> K> +};
>
> This can be written as:
>
> typedef enum {
>     PF_DIVERT_MTAG_DIR_IN = 1,
>     PF_DIVERT_MTAG_DIR_OUT = 2,
> } pf_mtag_dir;
> struct pf_divert_mtag {
>         pf_mtag_dir     idir;         /* Initial packet direction. */
>         union {
>                 pf_mtag_dir     ndir; /* New direction after re-enter. */
>                 uint16_t        port;    /* Initial divert(4) port. */
>         };
> };
>
> The benefit is that in the debugger you will see PF_DIVERT_MTAG_DIR_IN
> instead
> of 1 when looking at a structure. And compilation time failure if anybody
> sets
> it to a wrong value. Using "port" instead of "ndir" when assigning a port
> improves readability of code.
>
> This will grow structure from 4 bytes to 8, as enum is always an int.
> However,
> uma allocator backing m_tag_alloc() will allocate same amount of memory.
>
> Sorry for not messaing that in the phabricator.
>

I don't have an opinion on employing an enum, but if going that route
one can shrink it by using smaller enums -- grep for __enum_uint8 for
an example.

-- 
Mateusz Guzik <mjguzik gmail.com>