Re: [PATCH bpf-next v9 1/2] bpftool: Add recursive map dumping

[Quentin Monnet <qmo@kernel.org>]

2026-09-26 10:42 UTC+0900 ~ Tianyi Chen <hi@tychen.cc>
> Dumping a map-of-maps currently shows inner map IDs without their
> contents. Add a recursive keyword to map dump to include referenced
> inner maps, leaving the default output unchanged. Show selected maps
> followed by distinct inner maps, preserving plain and BTF formatting
> and using an array of map objects for JSON output.
> 
> Holding every discovered inner-map FD open would make descriptor use
> grow with the number of maps and could exhaust RLIMIT_NOFILE. Keep the
> selected map FDs open, queue distinct inner map IDs, and open, dump and
> close each queued map in turn. This needs only one additional map FD.
> The tradeoff is deferred ID resolution: concurrently removed inner
> maps can disappear before they are opened, so the dump is not atomic.
> 
> Report failure when an inner map cannot be opened, retain per-entry
> lookup errors, and close JSON containers before returning.
> 
> Link: https://github.com/libbpf/bpftool/issues/58
> Assisted-by: LLM
> Signed-off-by: Tianyi Chen <hi@tychen.cc>

> diff --git a/tools/bpf/bpftool/map.c b/tools/bpf/bpftool/map.c
> index 626882f2079e..015910a493db 100644
> --- a/tools/bpf/bpftool/map.c
> +++ b/tools/bpf/bpftool/map.c

> @@ -971,30 +1066,61 @@ static int do_dump(int argc, char **argv)
>  		}
>  	}
>  
> -	if (wtr && nb_fds > 1)
> +	if (wtr && (nb_fds > 1 || recursive_dump))
>  		jsonw_start_array(wtr);	/* root array */
>  	for (i = 0; i < nb_fds; i++) {
> +		memset(&info, 0, sizeof(info));
> +		len = sizeof(info);


Good catch, reading bpf-ci's review the other day I didn't realise there
were other locations in the file where we omitted to reset "info", which
means we may print wrong BTF info when dumping the maps, that's a bug.

I can see three other locations (do_show(), do_show_subset(),
maps_have_btf()) where we call bpf_map_get_info_by_fd() in loops without
resetting "info". We'll need to fix these, but this should be a
different patch or series. You're welcome to go at it if you want, I'll
send a patch otherwise, just let me know. I would maybe move this
memset() (the hunk above) to that dedicated patch.

This patch looks good to me otherwise:

Acked-by: Quentin Monnet <qmo@kernel.org>

You still have a conflict with Maxim's series on patch 2 though, both of
you picked the same test file name. Maybe use
bpftool_map_dump_recursive.c instead?

Thanks,
Quentin