Re: [PATCH 10/20] libdtrace: retain descriptors for cached module BTF IDs

[Nix <nix@esperi.org.uk>]

On 30 Oct 2026, Kris Van Hees outgrape:

> ---
>  libdtrace/dt_btf.c               | 79 ++++++++++++++++++++++++++++++--
>  libdtrace/dt_impl.h              |  1 +
>  libdtrace/dt_open.c              |  1 +
>  test/triggers/module-discovery.c | 11 +++++
>  4 files changed, 88 insertions(+), 4 deletions(-)
>
> diff --git a/libdtrace/dt_btf.c b/libdtrace/dt_btf.c
> index 617b0128..727e5f54 100644
> --- a/libdtrace/dt_btf.c
> +++ b/libdtrace/dt_btf.c
> @@ -1479,17 +1479,67 @@ dt_btf_lookup_name_kind(dtrace_hdl_t *dtp, dt_module_t *dmp, const char *name,
>  	return id;
>  }
>  
> +/*
> + * Retain a descriptor as well as the ID: otherwise an unloaded object's ID
> + * could be reused and a later enumeration could mistake it for a cached one.
> + * Only IDs assigned to modules are retained, bounding descriptor use by the
> + * number of modules discovered by this handle.
> + */

I suppose just not reusing IDs in general, and having them only
increment, would end up using too many? (I can't remember under what
circumstances they're reused, but I didn't think it was *common*.)

How it interacts with the earlier change to never remove dt_modules when
their corresponding kernel module is unloaded is unclear to me -- is
this the bit that ensures that an unloaded module never gets reused (and
instead we load a new one if the module is reloaded into the kernel), or
what?

-- 
NULL && (void)