Re: [PATCH bpf-next v2] bpf: Allow XDP dev-bound programs to perform XDP_REDIRECT into maps

[Lorenzo Bianconi <lorenzo@kernel.org>]

[-- Attachment #1: Type: text/plain, Size: 5206 bytes --]

> Lorenzo Bianconi <lorenzo@kernel.org> writes:
> 
> > In the current implementation if the program is dev-bound to a specific
> > device, it will not be possible to perform XDP_REDIRECT into a DEVMAP
> > or CPUMAP even if the program is running in the driver NAPI context and
> > it is not attached to any map entry. This seems in contrast with the
> > explanation available in bpf_prog_map_compatible routine.
> > Fix the issue introducing __bpf_prog_map_compatible utility routine in
> > order to avoid bpf_prog_is_dev_bound() check running bpf_check_tail_call()
> > at program load time (bpf_prog_select_runtime()).
> > Continue forbidding to attach a dev-bound program to XDP maps
> > (BPF_MAP_TYPE_PROG_ARRAY, BPF_MAP_TYPE_DEVMAP and BPF_MAP_TYPE_CPUMAP).
> >
> > Fixes: 3d76a4d3d4e59 ("bpf: XDP metadata RX kfuncs")
> > Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
> > ---
> > Changes in v2:
> > - Introduce __bpf_prog_map_compatible() utility routine in order to skip
> >   bpf_prog_is_dev_bound check in bpf_check_tail_call()
> > - Extend xdp_metadata selftest
> > - Link to v1: https://lore.kernel.org/r/20250422-xdp-prog-bound-fix-v1-1-0b581fa186fe@kernel.org
> > ---
> >  kernel/bpf/core.c                                  | 27 +++++++++++++---------
> >  .../selftests/bpf/prog_tests/xdp_metadata.c        | 22 +++++++++++++++++-
> >  tools/testing/selftests/bpf/progs/xdp_metadata.c   | 13 +++++++++++
> >  3 files changed, 50 insertions(+), 12 deletions(-)
> >
> > diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c
> > index ba6b6118cf504041278d05417c4212d57be6fca0..a3e571688421196c3ceaed62b3b59b62a0258a8c 100644
> > --- a/kernel/bpf/core.c
> > +++ b/kernel/bpf/core.c
> > @@ -2358,8 +2358,8 @@ static unsigned int __bpf_prog_ret0_warn(const void *ctx,
> >  	return 0;
> >  }
> >  
> > -bool bpf_prog_map_compatible(struct bpf_map *map,
> > -			     const struct bpf_prog *fp)
> > +static bool __bpf_prog_map_compatible(struct bpf_map *map,
> > +				      const struct bpf_prog *fp)
> >  {
> >  	enum bpf_prog_type prog_type = resolve_prog_type(fp);
> >  	bool ret;
> > @@ -2368,14 +2368,6 @@ bool bpf_prog_map_compatible(struct bpf_map *map,
> >  	if (fp->kprobe_override)
> >  		return false;
> >  
> > -	/* XDP programs inserted into maps are not guaranteed to run on
> > -	 * a particular netdev (and can run outside driver context entirely
> > -	 * in the case of devmap and cpumap). Until device checks
> > -	 * are implemented, prohibit adding dev-bound programs to program maps.
> > -	 */
> > -	if (bpf_prog_is_dev_bound(aux))
> > -		return false;
> > -
> >  	spin_lock(&map->owner.lock);
> >  	if (!map->owner.type) {
> >  		/* There's no owner yet where we could check for
> > @@ -2409,6 +2401,19 @@ bool bpf_prog_map_compatible(struct bpf_map *map,
> >  	return ret;
> >  }
> >  
> > +bool bpf_prog_map_compatible(struct bpf_map *map, const struct bpf_prog *fp)
> > +{
> > +	/* XDP programs inserted into maps are not guaranteed to run on
> > +	 * a particular netdev (and can run outside driver context entirely
> > +	 * in the case of devmap and cpumap). Until device checks
> > +	 * are implemented, prohibit adding dev-bound programs to program maps.
> > +	 */
> > +	if (bpf_prog_is_dev_bound(fp->aux))
> > +		return false;
> > +
> > +	return __bpf_prog_map_compatible(map, fp);
> > +}
> > +
> >  static int bpf_check_tail_call(const struct bpf_prog *fp)
> >  {
> >  	struct bpf_prog_aux *aux = fp->aux;
> > @@ -2421,7 +2426,7 @@ static int bpf_check_tail_call(const struct bpf_prog *fp)
> >  		if (!map_type_contains_progs(map))
> >  			continue;
> >  
> > -		if (!bpf_prog_map_compatible(map, fp)) {
> > +		if (!__bpf_prog_map_compatible(map, fp)) {
> 
> Hmm, so this allows devbound programs in tail call maps, right? But
> there's no guarantee that a tail call map will always be used for a
> particular device, is there? For instance, it could be shared between
> multiple XDP programs, bound to different devices, thus getting the
> wrong kfunc.

According to my understanding the following path will be executed just for
dev-bound program that performs XDP_REDIRECT into a BPF_MAP_TYPE_PROG_ARRAY:

bpf_prog_select_runtime() -> bpf_check_tail_call() -> __bpf_prog_map_compatible()

while for XDP program inserted into BPF_MAP_TYPE_PROG_ARRAY we will continue
running bpf_prog_map_compatible() so we will forbid inserting ev-bound programs.
This is even tested into xdp_metadata selftest:

https://github.com/torvalds/linux/blob/master/tools/testing/selftests/bpf/prog_tests/xdp_metadata.c#L416

It seems to me v2 is not more relaxed than v1. Am I missing something?

Regards,
Lorenzo

> 
> Or you could even have dev-bound programs tail-called from non-dev-bound
> programs with this change AFAICT?
> 
> In other words, I think this is too relaxed, your change in v1 that only
> relaxed cpumap and devmap checks here was better.
> 
> In fact, I don't really see why bpf_check_tail_call() needs to look at
> devmap/cpumap at all, so maybe just changing the
> map_type_contains_progs() call to only match tail call maps is better?
> 
> -Toke

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]