[bpf-next,v6,08/17] bpf: Support consuming XDP HW metadata from fext programs

Commit Message

Stanislav Fomichev Jan. 4, 2023, 9:59 p.m. UTC

From: Toke Høiland-Jørgensen <[e-mail]>

Instead of rejecting the attaching of PROG_TYPE_EXT programs to XDP
programs that consume HW metadata, implement support for propagating the
offload information. The extension program doesn't need to set a flag or
ifindex, these will just be propagated from the target by the verifier.
We need to create a separate offload object for the extension program,
though, since it can be reattached to a different program later (which
means we can't just inherit the offload information from the target).

An additional check is added on attach that the new target is compatible
with the offload information in the extension prog.

Signed-off-by: Toke Høiland-Jørgensen <[e-mail]>
Signed-off-by: Stanislav Fomichev <[e-mail]>
---
 include/linux/bpf.h   |  14 ++++++
 kernel/bpf/offload.c  | 113 ++++++++++++++++++++++++++++++++----------
 kernel/bpf/syscall.c  |  13 +++++
 kernel/bpf/verifier.c |   5 --
 4 files changed, 113 insertions(+), 32 deletions(-)

Comments

Martin KaFai Lau Jan. 6, 2023, 12:57 a.m. UTC | #1

On 1/4/23 1:59 PM, Stanislav Fomichev wrote:
> -int bpf_prog_dev_bound_init(struct bpf_prog *prog, union bpf_attr *attr)
> +static int __bpf_prog_dev_bound_init(struct bpf_prog *prog, struct net_device *netdev)
>   {
>   	struct bpf_offload_netdev *ondev;
>   	struct bpf_prog_offload *offload;
>   	int err;
>   
> -	if (attr->prog_type != BPF_PROG_TYPE_SCHED_CLS &&
> -	    attr->prog_type != BPF_PROG_TYPE_XDP)
> -		return -EINVAL;
> -
> -	if (attr->prog_flags & ~BPF_F_XDP_DEV_BOUND_ONLY)
> +	if (!netdev)

nit. I think this check is also unnecessary.

[ ... ]

> diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
> index 191a4312f4b7..2ec2f53eeff6 100644
> --- a/kernel/bpf/syscall.c
> +++ b/kernel/bpf/syscall.c
> @@ -2605,6 +2605,13 @@ static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr)
>   			goto free_prog_sec;
>   	}
>   
> +	if (type == BPF_PROG_TYPE_EXT && dst_prog &&
> +	    bpf_prog_is_dev_bound(dst_prog->aux)) {
> +		err = bpf_prog_dev_bound_inherit(prog, dst_prog);
> +		if (err)
> +			goto free_prog_sec;
> +	}
> +
>   	/* find program type: socket_filter vs tracing_filter */
>   	err = find_prog_type(type, prog);
>   	if (err < 0)
> @@ -3021,6 +3028,12 @@ static int bpf_tracing_prog_attach(struct bpf_prog *prog,
>   			goto out_put_prog;
>   		}
>   
> +		if (bpf_prog_is_dev_bound(prog->aux) &&
> +		    !bpf_prog_dev_bound_match(prog, tgt_prog)) {
> +			err = -EINVAL;
> +			goto out_put_prog;
> +		}

This looks good.  One minor comment...

> +
>   		key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
>   	}
>   
> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> index 0d0a49a2c5fd..8c1b1259f30b 100644
> --- a/kernel/bpf/verifier.c
> +++ b/kernel/bpf/verifier.c
> @@ -16531,11 +16531,6 @@ int bpf_check_attach_target(struct bpf_verifier_log *log,
>   	if (tgt_prog) {
>   		struct bpf_prog_aux *aux = tgt_prog->aux;
>   
> -		if (bpf_prog_is_dev_bound(tgt_prog->aux)) {
> -			bpf_log(log, "Replacing device-bound programs not supported\n");
> -			return -EINVAL;
> -		}

... can the above "bpf_prog_is_dev_bound(prog->aux) &&..." check in syscall.c be 
done in the bpf_check_attach_target() here?  Mentally that seems to belong more 
to bpf_check_attach_target().

> -
>   		for (i = 0; i < aux->func_info_cnt; i++)
>   			if (aux->func_info[i].type_id == btf_id) {
>   				subprog = i;

Stanislav Fomichev Jan. 6, 2023, 5:17 p.m. UTC | #2

On Thu, Jan 5, 2023 at 4:57 PM Martin KaFai Lau <[e-mail]> wrote:
>
> On 1/4/23 1:59 PM, Stanislav Fomichev wrote:
> > -int bpf_prog_dev_bound_init(struct bpf_prog *prog, union bpf_attr *attr)
> > +static int __bpf_prog_dev_bound_init(struct bpf_prog *prog, struct net_device *netdev)
> >   {
> >       struct bpf_offload_netdev *ondev;
> >       struct bpf_prog_offload *offload;
> >       int err;
> >
> > -     if (attr->prog_type != BPF_PROG_TYPE_SCHED_CLS &&
> > -         attr->prog_type != BPF_PROG_TYPE_XDP)
> > -             return -EINVAL;
> > -
> > -     if (attr->prog_flags & ~BPF_F_XDP_DEV_BOUND_ONLY)
> > +     if (!netdev)
>
> nit. I think this check is also unnecessary.
>
> [ ... ]
>
> > diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
> > index 191a4312f4b7..2ec2f53eeff6 100644
> > --- a/kernel/bpf/syscall.c
> > +++ b/kernel/bpf/syscall.c
> > @@ -2605,6 +2605,13 @@ static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr)
> >                       goto free_prog_sec;
> >       }
> >
> > +     if (type == BPF_PROG_TYPE_EXT && dst_prog &&
> > +         bpf_prog_is_dev_bound(dst_prog->aux)) {
> > +             err = bpf_prog_dev_bound_inherit(prog, dst_prog);
> > +             if (err)
> > +                     goto free_prog_sec;
> > +     }
> > +
> >       /* find program type: socket_filter vs tracing_filter */
> >       err = find_prog_type(type, prog);
> >       if (err < 0)
> > @@ -3021,6 +3028,12 @@ static int bpf_tracing_prog_attach(struct bpf_prog *prog,
> >                       goto out_put_prog;
> >               }
> >
> > +             if (bpf_prog_is_dev_bound(prog->aux) &&
> > +                 !bpf_prog_dev_bound_match(prog, tgt_prog)) {
> > +                     err = -EINVAL;
> > +                     goto out_put_prog;
> > +             }
>
> This looks good.  One minor comment...

True, fill drop, thanks!

> > +
> >               key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
> >       }
> >
> > diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> > index 0d0a49a2c5fd..8c1b1259f30b 100644
> > --- a/kernel/bpf/verifier.c
> > +++ b/kernel/bpf/verifier.c
> > @@ -16531,11 +16531,6 @@ int bpf_check_attach_target(struct bpf_verifier_log *log,
> >       if (tgt_prog) {
> >               struct bpf_prog_aux *aux = tgt_prog->aux;
> >
> > -             if (bpf_prog_is_dev_bound(tgt_prog->aux)) {
> > -                     bpf_log(log, "Replacing device-bound programs not supported\n");
> > -                     return -EINVAL;
> > -             }
>
> ... can the above "bpf_prog_is_dev_bound(prog->aux) &&..." check in syscall.c be
> done in the bpf_check_attach_target() here?  Mentally that seems to belong more
> to bpf_check_attach_target().

Sure, I don't see why not, will try!



> > -
> >               for (i = 0; i < aux->func_info_cnt; i++)
> >                       if (aux->func_info[i].type_id == btf_id) {
> >                               subprog = i;
>

Patch

diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index a9078f842b96..2ea32d242fc7 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -2484,6 +2484,7 @@  int bpf_dev_bound_kfunc_check(struct bpf_verifier_log *log,
 			      struct bpf_prog_aux *prog_aux);
 void *bpf_dev_bound_resolve_kfunc(struct bpf_prog *prog, u32 func_id);
 int bpf_prog_dev_bound_init(struct bpf_prog *prog, union bpf_attr *attr);
+int bpf_prog_dev_bound_inherit(struct bpf_prog *new_prog, struct bpf_prog *old_prog);
 void bpf_dev_bound_netdev_unregister(struct net_device *dev);
 
 static inline bool bpf_prog_is_dev_bound(const struct bpf_prog_aux *aux)
@@ -2496,6 +2497,8 @@  static inline bool bpf_prog_is_offloaded(const struct bpf_prog_aux *aux)
 	return aux->offload_requested;
 }
 
+bool bpf_prog_dev_bound_match(struct bpf_prog *lhs, struct bpf_prog *rhs);
+
 static inline bool bpf_map_is_offloaded(struct bpf_map *map)
 {
 	return unlikely(map->ops == &bpf_map_offload_ops);
@@ -2535,6 +2538,12 @@  static inline int bpf_prog_dev_bound_init(struct bpf_prog *prog,
 	return -EOPNOTSUPP;
 }
 
+static inline int bpf_prog_dev_bound_inherit(struct bpf_prog *new_prog,
+					     struct bpf_prog *old_prog)
+{
+	return -EOPNOTSUPP;
+}
+
 static inline void bpf_dev_bound_netdev_unregister(struct net_device *dev)
 {
 }
@@ -2549,6 +2558,11 @@  static inline bool bpf_prog_is_offloaded(struct bpf_prog_aux *aux)
 	return false;
 }
 
+static inline bool bpf_prog_dev_bound_match(struct bpf_prog *lhs, struct bpf_prog *rhs)
+{
+	return false;
+}
+
 static inline bool bpf_map_is_offloaded(struct bpf_map *map)
 {
 	return false;
diff --git a/kernel/bpf/offload.c b/kernel/bpf/offload.c
index 50ba391af281..e0956d524578 100644
--- a/kernel/bpf/offload.c
+++ b/kernel/bpf/offload.c
@@ -187,39 +187,27 @@  static void __bpf_offload_dev_netdev_unregister(struct bpf_offload_dev *offdev,
 	kfree(ondev);
 }
 
-int bpf_prog_dev_bound_init(struct bpf_prog *prog, union bpf_attr *attr)
+static int __bpf_prog_dev_bound_init(struct bpf_prog *prog, struct net_device *netdev)
 {
 	struct bpf_offload_netdev *ondev;
 	struct bpf_prog_offload *offload;
 	int err;
 
-	if (attr->prog_type != BPF_PROG_TYPE_SCHED_CLS &&
-	    attr->prog_type != BPF_PROG_TYPE_XDP)
-		return -EINVAL;
-
-	if (attr->prog_flags & ~BPF_F_XDP_DEV_BOUND_ONLY)
+	if (!netdev)
 		return -EINVAL;
 
-	if (attr->prog_type == BPF_PROG_TYPE_SCHED_CLS &&
-	    attr->prog_flags & BPF_F_XDP_DEV_BOUND_ONLY)
-		return -EINVAL;
+	offload = kzalloc(sizeof(*offload), GFP_USER);
+	if (!offload)
+		return -ENOMEM;
 
 	offload->prog = prog;
+	offload->netdev = netdev;
 
-	offload->netdev = dev_get_by_index(current->nsproxy->net_ns,
-					   attr->prog_ifindex);
-	err = bpf_dev_offload_check(offload->netdev);
-	if (err)
-		goto err_maybe_put;
-
-	prog->aux->offload_requested = !(attr->prog_flags & BPF_F_XDP_DEV_BOUND_ONLY);
-
-	down_write(&bpf_devs_lock);
 	ondev = bpf_offload_find_netdev(offload->netdev);
 	if (!ondev) {
 		if (bpf_prog_is_offloaded(prog->aux)) {
 			err = -EINVAL;
-			goto err_unlock;
+			goto err_free;
 		}
 
 		/* When only binding to the device, explicitly
@@ -227,25 +215,80 @@  int bpf_prog_dev_bound_init(struct bpf_prog *prog, union bpf_attr *attr)
 		 */
 		err = __bpf_offload_dev_netdev_register(NULL, offload->netdev);
 		if (err)
-			goto err_unlock;
+			goto err_free;
 		ondev = bpf_offload_find_netdev(offload->netdev);
 	}
 	offload->offdev = ondev->offdev;
 	prog->aux->offload = offload;
 	list_add_tail(&offload->offloads, &ondev->progs);
-	dev_put(offload->netdev);
-	up_write(&bpf_devs_lock);
 
 	return 0;
-err_unlock:
-	up_write(&bpf_devs_lock);
-err_maybe_put:
-	if (offload->netdev)
-		dev_put(offload->netdev);
+err_free:
 	kfree(offload);
 	return err;
 }
 
+int bpf_prog_dev_bound_init(struct bpf_prog *prog, union bpf_attr *attr)
+{
+	struct net_device *netdev;
+	int err;
+
+	if (attr->prog_type != BPF_PROG_TYPE_SCHED_CLS &&
+	    attr->prog_type != BPF_PROG_TYPE_XDP)
+		return -EINVAL;
+
+	if (attr->prog_flags & ~BPF_F_XDP_DEV_BOUND_ONLY)
+		return -EINVAL;
+
+	if (attr->prog_type == BPF_PROG_TYPE_SCHED_CLS &&
+	    attr->prog_flags & BPF_F_XDP_DEV_BOUND_ONLY)
+		return -EINVAL;
+
+	netdev = dev_get_by_index(current->nsproxy->net_ns, attr->prog_ifindex);
+	if (!netdev)
+		return -EINVAL;
+
+	err = bpf_dev_offload_check(netdev);
+	if (err)
+		goto out;
+
+	prog->aux->offload_requested = !(attr->prog_flags & BPF_F_XDP_DEV_BOUND_ONLY);
+
+	down_write(&bpf_devs_lock);
+	err = __bpf_prog_dev_bound_init(prog, netdev);
+	up_write(&bpf_devs_lock);
+
+out:
+	dev_put(netdev);
+	return err;
+}
+
+int bpf_prog_dev_bound_inherit(struct bpf_prog *new_prog, struct bpf_prog *old_prog)
+{
+	int err;
+
+	if (!bpf_prog_is_dev_bound(old_prog->aux))
+		return 0;
+
+	if (bpf_prog_is_offloaded(old_prog->aux))
+		return -EINVAL;
+
+	new_prog->aux->dev_bound = old_prog->aux->dev_bound;
+	new_prog->aux->offload_requested = old_prog->aux->offload_requested;
+
+	down_write(&bpf_devs_lock);
+	if (!old_prog->aux->offload) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	err = __bpf_prog_dev_bound_init(new_prog, old_prog->aux->offload->netdev);
+
+out:
+	up_write(&bpf_devs_lock);
+	return err;
+}
+
 int bpf_prog_offload_verifier_prep(struct bpf_prog *prog)
 {
 	struct bpf_prog_offload *offload;
@@ -671,6 +714,22 @@  bool bpf_offload_dev_match(struct bpf_prog *prog, struct net_device *netdev)
 }
 EXPORT_SYMBOL_GPL(bpf_offload_dev_match);
 
+bool bpf_prog_dev_bound_match(struct bpf_prog *lhs, struct bpf_prog *rhs)
+{
+	bool ret;
+
+	if (bpf_prog_is_offloaded(lhs->aux) != bpf_prog_is_offloaded(rhs->aux))
+		return false;
+
+	down_read(&bpf_devs_lock);
+	ret = lhs->aux->offload && rhs->aux->offload &&
+	      lhs->aux->offload->netdev &&
+	      lhs->aux->offload->netdev == rhs->aux->offload->netdev;
+	up_read(&bpf_devs_lock);
+
+	return ret;
+}
+
 bool bpf_offload_prog_map_match(struct bpf_prog *prog, struct bpf_map *map)
 {
 	struct bpf_offloaded_map *offmap;
diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
index 191a4312f4b7..2ec2f53eeff6 100644
--- a/kernel/bpf/syscall.c
+++ b/kernel/bpf/syscall.c
@@ -2605,6 +2605,13 @@  static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr)
 			goto free_prog_sec;
 	}
 
+	if (type == BPF_PROG_TYPE_EXT && dst_prog &&
+	    bpf_prog_is_dev_bound(dst_prog->aux)) {
+		err = bpf_prog_dev_bound_inherit(prog, dst_prog);
+		if (err)
+			goto free_prog_sec;
+	}
+
 	/* find program type: socket_filter vs tracing_filter */
 	err = find_prog_type(type, prog);
 	if (err < 0)
@@ -3021,6 +3028,12 @@  static int bpf_tracing_prog_attach(struct bpf_prog *prog,
 			goto out_put_prog;
 		}
 
+		if (bpf_prog_is_dev_bound(prog->aux) &&
+		    !bpf_prog_dev_bound_match(prog, tgt_prog)) {
+			err = -EINVAL;
+			goto out_put_prog;
+		}
+
 		key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
 	}
 
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index 0d0a49a2c5fd..8c1b1259f30b 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -16531,11 +16531,6 @@  int bpf_check_attach_target(struct bpf_verifier_log *log,
 	if (tgt_prog) {
 		struct bpf_prog_aux *aux = tgt_prog->aux;
 
-		if (bpf_prog_is_dev_bound(tgt_prog->aux)) {
-			bpf_log(log, "Replacing device-bound programs not supported\n");
-			return -EINVAL;
-		}
-
 		for (i = 0; i < aux->func_info_cnt; i++)
 			if (aux->func_info[i].type_id == btf_id) {
 				subprog = i;