libbpf: Add bpf_map__set_name()

[PATCH bpf-next] libbpf: Add bpf_map__set_name()

Posted by Joe Burton 3 years, 9 months ago

From: Joe Burton <jevburton@google.com>

Add the capability to set a `struct bpf_map` name.

bpf_map__reuse_fd(struct bpf_map *map, int fd) does the following:

1. get the bpf_map_info of the passed-in fd
2. strdup the name from the bpf_map_info
3. assign that name to the map
4. and some other stuff

While `map.name` may initially be arbitrarily long, this operation
truncates it after 15 characters.

We have some infrastructure that uses bpf_map__reuse_fd() to preserve
maps across upgrades. Some of our users have long map names, and are
seeing their maps 'disappear' after an upgrade, due to the name
truncation.

By invoking `bpf_map__set_name()` after `bpf_map__reuse_fd()`, we can
trivially work around the issue.

Signed-off-by: Joe Burton <jevburton@google.com>
---
 tools/lib/bpf/libbpf.c | 22 ++++++++++++++++++++++
 tools/lib/bpf/libbpf.h |  3 ++-
 2 files changed, 24 insertions(+), 1 deletion(-)

diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
index 72548798126b..725baf508e6f 100644
--- a/tools/lib/bpf/libbpf.c
+++ b/tools/lib/bpf/libbpf.c
@@ -9089,6 +9089,28 @@ const char *bpf_map__name(const struct bpf_map *map)
 	return map->name;
 }
 
+int bpf_map__set_name(struct bpf_map *map, const char *name)
+{
+	char *new_name;
+
+	if (!map)
+		return libbpf_err(-EINVAL);
+
+	new_name = strdup(name);
+	if (!new_name)
+		return libbpf_err(-ENOMEM);
+
+	if (map_uses_real_name(map)) {
+		free(map->real_name);
+		map->real_name = new_name;
+	} else {
+		free(map->name);
+		map->name = new_name;
+	}
+
+	return 0;
+}
+
 enum bpf_map_type bpf_map__type(const struct bpf_map *map)
 {
 	return map->def.type;
diff --git a/tools/lib/bpf/libbpf.h b/tools/lib/bpf/libbpf.h
index e4d5353f757b..e898c4cb514a 100644
--- a/tools/lib/bpf/libbpf.h
+++ b/tools/lib/bpf/libbpf.h
@@ -731,8 +731,9 @@ LIBBPF_API bool bpf_map__autocreate(const struct bpf_map *map);
  */
 LIBBPF_API int bpf_map__fd(const struct bpf_map *map);
 LIBBPF_API int bpf_map__reuse_fd(struct bpf_map *map, int fd);
-/* get map name */
+/* get/set map name */
 LIBBPF_API const char *bpf_map__name(const struct bpf_map *map);
+LIBBPF_API int bpf_map__set_name(struct bpf_map *map, const char *name);
 /* get/set map type */
 LIBBPF_API enum bpf_map_type bpf_map__type(const struct bpf_map *map);
 LIBBPF_API int bpf_map__set_type(struct bpf_map *map, enum bpf_map_type type);
-- 
2.37.0.144.g8ac04bfd2-goog

Re: [PATCH bpf-next] libbpf: Add bpf_map__set_name()

Posted by Stanislav Fomichev 3 years, 9 months ago

On Wed, Jul 13, 2022 at 2:43 PM Joe Burton <jevburton.kernel@gmail.com> wrote:
>
> From: Joe Burton <jevburton@google.com>
>
> Add the capability to set a `struct bpf_map` name.
>
> bpf_map__reuse_fd(struct bpf_map *map, int fd) does the following:
>
> 1. get the bpf_map_info of the passed-in fd
> 2. strdup the name from the bpf_map_info
> 3. assign that name to the map
> 4. and some other stuff
>
> While `map.name` may initially be arbitrarily long, this operation
> truncates it after 15 characters.
>
> We have some infrastructure that uses bpf_map__reuse_fd() to preserve
> maps across upgrades. Some of our users have long map names, and are
> seeing their maps 'disappear' after an upgrade, due to the name
> truncation.
>
> By invoking `bpf_map__set_name()` after `bpf_map__reuse_fd()`, we can
> trivially work around the issue.

Asked you internally, but not sure I follow. Can you share more on why
the following won't fix it for us:

https://lore.kernel.org/bpf/OSZP286MB1725CEA1C95C5CB8E7CCC53FB8869@OSZP286MB1725.JPNP286.PROD.OUTLOOK.COM/

?

The idea seems to be to get the supplied map name (from the obj)
instead of using pin name? So why is it not enough?

> Signed-off-by: Joe Burton <jevburton@google.com>
> ---
>  tools/lib/bpf/libbpf.c | 22 ++++++++++++++++++++++
>  tools/lib/bpf/libbpf.h |  3 ++-
>  2 files changed, 24 insertions(+), 1 deletion(-)
>
> diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
> index 72548798126b..725baf508e6f 100644
> --- a/tools/lib/bpf/libbpf.c
> +++ b/tools/lib/bpf/libbpf.c
> @@ -9089,6 +9089,28 @@ const char *bpf_map__name(const struct bpf_map *map)
>         return map->name;
>  }
>
> +int bpf_map__set_name(struct bpf_map *map, const char *name)
> +{
> +       char *new_name;
> +
> +       if (!map)
> +               return libbpf_err(-EINVAL);
> +
> +       new_name = strdup(name);
> +       if (!new_name)
> +               return libbpf_err(-ENOMEM);
> +
> +       if (map_uses_real_name(map)) {
> +               free(map->real_name);
> +               map->real_name = new_name;
> +       } else {
> +               free(map->name);
> +               map->name = new_name;
> +       }
> +
> +       return 0;
> +}
> +
>  enum bpf_map_type bpf_map__type(const struct bpf_map *map)
>  {
>         return map->def.type;
> diff --git a/tools/lib/bpf/libbpf.h b/tools/lib/bpf/libbpf.h
> index e4d5353f757b..e898c4cb514a 100644
> --- a/tools/lib/bpf/libbpf.h
> +++ b/tools/lib/bpf/libbpf.h
> @@ -731,8 +731,9 @@ LIBBPF_API bool bpf_map__autocreate(const struct bpf_map *map);
>   */
>  LIBBPF_API int bpf_map__fd(const struct bpf_map *map);
>  LIBBPF_API int bpf_map__reuse_fd(struct bpf_map *map, int fd);
> -/* get map name */
> +/* get/set map name */
>  LIBBPF_API const char *bpf_map__name(const struct bpf_map *map);
> +LIBBPF_API int bpf_map__set_name(struct bpf_map *map, const char *name);
>  /* get/set map type */
>  LIBBPF_API enum bpf_map_type bpf_map__type(const struct bpf_map *map);
>  LIBBPF_API int bpf_map__set_type(struct bpf_map *map, enum bpf_map_type type);
> --
> 2.37.0.144.g8ac04bfd2-goog
>

Re: [PATCH bpf-next] libbpf: Add bpf_map__set_name()

Posted by Joe Burton 3 years, 9 months ago

> Asked you internally, but not sure I follow. Can you share more on why
> the following won't fix it for us:
> 
> https://lore.kernel.org/bpf/OSZP286MB1725CEA1C95C5CB8E7CCC53FB8869@OSZP286MB1725.JPNP286.PROD.OUTLOOK.COM/
> 
> ?
> 
> The idea seems to be to get the supplied map name (from the obj)
> instead of using pin name? So why is it not enough?

You're correct, this approach also resolves the issue. No need for this
new API.

Re: [PATCH bpf-next] libbpf: Add bpf_map__set_name()

Posted by Stanislav Fomichev 3 years, 9 months ago

On Wed, Jul 13, 2022 at 3:32 PM Joe Burton <jevburton@google.com> wrote:
>
> > Asked you internally, but not sure I follow. Can you share more on why
> > the following won't fix it for us:
> >
> > https://lore.kernel.org/bpf/OSZP286MB1725CEA1C95C5CB8E7CCC53FB8869@OSZP286MB1725.JPNP286.PROD.OUTLOOK.COM/
> >
> > ?
> >
> > The idea seems to be to get the supplied map name (from the obj)
> > instead of using pin name? So why is it not enough?
>
> You're correct, this approach also resolves the issue. No need for this
> new API.

SG! New helper might still be useful, but I'm not sure how safe that
is, given how much we use the name internally in libbpf
(name/pin_path). So it might be safer to use Anquan's approach for
now.

Andrii, any concerns with [1] ? Should we pull that in?

[1] https://lore.kernel.org/bpf/OSZP286MB1725CEA1C95C5CB8E7CCC53FB8869@OSZP286MB1725.JPNP286.PROD.OUTLOOK.COM/

Re: [PATCH bpf-next] libbpf: Add bpf_map__set_name()

Posted by Andrii Nakryiko 3 years, 9 months ago

On Wed, Jul 13, 2022 at 4:02 PM Stanislav Fomichev <sdf@google.com> wrote:
>
> On Wed, Jul 13, 2022 at 3:32 PM Joe Burton <jevburton@google.com> wrote:
> >
> > > Asked you internally, but not sure I follow. Can you share more on why
> > > the following won't fix it for us:
> > >
> > > https://lore.kernel.org/bpf/OSZP286MB1725CEA1C95C5CB8E7CCC53FB8869@OSZP286MB1725.JPNP286.PROD.OUTLOOK.COM/
> > >
> > > ?
> > >
> > > The idea seems to be to get the supplied map name (from the obj)
> > > instead of using pin name? So why is it not enough?
> >
> > You're correct, this approach also resolves the issue. No need for this
> > new API.
>
> SG! New helper might still be useful, but I'm not sure how safe that
> is, given how much we use the name internally in libbpf
> (name/pin_path). So it might be safer to use Anquan's approach for
> now.
>
> Andrii, any concerns with [1] ? Should we pull that in?

I already applied [1]. I'm uncomfortable with bpf_map__set_name() in
general, because map name is tied into BTF and other things, so it
needs much more careful thinking how to support sudden map renames.
But given the problem was with bpf_map__reuse_fd(), I think [1] solves
it already.


>
> [1] https://lore.kernel.org/bpf/OSZP286MB1725CEA1C95C5CB8E7CCC53FB8869@OSZP286MB1725.JPNP286.PROD.OUTLOOK.COM/