2021-04-18 20:07:24

by Pedro Tammela

Subject: [PATCH] bpf: fix errno code for unsupported batch ops

ENOTSUPP is not a valid userland errno[1], which is annoying for
userland applications that implement a fallback to iterative, report
errors via 'strerror()' or both.

The batched ops return this errno whenever an operation
is not implemented for kernels that implement batched ops.

In older kernels, pre batched ops, it returns EINVAL as the arguments
are not supported in the syscall.

[1] https://lore.kernel.org/netdev/[email protected]/

Signed-off-by: Pedro Tammela <[email protected]>
---
kernel/bpf/syscall.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
index fd495190115e..88fe19c0aeb1 100644
--- a/kernel/bpf/syscall.c
+++ b/kernel/bpf/syscall.c
@@ -3961,7 +3961,7 @@ static int bpf_task_fd_query(const union bpf_attr *attr,
#define BPF_DO_BATCH(fn) \
do { \
if (!fn) { \
- err = -ENOTSUPP; \
+ err = -EOPNOTSUPP; \
goto err_put; \
} \
err = fn(map, attr, uattr); \
--
2.25.1

2021-04-18 22:58:48

by Alexei Starovoitov

Subject: Re: [PATCH] bpf: fix errno code for unsupported batch ops

On Sun, Apr 18, 2021 at 1:03 PM Pedro Tammela <[email protected]> wrote:
>
> ENOTSUPP is not a valid userland errno[1], which is annoying for
> userland applications that implement a fallback to iterative, report
> errors via 'strerror()' or both.
>
> The batched ops return this errno whenever an operation
> is not implemented for kernels that implement batched ops.
>
> In older kernels, pre batched ops, it returns EINVAL as the arguments
> are not supported in the syscall.
>
> [1] https://lore.kernel.org/netdev/[email protected]/
>
> Signed-off-by: Pedro Tammela <[email protected]>
> ---
> kernel/bpf/syscall.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
> index fd495190115e..88fe19c0aeb1 100644
> --- a/kernel/bpf/syscall.c
> +++ b/kernel/bpf/syscall.c
> @@ -3961,7 +3961,7 @@ static int bpf_task_fd_query(const union bpf_attr *attr,
> #define BPF_DO_BATCH(fn) \
> do { \
> if (!fn) { \
> - err = -ENOTSUPP; \
> + err = -EOPNOTSUPP; \

$ git grep EOPNOTSUPP kernel/bpf/|wc -l
11
$ git grep ENOTSUPP kernel/bpf/|wc -l
51

For new code EOPNOTSUPP is better, but I don't think changing all 51 case
is a good idea. Something might depend on it already.

2021-04-19 16:04:10

by Pedro Tammela

Subject: Re: [PATCH] bpf: fix errno code for unsupported batch ops

Em dom., 18 de abr. de 2021 às 19:56, Alexei Starovoitov
<[email protected]> escreveu:
>
> On Sun, Apr 18, 2021 at 1:03 PM Pedro Tammela <[email protected]> wrote:
> >
> > ENOTSUPP is not a valid userland errno[1], which is annoying for
> > userland applications that implement a fallback to iterative, report
> > errors via 'strerror()' or both.
> >
> > The batched ops return this errno whenever an operation
> > is not implemented for kernels that implement batched ops.
> >
> > In older kernels, pre batched ops, it returns EINVAL as the arguments
> > are not supported in the syscall.
> >
> > [1] https://lore.kernel.org/netdev/[email protected]/
> >
> > Signed-off-by: Pedro Tammela <[email protected]>
> > ---
> > kernel/bpf/syscall.c | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
> > index fd495190115e..88fe19c0aeb1 100644
> > --- a/kernel/bpf/syscall.c
> > +++ b/kernel/bpf/syscall.c
> > @@ -3961,7 +3961,7 @@ static int bpf_task_fd_query(const union bpf_attr *attr,
> > #define BPF_DO_BATCH(fn) \
> > do { \
> > if (!fn) { \
> > - err = -ENOTSUPP; \
> > + err = -EOPNOTSUPP; \
>
> $ git grep EOPNOTSUPP kernel/bpf/|wc -l
> 11
> $ git grep ENOTSUPP kernel/bpf/|wc -l
> 51
>
> For new code EOPNOTSUPP is better, but I don't think changing all 51 case
> is a good idea. Something might depend on it already.

OK, makes sense.

Perhaps, handle this errno in 'libbpf_strerror()'? So language
bindings don't get lost when dealing with this errno.

2021-04-19 16:07:03

by Alexei Starovoitov

Subject: Re: [PATCH] bpf: fix errno code for unsupported batch ops

On Mon, Apr 19, 2021 at 6:52 AM Pedro Tammela <[email protected]> wrote:
>
> Em dom., 18 de abr. de 2021 às 19:56, Alexei Starovoitov
> <[email protected]> escreveu:
> >
> > On Sun, Apr 18, 2021 at 1:03 PM Pedro Tammela <[email protected]> wrote:
> > >
> > > ENOTSUPP is not a valid userland errno[1], which is annoying for
> > > userland applications that implement a fallback to iterative, report
> > > errors via 'strerror()' or both.
> > >
> > > The batched ops return this errno whenever an operation
> > > is not implemented for kernels that implement batched ops.
> > >
> > > In older kernels, pre batched ops, it returns EINVAL as the arguments
> > > are not supported in the syscall.
> > >
> > > [1] https://lore.kernel.org/netdev/[email protected]/
> > >
> > > Signed-off-by: Pedro Tammela <[email protected]>
> > > ---
> > > kernel/bpf/syscall.c | 2 +-
> > > 1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
> > > index fd495190115e..88fe19c0aeb1 100644
> > > --- a/kernel/bpf/syscall.c
> > > +++ b/kernel/bpf/syscall.c
> > > @@ -3961,7 +3961,7 @@ static int bpf_task_fd_query(const union bpf_attr *attr,
> > > #define BPF_DO_BATCH(fn) \
> > > do { \
> > > if (!fn) { \
> > > - err = -ENOTSUPP; \
> > > + err = -EOPNOTSUPP; \
> >
> > $ git grep EOPNOTSUPP kernel/bpf/|wc -l
> > 11
> > $ git grep ENOTSUPP kernel/bpf/|wc -l
> > 51
> >
> > For new code EOPNOTSUPP is better, but I don't think changing all 51 case
> > is a good idea. Something might depend on it already.
>
> OK, makes sense.
>
> Perhaps, handle this errno in 'libbpf_strerror()'?

That's a good idea.

> So language
> bindings don't get lost when dealing with this errno.

I'm not sure what you mean by "language bindings".
In general, strerror is not that useful. The kernel aliases
multiple conditions into the same error code. The error string
is too generic in practice to be useful.