2022-03-01 14:54:00

by Cheng Li

[permalink] [raw]
Subject: [PATCH bpf] libbpf: unmap rings when umem deleted

xsk_umem__create() does mmap for fill/comp rings, but xsk_umem__delete()
doesn't do the unmap. This works fine for regular cases, because
xsk_socket__delete() does unmap for the rings. But for the case that
xsk_socket__create_shared() fails, umem rings are not unmapped.

fill_save/comp_save are checked to determine if rings have already be
unmapped by xsk. If fill_save and comp_save are NULL, it means that the
rings have already been used by xsk. Then they are supposed to be
unmapped by xsk_socket__delete(). Otherwise, xsk_umem__delete() does the
unmap.

Fixes: 2f6324a3937f ("libbpf: Support shared umems between queues and devices")
Signed-off-by: lic121 <[email protected]>
---
tools/lib/bpf/xsk.c | 11 +++++++++++
1 file changed, 11 insertions(+)

diff --git a/tools/lib/bpf/xsk.c b/tools/lib/bpf/xsk.c
index edafe56..32a2f57 100644
--- a/tools/lib/bpf/xsk.c
+++ b/tools/lib/bpf/xsk.c
@@ -1193,12 +1193,23 @@ int xsk_socket__create(struct xsk_socket **xsk_ptr, const char *ifname,

int xsk_umem__delete(struct xsk_umem *umem)
{
+ struct xdp_mmap_offsets off;
+ int err;
+
if (!umem)
return 0;

if (umem->refcount)
return -EBUSY;

+ err = xsk_get_mmap_offsets(umem->fd, &off);
+ if (!err && umem->fill_save && umem->comp_save) {
+ munmap(umem->fill_save->ring - off.fr.desc,
+ off.fr.desc + umem->config.fill_size * sizeof(__u64));
+ munmap(umem->comp_save->ring - off.cr.desc,
+ off.cr.desc + umem->config.comp_size * sizeof(__u64));
+ }
+
close(umem->fd);
free(umem);

--
1.8.3.1


2022-03-02 09:42:00

by Cheng Li

[permalink] [raw]
Subject: Re: [PATCH bpf] libbpf: unmap rings when umem deleted

On Wed, Mar 02, 2022 at 09:48:33AM +0100, Bj?rn T?pel wrote:
> On Wed, 2 Mar 2022 at 08:29, Magnus Karlsson <[email protected]> wrote:
> > On Tue, Mar 1, 2022 at 6:57 PM lic121 <[email protected]> wrote:
> [...]
> > > Signed-off-by: lic121 <[email protected]>
>
> In addition to Magnus' comments; Please use your full name, as
> outlined in Documentation/process/5.Posting.rst.

Thanks for the review Bj?rn.
Magnus, please let me know if you can correct the full name when you
apply the patch? Otherwise I can create a MR in libxdp repo. Thanks

full name: Cheng Li

>
> Cheers!
> Bj?rn

2022-03-02 10:04:27

by Björn Töpel

[permalink] [raw]
Subject: Re: [PATCH bpf] libbpf: unmap rings when umem deleted

On Wed, 2 Mar 2022 at 08:29, Magnus Karlsson <[email protected]> wrote:
> On Tue, Mar 1, 2022 at 6:57 PM lic121 <[email protected]> wrote:
[...]
> > Signed-off-by: lic121 <[email protected]>

In addition to Magnus' comments; Please use your full name, as
outlined in Documentation/process/5.Posting.rst.

Cheers!
Björn

2022-03-02 14:10:32

by Magnus Karlsson

[permalink] [raw]
Subject: Re: [PATCH bpf] libbpf: unmap rings when umem deleted

On Wed, Mar 2, 2022 at 10:06 AM lic121 <[email protected]> wrote:
>
> On Wed, Mar 02, 2022 at 09:48:33AM +0100, Björn Töpel wrote:
> > On Wed, 2 Mar 2022 at 08:29, Magnus Karlsson <[email protected]> wrote:
> > > On Tue, Mar 1, 2022 at 6:57 PM lic121 <[email protected]> wrote:
> > [...]
> > > > Signed-off-by: lic121 <[email protected]>
> >
> > In addition to Magnus' comments; Please use your full name, as
> > outlined in Documentation/process/5.Posting.rst.
>
> Thanks for the review Björn.
> Magnus, please let me know if you can correct the full name when you
> apply the patch? Otherwise I can create a MR in libxdp repo. Thanks

I will add it.

Thanks: Magnus

> full name: Cheng Li
>
> >
> > Cheers!
> > Björn

2022-03-02 18:45:21

by Magnus Karlsson

[permalink] [raw]
Subject: Re: [PATCH bpf] libbpf: unmap rings when umem deleted

On Tue, Mar 1, 2022 at 6:57 PM lic121 <[email protected]> wrote:
>
> xsk_umem__create() does mmap for fill/comp rings, but xsk_umem__delete()
> doesn't do the unmap. This works fine for regular cases, because
> xsk_socket__delete() does unmap for the rings. But for the case that
> xsk_socket__create_shared() fails, umem rings are not unmapped.
>
> fill_save/comp_save are checked to determine if rings have already be
> unmapped by xsk. If fill_save and comp_save are NULL, it means that the
> rings have already been used by xsk. Then they are supposed to be
> unmapped by xsk_socket__delete(). Otherwise, xsk_umem__delete() does the
> unmap.

Thanks for the fix. Please note that the AF_XDP support in libbpf has
been deprecated and moved to libxdp
(https://github.com/xdp-project/xdp-tools). The code will be
completely removed in the libbpf 1.0 release. Could I take your patch
and apply it to libxdp instead and fix the bug there? I have not
checked, but it is likely present there as well. And that is the code
base we will be using going forward.

> Fixes: 2f6324a3937f ("libbpf: Support shared umems between queues and devices")
> Signed-off-by: lic121 <[email protected]>
> ---
> tools/lib/bpf/xsk.c | 11 +++++++++++
> 1 file changed, 11 insertions(+)
>
> diff --git a/tools/lib/bpf/xsk.c b/tools/lib/bpf/xsk.c
> index edafe56..32a2f57 100644
> --- a/tools/lib/bpf/xsk.c
> +++ b/tools/lib/bpf/xsk.c
> @@ -1193,12 +1193,23 @@ int xsk_socket__create(struct xsk_socket **xsk_ptr, const char *ifname,
>
> int xsk_umem__delete(struct xsk_umem *umem)
> {
> + struct xdp_mmap_offsets off;
> + int err;
> +
> if (!umem)
> return 0;
>
> if (umem->refcount)
> return -EBUSY;
>
> + err = xsk_get_mmap_offsets(umem->fd, &off);
> + if (!err && umem->fill_save && umem->comp_save) {
> + munmap(umem->fill_save->ring - off.fr.desc,
> + off.fr.desc + umem->config.fill_size * sizeof(__u64));
> + munmap(umem->comp_save->ring - off.cr.desc,
> + off.cr.desc + umem->config.comp_size * sizeof(__u64));
> + }
> +
> close(umem->fd);
> free(umem);
>
> --
> 1.8.3.1
>

2022-03-08 08:52:30

by Andrii Nakryiko

[permalink] [raw]
Subject: Re: [PATCH bpf] libbpf: unmap rings when umem deleted

On Tue, Mar 1, 2022 at 5:26 AM lic121 <[email protected]> wrote:
>
> xsk_umem__create() does mmap for fill/comp rings, but xsk_umem__delete()
> doesn't do the unmap. This works fine for regular cases, because
> xsk_socket__delete() does unmap for the rings. But for the case that
> xsk_socket__create_shared() fails, umem rings are not unmapped.
>
> fill_save/comp_save are checked to determine if rings have already be
> unmapped by xsk. If fill_save and comp_save are NULL, it means that the
> rings have already been used by xsk. Then they are supposed to be
> unmapped by xsk_socket__delete(). Otherwise, xsk_umem__delete() does the
> unmap.
>
> Fixes: 2f6324a3937f ("libbpf: Support shared umems between queues and devices")
> Signed-off-by: lic121 <[email protected]>
> ---

Applied to bpf-next as well. Changed the name to Cheng Li while
applying. Thanks.

> tools/lib/bpf/xsk.c | 11 +++++++++++
> 1 file changed, 11 insertions(+)
>
> diff --git a/tools/lib/bpf/xsk.c b/tools/lib/bpf/xsk.c
> index edafe56..32a2f57 100644
> --- a/tools/lib/bpf/xsk.c
> +++ b/tools/lib/bpf/xsk.c
> @@ -1193,12 +1193,23 @@ int xsk_socket__create(struct xsk_socket **xsk_ptr, const char *ifname,
>
> int xsk_umem__delete(struct xsk_umem *umem)
> {
> + struct xdp_mmap_offsets off;
> + int err;
> +
> if (!umem)
> return 0;
>
> if (umem->refcount)
> return -EBUSY;
>
> + err = xsk_get_mmap_offsets(umem->fd, &off);
> + if (!err && umem->fill_save && umem->comp_save) {
> + munmap(umem->fill_save->ring - off.fr.desc,
> + off.fr.desc + umem->config.fill_size * sizeof(__u64));
> + munmap(umem->comp_save->ring - off.cr.desc,
> + off.cr.desc + umem->config.comp_size * sizeof(__u64));
> + }
> +
> close(umem->fd);
> free(umem);
>
> --
> 1.8.3.1
>