All of lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH bpf-next v2 0/1] Refactor cgroup_bpf internals to use more specific attach_type
@ 2021-08-19  9:24 Dave Marchevsky
  2021-08-19  9:24 ` [PATCH bpf-next v2 1/1] bpf: migrate cgroup_bpf to internal cgroup_bpf_attach_type enum Dave Marchevsky
  2021-08-24  1:00 ` [PATCH bpf-next v2 0/1] Refactor cgroup_bpf internals to use more specific attach_type patchwork-bot+netdevbpf
  0 siblings, 2 replies; 3+ messages in thread
From: Dave Marchevsky @ 2021-08-19  9:24 UTC (permalink / raw)
  To: bpf
  Cc: netdev, Alexei Starovoitov, Daniel Borkmann, Andrii Nakryiko,
	Martin KaFai Lau, Song Liu, Yonghong Song, John Fastabend,
	KP Singh, Dave Marchevsky

The cgroup_bpf struct has a few arrays (effective, progs, and flags) of
size MAX_BPF_ATTACH_TYPE. These are meant to separate progs by their
attach type, currently represented by the bpf_attach_type enum.

There are some bpf_attach_type values which are not valid attach types
for cgroup bpf programs. Programs with these attach types will never be
handled by cgroup_bpf_{attach,detach} and thus will never be held in
cgroup_bpf structs. Even if such programs did make it into their
reserved slot in those arrays, they would never be executed.

Accordingly we can migrate to a new internal cgroup_bpf-specific enum
for these arrays, saving some bytes per cgroup and making it more
obvious which BPF programs belong there. netns_bpf_attach_type is an
existing example of this pattern, let's do similar for cgroup_bpf.

v1->v2: Address Daniel's comments
	* Reverse xmas tree ordering for def changes
	* Helper macro to reduce to_cgroup_bpf_attach_type boilerplate
		* checkpatch.pl complains: "ERROR: Macros with complex values should
		be enclosed in parentheses". Found some existing macros (do 'git grep
		"define case"') which get same complaint. Think it's fine to keep
		as-is since it's immediately undef'd.
	* Remove CG_BPF_ prefix from cgroup_bpf_attach_type
		* Although I agree that the prefix is redundant, the de-prefixed
		names feel a bit too 'general' given the internal use of the enum.
		e.g. when someone sees CGROUP_INET6_BIND it's not obvious that it
		should only be used in certain ways internally.
		* Don't feel strongly about this, just my thoughts as a noob to the
		internals.
	* Rebase onto latest bpf-next/master
		* No significant conflicts, some small boilerplate adjustments
		needed to catch up to Andrii's "bpf: Refactor BPF_PROG_RUN_ARRAY
		family of macros into functions" change

Dave Marchevsky (1):
  bpf: migrate cgroup_bpf to internal cgroup_bpf_attach_type enum

 include/linux/bpf-cgroup.h     | 183 ++++++++++++++++++++++-----------
 include/uapi/linux/bpf.h       |   2 +-
 kernel/bpf/cgroup.c            | 156 ++++++++++++++++------------
 net/ipv4/af_inet.c             |   6 +-
 net/ipv4/udp.c                 |   2 +-
 net/ipv6/af_inet6.c            |   6 +-
 net/ipv6/udp.c                 |   2 +-
 tools/include/uapi/linux/bpf.h |   2 +-
 8 files changed, 227 insertions(+), 132 deletions(-)

-- 
2.30.2


^ permalink raw reply	[flat|nested] 3+ messages in thread
end of thread, other threads:[~2021-08-24  1:02 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-08-19  9:24 [PATCH bpf-next v2 0/1] Refactor cgroup_bpf internals to use more specific attach_type Dave Marchevsky
2021-08-19  9:24 ` [PATCH bpf-next v2 1/1] bpf: migrate cgroup_bpf to internal cgroup_bpf_attach_type enum Dave Marchevsky
2021-08-24  1:00 ` [PATCH bpf-next v2 0/1] Refactor cgroup_bpf internals to use more specific attach_type patchwork-bot+netdevbpf

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.