On Tue, Aug 13, 2024 at 4:03 PM Andrii Nakryiko <andrii@xxxxxxxxxx> wrote:
>
> This patch set extracts all the BPF-related changes done in [0] into
> a separate series based on top of stable-struct_fd branch ([1]) merged into
> bpf-next tree. There are also a few changes, additions, and adjustments:
>
> - patch subjects adjusted to use "bpf: " prefix consistently;
> - patch #2 is extracting bpf-related changes from original patch #19
> ("fdget_raw() users: switch to CLASS(fd_raw, ...)") and is ordered a bit
> earlier in this patch set;
> - patch #3 is reimplemented and replaces original patch #17
> ("bpf: resolve_pseudo_ldimm64(): take handling of a single ldimm64 insn into helper")
> completely;
> - in patch #4 ("bpf: switch maps to CLASS(fd, ...)"), which was originally
> patch #18 ("bpf maps: switch to CLASS(fd, ...)"), I've combined
> __bpf_get_map() and bpf_file_to_map() into __bpf_get_map(), as the latter
> is only used from it and makes no sense to keep separate;
> - as part of rebasing patch #4, I adjusted newly added in patch #3
> add_used_map_from_fd() function to use CLASS(fd, ...), as now
> __bpf_get_map() doesn't do its own fdput() anymore. This made unnecessary
> any further bpf_map_inc() changes, because we still rely on struct fd to
> keep map's file reference alive;
> - patches #5 and #6 are BPF-specific bits extracted from original patch #23
> ("fdget(), trivial conversions") and #24 ("fdget(), more trivial conversions");
> - patch #7 constifies security_bpf_token_create() LSM hook;
> - patch #8 is original patch #35 ("convert bpf_token_create()"), with
> path_get()+path_put() removed now that LSM hook above was adjusted.
>
> All these patches were pushed into a separate bpf-next/struct_fd branch ([2]).
> They were also merged into bpf-next/for-next so they can get early testing in
> linux-next.
>
> [0] https://lore.kernel.org/bpf/20240730050927.GC5334@ZenIV/
> [1] https://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs.git/log/?h=stable-struct_fd
> [2] https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next.git/log/?h=struct_fd
>
> Al Viro (6):
> bpf: convert __bpf_prog_get() to CLASS(fd, ...)
> bpf: switch fdget_raw() uses to CLASS(fd_raw, ...)
> bpf: switch maps to CLASS(fd, ...)
> bpf: trivial conversions for fdget()
> bpf: more trivial fdget() conversions
> bpf: convert bpf_token_create() to CLASS(fd, ...)
>
> Andrii Nakryiko (2):
> bpf: factor out fetching bpf_map from FD and adding it to used_maps
> list
> security,bpf: constify struct path in bpf_token_create() LSM hook
>
Al, Christian,
Can you guys please take a look and let us know if this looks sane and
fine to you? I kept Al's patches mostly intact (see my notes in the
cover letter above), and patch #3 does the refactoring I proposed
earlier, keeping explicit fdput() temporarily, until Al's
__bpf_map_get() refactoring which allows and nice and simple CLASS(fd)
conversion.
I think we end up at exactly what the end goal of the original series
is: using CLASS(fd, ...) throughout with all the benefits.
> include/linux/bpf.h | 11 +-
> include/linux/lsm_hook_defs.h | 2 +-
> include/linux/security.h | 4 +-
> kernel/bpf/bpf_inode_storage.c | 24 ++---
> kernel/bpf/btf.c | 11 +-
> kernel/bpf/map_in_map.c | 38 ++-----
> kernel/bpf/syscall.c | 181 +++++++++------------------------
> kernel/bpf/token.c | 74 +++++---------
> kernel/bpf/verifier.c | 110 +++++++++++---------
> net/core/sock_map.c | 23 ++---
> security/security.c | 2 +-
> security/selinux/hooks.c | 2 +-
> 12 files changed, 179 insertions(+), 303 deletions(-)
>
> --
> 2.43.5
>
