[PATCH bpf-next RFC v1 0/5] Implement file local storage

* [PATCH bpf-next RFC v1 0/5] Implement file local storage
@ 2021-08-21 18:48 Kumar Kartikeya Dwivedi
  2021-08-21 18:48 ` [PATCH bpf-next RFC v1 1/5] bpf: " Kumar Kartikeya Dwivedi
                   ` (4 more replies)
  0 siblings, 5 replies; 6+ messages in thread
From: Kumar Kartikeya Dwivedi @ 2021-08-21 18:48 UTC (permalink / raw)
  To: bpf
  Cc: Kumar Kartikeya Dwivedi, Alexei Starovoitov, Daniel Borkmann,
	Andrii Nakryiko, Martin KaFai Lau, KP Singh, Spencer Baugh,
	Andy Lutomirski, Pavel Emelyanov, Alexander Mihalicyn,
	Andrei Vagin, linux-security-module

This series implements a file local storage map for eBPF LSM programs. This
allows to tie lifetime of data in the map to an open file description (in POSIX
parlance). Like other local storage map types, lifetime of data is tied to the
struct file instance.

The main purpose is a general purpose map keyed by fd where the open file
underlying the fd (struct file *) serves as the key into the map. It is possible
to use struct file * from kernelspace, but sharing update access with userspace
means userspace has no way except kcmp-aring with another known fd with a key.
This is pretty wasteful.

It can also be used to treat the map as a set of files that have been added to
it, such that multiples sets can be looked up for matching purposes in O(1)
instead of O(n) using kcmp(2) from userspace (for same struct file *).

There are multiple other usecases served by this map. One of the motivating ones
is the ability to now implement a Capsicum [0] style capability based sandbox
using eBPF LSM, but the actual mechanism is much more generic and allows
applications to enforce rights of their own per open file that they delegate to
other users by conventional fd-passing on UNIX (dup/fork/SCM_RIGHTS).

The implementation is exactly the same as bpf_inode_storage, except some
modifications to use struct file * as the key instead of struct inode *.

[0]: https://www.usenix.org/legacy/event/sec10/tech/full_papers/Watson.pdf

Kumar Kartikeya Dwivedi (5):
  bpf: Implement file local storage
  tools: sync bpf.h header
  libbpf: Add bpf_probe_map_type support for file local storage
  tools: bpf: update bpftool for file_storage map
  tools: testing: Add selftest for file local storage map

 include/linux/bpf_lsm.h                       |  21 ++
 include/linux/bpf_types.h                     |   1 +
 include/uapi/linux/bpf.h                      |  39 +++
 kernel/bpf/Makefile                           |   2 +-
 kernel/bpf/bpf_file_storage.c                 | 244 ++++++++++++++++++
 kernel/bpf/bpf_lsm.c                          |   4 +
 kernel/bpf/syscall.c                          |   3 +-
 kernel/bpf/verifier.c                         |  10 +
 security/bpf/hooks.c                          |   2 +
 .../bpf/bpftool/Documentation/bpftool-map.rst |   2 +-
 tools/bpf/bpftool/bash-completion/bpftool     |   3 +-
 tools/bpf/bpftool/map.c                       |   3 +-
 tools/include/uapi/linux/bpf.h                |  39 +++
 tools/lib/bpf/libbpf_probes.c                 |   1 +
 .../bpf/prog_tests/test_local_storage.c       |  51 ++++
 .../selftests/bpf/progs/local_storage.c       |  23 ++
 16 files changed, 443 insertions(+), 5 deletions(-)
 create mode 100644 kernel/bpf/bpf_file_storage.c

-- 
2.33.0

^ permalink raw reply	[flat|nested] 6+ messages in thread