[0/2] bpf: Retry access to a map in read-only mode

Message

Roberto Sassu May 30, 2022, 8:45 a.m. UTC

If a map is write-protected, for example by an eBPF program implementing
the bpf_map security hook, some read-like operations like show and dump
cannot be performed by bpftool even if bpftool has the right to do so.

The reason is that bpftool sets the open flags to zero, at the time it gets
a map file descriptor. The kernel interprets this as a request for full
access to the map (with read and write permissions).

The simple solution is to set only the necessary open flags for a requested
operation, so that only those operations requiring more privileges than the
ones granted by the enforcing eBPF programs are denied.

There are different ways to solve the problem. One would be to introduce a
new function to acquire a read-only file descriptor and use it from the
functions implementing read-like operations.

Or more simply, another is to attempt to get a read-only file descriptor in
the original function when the first request with full permissions failed.

This patch set implements the second solution in patch 1, and adds a
corresponding test in patch 2. Depending on the feedback, the first
solution can be implemented.

Roberto Sassu (2):
  libbpf: Retry map access with read-only permission
  selftests/bpf: Add test for retrying access to map with read-only perm

 tools/lib/bpf/bpf.c                           |  5 ++
 .../bpf/prog_tests/test_map_retry_access.c    | 54 +++++++++++++++++++
 .../selftests/bpf/progs/map_retry_access.c    | 36 +++++++++++++
 3 files changed, 95 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/test_map_retry_access.c
 create mode 100644 tools/testing/selftests/bpf/progs/map_retry_access.c