1.. SPDX-License-Identifier: GPL-2.0
  2
  3============================
  4BPF_PROG_TYPE_CGROUP_SOCKOPT
  5============================
  6
  7``BPF_PROG_TYPE_CGROUP_SOCKOPT`` program type can be attached to two
  8cgroup hooks:
  9
 10* ``BPF_CGROUP_GETSOCKOPT`` - called every time process executes ``getsockopt``
 11  system call.
 12* ``BPF_CGROUP_SETSOCKOPT`` - called every time process executes ``setsockopt``
 13  system call.
 14
 15The context (``struct bpf_sockopt``) has associated socket (``sk``) and
 16all input arguments: ``level``, ``optname``, ``optval`` and ``optlen``.
 17
 18BPF_CGROUP_SETSOCKOPT
 19=====================
 20
 21``BPF_CGROUP_SETSOCKOPT`` is triggered *before* the kernel handling of
 22sockopt and it has writable context: it can modify the supplied arguments
 23before passing them down to the kernel. This hook has access to the cgroup
 24and socket local storage.
 25
 26If BPF program sets ``optlen`` to -1, the control will be returned
 27back to the userspace after all other BPF programs in the cgroup
 28chain finish (i.e. kernel ``setsockopt`` handling will *not* be executed).
 29
 30Note, that ``optlen`` can not be increased beyond the user-supplied
 31value. It can only be decreased or set to -1. Any other value will
 32trigger ``EFAULT``.
 33
 34Return Type
 35-----------
 36
 37* ``0`` - reject the syscall, ``EPERM`` will be returned to the userspace.
 38* ``1`` - success, continue with next BPF program in the cgroup chain.
 39
 40BPF_CGROUP_GETSOCKOPT
 41=====================
 42
 43``BPF_CGROUP_GETSOCKOPT`` is triggered *after* the kernel handing of
 44sockopt. The BPF hook can observe ``optval``, ``optlen`` and ``retval``
 45if it's interested in whatever kernel has returned. BPF hook can override
 46the values above, adjust ``optlen`` and reset ``retval`` to 0. If ``optlen``
 47has been increased above initial ``getsockopt`` value (i.e. userspace
 48buffer is too small), ``EFAULT`` is returned.
 49
 50This hook has access to the cgroup and socket local storage.
 51
 52Note, that the only acceptable value to set to ``retval`` is 0 and the
 53original value that the kernel returned. Any other value will trigger
 54``EFAULT``.
 55
 56Return Type
 57-----------
 58
 59* ``0`` - reject the syscall, ``EPERM`` will be returned to the userspace.
 60* ``1`` - success: copy ``optval`` and ``optlen`` to userspace, return
 61  ``retval`` from the syscall (note that this can be overwritten by
 62  the BPF program from the parent cgroup).
 63
 64Cgroup Inheritance
 65==================
 66
 67Suppose, there is the following cgroup hierarchy where each cgroup
 68has ``BPF_CGROUP_GETSOCKOPT`` attached at each level with
 69``BPF_F_ALLOW_MULTI`` flag::
 70
 71  A (root, parent)
 72   \
 73    B (child)
 74
 75When the application calls ``getsockopt`` syscall from the cgroup B,
 76the programs are executed from the bottom up: B, A. First program
 77(B) sees the result of kernel's ``getsockopt``. It can optionally
 78adjust ``optval``, ``optlen`` and reset ``retval`` to 0. After that
 79control will be passed to the second (A) program which will see the
 80same context as B including any potential modifications.
 81
 82Same for ``BPF_CGROUP_SETSOCKOPT``: if the program is attached to
 83A and B, the trigger order is B, then A. If B does any changes
 84to the input arguments (``level``, ``optname``, ``optval``, ``optlen``),
 85then the next program in the chain (A) will see those changes,
 86*not* the original input ``setsockopt`` arguments. The potentially
 87modified values will be then passed down to the kernel.
 88
 89Large optval
 90============
 91When the ``optval`` is greater than the ``PAGE_SIZE``, the BPF program
 92can access only the first ``PAGE_SIZE`` of that data. So it has to options:
 93
 94* Set ``optlen`` to zero, which indicates that the kernel should
 95  use the original buffer from the userspace. Any modifications
 96  done by the BPF program to the ``optval`` are ignored.
 97* Set ``optlen`` to the value less than ``PAGE_SIZE``, which
 98  indicates that the kernel should use BPF's trimmed ``optval``.
 99
100When the BPF program returns with the ``optlen`` greater than
101``PAGE_SIZE``, the userspace will receive ``EFAULT`` errno.
102
103Example
104=======
105
106See ``tools/testing/selftests/bpf/progs/sockopt_sk.c`` for an example
107of BPF program that handles socket options.