This patch series makes the following changes: * Adds a new bpf helper `bpf_access_process_vm` to read user space memory from a different task. * Adds the ability to create sleepable bpf iterator programs. As an example of how this will be used, at Meta we are using bpf task iterator programs and this new bpf helper to read C++ async stack traces of a running process for debugging C++ binaries in production. Changes since v4: * Make `flags` into u64. * Use `user_ptr` arg name to be consistent with `bpf_copy_from_user`. * Add an extra check in selftests to verify access_process_vm calls succeeded. Changes since v3: * Check if `flags` is 0 and return -EINVAL if not. * Rebase on latest bpf-next branch and fix merge conflicts. Changes since v2: * Reorder arguments in `bpf_access_process_vm` to match existing related bpf helpers (e.g. `bpf_probe_read_kernel`, `bpf_probe_read_user`, `bpf_copy_from_user`). * `flags` argument is provided for future extensibility and is not currently used, and we always invoke `access_process_vm` with no flags. * Merge bpf helper patch and `bpf_iter_run_prog` patch together for better bisectability in case of failures. * Clean up formatting and comments in selftests. Changes since v1: * Fixed "Invalid wait context" issue in `bpf_iter_run_prog` by using `rcu_read_lock_trace()` for sleepable bpf iterator programs. Kenny Yu (3): bpf: Add bpf_access_process_vm() helper libbpf: Add "iter.s" section for sleepable bpf iterator programs selftests/bpf: Add test for sleepable bpf iterator programs include/linux/bpf.h | 1 + include/uapi/linux/bpf.h | 11 +++++ kernel/bpf/bpf_iter.c | 20 ++++++--- kernel/bpf/helpers.c | 23 ++++++++++ kernel/trace/bpf_trace.c | 2 + tools/include/uapi/linux/bpf.h | 11 +++++ tools/lib/bpf/libbpf.c | 1 + .../selftests/bpf/prog_tests/bpf_iter.c | 16 +++++++ .../selftests/bpf/progs/bpf_iter_task.c | 44 +++++++++++++++++++ 9 files changed, 124 insertions(+), 5 deletions(-) -- 2.30.2
