selftests/bpf: test for bpf_get_file_path() from raw tracepoint
Subsystems (5)
PATCH v2 1/1 selftests/bpf: test for bpf_get_file_path() from raw tracepoint
trace fstat events by raw tracepoint sys_enter:newfstat, and handle events only produced by fd2path_loadgen, the fd2path_loadgen call fstat on several different types of files to test bpf_get_file_path's feature.
--- tools/testing/selftests/bpf/Makefile | 8 +- tools/testing/selftests/bpf/fd2path_loadgen.c | 75 ++++++++++ .../selftests/bpf/prog_tests/get_file_path.c | 130 ++++++++++++++++++ .../selftests/bpf/progs/test_get_file_path.c | 58 ++++++++ 4 files changed, 269 insertions(+), 2 deletions(-) create mode 100644 tools/testing/selftests/bpf/fd2path_loadgen.c create mode 100644 tools/testing/selftests/bpf/prog_tests/get_file_path.c create mode 100644 tools/testing/selftests/bpf/progs/test_get_file_path.c
tools/testing/selftests/bpf/Makefile +6 -2
@@ -72,7 +72,7 @@ TEST_GEN_PROGS_EXTENDED = test_sock_addr test_skb_cgroup_id_user \ flow_dissector_load test_flow_dissector test_tcp_check_syncookie_user \ test_lirc_mode2_user -TEST_CUSTOM_PROGS = urandom_read +TEST_CUSTOM_PROGS = urandom_read fd2path_loadgen include ../lib.mk
@@ -92,6 +92,9 @@ $(OUTPUT)/urandom_read: urandom_read.c $(OUTPUT)/test_stub.o: test_stub.c $(CC) -c $(CFLAGS) -o $@ $< +$(OUTPUT)/fd2path_loadgen: fd2path_loadgen.c + $(CC) -o $@ $< -Wl,--build-id + BPFOBJ := $(OUTPUT)/libbpf.a $(TEST_GEN_PROGS) $(TEST_GEN_PROGS_EXTENDED): $(OUTPUT)/test_stub.o $(BPFOBJ)
@@ -266,7 +269,8 @@ TRUNNER_TESTS_DIR := prog_tests TRUNNER_BPF_PROGS_DIR := progs TRUNNER_EXTRA_SOURCES := test_progs.c cgroup_helpers.c trace_helpers.c \ flow_dissector_load.h -TRUNNER_EXTRA_FILES := $(OUTPUT)/urandom_read \ +TRUNNER_EXTRA_FILES := $(OUTPUT)/urandom_read \ + $(OUTPUT)/fd2path_loadgen \ $(wildcard progs/btf_dump_test_case_*.c) TRUNNER_BPF_BUILD_RULE := CLANG_BPF_BUILD_RULE TRUNNER_BPF_CFLAGS := -I. -I$(OUTPUT) $(BPF_CFLAGS) $(CLANG_CFLAGS)
tools/testing/selftests/bpf/fd2path_loadgen.c new file +75 -0
75 lines not on this page · raw message
tools/testing/selftests/bpf/prog_tests/get_file_path.c new file +130 -0
130 lines not on this page · raw message
tools/testing/selftests/bpf/progs/test_get_file_path.c new file +58 -0
58 lines not on this page · raw message
Signature · 1 line
2.17.1
Quoted 5 lines · Wenbo Zhang
trace fstat events by raw tracepoint sys_enter:newfstat, and handle events only produced by
fd2path_loadgen, thefd2path_loadgencall fstat on several different types of files to test bpf_get_file_path's feature.
---
Unless there is a real reason for all this complexity (in which case, please spell it out in commit or comments), I think this could be so much simpler.
- you don't have to use perf_buffer to pass data back, just use global data;
- you can add a filter for PID to only capture data triggered by test process and avoid the noise;
- why all those set_affinity dances? Is it just because you used existing perf_buffer test which did that to specifically test perf_buffer delivering data across every CPU core? Seems like your test doesn't care about that...
- do we really need a separate binary generating hundreds of syscalls? It's hard to synchronize with test and it seems much simpler to just trigger necessary syscalls synchronously from the test itself, no?
I have a bunch of more minutia nits, but most of them will go away if you simplify your testing approach anyway, so I'll postpone them till then.
Quoted 9 lines
tools/testing/selftests/bpf/Makefile | 8 +- tools/testing/selftests/bpf/fd2path_loadgen.c | 75 ++++++++++ .../selftests/bpf/prog_tests/get_file_path.c | 130 ++++++++++++++++++ .../selftests/bpf/progs/test_get_file_path.c | 58 ++++++++ 4 files changed, 269 insertions(+), 2 deletions(-) create mode 100644 tools/testing/selftests/bpf/fd2path_loadgen.c create mode 100644 tools/testing/selftests/bpf/prog_tests/get_file_path.c create mode 100644 tools/testing/selftests/bpf/progs/test_get_file_path.c
[...]
Quoted 3 lines
Unless there is a real reason for all this complexity (in which case, please spell it out in commit or comments), I think this could be so much simpler.
Yes, it could be so much simpler than this implement.
Quoted 3 lines
- you don't have to use
perf_bufferto pass data back, just use global data;
- you can add a filter for PID to only capture data triggered by test process and avoid the noise;
Yes, I'll use map instead of perf buffer to communicate.
Quoted 4 lines
- why all those
set_affinitydances? Is it just because you used existingperf_buffertest which did that to specifically testperf_bufferdelivering data across every CPU core? Seems like your test doesn't care about that...
I used fd2path_loadgen to get hundreds of syscalls between multi usleep, which may cause it's sched to different cores, but as you say, this test doesn't care about
that... I'll remove them with perf buffer.
Quoted 3 lines
- do we really need a separate binary generating hundreds of syscalls? It's hard to synchronize with test and it seems much simpler to just trigger necessary syscalls synchronously from the test itself, no?
This is my fault, necessary syscalls synchronously from the test itself is enough.
Quoted 3 lines
I have a bunch of more minutia nits, but most of them will go away if you simplify your testing approach anyway, so I'll postpone them till then.
Thanks a lot. I'll modify a simplified version then recommitted.
Quoted 38 lines · Andrii Nakryiko <[email protected]> 于2019年11月2日周六 下午1:49写道:
Quoted 5 lines · Wenbo Zhang
trace fstat events by raw tracepoint sys_enter:newfstat, and handle events only produced by
fd2path_loadgen, thefd2path_loadgencall fstat on several different types of files to test bpf_get_file_path's feature.
---Unless there is a real reason for all this complexity (in which case, please spell it out in commit or comments), I think this could be so much simpler.
- you don't have to use
perf_bufferto pass data back, just use global data;
- you can add a filter for PID to only capture data triggered by test process and avoid the noise;
- why all thoseset_affinitydances? Is it just because you used existingperf_buffertest which did that to specifically testperf_bufferdelivering data across every CPU core? Seems like your test doesn't care about that...
- do we really need a separate binary generating hundreds of syscalls? It's hard to synchronize with test and it seems much simpler to just trigger necessary syscalls synchronously from the test itself, no?I have a bunch of more minutia nits, but most of them will go away if you simplify your testing approach anyway, so I'll postpone them till then.
Quoted 9 lines
tools/testing/selftests/bpf/Makefile | 8 +- tools/testing/selftests/bpf/fd2path_loadgen.c | 75 ++++++++++ .../selftests/bpf/prog_tests/get_file_path.c | 130 ++++++++++++++++++ .../selftests/bpf/progs/test_get_file_path.c | 58 ++++++++ 4 files changed, 269 insertions(+), 2 deletions(-) create mode 100644 tools/testing/selftests/bpf/fd2path_loadgen.c create mode 100644 tools/testing/selftests/bpf/prog_tests/get_file_path.c create mode 100644 tools/testing/selftests/bpf/progs/test_get_file_path.c[...]