test_verifier tests migration to inline assembly
Subsystems (3)
Eduard Zingerman
maintainercover test_verifier tests migration to inline assembly
As a part of the discussion started in [2] I developed a script [1] that can convert tests specified in test_verifier.c format to inline BPF assembly compatible for use with test_loader.c.
For example, test definition like below:
{
"invalid and of negative number",
.insns = {
BPF_ST_MEM(BPF_DW, BPF_REG_10, -8, 0),
BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),
BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -8),
BPF_LD_MAP_FD(BPF_REG_1, 0),
BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 5),
BPF_LDX_MEM(BPF_B, BPF_REG_1, BPF_REG_0, 0),
BPF_ALU64_IMM(BPF_AND, BPF_REG_1, -4),
BPF_ALU64_IMM(BPF_LSH, BPF_REG_1, 2),
BPF_ALU64_REG(BPF_ADD, BPF_REG_0, BPF_REG_1),
BPF_ST_MEM(BPF_DW, BPF_REG_0, 0, offsetof(struct test_val, foo)),
BPF_EXIT_INSN(),
},
.fixup_map_hash_48b = { 3 },
.errstr = "R0 max value is outside of the allowed memory range",
.result = REJECT,
.flags = F_NEEDS_EFFICIENT_UNALIGNED_ACCESS,
},Is converted to:
struct test_val { ... };
struct { ...} map_hash_48b SEC(".maps");
__description("invalid and of negative number")__failure __msg("R0 max value is outside of the allowed memory range") __failure_unpriv__flag(BPF_F_ANY_ALIGNMENT)SEC("socket")__naked void invalid_and_of_negative_number(void)
{
asm volatile (
" r1 = 0; \n\
*(u64*)(r10 - 8) = r1; \n\
r2 = r10; \n\
r2 += -8; \n\
r1 = %[map_hash_48b] ll; \n\
call %[bpf_map_lookup_elem]; \n\
if r0 == 0 goto l0_%=; \n\
r1 = *(u8*)(r0 + 0); \n\
r1 &= -4; \n\
r1 <<= 2; \n\
r0 += r1; \n\
r1 = %[test_val_foo_offset]; \n\
*(u64*)(r0 + 0) = r1; \n\
l0_%=: \n\
exit; \n\
" :
: [test_val_foo_offset]"i"(offsetof(struct test_val, foo)),
__imm(bpf_map_lookup_elem),
__imm_addr(map_hash_48b)
: __clobber_all);
}The script introduces labels for gotos and calls, immediate values for complex expressions like `offsetof(...)', takes care of map instructions patching, inserts map declarations used in the test, transfers comments from test, test fields and instructions. There are some issues with BPF_ST_MEM instruction support as described in [4], thus the script replaces such instructions with pairs of MOV/STX_MEM instructions.
This patch-set introduces changes necessary for test_loader.c and includes migration of a single test from test_verifier to test_progs format, here are descriptions for individual patches:
1. "Support custom per-test flags and multiple expected messages" This patch was shared by Andrii Nakryiko [3], it adds capability to specify flags required by the BPF program.
2. "Unprivileged tests for test_loader.c"
Extends test_loader.c by capability to load test programs in unprivileged mode and specify separate test outcomes for privileged and unprivileged modes.
Note: for a reason I do not understand I had to use different set of capabilities compared to test_verifier:
- test_verifier: CAP_NET_ADMIN, CAP_PERFMON, CAP_BPF;
- test_loader : CAP_SYS_ADMIN, CAP_PERFMON, CAP_BPF.
3. "Generate boilerplate code for test_loader-based tests" Extends selftests/bpf Makefile to automatically generate prog_tests/tests.h entry that uses test_loader for progs/*.c BPF programs if special marker is present.
4. "__imm_insn macro to embed raw insns in inline asm"
This macro can be generated by migration script for instructions that cannot be expressed in inline assembly, e.g. invalid instructions.
5. "convert jeq_infer_not_null tests to inline assembly" Shows an example of the test migration.
The test was updated slightly after automatic translation:
- unnecessary comments removed;
- functions renamed;
- some labels renamed.
The migration script has some limitations:
- Technical, test features that are not yet supported:
- few instructions like BPF_ENDIAN;
- macro like BPF_SK_LOOKUP or BPF_LD_MAP_VALUE;
- program types like BPF_PROG_TYPE_CGROUP_SOCK_ADDR and
BPF_PROG_TYPE_TRACING;
- tests that specify fields 'expected_attach_type' or 'insn_processed';
- a few tests expose complex structure that could not be automatically migrated, e.g.: 'atomic_fetch', 'lwt', 'bpf_loop_inline'.
- Tests that use .run field cannot be migrated as test_loader.c tests.
- Tests with generated bodies, e.g. test_verifier.c:bpf_fill_scale() cannot be migrated as test_loader.c tests.
- LLVM related:
- BPF_ST instruction is not supported by LLVM BPF assembly yet
(I'll take care of it);
- Issues with parsing of some assembly instructions like
"r0 = cmpxchg_64(r10 - 8, r0, r5)"
(I'll take care of it);
- libbpf related:
- libbpf does not support call instructions within a single
function, e.g.:
0: r1 = 1
1: r2 = 2
2: call 1
3: exit
4: r0 = r1
5: r0 += r2
6: exit
This pattern is common in verifier tests, I see two possible
mitigation:
(a) use an additional libbpf flag to allow such code;
(b) extend migration script to split such code in several functions.
I like (a) more.- libbpf does not allow empty programs.
All in all the current script stats are as follows:
- 62 out of 93 files from progs/*.c can be converted w/o warnings;
- 55 converted files could be compiled;
- 40 pass testing, 15 fail.
By submitting this RFC I seek the following feedback:
- is community interested in such migration?
- if yes, should I pursue partial or complete tests migration?
- in case of partial migration which tests should be prioritized?
- should I offer migrated tests one by one or in big butches?
Thanks,
Eduard
[1] https://github.com/eddyz87/verifier-tests-migrator
[2] https://lore.kernel.org/bpf/CAEf4BzYPsDWdRgx+ND1wiKAB62P=WwoLhr2uWkbVpQfbHqi1oA@mail.gmail.com/
[3] https://lore.kernel.org/bpf/CAEf4BzZH0ZxorCi7nPDbRqSK9f+410RooNwNJGwfw8=0a5i1nw@mail.gmail.com/
[4] https://lore.kernel.org/bpf/[email protected]/
Andrii Nakryiko (1):
selftests/bpf: support custom per-test flags and multiple expected
messages
Eduard Zingerman (4):
selftests/bpf: unprivileged tests for test_loader.c selftests/bpf: generate boilerplate code for test_loader-based tests selftests/bpf: __imm_insn macro to embed raw insns in inline asm selftests/bpf: convert jeq_infer_not_null tests to inline assembly
tools/testing/selftests/bpf/Makefile | 41 +- tools/testing/selftests/bpf/autoconf_helper.h | 9 + .../selftests/bpf/prog_tests/.gitignore | 1 + .../bpf/prog_tests/jeq_infer_not_null.c | 9 - tools/testing/selftests/bpf/progs/bpf_misc.h | 49 +++ .../selftests/bpf/progs/jeq_infer_not_null.c | 186 +++++++++ .../bpf/progs/jeq_infer_not_null_fail.c | 1 + tools/testing/selftests/bpf/test_loader.c | 370 +++++++++++++++--- tools/testing/selftests/bpf/test_progs.h | 1 + tools/testing/selftests/bpf/test_verifier.c | 25 +- tools/testing/selftests/bpf/unpriv_helpers.c | 26 ++ tools/testing/selftests/bpf/unpriv_helpers.h | 7 + .../bpf/verifier/jeq_infer_not_null.c | 174 -------- 13 files changed, 625 insertions(+), 274 deletions(-) create mode 100644 tools/testing/selftests/bpf/autoconf_helper.h delete mode 100644 tools/testing/selftests/bpf/prog_tests/jeq_infer_not_null.c create mode 100644 tools/testing/selftests/bpf/progs/jeq_infer_not_null.c create mode 100644 tools/testing/selftests/bpf/unpriv_helpers.c create mode 100644 tools/testing/selftests/bpf/unpriv_helpers.h delete mode 100644 tools/testing/selftests/bpf/verifier/jeq_infer_not_null.c
Signature · 1 line
2.39.0
Eduard Zingerman
maintainerRFC 1/5 selftests/bpf: support custom per-test flags and multiple expected messages
From: Andrii Nakryiko <[email protected]>
Extend __flag attribute by allowing to specify one of the following:
* BPF_F_STRICT_ALIGNMENT
* BPF_F_ANY_ALIGNMENT
* BPF_F_TEST_RND_HI32
* BPF_F_TEST_STATE_FREQ
* BPF_F_SLEEPABLE
* BPF_F_XDP_HAS_FRAGS
* Some numeric value
Extend __msg attribute by allowing to specify multiple exepcted messages. All messages are expected to be present in the verifier log in the order of application.
--- tools/testing/selftests/bpf/test_loader.c | 69 ++++++++++++++++++++--- tools/testing/selftests/bpf/test_progs.h | 1 + 2 files changed, 61 insertions(+), 9 deletions(-)
tools/testing/selftests/bpf/test_loader.c +60 -9
@@ -13,12 +13,15 @@#define TEST_TAG_EXPECT_SUCCESS "comment:test_expect_success"#define TEST_TAG_EXPECT_MSG_PFX "comment:test_expect_msg="#define TEST_TAG_LOG_LEVEL_PFX "comment:test_log_level=" +#define TEST_TAG_PROG_FLAGS_PFX "comment:test_prog_flags="struct test_spec {const char *name;bool expect_failure; -const char *expect_msg; +const char **expect_msgs; +size_t expect_msg_cnt;int log_level; +int prog_flags; };static int tester_init (struct test_loader *tester)
@@ -67,7 +70,8 @@ static int parse_test_spec(struct test_loader *tester,for (i =1 ; i <btf__type_cnt (btf); i++) {const struct btf_type *t; -const char *s; +const char *s, *val; +char *e; t =btf__type_by_id (btf, i);if (!btf_is_decl_tag (t))
@@ -82,14 +86,48 @@ static int parse_test_spec(struct test_loader *tester, }else if (strcmp (s,TEST_TAG_EXPECT_SUCCESS ) ==0 ) { spec->expect_failure = false; }else if (str_has_pfx (s,TEST_TAG_EXPECT_MSG_PFX )) { - spec->expect_msg = s +sizeof (TEST_TAG_EXPECT_MSG_PFX ) -1 ; +void *tmp; +const char **msg; + + tmp =realloc (spec->expect_msgs, + (1 + spec->expect_msg_cnt) *sizeof (void *)); +if (!tmp) { +ASSERT_FAIL ("failed to realloc memory for messages\n" ); +return -ENOMEM ; + } + spec->expect_msgs = tmp; + msg = &spec->expect_msgs[spec->expect_msg_cnt++]; + *msg = s +sizeof (TEST_TAG_EXPECT_MSG_PFX ) -1 ; }else if (str_has_pfx (s,TEST_TAG_LOG_LEVEL_PFX )) { + val = s +sizeof (TEST_TAG_LOG_LEVEL_PFX ) -1 ; errno =0 ; - spec->log_level =strtol (s +sizeof (TEST_TAG_LOG_LEVEL_PFX ) -1 ,NULL ,0 ); -if (errno) { + spec->log_level =strtol (val, &e,0 ); +if (errno || e[0 ] !='\0' ) {ASSERT_FAIL ("failed to parse test log level from '%s'" , s);return -EINVAL ; } + }else if (str_has_pfx (s,TEST_TAG_PROG_FLAGS_PFX )) { + val = s +sizeof (TEST_TAG_PROG_FLAGS_PFX ) -1 ; +if (strcmp (val,"BPF_F_STRICT_ALIGNMENT" ) ==0 ) { + spec->prog_flags |=BPF_F_STRICT_ALIGNMENT ; + }else if (strcmp (val,"BPF_F_ANY_ALIGNMENT" ) ==0 ) { + spec->prog_flags |=BPF_F_ANY_ALIGNMENT ; + }else if (strcmp (val,"BPF_F_TEST_RND_HI32" ) ==0 ) { + spec->prog_flags |=BPF_F_TEST_RND_HI32 ; + }else if (strcmp (val,"BPF_F_TEST_STATE_FREQ" ) ==0 ) { + spec->prog_flags |=BPF_F_TEST_STATE_FREQ ; + }else if (strcmp (val,"BPF_F_SLEEPABLE" ) ==0 ) { + spec->prog_flags |=BPF_F_SLEEPABLE ; + }else if (strcmp (val,"BPF_F_XDP_HAS_FRAGS" ) ==0 ) { + spec->prog_flags |=BPF_F_XDP_HAS_FRAGS ; + }else /* assume numeric value */ { + errno =0 ; + spec->prog_flags |=strtol (val, &e,0 ); +if (errno || e[0 ] !='\0' ) { +ASSERT_FAIL ("failed to parse test prog flags from '%s'" , s); +return -EINVAL ; + } + } } }
@@ -101,7 +139,7 @@ static void prepare_case(struct test_loader *tester,struct bpf_object *obj, struct bpf_program *prog) { - int min_log_level = 0; + int min_log_level = 0, prog_flags;if (env.verbosity >VERBOSE_NONE ) min_log_level =1 ;
@@ -119,7 +157,11 @@ static void prepare_case(struct test_loader *tester,else bpf_program__set_log_level (prog, spec->log_level); + prog_flags =bpf_program__flags (prog); +bpf_program__set_flags (prog, prog_flags | spec->prog_flags); + tester->log_buf[0 ] ='\0' ; + tester->next_match_pos =0 ; }static void emit_verifier_log (const char *log_buf,bool force)
@@ -135,17 +177,26 @@ static void validate_case(struct test_loader *tester,struct bpf_program *prog, int load_err) { -if (spec->expect_msg) { + int i, j; + +for (i =0 ; i < spec->expect_msg_cnt; i++) {char *match; +const char *expect_msg; + + expect_msg = spec->expect_msgs[i]; - match =strstr (tester->log_buf, spec->expect_msg); + match =strstr (tester->log_buf + tester->next_match_pos, expect_msg);if (!ASSERT_OK_PTR (match,"expect_msg" )) {/* if we are in verbose mode, we've already emitted log */ if (env.verbosity ==VERBOSE_NONE )emit_verifier_log (tester->log_buf, true/*force*/ ); -fprintf (stderr,"EXPECTED MSG: '%s'\n" , spec->expect_msg); +for (j =0 ; j < i; j++) +fprintf (stderr,"MATCHED MSG: '%s'\n" , spec->expect_msgs[j]); +fprintf (stderr,"EXPECTED MSG: '%s'\n" , expect_msg);return ; } + + tester->next_match_pos = match - tester->log_buf +strlen (expect_msg); } }
tools/testing/selftests/bpf/test_progs.h +1 -0
@@ -410,6 +410,7 @@ int write_sysctl(const char *sysctl, const char *value);struct test_loader {char *log_buf;size_t log_buf_sz; +size_t next_match_pos;struct bpf_object *obj; };
Signature · 1 line
2.39.0
Re: [RFC bpf-next 1/5] selftests/bpf: support custom per-test flags and multiple expected messages
Quoted 12 lines · Eduard Zingerman
From: Andrii Nakryiko <[email protected]>
Extend
__flagattribute by allowing to specify one of the following:
*BPF_F_STRICT_ALIGNMENT
*BPF_F_ANY_ALIGNMENT
*BPF_F_TEST_RND_HI32
*BPF_F_TEST_STATE_FREQ
*BPF_F_SLEEPABLE
*BPF_F_XDP_HAS_FRAGS
* Some numeric value
Quoted 8 lines
Extend
__msgattribute by allowing to specify multiple exepcted messages. All messages are expected to be present in the verifier log in the order of application.Signed-off-by Andrii Nakryiko <[email protected]>[ Eduard: added commit message, formatting ]Signed-off-by Eduard Zingerman <[email protected]>---
hey Eduard,
When you get a chance, can you please send this patch separately from the rest of the test_verifier rework patch set (it probably makes sense to also add #define __flags in this patch as well, given you are parsing its definition in this patch).
This would great help me with my work that uses all this assembly-level test facilities. Thanks!
Quoted 14 lines
tools/testing/selftests/bpf/test_loader.c | 69 ++++++++++++++++++++--- tools/testing/selftests/bpf/test_progs.h | 1 + 2 files changed, 61 insertions(+), 9 deletions(-)tools/testing/selftests/bpf/test_loader.c +1
-0@@ -13,12 +13,15 @@#define TEST_TAG_EXPECT_SUCCESS "comment:test_expect_success"#define TEST_TAG_EXPECT_MSG_PFX "comment:test_expect_msg="#define TEST_TAG_LOG_LEVEL_PFX "comment:test_log_level=" +#define TEST_TAG_PROG_FLAGS_PFX "comment:test_prog_flags="
[...]
Eduard Zingerman
maintainerRe: [RFC bpf-next 1/5] selftests/bpf: support custom per-test flags and multiple expected messages
Quoted 24 lines · Andrii Nakryiko
Quoted 12 lines · Eduard Zingerman
From: Andrii Nakryiko <[email protected]>
Extend
__flagattribute by allowing to specify one of the following:
*BPF_F_STRICT_ALIGNMENT
*BPF_F_ANY_ALIGNMENT
*BPF_F_TEST_RND_HI32
*BPF_F_TEST_STATE_FREQ
*BPF_F_SLEEPABLE
*BPF_F_XDP_HAS_FRAGS
* Some numeric valueQuoted 8 lines
Extend
__msgattribute by allowing to specify multiple exepcted messages. All messages are expected to be present in the verifier log in the order of application.Signed-off-by Andrii Nakryiko <[email protected]>[ Eduard: added commit message, formatting ]Signed-off-by Eduard Zingerman <[email protected]>---
hey Eduard,
Quoted 7 lines
When you get a chance, can you please send this patch separately from the rest of the
test_verifierrework patch set (it probably makes sense to also add #define__flagsin this patch as well, given you are parsing its definition in this patch).This would great help me with my work that uses all this assembly-level test facilities. Thanks!
Hi Andrii,
Rebase didn't change anything in the patch, I added __flags macro, some some comments, and started the CI job: [1].
Feels weird to post it, tbh, because it's 100% your code w/o added value from my side.
Thanks,
Eduard
[1] https://github.com/kernel-patches/bpf/pull/4688
Quoted 19 lines
Quoted 17 lines
tools/testing/selftests/bpf/test_loader.c | 69 ++++++++++++++++++++--- tools/testing/selftests/bpf/test_progs.h | 1 + 2 files changed, 61 insertions(+), 9 deletions(-)tools/testing/selftests/bpf/test_loader.c +1
-0@@ -13,12 +13,15 @@#define TEST_TAG_EXPECT_SUCCESS "comment:test_expect_success"#define TEST_TAG_EXPECT_MSG_PFX "comment:test_expect_msg="#define TEST_TAG_LOG_LEVEL_PFX "comment:test_log_level=" +#define TEST_TAG_PROG_FLAGS_PFX "comment:test_prog_flags="[...]
Re: [RFC bpf-next 1/5] selftests/bpf: support custom per-test flags and multiple expected messages
Quoted 34 lines · Eduard Zingerman
Quoted 24 lines · Andrii Nakryiko
Quoted 12 lines · Eduard Zingerman
From: Andrii Nakryiko <[email protected]>
Extend
__flagattribute by allowing to specify one of the following:
*BPF_F_STRICT_ALIGNMENT
*BPF_F_ANY_ALIGNMENT
*BPF_F_TEST_RND_HI32
*BPF_F_TEST_STATE_FREQ
*BPF_F_SLEEPABLE
*BPF_F_XDP_HAS_FRAGS
* Some numeric valueQuoted 8 lines
Extend
__msgattribute by allowing to specify multiple exepcted messages. All messages are expected to be present in the verifier log in the order of application.Signed-off-by Andrii Nakryiko <[email protected]>[ Eduard: added commit message, formatting ]Signed-off-by Eduard Zingerman <[email protected]>---
hey Eduard,
Quoted 7 lines
When you get a chance, can you please send this patch separately from the rest of the
test_verifierrework patch set (it probably makes sense to also add #define__flagsin this patch as well, given you are parsing its definition in this patch).This would great help me with my work that uses all this assembly-level test facilities. Thanks!
Quoted 7 lines
Hi Andrii,
Rebase didn't change anything in the patch, I added
__flagsmacro, some some comments, and started the CI job: [1].Feels weird to post it, tbh, because it's 100% your code w/o added value from my side.
you took the effort to prepare it for submission, testing, and integrating into your work, so feels well deserved
Quoted 5 lines
Thanks,
Eduard
apart from test flakiness, looks good, please send a patch "officially"
Quoted 20 lines
Quoted 19 lines
Quoted 17 lines
tools/testing/selftests/bpf/test_loader.c | 69 ++++++++++++++++++++--- tools/testing/selftests/bpf/test_progs.h | 1 + 2 files changed, 61 insertions(+), 9 deletions(-)tools/testing/selftests/bpf/test_loader.c +1
-0@@ -13,12 +13,15 @@#define TEST_TAG_EXPECT_SUCCESS "comment:test_expect_success"#define TEST_TAG_EXPECT_MSG_PFX "comment:test_expect_msg="#define TEST_TAG_LOG_LEVEL_PFX "comment:test_log_level=" +#define TEST_TAG_PROG_FLAGS_PFX "comment:test_prog_flags="[...]
Eduard Zingerman
maintainerRe: [RFC bpf-next 1/5] selftests/bpf: support custom per-test flags and multiple expected messages
On Wed, 2023-03-01 at 09:12 -0800, Andrii Nakryiko wrote:
[...]
Quoted 11 lines
Quoted 5 lines
Rebase didn't change anything in the patch, I added
__flagsmacro, some some comments, and started the CI job: [1].Feels weird to post it, tbh, because it's 100% your code w/o added value from my side.
you took the effort to prepare it for submission, testing, and integrating into your work, so feels well deserved
Quoted 1 line
Quoted 6 lines
Quoted 4 lines
Thanks,
Eduardapart from test flakiness, looks good, please send a patch "officially"
Done.
Local test run does not show any issues.
Quoted 21 lines
Quoted 21 lines
Quoted 20 lines
Quoted 18 lines
tools/testing/selftests/bpf/test_loader.c | 69 ++++++++++++++++++++--- tools/testing/selftests/bpf/test_progs.h | 1 + 2 files changed, 61 insertions(+), 9 deletions(-)tools/testing/selftests/bpf/test_loader.c +1
-0@@ -13,12 +13,15 @@#define TEST_TAG_EXPECT_SUCCESS "comment:test_expect_success"#define TEST_TAG_EXPECT_MSG_PFX "comment:test_expect_msg="#define TEST_TAG_LOG_LEVEL_PFX "comment:test_log_level=" +#define TEST_TAG_PROG_FLAGS_PFX "comment:test_prog_flags="[...]
Eduard Zingerman
maintainerRFC 2/5 selftests/bpf: unprivileged tests for test_loader.c
In reply to Eduard ZingermanExtends test_loader.c:test_loader__run_subtests() by allowing to execute tests in unprivileged mode, similar to test_verifier.c.
Adds the following new attributes controlling test_loader behavior:
__msg_unpriv
__success_unpriv
__failure_unpriv
* If any of these attributes is present the test would be loaded in unprivileged mode.
* If only "privileged" attributes are present the test would be loaded only in privileged mode.
* If both "privileged" and "unprivileged" attributes are present the test would be loaded in both modes.
* If test has to be executed in both modes, __msg(text) is specified and __msg_unpriv is not specified the behavior is the same as if __msg_unpriv(text) is specified.
* For test filtering purposes the name of the program loaded in unprivileged mode is derived from the usual program name by adding `@unpriv' suffix.
Also adds attribute '__description'. This attribute specifies text to be used instead of a program name for display and filtering purposes.
--- tools/testing/selftests/bpf/Makefile | 7 +- tools/testing/selftests/bpf/autoconf_helper.h | 9 + tools/testing/selftests/bpf/progs/bpf_misc.h | 48 +++ tools/testing/selftests/bpf/test_loader.c | 337 ++++++++++++++---- tools/testing/selftests/bpf/test_verifier.c | 25 +- tools/testing/selftests/bpf/unpriv_helpers.c | 26 ++ tools/testing/selftests/bpf/unpriv_helpers.h | 7 + 7 files changed, 359 insertions(+), 100 deletions(-) create mode 100644 tools/testing/selftests/bpf/autoconf_helper.h create mode 100644 tools/testing/selftests/bpf/unpriv_helpers.c create mode 100644 tools/testing/selftests/bpf/unpriv_helpers.h
tools/testing/selftests/bpf/Makefile +4 -3
@@ -218,8 +218,9 @@ TEST_GEN_PROGS_EXTENDED += $(DEFAULT_BPFTOOL) $(TEST_GEN_PROGS) $(TEST_GEN_PROGS_EXTENDED): $(BPFOBJ) -CGROUP_HELPERS := $(OUTPUT)/cgroup_helpers.o TESTING_HELPERS := $(OUTPUT)/testing_helpers.o +CGROUP_HELPERS := $(OUTPUT)/cgroup_helpers.o +UNPRIV_HELPERS := $(OUTPUT)/unpriv_helpers.o TRACE_HELPERS := $(OUTPUT)/trace_helpers.o CAP_HELPERS := $(OUTPUT)/cap_helpers.o
@@ -238,7 +239,7 @@ $(OUTPUT)/test_lirc_mode2_user: $(TESTING_HELPERS) $(OUTPUT)/xdping: $(TESTING_HELPERS) $(OUTPUT)/flow_dissector_load: $(TESTING_HELPERS) $(OUTPUT)/test_maps: $(TESTING_HELPERS) -$(OUTPUT)/test_verifier: $(TESTING_HELPERS) $(CAP_HELPERS) +$(OUTPUT)/test_verifier: $(TESTING_HELPERS) $(CAP_HELPERS) $(UNPRIV_HELPERS) $(OUTPUT)/xsk.o: $(BPFOBJ) BPFTOOL ?= $(DEFAULT_BPFTOOL)
@@ -527,7 +528,7 @@ TRUNNER_BPF_PROGS_DIR := progs TRUNNER_EXTRA_SOURCES := test_progs.c cgroup_helpers.c trace_helpers.c \ network_helpers.c testing_helpers.c \ btf_helpers.c flow_dissector_load.h \ - cap_helpers.c test_loader.c + cap_helpers.c test_loader.c unpriv_helpers.c TRUNNER_EXTRA_FILES := $(OUTPUT)/urandom_read $(OUTPUT)/bpf_testmod.ko \ $(OUTPUT)/liburandom_read.so \ $(OUTPUT)/xdp_synproxy \
tools/testing/selftests/bpf/autoconf_helper.h new file +9 -0
@@ -0,0 +1,9 @@ +// SPDX-License-Identifier: GPL-2.0-only + +#ifdef HAVE_GENHDR +# include "autoconf.h" +#else +# if defined(__i386) || defined(__x86_64) || defined(__s390x__) || defined(__aarch64__) +# define CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS 1 +# endif +#endif
tools/testing/selftests/bpf/progs/bpf_misc.h +48 -0
58 lines not on this page · raw message
tools/testing/selftests/bpf/test_loader.c +263 -74
470 lines not on this page · raw message
tools/testing/selftests/bpf/test_verifier.c +2 -23
@@ -33,13 +33,8 @@#include <bpf/bpf.h> #include <bpf/libbpf.h> -#ifdef HAVE_GENHDR -# include "autoconf.h" -#else -# if defined(__i386) || defined(__x86_64) || defined(__s390x__) || defined(__aarch64__) -# define CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS 1 -# endif -#endif +#include "autoconf_helper.h" +#include "unpriv_helpers.h" #include "cap_helpers.h" #include "bpf_rand.h" #include "bpf_util.h"
@@ -1665,22 +1660,6 @@ static bool is_admin(void)return (caps &ADMIN_CAPS ) ==ADMIN_CAPS ; } -static void get_unpriv_disabled () -{ -char buf[2 ]; -FILE *fd; - - fd =fopen ("/proc/sys/" UNPRIV_SYSCTL ,"r" ); -if (!fd) { -perror ("fopen /proc/sys/" UNPRIV_SYSCTL ); - unpriv_disabled = true; -return ; - } -if (fgets (buf,2 , fd) == buf &&atoi (buf)) - unpriv_disabled = true; -fclose (fd); -} -static bool test_as_unpriv (struct bpf_test *test) {#ifndef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
tools/testing/selftests/bpf/unpriv_helpers.c new file +26 -0
@@ -0,0 +1,26 @@ +// SPDX-License-Identifier: GPL-2.0-only + +#include <stdbool.h> +#include <stdlib.h> +#include <error.h> +#include <stdio.h> + +#include "unpriv_helpers.h" + +bool get_unpriv_disabled (void ) +{ +bool disabled; +char buf[2 ]; +FILE *fd; + + fd =fopen ("/proc/sys/" UNPRIV_SYSCTL ,"r" ); +if (fd) { + disabled = (fgets (buf,2 , fd) == buf &&atoi (buf)); +fclose (fd); + }else { +perror ("fopen /proc/sys/" UNPRIV_SYSCTL ); + disabled = true; + } + +return disabled; +}
tools/testing/selftests/bpf/unpriv_helpers.h new file +7 -0
@@ -0,0 +1,7 @@ +// SPDX-License-Identifier: GPL-2.0-only + +#include <stdbool.h> + +#define UNPRIV_SYSCTL "kernel/unprivileged_bpf_disabled" + +bool get_unpriv_disabled (void );
Signature · 1 line
2.39.0
Eduard Zingerman
maintainerRFC 3/5 selftests/bpf: generate boilerplate code for test_loader-based tests
In reply to Eduard ZingermanAutomatically generate boilerplate code necessary to run tests that use test_loader.c.
Adds a target 'prog_tests/test_loader_auto_wrappers.c' as part of rulesets for 'test_progs' and 'test_progs-no_alu32'. The content of this C file is generated by make and has the following structure:
#include <test_progs.h> #include "some_test_1.skel.h" #include "some_test_2.skel.h" ...
void test_some_test_1(void) { RUN_TESTS(some_test_1); } void test_some_test_2(void) { RUN_TESTS(some_test_2); }
...
Here RUN_TESTS is a macro defined in test_progs.h, it expands to a code that uses test_loader.c:test_loader__run_subtests() function to load tests specified by appropriate skel.h.
In order to get the list of tests included in 'test_loader_auto_wrappers.c' the generation script looks for 'progs/*.c' files that contain a special comment:
/* Use test_loader marker */
--- tools/testing/selftests/bpf/Makefile | 34 +++++++++++++++++++ .../selftests/bpf/prog_tests/.gitignore | 1 + 2 files changed, 35 insertions(+)
tools/testing/selftests/bpf/Makefile +34 -0
@@ -398,6 +398,11 @@ TRUNNER_OUTPUT := $(OUTPUT)$(if $2,/)$2 TRUNNER_BINARY := $1$(if $2,-)$2 TRUNNER_TEST_OBJS := $$(patsubst %.c,$$(TRUNNER_OUTPUT)/%.test.o, \ $$(notdir $$(wildcard $(TRUNNER_TESTS_DIR)/*.c))) +ifneq ($(TRUNNER_TEST_LOADER_AUTO_WRAPPERS),) +TRUNNER_TEST_OBJS := $$(filter-out $$(TRUNNER_OUTPUT)/test_loader_auto_wrappers.test.o, \ + $$(TRUNNER_TEST_OBJS)) +TRUNNER_TEST_OBJS += $$(TRUNNER_OUTPUT)/test_loader_auto_wrappers.test.o +endif TRUNNER_EXTRA_OBJS := $$(patsubst %.c,$$(TRUNNER_OUTPUT)/%.o, \ $$(filter %.c,$(TRUNNER_EXTRA_SOURCES))) TRUNNER_EXTRA_HDRS := $$(filter %.h,$(TRUNNER_EXTRA_SOURCES))
@@ -482,6 +487,32 @@ $(TRUNNER_TESTS_HDR): $(TRUNNER_TESTS_DIR)/*.c ) > $$@) endif +ifneq ($(TRUNNER_TEST_LOADER_AUTO_WRAPPERS),) +ifeq ($($(TRUNNER_TESTS_DIR)-test-loader-auto-wrappers-c),) +$(TRUNNER_TESTS_DIR)-test-loader-auto-wrappers-c := y +$(TRUNNER_TESTS_DIR)/test_loader_auto_wrappers.c: $(TRUNNER_BPF_PROGS_DIR)/*.c + $$(call msg,GEN-TEST,$(TRUNNER_BINARY),$$@) + $$(shell (echo '/* Generated source, do not edit */'; \ + tests=$$$$(grep --null -lF '/* Use test_loader marker */' \ + $$(TRUNNER_BPF_PROGS_DIR)/*.c \ + | xargs -0 -I {} basename {} .c); \ + echo "#include <test_progs.h>"; \ + echo ""; \ + for case in $$$$tests; \ + do \ + echo "#include \"$$$$case.skel.h\""; \ + done; \ + echo ""; \ + for case in $$$$tests; \ + do \ + printf "void %-50s { %-50s }\n" \ + "test_$$$$case(void)" \ + "RUN_TESTS($$$$case);"; \ + done) > $$@) +$(TRUNNER_TESTS_HDR): $(TRUNNER_TESTS_DIR)/test_loader_auto_wrappers.c +endif +endif # TRUNNER_TEST_LOADER_AUTO_WRAPPERS + # compile individual test files # Note: we cd into output directory to ensure embedded BPF object is found $(TRUNNER_TEST_OBJS): $(TRUNNER_OUTPUT)/%.test.o: \
@@ -537,6 +568,7 @@ TRUNNER_EXTRA_FILES := $(OUTPUT)/urandom_read $(OUTPUT)/bpf_testmod.ko \ verify_sig_setup.sh \ $(wildcard progs/btf_dump_test_case_*.c) \ $(wildcard progs/*.bpf.o) +TRUNNER_TEST_LOADER_AUTO_WRAPPERS := t TRUNNER_BPF_BUILD_RULE := CLANG_BPF_BUILD_RULE TRUNNER_BPF_CFLAGS := $(BPF_CFLAGS) $(CLANG_CFLAGS) -DENABLE_ATOMICS_TESTS $(eval $(call DEFINE_TEST_RUNNER,test_progs))
@@ -560,6 +592,7 @@ TRUNNER_EXTRA_SOURCES := test_maps.c TRUNNER_EXTRA_FILES := TRUNNER_BPF_BUILD_RULE := $$(error no BPF objects should be built) TRUNNER_BPF_CFLAGS := +TRUNNER_TEST_LOADER_AUTO_WRAPPERS := $(eval $(call DEFINE_TEST_RUNNER,test_maps)) # Define test_verifier test runner.
@@ -625,6 +658,7 @@ $(OUTPUT)/veristat: $(OUTPUT)/veristat.o EXTRA_CLEAN := $(TEST_CUSTOM_PROGS) $(SCRATCH_DIR) $(HOST_SCRATCH_DIR) \ prog_tests/tests.h map_tests/tests.h verifier/tests.h \ + prog_tests/test_loader_auto_wrappers.c \ feature bpftool \ $(addprefix $(OUTPUT)/,*.o *.skel.h *.lskel.h *.subskel.h \ no_alu32 bpf_gcc bpf_testmod.ko \
tools/testing/selftests/bpf/prog_tests/.gitignore +1 -0
@@ -1,2 +1,3 @@ # SPDX-License-Identifier: GPL-2.0-only tests.h +test_loader_auto_wrappers.c
Signature · 1 line
2.39.0
Re: [RFC bpf-next 3/5] selftests/bpf: generate boilerplate code for test_loader-based tests
Quoted 22 lines · Eduard Zingerman
Automatically generate boilerplate code necessary to run tests that use
test_loader.c.Adds a target '
prog_tests/test_loader_auto_wrappers.c' as part of rulesets for 'test_progs' and 'test_progs-no_alu32'. The content of this C file is generated by make and has the following structure:#include <test_progs.h> #include "some_test_1.skel.h" #include "some_test_2.skel.h" ...void
test_some_test_1(void){RUN_TESTS(some_test_1); } voidtest_some_test_2(void){RUN_TESTS(some_test_2); }
...Here
RUN_TESTSis a macro defined intest_progs.h, it expands to a code that uses test_loader.c:test_loader__run_subtests() function to load tests specified by appropriateskel.h.
Quoted 8 lines
In order to get the list of tests included in '
test_loader_auto_wrappers.c' the generation script looks for 'progs/*.c' files that contain a special comment:/* Use
test_loadermarker */Signed-off-by Eduard Zingerman <[email protected]>---
It feels like this is a bit of an overkill, tbh. There are
$ ls verifier/*.c | wc -l 94
files. We can move each migrated set of tests from verifier/xxx.c to progs/verifier_xxx.c. And then just have just manually maintained prog_tests/verifier.c file where for each converted test we have one #include and one void test_some_test_1(void) { RUN_TESTS(some_test_1); }.
It sometimes would useful to add some extra debugging printfs in such a file, so having it auto generated would be actually an inconvenience. And that on top of further Makefile complication.
For initial conversion we can auto-generate this file, of course. And then for each migrated file adding 2 lines manually doesn't seem like a big deal?
Quoted 4 lines
tools/testing/selftests/bpf/Makefile | 34 +++++++++++++++++++ .../selftests/bpf/prog_tests/.gitignore | 1 + 2 files changed, 35 insertions(+)
[...]
Eduard Zingerman
maintainerRe: [RFC bpf-next 3/5] selftests/bpf: generate boilerplate code for test_loader-based tests
Quoted 43 lines · Andrii Nakryiko
Quoted 22 lines · Eduard Zingerman
Automatically generate boilerplate code necessary to run tests that use
test_loader.c.Adds a target '
prog_tests/test_loader_auto_wrappers.c' as part of rulesets for 'test_progs' and 'test_progs-no_alu32'. The content of this C file is generated by make and has the following structure:#include <test_progs.h> #include "some_test_1.skel.h" #include "some_test_2.skel.h" ...void
test_some_test_1(void){RUN_TESTS(some_test_1); } voidtest_some_test_2(void){RUN_TESTS(some_test_2); }
...Here
RUN_TESTSis a macro defined intest_progs.h, it expands to a code that uses test_loader.c:test_loader__run_subtests() function to load tests specified by appropriateskel.h.Quoted 8 lines
In order to get the list of tests included in '
test_loader_auto_wrappers.c' the generation script looks for 'progs/*.c' files that contain a special comment:/* Use
test_loadermarker */Signed-off-by Eduard Zingerman <[email protected]>---
It feels like this is a bit of an overkill, tbh. There are
$ ls verifier/*.c | wc -l 94files. We can move each migrated set of tests from
verifier/xxx.ctoprogs/verifier_xxx.c. And then just have just manually maintainedprog_tests/verifier.cfile where for each converted test we have one #include and one voidtest_some_test_1(void){RUN_TESTS(some_test_1); }.
Quoted 7 lines
It sometimes would useful to add some extra debugging printfs in such a file, so having it auto generated would be actually an inconvenience. And that on top of further
Makefilecomplication.For initial conversion we can auto-generate this file, of course. And then for each migrated file adding 2 lines manually doesn't seem like a big deal?
Ok, I'll remove the makefile changes.
Quoted 9 lines
Quoted 7 lines
tools/testing/selftests/bpf/Makefile | 34 +++++++++++++++++++ .../selftests/bpf/prog_tests/.gitignore | 1 + 2 files changed, 35 insertions(+)[...]
Eduard Zingerman
maintainerRFC 4/5 selftests/bpf: __imm_insn macro to embed raw insns in inline asm
In reply to Eduard ZingermanA convenience macro to allow the following usage:
#include <linux/filter.h>
...
asm volatile (
...
".8byte %[raw_insn];"
...
:
: __imm_insn(raw_insn, BPF_RAW_INSN(...))
: __clobber_all);
--- tools/testing/selftests/bpf/progs/bpf_misc.h | 1 + 1 file changed, 1 insertion(+)
tools/testing/selftests/bpf/progs/bpf_misc.h +1 -0
@@ -61,6 +61,7 @@#define __clobber_common "r0", "r1", "r2", "r3", "r4", "r5", "memory"#define __imm (name) [name]"i"(name)#define __imm_addr (name) [name]"i"(&name) +#define __imm_insn (name, expr) [name]"i"(*(long *)&(expr))#if defined(__TARGET_ARCH_x86)#define SYSCALL_WRAPPER 1
Signature · 1 line
2.39.0
Re: [RFC bpf-next 4/5] selftests/bpf: __imm_insn macro to embed raw insns in inline asm
Quoted 5 lines · Eduard Zingerman
A convenience macro to allow the following usage:
#include <linux/filter.h>
Quoted 9 lines
...
asm volatile (
...
".8byte %[raw_insn];"
...
:
:__imm_insn(raw_insn, BPF_RAW_INSN(...))
:__clobber_all);
Ah, now I see why it's __imm_insns. Makes sense. Hopefully this won't be used very frequently anyways.
Quoted 20 lines
Signed-off-by Eduard Zingerman <[email protected]>--- tools/testing/selftests/bpf/progs/bpf_misc.h | 1 + 1 file changed, 1 insertion(+)tools/testing/selftests/bpf/progs/bpf_misc.h +1
-1@@ -61,6 +61,7 @@#define __clobber_common "r0", "r1", "r2", "r3", "r4", "r5", "memory"#define __imm (name) [name]"i"(name)#define __imm_addr (name) [name]"i"(&name) +#define __imm_insn (name, expr) [name]"i"(*(long *)&(expr))#if defined(__TARGET_ARCH_x86)#define SYSCALL_WRAPPER 1 --2.39.0
Eduard Zingerman
maintainerRFC 5/5 selftests/bpf: convert jeq_infer_not_null tests to inline assembly
In reply to Eduard ZingermanUse updated test_loader interface to convert progs/jeq_infer_not_null verifier test to inline assembly. Some redundant comments are removed in the process.
Existing test progs/jeq_infer_not_null_fail.c is updated to use "Use test_loader marker" to remove trivial progs_tests/jeq_infer_not_null.c boilerplate code.
--- .../bpf/prog_tests/jeq_infer_not_null.c | 9 - .../selftests/bpf/progs/jeq_infer_not_null.c | 186 ++++++++++++++++++ .../bpf/progs/jeq_infer_not_null_fail.c | 1 + .../bpf/verifier/jeq_infer_not_null.c | 174 ---------------- 4 files changed, 187 insertions(+), 183 deletions(-) delete mode 100644 tools/testing/selftests/bpf/prog_tests/jeq_infer_not_null.c create mode 100644 tools/testing/selftests/bpf/progs/jeq_infer_not_null.c delete mode 100644 tools/testing/selftests/bpf/verifier/jeq_infer_not_null.c
tools/testing/selftests/bpf/prog_tests/jeq_infer_not_null.c deleted +0 -9
@@ -1,9 +0,0 @@ -// SPDX-License-Identifier: GPL-2.0 - -#include <test_progs.h> -#include "jeq_infer_not_null_fail.skel.h" - -void test_jeq_infer_not_null (void ) -{ -RUN_TESTS (jeq_infer_not_null_fail); -}
tools/testing/selftests/bpf/progs/jeq_infer_not_null.c new file +186 -0
186 lines not on this page · raw message
tools/testing/selftests/bpf/progs/jeq_infer_not_null_fail.c +1 -0
@@ -1,4 +1,5 @@// SPDX-License-Identifier: GPL-2.0 +/* Use test_loader marker */ #include "vmlinux.h" #include <bpf/bpf_helpers.h>
tools/testing/selftests/bpf/verifier/jeq_infer_not_null.c deleted +0 -174
174 lines not on this page · raw message
Signature · 1 line
2.39.0
Quoted 33 lines · Eduard Zingerman
As a part of the discussion started in [2] I developed a script [1] that can convert tests specified in
test_verifier.cformat to inline BPF assembly compatible for use withtest_loader.c.For example, test definition like below:
{ "invalid and of negative number", .insns = { BPF_ST_MEM(BPF_DW, BPF_REG_10, -8, 0), BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -8), BPF_LD_MAP_FD(BPF_REG_1, 0), BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 5), BPF_LDX_MEM(BPF_B, BPF_REG_1, BPF_REG_0, 0), BPF_ALU64_IMM(BPF_AND, BPF_REG_1, -4), BPF_ALU64_IMM(BPF_LSH, BPF_REG_1, 2), BPF_ALU64_REG(BPF_ADD, BPF_REG_0, BPF_REG_1), BPF_ST_MEM(BPF_DW, BPF_REG_0, 0, offsetof(struct test_val, foo)), BPF_EXIT_INSN(), }, .fixup_map_hash_48b = { 3 }, .errstr = "R0 max value is outside of the allowed memory range", .result = REJECT, .flags = F_NEEDS_EFFICIENT_UNALIGNED_ACCESS, },Is converted to:
struct
test_val{ ... };
Quoted 7 lines
struct { ...}
map_hash_48bSEC(".maps");
__description("invalid and of negative number")__failure__msg("R0 max value is outside of the allowed memory range")__failure_unpriv__flag(BPF_F_ANY_ALIGNMENT)SEC("socket")
nit: let's make sure that SEC() is the first annotation, makes grepping easier and is consistent with how we use SEC() with BPF programs
Quoted 5 lines
__nakedvoidinvalid_and_of_negative_number(void){ asm volatile (" r1 = 0; \n\
Kumar recently landed similarly formatted inline asm-based test, let's make sure we stick to common style. \n at the end are pretty distracting, IMO (though helpful to debug syntax errors in asm, of course). I'd also move starting " into the same line as asm volatile:
asm volatile (" \
this will make adding/removing asm lines at the beginning simpler (and you already put closing quote on separate line, so that side is taken care of)
Quoted 8 lines
*(u64*)(r10 - 8) = r1; \n\
r2 = r10; \n\
r2 += -8; \n\
r1 = %[map_hash_48b] ll; \n\
call %[bpf_map_lookup_elem]; \n\
if r0 == 0 goto l0_%=; \n\
r1 = *(u8*)(r0 + 0); \n\
r1 &= -4; \n\Quoted 8 lines
r1 <<= 2; \n\
r0 += r1; \n\
r1 = %[test_val_foo_offset]; \n\
*(u64*)(r0 + 0) = r1; \n\
l0_%=: \n\
exit; \n\
" :
: [test_val_foo_offset]"i"(offsetof(struct test_val, foo)),should we use some __imm_const(name, value) macro (or __imm_named, or __imm_alias, or something like that) for cases like this?
Quoted 26 lines
__imm(bpf_map_lookup_elem), __imm_addr(map_hash_48b) : __clobber_all); }The script introduces labels for gotos and calls, immediate values for complex expressions like `offsetof(...)', takes care of map instructions patching, inserts map declarations used in the test, transfers comments from test, test fields and instructions. There are some issues with
BPF_ST_MEMinstruction support as described in [4], thus the script replaces such instructions with pairs of MOV/STX_MEM instructions.This patch-set introduces changes necessary for
test_loader.cand includes migration of a single test fromtest_verifiertotest_progsformat, here are descriptions for individual patches:1. "Support custom per-test flags and multiple expected messages" This patch was shared by Andrii Nakryiko [3], it adds capability to specify flags required by the BPF program.
2. "Unprivileged tests for
test_loader.c"
Extendstest_loader.cby capability to load test programs in unprivileged mode and specify separate test outcomes for privileged and unprivileged modes.
Quoted 4 lines
Note: for a reason I do not understand I had to use different set of capabilities compared to
test_verifier:
-test_verifier:CAP_NET_ADMIN,CAP_PERFMON,CAP_BPF;
-test_loader:CAP_SYS_ADMIN,CAP_PERFMON,CAP_BPF.
CAP_SYS_ADMIN is a superset of PERFMON and BPF, so it should be NET_ADMIN... Setting CAP_SYS_ADMIN makes PERFMON and BPF meaningless, so we should investigate.
Quoted 7 lines
3. "Generate boilerplate code for test_loader-based tests" Extends selftests/bpf
Makefileto automatically generateprog_tests/tests.hentry that usestest_loaderfor progs/*.c BPF programs if special marker is present.4. "
__imm_insnmacro to embed raw insns in inline asm"
nit: __raw_insns perhaps, otherwise confusing given __imm and __imm_addr macros
Quoted 25 lines
This macro can be generated by migration script for instructions that cannot be expressed in inline assembly, e.g. invalid instructions.
5. "convert
jeq_infer_not_nulltests to inline assembly" Shows an example of the test migration.
The test was updated slightly after automatic translation:
- unnecessary comments removed;
- functions renamed;
- some labels renamed.The migration script has some limitations:
- Technical, test features that are not yet supported:
- few instructions likeBPF_ENDIAN;
- macro likeBPF_SK_LOOKUPorBPF_LD_MAP_VALUE;
- program types likeBPF_PROG_TYPE_CGROUP_SOCK_ADDRandBPF_PROG_TYPE_TRACING;- tests that specify fields '
expected_attach_type' or 'insn_processed';
- a few tests expose complex structure that could not be automatically migrated, e.g.: 'atomic_fetch', 'lwt', 'bpf_loop_inline'.
- Tests that use .run field cannot be migrated astest_loader.ctests.
- Tests with generated bodies, e.g. test_verifier.c:bpf_fill_scale() cannot be migrated astest_loader.ctests.
- LLVM related:
-BPF_STinstruction is not supported by LLVM BPF assembly yet
Quoted 7 lines
(I'll take care of it);
- Issues with parsing of some assembly instructions like
"r0 = cmpxchg_64(r10 - 8, r0, r5)"
(I'll take care of it);
- libbpf related:
- libbpf does not support call instructions within a single
function, e.g.:it's more of an issue of not having valid relocation emitted and also corresponding .BTF.ext line/func info (that's my guess)
Quoted 8 lines
0: r1 = 1
1: r2 = 2
2: call 1
3: exit
4: r0 = r1
5: r0 += r2
6: exitso libbpf itself doesn't care, but given we use BTF and send .BTF.ext, I suspect kernel just doesn't like that it can't find func_info information for subprog. So we'll need to split into proper functions or disable .BTF.ext somehow.
Quoted 4 lines
This pattern is common in verifier tests, I see two possible mitigation:
(a) use an additional libbpf flag to allow such code;
I hope not :)
Quoted 2 lines
(b) extend migration script to split such code in several functions. I like (a) more.
see above, this won't be a libbpf flag, rather some code in test runner to strip away .BTF.ext information
Quoted 2 lines
- libbpf does not allow empty programs.
like:
SEC("kprobe")
int prog()
{
return 0;
}?
Quoted 8 lines
All in all the current script stats are as follows:
- 62 out of 93 files from progs/*.c can be converted w/o warnings;
- 55 converted files could be compiled;
- 40 pass testing, 15 fail.By submitting this RFC I seek the following feedback:
- is community interested in such migration?
+1
This is a great work!
Quoted 1 line
- if yes, should I pursue partial or complete tests migration?
I'd start with partial
Quoted 1 line
- in case of partial migration which tests should be prioritized?
those that work out of the box?
Quoted 1 line
- should I offer migrated tests one by one or in big butches?
they come grouped in files, would it be possible to live them in similar groupings?
Quoted 43 lines
Thanks,
Eduard[1] https://github.com/eddyz87/verifier-tests-migrator
[2] https://lore.kernel.org/bpf/CAEf4BzYPsDWdRgx+ND1wiKAB62P=WwoLhr2uWkbVpQfbHqi1oA@mail.gmail.com/
[3] https://lore.kernel.org/bpf/CAEf4BzZH0ZxorCi7nPDbRqSK9f+410RooNwNJGwfw8=0a5i1nw@mail.gmail.com/
[4] https://lore.kernel.org/bpf/[email protected]/Andrii Nakryiko (1):
selftests/bpf: support custom per-test flags and multiple expected
messagesEduard Zingerman (4):
selftests/bpf: unprivileged tests fortest_loader.cselftests/bpf: generate boilerplate code for test_loader-based tests selftests/bpf:__imm_insnmacro to embed raw insns in inline asm selftests/bpf: convertjeq_infer_not_nulltests to inline assemblytools/testing/selftests/bpf/Makefile | 41 +- tools/testing/selftests/bpf/autoconf_helper.h | 9 + .../selftests/bpf/prog_tests/.gitignore | 1 + .../bpf/prog_tests/jeq_infer_not_null.c | 9 - tools/testing/selftests/bpf/progs/bpf_misc.h | 49 +++ .../selftests/bpf/progs/jeq_infer_not_null.c | 186 +++++++++ .../bpf/progs/jeq_infer_not_null_fail.c | 1 + tools/testing/selftests/bpf/test_loader.c | 370 +++++++++++++++--- tools/testing/selftests/bpf/test_progs.h | 1 + tools/testing/selftests/bpf/test_verifier.c | 25 +- tools/testing/selftests/bpf/unpriv_helpers.c | 26 ++ tools/testing/selftests/bpf/unpriv_helpers.h | 7 + .../bpf/verifier/jeq_infer_not_null.c | 174 -------- 13 files changed, 625 insertions(+), 274 deletions(-) create mode 100644 tools/testing/selftests/bpf/autoconf_helper.h delete mode 100644 tools/testing/selftests/bpf/prog_tests/jeq_infer_not_null.c create mode 100644 tools/testing/selftests/bpf/progs/jeq_infer_not_null.c create mode 100644 tools/testing/selftests/bpf/unpriv_helpers.c create mode 100644 tools/testing/selftests/bpf/unpriv_helpers.h delete mode 100644 tools/testing/selftests/bpf/verifier/jeq_infer_not_null.c--
2.39.0
Quoted 11 lines · Andrii Nakryiko
Quoted 6 lines
__nakedvoidinvalid_and_of_negative_number(void){ asm volatile (" r1 = 0; \n\
Kumar recently landed similarly formatted inline asm-based test, let's make sure we stick to common style. \n at the end are pretty distracting, IMO (though helpful to debug syntax errors in asm, of course). I'd also move starting " into the same line as asm volatile:
+1. Pls drop \n.
You don't have \n anyway in migrator's README on github.
Quoted 5 lines
asm volatile (" \
this will make adding/removing asm lines at the beginning simpler (and you already put closing quote on separate line, so that side is taken care of)
+1
Also pls indent the asm code with two tabs the way Kumar did. I think it looks cleaner this way and single tab labels align with 'asm volatile ('.
Quoted 2 lines
Quoted 2 lines
All in all the current script stats are as follows:
- 62 out of 93 files from progs/*.c can be converted w/o warnings;
out of 98 in verifier/*.c ?
Quoted 2 lines
Quoted 2 lines
- 55 converted files could be compiled;
- 40 pass testing, 15 fail.
I would land this 40 now and continue step by step.
Quoted 7 lines
Quoted 3 lines
By submitting this RFC I seek the following feedback:
- is community interested in such migration?+1
This is a great work!
+1
Quoted 9 lines
Quoted 1 line
- if yes, should I pursue partial or complete tests migration?
I'd start with partial
Quoted 1 line
- in case of partial migration which tests should be prioritized?
those that work out of the box?
Quoted 1 line
- should I offer migrated tests one by one or in big butches?
Can you do one patch one file in verifier/*.c that would map to one new file in progs/ ?
Quoted 2 lines
Quoted 2 lines
Having this link in patch series is enough.
The 'migrator' itself doesn't need to be in the kernel tree.
Eduard Zingerman
maintainerQuoted 8 lines · Alexei Starovoitov
Quoted 7 lines · Andrii Nakryiko
Quoted 6 lines
__nakedvoidinvalid_and_of_negative_number(void){ asm volatile (" r1 = 0; \n\
Quoted 7 lines
Quoted 4 lines
Kumar recently landed similarly formatted inline asm-based test, let's make sure we stick to common style. \n at the end are pretty distracting, IMO (though helpful to debug syntax errors in asm, of course). I'd also move starting " into the same line as asm volatile:
+1. Pls drop \n.
You don't have \n anyway in migrator's README on github.
I have --newlines switch there :)
Quoted 7 lines
Quoted 6 lines
asm volatile (" \
this will make adding/removing asm lines at the beginning simpler (and you already put closing quote on separate line, so that side is taken care of)
Quoted 5 lines
+1
Also pls indent the asm code with two tabs the way Kumar did. I think it looks cleaner this way and single tab labels align with 'asm volatile ('.
Will do.
Quoted 5 lines
Quoted 3 lines
Quoted 3 lines
All in all the current script stats are as follows:
- 62 out of 93 files from progs/*.c can be converted w/o warnings;out of 98 in verifier/*.c ?
Sorry, yes, messed up twice in this statement:
- meant verifier/*.c not progs/*.c;
- counted 93 files after migrating one file picked to be an examle. I just double checked and there 94 *.c files in that directory on bpf-next master branch.
Quoted 7 lines
Quoted 3 lines
Quoted 3 lines
- 55 converted files could be compiled;
- 40 pass testing, 15 fail.I would land this 40 now and continue step by step.
Quoted 1 line
Quoted 1 line
Quoted 8 lines
Quoted 6 lines
Quoted 2 lines
By submitting this RFC I seek the following feedback:
- is community interested in such migration?+1
This is a great work!
+1
Thanks.
Quoted 5 lines
Quoted 5 lines
Quoted 2 lines
- if yes, should I pursue partial or complete tests migration?
I'd start with partial
Quoted 8 lines
Quoted 5 lines
Quoted 1 line
- in case of partial migration which tests should be prioritized?
those that work out of the box?
Quoted 1 line
- should I offer migrated tests one by one or in big butches?
Can you do one patch one file in verifier/*.c that would map to one new file in progs/ ?
Will do.
Quoted 6 lines
Quoted 3 lines
Quoted 3 lines
Having this link in patch series is enough.
The 'migrator' itself doesn't need to be in the kernel tree.
Eduard Zingerman
maintainerQuoted 42 lines · Andrii Nakryiko
Quoted 33 lines · Eduard Zingerman
As a part of the discussion started in [2] I developed a script [1] that can convert tests specified in
test_verifier.cformat to inline BPF assembly compatible for use withtest_loader.c.For example, test definition like below:
{ "invalid and of negative number", .insns = { BPF_ST_MEM(BPF_DW, BPF_REG_10, -8, 0), BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -8), BPF_LD_MAP_FD(BPF_REG_1, 0), BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 5), BPF_LDX_MEM(BPF_B, BPF_REG_1, BPF_REG_0, 0), BPF_ALU64_IMM(BPF_AND, BPF_REG_1, -4), BPF_ALU64_IMM(BPF_LSH, BPF_REG_1, 2), BPF_ALU64_REG(BPF_ADD, BPF_REG_0, BPF_REG_1), BPF_ST_MEM(BPF_DW, BPF_REG_0, 0, offsetof(struct test_val, foo)), BPF_EXIT_INSN(), }, .fixup_map_hash_48b = { 3 }, .errstr = "R0 max value is outside of the allowed memory range", .result = REJECT, .flags = F_NEEDS_EFFICIENT_UNALIGNED_ACCESS, },Is converted to:
struct
test_val{ ... };Quoted 7 lines
struct { ...}
map_hash_48bSEC(".maps");
__description("invalid and of negative number")__failure__msg("R0 max value is outside of the allowed memory range")__failure_unpriv__flag(BPF_F_ANY_ALIGNMENT)SEC("socket")
Quoted 3 lines
nit: let's make sure that
SEC()is the first annotation, makes grepping easier and is consistent with how we useSEC()with BPF programs
Ok, will move SEC to the top.
Quoted 10 lines
Quoted 7 lines
__nakedvoidinvalid_and_of_negative_number(void){ asm volatile (" r1 = 0; \n\
Kumar recently landed similarly formatted inline asm-based test, let's make sure we stick to common style. \n at the end are pretty
Tbh, the '\n\' at the end of the line is much less distracting compared to the first variant when each instruction had it's own string ('"r1 = 0;\n"'). And I find it quite useful when compiler points to a bad asm instruction.
But I'll remove it if you insist.
Quoted 2 lines
distracting, IMO (though helpful to debug syntax errors in asm, of course). I'd also move starting " into the same line as asm volatile:
Ok, no problem, will move the '"'.
Quoted 53 lines
asm volatile (" \
this will make adding/removing asm lines at the beginning simpler (and you already put closing quote on separate line, so that side is taken care of)
Quoted 8 lines
*(u64*)(r10 - 8) = r1; \n\ r2 = r10; \n\ r2 += -8; \n\ r1 = %[map_hash_48b] ll; \n\ call %[bpf_map_lookup_elem]; \n\ if r0 == 0 goto l0_%=; \n\ r1 = *(u8*)(r0 + 0); \n\ r1 &= -4; \n\Quoted 8 lines
r1 <<= 2; \n\ r0 += r1; \n\ r1 = %[test_val_foo_offset]; \n\ *(u64*)(r0 + 0) = r1; \n\ l0_%=: \n\ exit; \n\ " : : [test_val_foo_offset]"i"(offsetof(struct test_val, foo)),should we use some
__imm_const(name, value)macro (or__imm_named, or__imm_alias, or something like that) for cases like this?Quoted 26 lines
__imm(bpf_map_lookup_elem), __imm_addr(map_hash_48b) : __clobber_all); }The script introduces labels for gotos and calls, immediate values for complex expressions like `offsetof(...)', takes care of map instructions patching, inserts map declarations used in the test, transfers comments from test, test fields and instructions. There are some issues with
BPF_ST_MEMinstruction support as described in [4], thus the script replaces such instructions with pairs of MOV/STX_MEM instructions.This patch-set introduces changes necessary for
test_loader.cand includes migration of a single test fromtest_verifiertotest_progsformat, here are descriptions for individual patches:1. "Support custom per-test flags and multiple expected messages" This patch was shared by Andrii Nakryiko [3], it adds capability to specify flags required by the BPF program.
2. "Unprivileged tests for
test_loader.c"
Extendstest_loader.cby capability to load test programs in unprivileged mode and specify separate test outcomes for privileged and unprivileged modes.
Quoted 9 lines
Quoted 4 lines
Note: for a reason I do not understand I had to use different set of capabilities compared to
test_verifier:
-test_verifier:CAP_NET_ADMIN,CAP_PERFMON,CAP_BPF;
-test_loader:CAP_SYS_ADMIN,CAP_PERFMON,CAP_BPF.
CAP_SYS_ADMINis a superset of PERFMON and BPF, so it should beNET_ADMIN... SettingCAP_SYS_ADMINmakes PERFMON and BPF meaningless, so we should investigate.
Ok, I'll dig some more and ask for help in case if I would get stuck.
Quoted 9 lines
Quoted 7 lines
3. "Generate boilerplate code for test_loader-based tests" Extends selftests/bpf
Makefileto automatically generateprog_tests/tests.hentry that usestest_loaderfor progs/*.c BPF programs if special marker is present.4. "
__imm_insnmacro to embed raw insns in inline asm"nit:
__raw_insnsperhaps, otherwise confusing given__immand__imm_addrmacros
Sorry, I'm a bit confused, should I rename it to __raw_insn or __imm_insn is fine?
Quoted 46 lines
Quoted 26 lines
This macro can be generated by migration script for instructions that cannot be expressed in inline assembly, e.g. invalid instructions.
5. "convert
jeq_infer_not_nulltests to inline assembly" Shows an example of the test migration.
The test was updated slightly after automatic translation:
- unnecessary comments removed;
- functions renamed;
- some labels renamed.The migration script has some limitations:
- Technical, test features that are not yet supported:
- few instructions likeBPF_ENDIAN;
- macro likeBPF_SK_LOOKUPorBPF_LD_MAP_VALUE;
- program types likeBPF_PROG_TYPE_CGROUP_SOCK_ADDRandBPF_PROG_TYPE_TRACING;- tests that specify fields '
expected_attach_type' or 'insn_processed';
- a few tests expose complex structure that could not be automatically migrated, e.g.: 'atomic_fetch', 'lwt', 'bpf_loop_inline'.
- Tests that use .run field cannot be migrated astest_loader.ctests.
- Tests with generated bodies, e.g. test_verifier.c:bpf_fill_scale() cannot be migrated astest_loader.ctests.
- LLVM related:
-BPF_STinstruction is not supported by LLVM BPF assembly yetQuoted 7 lines
(I'll take care of it); - Issues with parsing of some assembly instructions like "r0 = cmpxchg_64(r10 - 8, r0, r5)" (I'll take care of it); - libbpf related: - libbpf does not support call instructions within a single function, e.g.:it's more of an issue of not having valid relocation emitted and also corresponding .BTF.ext line/func info (that's my guess)
Quoted 8 lines
0: r1 = 1 1: r2 = 2 2: call 1 3: exit 4: r0 = r1 5: r0 += r2 6: exit
Quoted 4 lines
so libbpf itself doesn't care, but given we use BTF and send .BTF.ext, I suspect kernel just doesn't like that it can't find
func_infoinformation for subprog. So we'll need to split into proper functions or disable .BTF.ext somehow.
Actual error that I see comes from libbpf itself and looks as follows:
libbpf: prog 'regalloc_in_callee': no .text section found yet sub-program call exists
libbpf: prog 'regalloc_in_callee': failed to relocate calls: -4005
libbpf: failed to load object 'verifier_regalloc'
It is reported from 'libbpf.c:bpf_object__reloc_code()' when processed instruction is 'insn_is_subprog_call(insn)' and relocation pointer is NULL. I would like to avoid automatically splitting test cases into functions as it means that tests with invalid CFG can't be represented when test_loader.c is used. However if you are completely against adding a flag to libbpf for this case I'll proceed with "function inference" approach.
Quoted 20 lines
Quoted 6 lines
This pattern is common in verifier tests, I see two possible mitigation:
(a) use an additional libbpf flag to allow such code;I hope not :)
Quoted 2 lines
(b) extend migration script to split such code in several functions. I like (a) more.
see above, this won't be a libbpf flag, rather some code in test runner to strip away .BTF.ext information
Quoted 2 lines
- libbpf does not allow empty programs.
like:
Quoted 7 lines
SEC("kprobe")
intprog(){ return 0; }?
Like this:
SEC("kprobe") int prog() {}
No body at all, we have such test. Probably not a big deal if it is not migrated.
Quoted 13 lines
Quoted 8 lines
All in all the current script stats are as follows:
- 62 out of 93 files from progs/*.c can be converted w/o warnings;
- 55 converted files could be compiled;
- 40 pass testing, 15 fail.By submitting this RFC I seek the following feedback:
- is community interested in such migration?+1
This is a great work!
Quoted 7 lines
Quoted 1 line
- if yes, should I pursue partial or complete tests migration?
I'd start with partial
Quoted 1 line
- in case of partial migration which tests should be prioritized?
those that work out of the box?
Ok.
Quoted 5 lines
Quoted 2 lines
- should I offer migrated tests one by one or in big butches?
they come grouped in files, would it be possible to live them in similar groupings?
Yes, file-by-file is how the tool currently works, I will keep the grouping. After this patch-set I can submit 40 patches one-by-one or as a single patch-set. The way I understand you and Alexei a single big patch-set is preferable, is that right?
Quoted 44 lines
Quoted 44 lines
Thanks,
Eduard[1] https://github.com/eddyz87/verifier-tests-migrator
[2] https://lore.kernel.org/bpf/CAEf4BzYPsDWdRgx+ND1wiKAB62P=WwoLhr2uWkbVpQfbHqi1oA@mail.gmail.com/
[3] https://lore.kernel.org/bpf/CAEf4BzZH0ZxorCi7nPDbRqSK9f+410RooNwNJGwfw8=0a5i1nw@mail.gmail.com/
[4] https://lore.kernel.org/bpf/[email protected]/Andrii Nakryiko (1):
selftests/bpf: support custom per-test flags and multiple expected
messagesEduard Zingerman (4):
selftests/bpf: unprivileged tests fortest_loader.cselftests/bpf: generate boilerplate code for test_loader-based tests selftests/bpf:__imm_insnmacro to embed raw insns in inline asm selftests/bpf: convertjeq_infer_not_nulltests to inline assemblytools/testing/selftests/bpf/Makefile | 41 +- tools/testing/selftests/bpf/autoconf_helper.h | 9 + .../selftests/bpf/prog_tests/.gitignore | 1 + .../bpf/prog_tests/jeq_infer_not_null.c | 9 - tools/testing/selftests/bpf/progs/bpf_misc.h | 49 +++ .../selftests/bpf/progs/jeq_infer_not_null.c | 186 +++++++++ .../bpf/progs/jeq_infer_not_null_fail.c | 1 + tools/testing/selftests/bpf/test_loader.c | 370 +++++++++++++++--- tools/testing/selftests/bpf/test_progs.h | 1 + tools/testing/selftests/bpf/test_verifier.c | 25 +- tools/testing/selftests/bpf/unpriv_helpers.c | 26 ++ tools/testing/selftests/bpf/unpriv_helpers.h | 7 + .../bpf/verifier/jeq_infer_not_null.c | 174 -------- 13 files changed, 625 insertions(+), 274 deletions(-) create mode 100644 tools/testing/selftests/bpf/autoconf_helper.h delete mode 100644 tools/testing/selftests/bpf/prog_tests/jeq_infer_not_null.c create mode 100644 tools/testing/selftests/bpf/progs/jeq_infer_not_null.c create mode 100644 tools/testing/selftests/bpf/unpriv_helpers.c create mode 100644 tools/testing/selftests/bpf/unpriv_helpers.h delete mode 100644 tools/testing/selftests/bpf/verifier/jeq_infer_not_null.c--
2.39.0
Quoted 61 lines · Eduard Zingerman
Quoted 42 lines · Andrii Nakryiko
Quoted 33 lines · Eduard Zingerman
As a part of the discussion started in [2] I developed a script [1] that can convert tests specified in
test_verifier.cformat to inline BPF assembly compatible for use withtest_loader.c.For example, test definition like below:
{ "invalid and of negative number", .insns = { BPF_ST_MEM(BPF_DW, BPF_REG_10, -8, 0), BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -8), BPF_LD_MAP_FD(BPF_REG_1, 0), BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 5), BPF_LDX_MEM(BPF_B, BPF_REG_1, BPF_REG_0, 0), BPF_ALU64_IMM(BPF_AND, BPF_REG_1, -4), BPF_ALU64_IMM(BPF_LSH, BPF_REG_1, 2), BPF_ALU64_REG(BPF_ADD, BPF_REG_0, BPF_REG_1), BPF_ST_MEM(BPF_DW, BPF_REG_0, 0, offsetof(struct test_val, foo)), BPF_EXIT_INSN(), }, .fixup_map_hash_48b = { 3 }, .errstr = "R0 max value is outside of the allowed memory range", .result = REJECT, .flags = F_NEEDS_EFFICIENT_UNALIGNED_ACCESS, },Is converted to:
struct
test_val{ ... };Quoted 7 lines
struct { ...}
map_hash_48bSEC(".maps");
__description("invalid and of negative number")__failure__msg("R0 max value is outside of the allowed memory range")__failure_unpriv__flag(BPF_F_ANY_ALIGNMENT)SEC("socket")Quoted 3 lines
nit: let's make sure that
SEC()is the first annotation, makes grepping easier and is consistent with how we useSEC()with BPF programsOk, will move SEC to the top.
Quoted 10 lines
Quoted 7 lines
__nakedvoidinvalid_and_of_negative_number(void){ asm volatile (" r1 = 0; \n\
Kumar recently landed similarly formatted inline asm-based test, let's make sure we stick to common style. \n at the end are pretty
Quoted 6 lines
Tbh, the '\n\' at the end of the line is much less distracting compared to the first variant when each instruction had it's own string ('"r1 = 0;\n"'). And I find it quite useful when compiler points to a bad asm instruction.
But I'll remove it if you insist.
I'm not strongly opposed, but when writing new tests by hand I wouldn't want to add all those \n's. So it's mostly style consistency.
Quoted 77 lines
Quoted 3 lines
distracting, IMO (though helpful to debug syntax errors in asm, of course). I'd also move starting " into the same line as asm volatile:
Ok, no problem, will move the '"'.
Quoted 53 lines
asm volatile (" \
this will make adding/removing asm lines at the beginning simpler (and you already put closing quote on separate line, so that side is taken care of)
Quoted 8 lines
*(u64*)(r10 - 8) = r1; \n\ r2 = r10; \n\ r2 += -8; \n\ r1 = %[map_hash_48b] ll; \n\ call %[bpf_map_lookup_elem]; \n\ if r0 == 0 goto l0_%=; \n\ r1 = *(u8*)(r0 + 0); \n\ r1 &= -4; \n\Quoted 8 lines
r1 <<= 2; \n\ r0 += r1; \n\ r1 = %[test_val_foo_offset]; \n\ *(u64*)(r0 + 0) = r1; \n\ l0_%=: \n\ exit; \n\ " : : [test_val_foo_offset]"i"(offsetof(struct test_val, foo)),should we use some
__imm_const(name, value)macro (or__imm_named, or__imm_alias, or something like that) for cases like this?Quoted 26 lines
__imm(bpf_map_lookup_elem), __imm_addr(map_hash_48b) : __clobber_all); }The script introduces labels for gotos and calls, immediate values for complex expressions like `offsetof(...)', takes care of map instructions patching, inserts map declarations used in the test, transfers comments from test, test fields and instructions. There are some issues with
BPF_ST_MEMinstruction support as described in [4], thus the script replaces such instructions with pairs of MOV/STX_MEM instructions.This patch-set introduces changes necessary for
test_loader.cand includes migration of a single test fromtest_verifiertotest_progsformat, here are descriptions for individual patches:1. "Support custom per-test flags and multiple expected messages" This patch was shared by Andrii Nakryiko [3], it adds capability to specify flags required by the BPF program.
2. "Unprivileged tests for
test_loader.c"
Extendstest_loader.cby capability to load test programs in unprivileged mode and specify separate test outcomes for privileged and unprivileged modes.Quoted 9 lines
Quoted 4 lines
Note: for a reason I do not understand I had to use different set of capabilities compared to
test_verifier:
-test_verifier:CAP_NET_ADMIN,CAP_PERFMON,CAP_BPF;
-test_loader:CAP_SYS_ADMIN,CAP_PERFMON,CAP_BPF.
CAP_SYS_ADMINis a superset of PERFMON and BPF, so it should beNET_ADMIN... SettingCAP_SYS_ADMINmakes PERFMON and BPF meaningless, so we should investigate.Ok, I'll dig some more and ask for help in case if I would get stuck.
Quoted 6 lines
Quoted 6 lines
3. "Generate boilerplate code for test_loader-based tests" Extends selftests/bpf
Makefileto automatically generateprog_tests/tests.hentry that usestest_loaderfor progs/*.c BPF programs if special marker is present.
Quoted 6 lines
Quoted 3 lines
Quoted 1 line
4. "
__imm_insnmacro to embed raw insns in inline asm"nit:
__raw_insnsperhaps, otherwise confusing given__immand__imm_addrmacrosSorry, I'm a bit confused, should I rename it to
__raw_insnor__imm_insnis fine?
nope, keep it as __imm_insn, I haven't seen patch 4 before replying here, I assumed this raw instruction macro will be used inline in the body of asm volatile, but it's done as passing immediate 8-byte value, so it makes sense to name it with __imm prefix
Quoted 58 lines
Quoted 47 lines
Quoted 27 lines
This macro can be generated by migration script for instructions that cannot be expressed in inline assembly, e.g. invalid instructions.
5. "convert
jeq_infer_not_nulltests to inline assembly" Shows an example of the test migration.
The test was updated slightly after automatic translation:
- unnecessary comments removed;
- functions renamed;
- some labels renamed.The migration script has some limitations:
- Technical, test features that are not yet supported:
- few instructions likeBPF_ENDIAN;
- macro likeBPF_SK_LOOKUPorBPF_LD_MAP_VALUE;
- program types likeBPF_PROG_TYPE_CGROUP_SOCK_ADDRandBPF_PROG_TYPE_TRACING;- tests that specify fields '
expected_attach_type' or 'insn_processed';
- a few tests expose complex structure that could not be automatically migrated, e.g.: 'atomic_fetch', 'lwt', 'bpf_loop_inline'.
- Tests that use .run field cannot be migrated astest_loader.ctests.
- Tests with generated bodies, e.g. test_verifier.c:bpf_fill_scale() cannot be migrated astest_loader.ctests.
- LLVM related:
-BPF_STinstruction is not supported by LLVM BPF assembly yetQuoted 7 lines
(I'll take care of it); - Issues with parsing of some assembly instructions like "r0 = cmpxchg_64(r10 - 8, r0, r5)" (I'll take care of it); - libbpf related: - libbpf does not support call instructions within a single function, e.g.:it's more of an issue of not having valid relocation emitted and also corresponding .BTF.ext line/func info (that's my guess)
Quoted 8 lines
0: r1 = 1 1: r2 = 2 2: call 1 3: exit 4: r0 = r1 5: r0 += r2 6: exitQuoted 4 lines
so libbpf itself doesn't care, but given we use BTF and send .BTF.ext, I suspect kernel just doesn't like that it can't find
func_infoinformation for subprog. So we'll need to split into proper functions or disable .BTF.ext somehow.Actual error that I see comes from libbpf itself and looks as follows:
libbpf: prog '
regalloc_in_callee': no .text section found yet sub-program call exists
libbpf: prog 'regalloc_in_callee': failed to relocate calls: -4005
libbpf: failed to load object 'verifier_regalloc'
Quoted 7 lines
It is reported from 'libbpf.c:bpf_object__reloc_code()' when processed instruction is '
insn_is_subprog_call(insn)' and relocation pointer is NULL. I would like to avoid automatically splitting test cases into functions as it means that tests with invalid CFG can't be represented whentest_loader.cis used. However if you are completely against adding a flag to libbpf for this case I'll proceed with "function inference" approach.
Ok, I see. Libbpf has restriction that subprograms have to be in .text, they can't be in SEC("kprobe") and whatnot. The latter is due to inability to distinguish which function in SEC("kprobe") (as one specific example of program section) is entry program and which one is subprog. So any subprogram is just an unannotated function in .text.
Have you tried .section directive to temporarily put subprogram into .text and then go back to the main program? This probably won't be enough, but I'm thinking something like
r1 = 1
r2 = 2
call 1
exit
.section .text, "rx"
r0 = r1
r0 += r2
exit
.section kprobe, "rx"
We might need more directives to properly annotate subprog (function) definition, we can check what compiler generates from C code for subprogs.
Quoted 27 lines
Quoted 27 lines
Quoted 7 lines
This pattern is common in verifier tests, I see two possible mitigation:
(a) use an additional libbpf flag to allow such code;I hope not :)
Quoted 2 lines
(b) extend migration script to split such code in several functions. I like (a) more.
see above, this won't be a libbpf flag, rather some code in test runner to strip away .BTF.ext information
Quoted 2 lines
- libbpf does not allow empty programs.
like:
SEC("kprobe")
intprog(){ return 0; }
Quoted 8 lines
Quoted 1 line
?
Like this:
SEC("kprobe")intprog(){}No body at all, we have such test. Probably not a big deal if it is not migrated.
prog returns int, so at the very least it should have return 0;. And in general, any BPF function should at least have exit BPF instruction, right?
Quoted 27 lines
Quoted 14 lines
Quoted 9 lines
All in all the current script stats are as follows:
- 62 out of 93 files from progs/*.c can be converted w/o warnings;
- 55 converted files could be compiled;
- 40 pass testing, 15 fail.By submitting this RFC I seek the following feedback:
- is community interested in such migration?+1
This is a great work!
Quoted 7 lines
Quoted 1 line
- if yes, should I pursue partial or complete tests migration?
I'd start with partial
Quoted 1 line
- in case of partial migration which tests should be prioritized?
those that work out of the box?
Ok.
Quoted 3 lines
Quoted 2 lines
- should I offer migrated tests one by one or in big butches?
Quoted 7 lines
Quoted 2 lines
they come grouped in files, would it be possible to live them in similar groupings?
Yes, file-by-file is how the tool currently works, I will keep the grouping. After this patch-set I can submit 40 patches one-by-one or as a single patch-set. The way I understand you and Alexei a single big patch-set is preferable, is that right?
yes, one larger patch set would be easier to manager than 40 separate ones in this particular case
Quoted 46 lines
Quoted 45 lines
Quoted 45 lines
Thanks,
Eduard[1] https://github.com/eddyz87/verifier-tests-migrator
[2] https://lore.kernel.org/bpf/CAEf4BzYPsDWdRgx+ND1wiKAB62P=WwoLhr2uWkbVpQfbHqi1oA@mail.gmail.com/
[3] https://lore.kernel.org/bpf/CAEf4BzZH0ZxorCi7nPDbRqSK9f+410RooNwNJGwfw8=0a5i1nw@mail.gmail.com/
[4] https://lore.kernel.org/bpf/[email protected]/Andrii Nakryiko (1):
selftests/bpf: support custom per-test flags and multiple expected
messagesEduard Zingerman (4):
selftests/bpf: unprivileged tests fortest_loader.cselftests/bpf: generate boilerplate code for test_loader-based tests selftests/bpf:__imm_insnmacro to embed raw insns in inline asm selftests/bpf: convertjeq_infer_not_nulltests to inline assemblytools/testing/selftests/bpf/Makefile | 41 +- tools/testing/selftests/bpf/autoconf_helper.h | 9 + .../selftests/bpf/prog_tests/.gitignore | 1 + .../bpf/prog_tests/jeq_infer_not_null.c | 9 - tools/testing/selftests/bpf/progs/bpf_misc.h | 49 +++ .../selftests/bpf/progs/jeq_infer_not_null.c | 186 +++++++++ .../bpf/progs/jeq_infer_not_null_fail.c | 1 + tools/testing/selftests/bpf/test_loader.c | 370 +++++++++++++++--- tools/testing/selftests/bpf/test_progs.h | 1 + tools/testing/selftests/bpf/test_verifier.c | 25 +- tools/testing/selftests/bpf/unpriv_helpers.c | 26 ++ tools/testing/selftests/bpf/unpriv_helpers.h | 7 + .../bpf/verifier/jeq_infer_not_null.c | 174 -------- 13 files changed, 625 insertions(+), 274 deletions(-) create mode 100644 tools/testing/selftests/bpf/autoconf_helper.h delete mode 100644 tools/testing/selftests/bpf/prog_tests/jeq_infer_not_null.c create mode 100644 tools/testing/selftests/bpf/progs/jeq_infer_not_null.c create mode 100644 tools/testing/selftests/bpf/unpriv_helpers.c create mode 100644 tools/testing/selftests/bpf/unpriv_helpers.h delete mode 100644 tools/testing/selftests/bpf/verifier/jeq_infer_not_null.c--
2.39.0