bpf: net: emit anonymous enum with BPF_TCP_CLOSE value explicitly
PATCH 1/1 bpf: net: emit anonymous enum with BPF_TCP_CLOSE value explicitly
The selftest failed to compile with clang-built bpf-next.
Adding LLVM=1 to your vmlinux and selftest build will use clang.
The error message is:
progs/test_sk_storage_tracing.c:38:18: error: use of undeclared identifier 'BPF_TCP_CLOSE'
if (newstate == BPF_TCP_CLOSE)
^ 1 error generated.
make: *** [Makefile:423: /bpf-next/tools/testing/selftests/bpf/test_sk_storage_tracing.o] Error 1
The reason for the failure is that BPF_TCP_CLOSE, a value of an anonymous enum defined in uapi bpf.h, is not defined in vmlinux.h. gcc does not have this problem. Since vmlinux.h is derived from BTF which is derived from vmlinux dwarf, that means gcc-produced vmlinux dwarf has BPF_TCP_CLOSE while llvm-produced vmlinux dwarf does not have.
BPF_TCP_CLOSE is referenced in net/ipv4/tcp.c as
BUILD_BUG_ON((int)BPF_TCP_CLOSE != (int)TCP_CLOSE);
The following test mimics the above BUILD_BUG_ON, preprocessed with clang compiler, and shows gcc dwarf contains BPF_TCP_CLOSE while llvm dwarf does not.
$ cat t.c
enum {
BPF_TCP_ESTABLISHED = 1,
BPF_TCP_CLOSE = 7,
};
enum {
TCP_ESTABLISHED = 1,
TCP_CLOSE = 7,
}; int test() {
do {
extern void __compiletime_assert_767(void) ;
if ((int)BPF_TCP_CLOSE != (int)TCP_CLOSE) __compiletime_assert_767();
} while (0);
return 0;
}
$ clang t.c -O2 -c -g && llvm-dwarfdump t.o | grep BPF_TCP_CLOSE
$ gcc t.c -O2 -c -g && llvm-dwarfdump t.o | grep BPF_TCP_CLOSE
DW_AT_name ("BPF_TCP_CLOSE")Further checking clang code find clang actually tried to evaluate condition at compile time. If it is definitely true/false, it will perform optimization and the whole if condition will be removed before generating IR/debuginfo.
This patch explicited add an expression like
(void)BPF_TCP_ESTABLISHED
to enable generation of debuginfo for the anonymous enum which also includes BPF_TCP_CLOSE. I put
this explicit type generation in kernel/bpf/core.c to (1) avoid polute net/ipv4/tcp.c and more importantly
(2) provide a central place to add other types (e.g. in bpf/btf uapi header) if they are not referenced in the kernel or generated in vmlinux dwarf.
--- include/linux/btf.h | 1 + kernel/bpf/core.c | 19 +++++++++++++++++++ 2 files changed, 20 insertions(+)
include/linux/btf.h +1 -0
@@ -9,6 +9,7 @@#include <uapi/linux/bpf.h> #define BTF_TYPE_EMIT (type) ((void)(type *)0) +#define BTF_TYPE_EMIT_ENUM (enum_val) ((void)enum_val)struct btf ;struct btf_member ;
kernel/bpf/core.c +19 -0
@@ -2378,3 +2378,22 @@ EXPORT_SYMBOL(bpf_stats_enabled_key);EXPORT_TRACEPOINT_SYMBOL_GPL (xdp_exception);EXPORT_TRACEPOINT_SYMBOL_GPL (xdp_bulk_tx); + +static int __init bpf_emit_btf_type (void ) +{ +/* bpf uapi header bpf.h defines an anonymous enum with values +* BPF_TCP_* used by bpf programs. Currently gcc built vmlinux +* is able to emit this enum in dwarf due to the following +* BUILD_BUG_ON test in net/ipv4/tcp.c: +* BUILD_BUG_ON((int)BPF_TCP_ESTABLISHED != (int)TCP_ESTABLISHED); +* clang built vmlinux does not have this enum in dwarf +* since clang removes the above code before generating IR/debuginfo. +* Let us explicitly emit the type debuginfo to ensure the +* above-mentioned anonymous enum in the vmlinux dwarf and hence BTF +* regardless of which compiler is used. +*/ +BTF_TYPE_EMIT_ENUM (BPF_TCP_ESTABLISHED ); + +return 0 ; +} +late_initcall (bpf_emit_btf_type);
Signature · 1 line
2.24.1
Daniel Borkmann
maintainerOn 3/14/21 4:58 AM, Yonghong Song wrote:
[...]
Quoted 45 lines
This patch explicited add an expression like
(void)BPF_TCP_ESTABLISHED
to enable generation of debuginfo for the anonymous enum which also includesBPF_TCP_CLOSE. I put
this explicit type generation inkernel/bpf/core.cto (1) avoid polutenet/ipv4/tcp.cand more importantly
(2) provide a central place to add other types (e.g. in bpf/btf uapi header) if they are not referenced in the kernel or generated in vmlinux dwarf.Signed-off-by Yonghong Song <[email protected]>--- include/linux/btf.h | 1 + kernel/bpf/core.c | 19 +++++++++++++++++++ 2 files changed, 20 insertions(+)include/linux/btf.h +1
-0@@ -9,6 +9,7 @@#include <uapi/linux/bpf.h> #define BTF_TYPE_EMIT (type) ((void)(type *)0) +#define BTF_TYPE_EMIT_ENUM (enum_val) ((void)enum_val)struct btf ;struct btf_member ;kernel/bpf/core.c +9
-0@@ -2378,3 +2378,22 @@ EXPORT_SYMBOL(bpf_stats_enabled_key);EXPORT_TRACEPOINT_SYMBOL_GPL (xdp_exception);EXPORT_TRACEPOINT_SYMBOL_GPL (xdp_bulk_tx); + +static int __init bpf_emit_btf_type (void ) +{ +/* bpf uapi header bpf.h defines an anonymous enum with values +* BPF_TCP_* used by bpf programs. Currently gcc built vmlinux +* is able to emit this enum in dwarf due to the following +* BUILD_BUG_ON test in net/ipv4/tcp.c: +* BUILD_BUG_ON((int)BPF_TCP_ESTABLISHED != (int)TCP_ESTABLISHED); +* clang built vmlinux does not have this enum in dwarf
kernel/bpf/core.c EXPORT_SYMBOL(bpf_stats_enabled_key); · line 2399
Quoted 10 lines
kernel/bpf/core.c +10
-0+* since clang removes the above code before generating IR/debuginfo. +* Let us explicitly emit the type debuginfo to ensure the +* above-mentioned anonymous enum in the vmlinux dwarf and hence BTF +* regardless of which compiler is used. +*/ +BTF_TYPE_EMIT_ENUM (BPF_TCP_ESTABLISHED ); + +return 0 ; +} +late_initcall (bpf_emit_btf_type);
Does this have to be late_initcall() given this adds minor init call overhead, what if this would be exported as symbol for modules instead?
Thanks,
Daniel
Quoted 50 lines · Daniel Borkmann
On 3/14/21 4:58 AM, Yonghong Song wrote:
[...]Quoted 48 lines
This patch explicited add an expression like (void)BPF_TCP_ESTABLISHED
to enable generation of debuginfo for the anonymous enum which also includesBPF_TCP_CLOSE. I put
this explicit type generation inkernel/bpf/core.cto (1) avoid polutenet/ipv4/tcp.cand more importantly
(2) provide a central place to add other types (e.g. in bpf/btf uapi header) if they are not referenced in the kernel or generated in vmlinux dwarf.Signed-off-by Yonghong Song <[email protected]>---
include/linux/btf.h| 1 +
kernel/bpf/core.c | 19 +++++++++++++++++++2 files changed, 20 insertions(+)include/linux/btf.h +1
-0@@ -9,6 +9,7 @@ #include <uapi/linux/bpf.h> #define BTF_TYPE_EMIT(type) ((void)(type *)0) +#define BTF_TYPE_EMIT_ENUM (enum_val) ((void)enum_val) struct btf; struct btf_member;kernel/bpf/core.c +10
-0@@ -2378,3 +2378,22 @@ EXPORT_SYMBOL(bpf_stats_enabled_key); EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_exception); EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_bulk_tx); + +static int __init bpf_emit_btf_type (void ) +{ + /* bpf uapi header bpf.h defines an anonymous enum with values +* BPF_TCP_* used by bpf programs. Currently gcc built vmlinux +* is able to emit this enum in dwarf due to the following +* BUILD_BUG_ON test in net/ipv4/tcp.c: +* BUILD_BUG_ON((int)BPF_TCP_ESTABLISHED != (int)TCP_ESTABLISHED); +* clang built vmlinux does not have this enum in dwarf +* since clang removes the above code before generating IR/debuginfo.
+ * Let us explicitly emit the type debuginfo to ensure the + * above-mentioned anonymous enum in the vmlinux dwarf and hence BTF + * regardless of which compiler is used.
+ */
Quoted 8 lines
kernel/bpf/core.c EXPORT_SYMBOL(bpf_stats_enabled_key); · line 2399
Quoted 5 lines
+
BTF_TYPE_EMIT_ENUM(BPF_TCP_ESTABLISHED);kernel/bpf/core.c +4
-0+ + return 0; +} +late_initcall (bpf_emit_btf_type);Does this have to be
late_initcall()given this adds minor init call overhead, what if this would be exported as symbol for modules instead?
If issuing types in module, if I understand correctly, it will not be in main vmlinux btf, so programs will not be able to use unless module is loaded. I would prefer such types always available in vmlinux btf.
I am using a separate late_initcall just to cleaner codes. But this BTF_TYPE_EMIT_ENUM can be in any init call.
$ grep _initcall *.c btf.c:fs_initcall(btf_module_init); core.c:pure_initcall(bpf_jit_charge_init); cpumap.c:subsys_initcall(cpu_map_init); devmap.c:subsys_initcall(dev_map_init); inode.c:fs_initcall(bpf_init); map_iter.c:late_initcall(bpf_map_iter_init); net_namespace.c:subsys_initcall(netns_bpf_init); prog_iter.c:late_initcall(bpf_prog_iter_init); stackmap.c:subsys_initcall(stack_map_init); sysfs_btf.c:subsys_initcall(btf_vmlinux_init); task_iter.c:late_initcall(task_iter_init); trampoline.c:late_initcall(init_trampolines); $
I think we can use any above in kernel/bpf directory. This way, we will have 0 runtime overhead as the code will be optimized away. Any preference?
Quoted 3 lines
Thanks,
Daniel
Quoted 55 lines · Yonghong Song
Quoted 52 lines · Daniel Borkmann
On 3/14/21 4:58 AM, Yonghong Song wrote:
[...]Quoted 50 lines
This patch explicited add an expression like (void)BPF_TCP_ESTABLISHED
to enable generation of debuginfo for the anonymous enum which also includesBPF_TCP_CLOSE. I put
this explicit type generation inkernel/bpf/core.cto (1) avoid polutenet/ipv4/tcp.cand more importantly
(2) provide a central place to add other types (e.g. in bpf/btf uapi header) if they are not referenced in the kernel or generated in vmlinux dwarf.Signed-off-by Yonghong Song <[email protected]>---
include/linux/btf.h| 1 +
kernel/bpf/core.c | 19 +++++++++++++++++++2 files changed, 20 insertions(+)include/linux/btf.h +1
-0@@ -9,6 +9,7 @@ #include <uapi/linux/bpf.h> #define BTF_TYPE_EMIT(type) ((void)(type *)0) +#define BTF_TYPE_EMIT_ENUM (enum_val) ((void)enum_val) struct btf; struct btf_member;kernel/bpf/core.c +10
-0@@ -2378,3 +2378,22 @@ EXPORT_SYMBOL(bpf_stats_enabled_key); EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_exception); EXPORT_TRACEPOINT_SYMBOL_GPL(xdp_bulk_tx); + +static int __init bpf_emit_btf_type (void ) +{ + /* bpf uapi header bpf.h defines an anonymous enum with values +* BPF_TCP_* used by bpf programs. Currently gcc built vmlinux +* is able to emit this enum in dwarf due to the following +* BUILD_BUG_ON test in net/ipv4/tcp.c: +* BUILD_BUG_ON((int)BPF_TCP_ESTABLISHED != (int)TCP_ESTABLISHED); +* clang built vmlinux does not have this enum in dwarf +* since clang removes the above code before generating IR/debuginfo.
+ * Let us explicitly emit the type debuginfo to ensure the + * above-mentioned anonymous enum in the vmlinux dwarf and hence BTF
+ * regardless of which compiler is used.
+ */
Quoted 8 lines
Quoted 8 lines
kernel/bpf/core.c EXPORT_SYMBOL(bpf_stats_enabled_key); · line 2399
Quoted 5 lines
+
BTF_TYPE_EMIT_ENUM(BPF_TCP_ESTABLISHED);kernel/bpf/core.c +4
-0+ + return 0; +} +late_initcall (bpf_emit_btf_type);Does this have to be
late_initcall()given this adds minor init call overhead, what if this would be exported as symbol for modules instead?
You mean EXPORT_SYMBOL, right? I think it should work, just one extra
unused empty global function, but a little bit misleading since the function is not intended to use by anybody. I am using init call and wants to remove this function after init...
Quoted 30 lines
If issuing types in module, if I understand correctly, it will not be in main vmlinux btf, so programs will not be able to use unless module is loaded. I would prefer such types always available in vmlinux btf.
I am using a separate
late_initcalljust to cleaner codes. But thisBTF_TYPE_EMIT_ENUMcan be in any init call.$ grep _initcall *.c btf.c:fs_initcall(btf_module_init); core.c:pure_initcall(bpf_jit_charge_init); cpumap.c:subsys_initcall(cpu_map_init); devmap.c:subsys_initcall(dev_map_init); inode.c:fs_initcall(bpf_init); map_iter.c:late_initcall(bpf_map_iter_init); net_namespace.c:subsys_initcall(netns_bpf_init); prog_iter.c:late_initcall(bpf_prog_iter_init); stackmap.c:subsys_initcall(stack_map_init); sysfs_btf.c:subsys_initcall(btf_vmlinux_init); task_iter.c:late_initcall(task_iter_init); trampoline.c:late_initcall(init_trampolines); $I think we can use any above in kernel/bpf directory. This way, we will have 0 runtime overhead as the code will be optimized away. Any preference?
Quoted 3 lines
Thanks,
Daniel
Quoted 59 lines · Yonghong Song
The selftest failed to compile with clang-built bpf-next.
Adding LLVM=1 to your vmlinux and selftest build will use clang.
The error message is:
progs/test_sk_storage_tracing.c:38:18: error: use of undeclared identifier 'BPF_TCP_CLOSE'if (newstate == BPF_TCP_CLOSE) ^1 error generated.
make: *** [Makefile:423: /bpf-next/tools/testing/selftests/bpf/test_sk_storage_tracing.o] Error 1The reason for the failure is that
BPF_TCP_CLOSE, a value of an anonymous enum defined in uapibpf.h, is not defined invmlinux.h. gcc does not have this problem. Sincevmlinux.his derived from BTF which is derived from vmlinux dwarf, that means gcc-produced vmlinux dwarf hasBPF_TCP_CLOSEwhile llvm-produced vmlinux dwarf does not have.
BPF_TCP_CLOSEis referenced innet/ipv4/tcp.cas
BUILD_BUG_ON((int)BPF_TCP_CLOSE != (int)TCP_CLOSE);
The following test mimics the aboveBUILD_BUG_ON, preprocessed with clang compiler, and shows gcc dwarf containsBPF_TCP_CLOSEwhile llvm dwarf does not.$ cat t.c enum { BPF_TCP_ESTABLISHED = 1, BPF_TCP_CLOSE = 7, }; enum { TCP_ESTABLISHED = 1, TCP_CLOSE = 7, };int test() { do { extern void __compiletime_assert_767(void) ; if ((int)BPF_TCP_CLOSE != (int)TCP_CLOSE) __compiletime_assert_767(); } while (0); return 0; } $ clang t.c -O2 -c -g && llvm-dwarfdump t.o | grep BPF_TCP_CLOSE $ gcc t.c -O2 -c -g && llvm-dwarfdump t.o | grep BPF_TCP_CLOSE DW_AT_name ("BPF_TCP_CLOSE")Further checking clang code find clang actually tried to evaluate condition at compile time. If it is definitely true/false, it will perform optimization and the whole if condition will be removed before generating IR/debuginfo.
This patch explicited add an expression like
(void)BPF_TCP_ESTABLISHED
to enable generation of debuginfo for the anonymous enum which also includesBPF_TCP_CLOSE. I put
this explicit type generation inkernel/bpf/core.cto (1) avoid polutenet/ipv4/tcp.cand more importantly
(2) provide a central place to add other types (e.g. in bpf/btf uapi header) if they are not referenced in the kernel or generated in vmlinux dwarf.
Quoted 2 lines
Signed-off-by Yonghong Song <[email protected]>---
But given
$ rg '\sdwarf\s' | wc -l 33 $ rg '\sDWARF\s' | wc -l 151
we should probably stick to using "DWARF", not "dwarf", everywhere.
Quoted 46 lines
include/linux/btf.h | 1 + kernel/bpf/core.c | 19 +++++++++++++++++++ 2 files changed, 20 insertions(+)include/linux/btf.h +1
-0@@ -9,6 +9,7 @@#include <uapi/linux/bpf.h> #define BTF_TYPE_EMIT (type) ((void)(type *)0) +#define BTF_TYPE_EMIT_ENUM (enum_val) ((void)enum_val)struct btf ;struct btf_member ;kernel/bpf/core.c +19
-1@@ -2378,3 +2378,22 @@ EXPORT_SYMBOL(bpf_stats_enabled_key);EXPORT_TRACEPOINT_SYMBOL_GPL (xdp_exception);EXPORT_TRACEPOINT_SYMBOL_GPL (xdp_bulk_tx); + +static int __init bpf_emit_btf_type (void ) +{ +/* bpf uapi header bpf.h defines an anonymous enum with values +* BPF_TCP_* used by bpf programs. Currently gcc built vmlinux +* is able to emit this enum in dwarf due to the following +* BUILD_BUG_ON test in net/ipv4/tcp.c: +* BUILD_BUG_ON((int)BPF_TCP_ESTABLISHED != (int)TCP_ESTABLISHED); +* clang built vmlinux does not have this enum in dwarf +* since clang removes the above code before generating IR/debuginfo. +* Let us explicitly emit the type debuginfo to ensure the +* above-mentioned anonymous enum in the vmlinux dwarf and hence BTF +* regardless of which compiler is used. +*/ +BTF_TYPE_EMIT_ENUM (BPF_TCP_ESTABLISHED ); + +return 0 ; +} +late_initcall (bpf_emit_btf_type); --2.24.1