traceevent support for kernels with CONFIG_GCC_PLUGIN_STRUCTLEAK=y
Hi!
In kernels built with CONFIG_GCC_PLUGIN_STRUCTLEAK=y definition of macros __user differs from the usual one:
If __CHECKER__ is false:
Quoted 5 lines
# ifdef STRUCTLEAK_PLUGIN # define __user __attribute__((user)) # else # define __user # endif
at linux-5.4.2-arch1/include/linux/compiler_types.h:28.
It seems that this fact leads the syscalls definitions macros linux-5.4.2-arch1/include/linux/syscalls.h
Quoted 3 lines
214:#define
SYSCALL_DEFINE1(name, ...)SYSCALL_DEFINEx(1, _##name, __VA_ARGS__)198:#defineSYSCALL_METADATA(sname, nb, ...)
135:#defineSYSCALL_TRACE_ENTER_EVENT(sname)
to include GGC attributes extensions definitions into the syscalls format metadata that is then delivered to the tracefs/syscalls/format.
This gets syscalls formats to have __attribute__ inside:
Quoted 8 lines
$ cat /sys/kernel/debug/tracing/events/syscalls/sys_enter_io_submit/format name: sys_enter_io_submit ID: 897 format: field:unsigned short common_type; offset:0; size:2; signed:0; ... field:struct iocb __attribute__((user)) * __attribute__((user)) * iocbpp; offset:32; size:8; signed:0; ...
Having __attribute__ leaked into syscalls format, it makes its impossible for the traceevent lib from the kernel/tools/lib to parse such fields, like in the example above.
This in its turn makes it impossible to use tracing for those syscalls:
Quoted 3 lines
$ sudo perf record -e syscalls:sys_enter_io_submit -aR libtraceevent: No such file or directory Error: expected type 4 but read 5
Thus, tracing does not work for some syscalls in Arch Linux kernels. And I suppose for all kernels that built with structleak plugin support.
Reproduce: https://github.com/sitano/traceevent_attribute
My question:
Is it a bug that the traceevent lib does not support parsing __attribute__ in syscalls formats?
or it is a bug of the SYSCALL_DEFINEx macroses / build system that they do allow C attributes to leak?
maybe this is already fixed in the latest kernel? or maybe I am missing something?
Thx,
Ivan
Quoted 26 lines · On Thu, 12 Dec 2019 19:07:46 +0200 John Koepi <[email protected]> wrote:
Quoted 8 lines
$ cat /sys/kernel/debug/tracing/events/syscalls/sys_enter_io_submit/format name: sys_enter_io_submit ID: 897 format: field:unsigned short common_type; offset:0; size:2; signed:0; ... field:struct iocb __attribute__((user)) * __attribute__((user)) * iocbpp; offset:32; size:8; signed:0; ...Having
__attribute__leaked into syscalls format, it makes its impossible for the traceevent lib from thekernel/tools/libto parse such fields, like in the example above.This in its turn makes it impossible to use tracing for those syscalls:
Quoted 3 lines
$ sudo perf record -e syscalls:sys_enter_io_submit -aR libtraceevent: No such file or directory Error: expected type 4 but read 5Thus, tracing does not work for some syscalls in Arch Linux kernels. And I suppose for all kernels that built with structleak plugin support.
Reproduce: https://github.com/sitano/traceevent_attribute
My question:
Quoted 8 lines
Is it a bug that the traceevent lib does not support parsing
__attribute__in syscalls formats?or it is a bug of the
SYSCALL_DEFINExmacroses / build system that they do allow C attributes to leak?maybe this is already fixed in the latest kernel? or maybe I am missing something?
Thanks for the report. This looks like something we can have libtraceveent handle, no need to change the kernel.
Could you file a bug report?
-- Steve
Filed at https://bugzilla.kernel.org/show_bug.cgi?id=205857
-- Ivan
Quoted 48 lines · Steven Rostedt
Quoted 26 lines · On Thu, 12 Dec 2019 19:07:46 +0200 John Koepi <[email protected]> wrote:
Quoted 8 lines
$ cat /sys/kernel/debug/tracing/events/syscalls/sys_enter_io_submit/format name: sys_enter_io_submit ID: 897 format: field:unsigned short common_type; offset:0; size:2; signed:0; ... field:struct iocb __attribute__((user)) * __attribute__((user)) * iocbpp; offset:32; size:8; signed:0; ...Having
__attribute__leaked into syscalls format, it makes its impossible for the traceevent lib from thekernel/tools/libto parse such fields, like in the example above.This in its turn makes it impossible to use tracing for those syscalls:
Quoted 3 lines
$ sudo perf record -e syscalls:sys_enter_io_submit -aR libtraceevent: No such file or directory Error: expected type 4 but read 5Thus, tracing does not work for some syscalls in Arch Linux kernels. And I suppose for all kernels that built with structleak plugin support.
Reproduce: https://github.com/sitano/traceevent_attribute
My question:
Quoted 8 lines
Is it a bug that the traceevent lib does not support parsing
__attribute__in syscalls formats?or it is a bug of the
SYSCALL_DEFINExmacroses / build system that they do allow C attributes to leak?maybe this is already fixed in the latest kernel? or maybe I am missing something?
Thanks for the report. This looks like something we can have libtraceveent handle, no need to change the kernel.
Could you file a bug report?
-- Steve
PATCH 1/1 traceevent: ignore __attribute__ in fields format
To support kernel (e.g. Arch linux) tracing that have events fields with C attributes (__attribute__((xxx))), traceevent must ignore __attribute__ parts when parsing fields types.
An example from Arch linux kernel 4.2:
$ cat /sys/kernel/.../sys_enter_io_submit/format
...
field:struct iocb __attribute__((user)) * ...
^^^This fix adds support for fields types C attributes parsing to event_read_fields function. When it sees __attribute__ ((attribute-list)) expression it skips it.
Fixes: https://bugzilla.kernel.org/show_bug.cgi?id=205857
Base: https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf.git/
--- tools/lib/traceevent/event-parse.c | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+)
tools/lib/traceevent/event-parse.c +22 -0
@@ -1486,6 +1486,28 @@ static int event_read_fields(struct tep_event *event, struct tep_format_field ** (event->flags & TEP_EVENT_FL_ISFTRACE && type == TEP_EVENT_OP && strcmp(token, ".") == 0)) { +/* ignore C attributes: __attribute__((expr)) */ +if (strcmp (token,"__attribute__" ) ==0 ) { +free (token); +for (int i =0 ; i <2 ; i++) { +if (read_expected_item (TEP_EVENT_DELIM ,"(" ) <0 ) { + goto fail_expect; + } + } +for (int brackets =2 ; brackets >0 ;) { +if (read_token (&token) ==TEP_EVENT_NONE ) { +do_warning_event (event,"%s: __attribute__ not full" , __func__); + goto fail_expect; + } +if (strcmp (token,"(" ) ==0 ) + brackets++; +else if (strcmp (token,")" ) ==0 ) + brackets--; +free (token); + } +continue ; + } +if (strcmp (token,"*" ) ==0 ) field->flags |=TEP_FIELD_IS_POINTER ;
Signature · 1 line
2.24.1
Re: [PATCH bpf] traceevent: ignore __attribute__ in fields format
I am not sure this is the right mailing list (I was not sure about @netdev). If it is not - I am sorry, I can resend it. The test for this change is at https://github.com/sitano/traceevent_attribute.
Quoted 61 lines · Ivan Prisyazhnyy
To support kernel (e.g. Arch linux) tracing that have events fields with C attributes (__attribute__((xxx))), traceevent must ignore
__attribute__parts when parsing fields types.An example from Arch linux kernel 4.2:
$ cat /sys/kernel/.../sys_enter_io_submit/format ... field:struct iocb __attribute__((user)) * ... ^^^This fix adds support for fields types C attributes parsing to
event_read_fieldsfunction. When it sees__attribute__((attribute-list)) expression it skips it.Fixes: https://bugzilla.kernel.org/show_bug.cgi?id=205857
Base: https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf.git/Signed-off-by Ivan Prisyazhnyy <[email protected]>--- tools/lib/traceevent/event-parse.c | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+)tools/lib/traceevent/event-parse.c +22
-1@@ -1486,6 +1486,28 @@ static int event_read_fields(struct tep_event *event, struct tep_format_field ** (event->flags & TEP_EVENT_FL_ISFTRACE && type == TEP_EVENT_OP && strcmp(token, ".") == 0)) { +/* ignore C attributes: __attribute__((expr)) */ +if (strcmp (token,"__attribute__" ) ==0 ) { +free (token); +for (int i =0 ; i <2 ; i++) { +if (read_expected_item (TEP_EVENT_DELIM ,"(" ) <0 ) { + goto fail_expect; + } + } +for (int brackets =2 ; brackets >0 ;) { +if (read_token (&token) ==TEP_EVENT_NONE ) { +do_warning_event (event,"%s: __attribute__ not full" , __func__); + goto fail_expect; + } +if (strcmp (token,"(" ) ==0 ) + brackets++; +else if (strcmp (token,")" ) ==0 ) + brackets--; +free (token); + } +continue ; + } +if (strcmp (token,"*" ) ==0 ) field->flags |=TEP_FIELD_IS_POINTER ; --2.24.1
Signature · 1 line
-- Ivan
Re: [PATCH bpf] traceevent: ignore __attribute__ in fields format
Quoted 4 lines · On Wed, 15 Jan 2020 20:47:47 +0200 John Koepi <[email protected]> wrote:
I am not sure this is the right mailing list (I was not sure about @netdev). If it is not - I am sorry, I can resend it. The test for this change is at https://github.com/sitano/traceevent_attribute.
It's fine to send it here. It's in my queue to look at.
Thanks!
-- Steve