Re: [PATCH v2 7/7] tools/testing/selftests/bpf: replace open-coded 16 with TASK_COMM_LEN

[Yafang Shao <laoar.shao@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 6197 bytes --]

On Thu, Feb 9, 2023 at 10:07 AM Mathieu Desnoyers
<mathieu.desnoyers@efficios.com> wrote:
>
> On 2023-02-08 19:54, John Stultz wrote:
> > On Wed, Feb 8, 2023 at 4:11 PM Alexei Starovoitov
> > <alexei.starovoitov@gmail.com> wrote:
> >>
> >> On Wed, Feb 8, 2023 at 2:01 PM John Stultz <jstultz@google.com> wrote:
> >>>
> >>> On Sat, Nov 20, 2021 at 11:27:38AM +0000, Yafang Shao wrote:
> >>>> As the sched:sched_switch tracepoint args are derived from the kernel,
> >>>> we'd better make it same with the kernel. So the macro TASK_COMM_LEN is
> >>>> converted to type enum, then all the BPF programs can get it through BTF.
> >>>>
> >>>> The BPF program which wants to use TASK_COMM_LEN should include the header
> >>>> vmlinux.h. Regarding the test_stacktrace_map and test_tracepoint, as the
> >>>> type defined in linux/bpf.h are also defined in vmlinux.h, so we don't
> >>>> need to include linux/bpf.h again.
> >>>>
> >>>> Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
> >>>> Acked-by: Andrii Nakryiko <andrii@kernel.org>
> >>>> Acked-by: David Hildenbrand <david@redhat.com>
> >>>> Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
> >>>> Cc: Arnaldo Carvalho de Melo <arnaldo.melo@gmail.com>
> >>>> Cc: Andrii Nakryiko <andrii.nakryiko@gmail.com>
> >>>> Cc: Michal Miroslaw <mirq-linux@rere.qmqm.pl>
> >>>> Cc: Peter Zijlstra <peterz@infradead.org>
> >>>> Cc: Steven Rostedt <rostedt@goodmis.org>
> >>>> Cc: Matthew Wilcox <willy@infradead.org>
> >>>> Cc: David Hildenbrand <david@redhat.com>
> >>>> Cc: Al Viro <viro@zeniv.linux.org.uk>
> >>>> Cc: Kees Cook <keescook@chromium.org>
> >>>> Cc: Petr Mladek <pmladek@suse.com>
> >>>> ---
> >>>>   include/linux/sched.h                                   | 9 +++++++--
> >>>>   tools/testing/selftests/bpf/progs/test_stacktrace_map.c | 6 +++---
> >>>>   tools/testing/selftests/bpf/progs/test_tracepoint.c     | 6 +++---
> >>>>   3 files changed, 13 insertions(+), 8 deletions(-)
> >>>
> >>> Hey all,
> >>>    I know this is a little late, but I recently got a report that
> >>> this change was causiing older versions of perfetto to stop
> >>> working.
> >>>
> >>> Apparently newer versions of perfetto has worked around this
> >>> via the following changes:
> >>>    https://android.googlesource.com/platform/external/perfetto/+/c717c93131b1b6e3705a11092a70ac47c78b731d%5E%21/
> >>>    https://android.googlesource.com/platform/external/perfetto/+/160a504ad5c91a227e55f84d3e5d3fe22af7c2bb%5E%21/
> >>>
> >>> But for older versions of perfetto, reverting upstream commit
> >>> 3087c61ed2c4 ("tools/testing/selftests/bpf: replace open-coded 16
> >>> with TASK_COMM_LEN") is necessary to get it back to working.
> >>>
> >>> I haven't dug very far into the details, and obviously this doesn't
> >>> break with the updated perfetto, but from a high level this does
> >>> seem to be a breaking-userland regression.
> >>>
> >>> So I wanted to reach out to see if there was more context for this
> >>> breakage? I don't want to raise a unnecessary stink if this was
> >>> an unfortuante but forced situation.
> >>
> >> Let me understand what you're saying...
> >>
> >> The commit 3087c61ed2c4 did
> >>
> >> -/* Task command name length: */
> >> -#define TASK_COMM_LEN                  16
> >> +/*
> >> + * Define the task command name length as enum, then it can be visible to
> >> + * BPF programs.
> >> + */
> >> +enum {
> >> +       TASK_COMM_LEN = 16,
> >> +};
> >>
> >>
> >> and that caused:
> >>
> >> cat /sys/kernel/debug/tracing/events/task/task_newtask/format
> >>
> >> to print
> >> field:char comm[TASK_COMM_LEN];    offset:12;    size:16;    signed:0;
> >> instead of
> >> field:char comm[16];    offset:12;    size:16;    signed:0;
> >>
> >> so the ftrace parsing android tracing tool had to do:
> >>
> >> -  if (Match(type_and_name.c_str(), R"(char [a-zA-Z_]+\[[0-9]+\])")) {
> >> +  if (Match(type_and_name.c_str(),
> >> +            R"(char [a-zA-Z_][a-zA-Z_0-9]*\[[a-zA-Z_0-9]+\])")) {
> >>
> >> to workaround this change.
> >> Right?
> >
> > I believe so.
> >
> >> And what are you proposing?
> >
> > I'm not proposing anything. I was just wanting to understand more
> > context around this, as it outwardly appears to be a user-breaking
> > change, and that is usually not done, so I figured it was an issue
> > worth raising.
> >
> > If the debug/tracing/*/format output is in the murky not-really-abi
> > space, that's fine, but I wanted to know if this was understood as
> > something that may require userland updates or if this was a
> > unexpected side-effect.
>
> If you are looking at the root cause in the kernel code generating this:
>
> kernel/trace/trace_events.c:f_show()
>
>          /*
>           * Smartly shows the array type(except dynamic array).
>           * Normal:
>           *      field:TYPE VAR
>           * If TYPE := TYPE[LEN], it is shown:
>           *      field:TYPE VAR[LEN]
>           */
>
> where it uses the content of field->type (a string) to format the VAR[LEN] part.
>
> This in turn is the result of the definition of the
> struct trace_event_fields done in:
>
> include/trace/trace_events.h at stage 4, thus with the context of those macros defined:
>
> include/trace/stages/stage4_event_fields.h:
>
> #undef __array
> #define __array(_type, _item, _len) {                                   \
>          .type = #_type"["__stringify(_len)"]", .name = #_item,          \
>          .size = sizeof(_type[_len]), .align = ALIGN_STRUCTFIELD(_type), \
>          .is_signed = is_signed_type(_type), .filter_type = FILTER_OTHER },
>
> I suspect the real culprit here is the use of __stringify(_len), which happens to work
> on macros, but not on enum labels.
>
> One possible solution to make this more robust would be to extend
> struct trace_event_fields with one more field that indicates the length
> of an array as an actual integer, without storing it in its stringified
> form in the type, and do the formatting in f_show where it belongs.
>
> This way everybody can stay happy and no ABI is broken.
>
> Thoughts ?

Many thanks for the detailed analysis. Seems it can work.

Hi John,

Could you pls. try the attached fix ? I have verified it in my test env.


--
Regards
Yafang

[-- Attachment #2: 0001-trace-fix-TASK_COMM_LEN-in-trace-event-format-file.patch --]
[-- Type: application/octet-stream, Size: 5019 bytes --]

From 4e6227e65716001aa7861168c3c510317a3774d4 Mon Sep 17 00:00:00 2001
From: Yafang Shao <laoar.shao@gmail.com>
Date: Thu, 9 Feb 2023 14:13:48 +0800
Subject: [PATCH 1/1] trace: fix TASK_COMM_LEN in trace event format file

Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
---
 include/linux/trace_events.h               |  1 +
 include/trace/stages/stage4_event_fields.h |  3 ++-
 kernel/trace/trace.h                       |  1 +
 kernel/trace/trace_events.c                | 32 ++++++++++++++++++++++--------
 4 files changed, 28 insertions(+), 9 deletions(-)

diff --git a/include/linux/trace_events.h b/include/linux/trace_events.h
index 4342e99..0e37322 100644
--- a/include/linux/trace_events.h
+++ b/include/linux/trace_events.h
@@ -270,6 +270,7 @@ struct trace_event_fields {
 			const int  align;
 			const int  is_signed;
 			const int  filter_type;
+			const int  len;
 		};
 		int (*define_fields)(struct trace_event_call *);
 	};
diff --git a/include/trace/stages/stage4_event_fields.h b/include/trace/stages/stage4_event_fields.h
index affd541..306f39a 100644
--- a/include/trace/stages/stage4_event_fields.h
+++ b/include/trace/stages/stage4_event_fields.h
@@ -26,7 +26,8 @@
 #define __array(_type, _item, _len) {					\
 	.type = #_type"["__stringify(_len)"]", .name = #_item,		\
 	.size = sizeof(_type[_len]), .align = ALIGN_STRUCTFIELD(_type),	\
-	.is_signed = is_signed_type(_type), .filter_type = FILTER_OTHER },
+	.is_signed = is_signed_type(_type), .filter_type = FILTER_OTHER, \
+	.len = _len},
 
 #undef __dynamic_array
 #define __dynamic_array(_type, _item, _len) {				\
diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
index e46a492..19caf15 100644
--- a/kernel/trace/trace.h
+++ b/kernel/trace/trace.h
@@ -1282,6 +1282,7 @@ struct ftrace_event_field {
 	int			offset;
 	int			size;
 	int			is_signed;
+	int			len;
 };
 
 struct prog_entry;
diff --git a/kernel/trace/trace_events.c b/kernel/trace/trace_events.c
index 33e0b4f..8419e48 100644
--- a/kernel/trace/trace_events.c
+++ b/kernel/trace/trace_events.c
@@ -114,7 +114,7 @@ struct ftrace_event_field *
 
 static int __trace_define_field(struct list_head *head, const char *type,
 				const char *name, int offset, int size,
-				int is_signed, int filter_type)
+				int is_signed, int filter_type, int len)
 {
 	struct ftrace_event_field *field;
 
@@ -133,6 +133,7 @@ static int __trace_define_field(struct list_head *head, const char *type,
 	field->offset = offset;
 	field->size = size;
 	field->is_signed = is_signed;
+	field->len = len;
 
 	list_add(&field->link, head);
 
@@ -150,14 +151,28 @@ int trace_define_field(struct trace_event_call *call, const char *type,
 
 	head = trace_get_fields(call);
 	return __trace_define_field(head, type, name, offset, size,
-				    is_signed, filter_type);
+				    is_signed, filter_type, 0);
 }
 EXPORT_SYMBOL_GPL(trace_define_field);
 
+int trace_define_field_ext(struct trace_event_call *call, const char *type,
+		       const char *name, int offset, int size, int is_signed,
+		       int filter_type, int len)
+{
+	struct list_head *head;
+
+	if (WARN_ON(!call->class))
+		return 0;
+
+	head = trace_get_fields(call);
+	return __trace_define_field(head, type, name, offset, size,
+				    is_signed, filter_type, len);
+}
+
 #define __generic_field(type, item, filter_type)			\
 	ret = __trace_define_field(&ftrace_generic_fields, #type,	\
 				   #item, 0, 0, is_signed_type(type),	\
-				   filter_type);			\
+				   filter_type, 0);			\
 	if (ret)							\
 		return ret;
 
@@ -166,7 +181,7 @@ int trace_define_field(struct trace_event_call *call, const char *type,
 				   "common_" #item,			\
 				   offsetof(typeof(ent), item),		\
 				   sizeof(ent.item),			\
-				   is_signed_type(type), FILTER_OTHER);	\
+				   is_signed_type(type), FILTER_OTHER, 0);	\
 	if (ret)							\
 		return ret;
 
@@ -1589,10 +1604,10 @@ static int f_show(struct seq_file *m, void *v)
 			   field->type, field->name, field->offset,
 			   field->size, !!field->is_signed);
 	else
-		seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
+		seq_printf(m, "\tfield:%.*s %s[%d];\toffset:%u;\tsize:%u;\tsigned:%d;\n",
 			   (int)(array_descriptor - field->type),
 			   field->type, field->name,
-			   array_descriptor, field->offset,
+			   field->len, field->offset,
 			   field->size, !!field->is_signed);
 
 	return 0;
@@ -2379,9 +2394,10 @@ static int ftrace_event_release(struct inode *inode, struct file *file)
 			}
 
 			offset = ALIGN(offset, field->align);
-			ret = trace_define_field(call, field->type, field->name,
+			ret = trace_define_field_ext(call, field->type, field->name,
 						 offset, field->size,
-						 field->is_signed, field->filter_type);
+						 field->is_signed, field->filter_type,
+						 field->len);
 			if (WARN_ON_ONCE(ret)) {
 				pr_err("error code is %d\n", ret);
 				break;
-- 
1.8.3.1