2010-07-13 18:45:21

by David Daney

Subject: [PATCH] trace-cmd: Don't SIGSEGV when event field format cannot be parsed.

>From the MIPS kernel we get things like:

print fmt: "page=%p pfn=%lu order=%d migratetype=%d", REC->page, ({ struct page *__pg = (REC->page); int __sec = page_to_section(__pg); (unsigned long)(__pg - __section_mem_map_addr(__nr_to_section(__sec))); }), REC->order, REC->migratetype

This cannot be parsed, leading to a NULL struct event_format* being
passed to pevent_get_common_field_val, which produces a SIGSEGV. It
would be good to get a parsable format from the kernel, but to
remediate the problem for legacy kernels, we can just return an error
indicator in this case. This allows some output from trace-cmd
report, although perhaps with some missing data. But this is better
than crashing.

Signed-off-by: David Daney <[email protected]>
---
parse-events.c | 3 +++
1 files changed, 3 insertions(+), 0 deletions(-)

diff --git a/parse-events.c b/parse-events.c
index 9d5a3fa..16fff12 100644
--- a/parse-events.c
+++ b/parse-events.c
@@ -4481,6 +4481,9 @@ int pevent_get_common_field_val(struct trace_seq *s, struct event_format *event,
{
struct format_field *field;

+ if (!event)
+ return -1;
+
field = pevent_find_common_field(event, name);

return get_field_val(s, field, name, record, val, err);
--
1.6.6.1

2010-07-22 20:08:48

by Steven Rostedt

Subject: Re: [PATCH] trace-cmd: Don't SIGSEGV when event field format cannot be parsed.

On Tue, 2010-07-13 at 11:44 -0700, David Daney wrote:
> >From the MIPS kernel we get things like:
>
> print fmt: "page=%p pfn=%lu order=%d migratetype=%d", REC->page, ({ struct page *__pg = (REC->page); int __sec = page_to_section(__pg); (unsigned long)(__pg - __section_mem_map_addr(__nr_to_section(__sec))); }), REC->order, REC->migratetype
>
> This cannot be parsed, leading to a NULL struct event_format* being
> passed to pevent_get_common_field_val, which produces a SIGSEGV. It
> would be good to get a parsable format from the kernel, but to
> remediate the problem for legacy kernels, we can just return an error
> indicator in this case. This allows some output from trace-cmd
> report, although perhaps with some missing data. But this is better
> than crashing.
>
> Signed-off-by: David Daney <[email protected]>
> ---
> parse-events.c | 3 +++
> 1 files changed, 3 insertions(+), 0 deletions(-)
>
> diff --git a/parse-events.c b/parse-events.c
> index 9d5a3fa..16fff12 100644
> --- a/parse-events.c
> +++ b/parse-events.c
> @@ -4481,6 +4481,9 @@ int pevent_get_common_field_val(struct trace_seq *s, struct event_format *event,
> {

To be consistent, probably all the pevent_get_*field* should test for
null event.

Did you want to resend a patch that does that?

-- Steve

> struct format_field *field;
>
> + if (!event)
> + return -1;
> +
> field = pevent_find_common_field(event, name);
>
> return get_field_val(s, field, name, record, val, err);

2010-07-22 21:05:17

by David Daney

Subject: [PATCH] trace-cmd: Don't SIGSEGV when event field format cannot be parsed (v2).

>From the MIPS kernel we get things like:

print fmt: "page=%p pfn=%lu order=%d migratetype=%d", REC->page, ({ struct page *__pg = (REC->page); int __sec = page_to_section(__pg); (unsigned long)(__pg - __section_mem_map_addr(__nr_to_section(__sec))); }), REC->order, REC->migratetype

This cannot be parsed, leading to a NULL struct event_format* being
passed to pevent_get_common_field_val, which produces a SIGSEGV. It
would be good to get a parsable format from the kernel, but to
remediate the problem for legacy kernels, we can just return an error
indicator in this case. This allows some output from trace-cmd
report, although perhaps with some missing data. But this is better
than crashing.

(v2): Do the check in all pevent_get_*

Signed-off-by: David Daney <[email protected]>
---
parse-events.c | 9 +++++++++
1 files changed, 9 insertions(+), 0 deletions(-)

diff --git a/parse-events.c b/parse-events.c
index 1e854e2..595ba90 100644
--- a/parse-events.c
+++ b/parse-events.c
@@ -4464,6 +4464,9 @@ int pevent_get_field_val(struct trace_seq *s, struct event_format *event,
{
struct format_field *field;

+ if (!event)
+ return -1;
+
field = pevent_find_field(event, name);

return get_field_val(s, field, name, record, val, err);
@@ -4486,6 +4489,9 @@ int pevent_get_common_field_val(struct trace_seq *s, struct event_format *event,
{
struct format_field *field;

+ if (!event)
+ return -1;
+
field = pevent_find_common_field(event, name);

return get_field_val(s, field, name, record, val, err);
@@ -4508,6 +4514,9 @@ int pevent_get_any_field_val(struct trace_seq *s, struct event_format *event,
{
struct format_field *field;

+ if (!event)
+ return -1;
+
field = pevent_find_any_field(event, name);

return get_field_val(s, field, name, record, val, err);
--
1.7.1.1

2010-07-23 15:10:45

by Steven Rostedt

Subject: Re: [PATCH] trace-cmd: Don't SIGSEGV when event field format cannot be parsed (v2).

On Thu, 2010-07-22 at 14:05 -0700, David Daney wrote:
> >From the MIPS kernel we get things like:
>
> print fmt: "page=%p pfn=%lu order=%d migratetype=%d", REC->page, ({ struct page *__pg = (REC->page); int __sec = page_to_section(__pg); (unsigned long)(__pg - __section_mem_map_addr(__nr_to_section(__sec))); }), REC->order, REC->migratetype
>
> This cannot be parsed, leading to a NULL struct event_format* being
> passed to pevent_get_common_field_val, which produces a SIGSEGV. It
> would be good to get a parsable format from the kernel, but to
> remediate the problem for legacy kernels, we can just return an error
> indicator in this case. This allows some output from trace-cmd
> report, although perhaps with some missing data. But this is better
> than crashing.
>
> (v2): Do the check in all pevent_get_*
>
> Signed-off-by: David Daney <[email protected]>

Applied, thanks David!

-- Steve