7:17 p.m.
From: Jackie Liu <liuyun01(a)kylinos.cn>
Some strange reasons may cause kcore to collect some strange
entries of ikconfig, such as CONFIG_SECU+[some hex data] causes
the 'val' to be NULL, and then crashes when strdup.
Signed-off-by: Jackie Liu <liuyun01(a)kylinos.cn>
---
kernel.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/kernel.c b/kernel.c
index 9871637..aa86f0d 100644
--- a/kernel.c
+++ b/kernel.c
@@ -10244,6 +10244,9 @@ static void add_ikconfig_entry(char *line, struct ikconfig_list
*ent)
sscanf(name, "CONFIG_%s", name);
val = strtok_r(NULL, "", &tokptr);
+ if (!val)
+ return;
+
ent->name = strdup(name);
ent->val = strdup(val);
}
--
2.17.1
9:48 p.m.
New subject: [PATCH] Fixed the segment fault when ikconfig passed nonstandard values
Some strange reasons may cause kcore to collect some strange
entries of ikconfig, such as CONFIG_SECU+[some hex data] causes
Could you tell me the example of such CONFIG_SEC+[some hex data] that
causes the val to have NULL. I tried to reproduce but failed.
the 'val' to be NULL, and then crashes when strdup.
Thanks.
HATAYAMA, Daisuke
11 p.m.
New subject: [PATCH] Fixed the segment fault when ikconfig passed nonstandard values
> Some strange reasons may cause kcore to collect some strange
> entries of ikconfig, such as CONFIG_SECU+[some hex data] causes
Could you tell me the example of such CONFIG_SEC+[some hex data] that
causes the val to have NULL. I tried to reproduce but failed.
I guess you mean CONFIG_SECU+[some hex data] to be **strings that don't
contain the delimitor string '='** and then strtok_r() interprets
it as consisting of a single token, hence val resulting in having NULL.
If so, I think it's better to write so.
OTOH, due to the cause of this issue, is there also possibility
that the broken line contains two or more '=' and broken name
and values are appended in the ikconfig list? It has to be less
important than here because it doesn't result in segfault, but
it might be necessary to deal with it too depending on its actual impact.
Thanks.
HATAYAMA, Daisuke
11:52 p.m.
在 2021/1/4 下午1:00, d.hatayama(a)fujitsu.com 写道:
>> Some strange reasons may cause kcore to collect some strange
>> entries of ikconfig, such as CONFIG_SECU+[some hex data] causes
> Could you tell me the example of such CONFIG_SEC+[some hex data] that
> causes the val to have NULL. I tried to reproduce but failed.
It's hard to say about that, because It's just appeared in the virtual
machine, when I debug
with vmcore (Not a dump.xxx file in my memory).
Ubuntu-kernel/debian/rules.d/2-binary-arch.mk has that:
$(stampdir)/stamp-prepare-tree-%: $(archconfdir)/kylinos_%_defconfig
debian/scripts/fix-filenames
@echo Debug: $@
install -d $(builddir)/build-$*
touch $(builddir)/build-$*/ubuntu-build
[ "$(do_full_source)" != 'true' ] && true || \
rsync -a --exclude debian --exclude debian.master
--exclude $(DEBIAN) * $(builddir)/build-$*
cat $(wordlist 1,3,$^) | sed -e
's/.*CONFIG_VERSION_SIGNATURE.*/CONFIG_VERSION_SIGNATURE="KYLINOS
$(release)-$(revision)-$* $(raw_kernelversion)"/' >
$(builddir)/build-$*/.config
find $(builddir)/build-$* -name "*.ko" | xargs rm -f
$(build_cd) $(kmake) $(build_O) -j1 olddefconfig syncconfig
prepare scripts
touch $@
Every time, '.config' has object file 'fix-filenames' hex data with
config, maybe that is an reason, but
I'm not sure. Maybe It's just any other strange reason cause
/proc/vmcore has CONFIG_SECU+[hex data].
I guess you mean CONFIG_SECU+[some hex data] to be **strings that
don't
contain the delimitor string '='** and then strtok_r() interprets
it as consisting of a single token, hence val resulting in having NULL.
If so, I think it's better to write so.
Yes, It is.
OTOH, due to the cause of this issue, is there also possibility
that the broken line contains two or more '=' and broken name
and values are appended in the ikconfig list? It has to be less
important than here because it doesn't result in segfault, but
it might be necessary to deal with it too depending on its actual impact.
Agree with you.
---
Thanks, BR.
Jackie Liu.
5:21 p.m.
New subject: [PATCH] Fixed the segment fault when ikconfig passed nonstandard values
-----Original Message-----
From: Jackie Liu <liuyun01(a)kylinos.cn>
Some strange reasons may cause kcore to collect some strange
entries of ikconfig, such as CONFIG_SECU+[some hex data] causes
the 'val' to be NULL, and then crashes when strdup.
Signed-off-by: Jackie Liu <liuyun01(a)kylinos.cn>
---
kernel.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/kernel.c b/kernel.c
index 9871637..aa86f0d 100644
--- a/kernel.c
+++ b/kernel.c
@@ -10244,6 +10244,9 @@ static void add_ikconfig_entry(char *line, struct ikconfig_list
*ent)
sscanf(name, "CONFIG_%s", name);
val = strtok_r(NULL, "", &tokptr);
+ if (!val)
+ return;
+
This looks harmless, but kt->ikconfig_ents is incremented unconditionally
in setup_ikconfig() and inconsistent with the following debug message
if (setup_ikconfig(pos)) {
kt->ikconfig_flags |= IKCONFIG_LOADED;
if (CRASHDEBUG(1))
fprintf(fp,
"ikconfig: %d valid configs.\n",
kt->ikconfig_ents);
and it consumes an element of the ikconfig_all array needlessly, so
how about something like this?:
if (!val) {
if (CRASHDEBUG(2))
error(WARNING, "invalid ikconfig entry: %s\n", line);
return FALSE;
}
...
return TRUE;
and in setup_ikconfig():
if (add_ikconfig_entry(ent, &ikconfig_all[kt->ikconfig_ents])
kt->ikconfig_ents++;
Thanks,
Kazu
ent->name = strdup(name);
ent->val = strdup(val);
}
--
2.17.1
--
Crash-utility mailing list
Crash-utility(a)redhat.com
https://www.redhat.com/mailman/listinfo/crash-utility
5:52 p.m.
New subject: [PATCH] Fixed the segment fault when ikconfig passed nonstandard values
Hi Jakie,
I replied only to the list but you look not subscribed?
so sending you this again just in case..
Thanks,
Kazu
-----Original Message-----
-----Original Message-----
> From: Jackie Liu <liuyun01(a)kylinos.cn>
>
> Some strange reasons may cause kcore to collect some strange
> entries of ikconfig, such as CONFIG_SECU+[some hex data] causes
> the 'val' to be NULL, and then crashes when strdup.
>
> Signed-off-by: Jackie Liu <liuyun01(a)kylinos.cn>
> ---
> kernel.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/kernel.c b/kernel.c
> index 9871637..aa86f0d 100644
> --- a/kernel.c
> +++ b/kernel.c
> @@ -10244,6 +10244,9 @@ static void add_ikconfig_entry(char *line, struct
ikconfig_list *ent)
> sscanf(name, "CONFIG_%s", name);
> val = strtok_r(NULL, "", &tokptr);
>
> + if (!val)
> + return;
> +
This looks harmless, but kt->ikconfig_ents is incremented unconditionally
in setup_ikconfig() and inconsistent with the following debug message
if (setup_ikconfig(pos)) {
kt->ikconfig_flags |= IKCONFIG_LOADED;
if (CRASHDEBUG(1))
fprintf(fp,
"ikconfig: %d valid configs.\n",
kt->ikconfig_ents);
and it consumes an element of the ikconfig_all array needlessly, so
how about something like this?:
if (!val) {
if (CRASHDEBUG(2))
error(WARNING, "invalid ikconfig entry: %s\n", line);
return FALSE;
}
...
return TRUE;
and in setup_ikconfig():
if (add_ikconfig_entry(ent, &ikconfig_all[kt->ikconfig_ents])
kt->ikconfig_ents++;
Thanks,
Kazu
> ent->name = strdup(name);
> ent->val = strdup(val);
> }
> --
> 2.17.1
>
>
1413
days inactive
1417
days old
devel@lists.crash-utility.osci.io
5 comments
4 participants
participants (4)
-
d.hatayama＠fujitsu.com -
HAGIO KAZUHITO(萩尾 一仁) -
Jackie Liu -
刘云