[PATCH] Fix buffer overflow in config parser

Hello together,

attached is a small patch to fix a buffer overflow in config.c.
Patch is against git master's HEAD.

I didn't send this one inline as I wanted to
preserve the 1024+ byte long line.

Best regards,
Thomas Jarosch

Thomas Jarosch, 09.04.2009:

Hello together,

attached is a small patch to fix a buffer overflow in config.c.
Patch is against git master's HEAD.

I didn't send this one inline as I wanted to
preserve the 1024+ byte long line.

You could send the patch inline and attach the example config.

At the beginning of the for loop, there is already an overflow guard.

		if (len >= sizeof(value))
			return NULL;

It would probably be better to fix it at this place.

		if (len >= sizeof(value)-1)
			return NULL;

Markus

Markus Heidelberg, 09.04.2009:

Thomas Jarosch, 09.04.2009:

quoted

Hello together,

attached is a small patch to fix a buffer overflow in config.c.
Patch is against git master's HEAD.

I didn't send this one inline as I wanted to
preserve the 1024+ byte long line.

You could send the patch inline and attach the example config.

quoted

diff --git a/config.c b/config.c
index b76fe4c..a9c67e8 100644
--- a/config.c
+++ b/config.c

@@ -72,7 +72,7 @@ static char *parse_value(void)
                        }
                }
                if (space) {
-                       if (len)
+                       if (len && len < sizeof(value)-1)
                                value[len++] = ' ';
                        space = 0;

Eh, or maybe better add a "continue;" here, so that only one char per
loop is read.

quoted

                }

On Thursday, 9. April 2009 01:15:17 Markus Heidelberg wrote:

quoted

quoted

diff --git a/config.c b/config.c
index b76fe4c..a9c67e8 100644
--- a/config.c
+++ b/config.c

@@ -72,7 +72,7 @@ static char *parse_value(void)
                        }
                }
                if (space) {
-                       if (len)
+                       if (len && len < sizeof(value)-1)
                                value[len++] = ' ';
                        space = 0;

Eh, or maybe better add a "continue;" here, so that only one char per
loop is read.

Thanks for the review.

If I understand the intention of the complete code correctly, the idea was
to read in 1+ spaces and put -one- space in the buffer as soon as the first
non-space character is encountered (if not inside quotes).

Adding a "continue" statement would eat up the first non-space character.

I guess it's ok to modify the first size check or keep to problem local and 
check the size before putting the space in the buffer. Guess that's up to
the maintainer which method he prefers.

Cheers,
Thomas

Thomas Jarosch, 09.04.2009:

On Thursday, 9. April 2009 01:15:17 Markus Heidelberg wrote:

quoted

quoted

quoted

diff --git a/config.c b/config.c
index b76fe4c..a9c67e8 100644
--- a/config.c
+++ b/config.c

@@ -72,7 +72,7 @@ static char *parse_value(void)
                        }
                }
                if (space) {
-                       if (len)
+                       if (len && len < sizeof(value)-1)
                                value[len++] = ' ';
                        space = 0;

Eh, or maybe better add a "continue;" here, so that only one char per
loop is read.

Thanks for the review.

If I understand the intention of the complete code correctly, the idea was
to read in 1+ spaces and put -one- space in the buffer as soon as the first
non-space character is encountered (if not inside quotes).

Adding a "continue" statement would eat up the first non-space character.

Yes, you are right, I judged to quickly.

I guess it's ok to modify the first size check or keep to problem local and 
check the size before putting the space in the buffer.

Keeping the problem local would mean to add this check to the end of the
"for" loop before "value[len++] = c;" where the overflow actually
happens.
OTOH this overflow only occurs within a whitespace area, so from this
POV it is local at this place.

I think the cleanest solution would be to decrease the existing check at
the top by 1, even if then the maximum length will be decreased. But
there is no hint in the docs about it anyway.

Guess that's up to
the maintainer which method he prefers.

Take what you think is the best solution and convince him :)

Markus