410587 - SSL_GetChannelInfo returns SECSuccess on invalid arguments

Status: RESOLVED FIXED

(NSS :: Libraries, defect)

Description

[Wan-Teh Chang]

Attachment: Proposed patch

The SSL_GetChannelInfo function (added in NSS 3.4) returns SECSuccess
if the input argument 'info' or 'len' is invalid.

It should set the error code and return SECFailure.

I rated the severity of this bug at minor because if 'info' is NULL,
we crash, and 'len < sizeof inf.length' can only be true when we
add new fields to SSLChannelInfo, and we can fix this bug before
we add new fields.

Comment 1

[Nelson Bolyard (seldom reads bugmail)]

Comment on attachment 295177 [details] [diff] [review]
Proposed patch

I'm surprised at this bug, because I wrote this code and I am usually 
meticulous about returning proper error codes. 

Regarding the comment:

> 'len < sizeof inf.length' can only be true when we add new fields to 
> SSLChannelInfo, and we can fix this bug before we add new fields.

That's not correct.  Adding new fields to the struct will not affect
the truth of that test.  Notice that that test does *not* ask 
"is len big enough to hold the entire "inf" structure?".  It only asks 
"is len big enough to hold the length field at the beginning of the inf 
structure?".  

This API outputs as much of the inf structure as the caller requests. 
That way, when we grow the structure, we don't create problems for 
callers that still request an older smaller subset.  

The output includes the length of the total inf structure, so that the
caller can call it twice, once to get the desired output length, and 
then again to get the entire structure, after allocating a buffer 
large enough to hold it.  

So, no change is required as we add new fields at the end.  But we 
can fix this failure to report error.

Comment 2

[Wan-Teh Chang]

You're right.  Would be nice to document how the function should
be used.  The caller can't call the function to get the desired
length as you described because we store

    inf.length = PR_MIN(sizeof inf, len);

as opposed to

    inf.length = sizeof inf;

Also, it's hard for the caller to handle the case where
len > sizeof inf, so we may want to treat that as an invalid
input argument.  (The caller would need to check the offset
of a field is < info->length before using it.)

I checked in the patch on the NSS trunk for NSS 3.12.
Checking in sslinfo.c;
/cvsroot/mozilla/security/nss/lib/ssl/sslinfo.c,v  <--  sslinfo.c
new revision: 1.17; previous revision: 1.16
done