49477 - CERT_OpenCertDB() leaks filename

Status: RESOLVED FIXED

Description

[John G. Myers]

CERT_OpenCertDB() leaks the filename returned from the filename callback.

Comment 1

[Wan-Teh Chang]

Ian, is there a duplicate for this bug?

Comment 2

[Wan-Teh Chang]

Created attachment 15921 [details] [diff] [review]
Proposed patch.

Comment 3

[Wan-Teh Chang]

Created attachment 15933 [details] [diff] [review]
Revised patch. certDBFilenameCallback needs to strdup the filename string.

Comment 4

[Wan-Teh Chang]

This bug is similar to, but not a duplicate of, bug #39476.

Comments on the revised patch (id=15933):
1. SEC_OpenPermCertDB (called by CERT_OpenCertDB) should
   free certdbname before it returns.
2. certDBFilenameCallback needs to duplicate the filename
   string so that its return value can be freed by
   SEC_OpenPermCertDB.

Please review the patch.  Thanks.

Comment 5

[Ian McGreer]

this looks fine to me.  Thanks Wan-Teh.

Comment 6

[Wan-Teh Chang]

I checked in the fix on the tip.
/cvsroot/mozilla/security/nss/lib/certdb/pcertdb.c, revision 1.4

Comment 7

[John G. Myers]

test_cdb_name_cb() in mailnews/mime/src/mimefilt.cpp needs to be changed to 
return the cert filename in malloced memory.

Same thing for CertDBFilenameCallback() in security/nss/lib/certdb/pcertdb.c

Comment 8

[Wan-Teh Chang]

mailnews/mime/src/mimefilt.cpp is not part of NSS.

My checkin already included the change you suggested
for certDBFilenameCallback() in
security/nss/lib/certdb/pcertdb.c.

Comment 9

[Wan-Teh Chang]

test_kdb_name_cb() in mailnews/mime/src/mimefilt.cpp
also needs the same change.

Comment 10

[Nelson Bolyard (seldom reads bugmail)]

Do any server products that use NSS also need similar changes?

Comment 11

[John G. Myers]

Possibly.  This needs to be mentioned in the release notes.

Comment 12

[Wan-Teh Chang]

Are CERT_OpenCertDB() and SECKEY_OpenKeyDB() typically
only called once in an application session?   If so, I'd
like to propose that we back out the PORT_Free() calls
from these two functions so that we do not break existing
filename callback functions that do not return their results
in malloc'ed memory.

On the other hand, since NSS 3.1 is likely to be the first
3.x release that most products will pick up, it might be
sufficient to mention this change in the release notes.

Comment 13

[John G. Myers]

SECKEY_OpenKeyDB() already had a PORT_Free() call for the failure case, so any 
existing filename callback function for that function which did not return its 
results in malloc'ed memory is already broken.

All of the CERT_OpenCertDB() calls from adminsdk use functions which return 
malloc'ed memory.

Comment 14

[Wan-Teh Chang]

> SECKEY_OpenKeyDB() already had a PORT_Free() call for
> the failure case, so any existing filename callback function
> for that function which did not return its results in
> malloc'ed memory is already broken.

Is it possible that the failure case code has never been exercised?

Comment 15

[John G. Myers]

No, because the server I work on ran into it extremely early.