summaryrefslogtreecommitdiffstats
path: root/src/conntrack/copy.c
diff options
context:
space:
mode:
authorFlorian Westphal <fw@strlen.de>2012-11-19 22:39:55 +0100
committerFlorian Westphal <fw@strlen.de>2012-11-28 22:00:06 +0100
commit525aa06da786f66709db4e1ec17bd635b8c5025a (patch)
tree1d546819294eaed4368c14b2deb55a214e6c3fa0 /src/conntrack/copy.c
parent4984b297a3fa6f79ab9162617b9c90aca76d6f70 (diff)
conntrack: fix nfct_clone with certain attribute data types
some attributes are pointers to malloc'd objects. Simply copying the pointer results in use-after free when the original or the clone is destroyed. Fix it by using nfct_copy instead of memcpy and add proper test case for cloned objects: - nfct_cmp of orig and clone should return 1 (equal) - freeing both the original and the clone should neither leak memory nor result in double-frees. the testsuite changes revealed a few more problems: - ct1->timeout == ct2->timeout returned 0, ie. same timeout was considered "not equal" by nfct_cmp - secctx comparision causes "Invalid address" valgrind warnings when pointer is NULL - NFCT_CP_OVERRIDE did not handle helper attribute and erronously freed ct1 secctx memory. While at it, bump qa_test data dummy to 256 (else, valgrind complains about move-depends-on-uninitialized-memory). Lastly, fix compilation of test_api by killing bogus ATTR_CONNLABEL. Signed-off-by: Florian Westphal <fw@strlen.de>
Diffstat (limited to 'src/conntrack/copy.c')
-rw-r--r--src/conntrack/copy.c3
1 files changed, 3 insertions, 0 deletions
diff --git a/src/conntrack/copy.c b/src/conntrack/copy.c
index a6aa9f7..e66c952 100644
--- a/src/conntrack/copy.c
+++ b/src/conntrack/copy.c
@@ -524,5 +524,8 @@ void __copy_fast(struct nf_conntrack *ct1, const struct nf_conntrack *ct2)
{
memcpy(ct1, ct2, sizeof(*ct1));
/* special case: secctx attribute is allocated dinamically. */
+ ct1->secctx = NULL; /* don't free: ct2 uses it */
+ ct1->helper_info = NULL;
copy_attr_secctx(ct1, ct2);
+ copy_attr_help_info(ct1, ct2);
}