NFS: Fix hash table key memory corruption

When the same (as determined by key_equal_func) key gets added to the
GHashTable, old value gets freed and replaced with the new one. This is
fine for hash tables where the key validity is not tightly coupled to
the actual data.

In the nfs_name_snoop_matched hash table the key becomes invalid once
the value gets destroyed (because it shares the data pointed to by fh,
which gets freed once the value is destroyed).

A problematic capture includes packets such that the matching fh gets
added twice to the nfs_name_snoop_matched hash table. Prior to this
change the hash table would end up in a state where the new value is
associated with the old key (which contains pointer to already freed
memory). According to the nfs_name_snoop_matched_equal(), the old key
was equal to the key intended for new value *at the time* of insertion.

This change fixes the bug by using g_hash_table_replace() which does
update the key in case it already exists in the GHashTable.

Bug: 16017
Bug: 16019
Change-Id: Ib3943f1e27e82c05d9abaa1e436554b37a98488e
Reviewed-on: https://code.wireshark.org/review/34360
Reviewed-by: Michael Mann <mmann78@netscape.net>
Petri-Dish: Michael Mann <mmann78@netscape.net>
Tested-by: Petri Dish Buildbot
Reviewed-by: Anders Broman <a.broman58@gmail.com>
This commit is contained in:
Tomasz Moń 2019-08-25 20:28:47 +02:00 committed by Anders Broman
parent 35056a6033
commit efe2926a66
1 changed files with 1 additions and 1 deletions

View File

@ -1211,7 +1211,7 @@ nfs_name_snoop_add_fh(int xid, tvbuff_t *tvb, int fh_offset, int fh_length)
key->fh = nns->fh;
g_hash_table_steal(nfs_name_snoop_unmatched, GINT_TO_POINTER(xid));
g_hash_table_insert(nfs_name_snoop_matched, key, nns);
g_hash_table_replace(nfs_name_snoop_matched, key, nns);
}