aboutsummaryrefslogtreecommitdiffstats
path: root/src
diff options
context:
space:
mode:
authorVsevolod Stakhov <vsevolod@highsecure.ru>2015-04-23 13:08:00 +0100
committerVsevolod Stakhov <vsevolod@highsecure.ru>2015-04-23 13:08:00 +0100
commitd14400e3ae7c9ccfd0d010a95f2d57add99a6868 (patch)
treeb17bda696139a61d2b92e6cbc3871b283e84f495 /src
parent31adf2042911d755510a1992018cd3bf479b9bc8 (diff)
downloadrspamd-d14400e3ae7c9ccfd0d010a95f2d57add99a6868.tar.gz
rspamd-d14400e3ae7c9ccfd0d010a95f2d57add99a6868.zip
Rework phishing detection to reduce false positives rate.
Now we use TLD to match phished URL's. Sometimes, we could miss real positives by not checking strictly for subdomains, however, this change should reduce false positives rate of this rule significantly.
Diffstat (limited to 'src')
-rw-r--r--src/libserver/html.c52
1 files changed, 9 insertions, 43 deletions
diff --git a/src/libserver/html.c b/src/libserver/html.c
index 563ac0825..f978ff1c7 100644
--- a/src/libserver/html.c
+++ b/src/libserver/html.c
@@ -676,7 +676,7 @@ check_phishing (struct rspamd_task *task,
gsize remain,
tag_id_t id)
{
- struct rspamd_url *new;
+ struct rspamd_url *text_url;
gchar *url_str;
const gchar *p, *c;
gchar tagbuf[128];
@@ -731,51 +731,17 @@ check_phishing (struct rspamd_task *task,
if (rspamd_url_find (task->task_pool, url_text, len, NULL, NULL, &url_str,
TRUE, &state) && url_str != NULL) {
- new = rspamd_mempool_alloc0 (task->task_pool, sizeof (struct rspamd_url));
- g_strstrip (url_str);
- rc = rspamd_url_parse (new, url_str, strlen (url_str), task->task_pool);
+ text_url = rspamd_mempool_alloc0 (task->task_pool, sizeof (struct rspamd_url));
+ rc = rspamd_url_parse (text_url, url_str, strlen (url_str), task->task_pool);
if (rc == URI_ERRNO_OK) {
- if (g_ascii_strncasecmp (href_url->host, new->host,
- MAX (href_url->hostlen, new->hostlen)) != 0) {
- /* Special check for urls beginning with 'www' */
- if (new->hostlen > 4 && href_url->hostlen > 4) {
- p = new->host;
- c = NULL;
- if ((p[0] == 'w' || p[0] == 'W') &&
- (p[1] == 'w' || p[1] == 'W') &&
- (p[2] == 'w' || p[2] == 'W') &&
- (p[3] == '.')) {
- p += 4;
- c = href_url->host;
- len = MAX (href_url->hostlen, new->hostlen - 4);
- }
- else {
- p = href_url->host;
- if ((p[0] == 'w' || p[0] == 'W') &&
- (p[1] == 'w' || p[1] == 'W') &&
- (p[2] == 'w' || p[2] == 'W') &&
- (p[3] == '.')) {
- p += 4;
- c = new->host;
- len = MAX (href_url->hostlen - 4, new->hostlen);
- }
- }
- /* Compare parts and check for phished hostname */
- if (c != NULL) {
- if (g_ascii_strncasecmp (p, c, len) != 0) {
- href_url->is_phished = TRUE;
- href_url->phished_url = new;
- }
- }
- else {
- href_url->is_phished = TRUE;
- href_url->phished_url = new;
- }
- }
- else {
+ if (href_url->hostlen != text_url->hostlen || memcmp (href_url->host,
+ text_url->host, href_url->hostlen) != 0) {
+
+ if (href_url->tldlen != text_url->tldlen || memcmp (href_url->tld,
+ text_url->tld, href_url->tldlen) != 0) {
href_url->is_phished = TRUE;
- href_url->phished_url = new;
+ href_url->phished_url = text_url;
}
}
}