Skip to contentSkip to conversation

net/ceph/messenger: support IPv6 Link-Local with scope identifier

PATCH In review Max Kellermann 2 participants 2 messages Aug 12 – Aug 19 Ceph common code (libceph)

lore
Share & tools
lore
#0 lore raw

PATCH v4 1/1 net/ceph/messenger: support IPv6 Link-Local with scope identifier

Implement ceph_pton() using inet_pton_with_scope(). This minimal patch preserves the ceph_pton() API and thus requires copying the parameter to a null-terminated string.

Since struct sockaddr_storage is part of the Ceph wire protocol and a non-zero scope id can break the memcmp() checks in the banner handlers, we need to clear all scope ids before using memcmp(). Therefore, this patch introduces the helper function ceph_addr_equal_no_scope().

---
v1->v2:
- check "ipend" before dereferencing it

v2->v3:
- fix indentation
- use the caller's network namespace instead of init_ns
- stop parsing at the port separator

v3->v4:
- fall back to DNS only on EINVAL
- use GFP_KERNEL instead of GFP_NOFS
- ignore the IPv6 scope id when checking the banner peer address

---
 include/linux/ceph/messenger.h |  6 ++++
 net/ceph/messenger.c           | 64 ++++++++++++++++++++++++++++------
 net/ceph/messenger_v1.c        |  3 +-
 net/ceph/messenger_v2.c        |  2 +-
 4 files changed, 62 insertions(+), 13 deletions(-)
include/linux/ceph/messenger.h +6 -0
index 6aa4c6478c9f..ab56a04993d0 100644
@@ -570,6 +570,12 @@ bool ceph_con_v2_opened(struct ceph_connection *con);
 void ceph_con_v2_reset_session(struct ceph_connection *con);
 void ceph_con_v2_reset_protocol(struct ceph_connection *con);
 
+/**
+ * Check whether the two addresses are equal, but ignore different
+ * IPv6 scope identifiers.
+ */
+bool ceph_addr_equal_no_scope(const struct ceph_entity_addr *lhs,
+			      const struct ceph_entity_addr *rhs);
 
 extern const char *ceph_pr_addr(const struct ceph_entity_addr *addr);
 
net/ceph/messenger.c +54 -10

86 lines not on this page · raw message

net/ceph/messenger_v1.c +1 -2
index c9e002d96319..309bca24b82f 100644
@@ -708,8 +708,7 @@ static int process_banner(struct ceph_connection *con)
 	 * end may not yet know their ip address, so if it's 0.0.0.0, give
 	 * them the benefit of the doubt.
 	 */
-	if (memcmp(&con->peer_addr, &con->v1.actual_peer_addr,
-		   sizeof(con->peer_addr)) != 0 &&
+	if (!ceph_addr_equal_no_scope(&con->peer_addr, &con->v1.actual_peer_addr) &&
 	    !(ceph_addr_is_blank(&con->v1.actual_peer_addr) &&
 	      con->v1.actual_peer_addr.nonce == con->peer_addr.nonce)) {
 		pr_warn("wrong peer, want %s/%u, got %s/%u\n",
net/ceph/messenger_v2.c +1 -1
index 05f6eea299fc..dbcf20d623a7 100644
@@ -2504,7 +2504,7 @@ static int process_server_ident(struct ceph_connection *con,
 	     global_id, global_seq, features, required_features, flags, cookie);
 
 	/* is this who we intended to talk to? */
-	if (memcmp(&addr, &con->peer_addr, sizeof(con->peer_addr))) {
+	if (!ceph_addr_equal_no_scope(&addr, &con->peer_addr)) {
 		pr_err("bad peer addr/nonce, want %s/%u, got %s/%u\n",
 		       ceph_pr_addr(&con->peer_addr),
 		       le32_to_cpu(con->peer_addr.nonce),
Signature · 1 line
2.47.3
maintainer
#1 lore raw

Hi Max,

The core issue from v3 is still present: sin6_scope_id is a local ifindex and must not go on the wire. v4 adds ceph_addr_equal_no_scope() for banner comparisons, but peer_addr still carries the scope from inet_pton_with_scope() and gets encoded into the msgr2 HELLO and client-ident. A peer with scope 0 or a different ifindex will still fail identity checks.

v4 also introduces new issues: the monmap update path drops scope when installing decoded addresses, so reconnect to link-local monitors breaks after the first map; ceph_addr_equal_no_type() still memcmps full addresses including scope, breaking monmap lookup; and ceph_addr_clear_scope() does an unaligned write through a cast on a const-qualified pointer.

The fix needs a clean wire/local split: scope lives only in the sockaddr used for kernel_connect(), cleared in everything that touches wire identity.

--
Alex Markuze