Skip to content

Commit

Permalink
netlink: fix errant route delete on NEWADDR
Browse files Browse the repository at this point in the history 
NEWADDR was errantly deleting route entities, specifically it happend
to delete IPv6 route entries causing a disconnect of all IPv6 clients.
NEWADDR can be issued on a DHCP lease renew, which does not change any
of the routes.

Code was also cleaned up a bit, and added some helpful logging.
  • Loading branch information
@mbutowski @lws-team
mbutowski authored and lws-team committed Sep 24, 2023
1 parent 0777878 commit 7ef2065
Showing 1 changed file with 137 additions and 61 deletions.
198 changes: 137 additions & 61 deletions lib/roles/netlink/ops-netlink.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -41,7 +41,8 @@
#define RTA_ALIGNTO 4U

//#define lwsl_netlink lwsl_notice
#define lwsl_cx_netlink lwsl_cx_info
#define lwsl_cx_netlink lwsl_cx_info
#define lwsl_cx_netlink_debug lwsl_cx_debug

static void
lws_netlink_coldplug_done_cb(lws_sorted_usec_list_t *sul)
Expand Down Expand Up @@ -69,11 +70,11 @@ rops_handle_POLLIN_netlink(struct lws_context_per_thread *pt, struct lws *wsi,
#endif
;
struct sockaddr_nl nladdr;
lws_route_t robj, *rou, *rmat;
lws_route_t robj, *rou;
struct nlmsghdr *h;
struct msghdr msg;
struct iovec iov;
unsigned int n;
unsigned int n, removed;
char buf[72];

if (!(pollfd->revents & LWS_POLLIN))
Expand Down Expand Up @@ -190,8 +191,50 @@ rops_handle_POLLIN_netlink(struct lws_context_per_thread *pt, struct lws *wsi,
ra_len = (unsigned int)IFA_PAYLOAD(h);

lwsl_cx_netlink(cx, "%s",
h->nlmsg_type == RTM_NEWADDR ?
"NEWADDR" : "DELADDR");
h->nlmsg_type == RTM_NEWADDR ? "NEWADDR" : "DELADDR");

// Parse attributes.
for ( ; RTA_OK(ra, ra_len); ra = RTA_NEXT(ra, ra_len)) {
//lwsl_cx_netlink_debug(cx, "%s: IFA %d\n", __func__, ra->rta_type);
switch (ra->rta_type) {
case IFA_LOCAL:
// Local address
lws_sa46_copy_address(&robj.src, RTA_DATA(ra), rm->rtm_family);
robj.src_len = rm->rtm_src_len;
lws_sa46_write_numeric_address(&robj.src, buf, sizeof(buf));
lwsl_cx_netlink_debug(cx, "IFA_LOCAL: %s/%d", buf, robj.src_len);
break;
case IFA_ADDRESS:
// Prefix address, not local interface.
lws_sa46_copy_address(&robj.dest, RTA_DATA(ra), rm->rtm_family);
robj.dest_len = rm->rtm_dst_len;
lws_sa46_write_numeric_address(&robj.dest, buf, sizeof(buf));
lwsl_cx_netlink_debug(cx, "IFA_ADDRESS: %s/%d", buf, robj.dest_len);
break;
case IFA_FLAGS:
lwsl_cx_netlink_debug(cx, "IFA_FLAGS: 0x%x (not handled)",
*(unsigned int*)RTA_DATA(ra));
break;
case IFA_BROADCAST:
lwsl_cx_netlink_debug(cx, "IFA_BROADCAST (not handled)");
break;
case IFA_ANYCAST:
lwsl_cx_netlink_debug(cx, "IFA_ANYCAST (not handled)");
break;
case IFA_CACHEINFO:
lwsl_cx_netlink_debug(cx, "IFA_CACHEINFO (not handled)");
break;
case IFA_LABEL:
strncpy(buf, RTA_DATA(ra), sizeof(buf));
buf[sizeof(buf)-1] = '\0';
lwsl_cx_netlink_debug(cx, "IFA_LABEL: %s (not used)", buf);
break;
default:
lwsl_cx_netlink_debug(cx, "unknown IFA attr type %d", ra->rta_type);
break;
}
//lwsl_cx_debug(cx, "rta payload length: %ld", RTA_PAYLOAD(ra));
} /* for */

/*
* almost nothing interesting within IFA_* attributes:
Expand Down Expand Up @@ -256,7 +299,10 @@ rops_handle_POLLIN_netlink(struct lws_context_per_thread *pt, struct lws *wsi,
rm->rtm_family);
robj.src_len = rm->rtm_src_len;
lws_sa46_write_numeric_address(&robj.src, buf, sizeof(buf));
lwsl_cx_netlink(cx, "RTA_SRC: %s", buf);
if (ra->rta_type == RTA_SRC)
lwsl_cx_netlink_debug(cx, "RTA_SRC: %s/%d", buf, robj.src_len);
else
lwsl_cx_netlink_debug(cx, "RTA_PREFSRC: %s/%d", buf, robj.src_len);
break;
case RTA_DST:
/* check if is local addr -> considering it as src addr too */
Expand All @@ -265,47 +311,53 @@ rops_handle_POLLIN_netlink(struct lws_context_per_thread *pt, struct lws *wsi,
(rm->rtm_family == AF_INET6 && rm->rtm_dst_len == 128))) {
lws_sa46_copy_address(&robj.src, RTA_DATA(ra),
rm->rtm_family);
lwsl_cx_netlink(cx, "Local addr: RTA_DST -> added to RTA_SRC");
lwsl_cx_netlink_debug(cx, "Local addr: RTA_DST -> added to RTA_SRC");
}

lws_sa46_copy_address(&robj.dest, RTA_DATA(ra),
rm->rtm_family);
rm->rtm_family);
robj.dest_len = rm->rtm_dst_len;
lws_sa46_write_numeric_address(&robj.dest, buf, sizeof(buf));
lwsl_cx_netlink(cx, "RTA_DST: %s", buf);
lwsl_cx_netlink_debug(cx, "RTA_DST: %s/%d", buf, robj.dest_len);
break;
case RTA_GATEWAY:
lws_sa46_copy_address(&robj.gateway,
RTA_DATA(ra),
lws_sa46_copy_address(&robj.gateway, RTA_DATA(ra),
rm->rtm_family);

lws_sa46_write_numeric_address(&robj.gateway, buf, sizeof(buf));
lwsl_cx_netlink_debug(cx, "RTA_GATEWAY: %s", buf);
#if defined(LWS_WITH_SYS_SMD)
gateway_change = 1;
#endif
break;
case RTA_IIF: /* int: input interface index */
case RTA_OIF: /* int: output interface index */
robj.if_idx = *(int *)RTA_DATA(ra);
lwsl_cx_netlink(cx, "ifidx %d", robj.if_idx);
lwsl_cx_netlink_debug(cx, "RTA_IIF/RTA_OIF: %d", robj.if_idx);
break;
case RTA_PRIORITY: /* int: priority of route */
p = RTA_DATA(ra);
robj.priority = p[3] << 24 | p[2] << 16 |
p[1] << 8 | p[0];
lwsl_cx_netlink_debug(cx, "RTA_PRIORITY: %d", robj.priority);
break;
case RTA_CACHEINFO: /* struct rta_cacheinfo */
lwsl_cx_netlink_debug(cx, "RTA_CACHEINFO (not handled)");
break;
#if defined(LWS_HAVE_RTA_PREF)
case RTA_PREF: /* char: RFC4191 v6 router preference */
lwsl_cx_netlink_debug(cx, "RTA_PREF (not handled)");
break;
#endif
case RTA_TABLE: /* int */
lwsl_cx_netlink_debug(cx, "RTA_TABLE (not handled)");
break;

default:
lwsl_cx_info(cx, "unknown attr type %d",
ra->rta_type);
lwsl_cx_netlink_debug(cx, "unknown attr type %d", ra->rta_type);
break;
}
//lwsl_cx_debug(cx, "rta payload length: %ld", RTA_PAYLOAD(ra));
} /* for */

/*
Expand All @@ -327,82 +379,79 @@ rops_handle_POLLIN_netlink(struct lws_context_per_thread *pt, struct lws *wsi,

case RTM_NEWROUTE:

lwsl_cx_netlink(cx, "NEWROUTE rtm_type %d",
rm->rtm_type);

/*
* We don't want any routing debris like /32 or broadcast
* in our routing table... we will collect source addresses
* bound to interfaces via NEWADDR
*/

if (rm->rtm_type != RTN_UNICAST &&
rm->rtm_type != RTN_LOCAL)
if (rm->rtm_type != RTN_UNICAST
&& rm->rtm_type != RTN_LOCAL) {
lwsl_cx_netlink(cx, "NEWROUTE: IGNORED (%s)",
rm->rtm_type == RTN_BROADCAST ? "broadcast" :
rm->rtm_type == RTN_ANYCAST ? "anycast" :
rm->rtm_type == RTN_MULTICAST ? "multicast" :
rm->rtm_type == RTN_UNREACHABLE ? "unreachable" :
rm->rtm_type == RTN_NAT ? "nat" :
rm->rtm_type == RTN_UNSPEC ? "unspecified" :
"other");
break;
}

if (rm->rtm_flags & RTM_F_CLONED)
if (rm->rtm_flags & RTM_F_CLONED) {
lwsl_cx_netlink(cx, "NEWROUTE: IGNORED (cloned)");
break;
}

goto ana;
lwsl_cx_netlink(cx, "NEWROUTE: ACCEPTED (if_idx %d)",
robj.if_idx);

case RTM_DELADDR:
lwsl_cx_notice(cx, "DELADDR");
#if defined(_DEBUG)
_lws_routing_entry_dump(cx, &robj);
#endif
lws_pt_lock(pt, __func__);
_lws_route_remove(pt, &robj, LRR_MATCH_SRC | LRR_IGNORE_PRI);
_lws_route_pt_close_unroutable(pt);
lws_pt_unlock(pt);
break;

case RTM_NEWADDR:

lwsl_cx_netlink(cx, "NEWADDR");
ana:

/*
* Is robj a dupe in the routing table already?
*
* match on pri ignore == set pri and skip
* no match == add
*/

lws_pt_lock(pt, __func__);

/* returns zero on match already in table */
rmat = _lws_route_remove(pt, &robj, h->nlmsg_type == RTM_NEWROUTE ?
LRR_MATCH_DST : LRR_MATCH_SRC | LRR_IGNORE_PRI);
lws_pt_unlock(pt);

if (rmat) {
rmat->priority = robj.priority;
break;
}

* 1. Allocate new route for linked-list.
* (robj is on stack, do NOT use)
*/
rou = lws_malloc(sizeof(*rou), __func__);
if (!rou) {
lwsl_cx_err(cx, "oom");
return LWS_HPI_RET_HANDLED;
}

*rou = robj;

// 2. Remove duplicates and add route (both under a lock).
lws_pt_lock(pt, __func__);

/*
* We lock the pt before getting the uidx, so it
* cannot race
* Is robj a dupe in the routing table already?
*
* match on pri ignore == set pri and skip
* no match == add
*
* returns zero ALWAYS
*
* We could be adding a route to the same destination with
* a higher or lower priority from a different source, so why
* all existing routes? Only remove if its the same source and
* destination, effectively a change in priority.
*/
_lws_route_remove(pt, &robj,
LRR_MATCH_DST | LRR_MATCH_SRC | LRR_IGNORE_PRI);

/* add route to linked-list */
rou->uidx = _lws_route_get_uidx(cx);
lws_dll2_add_tail(&rou->list, &cx->routing_table);
lwsl_cx_info(cx, "route list size %u",
cx->routing_table.count);
lws_pt_unlock(pt);

_lws_route_pt_close_unroutable(pt);
lwsl_cx_netlink_debug(cx, "route list size %u",
cx->routing_table.count);

lws_pt_unlock(pt);
/*
* 3. Close anyything we cant reach anymore due to the removal.
* (don't need to or want to do this under lock)
*/
_lws_route_pt_close_unroutable(pt);

inform:
#if defined(_DEBUG)
Expand All @@ -416,12 +465,39 @@ rops_handle_POLLIN_netlink(struct lws_context_per_thread *pt, struct lws *wsi,
*/
(void)lws_smd_msg_printf(cx, LWSSMDCL_NETWORK,
"{\"rt\":\"%s\"}\n",
(h->nlmsg_type == RTM_DELROUTE) ?
"del" : "add");
(h->nlmsg_type == RTM_NEWROUTE) ?
"add" : "del");
#endif

break;

case RTM_DELADDR:
lwsl_cx_notice(cx, "DELADDR");
#if defined(_DEBUG)
_lws_routing_entry_dump(cx, &robj);
#endif
lws_pt_lock(pt, __func__);
removed = cx->routing_table.count;
_lws_route_remove(pt, &robj, LRR_MATCH_SRC | LRR_IGNORE_PRI);
removed -= cx->routing_table.count;
lws_pt_unlock(pt);
_lws_route_pt_close_unroutable(pt);
if (removed > 0)
goto inform;
break;

case RTM_NEWADDR:
lwsl_cx_netlink(cx, "NEWADDR (nothing to do)");
#if defined(_DEBUG)
_lws_routing_entry_dump(cx, &robj);
#endif
/*
* An address alone does not provide new routes.
* NEWADDR will happen when the DHCP lease is being
* renewed, and will likely not change any routes.
*/
break;

default:
// lwsl_info("%s: unknown msg type %d\n", __func__,
// h->nlmsg_type);
Expand Down

0 comments on commit 7ef2065

Please sign in to comment.