[LU-15478] Regression in 005bd7075c LU-10391 lnet: Change lnet_send() to take large-addr nids

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: Lustre 2.15.0
• Fix Version/s: Lustre 2.15.0
• Severity: 3
• Assignee: Chris Horn
• Reporter: Chris Horn

Description

Routed, source-any sends were broken by https://review.whamcloud.com/43599

lnet_handle_find_routed_path() calls lnet_find_route_locked() passing LNET_NID_NET(src_nid) as an argument.

                best_route = lnet_find_route_locked(best_rnet,
                                                    LNET_NID_NET(src_nid),
                                                    sd->sd_best_lpni,
                                                    &last_route, &gwni);

This network ID is in turn passed to lnet_find_best_lpni() where it is compared against LNET_NET_ANY:

static inline struct lnet_peer_ni *
lnet_find_best_lpni(struct lnet_ni *lni, lnet_nid_t dst_nid,
                    struct lnet_peer *peer, __u32 net_id)
{
        struct lnet_peer_net *peer_net;

        /* find the best_lpni on any local network */
        if (net_id == LNET_NET_ANY) {

Where

#define LNET_NET_ANY LNET_NIDNET(LNET_NID_ANY)
 == LNET_NIDNET(-1)
 == 0xffffffff

In the case where a source NID was not specified, the network id passed to lnet_find_best_lpni() is equal to

LNET_NID_NET(LNET_ANY_NID)

Where:

static inline __u32 LNET_NID_NET(const struct lnet_nid *nid)
{
        return LNET_MKNET(nid->nid_type, __be16_to_cpu(nid->nid_num));
}

I think we need an "extended nid" version of LNET_NET_ANY (call it LNET_ANY_NET) such that

#define LNET_ANY_NET LNET_NID_NET(&LNET_ANY_NID)

or LNET_NID_NET could be modified to check for LNET_ANY_NID and return LNET_NET_ANY.

Issue Links

• is related to: LU-10391 LNET: Support IPv6