libnvme: Add nvme_getifaddrs()

[martin-belanger]

The POSIX getifaddrs() API returns the list of network interfaces on the system, but invoking it can be costly. On large-scale systems with hundreds of NVMe-over-TCP connections, this API may be called hundreds of times, resulting in increased latency.

This patch introduces nvme_getifaddrs(), a wrapper around getifaddrs() that caches the results on the first invocation and reuses the cached data on subsequent calls.

Fixes: linux-nvme/libnvme#1098

[igaw]

Let's keep this function private and don't expose it to user.

[igaw]

it would be indeed good to tie it to the global context. If I read it correctly, the users of this function are getting either a struct nvme_ctrl * or a struct nvme_subystem * pointer as input:

bool nvme_ctrl_config_match(struct nvme_ctrl *c, const char *transport,
			    const char *traddr, const char *trsvcid,
			    const char *subsysnqn, const char *host_traddr,
			    const char *host_iface)
{
	_cleanup_candidate_ struct candidate_args candidate = {};
	ctrl_match_t ctrl_match;

	/* Init candidate and get the matching function to use */
	ctrl_match = _candidate_init(&candidate, transport, traddr, trsvcid,
				     subsysnqn, host_traddr, host_iface);

	return ctrl_match(c, &candidate);
}

That means it is possible to access the global context via those two pointers, c->ctx or s->h->ctx. They should be valid at this point (famous last words :)).

[martin-belanger]

Ah! You're right. It's been so long since I've worked on libnvme that I forgot how to access the ctx from a ctrl or subsys object. Let me rework this.

[martin-belanger]

@igaw - Thanks for your suggestion. I am now saving the cache in the global context.

[igaw]

Yes, this looks way better! Thanks!