getline rows garbage, plus a leak

Garbage first, the leak second.

name, role and port are pointers into line. strtok just drops NULs into that buffer and hands back addresses inside it. getline then reuses the same buffer for the next row. So every host ends up pointing at the same memory, and you see whatever the last line left there. If line gets realloc'd it's worse, because the old pointers dangle.

Each row needs its own copy:

inv->hosts[inv->count].name = strdup(name);
inv->hosts[inv->count].role = strdup(role);

Check for NULL, and free both when you free the inventory. That's the lifetime boundary: line lives for one iteration, the host strings live as long as inv.

The leak: the early return -1 paths (bad row, failed grow) skip free(line). Only the happy path frees it. Either free it before each return or goto done to a single cleanup.

getline allocates the buffer, so the caller owns it. The POSIX page spells that out: https://pubs.opengroup.org/onlinepubs/9699919799/functions/getline.html

Garbage first. strtok gives pointers into line, so copy them with strdup. The getline page says the buffer is reallocated every call: https://pubs.opengroup.org/onlinepubs/9699919799/functions/getline.html

Thanks, that fixes it. Same cleanup for the early returns?

done:
	free(line);
	fclose(fp);
	return rc;

Agreed, copying with strdup is the right fix, and the getline page confirms the buffer gets reused. Pretty much what #2 said, too.

Yes, that works. Set rc = -1 and goto done at each early exit. Start with line = NULL.

	if (!name || !role) {
		fprintf(stderr, "%s:%d: expected 'name role [port]'\n",
			path, lineno);
		rc = -1;
		goto done;
	}

Set rc = -1 and goto done at each early exit.

Careful, the rows already copied stay in inv on failure, so those need freeing too, no?