Java动态替换InetAddress中DNS的做法简单分析1

在java.net包描述中, 简要说明了一些关键的接口. 其中负责networking identifiers的是Addresses. 这个类的具体实现类是InetAddress, 底层封装了Inet4Address与Inet6Address的异同, 可以看成一个Facade工具类.

1. A Low Level API, which deals with the following abstractions:

2. Addresses, which are networking identifiers, like IP addresses.
3. Sockets, which are basic bidirectional data communication mechanisms.
4. Interfaces, which describe network interfaces.

复制代码

在OpenJDK的InetAddress源码中, 根据dns或hostname解析IP的代码部分:

1. private static InetAddress[] getAllByName0 (String host, InetAddress reqAddr, boolean check)

2. throws UnknownHostException  {
4. /* If it gets here it is presumed to be a hostname */

5. /* Cache.get can return: null, unknownAddress, or InetAddress[] */
7. /* make sure the connection to the host is allowed, before we

8. * give out a hostname
9. */
10. if (check) {
11. SecurityManager security = System.getSecurityManager();
12. if (security != null) {
13. security.checkConnect(host, -1);
14. }
15. }
17. InetAddress[] addresses = getCachedAddresses(host);

19. /* If no entry in cache, then do the host lookup */

20. if (addresses == null) {
21. addresses = getAddressesFromNameService(host, reqAddr);
22. }
24. if (addresses == unknown_array)

25. throw new UnknownHostException(host);
27. return addresses.clone();

28. }

复制代码

关键的二个方法是:
getCachedAddresses(host);
getAddressesFromNameService(host, reqAddr);
前者从addressCache, 或negativeCache根据dns/hostname解析缓存的IP.
后者从遍历nameServices,调用每个NameService的lookupAllHostAddr(host)查找IP, 然后将host:IP缓存到前面的cache中.
根据上述, 实现动态解析DNS, 有二种方式:
1. 反射addressCache, 或negativeCache, 将host:IP通过Cache的put()方法放入.
2. 反射nameServices,将代理的NameService实例放入.
二种做法的难处:
1. addressCache, 或negativeCache都是 java.net.InetAddress.Cache, 其内部的CacheEntry受二组JVM选项影响:
networkaddress.cache.ttl
networkaddress.cache.negative.ttl
在ttl后, CacheEntry的get()只会返回null.
2. nameServices只是OpenJDK的实现. 换言之,只是SUN一家的. 其他JDK不用此属性名.
写段代码看看Jrockit与IBM JVM中InetAddress内部的属性:

1. Class<InetAddress> type = InetAddress.class;

3. Field[] fields = type.getDeclaredFields();

5. for (Field f : fields) {

6. System.out.println(f.getName() + ":" + f.getType());
7. }

复制代码

OpenJDK:

1. IPv4

2. IPv6
3. preferIPv6Address
4. holder
5. nameServices
6. canonicalHostName
7. serialVersionUID
8. addressCache
9. negativeCache
10. addressCacheInit
11. unknown_array
12. impl
13. lookupTable
14. cachedLocalHost
15. cacheTime
16. maxCacheTime
17. cacheLock
18. FIELDS_OFFSET
19. UNSAFE
20. serialPersistentFields
21. $assertionsDisabled

复制代码

JRockit:

1. IPv4

2. IPv6
3. preferIPv6Address
4. hostName
5. address
6. family
7. nameService
8. canonicalHostName
9. serialVersionUID
10. addressCache
11. negativeCache
12. addressCacheInit
13. unknown_array
14. impl
15. lookupTable
16. $assertionsDisabled

复制代码

IBM JDK

1. IPv4:int

2. IPv6:int
3. preferIPv6Address
4. hostName:class java.lang.String
5. address:int
6. family:int
7. nameService:interface sun.net.spi.nameservice.NameService
8. canonicalHostName:class java.lang.String
9. serialVersionUID:long
10. addressCache:class java.net.InetAddress$Cache
11. negativeCache:class java.net.InetAddress$Cache
12. localHostName:class java.lang.String
13. localHostNameLock:class java.lang.Object
14. cacheLocalHost:boolean
15. addressCacheInit:boolean
16. unknown_array:class [Ljava.net.InetAddress;
17. impl:interface java.net.InetAddressImpl
18. lookupTable:class java.util.HashMap
19. $assertionsDisabled:boolean

复制代码

看到这里, 知道蛋疼了吧. 三种JDK中,
OpenJDK中是nameservices是个List<NameService>,
Jrockit与IBM JVM中却是nameservice, 只是单独的NameService实例.
所以要用第2种做法, 你至少要满足这三种主流JDK的需求.
简单实现二种做法:
做法1, 动态替换AddressCache.
做法2, 动态代理NameService.
源码如下

1. 长度限制, 源码查看回贴.

复制代码

暂时测试到这吧, 有兴趣的同学可以一起完善. 争取同时满足OpenJDK, Jrockit, IBM JDK三种主流环境的DNS动态解析类.