Linux內核--網絡協議棧深入分析(三)--INET協議族socket和傳輸層sock

本文分析基於Linux Kernel 3.2.1

Linux內核中協議族有INET協議族,UNIX協議族等，我們還是以INET協議族為例。

更多請查看 Linux內核--網絡內核實現分析

下面是內核中的協議族聲明：

1. /* Supported address families. */
2. #define AF_UNSPEC 0
3. #define AF_UNIX 1 /* Unix domain sockets */
4. #define AF_LOCAL 1 /* POSIX name for AF_UNIX */
5. #define AF_INET 2 /* Internet IP Protocol */
6. #define AF_AX25 3 /* Amateur Radio AX.25 */
7. #define AF_IPX 4 /* Novell IPX */
8. #define AF_APPLETALK 5 /* AppleTalk DDP */
9. #define AF_NETROM 6 /* Amateur Radio NET/ROM */
10. #define AF_BRIDGE 7 /* Multiprotocol bridge */
11. #define AF_ATMPVC 8 /* ATM PVCs */
12. #define AF_X25 9 /* Reserved for X.25 project */
13. #define AF_INET6 10 /* IP version 6 */
14. #define AF_ROSE 11 /* Amateur Radio X.25 PLP */
15. #define AF_DECnet 12 /* Reserved for DECnet project */
16. #define AF_NETBEUI 13 /* Reserved for 802.2LLC project*/
17. #define AF_SECURITY 14 /* Security callback pseudo AF */
18. #define AF_KEY 15 /* PF_KEY key management API */
19. #define AF_NETLINK 16
20. #define AF_ROUTE AF_NETLINK /* Alias to emulate 4.4BSD */
21. #define AF_PACKET 17 /* Packet family */
22. #define AF_ASH 18 /* Ash */
23. #define AF_ECONET 19 /* Acorn Econet */
24. #define AF_ATMSVC 20 /* ATM SVCs */
25. #define AF_RDS 21 /* RDS sockets */
26. #define AF_SNA 22 /* Linux SNA Project (nutters!) */
27. #define AF_IRDA 23 /* IRDA sockets */
28. #define AF_PPPOX 24 /* PPPoX sockets */
29. #define AF_WANPIPE 25 /* Wanpipe API Sockets */
30. #define AF_LLC 26 /* Linux LLC */
31. #define AF_CAN 29 /* Controller Area Network */
32. #define AF_TIPC 30 /* TIPC sockets */
33. #define AF_BLUETOOTH 31 /* Bluetooth sockets */
34. #define AF_IUCV 32 /* IUCV sockets */
35. #define AF_RXRPC 33 /* RxRPC sockets */
36. #define AF_ISDN 34 /* mISDN sockets */
37. #define AF_PHONET 35 /* Phonet sockets */
38. #define AF_IEEE802154 36 /* IEEE802154 sockets */
39. #define AF_CAIF 37 /* CAIF sockets */
40. #define AF_ALG 38 /* Algorithm sockets */
41. #define AF_NFC 39 /* NFC sockets */
42. #define AF_MAX 40 /* For now.. */

內核中的PF_***和AF_***其實可以混用，它的宏定義如下：

1. /* Protocol families, same as address families. */
2. #define PF_UNSPEC AF_UNSPEC
3. #define PF_UNIX AF_UNIX
4. #define PF_LOCAL AF_LOCAL
5. #define PF_INET AF_INET
6. #define PF_AX25 AF_AX25
7. #define PF_IPX AF_IPX
8. #define PF_APPLETALK AF_APPLETALK
9. #define PF_NETROM AF_NETROM
10. #define PF_BRIDGE AF_BRIDGE
11. #define PF_ATMPVC AF_ATMPVC
12. #define PF_X25 AF_X25
13. #define PF_INET6 AF_INET6
14. #define PF_ROSE AF_ROSE
15. #define PF_DECnet AF_DECnet
16. #define PF_NETBEUI AF_NETBEUI
17. #define PF_SECURITY AF_SECURITY
18. #define PF_KEY AF_KEY
19. #define PF_NETLINK AF_NETLINK
20. #define PF_ROUTE AF_ROUTE
21. #define PF_PACKET AF_PACKET
22. #define PF_ASH AF_ASH
23. #define PF_ECONET AF_ECONET
24. #define PF_ATMSVC AF_ATMSVC
25. #define PF_RDS AF_RDS
26. #define PF_SNA AF_SNA
27. #define PF_IRDA AF_IRDA
28. #define PF_PPPOX AF_PPPOX
29. #define PF_WANPIPE AF_WANPIPE
30. #define PF_LLC AF_LLC
31. #define PF_CAN AF_CAN
32. #define PF_TIPC AF_TIPC
33. #define PF_BLUETOOTH AF_BLUETOOTH
34. #define PF_IUCV AF_IUCV
35. #define PF_RXRPC AF_RXRPC
36. #define PF_ISDN AF_ISDN
37. #define PF_PHONET AF_PHONET
38. #define PF_IEEE802154 AF_IEEE802154
39. #define PF_CAIF AF_CAIF
40. #define PF_ALG AF_ALG
41. #define PF_NFC AF_NFC
42. #define PF_MAX AF_MAX

以後的分析就是以INET協議族為例來分析的。

下面的結構體就是在系統初始化時用來管理協議族初始化的結構體：

1. struct net_proto_family {
2. int family;
3. int (*create)(struct net *net, struct socket *sock,
4. int protocol, int kern);
5. struct module *owner;
6. };

第一個屬性就是協議族的宏定義，如PF_INET；

第二個屬性就是協議族對應的初始化函數指針；

INET協議族對應該結構的定義如下：

1. static const struct net_proto_family inet_family_ops = {
2. .family = PF_INET,
3. .create = inet_create,
4. .owner = THIS_MODULE,
5. };

下面結構體是協議族操作集結構體定義：

1. struct proto_ops {
2. int family;
3. struct module *owner;
4. int (*release) (struct socket *sock);
5. int (*bind) (struct socket *sock,
6. struct sockaddr *myaddr,
7. int sockaddr_len);
8. int (*connect) (struct socket *sock,
9. struct sockaddr *vaddr,
10. int sockaddr_len, int flags);
11. int (*socketpair)(struct socket *sock1,
12. struct socket *sock2);
13. int (*accept) (struct socket *sock,
14. struct socket *newsock, int flags);
15. int (*getname) (struct socket *sock,
16. struct sockaddr *addr,
17. int *sockaddr_len, int peer);
18. unsigned int (*poll) (struct file *file, struct socket *sock,
19. struct poll_table_struct *wait);
20. int (*ioctl) (struct socket *sock, unsigned int cmd,
21. unsigned long arg);
22. #ifdef CONFIG_COMPAT
23. int (*compat_ioctl) (struct socket *sock, unsigned int cmd,
24. unsigned long arg);
25. #endif
26. int (*listen) (struct socket *sock, int len);
27. int (*shutdown) (struct socket *sock, int flags);
28. int (*setsockopt)(struct socket *sock, int level,
29. int optname, char __user *optval, unsigned int optlen);
30. int (*getsockopt)(struct socket *sock, int level,
31. int optname, char __user *optval, int __user *optlen);
32. #ifdef CONFIG_COMPAT
33. int (*compat_setsockopt)(struct socket *sock, int level,
34. int optname, char __user *optval, unsigned int optlen);
35. int (*compat_getsockopt)(struct socket *sock, int level,
36. int optname, char __user *optval, int __user *optlen);
37. #endif
38. int (*sendmsg) (struct kiocb *iocb, struct socket *sock,
39. struct msghdr *m, size_t total_len);
40. int (*recvmsg) (struct kiocb *iocb, struct socket *sock,
41. struct msghdr *m, size_t total_len,
42. int flags);
43. int (*mmap) (struct file *file, struct socket *sock,
44. struct vm_area_struct * vma);
45. ssize_t (*sendpage) (struct socket *sock, struct page *page,
46. int offset, size_t size, int flags);
47. ssize_t (*splice_read)(struct socket *sock, loff_t *ppos,
48. struct pipe_inode_info *pipe, size_t len, unsigned int flags);
49. };

INET協議族中TCP和UDP協議對應的上述操作集的定義不同：

TCP協議z在INET層操作集inet_stream_ops

1. const struct proto_ops inet_stream_ops = {
2. .family = PF_INET,
3. .owner = THIS_MODULE,
4. .release = inet_release,
5. .bind = inet_bind,
6. .connect = inet_stream_connect,
7. .socketpair = sock_no_socketpair,
8. .accept = inet_accept,
9. .getname = inet_getname,
10. .poll = tcp_poll,
11. .ioctl = inet_ioctl,
12. .listen = inet_listen,
13. .shutdown = inet_shutdown,
14. .setsockopt = sock_common_setsockopt,
15. .getsockopt = sock_common_getsockopt,
16. .sendmsg = inet_sendmsg,
17. .recvmsg = inet_recvmsg,
18. .mmap = sock_no_mmap,
19. .sendpage = inet_sendpage,
20. .splice_read = tcp_splice_read,
21. #ifdef CONFIG_COMPAT
22. .compat_setsockopt = compat_sock_common_setsockopt,
23. .compat_getsockopt = compat_sock_common_getsockopt,
24. .compat_ioctl = inet_compat_ioctl,
25. #endif
26. };

UDP協議在INET層操作集inet_dgram_ops

1. const struct proto_ops inet_dgram_ops = {
2. .family = PF_INET,
3. .owner = THIS_MODULE,
4. .release = inet_release,
5. .bind = inet_bind,
6. .connect = inet_dgram_connect,
7. .socketpair = sock_no_socketpair,
8. .accept = sock_no_accept,
9. .getname = inet_getname,
10. .poll = udp_poll,
11. .ioctl = inet_ioctl,
12. .listen = sock_no_listen,
13. .shutdown = inet_shutdown,
14. .setsockopt = sock_common_setsockopt,
15. .getsockopt = sock_common_getsockopt,
16. .sendmsg = inet_sendmsg,
17. .recvmsg = inet_recvmsg,
18. .mmap = sock_no_mmap,
19. .sendpage = inet_sendpage,
20. #ifdef CONFIG_COMPAT
21. .compat_setsockopt = compat_sock_common_setsockopt,
22. .compat_getsockopt = compat_sock_common_getsockopt,
23. .compat_ioctl = inet_compat_ioctl,
24. #endif
25. };

上面兩個操作集是屬於INET協議族層次，可以由協議族層套接字socket來管理，下面是協議族層析的套接字結構體（BSD Socket）定義：

1. /**
2. * struct socket - general BSD socket
3. * @state: socket state (%SS_CONNECTED, etc)
4. * @type: socket type (%SOCK_STREAM, etc)
5. * @flags: socket flags (%SOCK_ASYNC_NOSPACE, etc)
6. * @ops: protocol specific socket operations
7. * @file: File back pointer for gc
8. * @sk: internal networking protocol agnostic socket representation
9. * @wq: wait queue for several uses
10. */
11. struct socket {
12. socket_state state;
14. kmemcheck_bitfield_begin(type);
15. short type;
16. kmemcheck_bitfield_end(type);
18. unsigned long flags;
20. struct socket_wq __rcu *wq;
22. struct file *file;
23. struct sock *sk;
24. const struct proto_ops *ops;
25. };