tcp connection setup的实现(一)

bind的实现:

先来介绍几个地址结构.

struct sockaddr 其实相当于一个基类的地址结构,其他的结构都能够直接转到sockaddr.举个例子比如当sa_family为PF_INET时,sa_data就包含了端口号和ip地址(in_addr结构).

Java代码


struct sockaddr {

sa_family_t sa_family; /* address family, AF_xxx */

char sa_data[14]; /* 14 bytes of protocol address */

};

接下来就是sockaddr_in ,它表示了所有的ipv4的地址结构.可以看到他也就相当于sockaddr 的一个子类.

Java代码


struct sockaddr_in {

sa_family_t sin_family; /* Address family */

__be16 sin_port; /* Port number */

struct in_addr sin_addr; /* Internet address */

/* Pad to size of `struct sockaddr'. */

unsigned char __pad[__SOCK_SIZE__ - sizeof(short int) -

sizeof(unsigned short int) - sizeof(struct in_addr)];

};

这里还有一个内核比较新的地质结构sockaddr_storage,他可以容纳所有类型的套接口结构,比如ipv4,ipv6..可以看到它是强制对齐的,相比于sockaddr.

Java代码


struct __kernel_sockaddr_storage {

unsigned short ss_family; /* address family */

///每个协议实现自己的地址结构.

char __data[_K_SS_MAXSIZE - sizeof(unsigned short)];

/* space to achieve desired size, */

/* _SS_MAXSIZE value minus size of ss_family */

} __attribute__ ((aligned(_K_SS_ALIGNSIZE))); /* force desired alignment */

接下来看几个和bind相关的数据结构:

第一个是inet_hashinfo,它主要用来管理 tcp的bind hash bucket(在tcp的初始化函数中会将tcp_hashinfo初始化.然后在tcp_prot中会将tcp_hashinfo付给结构体h,然后相应的我们就可以通过sock中的sock_common域来存取这个值).后面我们会分析这个流程.

Java代码


struct inet_hashinfo {

/* This is for sockets with full identity only. Sockets here will

* always be without wildcards and will have the following invariant:

*

* TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE

*

* TIME_WAIT sockets use a separate chain (twchain).

*/

///下面会分析这个结构.

struct inet_ehash_bucket *ehash;

rwlock_t *ehash_locks;

unsigned int ehash_size;

unsigned int ehash_locks_mask;

/* Ok, let's try this, I give up, we do need a local binding

* TCP hash as well as the others for fast bind/connect.

*/

///表示所有的已经在使用的端口号的信息.这里bhash也就是一个hash链表,而链表的元素是inet_bind_bucket,紧接着我们会分析这个结构.

struct inet_bind_hashbucket *bhash;

unsigned int bhash_size;

/* Note : 4 bytes padding on 64 bit arches */

/* All sockets in TCP_LISTEN state will be in here. This is the only

* table where wildcard'd TCP sockets can exist. Hash function here

* is just local port number.

*/

///listening_hash表示所有的处于listen状态的socket.

struct hlist_head listening_hash[INET_LHTABLE_SIZE];

/* All the above members are written once at bootup and

* never written again _or_ are predominantly read-access.

*

* Now align to a new cache line as all the following members

* are often dirty.

*/

rwlock_t lhash_lock ____cacheline_aligned;

atomic_t lhash_users;

wait_queue_head_t lhash_wait;

struct kmem_cache *bind_bucket_cachep;

};

struct inet_ehash_bucket管理所有的tcp状态在TCP_ESTABLISHED和TCP_CLOSE之间的socket.这里要注意,twchain表示处于TIME_WAIT的socket.

Java代码


struct inet_ehash_bucket {

struct hlist_head chain;

struct hlist_head twchain;

};

inet_bind_bucket结构就是每个使用的端口的信息,最终会把它链接到bhash链表中.

Java代码


struct inet_bind_bucket {

struct net *ib_net;

///端口号

unsigned short port;

///表示这个端口是否能够被重复使用.

signed short fastreuse;

///指向下一个端口的inet_bind_bucket 结构.

struct hlist_node node;

///也就是使用这个端口的socket链表

struct hlist_head owners;

};

最后一个结构是tcp_hashinfo他在 tcp_init中被初始化,而tcp_init是在inet_init中被初始化的.然后tcp_hashinfo会被赋值给tcp_proto和sock的sk_prot域.

Java代码


struct inet_hashinfo __cacheline_aligned tcp_hashinfo = {

.lhash_lock = __RW_LOCK_UNLOCKED(tcp_hashinfo.lhash_lock),

.lhash_users = ATOMIC_INIT(0),

.lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(tcp_hashinfo.lhash_wait),

};

然后来看bind的实现,bind对应的系统调用是sys_bind:

Java代码


asmlinkage long sys_bind(int fd, struct sockaddr __user *umyaddr, int addrlen)

{

struct socket *sock;

struct sockaddr_storage address;

int err, fput_needed;

///通过fd查找相应的socket,如果不存在则返回错误.

sock = sockfd_lookup_light(fd, &err, &fput_needed);

if (sock) {

///用户空间和内核的地址拷贝.

err = move_addr_to_kernel(umyaddr, addrlen, (struct sockaddr *)&address);

if (err >= 0) {

err = security_socket_bind(sock,

(struct sockaddr *)&address,

addrlen);

if (!err)

///调用inet_bind方法.

err = sock->ops->bind(sock,

(struct sockaddr *)

&address, addrlen);

}

///将socket对应的file结构的引用计数.

fput_light(sock->file, fput_needed);

}

return err;

}

sockfd_lookup_light主要是查找fd对应的socket

Java代码


static struct socket *sockfd_lookup_light(int fd, int *err, int *fput_needed)

{

struct file *file;

struct socket *sock;

*err = -EBADF;

///通过fd得到对应的file结构

file = fget_light(fd, fput_needed);

if (file) {

///我们在sock_map_fd通过sock_attach_fd中已经把file的private域赋值为socket,因此这里就直接返回socket.

sock = sock_from_file(file, err);

if (sock)

return sock;

fput_light(file, *fput_needed);

}

return NULL;

}

然后来看inet_bind的实现.

Java代码


int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)

{

///取得绑定地址.以及相关的socket和inet_sock.

struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;

struct sock *sk = sock->sk;

struct inet_sock *inet = inet_sk(sk);

unsigned short snum;

int chk_addr_ret;

int err;

/* If the socket has its own bind function then use it. (RAW) */

if (sk->sk_prot->bind) {

err = sk->sk_prot->bind(sk, uaddr, addr_len);

goto out;

}

err = -EINVAL;

if (addr_len < sizeof(struct sockaddr_in))

goto out;

///得到地址类型,比如广播地址之类的.

chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);

err = -EADDRNOTAVAIL;

///主要是判断绑定的地址不是本地时的一些条件判断.

if (!sysctl_ip_nonlocal_bind &&

!inet->freebind &&

addr->sin_addr.s_addr != htonl(INADDR_ANY) &&

chk_addr_ret != RTN_LOCAL &&

chk_addr_ret != RTN_MULTICAST &&

chk_addr_ret != RTN_BROADCAST)

goto out;

///得到端口号.

snum = ntohs(addr->sin_port);

err = -EACCES;

///主要是端口号小于prot_sock(1024)必须得有root权限.如果没有则退出.capable就是用来判断权限的.

if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))

goto out;

/* We keep a pair of addresses. rcv_saddr is the one

* used by hash lookups, and saddr is used for transmit.

*

* In the BSD API these are the same except where it

* would be illegal to use them (multicast/broadcast) in

* which case the sending device address is used.

*/

lock_sock(sk);

/* Check these errors (active socket, double bind). */

err = -EINVAL;

///检测状态是否为close.如果是close状态,说明这个socket前面已经bind过了.而num只有当raw socket时才会不为0

if (sk->sk_state != TCP_CLOSE || inet->num)

goto out_release_sock;

///设置相应的地址.rcv_saddr是通过hash查找的源地址,而saddr是ip层使用的源地址(ip头的源地址).

inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;

///如果是多播或者广播,设置saddr.

if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)

inet->saddr = 0; /* Use device */

///这里get_port用来发现我们绑定的端口,是否被允许使用.而get_port在tcp中,被实例化为inet_csk_get_port,接近着我们会分析它的实现.

if (sk->sk_prot->get_port(sk, snum)) {

inet->saddr = inet->rcv_saddr = 0;

err = -EADDRINUSE;

goto out_release_sock;

}

///这两个锁不太理解.不知道谁能解释下.

if (inet->rcv_saddr)

sk->sk_userlocks |= SOCK_BINDADDR_LOCK;

if (snum)

sk->sk_userlocks |= SOCK_BINDPORT_LOCK;

///设置源端口

inet->sport = htons(inet->num);

///目的地址和目的端口,暂时设为0

inet->daddr = 0;

inet->dport = 0;

sk_dst_reset(sk);

err = 0;

out_release_sock:

release_sock(sk);

out:

return err;

}

这里我先来介绍下inet_csk_get_port的流程.

当绑定的port为0时,这时也就是说需要kernel来分配一个新的port.

1 首先得到系统的port范围.

2 随机分配一个port.

3 从bhash中得到当前随机分配的端口的链表(也就是inet_bind_bucket链表).

4 遍历这个链表(链表为空的话,也说明这个port没有被使用),如果这个端口已经被使用,则将端口号加一,继续循环,直到找到当前没有被使用的port,也就是没有在bhash中存在的port.

5 新建一个inet_bind_bucket,并插入到bhash中.

当指定port时.

1 从bhash中根据hash值(port计算的)取得当前指定端口对应的inet_bind_bucket结构.

2 如果bhash中存在,则说明,这个端口已经在使用,因此需要判断这个端口是否允许被reuse.

3 如果不存在,则步骤和上面的第5部一样.

Java代码


int inet_csk_get_port(struct sock *sk, unsigned short snum)

{

struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;

struct inet_bind_hashbucket *head;

struct hlist_node *node;

struct inet_bind_bucket *tb;

int ret;

struct net *net = sock_net(sk);

local_bh_disable();

if (!snum) {

///端口为0,也就是需要内核来分配端口.

int remaining, rover, low, high;

///得到端口范围.

inet_get_local_port_range(&low, &high);

remaining = (high - low) + 1;

rover = net_random() % remaining + low;

///循环来得到一个当前没有使用的端口.

do {

///通过端口为key,来得到相应的inet_bind_bucket

head = &hashinfo->bhash[inet_bhashfn(net, rover,

hashinfo->bhash_size)];

spin_lock(&head->lock);

inet_bind_bucket_for_each(tb, node, &head->chain)

if (tb->ib_net == net && tb->port == rover)

///说明这个端口已被使用,因此需要将端口加1,重新查找.

goto next;

break;

next:

spin_unlock(&head->lock);

///如果端口大于最大值,则将它赋值为最小值(这是因为我们这个端口是随机值,因此有可能很多端口就被跳过了),重新查找.

if (++rover > high)

rover = low;

} while (--remaining > 0);

/* Exhausted local port range during search? It is not

* possible for us to be holding one of the bind hash

* locks if this test triggers, because if 'remaining'

* drops to zero, we broke out of the do/while loop at

* the top level, not from the 'break;' statement.

*/

ret = 1;

if (remaining <= 0)

goto fail;

///将要分配的端口号.

snum = rover;

} else {

///指定端口号的情况.和上面的方法差不多,只不过只需要一次.

head = &hashinfo->bhash[inet_bhashfn(net, snum,

hashinfo->bhash_size)];

spin_lock(&head->lock);

inet_bind_bucket_for_each(tb, node, &head->chain)

if (tb->ib_net == net && tb->port == snum)

goto tb_found;

}

tb = NULL;

goto tb_not_found;

tb_found:

///用来处理端口号已经被使用的情况.他被使用的socket不为空的情况.

if (!hlist_empty(&tb->owners)) {

///fastreuse大于0说明其他的socket允许另外的socket也使用这个端口,而reuse表示当前的端口也允许和其他的端口分享这个port.并且socket的状态必须是TCP_LISTEN,才能做这个判断.

if (tb->fastreuse > 0 &&

sk->sk_reuse && sk->sk_state != TCP_LISTEN) {

goto success;

} else {

ret = 1;

///如果出错,调用inet_csk_bind_conflict.主要是有可能一些使用这个端口的socket,有可能使用不同的ip地址.此时,我们是可以使用这个端口的.

if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb))

goto fail_unlock;

}

}

tb_not_found:

ret = 1;

///重新分配一个inet_bind_bucket,并链接到bhash.

if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,

net, head, snum)) == NULL)

goto fail_unlock;

if (hlist_empty(&tb->owners)) {

///设置当前端口的fastreuse,这个域也只能是处于listen的socket才能设置.

if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)

tb->fastreuse = 1;

else

tb->fastreuse = 0;

} else if (tb->fastreuse &&

(!sk->sk_reuse || sk->sk_state == TCP_LISTEN))

tb->fastreuse = 0;

success:

///将这个socket加到这个端口的ower中.

if (!inet_csk(sk)->icsk_bind_hash)

inet_bind_hash(sk, tb, snum);

WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);

ret = 0;

fail_unlock:

spin_unlock(&head->lock);

fail:

local_bh_enable();

return ret;

}

在看listen的代码之前.我们也先来看相关的数据结构:

其中inet_connection_sock我们先前已经介绍过了,它包含了一个icsk_accept_queue的域,这个域是一个request_sock_queue类型,.我们就先来看这个结构:

request_sock_queue也就表示一个request_sock队列.这里我们知道,tcp中分为半连接队列(处于SYN_RECVD状态)和已完成连接队列(处于established状态).这两个一个是刚接到syn,等待三次握手完成,一个是已经完成三次握手,等待accept来读取.

这里每个syn分节到来都会新建一个request_sock结构,并将它加入到listen_sock的request_sock hash表中.然后3次握手完毕后,将它放入到request_sock_queue的rskq_accept_head和rskq_accept_tail队列中.这样当accept的时候就直接从这个队列中读取了.

Java代码


struct request_sock_queue {

///一个指向头,一个指向结尾.

struct request_sock *rskq_accept_head;

struct request_sock *rskq_accept_tail;

rwlock_t syn_wait_lock;

u8 rskq_defer_accept;

/* 3 bytes hole, try to pack */

///相应的listen_socket结构.

struct listen_sock *listen_opt;

};

listen_sock 表示一个处于listening状态的socket.

Java代码


struct listen_sock {

///log_2 of maximal queued SYNs/REQUESTs ,这里不太理解这个域的作用.

u8 max_qlen_log;

/* 3 bytes hole, try to use */

///当前的半连接队列的长度.

int qlen;

///也是指当前的半开连接队列长度,不过这个值会当重传syn/ack的时候(这里要注意是这个syn/ack第一次重传的时候才会减一)自动减一.

int qlen_young;

int clock_hand;

u32 hash_rnd;

///这个值表示了当前的syn_backlog(半开连接队列)的最大值

u32 nr_table_entries;

///半连接队列.

struct request_sock *syn_table[0];

};

最后来看下request_sock,它保存了tcp双方传输所必需的一些域,比如窗口大小,对端速率,对端数据包序列号等等这些值.

Java代码


struct request_sock {

struct request_sock *dl_next; /* Must be first member! */

///mss值.

u16 mss;

u8 retrans;

u8 cookie_ts; /* syncookie: encode tcpopts in timestamp */

/* The following two fields can be easily recomputed I think -AK */

u32 window_clamp; /* window clamp at creation time */

///窗口大小.

u32 rcv_wnd; /* rcv_wnd offered first time */

u32 ts_recent;

unsigned long expires;

///这个域包含了发送ack的操作集合.

const struct request_sock_ops *rsk_ops;

struct sock *sk;

u32 secid;

u32 peer_secid;

};

listen的对应的系统调用是sys_listen,它首先通过sockfd_lookup_light查找到相应的socket,然后调用inet_listen,大体流程和bind差不多,只不过中间调用的是inet_listen罢了.

这里还有一个概念那就是backlog,在linux中,backlog的大小指的是已完成连接队列的大小.而不是和半连接队列之和.而半开连接的大小一般是和backlog差不多大小.

而半开连接队列的最大长度是根据backlog计算的,我们后面会介绍这个.

因此我们直接来看inet_listen的实现,这个函数主要是进行一些合法性判断,然后调用inet_csk_listen_start来对相关域进行处理:

Java代码


int inet_listen(struct socket *sock, int backlog)

{

struct sock *sk = sock->sk;

unsigned char old_state;

int err;

lock_sock(sk);

err = -EINVAL;

///判断状态(非连接状态)以及socket类型.

if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)

goto out;

old_state = sk->sk_state;

///状态必须为close或者listen.

if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))

goto out;

/* Really, if the socket is already in listen state

* we can only allow the backlog to be adjusted.

*/

///非listen状态,需要我们处理.

if (old_state != TCP_LISTEN) {

err = inet_csk_listen_start(sk, backlog);

if (err)

goto out;

}

///将backlog赋值给sk_max_ack_backlog,也就是完全连接队列最大值.

sk->sk_max_ack_backlog = backlog;

err = 0;

out:

release_sock(sk);

return err;

}

然后来看inet_csk_listen_start的实现.

它的主要工作是新分配一个listen socket,将它加入到inet_connection_sock的icsk_accept_queue域的listen_opt中.然后对当前使用端口进行判断.最终返回:

Java代码


int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)

{

struct inet_sock *inet = inet_sk(sk);

struct inet_connection_sock *icsk = inet_csk(sk);

///新分配一个listen socket.

int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);

if (rc != 0)

return rc;

///先将这两个ack_backlog赋值为0.

sk->sk_max_ack_backlog = 0;

sk->sk_ack_backlog = 0;

inet_csk_delack_init(sk);

/* There is race window here: we announce ourselves listening,

* but this transition is still not validated by get_port().

* It is OK, because this socket enters to hash table only

* after validation is complete.

*/

///设置状态.

sk->sk_state = TCP_LISTEN;

///get_port上面已经分析过了.这里之所以还要再次判断一下端口,是为了防止多线程,也就是另一个线程在我们调用listen之前改变了这个端口的信息.

if (!sk->sk_prot->get_port(sk, inet->num)) {

//端口可用的情况,将端口值付给sport,并加入到inet_hashinfo(上面已经分析过)的listening_hash hash链表中.

inet->sport = htons(inet->num);

sk_dst_reset(sk);

///这里调用__inet_hash实现的.

sk->sk_prot->hash(sk);

return 0;

}

///不可用,则返回错误.

sk->sk_state = TCP_CLOSE;

__reqsk_queue_destroy(&icsk->icsk_accept_queue);

return -EADDRINUSE;

}

最后我们来看下reqsk_queue_alloc的实现:

Java代码


///半开连接的最大长度.

int sysctl_max_syn_backlog = 256;

int reqsk_queue_alloc(struct request_sock_queue *queue,

unsigned int nr_table_entries)

{

size_t lopt_size = sizeof(struct listen_sock);

struct listen_sock *lopt;

///在当前的nr_table_entries(也就是listen传进来的backlog)和sysctl_max_syn_backlog取一个较小的值.

nr_table_entries = min_t(u32, nr_table_entries, sysctl_max_syn_backlog);

///也就是说nr_table_entries不能小于8.

nr_table_entries = max_t(u32, nr_table_entries, 8);

///其实也就是使nr_table_entries更接近于2的次幂

nr_table_entries = roundup_pow_of_two(nr_table_entries + 1);

///最终所要分配的listen_sock 的大小.

lopt_size += nr_table_entries * sizeof(struct request_sock *);

if (lopt_size > PAGE_SIZE)

lopt = __vmalloc(lopt_size,

GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,

PAGE_KERNEL);

else

lopt = kzalloc(lopt_size, GFP_KERNEL);

if (lopt == NULL)

return -ENOMEM;

///计算max_qlen_log的值,他最小要为3,最大为对nr_table_entries求以2为低的log..

for (lopt->max_qlen_log = 3;

(1 << lopt->max_qlen_log) < nr_table_entries;

lopt->max_qlen_log++);

get_random_bytes(&lopt->hash_rnd, sizeof(lopt->hash_rnd));

rwlock_init(&queue->syn_wait_lock);

queue->rskq_accept_head = NULL;

///给nr_table_entries赋值.

lopt->nr_table_entries = nr_table_entries;

write_lock_bh(&queue->syn_wait_lock);

///将listen_socket赋值给queue->listen_opt

queue->listen_opt = lopt;

write_unlock_bh(&queue->syn_wait_lock);

return 0;

}