1.Linux网络子系统
系统调用接口层
为应用程序提供访问网络子系统的统一方法。
协议无关层
提供通用的方法来使用传输层协议。
协议栈的实现
实现具体的网络协议
设备无关层
协议与设备驱动之前通信的通用接口
设备驱动程序
2.重要数据结构
2.1 网卡描述结构
在Linux内核中，每个网卡都由一个net_device结构来描述，其中的一些重要成员有：
char name[IFNAMSIZ]
设备名,如：eth%d
unsigned long base_addr
I/O 基地址
const struct net_device_ops *netdev_ops;
记录了网卡所支持的操作
2.2 网卡操作集合
类似于字符设备驱动中的file_operations结构，net_device_ops结构记录了网卡所支持的操作。
static const struct net_device_ops dm9000_netdev_ops =
{
.ndo_open = dm9000_open,
.ndo_stop = dm9000_stop,
.ndo_start_xmit = dm9000_start_xmit,
.ndo_do_ioctl = dm9000_ioctl,
.ndo_validate_addr = eth_validate_addr,
.ndo_set_mac_address = eth_mac_addr,
};
2.3 网络数据包
Linux内核中的每个网络数据包都由一个套接字缓冲区结构 struct sk_buff 描述，即一个sk_buff结构就是一个网络包,指向sk_buff的指针通常被称做skb。
3.网卡驱动架构分析
cs89x0.c:
/* cs89x0.c: A Crystal Semiconductor (Now Cirrus Logic) CS89[02]0
* driver for linux.
*/
static const struct net_device_ops net_ops = { //net_device_ops结构，各种网卡操作函数接口
.ndo_open= net_open,
.ndo_stop= net_close,
.ndo_tx_timeout= net_timeout,
.ndo_start_xmit = net_send_packet,
.ndo_get_stats= net_get_stats,
.ndo_set_multicast_list = set_multicast_list,
.ndo_set_mac_address = set_mac_address,
#ifdef CONFIG_NET_POLL_CONTROLLER
.ndo_poll_controller= net_poll_controller,
#endif
.ndo_change_mtu= eth_change_mtu,
.ndo_validate_addr= eth_validate_addr,
};
static netdev_tx_t net_send_packet(struct sk_buff *skb,struct net_device *dev) //网卡的发送函数
{
struct net_local *lp = netdev_priv(dev);
unsigned long flags;
if (net_debug > 3) {
printk("%s: sent %d byte packet of type %xn",
dev->name, skb->len,
(skb->data[ETH_ALEN+ETH_ALEN] << 8) | skb->data[ETH_ALEN+ETH_ALEN+1]);
}
/* keep the upload from being interrupted, since we
ask the chip to start transmitting before the
whole packet has been completely uploaded. */
spin_lock_irqsave(&lp->lock, flags);
netif_stop_queue(dev); //驱动程序通知网络子系统暂停数据包传输，从来进行实现流量控制。
/* initiate a transmit sequence */
writeword(dev->base_addr, TX_CMD_PORT, lp->send_cmd);
writeword(dev->base_addr, TX_LEN_PORT, skb->len);
/* Test to see if the chip has allocated memory for the packet */
if ((readreg(dev, PP_BusST) & READY_FOR_TX_NOW) == 0) {
/*
* It hasn't. But that shouldn't happen since
* we're waiting for TxOk, so return 1 and requeue this packet.
*/
spin_unlock_irqrestore(&lp->lock, flags);
if (net_debug) printk("cs89x0: Tx buffer not free!n");
return NETDEV_TX_BUSY;
}
/* Write the contents of the packet */
writewords(dev->base_addr, TX_FRAME_PORT,skb->data,(skb->len+1) >>1); //将skb的内容写入寄存器，发送
spin_unlock_irqrestore(&lp->lock, flags);
dev->stats.tx_bytes += skb->len;
dev_kfree_skb (skb); //释放skb
/*
* We DO NOT call netif_wake_queue() here.
* We also DO NOT call netif_start_queue().
*
* Either of these would cause another bottom half run through
* net_send_packet() before this packet has fully gone out. That causes
* us to hit the "Gasp!" above and the send is rescheduled. it runs like
* a dog. We just return and wait for the Tx completion interrupt handler
* to restart the netdevice layer
*/
return NETDEV_TX_OK;
}
/* The typical workload of the driver:
Handle the network interface interrupts. */
static irqreturn_t net_interrupt(int irq, void *dev_id) //网卡中断处理函数
{
struct net_device *dev = dev_id;
struct net_local *lp;
int ioaddr, status;
int handled = 0;
ioaddr = dev->base_addr;
lp = netdev_priv(dev);
/* we MUST read all the events out of the ISQ, otherwise we'll never
get interrupted again. As a consequence, we can't have any limit
on the number of times we loop in the interrupt handler. The
hardware guarantees that eventually we'll run out of events. Of
course, if you're on a slow machine, and packets are arriving
faster than you can read them off, you're screwed. Hasta la
vista, */
while ((status = readword(dev->base_addr, ISQ_PORT))) {
if (net_debug > 4)printk("%s: event=%04xn", dev->name, status);
handled = 1;
switch(status & ISQ_EVENT_MASK) {
case ISQ_RECEIVER_EVENT:
/* Got a packet(s). */
net_rx(dev);
break;
case ISQ_TRANSMITTER_EVENT:
dev->stats.tx_packets++;
netif_wake_queue(dev);/* Inform upper layers. */ //发送中停止设备，在这里被唤醒
if ((status & (TX_OK |
TX_LOST_CRS |
TX_SQE_ERROR |
TX_LATE_COL |
TX_16_COL)) != TX_OK) {
if ((status & TX_OK) == 0)
dev->stats.tx_errors++;
if (status & TX_LOST_CRS)
dev->stats.tx_carrier_errors++;
if (status & TX_SQE_ERROR)
dev->stats.tx_heartbeat_errors++;
if (status & TX_LATE_COL)
dev->stats.tx_window_errors++;
if (status & TX_16_COL)
dev->stats.tx_aborted_errors++;
}
break;
case ISQ_BUFFER_EVENT:
if (status & READY_FOR_TX) {
/* we tried to transmit a packet earlier,
but inexplicably ran out of buffers.
That shouldn't happen since we only ever
load one packet. Shrug. Do the right
thing anyway. */
netif_wake_queue(dev);/* Inform upper layers. */
}
if (status & TX_UNDERRUN) {
if (net_debug > 0) printk("%s: transmit underrunn", dev->name);
lp->send_underrun++;
if (lp->send_underrun == 3) lp->send_cmd = TX_AFTER_381;
else if (lp->send_underrun == 6) lp->send_cmd = TX_AFTER_ALL;
/* transmit cycle is done, although
frame wasn't transmitted - this
avoids having to wait for the upper
layers to timeout on us, in the
event of a tx underrun */
netif_wake_queue(dev);/* Inform upper layers. */
}
#if ALLOW_DMA
if (lp->use_dma && (status & RX_DMA)) {
int count = readreg(dev, PP_DmaFrameCnt);
while(count) {
if (net_debug > 5)
printk("%s: receiving %d DMA framesn", dev->name, count);
if (net_debug > 2 && count >1)
printk("%s: receiving %d DMA framesn", dev->name, count);
dma_rx(dev);
if (--count == 0)
count = readreg(dev, PP_DmaFrameCnt);
if (net_debug > 2 && count > 0)
printk("%s: continuing with %d DMA framesn", dev->name, count);
}
}
#endif
break;
case ISQ_RX_MISS_EVENT:
dev->stats.rx_missed_errors += (status >> 6);
break;
case ISQ_TX_COL_EVENT:
dev->stats.collisions += (status >> 6);
break;
}
}
return IRQ_RETVAL(handled);
}
...........
/* We have a good packet(s), get it/them out of the buffers. */
static void
net_rx(struct net_device *dev) //网卡的接收函数
{
struct sk_buff *skb;
int status, length;
int ioaddr = dev->base_addr;
status = readword(ioaddr, RX_FRAME_PORT); //读取待接收的数据状态
length = readword(ioaddr, RX_FRAME_PORT); //读取接收数据的长度
if ((status & RX_OK) == 0) {
count_rx_errors(status, dev);
return;
}
/* Malloc up new buffer. */
skb = dev_alloc_skb(length + 2); //分配skb长度+2的结构
if (skb == NULL) {
#if 0/* Again, this seems a cruel thing to do */
printk(KERN_WARNING "%s: Memory squeeze, dropping packet.n", dev->name);
#endif
dev->stats.rx_dropped++;
return;
}
skb_reserve(skb, 2);/* longword align L3 header */
readwords(ioaddr, RX_FRAME_PORT, skb_put(skb, length), length >> 1); //从网卡寄存器中读出数据，存入skb
if (length & 1)
skb->data[length-1] = readword(ioaddr, RX_FRAME_PORT);
if (net_debug > 3) {
printk("%s: received %d byte packet of type %xn",
dev->name, length,
(skb->data[ETH_ALEN+ETH_ALEN] << 8) | skb->data[ETH_ALEN+ETH_ALEN+1]);
}
skb->protocol=eth_type_trans(skb,dev);
netif_rx(skb); //将接收到的数据包skb交给协议栈-netif_rx
dev->stats.rx_packets++;
dev->stats.rx_bytes += length;
}
#if ALLOW_DMA
static void release_dma_buff(struct net_local *lp)
{
if (lp->dma_buff) {
free_pages((unsigned long)(lp->dma_buff), get_order(lp->dmasize * 1024));
lp->dma_buff = NULL;
}
}
#endif
/* The inverse routine to net_open(). */
static int
net_close(struct net_device *dev)
{
#if ALLOW_DMA
struct net_local *lp = netdev_priv(dev);
#endif
netif_stop_queue(dev);
writereg(dev, PP_RxCFG, 0);
writereg(dev, PP_TxCFG, 0);
writereg(dev, PP_BufCFG, 0);
writereg(dev, PP_BusCTL, 0);
free_irq(dev->irq, dev);
#if ALLOW_DMA
if (lp->use_dma && lp->dma) {
free_dma(dev->dma);
release_dma_buff(lp);
}
#endif
/* Update the statistics here. */
return 0;
}
........
int __init init_module(void) //网卡初始化
{
struct net_device *dev = alloc_etherdev(sizeof(struct net_local)); //分配net_device结构空间
struct net_local *lp;
int ret = 0;
#if DEBUGGING
net_debug = debug;
#else
debug = 0;
#endif
if (!dev)
return -ENOMEM;
dev->irq = irq; //初始化分配的dev
dev->base_addr = io;
lp = netdev_priv(dev);
#if ALLOW_DMA
if (use_dma) {
lp->use_dma = use_dma;
lp->dma = dma;
lp->dmasize = dmasize;
}
#endif
spin_lock_init(&lp->lock);
/* boy, they'd better get these right */
if (!strcmp(media, "rj45"))
lp->adapter_cnf = A_CNF_MEDIA_10B_T | A_CNF_10B_T;
else if (!strcmp(media, "aui"))
lp->adapter_cnf = A_CNF_MEDIA_AUI | A_CNF_AUI;
else if (!strcmp(media, "bnc"))
lp->adapter_cnf = A_CNF_MEDIA_10B_2 | A_CNF_10B_2;
else
lp->adapter_cnf = A_CNF_MEDIA_10B_T | A_CNF_10B_T;
if (duplex==-1)
lp->auto_neg_cnf = AUTO_NEG_ENABLE;
if (io == 0) {
printk(KERN_ERR "cs89x0.c: Module autoprobing not allowed.n");
printk(KERN_ERR "cs89x0.c: Append io=0xNNNn");
ret = -EPERM;
goto out;
} else if (io <= 0x1ff) {
ret = -ENXIO;
goto out;
}
#if ALLOW_DMA
if (use_dma && dmasize != 16 && dmasize != 64) {
printk(KERN_ERR "cs89x0.c: dma size must be either 16K or 64K, not %dKn", dmasize);
ret = -EPERM;
goto out;
}
#endif
ret = cs89x0_probe1(dev, io, 1); //这里依然在初始化：硬件初始化、MAC地址、注册网卡驱动(register_netdev)
if (ret)
goto out;
dev_cs89x0 = dev;
return 0;
out:
free_netdev(dev);
return ret;
}
void __exit
cleanup_module(void)
{
unregister_netdev(dev_cs89x0);
writeword(dev_cs89x0->base_addr, ADD_PORT, PP_ChipID);
release_region(dev_cs89x0->base_addr, NETCARD_IO_EXTENT);
free_netdev(dev_cs89x0);
}
#endif /* MODULE */
/*
* Local variables:
* version-control: t
* kept-new-versions: 5
* c-indent-level: 8
* tab-width: 8
* End:
*
*/