学习Linux-4.12内核网路协议栈（1.7）——网络设备的初始化（struct net_device）

客官°小女子只卖身不卖艺 2022-06-12 04:55 1251阅读 0赞

在linux的网络设备里，其中一个最关键的结构体应该要算net\_device了，它由对应的网络设备驱动进行创建和初始化，服务于内核网络子系统。

## 1. struct net\_device 注释分析    ##

struct net\_device这个结构体比较大，在了解它之前，我们先看一下它的注释：

1433 /\*\*  
1434 \* struct net\_device - The DEVICE structure.  
1435 \* Actually, this whole structure is a big mistake. It mixes I/O //这个结构体的设计是一个很大的失误，它并没有对IO数据和高级别的数据进行区分，也就是说这个结构  
1436 \* data with strictly "high-level" data, and it has to know about //体并没有对数据的来源是普通内存还是高速缓存进行辨别，因此在INET模型里面，它不得不处理各种  
1437 \* almost every data structure used in the INET module. //不同的数据类型  
1438 \*  
1439 \* @name: This is the first field of the "visible" part of this structure //它代表一个接口的名字，在设备注册的时候，我们可以指定接口名字，如果没指定，他会自动申请  
1440 \* (i.e. as seen by users in the "Space.c" file). It is the name //一个自加1的名字，比如eth0，eth1，eth2...  
1441 \* of the interface.  
1442 \*  
1443 \* @name\_hlist: Device name hash chain, please keep it close to name\[\] //以名字为索引的哈希表  
1444 \* @ifalias: SNMP alias // snmp的别名  
1445 \* @mem\_end: Shared memory end //每一个设备都会分配一块内存区域，start和end指定了这块区域  
1446 \* @mem\_start: Shared memory start  
1447 \* @base\_addr: Device I/O address //网络硬件设备的基地址，内存管理系统将每一个外部设备都看作一块连续的地址，然后将它与内存中的一块地址进行映射，这样操作内存地址就相当于操作这块网络硬件设备的地址，而这里的基地址就是这个网络硬件设备的起始地址。他会在probe的时候初始化

1448 \* @irq: Device IRQ number //该设备对应的中断号  
1449 \*  
1450 \* @carrier\_changes: Stats to monitor carrier on<->off transitions  
1451 \*  
1452 \* @state: Generic network queuing layer state, see netdev\_state\_t //表示设备的状态，它很重要  
1453 \* @dev\_list: The global list of network devices //所有net\_device对象组成的一个链表，可以说系统中所有的网络设备都可以通过它查到  
1454 \* @napi\_list: List entry used for polling NAPI devices //如果该支持NAPI，会将它挂到这个链表上，CPU就可以更快的找到NAPI poll的设备  
1455 \* @unreg\_list: List entry when we are unregistering the //正在被卸载的设备会加到这个链表  
1456 \* device; see the function unregister\_netdev  
1457 \* @close\_list: List entry used when we are closing the device //正在被关闭的设备会加到这个链表  
1458 \* @ptype\_all: Device-specific packet handlers for all protocols //某些特定协议的处理函数会挂接在这里，但是未必是需要的  
1459 \* @ptype\_specific: Device-specific, protocol-specific packet handlers

1460 \*  
1461 \* @adj\_list: Directly linked devices, like slaves for bonding  
1462 \* @features: Currently active device features //用来标识接口的各种能力和特性  
1463 \* @hw\_features: User-changeable features //一些硬件相关的特性，这些是可以在用户空间修改的  
1464 \*  
1465 \* @wanted\_features: User-requested features  
1466 \* @vlan\_features: Mask of features inheritable by VLAN devices //是否支持vlan功能  
1467 \*  
1468 \* @hw\_enc\_features: Mask of features inherited by encapsulating devices //是否支持硬件封装功能  
1469 \* This field indicates what encapsulation  
1470 \* offloads the hardware is capable of doing,  
1471 \* and drivers will need to set them appropriately.  
1472 \*  
1473 \* @mpls\_features: Mask of features inheritable by MPLS  
1474 \*  
1475 \* @ifindex: interface index //内核指定的索引号，比如第一个，第二个设备等等  
1476 \* @group: The group the device belongs to //这个设备属于哪个组  
1477 \*  
1478 \* @stats: Statistics struct, which was left as a legacy, use //一些接口的信息，用于提供给旧接口的用户空间获取  
1479 \* rtnl\_link\_stats64 instead  
1480 \*  
1481 \* @rx\_dropped: Dropped packets by core network, //被内核丢掉的包，注意不是被driver丢的  
1482 \* do not use this in drivers  
1483 \* @tx\_dropped: Dropped packets by core network,  
1484 \* do not use this in drivers  
1485 \* @rx\_nohandler: nohandler dropped packets by core network on  
1486 \* inactive devices, do not use this in drivers  
1487 \*  
1488 \* @wireless\_handlers: List of functions to handle Wireless Extensions, //无线子系统的一些接口  
1489 \* instead of ioctl,  
1490 \* see <net/iw\_handler.h> for details.  
1491 \* @wireless\_data: Instance data managed by the core of wireless extensions  
1492 \*  
1493 \* @netdev\_ops: Includes several pointers to callbacks, //很重要！操作网络设备的函数都聚集在这里了，在网络初始化的时候被初始化，具体支持哪些操作函数，  
1494 \* if one wants to override the ndo\_\*() functions //请看这个函数struct net\_device\_ops（）  
1495 \* @ethtool\_ops: Management operations //ethtool的操作接口  
1496 \* @ndisc\_ops: Includes callbacks for different IPv6 neighbour  
1497 \* discovery handling. Necessary for e.g. 6LoWPAN.  
1498 \* @header\_ops: Includes callbacks for creating,parsing,caching,etc //对L2头部处理的函数  
1499 \* of Layer 2 headers.  
1500 \*  
1501 \* @flags: Interface flags (a la BSD) //标识接口的状态，比如UP/down等，可以通过用户空间修改  
1502 \* @priv\_flags: Like 'flags' but invisible to userspace, //和flags类似，但是用户空间不能修改  
1503 \* see if.h for the definitions  
1504 \* @gflags: Global flags ( kept as legacy ) //全局标识，和flags配合使用  
1505 \* @padded: How much padding added by alloc\_netdev() //对齐时使用的字节数，在申请net\_device的时候，需要进行对齐，它表示填充的字节数  
1506 \* @operstate: RFC2863 operstate  
1507 \* @link\_mode: Mapping policy to operstate  
1508 \* @if\_port: Selectable AUI, TP, ... 目前较少用，对于支持多介质的网络设备时，用来指定哪种设备的接口  
1509 \* @dma: DMA channel //为该设备分配的DMA通道，如果支持的话，目前来说应该都支持了  
1510 \* @mtu: Interface MTU value //这个不用说了，一般1500  
1511 \* @min\_mtu: Interface Minimum MTU value  
1512 \* @max\_mtu: Interface Maximum MTU value  
1513 \* @type: Interface hardware type //接口的硬件类型，目前来说主要都是以太网

1514 \* @hard\_header\_len: Maximum hardware header length.  
1515 \* @min\_header\_len: Minimum hardware header length  
1516 \*  
1517 \* @needed\_headroom: Extra headroom the hardware may need, but not in all //需要头部空间吗  
1518 \* cases can this be guaranteed  
1519 \* @needed\_tailroom: Extra tailroom the hardware may need, but not in all  
1520 \* cases can this be guaranteed. Some cases also use  
1521 \* LL\_MAX\_HEADER instead to allocate the skb  
1522 \*  
1523 \* interface address info:  
1524 \*  
1525 \* @perm\_addr: Permanent hw address //烧写在硬件中的地址，初始化的时候读取到这里  
1526 \* @addr\_assign\_type: Hw address assignment type //硬件地址分配类型，目前来说都是支持用户空间对硬件地址进行设置了  
1527 \* @addr\_len: Hardware address length //这个不用说了，14B  
1528 \* @neigh\_priv\_len: Used in neigh\_alloc()  
1529 \* @dev\_id: Used to differentiate devices that share //这个应该很少用了，如果有多个设备共用一个mac地址，就会有它的作用了，目前见过这样的产品，虽然mac  
1530 \* the same link layer address //地址一样，但是硬件设备不一样，工作是没有问题的  
1531 \* @dev\_port: Used to differentiate devices that share //如果有多个网络接口实现相同的功能就会用到  
1532 \* the same function  
1533 \* @addr\_list\_lock: XXX: need comments on this one  
1534 \* @uc\_promisc: Counter that indicates promiscuous mode //我们知道，如果不是在混杂模式下，网卡只会接收发往自己的单播地址， 但是如果同时想接收发往其他  
1535 \* has been enabled due to the need to listen to //mac的单播地址，就需要添加到这里让驱动不要过滤掉  
1536 \* additional unicast addresses in a device that  
1537 \* does not implement ndo\_set\_rx\_mode()  
1538 \* @uc: unicast mac addresses //自己的单播地址  
1539 \* @mc: multicast mac addresses //自己的广播地址  
1540 \* @dev\_addrs: list of device hw addresses //现在的设备可能同时使用多个mac地址，那么将会保留在这个链表里面  
1541 \* @queues\_kset: Group of all Kobjects in the Tx and RX queues //Tx和Rx链的对象  
1542 \* @promiscuity: Number of times the NIC is told to work in //是否工作在混杂模式  
1543 \* promiscuous mode; if it becomes 0 the NIC will  
1544 \* exit promiscuous mode  
1545 \* @allmulti: Counter, enables or disables allmulticast mode //开启或关闭allmulti功能，可以通过ifconfig命令设置  
1546 \*  
1547 \* @vlan\_info: VLAN info //顾名思义  
1548 \* @dsa\_ptr: dsa specific data //下面是各种不同类型包  
1549 \* @tipc\_ptr: TIPC specific data  
1550 \* @atalk\_ptr: AppleTalk link  
1552 \* @dn\_ptr: DECnet specific data  
1553 \* @ip6\_ptr: IPv6 specific data //  
1554 \* @ax25\_ptr: AX.25 specific data  
1555 \* @ieee80211\_ptr: IEEE 802.11 specific data, assign before registering  
1556 \*  
1557 \* @dev\_addr: Hw address (before bcast, //设备的mac地址  
1558 \* because most packets are unicast)  
1559 \*  
1560 \* @\_rx: Array of RX queues //与发包相关的一些设置  
1561 \* @num\_rx\_queues: Number of RX queues  
1562 \* allocated at register\_netdev() time  
1563 \* @real\_num\_rx\_queues: Number of RX queues currently active in device  
1564 \*  
1565 \* @rx\_handler: handler for received packets //收包处理函数  
1566 \* @rx\_handler\_data: XXX: need comments on this one  
1567 \* @ingress\_queue: XXX: need comments on this one  
1568 \* @broadcast: hw bcast address //广播地址  
1569 \*  
1570 \* @rx\_cpu\_rmap: CPU reverse-mapping for RX completion interrupts,  
1571 \* indexed by RX queue number. Assigned by driver.  
1572 \* This must only be set if the ndo\_rx\_flow\_steer  
1573 \* operation is defined  
1574 \* @index\_hlist: Device index hash chain  
1575 \*  
1576 \* @\_tx: Array of TX queues //与收报相关的以下设置  
1577 \* @num\_tx\_queues: Number of TX queues allocated at alloc\_netdev\_mq() time  
1578 \* @real\_num\_tx\_queues: Number of TX queues currently active in device  
1579 \* @qdisc: Root qdisc from userspace point of view  
1580 \* @tx\_queue\_len: Max frames per queue allowed  
1581 \* @tx\_global\_lock: XXX: need comments on this one  
1582 \*  
1583 \* @xps\_maps: XXX: need comments on this one  
1584 \*  
1585 \* @watchdog\_timeo: Represents the timeout that is used by //initial的时候该函数被初始化，网络层确定传输已经超时，将会调用driver中的tx\_timeout处理时间  
1586 \* the watchdog (see dev\_watchdog())  
1587 \* @watchdog\_timer: List of timers  
1588 \*  
1589 \* @pcpu\_refcnt: Number of references to this device //该设备被多少个CPU引用  
1590 \* @todo\_list: Delayed register/unregister //下面是和卸载相关的一些设定  
1591 \* @link\_watch\_list: XXX: need comments on this one  
1592 \*  
1593 \* @reg\_state: Register/unregister state machine  
1594 \* @dismantle: Device is going to be freed  
1595 \* @rtnl\_link\_state: This enum represents the phases of creating  
1596 \* a new link  
1597 \*  
1598 \* @needs\_free\_netdev: Should unregister perform free\_netdev?  
1599 \* @priv\_destructor: Called from unregister  
1600 \* @npinfo: XXX: need comments on this one  
1601 \* @nd\_net: Network namespace this network device is inside  
1602 \*  
1603 \* @ml\_priv: Mid-layer private //统计信息  
1604 \* @lstats: Loopback statistics  
1605 \* @tstats: Tunnel statistics  
1606 \* @dstats: Dummy statistics  
1607 \* @vstats: Virtual ethernet statistics  
1608 \*  
1609 \* @garp\_port: GARP //免费ARP接口  
1610 \* @mrp\_port: MRP //MAR接口  
1611 \*  
1612 \* @dev: Class/net/name entry //虽然是网络设备，它终究是普通设备，所以它也有普通设备该有的属性，也就是struct device结构体里面的属性  
1613 \* @sysfs\_groups: Space for optional device, statistics and wireless  
1614 \* sysfs groups  
1615 \*  
1616 \* @sysfs\_rx\_queue\_group: Space for optional per-rx queue attributes  
1617 \* @rtnl\_link\_ops: Rtnl\_link\_ops //netlink接口操作函数  
1618 \*  
1619 \* @gso\_max\_size: Maximum size of generic segmentation offload  
1620 \* @gso\_max\_segs: Maximum number of segments that can be passed to the  
1621 \* NIC for GSO  
1622 \*  
1623 \* @dcbnl\_ops: Data Center Bridging netlink ops //桥接操作函数  
1624 \* @num\_tc: Number of traffic classes in the net device  
1625 \* @tc\_to\_txq: XXX: need comments on this one  
1626 \* @prio\_tc\_map: XXX: need comments on this one  
1627 \*  
1628 \* @fcoe\_ddp\_xid: Max exchange id for FCoE LRO by ddp  
1629 \*  
1630 \* @priomap: XXX: need comments on this one  
1631 \* @phydev: Physical device may attach itself  
1632 \* for hardware timestamping  
1633 \*  
1634 \* @qdisc\_tx\_busylock: lockdep class annotating Qdisc->busylock spinlock  
1635 \* @qdisc\_running\_key: lockdep class annotating Qdisc->running seqcount  
1636 \*  
1637 \* @proto\_down: protocol port state information can be sent to the  
1638 \* switch driver and used to set the phys state of the  
1639 \* switch port.  
1640 \*  
1641 \* FIXME: cleanup struct net\_device such that network protocol info  
1642 \* moves out.  
1643 \*/  
1644

## 2. struct net\_device 结构体 ##

上面这些是对struct net\_device的基本介绍，下面将进一步介绍结构体的具体定义，需要说明的是，这个结构体很重要，所以了解越详细越好。

1645 struct net\_device \{  
1646 char name\[IFNAMSIZ\];  
1647 struct hlist\_node name\_hlist;  
1648 char \*ifalias;  
1649 /\*  
1650 \* I/O specific fields  
1651 \* FIXME: Merge these and struct ifmap into one  
1652 \*/  
1653 unsigned long mem\_end;  
1654 unsigned long mem\_start;  
1655 unsigned long base\_addr;  
1656 int irq;  
1657  
1658 atomic\_t carrier\_changes;  
1659  
1660 /\*  
1661 \* Some hardware also needs these fields (state,dev\_list,  
1662 \* napi\_list,unreg\_list,close\_list) but they are not  
1663 \* part of the usual set specified in Space.c.  
1664 \*/  
1665  
1666 unsigned long state;  
1667  
1668 struct list\_head dev\_list;  
1669 struct list\_head napi\_list;  
1670 struct list\_head unreg\_list;  
1671 struct list\_head close\_list;  
1672 struct list\_head ptype\_all;  
1673 struct list\_head ptype\_specific;  
1674  
1675 struct \{  
1676 struct list\_head upper;  
1677 struct list\_head lower;  
1678 \} adj\_list;  
1679  
1680 netdev\_features\_t features;  
1681 netdev\_features\_t hw\_features;  
1682 netdev\_features\_t wanted\_features;  
1683 netdev\_features\_t vlan\_features;  
1684 netdev\_features\_t hw\_enc\_features;  
1685 netdev\_features\_t mpls\_features;  
1686 netdev\_features\_t gso\_partial\_features;  
1687  
1688 int ifindex;  
1689 int group;  
1690  
1691 struct net\_device\_stats stats;  
1692  
1693 atomic\_long\_t rx\_dropped;  
1694 atomic\_long\_t tx\_dropped;  
1695 atomic\_long\_t rx\_nohandler;  
1696  
1697 \#ifdef CONFIG\_WIRELESS\_EXT  
1698 const struct iw\_handler\_def \*wireless\_handlers;  
1699 struct iw\_public\_data \*wireless\_data;  
1700 \#endif  
1701 const struct net\_device\_ops \*netdev\_ops;  
1702 const struct ethtool\_ops \*ethtool\_ops;  
1703 \#ifdef CONFIG\_NET\_SWITCHDEV  
1704 const struct switchdev\_ops \*switchdev\_ops;  
1705 \#endif  
1706 \#ifdef CONFIG\_NET\_L3\_MASTER\_DEV  
1707 const struct l3mdev\_ops \*l3mdev\_ops;  
1708 \#endif  
1709 \#if IS\_ENABLED(CONFIG\_IPV6)  
1710 const struct ndisc\_ops \*ndisc\_ops;  
1711 \#endif  
1712  
1713 \#ifdef CONFIG\_XFRM  
1714 const struct xfrmdev\_ops \*xfrmdev\_ops;  
1715 \#endif  
1716  
1717 const struct header\_ops \*header\_ops;  
1718  
1719 unsigned int flags;  
1720 unsigned int priv\_flags;  
1721  
1722 unsigned short gflags;  
1723 unsigned short padded;  
1724  
1725 unsigned char operstate;  
1726 unsigned char link\_mode;  
1727  
1728 unsigned char if\_port;  
1729 unsigned char dma;  
1730  
1731 unsigned int mtu;  
1732 unsigned int min\_mtu;  
1733 unsigned int max\_mtu;  
1734 unsigned short type;  
1735 unsigned short hard\_header\_len;  
1736 unsigned char min\_header\_len;  
1737  
1738 unsigned short needed\_headroom;  
1739 unsigned short needed\_tailroom;  
1740  
1741 /\* Interface address info. \*/  
1742 unsigned char perm\_addr\[MAX\_ADDR\_LEN\];  
1743 unsigned char addr\_assign\_type;  
1744 unsigned char addr\_len;  
1745 unsigned short neigh\_priv\_len;  
1746 unsigned short dev\_id;  
1747 unsigned short dev\_port;  
1748 spinlock\_t addr\_list\_lock;  
1749 unsigned char name\_assign\_type;  
1750 bool uc\_promisc;  
1751 struct netdev\_hw\_addr\_list uc;  
1752 struct netdev\_hw\_addr\_list mc;  
1753 struct netdev\_hw\_addr\_list dev\_addrs;  
1754  
1755 \#ifdef CONFIG\_SYSFS  
1756 struct kset \*queues\_kset;  
1757 \#endif  
1758 unsigned int promiscuity;  
1759 unsigned int allmulti;  
1760  
1761  
1762 /\* Protocol-specific pointers \*/  
1763  
1764 \#if IS\_ENABLED(CONFIG\_VLAN\_8021Q)  
1765 struct vlan\_info \_\_rcu \*vlan\_info;  
1766 \#endif  
1767 \#if IS\_ENABLED(CONFIG\_NET\_DSA)  
1768 struct dsa\_switch\_tree \*dsa\_ptr;  
1769 \#endif  
1770 \#if IS\_ENABLED(CONFIG\_TIPC)  
1771 struct tipc\_bearer \_\_rcu \*tipc\_ptr;  
1772 \#endif  
1773 void \*atalk\_ptr;  
1774  struct in\_device \_\_rcu \*ip\_ptr;  
1775 struct dn\_dev \_\_rcu \*dn\_ptr;  
1776 struct inet6\_dev \_\_rcu \*ip6\_ptr;  
1777 void \*ax25\_ptr;  
1778 struct wireless\_dev \*ieee80211\_ptr;  
1779 struct wpan\_dev \*ieee802154\_ptr;  
1780 \#if IS\_ENABLED(CONFIG\_MPLS\_ROUTING)  
1781 struct mpls\_dev \_\_rcu \*mpls\_ptr;  
1782 \#endif  
1783  
1784 /\*  
1785 \* Cache lines mostly used on receive path (including eth\_type\_trans())  
1786 \*/  
1787 /\* Interface address info used in eth\_type\_trans() \*/  
1788 unsigned char \*dev\_addr;  
1789  
1790 \#ifdef CONFIG\_SYSFS  
1791 struct netdev\_rx\_queue \*\_rx;  
1792  
1793 unsigned int num\_rx\_queues;  
1794 unsigned int real\_num\_rx\_queues;  
1795 \#endif  
1796  
1797 struct bpf\_prog \_\_rcu \*xdp\_prog;  
1798 unsigned long gro\_flush\_timeout;  
1799 rx\_handler\_func\_t \_\_rcu \*rx\_handler;  
1800 void \_\_rcu \*rx\_handler\_data;  
1801  
1802 \#ifdef CONFIG\_NET\_CLS\_ACT  
1803 struct tcf\_proto \_\_rcu \*ingress\_cl\_list;  
1804 \#endif  
1805 struct netdev\_queue \_\_rcu \*ingress\_queue;  
1806 \#ifdef CONFIG\_NETFILTER\_INGRESS  
1807 struct nf\_hook\_entry \_\_rcu \*nf\_hooks\_ingress;  
1808 \#endif  
1809  
1810 unsigned char broadcast\[MAX\_ADDR\_LEN\];  
1811 \#ifdef CONFIG\_RFS\_ACCEL  
1812 struct cpu\_rmap \*rx\_cpu\_rmap;  
1813 \#endif  
1814 struct hlist\_node index\_hlist;  
1815  
1816 /\*  
1817 \* Cache lines mostly used on transmit path  
1818 \*/  
1819 struct netdev\_queue \*\_tx \_\_\_\_cacheline\_aligned\_in\_smp;  
1820 unsigned int num\_tx\_queues;  
1821 unsigned int real\_num\_tx\_queues;  
1822 struct Qdisc \*qdisc;  
1823 \#ifdef CONFIG\_NET\_SCHED  
1824 DECLARE\_HASHTABLE (qdisc\_hash, 4);  
1825 \#endif  
1826 unsigned long tx\_queue\_len;  
1827 spinlock\_t tx\_global\_lock;  
1828 int watchdog\_timeo;  
1829  
1830 \#ifdef CONFIG\_XPS  
1831 struct xps\_dev\_maps \_\_rcu \*xps\_maps;  
1832 \#endif  
1833 \#ifdef CONFIG\_NET\_CLS\_ACT  
1834 struct tcf\_proto \_\_rcu \*egress\_cl\_list;  
1835 \#endif  
1836  
1837 /\* These may be needed for future network-power-down code. \*/  
1838 struct timer\_list watchdog\_timer;  
1839  
1840 int \_\_percpu \*pcpu\_refcnt;  
1841 struct list\_head todo\_list;  
1842  
1843 struct list\_head link\_watch\_list;  
1844  
1845 enum \{ NETREG\_UNINITIALIZED=0,  
1846 NETREG\_REGISTERED, /\* completed register\_netdevice \*/  
1847 NETREG\_UNREGISTERING, /\* called unregister\_netdevice \*/  
1848 NETREG\_UNREGISTERED, /\* completed unregister todo \*/  
1849 NETREG\_RELEASED, /\* called free\_netdev \*/  
1850 NETREG\_DUMMY, /\* dummy device for NAPI poll \*/  
1851 \} reg\_state:8;  
1852  
1853 bool dismantle;  
1854  
1855 enum \{  
1856 RTNL\_LINK\_INITIALIZED,  
1857 RTNL\_LINK\_INITIALIZING,  
1858 \} rtnl\_link\_state:16;  
1859  
1860 bool needs\_free\_netdev;  
1861 void (\*priv\_destructor)(struct net\_device \*dev);  
1862  
1863 \#ifdef CONFIG\_NETPOLL  
1864 struct netpoll\_info \_\_rcu \*npinfo;  
1865 \#endif  
1866  
1867 possible\_net\_t nd\_net;  
1868  
1869 /\* mid-layer private \*/  
1870 union \{  
1871 void \*ml\_priv;  
1872 struct pcpu\_lstats \_\_percpu \*lstats;  
1873 struct pcpu\_sw\_netstats \_\_percpu \*tstats;  
1874 struct pcpu\_dstats \_\_percpu \*dstats;  
1875 struct pcpu\_vstats \_\_percpu \*vstats;  
1876 \};  
1877  
1878 \#if IS\_ENABLED(CONFIG\_GARP)  
1879 struct garp\_port \_\_rcu \*garp\_port;  
1880 \#endif  
1881 \#if IS\_ENABLED(CONFIG\_MRP)  
1882 struct mrp\_port \_\_rcu \*mrp\_port;  
1883 \#endif  
1884  
1885 struct device dev;  
1886 const struct attribute\_group \*sysfs\_groups\[4\];  
1887 const struct attribute\_group \*sysfs\_rx\_queue\_group;  
1888  
1889 const struct rtnl\_link\_ops \*rtnl\_link\_ops;  
1890  
1891 /\* for setting kernel sock attribute on TCP connection setup \*/  
1892 \#define GSO\_MAX\_SIZE 65536  
1893 unsigned int gso\_max\_size;  
1894 \#define GSO\_MAX\_SEGS 65535  
1895 u16 gso\_max\_segs;  
1896  
1897 \#ifdef CONFIG\_DCB  
1898 const struct dcbnl\_rtnl\_ops \*dcbnl\_ops;  
1899 \#endif  
1900 u8 num\_tc;  
1901 struct netdev\_tc\_txq tc\_to\_txq\[TC\_MAX\_QUEUE\];  
1902 u8 prio\_tc\_map\[TC\_BITMASK + 1\];  
1903  
1904 \#if IS\_ENABLED(CONFIG\_FCOE)  
1905 unsigned int fcoe\_ddp\_xid;  
1906 \#endif  
1907 \#if IS\_ENABLED(CONFIG\_CGROUP\_NET\_PRIO)  
1908 struct netprio\_map \_\_rcu \*priomap;  
1909 \#endif  
1910 struct phy\_device \*phydev;  
1911 struct lock\_class\_key \*qdisc\_tx\_busylock;  
1912 struct lock\_class\_key \*qdisc\_running\_key;  
1913 bool proto\_down;  
1914 \};  
1915 \#define to\_net\_dev(d) container\_of(d, struct net\_device, dev)

## 3. 网络设备有关的结构组织 ##

net\_device结构包含了网络设备驱动相关的所有信息，按照信息的分类又把一些类型的信息组织到其他结构中，并嵌套在net\_device 里面，比如与ipv4相关的配置嵌套在 in\_device结构中，驱动的私有数据则嵌套在struct device中：

![20170716175316349][]  
网络设备是通过多条链表串连在一起的，具体怎么串连稍后再讲。我们前面看到了，每一个net\_device结构体都是由多个成员组成的，然而每个成员也有可能组成那么自己的链表，比如mc\_list和ip\_ptr，还有priv，虽然这个版本没有明确的定义priv这个指针，但是从alloc\_netdev函数可以知道仍然为它保留着，只要传进去的sizeof\_priv大于0.

下面我们看看其中一个很重要的成员ip\_ptr (struct in\_device \_\_rcu \*ip\_ptr)。它是一个头指针，指向struct in\_device对象，那它表示什么意思呢？我们知道，每一个网络设备都可以设置IP地址，而且这些参数也可以通过应用层进行修改，这些信息是每一个接口独有的，虽然并不是每一个都需要设置这些信息，但设置的时候，它的存放位置就是在in\_ptr指定链表里面。

下面我们来对比一下代码和实际输出：  
root:/\# ifconfig br-lan  
br-lan Link encap:Ethernet HWaddr 0A:02:8E:93:DD:3B  
 inet addr:192.168.1.129 Bcast:192.168.1.255 Mask:255.255.255.0  
inet6 addr: fe80::802:8eff:fe93:dd3b/64 Scope:Link  
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1  
RX packets:211672 errors:0 dropped:0 overruns:0 frame:0  
TX packets:120803 errors:0 dropped:0 overruns:0 carrier:0  
collisions:0 txqueuelen:0  
RX bytes:15794642 (15.0 MiB) TX bytes:24446287 (23.3 MiB)

23 struct in_device {
     24     struct net_device   *dev;  //它绕回去指向net_device结构体头部
     25     atomic_t        refcnt;   //这个对象被引用多少次
     26     int         dead;
     27     struct in_ifaddr    *ifa_list;  /* IP ifaddr chain      */
     28   为什么是链表里面？一个结构体对象不久够了吗？事实是一个接口往往不仅可以设置多个mac地址，当然也可以设置多个IP地址，最常见的是IPv4地址和IPv6地址。
     29     struct ip_mc_list __rcu *mc_list;   /* IP multicast filter chain    */
     30     struct ip_mc_list __rcu * __rcu *mc_hash;
     31
     32     int         mc_count;   /* Number of installed mcasts   */
     33     spinlock_t      mc_tomb_lock;
     34     struct ip_mc_list   *mc_tomb;
     35     unsigned long       mr_v1_seen;
     36     unsigned long       mr_v2_seen;
     37     unsigned long       mr_maxdelay;
     38     unsigned char       mr_qrv;
     39     unsigned char       mr_gq_running;
     40     unsigned char       mr_ifc_count;
     41     struct timer_list   mr_gq_timer;    /* general query timer */
     42     struct timer_list   mr_ifc_timer;   /* interface change timer */
     43
     44     struct neigh_parms  *arp_parms;
     45     struct ipv4_devconf cnf;
     46     struct rcu_head     rcu_head;
     47 };

下面这个图是关于ip\_ptr和priv两者的内存分配关系，这里需要注意的是，ip\_ptr指向的链表是内存随机分配空间的，但是priv则不一样，他的空间是紧紧接在net\_device结构体后面的！  
![20170716175320237][]  
1.设备无关层采用 in\_device\{\}数据结构保存 IP 地址和邻居信息——虽然是间接的  
2.网络抽象层采用 net\_device\{\}数据结构保存设备的名字、编号、地址等共性  
3.设备特定层的数据则有设备驱动开发人员自己定义，一般有硬件发送、接收缓冲区、芯片寄存器的信息等等。 这片内存区一般是紧跟在 net\_device\{\}后面，由驱动程序在创建 net\_device\{\}的时候顺带把这块内存也创建了。当然还是用 priv指针指向，以方便访问。  
虽然说priv指向的私有数据空间是紧接在net\_device后面，其实实际上更应该像这样添加了字节对齐：

![20170716182735364][]

为了更好的理解这一点，我们直接看代码：

7851 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
    7852         unsigned char name_assign_type,
    7853         void (*setup)(struct net_device *),
    7854         unsigned int txqs, unsigned int rxqs)
    7855 {
    7856     struct net_device *dev;
    7857     size_t alloc_size;
    7858     struct net_device *p;
    7859
             .......
    7873
    7874     alloc_size = sizeof(struct net_device);  //这里获取到net_device的大小
    7875     if (sizeof_priv) {   //看一下传进来的希望申请的私有空间大小是多少
    7876         /* ensure 32-byte alignment of private area */
    7877         alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);  //对齐
    7878         alloc_size += sizeof_priv;
    7879     }
    7880     /* ensure 32-byte alignment of whole construct */
    7881     alloc_size += NETDEV_ALIGN - 1; //32-1=31
    7882
    7883     p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_REPEAT); //这就是net_device和priv一起申请空间的地方

这样就组成了多个net\_device结构：  
![20170716183458267][]  
前面说过，net\_device是由多种链表串连在一起的，那么是由哪些链表呢？我们来看看：

![20170716183754536][]

从图中可以知道，一共有三个链表：

dev\_name\_head: 基于接口名字的查找, dev->name,对应的函数是dev\_get\_by\_name()

dev\_index\_head: 基于接口索引的查找，dev->ifindex， 对应的函数是dev\_get\_by\_index()

dev\_base: 基于其他参数的查找，比如设备类型，mac地址和标识等等

在了解了net\_device后，我们后面讲继续了解 设备驱动模块的加载，设备的注册和设备的启动

[20170716175316349]: /images/20220612/b671421b0d3542ad9833f6b1cb7c28eb.png
[20170716175320237]: /images/20220612/33938723d282426b9b47770903bff6bb.png
[20170716182735364]: /images/20220612/0620b2520b0046e29a17ae6a0bd98955.png
[20170716183458267]: /images/20220612/b2130ea05ad448eb8cf627ba2ab8a5f6.png
[20170716183754536]: /images/20220612/94833c6bebb44db28def53ec8d29f607.png