網(wǎng)卡收包

1. 內(nèi)核網(wǎng)絡(luò)模塊如何初始化？
2. 內(nèi)核如何通過(guò)網(wǎng)卡驅(qū)動(dòng)收發(fā)數(shù)據(jù)包？
3. 驅(qū)動(dòng)收到的數(shù)據(jù)怎么交給協(xié)議棧處理月培？

一，框架

網(wǎng)絡(luò)子系統(tǒng)中蒿往，在本文中我們關(guān)注的是驅(qū)動(dòng)和內(nèi)核的交互。也就是網(wǎng)卡收到數(shù)據(jù)包后怎么交給內(nèi)核湿弦，內(nèi)核收到數(shù)據(jù)包后怎么交給協(xié)議棧處理瓤漏。

在內(nèi)核中，網(wǎng)卡設(shè)備是被net_device結(jié)構(gòu)體描述的颊埃。驅(qū)動(dòng)需要通過(guò)net_device向內(nèi)核注冊(cè)一組操作網(wǎng)卡硬件的函數(shù)蔬充，這樣內(nèi)核便可以使用網(wǎng)卡了。而所有的數(shù)據(jù)包在內(nèi)核空間都是使用sk_buff結(jié)構(gòu)體來(lái)表示班利，所以將網(wǎng)卡硬件收到的數(shù)據(jù)轉(zhuǎn)換成內(nèi)核認(rèn)可的skb_buff也是驅(qū)動(dòng)的工作饥漫。

在這之后，還有兩個(gè)結(jié)構(gòu)體也發(fā)揮了非常重要的作用罗标。一個(gè)是為struct softnet_data庸队，另一個(gè)是struct napi_struct。為軟中斷的方式處理數(shù)據(jù)包提供了支持闯割。盜圖一張：

img

二彻消，初始化

一切的起源都是上電那一刻，當(dāng)系統(tǒng)初始化完畢后宙拉，我們的系統(tǒng)就應(yīng)該是可用的了宾尚。網(wǎng)絡(luò)子模塊的初始化也是在Linux啟動(dòng)經(jīng)歷兩階段的混沌boost自舉后，進(jìn)入的第一個(gè)C函數(shù)start_kernel谢澈。在這之前是Bootloader和Linux的故事煌贴，在這之后，便是Linux的單人秀了澳化。

網(wǎng)絡(luò)子設(shè)備初始化調(diào)用鏈：start_kernel->rest_init->kernel_init->kernel_init_freeable->do_basic_setup->do_initcalls->do_initcalls->net_dev_init。

上面調(diào)用關(guān)系中的kernel_init是一個(gè)內(nèi)核子線(xiàn)程中調(diào)用的：

pid = kernel_thread(kernel_init, NULL, CLONE_FS);

然后再一個(gè)問(wèn)題就是當(dāng)進(jìn)入do_initcalls后我們會(huì)發(fā)現(xiàn)畫(huà)風(fēng)突變：

static void __init do_initcalls(void)
{
    int level;

    for (level = 0; level < ARRAY_SIZE(initcall_levels) - 1; level++)
        do_initcall_level(level);
}

我是誰(shuí)稳吮，我來(lái)自哪缎谷，我要到哪去。

如果do_initcalls還給了我們一絲看下去的希望灶似，點(diǎn)開(kāi)do_initcall_level可能就真的絕望了列林。

static void __init do_initcall_level(int level)
{
    initcall_t *fn;

    strcpy(initcall_command_line, saved_command_line);
    parse_args(initcall_level_names[level],
           initcall_command_line, __start___param,
           __stop___param - __start___param,
           level, level,
           NULL, &repair_env_string);

    trace_initcall_level(initcall_level_names[level]);
    for (fn = initcall_levels[level]; fn < initcall_levels[level+1]; fn++)
        do_one_initcall(*fn);
}

全局一個(gè)fn指針，實(shí)現(xiàn)調(diào)用全靠猜酪惭。反正我不管希痴，我說(shuō)調(diào)用了net_dev_init就是調(diào)用了。偉大的google告訴我只要被下面這些宏定義包裹的函數(shù)就會(huì)被do_one_initcall調(diào)用春感，用了什么黑科技砌创，先不管：

#file：include/linux/init.h

#define pure_initcall(fn)       __define_initcall(fn, 0)

#define core_initcall(fn)       __define_initcall(fn, 1)
#define core_initcall_sync(fn)      __define_initcall(fn, 1s)
#define postcore_initcall(fn)       __define_initcall(fn, 2)
#define postcore_initcall_sync(fn)  __define_initcall(fn, 2s)
#define arch_initcall(fn)       __define_initcall(fn, 3)
#define arch_initcall_sync(fn)      __define_initcall(fn, 3s)
#define subsys_initcall(fn)     __define_initcall(fn, 4)
#define subsys_initcall_sync(fn)    __define_initcall(fn, 4s)
#define fs_initcall(fn)         __define_initcall(fn, 5)
#define fs_initcall_sync(fn)        __define_initcall(fn, 5s)
#define rootfs_initcall(fn)     __define_initcall(fn, rootfs)
#define device_initcall(fn)     __define_initcall(fn, 6)
#define device_initcall_sync(fn)    __define_initcall(fn, 6s)
#define late_initcall(fn)       __define_initcall(fn, 7)
#define late_initcall_sync(fn)      __define_initcall(fn, 7s)

在net_dev_init的定義下面虏缸，我們可以找到subsys_initcall(net_dev_init);。Ok嫩实，網(wǎng)絡(luò)子系統(tǒng)的初始化入口已找到到刽辙。

static int __init net_dev_init(void)
{
    int i, rc = -ENOMEM;

    BUG_ON(!dev_boot_phase);

    if (dev_proc_init())
        goto out;

    if (netdev_kobject_init())
        goto out;

    INIT_LIST_HEAD(&ptype_all);
    for (i = 0; i < PTYPE_HASH_SIZE; i++)
        INIT_LIST_HEAD(&ptype_base[i]);

    INIT_LIST_HEAD(&offload_base);

    if (register_pernet_subsys(&netdev_net_ops))
        goto out;

    /*
     *  Initialise the packet receive queues.
     */

    for_each_possible_cpu(i) {
        struct work_struct *flush = per_cpu_ptr(&flush_works, i);
        struct softnet_data *sd = &per_cpu(softnet_data, i);

        INIT_WORK(flush, flush_backlog);

        skb_queue_head_init(&sd->input_pkt_queue);
        skb_queue_head_init(&sd->process_queue);
#ifdef CONFIG_XFRM_OFFLOAD
        skb_queue_head_init(&sd->xfrm_backlog);
#endif
        INIT_LIST_HEAD(&sd->poll_list);
        sd->output_queue_tailp = &sd->output_queue;
#ifdef CONFIG_RPS
        sd->csd.func = rps_trigger_softirq;
        sd->csd.info = sd;
        sd->cpu = i;
#endif

        sd->backlog.poll = process_backlog;
        sd->backlog.weight = weight_p;
    }

    dev_boot_phase = 0;

    /* The loopback device is special if any other network devices
     * is present in a network namespace the loopback device must
     * be present. Since we now dynamically allocate and free the
     * loopback device ensure this invariant is maintained by
     * keeping the loopback device as the first device on the
     * list of network devices.  Ensuring the loopback devices
     * is the first device that appears and the last network device
     * that disappears.
     */
    if (register_pernet_device(&loopback_net_ops))
        goto out;

    if (register_pernet_device(&default_device_ops))
        goto out;

    open_softirq(NET_TX_SOFTIRQ, net_tx_action);
    open_softirq(NET_RX_SOFTIRQ, net_rx_action);

    rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
                       NULL, dev_cpu_dead);
    WARN_ON(rc < 0);
    rc = 0;
out:
    return rc;
}

在net_dev_init中，初始化了內(nèi)核收發(fā)包隊(duì)列甲献，開(kāi)啟了對(duì)應(yīng)的軟中斷NET_TX_SOFTIRQ和NET_RX_SOFTIRQ宰缤。在其中，該函數(shù)為每個(gè)CPU初始化了一個(gè)softnet_data來(lái)掛載需要處理設(shè)備的napi_struct晃洒。這個(gè)結(jié)構(gòu)非常重要慨灭，軟中斷的處理就是從這個(gè)鏈表上取napi_struct，然后收包的球及。這也是內(nèi)核和驅(qū)動(dòng)的接口之一氧骤。

再就是開(kāi)啟的兩個(gè)軟中斷，當(dāng)驅(qū)動(dòng)在硬終端完成必要的上半部工作后桶略，就會(huì)拉起對(duì)應(yīng)的軟中斷语淘。讓數(shù)據(jù)包下半部軟中斷中處理。

net_dev_init執(zhí)行完后际歼，我們內(nèi)核就有了處理數(shù)據(jù)包的能力惶翻，只要驅(qū)動(dòng)能向softnet_data掛載需要收包設(shè)備的napi_struct。內(nèi)核子線(xiàn)程ksoftirqd便會(huì)做后續(xù)的處理鹅心。接下來(lái)就是網(wǎng)卡驅(qū)動(dòng)的初始化了吕粗。

各種網(wǎng)卡肯定有不同的驅(qū)動(dòng)，各驅(qū)動(dòng)封裝各自硬件的差異旭愧，給內(nèi)核提供一個(gè)統(tǒng)一的接口颅筋。我們這不關(guān)心，網(wǎng)卡驅(qū)動(dòng)是怎么把數(shù)據(jù)發(fā)出去的输枯，如何收回來(lái)的议泵。而是探究網(wǎng)卡收到數(shù)據(jù)了，要怎么交給內(nèi)核桃熄，內(nèi)核如何將要發(fā)的數(shù)據(jù)給網(wǎng)卡先口。總之瞳收，驅(qū)動(dòng)需要給內(nèi)核提供哪些接口碉京，內(nèi)核又需要給網(wǎng)卡哪些支持。我們以e1000網(wǎng)卡為例子螟深⌒持妫看看它和內(nèi)核的纏綿故事。

e1000網(wǎng)卡是一塊PCI設(shè)備界弧。所以它首先得要讓內(nèi)核能通過(guò)PCI總線(xiàn)探測(cè)到凡蜻，需要向內(nèi)核注冊(cè)一個(gè)pci_driver結(jié)構(gòu)搭综，PCI設(shè)備的使用是另一個(gè)話(huà)題，這里不會(huì)探究咽瓷，我也不知道：

static struct pci_driver e1000_driver = {
    .name     = e1000_driver_name,
    .id_table = e1000_pci_tbl,
    .probe    = e1000_probe,
    .remove   = e1000_remove,
#ifdef CONFIG_PM
    /* Power Management Hooks */
    .suspend  = e1000_suspend,
    .resume   = e1000_resume,
#endif
    .shutdown = e1000_shutdown,
    .err_handler = &e1000_err_handler
};

其中e1000_probe就是給內(nèi)核的探測(cè)回調(diào)函數(shù)设凹，算是網(wǎng)卡的初始化函數(shù)吧，驅(qū)動(dòng)需要在這里初始化網(wǎng)卡設(shè)備茅姜。去掉總線(xiàn)相關(guān)的代碼闪朱，錯(cuò)誤處理的代碼，硬件相關(guān)的代碼：

static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
{
    struct net_device *netdev;
    netdev = alloc_etherdev(sizeof(struct e1000_adapter));//申請(qǐng)net_device設(shè)備
    netdev->netdev_ops = &e1000_netdev_ops; //注冊(cè)操作設(shè)備的回調(diào)函數(shù)
    e1000_set_ethtool_ops(netdev);
    netdev->watchdog_timeo = 5 * HZ;
    netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);//軟中斷里會(huì)調(diào)用poll鉤子函數(shù)
    strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
    err = register_netdev(netdev);
}

每一個(gè)網(wǎng)絡(luò)設(shè)備都有一個(gè)對(duì)應(yīng)的net_devie結(jié)構(gòu)體來(lái)描述钻洒。其中像設(shè)備文件操作一樣奋姿，保存了一種操作設(shè)備的接口函數(shù)netdev_ops，對(duì)e1000網(wǎng)卡是e1000_netdev_ops素标。當(dāng)通過(guò)終端輸入ifup称诗，ifdowm命令操作網(wǎng)卡時(shí)，對(duì)應(yīng)的open头遭，close函數(shù)就會(huì)被調(diào)用寓免。這段代碼最重要的還是netif_napi_add的調(diào)用，它向內(nèi)核注冊(cè)了e1000_clean函數(shù)计维，用來(lái)給上面的CPU收包隊(duì)列調(diào)用袜香。

通過(guò)初始化，驅(qū)動(dòng)注冊(cè)了網(wǎng)卡描述net_device, 內(nèi)核可以通過(guò)它操作到網(wǎng)卡設(shè)備鲫惶。通過(guò)e1000_clean函數(shù)內(nèi)核軟中斷也可以收包了蜈首。

三，驅(qū)動(dòng)收包

前面有一個(gè)內(nèi)核軟中斷來(lái)收包欠母，但這個(gè)軟中斷怎么觸發(fā)呢欢策？硬中斷。當(dāng)有數(shù)據(jù)到網(wǎng)卡時(shí)赏淌，會(huì)產(chǎn)生一個(gè)硬中斷踩寇。這中斷的注冊(cè)是上面，e1000_netdev_ops中的e1000_up函數(shù)調(diào)用的六水。也就是網(wǎng)卡up時(shí)會(huì)注冊(cè)這個(gè)硬中斷處理函數(shù)e1000_intr俺孙。

/**
 * e1000_intr - Interrupt Handler
 * @irq: interrupt number
 * @data: pointer to a network interface device structure
 **/
static irqreturn_t e1000_intr(int irq, void *data)
{
    struct net_device *netdev = data;
    struct e1000_adapter *adapter = netdev_priv(netdev);
    struct e1000_hw *hw = &adapter->hw;
    u32 icr = er32(ICR);

    /* disable interrupts, without the synchronize_irq bit */
    ew32(IMC, ~0);
    E1000_WRITE_FLUSH();

    if (likely(napi_schedule_prep(&adapter->napi))) {
        adapter->total_tx_bytes = 0;
        adapter->total_tx_packets = 0;
        adapter->total_rx_bytes = 0;
        adapter->total_rx_packets = 0;
        __napi_schedule(&adapter->napi);
    } else {
        /* this really should not happen! if it does it is basically a
         * bug, but not a hard error, so enable ints and continue
         */
        if (!test_bit(__E1000_DOWN, &adapter->flags))
            e1000_irq_enable(adapter);
    }

    return IRQ_HANDLED;
}

去掉unlikely的代碼，其中通過(guò)if (likely(napi_schedule_prep(&adapter->napi)))測(cè)試缩擂，網(wǎng)卡設(shè)備自己的napi是否正在被CPU使用鼠冕。沒(méi)有就調(diào)用__napi_schedule將自己的napi掛載到CPU的softnet_data上添寺。這樣軟中斷的內(nèi)核線(xiàn)程就能輪詢(xún)到這個(gè)軟中斷胯盯。

/**
 * __napi_schedule - schedule for receive
 * @n: entry to schedule
 *
 * The entry's receive function will be scheduled to run.
 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
 */
void __napi_schedule(struct napi_struct *n)
{
    unsigned long flags;

    local_irq_save(flags);
    ____napi_schedule(this_cpu_ptr(&softnet_data), n);
    local_irq_restore(flags);
}

/* Called with irq disabled */
static inline void ____napi_schedule(struct softnet_data *sd,
                     struct napi_struct *napi)
{
    list_add_tail(&napi->poll_list, &sd->poll_list);
    __raise_softirq_irqoff(NET_RX_SOFTIRQ); //設(shè)置軟中斷標(biāo)志位NET_RX_SOFTIRQ
}

這里的softnet_data就是前面net_dev_init函數(shù)為每個(gè)CPU初始化的。到這里硬件中斷就處理完了计露，但我們依然沒(méi)有發(fā)現(xiàn)任何有關(guān)數(shù)據(jù)包的處理博脑，只知道了有一個(gè)napi被掛載憎乙。這是因?yàn)橛布袛嗖荒茱@然太長(zhǎng)，的確不會(huì)去做數(shù)據(jù)的處理工作叉趣。這些都交給軟中斷的內(nèi)核線(xiàn)程來(lái)處理的泞边。

四，內(nèi)核處理

硬中斷將一個(gè)napi結(jié)構(gòu)體甩給了內(nèi)核疗杉，內(nèi)核要怎么根據(jù)它來(lái)接收數(shù)據(jù)呢阵谚？前面說(shuō)到，內(nèi)核為每個(gè)CPU核心都運(yùn)行了一個(gè)內(nèi)核線(xiàn)程ksoftirqd烟具。軟中斷也就是在這線(xiàn)程中處理的梢什。上面的硬件中斷函數(shù)設(shè)置了NET_RX_SOFTIRQ軟中斷標(biāo)志，這個(gè)字段處理函數(shù)還記得在哪注冊(cè)的么朝聋？是的嗡午，net_dev_init中。

    open_softirq(NET_TX_SOFTIRQ, net_tx_action);
    open_softirq(NET_RX_SOFTIRQ, net_rx_action);

顯然冀痕，后續(xù)處理肯定是由net_rx_action來(lái)完成荔睹。

static __latent_entropy void net_rx_action(struct softirq_action *h)
{
    struct softnet_data *sd = this_cpu_ptr(&softnet_data);
    unsigned long time_limit = jiffies +
        usecs_to_jiffies(netdev_budget_usecs);
    int budget = netdev_budget;
    LIST_HEAD(list);
    LIST_HEAD(repoll);

    local_irq_disable();
    list_splice_init(&sd->poll_list, &list);
    local_irq_enable();

    for (;;) {
        struct napi_struct *n;

        if (list_empty(&list)) {
            if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
                goto out;
            break;
        }

        n = list_first_entry(&list, struct napi_struct, poll_list);
        budget -= napi_poll(n, &repoll);    //在這回調(diào)驅(qū)動(dòng)的poll函數(shù)，這個(gè)函數(shù)在napi中

        /* If softirq window is exhausted then punt.
         * Allow this to run for 2 jiffies since which will allow
         * an average latency of 1.5/HZ.
         */
        if (unlikely(budget <= 0 ||
                 time_after_eq(jiffies, time_limit))) {
            sd->time_squeeze++;
            break;
        }
    }

    local_irq_disable();

    list_splice_tail_init(&sd->poll_list, &list);
    list_splice_tail(&repoll, &list);
    list_splice(&list, &sd->poll_list);
    if (!list_empty(&sd->poll_list))
        __raise_softirq_irqoff(NET_RX_SOFTIRQ);

    net_rps_action_and_irq_enable(sd);
out:
    __kfree_skb_flush();
}

上面看到budget -= napi_poll(n, &repoll);他會(huì)去調(diào)用我們驅(qū)動(dòng)初始化時(shí)注冊(cè)的poll函數(shù)言蛇，在e1000網(wǎng)卡中就是e1000_clean函數(shù)僻他。

/**
 * e1000_clean - NAPI Rx polling callback
 * @adapter: board private structure
 **/
static int e1000_clean(struct napi_struct *napi, int budget)
{
    struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter,
                             napi);
    int tx_clean_complete = 0, work_done = 0;

    tx_clean_complete = e1000_clean_tx_irq(adapter, &adapter->tx_ring[0]);

    adapter->clean_rx(adapter, &adapter->rx_ring[0], &work_done, budget);//將數(shù)據(jù)發(fā)給協(xié)議棧來(lái)處理。

    if (!tx_clean_complete)
        work_done = budget;

    /* If budget not fully consumed, exit the polling mode */
    if (work_done < budget) {
        if (likely(adapter->itr_setting & 3))
            e1000_set_itr(adapter);
        napi_complete_done(napi, work_done);
        if (!test_bit(__E1000_DOWN, &adapter->flags))
            e1000_irq_enable(adapter);
    }

    return work_done;
}

e1000_clean函數(shù)通過(guò)調(diào)用clean_rx函數(shù)指針來(lái)處理數(shù)據(jù)包猜极。

/**
 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
 * @adapter: board private structure
 * @rx_ring: ring to clean
 * @work_done: amount of napi work completed this call
 * @work_to_do: max amount of work allowed for this call to do
 *
 * the return value indicates whether actual cleaning was done, there
 * is no guarantee that everything was cleaned
 */
static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
                     struct e1000_rx_ring *rx_ring,
                     int *work_done, int work_to_do)
{
    struct net_device *netdev = adapter->netdev;
    struct pci_dev *pdev = adapter->pdev;
    struct e1000_rx_desc *rx_desc, *next_rxd;
    struct e1000_rx_buffer *buffer_info, *next_buffer;
    u32 length;
    unsigned int i;
    int cleaned_count = 0;
    bool cleaned = false;
    unsigned int total_rx_bytes = 0, total_rx_packets = 0;

    i = rx_ring->next_to_clean;
    rx_desc = E1000_RX_DESC(*rx_ring, i);
    buffer_info = &rx_ring->buffer_info[i];
    e1000_receive_skb(adapter, status, rx_desc->special, skb);
    napi_gro_frags(&adapter->napi);

    return cleaned;
}

/**
 * e1000_receive_skb - helper function to handle rx indications
 * @adapter: board private structure
 * @status: descriptor status field as written by hardware
 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
 * @skb: pointer to sk_buff to be indicated to stack
 */
static void e1000_receive_skb(struct e1000_adapter *adapter, u8 status,
                  __le16 vlan, struct sk_buff *skb)
{
    skb->protocol = eth_type_trans(skb, adapter->netdev);

    if (status & E1000_RXD_STAT_VP) {
        u16 vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;

        __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
    }
    napi_gro_receive(&adapter->napi, skb);
}

這個(gè)函數(shù)太長(zhǎng)中姜，我就保留了e1000_receive_skb函數(shù)的調(diào)用，它調(diào)用了napi_gro_receive跟伏，這個(gè)函數(shù)同樣是NAPI提供的函數(shù)丢胚，我們的skb從這里調(diào)用到netif_receive_skb協(xié)議棧的入口函數(shù)。調(diào)用路徑是napi_gro_receive->napi_frags_finish->netif_receive_skb_internal->__netif_receive_skb受扳。具體的流程先放放携龟。畢竟NAPI是內(nèi)核為了提高網(wǎng)卡收包性能而設(shè)計(jì)的一套框架。這就可以讓我先挖個(gè)坑以后在分析NAPI的時(shí)候在填上勘高∠矿總之有了NAPI后的收包流程和之前的區(qū)別如圖：

img

到這里，網(wǎng)卡驅(qū)動(dòng)到協(xié)議棧入口的處理過(guò)程就寫(xiě)完了华望。