Linux 文件系统注册过程

1. Linux文件系统的注册包括多种，可以将文件系统编译进内核，在系统启动时注册，也可以将文件系统编译为模块，在加载时注册。以编译进内核为例子，在init/main.c函数中:

asmlinkage void __init start_kernel(void)
{
    char * command_line;
    extern struct kernel_param __start___param[], __stop___param[];

    smp_setup_processor_id();

    /*
     * Need to run as early as possible, to initialize the
     * lockdep hash:
     */
    unwind_init();
    lockdep_init();

    local_irq_disable();
    early_boot_irqs_off();
    early_init_irq_lock_class();

/*
 * Interrupts are still disabled. Do necessary setups, then
 * enable them
 */
    lock_kernel();
    boot_cpu_init();
    page_address_init();
    printk(KERN_NOTICE);
    printk(linux_banner);
    setup_arch(&command_line);
    unwind_setup();
    setup_per_cpu_areas();
    smp_prepare_boot_cpu();    /* arch-specific boot-cpu hooks */

    /*
     * Set up the scheduler prior starting any interrupts (such as the
     * timer interrupt). Full topology setup happens at smp_init()
     * time - but meanwhile we still have a functioning scheduler.
     */
    sched_init();
    /*
     * Disable preemption - early bootup scheduling is extremely
     * fragile until we cpu_idle() for the first time.
     */
    preempt_disable();
    build_all_zonelists();
    page_alloc_init();
    printk(KERN_NOTICE "Kernel command line: %s\n", saved_command_line);
    parse_early_param();
    parse_args("Booting kernel", command_line, __start___param,
           __stop___param - __start___param,
           &unknown_bootoption);
    if (!irqs_disabled()) {
        printk(KERN_WARNING "start_kernel(): bug: interrupts were "
                "enabled *very* early, fixing it\n");
        local_irq_disable();
    }
    sort_main_extable();
    trap_init();
    rcu_init();
    init_IRQ();
    pidhash_init();
    init_timers();
    hrtimers_init();
    softirq_init();
    timekeeping_init();
    time_init();
    profile_init();
    if (!irqs_disabled())
        printk("start_kernel(): bug: interrupts were enabled early\n");
    early_boot_irqs_on();
    local_irq_enable();

    /*
     * HACK ALERT! This is early. We're enabling the console before
     * we've done PCI setups etc, and console_init() must be aware of
     * this. But we do want output early, in case something goes wrong.
     */
    console_init();
    if (panic_later)
        panic(panic_later, panic_param);

    lockdep_info();

    /*
     * Need to run this when irqs are enabled, because it wants
     * to self-test [hard/soft]-irqs on/off lock inversion bugs
     * too:
     */
    locking_selftest();

#ifdef CONFIG_BLK_DEV_INITRD
    if (initrd_start && !initrd_below_start_ok &&
            initrd_start < min_low_pfn << PAGE_SHIFT) {
        printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
            "disabling it.\n",initrd_start,min_low_pfn << PAGE_SHIFT);
        initrd_start = 0;
    }
#endif
    vfs_caches_init_early();
    cpuset_init_early();
    mem_init();
    kmem_cache_init();
    setup_per_cpu_pageset();
    numa_policy_init();
    if (late_time_init)
        late_time_init();
    calibrate_delay();
    pidmap_init();
    pgtable_cache_init();
    prio_tree_init();
    anon_vma_init();
#ifdef CONFIG_X86
    if (efi_enabled)
        efi_enter_virtual_mode();
#endif
    fork_init(num_physpages);
    proc_caches_init();
    buffer_init();
    unnamed_dev_init();
    key_init();
    security_init();
    vfs_caches_init(num_physpages);
    radix_tree_init();
    signals_init();
    /* rootfs populating might need page-writeback */
    page_writeback_init();
#ifdef CONFIG_PROC_FS
    proc_root_init();
#endif
    cpuset_init();
    taskstats_init_early();
    delayacct_init();

    check_bugs();

    acpi_early_init(); /* before LAPIC and SMP init */

    /* Do the rest non-__init'ed, we're now alive */
    rest_init();
}

这段代码主要进行内核刚启动的初始化操作，找到vfs_cache_init函数，接下来:

void __init vfs_caches_init(unsigned long mempages)
{
    unsigned long reserve;

    /* Base hash sizes on available memory, with a reserve equal to
           150% of current kernel size */

    reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
    mempages -= reserve;

    names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
            SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);

    filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
            SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);

    dcache_init(mempages);
    inode_init(mempages);
    files_init(mempages);
    mnt_init(mempages);
    bdev_cache_init();
    chrdev_init();
}

重点关注一下mnt_init函数，这个函数会调用init_rootfs函数，在这个函数中我们可以看到:

int __init init_rootfs(void)
{
    return register_filesystem(&rootfs_fs_type);
}

最终会调用register_filesystem, 传入参数为根文件系统类型，具体关注一下此函数:

/**将所有的文件系统类型组织在file_systesm链表，如ext2,ext3,gfs
 *    register_filesystem - register a new filesystem
 *    @fs: the file system structure
 *
 *    Adds the file system passed to the list of file systems the kernel
 *    is aware of for mount and other syscalls. Returns 0 on success,
 *    or a negative errno code on an error.
 *
 *    The &struct file_system_type that is passed is linked into the kernel
 *    structures and must not be freed until the file system has been
 *    unregistered.
 */
 
int register_filesystem(struct file_system_type * fs)
{
    int res = 0;
    struct file_system_type ** p;
    if (fs->next)
        return -EBUSY;
    INIT_LIST_HEAD(&fs->fs_supers);
    write_lock(&file_systems_lock);
    p = find_filesystem(fs->name);
    if (*p)
        res = -EBUSY;
    else
        *p = fs;
    write_unlock(&file_systems_lock);
    return res;
}

EXPORT_SYMBOL(register_filesystem);

第一步: 当fs->next不空，说明该文件系统已经注册了，那么就不能再注册了。

第二步: 初始化fs->fs_supers对象，指向同类型文件系统的链表头。

接下来，调用find_filesystem函数:

static struct file_system_type **find_filesystem(const char *name)
{
    struct file_system_type **p;
    for (p=&file_systems; *p; p=&(*p)->next)
        if (strcmp((*p)->name,name) == 0)
            break;
    return p;
}

file_systems存储了已注册所有的文件系统类型，查找此链表，如果存在，就不需要重新注册，否则，在链表尾加入新类型的文件系统。

简单的说，注册一个新的文件系统类型，就是将文件系统类型加入到file_systems链表尾。