1.什么是反向映射
是一种物理地址反向映射虚拟地址的方法;
正向映射:用户访问的虚拟地址,经过多级页表转化,最终映射到物理页面;
反向映射:根据物理页面,找到所有映射到这个页面的虚拟地址的过程;
2.ramp出现的背景
当物理内存短缺时: 虚拟内存常大于物理内存; 把暂时不用的物理内存swap到交换分区;
3. 反向映射的应用场景
3.1页面回收 同步内存回收:分配内存时,触发低水位; 异步内存回收:kswaped线程
3.2 页面迁移 3.3 KSM
4.RMAP相关的几个数据结构
4.1 mm_struct
mm_struct是进程task_struct中的一个成员,指向该进程的地址空间
struct mm_struct {
struct {
///进程里所有vma形成的一个单链表,mmap是表头
struct vm_area_struct *mmap; /* list of VMAs */
///vma红黑树的根节点
struct rb_root mm_rb;
#ifdef CONFIG_MMU
///判断虚拟内存空间是否有足够空间,返回一段没有映射过的虚拟空间起始地址
unsigned long (*get_unmapped_area) (struct file *filp,
unsigned long addr, unsigned long len,
unsigned long pgoff, unsigned long flags);
#endif
pgd_t * pgd; ///指向进程一级页表
atomic_t mm_count; ///mm_struct结构体的主引用计数
struct rw_semaphore mmap_lock; ///保护vma的读写信号量
///所有的mm_struct结构都连接到一个双向链表中,链表头是init_mm内存描述符
struct list_head mmlist; /* List of maybe swapped mm's. These
* are globally strung together off
* init_mm.mmlist, and are protected
* by mmlist_lock
*/
unsigned long start_code, end_code, start_data, end_data; ///代码段,数据段的起始地址和结束地址
unsigned long start_brk, brk, start_stack; ///start_brk:堆空间的起始地址,brk:当前堆中vma的结束地址
...
};
4.2 vm_area_struct
vm_area_struct是对一个内存段的抽象,简称vma,比如malloc分配一段内存,就会对应一个vma,一个代码段,数据段,都会对应一个vma;
VMA是linux管理内存的重要抽象,基本上所有的内存段都是通过VMA来描述的;
跟RMAP相关的,vma包含两个成员,anon_vma_chain, anon_vma;
struct vm_area_struct {
/* The first cache line has the info for VMA tree walking. */
///VMA在进程地址空间内的起始地址,结束地址
unsigned long vm_start; /* Our start address within vm_mm. */
unsigned long vm_end; /* The first byte after our end address within vm_mm. */
/* linked list of VM areas per task, sorted by address */
///进程的所有vma连接成一个链表
struct vm_area_struct *vm_next, *vm_prev;
///每个进程的mm_struct都有一个红黑树,VMA作为一个节点,加入该红黑树
struct rb_node vm_rb; ///
///指向vma所属进程的mm_struct
struct mm_struct *vm_mm; /* The address space we belong to. */
///vma的访问权限
pgprot_t vm_page_prot;
///描述该vma的一组标志位
unsigned long vm_flags; /* Flags, see mm.h. */
///指向avc
struct list_head anon_vma_chain; /* Serialized by mmap_lock & * page_table_lock */
///指向anon_vma
struct anon_vma *anon_vma; /* Serialized by page_table_lock */
/* Function pointers to deal with this struct. */
///指向操作方法集合,常用在文件映射
const struct vm_operations_struct *vm_ops;
/* Information about our backing store: */
///指定文件映射的偏移量,单位页
unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
units */
///指向映射的文件
struct file * vm_file; /* File we map to (can be NULL). */
...
} __randomize_layout;
task_struct, mm_struct, vma的关系图:
4.3 anon_vma
anon_vma,简单说,链接page和vma的桥梁,简称av;
其关系图:
/**************************************************
* func:链接物理页面的page结构和vma的vm_area_struct
*************************************************/
struct anon_vma {
///指向anon_vma结构的根节点
struct anon_vma *root; /* Root of this anon_vma tree */
///保护anon_vma数据结构的读写信号量
struct rw_semaphore rwsem; /* W: modification, R: walking the list */
///引用计数
atomic_t refcount;
///指向父anon_vma数据结构
struct anon_vma *parent; /* Parent of this anon_vma */
/* Interval tree of private "related" vmas */
///红黑树根节点,anon_vma内部有一颗红黑树
struct rb_root_cached rb_root;
...
};
4.4 anon_vma_chain
anon_vma_chain,链接vma和av的枢纽,简称avc;
avc作用有两个: (1)链接本进程的vma和av的枢纽; (2)链接本进程的vma和父系进程的av的枢纽;
/**************************************************
* func:链接枢纽
*************************************************/
struct anon_vma_chain {
///指向vma,可以指向父进程,也可以指向子进程
struct vm_area_struct *vma;
///指向anon_vma,可以指向父/子进程
struct anon_vma *anon_vma;
///把avc添加到vma的avc链表中
struct list_head same_vma; /* locked by mmap_lock & page_table_lock */
///把anon_vma添加到anon_vma的红黑树中
struct rb_node rb; /* locked by anon_vma->rwsem */
...
};
5. rmap历史
5.1 历史版本
a.原始内核版本Linux2.4
2.4版本,还没有RMAP机制,是如何解除映射关系呢? (1)从init_mm开始,遍历每一个进程; (2)遍历每一个进程所有的vma;
就这样简单粗暴,其性能明显是无法满足现代机器性能需求的,不深入研究;
b. Linux2.5版本引入rmap第一版:
在page数据结构中,增加一个链表指针,保存所有映射的pte;
这个方法,简单,但是会浪费大量内存;
c. Linux 2.6.11改进版的rmap
增加一个av数据结构,将page的mapping成员指向av; av的红黑树保存所有vma;
子进程的vma都添加到父进程的av链表中;
文件映射rmap:非常高效
匿名映射rmap:不高效,锁竞争激烈
该方法简单,高效;但面临的挑战和缺陷:
举个典型案例, 一个父进程fork了1000个子进程,每个子进程有1个vma,每个VMA里面有1000个匿名页面,当所有的子进程的VMA同时发生写时复制会是什么情况?
RMAP释放page核心函数:
try_to_unmap()
->rmap_walk()
->rmap_walk_anon()
rmap_walk_anon函数里需要获取avp->lock锁,由于有100w个页面共享了这个avp,锁竞争会非常激烈;
(1)很明显,锁的粒度太大了; (2)当子进程做rmap时,需要扫描所有vma,而链表上大部分vma并没有映射到这个页面上;且扫描过程是全程持有锁的,更加低效;
5.2 Linux2.6.32 成熟的ramp方案
增加一个avc,作为vma和av的枢纽;avc作用: (1)作为子进程vma和av的链接枢纽; (2)作为父系进程av和子进程vma链接枢纽; 按流程解析: (1)当一个进程建立好映射后,
(2)fork一个进程
(3)子进程在fork孙进程
源码解读:
fork()--->dup_mmap()->anon_vma_fork()
/***********************************************************************
* func:为子进程创建av数据结构,并构建av链接关系
* vma: 子进程vma
* pvma: 父进程的vma
***********************************************************************/
int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma)
{
struct anon_vma_chain *avc;
struct anon_vma *anon_vma;
int error;
/* Don't bother if the parent process has no anon_vma here. */
if (!pvma->anon_vma) ///若父进程没有av,就不需要绑定了
return 0;
/* Drop inherited anon_vma, we'll reuse existing or allocate new. */
vma->anon_vma = NULL;
/*
* First, attach the new VMA to the parent VMA's anon_vmas,
* so rmap can find non-COWed pages in child processes.
*/
error = anon_vma_clone(vma, pvma); ///把子进程的vma(通过avc)绑定到父进程vma的av链表中
if (error)
return error;
/* An existing anon_vma has been reused, all done then. */
if (vma->anon_vma) ///若子进程已经创建有anon_vma,说明绑定已完成
return 0;
/* Then add our own anon_vma. */
anon_vma = anon_vma_alloc(); ///分配子进程的av
if (!anon_vma)
goto out_error;
avc = anon_vma_chain_alloc(GFP_KERNEL); ///分配子进程的avc
if (!avc)
goto out_error_free_anon_vma;
/*
* The root anon_vma's rwsem is the lock actually used when we
* lock any of the anon_vmas in this anon_vma tree.
*/
anon_vma->root = pvma->anon_vma->root; ///子进程av的root,指向父进程av的root
anon_vma->parent = pvma->anon_vma; ///子进程av的parent,指向父进程的av
/*
* With refcounts, an anon_vma can stay around longer than the
* process it belongs to. The root anon_vma needs to be pinned until
* this anon_vma is freed, because the lock lives in the root.
*/
get_anon_vma(anon_vma->root); ///增加父进程的anon_vma的_refcount
/* Mark this anon_vma as the one where our new (COWed) pages go. */
vma->anon_vma = anon_vma;
anon_vma_lock_write(anon_vma);
anon_vma_chain_link(vma, avc, anon_vma); ///将子进程的avc分别添加到自己的av的rb, 和vma的avc链表中
anon_vma->parent->degree++;
anon_vma_unlock_write(anon_vma);
return 0;
out_error_free_anon_vma:
put_anon_vma(anon_vma);
out_error:
unlink_anon_vmas(vma);
return -ENOMEM;
}
int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src)
{
struct anon_vma_chain *avc, *pavc;
struct anon_vma *root = NULL;
///遍历父进程vma中的avc链表,寻找avc实例
list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) {
struct anon_vma *anon_vma;
avc = anon_vma_chain_alloc(GFP_NOWAIT | __GFP_NOWARN); ///分配一个新的avc,作为链接父子进程的枢纽
if (unlikely(!avc)) {
unlock_anon_vma_root(root);
root = NULL;
avc = anon_vma_chain_alloc(GFP_KERNEL);
if (!avc)
goto enomem_failure;
}
anon_vma = pavc->anon_vma;
root = lock_anon_vma_root(root, anon_vma);
anon_vma_chain_link(dst, avc, anon_vma); ///枢纽avc添加到父进程的rb,子进程的vma中avc链表中
/*
* Reuse existing anon_vma if its degree lower than two,
* that means it has no vma and only one anon_vma child.
*
* Do not chose parent anon_vma, otherwise first child
* will always reuse it. Root anon_vma is never reused:
* it has self-parent reference and at least one child.
*/
if (!dst->anon_vma && src->anon_vma &&
anon_vma != src->anon_vma && anon_vma->degree < 2)
dst->anon_vma = anon_vma;
}
if (dst->anon_vma)
dst->anon_vma->degree++;
unlock_anon_vma_root(root);
return 0;
enomem_failure:
/*
* dst->anon_vma is dropped here otherwise its degree can be incorrectly
* decremented in unlink_anon_vmas().
* We can safely do this because callers of anon_vma_clone() don't care
* about dst->anon_vma if anon_vma_clone() failed.
*/
dst->anon_vma = NULL;
unlink_anon_vmas(dst);
return -ENOMEM;
}
总结: (1)每个进程的vma中的av链表都保存所有子孙进程的avc;
(2)每个新建进程,(除创建正常的vma,av,avc外)要针对每一个父进程(包括祖父进程,...),创建一个avc_x保存在上一级进程的av红黑树中;
该avc_x也保存在本进程vma的avc链表中(供自己的子孙进程枚举,查找所有父系进程avc).
将兄弟进程分开了管理;这样,就将锁的竞争粒度减小了,只有同一个父进程的子孙才会竞争,兄弟进程隔离开了;
Linux 2.6.11版本的问题都得到解决: (1)只需要在子系进程中竞争锁,兄弟进程隔离开,锁粒度大大降低; (2)当发生写时复制时,新分配匿名页面cow_page->mapping指向子进程的AV结构,遍历时,不需要扫描所有子进程;
有两个问题: 1.当某个子进程触发缺页中断时,发生了什么? 2.当需要回收某个page时,做了哪些如何处理?
try_to_unmap()
->rmap_walk()
->rmap_walk_file
->vma_interval_tree_foreach()
rwc->rmap_one