Linux内存管理 (12)反向映射RMAP

staticvoidanon_vma_chain_link(structvm_area_struct*vma,
structanon_vma_chain*avc,
structanon_vma*anon_vma)
{
avc->vma=vma;--------------------------------------------建立structanon_vma_chain和structvm_area_struct关联
avc->anon_vma=anon_vma;----------------------------------建立structanon_vma_chain和structanon_vma关联
list_add(&avc->same_vma,&vma->anon_vma_chain);------------将AVC添加到structvm_area_struct->anon_vma_chain链表中。
anon_vma_interval_tree_insert(avc,&anon_vma->rb_root);----将AVC添加到structanon_vma->rb_root红黑树中。
}

voidpage_add_new_anon_rmap(structpage*page,
structvm_area_struct*vma,unsignedlongaddress)
{
VM_BUG_ON_VMA(address<vma->vm_start||address>=vma->vm_end,vma);
SetPageSwapBacked(page);----------------------------------------------------------设置PG_SwapBacked表示这个页面可以swap到磁盘。
atomic_set(&page->_mapcount,0);/*incrementcount(startsat-1)*/-------------设置_mapcount引用计数为0
if(PageTransHuge(page))
__inc_zone_page_state(page,NR_ANON_TRANSPARENT_HUGEPAGES);
__mod_zone_page_state(page_zone(page),NR_ANON_PAGES,-----------------------------增加页面所在zone的匿名页面计数
hpage_nr_pages(page));
__page_set_anon_rmap(page,vma,address,1);--------------------------------------设置这个页面位匿名映射
}

staticvoid__page_set_anon_rmap(structpage*page,
structvm_area_struct*vma,unsignedlongaddress,intexclusive)
{
structanon_vma*anon_vma=vma->anon_vma;

BUG_ON(!anon_vma);

if(PageAnon(page))---------------------------------------------------------------判断当前页面是否是匿名页面PAGE_MAPPING_ANON
return;

/*
*Ifthepageisn'texclusivelymappedintothisvma,
*wemustusethe_oldest_possibleanon_vmaforthe
*pagemapping!
*/
if(!exclusive)
anon_vma=anon_vma->root;

anon_vma=(void*)anon_vma+PAGE_MAPPING_ANON;
page->mapping=(structaddress_space*)anon_vma;------------------------------mapping指定页面所在的地址空间，这里指向匿名页面的地址空间数据结构structanon_vma。
page->index=linear_page_index(vma,address);
}

2.父进程创建子进程

staticintdup_mmap(structmm_struct*mm,structmm_struct*oldmm)
{
structvm_area_struct*mpnt,*tmp,*prev,**pprev;
structrb_node**rb_link,*rb_parent;
intretval;
unsignedlongcharge;

uprobe_start_dup_mmap();
down_write(&oldmm->mmap_sem);
flush_cache_dup_mm(oldmm);
uprobe_dup_mmap(oldmm,mm);
/*
*Notlinkedinyet-nodeadlockpotential:
*/
down_write_nested(&mm->mmap_sem,SINGLE_DEPTH_NESTING);

mm->total_vm=oldmm->total_vm;
mm->shared_vm=oldmm->shared_vm;
mm->exec_vm=oldmm->exec_vm;
mm->stack_vm=oldmm->stack_vm;

rb_link=&mm->mm_rb.rb_node;
rb_parent=NULL;
pprev=&mm->mmap;
retval=ksm_fork(mm,oldmm);
if(retval)
gotoout;
retval=khugepaged_fork(mm,oldmm);
if(retval)
gotoout;

prev=NULL;
for(mpnt=oldmm->mmap;mpnt;mpnt=mpnt->vm_next){-------------------for循环遍历父进程的进程地址空间VMA。
structfile*file;

if(mpnt->vm_flags&VM_DONTCOPY){
vm_stat_account(mm,mpnt->vm_flags,mpnt->vm_file,
-vma_pages(mpnt));
continue;
}
charge=0;
if(mpnt->vm_flags&VM_ACCOUNT){
unsignedlonglen=vma_pages(mpnt);

if(security_vm_enough_memory_mm(oldmm,len))/*sic*/
gotofail_nomem;
charge=len;
}
tmp=kmem_cache_alloc(vm_area_cachep,GFP_KERNEL);
if(!tmp)
gotofail_nomem;
*tmp=*mpnt;--------------------------------------------------------复制父进程地址空间VMA到刚创建的子进程tmp中。
INIT_LIST_HEAD(&tmp->anon_vma_chain);
retval=vma_dup_policy(mpnt,tmp);
if(retval)
gotofail_nomem_policy;
tmp->vm_mm=mm;
if(anon_vma_fork(tmp,mpnt))----------------------------------------为子进程创建相应的anon_vma数据结构。
gotofail_nomem_anon_vma_fork;
tmp->vm_flags&=~VM_LOCKED;
tmp->vm_next=tmp->vm_prev=NULL;
file=tmp->vm_file;
...
__vma_link_rb(mm,tmp,rb_link,rb_parent);--------------------------把VMA添加到子进程红黑树中。
rb_link=&tmp->vm_rb.rb_right;
rb_parent=&tmp->vm_rb;

mm->map_count++;
retval=copy_page_range(mm,oldmm,mpnt);---------------------------复制父进程的PTE页表项到子进程页表中。

if(tmp->vm_ops&&tmp->vm_ops->open)
tmp->vm_ops->open(tmp);

if(retval)
gotoout;
}
...
}

3.子进程发生COW

->发生缺页中断
->handle_pte_fault
->do_wp_page
->分配一个新的匿名页面
->__page_set_anon_rmap使用子进程的anon_vma来设置page->mapping

4.RMAP应用

/**
*try_to_unmap-trytoremoveallpagetablemappingstoapage
*@page:thepagetogetunmapped
*@flags:actionandflags
*
*Triestoremoveallthepagetableentrieswhicharemappingthis
*page,usedinthepageoutpath.Callermustholdthepagelock.
*Returnvaluesare:
*
*SWAP_SUCCESS-wesucceededinremovingallmappings------------成功解除了所有映射的PTE。
*SWAP_AGAIN-wemissedamapping,tryagainlater---------------可能错过了一个映射的PTE，需要重来一次。
*SWAP_FAIL-thepageisunswappable-----------------------------失败
*SWAP_MLOCK-pageismlocked.-----------------------------------页面被锁住了
*/
inttry_to_unmap(structpage*page,enumttu_flagsflags)
{
intret;
structrmap_walk_controlrwc={
.rmap_one=try_to_unmap_one,--------------------------------具体断开某个VMA上映射的pte
.arg=(void*)flags,
.done=page_not_mapped,-------------------------------------判断一个页面是否断开成功的条件
.anon_lock=page_lock_anon_vma_read,------------------------锁
};

VM_BUG_ON_PAGE(!PageHuge(page)&&PageTransHuge(page),page);

/*
*Duringexec,atemporaryVMAissetupandlatermoved.
*TheVMAismovedundertheanon_vmalockbutnotthe
*pagetablesleadingtoaracewheremigrationcannot
*findthemigrationptes.Ratherthanincreasingthe
*lockingrequirementsofexec(),migrationskips
*temporaryVMAsuntilafterexec()completes.
*/
if((flags&TTU_MIGRATION)&&!PageKsm(page)&&PageAnon(page))
rwc.invalid_vma=invalid_migration_vma;

ret=rmap_walk(page,&rwc);

if(ret!=SWAP_MLOCK&&!page_mapped(page))
ret=SWAP_SUCCESS;
returnret;
}

intrmap_walk(structpage*page,structrmap_walk_control*rwc)
{
if(unlikely(PageKsm(page)))
returnrmap_walk_ksm(page,rwc);
elseif(PageAnon(page))
returnrmap_walk_anon(page,rwc);
else
returnrmap_walk_file(page,rwc);
}

staticintrmap_walk_anon(structpage*page,structrmap_walk_control*rwc)
{
structanon_vma*anon_vma;
pgoff_tpgoff;
structanon_vma_chain*avc;
intret=SWAP_AGAIN;

anon_vma=rmap_walk_anon_lock(page,rwc);-----------------------------------获取页面page->mapping指向的anon_vma数据结构，并申请一个读者锁。
if(!anon_vma)
returnret;

pgoff=page_to_pgoff(page);
anon_vma_interval_tree_foreach(avc,&anon_vma->rb_root,pgoff,pgoff){------遍历anon_vma->rb_root红黑树中的AVC，从AVC得到相应的VMA。
structvm_area_struct*vma=avc->vma;
unsignedlongaddress=vma_address(page,vma);

if(rwc->invalid_vma&&rwc->invalid_vma(vma,rwc->arg))
continue;

ret=rwc->rmap_one(page,vma,address,rwc->arg);-----------------------实际的断开用户PTE页表项操作。
if(ret!=SWAP_AGAIN)
break;
if(rwc->done&&rwc->done(page))
break;
}
anon_vma_unlock_read(anon_vma);
returnret;
}