memory.c

LINUX1.0源代码,代码条理清晰

		if (!(PAGE_PRESENT & *dir)) {
			/* clearing page here, needed?  SRB. */
			if (!(page_table = (unsigned long*) get_free_page(GFP_KERNEL))) {
				invalidate();
				return -1;
			}
			*dir++ = ((unsigned long) page_table) | PAGE_TABLE;
		}
		else
			page_table = (unsigned long *)(PAGE_MASK & *dir++);
		if (poff) {
			page_table += poff;
			poff = 0;
		}

		for (size -= pcnt; pcnt-- ;) {
			if ((page = *page_table) != 0) {
				*page_table = 0;
				if (PAGE_PRESENT & page) {
					if (!(mem_map[MAP_NR(page)] & MAP_PAGE_RESERVED))
						if (current->rss > 0)
							--current->rss;
					free_page(PAGE_MASK & page);
				} else
					swap_free(page);
			}

			/*
			 * the first condition should return an invalid access
			 * when the page is referenced. current assumptions
			 * cause it to be treated as demand allocation in some
			 * cases.
			 */
			if (!mask)
				*page_table++ = 0;	/* not present */
			else if (to >= high_memory)
				*page_table++ = (to | mask);
			else if (!mem_map[MAP_NR(to)])
				*page_table++ = 0;	/* not present */
			else {
				*page_table++ = (to | mask);
				if (!(mem_map[MAP_NR(to)] & MAP_PAGE_RESERVED)) {
					++current->rss;
					mem_map[MAP_NR(to)]++;
				}
			}
			to += PAGE_SIZE;
		}
		pcnt = (size > PTRS_PER_PAGE ? PTRS_PER_PAGE : size);
	}
	invalidate();
	return 0;
}

/*
 * This function puts a page in memory at the wanted address.
 * It returns the physical address of the page gotten, 0 if
 * out of memory (either when trying to access page-table or
 * page.)
 */
unsigned long put_page(struct task_struct * tsk,unsigned long page,
	unsigned long address,int prot)
{
	unsigned long *page_table;

	if ((prot & (PAGE_MASK|PAGE_PRESENT)) != PAGE_PRESENT)
		printk("put_page: prot = %08x\n",prot);
	if (page >= high_memory) {
		printk("put_page: trying to put page %08lx at %08lx\n",page,address);
		return 0;
	}
	page_table = PAGE_DIR_OFFSET(tsk->tss.cr3,address);
	if ((*page_table) & PAGE_PRESENT)
		page_table = (unsigned long *) (PAGE_MASK & *page_table);
	else {
		printk("put_page: bad page directory entry\n");
		oom(tsk);
		*page_table = BAD_PAGETABLE | PAGE_TABLE;
		return 0;
	}
	page_table += (address >> PAGE_SHIFT) & (PTRS_PER_PAGE-1);
	if (*page_table) {
		printk("put_page: page already exists\n");
		*page_table = 0;
		invalidate();
	}
	*page_table = page | prot;
/* no need for invalidate */
	return page;
}

/*
 * The previous function doesn't work very well if you also want to mark
 * the page dirty: exec.c wants this, as it has earlier changed the page,
 * and we want the dirty-status to be correct (for VM). Thus the same
 * routine, but this time we mark it dirty too.
 */
unsigned long put_dirty_page(struct task_struct * tsk, unsigned long page, unsigned long address)
{
	unsigned long tmp, *page_table;

	if (page >= high_memory)
		printk("put_dirty_page: trying to put page %08lx at %08lx\n",page,address);
	if (mem_map[MAP_NR(page)] != 1)
		printk("mem_map disagrees with %08lx at %08lx\n",page,address);
	page_table = PAGE_DIR_OFFSET(tsk->tss.cr3,address);
	if (PAGE_PRESENT & *page_table)
		page_table = (unsigned long *) (PAGE_MASK & *page_table);
	else {
		if (!(tmp = get_free_page(GFP_KERNEL)))
			return 0;
		if (PAGE_PRESENT & *page_table) {
			free_page(tmp);
			page_table = (unsigned long *) (PAGE_MASK & *page_table);
		} else {
			*page_table = tmp | PAGE_TABLE;
			page_table = (unsigned long *) tmp;
		}
	}
	page_table += (address >> PAGE_SHIFT) & (PTRS_PER_PAGE-1);
	if (*page_table) {
		printk("put_dirty_page: page already exists\n");
		*page_table = 0;
		invalidate();
	}
	*page_table = page | (PAGE_DIRTY | PAGE_PRIVATE);
/* no need for invalidate */
	return page;
}

/*
 * This routine handles present pages, when users try to write
 * to a shared page. It is done by copying the page to a new address
 * and decrementing the shared-page counter for the old page.
 *
 * Note that we do many checks twice (look at do_wp_page()), as
 * we have to be careful about race-conditions.
 *
 * Goto-purists beware: the only reason for goto's here is that it results
 * in better assembly code.. The "default" path will see no jumps at all.
 */
static void __do_wp_page(unsigned long error_code, unsigned long address,
	struct task_struct * tsk, unsigned long user_esp)
{
	unsigned long *pde, pte, old_page, prot;
	unsigned long new_page;

	new_page = __get_free_page(GFP_KERNEL);
	pde = PAGE_DIR_OFFSET(tsk->tss.cr3,address);
	pte = *pde;
	if (!(pte & PAGE_PRESENT))
		goto end_wp_page;
	if ((pte & PAGE_TABLE) != PAGE_TABLE || pte >= high_memory)
		goto bad_wp_pagetable;
	pte &= PAGE_MASK;
	pte += PAGE_PTR(address);
	old_page = *(unsigned long *) pte;
	if (!(old_page & PAGE_PRESENT))
		goto end_wp_page;
	if (old_page >= high_memory)
		goto bad_wp_page;
	if (old_page & PAGE_RW)
		goto end_wp_page;
	tsk->min_flt++;
	prot = (old_page & ~PAGE_MASK) | PAGE_RW;
	old_page &= PAGE_MASK;
	if (mem_map[MAP_NR(old_page)] != 1) {
		if (new_page) {
			if (mem_map[MAP_NR(old_page)] & MAP_PAGE_RESERVED)
				++tsk->rss;
			copy_page(old_page,new_page);
			*(unsigned long *) pte = new_page | prot;
			free_page(old_page);
			invalidate();
			return;
		}
		free_page(old_page);
		oom(tsk);
		*(unsigned long *) pte = BAD_PAGE | prot;
		invalidate();
		return;
	}
	*(unsigned long *) pte |= PAGE_RW;
	invalidate();
	if (new_page)
		free_page(new_page);
	return;
bad_wp_page:
	printk("do_wp_page: bogus page at address %08lx (%08lx)\n",address,old_page);
	*(unsigned long *) pte = BAD_PAGE | PAGE_SHARED;
	send_sig(SIGKILL, tsk, 1);
	goto end_wp_page;
bad_wp_pagetable:
	printk("do_wp_page: bogus page-table at address %08lx (%08lx)\n",address,pte);
	*pde = BAD_PAGETABLE | PAGE_TABLE;
	send_sig(SIGKILL, tsk, 1);
end_wp_page:
	if (new_page)
		free_page(new_page);
	return;
}

/*
 * check that a page table change is actually needed, and call
 * the low-level function only in that case..
 */
void do_wp_page(unsigned long error_code, unsigned long address,
	struct task_struct * tsk, unsigned long user_esp)
{
	unsigned long page;
	unsigned long * pg_table;

	pg_table = PAGE_DIR_OFFSET(tsk->tss.cr3,address);
	page = *pg_table;
	if (!page)
		return;
	if ((page & PAGE_PRESENT) && page < high_memory) {
		pg_table = (unsigned long *) ((page & PAGE_MASK) + PAGE_PTR(address));
		page = *pg_table;
		if (!(page & PAGE_PRESENT))
			return;
		if (page & PAGE_RW)
			return;
		if (!(page & PAGE_COW)) {
			if (user_esp && tsk == current) {
				current->tss.cr2 = address;
				current->tss.error_code = error_code;
				current->tss.trap_no = 14;
				send_sig(SIGSEGV, tsk, 1);
				return;
			}
		}
		if (mem_map[MAP_NR(page)] == 1) {
			*pg_table |= PAGE_RW | PAGE_DIRTY;
			invalidate();
			return;
		}
		__do_wp_page(error_code, address, tsk, user_esp);
		return;
	}
	printk("bad page directory entry %08lx\n",page);
	*pg_table = 0;
}

int __verify_write(unsigned long start, unsigned long size)
{
	size--;
	size += start & ~PAGE_MASK;
	size >>= PAGE_SHIFT;
	start &= PAGE_MASK;
	do {
		do_wp_page(1,start,current,0);
		start += PAGE_SIZE;
	} while (size--);
	return 0;
}

static inline void get_empty_page(struct task_struct * tsk, unsigned long address)
{
	unsigned long tmp;

	if (!(tmp = get_free_page(GFP_KERNEL))) {
		oom(tsk);
		tmp = BAD_PAGE;
	}
	if (!put_page(tsk,tmp,address,PAGE_PRIVATE))
		free_page(tmp);
}

/*
 * try_to_share() checks the page at address "address" in the task "p",
 * to see if it exists, and if it is clean. If so, share it with the current
 * task.
 *
 * NOTE! This assumes we have checked that p != current, and that they
 * share the same executable or library.
 *
 * We may want to fix this to allow page sharing for PIC pages at different
 * addresses so that ELF will really perform properly. As long as the vast
 * majority of sharable libraries load at fixed addresses this is not a
 * big concern. Any sharing of pages between the buffer cache and the
 * code space reduces the need for this as well.  - ERY
 */
static int try_to_share(unsigned long address, struct task_struct * tsk,
	struct task_struct * p, unsigned long error_code, unsigned long newpage)
{
	unsigned long from;
	unsigned long to;
	unsigned long from_page;
	unsigned long to_page;

	from_page = (unsigned long)PAGE_DIR_OFFSET(p->tss.cr3,address);
	to_page = (unsigned long)PAGE_DIR_OFFSET(tsk->tss.cr3,address);
/* is there a page-directory at from? */
	from = *(unsigned long *) from_page;
	if (!(from & PAGE_PRESENT))
		return 0;
	from &= PAGE_MASK;
	from_page = from + PAGE_PTR(address);
	from = *(unsigned long *) from_page;
/* is the page clean and present? */
	if ((from & (PAGE_PRESENT | PAGE_DIRTY)) != PAGE_PRESENT)
		return 0;
	if (from >= high_memory)
		return 0;
	if (mem_map[MAP_NR(from)] & MAP_PAGE_RESERVED)
		return 0;
/* is the destination ok? */
	to = *(unsigned long *) to_page;
	if (!(to & PAGE_PRESENT))
		return 0;
	to &= PAGE_MASK;
	to_page = to + PAGE_PTR(address);
	if (*(unsigned long *) to_page)
		return 0;
/* share them if read - do COW immediately otherwise */
	if (error_code & PAGE_RW) {
		if(!newpage)	/* did the page exist?  SRB. */
			return 0;
		copy_page((from & PAGE_MASK),newpage);
		to = newpage | PAGE_PRIVATE;
	} else {
		mem_map[MAP_NR(from)]++;
		from &= ~PAGE_RW;
		to = from;
		if(newpage)	/* only if it existed. SRB. */
			free_page(newpage);
	}
	*(unsigned long *) from_page = from;
	*(unsigned long *) to_page = to;
	invalidate();
	return 1;
}

/*
 * share_page() tries to find a process that could share a page with
 * the current one. Address is the address of the wanted page relative
 * to the current data space.
 *
 * We first check if it is at all feasible by checking executable->i_count.
 * It should be >1 if there are other tasks sharing this inode.
 */
int share_page(struct vm_area_struct * area, struct task_struct * tsk,
	struct inode * inode,
	unsigned long address, unsigned long error_code, unsigned long newpage)
{
	struct task_struct ** p;

	if (!inode || inode->i_count < 2 || !area->vm_ops)
		return 0;
	for (p = &LAST_TASK ; p > &FIRST_TASK ; --p) {
		if (!*p)
			continue;
		if (tsk == *p)
			continue;
		if (inode != (*p)->executable) {
			  if(!area) continue;
			/* Now see if there is something in the VMM that
			   we can share pages with */
			if(area){
			  struct vm_area_struct * mpnt;
			  for (mpnt = (*p)->mmap; mpnt; mpnt = mpnt->vm_next) {
			    if (mpnt->vm_ops == area->vm_ops &&
			       mpnt->vm_inode->i_ino == area->vm_inode->i_ino&&
			       mpnt->vm_inode->i_dev == area->vm_inode->i_dev){
			      if (mpnt->vm_ops->share(mpnt, area, address))
				break;
			    };
			  };
			  if (!mpnt) continue;  /* Nope.  Nuthin here */
			};
		}
		if (try_to_share(address,tsk,*p,error_code,newpage))
			return 1;
	}
	return 0;
}

/*
 * fill in an empty page-table if none exists.
 */
static inline unsigned long get_empty_pgtable(struct task_struct * tsk,unsigned long address)
{
	unsigned long page;
	unsigned long *p;

	p = PAGE_DIR_OFFSET(tsk->tss.cr3,address);
	if (PAGE_PRESENT & *p)
		return *p;
	if (*p) {
		printk("get_empty_pgtable: bad page-directory entry \n");
		*p = 0;
	}
	page = get_free_page(GFP_KERNEL);
	p = PAGE_DIR_OFFSET(tsk->tss.cr3,address);
	if (PAGE_PRESENT & *p) {
		free_page(page);
		return *p;
	}
	if (*p) {
		printk("get_empty_pgtable: bad page-directory entry \n");
		*p = 0;
	}
	if (page) {
		*p = page | PAGE_TABLE;
		return *p;
	}
	oom(current);
	*p = BAD_PAGETABLE | PAGE_TABLE;