2 * mm/rmap.c - physical to virtual reverse mappings
4 * Copyright 2001, Rik van Riel <riel@conectiva.com.br>
5 * Released under the General Public License (GPL).
7 * Simple, low overhead reverse mapping scheme.
8 * Please try to keep this thing as modular as possible.
10 * Provides methods for unmapping each kind of mapped page:
11 * the anon methods track anonymous pages, and
12 * the file methods track pages belonging to an inode.
14 * Original design by Rik van Riel <riel@conectiva.com.br> 2001
15 * File methods by Dave McCracken <dmccr@us.ibm.com> 2003, 2004
16 * Anonymous methods by Andrea Arcangeli <andrea@suse.de> 2004
17 * Contributions by Hugh Dickins <hugh@veritas.com> 2003, 2004
21 * Lock ordering in mm:
23 * inode->i_mutex (while writing or truncating, not reading or faulting)
24 * inode->i_alloc_sem (vmtruncate_range)
26 * page->flags PG_locked (lock_page)
27 * mapping->i_mmap_lock
29 * mm->page_table_lock or pte_lock
30 * zone->lru_lock (in mark_page_accessed, isolate_lru_page)
31 * swap_lock (in swap_duplicate, swap_info_get)
32 * mmlist_lock (in mmput, drain_mmlist and others)
33 * mapping->private_lock (in __set_page_dirty_buffers)
34 * inode_lock (in set_page_dirty's __mark_inode_dirty)
35 * sb_lock (within inode_lock in fs/fs-writeback.c)
36 * mapping->tree_lock (widely used, in set_page_dirty,
37 * in arch-dependent flush_dcache_mmap_lock,
38 * within inode_lock in __sync_single_inode)
39 * zone->lock (within radix tree node alloc)
43 #include <linux/pagemap.h>
44 #include <linux/swap.h>
45 #include <linux/swapops.h>
46 #include <linux/slab.h>
47 #include <linux/init.h>
48 #include <linux/rmap.h>
49 #include <linux/rcupdate.h>
50 #include <linux/module.h>
51 #include <linux/kallsyms.h>
53 #include <asm/tlbflush.h>
55 struct kmem_cache *anon_vma_cachep;
57 /* This must be called under the mmap_sem. */
58 int anon_vma_prepare(struct vm_area_struct *vma)
60 struct anon_vma *anon_vma = vma->anon_vma;
63 if (unlikely(!anon_vma)) {
64 struct mm_struct *mm = vma->vm_mm;
65 struct anon_vma *allocated, *locked;
67 anon_vma = find_mergeable_anon_vma(vma);
71 spin_lock(&locked->lock);
73 anon_vma = anon_vma_alloc();
74 if (unlikely(!anon_vma))
80 /* page_table_lock to protect against threads */
81 spin_lock(&mm->page_table_lock);
82 if (likely(!vma->anon_vma)) {
83 vma->anon_vma = anon_vma;
84 list_add_tail(&vma->anon_vma_node, &anon_vma->head);
87 spin_unlock(&mm->page_table_lock);
90 spin_unlock(&locked->lock);
91 if (unlikely(allocated))
92 anon_vma_free(allocated);
97 void __anon_vma_merge(struct vm_area_struct *vma, struct vm_area_struct *next)
99 BUG_ON(vma->anon_vma != next->anon_vma);
100 list_del(&next->anon_vma_node);
103 void __anon_vma_link(struct vm_area_struct *vma)
105 struct anon_vma *anon_vma = vma->anon_vma;
108 list_add_tail(&vma->anon_vma_node, &anon_vma->head);
111 void anon_vma_link(struct vm_area_struct *vma)
113 struct anon_vma *anon_vma = vma->anon_vma;
116 spin_lock(&anon_vma->lock);
117 list_add_tail(&vma->anon_vma_node, &anon_vma->head);
118 spin_unlock(&anon_vma->lock);
122 void anon_vma_unlink(struct vm_area_struct *vma)
124 struct anon_vma *anon_vma = vma->anon_vma;
130 spin_lock(&anon_vma->lock);
131 list_del(&vma->anon_vma_node);
133 /* We must garbage collect the anon_vma if it's empty */
134 empty = list_empty(&anon_vma->head);
135 spin_unlock(&anon_vma->lock);
138 anon_vma_free(anon_vma);
141 static void anon_vma_ctor(struct kmem_cache *cachep, void *data)
143 struct anon_vma *anon_vma = data;
145 spin_lock_init(&anon_vma->lock);
146 INIT_LIST_HEAD(&anon_vma->head);
149 void __init anon_vma_init(void)
151 anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
152 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor);
156 * Getting a lock on a stable anon_vma from a page off the LRU is
157 * tricky: page_lock_anon_vma rely on RCU to guard against the races.
159 static struct anon_vma *page_lock_anon_vma(struct page *page)
161 struct anon_vma *anon_vma;
162 unsigned long anon_mapping;
165 anon_mapping = (unsigned long) page->mapping;
166 if (!(anon_mapping & PAGE_MAPPING_ANON))
168 if (!page_mapped(page))
171 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
172 spin_lock(&anon_vma->lock);
179 static void page_unlock_anon_vma(struct anon_vma *anon_vma)
181 spin_unlock(&anon_vma->lock);
186 * At what user virtual address is page expected in vma?
188 static inline unsigned long
189 vma_address(struct page *page, struct vm_area_struct *vma)
191 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
192 unsigned long address;
194 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
195 if (unlikely(address < vma->vm_start || address >= vma->vm_end)) {
196 /* page should be within any vma from prio_tree_next */
197 BUG_ON(!PageAnon(page));
204 * At what user virtual address is page expected in vma? checking that the
205 * page matches the vma: currently only used on anon pages, by unuse_vma;
207 unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
209 if (PageAnon(page)) {
210 if ((void *)vma->anon_vma !=
211 (void *)page->mapping - PAGE_MAPPING_ANON)
213 } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) {
215 vma->vm_file->f_mapping != page->mapping)
219 return vma_address(page, vma);
223 * Check that @page is mapped at @address into @mm.
225 * On success returns with pte mapped and locked.
227 pte_t *page_check_address(struct page *page, struct mm_struct *mm,
228 unsigned long address, spinlock_t **ptlp)
236 pgd = pgd_offset(mm, address);
237 if (!pgd_present(*pgd))
240 pud = pud_offset(pgd, address);
241 if (!pud_present(*pud))
244 pmd = pmd_offset(pud, address);
245 if (!pmd_present(*pmd))
248 pte = pte_offset_map(pmd, address);
249 /* Make a quick check before getting the lock */
250 if (!pte_present(*pte)) {
255 ptl = pte_lockptr(mm, pmd);
257 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
261 pte_unmap_unlock(pte, ptl);
266 * Subfunctions of page_referenced: page_referenced_one called
267 * repeatedly from either page_referenced_anon or page_referenced_file.
269 static int page_referenced_one(struct page *page,
270 struct vm_area_struct *vma, unsigned int *mapcount)
272 struct mm_struct *mm = vma->vm_mm;
273 unsigned long address;
278 address = vma_address(page, vma);
279 if (address == -EFAULT)
282 pte = page_check_address(page, mm, address, &ptl);
286 if (ptep_clear_flush_young(vma, address, pte))
289 /* Pretend the page is referenced if the task has the
290 swap token and is in the middle of a page fault. */
291 if (mm != current->mm && has_swap_token(mm) &&
292 rwsem_is_locked(&mm->mmap_sem))
296 pte_unmap_unlock(pte, ptl);
301 static int page_referenced_anon(struct page *page)
303 unsigned int mapcount;
304 struct anon_vma *anon_vma;
305 struct vm_area_struct *vma;
308 anon_vma = page_lock_anon_vma(page);
312 mapcount = page_mapcount(page);
313 list_for_each_entry(vma, &anon_vma->head, anon_vma_node) {
314 referenced += page_referenced_one(page, vma, &mapcount);
319 page_unlock_anon_vma(anon_vma);
324 * page_referenced_file - referenced check for object-based rmap
325 * @page: the page we're checking references on.
327 * For an object-based mapped page, find all the places it is mapped and
328 * check/clear the referenced flag. This is done by following the page->mapping
329 * pointer, then walking the chain of vmas it holds. It returns the number
330 * of references it found.
332 * This function is only called from page_referenced for object-based pages.
334 static int page_referenced_file(struct page *page)
336 unsigned int mapcount;
337 struct address_space *mapping = page->mapping;
338 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
339 struct vm_area_struct *vma;
340 struct prio_tree_iter iter;
344 * The caller's checks on page->mapping and !PageAnon have made
345 * sure that this is a file page: the check for page->mapping
346 * excludes the case just before it gets set on an anon page.
348 BUG_ON(PageAnon(page));
351 * The page lock not only makes sure that page->mapping cannot
352 * suddenly be NULLified by truncation, it makes sure that the
353 * structure at mapping cannot be freed and reused yet,
354 * so we can safely take mapping->i_mmap_lock.
356 BUG_ON(!PageLocked(page));
358 spin_lock(&mapping->i_mmap_lock);
361 * i_mmap_lock does not stabilize mapcount at all, but mapcount
362 * is more likely to be accurate if we note it after spinning.
364 mapcount = page_mapcount(page);
366 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
367 if ((vma->vm_flags & (VM_LOCKED|VM_MAYSHARE))
368 == (VM_LOCKED|VM_MAYSHARE)) {
372 referenced += page_referenced_one(page, vma, &mapcount);
377 spin_unlock(&mapping->i_mmap_lock);
382 * page_referenced - test if the page was referenced
383 * @page: the page to test
384 * @is_locked: caller holds lock on the page
386 * Quick test_and_clear_referenced for all mappings to a page,
387 * returns the number of ptes which referenced the page.
389 int page_referenced(struct page *page, int is_locked)
393 if (page_test_and_clear_young(page))
396 if (TestClearPageReferenced(page))
399 if (page_mapped(page) && page->mapping) {
401 referenced += page_referenced_anon(page);
403 referenced += page_referenced_file(page);
404 else if (TestSetPageLocked(page))
408 referenced += page_referenced_file(page);
415 static int page_mkclean_one(struct page *page, struct vm_area_struct *vma)
417 struct mm_struct *mm = vma->vm_mm;
418 unsigned long address;
423 address = vma_address(page, vma);
424 if (address == -EFAULT)
427 pte = page_check_address(page, mm, address, &ptl);
431 if (pte_dirty(*pte) || pte_write(*pte)) {
434 flush_cache_page(vma, address, pte_pfn(*pte));
435 entry = ptep_clear_flush(vma, address, pte);
436 entry = pte_wrprotect(entry);
437 entry = pte_mkclean(entry);
438 set_pte_at(mm, address, pte, entry);
442 pte_unmap_unlock(pte, ptl);
447 static int page_mkclean_file(struct address_space *mapping, struct page *page)
449 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
450 struct vm_area_struct *vma;
451 struct prio_tree_iter iter;
454 BUG_ON(PageAnon(page));
456 spin_lock(&mapping->i_mmap_lock);
457 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
458 if (vma->vm_flags & VM_SHARED)
459 ret += page_mkclean_one(page, vma);
461 spin_unlock(&mapping->i_mmap_lock);
465 int page_mkclean(struct page *page)
469 BUG_ON(!PageLocked(page));
471 if (page_mapped(page)) {
472 struct address_space *mapping = page_mapping(page);
474 ret = page_mkclean_file(mapping, page);
475 if (page_test_dirty(page)) {
476 page_clear_dirty(page);
483 EXPORT_SYMBOL_GPL(page_mkclean);
486 * page_set_anon_rmap - setup new anonymous rmap
487 * @page: the page to add the mapping to
488 * @vma: the vm area in which the mapping is added
489 * @address: the user virtual address mapped
491 static void __page_set_anon_rmap(struct page *page,
492 struct vm_area_struct *vma, unsigned long address)
494 struct anon_vma *anon_vma = vma->anon_vma;
497 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
498 page->mapping = (struct address_space *) anon_vma;
500 page->index = linear_page_index(vma, address);
503 * nr_mapped state can be updated without turning off
504 * interrupts because it is not modified via interrupt.
506 __inc_zone_page_state(page, NR_ANON_PAGES);
510 * page_set_anon_rmap - sanity check anonymous rmap addition
511 * @page: the page to add the mapping to
512 * @vma: the vm area in which the mapping is added
513 * @address: the user virtual address mapped
515 static void __page_check_anon_rmap(struct page *page,
516 struct vm_area_struct *vma, unsigned long address)
518 #ifdef CONFIG_DEBUG_VM
520 * The page's anon-rmap details (mapping and index) are guaranteed to
521 * be set up correctly at this point.
523 * We have exclusion against page_add_anon_rmap because the caller
524 * always holds the page locked, except if called from page_dup_rmap,
525 * in which case the page is already known to be setup.
527 * We have exclusion against page_add_new_anon_rmap because those pages
528 * are initially only visible via the pagetables, and the pte is locked
529 * over the call to page_add_new_anon_rmap.
531 struct anon_vma *anon_vma = vma->anon_vma;
532 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
533 BUG_ON(page->mapping != (struct address_space *)anon_vma);
534 BUG_ON(page->index != linear_page_index(vma, address));
539 * page_add_anon_rmap - add pte mapping to an anonymous page
540 * @page: the page to add the mapping to
541 * @vma: the vm area in which the mapping is added
542 * @address: the user virtual address mapped
544 * The caller needs to hold the pte lock and the page must be locked.
546 void page_add_anon_rmap(struct page *page,
547 struct vm_area_struct *vma, unsigned long address)
549 VM_BUG_ON(!PageLocked(page));
550 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
551 if (atomic_inc_and_test(&page->_mapcount))
552 __page_set_anon_rmap(page, vma, address);
554 __page_check_anon_rmap(page, vma, address);
558 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
559 * @page: the page to add the mapping to
560 * @vma: the vm area in which the mapping is added
561 * @address: the user virtual address mapped
563 * Same as page_add_anon_rmap but must only be called on *new* pages.
564 * This means the inc-and-test can be bypassed.
565 * Page does not have to be locked.
567 void page_add_new_anon_rmap(struct page *page,
568 struct vm_area_struct *vma, unsigned long address)
570 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
571 atomic_set(&page->_mapcount, 0); /* elevate count by 1 (starts at -1) */
572 __page_set_anon_rmap(page, vma, address);
576 * page_add_file_rmap - add pte mapping to a file page
577 * @page: the page to add the mapping to
579 * The caller needs to hold the pte lock.
581 void page_add_file_rmap(struct page *page)
583 if (atomic_inc_and_test(&page->_mapcount))
584 __inc_zone_page_state(page, NR_FILE_MAPPED);
587 #ifdef CONFIG_DEBUG_VM
589 * page_dup_rmap - duplicate pte mapping to a page
590 * @page: the page to add the mapping to
592 * For copy_page_range only: minimal extract from page_add_file_rmap /
593 * page_add_anon_rmap, avoiding unnecessary tests (already checked) so it's
596 * The caller needs to hold the pte lock.
598 void page_dup_rmap(struct page *page, struct vm_area_struct *vma, unsigned long address)
600 BUG_ON(page_mapcount(page) == 0);
602 __page_check_anon_rmap(page, vma, address);
603 atomic_inc(&page->_mapcount);
608 * page_remove_rmap - take down pte mapping from a page
609 * @page: page to remove mapping from
611 * The caller needs to hold the pte lock.
613 void page_remove_rmap(struct page *page, struct vm_area_struct *vma)
615 if (atomic_add_negative(-1, &page->_mapcount)) {
616 if (unlikely(page_mapcount(page) < 0)) {
617 printk (KERN_EMERG "Eeek! page_mapcount(page) went negative! (%d)\n", page_mapcount(page));
618 printk (KERN_EMERG " page pfn = %lx\n", page_to_pfn(page));
619 printk (KERN_EMERG " page->flags = %lx\n", page->flags);
620 printk (KERN_EMERG " page->count = %x\n", page_count(page));
621 printk (KERN_EMERG " page->mapping = %p\n", page->mapping);
622 print_symbol (KERN_EMERG " vma->vm_ops = %s\n", (unsigned long)vma->vm_ops);
624 print_symbol (KERN_EMERG " vma->vm_ops->nopage = %s\n", (unsigned long)vma->vm_ops->nopage);
625 print_symbol (KERN_EMERG " vma->vm_ops->fault = %s\n", (unsigned long)vma->vm_ops->fault);
627 if (vma->vm_file && vma->vm_file->f_op)
628 print_symbol (KERN_EMERG " vma->vm_file->f_op->mmap = %s\n", (unsigned long)vma->vm_file->f_op->mmap);
633 * It would be tidy to reset the PageAnon mapping here,
634 * but that might overwrite a racing page_add_anon_rmap
635 * which increments mapcount after us but sets mapping
636 * before us: so leave the reset to free_hot_cold_page,
637 * and remember that it's only reliable while mapped.
638 * Leaving it set also helps swapoff to reinstate ptes
639 * faster for those pages still in swapcache.
641 if (page_test_dirty(page)) {
642 page_clear_dirty(page);
643 set_page_dirty(page);
645 __dec_zone_page_state(page,
646 PageAnon(page) ? NR_ANON_PAGES : NR_FILE_MAPPED);
651 * Subfunctions of try_to_unmap: try_to_unmap_one called
652 * repeatedly from either try_to_unmap_anon or try_to_unmap_file.
654 static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
657 struct mm_struct *mm = vma->vm_mm;
658 unsigned long address;
662 int ret = SWAP_AGAIN;
664 address = vma_address(page, vma);
665 if (address == -EFAULT)
668 pte = page_check_address(page, mm, address, &ptl);
673 * If the page is mlock()d, we cannot swap it out.
674 * If it's recently referenced (perhaps page_referenced
675 * skipped over this mm) then we should reactivate it.
677 if (!migration && ((vma->vm_flags & VM_LOCKED) ||
678 (ptep_clear_flush_young(vma, address, pte)))) {
683 /* Nuke the page table entry. */
684 flush_cache_page(vma, address, page_to_pfn(page));
685 pteval = ptep_clear_flush(vma, address, pte);
687 /* Move the dirty bit to the physical page now the pte is gone. */
688 if (pte_dirty(pteval))
689 set_page_dirty(page);
691 /* Update high watermark before we lower rss */
692 update_hiwater_rss(mm);
694 if (PageAnon(page)) {
695 swp_entry_t entry = { .val = page_private(page) };
697 if (PageSwapCache(page)) {
699 * Store the swap location in the pte.
700 * See handle_pte_fault() ...
702 swap_duplicate(entry);
703 if (list_empty(&mm->mmlist)) {
704 spin_lock(&mmlist_lock);
705 if (list_empty(&mm->mmlist))
706 list_add(&mm->mmlist, &init_mm.mmlist);
707 spin_unlock(&mmlist_lock);
709 dec_mm_counter(mm, anon_rss);
710 #ifdef CONFIG_MIGRATION
713 * Store the pfn of the page in a special migration
714 * pte. do_swap_page() will wait until the migration
715 * pte is removed and then restart fault handling.
718 entry = make_migration_entry(page, pte_write(pteval));
721 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
722 BUG_ON(pte_file(*pte));
724 #ifdef CONFIG_MIGRATION
726 /* Establish migration entry for a file page */
728 entry = make_migration_entry(page, pte_write(pteval));
729 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
732 dec_mm_counter(mm, file_rss);
735 page_remove_rmap(page, vma);
736 page_cache_release(page);
739 pte_unmap_unlock(pte, ptl);
745 * objrmap doesn't work for nonlinear VMAs because the assumption that
746 * offset-into-file correlates with offset-into-virtual-addresses does not hold.
747 * Consequently, given a particular page and its ->index, we cannot locate the
748 * ptes which are mapping that page without an exhaustive linear search.
750 * So what this code does is a mini "virtual scan" of each nonlinear VMA which
751 * maps the file to which the target page belongs. The ->vm_private_data field
752 * holds the current cursor into that scan. Successive searches will circulate
753 * around the vma's virtual address space.
755 * So as more replacement pressure is applied to the pages in a nonlinear VMA,
756 * more scanning pressure is placed against them as well. Eventually pages
757 * will become fully unmapped and are eligible for eviction.
759 * For very sparsely populated VMAs this is a little inefficient - chances are
760 * there there won't be many ptes located within the scan cluster. In this case
761 * maybe we could scan further - to the end of the pte page, perhaps.
763 #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE)
764 #define CLUSTER_MASK (~(CLUSTER_SIZE - 1))
766 static void try_to_unmap_cluster(unsigned long cursor,
767 unsigned int *mapcount, struct vm_area_struct *vma)
769 struct mm_struct *mm = vma->vm_mm;
777 unsigned long address;
780 address = (vma->vm_start + cursor) & CLUSTER_MASK;
781 end = address + CLUSTER_SIZE;
782 if (address < vma->vm_start)
783 address = vma->vm_start;
784 if (end > vma->vm_end)
787 pgd = pgd_offset(mm, address);
788 if (!pgd_present(*pgd))
791 pud = pud_offset(pgd, address);
792 if (!pud_present(*pud))
795 pmd = pmd_offset(pud, address);
796 if (!pmd_present(*pmd))
799 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
801 /* Update high watermark before we lower rss */
802 update_hiwater_rss(mm);
804 for (; address < end; pte++, address += PAGE_SIZE) {
805 if (!pte_present(*pte))
807 page = vm_normal_page(vma, address, *pte);
808 BUG_ON(!page || PageAnon(page));
810 if (ptep_clear_flush_young(vma, address, pte))
813 /* Nuke the page table entry. */
814 flush_cache_page(vma, address, pte_pfn(*pte));
815 pteval = ptep_clear_flush(vma, address, pte);
817 /* If nonlinear, store the file page offset in the pte. */
818 if (page->index != linear_page_index(vma, address))
819 set_pte_at(mm, address, pte, pgoff_to_pte(page->index));
821 /* Move the dirty bit to the physical page now the pte is gone. */
822 if (pte_dirty(pteval))
823 set_page_dirty(page);
825 page_remove_rmap(page, vma);
826 page_cache_release(page);
827 dec_mm_counter(mm, file_rss);
830 pte_unmap_unlock(pte - 1, ptl);
833 static int try_to_unmap_anon(struct page *page, int migration)
835 struct anon_vma *anon_vma;
836 struct vm_area_struct *vma;
837 int ret = SWAP_AGAIN;
839 anon_vma = page_lock_anon_vma(page);
843 list_for_each_entry(vma, &anon_vma->head, anon_vma_node) {
844 ret = try_to_unmap_one(page, vma, migration);
845 if (ret == SWAP_FAIL || !page_mapped(page))
849 page_unlock_anon_vma(anon_vma);
854 * try_to_unmap_file - unmap file page using the object-based rmap method
855 * @page: the page to unmap
857 * Find all the mappings of a page using the mapping pointer and the vma chains
858 * contained in the address_space struct it points to.
860 * This function is only called from try_to_unmap for object-based pages.
862 static int try_to_unmap_file(struct page *page, int migration)
864 struct address_space *mapping = page->mapping;
865 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
866 struct vm_area_struct *vma;
867 struct prio_tree_iter iter;
868 int ret = SWAP_AGAIN;
869 unsigned long cursor;
870 unsigned long max_nl_cursor = 0;
871 unsigned long max_nl_size = 0;
872 unsigned int mapcount;
874 spin_lock(&mapping->i_mmap_lock);
875 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
876 ret = try_to_unmap_one(page, vma, migration);
877 if (ret == SWAP_FAIL || !page_mapped(page))
881 if (list_empty(&mapping->i_mmap_nonlinear))
884 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
885 shared.vm_set.list) {
886 if ((vma->vm_flags & VM_LOCKED) && !migration)
888 cursor = (unsigned long) vma->vm_private_data;
889 if (cursor > max_nl_cursor)
890 max_nl_cursor = cursor;
891 cursor = vma->vm_end - vma->vm_start;
892 if (cursor > max_nl_size)
893 max_nl_size = cursor;
896 if (max_nl_size == 0) { /* any nonlinears locked or reserved */
902 * We don't try to search for this page in the nonlinear vmas,
903 * and page_referenced wouldn't have found it anyway. Instead
904 * just walk the nonlinear vmas trying to age and unmap some.
905 * The mapcount of the page we came in with is irrelevant,
906 * but even so use it as a guide to how hard we should try?
908 mapcount = page_mapcount(page);
911 cond_resched_lock(&mapping->i_mmap_lock);
913 max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK;
914 if (max_nl_cursor == 0)
915 max_nl_cursor = CLUSTER_SIZE;
918 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
919 shared.vm_set.list) {
920 if ((vma->vm_flags & VM_LOCKED) && !migration)
922 cursor = (unsigned long) vma->vm_private_data;
923 while ( cursor < max_nl_cursor &&
924 cursor < vma->vm_end - vma->vm_start) {
925 try_to_unmap_cluster(cursor, &mapcount, vma);
926 cursor += CLUSTER_SIZE;
927 vma->vm_private_data = (void *) cursor;
928 if ((int)mapcount <= 0)
931 vma->vm_private_data = (void *) max_nl_cursor;
933 cond_resched_lock(&mapping->i_mmap_lock);
934 max_nl_cursor += CLUSTER_SIZE;
935 } while (max_nl_cursor <= max_nl_size);
938 * Don't loop forever (perhaps all the remaining pages are
939 * in locked vmas). Reset cursor on all unreserved nonlinear
940 * vmas, now forgetting on which ones it had fallen behind.
942 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
943 vma->vm_private_data = NULL;
945 spin_unlock(&mapping->i_mmap_lock);
950 * try_to_unmap - try to remove all page table mappings to a page
951 * @page: the page to get unmapped
953 * Tries to remove all the page table entries which are mapping this
954 * page, used in the pageout path. Caller must hold the page lock.
957 * SWAP_SUCCESS - we succeeded in removing all mappings
958 * SWAP_AGAIN - we missed a mapping, try again later
959 * SWAP_FAIL - the page is unswappable
961 int try_to_unmap(struct page *page, int migration)
965 BUG_ON(!PageLocked(page));
968 ret = try_to_unmap_anon(page, migration);
970 ret = try_to_unmap_file(page, migration);
972 if (!page_mapped(page))