11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * mm/rmap.c - physical to virtual reverse mappings 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright 2001, Rik van Riel <riel@conectiva.com.br> 51da177e4SLinus Torvalds * Released under the General Public License (GPL). 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * Simple, low overhead reverse mapping scheme. 81da177e4SLinus Torvalds * Please try to keep this thing as modular as possible. 91da177e4SLinus Torvalds * 101da177e4SLinus Torvalds * Provides methods for unmapping each kind of mapped page: 111da177e4SLinus Torvalds * the anon methods track anonymous pages, and 121da177e4SLinus Torvalds * the file methods track pages belonging to an inode. 131da177e4SLinus Torvalds * 141da177e4SLinus Torvalds * Original design by Rik van Riel <riel@conectiva.com.br> 2001 151da177e4SLinus Torvalds * File methods by Dave McCracken <dmccr@us.ibm.com> 2003, 2004 161da177e4SLinus Torvalds * Anonymous methods by Andrea Arcangeli <andrea@suse.de> 2004 1798f32602SHugh Dickins * Contributions by Hugh Dickins 2003, 2004 181da177e4SLinus Torvalds */ 191da177e4SLinus Torvalds 201da177e4SLinus Torvalds /* 211da177e4SLinus Torvalds * Lock ordering in mm: 221da177e4SLinus Torvalds * 231b1dcc1bSJes Sorensen * inode->i_mutex (while writing or truncating, not reading or faulting) 2482591e6eSNick Piggin * inode->i_alloc_sem (vmtruncate_range) 251da177e4SLinus Torvalds * mm->mmap_sem 261da177e4SLinus Torvalds * page->flags PG_locked (lock_page) 271da177e4SLinus Torvalds * mapping->i_mmap_lock 281da177e4SLinus Torvalds * anon_vma->lock 29b8072f09SHugh Dickins * mm->page_table_lock or pte_lock 30053837fcSNick Piggin * zone->lru_lock (in mark_page_accessed, isolate_lru_page) 315d337b91SHugh Dickins * swap_lock (in swap_duplicate, swap_info_get) 321da177e4SLinus Torvalds * mmlist_lock (in mmput, drain_mmlist and others) 331da177e4SLinus Torvalds * mapping->private_lock (in __set_page_dirty_buffers) 341da177e4SLinus Torvalds * inode_lock (in set_page_dirty's __mark_inode_dirty) 351da177e4SLinus Torvalds * sb_lock (within inode_lock in fs/fs-writeback.c) 361da177e4SLinus Torvalds * mapping->tree_lock (widely used, in set_page_dirty, 371da177e4SLinus Torvalds * in arch-dependent flush_dcache_mmap_lock, 381da177e4SLinus Torvalds * within inode_lock in __sync_single_inode) 396a46079cSAndi Kleen * 406a46079cSAndi Kleen * (code doesn't rely on that order so it could be switched around) 416a46079cSAndi Kleen * ->tasklist_lock 426a46079cSAndi Kleen * anon_vma->lock (memory_failure, collect_procs_anon) 436a46079cSAndi Kleen * pte map lock 441da177e4SLinus Torvalds */ 451da177e4SLinus Torvalds 461da177e4SLinus Torvalds #include <linux/mm.h> 471da177e4SLinus Torvalds #include <linux/pagemap.h> 481da177e4SLinus Torvalds #include <linux/swap.h> 491da177e4SLinus Torvalds #include <linux/swapops.h> 501da177e4SLinus Torvalds #include <linux/slab.h> 511da177e4SLinus Torvalds #include <linux/init.h> 521da177e4SLinus Torvalds #include <linux/rmap.h> 531da177e4SLinus Torvalds #include <linux/rcupdate.h> 54a48d07afSChristoph Lameter #include <linux/module.h> 558a9f3ccdSBalbir Singh #include <linux/memcontrol.h> 56cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 5764cdd548SKOSAKI Motohiro #include <linux/migrate.h> 581da177e4SLinus Torvalds 591da177e4SLinus Torvalds #include <asm/tlbflush.h> 601da177e4SLinus Torvalds 61b291f000SNick Piggin #include "internal.h" 62b291f000SNick Piggin 63fdd2e5f8SAdrian Bunk static struct kmem_cache *anon_vma_cachep; 64fdd2e5f8SAdrian Bunk 65fdd2e5f8SAdrian Bunk static inline struct anon_vma *anon_vma_alloc(void) 66fdd2e5f8SAdrian Bunk { 67fdd2e5f8SAdrian Bunk return kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL); 68fdd2e5f8SAdrian Bunk } 69fdd2e5f8SAdrian Bunk 70fdd2e5f8SAdrian Bunk static inline void anon_vma_free(struct anon_vma *anon_vma) 71fdd2e5f8SAdrian Bunk { 72fdd2e5f8SAdrian Bunk kmem_cache_free(anon_vma_cachep, anon_vma); 73fdd2e5f8SAdrian Bunk } 741da177e4SLinus Torvalds 75d9d332e0SLinus Torvalds /** 76d9d332e0SLinus Torvalds * anon_vma_prepare - attach an anon_vma to a memory region 77d9d332e0SLinus Torvalds * @vma: the memory region in question 78d9d332e0SLinus Torvalds * 79d9d332e0SLinus Torvalds * This makes sure the memory mapping described by 'vma' has 80d9d332e0SLinus Torvalds * an 'anon_vma' attached to it, so that we can associate the 81d9d332e0SLinus Torvalds * anonymous pages mapped into it with that anon_vma. 82d9d332e0SLinus Torvalds * 83d9d332e0SLinus Torvalds * The common case will be that we already have one, but if 84d9d332e0SLinus Torvalds * if not we either need to find an adjacent mapping that we 85d9d332e0SLinus Torvalds * can re-use the anon_vma from (very common when the only 86d9d332e0SLinus Torvalds * reason for splitting a vma has been mprotect()), or we 87d9d332e0SLinus Torvalds * allocate a new one. 88d9d332e0SLinus Torvalds * 89d9d332e0SLinus Torvalds * Anon-vma allocations are very subtle, because we may have 90d9d332e0SLinus Torvalds * optimistically looked up an anon_vma in page_lock_anon_vma() 91d9d332e0SLinus Torvalds * and that may actually touch the spinlock even in the newly 92d9d332e0SLinus Torvalds * allocated vma (it depends on RCU to make sure that the 93d9d332e0SLinus Torvalds * anon_vma isn't actually destroyed). 94d9d332e0SLinus Torvalds * 95d9d332e0SLinus Torvalds * As a result, we need to do proper anon_vma locking even 96d9d332e0SLinus Torvalds * for the new allocation. At the same time, we do not want 97d9d332e0SLinus Torvalds * to do any locking for the common case of already having 98d9d332e0SLinus Torvalds * an anon_vma. 99d9d332e0SLinus Torvalds * 100d9d332e0SLinus Torvalds * This must be called with the mmap_sem held for reading. 101d9d332e0SLinus Torvalds */ 1021da177e4SLinus Torvalds int anon_vma_prepare(struct vm_area_struct *vma) 1031da177e4SLinus Torvalds { 1041da177e4SLinus Torvalds struct anon_vma *anon_vma = vma->anon_vma; 1051da177e4SLinus Torvalds 1061da177e4SLinus Torvalds might_sleep(); 1071da177e4SLinus Torvalds if (unlikely(!anon_vma)) { 1081da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 109d9d332e0SLinus Torvalds struct anon_vma *allocated; 1101da177e4SLinus Torvalds 1111da177e4SLinus Torvalds anon_vma = find_mergeable_anon_vma(vma); 1121da177e4SLinus Torvalds allocated = NULL; 113d9d332e0SLinus Torvalds if (!anon_vma) { 1141da177e4SLinus Torvalds anon_vma = anon_vma_alloc(); 1151da177e4SLinus Torvalds if (unlikely(!anon_vma)) 1161da177e4SLinus Torvalds return -ENOMEM; 1171da177e4SLinus Torvalds allocated = anon_vma; 1181da177e4SLinus Torvalds } 119d9d332e0SLinus Torvalds spin_lock(&anon_vma->lock); 1201da177e4SLinus Torvalds 1211da177e4SLinus Torvalds /* page_table_lock to protect against threads */ 1221da177e4SLinus Torvalds spin_lock(&mm->page_table_lock); 1231da177e4SLinus Torvalds if (likely(!vma->anon_vma)) { 1241da177e4SLinus Torvalds vma->anon_vma = anon_vma; 1250697212aSChristoph Lameter list_add_tail(&vma->anon_vma_node, &anon_vma->head); 1261da177e4SLinus Torvalds allocated = NULL; 1271da177e4SLinus Torvalds } 1281da177e4SLinus Torvalds spin_unlock(&mm->page_table_lock); 1291da177e4SLinus Torvalds 130d9d332e0SLinus Torvalds spin_unlock(&anon_vma->lock); 1311da177e4SLinus Torvalds if (unlikely(allocated)) 1321da177e4SLinus Torvalds anon_vma_free(allocated); 1331da177e4SLinus Torvalds } 1341da177e4SLinus Torvalds return 0; 1351da177e4SLinus Torvalds } 1361da177e4SLinus Torvalds 1371da177e4SLinus Torvalds void __anon_vma_merge(struct vm_area_struct *vma, struct vm_area_struct *next) 1381da177e4SLinus Torvalds { 1391da177e4SLinus Torvalds BUG_ON(vma->anon_vma != next->anon_vma); 1401da177e4SLinus Torvalds list_del(&next->anon_vma_node); 1411da177e4SLinus Torvalds } 1421da177e4SLinus Torvalds 1431da177e4SLinus Torvalds void __anon_vma_link(struct vm_area_struct *vma) 1441da177e4SLinus Torvalds { 1451da177e4SLinus Torvalds struct anon_vma *anon_vma = vma->anon_vma; 1461da177e4SLinus Torvalds 14730acbabaSHugh Dickins if (anon_vma) 1480697212aSChristoph Lameter list_add_tail(&vma->anon_vma_node, &anon_vma->head); 1491da177e4SLinus Torvalds } 1501da177e4SLinus Torvalds 1511da177e4SLinus Torvalds void anon_vma_link(struct vm_area_struct *vma) 1521da177e4SLinus Torvalds { 1531da177e4SLinus Torvalds struct anon_vma *anon_vma = vma->anon_vma; 1541da177e4SLinus Torvalds 1551da177e4SLinus Torvalds if (anon_vma) { 1561da177e4SLinus Torvalds spin_lock(&anon_vma->lock); 1570697212aSChristoph Lameter list_add_tail(&vma->anon_vma_node, &anon_vma->head); 1581da177e4SLinus Torvalds spin_unlock(&anon_vma->lock); 1591da177e4SLinus Torvalds } 1601da177e4SLinus Torvalds } 1611da177e4SLinus Torvalds 1621da177e4SLinus Torvalds void anon_vma_unlink(struct vm_area_struct *vma) 1631da177e4SLinus Torvalds { 1641da177e4SLinus Torvalds struct anon_vma *anon_vma = vma->anon_vma; 1651da177e4SLinus Torvalds int empty; 1661da177e4SLinus Torvalds 1671da177e4SLinus Torvalds if (!anon_vma) 1681da177e4SLinus Torvalds return; 1691da177e4SLinus Torvalds 1701da177e4SLinus Torvalds spin_lock(&anon_vma->lock); 1711da177e4SLinus Torvalds list_del(&vma->anon_vma_node); 1721da177e4SLinus Torvalds 1731da177e4SLinus Torvalds /* We must garbage collect the anon_vma if it's empty */ 1741da177e4SLinus Torvalds empty = list_empty(&anon_vma->head); 1751da177e4SLinus Torvalds spin_unlock(&anon_vma->lock); 1761da177e4SLinus Torvalds 1771da177e4SLinus Torvalds if (empty) 1781da177e4SLinus Torvalds anon_vma_free(anon_vma); 1791da177e4SLinus Torvalds } 1801da177e4SLinus Torvalds 18151cc5068SAlexey Dobriyan static void anon_vma_ctor(void *data) 1821da177e4SLinus Torvalds { 1831da177e4SLinus Torvalds struct anon_vma *anon_vma = data; 1841da177e4SLinus Torvalds 1851da177e4SLinus Torvalds spin_lock_init(&anon_vma->lock); 1861da177e4SLinus Torvalds INIT_LIST_HEAD(&anon_vma->head); 1871da177e4SLinus Torvalds } 1881da177e4SLinus Torvalds 1891da177e4SLinus Torvalds void __init anon_vma_init(void) 1901da177e4SLinus Torvalds { 1911da177e4SLinus Torvalds anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), 19220c2df83SPaul Mundt 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor); 1931da177e4SLinus Torvalds } 1941da177e4SLinus Torvalds 1951da177e4SLinus Torvalds /* 1961da177e4SLinus Torvalds * Getting a lock on a stable anon_vma from a page off the LRU is 1971da177e4SLinus Torvalds * tricky: page_lock_anon_vma rely on RCU to guard against the races. 1981da177e4SLinus Torvalds */ 19910be22dfSAndi Kleen struct anon_vma *page_lock_anon_vma(struct page *page) 2001da177e4SLinus Torvalds { 20134bbd704SOleg Nesterov struct anon_vma *anon_vma; 2021da177e4SLinus Torvalds unsigned long anon_mapping; 2031da177e4SLinus Torvalds 2041da177e4SLinus Torvalds rcu_read_lock(); 2051da177e4SLinus Torvalds anon_mapping = (unsigned long) page->mapping; 206*3ca7b3c5SHugh Dickins if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) 2071da177e4SLinus Torvalds goto out; 2081da177e4SLinus Torvalds if (!page_mapped(page)) 2091da177e4SLinus Torvalds goto out; 2101da177e4SLinus Torvalds 2111da177e4SLinus Torvalds anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); 2121da177e4SLinus Torvalds spin_lock(&anon_vma->lock); 21334bbd704SOleg Nesterov return anon_vma; 2141da177e4SLinus Torvalds out: 2151da177e4SLinus Torvalds rcu_read_unlock(); 21634bbd704SOleg Nesterov return NULL; 21734bbd704SOleg Nesterov } 21834bbd704SOleg Nesterov 21910be22dfSAndi Kleen void page_unlock_anon_vma(struct anon_vma *anon_vma) 22034bbd704SOleg Nesterov { 22134bbd704SOleg Nesterov spin_unlock(&anon_vma->lock); 22234bbd704SOleg Nesterov rcu_read_unlock(); 2231da177e4SLinus Torvalds } 2241da177e4SLinus Torvalds 2251da177e4SLinus Torvalds /* 2263ad33b24SLee Schermerhorn * At what user virtual address is page expected in @vma? 2273ad33b24SLee Schermerhorn * Returns virtual address or -EFAULT if page's index/offset is not 2283ad33b24SLee Schermerhorn * within the range mapped the @vma. 2291da177e4SLinus Torvalds */ 2301da177e4SLinus Torvalds static inline unsigned long 2311da177e4SLinus Torvalds vma_address(struct page *page, struct vm_area_struct *vma) 2321da177e4SLinus Torvalds { 2331da177e4SLinus Torvalds pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); 2341da177e4SLinus Torvalds unsigned long address; 2351da177e4SLinus Torvalds 2361da177e4SLinus Torvalds address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); 2371da177e4SLinus Torvalds if (unlikely(address < vma->vm_start || address >= vma->vm_end)) { 2383ad33b24SLee Schermerhorn /* page should be within @vma mapping range */ 2391da177e4SLinus Torvalds return -EFAULT; 2401da177e4SLinus Torvalds } 2411da177e4SLinus Torvalds return address; 2421da177e4SLinus Torvalds } 2431da177e4SLinus Torvalds 2441da177e4SLinus Torvalds /* 245bf89c8c8SHuang Shijie * At what user virtual address is page expected in vma? 246bf89c8c8SHuang Shijie * checking that the page matches the vma. 2471da177e4SLinus Torvalds */ 2481da177e4SLinus Torvalds unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) 2491da177e4SLinus Torvalds { 2501da177e4SLinus Torvalds if (PageAnon(page)) { 251*3ca7b3c5SHugh Dickins if (vma->anon_vma != page_anon_vma(page)) 2521da177e4SLinus Torvalds return -EFAULT; 2531da177e4SLinus Torvalds } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) { 254ee498ed7SHugh Dickins if (!vma->vm_file || 255ee498ed7SHugh Dickins vma->vm_file->f_mapping != page->mapping) 2561da177e4SLinus Torvalds return -EFAULT; 2571da177e4SLinus Torvalds } else 2581da177e4SLinus Torvalds return -EFAULT; 2591da177e4SLinus Torvalds return vma_address(page, vma); 2601da177e4SLinus Torvalds } 2611da177e4SLinus Torvalds 2621da177e4SLinus Torvalds /* 26381b4082dSNikita Danilov * Check that @page is mapped at @address into @mm. 26481b4082dSNikita Danilov * 265479db0bfSNick Piggin * If @sync is false, page_check_address may perform a racy check to avoid 266479db0bfSNick Piggin * the page table lock when the pte is not present (helpful when reclaiming 267479db0bfSNick Piggin * highly shared pages). 268479db0bfSNick Piggin * 269b8072f09SHugh Dickins * On success returns with pte mapped and locked. 27081b4082dSNikita Danilov */ 271ceffc078SCarsten Otte pte_t *page_check_address(struct page *page, struct mm_struct *mm, 272479db0bfSNick Piggin unsigned long address, spinlock_t **ptlp, int sync) 27381b4082dSNikita Danilov { 27481b4082dSNikita Danilov pgd_t *pgd; 27581b4082dSNikita Danilov pud_t *pud; 27681b4082dSNikita Danilov pmd_t *pmd; 27781b4082dSNikita Danilov pte_t *pte; 278c0718806SHugh Dickins spinlock_t *ptl; 27981b4082dSNikita Danilov 28081b4082dSNikita Danilov pgd = pgd_offset(mm, address); 281c0718806SHugh Dickins if (!pgd_present(*pgd)) 282c0718806SHugh Dickins return NULL; 283c0718806SHugh Dickins 28481b4082dSNikita Danilov pud = pud_offset(pgd, address); 285c0718806SHugh Dickins if (!pud_present(*pud)) 286c0718806SHugh Dickins return NULL; 287c0718806SHugh Dickins 28881b4082dSNikita Danilov pmd = pmd_offset(pud, address); 289c0718806SHugh Dickins if (!pmd_present(*pmd)) 290c0718806SHugh Dickins return NULL; 291c0718806SHugh Dickins 29281b4082dSNikita Danilov pte = pte_offset_map(pmd, address); 293c0718806SHugh Dickins /* Make a quick check before getting the lock */ 294479db0bfSNick Piggin if (!sync && !pte_present(*pte)) { 29581b4082dSNikita Danilov pte_unmap(pte); 296c0718806SHugh Dickins return NULL; 29781b4082dSNikita Danilov } 298c0718806SHugh Dickins 2994c21e2f2SHugh Dickins ptl = pte_lockptr(mm, pmd); 300c0718806SHugh Dickins spin_lock(ptl); 301c0718806SHugh Dickins if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) { 302c0718806SHugh Dickins *ptlp = ptl; 303c0718806SHugh Dickins return pte; 30481b4082dSNikita Danilov } 305c0718806SHugh Dickins pte_unmap_unlock(pte, ptl); 306c0718806SHugh Dickins return NULL; 30781b4082dSNikita Danilov } 30881b4082dSNikita Danilov 309b291f000SNick Piggin /** 310b291f000SNick Piggin * page_mapped_in_vma - check whether a page is really mapped in a VMA 311b291f000SNick Piggin * @page: the page to test 312b291f000SNick Piggin * @vma: the VMA to test 313b291f000SNick Piggin * 314b291f000SNick Piggin * Returns 1 if the page is mapped into the page tables of the VMA, 0 315b291f000SNick Piggin * if the page is not mapped into the page tables of this VMA. Only 316b291f000SNick Piggin * valid for normal file or anonymous VMAs. 317b291f000SNick Piggin */ 3186a46079cSAndi Kleen int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma) 319b291f000SNick Piggin { 320b291f000SNick Piggin unsigned long address; 321b291f000SNick Piggin pte_t *pte; 322b291f000SNick Piggin spinlock_t *ptl; 323b291f000SNick Piggin 324b291f000SNick Piggin address = vma_address(page, vma); 325b291f000SNick Piggin if (address == -EFAULT) /* out of vma range */ 326b291f000SNick Piggin return 0; 327b291f000SNick Piggin pte = page_check_address(page, vma->vm_mm, address, &ptl, 1); 328b291f000SNick Piggin if (!pte) /* the page is not in this mm */ 329b291f000SNick Piggin return 0; 330b291f000SNick Piggin pte_unmap_unlock(pte, ptl); 331b291f000SNick Piggin 332b291f000SNick Piggin return 1; 333b291f000SNick Piggin } 334b291f000SNick Piggin 33581b4082dSNikita Danilov /* 3361da177e4SLinus Torvalds * Subfunctions of page_referenced: page_referenced_one called 3371da177e4SLinus Torvalds * repeatedly from either page_referenced_anon or page_referenced_file. 3381da177e4SLinus Torvalds */ 3391da177e4SLinus Torvalds static int page_referenced_one(struct page *page, 3406fe6b7e3SWu Fengguang struct vm_area_struct *vma, 3416fe6b7e3SWu Fengguang unsigned int *mapcount, 3426fe6b7e3SWu Fengguang unsigned long *vm_flags) 3431da177e4SLinus Torvalds { 3441da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 3451da177e4SLinus Torvalds unsigned long address; 3461da177e4SLinus Torvalds pte_t *pte; 347c0718806SHugh Dickins spinlock_t *ptl; 3481da177e4SLinus Torvalds int referenced = 0; 3491da177e4SLinus Torvalds 3501da177e4SLinus Torvalds address = vma_address(page, vma); 3511da177e4SLinus Torvalds if (address == -EFAULT) 3521da177e4SLinus Torvalds goto out; 3531da177e4SLinus Torvalds 354479db0bfSNick Piggin pte = page_check_address(page, mm, address, &ptl, 0); 355c0718806SHugh Dickins if (!pte) 356c0718806SHugh Dickins goto out; 357c0718806SHugh Dickins 358b291f000SNick Piggin /* 359b291f000SNick Piggin * Don't want to elevate referenced for mlocked page that gets this far, 360b291f000SNick Piggin * in order that it progresses to try_to_unmap and is moved to the 361b291f000SNick Piggin * unevictable list. 362b291f000SNick Piggin */ 3635a9bbdcdSHugh Dickins if (vma->vm_flags & VM_LOCKED) { 3645a9bbdcdSHugh Dickins *mapcount = 1; /* break early from loop */ 36503ef83afSMinchan Kim *vm_flags |= VM_LOCKED; 366b291f000SNick Piggin goto out_unmap; 367b291f000SNick Piggin } 368b291f000SNick Piggin 3694917e5d0SJohannes Weiner if (ptep_clear_flush_young_notify(vma, address, pte)) { 3704917e5d0SJohannes Weiner /* 3714917e5d0SJohannes Weiner * Don't treat a reference through a sequentially read 3724917e5d0SJohannes Weiner * mapping as such. If the page has been used in 3734917e5d0SJohannes Weiner * another mapping, we will catch it; if this other 3744917e5d0SJohannes Weiner * mapping is already gone, the unmap path will have 3754917e5d0SJohannes Weiner * set PG_referenced or activated the page. 3764917e5d0SJohannes Weiner */ 3774917e5d0SJohannes Weiner if (likely(!VM_SequentialReadHint(vma))) 3781da177e4SLinus Torvalds referenced++; 3794917e5d0SJohannes Weiner } 3801da177e4SLinus Torvalds 381fcdae29aSRik Van Riel /* Pretend the page is referenced if the task has the 382fcdae29aSRik Van Riel swap token and is in the middle of a page fault. */ 383f7b7fd8fSRik van Riel if (mm != current->mm && has_swap_token(mm) && 384fcdae29aSRik Van Riel rwsem_is_locked(&mm->mmap_sem)) 3851da177e4SLinus Torvalds referenced++; 3861da177e4SLinus Torvalds 387b291f000SNick Piggin out_unmap: 3881da177e4SLinus Torvalds (*mapcount)--; 389c0718806SHugh Dickins pte_unmap_unlock(pte, ptl); 390273f047eSHuang Shijie 3916fe6b7e3SWu Fengguang if (referenced) 3926fe6b7e3SWu Fengguang *vm_flags |= vma->vm_flags; 393273f047eSHuang Shijie out: 3941da177e4SLinus Torvalds return referenced; 3951da177e4SLinus Torvalds } 3961da177e4SLinus Torvalds 397bed7161aSBalbir Singh static int page_referenced_anon(struct page *page, 3986fe6b7e3SWu Fengguang struct mem_cgroup *mem_cont, 3996fe6b7e3SWu Fengguang unsigned long *vm_flags) 4001da177e4SLinus Torvalds { 4011da177e4SLinus Torvalds unsigned int mapcount; 4021da177e4SLinus Torvalds struct anon_vma *anon_vma; 4031da177e4SLinus Torvalds struct vm_area_struct *vma; 4041da177e4SLinus Torvalds int referenced = 0; 4051da177e4SLinus Torvalds 4061da177e4SLinus Torvalds anon_vma = page_lock_anon_vma(page); 4071da177e4SLinus Torvalds if (!anon_vma) 4081da177e4SLinus Torvalds return referenced; 4091da177e4SLinus Torvalds 4101da177e4SLinus Torvalds mapcount = page_mapcount(page); 4111da177e4SLinus Torvalds list_for_each_entry(vma, &anon_vma->head, anon_vma_node) { 412bed7161aSBalbir Singh /* 413bed7161aSBalbir Singh * If we are reclaiming on behalf of a cgroup, skip 414bed7161aSBalbir Singh * counting on behalf of references from different 415bed7161aSBalbir Singh * cgroups 416bed7161aSBalbir Singh */ 417bd845e38SHugh Dickins if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) 418bed7161aSBalbir Singh continue; 4196fe6b7e3SWu Fengguang referenced += page_referenced_one(page, vma, 4206fe6b7e3SWu Fengguang &mapcount, vm_flags); 4211da177e4SLinus Torvalds if (!mapcount) 4221da177e4SLinus Torvalds break; 4231da177e4SLinus Torvalds } 42434bbd704SOleg Nesterov 42534bbd704SOleg Nesterov page_unlock_anon_vma(anon_vma); 4261da177e4SLinus Torvalds return referenced; 4271da177e4SLinus Torvalds } 4281da177e4SLinus Torvalds 4291da177e4SLinus Torvalds /** 4301da177e4SLinus Torvalds * page_referenced_file - referenced check for object-based rmap 4311da177e4SLinus Torvalds * @page: the page we're checking references on. 43243d8eac4SRandy Dunlap * @mem_cont: target memory controller 4336fe6b7e3SWu Fengguang * @vm_flags: collect encountered vma->vm_flags who actually referenced the page 4341da177e4SLinus Torvalds * 4351da177e4SLinus Torvalds * For an object-based mapped page, find all the places it is mapped and 4361da177e4SLinus Torvalds * check/clear the referenced flag. This is done by following the page->mapping 4371da177e4SLinus Torvalds * pointer, then walking the chain of vmas it holds. It returns the number 4381da177e4SLinus Torvalds * of references it found. 4391da177e4SLinus Torvalds * 4401da177e4SLinus Torvalds * This function is only called from page_referenced for object-based pages. 4411da177e4SLinus Torvalds */ 442bed7161aSBalbir Singh static int page_referenced_file(struct page *page, 4436fe6b7e3SWu Fengguang struct mem_cgroup *mem_cont, 4446fe6b7e3SWu Fengguang unsigned long *vm_flags) 4451da177e4SLinus Torvalds { 4461da177e4SLinus Torvalds unsigned int mapcount; 4471da177e4SLinus Torvalds struct address_space *mapping = page->mapping; 4481da177e4SLinus Torvalds pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); 4491da177e4SLinus Torvalds struct vm_area_struct *vma; 4501da177e4SLinus Torvalds struct prio_tree_iter iter; 4511da177e4SLinus Torvalds int referenced = 0; 4521da177e4SLinus Torvalds 4531da177e4SLinus Torvalds /* 4541da177e4SLinus Torvalds * The caller's checks on page->mapping and !PageAnon have made 4551da177e4SLinus Torvalds * sure that this is a file page: the check for page->mapping 4561da177e4SLinus Torvalds * excludes the case just before it gets set on an anon page. 4571da177e4SLinus Torvalds */ 4581da177e4SLinus Torvalds BUG_ON(PageAnon(page)); 4591da177e4SLinus Torvalds 4601da177e4SLinus Torvalds /* 4611da177e4SLinus Torvalds * The page lock not only makes sure that page->mapping cannot 4621da177e4SLinus Torvalds * suddenly be NULLified by truncation, it makes sure that the 4631da177e4SLinus Torvalds * structure at mapping cannot be freed and reused yet, 4641da177e4SLinus Torvalds * so we can safely take mapping->i_mmap_lock. 4651da177e4SLinus Torvalds */ 4661da177e4SLinus Torvalds BUG_ON(!PageLocked(page)); 4671da177e4SLinus Torvalds 4681da177e4SLinus Torvalds spin_lock(&mapping->i_mmap_lock); 4691da177e4SLinus Torvalds 4701da177e4SLinus Torvalds /* 4711da177e4SLinus Torvalds * i_mmap_lock does not stabilize mapcount at all, but mapcount 4721da177e4SLinus Torvalds * is more likely to be accurate if we note it after spinning. 4731da177e4SLinus Torvalds */ 4741da177e4SLinus Torvalds mapcount = page_mapcount(page); 4751da177e4SLinus Torvalds 4761da177e4SLinus Torvalds vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { 477bed7161aSBalbir Singh /* 478bed7161aSBalbir Singh * If we are reclaiming on behalf of a cgroup, skip 479bed7161aSBalbir Singh * counting on behalf of references from different 480bed7161aSBalbir Singh * cgroups 481bed7161aSBalbir Singh */ 482bd845e38SHugh Dickins if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) 483bed7161aSBalbir Singh continue; 4846fe6b7e3SWu Fengguang referenced += page_referenced_one(page, vma, 4856fe6b7e3SWu Fengguang &mapcount, vm_flags); 4861da177e4SLinus Torvalds if (!mapcount) 4871da177e4SLinus Torvalds break; 4881da177e4SLinus Torvalds } 4891da177e4SLinus Torvalds 4901da177e4SLinus Torvalds spin_unlock(&mapping->i_mmap_lock); 4911da177e4SLinus Torvalds return referenced; 4921da177e4SLinus Torvalds } 4931da177e4SLinus Torvalds 4941da177e4SLinus Torvalds /** 4951da177e4SLinus Torvalds * page_referenced - test if the page was referenced 4961da177e4SLinus Torvalds * @page: the page to test 4971da177e4SLinus Torvalds * @is_locked: caller holds lock on the page 49843d8eac4SRandy Dunlap * @mem_cont: target memory controller 4996fe6b7e3SWu Fengguang * @vm_flags: collect encountered vma->vm_flags who actually referenced the page 5001da177e4SLinus Torvalds * 5011da177e4SLinus Torvalds * Quick test_and_clear_referenced for all mappings to a page, 5021da177e4SLinus Torvalds * returns the number of ptes which referenced the page. 5031da177e4SLinus Torvalds */ 5046fe6b7e3SWu Fengguang int page_referenced(struct page *page, 5056fe6b7e3SWu Fengguang int is_locked, 5066fe6b7e3SWu Fengguang struct mem_cgroup *mem_cont, 5076fe6b7e3SWu Fengguang unsigned long *vm_flags) 5081da177e4SLinus Torvalds { 5091da177e4SLinus Torvalds int referenced = 0; 5101da177e4SLinus Torvalds 5111da177e4SLinus Torvalds if (TestClearPageReferenced(page)) 5121da177e4SLinus Torvalds referenced++; 5131da177e4SLinus Torvalds 5146fe6b7e3SWu Fengguang *vm_flags = 0; 515*3ca7b3c5SHugh Dickins if (page_mapped(page) && page_rmapping(page)) { 5161da177e4SLinus Torvalds if (PageAnon(page)) 5176fe6b7e3SWu Fengguang referenced += page_referenced_anon(page, mem_cont, 5186fe6b7e3SWu Fengguang vm_flags); 5191da177e4SLinus Torvalds else if (is_locked) 5206fe6b7e3SWu Fengguang referenced += page_referenced_file(page, mem_cont, 5216fe6b7e3SWu Fengguang vm_flags); 522529ae9aaSNick Piggin else if (!trylock_page(page)) 5231da177e4SLinus Torvalds referenced++; 5241da177e4SLinus Torvalds else { 5251da177e4SLinus Torvalds if (page->mapping) 5266fe6b7e3SWu Fengguang referenced += page_referenced_file(page, 5276fe6b7e3SWu Fengguang mem_cont, vm_flags); 5281da177e4SLinus Torvalds unlock_page(page); 5291da177e4SLinus Torvalds } 5301da177e4SLinus Torvalds } 5315b7baf05SChristian Borntraeger 5325b7baf05SChristian Borntraeger if (page_test_and_clear_young(page)) 5335b7baf05SChristian Borntraeger referenced++; 5345b7baf05SChristian Borntraeger 5351da177e4SLinus Torvalds return referenced; 5361da177e4SLinus Torvalds } 5371da177e4SLinus Torvalds 538d08b3851SPeter Zijlstra static int page_mkclean_one(struct page *page, struct vm_area_struct *vma) 539d08b3851SPeter Zijlstra { 540d08b3851SPeter Zijlstra struct mm_struct *mm = vma->vm_mm; 541d08b3851SPeter Zijlstra unsigned long address; 542c2fda5feSPeter Zijlstra pte_t *pte; 543d08b3851SPeter Zijlstra spinlock_t *ptl; 544d08b3851SPeter Zijlstra int ret = 0; 545d08b3851SPeter Zijlstra 546d08b3851SPeter Zijlstra address = vma_address(page, vma); 547d08b3851SPeter Zijlstra if (address == -EFAULT) 548d08b3851SPeter Zijlstra goto out; 549d08b3851SPeter Zijlstra 550479db0bfSNick Piggin pte = page_check_address(page, mm, address, &ptl, 1); 551d08b3851SPeter Zijlstra if (!pte) 552d08b3851SPeter Zijlstra goto out; 553d08b3851SPeter Zijlstra 554c2fda5feSPeter Zijlstra if (pte_dirty(*pte) || pte_write(*pte)) { 555c2fda5feSPeter Zijlstra pte_t entry; 556d08b3851SPeter Zijlstra 557c2fda5feSPeter Zijlstra flush_cache_page(vma, address, pte_pfn(*pte)); 558cddb8a5cSAndrea Arcangeli entry = ptep_clear_flush_notify(vma, address, pte); 559d08b3851SPeter Zijlstra entry = pte_wrprotect(entry); 560c2fda5feSPeter Zijlstra entry = pte_mkclean(entry); 561d6e88e67SAl Viro set_pte_at(mm, address, pte, entry); 562d08b3851SPeter Zijlstra ret = 1; 563c2fda5feSPeter Zijlstra } 564d08b3851SPeter Zijlstra 565d08b3851SPeter Zijlstra pte_unmap_unlock(pte, ptl); 566d08b3851SPeter Zijlstra out: 567d08b3851SPeter Zijlstra return ret; 568d08b3851SPeter Zijlstra } 569d08b3851SPeter Zijlstra 570d08b3851SPeter Zijlstra static int page_mkclean_file(struct address_space *mapping, struct page *page) 571d08b3851SPeter Zijlstra { 572d08b3851SPeter Zijlstra pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); 573d08b3851SPeter Zijlstra struct vm_area_struct *vma; 574d08b3851SPeter Zijlstra struct prio_tree_iter iter; 575d08b3851SPeter Zijlstra int ret = 0; 576d08b3851SPeter Zijlstra 577d08b3851SPeter Zijlstra BUG_ON(PageAnon(page)); 578d08b3851SPeter Zijlstra 579d08b3851SPeter Zijlstra spin_lock(&mapping->i_mmap_lock); 580d08b3851SPeter Zijlstra vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { 581d08b3851SPeter Zijlstra if (vma->vm_flags & VM_SHARED) 582d08b3851SPeter Zijlstra ret += page_mkclean_one(page, vma); 583d08b3851SPeter Zijlstra } 584d08b3851SPeter Zijlstra spin_unlock(&mapping->i_mmap_lock); 585d08b3851SPeter Zijlstra return ret; 586d08b3851SPeter Zijlstra } 587d08b3851SPeter Zijlstra 588d08b3851SPeter Zijlstra int page_mkclean(struct page *page) 589d08b3851SPeter Zijlstra { 590d08b3851SPeter Zijlstra int ret = 0; 591d08b3851SPeter Zijlstra 592d08b3851SPeter Zijlstra BUG_ON(!PageLocked(page)); 593d08b3851SPeter Zijlstra 594d08b3851SPeter Zijlstra if (page_mapped(page)) { 595d08b3851SPeter Zijlstra struct address_space *mapping = page_mapping(page); 596ce7e9faeSChristian Borntraeger if (mapping) { 597d08b3851SPeter Zijlstra ret = page_mkclean_file(mapping, page); 5986c210482SMartin Schwidefsky if (page_test_dirty(page)) { 5996c210482SMartin Schwidefsky page_clear_dirty(page); 600c2fda5feSPeter Zijlstra ret = 1; 6016e1beb3cSMartin Schwidefsky } 6026c210482SMartin Schwidefsky } 603ce7e9faeSChristian Borntraeger } 604d08b3851SPeter Zijlstra 605d08b3851SPeter Zijlstra return ret; 606d08b3851SPeter Zijlstra } 60760b59beaSJaya Kumar EXPORT_SYMBOL_GPL(page_mkclean); 608d08b3851SPeter Zijlstra 6091da177e4SLinus Torvalds /** 61043d8eac4SRandy Dunlap * __page_set_anon_rmap - setup new anonymous rmap 6111da177e4SLinus Torvalds * @page: the page to add the mapping to 6121da177e4SLinus Torvalds * @vma: the vm area in which the mapping is added 6131da177e4SLinus Torvalds * @address: the user virtual address mapped 6141da177e4SLinus Torvalds */ 6159617d95eSNick Piggin static void __page_set_anon_rmap(struct page *page, 6161da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long address) 6171da177e4SLinus Torvalds { 6182822c1aaSNick Piggin struct anon_vma *anon_vma = vma->anon_vma; 6192822c1aaSNick Piggin 6202822c1aaSNick Piggin BUG_ON(!anon_vma); 6211da177e4SLinus Torvalds anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; 6222822c1aaSNick Piggin page->mapping = (struct address_space *) anon_vma; 6232822c1aaSNick Piggin 6244d7670e0SNick Piggin page->index = linear_page_index(vma, address); 6252822c1aaSNick Piggin 626a74609faSNick Piggin /* 627a74609faSNick Piggin * nr_mapped state can be updated without turning off 628a74609faSNick Piggin * interrupts because it is not modified via interrupt. 629a74609faSNick Piggin */ 630f3dbd344SChristoph Lameter __inc_zone_page_state(page, NR_ANON_PAGES); 6311da177e4SLinus Torvalds } 6329617d95eSNick Piggin 6339617d95eSNick Piggin /** 63443d8eac4SRandy Dunlap * __page_check_anon_rmap - sanity check anonymous rmap addition 635c97a9e10SNick Piggin * @page: the page to add the mapping to 636c97a9e10SNick Piggin * @vma: the vm area in which the mapping is added 637c97a9e10SNick Piggin * @address: the user virtual address mapped 638c97a9e10SNick Piggin */ 639c97a9e10SNick Piggin static void __page_check_anon_rmap(struct page *page, 640c97a9e10SNick Piggin struct vm_area_struct *vma, unsigned long address) 641c97a9e10SNick Piggin { 642c97a9e10SNick Piggin #ifdef CONFIG_DEBUG_VM 643c97a9e10SNick Piggin /* 644c97a9e10SNick Piggin * The page's anon-rmap details (mapping and index) are guaranteed to 645c97a9e10SNick Piggin * be set up correctly at this point. 646c97a9e10SNick Piggin * 647c97a9e10SNick Piggin * We have exclusion against page_add_anon_rmap because the caller 648c97a9e10SNick Piggin * always holds the page locked, except if called from page_dup_rmap, 649c97a9e10SNick Piggin * in which case the page is already known to be setup. 650c97a9e10SNick Piggin * 651c97a9e10SNick Piggin * We have exclusion against page_add_new_anon_rmap because those pages 652c97a9e10SNick Piggin * are initially only visible via the pagetables, and the pte is locked 653c97a9e10SNick Piggin * over the call to page_add_new_anon_rmap. 654c97a9e10SNick Piggin */ 655c97a9e10SNick Piggin struct anon_vma *anon_vma = vma->anon_vma; 656c97a9e10SNick Piggin anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; 657c97a9e10SNick Piggin BUG_ON(page->mapping != (struct address_space *)anon_vma); 658c97a9e10SNick Piggin BUG_ON(page->index != linear_page_index(vma, address)); 659c97a9e10SNick Piggin #endif 660c97a9e10SNick Piggin } 661c97a9e10SNick Piggin 662c97a9e10SNick Piggin /** 6639617d95eSNick Piggin * page_add_anon_rmap - add pte mapping to an anonymous page 6649617d95eSNick Piggin * @page: the page to add the mapping to 6659617d95eSNick Piggin * @vma: the vm area in which the mapping is added 6669617d95eSNick Piggin * @address: the user virtual address mapped 6679617d95eSNick Piggin * 668c97a9e10SNick Piggin * The caller needs to hold the pte lock and the page must be locked. 6699617d95eSNick Piggin */ 6709617d95eSNick Piggin void page_add_anon_rmap(struct page *page, 6719617d95eSNick Piggin struct vm_area_struct *vma, unsigned long address) 6729617d95eSNick Piggin { 673c97a9e10SNick Piggin VM_BUG_ON(!PageLocked(page)); 674c97a9e10SNick Piggin VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); 6759617d95eSNick Piggin if (atomic_inc_and_test(&page->_mapcount)) 6769617d95eSNick Piggin __page_set_anon_rmap(page, vma, address); 67769029cd5SKAMEZAWA Hiroyuki else 678c97a9e10SNick Piggin __page_check_anon_rmap(page, vma, address); 6791da177e4SLinus Torvalds } 6801da177e4SLinus Torvalds 68143d8eac4SRandy Dunlap /** 6829617d95eSNick Piggin * page_add_new_anon_rmap - add pte mapping to a new anonymous page 6839617d95eSNick Piggin * @page: the page to add the mapping to 6849617d95eSNick Piggin * @vma: the vm area in which the mapping is added 6859617d95eSNick Piggin * @address: the user virtual address mapped 6869617d95eSNick Piggin * 6879617d95eSNick Piggin * Same as page_add_anon_rmap but must only be called on *new* pages. 6889617d95eSNick Piggin * This means the inc-and-test can be bypassed. 689c97a9e10SNick Piggin * Page does not have to be locked. 6909617d95eSNick Piggin */ 6919617d95eSNick Piggin void page_add_new_anon_rmap(struct page *page, 6929617d95eSNick Piggin struct vm_area_struct *vma, unsigned long address) 6939617d95eSNick Piggin { 694b5934c53SHugh Dickins VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); 695cbf84b7aSHugh Dickins SetPageSwapBacked(page); 696cbf84b7aSHugh Dickins atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */ 6979617d95eSNick Piggin __page_set_anon_rmap(page, vma, address); 698b5934c53SHugh Dickins if (page_evictable(page, vma)) 699cbf84b7aSHugh Dickins lru_cache_add_lru(page, LRU_ACTIVE_ANON); 700b5934c53SHugh Dickins else 701b5934c53SHugh Dickins add_page_to_unevictable_list(page); 7029617d95eSNick Piggin } 7039617d95eSNick Piggin 7041da177e4SLinus Torvalds /** 7051da177e4SLinus Torvalds * page_add_file_rmap - add pte mapping to a file page 7061da177e4SLinus Torvalds * @page: the page to add the mapping to 7071da177e4SLinus Torvalds * 708b8072f09SHugh Dickins * The caller needs to hold the pte lock. 7091da177e4SLinus Torvalds */ 7101da177e4SLinus Torvalds void page_add_file_rmap(struct page *page) 7111da177e4SLinus Torvalds { 712d69b042fSBalbir Singh if (atomic_inc_and_test(&page->_mapcount)) { 71365ba55f5SChristoph Lameter __inc_zone_page_state(page, NR_FILE_MAPPED); 714d69b042fSBalbir Singh mem_cgroup_update_mapped_file_stat(page, 1); 715d69b042fSBalbir Singh } 7161da177e4SLinus Torvalds } 7171da177e4SLinus Torvalds 7181da177e4SLinus Torvalds /** 7191da177e4SLinus Torvalds * page_remove_rmap - take down pte mapping from a page 7201da177e4SLinus Torvalds * @page: page to remove mapping from 7211da177e4SLinus Torvalds * 722b8072f09SHugh Dickins * The caller needs to hold the pte lock. 7231da177e4SLinus Torvalds */ 724edc315fdSHugh Dickins void page_remove_rmap(struct page *page) 7251da177e4SLinus Torvalds { 726b904dcfeSKOSAKI Motohiro /* page still mapped by someone else? */ 727b904dcfeSKOSAKI Motohiro if (!atomic_add_negative(-1, &page->_mapcount)) 728b904dcfeSKOSAKI Motohiro return; 729b904dcfeSKOSAKI Motohiro 7301da177e4SLinus Torvalds /* 73116f8c5b2SHugh Dickins * Now that the last pte has gone, s390 must transfer dirty 73216f8c5b2SHugh Dickins * flag from storage key to struct page. We can usually skip 73316f8c5b2SHugh Dickins * this if the page is anon, so about to be freed; but perhaps 73416f8c5b2SHugh Dickins * not if it's in swapcache - there might be another pte slot 73516f8c5b2SHugh Dickins * containing the swap entry, but page not yet written to swap. 73616f8c5b2SHugh Dickins */ 737b904dcfeSKOSAKI Motohiro if ((!PageAnon(page) || PageSwapCache(page)) && page_test_dirty(page)) { 73816f8c5b2SHugh Dickins page_clear_dirty(page); 73916f8c5b2SHugh Dickins set_page_dirty(page); 74016f8c5b2SHugh Dickins } 741b904dcfeSKOSAKI Motohiro if (PageAnon(page)) { 74216f8c5b2SHugh Dickins mem_cgroup_uncharge_page(page); 743b904dcfeSKOSAKI Motohiro __dec_zone_page_state(page, NR_ANON_PAGES); 744b904dcfeSKOSAKI Motohiro } else { 745b904dcfeSKOSAKI Motohiro __dec_zone_page_state(page, NR_FILE_MAPPED); 746b904dcfeSKOSAKI Motohiro } 747d69b042fSBalbir Singh mem_cgroup_update_mapped_file_stat(page, -1); 74816f8c5b2SHugh Dickins /* 7491da177e4SLinus Torvalds * It would be tidy to reset the PageAnon mapping here, 7501da177e4SLinus Torvalds * but that might overwrite a racing page_add_anon_rmap 7511da177e4SLinus Torvalds * which increments mapcount after us but sets mapping 7521da177e4SLinus Torvalds * before us: so leave the reset to free_hot_cold_page, 7531da177e4SLinus Torvalds * and remember that it's only reliable while mapped. 7541da177e4SLinus Torvalds * Leaving it set also helps swapoff to reinstate ptes 7551da177e4SLinus Torvalds * faster for those pages still in swapcache. 7561da177e4SLinus Torvalds */ 7571da177e4SLinus Torvalds } 7581da177e4SLinus Torvalds 7591da177e4SLinus Torvalds /* 7601da177e4SLinus Torvalds * Subfunctions of try_to_unmap: try_to_unmap_one called 7611da177e4SLinus Torvalds * repeatedly from either try_to_unmap_anon or try_to_unmap_file. 7621da177e4SLinus Torvalds */ 763a48d07afSChristoph Lameter static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, 76414fa31b8SAndi Kleen enum ttu_flags flags) 7651da177e4SLinus Torvalds { 7661da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 7671da177e4SLinus Torvalds unsigned long address; 7681da177e4SLinus Torvalds pte_t *pte; 7691da177e4SLinus Torvalds pte_t pteval; 770c0718806SHugh Dickins spinlock_t *ptl; 7711da177e4SLinus Torvalds int ret = SWAP_AGAIN; 7721da177e4SLinus Torvalds 7731da177e4SLinus Torvalds address = vma_address(page, vma); 7741da177e4SLinus Torvalds if (address == -EFAULT) 7751da177e4SLinus Torvalds goto out; 7761da177e4SLinus Torvalds 777479db0bfSNick Piggin pte = page_check_address(page, mm, address, &ptl, 0); 778c0718806SHugh Dickins if (!pte) 77981b4082dSNikita Danilov goto out; 7801da177e4SLinus Torvalds 7811da177e4SLinus Torvalds /* 7821da177e4SLinus Torvalds * If the page is mlock()d, we cannot swap it out. 7831da177e4SLinus Torvalds * If it's recently referenced (perhaps page_referenced 7841da177e4SLinus Torvalds * skipped over this mm) then we should reactivate it. 7851da177e4SLinus Torvalds */ 78614fa31b8SAndi Kleen if (!(flags & TTU_IGNORE_MLOCK)) { 787b291f000SNick Piggin if (vma->vm_flags & VM_LOCKED) { 788b291f000SNick Piggin ret = SWAP_MLOCK; 789b291f000SNick Piggin goto out_unmap; 790b291f000SNick Piggin } 79114fa31b8SAndi Kleen } 79214fa31b8SAndi Kleen if (!(flags & TTU_IGNORE_ACCESS)) { 793b291f000SNick Piggin if (ptep_clear_flush_young_notify(vma, address, pte)) { 7941da177e4SLinus Torvalds ret = SWAP_FAIL; 7951da177e4SLinus Torvalds goto out_unmap; 7961da177e4SLinus Torvalds } 797b291f000SNick Piggin } 7981da177e4SLinus Torvalds 7991da177e4SLinus Torvalds /* Nuke the page table entry. */ 8001da177e4SLinus Torvalds flush_cache_page(vma, address, page_to_pfn(page)); 801cddb8a5cSAndrea Arcangeli pteval = ptep_clear_flush_notify(vma, address, pte); 8021da177e4SLinus Torvalds 8031da177e4SLinus Torvalds /* Move the dirty bit to the physical page now the pte is gone. */ 8041da177e4SLinus Torvalds if (pte_dirty(pteval)) 8051da177e4SLinus Torvalds set_page_dirty(page); 8061da177e4SLinus Torvalds 807365e9c87SHugh Dickins /* Update high watermark before we lower rss */ 808365e9c87SHugh Dickins update_hiwater_rss(mm); 809365e9c87SHugh Dickins 810888b9f7cSAndi Kleen if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) { 811888b9f7cSAndi Kleen if (PageAnon(page)) 812888b9f7cSAndi Kleen dec_mm_counter(mm, anon_rss); 813888b9f7cSAndi Kleen else 814888b9f7cSAndi Kleen dec_mm_counter(mm, file_rss); 815888b9f7cSAndi Kleen set_pte_at(mm, address, pte, 816888b9f7cSAndi Kleen swp_entry_to_pte(make_hwpoison_entry(page))); 817888b9f7cSAndi Kleen } else if (PageAnon(page)) { 8184c21e2f2SHugh Dickins swp_entry_t entry = { .val = page_private(page) }; 8190697212aSChristoph Lameter 8200697212aSChristoph Lameter if (PageSwapCache(page)) { 8211da177e4SLinus Torvalds /* 8221da177e4SLinus Torvalds * Store the swap location in the pte. 8231da177e4SLinus Torvalds * See handle_pte_fault() ... 8241da177e4SLinus Torvalds */ 825570a335bSHugh Dickins if (swap_duplicate(entry) < 0) { 826570a335bSHugh Dickins set_pte_at(mm, address, pte, pteval); 827570a335bSHugh Dickins ret = SWAP_FAIL; 828570a335bSHugh Dickins goto out_unmap; 829570a335bSHugh Dickins } 8301da177e4SLinus Torvalds if (list_empty(&mm->mmlist)) { 8311da177e4SLinus Torvalds spin_lock(&mmlist_lock); 832f412ac08SHugh Dickins if (list_empty(&mm->mmlist)) 8331da177e4SLinus Torvalds list_add(&mm->mmlist, &init_mm.mmlist); 8341da177e4SLinus Torvalds spin_unlock(&mmlist_lock); 8351da177e4SLinus Torvalds } 836442c9137SChristoph Lameter dec_mm_counter(mm, anon_rss); 83764cdd548SKOSAKI Motohiro } else if (PAGE_MIGRATION) { 8380697212aSChristoph Lameter /* 8390697212aSChristoph Lameter * Store the pfn of the page in a special migration 8400697212aSChristoph Lameter * pte. do_swap_page() will wait until the migration 8410697212aSChristoph Lameter * pte is removed and then restart fault handling. 8420697212aSChristoph Lameter */ 84314fa31b8SAndi Kleen BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION); 8440697212aSChristoph Lameter entry = make_migration_entry(page, pte_write(pteval)); 8450697212aSChristoph Lameter } 8461da177e4SLinus Torvalds set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); 8471da177e4SLinus Torvalds BUG_ON(pte_file(*pte)); 84814fa31b8SAndi Kleen } else if (PAGE_MIGRATION && (TTU_ACTION(flags) == TTU_MIGRATION)) { 84904e62a29SChristoph Lameter /* Establish migration entry for a file page */ 85004e62a29SChristoph Lameter swp_entry_t entry; 85104e62a29SChristoph Lameter entry = make_migration_entry(page, pte_write(pteval)); 85204e62a29SChristoph Lameter set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); 85304e62a29SChristoph Lameter } else 8544294621fSHugh Dickins dec_mm_counter(mm, file_rss); 8551da177e4SLinus Torvalds 85604e62a29SChristoph Lameter 857edc315fdSHugh Dickins page_remove_rmap(page); 8581da177e4SLinus Torvalds page_cache_release(page); 8591da177e4SLinus Torvalds 8601da177e4SLinus Torvalds out_unmap: 861c0718806SHugh Dickins pte_unmap_unlock(pte, ptl); 8621da177e4SLinus Torvalds out: 8631da177e4SLinus Torvalds return ret; 8641da177e4SLinus Torvalds } 8651da177e4SLinus Torvalds 8661da177e4SLinus Torvalds /* 8671da177e4SLinus Torvalds * objrmap doesn't work for nonlinear VMAs because the assumption that 8681da177e4SLinus Torvalds * offset-into-file correlates with offset-into-virtual-addresses does not hold. 8691da177e4SLinus Torvalds * Consequently, given a particular page and its ->index, we cannot locate the 8701da177e4SLinus Torvalds * ptes which are mapping that page without an exhaustive linear search. 8711da177e4SLinus Torvalds * 8721da177e4SLinus Torvalds * So what this code does is a mini "virtual scan" of each nonlinear VMA which 8731da177e4SLinus Torvalds * maps the file to which the target page belongs. The ->vm_private_data field 8741da177e4SLinus Torvalds * holds the current cursor into that scan. Successive searches will circulate 8751da177e4SLinus Torvalds * around the vma's virtual address space. 8761da177e4SLinus Torvalds * 8771da177e4SLinus Torvalds * So as more replacement pressure is applied to the pages in a nonlinear VMA, 8781da177e4SLinus Torvalds * more scanning pressure is placed against them as well. Eventually pages 8791da177e4SLinus Torvalds * will become fully unmapped and are eligible for eviction. 8801da177e4SLinus Torvalds * 8811da177e4SLinus Torvalds * For very sparsely populated VMAs this is a little inefficient - chances are 8821da177e4SLinus Torvalds * there there won't be many ptes located within the scan cluster. In this case 8831da177e4SLinus Torvalds * maybe we could scan further - to the end of the pte page, perhaps. 884b291f000SNick Piggin * 885b291f000SNick Piggin * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can 886b291f000SNick Piggin * acquire it without blocking. If vma locked, mlock the pages in the cluster, 887b291f000SNick Piggin * rather than unmapping them. If we encounter the "check_page" that vmscan is 888b291f000SNick Piggin * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN. 8891da177e4SLinus Torvalds */ 8901da177e4SLinus Torvalds #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE) 8911da177e4SLinus Torvalds #define CLUSTER_MASK (~(CLUSTER_SIZE - 1)) 8921da177e4SLinus Torvalds 893b291f000SNick Piggin static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount, 894b291f000SNick Piggin struct vm_area_struct *vma, struct page *check_page) 8951da177e4SLinus Torvalds { 8961da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 8971da177e4SLinus Torvalds pgd_t *pgd; 8981da177e4SLinus Torvalds pud_t *pud; 8991da177e4SLinus Torvalds pmd_t *pmd; 900c0718806SHugh Dickins pte_t *pte; 9011da177e4SLinus Torvalds pte_t pteval; 902c0718806SHugh Dickins spinlock_t *ptl; 9031da177e4SLinus Torvalds struct page *page; 9041da177e4SLinus Torvalds unsigned long address; 9051da177e4SLinus Torvalds unsigned long end; 906b291f000SNick Piggin int ret = SWAP_AGAIN; 907b291f000SNick Piggin int locked_vma = 0; 9081da177e4SLinus Torvalds 9091da177e4SLinus Torvalds address = (vma->vm_start + cursor) & CLUSTER_MASK; 9101da177e4SLinus Torvalds end = address + CLUSTER_SIZE; 9111da177e4SLinus Torvalds if (address < vma->vm_start) 9121da177e4SLinus Torvalds address = vma->vm_start; 9131da177e4SLinus Torvalds if (end > vma->vm_end) 9141da177e4SLinus Torvalds end = vma->vm_end; 9151da177e4SLinus Torvalds 9161da177e4SLinus Torvalds pgd = pgd_offset(mm, address); 9171da177e4SLinus Torvalds if (!pgd_present(*pgd)) 918b291f000SNick Piggin return ret; 9191da177e4SLinus Torvalds 9201da177e4SLinus Torvalds pud = pud_offset(pgd, address); 9211da177e4SLinus Torvalds if (!pud_present(*pud)) 922b291f000SNick Piggin return ret; 9231da177e4SLinus Torvalds 9241da177e4SLinus Torvalds pmd = pmd_offset(pud, address); 9251da177e4SLinus Torvalds if (!pmd_present(*pmd)) 926b291f000SNick Piggin return ret; 927b291f000SNick Piggin 928b291f000SNick Piggin /* 929b291f000SNick Piggin * MLOCK_PAGES => feature is configured. 930b291f000SNick Piggin * if we can acquire the mmap_sem for read, and vma is VM_LOCKED, 931b291f000SNick Piggin * keep the sem while scanning the cluster for mlocking pages. 932b291f000SNick Piggin */ 933b291f000SNick Piggin if (MLOCK_PAGES && down_read_trylock(&vma->vm_mm->mmap_sem)) { 934b291f000SNick Piggin locked_vma = (vma->vm_flags & VM_LOCKED); 935b291f000SNick Piggin if (!locked_vma) 936b291f000SNick Piggin up_read(&vma->vm_mm->mmap_sem); /* don't need it */ 937b291f000SNick Piggin } 938c0718806SHugh Dickins 939c0718806SHugh Dickins pte = pte_offset_map_lock(mm, pmd, address, &ptl); 9401da177e4SLinus Torvalds 941365e9c87SHugh Dickins /* Update high watermark before we lower rss */ 942365e9c87SHugh Dickins update_hiwater_rss(mm); 943365e9c87SHugh Dickins 944c0718806SHugh Dickins for (; address < end; pte++, address += PAGE_SIZE) { 9451da177e4SLinus Torvalds if (!pte_present(*pte)) 9461da177e4SLinus Torvalds continue; 9476aab341eSLinus Torvalds page = vm_normal_page(vma, address, *pte); 9486aab341eSLinus Torvalds BUG_ON(!page || PageAnon(page)); 9491da177e4SLinus Torvalds 950b291f000SNick Piggin if (locked_vma) { 951b291f000SNick Piggin mlock_vma_page(page); /* no-op if already mlocked */ 952b291f000SNick Piggin if (page == check_page) 953b291f000SNick Piggin ret = SWAP_MLOCK; 954b291f000SNick Piggin continue; /* don't unmap */ 955b291f000SNick Piggin } 956b291f000SNick Piggin 957cddb8a5cSAndrea Arcangeli if (ptep_clear_flush_young_notify(vma, address, pte)) 9581da177e4SLinus Torvalds continue; 9591da177e4SLinus Torvalds 9601da177e4SLinus Torvalds /* Nuke the page table entry. */ 961eca35133SBen Collins flush_cache_page(vma, address, pte_pfn(*pte)); 962cddb8a5cSAndrea Arcangeli pteval = ptep_clear_flush_notify(vma, address, pte); 9631da177e4SLinus Torvalds 9641da177e4SLinus Torvalds /* If nonlinear, store the file page offset in the pte. */ 9651da177e4SLinus Torvalds if (page->index != linear_page_index(vma, address)) 9661da177e4SLinus Torvalds set_pte_at(mm, address, pte, pgoff_to_pte(page->index)); 9671da177e4SLinus Torvalds 9681da177e4SLinus Torvalds /* Move the dirty bit to the physical page now the pte is gone. */ 9691da177e4SLinus Torvalds if (pte_dirty(pteval)) 9701da177e4SLinus Torvalds set_page_dirty(page); 9711da177e4SLinus Torvalds 972edc315fdSHugh Dickins page_remove_rmap(page); 9731da177e4SLinus Torvalds page_cache_release(page); 9744294621fSHugh Dickins dec_mm_counter(mm, file_rss); 9751da177e4SLinus Torvalds (*mapcount)--; 9761da177e4SLinus Torvalds } 977c0718806SHugh Dickins pte_unmap_unlock(pte - 1, ptl); 978b291f000SNick Piggin if (locked_vma) 979b291f000SNick Piggin up_read(&vma->vm_mm->mmap_sem); 980b291f000SNick Piggin return ret; 9811da177e4SLinus Torvalds } 9821da177e4SLinus Torvalds 983b291f000SNick Piggin /* 984b291f000SNick Piggin * common handling for pages mapped in VM_LOCKED vmas 985b291f000SNick Piggin */ 986b291f000SNick Piggin static int try_to_mlock_page(struct page *page, struct vm_area_struct *vma) 987b291f000SNick Piggin { 988b291f000SNick Piggin int mlocked = 0; 989b291f000SNick Piggin 990b291f000SNick Piggin if (down_read_trylock(&vma->vm_mm->mmap_sem)) { 991b291f000SNick Piggin if (vma->vm_flags & VM_LOCKED) { 992b291f000SNick Piggin mlock_vma_page(page); 993b291f000SNick Piggin mlocked++; /* really mlocked the page */ 994b291f000SNick Piggin } 995b291f000SNick Piggin up_read(&vma->vm_mm->mmap_sem); 996b291f000SNick Piggin } 997b291f000SNick Piggin return mlocked; 998b291f000SNick Piggin } 999b291f000SNick Piggin 1000b291f000SNick Piggin /** 1001b291f000SNick Piggin * try_to_unmap_anon - unmap or unlock anonymous page using the object-based 1002b291f000SNick Piggin * rmap method 1003b291f000SNick Piggin * @page: the page to unmap/unlock 10048051be5eSHuang Shijie * @flags: action and flags 1005b291f000SNick Piggin * 1006b291f000SNick Piggin * Find all the mappings of a page using the mapping pointer and the vma chains 1007b291f000SNick Piggin * contained in the anon_vma struct it points to. 1008b291f000SNick Piggin * 1009b291f000SNick Piggin * This function is only called from try_to_unmap/try_to_munlock for 1010b291f000SNick Piggin * anonymous pages. 1011b291f000SNick Piggin * When called from try_to_munlock(), the mmap_sem of the mm containing the vma 1012b291f000SNick Piggin * where the page was found will be held for write. So, we won't recheck 1013b291f000SNick Piggin * vm_flags for that VMA. That should be OK, because that vma shouldn't be 1014b291f000SNick Piggin * 'LOCKED. 1015b291f000SNick Piggin */ 101614fa31b8SAndi Kleen static int try_to_unmap_anon(struct page *page, enum ttu_flags flags) 10171da177e4SLinus Torvalds { 10181da177e4SLinus Torvalds struct anon_vma *anon_vma; 10191da177e4SLinus Torvalds struct vm_area_struct *vma; 1020b291f000SNick Piggin unsigned int mlocked = 0; 10211da177e4SLinus Torvalds int ret = SWAP_AGAIN; 102214fa31b8SAndi Kleen int unlock = TTU_ACTION(flags) == TTU_MUNLOCK; 10231da177e4SLinus Torvalds 1024b291f000SNick Piggin if (MLOCK_PAGES && unlikely(unlock)) 1025b291f000SNick Piggin ret = SWAP_SUCCESS; /* default for try_to_munlock() */ 1026b291f000SNick Piggin 10271da177e4SLinus Torvalds anon_vma = page_lock_anon_vma(page); 10281da177e4SLinus Torvalds if (!anon_vma) 10291da177e4SLinus Torvalds return ret; 10301da177e4SLinus Torvalds 10311da177e4SLinus Torvalds list_for_each_entry(vma, &anon_vma->head, anon_vma_node) { 1032b291f000SNick Piggin if (MLOCK_PAGES && unlikely(unlock)) { 1033b291f000SNick Piggin if (!((vma->vm_flags & VM_LOCKED) && 1034b291f000SNick Piggin page_mapped_in_vma(page, vma))) 1035b291f000SNick Piggin continue; /* must visit all unlocked vmas */ 1036b291f000SNick Piggin ret = SWAP_MLOCK; /* saw at least one mlocked vma */ 1037b291f000SNick Piggin } else { 103814fa31b8SAndi Kleen ret = try_to_unmap_one(page, vma, flags); 10391da177e4SLinus Torvalds if (ret == SWAP_FAIL || !page_mapped(page)) 10401da177e4SLinus Torvalds break; 10411da177e4SLinus Torvalds } 1042b291f000SNick Piggin if (ret == SWAP_MLOCK) { 1043b291f000SNick Piggin mlocked = try_to_mlock_page(page, vma); 1044b291f000SNick Piggin if (mlocked) 1045b291f000SNick Piggin break; /* stop if actually mlocked page */ 1046b291f000SNick Piggin } 1047b291f000SNick Piggin } 104834bbd704SOleg Nesterov 104934bbd704SOleg Nesterov page_unlock_anon_vma(anon_vma); 1050b291f000SNick Piggin 1051b291f000SNick Piggin if (mlocked) 1052b291f000SNick Piggin ret = SWAP_MLOCK; /* actually mlocked the page */ 1053b291f000SNick Piggin else if (ret == SWAP_MLOCK) 1054b291f000SNick Piggin ret = SWAP_AGAIN; /* saw VM_LOCKED vma */ 1055b291f000SNick Piggin 10561da177e4SLinus Torvalds return ret; 10571da177e4SLinus Torvalds } 10581da177e4SLinus Torvalds 10591da177e4SLinus Torvalds /** 1060b291f000SNick Piggin * try_to_unmap_file - unmap/unlock file page using the object-based rmap method 1061b291f000SNick Piggin * @page: the page to unmap/unlock 106214fa31b8SAndi Kleen * @flags: action and flags 10631da177e4SLinus Torvalds * 10641da177e4SLinus Torvalds * Find all the mappings of a page using the mapping pointer and the vma chains 10651da177e4SLinus Torvalds * contained in the address_space struct it points to. 10661da177e4SLinus Torvalds * 1067b291f000SNick Piggin * This function is only called from try_to_unmap/try_to_munlock for 1068b291f000SNick Piggin * object-based pages. 1069b291f000SNick Piggin * When called from try_to_munlock(), the mmap_sem of the mm containing the vma 1070b291f000SNick Piggin * where the page was found will be held for write. So, we won't recheck 1071b291f000SNick Piggin * vm_flags for that VMA. That should be OK, because that vma shouldn't be 1072b291f000SNick Piggin * 'LOCKED. 10731da177e4SLinus Torvalds */ 107414fa31b8SAndi Kleen static int try_to_unmap_file(struct page *page, enum ttu_flags flags) 10751da177e4SLinus Torvalds { 10761da177e4SLinus Torvalds struct address_space *mapping = page->mapping; 10771da177e4SLinus Torvalds pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); 10781da177e4SLinus Torvalds struct vm_area_struct *vma; 10791da177e4SLinus Torvalds struct prio_tree_iter iter; 10801da177e4SLinus Torvalds int ret = SWAP_AGAIN; 10811da177e4SLinus Torvalds unsigned long cursor; 10821da177e4SLinus Torvalds unsigned long max_nl_cursor = 0; 10831da177e4SLinus Torvalds unsigned long max_nl_size = 0; 10841da177e4SLinus Torvalds unsigned int mapcount; 1085b291f000SNick Piggin unsigned int mlocked = 0; 108614fa31b8SAndi Kleen int unlock = TTU_ACTION(flags) == TTU_MUNLOCK; 1087b291f000SNick Piggin 1088b291f000SNick Piggin if (MLOCK_PAGES && unlikely(unlock)) 1089b291f000SNick Piggin ret = SWAP_SUCCESS; /* default for try_to_munlock() */ 10901da177e4SLinus Torvalds 10911da177e4SLinus Torvalds spin_lock(&mapping->i_mmap_lock); 10921da177e4SLinus Torvalds vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { 1093b291f000SNick Piggin if (MLOCK_PAGES && unlikely(unlock)) { 1094508b9f8eSMinChan Kim if (!((vma->vm_flags & VM_LOCKED) && 1095508b9f8eSMinChan Kim page_mapped_in_vma(page, vma))) 1096b291f000SNick Piggin continue; /* must visit all vmas */ 1097b291f000SNick Piggin ret = SWAP_MLOCK; 1098b291f000SNick Piggin } else { 109914fa31b8SAndi Kleen ret = try_to_unmap_one(page, vma, flags); 11001da177e4SLinus Torvalds if (ret == SWAP_FAIL || !page_mapped(page)) 11011da177e4SLinus Torvalds goto out; 11021da177e4SLinus Torvalds } 1103b291f000SNick Piggin if (ret == SWAP_MLOCK) { 1104b291f000SNick Piggin mlocked = try_to_mlock_page(page, vma); 1105b291f000SNick Piggin if (mlocked) 11067b511594SHuang Shijie goto out; /* stop if actually mlocked page */ 1107b291f000SNick Piggin } 1108b291f000SNick Piggin } 1109b291f000SNick Piggin 11101da177e4SLinus Torvalds if (list_empty(&mapping->i_mmap_nonlinear)) 11111da177e4SLinus Torvalds goto out; 11121da177e4SLinus Torvalds 11131da177e4SLinus Torvalds list_for_each_entry(vma, &mapping->i_mmap_nonlinear, 11141da177e4SLinus Torvalds shared.vm_set.list) { 1115b291f000SNick Piggin if (MLOCK_PAGES && unlikely(unlock)) { 1116b291f000SNick Piggin if (!(vma->vm_flags & VM_LOCKED)) 1117b291f000SNick Piggin continue; /* must visit all vmas */ 1118b291f000SNick Piggin ret = SWAP_MLOCK; /* leave mlocked == 0 */ 1119b291f000SNick Piggin goto out; /* no need to look further */ 1120b291f000SNick Piggin } 112114fa31b8SAndi Kleen if (!MLOCK_PAGES && !(flags & TTU_IGNORE_MLOCK) && 112214fa31b8SAndi Kleen (vma->vm_flags & VM_LOCKED)) 11231da177e4SLinus Torvalds continue; 11241da177e4SLinus Torvalds cursor = (unsigned long) vma->vm_private_data; 11251da177e4SLinus Torvalds if (cursor > max_nl_cursor) 11261da177e4SLinus Torvalds max_nl_cursor = cursor; 11271da177e4SLinus Torvalds cursor = vma->vm_end - vma->vm_start; 11281da177e4SLinus Torvalds if (cursor > max_nl_size) 11291da177e4SLinus Torvalds max_nl_size = cursor; 11301da177e4SLinus Torvalds } 11311da177e4SLinus Torvalds 1132b291f000SNick Piggin if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */ 11331da177e4SLinus Torvalds ret = SWAP_FAIL; 11341da177e4SLinus Torvalds goto out; 11351da177e4SLinus Torvalds } 11361da177e4SLinus Torvalds 11371da177e4SLinus Torvalds /* 11381da177e4SLinus Torvalds * We don't try to search for this page in the nonlinear vmas, 11391da177e4SLinus Torvalds * and page_referenced wouldn't have found it anyway. Instead 11401da177e4SLinus Torvalds * just walk the nonlinear vmas trying to age and unmap some. 11411da177e4SLinus Torvalds * The mapcount of the page we came in with is irrelevant, 11421da177e4SLinus Torvalds * but even so use it as a guide to how hard we should try? 11431da177e4SLinus Torvalds */ 11441da177e4SLinus Torvalds mapcount = page_mapcount(page); 11451da177e4SLinus Torvalds if (!mapcount) 11461da177e4SLinus Torvalds goto out; 11471da177e4SLinus Torvalds cond_resched_lock(&mapping->i_mmap_lock); 11481da177e4SLinus Torvalds 11491da177e4SLinus Torvalds max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK; 11501da177e4SLinus Torvalds if (max_nl_cursor == 0) 11511da177e4SLinus Torvalds max_nl_cursor = CLUSTER_SIZE; 11521da177e4SLinus Torvalds 11531da177e4SLinus Torvalds do { 11541da177e4SLinus Torvalds list_for_each_entry(vma, &mapping->i_mmap_nonlinear, 11551da177e4SLinus Torvalds shared.vm_set.list) { 115614fa31b8SAndi Kleen if (!MLOCK_PAGES && !(flags & TTU_IGNORE_MLOCK) && 1157b291f000SNick Piggin (vma->vm_flags & VM_LOCKED)) 11581da177e4SLinus Torvalds continue; 11591da177e4SLinus Torvalds cursor = (unsigned long) vma->vm_private_data; 1160839b9685SHugh Dickins while ( cursor < max_nl_cursor && 11611da177e4SLinus Torvalds cursor < vma->vm_end - vma->vm_start) { 1162b291f000SNick Piggin ret = try_to_unmap_cluster(cursor, &mapcount, 1163b291f000SNick Piggin vma, page); 1164b291f000SNick Piggin if (ret == SWAP_MLOCK) 1165b291f000SNick Piggin mlocked = 2; /* to return below */ 11661da177e4SLinus Torvalds cursor += CLUSTER_SIZE; 11671da177e4SLinus Torvalds vma->vm_private_data = (void *) cursor; 11681da177e4SLinus Torvalds if ((int)mapcount <= 0) 11691da177e4SLinus Torvalds goto out; 11701da177e4SLinus Torvalds } 11711da177e4SLinus Torvalds vma->vm_private_data = (void *) max_nl_cursor; 11721da177e4SLinus Torvalds } 11731da177e4SLinus Torvalds cond_resched_lock(&mapping->i_mmap_lock); 11741da177e4SLinus Torvalds max_nl_cursor += CLUSTER_SIZE; 11751da177e4SLinus Torvalds } while (max_nl_cursor <= max_nl_size); 11761da177e4SLinus Torvalds 11771da177e4SLinus Torvalds /* 11781da177e4SLinus Torvalds * Don't loop forever (perhaps all the remaining pages are 11791da177e4SLinus Torvalds * in locked vmas). Reset cursor on all unreserved nonlinear 11801da177e4SLinus Torvalds * vmas, now forgetting on which ones it had fallen behind. 11811da177e4SLinus Torvalds */ 1182101d2be7SHugh Dickins list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) 11831da177e4SLinus Torvalds vma->vm_private_data = NULL; 11841da177e4SLinus Torvalds out: 11851da177e4SLinus Torvalds spin_unlock(&mapping->i_mmap_lock); 1186b291f000SNick Piggin if (mlocked) 1187b291f000SNick Piggin ret = SWAP_MLOCK; /* actually mlocked the page */ 1188b291f000SNick Piggin else if (ret == SWAP_MLOCK) 1189b291f000SNick Piggin ret = SWAP_AGAIN; /* saw VM_LOCKED vma */ 11901da177e4SLinus Torvalds return ret; 11911da177e4SLinus Torvalds } 11921da177e4SLinus Torvalds 11931da177e4SLinus Torvalds /** 11941da177e4SLinus Torvalds * try_to_unmap - try to remove all page table mappings to a page 11951da177e4SLinus Torvalds * @page: the page to get unmapped 119614fa31b8SAndi Kleen * @flags: action and flags 11971da177e4SLinus Torvalds * 11981da177e4SLinus Torvalds * Tries to remove all the page table entries which are mapping this 11991da177e4SLinus Torvalds * page, used in the pageout path. Caller must hold the page lock. 12001da177e4SLinus Torvalds * Return values are: 12011da177e4SLinus Torvalds * 12021da177e4SLinus Torvalds * SWAP_SUCCESS - we succeeded in removing all mappings 12031da177e4SLinus Torvalds * SWAP_AGAIN - we missed a mapping, try again later 12041da177e4SLinus Torvalds * SWAP_FAIL - the page is unswappable 1205b291f000SNick Piggin * SWAP_MLOCK - page is mlocked. 12061da177e4SLinus Torvalds */ 120714fa31b8SAndi Kleen int try_to_unmap(struct page *page, enum ttu_flags flags) 12081da177e4SLinus Torvalds { 12091da177e4SLinus Torvalds int ret; 12101da177e4SLinus Torvalds 12111da177e4SLinus Torvalds BUG_ON(!PageLocked(page)); 12121da177e4SLinus Torvalds 12131da177e4SLinus Torvalds if (PageAnon(page)) 121414fa31b8SAndi Kleen ret = try_to_unmap_anon(page, flags); 12151da177e4SLinus Torvalds else 121614fa31b8SAndi Kleen ret = try_to_unmap_file(page, flags); 1217b291f000SNick Piggin if (ret != SWAP_MLOCK && !page_mapped(page)) 12181da177e4SLinus Torvalds ret = SWAP_SUCCESS; 12191da177e4SLinus Torvalds return ret; 12201da177e4SLinus Torvalds } 122181b4082dSNikita Danilov 1222b291f000SNick Piggin /** 1223b291f000SNick Piggin * try_to_munlock - try to munlock a page 1224b291f000SNick Piggin * @page: the page to be munlocked 1225b291f000SNick Piggin * 1226b291f000SNick Piggin * Called from munlock code. Checks all of the VMAs mapping the page 1227b291f000SNick Piggin * to make sure nobody else has this page mlocked. The page will be 1228b291f000SNick Piggin * returned with PG_mlocked cleared if no other vmas have it mlocked. 1229b291f000SNick Piggin * 1230b291f000SNick Piggin * Return values are: 1231b291f000SNick Piggin * 1232b291f000SNick Piggin * SWAP_SUCCESS - no vma's holding page mlocked. 1233b291f000SNick Piggin * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem 1234b291f000SNick Piggin * SWAP_MLOCK - page is now mlocked. 1235b291f000SNick Piggin */ 1236b291f000SNick Piggin int try_to_munlock(struct page *page) 1237b291f000SNick Piggin { 1238b291f000SNick Piggin VM_BUG_ON(!PageLocked(page) || PageLRU(page)); 1239b291f000SNick Piggin 1240b291f000SNick Piggin if (PageAnon(page)) 124114fa31b8SAndi Kleen return try_to_unmap_anon(page, TTU_MUNLOCK); 1242b291f000SNick Piggin else 124314fa31b8SAndi Kleen return try_to_unmap_file(page, TTU_MUNLOCK); 1244b291f000SNick Piggin } 124568377659SKOSAKI Motohiro 1246