blob: 0feeef860a8f5b5d61081234f40e8e149e2696f5 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * mm/rmap.c - physical to virtual reverse mappings
3 *
4 * Copyright 2001, Rik van Riel <riel@conectiva.com.br>
5 * Released under the General Public License (GPL).
6 *
7 * Simple, low overhead reverse mapping scheme.
8 * Please try to keep this thing as modular as possible.
9 *
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.
13 *
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
Hugh Dickins98f32602009-05-21 20:33:58 +010017 * Contributions by Hugh Dickins 2003, 2004
Linus Torvalds1da177e2005-04-16 15:20:36 -070018 */
19
20/*
21 * Lock ordering in mm:
22 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080023 * inode->i_mutex (while writing or truncating, not reading or faulting)
Nick Piggin82591e62006-10-19 23:29:10 -070024 * inode->i_alloc_sem (vmtruncate_range)
25 * mm->mmap_sem
26 * page->flags PG_locked (lock_page)
27 * mapping->i_mmap_lock
28 * anon_vma->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)
Andi Kleen6a460792009-09-16 11:50:15 +020039 *
40 * (code doesn't rely on that order so it could be switched around)
41 * ->tasklist_lock
42 * anon_vma->lock (memory_failure, collect_procs_anon)
43 * pte map lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070044 */
45
46#include <linux/mm.h>
47#include <linux/pagemap.h>
48#include <linux/swap.h>
49#include <linux/swapops.h>
50#include <linux/slab.h>
51#include <linux/init.h>
Hugh Dickins5ad64682009-12-14 17:59:24 -080052#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070053#include <linux/rmap.h>
54#include <linux/rcupdate.h>
Christoph Lametera48d07a2006-02-01 03:05:38 -080055#include <linux/module.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080056#include <linux/memcontrol.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070057#include <linux/mmu_notifier.h>
KOSAKI Motohiro64cdd542009-01-06 14:39:16 -080058#include <linux/migrate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
60#include <asm/tlbflush.h>
61
Nick Pigginb291f002008-10-18 20:26:44 -070062#include "internal.h"
63
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070064static struct kmem_cache *anon_vma_cachep;
Rik van Riel5beb4932010-03-05 13:42:07 -080065static struct kmem_cache *anon_vma_chain_cachep;
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070066
67static inline struct anon_vma *anon_vma_alloc(void)
68{
69 return kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL);
70}
71
Hugh Dickinsdb114b82009-12-14 17:59:25 -080072void anon_vma_free(struct anon_vma *anon_vma)
Adrian Bunkfdd2e5f2008-10-18 20:28:38 -070073{
74 kmem_cache_free(anon_vma_cachep, anon_vma);
75}
Linus Torvalds1da177e2005-04-16 15:20:36 -070076
Rik van Riel5beb4932010-03-05 13:42:07 -080077static inline struct anon_vma_chain *anon_vma_chain_alloc(void)
78{
79 return kmem_cache_alloc(anon_vma_chain_cachep, GFP_KERNEL);
80}
81
82void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain)
83{
84 kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain);
85}
86
Linus Torvaldsd9d332e2008-10-19 10:32:20 -070087/**
88 * anon_vma_prepare - attach an anon_vma to a memory region
89 * @vma: the memory region in question
90 *
91 * This makes sure the memory mapping described by 'vma' has
92 * an 'anon_vma' attached to it, so that we can associate the
93 * anonymous pages mapped into it with that anon_vma.
94 *
95 * The common case will be that we already have one, but if
96 * if not we either need to find an adjacent mapping that we
97 * can re-use the anon_vma from (very common when the only
98 * reason for splitting a vma has been mprotect()), or we
99 * allocate a new one.
100 *
101 * Anon-vma allocations are very subtle, because we may have
102 * optimistically looked up an anon_vma in page_lock_anon_vma()
103 * and that may actually touch the spinlock even in the newly
104 * allocated vma (it depends on RCU to make sure that the
105 * anon_vma isn't actually destroyed).
106 *
107 * As a result, we need to do proper anon_vma locking even
108 * for the new allocation. At the same time, we do not want
109 * to do any locking for the common case of already having
110 * an anon_vma.
111 *
112 * This must be called with the mmap_sem held for reading.
113 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114int anon_vma_prepare(struct vm_area_struct *vma)
115{
116 struct anon_vma *anon_vma = vma->anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800117 struct anon_vma_chain *avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118
119 might_sleep();
120 if (unlikely(!anon_vma)) {
121 struct mm_struct *mm = vma->vm_mm;
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700122 struct anon_vma *allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123
Rik van Riel5beb4932010-03-05 13:42:07 -0800124 avc = anon_vma_chain_alloc();
125 if (!avc)
126 goto out_enomem;
127
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 anon_vma = find_mergeable_anon_vma(vma);
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700129 allocated = NULL;
130 if (!anon_vma) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 anon_vma = anon_vma_alloc();
132 if (unlikely(!anon_vma))
Rik van Riel5beb4932010-03-05 13:42:07 -0800133 goto out_enomem_free_avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 allocated = anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 }
136
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400137 spin_lock(&anon_vma->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 /* page_table_lock to protect against threads */
139 spin_lock(&mm->page_table_lock);
140 if (likely(!vma->anon_vma)) {
141 vma->anon_vma = anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800142 avc->anon_vma = anon_vma;
143 avc->vma = vma;
144 list_add(&avc->same_vma, &vma->anon_vma_chain);
145 list_add(&avc->same_anon_vma, &anon_vma->head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 allocated = NULL;
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400147 avc = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 }
149 spin_unlock(&mm->page_table_lock);
Linus Torvaldsd9d332e2008-10-19 10:32:20 -0700150 spin_unlock(&anon_vma->lock);
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400151
152 if (unlikely(allocated))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 anon_vma_free(allocated);
Oleg Nesterov31f2b0e2010-04-23 13:18:01 -0400154 if (unlikely(avc))
Rik van Riel5beb4932010-03-05 13:42:07 -0800155 anon_vma_chain_free(avc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 }
157 return 0;
Rik van Riel5beb4932010-03-05 13:42:07 -0800158
159 out_enomem_free_avc:
160 anon_vma_chain_free(avc);
161 out_enomem:
162 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163}
164
Rik van Riel5beb4932010-03-05 13:42:07 -0800165static void anon_vma_chain_link(struct vm_area_struct *vma,
166 struct anon_vma_chain *avc,
167 struct anon_vma *anon_vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168{
Rik van Riel5beb4932010-03-05 13:42:07 -0800169 avc->vma = vma;
170 avc->anon_vma = anon_vma;
171 list_add(&avc->same_vma, &vma->anon_vma_chain);
172
173 spin_lock(&anon_vma->lock);
174 list_add_tail(&avc->same_anon_vma, &anon_vma->head);
175 spin_unlock(&anon_vma->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176}
177
Rik van Riel5beb4932010-03-05 13:42:07 -0800178/*
179 * Attach the anon_vmas from src to dst.
180 * Returns 0 on success, -ENOMEM on failure.
181 */
182int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183{
Rik van Riel5beb4932010-03-05 13:42:07 -0800184 struct anon_vma_chain *avc, *pavc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
Linus Torvalds646d87b2010-04-11 17:15:03 -0700186 list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) {
Rik van Riel5beb4932010-03-05 13:42:07 -0800187 avc = anon_vma_chain_alloc();
188 if (!avc)
189 goto enomem_failure;
190 anon_vma_chain_link(dst, avc, pavc->anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 }
Rik van Riel5beb4932010-03-05 13:42:07 -0800192 return 0;
193
194 enomem_failure:
195 unlink_anon_vmas(dst);
196 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197}
198
Rik van Riel5beb4932010-03-05 13:42:07 -0800199/*
200 * Attach vma to its own anon_vma, as well as to the anon_vmas that
201 * the corresponding VMA in the parent process is attached to.
202 * Returns 0 on success, non-zero on failure.
203 */
204int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205{
Rik van Riel5beb4932010-03-05 13:42:07 -0800206 struct anon_vma_chain *avc;
207 struct anon_vma *anon_vma;
208
209 /* Don't bother if the parent process has no anon_vma here. */
210 if (!pvma->anon_vma)
211 return 0;
212
213 /*
214 * First, attach the new VMA to the parent VMA's anon_vmas,
215 * so rmap can find non-COWed pages in child processes.
216 */
217 if (anon_vma_clone(vma, pvma))
218 return -ENOMEM;
219
220 /* Then add our own anon_vma. */
221 anon_vma = anon_vma_alloc();
222 if (!anon_vma)
223 goto out_error;
224 avc = anon_vma_chain_alloc();
225 if (!avc)
226 goto out_error_free_anon_vma;
227 anon_vma_chain_link(vma, avc, anon_vma);
228 /* Mark this anon_vma as the one where our new (COWed) pages go. */
229 vma->anon_vma = anon_vma;
230
231 return 0;
232
233 out_error_free_anon_vma:
234 anon_vma_free(anon_vma);
235 out_error:
Rik van Riel4946d542010-04-05 12:13:33 -0400236 unlink_anon_vmas(vma);
Rik van Riel5beb4932010-03-05 13:42:07 -0800237 return -ENOMEM;
238}
239
240static void anon_vma_unlink(struct anon_vma_chain *anon_vma_chain)
241{
242 struct anon_vma *anon_vma = anon_vma_chain->anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 int empty;
244
Rik van Riel5beb4932010-03-05 13:42:07 -0800245 /* If anon_vma_fork fails, we can get an empty anon_vma_chain. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 if (!anon_vma)
247 return;
248
249 spin_lock(&anon_vma->lock);
Rik van Riel5beb4932010-03-05 13:42:07 -0800250 list_del(&anon_vma_chain->same_anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251
252 /* We must garbage collect the anon_vma if it's empty */
Hugh Dickinsdb114b82009-12-14 17:59:25 -0800253 empty = list_empty(&anon_vma->head) && !ksm_refcount(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 spin_unlock(&anon_vma->lock);
255
256 if (empty)
257 anon_vma_free(anon_vma);
258}
259
Rik van Riel5beb4932010-03-05 13:42:07 -0800260void unlink_anon_vmas(struct vm_area_struct *vma)
261{
262 struct anon_vma_chain *avc, *next;
263
264 /* Unlink each anon_vma chained to the VMA. */
265 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
266 anon_vma_unlink(avc);
267 list_del(&avc->same_vma);
268 anon_vma_chain_free(avc);
269 }
270}
271
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700272static void anon_vma_ctor(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273{
Christoph Lametera35afb82007-05-16 22:10:57 -0700274 struct anon_vma *anon_vma = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275
Christoph Lametera35afb82007-05-16 22:10:57 -0700276 spin_lock_init(&anon_vma->lock);
Hugh Dickinsdb114b82009-12-14 17:59:25 -0800277 ksm_refcount_init(anon_vma);
Christoph Lametera35afb82007-05-16 22:10:57 -0700278 INIT_LIST_HEAD(&anon_vma->head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279}
280
281void __init anon_vma_init(void)
282{
283 anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
Paul Mundt20c2df82007-07-20 10:11:58 +0900284 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor);
Rik van Riel5beb4932010-03-05 13:42:07 -0800285 anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286}
287
288/*
289 * Getting a lock on a stable anon_vma from a page off the LRU is
290 * tricky: page_lock_anon_vma rely on RCU to guard against the races.
291 */
Andi Kleen10be22d2009-09-16 11:50:04 +0200292struct anon_vma *page_lock_anon_vma(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293{
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800294 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 unsigned long anon_mapping;
296
297 rcu_read_lock();
Hugh Dickins80e14822009-12-14 17:59:29 -0800298 anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping);
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -0800299 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 goto out;
301 if (!page_mapped(page))
302 goto out;
303
304 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
305 spin_lock(&anon_vma->lock);
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800306 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307out:
308 rcu_read_unlock();
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800309 return NULL;
310}
311
Andi Kleen10be22d2009-09-16 11:50:04 +0200312void page_unlock_anon_vma(struct anon_vma *anon_vma)
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800313{
314 spin_unlock(&anon_vma->lock);
315 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316}
317
318/*
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -0800319 * At what user virtual address is page expected in @vma?
320 * Returns virtual address or -EFAULT if page's index/offset is not
321 * within the range mapped the @vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 */
323static inline unsigned long
324vma_address(struct page *page, struct vm_area_struct *vma)
325{
326 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
327 unsigned long address;
328
329 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
330 if (unlikely(address < vma->vm_start || address >= vma->vm_end)) {
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -0800331 /* page should be within @vma mapping range */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 return -EFAULT;
333 }
334 return address;
335}
336
337/*
Huang Shijiebf89c8c2009-10-01 15:44:04 -0700338 * At what user virtual address is page expected in vma?
Naoya Horiguchiab941e02010-05-11 14:06:55 -0700339 * Caller should check the page is actually part of the vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 */
341unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
342{
Naoya Horiguchiab941e02010-05-11 14:06:55 -0700343 if (PageAnon(page))
344 ;
345 else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800346 if (!vma->vm_file ||
347 vma->vm_file->f_mapping != page->mapping)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 return -EFAULT;
349 } else
350 return -EFAULT;
351 return vma_address(page, vma);
352}
353
354/*
Nikita Danilov81b40822005-05-01 08:58:36 -0700355 * Check that @page is mapped at @address into @mm.
356 *
Nick Piggin479db0b2008-08-20 14:09:18 -0700357 * If @sync is false, page_check_address may perform a racy check to avoid
358 * the page table lock when the pte is not present (helpful when reclaiming
359 * highly shared pages).
360 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700361 * On success returns with pte mapped and locked.
Nikita Danilov81b40822005-05-01 08:58:36 -0700362 */
Carsten Otteceffc072005-06-23 22:05:25 -0700363pte_t *page_check_address(struct page *page, struct mm_struct *mm,
Nick Piggin479db0b2008-08-20 14:09:18 -0700364 unsigned long address, spinlock_t **ptlp, int sync)
Nikita Danilov81b40822005-05-01 08:58:36 -0700365{
366 pgd_t *pgd;
367 pud_t *pud;
368 pmd_t *pmd;
369 pte_t *pte;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700370 spinlock_t *ptl;
Nikita Danilov81b40822005-05-01 08:58:36 -0700371
Nikita Danilov81b40822005-05-01 08:58:36 -0700372 pgd = pgd_offset(mm, address);
Hugh Dickinsc0718802005-10-29 18:16:31 -0700373 if (!pgd_present(*pgd))
374 return NULL;
375
376 pud = pud_offset(pgd, address);
377 if (!pud_present(*pud))
378 return NULL;
379
380 pmd = pmd_offset(pud, address);
381 if (!pmd_present(*pmd))
382 return NULL;
383
384 pte = pte_offset_map(pmd, address);
385 /* Make a quick check before getting the lock */
Nick Piggin479db0b2008-08-20 14:09:18 -0700386 if (!sync && !pte_present(*pte)) {
Hugh Dickinsc0718802005-10-29 18:16:31 -0700387 pte_unmap(pte);
388 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700389 }
Hugh Dickinsc0718802005-10-29 18:16:31 -0700390
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700391 ptl = pte_lockptr(mm, pmd);
Hugh Dickinsc0718802005-10-29 18:16:31 -0700392 spin_lock(ptl);
393 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
394 *ptlp = ptl;
395 return pte;
396 }
397 pte_unmap_unlock(pte, ptl);
398 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700399}
400
Nick Pigginb291f002008-10-18 20:26:44 -0700401/**
402 * page_mapped_in_vma - check whether a page is really mapped in a VMA
403 * @page: the page to test
404 * @vma: the VMA to test
405 *
406 * Returns 1 if the page is mapped into the page tables of the VMA, 0
407 * if the page is not mapped into the page tables of this VMA. Only
408 * valid for normal file or anonymous VMAs.
409 */
Andi Kleen6a460792009-09-16 11:50:15 +0200410int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
Nick Pigginb291f002008-10-18 20:26:44 -0700411{
412 unsigned long address;
413 pte_t *pte;
414 spinlock_t *ptl;
415
416 address = vma_address(page, vma);
417 if (address == -EFAULT) /* out of vma range */
418 return 0;
419 pte = page_check_address(page, vma->vm_mm, address, &ptl, 1);
420 if (!pte) /* the page is not in this mm */
421 return 0;
422 pte_unmap_unlock(pte, ptl);
423
424 return 1;
425}
426
Nikita Danilov81b40822005-05-01 08:58:36 -0700427/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 * Subfunctions of page_referenced: page_referenced_one called
429 * repeatedly from either page_referenced_anon or page_referenced_file.
430 */
Hugh Dickins5ad64682009-12-14 17:59:24 -0800431int page_referenced_one(struct page *page, struct vm_area_struct *vma,
432 unsigned long address, unsigned int *mapcount,
433 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434{
435 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 pte_t *pte;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700437 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 int referenced = 0;
439
Nick Piggin479db0b2008-08-20 14:09:18 -0700440 pte = page_check_address(page, mm, address, &ptl, 0);
Hugh Dickinsc0718802005-10-29 18:16:31 -0700441 if (!pte)
442 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443
Nick Pigginb291f002008-10-18 20:26:44 -0700444 /*
445 * Don't want to elevate referenced for mlocked page that gets this far,
446 * in order that it progresses to try_to_unmap and is moved to the
447 * unevictable list.
448 */
Hugh Dickins5a9bbdc2008-02-04 22:29:23 -0800449 if (vma->vm_flags & VM_LOCKED) {
Hugh Dickins5a9bbdc2008-02-04 22:29:23 -0800450 *mapcount = 1; /* break early from loop */
Minchan Kim03ef83a2009-08-26 14:29:23 -0700451 *vm_flags |= VM_LOCKED;
Nick Pigginb291f002008-10-18 20:26:44 -0700452 goto out_unmap;
453 }
454
Johannes Weiner4917e5d2009-01-06 14:39:17 -0800455 if (ptep_clear_flush_young_notify(vma, address, pte)) {
456 /*
457 * Don't treat a reference through a sequentially read
458 * mapping as such. If the page has been used in
459 * another mapping, we will catch it; if this other
460 * mapping is already gone, the unmap path will have
461 * set PG_referenced or activated the page.
462 */
463 if (likely(!VM_SequentialReadHint(vma)))
464 referenced++;
465 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
Hugh Dickinsc0718802005-10-29 18:16:31 -0700467 /* Pretend the page is referenced if the task has the
468 swap token and is in the middle of a page fault. */
Rik van Rielf7b7fd82005-11-28 13:44:07 -0800469 if (mm != current->mm && has_swap_token(mm) &&
Hugh Dickinsc0718802005-10-29 18:16:31 -0700470 rwsem_is_locked(&mm->mmap_sem))
471 referenced++;
472
Nick Pigginb291f002008-10-18 20:26:44 -0700473out_unmap:
Hugh Dickinsc0718802005-10-29 18:16:31 -0700474 (*mapcount)--;
475 pte_unmap_unlock(pte, ptl);
Huang Shijie273f0472009-12-14 17:58:51 -0800476
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700477 if (referenced)
478 *vm_flags |= vma->vm_flags;
Huang Shijie273f0472009-12-14 17:58:51 -0800479out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 return referenced;
481}
482
Balbir Singhbed71612008-02-07 00:14:01 -0800483static int page_referenced_anon(struct page *page,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700484 struct mem_cgroup *mem_cont,
485 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486{
487 unsigned int mapcount;
488 struct anon_vma *anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -0800489 struct anon_vma_chain *avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 int referenced = 0;
491
492 anon_vma = page_lock_anon_vma(page);
493 if (!anon_vma)
494 return referenced;
495
496 mapcount = page_mapcount(page);
Rik van Riel5beb4932010-03-05 13:42:07 -0800497 list_for_each_entry(avc, &anon_vma->head, same_anon_vma) {
498 struct vm_area_struct *vma = avc->vma;
Hugh Dickins1cb17292009-12-14 17:59:01 -0800499 unsigned long address = vma_address(page, vma);
500 if (address == -EFAULT)
501 continue;
Balbir Singhbed71612008-02-07 00:14:01 -0800502 /*
503 * If we are reclaiming on behalf of a cgroup, skip
504 * counting on behalf of references from different
505 * cgroups
506 */
Hugh Dickinsbd845e32008-03-04 14:29:01 -0800507 if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont))
Balbir Singhbed71612008-02-07 00:14:01 -0800508 continue;
Hugh Dickins1cb17292009-12-14 17:59:01 -0800509 referenced += page_referenced_one(page, vma, address,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700510 &mapcount, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 if (!mapcount)
512 break;
513 }
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800514
515 page_unlock_anon_vma(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 return referenced;
517}
518
519/**
520 * page_referenced_file - referenced check for object-based rmap
521 * @page: the page we're checking references on.
Randy Dunlap43d8eac2008-03-19 17:00:43 -0700522 * @mem_cont: target memory controller
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700523 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 *
525 * For an object-based mapped page, find all the places it is mapped and
526 * check/clear the referenced flag. This is done by following the page->mapping
527 * pointer, then walking the chain of vmas it holds. It returns the number
528 * of references it found.
529 *
530 * This function is only called from page_referenced for object-based pages.
531 */
Balbir Singhbed71612008-02-07 00:14:01 -0800532static int page_referenced_file(struct page *page,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700533 struct mem_cgroup *mem_cont,
534 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535{
536 unsigned int mapcount;
537 struct address_space *mapping = page->mapping;
538 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
539 struct vm_area_struct *vma;
540 struct prio_tree_iter iter;
541 int referenced = 0;
542
543 /*
544 * The caller's checks on page->mapping and !PageAnon have made
545 * sure that this is a file page: the check for page->mapping
546 * excludes the case just before it gets set on an anon page.
547 */
548 BUG_ON(PageAnon(page));
549
550 /*
551 * The page lock not only makes sure that page->mapping cannot
552 * suddenly be NULLified by truncation, it makes sure that the
553 * structure at mapping cannot be freed and reused yet,
554 * so we can safely take mapping->i_mmap_lock.
555 */
556 BUG_ON(!PageLocked(page));
557
558 spin_lock(&mapping->i_mmap_lock);
559
560 /*
561 * i_mmap_lock does not stabilize mapcount at all, but mapcount
562 * is more likely to be accurate if we note it after spinning.
563 */
564 mapcount = page_mapcount(page);
565
566 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
Hugh Dickins1cb17292009-12-14 17:59:01 -0800567 unsigned long address = vma_address(page, vma);
568 if (address == -EFAULT)
569 continue;
Balbir Singhbed71612008-02-07 00:14:01 -0800570 /*
571 * If we are reclaiming on behalf of a cgroup, skip
572 * counting on behalf of references from different
573 * cgroups
574 */
Hugh Dickinsbd845e32008-03-04 14:29:01 -0800575 if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont))
Balbir Singhbed71612008-02-07 00:14:01 -0800576 continue;
Hugh Dickins1cb17292009-12-14 17:59:01 -0800577 referenced += page_referenced_one(page, vma, address,
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700578 &mapcount, vm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 if (!mapcount)
580 break;
581 }
582
583 spin_unlock(&mapping->i_mmap_lock);
584 return referenced;
585}
586
587/**
588 * page_referenced - test if the page was referenced
589 * @page: the page to test
590 * @is_locked: caller holds lock on the page
Randy Dunlap43d8eac2008-03-19 17:00:43 -0700591 * @mem_cont: target memory controller
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700592 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 *
594 * Quick test_and_clear_referenced for all mappings to a page,
595 * returns the number of ptes which referenced the page.
596 */
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700597int page_referenced(struct page *page,
598 int is_locked,
599 struct mem_cgroup *mem_cont,
600 unsigned long *vm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601{
602 int referenced = 0;
Hugh Dickins5ad64682009-12-14 17:59:24 -0800603 int we_locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700605 *vm_flags = 0;
Hugh Dickins3ca7b3c2009-12-14 17:58:57 -0800606 if (page_mapped(page) && page_rmapping(page)) {
Hugh Dickins5ad64682009-12-14 17:59:24 -0800607 if (!is_locked && (!PageAnon(page) || PageKsm(page))) {
608 we_locked = trylock_page(page);
609 if (!we_locked) {
610 referenced++;
611 goto out;
612 }
613 }
614 if (unlikely(PageKsm(page)))
615 referenced += page_referenced_ksm(page, mem_cont,
616 vm_flags);
617 else if (PageAnon(page))
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700618 referenced += page_referenced_anon(page, mem_cont,
619 vm_flags);
Hugh Dickins5ad64682009-12-14 17:59:24 -0800620 else if (page->mapping)
Wu Fengguang6fe6b7e2009-06-16 15:33:05 -0700621 referenced += page_referenced_file(page, mem_cont,
622 vm_flags);
Hugh Dickins5ad64682009-12-14 17:59:24 -0800623 if (we_locked)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 }
Hugh Dickins5ad64682009-12-14 17:59:24 -0800626out:
Christian Borntraeger5b7baf02008-03-25 18:47:12 +0100627 if (page_test_and_clear_young(page))
628 referenced++;
629
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 return referenced;
631}
632
Hugh Dickins1cb17292009-12-14 17:59:01 -0800633static int page_mkclean_one(struct page *page, struct vm_area_struct *vma,
634 unsigned long address)
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700635{
636 struct mm_struct *mm = vma->vm_mm;
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100637 pte_t *pte;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700638 spinlock_t *ptl;
639 int ret = 0;
640
Nick Piggin479db0b2008-08-20 14:09:18 -0700641 pte = page_check_address(page, mm, address, &ptl, 1);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700642 if (!pte)
643 goto out;
644
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100645 if (pte_dirty(*pte) || pte_write(*pte)) {
646 pte_t entry;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700647
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100648 flush_cache_page(vma, address, pte_pfn(*pte));
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700649 entry = ptep_clear_flush_notify(vma, address, pte);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100650 entry = pte_wrprotect(entry);
651 entry = pte_mkclean(entry);
Al Virod6e88e62006-12-29 16:48:35 -0800652 set_pte_at(mm, address, pte, entry);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100653 ret = 1;
654 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700655
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700656 pte_unmap_unlock(pte, ptl);
657out:
658 return ret;
659}
660
661static int page_mkclean_file(struct address_space *mapping, struct page *page)
662{
663 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
664 struct vm_area_struct *vma;
665 struct prio_tree_iter iter;
666 int ret = 0;
667
668 BUG_ON(PageAnon(page));
669
670 spin_lock(&mapping->i_mmap_lock);
671 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
Hugh Dickins1cb17292009-12-14 17:59:01 -0800672 if (vma->vm_flags & VM_SHARED) {
673 unsigned long address = vma_address(page, vma);
674 if (address == -EFAULT)
675 continue;
676 ret += page_mkclean_one(page, vma, address);
677 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700678 }
679 spin_unlock(&mapping->i_mmap_lock);
680 return ret;
681}
682
683int page_mkclean(struct page *page)
684{
685 int ret = 0;
686
687 BUG_ON(!PageLocked(page));
688
689 if (page_mapped(page)) {
690 struct address_space *mapping = page_mapping(page);
Christian Borntraegerce7e9fa2007-11-20 11:13:36 +0100691 if (mapping) {
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700692 ret = page_mkclean_file(mapping, page);
Christian Borntraegerce7e9fa2007-11-20 11:13:36 +0100693 if (page_test_dirty(page)) {
694 page_clear_dirty(page);
695 ret = 1;
696 }
Martin Schwidefsky6c210482007-04-27 16:01:57 +0200697 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700698 }
699
700 return ret;
701}
Jaya Kumar60b59be2007-05-08 00:37:37 -0700702EXPORT_SYMBOL_GPL(page_mkclean);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700703
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704/**
Rik van Rielc44b6742010-03-05 13:42:09 -0800705 * page_move_anon_rmap - move a page to our anon_vma
706 * @page: the page to move to our anon_vma
707 * @vma: the vma the page belongs to
708 * @address: the user virtual address mapped
709 *
710 * When a page belongs exclusively to one process after a COW event,
711 * that page can be moved into the anon_vma that belongs to just that
712 * process, so the rmap code will not search the parent or sibling
713 * processes.
714 */
715void page_move_anon_rmap(struct page *page,
716 struct vm_area_struct *vma, unsigned long address)
717{
718 struct anon_vma *anon_vma = vma->anon_vma;
719
720 VM_BUG_ON(!PageLocked(page));
721 VM_BUG_ON(!anon_vma);
722 VM_BUG_ON(page->index != linear_page_index(vma, address));
723
724 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
725 page->mapping = (struct address_space *) anon_vma;
726}
727
728/**
Randy Dunlap43d8eac2008-03-19 17:00:43 -0700729 * __page_set_anon_rmap - setup new anonymous rmap
Nick Piggin9617d952006-01-06 00:11:12 -0800730 * @page: the page to add the mapping to
731 * @vma: the vm area in which the mapping is added
732 * @address: the user virtual address mapped
Rik van Riele8a03fe2010-04-14 17:59:28 -0400733 * @exclusive: the page is exclusively owned by the current process
Nick Piggin9617d952006-01-06 00:11:12 -0800734 */
735static void __page_set_anon_rmap(struct page *page,
Rik van Riele8a03fe2010-04-14 17:59:28 -0400736 struct vm_area_struct *vma, unsigned long address, int exclusive)
Nick Piggin9617d952006-01-06 00:11:12 -0800737{
Rik van Riele8a03fe2010-04-14 17:59:28 -0400738 struct anon_vma *anon_vma = vma->anon_vma;
Nick Piggin9617d952006-01-06 00:11:12 -0800739
Rik van Riele8a03fe2010-04-14 17:59:28 -0400740 BUG_ON(!anon_vma);
Linus Torvaldsea900022010-04-12 12:44:29 -0700741
742 /*
Rik van Riele8a03fe2010-04-14 17:59:28 -0400743 * If the page isn't exclusively mapped into this vma,
744 * we must use the _oldest_ possible anon_vma for the
745 * page mapping!
Linus Torvaldsea900022010-04-12 12:44:29 -0700746 *
Rik van Riele8a03fe2010-04-14 17:59:28 -0400747 * So take the last AVC chain entry in the vma, which is
748 * the deepest ancestor, and use the anon_vma from that.
Linus Torvaldsea900022010-04-12 12:44:29 -0700749 */
Rik van Riele8a03fe2010-04-14 17:59:28 -0400750 if (!exclusive) {
751 struct anon_vma_chain *avc;
752 avc = list_entry(vma->anon_vma_chain.prev, struct anon_vma_chain, same_vma);
753 anon_vma = avc->anon_vma;
754 }
Linus Torvaldsea900022010-04-12 12:44:29 -0700755
Nick Piggin9617d952006-01-06 00:11:12 -0800756 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
757 page->mapping = (struct address_space *) anon_vma;
Nick Piggin9617d952006-01-06 00:11:12 -0800758 page->index = linear_page_index(vma, address);
Nick Piggin9617d952006-01-06 00:11:12 -0800759}
760
761/**
Randy Dunlap43d8eac2008-03-19 17:00:43 -0700762 * __page_check_anon_rmap - sanity check anonymous rmap addition
Nick Pigginc97a9e12007-05-16 22:11:21 -0700763 * @page: the page to add the mapping to
764 * @vma: the vm area in which the mapping is added
765 * @address: the user virtual address mapped
766 */
767static void __page_check_anon_rmap(struct page *page,
768 struct vm_area_struct *vma, unsigned long address)
769{
770#ifdef CONFIG_DEBUG_VM
771 /*
772 * The page's anon-rmap details (mapping and index) are guaranteed to
773 * be set up correctly at this point.
774 *
775 * We have exclusion against page_add_anon_rmap because the caller
776 * always holds the page locked, except if called from page_dup_rmap,
777 * in which case the page is already known to be setup.
778 *
779 * We have exclusion against page_add_new_anon_rmap because those pages
780 * are initially only visible via the pagetables, and the pte is locked
781 * over the call to page_add_new_anon_rmap.
782 */
Nick Pigginc97a9e12007-05-16 22:11:21 -0700783 BUG_ON(page->index != linear_page_index(vma, address));
784#endif
785}
786
787/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 * page_add_anon_rmap - add pte mapping to an anonymous page
789 * @page: the page to add the mapping to
790 * @vma: the vm area in which the mapping is added
791 * @address: the user virtual address mapped
792 *
Hugh Dickins5ad64682009-12-14 17:59:24 -0800793 * The caller needs to hold the pte lock, and the page must be locked in
Hugh Dickins80e14822009-12-14 17:59:29 -0800794 * the anon_vma case: to serialize mapping,index checking after setting,
795 * and to ensure that PageAnon is not being upgraded racily to PageKsm
796 * (but PageKsm is never downgraded to PageAnon).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 */
798void page_add_anon_rmap(struct page *page,
799 struct vm_area_struct *vma, unsigned long address)
800{
Hugh Dickins5ad64682009-12-14 17:59:24 -0800801 int first = atomic_inc_and_test(&page->_mapcount);
802 if (first)
803 __inc_zone_page_state(page, NR_ANON_PAGES);
804 if (unlikely(PageKsm(page)))
805 return;
806
Nick Pigginc97a9e12007-05-16 22:11:21 -0700807 VM_BUG_ON(!PageLocked(page));
808 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
Hugh Dickins5ad64682009-12-14 17:59:24 -0800809 if (first)
Rik van Riele8a03fe2010-04-14 17:59:28 -0400810 __page_set_anon_rmap(page, vma, address, 0);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700811 else
Nick Pigginc97a9e12007-05-16 22:11:21 -0700812 __page_check_anon_rmap(page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813}
814
Randy Dunlap43d8eac2008-03-19 17:00:43 -0700815/**
Nick Piggin9617d952006-01-06 00:11:12 -0800816 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
817 * @page: the page to add the mapping to
818 * @vma: the vm area in which the mapping is added
819 * @address: the user virtual address mapped
820 *
821 * Same as page_add_anon_rmap but must only be called on *new* pages.
822 * This means the inc-and-test can be bypassed.
Nick Pigginc97a9e12007-05-16 22:11:21 -0700823 * Page does not have to be locked.
Nick Piggin9617d952006-01-06 00:11:12 -0800824 */
825void page_add_new_anon_rmap(struct page *page,
826 struct vm_area_struct *vma, unsigned long address)
827{
Hugh Dickinsb5934c52009-01-06 14:39:25 -0800828 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
Hugh Dickinscbf84b72009-01-06 14:39:27 -0800829 SetPageSwapBacked(page);
830 atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */
Hugh Dickins5ad64682009-12-14 17:59:24 -0800831 __inc_zone_page_state(page, NR_ANON_PAGES);
Rik van Riele8a03fe2010-04-14 17:59:28 -0400832 __page_set_anon_rmap(page, vma, address, 1);
Hugh Dickinsb5934c52009-01-06 14:39:25 -0800833 if (page_evictable(page, vma))
Hugh Dickinscbf84b72009-01-06 14:39:27 -0800834 lru_cache_add_lru(page, LRU_ACTIVE_ANON);
Hugh Dickinsb5934c52009-01-06 14:39:25 -0800835 else
836 add_page_to_unevictable_list(page);
Nick Piggin9617d952006-01-06 00:11:12 -0800837}
838
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839/**
840 * page_add_file_rmap - add pte mapping to a file page
841 * @page: the page to add the mapping to
842 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700843 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 */
845void page_add_file_rmap(struct page *page)
846{
Balbir Singhd69b0422009-06-17 16:26:34 -0700847 if (atomic_inc_and_test(&page->_mapcount)) {
Christoph Lameter65ba55f2006-06-30 01:55:34 -0700848 __inc_zone_page_state(page, NR_FILE_MAPPED);
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -0800849 mem_cgroup_update_file_mapped(page, 1);
Balbir Singhd69b0422009-06-17 16:26:34 -0700850 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851}
852
853/**
854 * page_remove_rmap - take down pte mapping from a page
855 * @page: page to remove mapping from
856 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700857 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 */
Hugh Dickinsedc315f2009-01-06 14:40:11 -0800859void page_remove_rmap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860{
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -0700861 /* page still mapped by someone else? */
862 if (!atomic_add_negative(-1, &page->_mapcount))
863 return;
864
865 /*
866 * Now that the last pte has gone, s390 must transfer dirty
867 * flag from storage key to struct page. We can usually skip
868 * this if the page is anon, so about to be freed; but perhaps
869 * not if it's in swapcache - there might be another pte slot
870 * containing the swap entry, but page not yet written to swap.
871 */
872 if ((!PageAnon(page) || PageSwapCache(page)) && page_test_dirty(page)) {
873 page_clear_dirty(page);
874 set_page_dirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 }
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -0700876 if (PageAnon(page)) {
877 mem_cgroup_uncharge_page(page);
878 __dec_zone_page_state(page, NR_ANON_PAGES);
879 } else {
880 __dec_zone_page_state(page, NR_FILE_MAPPED);
KAMEZAWA Hiroyukid8046582009-12-15 16:47:09 -0800881 mem_cgroup_update_file_mapped(page, -1);
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -0700882 }
KOSAKI Motohirob904dcf2009-09-21 17:01:28 -0700883 /*
884 * It would be tidy to reset the PageAnon mapping here,
885 * but that might overwrite a racing page_add_anon_rmap
886 * which increments mapcount after us but sets mapping
887 * before us: so leave the reset to free_hot_cold_page,
888 * and remember that it's only reliable while mapped.
889 * Leaving it set also helps swapoff to reinstate ptes
890 * faster for those pages still in swapcache.
891 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892}
893
894/*
895 * Subfunctions of try_to_unmap: try_to_unmap_one called
896 * repeatedly from either try_to_unmap_anon or try_to_unmap_file.
897 */
Hugh Dickins5ad64682009-12-14 17:59:24 -0800898int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
899 unsigned long address, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900{
901 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 pte_t *pte;
903 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700904 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 int ret = SWAP_AGAIN;
906
Nick Piggin479db0b2008-08-20 14:09:18 -0700907 pte = page_check_address(page, mm, address, &ptl, 0);
Hugh Dickinsc0718802005-10-29 18:16:31 -0700908 if (!pte)
Nikita Danilov81b40822005-05-01 08:58:36 -0700909 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910
911 /*
912 * If the page is mlock()d, we cannot swap it out.
913 * If it's recently referenced (perhaps page_referenced
914 * skipped over this mm) then we should reactivate it.
915 */
Andi Kleen14fa31b2009-09-16 11:50:10 +0200916 if (!(flags & TTU_IGNORE_MLOCK)) {
KOSAKI Motohirocaed0f42009-12-14 17:59:45 -0800917 if (vma->vm_flags & VM_LOCKED)
918 goto out_mlock;
919
Hugh Dickinsaf8e3352009-12-14 17:58:59 -0800920 if (TTU_ACTION(flags) == TTU_MUNLOCK)
Hugh Dickins53f79ac2009-12-14 17:58:58 -0800921 goto out_unmap;
Andi Kleen14fa31b2009-09-16 11:50:10 +0200922 }
923 if (!(flags & TTU_IGNORE_ACCESS)) {
Nick Pigginb291f002008-10-18 20:26:44 -0700924 if (ptep_clear_flush_young_notify(vma, address, pte)) {
925 ret = SWAP_FAIL;
926 goto out_unmap;
927 }
928 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 /* Nuke the page table entry. */
931 flush_cache_page(vma, address, page_to_pfn(page));
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700932 pteval = ptep_clear_flush_notify(vma, address, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933
934 /* Move the dirty bit to the physical page now the pte is gone. */
935 if (pte_dirty(pteval))
936 set_page_dirty(page);
937
Hugh Dickins365e9c872005-10-29 18:16:18 -0700938 /* Update high watermark before we lower rss */
939 update_hiwater_rss(mm);
940
Andi Kleen888b9f72009-09-16 11:50:11 +0200941 if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
942 if (PageAnon(page))
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800943 dec_mm_counter(mm, MM_ANONPAGES);
Andi Kleen888b9f72009-09-16 11:50:11 +0200944 else
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800945 dec_mm_counter(mm, MM_FILEPAGES);
Andi Kleen888b9f72009-09-16 11:50:11 +0200946 set_pte_at(mm, address, pte,
947 swp_entry_to_pte(make_hwpoison_entry(page)));
948 } else if (PageAnon(page)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700949 swp_entry_t entry = { .val = page_private(page) };
Christoph Lameter06972122006-06-23 02:03:35 -0700950
951 if (PageSwapCache(page)) {
952 /*
953 * Store the swap location in the pte.
954 * See handle_pte_fault() ...
955 */
Hugh Dickins570a3352009-12-14 17:58:46 -0800956 if (swap_duplicate(entry) < 0) {
957 set_pte_at(mm, address, pte, pteval);
958 ret = SWAP_FAIL;
959 goto out_unmap;
960 }
Christoph Lameter06972122006-06-23 02:03:35 -0700961 if (list_empty(&mm->mmlist)) {
962 spin_lock(&mmlist_lock);
963 if (list_empty(&mm->mmlist))
964 list_add(&mm->mmlist, &init_mm.mmlist);
965 spin_unlock(&mmlist_lock);
966 }
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800967 dec_mm_counter(mm, MM_ANONPAGES);
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -0800968 inc_mm_counter(mm, MM_SWAPENTS);
KOSAKI Motohiro64cdd542009-01-06 14:39:16 -0800969 } else if (PAGE_MIGRATION) {
Christoph Lameter06972122006-06-23 02:03:35 -0700970 /*
971 * Store the pfn of the page in a special migration
972 * pte. do_swap_page() will wait until the migration
973 * pte is removed and then restart fault handling.
974 */
Andi Kleen14fa31b2009-09-16 11:50:10 +0200975 BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION);
Christoph Lameter06972122006-06-23 02:03:35 -0700976 entry = make_migration_entry(page, pte_write(pteval));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 }
978 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
979 BUG_ON(pte_file(*pte));
Andi Kleen14fa31b2009-09-16 11:50:10 +0200980 } else if (PAGE_MIGRATION && (TTU_ACTION(flags) == TTU_MIGRATION)) {
Christoph Lameter04e62a22006-06-23 02:03:38 -0700981 /* Establish migration entry for a file page */
982 swp_entry_t entry;
983 entry = make_migration_entry(page, pte_write(pteval));
984 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
985 } else
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800986 dec_mm_counter(mm, MM_FILEPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987
Hugh Dickinsedc315f2009-01-06 14:40:11 -0800988 page_remove_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 page_cache_release(page);
990
991out_unmap:
Hugh Dickinsc0718802005-10-29 18:16:31 -0700992 pte_unmap_unlock(pte, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993out:
994 return ret;
KOSAKI Motohirocaed0f42009-12-14 17:59:45 -0800995
996out_mlock:
997 pte_unmap_unlock(pte, ptl);
998
999
1000 /*
1001 * We need mmap_sem locking, Otherwise VM_LOCKED check makes
1002 * unstable result and race. Plus, We can't wait here because
1003 * we now hold anon_vma->lock or mapping->i_mmap_lock.
1004 * if trylock failed, the page remain in evictable lru and later
1005 * vmscan could retry to move the page to unevictable lru if the
1006 * page is actually mlocked.
1007 */
1008 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
1009 if (vma->vm_flags & VM_LOCKED) {
1010 mlock_vma_page(page);
1011 ret = SWAP_MLOCK;
1012 }
1013 up_read(&vma->vm_mm->mmap_sem);
1014 }
1015 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016}
1017
1018/*
1019 * objrmap doesn't work for nonlinear VMAs because the assumption that
1020 * offset-into-file correlates with offset-into-virtual-addresses does not hold.
1021 * Consequently, given a particular page and its ->index, we cannot locate the
1022 * ptes which are mapping that page without an exhaustive linear search.
1023 *
1024 * So what this code does is a mini "virtual scan" of each nonlinear VMA which
1025 * maps the file to which the target page belongs. The ->vm_private_data field
1026 * holds the current cursor into that scan. Successive searches will circulate
1027 * around the vma's virtual address space.
1028 *
1029 * So as more replacement pressure is applied to the pages in a nonlinear VMA,
1030 * more scanning pressure is placed against them as well. Eventually pages
1031 * will become fully unmapped and are eligible for eviction.
1032 *
1033 * For very sparsely populated VMAs this is a little inefficient - chances are
1034 * there there won't be many ptes located within the scan cluster. In this case
1035 * maybe we could scan further - to the end of the pte page, perhaps.
Nick Pigginb291f002008-10-18 20:26:44 -07001036 *
1037 * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can
1038 * acquire it without blocking. If vma locked, mlock the pages in the cluster,
1039 * rather than unmapping them. If we encounter the "check_page" that vmscan is
1040 * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 */
1042#define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE)
1043#define CLUSTER_MASK (~(CLUSTER_SIZE - 1))
1044
Nick Pigginb291f002008-10-18 20:26:44 -07001045static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount,
1046 struct vm_area_struct *vma, struct page *check_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047{
1048 struct mm_struct *mm = vma->vm_mm;
1049 pgd_t *pgd;
1050 pud_t *pud;
1051 pmd_t *pmd;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001052 pte_t *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -07001054 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 struct page *page;
1056 unsigned long address;
1057 unsigned long end;
Nick Pigginb291f002008-10-18 20:26:44 -07001058 int ret = SWAP_AGAIN;
1059 int locked_vma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 address = (vma->vm_start + cursor) & CLUSTER_MASK;
1062 end = address + CLUSTER_SIZE;
1063 if (address < vma->vm_start)
1064 address = vma->vm_start;
1065 if (end > vma->vm_end)
1066 end = vma->vm_end;
1067
1068 pgd = pgd_offset(mm, address);
1069 if (!pgd_present(*pgd))
Nick Pigginb291f002008-10-18 20:26:44 -07001070 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071
1072 pud = pud_offset(pgd, address);
1073 if (!pud_present(*pud))
Nick Pigginb291f002008-10-18 20:26:44 -07001074 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075
1076 pmd = pmd_offset(pud, address);
1077 if (!pmd_present(*pmd))
Nick Pigginb291f002008-10-18 20:26:44 -07001078 return ret;
1079
1080 /*
Hugh Dickinsaf8e3352009-12-14 17:58:59 -08001081 * If we can acquire the mmap_sem for read, and vma is VM_LOCKED,
Nick Pigginb291f002008-10-18 20:26:44 -07001082 * keep the sem while scanning the cluster for mlocking pages.
1083 */
Hugh Dickinsaf8e3352009-12-14 17:58:59 -08001084 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
Nick Pigginb291f002008-10-18 20:26:44 -07001085 locked_vma = (vma->vm_flags & VM_LOCKED);
1086 if (!locked_vma)
1087 up_read(&vma->vm_mm->mmap_sem); /* don't need it */
1088 }
Hugh Dickinsc0718802005-10-29 18:16:31 -07001089
1090 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091
Hugh Dickins365e9c872005-10-29 18:16:18 -07001092 /* Update high watermark before we lower rss */
1093 update_hiwater_rss(mm);
1094
Hugh Dickinsc0718802005-10-29 18:16:31 -07001095 for (; address < end; pte++, address += PAGE_SIZE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 if (!pte_present(*pte))
1097 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001098 page = vm_normal_page(vma, address, *pte);
1099 BUG_ON(!page || PageAnon(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100
Nick Pigginb291f002008-10-18 20:26:44 -07001101 if (locked_vma) {
1102 mlock_vma_page(page); /* no-op if already mlocked */
1103 if (page == check_page)
1104 ret = SWAP_MLOCK;
1105 continue; /* don't unmap */
1106 }
1107
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001108 if (ptep_clear_flush_young_notify(vma, address, pte))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 continue;
1110
1111 /* Nuke the page table entry. */
Ben Collinseca35132005-11-29 11:45:26 -08001112 flush_cache_page(vma, address, pte_pfn(*pte));
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001113 pteval = ptep_clear_flush_notify(vma, address, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114
1115 /* If nonlinear, store the file page offset in the pte. */
1116 if (page->index != linear_page_index(vma, address))
1117 set_pte_at(mm, address, pte, pgoff_to_pte(page->index));
1118
1119 /* Move the dirty bit to the physical page now the pte is gone. */
1120 if (pte_dirty(pteval))
1121 set_page_dirty(page);
1122
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001123 page_remove_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 page_cache_release(page);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001125 dec_mm_counter(mm, MM_FILEPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 (*mapcount)--;
1127 }
Hugh Dickinsc0718802005-10-29 18:16:31 -07001128 pte_unmap_unlock(pte - 1, ptl);
Nick Pigginb291f002008-10-18 20:26:44 -07001129 if (locked_vma)
1130 up_read(&vma->vm_mm->mmap_sem);
1131 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132}
1133
Nick Pigginb291f002008-10-18 20:26:44 -07001134/**
1135 * try_to_unmap_anon - unmap or unlock anonymous page using the object-based
1136 * rmap method
1137 * @page: the page to unmap/unlock
Huang Shijie8051be52009-12-14 17:58:50 -08001138 * @flags: action and flags
Nick Pigginb291f002008-10-18 20:26:44 -07001139 *
1140 * Find all the mappings of a page using the mapping pointer and the vma chains
1141 * contained in the anon_vma struct it points to.
1142 *
1143 * This function is only called from try_to_unmap/try_to_munlock for
1144 * anonymous pages.
1145 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1146 * where the page was found will be held for write. So, we won't recheck
1147 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1148 * 'LOCKED.
1149 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001150static int try_to_unmap_anon(struct page *page, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151{
1152 struct anon_vma *anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08001153 struct anon_vma_chain *avc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 int ret = SWAP_AGAIN;
Nick Pigginb291f002008-10-18 20:26:44 -07001155
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 anon_vma = page_lock_anon_vma(page);
1157 if (!anon_vma)
1158 return ret;
1159
Rik van Riel5beb4932010-03-05 13:42:07 -08001160 list_for_each_entry(avc, &anon_vma->head, same_anon_vma) {
1161 struct vm_area_struct *vma = avc->vma;
Hugh Dickins1cb17292009-12-14 17:59:01 -08001162 unsigned long address = vma_address(page, vma);
1163 if (address == -EFAULT)
1164 continue;
1165 ret = try_to_unmap_one(page, vma, address, flags);
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001166 if (ret != SWAP_AGAIN || !page_mapped(page))
1167 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 }
Oleg Nesterov34bbd702007-02-28 20:13:49 -08001169
1170 page_unlock_anon_vma(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 return ret;
1172}
1173
1174/**
Nick Pigginb291f002008-10-18 20:26:44 -07001175 * try_to_unmap_file - unmap/unlock file page using the object-based rmap method
1176 * @page: the page to unmap/unlock
Andi Kleen14fa31b2009-09-16 11:50:10 +02001177 * @flags: action and flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 *
1179 * Find all the mappings of a page using the mapping pointer and the vma chains
1180 * contained in the address_space struct it points to.
1181 *
Nick Pigginb291f002008-10-18 20:26:44 -07001182 * This function is only called from try_to_unmap/try_to_munlock for
1183 * object-based pages.
1184 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1185 * where the page was found will be held for write. So, we won't recheck
1186 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1187 * 'LOCKED.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001189static int try_to_unmap_file(struct page *page, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190{
1191 struct address_space *mapping = page->mapping;
1192 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1193 struct vm_area_struct *vma;
1194 struct prio_tree_iter iter;
1195 int ret = SWAP_AGAIN;
1196 unsigned long cursor;
1197 unsigned long max_nl_cursor = 0;
1198 unsigned long max_nl_size = 0;
1199 unsigned int mapcount;
1200
1201 spin_lock(&mapping->i_mmap_lock);
1202 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
Hugh Dickins1cb17292009-12-14 17:59:01 -08001203 unsigned long address = vma_address(page, vma);
1204 if (address == -EFAULT)
1205 continue;
1206 ret = try_to_unmap_one(page, vma, address, flags);
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001207 if (ret != SWAP_AGAIN || !page_mapped(page))
1208 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 }
1210
1211 if (list_empty(&mapping->i_mmap_nonlinear))
1212 goto out;
1213
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001214 /*
1215 * We don't bother to try to find the munlocked page in nonlinears.
1216 * It's costly. Instead, later, page reclaim logic may call
1217 * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily.
1218 */
1219 if (TTU_ACTION(flags) == TTU_MUNLOCK)
1220 goto out;
1221
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
1223 shared.vm_set.list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 cursor = (unsigned long) vma->vm_private_data;
1225 if (cursor > max_nl_cursor)
1226 max_nl_cursor = cursor;
1227 cursor = vma->vm_end - vma->vm_start;
1228 if (cursor > max_nl_size)
1229 max_nl_size = cursor;
1230 }
1231
Nick Pigginb291f002008-10-18 20:26:44 -07001232 if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 ret = SWAP_FAIL;
1234 goto out;
1235 }
1236
1237 /*
1238 * We don't try to search for this page in the nonlinear vmas,
1239 * and page_referenced wouldn't have found it anyway. Instead
1240 * just walk the nonlinear vmas trying to age and unmap some.
1241 * The mapcount of the page we came in with is irrelevant,
1242 * but even so use it as a guide to how hard we should try?
1243 */
1244 mapcount = page_mapcount(page);
1245 if (!mapcount)
1246 goto out;
1247 cond_resched_lock(&mapping->i_mmap_lock);
1248
1249 max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK;
1250 if (max_nl_cursor == 0)
1251 max_nl_cursor = CLUSTER_SIZE;
1252
1253 do {
1254 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
1255 shared.vm_set.list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 cursor = (unsigned long) vma->vm_private_data;
Hugh Dickins839b9682005-09-03 15:54:43 -07001257 while ( cursor < max_nl_cursor &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 cursor < vma->vm_end - vma->vm_start) {
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001259 if (try_to_unmap_cluster(cursor, &mapcount,
1260 vma, page) == SWAP_MLOCK)
1261 ret = SWAP_MLOCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 cursor += CLUSTER_SIZE;
1263 vma->vm_private_data = (void *) cursor;
1264 if ((int)mapcount <= 0)
1265 goto out;
1266 }
1267 vma->vm_private_data = (void *) max_nl_cursor;
1268 }
1269 cond_resched_lock(&mapping->i_mmap_lock);
1270 max_nl_cursor += CLUSTER_SIZE;
1271 } while (max_nl_cursor <= max_nl_size);
1272
1273 /*
1274 * Don't loop forever (perhaps all the remaining pages are
1275 * in locked vmas). Reset cursor on all unreserved nonlinear
1276 * vmas, now forgetting on which ones it had fallen behind.
1277 */
Hugh Dickins101d2be2005-11-21 21:32:16 -08001278 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1279 vma->vm_private_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280out:
1281 spin_unlock(&mapping->i_mmap_lock);
1282 return ret;
1283}
1284
1285/**
1286 * try_to_unmap - try to remove all page table mappings to a page
1287 * @page: the page to get unmapped
Andi Kleen14fa31b2009-09-16 11:50:10 +02001288 * @flags: action and flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 *
1290 * Tries to remove all the page table entries which are mapping this
1291 * page, used in the pageout path. Caller must hold the page lock.
1292 * Return values are:
1293 *
1294 * SWAP_SUCCESS - we succeeded in removing all mappings
1295 * SWAP_AGAIN - we missed a mapping, try again later
1296 * SWAP_FAIL - the page is unswappable
Nick Pigginb291f002008-10-18 20:26:44 -07001297 * SWAP_MLOCK - page is mlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 */
Andi Kleen14fa31b2009-09-16 11:50:10 +02001299int try_to_unmap(struct page *page, enum ttu_flags flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300{
1301 int ret;
1302
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 BUG_ON(!PageLocked(page));
1304
Hugh Dickins5ad64682009-12-14 17:59:24 -08001305 if (unlikely(PageKsm(page)))
1306 ret = try_to_unmap_ksm(page, flags);
1307 else if (PageAnon(page))
Andi Kleen14fa31b2009-09-16 11:50:10 +02001308 ret = try_to_unmap_anon(page, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 else
Andi Kleen14fa31b2009-09-16 11:50:10 +02001310 ret = try_to_unmap_file(page, flags);
Nick Pigginb291f002008-10-18 20:26:44 -07001311 if (ret != SWAP_MLOCK && !page_mapped(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 ret = SWAP_SUCCESS;
1313 return ret;
1314}
Nikita Danilov81b40822005-05-01 08:58:36 -07001315
Nick Pigginb291f002008-10-18 20:26:44 -07001316/**
1317 * try_to_munlock - try to munlock a page
1318 * @page: the page to be munlocked
1319 *
1320 * Called from munlock code. Checks all of the VMAs mapping the page
1321 * to make sure nobody else has this page mlocked. The page will be
1322 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1323 *
1324 * Return values are:
1325 *
Hugh Dickins53f79ac2009-12-14 17:58:58 -08001326 * SWAP_AGAIN - no vma is holding page mlocked, or,
Nick Pigginb291f002008-10-18 20:26:44 -07001327 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
Hugh Dickins5ad64682009-12-14 17:59:24 -08001328 * SWAP_FAIL - page cannot be located at present
Nick Pigginb291f002008-10-18 20:26:44 -07001329 * SWAP_MLOCK - page is now mlocked.
1330 */
1331int try_to_munlock(struct page *page)
1332{
1333 VM_BUG_ON(!PageLocked(page) || PageLRU(page));
1334
Hugh Dickins5ad64682009-12-14 17:59:24 -08001335 if (unlikely(PageKsm(page)))
1336 return try_to_unmap_ksm(page, TTU_MUNLOCK);
1337 else if (PageAnon(page))
Andi Kleen14fa31b2009-09-16 11:50:10 +02001338 return try_to_unmap_anon(page, TTU_MUNLOCK);
Nick Pigginb291f002008-10-18 20:26:44 -07001339 else
Andi Kleen14fa31b2009-09-16 11:50:10 +02001340 return try_to_unmap_file(page, TTU_MUNLOCK);
Nick Pigginb291f002008-10-18 20:26:44 -07001341}
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001342
1343#ifdef CONFIG_MIGRATION
1344/*
1345 * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file():
1346 * Called by migrate.c to remove migration ptes, but might be used more later.
1347 */
1348static int rmap_walk_anon(struct page *page, int (*rmap_one)(struct page *,
1349 struct vm_area_struct *, unsigned long, void *), void *arg)
1350{
1351 struct anon_vma *anon_vma;
Rik van Riel5beb4932010-03-05 13:42:07 -08001352 struct anon_vma_chain *avc;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001353 int ret = SWAP_AGAIN;
1354
1355 /*
1356 * Note: remove_migration_ptes() cannot use page_lock_anon_vma()
1357 * because that depends on page_mapped(); but not all its usages
1358 * are holding mmap_sem, which also gave the necessary guarantee
1359 * (that this anon_vma's slab has not already been destroyed).
1360 * This needs to be reviewed later: avoiding page_lock_anon_vma()
1361 * is risky, and currently limits the usefulness of rmap_walk().
1362 */
1363 anon_vma = page_anon_vma(page);
1364 if (!anon_vma)
1365 return ret;
1366 spin_lock(&anon_vma->lock);
Rik van Riel5beb4932010-03-05 13:42:07 -08001367 list_for_each_entry(avc, &anon_vma->head, same_anon_vma) {
1368 struct vm_area_struct *vma = avc->vma;
Hugh Dickinse9995ef2009-12-14 17:59:31 -08001369 unsigned long address = vma_address(page, vma);
1370 if (address == -EFAULT)
1371 continue;
1372 ret = rmap_one(page, vma, address, arg);
1373 if (ret != SWAP_AGAIN)
1374 break;
1375 }
1376 spin_unlock(&anon_vma->lock);
1377 return ret;
1378}
1379
1380static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *,
1381 struct vm_area_struct *, unsigned long, void *), void *arg)
1382{
1383 struct address_space *mapping = page->mapping;
1384 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1385 struct vm_area_struct *vma;
1386 struct prio_tree_iter iter;
1387 int ret = SWAP_AGAIN;
1388
1389 if (!mapping)
1390 return ret;
1391 spin_lock(&mapping->i_mmap_lock);
1392 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
1393 unsigned long address = vma_address(page, vma);
1394 if (address == -EFAULT)
1395 continue;
1396 ret = rmap_one(page, vma, address, arg);
1397 if (ret != SWAP_AGAIN)
1398 break;
1399 }
1400 /*
1401 * No nonlinear handling: being always shared, nonlinear vmas
1402 * never contain migration ptes. Decide what to do about this
1403 * limitation to linear when we need rmap_walk() on nonlinear.
1404 */
1405 spin_unlock(&mapping->i_mmap_lock);
1406 return ret;
1407}
1408
1409int rmap_walk(struct page *page, int (*rmap_one)(struct page *,
1410 struct vm_area_struct *, unsigned long, void *), void *arg)
1411{
1412 VM_BUG_ON(!PageLocked(page));
1413
1414 if (unlikely(PageKsm(page)))
1415 return rmap_walk_ksm(page, rmap_one, arg);
1416 else if (PageAnon(page))
1417 return rmap_walk_anon(page, rmap_one, arg);
1418 else
1419 return rmap_walk_file(page, rmap_one, arg);
1420}
1421#endif /* CONFIG_MIGRATION */