rmap.c 45.8 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
/*
 * mm/rmap.c - physical to virtual reverse mappings
 *
 * Copyright 2001, Rik van Riel <riel@conectiva.com.br>
 * Released under the General Public License (GPL).
 *
 * Simple, low overhead reverse mapping scheme.
 * Please try to keep this thing as modular as possible.
 *
 * Provides methods for unmapping each kind of mapped page:
 * the anon methods track anonymous pages, and
 * the file methods track pages belonging to an inode.
 *
 * Original design by Rik van Riel <riel@conectiva.com.br> 2001
 * File methods by Dave McCracken <dmccr@us.ibm.com> 2003, 2004
 * Anonymous methods by Andrea Arcangeli <andrea@suse.de> 2004
H
Hugh Dickins 已提交
17
 * Contributions by Hugh Dickins 2003, 2004
L
Linus Torvalds 已提交
18 19 20 21 22
 */

/*
 * Lock ordering in mm:
 *
23
 * inode->i_mutex	(while writing or truncating, not reading or faulting)
24 25 26 27 28 29 30 31 32 33
 *   inode->i_alloc_sem (vmtruncate_range)
 *   mm->mmap_sem
 *     page->flags PG_locked (lock_page)
 *       mapping->i_mmap_lock
 *         anon_vma->lock
 *           mm->page_table_lock or pte_lock
 *             zone->lru_lock (in mark_page_accessed, isolate_lru_page)
 *             swap_lock (in swap_duplicate, swap_info_get)
 *               mmlist_lock (in mmput, drain_mmlist and others)
 *               mapping->private_lock (in __set_page_dirty_buffers)
34
 *               inode->i_lock (in set_page_dirty's __mark_inode_dirty)
35
 *               inode_wb_list_lock (in set_page_dirty's __mark_inode_dirty)
36 37 38
 *                 sb_lock (within inode_lock in fs/fs-writeback.c)
 *                 mapping->tree_lock (widely used, in set_page_dirty,
 *                           in arch-dependent flush_dcache_mmap_lock,
39
 *                           within inode_wb_list_lock in __sync_single_inode)
40 41 42 43 44
 *
 * (code doesn't rely on that order so it could be switched around)
 * ->tasklist_lock
 *   anon_vma->lock      (memory_failure, collect_procs_anon)
 *     pte map lock
L
Linus Torvalds 已提交
45 46 47 48 49 50 51 52
 */

#include <linux/mm.h>
#include <linux/pagemap.h>
#include <linux/swap.h>
#include <linux/swapops.h>
#include <linux/slab.h>
#include <linux/init.h>
H
Hugh Dickins 已提交
53
#include <linux/ksm.h>
L
Linus Torvalds 已提交
54 55
#include <linux/rmap.h>
#include <linux/rcupdate.h>
56
#include <linux/module.h>
57
#include <linux/memcontrol.h>
A
Andrea Arcangeli 已提交
58
#include <linux/mmu_notifier.h>
59
#include <linux/migrate.h>
60
#include <linux/hugetlb.h>
L
Linus Torvalds 已提交
61 62 63

#include <asm/tlbflush.h>

N
Nick Piggin 已提交
64 65
#include "internal.h"

66
static struct kmem_cache *anon_vma_cachep;
67
static struct kmem_cache *anon_vma_chain_cachep;
68 69 70

static inline struct anon_vma *anon_vma_alloc(void)
{
P
Peter Zijlstra 已提交
71 72 73 74 75 76 77 78 79 80 81 82 83
	struct anon_vma *anon_vma;

	anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL);
	if (anon_vma) {
		atomic_set(&anon_vma->refcount, 1);
		/*
		 * Initialise the anon_vma root to point to itself. If called
		 * from fork, the root will be reset to the parents anon_vma.
		 */
		anon_vma->root = anon_vma;
	}

	return anon_vma;
84 85
}

P
Peter Zijlstra 已提交
86
static inline void anon_vma_free(struct anon_vma *anon_vma)
87
{
P
Peter Zijlstra 已提交
88
	VM_BUG_ON(atomic_read(&anon_vma->refcount));
89 90
	kmem_cache_free(anon_vma_cachep, anon_vma);
}
L
Linus Torvalds 已提交
91

92 93 94 95 96
static inline struct anon_vma_chain *anon_vma_chain_alloc(void)
{
	return kmem_cache_alloc(anon_vma_chain_cachep, GFP_KERNEL);
}

97
static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain)
98 99 100 101
{
	kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain);
}

102 103 104 105 106 107 108 109 110
/**
 * anon_vma_prepare - attach an anon_vma to a memory region
 * @vma: the memory region in question
 *
 * This makes sure the memory mapping described by 'vma' has
 * an 'anon_vma' attached to it, so that we can associate the
 * anonymous pages mapped into it with that anon_vma.
 *
 * The common case will be that we already have one, but if
F
Figo.zhang 已提交
111
 * not we either need to find an adjacent mapping that we
112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128
 * can re-use the anon_vma from (very common when the only
 * reason for splitting a vma has been mprotect()), or we
 * allocate a new one.
 *
 * Anon-vma allocations are very subtle, because we may have
 * optimistically looked up an anon_vma in page_lock_anon_vma()
 * and that may actually touch the spinlock even in the newly
 * allocated vma (it depends on RCU to make sure that the
 * anon_vma isn't actually destroyed).
 *
 * As a result, we need to do proper anon_vma locking even
 * for the new allocation. At the same time, we do not want
 * to do any locking for the common case of already having
 * an anon_vma.
 *
 * This must be called with the mmap_sem held for reading.
 */
L
Linus Torvalds 已提交
129 130 131
int anon_vma_prepare(struct vm_area_struct *vma)
{
	struct anon_vma *anon_vma = vma->anon_vma;
132
	struct anon_vma_chain *avc;
L
Linus Torvalds 已提交
133 134 135 136

	might_sleep();
	if (unlikely(!anon_vma)) {
		struct mm_struct *mm = vma->vm_mm;
137
		struct anon_vma *allocated;
L
Linus Torvalds 已提交
138

139 140 141 142
		avc = anon_vma_chain_alloc();
		if (!avc)
			goto out_enomem;

L
Linus Torvalds 已提交
143
		anon_vma = find_mergeable_anon_vma(vma);
144 145
		allocated = NULL;
		if (!anon_vma) {
L
Linus Torvalds 已提交
146 147
			anon_vma = anon_vma_alloc();
			if (unlikely(!anon_vma))
148
				goto out_enomem_free_avc;
L
Linus Torvalds 已提交
149 150 151
			allocated = anon_vma;
		}

152
		anon_vma_lock(anon_vma);
L
Linus Torvalds 已提交
153 154 155 156
		/* page_table_lock to protect against threads */
		spin_lock(&mm->page_table_lock);
		if (likely(!vma->anon_vma)) {
			vma->anon_vma = anon_vma;
157 158 159
			avc->anon_vma = anon_vma;
			avc->vma = vma;
			list_add(&avc->same_vma, &vma->anon_vma_chain);
160
			list_add_tail(&avc->same_anon_vma, &anon_vma->head);
L
Linus Torvalds 已提交
161
			allocated = NULL;
162
			avc = NULL;
L
Linus Torvalds 已提交
163 164
		}
		spin_unlock(&mm->page_table_lock);
165
		anon_vma_unlock(anon_vma);
166 167

		if (unlikely(allocated))
P
Peter Zijlstra 已提交
168
			put_anon_vma(allocated);
169
		if (unlikely(avc))
170
			anon_vma_chain_free(avc);
L
Linus Torvalds 已提交
171 172
	}
	return 0;
173 174 175 176 177

 out_enomem_free_avc:
	anon_vma_chain_free(avc);
 out_enomem:
	return -ENOMEM;
L
Linus Torvalds 已提交
178 179
}

180 181 182
static void anon_vma_chain_link(struct vm_area_struct *vma,
				struct anon_vma_chain *avc,
				struct anon_vma *anon_vma)
L
Linus Torvalds 已提交
183
{
184 185 186 187
	avc->vma = vma;
	avc->anon_vma = anon_vma;
	list_add(&avc->same_vma, &vma->anon_vma_chain);

188
	anon_vma_lock(anon_vma);
189 190 191 192
	/*
	 * It's critical to add new vmas to the tail of the anon_vma,
	 * see comment in huge_memory.c:__split_huge_page().
	 */
193
	list_add_tail(&avc->same_anon_vma, &anon_vma->head);
194
	anon_vma_unlock(anon_vma);
L
Linus Torvalds 已提交
195 196
}

197 198 199 200 201
/*
 * Attach the anon_vmas from src to dst.
 * Returns 0 on success, -ENOMEM on failure.
 */
int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src)
L
Linus Torvalds 已提交
202
{
203 204
	struct anon_vma_chain *avc, *pavc;

205
	list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) {
206 207 208 209 210 211
		avc = anon_vma_chain_alloc();
		if (!avc)
			goto enomem_failure;
		anon_vma_chain_link(dst, avc, pavc->anon_vma);
	}
	return 0;
L
Linus Torvalds 已提交
212

213 214 215
 enomem_failure:
	unlink_anon_vmas(dst);
	return -ENOMEM;
L
Linus Torvalds 已提交
216 217
}

218 219 220 221 222 223
/*
 * Attach vma to its own anon_vma, as well as to the anon_vmas that
 * the corresponding VMA in the parent process is attached to.
 * Returns 0 on success, non-zero on failure.
 */
int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma)
L
Linus Torvalds 已提交
224
{
225 226
	struct anon_vma_chain *avc;
	struct anon_vma *anon_vma;
L
Linus Torvalds 已提交
227

228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245
	/* Don't bother if the parent process has no anon_vma here. */
	if (!pvma->anon_vma)
		return 0;

	/*
	 * First, attach the new VMA to the parent VMA's anon_vmas,
	 * so rmap can find non-COWed pages in child processes.
	 */
	if (anon_vma_clone(vma, pvma))
		return -ENOMEM;

	/* Then add our own anon_vma. */
	anon_vma = anon_vma_alloc();
	if (!anon_vma)
		goto out_error;
	avc = anon_vma_chain_alloc();
	if (!avc)
		goto out_error_free_anon_vma;
246 247 248 249 250 251

	/*
	 * The root anon_vma's spinlock is the lock actually used when we
	 * lock any of the anon_vmas in this anon_vma tree.
	 */
	anon_vma->root = pvma->anon_vma->root;
252
	/*
P
Peter Zijlstra 已提交
253 254 255
	 * With refcounts, an anon_vma can stay around longer than the
	 * process it belongs to. The root anon_vma needs to be pinned until
	 * this anon_vma is freed, because the lock lives in the root.
256 257
	 */
	get_anon_vma(anon_vma->root);
258 259
	/* Mark this anon_vma as the one where our new (COWed) pages go. */
	vma->anon_vma = anon_vma;
260
	anon_vma_chain_link(vma, avc, anon_vma);
261 262 263 264

	return 0;

 out_error_free_anon_vma:
P
Peter Zijlstra 已提交
265
	put_anon_vma(anon_vma);
266
 out_error:
267
	unlink_anon_vmas(vma);
268
	return -ENOMEM;
L
Linus Torvalds 已提交
269 270
}

271
static void anon_vma_unlink(struct anon_vma_chain *anon_vma_chain)
L
Linus Torvalds 已提交
272
{
273
	struct anon_vma *anon_vma = anon_vma_chain->anon_vma;
L
Linus Torvalds 已提交
274 275
	int empty;

276
	/* If anon_vma_fork fails, we can get an empty anon_vma_chain. */
L
Linus Torvalds 已提交
277 278 279
	if (!anon_vma)
		return;

280
	anon_vma_lock(anon_vma);
281
	list_del(&anon_vma_chain->same_anon_vma);
L
Linus Torvalds 已提交
282 283

	/* We must garbage collect the anon_vma if it's empty */
P
Peter Zijlstra 已提交
284
	empty = list_empty(&anon_vma->head);
285
	anon_vma_unlock(anon_vma);
L
Linus Torvalds 已提交
286

P
Peter Zijlstra 已提交
287 288
	if (empty)
		put_anon_vma(anon_vma);
L
Linus Torvalds 已提交
289 290
}

291 292 293 294
void unlink_anon_vmas(struct vm_area_struct *vma)
{
	struct anon_vma_chain *avc, *next;

295 296 297 298
	/*
	 * Unlink each anon_vma chained to the VMA.  This list is ordered
	 * from newest to oldest, ensuring the root anon_vma gets freed last.
	 */
299 300 301 302 303 304 305
	list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
		anon_vma_unlink(avc);
		list_del(&avc->same_vma);
		anon_vma_chain_free(avc);
	}
}

306
static void anon_vma_ctor(void *data)
L
Linus Torvalds 已提交
307
{
C
Christoph Lameter 已提交
308
	struct anon_vma *anon_vma = data;
L
Linus Torvalds 已提交
309

C
Christoph Lameter 已提交
310
	spin_lock_init(&anon_vma->lock);
311
	atomic_set(&anon_vma->refcount, 0);
C
Christoph Lameter 已提交
312
	INIT_LIST_HEAD(&anon_vma->head);
L
Linus Torvalds 已提交
313 314 315 316 317
}

void __init anon_vma_init(void)
{
	anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
318
			0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor);
319
	anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC);
L
Linus Torvalds 已提交
320 321 322 323 324 325
}

/*
 * Getting a lock on a stable anon_vma from a page off the LRU is
 * tricky: page_lock_anon_vma rely on RCU to guard against the races.
 */
326
struct anon_vma *__page_lock_anon_vma(struct page *page)
L
Linus Torvalds 已提交
327
{
328
	struct anon_vma *anon_vma, *root_anon_vma;
L
Linus Torvalds 已提交
329 330 331
	unsigned long anon_mapping;

	rcu_read_lock();
332
	anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping);
H
Hugh Dickins 已提交
333
	if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
L
Linus Torvalds 已提交
334 335 336 337 338
		goto out;
	if (!page_mapped(page))
		goto out;

	anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
339 340 341 342 343 344 345 346 347 348 349 350 351 352 353
	root_anon_vma = ACCESS_ONCE(anon_vma->root);
	spin_lock(&root_anon_vma->lock);

	/*
	 * If this page is still mapped, then its anon_vma cannot have been
	 * freed.  But if it has been unmapped, we have no security against
	 * the anon_vma structure being freed and reused (for another anon_vma:
	 * SLAB_DESTROY_BY_RCU guarantees that - so the spin_lock above cannot
	 * corrupt): with anon_vma_prepare() or anon_vma_fork() redirecting
	 * anon_vma->root before page_unlock_anon_vma() is called to unlock.
	 */
	if (page_mapped(page))
		return anon_vma;

	spin_unlock(&root_anon_vma->lock);
L
Linus Torvalds 已提交
354 355
out:
	rcu_read_unlock();
356 357 358
	return NULL;
}

359
void page_unlock_anon_vma(struct anon_vma *anon_vma)
360 361
	__releases(&anon_vma->root->lock)
	__releases(RCU)
362
{
363
	anon_vma_unlock(anon_vma);
364
	rcu_read_unlock();
L
Linus Torvalds 已提交
365 366 367
}

/*
368 369 370
 * At what user virtual address is page expected in @vma?
 * Returns virtual address or -EFAULT if page's index/offset is not
 * within the range mapped the @vma.
L
Linus Torvalds 已提交
371
 */
372
inline unsigned long
L
Linus Torvalds 已提交
373 374 375 376 377
vma_address(struct page *page, struct vm_area_struct *vma)
{
	pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
	unsigned long address;

378 379
	if (unlikely(is_vm_hugetlb_page(vma)))
		pgoff = page->index << huge_page_order(page_hstate(page));
L
Linus Torvalds 已提交
380 381
	address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
	if (unlikely(address < vma->vm_start || address >= vma->vm_end)) {
382
		/* page should be within @vma mapping range */
L
Linus Torvalds 已提交
383 384 385 386 387 388
		return -EFAULT;
	}
	return address;
}

/*
H
Huang Shijie 已提交
389
 * At what user virtual address is page expected in vma?
390
 * Caller should check the page is actually part of the vma.
L
Linus Torvalds 已提交
391 392 393
 */
unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
{
394
	if (PageAnon(page)) {
395 396 397 398 399 400 401
		struct anon_vma *page__anon_vma = page_anon_vma(page);
		/*
		 * Note: swapoff's unuse_vma() is more efficient with this
		 * check, and needs it to match anon_vma when KSM is active.
		 */
		if (!vma->anon_vma || !page__anon_vma ||
		    vma->anon_vma->root != page__anon_vma->root)
402 403
			return -EFAULT;
	} else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) {
H
Hugh Dickins 已提交
404 405
		if (!vma->vm_file ||
		    vma->vm_file->f_mapping != page->mapping)
L
Linus Torvalds 已提交
406 407 408 409 410 411
			return -EFAULT;
	} else
		return -EFAULT;
	return vma_address(page, vma);
}

N
Nikita Danilov 已提交
412 413 414
/*
 * Check that @page is mapped at @address into @mm.
 *
N
Nick Piggin 已提交
415 416 417 418
 * If @sync is false, page_check_address may perform a racy check to avoid
 * the page table lock when the pte is not present (helpful when reclaiming
 * highly shared pages).
 *
419
 * On success returns with pte mapped and locked.
N
Nikita Danilov 已提交
420
 */
421
pte_t *__page_check_address(struct page *page, struct mm_struct *mm,
N
Nick Piggin 已提交
422
			  unsigned long address, spinlock_t **ptlp, int sync)
N
Nikita Danilov 已提交
423 424 425 426 427
{
	pgd_t *pgd;
	pud_t *pud;
	pmd_t *pmd;
	pte_t *pte;
H
Hugh Dickins 已提交
428
	spinlock_t *ptl;
N
Nikita Danilov 已提交
429

430 431 432 433 434 435
	if (unlikely(PageHuge(page))) {
		pte = huge_pte_offset(mm, address);
		ptl = &mm->page_table_lock;
		goto check;
	}

N
Nikita Danilov 已提交
436
	pgd = pgd_offset(mm, address);
H
Hugh Dickins 已提交
437 438 439 440 441 442 443 444 445 446
	if (!pgd_present(*pgd))
		return NULL;

	pud = pud_offset(pgd, address);
	if (!pud_present(*pud))
		return NULL;

	pmd = pmd_offset(pud, address);
	if (!pmd_present(*pmd))
		return NULL;
447 448
	if (pmd_trans_huge(*pmd))
		return NULL;
H
Hugh Dickins 已提交
449 450 451

	pte = pte_offset_map(pmd, address);
	/* Make a quick check before getting the lock */
N
Nick Piggin 已提交
452
	if (!sync && !pte_present(*pte)) {
H
Hugh Dickins 已提交
453 454 455 456
		pte_unmap(pte);
		return NULL;
	}

H
Hugh Dickins 已提交
457
	ptl = pte_lockptr(mm, pmd);
458
check:
H
Hugh Dickins 已提交
459 460 461 462
	spin_lock(ptl);
	if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
		*ptlp = ptl;
		return pte;
N
Nikita Danilov 已提交
463
	}
H
Hugh Dickins 已提交
464 465
	pte_unmap_unlock(pte, ptl);
	return NULL;
N
Nikita Danilov 已提交
466 467
}

N
Nick Piggin 已提交
468 469 470 471 472 473 474 475 476
/**
 * page_mapped_in_vma - check whether a page is really mapped in a VMA
 * @page: the page to test
 * @vma: the VMA to test
 *
 * Returns 1 if the page is mapped into the page tables of the VMA, 0
 * if the page is not mapped into the page tables of this VMA.  Only
 * valid for normal file or anonymous VMAs.
 */
477
int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
N
Nick Piggin 已提交
478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493
{
	unsigned long address;
	pte_t *pte;
	spinlock_t *ptl;

	address = vma_address(page, vma);
	if (address == -EFAULT)		/* out of vma range */
		return 0;
	pte = page_check_address(page, vma->vm_mm, address, &ptl, 1);
	if (!pte)			/* the page is not in this mm */
		return 0;
	pte_unmap_unlock(pte, ptl);

	return 1;
}

L
Linus Torvalds 已提交
494 495 496 497
/*
 * Subfunctions of page_referenced: page_referenced_one called
 * repeatedly from either page_referenced_anon or page_referenced_file.
 */
H
Hugh Dickins 已提交
498 499 500
int page_referenced_one(struct page *page, struct vm_area_struct *vma,
			unsigned long address, unsigned int *mapcount,
			unsigned long *vm_flags)
L
Linus Torvalds 已提交
501 502 503 504
{
	struct mm_struct *mm = vma->vm_mm;
	int referenced = 0;

505 506 507 508
	if (unlikely(PageTransHuge(page))) {
		pmd_t *pmd;

		spin_lock(&mm->page_table_lock);
509 510 511 512
		/*
		 * rmap might return false positives; we must filter
		 * these out using page_check_address_pmd().
		 */
513 514
		pmd = page_check_address_pmd(page, mm, address,
					     PAGE_CHECK_ADDRESS_PMD_FLAG);
515 516 517 518 519 520 521 522 523 524 525 526 527 528
		if (!pmd) {
			spin_unlock(&mm->page_table_lock);
			goto out;
		}

		if (vma->vm_flags & VM_LOCKED) {
			spin_unlock(&mm->page_table_lock);
			*mapcount = 0;	/* break early from loop */
			*vm_flags |= VM_LOCKED;
			goto out;
		}

		/* go ahead even if the pmd is pmd_trans_splitting() */
		if (pmdp_clear_flush_young_notify(vma, address, pmd))
529 530 531 532 533 534
			referenced++;
		spin_unlock(&mm->page_table_lock);
	} else {
		pte_t *pte;
		spinlock_t *ptl;

535 536 537 538
		/*
		 * rmap might return false positives; we must filter
		 * these out using page_check_address().
		 */
539 540 541 542
		pte = page_check_address(page, mm, address, &ptl, 0);
		if (!pte)
			goto out;

543 544 545 546 547 548 549
		if (vma->vm_flags & VM_LOCKED) {
			pte_unmap_unlock(pte, ptl);
			*mapcount = 0;	/* break early from loop */
			*vm_flags |= VM_LOCKED;
			goto out;
		}

550 551 552 553 554 555 556 557 558 559 560 561 562 563
		if (ptep_clear_flush_young_notify(vma, address, pte)) {
			/*
			 * Don't treat a reference through a sequentially read
			 * mapping as such.  If the page has been used in
			 * another mapping, we will catch it; if this other
			 * mapping is already gone, the unmap path will have
			 * set PG_referenced or activated the page.
			 */
			if (likely(!VM_SequentialReadHint(vma)))
				referenced++;
		}
		pte_unmap_unlock(pte, ptl);
	}

564 565 566 567 568 569
	/* Pretend the page is referenced if the task has the
	   swap token and is in the middle of a page fault. */
	if (mm != current->mm && has_swap_token(mm) &&
			rwsem_is_locked(&mm->mmap_sem))
		referenced++;

H
Hugh Dickins 已提交
570
	(*mapcount)--;
571

572 573
	if (referenced)
		*vm_flags |= vma->vm_flags;
574
out:
L
Linus Torvalds 已提交
575 576 577
	return referenced;
}

578
static int page_referenced_anon(struct page *page,
579 580
				struct mem_cgroup *mem_cont,
				unsigned long *vm_flags)
L
Linus Torvalds 已提交
581 582 583
{
	unsigned int mapcount;
	struct anon_vma *anon_vma;
584
	struct anon_vma_chain *avc;
L
Linus Torvalds 已提交
585 586 587 588 589 590 591
	int referenced = 0;

	anon_vma = page_lock_anon_vma(page);
	if (!anon_vma)
		return referenced;

	mapcount = page_mapcount(page);
592 593
	list_for_each_entry(avc, &anon_vma->head, same_anon_vma) {
		struct vm_area_struct *vma = avc->vma;
H
Hugh Dickins 已提交
594 595 596
		unsigned long address = vma_address(page, vma);
		if (address == -EFAULT)
			continue;
597 598 599 600 601
		/*
		 * If we are reclaiming on behalf of a cgroup, skip
		 * counting on behalf of references from different
		 * cgroups
		 */
602
		if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont))
603
			continue;
H
Hugh Dickins 已提交
604
		referenced += page_referenced_one(page, vma, address,
605
						  &mapcount, vm_flags);
L
Linus Torvalds 已提交
606 607 608
		if (!mapcount)
			break;
	}
609 610

	page_unlock_anon_vma(anon_vma);
L
Linus Torvalds 已提交
611 612 613 614 615 616
	return referenced;
}

/**
 * page_referenced_file - referenced check for object-based rmap
 * @page: the page we're checking references on.
R
Randy Dunlap 已提交
617
 * @mem_cont: target memory controller
618
 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
L
Linus Torvalds 已提交
619 620 621 622 623 624 625 626
 *
 * For an object-based mapped page, find all the places it is mapped and
 * check/clear the referenced flag.  This is done by following the page->mapping
 * pointer, then walking the chain of vmas it holds.  It returns the number
 * of references it found.
 *
 * This function is only called from page_referenced for object-based pages.
 */
627
static int page_referenced_file(struct page *page,
628 629
				struct mem_cgroup *mem_cont,
				unsigned long *vm_flags)
L
Linus Torvalds 已提交
630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661
{
	unsigned int mapcount;
	struct address_space *mapping = page->mapping;
	pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
	struct vm_area_struct *vma;
	struct prio_tree_iter iter;
	int referenced = 0;

	/*
	 * The caller's checks on page->mapping and !PageAnon have made
	 * sure that this is a file page: the check for page->mapping
	 * excludes the case just before it gets set on an anon page.
	 */
	BUG_ON(PageAnon(page));

	/*
	 * The page lock not only makes sure that page->mapping cannot
	 * suddenly be NULLified by truncation, it makes sure that the
	 * structure at mapping cannot be freed and reused yet,
	 * so we can safely take mapping->i_mmap_lock.
	 */
	BUG_ON(!PageLocked(page));

	spin_lock(&mapping->i_mmap_lock);

	/*
	 * i_mmap_lock does not stabilize mapcount at all, but mapcount
	 * is more likely to be accurate if we note it after spinning.
	 */
	mapcount = page_mapcount(page);

	vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
H
Hugh Dickins 已提交
662 663 664
		unsigned long address = vma_address(page, vma);
		if (address == -EFAULT)
			continue;
665 666 667 668 669
		/*
		 * If we are reclaiming on behalf of a cgroup, skip
		 * counting on behalf of references from different
		 * cgroups
		 */
670
		if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont))
671
			continue;
H
Hugh Dickins 已提交
672
		referenced += page_referenced_one(page, vma, address,
673
						  &mapcount, vm_flags);
L
Linus Torvalds 已提交
674 675 676 677 678 679 680 681 682 683 684 685
		if (!mapcount)
			break;
	}

	spin_unlock(&mapping->i_mmap_lock);
	return referenced;
}

/**
 * page_referenced - test if the page was referenced
 * @page: the page to test
 * @is_locked: caller holds lock on the page
R
Randy Dunlap 已提交
686
 * @mem_cont: target memory controller
687
 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
L
Linus Torvalds 已提交
688 689 690 691
 *
 * Quick test_and_clear_referenced for all mappings to a page,
 * returns the number of ptes which referenced the page.
 */
692 693 694 695
int page_referenced(struct page *page,
		    int is_locked,
		    struct mem_cgroup *mem_cont,
		    unsigned long *vm_flags)
L
Linus Torvalds 已提交
696 697
{
	int referenced = 0;
H
Hugh Dickins 已提交
698
	int we_locked = 0;
L
Linus Torvalds 已提交
699

700
	*vm_flags = 0;
H
Hugh Dickins 已提交
701
	if (page_mapped(page) && page_rmapping(page)) {
H
Hugh Dickins 已提交
702 703 704 705 706 707 708 709 710 711 712
		if (!is_locked && (!PageAnon(page) || PageKsm(page))) {
			we_locked = trylock_page(page);
			if (!we_locked) {
				referenced++;
				goto out;
			}
		}
		if (unlikely(PageKsm(page)))
			referenced += page_referenced_ksm(page, mem_cont,
								vm_flags);
		else if (PageAnon(page))
713 714
			referenced += page_referenced_anon(page, mem_cont,
								vm_flags);
H
Hugh Dickins 已提交
715
		else if (page->mapping)
716 717
			referenced += page_referenced_file(page, mem_cont,
								vm_flags);
H
Hugh Dickins 已提交
718
		if (we_locked)
L
Linus Torvalds 已提交
719 720
			unlock_page(page);
	}
H
Hugh Dickins 已提交
721
out:
722
	if (page_test_and_clear_young(page_to_pfn(page)))
723 724
		referenced++;

L
Linus Torvalds 已提交
725 726 727
	return referenced;
}

H
Hugh Dickins 已提交
728 729
static int page_mkclean_one(struct page *page, struct vm_area_struct *vma,
			    unsigned long address)
730 731
{
	struct mm_struct *mm = vma->vm_mm;
732
	pte_t *pte;
733 734 735
	spinlock_t *ptl;
	int ret = 0;

N
Nick Piggin 已提交
736
	pte = page_check_address(page, mm, address, &ptl, 1);
737 738 739
	if (!pte)
		goto out;

740 741
	if (pte_dirty(*pte) || pte_write(*pte)) {
		pte_t entry;
742

743
		flush_cache_page(vma, address, pte_pfn(*pte));
A
Andrea Arcangeli 已提交
744
		entry = ptep_clear_flush_notify(vma, address, pte);
745 746
		entry = pte_wrprotect(entry);
		entry = pte_mkclean(entry);
747
		set_pte_at(mm, address, pte, entry);
748 749
		ret = 1;
	}
750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766

	pte_unmap_unlock(pte, ptl);
out:
	return ret;
}

static int page_mkclean_file(struct address_space *mapping, struct page *page)
{
	pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
	struct vm_area_struct *vma;
	struct prio_tree_iter iter;
	int ret = 0;

	BUG_ON(PageAnon(page));

	spin_lock(&mapping->i_mmap_lock);
	vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
H
Hugh Dickins 已提交
767 768 769 770 771 772
		if (vma->vm_flags & VM_SHARED) {
			unsigned long address = vma_address(page, vma);
			if (address == -EFAULT)
				continue;
			ret += page_mkclean_one(page, vma, address);
		}
773 774 775 776 777 778 779 780 781 782 783 784 785
	}
	spin_unlock(&mapping->i_mmap_lock);
	return ret;
}

int page_mkclean(struct page *page)
{
	int ret = 0;

	BUG_ON(!PageLocked(page));

	if (page_mapped(page)) {
		struct address_space *mapping = page_mapping(page);
786
		if (mapping) {
787
			ret = page_mkclean_file(mapping, page);
788
			if (page_test_and_clear_dirty(page_to_pfn(page), 1))
789
				ret = 1;
790
		}
791 792 793 794
	}

	return ret;
}
J
Jaya Kumar 已提交
795
EXPORT_SYMBOL_GPL(page_mkclean);
796

797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820
/**
 * page_move_anon_rmap - move a page to our anon_vma
 * @page:	the page to move to our anon_vma
 * @vma:	the vma the page belongs to
 * @address:	the user virtual address mapped
 *
 * When a page belongs exclusively to one process after a COW event,
 * that page can be moved into the anon_vma that belongs to just that
 * process, so the rmap code will not search the parent or sibling
 * processes.
 */
void page_move_anon_rmap(struct page *page,
	struct vm_area_struct *vma, unsigned long address)
{
	struct anon_vma *anon_vma = vma->anon_vma;

	VM_BUG_ON(!PageLocked(page));
	VM_BUG_ON(!anon_vma);
	VM_BUG_ON(page->index != linear_page_index(vma, address));

	anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
	page->mapping = (struct address_space *) anon_vma;
}

N
Nick Piggin 已提交
821
/**
A
Andi Kleen 已提交
822 823 824 825
 * __page_set_anon_rmap - set up new anonymous rmap
 * @page:	Page to add to rmap	
 * @vma:	VM area to add page to.
 * @address:	User virtual address of the mapping	
826
 * @exclusive:	the page is exclusively owned by the current process
N
Nick Piggin 已提交
827 828
 */
static void __page_set_anon_rmap(struct page *page,
829
	struct vm_area_struct *vma, unsigned long address, int exclusive)
N
Nick Piggin 已提交
830
{
831
	struct anon_vma *anon_vma = vma->anon_vma;
832

833
	BUG_ON(!anon_vma);
834

A
Andi Kleen 已提交
835 836 837
	if (PageAnon(page))
		return;

838
	/*
839 840 841
	 * If the page isn't exclusively mapped into this vma,
	 * we must use the _oldest_ possible anon_vma for the
	 * page mapping!
842
	 */
A
Andi Kleen 已提交
843
	if (!exclusive)
844
		anon_vma = anon_vma->root;
N
Nick Piggin 已提交
845 846 847 848 849 850

	anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
	page->mapping = (struct address_space *) anon_vma;
	page->index = linear_page_index(vma, address);
}

N
Nick Piggin 已提交
851
/**
R
Randy Dunlap 已提交
852
 * __page_check_anon_rmap - sanity check anonymous rmap addition
N
Nick Piggin 已提交
853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872
 * @page:	the page to add the mapping to
 * @vma:	the vm area in which the mapping is added
 * @address:	the user virtual address mapped
 */
static void __page_check_anon_rmap(struct page *page,
	struct vm_area_struct *vma, unsigned long address)
{
#ifdef CONFIG_DEBUG_VM
	/*
	 * The page's anon-rmap details (mapping and index) are guaranteed to
	 * be set up correctly at this point.
	 *
	 * We have exclusion against page_add_anon_rmap because the caller
	 * always holds the page locked, except if called from page_dup_rmap,
	 * in which case the page is already known to be setup.
	 *
	 * We have exclusion against page_add_new_anon_rmap because those pages
	 * are initially only visible via the pagetables, and the pte is locked
	 * over the call to page_add_new_anon_rmap.
	 */
A
Andrea Arcangeli 已提交
873
	BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root);
N
Nick Piggin 已提交
874 875 876 877
	BUG_ON(page->index != linear_page_index(vma, address));
#endif
}

L
Linus Torvalds 已提交
878 879 880 881 882 883
/**
 * page_add_anon_rmap - add pte mapping to an anonymous page
 * @page:	the page to add the mapping to
 * @vma:	the vm area in which the mapping is added
 * @address:	the user virtual address mapped
 *
H
Hugh Dickins 已提交
884
 * The caller needs to hold the pte lock, and the page must be locked in
885 886 887
 * the anon_vma case: to serialize mapping,index checking after setting,
 * and to ensure that PageAnon is not being upgraded racily to PageKsm
 * (but PageKsm is never downgraded to PageAnon).
L
Linus Torvalds 已提交
888 889 890
 */
void page_add_anon_rmap(struct page *page,
	struct vm_area_struct *vma, unsigned long address)
891 892 893 894 895 896 897 898 899 900 901
{
	do_page_add_anon_rmap(page, vma, address, 0);
}

/*
 * Special version of the above for do_swap_page, which often runs
 * into pages that are exclusively owned by the current process.
 * Everybody else should continue to use page_add_anon_rmap above.
 */
void do_page_add_anon_rmap(struct page *page,
	struct vm_area_struct *vma, unsigned long address, int exclusive)
L
Linus Torvalds 已提交
902
{
H
Hugh Dickins 已提交
903
	int first = atomic_inc_and_test(&page->_mapcount);
904 905 906 907 908 909 910
	if (first) {
		if (!PageTransHuge(page))
			__inc_zone_page_state(page, NR_ANON_PAGES);
		else
			__inc_zone_page_state(page,
					      NR_ANON_TRANSPARENT_HUGEPAGES);
	}
H
Hugh Dickins 已提交
911 912 913
	if (unlikely(PageKsm(page)))
		return;

N
Nick Piggin 已提交
914 915
	VM_BUG_ON(!PageLocked(page));
	VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
H
Hugh Dickins 已提交
916
	if (first)
917
		__page_set_anon_rmap(page, vma, address, exclusive);
918
	else
N
Nick Piggin 已提交
919
		__page_check_anon_rmap(page, vma, address);
L
Linus Torvalds 已提交
920 921
}

R
Randy Dunlap 已提交
922
/**
N
Nick Piggin 已提交
923 924 925 926 927 928 929
 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
 * @page:	the page to add the mapping to
 * @vma:	the vm area in which the mapping is added
 * @address:	the user virtual address mapped
 *
 * Same as page_add_anon_rmap but must only be called on *new* pages.
 * This means the inc-and-test can be bypassed.
N
Nick Piggin 已提交
930
 * Page does not have to be locked.
N
Nick Piggin 已提交
931 932 933 934
 */
void page_add_new_anon_rmap(struct page *page,
	struct vm_area_struct *vma, unsigned long address)
{
935
	VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
936 937
	SetPageSwapBacked(page);
	atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */
938 939 940 941
	if (!PageTransHuge(page))
		__inc_zone_page_state(page, NR_ANON_PAGES);
	else
		__inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
942
	__page_set_anon_rmap(page, vma, address, 1);
943
	if (page_evictable(page, vma))
944
		lru_cache_add_lru(page, LRU_ACTIVE_ANON);
945 946
	else
		add_page_to_unevictable_list(page);
N
Nick Piggin 已提交
947 948
}

L
Linus Torvalds 已提交
949 950 951 952
/**
 * page_add_file_rmap - add pte mapping to a file page
 * @page: the page to add the mapping to
 *
953
 * The caller needs to hold the pte lock.
L
Linus Torvalds 已提交
954 955 956
 */
void page_add_file_rmap(struct page *page)
{
957
	if (atomic_inc_and_test(&page->_mapcount)) {
958
		__inc_zone_page_state(page, NR_FILE_MAPPED);
959
		mem_cgroup_inc_page_stat(page, MEMCG_NR_FILE_MAPPED);
960
	}
L
Linus Torvalds 已提交
961 962 963 964 965 966
}

/**
 * page_remove_rmap - take down pte mapping from a page
 * @page: page to remove mapping from
 *
967
 * The caller needs to hold the pte lock.
L
Linus Torvalds 已提交
968
 */
969
void page_remove_rmap(struct page *page)
L
Linus Torvalds 已提交
970
{
971 972 973 974 975 976 977 978 979 980 981
	/* page still mapped by someone else? */
	if (!atomic_add_negative(-1, &page->_mapcount))
		return;

	/*
	 * Now that the last pte has gone, s390 must transfer dirty
	 * flag from storage key to struct page.  We can usually skip
	 * this if the page is anon, so about to be freed; but perhaps
	 * not if it's in swapcache - there might be another pte slot
	 * containing the swap entry, but page not yet written to swap.
	 */
982 983
	if ((!PageAnon(page) || PageSwapCache(page)) &&
	    page_test_and_clear_dirty(page_to_pfn(page), 1))
984
		set_page_dirty(page);
985 986 987 988 989 990
	/*
	 * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED
	 * and not charged by memcg for now.
	 */
	if (unlikely(PageHuge(page)))
		return;
991 992
	if (PageAnon(page)) {
		mem_cgroup_uncharge_page(page);
993 994 995 996 997
		if (!PageTransHuge(page))
			__dec_zone_page_state(page, NR_ANON_PAGES);
		else
			__dec_zone_page_state(page,
					      NR_ANON_TRANSPARENT_HUGEPAGES);
998 999
	} else {
		__dec_zone_page_state(page, NR_FILE_MAPPED);
1000
		mem_cgroup_dec_page_stat(page, MEMCG_NR_FILE_MAPPED);
1001 1002 1003 1004 1005 1006 1007 1008 1009 1010
	}
	/*
	 * It would be tidy to reset the PageAnon mapping here,
	 * but that might overwrite a racing page_add_anon_rmap
	 * which increments mapcount after us but sets mapping
	 * before us: so leave the reset to free_hot_cold_page,
	 * and remember that it's only reliable while mapped.
	 * Leaving it set also helps swapoff to reinstate ptes
	 * faster for those pages still in swapcache.
	 */
L
Linus Torvalds 已提交
1011 1012 1013 1014 1015 1016
}

/*
 * Subfunctions of try_to_unmap: try_to_unmap_one called
 * repeatedly from either try_to_unmap_anon or try_to_unmap_file.
 */
H
Hugh Dickins 已提交
1017 1018
int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
		     unsigned long address, enum ttu_flags flags)
L
Linus Torvalds 已提交
1019 1020 1021 1022
{
	struct mm_struct *mm = vma->vm_mm;
	pte_t *pte;
	pte_t pteval;
H
Hugh Dickins 已提交
1023
	spinlock_t *ptl;
L
Linus Torvalds 已提交
1024 1025
	int ret = SWAP_AGAIN;

N
Nick Piggin 已提交
1026
	pte = page_check_address(page, mm, address, &ptl, 0);
H
Hugh Dickins 已提交
1027
	if (!pte)
N
Nikita Danilov 已提交
1028
		goto out;
L
Linus Torvalds 已提交
1029 1030 1031 1032 1033 1034

	/*
	 * If the page is mlock()d, we cannot swap it out.
	 * If it's recently referenced (perhaps page_referenced
	 * skipped over this mm) then we should reactivate it.
	 */
1035
	if (!(flags & TTU_IGNORE_MLOCK)) {
K
KOSAKI Motohiro 已提交
1036 1037 1038
		if (vma->vm_flags & VM_LOCKED)
			goto out_mlock;

H
Hugh Dickins 已提交
1039
		if (TTU_ACTION(flags) == TTU_MUNLOCK)
H
Hugh Dickins 已提交
1040
			goto out_unmap;
1041 1042
	}
	if (!(flags & TTU_IGNORE_ACCESS)) {
N
Nick Piggin 已提交
1043 1044 1045 1046 1047
		if (ptep_clear_flush_young_notify(vma, address, pte)) {
			ret = SWAP_FAIL;
			goto out_unmap;
		}
  	}
L
Linus Torvalds 已提交
1048 1049 1050

	/* Nuke the page table entry. */
	flush_cache_page(vma, address, page_to_pfn(page));
A
Andrea Arcangeli 已提交
1051
	pteval = ptep_clear_flush_notify(vma, address, pte);
L
Linus Torvalds 已提交
1052 1053 1054 1055 1056

	/* Move the dirty bit to the physical page now the pte is gone. */
	if (pte_dirty(pteval))
		set_page_dirty(page);

1057 1058 1059
	/* Update high watermark before we lower rss */
	update_hiwater_rss(mm);

1060 1061
	if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
		if (PageAnon(page))
K
KAMEZAWA Hiroyuki 已提交
1062
			dec_mm_counter(mm, MM_ANONPAGES);
1063
		else
K
KAMEZAWA Hiroyuki 已提交
1064
			dec_mm_counter(mm, MM_FILEPAGES);
1065 1066 1067
		set_pte_at(mm, address, pte,
				swp_entry_to_pte(make_hwpoison_entry(page)));
	} else if (PageAnon(page)) {
H
Hugh Dickins 已提交
1068
		swp_entry_t entry = { .val = page_private(page) };
1069 1070 1071 1072 1073 1074

		if (PageSwapCache(page)) {
			/*
			 * Store the swap location in the pte.
			 * See handle_pte_fault() ...
			 */
H
Hugh Dickins 已提交
1075 1076 1077 1078 1079
			if (swap_duplicate(entry) < 0) {
				set_pte_at(mm, address, pte, pteval);
				ret = SWAP_FAIL;
				goto out_unmap;
			}
1080 1081 1082 1083 1084 1085
			if (list_empty(&mm->mmlist)) {
				spin_lock(&mmlist_lock);
				if (list_empty(&mm->mmlist))
					list_add(&mm->mmlist, &init_mm.mmlist);
				spin_unlock(&mmlist_lock);
			}
K
KAMEZAWA Hiroyuki 已提交
1086
			dec_mm_counter(mm, MM_ANONPAGES);
K
KAMEZAWA Hiroyuki 已提交
1087
			inc_mm_counter(mm, MM_SWAPENTS);
1088
		} else if (PAGE_MIGRATION) {
1089 1090 1091 1092 1093
			/*
			 * Store the pfn of the page in a special migration
			 * pte. do_swap_page() will wait until the migration
			 * pte is removed and then restart fault handling.
			 */
1094
			BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION);
1095
			entry = make_migration_entry(page, pte_write(pteval));
L
Linus Torvalds 已提交
1096 1097 1098
		}
		set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
		BUG_ON(pte_file(*pte));
1099
	} else if (PAGE_MIGRATION && (TTU_ACTION(flags) == TTU_MIGRATION)) {
1100 1101 1102 1103 1104
		/* Establish migration entry for a file page */
		swp_entry_t entry;
		entry = make_migration_entry(page, pte_write(pteval));
		set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
	} else
K
KAMEZAWA Hiroyuki 已提交
1105
		dec_mm_counter(mm, MM_FILEPAGES);
L
Linus Torvalds 已提交
1106

1107
	page_remove_rmap(page);
L
Linus Torvalds 已提交
1108 1109 1110
	page_cache_release(page);

out_unmap:
H
Hugh Dickins 已提交
1111
	pte_unmap_unlock(pte, ptl);
K
KOSAKI Motohiro 已提交
1112 1113
out:
	return ret;
H
Hugh Dickins 已提交
1114

K
KOSAKI Motohiro 已提交
1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130
out_mlock:
	pte_unmap_unlock(pte, ptl);


	/*
	 * We need mmap_sem locking, Otherwise VM_LOCKED check makes
	 * unstable result and race. Plus, We can't wait here because
	 * we now hold anon_vma->lock or mapping->i_mmap_lock.
	 * if trylock failed, the page remain in evictable lru and later
	 * vmscan could retry to move the page to unevictable lru if the
	 * page is actually mlocked.
	 */
	if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
		if (vma->vm_flags & VM_LOCKED) {
			mlock_vma_page(page);
			ret = SWAP_MLOCK;
H
Hugh Dickins 已提交
1131
		}
K
KOSAKI Motohiro 已提交
1132
		up_read(&vma->vm_mm->mmap_sem);
H
Hugh Dickins 已提交
1133
	}
L
Linus Torvalds 已提交
1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154
	return ret;
}

/*
 * objrmap doesn't work for nonlinear VMAs because the assumption that
 * offset-into-file correlates with offset-into-virtual-addresses does not hold.
 * Consequently, given a particular page and its ->index, we cannot locate the
 * ptes which are mapping that page without an exhaustive linear search.
 *
 * So what this code does is a mini "virtual scan" of each nonlinear VMA which
 * maps the file to which the target page belongs.  The ->vm_private_data field
 * holds the current cursor into that scan.  Successive searches will circulate
 * around the vma's virtual address space.
 *
 * So as more replacement pressure is applied to the pages in a nonlinear VMA,
 * more scanning pressure is placed against them as well.   Eventually pages
 * will become fully unmapped and are eligible for eviction.
 *
 * For very sparsely populated VMAs this is a little inefficient - chances are
 * there there won't be many ptes located within the scan cluster.  In this case
 * maybe we could scan further - to the end of the pte page, perhaps.
N
Nick Piggin 已提交
1155 1156 1157 1158 1159
 *
 * Mlocked pages:  check VM_LOCKED under mmap_sem held for read, if we can
 * acquire it without blocking.  If vma locked, mlock the pages in the cluster,
 * rather than unmapping them.  If we encounter the "check_page" that vmscan is
 * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN.
L
Linus Torvalds 已提交
1160 1161 1162 1163
 */
#define CLUSTER_SIZE	min(32*PAGE_SIZE, PMD_SIZE)
#define CLUSTER_MASK	(~(CLUSTER_SIZE - 1))

N
Nick Piggin 已提交
1164 1165
static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount,
		struct vm_area_struct *vma, struct page *check_page)
L
Linus Torvalds 已提交
1166 1167 1168 1169 1170
{
	struct mm_struct *mm = vma->vm_mm;
	pgd_t *pgd;
	pud_t *pud;
	pmd_t *pmd;
H
Hugh Dickins 已提交
1171
	pte_t *pte;
L
Linus Torvalds 已提交
1172
	pte_t pteval;
H
Hugh Dickins 已提交
1173
	spinlock_t *ptl;
L
Linus Torvalds 已提交
1174 1175 1176
	struct page *page;
	unsigned long address;
	unsigned long end;
N
Nick Piggin 已提交
1177 1178
	int ret = SWAP_AGAIN;
	int locked_vma = 0;
L
Linus Torvalds 已提交
1179 1180 1181 1182 1183 1184 1185 1186 1187 1188

	address = (vma->vm_start + cursor) & CLUSTER_MASK;
	end = address + CLUSTER_SIZE;
	if (address < vma->vm_start)
		address = vma->vm_start;
	if (end > vma->vm_end)
		end = vma->vm_end;

	pgd = pgd_offset(mm, address);
	if (!pgd_present(*pgd))
N
Nick Piggin 已提交
1189
		return ret;
L
Linus Torvalds 已提交
1190 1191 1192

	pud = pud_offset(pgd, address);
	if (!pud_present(*pud))
N
Nick Piggin 已提交
1193
		return ret;
L
Linus Torvalds 已提交
1194 1195 1196

	pmd = pmd_offset(pud, address);
	if (!pmd_present(*pmd))
N
Nick Piggin 已提交
1197 1198 1199
		return ret;

	/*
H
Hugh Dickins 已提交
1200
	 * If we can acquire the mmap_sem for read, and vma is VM_LOCKED,
N
Nick Piggin 已提交
1201 1202
	 * keep the sem while scanning the cluster for mlocking pages.
	 */
H
Hugh Dickins 已提交
1203
	if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
N
Nick Piggin 已提交
1204 1205 1206 1207
		locked_vma = (vma->vm_flags & VM_LOCKED);
		if (!locked_vma)
			up_read(&vma->vm_mm->mmap_sem); /* don't need it */
	}
H
Hugh Dickins 已提交
1208 1209

	pte = pte_offset_map_lock(mm, pmd, address, &ptl);
L
Linus Torvalds 已提交
1210

1211 1212 1213
	/* Update high watermark before we lower rss */
	update_hiwater_rss(mm);

H
Hugh Dickins 已提交
1214
	for (; address < end; pte++, address += PAGE_SIZE) {
L
Linus Torvalds 已提交
1215 1216
		if (!pte_present(*pte))
			continue;
1217 1218
		page = vm_normal_page(vma, address, *pte);
		BUG_ON(!page || PageAnon(page));
L
Linus Torvalds 已提交
1219

N
Nick Piggin 已提交
1220 1221 1222 1223 1224 1225 1226
		if (locked_vma) {
			mlock_vma_page(page);   /* no-op if already mlocked */
			if (page == check_page)
				ret = SWAP_MLOCK;
			continue;	/* don't unmap */
		}

A
Andrea Arcangeli 已提交
1227
		if (ptep_clear_flush_young_notify(vma, address, pte))
L
Linus Torvalds 已提交
1228 1229 1230
			continue;

		/* Nuke the page table entry. */
1231
		flush_cache_page(vma, address, pte_pfn(*pte));
A
Andrea Arcangeli 已提交
1232
		pteval = ptep_clear_flush_notify(vma, address, pte);
L
Linus Torvalds 已提交
1233 1234 1235 1236 1237 1238 1239 1240 1241

		/* If nonlinear, store the file page offset in the pte. */
		if (page->index != linear_page_index(vma, address))
			set_pte_at(mm, address, pte, pgoff_to_pte(page->index));

		/* Move the dirty bit to the physical page now the pte is gone. */
		if (pte_dirty(pteval))
			set_page_dirty(page);

1242
		page_remove_rmap(page);
L
Linus Torvalds 已提交
1243
		page_cache_release(page);
K
KAMEZAWA Hiroyuki 已提交
1244
		dec_mm_counter(mm, MM_FILEPAGES);
L
Linus Torvalds 已提交
1245 1246
		(*mapcount)--;
	}
H
Hugh Dickins 已提交
1247
	pte_unmap_unlock(pte - 1, ptl);
N
Nick Piggin 已提交
1248 1249 1250
	if (locked_vma)
		up_read(&vma->vm_mm->mmap_sem);
	return ret;
L
Linus Torvalds 已提交
1251 1252
}

1253
bool is_vma_temporary_stack(struct vm_area_struct *vma)
1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266
{
	int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP);

	if (!maybe_stack)
		return false;

	if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) ==
						VM_STACK_INCOMPLETE_SETUP)
		return true;

	return false;
}

N
Nick Piggin 已提交
1267 1268 1269 1270
/**
 * try_to_unmap_anon - unmap or unlock anonymous page using the object-based
 * rmap method
 * @page: the page to unmap/unlock
1271
 * @flags: action and flags
N
Nick Piggin 已提交
1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282
 *
 * Find all the mappings of a page using the mapping pointer and the vma chains
 * contained in the anon_vma struct it points to.
 *
 * This function is only called from try_to_unmap/try_to_munlock for
 * anonymous pages.
 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
 * where the page was found will be held for write.  So, we won't recheck
 * vm_flags for that VMA.  That should be OK, because that vma shouldn't be
 * 'LOCKED.
 */
1283
static int try_to_unmap_anon(struct page *page, enum ttu_flags flags)
L
Linus Torvalds 已提交
1284 1285
{
	struct anon_vma *anon_vma;
1286
	struct anon_vma_chain *avc;
L
Linus Torvalds 已提交
1287
	int ret = SWAP_AGAIN;
N
Nick Piggin 已提交
1288

L
Linus Torvalds 已提交
1289 1290 1291 1292
	anon_vma = page_lock_anon_vma(page);
	if (!anon_vma)
		return ret;

1293 1294
	list_for_each_entry(avc, &anon_vma->head, same_anon_vma) {
		struct vm_area_struct *vma = avc->vma;
1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309
		unsigned long address;

		/*
		 * During exec, a temporary VMA is setup and later moved.
		 * The VMA is moved under the anon_vma lock but not the
		 * page tables leading to a race where migration cannot
		 * find the migration ptes. Rather than increasing the
		 * locking requirements of exec(), migration skips
		 * temporary VMAs until after exec() completes.
		 */
		if (PAGE_MIGRATION && (flags & TTU_MIGRATION) &&
				is_vma_temporary_stack(vma))
			continue;

		address = vma_address(page, vma);
H
Hugh Dickins 已提交
1310 1311 1312
		if (address == -EFAULT)
			continue;
		ret = try_to_unmap_one(page, vma, address, flags);
H
Hugh Dickins 已提交
1313 1314
		if (ret != SWAP_AGAIN || !page_mapped(page))
			break;
L
Linus Torvalds 已提交
1315
	}
1316 1317

	page_unlock_anon_vma(anon_vma);
L
Linus Torvalds 已提交
1318 1319 1320 1321
	return ret;
}

/**
N
Nick Piggin 已提交
1322 1323
 * try_to_unmap_file - unmap/unlock file page using the object-based rmap method
 * @page: the page to unmap/unlock
1324
 * @flags: action and flags
L
Linus Torvalds 已提交
1325 1326 1327 1328
 *
 * Find all the mappings of a page using the mapping pointer and the vma chains
 * contained in the address_space struct it points to.
 *
N
Nick Piggin 已提交
1329 1330 1331 1332 1333 1334
 * This function is only called from try_to_unmap/try_to_munlock for
 * object-based pages.
 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
 * where the page was found will be held for write.  So, we won't recheck
 * vm_flags for that VMA.  That should be OK, because that vma shouldn't be
 * 'LOCKED.
L
Linus Torvalds 已提交
1335
 */
1336
static int try_to_unmap_file(struct page *page, enum ttu_flags flags)
L
Linus Torvalds 已提交
1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349
{
	struct address_space *mapping = page->mapping;
	pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
	struct vm_area_struct *vma;
	struct prio_tree_iter iter;
	int ret = SWAP_AGAIN;
	unsigned long cursor;
	unsigned long max_nl_cursor = 0;
	unsigned long max_nl_size = 0;
	unsigned int mapcount;

	spin_lock(&mapping->i_mmap_lock);
	vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
H
Hugh Dickins 已提交
1350 1351 1352 1353
		unsigned long address = vma_address(page, vma);
		if (address == -EFAULT)
			continue;
		ret = try_to_unmap_one(page, vma, address, flags);
H
Hugh Dickins 已提交
1354 1355
		if (ret != SWAP_AGAIN || !page_mapped(page))
			goto out;
L
Linus Torvalds 已提交
1356 1357 1358 1359 1360
	}

	if (list_empty(&mapping->i_mmap_nonlinear))
		goto out;

H
Hugh Dickins 已提交
1361 1362 1363 1364 1365 1366 1367 1368
	/*
	 * We don't bother to try to find the munlocked page in nonlinears.
	 * It's costly. Instead, later, page reclaim logic may call
	 * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily.
	 */
	if (TTU_ACTION(flags) == TTU_MUNLOCK)
		goto out;

L
Linus Torvalds 已提交
1369 1370 1371 1372 1373 1374 1375 1376 1377 1378
	list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
						shared.vm_set.list) {
		cursor = (unsigned long) vma->vm_private_data;
		if (cursor > max_nl_cursor)
			max_nl_cursor = cursor;
		cursor = vma->vm_end - vma->vm_start;
		if (cursor > max_nl_size)
			max_nl_size = cursor;
	}

N
Nick Piggin 已提交
1379
	if (max_nl_size == 0) {	/* all nonlinears locked or reserved ? */
L
Linus Torvalds 已提交
1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403
		ret = SWAP_FAIL;
		goto out;
	}

	/*
	 * We don't try to search for this page in the nonlinear vmas,
	 * and page_referenced wouldn't have found it anyway.  Instead
	 * just walk the nonlinear vmas trying to age and unmap some.
	 * The mapcount of the page we came in with is irrelevant,
	 * but even so use it as a guide to how hard we should try?
	 */
	mapcount = page_mapcount(page);
	if (!mapcount)
		goto out;
	cond_resched_lock(&mapping->i_mmap_lock);

	max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK;
	if (max_nl_cursor == 0)
		max_nl_cursor = CLUSTER_SIZE;

	do {
		list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
						shared.vm_set.list) {
			cursor = (unsigned long) vma->vm_private_data;
H
Hugh Dickins 已提交
1404
			while ( cursor < max_nl_cursor &&
L
Linus Torvalds 已提交
1405
				cursor < vma->vm_end - vma->vm_start) {
H
Hugh Dickins 已提交
1406 1407 1408
				if (try_to_unmap_cluster(cursor, &mapcount,
						vma, page) == SWAP_MLOCK)
					ret = SWAP_MLOCK;
L
Linus Torvalds 已提交
1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424
				cursor += CLUSTER_SIZE;
				vma->vm_private_data = (void *) cursor;
				if ((int)mapcount <= 0)
					goto out;
			}
			vma->vm_private_data = (void *) max_nl_cursor;
		}
		cond_resched_lock(&mapping->i_mmap_lock);
		max_nl_cursor += CLUSTER_SIZE;
	} while (max_nl_cursor <= max_nl_size);

	/*
	 * Don't loop forever (perhaps all the remaining pages are
	 * in locked vmas).  Reset cursor on all unreserved nonlinear
	 * vmas, now forgetting on which ones it had fallen behind.
	 */
1425 1426
	list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
		vma->vm_private_data = NULL;
L
Linus Torvalds 已提交
1427 1428 1429 1430 1431 1432 1433 1434
out:
	spin_unlock(&mapping->i_mmap_lock);
	return ret;
}

/**
 * try_to_unmap - try to remove all page table mappings to a page
 * @page: the page to get unmapped
1435
 * @flags: action and flags
L
Linus Torvalds 已提交
1436 1437 1438 1439 1440 1441 1442 1443
 *
 * Tries to remove all the page table entries which are mapping this
 * page, used in the pageout path.  Caller must hold the page lock.
 * Return values are:
 *
 * SWAP_SUCCESS	- we succeeded in removing all mappings
 * SWAP_AGAIN	- we missed a mapping, try again later
 * SWAP_FAIL	- the page is unswappable
N
Nick Piggin 已提交
1444
 * SWAP_MLOCK	- page is mlocked.
L
Linus Torvalds 已提交
1445
 */
1446
int try_to_unmap(struct page *page, enum ttu_flags flags)
L
Linus Torvalds 已提交
1447 1448 1449 1450
{
	int ret;

	BUG_ON(!PageLocked(page));
1451
	VM_BUG_ON(!PageHuge(page) && PageTransHuge(page));
L
Linus Torvalds 已提交
1452

H
Hugh Dickins 已提交
1453 1454 1455
	if (unlikely(PageKsm(page)))
		ret = try_to_unmap_ksm(page, flags);
	else if (PageAnon(page))
1456
		ret = try_to_unmap_anon(page, flags);
L
Linus Torvalds 已提交
1457
	else
1458
		ret = try_to_unmap_file(page, flags);
N
Nick Piggin 已提交
1459
	if (ret != SWAP_MLOCK && !page_mapped(page))
L
Linus Torvalds 已提交
1460 1461 1462
		ret = SWAP_SUCCESS;
	return ret;
}
N
Nikita Danilov 已提交
1463

N
Nick Piggin 已提交
1464 1465 1466 1467 1468 1469 1470 1471 1472 1473
/**
 * try_to_munlock - try to munlock a page
 * @page: the page to be munlocked
 *
 * Called from munlock code.  Checks all of the VMAs mapping the page
 * to make sure nobody else has this page mlocked. The page will be
 * returned with PG_mlocked cleared if no other vmas have it mlocked.
 *
 * Return values are:
 *
H
Hugh Dickins 已提交
1474
 * SWAP_AGAIN	- no vma is holding page mlocked, or,
N
Nick Piggin 已提交
1475
 * SWAP_AGAIN	- page mapped in mlocked vma -- couldn't acquire mmap sem
H
Hugh Dickins 已提交
1476
 * SWAP_FAIL	- page cannot be located at present
N
Nick Piggin 已提交
1477 1478 1479 1480 1481 1482
 * SWAP_MLOCK	- page is now mlocked.
 */
int try_to_munlock(struct page *page)
{
	VM_BUG_ON(!PageLocked(page) || PageLRU(page));

H
Hugh Dickins 已提交
1483 1484 1485
	if (unlikely(PageKsm(page)))
		return try_to_unmap_ksm(page, TTU_MUNLOCK);
	else if (PageAnon(page))
1486
		return try_to_unmap_anon(page, TTU_MUNLOCK);
N
Nick Piggin 已提交
1487
	else
1488
		return try_to_unmap_file(page, TTU_MUNLOCK);
N
Nick Piggin 已提交
1489
}
1490

P
Peter Zijlstra 已提交
1491
void __put_anon_vma(struct anon_vma *anon_vma)
1492
{
P
Peter Zijlstra 已提交
1493
	struct anon_vma *root = anon_vma->root;
1494

P
Peter Zijlstra 已提交
1495 1496
	if (root != anon_vma && atomic_dec_and_test(&root->refcount))
		anon_vma_free(root);
1497

P
Peter Zijlstra 已提交
1498
	anon_vma_free(anon_vma);
1499 1500
}

1501 1502 1503 1504 1505 1506 1507 1508 1509
#ifdef CONFIG_MIGRATION
/*
 * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file():
 * Called by migrate.c to remove migration ptes, but might be used more later.
 */
static int rmap_walk_anon(struct page *page, int (*rmap_one)(struct page *,
		struct vm_area_struct *, unsigned long, void *), void *arg)
{
	struct anon_vma *anon_vma;
1510
	struct anon_vma_chain *avc;
1511 1512 1513 1514 1515
	int ret = SWAP_AGAIN;

	/*
	 * Note: remove_migration_ptes() cannot use page_lock_anon_vma()
	 * because that depends on page_mapped(); but not all its usages
1516 1517
	 * are holding mmap_sem. Users without mmap_sem are required to
	 * take a reference count to prevent the anon_vma disappearing
1518 1519 1520 1521
	 */
	anon_vma = page_anon_vma(page);
	if (!anon_vma)
		return ret;
1522
	anon_vma_lock(anon_vma);
1523 1524
	list_for_each_entry(avc, &anon_vma->head, same_anon_vma) {
		struct vm_area_struct *vma = avc->vma;
1525 1526 1527 1528 1529 1530 1531
		unsigned long address = vma_address(page, vma);
		if (address == -EFAULT)
			continue;
		ret = rmap_one(page, vma, address, arg);
		if (ret != SWAP_AGAIN)
			break;
	}
1532
	anon_vma_unlock(anon_vma);
1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577
	return ret;
}

static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *,
		struct vm_area_struct *, unsigned long, void *), void *arg)
{
	struct address_space *mapping = page->mapping;
	pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
	struct vm_area_struct *vma;
	struct prio_tree_iter iter;
	int ret = SWAP_AGAIN;

	if (!mapping)
		return ret;
	spin_lock(&mapping->i_mmap_lock);
	vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
		unsigned long address = vma_address(page, vma);
		if (address == -EFAULT)
			continue;
		ret = rmap_one(page, vma, address, arg);
		if (ret != SWAP_AGAIN)
			break;
	}
	/*
	 * No nonlinear handling: being always shared, nonlinear vmas
	 * never contain migration ptes.  Decide what to do about this
	 * limitation to linear when we need rmap_walk() on nonlinear.
	 */
	spin_unlock(&mapping->i_mmap_lock);
	return ret;
}

int rmap_walk(struct page *page, int (*rmap_one)(struct page *,
		struct vm_area_struct *, unsigned long, void *), void *arg)
{
	VM_BUG_ON(!PageLocked(page));

	if (unlikely(PageKsm(page)))
		return rmap_walk_ksm(page, rmap_one, arg);
	else if (PageAnon(page))
		return rmap_walk_anon(page, rmap_one, arg);
	else
		return rmap_walk_file(page, rmap_one, arg);
}
#endif /* CONFIG_MIGRATION */
1578

N
Naoya Horiguchi 已提交
1579
#ifdef CONFIG_HUGETLB_PAGE
1580 1581 1582 1583 1584 1585 1586 1587 1588
/*
 * The following three functions are for anonymous (private mapped) hugepages.
 * Unlike common anonymous pages, anonymous hugepages have no accounting code
 * and no lru code, because we handle hugepages differently from common pages.
 */
static void __hugepage_set_anon_rmap(struct page *page,
	struct vm_area_struct *vma, unsigned long address, int exclusive)
{
	struct anon_vma *anon_vma = vma->anon_vma;
1589

1590
	BUG_ON(!anon_vma);
1591 1592 1593 1594 1595 1596

	if (PageAnon(page))
		return;
	if (!exclusive)
		anon_vma = anon_vma->root;

1597 1598 1599 1600 1601 1602 1603 1604 1605 1606
	anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
	page->mapping = (struct address_space *) anon_vma;
	page->index = linear_page_index(vma, address);
}

void hugepage_add_anon_rmap(struct page *page,
			    struct vm_area_struct *vma, unsigned long address)
{
	struct anon_vma *anon_vma = vma->anon_vma;
	int first;
1607 1608

	BUG_ON(!PageLocked(page));
1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622
	BUG_ON(!anon_vma);
	BUG_ON(address < vma->vm_start || address >= vma->vm_end);
	first = atomic_inc_and_test(&page->_mapcount);
	if (first)
		__hugepage_set_anon_rmap(page, vma, address, 0);
}

void hugepage_add_new_anon_rmap(struct page *page,
			struct vm_area_struct *vma, unsigned long address)
{
	BUG_ON(address < vma->vm_start || address >= vma->vm_end);
	atomic_set(&page->_mapcount, 0);
	__hugepage_set_anon_rmap(page, vma, address, 1);
}
N
Naoya Horiguchi 已提交
1623
#endif /* CONFIG_HUGETLB_PAGE */