rmap.c 45.4 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 34 35 36 37 38
 *   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)
 *               inode_lock (in set_page_dirty's __mark_inode_dirty)
 *                 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,
 *                           within inode_lock in __sync_single_inode)
39 40 41 42 43
 *
 * (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
44 45 46 47 48 49 50 51
 */

#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
52
#include <linux/ksm.h>
L
Linus Torvalds 已提交
53 54
#include <linux/rmap.h>
#include <linux/rcupdate.h>
55
#include <linux/module.h>
56
#include <linux/memcontrol.h>
A
Andrea Arcangeli 已提交
57
#include <linux/mmu_notifier.h>
58
#include <linux/migrate.h>
59
#include <linux/hugetlb.h>
L
Linus Torvalds 已提交
60 61 62

#include <asm/tlbflush.h>

N
Nick Piggin 已提交
63 64
#include "internal.h"

65
static struct kmem_cache *anon_vma_cachep;
66
static struct kmem_cache *anon_vma_chain_cachep;
67 68 69 70 71 72

static inline struct anon_vma *anon_vma_alloc(void)
{
	return kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL);
}

H
Hugh Dickins 已提交
73
void anon_vma_free(struct anon_vma *anon_vma)
74 75 76
{
	kmem_cache_free(anon_vma_cachep, anon_vma);
}
L
Linus Torvalds 已提交
77

78 79 80 81 82 83 84 85 86 87
static inline struct anon_vma_chain *anon_vma_chain_alloc(void)
{
	return kmem_cache_alloc(anon_vma_chain_cachep, GFP_KERNEL);
}

void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain)
{
	kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain);
}

88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114
/**
 * 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
 * if not we either need to find an adjacent mapping that we
 * 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
115 116 117
int anon_vma_prepare(struct vm_area_struct *vma)
{
	struct anon_vma *anon_vma = vma->anon_vma;
118
	struct anon_vma_chain *avc;
L
Linus Torvalds 已提交
119 120 121 122

	might_sleep();
	if (unlikely(!anon_vma)) {
		struct mm_struct *mm = vma->vm_mm;
123
		struct anon_vma *allocated;
L
Linus Torvalds 已提交
124

125 126 127 128
		avc = anon_vma_chain_alloc();
		if (!avc)
			goto out_enomem;

L
Linus Torvalds 已提交
129
		anon_vma = find_mergeable_anon_vma(vma);
130 131
		allocated = NULL;
		if (!anon_vma) {
L
Linus Torvalds 已提交
132 133
			anon_vma = anon_vma_alloc();
			if (unlikely(!anon_vma))
134
				goto out_enomem_free_avc;
L
Linus Torvalds 已提交
135
			allocated = anon_vma;
136 137 138 139 140
			/*
			 * This VMA had no anon_vma yet.  This anon_vma is
			 * the root of any anon_vma tree that might form.
			 */
			anon_vma->root = anon_vma;
L
Linus Torvalds 已提交
141 142
		}

143
		anon_vma_lock(anon_vma);
L
Linus Torvalds 已提交
144 145 146 147
		/* page_table_lock to protect against threads */
		spin_lock(&mm->page_table_lock);
		if (likely(!vma->anon_vma)) {
			vma->anon_vma = anon_vma;
148 149 150
			avc->anon_vma = anon_vma;
			avc->vma = vma;
			list_add(&avc->same_vma, &vma->anon_vma_chain);
151
			list_add_tail(&avc->same_anon_vma, &anon_vma->head);
L
Linus Torvalds 已提交
152
			allocated = NULL;
153
			avc = NULL;
L
Linus Torvalds 已提交
154 155
		}
		spin_unlock(&mm->page_table_lock);
156
		anon_vma_unlock(anon_vma);
157 158

		if (unlikely(allocated))
L
Linus Torvalds 已提交
159
			anon_vma_free(allocated);
160
		if (unlikely(avc))
161
			anon_vma_chain_free(avc);
L
Linus Torvalds 已提交
162 163
	}
	return 0;
164 165 166 167 168

 out_enomem_free_avc:
	anon_vma_chain_free(avc);
 out_enomem:
	return -ENOMEM;
L
Linus Torvalds 已提交
169 170
}

171 172 173
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
174
{
175 176 177 178
	avc->vma = vma;
	avc->anon_vma = anon_vma;
	list_add(&avc->same_vma, &vma->anon_vma_chain);

179
	anon_vma_lock(anon_vma);
180
	list_add_tail(&avc->same_anon_vma, &anon_vma->head);
181
	anon_vma_unlock(anon_vma);
L
Linus Torvalds 已提交
182 183
}

184 185 186 187 188
/*
 * 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
189
{
190 191
	struct anon_vma_chain *avc, *pavc;

192
	list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) {
193 194 195 196 197 198
		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
199

200 201 202
 enomem_failure:
	unlink_anon_vmas(dst);
	return -ENOMEM;
L
Linus Torvalds 已提交
203 204
}

205 206 207 208 209 210
/*
 * 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
211
{
212 213
	struct anon_vma_chain *avc;
	struct anon_vma *anon_vma;
L
Linus Torvalds 已提交
214

215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232
	/* 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;
233 234 235 236 237 238

	/*
	 * 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;
239 240 241 242 243 244
	/*
	 * With KSM 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.
	 */
	get_anon_vma(anon_vma->root);
245 246
	/* Mark this anon_vma as the one where our new (COWed) pages go. */
	vma->anon_vma = anon_vma;
247
	anon_vma_chain_link(vma, avc, anon_vma);
248 249 250 251 252 253

	return 0;

 out_error_free_anon_vma:
	anon_vma_free(anon_vma);
 out_error:
254
	unlink_anon_vmas(vma);
255
	return -ENOMEM;
L
Linus Torvalds 已提交
256 257
}

258
static void anon_vma_unlink(struct anon_vma_chain *anon_vma_chain)
L
Linus Torvalds 已提交
259
{
260
	struct anon_vma *anon_vma = anon_vma_chain->anon_vma;
L
Linus Torvalds 已提交
261 262
	int empty;

263
	/* If anon_vma_fork fails, we can get an empty anon_vma_chain. */
L
Linus Torvalds 已提交
264 265 266
	if (!anon_vma)
		return;

267
	anon_vma_lock(anon_vma);
268
	list_del(&anon_vma_chain->same_anon_vma);
L
Linus Torvalds 已提交
269 270

	/* We must garbage collect the anon_vma if it's empty */
271
	empty = list_empty(&anon_vma->head) && !anonvma_external_refcount(anon_vma);
272
	anon_vma_unlock(anon_vma);
L
Linus Torvalds 已提交
273

274 275 276 277
	if (empty) {
		/* We no longer need the root anon_vma */
		if (anon_vma->root != anon_vma)
			drop_anon_vma(anon_vma->root);
L
Linus Torvalds 已提交
278
		anon_vma_free(anon_vma);
279
	}
L
Linus Torvalds 已提交
280 281
}

282 283 284 285
void unlink_anon_vmas(struct vm_area_struct *vma)
{
	struct anon_vma_chain *avc, *next;

286 287 288 289
	/*
	 * Unlink each anon_vma chained to the VMA.  This list is ordered
	 * from newest to oldest, ensuring the root anon_vma gets freed last.
	 */
290 291 292 293 294 295 296
	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);
	}
}

297
static void anon_vma_ctor(void *data)
L
Linus Torvalds 已提交
298
{
C
Christoph Lameter 已提交
299
	struct anon_vma *anon_vma = data;
L
Linus Torvalds 已提交
300

C
Christoph Lameter 已提交
301
	spin_lock_init(&anon_vma->lock);
302
	anonvma_external_refcount_init(anon_vma);
C
Christoph Lameter 已提交
303
	INIT_LIST_HEAD(&anon_vma->head);
L
Linus Torvalds 已提交
304 305 306 307 308
}

void __init anon_vma_init(void)
{
	anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
309
			0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor);
310
	anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC);
L
Linus Torvalds 已提交
311 312 313 314 315 316
}

/*
 * 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.
 */
317
struct anon_vma *page_lock_anon_vma(struct page *page)
L
Linus Torvalds 已提交
318
{
319
	struct anon_vma *anon_vma, *root_anon_vma;
L
Linus Torvalds 已提交
320 321 322
	unsigned long anon_mapping;

	rcu_read_lock();
323
	anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping);
H
Hugh Dickins 已提交
324
	if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
L
Linus Torvalds 已提交
325 326 327 328 329
		goto out;
	if (!page_mapped(page))
		goto out;

	anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
330 331 332 333 334 335 336 337 338 339 340 341 342 343 344
	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
345 346
out:
	rcu_read_unlock();
347 348 349
	return NULL;
}

350
void page_unlock_anon_vma(struct anon_vma *anon_vma)
351
{
352
	anon_vma_unlock(anon_vma);
353
	rcu_read_unlock();
L
Linus Torvalds 已提交
354 355 356
}

/*
357 358 359
 * 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
360 361 362 363 364 365 366
 */
static inline unsigned long
vma_address(struct page *page, struct vm_area_struct *vma)
{
	pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
	unsigned long address;

367 368
	if (unlikely(is_vm_hugetlb_page(vma)))
		pgoff = page->index << huge_page_order(page_hstate(page));
L
Linus Torvalds 已提交
369 370
	address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
	if (unlikely(address < vma->vm_start || address >= vma->vm_end)) {
371
		/* page should be within @vma mapping range */
L
Linus Torvalds 已提交
372 373 374 375 376 377
		return -EFAULT;
	}
	return address;
}

/*
H
Huang Shijie 已提交
378
 * At what user virtual address is page expected in vma?
379
 * Caller should check the page is actually part of the vma.
L
Linus Torvalds 已提交
380 381 382
 */
unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
{
383 384 385 386
	if (PageAnon(page)) {
		if (vma->anon_vma->root != page_anon_vma(page)->root)
			return -EFAULT;
	} else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) {
H
Hugh Dickins 已提交
387 388
		if (!vma->vm_file ||
		    vma->vm_file->f_mapping != page->mapping)
L
Linus Torvalds 已提交
389 390 391 392 393 394
			return -EFAULT;
	} else
		return -EFAULT;
	return vma_address(page, vma);
}

N
Nikita Danilov 已提交
395 396 397
/*
 * Check that @page is mapped at @address into @mm.
 *
N
Nick Piggin 已提交
398 399 400 401
 * 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).
 *
402
 * On success returns with pte mapped and locked.
N
Nikita Danilov 已提交
403
 */
404
pte_t *page_check_address(struct page *page, struct mm_struct *mm,
N
Nick Piggin 已提交
405
			  unsigned long address, spinlock_t **ptlp, int sync)
N
Nikita Danilov 已提交
406 407 408 409 410
{
	pgd_t *pgd;
	pud_t *pud;
	pmd_t *pmd;
	pte_t *pte;
H
Hugh Dickins 已提交
411
	spinlock_t *ptl;
N
Nikita Danilov 已提交
412

413 414 415 416 417 418
	if (unlikely(PageHuge(page))) {
		pte = huge_pte_offset(mm, address);
		ptl = &mm->page_table_lock;
		goto check;
	}

N
Nikita Danilov 已提交
419
	pgd = pgd_offset(mm, address);
H
Hugh Dickins 已提交
420 421 422 423 424 425 426 427 428 429 430 431 432
	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;

	pte = pte_offset_map(pmd, address);
	/* Make a quick check before getting the lock */
N
Nick Piggin 已提交
433
	if (!sync && !pte_present(*pte)) {
H
Hugh Dickins 已提交
434 435 436 437
		pte_unmap(pte);
		return NULL;
	}

H
Hugh Dickins 已提交
438
	ptl = pte_lockptr(mm, pmd);
439
check:
H
Hugh Dickins 已提交
440 441 442 443
	spin_lock(ptl);
	if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
		*ptlp = ptl;
		return pte;
N
Nikita Danilov 已提交
444
	}
H
Hugh Dickins 已提交
445 446
	pte_unmap_unlock(pte, ptl);
	return NULL;
N
Nikita Danilov 已提交
447 448
}

N
Nick Piggin 已提交
449 450 451 452 453 454 455 456 457
/**
 * 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.
 */
458
int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
N
Nick Piggin 已提交
459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474
{
	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
475 476 477 478
/*
 * Subfunctions of page_referenced: page_referenced_one called
 * repeatedly from either page_referenced_anon or page_referenced_file.
 */
H
Hugh Dickins 已提交
479 480 481
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
482 483 484
{
	struct mm_struct *mm = vma->vm_mm;
	pte_t *pte;
H
Hugh Dickins 已提交
485
	spinlock_t *ptl;
L
Linus Torvalds 已提交
486 487
	int referenced = 0;

N
Nick Piggin 已提交
488
	pte = page_check_address(page, mm, address, &ptl, 0);
H
Hugh Dickins 已提交
489 490
	if (!pte)
		goto out;
L
Linus Torvalds 已提交
491

N
Nick Piggin 已提交
492 493 494 495 496
	/*
	 * Don't want to elevate referenced for mlocked page that gets this far,
	 * in order that it progresses to try_to_unmap and is moved to the
	 * unevictable list.
	 */
497 498
	if (vma->vm_flags & VM_LOCKED) {
		*mapcount = 1;	/* break early from loop */
499
		*vm_flags |= VM_LOCKED;
N
Nick Piggin 已提交
500 501 502
		goto out_unmap;
	}

503 504 505 506 507 508 509 510 511 512 513
	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++;
	}
L
Linus Torvalds 已提交
514

H
Hugh Dickins 已提交
515 516
	/* Pretend the page is referenced if the task has the
	   swap token and is in the middle of a page fault. */
517
	if (mm != current->mm && has_swap_token(mm) &&
H
Hugh Dickins 已提交
518 519 520
			rwsem_is_locked(&mm->mmap_sem))
		referenced++;

N
Nick Piggin 已提交
521
out_unmap:
H
Hugh Dickins 已提交
522 523
	(*mapcount)--;
	pte_unmap_unlock(pte, ptl);
524

525 526
	if (referenced)
		*vm_flags |= vma->vm_flags;
527
out:
L
Linus Torvalds 已提交
528 529 530
	return referenced;
}

531
static int page_referenced_anon(struct page *page,
532 533
				struct mem_cgroup *mem_cont,
				unsigned long *vm_flags)
L
Linus Torvalds 已提交
534 535 536
{
	unsigned int mapcount;
	struct anon_vma *anon_vma;
537
	struct anon_vma_chain *avc;
L
Linus Torvalds 已提交
538 539 540 541 542 543 544
	int referenced = 0;

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

	mapcount = page_mapcount(page);
545 546
	list_for_each_entry(avc, &anon_vma->head, same_anon_vma) {
		struct vm_area_struct *vma = avc->vma;
H
Hugh Dickins 已提交
547 548 549
		unsigned long address = vma_address(page, vma);
		if (address == -EFAULT)
			continue;
550 551 552 553 554
		/*
		 * If we are reclaiming on behalf of a cgroup, skip
		 * counting on behalf of references from different
		 * cgroups
		 */
555
		if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont))
556
			continue;
H
Hugh Dickins 已提交
557
		referenced += page_referenced_one(page, vma, address,
558
						  &mapcount, vm_flags);
L
Linus Torvalds 已提交
559 560 561
		if (!mapcount)
			break;
	}
562 563

	page_unlock_anon_vma(anon_vma);
L
Linus Torvalds 已提交
564 565 566 567 568 569
	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
570
 * @mem_cont: target memory controller
571
 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
L
Linus Torvalds 已提交
572 573 574 575 576 577 578 579
 *
 * 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.
 */
580
static int page_referenced_file(struct page *page,
581 582
				struct mem_cgroup *mem_cont,
				unsigned long *vm_flags)
L
Linus Torvalds 已提交
583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614
{
	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
615 616 617
		unsigned long address = vma_address(page, vma);
		if (address == -EFAULT)
			continue;
618 619 620 621 622
		/*
		 * If we are reclaiming on behalf of a cgroup, skip
		 * counting on behalf of references from different
		 * cgroups
		 */
623
		if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont))
624
			continue;
H
Hugh Dickins 已提交
625
		referenced += page_referenced_one(page, vma, address,
626
						  &mapcount, vm_flags);
L
Linus Torvalds 已提交
627 628 629 630 631 632 633 634 635 636 637 638
		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
639
 * @mem_cont: target memory controller
640
 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
L
Linus Torvalds 已提交
641 642 643 644
 *
 * Quick test_and_clear_referenced for all mappings to a page,
 * returns the number of ptes which referenced the page.
 */
645 646 647 648
int page_referenced(struct page *page,
		    int is_locked,
		    struct mem_cgroup *mem_cont,
		    unsigned long *vm_flags)
L
Linus Torvalds 已提交
649 650
{
	int referenced = 0;
H
Hugh Dickins 已提交
651
	int we_locked = 0;
L
Linus Torvalds 已提交
652

653
	*vm_flags = 0;
H
Hugh Dickins 已提交
654
	if (page_mapped(page) && page_rmapping(page)) {
H
Hugh Dickins 已提交
655 656 657 658 659 660 661 662 663 664 665
		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))
666 667
			referenced += page_referenced_anon(page, mem_cont,
								vm_flags);
H
Hugh Dickins 已提交
668
		else if (page->mapping)
669 670
			referenced += page_referenced_file(page, mem_cont,
								vm_flags);
H
Hugh Dickins 已提交
671
		if (we_locked)
L
Linus Torvalds 已提交
672 673
			unlock_page(page);
	}
H
Hugh Dickins 已提交
674
out:
675 676 677
	if (page_test_and_clear_young(page))
		referenced++;

L
Linus Torvalds 已提交
678 679 680
	return referenced;
}

H
Hugh Dickins 已提交
681 682
static int page_mkclean_one(struct page *page, struct vm_area_struct *vma,
			    unsigned long address)
683 684
{
	struct mm_struct *mm = vma->vm_mm;
685
	pte_t *pte;
686 687 688
	spinlock_t *ptl;
	int ret = 0;

N
Nick Piggin 已提交
689
	pte = page_check_address(page, mm, address, &ptl, 1);
690 691 692
	if (!pte)
		goto out;

693 694
	if (pte_dirty(*pte) || pte_write(*pte)) {
		pte_t entry;
695

696
		flush_cache_page(vma, address, pte_pfn(*pte));
A
Andrea Arcangeli 已提交
697
		entry = ptep_clear_flush_notify(vma, address, pte);
698 699
		entry = pte_wrprotect(entry);
		entry = pte_mkclean(entry);
700
		set_pte_at(mm, address, pte, entry);
701 702
		ret = 1;
	}
703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719

	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
720 721 722 723 724 725
		if (vma->vm_flags & VM_SHARED) {
			unsigned long address = vma_address(page, vma);
			if (address == -EFAULT)
				continue;
			ret += page_mkclean_one(page, vma, address);
		}
726 727 728 729 730 731 732 733 734 735 736 737 738
	}
	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);
739
		if (mapping) {
740
			ret = page_mkclean_file(mapping, page);
741 742 743 744
			if (page_test_dirty(page)) {
				page_clear_dirty(page);
				ret = 1;
			}
745
		}
746 747 748 749
	}

	return ret;
}
J
Jaya Kumar 已提交
750
EXPORT_SYMBOL_GPL(page_mkclean);
751

752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775
/**
 * 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
776
/**
R
Randy Dunlap 已提交
777
 * __page_set_anon_rmap - setup new anonymous rmap
N
Nick Piggin 已提交
778 779 780
 * @page:	the page to add the mapping to
 * @vma:	the vm area in which the mapping is added
 * @address:	the user virtual address mapped
781
 * @exclusive:	the page is exclusively owned by the current process
N
Nick Piggin 已提交
782 783
 */
static void __page_set_anon_rmap(struct page *page,
784
	struct vm_area_struct *vma, unsigned long address, int exclusive)
N
Nick Piggin 已提交
785
{
786
	struct anon_vma *anon_vma = vma->anon_vma;
787

788
	BUG_ON(!anon_vma);
789 790

	/*
791 792 793
	 * If the page isn't exclusively mapped into this vma,
	 * we must use the _oldest_ possible anon_vma for the
	 * page mapping!
794
	 */
795
	if (!exclusive) {
796 797 798 799 800 801 802 803 804 805 806 807
		if (PageAnon(page))
			return;
		anon_vma = anon_vma->root;
	} else {
		/*
		 * In this case, swapped-out-but-not-discarded swap-cache
		 * is remapped. So, no need to update page->mapping here.
		 * We convice anon_vma poitned by page->mapping is not obsolete
		 * because vma->anon_vma is necessary to be a family of it.
		 */
		if (PageAnon(page))
			return;
808
	}
N
Nick Piggin 已提交
809 810 811 812 813 814

	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
815
/**
R
Randy Dunlap 已提交
816
 * __page_check_anon_rmap - sanity check anonymous rmap addition
N
Nick Piggin 已提交
817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836
 * @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
837
	BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root);
N
Nick Piggin 已提交
838 839 840 841
	BUG_ON(page->index != linear_page_index(vma, address));
#endif
}

L
Linus Torvalds 已提交
842 843 844 845 846 847
/**
 * 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
848
 * The caller needs to hold the pte lock, and the page must be locked in
849 850 851
 * 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
852 853 854
 */
void page_add_anon_rmap(struct page *page,
	struct vm_area_struct *vma, unsigned long address)
855 856 857 858 859 860 861 862 863 864 865
{
	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
866
{
H
Hugh Dickins 已提交
867 868 869 870 871 872
	int first = atomic_inc_and_test(&page->_mapcount);
	if (first)
		__inc_zone_page_state(page, NR_ANON_PAGES);
	if (unlikely(PageKsm(page)))
		return;

N
Nick Piggin 已提交
873 874
	VM_BUG_ON(!PageLocked(page));
	VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
H
Hugh Dickins 已提交
875
	if (first)
876
		__page_set_anon_rmap(page, vma, address, exclusive);
877
	else
N
Nick Piggin 已提交
878
		__page_check_anon_rmap(page, vma, address);
L
Linus Torvalds 已提交
879 880
}

R
Randy Dunlap 已提交
881
/**
N
Nick Piggin 已提交
882 883 884 885 886 887 888
 * 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
889
 * Page does not have to be locked.
N
Nick Piggin 已提交
890 891 892 893
 */
void page_add_new_anon_rmap(struct page *page,
	struct vm_area_struct *vma, unsigned long address)
{
894
	VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
895 896
	SetPageSwapBacked(page);
	atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */
H
Hugh Dickins 已提交
897
	__inc_zone_page_state(page, NR_ANON_PAGES);
898
	__page_set_anon_rmap(page, vma, address, 1);
899
	if (page_evictable(page, vma))
900
		lru_cache_add_lru(page, LRU_ACTIVE_ANON);
901 902
	else
		add_page_to_unevictable_list(page);
N
Nick Piggin 已提交
903 904
}

L
Linus Torvalds 已提交
905 906 907 908
/**
 * page_add_file_rmap - add pte mapping to a file page
 * @page: the page to add the mapping to
 *
909
 * The caller needs to hold the pte lock.
L
Linus Torvalds 已提交
910 911 912
 */
void page_add_file_rmap(struct page *page)
{
913
	if (atomic_inc_and_test(&page->_mapcount)) {
914
		__inc_zone_page_state(page, NR_FILE_MAPPED);
915
		mem_cgroup_update_file_mapped(page, 1);
916
	}
L
Linus Torvalds 已提交
917 918 919 920 921 922
}

/**
 * page_remove_rmap - take down pte mapping from a page
 * @page: page to remove mapping from
 *
923
 * The caller needs to hold the pte lock.
L
Linus Torvalds 已提交
924
 */
925
void page_remove_rmap(struct page *page)
L
Linus Torvalds 已提交
926
{
927 928 929 930 931 932 933 934 935 936 937 938 939 940
	/* 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.
	 */
	if ((!PageAnon(page) || PageSwapCache(page)) && page_test_dirty(page)) {
		page_clear_dirty(page);
		set_page_dirty(page);
L
Linus Torvalds 已提交
941
	}
942 943 944 945 946 947
	/*
	 * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED
	 * and not charged by memcg for now.
	 */
	if (unlikely(PageHuge(page)))
		return;
948 949 950 951 952
	if (PageAnon(page)) {
		mem_cgroup_uncharge_page(page);
		__dec_zone_page_state(page, NR_ANON_PAGES);
	} else {
		__dec_zone_page_state(page, NR_FILE_MAPPED);
953
		mem_cgroup_update_file_mapped(page, -1);
954 955 956 957 958 959 960 961 962 963
	}
	/*
	 * 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
964 965 966 967 968 969
}

/*
 * 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
970 971
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
972 973 974 975
{
	struct mm_struct *mm = vma->vm_mm;
	pte_t *pte;
	pte_t pteval;
H
Hugh Dickins 已提交
976
	spinlock_t *ptl;
L
Linus Torvalds 已提交
977 978
	int ret = SWAP_AGAIN;

N
Nick Piggin 已提交
979
	pte = page_check_address(page, mm, address, &ptl, 0);
H
Hugh Dickins 已提交
980
	if (!pte)
N
Nikita Danilov 已提交
981
		goto out;
L
Linus Torvalds 已提交
982 983 984 985 986 987

	/*
	 * 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.
	 */
988
	if (!(flags & TTU_IGNORE_MLOCK)) {
K
KOSAKI Motohiro 已提交
989 990 991
		if (vma->vm_flags & VM_LOCKED)
			goto out_mlock;

H
Hugh Dickins 已提交
992
		if (TTU_ACTION(flags) == TTU_MUNLOCK)
H
Hugh Dickins 已提交
993
			goto out_unmap;
994 995
	}
	if (!(flags & TTU_IGNORE_ACCESS)) {
N
Nick Piggin 已提交
996 997 998 999 1000
		if (ptep_clear_flush_young_notify(vma, address, pte)) {
			ret = SWAP_FAIL;
			goto out_unmap;
		}
  	}
L
Linus Torvalds 已提交
1001 1002 1003

	/* Nuke the page table entry. */
	flush_cache_page(vma, address, page_to_pfn(page));
A
Andrea Arcangeli 已提交
1004
	pteval = ptep_clear_flush_notify(vma, address, pte);
L
Linus Torvalds 已提交
1005 1006 1007 1008 1009

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

1010 1011 1012
	/* Update high watermark before we lower rss */
	update_hiwater_rss(mm);

1013 1014
	if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
		if (PageAnon(page))
K
KAMEZAWA Hiroyuki 已提交
1015
			dec_mm_counter(mm, MM_ANONPAGES);
1016
		else
K
KAMEZAWA Hiroyuki 已提交
1017
			dec_mm_counter(mm, MM_FILEPAGES);
1018 1019 1020
		set_pte_at(mm, address, pte,
				swp_entry_to_pte(make_hwpoison_entry(page)));
	} else if (PageAnon(page)) {
H
Hugh Dickins 已提交
1021
		swp_entry_t entry = { .val = page_private(page) };
1022 1023 1024 1025 1026 1027

		if (PageSwapCache(page)) {
			/*
			 * Store the swap location in the pte.
			 * See handle_pte_fault() ...
			 */
H
Hugh Dickins 已提交
1028 1029 1030 1031 1032
			if (swap_duplicate(entry) < 0) {
				set_pte_at(mm, address, pte, pteval);
				ret = SWAP_FAIL;
				goto out_unmap;
			}
1033 1034 1035 1036 1037 1038
			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
1039
			dec_mm_counter(mm, MM_ANONPAGES);
K
KAMEZAWA Hiroyuki 已提交
1040
			inc_mm_counter(mm, MM_SWAPENTS);
1041
		} else if (PAGE_MIGRATION) {
1042 1043 1044 1045 1046
			/*
			 * 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.
			 */
1047
			BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION);
1048
			entry = make_migration_entry(page, pte_write(pteval));
L
Linus Torvalds 已提交
1049 1050 1051
		}
		set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
		BUG_ON(pte_file(*pte));
1052
	} else if (PAGE_MIGRATION && (TTU_ACTION(flags) == TTU_MIGRATION)) {
1053 1054 1055 1056 1057
		/* 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
1058
		dec_mm_counter(mm, MM_FILEPAGES);
L
Linus Torvalds 已提交
1059

1060
	page_remove_rmap(page);
L
Linus Torvalds 已提交
1061 1062 1063
	page_cache_release(page);

out_unmap:
H
Hugh Dickins 已提交
1064
	pte_unmap_unlock(pte, ptl);
K
KOSAKI Motohiro 已提交
1065 1066
out:
	return ret;
H
Hugh Dickins 已提交
1067

K
KOSAKI Motohiro 已提交
1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083
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
1084
		}
K
KOSAKI Motohiro 已提交
1085
		up_read(&vma->vm_mm->mmap_sem);
H
Hugh Dickins 已提交
1086
	}
L
Linus Torvalds 已提交
1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107
	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
1108 1109 1110 1111 1112
 *
 * 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
1113 1114 1115 1116
 */
#define CLUSTER_SIZE	min(32*PAGE_SIZE, PMD_SIZE)
#define CLUSTER_MASK	(~(CLUSTER_SIZE - 1))

N
Nick Piggin 已提交
1117 1118
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
1119 1120 1121 1122 1123
{
	struct mm_struct *mm = vma->vm_mm;
	pgd_t *pgd;
	pud_t *pud;
	pmd_t *pmd;
H
Hugh Dickins 已提交
1124
	pte_t *pte;
L
Linus Torvalds 已提交
1125
	pte_t pteval;
H
Hugh Dickins 已提交
1126
	spinlock_t *ptl;
L
Linus Torvalds 已提交
1127 1128 1129
	struct page *page;
	unsigned long address;
	unsigned long end;
N
Nick Piggin 已提交
1130 1131
	int ret = SWAP_AGAIN;
	int locked_vma = 0;
L
Linus Torvalds 已提交
1132 1133 1134 1135 1136 1137 1138 1139 1140 1141

	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
1142
		return ret;
L
Linus Torvalds 已提交
1143 1144 1145

	pud = pud_offset(pgd, address);
	if (!pud_present(*pud))
N
Nick Piggin 已提交
1146
		return ret;
L
Linus Torvalds 已提交
1147 1148 1149

	pmd = pmd_offset(pud, address);
	if (!pmd_present(*pmd))
N
Nick Piggin 已提交
1150 1151 1152
		return ret;

	/*
H
Hugh Dickins 已提交
1153
	 * If we can acquire the mmap_sem for read, and vma is VM_LOCKED,
N
Nick Piggin 已提交
1154 1155
	 * keep the sem while scanning the cluster for mlocking pages.
	 */
H
Hugh Dickins 已提交
1156
	if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
N
Nick Piggin 已提交
1157 1158 1159 1160
		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
1161 1162

	pte = pte_offset_map_lock(mm, pmd, address, &ptl);
L
Linus Torvalds 已提交
1163

1164 1165 1166
	/* Update high watermark before we lower rss */
	update_hiwater_rss(mm);

H
Hugh Dickins 已提交
1167
	for (; address < end; pte++, address += PAGE_SIZE) {
L
Linus Torvalds 已提交
1168 1169
		if (!pte_present(*pte))
			continue;
1170 1171
		page = vm_normal_page(vma, address, *pte);
		BUG_ON(!page || PageAnon(page));
L
Linus Torvalds 已提交
1172

N
Nick Piggin 已提交
1173 1174 1175 1176 1177 1178 1179
		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
1180
		if (ptep_clear_flush_young_notify(vma, address, pte))
L
Linus Torvalds 已提交
1181 1182 1183
			continue;

		/* Nuke the page table entry. */
1184
		flush_cache_page(vma, address, pte_pfn(*pte));
A
Andrea Arcangeli 已提交
1185
		pteval = ptep_clear_flush_notify(vma, address, pte);
L
Linus Torvalds 已提交
1186 1187 1188 1189 1190 1191 1192 1193 1194

		/* 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);

1195
		page_remove_rmap(page);
L
Linus Torvalds 已提交
1196
		page_cache_release(page);
K
KAMEZAWA Hiroyuki 已提交
1197
		dec_mm_counter(mm, MM_FILEPAGES);
L
Linus Torvalds 已提交
1198 1199
		(*mapcount)--;
	}
H
Hugh Dickins 已提交
1200
	pte_unmap_unlock(pte - 1, ptl);
N
Nick Piggin 已提交
1201 1202 1203
	if (locked_vma)
		up_read(&vma->vm_mm->mmap_sem);
	return ret;
L
Linus Torvalds 已提交
1204 1205
}

1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219
static bool is_vma_temporary_stack(struct vm_area_struct *vma)
{
	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
1220 1221 1222 1223
/**
 * try_to_unmap_anon - unmap or unlock anonymous page using the object-based
 * rmap method
 * @page: the page to unmap/unlock
1224
 * @flags: action and flags
N
Nick Piggin 已提交
1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235
 *
 * 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.
 */
1236
static int try_to_unmap_anon(struct page *page, enum ttu_flags flags)
L
Linus Torvalds 已提交
1237 1238
{
	struct anon_vma *anon_vma;
1239
	struct anon_vma_chain *avc;
L
Linus Torvalds 已提交
1240
	int ret = SWAP_AGAIN;
N
Nick Piggin 已提交
1241

L
Linus Torvalds 已提交
1242 1243 1244 1245
	anon_vma = page_lock_anon_vma(page);
	if (!anon_vma)
		return ret;

1246 1247
	list_for_each_entry(avc, &anon_vma->head, same_anon_vma) {
		struct vm_area_struct *vma = avc->vma;
1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262
		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
1263 1264 1265
		if (address == -EFAULT)
			continue;
		ret = try_to_unmap_one(page, vma, address, flags);
H
Hugh Dickins 已提交
1266 1267
		if (ret != SWAP_AGAIN || !page_mapped(page))
			break;
L
Linus Torvalds 已提交
1268
	}
1269 1270

	page_unlock_anon_vma(anon_vma);
L
Linus Torvalds 已提交
1271 1272 1273 1274
	return ret;
}

/**
N
Nick Piggin 已提交
1275 1276
 * try_to_unmap_file - unmap/unlock file page using the object-based rmap method
 * @page: the page to unmap/unlock
1277
 * @flags: action and flags
L
Linus Torvalds 已提交
1278 1279 1280 1281
 *
 * 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
1282 1283 1284 1285 1286 1287
 * 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
1288
 */
1289
static int try_to_unmap_file(struct page *page, enum ttu_flags flags)
L
Linus Torvalds 已提交
1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302
{
	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
1303 1304 1305 1306
		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
1307 1308
		if (ret != SWAP_AGAIN || !page_mapped(page))
			goto out;
L
Linus Torvalds 已提交
1309 1310 1311 1312 1313
	}

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

H
Hugh Dickins 已提交
1314 1315 1316 1317 1318 1319 1320 1321
	/*
	 * 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
1322 1323 1324 1325 1326 1327 1328 1329 1330 1331
	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
1332
	if (max_nl_size == 0) {	/* all nonlinears locked or reserved ? */
L
Linus Torvalds 已提交
1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356
		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
1357
			while ( cursor < max_nl_cursor &&
L
Linus Torvalds 已提交
1358
				cursor < vma->vm_end - vma->vm_start) {
H
Hugh Dickins 已提交
1359 1360 1361
				if (try_to_unmap_cluster(cursor, &mapcount,
						vma, page) == SWAP_MLOCK)
					ret = SWAP_MLOCK;
L
Linus Torvalds 已提交
1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377
				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.
	 */
1378 1379
	list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
		vma->vm_private_data = NULL;
L
Linus Torvalds 已提交
1380 1381 1382 1383 1384 1385 1386 1387
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
1388
 * @flags: action and flags
L
Linus Torvalds 已提交
1389 1390 1391 1392 1393 1394 1395 1396
 *
 * 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
1397
 * SWAP_MLOCK	- page is mlocked.
L
Linus Torvalds 已提交
1398
 */
1399
int try_to_unmap(struct page *page, enum ttu_flags flags)
L
Linus Torvalds 已提交
1400 1401 1402 1403 1404
{
	int ret;

	BUG_ON(!PageLocked(page));

H
Hugh Dickins 已提交
1405 1406 1407
	if (unlikely(PageKsm(page)))
		ret = try_to_unmap_ksm(page, flags);
	else if (PageAnon(page))
1408
		ret = try_to_unmap_anon(page, flags);
L
Linus Torvalds 已提交
1409
	else
1410
		ret = try_to_unmap_file(page, flags);
N
Nick Piggin 已提交
1411
	if (ret != SWAP_MLOCK && !page_mapped(page))
L
Linus Torvalds 已提交
1412 1413 1414
		ret = SWAP_SUCCESS;
	return ret;
}
N
Nikita Danilov 已提交
1415

N
Nick Piggin 已提交
1416 1417 1418 1419 1420 1421 1422 1423 1424 1425
/**
 * 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
1426
 * SWAP_AGAIN	- no vma is holding page mlocked, or,
N
Nick Piggin 已提交
1427
 * SWAP_AGAIN	- page mapped in mlocked vma -- couldn't acquire mmap sem
H
Hugh Dickins 已提交
1428
 * SWAP_FAIL	- page cannot be located at present
N
Nick Piggin 已提交
1429 1430 1431 1432 1433 1434
 * 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
1435 1436 1437
	if (unlikely(PageKsm(page)))
		return try_to_unmap_ksm(page, TTU_MUNLOCK);
	else if (PageAnon(page))
1438
		return try_to_unmap_anon(page, TTU_MUNLOCK);
N
Nick Piggin 已提交
1439
	else
1440
		return try_to_unmap_file(page, TTU_MUNLOCK);
N
Nick Piggin 已提交
1441
}
1442

1443 1444 1445 1446 1447 1448 1449 1450 1451
#if defined(CONFIG_KSM) || defined(CONFIG_MIGRATION)
/*
 * Drop an anon_vma refcount, freeing the anon_vma and anon_vma->root
 * if necessary.  Be careful to do all the tests under the lock.  Once
 * we know we are the last user, nobody else can get a reference and we
 * can do the freeing without the lock.
 */
void drop_anon_vma(struct anon_vma *anon_vma)
{
A
Andrea Arcangeli 已提交
1452
	BUG_ON(atomic_read(&anon_vma->external_refcount) <= 0);
1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463
	if (atomic_dec_and_lock(&anon_vma->external_refcount, &anon_vma->root->lock)) {
		struct anon_vma *root = anon_vma->root;
		int empty = list_empty(&anon_vma->head);
		int last_root_user = 0;
		int root_empty = 0;

		/*
		 * The refcount on a non-root anon_vma got dropped.  Drop
		 * the refcount on the root and check if we need to free it.
		 */
		if (empty && anon_vma != root) {
A
Andrea Arcangeli 已提交
1464
			BUG_ON(atomic_read(&root->external_refcount) <= 0);
1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478
			last_root_user = atomic_dec_and_test(&root->external_refcount);
			root_empty = list_empty(&root->head);
		}
		anon_vma_unlock(anon_vma);

		if (empty) {
			anon_vma_free(anon_vma);
			if (root_empty && last_root_user)
				anon_vma_free(root);
		}
	}
}
#endif

1479 1480 1481 1482 1483 1484 1485 1486 1487
#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;
1488
	struct anon_vma_chain *avc;
1489 1490 1491 1492 1493
	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
1494 1495
	 * are holding mmap_sem. Users without mmap_sem are required to
	 * take a reference count to prevent the anon_vma disappearing
1496 1497 1498 1499
	 */
	anon_vma = page_anon_vma(page);
	if (!anon_vma)
		return ret;
1500
	anon_vma_lock(anon_vma);
1501 1502
	list_for_each_entry(avc, &anon_vma->head, same_anon_vma) {
		struct vm_area_struct *vma = avc->vma;
1503 1504 1505 1506 1507 1508 1509
		unsigned long address = vma_address(page, vma);
		if (address == -EFAULT)
			continue;
		ret = rmap_one(page, vma, address, arg);
		if (ret != SWAP_AGAIN)
			break;
	}
1510
	anon_vma_unlock(anon_vma);
1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555
	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 */
1556

N
Naoya Horiguchi 已提交
1557
#ifdef CONFIG_HUGETLB_PAGE
1558 1559 1560 1561 1562 1563 1564 1565 1566
/*
 * 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;
1567

1568
	BUG_ON(!anon_vma);
1569 1570 1571 1572 1573 1574

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

1575 1576 1577 1578 1579 1580 1581 1582 1583 1584
	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;
1585 1586

	BUG_ON(!PageLocked(page));
1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600
	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
1601
#endif /* CONFIG_HUGETLB_PAGE */