1/*
   2 * mm/rmap.c - physical to virtual reverse mappings
   3 *
   4 * Copyright 2001, Rik van Riel <riel@conectiva.com.br>
   5 * Released under the General Public License (GPL).
   6 *
   7 * Simple, low overhead reverse mapping scheme.
   8 * Please try to keep this thing as modular as possible.
   9 *
  10 * Provides methods for unmapping each kind of mapped page:
  11 * the anon methods track anonymous pages, and
  12 * the file methods track pages belonging to an inode.
  13 *
  14 * Original design by Rik van Riel <riel@conectiva.com.br> 2001
  15 * File methods by Dave McCracken <dmccr@us.ibm.com> 2003, 2004
  16 * Anonymous methods by Andrea Arcangeli <andrea@suse.de> 2004
  17 * Contributions by Hugh Dickins 2003, 2004
  18 */
  19
  20/*
  21 * Lock ordering in mm:
  22 *
  23 * inode->i_mutex	(while writing or truncating, not reading or faulting)
  24 *   mm->mmap_sem
  25 *     page->flags PG_locked (lock_page)
  26 *       hugetlbfs_i_mmap_rwsem_key (in huge_pmd_share)
  27 *         mapping->i_mmap_rwsem
  28 *           anon_vma->rwsem
  29 *             mm->page_table_lock or pte_lock
  30 *               zone_lru_lock (in mark_page_accessed, isolate_lru_page)
  31 *               swap_lock (in swap_duplicate, swap_info_get)
  32 *                 mmlist_lock (in mmput, drain_mmlist and others)
  33 *                 mapping->private_lock (in __set_page_dirty_buffers)
  34 *                   mem_cgroup_{begin,end}_page_stat (memcg->move_lock)
  35 *                     i_pages lock (widely used)
  36 *                 inode->i_lock (in set_page_dirty's __mark_inode_dirty)
  37 *                 bdi.wb->list_lock (in set_page_dirty's __mark_inode_dirty)
  38 *                   sb_lock (within inode_lock in fs/fs-writeback.c)
  39 *                   i_pages lock (widely used, in set_page_dirty,
  40 *                             in arch-dependent flush_dcache_mmap_lock,
  41 *                             within bdi.wb->list_lock in __sync_single_inode)
 
  42 *
  43 * anon_vma->rwsem,mapping->i_mutex      (memory_failure, collect_procs_anon)
  44 *   ->tasklist_lock
  45 *     pte map lock
 
 
 
 
 
 
  46 */
  47
  48#include <linux/mm.h>
  49#include <linux/sched/mm.h>
  50#include <linux/sched/task.h>
  51#include <linux/pagemap.h>
  52#include <linux/swap.h>
  53#include <linux/swapops.h>
  54#include <linux/slab.h>
  55#include <linux/init.h>
  56#include <linux/ksm.h>
  57#include <linux/rmap.h>
  58#include <linux/rcupdate.h>
  59#include <linux/export.h>
  60#include <linux/memcontrol.h>
  61#include <linux/mmu_notifier.h>
  62#include <linux/migrate.h>
  63#include <linux/hugetlb.h>
 
  64#include <linux/backing-dev.h>
  65#include <linux/page_idle.h>
  66#include <linux/memremap.h>
 
 
  67
  68#include <asm/tlbflush.h>
  69
 
  70#include <trace/events/tlb.h>
 
  71
  72#include "internal.h"
  73
  74static struct kmem_cache *anon_vma_cachep;
  75static struct kmem_cache *anon_vma_chain_cachep;
  76
  77static inline struct anon_vma *anon_vma_alloc(void)
  78{
  79	struct anon_vma *anon_vma;
  80
  81	anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL);
  82	if (anon_vma) {
  83		atomic_set(&anon_vma->refcount, 1);
  84		anon_vma->degree = 1;	/* Reference for first vma */
 
  85		anon_vma->parent = anon_vma;
  86		/*
  87		 * Initialise the anon_vma root to point to itself. If called
  88		 * from fork, the root will be reset to the parents anon_vma.
  89		 */
  90		anon_vma->root = anon_vma;
  91	}
  92
  93	return anon_vma;
  94}
  95
  96static inline void anon_vma_free(struct anon_vma *anon_vma)
  97{
  98	VM_BUG_ON(atomic_read(&anon_vma->refcount));
  99
 100	/*
 101	 * Synchronize against page_lock_anon_vma_read() such that
 102	 * we can safely hold the lock without the anon_vma getting
 103	 * freed.
 104	 *
 105	 * Relies on the full mb implied by the atomic_dec_and_test() from
 106	 * put_anon_vma() against the acquire barrier implied by
 107	 * down_read_trylock() from page_lock_anon_vma_read(). This orders:
 108	 *
 109	 * page_lock_anon_vma_read()	VS	put_anon_vma()
 110	 *   down_read_trylock()		  atomic_dec_and_test()
 111	 *   LOCK				  MB
 112	 *   atomic_read()			  rwsem_is_locked()
 113	 *
 114	 * LOCK should suffice since the actual taking of the lock must
 115	 * happen _before_ what follows.
 116	 */
 117	might_sleep();
 118	if (rwsem_is_locked(&anon_vma->root->rwsem)) {
 119		anon_vma_lock_write(anon_vma);
 120		anon_vma_unlock_write(anon_vma);
 121	}
 122
 123	kmem_cache_free(anon_vma_cachep, anon_vma);
 124}
 125
 126static inline struct anon_vma_chain *anon_vma_chain_alloc(gfp_t gfp)
 127{
 128	return kmem_cache_alloc(anon_vma_chain_cachep, gfp);
 129}
 130
 131static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain)
 132{
 133	kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain);
 134}
 135
 136static void anon_vma_chain_link(struct vm_area_struct *vma,
 137				struct anon_vma_chain *avc,
 138				struct anon_vma *anon_vma)
 139{
 140	avc->vma = vma;
 141	avc->anon_vma = anon_vma;
 142	list_add(&avc->same_vma, &vma->anon_vma_chain);
 143	anon_vma_interval_tree_insert(avc, &anon_vma->rb_root);
 144}
 145
 146/**
 147 * __anon_vma_prepare - attach an anon_vma to a memory region
 148 * @vma: the memory region in question
 149 *
 150 * This makes sure the memory mapping described by 'vma' has
 151 * an 'anon_vma' attached to it, so that we can associate the
 152 * anonymous pages mapped into it with that anon_vma.
 153 *
 154 * The common case will be that we already have one, which
 155 * is handled inline by anon_vma_prepare(). But if
 156 * not we either need to find an adjacent mapping that we
 157 * can re-use the anon_vma from (very common when the only
 158 * reason for splitting a vma has been mprotect()), or we
 159 * allocate a new one.
 160 *
 161 * Anon-vma allocations are very subtle, because we may have
 162 * optimistically looked up an anon_vma in page_lock_anon_vma_read()
 163 * and that may actually touch the spinlock even in the newly
 164 * allocated vma (it depends on RCU to make sure that the
 165 * anon_vma isn't actually destroyed).
 166 *
 167 * As a result, we need to do proper anon_vma locking even
 168 * for the new allocation. At the same time, we do not want
 169 * to do any locking for the common case of already having
 170 * an anon_vma.
 171 *
 172 * This must be called with the mmap_sem held for reading.
 173 */
 174int __anon_vma_prepare(struct vm_area_struct *vma)
 175{
 176	struct mm_struct *mm = vma->vm_mm;
 177	struct anon_vma *anon_vma, *allocated;
 178	struct anon_vma_chain *avc;
 179
 180	might_sleep();
 181
 182	avc = anon_vma_chain_alloc(GFP_KERNEL);
 183	if (!avc)
 184		goto out_enomem;
 185
 186	anon_vma = find_mergeable_anon_vma(vma);
 187	allocated = NULL;
 188	if (!anon_vma) {
 189		anon_vma = anon_vma_alloc();
 190		if (unlikely(!anon_vma))
 191			goto out_enomem_free_avc;
 
 192		allocated = anon_vma;
 193	}
 194
 195	anon_vma_lock_write(anon_vma);
 196	/* page_table_lock to protect against threads */
 197	spin_lock(&mm->page_table_lock);
 198	if (likely(!vma->anon_vma)) {
 199		vma->anon_vma = anon_vma;
 200		anon_vma_chain_link(vma, avc, anon_vma);
 201		/* vma reference or self-parent link for new root */
 202		anon_vma->degree++;
 203		allocated = NULL;
 204		avc = NULL;
 205	}
 206	spin_unlock(&mm->page_table_lock);
 207	anon_vma_unlock_write(anon_vma);
 208
 209	if (unlikely(allocated))
 210		put_anon_vma(allocated);
 211	if (unlikely(avc))
 212		anon_vma_chain_free(avc);
 213
 214	return 0;
 215
 216 out_enomem_free_avc:
 217	anon_vma_chain_free(avc);
 218 out_enomem:
 219	return -ENOMEM;
 220}
 221
 222/*
 223 * This is a useful helper function for locking the anon_vma root as
 224 * we traverse the vma->anon_vma_chain, looping over anon_vma's that
 225 * have the same vma.
 226 *
 227 * Such anon_vma's should have the same root, so you'd expect to see
 228 * just a single mutex_lock for the whole traversal.
 229 */
 230static inline struct anon_vma *lock_anon_vma_root(struct anon_vma *root, struct anon_vma *anon_vma)
 231{
 232	struct anon_vma *new_root = anon_vma->root;
 233	if (new_root != root) {
 234		if (WARN_ON_ONCE(root))
 235			up_write(&root->rwsem);
 236		root = new_root;
 237		down_write(&root->rwsem);
 238	}
 239	return root;
 240}
 241
 242static inline void unlock_anon_vma_root(struct anon_vma *root)
 243{
 244	if (root)
 245		up_write(&root->rwsem);
 246}
 247
 248/*
 249 * Attach the anon_vmas from src to dst.
 250 * Returns 0 on success, -ENOMEM on failure.
 251 *
 252 * If dst->anon_vma is NULL this function tries to find and reuse existing
 253 * anon_vma which has no vmas and only one child anon_vma. This prevents
 254 * degradation of anon_vma hierarchy to endless linear chain in case of
 255 * constantly forking task. On the other hand, an anon_vma with more than one
 256 * child isn't reused even if there was no alive vma, thus rmap walker has a
 257 * good chance of avoiding scanning the whole hierarchy when it searches where
 258 * page is mapped.
 
 
 
 
 
 
 259 */
 260int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src)
 261{
 262	struct anon_vma_chain *avc, *pavc;
 263	struct anon_vma *root = NULL;
 264
 265	list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) {
 266		struct anon_vma *anon_vma;
 267
 268		avc = anon_vma_chain_alloc(GFP_NOWAIT | __GFP_NOWARN);
 269		if (unlikely(!avc)) {
 270			unlock_anon_vma_root(root);
 271			root = NULL;
 272			avc = anon_vma_chain_alloc(GFP_KERNEL);
 273			if (!avc)
 274				goto enomem_failure;
 275		}
 276		anon_vma = pavc->anon_vma;
 277		root = lock_anon_vma_root(root, anon_vma);
 278		anon_vma_chain_link(dst, avc, anon_vma);
 279
 280		/*
 281		 * Reuse existing anon_vma if its degree lower than two,
 282		 * that means it has no vma and only one anon_vma child.
 283		 *
 284		 * Do not chose parent anon_vma, otherwise first child
 285		 * will always reuse it. Root anon_vma is never reused:
 286		 * it has self-parent reference and at least one child.
 287		 */
 288		if (!dst->anon_vma && anon_vma != src->anon_vma &&
 289				anon_vma->degree < 2)
 
 290			dst->anon_vma = anon_vma;
 291	}
 292	if (dst->anon_vma)
 293		dst->anon_vma->degree++;
 294	unlock_anon_vma_root(root);
 295	return 0;
 296
 297 enomem_failure:
 298	/*
 299	 * dst->anon_vma is dropped here otherwise its degree can be incorrectly
 300	 * decremented in unlink_anon_vmas().
 301	 * We can safely do this because callers of anon_vma_clone() don't care
 302	 * about dst->anon_vma if anon_vma_clone() failed.
 303	 */
 304	dst->anon_vma = NULL;
 305	unlink_anon_vmas(dst);
 306	return -ENOMEM;
 307}
 308
 309/*
 310 * Attach vma to its own anon_vma, as well as to the anon_vmas that
 311 * the corresponding VMA in the parent process is attached to.
 312 * Returns 0 on success, non-zero on failure.
 313 */
 314int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma)
 315{
 316	struct anon_vma_chain *avc;
 317	struct anon_vma *anon_vma;
 318	int error;
 319
 320	/* Don't bother if the parent process has no anon_vma here. */
 321	if (!pvma->anon_vma)
 322		return 0;
 323
 324	/* Drop inherited anon_vma, we'll reuse existing or allocate new. */
 325	vma->anon_vma = NULL;
 326
 327	/*
 328	 * First, attach the new VMA to the parent VMA's anon_vmas,
 329	 * so rmap can find non-COWed pages in child processes.
 330	 */
 331	error = anon_vma_clone(vma, pvma);
 332	if (error)
 333		return error;
 334
 335	/* An existing anon_vma has been reused, all done then. */
 336	if (vma->anon_vma)
 337		return 0;
 338
 339	/* Then add our own anon_vma. */
 340	anon_vma = anon_vma_alloc();
 341	if (!anon_vma)
 342		goto out_error;
 
 343	avc = anon_vma_chain_alloc(GFP_KERNEL);
 344	if (!avc)
 345		goto out_error_free_anon_vma;
 346
 347	/*
 348	 * The root anon_vma's spinlock is the lock actually used when we
 349	 * lock any of the anon_vmas in this anon_vma tree.
 350	 */
 351	anon_vma->root = pvma->anon_vma->root;
 352	anon_vma->parent = pvma->anon_vma;
 353	/*
 354	 * With refcounts, an anon_vma can stay around longer than the
 355	 * process it belongs to. The root anon_vma needs to be pinned until
 356	 * this anon_vma is freed, because the lock lives in the root.
 357	 */
 358	get_anon_vma(anon_vma->root);
 359	/* Mark this anon_vma as the one where our new (COWed) pages go. */
 360	vma->anon_vma = anon_vma;
 361	anon_vma_lock_write(anon_vma);
 362	anon_vma_chain_link(vma, avc, anon_vma);
 363	anon_vma->parent->degree++;
 364	anon_vma_unlock_write(anon_vma);
 365
 366	return 0;
 367
 368 out_error_free_anon_vma:
 369	put_anon_vma(anon_vma);
 370 out_error:
 371	unlink_anon_vmas(vma);
 372	return -ENOMEM;
 373}
 374
 375void unlink_anon_vmas(struct vm_area_struct *vma)
 376{
 377	struct anon_vma_chain *avc, *next;
 378	struct anon_vma *root = NULL;
 379
 380	/*
 381	 * Unlink each anon_vma chained to the VMA.  This list is ordered
 382	 * from newest to oldest, ensuring the root anon_vma gets freed last.
 383	 */
 384	list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
 385		struct anon_vma *anon_vma = avc->anon_vma;
 386
 387		root = lock_anon_vma_root(root, anon_vma);
 388		anon_vma_interval_tree_remove(avc, &anon_vma->rb_root);
 389
 390		/*
 391		 * Leave empty anon_vmas on the list - we'll need
 392		 * to free them outside the lock.
 393		 */
 394		if (RB_EMPTY_ROOT(&anon_vma->rb_root.rb_root)) {
 395			anon_vma->parent->degree--;
 396			continue;
 397		}
 398
 399		list_del(&avc->same_vma);
 400		anon_vma_chain_free(avc);
 401	}
 402	if (vma->anon_vma)
 403		vma->anon_vma->degree--;
 
 
 
 
 
 
 
 404	unlock_anon_vma_root(root);
 405
 406	/*
 407	 * Iterate the list once more, it now only contains empty and unlinked
 408	 * anon_vmas, destroy them. Could not do before due to __put_anon_vma()
 409	 * needing to write-acquire the anon_vma->root->rwsem.
 410	 */
 411	list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
 412		struct anon_vma *anon_vma = avc->anon_vma;
 413
 414		VM_WARN_ON(anon_vma->degree);
 
 415		put_anon_vma(anon_vma);
 416
 417		list_del(&avc->same_vma);
 418		anon_vma_chain_free(avc);
 419	}
 420}
 421
 422static void anon_vma_ctor(void *data)
 423{
 424	struct anon_vma *anon_vma = data;
 425
 426	init_rwsem(&anon_vma->rwsem);
 427	atomic_set(&anon_vma->refcount, 0);
 428	anon_vma->rb_root = RB_ROOT_CACHED;
 429}
 430
 431void __init anon_vma_init(void)
 432{
 433	anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
 434			0, SLAB_TYPESAFE_BY_RCU|SLAB_PANIC|SLAB_ACCOUNT,
 435			anon_vma_ctor);
 436	anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain,
 437			SLAB_PANIC|SLAB_ACCOUNT);
 438}
 439
 440/*
 441 * Getting a lock on a stable anon_vma from a page off the LRU is tricky!
 442 *
 443 * Since there is no serialization what so ever against page_remove_rmap()
 444 * the best this function can do is return a locked anon_vma that might
 445 * have been relevant to this page.
 446 *
 447 * The page might have been remapped to a different anon_vma or the anon_vma
 448 * returned may already be freed (and even reused).
 449 *
 450 * In case it was remapped to a different anon_vma, the new anon_vma will be a
 451 * child of the old anon_vma, and the anon_vma lifetime rules will therefore
 452 * ensure that any anon_vma obtained from the page will still be valid for as
 453 * long as we observe page_mapped() [ hence all those page_mapped() tests ].
 454 *
 455 * All users of this function must be very careful when walking the anon_vma
 456 * chain and verify that the page in question is indeed mapped in it
 457 * [ something equivalent to page_mapped_in_vma() ].
 458 *
 459 * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap()
 460 * that the anon_vma pointer from page->mapping is valid if there is a
 461 * mapcount, we can dereference the anon_vma after observing those.
 
 462 */
 463struct anon_vma *page_get_anon_vma(struct page *page)
 464{
 465	struct anon_vma *anon_vma = NULL;
 466	unsigned long anon_mapping;
 467
 468	rcu_read_lock();
 469	anon_mapping = (unsigned long)READ_ONCE(page->mapping);
 470	if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
 471		goto out;
 472	if (!page_mapped(page))
 473		goto out;
 474
 475	anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
 476	if (!atomic_inc_not_zero(&anon_vma->refcount)) {
 477		anon_vma = NULL;
 478		goto out;
 479	}
 480
 481	/*
 482	 * If this page is still mapped, then its anon_vma cannot have been
 483	 * freed.  But if it has been unmapped, we have no security against the
 484	 * anon_vma structure being freed and reused (for another anon_vma:
 485	 * SLAB_TYPESAFE_BY_RCU guarantees that - so the atomic_inc_not_zero()
 486	 * above cannot corrupt).
 487	 */
 488	if (!page_mapped(page)) {
 489		rcu_read_unlock();
 490		put_anon_vma(anon_vma);
 491		return NULL;
 492	}
 493out:
 494	rcu_read_unlock();
 495
 496	return anon_vma;
 497}
 498
 499/*
 500 * Similar to page_get_anon_vma() except it locks the anon_vma.
 501 *
 502 * Its a little more complex as it tries to keep the fast path to a single
 503 * atomic op -- the trylock. If we fail the trylock, we fall back to getting a
 504 * reference like with page_get_anon_vma() and then block on the mutex.
 
 505 */
 506struct anon_vma *page_lock_anon_vma_read(struct page *page)
 
 507{
 508	struct anon_vma *anon_vma = NULL;
 509	struct anon_vma *root_anon_vma;
 510	unsigned long anon_mapping;
 511
 512	rcu_read_lock();
 513	anon_mapping = (unsigned long)READ_ONCE(page->mapping);
 514	if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
 515		goto out;
 516	if (!page_mapped(page))
 517		goto out;
 518
 519	anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
 520	root_anon_vma = READ_ONCE(anon_vma->root);
 521	if (down_read_trylock(&root_anon_vma->rwsem)) {
 522		/*
 523		 * If the page is still mapped, then this anon_vma is still
 524		 * its anon_vma, and holding the mutex ensures that it will
 525		 * not go away, see anon_vma_free().
 526		 */
 527		if (!page_mapped(page)) {
 528			up_read(&root_anon_vma->rwsem);
 529			anon_vma = NULL;
 530		}
 531		goto out;
 532	}
 533
 
 
 
 
 
 
 534	/* trylock failed, we got to sleep */
 535	if (!atomic_inc_not_zero(&anon_vma->refcount)) {
 536		anon_vma = NULL;
 537		goto out;
 538	}
 539
 540	if (!page_mapped(page)) {
 541		rcu_read_unlock();
 542		put_anon_vma(anon_vma);
 543		return NULL;
 544	}
 545
 546	/* we pinned the anon_vma, its safe to sleep */
 547	rcu_read_unlock();
 548	anon_vma_lock_read(anon_vma);
 549
 550	if (atomic_dec_and_test(&anon_vma->refcount)) {
 551		/*
 552		 * Oops, we held the last refcount, release the lock
 553		 * and bail -- can't simply use put_anon_vma() because
 554		 * we'll deadlock on the anon_vma_lock_write() recursion.
 555		 */
 556		anon_vma_unlock_read(anon_vma);
 557		__put_anon_vma(anon_vma);
 558		anon_vma = NULL;
 559	}
 560
 561	return anon_vma;
 562
 563out:
 564	rcu_read_unlock();
 565	return anon_vma;
 566}
 567
 568void page_unlock_anon_vma_read(struct anon_vma *anon_vma)
 569{
 570	anon_vma_unlock_read(anon_vma);
 571}
 572
 573#ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
 574/*
 575 * Flush TLB entries for recently unmapped pages from remote CPUs. It is
 576 * important if a PTE was dirty when it was unmapped that it's flushed
 577 * before any IO is initiated on the page to prevent lost writes. Similarly,
 578 * it must be flushed before freeing to prevent data leakage.
 579 */
 580void try_to_unmap_flush(void)
 581{
 582	struct tlbflush_unmap_batch *tlb_ubc = &current->tlb_ubc;
 583
 584	if (!tlb_ubc->flush_required)
 585		return;
 586
 587	arch_tlbbatch_flush(&tlb_ubc->arch);
 588	tlb_ubc->flush_required = false;
 589	tlb_ubc->writable = false;
 590}
 591
 592/* Flush iff there are potentially writable TLB entries that can race with IO */
 593void try_to_unmap_flush_dirty(void)
 594{
 595	struct tlbflush_unmap_batch *tlb_ubc = &current->tlb_ubc;
 596
 597	if (tlb_ubc->writable)
 598		try_to_unmap_flush();
 599}
 600
 
 
 
 
 
 
 
 
 
 
 601static void set_tlb_ubc_flush_pending(struct mm_struct *mm, bool writable)
 602{
 603	struct tlbflush_unmap_batch *tlb_ubc = &current->tlb_ubc;
 
 604
 605	arch_tlbbatch_add_mm(&tlb_ubc->arch, mm);
 606	tlb_ubc->flush_required = true;
 607
 608	/*
 609	 * Ensure compiler does not re-order the setting of tlb_flush_batched
 610	 * before the PTE is cleared.
 611	 */
 612	barrier();
 613	mm->tlb_flush_batched = true;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 614
 615	/*
 616	 * If the PTE was dirty then it's best to assume it's writable. The
 617	 * caller must use try_to_unmap_flush_dirty() or try_to_unmap_flush()
 618	 * before the page is queued for IO.
 619	 */
 620	if (writable)
 621		tlb_ubc->writable = true;
 622}
 623
 624/*
 625 * Returns true if the TLB flush should be deferred to the end of a batch of
 626 * unmap operations to reduce IPIs.
 627 */
 628static bool should_defer_flush(struct mm_struct *mm, enum ttu_flags flags)
 629{
 630	bool should_defer = false;
 631
 632	if (!(flags & TTU_BATCH_FLUSH))
 633		return false;
 634
 635	/* If remote CPUs need to be flushed then defer batch the flush */
 636	if (cpumask_any_but(mm_cpumask(mm), get_cpu()) < nr_cpu_ids)
 637		should_defer = true;
 638	put_cpu();
 639
 640	return should_defer;
 641}
 642
 643/*
 644 * Reclaim unmaps pages under the PTL but do not flush the TLB prior to
 645 * releasing the PTL if TLB flushes are batched. It's possible for a parallel
 646 * operation such as mprotect or munmap to race between reclaim unmapping
 647 * the page and flushing the page. If this race occurs, it potentially allows
 648 * access to data via a stale TLB entry. Tracking all mm's that have TLB
 649 * batching in flight would be expensive during reclaim so instead track
 650 * whether TLB batching occurred in the past and if so then do a flush here
 651 * if required. This will cost one additional flush per reclaim cycle paid
 652 * by the first operation at risk such as mprotect and mumap.
 653 *
 654 * This must be called under the PTL so that an access to tlb_flush_batched
 655 * that is potentially a "reclaim vs mprotect/munmap/etc" race will synchronise
 656 * via the PTL.
 657 */
 658void flush_tlb_batched_pending(struct mm_struct *mm)
 659{
 660	if (mm->tlb_flush_batched) {
 661		flush_tlb_mm(mm);
 
 662
 
 
 663		/*
 664		 * Do not allow the compiler to re-order the clearing of
 665		 * tlb_flush_batched before the tlb is flushed.
 666		 */
 667		barrier();
 668		mm->tlb_flush_batched = false;
 669	}
 670}
 671#else
 672static void set_tlb_ubc_flush_pending(struct mm_struct *mm, bool writable)
 673{
 674}
 675
 676static bool should_defer_flush(struct mm_struct *mm, enum ttu_flags flags)
 677{
 678	return false;
 679}
 680#endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
 681
 682/*
 683 * At what user virtual address is page expected in vma?
 684 * Caller should check the page is actually part of the vma.
 685 */
 686unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
 687{
 688	unsigned long address;
 689	if (PageAnon(page)) {
 690		struct anon_vma *page__anon_vma = page_anon_vma(page);
 691		/*
 692		 * Note: swapoff's unuse_vma() is more efficient with this
 693		 * check, and needs it to match anon_vma when KSM is active.
 694		 */
 695		if (!vma->anon_vma || !page__anon_vma ||
 696		    vma->anon_vma->root != page__anon_vma->root)
 697			return -EFAULT;
 698	} else if (page->mapping) {
 699		if (!vma->vm_file || vma->vm_file->f_mapping != page->mapping)
 700			return -EFAULT;
 701	} else
 702		return -EFAULT;
 703	address = __vma_address(page, vma);
 704	if (unlikely(address < vma->vm_start || address >= vma->vm_end))
 705		return -EFAULT;
 706	return address;
 
 
 707}
 708
 
 
 
 
 
 709pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address)
 710{
 711	pgd_t *pgd;
 712	p4d_t *p4d;
 713	pud_t *pud;
 714	pmd_t *pmd = NULL;
 715	pmd_t pmde;
 716
 717	pgd = pgd_offset(mm, address);
 718	if (!pgd_present(*pgd))
 719		goto out;
 720
 721	p4d = p4d_offset(pgd, address);
 722	if (!p4d_present(*p4d))
 723		goto out;
 724
 725	pud = pud_offset(p4d, address);
 726	if (!pud_present(*pud))
 727		goto out;
 728
 729	pmd = pmd_offset(pud, address);
 730	/*
 731	 * Some THP functions use the sequence pmdp_huge_clear_flush(), set_pmd_at()
 732	 * without holding anon_vma lock for write.  So when looking for a
 733	 * genuine pmde (in which to find pte), test present and !THP together.
 734	 */
 735	pmde = *pmd;
 736	barrier();
 737	if (!pmd_present(pmde) || pmd_trans_huge(pmde))
 738		pmd = NULL;
 739out:
 740	return pmd;
 741}
 742
 743struct page_referenced_arg {
 744	int mapcount;
 745	int referenced;
 746	unsigned long vm_flags;
 747	struct mem_cgroup *memcg;
 748};
 749/*
 750 * arg: page_referenced_arg will be passed
 751 */
 752static bool page_referenced_one(struct page *page, struct vm_area_struct *vma,
 753			unsigned long address, void *arg)
 754{
 755	struct page_referenced_arg *pra = arg;
 756	struct page_vma_mapped_walk pvmw = {
 757		.page = page,
 758		.vma = vma,
 759		.address = address,
 760	};
 761	int referenced = 0;
 762
 763	while (page_vma_mapped_walk(&pvmw)) {
 764		address = pvmw.address;
 765
 766		if (vma->vm_flags & VM_LOCKED) {
 
 
 
 767			page_vma_mapped_walk_done(&pvmw);
 768			pra->vm_flags |= VM_LOCKED;
 769			return false; /* To break the loop */
 770		}
 771
 772		if (pvmw.pte) {
 
 
 
 
 
 
 773			if (ptep_clear_flush_young_notify(vma, address,
 774						pvmw.pte)) {
 775				/*
 776				 * Don't treat a reference through
 777				 * a sequentially read mapping as such.
 778				 * If the page has been used in another mapping,
 779				 * we will catch it; if this other mapping is
 780				 * already gone, the unmap path will have set
 781				 * PG_referenced or activated the page.
 782				 */
 783				if (likely(!(vma->vm_flags & VM_SEQ_READ)))
 784					referenced++;
 785			}
 786		} else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
 787			if (pmdp_clear_flush_young_notify(vma, address,
 788						pvmw.pmd))
 789				referenced++;
 790		} else {
 791			/* unexpected pmd-mapped page? */
 792			WARN_ON_ONCE(1);
 793		}
 794
 795		pra->mapcount--;
 796	}
 797
 798	if (referenced)
 799		clear_page_idle(page);
 800	if (test_and_clear_page_young(page))
 801		referenced++;
 802
 803	if (referenced) {
 804		pra->referenced++;
 805		pra->vm_flags |= vma->vm_flags;
 806	}
 807
 808	if (!pra->mapcount)
 809		return false; /* To break the loop */
 810
 811	return true;
 812}
 813
 814static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg)
 815{
 816	struct page_referenced_arg *pra = arg;
 817	struct mem_cgroup *memcg = pra->memcg;
 818
 819	if (!mm_match_cgroup(vma->vm_mm, memcg))
 820		return true;
 821
 822	return false;
 823}
 824
 825/**
 826 * page_referenced - test if the page was referenced
 827 * @page: the page to test
 828 * @is_locked: caller holds lock on the page
 829 * @memcg: target memory cgroup
 830 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
 
 
 831 *
 832 * Quick test_and_clear_referenced for all mappings to a page,
 833 * returns the number of ptes which referenced the page.
 834 */
 835int page_referenced(struct page *page,
 836		    int is_locked,
 837		    struct mem_cgroup *memcg,
 838		    unsigned long *vm_flags)
 839{
 840	int we_locked = 0;
 841	struct page_referenced_arg pra = {
 842		.mapcount = total_mapcount(page),
 843		.memcg = memcg,
 844	};
 845	struct rmap_walk_control rwc = {
 846		.rmap_one = page_referenced_one,
 847		.arg = (void *)&pra,
 848		.anon_lock = page_lock_anon_vma_read,
 
 849	};
 850
 851	*vm_flags = 0;
 852	if (!page_mapped(page))
 853		return 0;
 854
 855	if (!page_rmapping(page))
 856		return 0;
 857
 858	if (!is_locked && (!PageAnon(page) || PageKsm(page))) {
 859		we_locked = trylock_page(page);
 860		if (!we_locked)
 861			return 1;
 862	}
 863
 864	/*
 865	 * If we are reclaiming on behalf of a cgroup, skip
 866	 * counting on behalf of references from different
 867	 * cgroups
 868	 */
 869	if (memcg) {
 870		rwc.invalid_vma = invalid_page_referenced_vma;
 871	}
 872
 873	rmap_walk(page, &rwc);
 874	*vm_flags = pra.vm_flags;
 875
 876	if (we_locked)
 877		unlock_page(page);
 878
 879	return pra.referenced;
 880}
 881
 882static bool page_mkclean_one(struct page *page, struct vm_area_struct *vma,
 883			    unsigned long address, void *arg)
 884{
 885	struct page_vma_mapped_walk pvmw = {
 886		.page = page,
 887		.vma = vma,
 888		.address = address,
 889		.flags = PVMW_SYNC,
 890	};
 891	unsigned long start = address, end;
 892	int *cleaned = arg;
 893
 894	/*
 895	 * We have to assume the worse case ie pmd for invalidation. Note that
 896	 * the page can not be free from this function.
 897	 */
 898	end = min(vma->vm_end, start + (PAGE_SIZE << compound_order(page)));
 899	mmu_notifier_invalidate_range_start(vma->vm_mm, start, end);
 
 
 900
 901	while (page_vma_mapped_walk(&pvmw)) {
 902		unsigned long cstart;
 903		int ret = 0;
 904
 905		cstart = address = pvmw.address;
 906		if (pvmw.pte) {
 907			pte_t entry;
 908			pte_t *pte = pvmw.pte;
 909
 910			if (!pte_dirty(*pte) && !pte_write(*pte))
 911				continue;
 912
 913			flush_cache_page(vma, address, pte_pfn(*pte));
 914			entry = ptep_clear_flush(vma, address, pte);
 915			entry = pte_wrprotect(entry);
 916			entry = pte_mkclean(entry);
 917			set_pte_at(vma->vm_mm, address, pte, entry);
 918			ret = 1;
 919		} else {
 920#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
 921			pmd_t *pmd = pvmw.pmd;
 922			pmd_t entry;
 923
 924			if (!pmd_dirty(*pmd) && !pmd_write(*pmd))
 925				continue;
 926
 927			flush_cache_page(vma, address, page_to_pfn(page));
 928			entry = pmdp_huge_clear_flush(vma, address, pmd);
 
 929			entry = pmd_wrprotect(entry);
 930			entry = pmd_mkclean(entry);
 931			set_pmd_at(vma->vm_mm, address, pmd, entry);
 932			cstart &= PMD_MASK;
 933			ret = 1;
 934#else
 935			/* unexpected pmd-mapped page? */
 936			WARN_ON_ONCE(1);
 937#endif
 938		}
 939
 940		/*
 941		 * No need to call mmu_notifier_invalidate_range() as we are
 942		 * downgrading page table protection not changing it to point
 943		 * to a new page.
 944		 *
 945		 * See Documentation/vm/mmu_notifier.txt
 946		 */
 947		if (ret)
 948			(*cleaned)++;
 949	}
 950
 951	mmu_notifier_invalidate_range_end(vma->vm_mm, start, end);
 
 
 
 
 
 
 
 
 
 
 
 952
 953	return true;
 954}
 955
 956static bool invalid_mkclean_vma(struct vm_area_struct *vma, void *arg)
 957{
 958	if (vma->vm_flags & VM_SHARED)
 959		return false;
 960
 961	return true;
 962}
 963
 964int page_mkclean(struct page *page)
 965{
 966	int cleaned = 0;
 967	struct address_space *mapping;
 968	struct rmap_walk_control rwc = {
 969		.arg = (void *)&cleaned,
 970		.rmap_one = page_mkclean_one,
 971		.invalid_vma = invalid_mkclean_vma,
 972	};
 973
 974	BUG_ON(!PageLocked(page));
 975
 976	if (!page_mapped(page))
 977		return 0;
 978
 979	mapping = page_mapping(page);
 980	if (!mapping)
 981		return 0;
 982
 983	rmap_walk(page, &rwc);
 984
 985	return cleaned;
 986}
 987EXPORT_SYMBOL_GPL(page_mkclean);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 988
 989/**
 990 * page_move_anon_rmap - move a page to our anon_vma
 991 * @page:	the page to move to our anon_vma
 992 * @vma:	the vma the page belongs to
 993 *
 994 * When a page belongs exclusively to one process after a COW event,
 995 * that page can be moved into the anon_vma that belongs to just that
 996 * process, so the rmap code will not search the parent or sibling
 997 * processes.
 998 */
 999void page_move_anon_rmap(struct page *page, struct vm_area_struct *vma)
1000{
1001	struct anon_vma *anon_vma = vma->anon_vma;
 
1002
1003	page = compound_head(page);
1004
1005	VM_BUG_ON_PAGE(!PageLocked(page), page);
1006	VM_BUG_ON_VMA(!anon_vma, vma);
1007
1008	anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1009	/*
1010	 * Ensure that anon_vma and the PAGE_MAPPING_ANON bit are written
1011	 * simultaneously, so a concurrent reader (eg page_referenced()'s
1012	 * PageAnon()) will not see one without the other.
1013	 */
1014	WRITE_ONCE(page->mapping, (struct address_space *) anon_vma);
 
1015}
1016
1017/**
1018 * __page_set_anon_rmap - set up new anonymous rmap
1019 * @page:	Page to add to rmap	
1020 * @vma:	VM area to add page to.
1021 * @address:	User virtual address of the mapping	
1022 * @exclusive:	the page is exclusively owned by the current process
1023 */
1024static void __page_set_anon_rmap(struct page *page,
1025	struct vm_area_struct *vma, unsigned long address, int exclusive)
1026{
1027	struct anon_vma *anon_vma = vma->anon_vma;
1028
1029	BUG_ON(!anon_vma);
1030
1031	if (PageAnon(page))
1032		return;
1033
1034	/*
1035	 * If the page isn't exclusively mapped into this vma,
1036	 * we must use the _oldest_ possible anon_vma for the
1037	 * page mapping!
1038	 */
1039	if (!exclusive)
1040		anon_vma = anon_vma->root;
1041
 
 
 
 
 
 
1042	anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1043	page->mapping = (struct address_space *) anon_vma;
1044	page->index = linear_page_index(vma, address);
 
 
 
1045}
1046
1047/**
1048 * __page_check_anon_rmap - sanity check anonymous rmap addition
1049 * @page:	the page to add the mapping to
1050 * @vma:	the vm area in which the mapping is added
1051 * @address:	the user virtual address mapped
1052 */
1053static void __page_check_anon_rmap(struct page *page,
1054	struct vm_area_struct *vma, unsigned long address)
1055{
1056#ifdef CONFIG_DEBUG_VM
1057	/*
1058	 * The page's anon-rmap details (mapping and index) are guaranteed to
1059	 * be set up correctly at this point.
1060	 *
1061	 * We have exclusion against page_add_anon_rmap because the caller
1062	 * always holds the page locked, except if called from page_dup_rmap,
1063	 * in which case the page is already known to be setup.
1064	 *
1065	 * We have exclusion against page_add_new_anon_rmap because those pages
1066	 * are initially only visible via the pagetables, and the pte is locked
1067	 * over the call to page_add_new_anon_rmap.
1068	 */
1069	BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root);
1070	BUG_ON(page_to_pgoff(page) != linear_page_index(vma, address));
1071#endif
 
1072}
1073
1074/**
1075 * page_add_anon_rmap - add pte mapping to an anonymous page
1076 * @page:	the page to add the mapping to
1077 * @vma:	the vm area in which the mapping is added
1078 * @address:	the user virtual address mapped
1079 * @compound:	charge the page as compound or small page
1080 *
1081 * The caller needs to hold the pte lock, and the page must be locked in
1082 * the anon_vma case: to serialize mapping,index checking after setting,
1083 * and to ensure that PageAnon is not being upgraded racily to PageKsm
1084 * (but PageKsm is never downgraded to PageAnon).
1085 */
1086void page_add_anon_rmap(struct page *page,
1087	struct vm_area_struct *vma, unsigned long address, bool compound)
1088{
1089	do_page_add_anon_rmap(page, vma, address, compound ? RMAP_COMPOUND : 0);
1090}
1091
1092/*
1093 * Special version of the above for do_swap_page, which often runs
1094 * into pages that are exclusively owned by the current process.
1095 * Everybody else should continue to use page_add_anon_rmap above.
1096 */
1097void do_page_add_anon_rmap(struct page *page,
1098	struct vm_area_struct *vma, unsigned long address, int flags)
1099{
 
 
1100	bool compound = flags & RMAP_COMPOUND;
1101	bool first;
1102
1103	if (compound) {
1104		atomic_t *mapcount;
1105		VM_BUG_ON_PAGE(!PageLocked(page), page);
1106		VM_BUG_ON_PAGE(!PageTransHuge(page), page);
1107		mapcount = compound_mapcount_ptr(page);
1108		first = atomic_inc_and_test(mapcount);
1109	} else {
1110		first = atomic_inc_and_test(&page->_mapcount);
1111	}
 
 
 
 
 
 
 
1112
1113	if (first) {
1114		int nr = compound ? hpage_nr_pages(page) : 1;
1115		/*
1116		 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1117		 * these counters are not modified in interrupt context, and
1118		 * pte lock(a spinlock) is held, which implies preemption
1119		 * disabled.
1120		 */
1121		if (compound)
1122			__inc_node_page_state(page, NR_ANON_THPS);
1123		__mod_node_page_state(page_pgdat(page), NR_ANON_MAPPED, nr);
 
 
 
 
1124	}
1125	if (unlikely(PageKsm(page)))
1126		return;
1127
1128	VM_BUG_ON_PAGE(!PageLocked(page), page);
 
 
 
 
 
 
 
 
 
1129
1130	/* address might be in next vma when migration races vma_adjust */
1131	if (first)
1132		__page_set_anon_rmap(page, vma, address,
1133				flags & RMAP_EXCLUSIVE);
1134	else
1135		__page_check_anon_rmap(page, vma, address);
 
 
1136}
1137
1138/**
1139 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
1140 * @page:	the page to add the mapping to
1141 * @vma:	the vm area in which the mapping is added
1142 * @address:	the user virtual address mapped
1143 * @compound:	charge the page as compound or small page
 
 
1144 *
1145 * Same as page_add_anon_rmap but must only be called on *new* pages.
1146 * This means the inc-and-test can be bypassed.
1147 * Page does not have to be locked.
1148 */
1149void page_add_new_anon_rmap(struct page *page,
1150	struct vm_area_struct *vma, unsigned long address, bool compound)
1151{
1152	int nr = compound ? hpage_nr_pages(page) : 1;
1153
1154	VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma);
1155	__SetPageSwapBacked(page);
1156	if (compound) {
1157		VM_BUG_ON_PAGE(!PageTransHuge(page), page);
1158		/* increment count (starts at -1) */
1159		atomic_set(compound_mapcount_ptr(page), 0);
1160		__inc_node_page_state(page, NR_ANON_THPS);
1161	} else {
1162		/* Anon THP always mapped first with PMD */
1163		VM_BUG_ON_PAGE(PageTransCompound(page), page);
1164		/* increment count (starts at -1) */
1165		atomic_set(&page->_mapcount, 0);
 
 
 
1166	}
1167	__mod_node_page_state(page_pgdat(page), NR_ANON_MAPPED, nr);
 
1168	__page_set_anon_rmap(page, vma, address, 1);
1169}
1170
1171/**
1172 * page_add_file_rmap - add pte mapping to a file page
1173 * @page: the page to add the mapping to
1174 * @compound: charge the page as compound or small page
 
1175 *
1176 * The caller needs to hold the pte lock.
1177 */
1178void page_add_file_rmap(struct page *page, bool compound)
 
1179{
1180	int i, nr = 1;
 
 
1181
1182	VM_BUG_ON_PAGE(compound && !PageTransHuge(page), page);
1183	lock_page_memcg(page);
1184	if (compound && PageTransHuge(page)) {
1185		for (i = 0, nr = 0; i < HPAGE_PMD_NR; i++) {
1186			if (atomic_inc_and_test(&page[i]._mapcount))
1187				nr++;
1188		}
1189		if (!atomic_inc_and_test(compound_mapcount_ptr(page)))
1190			goto out;
1191		VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
1192		__inc_node_page_state(page, NR_SHMEM_PMDMAPPED);
1193	} else {
1194		if (PageTransCompound(page) && page_mapping(page)) {
1195			VM_WARN_ON_ONCE(!PageLocked(page));
1196
1197			SetPageDoubleMap(compound_head(page));
1198			if (PageMlocked(page))
1199				clear_page_mlock(compound_head(page));
 
 
 
 
 
1200		}
1201		if (!atomic_inc_and_test(&page->_mapcount))
1202			goto out;
1203	}
1204	__mod_lruvec_page_state(page, NR_FILE_MAPPED, nr);
1205out:
1206	unlock_page_memcg(page);
1207}
1208
1209static void page_remove_file_rmap(struct page *page, bool compound)
1210{
1211	int i, nr = 1;
1212
1213	VM_BUG_ON_PAGE(compound && !PageHead(page), page);
1214	lock_page_memcg(page);
1215
1216	/* Hugepages are not counted in NR_FILE_MAPPED for now. */
1217	if (unlikely(PageHuge(page))) {
1218		/* hugetlb pages are always mapped with pmds */
1219		atomic_dec(compound_mapcount_ptr(page));
1220		goto out;
1221	}
1222
1223	/* page still mapped by someone else? */
1224	if (compound && PageTransHuge(page)) {
1225		for (i = 0, nr = 0; i < HPAGE_PMD_NR; i++) {
1226			if (atomic_add_negative(-1, &page[i]._mapcount))
1227				nr++;
1228		}
1229		if (!atomic_add_negative(-1, compound_mapcount_ptr(page)))
1230			goto out;
1231		VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
1232		__dec_node_page_state(page, NR_SHMEM_PMDMAPPED);
1233	} else {
1234		if (!atomic_add_negative(-1, &page->_mapcount))
1235			goto out;
1236	}
1237
1238	/*
1239	 * We use the irq-unsafe __{inc|mod}_lruvec_page_state because
1240	 * these counters are not modified in interrupt context, and
1241	 * pte lock(a spinlock) is held, which implies preemption disabled.
1242	 */
1243	__mod_lruvec_page_state(page, NR_FILE_MAPPED, -nr);
1244
1245	if (unlikely(PageMlocked(page)))
1246		clear_page_mlock(page);
1247out:
1248	unlock_page_memcg(page);
1249}
1250
1251static void page_remove_anon_compound_rmap(struct page *page)
1252{
1253	int i, nr;
1254
1255	if (!atomic_add_negative(-1, compound_mapcount_ptr(page)))
1256		return;
1257
1258	/* Hugepages are not counted in NR_ANON_PAGES for now. */
1259	if (unlikely(PageHuge(page)))
1260		return;
1261
1262	if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1263		return;
1264
1265	__dec_node_page_state(page, NR_ANON_THPS);
1266
1267	if (TestClearPageDoubleMap(page)) {
1268		/*
1269		 * Subpages can be mapped with PTEs too. Check how many of
1270		 * themi are still mapped.
1271		 */
1272		for (i = 0, nr = 0; i < HPAGE_PMD_NR; i++) {
1273			if (atomic_add_negative(-1, &page[i]._mapcount))
1274				nr++;
1275		}
1276	} else {
1277		nr = HPAGE_PMD_NR;
1278	}
1279
1280	if (unlikely(PageMlocked(page)))
1281		clear_page_mlock(page);
 
 
 
 
1282
1283	if (nr) {
1284		__mod_node_page_state(page_pgdat(page), NR_ANON_MAPPED, -nr);
1285		deferred_split_huge_page(page);
1286	}
1287}
1288
1289/**
1290 * page_remove_rmap - take down pte mapping from a page
1291 * @page:	page to remove mapping from
 
1292 * @compound:	uncharge the page as compound or small page
1293 *
1294 * The caller needs to hold the pte lock.
1295 */
1296void page_remove_rmap(struct page *page, bool compound)
 
1297{
1298	if (!PageAnon(page))
1299		return page_remove_file_rmap(page, compound);
 
1300
1301	if (compound)
1302		return page_remove_anon_compound_rmap(page);
1303
1304	/* page still mapped by someone else? */
1305	if (!atomic_add_negative(-1, &page->_mapcount))
 
 
1306		return;
 
1307
1308	/*
1309	 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1310	 * these counters are not modified in interrupt context, and
1311	 * pte lock(a spinlock) is held, which implies preemption disabled.
1312	 */
1313	__dec_node_page_state(page, NR_ANON_MAPPED);
1314
1315	if (unlikely(PageMlocked(page)))
1316		clear_page_mlock(page);
 
 
 
 
 
 
 
 
 
1317
1318	if (PageTransCompound(page))
1319		deferred_split_huge_page(compound_head(page));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1320
1321	/*
1322	 * It would be tidy to reset the PageAnon mapping here,
1323	 * but that might overwrite a racing page_add_anon_rmap
1324	 * which increments mapcount after us but sets mapping
1325	 * before us: so leave the reset to free_unref_page,
1326	 * and remember that it's only reliable while mapped.
1327	 * Leaving it set also helps swapoff to reinstate ptes
1328	 * faster for those pages still in swapcache.
1329	 */
 
 
 
 
1330}
1331
1332/*
1333 * @arg: enum ttu_flags will be passed to this argument
1334 */
1335static bool try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
1336		     unsigned long address, void *arg)
1337{
1338	struct mm_struct *mm = vma->vm_mm;
1339	struct page_vma_mapped_walk pvmw = {
1340		.page = page,
1341		.vma = vma,
1342		.address = address,
1343	};
1344	pte_t pteval;
1345	struct page *subpage;
1346	bool ret = true;
1347	unsigned long start = address, end;
1348	enum ttu_flags flags = (enum ttu_flags)arg;
1349
1350	/* munlock has nothing to gain from examining un-locked vmas */
1351	if ((flags & TTU_MUNLOCK) && !(vma->vm_flags & VM_LOCKED))
1352		return true;
1353
1354	if (IS_ENABLED(CONFIG_MIGRATION) && (flags & TTU_MIGRATION) &&
1355	    is_zone_device_page(page) && !is_device_private_page(page))
1356		return true;
 
 
 
 
 
1357
1358	if (flags & TTU_SPLIT_HUGE_PMD) {
1359		split_huge_pmd_address(vma, address,
1360				flags & TTU_SPLIT_FREEZE, page);
1361	}
1362
1363	/*
1364	 * We have to assume the worse case ie pmd for invalidation. Note that
1365	 * the page can not be free in this function as call of try_to_unmap()
1366	 * must hold a reference on the page.
 
 
 
1367	 */
1368	end = min(vma->vm_end, start + (PAGE_SIZE << compound_order(page)));
1369	mmu_notifier_invalidate_range_start(vma->vm_mm, start, end);
 
 
 
 
 
 
 
 
 
 
1370
1371	while (page_vma_mapped_walk(&pvmw)) {
1372#ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1373		/* PMD-mapped THP migration entry */
1374		if (!pvmw.pte && (flags & TTU_MIGRATION)) {
1375			VM_BUG_ON_PAGE(PageHuge(page) || !PageTransCompound(page), page);
1376
1377			set_pmd_migration_entry(&pvmw, page);
1378			continue;
1379		}
1380#endif
1381
1382		/*
1383		 * If the page is mlock()d, we cannot swap it out.
1384		 * If it's recently referenced (perhaps page_referenced
1385		 * skipped over this mm) then we should reactivate it.
1386		 */
1387		if (!(flags & TTU_IGNORE_MLOCK)) {
1388			if (vma->vm_flags & VM_LOCKED) {
1389				/* PTE-mapped THP are never mlocked */
1390				if (!PageTransCompound(page)) {
1391					/*
1392					 * Holding pte lock, we do *not* need
1393					 * mmap_sem here
1394					 */
1395					mlock_vma_page(page);
1396				}
1397				ret = false;
1398				page_vma_mapped_walk_done(&pvmw);
1399				break;
1400			}
1401			if (flags & TTU_MUNLOCK)
1402				continue;
1403		}
1404
1405		/* Unexpected PMD-mapped THP? */
1406		VM_BUG_ON_PAGE(!pvmw.pte, page);
1407
1408		subpage = page - page_to_pfn(page) + pte_pfn(*pvmw.pte);
1409		address = pvmw.address;
 
 
1410
1411
1412		if (IS_ENABLED(CONFIG_MIGRATION) &&
1413		    (flags & TTU_MIGRATION) &&
1414		    is_zone_device_page(page)) {
1415			swp_entry_t entry;
1416			pte_t swp_pte;
1417
1418			pteval = ptep_get_and_clear(mm, pvmw.address, pvmw.pte);
1419
1420			/*
1421			 * Store the pfn of the page in a special migration
1422			 * pte. do_swap_page() will wait until the migration
1423			 * pte is removed and then restart fault handling.
1424			 */
1425			entry = make_migration_entry(page, 0);
1426			swp_pte = swp_entry_to_pte(entry);
1427			if (pte_soft_dirty(pteval))
1428				swp_pte = pte_swp_mksoft_dirty(swp_pte);
1429			set_pte_at(mm, pvmw.address, pvmw.pte, swp_pte);
1430			/*
1431			 * No need to invalidate here it will synchronize on
1432			 * against the special swap migration pte.
 
 
 
1433			 */
1434			goto discard;
1435		}
1436
1437		if (!(flags & TTU_IGNORE_ACCESS)) {
1438			if (ptep_clear_flush_young_notify(vma, address,
1439						pvmw.pte)) {
1440				ret = false;
1441				page_vma_mapped_walk_done(&pvmw);
1442				break;
1443			}
1444		}
1445
1446		/* Nuke the page table entry. */
1447		flush_cache_page(vma, address, pte_pfn(*pvmw.pte));
1448		if (should_defer_flush(mm, flags)) {
1449			/*
1450			 * We clear the PTE but do not flush so potentially
1451			 * a remote CPU could still be writing to the page.
1452			 * If the entry was previously clean then the
1453			 * architecture must guarantee that a clear->dirty
1454			 * transition on a cached TLB entry is written through
1455			 * and traps if the PTE is unmapped.
 
 
1456			 */
1457			pteval = ptep_get_and_clear(mm, address, pvmw.pte);
1458
1459			set_tlb_ubc_flush_pending(mm, pte_dirty(pteval));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1460		} else {
1461			pteval = ptep_clear_flush(vma, address, pvmw.pte);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1462		}
1463
1464		/* Move the dirty bit to the page. Now the pte is gone. */
 
 
 
 
 
 
 
1465		if (pte_dirty(pteval))
1466			set_page_dirty(page);
1467
1468		/* Update high watermark before we lower rss */
1469		update_hiwater_rss(mm);
1470
1471		if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
1472			pteval = swp_entry_to_pte(make_hwpoison_entry(subpage));
1473			if (PageHuge(page)) {
1474				int nr = 1 << compound_order(page);
1475				hugetlb_count_sub(nr, mm);
1476				set_huge_swap_pte_at(mm, address,
1477						     pvmw.pte, pteval,
1478						     vma_mmu_pagesize(vma));
1479			} else {
1480				dec_mm_counter(mm, mm_counter(page));
1481				set_pte_at(mm, address, pvmw.pte, pteval);
1482			}
1483
1484		} else if (pte_unused(pteval)) {
1485			/*
1486			 * The guest indicated that the page content is of no
1487			 * interest anymore. Simply discard the pte, vmscan
1488			 * will take care of the rest.
 
 
 
 
 
1489			 */
1490			dec_mm_counter(mm, mm_counter(page));
1491			/* We have to invalidate as we cleared the pte */
1492			mmu_notifier_invalidate_range(mm, address,
1493						      address + PAGE_SIZE);
1494		} else if (IS_ENABLED(CONFIG_MIGRATION) &&
1495				(flags & (TTU_MIGRATION|TTU_SPLIT_FREEZE))) {
1496			swp_entry_t entry;
1497			pte_t swp_pte;
1498
1499			if (arch_unmap_one(mm, vma, address, pteval) < 0) {
1500				set_pte_at(mm, address, pvmw.pte, pteval);
1501				ret = false;
1502				page_vma_mapped_walk_done(&pvmw);
1503				break;
1504			}
1505
1506			/*
1507			 * Store the pfn of the page in a special migration
1508			 * pte. do_swap_page() will wait until the migration
1509			 * pte is removed and then restart fault handling.
1510			 */
1511			entry = make_migration_entry(subpage,
1512					pte_write(pteval));
1513			swp_pte = swp_entry_to_pte(entry);
1514			if (pte_soft_dirty(pteval))
1515				swp_pte = pte_swp_mksoft_dirty(swp_pte);
1516			set_pte_at(mm, address, pvmw.pte, swp_pte);
1517			/*
1518			 * No need to invalidate here it will synchronize on
1519			 * against the special swap migration pte.
1520			 */
1521		} else if (PageAnon(page)) {
1522			swp_entry_t entry = { .val = page_private(subpage) };
1523			pte_t swp_pte;
1524			/*
1525			 * Store the swap location in the pte.
1526			 * See handle_pte_fault() ...
1527			 */
1528			if (unlikely(PageSwapBacked(page) != PageSwapCache(page))) {
 
1529				WARN_ON_ONCE(1);
1530				ret = false;
1531				/* We have to invalidate as we cleared the pte */
1532				mmu_notifier_invalidate_range(mm, address,
1533							address + PAGE_SIZE);
1534				page_vma_mapped_walk_done(&pvmw);
1535				break;
1536			}
1537
1538			/* MADV_FREE page check */
1539			if (!PageSwapBacked(page)) {
1540				if (!PageDirty(page)) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1541					/* Invalidate as we cleared the pte */
1542					mmu_notifier_invalidate_range(mm,
1543						address, address + PAGE_SIZE);
1544					dec_mm_counter(mm, MM_ANONPAGES);
1545					goto discard;
1546				}
1547
1548				/*
1549				 * If the page was redirtied, it cannot be
1550				 * discarded. Remap the page to page table.
1551				 */
1552				set_pte_at(mm, address, pvmw.pte, pteval);
1553				SetPageSwapBacked(page);
1554				ret = false;
1555				page_vma_mapped_walk_done(&pvmw);
1556				break;
1557			}
1558
1559			if (swap_duplicate(entry) < 0) {
1560				set_pte_at(mm, address, pvmw.pte, pteval);
1561				ret = false;
1562				page_vma_mapped_walk_done(&pvmw);
1563				break;
1564			}
1565			if (arch_unmap_one(mm, vma, address, pteval) < 0) {
 
 
 
 
 
 
 
 
 
 
 
1566				set_pte_at(mm, address, pvmw.pte, pteval);
1567				ret = false;
1568				page_vma_mapped_walk_done(&pvmw);
1569				break;
1570			}
 
 
 
 
 
 
 
 
 
 
 
1571			if (list_empty(&mm->mmlist)) {
1572				spin_lock(&mmlist_lock);
1573				if (list_empty(&mm->mmlist))
1574					list_add(&mm->mmlist, &init_mm.mmlist);
1575				spin_unlock(&mmlist_lock);
1576			}
1577			dec_mm_counter(mm, MM_ANONPAGES);
1578			inc_mm_counter(mm, MM_SWAPENTS);
1579			swp_pte = swp_entry_to_pte(entry);
 
 
1580			if (pte_soft_dirty(pteval))
1581				swp_pte = pte_swp_mksoft_dirty(swp_pte);
 
 
1582			set_pte_at(mm, address, pvmw.pte, swp_pte);
1583			/* Invalidate as we cleared the pte */
1584			mmu_notifier_invalidate_range(mm, address,
1585						      address + PAGE_SIZE);
1586		} else {
1587			/*
1588			 * We should not need to notify here as we reach this
1589			 * case only from freeze_page() itself only call from
1590			 * split_huge_page_to_list() so everything below must
1591			 * be true:
1592			 *   - page is not anonymous
1593			 *   - page is locked
 
1594			 *
1595			 * So as it is a locked file back page thus it can not
1596			 * be remove from the page cache and replace by a new
1597			 * page before mmu_notifier_invalidate_range_end so no
1598			 * concurrent thread might update its page table to
1599			 * point at new page while a device still is using this
1600			 * page.
1601			 *
1602			 * See Documentation/vm/mmu_notifier.txt
1603			 */
1604			dec_mm_counter(mm, mm_counter_file(page));
1605		}
1606discard:
1607		/*
1608		 * No need to call mmu_notifier_invalidate_range() it has be
1609		 * done above for all cases requiring it to happen under page
1610		 * table lock before mmu_notifier_invalidate_range_end()
1611		 *
1612		 * See Documentation/vm/mmu_notifier.txt
1613		 */
1614		page_remove_rmap(subpage, PageHuge(page));
1615		put_page(page);
 
 
1616	}
1617
1618	mmu_notifier_invalidate_range_end(vma->vm_mm, start, end);
1619
1620	return ret;
1621}
1622
1623bool is_vma_temporary_stack(struct vm_area_struct *vma)
1624{
1625	int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP);
1626
1627	if (!maybe_stack)
1628		return false;
1629
1630	if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) ==
1631						VM_STACK_INCOMPLETE_SETUP)
1632		return true;
1633
1634	return false;
1635}
1636
1637static bool invalid_migration_vma(struct vm_area_struct *vma, void *arg)
1638{
1639	return is_vma_temporary_stack(vma);
1640}
1641
1642static int page_mapcount_is_zero(struct page *page)
1643{
1644	return !total_mapcount(page);
1645}
1646
1647/**
1648 * try_to_unmap - try to remove all page table mappings to a page
1649 * @page: the page to get unmapped
1650 * @flags: action and flags
1651 *
1652 * Tries to remove all the page table entries which are mapping this
1653 * page, used in the pageout path.  Caller must hold the page lock.
 
1654 *
1655 * If unmap is successful, return true. Otherwise, false.
1656 */
1657bool try_to_unmap(struct page *page, enum ttu_flags flags)
1658{
1659	struct rmap_walk_control rwc = {
1660		.rmap_one = try_to_unmap_one,
1661		.arg = (void *)flags,
1662		.done = page_mapcount_is_zero,
1663		.anon_lock = page_lock_anon_vma_read,
1664	};
1665
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1666	/*
1667	 * During exec, a temporary VMA is setup and later moved.
1668	 * The VMA is moved under the anon_vma lock but not the
1669	 * page tables leading to a race where migration cannot
1670	 * find the migration ptes. Rather than increasing the
1671	 * locking requirements of exec(), migration skips
1672	 * temporary VMAs until after exec() completes.
1673	 */
1674	if ((flags & (TTU_MIGRATION|TTU_SPLIT_FREEZE))
1675	    && !PageKsm(page) && PageAnon(page))
1676		rwc.invalid_vma = invalid_migration_vma;
1677
1678	if (flags & TTU_RMAP_LOCKED)
1679		rmap_walk_locked(page, &rwc);
1680	else
1681		rmap_walk(page, &rwc);
1682
1683	return !page_mapcount(page) ? true : false;
1684}
1685
1686static int page_not_mapped(struct page *page)
1687{
1688	return !page_mapped(page);
 
 
 
1689};
1690
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1691/**
1692 * try_to_munlock - try to munlock a page
1693 * @page: the page to be munlocked
 
 
 
 
 
 
 
1694 *
1695 * Called from munlock code.  Checks all of the VMAs mapping the page
1696 * to make sure nobody else has this page mlocked. The page will be
1697 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1698 */
1699
1700void try_to_munlock(struct page *page)
1701{
 
 
 
 
 
 
1702	struct rmap_walk_control rwc = {
1703		.rmap_one = try_to_unmap_one,
1704		.arg = (void *)TTU_MUNLOCK,
1705		.done = page_not_mapped,
1706		.anon_lock = page_lock_anon_vma_read,
1707
1708	};
1709
1710	VM_BUG_ON_PAGE(!PageLocked(page) || PageLRU(page), page);
1711	VM_BUG_ON_PAGE(PageCompound(page) && PageDoubleMap(page), page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1712
1713	rmap_walk(page, &rwc);
1714}
 
 
1715
1716void __put_anon_vma(struct anon_vma *anon_vma)
1717{
1718	struct anon_vma *root = anon_vma->root;
1719
1720	anon_vma_free(anon_vma);
1721	if (root != anon_vma && atomic_dec_and_test(&root->refcount))
1722		anon_vma_free(root);
1723}
1724
1725static struct anon_vma *rmap_walk_anon_lock(struct page *page,
1726					struct rmap_walk_control *rwc)
1727{
1728	struct anon_vma *anon_vma;
1729
1730	if (rwc->anon_lock)
1731		return rwc->anon_lock(page);
1732
1733	/*
1734	 * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read()
1735	 * because that depends on page_mapped(); but not all its usages
1736	 * are holding mmap_sem. Users without mmap_sem are required to
1737	 * take a reference count to prevent the anon_vma disappearing
1738	 */
1739	anon_vma = page_anon_vma(page);
1740	if (!anon_vma)
1741		return NULL;
1742
 
 
 
 
 
 
 
 
 
1743	anon_vma_lock_read(anon_vma);
 
1744	return anon_vma;
1745}
1746
1747/*
1748 * rmap_walk_anon - do something to anonymous page using the object-based
1749 * rmap method
1750 * @page: the page to be handled
1751 * @rwc: control variable according to each walk type
1752 *
1753 * Find all the mappings of a page using the mapping pointer and the vma chains
1754 * contained in the anon_vma struct it points to.
1755 *
1756 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1757 * where the page was found will be held for write.  So, we won't recheck
1758 * vm_flags for that VMA.  That should be OK, because that vma shouldn't be
1759 * LOCKED.
1760 */
1761static void rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc,
1762		bool locked)
1763{
1764	struct anon_vma *anon_vma;
1765	pgoff_t pgoff_start, pgoff_end;
1766	struct anon_vma_chain *avc;
1767
1768	if (locked) {
1769		anon_vma = page_anon_vma(page);
1770		/* anon_vma disappear under us? */
1771		VM_BUG_ON_PAGE(!anon_vma, page);
1772	} else {
1773		anon_vma = rmap_walk_anon_lock(page, rwc);
1774	}
1775	if (!anon_vma)
1776		return;
1777
1778	pgoff_start = page_to_pgoff(page);
1779	pgoff_end = pgoff_start + hpage_nr_pages(page) - 1;
1780	anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root,
1781			pgoff_start, pgoff_end) {
1782		struct vm_area_struct *vma = avc->vma;
1783		unsigned long address = vma_address(page, vma);
1784
 
1785		cond_resched();
1786
1787		if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
1788			continue;
1789
1790		if (!rwc->rmap_one(page, vma, address, rwc->arg))
1791			break;
1792		if (rwc->done && rwc->done(page))
1793			break;
1794	}
1795
1796	if (!locked)
1797		anon_vma_unlock_read(anon_vma);
1798}
1799
1800/*
1801 * rmap_walk_file - do something to file page using the object-based rmap method
1802 * @page: the page to be handled
1803 * @rwc: control variable according to each walk type
1804 *
1805 * Find all the mappings of a page using the mapping pointer and the vma chains
1806 * contained in the address_space struct it points to.
1807 *
1808 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1809 * where the page was found will be held for write.  So, we won't recheck
1810 * vm_flags for that VMA.  That should be OK, because that vma shouldn't be
1811 * LOCKED.
1812 */
1813static void rmap_walk_file(struct page *page, struct rmap_walk_control *rwc,
1814		bool locked)
1815{
1816	struct address_space *mapping = page_mapping(page);
1817	pgoff_t pgoff_start, pgoff_end;
1818	struct vm_area_struct *vma;
1819
1820	/*
1821	 * The page lock not only makes sure that page->mapping cannot
1822	 * suddenly be NULLified by truncation, it makes sure that the
1823	 * structure at mapping cannot be freed and reused yet,
1824	 * so we can safely take mapping->i_mmap_rwsem.
1825	 */
1826	VM_BUG_ON_PAGE(!PageLocked(page), page);
1827
1828	if (!mapping)
1829		return;
1830
1831	pgoff_start = page_to_pgoff(page);
1832	pgoff_end = pgoff_start + hpage_nr_pages(page) - 1;
1833	if (!locked)
 
 
 
 
 
 
 
 
1834		i_mmap_lock_read(mapping);
 
 
1835	vma_interval_tree_foreach(vma, &mapping->i_mmap,
1836			pgoff_start, pgoff_end) {
1837		unsigned long address = vma_address(page, vma);
1838
 
1839		cond_resched();
1840
1841		if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
1842			continue;
1843
1844		if (!rwc->rmap_one(page, vma, address, rwc->arg))
1845			goto done;
1846		if (rwc->done && rwc->done(page))
1847			goto done;
1848	}
1849
1850done:
1851	if (!locked)
1852		i_mmap_unlock_read(mapping);
1853}
1854
1855void rmap_walk(struct page *page, struct rmap_walk_control *rwc)
1856{
1857	if (unlikely(PageKsm(page)))
1858		rmap_walk_ksm(page, rwc);
1859	else if (PageAnon(page))
1860		rmap_walk_anon(page, rwc, false);
1861	else
1862		rmap_walk_file(page, rwc, false);
1863}
1864
1865/* Like rmap_walk, but caller holds relevant rmap lock */
1866void rmap_walk_locked(struct page *page, struct rmap_walk_control *rwc)
1867{
1868	/* no ksm support for now */
1869	VM_BUG_ON_PAGE(PageKsm(page), page);
1870	if (PageAnon(page))
1871		rmap_walk_anon(page, rwc, true);
1872	else
1873		rmap_walk_file(page, rwc, true);
1874}
1875
1876#ifdef CONFIG_HUGETLB_PAGE
1877/*
1878 * The following three functions are for anonymous (private mapped) hugepages.
1879 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1880 * and no lru code, because we handle hugepages differently from common pages.
 
 
1881 */
1882static void __hugepage_set_anon_rmap(struct page *page,
1883	struct vm_area_struct *vma, unsigned long address, int exclusive)
1884{
1885	struct anon_vma *anon_vma = vma->anon_vma;
1886
1887	BUG_ON(!anon_vma);
1888
1889	if (PageAnon(page))
1890		return;
1891	if (!exclusive)
1892		anon_vma = anon_vma->root;
1893
1894	anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1895	page->mapping = (struct address_space *) anon_vma;
1896	page->index = linear_page_index(vma, address);
1897}
1898
1899void hugepage_add_anon_rmap(struct page *page,
1900			    struct vm_area_struct *vma, unsigned long address)
1901{
1902	struct anon_vma *anon_vma = vma->anon_vma;
1903	int first;
1904
1905	BUG_ON(!PageLocked(page));
1906	BUG_ON(!anon_vma);
1907	/* address might be in next vma when migration races vma_adjust */
1908	first = atomic_inc_and_test(compound_mapcount_ptr(page));
 
 
1909	if (first)
1910		__hugepage_set_anon_rmap(page, vma, address, 0);
 
1911}
1912
1913void hugepage_add_new_anon_rmap(struct page *page,
1914			struct vm_area_struct *vma, unsigned long address)
1915{
1916	BUG_ON(address < vma->vm_start || address >= vma->vm_end);
 
1917	atomic_set(compound_mapcount_ptr(page), 0);
1918	__hugepage_set_anon_rmap(page, vma, address, 1);
 
1919}
1920#endif /* CONFIG_HUGETLB_PAGE */