11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * mm/rmap.c - physical to virtual reverse mappings 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright 2001, Rik van Riel <riel@conectiva.com.br> 51da177e4SLinus Torvalds * Released under the General Public License (GPL). 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * Simple, low overhead reverse mapping scheme. 81da177e4SLinus Torvalds * Please try to keep this thing as modular as possible. 91da177e4SLinus Torvalds * 101da177e4SLinus Torvalds * Provides methods for unmapping each kind of mapped page: 111da177e4SLinus Torvalds * the anon methods track anonymous pages, and 121da177e4SLinus Torvalds * the file methods track pages belonging to an inode. 131da177e4SLinus Torvalds * 141da177e4SLinus Torvalds * Original design by Rik van Riel <riel@conectiva.com.br> 2001 151da177e4SLinus Torvalds * File methods by Dave McCracken <dmccr@us.ibm.com> 2003, 2004 161da177e4SLinus Torvalds * Anonymous methods by Andrea Arcangeli <andrea@suse.de> 2004 1798f32602SHugh Dickins * Contributions by Hugh Dickins 2003, 2004 181da177e4SLinus Torvalds */ 191da177e4SLinus Torvalds 201da177e4SLinus Torvalds /* 211da177e4SLinus Torvalds * Lock ordering in mm: 221da177e4SLinus Torvalds * 231b1dcc1bSJes Sorensen * inode->i_mutex (while writing or truncating, not reading or faulting) 241da177e4SLinus Torvalds * mm->mmap_sem 251da177e4SLinus Torvalds * page->flags PG_locked (lock_page) 2688f306b6SKirill A. Shutemov * hugetlbfs_i_mmap_rwsem_key (in huge_pmd_share) 27c8c06efaSDavidlohr Bueso * mapping->i_mmap_rwsem 285a505085SIngo Molnar * anon_vma->rwsem 29b8072f09SHugh Dickins * mm->page_table_lock or pte_lock 30f4b7e272SAndrey Ryabinin * pgdat->lru_lock (in mark_page_accessed, isolate_lru_page) 315d337b91SHugh Dickins * swap_lock (in swap_duplicate, swap_info_get) 321da177e4SLinus Torvalds * mmlist_lock (in mmput, drain_mmlist and others) 331da177e4SLinus Torvalds * mapping->private_lock (in __set_page_dirty_buffers) 34c4843a75SGreg Thelen * mem_cgroup_{begin,end}_page_stat (memcg->move_lock) 35b93b0163SMatthew Wilcox * i_pages lock (widely used) 36250df6edSDave Chinner * inode->i_lock (in set_page_dirty's __mark_inode_dirty) 37f758eeabSChristoph Hellwig * bdi.wb->list_lock (in set_page_dirty's __mark_inode_dirty) 381da177e4SLinus Torvalds * sb_lock (within inode_lock in fs/fs-writeback.c) 39b93b0163SMatthew Wilcox * i_pages lock (widely used, in set_page_dirty, 401da177e4SLinus Torvalds * in arch-dependent flush_dcache_mmap_lock, 41f758eeabSChristoph Hellwig * within bdi.wb->list_lock in __sync_single_inode) 426a46079cSAndi Kleen * 435a505085SIngo Molnar * anon_vma->rwsem,mapping->i_mutex (memory_failure, collect_procs_anon) 446a46079cSAndi Kleen * ->tasklist_lock 456a46079cSAndi Kleen * pte map lock 461da177e4SLinus Torvalds */ 471da177e4SLinus Torvalds 481da177e4SLinus Torvalds #include <linux/mm.h> 496e84f315SIngo Molnar #include <linux/sched/mm.h> 5029930025SIngo Molnar #include <linux/sched/task.h> 511da177e4SLinus Torvalds #include <linux/pagemap.h> 521da177e4SLinus Torvalds #include <linux/swap.h> 531da177e4SLinus Torvalds #include <linux/swapops.h> 541da177e4SLinus Torvalds #include <linux/slab.h> 551da177e4SLinus Torvalds #include <linux/init.h> 565ad64688SHugh Dickins #include <linux/ksm.h> 571da177e4SLinus Torvalds #include <linux/rmap.h> 581da177e4SLinus Torvalds #include <linux/rcupdate.h> 59b95f1b31SPaul Gortmaker #include <linux/export.h> 608a9f3ccdSBalbir Singh #include <linux/memcontrol.h> 61cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 6264cdd548SKOSAKI Motohiro #include <linux/migrate.h> 630fe6e20bSNaoya Horiguchi #include <linux/hugetlb.h> 64ef5d437fSJan Kara #include <linux/backing-dev.h> 6533c3fc71SVladimir Davydov #include <linux/page_idle.h> 66a5430ddaSJérôme Glisse #include <linux/memremap.h> 67bce73e48SChristian Borntraeger #include <linux/userfaultfd_k.h> 681da177e4SLinus Torvalds 691da177e4SLinus Torvalds #include <asm/tlbflush.h> 701da177e4SLinus Torvalds 7172b252aeSMel Gorman #include <trace/events/tlb.h> 7272b252aeSMel Gorman 73b291f000SNick Piggin #include "internal.h" 74b291f000SNick Piggin 75fdd2e5f8SAdrian Bunk static struct kmem_cache *anon_vma_cachep; 765beb4930SRik van Riel static struct kmem_cache *anon_vma_chain_cachep; 77fdd2e5f8SAdrian Bunk 78fdd2e5f8SAdrian Bunk static inline struct anon_vma *anon_vma_alloc(void) 79fdd2e5f8SAdrian Bunk { 8001d8b20dSPeter Zijlstra struct anon_vma *anon_vma; 8101d8b20dSPeter Zijlstra 8201d8b20dSPeter Zijlstra anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL); 8301d8b20dSPeter Zijlstra if (anon_vma) { 8401d8b20dSPeter Zijlstra atomic_set(&anon_vma->refcount, 1); 857a3ef208SKonstantin Khlebnikov anon_vma->degree = 1; /* Reference for first vma */ 867a3ef208SKonstantin Khlebnikov anon_vma->parent = anon_vma; 8701d8b20dSPeter Zijlstra /* 8801d8b20dSPeter Zijlstra * Initialise the anon_vma root to point to itself. If called 8901d8b20dSPeter Zijlstra * from fork, the root will be reset to the parents anon_vma. 9001d8b20dSPeter Zijlstra */ 9101d8b20dSPeter Zijlstra anon_vma->root = anon_vma; 92fdd2e5f8SAdrian Bunk } 93fdd2e5f8SAdrian Bunk 9401d8b20dSPeter Zijlstra return anon_vma; 9501d8b20dSPeter Zijlstra } 9601d8b20dSPeter Zijlstra 9701d8b20dSPeter Zijlstra static inline void anon_vma_free(struct anon_vma *anon_vma) 98fdd2e5f8SAdrian Bunk { 9901d8b20dSPeter Zijlstra VM_BUG_ON(atomic_read(&anon_vma->refcount)); 10088c22088SPeter Zijlstra 10188c22088SPeter Zijlstra /* 1024fc3f1d6SIngo Molnar * Synchronize against page_lock_anon_vma_read() such that 10388c22088SPeter Zijlstra * we can safely hold the lock without the anon_vma getting 10488c22088SPeter Zijlstra * freed. 10588c22088SPeter Zijlstra * 10688c22088SPeter Zijlstra * Relies on the full mb implied by the atomic_dec_and_test() from 10788c22088SPeter Zijlstra * put_anon_vma() against the acquire barrier implied by 1084fc3f1d6SIngo Molnar * down_read_trylock() from page_lock_anon_vma_read(). This orders: 10988c22088SPeter Zijlstra * 1104fc3f1d6SIngo Molnar * page_lock_anon_vma_read() VS put_anon_vma() 1114fc3f1d6SIngo Molnar * down_read_trylock() atomic_dec_and_test() 11288c22088SPeter Zijlstra * LOCK MB 1134fc3f1d6SIngo Molnar * atomic_read() rwsem_is_locked() 11488c22088SPeter Zijlstra * 11588c22088SPeter Zijlstra * LOCK should suffice since the actual taking of the lock must 11688c22088SPeter Zijlstra * happen _before_ what follows. 11788c22088SPeter Zijlstra */ 1187f39dda9SHugh Dickins might_sleep(); 1195a505085SIngo Molnar if (rwsem_is_locked(&anon_vma->root->rwsem)) { 1204fc3f1d6SIngo Molnar anon_vma_lock_write(anon_vma); 12108b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 12288c22088SPeter Zijlstra } 12388c22088SPeter Zijlstra 124fdd2e5f8SAdrian Bunk kmem_cache_free(anon_vma_cachep, anon_vma); 125fdd2e5f8SAdrian Bunk } 1261da177e4SLinus Torvalds 127dd34739cSLinus Torvalds static inline struct anon_vma_chain *anon_vma_chain_alloc(gfp_t gfp) 1285beb4930SRik van Riel { 129dd34739cSLinus Torvalds return kmem_cache_alloc(anon_vma_chain_cachep, gfp); 1305beb4930SRik van Riel } 1315beb4930SRik van Riel 132e574b5fdSNamhyung Kim static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain) 1335beb4930SRik van Riel { 1345beb4930SRik van Riel kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain); 1355beb4930SRik van Riel } 1365beb4930SRik van Riel 1376583a843SKautuk Consul static void anon_vma_chain_link(struct vm_area_struct *vma, 1386583a843SKautuk Consul struct anon_vma_chain *avc, 1396583a843SKautuk Consul struct anon_vma *anon_vma) 1406583a843SKautuk Consul { 1416583a843SKautuk Consul avc->vma = vma; 1426583a843SKautuk Consul avc->anon_vma = anon_vma; 1436583a843SKautuk Consul list_add(&avc->same_vma, &vma->anon_vma_chain); 144bf181b9fSMichel Lespinasse anon_vma_interval_tree_insert(avc, &anon_vma->rb_root); 1456583a843SKautuk Consul } 1466583a843SKautuk Consul 147d9d332e0SLinus Torvalds /** 148d5a187daSVlastimil Babka * __anon_vma_prepare - attach an anon_vma to a memory region 149d9d332e0SLinus Torvalds * @vma: the memory region in question 150d9d332e0SLinus Torvalds * 151d9d332e0SLinus Torvalds * This makes sure the memory mapping described by 'vma' has 152d9d332e0SLinus Torvalds * an 'anon_vma' attached to it, so that we can associate the 153d9d332e0SLinus Torvalds * anonymous pages mapped into it with that anon_vma. 154d9d332e0SLinus Torvalds * 155d5a187daSVlastimil Babka * The common case will be that we already have one, which 156d5a187daSVlastimil Babka * is handled inline by anon_vma_prepare(). But if 15723a0790aSFigo.zhang * not we either need to find an adjacent mapping that we 158d9d332e0SLinus Torvalds * can re-use the anon_vma from (very common when the only 159d9d332e0SLinus Torvalds * reason for splitting a vma has been mprotect()), or we 160d9d332e0SLinus Torvalds * allocate a new one. 161d9d332e0SLinus Torvalds * 162d9d332e0SLinus Torvalds * Anon-vma allocations are very subtle, because we may have 1634fc3f1d6SIngo Molnar * optimistically looked up an anon_vma in page_lock_anon_vma_read() 164d9d332e0SLinus Torvalds * and that may actually touch the spinlock even in the newly 165d9d332e0SLinus Torvalds * allocated vma (it depends on RCU to make sure that the 166d9d332e0SLinus Torvalds * anon_vma isn't actually destroyed). 167d9d332e0SLinus Torvalds * 168d9d332e0SLinus Torvalds * As a result, we need to do proper anon_vma locking even 169d9d332e0SLinus Torvalds * for the new allocation. At the same time, we do not want 170d9d332e0SLinus Torvalds * to do any locking for the common case of already having 171d9d332e0SLinus Torvalds * an anon_vma. 172d9d332e0SLinus Torvalds * 173d9d332e0SLinus Torvalds * This must be called with the mmap_sem held for reading. 174d9d332e0SLinus Torvalds */ 175d5a187daSVlastimil Babka int __anon_vma_prepare(struct vm_area_struct *vma) 1761da177e4SLinus Torvalds { 177d5a187daSVlastimil Babka struct mm_struct *mm = vma->vm_mm; 178d5a187daSVlastimil Babka struct anon_vma *anon_vma, *allocated; 1795beb4930SRik van Riel struct anon_vma_chain *avc; 1801da177e4SLinus Torvalds 1811da177e4SLinus Torvalds might_sleep(); 1821da177e4SLinus Torvalds 183dd34739cSLinus Torvalds avc = anon_vma_chain_alloc(GFP_KERNEL); 1845beb4930SRik van Riel if (!avc) 1855beb4930SRik van Riel goto out_enomem; 1865beb4930SRik van Riel 1871da177e4SLinus Torvalds anon_vma = find_mergeable_anon_vma(vma); 1881da177e4SLinus Torvalds allocated = NULL; 189d9d332e0SLinus Torvalds if (!anon_vma) { 1901da177e4SLinus Torvalds anon_vma = anon_vma_alloc(); 1911da177e4SLinus Torvalds if (unlikely(!anon_vma)) 1925beb4930SRik van Riel goto out_enomem_free_avc; 1931da177e4SLinus Torvalds allocated = anon_vma; 1941da177e4SLinus Torvalds } 1951da177e4SLinus Torvalds 1964fc3f1d6SIngo Molnar anon_vma_lock_write(anon_vma); 1971da177e4SLinus Torvalds /* page_table_lock to protect against threads */ 1981da177e4SLinus Torvalds spin_lock(&mm->page_table_lock); 1991da177e4SLinus Torvalds if (likely(!vma->anon_vma)) { 2001da177e4SLinus Torvalds vma->anon_vma = anon_vma; 2016583a843SKautuk Consul anon_vma_chain_link(vma, avc, anon_vma); 2027a3ef208SKonstantin Khlebnikov /* vma reference or self-parent link for new root */ 2037a3ef208SKonstantin Khlebnikov anon_vma->degree++; 2041da177e4SLinus Torvalds allocated = NULL; 20531f2b0ebSOleg Nesterov avc = NULL; 2061da177e4SLinus Torvalds } 2071da177e4SLinus Torvalds spin_unlock(&mm->page_table_lock); 20808b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 20931f2b0ebSOleg Nesterov 21031f2b0ebSOleg Nesterov if (unlikely(allocated)) 21101d8b20dSPeter Zijlstra put_anon_vma(allocated); 21231f2b0ebSOleg Nesterov if (unlikely(avc)) 2135beb4930SRik van Riel anon_vma_chain_free(avc); 214d5a187daSVlastimil Babka 2151da177e4SLinus Torvalds return 0; 2165beb4930SRik van Riel 2175beb4930SRik van Riel out_enomem_free_avc: 2185beb4930SRik van Riel anon_vma_chain_free(avc); 2195beb4930SRik van Riel out_enomem: 2205beb4930SRik van Riel return -ENOMEM; 2211da177e4SLinus Torvalds } 2221da177e4SLinus Torvalds 223bb4aa396SLinus Torvalds /* 224bb4aa396SLinus Torvalds * This is a useful helper function for locking the anon_vma root as 225bb4aa396SLinus Torvalds * we traverse the vma->anon_vma_chain, looping over anon_vma's that 226bb4aa396SLinus Torvalds * have the same vma. 227bb4aa396SLinus Torvalds * 228bb4aa396SLinus Torvalds * Such anon_vma's should have the same root, so you'd expect to see 229bb4aa396SLinus Torvalds * just a single mutex_lock for the whole traversal. 230bb4aa396SLinus Torvalds */ 231bb4aa396SLinus Torvalds static inline struct anon_vma *lock_anon_vma_root(struct anon_vma *root, struct anon_vma *anon_vma) 232bb4aa396SLinus Torvalds { 233bb4aa396SLinus Torvalds struct anon_vma *new_root = anon_vma->root; 234bb4aa396SLinus Torvalds if (new_root != root) { 235bb4aa396SLinus Torvalds if (WARN_ON_ONCE(root)) 2365a505085SIngo Molnar up_write(&root->rwsem); 237bb4aa396SLinus Torvalds root = new_root; 2385a505085SIngo Molnar down_write(&root->rwsem); 239bb4aa396SLinus Torvalds } 240bb4aa396SLinus Torvalds return root; 241bb4aa396SLinus Torvalds } 242bb4aa396SLinus Torvalds 243bb4aa396SLinus Torvalds static inline void unlock_anon_vma_root(struct anon_vma *root) 244bb4aa396SLinus Torvalds { 245bb4aa396SLinus Torvalds if (root) 2465a505085SIngo Molnar up_write(&root->rwsem); 247bb4aa396SLinus Torvalds } 248bb4aa396SLinus Torvalds 2495beb4930SRik van Riel /* 2505beb4930SRik van Riel * Attach the anon_vmas from src to dst. 2515beb4930SRik van Riel * Returns 0 on success, -ENOMEM on failure. 2527a3ef208SKonstantin Khlebnikov * 2537a3ef208SKonstantin Khlebnikov * If dst->anon_vma is NULL this function tries to find and reuse existing 2547a3ef208SKonstantin Khlebnikov * anon_vma which has no vmas and only one child anon_vma. This prevents 2557a3ef208SKonstantin Khlebnikov * degradation of anon_vma hierarchy to endless linear chain in case of 2567a3ef208SKonstantin Khlebnikov * constantly forking task. On the other hand, an anon_vma with more than one 2577a3ef208SKonstantin Khlebnikov * child isn't reused even if there was no alive vma, thus rmap walker has a 2587a3ef208SKonstantin Khlebnikov * good chance of avoiding scanning the whole hierarchy when it searches where 2597a3ef208SKonstantin Khlebnikov * page is mapped. 2605beb4930SRik van Riel */ 2615beb4930SRik van Riel int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src) 2625beb4930SRik van Riel { 2635beb4930SRik van Riel struct anon_vma_chain *avc, *pavc; 264bb4aa396SLinus Torvalds struct anon_vma *root = NULL; 2655beb4930SRik van Riel 266646d87b4SLinus Torvalds list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) { 267bb4aa396SLinus Torvalds struct anon_vma *anon_vma; 268bb4aa396SLinus Torvalds 269dd34739cSLinus Torvalds avc = anon_vma_chain_alloc(GFP_NOWAIT | __GFP_NOWARN); 270dd34739cSLinus Torvalds if (unlikely(!avc)) { 271dd34739cSLinus Torvalds unlock_anon_vma_root(root); 272dd34739cSLinus Torvalds root = NULL; 273dd34739cSLinus Torvalds avc = anon_vma_chain_alloc(GFP_KERNEL); 2745beb4930SRik van Riel if (!avc) 2755beb4930SRik van Riel goto enomem_failure; 276dd34739cSLinus Torvalds } 277bb4aa396SLinus Torvalds anon_vma = pavc->anon_vma; 278bb4aa396SLinus Torvalds root = lock_anon_vma_root(root, anon_vma); 279bb4aa396SLinus Torvalds anon_vma_chain_link(dst, avc, anon_vma); 2807a3ef208SKonstantin Khlebnikov 2817a3ef208SKonstantin Khlebnikov /* 2827a3ef208SKonstantin Khlebnikov * Reuse existing anon_vma if its degree lower than two, 2837a3ef208SKonstantin Khlebnikov * that means it has no vma and only one anon_vma child. 2847a3ef208SKonstantin Khlebnikov * 2857a3ef208SKonstantin Khlebnikov * Do not chose parent anon_vma, otherwise first child 2867a3ef208SKonstantin Khlebnikov * will always reuse it. Root anon_vma is never reused: 2877a3ef208SKonstantin Khlebnikov * it has self-parent reference and at least one child. 2887a3ef208SKonstantin Khlebnikov */ 2897a3ef208SKonstantin Khlebnikov if (!dst->anon_vma && anon_vma != src->anon_vma && 2907a3ef208SKonstantin Khlebnikov anon_vma->degree < 2) 2917a3ef208SKonstantin Khlebnikov dst->anon_vma = anon_vma; 2925beb4930SRik van Riel } 2937a3ef208SKonstantin Khlebnikov if (dst->anon_vma) 2947a3ef208SKonstantin Khlebnikov dst->anon_vma->degree++; 295bb4aa396SLinus Torvalds unlock_anon_vma_root(root); 2965beb4930SRik van Riel return 0; 2975beb4930SRik van Riel 2985beb4930SRik van Riel enomem_failure: 2993fe89b3eSLeon Yu /* 3003fe89b3eSLeon Yu * dst->anon_vma is dropped here otherwise its degree can be incorrectly 3013fe89b3eSLeon Yu * decremented in unlink_anon_vmas(). 3023fe89b3eSLeon Yu * We can safely do this because callers of anon_vma_clone() don't care 3033fe89b3eSLeon Yu * about dst->anon_vma if anon_vma_clone() failed. 3043fe89b3eSLeon Yu */ 3053fe89b3eSLeon Yu dst->anon_vma = NULL; 3065beb4930SRik van Riel unlink_anon_vmas(dst); 3075beb4930SRik van Riel return -ENOMEM; 3081da177e4SLinus Torvalds } 3091da177e4SLinus Torvalds 3105beb4930SRik van Riel /* 3115beb4930SRik van Riel * Attach vma to its own anon_vma, as well as to the anon_vmas that 3125beb4930SRik van Riel * the corresponding VMA in the parent process is attached to. 3135beb4930SRik van Riel * Returns 0 on success, non-zero on failure. 3145beb4930SRik van Riel */ 3155beb4930SRik van Riel int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma) 3161da177e4SLinus Torvalds { 3175beb4930SRik van Riel struct anon_vma_chain *avc; 3185beb4930SRik van Riel struct anon_vma *anon_vma; 319c4ea95d7SDaniel Forrest int error; 3205beb4930SRik van Riel 3215beb4930SRik van Riel /* Don't bother if the parent process has no anon_vma here. */ 3225beb4930SRik van Riel if (!pvma->anon_vma) 3235beb4930SRik van Riel return 0; 3245beb4930SRik van Riel 3257a3ef208SKonstantin Khlebnikov /* Drop inherited anon_vma, we'll reuse existing or allocate new. */ 3267a3ef208SKonstantin Khlebnikov vma->anon_vma = NULL; 3277a3ef208SKonstantin Khlebnikov 3285beb4930SRik van Riel /* 3295beb4930SRik van Riel * First, attach the new VMA to the parent VMA's anon_vmas, 3305beb4930SRik van Riel * so rmap can find non-COWed pages in child processes. 3315beb4930SRik van Riel */ 332c4ea95d7SDaniel Forrest error = anon_vma_clone(vma, pvma); 333c4ea95d7SDaniel Forrest if (error) 334c4ea95d7SDaniel Forrest return error; 3355beb4930SRik van Riel 3367a3ef208SKonstantin Khlebnikov /* An existing anon_vma has been reused, all done then. */ 3377a3ef208SKonstantin Khlebnikov if (vma->anon_vma) 3387a3ef208SKonstantin Khlebnikov return 0; 3397a3ef208SKonstantin Khlebnikov 3405beb4930SRik van Riel /* Then add our own anon_vma. */ 3415beb4930SRik van Riel anon_vma = anon_vma_alloc(); 3425beb4930SRik van Riel if (!anon_vma) 3435beb4930SRik van Riel goto out_error; 344dd34739cSLinus Torvalds avc = anon_vma_chain_alloc(GFP_KERNEL); 3455beb4930SRik van Riel if (!avc) 3465beb4930SRik van Riel goto out_error_free_anon_vma; 3475c341ee1SRik van Riel 3485c341ee1SRik van Riel /* 3495c341ee1SRik van Riel * The root anon_vma's spinlock is the lock actually used when we 3505c341ee1SRik van Riel * lock any of the anon_vmas in this anon_vma tree. 3515c341ee1SRik van Riel */ 3525c341ee1SRik van Riel anon_vma->root = pvma->anon_vma->root; 3537a3ef208SKonstantin Khlebnikov anon_vma->parent = pvma->anon_vma; 35476545066SRik van Riel /* 35501d8b20dSPeter Zijlstra * With refcounts, an anon_vma can stay around longer than the 35601d8b20dSPeter Zijlstra * process it belongs to. The root anon_vma needs to be pinned until 35701d8b20dSPeter Zijlstra * this anon_vma is freed, because the lock lives in the root. 35876545066SRik van Riel */ 35976545066SRik van Riel get_anon_vma(anon_vma->root); 3605beb4930SRik van Riel /* Mark this anon_vma as the one where our new (COWed) pages go. */ 3615beb4930SRik van Riel vma->anon_vma = anon_vma; 3624fc3f1d6SIngo Molnar anon_vma_lock_write(anon_vma); 3635c341ee1SRik van Riel anon_vma_chain_link(vma, avc, anon_vma); 3647a3ef208SKonstantin Khlebnikov anon_vma->parent->degree++; 36508b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 3665beb4930SRik van Riel 3675beb4930SRik van Riel return 0; 3685beb4930SRik van Riel 3695beb4930SRik van Riel out_error_free_anon_vma: 37001d8b20dSPeter Zijlstra put_anon_vma(anon_vma); 3715beb4930SRik van Riel out_error: 3724946d54cSRik van Riel unlink_anon_vmas(vma); 3735beb4930SRik van Riel return -ENOMEM; 3745beb4930SRik van Riel } 3755beb4930SRik van Riel 3765beb4930SRik van Riel void unlink_anon_vmas(struct vm_area_struct *vma) 3775beb4930SRik van Riel { 3785beb4930SRik van Riel struct anon_vma_chain *avc, *next; 379eee2acbaSPeter Zijlstra struct anon_vma *root = NULL; 3805beb4930SRik van Riel 3815c341ee1SRik van Riel /* 3825c341ee1SRik van Riel * Unlink each anon_vma chained to the VMA. This list is ordered 3835c341ee1SRik van Riel * from newest to oldest, ensuring the root anon_vma gets freed last. 3845c341ee1SRik van Riel */ 3855beb4930SRik van Riel list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { 386eee2acbaSPeter Zijlstra struct anon_vma *anon_vma = avc->anon_vma; 387eee2acbaSPeter Zijlstra 388eee2acbaSPeter Zijlstra root = lock_anon_vma_root(root, anon_vma); 389bf181b9fSMichel Lespinasse anon_vma_interval_tree_remove(avc, &anon_vma->rb_root); 390eee2acbaSPeter Zijlstra 391eee2acbaSPeter Zijlstra /* 392eee2acbaSPeter Zijlstra * Leave empty anon_vmas on the list - we'll need 393eee2acbaSPeter Zijlstra * to free them outside the lock. 394eee2acbaSPeter Zijlstra */ 395f808c13fSDavidlohr Bueso if (RB_EMPTY_ROOT(&anon_vma->rb_root.rb_root)) { 3967a3ef208SKonstantin Khlebnikov anon_vma->parent->degree--; 397eee2acbaSPeter Zijlstra continue; 3987a3ef208SKonstantin Khlebnikov } 399eee2acbaSPeter Zijlstra 400eee2acbaSPeter Zijlstra list_del(&avc->same_vma); 401eee2acbaSPeter Zijlstra anon_vma_chain_free(avc); 402eee2acbaSPeter Zijlstra } 4037a3ef208SKonstantin Khlebnikov if (vma->anon_vma) 4047a3ef208SKonstantin Khlebnikov vma->anon_vma->degree--; 405eee2acbaSPeter Zijlstra unlock_anon_vma_root(root); 406eee2acbaSPeter Zijlstra 407eee2acbaSPeter Zijlstra /* 408eee2acbaSPeter Zijlstra * Iterate the list once more, it now only contains empty and unlinked 409eee2acbaSPeter Zijlstra * anon_vmas, destroy them. Could not do before due to __put_anon_vma() 4105a505085SIngo Molnar * needing to write-acquire the anon_vma->root->rwsem. 411eee2acbaSPeter Zijlstra */ 412eee2acbaSPeter Zijlstra list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { 413eee2acbaSPeter Zijlstra struct anon_vma *anon_vma = avc->anon_vma; 414eee2acbaSPeter Zijlstra 415e4c5800aSKonstantin Khlebnikov VM_WARN_ON(anon_vma->degree); 416eee2acbaSPeter Zijlstra put_anon_vma(anon_vma); 417eee2acbaSPeter Zijlstra 4185beb4930SRik van Riel list_del(&avc->same_vma); 4195beb4930SRik van Riel anon_vma_chain_free(avc); 4205beb4930SRik van Riel } 4215beb4930SRik van Riel } 4225beb4930SRik van Riel 42351cc5068SAlexey Dobriyan static void anon_vma_ctor(void *data) 4241da177e4SLinus Torvalds { 4251da177e4SLinus Torvalds struct anon_vma *anon_vma = data; 4261da177e4SLinus Torvalds 4275a505085SIngo Molnar init_rwsem(&anon_vma->rwsem); 42883813267SPeter Zijlstra atomic_set(&anon_vma->refcount, 0); 429f808c13fSDavidlohr Bueso anon_vma->rb_root = RB_ROOT_CACHED; 4301da177e4SLinus Torvalds } 4311da177e4SLinus Torvalds 4321da177e4SLinus Torvalds void __init anon_vma_init(void) 4331da177e4SLinus Torvalds { 4341da177e4SLinus Torvalds anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), 4355f0d5a3aSPaul E. McKenney 0, SLAB_TYPESAFE_BY_RCU|SLAB_PANIC|SLAB_ACCOUNT, 4365d097056SVladimir Davydov anon_vma_ctor); 4375d097056SVladimir Davydov anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, 4385d097056SVladimir Davydov SLAB_PANIC|SLAB_ACCOUNT); 4391da177e4SLinus Torvalds } 4401da177e4SLinus Torvalds 4411da177e4SLinus Torvalds /* 4426111e4caSPeter Zijlstra * Getting a lock on a stable anon_vma from a page off the LRU is tricky! 4436111e4caSPeter Zijlstra * 4446111e4caSPeter Zijlstra * Since there is no serialization what so ever against page_remove_rmap() 4456111e4caSPeter Zijlstra * the best this function can do is return a locked anon_vma that might 4466111e4caSPeter Zijlstra * have been relevant to this page. 4476111e4caSPeter Zijlstra * 4486111e4caSPeter Zijlstra * The page might have been remapped to a different anon_vma or the anon_vma 4496111e4caSPeter Zijlstra * returned may already be freed (and even reused). 4506111e4caSPeter Zijlstra * 451bc658c96SPeter Zijlstra * In case it was remapped to a different anon_vma, the new anon_vma will be a 452bc658c96SPeter Zijlstra * child of the old anon_vma, and the anon_vma lifetime rules will therefore 453bc658c96SPeter Zijlstra * ensure that any anon_vma obtained from the page will still be valid for as 454bc658c96SPeter Zijlstra * long as we observe page_mapped() [ hence all those page_mapped() tests ]. 455bc658c96SPeter Zijlstra * 4566111e4caSPeter Zijlstra * All users of this function must be very careful when walking the anon_vma 4576111e4caSPeter Zijlstra * chain and verify that the page in question is indeed mapped in it 4586111e4caSPeter Zijlstra * [ something equivalent to page_mapped_in_vma() ]. 4596111e4caSPeter Zijlstra * 4606111e4caSPeter Zijlstra * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap() 4616111e4caSPeter Zijlstra * that the anon_vma pointer from page->mapping is valid if there is a 4626111e4caSPeter Zijlstra * mapcount, we can dereference the anon_vma after observing those. 4631da177e4SLinus Torvalds */ 464746b18d4SPeter Zijlstra struct anon_vma *page_get_anon_vma(struct page *page) 4651da177e4SLinus Torvalds { 466746b18d4SPeter Zijlstra struct anon_vma *anon_vma = NULL; 4671da177e4SLinus Torvalds unsigned long anon_mapping; 4681da177e4SLinus Torvalds 4691da177e4SLinus Torvalds rcu_read_lock(); 4704db0c3c2SJason Low anon_mapping = (unsigned long)READ_ONCE(page->mapping); 4713ca7b3c5SHugh Dickins if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) 4721da177e4SLinus Torvalds goto out; 4731da177e4SLinus Torvalds if (!page_mapped(page)) 4741da177e4SLinus Torvalds goto out; 4751da177e4SLinus Torvalds 4761da177e4SLinus Torvalds anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); 477746b18d4SPeter Zijlstra if (!atomic_inc_not_zero(&anon_vma->refcount)) { 478746b18d4SPeter Zijlstra anon_vma = NULL; 479746b18d4SPeter Zijlstra goto out; 480746b18d4SPeter Zijlstra } 481f1819427SHugh Dickins 482f1819427SHugh Dickins /* 483f1819427SHugh Dickins * If this page is still mapped, then its anon_vma cannot have been 484746b18d4SPeter Zijlstra * freed. But if it has been unmapped, we have no security against the 485746b18d4SPeter Zijlstra * anon_vma structure being freed and reused (for another anon_vma: 4865f0d5a3aSPaul E. McKenney * SLAB_TYPESAFE_BY_RCU guarantees that - so the atomic_inc_not_zero() 487746b18d4SPeter Zijlstra * above cannot corrupt). 488f1819427SHugh Dickins */ 489746b18d4SPeter Zijlstra if (!page_mapped(page)) { 4907f39dda9SHugh Dickins rcu_read_unlock(); 491746b18d4SPeter Zijlstra put_anon_vma(anon_vma); 4927f39dda9SHugh Dickins return NULL; 493746b18d4SPeter Zijlstra } 4941da177e4SLinus Torvalds out: 4951da177e4SLinus Torvalds rcu_read_unlock(); 496746b18d4SPeter Zijlstra 497746b18d4SPeter Zijlstra return anon_vma; 498746b18d4SPeter Zijlstra } 499746b18d4SPeter Zijlstra 50088c22088SPeter Zijlstra /* 50188c22088SPeter Zijlstra * Similar to page_get_anon_vma() except it locks the anon_vma. 50288c22088SPeter Zijlstra * 50388c22088SPeter Zijlstra * Its a little more complex as it tries to keep the fast path to a single 50488c22088SPeter Zijlstra * atomic op -- the trylock. If we fail the trylock, we fall back to getting a 50588c22088SPeter Zijlstra * reference like with page_get_anon_vma() and then block on the mutex. 50688c22088SPeter Zijlstra */ 5074fc3f1d6SIngo Molnar struct anon_vma *page_lock_anon_vma_read(struct page *page) 508746b18d4SPeter Zijlstra { 50988c22088SPeter Zijlstra struct anon_vma *anon_vma = NULL; 510eee0f252SHugh Dickins struct anon_vma *root_anon_vma; 51188c22088SPeter Zijlstra unsigned long anon_mapping; 512746b18d4SPeter Zijlstra 51388c22088SPeter Zijlstra rcu_read_lock(); 5144db0c3c2SJason Low anon_mapping = (unsigned long)READ_ONCE(page->mapping); 51588c22088SPeter Zijlstra if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) 51688c22088SPeter Zijlstra goto out; 51788c22088SPeter Zijlstra if (!page_mapped(page)) 51888c22088SPeter Zijlstra goto out; 51988c22088SPeter Zijlstra 52088c22088SPeter Zijlstra anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); 5214db0c3c2SJason Low root_anon_vma = READ_ONCE(anon_vma->root); 5224fc3f1d6SIngo Molnar if (down_read_trylock(&root_anon_vma->rwsem)) { 52388c22088SPeter Zijlstra /* 524eee0f252SHugh Dickins * If the page is still mapped, then this anon_vma is still 525eee0f252SHugh Dickins * its anon_vma, and holding the mutex ensures that it will 526bc658c96SPeter Zijlstra * not go away, see anon_vma_free(). 52788c22088SPeter Zijlstra */ 528eee0f252SHugh Dickins if (!page_mapped(page)) { 5294fc3f1d6SIngo Molnar up_read(&root_anon_vma->rwsem); 53088c22088SPeter Zijlstra anon_vma = NULL; 53188c22088SPeter Zijlstra } 53288c22088SPeter Zijlstra goto out; 53388c22088SPeter Zijlstra } 53488c22088SPeter Zijlstra 53588c22088SPeter Zijlstra /* trylock failed, we got to sleep */ 53688c22088SPeter Zijlstra if (!atomic_inc_not_zero(&anon_vma->refcount)) { 53788c22088SPeter Zijlstra anon_vma = NULL; 53888c22088SPeter Zijlstra goto out; 53988c22088SPeter Zijlstra } 54088c22088SPeter Zijlstra 54188c22088SPeter Zijlstra if (!page_mapped(page)) { 5427f39dda9SHugh Dickins rcu_read_unlock(); 54388c22088SPeter Zijlstra put_anon_vma(anon_vma); 5447f39dda9SHugh Dickins return NULL; 54588c22088SPeter Zijlstra } 54688c22088SPeter Zijlstra 54788c22088SPeter Zijlstra /* we pinned the anon_vma, its safe to sleep */ 54888c22088SPeter Zijlstra rcu_read_unlock(); 5494fc3f1d6SIngo Molnar anon_vma_lock_read(anon_vma); 550746b18d4SPeter Zijlstra 55188c22088SPeter Zijlstra if (atomic_dec_and_test(&anon_vma->refcount)) { 55288c22088SPeter Zijlstra /* 55388c22088SPeter Zijlstra * Oops, we held the last refcount, release the lock 55488c22088SPeter Zijlstra * and bail -- can't simply use put_anon_vma() because 5554fc3f1d6SIngo Molnar * we'll deadlock on the anon_vma_lock_write() recursion. 55688c22088SPeter Zijlstra */ 5574fc3f1d6SIngo Molnar anon_vma_unlock_read(anon_vma); 55888c22088SPeter Zijlstra __put_anon_vma(anon_vma); 55988c22088SPeter Zijlstra anon_vma = NULL; 56088c22088SPeter Zijlstra } 56188c22088SPeter Zijlstra 56288c22088SPeter Zijlstra return anon_vma; 56388c22088SPeter Zijlstra 56488c22088SPeter Zijlstra out: 56588c22088SPeter Zijlstra rcu_read_unlock(); 566746b18d4SPeter Zijlstra return anon_vma; 56734bbd704SOleg Nesterov } 56834bbd704SOleg Nesterov 5694fc3f1d6SIngo Molnar void page_unlock_anon_vma_read(struct anon_vma *anon_vma) 57034bbd704SOleg Nesterov { 5714fc3f1d6SIngo Molnar anon_vma_unlock_read(anon_vma); 5721da177e4SLinus Torvalds } 5731da177e4SLinus Torvalds 57472b252aeSMel Gorman #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH 57572b252aeSMel Gorman /* 57672b252aeSMel Gorman * Flush TLB entries for recently unmapped pages from remote CPUs. It is 57772b252aeSMel Gorman * important if a PTE was dirty when it was unmapped that it's flushed 57872b252aeSMel Gorman * before any IO is initiated on the page to prevent lost writes. Similarly, 57972b252aeSMel Gorman * it must be flushed before freeing to prevent data leakage. 58072b252aeSMel Gorman */ 58172b252aeSMel Gorman void try_to_unmap_flush(void) 58272b252aeSMel Gorman { 58372b252aeSMel Gorman struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc; 58472b252aeSMel Gorman 58572b252aeSMel Gorman if (!tlb_ubc->flush_required) 58672b252aeSMel Gorman return; 58772b252aeSMel Gorman 588e73ad5ffSAndy Lutomirski arch_tlbbatch_flush(&tlb_ubc->arch); 58972b252aeSMel Gorman tlb_ubc->flush_required = false; 590d950c947SMel Gorman tlb_ubc->writable = false; 59172b252aeSMel Gorman } 59272b252aeSMel Gorman 593d950c947SMel Gorman /* Flush iff there are potentially writable TLB entries that can race with IO */ 594d950c947SMel Gorman void try_to_unmap_flush_dirty(void) 595d950c947SMel Gorman { 596d950c947SMel Gorman struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc; 597d950c947SMel Gorman 598d950c947SMel Gorman if (tlb_ubc->writable) 599d950c947SMel Gorman try_to_unmap_flush(); 600d950c947SMel Gorman } 601d950c947SMel Gorman 602c7ab0d2fSKirill A. Shutemov static void set_tlb_ubc_flush_pending(struct mm_struct *mm, bool writable) 60372b252aeSMel Gorman { 60472b252aeSMel Gorman struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc; 60572b252aeSMel Gorman 606e73ad5ffSAndy Lutomirski arch_tlbbatch_add_mm(&tlb_ubc->arch, mm); 60772b252aeSMel Gorman tlb_ubc->flush_required = true; 608d950c947SMel Gorman 609d950c947SMel Gorman /* 6103ea27719SMel Gorman * Ensure compiler does not re-order the setting of tlb_flush_batched 6113ea27719SMel Gorman * before the PTE is cleared. 6123ea27719SMel Gorman */ 6133ea27719SMel Gorman barrier(); 6143ea27719SMel Gorman mm->tlb_flush_batched = true; 6153ea27719SMel Gorman 6163ea27719SMel Gorman /* 617d950c947SMel Gorman * If the PTE was dirty then it's best to assume it's writable. The 618d950c947SMel Gorman * caller must use try_to_unmap_flush_dirty() or try_to_unmap_flush() 619d950c947SMel Gorman * before the page is queued for IO. 620d950c947SMel Gorman */ 621d950c947SMel Gorman if (writable) 622d950c947SMel Gorman tlb_ubc->writable = true; 62372b252aeSMel Gorman } 62472b252aeSMel Gorman 62572b252aeSMel Gorman /* 62672b252aeSMel Gorman * Returns true if the TLB flush should be deferred to the end of a batch of 62772b252aeSMel Gorman * unmap operations to reduce IPIs. 62872b252aeSMel Gorman */ 62972b252aeSMel Gorman static bool should_defer_flush(struct mm_struct *mm, enum ttu_flags flags) 63072b252aeSMel Gorman { 63172b252aeSMel Gorman bool should_defer = false; 63272b252aeSMel Gorman 63372b252aeSMel Gorman if (!(flags & TTU_BATCH_FLUSH)) 63472b252aeSMel Gorman return false; 63572b252aeSMel Gorman 63672b252aeSMel Gorman /* If remote CPUs need to be flushed then defer batch the flush */ 63772b252aeSMel Gorman if (cpumask_any_but(mm_cpumask(mm), get_cpu()) < nr_cpu_ids) 63872b252aeSMel Gorman should_defer = true; 63972b252aeSMel Gorman put_cpu(); 64072b252aeSMel Gorman 64172b252aeSMel Gorman return should_defer; 64272b252aeSMel Gorman } 6433ea27719SMel Gorman 6443ea27719SMel Gorman /* 6453ea27719SMel Gorman * Reclaim unmaps pages under the PTL but do not flush the TLB prior to 6463ea27719SMel Gorman * releasing the PTL if TLB flushes are batched. It's possible for a parallel 6473ea27719SMel Gorman * operation such as mprotect or munmap to race between reclaim unmapping 6483ea27719SMel Gorman * the page and flushing the page. If this race occurs, it potentially allows 6493ea27719SMel Gorman * access to data via a stale TLB entry. Tracking all mm's that have TLB 6503ea27719SMel Gorman * batching in flight would be expensive during reclaim so instead track 6513ea27719SMel Gorman * whether TLB batching occurred in the past and if so then do a flush here 6523ea27719SMel Gorman * if required. This will cost one additional flush per reclaim cycle paid 6533ea27719SMel Gorman * by the first operation at risk such as mprotect and mumap. 6543ea27719SMel Gorman * 6553ea27719SMel Gorman * This must be called under the PTL so that an access to tlb_flush_batched 6563ea27719SMel Gorman * that is potentially a "reclaim vs mprotect/munmap/etc" race will synchronise 6573ea27719SMel Gorman * via the PTL. 6583ea27719SMel Gorman */ 6593ea27719SMel Gorman void flush_tlb_batched_pending(struct mm_struct *mm) 6603ea27719SMel Gorman { 6613ea27719SMel Gorman if (mm->tlb_flush_batched) { 6623ea27719SMel Gorman flush_tlb_mm(mm); 6633ea27719SMel Gorman 6643ea27719SMel Gorman /* 6653ea27719SMel Gorman * Do not allow the compiler to re-order the clearing of 6663ea27719SMel Gorman * tlb_flush_batched before the tlb is flushed. 6673ea27719SMel Gorman */ 6683ea27719SMel Gorman barrier(); 6693ea27719SMel Gorman mm->tlb_flush_batched = false; 6703ea27719SMel Gorman } 6713ea27719SMel Gorman } 67272b252aeSMel Gorman #else 673c7ab0d2fSKirill A. Shutemov static void set_tlb_ubc_flush_pending(struct mm_struct *mm, bool writable) 67472b252aeSMel Gorman { 67572b252aeSMel Gorman } 67672b252aeSMel Gorman 67772b252aeSMel Gorman static bool should_defer_flush(struct mm_struct *mm, enum ttu_flags flags) 67872b252aeSMel Gorman { 67972b252aeSMel Gorman return false; 68072b252aeSMel Gorman } 68172b252aeSMel Gorman #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */ 68272b252aeSMel Gorman 6831da177e4SLinus Torvalds /* 684bf89c8c8SHuang Shijie * At what user virtual address is page expected in vma? 685ab941e0fSNaoya Horiguchi * Caller should check the page is actually part of the vma. 6861da177e4SLinus Torvalds */ 6871da177e4SLinus Torvalds unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) 6881da177e4SLinus Torvalds { 68986c2ad19SMichel Lespinasse unsigned long address; 69021d0d443SAndrea Arcangeli if (PageAnon(page)) { 6914829b906SHugh Dickins struct anon_vma *page__anon_vma = page_anon_vma(page); 6924829b906SHugh Dickins /* 6934829b906SHugh Dickins * Note: swapoff's unuse_vma() is more efficient with this 6944829b906SHugh Dickins * check, and needs it to match anon_vma when KSM is active. 6954829b906SHugh Dickins */ 6964829b906SHugh Dickins if (!vma->anon_vma || !page__anon_vma || 6974829b906SHugh Dickins vma->anon_vma->root != page__anon_vma->root) 69821d0d443SAndrea Arcangeli return -EFAULT; 69927ba0644SKirill A. Shutemov } else if (page->mapping) { 70027ba0644SKirill A. Shutemov if (!vma->vm_file || vma->vm_file->f_mapping != page->mapping) 7011da177e4SLinus Torvalds return -EFAULT; 7021da177e4SLinus Torvalds } else 7031da177e4SLinus Torvalds return -EFAULT; 70486c2ad19SMichel Lespinasse address = __vma_address(page, vma); 70586c2ad19SMichel Lespinasse if (unlikely(address < vma->vm_start || address >= vma->vm_end)) 70686c2ad19SMichel Lespinasse return -EFAULT; 70786c2ad19SMichel Lespinasse return address; 7081da177e4SLinus Torvalds } 7091da177e4SLinus Torvalds 7106219049aSBob Liu pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address) 7116219049aSBob Liu { 7126219049aSBob Liu pgd_t *pgd; 713c2febafcSKirill A. Shutemov p4d_t *p4d; 7146219049aSBob Liu pud_t *pud; 7156219049aSBob Liu pmd_t *pmd = NULL; 716f72e7dcdSHugh Dickins pmd_t pmde; 7176219049aSBob Liu 7186219049aSBob Liu pgd = pgd_offset(mm, address); 7196219049aSBob Liu if (!pgd_present(*pgd)) 7206219049aSBob Liu goto out; 7216219049aSBob Liu 722c2febafcSKirill A. Shutemov p4d = p4d_offset(pgd, address); 723c2febafcSKirill A. Shutemov if (!p4d_present(*p4d)) 724c2febafcSKirill A. Shutemov goto out; 725c2febafcSKirill A. Shutemov 726c2febafcSKirill A. Shutemov pud = pud_offset(p4d, address); 7276219049aSBob Liu if (!pud_present(*pud)) 7286219049aSBob Liu goto out; 7296219049aSBob Liu 7306219049aSBob Liu pmd = pmd_offset(pud, address); 731f72e7dcdSHugh Dickins /* 7328809aa2dSAneesh Kumar K.V * Some THP functions use the sequence pmdp_huge_clear_flush(), set_pmd_at() 733f72e7dcdSHugh Dickins * without holding anon_vma lock for write. So when looking for a 734f72e7dcdSHugh Dickins * genuine pmde (in which to find pte), test present and !THP together. 735f72e7dcdSHugh Dickins */ 736e37c6982SChristian Borntraeger pmde = *pmd; 737e37c6982SChristian Borntraeger barrier(); 738f72e7dcdSHugh Dickins if (!pmd_present(pmde) || pmd_trans_huge(pmde)) 7396219049aSBob Liu pmd = NULL; 7406219049aSBob Liu out: 7416219049aSBob Liu return pmd; 7426219049aSBob Liu } 7436219049aSBob Liu 7449f32624bSJoonsoo Kim struct page_referenced_arg { 7459f32624bSJoonsoo Kim int mapcount; 7469f32624bSJoonsoo Kim int referenced; 7479f32624bSJoonsoo Kim unsigned long vm_flags; 7489f32624bSJoonsoo Kim struct mem_cgroup *memcg; 7499f32624bSJoonsoo Kim }; 75081b4082dSNikita Danilov /* 7519f32624bSJoonsoo Kim * arg: page_referenced_arg will be passed 7521da177e4SLinus Torvalds */ 753e4b82222SMinchan Kim static bool page_referenced_one(struct page *page, struct vm_area_struct *vma, 7549f32624bSJoonsoo Kim unsigned long address, void *arg) 7551da177e4SLinus Torvalds { 7569f32624bSJoonsoo Kim struct page_referenced_arg *pra = arg; 7578eaededeSKirill A. Shutemov struct page_vma_mapped_walk pvmw = { 7588eaededeSKirill A. Shutemov .page = page, 7598eaededeSKirill A. Shutemov .vma = vma, 7608eaededeSKirill A. Shutemov .address = address, 7618eaededeSKirill A. Shutemov }; 7628749cfeaSVladimir Davydov int referenced = 0; 7632da28bfdSAndrea Arcangeli 7648eaededeSKirill A. Shutemov while (page_vma_mapped_walk(&pvmw)) { 7658eaededeSKirill A. Shutemov address = pvmw.address; 7662da28bfdSAndrea Arcangeli 767b20ce5e0SKirill A. Shutemov if (vma->vm_flags & VM_LOCKED) { 7688eaededeSKirill A. Shutemov page_vma_mapped_walk_done(&pvmw); 7699f32624bSJoonsoo Kim pra->vm_flags |= VM_LOCKED; 770e4b82222SMinchan Kim return false; /* To break the loop */ 7712da28bfdSAndrea Arcangeli } 7722da28bfdSAndrea Arcangeli 7738eaededeSKirill A. Shutemov if (pvmw.pte) { 7748eaededeSKirill A. Shutemov if (ptep_clear_flush_young_notify(vma, address, 7758eaededeSKirill A. Shutemov pvmw.pte)) { 7764917e5d0SJohannes Weiner /* 7778eaededeSKirill A. Shutemov * Don't treat a reference through 7788eaededeSKirill A. Shutemov * a sequentially read mapping as such. 7798eaededeSKirill A. Shutemov * If the page has been used in another mapping, 7808eaededeSKirill A. Shutemov * we will catch it; if this other mapping is 7818eaededeSKirill A. Shutemov * already gone, the unmap path will have set 7828eaededeSKirill A. Shutemov * PG_referenced or activated the page. 7834917e5d0SJohannes Weiner */ 78464363aadSJoe Perches if (likely(!(vma->vm_flags & VM_SEQ_READ))) 7851da177e4SLinus Torvalds referenced++; 7864917e5d0SJohannes Weiner } 7878749cfeaSVladimir Davydov } else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) { 7888eaededeSKirill A. Shutemov if (pmdp_clear_flush_young_notify(vma, address, 7898eaededeSKirill A. Shutemov pvmw.pmd)) 7908749cfeaSVladimir Davydov referenced++; 7918749cfeaSVladimir Davydov } else { 7928749cfeaSVladimir Davydov /* unexpected pmd-mapped page? */ 7938749cfeaSVladimir Davydov WARN_ON_ONCE(1); 7948749cfeaSVladimir Davydov } 7958eaededeSKirill A. Shutemov 7968eaededeSKirill A. Shutemov pra->mapcount--; 7978eaededeSKirill A. Shutemov } 79871e3aac0SAndrea Arcangeli 79933c3fc71SVladimir Davydov if (referenced) 80033c3fc71SVladimir Davydov clear_page_idle(page); 80133c3fc71SVladimir Davydov if (test_and_clear_page_young(page)) 80233c3fc71SVladimir Davydov referenced++; 80333c3fc71SVladimir Davydov 8049f32624bSJoonsoo Kim if (referenced) { 8059f32624bSJoonsoo Kim pra->referenced++; 8069f32624bSJoonsoo Kim pra->vm_flags |= vma->vm_flags; 8071da177e4SLinus Torvalds } 8081da177e4SLinus Torvalds 8099f32624bSJoonsoo Kim if (!pra->mapcount) 810e4b82222SMinchan Kim return false; /* To break the loop */ 8119f32624bSJoonsoo Kim 812e4b82222SMinchan Kim return true; 8139f32624bSJoonsoo Kim } 8149f32624bSJoonsoo Kim 8159f32624bSJoonsoo Kim static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg) 8161da177e4SLinus Torvalds { 8179f32624bSJoonsoo Kim struct page_referenced_arg *pra = arg; 8189f32624bSJoonsoo Kim struct mem_cgroup *memcg = pra->memcg; 8191da177e4SLinus Torvalds 8209f32624bSJoonsoo Kim if (!mm_match_cgroup(vma->vm_mm, memcg)) 8219f32624bSJoonsoo Kim return true; 8221da177e4SLinus Torvalds 8239f32624bSJoonsoo Kim return false; 8241da177e4SLinus Torvalds } 8251da177e4SLinus Torvalds 8261da177e4SLinus Torvalds /** 8271da177e4SLinus Torvalds * page_referenced - test if the page was referenced 8281da177e4SLinus Torvalds * @page: the page to test 8291da177e4SLinus Torvalds * @is_locked: caller holds lock on the page 83072835c86SJohannes Weiner * @memcg: target memory cgroup 8316fe6b7e3SWu Fengguang * @vm_flags: collect encountered vma->vm_flags who actually referenced the page 8321da177e4SLinus Torvalds * 8331da177e4SLinus Torvalds * Quick test_and_clear_referenced for all mappings to a page, 8341da177e4SLinus Torvalds * returns the number of ptes which referenced the page. 8351da177e4SLinus Torvalds */ 8366fe6b7e3SWu Fengguang int page_referenced(struct page *page, 8376fe6b7e3SWu Fengguang int is_locked, 83872835c86SJohannes Weiner struct mem_cgroup *memcg, 8396fe6b7e3SWu Fengguang unsigned long *vm_flags) 8401da177e4SLinus Torvalds { 8415ad64688SHugh Dickins int we_locked = 0; 8429f32624bSJoonsoo Kim struct page_referenced_arg pra = { 843b20ce5e0SKirill A. Shutemov .mapcount = total_mapcount(page), 8449f32624bSJoonsoo Kim .memcg = memcg, 8459f32624bSJoonsoo Kim }; 8469f32624bSJoonsoo Kim struct rmap_walk_control rwc = { 8479f32624bSJoonsoo Kim .rmap_one = page_referenced_one, 8489f32624bSJoonsoo Kim .arg = (void *)&pra, 8499f32624bSJoonsoo Kim .anon_lock = page_lock_anon_vma_read, 8509f32624bSJoonsoo Kim }; 8511da177e4SLinus Torvalds 8526fe6b7e3SWu Fengguang *vm_flags = 0; 8539f32624bSJoonsoo Kim if (!page_mapped(page)) 8549f32624bSJoonsoo Kim return 0; 8559f32624bSJoonsoo Kim 8569f32624bSJoonsoo Kim if (!page_rmapping(page)) 8579f32624bSJoonsoo Kim return 0; 8589f32624bSJoonsoo Kim 8595ad64688SHugh Dickins if (!is_locked && (!PageAnon(page) || PageKsm(page))) { 8605ad64688SHugh Dickins we_locked = trylock_page(page); 8619f32624bSJoonsoo Kim if (!we_locked) 8629f32624bSJoonsoo Kim return 1; 8635ad64688SHugh Dickins } 8649f32624bSJoonsoo Kim 8659f32624bSJoonsoo Kim /* 8669f32624bSJoonsoo Kim * If we are reclaiming on behalf of a cgroup, skip 8679f32624bSJoonsoo Kim * counting on behalf of references from different 8689f32624bSJoonsoo Kim * cgroups 8699f32624bSJoonsoo Kim */ 8709f32624bSJoonsoo Kim if (memcg) { 8719f32624bSJoonsoo Kim rwc.invalid_vma = invalid_page_referenced_vma; 8725ad64688SHugh Dickins } 8739f32624bSJoonsoo Kim 874c24f386cSMinchan Kim rmap_walk(page, &rwc); 8759f32624bSJoonsoo Kim *vm_flags = pra.vm_flags; 8769f32624bSJoonsoo Kim 8775ad64688SHugh Dickins if (we_locked) 8781da177e4SLinus Torvalds unlock_page(page); 8799f32624bSJoonsoo Kim 8809f32624bSJoonsoo Kim return pra.referenced; 8811da177e4SLinus Torvalds } 8821da177e4SLinus Torvalds 883e4b82222SMinchan Kim static bool page_mkclean_one(struct page *page, struct vm_area_struct *vma, 8849853a407SJoonsoo Kim unsigned long address, void *arg) 885d08b3851SPeter Zijlstra { 886f27176cfSKirill A. Shutemov struct page_vma_mapped_walk pvmw = { 887f27176cfSKirill A. Shutemov .page = page, 888f27176cfSKirill A. Shutemov .vma = vma, 889f27176cfSKirill A. Shutemov .address = address, 890f27176cfSKirill A. Shutemov .flags = PVMW_SYNC, 891f27176cfSKirill A. Shutemov }; 892ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 8939853a407SJoonsoo Kim int *cleaned = arg; 894d08b3851SPeter Zijlstra 895369ea824SJérôme Glisse /* 896369ea824SJérôme Glisse * We have to assume the worse case ie pmd for invalidation. Note that 897369ea824SJérôme Glisse * the page can not be free from this function. 898369ea824SJérôme Glisse */ 899ac46d4f3SJérôme Glisse mmu_notifier_range_init(&range, vma->vm_mm, address, 900ac46d4f3SJérôme Glisse min(vma->vm_end, address + 901ac46d4f3SJérôme Glisse (PAGE_SIZE << compound_order(page)))); 902ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 903369ea824SJérôme Glisse 904f27176cfSKirill A. Shutemov while (page_vma_mapped_walk(&pvmw)) { 905cdb07bdeSColin Ian King unsigned long cstart; 906f27176cfSKirill A. Shutemov int ret = 0; 907369ea824SJérôme Glisse 908369ea824SJérôme Glisse cstart = address = pvmw.address; 909f27176cfSKirill A. Shutemov if (pvmw.pte) { 910c2fda5feSPeter Zijlstra pte_t entry; 911f27176cfSKirill A. Shutemov pte_t *pte = pvmw.pte; 912f27176cfSKirill A. Shutemov 913f27176cfSKirill A. Shutemov if (!pte_dirty(*pte) && !pte_write(*pte)) 914f27176cfSKirill A. Shutemov continue; 915d08b3851SPeter Zijlstra 916785373b4SLinus Torvalds flush_cache_page(vma, address, pte_pfn(*pte)); 917785373b4SLinus Torvalds entry = ptep_clear_flush(vma, address, pte); 918d08b3851SPeter Zijlstra entry = pte_wrprotect(entry); 919c2fda5feSPeter Zijlstra entry = pte_mkclean(entry); 920785373b4SLinus Torvalds set_pte_at(vma->vm_mm, address, pte, entry); 921d08b3851SPeter Zijlstra ret = 1; 922f27176cfSKirill A. Shutemov } else { 923f27176cfSKirill A. Shutemov #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE 924f27176cfSKirill A. Shutemov pmd_t *pmd = pvmw.pmd; 925f27176cfSKirill A. Shutemov pmd_t entry; 926d08b3851SPeter Zijlstra 927f27176cfSKirill A. Shutemov if (!pmd_dirty(*pmd) && !pmd_write(*pmd)) 928f27176cfSKirill A. Shutemov continue; 929f27176cfSKirill A. Shutemov 930785373b4SLinus Torvalds flush_cache_page(vma, address, page_to_pfn(page)); 931*024eee0eSAneesh Kumar K.V entry = pmdp_invalidate(vma, address, pmd); 932f27176cfSKirill A. Shutemov entry = pmd_wrprotect(entry); 933f27176cfSKirill A. Shutemov entry = pmd_mkclean(entry); 934785373b4SLinus Torvalds set_pmd_at(vma->vm_mm, address, pmd, entry); 935369ea824SJérôme Glisse cstart &= PMD_MASK; 936f27176cfSKirill A. Shutemov ret = 1; 937f27176cfSKirill A. Shutemov #else 938f27176cfSKirill A. Shutemov /* unexpected pmd-mapped page? */ 939f27176cfSKirill A. Shutemov WARN_ON_ONCE(1); 940f27176cfSKirill A. Shutemov #endif 941f27176cfSKirill A. Shutemov } 9422ec74c3eSSagi Grimberg 9430f10851eSJérôme Glisse /* 9440f10851eSJérôme Glisse * No need to call mmu_notifier_invalidate_range() as we are 9450f10851eSJérôme Glisse * downgrading page table protection not changing it to point 9460f10851eSJérôme Glisse * to a new page. 9470f10851eSJérôme Glisse * 948ad56b738SMike Rapoport * See Documentation/vm/mmu_notifier.rst 9490f10851eSJérôme Glisse */ 9500f10851eSJérôme Glisse if (ret) 9519853a407SJoonsoo Kim (*cleaned)++; 9529853a407SJoonsoo Kim } 953f27176cfSKirill A. Shutemov 954ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(&range); 955369ea824SJérôme Glisse 956e4b82222SMinchan Kim return true; 957d08b3851SPeter Zijlstra } 958d08b3851SPeter Zijlstra 9599853a407SJoonsoo Kim static bool invalid_mkclean_vma(struct vm_area_struct *vma, void *arg) 960d08b3851SPeter Zijlstra { 9619853a407SJoonsoo Kim if (vma->vm_flags & VM_SHARED) 962871beb8cSFengguang Wu return false; 963d08b3851SPeter Zijlstra 964871beb8cSFengguang Wu return true; 965d08b3851SPeter Zijlstra } 966d08b3851SPeter Zijlstra 967d08b3851SPeter Zijlstra int page_mkclean(struct page *page) 968d08b3851SPeter Zijlstra { 9699853a407SJoonsoo Kim int cleaned = 0; 9709853a407SJoonsoo Kim struct address_space *mapping; 9719853a407SJoonsoo Kim struct rmap_walk_control rwc = { 9729853a407SJoonsoo Kim .arg = (void *)&cleaned, 9739853a407SJoonsoo Kim .rmap_one = page_mkclean_one, 9749853a407SJoonsoo Kim .invalid_vma = invalid_mkclean_vma, 9759853a407SJoonsoo Kim }; 976d08b3851SPeter Zijlstra 977d08b3851SPeter Zijlstra BUG_ON(!PageLocked(page)); 978d08b3851SPeter Zijlstra 9799853a407SJoonsoo Kim if (!page_mapped(page)) 9809853a407SJoonsoo Kim return 0; 981d08b3851SPeter Zijlstra 9829853a407SJoonsoo Kim mapping = page_mapping(page); 9839853a407SJoonsoo Kim if (!mapping) 9849853a407SJoonsoo Kim return 0; 9859853a407SJoonsoo Kim 9869853a407SJoonsoo Kim rmap_walk(page, &rwc); 9879853a407SJoonsoo Kim 9889853a407SJoonsoo Kim return cleaned; 989d08b3851SPeter Zijlstra } 99060b59beaSJaya Kumar EXPORT_SYMBOL_GPL(page_mkclean); 991d08b3851SPeter Zijlstra 9921da177e4SLinus Torvalds /** 993c44b6743SRik van Riel * page_move_anon_rmap - move a page to our anon_vma 994c44b6743SRik van Riel * @page: the page to move to our anon_vma 995c44b6743SRik van Riel * @vma: the vma the page belongs to 996c44b6743SRik van Riel * 997c44b6743SRik van Riel * When a page belongs exclusively to one process after a COW event, 998c44b6743SRik van Riel * that page can be moved into the anon_vma that belongs to just that 999c44b6743SRik van Riel * process, so the rmap code will not search the parent or sibling 1000c44b6743SRik van Riel * processes. 1001c44b6743SRik van Riel */ 10025a49973dSHugh Dickins void page_move_anon_rmap(struct page *page, struct vm_area_struct *vma) 1003c44b6743SRik van Riel { 1004c44b6743SRik van Riel struct anon_vma *anon_vma = vma->anon_vma; 1005c44b6743SRik van Riel 10065a49973dSHugh Dickins page = compound_head(page); 10075a49973dSHugh Dickins 1008309381feSSasha Levin VM_BUG_ON_PAGE(!PageLocked(page), page); 100981d1b09cSSasha Levin VM_BUG_ON_VMA(!anon_vma, vma); 1010c44b6743SRik van Riel 1011c44b6743SRik van Riel anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; 1012414e2fb8SVladimir Davydov /* 1013414e2fb8SVladimir Davydov * Ensure that anon_vma and the PAGE_MAPPING_ANON bit are written 1014414e2fb8SVladimir Davydov * simultaneously, so a concurrent reader (eg page_referenced()'s 1015414e2fb8SVladimir Davydov * PageAnon()) will not see one without the other. 1016414e2fb8SVladimir Davydov */ 1017414e2fb8SVladimir Davydov WRITE_ONCE(page->mapping, (struct address_space *) anon_vma); 1018c44b6743SRik van Riel } 1019c44b6743SRik van Riel 1020c44b6743SRik van Riel /** 102143d8eac4SRandy Dunlap * __page_set_anon_rmap - set up new anonymous rmap 1022451b9514SKirill Tkhai * @page: Page or Hugepage to add to rmap 10234e1c1975SAndi Kleen * @vma: VM area to add page to. 10244e1c1975SAndi Kleen * @address: User virtual address of the mapping 1025e8a03febSRik van Riel * @exclusive: the page is exclusively owned by the current process 10261da177e4SLinus Torvalds */ 10279617d95eSNick Piggin static void __page_set_anon_rmap(struct page *page, 1028e8a03febSRik van Riel struct vm_area_struct *vma, unsigned long address, int exclusive) 10291da177e4SLinus Torvalds { 1030e8a03febSRik van Riel struct anon_vma *anon_vma = vma->anon_vma; 10312822c1aaSNick Piggin 1032e8a03febSRik van Riel BUG_ON(!anon_vma); 1033ea90002bSLinus Torvalds 10344e1c1975SAndi Kleen if (PageAnon(page)) 10354e1c1975SAndi Kleen return; 10364e1c1975SAndi Kleen 1037ea90002bSLinus Torvalds /* 1038e8a03febSRik van Riel * If the page isn't exclusively mapped into this vma, 1039e8a03febSRik van Riel * we must use the _oldest_ possible anon_vma for the 1040e8a03febSRik van Riel * page mapping! 1041ea90002bSLinus Torvalds */ 10424e1c1975SAndi Kleen if (!exclusive) 1043288468c3SAndrea Arcangeli anon_vma = anon_vma->root; 1044ea90002bSLinus Torvalds 10451da177e4SLinus Torvalds anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; 10462822c1aaSNick Piggin page->mapping = (struct address_space *) anon_vma; 10474d7670e0SNick Piggin page->index = linear_page_index(vma, address); 10481da177e4SLinus Torvalds } 10499617d95eSNick Piggin 10509617d95eSNick Piggin /** 105143d8eac4SRandy Dunlap * __page_check_anon_rmap - sanity check anonymous rmap addition 1052c97a9e10SNick Piggin * @page: the page to add the mapping to 1053c97a9e10SNick Piggin * @vma: the vm area in which the mapping is added 1054c97a9e10SNick Piggin * @address: the user virtual address mapped 1055c97a9e10SNick Piggin */ 1056c97a9e10SNick Piggin static void __page_check_anon_rmap(struct page *page, 1057c97a9e10SNick Piggin struct vm_area_struct *vma, unsigned long address) 1058c97a9e10SNick Piggin { 1059c97a9e10SNick Piggin #ifdef CONFIG_DEBUG_VM 1060c97a9e10SNick Piggin /* 1061c97a9e10SNick Piggin * The page's anon-rmap details (mapping and index) are guaranteed to 1062c97a9e10SNick Piggin * be set up correctly at this point. 1063c97a9e10SNick Piggin * 1064c97a9e10SNick Piggin * We have exclusion against page_add_anon_rmap because the caller 1065c97a9e10SNick Piggin * always holds the page locked, except if called from page_dup_rmap, 1066c97a9e10SNick Piggin * in which case the page is already known to be setup. 1067c97a9e10SNick Piggin * 1068c97a9e10SNick Piggin * We have exclusion against page_add_new_anon_rmap because those pages 1069c97a9e10SNick Piggin * are initially only visible via the pagetables, and the pte is locked 1070c97a9e10SNick Piggin * over the call to page_add_new_anon_rmap. 1071c97a9e10SNick Piggin */ 107244ab57a0SAndrea Arcangeli BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root); 107353f9263bSKirill A. Shutemov BUG_ON(page_to_pgoff(page) != linear_page_index(vma, address)); 1074c97a9e10SNick Piggin #endif 1075c97a9e10SNick Piggin } 1076c97a9e10SNick Piggin 1077c97a9e10SNick Piggin /** 10789617d95eSNick Piggin * page_add_anon_rmap - add pte mapping to an anonymous page 10799617d95eSNick Piggin * @page: the page to add the mapping to 10809617d95eSNick Piggin * @vma: the vm area in which the mapping is added 10819617d95eSNick Piggin * @address: the user virtual address mapped 1082d281ee61SKirill A. Shutemov * @compound: charge the page as compound or small page 10839617d95eSNick Piggin * 10845ad64688SHugh Dickins * The caller needs to hold the pte lock, and the page must be locked in 108580e14822SHugh Dickins * the anon_vma case: to serialize mapping,index checking after setting, 108680e14822SHugh Dickins * and to ensure that PageAnon is not being upgraded racily to PageKsm 108780e14822SHugh Dickins * (but PageKsm is never downgraded to PageAnon). 10889617d95eSNick Piggin */ 10899617d95eSNick Piggin void page_add_anon_rmap(struct page *page, 1090d281ee61SKirill A. Shutemov struct vm_area_struct *vma, unsigned long address, bool compound) 10919617d95eSNick Piggin { 1092d281ee61SKirill A. Shutemov do_page_add_anon_rmap(page, vma, address, compound ? RMAP_COMPOUND : 0); 1093ad8c2ee8SRik van Riel } 1094ad8c2ee8SRik van Riel 1095ad8c2ee8SRik van Riel /* 1096ad8c2ee8SRik van Riel * Special version of the above for do_swap_page, which often runs 1097ad8c2ee8SRik van Riel * into pages that are exclusively owned by the current process. 1098ad8c2ee8SRik van Riel * Everybody else should continue to use page_add_anon_rmap above. 1099ad8c2ee8SRik van Riel */ 1100ad8c2ee8SRik van Riel void do_page_add_anon_rmap(struct page *page, 1101d281ee61SKirill A. Shutemov struct vm_area_struct *vma, unsigned long address, int flags) 1102ad8c2ee8SRik van Riel { 1103d281ee61SKirill A. Shutemov bool compound = flags & RMAP_COMPOUND; 110453f9263bSKirill A. Shutemov bool first; 110553f9263bSKirill A. Shutemov 110653f9263bSKirill A. Shutemov if (compound) { 110753f9263bSKirill A. Shutemov atomic_t *mapcount; 1108e9b61f19SKirill A. Shutemov VM_BUG_ON_PAGE(!PageLocked(page), page); 110953f9263bSKirill A. Shutemov VM_BUG_ON_PAGE(!PageTransHuge(page), page); 111053f9263bSKirill A. Shutemov mapcount = compound_mapcount_ptr(page); 111153f9263bSKirill A. Shutemov first = atomic_inc_and_test(mapcount); 111253f9263bSKirill A. Shutemov } else { 111353f9263bSKirill A. Shutemov first = atomic_inc_and_test(&page->_mapcount); 111453f9263bSKirill A. Shutemov } 111553f9263bSKirill A. Shutemov 111653f9263bSKirill A. Shutemov if (first) { 1117d281ee61SKirill A. Shutemov int nr = compound ? hpage_nr_pages(page) : 1; 1118bea04b07SJianyu Zhan /* 1119bea04b07SJianyu Zhan * We use the irq-unsafe __{inc|mod}_zone_page_stat because 1120bea04b07SJianyu Zhan * these counters are not modified in interrupt context, and 1121bea04b07SJianyu Zhan * pte lock(a spinlock) is held, which implies preemption 1122bea04b07SJianyu Zhan * disabled. 1123bea04b07SJianyu Zhan */ 112465c45377SKirill A. Shutemov if (compound) 112511fb9989SMel Gorman __inc_node_page_state(page, NR_ANON_THPS); 11264b9d0fabSMel Gorman __mod_node_page_state(page_pgdat(page), NR_ANON_MAPPED, nr); 112779134171SAndrea Arcangeli } 11285ad64688SHugh Dickins if (unlikely(PageKsm(page))) 11295ad64688SHugh Dickins return; 11305ad64688SHugh Dickins 1131309381feSSasha Levin VM_BUG_ON_PAGE(!PageLocked(page), page); 113253f9263bSKirill A. Shutemov 11335dbe0af4SHugh Dickins /* address might be in next vma when migration races vma_adjust */ 11345ad64688SHugh Dickins if (first) 1135d281ee61SKirill A. Shutemov __page_set_anon_rmap(page, vma, address, 1136d281ee61SKirill A. Shutemov flags & RMAP_EXCLUSIVE); 113769029cd5SKAMEZAWA Hiroyuki else 1138c97a9e10SNick Piggin __page_check_anon_rmap(page, vma, address); 11391da177e4SLinus Torvalds } 11401da177e4SLinus Torvalds 114143d8eac4SRandy Dunlap /** 11429617d95eSNick Piggin * page_add_new_anon_rmap - add pte mapping to a new anonymous page 11439617d95eSNick Piggin * @page: the page to add the mapping to 11449617d95eSNick Piggin * @vma: the vm area in which the mapping is added 11459617d95eSNick Piggin * @address: the user virtual address mapped 1146d281ee61SKirill A. Shutemov * @compound: charge the page as compound or small page 11479617d95eSNick Piggin * 11489617d95eSNick Piggin * Same as page_add_anon_rmap but must only be called on *new* pages. 11499617d95eSNick Piggin * This means the inc-and-test can be bypassed. 1150c97a9e10SNick Piggin * Page does not have to be locked. 11519617d95eSNick Piggin */ 11529617d95eSNick Piggin void page_add_new_anon_rmap(struct page *page, 1153d281ee61SKirill A. Shutemov struct vm_area_struct *vma, unsigned long address, bool compound) 11549617d95eSNick Piggin { 1155d281ee61SKirill A. Shutemov int nr = compound ? hpage_nr_pages(page) : 1; 1156d281ee61SKirill A. Shutemov 115781d1b09cSSasha Levin VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma); 1158fa9949daSHugh Dickins __SetPageSwapBacked(page); 1159d281ee61SKirill A. Shutemov if (compound) { 1160d281ee61SKirill A. Shutemov VM_BUG_ON_PAGE(!PageTransHuge(page), page); 116153f9263bSKirill A. Shutemov /* increment count (starts at -1) */ 116253f9263bSKirill A. Shutemov atomic_set(compound_mapcount_ptr(page), 0); 116311fb9989SMel Gorman __inc_node_page_state(page, NR_ANON_THPS); 116453f9263bSKirill A. Shutemov } else { 116553f9263bSKirill A. Shutemov /* Anon THP always mapped first with PMD */ 116653f9263bSKirill A. Shutemov VM_BUG_ON_PAGE(PageTransCompound(page), page); 116753f9263bSKirill A. Shutemov /* increment count (starts at -1) */ 116853f9263bSKirill A. Shutemov atomic_set(&page->_mapcount, 0); 1169d281ee61SKirill A. Shutemov } 11704b9d0fabSMel Gorman __mod_node_page_state(page_pgdat(page), NR_ANON_MAPPED, nr); 1171e8a03febSRik van Riel __page_set_anon_rmap(page, vma, address, 1); 11729617d95eSNick Piggin } 11739617d95eSNick Piggin 11741da177e4SLinus Torvalds /** 11751da177e4SLinus Torvalds * page_add_file_rmap - add pte mapping to a file page 11761da177e4SLinus Torvalds * @page: the page to add the mapping to 1177e8b098fcSMike Rapoport * @compound: charge the page as compound or small page 11781da177e4SLinus Torvalds * 1179b8072f09SHugh Dickins * The caller needs to hold the pte lock. 11801da177e4SLinus Torvalds */ 1181dd78feddSKirill A. Shutemov void page_add_file_rmap(struct page *page, bool compound) 11821da177e4SLinus Torvalds { 1183dd78feddSKirill A. Shutemov int i, nr = 1; 1184dd78feddSKirill A. Shutemov 1185dd78feddSKirill A. Shutemov VM_BUG_ON_PAGE(compound && !PageTransHuge(page), page); 118662cccb8cSJohannes Weiner lock_page_memcg(page); 1187dd78feddSKirill A. Shutemov if (compound && PageTransHuge(page)) { 1188dd78feddSKirill A. Shutemov for (i = 0, nr = 0; i < HPAGE_PMD_NR; i++) { 1189dd78feddSKirill A. Shutemov if (atomic_inc_and_test(&page[i]._mapcount)) 1190dd78feddSKirill A. Shutemov nr++; 1191d69b042fSBalbir Singh } 1192dd78feddSKirill A. Shutemov if (!atomic_inc_and_test(compound_mapcount_ptr(page))) 1193dd78feddSKirill A. Shutemov goto out; 119465c45377SKirill A. Shutemov VM_BUG_ON_PAGE(!PageSwapBacked(page), page); 119511fb9989SMel Gorman __inc_node_page_state(page, NR_SHMEM_PMDMAPPED); 1196dd78feddSKirill A. Shutemov } else { 1197c8efc390SKirill A. Shutemov if (PageTransCompound(page) && page_mapping(page)) { 1198c8efc390SKirill A. Shutemov VM_WARN_ON_ONCE(!PageLocked(page)); 1199c8efc390SKirill A. Shutemov 12009a73f61bSKirill A. Shutemov SetPageDoubleMap(compound_head(page)); 12019a73f61bSKirill A. Shutemov if (PageMlocked(page)) 12029a73f61bSKirill A. Shutemov clear_page_mlock(compound_head(page)); 12039a73f61bSKirill A. Shutemov } 1204dd78feddSKirill A. Shutemov if (!atomic_inc_and_test(&page->_mapcount)) 1205dd78feddSKirill A. Shutemov goto out; 1206dd78feddSKirill A. Shutemov } 120700f3ca2cSJohannes Weiner __mod_lruvec_page_state(page, NR_FILE_MAPPED, nr); 1208dd78feddSKirill A. Shutemov out: 120962cccb8cSJohannes Weiner unlock_page_memcg(page); 12101da177e4SLinus Torvalds } 12111da177e4SLinus Torvalds 1212dd78feddSKirill A. Shutemov static void page_remove_file_rmap(struct page *page, bool compound) 12138186eb6aSJohannes Weiner { 1214dd78feddSKirill A. Shutemov int i, nr = 1; 1215dd78feddSKirill A. Shutemov 121657dea93aSSteve Capper VM_BUG_ON_PAGE(compound && !PageHead(page), page); 121762cccb8cSJohannes Weiner lock_page_memcg(page); 12188186eb6aSJohannes Weiner 121953f9263bSKirill A. Shutemov /* Hugepages are not counted in NR_FILE_MAPPED for now. */ 122053f9263bSKirill A. Shutemov if (unlikely(PageHuge(page))) { 122153f9263bSKirill A. Shutemov /* hugetlb pages are always mapped with pmds */ 122253f9263bSKirill A. Shutemov atomic_dec(compound_mapcount_ptr(page)); 122353f9263bSKirill A. Shutemov goto out; 122453f9263bSKirill A. Shutemov } 122553f9263bSKirill A. Shutemov 12268186eb6aSJohannes Weiner /* page still mapped by someone else? */ 1227dd78feddSKirill A. Shutemov if (compound && PageTransHuge(page)) { 1228dd78feddSKirill A. Shutemov for (i = 0, nr = 0; i < HPAGE_PMD_NR; i++) { 1229dd78feddSKirill A. Shutemov if (atomic_add_negative(-1, &page[i]._mapcount)) 1230dd78feddSKirill A. Shutemov nr++; 1231dd78feddSKirill A. Shutemov } 1232dd78feddSKirill A. Shutemov if (!atomic_add_negative(-1, compound_mapcount_ptr(page))) 1233dd78feddSKirill A. Shutemov goto out; 123465c45377SKirill A. Shutemov VM_BUG_ON_PAGE(!PageSwapBacked(page), page); 123511fb9989SMel Gorman __dec_node_page_state(page, NR_SHMEM_PMDMAPPED); 1236dd78feddSKirill A. Shutemov } else { 12378186eb6aSJohannes Weiner if (!atomic_add_negative(-1, &page->_mapcount)) 12388186eb6aSJohannes Weiner goto out; 1239dd78feddSKirill A. Shutemov } 12408186eb6aSJohannes Weiner 12418186eb6aSJohannes Weiner /* 124200f3ca2cSJohannes Weiner * We use the irq-unsafe __{inc|mod}_lruvec_page_state because 12438186eb6aSJohannes Weiner * these counters are not modified in interrupt context, and 12448186eb6aSJohannes Weiner * pte lock(a spinlock) is held, which implies preemption disabled. 12458186eb6aSJohannes Weiner */ 124600f3ca2cSJohannes Weiner __mod_lruvec_page_state(page, NR_FILE_MAPPED, -nr); 12478186eb6aSJohannes Weiner 12488186eb6aSJohannes Weiner if (unlikely(PageMlocked(page))) 12498186eb6aSJohannes Weiner clear_page_mlock(page); 12508186eb6aSJohannes Weiner out: 125162cccb8cSJohannes Weiner unlock_page_memcg(page); 12528186eb6aSJohannes Weiner } 12538186eb6aSJohannes Weiner 125453f9263bSKirill A. Shutemov static void page_remove_anon_compound_rmap(struct page *page) 125553f9263bSKirill A. Shutemov { 125653f9263bSKirill A. Shutemov int i, nr; 125753f9263bSKirill A. Shutemov 125853f9263bSKirill A. Shutemov if (!atomic_add_negative(-1, compound_mapcount_ptr(page))) 125953f9263bSKirill A. Shutemov return; 126053f9263bSKirill A. Shutemov 126153f9263bSKirill A. Shutemov /* Hugepages are not counted in NR_ANON_PAGES for now. */ 126253f9263bSKirill A. Shutemov if (unlikely(PageHuge(page))) 126353f9263bSKirill A. Shutemov return; 126453f9263bSKirill A. Shutemov 126553f9263bSKirill A. Shutemov if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) 126653f9263bSKirill A. Shutemov return; 126753f9263bSKirill A. Shutemov 126811fb9989SMel Gorman __dec_node_page_state(page, NR_ANON_THPS); 126953f9263bSKirill A. Shutemov 127053f9263bSKirill A. Shutemov if (TestClearPageDoubleMap(page)) { 127153f9263bSKirill A. Shutemov /* 127253f9263bSKirill A. Shutemov * Subpages can be mapped with PTEs too. Check how many of 127353f9263bSKirill A. Shutemov * themi are still mapped. 127453f9263bSKirill A. Shutemov */ 127553f9263bSKirill A. Shutemov for (i = 0, nr = 0; i < HPAGE_PMD_NR; i++) { 127653f9263bSKirill A. Shutemov if (atomic_add_negative(-1, &page[i]._mapcount)) 127753f9263bSKirill A. Shutemov nr++; 127853f9263bSKirill A. Shutemov } 127953f9263bSKirill A. Shutemov } else { 128053f9263bSKirill A. Shutemov nr = HPAGE_PMD_NR; 128153f9263bSKirill A. Shutemov } 128253f9263bSKirill A. Shutemov 1283e90309c9SKirill A. Shutemov if (unlikely(PageMlocked(page))) 1284e90309c9SKirill A. Shutemov clear_page_mlock(page); 1285e90309c9SKirill A. Shutemov 12869a982250SKirill A. Shutemov if (nr) { 12874b9d0fabSMel Gorman __mod_node_page_state(page_pgdat(page), NR_ANON_MAPPED, -nr); 12889a982250SKirill A. Shutemov deferred_split_huge_page(page); 12899a982250SKirill A. Shutemov } 129053f9263bSKirill A. Shutemov } 129153f9263bSKirill A. Shutemov 12921da177e4SLinus Torvalds /** 12931da177e4SLinus Torvalds * page_remove_rmap - take down pte mapping from a page 12941da177e4SLinus Torvalds * @page: page to remove mapping from 1295d281ee61SKirill A. Shutemov * @compound: uncharge the page as compound or small page 12961da177e4SLinus Torvalds * 1297b8072f09SHugh Dickins * The caller needs to hold the pte lock. 12981da177e4SLinus Torvalds */ 1299d281ee61SKirill A. Shutemov void page_remove_rmap(struct page *page, bool compound) 13001da177e4SLinus Torvalds { 1301dd78feddSKirill A. Shutemov if (!PageAnon(page)) 1302dd78feddSKirill A. Shutemov return page_remove_file_rmap(page, compound); 130389c06bd5SKAMEZAWA Hiroyuki 130453f9263bSKirill A. Shutemov if (compound) 130553f9263bSKirill A. Shutemov return page_remove_anon_compound_rmap(page); 130653f9263bSKirill A. Shutemov 1307b904dcfeSKOSAKI Motohiro /* page still mapped by someone else? */ 1308b904dcfeSKOSAKI Motohiro if (!atomic_add_negative(-1, &page->_mapcount)) 13098186eb6aSJohannes Weiner return; 13108186eb6aSJohannes Weiner 13111da177e4SLinus Torvalds /* 1312bea04b07SJianyu Zhan * We use the irq-unsafe __{inc|mod}_zone_page_stat because 1313bea04b07SJianyu Zhan * these counters are not modified in interrupt context, and 1314bea04b07SJianyu Zhan * pte lock(a spinlock) is held, which implies preemption disabled. 13150fe6e20bSNaoya Horiguchi */ 13164b9d0fabSMel Gorman __dec_node_page_state(page, NR_ANON_MAPPED); 13178186eb6aSJohannes Weiner 1318e6c509f8SHugh Dickins if (unlikely(PageMlocked(page))) 1319e6c509f8SHugh Dickins clear_page_mlock(page); 13208186eb6aSJohannes Weiner 13219a982250SKirill A. Shutemov if (PageTransCompound(page)) 13229a982250SKirill A. Shutemov deferred_split_huge_page(compound_head(page)); 13239a982250SKirill A. Shutemov 132416f8c5b2SHugh Dickins /* 13251da177e4SLinus Torvalds * It would be tidy to reset the PageAnon mapping here, 13261da177e4SLinus Torvalds * but that might overwrite a racing page_add_anon_rmap 13271da177e4SLinus Torvalds * which increments mapcount after us but sets mapping 13282d4894b5SMel Gorman * before us: so leave the reset to free_unref_page, 13291da177e4SLinus Torvalds * and remember that it's only reliable while mapped. 13301da177e4SLinus Torvalds * Leaving it set also helps swapoff to reinstate ptes 13311da177e4SLinus Torvalds * faster for those pages still in swapcache. 13321da177e4SLinus Torvalds */ 13331da177e4SLinus Torvalds } 13341da177e4SLinus Torvalds 13351da177e4SLinus Torvalds /* 133652629506SJoonsoo Kim * @arg: enum ttu_flags will be passed to this argument 13371da177e4SLinus Torvalds */ 1338e4b82222SMinchan Kim static bool try_to_unmap_one(struct page *page, struct vm_area_struct *vma, 133952629506SJoonsoo Kim unsigned long address, void *arg) 13401da177e4SLinus Torvalds { 13411da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 1342c7ab0d2fSKirill A. Shutemov struct page_vma_mapped_walk pvmw = { 1343c7ab0d2fSKirill A. Shutemov .page = page, 1344c7ab0d2fSKirill A. Shutemov .vma = vma, 1345c7ab0d2fSKirill A. Shutemov .address = address, 1346c7ab0d2fSKirill A. Shutemov }; 13471da177e4SLinus Torvalds pte_t pteval; 1348c7ab0d2fSKirill A. Shutemov struct page *subpage; 1349785373b4SLinus Torvalds bool ret = true; 1350ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 1351802a3a92SShaohua Li enum ttu_flags flags = (enum ttu_flags)arg; 13521da177e4SLinus Torvalds 1353b87537d9SHugh Dickins /* munlock has nothing to gain from examining un-locked vmas */ 1354b87537d9SHugh Dickins if ((flags & TTU_MUNLOCK) && !(vma->vm_flags & VM_LOCKED)) 1355e4b82222SMinchan Kim return true; 1356b87537d9SHugh Dickins 1357a5430ddaSJérôme Glisse if (IS_ENABLED(CONFIG_MIGRATION) && (flags & TTU_MIGRATION) && 1358a5430ddaSJérôme Glisse is_zone_device_page(page) && !is_device_private_page(page)) 1359a5430ddaSJérôme Glisse return true; 1360a5430ddaSJérôme Glisse 1361fec89c10SKirill A. Shutemov if (flags & TTU_SPLIT_HUGE_PMD) { 1362fec89c10SKirill A. Shutemov split_huge_pmd_address(vma, address, 1363b5ff8161SNaoya Horiguchi flags & TTU_SPLIT_FREEZE, page); 1364fec89c10SKirill A. Shutemov } 1365fec89c10SKirill A. Shutemov 1366369ea824SJérôme Glisse /* 1367017b1660SMike Kravetz * For THP, we have to assume the worse case ie pmd for invalidation. 1368017b1660SMike Kravetz * For hugetlb, it could be much worse if we need to do pud 1369017b1660SMike Kravetz * invalidation in the case of pmd sharing. 1370017b1660SMike Kravetz * 1371017b1660SMike Kravetz * Note that the page can not be free in this function as call of 1372017b1660SMike Kravetz * try_to_unmap() must hold a reference on the page. 1373369ea824SJérôme Glisse */ 1374ba422731SSean Christopherson mmu_notifier_range_init(&range, vma->vm_mm, address, 1375ba422731SSean Christopherson min(vma->vm_end, address + 1376ac46d4f3SJérôme Glisse (PAGE_SIZE << compound_order(page)))); 1377017b1660SMike Kravetz if (PageHuge(page)) { 1378017b1660SMike Kravetz /* 1379017b1660SMike Kravetz * If sharing is possible, start and end will be adjusted 1380017b1660SMike Kravetz * accordingly. 1381017b1660SMike Kravetz */ 1382ac46d4f3SJérôme Glisse adjust_range_if_pmd_sharing_possible(vma, &range.start, 1383ac46d4f3SJérôme Glisse &range.end); 1384017b1660SMike Kravetz } 1385ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 1386369ea824SJérôme Glisse 1387c7ab0d2fSKirill A. Shutemov while (page_vma_mapped_walk(&pvmw)) { 1388616b8371SZi Yan #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION 1389616b8371SZi Yan /* PMD-mapped THP migration entry */ 1390616b8371SZi Yan if (!pvmw.pte && (flags & TTU_MIGRATION)) { 1391616b8371SZi Yan VM_BUG_ON_PAGE(PageHuge(page) || !PageTransCompound(page), page); 1392616b8371SZi Yan 1393616b8371SZi Yan set_pmd_migration_entry(&pvmw, page); 1394616b8371SZi Yan continue; 1395616b8371SZi Yan } 1396616b8371SZi Yan #endif 1397616b8371SZi Yan 13981da177e4SLinus Torvalds /* 13991da177e4SLinus Torvalds * If the page is mlock()d, we cannot swap it out. 14001da177e4SLinus Torvalds * If it's recently referenced (perhaps page_referenced 14011da177e4SLinus Torvalds * skipped over this mm) then we should reactivate it. 14021da177e4SLinus Torvalds */ 140314fa31b8SAndi Kleen if (!(flags & TTU_IGNORE_MLOCK)) { 1404b87537d9SHugh Dickins if (vma->vm_flags & VM_LOCKED) { 14059a73f61bSKirill A. Shutemov /* PTE-mapped THP are never mlocked */ 14069a73f61bSKirill A. Shutemov if (!PageTransCompound(page)) { 14079a73f61bSKirill A. Shutemov /* 14089a73f61bSKirill A. Shutemov * Holding pte lock, we do *not* need 14099a73f61bSKirill A. Shutemov * mmap_sem here 14109a73f61bSKirill A. Shutemov */ 1411b87537d9SHugh Dickins mlock_vma_page(page); 14129a73f61bSKirill A. Shutemov } 1413e4b82222SMinchan Kim ret = false; 1414c7ab0d2fSKirill A. Shutemov page_vma_mapped_walk_done(&pvmw); 1415c7ab0d2fSKirill A. Shutemov break; 1416b87537d9SHugh Dickins } 1417daa5ba76SKonstantin Khlebnikov if (flags & TTU_MUNLOCK) 1418c7ab0d2fSKirill A. Shutemov continue; 141914fa31b8SAndi Kleen } 1420c7ab0d2fSKirill A. Shutemov 14218346242aSKirill A. Shutemov /* Unexpected PMD-mapped THP? */ 14228346242aSKirill A. Shutemov VM_BUG_ON_PAGE(!pvmw.pte, page); 14238346242aSKirill A. Shutemov 14248346242aSKirill A. Shutemov subpage = page - page_to_pfn(page) + pte_pfn(*pvmw.pte); 1425785373b4SLinus Torvalds address = pvmw.address; 1426785373b4SLinus Torvalds 1427017b1660SMike Kravetz if (PageHuge(page)) { 1428017b1660SMike Kravetz if (huge_pmd_unshare(mm, &address, pvmw.pte)) { 1429017b1660SMike Kravetz /* 1430017b1660SMike Kravetz * huge_pmd_unshare unmapped an entire PMD 1431017b1660SMike Kravetz * page. There is no way of knowing exactly 1432017b1660SMike Kravetz * which PMDs may be cached for this mm, so 1433017b1660SMike Kravetz * we must flush them all. start/end were 1434017b1660SMike Kravetz * already adjusted above to cover this range. 1435017b1660SMike Kravetz */ 1436ac46d4f3SJérôme Glisse flush_cache_range(vma, range.start, range.end); 1437ac46d4f3SJérôme Glisse flush_tlb_range(vma, range.start, range.end); 1438ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range(mm, range.start, 1439ac46d4f3SJérôme Glisse range.end); 1440017b1660SMike Kravetz 1441017b1660SMike Kravetz /* 1442017b1660SMike Kravetz * The ref count of the PMD page was dropped 1443017b1660SMike Kravetz * which is part of the way map counting 1444017b1660SMike Kravetz * is done for shared PMDs. Return 'true' 1445017b1660SMike Kravetz * here. When there is no other sharing, 1446017b1660SMike Kravetz * huge_pmd_unshare returns false and we will 1447017b1660SMike Kravetz * unmap the actual page and drop map count 1448017b1660SMike Kravetz * to zero. 1449017b1660SMike Kravetz */ 1450017b1660SMike Kravetz page_vma_mapped_walk_done(&pvmw); 1451017b1660SMike Kravetz break; 1452017b1660SMike Kravetz } 1453017b1660SMike Kravetz } 14548346242aSKirill A. Shutemov 1455a5430ddaSJérôme Glisse if (IS_ENABLED(CONFIG_MIGRATION) && 1456a5430ddaSJérôme Glisse (flags & TTU_MIGRATION) && 1457a5430ddaSJérôme Glisse is_zone_device_page(page)) { 1458a5430ddaSJérôme Glisse swp_entry_t entry; 1459a5430ddaSJérôme Glisse pte_t swp_pte; 1460a5430ddaSJérôme Glisse 1461a5430ddaSJérôme Glisse pteval = ptep_get_and_clear(mm, pvmw.address, pvmw.pte); 1462a5430ddaSJérôme Glisse 1463a5430ddaSJérôme Glisse /* 1464a5430ddaSJérôme Glisse * Store the pfn of the page in a special migration 1465a5430ddaSJérôme Glisse * pte. do_swap_page() will wait until the migration 1466a5430ddaSJérôme Glisse * pte is removed and then restart fault handling. 1467a5430ddaSJérôme Glisse */ 1468a5430ddaSJérôme Glisse entry = make_migration_entry(page, 0); 1469a5430ddaSJérôme Glisse swp_pte = swp_entry_to_pte(entry); 1470a5430ddaSJérôme Glisse if (pte_soft_dirty(pteval)) 1471a5430ddaSJérôme Glisse swp_pte = pte_swp_mksoft_dirty(swp_pte); 1472a5430ddaSJérôme Glisse set_pte_at(mm, pvmw.address, pvmw.pte, swp_pte); 14730f10851eSJérôme Glisse /* 14740f10851eSJérôme Glisse * No need to invalidate here it will synchronize on 14750f10851eSJérôme Glisse * against the special swap migration pte. 14760f10851eSJérôme Glisse */ 1477a5430ddaSJérôme Glisse goto discard; 1478a5430ddaSJérôme Glisse } 1479a5430ddaSJérôme Glisse 148014fa31b8SAndi Kleen if (!(flags & TTU_IGNORE_ACCESS)) { 1481785373b4SLinus Torvalds if (ptep_clear_flush_young_notify(vma, address, 1482c7ab0d2fSKirill A. Shutemov pvmw.pte)) { 1483e4b82222SMinchan Kim ret = false; 1484c7ab0d2fSKirill A. Shutemov page_vma_mapped_walk_done(&pvmw); 1485c7ab0d2fSKirill A. Shutemov break; 14861da177e4SLinus Torvalds } 1487b291f000SNick Piggin } 14881da177e4SLinus Torvalds 14891da177e4SLinus Torvalds /* Nuke the page table entry. */ 1490785373b4SLinus Torvalds flush_cache_page(vma, address, pte_pfn(*pvmw.pte)); 149172b252aeSMel Gorman if (should_defer_flush(mm, flags)) { 149272b252aeSMel Gorman /* 1493c7ab0d2fSKirill A. Shutemov * We clear the PTE but do not flush so potentially 1494c7ab0d2fSKirill A. Shutemov * a remote CPU could still be writing to the page. 1495c7ab0d2fSKirill A. Shutemov * If the entry was previously clean then the 1496c7ab0d2fSKirill A. Shutemov * architecture must guarantee that a clear->dirty 1497c7ab0d2fSKirill A. Shutemov * transition on a cached TLB entry is written through 1498c7ab0d2fSKirill A. Shutemov * and traps if the PTE is unmapped. 149972b252aeSMel Gorman */ 1500785373b4SLinus Torvalds pteval = ptep_get_and_clear(mm, address, pvmw.pte); 150172b252aeSMel Gorman 1502c7ab0d2fSKirill A. Shutemov set_tlb_ubc_flush_pending(mm, pte_dirty(pteval)); 150372b252aeSMel Gorman } else { 1504785373b4SLinus Torvalds pteval = ptep_clear_flush(vma, address, pvmw.pte); 150572b252aeSMel Gorman } 15061da177e4SLinus Torvalds 1507c7ab0d2fSKirill A. Shutemov /* Move the dirty bit to the page. Now the pte is gone. */ 15081da177e4SLinus Torvalds if (pte_dirty(pteval)) 15091da177e4SLinus Torvalds set_page_dirty(page); 15101da177e4SLinus Torvalds 1511365e9c87SHugh Dickins /* Update high watermark before we lower rss */ 1512365e9c87SHugh Dickins update_hiwater_rss(mm); 1513365e9c87SHugh Dickins 1514888b9f7cSAndi Kleen if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) { 15155fd27b8eSPunit Agrawal pteval = swp_entry_to_pte(make_hwpoison_entry(subpage)); 15165d317b2bSNaoya Horiguchi if (PageHuge(page)) { 1517c7ab0d2fSKirill A. Shutemov int nr = 1 << compound_order(page); 1518c7ab0d2fSKirill A. Shutemov hugetlb_count_sub(nr, mm); 1519785373b4SLinus Torvalds set_huge_swap_pte_at(mm, address, 15205fd27b8eSPunit Agrawal pvmw.pte, pteval, 15215fd27b8eSPunit Agrawal vma_mmu_pagesize(vma)); 15225d317b2bSNaoya Horiguchi } else { 1523eca56ff9SJerome Marchand dec_mm_counter(mm, mm_counter(page)); 1524785373b4SLinus Torvalds set_pte_at(mm, address, pvmw.pte, pteval); 15255f24ae58SNaoya Horiguchi } 1526c7ab0d2fSKirill A. Shutemov 1527bce73e48SChristian Borntraeger } else if (pte_unused(pteval) && !userfaultfd_armed(vma)) { 152845961722SKonstantin Weitz /* 152945961722SKonstantin Weitz * The guest indicated that the page content is of no 153045961722SKonstantin Weitz * interest anymore. Simply discard the pte, vmscan 153145961722SKonstantin Weitz * will take care of the rest. 1532bce73e48SChristian Borntraeger * A future reference will then fault in a new zero 1533bce73e48SChristian Borntraeger * page. When userfaultfd is active, we must not drop 1534bce73e48SChristian Borntraeger * this page though, as its main user (postcopy 1535bce73e48SChristian Borntraeger * migration) will not expect userfaults on already 1536bce73e48SChristian Borntraeger * copied pages. 153745961722SKonstantin Weitz */ 1538eca56ff9SJerome Marchand dec_mm_counter(mm, mm_counter(page)); 15390f10851eSJérôme Glisse /* We have to invalidate as we cleared the pte */ 15400f10851eSJérôme Glisse mmu_notifier_invalidate_range(mm, address, 15410f10851eSJérôme Glisse address + PAGE_SIZE); 1542c7ab0d2fSKirill A. Shutemov } else if (IS_ENABLED(CONFIG_MIGRATION) && 1543b5ff8161SNaoya Horiguchi (flags & (TTU_MIGRATION|TTU_SPLIT_FREEZE))) { 1544470f119fSHugh Dickins swp_entry_t entry; 1545470f119fSHugh Dickins pte_t swp_pte; 1546ca827d55SKhalid Aziz 1547ca827d55SKhalid Aziz if (arch_unmap_one(mm, vma, address, pteval) < 0) { 1548ca827d55SKhalid Aziz set_pte_at(mm, address, pvmw.pte, pteval); 1549ca827d55SKhalid Aziz ret = false; 1550ca827d55SKhalid Aziz page_vma_mapped_walk_done(&pvmw); 1551ca827d55SKhalid Aziz break; 1552ca827d55SKhalid Aziz } 1553ca827d55SKhalid Aziz 1554470f119fSHugh Dickins /* 1555470f119fSHugh Dickins * Store the pfn of the page in a special migration 1556470f119fSHugh Dickins * pte. do_swap_page() will wait until the migration 1557470f119fSHugh Dickins * pte is removed and then restart fault handling. 1558470f119fSHugh Dickins */ 1559c7ab0d2fSKirill A. Shutemov entry = make_migration_entry(subpage, 1560c7ab0d2fSKirill A. Shutemov pte_write(pteval)); 1561470f119fSHugh Dickins swp_pte = swp_entry_to_pte(entry); 1562470f119fSHugh Dickins if (pte_soft_dirty(pteval)) 1563470f119fSHugh Dickins swp_pte = pte_swp_mksoft_dirty(swp_pte); 1564785373b4SLinus Torvalds set_pte_at(mm, address, pvmw.pte, swp_pte); 15650f10851eSJérôme Glisse /* 15660f10851eSJérôme Glisse * No need to invalidate here it will synchronize on 15670f10851eSJérôme Glisse * against the special swap migration pte. 15680f10851eSJérôme Glisse */ 1569888b9f7cSAndi Kleen } else if (PageAnon(page)) { 1570c7ab0d2fSKirill A. Shutemov swp_entry_t entry = { .val = page_private(subpage) }; 1571179ef71cSCyrill Gorcunov pte_t swp_pte; 15721da177e4SLinus Torvalds /* 15731da177e4SLinus Torvalds * Store the swap location in the pte. 15741da177e4SLinus Torvalds * See handle_pte_fault() ... 15751da177e4SLinus Torvalds */ 1576eb94a878SMinchan Kim if (unlikely(PageSwapBacked(page) != PageSwapCache(page))) { 1577eb94a878SMinchan Kim WARN_ON_ONCE(1); 157883612a94SMinchan Kim ret = false; 1579369ea824SJérôme Glisse /* We have to invalidate as we cleared the pte */ 15800f10851eSJérôme Glisse mmu_notifier_invalidate_range(mm, address, 15810f10851eSJérôme Glisse address + PAGE_SIZE); 1582eb94a878SMinchan Kim page_vma_mapped_walk_done(&pvmw); 1583eb94a878SMinchan Kim break; 1584eb94a878SMinchan Kim } 1585854e9ed0SMinchan Kim 1586802a3a92SShaohua Li /* MADV_FREE page check */ 1587802a3a92SShaohua Li if (!PageSwapBacked(page)) { 1588a128ca71SShaohua Li if (!PageDirty(page)) { 15890f10851eSJérôme Glisse /* Invalidate as we cleared the pte */ 15900f10851eSJérôme Glisse mmu_notifier_invalidate_range(mm, 15910f10851eSJérôme Glisse address, address + PAGE_SIZE); 1592854e9ed0SMinchan Kim dec_mm_counter(mm, MM_ANONPAGES); 1593854e9ed0SMinchan Kim goto discard; 1594854e9ed0SMinchan Kim } 1595854e9ed0SMinchan Kim 1596802a3a92SShaohua Li /* 1597802a3a92SShaohua Li * If the page was redirtied, it cannot be 1598802a3a92SShaohua Li * discarded. Remap the page to page table. 1599802a3a92SShaohua Li */ 1600785373b4SLinus Torvalds set_pte_at(mm, address, pvmw.pte, pteval); 160118863d3aSMinchan Kim SetPageSwapBacked(page); 1602e4b82222SMinchan Kim ret = false; 1603802a3a92SShaohua Li page_vma_mapped_walk_done(&pvmw); 1604802a3a92SShaohua Li break; 1605802a3a92SShaohua Li } 1606802a3a92SShaohua Li 1607570a335bSHugh Dickins if (swap_duplicate(entry) < 0) { 1608785373b4SLinus Torvalds set_pte_at(mm, address, pvmw.pte, pteval); 1609e4b82222SMinchan Kim ret = false; 1610c7ab0d2fSKirill A. Shutemov page_vma_mapped_walk_done(&pvmw); 1611c7ab0d2fSKirill A. Shutemov break; 1612570a335bSHugh Dickins } 1613ca827d55SKhalid Aziz if (arch_unmap_one(mm, vma, address, pteval) < 0) { 1614ca827d55SKhalid Aziz set_pte_at(mm, address, pvmw.pte, pteval); 1615ca827d55SKhalid Aziz ret = false; 1616ca827d55SKhalid Aziz page_vma_mapped_walk_done(&pvmw); 1617ca827d55SKhalid Aziz break; 1618ca827d55SKhalid Aziz } 16191da177e4SLinus Torvalds if (list_empty(&mm->mmlist)) { 16201da177e4SLinus Torvalds spin_lock(&mmlist_lock); 1621f412ac08SHugh Dickins if (list_empty(&mm->mmlist)) 16221da177e4SLinus Torvalds list_add(&mm->mmlist, &init_mm.mmlist); 16231da177e4SLinus Torvalds spin_unlock(&mmlist_lock); 16241da177e4SLinus Torvalds } 1625d559db08SKAMEZAWA Hiroyuki dec_mm_counter(mm, MM_ANONPAGES); 1626b084d435SKAMEZAWA Hiroyuki inc_mm_counter(mm, MM_SWAPENTS); 1627179ef71cSCyrill Gorcunov swp_pte = swp_entry_to_pte(entry); 1628179ef71cSCyrill Gorcunov if (pte_soft_dirty(pteval)) 1629179ef71cSCyrill Gorcunov swp_pte = pte_swp_mksoft_dirty(swp_pte); 1630785373b4SLinus Torvalds set_pte_at(mm, address, pvmw.pte, swp_pte); 16310f10851eSJérôme Glisse /* Invalidate as we cleared the pte */ 1632369ea824SJérôme Glisse mmu_notifier_invalidate_range(mm, address, 1633369ea824SJérôme Glisse address + PAGE_SIZE); 16340f10851eSJérôme Glisse } else { 16350f10851eSJérôme Glisse /* 1636906f9cdfSHugh Dickins * This is a locked file-backed page, thus it cannot 1637906f9cdfSHugh Dickins * be removed from the page cache and replaced by a new 1638906f9cdfSHugh Dickins * page before mmu_notifier_invalidate_range_end, so no 16390f10851eSJérôme Glisse * concurrent thread might update its page table to 16400f10851eSJérôme Glisse * point at new page while a device still is using this 16410f10851eSJérôme Glisse * page. 16420f10851eSJérôme Glisse * 1643ad56b738SMike Rapoport * See Documentation/vm/mmu_notifier.rst 16440f10851eSJérôme Glisse */ 16450f10851eSJérôme Glisse dec_mm_counter(mm, mm_counter_file(page)); 16460f10851eSJérôme Glisse } 16470f10851eSJérôme Glisse discard: 16480f10851eSJérôme Glisse /* 16490f10851eSJérôme Glisse * No need to call mmu_notifier_invalidate_range() it has be 16500f10851eSJérôme Glisse * done above for all cases requiring it to happen under page 16510f10851eSJérôme Glisse * table lock before mmu_notifier_invalidate_range_end() 16520f10851eSJérôme Glisse * 1653ad56b738SMike Rapoport * See Documentation/vm/mmu_notifier.rst 16540f10851eSJérôme Glisse */ 16550f10851eSJérôme Glisse page_remove_rmap(subpage, PageHuge(page)); 16560f10851eSJérôme Glisse put_page(page); 1657c7ab0d2fSKirill A. Shutemov } 1658369ea824SJérôme Glisse 1659ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(&range); 1660369ea824SJérôme Glisse 1661caed0f48SKOSAKI Motohiro return ret; 16621da177e4SLinus Torvalds } 16631da177e4SLinus Torvalds 166471e3aac0SAndrea Arcangeli bool is_vma_temporary_stack(struct vm_area_struct *vma) 1665a8bef8ffSMel Gorman { 1666a8bef8ffSMel Gorman int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP); 1667a8bef8ffSMel Gorman 1668a8bef8ffSMel Gorman if (!maybe_stack) 1669a8bef8ffSMel Gorman return false; 1670a8bef8ffSMel Gorman 1671a8bef8ffSMel Gorman if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) == 1672a8bef8ffSMel Gorman VM_STACK_INCOMPLETE_SETUP) 1673a8bef8ffSMel Gorman return true; 1674a8bef8ffSMel Gorman 1675a8bef8ffSMel Gorman return false; 1676a8bef8ffSMel Gorman } 1677a8bef8ffSMel Gorman 167852629506SJoonsoo Kim static bool invalid_migration_vma(struct vm_area_struct *vma, void *arg) 167952629506SJoonsoo Kim { 168052629506SJoonsoo Kim return is_vma_temporary_stack(vma); 168152629506SJoonsoo Kim } 168252629506SJoonsoo Kim 16832a52bcbcSKirill A. Shutemov static int page_mapcount_is_zero(struct page *page) 168452629506SJoonsoo Kim { 1685c7ab0d2fSKirill A. Shutemov return !total_mapcount(page); 16862a52bcbcSKirill A. Shutemov } 168752629506SJoonsoo Kim 16881da177e4SLinus Torvalds /** 16891da177e4SLinus Torvalds * try_to_unmap - try to remove all page table mappings to a page 16901da177e4SLinus Torvalds * @page: the page to get unmapped 169114fa31b8SAndi Kleen * @flags: action and flags 16921da177e4SLinus Torvalds * 16931da177e4SLinus Torvalds * Tries to remove all the page table entries which are mapping this 16941da177e4SLinus Torvalds * page, used in the pageout path. Caller must hold the page lock. 16951da177e4SLinus Torvalds * 1696666e5a40SMinchan Kim * If unmap is successful, return true. Otherwise, false. 16971da177e4SLinus Torvalds */ 1698666e5a40SMinchan Kim bool try_to_unmap(struct page *page, enum ttu_flags flags) 16991da177e4SLinus Torvalds { 170052629506SJoonsoo Kim struct rmap_walk_control rwc = { 170152629506SJoonsoo Kim .rmap_one = try_to_unmap_one, 1702802a3a92SShaohua Li .arg = (void *)flags, 17032a52bcbcSKirill A. Shutemov .done = page_mapcount_is_zero, 170452629506SJoonsoo Kim .anon_lock = page_lock_anon_vma_read, 170552629506SJoonsoo Kim }; 17061da177e4SLinus Torvalds 170752629506SJoonsoo Kim /* 170852629506SJoonsoo Kim * During exec, a temporary VMA is setup and later moved. 170952629506SJoonsoo Kim * The VMA is moved under the anon_vma lock but not the 171052629506SJoonsoo Kim * page tables leading to a race where migration cannot 171152629506SJoonsoo Kim * find the migration ptes. Rather than increasing the 171252629506SJoonsoo Kim * locking requirements of exec(), migration skips 171352629506SJoonsoo Kim * temporary VMAs until after exec() completes. 171452629506SJoonsoo Kim */ 1715b5ff8161SNaoya Horiguchi if ((flags & (TTU_MIGRATION|TTU_SPLIT_FREEZE)) 1716b5ff8161SNaoya Horiguchi && !PageKsm(page) && PageAnon(page)) 171752629506SJoonsoo Kim rwc.invalid_vma = invalid_migration_vma; 171852629506SJoonsoo Kim 17192a52bcbcSKirill A. Shutemov if (flags & TTU_RMAP_LOCKED) 172033fc80e2SMinchan Kim rmap_walk_locked(page, &rwc); 17212a52bcbcSKirill A. Shutemov else 172233fc80e2SMinchan Kim rmap_walk(page, &rwc); 172352629506SJoonsoo Kim 1724666e5a40SMinchan Kim return !page_mapcount(page) ? true : false; 17251da177e4SLinus Torvalds } 172681b4082dSNikita Danilov 17272a52bcbcSKirill A. Shutemov static int page_not_mapped(struct page *page) 17282a52bcbcSKirill A. Shutemov { 17292a52bcbcSKirill A. Shutemov return !page_mapped(page); 17302a52bcbcSKirill A. Shutemov }; 17312a52bcbcSKirill A. Shutemov 1732b291f000SNick Piggin /** 1733b291f000SNick Piggin * try_to_munlock - try to munlock a page 1734b291f000SNick Piggin * @page: the page to be munlocked 1735b291f000SNick Piggin * 1736b291f000SNick Piggin * Called from munlock code. Checks all of the VMAs mapping the page 1737b291f000SNick Piggin * to make sure nobody else has this page mlocked. The page will be 1738b291f000SNick Piggin * returned with PG_mlocked cleared if no other vmas have it mlocked. 1739b291f000SNick Piggin */ 1740854e9ed0SMinchan Kim 1741192d7232SMinchan Kim void try_to_munlock(struct page *page) 1742192d7232SMinchan Kim { 1743e8351ac9SJoonsoo Kim struct rmap_walk_control rwc = { 1744e8351ac9SJoonsoo Kim .rmap_one = try_to_unmap_one, 1745802a3a92SShaohua Li .arg = (void *)TTU_MUNLOCK, 1746e8351ac9SJoonsoo Kim .done = page_not_mapped, 1747e8351ac9SJoonsoo Kim .anon_lock = page_lock_anon_vma_read, 1748e8351ac9SJoonsoo Kim 1749e8351ac9SJoonsoo Kim }; 1750e8351ac9SJoonsoo Kim 1751309381feSSasha Levin VM_BUG_ON_PAGE(!PageLocked(page) || PageLRU(page), page); 1752192d7232SMinchan Kim VM_BUG_ON_PAGE(PageCompound(page) && PageDoubleMap(page), page); 1753b291f000SNick Piggin 1754192d7232SMinchan Kim rmap_walk(page, &rwc); 1755b291f000SNick Piggin } 1756e9995ef9SHugh Dickins 175701d8b20dSPeter Zijlstra void __put_anon_vma(struct anon_vma *anon_vma) 175876545066SRik van Riel { 175976545066SRik van Riel struct anon_vma *root = anon_vma->root; 176076545066SRik van Riel 1761624483f3SAndrey Ryabinin anon_vma_free(anon_vma); 176201d8b20dSPeter Zijlstra if (root != anon_vma && atomic_dec_and_test(&root->refcount)) 176376545066SRik van Riel anon_vma_free(root); 176476545066SRik van Riel } 176576545066SRik van Riel 17660dd1c7bbSJoonsoo Kim static struct anon_vma *rmap_walk_anon_lock(struct page *page, 17670dd1c7bbSJoonsoo Kim struct rmap_walk_control *rwc) 1768faecd8ddSJoonsoo Kim { 1769faecd8ddSJoonsoo Kim struct anon_vma *anon_vma; 1770faecd8ddSJoonsoo Kim 17710dd1c7bbSJoonsoo Kim if (rwc->anon_lock) 17720dd1c7bbSJoonsoo Kim return rwc->anon_lock(page); 17730dd1c7bbSJoonsoo Kim 1774faecd8ddSJoonsoo Kim /* 1775faecd8ddSJoonsoo Kim * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read() 1776faecd8ddSJoonsoo Kim * because that depends on page_mapped(); but not all its usages 1777faecd8ddSJoonsoo Kim * are holding mmap_sem. Users without mmap_sem are required to 1778faecd8ddSJoonsoo Kim * take a reference count to prevent the anon_vma disappearing 1779faecd8ddSJoonsoo Kim */ 1780faecd8ddSJoonsoo Kim anon_vma = page_anon_vma(page); 1781faecd8ddSJoonsoo Kim if (!anon_vma) 1782faecd8ddSJoonsoo Kim return NULL; 1783faecd8ddSJoonsoo Kim 1784faecd8ddSJoonsoo Kim anon_vma_lock_read(anon_vma); 1785faecd8ddSJoonsoo Kim return anon_vma; 1786faecd8ddSJoonsoo Kim } 1787faecd8ddSJoonsoo Kim 1788e9995ef9SHugh Dickins /* 1789e8351ac9SJoonsoo Kim * rmap_walk_anon - do something to anonymous page using the object-based 1790e8351ac9SJoonsoo Kim * rmap method 1791e8351ac9SJoonsoo Kim * @page: the page to be handled 1792e8351ac9SJoonsoo Kim * @rwc: control variable according to each walk type 1793e8351ac9SJoonsoo Kim * 1794e8351ac9SJoonsoo Kim * Find all the mappings of a page using the mapping pointer and the vma chains 1795e8351ac9SJoonsoo Kim * contained in the anon_vma struct it points to. 1796e8351ac9SJoonsoo Kim * 1797e8351ac9SJoonsoo Kim * When called from try_to_munlock(), the mmap_sem of the mm containing the vma 1798e8351ac9SJoonsoo Kim * where the page was found will be held for write. So, we won't recheck 1799e8351ac9SJoonsoo Kim * vm_flags for that VMA. That should be OK, because that vma shouldn't be 1800e8351ac9SJoonsoo Kim * LOCKED. 1801e9995ef9SHugh Dickins */ 18021df631aeSMinchan Kim static void rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc, 1803b9773199SKirill A. Shutemov bool locked) 1804e9995ef9SHugh Dickins { 1805e9995ef9SHugh Dickins struct anon_vma *anon_vma; 1806a8fa41adSKirill A. Shutemov pgoff_t pgoff_start, pgoff_end; 18075beb4930SRik van Riel struct anon_vma_chain *avc; 1808e9995ef9SHugh Dickins 1809b9773199SKirill A. Shutemov if (locked) { 1810b9773199SKirill A. Shutemov anon_vma = page_anon_vma(page); 1811b9773199SKirill A. Shutemov /* anon_vma disappear under us? */ 1812b9773199SKirill A. Shutemov VM_BUG_ON_PAGE(!anon_vma, page); 1813b9773199SKirill A. Shutemov } else { 18140dd1c7bbSJoonsoo Kim anon_vma = rmap_walk_anon_lock(page, rwc); 1815b9773199SKirill A. Shutemov } 1816e9995ef9SHugh Dickins if (!anon_vma) 18171df631aeSMinchan Kim return; 1818faecd8ddSJoonsoo Kim 1819a8fa41adSKirill A. Shutemov pgoff_start = page_to_pgoff(page); 1820a8fa41adSKirill A. Shutemov pgoff_end = pgoff_start + hpage_nr_pages(page) - 1; 1821a8fa41adSKirill A. Shutemov anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, 1822a8fa41adSKirill A. Shutemov pgoff_start, pgoff_end) { 18235beb4930SRik van Riel struct vm_area_struct *vma = avc->vma; 1824e9995ef9SHugh Dickins unsigned long address = vma_address(page, vma); 18250dd1c7bbSJoonsoo Kim 1826ad12695fSAndrea Arcangeli cond_resched(); 1827ad12695fSAndrea Arcangeli 18280dd1c7bbSJoonsoo Kim if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) 18290dd1c7bbSJoonsoo Kim continue; 18300dd1c7bbSJoonsoo Kim 1831e4b82222SMinchan Kim if (!rwc->rmap_one(page, vma, address, rwc->arg)) 1832e9995ef9SHugh Dickins break; 18330dd1c7bbSJoonsoo Kim if (rwc->done && rwc->done(page)) 18340dd1c7bbSJoonsoo Kim break; 1835e9995ef9SHugh Dickins } 1836b9773199SKirill A. Shutemov 1837b9773199SKirill A. Shutemov if (!locked) 18384fc3f1d6SIngo Molnar anon_vma_unlock_read(anon_vma); 1839e9995ef9SHugh Dickins } 1840e9995ef9SHugh Dickins 1841e8351ac9SJoonsoo Kim /* 1842e8351ac9SJoonsoo Kim * rmap_walk_file - do something to file page using the object-based rmap method 1843e8351ac9SJoonsoo Kim * @page: the page to be handled 1844e8351ac9SJoonsoo Kim * @rwc: control variable according to each walk type 1845e8351ac9SJoonsoo Kim * 1846e8351ac9SJoonsoo Kim * Find all the mappings of a page using the mapping pointer and the vma chains 1847e8351ac9SJoonsoo Kim * contained in the address_space struct it points to. 1848e8351ac9SJoonsoo Kim * 1849e8351ac9SJoonsoo Kim * When called from try_to_munlock(), the mmap_sem of the mm containing the vma 1850e8351ac9SJoonsoo Kim * where the page was found will be held for write. So, we won't recheck 1851e8351ac9SJoonsoo Kim * vm_flags for that VMA. That should be OK, because that vma shouldn't be 1852e8351ac9SJoonsoo Kim * LOCKED. 1853e8351ac9SJoonsoo Kim */ 18541df631aeSMinchan Kim static void rmap_walk_file(struct page *page, struct rmap_walk_control *rwc, 1855b9773199SKirill A. Shutemov bool locked) 1856e9995ef9SHugh Dickins { 1857b9773199SKirill A. Shutemov struct address_space *mapping = page_mapping(page); 1858a8fa41adSKirill A. Shutemov pgoff_t pgoff_start, pgoff_end; 1859e9995ef9SHugh Dickins struct vm_area_struct *vma; 1860e9995ef9SHugh Dickins 18619f32624bSJoonsoo Kim /* 18629f32624bSJoonsoo Kim * The page lock not only makes sure that page->mapping cannot 18639f32624bSJoonsoo Kim * suddenly be NULLified by truncation, it makes sure that the 18649f32624bSJoonsoo Kim * structure at mapping cannot be freed and reused yet, 1865c8c06efaSDavidlohr Bueso * so we can safely take mapping->i_mmap_rwsem. 18669f32624bSJoonsoo Kim */ 186781d1b09cSSasha Levin VM_BUG_ON_PAGE(!PageLocked(page), page); 18689f32624bSJoonsoo Kim 1869e9995ef9SHugh Dickins if (!mapping) 18701df631aeSMinchan Kim return; 18713dec0ba0SDavidlohr Bueso 1872a8fa41adSKirill A. Shutemov pgoff_start = page_to_pgoff(page); 1873a8fa41adSKirill A. Shutemov pgoff_end = pgoff_start + hpage_nr_pages(page) - 1; 1874b9773199SKirill A. Shutemov if (!locked) 18753dec0ba0SDavidlohr Bueso i_mmap_lock_read(mapping); 1876a8fa41adSKirill A. Shutemov vma_interval_tree_foreach(vma, &mapping->i_mmap, 1877a8fa41adSKirill A. Shutemov pgoff_start, pgoff_end) { 1878e9995ef9SHugh Dickins unsigned long address = vma_address(page, vma); 18790dd1c7bbSJoonsoo Kim 1880ad12695fSAndrea Arcangeli cond_resched(); 1881ad12695fSAndrea Arcangeli 18820dd1c7bbSJoonsoo Kim if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg)) 18830dd1c7bbSJoonsoo Kim continue; 18840dd1c7bbSJoonsoo Kim 1885e4b82222SMinchan Kim if (!rwc->rmap_one(page, vma, address, rwc->arg)) 18860dd1c7bbSJoonsoo Kim goto done; 18870dd1c7bbSJoonsoo Kim if (rwc->done && rwc->done(page)) 18880dd1c7bbSJoonsoo Kim goto done; 1889e9995ef9SHugh Dickins } 18900dd1c7bbSJoonsoo Kim 18910dd1c7bbSJoonsoo Kim done: 1892b9773199SKirill A. Shutemov if (!locked) 18933dec0ba0SDavidlohr Bueso i_mmap_unlock_read(mapping); 1894e9995ef9SHugh Dickins } 1895e9995ef9SHugh Dickins 18961df631aeSMinchan Kim void rmap_walk(struct page *page, struct rmap_walk_control *rwc) 1897e9995ef9SHugh Dickins { 1898e9995ef9SHugh Dickins if (unlikely(PageKsm(page))) 18991df631aeSMinchan Kim rmap_walk_ksm(page, rwc); 1900e9995ef9SHugh Dickins else if (PageAnon(page)) 19011df631aeSMinchan Kim rmap_walk_anon(page, rwc, false); 1902e9995ef9SHugh Dickins else 19031df631aeSMinchan Kim rmap_walk_file(page, rwc, false); 1904b9773199SKirill A. Shutemov } 1905b9773199SKirill A. Shutemov 1906b9773199SKirill A. Shutemov /* Like rmap_walk, but caller holds relevant rmap lock */ 19071df631aeSMinchan Kim void rmap_walk_locked(struct page *page, struct rmap_walk_control *rwc) 1908b9773199SKirill A. Shutemov { 1909b9773199SKirill A. Shutemov /* no ksm support for now */ 1910b9773199SKirill A. Shutemov VM_BUG_ON_PAGE(PageKsm(page), page); 1911b9773199SKirill A. Shutemov if (PageAnon(page)) 19121df631aeSMinchan Kim rmap_walk_anon(page, rwc, true); 1913b9773199SKirill A. Shutemov else 19141df631aeSMinchan Kim rmap_walk_file(page, rwc, true); 1915e9995ef9SHugh Dickins } 19160fe6e20bSNaoya Horiguchi 1917e3390f67SNaoya Horiguchi #ifdef CONFIG_HUGETLB_PAGE 19180fe6e20bSNaoya Horiguchi /* 1919451b9514SKirill Tkhai * The following two functions are for anonymous (private mapped) hugepages. 19200fe6e20bSNaoya Horiguchi * Unlike common anonymous pages, anonymous hugepages have no accounting code 19210fe6e20bSNaoya Horiguchi * and no lru code, because we handle hugepages differently from common pages. 19220fe6e20bSNaoya Horiguchi */ 19230fe6e20bSNaoya Horiguchi void hugepage_add_anon_rmap(struct page *page, 19240fe6e20bSNaoya Horiguchi struct vm_area_struct *vma, unsigned long address) 19250fe6e20bSNaoya Horiguchi { 19260fe6e20bSNaoya Horiguchi struct anon_vma *anon_vma = vma->anon_vma; 19270fe6e20bSNaoya Horiguchi int first; 1928a850ea30SNaoya Horiguchi 1929a850ea30SNaoya Horiguchi BUG_ON(!PageLocked(page)); 19300fe6e20bSNaoya Horiguchi BUG_ON(!anon_vma); 19315dbe0af4SHugh Dickins /* address might be in next vma when migration races vma_adjust */ 193253f9263bSKirill A. Shutemov first = atomic_inc_and_test(compound_mapcount_ptr(page)); 19330fe6e20bSNaoya Horiguchi if (first) 1934451b9514SKirill Tkhai __page_set_anon_rmap(page, vma, address, 0); 19350fe6e20bSNaoya Horiguchi } 19360fe6e20bSNaoya Horiguchi 19370fe6e20bSNaoya Horiguchi void hugepage_add_new_anon_rmap(struct page *page, 19380fe6e20bSNaoya Horiguchi struct vm_area_struct *vma, unsigned long address) 19390fe6e20bSNaoya Horiguchi { 19400fe6e20bSNaoya Horiguchi BUG_ON(address < vma->vm_start || address >= vma->vm_end); 194153f9263bSKirill A. Shutemov atomic_set(compound_mapcount_ptr(page), 0); 1942451b9514SKirill Tkhai __page_set_anon_rmap(page, vma, address, 1); 19430fe6e20bSNaoya Horiguchi } 1944e3390f67SNaoya Horiguchi #endif /* CONFIG_HUGETLB_PAGE */ 1945