Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * mm/rmap.c - physical to virtual reverse mappings |
| 3 | * |
| 4 | * Copyright 2001, Rik van Riel <riel@conectiva.com.br> |
| 5 | * Released under the General Public License (GPL). |
| 6 | * |
| 7 | * Simple, low overhead reverse mapping scheme. |
| 8 | * Please try to keep this thing as modular as possible. |
| 9 | * |
| 10 | * Provides methods for unmapping each kind of mapped page: |
| 11 | * the anon methods track anonymous pages, and |
| 12 | * the file methods track pages belonging to an inode. |
| 13 | * |
| 14 | * Original design by Rik van Riel <riel@conectiva.com.br> 2001 |
| 15 | * File methods by Dave McCracken <dmccr@us.ibm.com> 2003, 2004 |
| 16 | * Anonymous methods by Andrea Arcangeli <andrea@suse.de> 2004 |
Hugh Dickins | 98f3260 | 2009-05-21 20:33:58 +0100 | [diff] [blame] | 17 | * Contributions by Hugh Dickins 2003, 2004 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | */ |
| 19 | |
| 20 | /* |
| 21 | * Lock ordering in mm: |
| 22 | * |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 23 | * inode->i_mutex (while writing or truncating, not reading or faulting) |
Nick Piggin | 82591e6 | 2006-10-19 23:29:10 -0700 | [diff] [blame] | 24 | * inode->i_alloc_sem (vmtruncate_range) |
| 25 | * mm->mmap_sem |
| 26 | * page->flags PG_locked (lock_page) |
Peter Zijlstra | 3d48ae4 | 2011-05-24 17:12:06 -0700 | [diff] [blame] | 27 | * mapping->i_mmap_mutex |
Peter Zijlstra | 2b575eb | 2011-05-24 17:12:11 -0700 | [diff] [blame] | 28 | * anon_vma->mutex |
Nick Piggin | 82591e6 | 2006-10-19 23:29:10 -0700 | [diff] [blame] | 29 | * mm->page_table_lock or pte_lock |
| 30 | * zone->lru_lock (in mark_page_accessed, isolate_lru_page) |
| 31 | * swap_lock (in swap_duplicate, swap_info_get) |
| 32 | * mmlist_lock (in mmput, drain_mmlist and others) |
| 33 | * mapping->private_lock (in __set_page_dirty_buffers) |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 34 | * inode->i_lock (in set_page_dirty's __mark_inode_dirty) |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 35 | * inode_wb_list_lock (in set_page_dirty's __mark_inode_dirty) |
Nick Piggin | 82591e6 | 2006-10-19 23:29:10 -0700 | [diff] [blame] | 36 | * sb_lock (within inode_lock in fs/fs-writeback.c) |
| 37 | * mapping->tree_lock (widely used, in set_page_dirty, |
| 38 | * in arch-dependent flush_dcache_mmap_lock, |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 39 | * within inode_wb_list_lock in __sync_single_inode) |
Andi Kleen | 6a46079 | 2009-09-16 11:50:15 +0200 | [diff] [blame] | 40 | * |
| 41 | * (code doesn't rely on that order so it could be switched around) |
| 42 | * ->tasklist_lock |
Peter Zijlstra | 2b575eb | 2011-05-24 17:12:11 -0700 | [diff] [blame] | 43 | * anon_vma->mutex (memory_failure, collect_procs_anon) |
Andi Kleen | 6a46079 | 2009-09-16 11:50:15 +0200 | [diff] [blame] | 44 | * pte map lock |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 45 | */ |
| 46 | |
| 47 | #include <linux/mm.h> |
| 48 | #include <linux/pagemap.h> |
| 49 | #include <linux/swap.h> |
| 50 | #include <linux/swapops.h> |
| 51 | #include <linux/slab.h> |
| 52 | #include <linux/init.h> |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 53 | #include <linux/ksm.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 54 | #include <linux/rmap.h> |
| 55 | #include <linux/rcupdate.h> |
Christoph Lameter | a48d07a | 2006-02-01 03:05:38 -0800 | [diff] [blame] | 56 | #include <linux/module.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 57 | #include <linux/memcontrol.h> |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 58 | #include <linux/mmu_notifier.h> |
KOSAKI Motohiro | 64cdd54 | 2009-01-06 14:39:16 -0800 | [diff] [blame] | 59 | #include <linux/migrate.h> |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 60 | #include <linux/hugetlb.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 61 | |
| 62 | #include <asm/tlbflush.h> |
| 63 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 64 | #include "internal.h" |
| 65 | |
Adrian Bunk | fdd2e5f | 2008-10-18 20:28:38 -0700 | [diff] [blame] | 66 | static struct kmem_cache *anon_vma_cachep; |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 67 | static struct kmem_cache *anon_vma_chain_cachep; |
Adrian Bunk | fdd2e5f | 2008-10-18 20:28:38 -0700 | [diff] [blame] | 68 | |
| 69 | static inline struct anon_vma *anon_vma_alloc(void) |
| 70 | { |
Peter Zijlstra | 01d8b20 | 2011-03-22 16:32:49 -0700 | [diff] [blame] | 71 | struct anon_vma *anon_vma; |
| 72 | |
| 73 | anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL); |
| 74 | if (anon_vma) { |
| 75 | atomic_set(&anon_vma->refcount, 1); |
| 76 | /* |
| 77 | * Initialise the anon_vma root to point to itself. If called |
| 78 | * from fork, the root will be reset to the parents anon_vma. |
| 79 | */ |
| 80 | anon_vma->root = anon_vma; |
| 81 | } |
| 82 | |
| 83 | return anon_vma; |
Adrian Bunk | fdd2e5f | 2008-10-18 20:28:38 -0700 | [diff] [blame] | 84 | } |
| 85 | |
Peter Zijlstra | 01d8b20 | 2011-03-22 16:32:49 -0700 | [diff] [blame] | 86 | static inline void anon_vma_free(struct anon_vma *anon_vma) |
Adrian Bunk | fdd2e5f | 2008-10-18 20:28:38 -0700 | [diff] [blame] | 87 | { |
Peter Zijlstra | 01d8b20 | 2011-03-22 16:32:49 -0700 | [diff] [blame] | 88 | VM_BUG_ON(atomic_read(&anon_vma->refcount)); |
Peter Zijlstra | 88c2208 | 2011-05-24 17:12:13 -0700 | [diff] [blame] | 89 | |
| 90 | /* |
| 91 | * Synchronize against page_lock_anon_vma() such that |
| 92 | * we can safely hold the lock without the anon_vma getting |
| 93 | * freed. |
| 94 | * |
| 95 | * Relies on the full mb implied by the atomic_dec_and_test() from |
| 96 | * put_anon_vma() against the acquire barrier implied by |
| 97 | * mutex_trylock() from page_lock_anon_vma(). This orders: |
| 98 | * |
| 99 | * page_lock_anon_vma() VS put_anon_vma() |
| 100 | * mutex_trylock() atomic_dec_and_test() |
| 101 | * LOCK MB |
| 102 | * atomic_read() mutex_is_locked() |
| 103 | * |
| 104 | * LOCK should suffice since the actual taking of the lock must |
| 105 | * happen _before_ what follows. |
| 106 | */ |
| 107 | if (mutex_is_locked(&anon_vma->root->mutex)) { |
| 108 | anon_vma_lock(anon_vma); |
| 109 | anon_vma_unlock(anon_vma); |
| 110 | } |
| 111 | |
Adrian Bunk | fdd2e5f | 2008-10-18 20:28:38 -0700 | [diff] [blame] | 112 | kmem_cache_free(anon_vma_cachep, anon_vma); |
| 113 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 114 | |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 115 | static inline struct anon_vma_chain *anon_vma_chain_alloc(void) |
| 116 | { |
| 117 | return kmem_cache_alloc(anon_vma_chain_cachep, GFP_KERNEL); |
| 118 | } |
| 119 | |
Namhyung Kim | e574b5f | 2010-10-26 14:22:02 -0700 | [diff] [blame] | 120 | static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain) |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 121 | { |
| 122 | kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain); |
| 123 | } |
| 124 | |
Linus Torvalds | d9d332e | 2008-10-19 10:32:20 -0700 | [diff] [blame] | 125 | /** |
| 126 | * anon_vma_prepare - attach an anon_vma to a memory region |
| 127 | * @vma: the memory region in question |
| 128 | * |
| 129 | * This makes sure the memory mapping described by 'vma' has |
| 130 | * an 'anon_vma' attached to it, so that we can associate the |
| 131 | * anonymous pages mapped into it with that anon_vma. |
| 132 | * |
| 133 | * The common case will be that we already have one, but if |
Figo.zhang | 23a0790 | 2010-12-27 15:14:06 +0100 | [diff] [blame] | 134 | * not we either need to find an adjacent mapping that we |
Linus Torvalds | d9d332e | 2008-10-19 10:32:20 -0700 | [diff] [blame] | 135 | * can re-use the anon_vma from (very common when the only |
| 136 | * reason for splitting a vma has been mprotect()), or we |
| 137 | * allocate a new one. |
| 138 | * |
| 139 | * Anon-vma allocations are very subtle, because we may have |
| 140 | * optimistically looked up an anon_vma in page_lock_anon_vma() |
| 141 | * and that may actually touch the spinlock even in the newly |
| 142 | * allocated vma (it depends on RCU to make sure that the |
| 143 | * anon_vma isn't actually destroyed). |
| 144 | * |
| 145 | * As a result, we need to do proper anon_vma locking even |
| 146 | * for the new allocation. At the same time, we do not want |
| 147 | * to do any locking for the common case of already having |
| 148 | * an anon_vma. |
| 149 | * |
| 150 | * This must be called with the mmap_sem held for reading. |
| 151 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 152 | int anon_vma_prepare(struct vm_area_struct *vma) |
| 153 | { |
| 154 | struct anon_vma *anon_vma = vma->anon_vma; |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 155 | struct anon_vma_chain *avc; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 156 | |
| 157 | might_sleep(); |
| 158 | if (unlikely(!anon_vma)) { |
| 159 | struct mm_struct *mm = vma->vm_mm; |
Linus Torvalds | d9d332e | 2008-10-19 10:32:20 -0700 | [diff] [blame] | 160 | struct anon_vma *allocated; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 161 | |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 162 | avc = anon_vma_chain_alloc(); |
| 163 | if (!avc) |
| 164 | goto out_enomem; |
| 165 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 166 | anon_vma = find_mergeable_anon_vma(vma); |
Linus Torvalds | d9d332e | 2008-10-19 10:32:20 -0700 | [diff] [blame] | 167 | allocated = NULL; |
| 168 | if (!anon_vma) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 169 | anon_vma = anon_vma_alloc(); |
| 170 | if (unlikely(!anon_vma)) |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 171 | goto out_enomem_free_avc; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 172 | allocated = anon_vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 173 | } |
| 174 | |
Rik van Riel | cba48b9 | 2010-08-09 17:18:38 -0700 | [diff] [blame] | 175 | anon_vma_lock(anon_vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 176 | /* page_table_lock to protect against threads */ |
| 177 | spin_lock(&mm->page_table_lock); |
| 178 | if (likely(!vma->anon_vma)) { |
| 179 | vma->anon_vma = anon_vma; |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 180 | avc->anon_vma = anon_vma; |
| 181 | avc->vma = vma; |
| 182 | list_add(&avc->same_vma, &vma->anon_vma_chain); |
Andrea Arcangeli | 26ba0cb | 2010-08-09 17:19:08 -0700 | [diff] [blame] | 183 | list_add_tail(&avc->same_anon_vma, &anon_vma->head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 184 | allocated = NULL; |
Oleg Nesterov | 31f2b0e | 2010-04-23 13:18:01 -0400 | [diff] [blame] | 185 | avc = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 186 | } |
| 187 | spin_unlock(&mm->page_table_lock); |
Rik van Riel | cba48b9 | 2010-08-09 17:18:38 -0700 | [diff] [blame] | 188 | anon_vma_unlock(anon_vma); |
Oleg Nesterov | 31f2b0e | 2010-04-23 13:18:01 -0400 | [diff] [blame] | 189 | |
| 190 | if (unlikely(allocated)) |
Peter Zijlstra | 01d8b20 | 2011-03-22 16:32:49 -0700 | [diff] [blame] | 191 | put_anon_vma(allocated); |
Oleg Nesterov | 31f2b0e | 2010-04-23 13:18:01 -0400 | [diff] [blame] | 192 | if (unlikely(avc)) |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 193 | anon_vma_chain_free(avc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 194 | } |
| 195 | return 0; |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 196 | |
| 197 | out_enomem_free_avc: |
| 198 | anon_vma_chain_free(avc); |
| 199 | out_enomem: |
| 200 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 201 | } |
| 202 | |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 203 | static void anon_vma_chain_link(struct vm_area_struct *vma, |
| 204 | struct anon_vma_chain *avc, |
| 205 | struct anon_vma *anon_vma) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 206 | { |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 207 | avc->vma = vma; |
| 208 | avc->anon_vma = anon_vma; |
| 209 | list_add(&avc->same_vma, &vma->anon_vma_chain); |
| 210 | |
Rik van Riel | cba48b9 | 2010-08-09 17:18:38 -0700 | [diff] [blame] | 211 | anon_vma_lock(anon_vma); |
Andrea Arcangeli | 05759d3 | 2011-01-13 15:46:53 -0800 | [diff] [blame] | 212 | /* |
| 213 | * It's critical to add new vmas to the tail of the anon_vma, |
| 214 | * see comment in huge_memory.c:__split_huge_page(). |
| 215 | */ |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 216 | list_add_tail(&avc->same_anon_vma, &anon_vma->head); |
Rik van Riel | cba48b9 | 2010-08-09 17:18:38 -0700 | [diff] [blame] | 217 | anon_vma_unlock(anon_vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 218 | } |
| 219 | |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 220 | /* |
| 221 | * Attach the anon_vmas from src to dst. |
| 222 | * Returns 0 on success, -ENOMEM on failure. |
| 223 | */ |
| 224 | int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 225 | { |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 226 | struct anon_vma_chain *avc, *pavc; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 227 | |
Linus Torvalds | 646d87b | 2010-04-11 17:15:03 -0700 | [diff] [blame] | 228 | list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) { |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 229 | avc = anon_vma_chain_alloc(); |
| 230 | if (!avc) |
| 231 | goto enomem_failure; |
| 232 | anon_vma_chain_link(dst, avc, pavc->anon_vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 233 | } |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 234 | return 0; |
| 235 | |
| 236 | enomem_failure: |
| 237 | unlink_anon_vmas(dst); |
| 238 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 239 | } |
| 240 | |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 241 | /* |
| 242 | * Attach vma to its own anon_vma, as well as to the anon_vmas that |
| 243 | * the corresponding VMA in the parent process is attached to. |
| 244 | * Returns 0 on success, non-zero on failure. |
| 245 | */ |
| 246 | int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 247 | { |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 248 | struct anon_vma_chain *avc; |
| 249 | struct anon_vma *anon_vma; |
| 250 | |
| 251 | /* Don't bother if the parent process has no anon_vma here. */ |
| 252 | if (!pvma->anon_vma) |
| 253 | return 0; |
| 254 | |
| 255 | /* |
| 256 | * First, attach the new VMA to the parent VMA's anon_vmas, |
| 257 | * so rmap can find non-COWed pages in child processes. |
| 258 | */ |
| 259 | if (anon_vma_clone(vma, pvma)) |
| 260 | return -ENOMEM; |
| 261 | |
| 262 | /* Then add our own anon_vma. */ |
| 263 | anon_vma = anon_vma_alloc(); |
| 264 | if (!anon_vma) |
| 265 | goto out_error; |
| 266 | avc = anon_vma_chain_alloc(); |
| 267 | if (!avc) |
| 268 | goto out_error_free_anon_vma; |
Rik van Riel | 5c341ee | 2010-08-09 17:18:39 -0700 | [diff] [blame] | 269 | |
| 270 | /* |
| 271 | * The root anon_vma's spinlock is the lock actually used when we |
| 272 | * lock any of the anon_vmas in this anon_vma tree. |
| 273 | */ |
| 274 | anon_vma->root = pvma->anon_vma->root; |
Rik van Riel | 7654506 | 2010-08-09 17:18:41 -0700 | [diff] [blame] | 275 | /* |
Peter Zijlstra | 01d8b20 | 2011-03-22 16:32:49 -0700 | [diff] [blame] | 276 | * With refcounts, an anon_vma can stay around longer than the |
| 277 | * process it belongs to. The root anon_vma needs to be pinned until |
| 278 | * this anon_vma is freed, because the lock lives in the root. |
Rik van Riel | 7654506 | 2010-08-09 17:18:41 -0700 | [diff] [blame] | 279 | */ |
| 280 | get_anon_vma(anon_vma->root); |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 281 | /* Mark this anon_vma as the one where our new (COWed) pages go. */ |
| 282 | vma->anon_vma = anon_vma; |
Rik van Riel | 5c341ee | 2010-08-09 17:18:39 -0700 | [diff] [blame] | 283 | anon_vma_chain_link(vma, avc, anon_vma); |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 284 | |
| 285 | return 0; |
| 286 | |
| 287 | out_error_free_anon_vma: |
Peter Zijlstra | 01d8b20 | 2011-03-22 16:32:49 -0700 | [diff] [blame] | 288 | put_anon_vma(anon_vma); |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 289 | out_error: |
Rik van Riel | 4946d54 | 2010-04-05 12:13:33 -0400 | [diff] [blame] | 290 | unlink_anon_vmas(vma); |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 291 | return -ENOMEM; |
| 292 | } |
| 293 | |
| 294 | static void anon_vma_unlink(struct anon_vma_chain *anon_vma_chain) |
| 295 | { |
| 296 | struct anon_vma *anon_vma = anon_vma_chain->anon_vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 297 | int empty; |
| 298 | |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 299 | /* If anon_vma_fork fails, we can get an empty anon_vma_chain. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 300 | if (!anon_vma) |
| 301 | return; |
| 302 | |
Rik van Riel | cba48b9 | 2010-08-09 17:18:38 -0700 | [diff] [blame] | 303 | anon_vma_lock(anon_vma); |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 304 | list_del(&anon_vma_chain->same_anon_vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 305 | |
| 306 | /* We must garbage collect the anon_vma if it's empty */ |
Peter Zijlstra | 01d8b20 | 2011-03-22 16:32:49 -0700 | [diff] [blame] | 307 | empty = list_empty(&anon_vma->head); |
Rik van Riel | cba48b9 | 2010-08-09 17:18:38 -0700 | [diff] [blame] | 308 | anon_vma_unlock(anon_vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 309 | |
Peter Zijlstra | 01d8b20 | 2011-03-22 16:32:49 -0700 | [diff] [blame] | 310 | if (empty) |
| 311 | put_anon_vma(anon_vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 312 | } |
| 313 | |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 314 | void unlink_anon_vmas(struct vm_area_struct *vma) |
| 315 | { |
| 316 | struct anon_vma_chain *avc, *next; |
| 317 | |
Rik van Riel | 5c341ee | 2010-08-09 17:18:39 -0700 | [diff] [blame] | 318 | /* |
| 319 | * Unlink each anon_vma chained to the VMA. This list is ordered |
| 320 | * from newest to oldest, ensuring the root anon_vma gets freed last. |
| 321 | */ |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 322 | list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { |
| 323 | anon_vma_unlink(avc); |
| 324 | list_del(&avc->same_vma); |
| 325 | anon_vma_chain_free(avc); |
| 326 | } |
| 327 | } |
| 328 | |
Alexey Dobriyan | 51cc506 | 2008-07-25 19:45:34 -0700 | [diff] [blame] | 329 | static void anon_vma_ctor(void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 330 | { |
Christoph Lameter | a35afb8 | 2007-05-16 22:10:57 -0700 | [diff] [blame] | 331 | struct anon_vma *anon_vma = data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 332 | |
Peter Zijlstra | 2b575eb | 2011-05-24 17:12:11 -0700 | [diff] [blame] | 333 | mutex_init(&anon_vma->mutex); |
Peter Zijlstra | 8381326 | 2011-03-22 16:32:48 -0700 | [diff] [blame] | 334 | atomic_set(&anon_vma->refcount, 0); |
Christoph Lameter | a35afb8 | 2007-05-16 22:10:57 -0700 | [diff] [blame] | 335 | INIT_LIST_HEAD(&anon_vma->head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 336 | } |
| 337 | |
| 338 | void __init anon_vma_init(void) |
| 339 | { |
| 340 | anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), |
Paul Mundt | 20c2df8 | 2007-07-20 10:11:58 +0900 | [diff] [blame] | 341 | 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor); |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 342 | anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 343 | } |
| 344 | |
| 345 | /* |
Peter Zijlstra | 6111e4c | 2011-05-24 17:12:08 -0700 | [diff] [blame] | 346 | * Getting a lock on a stable anon_vma from a page off the LRU is tricky! |
| 347 | * |
| 348 | * Since there is no serialization what so ever against page_remove_rmap() |
| 349 | * the best this function can do is return a locked anon_vma that might |
| 350 | * have been relevant to this page. |
| 351 | * |
| 352 | * The page might have been remapped to a different anon_vma or the anon_vma |
| 353 | * returned may already be freed (and even reused). |
| 354 | * |
Peter Zijlstra | bc658c9 | 2011-05-29 10:33:44 +0200 | [diff] [blame^] | 355 | * In case it was remapped to a different anon_vma, the new anon_vma will be a |
| 356 | * child of the old anon_vma, and the anon_vma lifetime rules will therefore |
| 357 | * ensure that any anon_vma obtained from the page will still be valid for as |
| 358 | * long as we observe page_mapped() [ hence all those page_mapped() tests ]. |
| 359 | * |
Peter Zijlstra | 6111e4c | 2011-05-24 17:12:08 -0700 | [diff] [blame] | 360 | * All users of this function must be very careful when walking the anon_vma |
| 361 | * chain and verify that the page in question is indeed mapped in it |
| 362 | * [ something equivalent to page_mapped_in_vma() ]. |
| 363 | * |
| 364 | * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap() |
| 365 | * that the anon_vma pointer from page->mapping is valid if there is a |
| 366 | * mapcount, we can dereference the anon_vma after observing those. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 367 | */ |
Peter Zijlstra | 746b18d | 2011-05-24 17:12:10 -0700 | [diff] [blame] | 368 | struct anon_vma *page_get_anon_vma(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 369 | { |
Peter Zijlstra | 746b18d | 2011-05-24 17:12:10 -0700 | [diff] [blame] | 370 | struct anon_vma *anon_vma = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 371 | unsigned long anon_mapping; |
| 372 | |
| 373 | rcu_read_lock(); |
Hugh Dickins | 80e1482 | 2009-12-14 17:59:29 -0800 | [diff] [blame] | 374 | anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping); |
Hugh Dickins | 3ca7b3c | 2009-12-14 17:58:57 -0800 | [diff] [blame] | 375 | if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 376 | goto out; |
| 377 | if (!page_mapped(page)) |
| 378 | goto out; |
| 379 | |
| 380 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); |
Peter Zijlstra | 746b18d | 2011-05-24 17:12:10 -0700 | [diff] [blame] | 381 | if (!atomic_inc_not_zero(&anon_vma->refcount)) { |
| 382 | anon_vma = NULL; |
| 383 | goto out; |
| 384 | } |
Hugh Dickins | f181942 | 2010-08-25 23:12:54 -0700 | [diff] [blame] | 385 | |
| 386 | /* |
| 387 | * If this page is still mapped, then its anon_vma cannot have been |
Peter Zijlstra | 746b18d | 2011-05-24 17:12:10 -0700 | [diff] [blame] | 388 | * freed. But if it has been unmapped, we have no security against the |
| 389 | * anon_vma structure being freed and reused (for another anon_vma: |
| 390 | * SLAB_DESTROY_BY_RCU guarantees that - so the atomic_inc_not_zero() |
| 391 | * above cannot corrupt). |
Hugh Dickins | f181942 | 2010-08-25 23:12:54 -0700 | [diff] [blame] | 392 | */ |
Peter Zijlstra | 746b18d | 2011-05-24 17:12:10 -0700 | [diff] [blame] | 393 | if (!page_mapped(page)) { |
| 394 | put_anon_vma(anon_vma); |
| 395 | anon_vma = NULL; |
| 396 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 397 | out: |
| 398 | rcu_read_unlock(); |
Peter Zijlstra | 746b18d | 2011-05-24 17:12:10 -0700 | [diff] [blame] | 399 | |
| 400 | return anon_vma; |
| 401 | } |
| 402 | |
Peter Zijlstra | 88c2208 | 2011-05-24 17:12:13 -0700 | [diff] [blame] | 403 | /* |
| 404 | * Similar to page_get_anon_vma() except it locks the anon_vma. |
| 405 | * |
| 406 | * Its a little more complex as it tries to keep the fast path to a single |
| 407 | * atomic op -- the trylock. If we fail the trylock, we fall back to getting a |
| 408 | * reference like with page_get_anon_vma() and then block on the mutex. |
| 409 | */ |
Peter Zijlstra | 746b18d | 2011-05-24 17:12:10 -0700 | [diff] [blame] | 410 | struct anon_vma *page_lock_anon_vma(struct page *page) |
| 411 | { |
Peter Zijlstra | 88c2208 | 2011-05-24 17:12:13 -0700 | [diff] [blame] | 412 | struct anon_vma *anon_vma = NULL; |
Hugh Dickins | eee0f25 | 2011-05-28 13:20:21 -0700 | [diff] [blame] | 413 | struct anon_vma *root_anon_vma; |
Peter Zijlstra | 88c2208 | 2011-05-24 17:12:13 -0700 | [diff] [blame] | 414 | unsigned long anon_mapping; |
Peter Zijlstra | 746b18d | 2011-05-24 17:12:10 -0700 | [diff] [blame] | 415 | |
Peter Zijlstra | 88c2208 | 2011-05-24 17:12:13 -0700 | [diff] [blame] | 416 | rcu_read_lock(); |
| 417 | anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping); |
| 418 | if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) |
| 419 | goto out; |
| 420 | if (!page_mapped(page)) |
| 421 | goto out; |
Peter Zijlstra | 746b18d | 2011-05-24 17:12:10 -0700 | [diff] [blame] | 422 | |
Peter Zijlstra | 88c2208 | 2011-05-24 17:12:13 -0700 | [diff] [blame] | 423 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); |
Hugh Dickins | eee0f25 | 2011-05-28 13:20:21 -0700 | [diff] [blame] | 424 | root_anon_vma = ACCESS_ONCE(anon_vma->root); |
| 425 | if (mutex_trylock(&root_anon_vma->mutex)) { |
Peter Zijlstra | 88c2208 | 2011-05-24 17:12:13 -0700 | [diff] [blame] | 426 | /* |
Hugh Dickins | eee0f25 | 2011-05-28 13:20:21 -0700 | [diff] [blame] | 427 | * If the page is still mapped, then this anon_vma is still |
| 428 | * its anon_vma, and holding the mutex ensures that it will |
Peter Zijlstra | bc658c9 | 2011-05-29 10:33:44 +0200 | [diff] [blame^] | 429 | * not go away, see anon_vma_free(). |
Peter Zijlstra | 88c2208 | 2011-05-24 17:12:13 -0700 | [diff] [blame] | 430 | */ |
Hugh Dickins | eee0f25 | 2011-05-28 13:20:21 -0700 | [diff] [blame] | 431 | if (!page_mapped(page)) { |
| 432 | mutex_unlock(&root_anon_vma->mutex); |
Peter Zijlstra | 88c2208 | 2011-05-24 17:12:13 -0700 | [diff] [blame] | 433 | anon_vma = NULL; |
| 434 | } |
| 435 | goto out; |
| 436 | } |
| 437 | |
| 438 | /* trylock failed, we got to sleep */ |
| 439 | if (!atomic_inc_not_zero(&anon_vma->refcount)) { |
| 440 | anon_vma = NULL; |
| 441 | goto out; |
| 442 | } |
| 443 | |
| 444 | if (!page_mapped(page)) { |
| 445 | put_anon_vma(anon_vma); |
| 446 | anon_vma = NULL; |
| 447 | goto out; |
| 448 | } |
| 449 | |
| 450 | /* we pinned the anon_vma, its safe to sleep */ |
| 451 | rcu_read_unlock(); |
| 452 | anon_vma_lock(anon_vma); |
| 453 | |
| 454 | if (atomic_dec_and_test(&anon_vma->refcount)) { |
| 455 | /* |
| 456 | * Oops, we held the last refcount, release the lock |
| 457 | * and bail -- can't simply use put_anon_vma() because |
| 458 | * we'll deadlock on the anon_vma_lock() recursion. |
| 459 | */ |
| 460 | anon_vma_unlock(anon_vma); |
| 461 | __put_anon_vma(anon_vma); |
| 462 | anon_vma = NULL; |
| 463 | } |
| 464 | |
| 465 | return anon_vma; |
| 466 | |
| 467 | out: |
| 468 | rcu_read_unlock(); |
Peter Zijlstra | 746b18d | 2011-05-24 17:12:10 -0700 | [diff] [blame] | 469 | return anon_vma; |
Oleg Nesterov | 34bbd70 | 2007-02-28 20:13:49 -0800 | [diff] [blame] | 470 | } |
| 471 | |
Andi Kleen | 10be22d | 2009-09-16 11:50:04 +0200 | [diff] [blame] | 472 | void page_unlock_anon_vma(struct anon_vma *anon_vma) |
Oleg Nesterov | 34bbd70 | 2007-02-28 20:13:49 -0800 | [diff] [blame] | 473 | { |
Rik van Riel | cba48b9 | 2010-08-09 17:18:38 -0700 | [diff] [blame] | 474 | anon_vma_unlock(anon_vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 475 | } |
| 476 | |
| 477 | /* |
Lee Schermerhorn | 3ad33b2 | 2007-11-14 16:59:10 -0800 | [diff] [blame] | 478 | * At what user virtual address is page expected in @vma? |
| 479 | * Returns virtual address or -EFAULT if page's index/offset is not |
| 480 | * within the range mapped the @vma. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 481 | */ |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 482 | inline unsigned long |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 483 | vma_address(struct page *page, struct vm_area_struct *vma) |
| 484 | { |
| 485 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
| 486 | unsigned long address; |
| 487 | |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 488 | if (unlikely(is_vm_hugetlb_page(vma))) |
| 489 | pgoff = page->index << huge_page_order(page_hstate(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 490 | address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
| 491 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) { |
Lee Schermerhorn | 3ad33b2 | 2007-11-14 16:59:10 -0800 | [diff] [blame] | 492 | /* page should be within @vma mapping range */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 493 | return -EFAULT; |
| 494 | } |
| 495 | return address; |
| 496 | } |
| 497 | |
| 498 | /* |
Huang Shijie | bf89c8c | 2009-10-01 15:44:04 -0700 | [diff] [blame] | 499 | * At what user virtual address is page expected in vma? |
Naoya Horiguchi | ab941e0 | 2010-05-11 14:06:55 -0700 | [diff] [blame] | 500 | * Caller should check the page is actually part of the vma. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 501 | */ |
| 502 | unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) |
| 503 | { |
Andrea Arcangeli | 21d0d44 | 2010-08-09 17:19:10 -0700 | [diff] [blame] | 504 | if (PageAnon(page)) { |
Hugh Dickins | 4829b90 | 2010-10-02 17:46:06 -0700 | [diff] [blame] | 505 | struct anon_vma *page__anon_vma = page_anon_vma(page); |
| 506 | /* |
| 507 | * Note: swapoff's unuse_vma() is more efficient with this |
| 508 | * check, and needs it to match anon_vma when KSM is active. |
| 509 | */ |
| 510 | if (!vma->anon_vma || !page__anon_vma || |
| 511 | vma->anon_vma->root != page__anon_vma->root) |
Andrea Arcangeli | 21d0d44 | 2010-08-09 17:19:10 -0700 | [diff] [blame] | 512 | return -EFAULT; |
| 513 | } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) { |
Hugh Dickins | ee498ed | 2005-11-21 21:32:18 -0800 | [diff] [blame] | 514 | if (!vma->vm_file || |
| 515 | vma->vm_file->f_mapping != page->mapping) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 516 | return -EFAULT; |
| 517 | } else |
| 518 | return -EFAULT; |
| 519 | return vma_address(page, vma); |
| 520 | } |
| 521 | |
| 522 | /* |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 523 | * Check that @page is mapped at @address into @mm. |
| 524 | * |
Nick Piggin | 479db0b | 2008-08-20 14:09:18 -0700 | [diff] [blame] | 525 | * If @sync is false, page_check_address may perform a racy check to avoid |
| 526 | * the page table lock when the pte is not present (helpful when reclaiming |
| 527 | * highly shared pages). |
| 528 | * |
Hugh Dickins | b8072f0 | 2005-10-29 18:16:41 -0700 | [diff] [blame] | 529 | * On success returns with pte mapped and locked. |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 530 | */ |
Namhyung Kim | e9a81a8 | 2010-10-26 14:22:01 -0700 | [diff] [blame] | 531 | pte_t *__page_check_address(struct page *page, struct mm_struct *mm, |
Nick Piggin | 479db0b | 2008-08-20 14:09:18 -0700 | [diff] [blame] | 532 | unsigned long address, spinlock_t **ptlp, int sync) |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 533 | { |
| 534 | pgd_t *pgd; |
| 535 | pud_t *pud; |
| 536 | pmd_t *pmd; |
| 537 | pte_t *pte; |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 538 | spinlock_t *ptl; |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 539 | |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 540 | if (unlikely(PageHuge(page))) { |
| 541 | pte = huge_pte_offset(mm, address); |
| 542 | ptl = &mm->page_table_lock; |
| 543 | goto check; |
| 544 | } |
| 545 | |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 546 | pgd = pgd_offset(mm, address); |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 547 | if (!pgd_present(*pgd)) |
| 548 | return NULL; |
| 549 | |
| 550 | pud = pud_offset(pgd, address); |
| 551 | if (!pud_present(*pud)) |
| 552 | return NULL; |
| 553 | |
| 554 | pmd = pmd_offset(pud, address); |
| 555 | if (!pmd_present(*pmd)) |
| 556 | return NULL; |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 557 | if (pmd_trans_huge(*pmd)) |
| 558 | return NULL; |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 559 | |
| 560 | pte = pte_offset_map(pmd, address); |
| 561 | /* Make a quick check before getting the lock */ |
Nick Piggin | 479db0b | 2008-08-20 14:09:18 -0700 | [diff] [blame] | 562 | if (!sync && !pte_present(*pte)) { |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 563 | pte_unmap(pte); |
| 564 | return NULL; |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 565 | } |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 566 | |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 567 | ptl = pte_lockptr(mm, pmd); |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 568 | check: |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 569 | spin_lock(ptl); |
| 570 | if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) { |
| 571 | *ptlp = ptl; |
| 572 | return pte; |
| 573 | } |
| 574 | pte_unmap_unlock(pte, ptl); |
| 575 | return NULL; |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 576 | } |
| 577 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 578 | /** |
| 579 | * page_mapped_in_vma - check whether a page is really mapped in a VMA |
| 580 | * @page: the page to test |
| 581 | * @vma: the VMA to test |
| 582 | * |
| 583 | * Returns 1 if the page is mapped into the page tables of the VMA, 0 |
| 584 | * if the page is not mapped into the page tables of this VMA. Only |
| 585 | * valid for normal file or anonymous VMAs. |
| 586 | */ |
Andi Kleen | 6a46079 | 2009-09-16 11:50:15 +0200 | [diff] [blame] | 587 | int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 588 | { |
| 589 | unsigned long address; |
| 590 | pte_t *pte; |
| 591 | spinlock_t *ptl; |
| 592 | |
| 593 | address = vma_address(page, vma); |
| 594 | if (address == -EFAULT) /* out of vma range */ |
| 595 | return 0; |
| 596 | pte = page_check_address(page, vma->vm_mm, address, &ptl, 1); |
| 597 | if (!pte) /* the page is not in this mm */ |
| 598 | return 0; |
| 599 | pte_unmap_unlock(pte, ptl); |
| 600 | |
| 601 | return 1; |
| 602 | } |
| 603 | |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 604 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 605 | * Subfunctions of page_referenced: page_referenced_one called |
| 606 | * repeatedly from either page_referenced_anon or page_referenced_file. |
| 607 | */ |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 608 | int page_referenced_one(struct page *page, struct vm_area_struct *vma, |
| 609 | unsigned long address, unsigned int *mapcount, |
| 610 | unsigned long *vm_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 611 | { |
| 612 | struct mm_struct *mm = vma->vm_mm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 613 | int referenced = 0; |
| 614 | |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 615 | if (unlikely(PageTransHuge(page))) { |
| 616 | pmd_t *pmd; |
| 617 | |
| 618 | spin_lock(&mm->page_table_lock); |
Andrea Arcangeli | 2da28bf | 2011-03-11 14:58:29 -0800 | [diff] [blame] | 619 | /* |
| 620 | * rmap might return false positives; we must filter |
| 621 | * these out using page_check_address_pmd(). |
| 622 | */ |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 623 | pmd = page_check_address_pmd(page, mm, address, |
| 624 | PAGE_CHECK_ADDRESS_PMD_FLAG); |
Andrea Arcangeli | 2da28bf | 2011-03-11 14:58:29 -0800 | [diff] [blame] | 625 | if (!pmd) { |
| 626 | spin_unlock(&mm->page_table_lock); |
| 627 | goto out; |
| 628 | } |
| 629 | |
| 630 | if (vma->vm_flags & VM_LOCKED) { |
| 631 | spin_unlock(&mm->page_table_lock); |
| 632 | *mapcount = 0; /* break early from loop */ |
| 633 | *vm_flags |= VM_LOCKED; |
| 634 | goto out; |
| 635 | } |
| 636 | |
| 637 | /* go ahead even if the pmd is pmd_trans_splitting() */ |
| 638 | if (pmdp_clear_flush_young_notify(vma, address, pmd)) |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 639 | referenced++; |
| 640 | spin_unlock(&mm->page_table_lock); |
| 641 | } else { |
| 642 | pte_t *pte; |
| 643 | spinlock_t *ptl; |
| 644 | |
Andrea Arcangeli | 2da28bf | 2011-03-11 14:58:29 -0800 | [diff] [blame] | 645 | /* |
| 646 | * rmap might return false positives; we must filter |
| 647 | * these out using page_check_address(). |
| 648 | */ |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 649 | pte = page_check_address(page, mm, address, &ptl, 0); |
| 650 | if (!pte) |
| 651 | goto out; |
| 652 | |
Andrea Arcangeli | 2da28bf | 2011-03-11 14:58:29 -0800 | [diff] [blame] | 653 | if (vma->vm_flags & VM_LOCKED) { |
| 654 | pte_unmap_unlock(pte, ptl); |
| 655 | *mapcount = 0; /* break early from loop */ |
| 656 | *vm_flags |= VM_LOCKED; |
| 657 | goto out; |
| 658 | } |
| 659 | |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 660 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
| 661 | /* |
| 662 | * Don't treat a reference through a sequentially read |
| 663 | * mapping as such. If the page has been used in |
| 664 | * another mapping, we will catch it; if this other |
| 665 | * mapping is already gone, the unmap path will have |
| 666 | * set PG_referenced or activated the page. |
| 667 | */ |
| 668 | if (likely(!VM_SequentialReadHint(vma))) |
| 669 | referenced++; |
| 670 | } |
| 671 | pte_unmap_unlock(pte, ptl); |
| 672 | } |
| 673 | |
Andrea Arcangeli | 2da28bf | 2011-03-11 14:58:29 -0800 | [diff] [blame] | 674 | /* Pretend the page is referenced if the task has the |
| 675 | swap token and is in the middle of a page fault. */ |
| 676 | if (mm != current->mm && has_swap_token(mm) && |
| 677 | rwsem_is_locked(&mm->mmap_sem)) |
| 678 | referenced++; |
| 679 | |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 680 | (*mapcount)--; |
Huang Shijie | 273f047 | 2009-12-14 17:58:51 -0800 | [diff] [blame] | 681 | |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 682 | if (referenced) |
| 683 | *vm_flags |= vma->vm_flags; |
Huang Shijie | 273f047 | 2009-12-14 17:58:51 -0800 | [diff] [blame] | 684 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 685 | return referenced; |
| 686 | } |
| 687 | |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 688 | static int page_referenced_anon(struct page *page, |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 689 | struct mem_cgroup *mem_cont, |
| 690 | unsigned long *vm_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 691 | { |
| 692 | unsigned int mapcount; |
| 693 | struct anon_vma *anon_vma; |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 694 | struct anon_vma_chain *avc; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 695 | int referenced = 0; |
| 696 | |
| 697 | anon_vma = page_lock_anon_vma(page); |
| 698 | if (!anon_vma) |
| 699 | return referenced; |
| 700 | |
| 701 | mapcount = page_mapcount(page); |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 702 | list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { |
| 703 | struct vm_area_struct *vma = avc->vma; |
Hugh Dickins | 1cb1729 | 2009-12-14 17:59:01 -0800 | [diff] [blame] | 704 | unsigned long address = vma_address(page, vma); |
| 705 | if (address == -EFAULT) |
| 706 | continue; |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 707 | /* |
| 708 | * If we are reclaiming on behalf of a cgroup, skip |
| 709 | * counting on behalf of references from different |
| 710 | * cgroups |
| 711 | */ |
Hugh Dickins | bd845e3 | 2008-03-04 14:29:01 -0800 | [diff] [blame] | 712 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 713 | continue; |
Hugh Dickins | 1cb1729 | 2009-12-14 17:59:01 -0800 | [diff] [blame] | 714 | referenced += page_referenced_one(page, vma, address, |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 715 | &mapcount, vm_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 716 | if (!mapcount) |
| 717 | break; |
| 718 | } |
Oleg Nesterov | 34bbd70 | 2007-02-28 20:13:49 -0800 | [diff] [blame] | 719 | |
| 720 | page_unlock_anon_vma(anon_vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 721 | return referenced; |
| 722 | } |
| 723 | |
| 724 | /** |
| 725 | * page_referenced_file - referenced check for object-based rmap |
| 726 | * @page: the page we're checking references on. |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 727 | * @mem_cont: target memory controller |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 728 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 729 | * |
| 730 | * For an object-based mapped page, find all the places it is mapped and |
| 731 | * check/clear the referenced flag. This is done by following the page->mapping |
| 732 | * pointer, then walking the chain of vmas it holds. It returns the number |
| 733 | * of references it found. |
| 734 | * |
| 735 | * This function is only called from page_referenced for object-based pages. |
| 736 | */ |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 737 | static int page_referenced_file(struct page *page, |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 738 | struct mem_cgroup *mem_cont, |
| 739 | unsigned long *vm_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 740 | { |
| 741 | unsigned int mapcount; |
| 742 | struct address_space *mapping = page->mapping; |
| 743 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
| 744 | struct vm_area_struct *vma; |
| 745 | struct prio_tree_iter iter; |
| 746 | int referenced = 0; |
| 747 | |
| 748 | /* |
| 749 | * The caller's checks on page->mapping and !PageAnon have made |
| 750 | * sure that this is a file page: the check for page->mapping |
| 751 | * excludes the case just before it gets set on an anon page. |
| 752 | */ |
| 753 | BUG_ON(PageAnon(page)); |
| 754 | |
| 755 | /* |
| 756 | * The page lock not only makes sure that page->mapping cannot |
| 757 | * suddenly be NULLified by truncation, it makes sure that the |
| 758 | * structure at mapping cannot be freed and reused yet, |
Peter Zijlstra | 3d48ae4 | 2011-05-24 17:12:06 -0700 | [diff] [blame] | 759 | * so we can safely take mapping->i_mmap_mutex. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 760 | */ |
| 761 | BUG_ON(!PageLocked(page)); |
| 762 | |
Peter Zijlstra | 3d48ae4 | 2011-05-24 17:12:06 -0700 | [diff] [blame] | 763 | mutex_lock(&mapping->i_mmap_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 764 | |
| 765 | /* |
Peter Zijlstra | 3d48ae4 | 2011-05-24 17:12:06 -0700 | [diff] [blame] | 766 | * i_mmap_mutex does not stabilize mapcount at all, but mapcount |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 767 | * is more likely to be accurate if we note it after spinning. |
| 768 | */ |
| 769 | mapcount = page_mapcount(page); |
| 770 | |
| 771 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
Hugh Dickins | 1cb1729 | 2009-12-14 17:59:01 -0800 | [diff] [blame] | 772 | unsigned long address = vma_address(page, vma); |
| 773 | if (address == -EFAULT) |
| 774 | continue; |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 775 | /* |
| 776 | * If we are reclaiming on behalf of a cgroup, skip |
| 777 | * counting on behalf of references from different |
| 778 | * cgroups |
| 779 | */ |
Hugh Dickins | bd845e3 | 2008-03-04 14:29:01 -0800 | [diff] [blame] | 780 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 781 | continue; |
Hugh Dickins | 1cb1729 | 2009-12-14 17:59:01 -0800 | [diff] [blame] | 782 | referenced += page_referenced_one(page, vma, address, |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 783 | &mapcount, vm_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 784 | if (!mapcount) |
| 785 | break; |
| 786 | } |
| 787 | |
Peter Zijlstra | 3d48ae4 | 2011-05-24 17:12:06 -0700 | [diff] [blame] | 788 | mutex_unlock(&mapping->i_mmap_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 789 | return referenced; |
| 790 | } |
| 791 | |
| 792 | /** |
| 793 | * page_referenced - test if the page was referenced |
| 794 | * @page: the page to test |
| 795 | * @is_locked: caller holds lock on the page |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 796 | * @mem_cont: target memory controller |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 797 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 798 | * |
| 799 | * Quick test_and_clear_referenced for all mappings to a page, |
| 800 | * returns the number of ptes which referenced the page. |
| 801 | */ |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 802 | int page_referenced(struct page *page, |
| 803 | int is_locked, |
| 804 | struct mem_cgroup *mem_cont, |
| 805 | unsigned long *vm_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 806 | { |
| 807 | int referenced = 0; |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 808 | int we_locked = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 809 | |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 810 | *vm_flags = 0; |
Hugh Dickins | 3ca7b3c | 2009-12-14 17:58:57 -0800 | [diff] [blame] | 811 | if (page_mapped(page) && page_rmapping(page)) { |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 812 | if (!is_locked && (!PageAnon(page) || PageKsm(page))) { |
| 813 | we_locked = trylock_page(page); |
| 814 | if (!we_locked) { |
| 815 | referenced++; |
| 816 | goto out; |
| 817 | } |
| 818 | } |
| 819 | if (unlikely(PageKsm(page))) |
| 820 | referenced += page_referenced_ksm(page, mem_cont, |
| 821 | vm_flags); |
| 822 | else if (PageAnon(page)) |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 823 | referenced += page_referenced_anon(page, mem_cont, |
| 824 | vm_flags); |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 825 | else if (page->mapping) |
Wu Fengguang | 6fe6b7e | 2009-06-16 15:33:05 -0700 | [diff] [blame] | 826 | referenced += page_referenced_file(page, mem_cont, |
| 827 | vm_flags); |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 828 | if (we_locked) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 829 | unlock_page(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 830 | } |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 831 | out: |
Martin Schwidefsky | 2d42552 | 2011-05-23 10:24:39 +0200 | [diff] [blame] | 832 | if (page_test_and_clear_young(page_to_pfn(page))) |
Christian Borntraeger | 5b7baf0 | 2008-03-25 18:47:12 +0100 | [diff] [blame] | 833 | referenced++; |
| 834 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 835 | return referenced; |
| 836 | } |
| 837 | |
Hugh Dickins | 1cb1729 | 2009-12-14 17:59:01 -0800 | [diff] [blame] | 838 | static int page_mkclean_one(struct page *page, struct vm_area_struct *vma, |
| 839 | unsigned long address) |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 840 | { |
| 841 | struct mm_struct *mm = vma->vm_mm; |
Peter Zijlstra | c2fda5f | 2006-12-22 14:25:52 +0100 | [diff] [blame] | 842 | pte_t *pte; |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 843 | spinlock_t *ptl; |
| 844 | int ret = 0; |
| 845 | |
Nick Piggin | 479db0b | 2008-08-20 14:09:18 -0700 | [diff] [blame] | 846 | pte = page_check_address(page, mm, address, &ptl, 1); |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 847 | if (!pte) |
| 848 | goto out; |
| 849 | |
Peter Zijlstra | c2fda5f | 2006-12-22 14:25:52 +0100 | [diff] [blame] | 850 | if (pte_dirty(*pte) || pte_write(*pte)) { |
| 851 | pte_t entry; |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 852 | |
Peter Zijlstra | c2fda5f | 2006-12-22 14:25:52 +0100 | [diff] [blame] | 853 | flush_cache_page(vma, address, pte_pfn(*pte)); |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 854 | entry = ptep_clear_flush_notify(vma, address, pte); |
Peter Zijlstra | c2fda5f | 2006-12-22 14:25:52 +0100 | [diff] [blame] | 855 | entry = pte_wrprotect(entry); |
| 856 | entry = pte_mkclean(entry); |
Al Viro | d6e88e6 | 2006-12-29 16:48:35 -0800 | [diff] [blame] | 857 | set_pte_at(mm, address, pte, entry); |
Peter Zijlstra | c2fda5f | 2006-12-22 14:25:52 +0100 | [diff] [blame] | 858 | ret = 1; |
| 859 | } |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 860 | |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 861 | pte_unmap_unlock(pte, ptl); |
| 862 | out: |
| 863 | return ret; |
| 864 | } |
| 865 | |
| 866 | static int page_mkclean_file(struct address_space *mapping, struct page *page) |
| 867 | { |
| 868 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
| 869 | struct vm_area_struct *vma; |
| 870 | struct prio_tree_iter iter; |
| 871 | int ret = 0; |
| 872 | |
| 873 | BUG_ON(PageAnon(page)); |
| 874 | |
Peter Zijlstra | 3d48ae4 | 2011-05-24 17:12:06 -0700 | [diff] [blame] | 875 | mutex_lock(&mapping->i_mmap_mutex); |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 876 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
Hugh Dickins | 1cb1729 | 2009-12-14 17:59:01 -0800 | [diff] [blame] | 877 | if (vma->vm_flags & VM_SHARED) { |
| 878 | unsigned long address = vma_address(page, vma); |
| 879 | if (address == -EFAULT) |
| 880 | continue; |
| 881 | ret += page_mkclean_one(page, vma, address); |
| 882 | } |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 883 | } |
Peter Zijlstra | 3d48ae4 | 2011-05-24 17:12:06 -0700 | [diff] [blame] | 884 | mutex_unlock(&mapping->i_mmap_mutex); |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 885 | return ret; |
| 886 | } |
| 887 | |
| 888 | int page_mkclean(struct page *page) |
| 889 | { |
| 890 | int ret = 0; |
| 891 | |
| 892 | BUG_ON(!PageLocked(page)); |
| 893 | |
| 894 | if (page_mapped(page)) { |
| 895 | struct address_space *mapping = page_mapping(page); |
Christian Borntraeger | ce7e9fa | 2007-11-20 11:13:36 +0100 | [diff] [blame] | 896 | if (mapping) { |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 897 | ret = page_mkclean_file(mapping, page); |
Martin Schwidefsky | 2d42552 | 2011-05-23 10:24:39 +0200 | [diff] [blame] | 898 | if (page_test_and_clear_dirty(page_to_pfn(page), 1)) |
Christian Borntraeger | ce7e9fa | 2007-11-20 11:13:36 +0100 | [diff] [blame] | 899 | ret = 1; |
Martin Schwidefsky | 6c21048 | 2007-04-27 16:01:57 +0200 | [diff] [blame] | 900 | } |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 901 | } |
| 902 | |
| 903 | return ret; |
| 904 | } |
Jaya Kumar | 60b59be | 2007-05-08 00:37:37 -0700 | [diff] [blame] | 905 | EXPORT_SYMBOL_GPL(page_mkclean); |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 906 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 907 | /** |
Rik van Riel | c44b674 | 2010-03-05 13:42:09 -0800 | [diff] [blame] | 908 | * page_move_anon_rmap - move a page to our anon_vma |
| 909 | * @page: the page to move to our anon_vma |
| 910 | * @vma: the vma the page belongs to |
| 911 | * @address: the user virtual address mapped |
| 912 | * |
| 913 | * When a page belongs exclusively to one process after a COW event, |
| 914 | * that page can be moved into the anon_vma that belongs to just that |
| 915 | * process, so the rmap code will not search the parent or sibling |
| 916 | * processes. |
| 917 | */ |
| 918 | void page_move_anon_rmap(struct page *page, |
| 919 | struct vm_area_struct *vma, unsigned long address) |
| 920 | { |
| 921 | struct anon_vma *anon_vma = vma->anon_vma; |
| 922 | |
| 923 | VM_BUG_ON(!PageLocked(page)); |
| 924 | VM_BUG_ON(!anon_vma); |
| 925 | VM_BUG_ON(page->index != linear_page_index(vma, address)); |
| 926 | |
| 927 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
| 928 | page->mapping = (struct address_space *) anon_vma; |
| 929 | } |
| 930 | |
| 931 | /** |
Andi Kleen | 4e1c197 | 2010-09-22 12:43:56 +0200 | [diff] [blame] | 932 | * __page_set_anon_rmap - set up new anonymous rmap |
| 933 | * @page: Page to add to rmap |
| 934 | * @vma: VM area to add page to. |
| 935 | * @address: User virtual address of the mapping |
Rik van Riel | e8a03fe | 2010-04-14 17:59:28 -0400 | [diff] [blame] | 936 | * @exclusive: the page is exclusively owned by the current process |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 937 | */ |
| 938 | static void __page_set_anon_rmap(struct page *page, |
Rik van Riel | e8a03fe | 2010-04-14 17:59:28 -0400 | [diff] [blame] | 939 | struct vm_area_struct *vma, unsigned long address, int exclusive) |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 940 | { |
Rik van Riel | e8a03fe | 2010-04-14 17:59:28 -0400 | [diff] [blame] | 941 | struct anon_vma *anon_vma = vma->anon_vma; |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 942 | |
Rik van Riel | e8a03fe | 2010-04-14 17:59:28 -0400 | [diff] [blame] | 943 | BUG_ON(!anon_vma); |
Linus Torvalds | ea90002 | 2010-04-12 12:44:29 -0700 | [diff] [blame] | 944 | |
Andi Kleen | 4e1c197 | 2010-09-22 12:43:56 +0200 | [diff] [blame] | 945 | if (PageAnon(page)) |
| 946 | return; |
| 947 | |
Linus Torvalds | ea90002 | 2010-04-12 12:44:29 -0700 | [diff] [blame] | 948 | /* |
Rik van Riel | e8a03fe | 2010-04-14 17:59:28 -0400 | [diff] [blame] | 949 | * If the page isn't exclusively mapped into this vma, |
| 950 | * we must use the _oldest_ possible anon_vma for the |
| 951 | * page mapping! |
Linus Torvalds | ea90002 | 2010-04-12 12:44:29 -0700 | [diff] [blame] | 952 | */ |
Andi Kleen | 4e1c197 | 2010-09-22 12:43:56 +0200 | [diff] [blame] | 953 | if (!exclusive) |
Andrea Arcangeli | 288468c | 2010-08-09 17:19:09 -0700 | [diff] [blame] | 954 | anon_vma = anon_vma->root; |
Linus Torvalds | ea90002 | 2010-04-12 12:44:29 -0700 | [diff] [blame] | 955 | |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 956 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
| 957 | page->mapping = (struct address_space *) anon_vma; |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 958 | page->index = linear_page_index(vma, address); |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 959 | } |
| 960 | |
| 961 | /** |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 962 | * __page_check_anon_rmap - sanity check anonymous rmap addition |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 963 | * @page: the page to add the mapping to |
| 964 | * @vma: the vm area in which the mapping is added |
| 965 | * @address: the user virtual address mapped |
| 966 | */ |
| 967 | static void __page_check_anon_rmap(struct page *page, |
| 968 | struct vm_area_struct *vma, unsigned long address) |
| 969 | { |
| 970 | #ifdef CONFIG_DEBUG_VM |
| 971 | /* |
| 972 | * The page's anon-rmap details (mapping and index) are guaranteed to |
| 973 | * be set up correctly at this point. |
| 974 | * |
| 975 | * We have exclusion against page_add_anon_rmap because the caller |
| 976 | * always holds the page locked, except if called from page_dup_rmap, |
| 977 | * in which case the page is already known to be setup. |
| 978 | * |
| 979 | * We have exclusion against page_add_new_anon_rmap because those pages |
| 980 | * are initially only visible via the pagetables, and the pte is locked |
| 981 | * over the call to page_add_new_anon_rmap. |
| 982 | */ |
Andrea Arcangeli | 44ab57a | 2010-08-09 17:19:10 -0700 | [diff] [blame] | 983 | BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root); |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 984 | BUG_ON(page->index != linear_page_index(vma, address)); |
| 985 | #endif |
| 986 | } |
| 987 | |
| 988 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 989 | * page_add_anon_rmap - add pte mapping to an anonymous page |
| 990 | * @page: the page to add the mapping to |
| 991 | * @vma: the vm area in which the mapping is added |
| 992 | * @address: the user virtual address mapped |
| 993 | * |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 994 | * The caller needs to hold the pte lock, and the page must be locked in |
Hugh Dickins | 80e1482 | 2009-12-14 17:59:29 -0800 | [diff] [blame] | 995 | * the anon_vma case: to serialize mapping,index checking after setting, |
| 996 | * and to ensure that PageAnon is not being upgraded racily to PageKsm |
| 997 | * (but PageKsm is never downgraded to PageAnon). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 998 | */ |
| 999 | void page_add_anon_rmap(struct page *page, |
| 1000 | struct vm_area_struct *vma, unsigned long address) |
| 1001 | { |
Rik van Riel | ad8c2ee | 2010-08-09 17:19:48 -0700 | [diff] [blame] | 1002 | do_page_add_anon_rmap(page, vma, address, 0); |
| 1003 | } |
| 1004 | |
| 1005 | /* |
| 1006 | * Special version of the above for do_swap_page, which often runs |
| 1007 | * into pages that are exclusively owned by the current process. |
| 1008 | * Everybody else should continue to use page_add_anon_rmap above. |
| 1009 | */ |
| 1010 | void do_page_add_anon_rmap(struct page *page, |
| 1011 | struct vm_area_struct *vma, unsigned long address, int exclusive) |
| 1012 | { |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 1013 | int first = atomic_inc_and_test(&page->_mapcount); |
Andrea Arcangeli | 7913417 | 2011-01-13 15:46:58 -0800 | [diff] [blame] | 1014 | if (first) { |
| 1015 | if (!PageTransHuge(page)) |
| 1016 | __inc_zone_page_state(page, NR_ANON_PAGES); |
| 1017 | else |
| 1018 | __inc_zone_page_state(page, |
| 1019 | NR_ANON_TRANSPARENT_HUGEPAGES); |
| 1020 | } |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 1021 | if (unlikely(PageKsm(page))) |
| 1022 | return; |
| 1023 | |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 1024 | VM_BUG_ON(!PageLocked(page)); |
Hugh Dickins | 5dbe0af | 2011-05-28 13:17:04 -0700 | [diff] [blame] | 1025 | /* address might be in next vma when migration races vma_adjust */ |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 1026 | if (first) |
Rik van Riel | ad8c2ee | 2010-08-09 17:19:48 -0700 | [diff] [blame] | 1027 | __page_set_anon_rmap(page, vma, address, exclusive); |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 1028 | else |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 1029 | __page_check_anon_rmap(page, vma, address); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1030 | } |
| 1031 | |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 1032 | /** |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 1033 | * page_add_new_anon_rmap - add pte mapping to a new anonymous page |
| 1034 | * @page: the page to add the mapping to |
| 1035 | * @vma: the vm area in which the mapping is added |
| 1036 | * @address: the user virtual address mapped |
| 1037 | * |
| 1038 | * Same as page_add_anon_rmap but must only be called on *new* pages. |
| 1039 | * This means the inc-and-test can be bypassed. |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 1040 | * Page does not have to be locked. |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 1041 | */ |
| 1042 | void page_add_new_anon_rmap(struct page *page, |
| 1043 | struct vm_area_struct *vma, unsigned long address) |
| 1044 | { |
Hugh Dickins | b5934c5 | 2009-01-06 14:39:25 -0800 | [diff] [blame] | 1045 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
Hugh Dickins | cbf84b7 | 2009-01-06 14:39:27 -0800 | [diff] [blame] | 1046 | SetPageSwapBacked(page); |
| 1047 | atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */ |
Andrea Arcangeli | 7913417 | 2011-01-13 15:46:58 -0800 | [diff] [blame] | 1048 | if (!PageTransHuge(page)) |
| 1049 | __inc_zone_page_state(page, NR_ANON_PAGES); |
| 1050 | else |
| 1051 | __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES); |
Rik van Riel | e8a03fe | 2010-04-14 17:59:28 -0400 | [diff] [blame] | 1052 | __page_set_anon_rmap(page, vma, address, 1); |
Hugh Dickins | b5934c5 | 2009-01-06 14:39:25 -0800 | [diff] [blame] | 1053 | if (page_evictable(page, vma)) |
Hugh Dickins | cbf84b7 | 2009-01-06 14:39:27 -0800 | [diff] [blame] | 1054 | lru_cache_add_lru(page, LRU_ACTIVE_ANON); |
Hugh Dickins | b5934c5 | 2009-01-06 14:39:25 -0800 | [diff] [blame] | 1055 | else |
| 1056 | add_page_to_unevictable_list(page); |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 1057 | } |
| 1058 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1059 | /** |
| 1060 | * page_add_file_rmap - add pte mapping to a file page |
| 1061 | * @page: the page to add the mapping to |
| 1062 | * |
Hugh Dickins | b8072f0 | 2005-10-29 18:16:41 -0700 | [diff] [blame] | 1063 | * The caller needs to hold the pte lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1064 | */ |
| 1065 | void page_add_file_rmap(struct page *page) |
| 1066 | { |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1067 | if (atomic_inc_and_test(&page->_mapcount)) { |
Christoph Lameter | 65ba55f | 2006-06-30 01:55:34 -0700 | [diff] [blame] | 1068 | __inc_zone_page_state(page, NR_FILE_MAPPED); |
Greg Thelen | 2a7106f | 2011-01-13 15:47:37 -0800 | [diff] [blame] | 1069 | mem_cgroup_inc_page_stat(page, MEMCG_NR_FILE_MAPPED); |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1070 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1071 | } |
| 1072 | |
| 1073 | /** |
| 1074 | * page_remove_rmap - take down pte mapping from a page |
| 1075 | * @page: page to remove mapping from |
| 1076 | * |
Hugh Dickins | b8072f0 | 2005-10-29 18:16:41 -0700 | [diff] [blame] | 1077 | * The caller needs to hold the pte lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1078 | */ |
Hugh Dickins | edc315f | 2009-01-06 14:40:11 -0800 | [diff] [blame] | 1079 | void page_remove_rmap(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1080 | { |
KOSAKI Motohiro | b904dcf | 2009-09-21 17:01:28 -0700 | [diff] [blame] | 1081 | /* page still mapped by someone else? */ |
| 1082 | if (!atomic_add_negative(-1, &page->_mapcount)) |
| 1083 | return; |
| 1084 | |
| 1085 | /* |
| 1086 | * Now that the last pte has gone, s390 must transfer dirty |
| 1087 | * flag from storage key to struct page. We can usually skip |
| 1088 | * this if the page is anon, so about to be freed; but perhaps |
| 1089 | * not if it's in swapcache - there might be another pte slot |
| 1090 | * containing the swap entry, but page not yet written to swap. |
| 1091 | */ |
Martin Schwidefsky | 2d42552 | 2011-05-23 10:24:39 +0200 | [diff] [blame] | 1092 | if ((!PageAnon(page) || PageSwapCache(page)) && |
| 1093 | page_test_and_clear_dirty(page_to_pfn(page), 1)) |
KOSAKI Motohiro | b904dcf | 2009-09-21 17:01:28 -0700 | [diff] [blame] | 1094 | set_page_dirty(page); |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 1095 | /* |
| 1096 | * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED |
| 1097 | * and not charged by memcg for now. |
| 1098 | */ |
| 1099 | if (unlikely(PageHuge(page))) |
| 1100 | return; |
KOSAKI Motohiro | b904dcf | 2009-09-21 17:01:28 -0700 | [diff] [blame] | 1101 | if (PageAnon(page)) { |
| 1102 | mem_cgroup_uncharge_page(page); |
Andrea Arcangeli | 7913417 | 2011-01-13 15:46:58 -0800 | [diff] [blame] | 1103 | if (!PageTransHuge(page)) |
| 1104 | __dec_zone_page_state(page, NR_ANON_PAGES); |
| 1105 | else |
| 1106 | __dec_zone_page_state(page, |
| 1107 | NR_ANON_TRANSPARENT_HUGEPAGES); |
KOSAKI Motohiro | b904dcf | 2009-09-21 17:01:28 -0700 | [diff] [blame] | 1108 | } else { |
| 1109 | __dec_zone_page_state(page, NR_FILE_MAPPED); |
Greg Thelen | 2a7106f | 2011-01-13 15:47:37 -0800 | [diff] [blame] | 1110 | mem_cgroup_dec_page_stat(page, MEMCG_NR_FILE_MAPPED); |
KOSAKI Motohiro | b904dcf | 2009-09-21 17:01:28 -0700 | [diff] [blame] | 1111 | } |
KOSAKI Motohiro | b904dcf | 2009-09-21 17:01:28 -0700 | [diff] [blame] | 1112 | /* |
| 1113 | * It would be tidy to reset the PageAnon mapping here, |
| 1114 | * but that might overwrite a racing page_add_anon_rmap |
| 1115 | * which increments mapcount after us but sets mapping |
| 1116 | * before us: so leave the reset to free_hot_cold_page, |
| 1117 | * and remember that it's only reliable while mapped. |
| 1118 | * Leaving it set also helps swapoff to reinstate ptes |
| 1119 | * faster for those pages still in swapcache. |
| 1120 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1121 | } |
| 1122 | |
| 1123 | /* |
| 1124 | * Subfunctions of try_to_unmap: try_to_unmap_one called |
| 1125 | * repeatedly from either try_to_unmap_anon or try_to_unmap_file. |
| 1126 | */ |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 1127 | int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
| 1128 | unsigned long address, enum ttu_flags flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1129 | { |
| 1130 | struct mm_struct *mm = vma->vm_mm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1131 | pte_t *pte; |
| 1132 | pte_t pteval; |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 1133 | spinlock_t *ptl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1134 | int ret = SWAP_AGAIN; |
| 1135 | |
Nick Piggin | 479db0b | 2008-08-20 14:09:18 -0700 | [diff] [blame] | 1136 | pte = page_check_address(page, mm, address, &ptl, 0); |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 1137 | if (!pte) |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 1138 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1139 | |
| 1140 | /* |
| 1141 | * If the page is mlock()d, we cannot swap it out. |
| 1142 | * If it's recently referenced (perhaps page_referenced |
| 1143 | * skipped over this mm) then we should reactivate it. |
| 1144 | */ |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1145 | if (!(flags & TTU_IGNORE_MLOCK)) { |
KOSAKI Motohiro | caed0f4 | 2009-12-14 17:59:45 -0800 | [diff] [blame] | 1146 | if (vma->vm_flags & VM_LOCKED) |
| 1147 | goto out_mlock; |
| 1148 | |
Hugh Dickins | af8e335 | 2009-12-14 17:58:59 -0800 | [diff] [blame] | 1149 | if (TTU_ACTION(flags) == TTU_MUNLOCK) |
Hugh Dickins | 53f79ac | 2009-12-14 17:58:58 -0800 | [diff] [blame] | 1150 | goto out_unmap; |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1151 | } |
| 1152 | if (!(flags & TTU_IGNORE_ACCESS)) { |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1153 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
| 1154 | ret = SWAP_FAIL; |
| 1155 | goto out_unmap; |
| 1156 | } |
| 1157 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1158 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1159 | /* Nuke the page table entry. */ |
| 1160 | flush_cache_page(vma, address, page_to_pfn(page)); |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 1161 | pteval = ptep_clear_flush_notify(vma, address, pte); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1162 | |
| 1163 | /* Move the dirty bit to the physical page now the pte is gone. */ |
| 1164 | if (pte_dirty(pteval)) |
| 1165 | set_page_dirty(page); |
| 1166 | |
Hugh Dickins | 365e9c87 | 2005-10-29 18:16:18 -0700 | [diff] [blame] | 1167 | /* Update high watermark before we lower rss */ |
| 1168 | update_hiwater_rss(mm); |
| 1169 | |
Andi Kleen | 888b9f7 | 2009-09-16 11:50:11 +0200 | [diff] [blame] | 1170 | if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) { |
| 1171 | if (PageAnon(page)) |
KAMEZAWA Hiroyuki | d559db0 | 2010-03-05 13:41:39 -0800 | [diff] [blame] | 1172 | dec_mm_counter(mm, MM_ANONPAGES); |
Andi Kleen | 888b9f7 | 2009-09-16 11:50:11 +0200 | [diff] [blame] | 1173 | else |
KAMEZAWA Hiroyuki | d559db0 | 2010-03-05 13:41:39 -0800 | [diff] [blame] | 1174 | dec_mm_counter(mm, MM_FILEPAGES); |
Andi Kleen | 888b9f7 | 2009-09-16 11:50:11 +0200 | [diff] [blame] | 1175 | set_pte_at(mm, address, pte, |
| 1176 | swp_entry_to_pte(make_hwpoison_entry(page))); |
| 1177 | } else if (PageAnon(page)) { |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 1178 | swp_entry_t entry = { .val = page_private(page) }; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 1179 | |
| 1180 | if (PageSwapCache(page)) { |
| 1181 | /* |
| 1182 | * Store the swap location in the pte. |
| 1183 | * See handle_pte_fault() ... |
| 1184 | */ |
Hugh Dickins | 570a335 | 2009-12-14 17:58:46 -0800 | [diff] [blame] | 1185 | if (swap_duplicate(entry) < 0) { |
| 1186 | set_pte_at(mm, address, pte, pteval); |
| 1187 | ret = SWAP_FAIL; |
| 1188 | goto out_unmap; |
| 1189 | } |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 1190 | if (list_empty(&mm->mmlist)) { |
| 1191 | spin_lock(&mmlist_lock); |
| 1192 | if (list_empty(&mm->mmlist)) |
| 1193 | list_add(&mm->mmlist, &init_mm.mmlist); |
| 1194 | spin_unlock(&mmlist_lock); |
| 1195 | } |
KAMEZAWA Hiroyuki | d559db0 | 2010-03-05 13:41:39 -0800 | [diff] [blame] | 1196 | dec_mm_counter(mm, MM_ANONPAGES); |
KAMEZAWA Hiroyuki | b084d43 | 2010-03-05 13:41:42 -0800 | [diff] [blame] | 1197 | inc_mm_counter(mm, MM_SWAPENTS); |
KOSAKI Motohiro | 64cdd54 | 2009-01-06 14:39:16 -0800 | [diff] [blame] | 1198 | } else if (PAGE_MIGRATION) { |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 1199 | /* |
| 1200 | * Store the pfn of the page in a special migration |
| 1201 | * pte. do_swap_page() will wait until the migration |
| 1202 | * pte is removed and then restart fault handling. |
| 1203 | */ |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1204 | BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION); |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 1205 | entry = make_migration_entry(page, pte_write(pteval)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1206 | } |
| 1207 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); |
| 1208 | BUG_ON(pte_file(*pte)); |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1209 | } else if (PAGE_MIGRATION && (TTU_ACTION(flags) == TTU_MIGRATION)) { |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 1210 | /* Establish migration entry for a file page */ |
| 1211 | swp_entry_t entry; |
| 1212 | entry = make_migration_entry(page, pte_write(pteval)); |
| 1213 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); |
| 1214 | } else |
KAMEZAWA Hiroyuki | d559db0 | 2010-03-05 13:41:39 -0800 | [diff] [blame] | 1215 | dec_mm_counter(mm, MM_FILEPAGES); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1216 | |
Hugh Dickins | edc315f | 2009-01-06 14:40:11 -0800 | [diff] [blame] | 1217 | page_remove_rmap(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1218 | page_cache_release(page); |
| 1219 | |
| 1220 | out_unmap: |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 1221 | pte_unmap_unlock(pte, ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1222 | out: |
| 1223 | return ret; |
KOSAKI Motohiro | caed0f4 | 2009-12-14 17:59:45 -0800 | [diff] [blame] | 1224 | |
| 1225 | out_mlock: |
| 1226 | pte_unmap_unlock(pte, ptl); |
| 1227 | |
| 1228 | |
| 1229 | /* |
| 1230 | * We need mmap_sem locking, Otherwise VM_LOCKED check makes |
| 1231 | * unstable result and race. Plus, We can't wait here because |
Peter Zijlstra | 2b575eb | 2011-05-24 17:12:11 -0700 | [diff] [blame] | 1232 | * we now hold anon_vma->mutex or mapping->i_mmap_mutex. |
KOSAKI Motohiro | caed0f4 | 2009-12-14 17:59:45 -0800 | [diff] [blame] | 1233 | * if trylock failed, the page remain in evictable lru and later |
| 1234 | * vmscan could retry to move the page to unevictable lru if the |
| 1235 | * page is actually mlocked. |
| 1236 | */ |
| 1237 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { |
| 1238 | if (vma->vm_flags & VM_LOCKED) { |
| 1239 | mlock_vma_page(page); |
| 1240 | ret = SWAP_MLOCK; |
| 1241 | } |
| 1242 | up_read(&vma->vm_mm->mmap_sem); |
| 1243 | } |
| 1244 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1245 | } |
| 1246 | |
| 1247 | /* |
| 1248 | * objrmap doesn't work for nonlinear VMAs because the assumption that |
| 1249 | * offset-into-file correlates with offset-into-virtual-addresses does not hold. |
| 1250 | * Consequently, given a particular page and its ->index, we cannot locate the |
| 1251 | * ptes which are mapping that page without an exhaustive linear search. |
| 1252 | * |
| 1253 | * So what this code does is a mini "virtual scan" of each nonlinear VMA which |
| 1254 | * maps the file to which the target page belongs. The ->vm_private_data field |
| 1255 | * holds the current cursor into that scan. Successive searches will circulate |
| 1256 | * around the vma's virtual address space. |
| 1257 | * |
| 1258 | * So as more replacement pressure is applied to the pages in a nonlinear VMA, |
| 1259 | * more scanning pressure is placed against them as well. Eventually pages |
| 1260 | * will become fully unmapped and are eligible for eviction. |
| 1261 | * |
| 1262 | * For very sparsely populated VMAs this is a little inefficient - chances are |
| 1263 | * there there won't be many ptes located within the scan cluster. In this case |
| 1264 | * maybe we could scan further - to the end of the pte page, perhaps. |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1265 | * |
| 1266 | * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can |
| 1267 | * acquire it without blocking. If vma locked, mlock the pages in the cluster, |
| 1268 | * rather than unmapping them. If we encounter the "check_page" that vmscan is |
| 1269 | * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1270 | */ |
| 1271 | #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE) |
| 1272 | #define CLUSTER_MASK (~(CLUSTER_SIZE - 1)) |
| 1273 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1274 | static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount, |
| 1275 | struct vm_area_struct *vma, struct page *check_page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1276 | { |
| 1277 | struct mm_struct *mm = vma->vm_mm; |
| 1278 | pgd_t *pgd; |
| 1279 | pud_t *pud; |
| 1280 | pmd_t *pmd; |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 1281 | pte_t *pte; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1282 | pte_t pteval; |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 1283 | spinlock_t *ptl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1284 | struct page *page; |
| 1285 | unsigned long address; |
| 1286 | unsigned long end; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1287 | int ret = SWAP_AGAIN; |
| 1288 | int locked_vma = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1289 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1290 | address = (vma->vm_start + cursor) & CLUSTER_MASK; |
| 1291 | end = address + CLUSTER_SIZE; |
| 1292 | if (address < vma->vm_start) |
| 1293 | address = vma->vm_start; |
| 1294 | if (end > vma->vm_end) |
| 1295 | end = vma->vm_end; |
| 1296 | |
| 1297 | pgd = pgd_offset(mm, address); |
| 1298 | if (!pgd_present(*pgd)) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1299 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1300 | |
| 1301 | pud = pud_offset(pgd, address); |
| 1302 | if (!pud_present(*pud)) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1303 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1304 | |
| 1305 | pmd = pmd_offset(pud, address); |
| 1306 | if (!pmd_present(*pmd)) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1307 | return ret; |
| 1308 | |
| 1309 | /* |
Hugh Dickins | af8e335 | 2009-12-14 17:58:59 -0800 | [diff] [blame] | 1310 | * If we can acquire the mmap_sem for read, and vma is VM_LOCKED, |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1311 | * keep the sem while scanning the cluster for mlocking pages. |
| 1312 | */ |
Hugh Dickins | af8e335 | 2009-12-14 17:58:59 -0800 | [diff] [blame] | 1313 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1314 | locked_vma = (vma->vm_flags & VM_LOCKED); |
| 1315 | if (!locked_vma) |
| 1316 | up_read(&vma->vm_mm->mmap_sem); /* don't need it */ |
| 1317 | } |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 1318 | |
| 1319 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1320 | |
Hugh Dickins | 365e9c87 | 2005-10-29 18:16:18 -0700 | [diff] [blame] | 1321 | /* Update high watermark before we lower rss */ |
| 1322 | update_hiwater_rss(mm); |
| 1323 | |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 1324 | for (; address < end; pte++, address += PAGE_SIZE) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1325 | if (!pte_present(*pte)) |
| 1326 | continue; |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 1327 | page = vm_normal_page(vma, address, *pte); |
| 1328 | BUG_ON(!page || PageAnon(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1329 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1330 | if (locked_vma) { |
| 1331 | mlock_vma_page(page); /* no-op if already mlocked */ |
| 1332 | if (page == check_page) |
| 1333 | ret = SWAP_MLOCK; |
| 1334 | continue; /* don't unmap */ |
| 1335 | } |
| 1336 | |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 1337 | if (ptep_clear_flush_young_notify(vma, address, pte)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1338 | continue; |
| 1339 | |
| 1340 | /* Nuke the page table entry. */ |
Ben Collins | eca3513 | 2005-11-29 11:45:26 -0800 | [diff] [blame] | 1341 | flush_cache_page(vma, address, pte_pfn(*pte)); |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 1342 | pteval = ptep_clear_flush_notify(vma, address, pte); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1343 | |
| 1344 | /* If nonlinear, store the file page offset in the pte. */ |
| 1345 | if (page->index != linear_page_index(vma, address)) |
| 1346 | set_pte_at(mm, address, pte, pgoff_to_pte(page->index)); |
| 1347 | |
| 1348 | /* Move the dirty bit to the physical page now the pte is gone. */ |
| 1349 | if (pte_dirty(pteval)) |
| 1350 | set_page_dirty(page); |
| 1351 | |
Hugh Dickins | edc315f | 2009-01-06 14:40:11 -0800 | [diff] [blame] | 1352 | page_remove_rmap(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1353 | page_cache_release(page); |
KAMEZAWA Hiroyuki | d559db0 | 2010-03-05 13:41:39 -0800 | [diff] [blame] | 1354 | dec_mm_counter(mm, MM_FILEPAGES); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1355 | (*mapcount)--; |
| 1356 | } |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 1357 | pte_unmap_unlock(pte - 1, ptl); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1358 | if (locked_vma) |
| 1359 | up_read(&vma->vm_mm->mmap_sem); |
| 1360 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1361 | } |
| 1362 | |
Andrea Arcangeli | 71e3aac | 2011-01-13 15:46:52 -0800 | [diff] [blame] | 1363 | bool is_vma_temporary_stack(struct vm_area_struct *vma) |
Mel Gorman | a8bef8ff | 2010-05-24 14:32:24 -0700 | [diff] [blame] | 1364 | { |
| 1365 | int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP); |
| 1366 | |
| 1367 | if (!maybe_stack) |
| 1368 | return false; |
| 1369 | |
| 1370 | if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) == |
| 1371 | VM_STACK_INCOMPLETE_SETUP) |
| 1372 | return true; |
| 1373 | |
| 1374 | return false; |
| 1375 | } |
| 1376 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1377 | /** |
| 1378 | * try_to_unmap_anon - unmap or unlock anonymous page using the object-based |
| 1379 | * rmap method |
| 1380 | * @page: the page to unmap/unlock |
Huang Shijie | 8051be5 | 2009-12-14 17:58:50 -0800 | [diff] [blame] | 1381 | * @flags: action and flags |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1382 | * |
| 1383 | * Find all the mappings of a page using the mapping pointer and the vma chains |
| 1384 | * contained in the anon_vma struct it points to. |
| 1385 | * |
| 1386 | * This function is only called from try_to_unmap/try_to_munlock for |
| 1387 | * anonymous pages. |
| 1388 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma |
| 1389 | * where the page was found will be held for write. So, we won't recheck |
| 1390 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be |
| 1391 | * 'LOCKED. |
| 1392 | */ |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1393 | static int try_to_unmap_anon(struct page *page, enum ttu_flags flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1394 | { |
| 1395 | struct anon_vma *anon_vma; |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 1396 | struct anon_vma_chain *avc; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1397 | int ret = SWAP_AGAIN; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1398 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1399 | anon_vma = page_lock_anon_vma(page); |
| 1400 | if (!anon_vma) |
| 1401 | return ret; |
| 1402 | |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 1403 | list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { |
| 1404 | struct vm_area_struct *vma = avc->vma; |
Mel Gorman | a8bef8ff | 2010-05-24 14:32:24 -0700 | [diff] [blame] | 1405 | unsigned long address; |
| 1406 | |
| 1407 | /* |
| 1408 | * During exec, a temporary VMA is setup and later moved. |
| 1409 | * The VMA is moved under the anon_vma lock but not the |
| 1410 | * page tables leading to a race where migration cannot |
| 1411 | * find the migration ptes. Rather than increasing the |
| 1412 | * locking requirements of exec(), migration skips |
| 1413 | * temporary VMAs until after exec() completes. |
| 1414 | */ |
| 1415 | if (PAGE_MIGRATION && (flags & TTU_MIGRATION) && |
| 1416 | is_vma_temporary_stack(vma)) |
| 1417 | continue; |
| 1418 | |
| 1419 | address = vma_address(page, vma); |
Hugh Dickins | 1cb1729 | 2009-12-14 17:59:01 -0800 | [diff] [blame] | 1420 | if (address == -EFAULT) |
| 1421 | continue; |
| 1422 | ret = try_to_unmap_one(page, vma, address, flags); |
Hugh Dickins | 53f79ac | 2009-12-14 17:58:58 -0800 | [diff] [blame] | 1423 | if (ret != SWAP_AGAIN || !page_mapped(page)) |
| 1424 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1425 | } |
Oleg Nesterov | 34bbd70 | 2007-02-28 20:13:49 -0800 | [diff] [blame] | 1426 | |
| 1427 | page_unlock_anon_vma(anon_vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1428 | return ret; |
| 1429 | } |
| 1430 | |
| 1431 | /** |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1432 | * try_to_unmap_file - unmap/unlock file page using the object-based rmap method |
| 1433 | * @page: the page to unmap/unlock |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1434 | * @flags: action and flags |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1435 | * |
| 1436 | * Find all the mappings of a page using the mapping pointer and the vma chains |
| 1437 | * contained in the address_space struct it points to. |
| 1438 | * |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1439 | * This function is only called from try_to_unmap/try_to_munlock for |
| 1440 | * object-based pages. |
| 1441 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma |
| 1442 | * where the page was found will be held for write. So, we won't recheck |
| 1443 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be |
| 1444 | * 'LOCKED. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1445 | */ |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1446 | static int try_to_unmap_file(struct page *page, enum ttu_flags flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1447 | { |
| 1448 | struct address_space *mapping = page->mapping; |
| 1449 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
| 1450 | struct vm_area_struct *vma; |
| 1451 | struct prio_tree_iter iter; |
| 1452 | int ret = SWAP_AGAIN; |
| 1453 | unsigned long cursor; |
| 1454 | unsigned long max_nl_cursor = 0; |
| 1455 | unsigned long max_nl_size = 0; |
| 1456 | unsigned int mapcount; |
| 1457 | |
Peter Zijlstra | 3d48ae4 | 2011-05-24 17:12:06 -0700 | [diff] [blame] | 1458 | mutex_lock(&mapping->i_mmap_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1459 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
Hugh Dickins | 1cb1729 | 2009-12-14 17:59:01 -0800 | [diff] [blame] | 1460 | unsigned long address = vma_address(page, vma); |
| 1461 | if (address == -EFAULT) |
| 1462 | continue; |
| 1463 | ret = try_to_unmap_one(page, vma, address, flags); |
Hugh Dickins | 53f79ac | 2009-12-14 17:58:58 -0800 | [diff] [blame] | 1464 | if (ret != SWAP_AGAIN || !page_mapped(page)) |
| 1465 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1466 | } |
| 1467 | |
| 1468 | if (list_empty(&mapping->i_mmap_nonlinear)) |
| 1469 | goto out; |
| 1470 | |
Hugh Dickins | 53f79ac | 2009-12-14 17:58:58 -0800 | [diff] [blame] | 1471 | /* |
| 1472 | * We don't bother to try to find the munlocked page in nonlinears. |
| 1473 | * It's costly. Instead, later, page reclaim logic may call |
| 1474 | * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily. |
| 1475 | */ |
| 1476 | if (TTU_ACTION(flags) == TTU_MUNLOCK) |
| 1477 | goto out; |
| 1478 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1479 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, |
| 1480 | shared.vm_set.list) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1481 | cursor = (unsigned long) vma->vm_private_data; |
| 1482 | if (cursor > max_nl_cursor) |
| 1483 | max_nl_cursor = cursor; |
| 1484 | cursor = vma->vm_end - vma->vm_start; |
| 1485 | if (cursor > max_nl_size) |
| 1486 | max_nl_size = cursor; |
| 1487 | } |
| 1488 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1489 | if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1490 | ret = SWAP_FAIL; |
| 1491 | goto out; |
| 1492 | } |
| 1493 | |
| 1494 | /* |
| 1495 | * We don't try to search for this page in the nonlinear vmas, |
| 1496 | * and page_referenced wouldn't have found it anyway. Instead |
| 1497 | * just walk the nonlinear vmas trying to age and unmap some. |
| 1498 | * The mapcount of the page we came in with is irrelevant, |
| 1499 | * but even so use it as a guide to how hard we should try? |
| 1500 | */ |
| 1501 | mapcount = page_mapcount(page); |
| 1502 | if (!mapcount) |
| 1503 | goto out; |
Peter Zijlstra | 3d48ae4 | 2011-05-24 17:12:06 -0700 | [diff] [blame] | 1504 | cond_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1505 | |
| 1506 | max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK; |
| 1507 | if (max_nl_cursor == 0) |
| 1508 | max_nl_cursor = CLUSTER_SIZE; |
| 1509 | |
| 1510 | do { |
| 1511 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, |
| 1512 | shared.vm_set.list) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1513 | cursor = (unsigned long) vma->vm_private_data; |
Hugh Dickins | 839b968 | 2005-09-03 15:54:43 -0700 | [diff] [blame] | 1514 | while ( cursor < max_nl_cursor && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1515 | cursor < vma->vm_end - vma->vm_start) { |
Hugh Dickins | 53f79ac | 2009-12-14 17:58:58 -0800 | [diff] [blame] | 1516 | if (try_to_unmap_cluster(cursor, &mapcount, |
| 1517 | vma, page) == SWAP_MLOCK) |
| 1518 | ret = SWAP_MLOCK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1519 | cursor += CLUSTER_SIZE; |
| 1520 | vma->vm_private_data = (void *) cursor; |
| 1521 | if ((int)mapcount <= 0) |
| 1522 | goto out; |
| 1523 | } |
| 1524 | vma->vm_private_data = (void *) max_nl_cursor; |
| 1525 | } |
Peter Zijlstra | 3d48ae4 | 2011-05-24 17:12:06 -0700 | [diff] [blame] | 1526 | cond_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1527 | max_nl_cursor += CLUSTER_SIZE; |
| 1528 | } while (max_nl_cursor <= max_nl_size); |
| 1529 | |
| 1530 | /* |
| 1531 | * Don't loop forever (perhaps all the remaining pages are |
| 1532 | * in locked vmas). Reset cursor on all unreserved nonlinear |
| 1533 | * vmas, now forgetting on which ones it had fallen behind. |
| 1534 | */ |
Hugh Dickins | 101d2be | 2005-11-21 21:32:16 -0800 | [diff] [blame] | 1535 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) |
| 1536 | vma->vm_private_data = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1537 | out: |
Peter Zijlstra | 3d48ae4 | 2011-05-24 17:12:06 -0700 | [diff] [blame] | 1538 | mutex_unlock(&mapping->i_mmap_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1539 | return ret; |
| 1540 | } |
| 1541 | |
| 1542 | /** |
| 1543 | * try_to_unmap - try to remove all page table mappings to a page |
| 1544 | * @page: the page to get unmapped |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1545 | * @flags: action and flags |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1546 | * |
| 1547 | * Tries to remove all the page table entries which are mapping this |
| 1548 | * page, used in the pageout path. Caller must hold the page lock. |
| 1549 | * Return values are: |
| 1550 | * |
| 1551 | * SWAP_SUCCESS - we succeeded in removing all mappings |
| 1552 | * SWAP_AGAIN - we missed a mapping, try again later |
| 1553 | * SWAP_FAIL - the page is unswappable |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1554 | * SWAP_MLOCK - page is mlocked. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1555 | */ |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1556 | int try_to_unmap(struct page *page, enum ttu_flags flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1557 | { |
| 1558 | int ret; |
| 1559 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1560 | BUG_ON(!PageLocked(page)); |
Andrea Arcangeli | 91600e9 | 2011-01-13 15:47:16 -0800 | [diff] [blame] | 1561 | VM_BUG_ON(!PageHuge(page) && PageTransHuge(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1562 | |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 1563 | if (unlikely(PageKsm(page))) |
| 1564 | ret = try_to_unmap_ksm(page, flags); |
| 1565 | else if (PageAnon(page)) |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1566 | ret = try_to_unmap_anon(page, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1567 | else |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1568 | ret = try_to_unmap_file(page, flags); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1569 | if (ret != SWAP_MLOCK && !page_mapped(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1570 | ret = SWAP_SUCCESS; |
| 1571 | return ret; |
| 1572 | } |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 1573 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1574 | /** |
| 1575 | * try_to_munlock - try to munlock a page |
| 1576 | * @page: the page to be munlocked |
| 1577 | * |
| 1578 | * Called from munlock code. Checks all of the VMAs mapping the page |
| 1579 | * to make sure nobody else has this page mlocked. The page will be |
| 1580 | * returned with PG_mlocked cleared if no other vmas have it mlocked. |
| 1581 | * |
| 1582 | * Return values are: |
| 1583 | * |
Hugh Dickins | 53f79ac | 2009-12-14 17:58:58 -0800 | [diff] [blame] | 1584 | * SWAP_AGAIN - no vma is holding page mlocked, or, |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1585 | * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 1586 | * SWAP_FAIL - page cannot be located at present |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1587 | * SWAP_MLOCK - page is now mlocked. |
| 1588 | */ |
| 1589 | int try_to_munlock(struct page *page) |
| 1590 | { |
| 1591 | VM_BUG_ON(!PageLocked(page) || PageLRU(page)); |
| 1592 | |
Hugh Dickins | 5ad6468 | 2009-12-14 17:59:24 -0800 | [diff] [blame] | 1593 | if (unlikely(PageKsm(page))) |
| 1594 | return try_to_unmap_ksm(page, TTU_MUNLOCK); |
| 1595 | else if (PageAnon(page)) |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1596 | return try_to_unmap_anon(page, TTU_MUNLOCK); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1597 | else |
Andi Kleen | 14fa31b | 2009-09-16 11:50:10 +0200 | [diff] [blame] | 1598 | return try_to_unmap_file(page, TTU_MUNLOCK); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1599 | } |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 1600 | |
Peter Zijlstra | 01d8b20 | 2011-03-22 16:32:49 -0700 | [diff] [blame] | 1601 | void __put_anon_vma(struct anon_vma *anon_vma) |
Rik van Riel | 7654506 | 2010-08-09 17:18:41 -0700 | [diff] [blame] | 1602 | { |
Peter Zijlstra | 01d8b20 | 2011-03-22 16:32:49 -0700 | [diff] [blame] | 1603 | struct anon_vma *root = anon_vma->root; |
Rik van Riel | 7654506 | 2010-08-09 17:18:41 -0700 | [diff] [blame] | 1604 | |
Peter Zijlstra | 01d8b20 | 2011-03-22 16:32:49 -0700 | [diff] [blame] | 1605 | if (root != anon_vma && atomic_dec_and_test(&root->refcount)) |
| 1606 | anon_vma_free(root); |
Rik van Riel | 7654506 | 2010-08-09 17:18:41 -0700 | [diff] [blame] | 1607 | |
Peter Zijlstra | 01d8b20 | 2011-03-22 16:32:49 -0700 | [diff] [blame] | 1608 | anon_vma_free(anon_vma); |
Rik van Riel | 7654506 | 2010-08-09 17:18:41 -0700 | [diff] [blame] | 1609 | } |
Rik van Riel | 7654506 | 2010-08-09 17:18:41 -0700 | [diff] [blame] | 1610 | |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 1611 | #ifdef CONFIG_MIGRATION |
| 1612 | /* |
| 1613 | * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file(): |
| 1614 | * Called by migrate.c to remove migration ptes, but might be used more later. |
| 1615 | */ |
| 1616 | static int rmap_walk_anon(struct page *page, int (*rmap_one)(struct page *, |
| 1617 | struct vm_area_struct *, unsigned long, void *), void *arg) |
| 1618 | { |
| 1619 | struct anon_vma *anon_vma; |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 1620 | struct anon_vma_chain *avc; |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 1621 | int ret = SWAP_AGAIN; |
| 1622 | |
| 1623 | /* |
| 1624 | * Note: remove_migration_ptes() cannot use page_lock_anon_vma() |
| 1625 | * because that depends on page_mapped(); but not all its usages |
Mel Gorman | 3f6c827 | 2010-05-24 14:32:17 -0700 | [diff] [blame] | 1626 | * are holding mmap_sem. Users without mmap_sem are required to |
| 1627 | * take a reference count to prevent the anon_vma disappearing |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 1628 | */ |
| 1629 | anon_vma = page_anon_vma(page); |
| 1630 | if (!anon_vma) |
| 1631 | return ret; |
Rik van Riel | cba48b9 | 2010-08-09 17:18:38 -0700 | [diff] [blame] | 1632 | anon_vma_lock(anon_vma); |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 1633 | list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { |
| 1634 | struct vm_area_struct *vma = avc->vma; |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 1635 | unsigned long address = vma_address(page, vma); |
| 1636 | if (address == -EFAULT) |
| 1637 | continue; |
| 1638 | ret = rmap_one(page, vma, address, arg); |
| 1639 | if (ret != SWAP_AGAIN) |
| 1640 | break; |
| 1641 | } |
Rik van Riel | cba48b9 | 2010-08-09 17:18:38 -0700 | [diff] [blame] | 1642 | anon_vma_unlock(anon_vma); |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 1643 | return ret; |
| 1644 | } |
| 1645 | |
| 1646 | static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *, |
| 1647 | struct vm_area_struct *, unsigned long, void *), void *arg) |
| 1648 | { |
| 1649 | struct address_space *mapping = page->mapping; |
| 1650 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
| 1651 | struct vm_area_struct *vma; |
| 1652 | struct prio_tree_iter iter; |
| 1653 | int ret = SWAP_AGAIN; |
| 1654 | |
| 1655 | if (!mapping) |
| 1656 | return ret; |
Peter Zijlstra | 3d48ae4 | 2011-05-24 17:12:06 -0700 | [diff] [blame] | 1657 | mutex_lock(&mapping->i_mmap_mutex); |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 1658 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
| 1659 | unsigned long address = vma_address(page, vma); |
| 1660 | if (address == -EFAULT) |
| 1661 | continue; |
| 1662 | ret = rmap_one(page, vma, address, arg); |
| 1663 | if (ret != SWAP_AGAIN) |
| 1664 | break; |
| 1665 | } |
| 1666 | /* |
| 1667 | * No nonlinear handling: being always shared, nonlinear vmas |
| 1668 | * never contain migration ptes. Decide what to do about this |
| 1669 | * limitation to linear when we need rmap_walk() on nonlinear. |
| 1670 | */ |
Peter Zijlstra | 3d48ae4 | 2011-05-24 17:12:06 -0700 | [diff] [blame] | 1671 | mutex_unlock(&mapping->i_mmap_mutex); |
Hugh Dickins | e9995ef | 2009-12-14 17:59:31 -0800 | [diff] [blame] | 1672 | return ret; |
| 1673 | } |
| 1674 | |
| 1675 | int rmap_walk(struct page *page, int (*rmap_one)(struct page *, |
| 1676 | struct vm_area_struct *, unsigned long, void *), void *arg) |
| 1677 | { |
| 1678 | VM_BUG_ON(!PageLocked(page)); |
| 1679 | |
| 1680 | if (unlikely(PageKsm(page))) |
| 1681 | return rmap_walk_ksm(page, rmap_one, arg); |
| 1682 | else if (PageAnon(page)) |
| 1683 | return rmap_walk_anon(page, rmap_one, arg); |
| 1684 | else |
| 1685 | return rmap_walk_file(page, rmap_one, arg); |
| 1686 | } |
| 1687 | #endif /* CONFIG_MIGRATION */ |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 1688 | |
Naoya Horiguchi | e3390f6 | 2010-06-15 13:18:13 +0900 | [diff] [blame] | 1689 | #ifdef CONFIG_HUGETLB_PAGE |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 1690 | /* |
| 1691 | * The following three functions are for anonymous (private mapped) hugepages. |
| 1692 | * Unlike common anonymous pages, anonymous hugepages have no accounting code |
| 1693 | * and no lru code, because we handle hugepages differently from common pages. |
| 1694 | */ |
| 1695 | static void __hugepage_set_anon_rmap(struct page *page, |
| 1696 | struct vm_area_struct *vma, unsigned long address, int exclusive) |
| 1697 | { |
| 1698 | struct anon_vma *anon_vma = vma->anon_vma; |
Naoya Horiguchi | 433abed | 2010-09-10 13:23:03 +0900 | [diff] [blame] | 1699 | |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 1700 | BUG_ON(!anon_vma); |
Naoya Horiguchi | 433abed | 2010-09-10 13:23:03 +0900 | [diff] [blame] | 1701 | |
| 1702 | if (PageAnon(page)) |
| 1703 | return; |
| 1704 | if (!exclusive) |
| 1705 | anon_vma = anon_vma->root; |
| 1706 | |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 1707 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
| 1708 | page->mapping = (struct address_space *) anon_vma; |
| 1709 | page->index = linear_page_index(vma, address); |
| 1710 | } |
| 1711 | |
| 1712 | void hugepage_add_anon_rmap(struct page *page, |
| 1713 | struct vm_area_struct *vma, unsigned long address) |
| 1714 | { |
| 1715 | struct anon_vma *anon_vma = vma->anon_vma; |
| 1716 | int first; |
Naoya Horiguchi | a850ea3 | 2010-09-10 13:23:06 +0900 | [diff] [blame] | 1717 | |
| 1718 | BUG_ON(!PageLocked(page)); |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 1719 | BUG_ON(!anon_vma); |
Hugh Dickins | 5dbe0af | 2011-05-28 13:17:04 -0700 | [diff] [blame] | 1720 | /* address might be in next vma when migration races vma_adjust */ |
Naoya Horiguchi | 0fe6e20 | 2010-05-28 09:29:16 +0900 | [diff] [blame] | 1721 | first = atomic_inc_and_test(&page->_mapcount); |
| 1722 | if (first) |
| 1723 | __hugepage_set_anon_rmap(page, vma, address, 0); |
| 1724 | } |
| 1725 | |
| 1726 | void hugepage_add_new_anon_rmap(struct page *page, |
| 1727 | struct vm_area_struct *vma, unsigned long address) |
| 1728 | { |
| 1729 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
| 1730 | atomic_set(&page->_mapcount, 0); |
| 1731 | __hugepage_set_anon_rmap(page, vma, address, 1); |
| 1732 | } |
Naoya Horiguchi | e3390f6 | 2010-06-15 13:18:13 +0900 | [diff] [blame] | 1733 | #endif /* CONFIG_HUGETLB_PAGE */ |