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 |
| 17 | * Contributions by Hugh Dickins <hugh@veritas.com> 2003, 2004 |
| 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) |
| 27 | * mapping->i_mmap_lock |
| 28 | * anon_vma->lock |
| 29 | * mm->page_table_lock or pte_lock |
| 30 | * zone->lru_lock (in mark_page_accessed, isolate_lru_page) |
| 31 | * swap_lock (in swap_duplicate, swap_info_get) |
| 32 | * mmlist_lock (in mmput, drain_mmlist and others) |
| 33 | * mapping->private_lock (in __set_page_dirty_buffers) |
| 34 | * inode_lock (in set_page_dirty's __mark_inode_dirty) |
| 35 | * sb_lock (within inode_lock in fs/fs-writeback.c) |
| 36 | * mapping->tree_lock (widely used, in set_page_dirty, |
| 37 | * in arch-dependent flush_dcache_mmap_lock, |
| 38 | * within inode_lock in __sync_single_inode) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 39 | */ |
| 40 | |
| 41 | #include <linux/mm.h> |
| 42 | #include <linux/pagemap.h> |
| 43 | #include <linux/swap.h> |
| 44 | #include <linux/swapops.h> |
| 45 | #include <linux/slab.h> |
| 46 | #include <linux/init.h> |
| 47 | #include <linux/rmap.h> |
| 48 | #include <linux/rcupdate.h> |
Christoph Lameter | a48d07a | 2006-02-01 03:05:38 -0800 | [diff] [blame] | 49 | #include <linux/module.h> |
Hugh Dickins | b5934c5 | 2009-01-06 14:39:25 -0800 | [diff] [blame^] | 50 | #include <linux/mm_inline.h> |
Nick Piggin | 7de6b80 | 2006-12-22 01:09:33 -0800 | [diff] [blame] | 51 | #include <linux/kallsyms.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 52 | #include <linux/memcontrol.h> |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 53 | #include <linux/mmu_notifier.h> |
KOSAKI Motohiro | 64cdd54 | 2009-01-06 14:39:16 -0800 | [diff] [blame] | 54 | #include <linux/migrate.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 55 | |
| 56 | #include <asm/tlbflush.h> |
| 57 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 58 | #include "internal.h" |
| 59 | |
Adrian Bunk | fdd2e5f | 2008-10-18 20:28:38 -0700 | [diff] [blame] | 60 | static struct kmem_cache *anon_vma_cachep; |
| 61 | |
| 62 | static inline struct anon_vma *anon_vma_alloc(void) |
| 63 | { |
| 64 | return kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL); |
| 65 | } |
| 66 | |
| 67 | static inline void anon_vma_free(struct anon_vma *anon_vma) |
| 68 | { |
| 69 | kmem_cache_free(anon_vma_cachep, anon_vma); |
| 70 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 71 | |
Linus Torvalds | d9d332e | 2008-10-19 10:32:20 -0700 | [diff] [blame] | 72 | /** |
| 73 | * anon_vma_prepare - attach an anon_vma to a memory region |
| 74 | * @vma: the memory region in question |
| 75 | * |
| 76 | * This makes sure the memory mapping described by 'vma' has |
| 77 | * an 'anon_vma' attached to it, so that we can associate the |
| 78 | * anonymous pages mapped into it with that anon_vma. |
| 79 | * |
| 80 | * The common case will be that we already have one, but if |
| 81 | * if not we either need to find an adjacent mapping that we |
| 82 | * can re-use the anon_vma from (very common when the only |
| 83 | * reason for splitting a vma has been mprotect()), or we |
| 84 | * allocate a new one. |
| 85 | * |
| 86 | * Anon-vma allocations are very subtle, because we may have |
| 87 | * optimistically looked up an anon_vma in page_lock_anon_vma() |
| 88 | * and that may actually touch the spinlock even in the newly |
| 89 | * allocated vma (it depends on RCU to make sure that the |
| 90 | * anon_vma isn't actually destroyed). |
| 91 | * |
| 92 | * As a result, we need to do proper anon_vma locking even |
| 93 | * for the new allocation. At the same time, we do not want |
| 94 | * to do any locking for the common case of already having |
| 95 | * an anon_vma. |
| 96 | * |
| 97 | * This must be called with the mmap_sem held for reading. |
| 98 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 99 | int anon_vma_prepare(struct vm_area_struct *vma) |
| 100 | { |
| 101 | struct anon_vma *anon_vma = vma->anon_vma; |
| 102 | |
| 103 | might_sleep(); |
| 104 | if (unlikely(!anon_vma)) { |
| 105 | struct mm_struct *mm = vma->vm_mm; |
Linus Torvalds | d9d332e | 2008-10-19 10:32:20 -0700 | [diff] [blame] | 106 | struct anon_vma *allocated; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 107 | |
| 108 | anon_vma = find_mergeable_anon_vma(vma); |
Linus Torvalds | d9d332e | 2008-10-19 10:32:20 -0700 | [diff] [blame] | 109 | allocated = NULL; |
| 110 | if (!anon_vma) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 111 | anon_vma = anon_vma_alloc(); |
| 112 | if (unlikely(!anon_vma)) |
| 113 | return -ENOMEM; |
| 114 | allocated = anon_vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 115 | } |
Linus Torvalds | d9d332e | 2008-10-19 10:32:20 -0700 | [diff] [blame] | 116 | spin_lock(&anon_vma->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 117 | |
| 118 | /* page_table_lock to protect against threads */ |
| 119 | spin_lock(&mm->page_table_lock); |
| 120 | if (likely(!vma->anon_vma)) { |
| 121 | vma->anon_vma = anon_vma; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 122 | list_add_tail(&vma->anon_vma_node, &anon_vma->head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 123 | allocated = NULL; |
| 124 | } |
| 125 | spin_unlock(&mm->page_table_lock); |
| 126 | |
Linus Torvalds | d9d332e | 2008-10-19 10:32:20 -0700 | [diff] [blame] | 127 | spin_unlock(&anon_vma->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 128 | if (unlikely(allocated)) |
| 129 | anon_vma_free(allocated); |
| 130 | } |
| 131 | return 0; |
| 132 | } |
| 133 | |
| 134 | void __anon_vma_merge(struct vm_area_struct *vma, struct vm_area_struct *next) |
| 135 | { |
| 136 | BUG_ON(vma->anon_vma != next->anon_vma); |
| 137 | list_del(&next->anon_vma_node); |
| 138 | } |
| 139 | |
| 140 | void __anon_vma_link(struct vm_area_struct *vma) |
| 141 | { |
| 142 | struct anon_vma *anon_vma = vma->anon_vma; |
| 143 | |
Hugh Dickins | 30acbab | 2007-06-27 14:09:53 -0700 | [diff] [blame] | 144 | if (anon_vma) |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 145 | list_add_tail(&vma->anon_vma_node, &anon_vma->head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 146 | } |
| 147 | |
| 148 | void anon_vma_link(struct vm_area_struct *vma) |
| 149 | { |
| 150 | struct anon_vma *anon_vma = vma->anon_vma; |
| 151 | |
| 152 | if (anon_vma) { |
| 153 | spin_lock(&anon_vma->lock); |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 154 | list_add_tail(&vma->anon_vma_node, &anon_vma->head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 155 | spin_unlock(&anon_vma->lock); |
| 156 | } |
| 157 | } |
| 158 | |
| 159 | void anon_vma_unlink(struct vm_area_struct *vma) |
| 160 | { |
| 161 | struct anon_vma *anon_vma = vma->anon_vma; |
| 162 | int empty; |
| 163 | |
| 164 | if (!anon_vma) |
| 165 | return; |
| 166 | |
| 167 | spin_lock(&anon_vma->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 168 | list_del(&vma->anon_vma_node); |
| 169 | |
| 170 | /* We must garbage collect the anon_vma if it's empty */ |
| 171 | empty = list_empty(&anon_vma->head); |
| 172 | spin_unlock(&anon_vma->lock); |
| 173 | |
| 174 | if (empty) |
| 175 | anon_vma_free(anon_vma); |
| 176 | } |
| 177 | |
Alexey Dobriyan | 51cc506 | 2008-07-25 19:45:34 -0700 | [diff] [blame] | 178 | static void anon_vma_ctor(void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 179 | { |
Christoph Lameter | a35afb8 | 2007-05-16 22:10:57 -0700 | [diff] [blame] | 180 | struct anon_vma *anon_vma = data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 181 | |
Christoph Lameter | a35afb8 | 2007-05-16 22:10:57 -0700 | [diff] [blame] | 182 | spin_lock_init(&anon_vma->lock); |
| 183 | INIT_LIST_HEAD(&anon_vma->head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 184 | } |
| 185 | |
| 186 | void __init anon_vma_init(void) |
| 187 | { |
| 188 | anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), |
Paul Mundt | 20c2df8 | 2007-07-20 10:11:58 +0900 | [diff] [blame] | 189 | 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 190 | } |
| 191 | |
| 192 | /* |
| 193 | * Getting a lock on a stable anon_vma from a page off the LRU is |
| 194 | * tricky: page_lock_anon_vma rely on RCU to guard against the races. |
| 195 | */ |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 196 | struct anon_vma *page_lock_anon_vma(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 197 | { |
Oleg Nesterov | 34bbd70 | 2007-02-28 20:13:49 -0800 | [diff] [blame] | 198 | struct anon_vma *anon_vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 199 | unsigned long anon_mapping; |
| 200 | |
| 201 | rcu_read_lock(); |
| 202 | anon_mapping = (unsigned long) page->mapping; |
| 203 | if (!(anon_mapping & PAGE_MAPPING_ANON)) |
| 204 | goto out; |
| 205 | if (!page_mapped(page)) |
| 206 | goto out; |
| 207 | |
| 208 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); |
| 209 | spin_lock(&anon_vma->lock); |
Oleg Nesterov | 34bbd70 | 2007-02-28 20:13:49 -0800 | [diff] [blame] | 210 | return anon_vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 211 | out: |
| 212 | rcu_read_unlock(); |
Oleg Nesterov | 34bbd70 | 2007-02-28 20:13:49 -0800 | [diff] [blame] | 213 | return NULL; |
| 214 | } |
| 215 | |
Lee Schermerhorn | af936a1 | 2008-10-18 20:26:53 -0700 | [diff] [blame] | 216 | void page_unlock_anon_vma(struct anon_vma *anon_vma) |
Oleg Nesterov | 34bbd70 | 2007-02-28 20:13:49 -0800 | [diff] [blame] | 217 | { |
| 218 | spin_unlock(&anon_vma->lock); |
| 219 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 220 | } |
| 221 | |
| 222 | /* |
Lee Schermerhorn | 3ad33b2 | 2007-11-14 16:59:10 -0800 | [diff] [blame] | 223 | * At what user virtual address is page expected in @vma? |
| 224 | * Returns virtual address or -EFAULT if page's index/offset is not |
| 225 | * within the range mapped the @vma. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 226 | */ |
| 227 | static inline unsigned long |
| 228 | vma_address(struct page *page, struct vm_area_struct *vma) |
| 229 | { |
| 230 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
| 231 | unsigned long address; |
| 232 | |
| 233 | address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
| 234 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) { |
Lee Schermerhorn | 3ad33b2 | 2007-11-14 16:59:10 -0800 | [diff] [blame] | 235 | /* page should be within @vma mapping range */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 236 | return -EFAULT; |
| 237 | } |
| 238 | return address; |
| 239 | } |
| 240 | |
| 241 | /* |
| 242 | * At what user virtual address is page expected in vma? checking that the |
Hugh Dickins | ee498ed | 2005-11-21 21:32:18 -0800 | [diff] [blame] | 243 | * page matches the vma: currently only used on anon pages, by unuse_vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 244 | */ |
| 245 | unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) |
| 246 | { |
| 247 | if (PageAnon(page)) { |
| 248 | if ((void *)vma->anon_vma != |
| 249 | (void *)page->mapping - PAGE_MAPPING_ANON) |
| 250 | return -EFAULT; |
| 251 | } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) { |
Hugh Dickins | ee498ed | 2005-11-21 21:32:18 -0800 | [diff] [blame] | 252 | if (!vma->vm_file || |
| 253 | vma->vm_file->f_mapping != page->mapping) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 254 | return -EFAULT; |
| 255 | } else |
| 256 | return -EFAULT; |
| 257 | return vma_address(page, vma); |
| 258 | } |
| 259 | |
| 260 | /* |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 261 | * Check that @page is mapped at @address into @mm. |
| 262 | * |
Nick Piggin | 479db0b | 2008-08-20 14:09:18 -0700 | [diff] [blame] | 263 | * If @sync is false, page_check_address may perform a racy check to avoid |
| 264 | * the page table lock when the pte is not present (helpful when reclaiming |
| 265 | * highly shared pages). |
| 266 | * |
Hugh Dickins | b8072f0 | 2005-10-29 18:16:41 -0700 | [diff] [blame] | 267 | * On success returns with pte mapped and locked. |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 268 | */ |
Carsten Otte | ceffc07 | 2005-06-23 22:05:25 -0700 | [diff] [blame] | 269 | pte_t *page_check_address(struct page *page, struct mm_struct *mm, |
Nick Piggin | 479db0b | 2008-08-20 14:09:18 -0700 | [diff] [blame] | 270 | unsigned long address, spinlock_t **ptlp, int sync) |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 271 | { |
| 272 | pgd_t *pgd; |
| 273 | pud_t *pud; |
| 274 | pmd_t *pmd; |
| 275 | pte_t *pte; |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 276 | spinlock_t *ptl; |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 277 | |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 278 | pgd = pgd_offset(mm, address); |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 279 | if (!pgd_present(*pgd)) |
| 280 | return NULL; |
| 281 | |
| 282 | pud = pud_offset(pgd, address); |
| 283 | if (!pud_present(*pud)) |
| 284 | return NULL; |
| 285 | |
| 286 | pmd = pmd_offset(pud, address); |
| 287 | if (!pmd_present(*pmd)) |
| 288 | return NULL; |
| 289 | |
| 290 | pte = pte_offset_map(pmd, address); |
| 291 | /* Make a quick check before getting the lock */ |
Nick Piggin | 479db0b | 2008-08-20 14:09:18 -0700 | [diff] [blame] | 292 | if (!sync && !pte_present(*pte)) { |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 293 | pte_unmap(pte); |
| 294 | return NULL; |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 295 | } |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 296 | |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 297 | ptl = pte_lockptr(mm, pmd); |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 298 | spin_lock(ptl); |
| 299 | if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) { |
| 300 | *ptlp = ptl; |
| 301 | return pte; |
| 302 | } |
| 303 | pte_unmap_unlock(pte, ptl); |
| 304 | return NULL; |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 305 | } |
| 306 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 307 | /** |
| 308 | * page_mapped_in_vma - check whether a page is really mapped in a VMA |
| 309 | * @page: the page to test |
| 310 | * @vma: the VMA to test |
| 311 | * |
| 312 | * Returns 1 if the page is mapped into the page tables of the VMA, 0 |
| 313 | * if the page is not mapped into the page tables of this VMA. Only |
| 314 | * valid for normal file or anonymous VMAs. |
| 315 | */ |
| 316 | static int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma) |
| 317 | { |
| 318 | unsigned long address; |
| 319 | pte_t *pte; |
| 320 | spinlock_t *ptl; |
| 321 | |
| 322 | address = vma_address(page, vma); |
| 323 | if (address == -EFAULT) /* out of vma range */ |
| 324 | return 0; |
| 325 | pte = page_check_address(page, vma->vm_mm, address, &ptl, 1); |
| 326 | if (!pte) /* the page is not in this mm */ |
| 327 | return 0; |
| 328 | pte_unmap_unlock(pte, ptl); |
| 329 | |
| 330 | return 1; |
| 331 | } |
| 332 | |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 333 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 334 | * Subfunctions of page_referenced: page_referenced_one called |
| 335 | * repeatedly from either page_referenced_anon or page_referenced_file. |
| 336 | */ |
| 337 | static int page_referenced_one(struct page *page, |
Rik van Riel | f7b7fd8 | 2005-11-28 13:44:07 -0800 | [diff] [blame] | 338 | struct vm_area_struct *vma, unsigned int *mapcount) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 339 | { |
| 340 | struct mm_struct *mm = vma->vm_mm; |
| 341 | unsigned long address; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 342 | pte_t *pte; |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 343 | spinlock_t *ptl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 344 | int referenced = 0; |
| 345 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 346 | address = vma_address(page, vma); |
| 347 | if (address == -EFAULT) |
| 348 | goto out; |
| 349 | |
Nick Piggin | 479db0b | 2008-08-20 14:09:18 -0700 | [diff] [blame] | 350 | pte = page_check_address(page, mm, address, &ptl, 0); |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 351 | if (!pte) |
| 352 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 353 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 354 | /* |
| 355 | * Don't want to elevate referenced for mlocked page that gets this far, |
| 356 | * in order that it progresses to try_to_unmap and is moved to the |
| 357 | * unevictable list. |
| 358 | */ |
Hugh Dickins | 5a9bbdc | 2008-02-04 22:29:23 -0800 | [diff] [blame] | 359 | if (vma->vm_flags & VM_LOCKED) { |
Hugh Dickins | 5a9bbdc | 2008-02-04 22:29:23 -0800 | [diff] [blame] | 360 | *mapcount = 1; /* break early from loop */ |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 361 | goto out_unmap; |
| 362 | } |
| 363 | |
Johannes Weiner | 4917e5d | 2009-01-06 14:39:17 -0800 | [diff] [blame] | 364 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
| 365 | /* |
| 366 | * Don't treat a reference through a sequentially read |
| 367 | * mapping as such. If the page has been used in |
| 368 | * another mapping, we will catch it; if this other |
| 369 | * mapping is already gone, the unmap path will have |
| 370 | * set PG_referenced or activated the page. |
| 371 | */ |
| 372 | if (likely(!VM_SequentialReadHint(vma))) |
| 373 | referenced++; |
| 374 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 375 | |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 376 | /* Pretend the page is referenced if the task has the |
| 377 | swap token and is in the middle of a page fault. */ |
Rik van Riel | f7b7fd8 | 2005-11-28 13:44:07 -0800 | [diff] [blame] | 378 | if (mm != current->mm && has_swap_token(mm) && |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 379 | rwsem_is_locked(&mm->mmap_sem)) |
| 380 | referenced++; |
| 381 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 382 | out_unmap: |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 383 | (*mapcount)--; |
| 384 | pte_unmap_unlock(pte, ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 385 | out: |
| 386 | return referenced; |
| 387 | } |
| 388 | |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 389 | static int page_referenced_anon(struct page *page, |
| 390 | struct mem_cgroup *mem_cont) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 391 | { |
| 392 | unsigned int mapcount; |
| 393 | struct anon_vma *anon_vma; |
| 394 | struct vm_area_struct *vma; |
| 395 | int referenced = 0; |
| 396 | |
| 397 | anon_vma = page_lock_anon_vma(page); |
| 398 | if (!anon_vma) |
| 399 | return referenced; |
| 400 | |
| 401 | mapcount = page_mapcount(page); |
| 402 | list_for_each_entry(vma, &anon_vma->head, anon_vma_node) { |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 403 | /* |
| 404 | * If we are reclaiming on behalf of a cgroup, skip |
| 405 | * counting on behalf of references from different |
| 406 | * cgroups |
| 407 | */ |
Hugh Dickins | bd845e3 | 2008-03-04 14:29:01 -0800 | [diff] [blame] | 408 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 409 | continue; |
Rik van Riel | f7b7fd8 | 2005-11-28 13:44:07 -0800 | [diff] [blame] | 410 | referenced += page_referenced_one(page, vma, &mapcount); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 411 | if (!mapcount) |
| 412 | break; |
| 413 | } |
Oleg Nesterov | 34bbd70 | 2007-02-28 20:13:49 -0800 | [diff] [blame] | 414 | |
| 415 | page_unlock_anon_vma(anon_vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 416 | return referenced; |
| 417 | } |
| 418 | |
| 419 | /** |
| 420 | * page_referenced_file - referenced check for object-based rmap |
| 421 | * @page: the page we're checking references on. |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 422 | * @mem_cont: target memory controller |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 423 | * |
| 424 | * For an object-based mapped page, find all the places it is mapped and |
| 425 | * check/clear the referenced flag. This is done by following the page->mapping |
| 426 | * pointer, then walking the chain of vmas it holds. It returns the number |
| 427 | * of references it found. |
| 428 | * |
| 429 | * This function is only called from page_referenced for object-based pages. |
| 430 | */ |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 431 | static int page_referenced_file(struct page *page, |
| 432 | struct mem_cgroup *mem_cont) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 433 | { |
| 434 | unsigned int mapcount; |
| 435 | struct address_space *mapping = page->mapping; |
| 436 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
| 437 | struct vm_area_struct *vma; |
| 438 | struct prio_tree_iter iter; |
| 439 | int referenced = 0; |
| 440 | |
| 441 | /* |
| 442 | * The caller's checks on page->mapping and !PageAnon have made |
| 443 | * sure that this is a file page: the check for page->mapping |
| 444 | * excludes the case just before it gets set on an anon page. |
| 445 | */ |
| 446 | BUG_ON(PageAnon(page)); |
| 447 | |
| 448 | /* |
| 449 | * The page lock not only makes sure that page->mapping cannot |
| 450 | * suddenly be NULLified by truncation, it makes sure that the |
| 451 | * structure at mapping cannot be freed and reused yet, |
| 452 | * so we can safely take mapping->i_mmap_lock. |
| 453 | */ |
| 454 | BUG_ON(!PageLocked(page)); |
| 455 | |
| 456 | spin_lock(&mapping->i_mmap_lock); |
| 457 | |
| 458 | /* |
| 459 | * i_mmap_lock does not stabilize mapcount at all, but mapcount |
| 460 | * is more likely to be accurate if we note it after spinning. |
| 461 | */ |
| 462 | mapcount = page_mapcount(page); |
| 463 | |
| 464 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 465 | /* |
| 466 | * If we are reclaiming on behalf of a cgroup, skip |
| 467 | * counting on behalf of references from different |
| 468 | * cgroups |
| 469 | */ |
Hugh Dickins | bd845e3 | 2008-03-04 14:29:01 -0800 | [diff] [blame] | 470 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 471 | continue; |
Rik van Riel | f7b7fd8 | 2005-11-28 13:44:07 -0800 | [diff] [blame] | 472 | referenced += page_referenced_one(page, vma, &mapcount); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 473 | if (!mapcount) |
| 474 | break; |
| 475 | } |
| 476 | |
| 477 | spin_unlock(&mapping->i_mmap_lock); |
| 478 | return referenced; |
| 479 | } |
| 480 | |
| 481 | /** |
| 482 | * page_referenced - test if the page was referenced |
| 483 | * @page: the page to test |
| 484 | * @is_locked: caller holds lock on the page |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 485 | * @mem_cont: target memory controller |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 486 | * |
| 487 | * Quick test_and_clear_referenced for all mappings to a page, |
| 488 | * returns the number of ptes which referenced the page. |
| 489 | */ |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 490 | int page_referenced(struct page *page, int is_locked, |
| 491 | struct mem_cgroup *mem_cont) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 492 | { |
| 493 | int referenced = 0; |
| 494 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 495 | if (TestClearPageReferenced(page)) |
| 496 | referenced++; |
| 497 | |
| 498 | if (page_mapped(page) && page->mapping) { |
| 499 | if (PageAnon(page)) |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 500 | referenced += page_referenced_anon(page, mem_cont); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 501 | else if (is_locked) |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 502 | referenced += page_referenced_file(page, mem_cont); |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 503 | else if (!trylock_page(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 504 | referenced++; |
| 505 | else { |
| 506 | if (page->mapping) |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 507 | referenced += |
| 508 | page_referenced_file(page, mem_cont); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 509 | unlock_page(page); |
| 510 | } |
| 511 | } |
Christian Borntraeger | 5b7baf0 | 2008-03-25 18:47:12 +0100 | [diff] [blame] | 512 | |
| 513 | if (page_test_and_clear_young(page)) |
| 514 | referenced++; |
| 515 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 516 | return referenced; |
| 517 | } |
| 518 | |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 519 | static int page_mkclean_one(struct page *page, struct vm_area_struct *vma) |
| 520 | { |
| 521 | struct mm_struct *mm = vma->vm_mm; |
| 522 | unsigned long address; |
Peter Zijlstra | c2fda5f | 2006-12-22 14:25:52 +0100 | [diff] [blame] | 523 | pte_t *pte; |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 524 | spinlock_t *ptl; |
| 525 | int ret = 0; |
| 526 | |
| 527 | address = vma_address(page, vma); |
| 528 | if (address == -EFAULT) |
| 529 | goto out; |
| 530 | |
Nick Piggin | 479db0b | 2008-08-20 14:09:18 -0700 | [diff] [blame] | 531 | pte = page_check_address(page, mm, address, &ptl, 1); |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 532 | if (!pte) |
| 533 | goto out; |
| 534 | |
Peter Zijlstra | c2fda5f | 2006-12-22 14:25:52 +0100 | [diff] [blame] | 535 | if (pte_dirty(*pte) || pte_write(*pte)) { |
| 536 | pte_t entry; |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 537 | |
Peter Zijlstra | c2fda5f | 2006-12-22 14:25:52 +0100 | [diff] [blame] | 538 | flush_cache_page(vma, address, pte_pfn(*pte)); |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 539 | entry = ptep_clear_flush_notify(vma, address, pte); |
Peter Zijlstra | c2fda5f | 2006-12-22 14:25:52 +0100 | [diff] [blame] | 540 | entry = pte_wrprotect(entry); |
| 541 | entry = pte_mkclean(entry); |
Al Viro | d6e88e6 | 2006-12-29 16:48:35 -0800 | [diff] [blame] | 542 | set_pte_at(mm, address, pte, entry); |
Peter Zijlstra | c2fda5f | 2006-12-22 14:25:52 +0100 | [diff] [blame] | 543 | ret = 1; |
| 544 | } |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 545 | |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 546 | pte_unmap_unlock(pte, ptl); |
| 547 | out: |
| 548 | return ret; |
| 549 | } |
| 550 | |
| 551 | static int page_mkclean_file(struct address_space *mapping, struct page *page) |
| 552 | { |
| 553 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
| 554 | struct vm_area_struct *vma; |
| 555 | struct prio_tree_iter iter; |
| 556 | int ret = 0; |
| 557 | |
| 558 | BUG_ON(PageAnon(page)); |
| 559 | |
| 560 | spin_lock(&mapping->i_mmap_lock); |
| 561 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
| 562 | if (vma->vm_flags & VM_SHARED) |
| 563 | ret += page_mkclean_one(page, vma); |
| 564 | } |
| 565 | spin_unlock(&mapping->i_mmap_lock); |
| 566 | return ret; |
| 567 | } |
| 568 | |
| 569 | int page_mkclean(struct page *page) |
| 570 | { |
| 571 | int ret = 0; |
| 572 | |
| 573 | BUG_ON(!PageLocked(page)); |
| 574 | |
| 575 | if (page_mapped(page)) { |
| 576 | struct address_space *mapping = page_mapping(page); |
Christian Borntraeger | ce7e9fa | 2007-11-20 11:13:36 +0100 | [diff] [blame] | 577 | if (mapping) { |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 578 | ret = page_mkclean_file(mapping, page); |
Christian Borntraeger | ce7e9fa | 2007-11-20 11:13:36 +0100 | [diff] [blame] | 579 | if (page_test_dirty(page)) { |
| 580 | page_clear_dirty(page); |
| 581 | ret = 1; |
| 582 | } |
Martin Schwidefsky | 6c21048 | 2007-04-27 16:01:57 +0200 | [diff] [blame] | 583 | } |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 584 | } |
| 585 | |
| 586 | return ret; |
| 587 | } |
Jaya Kumar | 60b59be | 2007-05-08 00:37:37 -0700 | [diff] [blame] | 588 | EXPORT_SYMBOL_GPL(page_mkclean); |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 589 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 590 | /** |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 591 | * __page_set_anon_rmap - setup new anonymous rmap |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 592 | * @page: the page to add the mapping to |
| 593 | * @vma: the vm area in which the mapping is added |
| 594 | * @address: the user virtual address mapped |
| 595 | */ |
| 596 | static void __page_set_anon_rmap(struct page *page, |
| 597 | struct vm_area_struct *vma, unsigned long address) |
| 598 | { |
| 599 | struct anon_vma *anon_vma = vma->anon_vma; |
| 600 | |
| 601 | BUG_ON(!anon_vma); |
| 602 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
| 603 | page->mapping = (struct address_space *) anon_vma; |
| 604 | |
| 605 | page->index = linear_page_index(vma, address); |
| 606 | |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 607 | /* |
| 608 | * nr_mapped state can be updated without turning off |
| 609 | * interrupts because it is not modified via interrupt. |
| 610 | */ |
Christoph Lameter | f3dbd34 | 2006-06-30 01:55:36 -0700 | [diff] [blame] | 611 | __inc_zone_page_state(page, NR_ANON_PAGES); |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 612 | } |
| 613 | |
| 614 | /** |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 615 | * __page_check_anon_rmap - sanity check anonymous rmap addition |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 616 | * @page: the page to add the mapping to |
| 617 | * @vma: the vm area in which the mapping is added |
| 618 | * @address: the user virtual address mapped |
| 619 | */ |
| 620 | static void __page_check_anon_rmap(struct page *page, |
| 621 | struct vm_area_struct *vma, unsigned long address) |
| 622 | { |
| 623 | #ifdef CONFIG_DEBUG_VM |
| 624 | /* |
| 625 | * The page's anon-rmap details (mapping and index) are guaranteed to |
| 626 | * be set up correctly at this point. |
| 627 | * |
| 628 | * We have exclusion against page_add_anon_rmap because the caller |
| 629 | * always holds the page locked, except if called from page_dup_rmap, |
| 630 | * in which case the page is already known to be setup. |
| 631 | * |
| 632 | * We have exclusion against page_add_new_anon_rmap because those pages |
| 633 | * are initially only visible via the pagetables, and the pte is locked |
| 634 | * over the call to page_add_new_anon_rmap. |
| 635 | */ |
| 636 | struct anon_vma *anon_vma = vma->anon_vma; |
| 637 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
| 638 | BUG_ON(page->mapping != (struct address_space *)anon_vma); |
| 639 | BUG_ON(page->index != linear_page_index(vma, address)); |
| 640 | #endif |
| 641 | } |
| 642 | |
| 643 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 644 | * page_add_anon_rmap - add pte mapping to an anonymous page |
| 645 | * @page: the page to add the mapping to |
| 646 | * @vma: the vm area in which the mapping is added |
| 647 | * @address: the user virtual address mapped |
| 648 | * |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 649 | * The caller needs to hold the pte lock and the page must be locked. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 650 | */ |
| 651 | void page_add_anon_rmap(struct page *page, |
| 652 | struct vm_area_struct *vma, unsigned long address) |
| 653 | { |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 654 | VM_BUG_ON(!PageLocked(page)); |
| 655 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 656 | if (atomic_inc_and_test(&page->_mapcount)) |
| 657 | __page_set_anon_rmap(page, vma, address); |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 658 | else |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 659 | __page_check_anon_rmap(page, vma, address); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 660 | } |
| 661 | |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 662 | /** |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 663 | * page_add_new_anon_rmap - add pte mapping to a new anonymous page |
| 664 | * @page: the page to add the mapping to |
| 665 | * @vma: the vm area in which the mapping is added |
| 666 | * @address: the user virtual address mapped |
| 667 | * |
| 668 | * Same as page_add_anon_rmap but must only be called on *new* pages. |
| 669 | * This means the inc-and-test can be bypassed. |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 670 | * Page does not have to be locked. |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 671 | */ |
| 672 | void page_add_new_anon_rmap(struct page *page, |
| 673 | struct vm_area_struct *vma, unsigned long address) |
| 674 | { |
Hugh Dickins | b5934c5 | 2009-01-06 14:39:25 -0800 | [diff] [blame^] | 675 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 676 | atomic_set(&page->_mapcount, 0); /* elevate count by 1 (starts at -1) */ |
| 677 | __page_set_anon_rmap(page, vma, address); |
Hugh Dickins | b5934c5 | 2009-01-06 14:39:25 -0800 | [diff] [blame^] | 678 | if (page_evictable(page, vma)) |
| 679 | lru_cache_add_lru(page, LRU_ACTIVE + page_is_file_cache(page)); |
| 680 | else |
| 681 | add_page_to_unevictable_list(page); |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 682 | } |
| 683 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 684 | /** |
| 685 | * page_add_file_rmap - add pte mapping to a file page |
| 686 | * @page: the page to add the mapping to |
| 687 | * |
Hugh Dickins | b8072f0 | 2005-10-29 18:16:41 -0700 | [diff] [blame] | 688 | * The caller needs to hold the pte lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 689 | */ |
| 690 | void page_add_file_rmap(struct page *page) |
| 691 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 692 | if (atomic_inc_and_test(&page->_mapcount)) |
Christoph Lameter | 65ba55f | 2006-06-30 01:55:34 -0700 | [diff] [blame] | 693 | __inc_zone_page_state(page, NR_FILE_MAPPED); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 694 | } |
| 695 | |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 696 | #ifdef CONFIG_DEBUG_VM |
| 697 | /** |
| 698 | * page_dup_rmap - duplicate pte mapping to a page |
| 699 | * @page: the page to add the mapping to |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 700 | * @vma: the vm area being duplicated |
| 701 | * @address: the user virtual address mapped |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 702 | * |
| 703 | * For copy_page_range only: minimal extract from page_add_file_rmap / |
| 704 | * page_add_anon_rmap, avoiding unnecessary tests (already checked) so it's |
| 705 | * quicker. |
| 706 | * |
| 707 | * The caller needs to hold the pte lock. |
| 708 | */ |
| 709 | void page_dup_rmap(struct page *page, struct vm_area_struct *vma, unsigned long address) |
| 710 | { |
| 711 | BUG_ON(page_mapcount(page) == 0); |
| 712 | if (PageAnon(page)) |
| 713 | __page_check_anon_rmap(page, vma, address); |
| 714 | atomic_inc(&page->_mapcount); |
| 715 | } |
| 716 | #endif |
| 717 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 718 | /** |
| 719 | * page_remove_rmap - take down pte mapping from a page |
| 720 | * @page: page to remove mapping from |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 721 | * @vma: the vm area in which the mapping is removed |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 722 | * |
Hugh Dickins | b8072f0 | 2005-10-29 18:16:41 -0700 | [diff] [blame] | 723 | * The caller needs to hold the pte lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 724 | */ |
Nick Piggin | 7de6b80 | 2006-12-22 01:09:33 -0800 | [diff] [blame] | 725 | void page_remove_rmap(struct page *page, struct vm_area_struct *vma) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 726 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 727 | if (atomic_add_negative(-1, &page->_mapcount)) { |
Nick Piggin | b7ab795 | 2006-03-22 00:08:42 -0800 | [diff] [blame] | 728 | if (unlikely(page_mapcount(page) < 0)) { |
Dave Jones | ef2bf0d | 2006-01-08 01:01:00 -0800 | [diff] [blame] | 729 | printk (KERN_EMERG "Eeek! page_mapcount(page) went negative! (%d)\n", page_mapcount(page)); |
Nick Piggin | 7de6b80 | 2006-12-22 01:09:33 -0800 | [diff] [blame] | 730 | printk (KERN_EMERG " page pfn = %lx\n", page_to_pfn(page)); |
Dave Jones | ef2bf0d | 2006-01-08 01:01:00 -0800 | [diff] [blame] | 731 | printk (KERN_EMERG " page->flags = %lx\n", page->flags); |
| 732 | printk (KERN_EMERG " page->count = %x\n", page_count(page)); |
| 733 | printk (KERN_EMERG " page->mapping = %p\n", page->mapping); |
Nick Piggin | 7de6b80 | 2006-12-22 01:09:33 -0800 | [diff] [blame] | 734 | print_symbol (KERN_EMERG " vma->vm_ops = %s\n", (unsigned long)vma->vm_ops); |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 735 | if (vma->vm_ops) { |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 736 | print_symbol (KERN_EMERG " vma->vm_ops->fault = %s\n", (unsigned long)vma->vm_ops->fault); |
| 737 | } |
Nick Piggin | 7de6b80 | 2006-12-22 01:09:33 -0800 | [diff] [blame] | 738 | if (vma->vm_file && vma->vm_file->f_op) |
| 739 | print_symbol (KERN_EMERG " vma->vm_file->f_op->mmap = %s\n", (unsigned long)vma->vm_file->f_op->mmap); |
Dave Jones | b16bc64 | 2006-10-11 01:21:27 -0700 | [diff] [blame] | 740 | BUG(); |
Dave Jones | ef2bf0d | 2006-01-08 01:01:00 -0800 | [diff] [blame] | 741 | } |
Dave Jones | b16bc64 | 2006-10-11 01:21:27 -0700 | [diff] [blame] | 742 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 743 | /* |
Hugh Dickins | 16f8c5b | 2008-08-20 14:09:04 -0700 | [diff] [blame] | 744 | * Now that the last pte has gone, s390 must transfer dirty |
| 745 | * flag from storage key to struct page. We can usually skip |
| 746 | * this if the page is anon, so about to be freed; but perhaps |
| 747 | * not if it's in swapcache - there might be another pte slot |
| 748 | * containing the swap entry, but page not yet written to swap. |
| 749 | */ |
| 750 | if ((!PageAnon(page) || PageSwapCache(page)) && |
| 751 | page_test_dirty(page)) { |
| 752 | page_clear_dirty(page); |
| 753 | set_page_dirty(page); |
| 754 | } |
KAMEZAWA Hiroyuki | 5b4e655 | 2008-10-18 20:28:10 -0700 | [diff] [blame] | 755 | if (PageAnon(page)) |
| 756 | mem_cgroup_uncharge_page(page); |
Hugh Dickins | 16f8c5b | 2008-08-20 14:09:04 -0700 | [diff] [blame] | 757 | __dec_zone_page_state(page, |
| 758 | PageAnon(page) ? NR_ANON_PAGES : NR_FILE_MAPPED); |
| 759 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 760 | * It would be tidy to reset the PageAnon mapping here, |
| 761 | * but that might overwrite a racing page_add_anon_rmap |
| 762 | * which increments mapcount after us but sets mapping |
| 763 | * before us: so leave the reset to free_hot_cold_page, |
| 764 | * and remember that it's only reliable while mapped. |
| 765 | * Leaving it set also helps swapoff to reinstate ptes |
| 766 | * faster for those pages still in swapcache. |
| 767 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 768 | } |
| 769 | } |
| 770 | |
| 771 | /* |
| 772 | * Subfunctions of try_to_unmap: try_to_unmap_one called |
| 773 | * repeatedly from either try_to_unmap_anon or try_to_unmap_file. |
| 774 | */ |
Christoph Lameter | a48d07a | 2006-02-01 03:05:38 -0800 | [diff] [blame] | 775 | static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
Christoph Lameter | 7352349 | 2006-06-23 02:03:27 -0700 | [diff] [blame] | 776 | int migration) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 777 | { |
| 778 | struct mm_struct *mm = vma->vm_mm; |
| 779 | unsigned long address; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 780 | pte_t *pte; |
| 781 | pte_t pteval; |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 782 | spinlock_t *ptl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 783 | int ret = SWAP_AGAIN; |
| 784 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 785 | address = vma_address(page, vma); |
| 786 | if (address == -EFAULT) |
| 787 | goto out; |
| 788 | |
Nick Piggin | 479db0b | 2008-08-20 14:09:18 -0700 | [diff] [blame] | 789 | pte = page_check_address(page, mm, address, &ptl, 0); |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 790 | if (!pte) |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 791 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 792 | |
| 793 | /* |
| 794 | * If the page is mlock()d, we cannot swap it out. |
| 795 | * If it's recently referenced (perhaps page_referenced |
| 796 | * skipped over this mm) then we should reactivate it. |
| 797 | */ |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 798 | if (!migration) { |
| 799 | if (vma->vm_flags & VM_LOCKED) { |
| 800 | ret = SWAP_MLOCK; |
| 801 | goto out_unmap; |
| 802 | } |
| 803 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
| 804 | ret = SWAP_FAIL; |
| 805 | goto out_unmap; |
| 806 | } |
| 807 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 808 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 809 | /* Nuke the page table entry. */ |
| 810 | flush_cache_page(vma, address, page_to_pfn(page)); |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 811 | pteval = ptep_clear_flush_notify(vma, address, pte); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 812 | |
| 813 | /* Move the dirty bit to the physical page now the pte is gone. */ |
| 814 | if (pte_dirty(pteval)) |
| 815 | set_page_dirty(page); |
| 816 | |
Hugh Dickins | 365e9c87 | 2005-10-29 18:16:18 -0700 | [diff] [blame] | 817 | /* Update high watermark before we lower rss */ |
| 818 | update_hiwater_rss(mm); |
| 819 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 820 | if (PageAnon(page)) { |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 821 | swp_entry_t entry = { .val = page_private(page) }; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 822 | |
| 823 | if (PageSwapCache(page)) { |
| 824 | /* |
| 825 | * Store the swap location in the pte. |
| 826 | * See handle_pte_fault() ... |
| 827 | */ |
| 828 | swap_duplicate(entry); |
| 829 | if (list_empty(&mm->mmlist)) { |
| 830 | spin_lock(&mmlist_lock); |
| 831 | if (list_empty(&mm->mmlist)) |
| 832 | list_add(&mm->mmlist, &init_mm.mmlist); |
| 833 | spin_unlock(&mmlist_lock); |
| 834 | } |
Christoph Lameter | 442c913 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 835 | dec_mm_counter(mm, anon_rss); |
KOSAKI Motohiro | 64cdd54 | 2009-01-06 14:39:16 -0800 | [diff] [blame] | 836 | } else if (PAGE_MIGRATION) { |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 837 | /* |
| 838 | * Store the pfn of the page in a special migration |
| 839 | * pte. do_swap_page() will wait until the migration |
| 840 | * pte is removed and then restart fault handling. |
| 841 | */ |
| 842 | BUG_ON(!migration); |
| 843 | entry = make_migration_entry(page, pte_write(pteval)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 844 | } |
| 845 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); |
| 846 | BUG_ON(pte_file(*pte)); |
KOSAKI Motohiro | 64cdd54 | 2009-01-06 14:39:16 -0800 | [diff] [blame] | 847 | } else if (PAGE_MIGRATION && migration) { |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 848 | /* Establish migration entry for a file page */ |
| 849 | swp_entry_t entry; |
| 850 | entry = make_migration_entry(page, pte_write(pteval)); |
| 851 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); |
| 852 | } else |
Hugh Dickins | 4294621 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 853 | dec_mm_counter(mm, file_rss); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 854 | |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 855 | |
Nick Piggin | 7de6b80 | 2006-12-22 01:09:33 -0800 | [diff] [blame] | 856 | page_remove_rmap(page, vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 857 | page_cache_release(page); |
| 858 | |
| 859 | out_unmap: |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 860 | pte_unmap_unlock(pte, ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 861 | out: |
| 862 | return ret; |
| 863 | } |
| 864 | |
| 865 | /* |
| 866 | * objrmap doesn't work for nonlinear VMAs because the assumption that |
| 867 | * offset-into-file correlates with offset-into-virtual-addresses does not hold. |
| 868 | * Consequently, given a particular page and its ->index, we cannot locate the |
| 869 | * ptes which are mapping that page without an exhaustive linear search. |
| 870 | * |
| 871 | * So what this code does is a mini "virtual scan" of each nonlinear VMA which |
| 872 | * maps the file to which the target page belongs. The ->vm_private_data field |
| 873 | * holds the current cursor into that scan. Successive searches will circulate |
| 874 | * around the vma's virtual address space. |
| 875 | * |
| 876 | * So as more replacement pressure is applied to the pages in a nonlinear VMA, |
| 877 | * more scanning pressure is placed against them as well. Eventually pages |
| 878 | * will become fully unmapped and are eligible for eviction. |
| 879 | * |
| 880 | * For very sparsely populated VMAs this is a little inefficient - chances are |
| 881 | * there there won't be many ptes located within the scan cluster. In this case |
| 882 | * 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] | 883 | * |
| 884 | * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can |
| 885 | * acquire it without blocking. If vma locked, mlock the pages in the cluster, |
| 886 | * rather than unmapping them. If we encounter the "check_page" that vmscan is |
| 887 | * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 888 | */ |
| 889 | #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE) |
| 890 | #define CLUSTER_MASK (~(CLUSTER_SIZE - 1)) |
| 891 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 892 | static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount, |
| 893 | struct vm_area_struct *vma, struct page *check_page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 894 | { |
| 895 | struct mm_struct *mm = vma->vm_mm; |
| 896 | pgd_t *pgd; |
| 897 | pud_t *pud; |
| 898 | pmd_t *pmd; |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 899 | pte_t *pte; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 900 | pte_t pteval; |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 901 | spinlock_t *ptl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 902 | struct page *page; |
| 903 | unsigned long address; |
| 904 | unsigned long end; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 905 | int ret = SWAP_AGAIN; |
| 906 | int locked_vma = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 907 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 908 | address = (vma->vm_start + cursor) & CLUSTER_MASK; |
| 909 | end = address + CLUSTER_SIZE; |
| 910 | if (address < vma->vm_start) |
| 911 | address = vma->vm_start; |
| 912 | if (end > vma->vm_end) |
| 913 | end = vma->vm_end; |
| 914 | |
| 915 | pgd = pgd_offset(mm, address); |
| 916 | if (!pgd_present(*pgd)) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 917 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 918 | |
| 919 | pud = pud_offset(pgd, address); |
| 920 | if (!pud_present(*pud)) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 921 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 922 | |
| 923 | pmd = pmd_offset(pud, address); |
| 924 | if (!pmd_present(*pmd)) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 925 | return ret; |
| 926 | |
| 927 | /* |
| 928 | * MLOCK_PAGES => feature is configured. |
| 929 | * if we can acquire the mmap_sem for read, and vma is VM_LOCKED, |
| 930 | * keep the sem while scanning the cluster for mlocking pages. |
| 931 | */ |
| 932 | if (MLOCK_PAGES && down_read_trylock(&vma->vm_mm->mmap_sem)) { |
| 933 | locked_vma = (vma->vm_flags & VM_LOCKED); |
| 934 | if (!locked_vma) |
| 935 | up_read(&vma->vm_mm->mmap_sem); /* don't need it */ |
| 936 | } |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 937 | |
| 938 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 939 | |
Hugh Dickins | 365e9c87 | 2005-10-29 18:16:18 -0700 | [diff] [blame] | 940 | /* Update high watermark before we lower rss */ |
| 941 | update_hiwater_rss(mm); |
| 942 | |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 943 | for (; address < end; pte++, address += PAGE_SIZE) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 944 | if (!pte_present(*pte)) |
| 945 | continue; |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 946 | page = vm_normal_page(vma, address, *pte); |
| 947 | BUG_ON(!page || PageAnon(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 948 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 949 | if (locked_vma) { |
| 950 | mlock_vma_page(page); /* no-op if already mlocked */ |
| 951 | if (page == check_page) |
| 952 | ret = SWAP_MLOCK; |
| 953 | continue; /* don't unmap */ |
| 954 | } |
| 955 | |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 956 | if (ptep_clear_flush_young_notify(vma, address, pte)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 957 | continue; |
| 958 | |
| 959 | /* Nuke the page table entry. */ |
Ben Collins | eca3513 | 2005-11-29 11:45:26 -0800 | [diff] [blame] | 960 | flush_cache_page(vma, address, pte_pfn(*pte)); |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 961 | pteval = ptep_clear_flush_notify(vma, address, pte); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 962 | |
| 963 | /* If nonlinear, store the file page offset in the pte. */ |
| 964 | if (page->index != linear_page_index(vma, address)) |
| 965 | set_pte_at(mm, address, pte, pgoff_to_pte(page->index)); |
| 966 | |
| 967 | /* Move the dirty bit to the physical page now the pte is gone. */ |
| 968 | if (pte_dirty(pteval)) |
| 969 | set_page_dirty(page); |
| 970 | |
Nick Piggin | 7de6b80 | 2006-12-22 01:09:33 -0800 | [diff] [blame] | 971 | page_remove_rmap(page, vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 972 | page_cache_release(page); |
Hugh Dickins | 4294621 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 973 | dec_mm_counter(mm, file_rss); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 974 | (*mapcount)--; |
| 975 | } |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 976 | pte_unmap_unlock(pte - 1, ptl); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 977 | if (locked_vma) |
| 978 | up_read(&vma->vm_mm->mmap_sem); |
| 979 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 980 | } |
| 981 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 982 | /* |
| 983 | * common handling for pages mapped in VM_LOCKED vmas |
| 984 | */ |
| 985 | static int try_to_mlock_page(struct page *page, struct vm_area_struct *vma) |
| 986 | { |
| 987 | int mlocked = 0; |
| 988 | |
| 989 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { |
| 990 | if (vma->vm_flags & VM_LOCKED) { |
| 991 | mlock_vma_page(page); |
| 992 | mlocked++; /* really mlocked the page */ |
| 993 | } |
| 994 | up_read(&vma->vm_mm->mmap_sem); |
| 995 | } |
| 996 | return mlocked; |
| 997 | } |
| 998 | |
| 999 | /** |
| 1000 | * try_to_unmap_anon - unmap or unlock anonymous page using the object-based |
| 1001 | * rmap method |
| 1002 | * @page: the page to unmap/unlock |
| 1003 | * @unlock: request for unlock rather than unmap [unlikely] |
| 1004 | * @migration: unmapping for migration - ignored if @unlock |
| 1005 | * |
| 1006 | * Find all the mappings of a page using the mapping pointer and the vma chains |
| 1007 | * contained in the anon_vma struct it points to. |
| 1008 | * |
| 1009 | * This function is only called from try_to_unmap/try_to_munlock for |
| 1010 | * anonymous pages. |
| 1011 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma |
| 1012 | * where the page was found will be held for write. So, we won't recheck |
| 1013 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be |
| 1014 | * 'LOCKED. |
| 1015 | */ |
| 1016 | static int try_to_unmap_anon(struct page *page, int unlock, int migration) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1017 | { |
| 1018 | struct anon_vma *anon_vma; |
| 1019 | struct vm_area_struct *vma; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1020 | unsigned int mlocked = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1021 | int ret = SWAP_AGAIN; |
| 1022 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1023 | if (MLOCK_PAGES && unlikely(unlock)) |
| 1024 | ret = SWAP_SUCCESS; /* default for try_to_munlock() */ |
| 1025 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1026 | anon_vma = page_lock_anon_vma(page); |
| 1027 | if (!anon_vma) |
| 1028 | return ret; |
| 1029 | |
| 1030 | list_for_each_entry(vma, &anon_vma->head, anon_vma_node) { |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1031 | if (MLOCK_PAGES && unlikely(unlock)) { |
| 1032 | if (!((vma->vm_flags & VM_LOCKED) && |
| 1033 | page_mapped_in_vma(page, vma))) |
| 1034 | continue; /* must visit all unlocked vmas */ |
| 1035 | ret = SWAP_MLOCK; /* saw at least one mlocked vma */ |
| 1036 | } else { |
| 1037 | ret = try_to_unmap_one(page, vma, migration); |
| 1038 | if (ret == SWAP_FAIL || !page_mapped(page)) |
| 1039 | break; |
| 1040 | } |
| 1041 | if (ret == SWAP_MLOCK) { |
| 1042 | mlocked = try_to_mlock_page(page, vma); |
| 1043 | if (mlocked) |
| 1044 | break; /* stop if actually mlocked page */ |
| 1045 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1046 | } |
Oleg Nesterov | 34bbd70 | 2007-02-28 20:13:49 -0800 | [diff] [blame] | 1047 | |
| 1048 | page_unlock_anon_vma(anon_vma); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1049 | |
| 1050 | if (mlocked) |
| 1051 | ret = SWAP_MLOCK; /* actually mlocked the page */ |
| 1052 | else if (ret == SWAP_MLOCK) |
| 1053 | ret = SWAP_AGAIN; /* saw VM_LOCKED vma */ |
| 1054 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1055 | return ret; |
| 1056 | } |
| 1057 | |
| 1058 | /** |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1059 | * try_to_unmap_file - unmap/unlock file page using the object-based rmap method |
| 1060 | * @page: the page to unmap/unlock |
| 1061 | * @unlock: request for unlock rather than unmap [unlikely] |
| 1062 | * @migration: unmapping for migration - ignored if @unlock |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1063 | * |
| 1064 | * Find all the mappings of a page using the mapping pointer and the vma chains |
| 1065 | * contained in the address_space struct it points to. |
| 1066 | * |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1067 | * This function is only called from try_to_unmap/try_to_munlock for |
| 1068 | * object-based pages. |
| 1069 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma |
| 1070 | * where the page was found will be held for write. So, we won't recheck |
| 1071 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be |
| 1072 | * 'LOCKED. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1073 | */ |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1074 | static int try_to_unmap_file(struct page *page, int unlock, int migration) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1075 | { |
| 1076 | struct address_space *mapping = page->mapping; |
| 1077 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
| 1078 | struct vm_area_struct *vma; |
| 1079 | struct prio_tree_iter iter; |
| 1080 | int ret = SWAP_AGAIN; |
| 1081 | unsigned long cursor; |
| 1082 | unsigned long max_nl_cursor = 0; |
| 1083 | unsigned long max_nl_size = 0; |
| 1084 | unsigned int mapcount; |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1085 | unsigned int mlocked = 0; |
| 1086 | |
| 1087 | if (MLOCK_PAGES && unlikely(unlock)) |
| 1088 | ret = SWAP_SUCCESS; /* default for try_to_munlock() */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1089 | |
| 1090 | spin_lock(&mapping->i_mmap_lock); |
| 1091 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1092 | if (MLOCK_PAGES && unlikely(unlock)) { |
| 1093 | if (!(vma->vm_flags & VM_LOCKED)) |
| 1094 | continue; /* must visit all vmas */ |
| 1095 | ret = SWAP_MLOCK; |
| 1096 | } else { |
| 1097 | ret = try_to_unmap_one(page, vma, migration); |
| 1098 | if (ret == SWAP_FAIL || !page_mapped(page)) |
| 1099 | goto out; |
| 1100 | } |
| 1101 | if (ret == SWAP_MLOCK) { |
| 1102 | mlocked = try_to_mlock_page(page, vma); |
| 1103 | if (mlocked) |
| 1104 | break; /* stop if actually mlocked page */ |
| 1105 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1106 | } |
| 1107 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1108 | if (mlocked) |
| 1109 | goto out; |
| 1110 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1111 | if (list_empty(&mapping->i_mmap_nonlinear)) |
| 1112 | goto out; |
| 1113 | |
| 1114 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, |
| 1115 | shared.vm_set.list) { |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1116 | if (MLOCK_PAGES && unlikely(unlock)) { |
| 1117 | if (!(vma->vm_flags & VM_LOCKED)) |
| 1118 | continue; /* must visit all vmas */ |
| 1119 | ret = SWAP_MLOCK; /* leave mlocked == 0 */ |
| 1120 | goto out; /* no need to look further */ |
| 1121 | } |
| 1122 | if (!MLOCK_PAGES && !migration && (vma->vm_flags & VM_LOCKED)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1123 | continue; |
| 1124 | cursor = (unsigned long) vma->vm_private_data; |
| 1125 | if (cursor > max_nl_cursor) |
| 1126 | max_nl_cursor = cursor; |
| 1127 | cursor = vma->vm_end - vma->vm_start; |
| 1128 | if (cursor > max_nl_size) |
| 1129 | max_nl_size = cursor; |
| 1130 | } |
| 1131 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1132 | if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1133 | ret = SWAP_FAIL; |
| 1134 | goto out; |
| 1135 | } |
| 1136 | |
| 1137 | /* |
| 1138 | * We don't try to search for this page in the nonlinear vmas, |
| 1139 | * and page_referenced wouldn't have found it anyway. Instead |
| 1140 | * just walk the nonlinear vmas trying to age and unmap some. |
| 1141 | * The mapcount of the page we came in with is irrelevant, |
| 1142 | * but even so use it as a guide to how hard we should try? |
| 1143 | */ |
| 1144 | mapcount = page_mapcount(page); |
| 1145 | if (!mapcount) |
| 1146 | goto out; |
| 1147 | cond_resched_lock(&mapping->i_mmap_lock); |
| 1148 | |
| 1149 | max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK; |
| 1150 | if (max_nl_cursor == 0) |
| 1151 | max_nl_cursor = CLUSTER_SIZE; |
| 1152 | |
| 1153 | do { |
| 1154 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, |
| 1155 | shared.vm_set.list) { |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1156 | if (!MLOCK_PAGES && !migration && |
| 1157 | (vma->vm_flags & VM_LOCKED)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1158 | continue; |
| 1159 | cursor = (unsigned long) vma->vm_private_data; |
Hugh Dickins | 839b968 | 2005-09-03 15:54:43 -0700 | [diff] [blame] | 1160 | while ( cursor < max_nl_cursor && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1161 | cursor < vma->vm_end - vma->vm_start) { |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1162 | ret = try_to_unmap_cluster(cursor, &mapcount, |
| 1163 | vma, page); |
| 1164 | if (ret == SWAP_MLOCK) |
| 1165 | mlocked = 2; /* to return below */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1166 | cursor += CLUSTER_SIZE; |
| 1167 | vma->vm_private_data = (void *) cursor; |
| 1168 | if ((int)mapcount <= 0) |
| 1169 | goto out; |
| 1170 | } |
| 1171 | vma->vm_private_data = (void *) max_nl_cursor; |
| 1172 | } |
| 1173 | cond_resched_lock(&mapping->i_mmap_lock); |
| 1174 | max_nl_cursor += CLUSTER_SIZE; |
| 1175 | } while (max_nl_cursor <= max_nl_size); |
| 1176 | |
| 1177 | /* |
| 1178 | * Don't loop forever (perhaps all the remaining pages are |
| 1179 | * in locked vmas). Reset cursor on all unreserved nonlinear |
| 1180 | * vmas, now forgetting on which ones it had fallen behind. |
| 1181 | */ |
Hugh Dickins | 101d2be | 2005-11-21 21:32:16 -0800 | [diff] [blame] | 1182 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) |
| 1183 | vma->vm_private_data = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1184 | out: |
| 1185 | spin_unlock(&mapping->i_mmap_lock); |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1186 | if (mlocked) |
| 1187 | ret = SWAP_MLOCK; /* actually mlocked the page */ |
| 1188 | else if (ret == SWAP_MLOCK) |
| 1189 | ret = SWAP_AGAIN; /* saw VM_LOCKED vma */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1190 | return ret; |
| 1191 | } |
| 1192 | |
| 1193 | /** |
| 1194 | * try_to_unmap - try to remove all page table mappings to a page |
| 1195 | * @page: the page to get unmapped |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 1196 | * @migration: migration flag |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1197 | * |
| 1198 | * Tries to remove all the page table entries which are mapping this |
| 1199 | * page, used in the pageout path. Caller must hold the page lock. |
| 1200 | * Return values are: |
| 1201 | * |
| 1202 | * SWAP_SUCCESS - we succeeded in removing all mappings |
| 1203 | * SWAP_AGAIN - we missed a mapping, try again later |
| 1204 | * SWAP_FAIL - the page is unswappable |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1205 | * SWAP_MLOCK - page is mlocked. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1206 | */ |
Christoph Lameter | 7352349 | 2006-06-23 02:03:27 -0700 | [diff] [blame] | 1207 | int try_to_unmap(struct page *page, int migration) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1208 | { |
| 1209 | int ret; |
| 1210 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1211 | BUG_ON(!PageLocked(page)); |
| 1212 | |
| 1213 | if (PageAnon(page)) |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1214 | ret = try_to_unmap_anon(page, 0, migration); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1215 | else |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1216 | ret = try_to_unmap_file(page, 0, migration); |
| 1217 | if (ret != SWAP_MLOCK && !page_mapped(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1218 | ret = SWAP_SUCCESS; |
| 1219 | return ret; |
| 1220 | } |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 1221 | |
Nick Piggin | b291f00 | 2008-10-18 20:26:44 -0700 | [diff] [blame] | 1222 | #ifdef CONFIG_UNEVICTABLE_LRU |
| 1223 | /** |
| 1224 | * try_to_munlock - try to munlock a page |
| 1225 | * @page: the page to be munlocked |
| 1226 | * |
| 1227 | * Called from munlock code. Checks all of the VMAs mapping the page |
| 1228 | * to make sure nobody else has this page mlocked. The page will be |
| 1229 | * returned with PG_mlocked cleared if no other vmas have it mlocked. |
| 1230 | * |
| 1231 | * Return values are: |
| 1232 | * |
| 1233 | * SWAP_SUCCESS - no vma's holding page mlocked. |
| 1234 | * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem |
| 1235 | * SWAP_MLOCK - page is now mlocked. |
| 1236 | */ |
| 1237 | int try_to_munlock(struct page *page) |
| 1238 | { |
| 1239 | VM_BUG_ON(!PageLocked(page) || PageLRU(page)); |
| 1240 | |
| 1241 | if (PageAnon(page)) |
| 1242 | return try_to_unmap_anon(page, 1, 0); |
| 1243 | else |
| 1244 | return try_to_unmap_file(page, 1, 0); |
| 1245 | } |
| 1246 | #endif |