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> |
Nick Piggin | 7de6b80 | 2006-12-22 01:09:33 -0800 | [diff] [blame] | 50 | #include <linux/kallsyms.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 51 | #include <linux/memcontrol.h> |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 52 | #include <linux/mmu_notifier.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 53 | |
| 54 | #include <asm/tlbflush.h> |
| 55 | |
Pekka Enberg | fcc234f | 2006-03-22 00:08:13 -0800 | [diff] [blame] | 56 | struct kmem_cache *anon_vma_cachep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 57 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 58 | /* This must be called under the mmap_sem. */ |
| 59 | int anon_vma_prepare(struct vm_area_struct *vma) |
| 60 | { |
| 61 | struct anon_vma *anon_vma = vma->anon_vma; |
| 62 | |
| 63 | might_sleep(); |
| 64 | if (unlikely(!anon_vma)) { |
| 65 | struct mm_struct *mm = vma->vm_mm; |
| 66 | struct anon_vma *allocated, *locked; |
| 67 | |
| 68 | anon_vma = find_mergeable_anon_vma(vma); |
| 69 | if (anon_vma) { |
| 70 | allocated = NULL; |
| 71 | locked = anon_vma; |
| 72 | spin_lock(&locked->lock); |
| 73 | } else { |
| 74 | anon_vma = anon_vma_alloc(); |
| 75 | if (unlikely(!anon_vma)) |
| 76 | return -ENOMEM; |
| 77 | allocated = anon_vma; |
| 78 | locked = NULL; |
| 79 | } |
| 80 | |
| 81 | /* page_table_lock to protect against threads */ |
| 82 | spin_lock(&mm->page_table_lock); |
| 83 | if (likely(!vma->anon_vma)) { |
| 84 | vma->anon_vma = anon_vma; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 85 | list_add_tail(&vma->anon_vma_node, &anon_vma->head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 86 | allocated = NULL; |
| 87 | } |
| 88 | spin_unlock(&mm->page_table_lock); |
| 89 | |
| 90 | if (locked) |
| 91 | spin_unlock(&locked->lock); |
| 92 | if (unlikely(allocated)) |
| 93 | anon_vma_free(allocated); |
| 94 | } |
| 95 | return 0; |
| 96 | } |
| 97 | |
| 98 | void __anon_vma_merge(struct vm_area_struct *vma, struct vm_area_struct *next) |
| 99 | { |
| 100 | BUG_ON(vma->anon_vma != next->anon_vma); |
| 101 | list_del(&next->anon_vma_node); |
| 102 | } |
| 103 | |
| 104 | void __anon_vma_link(struct vm_area_struct *vma) |
| 105 | { |
| 106 | struct anon_vma *anon_vma = vma->anon_vma; |
| 107 | |
Hugh Dickins | 30acbab | 2007-06-27 14:09:53 -0700 | [diff] [blame] | 108 | if (anon_vma) |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 109 | list_add_tail(&vma->anon_vma_node, &anon_vma->head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 110 | } |
| 111 | |
| 112 | void anon_vma_link(struct vm_area_struct *vma) |
| 113 | { |
| 114 | struct anon_vma *anon_vma = vma->anon_vma; |
| 115 | |
| 116 | if (anon_vma) { |
| 117 | spin_lock(&anon_vma->lock); |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 118 | list_add_tail(&vma->anon_vma_node, &anon_vma->head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 119 | spin_unlock(&anon_vma->lock); |
| 120 | } |
| 121 | } |
| 122 | |
| 123 | void anon_vma_unlink(struct vm_area_struct *vma) |
| 124 | { |
| 125 | struct anon_vma *anon_vma = vma->anon_vma; |
| 126 | int empty; |
| 127 | |
| 128 | if (!anon_vma) |
| 129 | return; |
| 130 | |
| 131 | spin_lock(&anon_vma->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 132 | list_del(&vma->anon_vma_node); |
| 133 | |
| 134 | /* We must garbage collect the anon_vma if it's empty */ |
| 135 | empty = list_empty(&anon_vma->head); |
| 136 | spin_unlock(&anon_vma->lock); |
| 137 | |
| 138 | if (empty) |
| 139 | anon_vma_free(anon_vma); |
| 140 | } |
| 141 | |
Alexey Dobriyan | 51cc506 | 2008-07-25 19:45:34 -0700 | [diff] [blame] | 142 | static void anon_vma_ctor(void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 143 | { |
Christoph Lameter | a35afb8 | 2007-05-16 22:10:57 -0700 | [diff] [blame] | 144 | struct anon_vma *anon_vma = data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 145 | |
Christoph Lameter | a35afb8 | 2007-05-16 22:10:57 -0700 | [diff] [blame] | 146 | spin_lock_init(&anon_vma->lock); |
| 147 | INIT_LIST_HEAD(&anon_vma->head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 148 | } |
| 149 | |
| 150 | void __init anon_vma_init(void) |
| 151 | { |
| 152 | anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), |
Paul Mundt | 20c2df8 | 2007-07-20 10:11:58 +0900 | [diff] [blame] | 153 | 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 154 | } |
| 155 | |
| 156 | /* |
| 157 | * Getting a lock on a stable anon_vma from a page off the LRU is |
| 158 | * tricky: page_lock_anon_vma rely on RCU to guard against the races. |
| 159 | */ |
| 160 | static struct anon_vma *page_lock_anon_vma(struct page *page) |
| 161 | { |
Oleg Nesterov | 34bbd70 | 2007-02-28 20:13:49 -0800 | [diff] [blame] | 162 | struct anon_vma *anon_vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 163 | unsigned long anon_mapping; |
| 164 | |
| 165 | rcu_read_lock(); |
| 166 | anon_mapping = (unsigned long) page->mapping; |
| 167 | if (!(anon_mapping & PAGE_MAPPING_ANON)) |
| 168 | goto out; |
| 169 | if (!page_mapped(page)) |
| 170 | goto out; |
| 171 | |
| 172 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); |
| 173 | spin_lock(&anon_vma->lock); |
Oleg Nesterov | 34bbd70 | 2007-02-28 20:13:49 -0800 | [diff] [blame] | 174 | return anon_vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 175 | out: |
| 176 | rcu_read_unlock(); |
Oleg Nesterov | 34bbd70 | 2007-02-28 20:13:49 -0800 | [diff] [blame] | 177 | return NULL; |
| 178 | } |
| 179 | |
| 180 | static void page_unlock_anon_vma(struct anon_vma *anon_vma) |
| 181 | { |
| 182 | spin_unlock(&anon_vma->lock); |
| 183 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 184 | } |
| 185 | |
| 186 | /* |
Lee Schermerhorn | 3ad33b2 | 2007-11-14 16:59:10 -0800 | [diff] [blame] | 187 | * At what user virtual address is page expected in @vma? |
| 188 | * Returns virtual address or -EFAULT if page's index/offset is not |
| 189 | * within the range mapped the @vma. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 190 | */ |
| 191 | static inline unsigned long |
| 192 | vma_address(struct page *page, struct vm_area_struct *vma) |
| 193 | { |
| 194 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
| 195 | unsigned long address; |
| 196 | |
| 197 | address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
| 198 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) { |
Lee Schermerhorn | 3ad33b2 | 2007-11-14 16:59:10 -0800 | [diff] [blame] | 199 | /* page should be within @vma mapping range */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 200 | return -EFAULT; |
| 201 | } |
| 202 | return address; |
| 203 | } |
| 204 | |
| 205 | /* |
| 206 | * 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] | 207 | * 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] | 208 | */ |
| 209 | unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) |
| 210 | { |
| 211 | if (PageAnon(page)) { |
| 212 | if ((void *)vma->anon_vma != |
| 213 | (void *)page->mapping - PAGE_MAPPING_ANON) |
| 214 | return -EFAULT; |
| 215 | } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) { |
Hugh Dickins | ee498ed | 2005-11-21 21:32:18 -0800 | [diff] [blame] | 216 | if (!vma->vm_file || |
| 217 | vma->vm_file->f_mapping != page->mapping) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 218 | return -EFAULT; |
| 219 | } else |
| 220 | return -EFAULT; |
| 221 | return vma_address(page, vma); |
| 222 | } |
| 223 | |
| 224 | /* |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 225 | * Check that @page is mapped at @address into @mm. |
| 226 | * |
Hugh Dickins | b8072f0 | 2005-10-29 18:16:41 -0700 | [diff] [blame] | 227 | * On success returns with pte mapped and locked. |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 228 | */ |
Carsten Otte | ceffc07 | 2005-06-23 22:05:25 -0700 | [diff] [blame] | 229 | pte_t *page_check_address(struct page *page, struct mm_struct *mm, |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 230 | unsigned long address, spinlock_t **ptlp) |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 231 | { |
| 232 | pgd_t *pgd; |
| 233 | pud_t *pud; |
| 234 | pmd_t *pmd; |
| 235 | pte_t *pte; |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 236 | spinlock_t *ptl; |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 237 | |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 238 | pgd = pgd_offset(mm, address); |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 239 | if (!pgd_present(*pgd)) |
| 240 | return NULL; |
| 241 | |
| 242 | pud = pud_offset(pgd, address); |
| 243 | if (!pud_present(*pud)) |
| 244 | return NULL; |
| 245 | |
| 246 | pmd = pmd_offset(pud, address); |
| 247 | if (!pmd_present(*pmd)) |
| 248 | return NULL; |
| 249 | |
| 250 | pte = pte_offset_map(pmd, address); |
| 251 | /* Make a quick check before getting the lock */ |
| 252 | if (!pte_present(*pte)) { |
| 253 | pte_unmap(pte); |
| 254 | return NULL; |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 255 | } |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 256 | |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 257 | ptl = pte_lockptr(mm, pmd); |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 258 | spin_lock(ptl); |
| 259 | if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) { |
| 260 | *ptlp = ptl; |
| 261 | return pte; |
| 262 | } |
| 263 | pte_unmap_unlock(pte, ptl); |
| 264 | return NULL; |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 265 | } |
| 266 | |
| 267 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 268 | * Subfunctions of page_referenced: page_referenced_one called |
| 269 | * repeatedly from either page_referenced_anon or page_referenced_file. |
| 270 | */ |
| 271 | static int page_referenced_one(struct page *page, |
Rik van Riel | f7b7fd8 | 2005-11-28 13:44:07 -0800 | [diff] [blame] | 272 | struct vm_area_struct *vma, unsigned int *mapcount) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 273 | { |
| 274 | struct mm_struct *mm = vma->vm_mm; |
| 275 | unsigned long address; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 276 | pte_t *pte; |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 277 | spinlock_t *ptl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 278 | int referenced = 0; |
| 279 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 280 | address = vma_address(page, vma); |
| 281 | if (address == -EFAULT) |
| 282 | goto out; |
| 283 | |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 284 | pte = page_check_address(page, mm, address, &ptl); |
| 285 | if (!pte) |
| 286 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 287 | |
Hugh Dickins | 5a9bbdc | 2008-02-04 22:29:23 -0800 | [diff] [blame] | 288 | if (vma->vm_flags & VM_LOCKED) { |
| 289 | referenced++; |
| 290 | *mapcount = 1; /* break early from loop */ |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 291 | } else if (ptep_clear_flush_young_notify(vma, address, pte)) |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 292 | referenced++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 293 | |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 294 | /* Pretend the page is referenced if the task has the |
| 295 | swap token and is in the middle of a page fault. */ |
Rik van Riel | f7b7fd8 | 2005-11-28 13:44:07 -0800 | [diff] [blame] | 296 | if (mm != current->mm && has_swap_token(mm) && |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 297 | rwsem_is_locked(&mm->mmap_sem)) |
| 298 | referenced++; |
| 299 | |
| 300 | (*mapcount)--; |
| 301 | pte_unmap_unlock(pte, ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 302 | out: |
| 303 | return referenced; |
| 304 | } |
| 305 | |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 306 | static int page_referenced_anon(struct page *page, |
| 307 | struct mem_cgroup *mem_cont) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 308 | { |
| 309 | unsigned int mapcount; |
| 310 | struct anon_vma *anon_vma; |
| 311 | struct vm_area_struct *vma; |
| 312 | int referenced = 0; |
| 313 | |
| 314 | anon_vma = page_lock_anon_vma(page); |
| 315 | if (!anon_vma) |
| 316 | return referenced; |
| 317 | |
| 318 | mapcount = page_mapcount(page); |
| 319 | list_for_each_entry(vma, &anon_vma->head, anon_vma_node) { |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 320 | /* |
| 321 | * If we are reclaiming on behalf of a cgroup, skip |
| 322 | * counting on behalf of references from different |
| 323 | * cgroups |
| 324 | */ |
Hugh Dickins | bd845e3 | 2008-03-04 14:29:01 -0800 | [diff] [blame] | 325 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 326 | continue; |
Rik van Riel | f7b7fd8 | 2005-11-28 13:44:07 -0800 | [diff] [blame] | 327 | referenced += page_referenced_one(page, vma, &mapcount); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 328 | if (!mapcount) |
| 329 | break; |
| 330 | } |
Oleg Nesterov | 34bbd70 | 2007-02-28 20:13:49 -0800 | [diff] [blame] | 331 | |
| 332 | page_unlock_anon_vma(anon_vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 333 | return referenced; |
| 334 | } |
| 335 | |
| 336 | /** |
| 337 | * page_referenced_file - referenced check for object-based rmap |
| 338 | * @page: the page we're checking references on. |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 339 | * @mem_cont: target memory controller |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 340 | * |
| 341 | * For an object-based mapped page, find all the places it is mapped and |
| 342 | * check/clear the referenced flag. This is done by following the page->mapping |
| 343 | * pointer, then walking the chain of vmas it holds. It returns the number |
| 344 | * of references it found. |
| 345 | * |
| 346 | * This function is only called from page_referenced for object-based pages. |
| 347 | */ |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 348 | static int page_referenced_file(struct page *page, |
| 349 | struct mem_cgroup *mem_cont) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 350 | { |
| 351 | unsigned int mapcount; |
| 352 | struct address_space *mapping = page->mapping; |
| 353 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
| 354 | struct vm_area_struct *vma; |
| 355 | struct prio_tree_iter iter; |
| 356 | int referenced = 0; |
| 357 | |
| 358 | /* |
| 359 | * The caller's checks on page->mapping and !PageAnon have made |
| 360 | * sure that this is a file page: the check for page->mapping |
| 361 | * excludes the case just before it gets set on an anon page. |
| 362 | */ |
| 363 | BUG_ON(PageAnon(page)); |
| 364 | |
| 365 | /* |
| 366 | * The page lock not only makes sure that page->mapping cannot |
| 367 | * suddenly be NULLified by truncation, it makes sure that the |
| 368 | * structure at mapping cannot be freed and reused yet, |
| 369 | * so we can safely take mapping->i_mmap_lock. |
| 370 | */ |
| 371 | BUG_ON(!PageLocked(page)); |
| 372 | |
| 373 | spin_lock(&mapping->i_mmap_lock); |
| 374 | |
| 375 | /* |
| 376 | * i_mmap_lock does not stabilize mapcount at all, but mapcount |
| 377 | * is more likely to be accurate if we note it after spinning. |
| 378 | */ |
| 379 | mapcount = page_mapcount(page); |
| 380 | |
| 381 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 382 | /* |
| 383 | * If we are reclaiming on behalf of a cgroup, skip |
| 384 | * counting on behalf of references from different |
| 385 | * cgroups |
| 386 | */ |
Hugh Dickins | bd845e3 | 2008-03-04 14:29:01 -0800 | [diff] [blame] | 387 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 388 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 389 | if ((vma->vm_flags & (VM_LOCKED|VM_MAYSHARE)) |
| 390 | == (VM_LOCKED|VM_MAYSHARE)) { |
| 391 | referenced++; |
| 392 | break; |
| 393 | } |
Rik van Riel | f7b7fd8 | 2005-11-28 13:44:07 -0800 | [diff] [blame] | 394 | referenced += page_referenced_one(page, vma, &mapcount); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 395 | if (!mapcount) |
| 396 | break; |
| 397 | } |
| 398 | |
| 399 | spin_unlock(&mapping->i_mmap_lock); |
| 400 | return referenced; |
| 401 | } |
| 402 | |
| 403 | /** |
| 404 | * page_referenced - test if the page was referenced |
| 405 | * @page: the page to test |
| 406 | * @is_locked: caller holds lock on the page |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 407 | * @mem_cont: target memory controller |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 408 | * |
| 409 | * Quick test_and_clear_referenced for all mappings to a page, |
| 410 | * returns the number of ptes which referenced the page. |
| 411 | */ |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 412 | int page_referenced(struct page *page, int is_locked, |
| 413 | struct mem_cgroup *mem_cont) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 414 | { |
| 415 | int referenced = 0; |
| 416 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 417 | if (TestClearPageReferenced(page)) |
| 418 | referenced++; |
| 419 | |
| 420 | if (page_mapped(page) && page->mapping) { |
| 421 | if (PageAnon(page)) |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 422 | referenced += page_referenced_anon(page, mem_cont); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 423 | else if (is_locked) |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 424 | referenced += page_referenced_file(page, mem_cont); |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 425 | else if (!trylock_page(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 426 | referenced++; |
| 427 | else { |
| 428 | if (page->mapping) |
Balbir Singh | bed7161 | 2008-02-07 00:14:01 -0800 | [diff] [blame] | 429 | referenced += |
| 430 | page_referenced_file(page, mem_cont); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 431 | unlock_page(page); |
| 432 | } |
| 433 | } |
Christian Borntraeger | 5b7baf0 | 2008-03-25 18:47:12 +0100 | [diff] [blame] | 434 | |
| 435 | if (page_test_and_clear_young(page)) |
| 436 | referenced++; |
| 437 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 438 | return referenced; |
| 439 | } |
| 440 | |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 441 | static int page_mkclean_one(struct page *page, struct vm_area_struct *vma) |
| 442 | { |
| 443 | struct mm_struct *mm = vma->vm_mm; |
| 444 | unsigned long address; |
Peter Zijlstra | c2fda5f | 2006-12-22 14:25:52 +0100 | [diff] [blame] | 445 | pte_t *pte; |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 446 | spinlock_t *ptl; |
| 447 | int ret = 0; |
| 448 | |
| 449 | address = vma_address(page, vma); |
| 450 | if (address == -EFAULT) |
| 451 | goto out; |
| 452 | |
| 453 | pte = page_check_address(page, mm, address, &ptl); |
| 454 | if (!pte) |
| 455 | goto out; |
| 456 | |
Peter Zijlstra | c2fda5f | 2006-12-22 14:25:52 +0100 | [diff] [blame] | 457 | if (pte_dirty(*pte) || pte_write(*pte)) { |
| 458 | pte_t entry; |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 459 | |
Peter Zijlstra | c2fda5f | 2006-12-22 14:25:52 +0100 | [diff] [blame] | 460 | flush_cache_page(vma, address, pte_pfn(*pte)); |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 461 | entry = ptep_clear_flush_notify(vma, address, pte); |
Peter Zijlstra | c2fda5f | 2006-12-22 14:25:52 +0100 | [diff] [blame] | 462 | entry = pte_wrprotect(entry); |
| 463 | entry = pte_mkclean(entry); |
Al Viro | d6e88e6 | 2006-12-29 16:48:35 -0800 | [diff] [blame] | 464 | set_pte_at(mm, address, pte, entry); |
Peter Zijlstra | c2fda5f | 2006-12-22 14:25:52 +0100 | [diff] [blame] | 465 | ret = 1; |
| 466 | } |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 467 | |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 468 | pte_unmap_unlock(pte, ptl); |
| 469 | out: |
| 470 | return ret; |
| 471 | } |
| 472 | |
| 473 | static int page_mkclean_file(struct address_space *mapping, struct page *page) |
| 474 | { |
| 475 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
| 476 | struct vm_area_struct *vma; |
| 477 | struct prio_tree_iter iter; |
| 478 | int ret = 0; |
| 479 | |
| 480 | BUG_ON(PageAnon(page)); |
| 481 | |
| 482 | spin_lock(&mapping->i_mmap_lock); |
| 483 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
| 484 | if (vma->vm_flags & VM_SHARED) |
| 485 | ret += page_mkclean_one(page, vma); |
| 486 | } |
| 487 | spin_unlock(&mapping->i_mmap_lock); |
| 488 | return ret; |
| 489 | } |
| 490 | |
| 491 | int page_mkclean(struct page *page) |
| 492 | { |
| 493 | int ret = 0; |
| 494 | |
| 495 | BUG_ON(!PageLocked(page)); |
| 496 | |
| 497 | if (page_mapped(page)) { |
| 498 | struct address_space *mapping = page_mapping(page); |
Christian Borntraeger | ce7e9fa | 2007-11-20 11:13:36 +0100 | [diff] [blame] | 499 | if (mapping) { |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 500 | ret = page_mkclean_file(mapping, page); |
Christian Borntraeger | ce7e9fa | 2007-11-20 11:13:36 +0100 | [diff] [blame] | 501 | if (page_test_dirty(page)) { |
| 502 | page_clear_dirty(page); |
| 503 | ret = 1; |
| 504 | } |
Martin Schwidefsky | 6c21048 | 2007-04-27 16:01:57 +0200 | [diff] [blame] | 505 | } |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 506 | } |
| 507 | |
| 508 | return ret; |
| 509 | } |
Jaya Kumar | 60b59be | 2007-05-08 00:37:37 -0700 | [diff] [blame] | 510 | EXPORT_SYMBOL_GPL(page_mkclean); |
Peter Zijlstra | d08b385 | 2006-09-25 23:30:57 -0700 | [diff] [blame] | 511 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 512 | /** |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 513 | * __page_set_anon_rmap - setup new anonymous rmap |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 514 | * @page: the page to add the mapping to |
| 515 | * @vma: the vm area in which the mapping is added |
| 516 | * @address: the user virtual address mapped |
| 517 | */ |
| 518 | static void __page_set_anon_rmap(struct page *page, |
| 519 | struct vm_area_struct *vma, unsigned long address) |
| 520 | { |
| 521 | struct anon_vma *anon_vma = vma->anon_vma; |
| 522 | |
| 523 | BUG_ON(!anon_vma); |
| 524 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
| 525 | page->mapping = (struct address_space *) anon_vma; |
| 526 | |
| 527 | page->index = linear_page_index(vma, address); |
| 528 | |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 529 | /* |
| 530 | * nr_mapped state can be updated without turning off |
| 531 | * interrupts because it is not modified via interrupt. |
| 532 | */ |
Christoph Lameter | f3dbd34 | 2006-06-30 01:55:36 -0700 | [diff] [blame] | 533 | __inc_zone_page_state(page, NR_ANON_PAGES); |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 534 | } |
| 535 | |
| 536 | /** |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 537 | * __page_check_anon_rmap - sanity check anonymous rmap addition |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 538 | * @page: the page to add the mapping to |
| 539 | * @vma: the vm area in which the mapping is added |
| 540 | * @address: the user virtual address mapped |
| 541 | */ |
| 542 | static void __page_check_anon_rmap(struct page *page, |
| 543 | struct vm_area_struct *vma, unsigned long address) |
| 544 | { |
| 545 | #ifdef CONFIG_DEBUG_VM |
| 546 | /* |
| 547 | * The page's anon-rmap details (mapping and index) are guaranteed to |
| 548 | * be set up correctly at this point. |
| 549 | * |
| 550 | * We have exclusion against page_add_anon_rmap because the caller |
| 551 | * always holds the page locked, except if called from page_dup_rmap, |
| 552 | * in which case the page is already known to be setup. |
| 553 | * |
| 554 | * We have exclusion against page_add_new_anon_rmap because those pages |
| 555 | * are initially only visible via the pagetables, and the pte is locked |
| 556 | * over the call to page_add_new_anon_rmap. |
| 557 | */ |
| 558 | struct anon_vma *anon_vma = vma->anon_vma; |
| 559 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
| 560 | BUG_ON(page->mapping != (struct address_space *)anon_vma); |
| 561 | BUG_ON(page->index != linear_page_index(vma, address)); |
| 562 | #endif |
| 563 | } |
| 564 | |
| 565 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 566 | * page_add_anon_rmap - add pte mapping to an anonymous page |
| 567 | * @page: the page to add the mapping to |
| 568 | * @vma: the vm area in which the mapping is added |
| 569 | * @address: the user virtual address mapped |
| 570 | * |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 571 | * 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] | 572 | */ |
| 573 | void page_add_anon_rmap(struct page *page, |
| 574 | struct vm_area_struct *vma, unsigned long address) |
| 575 | { |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 576 | VM_BUG_ON(!PageLocked(page)); |
| 577 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 578 | if (atomic_inc_and_test(&page->_mapcount)) |
| 579 | __page_set_anon_rmap(page, vma, address); |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 580 | else |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 581 | __page_check_anon_rmap(page, vma, address); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 582 | } |
| 583 | |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 584 | /** |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 585 | * page_add_new_anon_rmap - add pte mapping to a new anonymous page |
| 586 | * @page: the page to add the mapping to |
| 587 | * @vma: the vm area in which the mapping is added |
| 588 | * @address: the user virtual address mapped |
| 589 | * |
| 590 | * Same as page_add_anon_rmap but must only be called on *new* pages. |
| 591 | * This means the inc-and-test can be bypassed. |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 592 | * Page does not have to be locked. |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 593 | */ |
| 594 | void page_add_new_anon_rmap(struct page *page, |
| 595 | struct vm_area_struct *vma, unsigned long address) |
| 596 | { |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 597 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
Nick Piggin | 9617d95 | 2006-01-06 00:11:12 -0800 | [diff] [blame] | 598 | atomic_set(&page->_mapcount, 0); /* elevate count by 1 (starts at -1) */ |
| 599 | __page_set_anon_rmap(page, vma, address); |
| 600 | } |
| 601 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 602 | /** |
| 603 | * page_add_file_rmap - add pte mapping to a file page |
| 604 | * @page: the page to add the mapping to |
| 605 | * |
Hugh Dickins | b8072f0 | 2005-10-29 18:16:41 -0700 | [diff] [blame] | 606 | * The caller needs to hold the pte lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 607 | */ |
| 608 | void page_add_file_rmap(struct page *page) |
| 609 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 610 | if (atomic_inc_and_test(&page->_mapcount)) |
Christoph Lameter | 65ba55f | 2006-06-30 01:55:34 -0700 | [diff] [blame] | 611 | __inc_zone_page_state(page, NR_FILE_MAPPED); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 612 | } |
| 613 | |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 614 | #ifdef CONFIG_DEBUG_VM |
| 615 | /** |
| 616 | * page_dup_rmap - duplicate pte mapping to a page |
| 617 | * @page: the page to add the mapping to |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 618 | * @vma: the vm area being duplicated |
| 619 | * @address: the user virtual address mapped |
Nick Piggin | c97a9e1 | 2007-05-16 22:11:21 -0700 | [diff] [blame] | 620 | * |
| 621 | * For copy_page_range only: minimal extract from page_add_file_rmap / |
| 622 | * page_add_anon_rmap, avoiding unnecessary tests (already checked) so it's |
| 623 | * quicker. |
| 624 | * |
| 625 | * The caller needs to hold the pte lock. |
| 626 | */ |
| 627 | void page_dup_rmap(struct page *page, struct vm_area_struct *vma, unsigned long address) |
| 628 | { |
| 629 | BUG_ON(page_mapcount(page) == 0); |
| 630 | if (PageAnon(page)) |
| 631 | __page_check_anon_rmap(page, vma, address); |
| 632 | atomic_inc(&page->_mapcount); |
| 633 | } |
| 634 | #endif |
| 635 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 636 | /** |
| 637 | * page_remove_rmap - take down pte mapping from a page |
| 638 | * @page: page to remove mapping from |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 639 | * @vma: the vm area in which the mapping is removed |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 640 | * |
Hugh Dickins | b8072f0 | 2005-10-29 18:16:41 -0700 | [diff] [blame] | 641 | * The caller needs to hold the pte lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 642 | */ |
Nick Piggin | 7de6b80 | 2006-12-22 01:09:33 -0800 | [diff] [blame] | 643 | void page_remove_rmap(struct page *page, struct vm_area_struct *vma) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 644 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 645 | if (atomic_add_negative(-1, &page->_mapcount)) { |
Nick Piggin | b7ab795 | 2006-03-22 00:08:42 -0800 | [diff] [blame] | 646 | if (unlikely(page_mapcount(page) < 0)) { |
Dave Jones | ef2bf0d | 2006-01-08 01:01:00 -0800 | [diff] [blame] | 647 | 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] | 648 | printk (KERN_EMERG " page pfn = %lx\n", page_to_pfn(page)); |
Dave Jones | ef2bf0d | 2006-01-08 01:01:00 -0800 | [diff] [blame] | 649 | printk (KERN_EMERG " page->flags = %lx\n", page->flags); |
| 650 | printk (KERN_EMERG " page->count = %x\n", page_count(page)); |
| 651 | printk (KERN_EMERG " page->mapping = %p\n", page->mapping); |
Nick Piggin | 7de6b80 | 2006-12-22 01:09:33 -0800 | [diff] [blame] | 652 | 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] | 653 | if (vma->vm_ops) { |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 654 | print_symbol (KERN_EMERG " vma->vm_ops->fault = %s\n", (unsigned long)vma->vm_ops->fault); |
| 655 | } |
Nick Piggin | 7de6b80 | 2006-12-22 01:09:33 -0800 | [diff] [blame] | 656 | if (vma->vm_file && vma->vm_file->f_op) |
| 657 | 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] | 658 | BUG(); |
Dave Jones | ef2bf0d | 2006-01-08 01:01:00 -0800 | [diff] [blame] | 659 | } |
Dave Jones | b16bc64 | 2006-10-11 01:21:27 -0700 | [diff] [blame] | 660 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 661 | /* |
Hugh Dickins | 16f8c5b | 2008-08-20 14:09:04 -0700 | [diff] [blame^] | 662 | * Now that the last pte has gone, s390 must transfer dirty |
| 663 | * flag from storage key to struct page. We can usually skip |
| 664 | * this if the page is anon, so about to be freed; but perhaps |
| 665 | * not if it's in swapcache - there might be another pte slot |
| 666 | * containing the swap entry, but page not yet written to swap. |
| 667 | */ |
| 668 | if ((!PageAnon(page) || PageSwapCache(page)) && |
| 669 | page_test_dirty(page)) { |
| 670 | page_clear_dirty(page); |
| 671 | set_page_dirty(page); |
| 672 | } |
| 673 | |
| 674 | mem_cgroup_uncharge_page(page); |
| 675 | __dec_zone_page_state(page, |
| 676 | PageAnon(page) ? NR_ANON_PAGES : NR_FILE_MAPPED); |
| 677 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 678 | * It would be tidy to reset the PageAnon mapping here, |
| 679 | * but that might overwrite a racing page_add_anon_rmap |
| 680 | * which increments mapcount after us but sets mapping |
| 681 | * before us: so leave the reset to free_hot_cold_page, |
| 682 | * and remember that it's only reliable while mapped. |
| 683 | * Leaving it set also helps swapoff to reinstate ptes |
| 684 | * faster for those pages still in swapcache. |
| 685 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 686 | } |
| 687 | } |
| 688 | |
| 689 | /* |
| 690 | * Subfunctions of try_to_unmap: try_to_unmap_one called |
| 691 | * repeatedly from either try_to_unmap_anon or try_to_unmap_file. |
| 692 | */ |
Christoph Lameter | a48d07a | 2006-02-01 03:05:38 -0800 | [diff] [blame] | 693 | 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] | 694 | int migration) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 695 | { |
| 696 | struct mm_struct *mm = vma->vm_mm; |
| 697 | unsigned long address; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 698 | pte_t *pte; |
| 699 | pte_t pteval; |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 700 | spinlock_t *ptl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 701 | int ret = SWAP_AGAIN; |
| 702 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 703 | address = vma_address(page, vma); |
| 704 | if (address == -EFAULT) |
| 705 | goto out; |
| 706 | |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 707 | pte = page_check_address(page, mm, address, &ptl); |
| 708 | if (!pte) |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 709 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 710 | |
| 711 | /* |
| 712 | * If the page is mlock()d, we cannot swap it out. |
| 713 | * If it's recently referenced (perhaps page_referenced |
| 714 | * skipped over this mm) then we should reactivate it. |
| 715 | */ |
Christoph Lameter | e6a1530 | 2006-06-25 05:46:49 -0700 | [diff] [blame] | 716 | if (!migration && ((vma->vm_flags & VM_LOCKED) || |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 717 | (ptep_clear_flush_young_notify(vma, address, pte)))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 718 | ret = SWAP_FAIL; |
| 719 | goto out_unmap; |
| 720 | } |
| 721 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 722 | /* Nuke the page table entry. */ |
| 723 | flush_cache_page(vma, address, page_to_pfn(page)); |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 724 | pteval = ptep_clear_flush_notify(vma, address, pte); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 725 | |
| 726 | /* Move the dirty bit to the physical page now the pte is gone. */ |
| 727 | if (pte_dirty(pteval)) |
| 728 | set_page_dirty(page); |
| 729 | |
Hugh Dickins | 365e9c87 | 2005-10-29 18:16:18 -0700 | [diff] [blame] | 730 | /* Update high watermark before we lower rss */ |
| 731 | update_hiwater_rss(mm); |
| 732 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 733 | if (PageAnon(page)) { |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 734 | swp_entry_t entry = { .val = page_private(page) }; |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 735 | |
| 736 | if (PageSwapCache(page)) { |
| 737 | /* |
| 738 | * Store the swap location in the pte. |
| 739 | * See handle_pte_fault() ... |
| 740 | */ |
| 741 | swap_duplicate(entry); |
| 742 | if (list_empty(&mm->mmlist)) { |
| 743 | spin_lock(&mmlist_lock); |
| 744 | if (list_empty(&mm->mmlist)) |
| 745 | list_add(&mm->mmlist, &init_mm.mmlist); |
| 746 | spin_unlock(&mmlist_lock); |
| 747 | } |
Christoph Lameter | 442c913 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 748 | dec_mm_counter(mm, anon_rss); |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 749 | #ifdef CONFIG_MIGRATION |
Christoph Lameter | 0697212 | 2006-06-23 02:03:35 -0700 | [diff] [blame] | 750 | } else { |
| 751 | /* |
| 752 | * Store the pfn of the page in a special migration |
| 753 | * pte. do_swap_page() will wait until the migration |
| 754 | * pte is removed and then restart fault handling. |
| 755 | */ |
| 756 | BUG_ON(!migration); |
| 757 | entry = make_migration_entry(page, pte_write(pteval)); |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 758 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 759 | } |
| 760 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); |
| 761 | BUG_ON(pte_file(*pte)); |
Hugh Dickins | 4294621 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 762 | } else |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 763 | #ifdef CONFIG_MIGRATION |
| 764 | if (migration) { |
| 765 | /* Establish migration entry for a file page */ |
| 766 | swp_entry_t entry; |
| 767 | entry = make_migration_entry(page, pte_write(pteval)); |
| 768 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); |
| 769 | } else |
| 770 | #endif |
Hugh Dickins | 4294621 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 771 | dec_mm_counter(mm, file_rss); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 772 | |
Christoph Lameter | 04e62a2 | 2006-06-23 02:03:38 -0700 | [diff] [blame] | 773 | |
Nick Piggin | 7de6b80 | 2006-12-22 01:09:33 -0800 | [diff] [blame] | 774 | page_remove_rmap(page, vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 775 | page_cache_release(page); |
| 776 | |
| 777 | out_unmap: |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 778 | pte_unmap_unlock(pte, ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 779 | out: |
| 780 | return ret; |
| 781 | } |
| 782 | |
| 783 | /* |
| 784 | * objrmap doesn't work for nonlinear VMAs because the assumption that |
| 785 | * offset-into-file correlates with offset-into-virtual-addresses does not hold. |
| 786 | * Consequently, given a particular page and its ->index, we cannot locate the |
| 787 | * ptes which are mapping that page without an exhaustive linear search. |
| 788 | * |
| 789 | * So what this code does is a mini "virtual scan" of each nonlinear VMA which |
| 790 | * maps the file to which the target page belongs. The ->vm_private_data field |
| 791 | * holds the current cursor into that scan. Successive searches will circulate |
| 792 | * around the vma's virtual address space. |
| 793 | * |
| 794 | * So as more replacement pressure is applied to the pages in a nonlinear VMA, |
| 795 | * more scanning pressure is placed against them as well. Eventually pages |
| 796 | * will become fully unmapped and are eligible for eviction. |
| 797 | * |
| 798 | * For very sparsely populated VMAs this is a little inefficient - chances are |
| 799 | * there there won't be many ptes located within the scan cluster. In this case |
| 800 | * maybe we could scan further - to the end of the pte page, perhaps. |
| 801 | */ |
| 802 | #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE) |
| 803 | #define CLUSTER_MASK (~(CLUSTER_SIZE - 1)) |
| 804 | |
| 805 | static void try_to_unmap_cluster(unsigned long cursor, |
| 806 | unsigned int *mapcount, struct vm_area_struct *vma) |
| 807 | { |
| 808 | struct mm_struct *mm = vma->vm_mm; |
| 809 | pgd_t *pgd; |
| 810 | pud_t *pud; |
| 811 | pmd_t *pmd; |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 812 | pte_t *pte; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 813 | pte_t pteval; |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 814 | spinlock_t *ptl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 815 | struct page *page; |
| 816 | unsigned long address; |
| 817 | unsigned long end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 818 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 819 | address = (vma->vm_start + cursor) & CLUSTER_MASK; |
| 820 | end = address + CLUSTER_SIZE; |
| 821 | if (address < vma->vm_start) |
| 822 | address = vma->vm_start; |
| 823 | if (end > vma->vm_end) |
| 824 | end = vma->vm_end; |
| 825 | |
| 826 | pgd = pgd_offset(mm, address); |
| 827 | if (!pgd_present(*pgd)) |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 828 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 829 | |
| 830 | pud = pud_offset(pgd, address); |
| 831 | if (!pud_present(*pud)) |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 832 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 833 | |
| 834 | pmd = pmd_offset(pud, address); |
| 835 | if (!pmd_present(*pmd)) |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 836 | return; |
| 837 | |
| 838 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 839 | |
Hugh Dickins | 365e9c87 | 2005-10-29 18:16:18 -0700 | [diff] [blame] | 840 | /* Update high watermark before we lower rss */ |
| 841 | update_hiwater_rss(mm); |
| 842 | |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 843 | for (; address < end; pte++, address += PAGE_SIZE) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 844 | if (!pte_present(*pte)) |
| 845 | continue; |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 846 | page = vm_normal_page(vma, address, *pte); |
| 847 | BUG_ON(!page || PageAnon(page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 848 | |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 849 | if (ptep_clear_flush_young_notify(vma, address, pte)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 850 | continue; |
| 851 | |
| 852 | /* Nuke the page table entry. */ |
Ben Collins | eca3513 | 2005-11-29 11:45:26 -0800 | [diff] [blame] | 853 | flush_cache_page(vma, address, pte_pfn(*pte)); |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 854 | pteval = ptep_clear_flush_notify(vma, address, pte); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 855 | |
| 856 | /* If nonlinear, store the file page offset in the pte. */ |
| 857 | if (page->index != linear_page_index(vma, address)) |
| 858 | set_pte_at(mm, address, pte, pgoff_to_pte(page->index)); |
| 859 | |
| 860 | /* Move the dirty bit to the physical page now the pte is gone. */ |
| 861 | if (pte_dirty(pteval)) |
| 862 | set_page_dirty(page); |
| 863 | |
Nick Piggin | 7de6b80 | 2006-12-22 01:09:33 -0800 | [diff] [blame] | 864 | page_remove_rmap(page, vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 865 | page_cache_release(page); |
Hugh Dickins | 4294621 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 866 | dec_mm_counter(mm, file_rss); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 867 | (*mapcount)--; |
| 868 | } |
Hugh Dickins | c071880 | 2005-10-29 18:16:31 -0700 | [diff] [blame] | 869 | pte_unmap_unlock(pte - 1, ptl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 870 | } |
| 871 | |
Christoph Lameter | 7352349 | 2006-06-23 02:03:27 -0700 | [diff] [blame] | 872 | static int try_to_unmap_anon(struct page *page, int migration) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 873 | { |
| 874 | struct anon_vma *anon_vma; |
| 875 | struct vm_area_struct *vma; |
| 876 | int ret = SWAP_AGAIN; |
| 877 | |
| 878 | anon_vma = page_lock_anon_vma(page); |
| 879 | if (!anon_vma) |
| 880 | return ret; |
| 881 | |
| 882 | list_for_each_entry(vma, &anon_vma->head, anon_vma_node) { |
Christoph Lameter | 7352349 | 2006-06-23 02:03:27 -0700 | [diff] [blame] | 883 | ret = try_to_unmap_one(page, vma, migration); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 884 | if (ret == SWAP_FAIL || !page_mapped(page)) |
| 885 | break; |
| 886 | } |
Oleg Nesterov | 34bbd70 | 2007-02-28 20:13:49 -0800 | [diff] [blame] | 887 | |
| 888 | page_unlock_anon_vma(anon_vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 889 | return ret; |
| 890 | } |
| 891 | |
| 892 | /** |
| 893 | * try_to_unmap_file - unmap file page using the object-based rmap method |
| 894 | * @page: the page to unmap |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 895 | * @migration: migration flag |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 896 | * |
| 897 | * Find all the mappings of a page using the mapping pointer and the vma chains |
| 898 | * contained in the address_space struct it points to. |
| 899 | * |
| 900 | * This function is only called from try_to_unmap for object-based pages. |
| 901 | */ |
Christoph Lameter | 7352349 | 2006-06-23 02:03:27 -0700 | [diff] [blame] | 902 | static int try_to_unmap_file(struct page *page, int migration) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 903 | { |
| 904 | struct address_space *mapping = page->mapping; |
| 905 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); |
| 906 | struct vm_area_struct *vma; |
| 907 | struct prio_tree_iter iter; |
| 908 | int ret = SWAP_AGAIN; |
| 909 | unsigned long cursor; |
| 910 | unsigned long max_nl_cursor = 0; |
| 911 | unsigned long max_nl_size = 0; |
| 912 | unsigned int mapcount; |
| 913 | |
| 914 | spin_lock(&mapping->i_mmap_lock); |
| 915 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { |
Christoph Lameter | 7352349 | 2006-06-23 02:03:27 -0700 | [diff] [blame] | 916 | ret = try_to_unmap_one(page, vma, migration); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 917 | if (ret == SWAP_FAIL || !page_mapped(page)) |
| 918 | goto out; |
| 919 | } |
| 920 | |
| 921 | if (list_empty(&mapping->i_mmap_nonlinear)) |
| 922 | goto out; |
| 923 | |
| 924 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, |
| 925 | shared.vm_set.list) { |
Christoph Lameter | e6a1530 | 2006-06-25 05:46:49 -0700 | [diff] [blame] | 926 | if ((vma->vm_flags & VM_LOCKED) && !migration) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 927 | continue; |
| 928 | cursor = (unsigned long) vma->vm_private_data; |
| 929 | if (cursor > max_nl_cursor) |
| 930 | max_nl_cursor = cursor; |
| 931 | cursor = vma->vm_end - vma->vm_start; |
| 932 | if (cursor > max_nl_size) |
| 933 | max_nl_size = cursor; |
| 934 | } |
| 935 | |
| 936 | if (max_nl_size == 0) { /* any nonlinears locked or reserved */ |
| 937 | ret = SWAP_FAIL; |
| 938 | goto out; |
| 939 | } |
| 940 | |
| 941 | /* |
| 942 | * We don't try to search for this page in the nonlinear vmas, |
| 943 | * and page_referenced wouldn't have found it anyway. Instead |
| 944 | * just walk the nonlinear vmas trying to age and unmap some. |
| 945 | * The mapcount of the page we came in with is irrelevant, |
| 946 | * but even so use it as a guide to how hard we should try? |
| 947 | */ |
| 948 | mapcount = page_mapcount(page); |
| 949 | if (!mapcount) |
| 950 | goto out; |
| 951 | cond_resched_lock(&mapping->i_mmap_lock); |
| 952 | |
| 953 | max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK; |
| 954 | if (max_nl_cursor == 0) |
| 955 | max_nl_cursor = CLUSTER_SIZE; |
| 956 | |
| 957 | do { |
| 958 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, |
| 959 | shared.vm_set.list) { |
Christoph Lameter | e6a1530 | 2006-06-25 05:46:49 -0700 | [diff] [blame] | 960 | if ((vma->vm_flags & VM_LOCKED) && !migration) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 961 | continue; |
| 962 | cursor = (unsigned long) vma->vm_private_data; |
Hugh Dickins | 839b968 | 2005-09-03 15:54:43 -0700 | [diff] [blame] | 963 | while ( cursor < max_nl_cursor && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 964 | cursor < vma->vm_end - vma->vm_start) { |
| 965 | try_to_unmap_cluster(cursor, &mapcount, vma); |
| 966 | cursor += CLUSTER_SIZE; |
| 967 | vma->vm_private_data = (void *) cursor; |
| 968 | if ((int)mapcount <= 0) |
| 969 | goto out; |
| 970 | } |
| 971 | vma->vm_private_data = (void *) max_nl_cursor; |
| 972 | } |
| 973 | cond_resched_lock(&mapping->i_mmap_lock); |
| 974 | max_nl_cursor += CLUSTER_SIZE; |
| 975 | } while (max_nl_cursor <= max_nl_size); |
| 976 | |
| 977 | /* |
| 978 | * Don't loop forever (perhaps all the remaining pages are |
| 979 | * in locked vmas). Reset cursor on all unreserved nonlinear |
| 980 | * vmas, now forgetting on which ones it had fallen behind. |
| 981 | */ |
Hugh Dickins | 101d2be | 2005-11-21 21:32:16 -0800 | [diff] [blame] | 982 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) |
| 983 | vma->vm_private_data = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 984 | out: |
| 985 | spin_unlock(&mapping->i_mmap_lock); |
| 986 | return ret; |
| 987 | } |
| 988 | |
| 989 | /** |
| 990 | * try_to_unmap - try to remove all page table mappings to a page |
| 991 | * @page: the page to get unmapped |
Randy Dunlap | 43d8eac | 2008-03-19 17:00:43 -0700 | [diff] [blame] | 992 | * @migration: migration flag |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 993 | * |
| 994 | * Tries to remove all the page table entries which are mapping this |
| 995 | * page, used in the pageout path. Caller must hold the page lock. |
| 996 | * Return values are: |
| 997 | * |
| 998 | * SWAP_SUCCESS - we succeeded in removing all mappings |
| 999 | * SWAP_AGAIN - we missed a mapping, try again later |
| 1000 | * SWAP_FAIL - the page is unswappable |
| 1001 | */ |
Christoph Lameter | 7352349 | 2006-06-23 02:03:27 -0700 | [diff] [blame] | 1002 | int try_to_unmap(struct page *page, int migration) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1003 | { |
| 1004 | int ret; |
| 1005 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1006 | BUG_ON(!PageLocked(page)); |
| 1007 | |
| 1008 | if (PageAnon(page)) |
Christoph Lameter | 7352349 | 2006-06-23 02:03:27 -0700 | [diff] [blame] | 1009 | ret = try_to_unmap_anon(page, migration); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1010 | else |
Christoph Lameter | 7352349 | 2006-06-23 02:03:27 -0700 | [diff] [blame] | 1011 | ret = try_to_unmap_file(page, migration); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1012 | |
| 1013 | if (!page_mapped(page)) |
| 1014 | ret = SWAP_SUCCESS; |
| 1015 | return ret; |
| 1016 | } |
Nikita Danilov | 81b4082 | 2005-05-01 08:58:36 -0700 | [diff] [blame] | 1017 | |