Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | |
98f32602 | 17 | * Contributions by Hugh Dickins 2003, 2004 |
1da177e4 LT |
18 | */ |
19 | ||
20 | /* | |
21 | * Lock ordering in mm: | |
22 | * | |
1b1dcc1b | 23 | * inode->i_mutex (while writing or truncating, not reading or faulting) |
82591e6e NP |
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) | |
6a46079c AK |
39 | * |
40 | * (code doesn't rely on that order so it could be switched around) | |
41 | * ->tasklist_lock | |
42 | * anon_vma->lock (memory_failure, collect_procs_anon) | |
43 | * pte map lock | |
1da177e4 LT |
44 | */ |
45 | ||
46 | #include <linux/mm.h> | |
47 | #include <linux/pagemap.h> | |
48 | #include <linux/swap.h> | |
49 | #include <linux/swapops.h> | |
50 | #include <linux/slab.h> | |
51 | #include <linux/init.h> | |
52 | #include <linux/rmap.h> | |
53 | #include <linux/rcupdate.h> | |
a48d07af | 54 | #include <linux/module.h> |
8a9f3ccd | 55 | #include <linux/memcontrol.h> |
cddb8a5c | 56 | #include <linux/mmu_notifier.h> |
64cdd548 | 57 | #include <linux/migrate.h> |
1da177e4 LT |
58 | |
59 | #include <asm/tlbflush.h> | |
60 | ||
b291f000 NP |
61 | #include "internal.h" |
62 | ||
fdd2e5f8 AB |
63 | static struct kmem_cache *anon_vma_cachep; |
64 | ||
65 | static inline struct anon_vma *anon_vma_alloc(void) | |
66 | { | |
67 | return kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL); | |
68 | } | |
69 | ||
70 | static inline void anon_vma_free(struct anon_vma *anon_vma) | |
71 | { | |
72 | kmem_cache_free(anon_vma_cachep, anon_vma); | |
73 | } | |
1da177e4 | 74 | |
d9d332e0 LT |
75 | /** |
76 | * anon_vma_prepare - attach an anon_vma to a memory region | |
77 | * @vma: the memory region in question | |
78 | * | |
79 | * This makes sure the memory mapping described by 'vma' has | |
80 | * an 'anon_vma' attached to it, so that we can associate the | |
81 | * anonymous pages mapped into it with that anon_vma. | |
82 | * | |
83 | * The common case will be that we already have one, but if | |
84 | * if not we either need to find an adjacent mapping that we | |
85 | * can re-use the anon_vma from (very common when the only | |
86 | * reason for splitting a vma has been mprotect()), or we | |
87 | * allocate a new one. | |
88 | * | |
89 | * Anon-vma allocations are very subtle, because we may have | |
90 | * optimistically looked up an anon_vma in page_lock_anon_vma() | |
91 | * and that may actually touch the spinlock even in the newly | |
92 | * allocated vma (it depends on RCU to make sure that the | |
93 | * anon_vma isn't actually destroyed). | |
94 | * | |
95 | * As a result, we need to do proper anon_vma locking even | |
96 | * for the new allocation. At the same time, we do not want | |
97 | * to do any locking for the common case of already having | |
98 | * an anon_vma. | |
99 | * | |
100 | * This must be called with the mmap_sem held for reading. | |
101 | */ | |
1da177e4 LT |
102 | int anon_vma_prepare(struct vm_area_struct *vma) |
103 | { | |
104 | struct anon_vma *anon_vma = vma->anon_vma; | |
105 | ||
106 | might_sleep(); | |
107 | if (unlikely(!anon_vma)) { | |
108 | struct mm_struct *mm = vma->vm_mm; | |
d9d332e0 | 109 | struct anon_vma *allocated; |
1da177e4 LT |
110 | |
111 | anon_vma = find_mergeable_anon_vma(vma); | |
d9d332e0 LT |
112 | allocated = NULL; |
113 | if (!anon_vma) { | |
1da177e4 LT |
114 | anon_vma = anon_vma_alloc(); |
115 | if (unlikely(!anon_vma)) | |
116 | return -ENOMEM; | |
117 | allocated = anon_vma; | |
1da177e4 | 118 | } |
d9d332e0 | 119 | spin_lock(&anon_vma->lock); |
1da177e4 LT |
120 | |
121 | /* page_table_lock to protect against threads */ | |
122 | spin_lock(&mm->page_table_lock); | |
123 | if (likely(!vma->anon_vma)) { | |
124 | vma->anon_vma = anon_vma; | |
0697212a | 125 | list_add_tail(&vma->anon_vma_node, &anon_vma->head); |
1da177e4 LT |
126 | allocated = NULL; |
127 | } | |
128 | spin_unlock(&mm->page_table_lock); | |
129 | ||
d9d332e0 | 130 | spin_unlock(&anon_vma->lock); |
1da177e4 LT |
131 | if (unlikely(allocated)) |
132 | anon_vma_free(allocated); | |
133 | } | |
134 | return 0; | |
135 | } | |
136 | ||
137 | void __anon_vma_merge(struct vm_area_struct *vma, struct vm_area_struct *next) | |
138 | { | |
139 | BUG_ON(vma->anon_vma != next->anon_vma); | |
140 | list_del(&next->anon_vma_node); | |
141 | } | |
142 | ||
143 | void __anon_vma_link(struct vm_area_struct *vma) | |
144 | { | |
145 | struct anon_vma *anon_vma = vma->anon_vma; | |
146 | ||
30acbaba | 147 | if (anon_vma) |
0697212a | 148 | list_add_tail(&vma->anon_vma_node, &anon_vma->head); |
1da177e4 LT |
149 | } |
150 | ||
151 | void anon_vma_link(struct vm_area_struct *vma) | |
152 | { | |
153 | struct anon_vma *anon_vma = vma->anon_vma; | |
154 | ||
155 | if (anon_vma) { | |
156 | spin_lock(&anon_vma->lock); | |
0697212a | 157 | list_add_tail(&vma->anon_vma_node, &anon_vma->head); |
1da177e4 LT |
158 | spin_unlock(&anon_vma->lock); |
159 | } | |
160 | } | |
161 | ||
162 | void anon_vma_unlink(struct vm_area_struct *vma) | |
163 | { | |
164 | struct anon_vma *anon_vma = vma->anon_vma; | |
165 | int empty; | |
166 | ||
167 | if (!anon_vma) | |
168 | return; | |
169 | ||
170 | spin_lock(&anon_vma->lock); | |
1da177e4 LT |
171 | list_del(&vma->anon_vma_node); |
172 | ||
173 | /* We must garbage collect the anon_vma if it's empty */ | |
174 | empty = list_empty(&anon_vma->head); | |
175 | spin_unlock(&anon_vma->lock); | |
176 | ||
177 | if (empty) | |
178 | anon_vma_free(anon_vma); | |
179 | } | |
180 | ||
51cc5068 | 181 | static void anon_vma_ctor(void *data) |
1da177e4 | 182 | { |
a35afb83 | 183 | struct anon_vma *anon_vma = data; |
1da177e4 | 184 | |
a35afb83 CL |
185 | spin_lock_init(&anon_vma->lock); |
186 | INIT_LIST_HEAD(&anon_vma->head); | |
1da177e4 LT |
187 | } |
188 | ||
189 | void __init anon_vma_init(void) | |
190 | { | |
191 | anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), | |
20c2df83 | 192 | 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor); |
1da177e4 LT |
193 | } |
194 | ||
195 | /* | |
196 | * Getting a lock on a stable anon_vma from a page off the LRU is | |
197 | * tricky: page_lock_anon_vma rely on RCU to guard against the races. | |
198 | */ | |
10be22df | 199 | struct anon_vma *page_lock_anon_vma(struct page *page) |
1da177e4 | 200 | { |
34bbd704 | 201 | struct anon_vma *anon_vma; |
1da177e4 LT |
202 | unsigned long anon_mapping; |
203 | ||
204 | rcu_read_lock(); | |
205 | anon_mapping = (unsigned long) page->mapping; | |
206 | if (!(anon_mapping & PAGE_MAPPING_ANON)) | |
207 | goto out; | |
208 | if (!page_mapped(page)) | |
209 | goto out; | |
210 | ||
211 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); | |
212 | spin_lock(&anon_vma->lock); | |
34bbd704 | 213 | return anon_vma; |
1da177e4 LT |
214 | out: |
215 | rcu_read_unlock(); | |
34bbd704 ON |
216 | return NULL; |
217 | } | |
218 | ||
10be22df | 219 | void page_unlock_anon_vma(struct anon_vma *anon_vma) |
34bbd704 ON |
220 | { |
221 | spin_unlock(&anon_vma->lock); | |
222 | rcu_read_unlock(); | |
1da177e4 LT |
223 | } |
224 | ||
225 | /* | |
3ad33b24 LS |
226 | * At what user virtual address is page expected in @vma? |
227 | * Returns virtual address or -EFAULT if page's index/offset is not | |
228 | * within the range mapped the @vma. | |
1da177e4 LT |
229 | */ |
230 | static inline unsigned long | |
231 | vma_address(struct page *page, struct vm_area_struct *vma) | |
232 | { | |
233 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
234 | unsigned long address; | |
235 | ||
236 | address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); | |
237 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) { | |
3ad33b24 | 238 | /* page should be within @vma mapping range */ |
1da177e4 LT |
239 | return -EFAULT; |
240 | } | |
241 | return address; | |
242 | } | |
243 | ||
244 | /* | |
bf89c8c8 HS |
245 | * At what user virtual address is page expected in vma? |
246 | * checking that the page matches the vma. | |
1da177e4 LT |
247 | */ |
248 | unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) | |
249 | { | |
250 | if (PageAnon(page)) { | |
251 | if ((void *)vma->anon_vma != | |
252 | (void *)page->mapping - PAGE_MAPPING_ANON) | |
253 | return -EFAULT; | |
254 | } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) { | |
ee498ed7 HD |
255 | if (!vma->vm_file || |
256 | vma->vm_file->f_mapping != page->mapping) | |
1da177e4 LT |
257 | return -EFAULT; |
258 | } else | |
259 | return -EFAULT; | |
260 | return vma_address(page, vma); | |
261 | } | |
262 | ||
81b4082d ND |
263 | /* |
264 | * Check that @page is mapped at @address into @mm. | |
265 | * | |
479db0bf NP |
266 | * If @sync is false, page_check_address may perform a racy check to avoid |
267 | * the page table lock when the pte is not present (helpful when reclaiming | |
268 | * highly shared pages). | |
269 | * | |
b8072f09 | 270 | * On success returns with pte mapped and locked. |
81b4082d | 271 | */ |
ceffc078 | 272 | pte_t *page_check_address(struct page *page, struct mm_struct *mm, |
479db0bf | 273 | unsigned long address, spinlock_t **ptlp, int sync) |
81b4082d ND |
274 | { |
275 | pgd_t *pgd; | |
276 | pud_t *pud; | |
277 | pmd_t *pmd; | |
278 | pte_t *pte; | |
c0718806 | 279 | spinlock_t *ptl; |
81b4082d | 280 | |
81b4082d | 281 | pgd = pgd_offset(mm, address); |
c0718806 HD |
282 | if (!pgd_present(*pgd)) |
283 | return NULL; | |
284 | ||
285 | pud = pud_offset(pgd, address); | |
286 | if (!pud_present(*pud)) | |
287 | return NULL; | |
288 | ||
289 | pmd = pmd_offset(pud, address); | |
290 | if (!pmd_present(*pmd)) | |
291 | return NULL; | |
292 | ||
293 | pte = pte_offset_map(pmd, address); | |
294 | /* Make a quick check before getting the lock */ | |
479db0bf | 295 | if (!sync && !pte_present(*pte)) { |
c0718806 HD |
296 | pte_unmap(pte); |
297 | return NULL; | |
298 | } | |
299 | ||
4c21e2f2 | 300 | ptl = pte_lockptr(mm, pmd); |
c0718806 HD |
301 | spin_lock(ptl); |
302 | if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) { | |
303 | *ptlp = ptl; | |
304 | return pte; | |
81b4082d | 305 | } |
c0718806 HD |
306 | pte_unmap_unlock(pte, ptl); |
307 | return NULL; | |
81b4082d ND |
308 | } |
309 | ||
b291f000 NP |
310 | /** |
311 | * page_mapped_in_vma - check whether a page is really mapped in a VMA | |
312 | * @page: the page to test | |
313 | * @vma: the VMA to test | |
314 | * | |
315 | * Returns 1 if the page is mapped into the page tables of the VMA, 0 | |
316 | * if the page is not mapped into the page tables of this VMA. Only | |
317 | * valid for normal file or anonymous VMAs. | |
318 | */ | |
6a46079c | 319 | int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma) |
b291f000 NP |
320 | { |
321 | unsigned long address; | |
322 | pte_t *pte; | |
323 | spinlock_t *ptl; | |
324 | ||
325 | address = vma_address(page, vma); | |
326 | if (address == -EFAULT) /* out of vma range */ | |
327 | return 0; | |
328 | pte = page_check_address(page, vma->vm_mm, address, &ptl, 1); | |
329 | if (!pte) /* the page is not in this mm */ | |
330 | return 0; | |
331 | pte_unmap_unlock(pte, ptl); | |
332 | ||
333 | return 1; | |
334 | } | |
335 | ||
1da177e4 LT |
336 | /* |
337 | * Subfunctions of page_referenced: page_referenced_one called | |
338 | * repeatedly from either page_referenced_anon or page_referenced_file. | |
339 | */ | |
340 | static int page_referenced_one(struct page *page, | |
6fe6b7e3 WF |
341 | struct vm_area_struct *vma, |
342 | unsigned int *mapcount, | |
343 | unsigned long *vm_flags) | |
1da177e4 LT |
344 | { |
345 | struct mm_struct *mm = vma->vm_mm; | |
346 | unsigned long address; | |
1da177e4 | 347 | pte_t *pte; |
c0718806 | 348 | spinlock_t *ptl; |
1da177e4 LT |
349 | int referenced = 0; |
350 | ||
1da177e4 LT |
351 | address = vma_address(page, vma); |
352 | if (address == -EFAULT) | |
353 | goto out; | |
354 | ||
479db0bf | 355 | pte = page_check_address(page, mm, address, &ptl, 0); |
c0718806 HD |
356 | if (!pte) |
357 | goto out; | |
1da177e4 | 358 | |
b291f000 NP |
359 | /* |
360 | * Don't want to elevate referenced for mlocked page that gets this far, | |
361 | * in order that it progresses to try_to_unmap and is moved to the | |
362 | * unevictable list. | |
363 | */ | |
5a9bbdcd | 364 | if (vma->vm_flags & VM_LOCKED) { |
5a9bbdcd | 365 | *mapcount = 1; /* break early from loop */ |
03ef83af | 366 | *vm_flags |= VM_LOCKED; |
b291f000 NP |
367 | goto out_unmap; |
368 | } | |
369 | ||
4917e5d0 JW |
370 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
371 | /* | |
372 | * Don't treat a reference through a sequentially read | |
373 | * mapping as such. If the page has been used in | |
374 | * another mapping, we will catch it; if this other | |
375 | * mapping is already gone, the unmap path will have | |
376 | * set PG_referenced or activated the page. | |
377 | */ | |
378 | if (likely(!VM_SequentialReadHint(vma))) | |
379 | referenced++; | |
380 | } | |
1da177e4 | 381 | |
c0718806 HD |
382 | /* Pretend the page is referenced if the task has the |
383 | swap token and is in the middle of a page fault. */ | |
f7b7fd8f | 384 | if (mm != current->mm && has_swap_token(mm) && |
c0718806 HD |
385 | rwsem_is_locked(&mm->mmap_sem)) |
386 | referenced++; | |
387 | ||
b291f000 | 388 | out_unmap: |
c0718806 HD |
389 | (*mapcount)--; |
390 | pte_unmap_unlock(pte, ptl); | |
273f047e | 391 | |
6fe6b7e3 WF |
392 | if (referenced) |
393 | *vm_flags |= vma->vm_flags; | |
273f047e | 394 | out: |
1da177e4 LT |
395 | return referenced; |
396 | } | |
397 | ||
bed7161a | 398 | static int page_referenced_anon(struct page *page, |
6fe6b7e3 WF |
399 | struct mem_cgroup *mem_cont, |
400 | unsigned long *vm_flags) | |
1da177e4 LT |
401 | { |
402 | unsigned int mapcount; | |
403 | struct anon_vma *anon_vma; | |
404 | struct vm_area_struct *vma; | |
405 | int referenced = 0; | |
406 | ||
407 | anon_vma = page_lock_anon_vma(page); | |
408 | if (!anon_vma) | |
409 | return referenced; | |
410 | ||
411 | mapcount = page_mapcount(page); | |
412 | list_for_each_entry(vma, &anon_vma->head, anon_vma_node) { | |
bed7161a BS |
413 | /* |
414 | * If we are reclaiming on behalf of a cgroup, skip | |
415 | * counting on behalf of references from different | |
416 | * cgroups | |
417 | */ | |
bd845e38 | 418 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
bed7161a | 419 | continue; |
6fe6b7e3 WF |
420 | referenced += page_referenced_one(page, vma, |
421 | &mapcount, vm_flags); | |
1da177e4 LT |
422 | if (!mapcount) |
423 | break; | |
424 | } | |
34bbd704 ON |
425 | |
426 | page_unlock_anon_vma(anon_vma); | |
1da177e4 LT |
427 | return referenced; |
428 | } | |
429 | ||
430 | /** | |
431 | * page_referenced_file - referenced check for object-based rmap | |
432 | * @page: the page we're checking references on. | |
43d8eac4 | 433 | * @mem_cont: target memory controller |
6fe6b7e3 | 434 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
1da177e4 LT |
435 | * |
436 | * For an object-based mapped page, find all the places it is mapped and | |
437 | * check/clear the referenced flag. This is done by following the page->mapping | |
438 | * pointer, then walking the chain of vmas it holds. It returns the number | |
439 | * of references it found. | |
440 | * | |
441 | * This function is only called from page_referenced for object-based pages. | |
442 | */ | |
bed7161a | 443 | static int page_referenced_file(struct page *page, |
6fe6b7e3 WF |
444 | struct mem_cgroup *mem_cont, |
445 | unsigned long *vm_flags) | |
1da177e4 LT |
446 | { |
447 | unsigned int mapcount; | |
448 | struct address_space *mapping = page->mapping; | |
449 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
450 | struct vm_area_struct *vma; | |
451 | struct prio_tree_iter iter; | |
452 | int referenced = 0; | |
453 | ||
454 | /* | |
455 | * The caller's checks on page->mapping and !PageAnon have made | |
456 | * sure that this is a file page: the check for page->mapping | |
457 | * excludes the case just before it gets set on an anon page. | |
458 | */ | |
459 | BUG_ON(PageAnon(page)); | |
460 | ||
461 | /* | |
462 | * The page lock not only makes sure that page->mapping cannot | |
463 | * suddenly be NULLified by truncation, it makes sure that the | |
464 | * structure at mapping cannot be freed and reused yet, | |
465 | * so we can safely take mapping->i_mmap_lock. | |
466 | */ | |
467 | BUG_ON(!PageLocked(page)); | |
468 | ||
469 | spin_lock(&mapping->i_mmap_lock); | |
470 | ||
471 | /* | |
472 | * i_mmap_lock does not stabilize mapcount at all, but mapcount | |
473 | * is more likely to be accurate if we note it after spinning. | |
474 | */ | |
475 | mapcount = page_mapcount(page); | |
476 | ||
477 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
bed7161a BS |
478 | /* |
479 | * If we are reclaiming on behalf of a cgroup, skip | |
480 | * counting on behalf of references from different | |
481 | * cgroups | |
482 | */ | |
bd845e38 | 483 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
bed7161a | 484 | continue; |
6fe6b7e3 WF |
485 | referenced += page_referenced_one(page, vma, |
486 | &mapcount, vm_flags); | |
1da177e4 LT |
487 | if (!mapcount) |
488 | break; | |
489 | } | |
490 | ||
491 | spin_unlock(&mapping->i_mmap_lock); | |
492 | return referenced; | |
493 | } | |
494 | ||
495 | /** | |
496 | * page_referenced - test if the page was referenced | |
497 | * @page: the page to test | |
498 | * @is_locked: caller holds lock on the page | |
43d8eac4 | 499 | * @mem_cont: target memory controller |
6fe6b7e3 | 500 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
1da177e4 LT |
501 | * |
502 | * Quick test_and_clear_referenced for all mappings to a page, | |
503 | * returns the number of ptes which referenced the page. | |
504 | */ | |
6fe6b7e3 WF |
505 | int page_referenced(struct page *page, |
506 | int is_locked, | |
507 | struct mem_cgroup *mem_cont, | |
508 | unsigned long *vm_flags) | |
1da177e4 LT |
509 | { |
510 | int referenced = 0; | |
511 | ||
1da177e4 LT |
512 | if (TestClearPageReferenced(page)) |
513 | referenced++; | |
514 | ||
6fe6b7e3 | 515 | *vm_flags = 0; |
1da177e4 LT |
516 | if (page_mapped(page) && page->mapping) { |
517 | if (PageAnon(page)) | |
6fe6b7e3 WF |
518 | referenced += page_referenced_anon(page, mem_cont, |
519 | vm_flags); | |
1da177e4 | 520 | else if (is_locked) |
6fe6b7e3 WF |
521 | referenced += page_referenced_file(page, mem_cont, |
522 | vm_flags); | |
529ae9aa | 523 | else if (!trylock_page(page)) |
1da177e4 LT |
524 | referenced++; |
525 | else { | |
526 | if (page->mapping) | |
6fe6b7e3 WF |
527 | referenced += page_referenced_file(page, |
528 | mem_cont, vm_flags); | |
1da177e4 LT |
529 | unlock_page(page); |
530 | } | |
531 | } | |
5b7baf05 CB |
532 | |
533 | if (page_test_and_clear_young(page)) | |
534 | referenced++; | |
535 | ||
1da177e4 LT |
536 | return referenced; |
537 | } | |
538 | ||
d08b3851 PZ |
539 | static int page_mkclean_one(struct page *page, struct vm_area_struct *vma) |
540 | { | |
541 | struct mm_struct *mm = vma->vm_mm; | |
542 | unsigned long address; | |
c2fda5fe | 543 | pte_t *pte; |
d08b3851 PZ |
544 | spinlock_t *ptl; |
545 | int ret = 0; | |
546 | ||
547 | address = vma_address(page, vma); | |
548 | if (address == -EFAULT) | |
549 | goto out; | |
550 | ||
479db0bf | 551 | pte = page_check_address(page, mm, address, &ptl, 1); |
d08b3851 PZ |
552 | if (!pte) |
553 | goto out; | |
554 | ||
c2fda5fe PZ |
555 | if (pte_dirty(*pte) || pte_write(*pte)) { |
556 | pte_t entry; | |
d08b3851 | 557 | |
c2fda5fe | 558 | flush_cache_page(vma, address, pte_pfn(*pte)); |
cddb8a5c | 559 | entry = ptep_clear_flush_notify(vma, address, pte); |
c2fda5fe PZ |
560 | entry = pte_wrprotect(entry); |
561 | entry = pte_mkclean(entry); | |
d6e88e67 | 562 | set_pte_at(mm, address, pte, entry); |
c2fda5fe PZ |
563 | ret = 1; |
564 | } | |
d08b3851 | 565 | |
d08b3851 PZ |
566 | pte_unmap_unlock(pte, ptl); |
567 | out: | |
568 | return ret; | |
569 | } | |
570 | ||
571 | static int page_mkclean_file(struct address_space *mapping, struct page *page) | |
572 | { | |
573 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
574 | struct vm_area_struct *vma; | |
575 | struct prio_tree_iter iter; | |
576 | int ret = 0; | |
577 | ||
578 | BUG_ON(PageAnon(page)); | |
579 | ||
580 | spin_lock(&mapping->i_mmap_lock); | |
581 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
582 | if (vma->vm_flags & VM_SHARED) | |
583 | ret += page_mkclean_one(page, vma); | |
584 | } | |
585 | spin_unlock(&mapping->i_mmap_lock); | |
586 | return ret; | |
587 | } | |
588 | ||
589 | int page_mkclean(struct page *page) | |
590 | { | |
591 | int ret = 0; | |
592 | ||
593 | BUG_ON(!PageLocked(page)); | |
594 | ||
595 | if (page_mapped(page)) { | |
596 | struct address_space *mapping = page_mapping(page); | |
ce7e9fae | 597 | if (mapping) { |
d08b3851 | 598 | ret = page_mkclean_file(mapping, page); |
ce7e9fae CB |
599 | if (page_test_dirty(page)) { |
600 | page_clear_dirty(page); | |
601 | ret = 1; | |
602 | } | |
6c210482 | 603 | } |
d08b3851 PZ |
604 | } |
605 | ||
606 | return ret; | |
607 | } | |
60b59bea | 608 | EXPORT_SYMBOL_GPL(page_mkclean); |
d08b3851 | 609 | |
9617d95e | 610 | /** |
43d8eac4 | 611 | * __page_set_anon_rmap - setup new anonymous rmap |
9617d95e NP |
612 | * @page: the page to add the mapping to |
613 | * @vma: the vm area in which the mapping is added | |
614 | * @address: the user virtual address mapped | |
615 | */ | |
616 | static void __page_set_anon_rmap(struct page *page, | |
617 | struct vm_area_struct *vma, unsigned long address) | |
618 | { | |
619 | struct anon_vma *anon_vma = vma->anon_vma; | |
620 | ||
621 | BUG_ON(!anon_vma); | |
622 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | |
623 | page->mapping = (struct address_space *) anon_vma; | |
624 | ||
625 | page->index = linear_page_index(vma, address); | |
626 | ||
a74609fa NP |
627 | /* |
628 | * nr_mapped state can be updated without turning off | |
629 | * interrupts because it is not modified via interrupt. | |
630 | */ | |
f3dbd344 | 631 | __inc_zone_page_state(page, NR_ANON_PAGES); |
9617d95e NP |
632 | } |
633 | ||
c97a9e10 | 634 | /** |
43d8eac4 | 635 | * __page_check_anon_rmap - sanity check anonymous rmap addition |
c97a9e10 NP |
636 | * @page: the page to add the mapping to |
637 | * @vma: the vm area in which the mapping is added | |
638 | * @address: the user virtual address mapped | |
639 | */ | |
640 | static void __page_check_anon_rmap(struct page *page, | |
641 | struct vm_area_struct *vma, unsigned long address) | |
642 | { | |
643 | #ifdef CONFIG_DEBUG_VM | |
644 | /* | |
645 | * The page's anon-rmap details (mapping and index) are guaranteed to | |
646 | * be set up correctly at this point. | |
647 | * | |
648 | * We have exclusion against page_add_anon_rmap because the caller | |
649 | * always holds the page locked, except if called from page_dup_rmap, | |
650 | * in which case the page is already known to be setup. | |
651 | * | |
652 | * We have exclusion against page_add_new_anon_rmap because those pages | |
653 | * are initially only visible via the pagetables, and the pte is locked | |
654 | * over the call to page_add_new_anon_rmap. | |
655 | */ | |
656 | struct anon_vma *anon_vma = vma->anon_vma; | |
657 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | |
658 | BUG_ON(page->mapping != (struct address_space *)anon_vma); | |
659 | BUG_ON(page->index != linear_page_index(vma, address)); | |
660 | #endif | |
661 | } | |
662 | ||
1da177e4 LT |
663 | /** |
664 | * page_add_anon_rmap - add pte mapping to an anonymous page | |
665 | * @page: the page to add the mapping to | |
666 | * @vma: the vm area in which the mapping is added | |
667 | * @address: the user virtual address mapped | |
668 | * | |
c97a9e10 | 669 | * The caller needs to hold the pte lock and the page must be locked. |
1da177e4 LT |
670 | */ |
671 | void page_add_anon_rmap(struct page *page, | |
672 | struct vm_area_struct *vma, unsigned long address) | |
673 | { | |
c97a9e10 NP |
674 | VM_BUG_ON(!PageLocked(page)); |
675 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
9617d95e NP |
676 | if (atomic_inc_and_test(&page->_mapcount)) |
677 | __page_set_anon_rmap(page, vma, address); | |
69029cd5 | 678 | else |
c97a9e10 | 679 | __page_check_anon_rmap(page, vma, address); |
1da177e4 LT |
680 | } |
681 | ||
43d8eac4 | 682 | /** |
9617d95e NP |
683 | * page_add_new_anon_rmap - add pte mapping to a new anonymous page |
684 | * @page: the page to add the mapping to | |
685 | * @vma: the vm area in which the mapping is added | |
686 | * @address: the user virtual address mapped | |
687 | * | |
688 | * Same as page_add_anon_rmap but must only be called on *new* pages. | |
689 | * This means the inc-and-test can be bypassed. | |
c97a9e10 | 690 | * Page does not have to be locked. |
9617d95e NP |
691 | */ |
692 | void page_add_new_anon_rmap(struct page *page, | |
693 | struct vm_area_struct *vma, unsigned long address) | |
694 | { | |
b5934c53 | 695 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
cbf84b7a HD |
696 | SetPageSwapBacked(page); |
697 | atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */ | |
9617d95e | 698 | __page_set_anon_rmap(page, vma, address); |
b5934c53 | 699 | if (page_evictable(page, vma)) |
cbf84b7a | 700 | lru_cache_add_lru(page, LRU_ACTIVE_ANON); |
b5934c53 HD |
701 | else |
702 | add_page_to_unevictable_list(page); | |
9617d95e NP |
703 | } |
704 | ||
1da177e4 LT |
705 | /** |
706 | * page_add_file_rmap - add pte mapping to a file page | |
707 | * @page: the page to add the mapping to | |
708 | * | |
b8072f09 | 709 | * The caller needs to hold the pte lock. |
1da177e4 LT |
710 | */ |
711 | void page_add_file_rmap(struct page *page) | |
712 | { | |
d69b042f | 713 | if (atomic_inc_and_test(&page->_mapcount)) { |
65ba55f5 | 714 | __inc_zone_page_state(page, NR_FILE_MAPPED); |
d69b042f BS |
715 | mem_cgroup_update_mapped_file_stat(page, 1); |
716 | } | |
1da177e4 LT |
717 | } |
718 | ||
719 | /** | |
720 | * page_remove_rmap - take down pte mapping from a page | |
721 | * @page: page to remove mapping from | |
722 | * | |
b8072f09 | 723 | * The caller needs to hold the pte lock. |
1da177e4 | 724 | */ |
edc315fd | 725 | void page_remove_rmap(struct page *page) |
1da177e4 | 726 | { |
b904dcfe KM |
727 | /* page still mapped by someone else? */ |
728 | if (!atomic_add_negative(-1, &page->_mapcount)) | |
729 | return; | |
730 | ||
731 | /* | |
732 | * Now that the last pte has gone, s390 must transfer dirty | |
733 | * flag from storage key to struct page. We can usually skip | |
734 | * this if the page is anon, so about to be freed; but perhaps | |
735 | * not if it's in swapcache - there might be another pte slot | |
736 | * containing the swap entry, but page not yet written to swap. | |
737 | */ | |
738 | if ((!PageAnon(page) || PageSwapCache(page)) && page_test_dirty(page)) { | |
739 | page_clear_dirty(page); | |
740 | set_page_dirty(page); | |
1da177e4 | 741 | } |
b904dcfe KM |
742 | if (PageAnon(page)) { |
743 | mem_cgroup_uncharge_page(page); | |
744 | __dec_zone_page_state(page, NR_ANON_PAGES); | |
745 | } else { | |
746 | __dec_zone_page_state(page, NR_FILE_MAPPED); | |
747 | } | |
748 | mem_cgroup_update_mapped_file_stat(page, -1); | |
749 | /* | |
750 | * It would be tidy to reset the PageAnon mapping here, | |
751 | * but that might overwrite a racing page_add_anon_rmap | |
752 | * which increments mapcount after us but sets mapping | |
753 | * before us: so leave the reset to free_hot_cold_page, | |
754 | * and remember that it's only reliable while mapped. | |
755 | * Leaving it set also helps swapoff to reinstate ptes | |
756 | * faster for those pages still in swapcache. | |
757 | */ | |
1da177e4 LT |
758 | } |
759 | ||
760 | /* | |
761 | * Subfunctions of try_to_unmap: try_to_unmap_one called | |
762 | * repeatedly from either try_to_unmap_anon or try_to_unmap_file. | |
763 | */ | |
a48d07af | 764 | static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
14fa31b8 | 765 | enum ttu_flags flags) |
1da177e4 LT |
766 | { |
767 | struct mm_struct *mm = vma->vm_mm; | |
768 | unsigned long address; | |
1da177e4 LT |
769 | pte_t *pte; |
770 | pte_t pteval; | |
c0718806 | 771 | spinlock_t *ptl; |
1da177e4 LT |
772 | int ret = SWAP_AGAIN; |
773 | ||
1da177e4 LT |
774 | address = vma_address(page, vma); |
775 | if (address == -EFAULT) | |
776 | goto out; | |
777 | ||
479db0bf | 778 | pte = page_check_address(page, mm, address, &ptl, 0); |
c0718806 | 779 | if (!pte) |
81b4082d | 780 | goto out; |
1da177e4 LT |
781 | |
782 | /* | |
783 | * If the page is mlock()d, we cannot swap it out. | |
784 | * If it's recently referenced (perhaps page_referenced | |
785 | * skipped over this mm) then we should reactivate it. | |
786 | */ | |
14fa31b8 | 787 | if (!(flags & TTU_IGNORE_MLOCK)) { |
b291f000 NP |
788 | if (vma->vm_flags & VM_LOCKED) { |
789 | ret = SWAP_MLOCK; | |
790 | goto out_unmap; | |
791 | } | |
14fa31b8 AK |
792 | } |
793 | if (!(flags & TTU_IGNORE_ACCESS)) { | |
b291f000 NP |
794 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
795 | ret = SWAP_FAIL; | |
796 | goto out_unmap; | |
797 | } | |
798 | } | |
1da177e4 | 799 | |
1da177e4 LT |
800 | /* Nuke the page table entry. */ |
801 | flush_cache_page(vma, address, page_to_pfn(page)); | |
cddb8a5c | 802 | pteval = ptep_clear_flush_notify(vma, address, pte); |
1da177e4 LT |
803 | |
804 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
805 | if (pte_dirty(pteval)) | |
806 | set_page_dirty(page); | |
807 | ||
365e9c87 HD |
808 | /* Update high watermark before we lower rss */ |
809 | update_hiwater_rss(mm); | |
810 | ||
888b9f7c AK |
811 | if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) { |
812 | if (PageAnon(page)) | |
813 | dec_mm_counter(mm, anon_rss); | |
814 | else | |
815 | dec_mm_counter(mm, file_rss); | |
816 | set_pte_at(mm, address, pte, | |
817 | swp_entry_to_pte(make_hwpoison_entry(page))); | |
818 | } else if (PageAnon(page)) { | |
4c21e2f2 | 819 | swp_entry_t entry = { .val = page_private(page) }; |
0697212a CL |
820 | |
821 | if (PageSwapCache(page)) { | |
822 | /* | |
823 | * Store the swap location in the pte. | |
824 | * See handle_pte_fault() ... | |
825 | */ | |
570a335b HD |
826 | if (swap_duplicate(entry) < 0) { |
827 | set_pte_at(mm, address, pte, pteval); | |
828 | ret = SWAP_FAIL; | |
829 | goto out_unmap; | |
830 | } | |
0697212a CL |
831 | if (list_empty(&mm->mmlist)) { |
832 | spin_lock(&mmlist_lock); | |
833 | if (list_empty(&mm->mmlist)) | |
834 | list_add(&mm->mmlist, &init_mm.mmlist); | |
835 | spin_unlock(&mmlist_lock); | |
836 | } | |
442c9137 | 837 | dec_mm_counter(mm, anon_rss); |
64cdd548 | 838 | } else if (PAGE_MIGRATION) { |
0697212a CL |
839 | /* |
840 | * Store the pfn of the page in a special migration | |
841 | * pte. do_swap_page() will wait until the migration | |
842 | * pte is removed and then restart fault handling. | |
843 | */ | |
14fa31b8 | 844 | BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION); |
0697212a | 845 | entry = make_migration_entry(page, pte_write(pteval)); |
1da177e4 LT |
846 | } |
847 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
848 | BUG_ON(pte_file(*pte)); | |
14fa31b8 | 849 | } else if (PAGE_MIGRATION && (TTU_ACTION(flags) == TTU_MIGRATION)) { |
04e62a29 CL |
850 | /* Establish migration entry for a file page */ |
851 | swp_entry_t entry; | |
852 | entry = make_migration_entry(page, pte_write(pteval)); | |
853 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
854 | } else | |
4294621f | 855 | dec_mm_counter(mm, file_rss); |
1da177e4 | 856 | |
04e62a29 | 857 | |
edc315fd | 858 | page_remove_rmap(page); |
1da177e4 LT |
859 | page_cache_release(page); |
860 | ||
861 | out_unmap: | |
c0718806 | 862 | pte_unmap_unlock(pte, ptl); |
1da177e4 LT |
863 | out: |
864 | return ret; | |
865 | } | |
866 | ||
867 | /* | |
868 | * objrmap doesn't work for nonlinear VMAs because the assumption that | |
869 | * offset-into-file correlates with offset-into-virtual-addresses does not hold. | |
870 | * Consequently, given a particular page and its ->index, we cannot locate the | |
871 | * ptes which are mapping that page without an exhaustive linear search. | |
872 | * | |
873 | * So what this code does is a mini "virtual scan" of each nonlinear VMA which | |
874 | * maps the file to which the target page belongs. The ->vm_private_data field | |
875 | * holds the current cursor into that scan. Successive searches will circulate | |
876 | * around the vma's virtual address space. | |
877 | * | |
878 | * So as more replacement pressure is applied to the pages in a nonlinear VMA, | |
879 | * more scanning pressure is placed against them as well. Eventually pages | |
880 | * will become fully unmapped and are eligible for eviction. | |
881 | * | |
882 | * For very sparsely populated VMAs this is a little inefficient - chances are | |
883 | * there there won't be many ptes located within the scan cluster. In this case | |
884 | * maybe we could scan further - to the end of the pte page, perhaps. | |
b291f000 NP |
885 | * |
886 | * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can | |
887 | * acquire it without blocking. If vma locked, mlock the pages in the cluster, | |
888 | * rather than unmapping them. If we encounter the "check_page" that vmscan is | |
889 | * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN. | |
1da177e4 LT |
890 | */ |
891 | #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE) | |
892 | #define CLUSTER_MASK (~(CLUSTER_SIZE - 1)) | |
893 | ||
b291f000 NP |
894 | static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount, |
895 | struct vm_area_struct *vma, struct page *check_page) | |
1da177e4 LT |
896 | { |
897 | struct mm_struct *mm = vma->vm_mm; | |
898 | pgd_t *pgd; | |
899 | pud_t *pud; | |
900 | pmd_t *pmd; | |
c0718806 | 901 | pte_t *pte; |
1da177e4 | 902 | pte_t pteval; |
c0718806 | 903 | spinlock_t *ptl; |
1da177e4 LT |
904 | struct page *page; |
905 | unsigned long address; | |
906 | unsigned long end; | |
b291f000 NP |
907 | int ret = SWAP_AGAIN; |
908 | int locked_vma = 0; | |
1da177e4 | 909 | |
1da177e4 LT |
910 | address = (vma->vm_start + cursor) & CLUSTER_MASK; |
911 | end = address + CLUSTER_SIZE; | |
912 | if (address < vma->vm_start) | |
913 | address = vma->vm_start; | |
914 | if (end > vma->vm_end) | |
915 | end = vma->vm_end; | |
916 | ||
917 | pgd = pgd_offset(mm, address); | |
918 | if (!pgd_present(*pgd)) | |
b291f000 | 919 | return ret; |
1da177e4 LT |
920 | |
921 | pud = pud_offset(pgd, address); | |
922 | if (!pud_present(*pud)) | |
b291f000 | 923 | return ret; |
1da177e4 LT |
924 | |
925 | pmd = pmd_offset(pud, address); | |
926 | if (!pmd_present(*pmd)) | |
b291f000 NP |
927 | return ret; |
928 | ||
929 | /* | |
930 | * MLOCK_PAGES => feature is configured. | |
931 | * if we can acquire the mmap_sem for read, and vma is VM_LOCKED, | |
932 | * keep the sem while scanning the cluster for mlocking pages. | |
933 | */ | |
934 | if (MLOCK_PAGES && down_read_trylock(&vma->vm_mm->mmap_sem)) { | |
935 | locked_vma = (vma->vm_flags & VM_LOCKED); | |
936 | if (!locked_vma) | |
937 | up_read(&vma->vm_mm->mmap_sem); /* don't need it */ | |
938 | } | |
c0718806 HD |
939 | |
940 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); | |
1da177e4 | 941 | |
365e9c87 HD |
942 | /* Update high watermark before we lower rss */ |
943 | update_hiwater_rss(mm); | |
944 | ||
c0718806 | 945 | for (; address < end; pte++, address += PAGE_SIZE) { |
1da177e4 LT |
946 | if (!pte_present(*pte)) |
947 | continue; | |
6aab341e LT |
948 | page = vm_normal_page(vma, address, *pte); |
949 | BUG_ON(!page || PageAnon(page)); | |
1da177e4 | 950 | |
b291f000 NP |
951 | if (locked_vma) { |
952 | mlock_vma_page(page); /* no-op if already mlocked */ | |
953 | if (page == check_page) | |
954 | ret = SWAP_MLOCK; | |
955 | continue; /* don't unmap */ | |
956 | } | |
957 | ||
cddb8a5c | 958 | if (ptep_clear_flush_young_notify(vma, address, pte)) |
1da177e4 LT |
959 | continue; |
960 | ||
961 | /* Nuke the page table entry. */ | |
eca35133 | 962 | flush_cache_page(vma, address, pte_pfn(*pte)); |
cddb8a5c | 963 | pteval = ptep_clear_flush_notify(vma, address, pte); |
1da177e4 LT |
964 | |
965 | /* If nonlinear, store the file page offset in the pte. */ | |
966 | if (page->index != linear_page_index(vma, address)) | |
967 | set_pte_at(mm, address, pte, pgoff_to_pte(page->index)); | |
968 | ||
969 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
970 | if (pte_dirty(pteval)) | |
971 | set_page_dirty(page); | |
972 | ||
edc315fd | 973 | page_remove_rmap(page); |
1da177e4 | 974 | page_cache_release(page); |
4294621f | 975 | dec_mm_counter(mm, file_rss); |
1da177e4 LT |
976 | (*mapcount)--; |
977 | } | |
c0718806 | 978 | pte_unmap_unlock(pte - 1, ptl); |
b291f000 NP |
979 | if (locked_vma) |
980 | up_read(&vma->vm_mm->mmap_sem); | |
981 | return ret; | |
1da177e4 LT |
982 | } |
983 | ||
b291f000 NP |
984 | /* |
985 | * common handling for pages mapped in VM_LOCKED vmas | |
986 | */ | |
987 | static int try_to_mlock_page(struct page *page, struct vm_area_struct *vma) | |
988 | { | |
989 | int mlocked = 0; | |
990 | ||
991 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { | |
992 | if (vma->vm_flags & VM_LOCKED) { | |
993 | mlock_vma_page(page); | |
994 | mlocked++; /* really mlocked the page */ | |
995 | } | |
996 | up_read(&vma->vm_mm->mmap_sem); | |
997 | } | |
998 | return mlocked; | |
999 | } | |
1000 | ||
1001 | /** | |
1002 | * try_to_unmap_anon - unmap or unlock anonymous page using the object-based | |
1003 | * rmap method | |
1004 | * @page: the page to unmap/unlock | |
8051be5e | 1005 | * @flags: action and flags |
b291f000 NP |
1006 | * |
1007 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1008 | * contained in the anon_vma struct it points to. | |
1009 | * | |
1010 | * This function is only called from try_to_unmap/try_to_munlock for | |
1011 | * anonymous pages. | |
1012 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1013 | * where the page was found will be held for write. So, we won't recheck | |
1014 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1015 | * 'LOCKED. | |
1016 | */ | |
14fa31b8 | 1017 | static int try_to_unmap_anon(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1018 | { |
1019 | struct anon_vma *anon_vma; | |
1020 | struct vm_area_struct *vma; | |
b291f000 | 1021 | unsigned int mlocked = 0; |
1da177e4 | 1022 | int ret = SWAP_AGAIN; |
14fa31b8 | 1023 | int unlock = TTU_ACTION(flags) == TTU_MUNLOCK; |
1da177e4 | 1024 | |
b291f000 NP |
1025 | if (MLOCK_PAGES && unlikely(unlock)) |
1026 | ret = SWAP_SUCCESS; /* default for try_to_munlock() */ | |
1027 | ||
1da177e4 LT |
1028 | anon_vma = page_lock_anon_vma(page); |
1029 | if (!anon_vma) | |
1030 | return ret; | |
1031 | ||
1032 | list_for_each_entry(vma, &anon_vma->head, anon_vma_node) { | |
b291f000 NP |
1033 | if (MLOCK_PAGES && unlikely(unlock)) { |
1034 | if (!((vma->vm_flags & VM_LOCKED) && | |
1035 | page_mapped_in_vma(page, vma))) | |
1036 | continue; /* must visit all unlocked vmas */ | |
1037 | ret = SWAP_MLOCK; /* saw at least one mlocked vma */ | |
1038 | } else { | |
14fa31b8 | 1039 | ret = try_to_unmap_one(page, vma, flags); |
b291f000 NP |
1040 | if (ret == SWAP_FAIL || !page_mapped(page)) |
1041 | break; | |
1042 | } | |
1043 | if (ret == SWAP_MLOCK) { | |
1044 | mlocked = try_to_mlock_page(page, vma); | |
1045 | if (mlocked) | |
1046 | break; /* stop if actually mlocked page */ | |
1047 | } | |
1da177e4 | 1048 | } |
34bbd704 ON |
1049 | |
1050 | page_unlock_anon_vma(anon_vma); | |
b291f000 NP |
1051 | |
1052 | if (mlocked) | |
1053 | ret = SWAP_MLOCK; /* actually mlocked the page */ | |
1054 | else if (ret == SWAP_MLOCK) | |
1055 | ret = SWAP_AGAIN; /* saw VM_LOCKED vma */ | |
1056 | ||
1da177e4 LT |
1057 | return ret; |
1058 | } | |
1059 | ||
1060 | /** | |
b291f000 NP |
1061 | * try_to_unmap_file - unmap/unlock file page using the object-based rmap method |
1062 | * @page: the page to unmap/unlock | |
14fa31b8 | 1063 | * @flags: action and flags |
1da177e4 LT |
1064 | * |
1065 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1066 | * contained in the address_space struct it points to. | |
1067 | * | |
b291f000 NP |
1068 | * This function is only called from try_to_unmap/try_to_munlock for |
1069 | * object-based pages. | |
1070 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1071 | * where the page was found will be held for write. So, we won't recheck | |
1072 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1073 | * 'LOCKED. | |
1da177e4 | 1074 | */ |
14fa31b8 | 1075 | static int try_to_unmap_file(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1076 | { |
1077 | struct address_space *mapping = page->mapping; | |
1078 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
1079 | struct vm_area_struct *vma; | |
1080 | struct prio_tree_iter iter; | |
1081 | int ret = SWAP_AGAIN; | |
1082 | unsigned long cursor; | |
1083 | unsigned long max_nl_cursor = 0; | |
1084 | unsigned long max_nl_size = 0; | |
1085 | unsigned int mapcount; | |
b291f000 | 1086 | unsigned int mlocked = 0; |
14fa31b8 | 1087 | int unlock = TTU_ACTION(flags) == TTU_MUNLOCK; |
b291f000 NP |
1088 | |
1089 | if (MLOCK_PAGES && unlikely(unlock)) | |
1090 | ret = SWAP_SUCCESS; /* default for try_to_munlock() */ | |
1da177e4 LT |
1091 | |
1092 | spin_lock(&mapping->i_mmap_lock); | |
1093 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
b291f000 | 1094 | if (MLOCK_PAGES && unlikely(unlock)) { |
508b9f8e MK |
1095 | if (!((vma->vm_flags & VM_LOCKED) && |
1096 | page_mapped_in_vma(page, vma))) | |
b291f000 NP |
1097 | continue; /* must visit all vmas */ |
1098 | ret = SWAP_MLOCK; | |
1099 | } else { | |
14fa31b8 | 1100 | ret = try_to_unmap_one(page, vma, flags); |
b291f000 NP |
1101 | if (ret == SWAP_FAIL || !page_mapped(page)) |
1102 | goto out; | |
1103 | } | |
1104 | if (ret == SWAP_MLOCK) { | |
1105 | mlocked = try_to_mlock_page(page, vma); | |
1106 | if (mlocked) | |
7b511594 | 1107 | goto out; /* stop if actually mlocked page */ |
b291f000 | 1108 | } |
1da177e4 LT |
1109 | } |
1110 | ||
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) { | |
b291f000 NP |
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 | } | |
14fa31b8 AK |
1122 | if (!MLOCK_PAGES && !(flags & TTU_IGNORE_MLOCK) && |
1123 | (vma->vm_flags & VM_LOCKED)) | |
1da177e4 LT |
1124 | continue; |
1125 | cursor = (unsigned long) vma->vm_private_data; | |
1126 | if (cursor > max_nl_cursor) | |
1127 | max_nl_cursor = cursor; | |
1128 | cursor = vma->vm_end - vma->vm_start; | |
1129 | if (cursor > max_nl_size) | |
1130 | max_nl_size = cursor; | |
1131 | } | |
1132 | ||
b291f000 | 1133 | if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */ |
1da177e4 LT |
1134 | ret = SWAP_FAIL; |
1135 | goto out; | |
1136 | } | |
1137 | ||
1138 | /* | |
1139 | * We don't try to search for this page in the nonlinear vmas, | |
1140 | * and page_referenced wouldn't have found it anyway. Instead | |
1141 | * just walk the nonlinear vmas trying to age and unmap some. | |
1142 | * The mapcount of the page we came in with is irrelevant, | |
1143 | * but even so use it as a guide to how hard we should try? | |
1144 | */ | |
1145 | mapcount = page_mapcount(page); | |
1146 | if (!mapcount) | |
1147 | goto out; | |
1148 | cond_resched_lock(&mapping->i_mmap_lock); | |
1149 | ||
1150 | max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK; | |
1151 | if (max_nl_cursor == 0) | |
1152 | max_nl_cursor = CLUSTER_SIZE; | |
1153 | ||
1154 | do { | |
1155 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, | |
1156 | shared.vm_set.list) { | |
14fa31b8 | 1157 | if (!MLOCK_PAGES && !(flags & TTU_IGNORE_MLOCK) && |
b291f000 | 1158 | (vma->vm_flags & VM_LOCKED)) |
1da177e4 LT |
1159 | continue; |
1160 | cursor = (unsigned long) vma->vm_private_data; | |
839b9685 | 1161 | while ( cursor < max_nl_cursor && |
1da177e4 | 1162 | cursor < vma->vm_end - vma->vm_start) { |
b291f000 NP |
1163 | ret = try_to_unmap_cluster(cursor, &mapcount, |
1164 | vma, page); | |
1165 | if (ret == SWAP_MLOCK) | |
1166 | mlocked = 2; /* to return below */ | |
1da177e4 LT |
1167 | cursor += CLUSTER_SIZE; |
1168 | vma->vm_private_data = (void *) cursor; | |
1169 | if ((int)mapcount <= 0) | |
1170 | goto out; | |
1171 | } | |
1172 | vma->vm_private_data = (void *) max_nl_cursor; | |
1173 | } | |
1174 | cond_resched_lock(&mapping->i_mmap_lock); | |
1175 | max_nl_cursor += CLUSTER_SIZE; | |
1176 | } while (max_nl_cursor <= max_nl_size); | |
1177 | ||
1178 | /* | |
1179 | * Don't loop forever (perhaps all the remaining pages are | |
1180 | * in locked vmas). Reset cursor on all unreserved nonlinear | |
1181 | * vmas, now forgetting on which ones it had fallen behind. | |
1182 | */ | |
101d2be7 HD |
1183 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) |
1184 | vma->vm_private_data = NULL; | |
1da177e4 LT |
1185 | out: |
1186 | spin_unlock(&mapping->i_mmap_lock); | |
b291f000 NP |
1187 | if (mlocked) |
1188 | ret = SWAP_MLOCK; /* actually mlocked the page */ | |
1189 | else if (ret == SWAP_MLOCK) | |
1190 | ret = SWAP_AGAIN; /* saw VM_LOCKED vma */ | |
1da177e4 LT |
1191 | return ret; |
1192 | } | |
1193 | ||
1194 | /** | |
1195 | * try_to_unmap - try to remove all page table mappings to a page | |
1196 | * @page: the page to get unmapped | |
14fa31b8 | 1197 | * @flags: action and flags |
1da177e4 LT |
1198 | * |
1199 | * Tries to remove all the page table entries which are mapping this | |
1200 | * page, used in the pageout path. Caller must hold the page lock. | |
1201 | * Return values are: | |
1202 | * | |
1203 | * SWAP_SUCCESS - we succeeded in removing all mappings | |
1204 | * SWAP_AGAIN - we missed a mapping, try again later | |
1205 | * SWAP_FAIL - the page is unswappable | |
b291f000 | 1206 | * SWAP_MLOCK - page is mlocked. |
1da177e4 | 1207 | */ |
14fa31b8 | 1208 | int try_to_unmap(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1209 | { |
1210 | int ret; | |
1211 | ||
1da177e4 LT |
1212 | BUG_ON(!PageLocked(page)); |
1213 | ||
1214 | if (PageAnon(page)) | |
14fa31b8 | 1215 | ret = try_to_unmap_anon(page, flags); |
1da177e4 | 1216 | else |
14fa31b8 | 1217 | ret = try_to_unmap_file(page, flags); |
b291f000 | 1218 | if (ret != SWAP_MLOCK && !page_mapped(page)) |
1da177e4 LT |
1219 | ret = SWAP_SUCCESS; |
1220 | return ret; | |
1221 | } | |
81b4082d | 1222 | |
b291f000 NP |
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)) | |
14fa31b8 | 1242 | return try_to_unmap_anon(page, TTU_MUNLOCK); |
b291f000 | 1243 | else |
14fa31b8 | 1244 | return try_to_unmap_file(page, TTU_MUNLOCK); |
b291f000 | 1245 | } |
68377659 | 1246 |