]>
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 | |
17 | * Contributions by Hugh Dickins <hugh@veritas.com> 2003, 2004 | |
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) | |
1da177e4 LT |
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> | |
a48d07af | 49 | #include <linux/module.h> |
7de6b805 | 50 | #include <linux/kallsyms.h> |
8a9f3ccd | 51 | #include <linux/memcontrol.h> |
cddb8a5c | 52 | #include <linux/mmu_notifier.h> |
1da177e4 LT |
53 | |
54 | #include <asm/tlbflush.h> | |
55 | ||
b291f000 NP |
56 | #include "internal.h" |
57 | ||
fcc234f8 | 58 | struct kmem_cache *anon_vma_cachep; |
1da177e4 | 59 | |
d9d332e0 LT |
60 | /** |
61 | * anon_vma_prepare - attach an anon_vma to a memory region | |
62 | * @vma: the memory region in question | |
63 | * | |
64 | * This makes sure the memory mapping described by 'vma' has | |
65 | * an 'anon_vma' attached to it, so that we can associate the | |
66 | * anonymous pages mapped into it with that anon_vma. | |
67 | * | |
68 | * The common case will be that we already have one, but if | |
69 | * if not we either need to find an adjacent mapping that we | |
70 | * can re-use the anon_vma from (very common when the only | |
71 | * reason for splitting a vma has been mprotect()), or we | |
72 | * allocate a new one. | |
73 | * | |
74 | * Anon-vma allocations are very subtle, because we may have | |
75 | * optimistically looked up an anon_vma in page_lock_anon_vma() | |
76 | * and that may actually touch the spinlock even in the newly | |
77 | * allocated vma (it depends on RCU to make sure that the | |
78 | * anon_vma isn't actually destroyed). | |
79 | * | |
80 | * As a result, we need to do proper anon_vma locking even | |
81 | * for the new allocation. At the same time, we do not want | |
82 | * to do any locking for the common case of already having | |
83 | * an anon_vma. | |
84 | * | |
85 | * This must be called with the mmap_sem held for reading. | |
86 | */ | |
1da177e4 LT |
87 | int anon_vma_prepare(struct vm_area_struct *vma) |
88 | { | |
89 | struct anon_vma *anon_vma = vma->anon_vma; | |
90 | ||
91 | might_sleep(); | |
92 | if (unlikely(!anon_vma)) { | |
93 | struct mm_struct *mm = vma->vm_mm; | |
d9d332e0 | 94 | struct anon_vma *allocated; |
1da177e4 LT |
95 | |
96 | anon_vma = find_mergeable_anon_vma(vma); | |
d9d332e0 LT |
97 | allocated = NULL; |
98 | if (!anon_vma) { | |
1da177e4 LT |
99 | anon_vma = anon_vma_alloc(); |
100 | if (unlikely(!anon_vma)) | |
101 | return -ENOMEM; | |
102 | allocated = anon_vma; | |
1da177e4 | 103 | } |
d9d332e0 | 104 | spin_lock(&anon_vma->lock); |
1da177e4 LT |
105 | |
106 | /* page_table_lock to protect against threads */ | |
107 | spin_lock(&mm->page_table_lock); | |
108 | if (likely(!vma->anon_vma)) { | |
109 | vma->anon_vma = anon_vma; | |
0697212a | 110 | list_add_tail(&vma->anon_vma_node, &anon_vma->head); |
1da177e4 LT |
111 | allocated = NULL; |
112 | } | |
113 | spin_unlock(&mm->page_table_lock); | |
114 | ||
d9d332e0 | 115 | spin_unlock(&anon_vma->lock); |
1da177e4 LT |
116 | if (unlikely(allocated)) |
117 | anon_vma_free(allocated); | |
118 | } | |
119 | return 0; | |
120 | } | |
121 | ||
122 | void __anon_vma_merge(struct vm_area_struct *vma, struct vm_area_struct *next) | |
123 | { | |
124 | BUG_ON(vma->anon_vma != next->anon_vma); | |
125 | list_del(&next->anon_vma_node); | |
126 | } | |
127 | ||
128 | void __anon_vma_link(struct vm_area_struct *vma) | |
129 | { | |
130 | struct anon_vma *anon_vma = vma->anon_vma; | |
131 | ||
30acbaba | 132 | if (anon_vma) |
0697212a | 133 | list_add_tail(&vma->anon_vma_node, &anon_vma->head); |
1da177e4 LT |
134 | } |
135 | ||
136 | void anon_vma_link(struct vm_area_struct *vma) | |
137 | { | |
138 | struct anon_vma *anon_vma = vma->anon_vma; | |
139 | ||
140 | if (anon_vma) { | |
141 | spin_lock(&anon_vma->lock); | |
0697212a | 142 | list_add_tail(&vma->anon_vma_node, &anon_vma->head); |
1da177e4 LT |
143 | spin_unlock(&anon_vma->lock); |
144 | } | |
145 | } | |
146 | ||
147 | void anon_vma_unlink(struct vm_area_struct *vma) | |
148 | { | |
149 | struct anon_vma *anon_vma = vma->anon_vma; | |
150 | int empty; | |
151 | ||
152 | if (!anon_vma) | |
153 | return; | |
154 | ||
155 | spin_lock(&anon_vma->lock); | |
1da177e4 LT |
156 | list_del(&vma->anon_vma_node); |
157 | ||
158 | /* We must garbage collect the anon_vma if it's empty */ | |
159 | empty = list_empty(&anon_vma->head); | |
160 | spin_unlock(&anon_vma->lock); | |
161 | ||
162 | if (empty) | |
163 | anon_vma_free(anon_vma); | |
164 | } | |
165 | ||
51cc5068 | 166 | static void anon_vma_ctor(void *data) |
1da177e4 | 167 | { |
a35afb83 | 168 | struct anon_vma *anon_vma = data; |
1da177e4 | 169 | |
a35afb83 CL |
170 | spin_lock_init(&anon_vma->lock); |
171 | INIT_LIST_HEAD(&anon_vma->head); | |
1da177e4 LT |
172 | } |
173 | ||
174 | void __init anon_vma_init(void) | |
175 | { | |
176 | anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), | |
20c2df83 | 177 | 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor); |
1da177e4 LT |
178 | } |
179 | ||
180 | /* | |
181 | * Getting a lock on a stable anon_vma from a page off the LRU is | |
182 | * tricky: page_lock_anon_vma rely on RCU to guard against the races. | |
183 | */ | |
af936a16 | 184 | struct anon_vma *page_lock_anon_vma(struct page *page) |
1da177e4 | 185 | { |
34bbd704 | 186 | struct anon_vma *anon_vma; |
1da177e4 LT |
187 | unsigned long anon_mapping; |
188 | ||
189 | rcu_read_lock(); | |
190 | anon_mapping = (unsigned long) page->mapping; | |
191 | if (!(anon_mapping & PAGE_MAPPING_ANON)) | |
192 | goto out; | |
193 | if (!page_mapped(page)) | |
194 | goto out; | |
195 | ||
196 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); | |
197 | spin_lock(&anon_vma->lock); | |
34bbd704 | 198 | return anon_vma; |
1da177e4 LT |
199 | out: |
200 | rcu_read_unlock(); | |
34bbd704 ON |
201 | return NULL; |
202 | } | |
203 | ||
af936a16 | 204 | void page_unlock_anon_vma(struct anon_vma *anon_vma) |
34bbd704 ON |
205 | { |
206 | spin_unlock(&anon_vma->lock); | |
207 | rcu_read_unlock(); | |
1da177e4 LT |
208 | } |
209 | ||
210 | /* | |
3ad33b24 LS |
211 | * At what user virtual address is page expected in @vma? |
212 | * Returns virtual address or -EFAULT if page's index/offset is not | |
213 | * within the range mapped the @vma. | |
1da177e4 LT |
214 | */ |
215 | static inline unsigned long | |
216 | vma_address(struct page *page, struct vm_area_struct *vma) | |
217 | { | |
218 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
219 | unsigned long address; | |
220 | ||
221 | address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); | |
222 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) { | |
3ad33b24 | 223 | /* page should be within @vma mapping range */ |
1da177e4 LT |
224 | return -EFAULT; |
225 | } | |
226 | return address; | |
227 | } | |
228 | ||
229 | /* | |
230 | * At what user virtual address is page expected in vma? checking that the | |
ee498ed7 | 231 | * page matches the vma: currently only used on anon pages, by unuse_vma; |
1da177e4 LT |
232 | */ |
233 | unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) | |
234 | { | |
235 | if (PageAnon(page)) { | |
236 | if ((void *)vma->anon_vma != | |
237 | (void *)page->mapping - PAGE_MAPPING_ANON) | |
238 | return -EFAULT; | |
239 | } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) { | |
ee498ed7 HD |
240 | if (!vma->vm_file || |
241 | vma->vm_file->f_mapping != page->mapping) | |
1da177e4 LT |
242 | return -EFAULT; |
243 | } else | |
244 | return -EFAULT; | |
245 | return vma_address(page, vma); | |
246 | } | |
247 | ||
81b4082d ND |
248 | /* |
249 | * Check that @page is mapped at @address into @mm. | |
250 | * | |
479db0bf NP |
251 | * If @sync is false, page_check_address may perform a racy check to avoid |
252 | * the page table lock when the pte is not present (helpful when reclaiming | |
253 | * highly shared pages). | |
254 | * | |
b8072f09 | 255 | * On success returns with pte mapped and locked. |
81b4082d | 256 | */ |
ceffc078 | 257 | pte_t *page_check_address(struct page *page, struct mm_struct *mm, |
479db0bf | 258 | unsigned long address, spinlock_t **ptlp, int sync) |
81b4082d ND |
259 | { |
260 | pgd_t *pgd; | |
261 | pud_t *pud; | |
262 | pmd_t *pmd; | |
263 | pte_t *pte; | |
c0718806 | 264 | spinlock_t *ptl; |
81b4082d | 265 | |
81b4082d | 266 | pgd = pgd_offset(mm, address); |
c0718806 HD |
267 | if (!pgd_present(*pgd)) |
268 | return NULL; | |
269 | ||
270 | pud = pud_offset(pgd, address); | |
271 | if (!pud_present(*pud)) | |
272 | return NULL; | |
273 | ||
274 | pmd = pmd_offset(pud, address); | |
275 | if (!pmd_present(*pmd)) | |
276 | return NULL; | |
277 | ||
278 | pte = pte_offset_map(pmd, address); | |
279 | /* Make a quick check before getting the lock */ | |
479db0bf | 280 | if (!sync && !pte_present(*pte)) { |
c0718806 HD |
281 | pte_unmap(pte); |
282 | return NULL; | |
283 | } | |
284 | ||
4c21e2f2 | 285 | ptl = pte_lockptr(mm, pmd); |
c0718806 HD |
286 | spin_lock(ptl); |
287 | if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) { | |
288 | *ptlp = ptl; | |
289 | return pte; | |
81b4082d | 290 | } |
c0718806 HD |
291 | pte_unmap_unlock(pte, ptl); |
292 | return NULL; | |
81b4082d ND |
293 | } |
294 | ||
b291f000 NP |
295 | /** |
296 | * page_mapped_in_vma - check whether a page is really mapped in a VMA | |
297 | * @page: the page to test | |
298 | * @vma: the VMA to test | |
299 | * | |
300 | * Returns 1 if the page is mapped into the page tables of the VMA, 0 | |
301 | * if the page is not mapped into the page tables of this VMA. Only | |
302 | * valid for normal file or anonymous VMAs. | |
303 | */ | |
304 | static int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma) | |
305 | { | |
306 | unsigned long address; | |
307 | pte_t *pte; | |
308 | spinlock_t *ptl; | |
309 | ||
310 | address = vma_address(page, vma); | |
311 | if (address == -EFAULT) /* out of vma range */ | |
312 | return 0; | |
313 | pte = page_check_address(page, vma->vm_mm, address, &ptl, 1); | |
314 | if (!pte) /* the page is not in this mm */ | |
315 | return 0; | |
316 | pte_unmap_unlock(pte, ptl); | |
317 | ||
318 | return 1; | |
319 | } | |
320 | ||
1da177e4 LT |
321 | /* |
322 | * Subfunctions of page_referenced: page_referenced_one called | |
323 | * repeatedly from either page_referenced_anon or page_referenced_file. | |
324 | */ | |
325 | static int page_referenced_one(struct page *page, | |
f7b7fd8f | 326 | struct vm_area_struct *vma, unsigned int *mapcount) |
1da177e4 LT |
327 | { |
328 | struct mm_struct *mm = vma->vm_mm; | |
329 | unsigned long address; | |
1da177e4 | 330 | pte_t *pte; |
c0718806 | 331 | spinlock_t *ptl; |
1da177e4 LT |
332 | int referenced = 0; |
333 | ||
1da177e4 LT |
334 | address = vma_address(page, vma); |
335 | if (address == -EFAULT) | |
336 | goto out; | |
337 | ||
479db0bf | 338 | pte = page_check_address(page, mm, address, &ptl, 0); |
c0718806 HD |
339 | if (!pte) |
340 | goto out; | |
1da177e4 | 341 | |
b291f000 NP |
342 | /* |
343 | * Don't want to elevate referenced for mlocked page that gets this far, | |
344 | * in order that it progresses to try_to_unmap and is moved to the | |
345 | * unevictable list. | |
346 | */ | |
5a9bbdcd | 347 | if (vma->vm_flags & VM_LOCKED) { |
5a9bbdcd | 348 | *mapcount = 1; /* break early from loop */ |
b291f000 NP |
349 | goto out_unmap; |
350 | } | |
351 | ||
352 | if (ptep_clear_flush_young_notify(vma, address, pte)) | |
c0718806 | 353 | referenced++; |
1da177e4 | 354 | |
c0718806 HD |
355 | /* Pretend the page is referenced if the task has the |
356 | swap token and is in the middle of a page fault. */ | |
f7b7fd8f | 357 | if (mm != current->mm && has_swap_token(mm) && |
c0718806 HD |
358 | rwsem_is_locked(&mm->mmap_sem)) |
359 | referenced++; | |
360 | ||
b291f000 | 361 | out_unmap: |
c0718806 HD |
362 | (*mapcount)--; |
363 | pte_unmap_unlock(pte, ptl); | |
1da177e4 LT |
364 | out: |
365 | return referenced; | |
366 | } | |
367 | ||
bed7161a BS |
368 | static int page_referenced_anon(struct page *page, |
369 | struct mem_cgroup *mem_cont) | |
1da177e4 LT |
370 | { |
371 | unsigned int mapcount; | |
372 | struct anon_vma *anon_vma; | |
373 | struct vm_area_struct *vma; | |
374 | int referenced = 0; | |
375 | ||
376 | anon_vma = page_lock_anon_vma(page); | |
377 | if (!anon_vma) | |
378 | return referenced; | |
379 | ||
380 | mapcount = page_mapcount(page); | |
381 | list_for_each_entry(vma, &anon_vma->head, anon_vma_node) { | |
bed7161a BS |
382 | /* |
383 | * If we are reclaiming on behalf of a cgroup, skip | |
384 | * counting on behalf of references from different | |
385 | * cgroups | |
386 | */ | |
bd845e38 | 387 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
bed7161a | 388 | continue; |
f7b7fd8f | 389 | referenced += page_referenced_one(page, vma, &mapcount); |
1da177e4 LT |
390 | if (!mapcount) |
391 | break; | |
392 | } | |
34bbd704 ON |
393 | |
394 | page_unlock_anon_vma(anon_vma); | |
1da177e4 LT |
395 | return referenced; |
396 | } | |
397 | ||
398 | /** | |
399 | * page_referenced_file - referenced check for object-based rmap | |
400 | * @page: the page we're checking references on. | |
43d8eac4 | 401 | * @mem_cont: target memory controller |
1da177e4 LT |
402 | * |
403 | * For an object-based mapped page, find all the places it is mapped and | |
404 | * check/clear the referenced flag. This is done by following the page->mapping | |
405 | * pointer, then walking the chain of vmas it holds. It returns the number | |
406 | * of references it found. | |
407 | * | |
408 | * This function is only called from page_referenced for object-based pages. | |
409 | */ | |
bed7161a BS |
410 | static int page_referenced_file(struct page *page, |
411 | struct mem_cgroup *mem_cont) | |
1da177e4 LT |
412 | { |
413 | unsigned int mapcount; | |
414 | struct address_space *mapping = page->mapping; | |
415 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
416 | struct vm_area_struct *vma; | |
417 | struct prio_tree_iter iter; | |
418 | int referenced = 0; | |
419 | ||
420 | /* | |
421 | * The caller's checks on page->mapping and !PageAnon have made | |
422 | * sure that this is a file page: the check for page->mapping | |
423 | * excludes the case just before it gets set on an anon page. | |
424 | */ | |
425 | BUG_ON(PageAnon(page)); | |
426 | ||
427 | /* | |
428 | * The page lock not only makes sure that page->mapping cannot | |
429 | * suddenly be NULLified by truncation, it makes sure that the | |
430 | * structure at mapping cannot be freed and reused yet, | |
431 | * so we can safely take mapping->i_mmap_lock. | |
432 | */ | |
433 | BUG_ON(!PageLocked(page)); | |
434 | ||
435 | spin_lock(&mapping->i_mmap_lock); | |
436 | ||
437 | /* | |
438 | * i_mmap_lock does not stabilize mapcount at all, but mapcount | |
439 | * is more likely to be accurate if we note it after spinning. | |
440 | */ | |
441 | mapcount = page_mapcount(page); | |
442 | ||
443 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
bed7161a BS |
444 | /* |
445 | * If we are reclaiming on behalf of a cgroup, skip | |
446 | * counting on behalf of references from different | |
447 | * cgroups | |
448 | */ | |
bd845e38 | 449 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
bed7161a | 450 | continue; |
f7b7fd8f | 451 | referenced += page_referenced_one(page, vma, &mapcount); |
1da177e4 LT |
452 | if (!mapcount) |
453 | break; | |
454 | } | |
455 | ||
456 | spin_unlock(&mapping->i_mmap_lock); | |
457 | return referenced; | |
458 | } | |
459 | ||
460 | /** | |
461 | * page_referenced - test if the page was referenced | |
462 | * @page: the page to test | |
463 | * @is_locked: caller holds lock on the page | |
43d8eac4 | 464 | * @mem_cont: target memory controller |
1da177e4 LT |
465 | * |
466 | * Quick test_and_clear_referenced for all mappings to a page, | |
467 | * returns the number of ptes which referenced the page. | |
468 | */ | |
bed7161a BS |
469 | int page_referenced(struct page *page, int is_locked, |
470 | struct mem_cgroup *mem_cont) | |
1da177e4 LT |
471 | { |
472 | int referenced = 0; | |
473 | ||
1da177e4 LT |
474 | if (TestClearPageReferenced(page)) |
475 | referenced++; | |
476 | ||
477 | if (page_mapped(page) && page->mapping) { | |
478 | if (PageAnon(page)) | |
bed7161a | 479 | referenced += page_referenced_anon(page, mem_cont); |
1da177e4 | 480 | else if (is_locked) |
bed7161a | 481 | referenced += page_referenced_file(page, mem_cont); |
529ae9aa | 482 | else if (!trylock_page(page)) |
1da177e4 LT |
483 | referenced++; |
484 | else { | |
485 | if (page->mapping) | |
bed7161a BS |
486 | referenced += |
487 | page_referenced_file(page, mem_cont); | |
1da177e4 LT |
488 | unlock_page(page); |
489 | } | |
490 | } | |
5b7baf05 CB |
491 | |
492 | if (page_test_and_clear_young(page)) | |
493 | referenced++; | |
494 | ||
1da177e4 LT |
495 | return referenced; |
496 | } | |
497 | ||
d08b3851 PZ |
498 | static int page_mkclean_one(struct page *page, struct vm_area_struct *vma) |
499 | { | |
500 | struct mm_struct *mm = vma->vm_mm; | |
501 | unsigned long address; | |
c2fda5fe | 502 | pte_t *pte; |
d08b3851 PZ |
503 | spinlock_t *ptl; |
504 | int ret = 0; | |
505 | ||
506 | address = vma_address(page, vma); | |
507 | if (address == -EFAULT) | |
508 | goto out; | |
509 | ||
479db0bf | 510 | pte = page_check_address(page, mm, address, &ptl, 1); |
d08b3851 PZ |
511 | if (!pte) |
512 | goto out; | |
513 | ||
c2fda5fe PZ |
514 | if (pte_dirty(*pte) || pte_write(*pte)) { |
515 | pte_t entry; | |
d08b3851 | 516 | |
c2fda5fe | 517 | flush_cache_page(vma, address, pte_pfn(*pte)); |
cddb8a5c | 518 | entry = ptep_clear_flush_notify(vma, address, pte); |
c2fda5fe PZ |
519 | entry = pte_wrprotect(entry); |
520 | entry = pte_mkclean(entry); | |
d6e88e67 | 521 | set_pte_at(mm, address, pte, entry); |
c2fda5fe PZ |
522 | ret = 1; |
523 | } | |
d08b3851 | 524 | |
d08b3851 PZ |
525 | pte_unmap_unlock(pte, ptl); |
526 | out: | |
527 | return ret; | |
528 | } | |
529 | ||
530 | static int page_mkclean_file(struct address_space *mapping, struct page *page) | |
531 | { | |
532 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
533 | struct vm_area_struct *vma; | |
534 | struct prio_tree_iter iter; | |
535 | int ret = 0; | |
536 | ||
537 | BUG_ON(PageAnon(page)); | |
538 | ||
539 | spin_lock(&mapping->i_mmap_lock); | |
540 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
541 | if (vma->vm_flags & VM_SHARED) | |
542 | ret += page_mkclean_one(page, vma); | |
543 | } | |
544 | spin_unlock(&mapping->i_mmap_lock); | |
545 | return ret; | |
546 | } | |
547 | ||
548 | int page_mkclean(struct page *page) | |
549 | { | |
550 | int ret = 0; | |
551 | ||
552 | BUG_ON(!PageLocked(page)); | |
553 | ||
554 | if (page_mapped(page)) { | |
555 | struct address_space *mapping = page_mapping(page); | |
ce7e9fae | 556 | if (mapping) { |
d08b3851 | 557 | ret = page_mkclean_file(mapping, page); |
ce7e9fae CB |
558 | if (page_test_dirty(page)) { |
559 | page_clear_dirty(page); | |
560 | ret = 1; | |
561 | } | |
6c210482 | 562 | } |
d08b3851 PZ |
563 | } |
564 | ||
565 | return ret; | |
566 | } | |
60b59bea | 567 | EXPORT_SYMBOL_GPL(page_mkclean); |
d08b3851 | 568 | |
9617d95e | 569 | /** |
43d8eac4 | 570 | * __page_set_anon_rmap - setup new anonymous rmap |
9617d95e NP |
571 | * @page: the page to add the mapping to |
572 | * @vma: the vm area in which the mapping is added | |
573 | * @address: the user virtual address mapped | |
574 | */ | |
575 | static void __page_set_anon_rmap(struct page *page, | |
576 | struct vm_area_struct *vma, unsigned long address) | |
577 | { | |
578 | struct anon_vma *anon_vma = vma->anon_vma; | |
579 | ||
580 | BUG_ON(!anon_vma); | |
581 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | |
582 | page->mapping = (struct address_space *) anon_vma; | |
583 | ||
584 | page->index = linear_page_index(vma, address); | |
585 | ||
a74609fa NP |
586 | /* |
587 | * nr_mapped state can be updated without turning off | |
588 | * interrupts because it is not modified via interrupt. | |
589 | */ | |
f3dbd344 | 590 | __inc_zone_page_state(page, NR_ANON_PAGES); |
9617d95e NP |
591 | } |
592 | ||
c97a9e10 | 593 | /** |
43d8eac4 | 594 | * __page_check_anon_rmap - sanity check anonymous rmap addition |
c97a9e10 NP |
595 | * @page: the page to add the mapping to |
596 | * @vma: the vm area in which the mapping is added | |
597 | * @address: the user virtual address mapped | |
598 | */ | |
599 | static void __page_check_anon_rmap(struct page *page, | |
600 | struct vm_area_struct *vma, unsigned long address) | |
601 | { | |
602 | #ifdef CONFIG_DEBUG_VM | |
603 | /* | |
604 | * The page's anon-rmap details (mapping and index) are guaranteed to | |
605 | * be set up correctly at this point. | |
606 | * | |
607 | * We have exclusion against page_add_anon_rmap because the caller | |
608 | * always holds the page locked, except if called from page_dup_rmap, | |
609 | * in which case the page is already known to be setup. | |
610 | * | |
611 | * We have exclusion against page_add_new_anon_rmap because those pages | |
612 | * are initially only visible via the pagetables, and the pte is locked | |
613 | * over the call to page_add_new_anon_rmap. | |
614 | */ | |
615 | struct anon_vma *anon_vma = vma->anon_vma; | |
616 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | |
617 | BUG_ON(page->mapping != (struct address_space *)anon_vma); | |
618 | BUG_ON(page->index != linear_page_index(vma, address)); | |
619 | #endif | |
620 | } | |
621 | ||
1da177e4 LT |
622 | /** |
623 | * page_add_anon_rmap - add pte mapping to an anonymous page | |
624 | * @page: the page to add the mapping to | |
625 | * @vma: the vm area in which the mapping is added | |
626 | * @address: the user virtual address mapped | |
627 | * | |
c97a9e10 | 628 | * The caller needs to hold the pte lock and the page must be locked. |
1da177e4 LT |
629 | */ |
630 | void page_add_anon_rmap(struct page *page, | |
631 | struct vm_area_struct *vma, unsigned long address) | |
632 | { | |
c97a9e10 NP |
633 | VM_BUG_ON(!PageLocked(page)); |
634 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
9617d95e NP |
635 | if (atomic_inc_and_test(&page->_mapcount)) |
636 | __page_set_anon_rmap(page, vma, address); | |
69029cd5 | 637 | else |
c97a9e10 | 638 | __page_check_anon_rmap(page, vma, address); |
1da177e4 LT |
639 | } |
640 | ||
43d8eac4 | 641 | /** |
9617d95e NP |
642 | * page_add_new_anon_rmap - add pte mapping to a new anonymous page |
643 | * @page: the page to add the mapping to | |
644 | * @vma: the vm area in which the mapping is added | |
645 | * @address: the user virtual address mapped | |
646 | * | |
647 | * Same as page_add_anon_rmap but must only be called on *new* pages. | |
648 | * This means the inc-and-test can be bypassed. | |
c97a9e10 | 649 | * Page does not have to be locked. |
9617d95e NP |
650 | */ |
651 | void page_add_new_anon_rmap(struct page *page, | |
652 | struct vm_area_struct *vma, unsigned long address) | |
653 | { | |
c97a9e10 | 654 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
9617d95e NP |
655 | atomic_set(&page->_mapcount, 0); /* elevate count by 1 (starts at -1) */ |
656 | __page_set_anon_rmap(page, vma, address); | |
657 | } | |
658 | ||
1da177e4 LT |
659 | /** |
660 | * page_add_file_rmap - add pte mapping to a file page | |
661 | * @page: the page to add the mapping to | |
662 | * | |
b8072f09 | 663 | * The caller needs to hold the pte lock. |
1da177e4 LT |
664 | */ |
665 | void page_add_file_rmap(struct page *page) | |
666 | { | |
1da177e4 | 667 | if (atomic_inc_and_test(&page->_mapcount)) |
65ba55f5 | 668 | __inc_zone_page_state(page, NR_FILE_MAPPED); |
1da177e4 LT |
669 | } |
670 | ||
c97a9e10 NP |
671 | #ifdef CONFIG_DEBUG_VM |
672 | /** | |
673 | * page_dup_rmap - duplicate pte mapping to a page | |
674 | * @page: the page to add the mapping to | |
43d8eac4 RD |
675 | * @vma: the vm area being duplicated |
676 | * @address: the user virtual address mapped | |
c97a9e10 NP |
677 | * |
678 | * For copy_page_range only: minimal extract from page_add_file_rmap / | |
679 | * page_add_anon_rmap, avoiding unnecessary tests (already checked) so it's | |
680 | * quicker. | |
681 | * | |
682 | * The caller needs to hold the pte lock. | |
683 | */ | |
684 | void page_dup_rmap(struct page *page, struct vm_area_struct *vma, unsigned long address) | |
685 | { | |
686 | BUG_ON(page_mapcount(page) == 0); | |
687 | if (PageAnon(page)) | |
688 | __page_check_anon_rmap(page, vma, address); | |
689 | atomic_inc(&page->_mapcount); | |
690 | } | |
691 | #endif | |
692 | ||
1da177e4 LT |
693 | /** |
694 | * page_remove_rmap - take down pte mapping from a page | |
695 | * @page: page to remove mapping from | |
43d8eac4 | 696 | * @vma: the vm area in which the mapping is removed |
1da177e4 | 697 | * |
b8072f09 | 698 | * The caller needs to hold the pte lock. |
1da177e4 | 699 | */ |
7de6b805 | 700 | void page_remove_rmap(struct page *page, struct vm_area_struct *vma) |
1da177e4 | 701 | { |
1da177e4 | 702 | if (atomic_add_negative(-1, &page->_mapcount)) { |
b7ab795b | 703 | if (unlikely(page_mapcount(page) < 0)) { |
ef2bf0dc | 704 | printk (KERN_EMERG "Eeek! page_mapcount(page) went negative! (%d)\n", page_mapcount(page)); |
7de6b805 | 705 | printk (KERN_EMERG " page pfn = %lx\n", page_to_pfn(page)); |
ef2bf0dc DJ |
706 | printk (KERN_EMERG " page->flags = %lx\n", page->flags); |
707 | printk (KERN_EMERG " page->count = %x\n", page_count(page)); | |
708 | printk (KERN_EMERG " page->mapping = %p\n", page->mapping); | |
7de6b805 | 709 | print_symbol (KERN_EMERG " vma->vm_ops = %s\n", (unsigned long)vma->vm_ops); |
54cb8821 | 710 | if (vma->vm_ops) { |
54cb8821 NP |
711 | print_symbol (KERN_EMERG " vma->vm_ops->fault = %s\n", (unsigned long)vma->vm_ops->fault); |
712 | } | |
7de6b805 NP |
713 | if (vma->vm_file && vma->vm_file->f_op) |
714 | print_symbol (KERN_EMERG " vma->vm_file->f_op->mmap = %s\n", (unsigned long)vma->vm_file->f_op->mmap); | |
b16bc64d | 715 | BUG(); |
ef2bf0dc | 716 | } |
b16bc64d | 717 | |
1da177e4 | 718 | /* |
16f8c5b2 HD |
719 | * Now that the last pte has gone, s390 must transfer dirty |
720 | * flag from storage key to struct page. We can usually skip | |
721 | * this if the page is anon, so about to be freed; but perhaps | |
722 | * not if it's in swapcache - there might be another pte slot | |
723 | * containing the swap entry, but page not yet written to swap. | |
1da177e4 | 724 | */ |
a4b526b3 MS |
725 | if ((!PageAnon(page) || PageSwapCache(page)) && |
726 | page_test_dirty(page)) { | |
6c210482 | 727 | page_clear_dirty(page); |
1da177e4 | 728 | set_page_dirty(page); |
6c210482 | 729 | } |
5b4e655e KH |
730 | if (PageAnon(page)) |
731 | mem_cgroup_uncharge_page(page); | |
f3dbd344 | 732 | __dec_zone_page_state(page, |
16f8c5b2 HD |
733 | PageAnon(page) ? NR_ANON_PAGES : NR_FILE_MAPPED); |
734 | /* | |
735 | * It would be tidy to reset the PageAnon mapping here, | |
736 | * but that might overwrite a racing page_add_anon_rmap | |
737 | * which increments mapcount after us but sets mapping | |
738 | * before us: so leave the reset to free_hot_cold_page, | |
739 | * and remember that it's only reliable while mapped. | |
740 | * Leaving it set also helps swapoff to reinstate ptes | |
741 | * faster for those pages still in swapcache. | |
742 | */ | |
1da177e4 LT |
743 | } |
744 | } | |
745 | ||
746 | /* | |
747 | * Subfunctions of try_to_unmap: try_to_unmap_one called | |
748 | * repeatedly from either try_to_unmap_anon or try_to_unmap_file. | |
749 | */ | |
a48d07af | 750 | static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
7352349a | 751 | int migration) |
1da177e4 LT |
752 | { |
753 | struct mm_struct *mm = vma->vm_mm; | |
754 | unsigned long address; | |
1da177e4 LT |
755 | pte_t *pte; |
756 | pte_t pteval; | |
c0718806 | 757 | spinlock_t *ptl; |
1da177e4 LT |
758 | int ret = SWAP_AGAIN; |
759 | ||
1da177e4 LT |
760 | address = vma_address(page, vma); |
761 | if (address == -EFAULT) | |
762 | goto out; | |
763 | ||
479db0bf | 764 | pte = page_check_address(page, mm, address, &ptl, 0); |
c0718806 | 765 | if (!pte) |
81b4082d | 766 | goto out; |
1da177e4 LT |
767 | |
768 | /* | |
769 | * If the page is mlock()d, we cannot swap it out. | |
770 | * If it's recently referenced (perhaps page_referenced | |
771 | * skipped over this mm) then we should reactivate it. | |
772 | */ | |
b291f000 NP |
773 | if (!migration) { |
774 | if (vma->vm_flags & VM_LOCKED) { | |
775 | ret = SWAP_MLOCK; | |
776 | goto out_unmap; | |
777 | } | |
778 | if (ptep_clear_flush_young_notify(vma, address, pte)) { | |
779 | ret = SWAP_FAIL; | |
780 | goto out_unmap; | |
781 | } | |
782 | } | |
1da177e4 | 783 | |
1da177e4 LT |
784 | /* Nuke the page table entry. */ |
785 | flush_cache_page(vma, address, page_to_pfn(page)); | |
cddb8a5c | 786 | pteval = ptep_clear_flush_notify(vma, address, pte); |
1da177e4 LT |
787 | |
788 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
789 | if (pte_dirty(pteval)) | |
790 | set_page_dirty(page); | |
791 | ||
365e9c87 HD |
792 | /* Update high watermark before we lower rss */ |
793 | update_hiwater_rss(mm); | |
794 | ||
1da177e4 | 795 | if (PageAnon(page)) { |
4c21e2f2 | 796 | swp_entry_t entry = { .val = page_private(page) }; |
0697212a CL |
797 | |
798 | if (PageSwapCache(page)) { | |
799 | /* | |
800 | * Store the swap location in the pte. | |
801 | * See handle_pte_fault() ... | |
802 | */ | |
803 | swap_duplicate(entry); | |
804 | if (list_empty(&mm->mmlist)) { | |
805 | spin_lock(&mmlist_lock); | |
806 | if (list_empty(&mm->mmlist)) | |
807 | list_add(&mm->mmlist, &init_mm.mmlist); | |
808 | spin_unlock(&mmlist_lock); | |
809 | } | |
442c9137 | 810 | dec_mm_counter(mm, anon_rss); |
04e62a29 | 811 | #ifdef CONFIG_MIGRATION |
0697212a CL |
812 | } else { |
813 | /* | |
814 | * Store the pfn of the page in a special migration | |
815 | * pte. do_swap_page() will wait until the migration | |
816 | * pte is removed and then restart fault handling. | |
817 | */ | |
818 | BUG_ON(!migration); | |
819 | entry = make_migration_entry(page, pte_write(pteval)); | |
04e62a29 | 820 | #endif |
1da177e4 LT |
821 | } |
822 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
823 | BUG_ON(pte_file(*pte)); | |
4294621f | 824 | } else |
04e62a29 CL |
825 | #ifdef CONFIG_MIGRATION |
826 | if (migration) { | |
827 | /* Establish migration entry for a file page */ | |
828 | swp_entry_t entry; | |
829 | entry = make_migration_entry(page, pte_write(pteval)); | |
830 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
831 | } else | |
832 | #endif | |
4294621f | 833 | dec_mm_counter(mm, file_rss); |
1da177e4 | 834 | |
04e62a29 | 835 | |
7de6b805 | 836 | page_remove_rmap(page, vma); |
1da177e4 LT |
837 | page_cache_release(page); |
838 | ||
839 | out_unmap: | |
c0718806 | 840 | pte_unmap_unlock(pte, ptl); |
1da177e4 LT |
841 | out: |
842 | return ret; | |
843 | } | |
844 | ||
845 | /* | |
846 | * objrmap doesn't work for nonlinear VMAs because the assumption that | |
847 | * offset-into-file correlates with offset-into-virtual-addresses does not hold. | |
848 | * Consequently, given a particular page and its ->index, we cannot locate the | |
849 | * ptes which are mapping that page without an exhaustive linear search. | |
850 | * | |
851 | * So what this code does is a mini "virtual scan" of each nonlinear VMA which | |
852 | * maps the file to which the target page belongs. The ->vm_private_data field | |
853 | * holds the current cursor into that scan. Successive searches will circulate | |
854 | * around the vma's virtual address space. | |
855 | * | |
856 | * So as more replacement pressure is applied to the pages in a nonlinear VMA, | |
857 | * more scanning pressure is placed against them as well. Eventually pages | |
858 | * will become fully unmapped and are eligible for eviction. | |
859 | * | |
860 | * For very sparsely populated VMAs this is a little inefficient - chances are | |
861 | * there there won't be many ptes located within the scan cluster. In this case | |
862 | * maybe we could scan further - to the end of the pte page, perhaps. | |
b291f000 NP |
863 | * |
864 | * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can | |
865 | * acquire it without blocking. If vma locked, mlock the pages in the cluster, | |
866 | * rather than unmapping them. If we encounter the "check_page" that vmscan is | |
867 | * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN. | |
1da177e4 LT |
868 | */ |
869 | #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE) | |
870 | #define CLUSTER_MASK (~(CLUSTER_SIZE - 1)) | |
871 | ||
b291f000 NP |
872 | static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount, |
873 | struct vm_area_struct *vma, struct page *check_page) | |
1da177e4 LT |
874 | { |
875 | struct mm_struct *mm = vma->vm_mm; | |
876 | pgd_t *pgd; | |
877 | pud_t *pud; | |
878 | pmd_t *pmd; | |
c0718806 | 879 | pte_t *pte; |
1da177e4 | 880 | pte_t pteval; |
c0718806 | 881 | spinlock_t *ptl; |
1da177e4 LT |
882 | struct page *page; |
883 | unsigned long address; | |
884 | unsigned long end; | |
b291f000 NP |
885 | int ret = SWAP_AGAIN; |
886 | int locked_vma = 0; | |
1da177e4 | 887 | |
1da177e4 LT |
888 | address = (vma->vm_start + cursor) & CLUSTER_MASK; |
889 | end = address + CLUSTER_SIZE; | |
890 | if (address < vma->vm_start) | |
891 | address = vma->vm_start; | |
892 | if (end > vma->vm_end) | |
893 | end = vma->vm_end; | |
894 | ||
895 | pgd = pgd_offset(mm, address); | |
896 | if (!pgd_present(*pgd)) | |
b291f000 | 897 | return ret; |
1da177e4 LT |
898 | |
899 | pud = pud_offset(pgd, address); | |
900 | if (!pud_present(*pud)) | |
b291f000 | 901 | return ret; |
1da177e4 LT |
902 | |
903 | pmd = pmd_offset(pud, address); | |
904 | if (!pmd_present(*pmd)) | |
b291f000 NP |
905 | return ret; |
906 | ||
907 | /* | |
908 | * MLOCK_PAGES => feature is configured. | |
909 | * if we can acquire the mmap_sem for read, and vma is VM_LOCKED, | |
910 | * keep the sem while scanning the cluster for mlocking pages. | |
911 | */ | |
912 | if (MLOCK_PAGES && down_read_trylock(&vma->vm_mm->mmap_sem)) { | |
913 | locked_vma = (vma->vm_flags & VM_LOCKED); | |
914 | if (!locked_vma) | |
915 | up_read(&vma->vm_mm->mmap_sem); /* don't need it */ | |
916 | } | |
c0718806 HD |
917 | |
918 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); | |
1da177e4 | 919 | |
365e9c87 HD |
920 | /* Update high watermark before we lower rss */ |
921 | update_hiwater_rss(mm); | |
922 | ||
c0718806 | 923 | for (; address < end; pte++, address += PAGE_SIZE) { |
1da177e4 LT |
924 | if (!pte_present(*pte)) |
925 | continue; | |
6aab341e LT |
926 | page = vm_normal_page(vma, address, *pte); |
927 | BUG_ON(!page || PageAnon(page)); | |
1da177e4 | 928 | |
b291f000 NP |
929 | if (locked_vma) { |
930 | mlock_vma_page(page); /* no-op if already mlocked */ | |
931 | if (page == check_page) | |
932 | ret = SWAP_MLOCK; | |
933 | continue; /* don't unmap */ | |
934 | } | |
935 | ||
cddb8a5c | 936 | if (ptep_clear_flush_young_notify(vma, address, pte)) |
1da177e4 LT |
937 | continue; |
938 | ||
939 | /* Nuke the page table entry. */ | |
eca35133 | 940 | flush_cache_page(vma, address, pte_pfn(*pte)); |
cddb8a5c | 941 | pteval = ptep_clear_flush_notify(vma, address, pte); |
1da177e4 LT |
942 | |
943 | /* If nonlinear, store the file page offset in the pte. */ | |
944 | if (page->index != linear_page_index(vma, address)) | |
945 | set_pte_at(mm, address, pte, pgoff_to_pte(page->index)); | |
946 | ||
947 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
948 | if (pte_dirty(pteval)) | |
949 | set_page_dirty(page); | |
950 | ||
7de6b805 | 951 | page_remove_rmap(page, vma); |
1da177e4 | 952 | page_cache_release(page); |
4294621f | 953 | dec_mm_counter(mm, file_rss); |
1da177e4 LT |
954 | (*mapcount)--; |
955 | } | |
c0718806 | 956 | pte_unmap_unlock(pte - 1, ptl); |
b291f000 NP |
957 | if (locked_vma) |
958 | up_read(&vma->vm_mm->mmap_sem); | |
959 | return ret; | |
1da177e4 LT |
960 | } |
961 | ||
b291f000 NP |
962 | /* |
963 | * common handling for pages mapped in VM_LOCKED vmas | |
964 | */ | |
965 | static int try_to_mlock_page(struct page *page, struct vm_area_struct *vma) | |
966 | { | |
967 | int mlocked = 0; | |
968 | ||
969 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { | |
970 | if (vma->vm_flags & VM_LOCKED) { | |
971 | mlock_vma_page(page); | |
972 | mlocked++; /* really mlocked the page */ | |
973 | } | |
974 | up_read(&vma->vm_mm->mmap_sem); | |
975 | } | |
976 | return mlocked; | |
977 | } | |
978 | ||
979 | /** | |
980 | * try_to_unmap_anon - unmap or unlock anonymous page using the object-based | |
981 | * rmap method | |
982 | * @page: the page to unmap/unlock | |
983 | * @unlock: request for unlock rather than unmap [unlikely] | |
984 | * @migration: unmapping for migration - ignored if @unlock | |
985 | * | |
986 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
987 | * contained in the anon_vma struct it points to. | |
988 | * | |
989 | * This function is only called from try_to_unmap/try_to_munlock for | |
990 | * anonymous pages. | |
991 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
992 | * where the page was found will be held for write. So, we won't recheck | |
993 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
994 | * 'LOCKED. | |
995 | */ | |
996 | static int try_to_unmap_anon(struct page *page, int unlock, int migration) | |
1da177e4 LT |
997 | { |
998 | struct anon_vma *anon_vma; | |
999 | struct vm_area_struct *vma; | |
b291f000 | 1000 | unsigned int mlocked = 0; |
1da177e4 LT |
1001 | int ret = SWAP_AGAIN; |
1002 | ||
b291f000 NP |
1003 | if (MLOCK_PAGES && unlikely(unlock)) |
1004 | ret = SWAP_SUCCESS; /* default for try_to_munlock() */ | |
1005 | ||
1da177e4 LT |
1006 | anon_vma = page_lock_anon_vma(page); |
1007 | if (!anon_vma) | |
1008 | return ret; | |
1009 | ||
1010 | list_for_each_entry(vma, &anon_vma->head, anon_vma_node) { | |
b291f000 NP |
1011 | if (MLOCK_PAGES && unlikely(unlock)) { |
1012 | if (!((vma->vm_flags & VM_LOCKED) && | |
1013 | page_mapped_in_vma(page, vma))) | |
1014 | continue; /* must visit all unlocked vmas */ | |
1015 | ret = SWAP_MLOCK; /* saw at least one mlocked vma */ | |
1016 | } else { | |
1017 | ret = try_to_unmap_one(page, vma, migration); | |
1018 | if (ret == SWAP_FAIL || !page_mapped(page)) | |
1019 | break; | |
1020 | } | |
1021 | if (ret == SWAP_MLOCK) { | |
1022 | mlocked = try_to_mlock_page(page, vma); | |
1023 | if (mlocked) | |
1024 | break; /* stop if actually mlocked page */ | |
1025 | } | |
1da177e4 | 1026 | } |
34bbd704 ON |
1027 | |
1028 | page_unlock_anon_vma(anon_vma); | |
b291f000 NP |
1029 | |
1030 | if (mlocked) | |
1031 | ret = SWAP_MLOCK; /* actually mlocked the page */ | |
1032 | else if (ret == SWAP_MLOCK) | |
1033 | ret = SWAP_AGAIN; /* saw VM_LOCKED vma */ | |
1034 | ||
1da177e4 LT |
1035 | return ret; |
1036 | } | |
1037 | ||
1038 | /** | |
b291f000 NP |
1039 | * try_to_unmap_file - unmap/unlock file page using the object-based rmap method |
1040 | * @page: the page to unmap/unlock | |
1041 | * @unlock: request for unlock rather than unmap [unlikely] | |
1042 | * @migration: unmapping for migration - ignored if @unlock | |
1da177e4 LT |
1043 | * |
1044 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1045 | * contained in the address_space struct it points to. | |
1046 | * | |
b291f000 NP |
1047 | * This function is only called from try_to_unmap/try_to_munlock for |
1048 | * object-based pages. | |
1049 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1050 | * where the page was found will be held for write. So, we won't recheck | |
1051 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1052 | * 'LOCKED. | |
1da177e4 | 1053 | */ |
b291f000 | 1054 | static int try_to_unmap_file(struct page *page, int unlock, int migration) |
1da177e4 LT |
1055 | { |
1056 | struct address_space *mapping = page->mapping; | |
1057 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
1058 | struct vm_area_struct *vma; | |
1059 | struct prio_tree_iter iter; | |
1060 | int ret = SWAP_AGAIN; | |
1061 | unsigned long cursor; | |
1062 | unsigned long max_nl_cursor = 0; | |
1063 | unsigned long max_nl_size = 0; | |
1064 | unsigned int mapcount; | |
b291f000 NP |
1065 | unsigned int mlocked = 0; |
1066 | ||
1067 | if (MLOCK_PAGES && unlikely(unlock)) | |
1068 | ret = SWAP_SUCCESS; /* default for try_to_munlock() */ | |
1da177e4 LT |
1069 | |
1070 | spin_lock(&mapping->i_mmap_lock); | |
1071 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
b291f000 NP |
1072 | if (MLOCK_PAGES && unlikely(unlock)) { |
1073 | if (!(vma->vm_flags & VM_LOCKED)) | |
1074 | continue; /* must visit all vmas */ | |
1075 | ret = SWAP_MLOCK; | |
1076 | } else { | |
1077 | ret = try_to_unmap_one(page, vma, migration); | |
1078 | if (ret == SWAP_FAIL || !page_mapped(page)) | |
1079 | goto out; | |
1080 | } | |
1081 | if (ret == SWAP_MLOCK) { | |
1082 | mlocked = try_to_mlock_page(page, vma); | |
1083 | if (mlocked) | |
1084 | break; /* stop if actually mlocked page */ | |
1085 | } | |
1da177e4 LT |
1086 | } |
1087 | ||
b291f000 NP |
1088 | if (mlocked) |
1089 | goto out; | |
1090 | ||
1da177e4 LT |
1091 | if (list_empty(&mapping->i_mmap_nonlinear)) |
1092 | goto out; | |
1093 | ||
1094 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, | |
1095 | shared.vm_set.list) { | |
b291f000 NP |
1096 | if (MLOCK_PAGES && unlikely(unlock)) { |
1097 | if (!(vma->vm_flags & VM_LOCKED)) | |
1098 | continue; /* must visit all vmas */ | |
1099 | ret = SWAP_MLOCK; /* leave mlocked == 0 */ | |
1100 | goto out; /* no need to look further */ | |
1101 | } | |
1102 | if (!MLOCK_PAGES && !migration && (vma->vm_flags & VM_LOCKED)) | |
1da177e4 LT |
1103 | continue; |
1104 | cursor = (unsigned long) vma->vm_private_data; | |
1105 | if (cursor > max_nl_cursor) | |
1106 | max_nl_cursor = cursor; | |
1107 | cursor = vma->vm_end - vma->vm_start; | |
1108 | if (cursor > max_nl_size) | |
1109 | max_nl_size = cursor; | |
1110 | } | |
1111 | ||
b291f000 | 1112 | if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */ |
1da177e4 LT |
1113 | ret = SWAP_FAIL; |
1114 | goto out; | |
1115 | } | |
1116 | ||
1117 | /* | |
1118 | * We don't try to search for this page in the nonlinear vmas, | |
1119 | * and page_referenced wouldn't have found it anyway. Instead | |
1120 | * just walk the nonlinear vmas trying to age and unmap some. | |
1121 | * The mapcount of the page we came in with is irrelevant, | |
1122 | * but even so use it as a guide to how hard we should try? | |
1123 | */ | |
1124 | mapcount = page_mapcount(page); | |
1125 | if (!mapcount) | |
1126 | goto out; | |
1127 | cond_resched_lock(&mapping->i_mmap_lock); | |
1128 | ||
1129 | max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK; | |
1130 | if (max_nl_cursor == 0) | |
1131 | max_nl_cursor = CLUSTER_SIZE; | |
1132 | ||
1133 | do { | |
1134 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, | |
1135 | shared.vm_set.list) { | |
b291f000 NP |
1136 | if (!MLOCK_PAGES && !migration && |
1137 | (vma->vm_flags & VM_LOCKED)) | |
1da177e4 LT |
1138 | continue; |
1139 | cursor = (unsigned long) vma->vm_private_data; | |
839b9685 | 1140 | while ( cursor < max_nl_cursor && |
1da177e4 | 1141 | cursor < vma->vm_end - vma->vm_start) { |
b291f000 NP |
1142 | ret = try_to_unmap_cluster(cursor, &mapcount, |
1143 | vma, page); | |
1144 | if (ret == SWAP_MLOCK) | |
1145 | mlocked = 2; /* to return below */ | |
1da177e4 LT |
1146 | cursor += CLUSTER_SIZE; |
1147 | vma->vm_private_data = (void *) cursor; | |
1148 | if ((int)mapcount <= 0) | |
1149 | goto out; | |
1150 | } | |
1151 | vma->vm_private_data = (void *) max_nl_cursor; | |
1152 | } | |
1153 | cond_resched_lock(&mapping->i_mmap_lock); | |
1154 | max_nl_cursor += CLUSTER_SIZE; | |
1155 | } while (max_nl_cursor <= max_nl_size); | |
1156 | ||
1157 | /* | |
1158 | * Don't loop forever (perhaps all the remaining pages are | |
1159 | * in locked vmas). Reset cursor on all unreserved nonlinear | |
1160 | * vmas, now forgetting on which ones it had fallen behind. | |
1161 | */ | |
101d2be7 HD |
1162 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) |
1163 | vma->vm_private_data = NULL; | |
1da177e4 LT |
1164 | out: |
1165 | spin_unlock(&mapping->i_mmap_lock); | |
b291f000 NP |
1166 | if (mlocked) |
1167 | ret = SWAP_MLOCK; /* actually mlocked the page */ | |
1168 | else if (ret == SWAP_MLOCK) | |
1169 | ret = SWAP_AGAIN; /* saw VM_LOCKED vma */ | |
1da177e4 LT |
1170 | return ret; |
1171 | } | |
1172 | ||
1173 | /** | |
1174 | * try_to_unmap - try to remove all page table mappings to a page | |
1175 | * @page: the page to get unmapped | |
43d8eac4 | 1176 | * @migration: migration flag |
1da177e4 LT |
1177 | * |
1178 | * Tries to remove all the page table entries which are mapping this | |
1179 | * page, used in the pageout path. Caller must hold the page lock. | |
1180 | * Return values are: | |
1181 | * | |
1182 | * SWAP_SUCCESS - we succeeded in removing all mappings | |
1183 | * SWAP_AGAIN - we missed a mapping, try again later | |
1184 | * SWAP_FAIL - the page is unswappable | |
b291f000 | 1185 | * SWAP_MLOCK - page is mlocked. |
1da177e4 | 1186 | */ |
7352349a | 1187 | int try_to_unmap(struct page *page, int migration) |
1da177e4 LT |
1188 | { |
1189 | int ret; | |
1190 | ||
1da177e4 LT |
1191 | BUG_ON(!PageLocked(page)); |
1192 | ||
1193 | if (PageAnon(page)) | |
b291f000 | 1194 | ret = try_to_unmap_anon(page, 0, migration); |
1da177e4 | 1195 | else |
b291f000 NP |
1196 | ret = try_to_unmap_file(page, 0, migration); |
1197 | if (ret != SWAP_MLOCK && !page_mapped(page)) | |
1da177e4 LT |
1198 | ret = SWAP_SUCCESS; |
1199 | return ret; | |
1200 | } | |
81b4082d | 1201 | |
b291f000 NP |
1202 | #ifdef CONFIG_UNEVICTABLE_LRU |
1203 | /** | |
1204 | * try_to_munlock - try to munlock a page | |
1205 | * @page: the page to be munlocked | |
1206 | * | |
1207 | * Called from munlock code. Checks all of the VMAs mapping the page | |
1208 | * to make sure nobody else has this page mlocked. The page will be | |
1209 | * returned with PG_mlocked cleared if no other vmas have it mlocked. | |
1210 | * | |
1211 | * Return values are: | |
1212 | * | |
1213 | * SWAP_SUCCESS - no vma's holding page mlocked. | |
1214 | * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem | |
1215 | * SWAP_MLOCK - page is now mlocked. | |
1216 | */ | |
1217 | int try_to_munlock(struct page *page) | |
1218 | { | |
1219 | VM_BUG_ON(!PageLocked(page) || PageLRU(page)); | |
1220 | ||
1221 | if (PageAnon(page)) | |
1222 | return try_to_unmap_anon(page, 1, 0); | |
1223 | else | |
1224 | return try_to_unmap_file(page, 1, 0); | |
1225 | } | |
1226 | #endif |