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> |
64cdd548 | 53 | #include <linux/migrate.h> |
1da177e4 LT |
54 | |
55 | #include <asm/tlbflush.h> | |
56 | ||
b291f000 NP |
57 | #include "internal.h" |
58 | ||
fdd2e5f8 AB |
59 | static struct kmem_cache *anon_vma_cachep; |
60 | ||
61 | static inline struct anon_vma *anon_vma_alloc(void) | |
62 | { | |
63 | return kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL); | |
64 | } | |
65 | ||
66 | static inline void anon_vma_free(struct anon_vma *anon_vma) | |
67 | { | |
68 | kmem_cache_free(anon_vma_cachep, anon_vma); | |
69 | } | |
1da177e4 | 70 | |
d9d332e0 LT |
71 | /** |
72 | * anon_vma_prepare - attach an anon_vma to a memory region | |
73 | * @vma: the memory region in question | |
74 | * | |
75 | * This makes sure the memory mapping described by 'vma' has | |
76 | * an 'anon_vma' attached to it, so that we can associate the | |
77 | * anonymous pages mapped into it with that anon_vma. | |
78 | * | |
79 | * The common case will be that we already have one, but if | |
80 | * if not we either need to find an adjacent mapping that we | |
81 | * can re-use the anon_vma from (very common when the only | |
82 | * reason for splitting a vma has been mprotect()), or we | |
83 | * allocate a new one. | |
84 | * | |
85 | * Anon-vma allocations are very subtle, because we may have | |
86 | * optimistically looked up an anon_vma in page_lock_anon_vma() | |
87 | * and that may actually touch the spinlock even in the newly | |
88 | * allocated vma (it depends on RCU to make sure that the | |
89 | * anon_vma isn't actually destroyed). | |
90 | * | |
91 | * As a result, we need to do proper anon_vma locking even | |
92 | * for the new allocation. At the same time, we do not want | |
93 | * to do any locking for the common case of already having | |
94 | * an anon_vma. | |
95 | * | |
96 | * This must be called with the mmap_sem held for reading. | |
97 | */ | |
1da177e4 LT |
98 | int anon_vma_prepare(struct vm_area_struct *vma) |
99 | { | |
100 | struct anon_vma *anon_vma = vma->anon_vma; | |
101 | ||
102 | might_sleep(); | |
103 | if (unlikely(!anon_vma)) { | |
104 | struct mm_struct *mm = vma->vm_mm; | |
d9d332e0 | 105 | struct anon_vma *allocated; |
1da177e4 LT |
106 | |
107 | anon_vma = find_mergeable_anon_vma(vma); | |
d9d332e0 LT |
108 | allocated = NULL; |
109 | if (!anon_vma) { | |
1da177e4 LT |
110 | anon_vma = anon_vma_alloc(); |
111 | if (unlikely(!anon_vma)) | |
112 | return -ENOMEM; | |
113 | allocated = anon_vma; | |
1da177e4 | 114 | } |
d9d332e0 | 115 | spin_lock(&anon_vma->lock); |
1da177e4 LT |
116 | |
117 | /* page_table_lock to protect against threads */ | |
118 | spin_lock(&mm->page_table_lock); | |
119 | if (likely(!vma->anon_vma)) { | |
120 | vma->anon_vma = anon_vma; | |
0697212a | 121 | list_add_tail(&vma->anon_vma_node, &anon_vma->head); |
1da177e4 LT |
122 | allocated = NULL; |
123 | } | |
124 | spin_unlock(&mm->page_table_lock); | |
125 | ||
d9d332e0 | 126 | spin_unlock(&anon_vma->lock); |
1da177e4 LT |
127 | if (unlikely(allocated)) |
128 | anon_vma_free(allocated); | |
129 | } | |
130 | return 0; | |
131 | } | |
132 | ||
133 | void __anon_vma_merge(struct vm_area_struct *vma, struct vm_area_struct *next) | |
134 | { | |
135 | BUG_ON(vma->anon_vma != next->anon_vma); | |
136 | list_del(&next->anon_vma_node); | |
137 | } | |
138 | ||
139 | void __anon_vma_link(struct vm_area_struct *vma) | |
140 | { | |
141 | struct anon_vma *anon_vma = vma->anon_vma; | |
142 | ||
30acbaba | 143 | if (anon_vma) |
0697212a | 144 | list_add_tail(&vma->anon_vma_node, &anon_vma->head); |
1da177e4 LT |
145 | } |
146 | ||
147 | void anon_vma_link(struct vm_area_struct *vma) | |
148 | { | |
149 | struct anon_vma *anon_vma = vma->anon_vma; | |
150 | ||
151 | if (anon_vma) { | |
152 | spin_lock(&anon_vma->lock); | |
0697212a | 153 | list_add_tail(&vma->anon_vma_node, &anon_vma->head); |
1da177e4 LT |
154 | spin_unlock(&anon_vma->lock); |
155 | } | |
156 | } | |
157 | ||
158 | void anon_vma_unlink(struct vm_area_struct *vma) | |
159 | { | |
160 | struct anon_vma *anon_vma = vma->anon_vma; | |
161 | int empty; | |
162 | ||
163 | if (!anon_vma) | |
164 | return; | |
165 | ||
166 | spin_lock(&anon_vma->lock); | |
1da177e4 LT |
167 | list_del(&vma->anon_vma_node); |
168 | ||
169 | /* We must garbage collect the anon_vma if it's empty */ | |
170 | empty = list_empty(&anon_vma->head); | |
171 | spin_unlock(&anon_vma->lock); | |
172 | ||
173 | if (empty) | |
174 | anon_vma_free(anon_vma); | |
175 | } | |
176 | ||
51cc5068 | 177 | static void anon_vma_ctor(void *data) |
1da177e4 | 178 | { |
a35afb83 | 179 | struct anon_vma *anon_vma = data; |
1da177e4 | 180 | |
a35afb83 CL |
181 | spin_lock_init(&anon_vma->lock); |
182 | INIT_LIST_HEAD(&anon_vma->head); | |
1da177e4 LT |
183 | } |
184 | ||
185 | void __init anon_vma_init(void) | |
186 | { | |
187 | anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), | |
20c2df83 | 188 | 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor); |
1da177e4 LT |
189 | } |
190 | ||
191 | /* | |
192 | * Getting a lock on a stable anon_vma from a page off the LRU is | |
193 | * tricky: page_lock_anon_vma rely on RCU to guard against the races. | |
194 | */ | |
af936a16 | 195 | struct anon_vma *page_lock_anon_vma(struct page *page) |
1da177e4 | 196 | { |
34bbd704 | 197 | struct anon_vma *anon_vma; |
1da177e4 LT |
198 | unsigned long anon_mapping; |
199 | ||
200 | rcu_read_lock(); | |
201 | anon_mapping = (unsigned long) page->mapping; | |
202 | if (!(anon_mapping & PAGE_MAPPING_ANON)) | |
203 | goto out; | |
204 | if (!page_mapped(page)) | |
205 | goto out; | |
206 | ||
207 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); | |
208 | spin_lock(&anon_vma->lock); | |
34bbd704 | 209 | return anon_vma; |
1da177e4 LT |
210 | out: |
211 | rcu_read_unlock(); | |
34bbd704 ON |
212 | return NULL; |
213 | } | |
214 | ||
af936a16 | 215 | void page_unlock_anon_vma(struct anon_vma *anon_vma) |
34bbd704 ON |
216 | { |
217 | spin_unlock(&anon_vma->lock); | |
218 | rcu_read_unlock(); | |
1da177e4 LT |
219 | } |
220 | ||
221 | /* | |
3ad33b24 LS |
222 | * At what user virtual address is page expected in @vma? |
223 | * Returns virtual address or -EFAULT if page's index/offset is not | |
224 | * within the range mapped the @vma. | |
1da177e4 LT |
225 | */ |
226 | static inline unsigned long | |
227 | vma_address(struct page *page, struct vm_area_struct *vma) | |
228 | { | |
229 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
230 | unsigned long address; | |
231 | ||
232 | address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); | |
233 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) { | |
3ad33b24 | 234 | /* page should be within @vma mapping range */ |
1da177e4 LT |
235 | return -EFAULT; |
236 | } | |
237 | return address; | |
238 | } | |
239 | ||
240 | /* | |
241 | * At what user virtual address is page expected in vma? checking that the | |
ee498ed7 | 242 | * page matches the vma: currently only used on anon pages, by unuse_vma; |
1da177e4 LT |
243 | */ |
244 | unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) | |
245 | { | |
246 | if (PageAnon(page)) { | |
247 | if ((void *)vma->anon_vma != | |
248 | (void *)page->mapping - PAGE_MAPPING_ANON) | |
249 | return -EFAULT; | |
250 | } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) { | |
ee498ed7 HD |
251 | if (!vma->vm_file || |
252 | vma->vm_file->f_mapping != page->mapping) | |
1da177e4 LT |
253 | return -EFAULT; |
254 | } else | |
255 | return -EFAULT; | |
256 | return vma_address(page, vma); | |
257 | } | |
258 | ||
81b4082d ND |
259 | /* |
260 | * Check that @page is mapped at @address into @mm. | |
261 | * | |
479db0bf NP |
262 | * If @sync is false, page_check_address may perform a racy check to avoid |
263 | * the page table lock when the pte is not present (helpful when reclaiming | |
264 | * highly shared pages). | |
265 | * | |
b8072f09 | 266 | * On success returns with pte mapped and locked. |
81b4082d | 267 | */ |
ceffc078 | 268 | pte_t *page_check_address(struct page *page, struct mm_struct *mm, |
479db0bf | 269 | unsigned long address, spinlock_t **ptlp, int sync) |
81b4082d ND |
270 | { |
271 | pgd_t *pgd; | |
272 | pud_t *pud; | |
273 | pmd_t *pmd; | |
274 | pte_t *pte; | |
c0718806 | 275 | spinlock_t *ptl; |
81b4082d | 276 | |
81b4082d | 277 | pgd = pgd_offset(mm, address); |
c0718806 HD |
278 | if (!pgd_present(*pgd)) |
279 | return NULL; | |
280 | ||
281 | pud = pud_offset(pgd, address); | |
282 | if (!pud_present(*pud)) | |
283 | return NULL; | |
284 | ||
285 | pmd = pmd_offset(pud, address); | |
286 | if (!pmd_present(*pmd)) | |
287 | return NULL; | |
288 | ||
289 | pte = pte_offset_map(pmd, address); | |
290 | /* Make a quick check before getting the lock */ | |
479db0bf | 291 | if (!sync && !pte_present(*pte)) { |
c0718806 HD |
292 | pte_unmap(pte); |
293 | return NULL; | |
294 | } | |
295 | ||
4c21e2f2 | 296 | ptl = pte_lockptr(mm, pmd); |
c0718806 HD |
297 | spin_lock(ptl); |
298 | if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) { | |
299 | *ptlp = ptl; | |
300 | return pte; | |
81b4082d | 301 | } |
c0718806 HD |
302 | pte_unmap_unlock(pte, ptl); |
303 | return NULL; | |
81b4082d ND |
304 | } |
305 | ||
b291f000 NP |
306 | /** |
307 | * page_mapped_in_vma - check whether a page is really mapped in a VMA | |
308 | * @page: the page to test | |
309 | * @vma: the VMA to test | |
310 | * | |
311 | * Returns 1 if the page is mapped into the page tables of the VMA, 0 | |
312 | * if the page is not mapped into the page tables of this VMA. Only | |
313 | * valid for normal file or anonymous VMAs. | |
314 | */ | |
315 | static int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma) | |
316 | { | |
317 | unsigned long address; | |
318 | pte_t *pte; | |
319 | spinlock_t *ptl; | |
320 | ||
321 | address = vma_address(page, vma); | |
322 | if (address == -EFAULT) /* out of vma range */ | |
323 | return 0; | |
324 | pte = page_check_address(page, vma->vm_mm, address, &ptl, 1); | |
325 | if (!pte) /* the page is not in this mm */ | |
326 | return 0; | |
327 | pte_unmap_unlock(pte, ptl); | |
328 | ||
329 | return 1; | |
330 | } | |
331 | ||
1da177e4 LT |
332 | /* |
333 | * Subfunctions of page_referenced: page_referenced_one called | |
334 | * repeatedly from either page_referenced_anon or page_referenced_file. | |
335 | */ | |
336 | static int page_referenced_one(struct page *page, | |
f7b7fd8f | 337 | struct vm_area_struct *vma, unsigned int *mapcount) |
1da177e4 LT |
338 | { |
339 | struct mm_struct *mm = vma->vm_mm; | |
340 | unsigned long address; | |
1da177e4 | 341 | pte_t *pte; |
c0718806 | 342 | spinlock_t *ptl; |
1da177e4 LT |
343 | int referenced = 0; |
344 | ||
1da177e4 LT |
345 | address = vma_address(page, vma); |
346 | if (address == -EFAULT) | |
347 | goto out; | |
348 | ||
479db0bf | 349 | pte = page_check_address(page, mm, address, &ptl, 0); |
c0718806 HD |
350 | if (!pte) |
351 | goto out; | |
1da177e4 | 352 | |
b291f000 NP |
353 | /* |
354 | * Don't want to elevate referenced for mlocked page that gets this far, | |
355 | * in order that it progresses to try_to_unmap and is moved to the | |
356 | * unevictable list. | |
357 | */ | |
5a9bbdcd | 358 | if (vma->vm_flags & VM_LOCKED) { |
5a9bbdcd | 359 | *mapcount = 1; /* break early from loop */ |
b291f000 NP |
360 | goto out_unmap; |
361 | } | |
362 | ||
4917e5d0 JW |
363 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
364 | /* | |
365 | * Don't treat a reference through a sequentially read | |
366 | * mapping as such. If the page has been used in | |
367 | * another mapping, we will catch it; if this other | |
368 | * mapping is already gone, the unmap path will have | |
369 | * set PG_referenced or activated the page. | |
370 | */ | |
371 | if (likely(!VM_SequentialReadHint(vma))) | |
372 | referenced++; | |
373 | } | |
1da177e4 | 374 | |
c0718806 HD |
375 | /* Pretend the page is referenced if the task has the |
376 | swap token and is in the middle of a page fault. */ | |
f7b7fd8f | 377 | if (mm != current->mm && has_swap_token(mm) && |
c0718806 HD |
378 | rwsem_is_locked(&mm->mmap_sem)) |
379 | referenced++; | |
380 | ||
b291f000 | 381 | out_unmap: |
c0718806 HD |
382 | (*mapcount)--; |
383 | pte_unmap_unlock(pte, ptl); | |
1da177e4 LT |
384 | out: |
385 | return referenced; | |
386 | } | |
387 | ||
bed7161a BS |
388 | static int page_referenced_anon(struct page *page, |
389 | struct mem_cgroup *mem_cont) | |
1da177e4 LT |
390 | { |
391 | unsigned int mapcount; | |
392 | struct anon_vma *anon_vma; | |
393 | struct vm_area_struct *vma; | |
394 | int referenced = 0; | |
395 | ||
396 | anon_vma = page_lock_anon_vma(page); | |
397 | if (!anon_vma) | |
398 | return referenced; | |
399 | ||
400 | mapcount = page_mapcount(page); | |
401 | list_for_each_entry(vma, &anon_vma->head, anon_vma_node) { | |
bed7161a BS |
402 | /* |
403 | * If we are reclaiming on behalf of a cgroup, skip | |
404 | * counting on behalf of references from different | |
405 | * cgroups | |
406 | */ | |
bd845e38 | 407 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
bed7161a | 408 | continue; |
f7b7fd8f | 409 | referenced += page_referenced_one(page, vma, &mapcount); |
1da177e4 LT |
410 | if (!mapcount) |
411 | break; | |
412 | } | |
34bbd704 ON |
413 | |
414 | page_unlock_anon_vma(anon_vma); | |
1da177e4 LT |
415 | return referenced; |
416 | } | |
417 | ||
418 | /** | |
419 | * page_referenced_file - referenced check for object-based rmap | |
420 | * @page: the page we're checking references on. | |
43d8eac4 | 421 | * @mem_cont: target memory controller |
1da177e4 LT |
422 | * |
423 | * For an object-based mapped page, find all the places it is mapped and | |
424 | * check/clear the referenced flag. This is done by following the page->mapping | |
425 | * pointer, then walking the chain of vmas it holds. It returns the number | |
426 | * of references it found. | |
427 | * | |
428 | * This function is only called from page_referenced for object-based pages. | |
429 | */ | |
bed7161a BS |
430 | static int page_referenced_file(struct page *page, |
431 | struct mem_cgroup *mem_cont) | |
1da177e4 LT |
432 | { |
433 | unsigned int mapcount; | |
434 | struct address_space *mapping = page->mapping; | |
435 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
436 | struct vm_area_struct *vma; | |
437 | struct prio_tree_iter iter; | |
438 | int referenced = 0; | |
439 | ||
440 | /* | |
441 | * The caller's checks on page->mapping and !PageAnon have made | |
442 | * sure that this is a file page: the check for page->mapping | |
443 | * excludes the case just before it gets set on an anon page. | |
444 | */ | |
445 | BUG_ON(PageAnon(page)); | |
446 | ||
447 | /* | |
448 | * The page lock not only makes sure that page->mapping cannot | |
449 | * suddenly be NULLified by truncation, it makes sure that the | |
450 | * structure at mapping cannot be freed and reused yet, | |
451 | * so we can safely take mapping->i_mmap_lock. | |
452 | */ | |
453 | BUG_ON(!PageLocked(page)); | |
454 | ||
455 | spin_lock(&mapping->i_mmap_lock); | |
456 | ||
457 | /* | |
458 | * i_mmap_lock does not stabilize mapcount at all, but mapcount | |
459 | * is more likely to be accurate if we note it after spinning. | |
460 | */ | |
461 | mapcount = page_mapcount(page); | |
462 | ||
463 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
bed7161a BS |
464 | /* |
465 | * If we are reclaiming on behalf of a cgroup, skip | |
466 | * counting on behalf of references from different | |
467 | * cgroups | |
468 | */ | |
bd845e38 | 469 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
bed7161a | 470 | continue; |
f7b7fd8f | 471 | referenced += page_referenced_one(page, vma, &mapcount); |
1da177e4 LT |
472 | if (!mapcount) |
473 | break; | |
474 | } | |
475 | ||
476 | spin_unlock(&mapping->i_mmap_lock); | |
477 | return referenced; | |
478 | } | |
479 | ||
480 | /** | |
481 | * page_referenced - test if the page was referenced | |
482 | * @page: the page to test | |
483 | * @is_locked: caller holds lock on the page | |
43d8eac4 | 484 | * @mem_cont: target memory controller |
1da177e4 LT |
485 | * |
486 | * Quick test_and_clear_referenced for all mappings to a page, | |
487 | * returns the number of ptes which referenced the page. | |
488 | */ | |
bed7161a BS |
489 | int page_referenced(struct page *page, int is_locked, |
490 | struct mem_cgroup *mem_cont) | |
1da177e4 LT |
491 | { |
492 | int referenced = 0; | |
493 | ||
1da177e4 LT |
494 | if (TestClearPageReferenced(page)) |
495 | referenced++; | |
496 | ||
497 | if (page_mapped(page) && page->mapping) { | |
498 | if (PageAnon(page)) | |
bed7161a | 499 | referenced += page_referenced_anon(page, mem_cont); |
1da177e4 | 500 | else if (is_locked) |
bed7161a | 501 | referenced += page_referenced_file(page, mem_cont); |
529ae9aa | 502 | else if (!trylock_page(page)) |
1da177e4 LT |
503 | referenced++; |
504 | else { | |
505 | if (page->mapping) | |
bed7161a BS |
506 | referenced += |
507 | page_referenced_file(page, mem_cont); | |
1da177e4 LT |
508 | unlock_page(page); |
509 | } | |
510 | } | |
5b7baf05 CB |
511 | |
512 | if (page_test_and_clear_young(page)) | |
513 | referenced++; | |
514 | ||
1da177e4 LT |
515 | return referenced; |
516 | } | |
517 | ||
d08b3851 PZ |
518 | static int page_mkclean_one(struct page *page, struct vm_area_struct *vma) |
519 | { | |
520 | struct mm_struct *mm = vma->vm_mm; | |
521 | unsigned long address; | |
c2fda5fe | 522 | pte_t *pte; |
d08b3851 PZ |
523 | spinlock_t *ptl; |
524 | int ret = 0; | |
525 | ||
526 | address = vma_address(page, vma); | |
527 | if (address == -EFAULT) | |
528 | goto out; | |
529 | ||
479db0bf | 530 | pte = page_check_address(page, mm, address, &ptl, 1); |
d08b3851 PZ |
531 | if (!pte) |
532 | goto out; | |
533 | ||
c2fda5fe PZ |
534 | if (pte_dirty(*pte) || pte_write(*pte)) { |
535 | pte_t entry; | |
d08b3851 | 536 | |
c2fda5fe | 537 | flush_cache_page(vma, address, pte_pfn(*pte)); |
cddb8a5c | 538 | entry = ptep_clear_flush_notify(vma, address, pte); |
c2fda5fe PZ |
539 | entry = pte_wrprotect(entry); |
540 | entry = pte_mkclean(entry); | |
d6e88e67 | 541 | set_pte_at(mm, address, pte, entry); |
c2fda5fe PZ |
542 | ret = 1; |
543 | } | |
d08b3851 | 544 | |
d08b3851 PZ |
545 | pte_unmap_unlock(pte, ptl); |
546 | out: | |
547 | return ret; | |
548 | } | |
549 | ||
550 | static int page_mkclean_file(struct address_space *mapping, struct page *page) | |
551 | { | |
552 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
553 | struct vm_area_struct *vma; | |
554 | struct prio_tree_iter iter; | |
555 | int ret = 0; | |
556 | ||
557 | BUG_ON(PageAnon(page)); | |
558 | ||
559 | spin_lock(&mapping->i_mmap_lock); | |
560 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
561 | if (vma->vm_flags & VM_SHARED) | |
562 | ret += page_mkclean_one(page, vma); | |
563 | } | |
564 | spin_unlock(&mapping->i_mmap_lock); | |
565 | return ret; | |
566 | } | |
567 | ||
568 | int page_mkclean(struct page *page) | |
569 | { | |
570 | int ret = 0; | |
571 | ||
572 | BUG_ON(!PageLocked(page)); | |
573 | ||
574 | if (page_mapped(page)) { | |
575 | struct address_space *mapping = page_mapping(page); | |
ce7e9fae | 576 | if (mapping) { |
d08b3851 | 577 | ret = page_mkclean_file(mapping, page); |
ce7e9fae CB |
578 | if (page_test_dirty(page)) { |
579 | page_clear_dirty(page); | |
580 | ret = 1; | |
581 | } | |
6c210482 | 582 | } |
d08b3851 PZ |
583 | } |
584 | ||
585 | return ret; | |
586 | } | |
60b59bea | 587 | EXPORT_SYMBOL_GPL(page_mkclean); |
d08b3851 | 588 | |
9617d95e | 589 | /** |
43d8eac4 | 590 | * __page_set_anon_rmap - setup new anonymous rmap |
9617d95e NP |
591 | * @page: the page to add the mapping to |
592 | * @vma: the vm area in which the mapping is added | |
593 | * @address: the user virtual address mapped | |
594 | */ | |
595 | static void __page_set_anon_rmap(struct page *page, | |
596 | struct vm_area_struct *vma, unsigned long address) | |
597 | { | |
598 | struct anon_vma *anon_vma = vma->anon_vma; | |
599 | ||
600 | BUG_ON(!anon_vma); | |
601 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | |
602 | page->mapping = (struct address_space *) anon_vma; | |
603 | ||
604 | page->index = linear_page_index(vma, address); | |
605 | ||
a74609fa NP |
606 | /* |
607 | * nr_mapped state can be updated without turning off | |
608 | * interrupts because it is not modified via interrupt. | |
609 | */ | |
f3dbd344 | 610 | __inc_zone_page_state(page, NR_ANON_PAGES); |
9617d95e NP |
611 | } |
612 | ||
c97a9e10 | 613 | /** |
43d8eac4 | 614 | * __page_check_anon_rmap - sanity check anonymous rmap addition |
c97a9e10 NP |
615 | * @page: the page to add the mapping to |
616 | * @vma: the vm area in which the mapping is added | |
617 | * @address: the user virtual address mapped | |
618 | */ | |
619 | static void __page_check_anon_rmap(struct page *page, | |
620 | struct vm_area_struct *vma, unsigned long address) | |
621 | { | |
622 | #ifdef CONFIG_DEBUG_VM | |
623 | /* | |
624 | * The page's anon-rmap details (mapping and index) are guaranteed to | |
625 | * be set up correctly at this point. | |
626 | * | |
627 | * We have exclusion against page_add_anon_rmap because the caller | |
628 | * always holds the page locked, except if called from page_dup_rmap, | |
629 | * in which case the page is already known to be setup. | |
630 | * | |
631 | * We have exclusion against page_add_new_anon_rmap because those pages | |
632 | * are initially only visible via the pagetables, and the pte is locked | |
633 | * over the call to page_add_new_anon_rmap. | |
634 | */ | |
635 | struct anon_vma *anon_vma = vma->anon_vma; | |
636 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | |
637 | BUG_ON(page->mapping != (struct address_space *)anon_vma); | |
638 | BUG_ON(page->index != linear_page_index(vma, address)); | |
639 | #endif | |
640 | } | |
641 | ||
1da177e4 LT |
642 | /** |
643 | * page_add_anon_rmap - add pte mapping to an anonymous page | |
644 | * @page: the page to add the mapping to | |
645 | * @vma: the vm area in which the mapping is added | |
646 | * @address: the user virtual address mapped | |
647 | * | |
c97a9e10 | 648 | * The caller needs to hold the pte lock and the page must be locked. |
1da177e4 LT |
649 | */ |
650 | void page_add_anon_rmap(struct page *page, | |
651 | struct vm_area_struct *vma, unsigned long address) | |
652 | { | |
c97a9e10 NP |
653 | VM_BUG_ON(!PageLocked(page)); |
654 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
9617d95e NP |
655 | if (atomic_inc_and_test(&page->_mapcount)) |
656 | __page_set_anon_rmap(page, vma, address); | |
69029cd5 | 657 | else |
c97a9e10 | 658 | __page_check_anon_rmap(page, vma, address); |
1da177e4 LT |
659 | } |
660 | ||
43d8eac4 | 661 | /** |
9617d95e NP |
662 | * page_add_new_anon_rmap - add pte mapping to a new anonymous page |
663 | * @page: the page to add the mapping to | |
664 | * @vma: the vm area in which the mapping is added | |
665 | * @address: the user virtual address mapped | |
666 | * | |
667 | * Same as page_add_anon_rmap but must only be called on *new* pages. | |
668 | * This means the inc-and-test can be bypassed. | |
c97a9e10 | 669 | * Page does not have to be locked. |
9617d95e NP |
670 | */ |
671 | void page_add_new_anon_rmap(struct page *page, | |
672 | struct vm_area_struct *vma, unsigned long address) | |
673 | { | |
c97a9e10 | 674 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
9617d95e NP |
675 | atomic_set(&page->_mapcount, 0); /* elevate count by 1 (starts at -1) */ |
676 | __page_set_anon_rmap(page, vma, address); | |
677 | } | |
678 | ||
1da177e4 LT |
679 | /** |
680 | * page_add_file_rmap - add pte mapping to a file page | |
681 | * @page: the page to add the mapping to | |
682 | * | |
b8072f09 | 683 | * The caller needs to hold the pte lock. |
1da177e4 LT |
684 | */ |
685 | void page_add_file_rmap(struct page *page) | |
686 | { | |
1da177e4 | 687 | if (atomic_inc_and_test(&page->_mapcount)) |
65ba55f5 | 688 | __inc_zone_page_state(page, NR_FILE_MAPPED); |
1da177e4 LT |
689 | } |
690 | ||
c97a9e10 NP |
691 | #ifdef CONFIG_DEBUG_VM |
692 | /** | |
693 | * page_dup_rmap - duplicate pte mapping to a page | |
694 | * @page: the page to add the mapping to | |
43d8eac4 RD |
695 | * @vma: the vm area being duplicated |
696 | * @address: the user virtual address mapped | |
c97a9e10 NP |
697 | * |
698 | * For copy_page_range only: minimal extract from page_add_file_rmap / | |
699 | * page_add_anon_rmap, avoiding unnecessary tests (already checked) so it's | |
700 | * quicker. | |
701 | * | |
702 | * The caller needs to hold the pte lock. | |
703 | */ | |
704 | void page_dup_rmap(struct page *page, struct vm_area_struct *vma, unsigned long address) | |
705 | { | |
706 | BUG_ON(page_mapcount(page) == 0); | |
707 | if (PageAnon(page)) | |
708 | __page_check_anon_rmap(page, vma, address); | |
709 | atomic_inc(&page->_mapcount); | |
710 | } | |
711 | #endif | |
712 | ||
1da177e4 LT |
713 | /** |
714 | * page_remove_rmap - take down pte mapping from a page | |
715 | * @page: page to remove mapping from | |
43d8eac4 | 716 | * @vma: the vm area in which the mapping is removed |
1da177e4 | 717 | * |
b8072f09 | 718 | * The caller needs to hold the pte lock. |
1da177e4 | 719 | */ |
7de6b805 | 720 | void page_remove_rmap(struct page *page, struct vm_area_struct *vma) |
1da177e4 | 721 | { |
1da177e4 | 722 | if (atomic_add_negative(-1, &page->_mapcount)) { |
b7ab795b | 723 | if (unlikely(page_mapcount(page) < 0)) { |
ef2bf0dc | 724 | printk (KERN_EMERG "Eeek! page_mapcount(page) went negative! (%d)\n", page_mapcount(page)); |
7de6b805 | 725 | printk (KERN_EMERG " page pfn = %lx\n", page_to_pfn(page)); |
ef2bf0dc DJ |
726 | printk (KERN_EMERG " page->flags = %lx\n", page->flags); |
727 | printk (KERN_EMERG " page->count = %x\n", page_count(page)); | |
728 | printk (KERN_EMERG " page->mapping = %p\n", page->mapping); | |
7de6b805 | 729 | print_symbol (KERN_EMERG " vma->vm_ops = %s\n", (unsigned long)vma->vm_ops); |
54cb8821 | 730 | if (vma->vm_ops) { |
54cb8821 NP |
731 | print_symbol (KERN_EMERG " vma->vm_ops->fault = %s\n", (unsigned long)vma->vm_ops->fault); |
732 | } | |
7de6b805 NP |
733 | if (vma->vm_file && vma->vm_file->f_op) |
734 | print_symbol (KERN_EMERG " vma->vm_file->f_op->mmap = %s\n", (unsigned long)vma->vm_file->f_op->mmap); | |
b16bc64d | 735 | BUG(); |
ef2bf0dc | 736 | } |
b16bc64d | 737 | |
1da177e4 | 738 | /* |
16f8c5b2 HD |
739 | * Now that the last pte has gone, s390 must transfer dirty |
740 | * flag from storage key to struct page. We can usually skip | |
741 | * this if the page is anon, so about to be freed; but perhaps | |
742 | * not if it's in swapcache - there might be another pte slot | |
743 | * containing the swap entry, but page not yet written to swap. | |
1da177e4 | 744 | */ |
a4b526b3 MS |
745 | if ((!PageAnon(page) || PageSwapCache(page)) && |
746 | page_test_dirty(page)) { | |
6c210482 | 747 | page_clear_dirty(page); |
1da177e4 | 748 | set_page_dirty(page); |
6c210482 | 749 | } |
5b4e655e KH |
750 | if (PageAnon(page)) |
751 | mem_cgroup_uncharge_page(page); | |
f3dbd344 | 752 | __dec_zone_page_state(page, |
16f8c5b2 HD |
753 | PageAnon(page) ? NR_ANON_PAGES : NR_FILE_MAPPED); |
754 | /* | |
755 | * It would be tidy to reset the PageAnon mapping here, | |
756 | * but that might overwrite a racing page_add_anon_rmap | |
757 | * which increments mapcount after us but sets mapping | |
758 | * before us: so leave the reset to free_hot_cold_page, | |
759 | * and remember that it's only reliable while mapped. | |
760 | * Leaving it set also helps swapoff to reinstate ptes | |
761 | * faster for those pages still in swapcache. | |
762 | */ | |
1da177e4 LT |
763 | } |
764 | } | |
765 | ||
766 | /* | |
767 | * Subfunctions of try_to_unmap: try_to_unmap_one called | |
768 | * repeatedly from either try_to_unmap_anon or try_to_unmap_file. | |
769 | */ | |
a48d07af | 770 | static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
7352349a | 771 | int migration) |
1da177e4 LT |
772 | { |
773 | struct mm_struct *mm = vma->vm_mm; | |
774 | unsigned long address; | |
1da177e4 LT |
775 | pte_t *pte; |
776 | pte_t pteval; | |
c0718806 | 777 | spinlock_t *ptl; |
1da177e4 LT |
778 | int ret = SWAP_AGAIN; |
779 | ||
1da177e4 LT |
780 | address = vma_address(page, vma); |
781 | if (address == -EFAULT) | |
782 | goto out; | |
783 | ||
479db0bf | 784 | pte = page_check_address(page, mm, address, &ptl, 0); |
c0718806 | 785 | if (!pte) |
81b4082d | 786 | goto out; |
1da177e4 LT |
787 | |
788 | /* | |
789 | * If the page is mlock()d, we cannot swap it out. | |
790 | * If it's recently referenced (perhaps page_referenced | |
791 | * skipped over this mm) then we should reactivate it. | |
792 | */ | |
b291f000 NP |
793 | if (!migration) { |
794 | if (vma->vm_flags & VM_LOCKED) { | |
795 | ret = SWAP_MLOCK; | |
796 | goto out_unmap; | |
797 | } | |
798 | if (ptep_clear_flush_young_notify(vma, address, pte)) { | |
799 | ret = SWAP_FAIL; | |
800 | goto out_unmap; | |
801 | } | |
802 | } | |
1da177e4 | 803 | |
1da177e4 LT |
804 | /* Nuke the page table entry. */ |
805 | flush_cache_page(vma, address, page_to_pfn(page)); | |
cddb8a5c | 806 | pteval = ptep_clear_flush_notify(vma, address, pte); |
1da177e4 LT |
807 | |
808 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
809 | if (pte_dirty(pteval)) | |
810 | set_page_dirty(page); | |
811 | ||
365e9c87 HD |
812 | /* Update high watermark before we lower rss */ |
813 | update_hiwater_rss(mm); | |
814 | ||
1da177e4 | 815 | if (PageAnon(page)) { |
4c21e2f2 | 816 | swp_entry_t entry = { .val = page_private(page) }; |
0697212a CL |
817 | |
818 | if (PageSwapCache(page)) { | |
819 | /* | |
820 | * Store the swap location in the pte. | |
821 | * See handle_pte_fault() ... | |
822 | */ | |
823 | swap_duplicate(entry); | |
824 | if (list_empty(&mm->mmlist)) { | |
825 | spin_lock(&mmlist_lock); | |
826 | if (list_empty(&mm->mmlist)) | |
827 | list_add(&mm->mmlist, &init_mm.mmlist); | |
828 | spin_unlock(&mmlist_lock); | |
829 | } | |
442c9137 | 830 | dec_mm_counter(mm, anon_rss); |
64cdd548 | 831 | } else if (PAGE_MIGRATION) { |
0697212a CL |
832 | /* |
833 | * Store the pfn of the page in a special migration | |
834 | * pte. do_swap_page() will wait until the migration | |
835 | * pte is removed and then restart fault handling. | |
836 | */ | |
837 | BUG_ON(!migration); | |
838 | entry = make_migration_entry(page, pte_write(pteval)); | |
1da177e4 LT |
839 | } |
840 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
841 | BUG_ON(pte_file(*pte)); | |
64cdd548 | 842 | } else if (PAGE_MIGRATION && migration) { |
04e62a29 CL |
843 | /* Establish migration entry for a file page */ |
844 | swp_entry_t entry; | |
845 | entry = make_migration_entry(page, pte_write(pteval)); | |
846 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
847 | } else | |
4294621f | 848 | dec_mm_counter(mm, file_rss); |
1da177e4 | 849 | |
04e62a29 | 850 | |
7de6b805 | 851 | page_remove_rmap(page, vma); |
1da177e4 LT |
852 | page_cache_release(page); |
853 | ||
854 | out_unmap: | |
c0718806 | 855 | pte_unmap_unlock(pte, ptl); |
1da177e4 LT |
856 | out: |
857 | return ret; | |
858 | } | |
859 | ||
860 | /* | |
861 | * objrmap doesn't work for nonlinear VMAs because the assumption that | |
862 | * offset-into-file correlates with offset-into-virtual-addresses does not hold. | |
863 | * Consequently, given a particular page and its ->index, we cannot locate the | |
864 | * ptes which are mapping that page without an exhaustive linear search. | |
865 | * | |
866 | * So what this code does is a mini "virtual scan" of each nonlinear VMA which | |
867 | * maps the file to which the target page belongs. The ->vm_private_data field | |
868 | * holds the current cursor into that scan. Successive searches will circulate | |
869 | * around the vma's virtual address space. | |
870 | * | |
871 | * So as more replacement pressure is applied to the pages in a nonlinear VMA, | |
872 | * more scanning pressure is placed against them as well. Eventually pages | |
873 | * will become fully unmapped and are eligible for eviction. | |
874 | * | |
875 | * For very sparsely populated VMAs this is a little inefficient - chances are | |
876 | * there there won't be many ptes located within the scan cluster. In this case | |
877 | * maybe we could scan further - to the end of the pte page, perhaps. | |
b291f000 NP |
878 | * |
879 | * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can | |
880 | * acquire it without blocking. If vma locked, mlock the pages in the cluster, | |
881 | * rather than unmapping them. If we encounter the "check_page" that vmscan is | |
882 | * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN. | |
1da177e4 LT |
883 | */ |
884 | #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE) | |
885 | #define CLUSTER_MASK (~(CLUSTER_SIZE - 1)) | |
886 | ||
b291f000 NP |
887 | static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount, |
888 | struct vm_area_struct *vma, struct page *check_page) | |
1da177e4 LT |
889 | { |
890 | struct mm_struct *mm = vma->vm_mm; | |
891 | pgd_t *pgd; | |
892 | pud_t *pud; | |
893 | pmd_t *pmd; | |
c0718806 | 894 | pte_t *pte; |
1da177e4 | 895 | pte_t pteval; |
c0718806 | 896 | spinlock_t *ptl; |
1da177e4 LT |
897 | struct page *page; |
898 | unsigned long address; | |
899 | unsigned long end; | |
b291f000 NP |
900 | int ret = SWAP_AGAIN; |
901 | int locked_vma = 0; | |
1da177e4 | 902 | |
1da177e4 LT |
903 | address = (vma->vm_start + cursor) & CLUSTER_MASK; |
904 | end = address + CLUSTER_SIZE; | |
905 | if (address < vma->vm_start) | |
906 | address = vma->vm_start; | |
907 | if (end > vma->vm_end) | |
908 | end = vma->vm_end; | |
909 | ||
910 | pgd = pgd_offset(mm, address); | |
911 | if (!pgd_present(*pgd)) | |
b291f000 | 912 | return ret; |
1da177e4 LT |
913 | |
914 | pud = pud_offset(pgd, address); | |
915 | if (!pud_present(*pud)) | |
b291f000 | 916 | return ret; |
1da177e4 LT |
917 | |
918 | pmd = pmd_offset(pud, address); | |
919 | if (!pmd_present(*pmd)) | |
b291f000 NP |
920 | return ret; |
921 | ||
922 | /* | |
923 | * MLOCK_PAGES => feature is configured. | |
924 | * if we can acquire the mmap_sem for read, and vma is VM_LOCKED, | |
925 | * keep the sem while scanning the cluster for mlocking pages. | |
926 | */ | |
927 | if (MLOCK_PAGES && down_read_trylock(&vma->vm_mm->mmap_sem)) { | |
928 | locked_vma = (vma->vm_flags & VM_LOCKED); | |
929 | if (!locked_vma) | |
930 | up_read(&vma->vm_mm->mmap_sem); /* don't need it */ | |
931 | } | |
c0718806 HD |
932 | |
933 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); | |
1da177e4 | 934 | |
365e9c87 HD |
935 | /* Update high watermark before we lower rss */ |
936 | update_hiwater_rss(mm); | |
937 | ||
c0718806 | 938 | for (; address < end; pte++, address += PAGE_SIZE) { |
1da177e4 LT |
939 | if (!pte_present(*pte)) |
940 | continue; | |
6aab341e LT |
941 | page = vm_normal_page(vma, address, *pte); |
942 | BUG_ON(!page || PageAnon(page)); | |
1da177e4 | 943 | |
b291f000 NP |
944 | if (locked_vma) { |
945 | mlock_vma_page(page); /* no-op if already mlocked */ | |
946 | if (page == check_page) | |
947 | ret = SWAP_MLOCK; | |
948 | continue; /* don't unmap */ | |
949 | } | |
950 | ||
cddb8a5c | 951 | if (ptep_clear_flush_young_notify(vma, address, pte)) |
1da177e4 LT |
952 | continue; |
953 | ||
954 | /* Nuke the page table entry. */ | |
eca35133 | 955 | flush_cache_page(vma, address, pte_pfn(*pte)); |
cddb8a5c | 956 | pteval = ptep_clear_flush_notify(vma, address, pte); |
1da177e4 LT |
957 | |
958 | /* If nonlinear, store the file page offset in the pte. */ | |
959 | if (page->index != linear_page_index(vma, address)) | |
960 | set_pte_at(mm, address, pte, pgoff_to_pte(page->index)); | |
961 | ||
962 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
963 | if (pte_dirty(pteval)) | |
964 | set_page_dirty(page); | |
965 | ||
7de6b805 | 966 | page_remove_rmap(page, vma); |
1da177e4 | 967 | page_cache_release(page); |
4294621f | 968 | dec_mm_counter(mm, file_rss); |
1da177e4 LT |
969 | (*mapcount)--; |
970 | } | |
c0718806 | 971 | pte_unmap_unlock(pte - 1, ptl); |
b291f000 NP |
972 | if (locked_vma) |
973 | up_read(&vma->vm_mm->mmap_sem); | |
974 | return ret; | |
1da177e4 LT |
975 | } |
976 | ||
b291f000 NP |
977 | /* |
978 | * common handling for pages mapped in VM_LOCKED vmas | |
979 | */ | |
980 | static int try_to_mlock_page(struct page *page, struct vm_area_struct *vma) | |
981 | { | |
982 | int mlocked = 0; | |
983 | ||
984 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { | |
985 | if (vma->vm_flags & VM_LOCKED) { | |
986 | mlock_vma_page(page); | |
987 | mlocked++; /* really mlocked the page */ | |
988 | } | |
989 | up_read(&vma->vm_mm->mmap_sem); | |
990 | } | |
991 | return mlocked; | |
992 | } | |
993 | ||
994 | /** | |
995 | * try_to_unmap_anon - unmap or unlock anonymous page using the object-based | |
996 | * rmap method | |
997 | * @page: the page to unmap/unlock | |
998 | * @unlock: request for unlock rather than unmap [unlikely] | |
999 | * @migration: unmapping for migration - ignored if @unlock | |
1000 | * | |
1001 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1002 | * contained in the anon_vma struct it points to. | |
1003 | * | |
1004 | * This function is only called from try_to_unmap/try_to_munlock for | |
1005 | * anonymous pages. | |
1006 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1007 | * where the page was found will be held for write. So, we won't recheck | |
1008 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1009 | * 'LOCKED. | |
1010 | */ | |
1011 | static int try_to_unmap_anon(struct page *page, int unlock, int migration) | |
1da177e4 LT |
1012 | { |
1013 | struct anon_vma *anon_vma; | |
1014 | struct vm_area_struct *vma; | |
b291f000 | 1015 | unsigned int mlocked = 0; |
1da177e4 LT |
1016 | int ret = SWAP_AGAIN; |
1017 | ||
b291f000 NP |
1018 | if (MLOCK_PAGES && unlikely(unlock)) |
1019 | ret = SWAP_SUCCESS; /* default for try_to_munlock() */ | |
1020 | ||
1da177e4 LT |
1021 | anon_vma = page_lock_anon_vma(page); |
1022 | if (!anon_vma) | |
1023 | return ret; | |
1024 | ||
1025 | list_for_each_entry(vma, &anon_vma->head, anon_vma_node) { | |
b291f000 NP |
1026 | if (MLOCK_PAGES && unlikely(unlock)) { |
1027 | if (!((vma->vm_flags & VM_LOCKED) && | |
1028 | page_mapped_in_vma(page, vma))) | |
1029 | continue; /* must visit all unlocked vmas */ | |
1030 | ret = SWAP_MLOCK; /* saw at least one mlocked vma */ | |
1031 | } else { | |
1032 | ret = try_to_unmap_one(page, vma, migration); | |
1033 | if (ret == SWAP_FAIL || !page_mapped(page)) | |
1034 | break; | |
1035 | } | |
1036 | if (ret == SWAP_MLOCK) { | |
1037 | mlocked = try_to_mlock_page(page, vma); | |
1038 | if (mlocked) | |
1039 | break; /* stop if actually mlocked page */ | |
1040 | } | |
1da177e4 | 1041 | } |
34bbd704 ON |
1042 | |
1043 | page_unlock_anon_vma(anon_vma); | |
b291f000 NP |
1044 | |
1045 | if (mlocked) | |
1046 | ret = SWAP_MLOCK; /* actually mlocked the page */ | |
1047 | else if (ret == SWAP_MLOCK) | |
1048 | ret = SWAP_AGAIN; /* saw VM_LOCKED vma */ | |
1049 | ||
1da177e4 LT |
1050 | return ret; |
1051 | } | |
1052 | ||
1053 | /** | |
b291f000 NP |
1054 | * try_to_unmap_file - unmap/unlock file page using the object-based rmap method |
1055 | * @page: the page to unmap/unlock | |
1056 | * @unlock: request for unlock rather than unmap [unlikely] | |
1057 | * @migration: unmapping for migration - ignored if @unlock | |
1da177e4 LT |
1058 | * |
1059 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1060 | * contained in the address_space struct it points to. | |
1061 | * | |
b291f000 NP |
1062 | * This function is only called from try_to_unmap/try_to_munlock for |
1063 | * object-based pages. | |
1064 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1065 | * where the page was found will be held for write. So, we won't recheck | |
1066 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1067 | * 'LOCKED. | |
1da177e4 | 1068 | */ |
b291f000 | 1069 | static int try_to_unmap_file(struct page *page, int unlock, int migration) |
1da177e4 LT |
1070 | { |
1071 | struct address_space *mapping = page->mapping; | |
1072 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
1073 | struct vm_area_struct *vma; | |
1074 | struct prio_tree_iter iter; | |
1075 | int ret = SWAP_AGAIN; | |
1076 | unsigned long cursor; | |
1077 | unsigned long max_nl_cursor = 0; | |
1078 | unsigned long max_nl_size = 0; | |
1079 | unsigned int mapcount; | |
b291f000 NP |
1080 | unsigned int mlocked = 0; |
1081 | ||
1082 | if (MLOCK_PAGES && unlikely(unlock)) | |
1083 | ret = SWAP_SUCCESS; /* default for try_to_munlock() */ | |
1da177e4 LT |
1084 | |
1085 | spin_lock(&mapping->i_mmap_lock); | |
1086 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
b291f000 NP |
1087 | if (MLOCK_PAGES && unlikely(unlock)) { |
1088 | if (!(vma->vm_flags & VM_LOCKED)) | |
1089 | continue; /* must visit all vmas */ | |
1090 | ret = SWAP_MLOCK; | |
1091 | } else { | |
1092 | ret = try_to_unmap_one(page, vma, migration); | |
1093 | if (ret == SWAP_FAIL || !page_mapped(page)) | |
1094 | goto out; | |
1095 | } | |
1096 | if (ret == SWAP_MLOCK) { | |
1097 | mlocked = try_to_mlock_page(page, vma); | |
1098 | if (mlocked) | |
1099 | break; /* stop if actually mlocked page */ | |
1100 | } | |
1da177e4 LT |
1101 | } |
1102 | ||
b291f000 NP |
1103 | if (mlocked) |
1104 | goto out; | |
1105 | ||
1da177e4 LT |
1106 | if (list_empty(&mapping->i_mmap_nonlinear)) |
1107 | goto out; | |
1108 | ||
1109 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, | |
1110 | shared.vm_set.list) { | |
b291f000 NP |
1111 | if (MLOCK_PAGES && unlikely(unlock)) { |
1112 | if (!(vma->vm_flags & VM_LOCKED)) | |
1113 | continue; /* must visit all vmas */ | |
1114 | ret = SWAP_MLOCK; /* leave mlocked == 0 */ | |
1115 | goto out; /* no need to look further */ | |
1116 | } | |
1117 | if (!MLOCK_PAGES && !migration && (vma->vm_flags & VM_LOCKED)) | |
1da177e4 LT |
1118 | continue; |
1119 | cursor = (unsigned long) vma->vm_private_data; | |
1120 | if (cursor > max_nl_cursor) | |
1121 | max_nl_cursor = cursor; | |
1122 | cursor = vma->vm_end - vma->vm_start; | |
1123 | if (cursor > max_nl_size) | |
1124 | max_nl_size = cursor; | |
1125 | } | |
1126 | ||
b291f000 | 1127 | if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */ |
1da177e4 LT |
1128 | ret = SWAP_FAIL; |
1129 | goto out; | |
1130 | } | |
1131 | ||
1132 | /* | |
1133 | * We don't try to search for this page in the nonlinear vmas, | |
1134 | * and page_referenced wouldn't have found it anyway. Instead | |
1135 | * just walk the nonlinear vmas trying to age and unmap some. | |
1136 | * The mapcount of the page we came in with is irrelevant, | |
1137 | * but even so use it as a guide to how hard we should try? | |
1138 | */ | |
1139 | mapcount = page_mapcount(page); | |
1140 | if (!mapcount) | |
1141 | goto out; | |
1142 | cond_resched_lock(&mapping->i_mmap_lock); | |
1143 | ||
1144 | max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK; | |
1145 | if (max_nl_cursor == 0) | |
1146 | max_nl_cursor = CLUSTER_SIZE; | |
1147 | ||
1148 | do { | |
1149 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, | |
1150 | shared.vm_set.list) { | |
b291f000 NP |
1151 | if (!MLOCK_PAGES && !migration && |
1152 | (vma->vm_flags & VM_LOCKED)) | |
1da177e4 LT |
1153 | continue; |
1154 | cursor = (unsigned long) vma->vm_private_data; | |
839b9685 | 1155 | while ( cursor < max_nl_cursor && |
1da177e4 | 1156 | cursor < vma->vm_end - vma->vm_start) { |
b291f000 NP |
1157 | ret = try_to_unmap_cluster(cursor, &mapcount, |
1158 | vma, page); | |
1159 | if (ret == SWAP_MLOCK) | |
1160 | mlocked = 2; /* to return below */ | |
1da177e4 LT |
1161 | cursor += CLUSTER_SIZE; |
1162 | vma->vm_private_data = (void *) cursor; | |
1163 | if ((int)mapcount <= 0) | |
1164 | goto out; | |
1165 | } | |
1166 | vma->vm_private_data = (void *) max_nl_cursor; | |
1167 | } | |
1168 | cond_resched_lock(&mapping->i_mmap_lock); | |
1169 | max_nl_cursor += CLUSTER_SIZE; | |
1170 | } while (max_nl_cursor <= max_nl_size); | |
1171 | ||
1172 | /* | |
1173 | * Don't loop forever (perhaps all the remaining pages are | |
1174 | * in locked vmas). Reset cursor on all unreserved nonlinear | |
1175 | * vmas, now forgetting on which ones it had fallen behind. | |
1176 | */ | |
101d2be7 HD |
1177 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) |
1178 | vma->vm_private_data = NULL; | |
1da177e4 LT |
1179 | out: |
1180 | spin_unlock(&mapping->i_mmap_lock); | |
b291f000 NP |
1181 | if (mlocked) |
1182 | ret = SWAP_MLOCK; /* actually mlocked the page */ | |
1183 | else if (ret == SWAP_MLOCK) | |
1184 | ret = SWAP_AGAIN; /* saw VM_LOCKED vma */ | |
1da177e4 LT |
1185 | return ret; |
1186 | } | |
1187 | ||
1188 | /** | |
1189 | * try_to_unmap - try to remove all page table mappings to a page | |
1190 | * @page: the page to get unmapped | |
43d8eac4 | 1191 | * @migration: migration flag |
1da177e4 LT |
1192 | * |
1193 | * Tries to remove all the page table entries which are mapping this | |
1194 | * page, used in the pageout path. Caller must hold the page lock. | |
1195 | * Return values are: | |
1196 | * | |
1197 | * SWAP_SUCCESS - we succeeded in removing all mappings | |
1198 | * SWAP_AGAIN - we missed a mapping, try again later | |
1199 | * SWAP_FAIL - the page is unswappable | |
b291f000 | 1200 | * SWAP_MLOCK - page is mlocked. |
1da177e4 | 1201 | */ |
7352349a | 1202 | int try_to_unmap(struct page *page, int migration) |
1da177e4 LT |
1203 | { |
1204 | int ret; | |
1205 | ||
1da177e4 LT |
1206 | BUG_ON(!PageLocked(page)); |
1207 | ||
1208 | if (PageAnon(page)) | |
b291f000 | 1209 | ret = try_to_unmap_anon(page, 0, migration); |
1da177e4 | 1210 | else |
b291f000 NP |
1211 | ret = try_to_unmap_file(page, 0, migration); |
1212 | if (ret != SWAP_MLOCK && !page_mapped(page)) | |
1da177e4 LT |
1213 | ret = SWAP_SUCCESS; |
1214 | return ret; | |
1215 | } | |
81b4082d | 1216 | |
b291f000 NP |
1217 | #ifdef CONFIG_UNEVICTABLE_LRU |
1218 | /** | |
1219 | * try_to_munlock - try to munlock a page | |
1220 | * @page: the page to be munlocked | |
1221 | * | |
1222 | * Called from munlock code. Checks all of the VMAs mapping the page | |
1223 | * to make sure nobody else has this page mlocked. The page will be | |
1224 | * returned with PG_mlocked cleared if no other vmas have it mlocked. | |
1225 | * | |
1226 | * Return values are: | |
1227 | * | |
1228 | * SWAP_SUCCESS - no vma's holding page mlocked. | |
1229 | * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem | |
1230 | * SWAP_MLOCK - page is now mlocked. | |
1231 | */ | |
1232 | int try_to_munlock(struct page *page) | |
1233 | { | |
1234 | VM_BUG_ON(!PageLocked(page) || PageLRU(page)); | |
1235 | ||
1236 | if (PageAnon(page)) | |
1237 | return try_to_unmap_anon(page, 1, 0); | |
1238 | else | |
1239 | return try_to_unmap_file(page, 1, 0); | |
1240 | } | |
1241 | #endif |