]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/memory.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * demand-loading started 01.12.91 - seems it is high on the list of | |
9 | * things wanted, and it should be easy to implement. - Linus | |
10 | */ | |
11 | ||
12 | /* | |
13 | * Ok, demand-loading was easy, shared pages a little bit tricker. Shared | |
14 | * pages started 02.12.91, seems to work. - Linus. | |
15 | * | |
16 | * Tested sharing by executing about 30 /bin/sh: under the old kernel it | |
17 | * would have taken more than the 6M I have free, but it worked well as | |
18 | * far as I could see. | |
19 | * | |
20 | * Also corrected some "invalidate()"s - I wasn't doing enough of them. | |
21 | */ | |
22 | ||
23 | /* | |
24 | * Real VM (paging to/from disk) started 18.12.91. Much more work and | |
25 | * thought has to go into this. Oh, well.. | |
26 | * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why. | |
27 | * Found it. Everything seems to work now. | |
28 | * 20.12.91 - Ok, making the swap-device changeable like the root. | |
29 | */ | |
30 | ||
31 | /* | |
32 | * 05.04.94 - Multi-page memory management added for v1.1. | |
33 | * Idea by Alex Bligh (alex@cconcepts.co.uk) | |
34 | * | |
35 | * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG | |
36 | * (Gerhard.Wichert@pdb.siemens.de) | |
37 | * | |
38 | * Aug/Sep 2004 Changed to four level page tables (Andi Kleen) | |
39 | */ | |
40 | ||
41 | #include <linux/kernel_stat.h> | |
42 | #include <linux/mm.h> | |
43 | #include <linux/hugetlb.h> | |
44 | #include <linux/mman.h> | |
45 | #include <linux/swap.h> | |
46 | #include <linux/highmem.h> | |
47 | #include <linux/pagemap.h> | |
48 | #include <linux/rmap.h> | |
49 | #include <linux/module.h> | |
50 | #include <linux/init.h> | |
51 | ||
52 | #include <asm/pgalloc.h> | |
53 | #include <asm/uaccess.h> | |
54 | #include <asm/tlb.h> | |
55 | #include <asm/tlbflush.h> | |
56 | #include <asm/pgtable.h> | |
57 | ||
58 | #include <linux/swapops.h> | |
59 | #include <linux/elf.h> | |
60 | ||
61 | #ifndef CONFIG_DISCONTIGMEM | |
62 | /* use the per-pgdat data instead for discontigmem - mbligh */ | |
63 | unsigned long max_mapnr; | |
64 | struct page *mem_map; | |
65 | ||
66 | EXPORT_SYMBOL(max_mapnr); | |
67 | EXPORT_SYMBOL(mem_map); | |
68 | #endif | |
69 | ||
70 | unsigned long num_physpages; | |
71 | /* | |
72 | * A number of key systems in x86 including ioremap() rely on the assumption | |
73 | * that high_memory defines the upper bound on direct map memory, then end | |
74 | * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and | |
75 | * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL | |
76 | * and ZONE_HIGHMEM. | |
77 | */ | |
78 | void * high_memory; | |
79 | unsigned long vmalloc_earlyreserve; | |
80 | ||
81 | EXPORT_SYMBOL(num_physpages); | |
82 | EXPORT_SYMBOL(high_memory); | |
83 | EXPORT_SYMBOL(vmalloc_earlyreserve); | |
84 | ||
85 | /* | |
86 | * If a p?d_bad entry is found while walking page tables, report | |
87 | * the error, before resetting entry to p?d_none. Usually (but | |
88 | * very seldom) called out from the p?d_none_or_clear_bad macros. | |
89 | */ | |
90 | ||
91 | void pgd_clear_bad(pgd_t *pgd) | |
92 | { | |
93 | pgd_ERROR(*pgd); | |
94 | pgd_clear(pgd); | |
95 | } | |
96 | ||
97 | void pud_clear_bad(pud_t *pud) | |
98 | { | |
99 | pud_ERROR(*pud); | |
100 | pud_clear(pud); | |
101 | } | |
102 | ||
103 | void pmd_clear_bad(pmd_t *pmd) | |
104 | { | |
105 | pmd_ERROR(*pmd); | |
106 | pmd_clear(pmd); | |
107 | } | |
108 | ||
109 | /* | |
110 | * Note: this doesn't free the actual pages themselves. That | |
111 | * has been handled earlier when unmapping all the memory regions. | |
112 | */ | |
113 | static inline void clear_pte_range(struct mmu_gather *tlb, pmd_t *pmd, | |
114 | unsigned long addr, unsigned long end) | |
115 | { | |
116 | if (!((addr | end) & ~PMD_MASK)) { | |
117 | /* Only free fully aligned ranges */ | |
118 | struct page *page = pmd_page(*pmd); | |
119 | pmd_clear(pmd); | |
120 | dec_page_state(nr_page_table_pages); | |
121 | tlb->mm->nr_ptes--; | |
122 | pte_free_tlb(tlb, page); | |
123 | } | |
124 | } | |
125 | ||
126 | static inline void clear_pmd_range(struct mmu_gather *tlb, pud_t *pud, | |
127 | unsigned long addr, unsigned long end) | |
128 | { | |
129 | pmd_t *pmd; | |
130 | unsigned long next; | |
131 | pmd_t *empty_pmd = NULL; | |
132 | ||
133 | pmd = pmd_offset(pud, addr); | |
134 | ||
135 | /* Only free fully aligned ranges */ | |
136 | if (!((addr | end) & ~PUD_MASK)) | |
137 | empty_pmd = pmd; | |
138 | do { | |
139 | next = pmd_addr_end(addr, end); | |
140 | if (pmd_none_or_clear_bad(pmd)) | |
141 | continue; | |
142 | clear_pte_range(tlb, pmd, addr, next); | |
143 | } while (pmd++, addr = next, addr != end); | |
144 | ||
145 | if (empty_pmd) { | |
146 | pud_clear(pud); | |
147 | pmd_free_tlb(tlb, empty_pmd); | |
148 | } | |
149 | } | |
150 | ||
151 | static inline void clear_pud_range(struct mmu_gather *tlb, pgd_t *pgd, | |
152 | unsigned long addr, unsigned long end) | |
153 | { | |
154 | pud_t *pud; | |
155 | unsigned long next; | |
156 | pud_t *empty_pud = NULL; | |
157 | ||
158 | pud = pud_offset(pgd, addr); | |
159 | ||
160 | /* Only free fully aligned ranges */ | |
161 | if (!((addr | end) & ~PGDIR_MASK)) | |
162 | empty_pud = pud; | |
163 | do { | |
164 | next = pud_addr_end(addr, end); | |
165 | if (pud_none_or_clear_bad(pud)) | |
166 | continue; | |
167 | clear_pmd_range(tlb, pud, addr, next); | |
168 | } while (pud++, addr = next, addr != end); | |
169 | ||
170 | if (empty_pud) { | |
171 | pgd_clear(pgd); | |
172 | pud_free_tlb(tlb, empty_pud); | |
173 | } | |
174 | } | |
175 | ||
176 | /* | |
177 | * This function clears user-level page tables of a process. | |
178 | * Unlike other pagetable walks, some memory layouts might give end 0. | |
179 | * Must be called with pagetable lock held. | |
180 | */ | |
181 | void clear_page_range(struct mmu_gather *tlb, | |
182 | unsigned long addr, unsigned long end) | |
183 | { | |
184 | pgd_t *pgd; | |
185 | unsigned long next; | |
186 | ||
187 | pgd = pgd_offset(tlb->mm, addr); | |
188 | do { | |
189 | next = pgd_addr_end(addr, end); | |
190 | if (pgd_none_or_clear_bad(pgd)) | |
191 | continue; | |
192 | clear_pud_range(tlb, pgd, addr, next); | |
193 | } while (pgd++, addr = next, addr != end); | |
194 | } | |
195 | ||
196 | pte_t fastcall * pte_alloc_map(struct mm_struct *mm, pmd_t *pmd, unsigned long address) | |
197 | { | |
198 | if (!pmd_present(*pmd)) { | |
199 | struct page *new; | |
200 | ||
201 | spin_unlock(&mm->page_table_lock); | |
202 | new = pte_alloc_one(mm, address); | |
203 | spin_lock(&mm->page_table_lock); | |
204 | if (!new) | |
205 | return NULL; | |
206 | /* | |
207 | * Because we dropped the lock, we should re-check the | |
208 | * entry, as somebody else could have populated it.. | |
209 | */ | |
210 | if (pmd_present(*pmd)) { | |
211 | pte_free(new); | |
212 | goto out; | |
213 | } | |
214 | mm->nr_ptes++; | |
215 | inc_page_state(nr_page_table_pages); | |
216 | pmd_populate(mm, pmd, new); | |
217 | } | |
218 | out: | |
219 | return pte_offset_map(pmd, address); | |
220 | } | |
221 | ||
222 | pte_t fastcall * pte_alloc_kernel(struct mm_struct *mm, pmd_t *pmd, unsigned long address) | |
223 | { | |
224 | if (!pmd_present(*pmd)) { | |
225 | pte_t *new; | |
226 | ||
227 | spin_unlock(&mm->page_table_lock); | |
228 | new = pte_alloc_one_kernel(mm, address); | |
229 | spin_lock(&mm->page_table_lock); | |
230 | if (!new) | |
231 | return NULL; | |
232 | ||
233 | /* | |
234 | * Because we dropped the lock, we should re-check the | |
235 | * entry, as somebody else could have populated it.. | |
236 | */ | |
237 | if (pmd_present(*pmd)) { | |
238 | pte_free_kernel(new); | |
239 | goto out; | |
240 | } | |
241 | pmd_populate_kernel(mm, pmd, new); | |
242 | } | |
243 | out: | |
244 | return pte_offset_kernel(pmd, address); | |
245 | } | |
246 | ||
247 | /* | |
248 | * copy one vm_area from one task to the other. Assumes the page tables | |
249 | * already present in the new task to be cleared in the whole range | |
250 | * covered by this vma. | |
251 | * | |
252 | * dst->page_table_lock is held on entry and exit, | |
253 | * but may be dropped within p[mg]d_alloc() and pte_alloc_map(). | |
254 | */ | |
255 | ||
256 | static inline void | |
257 | copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
258 | pte_t *dst_pte, pte_t *src_pte, unsigned long vm_flags, | |
259 | unsigned long addr) | |
260 | { | |
261 | pte_t pte = *src_pte; | |
262 | struct page *page; | |
263 | unsigned long pfn; | |
264 | ||
265 | /* pte contains position in swap or file, so copy. */ | |
266 | if (unlikely(!pte_present(pte))) { | |
267 | if (!pte_file(pte)) { | |
268 | swap_duplicate(pte_to_swp_entry(pte)); | |
269 | /* make sure dst_mm is on swapoff's mmlist. */ | |
270 | if (unlikely(list_empty(&dst_mm->mmlist))) { | |
271 | spin_lock(&mmlist_lock); | |
272 | list_add(&dst_mm->mmlist, &src_mm->mmlist); | |
273 | spin_unlock(&mmlist_lock); | |
274 | } | |
275 | } | |
276 | set_pte_at(dst_mm, addr, dst_pte, pte); | |
277 | return; | |
278 | } | |
279 | ||
280 | pfn = pte_pfn(pte); | |
281 | /* the pte points outside of valid memory, the | |
282 | * mapping is assumed to be good, meaningful | |
283 | * and not mapped via rmap - duplicate the | |
284 | * mapping as is. | |
285 | */ | |
286 | page = NULL; | |
287 | if (pfn_valid(pfn)) | |
288 | page = pfn_to_page(pfn); | |
289 | ||
290 | if (!page || PageReserved(page)) { | |
291 | set_pte_at(dst_mm, addr, dst_pte, pte); | |
292 | return; | |
293 | } | |
294 | ||
295 | /* | |
296 | * If it's a COW mapping, write protect it both | |
297 | * in the parent and the child | |
298 | */ | |
299 | if ((vm_flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE) { | |
300 | ptep_set_wrprotect(src_mm, addr, src_pte); | |
301 | pte = *src_pte; | |
302 | } | |
303 | ||
304 | /* | |
305 | * If it's a shared mapping, mark it clean in | |
306 | * the child | |
307 | */ | |
308 | if (vm_flags & VM_SHARED) | |
309 | pte = pte_mkclean(pte); | |
310 | pte = pte_mkold(pte); | |
311 | get_page(page); | |
312 | inc_mm_counter(dst_mm, rss); | |
313 | if (PageAnon(page)) | |
314 | inc_mm_counter(dst_mm, anon_rss); | |
315 | set_pte_at(dst_mm, addr, dst_pte, pte); | |
316 | page_dup_rmap(page); | |
317 | } | |
318 | ||
319 | static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
320 | pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, | |
321 | unsigned long addr, unsigned long end) | |
322 | { | |
323 | pte_t *src_pte, *dst_pte; | |
324 | unsigned long vm_flags = vma->vm_flags; | |
325 | int progress; | |
326 | ||
327 | again: | |
328 | dst_pte = pte_alloc_map(dst_mm, dst_pmd, addr); | |
329 | if (!dst_pte) | |
330 | return -ENOMEM; | |
331 | src_pte = pte_offset_map_nested(src_pmd, addr); | |
332 | ||
333 | progress = 0; | |
334 | spin_lock(&src_mm->page_table_lock); | |
335 | do { | |
336 | /* | |
337 | * We are holding two locks at this point - either of them | |
338 | * could generate latencies in another task on another CPU. | |
339 | */ | |
340 | if (progress >= 32 && (need_resched() || | |
341 | need_lockbreak(&src_mm->page_table_lock) || | |
342 | need_lockbreak(&dst_mm->page_table_lock))) | |
343 | break; | |
344 | if (pte_none(*src_pte)) { | |
345 | progress++; | |
346 | continue; | |
347 | } | |
348 | copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vm_flags, addr); | |
349 | progress += 8; | |
350 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); | |
351 | spin_unlock(&src_mm->page_table_lock); | |
352 | ||
353 | pte_unmap_nested(src_pte - 1); | |
354 | pte_unmap(dst_pte - 1); | |
355 | cond_resched_lock(&dst_mm->page_table_lock); | |
356 | if (addr != end) | |
357 | goto again; | |
358 | return 0; | |
359 | } | |
360 | ||
361 | static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
362 | pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, | |
363 | unsigned long addr, unsigned long end) | |
364 | { | |
365 | pmd_t *src_pmd, *dst_pmd; | |
366 | unsigned long next; | |
367 | ||
368 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); | |
369 | if (!dst_pmd) | |
370 | return -ENOMEM; | |
371 | src_pmd = pmd_offset(src_pud, addr); | |
372 | do { | |
373 | next = pmd_addr_end(addr, end); | |
374 | if (pmd_none_or_clear_bad(src_pmd)) | |
375 | continue; | |
376 | if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, | |
377 | vma, addr, next)) | |
378 | return -ENOMEM; | |
379 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); | |
380 | return 0; | |
381 | } | |
382 | ||
383 | static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
384 | pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, | |
385 | unsigned long addr, unsigned long end) | |
386 | { | |
387 | pud_t *src_pud, *dst_pud; | |
388 | unsigned long next; | |
389 | ||
390 | dst_pud = pud_alloc(dst_mm, dst_pgd, addr); | |
391 | if (!dst_pud) | |
392 | return -ENOMEM; | |
393 | src_pud = pud_offset(src_pgd, addr); | |
394 | do { | |
395 | next = pud_addr_end(addr, end); | |
396 | if (pud_none_or_clear_bad(src_pud)) | |
397 | continue; | |
398 | if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, | |
399 | vma, addr, next)) | |
400 | return -ENOMEM; | |
401 | } while (dst_pud++, src_pud++, addr = next, addr != end); | |
402 | return 0; | |
403 | } | |
404 | ||
405 | int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
406 | struct vm_area_struct *vma) | |
407 | { | |
408 | pgd_t *src_pgd, *dst_pgd; | |
409 | unsigned long next; | |
410 | unsigned long addr = vma->vm_start; | |
411 | unsigned long end = vma->vm_end; | |
412 | ||
413 | if (is_vm_hugetlb_page(vma)) | |
414 | return copy_hugetlb_page_range(dst_mm, src_mm, vma); | |
415 | ||
416 | dst_pgd = pgd_offset(dst_mm, addr); | |
417 | src_pgd = pgd_offset(src_mm, addr); | |
418 | do { | |
419 | next = pgd_addr_end(addr, end); | |
420 | if (pgd_none_or_clear_bad(src_pgd)) | |
421 | continue; | |
422 | if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, | |
423 | vma, addr, next)) | |
424 | return -ENOMEM; | |
425 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); | |
426 | return 0; | |
427 | } | |
428 | ||
429 | static void zap_pte_range(struct mmu_gather *tlb, pmd_t *pmd, | |
430 | unsigned long addr, unsigned long end, | |
431 | struct zap_details *details) | |
432 | { | |
433 | pte_t *pte; | |
434 | ||
435 | pte = pte_offset_map(pmd, addr); | |
436 | do { | |
437 | pte_t ptent = *pte; | |
438 | if (pte_none(ptent)) | |
439 | continue; | |
440 | if (pte_present(ptent)) { | |
441 | struct page *page = NULL; | |
442 | unsigned long pfn = pte_pfn(ptent); | |
443 | if (pfn_valid(pfn)) { | |
444 | page = pfn_to_page(pfn); | |
445 | if (PageReserved(page)) | |
446 | page = NULL; | |
447 | } | |
448 | if (unlikely(details) && page) { | |
449 | /* | |
450 | * unmap_shared_mapping_pages() wants to | |
451 | * invalidate cache without truncating: | |
452 | * unmap shared but keep private pages. | |
453 | */ | |
454 | if (details->check_mapping && | |
455 | details->check_mapping != page->mapping) | |
456 | continue; | |
457 | /* | |
458 | * Each page->index must be checked when | |
459 | * invalidating or truncating nonlinear. | |
460 | */ | |
461 | if (details->nonlinear_vma && | |
462 | (page->index < details->first_index || | |
463 | page->index > details->last_index)) | |
464 | continue; | |
465 | } | |
466 | ptent = ptep_get_and_clear(tlb->mm, addr, pte); | |
467 | tlb_remove_tlb_entry(tlb, pte, addr); | |
468 | if (unlikely(!page)) | |
469 | continue; | |
470 | if (unlikely(details) && details->nonlinear_vma | |
471 | && linear_page_index(details->nonlinear_vma, | |
472 | addr) != page->index) | |
473 | set_pte_at(tlb->mm, addr, pte, | |
474 | pgoff_to_pte(page->index)); | |
475 | if (pte_dirty(ptent)) | |
476 | set_page_dirty(page); | |
477 | if (PageAnon(page)) | |
478 | dec_mm_counter(tlb->mm, anon_rss); | |
479 | else if (pte_young(ptent)) | |
480 | mark_page_accessed(page); | |
481 | tlb->freed++; | |
482 | page_remove_rmap(page); | |
483 | tlb_remove_page(tlb, page); | |
484 | continue; | |
485 | } | |
486 | /* | |
487 | * If details->check_mapping, we leave swap entries; | |
488 | * if details->nonlinear_vma, we leave file entries. | |
489 | */ | |
490 | if (unlikely(details)) | |
491 | continue; | |
492 | if (!pte_file(ptent)) | |
493 | free_swap_and_cache(pte_to_swp_entry(ptent)); | |
494 | pte_clear(tlb->mm, addr, pte); | |
495 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
496 | pte_unmap(pte - 1); | |
497 | } | |
498 | ||
499 | static inline void zap_pmd_range(struct mmu_gather *tlb, pud_t *pud, | |
500 | unsigned long addr, unsigned long end, | |
501 | struct zap_details *details) | |
502 | { | |
503 | pmd_t *pmd; | |
504 | unsigned long next; | |
505 | ||
506 | pmd = pmd_offset(pud, addr); | |
507 | do { | |
508 | next = pmd_addr_end(addr, end); | |
509 | if (pmd_none_or_clear_bad(pmd)) | |
510 | continue; | |
511 | zap_pte_range(tlb, pmd, addr, next, details); | |
512 | } while (pmd++, addr = next, addr != end); | |
513 | } | |
514 | ||
515 | static inline void zap_pud_range(struct mmu_gather *tlb, pgd_t *pgd, | |
516 | unsigned long addr, unsigned long end, | |
517 | struct zap_details *details) | |
518 | { | |
519 | pud_t *pud; | |
520 | unsigned long next; | |
521 | ||
522 | pud = pud_offset(pgd, addr); | |
523 | do { | |
524 | next = pud_addr_end(addr, end); | |
525 | if (pud_none_or_clear_bad(pud)) | |
526 | continue; | |
527 | zap_pmd_range(tlb, pud, addr, next, details); | |
528 | } while (pud++, addr = next, addr != end); | |
529 | } | |
530 | ||
531 | static void unmap_page_range(struct mmu_gather *tlb, struct vm_area_struct *vma, | |
532 | unsigned long addr, unsigned long end, | |
533 | struct zap_details *details) | |
534 | { | |
535 | pgd_t *pgd; | |
536 | unsigned long next; | |
537 | ||
538 | if (details && !details->check_mapping && !details->nonlinear_vma) | |
539 | details = NULL; | |
540 | ||
541 | BUG_ON(addr >= end); | |
542 | tlb_start_vma(tlb, vma); | |
543 | pgd = pgd_offset(vma->vm_mm, addr); | |
544 | do { | |
545 | next = pgd_addr_end(addr, end); | |
546 | if (pgd_none_or_clear_bad(pgd)) | |
547 | continue; | |
548 | zap_pud_range(tlb, pgd, addr, next, details); | |
549 | } while (pgd++, addr = next, addr != end); | |
550 | tlb_end_vma(tlb, vma); | |
551 | } | |
552 | ||
553 | #ifdef CONFIG_PREEMPT | |
554 | # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE) | |
555 | #else | |
556 | /* No preempt: go for improved straight-line efficiency */ | |
557 | # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE) | |
558 | #endif | |
559 | ||
560 | /** | |
561 | * unmap_vmas - unmap a range of memory covered by a list of vma's | |
562 | * @tlbp: address of the caller's struct mmu_gather | |
563 | * @mm: the controlling mm_struct | |
564 | * @vma: the starting vma | |
565 | * @start_addr: virtual address at which to start unmapping | |
566 | * @end_addr: virtual address at which to end unmapping | |
567 | * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here | |
568 | * @details: details of nonlinear truncation or shared cache invalidation | |
569 | * | |
570 | * Returns the number of vma's which were covered by the unmapping. | |
571 | * | |
572 | * Unmap all pages in the vma list. Called under page_table_lock. | |
573 | * | |
574 | * We aim to not hold page_table_lock for too long (for scheduling latency | |
575 | * reasons). So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to | |
576 | * return the ending mmu_gather to the caller. | |
577 | * | |
578 | * Only addresses between `start' and `end' will be unmapped. | |
579 | * | |
580 | * The VMA list must be sorted in ascending virtual address order. | |
581 | * | |
582 | * unmap_vmas() assumes that the caller will flush the whole unmapped address | |
583 | * range after unmap_vmas() returns. So the only responsibility here is to | |
584 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() | |
585 | * drops the lock and schedules. | |
586 | */ | |
587 | int unmap_vmas(struct mmu_gather **tlbp, struct mm_struct *mm, | |
588 | struct vm_area_struct *vma, unsigned long start_addr, | |
589 | unsigned long end_addr, unsigned long *nr_accounted, | |
590 | struct zap_details *details) | |
591 | { | |
592 | unsigned long zap_bytes = ZAP_BLOCK_SIZE; | |
593 | unsigned long tlb_start = 0; /* For tlb_finish_mmu */ | |
594 | int tlb_start_valid = 0; | |
595 | int ret = 0; | |
596 | spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL; | |
597 | int fullmm = tlb_is_full_mm(*tlbp); | |
598 | ||
599 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) { | |
600 | unsigned long start; | |
601 | unsigned long end; | |
602 | ||
603 | start = max(vma->vm_start, start_addr); | |
604 | if (start >= vma->vm_end) | |
605 | continue; | |
606 | end = min(vma->vm_end, end_addr); | |
607 | if (end <= vma->vm_start) | |
608 | continue; | |
609 | ||
610 | if (vma->vm_flags & VM_ACCOUNT) | |
611 | *nr_accounted += (end - start) >> PAGE_SHIFT; | |
612 | ||
613 | ret++; | |
614 | while (start != end) { | |
615 | unsigned long block; | |
616 | ||
617 | if (!tlb_start_valid) { | |
618 | tlb_start = start; | |
619 | tlb_start_valid = 1; | |
620 | } | |
621 | ||
622 | if (is_vm_hugetlb_page(vma)) { | |
623 | block = end - start; | |
624 | unmap_hugepage_range(vma, start, end); | |
625 | } else { | |
626 | block = min(zap_bytes, end - start); | |
627 | unmap_page_range(*tlbp, vma, start, | |
628 | start + block, details); | |
629 | } | |
630 | ||
631 | start += block; | |
632 | zap_bytes -= block; | |
633 | if ((long)zap_bytes > 0) | |
634 | continue; | |
635 | ||
636 | tlb_finish_mmu(*tlbp, tlb_start, start); | |
637 | ||
638 | if (need_resched() || | |
639 | need_lockbreak(&mm->page_table_lock) || | |
640 | (i_mmap_lock && need_lockbreak(i_mmap_lock))) { | |
641 | if (i_mmap_lock) { | |
642 | /* must reset count of rss freed */ | |
643 | *tlbp = tlb_gather_mmu(mm, fullmm); | |
644 | details->break_addr = start; | |
645 | goto out; | |
646 | } | |
647 | spin_unlock(&mm->page_table_lock); | |
648 | cond_resched(); | |
649 | spin_lock(&mm->page_table_lock); | |
650 | } | |
651 | ||
652 | *tlbp = tlb_gather_mmu(mm, fullmm); | |
653 | tlb_start_valid = 0; | |
654 | zap_bytes = ZAP_BLOCK_SIZE; | |
655 | } | |
656 | } | |
657 | out: | |
658 | return ret; | |
659 | } | |
660 | ||
661 | /** | |
662 | * zap_page_range - remove user pages in a given range | |
663 | * @vma: vm_area_struct holding the applicable pages | |
664 | * @address: starting address of pages to zap | |
665 | * @size: number of bytes to zap | |
666 | * @details: details of nonlinear truncation or shared cache invalidation | |
667 | */ | |
668 | void zap_page_range(struct vm_area_struct *vma, unsigned long address, | |
669 | unsigned long size, struct zap_details *details) | |
670 | { | |
671 | struct mm_struct *mm = vma->vm_mm; | |
672 | struct mmu_gather *tlb; | |
673 | unsigned long end = address + size; | |
674 | unsigned long nr_accounted = 0; | |
675 | ||
676 | if (is_vm_hugetlb_page(vma)) { | |
677 | zap_hugepage_range(vma, address, size); | |
678 | return; | |
679 | } | |
680 | ||
681 | lru_add_drain(); | |
682 | spin_lock(&mm->page_table_lock); | |
683 | tlb = tlb_gather_mmu(mm, 0); | |
684 | unmap_vmas(&tlb, mm, vma, address, end, &nr_accounted, details); | |
685 | tlb_finish_mmu(tlb, address, end); | |
686 | spin_unlock(&mm->page_table_lock); | |
687 | } | |
688 | ||
689 | /* | |
690 | * Do a quick page-table lookup for a single page. | |
691 | * mm->page_table_lock must be held. | |
692 | */ | |
693 | static struct page * | |
694 | __follow_page(struct mm_struct *mm, unsigned long address, int read, int write) | |
695 | { | |
696 | pgd_t *pgd; | |
697 | pud_t *pud; | |
698 | pmd_t *pmd; | |
699 | pte_t *ptep, pte; | |
700 | unsigned long pfn; | |
701 | struct page *page; | |
702 | ||
703 | page = follow_huge_addr(mm, address, write); | |
704 | if (! IS_ERR(page)) | |
705 | return page; | |
706 | ||
707 | pgd = pgd_offset(mm, address); | |
708 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
709 | goto out; | |
710 | ||
711 | pud = pud_offset(pgd, address); | |
712 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) | |
713 | goto out; | |
714 | ||
715 | pmd = pmd_offset(pud, address); | |
716 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) | |
717 | goto out; | |
718 | if (pmd_huge(*pmd)) | |
719 | return follow_huge_pmd(mm, address, pmd, write); | |
720 | ||
721 | ptep = pte_offset_map(pmd, address); | |
722 | if (!ptep) | |
723 | goto out; | |
724 | ||
725 | pte = *ptep; | |
726 | pte_unmap(ptep); | |
727 | if (pte_present(pte)) { | |
728 | if (write && !pte_write(pte)) | |
729 | goto out; | |
730 | if (read && !pte_read(pte)) | |
731 | goto out; | |
732 | pfn = pte_pfn(pte); | |
733 | if (pfn_valid(pfn)) { | |
734 | page = pfn_to_page(pfn); | |
735 | if (write && !pte_dirty(pte) && !PageDirty(page)) | |
736 | set_page_dirty(page); | |
737 | mark_page_accessed(page); | |
738 | return page; | |
739 | } | |
740 | } | |
741 | ||
742 | out: | |
743 | return NULL; | |
744 | } | |
745 | ||
746 | struct page * | |
747 | follow_page(struct mm_struct *mm, unsigned long address, int write) | |
748 | { | |
749 | return __follow_page(mm, address, /*read*/0, write); | |
750 | } | |
751 | ||
752 | int | |
753 | check_user_page_readable(struct mm_struct *mm, unsigned long address) | |
754 | { | |
755 | return __follow_page(mm, address, /*read*/1, /*write*/0) != NULL; | |
756 | } | |
757 | ||
758 | EXPORT_SYMBOL(check_user_page_readable); | |
759 | ||
760 | /* | |
761 | * Given a physical address, is there a useful struct page pointing to | |
762 | * it? This may become more complex in the future if we start dealing | |
763 | * with IO-aperture pages for direct-IO. | |
764 | */ | |
765 | ||
766 | static inline struct page *get_page_map(struct page *page) | |
767 | { | |
768 | if (!pfn_valid(page_to_pfn(page))) | |
769 | return NULL; | |
770 | return page; | |
771 | } | |
772 | ||
773 | ||
774 | static inline int | |
775 | untouched_anonymous_page(struct mm_struct* mm, struct vm_area_struct *vma, | |
776 | unsigned long address) | |
777 | { | |
778 | pgd_t *pgd; | |
779 | pud_t *pud; | |
780 | pmd_t *pmd; | |
781 | ||
782 | /* Check if the vma is for an anonymous mapping. */ | |
783 | if (vma->vm_ops && vma->vm_ops->nopage) | |
784 | return 0; | |
785 | ||
786 | /* Check if page directory entry exists. */ | |
787 | pgd = pgd_offset(mm, address); | |
788 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
789 | return 1; | |
790 | ||
791 | pud = pud_offset(pgd, address); | |
792 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) | |
793 | return 1; | |
794 | ||
795 | /* Check if page middle directory entry exists. */ | |
796 | pmd = pmd_offset(pud, address); | |
797 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) | |
798 | return 1; | |
799 | ||
800 | /* There is a pte slot for 'address' in 'mm'. */ | |
801 | return 0; | |
802 | } | |
803 | ||
804 | ||
805 | int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, | |
806 | unsigned long start, int len, int write, int force, | |
807 | struct page **pages, struct vm_area_struct **vmas) | |
808 | { | |
809 | int i; | |
810 | unsigned int flags; | |
811 | ||
812 | /* | |
813 | * Require read or write permissions. | |
814 | * If 'force' is set, we only require the "MAY" flags. | |
815 | */ | |
816 | flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); | |
817 | flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); | |
818 | i = 0; | |
819 | ||
820 | do { | |
821 | struct vm_area_struct * vma; | |
822 | ||
823 | vma = find_extend_vma(mm, start); | |
824 | if (!vma && in_gate_area(tsk, start)) { | |
825 | unsigned long pg = start & PAGE_MASK; | |
826 | struct vm_area_struct *gate_vma = get_gate_vma(tsk); | |
827 | pgd_t *pgd; | |
828 | pud_t *pud; | |
829 | pmd_t *pmd; | |
830 | pte_t *pte; | |
831 | if (write) /* user gate pages are read-only */ | |
832 | return i ? : -EFAULT; | |
833 | if (pg > TASK_SIZE) | |
834 | pgd = pgd_offset_k(pg); | |
835 | else | |
836 | pgd = pgd_offset_gate(mm, pg); | |
837 | BUG_ON(pgd_none(*pgd)); | |
838 | pud = pud_offset(pgd, pg); | |
839 | BUG_ON(pud_none(*pud)); | |
840 | pmd = pmd_offset(pud, pg); | |
841 | BUG_ON(pmd_none(*pmd)); | |
842 | pte = pte_offset_map(pmd, pg); | |
843 | BUG_ON(pte_none(*pte)); | |
844 | if (pages) { | |
845 | pages[i] = pte_page(*pte); | |
846 | get_page(pages[i]); | |
847 | } | |
848 | pte_unmap(pte); | |
849 | if (vmas) | |
850 | vmas[i] = gate_vma; | |
851 | i++; | |
852 | start += PAGE_SIZE; | |
853 | len--; | |
854 | continue; | |
855 | } | |
856 | ||
857 | if (!vma || (vma->vm_flags & VM_IO) | |
858 | || !(flags & vma->vm_flags)) | |
859 | return i ? : -EFAULT; | |
860 | ||
861 | if (is_vm_hugetlb_page(vma)) { | |
862 | i = follow_hugetlb_page(mm, vma, pages, vmas, | |
863 | &start, &len, i); | |
864 | continue; | |
865 | } | |
866 | spin_lock(&mm->page_table_lock); | |
867 | do { | |
868 | struct page *map; | |
869 | int lookup_write = write; | |
870 | ||
871 | cond_resched_lock(&mm->page_table_lock); | |
872 | while (!(map = follow_page(mm, start, lookup_write))) { | |
873 | /* | |
874 | * Shortcut for anonymous pages. We don't want | |
875 | * to force the creation of pages tables for | |
876 | * insanly big anonymously mapped areas that | |
877 | * nobody touched so far. This is important | |
878 | * for doing a core dump for these mappings. | |
879 | */ | |
880 | if (!lookup_write && | |
881 | untouched_anonymous_page(mm,vma,start)) { | |
882 | map = ZERO_PAGE(start); | |
883 | break; | |
884 | } | |
885 | spin_unlock(&mm->page_table_lock); | |
886 | switch (handle_mm_fault(mm,vma,start,write)) { | |
887 | case VM_FAULT_MINOR: | |
888 | tsk->min_flt++; | |
889 | break; | |
890 | case VM_FAULT_MAJOR: | |
891 | tsk->maj_flt++; | |
892 | break; | |
893 | case VM_FAULT_SIGBUS: | |
894 | return i ? i : -EFAULT; | |
895 | case VM_FAULT_OOM: | |
896 | return i ? i : -ENOMEM; | |
897 | default: | |
898 | BUG(); | |
899 | } | |
900 | /* | |
901 | * Now that we have performed a write fault | |
902 | * and surely no longer have a shared page we | |
903 | * shouldn't write, we shouldn't ignore an | |
904 | * unwritable page in the page table if | |
905 | * we are forcing write access. | |
906 | */ | |
907 | lookup_write = write && !force; | |
908 | spin_lock(&mm->page_table_lock); | |
909 | } | |
910 | if (pages) { | |
911 | pages[i] = get_page_map(map); | |
912 | if (!pages[i]) { | |
913 | spin_unlock(&mm->page_table_lock); | |
914 | while (i--) | |
915 | page_cache_release(pages[i]); | |
916 | i = -EFAULT; | |
917 | goto out; | |
918 | } | |
919 | flush_dcache_page(pages[i]); | |
920 | if (!PageReserved(pages[i])) | |
921 | page_cache_get(pages[i]); | |
922 | } | |
923 | if (vmas) | |
924 | vmas[i] = vma; | |
925 | i++; | |
926 | start += PAGE_SIZE; | |
927 | len--; | |
928 | } while(len && start < vma->vm_end); | |
929 | spin_unlock(&mm->page_table_lock); | |
930 | } while(len); | |
931 | out: | |
932 | return i; | |
933 | } | |
934 | ||
935 | EXPORT_SYMBOL(get_user_pages); | |
936 | ||
937 | static int zeromap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
938 | unsigned long addr, unsigned long end, pgprot_t prot) | |
939 | { | |
940 | pte_t *pte; | |
941 | ||
942 | pte = pte_alloc_map(mm, pmd, addr); | |
943 | if (!pte) | |
944 | return -ENOMEM; | |
945 | do { | |
946 | pte_t zero_pte = pte_wrprotect(mk_pte(ZERO_PAGE(addr), prot)); | |
947 | BUG_ON(!pte_none(*pte)); | |
948 | set_pte_at(mm, addr, pte, zero_pte); | |
949 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
950 | pte_unmap(pte - 1); | |
951 | return 0; | |
952 | } | |
953 | ||
954 | static inline int zeromap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
955 | unsigned long addr, unsigned long end, pgprot_t prot) | |
956 | { | |
957 | pmd_t *pmd; | |
958 | unsigned long next; | |
959 | ||
960 | pmd = pmd_alloc(mm, pud, addr); | |
961 | if (!pmd) | |
962 | return -ENOMEM; | |
963 | do { | |
964 | next = pmd_addr_end(addr, end); | |
965 | if (zeromap_pte_range(mm, pmd, addr, next, prot)) | |
966 | return -ENOMEM; | |
967 | } while (pmd++, addr = next, addr != end); | |
968 | return 0; | |
969 | } | |
970 | ||
971 | static inline int zeromap_pud_range(struct mm_struct *mm, pgd_t *pgd, | |
972 | unsigned long addr, unsigned long end, pgprot_t prot) | |
973 | { | |
974 | pud_t *pud; | |
975 | unsigned long next; | |
976 | ||
977 | pud = pud_alloc(mm, pgd, addr); | |
978 | if (!pud) | |
979 | return -ENOMEM; | |
980 | do { | |
981 | next = pud_addr_end(addr, end); | |
982 | if (zeromap_pmd_range(mm, pud, addr, next, prot)) | |
983 | return -ENOMEM; | |
984 | } while (pud++, addr = next, addr != end); | |
985 | return 0; | |
986 | } | |
987 | ||
988 | int zeromap_page_range(struct vm_area_struct *vma, | |
989 | unsigned long addr, unsigned long size, pgprot_t prot) | |
990 | { | |
991 | pgd_t *pgd; | |
992 | unsigned long next; | |
993 | unsigned long end = addr + size; | |
994 | struct mm_struct *mm = vma->vm_mm; | |
995 | int err; | |
996 | ||
997 | BUG_ON(addr >= end); | |
998 | pgd = pgd_offset(mm, addr); | |
999 | flush_cache_range(vma, addr, end); | |
1000 | spin_lock(&mm->page_table_lock); | |
1001 | do { | |
1002 | next = pgd_addr_end(addr, end); | |
1003 | err = zeromap_pud_range(mm, pgd, addr, next, prot); | |
1004 | if (err) | |
1005 | break; | |
1006 | } while (pgd++, addr = next, addr != end); | |
1007 | spin_unlock(&mm->page_table_lock); | |
1008 | return err; | |
1009 | } | |
1010 | ||
1011 | /* | |
1012 | * maps a range of physical memory into the requested pages. the old | |
1013 | * mappings are removed. any references to nonexistent pages results | |
1014 | * in null mappings (currently treated as "copy-on-access") | |
1015 | */ | |
1016 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
1017 | unsigned long addr, unsigned long end, | |
1018 | unsigned long pfn, pgprot_t prot) | |
1019 | { | |
1020 | pte_t *pte; | |
1021 | ||
1022 | pte = pte_alloc_map(mm, pmd, addr); | |
1023 | if (!pte) | |
1024 | return -ENOMEM; | |
1025 | do { | |
1026 | BUG_ON(!pte_none(*pte)); | |
1027 | if (!pfn_valid(pfn) || PageReserved(pfn_to_page(pfn))) | |
1028 | set_pte_at(mm, addr, pte, pfn_pte(pfn, prot)); | |
1029 | pfn++; | |
1030 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
1031 | pte_unmap(pte - 1); | |
1032 | return 0; | |
1033 | } | |
1034 | ||
1035 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
1036 | unsigned long addr, unsigned long end, | |
1037 | unsigned long pfn, pgprot_t prot) | |
1038 | { | |
1039 | pmd_t *pmd; | |
1040 | unsigned long next; | |
1041 | ||
1042 | pfn -= addr >> PAGE_SHIFT; | |
1043 | pmd = pmd_alloc(mm, pud, addr); | |
1044 | if (!pmd) | |
1045 | return -ENOMEM; | |
1046 | do { | |
1047 | next = pmd_addr_end(addr, end); | |
1048 | if (remap_pte_range(mm, pmd, addr, next, | |
1049 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1050 | return -ENOMEM; | |
1051 | } while (pmd++, addr = next, addr != end); | |
1052 | return 0; | |
1053 | } | |
1054 | ||
1055 | static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, | |
1056 | unsigned long addr, unsigned long end, | |
1057 | unsigned long pfn, pgprot_t prot) | |
1058 | { | |
1059 | pud_t *pud; | |
1060 | unsigned long next; | |
1061 | ||
1062 | pfn -= addr >> PAGE_SHIFT; | |
1063 | pud = pud_alloc(mm, pgd, addr); | |
1064 | if (!pud) | |
1065 | return -ENOMEM; | |
1066 | do { | |
1067 | next = pud_addr_end(addr, end); | |
1068 | if (remap_pmd_range(mm, pud, addr, next, | |
1069 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1070 | return -ENOMEM; | |
1071 | } while (pud++, addr = next, addr != end); | |
1072 | return 0; | |
1073 | } | |
1074 | ||
1075 | /* Note: this is only safe if the mm semaphore is held when called. */ | |
1076 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, | |
1077 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
1078 | { | |
1079 | pgd_t *pgd; | |
1080 | unsigned long next; | |
1081 | unsigned long end = addr + size; | |
1082 | struct mm_struct *mm = vma->vm_mm; | |
1083 | int err; | |
1084 | ||
1085 | /* | |
1086 | * Physically remapped pages are special. Tell the | |
1087 | * rest of the world about it: | |
1088 | * VM_IO tells people not to look at these pages | |
1089 | * (accesses can have side effects). | |
1090 | * VM_RESERVED tells swapout not to try to touch | |
1091 | * this region. | |
1092 | */ | |
1093 | vma->vm_flags |= VM_IO | VM_RESERVED; | |
1094 | ||
1095 | BUG_ON(addr >= end); | |
1096 | pfn -= addr >> PAGE_SHIFT; | |
1097 | pgd = pgd_offset(mm, addr); | |
1098 | flush_cache_range(vma, addr, end); | |
1099 | spin_lock(&mm->page_table_lock); | |
1100 | do { | |
1101 | next = pgd_addr_end(addr, end); | |
1102 | err = remap_pud_range(mm, pgd, addr, next, | |
1103 | pfn + (addr >> PAGE_SHIFT), prot); | |
1104 | if (err) | |
1105 | break; | |
1106 | } while (pgd++, addr = next, addr != end); | |
1107 | spin_unlock(&mm->page_table_lock); | |
1108 | return err; | |
1109 | } | |
1110 | EXPORT_SYMBOL(remap_pfn_range); | |
1111 | ||
1112 | /* | |
1113 | * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when | |
1114 | * servicing faults for write access. In the normal case, do always want | |
1115 | * pte_mkwrite. But get_user_pages can cause write faults for mappings | |
1116 | * that do not have writing enabled, when used by access_process_vm. | |
1117 | */ | |
1118 | static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma) | |
1119 | { | |
1120 | if (likely(vma->vm_flags & VM_WRITE)) | |
1121 | pte = pte_mkwrite(pte); | |
1122 | return pte; | |
1123 | } | |
1124 | ||
1125 | /* | |
1126 | * We hold the mm semaphore for reading and vma->vm_mm->page_table_lock | |
1127 | */ | |
1128 | static inline void break_cow(struct vm_area_struct * vma, struct page * new_page, unsigned long address, | |
1129 | pte_t *page_table) | |
1130 | { | |
1131 | pte_t entry; | |
1132 | ||
1133 | entry = maybe_mkwrite(pte_mkdirty(mk_pte(new_page, vma->vm_page_prot)), | |
1134 | vma); | |
1135 | ptep_establish(vma, address, page_table, entry); | |
1136 | update_mmu_cache(vma, address, entry); | |
1137 | lazy_mmu_prot_update(entry); | |
1138 | } | |
1139 | ||
1140 | /* | |
1141 | * This routine handles present pages, when users try to write | |
1142 | * to a shared page. It is done by copying the page to a new address | |
1143 | * and decrementing the shared-page counter for the old page. | |
1144 | * | |
1145 | * Goto-purists beware: the only reason for goto's here is that it results | |
1146 | * in better assembly code.. The "default" path will see no jumps at all. | |
1147 | * | |
1148 | * Note that this routine assumes that the protection checks have been | |
1149 | * done by the caller (the low-level page fault routine in most cases). | |
1150 | * Thus we can safely just mark it writable once we've done any necessary | |
1151 | * COW. | |
1152 | * | |
1153 | * We also mark the page dirty at this point even though the page will | |
1154 | * change only once the write actually happens. This avoids a few races, | |
1155 | * and potentially makes it more efficient. | |
1156 | * | |
1157 | * We hold the mm semaphore and the page_table_lock on entry and exit | |
1158 | * with the page_table_lock released. | |
1159 | */ | |
1160 | static int do_wp_page(struct mm_struct *mm, struct vm_area_struct * vma, | |
1161 | unsigned long address, pte_t *page_table, pmd_t *pmd, pte_t pte) | |
1162 | { | |
1163 | struct page *old_page, *new_page; | |
1164 | unsigned long pfn = pte_pfn(pte); | |
1165 | pte_t entry; | |
1166 | ||
1167 | if (unlikely(!pfn_valid(pfn))) { | |
1168 | /* | |
1169 | * This should really halt the system so it can be debugged or | |
1170 | * at least the kernel stops what it's doing before it corrupts | |
1171 | * data, but for the moment just pretend this is OOM. | |
1172 | */ | |
1173 | pte_unmap(page_table); | |
1174 | printk(KERN_ERR "do_wp_page: bogus page at address %08lx\n", | |
1175 | address); | |
1176 | spin_unlock(&mm->page_table_lock); | |
1177 | return VM_FAULT_OOM; | |
1178 | } | |
1179 | old_page = pfn_to_page(pfn); | |
1180 | ||
1181 | if (!TestSetPageLocked(old_page)) { | |
1182 | int reuse = can_share_swap_page(old_page); | |
1183 | unlock_page(old_page); | |
1184 | if (reuse) { | |
1185 | flush_cache_page(vma, address, pfn); | |
1186 | entry = maybe_mkwrite(pte_mkyoung(pte_mkdirty(pte)), | |
1187 | vma); | |
1188 | ptep_set_access_flags(vma, address, page_table, entry, 1); | |
1189 | update_mmu_cache(vma, address, entry); | |
1190 | lazy_mmu_prot_update(entry); | |
1191 | pte_unmap(page_table); | |
1192 | spin_unlock(&mm->page_table_lock); | |
1193 | return VM_FAULT_MINOR; | |
1194 | } | |
1195 | } | |
1196 | pte_unmap(page_table); | |
1197 | ||
1198 | /* | |
1199 | * Ok, we need to copy. Oh, well.. | |
1200 | */ | |
1201 | if (!PageReserved(old_page)) | |
1202 | page_cache_get(old_page); | |
1203 | spin_unlock(&mm->page_table_lock); | |
1204 | ||
1205 | if (unlikely(anon_vma_prepare(vma))) | |
1206 | goto no_new_page; | |
1207 | if (old_page == ZERO_PAGE(address)) { | |
1208 | new_page = alloc_zeroed_user_highpage(vma, address); | |
1209 | if (!new_page) | |
1210 | goto no_new_page; | |
1211 | } else { | |
1212 | new_page = alloc_page_vma(GFP_HIGHUSER, vma, address); | |
1213 | if (!new_page) | |
1214 | goto no_new_page; | |
1215 | copy_user_highpage(new_page, old_page, address); | |
1216 | } | |
1217 | /* | |
1218 | * Re-check the pte - we dropped the lock | |
1219 | */ | |
1220 | spin_lock(&mm->page_table_lock); | |
1221 | page_table = pte_offset_map(pmd, address); | |
1222 | if (likely(pte_same(*page_table, pte))) { | |
1223 | if (PageAnon(old_page)) | |
1224 | dec_mm_counter(mm, anon_rss); | |
1225 | if (PageReserved(old_page)) | |
1226 | inc_mm_counter(mm, rss); | |
1227 | else | |
1228 | page_remove_rmap(old_page); | |
1229 | flush_cache_page(vma, address, pfn); | |
1230 | break_cow(vma, new_page, address, page_table); | |
1231 | lru_cache_add_active(new_page); | |
1232 | page_add_anon_rmap(new_page, vma, address); | |
1233 | ||
1234 | /* Free the old page.. */ | |
1235 | new_page = old_page; | |
1236 | } | |
1237 | pte_unmap(page_table); | |
1238 | page_cache_release(new_page); | |
1239 | page_cache_release(old_page); | |
1240 | spin_unlock(&mm->page_table_lock); | |
1241 | return VM_FAULT_MINOR; | |
1242 | ||
1243 | no_new_page: | |
1244 | page_cache_release(old_page); | |
1245 | return VM_FAULT_OOM; | |
1246 | } | |
1247 | ||
1248 | /* | |
1249 | * Helper functions for unmap_mapping_range(). | |
1250 | * | |
1251 | * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __ | |
1252 | * | |
1253 | * We have to restart searching the prio_tree whenever we drop the lock, | |
1254 | * since the iterator is only valid while the lock is held, and anyway | |
1255 | * a later vma might be split and reinserted earlier while lock dropped. | |
1256 | * | |
1257 | * The list of nonlinear vmas could be handled more efficiently, using | |
1258 | * a placeholder, but handle it in the same way until a need is shown. | |
1259 | * It is important to search the prio_tree before nonlinear list: a vma | |
1260 | * may become nonlinear and be shifted from prio_tree to nonlinear list | |
1261 | * while the lock is dropped; but never shifted from list to prio_tree. | |
1262 | * | |
1263 | * In order to make forward progress despite restarting the search, | |
1264 | * vm_truncate_count is used to mark a vma as now dealt with, so we can | |
1265 | * quickly skip it next time around. Since the prio_tree search only | |
1266 | * shows us those vmas affected by unmapping the range in question, we | |
1267 | * can't efficiently keep all vmas in step with mapping->truncate_count: | |
1268 | * so instead reset them all whenever it wraps back to 0 (then go to 1). | |
1269 | * mapping->truncate_count and vma->vm_truncate_count are protected by | |
1270 | * i_mmap_lock. | |
1271 | * | |
1272 | * In order to make forward progress despite repeatedly restarting some | |
1273 | * large vma, note the break_addr set by unmap_vmas when it breaks out: | |
1274 | * and restart from that address when we reach that vma again. It might | |
1275 | * have been split or merged, shrunk or extended, but never shifted: so | |
1276 | * restart_addr remains valid so long as it remains in the vma's range. | |
1277 | * unmap_mapping_range forces truncate_count to leap over page-aligned | |
1278 | * values so we can save vma's restart_addr in its truncate_count field. | |
1279 | */ | |
1280 | #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK)) | |
1281 | ||
1282 | static void reset_vma_truncate_counts(struct address_space *mapping) | |
1283 | { | |
1284 | struct vm_area_struct *vma; | |
1285 | struct prio_tree_iter iter; | |
1286 | ||
1287 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX) | |
1288 | vma->vm_truncate_count = 0; | |
1289 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) | |
1290 | vma->vm_truncate_count = 0; | |
1291 | } | |
1292 | ||
1293 | static int unmap_mapping_range_vma(struct vm_area_struct *vma, | |
1294 | unsigned long start_addr, unsigned long end_addr, | |
1295 | struct zap_details *details) | |
1296 | { | |
1297 | unsigned long restart_addr; | |
1298 | int need_break; | |
1299 | ||
1300 | again: | |
1301 | restart_addr = vma->vm_truncate_count; | |
1302 | if (is_restart_addr(restart_addr) && start_addr < restart_addr) { | |
1303 | start_addr = restart_addr; | |
1304 | if (start_addr >= end_addr) { | |
1305 | /* Top of vma has been split off since last time */ | |
1306 | vma->vm_truncate_count = details->truncate_count; | |
1307 | return 0; | |
1308 | } | |
1309 | } | |
1310 | ||
1311 | details->break_addr = end_addr; | |
1312 | zap_page_range(vma, start_addr, end_addr - start_addr, details); | |
1313 | ||
1314 | /* | |
1315 | * We cannot rely on the break test in unmap_vmas: | |
1316 | * on the one hand, we don't want to restart our loop | |
1317 | * just because that broke out for the page_table_lock; | |
1318 | * on the other hand, it does no test when vma is small. | |
1319 | */ | |
1320 | need_break = need_resched() || | |
1321 | need_lockbreak(details->i_mmap_lock); | |
1322 | ||
1323 | if (details->break_addr >= end_addr) { | |
1324 | /* We have now completed this vma: mark it so */ | |
1325 | vma->vm_truncate_count = details->truncate_count; | |
1326 | if (!need_break) | |
1327 | return 0; | |
1328 | } else { | |
1329 | /* Note restart_addr in vma's truncate_count field */ | |
1330 | vma->vm_truncate_count = details->break_addr; | |
1331 | if (!need_break) | |
1332 | goto again; | |
1333 | } | |
1334 | ||
1335 | spin_unlock(details->i_mmap_lock); | |
1336 | cond_resched(); | |
1337 | spin_lock(details->i_mmap_lock); | |
1338 | return -EINTR; | |
1339 | } | |
1340 | ||
1341 | static inline void unmap_mapping_range_tree(struct prio_tree_root *root, | |
1342 | struct zap_details *details) | |
1343 | { | |
1344 | struct vm_area_struct *vma; | |
1345 | struct prio_tree_iter iter; | |
1346 | pgoff_t vba, vea, zba, zea; | |
1347 | ||
1348 | restart: | |
1349 | vma_prio_tree_foreach(vma, &iter, root, | |
1350 | details->first_index, details->last_index) { | |
1351 | /* Skip quickly over those we have already dealt with */ | |
1352 | if (vma->vm_truncate_count == details->truncate_count) | |
1353 | continue; | |
1354 | ||
1355 | vba = vma->vm_pgoff; | |
1356 | vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1; | |
1357 | /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */ | |
1358 | zba = details->first_index; | |
1359 | if (zba < vba) | |
1360 | zba = vba; | |
1361 | zea = details->last_index; | |
1362 | if (zea > vea) | |
1363 | zea = vea; | |
1364 | ||
1365 | if (unmap_mapping_range_vma(vma, | |
1366 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, | |
1367 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, | |
1368 | details) < 0) | |
1369 | goto restart; | |
1370 | } | |
1371 | } | |
1372 | ||
1373 | static inline void unmap_mapping_range_list(struct list_head *head, | |
1374 | struct zap_details *details) | |
1375 | { | |
1376 | struct vm_area_struct *vma; | |
1377 | ||
1378 | /* | |
1379 | * In nonlinear VMAs there is no correspondence between virtual address | |
1380 | * offset and file offset. So we must perform an exhaustive search | |
1381 | * across *all* the pages in each nonlinear VMA, not just the pages | |
1382 | * whose virtual address lies outside the file truncation point. | |
1383 | */ | |
1384 | restart: | |
1385 | list_for_each_entry(vma, head, shared.vm_set.list) { | |
1386 | /* Skip quickly over those we have already dealt with */ | |
1387 | if (vma->vm_truncate_count == details->truncate_count) | |
1388 | continue; | |
1389 | details->nonlinear_vma = vma; | |
1390 | if (unmap_mapping_range_vma(vma, vma->vm_start, | |
1391 | vma->vm_end, details) < 0) | |
1392 | goto restart; | |
1393 | } | |
1394 | } | |
1395 | ||
1396 | /** | |
1397 | * unmap_mapping_range - unmap the portion of all mmaps | |
1398 | * in the specified address_space corresponding to the specified | |
1399 | * page range in the underlying file. | |
1400 | * @address_space: the address space containing mmaps to be unmapped. | |
1401 | * @holebegin: byte in first page to unmap, relative to the start of | |
1402 | * the underlying file. This will be rounded down to a PAGE_SIZE | |
1403 | * boundary. Note that this is different from vmtruncate(), which | |
1404 | * must keep the partial page. In contrast, we must get rid of | |
1405 | * partial pages. | |
1406 | * @holelen: size of prospective hole in bytes. This will be rounded | |
1407 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the | |
1408 | * end of the file. | |
1409 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; | |
1410 | * but 0 when invalidating pagecache, don't throw away private data. | |
1411 | */ | |
1412 | void unmap_mapping_range(struct address_space *mapping, | |
1413 | loff_t const holebegin, loff_t const holelen, int even_cows) | |
1414 | { | |
1415 | struct zap_details details; | |
1416 | pgoff_t hba = holebegin >> PAGE_SHIFT; | |
1417 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1418 | ||
1419 | /* Check for overflow. */ | |
1420 | if (sizeof(holelen) > sizeof(hlen)) { | |
1421 | long long holeend = | |
1422 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1423 | if (holeend & ~(long long)ULONG_MAX) | |
1424 | hlen = ULONG_MAX - hba + 1; | |
1425 | } | |
1426 | ||
1427 | details.check_mapping = even_cows? NULL: mapping; | |
1428 | details.nonlinear_vma = NULL; | |
1429 | details.first_index = hba; | |
1430 | details.last_index = hba + hlen - 1; | |
1431 | if (details.last_index < details.first_index) | |
1432 | details.last_index = ULONG_MAX; | |
1433 | details.i_mmap_lock = &mapping->i_mmap_lock; | |
1434 | ||
1435 | spin_lock(&mapping->i_mmap_lock); | |
1436 | ||
1437 | /* serialize i_size write against truncate_count write */ | |
1438 | smp_wmb(); | |
1439 | /* Protect against page faults, and endless unmapping loops */ | |
1440 | mapping->truncate_count++; | |
1441 | /* | |
1442 | * For archs where spin_lock has inclusive semantics like ia64 | |
1443 | * this smp_mb() will prevent to read pagetable contents | |
1444 | * before the truncate_count increment is visible to | |
1445 | * other cpus. | |
1446 | */ | |
1447 | smp_mb(); | |
1448 | if (unlikely(is_restart_addr(mapping->truncate_count))) { | |
1449 | if (mapping->truncate_count == 0) | |
1450 | reset_vma_truncate_counts(mapping); | |
1451 | mapping->truncate_count++; | |
1452 | } | |
1453 | details.truncate_count = mapping->truncate_count; | |
1454 | ||
1455 | if (unlikely(!prio_tree_empty(&mapping->i_mmap))) | |
1456 | unmap_mapping_range_tree(&mapping->i_mmap, &details); | |
1457 | if (unlikely(!list_empty(&mapping->i_mmap_nonlinear))) | |
1458 | unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details); | |
1459 | spin_unlock(&mapping->i_mmap_lock); | |
1460 | } | |
1461 | EXPORT_SYMBOL(unmap_mapping_range); | |
1462 | ||
1463 | /* | |
1464 | * Handle all mappings that got truncated by a "truncate()" | |
1465 | * system call. | |
1466 | * | |
1467 | * NOTE! We have to be ready to update the memory sharing | |
1468 | * between the file and the memory map for a potential last | |
1469 | * incomplete page. Ugly, but necessary. | |
1470 | */ | |
1471 | int vmtruncate(struct inode * inode, loff_t offset) | |
1472 | { | |
1473 | struct address_space *mapping = inode->i_mapping; | |
1474 | unsigned long limit; | |
1475 | ||
1476 | if (inode->i_size < offset) | |
1477 | goto do_expand; | |
1478 | /* | |
1479 | * truncation of in-use swapfiles is disallowed - it would cause | |
1480 | * subsequent swapout to scribble on the now-freed blocks. | |
1481 | */ | |
1482 | if (IS_SWAPFILE(inode)) | |
1483 | goto out_busy; | |
1484 | i_size_write(inode, offset); | |
1485 | unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1); | |
1486 | truncate_inode_pages(mapping, offset); | |
1487 | goto out_truncate; | |
1488 | ||
1489 | do_expand: | |
1490 | limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; | |
1491 | if (limit != RLIM_INFINITY && offset > limit) | |
1492 | goto out_sig; | |
1493 | if (offset > inode->i_sb->s_maxbytes) | |
1494 | goto out_big; | |
1495 | i_size_write(inode, offset); | |
1496 | ||
1497 | out_truncate: | |
1498 | if (inode->i_op && inode->i_op->truncate) | |
1499 | inode->i_op->truncate(inode); | |
1500 | return 0; | |
1501 | out_sig: | |
1502 | send_sig(SIGXFSZ, current, 0); | |
1503 | out_big: | |
1504 | return -EFBIG; | |
1505 | out_busy: | |
1506 | return -ETXTBSY; | |
1507 | } | |
1508 | ||
1509 | EXPORT_SYMBOL(vmtruncate); | |
1510 | ||
1511 | /* | |
1512 | * Primitive swap readahead code. We simply read an aligned block of | |
1513 | * (1 << page_cluster) entries in the swap area. This method is chosen | |
1514 | * because it doesn't cost us any seek time. We also make sure to queue | |
1515 | * the 'original' request together with the readahead ones... | |
1516 | * | |
1517 | * This has been extended to use the NUMA policies from the mm triggering | |
1518 | * the readahead. | |
1519 | * | |
1520 | * Caller must hold down_read on the vma->vm_mm if vma is not NULL. | |
1521 | */ | |
1522 | void swapin_readahead(swp_entry_t entry, unsigned long addr,struct vm_area_struct *vma) | |
1523 | { | |
1524 | #ifdef CONFIG_NUMA | |
1525 | struct vm_area_struct *next_vma = vma ? vma->vm_next : NULL; | |
1526 | #endif | |
1527 | int i, num; | |
1528 | struct page *new_page; | |
1529 | unsigned long offset; | |
1530 | ||
1531 | /* | |
1532 | * Get the number of handles we should do readahead io to. | |
1533 | */ | |
1534 | num = valid_swaphandles(entry, &offset); | |
1535 | for (i = 0; i < num; offset++, i++) { | |
1536 | /* Ok, do the async read-ahead now */ | |
1537 | new_page = read_swap_cache_async(swp_entry(swp_type(entry), | |
1538 | offset), vma, addr); | |
1539 | if (!new_page) | |
1540 | break; | |
1541 | page_cache_release(new_page); | |
1542 | #ifdef CONFIG_NUMA | |
1543 | /* | |
1544 | * Find the next applicable VMA for the NUMA policy. | |
1545 | */ | |
1546 | addr += PAGE_SIZE; | |
1547 | if (addr == 0) | |
1548 | vma = NULL; | |
1549 | if (vma) { | |
1550 | if (addr >= vma->vm_end) { | |
1551 | vma = next_vma; | |
1552 | next_vma = vma ? vma->vm_next : NULL; | |
1553 | } | |
1554 | if (vma && addr < vma->vm_start) | |
1555 | vma = NULL; | |
1556 | } else { | |
1557 | if (next_vma && addr >= next_vma->vm_start) { | |
1558 | vma = next_vma; | |
1559 | next_vma = vma->vm_next; | |
1560 | } | |
1561 | } | |
1562 | #endif | |
1563 | } | |
1564 | lru_add_drain(); /* Push any new pages onto the LRU now */ | |
1565 | } | |
1566 | ||
1567 | /* | |
1568 | * We hold the mm semaphore and the page_table_lock on entry and | |
1569 | * should release the pagetable lock on exit.. | |
1570 | */ | |
1571 | static int do_swap_page(struct mm_struct * mm, | |
1572 | struct vm_area_struct * vma, unsigned long address, | |
1573 | pte_t *page_table, pmd_t *pmd, pte_t orig_pte, int write_access) | |
1574 | { | |
1575 | struct page *page; | |
1576 | swp_entry_t entry = pte_to_swp_entry(orig_pte); | |
1577 | pte_t pte; | |
1578 | int ret = VM_FAULT_MINOR; | |
1579 | ||
1580 | pte_unmap(page_table); | |
1581 | spin_unlock(&mm->page_table_lock); | |
1582 | page = lookup_swap_cache(entry); | |
1583 | if (!page) { | |
1584 | swapin_readahead(entry, address, vma); | |
1585 | page = read_swap_cache_async(entry, vma, address); | |
1586 | if (!page) { | |
1587 | /* | |
1588 | * Back out if somebody else faulted in this pte while | |
1589 | * we released the page table lock. | |
1590 | */ | |
1591 | spin_lock(&mm->page_table_lock); | |
1592 | page_table = pte_offset_map(pmd, address); | |
1593 | if (likely(pte_same(*page_table, orig_pte))) | |
1594 | ret = VM_FAULT_OOM; | |
1595 | else | |
1596 | ret = VM_FAULT_MINOR; | |
1597 | pte_unmap(page_table); | |
1598 | spin_unlock(&mm->page_table_lock); | |
1599 | goto out; | |
1600 | } | |
1601 | ||
1602 | /* Had to read the page from swap area: Major fault */ | |
1603 | ret = VM_FAULT_MAJOR; | |
1604 | inc_page_state(pgmajfault); | |
1605 | grab_swap_token(); | |
1606 | } | |
1607 | ||
1608 | mark_page_accessed(page); | |
1609 | lock_page(page); | |
1610 | ||
1611 | /* | |
1612 | * Back out if somebody else faulted in this pte while we | |
1613 | * released the page table lock. | |
1614 | */ | |
1615 | spin_lock(&mm->page_table_lock); | |
1616 | page_table = pte_offset_map(pmd, address); | |
1617 | if (unlikely(!pte_same(*page_table, orig_pte))) { | |
1618 | pte_unmap(page_table); | |
1619 | spin_unlock(&mm->page_table_lock); | |
1620 | unlock_page(page); | |
1621 | page_cache_release(page); | |
1622 | ret = VM_FAULT_MINOR; | |
1623 | goto out; | |
1624 | } | |
1625 | ||
1626 | /* The page isn't present yet, go ahead with the fault. */ | |
1627 | ||
1628 | swap_free(entry); | |
1629 | if (vm_swap_full()) | |
1630 | remove_exclusive_swap_page(page); | |
1631 | ||
1632 | inc_mm_counter(mm, rss); | |
1633 | pte = mk_pte(page, vma->vm_page_prot); | |
1634 | if (write_access && can_share_swap_page(page)) { | |
1635 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); | |
1636 | write_access = 0; | |
1637 | } | |
1638 | unlock_page(page); | |
1639 | ||
1640 | flush_icache_page(vma, page); | |
1641 | set_pte_at(mm, address, page_table, pte); | |
1642 | page_add_anon_rmap(page, vma, address); | |
1643 | ||
1644 | if (write_access) { | |
1645 | if (do_wp_page(mm, vma, address, | |
1646 | page_table, pmd, pte) == VM_FAULT_OOM) | |
1647 | ret = VM_FAULT_OOM; | |
1648 | goto out; | |
1649 | } | |
1650 | ||
1651 | /* No need to invalidate - it was non-present before */ | |
1652 | update_mmu_cache(vma, address, pte); | |
1653 | lazy_mmu_prot_update(pte); | |
1654 | pte_unmap(page_table); | |
1655 | spin_unlock(&mm->page_table_lock); | |
1656 | out: | |
1657 | return ret; | |
1658 | } | |
1659 | ||
1660 | /* | |
1661 | * We are called with the MM semaphore and page_table_lock | |
1662 | * spinlock held to protect against concurrent faults in | |
1663 | * multithreaded programs. | |
1664 | */ | |
1665 | static int | |
1666 | do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma, | |
1667 | pte_t *page_table, pmd_t *pmd, int write_access, | |
1668 | unsigned long addr) | |
1669 | { | |
1670 | pte_t entry; | |
1671 | struct page * page = ZERO_PAGE(addr); | |
1672 | ||
1673 | /* Read-only mapping of ZERO_PAGE. */ | |
1674 | entry = pte_wrprotect(mk_pte(ZERO_PAGE(addr), vma->vm_page_prot)); | |
1675 | ||
1676 | /* ..except if it's a write access */ | |
1677 | if (write_access) { | |
1678 | /* Allocate our own private page. */ | |
1679 | pte_unmap(page_table); | |
1680 | spin_unlock(&mm->page_table_lock); | |
1681 | ||
1682 | if (unlikely(anon_vma_prepare(vma))) | |
1683 | goto no_mem; | |
1684 | page = alloc_zeroed_user_highpage(vma, addr); | |
1685 | if (!page) | |
1686 | goto no_mem; | |
1687 | ||
1688 | spin_lock(&mm->page_table_lock); | |
1689 | page_table = pte_offset_map(pmd, addr); | |
1690 | ||
1691 | if (!pte_none(*page_table)) { | |
1692 | pte_unmap(page_table); | |
1693 | page_cache_release(page); | |
1694 | spin_unlock(&mm->page_table_lock); | |
1695 | goto out; | |
1696 | } | |
1697 | inc_mm_counter(mm, rss); | |
1698 | entry = maybe_mkwrite(pte_mkdirty(mk_pte(page, | |
1699 | vma->vm_page_prot)), | |
1700 | vma); | |
1701 | lru_cache_add_active(page); | |
1702 | SetPageReferenced(page); | |
1703 | page_add_anon_rmap(page, vma, addr); | |
1704 | } | |
1705 | ||
1706 | set_pte_at(mm, addr, page_table, entry); | |
1707 | pte_unmap(page_table); | |
1708 | ||
1709 | /* No need to invalidate - it was non-present before */ | |
1710 | update_mmu_cache(vma, addr, entry); | |
1711 | lazy_mmu_prot_update(entry); | |
1712 | spin_unlock(&mm->page_table_lock); | |
1713 | out: | |
1714 | return VM_FAULT_MINOR; | |
1715 | no_mem: | |
1716 | return VM_FAULT_OOM; | |
1717 | } | |
1718 | ||
1719 | /* | |
1720 | * do_no_page() tries to create a new page mapping. It aggressively | |
1721 | * tries to share with existing pages, but makes a separate copy if | |
1722 | * the "write_access" parameter is true in order to avoid the next | |
1723 | * page fault. | |
1724 | * | |
1725 | * As this is called only for pages that do not currently exist, we | |
1726 | * do not need to flush old virtual caches or the TLB. | |
1727 | * | |
1728 | * This is called with the MM semaphore held and the page table | |
1729 | * spinlock held. Exit with the spinlock released. | |
1730 | */ | |
1731 | static int | |
1732 | do_no_page(struct mm_struct *mm, struct vm_area_struct *vma, | |
1733 | unsigned long address, int write_access, pte_t *page_table, pmd_t *pmd) | |
1734 | { | |
1735 | struct page * new_page; | |
1736 | struct address_space *mapping = NULL; | |
1737 | pte_t entry; | |
1738 | unsigned int sequence = 0; | |
1739 | int ret = VM_FAULT_MINOR; | |
1740 | int anon = 0; | |
1741 | ||
1742 | if (!vma->vm_ops || !vma->vm_ops->nopage) | |
1743 | return do_anonymous_page(mm, vma, page_table, | |
1744 | pmd, write_access, address); | |
1745 | pte_unmap(page_table); | |
1746 | spin_unlock(&mm->page_table_lock); | |
1747 | ||
1748 | if (vma->vm_file) { | |
1749 | mapping = vma->vm_file->f_mapping; | |
1750 | sequence = mapping->truncate_count; | |
1751 | smp_rmb(); /* serializes i_size against truncate_count */ | |
1752 | } | |
1753 | retry: | |
1754 | cond_resched(); | |
1755 | new_page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret); | |
1756 | /* | |
1757 | * No smp_rmb is needed here as long as there's a full | |
1758 | * spin_lock/unlock sequence inside the ->nopage callback | |
1759 | * (for the pagecache lookup) that acts as an implicit | |
1760 | * smp_mb() and prevents the i_size read to happen | |
1761 | * after the next truncate_count read. | |
1762 | */ | |
1763 | ||
1764 | /* no page was available -- either SIGBUS or OOM */ | |
1765 | if (new_page == NOPAGE_SIGBUS) | |
1766 | return VM_FAULT_SIGBUS; | |
1767 | if (new_page == NOPAGE_OOM) | |
1768 | return VM_FAULT_OOM; | |
1769 | ||
1770 | /* | |
1771 | * Should we do an early C-O-W break? | |
1772 | */ | |
1773 | if (write_access && !(vma->vm_flags & VM_SHARED)) { | |
1774 | struct page *page; | |
1775 | ||
1776 | if (unlikely(anon_vma_prepare(vma))) | |
1777 | goto oom; | |
1778 | page = alloc_page_vma(GFP_HIGHUSER, vma, address); | |
1779 | if (!page) | |
1780 | goto oom; | |
1781 | copy_user_highpage(page, new_page, address); | |
1782 | page_cache_release(new_page); | |
1783 | new_page = page; | |
1784 | anon = 1; | |
1785 | } | |
1786 | ||
1787 | spin_lock(&mm->page_table_lock); | |
1788 | /* | |
1789 | * For a file-backed vma, someone could have truncated or otherwise | |
1790 | * invalidated this page. If unmap_mapping_range got called, | |
1791 | * retry getting the page. | |
1792 | */ | |
1793 | if (mapping && unlikely(sequence != mapping->truncate_count)) { | |
1794 | sequence = mapping->truncate_count; | |
1795 | spin_unlock(&mm->page_table_lock); | |
1796 | page_cache_release(new_page); | |
1797 | goto retry; | |
1798 | } | |
1799 | page_table = pte_offset_map(pmd, address); | |
1800 | ||
1801 | /* | |
1802 | * This silly early PAGE_DIRTY setting removes a race | |
1803 | * due to the bad i386 page protection. But it's valid | |
1804 | * for other architectures too. | |
1805 | * | |
1806 | * Note that if write_access is true, we either now have | |
1807 | * an exclusive copy of the page, or this is a shared mapping, | |
1808 | * so we can make it writable and dirty to avoid having to | |
1809 | * handle that later. | |
1810 | */ | |
1811 | /* Only go through if we didn't race with anybody else... */ | |
1812 | if (pte_none(*page_table)) { | |
1813 | if (!PageReserved(new_page)) | |
1814 | inc_mm_counter(mm, rss); | |
1815 | ||
1816 | flush_icache_page(vma, new_page); | |
1817 | entry = mk_pte(new_page, vma->vm_page_prot); | |
1818 | if (write_access) | |
1819 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
1820 | set_pte_at(mm, address, page_table, entry); | |
1821 | if (anon) { | |
1822 | lru_cache_add_active(new_page); | |
1823 | page_add_anon_rmap(new_page, vma, address); | |
1824 | } else | |
1825 | page_add_file_rmap(new_page); | |
1826 | pte_unmap(page_table); | |
1827 | } else { | |
1828 | /* One of our sibling threads was faster, back out. */ | |
1829 | pte_unmap(page_table); | |
1830 | page_cache_release(new_page); | |
1831 | spin_unlock(&mm->page_table_lock); | |
1832 | goto out; | |
1833 | } | |
1834 | ||
1835 | /* no need to invalidate: a not-present page shouldn't be cached */ | |
1836 | update_mmu_cache(vma, address, entry); | |
1837 | lazy_mmu_prot_update(entry); | |
1838 | spin_unlock(&mm->page_table_lock); | |
1839 | out: | |
1840 | return ret; | |
1841 | oom: | |
1842 | page_cache_release(new_page); | |
1843 | ret = VM_FAULT_OOM; | |
1844 | goto out; | |
1845 | } | |
1846 | ||
1847 | /* | |
1848 | * Fault of a previously existing named mapping. Repopulate the pte | |
1849 | * from the encoded file_pte if possible. This enables swappable | |
1850 | * nonlinear vmas. | |
1851 | */ | |
1852 | static int do_file_page(struct mm_struct * mm, struct vm_area_struct * vma, | |
1853 | unsigned long address, int write_access, pte_t *pte, pmd_t *pmd) | |
1854 | { | |
1855 | unsigned long pgoff; | |
1856 | int err; | |
1857 | ||
1858 | BUG_ON(!vma->vm_ops || !vma->vm_ops->nopage); | |
1859 | /* | |
1860 | * Fall back to the linear mapping if the fs does not support | |
1861 | * ->populate: | |
1862 | */ | |
1863 | if (!vma->vm_ops || !vma->vm_ops->populate || | |
1864 | (write_access && !(vma->vm_flags & VM_SHARED))) { | |
1865 | pte_clear(mm, address, pte); | |
1866 | return do_no_page(mm, vma, address, write_access, pte, pmd); | |
1867 | } | |
1868 | ||
1869 | pgoff = pte_to_pgoff(*pte); | |
1870 | ||
1871 | pte_unmap(pte); | |
1872 | spin_unlock(&mm->page_table_lock); | |
1873 | ||
1874 | err = vma->vm_ops->populate(vma, address & PAGE_MASK, PAGE_SIZE, vma->vm_page_prot, pgoff, 0); | |
1875 | if (err == -ENOMEM) | |
1876 | return VM_FAULT_OOM; | |
1877 | if (err) | |
1878 | return VM_FAULT_SIGBUS; | |
1879 | return VM_FAULT_MAJOR; | |
1880 | } | |
1881 | ||
1882 | /* | |
1883 | * These routines also need to handle stuff like marking pages dirty | |
1884 | * and/or accessed for architectures that don't do it in hardware (most | |
1885 | * RISC architectures). The early dirtying is also good on the i386. | |
1886 | * | |
1887 | * There is also a hook called "update_mmu_cache()" that architectures | |
1888 | * with external mmu caches can use to update those (ie the Sparc or | |
1889 | * PowerPC hashed page tables that act as extended TLBs). | |
1890 | * | |
1891 | * Note the "page_table_lock". It is to protect against kswapd removing | |
1892 | * pages from under us. Note that kswapd only ever _removes_ pages, never | |
1893 | * adds them. As such, once we have noticed that the page is not present, | |
1894 | * we can drop the lock early. | |
1895 | * | |
1896 | * The adding of pages is protected by the MM semaphore (which we hold), | |
1897 | * so we don't need to worry about a page being suddenly been added into | |
1898 | * our VM. | |
1899 | * | |
1900 | * We enter with the pagetable spinlock held, we are supposed to | |
1901 | * release it when done. | |
1902 | */ | |
1903 | static inline int handle_pte_fault(struct mm_struct *mm, | |
1904 | struct vm_area_struct * vma, unsigned long address, | |
1905 | int write_access, pte_t *pte, pmd_t *pmd) | |
1906 | { | |
1907 | pte_t entry; | |
1908 | ||
1909 | entry = *pte; | |
1910 | if (!pte_present(entry)) { | |
1911 | /* | |
1912 | * If it truly wasn't present, we know that kswapd | |
1913 | * and the PTE updates will not touch it later. So | |
1914 | * drop the lock. | |
1915 | */ | |
1916 | if (pte_none(entry)) | |
1917 | return do_no_page(mm, vma, address, write_access, pte, pmd); | |
1918 | if (pte_file(entry)) | |
1919 | return do_file_page(mm, vma, address, write_access, pte, pmd); | |
1920 | return do_swap_page(mm, vma, address, pte, pmd, entry, write_access); | |
1921 | } | |
1922 | ||
1923 | if (write_access) { | |
1924 | if (!pte_write(entry)) | |
1925 | return do_wp_page(mm, vma, address, pte, pmd, entry); | |
1926 | ||
1927 | entry = pte_mkdirty(entry); | |
1928 | } | |
1929 | entry = pte_mkyoung(entry); | |
1930 | ptep_set_access_flags(vma, address, pte, entry, write_access); | |
1931 | update_mmu_cache(vma, address, entry); | |
1932 | lazy_mmu_prot_update(entry); | |
1933 | pte_unmap(pte); | |
1934 | spin_unlock(&mm->page_table_lock); | |
1935 | return VM_FAULT_MINOR; | |
1936 | } | |
1937 | ||
1938 | /* | |
1939 | * By the time we get here, we already hold the mm semaphore | |
1940 | */ | |
1941 | int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct * vma, | |
1942 | unsigned long address, int write_access) | |
1943 | { | |
1944 | pgd_t *pgd; | |
1945 | pud_t *pud; | |
1946 | pmd_t *pmd; | |
1947 | pte_t *pte; | |
1948 | ||
1949 | __set_current_state(TASK_RUNNING); | |
1950 | ||
1951 | inc_page_state(pgfault); | |
1952 | ||
1953 | if (is_vm_hugetlb_page(vma)) | |
1954 | return VM_FAULT_SIGBUS; /* mapping truncation does this. */ | |
1955 | ||
1956 | /* | |
1957 | * We need the page table lock to synchronize with kswapd | |
1958 | * and the SMP-safe atomic PTE updates. | |
1959 | */ | |
1960 | pgd = pgd_offset(mm, address); | |
1961 | spin_lock(&mm->page_table_lock); | |
1962 | ||
1963 | pud = pud_alloc(mm, pgd, address); | |
1964 | if (!pud) | |
1965 | goto oom; | |
1966 | ||
1967 | pmd = pmd_alloc(mm, pud, address); | |
1968 | if (!pmd) | |
1969 | goto oom; | |
1970 | ||
1971 | pte = pte_alloc_map(mm, pmd, address); | |
1972 | if (!pte) | |
1973 | goto oom; | |
1974 | ||
1975 | return handle_pte_fault(mm, vma, address, write_access, pte, pmd); | |
1976 | ||
1977 | oom: | |
1978 | spin_unlock(&mm->page_table_lock); | |
1979 | return VM_FAULT_OOM; | |
1980 | } | |
1981 | ||
1982 | #ifndef __PAGETABLE_PUD_FOLDED | |
1983 | /* | |
1984 | * Allocate page upper directory. | |
1985 | * | |
1986 | * We've already handled the fast-path in-line, and we own the | |
1987 | * page table lock. | |
1988 | */ | |
1989 | pud_t fastcall *__pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) | |
1990 | { | |
1991 | pud_t *new; | |
1992 | ||
1993 | spin_unlock(&mm->page_table_lock); | |
1994 | new = pud_alloc_one(mm, address); | |
1995 | spin_lock(&mm->page_table_lock); | |
1996 | if (!new) | |
1997 | return NULL; | |
1998 | ||
1999 | /* | |
2000 | * Because we dropped the lock, we should re-check the | |
2001 | * entry, as somebody else could have populated it.. | |
2002 | */ | |
2003 | if (pgd_present(*pgd)) { | |
2004 | pud_free(new); | |
2005 | goto out; | |
2006 | } | |
2007 | pgd_populate(mm, pgd, new); | |
2008 | out: | |
2009 | return pud_offset(pgd, address); | |
2010 | } | |
2011 | #endif /* __PAGETABLE_PUD_FOLDED */ | |
2012 | ||
2013 | #ifndef __PAGETABLE_PMD_FOLDED | |
2014 | /* | |
2015 | * Allocate page middle directory. | |
2016 | * | |
2017 | * We've already handled the fast-path in-line, and we own the | |
2018 | * page table lock. | |
2019 | */ | |
2020 | pmd_t fastcall *__pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) | |
2021 | { | |
2022 | pmd_t *new; | |
2023 | ||
2024 | spin_unlock(&mm->page_table_lock); | |
2025 | new = pmd_alloc_one(mm, address); | |
2026 | spin_lock(&mm->page_table_lock); | |
2027 | if (!new) | |
2028 | return NULL; | |
2029 | ||
2030 | /* | |
2031 | * Because we dropped the lock, we should re-check the | |
2032 | * entry, as somebody else could have populated it.. | |
2033 | */ | |
2034 | #ifndef __ARCH_HAS_4LEVEL_HACK | |
2035 | if (pud_present(*pud)) { | |
2036 | pmd_free(new); | |
2037 | goto out; | |
2038 | } | |
2039 | pud_populate(mm, pud, new); | |
2040 | #else | |
2041 | if (pgd_present(*pud)) { | |
2042 | pmd_free(new); | |
2043 | goto out; | |
2044 | } | |
2045 | pgd_populate(mm, pud, new); | |
2046 | #endif /* __ARCH_HAS_4LEVEL_HACK */ | |
2047 | ||
2048 | out: | |
2049 | return pmd_offset(pud, address); | |
2050 | } | |
2051 | #endif /* __PAGETABLE_PMD_FOLDED */ | |
2052 | ||
2053 | int make_pages_present(unsigned long addr, unsigned long end) | |
2054 | { | |
2055 | int ret, len, write; | |
2056 | struct vm_area_struct * vma; | |
2057 | ||
2058 | vma = find_vma(current->mm, addr); | |
2059 | if (!vma) | |
2060 | return -1; | |
2061 | write = (vma->vm_flags & VM_WRITE) != 0; | |
2062 | if (addr >= end) | |
2063 | BUG(); | |
2064 | if (end > vma->vm_end) | |
2065 | BUG(); | |
2066 | len = (end+PAGE_SIZE-1)/PAGE_SIZE-addr/PAGE_SIZE; | |
2067 | ret = get_user_pages(current, current->mm, addr, | |
2068 | len, write, 0, NULL, NULL); | |
2069 | if (ret < 0) | |
2070 | return ret; | |
2071 | return ret == len ? 0 : -1; | |
2072 | } | |
2073 | ||
2074 | /* | |
2075 | * Map a vmalloc()-space virtual address to the physical page. | |
2076 | */ | |
2077 | struct page * vmalloc_to_page(void * vmalloc_addr) | |
2078 | { | |
2079 | unsigned long addr = (unsigned long) vmalloc_addr; | |
2080 | struct page *page = NULL; | |
2081 | pgd_t *pgd = pgd_offset_k(addr); | |
2082 | pud_t *pud; | |
2083 | pmd_t *pmd; | |
2084 | pte_t *ptep, pte; | |
2085 | ||
2086 | if (!pgd_none(*pgd)) { | |
2087 | pud = pud_offset(pgd, addr); | |
2088 | if (!pud_none(*pud)) { | |
2089 | pmd = pmd_offset(pud, addr); | |
2090 | if (!pmd_none(*pmd)) { | |
2091 | ptep = pte_offset_map(pmd, addr); | |
2092 | pte = *ptep; | |
2093 | if (pte_present(pte)) | |
2094 | page = pte_page(pte); | |
2095 | pte_unmap(ptep); | |
2096 | } | |
2097 | } | |
2098 | } | |
2099 | return page; | |
2100 | } | |
2101 | ||
2102 | EXPORT_SYMBOL(vmalloc_to_page); | |
2103 | ||
2104 | /* | |
2105 | * Map a vmalloc()-space virtual address to the physical page frame number. | |
2106 | */ | |
2107 | unsigned long vmalloc_to_pfn(void * vmalloc_addr) | |
2108 | { | |
2109 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); | |
2110 | } | |
2111 | ||
2112 | EXPORT_SYMBOL(vmalloc_to_pfn); | |
2113 | ||
2114 | /* | |
2115 | * update_mem_hiwater | |
2116 | * - update per process rss and vm high water data | |
2117 | */ | |
2118 | void update_mem_hiwater(struct task_struct *tsk) | |
2119 | { | |
2120 | if (tsk->mm) { | |
2121 | unsigned long rss = get_mm_counter(tsk->mm, rss); | |
2122 | ||
2123 | if (tsk->mm->hiwater_rss < rss) | |
2124 | tsk->mm->hiwater_rss = rss; | |
2125 | if (tsk->mm->hiwater_vm < tsk->mm->total_vm) | |
2126 | tsk->mm->hiwater_vm = tsk->mm->total_vm; | |
2127 | } | |
2128 | } | |
2129 | ||
2130 | #if !defined(__HAVE_ARCH_GATE_AREA) | |
2131 | ||
2132 | #if defined(AT_SYSINFO_EHDR) | |
2133 | struct vm_area_struct gate_vma; | |
2134 | ||
2135 | static int __init gate_vma_init(void) | |
2136 | { | |
2137 | gate_vma.vm_mm = NULL; | |
2138 | gate_vma.vm_start = FIXADDR_USER_START; | |
2139 | gate_vma.vm_end = FIXADDR_USER_END; | |
2140 | gate_vma.vm_page_prot = PAGE_READONLY; | |
2141 | gate_vma.vm_flags = 0; | |
2142 | return 0; | |
2143 | } | |
2144 | __initcall(gate_vma_init); | |
2145 | #endif | |
2146 | ||
2147 | struct vm_area_struct *get_gate_vma(struct task_struct *tsk) | |
2148 | { | |
2149 | #ifdef AT_SYSINFO_EHDR | |
2150 | return &gate_vma; | |
2151 | #else | |
2152 | return NULL; | |
2153 | #endif | |
2154 | } | |
2155 | ||
2156 | int in_gate_area_no_task(unsigned long addr) | |
2157 | { | |
2158 | #ifdef AT_SYSINFO_EHDR | |
2159 | if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END)) | |
2160 | return 1; | |
2161 | #endif | |
2162 | return 0; | |
2163 | } | |
2164 | ||
2165 | #endif /* __HAVE_ARCH_GATE_AREA */ |