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> | |
0ff92245 | 50 | #include <linux/delayacct.h> |
1da177e4 | 51 | #include <linux/init.h> |
edc79b2a | 52 | #include <linux/writeback.h> |
1da177e4 LT |
53 | |
54 | #include <asm/pgalloc.h> | |
55 | #include <asm/uaccess.h> | |
56 | #include <asm/tlb.h> | |
57 | #include <asm/tlbflush.h> | |
58 | #include <asm/pgtable.h> | |
59 | ||
60 | #include <linux/swapops.h> | |
61 | #include <linux/elf.h> | |
62 | ||
d41dee36 | 63 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
64 | /* use the per-pgdat data instead for discontigmem - mbligh */ |
65 | unsigned long max_mapnr; | |
66 | struct page *mem_map; | |
67 | ||
68 | EXPORT_SYMBOL(max_mapnr); | |
69 | EXPORT_SYMBOL(mem_map); | |
70 | #endif | |
71 | ||
72 | unsigned long num_physpages; | |
73 | /* | |
74 | * A number of key systems in x86 including ioremap() rely on the assumption | |
75 | * that high_memory defines the upper bound on direct map memory, then end | |
76 | * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and | |
77 | * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL | |
78 | * and ZONE_HIGHMEM. | |
79 | */ | |
80 | void * high_memory; | |
1da177e4 LT |
81 | |
82 | EXPORT_SYMBOL(num_physpages); | |
83 | EXPORT_SYMBOL(high_memory); | |
1da177e4 | 84 | |
a62eaf15 AK |
85 | int randomize_va_space __read_mostly = 1; |
86 | ||
87 | static int __init disable_randmaps(char *s) | |
88 | { | |
89 | randomize_va_space = 0; | |
9b41046c | 90 | return 1; |
a62eaf15 AK |
91 | } |
92 | __setup("norandmaps", disable_randmaps); | |
93 | ||
94 | ||
1da177e4 LT |
95 | /* |
96 | * If a p?d_bad entry is found while walking page tables, report | |
97 | * the error, before resetting entry to p?d_none. Usually (but | |
98 | * very seldom) called out from the p?d_none_or_clear_bad macros. | |
99 | */ | |
100 | ||
101 | void pgd_clear_bad(pgd_t *pgd) | |
102 | { | |
103 | pgd_ERROR(*pgd); | |
104 | pgd_clear(pgd); | |
105 | } | |
106 | ||
107 | void pud_clear_bad(pud_t *pud) | |
108 | { | |
109 | pud_ERROR(*pud); | |
110 | pud_clear(pud); | |
111 | } | |
112 | ||
113 | void pmd_clear_bad(pmd_t *pmd) | |
114 | { | |
115 | pmd_ERROR(*pmd); | |
116 | pmd_clear(pmd); | |
117 | } | |
118 | ||
119 | /* | |
120 | * Note: this doesn't free the actual pages themselves. That | |
121 | * has been handled earlier when unmapping all the memory regions. | |
122 | */ | |
e0da382c | 123 | static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd) |
1da177e4 | 124 | { |
e0da382c HD |
125 | struct page *page = pmd_page(*pmd); |
126 | pmd_clear(pmd); | |
4c21e2f2 | 127 | pte_lock_deinit(page); |
e0da382c | 128 | pte_free_tlb(tlb, page); |
df849a15 | 129 | dec_zone_page_state(page, NR_PAGETABLE); |
e0da382c | 130 | tlb->mm->nr_ptes--; |
1da177e4 LT |
131 | } |
132 | ||
e0da382c HD |
133 | static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
134 | unsigned long addr, unsigned long end, | |
135 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
136 | { |
137 | pmd_t *pmd; | |
138 | unsigned long next; | |
e0da382c | 139 | unsigned long start; |
1da177e4 | 140 | |
e0da382c | 141 | start = addr; |
1da177e4 | 142 | pmd = pmd_offset(pud, addr); |
1da177e4 LT |
143 | do { |
144 | next = pmd_addr_end(addr, end); | |
145 | if (pmd_none_or_clear_bad(pmd)) | |
146 | continue; | |
e0da382c | 147 | free_pte_range(tlb, pmd); |
1da177e4 LT |
148 | } while (pmd++, addr = next, addr != end); |
149 | ||
e0da382c HD |
150 | start &= PUD_MASK; |
151 | if (start < floor) | |
152 | return; | |
153 | if (ceiling) { | |
154 | ceiling &= PUD_MASK; | |
155 | if (!ceiling) | |
156 | return; | |
1da177e4 | 157 | } |
e0da382c HD |
158 | if (end - 1 > ceiling - 1) |
159 | return; | |
160 | ||
161 | pmd = pmd_offset(pud, start); | |
162 | pud_clear(pud); | |
163 | pmd_free_tlb(tlb, pmd); | |
1da177e4 LT |
164 | } |
165 | ||
e0da382c HD |
166 | static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, |
167 | unsigned long addr, unsigned long end, | |
168 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
169 | { |
170 | pud_t *pud; | |
171 | unsigned long next; | |
e0da382c | 172 | unsigned long start; |
1da177e4 | 173 | |
e0da382c | 174 | start = addr; |
1da177e4 | 175 | pud = pud_offset(pgd, addr); |
1da177e4 LT |
176 | do { |
177 | next = pud_addr_end(addr, end); | |
178 | if (pud_none_or_clear_bad(pud)) | |
179 | continue; | |
e0da382c | 180 | free_pmd_range(tlb, pud, addr, next, floor, ceiling); |
1da177e4 LT |
181 | } while (pud++, addr = next, addr != end); |
182 | ||
e0da382c HD |
183 | start &= PGDIR_MASK; |
184 | if (start < floor) | |
185 | return; | |
186 | if (ceiling) { | |
187 | ceiling &= PGDIR_MASK; | |
188 | if (!ceiling) | |
189 | return; | |
1da177e4 | 190 | } |
e0da382c HD |
191 | if (end - 1 > ceiling - 1) |
192 | return; | |
193 | ||
194 | pud = pud_offset(pgd, start); | |
195 | pgd_clear(pgd); | |
196 | pud_free_tlb(tlb, pud); | |
1da177e4 LT |
197 | } |
198 | ||
199 | /* | |
e0da382c HD |
200 | * This function frees user-level page tables of a process. |
201 | * | |
1da177e4 LT |
202 | * Must be called with pagetable lock held. |
203 | */ | |
3bf5ee95 | 204 | void free_pgd_range(struct mmu_gather **tlb, |
e0da382c HD |
205 | unsigned long addr, unsigned long end, |
206 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
207 | { |
208 | pgd_t *pgd; | |
209 | unsigned long next; | |
e0da382c HD |
210 | unsigned long start; |
211 | ||
212 | /* | |
213 | * The next few lines have given us lots of grief... | |
214 | * | |
215 | * Why are we testing PMD* at this top level? Because often | |
216 | * there will be no work to do at all, and we'd prefer not to | |
217 | * go all the way down to the bottom just to discover that. | |
218 | * | |
219 | * Why all these "- 1"s? Because 0 represents both the bottom | |
220 | * of the address space and the top of it (using -1 for the | |
221 | * top wouldn't help much: the masks would do the wrong thing). | |
222 | * The rule is that addr 0 and floor 0 refer to the bottom of | |
223 | * the address space, but end 0 and ceiling 0 refer to the top | |
224 | * Comparisons need to use "end - 1" and "ceiling - 1" (though | |
225 | * that end 0 case should be mythical). | |
226 | * | |
227 | * Wherever addr is brought up or ceiling brought down, we must | |
228 | * be careful to reject "the opposite 0" before it confuses the | |
229 | * subsequent tests. But what about where end is brought down | |
230 | * by PMD_SIZE below? no, end can't go down to 0 there. | |
231 | * | |
232 | * Whereas we round start (addr) and ceiling down, by different | |
233 | * masks at different levels, in order to test whether a table | |
234 | * now has no other vmas using it, so can be freed, we don't | |
235 | * bother to round floor or end up - the tests don't need that. | |
236 | */ | |
1da177e4 | 237 | |
e0da382c HD |
238 | addr &= PMD_MASK; |
239 | if (addr < floor) { | |
240 | addr += PMD_SIZE; | |
241 | if (!addr) | |
242 | return; | |
243 | } | |
244 | if (ceiling) { | |
245 | ceiling &= PMD_MASK; | |
246 | if (!ceiling) | |
247 | return; | |
248 | } | |
249 | if (end - 1 > ceiling - 1) | |
250 | end -= PMD_SIZE; | |
251 | if (addr > end - 1) | |
252 | return; | |
253 | ||
254 | start = addr; | |
3bf5ee95 | 255 | pgd = pgd_offset((*tlb)->mm, addr); |
1da177e4 LT |
256 | do { |
257 | next = pgd_addr_end(addr, end); | |
258 | if (pgd_none_or_clear_bad(pgd)) | |
259 | continue; | |
3bf5ee95 | 260 | free_pud_range(*tlb, pgd, addr, next, floor, ceiling); |
1da177e4 | 261 | } while (pgd++, addr = next, addr != end); |
e0da382c | 262 | |
4d6ddfa9 | 263 | if (!(*tlb)->fullmm) |
3bf5ee95 | 264 | flush_tlb_pgtables((*tlb)->mm, start, end); |
e0da382c HD |
265 | } |
266 | ||
267 | void free_pgtables(struct mmu_gather **tlb, struct vm_area_struct *vma, | |
3bf5ee95 | 268 | unsigned long floor, unsigned long ceiling) |
e0da382c HD |
269 | { |
270 | while (vma) { | |
271 | struct vm_area_struct *next = vma->vm_next; | |
272 | unsigned long addr = vma->vm_start; | |
273 | ||
8f4f8c16 HD |
274 | /* |
275 | * Hide vma from rmap and vmtruncate before freeing pgtables | |
276 | */ | |
277 | anon_vma_unlink(vma); | |
278 | unlink_file_vma(vma); | |
279 | ||
9da61aef | 280 | if (is_vm_hugetlb_page(vma)) { |
3bf5ee95 | 281 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
e0da382c | 282 | floor, next? next->vm_start: ceiling); |
3bf5ee95 HD |
283 | } else { |
284 | /* | |
285 | * Optimization: gather nearby vmas into one call down | |
286 | */ | |
287 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE | |
4866920b | 288 | && !is_vm_hugetlb_page(next)) { |
3bf5ee95 HD |
289 | vma = next; |
290 | next = vma->vm_next; | |
8f4f8c16 HD |
291 | anon_vma_unlink(vma); |
292 | unlink_file_vma(vma); | |
3bf5ee95 HD |
293 | } |
294 | free_pgd_range(tlb, addr, vma->vm_end, | |
295 | floor, next? next->vm_start: ceiling); | |
296 | } | |
e0da382c HD |
297 | vma = next; |
298 | } | |
1da177e4 LT |
299 | } |
300 | ||
1bb3630e | 301 | int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address) |
1da177e4 | 302 | { |
c74df32c | 303 | struct page *new = pte_alloc_one(mm, address); |
1bb3630e HD |
304 | if (!new) |
305 | return -ENOMEM; | |
306 | ||
4c21e2f2 | 307 | pte_lock_init(new); |
c74df32c | 308 | spin_lock(&mm->page_table_lock); |
4c21e2f2 HD |
309 | if (pmd_present(*pmd)) { /* Another has populated it */ |
310 | pte_lock_deinit(new); | |
1bb3630e | 311 | pte_free(new); |
4c21e2f2 | 312 | } else { |
1da177e4 | 313 | mm->nr_ptes++; |
df849a15 | 314 | inc_zone_page_state(new, NR_PAGETABLE); |
1da177e4 LT |
315 | pmd_populate(mm, pmd, new); |
316 | } | |
c74df32c | 317 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 318 | return 0; |
1da177e4 LT |
319 | } |
320 | ||
1bb3630e | 321 | int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) |
1da177e4 | 322 | { |
1bb3630e HD |
323 | pte_t *new = pte_alloc_one_kernel(&init_mm, address); |
324 | if (!new) | |
325 | return -ENOMEM; | |
326 | ||
327 | spin_lock(&init_mm.page_table_lock); | |
328 | if (pmd_present(*pmd)) /* Another has populated it */ | |
329 | pte_free_kernel(new); | |
330 | else | |
331 | pmd_populate_kernel(&init_mm, pmd, new); | |
332 | spin_unlock(&init_mm.page_table_lock); | |
333 | return 0; | |
1da177e4 LT |
334 | } |
335 | ||
ae859762 HD |
336 | static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss) |
337 | { | |
338 | if (file_rss) | |
339 | add_mm_counter(mm, file_rss, file_rss); | |
340 | if (anon_rss) | |
341 | add_mm_counter(mm, anon_rss, anon_rss); | |
342 | } | |
343 | ||
b5810039 | 344 | /* |
6aab341e LT |
345 | * This function is called to print an error when a bad pte |
346 | * is found. For example, we might have a PFN-mapped pte in | |
347 | * a region that doesn't allow it. | |
b5810039 NP |
348 | * |
349 | * The calling function must still handle the error. | |
350 | */ | |
351 | void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr) | |
352 | { | |
353 | printk(KERN_ERR "Bad pte = %08llx, process = %s, " | |
354 | "vm_flags = %lx, vaddr = %lx\n", | |
355 | (long long)pte_val(pte), | |
356 | (vma->vm_mm == current->mm ? current->comm : "???"), | |
357 | vma->vm_flags, vaddr); | |
358 | dump_stack(); | |
359 | } | |
360 | ||
67121172 LT |
361 | static inline int is_cow_mapping(unsigned int flags) |
362 | { | |
363 | return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; | |
364 | } | |
365 | ||
ee498ed7 | 366 | /* |
6aab341e LT |
367 | * This function gets the "struct page" associated with a pte. |
368 | * | |
369 | * NOTE! Some mappings do not have "struct pages". A raw PFN mapping | |
370 | * will have each page table entry just pointing to a raw page frame | |
371 | * number, and as far as the VM layer is concerned, those do not have | |
372 | * pages associated with them - even if the PFN might point to memory | |
373 | * that otherwise is perfectly fine and has a "struct page". | |
374 | * | |
375 | * The way we recognize those mappings is through the rules set up | |
376 | * by "remap_pfn_range()": the vma will have the VM_PFNMAP bit set, | |
377 | * and the vm_pgoff will point to the first PFN mapped: thus every | |
378 | * page that is a raw mapping will always honor the rule | |
379 | * | |
380 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) | |
381 | * | |
382 | * and if that isn't true, the page has been COW'ed (in which case it | |
383 | * _does_ have a "struct page" associated with it even if it is in a | |
384 | * VM_PFNMAP range). | |
ee498ed7 | 385 | */ |
6aab341e | 386 | struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, pte_t pte) |
ee498ed7 | 387 | { |
6aab341e LT |
388 | unsigned long pfn = pte_pfn(pte); |
389 | ||
b7ab795b | 390 | if (unlikely(vma->vm_flags & VM_PFNMAP)) { |
6aab341e LT |
391 | unsigned long off = (addr - vma->vm_start) >> PAGE_SHIFT; |
392 | if (pfn == vma->vm_pgoff + off) | |
393 | return NULL; | |
67121172 | 394 | if (!is_cow_mapping(vma->vm_flags)) |
fb155c16 | 395 | return NULL; |
6aab341e LT |
396 | } |
397 | ||
315ab19a NP |
398 | /* |
399 | * Add some anal sanity checks for now. Eventually, | |
400 | * we should just do "return pfn_to_page(pfn)", but | |
401 | * in the meantime we check that we get a valid pfn, | |
402 | * and that the resulting page looks ok. | |
403 | */ | |
6aab341e LT |
404 | if (unlikely(!pfn_valid(pfn))) { |
405 | print_bad_pte(vma, pte, addr); | |
406 | return NULL; | |
407 | } | |
408 | ||
409 | /* | |
410 | * NOTE! We still have PageReserved() pages in the page | |
411 | * tables. | |
412 | * | |
413 | * The PAGE_ZERO() pages and various VDSO mappings can | |
414 | * cause them to exist. | |
415 | */ | |
416 | return pfn_to_page(pfn); | |
ee498ed7 HD |
417 | } |
418 | ||
1da177e4 LT |
419 | /* |
420 | * copy one vm_area from one task to the other. Assumes the page tables | |
421 | * already present in the new task to be cleared in the whole range | |
422 | * covered by this vma. | |
1da177e4 LT |
423 | */ |
424 | ||
8c103762 | 425 | static inline void |
1da177e4 | 426 | copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
b5810039 | 427 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, |
8c103762 | 428 | unsigned long addr, int *rss) |
1da177e4 | 429 | { |
b5810039 | 430 | unsigned long vm_flags = vma->vm_flags; |
1da177e4 LT |
431 | pte_t pte = *src_pte; |
432 | struct page *page; | |
1da177e4 LT |
433 | |
434 | /* pte contains position in swap or file, so copy. */ | |
435 | if (unlikely(!pte_present(pte))) { | |
436 | if (!pte_file(pte)) { | |
0697212a CL |
437 | swp_entry_t entry = pte_to_swp_entry(pte); |
438 | ||
439 | swap_duplicate(entry); | |
1da177e4 LT |
440 | /* make sure dst_mm is on swapoff's mmlist. */ |
441 | if (unlikely(list_empty(&dst_mm->mmlist))) { | |
442 | spin_lock(&mmlist_lock); | |
f412ac08 HD |
443 | if (list_empty(&dst_mm->mmlist)) |
444 | list_add(&dst_mm->mmlist, | |
445 | &src_mm->mmlist); | |
1da177e4 LT |
446 | spin_unlock(&mmlist_lock); |
447 | } | |
0697212a CL |
448 | if (is_write_migration_entry(entry) && |
449 | is_cow_mapping(vm_flags)) { | |
450 | /* | |
451 | * COW mappings require pages in both parent | |
452 | * and child to be set to read. | |
453 | */ | |
454 | make_migration_entry_read(&entry); | |
455 | pte = swp_entry_to_pte(entry); | |
456 | set_pte_at(src_mm, addr, src_pte, pte); | |
457 | } | |
1da177e4 | 458 | } |
ae859762 | 459 | goto out_set_pte; |
1da177e4 LT |
460 | } |
461 | ||
1da177e4 LT |
462 | /* |
463 | * If it's a COW mapping, write protect it both | |
464 | * in the parent and the child | |
465 | */ | |
67121172 | 466 | if (is_cow_mapping(vm_flags)) { |
1da177e4 | 467 | ptep_set_wrprotect(src_mm, addr, src_pte); |
3dc90795 | 468 | pte = pte_wrprotect(pte); |
1da177e4 LT |
469 | } |
470 | ||
471 | /* | |
472 | * If it's a shared mapping, mark it clean in | |
473 | * the child | |
474 | */ | |
475 | if (vm_flags & VM_SHARED) | |
476 | pte = pte_mkclean(pte); | |
477 | pte = pte_mkold(pte); | |
6aab341e LT |
478 | |
479 | page = vm_normal_page(vma, addr, pte); | |
480 | if (page) { | |
481 | get_page(page); | |
c97a9e10 | 482 | page_dup_rmap(page, vma, addr); |
6aab341e LT |
483 | rss[!!PageAnon(page)]++; |
484 | } | |
ae859762 HD |
485 | |
486 | out_set_pte: | |
487 | set_pte_at(dst_mm, addr, dst_pte, pte); | |
1da177e4 LT |
488 | } |
489 | ||
490 | static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
491 | pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, | |
492 | unsigned long addr, unsigned long end) | |
493 | { | |
494 | pte_t *src_pte, *dst_pte; | |
c74df32c | 495 | spinlock_t *src_ptl, *dst_ptl; |
e040f218 | 496 | int progress = 0; |
8c103762 | 497 | int rss[2]; |
1da177e4 LT |
498 | |
499 | again: | |
ae859762 | 500 | rss[1] = rss[0] = 0; |
c74df32c | 501 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
1da177e4 LT |
502 | if (!dst_pte) |
503 | return -ENOMEM; | |
504 | src_pte = pte_offset_map_nested(src_pmd, addr); | |
4c21e2f2 | 505 | src_ptl = pte_lockptr(src_mm, src_pmd); |
f20dc5f7 | 506 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
6606c3e0 | 507 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 508 | |
1da177e4 LT |
509 | do { |
510 | /* | |
511 | * We are holding two locks at this point - either of them | |
512 | * could generate latencies in another task on another CPU. | |
513 | */ | |
e040f218 HD |
514 | if (progress >= 32) { |
515 | progress = 0; | |
516 | if (need_resched() || | |
c74df32c HD |
517 | need_lockbreak(src_ptl) || |
518 | need_lockbreak(dst_ptl)) | |
e040f218 HD |
519 | break; |
520 | } | |
1da177e4 LT |
521 | if (pte_none(*src_pte)) { |
522 | progress++; | |
523 | continue; | |
524 | } | |
8c103762 | 525 | copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss); |
1da177e4 LT |
526 | progress += 8; |
527 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); | |
1da177e4 | 528 | |
6606c3e0 | 529 | arch_leave_lazy_mmu_mode(); |
c74df32c | 530 | spin_unlock(src_ptl); |
1da177e4 | 531 | pte_unmap_nested(src_pte - 1); |
ae859762 | 532 | add_mm_rss(dst_mm, rss[0], rss[1]); |
c74df32c HD |
533 | pte_unmap_unlock(dst_pte - 1, dst_ptl); |
534 | cond_resched(); | |
1da177e4 LT |
535 | if (addr != end) |
536 | goto again; | |
537 | return 0; | |
538 | } | |
539 | ||
540 | static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
541 | pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, | |
542 | unsigned long addr, unsigned long end) | |
543 | { | |
544 | pmd_t *src_pmd, *dst_pmd; | |
545 | unsigned long next; | |
546 | ||
547 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); | |
548 | if (!dst_pmd) | |
549 | return -ENOMEM; | |
550 | src_pmd = pmd_offset(src_pud, addr); | |
551 | do { | |
552 | next = pmd_addr_end(addr, end); | |
553 | if (pmd_none_or_clear_bad(src_pmd)) | |
554 | continue; | |
555 | if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, | |
556 | vma, addr, next)) | |
557 | return -ENOMEM; | |
558 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); | |
559 | return 0; | |
560 | } | |
561 | ||
562 | static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
563 | pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, | |
564 | unsigned long addr, unsigned long end) | |
565 | { | |
566 | pud_t *src_pud, *dst_pud; | |
567 | unsigned long next; | |
568 | ||
569 | dst_pud = pud_alloc(dst_mm, dst_pgd, addr); | |
570 | if (!dst_pud) | |
571 | return -ENOMEM; | |
572 | src_pud = pud_offset(src_pgd, addr); | |
573 | do { | |
574 | next = pud_addr_end(addr, end); | |
575 | if (pud_none_or_clear_bad(src_pud)) | |
576 | continue; | |
577 | if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, | |
578 | vma, addr, next)) | |
579 | return -ENOMEM; | |
580 | } while (dst_pud++, src_pud++, addr = next, addr != end); | |
581 | return 0; | |
582 | } | |
583 | ||
584 | int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
585 | struct vm_area_struct *vma) | |
586 | { | |
587 | pgd_t *src_pgd, *dst_pgd; | |
588 | unsigned long next; | |
589 | unsigned long addr = vma->vm_start; | |
590 | unsigned long end = vma->vm_end; | |
591 | ||
d992895b NP |
592 | /* |
593 | * Don't copy ptes where a page fault will fill them correctly. | |
594 | * Fork becomes much lighter when there are big shared or private | |
595 | * readonly mappings. The tradeoff is that copy_page_range is more | |
596 | * efficient than faulting. | |
597 | */ | |
4d7672b4 | 598 | if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) { |
d992895b NP |
599 | if (!vma->anon_vma) |
600 | return 0; | |
601 | } | |
602 | ||
1da177e4 LT |
603 | if (is_vm_hugetlb_page(vma)) |
604 | return copy_hugetlb_page_range(dst_mm, src_mm, vma); | |
605 | ||
606 | dst_pgd = pgd_offset(dst_mm, addr); | |
607 | src_pgd = pgd_offset(src_mm, addr); | |
608 | do { | |
609 | next = pgd_addr_end(addr, end); | |
610 | if (pgd_none_or_clear_bad(src_pgd)) | |
611 | continue; | |
612 | if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, | |
613 | vma, addr, next)) | |
614 | return -ENOMEM; | |
615 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); | |
616 | return 0; | |
617 | } | |
618 | ||
51c6f666 | 619 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
b5810039 | 620 | struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 | 621 | unsigned long addr, unsigned long end, |
51c6f666 | 622 | long *zap_work, struct zap_details *details) |
1da177e4 | 623 | { |
b5810039 | 624 | struct mm_struct *mm = tlb->mm; |
1da177e4 | 625 | pte_t *pte; |
508034a3 | 626 | spinlock_t *ptl; |
ae859762 HD |
627 | int file_rss = 0; |
628 | int anon_rss = 0; | |
1da177e4 | 629 | |
508034a3 | 630 | pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
6606c3e0 | 631 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
632 | do { |
633 | pte_t ptent = *pte; | |
51c6f666 RH |
634 | if (pte_none(ptent)) { |
635 | (*zap_work)--; | |
1da177e4 | 636 | continue; |
51c6f666 | 637 | } |
6f5e6b9e HD |
638 | |
639 | (*zap_work) -= PAGE_SIZE; | |
640 | ||
1da177e4 | 641 | if (pte_present(ptent)) { |
ee498ed7 | 642 | struct page *page; |
51c6f666 | 643 | |
6aab341e | 644 | page = vm_normal_page(vma, addr, ptent); |
1da177e4 LT |
645 | if (unlikely(details) && page) { |
646 | /* | |
647 | * unmap_shared_mapping_pages() wants to | |
648 | * invalidate cache without truncating: | |
649 | * unmap shared but keep private pages. | |
650 | */ | |
651 | if (details->check_mapping && | |
652 | details->check_mapping != page->mapping) | |
653 | continue; | |
654 | /* | |
655 | * Each page->index must be checked when | |
656 | * invalidating or truncating nonlinear. | |
657 | */ | |
658 | if (details->nonlinear_vma && | |
659 | (page->index < details->first_index || | |
660 | page->index > details->last_index)) | |
661 | continue; | |
662 | } | |
b5810039 | 663 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
a600388d | 664 | tlb->fullmm); |
1da177e4 LT |
665 | tlb_remove_tlb_entry(tlb, pte, addr); |
666 | if (unlikely(!page)) | |
667 | continue; | |
668 | if (unlikely(details) && details->nonlinear_vma | |
669 | && linear_page_index(details->nonlinear_vma, | |
670 | addr) != page->index) | |
b5810039 | 671 | set_pte_at(mm, addr, pte, |
1da177e4 | 672 | pgoff_to_pte(page->index)); |
1da177e4 | 673 | if (PageAnon(page)) |
86d912f4 | 674 | anon_rss--; |
6237bcd9 HD |
675 | else { |
676 | if (pte_dirty(ptent)) | |
677 | set_page_dirty(page); | |
678 | if (pte_young(ptent)) | |
daa88c8d | 679 | SetPageReferenced(page); |
86d912f4 | 680 | file_rss--; |
6237bcd9 | 681 | } |
7de6b805 | 682 | page_remove_rmap(page, vma); |
1da177e4 LT |
683 | tlb_remove_page(tlb, page); |
684 | continue; | |
685 | } | |
686 | /* | |
687 | * If details->check_mapping, we leave swap entries; | |
688 | * if details->nonlinear_vma, we leave file entries. | |
689 | */ | |
690 | if (unlikely(details)) | |
691 | continue; | |
692 | if (!pte_file(ptent)) | |
693 | free_swap_and_cache(pte_to_swp_entry(ptent)); | |
9888a1ca | 694 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
51c6f666 | 695 | } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0)); |
ae859762 | 696 | |
86d912f4 | 697 | add_mm_rss(mm, file_rss, anon_rss); |
6606c3e0 | 698 | arch_leave_lazy_mmu_mode(); |
508034a3 | 699 | pte_unmap_unlock(pte - 1, ptl); |
51c6f666 RH |
700 | |
701 | return addr; | |
1da177e4 LT |
702 | } |
703 | ||
51c6f666 | 704 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
b5810039 | 705 | struct vm_area_struct *vma, pud_t *pud, |
1da177e4 | 706 | unsigned long addr, unsigned long end, |
51c6f666 | 707 | long *zap_work, struct zap_details *details) |
1da177e4 LT |
708 | { |
709 | pmd_t *pmd; | |
710 | unsigned long next; | |
711 | ||
712 | pmd = pmd_offset(pud, addr); | |
713 | do { | |
714 | next = pmd_addr_end(addr, end); | |
51c6f666 RH |
715 | if (pmd_none_or_clear_bad(pmd)) { |
716 | (*zap_work)--; | |
1da177e4 | 717 | continue; |
51c6f666 RH |
718 | } |
719 | next = zap_pte_range(tlb, vma, pmd, addr, next, | |
720 | zap_work, details); | |
721 | } while (pmd++, addr = next, (addr != end && *zap_work > 0)); | |
722 | ||
723 | return addr; | |
1da177e4 LT |
724 | } |
725 | ||
51c6f666 | 726 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
b5810039 | 727 | struct vm_area_struct *vma, pgd_t *pgd, |
1da177e4 | 728 | unsigned long addr, unsigned long end, |
51c6f666 | 729 | long *zap_work, struct zap_details *details) |
1da177e4 LT |
730 | { |
731 | pud_t *pud; | |
732 | unsigned long next; | |
733 | ||
734 | pud = pud_offset(pgd, addr); | |
735 | do { | |
736 | next = pud_addr_end(addr, end); | |
51c6f666 RH |
737 | if (pud_none_or_clear_bad(pud)) { |
738 | (*zap_work)--; | |
1da177e4 | 739 | continue; |
51c6f666 RH |
740 | } |
741 | next = zap_pmd_range(tlb, vma, pud, addr, next, | |
742 | zap_work, details); | |
743 | } while (pud++, addr = next, (addr != end && *zap_work > 0)); | |
744 | ||
745 | return addr; | |
1da177e4 LT |
746 | } |
747 | ||
51c6f666 RH |
748 | static unsigned long unmap_page_range(struct mmu_gather *tlb, |
749 | struct vm_area_struct *vma, | |
1da177e4 | 750 | unsigned long addr, unsigned long end, |
51c6f666 | 751 | long *zap_work, struct zap_details *details) |
1da177e4 LT |
752 | { |
753 | pgd_t *pgd; | |
754 | unsigned long next; | |
755 | ||
756 | if (details && !details->check_mapping && !details->nonlinear_vma) | |
757 | details = NULL; | |
758 | ||
759 | BUG_ON(addr >= end); | |
760 | tlb_start_vma(tlb, vma); | |
761 | pgd = pgd_offset(vma->vm_mm, addr); | |
762 | do { | |
763 | next = pgd_addr_end(addr, end); | |
51c6f666 RH |
764 | if (pgd_none_or_clear_bad(pgd)) { |
765 | (*zap_work)--; | |
1da177e4 | 766 | continue; |
51c6f666 RH |
767 | } |
768 | next = zap_pud_range(tlb, vma, pgd, addr, next, | |
769 | zap_work, details); | |
770 | } while (pgd++, addr = next, (addr != end && *zap_work > 0)); | |
1da177e4 | 771 | tlb_end_vma(tlb, vma); |
51c6f666 RH |
772 | |
773 | return addr; | |
1da177e4 LT |
774 | } |
775 | ||
776 | #ifdef CONFIG_PREEMPT | |
777 | # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE) | |
778 | #else | |
779 | /* No preempt: go for improved straight-line efficiency */ | |
780 | # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE) | |
781 | #endif | |
782 | ||
783 | /** | |
784 | * unmap_vmas - unmap a range of memory covered by a list of vma's | |
785 | * @tlbp: address of the caller's struct mmu_gather | |
1da177e4 LT |
786 | * @vma: the starting vma |
787 | * @start_addr: virtual address at which to start unmapping | |
788 | * @end_addr: virtual address at which to end unmapping | |
789 | * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here | |
790 | * @details: details of nonlinear truncation or shared cache invalidation | |
791 | * | |
ee39b37b | 792 | * Returns the end address of the unmapping (restart addr if interrupted). |
1da177e4 | 793 | * |
508034a3 | 794 | * Unmap all pages in the vma list. |
1da177e4 | 795 | * |
508034a3 HD |
796 | * We aim to not hold locks for too long (for scheduling latency reasons). |
797 | * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to | |
1da177e4 LT |
798 | * return the ending mmu_gather to the caller. |
799 | * | |
800 | * Only addresses between `start' and `end' will be unmapped. | |
801 | * | |
802 | * The VMA list must be sorted in ascending virtual address order. | |
803 | * | |
804 | * unmap_vmas() assumes that the caller will flush the whole unmapped address | |
805 | * range after unmap_vmas() returns. So the only responsibility here is to | |
806 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() | |
807 | * drops the lock and schedules. | |
808 | */ | |
508034a3 | 809 | unsigned long unmap_vmas(struct mmu_gather **tlbp, |
1da177e4 LT |
810 | struct vm_area_struct *vma, unsigned long start_addr, |
811 | unsigned long end_addr, unsigned long *nr_accounted, | |
812 | struct zap_details *details) | |
813 | { | |
51c6f666 | 814 | long zap_work = ZAP_BLOCK_SIZE; |
1da177e4 LT |
815 | unsigned long tlb_start = 0; /* For tlb_finish_mmu */ |
816 | int tlb_start_valid = 0; | |
ee39b37b | 817 | unsigned long start = start_addr; |
1da177e4 | 818 | spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL; |
4d6ddfa9 | 819 | int fullmm = (*tlbp)->fullmm; |
1da177e4 LT |
820 | |
821 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) { | |
1da177e4 LT |
822 | unsigned long end; |
823 | ||
824 | start = max(vma->vm_start, start_addr); | |
825 | if (start >= vma->vm_end) | |
826 | continue; | |
827 | end = min(vma->vm_end, end_addr); | |
828 | if (end <= vma->vm_start) | |
829 | continue; | |
830 | ||
831 | if (vma->vm_flags & VM_ACCOUNT) | |
832 | *nr_accounted += (end - start) >> PAGE_SHIFT; | |
833 | ||
1da177e4 | 834 | while (start != end) { |
1da177e4 LT |
835 | if (!tlb_start_valid) { |
836 | tlb_start = start; | |
837 | tlb_start_valid = 1; | |
838 | } | |
839 | ||
51c6f666 | 840 | if (unlikely(is_vm_hugetlb_page(vma))) { |
1da177e4 | 841 | unmap_hugepage_range(vma, start, end); |
51c6f666 RH |
842 | zap_work -= (end - start) / |
843 | (HPAGE_SIZE / PAGE_SIZE); | |
844 | start = end; | |
845 | } else | |
846 | start = unmap_page_range(*tlbp, vma, | |
847 | start, end, &zap_work, details); | |
848 | ||
849 | if (zap_work > 0) { | |
850 | BUG_ON(start != end); | |
851 | break; | |
1da177e4 LT |
852 | } |
853 | ||
1da177e4 LT |
854 | tlb_finish_mmu(*tlbp, tlb_start, start); |
855 | ||
856 | if (need_resched() || | |
1da177e4 LT |
857 | (i_mmap_lock && need_lockbreak(i_mmap_lock))) { |
858 | if (i_mmap_lock) { | |
508034a3 | 859 | *tlbp = NULL; |
1da177e4 LT |
860 | goto out; |
861 | } | |
1da177e4 | 862 | cond_resched(); |
1da177e4 LT |
863 | } |
864 | ||
508034a3 | 865 | *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm); |
1da177e4 | 866 | tlb_start_valid = 0; |
51c6f666 | 867 | zap_work = ZAP_BLOCK_SIZE; |
1da177e4 LT |
868 | } |
869 | } | |
870 | out: | |
ee39b37b | 871 | return start; /* which is now the end (or restart) address */ |
1da177e4 LT |
872 | } |
873 | ||
874 | /** | |
875 | * zap_page_range - remove user pages in a given range | |
876 | * @vma: vm_area_struct holding the applicable pages | |
877 | * @address: starting address of pages to zap | |
878 | * @size: number of bytes to zap | |
879 | * @details: details of nonlinear truncation or shared cache invalidation | |
880 | */ | |
ee39b37b | 881 | unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address, |
1da177e4 LT |
882 | unsigned long size, struct zap_details *details) |
883 | { | |
884 | struct mm_struct *mm = vma->vm_mm; | |
885 | struct mmu_gather *tlb; | |
886 | unsigned long end = address + size; | |
887 | unsigned long nr_accounted = 0; | |
888 | ||
1da177e4 | 889 | lru_add_drain(); |
1da177e4 | 890 | tlb = tlb_gather_mmu(mm, 0); |
365e9c87 | 891 | update_hiwater_rss(mm); |
508034a3 HD |
892 | end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details); |
893 | if (tlb) | |
894 | tlb_finish_mmu(tlb, address, end); | |
ee39b37b | 895 | return end; |
1da177e4 LT |
896 | } |
897 | ||
898 | /* | |
899 | * Do a quick page-table lookup for a single page. | |
1da177e4 | 900 | */ |
6aab341e | 901 | struct page *follow_page(struct vm_area_struct *vma, unsigned long address, |
deceb6cd | 902 | unsigned int flags) |
1da177e4 LT |
903 | { |
904 | pgd_t *pgd; | |
905 | pud_t *pud; | |
906 | pmd_t *pmd; | |
907 | pte_t *ptep, pte; | |
deceb6cd | 908 | spinlock_t *ptl; |
1da177e4 | 909 | struct page *page; |
6aab341e | 910 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 911 | |
deceb6cd HD |
912 | page = follow_huge_addr(mm, address, flags & FOLL_WRITE); |
913 | if (!IS_ERR(page)) { | |
914 | BUG_ON(flags & FOLL_GET); | |
915 | goto out; | |
916 | } | |
1da177e4 | 917 | |
deceb6cd | 918 | page = NULL; |
1da177e4 LT |
919 | pgd = pgd_offset(mm, address); |
920 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
deceb6cd | 921 | goto no_page_table; |
1da177e4 LT |
922 | |
923 | pud = pud_offset(pgd, address); | |
924 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) | |
deceb6cd | 925 | goto no_page_table; |
1da177e4 LT |
926 | |
927 | pmd = pmd_offset(pud, address); | |
928 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) | |
deceb6cd HD |
929 | goto no_page_table; |
930 | ||
931 | if (pmd_huge(*pmd)) { | |
932 | BUG_ON(flags & FOLL_GET); | |
933 | page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE); | |
1da177e4 | 934 | goto out; |
deceb6cd | 935 | } |
1da177e4 | 936 | |
deceb6cd | 937 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); |
1da177e4 LT |
938 | if (!ptep) |
939 | goto out; | |
940 | ||
941 | pte = *ptep; | |
deceb6cd HD |
942 | if (!pte_present(pte)) |
943 | goto unlock; | |
944 | if ((flags & FOLL_WRITE) && !pte_write(pte)) | |
945 | goto unlock; | |
6aab341e LT |
946 | page = vm_normal_page(vma, address, pte); |
947 | if (unlikely(!page)) | |
deceb6cd | 948 | goto unlock; |
1da177e4 | 949 | |
deceb6cd HD |
950 | if (flags & FOLL_GET) |
951 | get_page(page); | |
952 | if (flags & FOLL_TOUCH) { | |
953 | if ((flags & FOLL_WRITE) && | |
954 | !pte_dirty(pte) && !PageDirty(page)) | |
955 | set_page_dirty(page); | |
956 | mark_page_accessed(page); | |
957 | } | |
958 | unlock: | |
959 | pte_unmap_unlock(ptep, ptl); | |
1da177e4 | 960 | out: |
deceb6cd | 961 | return page; |
1da177e4 | 962 | |
deceb6cd HD |
963 | no_page_table: |
964 | /* | |
965 | * When core dumping an enormous anonymous area that nobody | |
966 | * has touched so far, we don't want to allocate page tables. | |
967 | */ | |
968 | if (flags & FOLL_ANON) { | |
557ed1fa | 969 | page = ZERO_PAGE(0); |
deceb6cd HD |
970 | if (flags & FOLL_GET) |
971 | get_page(page); | |
972 | BUG_ON(flags & FOLL_WRITE); | |
973 | } | |
974 | return page; | |
1da177e4 LT |
975 | } |
976 | ||
1da177e4 LT |
977 | int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
978 | unsigned long start, int len, int write, int force, | |
979 | struct page **pages, struct vm_area_struct **vmas) | |
980 | { | |
981 | int i; | |
deceb6cd | 982 | unsigned int vm_flags; |
1da177e4 LT |
983 | |
984 | /* | |
985 | * Require read or write permissions. | |
986 | * If 'force' is set, we only require the "MAY" flags. | |
987 | */ | |
deceb6cd HD |
988 | vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); |
989 | vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); | |
1da177e4 LT |
990 | i = 0; |
991 | ||
992 | do { | |
deceb6cd HD |
993 | struct vm_area_struct *vma; |
994 | unsigned int foll_flags; | |
1da177e4 LT |
995 | |
996 | vma = find_extend_vma(mm, start); | |
997 | if (!vma && in_gate_area(tsk, start)) { | |
998 | unsigned long pg = start & PAGE_MASK; | |
999 | struct vm_area_struct *gate_vma = get_gate_vma(tsk); | |
1000 | pgd_t *pgd; | |
1001 | pud_t *pud; | |
1002 | pmd_t *pmd; | |
1003 | pte_t *pte; | |
1004 | if (write) /* user gate pages are read-only */ | |
1005 | return i ? : -EFAULT; | |
1006 | if (pg > TASK_SIZE) | |
1007 | pgd = pgd_offset_k(pg); | |
1008 | else | |
1009 | pgd = pgd_offset_gate(mm, pg); | |
1010 | BUG_ON(pgd_none(*pgd)); | |
1011 | pud = pud_offset(pgd, pg); | |
1012 | BUG_ON(pud_none(*pud)); | |
1013 | pmd = pmd_offset(pud, pg); | |
690dbe1c HD |
1014 | if (pmd_none(*pmd)) |
1015 | return i ? : -EFAULT; | |
1da177e4 | 1016 | pte = pte_offset_map(pmd, pg); |
690dbe1c HD |
1017 | if (pte_none(*pte)) { |
1018 | pte_unmap(pte); | |
1019 | return i ? : -EFAULT; | |
1020 | } | |
1da177e4 | 1021 | if (pages) { |
fa2a455b | 1022 | struct page *page = vm_normal_page(gate_vma, start, *pte); |
6aab341e LT |
1023 | pages[i] = page; |
1024 | if (page) | |
1025 | get_page(page); | |
1da177e4 LT |
1026 | } |
1027 | pte_unmap(pte); | |
1028 | if (vmas) | |
1029 | vmas[i] = gate_vma; | |
1030 | i++; | |
1031 | start += PAGE_SIZE; | |
1032 | len--; | |
1033 | continue; | |
1034 | } | |
1035 | ||
1ff80389 | 1036 | if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP)) |
deceb6cd | 1037 | || !(vm_flags & vma->vm_flags)) |
1da177e4 LT |
1038 | return i ? : -EFAULT; |
1039 | ||
1040 | if (is_vm_hugetlb_page(vma)) { | |
1041 | i = follow_hugetlb_page(mm, vma, pages, vmas, | |
1042 | &start, &len, i); | |
1043 | continue; | |
1044 | } | |
deceb6cd HD |
1045 | |
1046 | foll_flags = FOLL_TOUCH; | |
1047 | if (pages) | |
1048 | foll_flags |= FOLL_GET; | |
1049 | if (!write && !(vma->vm_flags & VM_LOCKED) && | |
54cb8821 NP |
1050 | (!vma->vm_ops || (!vma->vm_ops->nopage && |
1051 | !vma->vm_ops->fault))) | |
deceb6cd HD |
1052 | foll_flags |= FOLL_ANON; |
1053 | ||
1da177e4 | 1054 | do { |
08ef4729 | 1055 | struct page *page; |
1da177e4 | 1056 | |
462e00cc ES |
1057 | /* |
1058 | * If tsk is ooming, cut off its access to large memory | |
1059 | * allocations. It has a pending SIGKILL, but it can't | |
1060 | * be processed until returning to user space. | |
1061 | */ | |
1062 | if (unlikely(test_tsk_thread_flag(tsk, TIF_MEMDIE))) | |
1063 | return -ENOMEM; | |
1064 | ||
deceb6cd HD |
1065 | if (write) |
1066 | foll_flags |= FOLL_WRITE; | |
a68d2ebc | 1067 | |
deceb6cd | 1068 | cond_resched(); |
6aab341e | 1069 | while (!(page = follow_page(vma, start, foll_flags))) { |
deceb6cd | 1070 | int ret; |
83c54070 | 1071 | ret = handle_mm_fault(mm, vma, start, |
deceb6cd | 1072 | foll_flags & FOLL_WRITE); |
83c54070 NP |
1073 | if (ret & VM_FAULT_ERROR) { |
1074 | if (ret & VM_FAULT_OOM) | |
1075 | return i ? i : -ENOMEM; | |
1076 | else if (ret & VM_FAULT_SIGBUS) | |
1077 | return i ? i : -EFAULT; | |
1078 | BUG(); | |
1079 | } | |
1080 | if (ret & VM_FAULT_MAJOR) | |
1081 | tsk->maj_flt++; | |
1082 | else | |
1083 | tsk->min_flt++; | |
1084 | ||
a68d2ebc | 1085 | /* |
83c54070 NP |
1086 | * The VM_FAULT_WRITE bit tells us that |
1087 | * do_wp_page has broken COW when necessary, | |
1088 | * even if maybe_mkwrite decided not to set | |
1089 | * pte_write. We can thus safely do subsequent | |
1090 | * page lookups as if they were reads. | |
a68d2ebc LT |
1091 | */ |
1092 | if (ret & VM_FAULT_WRITE) | |
deceb6cd | 1093 | foll_flags &= ~FOLL_WRITE; |
83c54070 | 1094 | |
7f7bbbe5 | 1095 | cond_resched(); |
1da177e4 LT |
1096 | } |
1097 | if (pages) { | |
08ef4729 | 1098 | pages[i] = page; |
03beb076 | 1099 | |
a6f36be3 | 1100 | flush_anon_page(vma, page, start); |
08ef4729 | 1101 | flush_dcache_page(page); |
1da177e4 LT |
1102 | } |
1103 | if (vmas) | |
1104 | vmas[i] = vma; | |
1105 | i++; | |
1106 | start += PAGE_SIZE; | |
1107 | len--; | |
08ef4729 | 1108 | } while (len && start < vma->vm_end); |
08ef4729 | 1109 | } while (len); |
1da177e4 LT |
1110 | return i; |
1111 | } | |
1da177e4 LT |
1112 | EXPORT_SYMBOL(get_user_pages); |
1113 | ||
49c91fb0 | 1114 | pte_t * fastcall get_locked_pte(struct mm_struct *mm, unsigned long addr, spinlock_t **ptl) |
c9cfcddf LT |
1115 | { |
1116 | pgd_t * pgd = pgd_offset(mm, addr); | |
1117 | pud_t * pud = pud_alloc(mm, pgd, addr); | |
1118 | if (pud) { | |
49c91fb0 | 1119 | pmd_t * pmd = pmd_alloc(mm, pud, addr); |
c9cfcddf LT |
1120 | if (pmd) |
1121 | return pte_alloc_map_lock(mm, pmd, addr, ptl); | |
1122 | } | |
1123 | return NULL; | |
1124 | } | |
1125 | ||
238f58d8 LT |
1126 | /* |
1127 | * This is the old fallback for page remapping. | |
1128 | * | |
1129 | * For historical reasons, it only allows reserved pages. Only | |
1130 | * old drivers should use this, and they needed to mark their | |
1131 | * pages reserved for the old functions anyway. | |
1132 | */ | |
1133 | static int insert_page(struct mm_struct *mm, unsigned long addr, struct page *page, pgprot_t prot) | |
1134 | { | |
1135 | int retval; | |
c9cfcddf | 1136 | pte_t *pte; |
238f58d8 LT |
1137 | spinlock_t *ptl; |
1138 | ||
1139 | retval = -EINVAL; | |
a145dd41 | 1140 | if (PageAnon(page)) |
238f58d8 LT |
1141 | goto out; |
1142 | retval = -ENOMEM; | |
1143 | flush_dcache_page(page); | |
c9cfcddf | 1144 | pte = get_locked_pte(mm, addr, &ptl); |
238f58d8 LT |
1145 | if (!pte) |
1146 | goto out; | |
1147 | retval = -EBUSY; | |
1148 | if (!pte_none(*pte)) | |
1149 | goto out_unlock; | |
1150 | ||
1151 | /* Ok, finally just insert the thing.. */ | |
1152 | get_page(page); | |
1153 | inc_mm_counter(mm, file_rss); | |
1154 | page_add_file_rmap(page); | |
1155 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); | |
1156 | ||
1157 | retval = 0; | |
1158 | out_unlock: | |
1159 | pte_unmap_unlock(pte, ptl); | |
1160 | out: | |
1161 | return retval; | |
1162 | } | |
1163 | ||
bfa5bf6d REB |
1164 | /** |
1165 | * vm_insert_page - insert single page into user vma | |
1166 | * @vma: user vma to map to | |
1167 | * @addr: target user address of this page | |
1168 | * @page: source kernel page | |
1169 | * | |
a145dd41 LT |
1170 | * This allows drivers to insert individual pages they've allocated |
1171 | * into a user vma. | |
1172 | * | |
1173 | * The page has to be a nice clean _individual_ kernel allocation. | |
1174 | * If you allocate a compound page, you need to have marked it as | |
1175 | * such (__GFP_COMP), or manually just split the page up yourself | |
8dfcc9ba | 1176 | * (see split_page()). |
a145dd41 LT |
1177 | * |
1178 | * NOTE! Traditionally this was done with "remap_pfn_range()" which | |
1179 | * took an arbitrary page protection parameter. This doesn't allow | |
1180 | * that. Your vma protection will have to be set up correctly, which | |
1181 | * means that if you want a shared writable mapping, you'd better | |
1182 | * ask for a shared writable mapping! | |
1183 | * | |
1184 | * The page does not need to be reserved. | |
1185 | */ | |
1186 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, struct page *page) | |
1187 | { | |
1188 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
1189 | return -EFAULT; | |
1190 | if (!page_count(page)) | |
1191 | return -EINVAL; | |
4d7672b4 | 1192 | vma->vm_flags |= VM_INSERTPAGE; |
a145dd41 LT |
1193 | return insert_page(vma->vm_mm, addr, page, vma->vm_page_prot); |
1194 | } | |
e3c3374f | 1195 | EXPORT_SYMBOL(vm_insert_page); |
a145dd41 | 1196 | |
e0dc0d8f NP |
1197 | /** |
1198 | * vm_insert_pfn - insert single pfn into user vma | |
1199 | * @vma: user vma to map to | |
1200 | * @addr: target user address of this page | |
1201 | * @pfn: source kernel pfn | |
1202 | * | |
1203 | * Similar to vm_inert_page, this allows drivers to insert individual pages | |
1204 | * they've allocated into a user vma. Same comments apply. | |
1205 | * | |
1206 | * This function should only be called from a vm_ops->fault handler, and | |
1207 | * in that case the handler should return NULL. | |
1208 | */ | |
1209 | int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, | |
1210 | unsigned long pfn) | |
1211 | { | |
1212 | struct mm_struct *mm = vma->vm_mm; | |
1213 | int retval; | |
1214 | pte_t *pte, entry; | |
1215 | spinlock_t *ptl; | |
1216 | ||
1217 | BUG_ON(!(vma->vm_flags & VM_PFNMAP)); | |
1218 | BUG_ON(is_cow_mapping(vma->vm_flags)); | |
1219 | ||
1220 | retval = -ENOMEM; | |
1221 | pte = get_locked_pte(mm, addr, &ptl); | |
1222 | if (!pte) | |
1223 | goto out; | |
1224 | retval = -EBUSY; | |
1225 | if (!pte_none(*pte)) | |
1226 | goto out_unlock; | |
1227 | ||
1228 | /* Ok, finally just insert the thing.. */ | |
1229 | entry = pfn_pte(pfn, vma->vm_page_prot); | |
1230 | set_pte_at(mm, addr, pte, entry); | |
1231 | update_mmu_cache(vma, addr, entry); | |
1232 | ||
1233 | retval = 0; | |
1234 | out_unlock: | |
1235 | pte_unmap_unlock(pte, ptl); | |
1236 | ||
1237 | out: | |
1238 | return retval; | |
1239 | } | |
1240 | EXPORT_SYMBOL(vm_insert_pfn); | |
1241 | ||
1da177e4 LT |
1242 | /* |
1243 | * maps a range of physical memory into the requested pages. the old | |
1244 | * mappings are removed. any references to nonexistent pages results | |
1245 | * in null mappings (currently treated as "copy-on-access") | |
1246 | */ | |
1247 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
1248 | unsigned long addr, unsigned long end, | |
1249 | unsigned long pfn, pgprot_t prot) | |
1250 | { | |
1251 | pte_t *pte; | |
c74df32c | 1252 | spinlock_t *ptl; |
1da177e4 | 1253 | |
c74df32c | 1254 | pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1da177e4 LT |
1255 | if (!pte) |
1256 | return -ENOMEM; | |
6606c3e0 | 1257 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1258 | do { |
1259 | BUG_ON(!pte_none(*pte)); | |
b5810039 | 1260 | set_pte_at(mm, addr, pte, pfn_pte(pfn, prot)); |
1da177e4 LT |
1261 | pfn++; |
1262 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 1263 | arch_leave_lazy_mmu_mode(); |
c74df32c | 1264 | pte_unmap_unlock(pte - 1, ptl); |
1da177e4 LT |
1265 | return 0; |
1266 | } | |
1267 | ||
1268 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
1269 | unsigned long addr, unsigned long end, | |
1270 | unsigned long pfn, pgprot_t prot) | |
1271 | { | |
1272 | pmd_t *pmd; | |
1273 | unsigned long next; | |
1274 | ||
1275 | pfn -= addr >> PAGE_SHIFT; | |
1276 | pmd = pmd_alloc(mm, pud, addr); | |
1277 | if (!pmd) | |
1278 | return -ENOMEM; | |
1279 | do { | |
1280 | next = pmd_addr_end(addr, end); | |
1281 | if (remap_pte_range(mm, pmd, addr, next, | |
1282 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1283 | return -ENOMEM; | |
1284 | } while (pmd++, addr = next, addr != end); | |
1285 | return 0; | |
1286 | } | |
1287 | ||
1288 | static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, | |
1289 | unsigned long addr, unsigned long end, | |
1290 | unsigned long pfn, pgprot_t prot) | |
1291 | { | |
1292 | pud_t *pud; | |
1293 | unsigned long next; | |
1294 | ||
1295 | pfn -= addr >> PAGE_SHIFT; | |
1296 | pud = pud_alloc(mm, pgd, addr); | |
1297 | if (!pud) | |
1298 | return -ENOMEM; | |
1299 | do { | |
1300 | next = pud_addr_end(addr, end); | |
1301 | if (remap_pmd_range(mm, pud, addr, next, | |
1302 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1303 | return -ENOMEM; | |
1304 | } while (pud++, addr = next, addr != end); | |
1305 | return 0; | |
1306 | } | |
1307 | ||
bfa5bf6d REB |
1308 | /** |
1309 | * remap_pfn_range - remap kernel memory to userspace | |
1310 | * @vma: user vma to map to | |
1311 | * @addr: target user address to start at | |
1312 | * @pfn: physical address of kernel memory | |
1313 | * @size: size of map area | |
1314 | * @prot: page protection flags for this mapping | |
1315 | * | |
1316 | * Note: this is only safe if the mm semaphore is held when called. | |
1317 | */ | |
1da177e4 LT |
1318 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
1319 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
1320 | { | |
1321 | pgd_t *pgd; | |
1322 | unsigned long next; | |
2d15cab8 | 1323 | unsigned long end = addr + PAGE_ALIGN(size); |
1da177e4 LT |
1324 | struct mm_struct *mm = vma->vm_mm; |
1325 | int err; | |
1326 | ||
1327 | /* | |
1328 | * Physically remapped pages are special. Tell the | |
1329 | * rest of the world about it: | |
1330 | * VM_IO tells people not to look at these pages | |
1331 | * (accesses can have side effects). | |
0b14c179 HD |
1332 | * VM_RESERVED is specified all over the place, because |
1333 | * in 2.4 it kept swapout's vma scan off this vma; but | |
1334 | * in 2.6 the LRU scan won't even find its pages, so this | |
1335 | * flag means no more than count its pages in reserved_vm, | |
1336 | * and omit it from core dump, even when VM_IO turned off. | |
6aab341e LT |
1337 | * VM_PFNMAP tells the core MM that the base pages are just |
1338 | * raw PFN mappings, and do not have a "struct page" associated | |
1339 | * with them. | |
fb155c16 LT |
1340 | * |
1341 | * There's a horrible special case to handle copy-on-write | |
1342 | * behaviour that some programs depend on. We mark the "original" | |
1343 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". | |
1da177e4 | 1344 | */ |
67121172 | 1345 | if (is_cow_mapping(vma->vm_flags)) { |
fb155c16 | 1346 | if (addr != vma->vm_start || end != vma->vm_end) |
7fc7e2ee | 1347 | return -EINVAL; |
fb155c16 LT |
1348 | vma->vm_pgoff = pfn; |
1349 | } | |
1350 | ||
6aab341e | 1351 | vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP; |
1da177e4 LT |
1352 | |
1353 | BUG_ON(addr >= end); | |
1354 | pfn -= addr >> PAGE_SHIFT; | |
1355 | pgd = pgd_offset(mm, addr); | |
1356 | flush_cache_range(vma, addr, end); | |
1da177e4 LT |
1357 | do { |
1358 | next = pgd_addr_end(addr, end); | |
1359 | err = remap_pud_range(mm, pgd, addr, next, | |
1360 | pfn + (addr >> PAGE_SHIFT), prot); | |
1361 | if (err) | |
1362 | break; | |
1363 | } while (pgd++, addr = next, addr != end); | |
1da177e4 LT |
1364 | return err; |
1365 | } | |
1366 | EXPORT_SYMBOL(remap_pfn_range); | |
1367 | ||
aee16b3c JF |
1368 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
1369 | unsigned long addr, unsigned long end, | |
1370 | pte_fn_t fn, void *data) | |
1371 | { | |
1372 | pte_t *pte; | |
1373 | int err; | |
1374 | struct page *pmd_page; | |
94909914 | 1375 | spinlock_t *uninitialized_var(ptl); |
aee16b3c JF |
1376 | |
1377 | pte = (mm == &init_mm) ? | |
1378 | pte_alloc_kernel(pmd, addr) : | |
1379 | pte_alloc_map_lock(mm, pmd, addr, &ptl); | |
1380 | if (!pte) | |
1381 | return -ENOMEM; | |
1382 | ||
1383 | BUG_ON(pmd_huge(*pmd)); | |
1384 | ||
1385 | pmd_page = pmd_page(*pmd); | |
1386 | ||
1387 | do { | |
1388 | err = fn(pte, pmd_page, addr, data); | |
1389 | if (err) | |
1390 | break; | |
1391 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
1392 | ||
1393 | if (mm != &init_mm) | |
1394 | pte_unmap_unlock(pte-1, ptl); | |
1395 | return err; | |
1396 | } | |
1397 | ||
1398 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, | |
1399 | unsigned long addr, unsigned long end, | |
1400 | pte_fn_t fn, void *data) | |
1401 | { | |
1402 | pmd_t *pmd; | |
1403 | unsigned long next; | |
1404 | int err; | |
1405 | ||
1406 | pmd = pmd_alloc(mm, pud, addr); | |
1407 | if (!pmd) | |
1408 | return -ENOMEM; | |
1409 | do { | |
1410 | next = pmd_addr_end(addr, end); | |
1411 | err = apply_to_pte_range(mm, pmd, addr, next, fn, data); | |
1412 | if (err) | |
1413 | break; | |
1414 | } while (pmd++, addr = next, addr != end); | |
1415 | return err; | |
1416 | } | |
1417 | ||
1418 | static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd, | |
1419 | unsigned long addr, unsigned long end, | |
1420 | pte_fn_t fn, void *data) | |
1421 | { | |
1422 | pud_t *pud; | |
1423 | unsigned long next; | |
1424 | int err; | |
1425 | ||
1426 | pud = pud_alloc(mm, pgd, addr); | |
1427 | if (!pud) | |
1428 | return -ENOMEM; | |
1429 | do { | |
1430 | next = pud_addr_end(addr, end); | |
1431 | err = apply_to_pmd_range(mm, pud, addr, next, fn, data); | |
1432 | if (err) | |
1433 | break; | |
1434 | } while (pud++, addr = next, addr != end); | |
1435 | return err; | |
1436 | } | |
1437 | ||
1438 | /* | |
1439 | * Scan a region of virtual memory, filling in page tables as necessary | |
1440 | * and calling a provided function on each leaf page table. | |
1441 | */ | |
1442 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, | |
1443 | unsigned long size, pte_fn_t fn, void *data) | |
1444 | { | |
1445 | pgd_t *pgd; | |
1446 | unsigned long next; | |
1447 | unsigned long end = addr + size; | |
1448 | int err; | |
1449 | ||
1450 | BUG_ON(addr >= end); | |
1451 | pgd = pgd_offset(mm, addr); | |
1452 | do { | |
1453 | next = pgd_addr_end(addr, end); | |
1454 | err = apply_to_pud_range(mm, pgd, addr, next, fn, data); | |
1455 | if (err) | |
1456 | break; | |
1457 | } while (pgd++, addr = next, addr != end); | |
1458 | return err; | |
1459 | } | |
1460 | EXPORT_SYMBOL_GPL(apply_to_page_range); | |
1461 | ||
8f4e2101 HD |
1462 | /* |
1463 | * handle_pte_fault chooses page fault handler according to an entry | |
1464 | * which was read non-atomically. Before making any commitment, on | |
1465 | * those architectures or configurations (e.g. i386 with PAE) which | |
1466 | * might give a mix of unmatched parts, do_swap_page and do_file_page | |
1467 | * must check under lock before unmapping the pte and proceeding | |
1468 | * (but do_wp_page is only called after already making such a check; | |
1469 | * and do_anonymous_page and do_no_page can safely check later on). | |
1470 | */ | |
4c21e2f2 | 1471 | static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, |
8f4e2101 HD |
1472 | pte_t *page_table, pte_t orig_pte) |
1473 | { | |
1474 | int same = 1; | |
1475 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) | |
1476 | if (sizeof(pte_t) > sizeof(unsigned long)) { | |
4c21e2f2 HD |
1477 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
1478 | spin_lock(ptl); | |
8f4e2101 | 1479 | same = pte_same(*page_table, orig_pte); |
4c21e2f2 | 1480 | spin_unlock(ptl); |
8f4e2101 HD |
1481 | } |
1482 | #endif | |
1483 | pte_unmap(page_table); | |
1484 | return same; | |
1485 | } | |
1486 | ||
1da177e4 LT |
1487 | /* |
1488 | * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when | |
1489 | * servicing faults for write access. In the normal case, do always want | |
1490 | * pte_mkwrite. But get_user_pages can cause write faults for mappings | |
1491 | * that do not have writing enabled, when used by access_process_vm. | |
1492 | */ | |
1493 | static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma) | |
1494 | { | |
1495 | if (likely(vma->vm_flags & VM_WRITE)) | |
1496 | pte = pte_mkwrite(pte); | |
1497 | return pte; | |
1498 | } | |
1499 | ||
9de455b2 | 1500 | static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) |
6aab341e LT |
1501 | { |
1502 | /* | |
1503 | * If the source page was a PFN mapping, we don't have | |
1504 | * a "struct page" for it. We do a best-effort copy by | |
1505 | * just copying from the original user address. If that | |
1506 | * fails, we just zero-fill it. Live with it. | |
1507 | */ | |
1508 | if (unlikely(!src)) { | |
1509 | void *kaddr = kmap_atomic(dst, KM_USER0); | |
5d2a2dbb LT |
1510 | void __user *uaddr = (void __user *)(va & PAGE_MASK); |
1511 | ||
1512 | /* | |
1513 | * This really shouldn't fail, because the page is there | |
1514 | * in the page tables. But it might just be unreadable, | |
1515 | * in which case we just give up and fill the result with | |
1516 | * zeroes. | |
1517 | */ | |
1518 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) | |
6aab341e LT |
1519 | memset(kaddr, 0, PAGE_SIZE); |
1520 | kunmap_atomic(kaddr, KM_USER0); | |
c4ec7b0d | 1521 | flush_dcache_page(dst); |
6aab341e | 1522 | return; |
9de455b2 | 1523 | |
6aab341e | 1524 | } |
9de455b2 | 1525 | copy_user_highpage(dst, src, va, vma); |
6aab341e LT |
1526 | } |
1527 | ||
1da177e4 LT |
1528 | /* |
1529 | * This routine handles present pages, when users try to write | |
1530 | * to a shared page. It is done by copying the page to a new address | |
1531 | * and decrementing the shared-page counter for the old page. | |
1532 | * | |
1da177e4 LT |
1533 | * Note that this routine assumes that the protection checks have been |
1534 | * done by the caller (the low-level page fault routine in most cases). | |
1535 | * Thus we can safely just mark it writable once we've done any necessary | |
1536 | * COW. | |
1537 | * | |
1538 | * We also mark the page dirty at this point even though the page will | |
1539 | * change only once the write actually happens. This avoids a few races, | |
1540 | * and potentially makes it more efficient. | |
1541 | * | |
8f4e2101 HD |
1542 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
1543 | * but allow concurrent faults), with pte both mapped and locked. | |
1544 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 1545 | */ |
65500d23 HD |
1546 | static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, |
1547 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
8f4e2101 | 1548 | spinlock_t *ptl, pte_t orig_pte) |
1da177e4 | 1549 | { |
e5bbe4df | 1550 | struct page *old_page, *new_page; |
1da177e4 | 1551 | pte_t entry; |
83c54070 | 1552 | int reuse = 0, ret = 0; |
a200ee18 | 1553 | int page_mkwrite = 0; |
d08b3851 | 1554 | struct page *dirty_page = NULL; |
1da177e4 | 1555 | |
6aab341e | 1556 | old_page = vm_normal_page(vma, address, orig_pte); |
6aab341e LT |
1557 | if (!old_page) |
1558 | goto gotten; | |
1da177e4 | 1559 | |
d08b3851 | 1560 | /* |
ee6a6457 PZ |
1561 | * Take out anonymous pages first, anonymous shared vmas are |
1562 | * not dirty accountable. | |
d08b3851 | 1563 | */ |
ee6a6457 PZ |
1564 | if (PageAnon(old_page)) { |
1565 | if (!TestSetPageLocked(old_page)) { | |
1566 | reuse = can_share_swap_page(old_page); | |
1567 | unlock_page(old_page); | |
1568 | } | |
1569 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == | |
d08b3851 | 1570 | (VM_WRITE|VM_SHARED))) { |
ee6a6457 PZ |
1571 | /* |
1572 | * Only catch write-faults on shared writable pages, | |
1573 | * read-only shared pages can get COWed by | |
1574 | * get_user_pages(.write=1, .force=1). | |
1575 | */ | |
9637a5ef DH |
1576 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
1577 | /* | |
1578 | * Notify the address space that the page is about to | |
1579 | * become writable so that it can prohibit this or wait | |
1580 | * for the page to get into an appropriate state. | |
1581 | * | |
1582 | * We do this without the lock held, so that it can | |
1583 | * sleep if it needs to. | |
1584 | */ | |
1585 | page_cache_get(old_page); | |
1586 | pte_unmap_unlock(page_table, ptl); | |
1587 | ||
1588 | if (vma->vm_ops->page_mkwrite(vma, old_page) < 0) | |
1589 | goto unwritable_page; | |
1590 | ||
9637a5ef DH |
1591 | /* |
1592 | * Since we dropped the lock we need to revalidate | |
1593 | * the PTE as someone else may have changed it. If | |
1594 | * they did, we just return, as we can count on the | |
1595 | * MMU to tell us if they didn't also make it writable. | |
1596 | */ | |
1597 | page_table = pte_offset_map_lock(mm, pmd, address, | |
1598 | &ptl); | |
c3704ceb | 1599 | page_cache_release(old_page); |
9637a5ef DH |
1600 | if (!pte_same(*page_table, orig_pte)) |
1601 | goto unlock; | |
a200ee18 PZ |
1602 | |
1603 | page_mkwrite = 1; | |
1da177e4 | 1604 | } |
d08b3851 PZ |
1605 | dirty_page = old_page; |
1606 | get_page(dirty_page); | |
9637a5ef | 1607 | reuse = 1; |
9637a5ef DH |
1608 | } |
1609 | ||
1610 | if (reuse) { | |
1611 | flush_cache_page(vma, address, pte_pfn(orig_pte)); | |
1612 | entry = pte_mkyoung(orig_pte); | |
1613 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
954ffcb3 | 1614 | if (ptep_set_access_flags(vma, address, page_table, entry,1)) |
8dab5241 | 1615 | update_mmu_cache(vma, address, entry); |
9637a5ef DH |
1616 | ret |= VM_FAULT_WRITE; |
1617 | goto unlock; | |
1da177e4 | 1618 | } |
1da177e4 LT |
1619 | |
1620 | /* | |
1621 | * Ok, we need to copy. Oh, well.. | |
1622 | */ | |
b5810039 | 1623 | page_cache_get(old_page); |
920fc356 | 1624 | gotten: |
8f4e2101 | 1625 | pte_unmap_unlock(page_table, ptl); |
1da177e4 LT |
1626 | |
1627 | if (unlikely(anon_vma_prepare(vma))) | |
65500d23 | 1628 | goto oom; |
557ed1fa NP |
1629 | VM_BUG_ON(old_page == ZERO_PAGE(0)); |
1630 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); | |
1631 | if (!new_page) | |
1632 | goto oom; | |
1633 | cow_user_page(new_page, old_page, address, vma); | |
65500d23 | 1634 | |
1da177e4 LT |
1635 | /* |
1636 | * Re-check the pte - we dropped the lock | |
1637 | */ | |
8f4e2101 | 1638 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
65500d23 | 1639 | if (likely(pte_same(*page_table, orig_pte))) { |
920fc356 | 1640 | if (old_page) { |
7de6b805 | 1641 | page_remove_rmap(old_page, vma); |
920fc356 HD |
1642 | if (!PageAnon(old_page)) { |
1643 | dec_mm_counter(mm, file_rss); | |
1644 | inc_mm_counter(mm, anon_rss); | |
1645 | } | |
1646 | } else | |
4294621f | 1647 | inc_mm_counter(mm, anon_rss); |
eca35133 | 1648 | flush_cache_page(vma, address, pte_pfn(orig_pte)); |
65500d23 HD |
1649 | entry = mk_pte(new_page, vma->vm_page_prot); |
1650 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
4ce072f1 SS |
1651 | /* |
1652 | * Clear the pte entry and flush it first, before updating the | |
1653 | * pte with the new entry. This will avoid a race condition | |
1654 | * seen in the presence of one thread doing SMC and another | |
1655 | * thread doing COW. | |
1656 | */ | |
1657 | ptep_clear_flush(vma, address, page_table); | |
1658 | set_pte_at(mm, address, page_table, entry); | |
65500d23 | 1659 | update_mmu_cache(vma, address, entry); |
1da177e4 | 1660 | lru_cache_add_active(new_page); |
9617d95e | 1661 | page_add_new_anon_rmap(new_page, vma, address); |
1da177e4 LT |
1662 | |
1663 | /* Free the old page.. */ | |
1664 | new_page = old_page; | |
f33ea7f4 | 1665 | ret |= VM_FAULT_WRITE; |
1da177e4 | 1666 | } |
920fc356 HD |
1667 | if (new_page) |
1668 | page_cache_release(new_page); | |
1669 | if (old_page) | |
1670 | page_cache_release(old_page); | |
65500d23 | 1671 | unlock: |
8f4e2101 | 1672 | pte_unmap_unlock(page_table, ptl); |
d08b3851 | 1673 | if (dirty_page) { |
79352894 NP |
1674 | /* |
1675 | * Yes, Virginia, this is actually required to prevent a race | |
1676 | * with clear_page_dirty_for_io() from clearing the page dirty | |
1677 | * bit after it clear all dirty ptes, but before a racing | |
1678 | * do_wp_page installs a dirty pte. | |
1679 | * | |
1680 | * do_no_page is protected similarly. | |
1681 | */ | |
1682 | wait_on_page_locked(dirty_page); | |
a200ee18 | 1683 | set_page_dirty_balance(dirty_page, page_mkwrite); |
d08b3851 PZ |
1684 | put_page(dirty_page); |
1685 | } | |
f33ea7f4 | 1686 | return ret; |
65500d23 | 1687 | oom: |
920fc356 HD |
1688 | if (old_page) |
1689 | page_cache_release(old_page); | |
1da177e4 | 1690 | return VM_FAULT_OOM; |
9637a5ef DH |
1691 | |
1692 | unwritable_page: | |
1693 | page_cache_release(old_page); | |
1694 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
1695 | } |
1696 | ||
1697 | /* | |
1698 | * Helper functions for unmap_mapping_range(). | |
1699 | * | |
1700 | * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __ | |
1701 | * | |
1702 | * We have to restart searching the prio_tree whenever we drop the lock, | |
1703 | * since the iterator is only valid while the lock is held, and anyway | |
1704 | * a later vma might be split and reinserted earlier while lock dropped. | |
1705 | * | |
1706 | * The list of nonlinear vmas could be handled more efficiently, using | |
1707 | * a placeholder, but handle it in the same way until a need is shown. | |
1708 | * It is important to search the prio_tree before nonlinear list: a vma | |
1709 | * may become nonlinear and be shifted from prio_tree to nonlinear list | |
1710 | * while the lock is dropped; but never shifted from list to prio_tree. | |
1711 | * | |
1712 | * In order to make forward progress despite restarting the search, | |
1713 | * vm_truncate_count is used to mark a vma as now dealt with, so we can | |
1714 | * quickly skip it next time around. Since the prio_tree search only | |
1715 | * shows us those vmas affected by unmapping the range in question, we | |
1716 | * can't efficiently keep all vmas in step with mapping->truncate_count: | |
1717 | * so instead reset them all whenever it wraps back to 0 (then go to 1). | |
1718 | * mapping->truncate_count and vma->vm_truncate_count are protected by | |
1719 | * i_mmap_lock. | |
1720 | * | |
1721 | * In order to make forward progress despite repeatedly restarting some | |
ee39b37b | 1722 | * large vma, note the restart_addr from unmap_vmas when it breaks out: |
1da177e4 LT |
1723 | * and restart from that address when we reach that vma again. It might |
1724 | * have been split or merged, shrunk or extended, but never shifted: so | |
1725 | * restart_addr remains valid so long as it remains in the vma's range. | |
1726 | * unmap_mapping_range forces truncate_count to leap over page-aligned | |
1727 | * values so we can save vma's restart_addr in its truncate_count field. | |
1728 | */ | |
1729 | #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK)) | |
1730 | ||
1731 | static void reset_vma_truncate_counts(struct address_space *mapping) | |
1732 | { | |
1733 | struct vm_area_struct *vma; | |
1734 | struct prio_tree_iter iter; | |
1735 | ||
1736 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX) | |
1737 | vma->vm_truncate_count = 0; | |
1738 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) | |
1739 | vma->vm_truncate_count = 0; | |
1740 | } | |
1741 | ||
1742 | static int unmap_mapping_range_vma(struct vm_area_struct *vma, | |
1743 | unsigned long start_addr, unsigned long end_addr, | |
1744 | struct zap_details *details) | |
1745 | { | |
1746 | unsigned long restart_addr; | |
1747 | int need_break; | |
1748 | ||
d00806b1 NP |
1749 | /* |
1750 | * files that support invalidating or truncating portions of the | |
d0217ac0 | 1751 | * file from under mmaped areas must have their ->fault function |
83c54070 NP |
1752 | * return a locked page (and set VM_FAULT_LOCKED in the return). |
1753 | * This provides synchronisation against concurrent unmapping here. | |
d00806b1 | 1754 | */ |
d00806b1 | 1755 | |
1da177e4 LT |
1756 | again: |
1757 | restart_addr = vma->vm_truncate_count; | |
1758 | if (is_restart_addr(restart_addr) && start_addr < restart_addr) { | |
1759 | start_addr = restart_addr; | |
1760 | if (start_addr >= end_addr) { | |
1761 | /* Top of vma has been split off since last time */ | |
1762 | vma->vm_truncate_count = details->truncate_count; | |
1763 | return 0; | |
1764 | } | |
1765 | } | |
1766 | ||
ee39b37b HD |
1767 | restart_addr = zap_page_range(vma, start_addr, |
1768 | end_addr - start_addr, details); | |
1da177e4 LT |
1769 | need_break = need_resched() || |
1770 | need_lockbreak(details->i_mmap_lock); | |
1771 | ||
ee39b37b | 1772 | if (restart_addr >= end_addr) { |
1da177e4 LT |
1773 | /* We have now completed this vma: mark it so */ |
1774 | vma->vm_truncate_count = details->truncate_count; | |
1775 | if (!need_break) | |
1776 | return 0; | |
1777 | } else { | |
1778 | /* Note restart_addr in vma's truncate_count field */ | |
ee39b37b | 1779 | vma->vm_truncate_count = restart_addr; |
1da177e4 LT |
1780 | if (!need_break) |
1781 | goto again; | |
1782 | } | |
1783 | ||
1784 | spin_unlock(details->i_mmap_lock); | |
1785 | cond_resched(); | |
1786 | spin_lock(details->i_mmap_lock); | |
1787 | return -EINTR; | |
1788 | } | |
1789 | ||
1790 | static inline void unmap_mapping_range_tree(struct prio_tree_root *root, | |
1791 | struct zap_details *details) | |
1792 | { | |
1793 | struct vm_area_struct *vma; | |
1794 | struct prio_tree_iter iter; | |
1795 | pgoff_t vba, vea, zba, zea; | |
1796 | ||
1797 | restart: | |
1798 | vma_prio_tree_foreach(vma, &iter, root, | |
1799 | details->first_index, details->last_index) { | |
1800 | /* Skip quickly over those we have already dealt with */ | |
1801 | if (vma->vm_truncate_count == details->truncate_count) | |
1802 | continue; | |
1803 | ||
1804 | vba = vma->vm_pgoff; | |
1805 | vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1; | |
1806 | /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */ | |
1807 | zba = details->first_index; | |
1808 | if (zba < vba) | |
1809 | zba = vba; | |
1810 | zea = details->last_index; | |
1811 | if (zea > vea) | |
1812 | zea = vea; | |
1813 | ||
1814 | if (unmap_mapping_range_vma(vma, | |
1815 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, | |
1816 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, | |
1817 | details) < 0) | |
1818 | goto restart; | |
1819 | } | |
1820 | } | |
1821 | ||
1822 | static inline void unmap_mapping_range_list(struct list_head *head, | |
1823 | struct zap_details *details) | |
1824 | { | |
1825 | struct vm_area_struct *vma; | |
1826 | ||
1827 | /* | |
1828 | * In nonlinear VMAs there is no correspondence between virtual address | |
1829 | * offset and file offset. So we must perform an exhaustive search | |
1830 | * across *all* the pages in each nonlinear VMA, not just the pages | |
1831 | * whose virtual address lies outside the file truncation point. | |
1832 | */ | |
1833 | restart: | |
1834 | list_for_each_entry(vma, head, shared.vm_set.list) { | |
1835 | /* Skip quickly over those we have already dealt with */ | |
1836 | if (vma->vm_truncate_count == details->truncate_count) | |
1837 | continue; | |
1838 | details->nonlinear_vma = vma; | |
1839 | if (unmap_mapping_range_vma(vma, vma->vm_start, | |
1840 | vma->vm_end, details) < 0) | |
1841 | goto restart; | |
1842 | } | |
1843 | } | |
1844 | ||
1845 | /** | |
72fd4a35 | 1846 | * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file. |
3d41088f | 1847 | * @mapping: the address space containing mmaps to be unmapped. |
1da177e4 LT |
1848 | * @holebegin: byte in first page to unmap, relative to the start of |
1849 | * the underlying file. This will be rounded down to a PAGE_SIZE | |
1850 | * boundary. Note that this is different from vmtruncate(), which | |
1851 | * must keep the partial page. In contrast, we must get rid of | |
1852 | * partial pages. | |
1853 | * @holelen: size of prospective hole in bytes. This will be rounded | |
1854 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the | |
1855 | * end of the file. | |
1856 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; | |
1857 | * but 0 when invalidating pagecache, don't throw away private data. | |
1858 | */ | |
1859 | void unmap_mapping_range(struct address_space *mapping, | |
1860 | loff_t const holebegin, loff_t const holelen, int even_cows) | |
1861 | { | |
1862 | struct zap_details details; | |
1863 | pgoff_t hba = holebegin >> PAGE_SHIFT; | |
1864 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1865 | ||
1866 | /* Check for overflow. */ | |
1867 | if (sizeof(holelen) > sizeof(hlen)) { | |
1868 | long long holeend = | |
1869 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1870 | if (holeend & ~(long long)ULONG_MAX) | |
1871 | hlen = ULONG_MAX - hba + 1; | |
1872 | } | |
1873 | ||
1874 | details.check_mapping = even_cows? NULL: mapping; | |
1875 | details.nonlinear_vma = NULL; | |
1876 | details.first_index = hba; | |
1877 | details.last_index = hba + hlen - 1; | |
1878 | if (details.last_index < details.first_index) | |
1879 | details.last_index = ULONG_MAX; | |
1880 | details.i_mmap_lock = &mapping->i_mmap_lock; | |
1881 | ||
1882 | spin_lock(&mapping->i_mmap_lock); | |
1883 | ||
d00806b1 | 1884 | /* Protect against endless unmapping loops */ |
1da177e4 | 1885 | mapping->truncate_count++; |
1da177e4 LT |
1886 | if (unlikely(is_restart_addr(mapping->truncate_count))) { |
1887 | if (mapping->truncate_count == 0) | |
1888 | reset_vma_truncate_counts(mapping); | |
1889 | mapping->truncate_count++; | |
1890 | } | |
1891 | details.truncate_count = mapping->truncate_count; | |
1892 | ||
1893 | if (unlikely(!prio_tree_empty(&mapping->i_mmap))) | |
1894 | unmap_mapping_range_tree(&mapping->i_mmap, &details); | |
1895 | if (unlikely(!list_empty(&mapping->i_mmap_nonlinear))) | |
1896 | unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details); | |
1897 | spin_unlock(&mapping->i_mmap_lock); | |
1898 | } | |
1899 | EXPORT_SYMBOL(unmap_mapping_range); | |
1900 | ||
bfa5bf6d REB |
1901 | /** |
1902 | * vmtruncate - unmap mappings "freed" by truncate() syscall | |
1903 | * @inode: inode of the file used | |
1904 | * @offset: file offset to start truncating | |
1da177e4 LT |
1905 | * |
1906 | * NOTE! We have to be ready to update the memory sharing | |
1907 | * between the file and the memory map for a potential last | |
1908 | * incomplete page. Ugly, but necessary. | |
1909 | */ | |
1910 | int vmtruncate(struct inode * inode, loff_t offset) | |
1911 | { | |
1912 | struct address_space *mapping = inode->i_mapping; | |
1913 | unsigned long limit; | |
1914 | ||
1915 | if (inode->i_size < offset) | |
1916 | goto do_expand; | |
1917 | /* | |
1918 | * truncation of in-use swapfiles is disallowed - it would cause | |
1919 | * subsequent swapout to scribble on the now-freed blocks. | |
1920 | */ | |
1921 | if (IS_SWAPFILE(inode)) | |
1922 | goto out_busy; | |
1923 | i_size_write(inode, offset); | |
d00806b1 NP |
1924 | |
1925 | /* | |
1926 | * unmap_mapping_range is called twice, first simply for efficiency | |
1927 | * so that truncate_inode_pages does fewer single-page unmaps. However | |
1928 | * after this first call, and before truncate_inode_pages finishes, | |
1929 | * it is possible for private pages to be COWed, which remain after | |
1930 | * truncate_inode_pages finishes, hence the second unmap_mapping_range | |
1931 | * call must be made for correctness. | |
1932 | */ | |
1da177e4 LT |
1933 | unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1); |
1934 | truncate_inode_pages(mapping, offset); | |
d00806b1 | 1935 | unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1); |
1da177e4 LT |
1936 | goto out_truncate; |
1937 | ||
1938 | do_expand: | |
1939 | limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; | |
1940 | if (limit != RLIM_INFINITY && offset > limit) | |
1941 | goto out_sig; | |
1942 | if (offset > inode->i_sb->s_maxbytes) | |
1943 | goto out_big; | |
1944 | i_size_write(inode, offset); | |
1945 | ||
1946 | out_truncate: | |
1947 | if (inode->i_op && inode->i_op->truncate) | |
1948 | inode->i_op->truncate(inode); | |
1949 | return 0; | |
1950 | out_sig: | |
1951 | send_sig(SIGXFSZ, current, 0); | |
1952 | out_big: | |
1953 | return -EFBIG; | |
1954 | out_busy: | |
1955 | return -ETXTBSY; | |
1956 | } | |
1da177e4 LT |
1957 | EXPORT_SYMBOL(vmtruncate); |
1958 | ||
f6b3ec23 BP |
1959 | int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end) |
1960 | { | |
1961 | struct address_space *mapping = inode->i_mapping; | |
1962 | ||
1963 | /* | |
1964 | * If the underlying filesystem is not going to provide | |
1965 | * a way to truncate a range of blocks (punch a hole) - | |
1966 | * we should return failure right now. | |
1967 | */ | |
1968 | if (!inode->i_op || !inode->i_op->truncate_range) | |
1969 | return -ENOSYS; | |
1970 | ||
1b1dcc1b | 1971 | mutex_lock(&inode->i_mutex); |
f6b3ec23 BP |
1972 | down_write(&inode->i_alloc_sem); |
1973 | unmap_mapping_range(mapping, offset, (end - offset), 1); | |
1974 | truncate_inode_pages_range(mapping, offset, end); | |
d00806b1 | 1975 | unmap_mapping_range(mapping, offset, (end - offset), 1); |
f6b3ec23 BP |
1976 | inode->i_op->truncate_range(inode, offset, end); |
1977 | up_write(&inode->i_alloc_sem); | |
1b1dcc1b | 1978 | mutex_unlock(&inode->i_mutex); |
f6b3ec23 BP |
1979 | |
1980 | return 0; | |
1981 | } | |
f6b3ec23 | 1982 | |
bfa5bf6d REB |
1983 | /** |
1984 | * swapin_readahead - swap in pages in hope we need them soon | |
1985 | * @entry: swap entry of this memory | |
1986 | * @addr: address to start | |
1987 | * @vma: user vma this addresses belong to | |
1988 | * | |
1da177e4 LT |
1989 | * Primitive swap readahead code. We simply read an aligned block of |
1990 | * (1 << page_cluster) entries in the swap area. This method is chosen | |
1991 | * because it doesn't cost us any seek time. We also make sure to queue | |
bfa5bf6d | 1992 | * the 'original' request together with the readahead ones... |
1da177e4 LT |
1993 | * |
1994 | * This has been extended to use the NUMA policies from the mm triggering | |
1995 | * the readahead. | |
1996 | * | |
1997 | * Caller must hold down_read on the vma->vm_mm if vma is not NULL. | |
1998 | */ | |
1999 | void swapin_readahead(swp_entry_t entry, unsigned long addr,struct vm_area_struct *vma) | |
2000 | { | |
2001 | #ifdef CONFIG_NUMA | |
2002 | struct vm_area_struct *next_vma = vma ? vma->vm_next : NULL; | |
2003 | #endif | |
2004 | int i, num; | |
2005 | struct page *new_page; | |
2006 | unsigned long offset; | |
2007 | ||
2008 | /* | |
2009 | * Get the number of handles we should do readahead io to. | |
2010 | */ | |
2011 | num = valid_swaphandles(entry, &offset); | |
2012 | for (i = 0; i < num; offset++, i++) { | |
2013 | /* Ok, do the async read-ahead now */ | |
2014 | new_page = read_swap_cache_async(swp_entry(swp_type(entry), | |
2015 | offset), vma, addr); | |
2016 | if (!new_page) | |
2017 | break; | |
2018 | page_cache_release(new_page); | |
2019 | #ifdef CONFIG_NUMA | |
2020 | /* | |
2021 | * Find the next applicable VMA for the NUMA policy. | |
2022 | */ | |
2023 | addr += PAGE_SIZE; | |
2024 | if (addr == 0) | |
2025 | vma = NULL; | |
2026 | if (vma) { | |
2027 | if (addr >= vma->vm_end) { | |
2028 | vma = next_vma; | |
2029 | next_vma = vma ? vma->vm_next : NULL; | |
2030 | } | |
2031 | if (vma && addr < vma->vm_start) | |
2032 | vma = NULL; | |
2033 | } else { | |
2034 | if (next_vma && addr >= next_vma->vm_start) { | |
2035 | vma = next_vma; | |
2036 | next_vma = vma->vm_next; | |
2037 | } | |
2038 | } | |
2039 | #endif | |
2040 | } | |
2041 | lru_add_drain(); /* Push any new pages onto the LRU now */ | |
2042 | } | |
2043 | ||
2044 | /* | |
8f4e2101 HD |
2045 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2046 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2047 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2048 | */ |
65500d23 HD |
2049 | static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2050 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
2051 | int write_access, pte_t orig_pte) | |
1da177e4 | 2052 | { |
8f4e2101 | 2053 | spinlock_t *ptl; |
1da177e4 | 2054 | struct page *page; |
65500d23 | 2055 | swp_entry_t entry; |
1da177e4 | 2056 | pte_t pte; |
83c54070 | 2057 | int ret = 0; |
1da177e4 | 2058 | |
4c21e2f2 | 2059 | if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) |
8f4e2101 | 2060 | goto out; |
65500d23 HD |
2061 | |
2062 | entry = pte_to_swp_entry(orig_pte); | |
0697212a CL |
2063 | if (is_migration_entry(entry)) { |
2064 | migration_entry_wait(mm, pmd, address); | |
2065 | goto out; | |
2066 | } | |
0ff92245 | 2067 | delayacct_set_flag(DELAYACCT_PF_SWAPIN); |
1da177e4 LT |
2068 | page = lookup_swap_cache(entry); |
2069 | if (!page) { | |
098fe651 | 2070 | grab_swap_token(); /* Contend for token _before_ read-in */ |
1da177e4 LT |
2071 | swapin_readahead(entry, address, vma); |
2072 | page = read_swap_cache_async(entry, vma, address); | |
2073 | if (!page) { | |
2074 | /* | |
8f4e2101 HD |
2075 | * Back out if somebody else faulted in this pte |
2076 | * while we released the pte lock. | |
1da177e4 | 2077 | */ |
8f4e2101 | 2078 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
1da177e4 LT |
2079 | if (likely(pte_same(*page_table, orig_pte))) |
2080 | ret = VM_FAULT_OOM; | |
0ff92245 | 2081 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
65500d23 | 2082 | goto unlock; |
1da177e4 LT |
2083 | } |
2084 | ||
2085 | /* Had to read the page from swap area: Major fault */ | |
2086 | ret = VM_FAULT_MAJOR; | |
f8891e5e | 2087 | count_vm_event(PGMAJFAULT); |
1da177e4 LT |
2088 | } |
2089 | ||
0ff92245 | 2090 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
1da177e4 LT |
2091 | mark_page_accessed(page); |
2092 | lock_page(page); | |
2093 | ||
2094 | /* | |
8f4e2101 | 2095 | * Back out if somebody else already faulted in this pte. |
1da177e4 | 2096 | */ |
8f4e2101 | 2097 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
9e9bef07 | 2098 | if (unlikely(!pte_same(*page_table, orig_pte))) |
b8107480 | 2099 | goto out_nomap; |
b8107480 KK |
2100 | |
2101 | if (unlikely(!PageUptodate(page))) { | |
2102 | ret = VM_FAULT_SIGBUS; | |
2103 | goto out_nomap; | |
1da177e4 LT |
2104 | } |
2105 | ||
2106 | /* The page isn't present yet, go ahead with the fault. */ | |
1da177e4 | 2107 | |
4294621f | 2108 | inc_mm_counter(mm, anon_rss); |
1da177e4 LT |
2109 | pte = mk_pte(page, vma->vm_page_prot); |
2110 | if (write_access && can_share_swap_page(page)) { | |
2111 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); | |
2112 | write_access = 0; | |
2113 | } | |
1da177e4 LT |
2114 | |
2115 | flush_icache_page(vma, page); | |
2116 | set_pte_at(mm, address, page_table, pte); | |
2117 | page_add_anon_rmap(page, vma, address); | |
2118 | ||
c475a8ab HD |
2119 | swap_free(entry); |
2120 | if (vm_swap_full()) | |
2121 | remove_exclusive_swap_page(page); | |
2122 | unlock_page(page); | |
2123 | ||
1da177e4 | 2124 | if (write_access) { |
83c54070 | 2125 | /* XXX: We could OR the do_wp_page code with this one? */ |
1da177e4 | 2126 | if (do_wp_page(mm, vma, address, |
83c54070 | 2127 | page_table, pmd, ptl, pte) & VM_FAULT_OOM) |
1da177e4 LT |
2128 | ret = VM_FAULT_OOM; |
2129 | goto out; | |
2130 | } | |
2131 | ||
2132 | /* No need to invalidate - it was non-present before */ | |
2133 | update_mmu_cache(vma, address, pte); | |
65500d23 | 2134 | unlock: |
8f4e2101 | 2135 | pte_unmap_unlock(page_table, ptl); |
1da177e4 LT |
2136 | out: |
2137 | return ret; | |
b8107480 | 2138 | out_nomap: |
8f4e2101 | 2139 | pte_unmap_unlock(page_table, ptl); |
b8107480 KK |
2140 | unlock_page(page); |
2141 | page_cache_release(page); | |
65500d23 | 2142 | return ret; |
1da177e4 LT |
2143 | } |
2144 | ||
2145 | /* | |
8f4e2101 HD |
2146 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2147 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2148 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2149 | */ |
65500d23 HD |
2150 | static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2151 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
2152 | int write_access) | |
1da177e4 | 2153 | { |
8f4e2101 HD |
2154 | struct page *page; |
2155 | spinlock_t *ptl; | |
1da177e4 | 2156 | pte_t entry; |
1da177e4 | 2157 | |
557ed1fa NP |
2158 | /* Allocate our own private page. */ |
2159 | pte_unmap(page_table); | |
8f4e2101 | 2160 | |
557ed1fa NP |
2161 | if (unlikely(anon_vma_prepare(vma))) |
2162 | goto oom; | |
2163 | page = alloc_zeroed_user_highpage_movable(vma, address); | |
2164 | if (!page) | |
2165 | goto oom; | |
8f4e2101 | 2166 | |
557ed1fa NP |
2167 | entry = mk_pte(page, vma->vm_page_prot); |
2168 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
1da177e4 | 2169 | |
557ed1fa NP |
2170 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
2171 | if (!pte_none(*page_table)) | |
2172 | goto release; | |
2173 | inc_mm_counter(mm, anon_rss); | |
2174 | lru_cache_add_active(page); | |
2175 | page_add_new_anon_rmap(page, vma, address); | |
65500d23 | 2176 | set_pte_at(mm, address, page_table, entry); |
1da177e4 LT |
2177 | |
2178 | /* No need to invalidate - it was non-present before */ | |
65500d23 | 2179 | update_mmu_cache(vma, address, entry); |
65500d23 | 2180 | unlock: |
8f4e2101 | 2181 | pte_unmap_unlock(page_table, ptl); |
83c54070 | 2182 | return 0; |
8f4e2101 HD |
2183 | release: |
2184 | page_cache_release(page); | |
2185 | goto unlock; | |
65500d23 | 2186 | oom: |
1da177e4 LT |
2187 | return VM_FAULT_OOM; |
2188 | } | |
2189 | ||
2190 | /* | |
54cb8821 | 2191 | * __do_fault() tries to create a new page mapping. It aggressively |
1da177e4 | 2192 | * tries to share with existing pages, but makes a separate copy if |
54cb8821 NP |
2193 | * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid |
2194 | * the next page fault. | |
1da177e4 LT |
2195 | * |
2196 | * As this is called only for pages that do not currently exist, we | |
2197 | * do not need to flush old virtual caches or the TLB. | |
2198 | * | |
8f4e2101 | 2199 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
16abfa08 | 2200 | * but allow concurrent faults), and pte neither mapped nor locked. |
8f4e2101 | 2201 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
1da177e4 | 2202 | */ |
54cb8821 | 2203 | static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
16abfa08 | 2204 | unsigned long address, pmd_t *pmd, |
54cb8821 | 2205 | pgoff_t pgoff, unsigned int flags, pte_t orig_pte) |
1da177e4 | 2206 | { |
16abfa08 | 2207 | pte_t *page_table; |
8f4e2101 | 2208 | spinlock_t *ptl; |
d0217ac0 | 2209 | struct page *page; |
1da177e4 | 2210 | pte_t entry; |
1da177e4 | 2211 | int anon = 0; |
d08b3851 | 2212 | struct page *dirty_page = NULL; |
d0217ac0 NP |
2213 | struct vm_fault vmf; |
2214 | int ret; | |
a200ee18 | 2215 | int page_mkwrite = 0; |
54cb8821 | 2216 | |
d0217ac0 NP |
2217 | vmf.virtual_address = (void __user *)(address & PAGE_MASK); |
2218 | vmf.pgoff = pgoff; | |
2219 | vmf.flags = flags; | |
2220 | vmf.page = NULL; | |
1da177e4 | 2221 | |
325f04db HD |
2222 | BUG_ON(vma->vm_flags & VM_PFNMAP); |
2223 | ||
54cb8821 | 2224 | if (likely(vma->vm_ops->fault)) { |
d0217ac0 | 2225 | ret = vma->vm_ops->fault(vma, &vmf); |
83c54070 NP |
2226 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) |
2227 | return ret; | |
54cb8821 NP |
2228 | } else { |
2229 | /* Legacy ->nopage path */ | |
83c54070 | 2230 | ret = 0; |
d0217ac0 | 2231 | vmf.page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret); |
54cb8821 | 2232 | /* no page was available -- either SIGBUS or OOM */ |
d0217ac0 | 2233 | if (unlikely(vmf.page == NOPAGE_SIGBUS)) |
54cb8821 | 2234 | return VM_FAULT_SIGBUS; |
d0217ac0 | 2235 | else if (unlikely(vmf.page == NOPAGE_OOM)) |
54cb8821 NP |
2236 | return VM_FAULT_OOM; |
2237 | } | |
1da177e4 | 2238 | |
d00806b1 | 2239 | /* |
d0217ac0 | 2240 | * For consistency in subsequent calls, make the faulted page always |
d00806b1 NP |
2241 | * locked. |
2242 | */ | |
83c54070 | 2243 | if (unlikely(!(ret & VM_FAULT_LOCKED))) |
d0217ac0 | 2244 | lock_page(vmf.page); |
54cb8821 | 2245 | else |
d0217ac0 | 2246 | VM_BUG_ON(!PageLocked(vmf.page)); |
d00806b1 | 2247 | |
1da177e4 LT |
2248 | /* |
2249 | * Should we do an early C-O-W break? | |
2250 | */ | |
d0217ac0 | 2251 | page = vmf.page; |
54cb8821 | 2252 | if (flags & FAULT_FLAG_WRITE) { |
9637a5ef | 2253 | if (!(vma->vm_flags & VM_SHARED)) { |
54cb8821 | 2254 | anon = 1; |
d00806b1 | 2255 | if (unlikely(anon_vma_prepare(vma))) { |
d0217ac0 | 2256 | ret = VM_FAULT_OOM; |
54cb8821 | 2257 | goto out; |
d00806b1 | 2258 | } |
83c54070 NP |
2259 | page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, |
2260 | vma, address); | |
d00806b1 | 2261 | if (!page) { |
d0217ac0 | 2262 | ret = VM_FAULT_OOM; |
54cb8821 | 2263 | goto out; |
d00806b1 | 2264 | } |
d0217ac0 | 2265 | copy_user_highpage(page, vmf.page, address, vma); |
9637a5ef | 2266 | } else { |
54cb8821 NP |
2267 | /* |
2268 | * If the page will be shareable, see if the backing | |
9637a5ef | 2269 | * address space wants to know that the page is about |
54cb8821 NP |
2270 | * to become writable |
2271 | */ | |
69676147 MF |
2272 | if (vma->vm_ops->page_mkwrite) { |
2273 | unlock_page(page); | |
2274 | if (vma->vm_ops->page_mkwrite(vma, page) < 0) { | |
d0217ac0 NP |
2275 | ret = VM_FAULT_SIGBUS; |
2276 | anon = 1; /* no anon but release vmf.page */ | |
69676147 MF |
2277 | goto out_unlocked; |
2278 | } | |
2279 | lock_page(page); | |
d0217ac0 NP |
2280 | /* |
2281 | * XXX: this is not quite right (racy vs | |
2282 | * invalidate) to unlock and relock the page | |
2283 | * like this, however a better fix requires | |
2284 | * reworking page_mkwrite locking API, which | |
2285 | * is better done later. | |
2286 | */ | |
2287 | if (!page->mapping) { | |
83c54070 | 2288 | ret = 0; |
d0217ac0 NP |
2289 | anon = 1; /* no anon but release vmf.page */ |
2290 | goto out; | |
2291 | } | |
a200ee18 | 2292 | page_mkwrite = 1; |
9637a5ef DH |
2293 | } |
2294 | } | |
54cb8821 | 2295 | |
1da177e4 LT |
2296 | } |
2297 | ||
8f4e2101 | 2298 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
1da177e4 LT |
2299 | |
2300 | /* | |
2301 | * This silly early PAGE_DIRTY setting removes a race | |
2302 | * due to the bad i386 page protection. But it's valid | |
2303 | * for other architectures too. | |
2304 | * | |
2305 | * Note that if write_access is true, we either now have | |
2306 | * an exclusive copy of the page, or this is a shared mapping, | |
2307 | * so we can make it writable and dirty to avoid having to | |
2308 | * handle that later. | |
2309 | */ | |
2310 | /* Only go through if we didn't race with anybody else... */ | |
54cb8821 | 2311 | if (likely(pte_same(*page_table, orig_pte))) { |
d00806b1 NP |
2312 | flush_icache_page(vma, page); |
2313 | entry = mk_pte(page, vma->vm_page_prot); | |
54cb8821 | 2314 | if (flags & FAULT_FLAG_WRITE) |
1da177e4 LT |
2315 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
2316 | set_pte_at(mm, address, page_table, entry); | |
2317 | if (anon) { | |
d00806b1 NP |
2318 | inc_mm_counter(mm, anon_rss); |
2319 | lru_cache_add_active(page); | |
2320 | page_add_new_anon_rmap(page, vma, address); | |
f57e88a8 | 2321 | } else { |
4294621f | 2322 | inc_mm_counter(mm, file_rss); |
d00806b1 | 2323 | page_add_file_rmap(page); |
54cb8821 | 2324 | if (flags & FAULT_FLAG_WRITE) { |
d00806b1 | 2325 | dirty_page = page; |
d08b3851 PZ |
2326 | get_page(dirty_page); |
2327 | } | |
4294621f | 2328 | } |
d00806b1 NP |
2329 | |
2330 | /* no need to invalidate: a not-present page won't be cached */ | |
2331 | update_mmu_cache(vma, address, entry); | |
1da177e4 | 2332 | } else { |
d00806b1 NP |
2333 | if (anon) |
2334 | page_cache_release(page); | |
2335 | else | |
54cb8821 | 2336 | anon = 1; /* no anon but release faulted_page */ |
1da177e4 LT |
2337 | } |
2338 | ||
8f4e2101 | 2339 | pte_unmap_unlock(page_table, ptl); |
d00806b1 NP |
2340 | |
2341 | out: | |
d0217ac0 | 2342 | unlock_page(vmf.page); |
69676147 | 2343 | out_unlocked: |
d00806b1 | 2344 | if (anon) |
d0217ac0 | 2345 | page_cache_release(vmf.page); |
d00806b1 | 2346 | else if (dirty_page) { |
a200ee18 | 2347 | set_page_dirty_balance(dirty_page, page_mkwrite); |
d08b3851 PZ |
2348 | put_page(dirty_page); |
2349 | } | |
d00806b1 | 2350 | |
83c54070 | 2351 | return ret; |
54cb8821 | 2352 | } |
d00806b1 | 2353 | |
54cb8821 NP |
2354 | static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
2355 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
2356 | int write_access, pte_t orig_pte) | |
2357 | { | |
2358 | pgoff_t pgoff = (((address & PAGE_MASK) | |
0da7e01f | 2359 | - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
54cb8821 NP |
2360 | unsigned int flags = (write_access ? FAULT_FLAG_WRITE : 0); |
2361 | ||
16abfa08 HD |
2362 | pte_unmap(page_table); |
2363 | return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); | |
54cb8821 NP |
2364 | } |
2365 | ||
1da177e4 | 2366 | |
f4b81804 JS |
2367 | /* |
2368 | * do_no_pfn() tries to create a new page mapping for a page without | |
2369 | * a struct_page backing it | |
2370 | * | |
2371 | * As this is called only for pages that do not currently exist, we | |
2372 | * do not need to flush old virtual caches or the TLB. | |
2373 | * | |
2374 | * We enter with non-exclusive mmap_sem (to exclude vma changes, | |
2375 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2376 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
2377 | * | |
2378 | * It is expected that the ->nopfn handler always returns the same pfn | |
2379 | * for a given virtual mapping. | |
2380 | * | |
2381 | * Mark this `noinline' to prevent it from bloating the main pagefault code. | |
2382 | */ | |
2383 | static noinline int do_no_pfn(struct mm_struct *mm, struct vm_area_struct *vma, | |
2384 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
2385 | int write_access) | |
2386 | { | |
2387 | spinlock_t *ptl; | |
2388 | pte_t entry; | |
2389 | unsigned long pfn; | |
f4b81804 JS |
2390 | |
2391 | pte_unmap(page_table); | |
2392 | BUG_ON(!(vma->vm_flags & VM_PFNMAP)); | |
2393 | BUG_ON(is_cow_mapping(vma->vm_flags)); | |
2394 | ||
2395 | pfn = vma->vm_ops->nopfn(vma, address & PAGE_MASK); | |
22cd25ed | 2396 | if (unlikely(pfn == NOPFN_OOM)) |
f4b81804 | 2397 | return VM_FAULT_OOM; |
22cd25ed | 2398 | else if (unlikely(pfn == NOPFN_SIGBUS)) |
f4b81804 | 2399 | return VM_FAULT_SIGBUS; |
22cd25ed | 2400 | else if (unlikely(pfn == NOPFN_REFAULT)) |
83c54070 | 2401 | return 0; |
f4b81804 JS |
2402 | |
2403 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); | |
2404 | ||
2405 | /* Only go through if we didn't race with anybody else... */ | |
2406 | if (pte_none(*page_table)) { | |
2407 | entry = pfn_pte(pfn, vma->vm_page_prot); | |
2408 | if (write_access) | |
2409 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
2410 | set_pte_at(mm, address, page_table, entry); | |
2411 | } | |
2412 | pte_unmap_unlock(page_table, ptl); | |
83c54070 | 2413 | return 0; |
f4b81804 JS |
2414 | } |
2415 | ||
1da177e4 LT |
2416 | /* |
2417 | * Fault of a previously existing named mapping. Repopulate the pte | |
2418 | * from the encoded file_pte if possible. This enables swappable | |
2419 | * nonlinear vmas. | |
8f4e2101 HD |
2420 | * |
2421 | * We enter with non-exclusive mmap_sem (to exclude vma changes, | |
2422 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2423 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2424 | */ |
d0217ac0 | 2425 | static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
65500d23 HD |
2426 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
2427 | int write_access, pte_t orig_pte) | |
1da177e4 | 2428 | { |
d0217ac0 NP |
2429 | unsigned int flags = FAULT_FLAG_NONLINEAR | |
2430 | (write_access ? FAULT_FLAG_WRITE : 0); | |
65500d23 | 2431 | pgoff_t pgoff; |
1da177e4 | 2432 | |
4c21e2f2 | 2433 | if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) |
83c54070 | 2434 | return 0; |
1da177e4 | 2435 | |
d0217ac0 NP |
2436 | if (unlikely(!(vma->vm_flags & VM_NONLINEAR) || |
2437 | !(vma->vm_flags & VM_CAN_NONLINEAR))) { | |
65500d23 HD |
2438 | /* |
2439 | * Page table corrupted: show pte and kill process. | |
2440 | */ | |
b5810039 | 2441 | print_bad_pte(vma, orig_pte, address); |
65500d23 HD |
2442 | return VM_FAULT_OOM; |
2443 | } | |
65500d23 HD |
2444 | |
2445 | pgoff = pte_to_pgoff(orig_pte); | |
16abfa08 | 2446 | return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); |
1da177e4 LT |
2447 | } |
2448 | ||
2449 | /* | |
2450 | * These routines also need to handle stuff like marking pages dirty | |
2451 | * and/or accessed for architectures that don't do it in hardware (most | |
2452 | * RISC architectures). The early dirtying is also good on the i386. | |
2453 | * | |
2454 | * There is also a hook called "update_mmu_cache()" that architectures | |
2455 | * with external mmu caches can use to update those (ie the Sparc or | |
2456 | * PowerPC hashed page tables that act as extended TLBs). | |
2457 | * | |
c74df32c HD |
2458 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2459 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2460 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 LT |
2461 | */ |
2462 | static inline int handle_pte_fault(struct mm_struct *mm, | |
65500d23 HD |
2463 | struct vm_area_struct *vma, unsigned long address, |
2464 | pte_t *pte, pmd_t *pmd, int write_access) | |
1da177e4 LT |
2465 | { |
2466 | pte_t entry; | |
8f4e2101 | 2467 | spinlock_t *ptl; |
1da177e4 | 2468 | |
8dab5241 | 2469 | entry = *pte; |
1da177e4 | 2470 | if (!pte_present(entry)) { |
65500d23 | 2471 | if (pte_none(entry)) { |
f4b81804 | 2472 | if (vma->vm_ops) { |
54cb8821 NP |
2473 | if (vma->vm_ops->fault || vma->vm_ops->nopage) |
2474 | return do_linear_fault(mm, vma, address, | |
2475 | pte, pmd, write_access, entry); | |
f4b81804 JS |
2476 | if (unlikely(vma->vm_ops->nopfn)) |
2477 | return do_no_pfn(mm, vma, address, pte, | |
2478 | pmd, write_access); | |
2479 | } | |
2480 | return do_anonymous_page(mm, vma, address, | |
2481 | pte, pmd, write_access); | |
65500d23 | 2482 | } |
1da177e4 | 2483 | if (pte_file(entry)) |
d0217ac0 | 2484 | return do_nonlinear_fault(mm, vma, address, |
65500d23 HD |
2485 | pte, pmd, write_access, entry); |
2486 | return do_swap_page(mm, vma, address, | |
2487 | pte, pmd, write_access, entry); | |
1da177e4 LT |
2488 | } |
2489 | ||
4c21e2f2 | 2490 | ptl = pte_lockptr(mm, pmd); |
8f4e2101 HD |
2491 | spin_lock(ptl); |
2492 | if (unlikely(!pte_same(*pte, entry))) | |
2493 | goto unlock; | |
1da177e4 LT |
2494 | if (write_access) { |
2495 | if (!pte_write(entry)) | |
8f4e2101 HD |
2496 | return do_wp_page(mm, vma, address, |
2497 | pte, pmd, ptl, entry); | |
1da177e4 LT |
2498 | entry = pte_mkdirty(entry); |
2499 | } | |
2500 | entry = pte_mkyoung(entry); | |
8dab5241 | 2501 | if (ptep_set_access_flags(vma, address, pte, entry, write_access)) { |
1a44e149 | 2502 | update_mmu_cache(vma, address, entry); |
1a44e149 AA |
2503 | } else { |
2504 | /* | |
2505 | * This is needed only for protection faults but the arch code | |
2506 | * is not yet telling us if this is a protection fault or not. | |
2507 | * This still avoids useless tlb flushes for .text page faults | |
2508 | * with threads. | |
2509 | */ | |
2510 | if (write_access) | |
2511 | flush_tlb_page(vma, address); | |
2512 | } | |
8f4e2101 HD |
2513 | unlock: |
2514 | pte_unmap_unlock(pte, ptl); | |
83c54070 | 2515 | return 0; |
1da177e4 LT |
2516 | } |
2517 | ||
2518 | /* | |
2519 | * By the time we get here, we already hold the mm semaphore | |
2520 | */ | |
83c54070 | 2521 | int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
1da177e4 LT |
2522 | unsigned long address, int write_access) |
2523 | { | |
2524 | pgd_t *pgd; | |
2525 | pud_t *pud; | |
2526 | pmd_t *pmd; | |
2527 | pte_t *pte; | |
2528 | ||
2529 | __set_current_state(TASK_RUNNING); | |
2530 | ||
f8891e5e | 2531 | count_vm_event(PGFAULT); |
1da177e4 | 2532 | |
ac9b9c66 HD |
2533 | if (unlikely(is_vm_hugetlb_page(vma))) |
2534 | return hugetlb_fault(mm, vma, address, write_access); | |
1da177e4 | 2535 | |
1da177e4 | 2536 | pgd = pgd_offset(mm, address); |
1da177e4 LT |
2537 | pud = pud_alloc(mm, pgd, address); |
2538 | if (!pud) | |
c74df32c | 2539 | return VM_FAULT_OOM; |
1da177e4 LT |
2540 | pmd = pmd_alloc(mm, pud, address); |
2541 | if (!pmd) | |
c74df32c | 2542 | return VM_FAULT_OOM; |
1da177e4 LT |
2543 | pte = pte_alloc_map(mm, pmd, address); |
2544 | if (!pte) | |
c74df32c | 2545 | return VM_FAULT_OOM; |
1da177e4 | 2546 | |
c74df32c | 2547 | return handle_pte_fault(mm, vma, address, pte, pmd, write_access); |
1da177e4 LT |
2548 | } |
2549 | ||
2550 | #ifndef __PAGETABLE_PUD_FOLDED | |
2551 | /* | |
2552 | * Allocate page upper directory. | |
872fec16 | 2553 | * We've already handled the fast-path in-line. |
1da177e4 | 2554 | */ |
1bb3630e | 2555 | int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) |
1da177e4 | 2556 | { |
c74df32c HD |
2557 | pud_t *new = pud_alloc_one(mm, address); |
2558 | if (!new) | |
1bb3630e | 2559 | return -ENOMEM; |
1da177e4 | 2560 | |
872fec16 | 2561 | spin_lock(&mm->page_table_lock); |
1bb3630e | 2562 | if (pgd_present(*pgd)) /* Another has populated it */ |
1da177e4 | 2563 | pud_free(new); |
1bb3630e HD |
2564 | else |
2565 | pgd_populate(mm, pgd, new); | |
c74df32c | 2566 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 2567 | return 0; |
1da177e4 LT |
2568 | } |
2569 | #endif /* __PAGETABLE_PUD_FOLDED */ | |
2570 | ||
2571 | #ifndef __PAGETABLE_PMD_FOLDED | |
2572 | /* | |
2573 | * Allocate page middle directory. | |
872fec16 | 2574 | * We've already handled the fast-path in-line. |
1da177e4 | 2575 | */ |
1bb3630e | 2576 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
1da177e4 | 2577 | { |
c74df32c HD |
2578 | pmd_t *new = pmd_alloc_one(mm, address); |
2579 | if (!new) | |
1bb3630e | 2580 | return -ENOMEM; |
1da177e4 | 2581 | |
872fec16 | 2582 | spin_lock(&mm->page_table_lock); |
1da177e4 | 2583 | #ifndef __ARCH_HAS_4LEVEL_HACK |
1bb3630e | 2584 | if (pud_present(*pud)) /* Another has populated it */ |
1da177e4 | 2585 | pmd_free(new); |
1bb3630e HD |
2586 | else |
2587 | pud_populate(mm, pud, new); | |
1da177e4 | 2588 | #else |
1bb3630e | 2589 | if (pgd_present(*pud)) /* Another has populated it */ |
1da177e4 | 2590 | pmd_free(new); |
1bb3630e HD |
2591 | else |
2592 | pgd_populate(mm, pud, new); | |
1da177e4 | 2593 | #endif /* __ARCH_HAS_4LEVEL_HACK */ |
c74df32c | 2594 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 2595 | return 0; |
e0f39591 | 2596 | } |
1da177e4 LT |
2597 | #endif /* __PAGETABLE_PMD_FOLDED */ |
2598 | ||
2599 | int make_pages_present(unsigned long addr, unsigned long end) | |
2600 | { | |
2601 | int ret, len, write; | |
2602 | struct vm_area_struct * vma; | |
2603 | ||
2604 | vma = find_vma(current->mm, addr); | |
2605 | if (!vma) | |
2606 | return -1; | |
2607 | write = (vma->vm_flags & VM_WRITE) != 0; | |
5bcb28b1 ES |
2608 | BUG_ON(addr >= end); |
2609 | BUG_ON(end > vma->vm_end); | |
68e116a3 | 2610 | len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE; |
1da177e4 LT |
2611 | ret = get_user_pages(current, current->mm, addr, |
2612 | len, write, 0, NULL, NULL); | |
2613 | if (ret < 0) | |
2614 | return ret; | |
2615 | return ret == len ? 0 : -1; | |
2616 | } | |
2617 | ||
2618 | /* | |
2619 | * Map a vmalloc()-space virtual address to the physical page. | |
2620 | */ | |
2621 | struct page * vmalloc_to_page(void * vmalloc_addr) | |
2622 | { | |
2623 | unsigned long addr = (unsigned long) vmalloc_addr; | |
2624 | struct page *page = NULL; | |
2625 | pgd_t *pgd = pgd_offset_k(addr); | |
2626 | pud_t *pud; | |
2627 | pmd_t *pmd; | |
2628 | pte_t *ptep, pte; | |
2629 | ||
2630 | if (!pgd_none(*pgd)) { | |
2631 | pud = pud_offset(pgd, addr); | |
2632 | if (!pud_none(*pud)) { | |
2633 | pmd = pmd_offset(pud, addr); | |
2634 | if (!pmd_none(*pmd)) { | |
2635 | ptep = pte_offset_map(pmd, addr); | |
2636 | pte = *ptep; | |
2637 | if (pte_present(pte)) | |
2638 | page = pte_page(pte); | |
2639 | pte_unmap(ptep); | |
2640 | } | |
2641 | } | |
2642 | } | |
2643 | return page; | |
2644 | } | |
2645 | ||
2646 | EXPORT_SYMBOL(vmalloc_to_page); | |
2647 | ||
2648 | /* | |
2649 | * Map a vmalloc()-space virtual address to the physical page frame number. | |
2650 | */ | |
2651 | unsigned long vmalloc_to_pfn(void * vmalloc_addr) | |
2652 | { | |
2653 | return page_to_pfn(vmalloc_to_page(vmalloc_addr)); | |
2654 | } | |
2655 | ||
2656 | EXPORT_SYMBOL(vmalloc_to_pfn); | |
2657 | ||
1da177e4 LT |
2658 | #if !defined(__HAVE_ARCH_GATE_AREA) |
2659 | ||
2660 | #if defined(AT_SYSINFO_EHDR) | |
5ce7852c | 2661 | static struct vm_area_struct gate_vma; |
1da177e4 LT |
2662 | |
2663 | static int __init gate_vma_init(void) | |
2664 | { | |
2665 | gate_vma.vm_mm = NULL; | |
2666 | gate_vma.vm_start = FIXADDR_USER_START; | |
2667 | gate_vma.vm_end = FIXADDR_USER_END; | |
b6558c4a RM |
2668 | gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC; |
2669 | gate_vma.vm_page_prot = __P101; | |
f47aef55 RM |
2670 | /* |
2671 | * Make sure the vDSO gets into every core dump. | |
2672 | * Dumping its contents makes post-mortem fully interpretable later | |
2673 | * without matching up the same kernel and hardware config to see | |
2674 | * what PC values meant. | |
2675 | */ | |
2676 | gate_vma.vm_flags |= VM_ALWAYSDUMP; | |
1da177e4 LT |
2677 | return 0; |
2678 | } | |
2679 | __initcall(gate_vma_init); | |
2680 | #endif | |
2681 | ||
2682 | struct vm_area_struct *get_gate_vma(struct task_struct *tsk) | |
2683 | { | |
2684 | #ifdef AT_SYSINFO_EHDR | |
2685 | return &gate_vma; | |
2686 | #else | |
2687 | return NULL; | |
2688 | #endif | |
2689 | } | |
2690 | ||
2691 | int in_gate_area_no_task(unsigned long addr) | |
2692 | { | |
2693 | #ifdef AT_SYSINFO_EHDR | |
2694 | if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END)) | |
2695 | return 1; | |
2696 | #endif | |
2697 | return 0; | |
2698 | } | |
2699 | ||
2700 | #endif /* __HAVE_ARCH_GATE_AREA */ | |
0ec76a11 DH |
2701 | |
2702 | /* | |
2703 | * Access another process' address space. | |
2704 | * Source/target buffer must be kernel space, | |
2705 | * Do not walk the page table directly, use get_user_pages | |
2706 | */ | |
2707 | int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) | |
2708 | { | |
2709 | struct mm_struct *mm; | |
2710 | struct vm_area_struct *vma; | |
2711 | struct page *page; | |
2712 | void *old_buf = buf; | |
2713 | ||
2714 | mm = get_task_mm(tsk); | |
2715 | if (!mm) | |
2716 | return 0; | |
2717 | ||
2718 | down_read(&mm->mmap_sem); | |
2719 | /* ignore errors, just check how much was sucessfully transfered */ | |
2720 | while (len) { | |
2721 | int bytes, ret, offset; | |
2722 | void *maddr; | |
2723 | ||
2724 | ret = get_user_pages(tsk, mm, addr, 1, | |
2725 | write, 1, &page, &vma); | |
2726 | if (ret <= 0) | |
2727 | break; | |
2728 | ||
2729 | bytes = len; | |
2730 | offset = addr & (PAGE_SIZE-1); | |
2731 | if (bytes > PAGE_SIZE-offset) | |
2732 | bytes = PAGE_SIZE-offset; | |
2733 | ||
2734 | maddr = kmap(page); | |
2735 | if (write) { | |
2736 | copy_to_user_page(vma, page, addr, | |
2737 | maddr + offset, buf, bytes); | |
2738 | set_page_dirty_lock(page); | |
2739 | } else { | |
2740 | copy_from_user_page(vma, page, addr, | |
2741 | buf, maddr + offset, bytes); | |
2742 | } | |
2743 | kunmap(page); | |
2744 | page_cache_release(page); | |
2745 | len -= bytes; | |
2746 | buf += bytes; | |
2747 | addr += bytes; | |
2748 | } | |
2749 | up_read(&mm->mmap_sem); | |
2750 | mmput(mm); | |
2751 | ||
2752 | return buf - old_buf; | |
2753 | } | |
5992b6da | 2754 | EXPORT_SYMBOL_GPL(access_process_vm); |