]>
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> |
8a9f3ccd | 53 | #include <linux/memcontrol.h> |
1da177e4 LT |
54 | |
55 | #include <asm/pgalloc.h> | |
56 | #include <asm/uaccess.h> | |
57 | #include <asm/tlb.h> | |
58 | #include <asm/tlbflush.h> | |
59 | #include <asm/pgtable.h> | |
60 | ||
61 | #include <linux/swapops.h> | |
62 | #include <linux/elf.h> | |
63 | ||
d41dee36 | 64 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
65 | /* use the per-pgdat data instead for discontigmem - mbligh */ |
66 | unsigned long max_mapnr; | |
67 | struct page *mem_map; | |
68 | ||
69 | EXPORT_SYMBOL(max_mapnr); | |
70 | EXPORT_SYMBOL(mem_map); | |
71 | #endif | |
72 | ||
73 | unsigned long num_physpages; | |
74 | /* | |
75 | * A number of key systems in x86 including ioremap() rely on the assumption | |
76 | * that high_memory defines the upper bound on direct map memory, then end | |
77 | * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and | |
78 | * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL | |
79 | * and ZONE_HIGHMEM. | |
80 | */ | |
81 | void * high_memory; | |
1da177e4 LT |
82 | |
83 | EXPORT_SYMBOL(num_physpages); | |
84 | EXPORT_SYMBOL(high_memory); | |
1da177e4 | 85 | |
32a93233 IM |
86 | /* |
87 | * Randomize the address space (stacks, mmaps, brk, etc.). | |
88 | * | |
89 | * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, | |
90 | * as ancient (libc5 based) binaries can segfault. ) | |
91 | */ | |
92 | int randomize_va_space __read_mostly = | |
93 | #ifdef CONFIG_COMPAT_BRK | |
94 | 1; | |
95 | #else | |
96 | 2; | |
97 | #endif | |
a62eaf15 AK |
98 | |
99 | static int __init disable_randmaps(char *s) | |
100 | { | |
101 | randomize_va_space = 0; | |
9b41046c | 102 | return 1; |
a62eaf15 AK |
103 | } |
104 | __setup("norandmaps", disable_randmaps); | |
105 | ||
106 | ||
1da177e4 LT |
107 | /* |
108 | * If a p?d_bad entry is found while walking page tables, report | |
109 | * the error, before resetting entry to p?d_none. Usually (but | |
110 | * very seldom) called out from the p?d_none_or_clear_bad macros. | |
111 | */ | |
112 | ||
113 | void pgd_clear_bad(pgd_t *pgd) | |
114 | { | |
115 | pgd_ERROR(*pgd); | |
116 | pgd_clear(pgd); | |
117 | } | |
118 | ||
119 | void pud_clear_bad(pud_t *pud) | |
120 | { | |
121 | pud_ERROR(*pud); | |
122 | pud_clear(pud); | |
123 | } | |
124 | ||
125 | void pmd_clear_bad(pmd_t *pmd) | |
126 | { | |
127 | pmd_ERROR(*pmd); | |
128 | pmd_clear(pmd); | |
129 | } | |
130 | ||
131 | /* | |
132 | * Note: this doesn't free the actual pages themselves. That | |
133 | * has been handled earlier when unmapping all the memory regions. | |
134 | */ | |
e0da382c | 135 | static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd) |
1da177e4 | 136 | { |
2f569afd | 137 | pgtable_t token = pmd_pgtable(*pmd); |
e0da382c | 138 | pmd_clear(pmd); |
2f569afd | 139 | pte_free_tlb(tlb, token); |
e0da382c | 140 | tlb->mm->nr_ptes--; |
1da177e4 LT |
141 | } |
142 | ||
e0da382c HD |
143 | static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
144 | unsigned long addr, unsigned long end, | |
145 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
146 | { |
147 | pmd_t *pmd; | |
148 | unsigned long next; | |
e0da382c | 149 | unsigned long start; |
1da177e4 | 150 | |
e0da382c | 151 | start = addr; |
1da177e4 | 152 | pmd = pmd_offset(pud, addr); |
1da177e4 LT |
153 | do { |
154 | next = pmd_addr_end(addr, end); | |
155 | if (pmd_none_or_clear_bad(pmd)) | |
156 | continue; | |
e0da382c | 157 | free_pte_range(tlb, pmd); |
1da177e4 LT |
158 | } while (pmd++, addr = next, addr != end); |
159 | ||
e0da382c HD |
160 | start &= PUD_MASK; |
161 | if (start < floor) | |
162 | return; | |
163 | if (ceiling) { | |
164 | ceiling &= PUD_MASK; | |
165 | if (!ceiling) | |
166 | return; | |
1da177e4 | 167 | } |
e0da382c HD |
168 | if (end - 1 > ceiling - 1) |
169 | return; | |
170 | ||
171 | pmd = pmd_offset(pud, start); | |
172 | pud_clear(pud); | |
173 | pmd_free_tlb(tlb, pmd); | |
1da177e4 LT |
174 | } |
175 | ||
e0da382c HD |
176 | static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, |
177 | unsigned long addr, unsigned long end, | |
178 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
179 | { |
180 | pud_t *pud; | |
181 | unsigned long next; | |
e0da382c | 182 | unsigned long start; |
1da177e4 | 183 | |
e0da382c | 184 | start = addr; |
1da177e4 | 185 | pud = pud_offset(pgd, addr); |
1da177e4 LT |
186 | do { |
187 | next = pud_addr_end(addr, end); | |
188 | if (pud_none_or_clear_bad(pud)) | |
189 | continue; | |
e0da382c | 190 | free_pmd_range(tlb, pud, addr, next, floor, ceiling); |
1da177e4 LT |
191 | } while (pud++, addr = next, addr != end); |
192 | ||
e0da382c HD |
193 | start &= PGDIR_MASK; |
194 | if (start < floor) | |
195 | return; | |
196 | if (ceiling) { | |
197 | ceiling &= PGDIR_MASK; | |
198 | if (!ceiling) | |
199 | return; | |
1da177e4 | 200 | } |
e0da382c HD |
201 | if (end - 1 > ceiling - 1) |
202 | return; | |
203 | ||
204 | pud = pud_offset(pgd, start); | |
205 | pgd_clear(pgd); | |
206 | pud_free_tlb(tlb, pud); | |
1da177e4 LT |
207 | } |
208 | ||
209 | /* | |
e0da382c HD |
210 | * This function frees user-level page tables of a process. |
211 | * | |
1da177e4 LT |
212 | * Must be called with pagetable lock held. |
213 | */ | |
3bf5ee95 | 214 | void free_pgd_range(struct mmu_gather **tlb, |
e0da382c HD |
215 | unsigned long addr, unsigned long end, |
216 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
217 | { |
218 | pgd_t *pgd; | |
219 | unsigned long next; | |
e0da382c HD |
220 | unsigned long start; |
221 | ||
222 | /* | |
223 | * The next few lines have given us lots of grief... | |
224 | * | |
225 | * Why are we testing PMD* at this top level? Because often | |
226 | * there will be no work to do at all, and we'd prefer not to | |
227 | * go all the way down to the bottom just to discover that. | |
228 | * | |
229 | * Why all these "- 1"s? Because 0 represents both the bottom | |
230 | * of the address space and the top of it (using -1 for the | |
231 | * top wouldn't help much: the masks would do the wrong thing). | |
232 | * The rule is that addr 0 and floor 0 refer to the bottom of | |
233 | * the address space, but end 0 and ceiling 0 refer to the top | |
234 | * Comparisons need to use "end - 1" and "ceiling - 1" (though | |
235 | * that end 0 case should be mythical). | |
236 | * | |
237 | * Wherever addr is brought up or ceiling brought down, we must | |
238 | * be careful to reject "the opposite 0" before it confuses the | |
239 | * subsequent tests. But what about where end is brought down | |
240 | * by PMD_SIZE below? no, end can't go down to 0 there. | |
241 | * | |
242 | * Whereas we round start (addr) and ceiling down, by different | |
243 | * masks at different levels, in order to test whether a table | |
244 | * now has no other vmas using it, so can be freed, we don't | |
245 | * bother to round floor or end up - the tests don't need that. | |
246 | */ | |
1da177e4 | 247 | |
e0da382c HD |
248 | addr &= PMD_MASK; |
249 | if (addr < floor) { | |
250 | addr += PMD_SIZE; | |
251 | if (!addr) | |
252 | return; | |
253 | } | |
254 | if (ceiling) { | |
255 | ceiling &= PMD_MASK; | |
256 | if (!ceiling) | |
257 | return; | |
258 | } | |
259 | if (end - 1 > ceiling - 1) | |
260 | end -= PMD_SIZE; | |
261 | if (addr > end - 1) | |
262 | return; | |
263 | ||
264 | start = addr; | |
3bf5ee95 | 265 | pgd = pgd_offset((*tlb)->mm, addr); |
1da177e4 LT |
266 | do { |
267 | next = pgd_addr_end(addr, end); | |
268 | if (pgd_none_or_clear_bad(pgd)) | |
269 | continue; | |
3bf5ee95 | 270 | free_pud_range(*tlb, pgd, addr, next, floor, ceiling); |
1da177e4 | 271 | } while (pgd++, addr = next, addr != end); |
e0da382c HD |
272 | } |
273 | ||
274 | void free_pgtables(struct mmu_gather **tlb, struct vm_area_struct *vma, | |
3bf5ee95 | 275 | unsigned long floor, unsigned long ceiling) |
e0da382c HD |
276 | { |
277 | while (vma) { | |
278 | struct vm_area_struct *next = vma->vm_next; | |
279 | unsigned long addr = vma->vm_start; | |
280 | ||
8f4f8c16 HD |
281 | /* |
282 | * Hide vma from rmap and vmtruncate before freeing pgtables | |
283 | */ | |
284 | anon_vma_unlink(vma); | |
285 | unlink_file_vma(vma); | |
286 | ||
9da61aef | 287 | if (is_vm_hugetlb_page(vma)) { |
3bf5ee95 | 288 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
e0da382c | 289 | floor, next? next->vm_start: ceiling); |
3bf5ee95 HD |
290 | } else { |
291 | /* | |
292 | * Optimization: gather nearby vmas into one call down | |
293 | */ | |
294 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE | |
4866920b | 295 | && !is_vm_hugetlb_page(next)) { |
3bf5ee95 HD |
296 | vma = next; |
297 | next = vma->vm_next; | |
8f4f8c16 HD |
298 | anon_vma_unlink(vma); |
299 | unlink_file_vma(vma); | |
3bf5ee95 HD |
300 | } |
301 | free_pgd_range(tlb, addr, vma->vm_end, | |
302 | floor, next? next->vm_start: ceiling); | |
303 | } | |
e0da382c HD |
304 | vma = next; |
305 | } | |
1da177e4 LT |
306 | } |
307 | ||
1bb3630e | 308 | int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address) |
1da177e4 | 309 | { |
2f569afd | 310 | pgtable_t new = pte_alloc_one(mm, address); |
1bb3630e HD |
311 | if (!new) |
312 | return -ENOMEM; | |
313 | ||
c74df32c | 314 | spin_lock(&mm->page_table_lock); |
2f569afd | 315 | if (!pmd_present(*pmd)) { /* Has another populated it ? */ |
1da177e4 | 316 | mm->nr_ptes++; |
1da177e4 | 317 | pmd_populate(mm, pmd, new); |
2f569afd | 318 | new = NULL; |
1da177e4 | 319 | } |
c74df32c | 320 | spin_unlock(&mm->page_table_lock); |
2f569afd MS |
321 | if (new) |
322 | pte_free(mm, new); | |
1bb3630e | 323 | return 0; |
1da177e4 LT |
324 | } |
325 | ||
1bb3630e | 326 | int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) |
1da177e4 | 327 | { |
1bb3630e HD |
328 | pte_t *new = pte_alloc_one_kernel(&init_mm, address); |
329 | if (!new) | |
330 | return -ENOMEM; | |
331 | ||
332 | spin_lock(&init_mm.page_table_lock); | |
2f569afd | 333 | if (!pmd_present(*pmd)) { /* Has another populated it ? */ |
1bb3630e | 334 | pmd_populate_kernel(&init_mm, pmd, new); |
2f569afd MS |
335 | new = NULL; |
336 | } | |
1bb3630e | 337 | spin_unlock(&init_mm.page_table_lock); |
2f569afd MS |
338 | if (new) |
339 | pte_free_kernel(&init_mm, new); | |
1bb3630e | 340 | return 0; |
1da177e4 LT |
341 | } |
342 | ||
ae859762 HD |
343 | static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss) |
344 | { | |
345 | if (file_rss) | |
346 | add_mm_counter(mm, file_rss, file_rss); | |
347 | if (anon_rss) | |
348 | add_mm_counter(mm, anon_rss, anon_rss); | |
349 | } | |
350 | ||
b5810039 | 351 | /* |
6aab341e LT |
352 | * This function is called to print an error when a bad pte |
353 | * is found. For example, we might have a PFN-mapped pte in | |
354 | * a region that doesn't allow it. | |
b5810039 NP |
355 | * |
356 | * The calling function must still handle the error. | |
357 | */ | |
358 | void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr) | |
359 | { | |
360 | printk(KERN_ERR "Bad pte = %08llx, process = %s, " | |
361 | "vm_flags = %lx, vaddr = %lx\n", | |
362 | (long long)pte_val(pte), | |
363 | (vma->vm_mm == current->mm ? current->comm : "???"), | |
364 | vma->vm_flags, vaddr); | |
365 | dump_stack(); | |
366 | } | |
367 | ||
67121172 LT |
368 | static inline int is_cow_mapping(unsigned int flags) |
369 | { | |
370 | return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; | |
371 | } | |
372 | ||
ee498ed7 | 373 | /* |
7e675137 | 374 | * vm_normal_page -- This function gets the "struct page" associated with a pte. |
6aab341e | 375 | * |
7e675137 NP |
376 | * "Special" mappings do not wish to be associated with a "struct page" (either |
377 | * it doesn't exist, or it exists but they don't want to touch it). In this | |
378 | * case, NULL is returned here. "Normal" mappings do have a struct page. | |
b379d790 | 379 | * |
7e675137 NP |
380 | * There are 2 broad cases. Firstly, an architecture may define a pte_special() |
381 | * pte bit, in which case this function is trivial. Secondly, an architecture | |
382 | * may not have a spare pte bit, which requires a more complicated scheme, | |
383 | * described below. | |
384 | * | |
385 | * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a | |
386 | * special mapping (even if there are underlying and valid "struct pages"). | |
387 | * COWed pages of a VM_PFNMAP are always normal. | |
6aab341e | 388 | * |
b379d790 JH |
389 | * The way we recognize COWed pages within VM_PFNMAP mappings is through the |
390 | * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit | |
7e675137 NP |
391 | * set, and the vm_pgoff will point to the first PFN mapped: thus every special |
392 | * mapping will always honor the rule | |
6aab341e LT |
393 | * |
394 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) | |
395 | * | |
7e675137 NP |
396 | * And for normal mappings this is false. |
397 | * | |
398 | * This restricts such mappings to be a linear translation from virtual address | |
399 | * to pfn. To get around this restriction, we allow arbitrary mappings so long | |
400 | * as the vma is not a COW mapping; in that case, we know that all ptes are | |
401 | * special (because none can have been COWed). | |
b379d790 | 402 | * |
b379d790 | 403 | * |
7e675137 | 404 | * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. |
b379d790 JH |
405 | * |
406 | * VM_MIXEDMAP mappings can likewise contain memory with or without "struct | |
407 | * page" backing, however the difference is that _all_ pages with a struct | |
408 | * page (that is, those where pfn_valid is true) are refcounted and considered | |
409 | * normal pages by the VM. The disadvantage is that pages are refcounted | |
410 | * (which can be slower and simply not an option for some PFNMAP users). The | |
411 | * advantage is that we don't have to follow the strict linearity rule of | |
412 | * PFNMAP mappings in order to support COWable mappings. | |
413 | * | |
ee498ed7 | 414 | */ |
7e675137 NP |
415 | #ifdef __HAVE_ARCH_PTE_SPECIAL |
416 | # define HAVE_PTE_SPECIAL 1 | |
417 | #else | |
418 | # define HAVE_PTE_SPECIAL 0 | |
419 | #endif | |
420 | struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, | |
421 | pte_t pte) | |
ee498ed7 | 422 | { |
7e675137 NP |
423 | unsigned long pfn; |
424 | ||
425 | if (HAVE_PTE_SPECIAL) { | |
426 | if (likely(!pte_special(pte))) { | |
427 | VM_BUG_ON(!pfn_valid(pte_pfn(pte))); | |
428 | return pte_page(pte); | |
429 | } | |
430 | VM_BUG_ON(!(vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))); | |
431 | return NULL; | |
432 | } | |
433 | ||
434 | /* !HAVE_PTE_SPECIAL case follows: */ | |
435 | ||
436 | pfn = pte_pfn(pte); | |
6aab341e | 437 | |
b379d790 JH |
438 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
439 | if (vma->vm_flags & VM_MIXEDMAP) { | |
440 | if (!pfn_valid(pfn)) | |
441 | return NULL; | |
442 | goto out; | |
443 | } else { | |
7e675137 NP |
444 | unsigned long off; |
445 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
b379d790 JH |
446 | if (pfn == vma->vm_pgoff + off) |
447 | return NULL; | |
448 | if (!is_cow_mapping(vma->vm_flags)) | |
449 | return NULL; | |
450 | } | |
6aab341e LT |
451 | } |
452 | ||
7e675137 | 453 | VM_BUG_ON(!pfn_valid(pfn)); |
6aab341e LT |
454 | |
455 | /* | |
7e675137 | 456 | * NOTE! We still have PageReserved() pages in the page tables. |
6aab341e | 457 | * |
7e675137 | 458 | * eg. VDSO mappings can cause them to exist. |
6aab341e | 459 | */ |
b379d790 | 460 | out: |
6aab341e | 461 | return pfn_to_page(pfn); |
ee498ed7 HD |
462 | } |
463 | ||
1da177e4 LT |
464 | /* |
465 | * copy one vm_area from one task to the other. Assumes the page tables | |
466 | * already present in the new task to be cleared in the whole range | |
467 | * covered by this vma. | |
1da177e4 LT |
468 | */ |
469 | ||
8c103762 | 470 | static inline void |
1da177e4 | 471 | copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
b5810039 | 472 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, |
8c103762 | 473 | unsigned long addr, int *rss) |
1da177e4 | 474 | { |
b5810039 | 475 | unsigned long vm_flags = vma->vm_flags; |
1da177e4 LT |
476 | pte_t pte = *src_pte; |
477 | struct page *page; | |
1da177e4 LT |
478 | |
479 | /* pte contains position in swap or file, so copy. */ | |
480 | if (unlikely(!pte_present(pte))) { | |
481 | if (!pte_file(pte)) { | |
0697212a CL |
482 | swp_entry_t entry = pte_to_swp_entry(pte); |
483 | ||
484 | swap_duplicate(entry); | |
1da177e4 LT |
485 | /* make sure dst_mm is on swapoff's mmlist. */ |
486 | if (unlikely(list_empty(&dst_mm->mmlist))) { | |
487 | spin_lock(&mmlist_lock); | |
f412ac08 HD |
488 | if (list_empty(&dst_mm->mmlist)) |
489 | list_add(&dst_mm->mmlist, | |
490 | &src_mm->mmlist); | |
1da177e4 LT |
491 | spin_unlock(&mmlist_lock); |
492 | } | |
0697212a CL |
493 | if (is_write_migration_entry(entry) && |
494 | is_cow_mapping(vm_flags)) { | |
495 | /* | |
496 | * COW mappings require pages in both parent | |
497 | * and child to be set to read. | |
498 | */ | |
499 | make_migration_entry_read(&entry); | |
500 | pte = swp_entry_to_pte(entry); | |
501 | set_pte_at(src_mm, addr, src_pte, pte); | |
502 | } | |
1da177e4 | 503 | } |
ae859762 | 504 | goto out_set_pte; |
1da177e4 LT |
505 | } |
506 | ||
1da177e4 LT |
507 | /* |
508 | * If it's a COW mapping, write protect it both | |
509 | * in the parent and the child | |
510 | */ | |
67121172 | 511 | if (is_cow_mapping(vm_flags)) { |
1da177e4 | 512 | ptep_set_wrprotect(src_mm, addr, src_pte); |
3dc90795 | 513 | pte = pte_wrprotect(pte); |
1da177e4 LT |
514 | } |
515 | ||
516 | /* | |
517 | * If it's a shared mapping, mark it clean in | |
518 | * the child | |
519 | */ | |
520 | if (vm_flags & VM_SHARED) | |
521 | pte = pte_mkclean(pte); | |
522 | pte = pte_mkold(pte); | |
6aab341e LT |
523 | |
524 | page = vm_normal_page(vma, addr, pte); | |
525 | if (page) { | |
526 | get_page(page); | |
c97a9e10 | 527 | page_dup_rmap(page, vma, addr); |
6aab341e LT |
528 | rss[!!PageAnon(page)]++; |
529 | } | |
ae859762 HD |
530 | |
531 | out_set_pte: | |
532 | set_pte_at(dst_mm, addr, dst_pte, pte); | |
1da177e4 LT |
533 | } |
534 | ||
535 | static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
536 | pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, | |
537 | unsigned long addr, unsigned long end) | |
538 | { | |
539 | pte_t *src_pte, *dst_pte; | |
c74df32c | 540 | spinlock_t *src_ptl, *dst_ptl; |
e040f218 | 541 | int progress = 0; |
8c103762 | 542 | int rss[2]; |
1da177e4 LT |
543 | |
544 | again: | |
ae859762 | 545 | rss[1] = rss[0] = 0; |
c74df32c | 546 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
1da177e4 LT |
547 | if (!dst_pte) |
548 | return -ENOMEM; | |
549 | src_pte = pte_offset_map_nested(src_pmd, addr); | |
4c21e2f2 | 550 | src_ptl = pte_lockptr(src_mm, src_pmd); |
f20dc5f7 | 551 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
6606c3e0 | 552 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 553 | |
1da177e4 LT |
554 | do { |
555 | /* | |
556 | * We are holding two locks at this point - either of them | |
557 | * could generate latencies in another task on another CPU. | |
558 | */ | |
e040f218 HD |
559 | if (progress >= 32) { |
560 | progress = 0; | |
561 | if (need_resched() || | |
95c354fe | 562 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
e040f218 HD |
563 | break; |
564 | } | |
1da177e4 LT |
565 | if (pte_none(*src_pte)) { |
566 | progress++; | |
567 | continue; | |
568 | } | |
8c103762 | 569 | copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss); |
1da177e4 LT |
570 | progress += 8; |
571 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); | |
1da177e4 | 572 | |
6606c3e0 | 573 | arch_leave_lazy_mmu_mode(); |
c74df32c | 574 | spin_unlock(src_ptl); |
1da177e4 | 575 | pte_unmap_nested(src_pte - 1); |
ae859762 | 576 | add_mm_rss(dst_mm, rss[0], rss[1]); |
c74df32c HD |
577 | pte_unmap_unlock(dst_pte - 1, dst_ptl); |
578 | cond_resched(); | |
1da177e4 LT |
579 | if (addr != end) |
580 | goto again; | |
581 | return 0; | |
582 | } | |
583 | ||
584 | static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
585 | pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, | |
586 | unsigned long addr, unsigned long end) | |
587 | { | |
588 | pmd_t *src_pmd, *dst_pmd; | |
589 | unsigned long next; | |
590 | ||
591 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); | |
592 | if (!dst_pmd) | |
593 | return -ENOMEM; | |
594 | src_pmd = pmd_offset(src_pud, addr); | |
595 | do { | |
596 | next = pmd_addr_end(addr, end); | |
597 | if (pmd_none_or_clear_bad(src_pmd)) | |
598 | continue; | |
599 | if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, | |
600 | vma, addr, next)) | |
601 | return -ENOMEM; | |
602 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); | |
603 | return 0; | |
604 | } | |
605 | ||
606 | static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
607 | pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, | |
608 | unsigned long addr, unsigned long end) | |
609 | { | |
610 | pud_t *src_pud, *dst_pud; | |
611 | unsigned long next; | |
612 | ||
613 | dst_pud = pud_alloc(dst_mm, dst_pgd, addr); | |
614 | if (!dst_pud) | |
615 | return -ENOMEM; | |
616 | src_pud = pud_offset(src_pgd, addr); | |
617 | do { | |
618 | next = pud_addr_end(addr, end); | |
619 | if (pud_none_or_clear_bad(src_pud)) | |
620 | continue; | |
621 | if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, | |
622 | vma, addr, next)) | |
623 | return -ENOMEM; | |
624 | } while (dst_pud++, src_pud++, addr = next, addr != end); | |
625 | return 0; | |
626 | } | |
627 | ||
628 | int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
629 | struct vm_area_struct *vma) | |
630 | { | |
631 | pgd_t *src_pgd, *dst_pgd; | |
632 | unsigned long next; | |
633 | unsigned long addr = vma->vm_start; | |
634 | unsigned long end = vma->vm_end; | |
635 | ||
d992895b NP |
636 | /* |
637 | * Don't copy ptes where a page fault will fill them correctly. | |
638 | * Fork becomes much lighter when there are big shared or private | |
639 | * readonly mappings. The tradeoff is that copy_page_range is more | |
640 | * efficient than faulting. | |
641 | */ | |
4d7672b4 | 642 | if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) { |
d992895b NP |
643 | if (!vma->anon_vma) |
644 | return 0; | |
645 | } | |
646 | ||
1da177e4 LT |
647 | if (is_vm_hugetlb_page(vma)) |
648 | return copy_hugetlb_page_range(dst_mm, src_mm, vma); | |
649 | ||
650 | dst_pgd = pgd_offset(dst_mm, addr); | |
651 | src_pgd = pgd_offset(src_mm, addr); | |
652 | do { | |
653 | next = pgd_addr_end(addr, end); | |
654 | if (pgd_none_or_clear_bad(src_pgd)) | |
655 | continue; | |
656 | if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, | |
657 | vma, addr, next)) | |
658 | return -ENOMEM; | |
659 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); | |
660 | return 0; | |
661 | } | |
662 | ||
51c6f666 | 663 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
b5810039 | 664 | struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 | 665 | unsigned long addr, unsigned long end, |
51c6f666 | 666 | long *zap_work, struct zap_details *details) |
1da177e4 | 667 | { |
b5810039 | 668 | struct mm_struct *mm = tlb->mm; |
1da177e4 | 669 | pte_t *pte; |
508034a3 | 670 | spinlock_t *ptl; |
ae859762 HD |
671 | int file_rss = 0; |
672 | int anon_rss = 0; | |
1da177e4 | 673 | |
508034a3 | 674 | pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
6606c3e0 | 675 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
676 | do { |
677 | pte_t ptent = *pte; | |
51c6f666 RH |
678 | if (pte_none(ptent)) { |
679 | (*zap_work)--; | |
1da177e4 | 680 | continue; |
51c6f666 | 681 | } |
6f5e6b9e HD |
682 | |
683 | (*zap_work) -= PAGE_SIZE; | |
684 | ||
1da177e4 | 685 | if (pte_present(ptent)) { |
ee498ed7 | 686 | struct page *page; |
51c6f666 | 687 | |
6aab341e | 688 | page = vm_normal_page(vma, addr, ptent); |
1da177e4 LT |
689 | if (unlikely(details) && page) { |
690 | /* | |
691 | * unmap_shared_mapping_pages() wants to | |
692 | * invalidate cache without truncating: | |
693 | * unmap shared but keep private pages. | |
694 | */ | |
695 | if (details->check_mapping && | |
696 | details->check_mapping != page->mapping) | |
697 | continue; | |
698 | /* | |
699 | * Each page->index must be checked when | |
700 | * invalidating or truncating nonlinear. | |
701 | */ | |
702 | if (details->nonlinear_vma && | |
703 | (page->index < details->first_index || | |
704 | page->index > details->last_index)) | |
705 | continue; | |
706 | } | |
b5810039 | 707 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
a600388d | 708 | tlb->fullmm); |
1da177e4 LT |
709 | tlb_remove_tlb_entry(tlb, pte, addr); |
710 | if (unlikely(!page)) | |
711 | continue; | |
712 | if (unlikely(details) && details->nonlinear_vma | |
713 | && linear_page_index(details->nonlinear_vma, | |
714 | addr) != page->index) | |
b5810039 | 715 | set_pte_at(mm, addr, pte, |
1da177e4 | 716 | pgoff_to_pte(page->index)); |
1da177e4 | 717 | if (PageAnon(page)) |
86d912f4 | 718 | anon_rss--; |
6237bcd9 HD |
719 | else { |
720 | if (pte_dirty(ptent)) | |
721 | set_page_dirty(page); | |
722 | if (pte_young(ptent)) | |
daa88c8d | 723 | SetPageReferenced(page); |
86d912f4 | 724 | file_rss--; |
6237bcd9 | 725 | } |
7de6b805 | 726 | page_remove_rmap(page, vma); |
1da177e4 LT |
727 | tlb_remove_page(tlb, page); |
728 | continue; | |
729 | } | |
730 | /* | |
731 | * If details->check_mapping, we leave swap entries; | |
732 | * if details->nonlinear_vma, we leave file entries. | |
733 | */ | |
734 | if (unlikely(details)) | |
735 | continue; | |
736 | if (!pte_file(ptent)) | |
737 | free_swap_and_cache(pte_to_swp_entry(ptent)); | |
9888a1ca | 738 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
51c6f666 | 739 | } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0)); |
ae859762 | 740 | |
86d912f4 | 741 | add_mm_rss(mm, file_rss, anon_rss); |
6606c3e0 | 742 | arch_leave_lazy_mmu_mode(); |
508034a3 | 743 | pte_unmap_unlock(pte - 1, ptl); |
51c6f666 RH |
744 | |
745 | return addr; | |
1da177e4 LT |
746 | } |
747 | ||
51c6f666 | 748 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
b5810039 | 749 | struct vm_area_struct *vma, pud_t *pud, |
1da177e4 | 750 | unsigned long addr, unsigned long end, |
51c6f666 | 751 | long *zap_work, struct zap_details *details) |
1da177e4 LT |
752 | { |
753 | pmd_t *pmd; | |
754 | unsigned long next; | |
755 | ||
756 | pmd = pmd_offset(pud, addr); | |
757 | do { | |
758 | next = pmd_addr_end(addr, end); | |
51c6f666 RH |
759 | if (pmd_none_or_clear_bad(pmd)) { |
760 | (*zap_work)--; | |
1da177e4 | 761 | continue; |
51c6f666 RH |
762 | } |
763 | next = zap_pte_range(tlb, vma, pmd, addr, next, | |
764 | zap_work, details); | |
765 | } while (pmd++, addr = next, (addr != end && *zap_work > 0)); | |
766 | ||
767 | return addr; | |
1da177e4 LT |
768 | } |
769 | ||
51c6f666 | 770 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
b5810039 | 771 | struct vm_area_struct *vma, pgd_t *pgd, |
1da177e4 | 772 | unsigned long addr, unsigned long end, |
51c6f666 | 773 | long *zap_work, struct zap_details *details) |
1da177e4 LT |
774 | { |
775 | pud_t *pud; | |
776 | unsigned long next; | |
777 | ||
778 | pud = pud_offset(pgd, addr); | |
779 | do { | |
780 | next = pud_addr_end(addr, end); | |
51c6f666 RH |
781 | if (pud_none_or_clear_bad(pud)) { |
782 | (*zap_work)--; | |
1da177e4 | 783 | continue; |
51c6f666 RH |
784 | } |
785 | next = zap_pmd_range(tlb, vma, pud, addr, next, | |
786 | zap_work, details); | |
787 | } while (pud++, addr = next, (addr != end && *zap_work > 0)); | |
788 | ||
789 | return addr; | |
1da177e4 LT |
790 | } |
791 | ||
51c6f666 RH |
792 | static unsigned long unmap_page_range(struct mmu_gather *tlb, |
793 | struct vm_area_struct *vma, | |
1da177e4 | 794 | unsigned long addr, unsigned long end, |
51c6f666 | 795 | long *zap_work, struct zap_details *details) |
1da177e4 LT |
796 | { |
797 | pgd_t *pgd; | |
798 | unsigned long next; | |
799 | ||
800 | if (details && !details->check_mapping && !details->nonlinear_vma) | |
801 | details = NULL; | |
802 | ||
803 | BUG_ON(addr >= end); | |
804 | tlb_start_vma(tlb, vma); | |
805 | pgd = pgd_offset(vma->vm_mm, addr); | |
806 | do { | |
807 | next = pgd_addr_end(addr, end); | |
51c6f666 RH |
808 | if (pgd_none_or_clear_bad(pgd)) { |
809 | (*zap_work)--; | |
1da177e4 | 810 | continue; |
51c6f666 RH |
811 | } |
812 | next = zap_pud_range(tlb, vma, pgd, addr, next, | |
813 | zap_work, details); | |
814 | } while (pgd++, addr = next, (addr != end && *zap_work > 0)); | |
1da177e4 | 815 | tlb_end_vma(tlb, vma); |
51c6f666 RH |
816 | |
817 | return addr; | |
1da177e4 LT |
818 | } |
819 | ||
820 | #ifdef CONFIG_PREEMPT | |
821 | # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE) | |
822 | #else | |
823 | /* No preempt: go for improved straight-line efficiency */ | |
824 | # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE) | |
825 | #endif | |
826 | ||
827 | /** | |
828 | * unmap_vmas - unmap a range of memory covered by a list of vma's | |
829 | * @tlbp: address of the caller's struct mmu_gather | |
1da177e4 LT |
830 | * @vma: the starting vma |
831 | * @start_addr: virtual address at which to start unmapping | |
832 | * @end_addr: virtual address at which to end unmapping | |
833 | * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here | |
834 | * @details: details of nonlinear truncation or shared cache invalidation | |
835 | * | |
ee39b37b | 836 | * Returns the end address of the unmapping (restart addr if interrupted). |
1da177e4 | 837 | * |
508034a3 | 838 | * Unmap all pages in the vma list. |
1da177e4 | 839 | * |
508034a3 HD |
840 | * We aim to not hold locks for too long (for scheduling latency reasons). |
841 | * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to | |
1da177e4 LT |
842 | * return the ending mmu_gather to the caller. |
843 | * | |
844 | * Only addresses between `start' and `end' will be unmapped. | |
845 | * | |
846 | * The VMA list must be sorted in ascending virtual address order. | |
847 | * | |
848 | * unmap_vmas() assumes that the caller will flush the whole unmapped address | |
849 | * range after unmap_vmas() returns. So the only responsibility here is to | |
850 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() | |
851 | * drops the lock and schedules. | |
852 | */ | |
508034a3 | 853 | unsigned long unmap_vmas(struct mmu_gather **tlbp, |
1da177e4 LT |
854 | struct vm_area_struct *vma, unsigned long start_addr, |
855 | unsigned long end_addr, unsigned long *nr_accounted, | |
856 | struct zap_details *details) | |
857 | { | |
51c6f666 | 858 | long zap_work = ZAP_BLOCK_SIZE; |
1da177e4 LT |
859 | unsigned long tlb_start = 0; /* For tlb_finish_mmu */ |
860 | int tlb_start_valid = 0; | |
ee39b37b | 861 | unsigned long start = start_addr; |
1da177e4 | 862 | spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL; |
4d6ddfa9 | 863 | int fullmm = (*tlbp)->fullmm; |
1da177e4 LT |
864 | |
865 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) { | |
1da177e4 LT |
866 | unsigned long end; |
867 | ||
868 | start = max(vma->vm_start, start_addr); | |
869 | if (start >= vma->vm_end) | |
870 | continue; | |
871 | end = min(vma->vm_end, end_addr); | |
872 | if (end <= vma->vm_start) | |
873 | continue; | |
874 | ||
875 | if (vma->vm_flags & VM_ACCOUNT) | |
876 | *nr_accounted += (end - start) >> PAGE_SHIFT; | |
877 | ||
1da177e4 | 878 | while (start != end) { |
1da177e4 LT |
879 | if (!tlb_start_valid) { |
880 | tlb_start = start; | |
881 | tlb_start_valid = 1; | |
882 | } | |
883 | ||
51c6f666 | 884 | if (unlikely(is_vm_hugetlb_page(vma))) { |
1da177e4 | 885 | unmap_hugepage_range(vma, start, end); |
51c6f666 RH |
886 | zap_work -= (end - start) / |
887 | (HPAGE_SIZE / PAGE_SIZE); | |
888 | start = end; | |
889 | } else | |
890 | start = unmap_page_range(*tlbp, vma, | |
891 | start, end, &zap_work, details); | |
892 | ||
893 | if (zap_work > 0) { | |
894 | BUG_ON(start != end); | |
895 | break; | |
1da177e4 LT |
896 | } |
897 | ||
1da177e4 LT |
898 | tlb_finish_mmu(*tlbp, tlb_start, start); |
899 | ||
900 | if (need_resched() || | |
95c354fe | 901 | (i_mmap_lock && spin_needbreak(i_mmap_lock))) { |
1da177e4 | 902 | if (i_mmap_lock) { |
508034a3 | 903 | *tlbp = NULL; |
1da177e4 LT |
904 | goto out; |
905 | } | |
1da177e4 | 906 | cond_resched(); |
1da177e4 LT |
907 | } |
908 | ||
508034a3 | 909 | *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm); |
1da177e4 | 910 | tlb_start_valid = 0; |
51c6f666 | 911 | zap_work = ZAP_BLOCK_SIZE; |
1da177e4 LT |
912 | } |
913 | } | |
914 | out: | |
ee39b37b | 915 | return start; /* which is now the end (or restart) address */ |
1da177e4 LT |
916 | } |
917 | ||
918 | /** | |
919 | * zap_page_range - remove user pages in a given range | |
920 | * @vma: vm_area_struct holding the applicable pages | |
921 | * @address: starting address of pages to zap | |
922 | * @size: number of bytes to zap | |
923 | * @details: details of nonlinear truncation or shared cache invalidation | |
924 | */ | |
ee39b37b | 925 | unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address, |
1da177e4 LT |
926 | unsigned long size, struct zap_details *details) |
927 | { | |
928 | struct mm_struct *mm = vma->vm_mm; | |
929 | struct mmu_gather *tlb; | |
930 | unsigned long end = address + size; | |
931 | unsigned long nr_accounted = 0; | |
932 | ||
1da177e4 | 933 | lru_add_drain(); |
1da177e4 | 934 | tlb = tlb_gather_mmu(mm, 0); |
365e9c87 | 935 | update_hiwater_rss(mm); |
508034a3 HD |
936 | end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details); |
937 | if (tlb) | |
938 | tlb_finish_mmu(tlb, address, end); | |
ee39b37b | 939 | return end; |
1da177e4 LT |
940 | } |
941 | ||
942 | /* | |
943 | * Do a quick page-table lookup for a single page. | |
1da177e4 | 944 | */ |
6aab341e | 945 | struct page *follow_page(struct vm_area_struct *vma, unsigned long address, |
deceb6cd | 946 | unsigned int flags) |
1da177e4 LT |
947 | { |
948 | pgd_t *pgd; | |
949 | pud_t *pud; | |
950 | pmd_t *pmd; | |
951 | pte_t *ptep, pte; | |
deceb6cd | 952 | spinlock_t *ptl; |
1da177e4 | 953 | struct page *page; |
6aab341e | 954 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 955 | |
deceb6cd HD |
956 | page = follow_huge_addr(mm, address, flags & FOLL_WRITE); |
957 | if (!IS_ERR(page)) { | |
958 | BUG_ON(flags & FOLL_GET); | |
959 | goto out; | |
960 | } | |
1da177e4 | 961 | |
deceb6cd | 962 | page = NULL; |
1da177e4 LT |
963 | pgd = pgd_offset(mm, address); |
964 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
deceb6cd | 965 | goto no_page_table; |
1da177e4 LT |
966 | |
967 | pud = pud_offset(pgd, address); | |
968 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) | |
deceb6cd | 969 | goto no_page_table; |
1da177e4 LT |
970 | |
971 | pmd = pmd_offset(pud, address); | |
aeed5fce | 972 | if (pmd_none(*pmd)) |
deceb6cd HD |
973 | goto no_page_table; |
974 | ||
975 | if (pmd_huge(*pmd)) { | |
976 | BUG_ON(flags & FOLL_GET); | |
977 | page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE); | |
1da177e4 | 978 | goto out; |
deceb6cd | 979 | } |
1da177e4 | 980 | |
aeed5fce HD |
981 | if (unlikely(pmd_bad(*pmd))) |
982 | goto no_page_table; | |
983 | ||
deceb6cd | 984 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); |
1da177e4 LT |
985 | if (!ptep) |
986 | goto out; | |
987 | ||
988 | pte = *ptep; | |
deceb6cd HD |
989 | if (!pte_present(pte)) |
990 | goto unlock; | |
991 | if ((flags & FOLL_WRITE) && !pte_write(pte)) | |
992 | goto unlock; | |
6aab341e LT |
993 | page = vm_normal_page(vma, address, pte); |
994 | if (unlikely(!page)) | |
deceb6cd | 995 | goto unlock; |
1da177e4 | 996 | |
deceb6cd HD |
997 | if (flags & FOLL_GET) |
998 | get_page(page); | |
999 | if (flags & FOLL_TOUCH) { | |
1000 | if ((flags & FOLL_WRITE) && | |
1001 | !pte_dirty(pte) && !PageDirty(page)) | |
1002 | set_page_dirty(page); | |
1003 | mark_page_accessed(page); | |
1004 | } | |
1005 | unlock: | |
1006 | pte_unmap_unlock(ptep, ptl); | |
1da177e4 | 1007 | out: |
deceb6cd | 1008 | return page; |
1da177e4 | 1009 | |
deceb6cd HD |
1010 | no_page_table: |
1011 | /* | |
1012 | * When core dumping an enormous anonymous area that nobody | |
1013 | * has touched so far, we don't want to allocate page tables. | |
1014 | */ | |
1015 | if (flags & FOLL_ANON) { | |
557ed1fa | 1016 | page = ZERO_PAGE(0); |
deceb6cd HD |
1017 | if (flags & FOLL_GET) |
1018 | get_page(page); | |
1019 | BUG_ON(flags & FOLL_WRITE); | |
1020 | } | |
1021 | return page; | |
1da177e4 LT |
1022 | } |
1023 | ||
1da177e4 LT |
1024 | int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
1025 | unsigned long start, int len, int write, int force, | |
1026 | struct page **pages, struct vm_area_struct **vmas) | |
1027 | { | |
1028 | int i; | |
deceb6cd | 1029 | unsigned int vm_flags; |
1da177e4 | 1030 | |
900cf086 JC |
1031 | if (len <= 0) |
1032 | return 0; | |
1da177e4 LT |
1033 | /* |
1034 | * Require read or write permissions. | |
1035 | * If 'force' is set, we only require the "MAY" flags. | |
1036 | */ | |
deceb6cd HD |
1037 | vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); |
1038 | vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); | |
1da177e4 LT |
1039 | i = 0; |
1040 | ||
1041 | do { | |
deceb6cd HD |
1042 | struct vm_area_struct *vma; |
1043 | unsigned int foll_flags; | |
1da177e4 LT |
1044 | |
1045 | vma = find_extend_vma(mm, start); | |
1046 | if (!vma && in_gate_area(tsk, start)) { | |
1047 | unsigned long pg = start & PAGE_MASK; | |
1048 | struct vm_area_struct *gate_vma = get_gate_vma(tsk); | |
1049 | pgd_t *pgd; | |
1050 | pud_t *pud; | |
1051 | pmd_t *pmd; | |
1052 | pte_t *pte; | |
1053 | if (write) /* user gate pages are read-only */ | |
1054 | return i ? : -EFAULT; | |
1055 | if (pg > TASK_SIZE) | |
1056 | pgd = pgd_offset_k(pg); | |
1057 | else | |
1058 | pgd = pgd_offset_gate(mm, pg); | |
1059 | BUG_ON(pgd_none(*pgd)); | |
1060 | pud = pud_offset(pgd, pg); | |
1061 | BUG_ON(pud_none(*pud)); | |
1062 | pmd = pmd_offset(pud, pg); | |
690dbe1c HD |
1063 | if (pmd_none(*pmd)) |
1064 | return i ? : -EFAULT; | |
1da177e4 | 1065 | pte = pte_offset_map(pmd, pg); |
690dbe1c HD |
1066 | if (pte_none(*pte)) { |
1067 | pte_unmap(pte); | |
1068 | return i ? : -EFAULT; | |
1069 | } | |
1da177e4 | 1070 | if (pages) { |
fa2a455b | 1071 | struct page *page = vm_normal_page(gate_vma, start, *pte); |
6aab341e LT |
1072 | pages[i] = page; |
1073 | if (page) | |
1074 | get_page(page); | |
1da177e4 LT |
1075 | } |
1076 | pte_unmap(pte); | |
1077 | if (vmas) | |
1078 | vmas[i] = gate_vma; | |
1079 | i++; | |
1080 | start += PAGE_SIZE; | |
1081 | len--; | |
1082 | continue; | |
1083 | } | |
1084 | ||
1ff80389 | 1085 | if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP)) |
deceb6cd | 1086 | || !(vm_flags & vma->vm_flags)) |
1da177e4 LT |
1087 | return i ? : -EFAULT; |
1088 | ||
1089 | if (is_vm_hugetlb_page(vma)) { | |
1090 | i = follow_hugetlb_page(mm, vma, pages, vmas, | |
5b23dbe8 | 1091 | &start, &len, i, write); |
1da177e4 LT |
1092 | continue; |
1093 | } | |
deceb6cd HD |
1094 | |
1095 | foll_flags = FOLL_TOUCH; | |
1096 | if (pages) | |
1097 | foll_flags |= FOLL_GET; | |
1098 | if (!write && !(vma->vm_flags & VM_LOCKED) && | |
3c18ddd1 | 1099 | (!vma->vm_ops || !vma->vm_ops->fault)) |
deceb6cd HD |
1100 | foll_flags |= FOLL_ANON; |
1101 | ||
1da177e4 | 1102 | do { |
08ef4729 | 1103 | struct page *page; |
1da177e4 | 1104 | |
462e00cc ES |
1105 | /* |
1106 | * If tsk is ooming, cut off its access to large memory | |
1107 | * allocations. It has a pending SIGKILL, but it can't | |
1108 | * be processed until returning to user space. | |
1109 | */ | |
1110 | if (unlikely(test_tsk_thread_flag(tsk, TIF_MEMDIE))) | |
1111 | return -ENOMEM; | |
1112 | ||
deceb6cd HD |
1113 | if (write) |
1114 | foll_flags |= FOLL_WRITE; | |
a68d2ebc | 1115 | |
deceb6cd | 1116 | cond_resched(); |
6aab341e | 1117 | while (!(page = follow_page(vma, start, foll_flags))) { |
deceb6cd | 1118 | int ret; |
83c54070 | 1119 | ret = handle_mm_fault(mm, vma, start, |
deceb6cd | 1120 | foll_flags & FOLL_WRITE); |
83c54070 NP |
1121 | if (ret & VM_FAULT_ERROR) { |
1122 | if (ret & VM_FAULT_OOM) | |
1123 | return i ? i : -ENOMEM; | |
1124 | else if (ret & VM_FAULT_SIGBUS) | |
1125 | return i ? i : -EFAULT; | |
1126 | BUG(); | |
1127 | } | |
1128 | if (ret & VM_FAULT_MAJOR) | |
1129 | tsk->maj_flt++; | |
1130 | else | |
1131 | tsk->min_flt++; | |
1132 | ||
a68d2ebc | 1133 | /* |
83c54070 NP |
1134 | * The VM_FAULT_WRITE bit tells us that |
1135 | * do_wp_page has broken COW when necessary, | |
1136 | * even if maybe_mkwrite decided not to set | |
1137 | * pte_write. We can thus safely do subsequent | |
1138 | * page lookups as if they were reads. | |
a68d2ebc LT |
1139 | */ |
1140 | if (ret & VM_FAULT_WRITE) | |
deceb6cd | 1141 | foll_flags &= ~FOLL_WRITE; |
83c54070 | 1142 | |
7f7bbbe5 | 1143 | cond_resched(); |
1da177e4 LT |
1144 | } |
1145 | if (pages) { | |
08ef4729 | 1146 | pages[i] = page; |
03beb076 | 1147 | |
a6f36be3 | 1148 | flush_anon_page(vma, page, start); |
08ef4729 | 1149 | flush_dcache_page(page); |
1da177e4 LT |
1150 | } |
1151 | if (vmas) | |
1152 | vmas[i] = vma; | |
1153 | i++; | |
1154 | start += PAGE_SIZE; | |
1155 | len--; | |
08ef4729 | 1156 | } while (len && start < vma->vm_end); |
08ef4729 | 1157 | } while (len); |
1da177e4 LT |
1158 | return i; |
1159 | } | |
1da177e4 LT |
1160 | EXPORT_SYMBOL(get_user_pages); |
1161 | ||
920c7a5d HH |
1162 | pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr, |
1163 | spinlock_t **ptl) | |
c9cfcddf LT |
1164 | { |
1165 | pgd_t * pgd = pgd_offset(mm, addr); | |
1166 | pud_t * pud = pud_alloc(mm, pgd, addr); | |
1167 | if (pud) { | |
49c91fb0 | 1168 | pmd_t * pmd = pmd_alloc(mm, pud, addr); |
c9cfcddf LT |
1169 | if (pmd) |
1170 | return pte_alloc_map_lock(mm, pmd, addr, ptl); | |
1171 | } | |
1172 | return NULL; | |
1173 | } | |
1174 | ||
238f58d8 LT |
1175 | /* |
1176 | * This is the old fallback for page remapping. | |
1177 | * | |
1178 | * For historical reasons, it only allows reserved pages. Only | |
1179 | * old drivers should use this, and they needed to mark their | |
1180 | * pages reserved for the old functions anyway. | |
1181 | */ | |
423bad60 NP |
1182 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
1183 | struct page *page, pgprot_t prot) | |
238f58d8 | 1184 | { |
423bad60 | 1185 | struct mm_struct *mm = vma->vm_mm; |
238f58d8 | 1186 | int retval; |
c9cfcddf | 1187 | pte_t *pte; |
8a9f3ccd BS |
1188 | spinlock_t *ptl; |
1189 | ||
e1a1cd59 | 1190 | retval = mem_cgroup_charge(page, mm, GFP_KERNEL); |
8a9f3ccd BS |
1191 | if (retval) |
1192 | goto out; | |
238f58d8 LT |
1193 | |
1194 | retval = -EINVAL; | |
a145dd41 | 1195 | if (PageAnon(page)) |
8a9f3ccd | 1196 | goto out_uncharge; |
238f58d8 LT |
1197 | retval = -ENOMEM; |
1198 | flush_dcache_page(page); | |
c9cfcddf | 1199 | pte = get_locked_pte(mm, addr, &ptl); |
238f58d8 | 1200 | if (!pte) |
8a9f3ccd | 1201 | goto out_uncharge; |
238f58d8 LT |
1202 | retval = -EBUSY; |
1203 | if (!pte_none(*pte)) | |
1204 | goto out_unlock; | |
1205 | ||
1206 | /* Ok, finally just insert the thing.. */ | |
1207 | get_page(page); | |
1208 | inc_mm_counter(mm, file_rss); | |
1209 | page_add_file_rmap(page); | |
1210 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); | |
1211 | ||
1212 | retval = 0; | |
8a9f3ccd BS |
1213 | pte_unmap_unlock(pte, ptl); |
1214 | return retval; | |
238f58d8 LT |
1215 | out_unlock: |
1216 | pte_unmap_unlock(pte, ptl); | |
8a9f3ccd BS |
1217 | out_uncharge: |
1218 | mem_cgroup_uncharge_page(page); | |
238f58d8 LT |
1219 | out: |
1220 | return retval; | |
1221 | } | |
1222 | ||
bfa5bf6d REB |
1223 | /** |
1224 | * vm_insert_page - insert single page into user vma | |
1225 | * @vma: user vma to map to | |
1226 | * @addr: target user address of this page | |
1227 | * @page: source kernel page | |
1228 | * | |
a145dd41 LT |
1229 | * This allows drivers to insert individual pages they've allocated |
1230 | * into a user vma. | |
1231 | * | |
1232 | * The page has to be a nice clean _individual_ kernel allocation. | |
1233 | * If you allocate a compound page, you need to have marked it as | |
1234 | * such (__GFP_COMP), or manually just split the page up yourself | |
8dfcc9ba | 1235 | * (see split_page()). |
a145dd41 LT |
1236 | * |
1237 | * NOTE! Traditionally this was done with "remap_pfn_range()" which | |
1238 | * took an arbitrary page protection parameter. This doesn't allow | |
1239 | * that. Your vma protection will have to be set up correctly, which | |
1240 | * means that if you want a shared writable mapping, you'd better | |
1241 | * ask for a shared writable mapping! | |
1242 | * | |
1243 | * The page does not need to be reserved. | |
1244 | */ | |
423bad60 NP |
1245 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
1246 | struct page *page) | |
a145dd41 LT |
1247 | { |
1248 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
1249 | return -EFAULT; | |
1250 | if (!page_count(page)) | |
1251 | return -EINVAL; | |
4d7672b4 | 1252 | vma->vm_flags |= VM_INSERTPAGE; |
423bad60 | 1253 | return insert_page(vma, addr, page, vma->vm_page_prot); |
a145dd41 | 1254 | } |
e3c3374f | 1255 | EXPORT_SYMBOL(vm_insert_page); |
a145dd41 | 1256 | |
423bad60 NP |
1257 | static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
1258 | unsigned long pfn, pgprot_t prot) | |
1259 | { | |
1260 | struct mm_struct *mm = vma->vm_mm; | |
1261 | int retval; | |
1262 | pte_t *pte, entry; | |
1263 | spinlock_t *ptl; | |
1264 | ||
1265 | retval = -ENOMEM; | |
1266 | pte = get_locked_pte(mm, addr, &ptl); | |
1267 | if (!pte) | |
1268 | goto out; | |
1269 | retval = -EBUSY; | |
1270 | if (!pte_none(*pte)) | |
1271 | goto out_unlock; | |
1272 | ||
1273 | /* Ok, finally just insert the thing.. */ | |
1274 | entry = pte_mkspecial(pfn_pte(pfn, prot)); | |
1275 | set_pte_at(mm, addr, pte, entry); | |
1276 | update_mmu_cache(vma, addr, entry); /* XXX: why not for insert_page? */ | |
1277 | ||
1278 | retval = 0; | |
1279 | out_unlock: | |
1280 | pte_unmap_unlock(pte, ptl); | |
1281 | out: | |
1282 | return retval; | |
1283 | } | |
1284 | ||
e0dc0d8f NP |
1285 | /** |
1286 | * vm_insert_pfn - insert single pfn into user vma | |
1287 | * @vma: user vma to map to | |
1288 | * @addr: target user address of this page | |
1289 | * @pfn: source kernel pfn | |
1290 | * | |
1291 | * Similar to vm_inert_page, this allows drivers to insert individual pages | |
1292 | * they've allocated into a user vma. Same comments apply. | |
1293 | * | |
1294 | * This function should only be called from a vm_ops->fault handler, and | |
1295 | * in that case the handler should return NULL. | |
1296 | */ | |
1297 | int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, | |
423bad60 | 1298 | unsigned long pfn) |
e0dc0d8f | 1299 | { |
7e675137 NP |
1300 | /* |
1301 | * Technically, architectures with pte_special can avoid all these | |
1302 | * restrictions (same for remap_pfn_range). However we would like | |
1303 | * consistency in testing and feature parity among all, so we should | |
1304 | * try to keep these invariants in place for everybody. | |
1305 | */ | |
b379d790 JH |
1306 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); |
1307 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == | |
1308 | (VM_PFNMAP|VM_MIXEDMAP)); | |
1309 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); | |
1310 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); | |
e0dc0d8f | 1311 | |
423bad60 NP |
1312 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1313 | return -EFAULT; | |
1314 | return insert_pfn(vma, addr, pfn, vma->vm_page_prot); | |
1315 | } | |
1316 | EXPORT_SYMBOL(vm_insert_pfn); | |
e0dc0d8f | 1317 | |
423bad60 NP |
1318 | int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, |
1319 | unsigned long pfn) | |
1320 | { | |
1321 | BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); | |
e0dc0d8f | 1322 | |
423bad60 NP |
1323 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1324 | return -EFAULT; | |
e0dc0d8f | 1325 | |
423bad60 NP |
1326 | /* |
1327 | * If we don't have pte special, then we have to use the pfn_valid() | |
1328 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* | |
1329 | * refcount the page if pfn_valid is true (hence insert_page rather | |
1330 | * than insert_pfn). | |
1331 | */ | |
1332 | if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) { | |
1333 | struct page *page; | |
1334 | ||
1335 | page = pfn_to_page(pfn); | |
1336 | return insert_page(vma, addr, page, vma->vm_page_prot); | |
1337 | } | |
1338 | return insert_pfn(vma, addr, pfn, vma->vm_page_prot); | |
e0dc0d8f | 1339 | } |
423bad60 | 1340 | EXPORT_SYMBOL(vm_insert_mixed); |
e0dc0d8f | 1341 | |
1da177e4 LT |
1342 | /* |
1343 | * maps a range of physical memory into the requested pages. the old | |
1344 | * mappings are removed. any references to nonexistent pages results | |
1345 | * in null mappings (currently treated as "copy-on-access") | |
1346 | */ | |
1347 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
1348 | unsigned long addr, unsigned long end, | |
1349 | unsigned long pfn, pgprot_t prot) | |
1350 | { | |
1351 | pte_t *pte; | |
c74df32c | 1352 | spinlock_t *ptl; |
1da177e4 | 1353 | |
c74df32c | 1354 | pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1da177e4 LT |
1355 | if (!pte) |
1356 | return -ENOMEM; | |
6606c3e0 | 1357 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1358 | do { |
1359 | BUG_ON(!pte_none(*pte)); | |
7e675137 | 1360 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
1da177e4 LT |
1361 | pfn++; |
1362 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 1363 | arch_leave_lazy_mmu_mode(); |
c74df32c | 1364 | pte_unmap_unlock(pte - 1, ptl); |
1da177e4 LT |
1365 | return 0; |
1366 | } | |
1367 | ||
1368 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
1369 | unsigned long addr, unsigned long end, | |
1370 | unsigned long pfn, pgprot_t prot) | |
1371 | { | |
1372 | pmd_t *pmd; | |
1373 | unsigned long next; | |
1374 | ||
1375 | pfn -= addr >> PAGE_SHIFT; | |
1376 | pmd = pmd_alloc(mm, pud, addr); | |
1377 | if (!pmd) | |
1378 | return -ENOMEM; | |
1379 | do { | |
1380 | next = pmd_addr_end(addr, end); | |
1381 | if (remap_pte_range(mm, pmd, addr, next, | |
1382 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1383 | return -ENOMEM; | |
1384 | } while (pmd++, addr = next, addr != end); | |
1385 | return 0; | |
1386 | } | |
1387 | ||
1388 | static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, | |
1389 | unsigned long addr, unsigned long end, | |
1390 | unsigned long pfn, pgprot_t prot) | |
1391 | { | |
1392 | pud_t *pud; | |
1393 | unsigned long next; | |
1394 | ||
1395 | pfn -= addr >> PAGE_SHIFT; | |
1396 | pud = pud_alloc(mm, pgd, addr); | |
1397 | if (!pud) | |
1398 | return -ENOMEM; | |
1399 | do { | |
1400 | next = pud_addr_end(addr, end); | |
1401 | if (remap_pmd_range(mm, pud, addr, next, | |
1402 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1403 | return -ENOMEM; | |
1404 | } while (pud++, addr = next, addr != end); | |
1405 | return 0; | |
1406 | } | |
1407 | ||
bfa5bf6d REB |
1408 | /** |
1409 | * remap_pfn_range - remap kernel memory to userspace | |
1410 | * @vma: user vma to map to | |
1411 | * @addr: target user address to start at | |
1412 | * @pfn: physical address of kernel memory | |
1413 | * @size: size of map area | |
1414 | * @prot: page protection flags for this mapping | |
1415 | * | |
1416 | * Note: this is only safe if the mm semaphore is held when called. | |
1417 | */ | |
1da177e4 LT |
1418 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
1419 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
1420 | { | |
1421 | pgd_t *pgd; | |
1422 | unsigned long next; | |
2d15cab8 | 1423 | unsigned long end = addr + PAGE_ALIGN(size); |
1da177e4 LT |
1424 | struct mm_struct *mm = vma->vm_mm; |
1425 | int err; | |
1426 | ||
1427 | /* | |
1428 | * Physically remapped pages are special. Tell the | |
1429 | * rest of the world about it: | |
1430 | * VM_IO tells people not to look at these pages | |
1431 | * (accesses can have side effects). | |
0b14c179 HD |
1432 | * VM_RESERVED is specified all over the place, because |
1433 | * in 2.4 it kept swapout's vma scan off this vma; but | |
1434 | * in 2.6 the LRU scan won't even find its pages, so this | |
1435 | * flag means no more than count its pages in reserved_vm, | |
1436 | * and omit it from core dump, even when VM_IO turned off. | |
6aab341e LT |
1437 | * VM_PFNMAP tells the core MM that the base pages are just |
1438 | * raw PFN mappings, and do not have a "struct page" associated | |
1439 | * with them. | |
fb155c16 LT |
1440 | * |
1441 | * There's a horrible special case to handle copy-on-write | |
1442 | * behaviour that some programs depend on. We mark the "original" | |
1443 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". | |
1da177e4 | 1444 | */ |
67121172 | 1445 | if (is_cow_mapping(vma->vm_flags)) { |
fb155c16 | 1446 | if (addr != vma->vm_start || end != vma->vm_end) |
7fc7e2ee | 1447 | return -EINVAL; |
fb155c16 LT |
1448 | vma->vm_pgoff = pfn; |
1449 | } | |
1450 | ||
6aab341e | 1451 | vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP; |
1da177e4 LT |
1452 | |
1453 | BUG_ON(addr >= end); | |
1454 | pfn -= addr >> PAGE_SHIFT; | |
1455 | pgd = pgd_offset(mm, addr); | |
1456 | flush_cache_range(vma, addr, end); | |
1da177e4 LT |
1457 | do { |
1458 | next = pgd_addr_end(addr, end); | |
1459 | err = remap_pud_range(mm, pgd, addr, next, | |
1460 | pfn + (addr >> PAGE_SHIFT), prot); | |
1461 | if (err) | |
1462 | break; | |
1463 | } while (pgd++, addr = next, addr != end); | |
1da177e4 LT |
1464 | return err; |
1465 | } | |
1466 | EXPORT_SYMBOL(remap_pfn_range); | |
1467 | ||
aee16b3c JF |
1468 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
1469 | unsigned long addr, unsigned long end, | |
1470 | pte_fn_t fn, void *data) | |
1471 | { | |
1472 | pte_t *pte; | |
1473 | int err; | |
2f569afd | 1474 | pgtable_t token; |
94909914 | 1475 | spinlock_t *uninitialized_var(ptl); |
aee16b3c JF |
1476 | |
1477 | pte = (mm == &init_mm) ? | |
1478 | pte_alloc_kernel(pmd, addr) : | |
1479 | pte_alloc_map_lock(mm, pmd, addr, &ptl); | |
1480 | if (!pte) | |
1481 | return -ENOMEM; | |
1482 | ||
1483 | BUG_ON(pmd_huge(*pmd)); | |
1484 | ||
2f569afd | 1485 | token = pmd_pgtable(*pmd); |
aee16b3c JF |
1486 | |
1487 | do { | |
2f569afd | 1488 | err = fn(pte, token, addr, data); |
aee16b3c JF |
1489 | if (err) |
1490 | break; | |
1491 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
1492 | ||
1493 | if (mm != &init_mm) | |
1494 | pte_unmap_unlock(pte-1, ptl); | |
1495 | return err; | |
1496 | } | |
1497 | ||
1498 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, | |
1499 | unsigned long addr, unsigned long end, | |
1500 | pte_fn_t fn, void *data) | |
1501 | { | |
1502 | pmd_t *pmd; | |
1503 | unsigned long next; | |
1504 | int err; | |
1505 | ||
1506 | pmd = pmd_alloc(mm, pud, addr); | |
1507 | if (!pmd) | |
1508 | return -ENOMEM; | |
1509 | do { | |
1510 | next = pmd_addr_end(addr, end); | |
1511 | err = apply_to_pte_range(mm, pmd, addr, next, fn, data); | |
1512 | if (err) | |
1513 | break; | |
1514 | } while (pmd++, addr = next, addr != end); | |
1515 | return err; | |
1516 | } | |
1517 | ||
1518 | static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd, | |
1519 | unsigned long addr, unsigned long end, | |
1520 | pte_fn_t fn, void *data) | |
1521 | { | |
1522 | pud_t *pud; | |
1523 | unsigned long next; | |
1524 | int err; | |
1525 | ||
1526 | pud = pud_alloc(mm, pgd, addr); | |
1527 | if (!pud) | |
1528 | return -ENOMEM; | |
1529 | do { | |
1530 | next = pud_addr_end(addr, end); | |
1531 | err = apply_to_pmd_range(mm, pud, addr, next, fn, data); | |
1532 | if (err) | |
1533 | break; | |
1534 | } while (pud++, addr = next, addr != end); | |
1535 | return err; | |
1536 | } | |
1537 | ||
1538 | /* | |
1539 | * Scan a region of virtual memory, filling in page tables as necessary | |
1540 | * and calling a provided function on each leaf page table. | |
1541 | */ | |
1542 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, | |
1543 | unsigned long size, pte_fn_t fn, void *data) | |
1544 | { | |
1545 | pgd_t *pgd; | |
1546 | unsigned long next; | |
1547 | unsigned long end = addr + size; | |
1548 | int err; | |
1549 | ||
1550 | BUG_ON(addr >= end); | |
1551 | pgd = pgd_offset(mm, addr); | |
1552 | do { | |
1553 | next = pgd_addr_end(addr, end); | |
1554 | err = apply_to_pud_range(mm, pgd, addr, next, fn, data); | |
1555 | if (err) | |
1556 | break; | |
1557 | } while (pgd++, addr = next, addr != end); | |
1558 | return err; | |
1559 | } | |
1560 | EXPORT_SYMBOL_GPL(apply_to_page_range); | |
1561 | ||
8f4e2101 HD |
1562 | /* |
1563 | * handle_pte_fault chooses page fault handler according to an entry | |
1564 | * which was read non-atomically. Before making any commitment, on | |
1565 | * those architectures or configurations (e.g. i386 with PAE) which | |
1566 | * might give a mix of unmatched parts, do_swap_page and do_file_page | |
1567 | * must check under lock before unmapping the pte and proceeding | |
1568 | * (but do_wp_page is only called after already making such a check; | |
1569 | * and do_anonymous_page and do_no_page can safely check later on). | |
1570 | */ | |
4c21e2f2 | 1571 | static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, |
8f4e2101 HD |
1572 | pte_t *page_table, pte_t orig_pte) |
1573 | { | |
1574 | int same = 1; | |
1575 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) | |
1576 | if (sizeof(pte_t) > sizeof(unsigned long)) { | |
4c21e2f2 HD |
1577 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
1578 | spin_lock(ptl); | |
8f4e2101 | 1579 | same = pte_same(*page_table, orig_pte); |
4c21e2f2 | 1580 | spin_unlock(ptl); |
8f4e2101 HD |
1581 | } |
1582 | #endif | |
1583 | pte_unmap(page_table); | |
1584 | return same; | |
1585 | } | |
1586 | ||
1da177e4 LT |
1587 | /* |
1588 | * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when | |
1589 | * servicing faults for write access. In the normal case, do always want | |
1590 | * pte_mkwrite. But get_user_pages can cause write faults for mappings | |
1591 | * that do not have writing enabled, when used by access_process_vm. | |
1592 | */ | |
1593 | static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma) | |
1594 | { | |
1595 | if (likely(vma->vm_flags & VM_WRITE)) | |
1596 | pte = pte_mkwrite(pte); | |
1597 | return pte; | |
1598 | } | |
1599 | ||
9de455b2 | 1600 | static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) |
6aab341e LT |
1601 | { |
1602 | /* | |
1603 | * If the source page was a PFN mapping, we don't have | |
1604 | * a "struct page" for it. We do a best-effort copy by | |
1605 | * just copying from the original user address. If that | |
1606 | * fails, we just zero-fill it. Live with it. | |
1607 | */ | |
1608 | if (unlikely(!src)) { | |
1609 | void *kaddr = kmap_atomic(dst, KM_USER0); | |
5d2a2dbb LT |
1610 | void __user *uaddr = (void __user *)(va & PAGE_MASK); |
1611 | ||
1612 | /* | |
1613 | * This really shouldn't fail, because the page is there | |
1614 | * in the page tables. But it might just be unreadable, | |
1615 | * in which case we just give up and fill the result with | |
1616 | * zeroes. | |
1617 | */ | |
1618 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) | |
6aab341e LT |
1619 | memset(kaddr, 0, PAGE_SIZE); |
1620 | kunmap_atomic(kaddr, KM_USER0); | |
c4ec7b0d | 1621 | flush_dcache_page(dst); |
0ed361de NP |
1622 | } else |
1623 | copy_user_highpage(dst, src, va, vma); | |
6aab341e LT |
1624 | } |
1625 | ||
1da177e4 LT |
1626 | /* |
1627 | * This routine handles present pages, when users try to write | |
1628 | * to a shared page. It is done by copying the page to a new address | |
1629 | * and decrementing the shared-page counter for the old page. | |
1630 | * | |
1da177e4 LT |
1631 | * Note that this routine assumes that the protection checks have been |
1632 | * done by the caller (the low-level page fault routine in most cases). | |
1633 | * Thus we can safely just mark it writable once we've done any necessary | |
1634 | * COW. | |
1635 | * | |
1636 | * We also mark the page dirty at this point even though the page will | |
1637 | * change only once the write actually happens. This avoids a few races, | |
1638 | * and potentially makes it more efficient. | |
1639 | * | |
8f4e2101 HD |
1640 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
1641 | * but allow concurrent faults), with pte both mapped and locked. | |
1642 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 1643 | */ |
65500d23 HD |
1644 | static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, |
1645 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
8f4e2101 | 1646 | spinlock_t *ptl, pte_t orig_pte) |
1da177e4 | 1647 | { |
e5bbe4df | 1648 | struct page *old_page, *new_page; |
1da177e4 | 1649 | pte_t entry; |
83c54070 | 1650 | int reuse = 0, ret = 0; |
a200ee18 | 1651 | int page_mkwrite = 0; |
d08b3851 | 1652 | struct page *dirty_page = NULL; |
1da177e4 | 1653 | |
6aab341e | 1654 | old_page = vm_normal_page(vma, address, orig_pte); |
6aab341e LT |
1655 | if (!old_page) |
1656 | goto gotten; | |
1da177e4 | 1657 | |
d08b3851 | 1658 | /* |
ee6a6457 PZ |
1659 | * Take out anonymous pages first, anonymous shared vmas are |
1660 | * not dirty accountable. | |
d08b3851 | 1661 | */ |
ee6a6457 PZ |
1662 | if (PageAnon(old_page)) { |
1663 | if (!TestSetPageLocked(old_page)) { | |
1664 | reuse = can_share_swap_page(old_page); | |
1665 | unlock_page(old_page); | |
1666 | } | |
1667 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == | |
d08b3851 | 1668 | (VM_WRITE|VM_SHARED))) { |
ee6a6457 PZ |
1669 | /* |
1670 | * Only catch write-faults on shared writable pages, | |
1671 | * read-only shared pages can get COWed by | |
1672 | * get_user_pages(.write=1, .force=1). | |
1673 | */ | |
9637a5ef DH |
1674 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
1675 | /* | |
1676 | * Notify the address space that the page is about to | |
1677 | * become writable so that it can prohibit this or wait | |
1678 | * for the page to get into an appropriate state. | |
1679 | * | |
1680 | * We do this without the lock held, so that it can | |
1681 | * sleep if it needs to. | |
1682 | */ | |
1683 | page_cache_get(old_page); | |
1684 | pte_unmap_unlock(page_table, ptl); | |
1685 | ||
1686 | if (vma->vm_ops->page_mkwrite(vma, old_page) < 0) | |
1687 | goto unwritable_page; | |
1688 | ||
9637a5ef DH |
1689 | /* |
1690 | * Since we dropped the lock we need to revalidate | |
1691 | * the PTE as someone else may have changed it. If | |
1692 | * they did, we just return, as we can count on the | |
1693 | * MMU to tell us if they didn't also make it writable. | |
1694 | */ | |
1695 | page_table = pte_offset_map_lock(mm, pmd, address, | |
1696 | &ptl); | |
c3704ceb | 1697 | page_cache_release(old_page); |
9637a5ef DH |
1698 | if (!pte_same(*page_table, orig_pte)) |
1699 | goto unlock; | |
a200ee18 PZ |
1700 | |
1701 | page_mkwrite = 1; | |
1da177e4 | 1702 | } |
d08b3851 PZ |
1703 | dirty_page = old_page; |
1704 | get_page(dirty_page); | |
9637a5ef | 1705 | reuse = 1; |
9637a5ef DH |
1706 | } |
1707 | ||
1708 | if (reuse) { | |
1709 | flush_cache_page(vma, address, pte_pfn(orig_pte)); | |
1710 | entry = pte_mkyoung(orig_pte); | |
1711 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
954ffcb3 | 1712 | if (ptep_set_access_flags(vma, address, page_table, entry,1)) |
8dab5241 | 1713 | update_mmu_cache(vma, address, entry); |
9637a5ef DH |
1714 | ret |= VM_FAULT_WRITE; |
1715 | goto unlock; | |
1da177e4 | 1716 | } |
1da177e4 LT |
1717 | |
1718 | /* | |
1719 | * Ok, we need to copy. Oh, well.. | |
1720 | */ | |
b5810039 | 1721 | page_cache_get(old_page); |
920fc356 | 1722 | gotten: |
8f4e2101 | 1723 | pte_unmap_unlock(page_table, ptl); |
1da177e4 LT |
1724 | |
1725 | if (unlikely(anon_vma_prepare(vma))) | |
65500d23 | 1726 | goto oom; |
557ed1fa NP |
1727 | VM_BUG_ON(old_page == ZERO_PAGE(0)); |
1728 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); | |
1729 | if (!new_page) | |
1730 | goto oom; | |
1731 | cow_user_page(new_page, old_page, address, vma); | |
0ed361de | 1732 | __SetPageUptodate(new_page); |
65500d23 | 1733 | |
e1a1cd59 | 1734 | if (mem_cgroup_charge(new_page, mm, GFP_KERNEL)) |
8a9f3ccd BS |
1735 | goto oom_free_new; |
1736 | ||
1da177e4 LT |
1737 | /* |
1738 | * Re-check the pte - we dropped the lock | |
1739 | */ | |
8f4e2101 | 1740 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
65500d23 | 1741 | if (likely(pte_same(*page_table, orig_pte))) { |
920fc356 | 1742 | if (old_page) { |
7de6b805 | 1743 | page_remove_rmap(old_page, vma); |
920fc356 HD |
1744 | if (!PageAnon(old_page)) { |
1745 | dec_mm_counter(mm, file_rss); | |
1746 | inc_mm_counter(mm, anon_rss); | |
1747 | } | |
1748 | } else | |
4294621f | 1749 | inc_mm_counter(mm, anon_rss); |
eca35133 | 1750 | flush_cache_page(vma, address, pte_pfn(orig_pte)); |
65500d23 HD |
1751 | entry = mk_pte(new_page, vma->vm_page_prot); |
1752 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
4ce072f1 SS |
1753 | /* |
1754 | * Clear the pte entry and flush it first, before updating the | |
1755 | * pte with the new entry. This will avoid a race condition | |
1756 | * seen in the presence of one thread doing SMC and another | |
1757 | * thread doing COW. | |
1758 | */ | |
1759 | ptep_clear_flush(vma, address, page_table); | |
1760 | set_pte_at(mm, address, page_table, entry); | |
65500d23 | 1761 | update_mmu_cache(vma, address, entry); |
1da177e4 | 1762 | lru_cache_add_active(new_page); |
9617d95e | 1763 | page_add_new_anon_rmap(new_page, vma, address); |
1da177e4 LT |
1764 | |
1765 | /* Free the old page.. */ | |
1766 | new_page = old_page; | |
f33ea7f4 | 1767 | ret |= VM_FAULT_WRITE; |
8a9f3ccd BS |
1768 | } else |
1769 | mem_cgroup_uncharge_page(new_page); | |
1770 | ||
920fc356 HD |
1771 | if (new_page) |
1772 | page_cache_release(new_page); | |
1773 | if (old_page) | |
1774 | page_cache_release(old_page); | |
65500d23 | 1775 | unlock: |
8f4e2101 | 1776 | pte_unmap_unlock(page_table, ptl); |
d08b3851 | 1777 | if (dirty_page) { |
8f7b3d15 AS |
1778 | if (vma->vm_file) |
1779 | file_update_time(vma->vm_file); | |
1780 | ||
79352894 NP |
1781 | /* |
1782 | * Yes, Virginia, this is actually required to prevent a race | |
1783 | * with clear_page_dirty_for_io() from clearing the page dirty | |
1784 | * bit after it clear all dirty ptes, but before a racing | |
1785 | * do_wp_page installs a dirty pte. | |
1786 | * | |
1787 | * do_no_page is protected similarly. | |
1788 | */ | |
1789 | wait_on_page_locked(dirty_page); | |
a200ee18 | 1790 | set_page_dirty_balance(dirty_page, page_mkwrite); |
d08b3851 PZ |
1791 | put_page(dirty_page); |
1792 | } | |
f33ea7f4 | 1793 | return ret; |
8a9f3ccd | 1794 | oom_free_new: |
6dbf6d3b | 1795 | page_cache_release(new_page); |
65500d23 | 1796 | oom: |
920fc356 HD |
1797 | if (old_page) |
1798 | page_cache_release(old_page); | |
1da177e4 | 1799 | return VM_FAULT_OOM; |
9637a5ef DH |
1800 | |
1801 | unwritable_page: | |
1802 | page_cache_release(old_page); | |
1803 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
1804 | } |
1805 | ||
1806 | /* | |
1807 | * Helper functions for unmap_mapping_range(). | |
1808 | * | |
1809 | * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __ | |
1810 | * | |
1811 | * We have to restart searching the prio_tree whenever we drop the lock, | |
1812 | * since the iterator is only valid while the lock is held, and anyway | |
1813 | * a later vma might be split and reinserted earlier while lock dropped. | |
1814 | * | |
1815 | * The list of nonlinear vmas could be handled more efficiently, using | |
1816 | * a placeholder, but handle it in the same way until a need is shown. | |
1817 | * It is important to search the prio_tree before nonlinear list: a vma | |
1818 | * may become nonlinear and be shifted from prio_tree to nonlinear list | |
1819 | * while the lock is dropped; but never shifted from list to prio_tree. | |
1820 | * | |
1821 | * In order to make forward progress despite restarting the search, | |
1822 | * vm_truncate_count is used to mark a vma as now dealt with, so we can | |
1823 | * quickly skip it next time around. Since the prio_tree search only | |
1824 | * shows us those vmas affected by unmapping the range in question, we | |
1825 | * can't efficiently keep all vmas in step with mapping->truncate_count: | |
1826 | * so instead reset them all whenever it wraps back to 0 (then go to 1). | |
1827 | * mapping->truncate_count and vma->vm_truncate_count are protected by | |
1828 | * i_mmap_lock. | |
1829 | * | |
1830 | * In order to make forward progress despite repeatedly restarting some | |
ee39b37b | 1831 | * large vma, note the restart_addr from unmap_vmas when it breaks out: |
1da177e4 LT |
1832 | * and restart from that address when we reach that vma again. It might |
1833 | * have been split or merged, shrunk or extended, but never shifted: so | |
1834 | * restart_addr remains valid so long as it remains in the vma's range. | |
1835 | * unmap_mapping_range forces truncate_count to leap over page-aligned | |
1836 | * values so we can save vma's restart_addr in its truncate_count field. | |
1837 | */ | |
1838 | #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK)) | |
1839 | ||
1840 | static void reset_vma_truncate_counts(struct address_space *mapping) | |
1841 | { | |
1842 | struct vm_area_struct *vma; | |
1843 | struct prio_tree_iter iter; | |
1844 | ||
1845 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX) | |
1846 | vma->vm_truncate_count = 0; | |
1847 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) | |
1848 | vma->vm_truncate_count = 0; | |
1849 | } | |
1850 | ||
1851 | static int unmap_mapping_range_vma(struct vm_area_struct *vma, | |
1852 | unsigned long start_addr, unsigned long end_addr, | |
1853 | struct zap_details *details) | |
1854 | { | |
1855 | unsigned long restart_addr; | |
1856 | int need_break; | |
1857 | ||
d00806b1 NP |
1858 | /* |
1859 | * files that support invalidating or truncating portions of the | |
d0217ac0 | 1860 | * file from under mmaped areas must have their ->fault function |
83c54070 NP |
1861 | * return a locked page (and set VM_FAULT_LOCKED in the return). |
1862 | * This provides synchronisation against concurrent unmapping here. | |
d00806b1 | 1863 | */ |
d00806b1 | 1864 | |
1da177e4 LT |
1865 | again: |
1866 | restart_addr = vma->vm_truncate_count; | |
1867 | if (is_restart_addr(restart_addr) && start_addr < restart_addr) { | |
1868 | start_addr = restart_addr; | |
1869 | if (start_addr >= end_addr) { | |
1870 | /* Top of vma has been split off since last time */ | |
1871 | vma->vm_truncate_count = details->truncate_count; | |
1872 | return 0; | |
1873 | } | |
1874 | } | |
1875 | ||
ee39b37b HD |
1876 | restart_addr = zap_page_range(vma, start_addr, |
1877 | end_addr - start_addr, details); | |
95c354fe | 1878 | need_break = need_resched() || spin_needbreak(details->i_mmap_lock); |
1da177e4 | 1879 | |
ee39b37b | 1880 | if (restart_addr >= end_addr) { |
1da177e4 LT |
1881 | /* We have now completed this vma: mark it so */ |
1882 | vma->vm_truncate_count = details->truncate_count; | |
1883 | if (!need_break) | |
1884 | return 0; | |
1885 | } else { | |
1886 | /* Note restart_addr in vma's truncate_count field */ | |
ee39b37b | 1887 | vma->vm_truncate_count = restart_addr; |
1da177e4 LT |
1888 | if (!need_break) |
1889 | goto again; | |
1890 | } | |
1891 | ||
1892 | spin_unlock(details->i_mmap_lock); | |
1893 | cond_resched(); | |
1894 | spin_lock(details->i_mmap_lock); | |
1895 | return -EINTR; | |
1896 | } | |
1897 | ||
1898 | static inline void unmap_mapping_range_tree(struct prio_tree_root *root, | |
1899 | struct zap_details *details) | |
1900 | { | |
1901 | struct vm_area_struct *vma; | |
1902 | struct prio_tree_iter iter; | |
1903 | pgoff_t vba, vea, zba, zea; | |
1904 | ||
1905 | restart: | |
1906 | vma_prio_tree_foreach(vma, &iter, root, | |
1907 | details->first_index, details->last_index) { | |
1908 | /* Skip quickly over those we have already dealt with */ | |
1909 | if (vma->vm_truncate_count == details->truncate_count) | |
1910 | continue; | |
1911 | ||
1912 | vba = vma->vm_pgoff; | |
1913 | vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1; | |
1914 | /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */ | |
1915 | zba = details->first_index; | |
1916 | if (zba < vba) | |
1917 | zba = vba; | |
1918 | zea = details->last_index; | |
1919 | if (zea > vea) | |
1920 | zea = vea; | |
1921 | ||
1922 | if (unmap_mapping_range_vma(vma, | |
1923 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, | |
1924 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, | |
1925 | details) < 0) | |
1926 | goto restart; | |
1927 | } | |
1928 | } | |
1929 | ||
1930 | static inline void unmap_mapping_range_list(struct list_head *head, | |
1931 | struct zap_details *details) | |
1932 | { | |
1933 | struct vm_area_struct *vma; | |
1934 | ||
1935 | /* | |
1936 | * In nonlinear VMAs there is no correspondence between virtual address | |
1937 | * offset and file offset. So we must perform an exhaustive search | |
1938 | * across *all* the pages in each nonlinear VMA, not just the pages | |
1939 | * whose virtual address lies outside the file truncation point. | |
1940 | */ | |
1941 | restart: | |
1942 | list_for_each_entry(vma, head, shared.vm_set.list) { | |
1943 | /* Skip quickly over those we have already dealt with */ | |
1944 | if (vma->vm_truncate_count == details->truncate_count) | |
1945 | continue; | |
1946 | details->nonlinear_vma = vma; | |
1947 | if (unmap_mapping_range_vma(vma, vma->vm_start, | |
1948 | vma->vm_end, details) < 0) | |
1949 | goto restart; | |
1950 | } | |
1951 | } | |
1952 | ||
1953 | /** | |
72fd4a35 | 1954 | * 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 | 1955 | * @mapping: the address space containing mmaps to be unmapped. |
1da177e4 LT |
1956 | * @holebegin: byte in first page to unmap, relative to the start of |
1957 | * the underlying file. This will be rounded down to a PAGE_SIZE | |
1958 | * boundary. Note that this is different from vmtruncate(), which | |
1959 | * must keep the partial page. In contrast, we must get rid of | |
1960 | * partial pages. | |
1961 | * @holelen: size of prospective hole in bytes. This will be rounded | |
1962 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the | |
1963 | * end of the file. | |
1964 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; | |
1965 | * but 0 when invalidating pagecache, don't throw away private data. | |
1966 | */ | |
1967 | void unmap_mapping_range(struct address_space *mapping, | |
1968 | loff_t const holebegin, loff_t const holelen, int even_cows) | |
1969 | { | |
1970 | struct zap_details details; | |
1971 | pgoff_t hba = holebegin >> PAGE_SHIFT; | |
1972 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1973 | ||
1974 | /* Check for overflow. */ | |
1975 | if (sizeof(holelen) > sizeof(hlen)) { | |
1976 | long long holeend = | |
1977 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1978 | if (holeend & ~(long long)ULONG_MAX) | |
1979 | hlen = ULONG_MAX - hba + 1; | |
1980 | } | |
1981 | ||
1982 | details.check_mapping = even_cows? NULL: mapping; | |
1983 | details.nonlinear_vma = NULL; | |
1984 | details.first_index = hba; | |
1985 | details.last_index = hba + hlen - 1; | |
1986 | if (details.last_index < details.first_index) | |
1987 | details.last_index = ULONG_MAX; | |
1988 | details.i_mmap_lock = &mapping->i_mmap_lock; | |
1989 | ||
1990 | spin_lock(&mapping->i_mmap_lock); | |
1991 | ||
d00806b1 | 1992 | /* Protect against endless unmapping loops */ |
1da177e4 | 1993 | mapping->truncate_count++; |
1da177e4 LT |
1994 | if (unlikely(is_restart_addr(mapping->truncate_count))) { |
1995 | if (mapping->truncate_count == 0) | |
1996 | reset_vma_truncate_counts(mapping); | |
1997 | mapping->truncate_count++; | |
1998 | } | |
1999 | details.truncate_count = mapping->truncate_count; | |
2000 | ||
2001 | if (unlikely(!prio_tree_empty(&mapping->i_mmap))) | |
2002 | unmap_mapping_range_tree(&mapping->i_mmap, &details); | |
2003 | if (unlikely(!list_empty(&mapping->i_mmap_nonlinear))) | |
2004 | unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details); | |
2005 | spin_unlock(&mapping->i_mmap_lock); | |
2006 | } | |
2007 | EXPORT_SYMBOL(unmap_mapping_range); | |
2008 | ||
bfa5bf6d REB |
2009 | /** |
2010 | * vmtruncate - unmap mappings "freed" by truncate() syscall | |
2011 | * @inode: inode of the file used | |
2012 | * @offset: file offset to start truncating | |
1da177e4 LT |
2013 | * |
2014 | * NOTE! We have to be ready to update the memory sharing | |
2015 | * between the file and the memory map for a potential last | |
2016 | * incomplete page. Ugly, but necessary. | |
2017 | */ | |
2018 | int vmtruncate(struct inode * inode, loff_t offset) | |
2019 | { | |
61d5048f CH |
2020 | if (inode->i_size < offset) { |
2021 | unsigned long limit; | |
2022 | ||
2023 | limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; | |
2024 | if (limit != RLIM_INFINITY && offset > limit) | |
2025 | goto out_sig; | |
2026 | if (offset > inode->i_sb->s_maxbytes) | |
2027 | goto out_big; | |
2028 | i_size_write(inode, offset); | |
2029 | } else { | |
2030 | struct address_space *mapping = inode->i_mapping; | |
1da177e4 | 2031 | |
61d5048f CH |
2032 | /* |
2033 | * truncation of in-use swapfiles is disallowed - it would | |
2034 | * cause subsequent swapout to scribble on the now-freed | |
2035 | * blocks. | |
2036 | */ | |
2037 | if (IS_SWAPFILE(inode)) | |
2038 | return -ETXTBSY; | |
2039 | i_size_write(inode, offset); | |
2040 | ||
2041 | /* | |
2042 | * unmap_mapping_range is called twice, first simply for | |
2043 | * efficiency so that truncate_inode_pages does fewer | |
2044 | * single-page unmaps. However after this first call, and | |
2045 | * before truncate_inode_pages finishes, it is possible for | |
2046 | * private pages to be COWed, which remain after | |
2047 | * truncate_inode_pages finishes, hence the second | |
2048 | * unmap_mapping_range call must be made for correctness. | |
2049 | */ | |
2050 | unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1); | |
2051 | truncate_inode_pages(mapping, offset); | |
2052 | unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1); | |
2053 | } | |
d00806b1 | 2054 | |
1da177e4 LT |
2055 | if (inode->i_op && inode->i_op->truncate) |
2056 | inode->i_op->truncate(inode); | |
2057 | return 0; | |
61d5048f | 2058 | |
1da177e4 LT |
2059 | out_sig: |
2060 | send_sig(SIGXFSZ, current, 0); | |
2061 | out_big: | |
2062 | return -EFBIG; | |
1da177e4 | 2063 | } |
1da177e4 LT |
2064 | EXPORT_SYMBOL(vmtruncate); |
2065 | ||
f6b3ec23 BP |
2066 | int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end) |
2067 | { | |
2068 | struct address_space *mapping = inode->i_mapping; | |
2069 | ||
2070 | /* | |
2071 | * If the underlying filesystem is not going to provide | |
2072 | * a way to truncate a range of blocks (punch a hole) - | |
2073 | * we should return failure right now. | |
2074 | */ | |
2075 | if (!inode->i_op || !inode->i_op->truncate_range) | |
2076 | return -ENOSYS; | |
2077 | ||
1b1dcc1b | 2078 | mutex_lock(&inode->i_mutex); |
f6b3ec23 BP |
2079 | down_write(&inode->i_alloc_sem); |
2080 | unmap_mapping_range(mapping, offset, (end - offset), 1); | |
2081 | truncate_inode_pages_range(mapping, offset, end); | |
d00806b1 | 2082 | unmap_mapping_range(mapping, offset, (end - offset), 1); |
f6b3ec23 BP |
2083 | inode->i_op->truncate_range(inode, offset, end); |
2084 | up_write(&inode->i_alloc_sem); | |
1b1dcc1b | 2085 | mutex_unlock(&inode->i_mutex); |
f6b3ec23 BP |
2086 | |
2087 | return 0; | |
2088 | } | |
f6b3ec23 | 2089 | |
1da177e4 | 2090 | /* |
8f4e2101 HD |
2091 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2092 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2093 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2094 | */ |
65500d23 HD |
2095 | static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2096 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
2097 | int write_access, pte_t orig_pte) | |
1da177e4 | 2098 | { |
8f4e2101 | 2099 | spinlock_t *ptl; |
1da177e4 | 2100 | struct page *page; |
65500d23 | 2101 | swp_entry_t entry; |
1da177e4 | 2102 | pte_t pte; |
83c54070 | 2103 | int ret = 0; |
1da177e4 | 2104 | |
4c21e2f2 | 2105 | if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) |
8f4e2101 | 2106 | goto out; |
65500d23 HD |
2107 | |
2108 | entry = pte_to_swp_entry(orig_pte); | |
0697212a CL |
2109 | if (is_migration_entry(entry)) { |
2110 | migration_entry_wait(mm, pmd, address); | |
2111 | goto out; | |
2112 | } | |
0ff92245 | 2113 | delayacct_set_flag(DELAYACCT_PF_SWAPIN); |
1da177e4 LT |
2114 | page = lookup_swap_cache(entry); |
2115 | if (!page) { | |
098fe651 | 2116 | grab_swap_token(); /* Contend for token _before_ read-in */ |
02098fea HD |
2117 | page = swapin_readahead(entry, |
2118 | GFP_HIGHUSER_MOVABLE, vma, address); | |
1da177e4 LT |
2119 | if (!page) { |
2120 | /* | |
8f4e2101 HD |
2121 | * Back out if somebody else faulted in this pte |
2122 | * while we released the pte lock. | |
1da177e4 | 2123 | */ |
8f4e2101 | 2124 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
1da177e4 LT |
2125 | if (likely(pte_same(*page_table, orig_pte))) |
2126 | ret = VM_FAULT_OOM; | |
0ff92245 | 2127 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
65500d23 | 2128 | goto unlock; |
1da177e4 LT |
2129 | } |
2130 | ||
2131 | /* Had to read the page from swap area: Major fault */ | |
2132 | ret = VM_FAULT_MAJOR; | |
f8891e5e | 2133 | count_vm_event(PGMAJFAULT); |
1da177e4 LT |
2134 | } |
2135 | ||
e1a1cd59 | 2136 | if (mem_cgroup_charge(page, mm, GFP_KERNEL)) { |
8a9f3ccd BS |
2137 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
2138 | ret = VM_FAULT_OOM; | |
2139 | goto out; | |
2140 | } | |
2141 | ||
1da177e4 LT |
2142 | mark_page_accessed(page); |
2143 | lock_page(page); | |
20a1022d | 2144 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
1da177e4 LT |
2145 | |
2146 | /* | |
8f4e2101 | 2147 | * Back out if somebody else already faulted in this pte. |
1da177e4 | 2148 | */ |
8f4e2101 | 2149 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
9e9bef07 | 2150 | if (unlikely(!pte_same(*page_table, orig_pte))) |
b8107480 | 2151 | goto out_nomap; |
b8107480 KK |
2152 | |
2153 | if (unlikely(!PageUptodate(page))) { | |
2154 | ret = VM_FAULT_SIGBUS; | |
2155 | goto out_nomap; | |
1da177e4 LT |
2156 | } |
2157 | ||
2158 | /* The page isn't present yet, go ahead with the fault. */ | |
1da177e4 | 2159 | |
4294621f | 2160 | inc_mm_counter(mm, anon_rss); |
1da177e4 LT |
2161 | pte = mk_pte(page, vma->vm_page_prot); |
2162 | if (write_access && can_share_swap_page(page)) { | |
2163 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); | |
2164 | write_access = 0; | |
2165 | } | |
1da177e4 LT |
2166 | |
2167 | flush_icache_page(vma, page); | |
2168 | set_pte_at(mm, address, page_table, pte); | |
2169 | page_add_anon_rmap(page, vma, address); | |
2170 | ||
c475a8ab HD |
2171 | swap_free(entry); |
2172 | if (vm_swap_full()) | |
2173 | remove_exclusive_swap_page(page); | |
2174 | unlock_page(page); | |
2175 | ||
1da177e4 | 2176 | if (write_access) { |
61469f1d HD |
2177 | ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte); |
2178 | if (ret & VM_FAULT_ERROR) | |
2179 | ret &= VM_FAULT_ERROR; | |
1da177e4 LT |
2180 | goto out; |
2181 | } | |
2182 | ||
2183 | /* No need to invalidate - it was non-present before */ | |
2184 | update_mmu_cache(vma, address, pte); | |
65500d23 | 2185 | unlock: |
8f4e2101 | 2186 | pte_unmap_unlock(page_table, ptl); |
1da177e4 LT |
2187 | out: |
2188 | return ret; | |
b8107480 | 2189 | out_nomap: |
8a9f3ccd | 2190 | mem_cgroup_uncharge_page(page); |
8f4e2101 | 2191 | pte_unmap_unlock(page_table, ptl); |
b8107480 KK |
2192 | unlock_page(page); |
2193 | page_cache_release(page); | |
65500d23 | 2194 | return ret; |
1da177e4 LT |
2195 | } |
2196 | ||
2197 | /* | |
8f4e2101 HD |
2198 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2199 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2200 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2201 | */ |
65500d23 HD |
2202 | static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2203 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
2204 | int write_access) | |
1da177e4 | 2205 | { |
8f4e2101 HD |
2206 | struct page *page; |
2207 | spinlock_t *ptl; | |
1da177e4 | 2208 | pte_t entry; |
1da177e4 | 2209 | |
557ed1fa NP |
2210 | /* Allocate our own private page. */ |
2211 | pte_unmap(page_table); | |
8f4e2101 | 2212 | |
557ed1fa NP |
2213 | if (unlikely(anon_vma_prepare(vma))) |
2214 | goto oom; | |
2215 | page = alloc_zeroed_user_highpage_movable(vma, address); | |
2216 | if (!page) | |
2217 | goto oom; | |
0ed361de | 2218 | __SetPageUptodate(page); |
8f4e2101 | 2219 | |
e1a1cd59 | 2220 | if (mem_cgroup_charge(page, mm, GFP_KERNEL)) |
8a9f3ccd BS |
2221 | goto oom_free_page; |
2222 | ||
557ed1fa NP |
2223 | entry = mk_pte(page, vma->vm_page_prot); |
2224 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
1da177e4 | 2225 | |
557ed1fa NP |
2226 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
2227 | if (!pte_none(*page_table)) | |
2228 | goto release; | |
2229 | inc_mm_counter(mm, anon_rss); | |
2230 | lru_cache_add_active(page); | |
2231 | page_add_new_anon_rmap(page, vma, address); | |
65500d23 | 2232 | set_pte_at(mm, address, page_table, entry); |
1da177e4 LT |
2233 | |
2234 | /* No need to invalidate - it was non-present before */ | |
65500d23 | 2235 | update_mmu_cache(vma, address, entry); |
65500d23 | 2236 | unlock: |
8f4e2101 | 2237 | pte_unmap_unlock(page_table, ptl); |
83c54070 | 2238 | return 0; |
8f4e2101 | 2239 | release: |
8a9f3ccd | 2240 | mem_cgroup_uncharge_page(page); |
8f4e2101 HD |
2241 | page_cache_release(page); |
2242 | goto unlock; | |
8a9f3ccd | 2243 | oom_free_page: |
6dbf6d3b | 2244 | page_cache_release(page); |
65500d23 | 2245 | oom: |
1da177e4 LT |
2246 | return VM_FAULT_OOM; |
2247 | } | |
2248 | ||
2249 | /* | |
54cb8821 | 2250 | * __do_fault() tries to create a new page mapping. It aggressively |
1da177e4 | 2251 | * tries to share with existing pages, but makes a separate copy if |
54cb8821 NP |
2252 | * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid |
2253 | * the next page fault. | |
1da177e4 LT |
2254 | * |
2255 | * As this is called only for pages that do not currently exist, we | |
2256 | * do not need to flush old virtual caches or the TLB. | |
2257 | * | |
8f4e2101 | 2258 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
16abfa08 | 2259 | * but allow concurrent faults), and pte neither mapped nor locked. |
8f4e2101 | 2260 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
1da177e4 | 2261 | */ |
54cb8821 | 2262 | static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
16abfa08 | 2263 | unsigned long address, pmd_t *pmd, |
54cb8821 | 2264 | pgoff_t pgoff, unsigned int flags, pte_t orig_pte) |
1da177e4 | 2265 | { |
16abfa08 | 2266 | pte_t *page_table; |
8f4e2101 | 2267 | spinlock_t *ptl; |
d0217ac0 | 2268 | struct page *page; |
1da177e4 | 2269 | pte_t entry; |
1da177e4 | 2270 | int anon = 0; |
d08b3851 | 2271 | struct page *dirty_page = NULL; |
d0217ac0 NP |
2272 | struct vm_fault vmf; |
2273 | int ret; | |
a200ee18 | 2274 | int page_mkwrite = 0; |
54cb8821 | 2275 | |
d0217ac0 NP |
2276 | vmf.virtual_address = (void __user *)(address & PAGE_MASK); |
2277 | vmf.pgoff = pgoff; | |
2278 | vmf.flags = flags; | |
2279 | vmf.page = NULL; | |
1da177e4 | 2280 | |
325f04db HD |
2281 | BUG_ON(vma->vm_flags & VM_PFNMAP); |
2282 | ||
3c18ddd1 NP |
2283 | ret = vma->vm_ops->fault(vma, &vmf); |
2284 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) | |
2285 | return ret; | |
1da177e4 | 2286 | |
d00806b1 | 2287 | /* |
d0217ac0 | 2288 | * For consistency in subsequent calls, make the faulted page always |
d00806b1 NP |
2289 | * locked. |
2290 | */ | |
83c54070 | 2291 | if (unlikely(!(ret & VM_FAULT_LOCKED))) |
d0217ac0 | 2292 | lock_page(vmf.page); |
54cb8821 | 2293 | else |
d0217ac0 | 2294 | VM_BUG_ON(!PageLocked(vmf.page)); |
d00806b1 | 2295 | |
1da177e4 LT |
2296 | /* |
2297 | * Should we do an early C-O-W break? | |
2298 | */ | |
d0217ac0 | 2299 | page = vmf.page; |
54cb8821 | 2300 | if (flags & FAULT_FLAG_WRITE) { |
9637a5ef | 2301 | if (!(vma->vm_flags & VM_SHARED)) { |
54cb8821 | 2302 | anon = 1; |
d00806b1 | 2303 | if (unlikely(anon_vma_prepare(vma))) { |
d0217ac0 | 2304 | ret = VM_FAULT_OOM; |
54cb8821 | 2305 | goto out; |
d00806b1 | 2306 | } |
83c54070 NP |
2307 | page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, |
2308 | vma, address); | |
d00806b1 | 2309 | if (!page) { |
d0217ac0 | 2310 | ret = VM_FAULT_OOM; |
54cb8821 | 2311 | goto out; |
d00806b1 | 2312 | } |
d0217ac0 | 2313 | copy_user_highpage(page, vmf.page, address, vma); |
0ed361de | 2314 | __SetPageUptodate(page); |
9637a5ef | 2315 | } else { |
54cb8821 NP |
2316 | /* |
2317 | * If the page will be shareable, see if the backing | |
9637a5ef | 2318 | * address space wants to know that the page is about |
54cb8821 NP |
2319 | * to become writable |
2320 | */ | |
69676147 MF |
2321 | if (vma->vm_ops->page_mkwrite) { |
2322 | unlock_page(page); | |
2323 | if (vma->vm_ops->page_mkwrite(vma, page) < 0) { | |
d0217ac0 NP |
2324 | ret = VM_FAULT_SIGBUS; |
2325 | anon = 1; /* no anon but release vmf.page */ | |
69676147 MF |
2326 | goto out_unlocked; |
2327 | } | |
2328 | lock_page(page); | |
d0217ac0 NP |
2329 | /* |
2330 | * XXX: this is not quite right (racy vs | |
2331 | * invalidate) to unlock and relock the page | |
2332 | * like this, however a better fix requires | |
2333 | * reworking page_mkwrite locking API, which | |
2334 | * is better done later. | |
2335 | */ | |
2336 | if (!page->mapping) { | |
83c54070 | 2337 | ret = 0; |
d0217ac0 NP |
2338 | anon = 1; /* no anon but release vmf.page */ |
2339 | goto out; | |
2340 | } | |
a200ee18 | 2341 | page_mkwrite = 1; |
9637a5ef DH |
2342 | } |
2343 | } | |
54cb8821 | 2344 | |
1da177e4 LT |
2345 | } |
2346 | ||
e1a1cd59 | 2347 | if (mem_cgroup_charge(page, mm, GFP_KERNEL)) { |
8a9f3ccd BS |
2348 | ret = VM_FAULT_OOM; |
2349 | goto out; | |
2350 | } | |
2351 | ||
8f4e2101 | 2352 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
1da177e4 LT |
2353 | |
2354 | /* | |
2355 | * This silly early PAGE_DIRTY setting removes a race | |
2356 | * due to the bad i386 page protection. But it's valid | |
2357 | * for other architectures too. | |
2358 | * | |
2359 | * Note that if write_access is true, we either now have | |
2360 | * an exclusive copy of the page, or this is a shared mapping, | |
2361 | * so we can make it writable and dirty to avoid having to | |
2362 | * handle that later. | |
2363 | */ | |
2364 | /* Only go through if we didn't race with anybody else... */ | |
54cb8821 | 2365 | if (likely(pte_same(*page_table, orig_pte))) { |
d00806b1 NP |
2366 | flush_icache_page(vma, page); |
2367 | entry = mk_pte(page, vma->vm_page_prot); | |
54cb8821 | 2368 | if (flags & FAULT_FLAG_WRITE) |
1da177e4 LT |
2369 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
2370 | set_pte_at(mm, address, page_table, entry); | |
2371 | if (anon) { | |
d00806b1 NP |
2372 | inc_mm_counter(mm, anon_rss); |
2373 | lru_cache_add_active(page); | |
2374 | page_add_new_anon_rmap(page, vma, address); | |
f57e88a8 | 2375 | } else { |
4294621f | 2376 | inc_mm_counter(mm, file_rss); |
d00806b1 | 2377 | page_add_file_rmap(page); |
54cb8821 | 2378 | if (flags & FAULT_FLAG_WRITE) { |
d00806b1 | 2379 | dirty_page = page; |
d08b3851 PZ |
2380 | get_page(dirty_page); |
2381 | } | |
4294621f | 2382 | } |
d00806b1 NP |
2383 | |
2384 | /* no need to invalidate: a not-present page won't be cached */ | |
2385 | update_mmu_cache(vma, address, entry); | |
1da177e4 | 2386 | } else { |
8a9f3ccd | 2387 | mem_cgroup_uncharge_page(page); |
d00806b1 NP |
2388 | if (anon) |
2389 | page_cache_release(page); | |
2390 | else | |
54cb8821 | 2391 | anon = 1; /* no anon but release faulted_page */ |
1da177e4 LT |
2392 | } |
2393 | ||
8f4e2101 | 2394 | pte_unmap_unlock(page_table, ptl); |
d00806b1 NP |
2395 | |
2396 | out: | |
d0217ac0 | 2397 | unlock_page(vmf.page); |
69676147 | 2398 | out_unlocked: |
d00806b1 | 2399 | if (anon) |
d0217ac0 | 2400 | page_cache_release(vmf.page); |
d00806b1 | 2401 | else if (dirty_page) { |
8f7b3d15 AS |
2402 | if (vma->vm_file) |
2403 | file_update_time(vma->vm_file); | |
2404 | ||
a200ee18 | 2405 | set_page_dirty_balance(dirty_page, page_mkwrite); |
d08b3851 PZ |
2406 | put_page(dirty_page); |
2407 | } | |
d00806b1 | 2408 | |
83c54070 | 2409 | return ret; |
54cb8821 | 2410 | } |
d00806b1 | 2411 | |
54cb8821 NP |
2412 | static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
2413 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
2414 | int write_access, pte_t orig_pte) | |
2415 | { | |
2416 | pgoff_t pgoff = (((address & PAGE_MASK) | |
0da7e01f | 2417 | - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
54cb8821 NP |
2418 | unsigned int flags = (write_access ? FAULT_FLAG_WRITE : 0); |
2419 | ||
16abfa08 HD |
2420 | pte_unmap(page_table); |
2421 | return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); | |
54cb8821 NP |
2422 | } |
2423 | ||
1da177e4 | 2424 | |
f4b81804 JS |
2425 | /* |
2426 | * do_no_pfn() tries to create a new page mapping for a page without | |
2427 | * a struct_page backing it | |
2428 | * | |
2429 | * As this is called only for pages that do not currently exist, we | |
2430 | * do not need to flush old virtual caches or the TLB. | |
2431 | * | |
2432 | * We enter with non-exclusive mmap_sem (to exclude vma changes, | |
2433 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2434 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
2435 | * | |
2436 | * It is expected that the ->nopfn handler always returns the same pfn | |
2437 | * for a given virtual mapping. | |
2438 | * | |
2439 | * Mark this `noinline' to prevent it from bloating the main pagefault code. | |
2440 | */ | |
2441 | static noinline int do_no_pfn(struct mm_struct *mm, struct vm_area_struct *vma, | |
2442 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
2443 | int write_access) | |
2444 | { | |
2445 | spinlock_t *ptl; | |
2446 | pte_t entry; | |
2447 | unsigned long pfn; | |
f4b81804 JS |
2448 | |
2449 | pte_unmap(page_table); | |
b379d790 JH |
2450 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); |
2451 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); | |
f4b81804 JS |
2452 | |
2453 | pfn = vma->vm_ops->nopfn(vma, address & PAGE_MASK); | |
b379d790 JH |
2454 | |
2455 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); | |
2456 | ||
22cd25ed | 2457 | if (unlikely(pfn == NOPFN_OOM)) |
f4b81804 | 2458 | return VM_FAULT_OOM; |
22cd25ed | 2459 | else if (unlikely(pfn == NOPFN_SIGBUS)) |
f4b81804 | 2460 | return VM_FAULT_SIGBUS; |
22cd25ed | 2461 | else if (unlikely(pfn == NOPFN_REFAULT)) |
83c54070 | 2462 | return 0; |
f4b81804 JS |
2463 | |
2464 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); | |
2465 | ||
2466 | /* Only go through if we didn't race with anybody else... */ | |
2467 | if (pte_none(*page_table)) { | |
2468 | entry = pfn_pte(pfn, vma->vm_page_prot); | |
2469 | if (write_access) | |
2470 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
2471 | set_pte_at(mm, address, page_table, entry); | |
2472 | } | |
2473 | pte_unmap_unlock(page_table, ptl); | |
83c54070 | 2474 | return 0; |
f4b81804 JS |
2475 | } |
2476 | ||
1da177e4 LT |
2477 | /* |
2478 | * Fault of a previously existing named mapping. Repopulate the pte | |
2479 | * from the encoded file_pte if possible. This enables swappable | |
2480 | * nonlinear vmas. | |
8f4e2101 HD |
2481 | * |
2482 | * We enter with non-exclusive mmap_sem (to exclude vma changes, | |
2483 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2484 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2485 | */ |
d0217ac0 | 2486 | static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
65500d23 HD |
2487 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
2488 | int write_access, pte_t orig_pte) | |
1da177e4 | 2489 | { |
d0217ac0 NP |
2490 | unsigned int flags = FAULT_FLAG_NONLINEAR | |
2491 | (write_access ? FAULT_FLAG_WRITE : 0); | |
65500d23 | 2492 | pgoff_t pgoff; |
1da177e4 | 2493 | |
4c21e2f2 | 2494 | if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) |
83c54070 | 2495 | return 0; |
1da177e4 | 2496 | |
d0217ac0 NP |
2497 | if (unlikely(!(vma->vm_flags & VM_NONLINEAR) || |
2498 | !(vma->vm_flags & VM_CAN_NONLINEAR))) { | |
65500d23 HD |
2499 | /* |
2500 | * Page table corrupted: show pte and kill process. | |
2501 | */ | |
b5810039 | 2502 | print_bad_pte(vma, orig_pte, address); |
65500d23 HD |
2503 | return VM_FAULT_OOM; |
2504 | } | |
65500d23 HD |
2505 | |
2506 | pgoff = pte_to_pgoff(orig_pte); | |
16abfa08 | 2507 | return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); |
1da177e4 LT |
2508 | } |
2509 | ||
2510 | /* | |
2511 | * These routines also need to handle stuff like marking pages dirty | |
2512 | * and/or accessed for architectures that don't do it in hardware (most | |
2513 | * RISC architectures). The early dirtying is also good on the i386. | |
2514 | * | |
2515 | * There is also a hook called "update_mmu_cache()" that architectures | |
2516 | * with external mmu caches can use to update those (ie the Sparc or | |
2517 | * PowerPC hashed page tables that act as extended TLBs). | |
2518 | * | |
c74df32c HD |
2519 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2520 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2521 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 LT |
2522 | */ |
2523 | static inline int handle_pte_fault(struct mm_struct *mm, | |
65500d23 HD |
2524 | struct vm_area_struct *vma, unsigned long address, |
2525 | pte_t *pte, pmd_t *pmd, int write_access) | |
1da177e4 LT |
2526 | { |
2527 | pte_t entry; | |
8f4e2101 | 2528 | spinlock_t *ptl; |
1da177e4 | 2529 | |
8dab5241 | 2530 | entry = *pte; |
1da177e4 | 2531 | if (!pte_present(entry)) { |
65500d23 | 2532 | if (pte_none(entry)) { |
f4b81804 | 2533 | if (vma->vm_ops) { |
3c18ddd1 | 2534 | if (likely(vma->vm_ops->fault)) |
54cb8821 NP |
2535 | return do_linear_fault(mm, vma, address, |
2536 | pte, pmd, write_access, entry); | |
f4b81804 JS |
2537 | if (unlikely(vma->vm_ops->nopfn)) |
2538 | return do_no_pfn(mm, vma, address, pte, | |
2539 | pmd, write_access); | |
2540 | } | |
2541 | return do_anonymous_page(mm, vma, address, | |
2542 | pte, pmd, write_access); | |
65500d23 | 2543 | } |
1da177e4 | 2544 | if (pte_file(entry)) |
d0217ac0 | 2545 | return do_nonlinear_fault(mm, vma, address, |
65500d23 HD |
2546 | pte, pmd, write_access, entry); |
2547 | return do_swap_page(mm, vma, address, | |
2548 | pte, pmd, write_access, entry); | |
1da177e4 LT |
2549 | } |
2550 | ||
4c21e2f2 | 2551 | ptl = pte_lockptr(mm, pmd); |
8f4e2101 HD |
2552 | spin_lock(ptl); |
2553 | if (unlikely(!pte_same(*pte, entry))) | |
2554 | goto unlock; | |
1da177e4 LT |
2555 | if (write_access) { |
2556 | if (!pte_write(entry)) | |
8f4e2101 HD |
2557 | return do_wp_page(mm, vma, address, |
2558 | pte, pmd, ptl, entry); | |
1da177e4 LT |
2559 | entry = pte_mkdirty(entry); |
2560 | } | |
2561 | entry = pte_mkyoung(entry); | |
8dab5241 | 2562 | if (ptep_set_access_flags(vma, address, pte, entry, write_access)) { |
1a44e149 | 2563 | update_mmu_cache(vma, address, entry); |
1a44e149 AA |
2564 | } else { |
2565 | /* | |
2566 | * This is needed only for protection faults but the arch code | |
2567 | * is not yet telling us if this is a protection fault or not. | |
2568 | * This still avoids useless tlb flushes for .text page faults | |
2569 | * with threads. | |
2570 | */ | |
2571 | if (write_access) | |
2572 | flush_tlb_page(vma, address); | |
2573 | } | |
8f4e2101 HD |
2574 | unlock: |
2575 | pte_unmap_unlock(pte, ptl); | |
83c54070 | 2576 | return 0; |
1da177e4 LT |
2577 | } |
2578 | ||
2579 | /* | |
2580 | * By the time we get here, we already hold the mm semaphore | |
2581 | */ | |
83c54070 | 2582 | int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
1da177e4 LT |
2583 | unsigned long address, int write_access) |
2584 | { | |
2585 | pgd_t *pgd; | |
2586 | pud_t *pud; | |
2587 | pmd_t *pmd; | |
2588 | pte_t *pte; | |
2589 | ||
2590 | __set_current_state(TASK_RUNNING); | |
2591 | ||
f8891e5e | 2592 | count_vm_event(PGFAULT); |
1da177e4 | 2593 | |
ac9b9c66 HD |
2594 | if (unlikely(is_vm_hugetlb_page(vma))) |
2595 | return hugetlb_fault(mm, vma, address, write_access); | |
1da177e4 | 2596 | |
1da177e4 | 2597 | pgd = pgd_offset(mm, address); |
1da177e4 LT |
2598 | pud = pud_alloc(mm, pgd, address); |
2599 | if (!pud) | |
c74df32c | 2600 | return VM_FAULT_OOM; |
1da177e4 LT |
2601 | pmd = pmd_alloc(mm, pud, address); |
2602 | if (!pmd) | |
c74df32c | 2603 | return VM_FAULT_OOM; |
1da177e4 LT |
2604 | pte = pte_alloc_map(mm, pmd, address); |
2605 | if (!pte) | |
c74df32c | 2606 | return VM_FAULT_OOM; |
1da177e4 | 2607 | |
c74df32c | 2608 | return handle_pte_fault(mm, vma, address, pte, pmd, write_access); |
1da177e4 LT |
2609 | } |
2610 | ||
2611 | #ifndef __PAGETABLE_PUD_FOLDED | |
2612 | /* | |
2613 | * Allocate page upper directory. | |
872fec16 | 2614 | * We've already handled the fast-path in-line. |
1da177e4 | 2615 | */ |
1bb3630e | 2616 | int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) |
1da177e4 | 2617 | { |
c74df32c HD |
2618 | pud_t *new = pud_alloc_one(mm, address); |
2619 | if (!new) | |
1bb3630e | 2620 | return -ENOMEM; |
1da177e4 | 2621 | |
872fec16 | 2622 | spin_lock(&mm->page_table_lock); |
1bb3630e | 2623 | if (pgd_present(*pgd)) /* Another has populated it */ |
5e541973 | 2624 | pud_free(mm, new); |
1bb3630e HD |
2625 | else |
2626 | pgd_populate(mm, pgd, new); | |
c74df32c | 2627 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 2628 | return 0; |
1da177e4 LT |
2629 | } |
2630 | #endif /* __PAGETABLE_PUD_FOLDED */ | |
2631 | ||
2632 | #ifndef __PAGETABLE_PMD_FOLDED | |
2633 | /* | |
2634 | * Allocate page middle directory. | |
872fec16 | 2635 | * We've already handled the fast-path in-line. |
1da177e4 | 2636 | */ |
1bb3630e | 2637 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
1da177e4 | 2638 | { |
c74df32c HD |
2639 | pmd_t *new = pmd_alloc_one(mm, address); |
2640 | if (!new) | |
1bb3630e | 2641 | return -ENOMEM; |
1da177e4 | 2642 | |
872fec16 | 2643 | spin_lock(&mm->page_table_lock); |
1da177e4 | 2644 | #ifndef __ARCH_HAS_4LEVEL_HACK |
1bb3630e | 2645 | if (pud_present(*pud)) /* Another has populated it */ |
5e541973 | 2646 | pmd_free(mm, new); |
1bb3630e HD |
2647 | else |
2648 | pud_populate(mm, pud, new); | |
1da177e4 | 2649 | #else |
1bb3630e | 2650 | if (pgd_present(*pud)) /* Another has populated it */ |
5e541973 | 2651 | pmd_free(mm, new); |
1bb3630e HD |
2652 | else |
2653 | pgd_populate(mm, pud, new); | |
1da177e4 | 2654 | #endif /* __ARCH_HAS_4LEVEL_HACK */ |
c74df32c | 2655 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 2656 | return 0; |
e0f39591 | 2657 | } |
1da177e4 LT |
2658 | #endif /* __PAGETABLE_PMD_FOLDED */ |
2659 | ||
2660 | int make_pages_present(unsigned long addr, unsigned long end) | |
2661 | { | |
2662 | int ret, len, write; | |
2663 | struct vm_area_struct * vma; | |
2664 | ||
2665 | vma = find_vma(current->mm, addr); | |
2666 | if (!vma) | |
2667 | return -1; | |
2668 | write = (vma->vm_flags & VM_WRITE) != 0; | |
5bcb28b1 ES |
2669 | BUG_ON(addr >= end); |
2670 | BUG_ON(end > vma->vm_end); | |
68e116a3 | 2671 | len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE; |
1da177e4 LT |
2672 | ret = get_user_pages(current, current->mm, addr, |
2673 | len, write, 0, NULL, NULL); | |
2674 | if (ret < 0) | |
2675 | return ret; | |
2676 | return ret == len ? 0 : -1; | |
2677 | } | |
2678 | ||
1da177e4 LT |
2679 | #if !defined(__HAVE_ARCH_GATE_AREA) |
2680 | ||
2681 | #if defined(AT_SYSINFO_EHDR) | |
5ce7852c | 2682 | static struct vm_area_struct gate_vma; |
1da177e4 LT |
2683 | |
2684 | static int __init gate_vma_init(void) | |
2685 | { | |
2686 | gate_vma.vm_mm = NULL; | |
2687 | gate_vma.vm_start = FIXADDR_USER_START; | |
2688 | gate_vma.vm_end = FIXADDR_USER_END; | |
b6558c4a RM |
2689 | gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC; |
2690 | gate_vma.vm_page_prot = __P101; | |
f47aef55 RM |
2691 | /* |
2692 | * Make sure the vDSO gets into every core dump. | |
2693 | * Dumping its contents makes post-mortem fully interpretable later | |
2694 | * without matching up the same kernel and hardware config to see | |
2695 | * what PC values meant. | |
2696 | */ | |
2697 | gate_vma.vm_flags |= VM_ALWAYSDUMP; | |
1da177e4 LT |
2698 | return 0; |
2699 | } | |
2700 | __initcall(gate_vma_init); | |
2701 | #endif | |
2702 | ||
2703 | struct vm_area_struct *get_gate_vma(struct task_struct *tsk) | |
2704 | { | |
2705 | #ifdef AT_SYSINFO_EHDR | |
2706 | return &gate_vma; | |
2707 | #else | |
2708 | return NULL; | |
2709 | #endif | |
2710 | } | |
2711 | ||
2712 | int in_gate_area_no_task(unsigned long addr) | |
2713 | { | |
2714 | #ifdef AT_SYSINFO_EHDR | |
2715 | if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END)) | |
2716 | return 1; | |
2717 | #endif | |
2718 | return 0; | |
2719 | } | |
2720 | ||
2721 | #endif /* __HAVE_ARCH_GATE_AREA */ | |
0ec76a11 DH |
2722 | |
2723 | /* | |
2724 | * Access another process' address space. | |
2725 | * Source/target buffer must be kernel space, | |
2726 | * Do not walk the page table directly, use get_user_pages | |
2727 | */ | |
2728 | int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) | |
2729 | { | |
2730 | struct mm_struct *mm; | |
2731 | struct vm_area_struct *vma; | |
2732 | struct page *page; | |
2733 | void *old_buf = buf; | |
2734 | ||
2735 | mm = get_task_mm(tsk); | |
2736 | if (!mm) | |
2737 | return 0; | |
2738 | ||
2739 | down_read(&mm->mmap_sem); | |
183ff22b | 2740 | /* ignore errors, just check how much was successfully transferred */ |
0ec76a11 DH |
2741 | while (len) { |
2742 | int bytes, ret, offset; | |
2743 | void *maddr; | |
2744 | ||
2745 | ret = get_user_pages(tsk, mm, addr, 1, | |
2746 | write, 1, &page, &vma); | |
2747 | if (ret <= 0) | |
2748 | break; | |
2749 | ||
2750 | bytes = len; | |
2751 | offset = addr & (PAGE_SIZE-1); | |
2752 | if (bytes > PAGE_SIZE-offset) | |
2753 | bytes = PAGE_SIZE-offset; | |
2754 | ||
2755 | maddr = kmap(page); | |
2756 | if (write) { | |
2757 | copy_to_user_page(vma, page, addr, | |
2758 | maddr + offset, buf, bytes); | |
2759 | set_page_dirty_lock(page); | |
2760 | } else { | |
2761 | copy_from_user_page(vma, page, addr, | |
2762 | buf, maddr + offset, bytes); | |
2763 | } | |
2764 | kunmap(page); | |
2765 | page_cache_release(page); | |
2766 | len -= bytes; | |
2767 | buf += bytes; | |
2768 | addr += bytes; | |
2769 | } | |
2770 | up_read(&mm->mmap_sem); | |
2771 | mmput(mm); | |
2772 | ||
2773 | return buf - old_buf; | |
2774 | } | |
03252919 AK |
2775 | |
2776 | /* | |
2777 | * Print the name of a VMA. | |
2778 | */ | |
2779 | void print_vma_addr(char *prefix, unsigned long ip) | |
2780 | { | |
2781 | struct mm_struct *mm = current->mm; | |
2782 | struct vm_area_struct *vma; | |
2783 | ||
e8bff74a IM |
2784 | /* |
2785 | * Do not print if we are in atomic | |
2786 | * contexts (in exception stacks, etc.): | |
2787 | */ | |
2788 | if (preempt_count()) | |
2789 | return; | |
2790 | ||
03252919 AK |
2791 | down_read(&mm->mmap_sem); |
2792 | vma = find_vma(mm, ip); | |
2793 | if (vma && vma->vm_file) { | |
2794 | struct file *f = vma->vm_file; | |
2795 | char *buf = (char *)__get_free_page(GFP_KERNEL); | |
2796 | if (buf) { | |
2797 | char *p, *s; | |
2798 | ||
cf28b486 | 2799 | p = d_path(&f->f_path, buf, PAGE_SIZE); |
03252919 AK |
2800 | if (IS_ERR(p)) |
2801 | p = "?"; | |
2802 | s = strrchr(p, '/'); | |
2803 | if (s) | |
2804 | p = s+1; | |
2805 | printk("%s%s[%lx+%lx]", prefix, p, | |
2806 | vma->vm_start, | |
2807 | vma->vm_end - vma->vm_start); | |
2808 | free_page((unsigned long)buf); | |
2809 | } | |
2810 | } | |
2811 | up_read(¤t->mm->mmap_sem); | |
2812 | } |