]>
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> | |
9a840895 | 48 | #include <linux/ksm.h> |
1da177e4 LT |
49 | #include <linux/rmap.h> |
50 | #include <linux/module.h> | |
0ff92245 | 51 | #include <linux/delayacct.h> |
1da177e4 | 52 | #include <linux/init.h> |
edc79b2a | 53 | #include <linux/writeback.h> |
8a9f3ccd | 54 | #include <linux/memcontrol.h> |
cddb8a5c | 55 | #include <linux/mmu_notifier.h> |
3dc14741 HD |
56 | #include <linux/kallsyms.h> |
57 | #include <linux/swapops.h> | |
58 | #include <linux/elf.h> | |
1da177e4 | 59 | |
6952b61d | 60 | #include <asm/io.h> |
1da177e4 LT |
61 | #include <asm/pgalloc.h> |
62 | #include <asm/uaccess.h> | |
63 | #include <asm/tlb.h> | |
64 | #include <asm/tlbflush.h> | |
65 | #include <asm/pgtable.h> | |
66 | ||
42b77728 JB |
67 | #include "internal.h" |
68 | ||
d41dee36 | 69 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
70 | /* use the per-pgdat data instead for discontigmem - mbligh */ |
71 | unsigned long max_mapnr; | |
72 | struct page *mem_map; | |
73 | ||
74 | EXPORT_SYMBOL(max_mapnr); | |
75 | EXPORT_SYMBOL(mem_map); | |
76 | #endif | |
77 | ||
78 | unsigned long num_physpages; | |
79 | /* | |
80 | * A number of key systems in x86 including ioremap() rely on the assumption | |
81 | * that high_memory defines the upper bound on direct map memory, then end | |
82 | * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and | |
83 | * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL | |
84 | * and ZONE_HIGHMEM. | |
85 | */ | |
86 | void * high_memory; | |
1da177e4 LT |
87 | |
88 | EXPORT_SYMBOL(num_physpages); | |
89 | EXPORT_SYMBOL(high_memory); | |
1da177e4 | 90 | |
32a93233 IM |
91 | /* |
92 | * Randomize the address space (stacks, mmaps, brk, etc.). | |
93 | * | |
94 | * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, | |
95 | * as ancient (libc5 based) binaries can segfault. ) | |
96 | */ | |
97 | int randomize_va_space __read_mostly = | |
98 | #ifdef CONFIG_COMPAT_BRK | |
99 | 1; | |
100 | #else | |
101 | 2; | |
102 | #endif | |
a62eaf15 AK |
103 | |
104 | static int __init disable_randmaps(char *s) | |
105 | { | |
106 | randomize_va_space = 0; | |
9b41046c | 107 | return 1; |
a62eaf15 AK |
108 | } |
109 | __setup("norandmaps", disable_randmaps); | |
110 | ||
62eede62 | 111 | unsigned long zero_pfn __read_mostly; |
03f6462a | 112 | unsigned long highest_memmap_pfn __read_mostly; |
a13ea5b7 HD |
113 | |
114 | /* | |
115 | * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() | |
116 | */ | |
117 | static int __init init_zero_pfn(void) | |
118 | { | |
119 | zero_pfn = page_to_pfn(ZERO_PAGE(0)); | |
120 | return 0; | |
121 | } | |
122 | core_initcall(init_zero_pfn); | |
a62eaf15 | 123 | |
1da177e4 LT |
124 | /* |
125 | * If a p?d_bad entry is found while walking page tables, report | |
126 | * the error, before resetting entry to p?d_none. Usually (but | |
127 | * very seldom) called out from the p?d_none_or_clear_bad macros. | |
128 | */ | |
129 | ||
130 | void pgd_clear_bad(pgd_t *pgd) | |
131 | { | |
132 | pgd_ERROR(*pgd); | |
133 | pgd_clear(pgd); | |
134 | } | |
135 | ||
136 | void pud_clear_bad(pud_t *pud) | |
137 | { | |
138 | pud_ERROR(*pud); | |
139 | pud_clear(pud); | |
140 | } | |
141 | ||
142 | void pmd_clear_bad(pmd_t *pmd) | |
143 | { | |
144 | pmd_ERROR(*pmd); | |
145 | pmd_clear(pmd); | |
146 | } | |
147 | ||
148 | /* | |
149 | * Note: this doesn't free the actual pages themselves. That | |
150 | * has been handled earlier when unmapping all the memory regions. | |
151 | */ | |
9e1b32ca BH |
152 | static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, |
153 | unsigned long addr) | |
1da177e4 | 154 | { |
2f569afd | 155 | pgtable_t token = pmd_pgtable(*pmd); |
e0da382c | 156 | pmd_clear(pmd); |
9e1b32ca | 157 | pte_free_tlb(tlb, token, addr); |
e0da382c | 158 | tlb->mm->nr_ptes--; |
1da177e4 LT |
159 | } |
160 | ||
e0da382c HD |
161 | static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
162 | unsigned long addr, unsigned long end, | |
163 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
164 | { |
165 | pmd_t *pmd; | |
166 | unsigned long next; | |
e0da382c | 167 | unsigned long start; |
1da177e4 | 168 | |
e0da382c | 169 | start = addr; |
1da177e4 | 170 | pmd = pmd_offset(pud, addr); |
1da177e4 LT |
171 | do { |
172 | next = pmd_addr_end(addr, end); | |
173 | if (pmd_none_or_clear_bad(pmd)) | |
174 | continue; | |
9e1b32ca | 175 | free_pte_range(tlb, pmd, addr); |
1da177e4 LT |
176 | } while (pmd++, addr = next, addr != end); |
177 | ||
e0da382c HD |
178 | start &= PUD_MASK; |
179 | if (start < floor) | |
180 | return; | |
181 | if (ceiling) { | |
182 | ceiling &= PUD_MASK; | |
183 | if (!ceiling) | |
184 | return; | |
1da177e4 | 185 | } |
e0da382c HD |
186 | if (end - 1 > ceiling - 1) |
187 | return; | |
188 | ||
189 | pmd = pmd_offset(pud, start); | |
190 | pud_clear(pud); | |
9e1b32ca | 191 | pmd_free_tlb(tlb, pmd, start); |
1da177e4 LT |
192 | } |
193 | ||
e0da382c HD |
194 | static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, |
195 | unsigned long addr, unsigned long end, | |
196 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
197 | { |
198 | pud_t *pud; | |
199 | unsigned long next; | |
e0da382c | 200 | unsigned long start; |
1da177e4 | 201 | |
e0da382c | 202 | start = addr; |
1da177e4 | 203 | pud = pud_offset(pgd, addr); |
1da177e4 LT |
204 | do { |
205 | next = pud_addr_end(addr, end); | |
206 | if (pud_none_or_clear_bad(pud)) | |
207 | continue; | |
e0da382c | 208 | free_pmd_range(tlb, pud, addr, next, floor, ceiling); |
1da177e4 LT |
209 | } while (pud++, addr = next, addr != end); |
210 | ||
e0da382c HD |
211 | start &= PGDIR_MASK; |
212 | if (start < floor) | |
213 | return; | |
214 | if (ceiling) { | |
215 | ceiling &= PGDIR_MASK; | |
216 | if (!ceiling) | |
217 | return; | |
1da177e4 | 218 | } |
e0da382c HD |
219 | if (end - 1 > ceiling - 1) |
220 | return; | |
221 | ||
222 | pud = pud_offset(pgd, start); | |
223 | pgd_clear(pgd); | |
9e1b32ca | 224 | pud_free_tlb(tlb, pud, start); |
1da177e4 LT |
225 | } |
226 | ||
227 | /* | |
e0da382c HD |
228 | * This function frees user-level page tables of a process. |
229 | * | |
1da177e4 LT |
230 | * Must be called with pagetable lock held. |
231 | */ | |
42b77728 | 232 | void free_pgd_range(struct mmu_gather *tlb, |
e0da382c HD |
233 | unsigned long addr, unsigned long end, |
234 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
235 | { |
236 | pgd_t *pgd; | |
237 | unsigned long next; | |
e0da382c HD |
238 | unsigned long start; |
239 | ||
240 | /* | |
241 | * The next few lines have given us lots of grief... | |
242 | * | |
243 | * Why are we testing PMD* at this top level? Because often | |
244 | * there will be no work to do at all, and we'd prefer not to | |
245 | * go all the way down to the bottom just to discover that. | |
246 | * | |
247 | * Why all these "- 1"s? Because 0 represents both the bottom | |
248 | * of the address space and the top of it (using -1 for the | |
249 | * top wouldn't help much: the masks would do the wrong thing). | |
250 | * The rule is that addr 0 and floor 0 refer to the bottom of | |
251 | * the address space, but end 0 and ceiling 0 refer to the top | |
252 | * Comparisons need to use "end - 1" and "ceiling - 1" (though | |
253 | * that end 0 case should be mythical). | |
254 | * | |
255 | * Wherever addr is brought up or ceiling brought down, we must | |
256 | * be careful to reject "the opposite 0" before it confuses the | |
257 | * subsequent tests. But what about where end is brought down | |
258 | * by PMD_SIZE below? no, end can't go down to 0 there. | |
259 | * | |
260 | * Whereas we round start (addr) and ceiling down, by different | |
261 | * masks at different levels, in order to test whether a table | |
262 | * now has no other vmas using it, so can be freed, we don't | |
263 | * bother to round floor or end up - the tests don't need that. | |
264 | */ | |
1da177e4 | 265 | |
e0da382c HD |
266 | addr &= PMD_MASK; |
267 | if (addr < floor) { | |
268 | addr += PMD_SIZE; | |
269 | if (!addr) | |
270 | return; | |
271 | } | |
272 | if (ceiling) { | |
273 | ceiling &= PMD_MASK; | |
274 | if (!ceiling) | |
275 | return; | |
276 | } | |
277 | if (end - 1 > ceiling - 1) | |
278 | end -= PMD_SIZE; | |
279 | if (addr > end - 1) | |
280 | return; | |
281 | ||
282 | start = addr; | |
42b77728 | 283 | pgd = pgd_offset(tlb->mm, addr); |
1da177e4 LT |
284 | do { |
285 | next = pgd_addr_end(addr, end); | |
286 | if (pgd_none_or_clear_bad(pgd)) | |
287 | continue; | |
42b77728 | 288 | free_pud_range(tlb, pgd, addr, next, floor, ceiling); |
1da177e4 | 289 | } while (pgd++, addr = next, addr != end); |
e0da382c HD |
290 | } |
291 | ||
42b77728 | 292 | void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, |
3bf5ee95 | 293 | unsigned long floor, unsigned long ceiling) |
e0da382c HD |
294 | { |
295 | while (vma) { | |
296 | struct vm_area_struct *next = vma->vm_next; | |
297 | unsigned long addr = vma->vm_start; | |
298 | ||
8f4f8c16 | 299 | /* |
25d9e2d1 NP |
300 | * Hide vma from rmap and truncate_pagecache before freeing |
301 | * pgtables | |
8f4f8c16 HD |
302 | */ |
303 | anon_vma_unlink(vma); | |
304 | unlink_file_vma(vma); | |
305 | ||
9da61aef | 306 | if (is_vm_hugetlb_page(vma)) { |
3bf5ee95 | 307 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
e0da382c | 308 | floor, next? next->vm_start: ceiling); |
3bf5ee95 HD |
309 | } else { |
310 | /* | |
311 | * Optimization: gather nearby vmas into one call down | |
312 | */ | |
313 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE | |
4866920b | 314 | && !is_vm_hugetlb_page(next)) { |
3bf5ee95 HD |
315 | vma = next; |
316 | next = vma->vm_next; | |
8f4f8c16 HD |
317 | anon_vma_unlink(vma); |
318 | unlink_file_vma(vma); | |
3bf5ee95 HD |
319 | } |
320 | free_pgd_range(tlb, addr, vma->vm_end, | |
321 | floor, next? next->vm_start: ceiling); | |
322 | } | |
e0da382c HD |
323 | vma = next; |
324 | } | |
1da177e4 LT |
325 | } |
326 | ||
1bb3630e | 327 | int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address) |
1da177e4 | 328 | { |
2f569afd | 329 | pgtable_t new = pte_alloc_one(mm, address); |
1bb3630e HD |
330 | if (!new) |
331 | return -ENOMEM; | |
332 | ||
362a61ad NP |
333 | /* |
334 | * Ensure all pte setup (eg. pte page lock and page clearing) are | |
335 | * visible before the pte is made visible to other CPUs by being | |
336 | * put into page tables. | |
337 | * | |
338 | * The other side of the story is the pointer chasing in the page | |
339 | * table walking code (when walking the page table without locking; | |
340 | * ie. most of the time). Fortunately, these data accesses consist | |
341 | * of a chain of data-dependent loads, meaning most CPUs (alpha | |
342 | * being the notable exception) will already guarantee loads are | |
343 | * seen in-order. See the alpha page table accessors for the | |
344 | * smp_read_barrier_depends() barriers in page table walking code. | |
345 | */ | |
346 | smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ | |
347 | ||
c74df32c | 348 | spin_lock(&mm->page_table_lock); |
2f569afd | 349 | if (!pmd_present(*pmd)) { /* Has another populated it ? */ |
1da177e4 | 350 | mm->nr_ptes++; |
1da177e4 | 351 | pmd_populate(mm, pmd, new); |
2f569afd | 352 | new = NULL; |
1da177e4 | 353 | } |
c74df32c | 354 | spin_unlock(&mm->page_table_lock); |
2f569afd MS |
355 | if (new) |
356 | pte_free(mm, new); | |
1bb3630e | 357 | return 0; |
1da177e4 LT |
358 | } |
359 | ||
1bb3630e | 360 | int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) |
1da177e4 | 361 | { |
1bb3630e HD |
362 | pte_t *new = pte_alloc_one_kernel(&init_mm, address); |
363 | if (!new) | |
364 | return -ENOMEM; | |
365 | ||
362a61ad NP |
366 | smp_wmb(); /* See comment in __pte_alloc */ |
367 | ||
1bb3630e | 368 | spin_lock(&init_mm.page_table_lock); |
2f569afd | 369 | if (!pmd_present(*pmd)) { /* Has another populated it ? */ |
1bb3630e | 370 | pmd_populate_kernel(&init_mm, pmd, new); |
2f569afd MS |
371 | new = NULL; |
372 | } | |
1bb3630e | 373 | spin_unlock(&init_mm.page_table_lock); |
2f569afd MS |
374 | if (new) |
375 | pte_free_kernel(&init_mm, new); | |
1bb3630e | 376 | return 0; |
1da177e4 LT |
377 | } |
378 | ||
ae859762 HD |
379 | static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss) |
380 | { | |
381 | if (file_rss) | |
382 | add_mm_counter(mm, file_rss, file_rss); | |
383 | if (anon_rss) | |
384 | add_mm_counter(mm, anon_rss, anon_rss); | |
385 | } | |
386 | ||
b5810039 | 387 | /* |
6aab341e LT |
388 | * This function is called to print an error when a bad pte |
389 | * is found. For example, we might have a PFN-mapped pte in | |
390 | * a region that doesn't allow it. | |
b5810039 NP |
391 | * |
392 | * The calling function must still handle the error. | |
393 | */ | |
3dc14741 HD |
394 | static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, |
395 | pte_t pte, struct page *page) | |
b5810039 | 396 | { |
3dc14741 HD |
397 | pgd_t *pgd = pgd_offset(vma->vm_mm, addr); |
398 | pud_t *pud = pud_offset(pgd, addr); | |
399 | pmd_t *pmd = pmd_offset(pud, addr); | |
400 | struct address_space *mapping; | |
401 | pgoff_t index; | |
d936cf9b HD |
402 | static unsigned long resume; |
403 | static unsigned long nr_shown; | |
404 | static unsigned long nr_unshown; | |
405 | ||
406 | /* | |
407 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
408 | * or allow a steady drip of one report per second. | |
409 | */ | |
410 | if (nr_shown == 60) { | |
411 | if (time_before(jiffies, resume)) { | |
412 | nr_unshown++; | |
413 | return; | |
414 | } | |
415 | if (nr_unshown) { | |
1e9e6365 HD |
416 | printk(KERN_ALERT |
417 | "BUG: Bad page map: %lu messages suppressed\n", | |
d936cf9b HD |
418 | nr_unshown); |
419 | nr_unshown = 0; | |
420 | } | |
421 | nr_shown = 0; | |
422 | } | |
423 | if (nr_shown++ == 0) | |
424 | resume = jiffies + 60 * HZ; | |
3dc14741 HD |
425 | |
426 | mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; | |
427 | index = linear_page_index(vma, addr); | |
428 | ||
1e9e6365 HD |
429 | printk(KERN_ALERT |
430 | "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", | |
3dc14741 HD |
431 | current->comm, |
432 | (long long)pte_val(pte), (long long)pmd_val(*pmd)); | |
433 | if (page) { | |
1e9e6365 | 434 | printk(KERN_ALERT |
3dc14741 HD |
435 | "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n", |
436 | page, (void *)page->flags, page_count(page), | |
437 | page_mapcount(page), page->mapping, page->index); | |
438 | } | |
1e9e6365 | 439 | printk(KERN_ALERT |
3dc14741 HD |
440 | "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n", |
441 | (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); | |
442 | /* | |
443 | * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y | |
444 | */ | |
445 | if (vma->vm_ops) | |
1e9e6365 | 446 | print_symbol(KERN_ALERT "vma->vm_ops->fault: %s\n", |
3dc14741 HD |
447 | (unsigned long)vma->vm_ops->fault); |
448 | if (vma->vm_file && vma->vm_file->f_op) | |
1e9e6365 | 449 | print_symbol(KERN_ALERT "vma->vm_file->f_op->mmap: %s\n", |
3dc14741 | 450 | (unsigned long)vma->vm_file->f_op->mmap); |
b5810039 | 451 | dump_stack(); |
3dc14741 | 452 | add_taint(TAINT_BAD_PAGE); |
b5810039 NP |
453 | } |
454 | ||
67121172 LT |
455 | static inline int is_cow_mapping(unsigned int flags) |
456 | { | |
457 | return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; | |
458 | } | |
459 | ||
62eede62 HD |
460 | #ifndef is_zero_pfn |
461 | static inline int is_zero_pfn(unsigned long pfn) | |
462 | { | |
463 | return pfn == zero_pfn; | |
464 | } | |
465 | #endif | |
466 | ||
467 | #ifndef my_zero_pfn | |
468 | static inline unsigned long my_zero_pfn(unsigned long addr) | |
469 | { | |
470 | return zero_pfn; | |
471 | } | |
472 | #endif | |
473 | ||
ee498ed7 | 474 | /* |
7e675137 | 475 | * vm_normal_page -- This function gets the "struct page" associated with a pte. |
6aab341e | 476 | * |
7e675137 NP |
477 | * "Special" mappings do not wish to be associated with a "struct page" (either |
478 | * it doesn't exist, or it exists but they don't want to touch it). In this | |
479 | * case, NULL is returned here. "Normal" mappings do have a struct page. | |
b379d790 | 480 | * |
7e675137 NP |
481 | * There are 2 broad cases. Firstly, an architecture may define a pte_special() |
482 | * pte bit, in which case this function is trivial. Secondly, an architecture | |
483 | * may not have a spare pte bit, which requires a more complicated scheme, | |
484 | * described below. | |
485 | * | |
486 | * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a | |
487 | * special mapping (even if there are underlying and valid "struct pages"). | |
488 | * COWed pages of a VM_PFNMAP are always normal. | |
6aab341e | 489 | * |
b379d790 JH |
490 | * The way we recognize COWed pages within VM_PFNMAP mappings is through the |
491 | * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit | |
7e675137 NP |
492 | * set, and the vm_pgoff will point to the first PFN mapped: thus every special |
493 | * mapping will always honor the rule | |
6aab341e LT |
494 | * |
495 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) | |
496 | * | |
7e675137 NP |
497 | * And for normal mappings this is false. |
498 | * | |
499 | * This restricts such mappings to be a linear translation from virtual address | |
500 | * to pfn. To get around this restriction, we allow arbitrary mappings so long | |
501 | * as the vma is not a COW mapping; in that case, we know that all ptes are | |
502 | * special (because none can have been COWed). | |
b379d790 | 503 | * |
b379d790 | 504 | * |
7e675137 | 505 | * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. |
b379d790 JH |
506 | * |
507 | * VM_MIXEDMAP mappings can likewise contain memory with or without "struct | |
508 | * page" backing, however the difference is that _all_ pages with a struct | |
509 | * page (that is, those where pfn_valid is true) are refcounted and considered | |
510 | * normal pages by the VM. The disadvantage is that pages are refcounted | |
511 | * (which can be slower and simply not an option for some PFNMAP users). The | |
512 | * advantage is that we don't have to follow the strict linearity rule of | |
513 | * PFNMAP mappings in order to support COWable mappings. | |
514 | * | |
ee498ed7 | 515 | */ |
7e675137 NP |
516 | #ifdef __HAVE_ARCH_PTE_SPECIAL |
517 | # define HAVE_PTE_SPECIAL 1 | |
518 | #else | |
519 | # define HAVE_PTE_SPECIAL 0 | |
520 | #endif | |
521 | struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, | |
522 | pte_t pte) | |
ee498ed7 | 523 | { |
22b31eec | 524 | unsigned long pfn = pte_pfn(pte); |
7e675137 NP |
525 | |
526 | if (HAVE_PTE_SPECIAL) { | |
22b31eec HD |
527 | if (likely(!pte_special(pte))) |
528 | goto check_pfn; | |
a13ea5b7 HD |
529 | if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) |
530 | return NULL; | |
62eede62 | 531 | if (!is_zero_pfn(pfn)) |
22b31eec | 532 | print_bad_pte(vma, addr, pte, NULL); |
7e675137 NP |
533 | return NULL; |
534 | } | |
535 | ||
536 | /* !HAVE_PTE_SPECIAL case follows: */ | |
537 | ||
b379d790 JH |
538 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
539 | if (vma->vm_flags & VM_MIXEDMAP) { | |
540 | if (!pfn_valid(pfn)) | |
541 | return NULL; | |
542 | goto out; | |
543 | } else { | |
7e675137 NP |
544 | unsigned long off; |
545 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
b379d790 JH |
546 | if (pfn == vma->vm_pgoff + off) |
547 | return NULL; | |
548 | if (!is_cow_mapping(vma->vm_flags)) | |
549 | return NULL; | |
550 | } | |
6aab341e LT |
551 | } |
552 | ||
62eede62 HD |
553 | if (is_zero_pfn(pfn)) |
554 | return NULL; | |
22b31eec HD |
555 | check_pfn: |
556 | if (unlikely(pfn > highest_memmap_pfn)) { | |
557 | print_bad_pte(vma, addr, pte, NULL); | |
558 | return NULL; | |
559 | } | |
6aab341e LT |
560 | |
561 | /* | |
7e675137 | 562 | * NOTE! We still have PageReserved() pages in the page tables. |
7e675137 | 563 | * eg. VDSO mappings can cause them to exist. |
6aab341e | 564 | */ |
b379d790 | 565 | out: |
6aab341e | 566 | return pfn_to_page(pfn); |
ee498ed7 HD |
567 | } |
568 | ||
1da177e4 LT |
569 | /* |
570 | * copy one vm_area from one task to the other. Assumes the page tables | |
571 | * already present in the new task to be cleared in the whole range | |
572 | * covered by this vma. | |
1da177e4 LT |
573 | */ |
574 | ||
8c103762 | 575 | static inline void |
1da177e4 | 576 | copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
b5810039 | 577 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, |
8c103762 | 578 | unsigned long addr, int *rss) |
1da177e4 | 579 | { |
b5810039 | 580 | unsigned long vm_flags = vma->vm_flags; |
1da177e4 LT |
581 | pte_t pte = *src_pte; |
582 | struct page *page; | |
1da177e4 LT |
583 | |
584 | /* pte contains position in swap or file, so copy. */ | |
585 | if (unlikely(!pte_present(pte))) { | |
586 | if (!pte_file(pte)) { | |
0697212a CL |
587 | swp_entry_t entry = pte_to_swp_entry(pte); |
588 | ||
589 | swap_duplicate(entry); | |
1da177e4 LT |
590 | /* make sure dst_mm is on swapoff's mmlist. */ |
591 | if (unlikely(list_empty(&dst_mm->mmlist))) { | |
592 | spin_lock(&mmlist_lock); | |
f412ac08 HD |
593 | if (list_empty(&dst_mm->mmlist)) |
594 | list_add(&dst_mm->mmlist, | |
595 | &src_mm->mmlist); | |
1da177e4 LT |
596 | spin_unlock(&mmlist_lock); |
597 | } | |
0697212a CL |
598 | if (is_write_migration_entry(entry) && |
599 | is_cow_mapping(vm_flags)) { | |
600 | /* | |
601 | * COW mappings require pages in both parent | |
602 | * and child to be set to read. | |
603 | */ | |
604 | make_migration_entry_read(&entry); | |
605 | pte = swp_entry_to_pte(entry); | |
606 | set_pte_at(src_mm, addr, src_pte, pte); | |
607 | } | |
1da177e4 | 608 | } |
ae859762 | 609 | goto out_set_pte; |
1da177e4 LT |
610 | } |
611 | ||
1da177e4 LT |
612 | /* |
613 | * If it's a COW mapping, write protect it both | |
614 | * in the parent and the child | |
615 | */ | |
67121172 | 616 | if (is_cow_mapping(vm_flags)) { |
1da177e4 | 617 | ptep_set_wrprotect(src_mm, addr, src_pte); |
3dc90795 | 618 | pte = pte_wrprotect(pte); |
1da177e4 LT |
619 | } |
620 | ||
621 | /* | |
622 | * If it's a shared mapping, mark it clean in | |
623 | * the child | |
624 | */ | |
625 | if (vm_flags & VM_SHARED) | |
626 | pte = pte_mkclean(pte); | |
627 | pte = pte_mkold(pte); | |
6aab341e LT |
628 | |
629 | page = vm_normal_page(vma, addr, pte); | |
630 | if (page) { | |
631 | get_page(page); | |
21333b2b | 632 | page_dup_rmap(page); |
b7c46d15 | 633 | rss[PageAnon(page)]++; |
6aab341e | 634 | } |
ae859762 HD |
635 | |
636 | out_set_pte: | |
637 | set_pte_at(dst_mm, addr, dst_pte, pte); | |
1da177e4 LT |
638 | } |
639 | ||
640 | static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
641 | pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, | |
642 | unsigned long addr, unsigned long end) | |
643 | { | |
c36987e2 | 644 | pte_t *orig_src_pte, *orig_dst_pte; |
1da177e4 | 645 | pte_t *src_pte, *dst_pte; |
c74df32c | 646 | spinlock_t *src_ptl, *dst_ptl; |
e040f218 | 647 | int progress = 0; |
8c103762 | 648 | int rss[2]; |
1da177e4 LT |
649 | |
650 | again: | |
ae859762 | 651 | rss[1] = rss[0] = 0; |
c74df32c | 652 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
1da177e4 LT |
653 | if (!dst_pte) |
654 | return -ENOMEM; | |
655 | src_pte = pte_offset_map_nested(src_pmd, addr); | |
4c21e2f2 | 656 | src_ptl = pte_lockptr(src_mm, src_pmd); |
f20dc5f7 | 657 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
c36987e2 DN |
658 | orig_src_pte = src_pte; |
659 | orig_dst_pte = dst_pte; | |
6606c3e0 | 660 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 661 | |
1da177e4 LT |
662 | do { |
663 | /* | |
664 | * We are holding two locks at this point - either of them | |
665 | * could generate latencies in another task on another CPU. | |
666 | */ | |
e040f218 HD |
667 | if (progress >= 32) { |
668 | progress = 0; | |
669 | if (need_resched() || | |
95c354fe | 670 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
e040f218 HD |
671 | break; |
672 | } | |
1da177e4 LT |
673 | if (pte_none(*src_pte)) { |
674 | progress++; | |
675 | continue; | |
676 | } | |
8c103762 | 677 | copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss); |
1da177e4 LT |
678 | progress += 8; |
679 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); | |
1da177e4 | 680 | |
6606c3e0 | 681 | arch_leave_lazy_mmu_mode(); |
c74df32c | 682 | spin_unlock(src_ptl); |
c36987e2 | 683 | pte_unmap_nested(orig_src_pte); |
ae859762 | 684 | add_mm_rss(dst_mm, rss[0], rss[1]); |
c36987e2 | 685 | pte_unmap_unlock(orig_dst_pte, dst_ptl); |
c74df32c | 686 | cond_resched(); |
1da177e4 LT |
687 | if (addr != end) |
688 | goto again; | |
689 | return 0; | |
690 | } | |
691 | ||
692 | static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
693 | pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, | |
694 | unsigned long addr, unsigned long end) | |
695 | { | |
696 | pmd_t *src_pmd, *dst_pmd; | |
697 | unsigned long next; | |
698 | ||
699 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); | |
700 | if (!dst_pmd) | |
701 | return -ENOMEM; | |
702 | src_pmd = pmd_offset(src_pud, addr); | |
703 | do { | |
704 | next = pmd_addr_end(addr, end); | |
705 | if (pmd_none_or_clear_bad(src_pmd)) | |
706 | continue; | |
707 | if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, | |
708 | vma, addr, next)) | |
709 | return -ENOMEM; | |
710 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); | |
711 | return 0; | |
712 | } | |
713 | ||
714 | static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
715 | pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, | |
716 | unsigned long addr, unsigned long end) | |
717 | { | |
718 | pud_t *src_pud, *dst_pud; | |
719 | unsigned long next; | |
720 | ||
721 | dst_pud = pud_alloc(dst_mm, dst_pgd, addr); | |
722 | if (!dst_pud) | |
723 | return -ENOMEM; | |
724 | src_pud = pud_offset(src_pgd, addr); | |
725 | do { | |
726 | next = pud_addr_end(addr, end); | |
727 | if (pud_none_or_clear_bad(src_pud)) | |
728 | continue; | |
729 | if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, | |
730 | vma, addr, next)) | |
731 | return -ENOMEM; | |
732 | } while (dst_pud++, src_pud++, addr = next, addr != end); | |
733 | return 0; | |
734 | } | |
735 | ||
736 | int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
737 | struct vm_area_struct *vma) | |
738 | { | |
739 | pgd_t *src_pgd, *dst_pgd; | |
740 | unsigned long next; | |
741 | unsigned long addr = vma->vm_start; | |
742 | unsigned long end = vma->vm_end; | |
cddb8a5c | 743 | int ret; |
1da177e4 | 744 | |
d992895b NP |
745 | /* |
746 | * Don't copy ptes where a page fault will fill them correctly. | |
747 | * Fork becomes much lighter when there are big shared or private | |
748 | * readonly mappings. The tradeoff is that copy_page_range is more | |
749 | * efficient than faulting. | |
750 | */ | |
4d7672b4 | 751 | if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) { |
d992895b NP |
752 | if (!vma->anon_vma) |
753 | return 0; | |
754 | } | |
755 | ||
1da177e4 LT |
756 | if (is_vm_hugetlb_page(vma)) |
757 | return copy_hugetlb_page_range(dst_mm, src_mm, vma); | |
758 | ||
34801ba9 | 759 | if (unlikely(is_pfn_mapping(vma))) { |
2ab64037 | 760 | /* |
761 | * We do not free on error cases below as remove_vma | |
762 | * gets called on error from higher level routine | |
763 | */ | |
764 | ret = track_pfn_vma_copy(vma); | |
765 | if (ret) | |
766 | return ret; | |
767 | } | |
768 | ||
cddb8a5c AA |
769 | /* |
770 | * We need to invalidate the secondary MMU mappings only when | |
771 | * there could be a permission downgrade on the ptes of the | |
772 | * parent mm. And a permission downgrade will only happen if | |
773 | * is_cow_mapping() returns true. | |
774 | */ | |
775 | if (is_cow_mapping(vma->vm_flags)) | |
776 | mmu_notifier_invalidate_range_start(src_mm, addr, end); | |
777 | ||
778 | ret = 0; | |
1da177e4 LT |
779 | dst_pgd = pgd_offset(dst_mm, addr); |
780 | src_pgd = pgd_offset(src_mm, addr); | |
781 | do { | |
782 | next = pgd_addr_end(addr, end); | |
783 | if (pgd_none_or_clear_bad(src_pgd)) | |
784 | continue; | |
cddb8a5c AA |
785 | if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, |
786 | vma, addr, next))) { | |
787 | ret = -ENOMEM; | |
788 | break; | |
789 | } | |
1da177e4 | 790 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); |
cddb8a5c AA |
791 | |
792 | if (is_cow_mapping(vma->vm_flags)) | |
793 | mmu_notifier_invalidate_range_end(src_mm, | |
794 | vma->vm_start, end); | |
795 | return ret; | |
1da177e4 LT |
796 | } |
797 | ||
51c6f666 | 798 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
b5810039 | 799 | struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 | 800 | unsigned long addr, unsigned long end, |
51c6f666 | 801 | long *zap_work, struct zap_details *details) |
1da177e4 | 802 | { |
b5810039 | 803 | struct mm_struct *mm = tlb->mm; |
1da177e4 | 804 | pte_t *pte; |
508034a3 | 805 | spinlock_t *ptl; |
ae859762 HD |
806 | int file_rss = 0; |
807 | int anon_rss = 0; | |
1da177e4 | 808 | |
508034a3 | 809 | pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
6606c3e0 | 810 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
811 | do { |
812 | pte_t ptent = *pte; | |
51c6f666 RH |
813 | if (pte_none(ptent)) { |
814 | (*zap_work)--; | |
1da177e4 | 815 | continue; |
51c6f666 | 816 | } |
6f5e6b9e HD |
817 | |
818 | (*zap_work) -= PAGE_SIZE; | |
819 | ||
1da177e4 | 820 | if (pte_present(ptent)) { |
ee498ed7 | 821 | struct page *page; |
51c6f666 | 822 | |
6aab341e | 823 | page = vm_normal_page(vma, addr, ptent); |
1da177e4 LT |
824 | if (unlikely(details) && page) { |
825 | /* | |
826 | * unmap_shared_mapping_pages() wants to | |
827 | * invalidate cache without truncating: | |
828 | * unmap shared but keep private pages. | |
829 | */ | |
830 | if (details->check_mapping && | |
831 | details->check_mapping != page->mapping) | |
832 | continue; | |
833 | /* | |
834 | * Each page->index must be checked when | |
835 | * invalidating or truncating nonlinear. | |
836 | */ | |
837 | if (details->nonlinear_vma && | |
838 | (page->index < details->first_index || | |
839 | page->index > details->last_index)) | |
840 | continue; | |
841 | } | |
b5810039 | 842 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
a600388d | 843 | tlb->fullmm); |
1da177e4 LT |
844 | tlb_remove_tlb_entry(tlb, pte, addr); |
845 | if (unlikely(!page)) | |
846 | continue; | |
847 | if (unlikely(details) && details->nonlinear_vma | |
848 | && linear_page_index(details->nonlinear_vma, | |
849 | addr) != page->index) | |
b5810039 | 850 | set_pte_at(mm, addr, pte, |
1da177e4 | 851 | pgoff_to_pte(page->index)); |
1da177e4 | 852 | if (PageAnon(page)) |
86d912f4 | 853 | anon_rss--; |
6237bcd9 HD |
854 | else { |
855 | if (pte_dirty(ptent)) | |
856 | set_page_dirty(page); | |
4917e5d0 JW |
857 | if (pte_young(ptent) && |
858 | likely(!VM_SequentialReadHint(vma))) | |
bf3f3bc5 | 859 | mark_page_accessed(page); |
86d912f4 | 860 | file_rss--; |
6237bcd9 | 861 | } |
edc315fd | 862 | page_remove_rmap(page); |
3dc14741 HD |
863 | if (unlikely(page_mapcount(page) < 0)) |
864 | print_bad_pte(vma, addr, ptent, page); | |
1da177e4 LT |
865 | tlb_remove_page(tlb, page); |
866 | continue; | |
867 | } | |
868 | /* | |
869 | * If details->check_mapping, we leave swap entries; | |
870 | * if details->nonlinear_vma, we leave file entries. | |
871 | */ | |
872 | if (unlikely(details)) | |
873 | continue; | |
2509ef26 HD |
874 | if (pte_file(ptent)) { |
875 | if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) | |
876 | print_bad_pte(vma, addr, ptent, NULL); | |
877 | } else if | |
878 | (unlikely(!free_swap_and_cache(pte_to_swp_entry(ptent)))) | |
879 | print_bad_pte(vma, addr, ptent, NULL); | |
9888a1ca | 880 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
51c6f666 | 881 | } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0)); |
ae859762 | 882 | |
86d912f4 | 883 | add_mm_rss(mm, file_rss, anon_rss); |
6606c3e0 | 884 | arch_leave_lazy_mmu_mode(); |
508034a3 | 885 | pte_unmap_unlock(pte - 1, ptl); |
51c6f666 RH |
886 | |
887 | return addr; | |
1da177e4 LT |
888 | } |
889 | ||
51c6f666 | 890 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
b5810039 | 891 | struct vm_area_struct *vma, pud_t *pud, |
1da177e4 | 892 | unsigned long addr, unsigned long end, |
51c6f666 | 893 | long *zap_work, struct zap_details *details) |
1da177e4 LT |
894 | { |
895 | pmd_t *pmd; | |
896 | unsigned long next; | |
897 | ||
898 | pmd = pmd_offset(pud, addr); | |
899 | do { | |
900 | next = pmd_addr_end(addr, end); | |
51c6f666 RH |
901 | if (pmd_none_or_clear_bad(pmd)) { |
902 | (*zap_work)--; | |
1da177e4 | 903 | continue; |
51c6f666 RH |
904 | } |
905 | next = zap_pte_range(tlb, vma, pmd, addr, next, | |
906 | zap_work, details); | |
907 | } while (pmd++, addr = next, (addr != end && *zap_work > 0)); | |
908 | ||
909 | return addr; | |
1da177e4 LT |
910 | } |
911 | ||
51c6f666 | 912 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
b5810039 | 913 | struct vm_area_struct *vma, pgd_t *pgd, |
1da177e4 | 914 | unsigned long addr, unsigned long end, |
51c6f666 | 915 | long *zap_work, struct zap_details *details) |
1da177e4 LT |
916 | { |
917 | pud_t *pud; | |
918 | unsigned long next; | |
919 | ||
920 | pud = pud_offset(pgd, addr); | |
921 | do { | |
922 | next = pud_addr_end(addr, end); | |
51c6f666 RH |
923 | if (pud_none_or_clear_bad(pud)) { |
924 | (*zap_work)--; | |
1da177e4 | 925 | continue; |
51c6f666 RH |
926 | } |
927 | next = zap_pmd_range(tlb, vma, pud, addr, next, | |
928 | zap_work, details); | |
929 | } while (pud++, addr = next, (addr != end && *zap_work > 0)); | |
930 | ||
931 | return addr; | |
1da177e4 LT |
932 | } |
933 | ||
51c6f666 RH |
934 | static unsigned long unmap_page_range(struct mmu_gather *tlb, |
935 | struct vm_area_struct *vma, | |
1da177e4 | 936 | unsigned long addr, unsigned long end, |
51c6f666 | 937 | long *zap_work, struct zap_details *details) |
1da177e4 LT |
938 | { |
939 | pgd_t *pgd; | |
940 | unsigned long next; | |
941 | ||
942 | if (details && !details->check_mapping && !details->nonlinear_vma) | |
943 | details = NULL; | |
944 | ||
945 | BUG_ON(addr >= end); | |
946 | tlb_start_vma(tlb, vma); | |
947 | pgd = pgd_offset(vma->vm_mm, addr); | |
948 | do { | |
949 | next = pgd_addr_end(addr, end); | |
51c6f666 RH |
950 | if (pgd_none_or_clear_bad(pgd)) { |
951 | (*zap_work)--; | |
1da177e4 | 952 | continue; |
51c6f666 RH |
953 | } |
954 | next = zap_pud_range(tlb, vma, pgd, addr, next, | |
955 | zap_work, details); | |
956 | } while (pgd++, addr = next, (addr != end && *zap_work > 0)); | |
1da177e4 | 957 | tlb_end_vma(tlb, vma); |
51c6f666 RH |
958 | |
959 | return addr; | |
1da177e4 LT |
960 | } |
961 | ||
962 | #ifdef CONFIG_PREEMPT | |
963 | # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE) | |
964 | #else | |
965 | /* No preempt: go for improved straight-line efficiency */ | |
966 | # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE) | |
967 | #endif | |
968 | ||
969 | /** | |
970 | * unmap_vmas - unmap a range of memory covered by a list of vma's | |
971 | * @tlbp: address of the caller's struct mmu_gather | |
1da177e4 LT |
972 | * @vma: the starting vma |
973 | * @start_addr: virtual address at which to start unmapping | |
974 | * @end_addr: virtual address at which to end unmapping | |
975 | * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here | |
976 | * @details: details of nonlinear truncation or shared cache invalidation | |
977 | * | |
ee39b37b | 978 | * Returns the end address of the unmapping (restart addr if interrupted). |
1da177e4 | 979 | * |
508034a3 | 980 | * Unmap all pages in the vma list. |
1da177e4 | 981 | * |
508034a3 HD |
982 | * We aim to not hold locks for too long (for scheduling latency reasons). |
983 | * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to | |
1da177e4 LT |
984 | * return the ending mmu_gather to the caller. |
985 | * | |
986 | * Only addresses between `start' and `end' will be unmapped. | |
987 | * | |
988 | * The VMA list must be sorted in ascending virtual address order. | |
989 | * | |
990 | * unmap_vmas() assumes that the caller will flush the whole unmapped address | |
991 | * range after unmap_vmas() returns. So the only responsibility here is to | |
992 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() | |
993 | * drops the lock and schedules. | |
994 | */ | |
508034a3 | 995 | unsigned long unmap_vmas(struct mmu_gather **tlbp, |
1da177e4 LT |
996 | struct vm_area_struct *vma, unsigned long start_addr, |
997 | unsigned long end_addr, unsigned long *nr_accounted, | |
998 | struct zap_details *details) | |
999 | { | |
51c6f666 | 1000 | long zap_work = ZAP_BLOCK_SIZE; |
1da177e4 LT |
1001 | unsigned long tlb_start = 0; /* For tlb_finish_mmu */ |
1002 | int tlb_start_valid = 0; | |
ee39b37b | 1003 | unsigned long start = start_addr; |
1da177e4 | 1004 | spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL; |
4d6ddfa9 | 1005 | int fullmm = (*tlbp)->fullmm; |
cddb8a5c | 1006 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 1007 | |
cddb8a5c | 1008 | mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); |
1da177e4 | 1009 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) { |
1da177e4 LT |
1010 | unsigned long end; |
1011 | ||
1012 | start = max(vma->vm_start, start_addr); | |
1013 | if (start >= vma->vm_end) | |
1014 | continue; | |
1015 | end = min(vma->vm_end, end_addr); | |
1016 | if (end <= vma->vm_start) | |
1017 | continue; | |
1018 | ||
1019 | if (vma->vm_flags & VM_ACCOUNT) | |
1020 | *nr_accounted += (end - start) >> PAGE_SHIFT; | |
1021 | ||
34801ba9 | 1022 | if (unlikely(is_pfn_mapping(vma))) |
2ab64037 | 1023 | untrack_pfn_vma(vma, 0, 0); |
1024 | ||
1da177e4 | 1025 | while (start != end) { |
1da177e4 LT |
1026 | if (!tlb_start_valid) { |
1027 | tlb_start = start; | |
1028 | tlb_start_valid = 1; | |
1029 | } | |
1030 | ||
51c6f666 | 1031 | if (unlikely(is_vm_hugetlb_page(vma))) { |
a137e1cc AK |
1032 | /* |
1033 | * It is undesirable to test vma->vm_file as it | |
1034 | * should be non-null for valid hugetlb area. | |
1035 | * However, vm_file will be NULL in the error | |
1036 | * cleanup path of do_mmap_pgoff. When | |
1037 | * hugetlbfs ->mmap method fails, | |
1038 | * do_mmap_pgoff() nullifies vma->vm_file | |
1039 | * before calling this function to clean up. | |
1040 | * Since no pte has actually been setup, it is | |
1041 | * safe to do nothing in this case. | |
1042 | */ | |
1043 | if (vma->vm_file) { | |
1044 | unmap_hugepage_range(vma, start, end, NULL); | |
1045 | zap_work -= (end - start) / | |
a5516438 | 1046 | pages_per_huge_page(hstate_vma(vma)); |
a137e1cc AK |
1047 | } |
1048 | ||
51c6f666 RH |
1049 | start = end; |
1050 | } else | |
1051 | start = unmap_page_range(*tlbp, vma, | |
1052 | start, end, &zap_work, details); | |
1053 | ||
1054 | if (zap_work > 0) { | |
1055 | BUG_ON(start != end); | |
1056 | break; | |
1da177e4 LT |
1057 | } |
1058 | ||
1da177e4 LT |
1059 | tlb_finish_mmu(*tlbp, tlb_start, start); |
1060 | ||
1061 | if (need_resched() || | |
95c354fe | 1062 | (i_mmap_lock && spin_needbreak(i_mmap_lock))) { |
1da177e4 | 1063 | if (i_mmap_lock) { |
508034a3 | 1064 | *tlbp = NULL; |
1da177e4 LT |
1065 | goto out; |
1066 | } | |
1da177e4 | 1067 | cond_resched(); |
1da177e4 LT |
1068 | } |
1069 | ||
508034a3 | 1070 | *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm); |
1da177e4 | 1071 | tlb_start_valid = 0; |
51c6f666 | 1072 | zap_work = ZAP_BLOCK_SIZE; |
1da177e4 LT |
1073 | } |
1074 | } | |
1075 | out: | |
cddb8a5c | 1076 | mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); |
ee39b37b | 1077 | return start; /* which is now the end (or restart) address */ |
1da177e4 LT |
1078 | } |
1079 | ||
1080 | /** | |
1081 | * zap_page_range - remove user pages in a given range | |
1082 | * @vma: vm_area_struct holding the applicable pages | |
1083 | * @address: starting address of pages to zap | |
1084 | * @size: number of bytes to zap | |
1085 | * @details: details of nonlinear truncation or shared cache invalidation | |
1086 | */ | |
ee39b37b | 1087 | unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address, |
1da177e4 LT |
1088 | unsigned long size, struct zap_details *details) |
1089 | { | |
1090 | struct mm_struct *mm = vma->vm_mm; | |
1091 | struct mmu_gather *tlb; | |
1092 | unsigned long end = address + size; | |
1093 | unsigned long nr_accounted = 0; | |
1094 | ||
1da177e4 | 1095 | lru_add_drain(); |
1da177e4 | 1096 | tlb = tlb_gather_mmu(mm, 0); |
365e9c87 | 1097 | update_hiwater_rss(mm); |
508034a3 HD |
1098 | end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details); |
1099 | if (tlb) | |
1100 | tlb_finish_mmu(tlb, address, end); | |
ee39b37b | 1101 | return end; |
1da177e4 LT |
1102 | } |
1103 | ||
c627f9cc JS |
1104 | /** |
1105 | * zap_vma_ptes - remove ptes mapping the vma | |
1106 | * @vma: vm_area_struct holding ptes to be zapped | |
1107 | * @address: starting address of pages to zap | |
1108 | * @size: number of bytes to zap | |
1109 | * | |
1110 | * This function only unmaps ptes assigned to VM_PFNMAP vmas. | |
1111 | * | |
1112 | * The entire address range must be fully contained within the vma. | |
1113 | * | |
1114 | * Returns 0 if successful. | |
1115 | */ | |
1116 | int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, | |
1117 | unsigned long size) | |
1118 | { | |
1119 | if (address < vma->vm_start || address + size > vma->vm_end || | |
1120 | !(vma->vm_flags & VM_PFNMAP)) | |
1121 | return -1; | |
1122 | zap_page_range(vma, address, size, NULL); | |
1123 | return 0; | |
1124 | } | |
1125 | EXPORT_SYMBOL_GPL(zap_vma_ptes); | |
1126 | ||
1da177e4 LT |
1127 | /* |
1128 | * Do a quick page-table lookup for a single page. | |
1da177e4 | 1129 | */ |
6aab341e | 1130 | struct page *follow_page(struct vm_area_struct *vma, unsigned long address, |
deceb6cd | 1131 | unsigned int flags) |
1da177e4 LT |
1132 | { |
1133 | pgd_t *pgd; | |
1134 | pud_t *pud; | |
1135 | pmd_t *pmd; | |
1136 | pte_t *ptep, pte; | |
deceb6cd | 1137 | spinlock_t *ptl; |
1da177e4 | 1138 | struct page *page; |
6aab341e | 1139 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 1140 | |
deceb6cd HD |
1141 | page = follow_huge_addr(mm, address, flags & FOLL_WRITE); |
1142 | if (!IS_ERR(page)) { | |
1143 | BUG_ON(flags & FOLL_GET); | |
1144 | goto out; | |
1145 | } | |
1da177e4 | 1146 | |
deceb6cd | 1147 | page = NULL; |
1da177e4 LT |
1148 | pgd = pgd_offset(mm, address); |
1149 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
deceb6cd | 1150 | goto no_page_table; |
1da177e4 LT |
1151 | |
1152 | pud = pud_offset(pgd, address); | |
ceb86879 | 1153 | if (pud_none(*pud)) |
deceb6cd | 1154 | goto no_page_table; |
ceb86879 AK |
1155 | if (pud_huge(*pud)) { |
1156 | BUG_ON(flags & FOLL_GET); | |
1157 | page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE); | |
1158 | goto out; | |
1159 | } | |
1160 | if (unlikely(pud_bad(*pud))) | |
1161 | goto no_page_table; | |
1162 | ||
1da177e4 | 1163 | pmd = pmd_offset(pud, address); |
aeed5fce | 1164 | if (pmd_none(*pmd)) |
deceb6cd | 1165 | goto no_page_table; |
deceb6cd HD |
1166 | if (pmd_huge(*pmd)) { |
1167 | BUG_ON(flags & FOLL_GET); | |
1168 | page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE); | |
1da177e4 | 1169 | goto out; |
deceb6cd | 1170 | } |
aeed5fce HD |
1171 | if (unlikely(pmd_bad(*pmd))) |
1172 | goto no_page_table; | |
1173 | ||
deceb6cd | 1174 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); |
1da177e4 LT |
1175 | |
1176 | pte = *ptep; | |
deceb6cd | 1177 | if (!pte_present(pte)) |
89f5b7da | 1178 | goto no_page; |
deceb6cd HD |
1179 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
1180 | goto unlock; | |
a13ea5b7 | 1181 | |
6aab341e | 1182 | page = vm_normal_page(vma, address, pte); |
a13ea5b7 HD |
1183 | if (unlikely(!page)) { |
1184 | if ((flags & FOLL_DUMP) || | |
62eede62 | 1185 | !is_zero_pfn(pte_pfn(pte))) |
a13ea5b7 HD |
1186 | goto bad_page; |
1187 | page = pte_page(pte); | |
1188 | } | |
1da177e4 | 1189 | |
deceb6cd HD |
1190 | if (flags & FOLL_GET) |
1191 | get_page(page); | |
1192 | if (flags & FOLL_TOUCH) { | |
1193 | if ((flags & FOLL_WRITE) && | |
1194 | !pte_dirty(pte) && !PageDirty(page)) | |
1195 | set_page_dirty(page); | |
bd775c42 KM |
1196 | /* |
1197 | * pte_mkyoung() would be more correct here, but atomic care | |
1198 | * is needed to avoid losing the dirty bit: it is easier to use | |
1199 | * mark_page_accessed(). | |
1200 | */ | |
deceb6cd HD |
1201 | mark_page_accessed(page); |
1202 | } | |
1203 | unlock: | |
1204 | pte_unmap_unlock(ptep, ptl); | |
1da177e4 | 1205 | out: |
deceb6cd | 1206 | return page; |
1da177e4 | 1207 | |
89f5b7da LT |
1208 | bad_page: |
1209 | pte_unmap_unlock(ptep, ptl); | |
1210 | return ERR_PTR(-EFAULT); | |
1211 | ||
1212 | no_page: | |
1213 | pte_unmap_unlock(ptep, ptl); | |
1214 | if (!pte_none(pte)) | |
1215 | return page; | |
8e4b9a60 | 1216 | |
deceb6cd HD |
1217 | no_page_table: |
1218 | /* | |
1219 | * When core dumping an enormous anonymous area that nobody | |
8e4b9a60 HD |
1220 | * has touched so far, we don't want to allocate unnecessary pages or |
1221 | * page tables. Return error instead of NULL to skip handle_mm_fault, | |
1222 | * then get_dump_page() will return NULL to leave a hole in the dump. | |
1223 | * But we can only make this optimization where a hole would surely | |
1224 | * be zero-filled if handle_mm_fault() actually did handle it. | |
deceb6cd | 1225 | */ |
8e4b9a60 HD |
1226 | if ((flags & FOLL_DUMP) && |
1227 | (!vma->vm_ops || !vma->vm_ops->fault)) | |
1228 | return ERR_PTR(-EFAULT); | |
deceb6cd | 1229 | return page; |
1da177e4 LT |
1230 | } |
1231 | ||
b291f000 | 1232 | int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
58fa879e | 1233 | unsigned long start, int nr_pages, unsigned int gup_flags, |
9d73777e | 1234 | struct page **pages, struct vm_area_struct **vmas) |
1da177e4 LT |
1235 | { |
1236 | int i; | |
58fa879e | 1237 | unsigned long vm_flags; |
1da177e4 | 1238 | |
9d73777e | 1239 | if (nr_pages <= 0) |
900cf086 | 1240 | return 0; |
58fa879e HD |
1241 | |
1242 | VM_BUG_ON(!!pages != !!(gup_flags & FOLL_GET)); | |
1243 | ||
1da177e4 LT |
1244 | /* |
1245 | * Require read or write permissions. | |
58fa879e | 1246 | * If FOLL_FORCE is set, we only require the "MAY" flags. |
1da177e4 | 1247 | */ |
58fa879e HD |
1248 | vm_flags = (gup_flags & FOLL_WRITE) ? |
1249 | (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); | |
1250 | vm_flags &= (gup_flags & FOLL_FORCE) ? | |
1251 | (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); | |
1da177e4 LT |
1252 | i = 0; |
1253 | ||
1254 | do { | |
deceb6cd | 1255 | struct vm_area_struct *vma; |
1da177e4 LT |
1256 | |
1257 | vma = find_extend_vma(mm, start); | |
1258 | if (!vma && in_gate_area(tsk, start)) { | |
1259 | unsigned long pg = start & PAGE_MASK; | |
1260 | struct vm_area_struct *gate_vma = get_gate_vma(tsk); | |
1261 | pgd_t *pgd; | |
1262 | pud_t *pud; | |
1263 | pmd_t *pmd; | |
1264 | pte_t *pte; | |
b291f000 NP |
1265 | |
1266 | /* user gate pages are read-only */ | |
58fa879e | 1267 | if (gup_flags & FOLL_WRITE) |
1da177e4 LT |
1268 | return i ? : -EFAULT; |
1269 | if (pg > TASK_SIZE) | |
1270 | pgd = pgd_offset_k(pg); | |
1271 | else | |
1272 | pgd = pgd_offset_gate(mm, pg); | |
1273 | BUG_ON(pgd_none(*pgd)); | |
1274 | pud = pud_offset(pgd, pg); | |
1275 | BUG_ON(pud_none(*pud)); | |
1276 | pmd = pmd_offset(pud, pg); | |
690dbe1c HD |
1277 | if (pmd_none(*pmd)) |
1278 | return i ? : -EFAULT; | |
1da177e4 | 1279 | pte = pte_offset_map(pmd, pg); |
690dbe1c HD |
1280 | if (pte_none(*pte)) { |
1281 | pte_unmap(pte); | |
1282 | return i ? : -EFAULT; | |
1283 | } | |
1da177e4 | 1284 | if (pages) { |
fa2a455b | 1285 | struct page *page = vm_normal_page(gate_vma, start, *pte); |
6aab341e LT |
1286 | pages[i] = page; |
1287 | if (page) | |
1288 | get_page(page); | |
1da177e4 LT |
1289 | } |
1290 | pte_unmap(pte); | |
1291 | if (vmas) | |
1292 | vmas[i] = gate_vma; | |
1293 | i++; | |
1294 | start += PAGE_SIZE; | |
9d73777e | 1295 | nr_pages--; |
1da177e4 LT |
1296 | continue; |
1297 | } | |
1298 | ||
b291f000 NP |
1299 | if (!vma || |
1300 | (vma->vm_flags & (VM_IO | VM_PFNMAP)) || | |
1c3aff1c | 1301 | !(vm_flags & vma->vm_flags)) |
1da177e4 LT |
1302 | return i ? : -EFAULT; |
1303 | ||
2a15efc9 HD |
1304 | if (is_vm_hugetlb_page(vma)) { |
1305 | i = follow_hugetlb_page(mm, vma, pages, vmas, | |
58fa879e | 1306 | &start, &nr_pages, i, gup_flags); |
2a15efc9 HD |
1307 | continue; |
1308 | } | |
deceb6cd | 1309 | |
1da177e4 | 1310 | do { |
08ef4729 | 1311 | struct page *page; |
58fa879e | 1312 | unsigned int foll_flags = gup_flags; |
1da177e4 | 1313 | |
462e00cc | 1314 | /* |
4779280d | 1315 | * If we have a pending SIGKILL, don't keep faulting |
1c3aff1c | 1316 | * pages and potentially allocating memory. |
462e00cc | 1317 | */ |
1c3aff1c | 1318 | if (unlikely(fatal_signal_pending(current))) |
4779280d | 1319 | return i ? i : -ERESTARTSYS; |
462e00cc | 1320 | |
deceb6cd | 1321 | cond_resched(); |
6aab341e | 1322 | while (!(page = follow_page(vma, start, foll_flags))) { |
deceb6cd | 1323 | int ret; |
d06063cc | 1324 | |
d26ed650 HD |
1325 | ret = handle_mm_fault(mm, vma, start, |
1326 | (foll_flags & FOLL_WRITE) ? | |
1327 | FAULT_FLAG_WRITE : 0); | |
1328 | ||
83c54070 NP |
1329 | if (ret & VM_FAULT_ERROR) { |
1330 | if (ret & VM_FAULT_OOM) | |
1331 | return i ? i : -ENOMEM; | |
d1737fdb AK |
1332 | if (ret & |
1333 | (VM_FAULT_HWPOISON|VM_FAULT_SIGBUS)) | |
83c54070 NP |
1334 | return i ? i : -EFAULT; |
1335 | BUG(); | |
1336 | } | |
1337 | if (ret & VM_FAULT_MAJOR) | |
1338 | tsk->maj_flt++; | |
1339 | else | |
1340 | tsk->min_flt++; | |
1341 | ||
a68d2ebc | 1342 | /* |
83c54070 NP |
1343 | * The VM_FAULT_WRITE bit tells us that |
1344 | * do_wp_page has broken COW when necessary, | |
1345 | * even if maybe_mkwrite decided not to set | |
1346 | * pte_write. We can thus safely do subsequent | |
878b63ac HD |
1347 | * page lookups as if they were reads. But only |
1348 | * do so when looping for pte_write is futile: | |
1349 | * in some cases userspace may also be wanting | |
1350 | * to write to the gotten user page, which a | |
1351 | * read fault here might prevent (a readonly | |
1352 | * page might get reCOWed by userspace write). | |
a68d2ebc | 1353 | */ |
878b63ac HD |
1354 | if ((ret & VM_FAULT_WRITE) && |
1355 | !(vma->vm_flags & VM_WRITE)) | |
deceb6cd | 1356 | foll_flags &= ~FOLL_WRITE; |
83c54070 | 1357 | |
7f7bbbe5 | 1358 | cond_resched(); |
1da177e4 | 1359 | } |
89f5b7da LT |
1360 | if (IS_ERR(page)) |
1361 | return i ? i : PTR_ERR(page); | |
1da177e4 | 1362 | if (pages) { |
08ef4729 | 1363 | pages[i] = page; |
03beb076 | 1364 | |
a6f36be3 | 1365 | flush_anon_page(vma, page, start); |
08ef4729 | 1366 | flush_dcache_page(page); |
1da177e4 LT |
1367 | } |
1368 | if (vmas) | |
1369 | vmas[i] = vma; | |
1370 | i++; | |
1371 | start += PAGE_SIZE; | |
9d73777e PZ |
1372 | nr_pages--; |
1373 | } while (nr_pages && start < vma->vm_end); | |
1374 | } while (nr_pages); | |
1da177e4 LT |
1375 | return i; |
1376 | } | |
b291f000 | 1377 | |
d2bf6be8 NP |
1378 | /** |
1379 | * get_user_pages() - pin user pages in memory | |
1380 | * @tsk: task_struct of target task | |
1381 | * @mm: mm_struct of target mm | |
1382 | * @start: starting user address | |
9d73777e | 1383 | * @nr_pages: number of pages from start to pin |
d2bf6be8 NP |
1384 | * @write: whether pages will be written to by the caller |
1385 | * @force: whether to force write access even if user mapping is | |
1386 | * readonly. This will result in the page being COWed even | |
1387 | * in MAP_SHARED mappings. You do not want this. | |
1388 | * @pages: array that receives pointers to the pages pinned. | |
1389 | * Should be at least nr_pages long. Or NULL, if caller | |
1390 | * only intends to ensure the pages are faulted in. | |
1391 | * @vmas: array of pointers to vmas corresponding to each page. | |
1392 | * Or NULL if the caller does not require them. | |
1393 | * | |
1394 | * Returns number of pages pinned. This may be fewer than the number | |
9d73777e | 1395 | * requested. If nr_pages is 0 or negative, returns 0. If no pages |
d2bf6be8 NP |
1396 | * were pinned, returns -errno. Each page returned must be released |
1397 | * with a put_page() call when it is finished with. vmas will only | |
1398 | * remain valid while mmap_sem is held. | |
1399 | * | |
1400 | * Must be called with mmap_sem held for read or write. | |
1401 | * | |
1402 | * get_user_pages walks a process's page tables and takes a reference to | |
1403 | * each struct page that each user address corresponds to at a given | |
1404 | * instant. That is, it takes the page that would be accessed if a user | |
1405 | * thread accesses the given user virtual address at that instant. | |
1406 | * | |
1407 | * This does not guarantee that the page exists in the user mappings when | |
1408 | * get_user_pages returns, and there may even be a completely different | |
1409 | * page there in some cases (eg. if mmapped pagecache has been invalidated | |
1410 | * and subsequently re faulted). However it does guarantee that the page | |
1411 | * won't be freed completely. And mostly callers simply care that the page | |
1412 | * contains data that was valid *at some point in time*. Typically, an IO | |
1413 | * or similar operation cannot guarantee anything stronger anyway because | |
1414 | * locks can't be held over the syscall boundary. | |
1415 | * | |
1416 | * If write=0, the page must not be written to. If the page is written to, | |
1417 | * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called | |
1418 | * after the page is finished with, and before put_page is called. | |
1419 | * | |
1420 | * get_user_pages is typically used for fewer-copy IO operations, to get a | |
1421 | * handle on the memory by some means other than accesses via the user virtual | |
1422 | * addresses. The pages may be submitted for DMA to devices or accessed via | |
1423 | * their kernel linear mapping (via the kmap APIs). Care should be taken to | |
1424 | * use the correct cache flushing APIs. | |
1425 | * | |
1426 | * See also get_user_pages_fast, for performance critical applications. | |
1427 | */ | |
b291f000 | 1428 | int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
9d73777e | 1429 | unsigned long start, int nr_pages, int write, int force, |
b291f000 NP |
1430 | struct page **pages, struct vm_area_struct **vmas) |
1431 | { | |
58fa879e | 1432 | int flags = FOLL_TOUCH; |
b291f000 | 1433 | |
58fa879e HD |
1434 | if (pages) |
1435 | flags |= FOLL_GET; | |
b291f000 | 1436 | if (write) |
58fa879e | 1437 | flags |= FOLL_WRITE; |
b291f000 | 1438 | if (force) |
58fa879e | 1439 | flags |= FOLL_FORCE; |
b291f000 | 1440 | |
9d73777e | 1441 | return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas); |
b291f000 | 1442 | } |
1da177e4 LT |
1443 | EXPORT_SYMBOL(get_user_pages); |
1444 | ||
f3e8fccd HD |
1445 | /** |
1446 | * get_dump_page() - pin user page in memory while writing it to core dump | |
1447 | * @addr: user address | |
1448 | * | |
1449 | * Returns struct page pointer of user page pinned for dump, | |
1450 | * to be freed afterwards by page_cache_release() or put_page(). | |
1451 | * | |
1452 | * Returns NULL on any kind of failure - a hole must then be inserted into | |
1453 | * the corefile, to preserve alignment with its headers; and also returns | |
1454 | * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found - | |
1455 | * allowing a hole to be left in the corefile to save diskspace. | |
1456 | * | |
1457 | * Called without mmap_sem, but after all other threads have been killed. | |
1458 | */ | |
1459 | #ifdef CONFIG_ELF_CORE | |
1460 | struct page *get_dump_page(unsigned long addr) | |
1461 | { | |
1462 | struct vm_area_struct *vma; | |
1463 | struct page *page; | |
1464 | ||
1465 | if (__get_user_pages(current, current->mm, addr, 1, | |
58fa879e | 1466 | FOLL_FORCE | FOLL_DUMP | FOLL_GET, &page, &vma) < 1) |
f3e8fccd | 1467 | return NULL; |
f3e8fccd HD |
1468 | flush_cache_page(vma, addr, page_to_pfn(page)); |
1469 | return page; | |
1470 | } | |
1471 | #endif /* CONFIG_ELF_CORE */ | |
1472 | ||
920c7a5d HH |
1473 | pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr, |
1474 | spinlock_t **ptl) | |
c9cfcddf LT |
1475 | { |
1476 | pgd_t * pgd = pgd_offset(mm, addr); | |
1477 | pud_t * pud = pud_alloc(mm, pgd, addr); | |
1478 | if (pud) { | |
49c91fb0 | 1479 | pmd_t * pmd = pmd_alloc(mm, pud, addr); |
c9cfcddf LT |
1480 | if (pmd) |
1481 | return pte_alloc_map_lock(mm, pmd, addr, ptl); | |
1482 | } | |
1483 | return NULL; | |
1484 | } | |
1485 | ||
238f58d8 LT |
1486 | /* |
1487 | * This is the old fallback for page remapping. | |
1488 | * | |
1489 | * For historical reasons, it only allows reserved pages. Only | |
1490 | * old drivers should use this, and they needed to mark their | |
1491 | * pages reserved for the old functions anyway. | |
1492 | */ | |
423bad60 NP |
1493 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
1494 | struct page *page, pgprot_t prot) | |
238f58d8 | 1495 | { |
423bad60 | 1496 | struct mm_struct *mm = vma->vm_mm; |
238f58d8 | 1497 | int retval; |
c9cfcddf | 1498 | pte_t *pte; |
8a9f3ccd BS |
1499 | spinlock_t *ptl; |
1500 | ||
238f58d8 | 1501 | retval = -EINVAL; |
a145dd41 | 1502 | if (PageAnon(page)) |
5b4e655e | 1503 | goto out; |
238f58d8 LT |
1504 | retval = -ENOMEM; |
1505 | flush_dcache_page(page); | |
c9cfcddf | 1506 | pte = get_locked_pte(mm, addr, &ptl); |
238f58d8 | 1507 | if (!pte) |
5b4e655e | 1508 | goto out; |
238f58d8 LT |
1509 | retval = -EBUSY; |
1510 | if (!pte_none(*pte)) | |
1511 | goto out_unlock; | |
1512 | ||
1513 | /* Ok, finally just insert the thing.. */ | |
1514 | get_page(page); | |
1515 | inc_mm_counter(mm, file_rss); | |
1516 | page_add_file_rmap(page); | |
1517 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); | |
1518 | ||
1519 | retval = 0; | |
8a9f3ccd BS |
1520 | pte_unmap_unlock(pte, ptl); |
1521 | return retval; | |
238f58d8 LT |
1522 | out_unlock: |
1523 | pte_unmap_unlock(pte, ptl); | |
1524 | out: | |
1525 | return retval; | |
1526 | } | |
1527 | ||
bfa5bf6d REB |
1528 | /** |
1529 | * vm_insert_page - insert single page into user vma | |
1530 | * @vma: user vma to map to | |
1531 | * @addr: target user address of this page | |
1532 | * @page: source kernel page | |
1533 | * | |
a145dd41 LT |
1534 | * This allows drivers to insert individual pages they've allocated |
1535 | * into a user vma. | |
1536 | * | |
1537 | * The page has to be a nice clean _individual_ kernel allocation. | |
1538 | * If you allocate a compound page, you need to have marked it as | |
1539 | * such (__GFP_COMP), or manually just split the page up yourself | |
8dfcc9ba | 1540 | * (see split_page()). |
a145dd41 LT |
1541 | * |
1542 | * NOTE! Traditionally this was done with "remap_pfn_range()" which | |
1543 | * took an arbitrary page protection parameter. This doesn't allow | |
1544 | * that. Your vma protection will have to be set up correctly, which | |
1545 | * means that if you want a shared writable mapping, you'd better | |
1546 | * ask for a shared writable mapping! | |
1547 | * | |
1548 | * The page does not need to be reserved. | |
1549 | */ | |
423bad60 NP |
1550 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
1551 | struct page *page) | |
a145dd41 LT |
1552 | { |
1553 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
1554 | return -EFAULT; | |
1555 | if (!page_count(page)) | |
1556 | return -EINVAL; | |
4d7672b4 | 1557 | vma->vm_flags |= VM_INSERTPAGE; |
423bad60 | 1558 | return insert_page(vma, addr, page, vma->vm_page_prot); |
a145dd41 | 1559 | } |
e3c3374f | 1560 | EXPORT_SYMBOL(vm_insert_page); |
a145dd41 | 1561 | |
423bad60 NP |
1562 | static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
1563 | unsigned long pfn, pgprot_t prot) | |
1564 | { | |
1565 | struct mm_struct *mm = vma->vm_mm; | |
1566 | int retval; | |
1567 | pte_t *pte, entry; | |
1568 | spinlock_t *ptl; | |
1569 | ||
1570 | retval = -ENOMEM; | |
1571 | pte = get_locked_pte(mm, addr, &ptl); | |
1572 | if (!pte) | |
1573 | goto out; | |
1574 | retval = -EBUSY; | |
1575 | if (!pte_none(*pte)) | |
1576 | goto out_unlock; | |
1577 | ||
1578 | /* Ok, finally just insert the thing.. */ | |
1579 | entry = pte_mkspecial(pfn_pte(pfn, prot)); | |
1580 | set_pte_at(mm, addr, pte, entry); | |
1581 | update_mmu_cache(vma, addr, entry); /* XXX: why not for insert_page? */ | |
1582 | ||
1583 | retval = 0; | |
1584 | out_unlock: | |
1585 | pte_unmap_unlock(pte, ptl); | |
1586 | out: | |
1587 | return retval; | |
1588 | } | |
1589 | ||
e0dc0d8f NP |
1590 | /** |
1591 | * vm_insert_pfn - insert single pfn into user vma | |
1592 | * @vma: user vma to map to | |
1593 | * @addr: target user address of this page | |
1594 | * @pfn: source kernel pfn | |
1595 | * | |
1596 | * Similar to vm_inert_page, this allows drivers to insert individual pages | |
1597 | * they've allocated into a user vma. Same comments apply. | |
1598 | * | |
1599 | * This function should only be called from a vm_ops->fault handler, and | |
1600 | * in that case the handler should return NULL. | |
0d71d10a NP |
1601 | * |
1602 | * vma cannot be a COW mapping. | |
1603 | * | |
1604 | * As this is called only for pages that do not currently exist, we | |
1605 | * do not need to flush old virtual caches or the TLB. | |
e0dc0d8f NP |
1606 | */ |
1607 | int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, | |
423bad60 | 1608 | unsigned long pfn) |
e0dc0d8f | 1609 | { |
2ab64037 | 1610 | int ret; |
e4b866ed | 1611 | pgprot_t pgprot = vma->vm_page_prot; |
7e675137 NP |
1612 | /* |
1613 | * Technically, architectures with pte_special can avoid all these | |
1614 | * restrictions (same for remap_pfn_range). However we would like | |
1615 | * consistency in testing and feature parity among all, so we should | |
1616 | * try to keep these invariants in place for everybody. | |
1617 | */ | |
b379d790 JH |
1618 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); |
1619 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == | |
1620 | (VM_PFNMAP|VM_MIXEDMAP)); | |
1621 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); | |
1622 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); | |
e0dc0d8f | 1623 | |
423bad60 NP |
1624 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1625 | return -EFAULT; | |
e4b866ed | 1626 | if (track_pfn_vma_new(vma, &pgprot, pfn, PAGE_SIZE)) |
2ab64037 | 1627 | return -EINVAL; |
1628 | ||
e4b866ed | 1629 | ret = insert_pfn(vma, addr, pfn, pgprot); |
2ab64037 | 1630 | |
1631 | if (ret) | |
1632 | untrack_pfn_vma(vma, pfn, PAGE_SIZE); | |
1633 | ||
1634 | return ret; | |
423bad60 NP |
1635 | } |
1636 | EXPORT_SYMBOL(vm_insert_pfn); | |
e0dc0d8f | 1637 | |
423bad60 NP |
1638 | int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, |
1639 | unsigned long pfn) | |
1640 | { | |
1641 | BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); | |
e0dc0d8f | 1642 | |
423bad60 NP |
1643 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1644 | return -EFAULT; | |
e0dc0d8f | 1645 | |
423bad60 NP |
1646 | /* |
1647 | * If we don't have pte special, then we have to use the pfn_valid() | |
1648 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* | |
1649 | * refcount the page if pfn_valid is true (hence insert_page rather | |
62eede62 HD |
1650 | * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP |
1651 | * without pte special, it would there be refcounted as a normal page. | |
423bad60 NP |
1652 | */ |
1653 | if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) { | |
1654 | struct page *page; | |
1655 | ||
1656 | page = pfn_to_page(pfn); | |
1657 | return insert_page(vma, addr, page, vma->vm_page_prot); | |
1658 | } | |
1659 | return insert_pfn(vma, addr, pfn, vma->vm_page_prot); | |
e0dc0d8f | 1660 | } |
423bad60 | 1661 | EXPORT_SYMBOL(vm_insert_mixed); |
e0dc0d8f | 1662 | |
1da177e4 LT |
1663 | /* |
1664 | * maps a range of physical memory into the requested pages. the old | |
1665 | * mappings are removed. any references to nonexistent pages results | |
1666 | * in null mappings (currently treated as "copy-on-access") | |
1667 | */ | |
1668 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
1669 | unsigned long addr, unsigned long end, | |
1670 | unsigned long pfn, pgprot_t prot) | |
1671 | { | |
1672 | pte_t *pte; | |
c74df32c | 1673 | spinlock_t *ptl; |
1da177e4 | 1674 | |
c74df32c | 1675 | pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1da177e4 LT |
1676 | if (!pte) |
1677 | return -ENOMEM; | |
6606c3e0 | 1678 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1679 | do { |
1680 | BUG_ON(!pte_none(*pte)); | |
7e675137 | 1681 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
1da177e4 LT |
1682 | pfn++; |
1683 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 1684 | arch_leave_lazy_mmu_mode(); |
c74df32c | 1685 | pte_unmap_unlock(pte - 1, ptl); |
1da177e4 LT |
1686 | return 0; |
1687 | } | |
1688 | ||
1689 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
1690 | unsigned long addr, unsigned long end, | |
1691 | unsigned long pfn, pgprot_t prot) | |
1692 | { | |
1693 | pmd_t *pmd; | |
1694 | unsigned long next; | |
1695 | ||
1696 | pfn -= addr >> PAGE_SHIFT; | |
1697 | pmd = pmd_alloc(mm, pud, addr); | |
1698 | if (!pmd) | |
1699 | return -ENOMEM; | |
1700 | do { | |
1701 | next = pmd_addr_end(addr, end); | |
1702 | if (remap_pte_range(mm, pmd, addr, next, | |
1703 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1704 | return -ENOMEM; | |
1705 | } while (pmd++, addr = next, addr != end); | |
1706 | return 0; | |
1707 | } | |
1708 | ||
1709 | static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, | |
1710 | unsigned long addr, unsigned long end, | |
1711 | unsigned long pfn, pgprot_t prot) | |
1712 | { | |
1713 | pud_t *pud; | |
1714 | unsigned long next; | |
1715 | ||
1716 | pfn -= addr >> PAGE_SHIFT; | |
1717 | pud = pud_alloc(mm, pgd, addr); | |
1718 | if (!pud) | |
1719 | return -ENOMEM; | |
1720 | do { | |
1721 | next = pud_addr_end(addr, end); | |
1722 | if (remap_pmd_range(mm, pud, addr, next, | |
1723 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1724 | return -ENOMEM; | |
1725 | } while (pud++, addr = next, addr != end); | |
1726 | return 0; | |
1727 | } | |
1728 | ||
bfa5bf6d REB |
1729 | /** |
1730 | * remap_pfn_range - remap kernel memory to userspace | |
1731 | * @vma: user vma to map to | |
1732 | * @addr: target user address to start at | |
1733 | * @pfn: physical address of kernel memory | |
1734 | * @size: size of map area | |
1735 | * @prot: page protection flags for this mapping | |
1736 | * | |
1737 | * Note: this is only safe if the mm semaphore is held when called. | |
1738 | */ | |
1da177e4 LT |
1739 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
1740 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
1741 | { | |
1742 | pgd_t *pgd; | |
1743 | unsigned long next; | |
2d15cab8 | 1744 | unsigned long end = addr + PAGE_ALIGN(size); |
1da177e4 LT |
1745 | struct mm_struct *mm = vma->vm_mm; |
1746 | int err; | |
1747 | ||
1748 | /* | |
1749 | * Physically remapped pages are special. Tell the | |
1750 | * rest of the world about it: | |
1751 | * VM_IO tells people not to look at these pages | |
1752 | * (accesses can have side effects). | |
0b14c179 HD |
1753 | * VM_RESERVED is specified all over the place, because |
1754 | * in 2.4 it kept swapout's vma scan off this vma; but | |
1755 | * in 2.6 the LRU scan won't even find its pages, so this | |
1756 | * flag means no more than count its pages in reserved_vm, | |
1757 | * and omit it from core dump, even when VM_IO turned off. | |
6aab341e LT |
1758 | * VM_PFNMAP tells the core MM that the base pages are just |
1759 | * raw PFN mappings, and do not have a "struct page" associated | |
1760 | * with them. | |
fb155c16 LT |
1761 | * |
1762 | * There's a horrible special case to handle copy-on-write | |
1763 | * behaviour that some programs depend on. We mark the "original" | |
1764 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". | |
1da177e4 | 1765 | */ |
4bb9c5c0 | 1766 | if (addr == vma->vm_start && end == vma->vm_end) { |
fb155c16 | 1767 | vma->vm_pgoff = pfn; |
895791da | 1768 | vma->vm_flags |= VM_PFN_AT_MMAP; |
4bb9c5c0 | 1769 | } else if (is_cow_mapping(vma->vm_flags)) |
3c8bb73a | 1770 | return -EINVAL; |
fb155c16 | 1771 | |
6aab341e | 1772 | vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP; |
1da177e4 | 1773 | |
e4b866ed | 1774 | err = track_pfn_vma_new(vma, &prot, pfn, PAGE_ALIGN(size)); |
a3670613 | 1775 | if (err) { |
1776 | /* | |
1777 | * To indicate that track_pfn related cleanup is not | |
1778 | * needed from higher level routine calling unmap_vmas | |
1779 | */ | |
1780 | vma->vm_flags &= ~(VM_IO | VM_RESERVED | VM_PFNMAP); | |
895791da | 1781 | vma->vm_flags &= ~VM_PFN_AT_MMAP; |
2ab64037 | 1782 | return -EINVAL; |
a3670613 | 1783 | } |
2ab64037 | 1784 | |
1da177e4 LT |
1785 | BUG_ON(addr >= end); |
1786 | pfn -= addr >> PAGE_SHIFT; | |
1787 | pgd = pgd_offset(mm, addr); | |
1788 | flush_cache_range(vma, addr, end); | |
1da177e4 LT |
1789 | do { |
1790 | next = pgd_addr_end(addr, end); | |
1791 | err = remap_pud_range(mm, pgd, addr, next, | |
1792 | pfn + (addr >> PAGE_SHIFT), prot); | |
1793 | if (err) | |
1794 | break; | |
1795 | } while (pgd++, addr = next, addr != end); | |
2ab64037 | 1796 | |
1797 | if (err) | |
1798 | untrack_pfn_vma(vma, pfn, PAGE_ALIGN(size)); | |
1799 | ||
1da177e4 LT |
1800 | return err; |
1801 | } | |
1802 | EXPORT_SYMBOL(remap_pfn_range); | |
1803 | ||
aee16b3c JF |
1804 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
1805 | unsigned long addr, unsigned long end, | |
1806 | pte_fn_t fn, void *data) | |
1807 | { | |
1808 | pte_t *pte; | |
1809 | int err; | |
2f569afd | 1810 | pgtable_t token; |
94909914 | 1811 | spinlock_t *uninitialized_var(ptl); |
aee16b3c JF |
1812 | |
1813 | pte = (mm == &init_mm) ? | |
1814 | pte_alloc_kernel(pmd, addr) : | |
1815 | pte_alloc_map_lock(mm, pmd, addr, &ptl); | |
1816 | if (!pte) | |
1817 | return -ENOMEM; | |
1818 | ||
1819 | BUG_ON(pmd_huge(*pmd)); | |
1820 | ||
38e0edb1 JF |
1821 | arch_enter_lazy_mmu_mode(); |
1822 | ||
2f569afd | 1823 | token = pmd_pgtable(*pmd); |
aee16b3c JF |
1824 | |
1825 | do { | |
c36987e2 | 1826 | err = fn(pte++, token, addr, data); |
aee16b3c JF |
1827 | if (err) |
1828 | break; | |
c36987e2 | 1829 | } while (addr += PAGE_SIZE, addr != end); |
aee16b3c | 1830 | |
38e0edb1 JF |
1831 | arch_leave_lazy_mmu_mode(); |
1832 | ||
aee16b3c JF |
1833 | if (mm != &init_mm) |
1834 | pte_unmap_unlock(pte-1, ptl); | |
1835 | return err; | |
1836 | } | |
1837 | ||
1838 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, | |
1839 | unsigned long addr, unsigned long end, | |
1840 | pte_fn_t fn, void *data) | |
1841 | { | |
1842 | pmd_t *pmd; | |
1843 | unsigned long next; | |
1844 | int err; | |
1845 | ||
ceb86879 AK |
1846 | BUG_ON(pud_huge(*pud)); |
1847 | ||
aee16b3c JF |
1848 | pmd = pmd_alloc(mm, pud, addr); |
1849 | if (!pmd) | |
1850 | return -ENOMEM; | |
1851 | do { | |
1852 | next = pmd_addr_end(addr, end); | |
1853 | err = apply_to_pte_range(mm, pmd, addr, next, fn, data); | |
1854 | if (err) | |
1855 | break; | |
1856 | } while (pmd++, addr = next, addr != end); | |
1857 | return err; | |
1858 | } | |
1859 | ||
1860 | static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd, | |
1861 | unsigned long addr, unsigned long end, | |
1862 | pte_fn_t fn, void *data) | |
1863 | { | |
1864 | pud_t *pud; | |
1865 | unsigned long next; | |
1866 | int err; | |
1867 | ||
1868 | pud = pud_alloc(mm, pgd, addr); | |
1869 | if (!pud) | |
1870 | return -ENOMEM; | |
1871 | do { | |
1872 | next = pud_addr_end(addr, end); | |
1873 | err = apply_to_pmd_range(mm, pud, addr, next, fn, data); | |
1874 | if (err) | |
1875 | break; | |
1876 | } while (pud++, addr = next, addr != end); | |
1877 | return err; | |
1878 | } | |
1879 | ||
1880 | /* | |
1881 | * Scan a region of virtual memory, filling in page tables as necessary | |
1882 | * and calling a provided function on each leaf page table. | |
1883 | */ | |
1884 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, | |
1885 | unsigned long size, pte_fn_t fn, void *data) | |
1886 | { | |
1887 | pgd_t *pgd; | |
1888 | unsigned long next; | |
cddb8a5c | 1889 | unsigned long start = addr, end = addr + size; |
aee16b3c JF |
1890 | int err; |
1891 | ||
1892 | BUG_ON(addr >= end); | |
cddb8a5c | 1893 | mmu_notifier_invalidate_range_start(mm, start, end); |
aee16b3c JF |
1894 | pgd = pgd_offset(mm, addr); |
1895 | do { | |
1896 | next = pgd_addr_end(addr, end); | |
1897 | err = apply_to_pud_range(mm, pgd, addr, next, fn, data); | |
1898 | if (err) | |
1899 | break; | |
1900 | } while (pgd++, addr = next, addr != end); | |
cddb8a5c | 1901 | mmu_notifier_invalidate_range_end(mm, start, end); |
aee16b3c JF |
1902 | return err; |
1903 | } | |
1904 | EXPORT_SYMBOL_GPL(apply_to_page_range); | |
1905 | ||
8f4e2101 HD |
1906 | /* |
1907 | * handle_pte_fault chooses page fault handler according to an entry | |
1908 | * which was read non-atomically. Before making any commitment, on | |
1909 | * those architectures or configurations (e.g. i386 with PAE) which | |
1910 | * might give a mix of unmatched parts, do_swap_page and do_file_page | |
1911 | * must check under lock before unmapping the pte and proceeding | |
1912 | * (but do_wp_page is only called after already making such a check; | |
1913 | * and do_anonymous_page and do_no_page can safely check later on). | |
1914 | */ | |
4c21e2f2 | 1915 | static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, |
8f4e2101 HD |
1916 | pte_t *page_table, pte_t orig_pte) |
1917 | { | |
1918 | int same = 1; | |
1919 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) | |
1920 | if (sizeof(pte_t) > sizeof(unsigned long)) { | |
4c21e2f2 HD |
1921 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
1922 | spin_lock(ptl); | |
8f4e2101 | 1923 | same = pte_same(*page_table, orig_pte); |
4c21e2f2 | 1924 | spin_unlock(ptl); |
8f4e2101 HD |
1925 | } |
1926 | #endif | |
1927 | pte_unmap(page_table); | |
1928 | return same; | |
1929 | } | |
1930 | ||
1da177e4 LT |
1931 | /* |
1932 | * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when | |
1933 | * servicing faults for write access. In the normal case, do always want | |
1934 | * pte_mkwrite. But get_user_pages can cause write faults for mappings | |
1935 | * that do not have writing enabled, when used by access_process_vm. | |
1936 | */ | |
1937 | static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma) | |
1938 | { | |
1939 | if (likely(vma->vm_flags & VM_WRITE)) | |
1940 | pte = pte_mkwrite(pte); | |
1941 | return pte; | |
1942 | } | |
1943 | ||
9de455b2 | 1944 | static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) |
6aab341e LT |
1945 | { |
1946 | /* | |
1947 | * If the source page was a PFN mapping, we don't have | |
1948 | * a "struct page" for it. We do a best-effort copy by | |
1949 | * just copying from the original user address. If that | |
1950 | * fails, we just zero-fill it. Live with it. | |
1951 | */ | |
1952 | if (unlikely(!src)) { | |
1953 | void *kaddr = kmap_atomic(dst, KM_USER0); | |
5d2a2dbb LT |
1954 | void __user *uaddr = (void __user *)(va & PAGE_MASK); |
1955 | ||
1956 | /* | |
1957 | * This really shouldn't fail, because the page is there | |
1958 | * in the page tables. But it might just be unreadable, | |
1959 | * in which case we just give up and fill the result with | |
1960 | * zeroes. | |
1961 | */ | |
1962 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) | |
6aab341e LT |
1963 | memset(kaddr, 0, PAGE_SIZE); |
1964 | kunmap_atomic(kaddr, KM_USER0); | |
c4ec7b0d | 1965 | flush_dcache_page(dst); |
0ed361de NP |
1966 | } else |
1967 | copy_user_highpage(dst, src, va, vma); | |
6aab341e LT |
1968 | } |
1969 | ||
1da177e4 LT |
1970 | /* |
1971 | * This routine handles present pages, when users try to write | |
1972 | * to a shared page. It is done by copying the page to a new address | |
1973 | * and decrementing the shared-page counter for the old page. | |
1974 | * | |
1da177e4 LT |
1975 | * Note that this routine assumes that the protection checks have been |
1976 | * done by the caller (the low-level page fault routine in most cases). | |
1977 | * Thus we can safely just mark it writable once we've done any necessary | |
1978 | * COW. | |
1979 | * | |
1980 | * We also mark the page dirty at this point even though the page will | |
1981 | * change only once the write actually happens. This avoids a few races, | |
1982 | * and potentially makes it more efficient. | |
1983 | * | |
8f4e2101 HD |
1984 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
1985 | * but allow concurrent faults), with pte both mapped and locked. | |
1986 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 1987 | */ |
65500d23 HD |
1988 | static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, |
1989 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
8f4e2101 | 1990 | spinlock_t *ptl, pte_t orig_pte) |
1da177e4 | 1991 | { |
e5bbe4df | 1992 | struct page *old_page, *new_page; |
1da177e4 | 1993 | pte_t entry; |
83c54070 | 1994 | int reuse = 0, ret = 0; |
a200ee18 | 1995 | int page_mkwrite = 0; |
d08b3851 | 1996 | struct page *dirty_page = NULL; |
1da177e4 | 1997 | |
6aab341e | 1998 | old_page = vm_normal_page(vma, address, orig_pte); |
251b97f5 PZ |
1999 | if (!old_page) { |
2000 | /* | |
2001 | * VM_MIXEDMAP !pfn_valid() case | |
2002 | * | |
2003 | * We should not cow pages in a shared writeable mapping. | |
2004 | * Just mark the pages writable as we can't do any dirty | |
2005 | * accounting on raw pfn maps. | |
2006 | */ | |
2007 | if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == | |
2008 | (VM_WRITE|VM_SHARED)) | |
2009 | goto reuse; | |
6aab341e | 2010 | goto gotten; |
251b97f5 | 2011 | } |
1da177e4 | 2012 | |
d08b3851 | 2013 | /* |
ee6a6457 PZ |
2014 | * Take out anonymous pages first, anonymous shared vmas are |
2015 | * not dirty accountable. | |
d08b3851 | 2016 | */ |
9a840895 | 2017 | if (PageAnon(old_page) && !PageKsm(old_page)) { |
ab967d86 HD |
2018 | if (!trylock_page(old_page)) { |
2019 | page_cache_get(old_page); | |
2020 | pte_unmap_unlock(page_table, ptl); | |
2021 | lock_page(old_page); | |
2022 | page_table = pte_offset_map_lock(mm, pmd, address, | |
2023 | &ptl); | |
2024 | if (!pte_same(*page_table, orig_pte)) { | |
2025 | unlock_page(old_page); | |
2026 | page_cache_release(old_page); | |
2027 | goto unlock; | |
2028 | } | |
2029 | page_cache_release(old_page); | |
ee6a6457 | 2030 | } |
7b1fe597 | 2031 | reuse = reuse_swap_page(old_page); |
ab967d86 | 2032 | unlock_page(old_page); |
ee6a6457 | 2033 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
d08b3851 | 2034 | (VM_WRITE|VM_SHARED))) { |
ee6a6457 PZ |
2035 | /* |
2036 | * Only catch write-faults on shared writable pages, | |
2037 | * read-only shared pages can get COWed by | |
2038 | * get_user_pages(.write=1, .force=1). | |
2039 | */ | |
9637a5ef | 2040 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
c2ec175c NP |
2041 | struct vm_fault vmf; |
2042 | int tmp; | |
2043 | ||
2044 | vmf.virtual_address = (void __user *)(address & | |
2045 | PAGE_MASK); | |
2046 | vmf.pgoff = old_page->index; | |
2047 | vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; | |
2048 | vmf.page = old_page; | |
2049 | ||
9637a5ef DH |
2050 | /* |
2051 | * Notify the address space that the page is about to | |
2052 | * become writable so that it can prohibit this or wait | |
2053 | * for the page to get into an appropriate state. | |
2054 | * | |
2055 | * We do this without the lock held, so that it can | |
2056 | * sleep if it needs to. | |
2057 | */ | |
2058 | page_cache_get(old_page); | |
2059 | pte_unmap_unlock(page_table, ptl); | |
2060 | ||
c2ec175c NP |
2061 | tmp = vma->vm_ops->page_mkwrite(vma, &vmf); |
2062 | if (unlikely(tmp & | |
2063 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { | |
2064 | ret = tmp; | |
9637a5ef | 2065 | goto unwritable_page; |
c2ec175c | 2066 | } |
b827e496 NP |
2067 | if (unlikely(!(tmp & VM_FAULT_LOCKED))) { |
2068 | lock_page(old_page); | |
2069 | if (!old_page->mapping) { | |
2070 | ret = 0; /* retry the fault */ | |
2071 | unlock_page(old_page); | |
2072 | goto unwritable_page; | |
2073 | } | |
2074 | } else | |
2075 | VM_BUG_ON(!PageLocked(old_page)); | |
9637a5ef | 2076 | |
9637a5ef DH |
2077 | /* |
2078 | * Since we dropped the lock we need to revalidate | |
2079 | * the PTE as someone else may have changed it. If | |
2080 | * they did, we just return, as we can count on the | |
2081 | * MMU to tell us if they didn't also make it writable. | |
2082 | */ | |
2083 | page_table = pte_offset_map_lock(mm, pmd, address, | |
2084 | &ptl); | |
b827e496 NP |
2085 | if (!pte_same(*page_table, orig_pte)) { |
2086 | unlock_page(old_page); | |
2087 | page_cache_release(old_page); | |
9637a5ef | 2088 | goto unlock; |
b827e496 | 2089 | } |
a200ee18 PZ |
2090 | |
2091 | page_mkwrite = 1; | |
1da177e4 | 2092 | } |
d08b3851 PZ |
2093 | dirty_page = old_page; |
2094 | get_page(dirty_page); | |
9637a5ef | 2095 | reuse = 1; |
9637a5ef DH |
2096 | } |
2097 | ||
2098 | if (reuse) { | |
251b97f5 | 2099 | reuse: |
9637a5ef DH |
2100 | flush_cache_page(vma, address, pte_pfn(orig_pte)); |
2101 | entry = pte_mkyoung(orig_pte); | |
2102 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
954ffcb3 | 2103 | if (ptep_set_access_flags(vma, address, page_table, entry,1)) |
8dab5241 | 2104 | update_mmu_cache(vma, address, entry); |
9637a5ef DH |
2105 | ret |= VM_FAULT_WRITE; |
2106 | goto unlock; | |
1da177e4 | 2107 | } |
1da177e4 LT |
2108 | |
2109 | /* | |
2110 | * Ok, we need to copy. Oh, well.. | |
2111 | */ | |
b5810039 | 2112 | page_cache_get(old_page); |
920fc356 | 2113 | gotten: |
8f4e2101 | 2114 | pte_unmap_unlock(page_table, ptl); |
1da177e4 LT |
2115 | |
2116 | if (unlikely(anon_vma_prepare(vma))) | |
65500d23 | 2117 | goto oom; |
a13ea5b7 | 2118 | |
62eede62 | 2119 | if (is_zero_pfn(pte_pfn(orig_pte))) { |
a13ea5b7 HD |
2120 | new_page = alloc_zeroed_user_highpage_movable(vma, address); |
2121 | if (!new_page) | |
2122 | goto oom; | |
2123 | } else { | |
2124 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); | |
2125 | if (!new_page) | |
2126 | goto oom; | |
2127 | cow_user_page(new_page, old_page, address, vma); | |
2128 | } | |
2129 | __SetPageUptodate(new_page); | |
2130 | ||
b291f000 NP |
2131 | /* |
2132 | * Don't let another task, with possibly unlocked vma, | |
2133 | * keep the mlocked page. | |
2134 | */ | |
ab92661d | 2135 | if ((vma->vm_flags & VM_LOCKED) && old_page) { |
b291f000 NP |
2136 | lock_page(old_page); /* for LRU manipulation */ |
2137 | clear_page_mlock(old_page); | |
2138 | unlock_page(old_page); | |
2139 | } | |
65500d23 | 2140 | |
2c26fdd7 | 2141 | if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL)) |
8a9f3ccd BS |
2142 | goto oom_free_new; |
2143 | ||
1da177e4 LT |
2144 | /* |
2145 | * Re-check the pte - we dropped the lock | |
2146 | */ | |
8f4e2101 | 2147 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
65500d23 | 2148 | if (likely(pte_same(*page_table, orig_pte))) { |
920fc356 | 2149 | if (old_page) { |
920fc356 HD |
2150 | if (!PageAnon(old_page)) { |
2151 | dec_mm_counter(mm, file_rss); | |
2152 | inc_mm_counter(mm, anon_rss); | |
2153 | } | |
2154 | } else | |
4294621f | 2155 | inc_mm_counter(mm, anon_rss); |
eca35133 | 2156 | flush_cache_page(vma, address, pte_pfn(orig_pte)); |
65500d23 HD |
2157 | entry = mk_pte(new_page, vma->vm_page_prot); |
2158 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
4ce072f1 SS |
2159 | /* |
2160 | * Clear the pte entry and flush it first, before updating the | |
2161 | * pte with the new entry. This will avoid a race condition | |
2162 | * seen in the presence of one thread doing SMC and another | |
2163 | * thread doing COW. | |
2164 | */ | |
828502d3 | 2165 | ptep_clear_flush(vma, address, page_table); |
9617d95e | 2166 | page_add_new_anon_rmap(new_page, vma, address); |
828502d3 IE |
2167 | /* |
2168 | * We call the notify macro here because, when using secondary | |
2169 | * mmu page tables (such as kvm shadow page tables), we want the | |
2170 | * new page to be mapped directly into the secondary page table. | |
2171 | */ | |
2172 | set_pte_at_notify(mm, address, page_table, entry); | |
64d6519d | 2173 | update_mmu_cache(vma, address, entry); |
945754a1 NP |
2174 | if (old_page) { |
2175 | /* | |
2176 | * Only after switching the pte to the new page may | |
2177 | * we remove the mapcount here. Otherwise another | |
2178 | * process may come and find the rmap count decremented | |
2179 | * before the pte is switched to the new page, and | |
2180 | * "reuse" the old page writing into it while our pte | |
2181 | * here still points into it and can be read by other | |
2182 | * threads. | |
2183 | * | |
2184 | * The critical issue is to order this | |
2185 | * page_remove_rmap with the ptp_clear_flush above. | |
2186 | * Those stores are ordered by (if nothing else,) | |
2187 | * the barrier present in the atomic_add_negative | |
2188 | * in page_remove_rmap. | |
2189 | * | |
2190 | * Then the TLB flush in ptep_clear_flush ensures that | |
2191 | * no process can access the old page before the | |
2192 | * decremented mapcount is visible. And the old page | |
2193 | * cannot be reused until after the decremented | |
2194 | * mapcount is visible. So transitively, TLBs to | |
2195 | * old page will be flushed before it can be reused. | |
2196 | */ | |
edc315fd | 2197 | page_remove_rmap(old_page); |
945754a1 NP |
2198 | } |
2199 | ||
1da177e4 LT |
2200 | /* Free the old page.. */ |
2201 | new_page = old_page; | |
f33ea7f4 | 2202 | ret |= VM_FAULT_WRITE; |
8a9f3ccd BS |
2203 | } else |
2204 | mem_cgroup_uncharge_page(new_page); | |
2205 | ||
920fc356 HD |
2206 | if (new_page) |
2207 | page_cache_release(new_page); | |
2208 | if (old_page) | |
2209 | page_cache_release(old_page); | |
65500d23 | 2210 | unlock: |
8f4e2101 | 2211 | pte_unmap_unlock(page_table, ptl); |
d08b3851 | 2212 | if (dirty_page) { |
79352894 NP |
2213 | /* |
2214 | * Yes, Virginia, this is actually required to prevent a race | |
2215 | * with clear_page_dirty_for_io() from clearing the page dirty | |
2216 | * bit after it clear all dirty ptes, but before a racing | |
2217 | * do_wp_page installs a dirty pte. | |
2218 | * | |
2219 | * do_no_page is protected similarly. | |
2220 | */ | |
b827e496 NP |
2221 | if (!page_mkwrite) { |
2222 | wait_on_page_locked(dirty_page); | |
2223 | set_page_dirty_balance(dirty_page, page_mkwrite); | |
2224 | } | |
d08b3851 | 2225 | put_page(dirty_page); |
b827e496 NP |
2226 | if (page_mkwrite) { |
2227 | struct address_space *mapping = dirty_page->mapping; | |
2228 | ||
2229 | set_page_dirty(dirty_page); | |
2230 | unlock_page(dirty_page); | |
2231 | page_cache_release(dirty_page); | |
2232 | if (mapping) { | |
2233 | /* | |
2234 | * Some device drivers do not set page.mapping | |
2235 | * but still dirty their pages | |
2236 | */ | |
2237 | balance_dirty_pages_ratelimited(mapping); | |
2238 | } | |
2239 | } | |
2240 | ||
2241 | /* file_update_time outside page_lock */ | |
2242 | if (vma->vm_file) | |
2243 | file_update_time(vma->vm_file); | |
d08b3851 | 2244 | } |
f33ea7f4 | 2245 | return ret; |
8a9f3ccd | 2246 | oom_free_new: |
6dbf6d3b | 2247 | page_cache_release(new_page); |
65500d23 | 2248 | oom: |
b827e496 NP |
2249 | if (old_page) { |
2250 | if (page_mkwrite) { | |
2251 | unlock_page(old_page); | |
2252 | page_cache_release(old_page); | |
2253 | } | |
920fc356 | 2254 | page_cache_release(old_page); |
b827e496 | 2255 | } |
1da177e4 | 2256 | return VM_FAULT_OOM; |
9637a5ef DH |
2257 | |
2258 | unwritable_page: | |
2259 | page_cache_release(old_page); | |
c2ec175c | 2260 | return ret; |
1da177e4 LT |
2261 | } |
2262 | ||
2263 | /* | |
2264 | * Helper functions for unmap_mapping_range(). | |
2265 | * | |
2266 | * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __ | |
2267 | * | |
2268 | * We have to restart searching the prio_tree whenever we drop the lock, | |
2269 | * since the iterator is only valid while the lock is held, and anyway | |
2270 | * a later vma might be split and reinserted earlier while lock dropped. | |
2271 | * | |
2272 | * The list of nonlinear vmas could be handled more efficiently, using | |
2273 | * a placeholder, but handle it in the same way until a need is shown. | |
2274 | * It is important to search the prio_tree before nonlinear list: a vma | |
2275 | * may become nonlinear and be shifted from prio_tree to nonlinear list | |
2276 | * while the lock is dropped; but never shifted from list to prio_tree. | |
2277 | * | |
2278 | * In order to make forward progress despite restarting the search, | |
2279 | * vm_truncate_count is used to mark a vma as now dealt with, so we can | |
2280 | * quickly skip it next time around. Since the prio_tree search only | |
2281 | * shows us those vmas affected by unmapping the range in question, we | |
2282 | * can't efficiently keep all vmas in step with mapping->truncate_count: | |
2283 | * so instead reset them all whenever it wraps back to 0 (then go to 1). | |
2284 | * mapping->truncate_count and vma->vm_truncate_count are protected by | |
2285 | * i_mmap_lock. | |
2286 | * | |
2287 | * In order to make forward progress despite repeatedly restarting some | |
ee39b37b | 2288 | * large vma, note the restart_addr from unmap_vmas when it breaks out: |
1da177e4 LT |
2289 | * and restart from that address when we reach that vma again. It might |
2290 | * have been split or merged, shrunk or extended, but never shifted: so | |
2291 | * restart_addr remains valid so long as it remains in the vma's range. | |
2292 | * unmap_mapping_range forces truncate_count to leap over page-aligned | |
2293 | * values so we can save vma's restart_addr in its truncate_count field. | |
2294 | */ | |
2295 | #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK)) | |
2296 | ||
2297 | static void reset_vma_truncate_counts(struct address_space *mapping) | |
2298 | { | |
2299 | struct vm_area_struct *vma; | |
2300 | struct prio_tree_iter iter; | |
2301 | ||
2302 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX) | |
2303 | vma->vm_truncate_count = 0; | |
2304 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) | |
2305 | vma->vm_truncate_count = 0; | |
2306 | } | |
2307 | ||
2308 | static int unmap_mapping_range_vma(struct vm_area_struct *vma, | |
2309 | unsigned long start_addr, unsigned long end_addr, | |
2310 | struct zap_details *details) | |
2311 | { | |
2312 | unsigned long restart_addr; | |
2313 | int need_break; | |
2314 | ||
d00806b1 NP |
2315 | /* |
2316 | * files that support invalidating or truncating portions of the | |
d0217ac0 | 2317 | * file from under mmaped areas must have their ->fault function |
83c54070 NP |
2318 | * return a locked page (and set VM_FAULT_LOCKED in the return). |
2319 | * This provides synchronisation against concurrent unmapping here. | |
d00806b1 | 2320 | */ |
d00806b1 | 2321 | |
1da177e4 LT |
2322 | again: |
2323 | restart_addr = vma->vm_truncate_count; | |
2324 | if (is_restart_addr(restart_addr) && start_addr < restart_addr) { | |
2325 | start_addr = restart_addr; | |
2326 | if (start_addr >= end_addr) { | |
2327 | /* Top of vma has been split off since last time */ | |
2328 | vma->vm_truncate_count = details->truncate_count; | |
2329 | return 0; | |
2330 | } | |
2331 | } | |
2332 | ||
ee39b37b HD |
2333 | restart_addr = zap_page_range(vma, start_addr, |
2334 | end_addr - start_addr, details); | |
95c354fe | 2335 | need_break = need_resched() || spin_needbreak(details->i_mmap_lock); |
1da177e4 | 2336 | |
ee39b37b | 2337 | if (restart_addr >= end_addr) { |
1da177e4 LT |
2338 | /* We have now completed this vma: mark it so */ |
2339 | vma->vm_truncate_count = details->truncate_count; | |
2340 | if (!need_break) | |
2341 | return 0; | |
2342 | } else { | |
2343 | /* Note restart_addr in vma's truncate_count field */ | |
ee39b37b | 2344 | vma->vm_truncate_count = restart_addr; |
1da177e4 LT |
2345 | if (!need_break) |
2346 | goto again; | |
2347 | } | |
2348 | ||
2349 | spin_unlock(details->i_mmap_lock); | |
2350 | cond_resched(); | |
2351 | spin_lock(details->i_mmap_lock); | |
2352 | return -EINTR; | |
2353 | } | |
2354 | ||
2355 | static inline void unmap_mapping_range_tree(struct prio_tree_root *root, | |
2356 | struct zap_details *details) | |
2357 | { | |
2358 | struct vm_area_struct *vma; | |
2359 | struct prio_tree_iter iter; | |
2360 | pgoff_t vba, vea, zba, zea; | |
2361 | ||
2362 | restart: | |
2363 | vma_prio_tree_foreach(vma, &iter, root, | |
2364 | details->first_index, details->last_index) { | |
2365 | /* Skip quickly over those we have already dealt with */ | |
2366 | if (vma->vm_truncate_count == details->truncate_count) | |
2367 | continue; | |
2368 | ||
2369 | vba = vma->vm_pgoff; | |
2370 | vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1; | |
2371 | /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */ | |
2372 | zba = details->first_index; | |
2373 | if (zba < vba) | |
2374 | zba = vba; | |
2375 | zea = details->last_index; | |
2376 | if (zea > vea) | |
2377 | zea = vea; | |
2378 | ||
2379 | if (unmap_mapping_range_vma(vma, | |
2380 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, | |
2381 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, | |
2382 | details) < 0) | |
2383 | goto restart; | |
2384 | } | |
2385 | } | |
2386 | ||
2387 | static inline void unmap_mapping_range_list(struct list_head *head, | |
2388 | struct zap_details *details) | |
2389 | { | |
2390 | struct vm_area_struct *vma; | |
2391 | ||
2392 | /* | |
2393 | * In nonlinear VMAs there is no correspondence between virtual address | |
2394 | * offset and file offset. So we must perform an exhaustive search | |
2395 | * across *all* the pages in each nonlinear VMA, not just the pages | |
2396 | * whose virtual address lies outside the file truncation point. | |
2397 | */ | |
2398 | restart: | |
2399 | list_for_each_entry(vma, head, shared.vm_set.list) { | |
2400 | /* Skip quickly over those we have already dealt with */ | |
2401 | if (vma->vm_truncate_count == details->truncate_count) | |
2402 | continue; | |
2403 | details->nonlinear_vma = vma; | |
2404 | if (unmap_mapping_range_vma(vma, vma->vm_start, | |
2405 | vma->vm_end, details) < 0) | |
2406 | goto restart; | |
2407 | } | |
2408 | } | |
2409 | ||
2410 | /** | |
72fd4a35 | 2411 | * 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 | 2412 | * @mapping: the address space containing mmaps to be unmapped. |
1da177e4 LT |
2413 | * @holebegin: byte in first page to unmap, relative to the start of |
2414 | * the underlying file. This will be rounded down to a PAGE_SIZE | |
25d9e2d1 | 2415 | * boundary. Note that this is different from truncate_pagecache(), which |
1da177e4 LT |
2416 | * must keep the partial page. In contrast, we must get rid of |
2417 | * partial pages. | |
2418 | * @holelen: size of prospective hole in bytes. This will be rounded | |
2419 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the | |
2420 | * end of the file. | |
2421 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; | |
2422 | * but 0 when invalidating pagecache, don't throw away private data. | |
2423 | */ | |
2424 | void unmap_mapping_range(struct address_space *mapping, | |
2425 | loff_t const holebegin, loff_t const holelen, int even_cows) | |
2426 | { | |
2427 | struct zap_details details; | |
2428 | pgoff_t hba = holebegin >> PAGE_SHIFT; | |
2429 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2430 | ||
2431 | /* Check for overflow. */ | |
2432 | if (sizeof(holelen) > sizeof(hlen)) { | |
2433 | long long holeend = | |
2434 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2435 | if (holeend & ~(long long)ULONG_MAX) | |
2436 | hlen = ULONG_MAX - hba + 1; | |
2437 | } | |
2438 | ||
2439 | details.check_mapping = even_cows? NULL: mapping; | |
2440 | details.nonlinear_vma = NULL; | |
2441 | details.first_index = hba; | |
2442 | details.last_index = hba + hlen - 1; | |
2443 | if (details.last_index < details.first_index) | |
2444 | details.last_index = ULONG_MAX; | |
2445 | details.i_mmap_lock = &mapping->i_mmap_lock; | |
2446 | ||
2447 | spin_lock(&mapping->i_mmap_lock); | |
2448 | ||
d00806b1 | 2449 | /* Protect against endless unmapping loops */ |
1da177e4 | 2450 | mapping->truncate_count++; |
1da177e4 LT |
2451 | if (unlikely(is_restart_addr(mapping->truncate_count))) { |
2452 | if (mapping->truncate_count == 0) | |
2453 | reset_vma_truncate_counts(mapping); | |
2454 | mapping->truncate_count++; | |
2455 | } | |
2456 | details.truncate_count = mapping->truncate_count; | |
2457 | ||
2458 | if (unlikely(!prio_tree_empty(&mapping->i_mmap))) | |
2459 | unmap_mapping_range_tree(&mapping->i_mmap, &details); | |
2460 | if (unlikely(!list_empty(&mapping->i_mmap_nonlinear))) | |
2461 | unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details); | |
2462 | spin_unlock(&mapping->i_mmap_lock); | |
2463 | } | |
2464 | EXPORT_SYMBOL(unmap_mapping_range); | |
2465 | ||
f6b3ec23 BP |
2466 | int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end) |
2467 | { | |
2468 | struct address_space *mapping = inode->i_mapping; | |
2469 | ||
2470 | /* | |
2471 | * If the underlying filesystem is not going to provide | |
2472 | * a way to truncate a range of blocks (punch a hole) - | |
2473 | * we should return failure right now. | |
2474 | */ | |
acfa4380 | 2475 | if (!inode->i_op->truncate_range) |
f6b3ec23 BP |
2476 | return -ENOSYS; |
2477 | ||
1b1dcc1b | 2478 | mutex_lock(&inode->i_mutex); |
f6b3ec23 BP |
2479 | down_write(&inode->i_alloc_sem); |
2480 | unmap_mapping_range(mapping, offset, (end - offset), 1); | |
2481 | truncate_inode_pages_range(mapping, offset, end); | |
d00806b1 | 2482 | unmap_mapping_range(mapping, offset, (end - offset), 1); |
f6b3ec23 BP |
2483 | inode->i_op->truncate_range(inode, offset, end); |
2484 | up_write(&inode->i_alloc_sem); | |
1b1dcc1b | 2485 | mutex_unlock(&inode->i_mutex); |
f6b3ec23 BP |
2486 | |
2487 | return 0; | |
2488 | } | |
f6b3ec23 | 2489 | |
1da177e4 | 2490 | /* |
8f4e2101 HD |
2491 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2492 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2493 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2494 | */ |
65500d23 HD |
2495 | static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2496 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
30c9f3a9 | 2497 | unsigned int flags, pte_t orig_pte) |
1da177e4 | 2498 | { |
8f4e2101 | 2499 | spinlock_t *ptl; |
1da177e4 | 2500 | struct page *page; |
65500d23 | 2501 | swp_entry_t entry; |
1da177e4 | 2502 | pte_t pte; |
7a81b88c | 2503 | struct mem_cgroup *ptr = NULL; |
83c54070 | 2504 | int ret = 0; |
1da177e4 | 2505 | |
4c21e2f2 | 2506 | if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) |
8f4e2101 | 2507 | goto out; |
65500d23 HD |
2508 | |
2509 | entry = pte_to_swp_entry(orig_pte); | |
d1737fdb AK |
2510 | if (unlikely(non_swap_entry(entry))) { |
2511 | if (is_migration_entry(entry)) { | |
2512 | migration_entry_wait(mm, pmd, address); | |
2513 | } else if (is_hwpoison_entry(entry)) { | |
2514 | ret = VM_FAULT_HWPOISON; | |
2515 | } else { | |
2516 | print_bad_pte(vma, address, orig_pte, NULL); | |
2517 | ret = VM_FAULT_OOM; | |
2518 | } | |
0697212a CL |
2519 | goto out; |
2520 | } | |
0ff92245 | 2521 | delayacct_set_flag(DELAYACCT_PF_SWAPIN); |
1da177e4 LT |
2522 | page = lookup_swap_cache(entry); |
2523 | if (!page) { | |
a5c9b696 | 2524 | grab_swap_token(mm); /* Contend for token _before_ read-in */ |
02098fea HD |
2525 | page = swapin_readahead(entry, |
2526 | GFP_HIGHUSER_MOVABLE, vma, address); | |
1da177e4 LT |
2527 | if (!page) { |
2528 | /* | |
8f4e2101 HD |
2529 | * Back out if somebody else faulted in this pte |
2530 | * while we released the pte lock. | |
1da177e4 | 2531 | */ |
8f4e2101 | 2532 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
1da177e4 LT |
2533 | if (likely(pte_same(*page_table, orig_pte))) |
2534 | ret = VM_FAULT_OOM; | |
0ff92245 | 2535 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
65500d23 | 2536 | goto unlock; |
1da177e4 LT |
2537 | } |
2538 | ||
2539 | /* Had to read the page from swap area: Major fault */ | |
2540 | ret = VM_FAULT_MAJOR; | |
f8891e5e | 2541 | count_vm_event(PGMAJFAULT); |
d1737fdb AK |
2542 | } else if (PageHWPoison(page)) { |
2543 | ret = VM_FAULT_HWPOISON; | |
2544 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); | |
4779cb31 | 2545 | goto out_release; |
1da177e4 LT |
2546 | } |
2547 | ||
073e587e KH |
2548 | lock_page(page); |
2549 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); | |
2550 | ||
2c26fdd7 | 2551 | if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) { |
8a9f3ccd | 2552 | ret = VM_FAULT_OOM; |
bc43f75c | 2553 | goto out_page; |
8a9f3ccd BS |
2554 | } |
2555 | ||
1da177e4 | 2556 | /* |
8f4e2101 | 2557 | * Back out if somebody else already faulted in this pte. |
1da177e4 | 2558 | */ |
8f4e2101 | 2559 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
9e9bef07 | 2560 | if (unlikely(!pte_same(*page_table, orig_pte))) |
b8107480 | 2561 | goto out_nomap; |
b8107480 KK |
2562 | |
2563 | if (unlikely(!PageUptodate(page))) { | |
2564 | ret = VM_FAULT_SIGBUS; | |
2565 | goto out_nomap; | |
1da177e4 LT |
2566 | } |
2567 | ||
8c7c6e34 KH |
2568 | /* |
2569 | * The page isn't present yet, go ahead with the fault. | |
2570 | * | |
2571 | * Be careful about the sequence of operations here. | |
2572 | * To get its accounting right, reuse_swap_page() must be called | |
2573 | * while the page is counted on swap but not yet in mapcount i.e. | |
2574 | * before page_add_anon_rmap() and swap_free(); try_to_free_swap() | |
2575 | * must be called after the swap_free(), or it will never succeed. | |
03f3c433 KH |
2576 | * Because delete_from_swap_page() may be called by reuse_swap_page(), |
2577 | * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry | |
2578 | * in page->private. In this case, a record in swap_cgroup is silently | |
2579 | * discarded at swap_free(). | |
8c7c6e34 | 2580 | */ |
1da177e4 | 2581 | |
4294621f | 2582 | inc_mm_counter(mm, anon_rss); |
1da177e4 | 2583 | pte = mk_pte(page, vma->vm_page_prot); |
30c9f3a9 | 2584 | if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) { |
1da177e4 | 2585 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); |
30c9f3a9 | 2586 | flags &= ~FAULT_FLAG_WRITE; |
1da177e4 | 2587 | } |
1da177e4 LT |
2588 | flush_icache_page(vma, page); |
2589 | set_pte_at(mm, address, page_table, pte); | |
2590 | page_add_anon_rmap(page, vma, address); | |
03f3c433 KH |
2591 | /* It's better to call commit-charge after rmap is established */ |
2592 | mem_cgroup_commit_charge_swapin(page, ptr); | |
1da177e4 | 2593 | |
c475a8ab | 2594 | swap_free(entry); |
b291f000 | 2595 | if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) |
a2c43eed | 2596 | try_to_free_swap(page); |
c475a8ab HD |
2597 | unlock_page(page); |
2598 | ||
30c9f3a9 | 2599 | if (flags & FAULT_FLAG_WRITE) { |
61469f1d HD |
2600 | ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte); |
2601 | if (ret & VM_FAULT_ERROR) | |
2602 | ret &= VM_FAULT_ERROR; | |
1da177e4 LT |
2603 | goto out; |
2604 | } | |
2605 | ||
2606 | /* No need to invalidate - it was non-present before */ | |
2607 | update_mmu_cache(vma, address, pte); | |
65500d23 | 2608 | unlock: |
8f4e2101 | 2609 | pte_unmap_unlock(page_table, ptl); |
1da177e4 LT |
2610 | out: |
2611 | return ret; | |
b8107480 | 2612 | out_nomap: |
7a81b88c | 2613 | mem_cgroup_cancel_charge_swapin(ptr); |
8f4e2101 | 2614 | pte_unmap_unlock(page_table, ptl); |
bc43f75c | 2615 | out_page: |
b8107480 | 2616 | unlock_page(page); |
4779cb31 | 2617 | out_release: |
b8107480 | 2618 | page_cache_release(page); |
65500d23 | 2619 | return ret; |
1da177e4 LT |
2620 | } |
2621 | ||
2622 | /* | |
8f4e2101 HD |
2623 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2624 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2625 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2626 | */ |
65500d23 HD |
2627 | static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2628 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
30c9f3a9 | 2629 | unsigned int flags) |
1da177e4 | 2630 | { |
8f4e2101 HD |
2631 | struct page *page; |
2632 | spinlock_t *ptl; | |
1da177e4 | 2633 | pte_t entry; |
1da177e4 | 2634 | |
62eede62 HD |
2635 | if (!(flags & FAULT_FLAG_WRITE)) { |
2636 | entry = pte_mkspecial(pfn_pte(my_zero_pfn(address), | |
2637 | vma->vm_page_prot)); | |
a13ea5b7 HD |
2638 | ptl = pte_lockptr(mm, pmd); |
2639 | spin_lock(ptl); | |
2640 | if (!pte_none(*page_table)) | |
2641 | goto unlock; | |
2642 | goto setpte; | |
2643 | } | |
2644 | ||
557ed1fa NP |
2645 | /* Allocate our own private page. */ |
2646 | pte_unmap(page_table); | |
8f4e2101 | 2647 | |
557ed1fa NP |
2648 | if (unlikely(anon_vma_prepare(vma))) |
2649 | goto oom; | |
2650 | page = alloc_zeroed_user_highpage_movable(vma, address); | |
2651 | if (!page) | |
2652 | goto oom; | |
0ed361de | 2653 | __SetPageUptodate(page); |
8f4e2101 | 2654 | |
2c26fdd7 | 2655 | if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) |
8a9f3ccd BS |
2656 | goto oom_free_page; |
2657 | ||
557ed1fa | 2658 | entry = mk_pte(page, vma->vm_page_prot); |
1ac0cb5d HD |
2659 | if (vma->vm_flags & VM_WRITE) |
2660 | entry = pte_mkwrite(pte_mkdirty(entry)); | |
1da177e4 | 2661 | |
557ed1fa | 2662 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
1c2fb7a4 | 2663 | if (!pte_none(*page_table)) |
557ed1fa | 2664 | goto release; |
9ba69294 | 2665 | |
557ed1fa | 2666 | inc_mm_counter(mm, anon_rss); |
557ed1fa | 2667 | page_add_new_anon_rmap(page, vma, address); |
a13ea5b7 | 2668 | setpte: |
65500d23 | 2669 | set_pte_at(mm, address, page_table, entry); |
1da177e4 LT |
2670 | |
2671 | /* No need to invalidate - it was non-present before */ | |
65500d23 | 2672 | update_mmu_cache(vma, address, entry); |
65500d23 | 2673 | unlock: |
8f4e2101 | 2674 | pte_unmap_unlock(page_table, ptl); |
83c54070 | 2675 | return 0; |
8f4e2101 | 2676 | release: |
8a9f3ccd | 2677 | mem_cgroup_uncharge_page(page); |
8f4e2101 HD |
2678 | page_cache_release(page); |
2679 | goto unlock; | |
8a9f3ccd | 2680 | oom_free_page: |
6dbf6d3b | 2681 | page_cache_release(page); |
65500d23 | 2682 | oom: |
1da177e4 LT |
2683 | return VM_FAULT_OOM; |
2684 | } | |
2685 | ||
2686 | /* | |
54cb8821 | 2687 | * __do_fault() tries to create a new page mapping. It aggressively |
1da177e4 | 2688 | * tries to share with existing pages, but makes a separate copy if |
54cb8821 NP |
2689 | * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid |
2690 | * the next page fault. | |
1da177e4 LT |
2691 | * |
2692 | * As this is called only for pages that do not currently exist, we | |
2693 | * do not need to flush old virtual caches or the TLB. | |
2694 | * | |
8f4e2101 | 2695 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
16abfa08 | 2696 | * but allow concurrent faults), and pte neither mapped nor locked. |
8f4e2101 | 2697 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
1da177e4 | 2698 | */ |
54cb8821 | 2699 | static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
16abfa08 | 2700 | unsigned long address, pmd_t *pmd, |
54cb8821 | 2701 | pgoff_t pgoff, unsigned int flags, pte_t orig_pte) |
1da177e4 | 2702 | { |
16abfa08 | 2703 | pte_t *page_table; |
8f4e2101 | 2704 | spinlock_t *ptl; |
d0217ac0 | 2705 | struct page *page; |
1da177e4 | 2706 | pte_t entry; |
1da177e4 | 2707 | int anon = 0; |
5b4e655e | 2708 | int charged = 0; |
d08b3851 | 2709 | struct page *dirty_page = NULL; |
d0217ac0 NP |
2710 | struct vm_fault vmf; |
2711 | int ret; | |
a200ee18 | 2712 | int page_mkwrite = 0; |
54cb8821 | 2713 | |
d0217ac0 NP |
2714 | vmf.virtual_address = (void __user *)(address & PAGE_MASK); |
2715 | vmf.pgoff = pgoff; | |
2716 | vmf.flags = flags; | |
2717 | vmf.page = NULL; | |
1da177e4 | 2718 | |
3c18ddd1 NP |
2719 | ret = vma->vm_ops->fault(vma, &vmf); |
2720 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) | |
2721 | return ret; | |
1da177e4 | 2722 | |
a3b947ea AK |
2723 | if (unlikely(PageHWPoison(vmf.page))) { |
2724 | if (ret & VM_FAULT_LOCKED) | |
2725 | unlock_page(vmf.page); | |
2726 | return VM_FAULT_HWPOISON; | |
2727 | } | |
2728 | ||
d00806b1 | 2729 | /* |
d0217ac0 | 2730 | * For consistency in subsequent calls, make the faulted page always |
d00806b1 NP |
2731 | * locked. |
2732 | */ | |
83c54070 | 2733 | if (unlikely(!(ret & VM_FAULT_LOCKED))) |
d0217ac0 | 2734 | lock_page(vmf.page); |
54cb8821 | 2735 | else |
d0217ac0 | 2736 | VM_BUG_ON(!PageLocked(vmf.page)); |
d00806b1 | 2737 | |
1da177e4 LT |
2738 | /* |
2739 | * Should we do an early C-O-W break? | |
2740 | */ | |
d0217ac0 | 2741 | page = vmf.page; |
54cb8821 | 2742 | if (flags & FAULT_FLAG_WRITE) { |
9637a5ef | 2743 | if (!(vma->vm_flags & VM_SHARED)) { |
54cb8821 | 2744 | anon = 1; |
d00806b1 | 2745 | if (unlikely(anon_vma_prepare(vma))) { |
d0217ac0 | 2746 | ret = VM_FAULT_OOM; |
54cb8821 | 2747 | goto out; |
d00806b1 | 2748 | } |
83c54070 NP |
2749 | page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, |
2750 | vma, address); | |
d00806b1 | 2751 | if (!page) { |
d0217ac0 | 2752 | ret = VM_FAULT_OOM; |
54cb8821 | 2753 | goto out; |
d00806b1 | 2754 | } |
2c26fdd7 | 2755 | if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) { |
5b4e655e KH |
2756 | ret = VM_FAULT_OOM; |
2757 | page_cache_release(page); | |
2758 | goto out; | |
2759 | } | |
2760 | charged = 1; | |
b291f000 NP |
2761 | /* |
2762 | * Don't let another task, with possibly unlocked vma, | |
2763 | * keep the mlocked page. | |
2764 | */ | |
2765 | if (vma->vm_flags & VM_LOCKED) | |
2766 | clear_page_mlock(vmf.page); | |
d0217ac0 | 2767 | copy_user_highpage(page, vmf.page, address, vma); |
0ed361de | 2768 | __SetPageUptodate(page); |
9637a5ef | 2769 | } else { |
54cb8821 NP |
2770 | /* |
2771 | * If the page will be shareable, see if the backing | |
9637a5ef | 2772 | * address space wants to know that the page is about |
54cb8821 NP |
2773 | * to become writable |
2774 | */ | |
69676147 | 2775 | if (vma->vm_ops->page_mkwrite) { |
c2ec175c NP |
2776 | int tmp; |
2777 | ||
69676147 | 2778 | unlock_page(page); |
b827e496 | 2779 | vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; |
c2ec175c NP |
2780 | tmp = vma->vm_ops->page_mkwrite(vma, &vmf); |
2781 | if (unlikely(tmp & | |
2782 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { | |
2783 | ret = tmp; | |
b827e496 | 2784 | goto unwritable_page; |
d0217ac0 | 2785 | } |
b827e496 NP |
2786 | if (unlikely(!(tmp & VM_FAULT_LOCKED))) { |
2787 | lock_page(page); | |
2788 | if (!page->mapping) { | |
2789 | ret = 0; /* retry the fault */ | |
2790 | unlock_page(page); | |
2791 | goto unwritable_page; | |
2792 | } | |
2793 | } else | |
2794 | VM_BUG_ON(!PageLocked(page)); | |
a200ee18 | 2795 | page_mkwrite = 1; |
9637a5ef DH |
2796 | } |
2797 | } | |
54cb8821 | 2798 | |
1da177e4 LT |
2799 | } |
2800 | ||
8f4e2101 | 2801 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
1da177e4 LT |
2802 | |
2803 | /* | |
2804 | * This silly early PAGE_DIRTY setting removes a race | |
2805 | * due to the bad i386 page protection. But it's valid | |
2806 | * for other architectures too. | |
2807 | * | |
30c9f3a9 | 2808 | * Note that if FAULT_FLAG_WRITE is set, we either now have |
1da177e4 LT |
2809 | * an exclusive copy of the page, or this is a shared mapping, |
2810 | * so we can make it writable and dirty to avoid having to | |
2811 | * handle that later. | |
2812 | */ | |
2813 | /* Only go through if we didn't race with anybody else... */ | |
1c2fb7a4 | 2814 | if (likely(pte_same(*page_table, orig_pte))) { |
d00806b1 NP |
2815 | flush_icache_page(vma, page); |
2816 | entry = mk_pte(page, vma->vm_page_prot); | |
54cb8821 | 2817 | if (flags & FAULT_FLAG_WRITE) |
1da177e4 | 2818 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
1da177e4 | 2819 | if (anon) { |
64d6519d | 2820 | inc_mm_counter(mm, anon_rss); |
64d6519d | 2821 | page_add_new_anon_rmap(page, vma, address); |
f57e88a8 | 2822 | } else { |
4294621f | 2823 | inc_mm_counter(mm, file_rss); |
d00806b1 | 2824 | page_add_file_rmap(page); |
54cb8821 | 2825 | if (flags & FAULT_FLAG_WRITE) { |
d00806b1 | 2826 | dirty_page = page; |
d08b3851 PZ |
2827 | get_page(dirty_page); |
2828 | } | |
4294621f | 2829 | } |
64d6519d | 2830 | set_pte_at(mm, address, page_table, entry); |
d00806b1 NP |
2831 | |
2832 | /* no need to invalidate: a not-present page won't be cached */ | |
2833 | update_mmu_cache(vma, address, entry); | |
1da177e4 | 2834 | } else { |
5b4e655e KH |
2835 | if (charged) |
2836 | mem_cgroup_uncharge_page(page); | |
d00806b1 NP |
2837 | if (anon) |
2838 | page_cache_release(page); | |
2839 | else | |
54cb8821 | 2840 | anon = 1; /* no anon but release faulted_page */ |
1da177e4 LT |
2841 | } |
2842 | ||
8f4e2101 | 2843 | pte_unmap_unlock(page_table, ptl); |
d00806b1 NP |
2844 | |
2845 | out: | |
b827e496 NP |
2846 | if (dirty_page) { |
2847 | struct address_space *mapping = page->mapping; | |
8f7b3d15 | 2848 | |
b827e496 NP |
2849 | if (set_page_dirty(dirty_page)) |
2850 | page_mkwrite = 1; | |
2851 | unlock_page(dirty_page); | |
d08b3851 | 2852 | put_page(dirty_page); |
b827e496 NP |
2853 | if (page_mkwrite && mapping) { |
2854 | /* | |
2855 | * Some device drivers do not set page.mapping but still | |
2856 | * dirty their pages | |
2857 | */ | |
2858 | balance_dirty_pages_ratelimited(mapping); | |
2859 | } | |
2860 | ||
2861 | /* file_update_time outside page_lock */ | |
2862 | if (vma->vm_file) | |
2863 | file_update_time(vma->vm_file); | |
2864 | } else { | |
2865 | unlock_page(vmf.page); | |
2866 | if (anon) | |
2867 | page_cache_release(vmf.page); | |
d08b3851 | 2868 | } |
d00806b1 | 2869 | |
83c54070 | 2870 | return ret; |
b827e496 NP |
2871 | |
2872 | unwritable_page: | |
2873 | page_cache_release(page); | |
2874 | return ret; | |
54cb8821 | 2875 | } |
d00806b1 | 2876 | |
54cb8821 NP |
2877 | static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
2878 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
30c9f3a9 | 2879 | unsigned int flags, pte_t orig_pte) |
54cb8821 NP |
2880 | { |
2881 | pgoff_t pgoff = (((address & PAGE_MASK) | |
0da7e01f | 2882 | - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
54cb8821 | 2883 | |
16abfa08 HD |
2884 | pte_unmap(page_table); |
2885 | return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); | |
54cb8821 NP |
2886 | } |
2887 | ||
1da177e4 LT |
2888 | /* |
2889 | * Fault of a previously existing named mapping. Repopulate the pte | |
2890 | * from the encoded file_pte if possible. This enables swappable | |
2891 | * nonlinear vmas. | |
8f4e2101 HD |
2892 | * |
2893 | * We enter with non-exclusive mmap_sem (to exclude vma changes, | |
2894 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2895 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2896 | */ |
d0217ac0 | 2897 | static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
65500d23 | 2898 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
30c9f3a9 | 2899 | unsigned int flags, pte_t orig_pte) |
1da177e4 | 2900 | { |
65500d23 | 2901 | pgoff_t pgoff; |
1da177e4 | 2902 | |
30c9f3a9 LT |
2903 | flags |= FAULT_FLAG_NONLINEAR; |
2904 | ||
4c21e2f2 | 2905 | if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) |
83c54070 | 2906 | return 0; |
1da177e4 | 2907 | |
2509ef26 | 2908 | if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) { |
65500d23 HD |
2909 | /* |
2910 | * Page table corrupted: show pte and kill process. | |
2911 | */ | |
3dc14741 | 2912 | print_bad_pte(vma, address, orig_pte, NULL); |
65500d23 HD |
2913 | return VM_FAULT_OOM; |
2914 | } | |
65500d23 HD |
2915 | |
2916 | pgoff = pte_to_pgoff(orig_pte); | |
16abfa08 | 2917 | return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); |
1da177e4 LT |
2918 | } |
2919 | ||
2920 | /* | |
2921 | * These routines also need to handle stuff like marking pages dirty | |
2922 | * and/or accessed for architectures that don't do it in hardware (most | |
2923 | * RISC architectures). The early dirtying is also good on the i386. | |
2924 | * | |
2925 | * There is also a hook called "update_mmu_cache()" that architectures | |
2926 | * with external mmu caches can use to update those (ie the Sparc or | |
2927 | * PowerPC hashed page tables that act as extended TLBs). | |
2928 | * | |
c74df32c HD |
2929 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2930 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2931 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 LT |
2932 | */ |
2933 | static inline int handle_pte_fault(struct mm_struct *mm, | |
65500d23 | 2934 | struct vm_area_struct *vma, unsigned long address, |
30c9f3a9 | 2935 | pte_t *pte, pmd_t *pmd, unsigned int flags) |
1da177e4 LT |
2936 | { |
2937 | pte_t entry; | |
8f4e2101 | 2938 | spinlock_t *ptl; |
1da177e4 | 2939 | |
8dab5241 | 2940 | entry = *pte; |
1da177e4 | 2941 | if (!pte_present(entry)) { |
65500d23 | 2942 | if (pte_none(entry)) { |
f4b81804 | 2943 | if (vma->vm_ops) { |
3c18ddd1 | 2944 | if (likely(vma->vm_ops->fault)) |
54cb8821 | 2945 | return do_linear_fault(mm, vma, address, |
30c9f3a9 | 2946 | pte, pmd, flags, entry); |
f4b81804 JS |
2947 | } |
2948 | return do_anonymous_page(mm, vma, address, | |
30c9f3a9 | 2949 | pte, pmd, flags); |
65500d23 | 2950 | } |
1da177e4 | 2951 | if (pte_file(entry)) |
d0217ac0 | 2952 | return do_nonlinear_fault(mm, vma, address, |
30c9f3a9 | 2953 | pte, pmd, flags, entry); |
65500d23 | 2954 | return do_swap_page(mm, vma, address, |
30c9f3a9 | 2955 | pte, pmd, flags, entry); |
1da177e4 LT |
2956 | } |
2957 | ||
4c21e2f2 | 2958 | ptl = pte_lockptr(mm, pmd); |
8f4e2101 HD |
2959 | spin_lock(ptl); |
2960 | if (unlikely(!pte_same(*pte, entry))) | |
2961 | goto unlock; | |
30c9f3a9 | 2962 | if (flags & FAULT_FLAG_WRITE) { |
1da177e4 | 2963 | if (!pte_write(entry)) |
8f4e2101 HD |
2964 | return do_wp_page(mm, vma, address, |
2965 | pte, pmd, ptl, entry); | |
1da177e4 LT |
2966 | entry = pte_mkdirty(entry); |
2967 | } | |
2968 | entry = pte_mkyoung(entry); | |
30c9f3a9 | 2969 | if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) { |
1a44e149 | 2970 | update_mmu_cache(vma, address, entry); |
1a44e149 AA |
2971 | } else { |
2972 | /* | |
2973 | * This is needed only for protection faults but the arch code | |
2974 | * is not yet telling us if this is a protection fault or not. | |
2975 | * This still avoids useless tlb flushes for .text page faults | |
2976 | * with threads. | |
2977 | */ | |
30c9f3a9 | 2978 | if (flags & FAULT_FLAG_WRITE) |
1a44e149 AA |
2979 | flush_tlb_page(vma, address); |
2980 | } | |
8f4e2101 HD |
2981 | unlock: |
2982 | pte_unmap_unlock(pte, ptl); | |
83c54070 | 2983 | return 0; |
1da177e4 LT |
2984 | } |
2985 | ||
2986 | /* | |
2987 | * By the time we get here, we already hold the mm semaphore | |
2988 | */ | |
83c54070 | 2989 | int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
d06063cc | 2990 | unsigned long address, unsigned int flags) |
1da177e4 LT |
2991 | { |
2992 | pgd_t *pgd; | |
2993 | pud_t *pud; | |
2994 | pmd_t *pmd; | |
2995 | pte_t *pte; | |
2996 | ||
2997 | __set_current_state(TASK_RUNNING); | |
2998 | ||
f8891e5e | 2999 | count_vm_event(PGFAULT); |
1da177e4 | 3000 | |
ac9b9c66 | 3001 | if (unlikely(is_vm_hugetlb_page(vma))) |
30c9f3a9 | 3002 | return hugetlb_fault(mm, vma, address, flags); |
1da177e4 | 3003 | |
1da177e4 | 3004 | pgd = pgd_offset(mm, address); |
1da177e4 LT |
3005 | pud = pud_alloc(mm, pgd, address); |
3006 | if (!pud) | |
c74df32c | 3007 | return VM_FAULT_OOM; |
1da177e4 LT |
3008 | pmd = pmd_alloc(mm, pud, address); |
3009 | if (!pmd) | |
c74df32c | 3010 | return VM_FAULT_OOM; |
1da177e4 LT |
3011 | pte = pte_alloc_map(mm, pmd, address); |
3012 | if (!pte) | |
c74df32c | 3013 | return VM_FAULT_OOM; |
1da177e4 | 3014 | |
30c9f3a9 | 3015 | return handle_pte_fault(mm, vma, address, pte, pmd, flags); |
1da177e4 LT |
3016 | } |
3017 | ||
3018 | #ifndef __PAGETABLE_PUD_FOLDED | |
3019 | /* | |
3020 | * Allocate page upper directory. | |
872fec16 | 3021 | * We've already handled the fast-path in-line. |
1da177e4 | 3022 | */ |
1bb3630e | 3023 | int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) |
1da177e4 | 3024 | { |
c74df32c HD |
3025 | pud_t *new = pud_alloc_one(mm, address); |
3026 | if (!new) | |
1bb3630e | 3027 | return -ENOMEM; |
1da177e4 | 3028 | |
362a61ad NP |
3029 | smp_wmb(); /* See comment in __pte_alloc */ |
3030 | ||
872fec16 | 3031 | spin_lock(&mm->page_table_lock); |
1bb3630e | 3032 | if (pgd_present(*pgd)) /* Another has populated it */ |
5e541973 | 3033 | pud_free(mm, new); |
1bb3630e HD |
3034 | else |
3035 | pgd_populate(mm, pgd, new); | |
c74df32c | 3036 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 3037 | return 0; |
1da177e4 LT |
3038 | } |
3039 | #endif /* __PAGETABLE_PUD_FOLDED */ | |
3040 | ||
3041 | #ifndef __PAGETABLE_PMD_FOLDED | |
3042 | /* | |
3043 | * Allocate page middle directory. | |
872fec16 | 3044 | * We've already handled the fast-path in-line. |
1da177e4 | 3045 | */ |
1bb3630e | 3046 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
1da177e4 | 3047 | { |
c74df32c HD |
3048 | pmd_t *new = pmd_alloc_one(mm, address); |
3049 | if (!new) | |
1bb3630e | 3050 | return -ENOMEM; |
1da177e4 | 3051 | |
362a61ad NP |
3052 | smp_wmb(); /* See comment in __pte_alloc */ |
3053 | ||
872fec16 | 3054 | spin_lock(&mm->page_table_lock); |
1da177e4 | 3055 | #ifndef __ARCH_HAS_4LEVEL_HACK |
1bb3630e | 3056 | if (pud_present(*pud)) /* Another has populated it */ |
5e541973 | 3057 | pmd_free(mm, new); |
1bb3630e HD |
3058 | else |
3059 | pud_populate(mm, pud, new); | |
1da177e4 | 3060 | #else |
1bb3630e | 3061 | if (pgd_present(*pud)) /* Another has populated it */ |
5e541973 | 3062 | pmd_free(mm, new); |
1bb3630e HD |
3063 | else |
3064 | pgd_populate(mm, pud, new); | |
1da177e4 | 3065 | #endif /* __ARCH_HAS_4LEVEL_HACK */ |
c74df32c | 3066 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 3067 | return 0; |
e0f39591 | 3068 | } |
1da177e4 LT |
3069 | #endif /* __PAGETABLE_PMD_FOLDED */ |
3070 | ||
3071 | int make_pages_present(unsigned long addr, unsigned long end) | |
3072 | { | |
3073 | int ret, len, write; | |
3074 | struct vm_area_struct * vma; | |
3075 | ||
3076 | vma = find_vma(current->mm, addr); | |
3077 | if (!vma) | |
a477097d | 3078 | return -ENOMEM; |
1da177e4 | 3079 | write = (vma->vm_flags & VM_WRITE) != 0; |
5bcb28b1 ES |
3080 | BUG_ON(addr >= end); |
3081 | BUG_ON(end > vma->vm_end); | |
68e116a3 | 3082 | len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE; |
1da177e4 LT |
3083 | ret = get_user_pages(current, current->mm, addr, |
3084 | len, write, 0, NULL, NULL); | |
c11d69d8 | 3085 | if (ret < 0) |
1da177e4 | 3086 | return ret; |
9978ad58 | 3087 | return ret == len ? 0 : -EFAULT; |
1da177e4 LT |
3088 | } |
3089 | ||
1da177e4 LT |
3090 | #if !defined(__HAVE_ARCH_GATE_AREA) |
3091 | ||
3092 | #if defined(AT_SYSINFO_EHDR) | |
5ce7852c | 3093 | static struct vm_area_struct gate_vma; |
1da177e4 LT |
3094 | |
3095 | static int __init gate_vma_init(void) | |
3096 | { | |
3097 | gate_vma.vm_mm = NULL; | |
3098 | gate_vma.vm_start = FIXADDR_USER_START; | |
3099 | gate_vma.vm_end = FIXADDR_USER_END; | |
b6558c4a RM |
3100 | gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC; |
3101 | gate_vma.vm_page_prot = __P101; | |
f47aef55 RM |
3102 | /* |
3103 | * Make sure the vDSO gets into every core dump. | |
3104 | * Dumping its contents makes post-mortem fully interpretable later | |
3105 | * without matching up the same kernel and hardware config to see | |
3106 | * what PC values meant. | |
3107 | */ | |
3108 | gate_vma.vm_flags |= VM_ALWAYSDUMP; | |
1da177e4 LT |
3109 | return 0; |
3110 | } | |
3111 | __initcall(gate_vma_init); | |
3112 | #endif | |
3113 | ||
3114 | struct vm_area_struct *get_gate_vma(struct task_struct *tsk) | |
3115 | { | |
3116 | #ifdef AT_SYSINFO_EHDR | |
3117 | return &gate_vma; | |
3118 | #else | |
3119 | return NULL; | |
3120 | #endif | |
3121 | } | |
3122 | ||
3123 | int in_gate_area_no_task(unsigned long addr) | |
3124 | { | |
3125 | #ifdef AT_SYSINFO_EHDR | |
3126 | if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END)) | |
3127 | return 1; | |
3128 | #endif | |
3129 | return 0; | |
3130 | } | |
3131 | ||
3132 | #endif /* __HAVE_ARCH_GATE_AREA */ | |
0ec76a11 | 3133 | |
f8ad0f49 JW |
3134 | static int follow_pte(struct mm_struct *mm, unsigned long address, |
3135 | pte_t **ptepp, spinlock_t **ptlp) | |
3136 | { | |
3137 | pgd_t *pgd; | |
3138 | pud_t *pud; | |
3139 | pmd_t *pmd; | |
3140 | pte_t *ptep; | |
3141 | ||
3142 | pgd = pgd_offset(mm, address); | |
3143 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
3144 | goto out; | |
3145 | ||
3146 | pud = pud_offset(pgd, address); | |
3147 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) | |
3148 | goto out; | |
3149 | ||
3150 | pmd = pmd_offset(pud, address); | |
3151 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) | |
3152 | goto out; | |
3153 | ||
3154 | /* We cannot handle huge page PFN maps. Luckily they don't exist. */ | |
3155 | if (pmd_huge(*pmd)) | |
3156 | goto out; | |
3157 | ||
3158 | ptep = pte_offset_map_lock(mm, pmd, address, ptlp); | |
3159 | if (!ptep) | |
3160 | goto out; | |
3161 | if (!pte_present(*ptep)) | |
3162 | goto unlock; | |
3163 | *ptepp = ptep; | |
3164 | return 0; | |
3165 | unlock: | |
3166 | pte_unmap_unlock(ptep, *ptlp); | |
3167 | out: | |
3168 | return -EINVAL; | |
3169 | } | |
3170 | ||
3b6748e2 JW |
3171 | /** |
3172 | * follow_pfn - look up PFN at a user virtual address | |
3173 | * @vma: memory mapping | |
3174 | * @address: user virtual address | |
3175 | * @pfn: location to store found PFN | |
3176 | * | |
3177 | * Only IO mappings and raw PFN mappings are allowed. | |
3178 | * | |
3179 | * Returns zero and the pfn at @pfn on success, -ve otherwise. | |
3180 | */ | |
3181 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
3182 | unsigned long *pfn) | |
3183 | { | |
3184 | int ret = -EINVAL; | |
3185 | spinlock_t *ptl; | |
3186 | pte_t *ptep; | |
3187 | ||
3188 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
3189 | return ret; | |
3190 | ||
3191 | ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); | |
3192 | if (ret) | |
3193 | return ret; | |
3194 | *pfn = pte_pfn(*ptep); | |
3195 | pte_unmap_unlock(ptep, ptl); | |
3196 | return 0; | |
3197 | } | |
3198 | EXPORT_SYMBOL(follow_pfn); | |
3199 | ||
28b2ee20 | 3200 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
d87fe660 | 3201 | int follow_phys(struct vm_area_struct *vma, |
3202 | unsigned long address, unsigned int flags, | |
3203 | unsigned long *prot, resource_size_t *phys) | |
28b2ee20 | 3204 | { |
03668a4d | 3205 | int ret = -EINVAL; |
28b2ee20 RR |
3206 | pte_t *ptep, pte; |
3207 | spinlock_t *ptl; | |
28b2ee20 | 3208 | |
d87fe660 | 3209 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
3210 | goto out; | |
28b2ee20 | 3211 | |
03668a4d | 3212 | if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) |
d87fe660 | 3213 | goto out; |
28b2ee20 | 3214 | pte = *ptep; |
03668a4d | 3215 | |
28b2ee20 RR |
3216 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
3217 | goto unlock; | |
28b2ee20 RR |
3218 | |
3219 | *prot = pgprot_val(pte_pgprot(pte)); | |
03668a4d | 3220 | *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; |
28b2ee20 | 3221 | |
03668a4d | 3222 | ret = 0; |
28b2ee20 RR |
3223 | unlock: |
3224 | pte_unmap_unlock(ptep, ptl); | |
3225 | out: | |
d87fe660 | 3226 | return ret; |
28b2ee20 RR |
3227 | } |
3228 | ||
3229 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, | |
3230 | void *buf, int len, int write) | |
3231 | { | |
3232 | resource_size_t phys_addr; | |
3233 | unsigned long prot = 0; | |
2bc7273b | 3234 | void __iomem *maddr; |
28b2ee20 RR |
3235 | int offset = addr & (PAGE_SIZE-1); |
3236 | ||
d87fe660 | 3237 | if (follow_phys(vma, addr, write, &prot, &phys_addr)) |
28b2ee20 RR |
3238 | return -EINVAL; |
3239 | ||
3240 | maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot); | |
3241 | if (write) | |
3242 | memcpy_toio(maddr + offset, buf, len); | |
3243 | else | |
3244 | memcpy_fromio(buf, maddr + offset, len); | |
3245 | iounmap(maddr); | |
3246 | ||
3247 | return len; | |
3248 | } | |
3249 | #endif | |
3250 | ||
0ec76a11 DH |
3251 | /* |
3252 | * Access another process' address space. | |
3253 | * Source/target buffer must be kernel space, | |
3254 | * Do not walk the page table directly, use get_user_pages | |
3255 | */ | |
3256 | int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) | |
3257 | { | |
3258 | struct mm_struct *mm; | |
3259 | struct vm_area_struct *vma; | |
0ec76a11 DH |
3260 | void *old_buf = buf; |
3261 | ||
3262 | mm = get_task_mm(tsk); | |
3263 | if (!mm) | |
3264 | return 0; | |
3265 | ||
3266 | down_read(&mm->mmap_sem); | |
183ff22b | 3267 | /* ignore errors, just check how much was successfully transferred */ |
0ec76a11 DH |
3268 | while (len) { |
3269 | int bytes, ret, offset; | |
3270 | void *maddr; | |
28b2ee20 | 3271 | struct page *page = NULL; |
0ec76a11 DH |
3272 | |
3273 | ret = get_user_pages(tsk, mm, addr, 1, | |
3274 | write, 1, &page, &vma); | |
28b2ee20 RR |
3275 | if (ret <= 0) { |
3276 | /* | |
3277 | * Check if this is a VM_IO | VM_PFNMAP VMA, which | |
3278 | * we can access using slightly different code. | |
3279 | */ | |
3280 | #ifdef CONFIG_HAVE_IOREMAP_PROT | |
3281 | vma = find_vma(mm, addr); | |
3282 | if (!vma) | |
3283 | break; | |
3284 | if (vma->vm_ops && vma->vm_ops->access) | |
3285 | ret = vma->vm_ops->access(vma, addr, buf, | |
3286 | len, write); | |
3287 | if (ret <= 0) | |
3288 | #endif | |
3289 | break; | |
3290 | bytes = ret; | |
0ec76a11 | 3291 | } else { |
28b2ee20 RR |
3292 | bytes = len; |
3293 | offset = addr & (PAGE_SIZE-1); | |
3294 | if (bytes > PAGE_SIZE-offset) | |
3295 | bytes = PAGE_SIZE-offset; | |
3296 | ||
3297 | maddr = kmap(page); | |
3298 | if (write) { | |
3299 | copy_to_user_page(vma, page, addr, | |
3300 | maddr + offset, buf, bytes); | |
3301 | set_page_dirty_lock(page); | |
3302 | } else { | |
3303 | copy_from_user_page(vma, page, addr, | |
3304 | buf, maddr + offset, bytes); | |
3305 | } | |
3306 | kunmap(page); | |
3307 | page_cache_release(page); | |
0ec76a11 | 3308 | } |
0ec76a11 DH |
3309 | len -= bytes; |
3310 | buf += bytes; | |
3311 | addr += bytes; | |
3312 | } | |
3313 | up_read(&mm->mmap_sem); | |
3314 | mmput(mm); | |
3315 | ||
3316 | return buf - old_buf; | |
3317 | } | |
03252919 AK |
3318 | |
3319 | /* | |
3320 | * Print the name of a VMA. | |
3321 | */ | |
3322 | void print_vma_addr(char *prefix, unsigned long ip) | |
3323 | { | |
3324 | struct mm_struct *mm = current->mm; | |
3325 | struct vm_area_struct *vma; | |
3326 | ||
e8bff74a IM |
3327 | /* |
3328 | * Do not print if we are in atomic | |
3329 | * contexts (in exception stacks, etc.): | |
3330 | */ | |
3331 | if (preempt_count()) | |
3332 | return; | |
3333 | ||
03252919 AK |
3334 | down_read(&mm->mmap_sem); |
3335 | vma = find_vma(mm, ip); | |
3336 | if (vma && vma->vm_file) { | |
3337 | struct file *f = vma->vm_file; | |
3338 | char *buf = (char *)__get_free_page(GFP_KERNEL); | |
3339 | if (buf) { | |
3340 | char *p, *s; | |
3341 | ||
cf28b486 | 3342 | p = d_path(&f->f_path, buf, PAGE_SIZE); |
03252919 AK |
3343 | if (IS_ERR(p)) |
3344 | p = "?"; | |
3345 | s = strrchr(p, '/'); | |
3346 | if (s) | |
3347 | p = s+1; | |
3348 | printk("%s%s[%lx+%lx]", prefix, p, | |
3349 | vma->vm_start, | |
3350 | vma->vm_end - vma->vm_start); | |
3351 | free_page((unsigned long)buf); | |
3352 | } | |
3353 | } | |
3354 | up_read(¤t->mm->mmap_sem); | |
3355 | } | |
3ee1afa3 NP |
3356 | |
3357 | #ifdef CONFIG_PROVE_LOCKING | |
3358 | void might_fault(void) | |
3359 | { | |
95156f00 PZ |
3360 | /* |
3361 | * Some code (nfs/sunrpc) uses socket ops on kernel memory while | |
3362 | * holding the mmap_sem, this is safe because kernel memory doesn't | |
3363 | * get paged out, therefore we'll never actually fault, and the | |
3364 | * below annotations will generate false positives. | |
3365 | */ | |
3366 | if (segment_eq(get_fs(), KERNEL_DS)) | |
3367 | return; | |
3368 | ||
3ee1afa3 NP |
3369 | might_sleep(); |
3370 | /* | |
3371 | * it would be nicer only to annotate paths which are not under | |
3372 | * pagefault_disable, however that requires a larger audit and | |
3373 | * providing helpers like get_user_atomic. | |
3374 | */ | |
3375 | if (!in_atomic() && current->mm) | |
3376 | might_lock_read(¤t->mm->mmap_sem); | |
3377 | } | |
3378 | EXPORT_SYMBOL(might_fault); | |
3379 | #endif |