]>
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> | |
5a0e3ad6 | 59 | #include <linux/gfp.h> |
1da177e4 | 60 | |
6952b61d | 61 | #include <asm/io.h> |
1da177e4 LT |
62 | #include <asm/pgalloc.h> |
63 | #include <asm/uaccess.h> | |
64 | #include <asm/tlb.h> | |
65 | #include <asm/tlbflush.h> | |
66 | #include <asm/pgtable.h> | |
67 | ||
42b77728 JB |
68 | #include "internal.h" |
69 | ||
d41dee36 | 70 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
71 | /* use the per-pgdat data instead for discontigmem - mbligh */ |
72 | unsigned long max_mapnr; | |
73 | struct page *mem_map; | |
74 | ||
75 | EXPORT_SYMBOL(max_mapnr); | |
76 | EXPORT_SYMBOL(mem_map); | |
77 | #endif | |
78 | ||
79 | unsigned long num_physpages; | |
80 | /* | |
81 | * A number of key systems in x86 including ioremap() rely on the assumption | |
82 | * that high_memory defines the upper bound on direct map memory, then end | |
83 | * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and | |
84 | * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL | |
85 | * and ZONE_HIGHMEM. | |
86 | */ | |
87 | void * high_memory; | |
1da177e4 LT |
88 | |
89 | EXPORT_SYMBOL(num_physpages); | |
90 | EXPORT_SYMBOL(high_memory); | |
1da177e4 | 91 | |
32a93233 IM |
92 | /* |
93 | * Randomize the address space (stacks, mmaps, brk, etc.). | |
94 | * | |
95 | * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, | |
96 | * as ancient (libc5 based) binaries can segfault. ) | |
97 | */ | |
98 | int randomize_va_space __read_mostly = | |
99 | #ifdef CONFIG_COMPAT_BRK | |
100 | 1; | |
101 | #else | |
102 | 2; | |
103 | #endif | |
a62eaf15 AK |
104 | |
105 | static int __init disable_randmaps(char *s) | |
106 | { | |
107 | randomize_va_space = 0; | |
9b41046c | 108 | return 1; |
a62eaf15 AK |
109 | } |
110 | __setup("norandmaps", disable_randmaps); | |
111 | ||
62eede62 | 112 | unsigned long zero_pfn __read_mostly; |
03f6462a | 113 | unsigned long highest_memmap_pfn __read_mostly; |
a13ea5b7 HD |
114 | |
115 | /* | |
116 | * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() | |
117 | */ | |
118 | static int __init init_zero_pfn(void) | |
119 | { | |
120 | zero_pfn = page_to_pfn(ZERO_PAGE(0)); | |
121 | return 0; | |
122 | } | |
123 | core_initcall(init_zero_pfn); | |
a62eaf15 | 124 | |
d559db08 | 125 | |
34e55232 KH |
126 | #if defined(SPLIT_RSS_COUNTING) |
127 | ||
a3a2e76c | 128 | static void __sync_task_rss_stat(struct task_struct *task, struct mm_struct *mm) |
34e55232 KH |
129 | { |
130 | int i; | |
131 | ||
132 | for (i = 0; i < NR_MM_COUNTERS; i++) { | |
133 | if (task->rss_stat.count[i]) { | |
134 | add_mm_counter(mm, i, task->rss_stat.count[i]); | |
135 | task->rss_stat.count[i] = 0; | |
136 | } | |
137 | } | |
138 | task->rss_stat.events = 0; | |
139 | } | |
140 | ||
141 | static void add_mm_counter_fast(struct mm_struct *mm, int member, int val) | |
142 | { | |
143 | struct task_struct *task = current; | |
144 | ||
145 | if (likely(task->mm == mm)) | |
146 | task->rss_stat.count[member] += val; | |
147 | else | |
148 | add_mm_counter(mm, member, val); | |
149 | } | |
150 | #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1) | |
151 | #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1) | |
152 | ||
153 | /* sync counter once per 64 page faults */ | |
154 | #define TASK_RSS_EVENTS_THRESH (64) | |
155 | static void check_sync_rss_stat(struct task_struct *task) | |
156 | { | |
157 | if (unlikely(task != current)) | |
158 | return; | |
159 | if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH)) | |
160 | __sync_task_rss_stat(task, task->mm); | |
161 | } | |
162 | ||
163 | unsigned long get_mm_counter(struct mm_struct *mm, int member) | |
164 | { | |
165 | long val = 0; | |
166 | ||
167 | /* | |
168 | * Don't use task->mm here...for avoiding to use task_get_mm().. | |
169 | * The caller must guarantee task->mm is not invalid. | |
170 | */ | |
171 | val = atomic_long_read(&mm->rss_stat.count[member]); | |
172 | /* | |
173 | * counter is updated in asynchronous manner and may go to minus. | |
174 | * But it's never be expected number for users. | |
175 | */ | |
176 | if (val < 0) | |
177 | return 0; | |
178 | return (unsigned long)val; | |
179 | } | |
180 | ||
181 | void sync_mm_rss(struct task_struct *task, struct mm_struct *mm) | |
182 | { | |
183 | __sync_task_rss_stat(task, mm); | |
184 | } | |
185 | #else | |
186 | ||
187 | #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member) | |
188 | #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member) | |
189 | ||
190 | static void check_sync_rss_stat(struct task_struct *task) | |
191 | { | |
192 | } | |
193 | ||
34e55232 KH |
194 | #endif |
195 | ||
26723911 PZ |
196 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE |
197 | ||
198 | /* | |
199 | * See the comment near struct mmu_table_batch. | |
200 | */ | |
201 | ||
202 | static void tlb_remove_table_smp_sync(void *arg) | |
203 | { | |
204 | /* Simply deliver the interrupt */ | |
205 | } | |
206 | ||
207 | static void tlb_remove_table_one(void *table) | |
208 | { | |
209 | /* | |
210 | * This isn't an RCU grace period and hence the page-tables cannot be | |
211 | * assumed to be actually RCU-freed. | |
212 | * | |
213 | * It is however sufficient for software page-table walkers that rely on | |
214 | * IRQ disabling. See the comment near struct mmu_table_batch. | |
215 | */ | |
216 | smp_call_function(tlb_remove_table_smp_sync, NULL, 1); | |
217 | __tlb_remove_table(table); | |
218 | } | |
219 | ||
220 | static void tlb_remove_table_rcu(struct rcu_head *head) | |
221 | { | |
222 | struct mmu_table_batch *batch; | |
223 | int i; | |
224 | ||
225 | batch = container_of(head, struct mmu_table_batch, rcu); | |
226 | ||
227 | for (i = 0; i < batch->nr; i++) | |
228 | __tlb_remove_table(batch->tables[i]); | |
229 | ||
230 | free_page((unsigned long)batch); | |
231 | } | |
232 | ||
233 | void tlb_table_flush(struct mmu_gather *tlb) | |
234 | { | |
235 | struct mmu_table_batch **batch = &tlb->batch; | |
236 | ||
237 | if (*batch) { | |
238 | call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu); | |
239 | *batch = NULL; | |
240 | } | |
241 | } | |
242 | ||
243 | void tlb_remove_table(struct mmu_gather *tlb, void *table) | |
244 | { | |
245 | struct mmu_table_batch **batch = &tlb->batch; | |
246 | ||
247 | tlb->need_flush = 1; | |
248 | ||
249 | /* | |
250 | * When there's less then two users of this mm there cannot be a | |
251 | * concurrent page-table walk. | |
252 | */ | |
253 | if (atomic_read(&tlb->mm->mm_users) < 2) { | |
254 | __tlb_remove_table(table); | |
255 | return; | |
256 | } | |
257 | ||
258 | if (*batch == NULL) { | |
259 | *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN); | |
260 | if (*batch == NULL) { | |
261 | tlb_remove_table_one(table); | |
262 | return; | |
263 | } | |
264 | (*batch)->nr = 0; | |
265 | } | |
266 | (*batch)->tables[(*batch)->nr++] = table; | |
267 | if ((*batch)->nr == MAX_TABLE_BATCH) | |
268 | tlb_table_flush(tlb); | |
269 | } | |
270 | ||
271 | #endif | |
272 | ||
1da177e4 LT |
273 | /* |
274 | * If a p?d_bad entry is found while walking page tables, report | |
275 | * the error, before resetting entry to p?d_none. Usually (but | |
276 | * very seldom) called out from the p?d_none_or_clear_bad macros. | |
277 | */ | |
278 | ||
279 | void pgd_clear_bad(pgd_t *pgd) | |
280 | { | |
281 | pgd_ERROR(*pgd); | |
282 | pgd_clear(pgd); | |
283 | } | |
284 | ||
285 | void pud_clear_bad(pud_t *pud) | |
286 | { | |
287 | pud_ERROR(*pud); | |
288 | pud_clear(pud); | |
289 | } | |
290 | ||
291 | void pmd_clear_bad(pmd_t *pmd) | |
292 | { | |
293 | pmd_ERROR(*pmd); | |
294 | pmd_clear(pmd); | |
295 | } | |
296 | ||
297 | /* | |
298 | * Note: this doesn't free the actual pages themselves. That | |
299 | * has been handled earlier when unmapping all the memory regions. | |
300 | */ | |
9e1b32ca BH |
301 | static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, |
302 | unsigned long addr) | |
1da177e4 | 303 | { |
2f569afd | 304 | pgtable_t token = pmd_pgtable(*pmd); |
e0da382c | 305 | pmd_clear(pmd); |
9e1b32ca | 306 | pte_free_tlb(tlb, token, addr); |
e0da382c | 307 | tlb->mm->nr_ptes--; |
1da177e4 LT |
308 | } |
309 | ||
e0da382c HD |
310 | static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
311 | unsigned long addr, unsigned long end, | |
312 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
313 | { |
314 | pmd_t *pmd; | |
315 | unsigned long next; | |
e0da382c | 316 | unsigned long start; |
1da177e4 | 317 | |
e0da382c | 318 | start = addr; |
1da177e4 | 319 | pmd = pmd_offset(pud, addr); |
1da177e4 LT |
320 | do { |
321 | next = pmd_addr_end(addr, end); | |
322 | if (pmd_none_or_clear_bad(pmd)) | |
323 | continue; | |
9e1b32ca | 324 | free_pte_range(tlb, pmd, addr); |
1da177e4 LT |
325 | } while (pmd++, addr = next, addr != end); |
326 | ||
e0da382c HD |
327 | start &= PUD_MASK; |
328 | if (start < floor) | |
329 | return; | |
330 | if (ceiling) { | |
331 | ceiling &= PUD_MASK; | |
332 | if (!ceiling) | |
333 | return; | |
1da177e4 | 334 | } |
e0da382c HD |
335 | if (end - 1 > ceiling - 1) |
336 | return; | |
337 | ||
338 | pmd = pmd_offset(pud, start); | |
339 | pud_clear(pud); | |
9e1b32ca | 340 | pmd_free_tlb(tlb, pmd, start); |
1da177e4 LT |
341 | } |
342 | ||
e0da382c HD |
343 | static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, |
344 | unsigned long addr, unsigned long end, | |
345 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
346 | { |
347 | pud_t *pud; | |
348 | unsigned long next; | |
e0da382c | 349 | unsigned long start; |
1da177e4 | 350 | |
e0da382c | 351 | start = addr; |
1da177e4 | 352 | pud = pud_offset(pgd, addr); |
1da177e4 LT |
353 | do { |
354 | next = pud_addr_end(addr, end); | |
355 | if (pud_none_or_clear_bad(pud)) | |
356 | continue; | |
e0da382c | 357 | free_pmd_range(tlb, pud, addr, next, floor, ceiling); |
1da177e4 LT |
358 | } while (pud++, addr = next, addr != end); |
359 | ||
e0da382c HD |
360 | start &= PGDIR_MASK; |
361 | if (start < floor) | |
362 | return; | |
363 | if (ceiling) { | |
364 | ceiling &= PGDIR_MASK; | |
365 | if (!ceiling) | |
366 | return; | |
1da177e4 | 367 | } |
e0da382c HD |
368 | if (end - 1 > ceiling - 1) |
369 | return; | |
370 | ||
371 | pud = pud_offset(pgd, start); | |
372 | pgd_clear(pgd); | |
9e1b32ca | 373 | pud_free_tlb(tlb, pud, start); |
1da177e4 LT |
374 | } |
375 | ||
376 | /* | |
e0da382c HD |
377 | * This function frees user-level page tables of a process. |
378 | * | |
1da177e4 LT |
379 | * Must be called with pagetable lock held. |
380 | */ | |
42b77728 | 381 | void free_pgd_range(struct mmu_gather *tlb, |
e0da382c HD |
382 | unsigned long addr, unsigned long end, |
383 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
384 | { |
385 | pgd_t *pgd; | |
386 | unsigned long next; | |
e0da382c HD |
387 | |
388 | /* | |
389 | * The next few lines have given us lots of grief... | |
390 | * | |
391 | * Why are we testing PMD* at this top level? Because often | |
392 | * there will be no work to do at all, and we'd prefer not to | |
393 | * go all the way down to the bottom just to discover that. | |
394 | * | |
395 | * Why all these "- 1"s? Because 0 represents both the bottom | |
396 | * of the address space and the top of it (using -1 for the | |
397 | * top wouldn't help much: the masks would do the wrong thing). | |
398 | * The rule is that addr 0 and floor 0 refer to the bottom of | |
399 | * the address space, but end 0 and ceiling 0 refer to the top | |
400 | * Comparisons need to use "end - 1" and "ceiling - 1" (though | |
401 | * that end 0 case should be mythical). | |
402 | * | |
403 | * Wherever addr is brought up or ceiling brought down, we must | |
404 | * be careful to reject "the opposite 0" before it confuses the | |
405 | * subsequent tests. But what about where end is brought down | |
406 | * by PMD_SIZE below? no, end can't go down to 0 there. | |
407 | * | |
408 | * Whereas we round start (addr) and ceiling down, by different | |
409 | * masks at different levels, in order to test whether a table | |
410 | * now has no other vmas using it, so can be freed, we don't | |
411 | * bother to round floor or end up - the tests don't need that. | |
412 | */ | |
1da177e4 | 413 | |
e0da382c HD |
414 | addr &= PMD_MASK; |
415 | if (addr < floor) { | |
416 | addr += PMD_SIZE; | |
417 | if (!addr) | |
418 | return; | |
419 | } | |
420 | if (ceiling) { | |
421 | ceiling &= PMD_MASK; | |
422 | if (!ceiling) | |
423 | return; | |
424 | } | |
425 | if (end - 1 > ceiling - 1) | |
426 | end -= PMD_SIZE; | |
427 | if (addr > end - 1) | |
428 | return; | |
429 | ||
42b77728 | 430 | pgd = pgd_offset(tlb->mm, addr); |
1da177e4 LT |
431 | do { |
432 | next = pgd_addr_end(addr, end); | |
433 | if (pgd_none_or_clear_bad(pgd)) | |
434 | continue; | |
42b77728 | 435 | free_pud_range(tlb, pgd, addr, next, floor, ceiling); |
1da177e4 | 436 | } while (pgd++, addr = next, addr != end); |
e0da382c HD |
437 | } |
438 | ||
42b77728 | 439 | void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, |
3bf5ee95 | 440 | unsigned long floor, unsigned long ceiling) |
e0da382c HD |
441 | { |
442 | while (vma) { | |
443 | struct vm_area_struct *next = vma->vm_next; | |
444 | unsigned long addr = vma->vm_start; | |
445 | ||
8f4f8c16 | 446 | /* |
25d9e2d1 | 447 | * Hide vma from rmap and truncate_pagecache before freeing |
448 | * pgtables | |
8f4f8c16 | 449 | */ |
5beb4930 | 450 | unlink_anon_vmas(vma); |
8f4f8c16 HD |
451 | unlink_file_vma(vma); |
452 | ||
9da61aef | 453 | if (is_vm_hugetlb_page(vma)) { |
3bf5ee95 | 454 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
e0da382c | 455 | floor, next? next->vm_start: ceiling); |
3bf5ee95 HD |
456 | } else { |
457 | /* | |
458 | * Optimization: gather nearby vmas into one call down | |
459 | */ | |
460 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE | |
4866920b | 461 | && !is_vm_hugetlb_page(next)) { |
3bf5ee95 HD |
462 | vma = next; |
463 | next = vma->vm_next; | |
5beb4930 | 464 | unlink_anon_vmas(vma); |
8f4f8c16 | 465 | unlink_file_vma(vma); |
3bf5ee95 HD |
466 | } |
467 | free_pgd_range(tlb, addr, vma->vm_end, | |
468 | floor, next? next->vm_start: ceiling); | |
469 | } | |
e0da382c HD |
470 | vma = next; |
471 | } | |
1da177e4 LT |
472 | } |
473 | ||
8ac1f832 AA |
474 | int __pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma, |
475 | pmd_t *pmd, unsigned long address) | |
1da177e4 | 476 | { |
2f569afd | 477 | pgtable_t new = pte_alloc_one(mm, address); |
8ac1f832 | 478 | int wait_split_huge_page; |
1bb3630e HD |
479 | if (!new) |
480 | return -ENOMEM; | |
481 | ||
362a61ad NP |
482 | /* |
483 | * Ensure all pte setup (eg. pte page lock and page clearing) are | |
484 | * visible before the pte is made visible to other CPUs by being | |
485 | * put into page tables. | |
486 | * | |
487 | * The other side of the story is the pointer chasing in the page | |
488 | * table walking code (when walking the page table without locking; | |
489 | * ie. most of the time). Fortunately, these data accesses consist | |
490 | * of a chain of data-dependent loads, meaning most CPUs (alpha | |
491 | * being the notable exception) will already guarantee loads are | |
492 | * seen in-order. See the alpha page table accessors for the | |
493 | * smp_read_barrier_depends() barriers in page table walking code. | |
494 | */ | |
495 | smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ | |
496 | ||
c74df32c | 497 | spin_lock(&mm->page_table_lock); |
8ac1f832 AA |
498 | wait_split_huge_page = 0; |
499 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ | |
1da177e4 | 500 | mm->nr_ptes++; |
1da177e4 | 501 | pmd_populate(mm, pmd, new); |
2f569afd | 502 | new = NULL; |
8ac1f832 AA |
503 | } else if (unlikely(pmd_trans_splitting(*pmd))) |
504 | wait_split_huge_page = 1; | |
c74df32c | 505 | spin_unlock(&mm->page_table_lock); |
2f569afd MS |
506 | if (new) |
507 | pte_free(mm, new); | |
8ac1f832 AA |
508 | if (wait_split_huge_page) |
509 | wait_split_huge_page(vma->anon_vma, pmd); | |
1bb3630e | 510 | return 0; |
1da177e4 LT |
511 | } |
512 | ||
1bb3630e | 513 | int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) |
1da177e4 | 514 | { |
1bb3630e HD |
515 | pte_t *new = pte_alloc_one_kernel(&init_mm, address); |
516 | if (!new) | |
517 | return -ENOMEM; | |
518 | ||
362a61ad NP |
519 | smp_wmb(); /* See comment in __pte_alloc */ |
520 | ||
1bb3630e | 521 | spin_lock(&init_mm.page_table_lock); |
8ac1f832 | 522 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
1bb3630e | 523 | pmd_populate_kernel(&init_mm, pmd, new); |
2f569afd | 524 | new = NULL; |
8ac1f832 AA |
525 | } else |
526 | VM_BUG_ON(pmd_trans_splitting(*pmd)); | |
1bb3630e | 527 | spin_unlock(&init_mm.page_table_lock); |
2f569afd MS |
528 | if (new) |
529 | pte_free_kernel(&init_mm, new); | |
1bb3630e | 530 | return 0; |
1da177e4 LT |
531 | } |
532 | ||
d559db08 KH |
533 | static inline void init_rss_vec(int *rss) |
534 | { | |
535 | memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); | |
536 | } | |
537 | ||
538 | static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) | |
ae859762 | 539 | { |
d559db08 KH |
540 | int i; |
541 | ||
34e55232 KH |
542 | if (current->mm == mm) |
543 | sync_mm_rss(current, mm); | |
d559db08 KH |
544 | for (i = 0; i < NR_MM_COUNTERS; i++) |
545 | if (rss[i]) | |
546 | add_mm_counter(mm, i, rss[i]); | |
ae859762 HD |
547 | } |
548 | ||
b5810039 | 549 | /* |
6aab341e LT |
550 | * This function is called to print an error when a bad pte |
551 | * is found. For example, we might have a PFN-mapped pte in | |
552 | * a region that doesn't allow it. | |
b5810039 NP |
553 | * |
554 | * The calling function must still handle the error. | |
555 | */ | |
3dc14741 HD |
556 | static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, |
557 | pte_t pte, struct page *page) | |
b5810039 | 558 | { |
3dc14741 HD |
559 | pgd_t *pgd = pgd_offset(vma->vm_mm, addr); |
560 | pud_t *pud = pud_offset(pgd, addr); | |
561 | pmd_t *pmd = pmd_offset(pud, addr); | |
562 | struct address_space *mapping; | |
563 | pgoff_t index; | |
d936cf9b HD |
564 | static unsigned long resume; |
565 | static unsigned long nr_shown; | |
566 | static unsigned long nr_unshown; | |
567 | ||
568 | /* | |
569 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
570 | * or allow a steady drip of one report per second. | |
571 | */ | |
572 | if (nr_shown == 60) { | |
573 | if (time_before(jiffies, resume)) { | |
574 | nr_unshown++; | |
575 | return; | |
576 | } | |
577 | if (nr_unshown) { | |
1e9e6365 HD |
578 | printk(KERN_ALERT |
579 | "BUG: Bad page map: %lu messages suppressed\n", | |
d936cf9b HD |
580 | nr_unshown); |
581 | nr_unshown = 0; | |
582 | } | |
583 | nr_shown = 0; | |
584 | } | |
585 | if (nr_shown++ == 0) | |
586 | resume = jiffies + 60 * HZ; | |
3dc14741 HD |
587 | |
588 | mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; | |
589 | index = linear_page_index(vma, addr); | |
590 | ||
1e9e6365 HD |
591 | printk(KERN_ALERT |
592 | "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", | |
3dc14741 HD |
593 | current->comm, |
594 | (long long)pte_val(pte), (long long)pmd_val(*pmd)); | |
718a3821 WF |
595 | if (page) |
596 | dump_page(page); | |
1e9e6365 | 597 | printk(KERN_ALERT |
3dc14741 HD |
598 | "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n", |
599 | (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); | |
600 | /* | |
601 | * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y | |
602 | */ | |
603 | if (vma->vm_ops) | |
1e9e6365 | 604 | print_symbol(KERN_ALERT "vma->vm_ops->fault: %s\n", |
3dc14741 HD |
605 | (unsigned long)vma->vm_ops->fault); |
606 | if (vma->vm_file && vma->vm_file->f_op) | |
1e9e6365 | 607 | print_symbol(KERN_ALERT "vma->vm_file->f_op->mmap: %s\n", |
3dc14741 | 608 | (unsigned long)vma->vm_file->f_op->mmap); |
b5810039 | 609 | dump_stack(); |
3dc14741 | 610 | add_taint(TAINT_BAD_PAGE); |
b5810039 NP |
611 | } |
612 | ||
67121172 LT |
613 | static inline int is_cow_mapping(unsigned int flags) |
614 | { | |
615 | return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; | |
616 | } | |
617 | ||
62eede62 HD |
618 | #ifndef is_zero_pfn |
619 | static inline int is_zero_pfn(unsigned long pfn) | |
620 | { | |
621 | return pfn == zero_pfn; | |
622 | } | |
623 | #endif | |
624 | ||
625 | #ifndef my_zero_pfn | |
626 | static inline unsigned long my_zero_pfn(unsigned long addr) | |
627 | { | |
628 | return zero_pfn; | |
629 | } | |
630 | #endif | |
631 | ||
ee498ed7 | 632 | /* |
7e675137 | 633 | * vm_normal_page -- This function gets the "struct page" associated with a pte. |
6aab341e | 634 | * |
7e675137 NP |
635 | * "Special" mappings do not wish to be associated with a "struct page" (either |
636 | * it doesn't exist, or it exists but they don't want to touch it). In this | |
637 | * case, NULL is returned here. "Normal" mappings do have a struct page. | |
b379d790 | 638 | * |
7e675137 NP |
639 | * There are 2 broad cases. Firstly, an architecture may define a pte_special() |
640 | * pte bit, in which case this function is trivial. Secondly, an architecture | |
641 | * may not have a spare pte bit, which requires a more complicated scheme, | |
642 | * described below. | |
643 | * | |
644 | * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a | |
645 | * special mapping (even if there are underlying and valid "struct pages"). | |
646 | * COWed pages of a VM_PFNMAP are always normal. | |
6aab341e | 647 | * |
b379d790 JH |
648 | * The way we recognize COWed pages within VM_PFNMAP mappings is through the |
649 | * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit | |
7e675137 NP |
650 | * set, and the vm_pgoff will point to the first PFN mapped: thus every special |
651 | * mapping will always honor the rule | |
6aab341e LT |
652 | * |
653 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) | |
654 | * | |
7e675137 NP |
655 | * And for normal mappings this is false. |
656 | * | |
657 | * This restricts such mappings to be a linear translation from virtual address | |
658 | * to pfn. To get around this restriction, we allow arbitrary mappings so long | |
659 | * as the vma is not a COW mapping; in that case, we know that all ptes are | |
660 | * special (because none can have been COWed). | |
b379d790 | 661 | * |
b379d790 | 662 | * |
7e675137 | 663 | * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. |
b379d790 JH |
664 | * |
665 | * VM_MIXEDMAP mappings can likewise contain memory with or without "struct | |
666 | * page" backing, however the difference is that _all_ pages with a struct | |
667 | * page (that is, those where pfn_valid is true) are refcounted and considered | |
668 | * normal pages by the VM. The disadvantage is that pages are refcounted | |
669 | * (which can be slower and simply not an option for some PFNMAP users). The | |
670 | * advantage is that we don't have to follow the strict linearity rule of | |
671 | * PFNMAP mappings in order to support COWable mappings. | |
672 | * | |
ee498ed7 | 673 | */ |
7e675137 NP |
674 | #ifdef __HAVE_ARCH_PTE_SPECIAL |
675 | # define HAVE_PTE_SPECIAL 1 | |
676 | #else | |
677 | # define HAVE_PTE_SPECIAL 0 | |
678 | #endif | |
679 | struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, | |
680 | pte_t pte) | |
ee498ed7 | 681 | { |
22b31eec | 682 | unsigned long pfn = pte_pfn(pte); |
7e675137 NP |
683 | |
684 | if (HAVE_PTE_SPECIAL) { | |
22b31eec HD |
685 | if (likely(!pte_special(pte))) |
686 | goto check_pfn; | |
a13ea5b7 HD |
687 | if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) |
688 | return NULL; | |
62eede62 | 689 | if (!is_zero_pfn(pfn)) |
22b31eec | 690 | print_bad_pte(vma, addr, pte, NULL); |
7e675137 NP |
691 | return NULL; |
692 | } | |
693 | ||
694 | /* !HAVE_PTE_SPECIAL case follows: */ | |
695 | ||
b379d790 JH |
696 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
697 | if (vma->vm_flags & VM_MIXEDMAP) { | |
698 | if (!pfn_valid(pfn)) | |
699 | return NULL; | |
700 | goto out; | |
701 | } else { | |
7e675137 NP |
702 | unsigned long off; |
703 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
b379d790 JH |
704 | if (pfn == vma->vm_pgoff + off) |
705 | return NULL; | |
706 | if (!is_cow_mapping(vma->vm_flags)) | |
707 | return NULL; | |
708 | } | |
6aab341e LT |
709 | } |
710 | ||
62eede62 HD |
711 | if (is_zero_pfn(pfn)) |
712 | return NULL; | |
22b31eec HD |
713 | check_pfn: |
714 | if (unlikely(pfn > highest_memmap_pfn)) { | |
715 | print_bad_pte(vma, addr, pte, NULL); | |
716 | return NULL; | |
717 | } | |
6aab341e LT |
718 | |
719 | /* | |
7e675137 | 720 | * NOTE! We still have PageReserved() pages in the page tables. |
7e675137 | 721 | * eg. VDSO mappings can cause them to exist. |
6aab341e | 722 | */ |
b379d790 | 723 | out: |
6aab341e | 724 | return pfn_to_page(pfn); |
ee498ed7 HD |
725 | } |
726 | ||
1da177e4 LT |
727 | /* |
728 | * copy one vm_area from one task to the other. Assumes the page tables | |
729 | * already present in the new task to be cleared in the whole range | |
730 | * covered by this vma. | |
1da177e4 LT |
731 | */ |
732 | ||
570a335b | 733 | static inline unsigned long |
1da177e4 | 734 | copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
b5810039 | 735 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, |
8c103762 | 736 | unsigned long addr, int *rss) |
1da177e4 | 737 | { |
b5810039 | 738 | unsigned long vm_flags = vma->vm_flags; |
1da177e4 LT |
739 | pte_t pte = *src_pte; |
740 | struct page *page; | |
1da177e4 LT |
741 | |
742 | /* pte contains position in swap or file, so copy. */ | |
743 | if (unlikely(!pte_present(pte))) { | |
744 | if (!pte_file(pte)) { | |
0697212a CL |
745 | swp_entry_t entry = pte_to_swp_entry(pte); |
746 | ||
570a335b HD |
747 | if (swap_duplicate(entry) < 0) |
748 | return entry.val; | |
749 | ||
1da177e4 LT |
750 | /* make sure dst_mm is on swapoff's mmlist. */ |
751 | if (unlikely(list_empty(&dst_mm->mmlist))) { | |
752 | spin_lock(&mmlist_lock); | |
f412ac08 HD |
753 | if (list_empty(&dst_mm->mmlist)) |
754 | list_add(&dst_mm->mmlist, | |
755 | &src_mm->mmlist); | |
1da177e4 LT |
756 | spin_unlock(&mmlist_lock); |
757 | } | |
b084d435 KH |
758 | if (likely(!non_swap_entry(entry))) |
759 | rss[MM_SWAPENTS]++; | |
760 | else if (is_write_migration_entry(entry) && | |
0697212a CL |
761 | is_cow_mapping(vm_flags)) { |
762 | /* | |
763 | * COW mappings require pages in both parent | |
764 | * and child to be set to read. | |
765 | */ | |
766 | make_migration_entry_read(&entry); | |
767 | pte = swp_entry_to_pte(entry); | |
768 | set_pte_at(src_mm, addr, src_pte, pte); | |
769 | } | |
1da177e4 | 770 | } |
ae859762 | 771 | goto out_set_pte; |
1da177e4 LT |
772 | } |
773 | ||
1da177e4 LT |
774 | /* |
775 | * If it's a COW mapping, write protect it both | |
776 | * in the parent and the child | |
777 | */ | |
67121172 | 778 | if (is_cow_mapping(vm_flags)) { |
1da177e4 | 779 | ptep_set_wrprotect(src_mm, addr, src_pte); |
3dc90795 | 780 | pte = pte_wrprotect(pte); |
1da177e4 LT |
781 | } |
782 | ||
783 | /* | |
784 | * If it's a shared mapping, mark it clean in | |
785 | * the child | |
786 | */ | |
787 | if (vm_flags & VM_SHARED) | |
788 | pte = pte_mkclean(pte); | |
789 | pte = pte_mkold(pte); | |
6aab341e LT |
790 | |
791 | page = vm_normal_page(vma, addr, pte); | |
792 | if (page) { | |
793 | get_page(page); | |
21333b2b | 794 | page_dup_rmap(page); |
d559db08 KH |
795 | if (PageAnon(page)) |
796 | rss[MM_ANONPAGES]++; | |
797 | else | |
798 | rss[MM_FILEPAGES]++; | |
6aab341e | 799 | } |
ae859762 HD |
800 | |
801 | out_set_pte: | |
802 | set_pte_at(dst_mm, addr, dst_pte, pte); | |
570a335b | 803 | return 0; |
1da177e4 LT |
804 | } |
805 | ||
71e3aac0 AA |
806 | int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
807 | pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, | |
808 | unsigned long addr, unsigned long end) | |
1da177e4 | 809 | { |
c36987e2 | 810 | pte_t *orig_src_pte, *orig_dst_pte; |
1da177e4 | 811 | pte_t *src_pte, *dst_pte; |
c74df32c | 812 | spinlock_t *src_ptl, *dst_ptl; |
e040f218 | 813 | int progress = 0; |
d559db08 | 814 | int rss[NR_MM_COUNTERS]; |
570a335b | 815 | swp_entry_t entry = (swp_entry_t){0}; |
1da177e4 LT |
816 | |
817 | again: | |
d559db08 KH |
818 | init_rss_vec(rss); |
819 | ||
c74df32c | 820 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
1da177e4 LT |
821 | if (!dst_pte) |
822 | return -ENOMEM; | |
ece0e2b6 | 823 | src_pte = pte_offset_map(src_pmd, addr); |
4c21e2f2 | 824 | src_ptl = pte_lockptr(src_mm, src_pmd); |
f20dc5f7 | 825 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
c36987e2 DN |
826 | orig_src_pte = src_pte; |
827 | orig_dst_pte = dst_pte; | |
6606c3e0 | 828 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 829 | |
1da177e4 LT |
830 | do { |
831 | /* | |
832 | * We are holding two locks at this point - either of them | |
833 | * could generate latencies in another task on another CPU. | |
834 | */ | |
e040f218 HD |
835 | if (progress >= 32) { |
836 | progress = 0; | |
837 | if (need_resched() || | |
95c354fe | 838 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
e040f218 HD |
839 | break; |
840 | } | |
1da177e4 LT |
841 | if (pte_none(*src_pte)) { |
842 | progress++; | |
843 | continue; | |
844 | } | |
570a335b HD |
845 | entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, |
846 | vma, addr, rss); | |
847 | if (entry.val) | |
848 | break; | |
1da177e4 LT |
849 | progress += 8; |
850 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); | |
1da177e4 | 851 | |
6606c3e0 | 852 | arch_leave_lazy_mmu_mode(); |
c74df32c | 853 | spin_unlock(src_ptl); |
ece0e2b6 | 854 | pte_unmap(orig_src_pte); |
d559db08 | 855 | add_mm_rss_vec(dst_mm, rss); |
c36987e2 | 856 | pte_unmap_unlock(orig_dst_pte, dst_ptl); |
c74df32c | 857 | cond_resched(); |
570a335b HD |
858 | |
859 | if (entry.val) { | |
860 | if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) | |
861 | return -ENOMEM; | |
862 | progress = 0; | |
863 | } | |
1da177e4 LT |
864 | if (addr != end) |
865 | goto again; | |
866 | return 0; | |
867 | } | |
868 | ||
869 | static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
870 | pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, | |
871 | unsigned long addr, unsigned long end) | |
872 | { | |
873 | pmd_t *src_pmd, *dst_pmd; | |
874 | unsigned long next; | |
875 | ||
876 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); | |
877 | if (!dst_pmd) | |
878 | return -ENOMEM; | |
879 | src_pmd = pmd_offset(src_pud, addr); | |
880 | do { | |
881 | next = pmd_addr_end(addr, end); | |
71e3aac0 AA |
882 | if (pmd_trans_huge(*src_pmd)) { |
883 | int err; | |
14d1a55c | 884 | VM_BUG_ON(next-addr != HPAGE_PMD_SIZE); |
71e3aac0 AA |
885 | err = copy_huge_pmd(dst_mm, src_mm, |
886 | dst_pmd, src_pmd, addr, vma); | |
887 | if (err == -ENOMEM) | |
888 | return -ENOMEM; | |
889 | if (!err) | |
890 | continue; | |
891 | /* fall through */ | |
892 | } | |
1da177e4 LT |
893 | if (pmd_none_or_clear_bad(src_pmd)) |
894 | continue; | |
895 | if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, | |
896 | vma, addr, next)) | |
897 | return -ENOMEM; | |
898 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); | |
899 | return 0; | |
900 | } | |
901 | ||
902 | static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
903 | pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, | |
904 | unsigned long addr, unsigned long end) | |
905 | { | |
906 | pud_t *src_pud, *dst_pud; | |
907 | unsigned long next; | |
908 | ||
909 | dst_pud = pud_alloc(dst_mm, dst_pgd, addr); | |
910 | if (!dst_pud) | |
911 | return -ENOMEM; | |
912 | src_pud = pud_offset(src_pgd, addr); | |
913 | do { | |
914 | next = pud_addr_end(addr, end); | |
915 | if (pud_none_or_clear_bad(src_pud)) | |
916 | continue; | |
917 | if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, | |
918 | vma, addr, next)) | |
919 | return -ENOMEM; | |
920 | } while (dst_pud++, src_pud++, addr = next, addr != end); | |
921 | return 0; | |
922 | } | |
923 | ||
924 | int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
925 | struct vm_area_struct *vma) | |
926 | { | |
927 | pgd_t *src_pgd, *dst_pgd; | |
928 | unsigned long next; | |
929 | unsigned long addr = vma->vm_start; | |
930 | unsigned long end = vma->vm_end; | |
cddb8a5c | 931 | int ret; |
1da177e4 | 932 | |
d992895b NP |
933 | /* |
934 | * Don't copy ptes where a page fault will fill them correctly. | |
935 | * Fork becomes much lighter when there are big shared or private | |
936 | * readonly mappings. The tradeoff is that copy_page_range is more | |
937 | * efficient than faulting. | |
938 | */ | |
4d7672b4 | 939 | if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) { |
d992895b NP |
940 | if (!vma->anon_vma) |
941 | return 0; | |
942 | } | |
943 | ||
1da177e4 LT |
944 | if (is_vm_hugetlb_page(vma)) |
945 | return copy_hugetlb_page_range(dst_mm, src_mm, vma); | |
946 | ||
34801ba9 | 947 | if (unlikely(is_pfn_mapping(vma))) { |
2ab64037 | 948 | /* |
949 | * We do not free on error cases below as remove_vma | |
950 | * gets called on error from higher level routine | |
951 | */ | |
952 | ret = track_pfn_vma_copy(vma); | |
953 | if (ret) | |
954 | return ret; | |
955 | } | |
956 | ||
cddb8a5c AA |
957 | /* |
958 | * We need to invalidate the secondary MMU mappings only when | |
959 | * there could be a permission downgrade on the ptes of the | |
960 | * parent mm. And a permission downgrade will only happen if | |
961 | * is_cow_mapping() returns true. | |
962 | */ | |
963 | if (is_cow_mapping(vma->vm_flags)) | |
964 | mmu_notifier_invalidate_range_start(src_mm, addr, end); | |
965 | ||
966 | ret = 0; | |
1da177e4 LT |
967 | dst_pgd = pgd_offset(dst_mm, addr); |
968 | src_pgd = pgd_offset(src_mm, addr); | |
969 | do { | |
970 | next = pgd_addr_end(addr, end); | |
971 | if (pgd_none_or_clear_bad(src_pgd)) | |
972 | continue; | |
cddb8a5c AA |
973 | if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, |
974 | vma, addr, next))) { | |
975 | ret = -ENOMEM; | |
976 | break; | |
977 | } | |
1da177e4 | 978 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); |
cddb8a5c AA |
979 | |
980 | if (is_cow_mapping(vma->vm_flags)) | |
981 | mmu_notifier_invalidate_range_end(src_mm, | |
982 | vma->vm_start, end); | |
983 | return ret; | |
1da177e4 LT |
984 | } |
985 | ||
51c6f666 | 986 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
b5810039 | 987 | struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 | 988 | unsigned long addr, unsigned long end, |
97a89413 | 989 | struct zap_details *details) |
1da177e4 | 990 | { |
b5810039 | 991 | struct mm_struct *mm = tlb->mm; |
d16dfc55 | 992 | int force_flush = 0; |
d559db08 | 993 | int rss[NR_MM_COUNTERS]; |
97a89413 PZ |
994 | spinlock_t *ptl; |
995 | pte_t *pte; | |
d559db08 | 996 | |
d16dfc55 | 997 | again: |
e303297e | 998 | init_rss_vec(rss); |
508034a3 | 999 | pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
6606c3e0 | 1000 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1001 | do { |
1002 | pte_t ptent = *pte; | |
51c6f666 | 1003 | if (pte_none(ptent)) { |
1da177e4 | 1004 | continue; |
51c6f666 | 1005 | } |
6f5e6b9e | 1006 | |
1da177e4 | 1007 | if (pte_present(ptent)) { |
ee498ed7 | 1008 | struct page *page; |
51c6f666 | 1009 | |
6aab341e | 1010 | page = vm_normal_page(vma, addr, ptent); |
1da177e4 LT |
1011 | if (unlikely(details) && page) { |
1012 | /* | |
1013 | * unmap_shared_mapping_pages() wants to | |
1014 | * invalidate cache without truncating: | |
1015 | * unmap shared but keep private pages. | |
1016 | */ | |
1017 | if (details->check_mapping && | |
1018 | details->check_mapping != page->mapping) | |
1019 | continue; | |
1020 | /* | |
1021 | * Each page->index must be checked when | |
1022 | * invalidating or truncating nonlinear. | |
1023 | */ | |
1024 | if (details->nonlinear_vma && | |
1025 | (page->index < details->first_index || | |
1026 | page->index > details->last_index)) | |
1027 | continue; | |
1028 | } | |
b5810039 | 1029 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
a600388d | 1030 | tlb->fullmm); |
1da177e4 LT |
1031 | tlb_remove_tlb_entry(tlb, pte, addr); |
1032 | if (unlikely(!page)) | |
1033 | continue; | |
1034 | if (unlikely(details) && details->nonlinear_vma | |
1035 | && linear_page_index(details->nonlinear_vma, | |
1036 | addr) != page->index) | |
b5810039 | 1037 | set_pte_at(mm, addr, pte, |
1da177e4 | 1038 | pgoff_to_pte(page->index)); |
1da177e4 | 1039 | if (PageAnon(page)) |
d559db08 | 1040 | rss[MM_ANONPAGES]--; |
6237bcd9 HD |
1041 | else { |
1042 | if (pte_dirty(ptent)) | |
1043 | set_page_dirty(page); | |
4917e5d0 JW |
1044 | if (pte_young(ptent) && |
1045 | likely(!VM_SequentialReadHint(vma))) | |
bf3f3bc5 | 1046 | mark_page_accessed(page); |
d559db08 | 1047 | rss[MM_FILEPAGES]--; |
6237bcd9 | 1048 | } |
edc315fd | 1049 | page_remove_rmap(page); |
3dc14741 HD |
1050 | if (unlikely(page_mapcount(page) < 0)) |
1051 | print_bad_pte(vma, addr, ptent, page); | |
d16dfc55 PZ |
1052 | force_flush = !__tlb_remove_page(tlb, page); |
1053 | if (force_flush) | |
1054 | break; | |
1da177e4 LT |
1055 | continue; |
1056 | } | |
1057 | /* | |
1058 | * If details->check_mapping, we leave swap entries; | |
1059 | * if details->nonlinear_vma, we leave file entries. | |
1060 | */ | |
1061 | if (unlikely(details)) | |
1062 | continue; | |
2509ef26 HD |
1063 | if (pte_file(ptent)) { |
1064 | if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) | |
1065 | print_bad_pte(vma, addr, ptent, NULL); | |
b084d435 KH |
1066 | } else { |
1067 | swp_entry_t entry = pte_to_swp_entry(ptent); | |
1068 | ||
1069 | if (!non_swap_entry(entry)) | |
1070 | rss[MM_SWAPENTS]--; | |
1071 | if (unlikely(!free_swap_and_cache(entry))) | |
1072 | print_bad_pte(vma, addr, ptent, NULL); | |
1073 | } | |
9888a1ca | 1074 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
97a89413 | 1075 | } while (pte++, addr += PAGE_SIZE, addr != end); |
ae859762 | 1076 | |
d559db08 | 1077 | add_mm_rss_vec(mm, rss); |
6606c3e0 | 1078 | arch_leave_lazy_mmu_mode(); |
508034a3 | 1079 | pte_unmap_unlock(pte - 1, ptl); |
51c6f666 | 1080 | |
d16dfc55 PZ |
1081 | /* |
1082 | * mmu_gather ran out of room to batch pages, we break out of | |
1083 | * the PTE lock to avoid doing the potential expensive TLB invalidate | |
1084 | * and page-free while holding it. | |
1085 | */ | |
1086 | if (force_flush) { | |
1087 | force_flush = 0; | |
1088 | tlb_flush_mmu(tlb); | |
1089 | if (addr != end) | |
1090 | goto again; | |
1091 | } | |
1092 | ||
51c6f666 | 1093 | return addr; |
1da177e4 LT |
1094 | } |
1095 | ||
51c6f666 | 1096 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
b5810039 | 1097 | struct vm_area_struct *vma, pud_t *pud, |
1da177e4 | 1098 | unsigned long addr, unsigned long end, |
97a89413 | 1099 | struct zap_details *details) |
1da177e4 LT |
1100 | { |
1101 | pmd_t *pmd; | |
1102 | unsigned long next; | |
1103 | ||
1104 | pmd = pmd_offset(pud, addr); | |
1105 | do { | |
1106 | next = pmd_addr_end(addr, end); | |
71e3aac0 | 1107 | if (pmd_trans_huge(*pmd)) { |
14d1a55c AA |
1108 | if (next-addr != HPAGE_PMD_SIZE) { |
1109 | VM_BUG_ON(!rwsem_is_locked(&tlb->mm->mmap_sem)); | |
71e3aac0 | 1110 | split_huge_page_pmd(vma->vm_mm, pmd); |
97a89413 | 1111 | } else if (zap_huge_pmd(tlb, vma, pmd)) |
71e3aac0 | 1112 | continue; |
71e3aac0 AA |
1113 | /* fall through */ |
1114 | } | |
97a89413 | 1115 | if (pmd_none_or_clear_bad(pmd)) |
1da177e4 | 1116 | continue; |
97a89413 PZ |
1117 | next = zap_pte_range(tlb, vma, pmd, addr, next, details); |
1118 | cond_resched(); | |
1119 | } while (pmd++, addr = next, addr != end); | |
51c6f666 RH |
1120 | |
1121 | return addr; | |
1da177e4 LT |
1122 | } |
1123 | ||
51c6f666 | 1124 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
b5810039 | 1125 | struct vm_area_struct *vma, pgd_t *pgd, |
1da177e4 | 1126 | unsigned long addr, unsigned long end, |
97a89413 | 1127 | struct zap_details *details) |
1da177e4 LT |
1128 | { |
1129 | pud_t *pud; | |
1130 | unsigned long next; | |
1131 | ||
1132 | pud = pud_offset(pgd, addr); | |
1133 | do { | |
1134 | next = pud_addr_end(addr, end); | |
97a89413 | 1135 | if (pud_none_or_clear_bad(pud)) |
1da177e4 | 1136 | continue; |
97a89413 PZ |
1137 | next = zap_pmd_range(tlb, vma, pud, addr, next, details); |
1138 | } while (pud++, addr = next, addr != end); | |
51c6f666 RH |
1139 | |
1140 | return addr; | |
1da177e4 LT |
1141 | } |
1142 | ||
51c6f666 RH |
1143 | static unsigned long unmap_page_range(struct mmu_gather *tlb, |
1144 | struct vm_area_struct *vma, | |
1da177e4 | 1145 | unsigned long addr, unsigned long end, |
97a89413 | 1146 | struct zap_details *details) |
1da177e4 LT |
1147 | { |
1148 | pgd_t *pgd; | |
1149 | unsigned long next; | |
1150 | ||
1151 | if (details && !details->check_mapping && !details->nonlinear_vma) | |
1152 | details = NULL; | |
1153 | ||
1154 | BUG_ON(addr >= end); | |
569b846d | 1155 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
1156 | tlb_start_vma(tlb, vma); |
1157 | pgd = pgd_offset(vma->vm_mm, addr); | |
1158 | do { | |
1159 | next = pgd_addr_end(addr, end); | |
97a89413 | 1160 | if (pgd_none_or_clear_bad(pgd)) |
1da177e4 | 1161 | continue; |
97a89413 PZ |
1162 | next = zap_pud_range(tlb, vma, pgd, addr, next, details); |
1163 | } while (pgd++, addr = next, addr != end); | |
1da177e4 | 1164 | tlb_end_vma(tlb, vma); |
569b846d | 1165 | mem_cgroup_uncharge_end(); |
51c6f666 RH |
1166 | |
1167 | return addr; | |
1da177e4 LT |
1168 | } |
1169 | ||
1170 | #ifdef CONFIG_PREEMPT | |
1171 | # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE) | |
1172 | #else | |
1173 | /* No preempt: go for improved straight-line efficiency */ | |
1174 | # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE) | |
1175 | #endif | |
1176 | ||
1177 | /** | |
1178 | * unmap_vmas - unmap a range of memory covered by a list of vma's | |
1179 | * @tlbp: address of the caller's struct mmu_gather | |
1da177e4 LT |
1180 | * @vma: the starting vma |
1181 | * @start_addr: virtual address at which to start unmapping | |
1182 | * @end_addr: virtual address at which to end unmapping | |
1183 | * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here | |
1184 | * @details: details of nonlinear truncation or shared cache invalidation | |
1185 | * | |
ee39b37b | 1186 | * Returns the end address of the unmapping (restart addr if interrupted). |
1da177e4 | 1187 | * |
508034a3 | 1188 | * Unmap all pages in the vma list. |
1da177e4 | 1189 | * |
508034a3 HD |
1190 | * We aim to not hold locks for too long (for scheduling latency reasons). |
1191 | * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to | |
1da177e4 LT |
1192 | * return the ending mmu_gather to the caller. |
1193 | * | |
1194 | * Only addresses between `start' and `end' will be unmapped. | |
1195 | * | |
1196 | * The VMA list must be sorted in ascending virtual address order. | |
1197 | * | |
1198 | * unmap_vmas() assumes that the caller will flush the whole unmapped address | |
1199 | * range after unmap_vmas() returns. So the only responsibility here is to | |
1200 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() | |
1201 | * drops the lock and schedules. | |
1202 | */ | |
d16dfc55 | 1203 | unsigned long unmap_vmas(struct mmu_gather *tlb, |
1da177e4 LT |
1204 | struct vm_area_struct *vma, unsigned long start_addr, |
1205 | unsigned long end_addr, unsigned long *nr_accounted, | |
1206 | struct zap_details *details) | |
1207 | { | |
ee39b37b | 1208 | unsigned long start = start_addr; |
cddb8a5c | 1209 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 1210 | |
cddb8a5c | 1211 | mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); |
1da177e4 | 1212 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) { |
1da177e4 LT |
1213 | unsigned long end; |
1214 | ||
1215 | start = max(vma->vm_start, start_addr); | |
1216 | if (start >= vma->vm_end) | |
1217 | continue; | |
1218 | end = min(vma->vm_end, end_addr); | |
1219 | if (end <= vma->vm_start) | |
1220 | continue; | |
1221 | ||
1222 | if (vma->vm_flags & VM_ACCOUNT) | |
1223 | *nr_accounted += (end - start) >> PAGE_SHIFT; | |
1224 | ||
34801ba9 | 1225 | if (unlikely(is_pfn_mapping(vma))) |
2ab64037 | 1226 | untrack_pfn_vma(vma, 0, 0); |
1227 | ||
1da177e4 | 1228 | while (start != end) { |
51c6f666 | 1229 | if (unlikely(is_vm_hugetlb_page(vma))) { |
a137e1cc AK |
1230 | /* |
1231 | * It is undesirable to test vma->vm_file as it | |
1232 | * should be non-null for valid hugetlb area. | |
1233 | * However, vm_file will be NULL in the error | |
1234 | * cleanup path of do_mmap_pgoff. When | |
1235 | * hugetlbfs ->mmap method fails, | |
1236 | * do_mmap_pgoff() nullifies vma->vm_file | |
1237 | * before calling this function to clean up. | |
1238 | * Since no pte has actually been setup, it is | |
1239 | * safe to do nothing in this case. | |
1240 | */ | |
97a89413 | 1241 | if (vma->vm_file) |
a137e1cc | 1242 | unmap_hugepage_range(vma, start, end, NULL); |
a137e1cc | 1243 | |
51c6f666 RH |
1244 | start = end; |
1245 | } else | |
97a89413 | 1246 | start = unmap_page_range(tlb, vma, start, end, details); |
1da177e4 LT |
1247 | } |
1248 | } | |
97a89413 | 1249 | |
cddb8a5c | 1250 | mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); |
ee39b37b | 1251 | return start; /* which is now the end (or restart) address */ |
1da177e4 LT |
1252 | } |
1253 | ||
1254 | /** | |
1255 | * zap_page_range - remove user pages in a given range | |
1256 | * @vma: vm_area_struct holding the applicable pages | |
1257 | * @address: starting address of pages to zap | |
1258 | * @size: number of bytes to zap | |
1259 | * @details: details of nonlinear truncation or shared cache invalidation | |
1260 | */ | |
ee39b37b | 1261 | unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address, |
1da177e4 LT |
1262 | unsigned long size, struct zap_details *details) |
1263 | { | |
1264 | struct mm_struct *mm = vma->vm_mm; | |
d16dfc55 | 1265 | struct mmu_gather tlb; |
1da177e4 LT |
1266 | unsigned long end = address + size; |
1267 | unsigned long nr_accounted = 0; | |
1268 | ||
1da177e4 | 1269 | lru_add_drain(); |
d16dfc55 | 1270 | tlb_gather_mmu(&tlb, mm, 0); |
365e9c87 | 1271 | update_hiwater_rss(mm); |
508034a3 | 1272 | end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details); |
d16dfc55 | 1273 | tlb_finish_mmu(&tlb, address, end); |
ee39b37b | 1274 | return end; |
1da177e4 LT |
1275 | } |
1276 | ||
c627f9cc JS |
1277 | /** |
1278 | * zap_vma_ptes - remove ptes mapping the vma | |
1279 | * @vma: vm_area_struct holding ptes to be zapped | |
1280 | * @address: starting address of pages to zap | |
1281 | * @size: number of bytes to zap | |
1282 | * | |
1283 | * This function only unmaps ptes assigned to VM_PFNMAP vmas. | |
1284 | * | |
1285 | * The entire address range must be fully contained within the vma. | |
1286 | * | |
1287 | * Returns 0 if successful. | |
1288 | */ | |
1289 | int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, | |
1290 | unsigned long size) | |
1291 | { | |
1292 | if (address < vma->vm_start || address + size > vma->vm_end || | |
1293 | !(vma->vm_flags & VM_PFNMAP)) | |
1294 | return -1; | |
1295 | zap_page_range(vma, address, size, NULL); | |
1296 | return 0; | |
1297 | } | |
1298 | EXPORT_SYMBOL_GPL(zap_vma_ptes); | |
1299 | ||
142762bd JW |
1300 | /** |
1301 | * follow_page - look up a page descriptor from a user-virtual address | |
1302 | * @vma: vm_area_struct mapping @address | |
1303 | * @address: virtual address to look up | |
1304 | * @flags: flags modifying lookup behaviour | |
1305 | * | |
1306 | * @flags can have FOLL_ flags set, defined in <linux/mm.h> | |
1307 | * | |
1308 | * Returns the mapped (struct page *), %NULL if no mapping exists, or | |
1309 | * an error pointer if there is a mapping to something not represented | |
1310 | * by a page descriptor (see also vm_normal_page()). | |
1da177e4 | 1311 | */ |
6aab341e | 1312 | struct page *follow_page(struct vm_area_struct *vma, unsigned long address, |
deceb6cd | 1313 | unsigned int flags) |
1da177e4 LT |
1314 | { |
1315 | pgd_t *pgd; | |
1316 | pud_t *pud; | |
1317 | pmd_t *pmd; | |
1318 | pte_t *ptep, pte; | |
deceb6cd | 1319 | spinlock_t *ptl; |
1da177e4 | 1320 | struct page *page; |
6aab341e | 1321 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 1322 | |
deceb6cd HD |
1323 | page = follow_huge_addr(mm, address, flags & FOLL_WRITE); |
1324 | if (!IS_ERR(page)) { | |
1325 | BUG_ON(flags & FOLL_GET); | |
1326 | goto out; | |
1327 | } | |
1da177e4 | 1328 | |
deceb6cd | 1329 | page = NULL; |
1da177e4 LT |
1330 | pgd = pgd_offset(mm, address); |
1331 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
deceb6cd | 1332 | goto no_page_table; |
1da177e4 LT |
1333 | |
1334 | pud = pud_offset(pgd, address); | |
ceb86879 | 1335 | if (pud_none(*pud)) |
deceb6cd | 1336 | goto no_page_table; |
8a07651e | 1337 | if (pud_huge(*pud) && vma->vm_flags & VM_HUGETLB) { |
ceb86879 AK |
1338 | BUG_ON(flags & FOLL_GET); |
1339 | page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE); | |
1340 | goto out; | |
1341 | } | |
1342 | if (unlikely(pud_bad(*pud))) | |
1343 | goto no_page_table; | |
1344 | ||
1da177e4 | 1345 | pmd = pmd_offset(pud, address); |
aeed5fce | 1346 | if (pmd_none(*pmd)) |
deceb6cd | 1347 | goto no_page_table; |
71e3aac0 | 1348 | if (pmd_huge(*pmd) && vma->vm_flags & VM_HUGETLB) { |
deceb6cd HD |
1349 | BUG_ON(flags & FOLL_GET); |
1350 | page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE); | |
1da177e4 | 1351 | goto out; |
deceb6cd | 1352 | } |
71e3aac0 | 1353 | if (pmd_trans_huge(*pmd)) { |
500d65d4 AA |
1354 | if (flags & FOLL_SPLIT) { |
1355 | split_huge_page_pmd(mm, pmd); | |
1356 | goto split_fallthrough; | |
1357 | } | |
71e3aac0 AA |
1358 | spin_lock(&mm->page_table_lock); |
1359 | if (likely(pmd_trans_huge(*pmd))) { | |
1360 | if (unlikely(pmd_trans_splitting(*pmd))) { | |
1361 | spin_unlock(&mm->page_table_lock); | |
1362 | wait_split_huge_page(vma->anon_vma, pmd); | |
1363 | } else { | |
1364 | page = follow_trans_huge_pmd(mm, address, | |
1365 | pmd, flags); | |
1366 | spin_unlock(&mm->page_table_lock); | |
1367 | goto out; | |
1368 | } | |
1369 | } else | |
1370 | spin_unlock(&mm->page_table_lock); | |
1371 | /* fall through */ | |
1372 | } | |
500d65d4 | 1373 | split_fallthrough: |
aeed5fce HD |
1374 | if (unlikely(pmd_bad(*pmd))) |
1375 | goto no_page_table; | |
1376 | ||
deceb6cd | 1377 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); |
1da177e4 LT |
1378 | |
1379 | pte = *ptep; | |
deceb6cd | 1380 | if (!pte_present(pte)) |
89f5b7da | 1381 | goto no_page; |
deceb6cd HD |
1382 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
1383 | goto unlock; | |
a13ea5b7 | 1384 | |
6aab341e | 1385 | page = vm_normal_page(vma, address, pte); |
a13ea5b7 HD |
1386 | if (unlikely(!page)) { |
1387 | if ((flags & FOLL_DUMP) || | |
62eede62 | 1388 | !is_zero_pfn(pte_pfn(pte))) |
a13ea5b7 HD |
1389 | goto bad_page; |
1390 | page = pte_page(pte); | |
1391 | } | |
1da177e4 | 1392 | |
deceb6cd HD |
1393 | if (flags & FOLL_GET) |
1394 | get_page(page); | |
1395 | if (flags & FOLL_TOUCH) { | |
1396 | if ((flags & FOLL_WRITE) && | |
1397 | !pte_dirty(pte) && !PageDirty(page)) | |
1398 | set_page_dirty(page); | |
bd775c42 KM |
1399 | /* |
1400 | * pte_mkyoung() would be more correct here, but atomic care | |
1401 | * is needed to avoid losing the dirty bit: it is easier to use | |
1402 | * mark_page_accessed(). | |
1403 | */ | |
deceb6cd HD |
1404 | mark_page_accessed(page); |
1405 | } | |
a1fde08c | 1406 | if ((flags & FOLL_MLOCK) && (vma->vm_flags & VM_LOCKED)) { |
110d74a9 ML |
1407 | /* |
1408 | * The preliminary mapping check is mainly to avoid the | |
1409 | * pointless overhead of lock_page on the ZERO_PAGE | |
1410 | * which might bounce very badly if there is contention. | |
1411 | * | |
1412 | * If the page is already locked, we don't need to | |
1413 | * handle it now - vmscan will handle it later if and | |
1414 | * when it attempts to reclaim the page. | |
1415 | */ | |
1416 | if (page->mapping && trylock_page(page)) { | |
1417 | lru_add_drain(); /* push cached pages to LRU */ | |
1418 | /* | |
1419 | * Because we lock page here and migration is | |
1420 | * blocked by the pte's page reference, we need | |
1421 | * only check for file-cache page truncation. | |
1422 | */ | |
1423 | if (page->mapping) | |
1424 | mlock_vma_page(page); | |
1425 | unlock_page(page); | |
1426 | } | |
1427 | } | |
deceb6cd HD |
1428 | unlock: |
1429 | pte_unmap_unlock(ptep, ptl); | |
1da177e4 | 1430 | out: |
deceb6cd | 1431 | return page; |
1da177e4 | 1432 | |
89f5b7da LT |
1433 | bad_page: |
1434 | pte_unmap_unlock(ptep, ptl); | |
1435 | return ERR_PTR(-EFAULT); | |
1436 | ||
1437 | no_page: | |
1438 | pte_unmap_unlock(ptep, ptl); | |
1439 | if (!pte_none(pte)) | |
1440 | return page; | |
8e4b9a60 | 1441 | |
deceb6cd HD |
1442 | no_page_table: |
1443 | /* | |
1444 | * When core dumping an enormous anonymous area that nobody | |
8e4b9a60 HD |
1445 | * has touched so far, we don't want to allocate unnecessary pages or |
1446 | * page tables. Return error instead of NULL to skip handle_mm_fault, | |
1447 | * then get_dump_page() will return NULL to leave a hole in the dump. | |
1448 | * But we can only make this optimization where a hole would surely | |
1449 | * be zero-filled if handle_mm_fault() actually did handle it. | |
deceb6cd | 1450 | */ |
8e4b9a60 HD |
1451 | if ((flags & FOLL_DUMP) && |
1452 | (!vma->vm_ops || !vma->vm_ops->fault)) | |
1453 | return ERR_PTR(-EFAULT); | |
deceb6cd | 1454 | return page; |
1da177e4 LT |
1455 | } |
1456 | ||
95042f9e LT |
1457 | static inline int stack_guard_page(struct vm_area_struct *vma, unsigned long addr) |
1458 | { | |
a09a79f6 MP |
1459 | return stack_guard_page_start(vma, addr) || |
1460 | stack_guard_page_end(vma, addr+PAGE_SIZE); | |
95042f9e LT |
1461 | } |
1462 | ||
0014bd99 HY |
1463 | /** |
1464 | * __get_user_pages() - pin user pages in memory | |
1465 | * @tsk: task_struct of target task | |
1466 | * @mm: mm_struct of target mm | |
1467 | * @start: starting user address | |
1468 | * @nr_pages: number of pages from start to pin | |
1469 | * @gup_flags: flags modifying pin behaviour | |
1470 | * @pages: array that receives pointers to the pages pinned. | |
1471 | * Should be at least nr_pages long. Or NULL, if caller | |
1472 | * only intends to ensure the pages are faulted in. | |
1473 | * @vmas: array of pointers to vmas corresponding to each page. | |
1474 | * Or NULL if the caller does not require them. | |
1475 | * @nonblocking: whether waiting for disk IO or mmap_sem contention | |
1476 | * | |
1477 | * Returns number of pages pinned. This may be fewer than the number | |
1478 | * requested. If nr_pages is 0 or negative, returns 0. If no pages | |
1479 | * were pinned, returns -errno. Each page returned must be released | |
1480 | * with a put_page() call when it is finished with. vmas will only | |
1481 | * remain valid while mmap_sem is held. | |
1482 | * | |
1483 | * Must be called with mmap_sem held for read or write. | |
1484 | * | |
1485 | * __get_user_pages walks a process's page tables and takes a reference to | |
1486 | * each struct page that each user address corresponds to at a given | |
1487 | * instant. That is, it takes the page that would be accessed if a user | |
1488 | * thread accesses the given user virtual address at that instant. | |
1489 | * | |
1490 | * This does not guarantee that the page exists in the user mappings when | |
1491 | * __get_user_pages returns, and there may even be a completely different | |
1492 | * page there in some cases (eg. if mmapped pagecache has been invalidated | |
1493 | * and subsequently re faulted). However it does guarantee that the page | |
1494 | * won't be freed completely. And mostly callers simply care that the page | |
1495 | * contains data that was valid *at some point in time*. Typically, an IO | |
1496 | * or similar operation cannot guarantee anything stronger anyway because | |
1497 | * locks can't be held over the syscall boundary. | |
1498 | * | |
1499 | * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If | |
1500 | * the page is written to, set_page_dirty (or set_page_dirty_lock, as | |
1501 | * appropriate) must be called after the page is finished with, and | |
1502 | * before put_page is called. | |
1503 | * | |
1504 | * If @nonblocking != NULL, __get_user_pages will not wait for disk IO | |
1505 | * or mmap_sem contention, and if waiting is needed to pin all pages, | |
1506 | * *@nonblocking will be set to 0. | |
1507 | * | |
1508 | * In most cases, get_user_pages or get_user_pages_fast should be used | |
1509 | * instead of __get_user_pages. __get_user_pages should be used only if | |
1510 | * you need some special @gup_flags. | |
1511 | */ | |
b291f000 | 1512 | int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
58fa879e | 1513 | unsigned long start, int nr_pages, unsigned int gup_flags, |
53a7706d ML |
1514 | struct page **pages, struct vm_area_struct **vmas, |
1515 | int *nonblocking) | |
1da177e4 LT |
1516 | { |
1517 | int i; | |
58fa879e | 1518 | unsigned long vm_flags; |
1da177e4 | 1519 | |
9d73777e | 1520 | if (nr_pages <= 0) |
900cf086 | 1521 | return 0; |
58fa879e HD |
1522 | |
1523 | VM_BUG_ON(!!pages != !!(gup_flags & FOLL_GET)); | |
1524 | ||
1da177e4 LT |
1525 | /* |
1526 | * Require read or write permissions. | |
58fa879e | 1527 | * If FOLL_FORCE is set, we only require the "MAY" flags. |
1da177e4 | 1528 | */ |
58fa879e HD |
1529 | vm_flags = (gup_flags & FOLL_WRITE) ? |
1530 | (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); | |
1531 | vm_flags &= (gup_flags & FOLL_FORCE) ? | |
1532 | (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); | |
1da177e4 LT |
1533 | i = 0; |
1534 | ||
1535 | do { | |
deceb6cd | 1536 | struct vm_area_struct *vma; |
1da177e4 LT |
1537 | |
1538 | vma = find_extend_vma(mm, start); | |
e7f22e20 | 1539 | if (!vma && in_gate_area(mm, start)) { |
1da177e4 | 1540 | unsigned long pg = start & PAGE_MASK; |
1da177e4 LT |
1541 | pgd_t *pgd; |
1542 | pud_t *pud; | |
1543 | pmd_t *pmd; | |
1544 | pte_t *pte; | |
b291f000 NP |
1545 | |
1546 | /* user gate pages are read-only */ | |
58fa879e | 1547 | if (gup_flags & FOLL_WRITE) |
1da177e4 LT |
1548 | return i ? : -EFAULT; |
1549 | if (pg > TASK_SIZE) | |
1550 | pgd = pgd_offset_k(pg); | |
1551 | else | |
1552 | pgd = pgd_offset_gate(mm, pg); | |
1553 | BUG_ON(pgd_none(*pgd)); | |
1554 | pud = pud_offset(pgd, pg); | |
1555 | BUG_ON(pud_none(*pud)); | |
1556 | pmd = pmd_offset(pud, pg); | |
690dbe1c HD |
1557 | if (pmd_none(*pmd)) |
1558 | return i ? : -EFAULT; | |
f66055ab | 1559 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
1da177e4 | 1560 | pte = pte_offset_map(pmd, pg); |
690dbe1c HD |
1561 | if (pte_none(*pte)) { |
1562 | pte_unmap(pte); | |
1563 | return i ? : -EFAULT; | |
1564 | } | |
95042f9e | 1565 | vma = get_gate_vma(mm); |
1da177e4 | 1566 | if (pages) { |
de51257a HD |
1567 | struct page *page; |
1568 | ||
95042f9e | 1569 | page = vm_normal_page(vma, start, *pte); |
de51257a HD |
1570 | if (!page) { |
1571 | if (!(gup_flags & FOLL_DUMP) && | |
1572 | is_zero_pfn(pte_pfn(*pte))) | |
1573 | page = pte_page(*pte); | |
1574 | else { | |
1575 | pte_unmap(pte); | |
1576 | return i ? : -EFAULT; | |
1577 | } | |
1578 | } | |
6aab341e | 1579 | pages[i] = page; |
de51257a | 1580 | get_page(page); |
1da177e4 LT |
1581 | } |
1582 | pte_unmap(pte); | |
95042f9e | 1583 | goto next_page; |
1da177e4 LT |
1584 | } |
1585 | ||
b291f000 NP |
1586 | if (!vma || |
1587 | (vma->vm_flags & (VM_IO | VM_PFNMAP)) || | |
1c3aff1c | 1588 | !(vm_flags & vma->vm_flags)) |
1da177e4 LT |
1589 | return i ? : -EFAULT; |
1590 | ||
2a15efc9 HD |
1591 | if (is_vm_hugetlb_page(vma)) { |
1592 | i = follow_hugetlb_page(mm, vma, pages, vmas, | |
58fa879e | 1593 | &start, &nr_pages, i, gup_flags); |
2a15efc9 HD |
1594 | continue; |
1595 | } | |
deceb6cd | 1596 | |
1da177e4 | 1597 | do { |
08ef4729 | 1598 | struct page *page; |
58fa879e | 1599 | unsigned int foll_flags = gup_flags; |
1da177e4 | 1600 | |
462e00cc | 1601 | /* |
4779280d | 1602 | * If we have a pending SIGKILL, don't keep faulting |
1c3aff1c | 1603 | * pages and potentially allocating memory. |
462e00cc | 1604 | */ |
1c3aff1c | 1605 | if (unlikely(fatal_signal_pending(current))) |
4779280d | 1606 | return i ? i : -ERESTARTSYS; |
462e00cc | 1607 | |
deceb6cd | 1608 | cond_resched(); |
6aab341e | 1609 | while (!(page = follow_page(vma, start, foll_flags))) { |
deceb6cd | 1610 | int ret; |
53a7706d ML |
1611 | unsigned int fault_flags = 0; |
1612 | ||
a09a79f6 MP |
1613 | /* For mlock, just skip the stack guard page. */ |
1614 | if (foll_flags & FOLL_MLOCK) { | |
1615 | if (stack_guard_page(vma, start)) | |
1616 | goto next_page; | |
1617 | } | |
53a7706d ML |
1618 | if (foll_flags & FOLL_WRITE) |
1619 | fault_flags |= FAULT_FLAG_WRITE; | |
1620 | if (nonblocking) | |
1621 | fault_flags |= FAULT_FLAG_ALLOW_RETRY; | |
318b275f GN |
1622 | if (foll_flags & FOLL_NOWAIT) |
1623 | fault_flags |= (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT); | |
d06063cc | 1624 | |
d26ed650 | 1625 | ret = handle_mm_fault(mm, vma, start, |
53a7706d | 1626 | fault_flags); |
d26ed650 | 1627 | |
83c54070 NP |
1628 | if (ret & VM_FAULT_ERROR) { |
1629 | if (ret & VM_FAULT_OOM) | |
1630 | return i ? i : -ENOMEM; | |
69ebb83e HY |
1631 | if (ret & (VM_FAULT_HWPOISON | |
1632 | VM_FAULT_HWPOISON_LARGE)) { | |
1633 | if (i) | |
1634 | return i; | |
1635 | else if (gup_flags & FOLL_HWPOISON) | |
1636 | return -EHWPOISON; | |
1637 | else | |
1638 | return -EFAULT; | |
1639 | } | |
1640 | if (ret & VM_FAULT_SIGBUS) | |
83c54070 NP |
1641 | return i ? i : -EFAULT; |
1642 | BUG(); | |
1643 | } | |
e7f22e20 SW |
1644 | |
1645 | if (tsk) { | |
1646 | if (ret & VM_FAULT_MAJOR) | |
1647 | tsk->maj_flt++; | |
1648 | else | |
1649 | tsk->min_flt++; | |
1650 | } | |
83c54070 | 1651 | |
53a7706d | 1652 | if (ret & VM_FAULT_RETRY) { |
318b275f GN |
1653 | if (nonblocking) |
1654 | *nonblocking = 0; | |
53a7706d ML |
1655 | return i; |
1656 | } | |
1657 | ||
a68d2ebc | 1658 | /* |
83c54070 NP |
1659 | * The VM_FAULT_WRITE bit tells us that |
1660 | * do_wp_page has broken COW when necessary, | |
1661 | * even if maybe_mkwrite decided not to set | |
1662 | * pte_write. We can thus safely do subsequent | |
878b63ac HD |
1663 | * page lookups as if they were reads. But only |
1664 | * do so when looping for pte_write is futile: | |
1665 | * in some cases userspace may also be wanting | |
1666 | * to write to the gotten user page, which a | |
1667 | * read fault here might prevent (a readonly | |
1668 | * page might get reCOWed by userspace write). | |
a68d2ebc | 1669 | */ |
878b63ac HD |
1670 | if ((ret & VM_FAULT_WRITE) && |
1671 | !(vma->vm_flags & VM_WRITE)) | |
deceb6cd | 1672 | foll_flags &= ~FOLL_WRITE; |
83c54070 | 1673 | |
7f7bbbe5 | 1674 | cond_resched(); |
1da177e4 | 1675 | } |
89f5b7da LT |
1676 | if (IS_ERR(page)) |
1677 | return i ? i : PTR_ERR(page); | |
1da177e4 | 1678 | if (pages) { |
08ef4729 | 1679 | pages[i] = page; |
03beb076 | 1680 | |
a6f36be3 | 1681 | flush_anon_page(vma, page, start); |
08ef4729 | 1682 | flush_dcache_page(page); |
1da177e4 | 1683 | } |
95042f9e | 1684 | next_page: |
1da177e4 LT |
1685 | if (vmas) |
1686 | vmas[i] = vma; | |
1687 | i++; | |
1688 | start += PAGE_SIZE; | |
9d73777e PZ |
1689 | nr_pages--; |
1690 | } while (nr_pages && start < vma->vm_end); | |
1691 | } while (nr_pages); | |
1da177e4 LT |
1692 | return i; |
1693 | } | |
0014bd99 | 1694 | EXPORT_SYMBOL(__get_user_pages); |
b291f000 | 1695 | |
d2bf6be8 NP |
1696 | /** |
1697 | * get_user_pages() - pin user pages in memory | |
e7f22e20 SW |
1698 | * @tsk: the task_struct to use for page fault accounting, or |
1699 | * NULL if faults are not to be recorded. | |
d2bf6be8 NP |
1700 | * @mm: mm_struct of target mm |
1701 | * @start: starting user address | |
9d73777e | 1702 | * @nr_pages: number of pages from start to pin |
d2bf6be8 NP |
1703 | * @write: whether pages will be written to by the caller |
1704 | * @force: whether to force write access even if user mapping is | |
1705 | * readonly. This will result in the page being COWed even | |
1706 | * in MAP_SHARED mappings. You do not want this. | |
1707 | * @pages: array that receives pointers to the pages pinned. | |
1708 | * Should be at least nr_pages long. Or NULL, if caller | |
1709 | * only intends to ensure the pages are faulted in. | |
1710 | * @vmas: array of pointers to vmas corresponding to each page. | |
1711 | * Or NULL if the caller does not require them. | |
1712 | * | |
1713 | * Returns number of pages pinned. This may be fewer than the number | |
9d73777e | 1714 | * requested. If nr_pages is 0 or negative, returns 0. If no pages |
d2bf6be8 NP |
1715 | * were pinned, returns -errno. Each page returned must be released |
1716 | * with a put_page() call when it is finished with. vmas will only | |
1717 | * remain valid while mmap_sem is held. | |
1718 | * | |
1719 | * Must be called with mmap_sem held for read or write. | |
1720 | * | |
1721 | * get_user_pages walks a process's page tables and takes a reference to | |
1722 | * each struct page that each user address corresponds to at a given | |
1723 | * instant. That is, it takes the page that would be accessed if a user | |
1724 | * thread accesses the given user virtual address at that instant. | |
1725 | * | |
1726 | * This does not guarantee that the page exists in the user mappings when | |
1727 | * get_user_pages returns, and there may even be a completely different | |
1728 | * page there in some cases (eg. if mmapped pagecache has been invalidated | |
1729 | * and subsequently re faulted). However it does guarantee that the page | |
1730 | * won't be freed completely. And mostly callers simply care that the page | |
1731 | * contains data that was valid *at some point in time*. Typically, an IO | |
1732 | * or similar operation cannot guarantee anything stronger anyway because | |
1733 | * locks can't be held over the syscall boundary. | |
1734 | * | |
1735 | * If write=0, the page must not be written to. If the page is written to, | |
1736 | * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called | |
1737 | * after the page is finished with, and before put_page is called. | |
1738 | * | |
1739 | * get_user_pages is typically used for fewer-copy IO operations, to get a | |
1740 | * handle on the memory by some means other than accesses via the user virtual | |
1741 | * addresses. The pages may be submitted for DMA to devices or accessed via | |
1742 | * their kernel linear mapping (via the kmap APIs). Care should be taken to | |
1743 | * use the correct cache flushing APIs. | |
1744 | * | |
1745 | * See also get_user_pages_fast, for performance critical applications. | |
1746 | */ | |
b291f000 | 1747 | int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
9d73777e | 1748 | unsigned long start, int nr_pages, int write, int force, |
b291f000 NP |
1749 | struct page **pages, struct vm_area_struct **vmas) |
1750 | { | |
58fa879e | 1751 | int flags = FOLL_TOUCH; |
b291f000 | 1752 | |
58fa879e HD |
1753 | if (pages) |
1754 | flags |= FOLL_GET; | |
b291f000 | 1755 | if (write) |
58fa879e | 1756 | flags |= FOLL_WRITE; |
b291f000 | 1757 | if (force) |
58fa879e | 1758 | flags |= FOLL_FORCE; |
b291f000 | 1759 | |
53a7706d ML |
1760 | return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas, |
1761 | NULL); | |
b291f000 | 1762 | } |
1da177e4 LT |
1763 | EXPORT_SYMBOL(get_user_pages); |
1764 | ||
f3e8fccd HD |
1765 | /** |
1766 | * get_dump_page() - pin user page in memory while writing it to core dump | |
1767 | * @addr: user address | |
1768 | * | |
1769 | * Returns struct page pointer of user page pinned for dump, | |
1770 | * to be freed afterwards by page_cache_release() or put_page(). | |
1771 | * | |
1772 | * Returns NULL on any kind of failure - a hole must then be inserted into | |
1773 | * the corefile, to preserve alignment with its headers; and also returns | |
1774 | * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found - | |
1775 | * allowing a hole to be left in the corefile to save diskspace. | |
1776 | * | |
1777 | * Called without mmap_sem, but after all other threads have been killed. | |
1778 | */ | |
1779 | #ifdef CONFIG_ELF_CORE | |
1780 | struct page *get_dump_page(unsigned long addr) | |
1781 | { | |
1782 | struct vm_area_struct *vma; | |
1783 | struct page *page; | |
1784 | ||
1785 | if (__get_user_pages(current, current->mm, addr, 1, | |
53a7706d ML |
1786 | FOLL_FORCE | FOLL_DUMP | FOLL_GET, &page, &vma, |
1787 | NULL) < 1) | |
f3e8fccd | 1788 | return NULL; |
f3e8fccd HD |
1789 | flush_cache_page(vma, addr, page_to_pfn(page)); |
1790 | return page; | |
1791 | } | |
1792 | #endif /* CONFIG_ELF_CORE */ | |
1793 | ||
25ca1d6c | 1794 | pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, |
920c7a5d | 1795 | spinlock_t **ptl) |
c9cfcddf LT |
1796 | { |
1797 | pgd_t * pgd = pgd_offset(mm, addr); | |
1798 | pud_t * pud = pud_alloc(mm, pgd, addr); | |
1799 | if (pud) { | |
49c91fb0 | 1800 | pmd_t * pmd = pmd_alloc(mm, pud, addr); |
f66055ab AA |
1801 | if (pmd) { |
1802 | VM_BUG_ON(pmd_trans_huge(*pmd)); | |
c9cfcddf | 1803 | return pte_alloc_map_lock(mm, pmd, addr, ptl); |
f66055ab | 1804 | } |
c9cfcddf LT |
1805 | } |
1806 | return NULL; | |
1807 | } | |
1808 | ||
238f58d8 LT |
1809 | /* |
1810 | * This is the old fallback for page remapping. | |
1811 | * | |
1812 | * For historical reasons, it only allows reserved pages. Only | |
1813 | * old drivers should use this, and they needed to mark their | |
1814 | * pages reserved for the old functions anyway. | |
1815 | */ | |
423bad60 NP |
1816 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
1817 | struct page *page, pgprot_t prot) | |
238f58d8 | 1818 | { |
423bad60 | 1819 | struct mm_struct *mm = vma->vm_mm; |
238f58d8 | 1820 | int retval; |
c9cfcddf | 1821 | pte_t *pte; |
8a9f3ccd BS |
1822 | spinlock_t *ptl; |
1823 | ||
238f58d8 | 1824 | retval = -EINVAL; |
a145dd41 | 1825 | if (PageAnon(page)) |
5b4e655e | 1826 | goto out; |
238f58d8 LT |
1827 | retval = -ENOMEM; |
1828 | flush_dcache_page(page); | |
c9cfcddf | 1829 | pte = get_locked_pte(mm, addr, &ptl); |
238f58d8 | 1830 | if (!pte) |
5b4e655e | 1831 | goto out; |
238f58d8 LT |
1832 | retval = -EBUSY; |
1833 | if (!pte_none(*pte)) | |
1834 | goto out_unlock; | |
1835 | ||
1836 | /* Ok, finally just insert the thing.. */ | |
1837 | get_page(page); | |
34e55232 | 1838 | inc_mm_counter_fast(mm, MM_FILEPAGES); |
238f58d8 LT |
1839 | page_add_file_rmap(page); |
1840 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); | |
1841 | ||
1842 | retval = 0; | |
8a9f3ccd BS |
1843 | pte_unmap_unlock(pte, ptl); |
1844 | return retval; | |
238f58d8 LT |
1845 | out_unlock: |
1846 | pte_unmap_unlock(pte, ptl); | |
1847 | out: | |
1848 | return retval; | |
1849 | } | |
1850 | ||
bfa5bf6d REB |
1851 | /** |
1852 | * vm_insert_page - insert single page into user vma | |
1853 | * @vma: user vma to map to | |
1854 | * @addr: target user address of this page | |
1855 | * @page: source kernel page | |
1856 | * | |
a145dd41 LT |
1857 | * This allows drivers to insert individual pages they've allocated |
1858 | * into a user vma. | |
1859 | * | |
1860 | * The page has to be a nice clean _individual_ kernel allocation. | |
1861 | * If you allocate a compound page, you need to have marked it as | |
1862 | * such (__GFP_COMP), or manually just split the page up yourself | |
8dfcc9ba | 1863 | * (see split_page()). |
a145dd41 LT |
1864 | * |
1865 | * NOTE! Traditionally this was done with "remap_pfn_range()" which | |
1866 | * took an arbitrary page protection parameter. This doesn't allow | |
1867 | * that. Your vma protection will have to be set up correctly, which | |
1868 | * means that if you want a shared writable mapping, you'd better | |
1869 | * ask for a shared writable mapping! | |
1870 | * | |
1871 | * The page does not need to be reserved. | |
1872 | */ | |
423bad60 NP |
1873 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
1874 | struct page *page) | |
a145dd41 LT |
1875 | { |
1876 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
1877 | return -EFAULT; | |
1878 | if (!page_count(page)) | |
1879 | return -EINVAL; | |
4d7672b4 | 1880 | vma->vm_flags |= VM_INSERTPAGE; |
423bad60 | 1881 | return insert_page(vma, addr, page, vma->vm_page_prot); |
a145dd41 | 1882 | } |
e3c3374f | 1883 | EXPORT_SYMBOL(vm_insert_page); |
a145dd41 | 1884 | |
423bad60 NP |
1885 | static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
1886 | unsigned long pfn, pgprot_t prot) | |
1887 | { | |
1888 | struct mm_struct *mm = vma->vm_mm; | |
1889 | int retval; | |
1890 | pte_t *pte, entry; | |
1891 | spinlock_t *ptl; | |
1892 | ||
1893 | retval = -ENOMEM; | |
1894 | pte = get_locked_pte(mm, addr, &ptl); | |
1895 | if (!pte) | |
1896 | goto out; | |
1897 | retval = -EBUSY; | |
1898 | if (!pte_none(*pte)) | |
1899 | goto out_unlock; | |
1900 | ||
1901 | /* Ok, finally just insert the thing.. */ | |
1902 | entry = pte_mkspecial(pfn_pte(pfn, prot)); | |
1903 | set_pte_at(mm, addr, pte, entry); | |
4b3073e1 | 1904 | update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ |
423bad60 NP |
1905 | |
1906 | retval = 0; | |
1907 | out_unlock: | |
1908 | pte_unmap_unlock(pte, ptl); | |
1909 | out: | |
1910 | return retval; | |
1911 | } | |
1912 | ||
e0dc0d8f NP |
1913 | /** |
1914 | * vm_insert_pfn - insert single pfn into user vma | |
1915 | * @vma: user vma to map to | |
1916 | * @addr: target user address of this page | |
1917 | * @pfn: source kernel pfn | |
1918 | * | |
1919 | * Similar to vm_inert_page, this allows drivers to insert individual pages | |
1920 | * they've allocated into a user vma. Same comments apply. | |
1921 | * | |
1922 | * This function should only be called from a vm_ops->fault handler, and | |
1923 | * in that case the handler should return NULL. | |
0d71d10a NP |
1924 | * |
1925 | * vma cannot be a COW mapping. | |
1926 | * | |
1927 | * As this is called only for pages that do not currently exist, we | |
1928 | * do not need to flush old virtual caches or the TLB. | |
e0dc0d8f NP |
1929 | */ |
1930 | int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, | |
423bad60 | 1931 | unsigned long pfn) |
e0dc0d8f | 1932 | { |
2ab64037 | 1933 | int ret; |
e4b866ed | 1934 | pgprot_t pgprot = vma->vm_page_prot; |
7e675137 NP |
1935 | /* |
1936 | * Technically, architectures with pte_special can avoid all these | |
1937 | * restrictions (same for remap_pfn_range). However we would like | |
1938 | * consistency in testing and feature parity among all, so we should | |
1939 | * try to keep these invariants in place for everybody. | |
1940 | */ | |
b379d790 JH |
1941 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); |
1942 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == | |
1943 | (VM_PFNMAP|VM_MIXEDMAP)); | |
1944 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); | |
1945 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); | |
e0dc0d8f | 1946 | |
423bad60 NP |
1947 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1948 | return -EFAULT; | |
e4b866ed | 1949 | if (track_pfn_vma_new(vma, &pgprot, pfn, PAGE_SIZE)) |
2ab64037 | 1950 | return -EINVAL; |
1951 | ||
e4b866ed | 1952 | ret = insert_pfn(vma, addr, pfn, pgprot); |
2ab64037 | 1953 | |
1954 | if (ret) | |
1955 | untrack_pfn_vma(vma, pfn, PAGE_SIZE); | |
1956 | ||
1957 | return ret; | |
423bad60 NP |
1958 | } |
1959 | EXPORT_SYMBOL(vm_insert_pfn); | |
e0dc0d8f | 1960 | |
423bad60 NP |
1961 | int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, |
1962 | unsigned long pfn) | |
1963 | { | |
1964 | BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); | |
e0dc0d8f | 1965 | |
423bad60 NP |
1966 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1967 | return -EFAULT; | |
e0dc0d8f | 1968 | |
423bad60 NP |
1969 | /* |
1970 | * If we don't have pte special, then we have to use the pfn_valid() | |
1971 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* | |
1972 | * refcount the page if pfn_valid is true (hence insert_page rather | |
62eede62 HD |
1973 | * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP |
1974 | * without pte special, it would there be refcounted as a normal page. | |
423bad60 NP |
1975 | */ |
1976 | if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) { | |
1977 | struct page *page; | |
1978 | ||
1979 | page = pfn_to_page(pfn); | |
1980 | return insert_page(vma, addr, page, vma->vm_page_prot); | |
1981 | } | |
1982 | return insert_pfn(vma, addr, pfn, vma->vm_page_prot); | |
e0dc0d8f | 1983 | } |
423bad60 | 1984 | EXPORT_SYMBOL(vm_insert_mixed); |
e0dc0d8f | 1985 | |
1da177e4 LT |
1986 | /* |
1987 | * maps a range of physical memory into the requested pages. the old | |
1988 | * mappings are removed. any references to nonexistent pages results | |
1989 | * in null mappings (currently treated as "copy-on-access") | |
1990 | */ | |
1991 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
1992 | unsigned long addr, unsigned long end, | |
1993 | unsigned long pfn, pgprot_t prot) | |
1994 | { | |
1995 | pte_t *pte; | |
c74df32c | 1996 | spinlock_t *ptl; |
1da177e4 | 1997 | |
c74df32c | 1998 | pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1da177e4 LT |
1999 | if (!pte) |
2000 | return -ENOMEM; | |
6606c3e0 | 2001 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
2002 | do { |
2003 | BUG_ON(!pte_none(*pte)); | |
7e675137 | 2004 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
1da177e4 LT |
2005 | pfn++; |
2006 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 2007 | arch_leave_lazy_mmu_mode(); |
c74df32c | 2008 | pte_unmap_unlock(pte - 1, ptl); |
1da177e4 LT |
2009 | return 0; |
2010 | } | |
2011 | ||
2012 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
2013 | unsigned long addr, unsigned long end, | |
2014 | unsigned long pfn, pgprot_t prot) | |
2015 | { | |
2016 | pmd_t *pmd; | |
2017 | unsigned long next; | |
2018 | ||
2019 | pfn -= addr >> PAGE_SHIFT; | |
2020 | pmd = pmd_alloc(mm, pud, addr); | |
2021 | if (!pmd) | |
2022 | return -ENOMEM; | |
f66055ab | 2023 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
1da177e4 LT |
2024 | do { |
2025 | next = pmd_addr_end(addr, end); | |
2026 | if (remap_pte_range(mm, pmd, addr, next, | |
2027 | pfn + (addr >> PAGE_SHIFT), prot)) | |
2028 | return -ENOMEM; | |
2029 | } while (pmd++, addr = next, addr != end); | |
2030 | return 0; | |
2031 | } | |
2032 | ||
2033 | static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, | |
2034 | unsigned long addr, unsigned long end, | |
2035 | unsigned long pfn, pgprot_t prot) | |
2036 | { | |
2037 | pud_t *pud; | |
2038 | unsigned long next; | |
2039 | ||
2040 | pfn -= addr >> PAGE_SHIFT; | |
2041 | pud = pud_alloc(mm, pgd, addr); | |
2042 | if (!pud) | |
2043 | return -ENOMEM; | |
2044 | do { | |
2045 | next = pud_addr_end(addr, end); | |
2046 | if (remap_pmd_range(mm, pud, addr, next, | |
2047 | pfn + (addr >> PAGE_SHIFT), prot)) | |
2048 | return -ENOMEM; | |
2049 | } while (pud++, addr = next, addr != end); | |
2050 | return 0; | |
2051 | } | |
2052 | ||
bfa5bf6d REB |
2053 | /** |
2054 | * remap_pfn_range - remap kernel memory to userspace | |
2055 | * @vma: user vma to map to | |
2056 | * @addr: target user address to start at | |
2057 | * @pfn: physical address of kernel memory | |
2058 | * @size: size of map area | |
2059 | * @prot: page protection flags for this mapping | |
2060 | * | |
2061 | * Note: this is only safe if the mm semaphore is held when called. | |
2062 | */ | |
1da177e4 LT |
2063 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
2064 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
2065 | { | |
2066 | pgd_t *pgd; | |
2067 | unsigned long next; | |
2d15cab8 | 2068 | unsigned long end = addr + PAGE_ALIGN(size); |
1da177e4 LT |
2069 | struct mm_struct *mm = vma->vm_mm; |
2070 | int err; | |
2071 | ||
2072 | /* | |
2073 | * Physically remapped pages are special. Tell the | |
2074 | * rest of the world about it: | |
2075 | * VM_IO tells people not to look at these pages | |
2076 | * (accesses can have side effects). | |
0b14c179 HD |
2077 | * VM_RESERVED is specified all over the place, because |
2078 | * in 2.4 it kept swapout's vma scan off this vma; but | |
2079 | * in 2.6 the LRU scan won't even find its pages, so this | |
2080 | * flag means no more than count its pages in reserved_vm, | |
2081 | * and omit it from core dump, even when VM_IO turned off. | |
6aab341e LT |
2082 | * VM_PFNMAP tells the core MM that the base pages are just |
2083 | * raw PFN mappings, and do not have a "struct page" associated | |
2084 | * with them. | |
fb155c16 LT |
2085 | * |
2086 | * There's a horrible special case to handle copy-on-write | |
2087 | * behaviour that some programs depend on. We mark the "original" | |
2088 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". | |
1da177e4 | 2089 | */ |
4bb9c5c0 | 2090 | if (addr == vma->vm_start && end == vma->vm_end) { |
fb155c16 | 2091 | vma->vm_pgoff = pfn; |
895791da | 2092 | vma->vm_flags |= VM_PFN_AT_MMAP; |
4bb9c5c0 | 2093 | } else if (is_cow_mapping(vma->vm_flags)) |
3c8bb73a | 2094 | return -EINVAL; |
fb155c16 | 2095 | |
6aab341e | 2096 | vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP; |
1da177e4 | 2097 | |
e4b866ed | 2098 | err = track_pfn_vma_new(vma, &prot, pfn, PAGE_ALIGN(size)); |
a3670613 | 2099 | if (err) { |
2100 | /* | |
2101 | * To indicate that track_pfn related cleanup is not | |
2102 | * needed from higher level routine calling unmap_vmas | |
2103 | */ | |
2104 | vma->vm_flags &= ~(VM_IO | VM_RESERVED | VM_PFNMAP); | |
895791da | 2105 | vma->vm_flags &= ~VM_PFN_AT_MMAP; |
2ab64037 | 2106 | return -EINVAL; |
a3670613 | 2107 | } |
2ab64037 | 2108 | |
1da177e4 LT |
2109 | BUG_ON(addr >= end); |
2110 | pfn -= addr >> PAGE_SHIFT; | |
2111 | pgd = pgd_offset(mm, addr); | |
2112 | flush_cache_range(vma, addr, end); | |
1da177e4 LT |
2113 | do { |
2114 | next = pgd_addr_end(addr, end); | |
2115 | err = remap_pud_range(mm, pgd, addr, next, | |
2116 | pfn + (addr >> PAGE_SHIFT), prot); | |
2117 | if (err) | |
2118 | break; | |
2119 | } while (pgd++, addr = next, addr != end); | |
2ab64037 | 2120 | |
2121 | if (err) | |
2122 | untrack_pfn_vma(vma, pfn, PAGE_ALIGN(size)); | |
2123 | ||
1da177e4 LT |
2124 | return err; |
2125 | } | |
2126 | EXPORT_SYMBOL(remap_pfn_range); | |
2127 | ||
aee16b3c JF |
2128 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
2129 | unsigned long addr, unsigned long end, | |
2130 | pte_fn_t fn, void *data) | |
2131 | { | |
2132 | pte_t *pte; | |
2133 | int err; | |
2f569afd | 2134 | pgtable_t token; |
94909914 | 2135 | spinlock_t *uninitialized_var(ptl); |
aee16b3c JF |
2136 | |
2137 | pte = (mm == &init_mm) ? | |
2138 | pte_alloc_kernel(pmd, addr) : | |
2139 | pte_alloc_map_lock(mm, pmd, addr, &ptl); | |
2140 | if (!pte) | |
2141 | return -ENOMEM; | |
2142 | ||
2143 | BUG_ON(pmd_huge(*pmd)); | |
2144 | ||
38e0edb1 JF |
2145 | arch_enter_lazy_mmu_mode(); |
2146 | ||
2f569afd | 2147 | token = pmd_pgtable(*pmd); |
aee16b3c JF |
2148 | |
2149 | do { | |
c36987e2 | 2150 | err = fn(pte++, token, addr, data); |
aee16b3c JF |
2151 | if (err) |
2152 | break; | |
c36987e2 | 2153 | } while (addr += PAGE_SIZE, addr != end); |
aee16b3c | 2154 | |
38e0edb1 JF |
2155 | arch_leave_lazy_mmu_mode(); |
2156 | ||
aee16b3c JF |
2157 | if (mm != &init_mm) |
2158 | pte_unmap_unlock(pte-1, ptl); | |
2159 | return err; | |
2160 | } | |
2161 | ||
2162 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, | |
2163 | unsigned long addr, unsigned long end, | |
2164 | pte_fn_t fn, void *data) | |
2165 | { | |
2166 | pmd_t *pmd; | |
2167 | unsigned long next; | |
2168 | int err; | |
2169 | ||
ceb86879 AK |
2170 | BUG_ON(pud_huge(*pud)); |
2171 | ||
aee16b3c JF |
2172 | pmd = pmd_alloc(mm, pud, addr); |
2173 | if (!pmd) | |
2174 | return -ENOMEM; | |
2175 | do { | |
2176 | next = pmd_addr_end(addr, end); | |
2177 | err = apply_to_pte_range(mm, pmd, addr, next, fn, data); | |
2178 | if (err) | |
2179 | break; | |
2180 | } while (pmd++, addr = next, addr != end); | |
2181 | return err; | |
2182 | } | |
2183 | ||
2184 | static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd, | |
2185 | unsigned long addr, unsigned long end, | |
2186 | pte_fn_t fn, void *data) | |
2187 | { | |
2188 | pud_t *pud; | |
2189 | unsigned long next; | |
2190 | int err; | |
2191 | ||
2192 | pud = pud_alloc(mm, pgd, addr); | |
2193 | if (!pud) | |
2194 | return -ENOMEM; | |
2195 | do { | |
2196 | next = pud_addr_end(addr, end); | |
2197 | err = apply_to_pmd_range(mm, pud, addr, next, fn, data); | |
2198 | if (err) | |
2199 | break; | |
2200 | } while (pud++, addr = next, addr != end); | |
2201 | return err; | |
2202 | } | |
2203 | ||
2204 | /* | |
2205 | * Scan a region of virtual memory, filling in page tables as necessary | |
2206 | * and calling a provided function on each leaf page table. | |
2207 | */ | |
2208 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, | |
2209 | unsigned long size, pte_fn_t fn, void *data) | |
2210 | { | |
2211 | pgd_t *pgd; | |
2212 | unsigned long next; | |
57250a5b | 2213 | unsigned long end = addr + size; |
aee16b3c JF |
2214 | int err; |
2215 | ||
2216 | BUG_ON(addr >= end); | |
2217 | pgd = pgd_offset(mm, addr); | |
2218 | do { | |
2219 | next = pgd_addr_end(addr, end); | |
2220 | err = apply_to_pud_range(mm, pgd, addr, next, fn, data); | |
2221 | if (err) | |
2222 | break; | |
2223 | } while (pgd++, addr = next, addr != end); | |
57250a5b | 2224 | |
aee16b3c JF |
2225 | return err; |
2226 | } | |
2227 | EXPORT_SYMBOL_GPL(apply_to_page_range); | |
2228 | ||
8f4e2101 HD |
2229 | /* |
2230 | * handle_pte_fault chooses page fault handler according to an entry | |
2231 | * which was read non-atomically. Before making any commitment, on | |
2232 | * those architectures or configurations (e.g. i386 with PAE) which | |
a335b2e1 | 2233 | * might give a mix of unmatched parts, do_swap_page and do_nonlinear_fault |
8f4e2101 HD |
2234 | * must check under lock before unmapping the pte and proceeding |
2235 | * (but do_wp_page is only called after already making such a check; | |
a335b2e1 | 2236 | * and do_anonymous_page can safely check later on). |
8f4e2101 | 2237 | */ |
4c21e2f2 | 2238 | static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, |
8f4e2101 HD |
2239 | pte_t *page_table, pte_t orig_pte) |
2240 | { | |
2241 | int same = 1; | |
2242 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) | |
2243 | if (sizeof(pte_t) > sizeof(unsigned long)) { | |
4c21e2f2 HD |
2244 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
2245 | spin_lock(ptl); | |
8f4e2101 | 2246 | same = pte_same(*page_table, orig_pte); |
4c21e2f2 | 2247 | spin_unlock(ptl); |
8f4e2101 HD |
2248 | } |
2249 | #endif | |
2250 | pte_unmap(page_table); | |
2251 | return same; | |
2252 | } | |
2253 | ||
9de455b2 | 2254 | static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) |
6aab341e LT |
2255 | { |
2256 | /* | |
2257 | * If the source page was a PFN mapping, we don't have | |
2258 | * a "struct page" for it. We do a best-effort copy by | |
2259 | * just copying from the original user address. If that | |
2260 | * fails, we just zero-fill it. Live with it. | |
2261 | */ | |
2262 | if (unlikely(!src)) { | |
2263 | void *kaddr = kmap_atomic(dst, KM_USER0); | |
5d2a2dbb LT |
2264 | void __user *uaddr = (void __user *)(va & PAGE_MASK); |
2265 | ||
2266 | /* | |
2267 | * This really shouldn't fail, because the page is there | |
2268 | * in the page tables. But it might just be unreadable, | |
2269 | * in which case we just give up and fill the result with | |
2270 | * zeroes. | |
2271 | */ | |
2272 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) | |
3ecb01df | 2273 | clear_page(kaddr); |
6aab341e | 2274 | kunmap_atomic(kaddr, KM_USER0); |
c4ec7b0d | 2275 | flush_dcache_page(dst); |
0ed361de NP |
2276 | } else |
2277 | copy_user_highpage(dst, src, va, vma); | |
6aab341e LT |
2278 | } |
2279 | ||
1da177e4 LT |
2280 | /* |
2281 | * This routine handles present pages, when users try to write | |
2282 | * to a shared page. It is done by copying the page to a new address | |
2283 | * and decrementing the shared-page counter for the old page. | |
2284 | * | |
1da177e4 LT |
2285 | * Note that this routine assumes that the protection checks have been |
2286 | * done by the caller (the low-level page fault routine in most cases). | |
2287 | * Thus we can safely just mark it writable once we've done any necessary | |
2288 | * COW. | |
2289 | * | |
2290 | * We also mark the page dirty at this point even though the page will | |
2291 | * change only once the write actually happens. This avoids a few races, | |
2292 | * and potentially makes it more efficient. | |
2293 | * | |
8f4e2101 HD |
2294 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2295 | * but allow concurrent faults), with pte both mapped and locked. | |
2296 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2297 | */ |
65500d23 HD |
2298 | static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2299 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
8f4e2101 | 2300 | spinlock_t *ptl, pte_t orig_pte) |
e6219ec8 | 2301 | __releases(ptl) |
1da177e4 | 2302 | { |
e5bbe4df | 2303 | struct page *old_page, *new_page; |
1da177e4 | 2304 | pte_t entry; |
b009c024 | 2305 | int ret = 0; |
a200ee18 | 2306 | int page_mkwrite = 0; |
d08b3851 | 2307 | struct page *dirty_page = NULL; |
1da177e4 | 2308 | |
6aab341e | 2309 | old_page = vm_normal_page(vma, address, orig_pte); |
251b97f5 PZ |
2310 | if (!old_page) { |
2311 | /* | |
2312 | * VM_MIXEDMAP !pfn_valid() case | |
2313 | * | |
2314 | * We should not cow pages in a shared writeable mapping. | |
2315 | * Just mark the pages writable as we can't do any dirty | |
2316 | * accounting on raw pfn maps. | |
2317 | */ | |
2318 | if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == | |
2319 | (VM_WRITE|VM_SHARED)) | |
2320 | goto reuse; | |
6aab341e | 2321 | goto gotten; |
251b97f5 | 2322 | } |
1da177e4 | 2323 | |
d08b3851 | 2324 | /* |
ee6a6457 PZ |
2325 | * Take out anonymous pages first, anonymous shared vmas are |
2326 | * not dirty accountable. | |
d08b3851 | 2327 | */ |
9a840895 | 2328 | if (PageAnon(old_page) && !PageKsm(old_page)) { |
ab967d86 HD |
2329 | if (!trylock_page(old_page)) { |
2330 | page_cache_get(old_page); | |
2331 | pte_unmap_unlock(page_table, ptl); | |
2332 | lock_page(old_page); | |
2333 | page_table = pte_offset_map_lock(mm, pmd, address, | |
2334 | &ptl); | |
2335 | if (!pte_same(*page_table, orig_pte)) { | |
2336 | unlock_page(old_page); | |
ab967d86 HD |
2337 | goto unlock; |
2338 | } | |
2339 | page_cache_release(old_page); | |
ee6a6457 | 2340 | } |
b009c024 | 2341 | if (reuse_swap_page(old_page)) { |
c44b6743 RR |
2342 | /* |
2343 | * The page is all ours. Move it to our anon_vma so | |
2344 | * the rmap code will not search our parent or siblings. | |
2345 | * Protected against the rmap code by the page lock. | |
2346 | */ | |
2347 | page_move_anon_rmap(old_page, vma, address); | |
b009c024 ML |
2348 | unlock_page(old_page); |
2349 | goto reuse; | |
2350 | } | |
ab967d86 | 2351 | unlock_page(old_page); |
ee6a6457 | 2352 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
d08b3851 | 2353 | (VM_WRITE|VM_SHARED))) { |
ee6a6457 PZ |
2354 | /* |
2355 | * Only catch write-faults on shared writable pages, | |
2356 | * read-only shared pages can get COWed by | |
2357 | * get_user_pages(.write=1, .force=1). | |
2358 | */ | |
9637a5ef | 2359 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
c2ec175c NP |
2360 | struct vm_fault vmf; |
2361 | int tmp; | |
2362 | ||
2363 | vmf.virtual_address = (void __user *)(address & | |
2364 | PAGE_MASK); | |
2365 | vmf.pgoff = old_page->index; | |
2366 | vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; | |
2367 | vmf.page = old_page; | |
2368 | ||
9637a5ef DH |
2369 | /* |
2370 | * Notify the address space that the page is about to | |
2371 | * become writable so that it can prohibit this or wait | |
2372 | * for the page to get into an appropriate state. | |
2373 | * | |
2374 | * We do this without the lock held, so that it can | |
2375 | * sleep if it needs to. | |
2376 | */ | |
2377 | page_cache_get(old_page); | |
2378 | pte_unmap_unlock(page_table, ptl); | |
2379 | ||
c2ec175c NP |
2380 | tmp = vma->vm_ops->page_mkwrite(vma, &vmf); |
2381 | if (unlikely(tmp & | |
2382 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { | |
2383 | ret = tmp; | |
9637a5ef | 2384 | goto unwritable_page; |
c2ec175c | 2385 | } |
b827e496 NP |
2386 | if (unlikely(!(tmp & VM_FAULT_LOCKED))) { |
2387 | lock_page(old_page); | |
2388 | if (!old_page->mapping) { | |
2389 | ret = 0; /* retry the fault */ | |
2390 | unlock_page(old_page); | |
2391 | goto unwritable_page; | |
2392 | } | |
2393 | } else | |
2394 | VM_BUG_ON(!PageLocked(old_page)); | |
9637a5ef | 2395 | |
9637a5ef DH |
2396 | /* |
2397 | * Since we dropped the lock we need to revalidate | |
2398 | * the PTE as someone else may have changed it. If | |
2399 | * they did, we just return, as we can count on the | |
2400 | * MMU to tell us if they didn't also make it writable. | |
2401 | */ | |
2402 | page_table = pte_offset_map_lock(mm, pmd, address, | |
2403 | &ptl); | |
b827e496 NP |
2404 | if (!pte_same(*page_table, orig_pte)) { |
2405 | unlock_page(old_page); | |
9637a5ef | 2406 | goto unlock; |
b827e496 | 2407 | } |
a200ee18 PZ |
2408 | |
2409 | page_mkwrite = 1; | |
1da177e4 | 2410 | } |
d08b3851 PZ |
2411 | dirty_page = old_page; |
2412 | get_page(dirty_page); | |
9637a5ef | 2413 | |
251b97f5 | 2414 | reuse: |
9637a5ef DH |
2415 | flush_cache_page(vma, address, pte_pfn(orig_pte)); |
2416 | entry = pte_mkyoung(orig_pte); | |
2417 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
954ffcb3 | 2418 | if (ptep_set_access_flags(vma, address, page_table, entry,1)) |
4b3073e1 | 2419 | update_mmu_cache(vma, address, page_table); |
72ddc8f7 | 2420 | pte_unmap_unlock(page_table, ptl); |
9637a5ef | 2421 | ret |= VM_FAULT_WRITE; |
72ddc8f7 ML |
2422 | |
2423 | if (!dirty_page) | |
2424 | return ret; | |
2425 | ||
2426 | /* | |
2427 | * Yes, Virginia, this is actually required to prevent a race | |
2428 | * with clear_page_dirty_for_io() from clearing the page dirty | |
2429 | * bit after it clear all dirty ptes, but before a racing | |
2430 | * do_wp_page installs a dirty pte. | |
2431 | * | |
a335b2e1 | 2432 | * __do_fault is protected similarly. |
72ddc8f7 ML |
2433 | */ |
2434 | if (!page_mkwrite) { | |
2435 | wait_on_page_locked(dirty_page); | |
2436 | set_page_dirty_balance(dirty_page, page_mkwrite); | |
2437 | } | |
2438 | put_page(dirty_page); | |
2439 | if (page_mkwrite) { | |
2440 | struct address_space *mapping = dirty_page->mapping; | |
2441 | ||
2442 | set_page_dirty(dirty_page); | |
2443 | unlock_page(dirty_page); | |
2444 | page_cache_release(dirty_page); | |
2445 | if (mapping) { | |
2446 | /* | |
2447 | * Some device drivers do not set page.mapping | |
2448 | * but still dirty their pages | |
2449 | */ | |
2450 | balance_dirty_pages_ratelimited(mapping); | |
2451 | } | |
2452 | } | |
2453 | ||
2454 | /* file_update_time outside page_lock */ | |
2455 | if (vma->vm_file) | |
2456 | file_update_time(vma->vm_file); | |
2457 | ||
2458 | return ret; | |
1da177e4 | 2459 | } |
1da177e4 LT |
2460 | |
2461 | /* | |
2462 | * Ok, we need to copy. Oh, well.. | |
2463 | */ | |
b5810039 | 2464 | page_cache_get(old_page); |
920fc356 | 2465 | gotten: |
8f4e2101 | 2466 | pte_unmap_unlock(page_table, ptl); |
1da177e4 LT |
2467 | |
2468 | if (unlikely(anon_vma_prepare(vma))) | |
65500d23 | 2469 | goto oom; |
a13ea5b7 | 2470 | |
62eede62 | 2471 | if (is_zero_pfn(pte_pfn(orig_pte))) { |
a13ea5b7 HD |
2472 | new_page = alloc_zeroed_user_highpage_movable(vma, address); |
2473 | if (!new_page) | |
2474 | goto oom; | |
2475 | } else { | |
2476 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); | |
2477 | if (!new_page) | |
2478 | goto oom; | |
2479 | cow_user_page(new_page, old_page, address, vma); | |
2480 | } | |
2481 | __SetPageUptodate(new_page); | |
2482 | ||
2c26fdd7 | 2483 | if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL)) |
8a9f3ccd BS |
2484 | goto oom_free_new; |
2485 | ||
1da177e4 LT |
2486 | /* |
2487 | * Re-check the pte - we dropped the lock | |
2488 | */ | |
8f4e2101 | 2489 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
65500d23 | 2490 | if (likely(pte_same(*page_table, orig_pte))) { |
920fc356 | 2491 | if (old_page) { |
920fc356 | 2492 | if (!PageAnon(old_page)) { |
34e55232 KH |
2493 | dec_mm_counter_fast(mm, MM_FILEPAGES); |
2494 | inc_mm_counter_fast(mm, MM_ANONPAGES); | |
920fc356 HD |
2495 | } |
2496 | } else | |
34e55232 | 2497 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
eca35133 | 2498 | flush_cache_page(vma, address, pte_pfn(orig_pte)); |
65500d23 HD |
2499 | entry = mk_pte(new_page, vma->vm_page_prot); |
2500 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
4ce072f1 SS |
2501 | /* |
2502 | * Clear the pte entry and flush it first, before updating the | |
2503 | * pte with the new entry. This will avoid a race condition | |
2504 | * seen in the presence of one thread doing SMC and another | |
2505 | * thread doing COW. | |
2506 | */ | |
828502d3 | 2507 | ptep_clear_flush(vma, address, page_table); |
9617d95e | 2508 | page_add_new_anon_rmap(new_page, vma, address); |
828502d3 IE |
2509 | /* |
2510 | * We call the notify macro here because, when using secondary | |
2511 | * mmu page tables (such as kvm shadow page tables), we want the | |
2512 | * new page to be mapped directly into the secondary page table. | |
2513 | */ | |
2514 | set_pte_at_notify(mm, address, page_table, entry); | |
4b3073e1 | 2515 | update_mmu_cache(vma, address, page_table); |
945754a1 NP |
2516 | if (old_page) { |
2517 | /* | |
2518 | * Only after switching the pte to the new page may | |
2519 | * we remove the mapcount here. Otherwise another | |
2520 | * process may come and find the rmap count decremented | |
2521 | * before the pte is switched to the new page, and | |
2522 | * "reuse" the old page writing into it while our pte | |
2523 | * here still points into it and can be read by other | |
2524 | * threads. | |
2525 | * | |
2526 | * The critical issue is to order this | |
2527 | * page_remove_rmap with the ptp_clear_flush above. | |
2528 | * Those stores are ordered by (if nothing else,) | |
2529 | * the barrier present in the atomic_add_negative | |
2530 | * in page_remove_rmap. | |
2531 | * | |
2532 | * Then the TLB flush in ptep_clear_flush ensures that | |
2533 | * no process can access the old page before the | |
2534 | * decremented mapcount is visible. And the old page | |
2535 | * cannot be reused until after the decremented | |
2536 | * mapcount is visible. So transitively, TLBs to | |
2537 | * old page will be flushed before it can be reused. | |
2538 | */ | |
edc315fd | 2539 | page_remove_rmap(old_page); |
945754a1 NP |
2540 | } |
2541 | ||
1da177e4 LT |
2542 | /* Free the old page.. */ |
2543 | new_page = old_page; | |
f33ea7f4 | 2544 | ret |= VM_FAULT_WRITE; |
8a9f3ccd BS |
2545 | } else |
2546 | mem_cgroup_uncharge_page(new_page); | |
2547 | ||
920fc356 HD |
2548 | if (new_page) |
2549 | page_cache_release(new_page); | |
65500d23 | 2550 | unlock: |
8f4e2101 | 2551 | pte_unmap_unlock(page_table, ptl); |
e15f8c01 ML |
2552 | if (old_page) { |
2553 | /* | |
2554 | * Don't let another task, with possibly unlocked vma, | |
2555 | * keep the mlocked page. | |
2556 | */ | |
2557 | if ((ret & VM_FAULT_WRITE) && (vma->vm_flags & VM_LOCKED)) { | |
2558 | lock_page(old_page); /* LRU manipulation */ | |
2559 | munlock_vma_page(old_page); | |
2560 | unlock_page(old_page); | |
2561 | } | |
2562 | page_cache_release(old_page); | |
2563 | } | |
f33ea7f4 | 2564 | return ret; |
8a9f3ccd | 2565 | oom_free_new: |
6dbf6d3b | 2566 | page_cache_release(new_page); |
65500d23 | 2567 | oom: |
b827e496 NP |
2568 | if (old_page) { |
2569 | if (page_mkwrite) { | |
2570 | unlock_page(old_page); | |
2571 | page_cache_release(old_page); | |
2572 | } | |
920fc356 | 2573 | page_cache_release(old_page); |
b827e496 | 2574 | } |
1da177e4 | 2575 | return VM_FAULT_OOM; |
9637a5ef DH |
2576 | |
2577 | unwritable_page: | |
2578 | page_cache_release(old_page); | |
c2ec175c | 2579 | return ret; |
1da177e4 LT |
2580 | } |
2581 | ||
97a89413 | 2582 | static void unmap_mapping_range_vma(struct vm_area_struct *vma, |
1da177e4 LT |
2583 | unsigned long start_addr, unsigned long end_addr, |
2584 | struct zap_details *details) | |
2585 | { | |
97a89413 | 2586 | zap_page_range(vma, start_addr, end_addr - start_addr, details); |
1da177e4 LT |
2587 | } |
2588 | ||
2589 | static inline void unmap_mapping_range_tree(struct prio_tree_root *root, | |
2590 | struct zap_details *details) | |
2591 | { | |
2592 | struct vm_area_struct *vma; | |
2593 | struct prio_tree_iter iter; | |
2594 | pgoff_t vba, vea, zba, zea; | |
2595 | ||
1da177e4 LT |
2596 | vma_prio_tree_foreach(vma, &iter, root, |
2597 | details->first_index, details->last_index) { | |
1da177e4 LT |
2598 | |
2599 | vba = vma->vm_pgoff; | |
2600 | vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1; | |
2601 | /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */ | |
2602 | zba = details->first_index; | |
2603 | if (zba < vba) | |
2604 | zba = vba; | |
2605 | zea = details->last_index; | |
2606 | if (zea > vea) | |
2607 | zea = vea; | |
2608 | ||
97a89413 | 2609 | unmap_mapping_range_vma(vma, |
1da177e4 LT |
2610 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, |
2611 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, | |
97a89413 | 2612 | details); |
1da177e4 LT |
2613 | } |
2614 | } | |
2615 | ||
2616 | static inline void unmap_mapping_range_list(struct list_head *head, | |
2617 | struct zap_details *details) | |
2618 | { | |
2619 | struct vm_area_struct *vma; | |
2620 | ||
2621 | /* | |
2622 | * In nonlinear VMAs there is no correspondence between virtual address | |
2623 | * offset and file offset. So we must perform an exhaustive search | |
2624 | * across *all* the pages in each nonlinear VMA, not just the pages | |
2625 | * whose virtual address lies outside the file truncation point. | |
2626 | */ | |
1da177e4 | 2627 | list_for_each_entry(vma, head, shared.vm_set.list) { |
1da177e4 | 2628 | details->nonlinear_vma = vma; |
97a89413 | 2629 | unmap_mapping_range_vma(vma, vma->vm_start, vma->vm_end, details); |
1da177e4 LT |
2630 | } |
2631 | } | |
2632 | ||
2633 | /** | |
72fd4a35 | 2634 | * 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 | 2635 | * @mapping: the address space containing mmaps to be unmapped. |
1da177e4 LT |
2636 | * @holebegin: byte in first page to unmap, relative to the start of |
2637 | * the underlying file. This will be rounded down to a PAGE_SIZE | |
25d9e2d1 | 2638 | * boundary. Note that this is different from truncate_pagecache(), which |
1da177e4 LT |
2639 | * must keep the partial page. In contrast, we must get rid of |
2640 | * partial pages. | |
2641 | * @holelen: size of prospective hole in bytes. This will be rounded | |
2642 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the | |
2643 | * end of the file. | |
2644 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; | |
2645 | * but 0 when invalidating pagecache, don't throw away private data. | |
2646 | */ | |
2647 | void unmap_mapping_range(struct address_space *mapping, | |
2648 | loff_t const holebegin, loff_t const holelen, int even_cows) | |
2649 | { | |
2650 | struct zap_details details; | |
2651 | pgoff_t hba = holebegin >> PAGE_SHIFT; | |
2652 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2653 | ||
2654 | /* Check for overflow. */ | |
2655 | if (sizeof(holelen) > sizeof(hlen)) { | |
2656 | long long holeend = | |
2657 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2658 | if (holeend & ~(long long)ULONG_MAX) | |
2659 | hlen = ULONG_MAX - hba + 1; | |
2660 | } | |
2661 | ||
2662 | details.check_mapping = even_cows? NULL: mapping; | |
2663 | details.nonlinear_vma = NULL; | |
2664 | details.first_index = hba; | |
2665 | details.last_index = hba + hlen - 1; | |
2666 | if (details.last_index < details.first_index) | |
2667 | details.last_index = ULONG_MAX; | |
1da177e4 | 2668 | |
1da177e4 | 2669 | |
3d48ae45 | 2670 | mutex_lock(&mapping->i_mmap_mutex); |
1da177e4 LT |
2671 | if (unlikely(!prio_tree_empty(&mapping->i_mmap))) |
2672 | unmap_mapping_range_tree(&mapping->i_mmap, &details); | |
2673 | if (unlikely(!list_empty(&mapping->i_mmap_nonlinear))) | |
2674 | unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details); | |
3d48ae45 | 2675 | mutex_unlock(&mapping->i_mmap_mutex); |
1da177e4 LT |
2676 | } |
2677 | EXPORT_SYMBOL(unmap_mapping_range); | |
2678 | ||
f6b3ec23 BP |
2679 | int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end) |
2680 | { | |
2681 | struct address_space *mapping = inode->i_mapping; | |
2682 | ||
2683 | /* | |
2684 | * If the underlying filesystem is not going to provide | |
2685 | * a way to truncate a range of blocks (punch a hole) - | |
2686 | * we should return failure right now. | |
2687 | */ | |
acfa4380 | 2688 | if (!inode->i_op->truncate_range) |
f6b3ec23 BP |
2689 | return -ENOSYS; |
2690 | ||
1b1dcc1b | 2691 | mutex_lock(&inode->i_mutex); |
f6b3ec23 BP |
2692 | down_write(&inode->i_alloc_sem); |
2693 | unmap_mapping_range(mapping, offset, (end - offset), 1); | |
2694 | truncate_inode_pages_range(mapping, offset, end); | |
d00806b1 | 2695 | unmap_mapping_range(mapping, offset, (end - offset), 1); |
f6b3ec23 BP |
2696 | inode->i_op->truncate_range(inode, offset, end); |
2697 | up_write(&inode->i_alloc_sem); | |
1b1dcc1b | 2698 | mutex_unlock(&inode->i_mutex); |
f6b3ec23 BP |
2699 | |
2700 | return 0; | |
2701 | } | |
f6b3ec23 | 2702 | |
1da177e4 | 2703 | /* |
8f4e2101 HD |
2704 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2705 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2706 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2707 | */ |
65500d23 HD |
2708 | static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2709 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
30c9f3a9 | 2710 | unsigned int flags, pte_t orig_pte) |
1da177e4 | 2711 | { |
8f4e2101 | 2712 | spinlock_t *ptl; |
4969c119 | 2713 | struct page *page, *swapcache = NULL; |
65500d23 | 2714 | swp_entry_t entry; |
1da177e4 | 2715 | pte_t pte; |
d065bd81 | 2716 | int locked; |
56039efa | 2717 | struct mem_cgroup *ptr; |
ad8c2ee8 | 2718 | int exclusive = 0; |
83c54070 | 2719 | int ret = 0; |
1da177e4 | 2720 | |
4c21e2f2 | 2721 | if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) |
8f4e2101 | 2722 | goto out; |
65500d23 HD |
2723 | |
2724 | entry = pte_to_swp_entry(orig_pte); | |
d1737fdb AK |
2725 | if (unlikely(non_swap_entry(entry))) { |
2726 | if (is_migration_entry(entry)) { | |
2727 | migration_entry_wait(mm, pmd, address); | |
2728 | } else if (is_hwpoison_entry(entry)) { | |
2729 | ret = VM_FAULT_HWPOISON; | |
2730 | } else { | |
2731 | print_bad_pte(vma, address, orig_pte, NULL); | |
d99be1a8 | 2732 | ret = VM_FAULT_SIGBUS; |
d1737fdb | 2733 | } |
0697212a CL |
2734 | goto out; |
2735 | } | |
0ff92245 | 2736 | delayacct_set_flag(DELAYACCT_PF_SWAPIN); |
1da177e4 LT |
2737 | page = lookup_swap_cache(entry); |
2738 | if (!page) { | |
a5c9b696 | 2739 | grab_swap_token(mm); /* Contend for token _before_ read-in */ |
02098fea HD |
2740 | page = swapin_readahead(entry, |
2741 | GFP_HIGHUSER_MOVABLE, vma, address); | |
1da177e4 LT |
2742 | if (!page) { |
2743 | /* | |
8f4e2101 HD |
2744 | * Back out if somebody else faulted in this pte |
2745 | * while we released the pte lock. | |
1da177e4 | 2746 | */ |
8f4e2101 | 2747 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
1da177e4 LT |
2748 | if (likely(pte_same(*page_table, orig_pte))) |
2749 | ret = VM_FAULT_OOM; | |
0ff92245 | 2750 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
65500d23 | 2751 | goto unlock; |
1da177e4 LT |
2752 | } |
2753 | ||
2754 | /* Had to read the page from swap area: Major fault */ | |
2755 | ret = VM_FAULT_MAJOR; | |
f8891e5e | 2756 | count_vm_event(PGMAJFAULT); |
d1737fdb | 2757 | } else if (PageHWPoison(page)) { |
71f72525 WF |
2758 | /* |
2759 | * hwpoisoned dirty swapcache pages are kept for killing | |
2760 | * owner processes (which may be unknown at hwpoison time) | |
2761 | */ | |
d1737fdb AK |
2762 | ret = VM_FAULT_HWPOISON; |
2763 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); | |
4779cb31 | 2764 | goto out_release; |
1da177e4 LT |
2765 | } |
2766 | ||
d065bd81 | 2767 | locked = lock_page_or_retry(page, mm, flags); |
073e587e | 2768 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
d065bd81 ML |
2769 | if (!locked) { |
2770 | ret |= VM_FAULT_RETRY; | |
2771 | goto out_release; | |
2772 | } | |
073e587e | 2773 | |
4969c119 | 2774 | /* |
31c4a3d3 HD |
2775 | * Make sure try_to_free_swap or reuse_swap_page or swapoff did not |
2776 | * release the swapcache from under us. The page pin, and pte_same | |
2777 | * test below, are not enough to exclude that. Even if it is still | |
2778 | * swapcache, we need to check that the page's swap has not changed. | |
4969c119 | 2779 | */ |
31c4a3d3 | 2780 | if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val)) |
4969c119 AA |
2781 | goto out_page; |
2782 | ||
2783 | if (ksm_might_need_to_copy(page, vma, address)) { | |
2784 | swapcache = page; | |
2785 | page = ksm_does_need_to_copy(page, vma, address); | |
2786 | ||
2787 | if (unlikely(!page)) { | |
2788 | ret = VM_FAULT_OOM; | |
2789 | page = swapcache; | |
2790 | swapcache = NULL; | |
2791 | goto out_page; | |
2792 | } | |
5ad64688 HD |
2793 | } |
2794 | ||
2c26fdd7 | 2795 | if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) { |
8a9f3ccd | 2796 | ret = VM_FAULT_OOM; |
bc43f75c | 2797 | goto out_page; |
8a9f3ccd BS |
2798 | } |
2799 | ||
1da177e4 | 2800 | /* |
8f4e2101 | 2801 | * Back out if somebody else already faulted in this pte. |
1da177e4 | 2802 | */ |
8f4e2101 | 2803 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
9e9bef07 | 2804 | if (unlikely(!pte_same(*page_table, orig_pte))) |
b8107480 | 2805 | goto out_nomap; |
b8107480 KK |
2806 | |
2807 | if (unlikely(!PageUptodate(page))) { | |
2808 | ret = VM_FAULT_SIGBUS; | |
2809 | goto out_nomap; | |
1da177e4 LT |
2810 | } |
2811 | ||
8c7c6e34 KH |
2812 | /* |
2813 | * The page isn't present yet, go ahead with the fault. | |
2814 | * | |
2815 | * Be careful about the sequence of operations here. | |
2816 | * To get its accounting right, reuse_swap_page() must be called | |
2817 | * while the page is counted on swap but not yet in mapcount i.e. | |
2818 | * before page_add_anon_rmap() and swap_free(); try_to_free_swap() | |
2819 | * must be called after the swap_free(), or it will never succeed. | |
03f3c433 KH |
2820 | * Because delete_from_swap_page() may be called by reuse_swap_page(), |
2821 | * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry | |
2822 | * in page->private. In this case, a record in swap_cgroup is silently | |
2823 | * discarded at swap_free(). | |
8c7c6e34 | 2824 | */ |
1da177e4 | 2825 | |
34e55232 | 2826 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
b084d435 | 2827 | dec_mm_counter_fast(mm, MM_SWAPENTS); |
1da177e4 | 2828 | pte = mk_pte(page, vma->vm_page_prot); |
30c9f3a9 | 2829 | if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) { |
1da177e4 | 2830 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); |
30c9f3a9 | 2831 | flags &= ~FAULT_FLAG_WRITE; |
9a5b489b | 2832 | ret |= VM_FAULT_WRITE; |
ad8c2ee8 | 2833 | exclusive = 1; |
1da177e4 | 2834 | } |
1da177e4 LT |
2835 | flush_icache_page(vma, page); |
2836 | set_pte_at(mm, address, page_table, pte); | |
ad8c2ee8 | 2837 | do_page_add_anon_rmap(page, vma, address, exclusive); |
03f3c433 KH |
2838 | /* It's better to call commit-charge after rmap is established */ |
2839 | mem_cgroup_commit_charge_swapin(page, ptr); | |
1da177e4 | 2840 | |
c475a8ab | 2841 | swap_free(entry); |
b291f000 | 2842 | if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) |
a2c43eed | 2843 | try_to_free_swap(page); |
c475a8ab | 2844 | unlock_page(page); |
4969c119 AA |
2845 | if (swapcache) { |
2846 | /* | |
2847 | * Hold the lock to avoid the swap entry to be reused | |
2848 | * until we take the PT lock for the pte_same() check | |
2849 | * (to avoid false positives from pte_same). For | |
2850 | * further safety release the lock after the swap_free | |
2851 | * so that the swap count won't change under a | |
2852 | * parallel locked swapcache. | |
2853 | */ | |
2854 | unlock_page(swapcache); | |
2855 | page_cache_release(swapcache); | |
2856 | } | |
c475a8ab | 2857 | |
30c9f3a9 | 2858 | if (flags & FAULT_FLAG_WRITE) { |
61469f1d HD |
2859 | ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte); |
2860 | if (ret & VM_FAULT_ERROR) | |
2861 | ret &= VM_FAULT_ERROR; | |
1da177e4 LT |
2862 | goto out; |
2863 | } | |
2864 | ||
2865 | /* No need to invalidate - it was non-present before */ | |
4b3073e1 | 2866 | update_mmu_cache(vma, address, page_table); |
65500d23 | 2867 | unlock: |
8f4e2101 | 2868 | pte_unmap_unlock(page_table, ptl); |
1da177e4 LT |
2869 | out: |
2870 | return ret; | |
b8107480 | 2871 | out_nomap: |
7a81b88c | 2872 | mem_cgroup_cancel_charge_swapin(ptr); |
8f4e2101 | 2873 | pte_unmap_unlock(page_table, ptl); |
bc43f75c | 2874 | out_page: |
b8107480 | 2875 | unlock_page(page); |
4779cb31 | 2876 | out_release: |
b8107480 | 2877 | page_cache_release(page); |
4969c119 AA |
2878 | if (swapcache) { |
2879 | unlock_page(swapcache); | |
2880 | page_cache_release(swapcache); | |
2881 | } | |
65500d23 | 2882 | return ret; |
1da177e4 LT |
2883 | } |
2884 | ||
320b2b8d | 2885 | /* |
8ca3eb08 TL |
2886 | * This is like a special single-page "expand_{down|up}wards()", |
2887 | * except we must first make sure that 'address{-|+}PAGE_SIZE' | |
320b2b8d | 2888 | * doesn't hit another vma. |
320b2b8d LT |
2889 | */ |
2890 | static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address) | |
2891 | { | |
2892 | address &= PAGE_MASK; | |
2893 | if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) { | |
0e8e50e2 LT |
2894 | struct vm_area_struct *prev = vma->vm_prev; |
2895 | ||
2896 | /* | |
2897 | * Is there a mapping abutting this one below? | |
2898 | * | |
2899 | * That's only ok if it's the same stack mapping | |
2900 | * that has gotten split.. | |
2901 | */ | |
2902 | if (prev && prev->vm_end == address) | |
2903 | return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM; | |
320b2b8d | 2904 | |
d05f3169 | 2905 | expand_downwards(vma, address - PAGE_SIZE); |
320b2b8d | 2906 | } |
8ca3eb08 TL |
2907 | if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) { |
2908 | struct vm_area_struct *next = vma->vm_next; | |
2909 | ||
2910 | /* As VM_GROWSDOWN but s/below/above/ */ | |
2911 | if (next && next->vm_start == address + PAGE_SIZE) | |
2912 | return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM; | |
2913 | ||
2914 | expand_upwards(vma, address + PAGE_SIZE); | |
2915 | } | |
320b2b8d LT |
2916 | return 0; |
2917 | } | |
2918 | ||
1da177e4 | 2919 | /* |
8f4e2101 HD |
2920 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2921 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2922 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2923 | */ |
65500d23 HD |
2924 | static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2925 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
30c9f3a9 | 2926 | unsigned int flags) |
1da177e4 | 2927 | { |
8f4e2101 HD |
2928 | struct page *page; |
2929 | spinlock_t *ptl; | |
1da177e4 | 2930 | pte_t entry; |
1da177e4 | 2931 | |
11ac5524 LT |
2932 | pte_unmap(page_table); |
2933 | ||
2934 | /* Check if we need to add a guard page to the stack */ | |
2935 | if (check_stack_guard_page(vma, address) < 0) | |
320b2b8d LT |
2936 | return VM_FAULT_SIGBUS; |
2937 | ||
11ac5524 | 2938 | /* Use the zero-page for reads */ |
62eede62 HD |
2939 | if (!(flags & FAULT_FLAG_WRITE)) { |
2940 | entry = pte_mkspecial(pfn_pte(my_zero_pfn(address), | |
2941 | vma->vm_page_prot)); | |
11ac5524 | 2942 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
a13ea5b7 HD |
2943 | if (!pte_none(*page_table)) |
2944 | goto unlock; | |
2945 | goto setpte; | |
2946 | } | |
2947 | ||
557ed1fa | 2948 | /* Allocate our own private page. */ |
557ed1fa NP |
2949 | if (unlikely(anon_vma_prepare(vma))) |
2950 | goto oom; | |
2951 | page = alloc_zeroed_user_highpage_movable(vma, address); | |
2952 | if (!page) | |
2953 | goto oom; | |
0ed361de | 2954 | __SetPageUptodate(page); |
8f4e2101 | 2955 | |
2c26fdd7 | 2956 | if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) |
8a9f3ccd BS |
2957 | goto oom_free_page; |
2958 | ||
557ed1fa | 2959 | entry = mk_pte(page, vma->vm_page_prot); |
1ac0cb5d HD |
2960 | if (vma->vm_flags & VM_WRITE) |
2961 | entry = pte_mkwrite(pte_mkdirty(entry)); | |
1da177e4 | 2962 | |
557ed1fa | 2963 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
1c2fb7a4 | 2964 | if (!pte_none(*page_table)) |
557ed1fa | 2965 | goto release; |
9ba69294 | 2966 | |
34e55232 | 2967 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
557ed1fa | 2968 | page_add_new_anon_rmap(page, vma, address); |
a13ea5b7 | 2969 | setpte: |
65500d23 | 2970 | set_pte_at(mm, address, page_table, entry); |
1da177e4 LT |
2971 | |
2972 | /* No need to invalidate - it was non-present before */ | |
4b3073e1 | 2973 | update_mmu_cache(vma, address, page_table); |
65500d23 | 2974 | unlock: |
8f4e2101 | 2975 | pte_unmap_unlock(page_table, ptl); |
83c54070 | 2976 | return 0; |
8f4e2101 | 2977 | release: |
8a9f3ccd | 2978 | mem_cgroup_uncharge_page(page); |
8f4e2101 HD |
2979 | page_cache_release(page); |
2980 | goto unlock; | |
8a9f3ccd | 2981 | oom_free_page: |
6dbf6d3b | 2982 | page_cache_release(page); |
65500d23 | 2983 | oom: |
1da177e4 LT |
2984 | return VM_FAULT_OOM; |
2985 | } | |
2986 | ||
2987 | /* | |
54cb8821 | 2988 | * __do_fault() tries to create a new page mapping. It aggressively |
1da177e4 | 2989 | * tries to share with existing pages, but makes a separate copy if |
54cb8821 NP |
2990 | * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid |
2991 | * the next page fault. | |
1da177e4 LT |
2992 | * |
2993 | * As this is called only for pages that do not currently exist, we | |
2994 | * do not need to flush old virtual caches or the TLB. | |
2995 | * | |
8f4e2101 | 2996 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
16abfa08 | 2997 | * but allow concurrent faults), and pte neither mapped nor locked. |
8f4e2101 | 2998 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
1da177e4 | 2999 | */ |
54cb8821 | 3000 | static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
16abfa08 | 3001 | unsigned long address, pmd_t *pmd, |
54cb8821 | 3002 | pgoff_t pgoff, unsigned int flags, pte_t orig_pte) |
1da177e4 | 3003 | { |
16abfa08 | 3004 | pte_t *page_table; |
8f4e2101 | 3005 | spinlock_t *ptl; |
d0217ac0 | 3006 | struct page *page; |
1da177e4 | 3007 | pte_t entry; |
1da177e4 | 3008 | int anon = 0; |
5b4e655e | 3009 | int charged = 0; |
d08b3851 | 3010 | struct page *dirty_page = NULL; |
d0217ac0 NP |
3011 | struct vm_fault vmf; |
3012 | int ret; | |
a200ee18 | 3013 | int page_mkwrite = 0; |
54cb8821 | 3014 | |
d0217ac0 NP |
3015 | vmf.virtual_address = (void __user *)(address & PAGE_MASK); |
3016 | vmf.pgoff = pgoff; | |
3017 | vmf.flags = flags; | |
3018 | vmf.page = NULL; | |
1da177e4 | 3019 | |
3c18ddd1 | 3020 | ret = vma->vm_ops->fault(vma, &vmf); |
d065bd81 ML |
3021 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | |
3022 | VM_FAULT_RETRY))) | |
3c18ddd1 | 3023 | return ret; |
1da177e4 | 3024 | |
a3b947ea AK |
3025 | if (unlikely(PageHWPoison(vmf.page))) { |
3026 | if (ret & VM_FAULT_LOCKED) | |
3027 | unlock_page(vmf.page); | |
3028 | return VM_FAULT_HWPOISON; | |
3029 | } | |
3030 | ||
d00806b1 | 3031 | /* |
d0217ac0 | 3032 | * For consistency in subsequent calls, make the faulted page always |
d00806b1 NP |
3033 | * locked. |
3034 | */ | |
83c54070 | 3035 | if (unlikely(!(ret & VM_FAULT_LOCKED))) |
d0217ac0 | 3036 | lock_page(vmf.page); |
54cb8821 | 3037 | else |
d0217ac0 | 3038 | VM_BUG_ON(!PageLocked(vmf.page)); |
d00806b1 | 3039 | |
1da177e4 LT |
3040 | /* |
3041 | * Should we do an early C-O-W break? | |
3042 | */ | |
d0217ac0 | 3043 | page = vmf.page; |
54cb8821 | 3044 | if (flags & FAULT_FLAG_WRITE) { |
9637a5ef | 3045 | if (!(vma->vm_flags & VM_SHARED)) { |
54cb8821 | 3046 | anon = 1; |
d00806b1 | 3047 | if (unlikely(anon_vma_prepare(vma))) { |
d0217ac0 | 3048 | ret = VM_FAULT_OOM; |
54cb8821 | 3049 | goto out; |
d00806b1 | 3050 | } |
83c54070 NP |
3051 | page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, |
3052 | vma, address); | |
d00806b1 | 3053 | if (!page) { |
d0217ac0 | 3054 | ret = VM_FAULT_OOM; |
54cb8821 | 3055 | goto out; |
d00806b1 | 3056 | } |
2c26fdd7 | 3057 | if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) { |
5b4e655e KH |
3058 | ret = VM_FAULT_OOM; |
3059 | page_cache_release(page); | |
3060 | goto out; | |
3061 | } | |
3062 | charged = 1; | |
d0217ac0 | 3063 | copy_user_highpage(page, vmf.page, address, vma); |
0ed361de | 3064 | __SetPageUptodate(page); |
9637a5ef | 3065 | } else { |
54cb8821 NP |
3066 | /* |
3067 | * If the page will be shareable, see if the backing | |
9637a5ef | 3068 | * address space wants to know that the page is about |
54cb8821 NP |
3069 | * to become writable |
3070 | */ | |
69676147 | 3071 | if (vma->vm_ops->page_mkwrite) { |
c2ec175c NP |
3072 | int tmp; |
3073 | ||
69676147 | 3074 | unlock_page(page); |
b827e496 | 3075 | vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; |
c2ec175c NP |
3076 | tmp = vma->vm_ops->page_mkwrite(vma, &vmf); |
3077 | if (unlikely(tmp & | |
3078 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { | |
3079 | ret = tmp; | |
b827e496 | 3080 | goto unwritable_page; |
d0217ac0 | 3081 | } |
b827e496 NP |
3082 | if (unlikely(!(tmp & VM_FAULT_LOCKED))) { |
3083 | lock_page(page); | |
3084 | if (!page->mapping) { | |
3085 | ret = 0; /* retry the fault */ | |
3086 | unlock_page(page); | |
3087 | goto unwritable_page; | |
3088 | } | |
3089 | } else | |
3090 | VM_BUG_ON(!PageLocked(page)); | |
a200ee18 | 3091 | page_mkwrite = 1; |
9637a5ef DH |
3092 | } |
3093 | } | |
54cb8821 | 3094 | |
1da177e4 LT |
3095 | } |
3096 | ||
8f4e2101 | 3097 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
1da177e4 LT |
3098 | |
3099 | /* | |
3100 | * This silly early PAGE_DIRTY setting removes a race | |
3101 | * due to the bad i386 page protection. But it's valid | |
3102 | * for other architectures too. | |
3103 | * | |
30c9f3a9 | 3104 | * Note that if FAULT_FLAG_WRITE is set, we either now have |
1da177e4 LT |
3105 | * an exclusive copy of the page, or this is a shared mapping, |
3106 | * so we can make it writable and dirty to avoid having to | |
3107 | * handle that later. | |
3108 | */ | |
3109 | /* Only go through if we didn't race with anybody else... */ | |
1c2fb7a4 | 3110 | if (likely(pte_same(*page_table, orig_pte))) { |
d00806b1 NP |
3111 | flush_icache_page(vma, page); |
3112 | entry = mk_pte(page, vma->vm_page_prot); | |
54cb8821 | 3113 | if (flags & FAULT_FLAG_WRITE) |
1da177e4 | 3114 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
1da177e4 | 3115 | if (anon) { |
34e55232 | 3116 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
64d6519d | 3117 | page_add_new_anon_rmap(page, vma, address); |
f57e88a8 | 3118 | } else { |
34e55232 | 3119 | inc_mm_counter_fast(mm, MM_FILEPAGES); |
d00806b1 | 3120 | page_add_file_rmap(page); |
54cb8821 | 3121 | if (flags & FAULT_FLAG_WRITE) { |
d00806b1 | 3122 | dirty_page = page; |
d08b3851 PZ |
3123 | get_page(dirty_page); |
3124 | } | |
4294621f | 3125 | } |
64d6519d | 3126 | set_pte_at(mm, address, page_table, entry); |
d00806b1 NP |
3127 | |
3128 | /* no need to invalidate: a not-present page won't be cached */ | |
4b3073e1 | 3129 | update_mmu_cache(vma, address, page_table); |
1da177e4 | 3130 | } else { |
5b4e655e KH |
3131 | if (charged) |
3132 | mem_cgroup_uncharge_page(page); | |
d00806b1 NP |
3133 | if (anon) |
3134 | page_cache_release(page); | |
3135 | else | |
54cb8821 | 3136 | anon = 1; /* no anon but release faulted_page */ |
1da177e4 LT |
3137 | } |
3138 | ||
8f4e2101 | 3139 | pte_unmap_unlock(page_table, ptl); |
d00806b1 NP |
3140 | |
3141 | out: | |
b827e496 NP |
3142 | if (dirty_page) { |
3143 | struct address_space *mapping = page->mapping; | |
8f7b3d15 | 3144 | |
b827e496 NP |
3145 | if (set_page_dirty(dirty_page)) |
3146 | page_mkwrite = 1; | |
3147 | unlock_page(dirty_page); | |
d08b3851 | 3148 | put_page(dirty_page); |
b827e496 NP |
3149 | if (page_mkwrite && mapping) { |
3150 | /* | |
3151 | * Some device drivers do not set page.mapping but still | |
3152 | * dirty their pages | |
3153 | */ | |
3154 | balance_dirty_pages_ratelimited(mapping); | |
3155 | } | |
3156 | ||
3157 | /* file_update_time outside page_lock */ | |
3158 | if (vma->vm_file) | |
3159 | file_update_time(vma->vm_file); | |
3160 | } else { | |
3161 | unlock_page(vmf.page); | |
3162 | if (anon) | |
3163 | page_cache_release(vmf.page); | |
d08b3851 | 3164 | } |
d00806b1 | 3165 | |
83c54070 | 3166 | return ret; |
b827e496 NP |
3167 | |
3168 | unwritable_page: | |
3169 | page_cache_release(page); | |
3170 | return ret; | |
54cb8821 | 3171 | } |
d00806b1 | 3172 | |
54cb8821 NP |
3173 | static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
3174 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
30c9f3a9 | 3175 | unsigned int flags, pte_t orig_pte) |
54cb8821 NP |
3176 | { |
3177 | pgoff_t pgoff = (((address & PAGE_MASK) | |
0da7e01f | 3178 | - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
54cb8821 | 3179 | |
16abfa08 HD |
3180 | pte_unmap(page_table); |
3181 | return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); | |
54cb8821 NP |
3182 | } |
3183 | ||
1da177e4 LT |
3184 | /* |
3185 | * Fault of a previously existing named mapping. Repopulate the pte | |
3186 | * from the encoded file_pte if possible. This enables swappable | |
3187 | * nonlinear vmas. | |
8f4e2101 HD |
3188 | * |
3189 | * We enter with non-exclusive mmap_sem (to exclude vma changes, | |
3190 | * but allow concurrent faults), and pte mapped but not yet locked. | |
3191 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 3192 | */ |
d0217ac0 | 3193 | static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
65500d23 | 3194 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
30c9f3a9 | 3195 | unsigned int flags, pte_t orig_pte) |
1da177e4 | 3196 | { |
65500d23 | 3197 | pgoff_t pgoff; |
1da177e4 | 3198 | |
30c9f3a9 LT |
3199 | flags |= FAULT_FLAG_NONLINEAR; |
3200 | ||
4c21e2f2 | 3201 | if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) |
83c54070 | 3202 | return 0; |
1da177e4 | 3203 | |
2509ef26 | 3204 | if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) { |
65500d23 HD |
3205 | /* |
3206 | * Page table corrupted: show pte and kill process. | |
3207 | */ | |
3dc14741 | 3208 | print_bad_pte(vma, address, orig_pte, NULL); |
d99be1a8 | 3209 | return VM_FAULT_SIGBUS; |
65500d23 | 3210 | } |
65500d23 HD |
3211 | |
3212 | pgoff = pte_to_pgoff(orig_pte); | |
16abfa08 | 3213 | return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); |
1da177e4 LT |
3214 | } |
3215 | ||
3216 | /* | |
3217 | * These routines also need to handle stuff like marking pages dirty | |
3218 | * and/or accessed for architectures that don't do it in hardware (most | |
3219 | * RISC architectures). The early dirtying is also good on the i386. | |
3220 | * | |
3221 | * There is also a hook called "update_mmu_cache()" that architectures | |
3222 | * with external mmu caches can use to update those (ie the Sparc or | |
3223 | * PowerPC hashed page tables that act as extended TLBs). | |
3224 | * | |
c74df32c HD |
3225 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
3226 | * but allow concurrent faults), and pte mapped but not yet locked. | |
3227 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 3228 | */ |
71e3aac0 AA |
3229 | int handle_pte_fault(struct mm_struct *mm, |
3230 | struct vm_area_struct *vma, unsigned long address, | |
3231 | pte_t *pte, pmd_t *pmd, unsigned int flags) | |
1da177e4 LT |
3232 | { |
3233 | pte_t entry; | |
8f4e2101 | 3234 | spinlock_t *ptl; |
1da177e4 | 3235 | |
8dab5241 | 3236 | entry = *pte; |
1da177e4 | 3237 | if (!pte_present(entry)) { |
65500d23 | 3238 | if (pte_none(entry)) { |
f4b81804 | 3239 | if (vma->vm_ops) { |
3c18ddd1 | 3240 | if (likely(vma->vm_ops->fault)) |
54cb8821 | 3241 | return do_linear_fault(mm, vma, address, |
30c9f3a9 | 3242 | pte, pmd, flags, entry); |
f4b81804 JS |
3243 | } |
3244 | return do_anonymous_page(mm, vma, address, | |
30c9f3a9 | 3245 | pte, pmd, flags); |
65500d23 | 3246 | } |
1da177e4 | 3247 | if (pte_file(entry)) |
d0217ac0 | 3248 | return do_nonlinear_fault(mm, vma, address, |
30c9f3a9 | 3249 | pte, pmd, flags, entry); |
65500d23 | 3250 | return do_swap_page(mm, vma, address, |
30c9f3a9 | 3251 | pte, pmd, flags, entry); |
1da177e4 LT |
3252 | } |
3253 | ||
4c21e2f2 | 3254 | ptl = pte_lockptr(mm, pmd); |
8f4e2101 HD |
3255 | spin_lock(ptl); |
3256 | if (unlikely(!pte_same(*pte, entry))) | |
3257 | goto unlock; | |
30c9f3a9 | 3258 | if (flags & FAULT_FLAG_WRITE) { |
1da177e4 | 3259 | if (!pte_write(entry)) |
8f4e2101 HD |
3260 | return do_wp_page(mm, vma, address, |
3261 | pte, pmd, ptl, entry); | |
1da177e4 LT |
3262 | entry = pte_mkdirty(entry); |
3263 | } | |
3264 | entry = pte_mkyoung(entry); | |
30c9f3a9 | 3265 | if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) { |
4b3073e1 | 3266 | update_mmu_cache(vma, address, pte); |
1a44e149 AA |
3267 | } else { |
3268 | /* | |
3269 | * This is needed only for protection faults but the arch code | |
3270 | * is not yet telling us if this is a protection fault or not. | |
3271 | * This still avoids useless tlb flushes for .text page faults | |
3272 | * with threads. | |
3273 | */ | |
30c9f3a9 | 3274 | if (flags & FAULT_FLAG_WRITE) |
61c77326 | 3275 | flush_tlb_fix_spurious_fault(vma, address); |
1a44e149 | 3276 | } |
8f4e2101 HD |
3277 | unlock: |
3278 | pte_unmap_unlock(pte, ptl); | |
83c54070 | 3279 | return 0; |
1da177e4 LT |
3280 | } |
3281 | ||
3282 | /* | |
3283 | * By the time we get here, we already hold the mm semaphore | |
3284 | */ | |
83c54070 | 3285 | int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
d06063cc | 3286 | unsigned long address, unsigned int flags) |
1da177e4 LT |
3287 | { |
3288 | pgd_t *pgd; | |
3289 | pud_t *pud; | |
3290 | pmd_t *pmd; | |
3291 | pte_t *pte; | |
3292 | ||
3293 | __set_current_state(TASK_RUNNING); | |
3294 | ||
f8891e5e | 3295 | count_vm_event(PGFAULT); |
1da177e4 | 3296 | |
34e55232 KH |
3297 | /* do counter updates before entering really critical section. */ |
3298 | check_sync_rss_stat(current); | |
3299 | ||
ac9b9c66 | 3300 | if (unlikely(is_vm_hugetlb_page(vma))) |
30c9f3a9 | 3301 | return hugetlb_fault(mm, vma, address, flags); |
1da177e4 | 3302 | |
1da177e4 | 3303 | pgd = pgd_offset(mm, address); |
1da177e4 LT |
3304 | pud = pud_alloc(mm, pgd, address); |
3305 | if (!pud) | |
c74df32c | 3306 | return VM_FAULT_OOM; |
1da177e4 LT |
3307 | pmd = pmd_alloc(mm, pud, address); |
3308 | if (!pmd) | |
c74df32c | 3309 | return VM_FAULT_OOM; |
71e3aac0 AA |
3310 | if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) { |
3311 | if (!vma->vm_ops) | |
3312 | return do_huge_pmd_anonymous_page(mm, vma, address, | |
3313 | pmd, flags); | |
3314 | } else { | |
3315 | pmd_t orig_pmd = *pmd; | |
3316 | barrier(); | |
3317 | if (pmd_trans_huge(orig_pmd)) { | |
3318 | if (flags & FAULT_FLAG_WRITE && | |
3319 | !pmd_write(orig_pmd) && | |
3320 | !pmd_trans_splitting(orig_pmd)) | |
3321 | return do_huge_pmd_wp_page(mm, vma, address, | |
3322 | pmd, orig_pmd); | |
3323 | return 0; | |
3324 | } | |
3325 | } | |
3326 | ||
3327 | /* | |
3328 | * Use __pte_alloc instead of pte_alloc_map, because we can't | |
3329 | * run pte_offset_map on the pmd, if an huge pmd could | |
3330 | * materialize from under us from a different thread. | |
3331 | */ | |
cc03638d | 3332 | if (unlikely(pmd_none(*pmd)) && __pte_alloc(mm, vma, pmd, address)) |
c74df32c | 3333 | return VM_FAULT_OOM; |
71e3aac0 AA |
3334 | /* if an huge pmd materialized from under us just retry later */ |
3335 | if (unlikely(pmd_trans_huge(*pmd))) | |
3336 | return 0; | |
3337 | /* | |
3338 | * A regular pmd is established and it can't morph into a huge pmd | |
3339 | * from under us anymore at this point because we hold the mmap_sem | |
3340 | * read mode and khugepaged takes it in write mode. So now it's | |
3341 | * safe to run pte_offset_map(). | |
3342 | */ | |
3343 | pte = pte_offset_map(pmd, address); | |
1da177e4 | 3344 | |
30c9f3a9 | 3345 | return handle_pte_fault(mm, vma, address, pte, pmd, flags); |
1da177e4 LT |
3346 | } |
3347 | ||
3348 | #ifndef __PAGETABLE_PUD_FOLDED | |
3349 | /* | |
3350 | * Allocate page upper directory. | |
872fec16 | 3351 | * We've already handled the fast-path in-line. |
1da177e4 | 3352 | */ |
1bb3630e | 3353 | int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) |
1da177e4 | 3354 | { |
c74df32c HD |
3355 | pud_t *new = pud_alloc_one(mm, address); |
3356 | if (!new) | |
1bb3630e | 3357 | return -ENOMEM; |
1da177e4 | 3358 | |
362a61ad NP |
3359 | smp_wmb(); /* See comment in __pte_alloc */ |
3360 | ||
872fec16 | 3361 | spin_lock(&mm->page_table_lock); |
1bb3630e | 3362 | if (pgd_present(*pgd)) /* Another has populated it */ |
5e541973 | 3363 | pud_free(mm, new); |
1bb3630e HD |
3364 | else |
3365 | pgd_populate(mm, pgd, new); | |
c74df32c | 3366 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 3367 | return 0; |
1da177e4 LT |
3368 | } |
3369 | #endif /* __PAGETABLE_PUD_FOLDED */ | |
3370 | ||
3371 | #ifndef __PAGETABLE_PMD_FOLDED | |
3372 | /* | |
3373 | * Allocate page middle directory. | |
872fec16 | 3374 | * We've already handled the fast-path in-line. |
1da177e4 | 3375 | */ |
1bb3630e | 3376 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
1da177e4 | 3377 | { |
c74df32c HD |
3378 | pmd_t *new = pmd_alloc_one(mm, address); |
3379 | if (!new) | |
1bb3630e | 3380 | return -ENOMEM; |
1da177e4 | 3381 | |
362a61ad NP |
3382 | smp_wmb(); /* See comment in __pte_alloc */ |
3383 | ||
872fec16 | 3384 | spin_lock(&mm->page_table_lock); |
1da177e4 | 3385 | #ifndef __ARCH_HAS_4LEVEL_HACK |
1bb3630e | 3386 | if (pud_present(*pud)) /* Another has populated it */ |
5e541973 | 3387 | pmd_free(mm, new); |
1bb3630e HD |
3388 | else |
3389 | pud_populate(mm, pud, new); | |
1da177e4 | 3390 | #else |
1bb3630e | 3391 | if (pgd_present(*pud)) /* Another has populated it */ |
5e541973 | 3392 | pmd_free(mm, new); |
1bb3630e HD |
3393 | else |
3394 | pgd_populate(mm, pud, new); | |
1da177e4 | 3395 | #endif /* __ARCH_HAS_4LEVEL_HACK */ |
c74df32c | 3396 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 3397 | return 0; |
e0f39591 | 3398 | } |
1da177e4 LT |
3399 | #endif /* __PAGETABLE_PMD_FOLDED */ |
3400 | ||
3401 | int make_pages_present(unsigned long addr, unsigned long end) | |
3402 | { | |
3403 | int ret, len, write; | |
3404 | struct vm_area_struct * vma; | |
3405 | ||
3406 | vma = find_vma(current->mm, addr); | |
3407 | if (!vma) | |
a477097d | 3408 | return -ENOMEM; |
5ecfda04 ML |
3409 | /* |
3410 | * We want to touch writable mappings with a write fault in order | |
3411 | * to break COW, except for shared mappings because these don't COW | |
3412 | * and we would not want to dirty them for nothing. | |
3413 | */ | |
3414 | write = (vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE; | |
5bcb28b1 ES |
3415 | BUG_ON(addr >= end); |
3416 | BUG_ON(end > vma->vm_end); | |
68e116a3 | 3417 | len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE; |
1da177e4 LT |
3418 | ret = get_user_pages(current, current->mm, addr, |
3419 | len, write, 0, NULL, NULL); | |
c11d69d8 | 3420 | if (ret < 0) |
1da177e4 | 3421 | return ret; |
9978ad58 | 3422 | return ret == len ? 0 : -EFAULT; |
1da177e4 LT |
3423 | } |
3424 | ||
1da177e4 LT |
3425 | #if !defined(__HAVE_ARCH_GATE_AREA) |
3426 | ||
3427 | #if defined(AT_SYSINFO_EHDR) | |
5ce7852c | 3428 | static struct vm_area_struct gate_vma; |
1da177e4 LT |
3429 | |
3430 | static int __init gate_vma_init(void) | |
3431 | { | |
3432 | gate_vma.vm_mm = NULL; | |
3433 | gate_vma.vm_start = FIXADDR_USER_START; | |
3434 | gate_vma.vm_end = FIXADDR_USER_END; | |
b6558c4a RM |
3435 | gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC; |
3436 | gate_vma.vm_page_prot = __P101; | |
f47aef55 RM |
3437 | /* |
3438 | * Make sure the vDSO gets into every core dump. | |
3439 | * Dumping its contents makes post-mortem fully interpretable later | |
3440 | * without matching up the same kernel and hardware config to see | |
3441 | * what PC values meant. | |
3442 | */ | |
3443 | gate_vma.vm_flags |= VM_ALWAYSDUMP; | |
1da177e4 LT |
3444 | return 0; |
3445 | } | |
3446 | __initcall(gate_vma_init); | |
3447 | #endif | |
3448 | ||
31db58b3 | 3449 | struct vm_area_struct *get_gate_vma(struct mm_struct *mm) |
1da177e4 LT |
3450 | { |
3451 | #ifdef AT_SYSINFO_EHDR | |
3452 | return &gate_vma; | |
3453 | #else | |
3454 | return NULL; | |
3455 | #endif | |
3456 | } | |
3457 | ||
cae5d390 | 3458 | int in_gate_area_no_mm(unsigned long addr) |
1da177e4 LT |
3459 | { |
3460 | #ifdef AT_SYSINFO_EHDR | |
3461 | if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END)) | |
3462 | return 1; | |
3463 | #endif | |
3464 | return 0; | |
3465 | } | |
3466 | ||
3467 | #endif /* __HAVE_ARCH_GATE_AREA */ | |
0ec76a11 | 3468 | |
1b36ba81 | 3469 | static int __follow_pte(struct mm_struct *mm, unsigned long address, |
f8ad0f49 JW |
3470 | pte_t **ptepp, spinlock_t **ptlp) |
3471 | { | |
3472 | pgd_t *pgd; | |
3473 | pud_t *pud; | |
3474 | pmd_t *pmd; | |
3475 | pte_t *ptep; | |
3476 | ||
3477 | pgd = pgd_offset(mm, address); | |
3478 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
3479 | goto out; | |
3480 | ||
3481 | pud = pud_offset(pgd, address); | |
3482 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) | |
3483 | goto out; | |
3484 | ||
3485 | pmd = pmd_offset(pud, address); | |
f66055ab | 3486 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
f8ad0f49 JW |
3487 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) |
3488 | goto out; | |
3489 | ||
3490 | /* We cannot handle huge page PFN maps. Luckily they don't exist. */ | |
3491 | if (pmd_huge(*pmd)) | |
3492 | goto out; | |
3493 | ||
3494 | ptep = pte_offset_map_lock(mm, pmd, address, ptlp); | |
3495 | if (!ptep) | |
3496 | goto out; | |
3497 | if (!pte_present(*ptep)) | |
3498 | goto unlock; | |
3499 | *ptepp = ptep; | |
3500 | return 0; | |
3501 | unlock: | |
3502 | pte_unmap_unlock(ptep, *ptlp); | |
3503 | out: | |
3504 | return -EINVAL; | |
3505 | } | |
3506 | ||
1b36ba81 NK |
3507 | static inline int follow_pte(struct mm_struct *mm, unsigned long address, |
3508 | pte_t **ptepp, spinlock_t **ptlp) | |
3509 | { | |
3510 | int res; | |
3511 | ||
3512 | /* (void) is needed to make gcc happy */ | |
3513 | (void) __cond_lock(*ptlp, | |
3514 | !(res = __follow_pte(mm, address, ptepp, ptlp))); | |
3515 | return res; | |
3516 | } | |
3517 | ||
3b6748e2 JW |
3518 | /** |
3519 | * follow_pfn - look up PFN at a user virtual address | |
3520 | * @vma: memory mapping | |
3521 | * @address: user virtual address | |
3522 | * @pfn: location to store found PFN | |
3523 | * | |
3524 | * Only IO mappings and raw PFN mappings are allowed. | |
3525 | * | |
3526 | * Returns zero and the pfn at @pfn on success, -ve otherwise. | |
3527 | */ | |
3528 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
3529 | unsigned long *pfn) | |
3530 | { | |
3531 | int ret = -EINVAL; | |
3532 | spinlock_t *ptl; | |
3533 | pte_t *ptep; | |
3534 | ||
3535 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
3536 | return ret; | |
3537 | ||
3538 | ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); | |
3539 | if (ret) | |
3540 | return ret; | |
3541 | *pfn = pte_pfn(*ptep); | |
3542 | pte_unmap_unlock(ptep, ptl); | |
3543 | return 0; | |
3544 | } | |
3545 | EXPORT_SYMBOL(follow_pfn); | |
3546 | ||
28b2ee20 | 3547 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
d87fe660 | 3548 | int follow_phys(struct vm_area_struct *vma, |
3549 | unsigned long address, unsigned int flags, | |
3550 | unsigned long *prot, resource_size_t *phys) | |
28b2ee20 | 3551 | { |
03668a4d | 3552 | int ret = -EINVAL; |
28b2ee20 RR |
3553 | pte_t *ptep, pte; |
3554 | spinlock_t *ptl; | |
28b2ee20 | 3555 | |
d87fe660 | 3556 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
3557 | goto out; | |
28b2ee20 | 3558 | |
03668a4d | 3559 | if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) |
d87fe660 | 3560 | goto out; |
28b2ee20 | 3561 | pte = *ptep; |
03668a4d | 3562 | |
28b2ee20 RR |
3563 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
3564 | goto unlock; | |
28b2ee20 RR |
3565 | |
3566 | *prot = pgprot_val(pte_pgprot(pte)); | |
03668a4d | 3567 | *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; |
28b2ee20 | 3568 | |
03668a4d | 3569 | ret = 0; |
28b2ee20 RR |
3570 | unlock: |
3571 | pte_unmap_unlock(ptep, ptl); | |
3572 | out: | |
d87fe660 | 3573 | return ret; |
28b2ee20 RR |
3574 | } |
3575 | ||
3576 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, | |
3577 | void *buf, int len, int write) | |
3578 | { | |
3579 | resource_size_t phys_addr; | |
3580 | unsigned long prot = 0; | |
2bc7273b | 3581 | void __iomem *maddr; |
28b2ee20 RR |
3582 | int offset = addr & (PAGE_SIZE-1); |
3583 | ||
d87fe660 | 3584 | if (follow_phys(vma, addr, write, &prot, &phys_addr)) |
28b2ee20 RR |
3585 | return -EINVAL; |
3586 | ||
3587 | maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot); | |
3588 | if (write) | |
3589 | memcpy_toio(maddr + offset, buf, len); | |
3590 | else | |
3591 | memcpy_fromio(buf, maddr + offset, len); | |
3592 | iounmap(maddr); | |
3593 | ||
3594 | return len; | |
3595 | } | |
3596 | #endif | |
3597 | ||
0ec76a11 | 3598 | /* |
206cb636 SW |
3599 | * Access another process' address space as given in mm. If non-NULL, use the |
3600 | * given task for page fault accounting. | |
0ec76a11 | 3601 | */ |
206cb636 SW |
3602 | static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, |
3603 | unsigned long addr, void *buf, int len, int write) | |
0ec76a11 | 3604 | { |
0ec76a11 | 3605 | struct vm_area_struct *vma; |
0ec76a11 DH |
3606 | void *old_buf = buf; |
3607 | ||
0ec76a11 | 3608 | down_read(&mm->mmap_sem); |
183ff22b | 3609 | /* ignore errors, just check how much was successfully transferred */ |
0ec76a11 DH |
3610 | while (len) { |
3611 | int bytes, ret, offset; | |
3612 | void *maddr; | |
28b2ee20 | 3613 | struct page *page = NULL; |
0ec76a11 DH |
3614 | |
3615 | ret = get_user_pages(tsk, mm, addr, 1, | |
3616 | write, 1, &page, &vma); | |
28b2ee20 RR |
3617 | if (ret <= 0) { |
3618 | /* | |
3619 | * Check if this is a VM_IO | VM_PFNMAP VMA, which | |
3620 | * we can access using slightly different code. | |
3621 | */ | |
3622 | #ifdef CONFIG_HAVE_IOREMAP_PROT | |
3623 | vma = find_vma(mm, addr); | |
fe936dfc | 3624 | if (!vma || vma->vm_start > addr) |
28b2ee20 RR |
3625 | break; |
3626 | if (vma->vm_ops && vma->vm_ops->access) | |
3627 | ret = vma->vm_ops->access(vma, addr, buf, | |
3628 | len, write); | |
3629 | if (ret <= 0) | |
3630 | #endif | |
3631 | break; | |
3632 | bytes = ret; | |
0ec76a11 | 3633 | } else { |
28b2ee20 RR |
3634 | bytes = len; |
3635 | offset = addr & (PAGE_SIZE-1); | |
3636 | if (bytes > PAGE_SIZE-offset) | |
3637 | bytes = PAGE_SIZE-offset; | |
3638 | ||
3639 | maddr = kmap(page); | |
3640 | if (write) { | |
3641 | copy_to_user_page(vma, page, addr, | |
3642 | maddr + offset, buf, bytes); | |
3643 | set_page_dirty_lock(page); | |
3644 | } else { | |
3645 | copy_from_user_page(vma, page, addr, | |
3646 | buf, maddr + offset, bytes); | |
3647 | } | |
3648 | kunmap(page); | |
3649 | page_cache_release(page); | |
0ec76a11 | 3650 | } |
0ec76a11 DH |
3651 | len -= bytes; |
3652 | buf += bytes; | |
3653 | addr += bytes; | |
3654 | } | |
3655 | up_read(&mm->mmap_sem); | |
0ec76a11 DH |
3656 | |
3657 | return buf - old_buf; | |
3658 | } | |
03252919 | 3659 | |
5ddd36b9 | 3660 | /** |
ae91dbfc | 3661 | * access_remote_vm - access another process' address space |
5ddd36b9 SW |
3662 | * @mm: the mm_struct of the target address space |
3663 | * @addr: start address to access | |
3664 | * @buf: source or destination buffer | |
3665 | * @len: number of bytes to transfer | |
3666 | * @write: whether the access is a write | |
3667 | * | |
3668 | * The caller must hold a reference on @mm. | |
3669 | */ | |
3670 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, | |
3671 | void *buf, int len, int write) | |
3672 | { | |
3673 | return __access_remote_vm(NULL, mm, addr, buf, len, write); | |
3674 | } | |
3675 | ||
206cb636 SW |
3676 | /* |
3677 | * Access another process' address space. | |
3678 | * Source/target buffer must be kernel space, | |
3679 | * Do not walk the page table directly, use get_user_pages | |
3680 | */ | |
3681 | int access_process_vm(struct task_struct *tsk, unsigned long addr, | |
3682 | void *buf, int len, int write) | |
3683 | { | |
3684 | struct mm_struct *mm; | |
3685 | int ret; | |
3686 | ||
3687 | mm = get_task_mm(tsk); | |
3688 | if (!mm) | |
3689 | return 0; | |
3690 | ||
3691 | ret = __access_remote_vm(tsk, mm, addr, buf, len, write); | |
3692 | mmput(mm); | |
3693 | ||
3694 | return ret; | |
3695 | } | |
3696 | ||
03252919 AK |
3697 | /* |
3698 | * Print the name of a VMA. | |
3699 | */ | |
3700 | void print_vma_addr(char *prefix, unsigned long ip) | |
3701 | { | |
3702 | struct mm_struct *mm = current->mm; | |
3703 | struct vm_area_struct *vma; | |
3704 | ||
e8bff74a IM |
3705 | /* |
3706 | * Do not print if we are in atomic | |
3707 | * contexts (in exception stacks, etc.): | |
3708 | */ | |
3709 | if (preempt_count()) | |
3710 | return; | |
3711 | ||
03252919 AK |
3712 | down_read(&mm->mmap_sem); |
3713 | vma = find_vma(mm, ip); | |
3714 | if (vma && vma->vm_file) { | |
3715 | struct file *f = vma->vm_file; | |
3716 | char *buf = (char *)__get_free_page(GFP_KERNEL); | |
3717 | if (buf) { | |
3718 | char *p, *s; | |
3719 | ||
cf28b486 | 3720 | p = d_path(&f->f_path, buf, PAGE_SIZE); |
03252919 AK |
3721 | if (IS_ERR(p)) |
3722 | p = "?"; | |
3723 | s = strrchr(p, '/'); | |
3724 | if (s) | |
3725 | p = s+1; | |
3726 | printk("%s%s[%lx+%lx]", prefix, p, | |
3727 | vma->vm_start, | |
3728 | vma->vm_end - vma->vm_start); | |
3729 | free_page((unsigned long)buf); | |
3730 | } | |
3731 | } | |
3732 | up_read(¤t->mm->mmap_sem); | |
3733 | } | |
3ee1afa3 NP |
3734 | |
3735 | #ifdef CONFIG_PROVE_LOCKING | |
3736 | void might_fault(void) | |
3737 | { | |
95156f00 PZ |
3738 | /* |
3739 | * Some code (nfs/sunrpc) uses socket ops on kernel memory while | |
3740 | * holding the mmap_sem, this is safe because kernel memory doesn't | |
3741 | * get paged out, therefore we'll never actually fault, and the | |
3742 | * below annotations will generate false positives. | |
3743 | */ | |
3744 | if (segment_eq(get_fs(), KERNEL_DS)) | |
3745 | return; | |
3746 | ||
3ee1afa3 NP |
3747 | might_sleep(); |
3748 | /* | |
3749 | * it would be nicer only to annotate paths which are not under | |
3750 | * pagefault_disable, however that requires a larger audit and | |
3751 | * providing helpers like get_user_atomic. | |
3752 | */ | |
3753 | if (!in_atomic() && current->mm) | |
3754 | might_lock_read(¤t->mm->mmap_sem); | |
3755 | } | |
3756 | EXPORT_SYMBOL(might_fault); | |
3757 | #endif | |
47ad8475 AA |
3758 | |
3759 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) | |
3760 | static void clear_gigantic_page(struct page *page, | |
3761 | unsigned long addr, | |
3762 | unsigned int pages_per_huge_page) | |
3763 | { | |
3764 | int i; | |
3765 | struct page *p = page; | |
3766 | ||
3767 | might_sleep(); | |
3768 | for (i = 0; i < pages_per_huge_page; | |
3769 | i++, p = mem_map_next(p, page, i)) { | |
3770 | cond_resched(); | |
3771 | clear_user_highpage(p, addr + i * PAGE_SIZE); | |
3772 | } | |
3773 | } | |
3774 | void clear_huge_page(struct page *page, | |
3775 | unsigned long addr, unsigned int pages_per_huge_page) | |
3776 | { | |
3777 | int i; | |
3778 | ||
3779 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
3780 | clear_gigantic_page(page, addr, pages_per_huge_page); | |
3781 | return; | |
3782 | } | |
3783 | ||
3784 | might_sleep(); | |
3785 | for (i = 0; i < pages_per_huge_page; i++) { | |
3786 | cond_resched(); | |
3787 | clear_user_highpage(page + i, addr + i * PAGE_SIZE); | |
3788 | } | |
3789 | } | |
3790 | ||
3791 | static void copy_user_gigantic_page(struct page *dst, struct page *src, | |
3792 | unsigned long addr, | |
3793 | struct vm_area_struct *vma, | |
3794 | unsigned int pages_per_huge_page) | |
3795 | { | |
3796 | int i; | |
3797 | struct page *dst_base = dst; | |
3798 | struct page *src_base = src; | |
3799 | ||
3800 | for (i = 0; i < pages_per_huge_page; ) { | |
3801 | cond_resched(); | |
3802 | copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); | |
3803 | ||
3804 | i++; | |
3805 | dst = mem_map_next(dst, dst_base, i); | |
3806 | src = mem_map_next(src, src_base, i); | |
3807 | } | |
3808 | } | |
3809 | ||
3810 | void copy_user_huge_page(struct page *dst, struct page *src, | |
3811 | unsigned long addr, struct vm_area_struct *vma, | |
3812 | unsigned int pages_per_huge_page) | |
3813 | { | |
3814 | int i; | |
3815 | ||
3816 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
3817 | copy_user_gigantic_page(dst, src, addr, vma, | |
3818 | pages_per_huge_page); | |
3819 | return; | |
3820 | } | |
3821 | ||
3822 | might_sleep(); | |
3823 | for (i = 0; i < pages_per_huge_page; i++) { | |
3824 | cond_resched(); | |
3825 | copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma); | |
3826 | } | |
3827 | } | |
3828 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ |