]>
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 | 49 | #include <linux/rmap.h> |
b95f1b31 | 50 | #include <linux/export.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> |
4daae3b4 | 60 | #include <linux/migrate.h> |
2fbc57c5 | 61 | #include <linux/string.h> |
0abdd7a8 | 62 | #include <linux/dma-debug.h> |
1da177e4 | 63 | |
6952b61d | 64 | #include <asm/io.h> |
1da177e4 LT |
65 | #include <asm/pgalloc.h> |
66 | #include <asm/uaccess.h> | |
67 | #include <asm/tlb.h> | |
68 | #include <asm/tlbflush.h> | |
69 | #include <asm/pgtable.h> | |
70 | ||
42b77728 JB |
71 | #include "internal.h" |
72 | ||
90572890 PZ |
73 | #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS |
74 | #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid. | |
75980e97 PZ |
75 | #endif |
76 | ||
d41dee36 | 77 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
78 | /* use the per-pgdat data instead for discontigmem - mbligh */ |
79 | unsigned long max_mapnr; | |
80 | struct page *mem_map; | |
81 | ||
82 | EXPORT_SYMBOL(max_mapnr); | |
83 | EXPORT_SYMBOL(mem_map); | |
84 | #endif | |
85 | ||
1da177e4 LT |
86 | /* |
87 | * A number of key systems in x86 including ioremap() rely on the assumption | |
88 | * that high_memory defines the upper bound on direct map memory, then end | |
89 | * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and | |
90 | * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL | |
91 | * and ZONE_HIGHMEM. | |
92 | */ | |
93 | void * high_memory; | |
1da177e4 | 94 | |
1da177e4 | 95 | EXPORT_SYMBOL(high_memory); |
1da177e4 | 96 | |
32a93233 IM |
97 | /* |
98 | * Randomize the address space (stacks, mmaps, brk, etc.). | |
99 | * | |
100 | * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, | |
101 | * as ancient (libc5 based) binaries can segfault. ) | |
102 | */ | |
103 | int randomize_va_space __read_mostly = | |
104 | #ifdef CONFIG_COMPAT_BRK | |
105 | 1; | |
106 | #else | |
107 | 2; | |
108 | #endif | |
a62eaf15 AK |
109 | |
110 | static int __init disable_randmaps(char *s) | |
111 | { | |
112 | randomize_va_space = 0; | |
9b41046c | 113 | return 1; |
a62eaf15 AK |
114 | } |
115 | __setup("norandmaps", disable_randmaps); | |
116 | ||
62eede62 | 117 | unsigned long zero_pfn __read_mostly; |
03f6462a | 118 | unsigned long highest_memmap_pfn __read_mostly; |
a13ea5b7 HD |
119 | |
120 | /* | |
121 | * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() | |
122 | */ | |
123 | static int __init init_zero_pfn(void) | |
124 | { | |
125 | zero_pfn = page_to_pfn(ZERO_PAGE(0)); | |
126 | return 0; | |
127 | } | |
128 | core_initcall(init_zero_pfn); | |
a62eaf15 | 129 | |
d559db08 | 130 | |
34e55232 KH |
131 | #if defined(SPLIT_RSS_COUNTING) |
132 | ||
ea48cf78 | 133 | void sync_mm_rss(struct mm_struct *mm) |
34e55232 KH |
134 | { |
135 | int i; | |
136 | ||
137 | for (i = 0; i < NR_MM_COUNTERS; i++) { | |
05af2e10 DR |
138 | if (current->rss_stat.count[i]) { |
139 | add_mm_counter(mm, i, current->rss_stat.count[i]); | |
140 | current->rss_stat.count[i] = 0; | |
34e55232 KH |
141 | } |
142 | } | |
05af2e10 | 143 | current->rss_stat.events = 0; |
34e55232 KH |
144 | } |
145 | ||
146 | static void add_mm_counter_fast(struct mm_struct *mm, int member, int val) | |
147 | { | |
148 | struct task_struct *task = current; | |
149 | ||
150 | if (likely(task->mm == mm)) | |
151 | task->rss_stat.count[member] += val; | |
152 | else | |
153 | add_mm_counter(mm, member, val); | |
154 | } | |
155 | #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1) | |
156 | #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1) | |
157 | ||
158 | /* sync counter once per 64 page faults */ | |
159 | #define TASK_RSS_EVENTS_THRESH (64) | |
160 | static void check_sync_rss_stat(struct task_struct *task) | |
161 | { | |
162 | if (unlikely(task != current)) | |
163 | return; | |
164 | if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH)) | |
ea48cf78 | 165 | sync_mm_rss(task->mm); |
34e55232 | 166 | } |
9547d01b | 167 | #else /* SPLIT_RSS_COUNTING */ |
34e55232 KH |
168 | |
169 | #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member) | |
170 | #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member) | |
171 | ||
172 | static void check_sync_rss_stat(struct task_struct *task) | |
173 | { | |
174 | } | |
175 | ||
9547d01b PZ |
176 | #endif /* SPLIT_RSS_COUNTING */ |
177 | ||
178 | #ifdef HAVE_GENERIC_MMU_GATHER | |
179 | ||
180 | static int tlb_next_batch(struct mmu_gather *tlb) | |
181 | { | |
182 | struct mmu_gather_batch *batch; | |
183 | ||
184 | batch = tlb->active; | |
185 | if (batch->next) { | |
186 | tlb->active = batch->next; | |
187 | return 1; | |
188 | } | |
189 | ||
53a59fc6 MH |
190 | if (tlb->batch_count == MAX_GATHER_BATCH_COUNT) |
191 | return 0; | |
192 | ||
9547d01b PZ |
193 | batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0); |
194 | if (!batch) | |
195 | return 0; | |
196 | ||
53a59fc6 | 197 | tlb->batch_count++; |
9547d01b PZ |
198 | batch->next = NULL; |
199 | batch->nr = 0; | |
200 | batch->max = MAX_GATHER_BATCH; | |
201 | ||
202 | tlb->active->next = batch; | |
203 | tlb->active = batch; | |
204 | ||
205 | return 1; | |
206 | } | |
207 | ||
208 | /* tlb_gather_mmu | |
209 | * Called to initialize an (on-stack) mmu_gather structure for page-table | |
210 | * tear-down from @mm. The @fullmm argument is used when @mm is without | |
211 | * users and we're going to destroy the full address space (exit/execve). | |
212 | */ | |
2b047252 | 213 | void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, unsigned long start, unsigned long end) |
9547d01b PZ |
214 | { |
215 | tlb->mm = mm; | |
216 | ||
2b047252 LT |
217 | /* Is it from 0 to ~0? */ |
218 | tlb->fullmm = !(start | (end+1)); | |
1de14c3c | 219 | tlb->need_flush_all = 0; |
2b047252 LT |
220 | tlb->start = start; |
221 | tlb->end = end; | |
9547d01b | 222 | tlb->need_flush = 0; |
9547d01b PZ |
223 | tlb->local.next = NULL; |
224 | tlb->local.nr = 0; | |
225 | tlb->local.max = ARRAY_SIZE(tlb->__pages); | |
226 | tlb->active = &tlb->local; | |
53a59fc6 | 227 | tlb->batch_count = 0; |
9547d01b PZ |
228 | |
229 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE | |
230 | tlb->batch = NULL; | |
231 | #endif | |
232 | } | |
233 | ||
234 | void tlb_flush_mmu(struct mmu_gather *tlb) | |
235 | { | |
236 | struct mmu_gather_batch *batch; | |
237 | ||
238 | if (!tlb->need_flush) | |
239 | return; | |
240 | tlb->need_flush = 0; | |
241 | tlb_flush(tlb); | |
242 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE | |
243 | tlb_table_flush(tlb); | |
34e55232 KH |
244 | #endif |
245 | ||
9547d01b PZ |
246 | for (batch = &tlb->local; batch; batch = batch->next) { |
247 | free_pages_and_swap_cache(batch->pages, batch->nr); | |
248 | batch->nr = 0; | |
249 | } | |
250 | tlb->active = &tlb->local; | |
251 | } | |
252 | ||
253 | /* tlb_finish_mmu | |
254 | * Called at the end of the shootdown operation to free up any resources | |
255 | * that were required. | |
256 | */ | |
257 | void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end) | |
258 | { | |
259 | struct mmu_gather_batch *batch, *next; | |
260 | ||
261 | tlb_flush_mmu(tlb); | |
262 | ||
263 | /* keep the page table cache within bounds */ | |
264 | check_pgt_cache(); | |
265 | ||
266 | for (batch = tlb->local.next; batch; batch = next) { | |
267 | next = batch->next; | |
268 | free_pages((unsigned long)batch, 0); | |
269 | } | |
270 | tlb->local.next = NULL; | |
271 | } | |
272 | ||
273 | /* __tlb_remove_page | |
274 | * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while | |
275 | * handling the additional races in SMP caused by other CPUs caching valid | |
276 | * mappings in their TLBs. Returns the number of free page slots left. | |
277 | * When out of page slots we must call tlb_flush_mmu(). | |
278 | */ | |
279 | int __tlb_remove_page(struct mmu_gather *tlb, struct page *page) | |
280 | { | |
281 | struct mmu_gather_batch *batch; | |
282 | ||
f21760b1 | 283 | VM_BUG_ON(!tlb->need_flush); |
9547d01b | 284 | |
9547d01b PZ |
285 | batch = tlb->active; |
286 | batch->pages[batch->nr++] = page; | |
287 | if (batch->nr == batch->max) { | |
288 | if (!tlb_next_batch(tlb)) | |
289 | return 0; | |
0b43c3aa | 290 | batch = tlb->active; |
9547d01b | 291 | } |
309381fe | 292 | VM_BUG_ON_PAGE(batch->nr > batch->max, page); |
9547d01b PZ |
293 | |
294 | return batch->max - batch->nr; | |
295 | } | |
296 | ||
297 | #endif /* HAVE_GENERIC_MMU_GATHER */ | |
298 | ||
26723911 PZ |
299 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE |
300 | ||
301 | /* | |
302 | * See the comment near struct mmu_table_batch. | |
303 | */ | |
304 | ||
305 | static void tlb_remove_table_smp_sync(void *arg) | |
306 | { | |
307 | /* Simply deliver the interrupt */ | |
308 | } | |
309 | ||
310 | static void tlb_remove_table_one(void *table) | |
311 | { | |
312 | /* | |
313 | * This isn't an RCU grace period and hence the page-tables cannot be | |
314 | * assumed to be actually RCU-freed. | |
315 | * | |
316 | * It is however sufficient for software page-table walkers that rely on | |
317 | * IRQ disabling. See the comment near struct mmu_table_batch. | |
318 | */ | |
319 | smp_call_function(tlb_remove_table_smp_sync, NULL, 1); | |
320 | __tlb_remove_table(table); | |
321 | } | |
322 | ||
323 | static void tlb_remove_table_rcu(struct rcu_head *head) | |
324 | { | |
325 | struct mmu_table_batch *batch; | |
326 | int i; | |
327 | ||
328 | batch = container_of(head, struct mmu_table_batch, rcu); | |
329 | ||
330 | for (i = 0; i < batch->nr; i++) | |
331 | __tlb_remove_table(batch->tables[i]); | |
332 | ||
333 | free_page((unsigned long)batch); | |
334 | } | |
335 | ||
336 | void tlb_table_flush(struct mmu_gather *tlb) | |
337 | { | |
338 | struct mmu_table_batch **batch = &tlb->batch; | |
339 | ||
340 | if (*batch) { | |
341 | call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu); | |
342 | *batch = NULL; | |
343 | } | |
344 | } | |
345 | ||
346 | void tlb_remove_table(struct mmu_gather *tlb, void *table) | |
347 | { | |
348 | struct mmu_table_batch **batch = &tlb->batch; | |
349 | ||
350 | tlb->need_flush = 1; | |
351 | ||
352 | /* | |
353 | * When there's less then two users of this mm there cannot be a | |
354 | * concurrent page-table walk. | |
355 | */ | |
356 | if (atomic_read(&tlb->mm->mm_users) < 2) { | |
357 | __tlb_remove_table(table); | |
358 | return; | |
359 | } | |
360 | ||
361 | if (*batch == NULL) { | |
362 | *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN); | |
363 | if (*batch == NULL) { | |
364 | tlb_remove_table_one(table); | |
365 | return; | |
366 | } | |
367 | (*batch)->nr = 0; | |
368 | } | |
369 | (*batch)->tables[(*batch)->nr++] = table; | |
370 | if ((*batch)->nr == MAX_TABLE_BATCH) | |
371 | tlb_table_flush(tlb); | |
372 | } | |
373 | ||
9547d01b | 374 | #endif /* CONFIG_HAVE_RCU_TABLE_FREE */ |
26723911 | 375 | |
1da177e4 LT |
376 | /* |
377 | * Note: this doesn't free the actual pages themselves. That | |
378 | * has been handled earlier when unmapping all the memory regions. | |
379 | */ | |
9e1b32ca BH |
380 | static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, |
381 | unsigned long addr) | |
1da177e4 | 382 | { |
2f569afd | 383 | pgtable_t token = pmd_pgtable(*pmd); |
e0da382c | 384 | pmd_clear(pmd); |
9e1b32ca | 385 | pte_free_tlb(tlb, token, addr); |
e1f56c89 | 386 | atomic_long_dec(&tlb->mm->nr_ptes); |
1da177e4 LT |
387 | } |
388 | ||
e0da382c HD |
389 | static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
390 | unsigned long addr, unsigned long end, | |
391 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
392 | { |
393 | pmd_t *pmd; | |
394 | unsigned long next; | |
e0da382c | 395 | unsigned long start; |
1da177e4 | 396 | |
e0da382c | 397 | start = addr; |
1da177e4 | 398 | pmd = pmd_offset(pud, addr); |
1da177e4 LT |
399 | do { |
400 | next = pmd_addr_end(addr, end); | |
401 | if (pmd_none_or_clear_bad(pmd)) | |
402 | continue; | |
9e1b32ca | 403 | free_pte_range(tlb, pmd, addr); |
1da177e4 LT |
404 | } while (pmd++, addr = next, addr != end); |
405 | ||
e0da382c HD |
406 | start &= PUD_MASK; |
407 | if (start < floor) | |
408 | return; | |
409 | if (ceiling) { | |
410 | ceiling &= PUD_MASK; | |
411 | if (!ceiling) | |
412 | return; | |
1da177e4 | 413 | } |
e0da382c HD |
414 | if (end - 1 > ceiling - 1) |
415 | return; | |
416 | ||
417 | pmd = pmd_offset(pud, start); | |
418 | pud_clear(pud); | |
9e1b32ca | 419 | pmd_free_tlb(tlb, pmd, start); |
1da177e4 LT |
420 | } |
421 | ||
e0da382c HD |
422 | static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, |
423 | unsigned long addr, unsigned long end, | |
424 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
425 | { |
426 | pud_t *pud; | |
427 | unsigned long next; | |
e0da382c | 428 | unsigned long start; |
1da177e4 | 429 | |
e0da382c | 430 | start = addr; |
1da177e4 | 431 | pud = pud_offset(pgd, addr); |
1da177e4 LT |
432 | do { |
433 | next = pud_addr_end(addr, end); | |
434 | if (pud_none_or_clear_bad(pud)) | |
435 | continue; | |
e0da382c | 436 | free_pmd_range(tlb, pud, addr, next, floor, ceiling); |
1da177e4 LT |
437 | } while (pud++, addr = next, addr != end); |
438 | ||
e0da382c HD |
439 | start &= PGDIR_MASK; |
440 | if (start < floor) | |
441 | return; | |
442 | if (ceiling) { | |
443 | ceiling &= PGDIR_MASK; | |
444 | if (!ceiling) | |
445 | return; | |
1da177e4 | 446 | } |
e0da382c HD |
447 | if (end - 1 > ceiling - 1) |
448 | return; | |
449 | ||
450 | pud = pud_offset(pgd, start); | |
451 | pgd_clear(pgd); | |
9e1b32ca | 452 | pud_free_tlb(tlb, pud, start); |
1da177e4 LT |
453 | } |
454 | ||
455 | /* | |
e0da382c | 456 | * This function frees user-level page tables of a process. |
1da177e4 | 457 | */ |
42b77728 | 458 | void free_pgd_range(struct mmu_gather *tlb, |
e0da382c HD |
459 | unsigned long addr, unsigned long end, |
460 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
461 | { |
462 | pgd_t *pgd; | |
463 | unsigned long next; | |
e0da382c HD |
464 | |
465 | /* | |
466 | * The next few lines have given us lots of grief... | |
467 | * | |
468 | * Why are we testing PMD* at this top level? Because often | |
469 | * there will be no work to do at all, and we'd prefer not to | |
470 | * go all the way down to the bottom just to discover that. | |
471 | * | |
472 | * Why all these "- 1"s? Because 0 represents both the bottom | |
473 | * of the address space and the top of it (using -1 for the | |
474 | * top wouldn't help much: the masks would do the wrong thing). | |
475 | * The rule is that addr 0 and floor 0 refer to the bottom of | |
476 | * the address space, but end 0 and ceiling 0 refer to the top | |
477 | * Comparisons need to use "end - 1" and "ceiling - 1" (though | |
478 | * that end 0 case should be mythical). | |
479 | * | |
480 | * Wherever addr is brought up or ceiling brought down, we must | |
481 | * be careful to reject "the opposite 0" before it confuses the | |
482 | * subsequent tests. But what about where end is brought down | |
483 | * by PMD_SIZE below? no, end can't go down to 0 there. | |
484 | * | |
485 | * Whereas we round start (addr) and ceiling down, by different | |
486 | * masks at different levels, in order to test whether a table | |
487 | * now has no other vmas using it, so can be freed, we don't | |
488 | * bother to round floor or end up - the tests don't need that. | |
489 | */ | |
1da177e4 | 490 | |
e0da382c HD |
491 | addr &= PMD_MASK; |
492 | if (addr < floor) { | |
493 | addr += PMD_SIZE; | |
494 | if (!addr) | |
495 | return; | |
496 | } | |
497 | if (ceiling) { | |
498 | ceiling &= PMD_MASK; | |
499 | if (!ceiling) | |
500 | return; | |
501 | } | |
502 | if (end - 1 > ceiling - 1) | |
503 | end -= PMD_SIZE; | |
504 | if (addr > end - 1) | |
505 | return; | |
506 | ||
42b77728 | 507 | pgd = pgd_offset(tlb->mm, addr); |
1da177e4 LT |
508 | do { |
509 | next = pgd_addr_end(addr, end); | |
510 | if (pgd_none_or_clear_bad(pgd)) | |
511 | continue; | |
42b77728 | 512 | free_pud_range(tlb, pgd, addr, next, floor, ceiling); |
1da177e4 | 513 | } while (pgd++, addr = next, addr != end); |
e0da382c HD |
514 | } |
515 | ||
42b77728 | 516 | void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, |
3bf5ee95 | 517 | unsigned long floor, unsigned long ceiling) |
e0da382c HD |
518 | { |
519 | while (vma) { | |
520 | struct vm_area_struct *next = vma->vm_next; | |
521 | unsigned long addr = vma->vm_start; | |
522 | ||
8f4f8c16 | 523 | /* |
25d9e2d1 | 524 | * Hide vma from rmap and truncate_pagecache before freeing |
525 | * pgtables | |
8f4f8c16 | 526 | */ |
5beb4930 | 527 | unlink_anon_vmas(vma); |
8f4f8c16 HD |
528 | unlink_file_vma(vma); |
529 | ||
9da61aef | 530 | if (is_vm_hugetlb_page(vma)) { |
3bf5ee95 | 531 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
e0da382c | 532 | floor, next? next->vm_start: ceiling); |
3bf5ee95 HD |
533 | } else { |
534 | /* | |
535 | * Optimization: gather nearby vmas into one call down | |
536 | */ | |
537 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE | |
4866920b | 538 | && !is_vm_hugetlb_page(next)) { |
3bf5ee95 HD |
539 | vma = next; |
540 | next = vma->vm_next; | |
5beb4930 | 541 | unlink_anon_vmas(vma); |
8f4f8c16 | 542 | unlink_file_vma(vma); |
3bf5ee95 HD |
543 | } |
544 | free_pgd_range(tlb, addr, vma->vm_end, | |
545 | floor, next? next->vm_start: ceiling); | |
546 | } | |
e0da382c HD |
547 | vma = next; |
548 | } | |
1da177e4 LT |
549 | } |
550 | ||
8ac1f832 AA |
551 | int __pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma, |
552 | pmd_t *pmd, unsigned long address) | |
1da177e4 | 553 | { |
c4088ebd | 554 | spinlock_t *ptl; |
2f569afd | 555 | pgtable_t new = pte_alloc_one(mm, address); |
8ac1f832 | 556 | int wait_split_huge_page; |
1bb3630e HD |
557 | if (!new) |
558 | return -ENOMEM; | |
559 | ||
362a61ad NP |
560 | /* |
561 | * Ensure all pte setup (eg. pte page lock and page clearing) are | |
562 | * visible before the pte is made visible to other CPUs by being | |
563 | * put into page tables. | |
564 | * | |
565 | * The other side of the story is the pointer chasing in the page | |
566 | * table walking code (when walking the page table without locking; | |
567 | * ie. most of the time). Fortunately, these data accesses consist | |
568 | * of a chain of data-dependent loads, meaning most CPUs (alpha | |
569 | * being the notable exception) will already guarantee loads are | |
570 | * seen in-order. See the alpha page table accessors for the | |
571 | * smp_read_barrier_depends() barriers in page table walking code. | |
572 | */ | |
573 | smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ | |
574 | ||
c4088ebd | 575 | ptl = pmd_lock(mm, pmd); |
8ac1f832 AA |
576 | wait_split_huge_page = 0; |
577 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ | |
e1f56c89 | 578 | atomic_long_inc(&mm->nr_ptes); |
1da177e4 | 579 | pmd_populate(mm, pmd, new); |
2f569afd | 580 | new = NULL; |
8ac1f832 AA |
581 | } else if (unlikely(pmd_trans_splitting(*pmd))) |
582 | wait_split_huge_page = 1; | |
c4088ebd | 583 | spin_unlock(ptl); |
2f569afd MS |
584 | if (new) |
585 | pte_free(mm, new); | |
8ac1f832 AA |
586 | if (wait_split_huge_page) |
587 | wait_split_huge_page(vma->anon_vma, pmd); | |
1bb3630e | 588 | return 0; |
1da177e4 LT |
589 | } |
590 | ||
1bb3630e | 591 | int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) |
1da177e4 | 592 | { |
1bb3630e HD |
593 | pte_t *new = pte_alloc_one_kernel(&init_mm, address); |
594 | if (!new) | |
595 | return -ENOMEM; | |
596 | ||
362a61ad NP |
597 | smp_wmb(); /* See comment in __pte_alloc */ |
598 | ||
1bb3630e | 599 | spin_lock(&init_mm.page_table_lock); |
8ac1f832 | 600 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
1bb3630e | 601 | pmd_populate_kernel(&init_mm, pmd, new); |
2f569afd | 602 | new = NULL; |
8ac1f832 AA |
603 | } else |
604 | VM_BUG_ON(pmd_trans_splitting(*pmd)); | |
1bb3630e | 605 | spin_unlock(&init_mm.page_table_lock); |
2f569afd MS |
606 | if (new) |
607 | pte_free_kernel(&init_mm, new); | |
1bb3630e | 608 | return 0; |
1da177e4 LT |
609 | } |
610 | ||
d559db08 KH |
611 | static inline void init_rss_vec(int *rss) |
612 | { | |
613 | memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); | |
614 | } | |
615 | ||
616 | static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) | |
ae859762 | 617 | { |
d559db08 KH |
618 | int i; |
619 | ||
34e55232 | 620 | if (current->mm == mm) |
05af2e10 | 621 | sync_mm_rss(mm); |
d559db08 KH |
622 | for (i = 0; i < NR_MM_COUNTERS; i++) |
623 | if (rss[i]) | |
624 | add_mm_counter(mm, i, rss[i]); | |
ae859762 HD |
625 | } |
626 | ||
b5810039 | 627 | /* |
6aab341e LT |
628 | * This function is called to print an error when a bad pte |
629 | * is found. For example, we might have a PFN-mapped pte in | |
630 | * a region that doesn't allow it. | |
b5810039 NP |
631 | * |
632 | * The calling function must still handle the error. | |
633 | */ | |
3dc14741 HD |
634 | static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, |
635 | pte_t pte, struct page *page) | |
b5810039 | 636 | { |
3dc14741 HD |
637 | pgd_t *pgd = pgd_offset(vma->vm_mm, addr); |
638 | pud_t *pud = pud_offset(pgd, addr); | |
639 | pmd_t *pmd = pmd_offset(pud, addr); | |
640 | struct address_space *mapping; | |
641 | pgoff_t index; | |
d936cf9b HD |
642 | static unsigned long resume; |
643 | static unsigned long nr_shown; | |
644 | static unsigned long nr_unshown; | |
645 | ||
646 | /* | |
647 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
648 | * or allow a steady drip of one report per second. | |
649 | */ | |
650 | if (nr_shown == 60) { | |
651 | if (time_before(jiffies, resume)) { | |
652 | nr_unshown++; | |
653 | return; | |
654 | } | |
655 | if (nr_unshown) { | |
1e9e6365 HD |
656 | printk(KERN_ALERT |
657 | "BUG: Bad page map: %lu messages suppressed\n", | |
d936cf9b HD |
658 | nr_unshown); |
659 | nr_unshown = 0; | |
660 | } | |
661 | nr_shown = 0; | |
662 | } | |
663 | if (nr_shown++ == 0) | |
664 | resume = jiffies + 60 * HZ; | |
3dc14741 HD |
665 | |
666 | mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; | |
667 | index = linear_page_index(vma, addr); | |
668 | ||
1e9e6365 HD |
669 | printk(KERN_ALERT |
670 | "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", | |
3dc14741 HD |
671 | current->comm, |
672 | (long long)pte_val(pte), (long long)pmd_val(*pmd)); | |
718a3821 | 673 | if (page) |
f0b791a3 | 674 | dump_page(page, "bad pte"); |
1e9e6365 | 675 | printk(KERN_ALERT |
3dc14741 HD |
676 | "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n", |
677 | (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); | |
678 | /* | |
679 | * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y | |
680 | */ | |
681 | if (vma->vm_ops) | |
071361d3 JP |
682 | printk(KERN_ALERT "vma->vm_ops->fault: %pSR\n", |
683 | vma->vm_ops->fault); | |
72c2d531 | 684 | if (vma->vm_file) |
071361d3 JP |
685 | printk(KERN_ALERT "vma->vm_file->f_op->mmap: %pSR\n", |
686 | vma->vm_file->f_op->mmap); | |
b5810039 | 687 | dump_stack(); |
373d4d09 | 688 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
b5810039 NP |
689 | } |
690 | ||
2ec74c3e | 691 | static inline bool is_cow_mapping(vm_flags_t flags) |
67121172 LT |
692 | { |
693 | return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; | |
694 | } | |
695 | ||
ee498ed7 | 696 | /* |
7e675137 | 697 | * vm_normal_page -- This function gets the "struct page" associated with a pte. |
6aab341e | 698 | * |
7e675137 NP |
699 | * "Special" mappings do not wish to be associated with a "struct page" (either |
700 | * it doesn't exist, or it exists but they don't want to touch it). In this | |
701 | * case, NULL is returned here. "Normal" mappings do have a struct page. | |
b379d790 | 702 | * |
7e675137 NP |
703 | * There are 2 broad cases. Firstly, an architecture may define a pte_special() |
704 | * pte bit, in which case this function is trivial. Secondly, an architecture | |
705 | * may not have a spare pte bit, which requires a more complicated scheme, | |
706 | * described below. | |
707 | * | |
708 | * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a | |
709 | * special mapping (even if there are underlying and valid "struct pages"). | |
710 | * COWed pages of a VM_PFNMAP are always normal. | |
6aab341e | 711 | * |
b379d790 JH |
712 | * The way we recognize COWed pages within VM_PFNMAP mappings is through the |
713 | * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit | |
7e675137 NP |
714 | * set, and the vm_pgoff will point to the first PFN mapped: thus every special |
715 | * mapping will always honor the rule | |
6aab341e LT |
716 | * |
717 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) | |
718 | * | |
7e675137 NP |
719 | * And for normal mappings this is false. |
720 | * | |
721 | * This restricts such mappings to be a linear translation from virtual address | |
722 | * to pfn. To get around this restriction, we allow arbitrary mappings so long | |
723 | * as the vma is not a COW mapping; in that case, we know that all ptes are | |
724 | * special (because none can have been COWed). | |
b379d790 | 725 | * |
b379d790 | 726 | * |
7e675137 | 727 | * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. |
b379d790 JH |
728 | * |
729 | * VM_MIXEDMAP mappings can likewise contain memory with or without "struct | |
730 | * page" backing, however the difference is that _all_ pages with a struct | |
731 | * page (that is, those where pfn_valid is true) are refcounted and considered | |
732 | * normal pages by the VM. The disadvantage is that pages are refcounted | |
733 | * (which can be slower and simply not an option for some PFNMAP users). The | |
734 | * advantage is that we don't have to follow the strict linearity rule of | |
735 | * PFNMAP mappings in order to support COWable mappings. | |
736 | * | |
ee498ed7 | 737 | */ |
7e675137 NP |
738 | #ifdef __HAVE_ARCH_PTE_SPECIAL |
739 | # define HAVE_PTE_SPECIAL 1 | |
740 | #else | |
741 | # define HAVE_PTE_SPECIAL 0 | |
742 | #endif | |
743 | struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, | |
744 | pte_t pte) | |
ee498ed7 | 745 | { |
22b31eec | 746 | unsigned long pfn = pte_pfn(pte); |
7e675137 NP |
747 | |
748 | if (HAVE_PTE_SPECIAL) { | |
22b31eec HD |
749 | if (likely(!pte_special(pte))) |
750 | goto check_pfn; | |
a13ea5b7 HD |
751 | if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) |
752 | return NULL; | |
62eede62 | 753 | if (!is_zero_pfn(pfn)) |
22b31eec | 754 | print_bad_pte(vma, addr, pte, NULL); |
7e675137 NP |
755 | return NULL; |
756 | } | |
757 | ||
758 | /* !HAVE_PTE_SPECIAL case follows: */ | |
759 | ||
b379d790 JH |
760 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
761 | if (vma->vm_flags & VM_MIXEDMAP) { | |
762 | if (!pfn_valid(pfn)) | |
763 | return NULL; | |
764 | goto out; | |
765 | } else { | |
7e675137 NP |
766 | unsigned long off; |
767 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
b379d790 JH |
768 | if (pfn == vma->vm_pgoff + off) |
769 | return NULL; | |
770 | if (!is_cow_mapping(vma->vm_flags)) | |
771 | return NULL; | |
772 | } | |
6aab341e LT |
773 | } |
774 | ||
62eede62 HD |
775 | if (is_zero_pfn(pfn)) |
776 | return NULL; | |
22b31eec HD |
777 | check_pfn: |
778 | if (unlikely(pfn > highest_memmap_pfn)) { | |
779 | print_bad_pte(vma, addr, pte, NULL); | |
780 | return NULL; | |
781 | } | |
6aab341e LT |
782 | |
783 | /* | |
7e675137 | 784 | * NOTE! We still have PageReserved() pages in the page tables. |
7e675137 | 785 | * eg. VDSO mappings can cause them to exist. |
6aab341e | 786 | */ |
b379d790 | 787 | out: |
6aab341e | 788 | return pfn_to_page(pfn); |
ee498ed7 HD |
789 | } |
790 | ||
1da177e4 LT |
791 | /* |
792 | * copy one vm_area from one task to the other. Assumes the page tables | |
793 | * already present in the new task to be cleared in the whole range | |
794 | * covered by this vma. | |
1da177e4 LT |
795 | */ |
796 | ||
570a335b | 797 | static inline unsigned long |
1da177e4 | 798 | copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
b5810039 | 799 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, |
8c103762 | 800 | unsigned long addr, int *rss) |
1da177e4 | 801 | { |
b5810039 | 802 | unsigned long vm_flags = vma->vm_flags; |
1da177e4 LT |
803 | pte_t pte = *src_pte; |
804 | struct page *page; | |
1da177e4 LT |
805 | |
806 | /* pte contains position in swap or file, so copy. */ | |
807 | if (unlikely(!pte_present(pte))) { | |
808 | if (!pte_file(pte)) { | |
0697212a CL |
809 | swp_entry_t entry = pte_to_swp_entry(pte); |
810 | ||
570a335b HD |
811 | if (swap_duplicate(entry) < 0) |
812 | return entry.val; | |
813 | ||
1da177e4 LT |
814 | /* make sure dst_mm is on swapoff's mmlist. */ |
815 | if (unlikely(list_empty(&dst_mm->mmlist))) { | |
816 | spin_lock(&mmlist_lock); | |
f412ac08 HD |
817 | if (list_empty(&dst_mm->mmlist)) |
818 | list_add(&dst_mm->mmlist, | |
819 | &src_mm->mmlist); | |
1da177e4 LT |
820 | spin_unlock(&mmlist_lock); |
821 | } | |
b084d435 KH |
822 | if (likely(!non_swap_entry(entry))) |
823 | rss[MM_SWAPENTS]++; | |
9f9f1acd KK |
824 | else if (is_migration_entry(entry)) { |
825 | page = migration_entry_to_page(entry); | |
826 | ||
827 | if (PageAnon(page)) | |
828 | rss[MM_ANONPAGES]++; | |
829 | else | |
830 | rss[MM_FILEPAGES]++; | |
831 | ||
832 | if (is_write_migration_entry(entry) && | |
833 | is_cow_mapping(vm_flags)) { | |
834 | /* | |
835 | * COW mappings require pages in both | |
836 | * parent and child to be set to read. | |
837 | */ | |
838 | make_migration_entry_read(&entry); | |
839 | pte = swp_entry_to_pte(entry); | |
c3d16e16 CG |
840 | if (pte_swp_soft_dirty(*src_pte)) |
841 | pte = pte_swp_mksoft_dirty(pte); | |
9f9f1acd KK |
842 | set_pte_at(src_mm, addr, src_pte, pte); |
843 | } | |
0697212a | 844 | } |
1da177e4 | 845 | } |
ae859762 | 846 | goto out_set_pte; |
1da177e4 LT |
847 | } |
848 | ||
1da177e4 LT |
849 | /* |
850 | * If it's a COW mapping, write protect it both | |
851 | * in the parent and the child | |
852 | */ | |
67121172 | 853 | if (is_cow_mapping(vm_flags)) { |
1da177e4 | 854 | ptep_set_wrprotect(src_mm, addr, src_pte); |
3dc90795 | 855 | pte = pte_wrprotect(pte); |
1da177e4 LT |
856 | } |
857 | ||
858 | /* | |
859 | * If it's a shared mapping, mark it clean in | |
860 | * the child | |
861 | */ | |
862 | if (vm_flags & VM_SHARED) | |
863 | pte = pte_mkclean(pte); | |
864 | pte = pte_mkold(pte); | |
6aab341e LT |
865 | |
866 | page = vm_normal_page(vma, addr, pte); | |
867 | if (page) { | |
868 | get_page(page); | |
21333b2b | 869 | page_dup_rmap(page); |
d559db08 KH |
870 | if (PageAnon(page)) |
871 | rss[MM_ANONPAGES]++; | |
872 | else | |
873 | rss[MM_FILEPAGES]++; | |
6aab341e | 874 | } |
ae859762 HD |
875 | |
876 | out_set_pte: | |
877 | set_pte_at(dst_mm, addr, dst_pte, pte); | |
570a335b | 878 | return 0; |
1da177e4 LT |
879 | } |
880 | ||
71e3aac0 AA |
881 | int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
882 | pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, | |
883 | unsigned long addr, unsigned long end) | |
1da177e4 | 884 | { |
c36987e2 | 885 | pte_t *orig_src_pte, *orig_dst_pte; |
1da177e4 | 886 | pte_t *src_pte, *dst_pte; |
c74df32c | 887 | spinlock_t *src_ptl, *dst_ptl; |
e040f218 | 888 | int progress = 0; |
d559db08 | 889 | int rss[NR_MM_COUNTERS]; |
570a335b | 890 | swp_entry_t entry = (swp_entry_t){0}; |
1da177e4 LT |
891 | |
892 | again: | |
d559db08 KH |
893 | init_rss_vec(rss); |
894 | ||
c74df32c | 895 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
1da177e4 LT |
896 | if (!dst_pte) |
897 | return -ENOMEM; | |
ece0e2b6 | 898 | src_pte = pte_offset_map(src_pmd, addr); |
4c21e2f2 | 899 | src_ptl = pte_lockptr(src_mm, src_pmd); |
f20dc5f7 | 900 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
c36987e2 DN |
901 | orig_src_pte = src_pte; |
902 | orig_dst_pte = dst_pte; | |
6606c3e0 | 903 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 904 | |
1da177e4 LT |
905 | do { |
906 | /* | |
907 | * We are holding two locks at this point - either of them | |
908 | * could generate latencies in another task on another CPU. | |
909 | */ | |
e040f218 HD |
910 | if (progress >= 32) { |
911 | progress = 0; | |
912 | if (need_resched() || | |
95c354fe | 913 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
e040f218 HD |
914 | break; |
915 | } | |
1da177e4 LT |
916 | if (pte_none(*src_pte)) { |
917 | progress++; | |
918 | continue; | |
919 | } | |
570a335b HD |
920 | entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, |
921 | vma, addr, rss); | |
922 | if (entry.val) | |
923 | break; | |
1da177e4 LT |
924 | progress += 8; |
925 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); | |
1da177e4 | 926 | |
6606c3e0 | 927 | arch_leave_lazy_mmu_mode(); |
c74df32c | 928 | spin_unlock(src_ptl); |
ece0e2b6 | 929 | pte_unmap(orig_src_pte); |
d559db08 | 930 | add_mm_rss_vec(dst_mm, rss); |
c36987e2 | 931 | pte_unmap_unlock(orig_dst_pte, dst_ptl); |
c74df32c | 932 | cond_resched(); |
570a335b HD |
933 | |
934 | if (entry.val) { | |
935 | if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) | |
936 | return -ENOMEM; | |
937 | progress = 0; | |
938 | } | |
1da177e4 LT |
939 | if (addr != end) |
940 | goto again; | |
941 | return 0; | |
942 | } | |
943 | ||
944 | static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
945 | pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, | |
946 | unsigned long addr, unsigned long end) | |
947 | { | |
948 | pmd_t *src_pmd, *dst_pmd; | |
949 | unsigned long next; | |
950 | ||
951 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); | |
952 | if (!dst_pmd) | |
953 | return -ENOMEM; | |
954 | src_pmd = pmd_offset(src_pud, addr); | |
955 | do { | |
956 | next = pmd_addr_end(addr, end); | |
71e3aac0 AA |
957 | if (pmd_trans_huge(*src_pmd)) { |
958 | int err; | |
14d1a55c | 959 | VM_BUG_ON(next-addr != HPAGE_PMD_SIZE); |
71e3aac0 AA |
960 | err = copy_huge_pmd(dst_mm, src_mm, |
961 | dst_pmd, src_pmd, addr, vma); | |
962 | if (err == -ENOMEM) | |
963 | return -ENOMEM; | |
964 | if (!err) | |
965 | continue; | |
966 | /* fall through */ | |
967 | } | |
1da177e4 LT |
968 | if (pmd_none_or_clear_bad(src_pmd)) |
969 | continue; | |
970 | if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, | |
971 | vma, addr, next)) | |
972 | return -ENOMEM; | |
973 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); | |
974 | return 0; | |
975 | } | |
976 | ||
977 | static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
978 | pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, | |
979 | unsigned long addr, unsigned long end) | |
980 | { | |
981 | pud_t *src_pud, *dst_pud; | |
982 | unsigned long next; | |
983 | ||
984 | dst_pud = pud_alloc(dst_mm, dst_pgd, addr); | |
985 | if (!dst_pud) | |
986 | return -ENOMEM; | |
987 | src_pud = pud_offset(src_pgd, addr); | |
988 | do { | |
989 | next = pud_addr_end(addr, end); | |
990 | if (pud_none_or_clear_bad(src_pud)) | |
991 | continue; | |
992 | if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, | |
993 | vma, addr, next)) | |
994 | return -ENOMEM; | |
995 | } while (dst_pud++, src_pud++, addr = next, addr != end); | |
996 | return 0; | |
997 | } | |
998 | ||
999 | int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
1000 | struct vm_area_struct *vma) | |
1001 | { | |
1002 | pgd_t *src_pgd, *dst_pgd; | |
1003 | unsigned long next; | |
1004 | unsigned long addr = vma->vm_start; | |
1005 | unsigned long end = vma->vm_end; | |
2ec74c3e SG |
1006 | unsigned long mmun_start; /* For mmu_notifiers */ |
1007 | unsigned long mmun_end; /* For mmu_notifiers */ | |
1008 | bool is_cow; | |
cddb8a5c | 1009 | int ret; |
1da177e4 | 1010 | |
d992895b NP |
1011 | /* |
1012 | * Don't copy ptes where a page fault will fill them correctly. | |
1013 | * Fork becomes much lighter when there are big shared or private | |
1014 | * readonly mappings. The tradeoff is that copy_page_range is more | |
1015 | * efficient than faulting. | |
1016 | */ | |
4b6e1e37 KK |
1017 | if (!(vma->vm_flags & (VM_HUGETLB | VM_NONLINEAR | |
1018 | VM_PFNMAP | VM_MIXEDMAP))) { | |
d992895b NP |
1019 | if (!vma->anon_vma) |
1020 | return 0; | |
1021 | } | |
1022 | ||
1da177e4 LT |
1023 | if (is_vm_hugetlb_page(vma)) |
1024 | return copy_hugetlb_page_range(dst_mm, src_mm, vma); | |
1025 | ||
b3b9c293 | 1026 | if (unlikely(vma->vm_flags & VM_PFNMAP)) { |
2ab64037 | 1027 | /* |
1028 | * We do not free on error cases below as remove_vma | |
1029 | * gets called on error from higher level routine | |
1030 | */ | |
5180da41 | 1031 | ret = track_pfn_copy(vma); |
2ab64037 | 1032 | if (ret) |
1033 | return ret; | |
1034 | } | |
1035 | ||
cddb8a5c AA |
1036 | /* |
1037 | * We need to invalidate the secondary MMU mappings only when | |
1038 | * there could be a permission downgrade on the ptes of the | |
1039 | * parent mm. And a permission downgrade will only happen if | |
1040 | * is_cow_mapping() returns true. | |
1041 | */ | |
2ec74c3e SG |
1042 | is_cow = is_cow_mapping(vma->vm_flags); |
1043 | mmun_start = addr; | |
1044 | mmun_end = end; | |
1045 | if (is_cow) | |
1046 | mmu_notifier_invalidate_range_start(src_mm, mmun_start, | |
1047 | mmun_end); | |
cddb8a5c AA |
1048 | |
1049 | ret = 0; | |
1da177e4 LT |
1050 | dst_pgd = pgd_offset(dst_mm, addr); |
1051 | src_pgd = pgd_offset(src_mm, addr); | |
1052 | do { | |
1053 | next = pgd_addr_end(addr, end); | |
1054 | if (pgd_none_or_clear_bad(src_pgd)) | |
1055 | continue; | |
cddb8a5c AA |
1056 | if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, |
1057 | vma, addr, next))) { | |
1058 | ret = -ENOMEM; | |
1059 | break; | |
1060 | } | |
1da177e4 | 1061 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); |
cddb8a5c | 1062 | |
2ec74c3e SG |
1063 | if (is_cow) |
1064 | mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end); | |
cddb8a5c | 1065 | return ret; |
1da177e4 LT |
1066 | } |
1067 | ||
51c6f666 | 1068 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
b5810039 | 1069 | struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 | 1070 | unsigned long addr, unsigned long end, |
97a89413 | 1071 | struct zap_details *details) |
1da177e4 | 1072 | { |
b5810039 | 1073 | struct mm_struct *mm = tlb->mm; |
d16dfc55 | 1074 | int force_flush = 0; |
d559db08 | 1075 | int rss[NR_MM_COUNTERS]; |
97a89413 | 1076 | spinlock_t *ptl; |
5f1a1907 | 1077 | pte_t *start_pte; |
97a89413 | 1078 | pte_t *pte; |
d559db08 | 1079 | |
d16dfc55 | 1080 | again: |
e303297e | 1081 | init_rss_vec(rss); |
5f1a1907 SR |
1082 | start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
1083 | pte = start_pte; | |
6606c3e0 | 1084 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1085 | do { |
1086 | pte_t ptent = *pte; | |
51c6f666 | 1087 | if (pte_none(ptent)) { |
1da177e4 | 1088 | continue; |
51c6f666 | 1089 | } |
6f5e6b9e | 1090 | |
1da177e4 | 1091 | if (pte_present(ptent)) { |
ee498ed7 | 1092 | struct page *page; |
51c6f666 | 1093 | |
6aab341e | 1094 | page = vm_normal_page(vma, addr, ptent); |
1da177e4 LT |
1095 | if (unlikely(details) && page) { |
1096 | /* | |
1097 | * unmap_shared_mapping_pages() wants to | |
1098 | * invalidate cache without truncating: | |
1099 | * unmap shared but keep private pages. | |
1100 | */ | |
1101 | if (details->check_mapping && | |
1102 | details->check_mapping != page->mapping) | |
1103 | continue; | |
1104 | /* | |
1105 | * Each page->index must be checked when | |
1106 | * invalidating or truncating nonlinear. | |
1107 | */ | |
1108 | if (details->nonlinear_vma && | |
1109 | (page->index < details->first_index || | |
1110 | page->index > details->last_index)) | |
1111 | continue; | |
1112 | } | |
b5810039 | 1113 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
a600388d | 1114 | tlb->fullmm); |
1da177e4 LT |
1115 | tlb_remove_tlb_entry(tlb, pte, addr); |
1116 | if (unlikely(!page)) | |
1117 | continue; | |
1118 | if (unlikely(details) && details->nonlinear_vma | |
1119 | && linear_page_index(details->nonlinear_vma, | |
41bb3476 CG |
1120 | addr) != page->index) { |
1121 | pte_t ptfile = pgoff_to_pte(page->index); | |
1122 | if (pte_soft_dirty(ptent)) | |
1123 | pte_file_mksoft_dirty(ptfile); | |
1124 | set_pte_at(mm, addr, pte, ptfile); | |
1125 | } | |
1da177e4 | 1126 | if (PageAnon(page)) |
d559db08 | 1127 | rss[MM_ANONPAGES]--; |
6237bcd9 HD |
1128 | else { |
1129 | if (pte_dirty(ptent)) | |
1130 | set_page_dirty(page); | |
4917e5d0 | 1131 | if (pte_young(ptent) && |
64363aad | 1132 | likely(!(vma->vm_flags & VM_SEQ_READ))) |
bf3f3bc5 | 1133 | mark_page_accessed(page); |
d559db08 | 1134 | rss[MM_FILEPAGES]--; |
6237bcd9 | 1135 | } |
edc315fd | 1136 | page_remove_rmap(page); |
3dc14741 HD |
1137 | if (unlikely(page_mapcount(page) < 0)) |
1138 | print_bad_pte(vma, addr, ptent, page); | |
d16dfc55 PZ |
1139 | force_flush = !__tlb_remove_page(tlb, page); |
1140 | if (force_flush) | |
1141 | break; | |
1da177e4 LT |
1142 | continue; |
1143 | } | |
1144 | /* | |
1145 | * If details->check_mapping, we leave swap entries; | |
1146 | * if details->nonlinear_vma, we leave file entries. | |
1147 | */ | |
1148 | if (unlikely(details)) | |
1149 | continue; | |
2509ef26 HD |
1150 | if (pte_file(ptent)) { |
1151 | if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) | |
1152 | print_bad_pte(vma, addr, ptent, NULL); | |
b084d435 KH |
1153 | } else { |
1154 | swp_entry_t entry = pte_to_swp_entry(ptent); | |
1155 | ||
1156 | if (!non_swap_entry(entry)) | |
1157 | rss[MM_SWAPENTS]--; | |
9f9f1acd KK |
1158 | else if (is_migration_entry(entry)) { |
1159 | struct page *page; | |
1160 | ||
1161 | page = migration_entry_to_page(entry); | |
1162 | ||
1163 | if (PageAnon(page)) | |
1164 | rss[MM_ANONPAGES]--; | |
1165 | else | |
1166 | rss[MM_FILEPAGES]--; | |
1167 | } | |
b084d435 KH |
1168 | if (unlikely(!free_swap_and_cache(entry))) |
1169 | print_bad_pte(vma, addr, ptent, NULL); | |
1170 | } | |
9888a1ca | 1171 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
97a89413 | 1172 | } while (pte++, addr += PAGE_SIZE, addr != end); |
ae859762 | 1173 | |
d559db08 | 1174 | add_mm_rss_vec(mm, rss); |
6606c3e0 | 1175 | arch_leave_lazy_mmu_mode(); |
5f1a1907 | 1176 | pte_unmap_unlock(start_pte, ptl); |
51c6f666 | 1177 | |
d16dfc55 PZ |
1178 | /* |
1179 | * mmu_gather ran out of room to batch pages, we break out of | |
1180 | * the PTE lock to avoid doing the potential expensive TLB invalidate | |
1181 | * and page-free while holding it. | |
1182 | */ | |
1183 | if (force_flush) { | |
2b047252 LT |
1184 | unsigned long old_end; |
1185 | ||
d16dfc55 | 1186 | force_flush = 0; |
597e1c35 | 1187 | |
2b047252 LT |
1188 | /* |
1189 | * Flush the TLB just for the previous segment, | |
1190 | * then update the range to be the remaining | |
1191 | * TLB range. | |
1192 | */ | |
1193 | old_end = tlb->end; | |
e6c495a9 | 1194 | tlb->end = addr; |
2b047252 | 1195 | |
d16dfc55 | 1196 | tlb_flush_mmu(tlb); |
2b047252 LT |
1197 | |
1198 | tlb->start = addr; | |
1199 | tlb->end = old_end; | |
1200 | ||
1201 | if (addr != end) | |
d16dfc55 PZ |
1202 | goto again; |
1203 | } | |
1204 | ||
51c6f666 | 1205 | return addr; |
1da177e4 LT |
1206 | } |
1207 | ||
51c6f666 | 1208 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
b5810039 | 1209 | struct vm_area_struct *vma, pud_t *pud, |
1da177e4 | 1210 | unsigned long addr, unsigned long end, |
97a89413 | 1211 | struct zap_details *details) |
1da177e4 LT |
1212 | { |
1213 | pmd_t *pmd; | |
1214 | unsigned long next; | |
1215 | ||
1216 | pmd = pmd_offset(pud, addr); | |
1217 | do { | |
1218 | next = pmd_addr_end(addr, end); | |
71e3aac0 | 1219 | if (pmd_trans_huge(*pmd)) { |
1a5a9906 | 1220 | if (next - addr != HPAGE_PMD_SIZE) { |
e0897d75 DR |
1221 | #ifdef CONFIG_DEBUG_VM |
1222 | if (!rwsem_is_locked(&tlb->mm->mmap_sem)) { | |
1223 | pr_err("%s: mmap_sem is unlocked! addr=0x%lx end=0x%lx vma->vm_start=0x%lx vma->vm_end=0x%lx\n", | |
1224 | __func__, addr, end, | |
1225 | vma->vm_start, | |
1226 | vma->vm_end); | |
1227 | BUG(); | |
1228 | } | |
1229 | #endif | |
e180377f | 1230 | split_huge_page_pmd(vma, addr, pmd); |
f21760b1 | 1231 | } else if (zap_huge_pmd(tlb, vma, pmd, addr)) |
1a5a9906 | 1232 | goto next; |
71e3aac0 AA |
1233 | /* fall through */ |
1234 | } | |
1a5a9906 AA |
1235 | /* |
1236 | * Here there can be other concurrent MADV_DONTNEED or | |
1237 | * trans huge page faults running, and if the pmd is | |
1238 | * none or trans huge it can change under us. This is | |
1239 | * because MADV_DONTNEED holds the mmap_sem in read | |
1240 | * mode. | |
1241 | */ | |
1242 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) | |
1243 | goto next; | |
97a89413 | 1244 | next = zap_pte_range(tlb, vma, pmd, addr, next, details); |
1a5a9906 | 1245 | next: |
97a89413 PZ |
1246 | cond_resched(); |
1247 | } while (pmd++, addr = next, addr != end); | |
51c6f666 RH |
1248 | |
1249 | return addr; | |
1da177e4 LT |
1250 | } |
1251 | ||
51c6f666 | 1252 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
b5810039 | 1253 | struct vm_area_struct *vma, pgd_t *pgd, |
1da177e4 | 1254 | unsigned long addr, unsigned long end, |
97a89413 | 1255 | struct zap_details *details) |
1da177e4 LT |
1256 | { |
1257 | pud_t *pud; | |
1258 | unsigned long next; | |
1259 | ||
1260 | pud = pud_offset(pgd, addr); | |
1261 | do { | |
1262 | next = pud_addr_end(addr, end); | |
97a89413 | 1263 | if (pud_none_or_clear_bad(pud)) |
1da177e4 | 1264 | continue; |
97a89413 PZ |
1265 | next = zap_pmd_range(tlb, vma, pud, addr, next, details); |
1266 | } while (pud++, addr = next, addr != end); | |
51c6f666 RH |
1267 | |
1268 | return addr; | |
1da177e4 LT |
1269 | } |
1270 | ||
038c7aa1 AV |
1271 | static void unmap_page_range(struct mmu_gather *tlb, |
1272 | struct vm_area_struct *vma, | |
1273 | unsigned long addr, unsigned long end, | |
1274 | struct zap_details *details) | |
1da177e4 LT |
1275 | { |
1276 | pgd_t *pgd; | |
1277 | unsigned long next; | |
1278 | ||
1279 | if (details && !details->check_mapping && !details->nonlinear_vma) | |
1280 | details = NULL; | |
1281 | ||
1282 | BUG_ON(addr >= end); | |
569b846d | 1283 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
1284 | tlb_start_vma(tlb, vma); |
1285 | pgd = pgd_offset(vma->vm_mm, addr); | |
1286 | do { | |
1287 | next = pgd_addr_end(addr, end); | |
97a89413 | 1288 | if (pgd_none_or_clear_bad(pgd)) |
1da177e4 | 1289 | continue; |
97a89413 PZ |
1290 | next = zap_pud_range(tlb, vma, pgd, addr, next, details); |
1291 | } while (pgd++, addr = next, addr != end); | |
1da177e4 | 1292 | tlb_end_vma(tlb, vma); |
569b846d | 1293 | mem_cgroup_uncharge_end(); |
1da177e4 | 1294 | } |
51c6f666 | 1295 | |
f5cc4eef AV |
1296 | |
1297 | static void unmap_single_vma(struct mmu_gather *tlb, | |
1298 | struct vm_area_struct *vma, unsigned long start_addr, | |
4f74d2c8 | 1299 | unsigned long end_addr, |
f5cc4eef AV |
1300 | struct zap_details *details) |
1301 | { | |
1302 | unsigned long start = max(vma->vm_start, start_addr); | |
1303 | unsigned long end; | |
1304 | ||
1305 | if (start >= vma->vm_end) | |
1306 | return; | |
1307 | end = min(vma->vm_end, end_addr); | |
1308 | if (end <= vma->vm_start) | |
1309 | return; | |
1310 | ||
cbc91f71 SD |
1311 | if (vma->vm_file) |
1312 | uprobe_munmap(vma, start, end); | |
1313 | ||
b3b9c293 | 1314 | if (unlikely(vma->vm_flags & VM_PFNMAP)) |
5180da41 | 1315 | untrack_pfn(vma, 0, 0); |
f5cc4eef AV |
1316 | |
1317 | if (start != end) { | |
1318 | if (unlikely(is_vm_hugetlb_page(vma))) { | |
1319 | /* | |
1320 | * It is undesirable to test vma->vm_file as it | |
1321 | * should be non-null for valid hugetlb area. | |
1322 | * However, vm_file will be NULL in the error | |
7aa6b4ad | 1323 | * cleanup path of mmap_region. When |
f5cc4eef | 1324 | * hugetlbfs ->mmap method fails, |
7aa6b4ad | 1325 | * mmap_region() nullifies vma->vm_file |
f5cc4eef AV |
1326 | * before calling this function to clean up. |
1327 | * Since no pte has actually been setup, it is | |
1328 | * safe to do nothing in this case. | |
1329 | */ | |
24669e58 AK |
1330 | if (vma->vm_file) { |
1331 | mutex_lock(&vma->vm_file->f_mapping->i_mmap_mutex); | |
d833352a | 1332 | __unmap_hugepage_range_final(tlb, vma, start, end, NULL); |
24669e58 AK |
1333 | mutex_unlock(&vma->vm_file->f_mapping->i_mmap_mutex); |
1334 | } | |
f5cc4eef AV |
1335 | } else |
1336 | unmap_page_range(tlb, vma, start, end, details); | |
1337 | } | |
1da177e4 LT |
1338 | } |
1339 | ||
1da177e4 LT |
1340 | /** |
1341 | * unmap_vmas - unmap a range of memory covered by a list of vma's | |
0164f69d | 1342 | * @tlb: address of the caller's struct mmu_gather |
1da177e4 LT |
1343 | * @vma: the starting vma |
1344 | * @start_addr: virtual address at which to start unmapping | |
1345 | * @end_addr: virtual address at which to end unmapping | |
1da177e4 | 1346 | * |
508034a3 | 1347 | * Unmap all pages in the vma list. |
1da177e4 | 1348 | * |
1da177e4 LT |
1349 | * Only addresses between `start' and `end' will be unmapped. |
1350 | * | |
1351 | * The VMA list must be sorted in ascending virtual address order. | |
1352 | * | |
1353 | * unmap_vmas() assumes that the caller will flush the whole unmapped address | |
1354 | * range after unmap_vmas() returns. So the only responsibility here is to | |
1355 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() | |
1356 | * drops the lock and schedules. | |
1357 | */ | |
6e8bb019 | 1358 | void unmap_vmas(struct mmu_gather *tlb, |
1da177e4 | 1359 | struct vm_area_struct *vma, unsigned long start_addr, |
4f74d2c8 | 1360 | unsigned long end_addr) |
1da177e4 | 1361 | { |
cddb8a5c | 1362 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 1363 | |
cddb8a5c | 1364 | mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); |
f5cc4eef | 1365 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) |
4f74d2c8 | 1366 | unmap_single_vma(tlb, vma, start_addr, end_addr, NULL); |
cddb8a5c | 1367 | mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); |
1da177e4 LT |
1368 | } |
1369 | ||
1370 | /** | |
1371 | * zap_page_range - remove user pages in a given range | |
1372 | * @vma: vm_area_struct holding the applicable pages | |
eb4546bb | 1373 | * @start: starting address of pages to zap |
1da177e4 LT |
1374 | * @size: number of bytes to zap |
1375 | * @details: details of nonlinear truncation or shared cache invalidation | |
f5cc4eef AV |
1376 | * |
1377 | * Caller must protect the VMA list | |
1da177e4 | 1378 | */ |
7e027b14 | 1379 | void zap_page_range(struct vm_area_struct *vma, unsigned long start, |
1da177e4 LT |
1380 | unsigned long size, struct zap_details *details) |
1381 | { | |
1382 | struct mm_struct *mm = vma->vm_mm; | |
d16dfc55 | 1383 | struct mmu_gather tlb; |
7e027b14 | 1384 | unsigned long end = start + size; |
1da177e4 | 1385 | |
1da177e4 | 1386 | lru_add_drain(); |
2b047252 | 1387 | tlb_gather_mmu(&tlb, mm, start, end); |
365e9c87 | 1388 | update_hiwater_rss(mm); |
7e027b14 LT |
1389 | mmu_notifier_invalidate_range_start(mm, start, end); |
1390 | for ( ; vma && vma->vm_start < end; vma = vma->vm_next) | |
4f74d2c8 | 1391 | unmap_single_vma(&tlb, vma, start, end, details); |
7e027b14 LT |
1392 | mmu_notifier_invalidate_range_end(mm, start, end); |
1393 | tlb_finish_mmu(&tlb, start, end); | |
1da177e4 LT |
1394 | } |
1395 | ||
f5cc4eef AV |
1396 | /** |
1397 | * zap_page_range_single - remove user pages in a given range | |
1398 | * @vma: vm_area_struct holding the applicable pages | |
1399 | * @address: starting address of pages to zap | |
1400 | * @size: number of bytes to zap | |
1401 | * @details: details of nonlinear truncation or shared cache invalidation | |
1402 | * | |
1403 | * The range must fit into one VMA. | |
1da177e4 | 1404 | */ |
f5cc4eef | 1405 | static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, |
1da177e4 LT |
1406 | unsigned long size, struct zap_details *details) |
1407 | { | |
1408 | struct mm_struct *mm = vma->vm_mm; | |
d16dfc55 | 1409 | struct mmu_gather tlb; |
1da177e4 | 1410 | unsigned long end = address + size; |
1da177e4 | 1411 | |
1da177e4 | 1412 | lru_add_drain(); |
2b047252 | 1413 | tlb_gather_mmu(&tlb, mm, address, end); |
365e9c87 | 1414 | update_hiwater_rss(mm); |
f5cc4eef | 1415 | mmu_notifier_invalidate_range_start(mm, address, end); |
4f74d2c8 | 1416 | unmap_single_vma(&tlb, vma, address, end, details); |
f5cc4eef | 1417 | mmu_notifier_invalidate_range_end(mm, address, end); |
d16dfc55 | 1418 | tlb_finish_mmu(&tlb, address, end); |
1da177e4 LT |
1419 | } |
1420 | ||
c627f9cc JS |
1421 | /** |
1422 | * zap_vma_ptes - remove ptes mapping the vma | |
1423 | * @vma: vm_area_struct holding ptes to be zapped | |
1424 | * @address: starting address of pages to zap | |
1425 | * @size: number of bytes to zap | |
1426 | * | |
1427 | * This function only unmaps ptes assigned to VM_PFNMAP vmas. | |
1428 | * | |
1429 | * The entire address range must be fully contained within the vma. | |
1430 | * | |
1431 | * Returns 0 if successful. | |
1432 | */ | |
1433 | int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, | |
1434 | unsigned long size) | |
1435 | { | |
1436 | if (address < vma->vm_start || address + size > vma->vm_end || | |
1437 | !(vma->vm_flags & VM_PFNMAP)) | |
1438 | return -1; | |
f5cc4eef | 1439 | zap_page_range_single(vma, address, size, NULL); |
c627f9cc JS |
1440 | return 0; |
1441 | } | |
1442 | EXPORT_SYMBOL_GPL(zap_vma_ptes); | |
1443 | ||
142762bd | 1444 | /** |
240aadee | 1445 | * follow_page_mask - look up a page descriptor from a user-virtual address |
142762bd JW |
1446 | * @vma: vm_area_struct mapping @address |
1447 | * @address: virtual address to look up | |
1448 | * @flags: flags modifying lookup behaviour | |
240aadee | 1449 | * @page_mask: on output, *page_mask is set according to the size of the page |
142762bd JW |
1450 | * |
1451 | * @flags can have FOLL_ flags set, defined in <linux/mm.h> | |
1452 | * | |
1453 | * Returns the mapped (struct page *), %NULL if no mapping exists, or | |
1454 | * an error pointer if there is a mapping to something not represented | |
1455 | * by a page descriptor (see also vm_normal_page()). | |
1da177e4 | 1456 | */ |
240aadee ML |
1457 | struct page *follow_page_mask(struct vm_area_struct *vma, |
1458 | unsigned long address, unsigned int flags, | |
1459 | unsigned int *page_mask) | |
1da177e4 LT |
1460 | { |
1461 | pgd_t *pgd; | |
1462 | pud_t *pud; | |
1463 | pmd_t *pmd; | |
1464 | pte_t *ptep, pte; | |
deceb6cd | 1465 | spinlock_t *ptl; |
1da177e4 | 1466 | struct page *page; |
6aab341e | 1467 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 1468 | |
240aadee ML |
1469 | *page_mask = 0; |
1470 | ||
deceb6cd HD |
1471 | page = follow_huge_addr(mm, address, flags & FOLL_WRITE); |
1472 | if (!IS_ERR(page)) { | |
1473 | BUG_ON(flags & FOLL_GET); | |
1474 | goto out; | |
1475 | } | |
1da177e4 | 1476 | |
deceb6cd | 1477 | page = NULL; |
1da177e4 LT |
1478 | pgd = pgd_offset(mm, address); |
1479 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
deceb6cd | 1480 | goto no_page_table; |
1da177e4 LT |
1481 | |
1482 | pud = pud_offset(pgd, address); | |
ceb86879 | 1483 | if (pud_none(*pud)) |
deceb6cd | 1484 | goto no_page_table; |
8a07651e | 1485 | if (pud_huge(*pud) && vma->vm_flags & VM_HUGETLB) { |
e632a938 NH |
1486 | if (flags & FOLL_GET) |
1487 | goto out; | |
ceb86879 AK |
1488 | page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE); |
1489 | goto out; | |
1490 | } | |
1491 | if (unlikely(pud_bad(*pud))) | |
1492 | goto no_page_table; | |
1493 | ||
1da177e4 | 1494 | pmd = pmd_offset(pud, address); |
aeed5fce | 1495 | if (pmd_none(*pmd)) |
deceb6cd | 1496 | goto no_page_table; |
71e3aac0 | 1497 | if (pmd_huge(*pmd) && vma->vm_flags & VM_HUGETLB) { |
deceb6cd | 1498 | page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE); |
e632a938 NH |
1499 | if (flags & FOLL_GET) { |
1500 | /* | |
1501 | * Refcount on tail pages are not well-defined and | |
1502 | * shouldn't be taken. The caller should handle a NULL | |
1503 | * return when trying to follow tail pages. | |
1504 | */ | |
1505 | if (PageHead(page)) | |
1506 | get_page(page); | |
1507 | else { | |
1508 | page = NULL; | |
1509 | goto out; | |
1510 | } | |
1511 | } | |
1da177e4 | 1512 | goto out; |
deceb6cd | 1513 | } |
0b9d7052 AA |
1514 | if ((flags & FOLL_NUMA) && pmd_numa(*pmd)) |
1515 | goto no_page_table; | |
71e3aac0 | 1516 | if (pmd_trans_huge(*pmd)) { |
500d65d4 | 1517 | if (flags & FOLL_SPLIT) { |
e180377f | 1518 | split_huge_page_pmd(vma, address, pmd); |
500d65d4 AA |
1519 | goto split_fallthrough; |
1520 | } | |
c4088ebd | 1521 | ptl = pmd_lock(mm, pmd); |
71e3aac0 AA |
1522 | if (likely(pmd_trans_huge(*pmd))) { |
1523 | if (unlikely(pmd_trans_splitting(*pmd))) { | |
c4088ebd | 1524 | spin_unlock(ptl); |
71e3aac0 AA |
1525 | wait_split_huge_page(vma->anon_vma, pmd); |
1526 | } else { | |
b676b293 | 1527 | page = follow_trans_huge_pmd(vma, address, |
71e3aac0 | 1528 | pmd, flags); |
c4088ebd | 1529 | spin_unlock(ptl); |
240aadee | 1530 | *page_mask = HPAGE_PMD_NR - 1; |
71e3aac0 AA |
1531 | goto out; |
1532 | } | |
1533 | } else | |
c4088ebd | 1534 | spin_unlock(ptl); |
71e3aac0 AA |
1535 | /* fall through */ |
1536 | } | |
500d65d4 | 1537 | split_fallthrough: |
aeed5fce HD |
1538 | if (unlikely(pmd_bad(*pmd))) |
1539 | goto no_page_table; | |
1540 | ||
deceb6cd | 1541 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); |
1da177e4 LT |
1542 | |
1543 | pte = *ptep; | |
5117b3b8 HD |
1544 | if (!pte_present(pte)) { |
1545 | swp_entry_t entry; | |
1546 | /* | |
1547 | * KSM's break_ksm() relies upon recognizing a ksm page | |
1548 | * even while it is being migrated, so for that case we | |
1549 | * need migration_entry_wait(). | |
1550 | */ | |
1551 | if (likely(!(flags & FOLL_MIGRATION))) | |
1552 | goto no_page; | |
1553 | if (pte_none(pte) || pte_file(pte)) | |
1554 | goto no_page; | |
1555 | entry = pte_to_swp_entry(pte); | |
1556 | if (!is_migration_entry(entry)) | |
1557 | goto no_page; | |
1558 | pte_unmap_unlock(ptep, ptl); | |
1559 | migration_entry_wait(mm, pmd, address); | |
1560 | goto split_fallthrough; | |
1561 | } | |
0b9d7052 AA |
1562 | if ((flags & FOLL_NUMA) && pte_numa(pte)) |
1563 | goto no_page; | |
deceb6cd HD |
1564 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
1565 | goto unlock; | |
a13ea5b7 | 1566 | |
6aab341e | 1567 | page = vm_normal_page(vma, address, pte); |
a13ea5b7 HD |
1568 | if (unlikely(!page)) { |
1569 | if ((flags & FOLL_DUMP) || | |
62eede62 | 1570 | !is_zero_pfn(pte_pfn(pte))) |
a13ea5b7 HD |
1571 | goto bad_page; |
1572 | page = pte_page(pte); | |
1573 | } | |
1da177e4 | 1574 | |
deceb6cd | 1575 | if (flags & FOLL_GET) |
70b50f94 | 1576 | get_page_foll(page); |
deceb6cd HD |
1577 | if (flags & FOLL_TOUCH) { |
1578 | if ((flags & FOLL_WRITE) && | |
1579 | !pte_dirty(pte) && !PageDirty(page)) | |
1580 | set_page_dirty(page); | |
bd775c42 KM |
1581 | /* |
1582 | * pte_mkyoung() would be more correct here, but atomic care | |
1583 | * is needed to avoid losing the dirty bit: it is easier to use | |
1584 | * mark_page_accessed(). | |
1585 | */ | |
deceb6cd HD |
1586 | mark_page_accessed(page); |
1587 | } | |
a1fde08c | 1588 | if ((flags & FOLL_MLOCK) && (vma->vm_flags & VM_LOCKED)) { |
110d74a9 ML |
1589 | /* |
1590 | * The preliminary mapping check is mainly to avoid the | |
1591 | * pointless overhead of lock_page on the ZERO_PAGE | |
1592 | * which might bounce very badly if there is contention. | |
1593 | * | |
1594 | * If the page is already locked, we don't need to | |
1595 | * handle it now - vmscan will handle it later if and | |
1596 | * when it attempts to reclaim the page. | |
1597 | */ | |
1598 | if (page->mapping && trylock_page(page)) { | |
1599 | lru_add_drain(); /* push cached pages to LRU */ | |
1600 | /* | |
e6c509f8 HD |
1601 | * Because we lock page here, and migration is |
1602 | * blocked by the pte's page reference, and we | |
1603 | * know the page is still mapped, we don't even | |
1604 | * need to check for file-cache page truncation. | |
110d74a9 | 1605 | */ |
e6c509f8 | 1606 | mlock_vma_page(page); |
110d74a9 ML |
1607 | unlock_page(page); |
1608 | } | |
1609 | } | |
deceb6cd HD |
1610 | unlock: |
1611 | pte_unmap_unlock(ptep, ptl); | |
1da177e4 | 1612 | out: |
deceb6cd | 1613 | return page; |
1da177e4 | 1614 | |
89f5b7da LT |
1615 | bad_page: |
1616 | pte_unmap_unlock(ptep, ptl); | |
1617 | return ERR_PTR(-EFAULT); | |
1618 | ||
1619 | no_page: | |
1620 | pte_unmap_unlock(ptep, ptl); | |
1621 | if (!pte_none(pte)) | |
1622 | return page; | |
8e4b9a60 | 1623 | |
deceb6cd HD |
1624 | no_page_table: |
1625 | /* | |
1626 | * When core dumping an enormous anonymous area that nobody | |
8e4b9a60 HD |
1627 | * has touched so far, we don't want to allocate unnecessary pages or |
1628 | * page tables. Return error instead of NULL to skip handle_mm_fault, | |
1629 | * then get_dump_page() will return NULL to leave a hole in the dump. | |
1630 | * But we can only make this optimization where a hole would surely | |
1631 | * be zero-filled if handle_mm_fault() actually did handle it. | |
deceb6cd | 1632 | */ |
8e4b9a60 HD |
1633 | if ((flags & FOLL_DUMP) && |
1634 | (!vma->vm_ops || !vma->vm_ops->fault)) | |
1635 | return ERR_PTR(-EFAULT); | |
deceb6cd | 1636 | return page; |
1da177e4 LT |
1637 | } |
1638 | ||
95042f9e LT |
1639 | static inline int stack_guard_page(struct vm_area_struct *vma, unsigned long addr) |
1640 | { | |
a09a79f6 MP |
1641 | return stack_guard_page_start(vma, addr) || |
1642 | stack_guard_page_end(vma, addr+PAGE_SIZE); | |
95042f9e LT |
1643 | } |
1644 | ||
0014bd99 HY |
1645 | /** |
1646 | * __get_user_pages() - pin user pages in memory | |
1647 | * @tsk: task_struct of target task | |
1648 | * @mm: mm_struct of target mm | |
1649 | * @start: starting user address | |
1650 | * @nr_pages: number of pages from start to pin | |
1651 | * @gup_flags: flags modifying pin behaviour | |
1652 | * @pages: array that receives pointers to the pages pinned. | |
1653 | * Should be at least nr_pages long. Or NULL, if caller | |
1654 | * only intends to ensure the pages are faulted in. | |
1655 | * @vmas: array of pointers to vmas corresponding to each page. | |
1656 | * Or NULL if the caller does not require them. | |
1657 | * @nonblocking: whether waiting for disk IO or mmap_sem contention | |
1658 | * | |
1659 | * Returns number of pages pinned. This may be fewer than the number | |
1660 | * requested. If nr_pages is 0 or negative, returns 0. If no pages | |
1661 | * were pinned, returns -errno. Each page returned must be released | |
1662 | * with a put_page() call when it is finished with. vmas will only | |
1663 | * remain valid while mmap_sem is held. | |
1664 | * | |
1665 | * Must be called with mmap_sem held for read or write. | |
1666 | * | |
1667 | * __get_user_pages walks a process's page tables and takes a reference to | |
1668 | * each struct page that each user address corresponds to at a given | |
1669 | * instant. That is, it takes the page that would be accessed if a user | |
1670 | * thread accesses the given user virtual address at that instant. | |
1671 | * | |
1672 | * This does not guarantee that the page exists in the user mappings when | |
1673 | * __get_user_pages returns, and there may even be a completely different | |
1674 | * page there in some cases (eg. if mmapped pagecache has been invalidated | |
1675 | * and subsequently re faulted). However it does guarantee that the page | |
1676 | * won't be freed completely. And mostly callers simply care that the page | |
1677 | * contains data that was valid *at some point in time*. Typically, an IO | |
1678 | * or similar operation cannot guarantee anything stronger anyway because | |
1679 | * locks can't be held over the syscall boundary. | |
1680 | * | |
1681 | * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If | |
1682 | * the page is written to, set_page_dirty (or set_page_dirty_lock, as | |
1683 | * appropriate) must be called after the page is finished with, and | |
1684 | * before put_page is called. | |
1685 | * | |
1686 | * If @nonblocking != NULL, __get_user_pages will not wait for disk IO | |
1687 | * or mmap_sem contention, and if waiting is needed to pin all pages, | |
1688 | * *@nonblocking will be set to 0. | |
1689 | * | |
1690 | * In most cases, get_user_pages or get_user_pages_fast should be used | |
1691 | * instead of __get_user_pages. __get_user_pages should be used only if | |
1692 | * you need some special @gup_flags. | |
1693 | */ | |
28a35716 ML |
1694 | long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
1695 | unsigned long start, unsigned long nr_pages, | |
1696 | unsigned int gup_flags, struct page **pages, | |
1697 | struct vm_area_struct **vmas, int *nonblocking) | |
1da177e4 | 1698 | { |
28a35716 | 1699 | long i; |
58fa879e | 1700 | unsigned long vm_flags; |
240aadee | 1701 | unsigned int page_mask; |
1da177e4 | 1702 | |
28a35716 | 1703 | if (!nr_pages) |
900cf086 | 1704 | return 0; |
58fa879e HD |
1705 | |
1706 | VM_BUG_ON(!!pages != !!(gup_flags & FOLL_GET)); | |
1707 | ||
0b9d7052 AA |
1708 | /* |
1709 | * If FOLL_FORCE and FOLL_NUMA are both set, handle_mm_fault | |
1710 | * would be called on PROT_NONE ranges. We must never invoke | |
1711 | * handle_mm_fault on PROT_NONE ranges or the NUMA hinting | |
1712 | * page faults would unprotect the PROT_NONE ranges if | |
1713 | * _PAGE_NUMA and _PAGE_PROTNONE are sharing the same pte/pmd | |
1714 | * bitflag. So to avoid that, don't set FOLL_NUMA if | |
1715 | * FOLL_FORCE is set. | |
1716 | */ | |
1717 | if (!(gup_flags & FOLL_FORCE)) | |
1718 | gup_flags |= FOLL_NUMA; | |
1719 | ||
1da177e4 LT |
1720 | i = 0; |
1721 | ||
1722 | do { | |
deceb6cd | 1723 | struct vm_area_struct *vma; |
1da177e4 LT |
1724 | |
1725 | vma = find_extend_vma(mm, start); | |
e7f22e20 | 1726 | if (!vma && in_gate_area(mm, start)) { |
1da177e4 | 1727 | unsigned long pg = start & PAGE_MASK; |
1da177e4 LT |
1728 | pgd_t *pgd; |
1729 | pud_t *pud; | |
1730 | pmd_t *pmd; | |
1731 | pte_t *pte; | |
b291f000 NP |
1732 | |
1733 | /* user gate pages are read-only */ | |
58fa879e | 1734 | if (gup_flags & FOLL_WRITE) |
cda540ac | 1735 | goto efault; |
1da177e4 LT |
1736 | if (pg > TASK_SIZE) |
1737 | pgd = pgd_offset_k(pg); | |
1738 | else | |
1739 | pgd = pgd_offset_gate(mm, pg); | |
1740 | BUG_ON(pgd_none(*pgd)); | |
1741 | pud = pud_offset(pgd, pg); | |
1742 | BUG_ON(pud_none(*pud)); | |
1743 | pmd = pmd_offset(pud, pg); | |
690dbe1c | 1744 | if (pmd_none(*pmd)) |
cda540ac | 1745 | goto efault; |
f66055ab | 1746 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
1da177e4 | 1747 | pte = pte_offset_map(pmd, pg); |
690dbe1c HD |
1748 | if (pte_none(*pte)) { |
1749 | pte_unmap(pte); | |
cda540ac | 1750 | goto efault; |
690dbe1c | 1751 | } |
95042f9e | 1752 | vma = get_gate_vma(mm); |
1da177e4 | 1753 | if (pages) { |
de51257a HD |
1754 | struct page *page; |
1755 | ||
95042f9e | 1756 | page = vm_normal_page(vma, start, *pte); |
de51257a HD |
1757 | if (!page) { |
1758 | if (!(gup_flags & FOLL_DUMP) && | |
1759 | is_zero_pfn(pte_pfn(*pte))) | |
1760 | page = pte_page(*pte); | |
1761 | else { | |
1762 | pte_unmap(pte); | |
cda540ac | 1763 | goto efault; |
de51257a HD |
1764 | } |
1765 | } | |
6aab341e | 1766 | pages[i] = page; |
de51257a | 1767 | get_page(page); |
1da177e4 LT |
1768 | } |
1769 | pte_unmap(pte); | |
240aadee | 1770 | page_mask = 0; |
95042f9e | 1771 | goto next_page; |
1da177e4 LT |
1772 | } |
1773 | ||
cda540ac HD |
1774 | if (!vma) |
1775 | goto efault; | |
1776 | vm_flags = vma->vm_flags; | |
1777 | if (vm_flags & (VM_IO | VM_PFNMAP)) | |
1778 | goto efault; | |
1779 | ||
1780 | if (gup_flags & FOLL_WRITE) { | |
1781 | if (!(vm_flags & VM_WRITE)) { | |
1782 | if (!(gup_flags & FOLL_FORCE)) | |
1783 | goto efault; | |
1784 | /* | |
1785 | * We used to let the write,force case do COW | |
1786 | * in a VM_MAYWRITE VM_SHARED !VM_WRITE vma, so | |
1787 | * ptrace could set a breakpoint in a read-only | |
1788 | * mapping of an executable, without corrupting | |
1789 | * the file (yet only when that file had been | |
1790 | * opened for writing!). Anon pages in shared | |
1791 | * mappings are surprising: now just reject it. | |
1792 | */ | |
1793 | if (!is_cow_mapping(vm_flags)) { | |
1794 | WARN_ON_ONCE(vm_flags & VM_MAYWRITE); | |
1795 | goto efault; | |
1796 | } | |
1797 | } | |
1798 | } else { | |
1799 | if (!(vm_flags & VM_READ)) { | |
1800 | if (!(gup_flags & FOLL_FORCE)) | |
1801 | goto efault; | |
1802 | /* | |
1803 | * Is there actually any vma we can reach here | |
1804 | * which does not have VM_MAYREAD set? | |
1805 | */ | |
1806 | if (!(vm_flags & VM_MAYREAD)) | |
1807 | goto efault; | |
1808 | } | |
1809 | } | |
1da177e4 | 1810 | |
2a15efc9 HD |
1811 | if (is_vm_hugetlb_page(vma)) { |
1812 | i = follow_hugetlb_page(mm, vma, pages, vmas, | |
58fa879e | 1813 | &start, &nr_pages, i, gup_flags); |
2a15efc9 HD |
1814 | continue; |
1815 | } | |
deceb6cd | 1816 | |
1da177e4 | 1817 | do { |
08ef4729 | 1818 | struct page *page; |
58fa879e | 1819 | unsigned int foll_flags = gup_flags; |
240aadee | 1820 | unsigned int page_increm; |
1da177e4 | 1821 | |
462e00cc | 1822 | /* |
4779280d | 1823 | * If we have a pending SIGKILL, don't keep faulting |
1c3aff1c | 1824 | * pages and potentially allocating memory. |
462e00cc | 1825 | */ |
1c3aff1c | 1826 | if (unlikely(fatal_signal_pending(current))) |
4779280d | 1827 | return i ? i : -ERESTARTSYS; |
462e00cc | 1828 | |
deceb6cd | 1829 | cond_resched(); |
240aadee ML |
1830 | while (!(page = follow_page_mask(vma, start, |
1831 | foll_flags, &page_mask))) { | |
deceb6cd | 1832 | int ret; |
53a7706d ML |
1833 | unsigned int fault_flags = 0; |
1834 | ||
a09a79f6 MP |
1835 | /* For mlock, just skip the stack guard page. */ |
1836 | if (foll_flags & FOLL_MLOCK) { | |
1837 | if (stack_guard_page(vma, start)) | |
1838 | goto next_page; | |
1839 | } | |
53a7706d ML |
1840 | if (foll_flags & FOLL_WRITE) |
1841 | fault_flags |= FAULT_FLAG_WRITE; | |
1842 | if (nonblocking) | |
1843 | fault_flags |= FAULT_FLAG_ALLOW_RETRY; | |
318b275f GN |
1844 | if (foll_flags & FOLL_NOWAIT) |
1845 | fault_flags |= (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT); | |
d06063cc | 1846 | |
d26ed650 | 1847 | ret = handle_mm_fault(mm, vma, start, |
53a7706d | 1848 | fault_flags); |
d26ed650 | 1849 | |
83c54070 NP |
1850 | if (ret & VM_FAULT_ERROR) { |
1851 | if (ret & VM_FAULT_OOM) | |
1852 | return i ? i : -ENOMEM; | |
69ebb83e HY |
1853 | if (ret & (VM_FAULT_HWPOISON | |
1854 | VM_FAULT_HWPOISON_LARGE)) { | |
1855 | if (i) | |
1856 | return i; | |
1857 | else if (gup_flags & FOLL_HWPOISON) | |
1858 | return -EHWPOISON; | |
1859 | else | |
1860 | return -EFAULT; | |
1861 | } | |
1862 | if (ret & VM_FAULT_SIGBUS) | |
cda540ac | 1863 | goto efault; |
83c54070 NP |
1864 | BUG(); |
1865 | } | |
e7f22e20 SW |
1866 | |
1867 | if (tsk) { | |
1868 | if (ret & VM_FAULT_MAJOR) | |
1869 | tsk->maj_flt++; | |
1870 | else | |
1871 | tsk->min_flt++; | |
1872 | } | |
83c54070 | 1873 | |
53a7706d | 1874 | if (ret & VM_FAULT_RETRY) { |
318b275f GN |
1875 | if (nonblocking) |
1876 | *nonblocking = 0; | |
53a7706d ML |
1877 | return i; |
1878 | } | |
1879 | ||
a68d2ebc | 1880 | /* |
83c54070 NP |
1881 | * The VM_FAULT_WRITE bit tells us that |
1882 | * do_wp_page has broken COW when necessary, | |
1883 | * even if maybe_mkwrite decided not to set | |
1884 | * pte_write. We can thus safely do subsequent | |
878b63ac HD |
1885 | * page lookups as if they were reads. But only |
1886 | * do so when looping for pte_write is futile: | |
1887 | * in some cases userspace may also be wanting | |
1888 | * to write to the gotten user page, which a | |
1889 | * read fault here might prevent (a readonly | |
1890 | * page might get reCOWed by userspace write). | |
a68d2ebc | 1891 | */ |
878b63ac HD |
1892 | if ((ret & VM_FAULT_WRITE) && |
1893 | !(vma->vm_flags & VM_WRITE)) | |
deceb6cd | 1894 | foll_flags &= ~FOLL_WRITE; |
83c54070 | 1895 | |
7f7bbbe5 | 1896 | cond_resched(); |
1da177e4 | 1897 | } |
89f5b7da LT |
1898 | if (IS_ERR(page)) |
1899 | return i ? i : PTR_ERR(page); | |
1da177e4 | 1900 | if (pages) { |
08ef4729 | 1901 | pages[i] = page; |
03beb076 | 1902 | |
a6f36be3 | 1903 | flush_anon_page(vma, page, start); |
08ef4729 | 1904 | flush_dcache_page(page); |
240aadee | 1905 | page_mask = 0; |
1da177e4 | 1906 | } |
95042f9e | 1907 | next_page: |
240aadee | 1908 | if (vmas) { |
1da177e4 | 1909 | vmas[i] = vma; |
240aadee ML |
1910 | page_mask = 0; |
1911 | } | |
1912 | page_increm = 1 + (~(start >> PAGE_SHIFT) & page_mask); | |
1913 | if (page_increm > nr_pages) | |
1914 | page_increm = nr_pages; | |
1915 | i += page_increm; | |
1916 | start += page_increm * PAGE_SIZE; | |
1917 | nr_pages -= page_increm; | |
9d73777e PZ |
1918 | } while (nr_pages && start < vma->vm_end); |
1919 | } while (nr_pages); | |
1da177e4 | 1920 | return i; |
cda540ac HD |
1921 | efault: |
1922 | return i ? : -EFAULT; | |
1da177e4 | 1923 | } |
0014bd99 | 1924 | EXPORT_SYMBOL(__get_user_pages); |
b291f000 | 1925 | |
2efaca92 BH |
1926 | /* |
1927 | * fixup_user_fault() - manually resolve a user page fault | |
1928 | * @tsk: the task_struct to use for page fault accounting, or | |
1929 | * NULL if faults are not to be recorded. | |
1930 | * @mm: mm_struct of target mm | |
1931 | * @address: user address | |
1932 | * @fault_flags:flags to pass down to handle_mm_fault() | |
1933 | * | |
1934 | * This is meant to be called in the specific scenario where for locking reasons | |
1935 | * we try to access user memory in atomic context (within a pagefault_disable() | |
1936 | * section), this returns -EFAULT, and we want to resolve the user fault before | |
1937 | * trying again. | |
1938 | * | |
1939 | * Typically this is meant to be used by the futex code. | |
1940 | * | |
1941 | * The main difference with get_user_pages() is that this function will | |
1942 | * unconditionally call handle_mm_fault() which will in turn perform all the | |
1943 | * necessary SW fixup of the dirty and young bits in the PTE, while | |
1944 | * handle_mm_fault() only guarantees to update these in the struct page. | |
1945 | * | |
1946 | * This is important for some architectures where those bits also gate the | |
1947 | * access permission to the page because they are maintained in software. On | |
1948 | * such architectures, gup() will not be enough to make a subsequent access | |
1949 | * succeed. | |
1950 | * | |
1951 | * This should be called with the mm_sem held for read. | |
1952 | */ | |
1953 | int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm, | |
1954 | unsigned long address, unsigned int fault_flags) | |
1955 | { | |
1956 | struct vm_area_struct *vma; | |
1957 | int ret; | |
1958 | ||
1959 | vma = find_extend_vma(mm, address); | |
1960 | if (!vma || address < vma->vm_start) | |
1961 | return -EFAULT; | |
1962 | ||
1963 | ret = handle_mm_fault(mm, vma, address, fault_flags); | |
1964 | if (ret & VM_FAULT_ERROR) { | |
1965 | if (ret & VM_FAULT_OOM) | |
1966 | return -ENOMEM; | |
1967 | if (ret & (VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE)) | |
1968 | return -EHWPOISON; | |
1969 | if (ret & VM_FAULT_SIGBUS) | |
1970 | return -EFAULT; | |
1971 | BUG(); | |
1972 | } | |
1973 | if (tsk) { | |
1974 | if (ret & VM_FAULT_MAJOR) | |
1975 | tsk->maj_flt++; | |
1976 | else | |
1977 | tsk->min_flt++; | |
1978 | } | |
1979 | return 0; | |
1980 | } | |
1981 | ||
1982 | /* | |
d2bf6be8 | 1983 | * get_user_pages() - pin user pages in memory |
e7f22e20 SW |
1984 | * @tsk: the task_struct to use for page fault accounting, or |
1985 | * NULL if faults are not to be recorded. | |
d2bf6be8 NP |
1986 | * @mm: mm_struct of target mm |
1987 | * @start: starting user address | |
9d73777e | 1988 | * @nr_pages: number of pages from start to pin |
d2bf6be8 | 1989 | * @write: whether pages will be written to by the caller |
cda540ac HD |
1990 | * @force: whether to force access even when user mapping is currently |
1991 | * protected (but never forces write access to shared mapping). | |
d2bf6be8 NP |
1992 | * @pages: array that receives pointers to the pages pinned. |
1993 | * Should be at least nr_pages long. Or NULL, if caller | |
1994 | * only intends to ensure the pages are faulted in. | |
1995 | * @vmas: array of pointers to vmas corresponding to each page. | |
1996 | * Or NULL if the caller does not require them. | |
1997 | * | |
1998 | * Returns number of pages pinned. This may be fewer than the number | |
9d73777e | 1999 | * requested. If nr_pages is 0 or negative, returns 0. If no pages |
d2bf6be8 NP |
2000 | * were pinned, returns -errno. Each page returned must be released |
2001 | * with a put_page() call when it is finished with. vmas will only | |
2002 | * remain valid while mmap_sem is held. | |
2003 | * | |
2004 | * Must be called with mmap_sem held for read or write. | |
2005 | * | |
2006 | * get_user_pages walks a process's page tables and takes a reference to | |
2007 | * each struct page that each user address corresponds to at a given | |
2008 | * instant. That is, it takes the page that would be accessed if a user | |
2009 | * thread accesses the given user virtual address at that instant. | |
2010 | * | |
2011 | * This does not guarantee that the page exists in the user mappings when | |
2012 | * get_user_pages returns, and there may even be a completely different | |
2013 | * page there in some cases (eg. if mmapped pagecache has been invalidated | |
2014 | * and subsequently re faulted). However it does guarantee that the page | |
2015 | * won't be freed completely. And mostly callers simply care that the page | |
2016 | * contains data that was valid *at some point in time*. Typically, an IO | |
2017 | * or similar operation cannot guarantee anything stronger anyway because | |
2018 | * locks can't be held over the syscall boundary. | |
2019 | * | |
2020 | * If write=0, the page must not be written to. If the page is written to, | |
2021 | * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called | |
2022 | * after the page is finished with, and before put_page is called. | |
2023 | * | |
2024 | * get_user_pages is typically used for fewer-copy IO operations, to get a | |
2025 | * handle on the memory by some means other than accesses via the user virtual | |
2026 | * addresses. The pages may be submitted for DMA to devices or accessed via | |
2027 | * their kernel linear mapping (via the kmap APIs). Care should be taken to | |
2028 | * use the correct cache flushing APIs. | |
2029 | * | |
2030 | * See also get_user_pages_fast, for performance critical applications. | |
2031 | */ | |
28a35716 ML |
2032 | long get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
2033 | unsigned long start, unsigned long nr_pages, int write, | |
2034 | int force, struct page **pages, struct vm_area_struct **vmas) | |
b291f000 | 2035 | { |
58fa879e | 2036 | int flags = FOLL_TOUCH; |
b291f000 | 2037 | |
58fa879e HD |
2038 | if (pages) |
2039 | flags |= FOLL_GET; | |
b291f000 | 2040 | if (write) |
58fa879e | 2041 | flags |= FOLL_WRITE; |
b291f000 | 2042 | if (force) |
58fa879e | 2043 | flags |= FOLL_FORCE; |
b291f000 | 2044 | |
53a7706d ML |
2045 | return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas, |
2046 | NULL); | |
b291f000 | 2047 | } |
1da177e4 LT |
2048 | EXPORT_SYMBOL(get_user_pages); |
2049 | ||
f3e8fccd HD |
2050 | /** |
2051 | * get_dump_page() - pin user page in memory while writing it to core dump | |
2052 | * @addr: user address | |
2053 | * | |
2054 | * Returns struct page pointer of user page pinned for dump, | |
2055 | * to be freed afterwards by page_cache_release() or put_page(). | |
2056 | * | |
2057 | * Returns NULL on any kind of failure - a hole must then be inserted into | |
2058 | * the corefile, to preserve alignment with its headers; and also returns | |
2059 | * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found - | |
2060 | * allowing a hole to be left in the corefile to save diskspace. | |
2061 | * | |
2062 | * Called without mmap_sem, but after all other threads have been killed. | |
2063 | */ | |
2064 | #ifdef CONFIG_ELF_CORE | |
2065 | struct page *get_dump_page(unsigned long addr) | |
2066 | { | |
2067 | struct vm_area_struct *vma; | |
2068 | struct page *page; | |
2069 | ||
2070 | if (__get_user_pages(current, current->mm, addr, 1, | |
53a7706d ML |
2071 | FOLL_FORCE | FOLL_DUMP | FOLL_GET, &page, &vma, |
2072 | NULL) < 1) | |
f3e8fccd | 2073 | return NULL; |
f3e8fccd HD |
2074 | flush_cache_page(vma, addr, page_to_pfn(page)); |
2075 | return page; | |
2076 | } | |
2077 | #endif /* CONFIG_ELF_CORE */ | |
2078 | ||
25ca1d6c | 2079 | pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, |
920c7a5d | 2080 | spinlock_t **ptl) |
c9cfcddf LT |
2081 | { |
2082 | pgd_t * pgd = pgd_offset(mm, addr); | |
2083 | pud_t * pud = pud_alloc(mm, pgd, addr); | |
2084 | if (pud) { | |
49c91fb0 | 2085 | pmd_t * pmd = pmd_alloc(mm, pud, addr); |
f66055ab AA |
2086 | if (pmd) { |
2087 | VM_BUG_ON(pmd_trans_huge(*pmd)); | |
c9cfcddf | 2088 | return pte_alloc_map_lock(mm, pmd, addr, ptl); |
f66055ab | 2089 | } |
c9cfcddf LT |
2090 | } |
2091 | return NULL; | |
2092 | } | |
2093 | ||
238f58d8 LT |
2094 | /* |
2095 | * This is the old fallback for page remapping. | |
2096 | * | |
2097 | * For historical reasons, it only allows reserved pages. Only | |
2098 | * old drivers should use this, and they needed to mark their | |
2099 | * pages reserved for the old functions anyway. | |
2100 | */ | |
423bad60 NP |
2101 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
2102 | struct page *page, pgprot_t prot) | |
238f58d8 | 2103 | { |
423bad60 | 2104 | struct mm_struct *mm = vma->vm_mm; |
238f58d8 | 2105 | int retval; |
c9cfcddf | 2106 | pte_t *pte; |
8a9f3ccd BS |
2107 | spinlock_t *ptl; |
2108 | ||
238f58d8 | 2109 | retval = -EINVAL; |
a145dd41 | 2110 | if (PageAnon(page)) |
5b4e655e | 2111 | goto out; |
238f58d8 LT |
2112 | retval = -ENOMEM; |
2113 | flush_dcache_page(page); | |
c9cfcddf | 2114 | pte = get_locked_pte(mm, addr, &ptl); |
238f58d8 | 2115 | if (!pte) |
5b4e655e | 2116 | goto out; |
238f58d8 LT |
2117 | retval = -EBUSY; |
2118 | if (!pte_none(*pte)) | |
2119 | goto out_unlock; | |
2120 | ||
2121 | /* Ok, finally just insert the thing.. */ | |
2122 | get_page(page); | |
34e55232 | 2123 | inc_mm_counter_fast(mm, MM_FILEPAGES); |
238f58d8 LT |
2124 | page_add_file_rmap(page); |
2125 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); | |
2126 | ||
2127 | retval = 0; | |
8a9f3ccd BS |
2128 | pte_unmap_unlock(pte, ptl); |
2129 | return retval; | |
238f58d8 LT |
2130 | out_unlock: |
2131 | pte_unmap_unlock(pte, ptl); | |
2132 | out: | |
2133 | return retval; | |
2134 | } | |
2135 | ||
bfa5bf6d REB |
2136 | /** |
2137 | * vm_insert_page - insert single page into user vma | |
2138 | * @vma: user vma to map to | |
2139 | * @addr: target user address of this page | |
2140 | * @page: source kernel page | |
2141 | * | |
a145dd41 LT |
2142 | * This allows drivers to insert individual pages they've allocated |
2143 | * into a user vma. | |
2144 | * | |
2145 | * The page has to be a nice clean _individual_ kernel allocation. | |
2146 | * If you allocate a compound page, you need to have marked it as | |
2147 | * such (__GFP_COMP), or manually just split the page up yourself | |
8dfcc9ba | 2148 | * (see split_page()). |
a145dd41 LT |
2149 | * |
2150 | * NOTE! Traditionally this was done with "remap_pfn_range()" which | |
2151 | * took an arbitrary page protection parameter. This doesn't allow | |
2152 | * that. Your vma protection will have to be set up correctly, which | |
2153 | * means that if you want a shared writable mapping, you'd better | |
2154 | * ask for a shared writable mapping! | |
2155 | * | |
2156 | * The page does not need to be reserved. | |
4b6e1e37 KK |
2157 | * |
2158 | * Usually this function is called from f_op->mmap() handler | |
2159 | * under mm->mmap_sem write-lock, so it can change vma->vm_flags. | |
2160 | * Caller must set VM_MIXEDMAP on vma if it wants to call this | |
2161 | * function from other places, for example from page-fault handler. | |
a145dd41 | 2162 | */ |
423bad60 NP |
2163 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
2164 | struct page *page) | |
a145dd41 LT |
2165 | { |
2166 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
2167 | return -EFAULT; | |
2168 | if (!page_count(page)) | |
2169 | return -EINVAL; | |
4b6e1e37 KK |
2170 | if (!(vma->vm_flags & VM_MIXEDMAP)) { |
2171 | BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem)); | |
2172 | BUG_ON(vma->vm_flags & VM_PFNMAP); | |
2173 | vma->vm_flags |= VM_MIXEDMAP; | |
2174 | } | |
423bad60 | 2175 | return insert_page(vma, addr, page, vma->vm_page_prot); |
a145dd41 | 2176 | } |
e3c3374f | 2177 | EXPORT_SYMBOL(vm_insert_page); |
a145dd41 | 2178 | |
423bad60 NP |
2179 | static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
2180 | unsigned long pfn, pgprot_t prot) | |
2181 | { | |
2182 | struct mm_struct *mm = vma->vm_mm; | |
2183 | int retval; | |
2184 | pte_t *pte, entry; | |
2185 | spinlock_t *ptl; | |
2186 | ||
2187 | retval = -ENOMEM; | |
2188 | pte = get_locked_pte(mm, addr, &ptl); | |
2189 | if (!pte) | |
2190 | goto out; | |
2191 | retval = -EBUSY; | |
2192 | if (!pte_none(*pte)) | |
2193 | goto out_unlock; | |
2194 | ||
2195 | /* Ok, finally just insert the thing.. */ | |
2196 | entry = pte_mkspecial(pfn_pte(pfn, prot)); | |
2197 | set_pte_at(mm, addr, pte, entry); | |
4b3073e1 | 2198 | update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ |
423bad60 NP |
2199 | |
2200 | retval = 0; | |
2201 | out_unlock: | |
2202 | pte_unmap_unlock(pte, ptl); | |
2203 | out: | |
2204 | return retval; | |
2205 | } | |
2206 | ||
e0dc0d8f NP |
2207 | /** |
2208 | * vm_insert_pfn - insert single pfn into user vma | |
2209 | * @vma: user vma to map to | |
2210 | * @addr: target user address of this page | |
2211 | * @pfn: source kernel pfn | |
2212 | * | |
c462f179 | 2213 | * Similar to vm_insert_page, this allows drivers to insert individual pages |
e0dc0d8f NP |
2214 | * they've allocated into a user vma. Same comments apply. |
2215 | * | |
2216 | * This function should only be called from a vm_ops->fault handler, and | |
2217 | * in that case the handler should return NULL. | |
0d71d10a NP |
2218 | * |
2219 | * vma cannot be a COW mapping. | |
2220 | * | |
2221 | * As this is called only for pages that do not currently exist, we | |
2222 | * do not need to flush old virtual caches or the TLB. | |
e0dc0d8f NP |
2223 | */ |
2224 | int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, | |
423bad60 | 2225 | unsigned long pfn) |
e0dc0d8f | 2226 | { |
2ab64037 | 2227 | int ret; |
e4b866ed | 2228 | pgprot_t pgprot = vma->vm_page_prot; |
7e675137 NP |
2229 | /* |
2230 | * Technically, architectures with pte_special can avoid all these | |
2231 | * restrictions (same for remap_pfn_range). However we would like | |
2232 | * consistency in testing and feature parity among all, so we should | |
2233 | * try to keep these invariants in place for everybody. | |
2234 | */ | |
b379d790 JH |
2235 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); |
2236 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == | |
2237 | (VM_PFNMAP|VM_MIXEDMAP)); | |
2238 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); | |
2239 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); | |
e0dc0d8f | 2240 | |
423bad60 NP |
2241 | if (addr < vma->vm_start || addr >= vma->vm_end) |
2242 | return -EFAULT; | |
5180da41 | 2243 | if (track_pfn_insert(vma, &pgprot, pfn)) |
2ab64037 | 2244 | return -EINVAL; |
2245 | ||
e4b866ed | 2246 | ret = insert_pfn(vma, addr, pfn, pgprot); |
2ab64037 | 2247 | |
2ab64037 | 2248 | return ret; |
423bad60 NP |
2249 | } |
2250 | EXPORT_SYMBOL(vm_insert_pfn); | |
e0dc0d8f | 2251 | |
423bad60 NP |
2252 | int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, |
2253 | unsigned long pfn) | |
2254 | { | |
2255 | BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); | |
e0dc0d8f | 2256 | |
423bad60 NP |
2257 | if (addr < vma->vm_start || addr >= vma->vm_end) |
2258 | return -EFAULT; | |
e0dc0d8f | 2259 | |
423bad60 NP |
2260 | /* |
2261 | * If we don't have pte special, then we have to use the pfn_valid() | |
2262 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* | |
2263 | * refcount the page if pfn_valid is true (hence insert_page rather | |
62eede62 HD |
2264 | * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP |
2265 | * without pte special, it would there be refcounted as a normal page. | |
423bad60 NP |
2266 | */ |
2267 | if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) { | |
2268 | struct page *page; | |
2269 | ||
2270 | page = pfn_to_page(pfn); | |
2271 | return insert_page(vma, addr, page, vma->vm_page_prot); | |
2272 | } | |
2273 | return insert_pfn(vma, addr, pfn, vma->vm_page_prot); | |
e0dc0d8f | 2274 | } |
423bad60 | 2275 | EXPORT_SYMBOL(vm_insert_mixed); |
e0dc0d8f | 2276 | |
1da177e4 LT |
2277 | /* |
2278 | * maps a range of physical memory into the requested pages. the old | |
2279 | * mappings are removed. any references to nonexistent pages results | |
2280 | * in null mappings (currently treated as "copy-on-access") | |
2281 | */ | |
2282 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
2283 | unsigned long addr, unsigned long end, | |
2284 | unsigned long pfn, pgprot_t prot) | |
2285 | { | |
2286 | pte_t *pte; | |
c74df32c | 2287 | spinlock_t *ptl; |
1da177e4 | 2288 | |
c74df32c | 2289 | pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1da177e4 LT |
2290 | if (!pte) |
2291 | return -ENOMEM; | |
6606c3e0 | 2292 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
2293 | do { |
2294 | BUG_ON(!pte_none(*pte)); | |
7e675137 | 2295 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
1da177e4 LT |
2296 | pfn++; |
2297 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 2298 | arch_leave_lazy_mmu_mode(); |
c74df32c | 2299 | pte_unmap_unlock(pte - 1, ptl); |
1da177e4 LT |
2300 | return 0; |
2301 | } | |
2302 | ||
2303 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
2304 | unsigned long addr, unsigned long end, | |
2305 | unsigned long pfn, pgprot_t prot) | |
2306 | { | |
2307 | pmd_t *pmd; | |
2308 | unsigned long next; | |
2309 | ||
2310 | pfn -= addr >> PAGE_SHIFT; | |
2311 | pmd = pmd_alloc(mm, pud, addr); | |
2312 | if (!pmd) | |
2313 | return -ENOMEM; | |
f66055ab | 2314 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
1da177e4 LT |
2315 | do { |
2316 | next = pmd_addr_end(addr, end); | |
2317 | if (remap_pte_range(mm, pmd, addr, next, | |
2318 | pfn + (addr >> PAGE_SHIFT), prot)) | |
2319 | return -ENOMEM; | |
2320 | } while (pmd++, addr = next, addr != end); | |
2321 | return 0; | |
2322 | } | |
2323 | ||
2324 | static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, | |
2325 | unsigned long addr, unsigned long end, | |
2326 | unsigned long pfn, pgprot_t prot) | |
2327 | { | |
2328 | pud_t *pud; | |
2329 | unsigned long next; | |
2330 | ||
2331 | pfn -= addr >> PAGE_SHIFT; | |
2332 | pud = pud_alloc(mm, pgd, addr); | |
2333 | if (!pud) | |
2334 | return -ENOMEM; | |
2335 | do { | |
2336 | next = pud_addr_end(addr, end); | |
2337 | if (remap_pmd_range(mm, pud, addr, next, | |
2338 | pfn + (addr >> PAGE_SHIFT), prot)) | |
2339 | return -ENOMEM; | |
2340 | } while (pud++, addr = next, addr != end); | |
2341 | return 0; | |
2342 | } | |
2343 | ||
bfa5bf6d REB |
2344 | /** |
2345 | * remap_pfn_range - remap kernel memory to userspace | |
2346 | * @vma: user vma to map to | |
2347 | * @addr: target user address to start at | |
2348 | * @pfn: physical address of kernel memory | |
2349 | * @size: size of map area | |
2350 | * @prot: page protection flags for this mapping | |
2351 | * | |
2352 | * Note: this is only safe if the mm semaphore is held when called. | |
2353 | */ | |
1da177e4 LT |
2354 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
2355 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
2356 | { | |
2357 | pgd_t *pgd; | |
2358 | unsigned long next; | |
2d15cab8 | 2359 | unsigned long end = addr + PAGE_ALIGN(size); |
1da177e4 LT |
2360 | struct mm_struct *mm = vma->vm_mm; |
2361 | int err; | |
2362 | ||
2363 | /* | |
2364 | * Physically remapped pages are special. Tell the | |
2365 | * rest of the world about it: | |
2366 | * VM_IO tells people not to look at these pages | |
2367 | * (accesses can have side effects). | |
6aab341e LT |
2368 | * VM_PFNMAP tells the core MM that the base pages are just |
2369 | * raw PFN mappings, and do not have a "struct page" associated | |
2370 | * with them. | |
314e51b9 KK |
2371 | * VM_DONTEXPAND |
2372 | * Disable vma merging and expanding with mremap(). | |
2373 | * VM_DONTDUMP | |
2374 | * Omit vma from core dump, even when VM_IO turned off. | |
fb155c16 LT |
2375 | * |
2376 | * There's a horrible special case to handle copy-on-write | |
2377 | * behaviour that some programs depend on. We mark the "original" | |
2378 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". | |
b3b9c293 | 2379 | * See vm_normal_page() for details. |
1da177e4 | 2380 | */ |
b3b9c293 KK |
2381 | if (is_cow_mapping(vma->vm_flags)) { |
2382 | if (addr != vma->vm_start || end != vma->vm_end) | |
2383 | return -EINVAL; | |
fb155c16 | 2384 | vma->vm_pgoff = pfn; |
b3b9c293 KK |
2385 | } |
2386 | ||
2387 | err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size)); | |
2388 | if (err) | |
3c8bb73a | 2389 | return -EINVAL; |
fb155c16 | 2390 | |
314e51b9 | 2391 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; |
1da177e4 LT |
2392 | |
2393 | BUG_ON(addr >= end); | |
2394 | pfn -= addr >> PAGE_SHIFT; | |
2395 | pgd = pgd_offset(mm, addr); | |
2396 | flush_cache_range(vma, addr, end); | |
1da177e4 LT |
2397 | do { |
2398 | next = pgd_addr_end(addr, end); | |
2399 | err = remap_pud_range(mm, pgd, addr, next, | |
2400 | pfn + (addr >> PAGE_SHIFT), prot); | |
2401 | if (err) | |
2402 | break; | |
2403 | } while (pgd++, addr = next, addr != end); | |
2ab64037 | 2404 | |
2405 | if (err) | |
5180da41 | 2406 | untrack_pfn(vma, pfn, PAGE_ALIGN(size)); |
2ab64037 | 2407 | |
1da177e4 LT |
2408 | return err; |
2409 | } | |
2410 | EXPORT_SYMBOL(remap_pfn_range); | |
2411 | ||
b4cbb197 LT |
2412 | /** |
2413 | * vm_iomap_memory - remap memory to userspace | |
2414 | * @vma: user vma to map to | |
2415 | * @start: start of area | |
2416 | * @len: size of area | |
2417 | * | |
2418 | * This is a simplified io_remap_pfn_range() for common driver use. The | |
2419 | * driver just needs to give us the physical memory range to be mapped, | |
2420 | * we'll figure out the rest from the vma information. | |
2421 | * | |
2422 | * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get | |
2423 | * whatever write-combining details or similar. | |
2424 | */ | |
2425 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) | |
2426 | { | |
2427 | unsigned long vm_len, pfn, pages; | |
2428 | ||
2429 | /* Check that the physical memory area passed in looks valid */ | |
2430 | if (start + len < start) | |
2431 | return -EINVAL; | |
2432 | /* | |
2433 | * You *really* shouldn't map things that aren't page-aligned, | |
2434 | * but we've historically allowed it because IO memory might | |
2435 | * just have smaller alignment. | |
2436 | */ | |
2437 | len += start & ~PAGE_MASK; | |
2438 | pfn = start >> PAGE_SHIFT; | |
2439 | pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; | |
2440 | if (pfn + pages < pfn) | |
2441 | return -EINVAL; | |
2442 | ||
2443 | /* We start the mapping 'vm_pgoff' pages into the area */ | |
2444 | if (vma->vm_pgoff > pages) | |
2445 | return -EINVAL; | |
2446 | pfn += vma->vm_pgoff; | |
2447 | pages -= vma->vm_pgoff; | |
2448 | ||
2449 | /* Can we fit all of the mapping? */ | |
2450 | vm_len = vma->vm_end - vma->vm_start; | |
2451 | if (vm_len >> PAGE_SHIFT > pages) | |
2452 | return -EINVAL; | |
2453 | ||
2454 | /* Ok, let it rip */ | |
2455 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); | |
2456 | } | |
2457 | EXPORT_SYMBOL(vm_iomap_memory); | |
2458 | ||
aee16b3c JF |
2459 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
2460 | unsigned long addr, unsigned long end, | |
2461 | pte_fn_t fn, void *data) | |
2462 | { | |
2463 | pte_t *pte; | |
2464 | int err; | |
2f569afd | 2465 | pgtable_t token; |
94909914 | 2466 | spinlock_t *uninitialized_var(ptl); |
aee16b3c JF |
2467 | |
2468 | pte = (mm == &init_mm) ? | |
2469 | pte_alloc_kernel(pmd, addr) : | |
2470 | pte_alloc_map_lock(mm, pmd, addr, &ptl); | |
2471 | if (!pte) | |
2472 | return -ENOMEM; | |
2473 | ||
2474 | BUG_ON(pmd_huge(*pmd)); | |
2475 | ||
38e0edb1 JF |
2476 | arch_enter_lazy_mmu_mode(); |
2477 | ||
2f569afd | 2478 | token = pmd_pgtable(*pmd); |
aee16b3c JF |
2479 | |
2480 | do { | |
c36987e2 | 2481 | err = fn(pte++, token, addr, data); |
aee16b3c JF |
2482 | if (err) |
2483 | break; | |
c36987e2 | 2484 | } while (addr += PAGE_SIZE, addr != end); |
aee16b3c | 2485 | |
38e0edb1 JF |
2486 | arch_leave_lazy_mmu_mode(); |
2487 | ||
aee16b3c JF |
2488 | if (mm != &init_mm) |
2489 | pte_unmap_unlock(pte-1, ptl); | |
2490 | return err; | |
2491 | } | |
2492 | ||
2493 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, | |
2494 | unsigned long addr, unsigned long end, | |
2495 | pte_fn_t fn, void *data) | |
2496 | { | |
2497 | pmd_t *pmd; | |
2498 | unsigned long next; | |
2499 | int err; | |
2500 | ||
ceb86879 AK |
2501 | BUG_ON(pud_huge(*pud)); |
2502 | ||
aee16b3c JF |
2503 | pmd = pmd_alloc(mm, pud, addr); |
2504 | if (!pmd) | |
2505 | return -ENOMEM; | |
2506 | do { | |
2507 | next = pmd_addr_end(addr, end); | |
2508 | err = apply_to_pte_range(mm, pmd, addr, next, fn, data); | |
2509 | if (err) | |
2510 | break; | |
2511 | } while (pmd++, addr = next, addr != end); | |
2512 | return err; | |
2513 | } | |
2514 | ||
2515 | static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd, | |
2516 | unsigned long addr, unsigned long end, | |
2517 | pte_fn_t fn, void *data) | |
2518 | { | |
2519 | pud_t *pud; | |
2520 | unsigned long next; | |
2521 | int err; | |
2522 | ||
2523 | pud = pud_alloc(mm, pgd, addr); | |
2524 | if (!pud) | |
2525 | return -ENOMEM; | |
2526 | do { | |
2527 | next = pud_addr_end(addr, end); | |
2528 | err = apply_to_pmd_range(mm, pud, addr, next, fn, data); | |
2529 | if (err) | |
2530 | break; | |
2531 | } while (pud++, addr = next, addr != end); | |
2532 | return err; | |
2533 | } | |
2534 | ||
2535 | /* | |
2536 | * Scan a region of virtual memory, filling in page tables as necessary | |
2537 | * and calling a provided function on each leaf page table. | |
2538 | */ | |
2539 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, | |
2540 | unsigned long size, pte_fn_t fn, void *data) | |
2541 | { | |
2542 | pgd_t *pgd; | |
2543 | unsigned long next; | |
57250a5b | 2544 | unsigned long end = addr + size; |
aee16b3c JF |
2545 | int err; |
2546 | ||
2547 | BUG_ON(addr >= end); | |
2548 | pgd = pgd_offset(mm, addr); | |
2549 | do { | |
2550 | next = pgd_addr_end(addr, end); | |
2551 | err = apply_to_pud_range(mm, pgd, addr, next, fn, data); | |
2552 | if (err) | |
2553 | break; | |
2554 | } while (pgd++, addr = next, addr != end); | |
57250a5b | 2555 | |
aee16b3c JF |
2556 | return err; |
2557 | } | |
2558 | EXPORT_SYMBOL_GPL(apply_to_page_range); | |
2559 | ||
8f4e2101 HD |
2560 | /* |
2561 | * handle_pte_fault chooses page fault handler according to an entry | |
2562 | * which was read non-atomically. Before making any commitment, on | |
2563 | * those architectures or configurations (e.g. i386 with PAE) which | |
a335b2e1 | 2564 | * might give a mix of unmatched parts, do_swap_page and do_nonlinear_fault |
8f4e2101 HD |
2565 | * must check under lock before unmapping the pte and proceeding |
2566 | * (but do_wp_page is only called after already making such a check; | |
a335b2e1 | 2567 | * and do_anonymous_page can safely check later on). |
8f4e2101 | 2568 | */ |
4c21e2f2 | 2569 | static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, |
8f4e2101 HD |
2570 | pte_t *page_table, pte_t orig_pte) |
2571 | { | |
2572 | int same = 1; | |
2573 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) | |
2574 | if (sizeof(pte_t) > sizeof(unsigned long)) { | |
4c21e2f2 HD |
2575 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
2576 | spin_lock(ptl); | |
8f4e2101 | 2577 | same = pte_same(*page_table, orig_pte); |
4c21e2f2 | 2578 | spin_unlock(ptl); |
8f4e2101 HD |
2579 | } |
2580 | #endif | |
2581 | pte_unmap(page_table); | |
2582 | return same; | |
2583 | } | |
2584 | ||
9de455b2 | 2585 | static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) |
6aab341e | 2586 | { |
0abdd7a8 DW |
2587 | debug_dma_assert_idle(src); |
2588 | ||
6aab341e LT |
2589 | /* |
2590 | * If the source page was a PFN mapping, we don't have | |
2591 | * a "struct page" for it. We do a best-effort copy by | |
2592 | * just copying from the original user address. If that | |
2593 | * fails, we just zero-fill it. Live with it. | |
2594 | */ | |
2595 | if (unlikely(!src)) { | |
9b04c5fe | 2596 | void *kaddr = kmap_atomic(dst); |
5d2a2dbb LT |
2597 | void __user *uaddr = (void __user *)(va & PAGE_MASK); |
2598 | ||
2599 | /* | |
2600 | * This really shouldn't fail, because the page is there | |
2601 | * in the page tables. But it might just be unreadable, | |
2602 | * in which case we just give up and fill the result with | |
2603 | * zeroes. | |
2604 | */ | |
2605 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) | |
3ecb01df | 2606 | clear_page(kaddr); |
9b04c5fe | 2607 | kunmap_atomic(kaddr); |
c4ec7b0d | 2608 | flush_dcache_page(dst); |
0ed361de NP |
2609 | } else |
2610 | copy_user_highpage(dst, src, va, vma); | |
6aab341e LT |
2611 | } |
2612 | ||
fb09a464 KS |
2613 | /* |
2614 | * Notify the address space that the page is about to become writable so that | |
2615 | * it can prohibit this or wait for the page to get into an appropriate state. | |
2616 | * | |
2617 | * We do this without the lock held, so that it can sleep if it needs to. | |
2618 | */ | |
2619 | static int do_page_mkwrite(struct vm_area_struct *vma, struct page *page, | |
2620 | unsigned long address) | |
2621 | { | |
2622 | struct vm_fault vmf; | |
2623 | int ret; | |
2624 | ||
2625 | vmf.virtual_address = (void __user *)(address & PAGE_MASK); | |
2626 | vmf.pgoff = page->index; | |
2627 | vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; | |
2628 | vmf.page = page; | |
2629 | ||
2630 | ret = vma->vm_ops->page_mkwrite(vma, &vmf); | |
2631 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) | |
2632 | return ret; | |
2633 | if (unlikely(!(ret & VM_FAULT_LOCKED))) { | |
2634 | lock_page(page); | |
2635 | if (!page->mapping) { | |
2636 | unlock_page(page); | |
2637 | return 0; /* retry */ | |
2638 | } | |
2639 | ret |= VM_FAULT_LOCKED; | |
2640 | } else | |
2641 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
2642 | return ret; | |
2643 | } | |
2644 | ||
1da177e4 LT |
2645 | /* |
2646 | * This routine handles present pages, when users try to write | |
2647 | * to a shared page. It is done by copying the page to a new address | |
2648 | * and decrementing the shared-page counter for the old page. | |
2649 | * | |
1da177e4 LT |
2650 | * Note that this routine assumes that the protection checks have been |
2651 | * done by the caller (the low-level page fault routine in most cases). | |
2652 | * Thus we can safely just mark it writable once we've done any necessary | |
2653 | * COW. | |
2654 | * | |
2655 | * We also mark the page dirty at this point even though the page will | |
2656 | * change only once the write actually happens. This avoids a few races, | |
2657 | * and potentially makes it more efficient. | |
2658 | * | |
8f4e2101 HD |
2659 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2660 | * but allow concurrent faults), with pte both mapped and locked. | |
2661 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2662 | */ |
65500d23 HD |
2663 | static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2664 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
8f4e2101 | 2665 | spinlock_t *ptl, pte_t orig_pte) |
e6219ec8 | 2666 | __releases(ptl) |
1da177e4 | 2667 | { |
2ec74c3e | 2668 | struct page *old_page, *new_page = NULL; |
1da177e4 | 2669 | pte_t entry; |
b009c024 | 2670 | int ret = 0; |
a200ee18 | 2671 | int page_mkwrite = 0; |
d08b3851 | 2672 | struct page *dirty_page = NULL; |
1756954c DR |
2673 | unsigned long mmun_start = 0; /* For mmu_notifiers */ |
2674 | unsigned long mmun_end = 0; /* For mmu_notifiers */ | |
1da177e4 | 2675 | |
6aab341e | 2676 | old_page = vm_normal_page(vma, address, orig_pte); |
251b97f5 PZ |
2677 | if (!old_page) { |
2678 | /* | |
2679 | * VM_MIXEDMAP !pfn_valid() case | |
2680 | * | |
2681 | * We should not cow pages in a shared writeable mapping. | |
2682 | * Just mark the pages writable as we can't do any dirty | |
2683 | * accounting on raw pfn maps. | |
2684 | */ | |
2685 | if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == | |
2686 | (VM_WRITE|VM_SHARED)) | |
2687 | goto reuse; | |
6aab341e | 2688 | goto gotten; |
251b97f5 | 2689 | } |
1da177e4 | 2690 | |
d08b3851 | 2691 | /* |
ee6a6457 PZ |
2692 | * Take out anonymous pages first, anonymous shared vmas are |
2693 | * not dirty accountable. | |
d08b3851 | 2694 | */ |
9a840895 | 2695 | if (PageAnon(old_page) && !PageKsm(old_page)) { |
ab967d86 HD |
2696 | if (!trylock_page(old_page)) { |
2697 | page_cache_get(old_page); | |
2698 | pte_unmap_unlock(page_table, ptl); | |
2699 | lock_page(old_page); | |
2700 | page_table = pte_offset_map_lock(mm, pmd, address, | |
2701 | &ptl); | |
2702 | if (!pte_same(*page_table, orig_pte)) { | |
2703 | unlock_page(old_page); | |
ab967d86 HD |
2704 | goto unlock; |
2705 | } | |
2706 | page_cache_release(old_page); | |
ee6a6457 | 2707 | } |
b009c024 | 2708 | if (reuse_swap_page(old_page)) { |
c44b6743 RR |
2709 | /* |
2710 | * The page is all ours. Move it to our anon_vma so | |
2711 | * the rmap code will not search our parent or siblings. | |
2712 | * Protected against the rmap code by the page lock. | |
2713 | */ | |
2714 | page_move_anon_rmap(old_page, vma, address); | |
b009c024 ML |
2715 | unlock_page(old_page); |
2716 | goto reuse; | |
2717 | } | |
ab967d86 | 2718 | unlock_page(old_page); |
ee6a6457 | 2719 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
d08b3851 | 2720 | (VM_WRITE|VM_SHARED))) { |
ee6a6457 PZ |
2721 | /* |
2722 | * Only catch write-faults on shared writable pages, | |
2723 | * read-only shared pages can get COWed by | |
2724 | * get_user_pages(.write=1, .force=1). | |
2725 | */ | |
9637a5ef | 2726 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
c2ec175c | 2727 | int tmp; |
9637a5ef DH |
2728 | page_cache_get(old_page); |
2729 | pte_unmap_unlock(page_table, ptl); | |
fb09a464 KS |
2730 | tmp = do_page_mkwrite(vma, old_page, address); |
2731 | if (unlikely(!tmp || (tmp & | |
2732 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
2733 | page_cache_release(old_page); | |
2734 | return tmp; | |
c2ec175c | 2735 | } |
9637a5ef DH |
2736 | /* |
2737 | * Since we dropped the lock we need to revalidate | |
2738 | * the PTE as someone else may have changed it. If | |
2739 | * they did, we just return, as we can count on the | |
2740 | * MMU to tell us if they didn't also make it writable. | |
2741 | */ | |
2742 | page_table = pte_offset_map_lock(mm, pmd, address, | |
2743 | &ptl); | |
b827e496 NP |
2744 | if (!pte_same(*page_table, orig_pte)) { |
2745 | unlock_page(old_page); | |
9637a5ef | 2746 | goto unlock; |
b827e496 | 2747 | } |
a200ee18 PZ |
2748 | |
2749 | page_mkwrite = 1; | |
1da177e4 | 2750 | } |
d08b3851 PZ |
2751 | dirty_page = old_page; |
2752 | get_page(dirty_page); | |
9637a5ef | 2753 | |
251b97f5 | 2754 | reuse: |
8c8a743c PZ |
2755 | /* |
2756 | * Clear the pages cpupid information as the existing | |
2757 | * information potentially belongs to a now completely | |
2758 | * unrelated process. | |
2759 | */ | |
2760 | if (old_page) | |
2761 | page_cpupid_xchg_last(old_page, (1 << LAST_CPUPID_SHIFT) - 1); | |
2762 | ||
9637a5ef DH |
2763 | flush_cache_page(vma, address, pte_pfn(orig_pte)); |
2764 | entry = pte_mkyoung(orig_pte); | |
2765 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
954ffcb3 | 2766 | if (ptep_set_access_flags(vma, address, page_table, entry,1)) |
4b3073e1 | 2767 | update_mmu_cache(vma, address, page_table); |
72ddc8f7 | 2768 | pte_unmap_unlock(page_table, ptl); |
9637a5ef | 2769 | ret |= VM_FAULT_WRITE; |
72ddc8f7 ML |
2770 | |
2771 | if (!dirty_page) | |
2772 | return ret; | |
2773 | ||
2774 | /* | |
2775 | * Yes, Virginia, this is actually required to prevent a race | |
2776 | * with clear_page_dirty_for_io() from clearing the page dirty | |
2777 | * bit after it clear all dirty ptes, but before a racing | |
2778 | * do_wp_page installs a dirty pte. | |
2779 | * | |
f0c6d4d2 | 2780 | * do_shared_fault is protected similarly. |
72ddc8f7 ML |
2781 | */ |
2782 | if (!page_mkwrite) { | |
2783 | wait_on_page_locked(dirty_page); | |
2784 | set_page_dirty_balance(dirty_page, page_mkwrite); | |
41c4d25f JK |
2785 | /* file_update_time outside page_lock */ |
2786 | if (vma->vm_file) | |
2787 | file_update_time(vma->vm_file); | |
72ddc8f7 ML |
2788 | } |
2789 | put_page(dirty_page); | |
2790 | if (page_mkwrite) { | |
2791 | struct address_space *mapping = dirty_page->mapping; | |
2792 | ||
2793 | set_page_dirty(dirty_page); | |
2794 | unlock_page(dirty_page); | |
2795 | page_cache_release(dirty_page); | |
2796 | if (mapping) { | |
2797 | /* | |
2798 | * Some device drivers do not set page.mapping | |
2799 | * but still dirty their pages | |
2800 | */ | |
2801 | balance_dirty_pages_ratelimited(mapping); | |
2802 | } | |
2803 | } | |
2804 | ||
72ddc8f7 | 2805 | return ret; |
1da177e4 | 2806 | } |
1da177e4 LT |
2807 | |
2808 | /* | |
2809 | * Ok, we need to copy. Oh, well.. | |
2810 | */ | |
b5810039 | 2811 | page_cache_get(old_page); |
920fc356 | 2812 | gotten: |
8f4e2101 | 2813 | pte_unmap_unlock(page_table, ptl); |
1da177e4 LT |
2814 | |
2815 | if (unlikely(anon_vma_prepare(vma))) | |
65500d23 | 2816 | goto oom; |
a13ea5b7 | 2817 | |
62eede62 | 2818 | if (is_zero_pfn(pte_pfn(orig_pte))) { |
a13ea5b7 HD |
2819 | new_page = alloc_zeroed_user_highpage_movable(vma, address); |
2820 | if (!new_page) | |
2821 | goto oom; | |
2822 | } else { | |
2823 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); | |
2824 | if (!new_page) | |
2825 | goto oom; | |
2826 | cow_user_page(new_page, old_page, address, vma); | |
2827 | } | |
2828 | __SetPageUptodate(new_page); | |
2829 | ||
2c26fdd7 | 2830 | if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL)) |
8a9f3ccd BS |
2831 | goto oom_free_new; |
2832 | ||
6bdb913f | 2833 | mmun_start = address & PAGE_MASK; |
1756954c | 2834 | mmun_end = mmun_start + PAGE_SIZE; |
6bdb913f HE |
2835 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
2836 | ||
1da177e4 LT |
2837 | /* |
2838 | * Re-check the pte - we dropped the lock | |
2839 | */ | |
8f4e2101 | 2840 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
65500d23 | 2841 | if (likely(pte_same(*page_table, orig_pte))) { |
920fc356 | 2842 | if (old_page) { |
920fc356 | 2843 | if (!PageAnon(old_page)) { |
34e55232 KH |
2844 | dec_mm_counter_fast(mm, MM_FILEPAGES); |
2845 | inc_mm_counter_fast(mm, MM_ANONPAGES); | |
920fc356 HD |
2846 | } |
2847 | } else | |
34e55232 | 2848 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
eca35133 | 2849 | flush_cache_page(vma, address, pte_pfn(orig_pte)); |
65500d23 HD |
2850 | entry = mk_pte(new_page, vma->vm_page_prot); |
2851 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
4ce072f1 SS |
2852 | /* |
2853 | * Clear the pte entry and flush it first, before updating the | |
2854 | * pte with the new entry. This will avoid a race condition | |
2855 | * seen in the presence of one thread doing SMC and another | |
2856 | * thread doing COW. | |
2857 | */ | |
828502d3 | 2858 | ptep_clear_flush(vma, address, page_table); |
9617d95e | 2859 | page_add_new_anon_rmap(new_page, vma, address); |
828502d3 IE |
2860 | /* |
2861 | * We call the notify macro here because, when using secondary | |
2862 | * mmu page tables (such as kvm shadow page tables), we want the | |
2863 | * new page to be mapped directly into the secondary page table. | |
2864 | */ | |
2865 | set_pte_at_notify(mm, address, page_table, entry); | |
4b3073e1 | 2866 | update_mmu_cache(vma, address, page_table); |
945754a1 NP |
2867 | if (old_page) { |
2868 | /* | |
2869 | * Only after switching the pte to the new page may | |
2870 | * we remove the mapcount here. Otherwise another | |
2871 | * process may come and find the rmap count decremented | |
2872 | * before the pte is switched to the new page, and | |
2873 | * "reuse" the old page writing into it while our pte | |
2874 | * here still points into it and can be read by other | |
2875 | * threads. | |
2876 | * | |
2877 | * The critical issue is to order this | |
2878 | * page_remove_rmap with the ptp_clear_flush above. | |
2879 | * Those stores are ordered by (if nothing else,) | |
2880 | * the barrier present in the atomic_add_negative | |
2881 | * in page_remove_rmap. | |
2882 | * | |
2883 | * Then the TLB flush in ptep_clear_flush ensures that | |
2884 | * no process can access the old page before the | |
2885 | * decremented mapcount is visible. And the old page | |
2886 | * cannot be reused until after the decremented | |
2887 | * mapcount is visible. So transitively, TLBs to | |
2888 | * old page will be flushed before it can be reused. | |
2889 | */ | |
edc315fd | 2890 | page_remove_rmap(old_page); |
945754a1 NP |
2891 | } |
2892 | ||
1da177e4 LT |
2893 | /* Free the old page.. */ |
2894 | new_page = old_page; | |
f33ea7f4 | 2895 | ret |= VM_FAULT_WRITE; |
8a9f3ccd BS |
2896 | } else |
2897 | mem_cgroup_uncharge_page(new_page); | |
2898 | ||
6bdb913f HE |
2899 | if (new_page) |
2900 | page_cache_release(new_page); | |
65500d23 | 2901 | unlock: |
8f4e2101 | 2902 | pte_unmap_unlock(page_table, ptl); |
1756954c | 2903 | if (mmun_end > mmun_start) |
6bdb913f | 2904 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); |
e15f8c01 ML |
2905 | if (old_page) { |
2906 | /* | |
2907 | * Don't let another task, with possibly unlocked vma, | |
2908 | * keep the mlocked page. | |
2909 | */ | |
2910 | if ((ret & VM_FAULT_WRITE) && (vma->vm_flags & VM_LOCKED)) { | |
2911 | lock_page(old_page); /* LRU manipulation */ | |
2912 | munlock_vma_page(old_page); | |
2913 | unlock_page(old_page); | |
2914 | } | |
2915 | page_cache_release(old_page); | |
2916 | } | |
f33ea7f4 | 2917 | return ret; |
8a9f3ccd | 2918 | oom_free_new: |
6dbf6d3b | 2919 | page_cache_release(new_page); |
65500d23 | 2920 | oom: |
66521d5a | 2921 | if (old_page) |
920fc356 | 2922 | page_cache_release(old_page); |
1da177e4 LT |
2923 | return VM_FAULT_OOM; |
2924 | } | |
2925 | ||
97a89413 | 2926 | static void unmap_mapping_range_vma(struct vm_area_struct *vma, |
1da177e4 LT |
2927 | unsigned long start_addr, unsigned long end_addr, |
2928 | struct zap_details *details) | |
2929 | { | |
f5cc4eef | 2930 | zap_page_range_single(vma, start_addr, end_addr - start_addr, details); |
1da177e4 LT |
2931 | } |
2932 | ||
6b2dbba8 | 2933 | static inline void unmap_mapping_range_tree(struct rb_root *root, |
1da177e4 LT |
2934 | struct zap_details *details) |
2935 | { | |
2936 | struct vm_area_struct *vma; | |
1da177e4 LT |
2937 | pgoff_t vba, vea, zba, zea; |
2938 | ||
6b2dbba8 | 2939 | vma_interval_tree_foreach(vma, root, |
1da177e4 | 2940 | details->first_index, details->last_index) { |
1da177e4 LT |
2941 | |
2942 | vba = vma->vm_pgoff; | |
d6e93217 | 2943 | vea = vba + vma_pages(vma) - 1; |
1da177e4 LT |
2944 | /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */ |
2945 | zba = details->first_index; | |
2946 | if (zba < vba) | |
2947 | zba = vba; | |
2948 | zea = details->last_index; | |
2949 | if (zea > vea) | |
2950 | zea = vea; | |
2951 | ||
97a89413 | 2952 | unmap_mapping_range_vma(vma, |
1da177e4 LT |
2953 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, |
2954 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, | |
97a89413 | 2955 | details); |
1da177e4 LT |
2956 | } |
2957 | } | |
2958 | ||
2959 | static inline void unmap_mapping_range_list(struct list_head *head, | |
2960 | struct zap_details *details) | |
2961 | { | |
2962 | struct vm_area_struct *vma; | |
2963 | ||
2964 | /* | |
2965 | * In nonlinear VMAs there is no correspondence between virtual address | |
2966 | * offset and file offset. So we must perform an exhaustive search | |
2967 | * across *all* the pages in each nonlinear VMA, not just the pages | |
2968 | * whose virtual address lies outside the file truncation point. | |
2969 | */ | |
6b2dbba8 | 2970 | list_for_each_entry(vma, head, shared.nonlinear) { |
1da177e4 | 2971 | details->nonlinear_vma = vma; |
97a89413 | 2972 | unmap_mapping_range_vma(vma, vma->vm_start, vma->vm_end, details); |
1da177e4 LT |
2973 | } |
2974 | } | |
2975 | ||
2976 | /** | |
72fd4a35 | 2977 | * 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 | 2978 | * @mapping: the address space containing mmaps to be unmapped. |
1da177e4 LT |
2979 | * @holebegin: byte in first page to unmap, relative to the start of |
2980 | * the underlying file. This will be rounded down to a PAGE_SIZE | |
25d9e2d1 | 2981 | * boundary. Note that this is different from truncate_pagecache(), which |
1da177e4 LT |
2982 | * must keep the partial page. In contrast, we must get rid of |
2983 | * partial pages. | |
2984 | * @holelen: size of prospective hole in bytes. This will be rounded | |
2985 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the | |
2986 | * end of the file. | |
2987 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; | |
2988 | * but 0 when invalidating pagecache, don't throw away private data. | |
2989 | */ | |
2990 | void unmap_mapping_range(struct address_space *mapping, | |
2991 | loff_t const holebegin, loff_t const holelen, int even_cows) | |
2992 | { | |
2993 | struct zap_details details; | |
2994 | pgoff_t hba = holebegin >> PAGE_SHIFT; | |
2995 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2996 | ||
2997 | /* Check for overflow. */ | |
2998 | if (sizeof(holelen) > sizeof(hlen)) { | |
2999 | long long holeend = | |
3000 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
3001 | if (holeend & ~(long long)ULONG_MAX) | |
3002 | hlen = ULONG_MAX - hba + 1; | |
3003 | } | |
3004 | ||
3005 | details.check_mapping = even_cows? NULL: mapping; | |
3006 | details.nonlinear_vma = NULL; | |
3007 | details.first_index = hba; | |
3008 | details.last_index = hba + hlen - 1; | |
3009 | if (details.last_index < details.first_index) | |
3010 | details.last_index = ULONG_MAX; | |
1da177e4 | 3011 | |
1da177e4 | 3012 | |
3d48ae45 | 3013 | mutex_lock(&mapping->i_mmap_mutex); |
6b2dbba8 | 3014 | if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap))) |
1da177e4 LT |
3015 | unmap_mapping_range_tree(&mapping->i_mmap, &details); |
3016 | if (unlikely(!list_empty(&mapping->i_mmap_nonlinear))) | |
3017 | unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details); | |
3d48ae45 | 3018 | mutex_unlock(&mapping->i_mmap_mutex); |
1da177e4 LT |
3019 | } |
3020 | EXPORT_SYMBOL(unmap_mapping_range); | |
3021 | ||
1da177e4 | 3022 | /* |
8f4e2101 HD |
3023 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
3024 | * but allow concurrent faults), and pte mapped but not yet locked. | |
3025 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 3026 | */ |
65500d23 HD |
3027 | static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, |
3028 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
30c9f3a9 | 3029 | unsigned int flags, pte_t orig_pte) |
1da177e4 | 3030 | { |
8f4e2101 | 3031 | spinlock_t *ptl; |
56f31801 | 3032 | struct page *page, *swapcache; |
65500d23 | 3033 | swp_entry_t entry; |
1da177e4 | 3034 | pte_t pte; |
d065bd81 | 3035 | int locked; |
56039efa | 3036 | struct mem_cgroup *ptr; |
ad8c2ee8 | 3037 | int exclusive = 0; |
83c54070 | 3038 | int ret = 0; |
1da177e4 | 3039 | |
4c21e2f2 | 3040 | if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) |
8f4e2101 | 3041 | goto out; |
65500d23 HD |
3042 | |
3043 | entry = pte_to_swp_entry(orig_pte); | |
d1737fdb AK |
3044 | if (unlikely(non_swap_entry(entry))) { |
3045 | if (is_migration_entry(entry)) { | |
3046 | migration_entry_wait(mm, pmd, address); | |
3047 | } else if (is_hwpoison_entry(entry)) { | |
3048 | ret = VM_FAULT_HWPOISON; | |
3049 | } else { | |
3050 | print_bad_pte(vma, address, orig_pte, NULL); | |
d99be1a8 | 3051 | ret = VM_FAULT_SIGBUS; |
d1737fdb | 3052 | } |
0697212a CL |
3053 | goto out; |
3054 | } | |
0ff92245 | 3055 | delayacct_set_flag(DELAYACCT_PF_SWAPIN); |
1da177e4 LT |
3056 | page = lookup_swap_cache(entry); |
3057 | if (!page) { | |
02098fea HD |
3058 | page = swapin_readahead(entry, |
3059 | GFP_HIGHUSER_MOVABLE, vma, address); | |
1da177e4 LT |
3060 | if (!page) { |
3061 | /* | |
8f4e2101 HD |
3062 | * Back out if somebody else faulted in this pte |
3063 | * while we released the pte lock. | |
1da177e4 | 3064 | */ |
8f4e2101 | 3065 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
1da177e4 LT |
3066 | if (likely(pte_same(*page_table, orig_pte))) |
3067 | ret = VM_FAULT_OOM; | |
0ff92245 | 3068 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
65500d23 | 3069 | goto unlock; |
1da177e4 LT |
3070 | } |
3071 | ||
3072 | /* Had to read the page from swap area: Major fault */ | |
3073 | ret = VM_FAULT_MAJOR; | |
f8891e5e | 3074 | count_vm_event(PGMAJFAULT); |
456f998e | 3075 | mem_cgroup_count_vm_event(mm, PGMAJFAULT); |
d1737fdb | 3076 | } else if (PageHWPoison(page)) { |
71f72525 WF |
3077 | /* |
3078 | * hwpoisoned dirty swapcache pages are kept for killing | |
3079 | * owner processes (which may be unknown at hwpoison time) | |
3080 | */ | |
d1737fdb AK |
3081 | ret = VM_FAULT_HWPOISON; |
3082 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); | |
56f31801 | 3083 | swapcache = page; |
4779cb31 | 3084 | goto out_release; |
1da177e4 LT |
3085 | } |
3086 | ||
56f31801 | 3087 | swapcache = page; |
d065bd81 | 3088 | locked = lock_page_or_retry(page, mm, flags); |
e709ffd6 | 3089 | |
073e587e | 3090 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
d065bd81 ML |
3091 | if (!locked) { |
3092 | ret |= VM_FAULT_RETRY; | |
3093 | goto out_release; | |
3094 | } | |
073e587e | 3095 | |
4969c119 | 3096 | /* |
31c4a3d3 HD |
3097 | * Make sure try_to_free_swap or reuse_swap_page or swapoff did not |
3098 | * release the swapcache from under us. The page pin, and pte_same | |
3099 | * test below, are not enough to exclude that. Even if it is still | |
3100 | * swapcache, we need to check that the page's swap has not changed. | |
4969c119 | 3101 | */ |
31c4a3d3 | 3102 | if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val)) |
4969c119 AA |
3103 | goto out_page; |
3104 | ||
cbf86cfe HD |
3105 | page = ksm_might_need_to_copy(page, vma, address); |
3106 | if (unlikely(!page)) { | |
3107 | ret = VM_FAULT_OOM; | |
3108 | page = swapcache; | |
cbf86cfe | 3109 | goto out_page; |
5ad64688 HD |
3110 | } |
3111 | ||
2c26fdd7 | 3112 | if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) { |
8a9f3ccd | 3113 | ret = VM_FAULT_OOM; |
bc43f75c | 3114 | goto out_page; |
8a9f3ccd BS |
3115 | } |
3116 | ||
1da177e4 | 3117 | /* |
8f4e2101 | 3118 | * Back out if somebody else already faulted in this pte. |
1da177e4 | 3119 | */ |
8f4e2101 | 3120 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
9e9bef07 | 3121 | if (unlikely(!pte_same(*page_table, orig_pte))) |
b8107480 | 3122 | goto out_nomap; |
b8107480 KK |
3123 | |
3124 | if (unlikely(!PageUptodate(page))) { | |
3125 | ret = VM_FAULT_SIGBUS; | |
3126 | goto out_nomap; | |
1da177e4 LT |
3127 | } |
3128 | ||
8c7c6e34 KH |
3129 | /* |
3130 | * The page isn't present yet, go ahead with the fault. | |
3131 | * | |
3132 | * Be careful about the sequence of operations here. | |
3133 | * To get its accounting right, reuse_swap_page() must be called | |
3134 | * while the page is counted on swap but not yet in mapcount i.e. | |
3135 | * before page_add_anon_rmap() and swap_free(); try_to_free_swap() | |
3136 | * must be called after the swap_free(), or it will never succeed. | |
03f3c433 KH |
3137 | * Because delete_from_swap_page() may be called by reuse_swap_page(), |
3138 | * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry | |
3139 | * in page->private. In this case, a record in swap_cgroup is silently | |
3140 | * discarded at swap_free(). | |
8c7c6e34 | 3141 | */ |
1da177e4 | 3142 | |
34e55232 | 3143 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
b084d435 | 3144 | dec_mm_counter_fast(mm, MM_SWAPENTS); |
1da177e4 | 3145 | pte = mk_pte(page, vma->vm_page_prot); |
30c9f3a9 | 3146 | if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) { |
1da177e4 | 3147 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); |
30c9f3a9 | 3148 | flags &= ~FAULT_FLAG_WRITE; |
9a5b489b | 3149 | ret |= VM_FAULT_WRITE; |
ad8c2ee8 | 3150 | exclusive = 1; |
1da177e4 | 3151 | } |
1da177e4 | 3152 | flush_icache_page(vma, page); |
179ef71c CG |
3153 | if (pte_swp_soft_dirty(orig_pte)) |
3154 | pte = pte_mksoft_dirty(pte); | |
1da177e4 | 3155 | set_pte_at(mm, address, page_table, pte); |
56f31801 | 3156 | if (page == swapcache) |
af34770e | 3157 | do_page_add_anon_rmap(page, vma, address, exclusive); |
56f31801 HD |
3158 | else /* ksm created a completely new copy */ |
3159 | page_add_new_anon_rmap(page, vma, address); | |
03f3c433 KH |
3160 | /* It's better to call commit-charge after rmap is established */ |
3161 | mem_cgroup_commit_charge_swapin(page, ptr); | |
1da177e4 | 3162 | |
c475a8ab | 3163 | swap_free(entry); |
b291f000 | 3164 | if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) |
a2c43eed | 3165 | try_to_free_swap(page); |
c475a8ab | 3166 | unlock_page(page); |
56f31801 | 3167 | if (page != swapcache) { |
4969c119 AA |
3168 | /* |
3169 | * Hold the lock to avoid the swap entry to be reused | |
3170 | * until we take the PT lock for the pte_same() check | |
3171 | * (to avoid false positives from pte_same). For | |
3172 | * further safety release the lock after the swap_free | |
3173 | * so that the swap count won't change under a | |
3174 | * parallel locked swapcache. | |
3175 | */ | |
3176 | unlock_page(swapcache); | |
3177 | page_cache_release(swapcache); | |
3178 | } | |
c475a8ab | 3179 | |
30c9f3a9 | 3180 | if (flags & FAULT_FLAG_WRITE) { |
61469f1d HD |
3181 | ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte); |
3182 | if (ret & VM_FAULT_ERROR) | |
3183 | ret &= VM_FAULT_ERROR; | |
1da177e4 LT |
3184 | goto out; |
3185 | } | |
3186 | ||
3187 | /* No need to invalidate - it was non-present before */ | |
4b3073e1 | 3188 | update_mmu_cache(vma, address, page_table); |
65500d23 | 3189 | unlock: |
8f4e2101 | 3190 | pte_unmap_unlock(page_table, ptl); |
1da177e4 LT |
3191 | out: |
3192 | return ret; | |
b8107480 | 3193 | out_nomap: |
7a81b88c | 3194 | mem_cgroup_cancel_charge_swapin(ptr); |
8f4e2101 | 3195 | pte_unmap_unlock(page_table, ptl); |
bc43f75c | 3196 | out_page: |
b8107480 | 3197 | unlock_page(page); |
4779cb31 | 3198 | out_release: |
b8107480 | 3199 | page_cache_release(page); |
56f31801 | 3200 | if (page != swapcache) { |
4969c119 AA |
3201 | unlock_page(swapcache); |
3202 | page_cache_release(swapcache); | |
3203 | } | |
65500d23 | 3204 | return ret; |
1da177e4 LT |
3205 | } |
3206 | ||
320b2b8d | 3207 | /* |
8ca3eb08 TL |
3208 | * This is like a special single-page "expand_{down|up}wards()", |
3209 | * except we must first make sure that 'address{-|+}PAGE_SIZE' | |
320b2b8d | 3210 | * doesn't hit another vma. |
320b2b8d LT |
3211 | */ |
3212 | static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address) | |
3213 | { | |
3214 | address &= PAGE_MASK; | |
3215 | if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) { | |
0e8e50e2 LT |
3216 | struct vm_area_struct *prev = vma->vm_prev; |
3217 | ||
3218 | /* | |
3219 | * Is there a mapping abutting this one below? | |
3220 | * | |
3221 | * That's only ok if it's the same stack mapping | |
3222 | * that has gotten split.. | |
3223 | */ | |
3224 | if (prev && prev->vm_end == address) | |
3225 | return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM; | |
320b2b8d | 3226 | |
d05f3169 | 3227 | expand_downwards(vma, address - PAGE_SIZE); |
320b2b8d | 3228 | } |
8ca3eb08 TL |
3229 | if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) { |
3230 | struct vm_area_struct *next = vma->vm_next; | |
3231 | ||
3232 | /* As VM_GROWSDOWN but s/below/above/ */ | |
3233 | if (next && next->vm_start == address + PAGE_SIZE) | |
3234 | return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM; | |
3235 | ||
3236 | expand_upwards(vma, address + PAGE_SIZE); | |
3237 | } | |
320b2b8d LT |
3238 | return 0; |
3239 | } | |
3240 | ||
1da177e4 | 3241 | /* |
8f4e2101 HD |
3242 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
3243 | * but allow concurrent faults), and pte mapped but not yet locked. | |
3244 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 3245 | */ |
65500d23 HD |
3246 | static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma, |
3247 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
30c9f3a9 | 3248 | unsigned int flags) |
1da177e4 | 3249 | { |
8f4e2101 HD |
3250 | struct page *page; |
3251 | spinlock_t *ptl; | |
1da177e4 | 3252 | pte_t entry; |
1da177e4 | 3253 | |
11ac5524 LT |
3254 | pte_unmap(page_table); |
3255 | ||
3256 | /* Check if we need to add a guard page to the stack */ | |
3257 | if (check_stack_guard_page(vma, address) < 0) | |
320b2b8d LT |
3258 | return VM_FAULT_SIGBUS; |
3259 | ||
11ac5524 | 3260 | /* Use the zero-page for reads */ |
62eede62 HD |
3261 | if (!(flags & FAULT_FLAG_WRITE)) { |
3262 | entry = pte_mkspecial(pfn_pte(my_zero_pfn(address), | |
3263 | vma->vm_page_prot)); | |
11ac5524 | 3264 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
a13ea5b7 HD |
3265 | if (!pte_none(*page_table)) |
3266 | goto unlock; | |
3267 | goto setpte; | |
3268 | } | |
3269 | ||
557ed1fa | 3270 | /* Allocate our own private page. */ |
557ed1fa NP |
3271 | if (unlikely(anon_vma_prepare(vma))) |
3272 | goto oom; | |
3273 | page = alloc_zeroed_user_highpage_movable(vma, address); | |
3274 | if (!page) | |
3275 | goto oom; | |
52f37629 MK |
3276 | /* |
3277 | * The memory barrier inside __SetPageUptodate makes sure that | |
3278 | * preceeding stores to the page contents become visible before | |
3279 | * the set_pte_at() write. | |
3280 | */ | |
0ed361de | 3281 | __SetPageUptodate(page); |
8f4e2101 | 3282 | |
2c26fdd7 | 3283 | if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) |
8a9f3ccd BS |
3284 | goto oom_free_page; |
3285 | ||
557ed1fa | 3286 | entry = mk_pte(page, vma->vm_page_prot); |
1ac0cb5d HD |
3287 | if (vma->vm_flags & VM_WRITE) |
3288 | entry = pte_mkwrite(pte_mkdirty(entry)); | |
1da177e4 | 3289 | |
557ed1fa | 3290 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
1c2fb7a4 | 3291 | if (!pte_none(*page_table)) |
557ed1fa | 3292 | goto release; |
9ba69294 | 3293 | |
34e55232 | 3294 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
557ed1fa | 3295 | page_add_new_anon_rmap(page, vma, address); |
a13ea5b7 | 3296 | setpte: |
65500d23 | 3297 | set_pte_at(mm, address, page_table, entry); |
1da177e4 LT |
3298 | |
3299 | /* No need to invalidate - it was non-present before */ | |
4b3073e1 | 3300 | update_mmu_cache(vma, address, page_table); |
65500d23 | 3301 | unlock: |
8f4e2101 | 3302 | pte_unmap_unlock(page_table, ptl); |
83c54070 | 3303 | return 0; |
8f4e2101 | 3304 | release: |
8a9f3ccd | 3305 | mem_cgroup_uncharge_page(page); |
8f4e2101 HD |
3306 | page_cache_release(page); |
3307 | goto unlock; | |
8a9f3ccd | 3308 | oom_free_page: |
6dbf6d3b | 3309 | page_cache_release(page); |
65500d23 | 3310 | oom: |
1da177e4 LT |
3311 | return VM_FAULT_OOM; |
3312 | } | |
3313 | ||
7eae74af KS |
3314 | static int __do_fault(struct vm_area_struct *vma, unsigned long address, |
3315 | pgoff_t pgoff, unsigned int flags, struct page **page) | |
3316 | { | |
3317 | struct vm_fault vmf; | |
3318 | int ret; | |
3319 | ||
3320 | vmf.virtual_address = (void __user *)(address & PAGE_MASK); | |
3321 | vmf.pgoff = pgoff; | |
3322 | vmf.flags = flags; | |
3323 | vmf.page = NULL; | |
3324 | ||
3325 | ret = vma->vm_ops->fault(vma, &vmf); | |
3326 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) | |
3327 | return ret; | |
3328 | ||
3329 | if (unlikely(PageHWPoison(vmf.page))) { | |
3330 | if (ret & VM_FAULT_LOCKED) | |
3331 | unlock_page(vmf.page); | |
3332 | page_cache_release(vmf.page); | |
3333 | return VM_FAULT_HWPOISON; | |
3334 | } | |
3335 | ||
3336 | if (unlikely(!(ret & VM_FAULT_LOCKED))) | |
3337 | lock_page(vmf.page); | |
3338 | else | |
3339 | VM_BUG_ON_PAGE(!PageLocked(vmf.page), vmf.page); | |
3340 | ||
3341 | *page = vmf.page; | |
3342 | return ret; | |
3343 | } | |
3344 | ||
3bb97794 KS |
3345 | static void do_set_pte(struct vm_area_struct *vma, unsigned long address, |
3346 | struct page *page, pte_t *pte, bool write, bool anon) | |
3347 | { | |
3348 | pte_t entry; | |
3349 | ||
3350 | flush_icache_page(vma, page); | |
3351 | entry = mk_pte(page, vma->vm_page_prot); | |
3352 | if (write) | |
3353 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
3354 | else if (pte_file(*pte) && pte_file_soft_dirty(*pte)) | |
3355 | pte_mksoft_dirty(entry); | |
3356 | if (anon) { | |
3357 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); | |
3358 | page_add_new_anon_rmap(page, vma, address); | |
3359 | } else { | |
3360 | inc_mm_counter_fast(vma->vm_mm, MM_FILEPAGES); | |
3361 | page_add_file_rmap(page); | |
3362 | } | |
3363 | set_pte_at(vma->vm_mm, address, pte, entry); | |
3364 | ||
3365 | /* no need to invalidate: a not-present page won't be cached */ | |
3366 | update_mmu_cache(vma, address, pte); | |
3367 | } | |
3368 | ||
e655fb29 KS |
3369 | static int do_read_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
3370 | unsigned long address, pmd_t *pmd, | |
3371 | pgoff_t pgoff, unsigned int flags, pte_t orig_pte) | |
3372 | { | |
3373 | struct page *fault_page; | |
3374 | spinlock_t *ptl; | |
3bb97794 | 3375 | pte_t *pte; |
e655fb29 KS |
3376 | int ret; |
3377 | ||
3378 | ret = __do_fault(vma, address, pgoff, flags, &fault_page); | |
3379 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) | |
3380 | return ret; | |
3381 | ||
3382 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); | |
3383 | if (unlikely(!pte_same(*pte, orig_pte))) { | |
3384 | pte_unmap_unlock(pte, ptl); | |
3385 | unlock_page(fault_page); | |
3386 | page_cache_release(fault_page); | |
3387 | return ret; | |
3388 | } | |
3bb97794 | 3389 | do_set_pte(vma, address, fault_page, pte, false, false); |
e655fb29 KS |
3390 | pte_unmap_unlock(pte, ptl); |
3391 | unlock_page(fault_page); | |
e655fb29 KS |
3392 | return ret; |
3393 | } | |
3394 | ||
ec47c3b9 KS |
3395 | static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
3396 | unsigned long address, pmd_t *pmd, | |
3397 | pgoff_t pgoff, unsigned int flags, pte_t orig_pte) | |
3398 | { | |
3399 | struct page *fault_page, *new_page; | |
3400 | spinlock_t *ptl; | |
3bb97794 | 3401 | pte_t *pte; |
ec47c3b9 KS |
3402 | int ret; |
3403 | ||
3404 | if (unlikely(anon_vma_prepare(vma))) | |
3405 | return VM_FAULT_OOM; | |
3406 | ||
3407 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); | |
3408 | if (!new_page) | |
3409 | return VM_FAULT_OOM; | |
3410 | ||
3411 | if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL)) { | |
3412 | page_cache_release(new_page); | |
3413 | return VM_FAULT_OOM; | |
3414 | } | |
3415 | ||
3416 | ret = __do_fault(vma, address, pgoff, flags, &fault_page); | |
3417 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) | |
3418 | goto uncharge_out; | |
3419 | ||
3420 | copy_user_highpage(new_page, fault_page, address, vma); | |
3421 | __SetPageUptodate(new_page); | |
3422 | ||
3423 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); | |
3424 | if (unlikely(!pte_same(*pte, orig_pte))) { | |
3425 | pte_unmap_unlock(pte, ptl); | |
3426 | unlock_page(fault_page); | |
3427 | page_cache_release(fault_page); | |
3428 | goto uncharge_out; | |
3429 | } | |
3bb97794 | 3430 | do_set_pte(vma, address, new_page, pte, true, true); |
ec47c3b9 KS |
3431 | pte_unmap_unlock(pte, ptl); |
3432 | unlock_page(fault_page); | |
3433 | page_cache_release(fault_page); | |
3434 | return ret; | |
3435 | uncharge_out: | |
3436 | mem_cgroup_uncharge_page(new_page); | |
3437 | page_cache_release(new_page); | |
3438 | return ret; | |
3439 | } | |
3440 | ||
f0c6d4d2 | 3441 | static int do_shared_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
16abfa08 | 3442 | unsigned long address, pmd_t *pmd, |
54cb8821 | 3443 | pgoff_t pgoff, unsigned int flags, pte_t orig_pte) |
1da177e4 | 3444 | { |
f0c6d4d2 KS |
3445 | struct page *fault_page; |
3446 | struct address_space *mapping; | |
8f4e2101 | 3447 | spinlock_t *ptl; |
3bb97794 | 3448 | pte_t *pte; |
f0c6d4d2 | 3449 | int dirtied = 0; |
f0c6d4d2 | 3450 | int ret, tmp; |
1d65f86d | 3451 | |
7eae74af KS |
3452 | ret = __do_fault(vma, address, pgoff, flags, &fault_page); |
3453 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) | |
f0c6d4d2 | 3454 | return ret; |
1da177e4 LT |
3455 | |
3456 | /* | |
f0c6d4d2 KS |
3457 | * Check if the backing address space wants to know that the page is |
3458 | * about to become writable | |
1da177e4 | 3459 | */ |
fb09a464 KS |
3460 | if (vma->vm_ops->page_mkwrite) { |
3461 | unlock_page(fault_page); | |
3462 | tmp = do_page_mkwrite(vma, fault_page, address); | |
3463 | if (unlikely(!tmp || | |
3464 | (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
f0c6d4d2 | 3465 | page_cache_release(fault_page); |
fb09a464 | 3466 | return tmp; |
4294621f | 3467 | } |
fb09a464 KS |
3468 | } |
3469 | ||
f0c6d4d2 KS |
3470 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); |
3471 | if (unlikely(!pte_same(*pte, orig_pte))) { | |
3472 | pte_unmap_unlock(pte, ptl); | |
3473 | unlock_page(fault_page); | |
3474 | page_cache_release(fault_page); | |
3475 | return ret; | |
1da177e4 | 3476 | } |
3bb97794 | 3477 | do_set_pte(vma, address, fault_page, pte, true, false); |
f0c6d4d2 | 3478 | pte_unmap_unlock(pte, ptl); |
b827e496 | 3479 | |
f0c6d4d2 KS |
3480 | if (set_page_dirty(fault_page)) |
3481 | dirtied = 1; | |
3482 | mapping = fault_page->mapping; | |
3483 | unlock_page(fault_page); | |
3484 | if ((dirtied || vma->vm_ops->page_mkwrite) && mapping) { | |
3485 | /* | |
3486 | * Some device drivers do not set page.mapping but still | |
3487 | * dirty their pages | |
3488 | */ | |
3489 | balance_dirty_pages_ratelimited(mapping); | |
d08b3851 | 3490 | } |
d00806b1 | 3491 | |
f0c6d4d2 KS |
3492 | /* file_update_time outside page_lock */ |
3493 | if (vma->vm_file && !vma->vm_ops->page_mkwrite) | |
3494 | file_update_time(vma->vm_file); | |
b827e496 | 3495 | |
1d65f86d | 3496 | return ret; |
54cb8821 | 3497 | } |
d00806b1 | 3498 | |
54cb8821 NP |
3499 | static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
3500 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
30c9f3a9 | 3501 | unsigned int flags, pte_t orig_pte) |
54cb8821 NP |
3502 | { |
3503 | pgoff_t pgoff = (((address & PAGE_MASK) | |
0da7e01f | 3504 | - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
54cb8821 | 3505 | |
16abfa08 | 3506 | pte_unmap(page_table); |
e655fb29 KS |
3507 | if (!(flags & FAULT_FLAG_WRITE)) |
3508 | return do_read_fault(mm, vma, address, pmd, pgoff, flags, | |
3509 | orig_pte); | |
ec47c3b9 KS |
3510 | if (!(vma->vm_flags & VM_SHARED)) |
3511 | return do_cow_fault(mm, vma, address, pmd, pgoff, flags, | |
3512 | orig_pte); | |
f0c6d4d2 | 3513 | return do_shared_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); |
54cb8821 NP |
3514 | } |
3515 | ||
1da177e4 LT |
3516 | /* |
3517 | * Fault of a previously existing named mapping. Repopulate the pte | |
3518 | * from the encoded file_pte if possible. This enables swappable | |
3519 | * nonlinear vmas. | |
8f4e2101 HD |
3520 | * |
3521 | * We enter with non-exclusive mmap_sem (to exclude vma changes, | |
3522 | * but allow concurrent faults), and pte mapped but not yet locked. | |
3523 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 3524 | */ |
d0217ac0 | 3525 | static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
65500d23 | 3526 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
30c9f3a9 | 3527 | unsigned int flags, pte_t orig_pte) |
1da177e4 | 3528 | { |
65500d23 | 3529 | pgoff_t pgoff; |
1da177e4 | 3530 | |
30c9f3a9 LT |
3531 | flags |= FAULT_FLAG_NONLINEAR; |
3532 | ||
4c21e2f2 | 3533 | if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) |
83c54070 | 3534 | return 0; |
1da177e4 | 3535 | |
2509ef26 | 3536 | if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) { |
65500d23 HD |
3537 | /* |
3538 | * Page table corrupted: show pte and kill process. | |
3539 | */ | |
3dc14741 | 3540 | print_bad_pte(vma, address, orig_pte, NULL); |
d99be1a8 | 3541 | return VM_FAULT_SIGBUS; |
65500d23 | 3542 | } |
65500d23 HD |
3543 | |
3544 | pgoff = pte_to_pgoff(orig_pte); | |
e655fb29 KS |
3545 | if (!(flags & FAULT_FLAG_WRITE)) |
3546 | return do_read_fault(mm, vma, address, pmd, pgoff, flags, | |
3547 | orig_pte); | |
ec47c3b9 KS |
3548 | if (!(vma->vm_flags & VM_SHARED)) |
3549 | return do_cow_fault(mm, vma, address, pmd, pgoff, flags, | |
3550 | orig_pte); | |
f0c6d4d2 | 3551 | return do_shared_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); |
1da177e4 LT |
3552 | } |
3553 | ||
b19a9939 | 3554 | static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, |
04bb2f94 RR |
3555 | unsigned long addr, int page_nid, |
3556 | int *flags) | |
9532fec1 MG |
3557 | { |
3558 | get_page(page); | |
3559 | ||
3560 | count_vm_numa_event(NUMA_HINT_FAULTS); | |
04bb2f94 | 3561 | if (page_nid == numa_node_id()) { |
9532fec1 | 3562 | count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); |
04bb2f94 RR |
3563 | *flags |= TNF_FAULT_LOCAL; |
3564 | } | |
9532fec1 MG |
3565 | |
3566 | return mpol_misplaced(page, vma, addr); | |
3567 | } | |
3568 | ||
b19a9939 | 3569 | static int do_numa_page(struct mm_struct *mm, struct vm_area_struct *vma, |
d10e63f2 MG |
3570 | unsigned long addr, pte_t pte, pte_t *ptep, pmd_t *pmd) |
3571 | { | |
4daae3b4 | 3572 | struct page *page = NULL; |
d10e63f2 | 3573 | spinlock_t *ptl; |
8191acbd | 3574 | int page_nid = -1; |
90572890 | 3575 | int last_cpupid; |
cbee9f88 | 3576 | int target_nid; |
b8593bfd | 3577 | bool migrated = false; |
6688cc05 | 3578 | int flags = 0; |
d10e63f2 MG |
3579 | |
3580 | /* | |
3581 | * The "pte" at this point cannot be used safely without | |
3582 | * validation through pte_unmap_same(). It's of NUMA type but | |
3583 | * the pfn may be screwed if the read is non atomic. | |
3584 | * | |
3585 | * ptep_modify_prot_start is not called as this is clearing | |
3586 | * the _PAGE_NUMA bit and it is not really expected that there | |
3587 | * would be concurrent hardware modifications to the PTE. | |
3588 | */ | |
3589 | ptl = pte_lockptr(mm, pmd); | |
3590 | spin_lock(ptl); | |
4daae3b4 MG |
3591 | if (unlikely(!pte_same(*ptep, pte))) { |
3592 | pte_unmap_unlock(ptep, ptl); | |
3593 | goto out; | |
3594 | } | |
3595 | ||
d10e63f2 MG |
3596 | pte = pte_mknonnuma(pte); |
3597 | set_pte_at(mm, addr, ptep, pte); | |
3598 | update_mmu_cache(vma, addr, ptep); | |
3599 | ||
3600 | page = vm_normal_page(vma, addr, pte); | |
3601 | if (!page) { | |
3602 | pte_unmap_unlock(ptep, ptl); | |
3603 | return 0; | |
3604 | } | |
a1a46184 | 3605 | BUG_ON(is_zero_pfn(page_to_pfn(page))); |
d10e63f2 | 3606 | |
6688cc05 PZ |
3607 | /* |
3608 | * Avoid grouping on DSO/COW pages in specific and RO pages | |
3609 | * in general, RO pages shouldn't hurt as much anyway since | |
3610 | * they can be in shared cache state. | |
3611 | */ | |
3612 | if (!pte_write(pte)) | |
3613 | flags |= TNF_NO_GROUP; | |
3614 | ||
dabe1d99 RR |
3615 | /* |
3616 | * Flag if the page is shared between multiple address spaces. This | |
3617 | * is later used when determining whether to group tasks together | |
3618 | */ | |
3619 | if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) | |
3620 | flags |= TNF_SHARED; | |
3621 | ||
90572890 | 3622 | last_cpupid = page_cpupid_last(page); |
8191acbd | 3623 | page_nid = page_to_nid(page); |
04bb2f94 | 3624 | target_nid = numa_migrate_prep(page, vma, addr, page_nid, &flags); |
d10e63f2 | 3625 | pte_unmap_unlock(ptep, ptl); |
4daae3b4 | 3626 | if (target_nid == -1) { |
4daae3b4 MG |
3627 | put_page(page); |
3628 | goto out; | |
3629 | } | |
3630 | ||
3631 | /* Migrate to the requested node */ | |
1bc115d8 | 3632 | migrated = migrate_misplaced_page(page, vma, target_nid); |
6688cc05 | 3633 | if (migrated) { |
8191acbd | 3634 | page_nid = target_nid; |
6688cc05 PZ |
3635 | flags |= TNF_MIGRATED; |
3636 | } | |
4daae3b4 MG |
3637 | |
3638 | out: | |
8191acbd | 3639 | if (page_nid != -1) |
6688cc05 | 3640 | task_numa_fault(last_cpupid, page_nid, 1, flags); |
d10e63f2 MG |
3641 | return 0; |
3642 | } | |
3643 | ||
1da177e4 LT |
3644 | /* |
3645 | * These routines also need to handle stuff like marking pages dirty | |
3646 | * and/or accessed for architectures that don't do it in hardware (most | |
3647 | * RISC architectures). The early dirtying is also good on the i386. | |
3648 | * | |
3649 | * There is also a hook called "update_mmu_cache()" that architectures | |
3650 | * with external mmu caches can use to update those (ie the Sparc or | |
3651 | * PowerPC hashed page tables that act as extended TLBs). | |
3652 | * | |
c74df32c HD |
3653 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
3654 | * but allow concurrent faults), and pte mapped but not yet locked. | |
3655 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 3656 | */ |
c0292554 | 3657 | static int handle_pte_fault(struct mm_struct *mm, |
71e3aac0 AA |
3658 | struct vm_area_struct *vma, unsigned long address, |
3659 | pte_t *pte, pmd_t *pmd, unsigned int flags) | |
1da177e4 LT |
3660 | { |
3661 | pte_t entry; | |
8f4e2101 | 3662 | spinlock_t *ptl; |
1da177e4 | 3663 | |
8dab5241 | 3664 | entry = *pte; |
1da177e4 | 3665 | if (!pte_present(entry)) { |
65500d23 | 3666 | if (pte_none(entry)) { |
f4b81804 | 3667 | if (vma->vm_ops) { |
3c18ddd1 | 3668 | if (likely(vma->vm_ops->fault)) |
54cb8821 | 3669 | return do_linear_fault(mm, vma, address, |
30c9f3a9 | 3670 | pte, pmd, flags, entry); |
f4b81804 JS |
3671 | } |
3672 | return do_anonymous_page(mm, vma, address, | |
30c9f3a9 | 3673 | pte, pmd, flags); |
65500d23 | 3674 | } |
1da177e4 | 3675 | if (pte_file(entry)) |
d0217ac0 | 3676 | return do_nonlinear_fault(mm, vma, address, |
30c9f3a9 | 3677 | pte, pmd, flags, entry); |
65500d23 | 3678 | return do_swap_page(mm, vma, address, |
30c9f3a9 | 3679 | pte, pmd, flags, entry); |
1da177e4 LT |
3680 | } |
3681 | ||
d10e63f2 MG |
3682 | if (pte_numa(entry)) |
3683 | return do_numa_page(mm, vma, address, entry, pte, pmd); | |
3684 | ||
4c21e2f2 | 3685 | ptl = pte_lockptr(mm, pmd); |
8f4e2101 HD |
3686 | spin_lock(ptl); |
3687 | if (unlikely(!pte_same(*pte, entry))) | |
3688 | goto unlock; | |
30c9f3a9 | 3689 | if (flags & FAULT_FLAG_WRITE) { |
1da177e4 | 3690 | if (!pte_write(entry)) |
8f4e2101 HD |
3691 | return do_wp_page(mm, vma, address, |
3692 | pte, pmd, ptl, entry); | |
1da177e4 LT |
3693 | entry = pte_mkdirty(entry); |
3694 | } | |
3695 | entry = pte_mkyoung(entry); | |
30c9f3a9 | 3696 | if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) { |
4b3073e1 | 3697 | update_mmu_cache(vma, address, pte); |
1a44e149 AA |
3698 | } else { |
3699 | /* | |
3700 | * This is needed only for protection faults but the arch code | |
3701 | * is not yet telling us if this is a protection fault or not. | |
3702 | * This still avoids useless tlb flushes for .text page faults | |
3703 | * with threads. | |
3704 | */ | |
30c9f3a9 | 3705 | if (flags & FAULT_FLAG_WRITE) |
61c77326 | 3706 | flush_tlb_fix_spurious_fault(vma, address); |
1a44e149 | 3707 | } |
8f4e2101 HD |
3708 | unlock: |
3709 | pte_unmap_unlock(pte, ptl); | |
83c54070 | 3710 | return 0; |
1da177e4 LT |
3711 | } |
3712 | ||
3713 | /* | |
3714 | * By the time we get here, we already hold the mm semaphore | |
3715 | */ | |
519e5247 JW |
3716 | static int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
3717 | unsigned long address, unsigned int flags) | |
1da177e4 LT |
3718 | { |
3719 | pgd_t *pgd; | |
3720 | pud_t *pud; | |
3721 | pmd_t *pmd; | |
3722 | pte_t *pte; | |
3723 | ||
ac9b9c66 | 3724 | if (unlikely(is_vm_hugetlb_page(vma))) |
30c9f3a9 | 3725 | return hugetlb_fault(mm, vma, address, flags); |
1da177e4 | 3726 | |
1da177e4 | 3727 | pgd = pgd_offset(mm, address); |
1da177e4 LT |
3728 | pud = pud_alloc(mm, pgd, address); |
3729 | if (!pud) | |
c74df32c | 3730 | return VM_FAULT_OOM; |
1da177e4 LT |
3731 | pmd = pmd_alloc(mm, pud, address); |
3732 | if (!pmd) | |
c74df32c | 3733 | return VM_FAULT_OOM; |
71e3aac0 | 3734 | if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) { |
c0292554 | 3735 | int ret = VM_FAULT_FALLBACK; |
71e3aac0 | 3736 | if (!vma->vm_ops) |
c0292554 KS |
3737 | ret = do_huge_pmd_anonymous_page(mm, vma, address, |
3738 | pmd, flags); | |
3739 | if (!(ret & VM_FAULT_FALLBACK)) | |
3740 | return ret; | |
71e3aac0 AA |
3741 | } else { |
3742 | pmd_t orig_pmd = *pmd; | |
1f1d06c3 DR |
3743 | int ret; |
3744 | ||
71e3aac0 AA |
3745 | barrier(); |
3746 | if (pmd_trans_huge(orig_pmd)) { | |
a1dd450b WD |
3747 | unsigned int dirty = flags & FAULT_FLAG_WRITE; |
3748 | ||
e53289c0 LT |
3749 | /* |
3750 | * If the pmd is splitting, return and retry the | |
3751 | * the fault. Alternative: wait until the split | |
3752 | * is done, and goto retry. | |
3753 | */ | |
3754 | if (pmd_trans_splitting(orig_pmd)) | |
3755 | return 0; | |
3756 | ||
3d59eebc | 3757 | if (pmd_numa(orig_pmd)) |
4daae3b4 | 3758 | return do_huge_pmd_numa_page(mm, vma, address, |
d10e63f2 MG |
3759 | orig_pmd, pmd); |
3760 | ||
3d59eebc | 3761 | if (dirty && !pmd_write(orig_pmd)) { |
1f1d06c3 DR |
3762 | ret = do_huge_pmd_wp_page(mm, vma, address, pmd, |
3763 | orig_pmd); | |
9845cbbd KS |
3764 | if (!(ret & VM_FAULT_FALLBACK)) |
3765 | return ret; | |
a1dd450b WD |
3766 | } else { |
3767 | huge_pmd_set_accessed(mm, vma, address, pmd, | |
3768 | orig_pmd, dirty); | |
9845cbbd | 3769 | return 0; |
1f1d06c3 | 3770 | } |
71e3aac0 AA |
3771 | } |
3772 | } | |
3773 | ||
0f19c179 MG |
3774 | /* THP should already have been handled */ |
3775 | BUG_ON(pmd_numa(*pmd)); | |
d10e63f2 | 3776 | |
71e3aac0 AA |
3777 | /* |
3778 | * Use __pte_alloc instead of pte_alloc_map, because we can't | |
3779 | * run pte_offset_map on the pmd, if an huge pmd could | |
3780 | * materialize from under us from a different thread. | |
3781 | */ | |
4fd01770 MG |
3782 | if (unlikely(pmd_none(*pmd)) && |
3783 | unlikely(__pte_alloc(mm, vma, pmd, address))) | |
c74df32c | 3784 | return VM_FAULT_OOM; |
71e3aac0 AA |
3785 | /* if an huge pmd materialized from under us just retry later */ |
3786 | if (unlikely(pmd_trans_huge(*pmd))) | |
3787 | return 0; | |
3788 | /* | |
3789 | * A regular pmd is established and it can't morph into a huge pmd | |
3790 | * from under us anymore at this point because we hold the mmap_sem | |
3791 | * read mode and khugepaged takes it in write mode. So now it's | |
3792 | * safe to run pte_offset_map(). | |
3793 | */ | |
3794 | pte = pte_offset_map(pmd, address); | |
1da177e4 | 3795 | |
30c9f3a9 | 3796 | return handle_pte_fault(mm, vma, address, pte, pmd, flags); |
1da177e4 LT |
3797 | } |
3798 | ||
519e5247 JW |
3799 | int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
3800 | unsigned long address, unsigned int flags) | |
3801 | { | |
3802 | int ret; | |
3803 | ||
3804 | __set_current_state(TASK_RUNNING); | |
3805 | ||
3806 | count_vm_event(PGFAULT); | |
3807 | mem_cgroup_count_vm_event(mm, PGFAULT); | |
3808 | ||
3809 | /* do counter updates before entering really critical section. */ | |
3810 | check_sync_rss_stat(current); | |
3811 | ||
3812 | /* | |
3813 | * Enable the memcg OOM handling for faults triggered in user | |
3814 | * space. Kernel faults are handled more gracefully. | |
3815 | */ | |
3816 | if (flags & FAULT_FLAG_USER) | |
49426420 | 3817 | mem_cgroup_oom_enable(); |
519e5247 JW |
3818 | |
3819 | ret = __handle_mm_fault(mm, vma, address, flags); | |
3820 | ||
49426420 JW |
3821 | if (flags & FAULT_FLAG_USER) { |
3822 | mem_cgroup_oom_disable(); | |
3823 | /* | |
3824 | * The task may have entered a memcg OOM situation but | |
3825 | * if the allocation error was handled gracefully (no | |
3826 | * VM_FAULT_OOM), there is no need to kill anything. | |
3827 | * Just clean up the OOM state peacefully. | |
3828 | */ | |
3829 | if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) | |
3830 | mem_cgroup_oom_synchronize(false); | |
3831 | } | |
3812c8c8 | 3832 | |
519e5247 JW |
3833 | return ret; |
3834 | } | |
3835 | ||
1da177e4 LT |
3836 | #ifndef __PAGETABLE_PUD_FOLDED |
3837 | /* | |
3838 | * Allocate page upper directory. | |
872fec16 | 3839 | * We've already handled the fast-path in-line. |
1da177e4 | 3840 | */ |
1bb3630e | 3841 | int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) |
1da177e4 | 3842 | { |
c74df32c HD |
3843 | pud_t *new = pud_alloc_one(mm, address); |
3844 | if (!new) | |
1bb3630e | 3845 | return -ENOMEM; |
1da177e4 | 3846 | |
362a61ad NP |
3847 | smp_wmb(); /* See comment in __pte_alloc */ |
3848 | ||
872fec16 | 3849 | spin_lock(&mm->page_table_lock); |
1bb3630e | 3850 | if (pgd_present(*pgd)) /* Another has populated it */ |
5e541973 | 3851 | pud_free(mm, new); |
1bb3630e HD |
3852 | else |
3853 | pgd_populate(mm, pgd, new); | |
c74df32c | 3854 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 3855 | return 0; |
1da177e4 LT |
3856 | } |
3857 | #endif /* __PAGETABLE_PUD_FOLDED */ | |
3858 | ||
3859 | #ifndef __PAGETABLE_PMD_FOLDED | |
3860 | /* | |
3861 | * Allocate page middle directory. | |
872fec16 | 3862 | * We've already handled the fast-path in-line. |
1da177e4 | 3863 | */ |
1bb3630e | 3864 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
1da177e4 | 3865 | { |
c74df32c HD |
3866 | pmd_t *new = pmd_alloc_one(mm, address); |
3867 | if (!new) | |
1bb3630e | 3868 | return -ENOMEM; |
1da177e4 | 3869 | |
362a61ad NP |
3870 | smp_wmb(); /* See comment in __pte_alloc */ |
3871 | ||
872fec16 | 3872 | spin_lock(&mm->page_table_lock); |
1da177e4 | 3873 | #ifndef __ARCH_HAS_4LEVEL_HACK |
1bb3630e | 3874 | if (pud_present(*pud)) /* Another has populated it */ |
5e541973 | 3875 | pmd_free(mm, new); |
1bb3630e HD |
3876 | else |
3877 | pud_populate(mm, pud, new); | |
1da177e4 | 3878 | #else |
1bb3630e | 3879 | if (pgd_present(*pud)) /* Another has populated it */ |
5e541973 | 3880 | pmd_free(mm, new); |
1bb3630e HD |
3881 | else |
3882 | pgd_populate(mm, pud, new); | |
1da177e4 | 3883 | #endif /* __ARCH_HAS_4LEVEL_HACK */ |
c74df32c | 3884 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 3885 | return 0; |
e0f39591 | 3886 | } |
1da177e4 LT |
3887 | #endif /* __PAGETABLE_PMD_FOLDED */ |
3888 | ||
1da177e4 LT |
3889 | #if !defined(__HAVE_ARCH_GATE_AREA) |
3890 | ||
3891 | #if defined(AT_SYSINFO_EHDR) | |
5ce7852c | 3892 | static struct vm_area_struct gate_vma; |
1da177e4 LT |
3893 | |
3894 | static int __init gate_vma_init(void) | |
3895 | { | |
3896 | gate_vma.vm_mm = NULL; | |
3897 | gate_vma.vm_start = FIXADDR_USER_START; | |
3898 | gate_vma.vm_end = FIXADDR_USER_END; | |
b6558c4a RM |
3899 | gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC; |
3900 | gate_vma.vm_page_prot = __P101; | |
909af768 | 3901 | |
1da177e4 LT |
3902 | return 0; |
3903 | } | |
3904 | __initcall(gate_vma_init); | |
3905 | #endif | |
3906 | ||
31db58b3 | 3907 | struct vm_area_struct *get_gate_vma(struct mm_struct *mm) |
1da177e4 LT |
3908 | { |
3909 | #ifdef AT_SYSINFO_EHDR | |
3910 | return &gate_vma; | |
3911 | #else | |
3912 | return NULL; | |
3913 | #endif | |
3914 | } | |
3915 | ||
cae5d390 | 3916 | int in_gate_area_no_mm(unsigned long addr) |
1da177e4 LT |
3917 | { |
3918 | #ifdef AT_SYSINFO_EHDR | |
3919 | if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END)) | |
3920 | return 1; | |
3921 | #endif | |
3922 | return 0; | |
3923 | } | |
3924 | ||
3925 | #endif /* __HAVE_ARCH_GATE_AREA */ | |
0ec76a11 | 3926 | |
1b36ba81 | 3927 | static int __follow_pte(struct mm_struct *mm, unsigned long address, |
f8ad0f49 JW |
3928 | pte_t **ptepp, spinlock_t **ptlp) |
3929 | { | |
3930 | pgd_t *pgd; | |
3931 | pud_t *pud; | |
3932 | pmd_t *pmd; | |
3933 | pte_t *ptep; | |
3934 | ||
3935 | pgd = pgd_offset(mm, address); | |
3936 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
3937 | goto out; | |
3938 | ||
3939 | pud = pud_offset(pgd, address); | |
3940 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) | |
3941 | goto out; | |
3942 | ||
3943 | pmd = pmd_offset(pud, address); | |
f66055ab | 3944 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
f8ad0f49 JW |
3945 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) |
3946 | goto out; | |
3947 | ||
3948 | /* We cannot handle huge page PFN maps. Luckily they don't exist. */ | |
3949 | if (pmd_huge(*pmd)) | |
3950 | goto out; | |
3951 | ||
3952 | ptep = pte_offset_map_lock(mm, pmd, address, ptlp); | |
3953 | if (!ptep) | |
3954 | goto out; | |
3955 | if (!pte_present(*ptep)) | |
3956 | goto unlock; | |
3957 | *ptepp = ptep; | |
3958 | return 0; | |
3959 | unlock: | |
3960 | pte_unmap_unlock(ptep, *ptlp); | |
3961 | out: | |
3962 | return -EINVAL; | |
3963 | } | |
3964 | ||
1b36ba81 NK |
3965 | static inline int follow_pte(struct mm_struct *mm, unsigned long address, |
3966 | pte_t **ptepp, spinlock_t **ptlp) | |
3967 | { | |
3968 | int res; | |
3969 | ||
3970 | /* (void) is needed to make gcc happy */ | |
3971 | (void) __cond_lock(*ptlp, | |
3972 | !(res = __follow_pte(mm, address, ptepp, ptlp))); | |
3973 | return res; | |
3974 | } | |
3975 | ||
3b6748e2 JW |
3976 | /** |
3977 | * follow_pfn - look up PFN at a user virtual address | |
3978 | * @vma: memory mapping | |
3979 | * @address: user virtual address | |
3980 | * @pfn: location to store found PFN | |
3981 | * | |
3982 | * Only IO mappings and raw PFN mappings are allowed. | |
3983 | * | |
3984 | * Returns zero and the pfn at @pfn on success, -ve otherwise. | |
3985 | */ | |
3986 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
3987 | unsigned long *pfn) | |
3988 | { | |
3989 | int ret = -EINVAL; | |
3990 | spinlock_t *ptl; | |
3991 | pte_t *ptep; | |
3992 | ||
3993 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
3994 | return ret; | |
3995 | ||
3996 | ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); | |
3997 | if (ret) | |
3998 | return ret; | |
3999 | *pfn = pte_pfn(*ptep); | |
4000 | pte_unmap_unlock(ptep, ptl); | |
4001 | return 0; | |
4002 | } | |
4003 | EXPORT_SYMBOL(follow_pfn); | |
4004 | ||
28b2ee20 | 4005 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
d87fe660 | 4006 | int follow_phys(struct vm_area_struct *vma, |
4007 | unsigned long address, unsigned int flags, | |
4008 | unsigned long *prot, resource_size_t *phys) | |
28b2ee20 | 4009 | { |
03668a4d | 4010 | int ret = -EINVAL; |
28b2ee20 RR |
4011 | pte_t *ptep, pte; |
4012 | spinlock_t *ptl; | |
28b2ee20 | 4013 | |
d87fe660 | 4014 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
4015 | goto out; | |
28b2ee20 | 4016 | |
03668a4d | 4017 | if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) |
d87fe660 | 4018 | goto out; |
28b2ee20 | 4019 | pte = *ptep; |
03668a4d | 4020 | |
28b2ee20 RR |
4021 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
4022 | goto unlock; | |
28b2ee20 RR |
4023 | |
4024 | *prot = pgprot_val(pte_pgprot(pte)); | |
03668a4d | 4025 | *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; |
28b2ee20 | 4026 | |
03668a4d | 4027 | ret = 0; |
28b2ee20 RR |
4028 | unlock: |
4029 | pte_unmap_unlock(ptep, ptl); | |
4030 | out: | |
d87fe660 | 4031 | return ret; |
28b2ee20 RR |
4032 | } |
4033 | ||
4034 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, | |
4035 | void *buf, int len, int write) | |
4036 | { | |
4037 | resource_size_t phys_addr; | |
4038 | unsigned long prot = 0; | |
2bc7273b | 4039 | void __iomem *maddr; |
28b2ee20 RR |
4040 | int offset = addr & (PAGE_SIZE-1); |
4041 | ||
d87fe660 | 4042 | if (follow_phys(vma, addr, write, &prot, &phys_addr)) |
28b2ee20 RR |
4043 | return -EINVAL; |
4044 | ||
4045 | maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot); | |
4046 | if (write) | |
4047 | memcpy_toio(maddr + offset, buf, len); | |
4048 | else | |
4049 | memcpy_fromio(buf, maddr + offset, len); | |
4050 | iounmap(maddr); | |
4051 | ||
4052 | return len; | |
4053 | } | |
5a73633e | 4054 | EXPORT_SYMBOL_GPL(generic_access_phys); |
28b2ee20 RR |
4055 | #endif |
4056 | ||
0ec76a11 | 4057 | /* |
206cb636 SW |
4058 | * Access another process' address space as given in mm. If non-NULL, use the |
4059 | * given task for page fault accounting. | |
0ec76a11 | 4060 | */ |
206cb636 SW |
4061 | static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, |
4062 | unsigned long addr, void *buf, int len, int write) | |
0ec76a11 | 4063 | { |
0ec76a11 | 4064 | struct vm_area_struct *vma; |
0ec76a11 DH |
4065 | void *old_buf = buf; |
4066 | ||
0ec76a11 | 4067 | down_read(&mm->mmap_sem); |
183ff22b | 4068 | /* ignore errors, just check how much was successfully transferred */ |
0ec76a11 DH |
4069 | while (len) { |
4070 | int bytes, ret, offset; | |
4071 | void *maddr; | |
28b2ee20 | 4072 | struct page *page = NULL; |
0ec76a11 DH |
4073 | |
4074 | ret = get_user_pages(tsk, mm, addr, 1, | |
4075 | write, 1, &page, &vma); | |
28b2ee20 RR |
4076 | if (ret <= 0) { |
4077 | /* | |
4078 | * Check if this is a VM_IO | VM_PFNMAP VMA, which | |
4079 | * we can access using slightly different code. | |
4080 | */ | |
4081 | #ifdef CONFIG_HAVE_IOREMAP_PROT | |
4082 | vma = find_vma(mm, addr); | |
fe936dfc | 4083 | if (!vma || vma->vm_start > addr) |
28b2ee20 RR |
4084 | break; |
4085 | if (vma->vm_ops && vma->vm_ops->access) | |
4086 | ret = vma->vm_ops->access(vma, addr, buf, | |
4087 | len, write); | |
4088 | if (ret <= 0) | |
4089 | #endif | |
4090 | break; | |
4091 | bytes = ret; | |
0ec76a11 | 4092 | } else { |
28b2ee20 RR |
4093 | bytes = len; |
4094 | offset = addr & (PAGE_SIZE-1); | |
4095 | if (bytes > PAGE_SIZE-offset) | |
4096 | bytes = PAGE_SIZE-offset; | |
4097 | ||
4098 | maddr = kmap(page); | |
4099 | if (write) { | |
4100 | copy_to_user_page(vma, page, addr, | |
4101 | maddr + offset, buf, bytes); | |
4102 | set_page_dirty_lock(page); | |
4103 | } else { | |
4104 | copy_from_user_page(vma, page, addr, | |
4105 | buf, maddr + offset, bytes); | |
4106 | } | |
4107 | kunmap(page); | |
4108 | page_cache_release(page); | |
0ec76a11 | 4109 | } |
0ec76a11 DH |
4110 | len -= bytes; |
4111 | buf += bytes; | |
4112 | addr += bytes; | |
4113 | } | |
4114 | up_read(&mm->mmap_sem); | |
0ec76a11 DH |
4115 | |
4116 | return buf - old_buf; | |
4117 | } | |
03252919 | 4118 | |
5ddd36b9 | 4119 | /** |
ae91dbfc | 4120 | * access_remote_vm - access another process' address space |
5ddd36b9 SW |
4121 | * @mm: the mm_struct of the target address space |
4122 | * @addr: start address to access | |
4123 | * @buf: source or destination buffer | |
4124 | * @len: number of bytes to transfer | |
4125 | * @write: whether the access is a write | |
4126 | * | |
4127 | * The caller must hold a reference on @mm. | |
4128 | */ | |
4129 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, | |
4130 | void *buf, int len, int write) | |
4131 | { | |
4132 | return __access_remote_vm(NULL, mm, addr, buf, len, write); | |
4133 | } | |
4134 | ||
206cb636 SW |
4135 | /* |
4136 | * Access another process' address space. | |
4137 | * Source/target buffer must be kernel space, | |
4138 | * Do not walk the page table directly, use get_user_pages | |
4139 | */ | |
4140 | int access_process_vm(struct task_struct *tsk, unsigned long addr, | |
4141 | void *buf, int len, int write) | |
4142 | { | |
4143 | struct mm_struct *mm; | |
4144 | int ret; | |
4145 | ||
4146 | mm = get_task_mm(tsk); | |
4147 | if (!mm) | |
4148 | return 0; | |
4149 | ||
4150 | ret = __access_remote_vm(tsk, mm, addr, buf, len, write); | |
4151 | mmput(mm); | |
4152 | ||
4153 | return ret; | |
4154 | } | |
4155 | ||
03252919 AK |
4156 | /* |
4157 | * Print the name of a VMA. | |
4158 | */ | |
4159 | void print_vma_addr(char *prefix, unsigned long ip) | |
4160 | { | |
4161 | struct mm_struct *mm = current->mm; | |
4162 | struct vm_area_struct *vma; | |
4163 | ||
e8bff74a IM |
4164 | /* |
4165 | * Do not print if we are in atomic | |
4166 | * contexts (in exception stacks, etc.): | |
4167 | */ | |
4168 | if (preempt_count()) | |
4169 | return; | |
4170 | ||
03252919 AK |
4171 | down_read(&mm->mmap_sem); |
4172 | vma = find_vma(mm, ip); | |
4173 | if (vma && vma->vm_file) { | |
4174 | struct file *f = vma->vm_file; | |
4175 | char *buf = (char *)__get_free_page(GFP_KERNEL); | |
4176 | if (buf) { | |
2fbc57c5 | 4177 | char *p; |
03252919 | 4178 | |
cf28b486 | 4179 | p = d_path(&f->f_path, buf, PAGE_SIZE); |
03252919 AK |
4180 | if (IS_ERR(p)) |
4181 | p = "?"; | |
2fbc57c5 | 4182 | printk("%s%s[%lx+%lx]", prefix, kbasename(p), |
03252919 AK |
4183 | vma->vm_start, |
4184 | vma->vm_end - vma->vm_start); | |
4185 | free_page((unsigned long)buf); | |
4186 | } | |
4187 | } | |
51a07e50 | 4188 | up_read(&mm->mmap_sem); |
03252919 | 4189 | } |
3ee1afa3 | 4190 | |
662bbcb2 | 4191 | #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) |
3ee1afa3 NP |
4192 | void might_fault(void) |
4193 | { | |
95156f00 PZ |
4194 | /* |
4195 | * Some code (nfs/sunrpc) uses socket ops on kernel memory while | |
4196 | * holding the mmap_sem, this is safe because kernel memory doesn't | |
4197 | * get paged out, therefore we'll never actually fault, and the | |
4198 | * below annotations will generate false positives. | |
4199 | */ | |
4200 | if (segment_eq(get_fs(), KERNEL_DS)) | |
4201 | return; | |
4202 | ||
3ee1afa3 NP |
4203 | /* |
4204 | * it would be nicer only to annotate paths which are not under | |
4205 | * pagefault_disable, however that requires a larger audit and | |
4206 | * providing helpers like get_user_atomic. | |
4207 | */ | |
662bbcb2 MT |
4208 | if (in_atomic()) |
4209 | return; | |
4210 | ||
4211 | __might_sleep(__FILE__, __LINE__, 0); | |
4212 | ||
4213 | if (current->mm) | |
3ee1afa3 NP |
4214 | might_lock_read(¤t->mm->mmap_sem); |
4215 | } | |
4216 | EXPORT_SYMBOL(might_fault); | |
4217 | #endif | |
47ad8475 AA |
4218 | |
4219 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) | |
4220 | static void clear_gigantic_page(struct page *page, | |
4221 | unsigned long addr, | |
4222 | unsigned int pages_per_huge_page) | |
4223 | { | |
4224 | int i; | |
4225 | struct page *p = page; | |
4226 | ||
4227 | might_sleep(); | |
4228 | for (i = 0; i < pages_per_huge_page; | |
4229 | i++, p = mem_map_next(p, page, i)) { | |
4230 | cond_resched(); | |
4231 | clear_user_highpage(p, addr + i * PAGE_SIZE); | |
4232 | } | |
4233 | } | |
4234 | void clear_huge_page(struct page *page, | |
4235 | unsigned long addr, unsigned int pages_per_huge_page) | |
4236 | { | |
4237 | int i; | |
4238 | ||
4239 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
4240 | clear_gigantic_page(page, addr, pages_per_huge_page); | |
4241 | return; | |
4242 | } | |
4243 | ||
4244 | might_sleep(); | |
4245 | for (i = 0; i < pages_per_huge_page; i++) { | |
4246 | cond_resched(); | |
4247 | clear_user_highpage(page + i, addr + i * PAGE_SIZE); | |
4248 | } | |
4249 | } | |
4250 | ||
4251 | static void copy_user_gigantic_page(struct page *dst, struct page *src, | |
4252 | unsigned long addr, | |
4253 | struct vm_area_struct *vma, | |
4254 | unsigned int pages_per_huge_page) | |
4255 | { | |
4256 | int i; | |
4257 | struct page *dst_base = dst; | |
4258 | struct page *src_base = src; | |
4259 | ||
4260 | for (i = 0; i < pages_per_huge_page; ) { | |
4261 | cond_resched(); | |
4262 | copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); | |
4263 | ||
4264 | i++; | |
4265 | dst = mem_map_next(dst, dst_base, i); | |
4266 | src = mem_map_next(src, src_base, i); | |
4267 | } | |
4268 | } | |
4269 | ||
4270 | void copy_user_huge_page(struct page *dst, struct page *src, | |
4271 | unsigned long addr, struct vm_area_struct *vma, | |
4272 | unsigned int pages_per_huge_page) | |
4273 | { | |
4274 | int i; | |
4275 | ||
4276 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
4277 | copy_user_gigantic_page(dst, src, addr, vma, | |
4278 | pages_per_huge_page); | |
4279 | return; | |
4280 | } | |
4281 | ||
4282 | might_sleep(); | |
4283 | for (i = 0; i < pages_per_huge_page; i++) { | |
4284 | cond_resched(); | |
4285 | copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma); | |
4286 | } | |
4287 | } | |
4288 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ | |
49076ec2 | 4289 | |
40b64acd | 4290 | #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS |
b35f1819 KS |
4291 | |
4292 | static struct kmem_cache *page_ptl_cachep; | |
4293 | ||
4294 | void __init ptlock_cache_init(void) | |
4295 | { | |
4296 | page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, | |
4297 | SLAB_PANIC, NULL); | |
4298 | } | |
4299 | ||
539edb58 | 4300 | bool ptlock_alloc(struct page *page) |
49076ec2 KS |
4301 | { |
4302 | spinlock_t *ptl; | |
4303 | ||
b35f1819 | 4304 | ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); |
49076ec2 KS |
4305 | if (!ptl) |
4306 | return false; | |
539edb58 | 4307 | page->ptl = ptl; |
49076ec2 KS |
4308 | return true; |
4309 | } | |
4310 | ||
539edb58 | 4311 | void ptlock_free(struct page *page) |
49076ec2 | 4312 | { |
b35f1819 | 4313 | kmem_cache_free(page_ptl_cachep, page->ptl); |
49076ec2 KS |
4314 | } |
4315 | #endif |