]>
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> |
01c8f1c4 | 53 | #include <linux/pfn_t.h> |
edc79b2a | 54 | #include <linux/writeback.h> |
8a9f3ccd | 55 | #include <linux/memcontrol.h> |
cddb8a5c | 56 | #include <linux/mmu_notifier.h> |
3dc14741 HD |
57 | #include <linux/kallsyms.h> |
58 | #include <linux/swapops.h> | |
59 | #include <linux/elf.h> | |
5a0e3ad6 | 60 | #include <linux/gfp.h> |
4daae3b4 | 61 | #include <linux/migrate.h> |
2fbc57c5 | 62 | #include <linux/string.h> |
0abdd7a8 | 63 | #include <linux/dma-debug.h> |
1592eef0 | 64 | #include <linux/debugfs.h> |
6b251fc9 | 65 | #include <linux/userfaultfd_k.h> |
1da177e4 | 66 | |
6952b61d | 67 | #include <asm/io.h> |
1da177e4 LT |
68 | #include <asm/pgalloc.h> |
69 | #include <asm/uaccess.h> | |
70 | #include <asm/tlb.h> | |
71 | #include <asm/tlbflush.h> | |
72 | #include <asm/pgtable.h> | |
73 | ||
42b77728 JB |
74 | #include "internal.h" |
75 | ||
90572890 PZ |
76 | #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS |
77 | #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid. | |
75980e97 PZ |
78 | #endif |
79 | ||
d41dee36 | 80 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
81 | /* use the per-pgdat data instead for discontigmem - mbligh */ |
82 | unsigned long max_mapnr; | |
83 | struct page *mem_map; | |
84 | ||
85 | EXPORT_SYMBOL(max_mapnr); | |
86 | EXPORT_SYMBOL(mem_map); | |
87 | #endif | |
88 | ||
1da177e4 LT |
89 | /* |
90 | * A number of key systems in x86 including ioremap() rely on the assumption | |
91 | * that high_memory defines the upper bound on direct map memory, then end | |
92 | * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and | |
93 | * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL | |
94 | * and ZONE_HIGHMEM. | |
95 | */ | |
96 | void * high_memory; | |
1da177e4 | 97 | |
1da177e4 | 98 | EXPORT_SYMBOL(high_memory); |
1da177e4 | 99 | |
32a93233 IM |
100 | /* |
101 | * Randomize the address space (stacks, mmaps, brk, etc.). | |
102 | * | |
103 | * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, | |
104 | * as ancient (libc5 based) binaries can segfault. ) | |
105 | */ | |
106 | int randomize_va_space __read_mostly = | |
107 | #ifdef CONFIG_COMPAT_BRK | |
108 | 1; | |
109 | #else | |
110 | 2; | |
111 | #endif | |
a62eaf15 AK |
112 | |
113 | static int __init disable_randmaps(char *s) | |
114 | { | |
115 | randomize_va_space = 0; | |
9b41046c | 116 | return 1; |
a62eaf15 AK |
117 | } |
118 | __setup("norandmaps", disable_randmaps); | |
119 | ||
62eede62 | 120 | unsigned long zero_pfn __read_mostly; |
03f6462a | 121 | unsigned long highest_memmap_pfn __read_mostly; |
a13ea5b7 | 122 | |
0b70068e AB |
123 | EXPORT_SYMBOL(zero_pfn); |
124 | ||
a13ea5b7 HD |
125 | /* |
126 | * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() | |
127 | */ | |
128 | static int __init init_zero_pfn(void) | |
129 | { | |
130 | zero_pfn = page_to_pfn(ZERO_PAGE(0)); | |
131 | return 0; | |
132 | } | |
133 | core_initcall(init_zero_pfn); | |
a62eaf15 | 134 | |
d559db08 | 135 | |
34e55232 KH |
136 | #if defined(SPLIT_RSS_COUNTING) |
137 | ||
ea48cf78 | 138 | void sync_mm_rss(struct mm_struct *mm) |
34e55232 KH |
139 | { |
140 | int i; | |
141 | ||
142 | for (i = 0; i < NR_MM_COUNTERS; i++) { | |
05af2e10 DR |
143 | if (current->rss_stat.count[i]) { |
144 | add_mm_counter(mm, i, current->rss_stat.count[i]); | |
145 | current->rss_stat.count[i] = 0; | |
34e55232 KH |
146 | } |
147 | } | |
05af2e10 | 148 | current->rss_stat.events = 0; |
34e55232 KH |
149 | } |
150 | ||
151 | static void add_mm_counter_fast(struct mm_struct *mm, int member, int val) | |
152 | { | |
153 | struct task_struct *task = current; | |
154 | ||
155 | if (likely(task->mm == mm)) | |
156 | task->rss_stat.count[member] += val; | |
157 | else | |
158 | add_mm_counter(mm, member, val); | |
159 | } | |
160 | #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1) | |
161 | #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1) | |
162 | ||
163 | /* sync counter once per 64 page faults */ | |
164 | #define TASK_RSS_EVENTS_THRESH (64) | |
165 | static void check_sync_rss_stat(struct task_struct *task) | |
166 | { | |
167 | if (unlikely(task != current)) | |
168 | return; | |
169 | if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH)) | |
ea48cf78 | 170 | sync_mm_rss(task->mm); |
34e55232 | 171 | } |
9547d01b | 172 | #else /* SPLIT_RSS_COUNTING */ |
34e55232 KH |
173 | |
174 | #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member) | |
175 | #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member) | |
176 | ||
177 | static void check_sync_rss_stat(struct task_struct *task) | |
178 | { | |
179 | } | |
180 | ||
9547d01b PZ |
181 | #endif /* SPLIT_RSS_COUNTING */ |
182 | ||
183 | #ifdef HAVE_GENERIC_MMU_GATHER | |
184 | ||
ca1d6c7d | 185 | static bool tlb_next_batch(struct mmu_gather *tlb) |
9547d01b PZ |
186 | { |
187 | struct mmu_gather_batch *batch; | |
188 | ||
189 | batch = tlb->active; | |
190 | if (batch->next) { | |
191 | tlb->active = batch->next; | |
ca1d6c7d | 192 | return true; |
9547d01b PZ |
193 | } |
194 | ||
53a59fc6 | 195 | if (tlb->batch_count == MAX_GATHER_BATCH_COUNT) |
ca1d6c7d | 196 | return false; |
53a59fc6 | 197 | |
9547d01b PZ |
198 | batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0); |
199 | if (!batch) | |
ca1d6c7d | 200 | return false; |
9547d01b | 201 | |
53a59fc6 | 202 | tlb->batch_count++; |
9547d01b PZ |
203 | batch->next = NULL; |
204 | batch->nr = 0; | |
205 | batch->max = MAX_GATHER_BATCH; | |
206 | ||
207 | tlb->active->next = batch; | |
208 | tlb->active = batch; | |
209 | ||
ca1d6c7d | 210 | return true; |
9547d01b PZ |
211 | } |
212 | ||
213 | /* tlb_gather_mmu | |
214 | * Called to initialize an (on-stack) mmu_gather structure for page-table | |
215 | * tear-down from @mm. The @fullmm argument is used when @mm is without | |
216 | * users and we're going to destroy the full address space (exit/execve). | |
217 | */ | |
2b047252 | 218 | void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, unsigned long start, unsigned long end) |
9547d01b PZ |
219 | { |
220 | tlb->mm = mm; | |
221 | ||
2b047252 LT |
222 | /* Is it from 0 to ~0? */ |
223 | tlb->fullmm = !(start | (end+1)); | |
1de14c3c | 224 | tlb->need_flush_all = 0; |
9547d01b PZ |
225 | tlb->local.next = NULL; |
226 | tlb->local.nr = 0; | |
227 | tlb->local.max = ARRAY_SIZE(tlb->__pages); | |
228 | tlb->active = &tlb->local; | |
53a59fc6 | 229 | tlb->batch_count = 0; |
9547d01b PZ |
230 | |
231 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE | |
232 | tlb->batch = NULL; | |
233 | #endif | |
fb7332a9 WD |
234 | |
235 | __tlb_reset_range(tlb); | |
9547d01b PZ |
236 | } |
237 | ||
1cf35d47 | 238 | static void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb) |
9547d01b | 239 | { |
721c21c1 WD |
240 | if (!tlb->end) |
241 | return; | |
242 | ||
9547d01b | 243 | tlb_flush(tlb); |
34ee645e | 244 | mmu_notifier_invalidate_range(tlb->mm, tlb->start, tlb->end); |
9547d01b PZ |
245 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE |
246 | tlb_table_flush(tlb); | |
34e55232 | 247 | #endif |
fb7332a9 | 248 | __tlb_reset_range(tlb); |
1cf35d47 LT |
249 | } |
250 | ||
251 | static void tlb_flush_mmu_free(struct mmu_gather *tlb) | |
252 | { | |
253 | struct mmu_gather_batch *batch; | |
34e55232 | 254 | |
721c21c1 | 255 | for (batch = &tlb->local; batch && batch->nr; batch = batch->next) { |
9547d01b PZ |
256 | free_pages_and_swap_cache(batch->pages, batch->nr); |
257 | batch->nr = 0; | |
258 | } | |
259 | tlb->active = &tlb->local; | |
260 | } | |
261 | ||
1cf35d47 LT |
262 | void tlb_flush_mmu(struct mmu_gather *tlb) |
263 | { | |
1cf35d47 LT |
264 | tlb_flush_mmu_tlbonly(tlb); |
265 | tlb_flush_mmu_free(tlb); | |
266 | } | |
267 | ||
9547d01b PZ |
268 | /* tlb_finish_mmu |
269 | * Called at the end of the shootdown operation to free up any resources | |
270 | * that were required. | |
271 | */ | |
272 | void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end) | |
273 | { | |
274 | struct mmu_gather_batch *batch, *next; | |
275 | ||
276 | tlb_flush_mmu(tlb); | |
277 | ||
278 | /* keep the page table cache within bounds */ | |
279 | check_pgt_cache(); | |
280 | ||
281 | for (batch = tlb->local.next; batch; batch = next) { | |
282 | next = batch->next; | |
283 | free_pages((unsigned long)batch, 0); | |
284 | } | |
285 | tlb->local.next = NULL; | |
286 | } | |
287 | ||
288 | /* __tlb_remove_page | |
289 | * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while | |
290 | * handling the additional races in SMP caused by other CPUs caching valid | |
291 | * mappings in their TLBs. Returns the number of free page slots left. | |
292 | * When out of page slots we must call tlb_flush_mmu(). | |
293 | */ | |
294 | int __tlb_remove_page(struct mmu_gather *tlb, struct page *page) | |
295 | { | |
296 | struct mmu_gather_batch *batch; | |
297 | ||
fb7332a9 | 298 | VM_BUG_ON(!tlb->end); |
9547d01b | 299 | |
9547d01b PZ |
300 | batch = tlb->active; |
301 | batch->pages[batch->nr++] = page; | |
302 | if (batch->nr == batch->max) { | |
303 | if (!tlb_next_batch(tlb)) | |
304 | return 0; | |
0b43c3aa | 305 | batch = tlb->active; |
9547d01b | 306 | } |
309381fe | 307 | VM_BUG_ON_PAGE(batch->nr > batch->max, page); |
9547d01b PZ |
308 | |
309 | return batch->max - batch->nr; | |
310 | } | |
311 | ||
312 | #endif /* HAVE_GENERIC_MMU_GATHER */ | |
313 | ||
26723911 PZ |
314 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE |
315 | ||
316 | /* | |
317 | * See the comment near struct mmu_table_batch. | |
318 | */ | |
319 | ||
320 | static void tlb_remove_table_smp_sync(void *arg) | |
321 | { | |
322 | /* Simply deliver the interrupt */ | |
323 | } | |
324 | ||
325 | static void tlb_remove_table_one(void *table) | |
326 | { | |
327 | /* | |
328 | * This isn't an RCU grace period and hence the page-tables cannot be | |
329 | * assumed to be actually RCU-freed. | |
330 | * | |
331 | * It is however sufficient for software page-table walkers that rely on | |
332 | * IRQ disabling. See the comment near struct mmu_table_batch. | |
333 | */ | |
334 | smp_call_function(tlb_remove_table_smp_sync, NULL, 1); | |
335 | __tlb_remove_table(table); | |
336 | } | |
337 | ||
338 | static void tlb_remove_table_rcu(struct rcu_head *head) | |
339 | { | |
340 | struct mmu_table_batch *batch; | |
341 | int i; | |
342 | ||
343 | batch = container_of(head, struct mmu_table_batch, rcu); | |
344 | ||
345 | for (i = 0; i < batch->nr; i++) | |
346 | __tlb_remove_table(batch->tables[i]); | |
347 | ||
348 | free_page((unsigned long)batch); | |
349 | } | |
350 | ||
351 | void tlb_table_flush(struct mmu_gather *tlb) | |
352 | { | |
353 | struct mmu_table_batch **batch = &tlb->batch; | |
354 | ||
355 | if (*batch) { | |
356 | call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu); | |
357 | *batch = NULL; | |
358 | } | |
359 | } | |
360 | ||
361 | void tlb_remove_table(struct mmu_gather *tlb, void *table) | |
362 | { | |
363 | struct mmu_table_batch **batch = &tlb->batch; | |
364 | ||
26723911 PZ |
365 | /* |
366 | * When there's less then two users of this mm there cannot be a | |
367 | * concurrent page-table walk. | |
368 | */ | |
369 | if (atomic_read(&tlb->mm->mm_users) < 2) { | |
370 | __tlb_remove_table(table); | |
371 | return; | |
372 | } | |
373 | ||
374 | if (*batch == NULL) { | |
375 | *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN); | |
376 | if (*batch == NULL) { | |
377 | tlb_remove_table_one(table); | |
378 | return; | |
379 | } | |
380 | (*batch)->nr = 0; | |
381 | } | |
382 | (*batch)->tables[(*batch)->nr++] = table; | |
383 | if ((*batch)->nr == MAX_TABLE_BATCH) | |
384 | tlb_table_flush(tlb); | |
385 | } | |
386 | ||
9547d01b | 387 | #endif /* CONFIG_HAVE_RCU_TABLE_FREE */ |
26723911 | 388 | |
1da177e4 LT |
389 | /* |
390 | * Note: this doesn't free the actual pages themselves. That | |
391 | * has been handled earlier when unmapping all the memory regions. | |
392 | */ | |
9e1b32ca BH |
393 | static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, |
394 | unsigned long addr) | |
1da177e4 | 395 | { |
2f569afd | 396 | pgtable_t token = pmd_pgtable(*pmd); |
e0da382c | 397 | pmd_clear(pmd); |
9e1b32ca | 398 | pte_free_tlb(tlb, token, addr); |
e1f56c89 | 399 | atomic_long_dec(&tlb->mm->nr_ptes); |
1da177e4 LT |
400 | } |
401 | ||
e0da382c HD |
402 | static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
403 | unsigned long addr, unsigned long end, | |
404 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
405 | { |
406 | pmd_t *pmd; | |
407 | unsigned long next; | |
e0da382c | 408 | unsigned long start; |
1da177e4 | 409 | |
e0da382c | 410 | start = addr; |
1da177e4 | 411 | pmd = pmd_offset(pud, addr); |
1da177e4 LT |
412 | do { |
413 | next = pmd_addr_end(addr, end); | |
414 | if (pmd_none_or_clear_bad(pmd)) | |
415 | continue; | |
9e1b32ca | 416 | free_pte_range(tlb, pmd, addr); |
1da177e4 LT |
417 | } while (pmd++, addr = next, addr != end); |
418 | ||
e0da382c HD |
419 | start &= PUD_MASK; |
420 | if (start < floor) | |
421 | return; | |
422 | if (ceiling) { | |
423 | ceiling &= PUD_MASK; | |
424 | if (!ceiling) | |
425 | return; | |
1da177e4 | 426 | } |
e0da382c HD |
427 | if (end - 1 > ceiling - 1) |
428 | return; | |
429 | ||
430 | pmd = pmd_offset(pud, start); | |
431 | pud_clear(pud); | |
9e1b32ca | 432 | pmd_free_tlb(tlb, pmd, start); |
dc6c9a35 | 433 | mm_dec_nr_pmds(tlb->mm); |
1da177e4 LT |
434 | } |
435 | ||
e0da382c HD |
436 | static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, |
437 | unsigned long addr, unsigned long end, | |
438 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
439 | { |
440 | pud_t *pud; | |
441 | unsigned long next; | |
e0da382c | 442 | unsigned long start; |
1da177e4 | 443 | |
e0da382c | 444 | start = addr; |
1da177e4 | 445 | pud = pud_offset(pgd, addr); |
1da177e4 LT |
446 | do { |
447 | next = pud_addr_end(addr, end); | |
448 | if (pud_none_or_clear_bad(pud)) | |
449 | continue; | |
e0da382c | 450 | free_pmd_range(tlb, pud, addr, next, floor, ceiling); |
1da177e4 LT |
451 | } while (pud++, addr = next, addr != end); |
452 | ||
e0da382c HD |
453 | start &= PGDIR_MASK; |
454 | if (start < floor) | |
455 | return; | |
456 | if (ceiling) { | |
457 | ceiling &= PGDIR_MASK; | |
458 | if (!ceiling) | |
459 | return; | |
1da177e4 | 460 | } |
e0da382c HD |
461 | if (end - 1 > ceiling - 1) |
462 | return; | |
463 | ||
464 | pud = pud_offset(pgd, start); | |
465 | pgd_clear(pgd); | |
9e1b32ca | 466 | pud_free_tlb(tlb, pud, start); |
1da177e4 LT |
467 | } |
468 | ||
469 | /* | |
e0da382c | 470 | * This function frees user-level page tables of a process. |
1da177e4 | 471 | */ |
42b77728 | 472 | void free_pgd_range(struct mmu_gather *tlb, |
e0da382c HD |
473 | unsigned long addr, unsigned long end, |
474 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
475 | { |
476 | pgd_t *pgd; | |
477 | unsigned long next; | |
e0da382c HD |
478 | |
479 | /* | |
480 | * The next few lines have given us lots of grief... | |
481 | * | |
482 | * Why are we testing PMD* at this top level? Because often | |
483 | * there will be no work to do at all, and we'd prefer not to | |
484 | * go all the way down to the bottom just to discover that. | |
485 | * | |
486 | * Why all these "- 1"s? Because 0 represents both the bottom | |
487 | * of the address space and the top of it (using -1 for the | |
488 | * top wouldn't help much: the masks would do the wrong thing). | |
489 | * The rule is that addr 0 and floor 0 refer to the bottom of | |
490 | * the address space, but end 0 and ceiling 0 refer to the top | |
491 | * Comparisons need to use "end - 1" and "ceiling - 1" (though | |
492 | * that end 0 case should be mythical). | |
493 | * | |
494 | * Wherever addr is brought up or ceiling brought down, we must | |
495 | * be careful to reject "the opposite 0" before it confuses the | |
496 | * subsequent tests. But what about where end is brought down | |
497 | * by PMD_SIZE below? no, end can't go down to 0 there. | |
498 | * | |
499 | * Whereas we round start (addr) and ceiling down, by different | |
500 | * masks at different levels, in order to test whether a table | |
501 | * now has no other vmas using it, so can be freed, we don't | |
502 | * bother to round floor or end up - the tests don't need that. | |
503 | */ | |
1da177e4 | 504 | |
e0da382c HD |
505 | addr &= PMD_MASK; |
506 | if (addr < floor) { | |
507 | addr += PMD_SIZE; | |
508 | if (!addr) | |
509 | return; | |
510 | } | |
511 | if (ceiling) { | |
512 | ceiling &= PMD_MASK; | |
513 | if (!ceiling) | |
514 | return; | |
515 | } | |
516 | if (end - 1 > ceiling - 1) | |
517 | end -= PMD_SIZE; | |
518 | if (addr > end - 1) | |
519 | return; | |
520 | ||
42b77728 | 521 | pgd = pgd_offset(tlb->mm, addr); |
1da177e4 LT |
522 | do { |
523 | next = pgd_addr_end(addr, end); | |
524 | if (pgd_none_or_clear_bad(pgd)) | |
525 | continue; | |
42b77728 | 526 | free_pud_range(tlb, pgd, addr, next, floor, ceiling); |
1da177e4 | 527 | } while (pgd++, addr = next, addr != end); |
e0da382c HD |
528 | } |
529 | ||
42b77728 | 530 | void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, |
3bf5ee95 | 531 | unsigned long floor, unsigned long ceiling) |
e0da382c HD |
532 | { |
533 | while (vma) { | |
534 | struct vm_area_struct *next = vma->vm_next; | |
535 | unsigned long addr = vma->vm_start; | |
536 | ||
8f4f8c16 | 537 | /* |
25d9e2d1 | 538 | * Hide vma from rmap and truncate_pagecache before freeing |
539 | * pgtables | |
8f4f8c16 | 540 | */ |
5beb4930 | 541 | unlink_anon_vmas(vma); |
8f4f8c16 HD |
542 | unlink_file_vma(vma); |
543 | ||
9da61aef | 544 | if (is_vm_hugetlb_page(vma)) { |
3bf5ee95 | 545 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
e0da382c | 546 | floor, next? next->vm_start: ceiling); |
3bf5ee95 HD |
547 | } else { |
548 | /* | |
549 | * Optimization: gather nearby vmas into one call down | |
550 | */ | |
551 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE | |
4866920b | 552 | && !is_vm_hugetlb_page(next)) { |
3bf5ee95 HD |
553 | vma = next; |
554 | next = vma->vm_next; | |
5beb4930 | 555 | unlink_anon_vmas(vma); |
8f4f8c16 | 556 | unlink_file_vma(vma); |
3bf5ee95 HD |
557 | } |
558 | free_pgd_range(tlb, addr, vma->vm_end, | |
559 | floor, next? next->vm_start: ceiling); | |
560 | } | |
e0da382c HD |
561 | vma = next; |
562 | } | |
1da177e4 LT |
563 | } |
564 | ||
8ac1f832 AA |
565 | int __pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma, |
566 | pmd_t *pmd, unsigned long address) | |
1da177e4 | 567 | { |
c4088ebd | 568 | spinlock_t *ptl; |
2f569afd | 569 | pgtable_t new = pte_alloc_one(mm, address); |
1bb3630e HD |
570 | if (!new) |
571 | return -ENOMEM; | |
572 | ||
362a61ad NP |
573 | /* |
574 | * Ensure all pte setup (eg. pte page lock and page clearing) are | |
575 | * visible before the pte is made visible to other CPUs by being | |
576 | * put into page tables. | |
577 | * | |
578 | * The other side of the story is the pointer chasing in the page | |
579 | * table walking code (when walking the page table without locking; | |
580 | * ie. most of the time). Fortunately, these data accesses consist | |
581 | * of a chain of data-dependent loads, meaning most CPUs (alpha | |
582 | * being the notable exception) will already guarantee loads are | |
583 | * seen in-order. See the alpha page table accessors for the | |
584 | * smp_read_barrier_depends() barriers in page table walking code. | |
585 | */ | |
586 | smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ | |
587 | ||
c4088ebd | 588 | ptl = pmd_lock(mm, pmd); |
8ac1f832 | 589 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
e1f56c89 | 590 | atomic_long_inc(&mm->nr_ptes); |
1da177e4 | 591 | pmd_populate(mm, pmd, new); |
2f569afd | 592 | new = NULL; |
4b471e88 | 593 | } |
c4088ebd | 594 | spin_unlock(ptl); |
2f569afd MS |
595 | if (new) |
596 | pte_free(mm, new); | |
1bb3630e | 597 | return 0; |
1da177e4 LT |
598 | } |
599 | ||
1bb3630e | 600 | int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) |
1da177e4 | 601 | { |
1bb3630e HD |
602 | pte_t *new = pte_alloc_one_kernel(&init_mm, address); |
603 | if (!new) | |
604 | return -ENOMEM; | |
605 | ||
362a61ad NP |
606 | smp_wmb(); /* See comment in __pte_alloc */ |
607 | ||
1bb3630e | 608 | spin_lock(&init_mm.page_table_lock); |
8ac1f832 | 609 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
1bb3630e | 610 | pmd_populate_kernel(&init_mm, pmd, new); |
2f569afd | 611 | new = NULL; |
4b471e88 | 612 | } |
1bb3630e | 613 | spin_unlock(&init_mm.page_table_lock); |
2f569afd MS |
614 | if (new) |
615 | pte_free_kernel(&init_mm, new); | |
1bb3630e | 616 | return 0; |
1da177e4 LT |
617 | } |
618 | ||
d559db08 KH |
619 | static inline void init_rss_vec(int *rss) |
620 | { | |
621 | memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); | |
622 | } | |
623 | ||
624 | static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) | |
ae859762 | 625 | { |
d559db08 KH |
626 | int i; |
627 | ||
34e55232 | 628 | if (current->mm == mm) |
05af2e10 | 629 | sync_mm_rss(mm); |
d559db08 KH |
630 | for (i = 0; i < NR_MM_COUNTERS; i++) |
631 | if (rss[i]) | |
632 | add_mm_counter(mm, i, rss[i]); | |
ae859762 HD |
633 | } |
634 | ||
b5810039 | 635 | /* |
6aab341e LT |
636 | * This function is called to print an error when a bad pte |
637 | * is found. For example, we might have a PFN-mapped pte in | |
638 | * a region that doesn't allow it. | |
b5810039 NP |
639 | * |
640 | * The calling function must still handle the error. | |
641 | */ | |
3dc14741 HD |
642 | static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, |
643 | pte_t pte, struct page *page) | |
b5810039 | 644 | { |
3dc14741 HD |
645 | pgd_t *pgd = pgd_offset(vma->vm_mm, addr); |
646 | pud_t *pud = pud_offset(pgd, addr); | |
647 | pmd_t *pmd = pmd_offset(pud, addr); | |
648 | struct address_space *mapping; | |
649 | pgoff_t index; | |
d936cf9b HD |
650 | static unsigned long resume; |
651 | static unsigned long nr_shown; | |
652 | static unsigned long nr_unshown; | |
653 | ||
654 | /* | |
655 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
656 | * or allow a steady drip of one report per second. | |
657 | */ | |
658 | if (nr_shown == 60) { | |
659 | if (time_before(jiffies, resume)) { | |
660 | nr_unshown++; | |
661 | return; | |
662 | } | |
663 | if (nr_unshown) { | |
1e9e6365 HD |
664 | printk(KERN_ALERT |
665 | "BUG: Bad page map: %lu messages suppressed\n", | |
d936cf9b HD |
666 | nr_unshown); |
667 | nr_unshown = 0; | |
668 | } | |
669 | nr_shown = 0; | |
670 | } | |
671 | if (nr_shown++ == 0) | |
672 | resume = jiffies + 60 * HZ; | |
3dc14741 HD |
673 | |
674 | mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; | |
675 | index = linear_page_index(vma, addr); | |
676 | ||
1e9e6365 HD |
677 | printk(KERN_ALERT |
678 | "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", | |
3dc14741 HD |
679 | current->comm, |
680 | (long long)pte_val(pte), (long long)pmd_val(*pmd)); | |
718a3821 | 681 | if (page) |
f0b791a3 | 682 | dump_page(page, "bad pte"); |
1e9e6365 | 683 | printk(KERN_ALERT |
3dc14741 HD |
684 | "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n", |
685 | (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); | |
686 | /* | |
687 | * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y | |
688 | */ | |
2682582a KK |
689 | pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n", |
690 | vma->vm_file, | |
691 | vma->vm_ops ? vma->vm_ops->fault : NULL, | |
692 | vma->vm_file ? vma->vm_file->f_op->mmap : NULL, | |
693 | mapping ? mapping->a_ops->readpage : NULL); | |
b5810039 | 694 | dump_stack(); |
373d4d09 | 695 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
b5810039 NP |
696 | } |
697 | ||
ee498ed7 | 698 | /* |
7e675137 | 699 | * vm_normal_page -- This function gets the "struct page" associated with a pte. |
6aab341e | 700 | * |
7e675137 NP |
701 | * "Special" mappings do not wish to be associated with a "struct page" (either |
702 | * it doesn't exist, or it exists but they don't want to touch it). In this | |
703 | * case, NULL is returned here. "Normal" mappings do have a struct page. | |
b379d790 | 704 | * |
7e675137 NP |
705 | * There are 2 broad cases. Firstly, an architecture may define a pte_special() |
706 | * pte bit, in which case this function is trivial. Secondly, an architecture | |
707 | * may not have a spare pte bit, which requires a more complicated scheme, | |
708 | * described below. | |
709 | * | |
710 | * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a | |
711 | * special mapping (even if there are underlying and valid "struct pages"). | |
712 | * COWed pages of a VM_PFNMAP are always normal. | |
6aab341e | 713 | * |
b379d790 JH |
714 | * The way we recognize COWed pages within VM_PFNMAP mappings is through the |
715 | * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit | |
7e675137 NP |
716 | * set, and the vm_pgoff will point to the first PFN mapped: thus every special |
717 | * mapping will always honor the rule | |
6aab341e LT |
718 | * |
719 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) | |
720 | * | |
7e675137 NP |
721 | * And for normal mappings this is false. |
722 | * | |
723 | * This restricts such mappings to be a linear translation from virtual address | |
724 | * to pfn. To get around this restriction, we allow arbitrary mappings so long | |
725 | * as the vma is not a COW mapping; in that case, we know that all ptes are | |
726 | * special (because none can have been COWed). | |
b379d790 | 727 | * |
b379d790 | 728 | * |
7e675137 | 729 | * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. |
b379d790 JH |
730 | * |
731 | * VM_MIXEDMAP mappings can likewise contain memory with or without "struct | |
732 | * page" backing, however the difference is that _all_ pages with a struct | |
733 | * page (that is, those where pfn_valid is true) are refcounted and considered | |
734 | * normal pages by the VM. The disadvantage is that pages are refcounted | |
735 | * (which can be slower and simply not an option for some PFNMAP users). The | |
736 | * advantage is that we don't have to follow the strict linearity rule of | |
737 | * PFNMAP mappings in order to support COWable mappings. | |
738 | * | |
ee498ed7 | 739 | */ |
7e675137 NP |
740 | #ifdef __HAVE_ARCH_PTE_SPECIAL |
741 | # define HAVE_PTE_SPECIAL 1 | |
742 | #else | |
743 | # define HAVE_PTE_SPECIAL 0 | |
744 | #endif | |
745 | struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, | |
746 | pte_t pte) | |
ee498ed7 | 747 | { |
22b31eec | 748 | unsigned long pfn = pte_pfn(pte); |
7e675137 NP |
749 | |
750 | if (HAVE_PTE_SPECIAL) { | |
b38af472 | 751 | if (likely(!pte_special(pte))) |
22b31eec | 752 | goto check_pfn; |
667a0a06 DV |
753 | if (vma->vm_ops && vma->vm_ops->find_special_page) |
754 | return vma->vm_ops->find_special_page(vma, addr); | |
a13ea5b7 HD |
755 | if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) |
756 | return NULL; | |
62eede62 | 757 | if (!is_zero_pfn(pfn)) |
22b31eec | 758 | print_bad_pte(vma, addr, pte, NULL); |
7e675137 NP |
759 | return NULL; |
760 | } | |
761 | ||
762 | /* !HAVE_PTE_SPECIAL case follows: */ | |
763 | ||
b379d790 JH |
764 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
765 | if (vma->vm_flags & VM_MIXEDMAP) { | |
766 | if (!pfn_valid(pfn)) | |
767 | return NULL; | |
768 | goto out; | |
769 | } else { | |
7e675137 NP |
770 | unsigned long off; |
771 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
b379d790 JH |
772 | if (pfn == vma->vm_pgoff + off) |
773 | return NULL; | |
774 | if (!is_cow_mapping(vma->vm_flags)) | |
775 | return NULL; | |
776 | } | |
6aab341e LT |
777 | } |
778 | ||
b38af472 HD |
779 | if (is_zero_pfn(pfn)) |
780 | return NULL; | |
22b31eec HD |
781 | check_pfn: |
782 | if (unlikely(pfn > highest_memmap_pfn)) { | |
783 | print_bad_pte(vma, addr, pte, NULL); | |
784 | return NULL; | |
785 | } | |
6aab341e LT |
786 | |
787 | /* | |
7e675137 | 788 | * NOTE! We still have PageReserved() pages in the page tables. |
7e675137 | 789 | * eg. VDSO mappings can cause them to exist. |
6aab341e | 790 | */ |
b379d790 | 791 | out: |
6aab341e | 792 | return pfn_to_page(pfn); |
ee498ed7 HD |
793 | } |
794 | ||
1da177e4 LT |
795 | /* |
796 | * copy one vm_area from one task to the other. Assumes the page tables | |
797 | * already present in the new task to be cleared in the whole range | |
798 | * covered by this vma. | |
1da177e4 LT |
799 | */ |
800 | ||
570a335b | 801 | static inline unsigned long |
1da177e4 | 802 | copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
b5810039 | 803 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, |
8c103762 | 804 | unsigned long addr, int *rss) |
1da177e4 | 805 | { |
b5810039 | 806 | unsigned long vm_flags = vma->vm_flags; |
1da177e4 LT |
807 | pte_t pte = *src_pte; |
808 | struct page *page; | |
1da177e4 LT |
809 | |
810 | /* pte contains position in swap or file, so copy. */ | |
811 | if (unlikely(!pte_present(pte))) { | |
0661a336 KS |
812 | swp_entry_t entry = pte_to_swp_entry(pte); |
813 | ||
814 | if (likely(!non_swap_entry(entry))) { | |
815 | if (swap_duplicate(entry) < 0) | |
816 | return entry.val; | |
817 | ||
818 | /* make sure dst_mm is on swapoff's mmlist. */ | |
819 | if (unlikely(list_empty(&dst_mm->mmlist))) { | |
820 | spin_lock(&mmlist_lock); | |
821 | if (list_empty(&dst_mm->mmlist)) | |
822 | list_add(&dst_mm->mmlist, | |
823 | &src_mm->mmlist); | |
824 | spin_unlock(&mmlist_lock); | |
825 | } | |
826 | rss[MM_SWAPENTS]++; | |
827 | } else if (is_migration_entry(entry)) { | |
828 | page = migration_entry_to_page(entry); | |
829 | ||
eca56ff9 | 830 | rss[mm_counter(page)]++; |
0661a336 KS |
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); | |
840 | if (pte_swp_soft_dirty(*src_pte)) | |
841 | pte = pte_swp_mksoft_dirty(pte); | |
842 | set_pte_at(src_mm, addr, src_pte, pte); | |
0697212a | 843 | } |
1da177e4 | 844 | } |
ae859762 | 845 | goto out_set_pte; |
1da177e4 LT |
846 | } |
847 | ||
1da177e4 LT |
848 | /* |
849 | * If it's a COW mapping, write protect it both | |
850 | * in the parent and the child | |
851 | */ | |
67121172 | 852 | if (is_cow_mapping(vm_flags)) { |
1da177e4 | 853 | ptep_set_wrprotect(src_mm, addr, src_pte); |
3dc90795 | 854 | pte = pte_wrprotect(pte); |
1da177e4 LT |
855 | } |
856 | ||
857 | /* | |
858 | * If it's a shared mapping, mark it clean in | |
859 | * the child | |
860 | */ | |
861 | if (vm_flags & VM_SHARED) | |
862 | pte = pte_mkclean(pte); | |
863 | pte = pte_mkold(pte); | |
6aab341e LT |
864 | |
865 | page = vm_normal_page(vma, addr, pte); | |
866 | if (page) { | |
867 | get_page(page); | |
53f9263b | 868 | page_dup_rmap(page, false); |
eca56ff9 | 869 | rss[mm_counter(page)]++; |
6aab341e | 870 | } |
ae859762 HD |
871 | |
872 | out_set_pte: | |
873 | set_pte_at(dst_mm, addr, dst_pte, pte); | |
570a335b | 874 | return 0; |
1da177e4 LT |
875 | } |
876 | ||
21bda264 | 877 | static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
71e3aac0 AA |
878 | pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, |
879 | unsigned long addr, unsigned long end) | |
1da177e4 | 880 | { |
c36987e2 | 881 | pte_t *orig_src_pte, *orig_dst_pte; |
1da177e4 | 882 | pte_t *src_pte, *dst_pte; |
c74df32c | 883 | spinlock_t *src_ptl, *dst_ptl; |
e040f218 | 884 | int progress = 0; |
d559db08 | 885 | int rss[NR_MM_COUNTERS]; |
570a335b | 886 | swp_entry_t entry = (swp_entry_t){0}; |
1da177e4 LT |
887 | |
888 | again: | |
d559db08 KH |
889 | init_rss_vec(rss); |
890 | ||
c74df32c | 891 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
1da177e4 LT |
892 | if (!dst_pte) |
893 | return -ENOMEM; | |
ece0e2b6 | 894 | src_pte = pte_offset_map(src_pmd, addr); |
4c21e2f2 | 895 | src_ptl = pte_lockptr(src_mm, src_pmd); |
f20dc5f7 | 896 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
c36987e2 DN |
897 | orig_src_pte = src_pte; |
898 | orig_dst_pte = dst_pte; | |
6606c3e0 | 899 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 900 | |
1da177e4 LT |
901 | do { |
902 | /* | |
903 | * We are holding two locks at this point - either of them | |
904 | * could generate latencies in another task on another CPU. | |
905 | */ | |
e040f218 HD |
906 | if (progress >= 32) { |
907 | progress = 0; | |
908 | if (need_resched() || | |
95c354fe | 909 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
e040f218 HD |
910 | break; |
911 | } | |
1da177e4 LT |
912 | if (pte_none(*src_pte)) { |
913 | progress++; | |
914 | continue; | |
915 | } | |
570a335b HD |
916 | entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, |
917 | vma, addr, rss); | |
918 | if (entry.val) | |
919 | break; | |
1da177e4 LT |
920 | progress += 8; |
921 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); | |
1da177e4 | 922 | |
6606c3e0 | 923 | arch_leave_lazy_mmu_mode(); |
c74df32c | 924 | spin_unlock(src_ptl); |
ece0e2b6 | 925 | pte_unmap(orig_src_pte); |
d559db08 | 926 | add_mm_rss_vec(dst_mm, rss); |
c36987e2 | 927 | pte_unmap_unlock(orig_dst_pte, dst_ptl); |
c74df32c | 928 | cond_resched(); |
570a335b HD |
929 | |
930 | if (entry.val) { | |
931 | if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) | |
932 | return -ENOMEM; | |
933 | progress = 0; | |
934 | } | |
1da177e4 LT |
935 | if (addr != end) |
936 | goto again; | |
937 | return 0; | |
938 | } | |
939 | ||
940 | static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
941 | pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, | |
942 | unsigned long addr, unsigned long end) | |
943 | { | |
944 | pmd_t *src_pmd, *dst_pmd; | |
945 | unsigned long next; | |
946 | ||
947 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); | |
948 | if (!dst_pmd) | |
949 | return -ENOMEM; | |
950 | src_pmd = pmd_offset(src_pud, addr); | |
951 | do { | |
952 | next = pmd_addr_end(addr, end); | |
5c7fb56e | 953 | if (pmd_trans_huge(*src_pmd) || pmd_devmap(*src_pmd)) { |
71e3aac0 | 954 | int err; |
14d1a55c | 955 | VM_BUG_ON(next-addr != HPAGE_PMD_SIZE); |
71e3aac0 AA |
956 | err = copy_huge_pmd(dst_mm, src_mm, |
957 | dst_pmd, src_pmd, addr, vma); | |
958 | if (err == -ENOMEM) | |
959 | return -ENOMEM; | |
960 | if (!err) | |
961 | continue; | |
962 | /* fall through */ | |
963 | } | |
1da177e4 LT |
964 | if (pmd_none_or_clear_bad(src_pmd)) |
965 | continue; | |
966 | if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, | |
967 | vma, addr, next)) | |
968 | return -ENOMEM; | |
969 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); | |
970 | return 0; | |
971 | } | |
972 | ||
973 | static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
974 | pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, | |
975 | unsigned long addr, unsigned long end) | |
976 | { | |
977 | pud_t *src_pud, *dst_pud; | |
978 | unsigned long next; | |
979 | ||
980 | dst_pud = pud_alloc(dst_mm, dst_pgd, addr); | |
981 | if (!dst_pud) | |
982 | return -ENOMEM; | |
983 | src_pud = pud_offset(src_pgd, addr); | |
984 | do { | |
985 | next = pud_addr_end(addr, end); | |
986 | if (pud_none_or_clear_bad(src_pud)) | |
987 | continue; | |
988 | if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, | |
989 | vma, addr, next)) | |
990 | return -ENOMEM; | |
991 | } while (dst_pud++, src_pud++, addr = next, addr != end); | |
992 | return 0; | |
993 | } | |
994 | ||
995 | int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
996 | struct vm_area_struct *vma) | |
997 | { | |
998 | pgd_t *src_pgd, *dst_pgd; | |
999 | unsigned long next; | |
1000 | unsigned long addr = vma->vm_start; | |
1001 | unsigned long end = vma->vm_end; | |
2ec74c3e SG |
1002 | unsigned long mmun_start; /* For mmu_notifiers */ |
1003 | unsigned long mmun_end; /* For mmu_notifiers */ | |
1004 | bool is_cow; | |
cddb8a5c | 1005 | int ret; |
1da177e4 | 1006 | |
d992895b NP |
1007 | /* |
1008 | * Don't copy ptes where a page fault will fill them correctly. | |
1009 | * Fork becomes much lighter when there are big shared or private | |
1010 | * readonly mappings. The tradeoff is that copy_page_range is more | |
1011 | * efficient than faulting. | |
1012 | */ | |
0661a336 KS |
1013 | if (!(vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) && |
1014 | !vma->anon_vma) | |
1015 | return 0; | |
d992895b | 1016 | |
1da177e4 LT |
1017 | if (is_vm_hugetlb_page(vma)) |
1018 | return copy_hugetlb_page_range(dst_mm, src_mm, vma); | |
1019 | ||
b3b9c293 | 1020 | if (unlikely(vma->vm_flags & VM_PFNMAP)) { |
2ab64037 | 1021 | /* |
1022 | * We do not free on error cases below as remove_vma | |
1023 | * gets called on error from higher level routine | |
1024 | */ | |
5180da41 | 1025 | ret = track_pfn_copy(vma); |
2ab64037 | 1026 | if (ret) |
1027 | return ret; | |
1028 | } | |
1029 | ||
cddb8a5c AA |
1030 | /* |
1031 | * We need to invalidate the secondary MMU mappings only when | |
1032 | * there could be a permission downgrade on the ptes of the | |
1033 | * parent mm. And a permission downgrade will only happen if | |
1034 | * is_cow_mapping() returns true. | |
1035 | */ | |
2ec74c3e SG |
1036 | is_cow = is_cow_mapping(vma->vm_flags); |
1037 | mmun_start = addr; | |
1038 | mmun_end = end; | |
1039 | if (is_cow) | |
1040 | mmu_notifier_invalidate_range_start(src_mm, mmun_start, | |
1041 | mmun_end); | |
cddb8a5c AA |
1042 | |
1043 | ret = 0; | |
1da177e4 LT |
1044 | dst_pgd = pgd_offset(dst_mm, addr); |
1045 | src_pgd = pgd_offset(src_mm, addr); | |
1046 | do { | |
1047 | next = pgd_addr_end(addr, end); | |
1048 | if (pgd_none_or_clear_bad(src_pgd)) | |
1049 | continue; | |
cddb8a5c AA |
1050 | if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, |
1051 | vma, addr, next))) { | |
1052 | ret = -ENOMEM; | |
1053 | break; | |
1054 | } | |
1da177e4 | 1055 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); |
cddb8a5c | 1056 | |
2ec74c3e SG |
1057 | if (is_cow) |
1058 | mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end); | |
cddb8a5c | 1059 | return ret; |
1da177e4 LT |
1060 | } |
1061 | ||
51c6f666 | 1062 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
b5810039 | 1063 | struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 | 1064 | unsigned long addr, unsigned long end, |
97a89413 | 1065 | struct zap_details *details) |
1da177e4 | 1066 | { |
b5810039 | 1067 | struct mm_struct *mm = tlb->mm; |
d16dfc55 | 1068 | int force_flush = 0; |
d559db08 | 1069 | int rss[NR_MM_COUNTERS]; |
97a89413 | 1070 | spinlock_t *ptl; |
5f1a1907 | 1071 | pte_t *start_pte; |
97a89413 | 1072 | pte_t *pte; |
8a5f14a2 | 1073 | swp_entry_t entry; |
d559db08 | 1074 | |
d16dfc55 | 1075 | again: |
e303297e | 1076 | init_rss_vec(rss); |
5f1a1907 SR |
1077 | start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
1078 | pte = start_pte; | |
6606c3e0 | 1079 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1080 | do { |
1081 | pte_t ptent = *pte; | |
51c6f666 | 1082 | if (pte_none(ptent)) { |
1da177e4 | 1083 | continue; |
51c6f666 | 1084 | } |
6f5e6b9e | 1085 | |
1da177e4 | 1086 | if (pte_present(ptent)) { |
ee498ed7 | 1087 | struct page *page; |
51c6f666 | 1088 | |
6aab341e | 1089 | page = vm_normal_page(vma, addr, ptent); |
1da177e4 LT |
1090 | if (unlikely(details) && page) { |
1091 | /* | |
1092 | * unmap_shared_mapping_pages() wants to | |
1093 | * invalidate cache without truncating: | |
1094 | * unmap shared but keep private pages. | |
1095 | */ | |
1096 | if (details->check_mapping && | |
1097 | details->check_mapping != page->mapping) | |
1098 | continue; | |
1da177e4 | 1099 | } |
b5810039 | 1100 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
a600388d | 1101 | tlb->fullmm); |
1da177e4 LT |
1102 | tlb_remove_tlb_entry(tlb, pte, addr); |
1103 | if (unlikely(!page)) | |
1104 | continue; | |
eca56ff9 JM |
1105 | |
1106 | if (!PageAnon(page)) { | |
1cf35d47 LT |
1107 | if (pte_dirty(ptent)) { |
1108 | force_flush = 1; | |
6237bcd9 | 1109 | set_page_dirty(page); |
1cf35d47 | 1110 | } |
4917e5d0 | 1111 | if (pte_young(ptent) && |
64363aad | 1112 | likely(!(vma->vm_flags & VM_SEQ_READ))) |
bf3f3bc5 | 1113 | mark_page_accessed(page); |
6237bcd9 | 1114 | } |
eca56ff9 | 1115 | rss[mm_counter(page)]--; |
d281ee61 | 1116 | page_remove_rmap(page, false); |
3dc14741 HD |
1117 | if (unlikely(page_mapcount(page) < 0)) |
1118 | print_bad_pte(vma, addr, ptent, page); | |
1cf35d47 LT |
1119 | if (unlikely(!__tlb_remove_page(tlb, page))) { |
1120 | force_flush = 1; | |
ce9ec37b | 1121 | addr += PAGE_SIZE; |
d16dfc55 | 1122 | break; |
1cf35d47 | 1123 | } |
1da177e4 LT |
1124 | continue; |
1125 | } | |
8a5f14a2 | 1126 | /* If details->check_mapping, we leave swap entries. */ |
1da177e4 LT |
1127 | if (unlikely(details)) |
1128 | continue; | |
b084d435 | 1129 | |
8a5f14a2 KS |
1130 | entry = pte_to_swp_entry(ptent); |
1131 | if (!non_swap_entry(entry)) | |
1132 | rss[MM_SWAPENTS]--; | |
1133 | else if (is_migration_entry(entry)) { | |
1134 | struct page *page; | |
9f9f1acd | 1135 | |
8a5f14a2 | 1136 | page = migration_entry_to_page(entry); |
eca56ff9 | 1137 | rss[mm_counter(page)]--; |
b084d435 | 1138 | } |
8a5f14a2 KS |
1139 | if (unlikely(!free_swap_and_cache(entry))) |
1140 | print_bad_pte(vma, addr, ptent, NULL); | |
9888a1ca | 1141 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
97a89413 | 1142 | } while (pte++, addr += PAGE_SIZE, addr != end); |
ae859762 | 1143 | |
d559db08 | 1144 | add_mm_rss_vec(mm, rss); |
6606c3e0 | 1145 | arch_leave_lazy_mmu_mode(); |
51c6f666 | 1146 | |
1cf35d47 | 1147 | /* Do the actual TLB flush before dropping ptl */ |
fb7332a9 | 1148 | if (force_flush) |
1cf35d47 | 1149 | tlb_flush_mmu_tlbonly(tlb); |
1cf35d47 LT |
1150 | pte_unmap_unlock(start_pte, ptl); |
1151 | ||
1152 | /* | |
1153 | * If we forced a TLB flush (either due to running out of | |
1154 | * batch buffers or because we needed to flush dirty TLB | |
1155 | * entries before releasing the ptl), free the batched | |
1156 | * memory too. Restart if we didn't do everything. | |
1157 | */ | |
1158 | if (force_flush) { | |
1159 | force_flush = 0; | |
1160 | tlb_flush_mmu_free(tlb); | |
2b047252 LT |
1161 | |
1162 | if (addr != end) | |
d16dfc55 PZ |
1163 | goto again; |
1164 | } | |
1165 | ||
51c6f666 | 1166 | return addr; |
1da177e4 LT |
1167 | } |
1168 | ||
51c6f666 | 1169 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
b5810039 | 1170 | struct vm_area_struct *vma, pud_t *pud, |
1da177e4 | 1171 | unsigned long addr, unsigned long end, |
97a89413 | 1172 | struct zap_details *details) |
1da177e4 LT |
1173 | { |
1174 | pmd_t *pmd; | |
1175 | unsigned long next; | |
1176 | ||
1177 | pmd = pmd_offset(pud, addr); | |
1178 | do { | |
1179 | next = pmd_addr_end(addr, end); | |
5c7fb56e | 1180 | if (pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { |
1a5a9906 | 1181 | if (next - addr != HPAGE_PMD_SIZE) { |
e0897d75 DR |
1182 | #ifdef CONFIG_DEBUG_VM |
1183 | if (!rwsem_is_locked(&tlb->mm->mmap_sem)) { | |
1184 | pr_err("%s: mmap_sem is unlocked! addr=0x%lx end=0x%lx vma->vm_start=0x%lx vma->vm_end=0x%lx\n", | |
1185 | __func__, addr, end, | |
1186 | vma->vm_start, | |
1187 | vma->vm_end); | |
1188 | BUG(); | |
1189 | } | |
1190 | #endif | |
78ddc534 | 1191 | split_huge_pmd(vma, pmd, addr); |
f21760b1 | 1192 | } else if (zap_huge_pmd(tlb, vma, pmd, addr)) |
1a5a9906 | 1193 | goto next; |
71e3aac0 AA |
1194 | /* fall through */ |
1195 | } | |
1a5a9906 AA |
1196 | /* |
1197 | * Here there can be other concurrent MADV_DONTNEED or | |
1198 | * trans huge page faults running, and if the pmd is | |
1199 | * none or trans huge it can change under us. This is | |
1200 | * because MADV_DONTNEED holds the mmap_sem in read | |
1201 | * mode. | |
1202 | */ | |
1203 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) | |
1204 | goto next; | |
97a89413 | 1205 | next = zap_pte_range(tlb, vma, pmd, addr, next, details); |
1a5a9906 | 1206 | next: |
97a89413 PZ |
1207 | cond_resched(); |
1208 | } while (pmd++, addr = next, addr != end); | |
51c6f666 RH |
1209 | |
1210 | return addr; | |
1da177e4 LT |
1211 | } |
1212 | ||
51c6f666 | 1213 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
b5810039 | 1214 | struct vm_area_struct *vma, pgd_t *pgd, |
1da177e4 | 1215 | unsigned long addr, unsigned long end, |
97a89413 | 1216 | struct zap_details *details) |
1da177e4 LT |
1217 | { |
1218 | pud_t *pud; | |
1219 | unsigned long next; | |
1220 | ||
1221 | pud = pud_offset(pgd, addr); | |
1222 | do { | |
1223 | next = pud_addr_end(addr, end); | |
97a89413 | 1224 | if (pud_none_or_clear_bad(pud)) |
1da177e4 | 1225 | continue; |
97a89413 PZ |
1226 | next = zap_pmd_range(tlb, vma, pud, addr, next, details); |
1227 | } while (pud++, addr = next, addr != end); | |
51c6f666 RH |
1228 | |
1229 | return addr; | |
1da177e4 LT |
1230 | } |
1231 | ||
038c7aa1 AV |
1232 | static void unmap_page_range(struct mmu_gather *tlb, |
1233 | struct vm_area_struct *vma, | |
1234 | unsigned long addr, unsigned long end, | |
1235 | struct zap_details *details) | |
1da177e4 LT |
1236 | { |
1237 | pgd_t *pgd; | |
1238 | unsigned long next; | |
1239 | ||
8a5f14a2 | 1240 | if (details && !details->check_mapping) |
1da177e4 LT |
1241 | details = NULL; |
1242 | ||
1243 | BUG_ON(addr >= end); | |
1244 | tlb_start_vma(tlb, vma); | |
1245 | pgd = pgd_offset(vma->vm_mm, addr); | |
1246 | do { | |
1247 | next = pgd_addr_end(addr, end); | |
97a89413 | 1248 | if (pgd_none_or_clear_bad(pgd)) |
1da177e4 | 1249 | continue; |
97a89413 PZ |
1250 | next = zap_pud_range(tlb, vma, pgd, addr, next, details); |
1251 | } while (pgd++, addr = next, addr != end); | |
1da177e4 LT |
1252 | tlb_end_vma(tlb, vma); |
1253 | } | |
51c6f666 | 1254 | |
f5cc4eef AV |
1255 | |
1256 | static void unmap_single_vma(struct mmu_gather *tlb, | |
1257 | struct vm_area_struct *vma, unsigned long start_addr, | |
4f74d2c8 | 1258 | unsigned long end_addr, |
f5cc4eef AV |
1259 | struct zap_details *details) |
1260 | { | |
1261 | unsigned long start = max(vma->vm_start, start_addr); | |
1262 | unsigned long end; | |
1263 | ||
1264 | if (start >= vma->vm_end) | |
1265 | return; | |
1266 | end = min(vma->vm_end, end_addr); | |
1267 | if (end <= vma->vm_start) | |
1268 | return; | |
1269 | ||
cbc91f71 SD |
1270 | if (vma->vm_file) |
1271 | uprobe_munmap(vma, start, end); | |
1272 | ||
b3b9c293 | 1273 | if (unlikely(vma->vm_flags & VM_PFNMAP)) |
5180da41 | 1274 | untrack_pfn(vma, 0, 0); |
f5cc4eef AV |
1275 | |
1276 | if (start != end) { | |
1277 | if (unlikely(is_vm_hugetlb_page(vma))) { | |
1278 | /* | |
1279 | * It is undesirable to test vma->vm_file as it | |
1280 | * should be non-null for valid hugetlb area. | |
1281 | * However, vm_file will be NULL in the error | |
7aa6b4ad | 1282 | * cleanup path of mmap_region. When |
f5cc4eef | 1283 | * hugetlbfs ->mmap method fails, |
7aa6b4ad | 1284 | * mmap_region() nullifies vma->vm_file |
f5cc4eef AV |
1285 | * before calling this function to clean up. |
1286 | * Since no pte has actually been setup, it is | |
1287 | * safe to do nothing in this case. | |
1288 | */ | |
24669e58 | 1289 | if (vma->vm_file) { |
83cde9e8 | 1290 | i_mmap_lock_write(vma->vm_file->f_mapping); |
d833352a | 1291 | __unmap_hugepage_range_final(tlb, vma, start, end, NULL); |
83cde9e8 | 1292 | i_mmap_unlock_write(vma->vm_file->f_mapping); |
24669e58 | 1293 | } |
f5cc4eef AV |
1294 | } else |
1295 | unmap_page_range(tlb, vma, start, end, details); | |
1296 | } | |
1da177e4 LT |
1297 | } |
1298 | ||
1da177e4 LT |
1299 | /** |
1300 | * unmap_vmas - unmap a range of memory covered by a list of vma's | |
0164f69d | 1301 | * @tlb: address of the caller's struct mmu_gather |
1da177e4 LT |
1302 | * @vma: the starting vma |
1303 | * @start_addr: virtual address at which to start unmapping | |
1304 | * @end_addr: virtual address at which to end unmapping | |
1da177e4 | 1305 | * |
508034a3 | 1306 | * Unmap all pages in the vma list. |
1da177e4 | 1307 | * |
1da177e4 LT |
1308 | * Only addresses between `start' and `end' will be unmapped. |
1309 | * | |
1310 | * The VMA list must be sorted in ascending virtual address order. | |
1311 | * | |
1312 | * unmap_vmas() assumes that the caller will flush the whole unmapped address | |
1313 | * range after unmap_vmas() returns. So the only responsibility here is to | |
1314 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() | |
1315 | * drops the lock and schedules. | |
1316 | */ | |
6e8bb019 | 1317 | void unmap_vmas(struct mmu_gather *tlb, |
1da177e4 | 1318 | struct vm_area_struct *vma, unsigned long start_addr, |
4f74d2c8 | 1319 | unsigned long end_addr) |
1da177e4 | 1320 | { |
cddb8a5c | 1321 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 1322 | |
cddb8a5c | 1323 | mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); |
f5cc4eef | 1324 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) |
4f74d2c8 | 1325 | unmap_single_vma(tlb, vma, start_addr, end_addr, NULL); |
cddb8a5c | 1326 | mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); |
1da177e4 LT |
1327 | } |
1328 | ||
1329 | /** | |
1330 | * zap_page_range - remove user pages in a given range | |
1331 | * @vma: vm_area_struct holding the applicable pages | |
eb4546bb | 1332 | * @start: starting address of pages to zap |
1da177e4 | 1333 | * @size: number of bytes to zap |
8a5f14a2 | 1334 | * @details: details of shared cache invalidation |
f5cc4eef AV |
1335 | * |
1336 | * Caller must protect the VMA list | |
1da177e4 | 1337 | */ |
7e027b14 | 1338 | void zap_page_range(struct vm_area_struct *vma, unsigned long start, |
1da177e4 LT |
1339 | unsigned long size, struct zap_details *details) |
1340 | { | |
1341 | struct mm_struct *mm = vma->vm_mm; | |
d16dfc55 | 1342 | struct mmu_gather tlb; |
7e027b14 | 1343 | unsigned long end = start + size; |
1da177e4 | 1344 | |
1da177e4 | 1345 | lru_add_drain(); |
2b047252 | 1346 | tlb_gather_mmu(&tlb, mm, start, end); |
365e9c87 | 1347 | update_hiwater_rss(mm); |
7e027b14 LT |
1348 | mmu_notifier_invalidate_range_start(mm, start, end); |
1349 | for ( ; vma && vma->vm_start < end; vma = vma->vm_next) | |
4f74d2c8 | 1350 | unmap_single_vma(&tlb, vma, start, end, details); |
7e027b14 LT |
1351 | mmu_notifier_invalidate_range_end(mm, start, end); |
1352 | tlb_finish_mmu(&tlb, start, end); | |
1da177e4 LT |
1353 | } |
1354 | ||
f5cc4eef AV |
1355 | /** |
1356 | * zap_page_range_single - remove user pages in a given range | |
1357 | * @vma: vm_area_struct holding the applicable pages | |
1358 | * @address: starting address of pages to zap | |
1359 | * @size: number of bytes to zap | |
8a5f14a2 | 1360 | * @details: details of shared cache invalidation |
f5cc4eef AV |
1361 | * |
1362 | * The range must fit into one VMA. | |
1da177e4 | 1363 | */ |
f5cc4eef | 1364 | static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, |
1da177e4 LT |
1365 | unsigned long size, struct zap_details *details) |
1366 | { | |
1367 | struct mm_struct *mm = vma->vm_mm; | |
d16dfc55 | 1368 | struct mmu_gather tlb; |
1da177e4 | 1369 | unsigned long end = address + size; |
1da177e4 | 1370 | |
1da177e4 | 1371 | lru_add_drain(); |
2b047252 | 1372 | tlb_gather_mmu(&tlb, mm, address, end); |
365e9c87 | 1373 | update_hiwater_rss(mm); |
f5cc4eef | 1374 | mmu_notifier_invalidate_range_start(mm, address, end); |
4f74d2c8 | 1375 | unmap_single_vma(&tlb, vma, address, end, details); |
f5cc4eef | 1376 | mmu_notifier_invalidate_range_end(mm, address, end); |
d16dfc55 | 1377 | tlb_finish_mmu(&tlb, address, end); |
1da177e4 LT |
1378 | } |
1379 | ||
c627f9cc JS |
1380 | /** |
1381 | * zap_vma_ptes - remove ptes mapping the vma | |
1382 | * @vma: vm_area_struct holding ptes to be zapped | |
1383 | * @address: starting address of pages to zap | |
1384 | * @size: number of bytes to zap | |
1385 | * | |
1386 | * This function only unmaps ptes assigned to VM_PFNMAP vmas. | |
1387 | * | |
1388 | * The entire address range must be fully contained within the vma. | |
1389 | * | |
1390 | * Returns 0 if successful. | |
1391 | */ | |
1392 | int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, | |
1393 | unsigned long size) | |
1394 | { | |
1395 | if (address < vma->vm_start || address + size > vma->vm_end || | |
1396 | !(vma->vm_flags & VM_PFNMAP)) | |
1397 | return -1; | |
f5cc4eef | 1398 | zap_page_range_single(vma, address, size, NULL); |
c627f9cc JS |
1399 | return 0; |
1400 | } | |
1401 | EXPORT_SYMBOL_GPL(zap_vma_ptes); | |
1402 | ||
25ca1d6c | 1403 | pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, |
920c7a5d | 1404 | spinlock_t **ptl) |
c9cfcddf LT |
1405 | { |
1406 | pgd_t * pgd = pgd_offset(mm, addr); | |
1407 | pud_t * pud = pud_alloc(mm, pgd, addr); | |
1408 | if (pud) { | |
49c91fb0 | 1409 | pmd_t * pmd = pmd_alloc(mm, pud, addr); |
f66055ab AA |
1410 | if (pmd) { |
1411 | VM_BUG_ON(pmd_trans_huge(*pmd)); | |
c9cfcddf | 1412 | return pte_alloc_map_lock(mm, pmd, addr, ptl); |
f66055ab | 1413 | } |
c9cfcddf LT |
1414 | } |
1415 | return NULL; | |
1416 | } | |
1417 | ||
238f58d8 LT |
1418 | /* |
1419 | * This is the old fallback for page remapping. | |
1420 | * | |
1421 | * For historical reasons, it only allows reserved pages. Only | |
1422 | * old drivers should use this, and they needed to mark their | |
1423 | * pages reserved for the old functions anyway. | |
1424 | */ | |
423bad60 NP |
1425 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
1426 | struct page *page, pgprot_t prot) | |
238f58d8 | 1427 | { |
423bad60 | 1428 | struct mm_struct *mm = vma->vm_mm; |
238f58d8 | 1429 | int retval; |
c9cfcddf | 1430 | pte_t *pte; |
8a9f3ccd BS |
1431 | spinlock_t *ptl; |
1432 | ||
238f58d8 | 1433 | retval = -EINVAL; |
a145dd41 | 1434 | if (PageAnon(page)) |
5b4e655e | 1435 | goto out; |
238f58d8 LT |
1436 | retval = -ENOMEM; |
1437 | flush_dcache_page(page); | |
c9cfcddf | 1438 | pte = get_locked_pte(mm, addr, &ptl); |
238f58d8 | 1439 | if (!pte) |
5b4e655e | 1440 | goto out; |
238f58d8 LT |
1441 | retval = -EBUSY; |
1442 | if (!pte_none(*pte)) | |
1443 | goto out_unlock; | |
1444 | ||
1445 | /* Ok, finally just insert the thing.. */ | |
1446 | get_page(page); | |
eca56ff9 | 1447 | inc_mm_counter_fast(mm, mm_counter_file(page)); |
238f58d8 LT |
1448 | page_add_file_rmap(page); |
1449 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); | |
1450 | ||
1451 | retval = 0; | |
8a9f3ccd BS |
1452 | pte_unmap_unlock(pte, ptl); |
1453 | return retval; | |
238f58d8 LT |
1454 | out_unlock: |
1455 | pte_unmap_unlock(pte, ptl); | |
1456 | out: | |
1457 | return retval; | |
1458 | } | |
1459 | ||
bfa5bf6d REB |
1460 | /** |
1461 | * vm_insert_page - insert single page into user vma | |
1462 | * @vma: user vma to map to | |
1463 | * @addr: target user address of this page | |
1464 | * @page: source kernel page | |
1465 | * | |
a145dd41 LT |
1466 | * This allows drivers to insert individual pages they've allocated |
1467 | * into a user vma. | |
1468 | * | |
1469 | * The page has to be a nice clean _individual_ kernel allocation. | |
1470 | * If you allocate a compound page, you need to have marked it as | |
1471 | * such (__GFP_COMP), or manually just split the page up yourself | |
8dfcc9ba | 1472 | * (see split_page()). |
a145dd41 LT |
1473 | * |
1474 | * NOTE! Traditionally this was done with "remap_pfn_range()" which | |
1475 | * took an arbitrary page protection parameter. This doesn't allow | |
1476 | * that. Your vma protection will have to be set up correctly, which | |
1477 | * means that if you want a shared writable mapping, you'd better | |
1478 | * ask for a shared writable mapping! | |
1479 | * | |
1480 | * The page does not need to be reserved. | |
4b6e1e37 KK |
1481 | * |
1482 | * Usually this function is called from f_op->mmap() handler | |
1483 | * under mm->mmap_sem write-lock, so it can change vma->vm_flags. | |
1484 | * Caller must set VM_MIXEDMAP on vma if it wants to call this | |
1485 | * function from other places, for example from page-fault handler. | |
a145dd41 | 1486 | */ |
423bad60 NP |
1487 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
1488 | struct page *page) | |
a145dd41 LT |
1489 | { |
1490 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
1491 | return -EFAULT; | |
1492 | if (!page_count(page)) | |
1493 | return -EINVAL; | |
4b6e1e37 KK |
1494 | if (!(vma->vm_flags & VM_MIXEDMAP)) { |
1495 | BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem)); | |
1496 | BUG_ON(vma->vm_flags & VM_PFNMAP); | |
1497 | vma->vm_flags |= VM_MIXEDMAP; | |
1498 | } | |
423bad60 | 1499 | return insert_page(vma, addr, page, vma->vm_page_prot); |
a145dd41 | 1500 | } |
e3c3374f | 1501 | EXPORT_SYMBOL(vm_insert_page); |
a145dd41 | 1502 | |
423bad60 | 1503 | static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
01c8f1c4 | 1504 | pfn_t pfn, pgprot_t prot) |
423bad60 NP |
1505 | { |
1506 | struct mm_struct *mm = vma->vm_mm; | |
1507 | int retval; | |
1508 | pte_t *pte, entry; | |
1509 | spinlock_t *ptl; | |
1510 | ||
1511 | retval = -ENOMEM; | |
1512 | pte = get_locked_pte(mm, addr, &ptl); | |
1513 | if (!pte) | |
1514 | goto out; | |
1515 | retval = -EBUSY; | |
1516 | if (!pte_none(*pte)) | |
1517 | goto out_unlock; | |
1518 | ||
1519 | /* Ok, finally just insert the thing.. */ | |
01c8f1c4 DW |
1520 | if (pfn_t_devmap(pfn)) |
1521 | entry = pte_mkdevmap(pfn_t_pte(pfn, prot)); | |
1522 | else | |
1523 | entry = pte_mkspecial(pfn_t_pte(pfn, prot)); | |
423bad60 | 1524 | set_pte_at(mm, addr, pte, entry); |
4b3073e1 | 1525 | update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ |
423bad60 NP |
1526 | |
1527 | retval = 0; | |
1528 | out_unlock: | |
1529 | pte_unmap_unlock(pte, ptl); | |
1530 | out: | |
1531 | return retval; | |
1532 | } | |
1533 | ||
e0dc0d8f NP |
1534 | /** |
1535 | * vm_insert_pfn - insert single pfn into user vma | |
1536 | * @vma: user vma to map to | |
1537 | * @addr: target user address of this page | |
1538 | * @pfn: source kernel pfn | |
1539 | * | |
c462f179 | 1540 | * Similar to vm_insert_page, this allows drivers to insert individual pages |
e0dc0d8f NP |
1541 | * they've allocated into a user vma. Same comments apply. |
1542 | * | |
1543 | * This function should only be called from a vm_ops->fault handler, and | |
1544 | * in that case the handler should return NULL. | |
0d71d10a NP |
1545 | * |
1546 | * vma cannot be a COW mapping. | |
1547 | * | |
1548 | * As this is called only for pages that do not currently exist, we | |
1549 | * do not need to flush old virtual caches or the TLB. | |
e0dc0d8f NP |
1550 | */ |
1551 | int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, | |
423bad60 | 1552 | unsigned long pfn) |
e0dc0d8f | 1553 | { |
2ab64037 | 1554 | int ret; |
e4b866ed | 1555 | pgprot_t pgprot = vma->vm_page_prot; |
7e675137 NP |
1556 | /* |
1557 | * Technically, architectures with pte_special can avoid all these | |
1558 | * restrictions (same for remap_pfn_range). However we would like | |
1559 | * consistency in testing and feature parity among all, so we should | |
1560 | * try to keep these invariants in place for everybody. | |
1561 | */ | |
b379d790 JH |
1562 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); |
1563 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == | |
1564 | (VM_PFNMAP|VM_MIXEDMAP)); | |
1565 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); | |
1566 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); | |
e0dc0d8f | 1567 | |
423bad60 NP |
1568 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1569 | return -EFAULT; | |
f25748e3 | 1570 | if (track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV))) |
2ab64037 | 1571 | return -EINVAL; |
1572 | ||
01c8f1c4 | 1573 | ret = insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot); |
2ab64037 | 1574 | |
2ab64037 | 1575 | return ret; |
423bad60 NP |
1576 | } |
1577 | EXPORT_SYMBOL(vm_insert_pfn); | |
e0dc0d8f | 1578 | |
423bad60 | 1579 | int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, |
01c8f1c4 | 1580 | pfn_t pfn) |
423bad60 NP |
1581 | { |
1582 | BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); | |
e0dc0d8f | 1583 | |
423bad60 NP |
1584 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1585 | return -EFAULT; | |
e0dc0d8f | 1586 | |
423bad60 NP |
1587 | /* |
1588 | * If we don't have pte special, then we have to use the pfn_valid() | |
1589 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* | |
1590 | * refcount the page if pfn_valid is true (hence insert_page rather | |
62eede62 HD |
1591 | * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP |
1592 | * without pte special, it would there be refcounted as a normal page. | |
423bad60 | 1593 | */ |
03fc2da6 | 1594 | if (!HAVE_PTE_SPECIAL && !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) { |
423bad60 NP |
1595 | struct page *page; |
1596 | ||
03fc2da6 DW |
1597 | /* |
1598 | * At this point we are committed to insert_page() | |
1599 | * regardless of whether the caller specified flags that | |
1600 | * result in pfn_t_has_page() == false. | |
1601 | */ | |
1602 | page = pfn_to_page(pfn_t_to_pfn(pfn)); | |
423bad60 NP |
1603 | return insert_page(vma, addr, page, vma->vm_page_prot); |
1604 | } | |
1605 | return insert_pfn(vma, addr, pfn, vma->vm_page_prot); | |
e0dc0d8f | 1606 | } |
423bad60 | 1607 | EXPORT_SYMBOL(vm_insert_mixed); |
e0dc0d8f | 1608 | |
1da177e4 LT |
1609 | /* |
1610 | * maps a range of physical memory into the requested pages. the old | |
1611 | * mappings are removed. any references to nonexistent pages results | |
1612 | * in null mappings (currently treated as "copy-on-access") | |
1613 | */ | |
1614 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
1615 | unsigned long addr, unsigned long end, | |
1616 | unsigned long pfn, pgprot_t prot) | |
1617 | { | |
1618 | pte_t *pte; | |
c74df32c | 1619 | spinlock_t *ptl; |
1da177e4 | 1620 | |
c74df32c | 1621 | pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1da177e4 LT |
1622 | if (!pte) |
1623 | return -ENOMEM; | |
6606c3e0 | 1624 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1625 | do { |
1626 | BUG_ON(!pte_none(*pte)); | |
7e675137 | 1627 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
1da177e4 LT |
1628 | pfn++; |
1629 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 1630 | arch_leave_lazy_mmu_mode(); |
c74df32c | 1631 | pte_unmap_unlock(pte - 1, ptl); |
1da177e4 LT |
1632 | return 0; |
1633 | } | |
1634 | ||
1635 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
1636 | unsigned long addr, unsigned long end, | |
1637 | unsigned long pfn, pgprot_t prot) | |
1638 | { | |
1639 | pmd_t *pmd; | |
1640 | unsigned long next; | |
1641 | ||
1642 | pfn -= addr >> PAGE_SHIFT; | |
1643 | pmd = pmd_alloc(mm, pud, addr); | |
1644 | if (!pmd) | |
1645 | return -ENOMEM; | |
f66055ab | 1646 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
1da177e4 LT |
1647 | do { |
1648 | next = pmd_addr_end(addr, end); | |
1649 | if (remap_pte_range(mm, pmd, addr, next, | |
1650 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1651 | return -ENOMEM; | |
1652 | } while (pmd++, addr = next, addr != end); | |
1653 | return 0; | |
1654 | } | |
1655 | ||
1656 | static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, | |
1657 | unsigned long addr, unsigned long end, | |
1658 | unsigned long pfn, pgprot_t prot) | |
1659 | { | |
1660 | pud_t *pud; | |
1661 | unsigned long next; | |
1662 | ||
1663 | pfn -= addr >> PAGE_SHIFT; | |
1664 | pud = pud_alloc(mm, pgd, addr); | |
1665 | if (!pud) | |
1666 | return -ENOMEM; | |
1667 | do { | |
1668 | next = pud_addr_end(addr, end); | |
1669 | if (remap_pmd_range(mm, pud, addr, next, | |
1670 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1671 | return -ENOMEM; | |
1672 | } while (pud++, addr = next, addr != end); | |
1673 | return 0; | |
1674 | } | |
1675 | ||
bfa5bf6d REB |
1676 | /** |
1677 | * remap_pfn_range - remap kernel memory to userspace | |
1678 | * @vma: user vma to map to | |
1679 | * @addr: target user address to start at | |
1680 | * @pfn: physical address of kernel memory | |
1681 | * @size: size of map area | |
1682 | * @prot: page protection flags for this mapping | |
1683 | * | |
1684 | * Note: this is only safe if the mm semaphore is held when called. | |
1685 | */ | |
1da177e4 LT |
1686 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
1687 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
1688 | { | |
1689 | pgd_t *pgd; | |
1690 | unsigned long next; | |
2d15cab8 | 1691 | unsigned long end = addr + PAGE_ALIGN(size); |
1da177e4 LT |
1692 | struct mm_struct *mm = vma->vm_mm; |
1693 | int err; | |
1694 | ||
1695 | /* | |
1696 | * Physically remapped pages are special. Tell the | |
1697 | * rest of the world about it: | |
1698 | * VM_IO tells people not to look at these pages | |
1699 | * (accesses can have side effects). | |
6aab341e LT |
1700 | * VM_PFNMAP tells the core MM that the base pages are just |
1701 | * raw PFN mappings, and do not have a "struct page" associated | |
1702 | * with them. | |
314e51b9 KK |
1703 | * VM_DONTEXPAND |
1704 | * Disable vma merging and expanding with mremap(). | |
1705 | * VM_DONTDUMP | |
1706 | * Omit vma from core dump, even when VM_IO turned off. | |
fb155c16 LT |
1707 | * |
1708 | * There's a horrible special case to handle copy-on-write | |
1709 | * behaviour that some programs depend on. We mark the "original" | |
1710 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". | |
b3b9c293 | 1711 | * See vm_normal_page() for details. |
1da177e4 | 1712 | */ |
b3b9c293 KK |
1713 | if (is_cow_mapping(vma->vm_flags)) { |
1714 | if (addr != vma->vm_start || end != vma->vm_end) | |
1715 | return -EINVAL; | |
fb155c16 | 1716 | vma->vm_pgoff = pfn; |
b3b9c293 KK |
1717 | } |
1718 | ||
1719 | err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size)); | |
1720 | if (err) | |
3c8bb73a | 1721 | return -EINVAL; |
fb155c16 | 1722 | |
314e51b9 | 1723 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; |
1da177e4 LT |
1724 | |
1725 | BUG_ON(addr >= end); | |
1726 | pfn -= addr >> PAGE_SHIFT; | |
1727 | pgd = pgd_offset(mm, addr); | |
1728 | flush_cache_range(vma, addr, end); | |
1da177e4 LT |
1729 | do { |
1730 | next = pgd_addr_end(addr, end); | |
1731 | err = remap_pud_range(mm, pgd, addr, next, | |
1732 | pfn + (addr >> PAGE_SHIFT), prot); | |
1733 | if (err) | |
1734 | break; | |
1735 | } while (pgd++, addr = next, addr != end); | |
2ab64037 | 1736 | |
1737 | if (err) | |
5180da41 | 1738 | untrack_pfn(vma, pfn, PAGE_ALIGN(size)); |
2ab64037 | 1739 | |
1da177e4 LT |
1740 | return err; |
1741 | } | |
1742 | EXPORT_SYMBOL(remap_pfn_range); | |
1743 | ||
b4cbb197 LT |
1744 | /** |
1745 | * vm_iomap_memory - remap memory to userspace | |
1746 | * @vma: user vma to map to | |
1747 | * @start: start of area | |
1748 | * @len: size of area | |
1749 | * | |
1750 | * This is a simplified io_remap_pfn_range() for common driver use. The | |
1751 | * driver just needs to give us the physical memory range to be mapped, | |
1752 | * we'll figure out the rest from the vma information. | |
1753 | * | |
1754 | * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get | |
1755 | * whatever write-combining details or similar. | |
1756 | */ | |
1757 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) | |
1758 | { | |
1759 | unsigned long vm_len, pfn, pages; | |
1760 | ||
1761 | /* Check that the physical memory area passed in looks valid */ | |
1762 | if (start + len < start) | |
1763 | return -EINVAL; | |
1764 | /* | |
1765 | * You *really* shouldn't map things that aren't page-aligned, | |
1766 | * but we've historically allowed it because IO memory might | |
1767 | * just have smaller alignment. | |
1768 | */ | |
1769 | len += start & ~PAGE_MASK; | |
1770 | pfn = start >> PAGE_SHIFT; | |
1771 | pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; | |
1772 | if (pfn + pages < pfn) | |
1773 | return -EINVAL; | |
1774 | ||
1775 | /* We start the mapping 'vm_pgoff' pages into the area */ | |
1776 | if (vma->vm_pgoff > pages) | |
1777 | return -EINVAL; | |
1778 | pfn += vma->vm_pgoff; | |
1779 | pages -= vma->vm_pgoff; | |
1780 | ||
1781 | /* Can we fit all of the mapping? */ | |
1782 | vm_len = vma->vm_end - vma->vm_start; | |
1783 | if (vm_len >> PAGE_SHIFT > pages) | |
1784 | return -EINVAL; | |
1785 | ||
1786 | /* Ok, let it rip */ | |
1787 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); | |
1788 | } | |
1789 | EXPORT_SYMBOL(vm_iomap_memory); | |
1790 | ||
aee16b3c JF |
1791 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
1792 | unsigned long addr, unsigned long end, | |
1793 | pte_fn_t fn, void *data) | |
1794 | { | |
1795 | pte_t *pte; | |
1796 | int err; | |
2f569afd | 1797 | pgtable_t token; |
94909914 | 1798 | spinlock_t *uninitialized_var(ptl); |
aee16b3c JF |
1799 | |
1800 | pte = (mm == &init_mm) ? | |
1801 | pte_alloc_kernel(pmd, addr) : | |
1802 | pte_alloc_map_lock(mm, pmd, addr, &ptl); | |
1803 | if (!pte) | |
1804 | return -ENOMEM; | |
1805 | ||
1806 | BUG_ON(pmd_huge(*pmd)); | |
1807 | ||
38e0edb1 JF |
1808 | arch_enter_lazy_mmu_mode(); |
1809 | ||
2f569afd | 1810 | token = pmd_pgtable(*pmd); |
aee16b3c JF |
1811 | |
1812 | do { | |
c36987e2 | 1813 | err = fn(pte++, token, addr, data); |
aee16b3c JF |
1814 | if (err) |
1815 | break; | |
c36987e2 | 1816 | } while (addr += PAGE_SIZE, addr != end); |
aee16b3c | 1817 | |
38e0edb1 JF |
1818 | arch_leave_lazy_mmu_mode(); |
1819 | ||
aee16b3c JF |
1820 | if (mm != &init_mm) |
1821 | pte_unmap_unlock(pte-1, ptl); | |
1822 | return err; | |
1823 | } | |
1824 | ||
1825 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, | |
1826 | unsigned long addr, unsigned long end, | |
1827 | pte_fn_t fn, void *data) | |
1828 | { | |
1829 | pmd_t *pmd; | |
1830 | unsigned long next; | |
1831 | int err; | |
1832 | ||
ceb86879 AK |
1833 | BUG_ON(pud_huge(*pud)); |
1834 | ||
aee16b3c JF |
1835 | pmd = pmd_alloc(mm, pud, addr); |
1836 | if (!pmd) | |
1837 | return -ENOMEM; | |
1838 | do { | |
1839 | next = pmd_addr_end(addr, end); | |
1840 | err = apply_to_pte_range(mm, pmd, addr, next, fn, data); | |
1841 | if (err) | |
1842 | break; | |
1843 | } while (pmd++, addr = next, addr != end); | |
1844 | return err; | |
1845 | } | |
1846 | ||
1847 | static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd, | |
1848 | unsigned long addr, unsigned long end, | |
1849 | pte_fn_t fn, void *data) | |
1850 | { | |
1851 | pud_t *pud; | |
1852 | unsigned long next; | |
1853 | int err; | |
1854 | ||
1855 | pud = pud_alloc(mm, pgd, addr); | |
1856 | if (!pud) | |
1857 | return -ENOMEM; | |
1858 | do { | |
1859 | next = pud_addr_end(addr, end); | |
1860 | err = apply_to_pmd_range(mm, pud, addr, next, fn, data); | |
1861 | if (err) | |
1862 | break; | |
1863 | } while (pud++, addr = next, addr != end); | |
1864 | return err; | |
1865 | } | |
1866 | ||
1867 | /* | |
1868 | * Scan a region of virtual memory, filling in page tables as necessary | |
1869 | * and calling a provided function on each leaf page table. | |
1870 | */ | |
1871 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, | |
1872 | unsigned long size, pte_fn_t fn, void *data) | |
1873 | { | |
1874 | pgd_t *pgd; | |
1875 | unsigned long next; | |
57250a5b | 1876 | unsigned long end = addr + size; |
aee16b3c JF |
1877 | int err; |
1878 | ||
9cb65bc3 MP |
1879 | if (WARN_ON(addr >= end)) |
1880 | return -EINVAL; | |
1881 | ||
aee16b3c JF |
1882 | pgd = pgd_offset(mm, addr); |
1883 | do { | |
1884 | next = pgd_addr_end(addr, end); | |
1885 | err = apply_to_pud_range(mm, pgd, addr, next, fn, data); | |
1886 | if (err) | |
1887 | break; | |
1888 | } while (pgd++, addr = next, addr != end); | |
57250a5b | 1889 | |
aee16b3c JF |
1890 | return err; |
1891 | } | |
1892 | EXPORT_SYMBOL_GPL(apply_to_page_range); | |
1893 | ||
8f4e2101 | 1894 | /* |
9b4bdd2f KS |
1895 | * handle_pte_fault chooses page fault handler according to an entry which was |
1896 | * read non-atomically. Before making any commitment, on those architectures | |
1897 | * or configurations (e.g. i386 with PAE) which might give a mix of unmatched | |
1898 | * parts, do_swap_page must check under lock before unmapping the pte and | |
1899 | * proceeding (but do_wp_page is only called after already making such a check; | |
a335b2e1 | 1900 | * and do_anonymous_page can safely check later on). |
8f4e2101 | 1901 | */ |
4c21e2f2 | 1902 | static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, |
8f4e2101 HD |
1903 | pte_t *page_table, pte_t orig_pte) |
1904 | { | |
1905 | int same = 1; | |
1906 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) | |
1907 | if (sizeof(pte_t) > sizeof(unsigned long)) { | |
4c21e2f2 HD |
1908 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
1909 | spin_lock(ptl); | |
8f4e2101 | 1910 | same = pte_same(*page_table, orig_pte); |
4c21e2f2 | 1911 | spin_unlock(ptl); |
8f4e2101 HD |
1912 | } |
1913 | #endif | |
1914 | pte_unmap(page_table); | |
1915 | return same; | |
1916 | } | |
1917 | ||
9de455b2 | 1918 | static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) |
6aab341e | 1919 | { |
0abdd7a8 DW |
1920 | debug_dma_assert_idle(src); |
1921 | ||
6aab341e LT |
1922 | /* |
1923 | * If the source page was a PFN mapping, we don't have | |
1924 | * a "struct page" for it. We do a best-effort copy by | |
1925 | * just copying from the original user address. If that | |
1926 | * fails, we just zero-fill it. Live with it. | |
1927 | */ | |
1928 | if (unlikely(!src)) { | |
9b04c5fe | 1929 | void *kaddr = kmap_atomic(dst); |
5d2a2dbb LT |
1930 | void __user *uaddr = (void __user *)(va & PAGE_MASK); |
1931 | ||
1932 | /* | |
1933 | * This really shouldn't fail, because the page is there | |
1934 | * in the page tables. But it might just be unreadable, | |
1935 | * in which case we just give up and fill the result with | |
1936 | * zeroes. | |
1937 | */ | |
1938 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) | |
3ecb01df | 1939 | clear_page(kaddr); |
9b04c5fe | 1940 | kunmap_atomic(kaddr); |
c4ec7b0d | 1941 | flush_dcache_page(dst); |
0ed361de NP |
1942 | } else |
1943 | copy_user_highpage(dst, src, va, vma); | |
6aab341e LT |
1944 | } |
1945 | ||
c20cd45e MH |
1946 | static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma) |
1947 | { | |
1948 | struct file *vm_file = vma->vm_file; | |
1949 | ||
1950 | if (vm_file) | |
1951 | return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO; | |
1952 | ||
1953 | /* | |
1954 | * Special mappings (e.g. VDSO) do not have any file so fake | |
1955 | * a default GFP_KERNEL for them. | |
1956 | */ | |
1957 | return GFP_KERNEL; | |
1958 | } | |
1959 | ||
fb09a464 KS |
1960 | /* |
1961 | * Notify the address space that the page is about to become writable so that | |
1962 | * it can prohibit this or wait for the page to get into an appropriate state. | |
1963 | * | |
1964 | * We do this without the lock held, so that it can sleep if it needs to. | |
1965 | */ | |
1966 | static int do_page_mkwrite(struct vm_area_struct *vma, struct page *page, | |
1967 | unsigned long address) | |
1968 | { | |
1969 | struct vm_fault vmf; | |
1970 | int ret; | |
1971 | ||
1972 | vmf.virtual_address = (void __user *)(address & PAGE_MASK); | |
1973 | vmf.pgoff = page->index; | |
1974 | vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; | |
c20cd45e | 1975 | vmf.gfp_mask = __get_fault_gfp_mask(vma); |
fb09a464 | 1976 | vmf.page = page; |
2e4cdab0 | 1977 | vmf.cow_page = NULL; |
fb09a464 KS |
1978 | |
1979 | ret = vma->vm_ops->page_mkwrite(vma, &vmf); | |
1980 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) | |
1981 | return ret; | |
1982 | if (unlikely(!(ret & VM_FAULT_LOCKED))) { | |
1983 | lock_page(page); | |
1984 | if (!page->mapping) { | |
1985 | unlock_page(page); | |
1986 | return 0; /* retry */ | |
1987 | } | |
1988 | ret |= VM_FAULT_LOCKED; | |
1989 | } else | |
1990 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
1991 | return ret; | |
1992 | } | |
1993 | ||
4e047f89 SR |
1994 | /* |
1995 | * Handle write page faults for pages that can be reused in the current vma | |
1996 | * | |
1997 | * This can happen either due to the mapping being with the VM_SHARED flag, | |
1998 | * or due to us being the last reference standing to the page. In either | |
1999 | * case, all we need to do here is to mark the page as writable and update | |
2000 | * any related book-keeping. | |
2001 | */ | |
2002 | static inline int wp_page_reuse(struct mm_struct *mm, | |
2003 | struct vm_area_struct *vma, unsigned long address, | |
2004 | pte_t *page_table, spinlock_t *ptl, pte_t orig_pte, | |
2005 | struct page *page, int page_mkwrite, | |
2006 | int dirty_shared) | |
2007 | __releases(ptl) | |
2008 | { | |
2009 | pte_t entry; | |
2010 | /* | |
2011 | * Clear the pages cpupid information as the existing | |
2012 | * information potentially belongs to a now completely | |
2013 | * unrelated process. | |
2014 | */ | |
2015 | if (page) | |
2016 | page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1); | |
2017 | ||
2018 | flush_cache_page(vma, address, pte_pfn(orig_pte)); | |
2019 | entry = pte_mkyoung(orig_pte); | |
2020 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
2021 | if (ptep_set_access_flags(vma, address, page_table, entry, 1)) | |
2022 | update_mmu_cache(vma, address, page_table); | |
2023 | pte_unmap_unlock(page_table, ptl); | |
2024 | ||
2025 | if (dirty_shared) { | |
2026 | struct address_space *mapping; | |
2027 | int dirtied; | |
2028 | ||
2029 | if (!page_mkwrite) | |
2030 | lock_page(page); | |
2031 | ||
2032 | dirtied = set_page_dirty(page); | |
2033 | VM_BUG_ON_PAGE(PageAnon(page), page); | |
2034 | mapping = page->mapping; | |
2035 | unlock_page(page); | |
2036 | page_cache_release(page); | |
2037 | ||
2038 | if ((dirtied || page_mkwrite) && mapping) { | |
2039 | /* | |
2040 | * Some device drivers do not set page.mapping | |
2041 | * but still dirty their pages | |
2042 | */ | |
2043 | balance_dirty_pages_ratelimited(mapping); | |
2044 | } | |
2045 | ||
2046 | if (!page_mkwrite) | |
2047 | file_update_time(vma->vm_file); | |
2048 | } | |
2049 | ||
2050 | return VM_FAULT_WRITE; | |
2051 | } | |
2052 | ||
2f38ab2c SR |
2053 | /* |
2054 | * Handle the case of a page which we actually need to copy to a new page. | |
2055 | * | |
2056 | * Called with mmap_sem locked and the old page referenced, but | |
2057 | * without the ptl held. | |
2058 | * | |
2059 | * High level logic flow: | |
2060 | * | |
2061 | * - Allocate a page, copy the content of the old page to the new one. | |
2062 | * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. | |
2063 | * - Take the PTL. If the pte changed, bail out and release the allocated page | |
2064 | * - If the pte is still the way we remember it, update the page table and all | |
2065 | * relevant references. This includes dropping the reference the page-table | |
2066 | * held to the old page, as well as updating the rmap. | |
2067 | * - In any case, unlock the PTL and drop the reference we took to the old page. | |
2068 | */ | |
2069 | static int wp_page_copy(struct mm_struct *mm, struct vm_area_struct *vma, | |
2070 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
2071 | pte_t orig_pte, struct page *old_page) | |
2072 | { | |
2073 | struct page *new_page = NULL; | |
2074 | spinlock_t *ptl = NULL; | |
2075 | pte_t entry; | |
2076 | int page_copied = 0; | |
2077 | const unsigned long mmun_start = address & PAGE_MASK; /* For mmu_notifiers */ | |
2078 | const unsigned long mmun_end = mmun_start + PAGE_SIZE; /* For mmu_notifiers */ | |
2079 | struct mem_cgroup *memcg; | |
2080 | ||
2081 | if (unlikely(anon_vma_prepare(vma))) | |
2082 | goto oom; | |
2083 | ||
2084 | if (is_zero_pfn(pte_pfn(orig_pte))) { | |
2085 | new_page = alloc_zeroed_user_highpage_movable(vma, address); | |
2086 | if (!new_page) | |
2087 | goto oom; | |
2088 | } else { | |
2089 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); | |
2090 | if (!new_page) | |
2091 | goto oom; | |
2092 | cow_user_page(new_page, old_page, address, vma); | |
2093 | } | |
2f38ab2c | 2094 | |
f627c2f5 | 2095 | if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg, false)) |
2f38ab2c SR |
2096 | goto oom_free_new; |
2097 | ||
eb3c24f3 MG |
2098 | __SetPageUptodate(new_page); |
2099 | ||
2f38ab2c SR |
2100 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
2101 | ||
2102 | /* | |
2103 | * Re-check the pte - we dropped the lock | |
2104 | */ | |
2105 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); | |
2106 | if (likely(pte_same(*page_table, orig_pte))) { | |
2107 | if (old_page) { | |
2108 | if (!PageAnon(old_page)) { | |
eca56ff9 JM |
2109 | dec_mm_counter_fast(mm, |
2110 | mm_counter_file(old_page)); | |
2f38ab2c SR |
2111 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
2112 | } | |
2113 | } else { | |
2114 | inc_mm_counter_fast(mm, MM_ANONPAGES); | |
2115 | } | |
2116 | flush_cache_page(vma, address, pte_pfn(orig_pte)); | |
2117 | entry = mk_pte(new_page, vma->vm_page_prot); | |
2118 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
2119 | /* | |
2120 | * Clear the pte entry and flush it first, before updating the | |
2121 | * pte with the new entry. This will avoid a race condition | |
2122 | * seen in the presence of one thread doing SMC and another | |
2123 | * thread doing COW. | |
2124 | */ | |
2125 | ptep_clear_flush_notify(vma, address, page_table); | |
d281ee61 | 2126 | page_add_new_anon_rmap(new_page, vma, address, false); |
f627c2f5 | 2127 | mem_cgroup_commit_charge(new_page, memcg, false, false); |
2f38ab2c SR |
2128 | lru_cache_add_active_or_unevictable(new_page, vma); |
2129 | /* | |
2130 | * We call the notify macro here because, when using secondary | |
2131 | * mmu page tables (such as kvm shadow page tables), we want the | |
2132 | * new page to be mapped directly into the secondary page table. | |
2133 | */ | |
2134 | set_pte_at_notify(mm, address, page_table, entry); | |
2135 | update_mmu_cache(vma, address, page_table); | |
2136 | if (old_page) { | |
2137 | /* | |
2138 | * Only after switching the pte to the new page may | |
2139 | * we remove the mapcount here. Otherwise another | |
2140 | * process may come and find the rmap count decremented | |
2141 | * before the pte is switched to the new page, and | |
2142 | * "reuse" the old page writing into it while our pte | |
2143 | * here still points into it and can be read by other | |
2144 | * threads. | |
2145 | * | |
2146 | * The critical issue is to order this | |
2147 | * page_remove_rmap with the ptp_clear_flush above. | |
2148 | * Those stores are ordered by (if nothing else,) | |
2149 | * the barrier present in the atomic_add_negative | |
2150 | * in page_remove_rmap. | |
2151 | * | |
2152 | * Then the TLB flush in ptep_clear_flush ensures that | |
2153 | * no process can access the old page before the | |
2154 | * decremented mapcount is visible. And the old page | |
2155 | * cannot be reused until after the decremented | |
2156 | * mapcount is visible. So transitively, TLBs to | |
2157 | * old page will be flushed before it can be reused. | |
2158 | */ | |
d281ee61 | 2159 | page_remove_rmap(old_page, false); |
2f38ab2c SR |
2160 | } |
2161 | ||
2162 | /* Free the old page.. */ | |
2163 | new_page = old_page; | |
2164 | page_copied = 1; | |
2165 | } else { | |
f627c2f5 | 2166 | mem_cgroup_cancel_charge(new_page, memcg, false); |
2f38ab2c SR |
2167 | } |
2168 | ||
2169 | if (new_page) | |
2170 | page_cache_release(new_page); | |
2171 | ||
2172 | pte_unmap_unlock(page_table, ptl); | |
2173 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); | |
e90309c9 | 2174 | if (old_page) { |
2f38ab2c SR |
2175 | /* |
2176 | * Don't let another task, with possibly unlocked vma, | |
2177 | * keep the mlocked page. | |
2178 | */ | |
2179 | if (page_copied && (vma->vm_flags & VM_LOCKED)) { | |
2180 | lock_page(old_page); /* LRU manipulation */ | |
e90309c9 KS |
2181 | if (PageMlocked(old_page)) |
2182 | munlock_vma_page(old_page); | |
2f38ab2c SR |
2183 | unlock_page(old_page); |
2184 | } | |
2185 | page_cache_release(old_page); | |
2186 | } | |
2187 | return page_copied ? VM_FAULT_WRITE : 0; | |
2188 | oom_free_new: | |
2189 | page_cache_release(new_page); | |
2190 | oom: | |
2191 | if (old_page) | |
2192 | page_cache_release(old_page); | |
2193 | return VM_FAULT_OOM; | |
2194 | } | |
2195 | ||
dd906184 BH |
2196 | /* |
2197 | * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED | |
2198 | * mapping | |
2199 | */ | |
2200 | static int wp_pfn_shared(struct mm_struct *mm, | |
2201 | struct vm_area_struct *vma, unsigned long address, | |
2202 | pte_t *page_table, spinlock_t *ptl, pte_t orig_pte, | |
2203 | pmd_t *pmd) | |
2204 | { | |
2205 | if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { | |
2206 | struct vm_fault vmf = { | |
2207 | .page = NULL, | |
2208 | .pgoff = linear_page_index(vma, address), | |
2209 | .virtual_address = (void __user *)(address & PAGE_MASK), | |
2210 | .flags = FAULT_FLAG_WRITE | FAULT_FLAG_MKWRITE, | |
2211 | }; | |
2212 | int ret; | |
2213 | ||
2214 | pte_unmap_unlock(page_table, ptl); | |
2215 | ret = vma->vm_ops->pfn_mkwrite(vma, &vmf); | |
2216 | if (ret & VM_FAULT_ERROR) | |
2217 | return ret; | |
2218 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); | |
2219 | /* | |
2220 | * We might have raced with another page fault while we | |
2221 | * released the pte_offset_map_lock. | |
2222 | */ | |
2223 | if (!pte_same(*page_table, orig_pte)) { | |
2224 | pte_unmap_unlock(page_table, ptl); | |
2225 | return 0; | |
2226 | } | |
2227 | } | |
2228 | return wp_page_reuse(mm, vma, address, page_table, ptl, orig_pte, | |
2229 | NULL, 0, 0); | |
2230 | } | |
2231 | ||
93e478d4 SR |
2232 | static int wp_page_shared(struct mm_struct *mm, struct vm_area_struct *vma, |
2233 | unsigned long address, pte_t *page_table, | |
2234 | pmd_t *pmd, spinlock_t *ptl, pte_t orig_pte, | |
2235 | struct page *old_page) | |
2236 | __releases(ptl) | |
2237 | { | |
2238 | int page_mkwrite = 0; | |
2239 | ||
2240 | page_cache_get(old_page); | |
2241 | ||
93e478d4 SR |
2242 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
2243 | int tmp; | |
2244 | ||
2245 | pte_unmap_unlock(page_table, ptl); | |
2246 | tmp = do_page_mkwrite(vma, old_page, address); | |
2247 | if (unlikely(!tmp || (tmp & | |
2248 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
2249 | page_cache_release(old_page); | |
2250 | return tmp; | |
2251 | } | |
2252 | /* | |
2253 | * Since we dropped the lock we need to revalidate | |
2254 | * the PTE as someone else may have changed it. If | |
2255 | * they did, we just return, as we can count on the | |
2256 | * MMU to tell us if they didn't also make it writable. | |
2257 | */ | |
2258 | page_table = pte_offset_map_lock(mm, pmd, address, | |
2259 | &ptl); | |
2260 | if (!pte_same(*page_table, orig_pte)) { | |
2261 | unlock_page(old_page); | |
2262 | pte_unmap_unlock(page_table, ptl); | |
2263 | page_cache_release(old_page); | |
2264 | return 0; | |
2265 | } | |
2266 | page_mkwrite = 1; | |
2267 | } | |
2268 | ||
2269 | return wp_page_reuse(mm, vma, address, page_table, ptl, | |
2270 | orig_pte, old_page, page_mkwrite, 1); | |
2271 | } | |
2272 | ||
1da177e4 LT |
2273 | /* |
2274 | * This routine handles present pages, when users try to write | |
2275 | * to a shared page. It is done by copying the page to a new address | |
2276 | * and decrementing the shared-page counter for the old page. | |
2277 | * | |
1da177e4 LT |
2278 | * Note that this routine assumes that the protection checks have been |
2279 | * done by the caller (the low-level page fault routine in most cases). | |
2280 | * Thus we can safely just mark it writable once we've done any necessary | |
2281 | * COW. | |
2282 | * | |
2283 | * We also mark the page dirty at this point even though the page will | |
2284 | * change only once the write actually happens. This avoids a few races, | |
2285 | * and potentially makes it more efficient. | |
2286 | * | |
8f4e2101 HD |
2287 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2288 | * but allow concurrent faults), with pte both mapped and locked. | |
2289 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2290 | */ |
65500d23 HD |
2291 | static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2292 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
8f4e2101 | 2293 | spinlock_t *ptl, pte_t orig_pte) |
e6219ec8 | 2294 | __releases(ptl) |
1da177e4 | 2295 | { |
2f38ab2c | 2296 | struct page *old_page; |
1da177e4 | 2297 | |
6aab341e | 2298 | old_page = vm_normal_page(vma, address, orig_pte); |
251b97f5 PZ |
2299 | if (!old_page) { |
2300 | /* | |
64e45507 PF |
2301 | * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a |
2302 | * VM_PFNMAP VMA. | |
251b97f5 PZ |
2303 | * |
2304 | * We should not cow pages in a shared writeable mapping. | |
dd906184 | 2305 | * Just mark the pages writable and/or call ops->pfn_mkwrite. |
251b97f5 PZ |
2306 | */ |
2307 | if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == | |
2308 | (VM_WRITE|VM_SHARED)) | |
dd906184 BH |
2309 | return wp_pfn_shared(mm, vma, address, page_table, ptl, |
2310 | orig_pte, pmd); | |
2f38ab2c SR |
2311 | |
2312 | pte_unmap_unlock(page_table, ptl); | |
2313 | return wp_page_copy(mm, vma, address, page_table, pmd, | |
2314 | orig_pte, old_page); | |
251b97f5 | 2315 | } |
1da177e4 | 2316 | |
d08b3851 | 2317 | /* |
ee6a6457 PZ |
2318 | * Take out anonymous pages first, anonymous shared vmas are |
2319 | * not dirty accountable. | |
d08b3851 | 2320 | */ |
9a840895 | 2321 | if (PageAnon(old_page) && !PageKsm(old_page)) { |
ab967d86 HD |
2322 | if (!trylock_page(old_page)) { |
2323 | page_cache_get(old_page); | |
2324 | pte_unmap_unlock(page_table, ptl); | |
2325 | lock_page(old_page); | |
2326 | page_table = pte_offset_map_lock(mm, pmd, address, | |
2327 | &ptl); | |
2328 | if (!pte_same(*page_table, orig_pte)) { | |
2329 | unlock_page(old_page); | |
28766805 SR |
2330 | pte_unmap_unlock(page_table, ptl); |
2331 | page_cache_release(old_page); | |
2332 | return 0; | |
ab967d86 HD |
2333 | } |
2334 | page_cache_release(old_page); | |
ee6a6457 | 2335 | } |
b009c024 | 2336 | if (reuse_swap_page(old_page)) { |
c44b6743 RR |
2337 | /* |
2338 | * The page is all ours. Move it to our anon_vma so | |
2339 | * the rmap code will not search our parent or siblings. | |
2340 | * Protected against the rmap code by the page lock. | |
2341 | */ | |
2342 | page_move_anon_rmap(old_page, vma, address); | |
b009c024 | 2343 | unlock_page(old_page); |
4e047f89 SR |
2344 | return wp_page_reuse(mm, vma, address, page_table, ptl, |
2345 | orig_pte, old_page, 0, 0); | |
b009c024 | 2346 | } |
ab967d86 | 2347 | unlock_page(old_page); |
ee6a6457 | 2348 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
d08b3851 | 2349 | (VM_WRITE|VM_SHARED))) { |
93e478d4 SR |
2350 | return wp_page_shared(mm, vma, address, page_table, pmd, |
2351 | ptl, orig_pte, old_page); | |
1da177e4 | 2352 | } |
1da177e4 LT |
2353 | |
2354 | /* | |
2355 | * Ok, we need to copy. Oh, well.. | |
2356 | */ | |
b5810039 | 2357 | page_cache_get(old_page); |
28766805 | 2358 | |
8f4e2101 | 2359 | pte_unmap_unlock(page_table, ptl); |
2f38ab2c SR |
2360 | return wp_page_copy(mm, vma, address, page_table, pmd, |
2361 | orig_pte, old_page); | |
1da177e4 LT |
2362 | } |
2363 | ||
97a89413 | 2364 | static void unmap_mapping_range_vma(struct vm_area_struct *vma, |
1da177e4 LT |
2365 | unsigned long start_addr, unsigned long end_addr, |
2366 | struct zap_details *details) | |
2367 | { | |
f5cc4eef | 2368 | zap_page_range_single(vma, start_addr, end_addr - start_addr, details); |
1da177e4 LT |
2369 | } |
2370 | ||
6b2dbba8 | 2371 | static inline void unmap_mapping_range_tree(struct rb_root *root, |
1da177e4 LT |
2372 | struct zap_details *details) |
2373 | { | |
2374 | struct vm_area_struct *vma; | |
1da177e4 LT |
2375 | pgoff_t vba, vea, zba, zea; |
2376 | ||
6b2dbba8 | 2377 | vma_interval_tree_foreach(vma, root, |
1da177e4 | 2378 | details->first_index, details->last_index) { |
1da177e4 LT |
2379 | |
2380 | vba = vma->vm_pgoff; | |
d6e93217 | 2381 | vea = vba + vma_pages(vma) - 1; |
1da177e4 LT |
2382 | /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */ |
2383 | zba = details->first_index; | |
2384 | if (zba < vba) | |
2385 | zba = vba; | |
2386 | zea = details->last_index; | |
2387 | if (zea > vea) | |
2388 | zea = vea; | |
2389 | ||
97a89413 | 2390 | unmap_mapping_range_vma(vma, |
1da177e4 LT |
2391 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, |
2392 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, | |
97a89413 | 2393 | details); |
1da177e4 LT |
2394 | } |
2395 | } | |
2396 | ||
1da177e4 | 2397 | /** |
8a5f14a2 KS |
2398 | * unmap_mapping_range - unmap the portion of all mmaps in the specified |
2399 | * address_space corresponding to the specified page range in the underlying | |
2400 | * file. | |
2401 | * | |
3d41088f | 2402 | * @mapping: the address space containing mmaps to be unmapped. |
1da177e4 LT |
2403 | * @holebegin: byte in first page to unmap, relative to the start of |
2404 | * the underlying file. This will be rounded down to a PAGE_SIZE | |
25d9e2d1 | 2405 | * boundary. Note that this is different from truncate_pagecache(), which |
1da177e4 LT |
2406 | * must keep the partial page. In contrast, we must get rid of |
2407 | * partial pages. | |
2408 | * @holelen: size of prospective hole in bytes. This will be rounded | |
2409 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the | |
2410 | * end of the file. | |
2411 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; | |
2412 | * but 0 when invalidating pagecache, don't throw away private data. | |
2413 | */ | |
2414 | void unmap_mapping_range(struct address_space *mapping, | |
2415 | loff_t const holebegin, loff_t const holelen, int even_cows) | |
2416 | { | |
2417 | struct zap_details details; | |
2418 | pgoff_t hba = holebegin >> PAGE_SHIFT; | |
2419 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2420 | ||
2421 | /* Check for overflow. */ | |
2422 | if (sizeof(holelen) > sizeof(hlen)) { | |
2423 | long long holeend = | |
2424 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2425 | if (holeend & ~(long long)ULONG_MAX) | |
2426 | hlen = ULONG_MAX - hba + 1; | |
2427 | } | |
2428 | ||
2429 | details.check_mapping = even_cows? NULL: mapping; | |
1da177e4 LT |
2430 | details.first_index = hba; |
2431 | details.last_index = hba + hlen - 1; | |
2432 | if (details.last_index < details.first_index) | |
2433 | details.last_index = ULONG_MAX; | |
1da177e4 | 2434 | |
0f90cc66 RZ |
2435 | |
2436 | /* DAX uses i_mmap_lock to serialise file truncate vs page fault */ | |
46c043ed | 2437 | i_mmap_lock_write(mapping); |
6b2dbba8 | 2438 | if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap))) |
1da177e4 | 2439 | unmap_mapping_range_tree(&mapping->i_mmap, &details); |
46c043ed | 2440 | i_mmap_unlock_write(mapping); |
1da177e4 LT |
2441 | } |
2442 | EXPORT_SYMBOL(unmap_mapping_range); | |
2443 | ||
1da177e4 | 2444 | /* |
8f4e2101 HD |
2445 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2446 | * but allow concurrent faults), and pte mapped but not yet locked. | |
9a95f3cf PC |
2447 | * We return with pte unmapped and unlocked. |
2448 | * | |
2449 | * We return with the mmap_sem locked or unlocked in the same cases | |
2450 | * as does filemap_fault(). | |
1da177e4 | 2451 | */ |
65500d23 HD |
2452 | static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2453 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
30c9f3a9 | 2454 | unsigned int flags, pte_t orig_pte) |
1da177e4 | 2455 | { |
8f4e2101 | 2456 | spinlock_t *ptl; |
56f31801 | 2457 | struct page *page, *swapcache; |
00501b53 | 2458 | struct mem_cgroup *memcg; |
65500d23 | 2459 | swp_entry_t entry; |
1da177e4 | 2460 | pte_t pte; |
d065bd81 | 2461 | int locked; |
ad8c2ee8 | 2462 | int exclusive = 0; |
83c54070 | 2463 | int ret = 0; |
1da177e4 | 2464 | |
4c21e2f2 | 2465 | if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) |
8f4e2101 | 2466 | goto out; |
65500d23 HD |
2467 | |
2468 | entry = pte_to_swp_entry(orig_pte); | |
d1737fdb AK |
2469 | if (unlikely(non_swap_entry(entry))) { |
2470 | if (is_migration_entry(entry)) { | |
2471 | migration_entry_wait(mm, pmd, address); | |
2472 | } else if (is_hwpoison_entry(entry)) { | |
2473 | ret = VM_FAULT_HWPOISON; | |
2474 | } else { | |
2475 | print_bad_pte(vma, address, orig_pte, NULL); | |
d99be1a8 | 2476 | ret = VM_FAULT_SIGBUS; |
d1737fdb | 2477 | } |
0697212a CL |
2478 | goto out; |
2479 | } | |
0ff92245 | 2480 | delayacct_set_flag(DELAYACCT_PF_SWAPIN); |
1da177e4 LT |
2481 | page = lookup_swap_cache(entry); |
2482 | if (!page) { | |
02098fea HD |
2483 | page = swapin_readahead(entry, |
2484 | GFP_HIGHUSER_MOVABLE, vma, address); | |
1da177e4 LT |
2485 | if (!page) { |
2486 | /* | |
8f4e2101 HD |
2487 | * Back out if somebody else faulted in this pte |
2488 | * while we released the pte lock. | |
1da177e4 | 2489 | */ |
8f4e2101 | 2490 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
1da177e4 LT |
2491 | if (likely(pte_same(*page_table, orig_pte))) |
2492 | ret = VM_FAULT_OOM; | |
0ff92245 | 2493 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
65500d23 | 2494 | goto unlock; |
1da177e4 LT |
2495 | } |
2496 | ||
2497 | /* Had to read the page from swap area: Major fault */ | |
2498 | ret = VM_FAULT_MAJOR; | |
f8891e5e | 2499 | count_vm_event(PGMAJFAULT); |
456f998e | 2500 | mem_cgroup_count_vm_event(mm, PGMAJFAULT); |
d1737fdb | 2501 | } else if (PageHWPoison(page)) { |
71f72525 WF |
2502 | /* |
2503 | * hwpoisoned dirty swapcache pages are kept for killing | |
2504 | * owner processes (which may be unknown at hwpoison time) | |
2505 | */ | |
d1737fdb AK |
2506 | ret = VM_FAULT_HWPOISON; |
2507 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); | |
56f31801 | 2508 | swapcache = page; |
4779cb31 | 2509 | goto out_release; |
1da177e4 LT |
2510 | } |
2511 | ||
56f31801 | 2512 | swapcache = page; |
d065bd81 | 2513 | locked = lock_page_or_retry(page, mm, flags); |
e709ffd6 | 2514 | |
073e587e | 2515 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
d065bd81 ML |
2516 | if (!locked) { |
2517 | ret |= VM_FAULT_RETRY; | |
2518 | goto out_release; | |
2519 | } | |
073e587e | 2520 | |
4969c119 | 2521 | /* |
31c4a3d3 HD |
2522 | * Make sure try_to_free_swap or reuse_swap_page or swapoff did not |
2523 | * release the swapcache from under us. The page pin, and pte_same | |
2524 | * test below, are not enough to exclude that. Even if it is still | |
2525 | * swapcache, we need to check that the page's swap has not changed. | |
4969c119 | 2526 | */ |
31c4a3d3 | 2527 | if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val)) |
4969c119 AA |
2528 | goto out_page; |
2529 | ||
cbf86cfe HD |
2530 | page = ksm_might_need_to_copy(page, vma, address); |
2531 | if (unlikely(!page)) { | |
2532 | ret = VM_FAULT_OOM; | |
2533 | page = swapcache; | |
cbf86cfe | 2534 | goto out_page; |
5ad64688 HD |
2535 | } |
2536 | ||
f627c2f5 | 2537 | if (mem_cgroup_try_charge(page, mm, GFP_KERNEL, &memcg, false)) { |
8a9f3ccd | 2538 | ret = VM_FAULT_OOM; |
bc43f75c | 2539 | goto out_page; |
8a9f3ccd BS |
2540 | } |
2541 | ||
1da177e4 | 2542 | /* |
8f4e2101 | 2543 | * Back out if somebody else already faulted in this pte. |
1da177e4 | 2544 | */ |
8f4e2101 | 2545 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
9e9bef07 | 2546 | if (unlikely(!pte_same(*page_table, orig_pte))) |
b8107480 | 2547 | goto out_nomap; |
b8107480 KK |
2548 | |
2549 | if (unlikely(!PageUptodate(page))) { | |
2550 | ret = VM_FAULT_SIGBUS; | |
2551 | goto out_nomap; | |
1da177e4 LT |
2552 | } |
2553 | ||
8c7c6e34 KH |
2554 | /* |
2555 | * The page isn't present yet, go ahead with the fault. | |
2556 | * | |
2557 | * Be careful about the sequence of operations here. | |
2558 | * To get its accounting right, reuse_swap_page() must be called | |
2559 | * while the page is counted on swap but not yet in mapcount i.e. | |
2560 | * before page_add_anon_rmap() and swap_free(); try_to_free_swap() | |
2561 | * must be called after the swap_free(), or it will never succeed. | |
8c7c6e34 | 2562 | */ |
1da177e4 | 2563 | |
34e55232 | 2564 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
b084d435 | 2565 | dec_mm_counter_fast(mm, MM_SWAPENTS); |
1da177e4 | 2566 | pte = mk_pte(page, vma->vm_page_prot); |
30c9f3a9 | 2567 | if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) { |
1da177e4 | 2568 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); |
30c9f3a9 | 2569 | flags &= ~FAULT_FLAG_WRITE; |
9a5b489b | 2570 | ret |= VM_FAULT_WRITE; |
d281ee61 | 2571 | exclusive = RMAP_EXCLUSIVE; |
1da177e4 | 2572 | } |
1da177e4 | 2573 | flush_icache_page(vma, page); |
179ef71c CG |
2574 | if (pte_swp_soft_dirty(orig_pte)) |
2575 | pte = pte_mksoft_dirty(pte); | |
1da177e4 | 2576 | set_pte_at(mm, address, page_table, pte); |
00501b53 | 2577 | if (page == swapcache) { |
af34770e | 2578 | do_page_add_anon_rmap(page, vma, address, exclusive); |
f627c2f5 | 2579 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 2580 | } else { /* ksm created a completely new copy */ |
d281ee61 | 2581 | page_add_new_anon_rmap(page, vma, address, false); |
f627c2f5 | 2582 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 JW |
2583 | lru_cache_add_active_or_unevictable(page, vma); |
2584 | } | |
1da177e4 | 2585 | |
c475a8ab | 2586 | swap_free(entry); |
5ccc5aba VD |
2587 | if (mem_cgroup_swap_full(page) || |
2588 | (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) | |
a2c43eed | 2589 | try_to_free_swap(page); |
c475a8ab | 2590 | unlock_page(page); |
56f31801 | 2591 | if (page != swapcache) { |
4969c119 AA |
2592 | /* |
2593 | * Hold the lock to avoid the swap entry to be reused | |
2594 | * until we take the PT lock for the pte_same() check | |
2595 | * (to avoid false positives from pte_same). For | |
2596 | * further safety release the lock after the swap_free | |
2597 | * so that the swap count won't change under a | |
2598 | * parallel locked swapcache. | |
2599 | */ | |
2600 | unlock_page(swapcache); | |
2601 | page_cache_release(swapcache); | |
2602 | } | |
c475a8ab | 2603 | |
30c9f3a9 | 2604 | if (flags & FAULT_FLAG_WRITE) { |
61469f1d HD |
2605 | ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte); |
2606 | if (ret & VM_FAULT_ERROR) | |
2607 | ret &= VM_FAULT_ERROR; | |
1da177e4 LT |
2608 | goto out; |
2609 | } | |
2610 | ||
2611 | /* No need to invalidate - it was non-present before */ | |
4b3073e1 | 2612 | update_mmu_cache(vma, address, page_table); |
65500d23 | 2613 | unlock: |
8f4e2101 | 2614 | pte_unmap_unlock(page_table, ptl); |
1da177e4 LT |
2615 | out: |
2616 | return ret; | |
b8107480 | 2617 | out_nomap: |
f627c2f5 | 2618 | mem_cgroup_cancel_charge(page, memcg, false); |
8f4e2101 | 2619 | pte_unmap_unlock(page_table, ptl); |
bc43f75c | 2620 | out_page: |
b8107480 | 2621 | unlock_page(page); |
4779cb31 | 2622 | out_release: |
b8107480 | 2623 | page_cache_release(page); |
56f31801 | 2624 | if (page != swapcache) { |
4969c119 AA |
2625 | unlock_page(swapcache); |
2626 | page_cache_release(swapcache); | |
2627 | } | |
65500d23 | 2628 | return ret; |
1da177e4 LT |
2629 | } |
2630 | ||
320b2b8d | 2631 | /* |
8ca3eb08 TL |
2632 | * This is like a special single-page "expand_{down|up}wards()", |
2633 | * except we must first make sure that 'address{-|+}PAGE_SIZE' | |
320b2b8d | 2634 | * doesn't hit another vma. |
320b2b8d LT |
2635 | */ |
2636 | static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address) | |
2637 | { | |
2638 | address &= PAGE_MASK; | |
2639 | if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) { | |
0e8e50e2 LT |
2640 | struct vm_area_struct *prev = vma->vm_prev; |
2641 | ||
2642 | /* | |
2643 | * Is there a mapping abutting this one below? | |
2644 | * | |
2645 | * That's only ok if it's the same stack mapping | |
2646 | * that has gotten split.. | |
2647 | */ | |
2648 | if (prev && prev->vm_end == address) | |
2649 | return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM; | |
320b2b8d | 2650 | |
fee7e49d | 2651 | return expand_downwards(vma, address - PAGE_SIZE); |
320b2b8d | 2652 | } |
8ca3eb08 TL |
2653 | if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) { |
2654 | struct vm_area_struct *next = vma->vm_next; | |
2655 | ||
2656 | /* As VM_GROWSDOWN but s/below/above/ */ | |
2657 | if (next && next->vm_start == address + PAGE_SIZE) | |
2658 | return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM; | |
2659 | ||
fee7e49d | 2660 | return expand_upwards(vma, address + PAGE_SIZE); |
8ca3eb08 | 2661 | } |
320b2b8d LT |
2662 | return 0; |
2663 | } | |
2664 | ||
1da177e4 | 2665 | /* |
8f4e2101 HD |
2666 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2667 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2668 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2669 | */ |
65500d23 HD |
2670 | static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2671 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
30c9f3a9 | 2672 | unsigned int flags) |
1da177e4 | 2673 | { |
00501b53 | 2674 | struct mem_cgroup *memcg; |
8f4e2101 HD |
2675 | struct page *page; |
2676 | spinlock_t *ptl; | |
1da177e4 | 2677 | pte_t entry; |
1da177e4 | 2678 | |
11ac5524 LT |
2679 | pte_unmap(page_table); |
2680 | ||
6b7339f4 KS |
2681 | /* File mapping without ->vm_ops ? */ |
2682 | if (vma->vm_flags & VM_SHARED) | |
2683 | return VM_FAULT_SIGBUS; | |
2684 | ||
11ac5524 LT |
2685 | /* Check if we need to add a guard page to the stack */ |
2686 | if (check_stack_guard_page(vma, address) < 0) | |
9c145c56 | 2687 | return VM_FAULT_SIGSEGV; |
320b2b8d | 2688 | |
11ac5524 | 2689 | /* Use the zero-page for reads */ |
593befa6 | 2690 | if (!(flags & FAULT_FLAG_WRITE) && !mm_forbids_zeropage(mm)) { |
62eede62 HD |
2691 | entry = pte_mkspecial(pfn_pte(my_zero_pfn(address), |
2692 | vma->vm_page_prot)); | |
11ac5524 | 2693 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
a13ea5b7 HD |
2694 | if (!pte_none(*page_table)) |
2695 | goto unlock; | |
6b251fc9 AA |
2696 | /* Deliver the page fault to userland, check inside PT lock */ |
2697 | if (userfaultfd_missing(vma)) { | |
2698 | pte_unmap_unlock(page_table, ptl); | |
2699 | return handle_userfault(vma, address, flags, | |
2700 | VM_UFFD_MISSING); | |
2701 | } | |
a13ea5b7 HD |
2702 | goto setpte; |
2703 | } | |
2704 | ||
557ed1fa | 2705 | /* Allocate our own private page. */ |
557ed1fa NP |
2706 | if (unlikely(anon_vma_prepare(vma))) |
2707 | goto oom; | |
2708 | page = alloc_zeroed_user_highpage_movable(vma, address); | |
2709 | if (!page) | |
2710 | goto oom; | |
eb3c24f3 | 2711 | |
f627c2f5 | 2712 | if (mem_cgroup_try_charge(page, mm, GFP_KERNEL, &memcg, false)) |
eb3c24f3 MG |
2713 | goto oom_free_page; |
2714 | ||
52f37629 MK |
2715 | /* |
2716 | * The memory barrier inside __SetPageUptodate makes sure that | |
2717 | * preceeding stores to the page contents become visible before | |
2718 | * the set_pte_at() write. | |
2719 | */ | |
0ed361de | 2720 | __SetPageUptodate(page); |
8f4e2101 | 2721 | |
557ed1fa | 2722 | entry = mk_pte(page, vma->vm_page_prot); |
1ac0cb5d HD |
2723 | if (vma->vm_flags & VM_WRITE) |
2724 | entry = pte_mkwrite(pte_mkdirty(entry)); | |
1da177e4 | 2725 | |
557ed1fa | 2726 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
1c2fb7a4 | 2727 | if (!pte_none(*page_table)) |
557ed1fa | 2728 | goto release; |
9ba69294 | 2729 | |
6b251fc9 AA |
2730 | /* Deliver the page fault to userland, check inside PT lock */ |
2731 | if (userfaultfd_missing(vma)) { | |
2732 | pte_unmap_unlock(page_table, ptl); | |
f627c2f5 | 2733 | mem_cgroup_cancel_charge(page, memcg, false); |
6b251fc9 AA |
2734 | page_cache_release(page); |
2735 | return handle_userfault(vma, address, flags, | |
2736 | VM_UFFD_MISSING); | |
2737 | } | |
2738 | ||
34e55232 | 2739 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
d281ee61 | 2740 | page_add_new_anon_rmap(page, vma, address, false); |
f627c2f5 | 2741 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 | 2742 | lru_cache_add_active_or_unevictable(page, vma); |
a13ea5b7 | 2743 | setpte: |
65500d23 | 2744 | set_pte_at(mm, address, page_table, entry); |
1da177e4 LT |
2745 | |
2746 | /* No need to invalidate - it was non-present before */ | |
4b3073e1 | 2747 | update_mmu_cache(vma, address, page_table); |
65500d23 | 2748 | unlock: |
8f4e2101 | 2749 | pte_unmap_unlock(page_table, ptl); |
83c54070 | 2750 | return 0; |
8f4e2101 | 2751 | release: |
f627c2f5 | 2752 | mem_cgroup_cancel_charge(page, memcg, false); |
8f4e2101 HD |
2753 | page_cache_release(page); |
2754 | goto unlock; | |
8a9f3ccd | 2755 | oom_free_page: |
6dbf6d3b | 2756 | page_cache_release(page); |
65500d23 | 2757 | oom: |
1da177e4 LT |
2758 | return VM_FAULT_OOM; |
2759 | } | |
2760 | ||
9a95f3cf PC |
2761 | /* |
2762 | * The mmap_sem must have been held on entry, and may have been | |
2763 | * released depending on flags and vma->vm_ops->fault() return value. | |
2764 | * See filemap_fault() and __lock_page_retry(). | |
2765 | */ | |
7eae74af | 2766 | static int __do_fault(struct vm_area_struct *vma, unsigned long address, |
2e4cdab0 MW |
2767 | pgoff_t pgoff, unsigned int flags, |
2768 | struct page *cow_page, struct page **page) | |
7eae74af KS |
2769 | { |
2770 | struct vm_fault vmf; | |
2771 | int ret; | |
2772 | ||
2773 | vmf.virtual_address = (void __user *)(address & PAGE_MASK); | |
2774 | vmf.pgoff = pgoff; | |
2775 | vmf.flags = flags; | |
2776 | vmf.page = NULL; | |
c20cd45e | 2777 | vmf.gfp_mask = __get_fault_gfp_mask(vma); |
2e4cdab0 | 2778 | vmf.cow_page = cow_page; |
7eae74af KS |
2779 | |
2780 | ret = vma->vm_ops->fault(vma, &vmf); | |
2781 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) | |
2782 | return ret; | |
2e4cdab0 MW |
2783 | if (!vmf.page) |
2784 | goto out; | |
7eae74af KS |
2785 | |
2786 | if (unlikely(PageHWPoison(vmf.page))) { | |
2787 | if (ret & VM_FAULT_LOCKED) | |
2788 | unlock_page(vmf.page); | |
2789 | page_cache_release(vmf.page); | |
2790 | return VM_FAULT_HWPOISON; | |
2791 | } | |
2792 | ||
2793 | if (unlikely(!(ret & VM_FAULT_LOCKED))) | |
2794 | lock_page(vmf.page); | |
2795 | else | |
2796 | VM_BUG_ON_PAGE(!PageLocked(vmf.page), vmf.page); | |
2797 | ||
2e4cdab0 | 2798 | out: |
7eae74af KS |
2799 | *page = vmf.page; |
2800 | return ret; | |
2801 | } | |
2802 | ||
8c6e50b0 KS |
2803 | /** |
2804 | * do_set_pte - setup new PTE entry for given page and add reverse page mapping. | |
2805 | * | |
2806 | * @vma: virtual memory area | |
2807 | * @address: user virtual address | |
2808 | * @page: page to map | |
2809 | * @pte: pointer to target page table entry | |
2810 | * @write: true, if new entry is writable | |
2811 | * @anon: true, if it's anonymous page | |
2812 | * | |
2813 | * Caller must hold page table lock relevant for @pte. | |
2814 | * | |
2815 | * Target users are page handler itself and implementations of | |
2816 | * vm_ops->map_pages. | |
2817 | */ | |
2818 | void do_set_pte(struct vm_area_struct *vma, unsigned long address, | |
3bb97794 KS |
2819 | struct page *page, pte_t *pte, bool write, bool anon) |
2820 | { | |
2821 | pte_t entry; | |
2822 | ||
2823 | flush_icache_page(vma, page); | |
2824 | entry = mk_pte(page, vma->vm_page_prot); | |
2825 | if (write) | |
2826 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
3bb97794 KS |
2827 | if (anon) { |
2828 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); | |
d281ee61 | 2829 | page_add_new_anon_rmap(page, vma, address, false); |
3bb97794 | 2830 | } else { |
eca56ff9 | 2831 | inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page)); |
3bb97794 KS |
2832 | page_add_file_rmap(page); |
2833 | } | |
2834 | set_pte_at(vma->vm_mm, address, pte, entry); | |
2835 | ||
2836 | /* no need to invalidate: a not-present page won't be cached */ | |
2837 | update_mmu_cache(vma, address, pte); | |
2838 | } | |
2839 | ||
3a91053a KS |
2840 | static unsigned long fault_around_bytes __read_mostly = |
2841 | rounddown_pow_of_two(65536); | |
a9b0f861 | 2842 | |
a9b0f861 KS |
2843 | #ifdef CONFIG_DEBUG_FS |
2844 | static int fault_around_bytes_get(void *data, u64 *val) | |
1592eef0 | 2845 | { |
a9b0f861 | 2846 | *val = fault_around_bytes; |
1592eef0 KS |
2847 | return 0; |
2848 | } | |
2849 | ||
b4903d6e AR |
2850 | /* |
2851 | * fault_around_pages() and fault_around_mask() expects fault_around_bytes | |
2852 | * rounded down to nearest page order. It's what do_fault_around() expects to | |
2853 | * see. | |
2854 | */ | |
a9b0f861 | 2855 | static int fault_around_bytes_set(void *data, u64 val) |
1592eef0 | 2856 | { |
a9b0f861 | 2857 | if (val / PAGE_SIZE > PTRS_PER_PTE) |
1592eef0 | 2858 | return -EINVAL; |
b4903d6e AR |
2859 | if (val > PAGE_SIZE) |
2860 | fault_around_bytes = rounddown_pow_of_two(val); | |
2861 | else | |
2862 | fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */ | |
1592eef0 KS |
2863 | return 0; |
2864 | } | |
a9b0f861 KS |
2865 | DEFINE_SIMPLE_ATTRIBUTE(fault_around_bytes_fops, |
2866 | fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); | |
1592eef0 KS |
2867 | |
2868 | static int __init fault_around_debugfs(void) | |
2869 | { | |
2870 | void *ret; | |
2871 | ||
a9b0f861 KS |
2872 | ret = debugfs_create_file("fault_around_bytes", 0644, NULL, NULL, |
2873 | &fault_around_bytes_fops); | |
1592eef0 | 2874 | if (!ret) |
a9b0f861 | 2875 | pr_warn("Failed to create fault_around_bytes in debugfs"); |
1592eef0 KS |
2876 | return 0; |
2877 | } | |
2878 | late_initcall(fault_around_debugfs); | |
1592eef0 | 2879 | #endif |
8c6e50b0 | 2880 | |
1fdb412b KS |
2881 | /* |
2882 | * do_fault_around() tries to map few pages around the fault address. The hope | |
2883 | * is that the pages will be needed soon and this will lower the number of | |
2884 | * faults to handle. | |
2885 | * | |
2886 | * It uses vm_ops->map_pages() to map the pages, which skips the page if it's | |
2887 | * not ready to be mapped: not up-to-date, locked, etc. | |
2888 | * | |
2889 | * This function is called with the page table lock taken. In the split ptlock | |
2890 | * case the page table lock only protects only those entries which belong to | |
2891 | * the page table corresponding to the fault address. | |
2892 | * | |
2893 | * This function doesn't cross the VMA boundaries, in order to call map_pages() | |
2894 | * only once. | |
2895 | * | |
2896 | * fault_around_pages() defines how many pages we'll try to map. | |
2897 | * do_fault_around() expects it to return a power of two less than or equal to | |
2898 | * PTRS_PER_PTE. | |
2899 | * | |
2900 | * The virtual address of the area that we map is naturally aligned to the | |
2901 | * fault_around_pages() value (and therefore to page order). This way it's | |
2902 | * easier to guarantee that we don't cross page table boundaries. | |
2903 | */ | |
8c6e50b0 KS |
2904 | static void do_fault_around(struct vm_area_struct *vma, unsigned long address, |
2905 | pte_t *pte, pgoff_t pgoff, unsigned int flags) | |
2906 | { | |
aecd6f44 | 2907 | unsigned long start_addr, nr_pages, mask; |
8c6e50b0 KS |
2908 | pgoff_t max_pgoff; |
2909 | struct vm_fault vmf; | |
2910 | int off; | |
2911 | ||
4db0c3c2 | 2912 | nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT; |
aecd6f44 KS |
2913 | mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK; |
2914 | ||
2915 | start_addr = max(address & mask, vma->vm_start); | |
8c6e50b0 KS |
2916 | off = ((address - start_addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1); |
2917 | pte -= off; | |
2918 | pgoff -= off; | |
2919 | ||
2920 | /* | |
2921 | * max_pgoff is either end of page table or end of vma | |
850e9c69 | 2922 | * or fault_around_pages() from pgoff, depending what is nearest. |
8c6e50b0 KS |
2923 | */ |
2924 | max_pgoff = pgoff - ((start_addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) + | |
2925 | PTRS_PER_PTE - 1; | |
2926 | max_pgoff = min3(max_pgoff, vma_pages(vma) + vma->vm_pgoff - 1, | |
aecd6f44 | 2927 | pgoff + nr_pages - 1); |
8c6e50b0 KS |
2928 | |
2929 | /* Check if it makes any sense to call ->map_pages */ | |
2930 | while (!pte_none(*pte)) { | |
2931 | if (++pgoff > max_pgoff) | |
2932 | return; | |
2933 | start_addr += PAGE_SIZE; | |
2934 | if (start_addr >= vma->vm_end) | |
2935 | return; | |
2936 | pte++; | |
2937 | } | |
2938 | ||
2939 | vmf.virtual_address = (void __user *) start_addr; | |
2940 | vmf.pte = pte; | |
2941 | vmf.pgoff = pgoff; | |
2942 | vmf.max_pgoff = max_pgoff; | |
2943 | vmf.flags = flags; | |
c20cd45e | 2944 | vmf.gfp_mask = __get_fault_gfp_mask(vma); |
8c6e50b0 KS |
2945 | vma->vm_ops->map_pages(vma, &vmf); |
2946 | } | |
2947 | ||
e655fb29 KS |
2948 | static int do_read_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
2949 | unsigned long address, pmd_t *pmd, | |
2950 | pgoff_t pgoff, unsigned int flags, pte_t orig_pte) | |
2951 | { | |
2952 | struct page *fault_page; | |
2953 | spinlock_t *ptl; | |
3bb97794 | 2954 | pte_t *pte; |
8c6e50b0 KS |
2955 | int ret = 0; |
2956 | ||
2957 | /* | |
2958 | * Let's call ->map_pages() first and use ->fault() as fallback | |
2959 | * if page by the offset is not ready to be mapped (cold cache or | |
2960 | * something). | |
2961 | */ | |
9b4bdd2f | 2962 | if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) { |
8c6e50b0 KS |
2963 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); |
2964 | do_fault_around(vma, address, pte, pgoff, flags); | |
2965 | if (!pte_same(*pte, orig_pte)) | |
2966 | goto unlock_out; | |
2967 | pte_unmap_unlock(pte, ptl); | |
2968 | } | |
e655fb29 | 2969 | |
2e4cdab0 | 2970 | ret = __do_fault(vma, address, pgoff, flags, NULL, &fault_page); |
e655fb29 KS |
2971 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
2972 | return ret; | |
2973 | ||
2974 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); | |
2975 | if (unlikely(!pte_same(*pte, orig_pte))) { | |
2976 | pte_unmap_unlock(pte, ptl); | |
2977 | unlock_page(fault_page); | |
2978 | page_cache_release(fault_page); | |
2979 | return ret; | |
2980 | } | |
3bb97794 | 2981 | do_set_pte(vma, address, fault_page, pte, false, false); |
e655fb29 | 2982 | unlock_page(fault_page); |
8c6e50b0 KS |
2983 | unlock_out: |
2984 | pte_unmap_unlock(pte, ptl); | |
e655fb29 KS |
2985 | return ret; |
2986 | } | |
2987 | ||
ec47c3b9 KS |
2988 | static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
2989 | unsigned long address, pmd_t *pmd, | |
2990 | pgoff_t pgoff, unsigned int flags, pte_t orig_pte) | |
2991 | { | |
2992 | struct page *fault_page, *new_page; | |
00501b53 | 2993 | struct mem_cgroup *memcg; |
ec47c3b9 | 2994 | spinlock_t *ptl; |
3bb97794 | 2995 | pte_t *pte; |
ec47c3b9 KS |
2996 | int ret; |
2997 | ||
2998 | if (unlikely(anon_vma_prepare(vma))) | |
2999 | return VM_FAULT_OOM; | |
3000 | ||
3001 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); | |
3002 | if (!new_page) | |
3003 | return VM_FAULT_OOM; | |
3004 | ||
f627c2f5 | 3005 | if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg, false)) { |
ec47c3b9 KS |
3006 | page_cache_release(new_page); |
3007 | return VM_FAULT_OOM; | |
3008 | } | |
3009 | ||
2e4cdab0 | 3010 | ret = __do_fault(vma, address, pgoff, flags, new_page, &fault_page); |
ec47c3b9 KS |
3011 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3012 | goto uncharge_out; | |
3013 | ||
2e4cdab0 MW |
3014 | if (fault_page) |
3015 | copy_user_highpage(new_page, fault_page, address, vma); | |
ec47c3b9 KS |
3016 | __SetPageUptodate(new_page); |
3017 | ||
3018 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); | |
3019 | if (unlikely(!pte_same(*pte, orig_pte))) { | |
3020 | pte_unmap_unlock(pte, ptl); | |
2e4cdab0 MW |
3021 | if (fault_page) { |
3022 | unlock_page(fault_page); | |
3023 | page_cache_release(fault_page); | |
3024 | } else { | |
3025 | /* | |
3026 | * The fault handler has no page to lock, so it holds | |
0df9d41a | 3027 | * i_mmap_lock for read to protect against truncate. |
2e4cdab0 | 3028 | */ |
0df9d41a | 3029 | i_mmap_unlock_read(vma->vm_file->f_mapping); |
2e4cdab0 | 3030 | } |
ec47c3b9 KS |
3031 | goto uncharge_out; |
3032 | } | |
3bb97794 | 3033 | do_set_pte(vma, address, new_page, pte, true, true); |
f627c2f5 | 3034 | mem_cgroup_commit_charge(new_page, memcg, false, false); |
00501b53 | 3035 | lru_cache_add_active_or_unevictable(new_page, vma); |
ec47c3b9 | 3036 | pte_unmap_unlock(pte, ptl); |
2e4cdab0 MW |
3037 | if (fault_page) { |
3038 | unlock_page(fault_page); | |
3039 | page_cache_release(fault_page); | |
3040 | } else { | |
3041 | /* | |
3042 | * The fault handler has no page to lock, so it holds | |
0df9d41a | 3043 | * i_mmap_lock for read to protect against truncate. |
2e4cdab0 | 3044 | */ |
0df9d41a | 3045 | i_mmap_unlock_read(vma->vm_file->f_mapping); |
2e4cdab0 | 3046 | } |
ec47c3b9 KS |
3047 | return ret; |
3048 | uncharge_out: | |
f627c2f5 | 3049 | mem_cgroup_cancel_charge(new_page, memcg, false); |
ec47c3b9 KS |
3050 | page_cache_release(new_page); |
3051 | return ret; | |
3052 | } | |
3053 | ||
f0c6d4d2 | 3054 | static int do_shared_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
16abfa08 | 3055 | unsigned long address, pmd_t *pmd, |
54cb8821 | 3056 | pgoff_t pgoff, unsigned int flags, pte_t orig_pte) |
1da177e4 | 3057 | { |
f0c6d4d2 KS |
3058 | struct page *fault_page; |
3059 | struct address_space *mapping; | |
8f4e2101 | 3060 | spinlock_t *ptl; |
3bb97794 | 3061 | pte_t *pte; |
f0c6d4d2 | 3062 | int dirtied = 0; |
f0c6d4d2 | 3063 | int ret, tmp; |
1d65f86d | 3064 | |
2e4cdab0 | 3065 | ret = __do_fault(vma, address, pgoff, flags, NULL, &fault_page); |
7eae74af | 3066 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
f0c6d4d2 | 3067 | return ret; |
1da177e4 LT |
3068 | |
3069 | /* | |
f0c6d4d2 KS |
3070 | * Check if the backing address space wants to know that the page is |
3071 | * about to become writable | |
1da177e4 | 3072 | */ |
fb09a464 KS |
3073 | if (vma->vm_ops->page_mkwrite) { |
3074 | unlock_page(fault_page); | |
3075 | tmp = do_page_mkwrite(vma, fault_page, address); | |
3076 | if (unlikely(!tmp || | |
3077 | (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
f0c6d4d2 | 3078 | page_cache_release(fault_page); |
fb09a464 | 3079 | return tmp; |
4294621f | 3080 | } |
fb09a464 KS |
3081 | } |
3082 | ||
f0c6d4d2 KS |
3083 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); |
3084 | if (unlikely(!pte_same(*pte, orig_pte))) { | |
3085 | pte_unmap_unlock(pte, ptl); | |
3086 | unlock_page(fault_page); | |
3087 | page_cache_release(fault_page); | |
3088 | return ret; | |
1da177e4 | 3089 | } |
3bb97794 | 3090 | do_set_pte(vma, address, fault_page, pte, true, false); |
f0c6d4d2 | 3091 | pte_unmap_unlock(pte, ptl); |
b827e496 | 3092 | |
f0c6d4d2 KS |
3093 | if (set_page_dirty(fault_page)) |
3094 | dirtied = 1; | |
d82fa87d AM |
3095 | /* |
3096 | * Take a local copy of the address_space - page.mapping may be zeroed | |
3097 | * by truncate after unlock_page(). The address_space itself remains | |
3098 | * pinned by vma->vm_file's reference. We rely on unlock_page()'s | |
3099 | * release semantics to prevent the compiler from undoing this copying. | |
3100 | */ | |
1c290f64 | 3101 | mapping = page_rmapping(fault_page); |
f0c6d4d2 KS |
3102 | unlock_page(fault_page); |
3103 | if ((dirtied || vma->vm_ops->page_mkwrite) && mapping) { | |
3104 | /* | |
3105 | * Some device drivers do not set page.mapping but still | |
3106 | * dirty their pages | |
3107 | */ | |
3108 | balance_dirty_pages_ratelimited(mapping); | |
d08b3851 | 3109 | } |
d00806b1 | 3110 | |
74ec6751 | 3111 | if (!vma->vm_ops->page_mkwrite) |
f0c6d4d2 | 3112 | file_update_time(vma->vm_file); |
b827e496 | 3113 | |
1d65f86d | 3114 | return ret; |
54cb8821 | 3115 | } |
d00806b1 | 3116 | |
9a95f3cf PC |
3117 | /* |
3118 | * We enter with non-exclusive mmap_sem (to exclude vma changes, | |
3119 | * but allow concurrent faults). | |
3120 | * The mmap_sem may have been released depending on flags and our | |
3121 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
3122 | */ | |
9b4bdd2f | 3123 | static int do_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
54cb8821 | 3124 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
30c9f3a9 | 3125 | unsigned int flags, pte_t orig_pte) |
54cb8821 | 3126 | { |
88193f7c | 3127 | pgoff_t pgoff = linear_page_index(vma, address); |
54cb8821 | 3128 | |
16abfa08 | 3129 | pte_unmap(page_table); |
6b7339f4 KS |
3130 | /* The VMA was not fully populated on mmap() or missing VM_DONTEXPAND */ |
3131 | if (!vma->vm_ops->fault) | |
3132 | return VM_FAULT_SIGBUS; | |
e655fb29 KS |
3133 | if (!(flags & FAULT_FLAG_WRITE)) |
3134 | return do_read_fault(mm, vma, address, pmd, pgoff, flags, | |
3135 | orig_pte); | |
ec47c3b9 KS |
3136 | if (!(vma->vm_flags & VM_SHARED)) |
3137 | return do_cow_fault(mm, vma, address, pmd, pgoff, flags, | |
3138 | orig_pte); | |
f0c6d4d2 | 3139 | return do_shared_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); |
54cb8821 NP |
3140 | } |
3141 | ||
b19a9939 | 3142 | static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, |
04bb2f94 RR |
3143 | unsigned long addr, int page_nid, |
3144 | int *flags) | |
9532fec1 MG |
3145 | { |
3146 | get_page(page); | |
3147 | ||
3148 | count_vm_numa_event(NUMA_HINT_FAULTS); | |
04bb2f94 | 3149 | if (page_nid == numa_node_id()) { |
9532fec1 | 3150 | count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); |
04bb2f94 RR |
3151 | *flags |= TNF_FAULT_LOCAL; |
3152 | } | |
9532fec1 MG |
3153 | |
3154 | return mpol_misplaced(page, vma, addr); | |
3155 | } | |
3156 | ||
b19a9939 | 3157 | static int do_numa_page(struct mm_struct *mm, struct vm_area_struct *vma, |
d10e63f2 MG |
3158 | unsigned long addr, pte_t pte, pte_t *ptep, pmd_t *pmd) |
3159 | { | |
4daae3b4 | 3160 | struct page *page = NULL; |
d10e63f2 | 3161 | spinlock_t *ptl; |
8191acbd | 3162 | int page_nid = -1; |
90572890 | 3163 | int last_cpupid; |
cbee9f88 | 3164 | int target_nid; |
b8593bfd | 3165 | bool migrated = false; |
b191f9b1 | 3166 | bool was_writable = pte_write(pte); |
6688cc05 | 3167 | int flags = 0; |
d10e63f2 | 3168 | |
c0e7cad9 MG |
3169 | /* A PROT_NONE fault should not end up here */ |
3170 | BUG_ON(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))); | |
3171 | ||
d10e63f2 MG |
3172 | /* |
3173 | * The "pte" at this point cannot be used safely without | |
3174 | * validation through pte_unmap_same(). It's of NUMA type but | |
3175 | * the pfn may be screwed if the read is non atomic. | |
3176 | * | |
4d942466 MG |
3177 | * We can safely just do a "set_pte_at()", because the old |
3178 | * page table entry is not accessible, so there would be no | |
3179 | * concurrent hardware modifications to the PTE. | |
d10e63f2 MG |
3180 | */ |
3181 | ptl = pte_lockptr(mm, pmd); | |
3182 | spin_lock(ptl); | |
4daae3b4 MG |
3183 | if (unlikely(!pte_same(*ptep, pte))) { |
3184 | pte_unmap_unlock(ptep, ptl); | |
3185 | goto out; | |
3186 | } | |
3187 | ||
4d942466 MG |
3188 | /* Make it present again */ |
3189 | pte = pte_modify(pte, vma->vm_page_prot); | |
3190 | pte = pte_mkyoung(pte); | |
b191f9b1 MG |
3191 | if (was_writable) |
3192 | pte = pte_mkwrite(pte); | |
d10e63f2 MG |
3193 | set_pte_at(mm, addr, ptep, pte); |
3194 | update_mmu_cache(vma, addr, ptep); | |
3195 | ||
3196 | page = vm_normal_page(vma, addr, pte); | |
3197 | if (!page) { | |
3198 | pte_unmap_unlock(ptep, ptl); | |
3199 | return 0; | |
3200 | } | |
3201 | ||
e81c4802 KS |
3202 | /* TODO: handle PTE-mapped THP */ |
3203 | if (PageCompound(page)) { | |
3204 | pte_unmap_unlock(ptep, ptl); | |
3205 | return 0; | |
3206 | } | |
3207 | ||
6688cc05 | 3208 | /* |
bea66fbd MG |
3209 | * Avoid grouping on RO pages in general. RO pages shouldn't hurt as |
3210 | * much anyway since they can be in shared cache state. This misses | |
3211 | * the case where a mapping is writable but the process never writes | |
3212 | * to it but pte_write gets cleared during protection updates and | |
3213 | * pte_dirty has unpredictable behaviour between PTE scan updates, | |
3214 | * background writeback, dirty balancing and application behaviour. | |
6688cc05 | 3215 | */ |
bea66fbd | 3216 | if (!(vma->vm_flags & VM_WRITE)) |
6688cc05 PZ |
3217 | flags |= TNF_NO_GROUP; |
3218 | ||
dabe1d99 RR |
3219 | /* |
3220 | * Flag if the page is shared between multiple address spaces. This | |
3221 | * is later used when determining whether to group tasks together | |
3222 | */ | |
3223 | if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) | |
3224 | flags |= TNF_SHARED; | |
3225 | ||
90572890 | 3226 | last_cpupid = page_cpupid_last(page); |
8191acbd | 3227 | page_nid = page_to_nid(page); |
04bb2f94 | 3228 | target_nid = numa_migrate_prep(page, vma, addr, page_nid, &flags); |
d10e63f2 | 3229 | pte_unmap_unlock(ptep, ptl); |
4daae3b4 | 3230 | if (target_nid == -1) { |
4daae3b4 MG |
3231 | put_page(page); |
3232 | goto out; | |
3233 | } | |
3234 | ||
3235 | /* Migrate to the requested node */ | |
1bc115d8 | 3236 | migrated = migrate_misplaced_page(page, vma, target_nid); |
6688cc05 | 3237 | if (migrated) { |
8191acbd | 3238 | page_nid = target_nid; |
6688cc05 | 3239 | flags |= TNF_MIGRATED; |
074c2381 MG |
3240 | } else |
3241 | flags |= TNF_MIGRATE_FAIL; | |
4daae3b4 MG |
3242 | |
3243 | out: | |
8191acbd | 3244 | if (page_nid != -1) |
6688cc05 | 3245 | task_numa_fault(last_cpupid, page_nid, 1, flags); |
d10e63f2 MG |
3246 | return 0; |
3247 | } | |
3248 | ||
b96375f7 MW |
3249 | static int create_huge_pmd(struct mm_struct *mm, struct vm_area_struct *vma, |
3250 | unsigned long address, pmd_t *pmd, unsigned int flags) | |
3251 | { | |
fb6dd5fa | 3252 | if (vma_is_anonymous(vma)) |
b96375f7 MW |
3253 | return do_huge_pmd_anonymous_page(mm, vma, address, pmd, flags); |
3254 | if (vma->vm_ops->pmd_fault) | |
3255 | return vma->vm_ops->pmd_fault(vma, address, pmd, flags); | |
3256 | return VM_FAULT_FALLBACK; | |
3257 | } | |
3258 | ||
3259 | static int wp_huge_pmd(struct mm_struct *mm, struct vm_area_struct *vma, | |
3260 | unsigned long address, pmd_t *pmd, pmd_t orig_pmd, | |
3261 | unsigned int flags) | |
3262 | { | |
fb6dd5fa | 3263 | if (vma_is_anonymous(vma)) |
b96375f7 MW |
3264 | return do_huge_pmd_wp_page(mm, vma, address, pmd, orig_pmd); |
3265 | if (vma->vm_ops->pmd_fault) | |
3266 | return vma->vm_ops->pmd_fault(vma, address, pmd, flags); | |
3267 | return VM_FAULT_FALLBACK; | |
3268 | } | |
3269 | ||
1da177e4 LT |
3270 | /* |
3271 | * These routines also need to handle stuff like marking pages dirty | |
3272 | * and/or accessed for architectures that don't do it in hardware (most | |
3273 | * RISC architectures). The early dirtying is also good on the i386. | |
3274 | * | |
3275 | * There is also a hook called "update_mmu_cache()" that architectures | |
3276 | * with external mmu caches can use to update those (ie the Sparc or | |
3277 | * PowerPC hashed page tables that act as extended TLBs). | |
3278 | * | |
c74df32c HD |
3279 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
3280 | * but allow concurrent faults), and pte mapped but not yet locked. | |
9a95f3cf PC |
3281 | * We return with pte unmapped and unlocked. |
3282 | * | |
3283 | * The mmap_sem may have been released depending on flags and our | |
3284 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 3285 | */ |
c0292554 | 3286 | static int handle_pte_fault(struct mm_struct *mm, |
71e3aac0 AA |
3287 | struct vm_area_struct *vma, unsigned long address, |
3288 | pte_t *pte, pmd_t *pmd, unsigned int flags) | |
1da177e4 LT |
3289 | { |
3290 | pte_t entry; | |
8f4e2101 | 3291 | spinlock_t *ptl; |
1da177e4 | 3292 | |
e37c6982 CB |
3293 | /* |
3294 | * some architectures can have larger ptes than wordsize, | |
3295 | * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and CONFIG_32BIT=y, | |
3296 | * so READ_ONCE or ACCESS_ONCE cannot guarantee atomic accesses. | |
3297 | * The code below just needs a consistent view for the ifs and | |
3298 | * we later double check anyway with the ptl lock held. So here | |
3299 | * a barrier will do. | |
3300 | */ | |
3301 | entry = *pte; | |
3302 | barrier(); | |
1da177e4 | 3303 | if (!pte_present(entry)) { |
65500d23 | 3304 | if (pte_none(entry)) { |
b5330628 ON |
3305 | if (vma_is_anonymous(vma)) |
3306 | return do_anonymous_page(mm, vma, address, | |
3307 | pte, pmd, flags); | |
3308 | else | |
6b7339f4 KS |
3309 | return do_fault(mm, vma, address, pte, pmd, |
3310 | flags, entry); | |
65500d23 | 3311 | } |
65500d23 | 3312 | return do_swap_page(mm, vma, address, |
30c9f3a9 | 3313 | pte, pmd, flags, entry); |
1da177e4 LT |
3314 | } |
3315 | ||
8a0516ed | 3316 | if (pte_protnone(entry)) |
d10e63f2 MG |
3317 | return do_numa_page(mm, vma, address, entry, pte, pmd); |
3318 | ||
4c21e2f2 | 3319 | ptl = pte_lockptr(mm, pmd); |
8f4e2101 HD |
3320 | spin_lock(ptl); |
3321 | if (unlikely(!pte_same(*pte, entry))) | |
3322 | goto unlock; | |
30c9f3a9 | 3323 | if (flags & FAULT_FLAG_WRITE) { |
1da177e4 | 3324 | if (!pte_write(entry)) |
8f4e2101 HD |
3325 | return do_wp_page(mm, vma, address, |
3326 | pte, pmd, ptl, entry); | |
1da177e4 LT |
3327 | entry = pte_mkdirty(entry); |
3328 | } | |
3329 | entry = pte_mkyoung(entry); | |
30c9f3a9 | 3330 | if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) { |
4b3073e1 | 3331 | update_mmu_cache(vma, address, pte); |
1a44e149 AA |
3332 | } else { |
3333 | /* | |
3334 | * This is needed only for protection faults but the arch code | |
3335 | * is not yet telling us if this is a protection fault or not. | |
3336 | * This still avoids useless tlb flushes for .text page faults | |
3337 | * with threads. | |
3338 | */ | |
30c9f3a9 | 3339 | if (flags & FAULT_FLAG_WRITE) |
61c77326 | 3340 | flush_tlb_fix_spurious_fault(vma, address); |
1a44e149 | 3341 | } |
8f4e2101 HD |
3342 | unlock: |
3343 | pte_unmap_unlock(pte, ptl); | |
83c54070 | 3344 | return 0; |
1da177e4 LT |
3345 | } |
3346 | ||
3347 | /* | |
3348 | * By the time we get here, we already hold the mm semaphore | |
9a95f3cf PC |
3349 | * |
3350 | * The mmap_sem may have been released depending on flags and our | |
3351 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 3352 | */ |
519e5247 JW |
3353 | static int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
3354 | unsigned long address, unsigned int flags) | |
1da177e4 LT |
3355 | { |
3356 | pgd_t *pgd; | |
3357 | pud_t *pud; | |
3358 | pmd_t *pmd; | |
3359 | pte_t *pte; | |
3360 | ||
ac9b9c66 | 3361 | if (unlikely(is_vm_hugetlb_page(vma))) |
30c9f3a9 | 3362 | return hugetlb_fault(mm, vma, address, flags); |
1da177e4 | 3363 | |
1da177e4 | 3364 | pgd = pgd_offset(mm, address); |
1da177e4 LT |
3365 | pud = pud_alloc(mm, pgd, address); |
3366 | if (!pud) | |
c74df32c | 3367 | return VM_FAULT_OOM; |
1da177e4 LT |
3368 | pmd = pmd_alloc(mm, pud, address); |
3369 | if (!pmd) | |
c74df32c | 3370 | return VM_FAULT_OOM; |
71e3aac0 | 3371 | if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) { |
b96375f7 | 3372 | int ret = create_huge_pmd(mm, vma, address, pmd, flags); |
c0292554 KS |
3373 | if (!(ret & VM_FAULT_FALLBACK)) |
3374 | return ret; | |
71e3aac0 AA |
3375 | } else { |
3376 | pmd_t orig_pmd = *pmd; | |
1f1d06c3 DR |
3377 | int ret; |
3378 | ||
71e3aac0 | 3379 | barrier(); |
5c7fb56e | 3380 | if (pmd_trans_huge(orig_pmd) || pmd_devmap(orig_pmd)) { |
a1dd450b WD |
3381 | unsigned int dirty = flags & FAULT_FLAG_WRITE; |
3382 | ||
8a0516ed | 3383 | if (pmd_protnone(orig_pmd)) |
4daae3b4 | 3384 | return do_huge_pmd_numa_page(mm, vma, address, |
d10e63f2 MG |
3385 | orig_pmd, pmd); |
3386 | ||
3d59eebc | 3387 | if (dirty && !pmd_write(orig_pmd)) { |
b96375f7 MW |
3388 | ret = wp_huge_pmd(mm, vma, address, pmd, |
3389 | orig_pmd, flags); | |
9845cbbd KS |
3390 | if (!(ret & VM_FAULT_FALLBACK)) |
3391 | return ret; | |
a1dd450b WD |
3392 | } else { |
3393 | huge_pmd_set_accessed(mm, vma, address, pmd, | |
3394 | orig_pmd, dirty); | |
9845cbbd | 3395 | return 0; |
1f1d06c3 | 3396 | } |
71e3aac0 AA |
3397 | } |
3398 | } | |
3399 | ||
3400 | /* | |
3401 | * Use __pte_alloc instead of pte_alloc_map, because we can't | |
3402 | * run pte_offset_map on the pmd, if an huge pmd could | |
3403 | * materialize from under us from a different thread. | |
3404 | */ | |
4fd01770 MG |
3405 | if (unlikely(pmd_none(*pmd)) && |
3406 | unlikely(__pte_alloc(mm, vma, pmd, address))) | |
c74df32c | 3407 | return VM_FAULT_OOM; |
ad33bb04 AA |
3408 | /* |
3409 | * If a huge pmd materialized under us just retry later. Use | |
3410 | * pmd_trans_unstable() instead of pmd_trans_huge() to ensure the pmd | |
3411 | * didn't become pmd_trans_huge under us and then back to pmd_none, as | |
3412 | * a result of MADV_DONTNEED running immediately after a huge pmd fault | |
3413 | * in a different thread of this mm, in turn leading to a misleading | |
3414 | * pmd_trans_huge() retval. All we have to ensure is that it is a | |
3415 | * regular pmd that we can walk with pte_offset_map() and we can do that | |
3416 | * through an atomic read in C, which is what pmd_trans_unstable() | |
3417 | * provides. | |
3418 | */ | |
3419 | if (unlikely(pmd_trans_unstable(pmd) || pmd_devmap(*pmd))) | |
71e3aac0 AA |
3420 | return 0; |
3421 | /* | |
3422 | * A regular pmd is established and it can't morph into a huge pmd | |
3423 | * from under us anymore at this point because we hold the mmap_sem | |
3424 | * read mode and khugepaged takes it in write mode. So now it's | |
3425 | * safe to run pte_offset_map(). | |
3426 | */ | |
3427 | pte = pte_offset_map(pmd, address); | |
1da177e4 | 3428 | |
30c9f3a9 | 3429 | return handle_pte_fault(mm, vma, address, pte, pmd, flags); |
1da177e4 LT |
3430 | } |
3431 | ||
9a95f3cf PC |
3432 | /* |
3433 | * By the time we get here, we already hold the mm semaphore | |
3434 | * | |
3435 | * The mmap_sem may have been released depending on flags and our | |
3436 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
3437 | */ | |
519e5247 JW |
3438 | int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
3439 | unsigned long address, unsigned int flags) | |
3440 | { | |
3441 | int ret; | |
3442 | ||
3443 | __set_current_state(TASK_RUNNING); | |
3444 | ||
3445 | count_vm_event(PGFAULT); | |
3446 | mem_cgroup_count_vm_event(mm, PGFAULT); | |
3447 | ||
3448 | /* do counter updates before entering really critical section. */ | |
3449 | check_sync_rss_stat(current); | |
3450 | ||
3451 | /* | |
3452 | * Enable the memcg OOM handling for faults triggered in user | |
3453 | * space. Kernel faults are handled more gracefully. | |
3454 | */ | |
3455 | if (flags & FAULT_FLAG_USER) | |
49426420 | 3456 | mem_cgroup_oom_enable(); |
519e5247 JW |
3457 | |
3458 | ret = __handle_mm_fault(mm, vma, address, flags); | |
3459 | ||
49426420 JW |
3460 | if (flags & FAULT_FLAG_USER) { |
3461 | mem_cgroup_oom_disable(); | |
3462 | /* | |
3463 | * The task may have entered a memcg OOM situation but | |
3464 | * if the allocation error was handled gracefully (no | |
3465 | * VM_FAULT_OOM), there is no need to kill anything. | |
3466 | * Just clean up the OOM state peacefully. | |
3467 | */ | |
3468 | if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) | |
3469 | mem_cgroup_oom_synchronize(false); | |
3470 | } | |
3812c8c8 | 3471 | |
519e5247 JW |
3472 | return ret; |
3473 | } | |
e1d6d01a | 3474 | EXPORT_SYMBOL_GPL(handle_mm_fault); |
519e5247 | 3475 | |
1da177e4 LT |
3476 | #ifndef __PAGETABLE_PUD_FOLDED |
3477 | /* | |
3478 | * Allocate page upper directory. | |
872fec16 | 3479 | * We've already handled the fast-path in-line. |
1da177e4 | 3480 | */ |
1bb3630e | 3481 | int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) |
1da177e4 | 3482 | { |
c74df32c HD |
3483 | pud_t *new = pud_alloc_one(mm, address); |
3484 | if (!new) | |
1bb3630e | 3485 | return -ENOMEM; |
1da177e4 | 3486 | |
362a61ad NP |
3487 | smp_wmb(); /* See comment in __pte_alloc */ |
3488 | ||
872fec16 | 3489 | spin_lock(&mm->page_table_lock); |
1bb3630e | 3490 | if (pgd_present(*pgd)) /* Another has populated it */ |
5e541973 | 3491 | pud_free(mm, new); |
1bb3630e HD |
3492 | else |
3493 | pgd_populate(mm, pgd, new); | |
c74df32c | 3494 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 3495 | return 0; |
1da177e4 LT |
3496 | } |
3497 | #endif /* __PAGETABLE_PUD_FOLDED */ | |
3498 | ||
3499 | #ifndef __PAGETABLE_PMD_FOLDED | |
3500 | /* | |
3501 | * Allocate page middle directory. | |
872fec16 | 3502 | * We've already handled the fast-path in-line. |
1da177e4 | 3503 | */ |
1bb3630e | 3504 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
1da177e4 | 3505 | { |
c74df32c HD |
3506 | pmd_t *new = pmd_alloc_one(mm, address); |
3507 | if (!new) | |
1bb3630e | 3508 | return -ENOMEM; |
1da177e4 | 3509 | |
362a61ad NP |
3510 | smp_wmb(); /* See comment in __pte_alloc */ |
3511 | ||
872fec16 | 3512 | spin_lock(&mm->page_table_lock); |
1da177e4 | 3513 | #ifndef __ARCH_HAS_4LEVEL_HACK |
dc6c9a35 KS |
3514 | if (!pud_present(*pud)) { |
3515 | mm_inc_nr_pmds(mm); | |
1bb3630e | 3516 | pud_populate(mm, pud, new); |
dc6c9a35 | 3517 | } else /* Another has populated it */ |
5e541973 | 3518 | pmd_free(mm, new); |
dc6c9a35 KS |
3519 | #else |
3520 | if (!pgd_present(*pud)) { | |
3521 | mm_inc_nr_pmds(mm); | |
1bb3630e | 3522 | pgd_populate(mm, pud, new); |
dc6c9a35 KS |
3523 | } else /* Another has populated it */ |
3524 | pmd_free(mm, new); | |
1da177e4 | 3525 | #endif /* __ARCH_HAS_4LEVEL_HACK */ |
c74df32c | 3526 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 3527 | return 0; |
e0f39591 | 3528 | } |
1da177e4 LT |
3529 | #endif /* __PAGETABLE_PMD_FOLDED */ |
3530 | ||
1b36ba81 | 3531 | static int __follow_pte(struct mm_struct *mm, unsigned long address, |
f8ad0f49 JW |
3532 | pte_t **ptepp, spinlock_t **ptlp) |
3533 | { | |
3534 | pgd_t *pgd; | |
3535 | pud_t *pud; | |
3536 | pmd_t *pmd; | |
3537 | pte_t *ptep; | |
3538 | ||
3539 | pgd = pgd_offset(mm, address); | |
3540 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
3541 | goto out; | |
3542 | ||
3543 | pud = pud_offset(pgd, address); | |
3544 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) | |
3545 | goto out; | |
3546 | ||
3547 | pmd = pmd_offset(pud, address); | |
f66055ab | 3548 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
f8ad0f49 JW |
3549 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) |
3550 | goto out; | |
3551 | ||
3552 | /* We cannot handle huge page PFN maps. Luckily they don't exist. */ | |
3553 | if (pmd_huge(*pmd)) | |
3554 | goto out; | |
3555 | ||
3556 | ptep = pte_offset_map_lock(mm, pmd, address, ptlp); | |
3557 | if (!ptep) | |
3558 | goto out; | |
3559 | if (!pte_present(*ptep)) | |
3560 | goto unlock; | |
3561 | *ptepp = ptep; | |
3562 | return 0; | |
3563 | unlock: | |
3564 | pte_unmap_unlock(ptep, *ptlp); | |
3565 | out: | |
3566 | return -EINVAL; | |
3567 | } | |
3568 | ||
1b36ba81 NK |
3569 | static inline int follow_pte(struct mm_struct *mm, unsigned long address, |
3570 | pte_t **ptepp, spinlock_t **ptlp) | |
3571 | { | |
3572 | int res; | |
3573 | ||
3574 | /* (void) is needed to make gcc happy */ | |
3575 | (void) __cond_lock(*ptlp, | |
3576 | !(res = __follow_pte(mm, address, ptepp, ptlp))); | |
3577 | return res; | |
3578 | } | |
3579 | ||
3b6748e2 JW |
3580 | /** |
3581 | * follow_pfn - look up PFN at a user virtual address | |
3582 | * @vma: memory mapping | |
3583 | * @address: user virtual address | |
3584 | * @pfn: location to store found PFN | |
3585 | * | |
3586 | * Only IO mappings and raw PFN mappings are allowed. | |
3587 | * | |
3588 | * Returns zero and the pfn at @pfn on success, -ve otherwise. | |
3589 | */ | |
3590 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
3591 | unsigned long *pfn) | |
3592 | { | |
3593 | int ret = -EINVAL; | |
3594 | spinlock_t *ptl; | |
3595 | pte_t *ptep; | |
3596 | ||
3597 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
3598 | return ret; | |
3599 | ||
3600 | ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); | |
3601 | if (ret) | |
3602 | return ret; | |
3603 | *pfn = pte_pfn(*ptep); | |
3604 | pte_unmap_unlock(ptep, ptl); | |
3605 | return 0; | |
3606 | } | |
3607 | EXPORT_SYMBOL(follow_pfn); | |
3608 | ||
28b2ee20 | 3609 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
d87fe660 | 3610 | int follow_phys(struct vm_area_struct *vma, |
3611 | unsigned long address, unsigned int flags, | |
3612 | unsigned long *prot, resource_size_t *phys) | |
28b2ee20 | 3613 | { |
03668a4d | 3614 | int ret = -EINVAL; |
28b2ee20 RR |
3615 | pte_t *ptep, pte; |
3616 | spinlock_t *ptl; | |
28b2ee20 | 3617 | |
d87fe660 | 3618 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
3619 | goto out; | |
28b2ee20 | 3620 | |
03668a4d | 3621 | if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) |
d87fe660 | 3622 | goto out; |
28b2ee20 | 3623 | pte = *ptep; |
03668a4d | 3624 | |
28b2ee20 RR |
3625 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
3626 | goto unlock; | |
28b2ee20 RR |
3627 | |
3628 | *prot = pgprot_val(pte_pgprot(pte)); | |
03668a4d | 3629 | *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; |
28b2ee20 | 3630 | |
03668a4d | 3631 | ret = 0; |
28b2ee20 RR |
3632 | unlock: |
3633 | pte_unmap_unlock(ptep, ptl); | |
3634 | out: | |
d87fe660 | 3635 | return ret; |
28b2ee20 RR |
3636 | } |
3637 | ||
3638 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, | |
3639 | void *buf, int len, int write) | |
3640 | { | |
3641 | resource_size_t phys_addr; | |
3642 | unsigned long prot = 0; | |
2bc7273b | 3643 | void __iomem *maddr; |
28b2ee20 RR |
3644 | int offset = addr & (PAGE_SIZE-1); |
3645 | ||
d87fe660 | 3646 | if (follow_phys(vma, addr, write, &prot, &phys_addr)) |
28b2ee20 RR |
3647 | return -EINVAL; |
3648 | ||
9cb12d7b | 3649 | maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); |
28b2ee20 RR |
3650 | if (write) |
3651 | memcpy_toio(maddr + offset, buf, len); | |
3652 | else | |
3653 | memcpy_fromio(buf, maddr + offset, len); | |
3654 | iounmap(maddr); | |
3655 | ||
3656 | return len; | |
3657 | } | |
5a73633e | 3658 | EXPORT_SYMBOL_GPL(generic_access_phys); |
28b2ee20 RR |
3659 | #endif |
3660 | ||
0ec76a11 | 3661 | /* |
206cb636 SW |
3662 | * Access another process' address space as given in mm. If non-NULL, use the |
3663 | * given task for page fault accounting. | |
0ec76a11 | 3664 | */ |
206cb636 SW |
3665 | static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, |
3666 | unsigned long addr, void *buf, int len, int write) | |
0ec76a11 | 3667 | { |
0ec76a11 | 3668 | struct vm_area_struct *vma; |
0ec76a11 DH |
3669 | void *old_buf = buf; |
3670 | ||
0ec76a11 | 3671 | down_read(&mm->mmap_sem); |
183ff22b | 3672 | /* ignore errors, just check how much was successfully transferred */ |
0ec76a11 DH |
3673 | while (len) { |
3674 | int bytes, ret, offset; | |
3675 | void *maddr; | |
28b2ee20 | 3676 | struct page *page = NULL; |
0ec76a11 DH |
3677 | |
3678 | ret = get_user_pages(tsk, mm, addr, 1, | |
3679 | write, 1, &page, &vma); | |
28b2ee20 | 3680 | if (ret <= 0) { |
dbffcd03 RR |
3681 | #ifndef CONFIG_HAVE_IOREMAP_PROT |
3682 | break; | |
3683 | #else | |
28b2ee20 RR |
3684 | /* |
3685 | * Check if this is a VM_IO | VM_PFNMAP VMA, which | |
3686 | * we can access using slightly different code. | |
3687 | */ | |
28b2ee20 | 3688 | vma = find_vma(mm, addr); |
fe936dfc | 3689 | if (!vma || vma->vm_start > addr) |
28b2ee20 RR |
3690 | break; |
3691 | if (vma->vm_ops && vma->vm_ops->access) | |
3692 | ret = vma->vm_ops->access(vma, addr, buf, | |
3693 | len, write); | |
3694 | if (ret <= 0) | |
28b2ee20 RR |
3695 | break; |
3696 | bytes = ret; | |
dbffcd03 | 3697 | #endif |
0ec76a11 | 3698 | } else { |
28b2ee20 RR |
3699 | bytes = len; |
3700 | offset = addr & (PAGE_SIZE-1); | |
3701 | if (bytes > PAGE_SIZE-offset) | |
3702 | bytes = PAGE_SIZE-offset; | |
3703 | ||
3704 | maddr = kmap(page); | |
3705 | if (write) { | |
3706 | copy_to_user_page(vma, page, addr, | |
3707 | maddr + offset, buf, bytes); | |
3708 | set_page_dirty_lock(page); | |
3709 | } else { | |
3710 | copy_from_user_page(vma, page, addr, | |
3711 | buf, maddr + offset, bytes); | |
3712 | } | |
3713 | kunmap(page); | |
3714 | page_cache_release(page); | |
0ec76a11 | 3715 | } |
0ec76a11 DH |
3716 | len -= bytes; |
3717 | buf += bytes; | |
3718 | addr += bytes; | |
3719 | } | |
3720 | up_read(&mm->mmap_sem); | |
0ec76a11 DH |
3721 | |
3722 | return buf - old_buf; | |
3723 | } | |
03252919 | 3724 | |
5ddd36b9 | 3725 | /** |
ae91dbfc | 3726 | * access_remote_vm - access another process' address space |
5ddd36b9 SW |
3727 | * @mm: the mm_struct of the target address space |
3728 | * @addr: start address to access | |
3729 | * @buf: source or destination buffer | |
3730 | * @len: number of bytes to transfer | |
3731 | * @write: whether the access is a write | |
3732 | * | |
3733 | * The caller must hold a reference on @mm. | |
3734 | */ | |
3735 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, | |
3736 | void *buf, int len, int write) | |
3737 | { | |
3738 | return __access_remote_vm(NULL, mm, addr, buf, len, write); | |
3739 | } | |
3740 | ||
206cb636 SW |
3741 | /* |
3742 | * Access another process' address space. | |
3743 | * Source/target buffer must be kernel space, | |
3744 | * Do not walk the page table directly, use get_user_pages | |
3745 | */ | |
3746 | int access_process_vm(struct task_struct *tsk, unsigned long addr, | |
3747 | void *buf, int len, int write) | |
3748 | { | |
3749 | struct mm_struct *mm; | |
3750 | int ret; | |
3751 | ||
3752 | mm = get_task_mm(tsk); | |
3753 | if (!mm) | |
3754 | return 0; | |
3755 | ||
3756 | ret = __access_remote_vm(tsk, mm, addr, buf, len, write); | |
3757 | mmput(mm); | |
3758 | ||
3759 | return ret; | |
3760 | } | |
3761 | ||
03252919 AK |
3762 | /* |
3763 | * Print the name of a VMA. | |
3764 | */ | |
3765 | void print_vma_addr(char *prefix, unsigned long ip) | |
3766 | { | |
3767 | struct mm_struct *mm = current->mm; | |
3768 | struct vm_area_struct *vma; | |
3769 | ||
e8bff74a IM |
3770 | /* |
3771 | * Do not print if we are in atomic | |
3772 | * contexts (in exception stacks, etc.): | |
3773 | */ | |
3774 | if (preempt_count()) | |
3775 | return; | |
3776 | ||
03252919 AK |
3777 | down_read(&mm->mmap_sem); |
3778 | vma = find_vma(mm, ip); | |
3779 | if (vma && vma->vm_file) { | |
3780 | struct file *f = vma->vm_file; | |
3781 | char *buf = (char *)__get_free_page(GFP_KERNEL); | |
3782 | if (buf) { | |
2fbc57c5 | 3783 | char *p; |
03252919 | 3784 | |
9bf39ab2 | 3785 | p = file_path(f, buf, PAGE_SIZE); |
03252919 AK |
3786 | if (IS_ERR(p)) |
3787 | p = "?"; | |
2fbc57c5 | 3788 | printk("%s%s[%lx+%lx]", prefix, kbasename(p), |
03252919 AK |
3789 | vma->vm_start, |
3790 | vma->vm_end - vma->vm_start); | |
3791 | free_page((unsigned long)buf); | |
3792 | } | |
3793 | } | |
51a07e50 | 3794 | up_read(&mm->mmap_sem); |
03252919 | 3795 | } |
3ee1afa3 | 3796 | |
662bbcb2 | 3797 | #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) |
9ec23531 | 3798 | void __might_fault(const char *file, int line) |
3ee1afa3 | 3799 | { |
95156f00 PZ |
3800 | /* |
3801 | * Some code (nfs/sunrpc) uses socket ops on kernel memory while | |
3802 | * holding the mmap_sem, this is safe because kernel memory doesn't | |
3803 | * get paged out, therefore we'll never actually fault, and the | |
3804 | * below annotations will generate false positives. | |
3805 | */ | |
3806 | if (segment_eq(get_fs(), KERNEL_DS)) | |
3807 | return; | |
9ec23531 | 3808 | if (pagefault_disabled()) |
662bbcb2 | 3809 | return; |
9ec23531 DH |
3810 | __might_sleep(file, line, 0); |
3811 | #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) | |
662bbcb2 | 3812 | if (current->mm) |
3ee1afa3 | 3813 | might_lock_read(¤t->mm->mmap_sem); |
9ec23531 | 3814 | #endif |
3ee1afa3 | 3815 | } |
9ec23531 | 3816 | EXPORT_SYMBOL(__might_fault); |
3ee1afa3 | 3817 | #endif |
47ad8475 AA |
3818 | |
3819 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) | |
3820 | static void clear_gigantic_page(struct page *page, | |
3821 | unsigned long addr, | |
3822 | unsigned int pages_per_huge_page) | |
3823 | { | |
3824 | int i; | |
3825 | struct page *p = page; | |
3826 | ||
3827 | might_sleep(); | |
3828 | for (i = 0; i < pages_per_huge_page; | |
3829 | i++, p = mem_map_next(p, page, i)) { | |
3830 | cond_resched(); | |
3831 | clear_user_highpage(p, addr + i * PAGE_SIZE); | |
3832 | } | |
3833 | } | |
3834 | void clear_huge_page(struct page *page, | |
3835 | unsigned long addr, unsigned int pages_per_huge_page) | |
3836 | { | |
3837 | int i; | |
3838 | ||
3839 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
3840 | clear_gigantic_page(page, addr, pages_per_huge_page); | |
3841 | return; | |
3842 | } | |
3843 | ||
3844 | might_sleep(); | |
3845 | for (i = 0; i < pages_per_huge_page; i++) { | |
3846 | cond_resched(); | |
3847 | clear_user_highpage(page + i, addr + i * PAGE_SIZE); | |
3848 | } | |
3849 | } | |
3850 | ||
3851 | static void copy_user_gigantic_page(struct page *dst, struct page *src, | |
3852 | unsigned long addr, | |
3853 | struct vm_area_struct *vma, | |
3854 | unsigned int pages_per_huge_page) | |
3855 | { | |
3856 | int i; | |
3857 | struct page *dst_base = dst; | |
3858 | struct page *src_base = src; | |
3859 | ||
3860 | for (i = 0; i < pages_per_huge_page; ) { | |
3861 | cond_resched(); | |
3862 | copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); | |
3863 | ||
3864 | i++; | |
3865 | dst = mem_map_next(dst, dst_base, i); | |
3866 | src = mem_map_next(src, src_base, i); | |
3867 | } | |
3868 | } | |
3869 | ||
3870 | void copy_user_huge_page(struct page *dst, struct page *src, | |
3871 | unsigned long addr, struct vm_area_struct *vma, | |
3872 | unsigned int pages_per_huge_page) | |
3873 | { | |
3874 | int i; | |
3875 | ||
3876 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
3877 | copy_user_gigantic_page(dst, src, addr, vma, | |
3878 | pages_per_huge_page); | |
3879 | return; | |
3880 | } | |
3881 | ||
3882 | might_sleep(); | |
3883 | for (i = 0; i < pages_per_huge_page; i++) { | |
3884 | cond_resched(); | |
3885 | copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma); | |
3886 | } | |
3887 | } | |
3888 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ | |
49076ec2 | 3889 | |
40b64acd | 3890 | #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS |
b35f1819 KS |
3891 | |
3892 | static struct kmem_cache *page_ptl_cachep; | |
3893 | ||
3894 | void __init ptlock_cache_init(void) | |
3895 | { | |
3896 | page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, | |
3897 | SLAB_PANIC, NULL); | |
3898 | } | |
3899 | ||
539edb58 | 3900 | bool ptlock_alloc(struct page *page) |
49076ec2 KS |
3901 | { |
3902 | spinlock_t *ptl; | |
3903 | ||
b35f1819 | 3904 | ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); |
49076ec2 KS |
3905 | if (!ptl) |
3906 | return false; | |
539edb58 | 3907 | page->ptl = ptl; |
49076ec2 KS |
3908 | return true; |
3909 | } | |
3910 | ||
539edb58 | 3911 | void ptlock_free(struct page *page) |
49076ec2 | 3912 | { |
b35f1819 | 3913 | kmem_cache_free(page_ptl_cachep, page->ptl); |
49076ec2 KS |
3914 | } |
3915 | #endif |