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> |
33a709b2 | 68 | #include <asm/mmu_context.h> |
1da177e4 LT |
69 | #include <asm/pgalloc.h> |
70 | #include <asm/uaccess.h> | |
71 | #include <asm/tlb.h> | |
72 | #include <asm/tlbflush.h> | |
73 | #include <asm/pgtable.h> | |
74 | ||
42b77728 JB |
75 | #include "internal.h" |
76 | ||
90572890 PZ |
77 | #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS |
78 | #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid. | |
75980e97 PZ |
79 | #endif |
80 | ||
d41dee36 | 81 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
82 | /* use the per-pgdat data instead for discontigmem - mbligh */ |
83 | unsigned long max_mapnr; | |
84 | struct page *mem_map; | |
85 | ||
86 | EXPORT_SYMBOL(max_mapnr); | |
87 | EXPORT_SYMBOL(mem_map); | |
88 | #endif | |
89 | ||
1da177e4 LT |
90 | /* |
91 | * A number of key systems in x86 including ioremap() rely on the assumption | |
92 | * that high_memory defines the upper bound on direct map memory, then end | |
93 | * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and | |
94 | * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL | |
95 | * and ZONE_HIGHMEM. | |
96 | */ | |
97 | void * high_memory; | |
1da177e4 | 98 | |
1da177e4 | 99 | EXPORT_SYMBOL(high_memory); |
1da177e4 | 100 | |
32a93233 IM |
101 | /* |
102 | * Randomize the address space (stacks, mmaps, brk, etc.). | |
103 | * | |
104 | * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, | |
105 | * as ancient (libc5 based) binaries can segfault. ) | |
106 | */ | |
107 | int randomize_va_space __read_mostly = | |
108 | #ifdef CONFIG_COMPAT_BRK | |
109 | 1; | |
110 | #else | |
111 | 2; | |
112 | #endif | |
a62eaf15 AK |
113 | |
114 | static int __init disable_randmaps(char *s) | |
115 | { | |
116 | randomize_va_space = 0; | |
9b41046c | 117 | return 1; |
a62eaf15 AK |
118 | } |
119 | __setup("norandmaps", disable_randmaps); | |
120 | ||
62eede62 | 121 | unsigned long zero_pfn __read_mostly; |
03f6462a | 122 | unsigned long highest_memmap_pfn __read_mostly; |
a13ea5b7 | 123 | |
0b70068e AB |
124 | EXPORT_SYMBOL(zero_pfn); |
125 | ||
a13ea5b7 HD |
126 | /* |
127 | * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() | |
128 | */ | |
129 | static int __init init_zero_pfn(void) | |
130 | { | |
131 | zero_pfn = page_to_pfn(ZERO_PAGE(0)); | |
132 | return 0; | |
133 | } | |
134 | core_initcall(init_zero_pfn); | |
a62eaf15 | 135 | |
d559db08 | 136 | |
34e55232 KH |
137 | #if defined(SPLIT_RSS_COUNTING) |
138 | ||
ea48cf78 | 139 | void sync_mm_rss(struct mm_struct *mm) |
34e55232 KH |
140 | { |
141 | int i; | |
142 | ||
143 | for (i = 0; i < NR_MM_COUNTERS; i++) { | |
05af2e10 DR |
144 | if (current->rss_stat.count[i]) { |
145 | add_mm_counter(mm, i, current->rss_stat.count[i]); | |
146 | current->rss_stat.count[i] = 0; | |
34e55232 KH |
147 | } |
148 | } | |
05af2e10 | 149 | current->rss_stat.events = 0; |
34e55232 KH |
150 | } |
151 | ||
152 | static void add_mm_counter_fast(struct mm_struct *mm, int member, int val) | |
153 | { | |
154 | struct task_struct *task = current; | |
155 | ||
156 | if (likely(task->mm == mm)) | |
157 | task->rss_stat.count[member] += val; | |
158 | else | |
159 | add_mm_counter(mm, member, val); | |
160 | } | |
161 | #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1) | |
162 | #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1) | |
163 | ||
164 | /* sync counter once per 64 page faults */ | |
165 | #define TASK_RSS_EVENTS_THRESH (64) | |
166 | static void check_sync_rss_stat(struct task_struct *task) | |
167 | { | |
168 | if (unlikely(task != current)) | |
169 | return; | |
170 | if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH)) | |
ea48cf78 | 171 | sync_mm_rss(task->mm); |
34e55232 | 172 | } |
9547d01b | 173 | #else /* SPLIT_RSS_COUNTING */ |
34e55232 KH |
174 | |
175 | #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member) | |
176 | #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member) | |
177 | ||
178 | static void check_sync_rss_stat(struct task_struct *task) | |
179 | { | |
180 | } | |
181 | ||
9547d01b PZ |
182 | #endif /* SPLIT_RSS_COUNTING */ |
183 | ||
184 | #ifdef HAVE_GENERIC_MMU_GATHER | |
185 | ||
ca1d6c7d | 186 | static bool tlb_next_batch(struct mmu_gather *tlb) |
9547d01b PZ |
187 | { |
188 | struct mmu_gather_batch *batch; | |
189 | ||
190 | batch = tlb->active; | |
191 | if (batch->next) { | |
192 | tlb->active = batch->next; | |
ca1d6c7d | 193 | return true; |
9547d01b PZ |
194 | } |
195 | ||
53a59fc6 | 196 | if (tlb->batch_count == MAX_GATHER_BATCH_COUNT) |
ca1d6c7d | 197 | return false; |
53a59fc6 | 198 | |
9547d01b PZ |
199 | batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0); |
200 | if (!batch) | |
ca1d6c7d | 201 | return false; |
9547d01b | 202 | |
53a59fc6 | 203 | tlb->batch_count++; |
9547d01b PZ |
204 | batch->next = NULL; |
205 | batch->nr = 0; | |
206 | batch->max = MAX_GATHER_BATCH; | |
207 | ||
208 | tlb->active->next = batch; | |
209 | tlb->active = batch; | |
210 | ||
ca1d6c7d | 211 | return true; |
9547d01b PZ |
212 | } |
213 | ||
214 | /* tlb_gather_mmu | |
215 | * Called to initialize an (on-stack) mmu_gather structure for page-table | |
216 | * tear-down from @mm. The @fullmm argument is used when @mm is without | |
217 | * users and we're going to destroy the full address space (exit/execve). | |
218 | */ | |
2b047252 | 219 | void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, unsigned long start, unsigned long end) |
9547d01b PZ |
220 | { |
221 | tlb->mm = mm; | |
222 | ||
2b047252 LT |
223 | /* Is it from 0 to ~0? */ |
224 | tlb->fullmm = !(start | (end+1)); | |
1de14c3c | 225 | tlb->need_flush_all = 0; |
9547d01b PZ |
226 | tlb->local.next = NULL; |
227 | tlb->local.nr = 0; | |
228 | tlb->local.max = ARRAY_SIZE(tlb->__pages); | |
229 | tlb->active = &tlb->local; | |
53a59fc6 | 230 | tlb->batch_count = 0; |
9547d01b PZ |
231 | |
232 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE | |
233 | tlb->batch = NULL; | |
234 | #endif | |
fb7332a9 WD |
235 | |
236 | __tlb_reset_range(tlb); | |
9547d01b PZ |
237 | } |
238 | ||
1cf35d47 | 239 | static void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb) |
9547d01b | 240 | { |
721c21c1 WD |
241 | if (!tlb->end) |
242 | return; | |
243 | ||
9547d01b | 244 | tlb_flush(tlb); |
34ee645e | 245 | mmu_notifier_invalidate_range(tlb->mm, tlb->start, tlb->end); |
9547d01b PZ |
246 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE |
247 | tlb_table_flush(tlb); | |
34e55232 | 248 | #endif |
fb7332a9 | 249 | __tlb_reset_range(tlb); |
1cf35d47 LT |
250 | } |
251 | ||
252 | static void tlb_flush_mmu_free(struct mmu_gather *tlb) | |
253 | { | |
254 | struct mmu_gather_batch *batch; | |
34e55232 | 255 | |
721c21c1 | 256 | for (batch = &tlb->local; batch && batch->nr; batch = batch->next) { |
9547d01b PZ |
257 | free_pages_and_swap_cache(batch->pages, batch->nr); |
258 | batch->nr = 0; | |
259 | } | |
260 | tlb->active = &tlb->local; | |
261 | } | |
262 | ||
1cf35d47 LT |
263 | void tlb_flush_mmu(struct mmu_gather *tlb) |
264 | { | |
1cf35d47 LT |
265 | tlb_flush_mmu_tlbonly(tlb); |
266 | tlb_flush_mmu_free(tlb); | |
267 | } | |
268 | ||
9547d01b PZ |
269 | /* tlb_finish_mmu |
270 | * Called at the end of the shootdown operation to free up any resources | |
271 | * that were required. | |
272 | */ | |
273 | void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end) | |
274 | { | |
275 | struct mmu_gather_batch *batch, *next; | |
276 | ||
277 | tlb_flush_mmu(tlb); | |
278 | ||
279 | /* keep the page table cache within bounds */ | |
280 | check_pgt_cache(); | |
281 | ||
282 | for (batch = tlb->local.next; batch; batch = next) { | |
283 | next = batch->next; | |
284 | free_pages((unsigned long)batch, 0); | |
285 | } | |
286 | tlb->local.next = NULL; | |
287 | } | |
288 | ||
289 | /* __tlb_remove_page | |
290 | * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while | |
291 | * handling the additional races in SMP caused by other CPUs caching valid | |
292 | * mappings in their TLBs. Returns the number of free page slots left. | |
293 | * When out of page slots we must call tlb_flush_mmu(). | |
294 | */ | |
295 | int __tlb_remove_page(struct mmu_gather *tlb, struct page *page) | |
296 | { | |
297 | struct mmu_gather_batch *batch; | |
298 | ||
fb7332a9 | 299 | VM_BUG_ON(!tlb->end); |
9547d01b | 300 | |
9547d01b PZ |
301 | batch = tlb->active; |
302 | batch->pages[batch->nr++] = page; | |
303 | if (batch->nr == batch->max) { | |
304 | if (!tlb_next_batch(tlb)) | |
305 | return 0; | |
0b43c3aa | 306 | batch = tlb->active; |
9547d01b | 307 | } |
309381fe | 308 | VM_BUG_ON_PAGE(batch->nr > batch->max, page); |
9547d01b PZ |
309 | |
310 | return batch->max - batch->nr; | |
311 | } | |
312 | ||
313 | #endif /* HAVE_GENERIC_MMU_GATHER */ | |
314 | ||
26723911 PZ |
315 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE |
316 | ||
317 | /* | |
318 | * See the comment near struct mmu_table_batch. | |
319 | */ | |
320 | ||
321 | static void tlb_remove_table_smp_sync(void *arg) | |
322 | { | |
323 | /* Simply deliver the interrupt */ | |
324 | } | |
325 | ||
326 | static void tlb_remove_table_one(void *table) | |
327 | { | |
328 | /* | |
329 | * This isn't an RCU grace period and hence the page-tables cannot be | |
330 | * assumed to be actually RCU-freed. | |
331 | * | |
332 | * It is however sufficient for software page-table walkers that rely on | |
333 | * IRQ disabling. See the comment near struct mmu_table_batch. | |
334 | */ | |
335 | smp_call_function(tlb_remove_table_smp_sync, NULL, 1); | |
336 | __tlb_remove_table(table); | |
337 | } | |
338 | ||
339 | static void tlb_remove_table_rcu(struct rcu_head *head) | |
340 | { | |
341 | struct mmu_table_batch *batch; | |
342 | int i; | |
343 | ||
344 | batch = container_of(head, struct mmu_table_batch, rcu); | |
345 | ||
346 | for (i = 0; i < batch->nr; i++) | |
347 | __tlb_remove_table(batch->tables[i]); | |
348 | ||
349 | free_page((unsigned long)batch); | |
350 | } | |
351 | ||
352 | void tlb_table_flush(struct mmu_gather *tlb) | |
353 | { | |
354 | struct mmu_table_batch **batch = &tlb->batch; | |
355 | ||
356 | if (*batch) { | |
357 | call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu); | |
358 | *batch = NULL; | |
359 | } | |
360 | } | |
361 | ||
362 | void tlb_remove_table(struct mmu_gather *tlb, void *table) | |
363 | { | |
364 | struct mmu_table_batch **batch = &tlb->batch; | |
365 | ||
26723911 PZ |
366 | /* |
367 | * When there's less then two users of this mm there cannot be a | |
368 | * concurrent page-table walk. | |
369 | */ | |
370 | if (atomic_read(&tlb->mm->mm_users) < 2) { | |
371 | __tlb_remove_table(table); | |
372 | return; | |
373 | } | |
374 | ||
375 | if (*batch == NULL) { | |
376 | *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN); | |
377 | if (*batch == NULL) { | |
378 | tlb_remove_table_one(table); | |
379 | return; | |
380 | } | |
381 | (*batch)->nr = 0; | |
382 | } | |
383 | (*batch)->tables[(*batch)->nr++] = table; | |
384 | if ((*batch)->nr == MAX_TABLE_BATCH) | |
385 | tlb_table_flush(tlb); | |
386 | } | |
387 | ||
9547d01b | 388 | #endif /* CONFIG_HAVE_RCU_TABLE_FREE */ |
26723911 | 389 | |
1da177e4 LT |
390 | /* |
391 | * Note: this doesn't free the actual pages themselves. That | |
392 | * has been handled earlier when unmapping all the memory regions. | |
393 | */ | |
9e1b32ca BH |
394 | static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, |
395 | unsigned long addr) | |
1da177e4 | 396 | { |
2f569afd | 397 | pgtable_t token = pmd_pgtable(*pmd); |
e0da382c | 398 | pmd_clear(pmd); |
9e1b32ca | 399 | pte_free_tlb(tlb, token, addr); |
e1f56c89 | 400 | atomic_long_dec(&tlb->mm->nr_ptes); |
1da177e4 LT |
401 | } |
402 | ||
e0da382c HD |
403 | static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
404 | unsigned long addr, unsigned long end, | |
405 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
406 | { |
407 | pmd_t *pmd; | |
408 | unsigned long next; | |
e0da382c | 409 | unsigned long start; |
1da177e4 | 410 | |
e0da382c | 411 | start = addr; |
1da177e4 | 412 | pmd = pmd_offset(pud, addr); |
1da177e4 LT |
413 | do { |
414 | next = pmd_addr_end(addr, end); | |
415 | if (pmd_none_or_clear_bad(pmd)) | |
416 | continue; | |
9e1b32ca | 417 | free_pte_range(tlb, pmd, addr); |
1da177e4 LT |
418 | } while (pmd++, addr = next, addr != end); |
419 | ||
e0da382c HD |
420 | start &= PUD_MASK; |
421 | if (start < floor) | |
422 | return; | |
423 | if (ceiling) { | |
424 | ceiling &= PUD_MASK; | |
425 | if (!ceiling) | |
426 | return; | |
1da177e4 | 427 | } |
e0da382c HD |
428 | if (end - 1 > ceiling - 1) |
429 | return; | |
430 | ||
431 | pmd = pmd_offset(pud, start); | |
432 | pud_clear(pud); | |
9e1b32ca | 433 | pmd_free_tlb(tlb, pmd, start); |
dc6c9a35 | 434 | mm_dec_nr_pmds(tlb->mm); |
1da177e4 LT |
435 | } |
436 | ||
e0da382c HD |
437 | static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, |
438 | unsigned long addr, unsigned long end, | |
439 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
440 | { |
441 | pud_t *pud; | |
442 | unsigned long next; | |
e0da382c | 443 | unsigned long start; |
1da177e4 | 444 | |
e0da382c | 445 | start = addr; |
1da177e4 | 446 | pud = pud_offset(pgd, addr); |
1da177e4 LT |
447 | do { |
448 | next = pud_addr_end(addr, end); | |
449 | if (pud_none_or_clear_bad(pud)) | |
450 | continue; | |
e0da382c | 451 | free_pmd_range(tlb, pud, addr, next, floor, ceiling); |
1da177e4 LT |
452 | } while (pud++, addr = next, addr != end); |
453 | ||
e0da382c HD |
454 | start &= PGDIR_MASK; |
455 | if (start < floor) | |
456 | return; | |
457 | if (ceiling) { | |
458 | ceiling &= PGDIR_MASK; | |
459 | if (!ceiling) | |
460 | return; | |
1da177e4 | 461 | } |
e0da382c HD |
462 | if (end - 1 > ceiling - 1) |
463 | return; | |
464 | ||
465 | pud = pud_offset(pgd, start); | |
466 | pgd_clear(pgd); | |
9e1b32ca | 467 | pud_free_tlb(tlb, pud, start); |
1da177e4 LT |
468 | } |
469 | ||
470 | /* | |
e0da382c | 471 | * This function frees user-level page tables of a process. |
1da177e4 | 472 | */ |
42b77728 | 473 | void free_pgd_range(struct mmu_gather *tlb, |
e0da382c HD |
474 | unsigned long addr, unsigned long end, |
475 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
476 | { |
477 | pgd_t *pgd; | |
478 | unsigned long next; | |
e0da382c HD |
479 | |
480 | /* | |
481 | * The next few lines have given us lots of grief... | |
482 | * | |
483 | * Why are we testing PMD* at this top level? Because often | |
484 | * there will be no work to do at all, and we'd prefer not to | |
485 | * go all the way down to the bottom just to discover that. | |
486 | * | |
487 | * Why all these "- 1"s? Because 0 represents both the bottom | |
488 | * of the address space and the top of it (using -1 for the | |
489 | * top wouldn't help much: the masks would do the wrong thing). | |
490 | * The rule is that addr 0 and floor 0 refer to the bottom of | |
491 | * the address space, but end 0 and ceiling 0 refer to the top | |
492 | * Comparisons need to use "end - 1" and "ceiling - 1" (though | |
493 | * that end 0 case should be mythical). | |
494 | * | |
495 | * Wherever addr is brought up or ceiling brought down, we must | |
496 | * be careful to reject "the opposite 0" before it confuses the | |
497 | * subsequent tests. But what about where end is brought down | |
498 | * by PMD_SIZE below? no, end can't go down to 0 there. | |
499 | * | |
500 | * Whereas we round start (addr) and ceiling down, by different | |
501 | * masks at different levels, in order to test whether a table | |
502 | * now has no other vmas using it, so can be freed, we don't | |
503 | * bother to round floor or end up - the tests don't need that. | |
504 | */ | |
1da177e4 | 505 | |
e0da382c HD |
506 | addr &= PMD_MASK; |
507 | if (addr < floor) { | |
508 | addr += PMD_SIZE; | |
509 | if (!addr) | |
510 | return; | |
511 | } | |
512 | if (ceiling) { | |
513 | ceiling &= PMD_MASK; | |
514 | if (!ceiling) | |
515 | return; | |
516 | } | |
517 | if (end - 1 > ceiling - 1) | |
518 | end -= PMD_SIZE; | |
519 | if (addr > end - 1) | |
520 | return; | |
521 | ||
42b77728 | 522 | pgd = pgd_offset(tlb->mm, addr); |
1da177e4 LT |
523 | do { |
524 | next = pgd_addr_end(addr, end); | |
525 | if (pgd_none_or_clear_bad(pgd)) | |
526 | continue; | |
42b77728 | 527 | free_pud_range(tlb, pgd, addr, next, floor, ceiling); |
1da177e4 | 528 | } while (pgd++, addr = next, addr != end); |
e0da382c HD |
529 | } |
530 | ||
42b77728 | 531 | void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, |
3bf5ee95 | 532 | unsigned long floor, unsigned long ceiling) |
e0da382c HD |
533 | { |
534 | while (vma) { | |
535 | struct vm_area_struct *next = vma->vm_next; | |
536 | unsigned long addr = vma->vm_start; | |
537 | ||
8f4f8c16 | 538 | /* |
25d9e2d1 | 539 | * Hide vma from rmap and truncate_pagecache before freeing |
540 | * pgtables | |
8f4f8c16 | 541 | */ |
5beb4930 | 542 | unlink_anon_vmas(vma); |
8f4f8c16 HD |
543 | unlink_file_vma(vma); |
544 | ||
9da61aef | 545 | if (is_vm_hugetlb_page(vma)) { |
3bf5ee95 | 546 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
e0da382c | 547 | floor, next? next->vm_start: ceiling); |
3bf5ee95 HD |
548 | } else { |
549 | /* | |
550 | * Optimization: gather nearby vmas into one call down | |
551 | */ | |
552 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE | |
4866920b | 553 | && !is_vm_hugetlb_page(next)) { |
3bf5ee95 HD |
554 | vma = next; |
555 | next = vma->vm_next; | |
5beb4930 | 556 | unlink_anon_vmas(vma); |
8f4f8c16 | 557 | unlink_file_vma(vma); |
3bf5ee95 HD |
558 | } |
559 | free_pgd_range(tlb, addr, vma->vm_end, | |
560 | floor, next? next->vm_start: ceiling); | |
561 | } | |
e0da382c HD |
562 | vma = next; |
563 | } | |
1da177e4 LT |
564 | } |
565 | ||
3ed3a4f0 | 566 | int __pte_alloc(struct mm_struct *mm, 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) { | |
1170532b JP |
664 | pr_alert("BUG: Bad page map: %lu messages suppressed\n", |
665 | nr_unshown); | |
d936cf9b HD |
666 | nr_unshown = 0; |
667 | } | |
668 | nr_shown = 0; | |
669 | } | |
670 | if (nr_shown++ == 0) | |
671 | resume = jiffies + 60 * HZ; | |
3dc14741 HD |
672 | |
673 | mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; | |
674 | index = linear_page_index(vma, addr); | |
675 | ||
1170532b JP |
676 | pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", |
677 | current->comm, | |
678 | (long long)pte_val(pte), (long long)pmd_val(*pmd)); | |
718a3821 | 679 | if (page) |
f0b791a3 | 680 | dump_page(page, "bad pte"); |
1170532b JP |
681 | pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n", |
682 | (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); | |
3dc14741 HD |
683 | /* |
684 | * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y | |
685 | */ | |
2682582a KK |
686 | pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n", |
687 | vma->vm_file, | |
688 | vma->vm_ops ? vma->vm_ops->fault : NULL, | |
689 | vma->vm_file ? vma->vm_file->f_op->mmap : NULL, | |
690 | mapping ? mapping->a_ops->readpage : NULL); | |
b5810039 | 691 | dump_stack(); |
373d4d09 | 692 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
b5810039 NP |
693 | } |
694 | ||
ee498ed7 | 695 | /* |
7e675137 | 696 | * vm_normal_page -- This function gets the "struct page" associated with a pte. |
6aab341e | 697 | * |
7e675137 NP |
698 | * "Special" mappings do not wish to be associated with a "struct page" (either |
699 | * it doesn't exist, or it exists but they don't want to touch it). In this | |
700 | * case, NULL is returned here. "Normal" mappings do have a struct page. | |
b379d790 | 701 | * |
7e675137 NP |
702 | * There are 2 broad cases. Firstly, an architecture may define a pte_special() |
703 | * pte bit, in which case this function is trivial. Secondly, an architecture | |
704 | * may not have a spare pte bit, which requires a more complicated scheme, | |
705 | * described below. | |
706 | * | |
707 | * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a | |
708 | * special mapping (even if there are underlying and valid "struct pages"). | |
709 | * COWed pages of a VM_PFNMAP are always normal. | |
6aab341e | 710 | * |
b379d790 JH |
711 | * The way we recognize COWed pages within VM_PFNMAP mappings is through the |
712 | * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit | |
7e675137 NP |
713 | * set, and the vm_pgoff will point to the first PFN mapped: thus every special |
714 | * mapping will always honor the rule | |
6aab341e LT |
715 | * |
716 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) | |
717 | * | |
7e675137 NP |
718 | * And for normal mappings this is false. |
719 | * | |
720 | * This restricts such mappings to be a linear translation from virtual address | |
721 | * to pfn. To get around this restriction, we allow arbitrary mappings so long | |
722 | * as the vma is not a COW mapping; in that case, we know that all ptes are | |
723 | * special (because none can have been COWed). | |
b379d790 | 724 | * |
b379d790 | 725 | * |
7e675137 | 726 | * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. |
b379d790 JH |
727 | * |
728 | * VM_MIXEDMAP mappings can likewise contain memory with or without "struct | |
729 | * page" backing, however the difference is that _all_ pages with a struct | |
730 | * page (that is, those where pfn_valid is true) are refcounted and considered | |
731 | * normal pages by the VM. The disadvantage is that pages are refcounted | |
732 | * (which can be slower and simply not an option for some PFNMAP users). The | |
733 | * advantage is that we don't have to follow the strict linearity rule of | |
734 | * PFNMAP mappings in order to support COWable mappings. | |
735 | * | |
ee498ed7 | 736 | */ |
7e675137 NP |
737 | #ifdef __HAVE_ARCH_PTE_SPECIAL |
738 | # define HAVE_PTE_SPECIAL 1 | |
739 | #else | |
740 | # define HAVE_PTE_SPECIAL 0 | |
741 | #endif | |
742 | struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, | |
743 | pte_t pte) | |
ee498ed7 | 744 | { |
22b31eec | 745 | unsigned long pfn = pte_pfn(pte); |
7e675137 NP |
746 | |
747 | if (HAVE_PTE_SPECIAL) { | |
b38af472 | 748 | if (likely(!pte_special(pte))) |
22b31eec | 749 | goto check_pfn; |
667a0a06 DV |
750 | if (vma->vm_ops && vma->vm_ops->find_special_page) |
751 | return vma->vm_ops->find_special_page(vma, addr); | |
a13ea5b7 HD |
752 | if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) |
753 | return NULL; | |
62eede62 | 754 | if (!is_zero_pfn(pfn)) |
22b31eec | 755 | print_bad_pte(vma, addr, pte, NULL); |
7e675137 NP |
756 | return NULL; |
757 | } | |
758 | ||
759 | /* !HAVE_PTE_SPECIAL case follows: */ | |
760 | ||
b379d790 JH |
761 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
762 | if (vma->vm_flags & VM_MIXEDMAP) { | |
763 | if (!pfn_valid(pfn)) | |
764 | return NULL; | |
765 | goto out; | |
766 | } else { | |
7e675137 NP |
767 | unsigned long off; |
768 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
b379d790 JH |
769 | if (pfn == vma->vm_pgoff + off) |
770 | return NULL; | |
771 | if (!is_cow_mapping(vma->vm_flags)) | |
772 | return NULL; | |
773 | } | |
6aab341e LT |
774 | } |
775 | ||
b38af472 HD |
776 | if (is_zero_pfn(pfn)) |
777 | return NULL; | |
22b31eec HD |
778 | check_pfn: |
779 | if (unlikely(pfn > highest_memmap_pfn)) { | |
780 | print_bad_pte(vma, addr, pte, NULL); | |
781 | return NULL; | |
782 | } | |
6aab341e LT |
783 | |
784 | /* | |
7e675137 | 785 | * NOTE! We still have PageReserved() pages in the page tables. |
7e675137 | 786 | * eg. VDSO mappings can cause them to exist. |
6aab341e | 787 | */ |
b379d790 | 788 | out: |
6aab341e | 789 | return pfn_to_page(pfn); |
ee498ed7 HD |
790 | } |
791 | ||
28093f9f GS |
792 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
793 | struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr, | |
794 | pmd_t pmd) | |
795 | { | |
796 | unsigned long pfn = pmd_pfn(pmd); | |
797 | ||
798 | /* | |
799 | * There is no pmd_special() but there may be special pmds, e.g. | |
800 | * in a direct-access (dax) mapping, so let's just replicate the | |
801 | * !HAVE_PTE_SPECIAL case from vm_normal_page() here. | |
802 | */ | |
803 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { | |
804 | if (vma->vm_flags & VM_MIXEDMAP) { | |
805 | if (!pfn_valid(pfn)) | |
806 | return NULL; | |
807 | goto out; | |
808 | } else { | |
809 | unsigned long off; | |
810 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
811 | if (pfn == vma->vm_pgoff + off) | |
812 | return NULL; | |
813 | if (!is_cow_mapping(vma->vm_flags)) | |
814 | return NULL; | |
815 | } | |
816 | } | |
817 | ||
818 | if (is_zero_pfn(pfn)) | |
819 | return NULL; | |
820 | if (unlikely(pfn > highest_memmap_pfn)) | |
821 | return NULL; | |
822 | ||
823 | /* | |
824 | * NOTE! We still have PageReserved() pages in the page tables. | |
825 | * eg. VDSO mappings can cause them to exist. | |
826 | */ | |
827 | out: | |
828 | return pfn_to_page(pfn); | |
829 | } | |
830 | #endif | |
831 | ||
1da177e4 LT |
832 | /* |
833 | * copy one vm_area from one task to the other. Assumes the page tables | |
834 | * already present in the new task to be cleared in the whole range | |
835 | * covered by this vma. | |
1da177e4 LT |
836 | */ |
837 | ||
570a335b | 838 | static inline unsigned long |
1da177e4 | 839 | copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
b5810039 | 840 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, |
8c103762 | 841 | unsigned long addr, int *rss) |
1da177e4 | 842 | { |
b5810039 | 843 | unsigned long vm_flags = vma->vm_flags; |
1da177e4 LT |
844 | pte_t pte = *src_pte; |
845 | struct page *page; | |
1da177e4 LT |
846 | |
847 | /* pte contains position in swap or file, so copy. */ | |
848 | if (unlikely(!pte_present(pte))) { | |
0661a336 KS |
849 | swp_entry_t entry = pte_to_swp_entry(pte); |
850 | ||
851 | if (likely(!non_swap_entry(entry))) { | |
852 | if (swap_duplicate(entry) < 0) | |
853 | return entry.val; | |
854 | ||
855 | /* make sure dst_mm is on swapoff's mmlist. */ | |
856 | if (unlikely(list_empty(&dst_mm->mmlist))) { | |
857 | spin_lock(&mmlist_lock); | |
858 | if (list_empty(&dst_mm->mmlist)) | |
859 | list_add(&dst_mm->mmlist, | |
860 | &src_mm->mmlist); | |
861 | spin_unlock(&mmlist_lock); | |
862 | } | |
863 | rss[MM_SWAPENTS]++; | |
864 | } else if (is_migration_entry(entry)) { | |
865 | page = migration_entry_to_page(entry); | |
866 | ||
eca56ff9 | 867 | rss[mm_counter(page)]++; |
0661a336 KS |
868 | |
869 | if (is_write_migration_entry(entry) && | |
870 | is_cow_mapping(vm_flags)) { | |
871 | /* | |
872 | * COW mappings require pages in both | |
873 | * parent and child to be set to read. | |
874 | */ | |
875 | make_migration_entry_read(&entry); | |
876 | pte = swp_entry_to_pte(entry); | |
877 | if (pte_swp_soft_dirty(*src_pte)) | |
878 | pte = pte_swp_mksoft_dirty(pte); | |
879 | set_pte_at(src_mm, addr, src_pte, pte); | |
0697212a | 880 | } |
1da177e4 | 881 | } |
ae859762 | 882 | goto out_set_pte; |
1da177e4 LT |
883 | } |
884 | ||
1da177e4 LT |
885 | /* |
886 | * If it's a COW mapping, write protect it both | |
887 | * in the parent and the child | |
888 | */ | |
67121172 | 889 | if (is_cow_mapping(vm_flags)) { |
1da177e4 | 890 | ptep_set_wrprotect(src_mm, addr, src_pte); |
3dc90795 | 891 | pte = pte_wrprotect(pte); |
1da177e4 LT |
892 | } |
893 | ||
894 | /* | |
895 | * If it's a shared mapping, mark it clean in | |
896 | * the child | |
897 | */ | |
898 | if (vm_flags & VM_SHARED) | |
899 | pte = pte_mkclean(pte); | |
900 | pte = pte_mkold(pte); | |
6aab341e LT |
901 | |
902 | page = vm_normal_page(vma, addr, pte); | |
903 | if (page) { | |
904 | get_page(page); | |
53f9263b | 905 | page_dup_rmap(page, false); |
eca56ff9 | 906 | rss[mm_counter(page)]++; |
6aab341e | 907 | } |
ae859762 HD |
908 | |
909 | out_set_pte: | |
910 | set_pte_at(dst_mm, addr, dst_pte, pte); | |
570a335b | 911 | return 0; |
1da177e4 LT |
912 | } |
913 | ||
21bda264 | 914 | static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
71e3aac0 AA |
915 | pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, |
916 | unsigned long addr, unsigned long end) | |
1da177e4 | 917 | { |
c36987e2 | 918 | pte_t *orig_src_pte, *orig_dst_pte; |
1da177e4 | 919 | pte_t *src_pte, *dst_pte; |
c74df32c | 920 | spinlock_t *src_ptl, *dst_ptl; |
e040f218 | 921 | int progress = 0; |
d559db08 | 922 | int rss[NR_MM_COUNTERS]; |
570a335b | 923 | swp_entry_t entry = (swp_entry_t){0}; |
1da177e4 LT |
924 | |
925 | again: | |
d559db08 KH |
926 | init_rss_vec(rss); |
927 | ||
c74df32c | 928 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
1da177e4 LT |
929 | if (!dst_pte) |
930 | return -ENOMEM; | |
ece0e2b6 | 931 | src_pte = pte_offset_map(src_pmd, addr); |
4c21e2f2 | 932 | src_ptl = pte_lockptr(src_mm, src_pmd); |
f20dc5f7 | 933 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
c36987e2 DN |
934 | orig_src_pte = src_pte; |
935 | orig_dst_pte = dst_pte; | |
6606c3e0 | 936 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 937 | |
1da177e4 LT |
938 | do { |
939 | /* | |
940 | * We are holding two locks at this point - either of them | |
941 | * could generate latencies in another task on another CPU. | |
942 | */ | |
e040f218 HD |
943 | if (progress >= 32) { |
944 | progress = 0; | |
945 | if (need_resched() || | |
95c354fe | 946 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
e040f218 HD |
947 | break; |
948 | } | |
1da177e4 LT |
949 | if (pte_none(*src_pte)) { |
950 | progress++; | |
951 | continue; | |
952 | } | |
570a335b HD |
953 | entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, |
954 | vma, addr, rss); | |
955 | if (entry.val) | |
956 | break; | |
1da177e4 LT |
957 | progress += 8; |
958 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); | |
1da177e4 | 959 | |
6606c3e0 | 960 | arch_leave_lazy_mmu_mode(); |
c74df32c | 961 | spin_unlock(src_ptl); |
ece0e2b6 | 962 | pte_unmap(orig_src_pte); |
d559db08 | 963 | add_mm_rss_vec(dst_mm, rss); |
c36987e2 | 964 | pte_unmap_unlock(orig_dst_pte, dst_ptl); |
c74df32c | 965 | cond_resched(); |
570a335b HD |
966 | |
967 | if (entry.val) { | |
968 | if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) | |
969 | return -ENOMEM; | |
970 | progress = 0; | |
971 | } | |
1da177e4 LT |
972 | if (addr != end) |
973 | goto again; | |
974 | return 0; | |
975 | } | |
976 | ||
977 | static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
978 | pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, | |
979 | unsigned long addr, unsigned long end) | |
980 | { | |
981 | pmd_t *src_pmd, *dst_pmd; | |
982 | unsigned long next; | |
983 | ||
984 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); | |
985 | if (!dst_pmd) | |
986 | return -ENOMEM; | |
987 | src_pmd = pmd_offset(src_pud, addr); | |
988 | do { | |
989 | next = pmd_addr_end(addr, end); | |
5c7fb56e | 990 | if (pmd_trans_huge(*src_pmd) || pmd_devmap(*src_pmd)) { |
71e3aac0 | 991 | int err; |
14d1a55c | 992 | VM_BUG_ON(next-addr != HPAGE_PMD_SIZE); |
71e3aac0 AA |
993 | err = copy_huge_pmd(dst_mm, src_mm, |
994 | dst_pmd, src_pmd, addr, vma); | |
995 | if (err == -ENOMEM) | |
996 | return -ENOMEM; | |
997 | if (!err) | |
998 | continue; | |
999 | /* fall through */ | |
1000 | } | |
1da177e4 LT |
1001 | if (pmd_none_or_clear_bad(src_pmd)) |
1002 | continue; | |
1003 | if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, | |
1004 | vma, addr, next)) | |
1005 | return -ENOMEM; | |
1006 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); | |
1007 | return 0; | |
1008 | } | |
1009 | ||
1010 | static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
1011 | pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, | |
1012 | unsigned long addr, unsigned long end) | |
1013 | { | |
1014 | pud_t *src_pud, *dst_pud; | |
1015 | unsigned long next; | |
1016 | ||
1017 | dst_pud = pud_alloc(dst_mm, dst_pgd, addr); | |
1018 | if (!dst_pud) | |
1019 | return -ENOMEM; | |
1020 | src_pud = pud_offset(src_pgd, addr); | |
1021 | do { | |
1022 | next = pud_addr_end(addr, end); | |
1023 | if (pud_none_or_clear_bad(src_pud)) | |
1024 | continue; | |
1025 | if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, | |
1026 | vma, addr, next)) | |
1027 | return -ENOMEM; | |
1028 | } while (dst_pud++, src_pud++, addr = next, addr != end); | |
1029 | return 0; | |
1030 | } | |
1031 | ||
1032 | int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
1033 | struct vm_area_struct *vma) | |
1034 | { | |
1035 | pgd_t *src_pgd, *dst_pgd; | |
1036 | unsigned long next; | |
1037 | unsigned long addr = vma->vm_start; | |
1038 | unsigned long end = vma->vm_end; | |
2ec74c3e SG |
1039 | unsigned long mmun_start; /* For mmu_notifiers */ |
1040 | unsigned long mmun_end; /* For mmu_notifiers */ | |
1041 | bool is_cow; | |
cddb8a5c | 1042 | int ret; |
1da177e4 | 1043 | |
d992895b NP |
1044 | /* |
1045 | * Don't copy ptes where a page fault will fill them correctly. | |
1046 | * Fork becomes much lighter when there are big shared or private | |
1047 | * readonly mappings. The tradeoff is that copy_page_range is more | |
1048 | * efficient than faulting. | |
1049 | */ | |
0661a336 KS |
1050 | if (!(vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) && |
1051 | !vma->anon_vma) | |
1052 | return 0; | |
d992895b | 1053 | |
1da177e4 LT |
1054 | if (is_vm_hugetlb_page(vma)) |
1055 | return copy_hugetlb_page_range(dst_mm, src_mm, vma); | |
1056 | ||
b3b9c293 | 1057 | if (unlikely(vma->vm_flags & VM_PFNMAP)) { |
2ab64037 | 1058 | /* |
1059 | * We do not free on error cases below as remove_vma | |
1060 | * gets called on error from higher level routine | |
1061 | */ | |
5180da41 | 1062 | ret = track_pfn_copy(vma); |
2ab64037 | 1063 | if (ret) |
1064 | return ret; | |
1065 | } | |
1066 | ||
cddb8a5c AA |
1067 | /* |
1068 | * We need to invalidate the secondary MMU mappings only when | |
1069 | * there could be a permission downgrade on the ptes of the | |
1070 | * parent mm. And a permission downgrade will only happen if | |
1071 | * is_cow_mapping() returns true. | |
1072 | */ | |
2ec74c3e SG |
1073 | is_cow = is_cow_mapping(vma->vm_flags); |
1074 | mmun_start = addr; | |
1075 | mmun_end = end; | |
1076 | if (is_cow) | |
1077 | mmu_notifier_invalidate_range_start(src_mm, mmun_start, | |
1078 | mmun_end); | |
cddb8a5c AA |
1079 | |
1080 | ret = 0; | |
1da177e4 LT |
1081 | dst_pgd = pgd_offset(dst_mm, addr); |
1082 | src_pgd = pgd_offset(src_mm, addr); | |
1083 | do { | |
1084 | next = pgd_addr_end(addr, end); | |
1085 | if (pgd_none_or_clear_bad(src_pgd)) | |
1086 | continue; | |
cddb8a5c AA |
1087 | if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, |
1088 | vma, addr, next))) { | |
1089 | ret = -ENOMEM; | |
1090 | break; | |
1091 | } | |
1da177e4 | 1092 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); |
cddb8a5c | 1093 | |
2ec74c3e SG |
1094 | if (is_cow) |
1095 | mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end); | |
cddb8a5c | 1096 | return ret; |
1da177e4 LT |
1097 | } |
1098 | ||
51c6f666 | 1099 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
b5810039 | 1100 | struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 | 1101 | unsigned long addr, unsigned long end, |
97a89413 | 1102 | struct zap_details *details) |
1da177e4 | 1103 | { |
b5810039 | 1104 | struct mm_struct *mm = tlb->mm; |
d16dfc55 | 1105 | int force_flush = 0; |
d559db08 | 1106 | int rss[NR_MM_COUNTERS]; |
97a89413 | 1107 | spinlock_t *ptl; |
5f1a1907 | 1108 | pte_t *start_pte; |
97a89413 | 1109 | pte_t *pte; |
8a5f14a2 | 1110 | swp_entry_t entry; |
d559db08 | 1111 | |
d16dfc55 | 1112 | again: |
e303297e | 1113 | init_rss_vec(rss); |
5f1a1907 SR |
1114 | start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
1115 | pte = start_pte; | |
6606c3e0 | 1116 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1117 | do { |
1118 | pte_t ptent = *pte; | |
51c6f666 | 1119 | if (pte_none(ptent)) { |
1da177e4 | 1120 | continue; |
51c6f666 | 1121 | } |
6f5e6b9e | 1122 | |
1da177e4 | 1123 | if (pte_present(ptent)) { |
ee498ed7 | 1124 | struct page *page; |
51c6f666 | 1125 | |
6aab341e | 1126 | page = vm_normal_page(vma, addr, ptent); |
1da177e4 LT |
1127 | if (unlikely(details) && page) { |
1128 | /* | |
1129 | * unmap_shared_mapping_pages() wants to | |
1130 | * invalidate cache without truncating: | |
1131 | * unmap shared but keep private pages. | |
1132 | */ | |
1133 | if (details->check_mapping && | |
1134 | details->check_mapping != page->mapping) | |
1135 | continue; | |
1da177e4 | 1136 | } |
b5810039 | 1137 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
a600388d | 1138 | tlb->fullmm); |
1da177e4 LT |
1139 | tlb_remove_tlb_entry(tlb, pte, addr); |
1140 | if (unlikely(!page)) | |
1141 | continue; | |
eca56ff9 JM |
1142 | |
1143 | if (!PageAnon(page)) { | |
1cf35d47 | 1144 | if (pte_dirty(ptent)) { |
aac45363 MH |
1145 | /* |
1146 | * oom_reaper cannot tear down dirty | |
1147 | * pages | |
1148 | */ | |
1149 | if (unlikely(details && details->ignore_dirty)) | |
1150 | continue; | |
1cf35d47 | 1151 | force_flush = 1; |
6237bcd9 | 1152 | set_page_dirty(page); |
1cf35d47 | 1153 | } |
4917e5d0 | 1154 | if (pte_young(ptent) && |
64363aad | 1155 | likely(!(vma->vm_flags & VM_SEQ_READ))) |
bf3f3bc5 | 1156 | mark_page_accessed(page); |
6237bcd9 | 1157 | } |
eca56ff9 | 1158 | rss[mm_counter(page)]--; |
d281ee61 | 1159 | page_remove_rmap(page, false); |
3dc14741 HD |
1160 | if (unlikely(page_mapcount(page) < 0)) |
1161 | print_bad_pte(vma, addr, ptent, page); | |
1cf35d47 LT |
1162 | if (unlikely(!__tlb_remove_page(tlb, page))) { |
1163 | force_flush = 1; | |
ce9ec37b | 1164 | addr += PAGE_SIZE; |
d16dfc55 | 1165 | break; |
1cf35d47 | 1166 | } |
1da177e4 LT |
1167 | continue; |
1168 | } | |
aac45363 MH |
1169 | /* only check swap_entries if explicitly asked for in details */ |
1170 | if (unlikely(details && !details->check_swap_entries)) | |
1da177e4 | 1171 | continue; |
b084d435 | 1172 | |
8a5f14a2 KS |
1173 | entry = pte_to_swp_entry(ptent); |
1174 | if (!non_swap_entry(entry)) | |
1175 | rss[MM_SWAPENTS]--; | |
1176 | else if (is_migration_entry(entry)) { | |
1177 | struct page *page; | |
9f9f1acd | 1178 | |
8a5f14a2 | 1179 | page = migration_entry_to_page(entry); |
eca56ff9 | 1180 | rss[mm_counter(page)]--; |
b084d435 | 1181 | } |
8a5f14a2 KS |
1182 | if (unlikely(!free_swap_and_cache(entry))) |
1183 | print_bad_pte(vma, addr, ptent, NULL); | |
9888a1ca | 1184 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
97a89413 | 1185 | } while (pte++, addr += PAGE_SIZE, addr != end); |
ae859762 | 1186 | |
d559db08 | 1187 | add_mm_rss_vec(mm, rss); |
6606c3e0 | 1188 | arch_leave_lazy_mmu_mode(); |
51c6f666 | 1189 | |
1cf35d47 | 1190 | /* Do the actual TLB flush before dropping ptl */ |
fb7332a9 | 1191 | if (force_flush) |
1cf35d47 | 1192 | tlb_flush_mmu_tlbonly(tlb); |
1cf35d47 LT |
1193 | pte_unmap_unlock(start_pte, ptl); |
1194 | ||
1195 | /* | |
1196 | * If we forced a TLB flush (either due to running out of | |
1197 | * batch buffers or because we needed to flush dirty TLB | |
1198 | * entries before releasing the ptl), free the batched | |
1199 | * memory too. Restart if we didn't do everything. | |
1200 | */ | |
1201 | if (force_flush) { | |
1202 | force_flush = 0; | |
1203 | tlb_flush_mmu_free(tlb); | |
2b047252 LT |
1204 | |
1205 | if (addr != end) | |
d16dfc55 PZ |
1206 | goto again; |
1207 | } | |
1208 | ||
51c6f666 | 1209 | return addr; |
1da177e4 LT |
1210 | } |
1211 | ||
51c6f666 | 1212 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
b5810039 | 1213 | struct vm_area_struct *vma, pud_t *pud, |
1da177e4 | 1214 | unsigned long addr, unsigned long end, |
97a89413 | 1215 | struct zap_details *details) |
1da177e4 LT |
1216 | { |
1217 | pmd_t *pmd; | |
1218 | unsigned long next; | |
1219 | ||
1220 | pmd = pmd_offset(pud, addr); | |
1221 | do { | |
1222 | next = pmd_addr_end(addr, end); | |
5c7fb56e | 1223 | if (pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { |
1a5a9906 | 1224 | if (next - addr != HPAGE_PMD_SIZE) { |
68428398 HD |
1225 | VM_BUG_ON_VMA(vma_is_anonymous(vma) && |
1226 | !rwsem_is_locked(&tlb->mm->mmap_sem), vma); | |
78ddc534 | 1227 | split_huge_pmd(vma, pmd, addr); |
f21760b1 | 1228 | } else if (zap_huge_pmd(tlb, vma, pmd, addr)) |
1a5a9906 | 1229 | goto next; |
71e3aac0 AA |
1230 | /* fall through */ |
1231 | } | |
1a5a9906 AA |
1232 | /* |
1233 | * Here there can be other concurrent MADV_DONTNEED or | |
1234 | * trans huge page faults running, and if the pmd is | |
1235 | * none or trans huge it can change under us. This is | |
1236 | * because MADV_DONTNEED holds the mmap_sem in read | |
1237 | * mode. | |
1238 | */ | |
1239 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) | |
1240 | goto next; | |
97a89413 | 1241 | next = zap_pte_range(tlb, vma, pmd, addr, next, details); |
1a5a9906 | 1242 | next: |
97a89413 PZ |
1243 | cond_resched(); |
1244 | } while (pmd++, addr = next, addr != end); | |
51c6f666 RH |
1245 | |
1246 | return addr; | |
1da177e4 LT |
1247 | } |
1248 | ||
51c6f666 | 1249 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
b5810039 | 1250 | struct vm_area_struct *vma, pgd_t *pgd, |
1da177e4 | 1251 | unsigned long addr, unsigned long end, |
97a89413 | 1252 | struct zap_details *details) |
1da177e4 LT |
1253 | { |
1254 | pud_t *pud; | |
1255 | unsigned long next; | |
1256 | ||
1257 | pud = pud_offset(pgd, addr); | |
1258 | do { | |
1259 | next = pud_addr_end(addr, end); | |
97a89413 | 1260 | if (pud_none_or_clear_bad(pud)) |
1da177e4 | 1261 | continue; |
97a89413 PZ |
1262 | next = zap_pmd_range(tlb, vma, pud, addr, next, details); |
1263 | } while (pud++, addr = next, addr != end); | |
51c6f666 RH |
1264 | |
1265 | return addr; | |
1da177e4 LT |
1266 | } |
1267 | ||
aac45363 | 1268 | void unmap_page_range(struct mmu_gather *tlb, |
038c7aa1 AV |
1269 | struct vm_area_struct *vma, |
1270 | unsigned long addr, unsigned long end, | |
1271 | struct zap_details *details) | |
1da177e4 LT |
1272 | { |
1273 | pgd_t *pgd; | |
1274 | unsigned long next; | |
1275 | ||
1da177e4 LT |
1276 | BUG_ON(addr >= end); |
1277 | tlb_start_vma(tlb, vma); | |
1278 | pgd = pgd_offset(vma->vm_mm, addr); | |
1279 | do { | |
1280 | next = pgd_addr_end(addr, end); | |
97a89413 | 1281 | if (pgd_none_or_clear_bad(pgd)) |
1da177e4 | 1282 | continue; |
97a89413 PZ |
1283 | next = zap_pud_range(tlb, vma, pgd, addr, next, details); |
1284 | } while (pgd++, addr = next, addr != end); | |
1da177e4 LT |
1285 | tlb_end_vma(tlb, vma); |
1286 | } | |
51c6f666 | 1287 | |
f5cc4eef AV |
1288 | |
1289 | static void unmap_single_vma(struct mmu_gather *tlb, | |
1290 | struct vm_area_struct *vma, unsigned long start_addr, | |
4f74d2c8 | 1291 | unsigned long end_addr, |
f5cc4eef AV |
1292 | struct zap_details *details) |
1293 | { | |
1294 | unsigned long start = max(vma->vm_start, start_addr); | |
1295 | unsigned long end; | |
1296 | ||
1297 | if (start >= vma->vm_end) | |
1298 | return; | |
1299 | end = min(vma->vm_end, end_addr); | |
1300 | if (end <= vma->vm_start) | |
1301 | return; | |
1302 | ||
cbc91f71 SD |
1303 | if (vma->vm_file) |
1304 | uprobe_munmap(vma, start, end); | |
1305 | ||
b3b9c293 | 1306 | if (unlikely(vma->vm_flags & VM_PFNMAP)) |
5180da41 | 1307 | untrack_pfn(vma, 0, 0); |
f5cc4eef AV |
1308 | |
1309 | if (start != end) { | |
1310 | if (unlikely(is_vm_hugetlb_page(vma))) { | |
1311 | /* | |
1312 | * It is undesirable to test vma->vm_file as it | |
1313 | * should be non-null for valid hugetlb area. | |
1314 | * However, vm_file will be NULL in the error | |
7aa6b4ad | 1315 | * cleanup path of mmap_region. When |
f5cc4eef | 1316 | * hugetlbfs ->mmap method fails, |
7aa6b4ad | 1317 | * mmap_region() nullifies vma->vm_file |
f5cc4eef AV |
1318 | * before calling this function to clean up. |
1319 | * Since no pte has actually been setup, it is | |
1320 | * safe to do nothing in this case. | |
1321 | */ | |
24669e58 | 1322 | if (vma->vm_file) { |
83cde9e8 | 1323 | i_mmap_lock_write(vma->vm_file->f_mapping); |
d833352a | 1324 | __unmap_hugepage_range_final(tlb, vma, start, end, NULL); |
83cde9e8 | 1325 | i_mmap_unlock_write(vma->vm_file->f_mapping); |
24669e58 | 1326 | } |
f5cc4eef AV |
1327 | } else |
1328 | unmap_page_range(tlb, vma, start, end, details); | |
1329 | } | |
1da177e4 LT |
1330 | } |
1331 | ||
1da177e4 LT |
1332 | /** |
1333 | * unmap_vmas - unmap a range of memory covered by a list of vma's | |
0164f69d | 1334 | * @tlb: address of the caller's struct mmu_gather |
1da177e4 LT |
1335 | * @vma: the starting vma |
1336 | * @start_addr: virtual address at which to start unmapping | |
1337 | * @end_addr: virtual address at which to end unmapping | |
1da177e4 | 1338 | * |
508034a3 | 1339 | * Unmap all pages in the vma list. |
1da177e4 | 1340 | * |
1da177e4 LT |
1341 | * Only addresses between `start' and `end' will be unmapped. |
1342 | * | |
1343 | * The VMA list must be sorted in ascending virtual address order. | |
1344 | * | |
1345 | * unmap_vmas() assumes that the caller will flush the whole unmapped address | |
1346 | * range after unmap_vmas() returns. So the only responsibility here is to | |
1347 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() | |
1348 | * drops the lock and schedules. | |
1349 | */ | |
6e8bb019 | 1350 | void unmap_vmas(struct mmu_gather *tlb, |
1da177e4 | 1351 | struct vm_area_struct *vma, unsigned long start_addr, |
4f74d2c8 | 1352 | unsigned long end_addr) |
1da177e4 | 1353 | { |
cddb8a5c | 1354 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 1355 | |
cddb8a5c | 1356 | mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); |
f5cc4eef | 1357 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) |
4f74d2c8 | 1358 | unmap_single_vma(tlb, vma, start_addr, end_addr, NULL); |
cddb8a5c | 1359 | mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); |
1da177e4 LT |
1360 | } |
1361 | ||
1362 | /** | |
1363 | * zap_page_range - remove user pages in a given range | |
1364 | * @vma: vm_area_struct holding the applicable pages | |
eb4546bb | 1365 | * @start: starting address of pages to zap |
1da177e4 | 1366 | * @size: number of bytes to zap |
8a5f14a2 | 1367 | * @details: details of shared cache invalidation |
f5cc4eef AV |
1368 | * |
1369 | * Caller must protect the VMA list | |
1da177e4 | 1370 | */ |
7e027b14 | 1371 | void zap_page_range(struct vm_area_struct *vma, unsigned long start, |
1da177e4 LT |
1372 | unsigned long size, struct zap_details *details) |
1373 | { | |
1374 | struct mm_struct *mm = vma->vm_mm; | |
d16dfc55 | 1375 | struct mmu_gather tlb; |
7e027b14 | 1376 | unsigned long end = start + size; |
1da177e4 | 1377 | |
1da177e4 | 1378 | lru_add_drain(); |
2b047252 | 1379 | tlb_gather_mmu(&tlb, mm, start, end); |
365e9c87 | 1380 | update_hiwater_rss(mm); |
7e027b14 LT |
1381 | mmu_notifier_invalidate_range_start(mm, start, end); |
1382 | for ( ; vma && vma->vm_start < end; vma = vma->vm_next) | |
4f74d2c8 | 1383 | unmap_single_vma(&tlb, vma, start, end, details); |
7e027b14 LT |
1384 | mmu_notifier_invalidate_range_end(mm, start, end); |
1385 | tlb_finish_mmu(&tlb, start, end); | |
1da177e4 LT |
1386 | } |
1387 | ||
f5cc4eef AV |
1388 | /** |
1389 | * zap_page_range_single - remove user pages in a given range | |
1390 | * @vma: vm_area_struct holding the applicable pages | |
1391 | * @address: starting address of pages to zap | |
1392 | * @size: number of bytes to zap | |
8a5f14a2 | 1393 | * @details: details of shared cache invalidation |
f5cc4eef AV |
1394 | * |
1395 | * The range must fit into one VMA. | |
1da177e4 | 1396 | */ |
f5cc4eef | 1397 | static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, |
1da177e4 LT |
1398 | unsigned long size, struct zap_details *details) |
1399 | { | |
1400 | struct mm_struct *mm = vma->vm_mm; | |
d16dfc55 | 1401 | struct mmu_gather tlb; |
1da177e4 | 1402 | unsigned long end = address + size; |
1da177e4 | 1403 | |
1da177e4 | 1404 | lru_add_drain(); |
2b047252 | 1405 | tlb_gather_mmu(&tlb, mm, address, end); |
365e9c87 | 1406 | update_hiwater_rss(mm); |
f5cc4eef | 1407 | mmu_notifier_invalidate_range_start(mm, address, end); |
4f74d2c8 | 1408 | unmap_single_vma(&tlb, vma, address, end, details); |
f5cc4eef | 1409 | mmu_notifier_invalidate_range_end(mm, address, end); |
d16dfc55 | 1410 | tlb_finish_mmu(&tlb, address, end); |
1da177e4 LT |
1411 | } |
1412 | ||
c627f9cc JS |
1413 | /** |
1414 | * zap_vma_ptes - remove ptes mapping the vma | |
1415 | * @vma: vm_area_struct holding ptes to be zapped | |
1416 | * @address: starting address of pages to zap | |
1417 | * @size: number of bytes to zap | |
1418 | * | |
1419 | * This function only unmaps ptes assigned to VM_PFNMAP vmas. | |
1420 | * | |
1421 | * The entire address range must be fully contained within the vma. | |
1422 | * | |
1423 | * Returns 0 if successful. | |
1424 | */ | |
1425 | int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, | |
1426 | unsigned long size) | |
1427 | { | |
1428 | if (address < vma->vm_start || address + size > vma->vm_end || | |
1429 | !(vma->vm_flags & VM_PFNMAP)) | |
1430 | return -1; | |
f5cc4eef | 1431 | zap_page_range_single(vma, address, size, NULL); |
c627f9cc JS |
1432 | return 0; |
1433 | } | |
1434 | EXPORT_SYMBOL_GPL(zap_vma_ptes); | |
1435 | ||
25ca1d6c | 1436 | pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, |
920c7a5d | 1437 | spinlock_t **ptl) |
c9cfcddf LT |
1438 | { |
1439 | pgd_t * pgd = pgd_offset(mm, addr); | |
1440 | pud_t * pud = pud_alloc(mm, pgd, addr); | |
1441 | if (pud) { | |
49c91fb0 | 1442 | pmd_t * pmd = pmd_alloc(mm, pud, addr); |
f66055ab AA |
1443 | if (pmd) { |
1444 | VM_BUG_ON(pmd_trans_huge(*pmd)); | |
c9cfcddf | 1445 | return pte_alloc_map_lock(mm, pmd, addr, ptl); |
f66055ab | 1446 | } |
c9cfcddf LT |
1447 | } |
1448 | return NULL; | |
1449 | } | |
1450 | ||
238f58d8 LT |
1451 | /* |
1452 | * This is the old fallback for page remapping. | |
1453 | * | |
1454 | * For historical reasons, it only allows reserved pages. Only | |
1455 | * old drivers should use this, and they needed to mark their | |
1456 | * pages reserved for the old functions anyway. | |
1457 | */ | |
423bad60 NP |
1458 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
1459 | struct page *page, pgprot_t prot) | |
238f58d8 | 1460 | { |
423bad60 | 1461 | struct mm_struct *mm = vma->vm_mm; |
238f58d8 | 1462 | int retval; |
c9cfcddf | 1463 | pte_t *pte; |
8a9f3ccd BS |
1464 | spinlock_t *ptl; |
1465 | ||
238f58d8 | 1466 | retval = -EINVAL; |
a145dd41 | 1467 | if (PageAnon(page)) |
5b4e655e | 1468 | goto out; |
238f58d8 LT |
1469 | retval = -ENOMEM; |
1470 | flush_dcache_page(page); | |
c9cfcddf | 1471 | pte = get_locked_pte(mm, addr, &ptl); |
238f58d8 | 1472 | if (!pte) |
5b4e655e | 1473 | goto out; |
238f58d8 LT |
1474 | retval = -EBUSY; |
1475 | if (!pte_none(*pte)) | |
1476 | goto out_unlock; | |
1477 | ||
1478 | /* Ok, finally just insert the thing.. */ | |
1479 | get_page(page); | |
eca56ff9 | 1480 | inc_mm_counter_fast(mm, mm_counter_file(page)); |
238f58d8 LT |
1481 | page_add_file_rmap(page); |
1482 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); | |
1483 | ||
1484 | retval = 0; | |
8a9f3ccd BS |
1485 | pte_unmap_unlock(pte, ptl); |
1486 | return retval; | |
238f58d8 LT |
1487 | out_unlock: |
1488 | pte_unmap_unlock(pte, ptl); | |
1489 | out: | |
1490 | return retval; | |
1491 | } | |
1492 | ||
bfa5bf6d REB |
1493 | /** |
1494 | * vm_insert_page - insert single page into user vma | |
1495 | * @vma: user vma to map to | |
1496 | * @addr: target user address of this page | |
1497 | * @page: source kernel page | |
1498 | * | |
a145dd41 LT |
1499 | * This allows drivers to insert individual pages they've allocated |
1500 | * into a user vma. | |
1501 | * | |
1502 | * The page has to be a nice clean _individual_ kernel allocation. | |
1503 | * If you allocate a compound page, you need to have marked it as | |
1504 | * such (__GFP_COMP), or manually just split the page up yourself | |
8dfcc9ba | 1505 | * (see split_page()). |
a145dd41 LT |
1506 | * |
1507 | * NOTE! Traditionally this was done with "remap_pfn_range()" which | |
1508 | * took an arbitrary page protection parameter. This doesn't allow | |
1509 | * that. Your vma protection will have to be set up correctly, which | |
1510 | * means that if you want a shared writable mapping, you'd better | |
1511 | * ask for a shared writable mapping! | |
1512 | * | |
1513 | * The page does not need to be reserved. | |
4b6e1e37 KK |
1514 | * |
1515 | * Usually this function is called from f_op->mmap() handler | |
1516 | * under mm->mmap_sem write-lock, so it can change vma->vm_flags. | |
1517 | * Caller must set VM_MIXEDMAP on vma if it wants to call this | |
1518 | * function from other places, for example from page-fault handler. | |
a145dd41 | 1519 | */ |
423bad60 NP |
1520 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
1521 | struct page *page) | |
a145dd41 LT |
1522 | { |
1523 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
1524 | return -EFAULT; | |
1525 | if (!page_count(page)) | |
1526 | return -EINVAL; | |
4b6e1e37 KK |
1527 | if (!(vma->vm_flags & VM_MIXEDMAP)) { |
1528 | BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem)); | |
1529 | BUG_ON(vma->vm_flags & VM_PFNMAP); | |
1530 | vma->vm_flags |= VM_MIXEDMAP; | |
1531 | } | |
423bad60 | 1532 | return insert_page(vma, addr, page, vma->vm_page_prot); |
a145dd41 | 1533 | } |
e3c3374f | 1534 | EXPORT_SYMBOL(vm_insert_page); |
a145dd41 | 1535 | |
423bad60 | 1536 | static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
01c8f1c4 | 1537 | pfn_t pfn, pgprot_t prot) |
423bad60 NP |
1538 | { |
1539 | struct mm_struct *mm = vma->vm_mm; | |
1540 | int retval; | |
1541 | pte_t *pte, entry; | |
1542 | spinlock_t *ptl; | |
1543 | ||
1544 | retval = -ENOMEM; | |
1545 | pte = get_locked_pte(mm, addr, &ptl); | |
1546 | if (!pte) | |
1547 | goto out; | |
1548 | retval = -EBUSY; | |
1549 | if (!pte_none(*pte)) | |
1550 | goto out_unlock; | |
1551 | ||
1552 | /* Ok, finally just insert the thing.. */ | |
01c8f1c4 DW |
1553 | if (pfn_t_devmap(pfn)) |
1554 | entry = pte_mkdevmap(pfn_t_pte(pfn, prot)); | |
1555 | else | |
1556 | entry = pte_mkspecial(pfn_t_pte(pfn, prot)); | |
423bad60 | 1557 | set_pte_at(mm, addr, pte, entry); |
4b3073e1 | 1558 | update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ |
423bad60 NP |
1559 | |
1560 | retval = 0; | |
1561 | out_unlock: | |
1562 | pte_unmap_unlock(pte, ptl); | |
1563 | out: | |
1564 | return retval; | |
1565 | } | |
1566 | ||
e0dc0d8f NP |
1567 | /** |
1568 | * vm_insert_pfn - insert single pfn into user vma | |
1569 | * @vma: user vma to map to | |
1570 | * @addr: target user address of this page | |
1571 | * @pfn: source kernel pfn | |
1572 | * | |
c462f179 | 1573 | * Similar to vm_insert_page, this allows drivers to insert individual pages |
e0dc0d8f NP |
1574 | * they've allocated into a user vma. Same comments apply. |
1575 | * | |
1576 | * This function should only be called from a vm_ops->fault handler, and | |
1577 | * in that case the handler should return NULL. | |
0d71d10a NP |
1578 | * |
1579 | * vma cannot be a COW mapping. | |
1580 | * | |
1581 | * As this is called only for pages that do not currently exist, we | |
1582 | * do not need to flush old virtual caches or the TLB. | |
e0dc0d8f NP |
1583 | */ |
1584 | int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, | |
423bad60 | 1585 | unsigned long pfn) |
1745cbc5 AL |
1586 | { |
1587 | return vm_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot); | |
1588 | } | |
1589 | EXPORT_SYMBOL(vm_insert_pfn); | |
1590 | ||
1591 | /** | |
1592 | * vm_insert_pfn_prot - insert single pfn into user vma with specified pgprot | |
1593 | * @vma: user vma to map to | |
1594 | * @addr: target user address of this page | |
1595 | * @pfn: source kernel pfn | |
1596 | * @pgprot: pgprot flags for the inserted page | |
1597 | * | |
1598 | * This is exactly like vm_insert_pfn, except that it allows drivers to | |
1599 | * to override pgprot on a per-page basis. | |
1600 | * | |
1601 | * This only makes sense for IO mappings, and it makes no sense for | |
1602 | * cow mappings. In general, using multiple vmas is preferable; | |
1603 | * vm_insert_pfn_prot should only be used if using multiple VMAs is | |
1604 | * impractical. | |
1605 | */ | |
1606 | int vm_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr, | |
1607 | unsigned long pfn, pgprot_t pgprot) | |
e0dc0d8f | 1608 | { |
2ab64037 | 1609 | int ret; |
7e675137 NP |
1610 | /* |
1611 | * Technically, architectures with pte_special can avoid all these | |
1612 | * restrictions (same for remap_pfn_range). However we would like | |
1613 | * consistency in testing and feature parity among all, so we should | |
1614 | * try to keep these invariants in place for everybody. | |
1615 | */ | |
b379d790 JH |
1616 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); |
1617 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == | |
1618 | (VM_PFNMAP|VM_MIXEDMAP)); | |
1619 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); | |
1620 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); | |
e0dc0d8f | 1621 | |
423bad60 NP |
1622 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1623 | return -EFAULT; | |
f25748e3 | 1624 | if (track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV))) |
2ab64037 | 1625 | return -EINVAL; |
1626 | ||
01c8f1c4 | 1627 | ret = insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot); |
2ab64037 | 1628 | |
2ab64037 | 1629 | return ret; |
423bad60 | 1630 | } |
1745cbc5 | 1631 | EXPORT_SYMBOL(vm_insert_pfn_prot); |
e0dc0d8f | 1632 | |
423bad60 | 1633 | int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, |
01c8f1c4 | 1634 | pfn_t pfn) |
423bad60 NP |
1635 | { |
1636 | BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); | |
e0dc0d8f | 1637 | |
423bad60 NP |
1638 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1639 | return -EFAULT; | |
e0dc0d8f | 1640 | |
423bad60 NP |
1641 | /* |
1642 | * If we don't have pte special, then we have to use the pfn_valid() | |
1643 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* | |
1644 | * refcount the page if pfn_valid is true (hence insert_page rather | |
62eede62 HD |
1645 | * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP |
1646 | * without pte special, it would there be refcounted as a normal page. | |
423bad60 | 1647 | */ |
03fc2da6 | 1648 | if (!HAVE_PTE_SPECIAL && !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) { |
423bad60 NP |
1649 | struct page *page; |
1650 | ||
03fc2da6 DW |
1651 | /* |
1652 | * At this point we are committed to insert_page() | |
1653 | * regardless of whether the caller specified flags that | |
1654 | * result in pfn_t_has_page() == false. | |
1655 | */ | |
1656 | page = pfn_to_page(pfn_t_to_pfn(pfn)); | |
423bad60 NP |
1657 | return insert_page(vma, addr, page, vma->vm_page_prot); |
1658 | } | |
1659 | return insert_pfn(vma, addr, pfn, vma->vm_page_prot); | |
e0dc0d8f | 1660 | } |
423bad60 | 1661 | EXPORT_SYMBOL(vm_insert_mixed); |
e0dc0d8f | 1662 | |
1da177e4 LT |
1663 | /* |
1664 | * maps a range of physical memory into the requested pages. the old | |
1665 | * mappings are removed. any references to nonexistent pages results | |
1666 | * in null mappings (currently treated as "copy-on-access") | |
1667 | */ | |
1668 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
1669 | unsigned long addr, unsigned long end, | |
1670 | unsigned long pfn, pgprot_t prot) | |
1671 | { | |
1672 | pte_t *pte; | |
c74df32c | 1673 | spinlock_t *ptl; |
1da177e4 | 1674 | |
c74df32c | 1675 | pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1da177e4 LT |
1676 | if (!pte) |
1677 | return -ENOMEM; | |
6606c3e0 | 1678 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1679 | do { |
1680 | BUG_ON(!pte_none(*pte)); | |
7e675137 | 1681 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
1da177e4 LT |
1682 | pfn++; |
1683 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 1684 | arch_leave_lazy_mmu_mode(); |
c74df32c | 1685 | pte_unmap_unlock(pte - 1, ptl); |
1da177e4 LT |
1686 | return 0; |
1687 | } | |
1688 | ||
1689 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
1690 | unsigned long addr, unsigned long end, | |
1691 | unsigned long pfn, pgprot_t prot) | |
1692 | { | |
1693 | pmd_t *pmd; | |
1694 | unsigned long next; | |
1695 | ||
1696 | pfn -= addr >> PAGE_SHIFT; | |
1697 | pmd = pmd_alloc(mm, pud, addr); | |
1698 | if (!pmd) | |
1699 | return -ENOMEM; | |
f66055ab | 1700 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
1da177e4 LT |
1701 | do { |
1702 | next = pmd_addr_end(addr, end); | |
1703 | if (remap_pte_range(mm, pmd, addr, next, | |
1704 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1705 | return -ENOMEM; | |
1706 | } while (pmd++, addr = next, addr != end); | |
1707 | return 0; | |
1708 | } | |
1709 | ||
1710 | static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, | |
1711 | unsigned long addr, unsigned long end, | |
1712 | unsigned long pfn, pgprot_t prot) | |
1713 | { | |
1714 | pud_t *pud; | |
1715 | unsigned long next; | |
1716 | ||
1717 | pfn -= addr >> PAGE_SHIFT; | |
1718 | pud = pud_alloc(mm, pgd, addr); | |
1719 | if (!pud) | |
1720 | return -ENOMEM; | |
1721 | do { | |
1722 | next = pud_addr_end(addr, end); | |
1723 | if (remap_pmd_range(mm, pud, addr, next, | |
1724 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1725 | return -ENOMEM; | |
1726 | } while (pud++, addr = next, addr != end); | |
1727 | return 0; | |
1728 | } | |
1729 | ||
bfa5bf6d REB |
1730 | /** |
1731 | * remap_pfn_range - remap kernel memory to userspace | |
1732 | * @vma: user vma to map to | |
1733 | * @addr: target user address to start at | |
1734 | * @pfn: physical address of kernel memory | |
1735 | * @size: size of map area | |
1736 | * @prot: page protection flags for this mapping | |
1737 | * | |
1738 | * Note: this is only safe if the mm semaphore is held when called. | |
1739 | */ | |
1da177e4 LT |
1740 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
1741 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
1742 | { | |
1743 | pgd_t *pgd; | |
1744 | unsigned long next; | |
2d15cab8 | 1745 | unsigned long end = addr + PAGE_ALIGN(size); |
1da177e4 | 1746 | struct mm_struct *mm = vma->vm_mm; |
d5957d2f | 1747 | unsigned long remap_pfn = pfn; |
1da177e4 LT |
1748 | int err; |
1749 | ||
1750 | /* | |
1751 | * Physically remapped pages are special. Tell the | |
1752 | * rest of the world about it: | |
1753 | * VM_IO tells people not to look at these pages | |
1754 | * (accesses can have side effects). | |
6aab341e LT |
1755 | * VM_PFNMAP tells the core MM that the base pages are just |
1756 | * raw PFN mappings, and do not have a "struct page" associated | |
1757 | * with them. | |
314e51b9 KK |
1758 | * VM_DONTEXPAND |
1759 | * Disable vma merging and expanding with mremap(). | |
1760 | * VM_DONTDUMP | |
1761 | * Omit vma from core dump, even when VM_IO turned off. | |
fb155c16 LT |
1762 | * |
1763 | * There's a horrible special case to handle copy-on-write | |
1764 | * behaviour that some programs depend on. We mark the "original" | |
1765 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". | |
b3b9c293 | 1766 | * See vm_normal_page() for details. |
1da177e4 | 1767 | */ |
b3b9c293 KK |
1768 | if (is_cow_mapping(vma->vm_flags)) { |
1769 | if (addr != vma->vm_start || end != vma->vm_end) | |
1770 | return -EINVAL; | |
fb155c16 | 1771 | vma->vm_pgoff = pfn; |
b3b9c293 KK |
1772 | } |
1773 | ||
d5957d2f | 1774 | err = track_pfn_remap(vma, &prot, remap_pfn, addr, PAGE_ALIGN(size)); |
b3b9c293 | 1775 | if (err) |
3c8bb73a | 1776 | return -EINVAL; |
fb155c16 | 1777 | |
314e51b9 | 1778 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; |
1da177e4 LT |
1779 | |
1780 | BUG_ON(addr >= end); | |
1781 | pfn -= addr >> PAGE_SHIFT; | |
1782 | pgd = pgd_offset(mm, addr); | |
1783 | flush_cache_range(vma, addr, end); | |
1da177e4 LT |
1784 | do { |
1785 | next = pgd_addr_end(addr, end); | |
1786 | err = remap_pud_range(mm, pgd, addr, next, | |
1787 | pfn + (addr >> PAGE_SHIFT), prot); | |
1788 | if (err) | |
1789 | break; | |
1790 | } while (pgd++, addr = next, addr != end); | |
2ab64037 | 1791 | |
1792 | if (err) | |
d5957d2f | 1793 | untrack_pfn(vma, remap_pfn, PAGE_ALIGN(size)); |
2ab64037 | 1794 | |
1da177e4 LT |
1795 | return err; |
1796 | } | |
1797 | EXPORT_SYMBOL(remap_pfn_range); | |
1798 | ||
b4cbb197 LT |
1799 | /** |
1800 | * vm_iomap_memory - remap memory to userspace | |
1801 | * @vma: user vma to map to | |
1802 | * @start: start of area | |
1803 | * @len: size of area | |
1804 | * | |
1805 | * This is a simplified io_remap_pfn_range() for common driver use. The | |
1806 | * driver just needs to give us the physical memory range to be mapped, | |
1807 | * we'll figure out the rest from the vma information. | |
1808 | * | |
1809 | * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get | |
1810 | * whatever write-combining details or similar. | |
1811 | */ | |
1812 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) | |
1813 | { | |
1814 | unsigned long vm_len, pfn, pages; | |
1815 | ||
1816 | /* Check that the physical memory area passed in looks valid */ | |
1817 | if (start + len < start) | |
1818 | return -EINVAL; | |
1819 | /* | |
1820 | * You *really* shouldn't map things that aren't page-aligned, | |
1821 | * but we've historically allowed it because IO memory might | |
1822 | * just have smaller alignment. | |
1823 | */ | |
1824 | len += start & ~PAGE_MASK; | |
1825 | pfn = start >> PAGE_SHIFT; | |
1826 | pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; | |
1827 | if (pfn + pages < pfn) | |
1828 | return -EINVAL; | |
1829 | ||
1830 | /* We start the mapping 'vm_pgoff' pages into the area */ | |
1831 | if (vma->vm_pgoff > pages) | |
1832 | return -EINVAL; | |
1833 | pfn += vma->vm_pgoff; | |
1834 | pages -= vma->vm_pgoff; | |
1835 | ||
1836 | /* Can we fit all of the mapping? */ | |
1837 | vm_len = vma->vm_end - vma->vm_start; | |
1838 | if (vm_len >> PAGE_SHIFT > pages) | |
1839 | return -EINVAL; | |
1840 | ||
1841 | /* Ok, let it rip */ | |
1842 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); | |
1843 | } | |
1844 | EXPORT_SYMBOL(vm_iomap_memory); | |
1845 | ||
aee16b3c JF |
1846 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
1847 | unsigned long addr, unsigned long end, | |
1848 | pte_fn_t fn, void *data) | |
1849 | { | |
1850 | pte_t *pte; | |
1851 | int err; | |
2f569afd | 1852 | pgtable_t token; |
94909914 | 1853 | spinlock_t *uninitialized_var(ptl); |
aee16b3c JF |
1854 | |
1855 | pte = (mm == &init_mm) ? | |
1856 | pte_alloc_kernel(pmd, addr) : | |
1857 | pte_alloc_map_lock(mm, pmd, addr, &ptl); | |
1858 | if (!pte) | |
1859 | return -ENOMEM; | |
1860 | ||
1861 | BUG_ON(pmd_huge(*pmd)); | |
1862 | ||
38e0edb1 JF |
1863 | arch_enter_lazy_mmu_mode(); |
1864 | ||
2f569afd | 1865 | token = pmd_pgtable(*pmd); |
aee16b3c JF |
1866 | |
1867 | do { | |
c36987e2 | 1868 | err = fn(pte++, token, addr, data); |
aee16b3c JF |
1869 | if (err) |
1870 | break; | |
c36987e2 | 1871 | } while (addr += PAGE_SIZE, addr != end); |
aee16b3c | 1872 | |
38e0edb1 JF |
1873 | arch_leave_lazy_mmu_mode(); |
1874 | ||
aee16b3c JF |
1875 | if (mm != &init_mm) |
1876 | pte_unmap_unlock(pte-1, ptl); | |
1877 | return err; | |
1878 | } | |
1879 | ||
1880 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, | |
1881 | unsigned long addr, unsigned long end, | |
1882 | pte_fn_t fn, void *data) | |
1883 | { | |
1884 | pmd_t *pmd; | |
1885 | unsigned long next; | |
1886 | int err; | |
1887 | ||
ceb86879 AK |
1888 | BUG_ON(pud_huge(*pud)); |
1889 | ||
aee16b3c JF |
1890 | pmd = pmd_alloc(mm, pud, addr); |
1891 | if (!pmd) | |
1892 | return -ENOMEM; | |
1893 | do { | |
1894 | next = pmd_addr_end(addr, end); | |
1895 | err = apply_to_pte_range(mm, pmd, addr, next, fn, data); | |
1896 | if (err) | |
1897 | break; | |
1898 | } while (pmd++, addr = next, addr != end); | |
1899 | return err; | |
1900 | } | |
1901 | ||
1902 | static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd, | |
1903 | unsigned long addr, unsigned long end, | |
1904 | pte_fn_t fn, void *data) | |
1905 | { | |
1906 | pud_t *pud; | |
1907 | unsigned long next; | |
1908 | int err; | |
1909 | ||
1910 | pud = pud_alloc(mm, pgd, addr); | |
1911 | if (!pud) | |
1912 | return -ENOMEM; | |
1913 | do { | |
1914 | next = pud_addr_end(addr, end); | |
1915 | err = apply_to_pmd_range(mm, pud, addr, next, fn, data); | |
1916 | if (err) | |
1917 | break; | |
1918 | } while (pud++, addr = next, addr != end); | |
1919 | return err; | |
1920 | } | |
1921 | ||
1922 | /* | |
1923 | * Scan a region of virtual memory, filling in page tables as necessary | |
1924 | * and calling a provided function on each leaf page table. | |
1925 | */ | |
1926 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, | |
1927 | unsigned long size, pte_fn_t fn, void *data) | |
1928 | { | |
1929 | pgd_t *pgd; | |
1930 | unsigned long next; | |
57250a5b | 1931 | unsigned long end = addr + size; |
aee16b3c JF |
1932 | int err; |
1933 | ||
9cb65bc3 MP |
1934 | if (WARN_ON(addr >= end)) |
1935 | return -EINVAL; | |
1936 | ||
aee16b3c JF |
1937 | pgd = pgd_offset(mm, addr); |
1938 | do { | |
1939 | next = pgd_addr_end(addr, end); | |
1940 | err = apply_to_pud_range(mm, pgd, addr, next, fn, data); | |
1941 | if (err) | |
1942 | break; | |
1943 | } while (pgd++, addr = next, addr != end); | |
57250a5b | 1944 | |
aee16b3c JF |
1945 | return err; |
1946 | } | |
1947 | EXPORT_SYMBOL_GPL(apply_to_page_range); | |
1948 | ||
8f4e2101 | 1949 | /* |
9b4bdd2f KS |
1950 | * handle_pte_fault chooses page fault handler according to an entry which was |
1951 | * read non-atomically. Before making any commitment, on those architectures | |
1952 | * or configurations (e.g. i386 with PAE) which might give a mix of unmatched | |
1953 | * parts, do_swap_page must check under lock before unmapping the pte and | |
1954 | * proceeding (but do_wp_page is only called after already making such a check; | |
a335b2e1 | 1955 | * and do_anonymous_page can safely check later on). |
8f4e2101 | 1956 | */ |
4c21e2f2 | 1957 | static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, |
8f4e2101 HD |
1958 | pte_t *page_table, pte_t orig_pte) |
1959 | { | |
1960 | int same = 1; | |
1961 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) | |
1962 | if (sizeof(pte_t) > sizeof(unsigned long)) { | |
4c21e2f2 HD |
1963 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
1964 | spin_lock(ptl); | |
8f4e2101 | 1965 | same = pte_same(*page_table, orig_pte); |
4c21e2f2 | 1966 | spin_unlock(ptl); |
8f4e2101 HD |
1967 | } |
1968 | #endif | |
1969 | pte_unmap(page_table); | |
1970 | return same; | |
1971 | } | |
1972 | ||
9de455b2 | 1973 | static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) |
6aab341e | 1974 | { |
0abdd7a8 DW |
1975 | debug_dma_assert_idle(src); |
1976 | ||
6aab341e LT |
1977 | /* |
1978 | * If the source page was a PFN mapping, we don't have | |
1979 | * a "struct page" for it. We do a best-effort copy by | |
1980 | * just copying from the original user address. If that | |
1981 | * fails, we just zero-fill it. Live with it. | |
1982 | */ | |
1983 | if (unlikely(!src)) { | |
9b04c5fe | 1984 | void *kaddr = kmap_atomic(dst); |
5d2a2dbb LT |
1985 | void __user *uaddr = (void __user *)(va & PAGE_MASK); |
1986 | ||
1987 | /* | |
1988 | * This really shouldn't fail, because the page is there | |
1989 | * in the page tables. But it might just be unreadable, | |
1990 | * in which case we just give up and fill the result with | |
1991 | * zeroes. | |
1992 | */ | |
1993 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) | |
3ecb01df | 1994 | clear_page(kaddr); |
9b04c5fe | 1995 | kunmap_atomic(kaddr); |
c4ec7b0d | 1996 | flush_dcache_page(dst); |
0ed361de NP |
1997 | } else |
1998 | copy_user_highpage(dst, src, va, vma); | |
6aab341e LT |
1999 | } |
2000 | ||
c20cd45e MH |
2001 | static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma) |
2002 | { | |
2003 | struct file *vm_file = vma->vm_file; | |
2004 | ||
2005 | if (vm_file) | |
2006 | return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO; | |
2007 | ||
2008 | /* | |
2009 | * Special mappings (e.g. VDSO) do not have any file so fake | |
2010 | * a default GFP_KERNEL for them. | |
2011 | */ | |
2012 | return GFP_KERNEL; | |
2013 | } | |
2014 | ||
fb09a464 KS |
2015 | /* |
2016 | * Notify the address space that the page is about to become writable so that | |
2017 | * it can prohibit this or wait for the page to get into an appropriate state. | |
2018 | * | |
2019 | * We do this without the lock held, so that it can sleep if it needs to. | |
2020 | */ | |
2021 | static int do_page_mkwrite(struct vm_area_struct *vma, struct page *page, | |
2022 | unsigned long address) | |
2023 | { | |
2024 | struct vm_fault vmf; | |
2025 | int ret; | |
2026 | ||
2027 | vmf.virtual_address = (void __user *)(address & PAGE_MASK); | |
2028 | vmf.pgoff = page->index; | |
2029 | vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; | |
c20cd45e | 2030 | vmf.gfp_mask = __get_fault_gfp_mask(vma); |
fb09a464 | 2031 | vmf.page = page; |
2e4cdab0 | 2032 | vmf.cow_page = NULL; |
fb09a464 KS |
2033 | |
2034 | ret = vma->vm_ops->page_mkwrite(vma, &vmf); | |
2035 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) | |
2036 | return ret; | |
2037 | if (unlikely(!(ret & VM_FAULT_LOCKED))) { | |
2038 | lock_page(page); | |
2039 | if (!page->mapping) { | |
2040 | unlock_page(page); | |
2041 | return 0; /* retry */ | |
2042 | } | |
2043 | ret |= VM_FAULT_LOCKED; | |
2044 | } else | |
2045 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
2046 | return ret; | |
2047 | } | |
2048 | ||
4e047f89 SR |
2049 | /* |
2050 | * Handle write page faults for pages that can be reused in the current vma | |
2051 | * | |
2052 | * This can happen either due to the mapping being with the VM_SHARED flag, | |
2053 | * or due to us being the last reference standing to the page. In either | |
2054 | * case, all we need to do here is to mark the page as writable and update | |
2055 | * any related book-keeping. | |
2056 | */ | |
2057 | static inline int wp_page_reuse(struct mm_struct *mm, | |
2058 | struct vm_area_struct *vma, unsigned long address, | |
2059 | pte_t *page_table, spinlock_t *ptl, pte_t orig_pte, | |
2060 | struct page *page, int page_mkwrite, | |
2061 | int dirty_shared) | |
2062 | __releases(ptl) | |
2063 | { | |
2064 | pte_t entry; | |
2065 | /* | |
2066 | * Clear the pages cpupid information as the existing | |
2067 | * information potentially belongs to a now completely | |
2068 | * unrelated process. | |
2069 | */ | |
2070 | if (page) | |
2071 | page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1); | |
2072 | ||
2073 | flush_cache_page(vma, address, pte_pfn(orig_pte)); | |
2074 | entry = pte_mkyoung(orig_pte); | |
2075 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
2076 | if (ptep_set_access_flags(vma, address, page_table, entry, 1)) | |
2077 | update_mmu_cache(vma, address, page_table); | |
2078 | pte_unmap_unlock(page_table, ptl); | |
2079 | ||
2080 | if (dirty_shared) { | |
2081 | struct address_space *mapping; | |
2082 | int dirtied; | |
2083 | ||
2084 | if (!page_mkwrite) | |
2085 | lock_page(page); | |
2086 | ||
2087 | dirtied = set_page_dirty(page); | |
2088 | VM_BUG_ON_PAGE(PageAnon(page), page); | |
2089 | mapping = page->mapping; | |
2090 | unlock_page(page); | |
09cbfeaf | 2091 | put_page(page); |
4e047f89 SR |
2092 | |
2093 | if ((dirtied || page_mkwrite) && mapping) { | |
2094 | /* | |
2095 | * Some device drivers do not set page.mapping | |
2096 | * but still dirty their pages | |
2097 | */ | |
2098 | balance_dirty_pages_ratelimited(mapping); | |
2099 | } | |
2100 | ||
2101 | if (!page_mkwrite) | |
2102 | file_update_time(vma->vm_file); | |
2103 | } | |
2104 | ||
2105 | return VM_FAULT_WRITE; | |
2106 | } | |
2107 | ||
2f38ab2c SR |
2108 | /* |
2109 | * Handle the case of a page which we actually need to copy to a new page. | |
2110 | * | |
2111 | * Called with mmap_sem locked and the old page referenced, but | |
2112 | * without the ptl held. | |
2113 | * | |
2114 | * High level logic flow: | |
2115 | * | |
2116 | * - Allocate a page, copy the content of the old page to the new one. | |
2117 | * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. | |
2118 | * - Take the PTL. If the pte changed, bail out and release the allocated page | |
2119 | * - If the pte is still the way we remember it, update the page table and all | |
2120 | * relevant references. This includes dropping the reference the page-table | |
2121 | * held to the old page, as well as updating the rmap. | |
2122 | * - In any case, unlock the PTL and drop the reference we took to the old page. | |
2123 | */ | |
2124 | static int wp_page_copy(struct mm_struct *mm, struct vm_area_struct *vma, | |
2125 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
2126 | pte_t orig_pte, struct page *old_page) | |
2127 | { | |
2128 | struct page *new_page = NULL; | |
2129 | spinlock_t *ptl = NULL; | |
2130 | pte_t entry; | |
2131 | int page_copied = 0; | |
2132 | const unsigned long mmun_start = address & PAGE_MASK; /* For mmu_notifiers */ | |
2133 | const unsigned long mmun_end = mmun_start + PAGE_SIZE; /* For mmu_notifiers */ | |
2134 | struct mem_cgroup *memcg; | |
2135 | ||
2136 | if (unlikely(anon_vma_prepare(vma))) | |
2137 | goto oom; | |
2138 | ||
2139 | if (is_zero_pfn(pte_pfn(orig_pte))) { | |
2140 | new_page = alloc_zeroed_user_highpage_movable(vma, address); | |
2141 | if (!new_page) | |
2142 | goto oom; | |
2143 | } else { | |
2144 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); | |
2145 | if (!new_page) | |
2146 | goto oom; | |
2147 | cow_user_page(new_page, old_page, address, vma); | |
2148 | } | |
2f38ab2c | 2149 | |
f627c2f5 | 2150 | if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg, false)) |
2f38ab2c SR |
2151 | goto oom_free_new; |
2152 | ||
eb3c24f3 MG |
2153 | __SetPageUptodate(new_page); |
2154 | ||
2f38ab2c SR |
2155 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
2156 | ||
2157 | /* | |
2158 | * Re-check the pte - we dropped the lock | |
2159 | */ | |
2160 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); | |
2161 | if (likely(pte_same(*page_table, orig_pte))) { | |
2162 | if (old_page) { | |
2163 | if (!PageAnon(old_page)) { | |
eca56ff9 JM |
2164 | dec_mm_counter_fast(mm, |
2165 | mm_counter_file(old_page)); | |
2f38ab2c SR |
2166 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
2167 | } | |
2168 | } else { | |
2169 | inc_mm_counter_fast(mm, MM_ANONPAGES); | |
2170 | } | |
2171 | flush_cache_page(vma, address, pte_pfn(orig_pte)); | |
2172 | entry = mk_pte(new_page, vma->vm_page_prot); | |
2173 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
2174 | /* | |
2175 | * Clear the pte entry and flush it first, before updating the | |
2176 | * pte with the new entry. This will avoid a race condition | |
2177 | * seen in the presence of one thread doing SMC and another | |
2178 | * thread doing COW. | |
2179 | */ | |
2180 | ptep_clear_flush_notify(vma, address, page_table); | |
d281ee61 | 2181 | page_add_new_anon_rmap(new_page, vma, address, false); |
f627c2f5 | 2182 | mem_cgroup_commit_charge(new_page, memcg, false, false); |
2f38ab2c SR |
2183 | lru_cache_add_active_or_unevictable(new_page, vma); |
2184 | /* | |
2185 | * We call the notify macro here because, when using secondary | |
2186 | * mmu page tables (such as kvm shadow page tables), we want the | |
2187 | * new page to be mapped directly into the secondary page table. | |
2188 | */ | |
2189 | set_pte_at_notify(mm, address, page_table, entry); | |
2190 | update_mmu_cache(vma, address, page_table); | |
2191 | if (old_page) { | |
2192 | /* | |
2193 | * Only after switching the pte to the new page may | |
2194 | * we remove the mapcount here. Otherwise another | |
2195 | * process may come and find the rmap count decremented | |
2196 | * before the pte is switched to the new page, and | |
2197 | * "reuse" the old page writing into it while our pte | |
2198 | * here still points into it and can be read by other | |
2199 | * threads. | |
2200 | * | |
2201 | * The critical issue is to order this | |
2202 | * page_remove_rmap with the ptp_clear_flush above. | |
2203 | * Those stores are ordered by (if nothing else,) | |
2204 | * the barrier present in the atomic_add_negative | |
2205 | * in page_remove_rmap. | |
2206 | * | |
2207 | * Then the TLB flush in ptep_clear_flush ensures that | |
2208 | * no process can access the old page before the | |
2209 | * decremented mapcount is visible. And the old page | |
2210 | * cannot be reused until after the decremented | |
2211 | * mapcount is visible. So transitively, TLBs to | |
2212 | * old page will be flushed before it can be reused. | |
2213 | */ | |
d281ee61 | 2214 | page_remove_rmap(old_page, false); |
2f38ab2c SR |
2215 | } |
2216 | ||
2217 | /* Free the old page.. */ | |
2218 | new_page = old_page; | |
2219 | page_copied = 1; | |
2220 | } else { | |
f627c2f5 | 2221 | mem_cgroup_cancel_charge(new_page, memcg, false); |
2f38ab2c SR |
2222 | } |
2223 | ||
2224 | if (new_page) | |
09cbfeaf | 2225 | put_page(new_page); |
2f38ab2c SR |
2226 | |
2227 | pte_unmap_unlock(page_table, ptl); | |
2228 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); | |
2229 | if (old_page) { | |
2230 | /* | |
2231 | * Don't let another task, with possibly unlocked vma, | |
2232 | * keep the mlocked page. | |
2233 | */ | |
2234 | if (page_copied && (vma->vm_flags & VM_LOCKED)) { | |
2235 | lock_page(old_page); /* LRU manipulation */ | |
e90309c9 KS |
2236 | if (PageMlocked(old_page)) |
2237 | munlock_vma_page(old_page); | |
2f38ab2c SR |
2238 | unlock_page(old_page); |
2239 | } | |
09cbfeaf | 2240 | put_page(old_page); |
2f38ab2c SR |
2241 | } |
2242 | return page_copied ? VM_FAULT_WRITE : 0; | |
2243 | oom_free_new: | |
09cbfeaf | 2244 | put_page(new_page); |
2f38ab2c SR |
2245 | oom: |
2246 | if (old_page) | |
09cbfeaf | 2247 | put_page(old_page); |
2f38ab2c SR |
2248 | return VM_FAULT_OOM; |
2249 | } | |
2250 | ||
dd906184 BH |
2251 | /* |
2252 | * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED | |
2253 | * mapping | |
2254 | */ | |
2255 | static int wp_pfn_shared(struct mm_struct *mm, | |
2256 | struct vm_area_struct *vma, unsigned long address, | |
2257 | pte_t *page_table, spinlock_t *ptl, pte_t orig_pte, | |
2258 | pmd_t *pmd) | |
2259 | { | |
2260 | if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { | |
2261 | struct vm_fault vmf = { | |
2262 | .page = NULL, | |
2263 | .pgoff = linear_page_index(vma, address), | |
2264 | .virtual_address = (void __user *)(address & PAGE_MASK), | |
2265 | .flags = FAULT_FLAG_WRITE | FAULT_FLAG_MKWRITE, | |
2266 | }; | |
2267 | int ret; | |
2268 | ||
2269 | pte_unmap_unlock(page_table, ptl); | |
2270 | ret = vma->vm_ops->pfn_mkwrite(vma, &vmf); | |
2271 | if (ret & VM_FAULT_ERROR) | |
2272 | return ret; | |
2273 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); | |
2274 | /* | |
2275 | * We might have raced with another page fault while we | |
2276 | * released the pte_offset_map_lock. | |
2277 | */ | |
2278 | if (!pte_same(*page_table, orig_pte)) { | |
2279 | pte_unmap_unlock(page_table, ptl); | |
2280 | return 0; | |
2281 | } | |
2282 | } | |
2283 | return wp_page_reuse(mm, vma, address, page_table, ptl, orig_pte, | |
2284 | NULL, 0, 0); | |
2285 | } | |
2286 | ||
93e478d4 SR |
2287 | static int wp_page_shared(struct mm_struct *mm, struct vm_area_struct *vma, |
2288 | unsigned long address, pte_t *page_table, | |
2289 | pmd_t *pmd, spinlock_t *ptl, pte_t orig_pte, | |
2290 | struct page *old_page) | |
2291 | __releases(ptl) | |
2292 | { | |
2293 | int page_mkwrite = 0; | |
2294 | ||
09cbfeaf | 2295 | get_page(old_page); |
93e478d4 | 2296 | |
93e478d4 SR |
2297 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
2298 | int tmp; | |
2299 | ||
2300 | pte_unmap_unlock(page_table, ptl); | |
2301 | tmp = do_page_mkwrite(vma, old_page, address); | |
2302 | if (unlikely(!tmp || (tmp & | |
2303 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
09cbfeaf | 2304 | put_page(old_page); |
93e478d4 SR |
2305 | return tmp; |
2306 | } | |
2307 | /* | |
2308 | * Since we dropped the lock we need to revalidate | |
2309 | * the PTE as someone else may have changed it. If | |
2310 | * they did, we just return, as we can count on the | |
2311 | * MMU to tell us if they didn't also make it writable. | |
2312 | */ | |
2313 | page_table = pte_offset_map_lock(mm, pmd, address, | |
2314 | &ptl); | |
2315 | if (!pte_same(*page_table, orig_pte)) { | |
2316 | unlock_page(old_page); | |
2317 | pte_unmap_unlock(page_table, ptl); | |
09cbfeaf | 2318 | put_page(old_page); |
93e478d4 SR |
2319 | return 0; |
2320 | } | |
2321 | page_mkwrite = 1; | |
2322 | } | |
2323 | ||
2324 | return wp_page_reuse(mm, vma, address, page_table, ptl, | |
2325 | orig_pte, old_page, page_mkwrite, 1); | |
2326 | } | |
2327 | ||
1da177e4 LT |
2328 | /* |
2329 | * This routine handles present pages, when users try to write | |
2330 | * to a shared page. It is done by copying the page to a new address | |
2331 | * and decrementing the shared-page counter for the old page. | |
2332 | * | |
1da177e4 LT |
2333 | * Note that this routine assumes that the protection checks have been |
2334 | * done by the caller (the low-level page fault routine in most cases). | |
2335 | * Thus we can safely just mark it writable once we've done any necessary | |
2336 | * COW. | |
2337 | * | |
2338 | * We also mark the page dirty at this point even though the page will | |
2339 | * change only once the write actually happens. This avoids a few races, | |
2340 | * and potentially makes it more efficient. | |
2341 | * | |
8f4e2101 HD |
2342 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2343 | * but allow concurrent faults), with pte both mapped and locked. | |
2344 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2345 | */ |
65500d23 HD |
2346 | static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2347 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
8f4e2101 | 2348 | spinlock_t *ptl, pte_t orig_pte) |
e6219ec8 | 2349 | __releases(ptl) |
1da177e4 | 2350 | { |
2f38ab2c | 2351 | struct page *old_page; |
1da177e4 | 2352 | |
6aab341e | 2353 | old_page = vm_normal_page(vma, address, orig_pte); |
251b97f5 PZ |
2354 | if (!old_page) { |
2355 | /* | |
64e45507 PF |
2356 | * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a |
2357 | * VM_PFNMAP VMA. | |
251b97f5 PZ |
2358 | * |
2359 | * We should not cow pages in a shared writeable mapping. | |
dd906184 | 2360 | * Just mark the pages writable and/or call ops->pfn_mkwrite. |
251b97f5 PZ |
2361 | */ |
2362 | if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == | |
2363 | (VM_WRITE|VM_SHARED)) | |
dd906184 BH |
2364 | return wp_pfn_shared(mm, vma, address, page_table, ptl, |
2365 | orig_pte, pmd); | |
2f38ab2c SR |
2366 | |
2367 | pte_unmap_unlock(page_table, ptl); | |
2368 | return wp_page_copy(mm, vma, address, page_table, pmd, | |
2369 | orig_pte, old_page); | |
251b97f5 | 2370 | } |
1da177e4 | 2371 | |
d08b3851 | 2372 | /* |
ee6a6457 PZ |
2373 | * Take out anonymous pages first, anonymous shared vmas are |
2374 | * not dirty accountable. | |
d08b3851 | 2375 | */ |
9a840895 | 2376 | if (PageAnon(old_page) && !PageKsm(old_page)) { |
6d0a07ed | 2377 | int total_mapcount; |
ab967d86 | 2378 | if (!trylock_page(old_page)) { |
09cbfeaf | 2379 | get_page(old_page); |
ab967d86 HD |
2380 | pte_unmap_unlock(page_table, ptl); |
2381 | lock_page(old_page); | |
2382 | page_table = pte_offset_map_lock(mm, pmd, address, | |
2383 | &ptl); | |
2384 | if (!pte_same(*page_table, orig_pte)) { | |
2385 | unlock_page(old_page); | |
28766805 | 2386 | pte_unmap_unlock(page_table, ptl); |
09cbfeaf | 2387 | put_page(old_page); |
28766805 | 2388 | return 0; |
ab967d86 | 2389 | } |
09cbfeaf | 2390 | put_page(old_page); |
ee6a6457 | 2391 | } |
6d0a07ed AA |
2392 | if (reuse_swap_page(old_page, &total_mapcount)) { |
2393 | if (total_mapcount == 1) { | |
2394 | /* | |
2395 | * The page is all ours. Move it to | |
2396 | * our anon_vma so the rmap code will | |
2397 | * not search our parent or siblings. | |
2398 | * Protected against the rmap code by | |
2399 | * the page lock. | |
2400 | */ | |
2401 | page_move_anon_rmap(compound_head(old_page), | |
2402 | vma, address); | |
2403 | } | |
b009c024 | 2404 | unlock_page(old_page); |
4e047f89 SR |
2405 | return wp_page_reuse(mm, vma, address, page_table, ptl, |
2406 | orig_pte, old_page, 0, 0); | |
b009c024 | 2407 | } |
ab967d86 | 2408 | unlock_page(old_page); |
ee6a6457 | 2409 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
d08b3851 | 2410 | (VM_WRITE|VM_SHARED))) { |
93e478d4 SR |
2411 | return wp_page_shared(mm, vma, address, page_table, pmd, |
2412 | ptl, orig_pte, old_page); | |
1da177e4 | 2413 | } |
1da177e4 LT |
2414 | |
2415 | /* | |
2416 | * Ok, we need to copy. Oh, well.. | |
2417 | */ | |
09cbfeaf | 2418 | get_page(old_page); |
28766805 | 2419 | |
8f4e2101 | 2420 | pte_unmap_unlock(page_table, ptl); |
2f38ab2c SR |
2421 | return wp_page_copy(mm, vma, address, page_table, pmd, |
2422 | orig_pte, old_page); | |
1da177e4 LT |
2423 | } |
2424 | ||
97a89413 | 2425 | static void unmap_mapping_range_vma(struct vm_area_struct *vma, |
1da177e4 LT |
2426 | unsigned long start_addr, unsigned long end_addr, |
2427 | struct zap_details *details) | |
2428 | { | |
f5cc4eef | 2429 | zap_page_range_single(vma, start_addr, end_addr - start_addr, details); |
1da177e4 LT |
2430 | } |
2431 | ||
6b2dbba8 | 2432 | static inline void unmap_mapping_range_tree(struct rb_root *root, |
1da177e4 LT |
2433 | struct zap_details *details) |
2434 | { | |
2435 | struct vm_area_struct *vma; | |
1da177e4 LT |
2436 | pgoff_t vba, vea, zba, zea; |
2437 | ||
6b2dbba8 | 2438 | vma_interval_tree_foreach(vma, root, |
1da177e4 | 2439 | details->first_index, details->last_index) { |
1da177e4 LT |
2440 | |
2441 | vba = vma->vm_pgoff; | |
d6e93217 | 2442 | vea = vba + vma_pages(vma) - 1; |
1da177e4 LT |
2443 | zba = details->first_index; |
2444 | if (zba < vba) | |
2445 | zba = vba; | |
2446 | zea = details->last_index; | |
2447 | if (zea > vea) | |
2448 | zea = vea; | |
2449 | ||
97a89413 | 2450 | unmap_mapping_range_vma(vma, |
1da177e4 LT |
2451 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, |
2452 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, | |
97a89413 | 2453 | details); |
1da177e4 LT |
2454 | } |
2455 | } | |
2456 | ||
1da177e4 | 2457 | /** |
8a5f14a2 KS |
2458 | * unmap_mapping_range - unmap the portion of all mmaps in the specified |
2459 | * address_space corresponding to the specified page range in the underlying | |
2460 | * file. | |
2461 | * | |
3d41088f | 2462 | * @mapping: the address space containing mmaps to be unmapped. |
1da177e4 LT |
2463 | * @holebegin: byte in first page to unmap, relative to the start of |
2464 | * the underlying file. This will be rounded down to a PAGE_SIZE | |
25d9e2d1 | 2465 | * boundary. Note that this is different from truncate_pagecache(), which |
1da177e4 LT |
2466 | * must keep the partial page. In contrast, we must get rid of |
2467 | * partial pages. | |
2468 | * @holelen: size of prospective hole in bytes. This will be rounded | |
2469 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the | |
2470 | * end of the file. | |
2471 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; | |
2472 | * but 0 when invalidating pagecache, don't throw away private data. | |
2473 | */ | |
2474 | void unmap_mapping_range(struct address_space *mapping, | |
2475 | loff_t const holebegin, loff_t const holelen, int even_cows) | |
2476 | { | |
aac45363 | 2477 | struct zap_details details = { }; |
1da177e4 LT |
2478 | pgoff_t hba = holebegin >> PAGE_SHIFT; |
2479 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2480 | ||
2481 | /* Check for overflow. */ | |
2482 | if (sizeof(holelen) > sizeof(hlen)) { | |
2483 | long long holeend = | |
2484 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2485 | if (holeend & ~(long long)ULONG_MAX) | |
2486 | hlen = ULONG_MAX - hba + 1; | |
2487 | } | |
2488 | ||
2489 | details.check_mapping = even_cows? NULL: mapping; | |
1da177e4 LT |
2490 | details.first_index = hba; |
2491 | details.last_index = hba + hlen - 1; | |
2492 | if (details.last_index < details.first_index) | |
2493 | details.last_index = ULONG_MAX; | |
1da177e4 | 2494 | |
0f90cc66 RZ |
2495 | |
2496 | /* DAX uses i_mmap_lock to serialise file truncate vs page fault */ | |
46c043ed | 2497 | i_mmap_lock_write(mapping); |
6b2dbba8 | 2498 | if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap))) |
1da177e4 | 2499 | unmap_mapping_range_tree(&mapping->i_mmap, &details); |
46c043ed | 2500 | i_mmap_unlock_write(mapping); |
1da177e4 LT |
2501 | } |
2502 | EXPORT_SYMBOL(unmap_mapping_range); | |
2503 | ||
1da177e4 | 2504 | /* |
8f4e2101 HD |
2505 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2506 | * but allow concurrent faults), and pte mapped but not yet locked. | |
9a95f3cf PC |
2507 | * We return with pte unmapped and unlocked. |
2508 | * | |
2509 | * We return with the mmap_sem locked or unlocked in the same cases | |
2510 | * as does filemap_fault(). | |
1da177e4 | 2511 | */ |
65500d23 HD |
2512 | static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2513 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
30c9f3a9 | 2514 | unsigned int flags, pte_t orig_pte) |
1da177e4 | 2515 | { |
8f4e2101 | 2516 | spinlock_t *ptl; |
56f31801 | 2517 | struct page *page, *swapcache; |
00501b53 | 2518 | struct mem_cgroup *memcg; |
65500d23 | 2519 | swp_entry_t entry; |
1da177e4 | 2520 | pte_t pte; |
d065bd81 | 2521 | int locked; |
ad8c2ee8 | 2522 | int exclusive = 0; |
83c54070 | 2523 | int ret = 0; |
1da177e4 | 2524 | |
4c21e2f2 | 2525 | if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) |
8f4e2101 | 2526 | goto out; |
65500d23 HD |
2527 | |
2528 | entry = pte_to_swp_entry(orig_pte); | |
d1737fdb AK |
2529 | if (unlikely(non_swap_entry(entry))) { |
2530 | if (is_migration_entry(entry)) { | |
2531 | migration_entry_wait(mm, pmd, address); | |
2532 | } else if (is_hwpoison_entry(entry)) { | |
2533 | ret = VM_FAULT_HWPOISON; | |
2534 | } else { | |
2535 | print_bad_pte(vma, address, orig_pte, NULL); | |
d99be1a8 | 2536 | ret = VM_FAULT_SIGBUS; |
d1737fdb | 2537 | } |
0697212a CL |
2538 | goto out; |
2539 | } | |
0ff92245 | 2540 | delayacct_set_flag(DELAYACCT_PF_SWAPIN); |
1da177e4 LT |
2541 | page = lookup_swap_cache(entry); |
2542 | if (!page) { | |
02098fea HD |
2543 | page = swapin_readahead(entry, |
2544 | GFP_HIGHUSER_MOVABLE, vma, address); | |
1da177e4 LT |
2545 | if (!page) { |
2546 | /* | |
8f4e2101 HD |
2547 | * Back out if somebody else faulted in this pte |
2548 | * while we released the pte lock. | |
1da177e4 | 2549 | */ |
8f4e2101 | 2550 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
1da177e4 LT |
2551 | if (likely(pte_same(*page_table, orig_pte))) |
2552 | ret = VM_FAULT_OOM; | |
0ff92245 | 2553 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
65500d23 | 2554 | goto unlock; |
1da177e4 LT |
2555 | } |
2556 | ||
2557 | /* Had to read the page from swap area: Major fault */ | |
2558 | ret = VM_FAULT_MAJOR; | |
f8891e5e | 2559 | count_vm_event(PGMAJFAULT); |
456f998e | 2560 | mem_cgroup_count_vm_event(mm, PGMAJFAULT); |
d1737fdb | 2561 | } else if (PageHWPoison(page)) { |
71f72525 WF |
2562 | /* |
2563 | * hwpoisoned dirty swapcache pages are kept for killing | |
2564 | * owner processes (which may be unknown at hwpoison time) | |
2565 | */ | |
d1737fdb AK |
2566 | ret = VM_FAULT_HWPOISON; |
2567 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); | |
56f31801 | 2568 | swapcache = page; |
4779cb31 | 2569 | goto out_release; |
1da177e4 LT |
2570 | } |
2571 | ||
56f31801 | 2572 | swapcache = page; |
d065bd81 | 2573 | locked = lock_page_or_retry(page, mm, flags); |
e709ffd6 | 2574 | |
073e587e | 2575 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
d065bd81 ML |
2576 | if (!locked) { |
2577 | ret |= VM_FAULT_RETRY; | |
2578 | goto out_release; | |
2579 | } | |
073e587e | 2580 | |
4969c119 | 2581 | /* |
31c4a3d3 HD |
2582 | * Make sure try_to_free_swap or reuse_swap_page or swapoff did not |
2583 | * release the swapcache from under us. The page pin, and pte_same | |
2584 | * test below, are not enough to exclude that. Even if it is still | |
2585 | * swapcache, we need to check that the page's swap has not changed. | |
4969c119 | 2586 | */ |
31c4a3d3 | 2587 | if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val)) |
4969c119 AA |
2588 | goto out_page; |
2589 | ||
cbf86cfe HD |
2590 | page = ksm_might_need_to_copy(page, vma, address); |
2591 | if (unlikely(!page)) { | |
2592 | ret = VM_FAULT_OOM; | |
2593 | page = swapcache; | |
cbf86cfe | 2594 | goto out_page; |
5ad64688 HD |
2595 | } |
2596 | ||
f627c2f5 | 2597 | if (mem_cgroup_try_charge(page, mm, GFP_KERNEL, &memcg, false)) { |
8a9f3ccd | 2598 | ret = VM_FAULT_OOM; |
bc43f75c | 2599 | goto out_page; |
8a9f3ccd BS |
2600 | } |
2601 | ||
1da177e4 | 2602 | /* |
8f4e2101 | 2603 | * Back out if somebody else already faulted in this pte. |
1da177e4 | 2604 | */ |
8f4e2101 | 2605 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
9e9bef07 | 2606 | if (unlikely(!pte_same(*page_table, orig_pte))) |
b8107480 | 2607 | goto out_nomap; |
b8107480 KK |
2608 | |
2609 | if (unlikely(!PageUptodate(page))) { | |
2610 | ret = VM_FAULT_SIGBUS; | |
2611 | goto out_nomap; | |
1da177e4 LT |
2612 | } |
2613 | ||
8c7c6e34 KH |
2614 | /* |
2615 | * The page isn't present yet, go ahead with the fault. | |
2616 | * | |
2617 | * Be careful about the sequence of operations here. | |
2618 | * To get its accounting right, reuse_swap_page() must be called | |
2619 | * while the page is counted on swap but not yet in mapcount i.e. | |
2620 | * before page_add_anon_rmap() and swap_free(); try_to_free_swap() | |
2621 | * must be called after the swap_free(), or it will never succeed. | |
8c7c6e34 | 2622 | */ |
1da177e4 | 2623 | |
34e55232 | 2624 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
b084d435 | 2625 | dec_mm_counter_fast(mm, MM_SWAPENTS); |
1da177e4 | 2626 | pte = mk_pte(page, vma->vm_page_prot); |
6d0a07ed | 2627 | if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page, NULL)) { |
1da177e4 | 2628 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); |
30c9f3a9 | 2629 | flags &= ~FAULT_FLAG_WRITE; |
9a5b489b | 2630 | ret |= VM_FAULT_WRITE; |
d281ee61 | 2631 | exclusive = RMAP_EXCLUSIVE; |
1da177e4 | 2632 | } |
1da177e4 | 2633 | flush_icache_page(vma, page); |
179ef71c CG |
2634 | if (pte_swp_soft_dirty(orig_pte)) |
2635 | pte = pte_mksoft_dirty(pte); | |
1da177e4 | 2636 | set_pte_at(mm, address, page_table, pte); |
00501b53 | 2637 | if (page == swapcache) { |
af34770e | 2638 | do_page_add_anon_rmap(page, vma, address, exclusive); |
f627c2f5 | 2639 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 2640 | } else { /* ksm created a completely new copy */ |
d281ee61 | 2641 | page_add_new_anon_rmap(page, vma, address, false); |
f627c2f5 | 2642 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 JW |
2643 | lru_cache_add_active_or_unevictable(page, vma); |
2644 | } | |
1da177e4 | 2645 | |
c475a8ab | 2646 | swap_free(entry); |
5ccc5aba VD |
2647 | if (mem_cgroup_swap_full(page) || |
2648 | (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) | |
a2c43eed | 2649 | try_to_free_swap(page); |
c475a8ab | 2650 | unlock_page(page); |
56f31801 | 2651 | if (page != swapcache) { |
4969c119 AA |
2652 | /* |
2653 | * Hold the lock to avoid the swap entry to be reused | |
2654 | * until we take the PT lock for the pte_same() check | |
2655 | * (to avoid false positives from pte_same). For | |
2656 | * further safety release the lock after the swap_free | |
2657 | * so that the swap count won't change under a | |
2658 | * parallel locked swapcache. | |
2659 | */ | |
2660 | unlock_page(swapcache); | |
09cbfeaf | 2661 | put_page(swapcache); |
4969c119 | 2662 | } |
c475a8ab | 2663 | |
30c9f3a9 | 2664 | if (flags & FAULT_FLAG_WRITE) { |
61469f1d HD |
2665 | ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte); |
2666 | if (ret & VM_FAULT_ERROR) | |
2667 | ret &= VM_FAULT_ERROR; | |
1da177e4 LT |
2668 | goto out; |
2669 | } | |
2670 | ||
2671 | /* No need to invalidate - it was non-present before */ | |
4b3073e1 | 2672 | update_mmu_cache(vma, address, page_table); |
65500d23 | 2673 | unlock: |
8f4e2101 | 2674 | pte_unmap_unlock(page_table, ptl); |
1da177e4 LT |
2675 | out: |
2676 | return ret; | |
b8107480 | 2677 | out_nomap: |
f627c2f5 | 2678 | mem_cgroup_cancel_charge(page, memcg, false); |
8f4e2101 | 2679 | pte_unmap_unlock(page_table, ptl); |
bc43f75c | 2680 | out_page: |
b8107480 | 2681 | unlock_page(page); |
4779cb31 | 2682 | out_release: |
09cbfeaf | 2683 | put_page(page); |
56f31801 | 2684 | if (page != swapcache) { |
4969c119 | 2685 | unlock_page(swapcache); |
09cbfeaf | 2686 | put_page(swapcache); |
4969c119 | 2687 | } |
65500d23 | 2688 | return ret; |
1da177e4 LT |
2689 | } |
2690 | ||
320b2b8d | 2691 | /* |
8ca3eb08 LT |
2692 | * This is like a special single-page "expand_{down|up}wards()", |
2693 | * except we must first make sure that 'address{-|+}PAGE_SIZE' | |
320b2b8d | 2694 | * doesn't hit another vma. |
320b2b8d LT |
2695 | */ |
2696 | static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address) | |
2697 | { | |
2698 | address &= PAGE_MASK; | |
2699 | if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) { | |
0e8e50e2 LT |
2700 | struct vm_area_struct *prev = vma->vm_prev; |
2701 | ||
2702 | /* | |
2703 | * Is there a mapping abutting this one below? | |
2704 | * | |
2705 | * That's only ok if it's the same stack mapping | |
2706 | * that has gotten split.. | |
2707 | */ | |
2708 | if (prev && prev->vm_end == address) | |
2709 | return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM; | |
320b2b8d | 2710 | |
fee7e49d | 2711 | return expand_downwards(vma, address - PAGE_SIZE); |
320b2b8d | 2712 | } |
8ca3eb08 LT |
2713 | if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) { |
2714 | struct vm_area_struct *next = vma->vm_next; | |
2715 | ||
2716 | /* As VM_GROWSDOWN but s/below/above/ */ | |
2717 | if (next && next->vm_start == address + PAGE_SIZE) | |
2718 | return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM; | |
2719 | ||
fee7e49d | 2720 | return expand_upwards(vma, address + PAGE_SIZE); |
8ca3eb08 | 2721 | } |
320b2b8d LT |
2722 | return 0; |
2723 | } | |
2724 | ||
1da177e4 | 2725 | /* |
8f4e2101 HD |
2726 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2727 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2728 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2729 | */ |
65500d23 HD |
2730 | static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2731 | unsigned long address, pte_t *page_table, pmd_t *pmd, | |
30c9f3a9 | 2732 | unsigned int flags) |
1da177e4 | 2733 | { |
00501b53 | 2734 | struct mem_cgroup *memcg; |
8f4e2101 HD |
2735 | struct page *page; |
2736 | spinlock_t *ptl; | |
1da177e4 | 2737 | pte_t entry; |
1da177e4 | 2738 | |
11ac5524 LT |
2739 | pte_unmap(page_table); |
2740 | ||
6b7339f4 KS |
2741 | /* File mapping without ->vm_ops ? */ |
2742 | if (vma->vm_flags & VM_SHARED) | |
2743 | return VM_FAULT_SIGBUS; | |
2744 | ||
11ac5524 LT |
2745 | /* Check if we need to add a guard page to the stack */ |
2746 | if (check_stack_guard_page(vma, address) < 0) | |
9c145c56 | 2747 | return VM_FAULT_SIGSEGV; |
320b2b8d | 2748 | |
11ac5524 | 2749 | /* Use the zero-page for reads */ |
593befa6 | 2750 | if (!(flags & FAULT_FLAG_WRITE) && !mm_forbids_zeropage(mm)) { |
62eede62 HD |
2751 | entry = pte_mkspecial(pfn_pte(my_zero_pfn(address), |
2752 | vma->vm_page_prot)); | |
11ac5524 | 2753 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
a13ea5b7 HD |
2754 | if (!pte_none(*page_table)) |
2755 | goto unlock; | |
6b251fc9 AA |
2756 | /* Deliver the page fault to userland, check inside PT lock */ |
2757 | if (userfaultfd_missing(vma)) { | |
2758 | pte_unmap_unlock(page_table, ptl); | |
2759 | return handle_userfault(vma, address, flags, | |
2760 | VM_UFFD_MISSING); | |
2761 | } | |
a13ea5b7 HD |
2762 | goto setpte; |
2763 | } | |
2764 | ||
557ed1fa | 2765 | /* Allocate our own private page. */ |
557ed1fa NP |
2766 | if (unlikely(anon_vma_prepare(vma))) |
2767 | goto oom; | |
2768 | page = alloc_zeroed_user_highpage_movable(vma, address); | |
2769 | if (!page) | |
2770 | goto oom; | |
eb3c24f3 | 2771 | |
f627c2f5 | 2772 | if (mem_cgroup_try_charge(page, mm, GFP_KERNEL, &memcg, false)) |
eb3c24f3 MG |
2773 | goto oom_free_page; |
2774 | ||
52f37629 MK |
2775 | /* |
2776 | * The memory barrier inside __SetPageUptodate makes sure that | |
2777 | * preceeding stores to the page contents become visible before | |
2778 | * the set_pte_at() write. | |
2779 | */ | |
0ed361de | 2780 | __SetPageUptodate(page); |
8f4e2101 | 2781 | |
557ed1fa | 2782 | entry = mk_pte(page, vma->vm_page_prot); |
1ac0cb5d HD |
2783 | if (vma->vm_flags & VM_WRITE) |
2784 | entry = pte_mkwrite(pte_mkdirty(entry)); | |
1da177e4 | 2785 | |
557ed1fa | 2786 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
1c2fb7a4 | 2787 | if (!pte_none(*page_table)) |
557ed1fa | 2788 | goto release; |
9ba69294 | 2789 | |
6b251fc9 AA |
2790 | /* Deliver the page fault to userland, check inside PT lock */ |
2791 | if (userfaultfd_missing(vma)) { | |
2792 | pte_unmap_unlock(page_table, ptl); | |
f627c2f5 | 2793 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 2794 | put_page(page); |
6b251fc9 AA |
2795 | return handle_userfault(vma, address, flags, |
2796 | VM_UFFD_MISSING); | |
2797 | } | |
2798 | ||
34e55232 | 2799 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
d281ee61 | 2800 | page_add_new_anon_rmap(page, vma, address, false); |
f627c2f5 | 2801 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 | 2802 | lru_cache_add_active_or_unevictable(page, vma); |
a13ea5b7 | 2803 | setpte: |
65500d23 | 2804 | set_pte_at(mm, address, page_table, entry); |
1da177e4 LT |
2805 | |
2806 | /* No need to invalidate - it was non-present before */ | |
4b3073e1 | 2807 | update_mmu_cache(vma, address, page_table); |
65500d23 | 2808 | unlock: |
8f4e2101 | 2809 | pte_unmap_unlock(page_table, ptl); |
83c54070 | 2810 | return 0; |
8f4e2101 | 2811 | release: |
f627c2f5 | 2812 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 2813 | put_page(page); |
8f4e2101 | 2814 | goto unlock; |
8a9f3ccd | 2815 | oom_free_page: |
09cbfeaf | 2816 | put_page(page); |
65500d23 | 2817 | oom: |
1da177e4 LT |
2818 | return VM_FAULT_OOM; |
2819 | } | |
2820 | ||
9a95f3cf PC |
2821 | /* |
2822 | * The mmap_sem must have been held on entry, and may have been | |
2823 | * released depending on flags and vma->vm_ops->fault() return value. | |
2824 | * See filemap_fault() and __lock_page_retry(). | |
2825 | */ | |
7eae74af | 2826 | static int __do_fault(struct vm_area_struct *vma, unsigned long address, |
2e4cdab0 MW |
2827 | pgoff_t pgoff, unsigned int flags, |
2828 | struct page *cow_page, struct page **page) | |
7eae74af KS |
2829 | { |
2830 | struct vm_fault vmf; | |
2831 | int ret; | |
2832 | ||
2833 | vmf.virtual_address = (void __user *)(address & PAGE_MASK); | |
2834 | vmf.pgoff = pgoff; | |
2835 | vmf.flags = flags; | |
2836 | vmf.page = NULL; | |
c20cd45e | 2837 | vmf.gfp_mask = __get_fault_gfp_mask(vma); |
2e4cdab0 | 2838 | vmf.cow_page = cow_page; |
7eae74af KS |
2839 | |
2840 | ret = vma->vm_ops->fault(vma, &vmf); | |
2841 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) | |
2842 | return ret; | |
2e4cdab0 MW |
2843 | if (!vmf.page) |
2844 | goto out; | |
7eae74af KS |
2845 | |
2846 | if (unlikely(PageHWPoison(vmf.page))) { | |
2847 | if (ret & VM_FAULT_LOCKED) | |
2848 | unlock_page(vmf.page); | |
09cbfeaf | 2849 | put_page(vmf.page); |
7eae74af KS |
2850 | return VM_FAULT_HWPOISON; |
2851 | } | |
2852 | ||
2853 | if (unlikely(!(ret & VM_FAULT_LOCKED))) | |
2854 | lock_page(vmf.page); | |
2855 | else | |
2856 | VM_BUG_ON_PAGE(!PageLocked(vmf.page), vmf.page); | |
2857 | ||
2e4cdab0 | 2858 | out: |
7eae74af KS |
2859 | *page = vmf.page; |
2860 | return ret; | |
2861 | } | |
2862 | ||
8c6e50b0 KS |
2863 | /** |
2864 | * do_set_pte - setup new PTE entry for given page and add reverse page mapping. | |
2865 | * | |
2866 | * @vma: virtual memory area | |
2867 | * @address: user virtual address | |
2868 | * @page: page to map | |
2869 | * @pte: pointer to target page table entry | |
2870 | * @write: true, if new entry is writable | |
2871 | * @anon: true, if it's anonymous page | |
2872 | * | |
2873 | * Caller must hold page table lock relevant for @pte. | |
2874 | * | |
2875 | * Target users are page handler itself and implementations of | |
2876 | * vm_ops->map_pages. | |
2877 | */ | |
2878 | void do_set_pte(struct vm_area_struct *vma, unsigned long address, | |
5c0a85fa | 2879 | struct page *page, pte_t *pte, bool write, bool anon, bool old) |
3bb97794 KS |
2880 | { |
2881 | pte_t entry; | |
2882 | ||
2883 | flush_icache_page(vma, page); | |
2884 | entry = mk_pte(page, vma->vm_page_prot); | |
2885 | if (write) | |
2886 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
5c0a85fa KS |
2887 | if (old) |
2888 | entry = pte_mkold(entry); | |
3bb97794 KS |
2889 | if (anon) { |
2890 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); | |
d281ee61 | 2891 | page_add_new_anon_rmap(page, vma, address, false); |
3bb97794 | 2892 | } else { |
eca56ff9 | 2893 | inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page)); |
3bb97794 KS |
2894 | page_add_file_rmap(page); |
2895 | } | |
2896 | set_pte_at(vma->vm_mm, address, pte, entry); | |
2897 | ||
2898 | /* no need to invalidate: a not-present page won't be cached */ | |
2899 | update_mmu_cache(vma, address, pte); | |
2900 | } | |
2901 | ||
d0834a6c MK |
2902 | /* |
2903 | * If architecture emulates "accessed" or "young" bit without HW support, | |
2904 | * there is no much gain with fault_around. | |
2905 | */ | |
3a91053a | 2906 | static unsigned long fault_around_bytes __read_mostly = |
d0834a6c MK |
2907 | #ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS |
2908 | PAGE_SIZE; | |
2909 | #else | |
3a91053a | 2910 | rounddown_pow_of_two(65536); |
d0834a6c | 2911 | #endif |
a9b0f861 | 2912 | |
a9b0f861 KS |
2913 | #ifdef CONFIG_DEBUG_FS |
2914 | static int fault_around_bytes_get(void *data, u64 *val) | |
1592eef0 | 2915 | { |
a9b0f861 | 2916 | *val = fault_around_bytes; |
1592eef0 KS |
2917 | return 0; |
2918 | } | |
2919 | ||
b4903d6e AR |
2920 | /* |
2921 | * fault_around_pages() and fault_around_mask() expects fault_around_bytes | |
2922 | * rounded down to nearest page order. It's what do_fault_around() expects to | |
2923 | * see. | |
2924 | */ | |
a9b0f861 | 2925 | static int fault_around_bytes_set(void *data, u64 val) |
1592eef0 | 2926 | { |
a9b0f861 | 2927 | if (val / PAGE_SIZE > PTRS_PER_PTE) |
1592eef0 | 2928 | return -EINVAL; |
b4903d6e AR |
2929 | if (val > PAGE_SIZE) |
2930 | fault_around_bytes = rounddown_pow_of_two(val); | |
2931 | else | |
2932 | fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */ | |
1592eef0 KS |
2933 | return 0; |
2934 | } | |
a9b0f861 KS |
2935 | DEFINE_SIMPLE_ATTRIBUTE(fault_around_bytes_fops, |
2936 | fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); | |
1592eef0 KS |
2937 | |
2938 | static int __init fault_around_debugfs(void) | |
2939 | { | |
2940 | void *ret; | |
2941 | ||
a9b0f861 KS |
2942 | ret = debugfs_create_file("fault_around_bytes", 0644, NULL, NULL, |
2943 | &fault_around_bytes_fops); | |
1592eef0 | 2944 | if (!ret) |
a9b0f861 | 2945 | pr_warn("Failed to create fault_around_bytes in debugfs"); |
1592eef0 KS |
2946 | return 0; |
2947 | } | |
2948 | late_initcall(fault_around_debugfs); | |
1592eef0 | 2949 | #endif |
8c6e50b0 | 2950 | |
1fdb412b KS |
2951 | /* |
2952 | * do_fault_around() tries to map few pages around the fault address. The hope | |
2953 | * is that the pages will be needed soon and this will lower the number of | |
2954 | * faults to handle. | |
2955 | * | |
2956 | * It uses vm_ops->map_pages() to map the pages, which skips the page if it's | |
2957 | * not ready to be mapped: not up-to-date, locked, etc. | |
2958 | * | |
2959 | * This function is called with the page table lock taken. In the split ptlock | |
2960 | * case the page table lock only protects only those entries which belong to | |
2961 | * the page table corresponding to the fault address. | |
2962 | * | |
2963 | * This function doesn't cross the VMA boundaries, in order to call map_pages() | |
2964 | * only once. | |
2965 | * | |
2966 | * fault_around_pages() defines how many pages we'll try to map. | |
2967 | * do_fault_around() expects it to return a power of two less than or equal to | |
2968 | * PTRS_PER_PTE. | |
2969 | * | |
2970 | * The virtual address of the area that we map is naturally aligned to the | |
2971 | * fault_around_pages() value (and therefore to page order). This way it's | |
2972 | * easier to guarantee that we don't cross page table boundaries. | |
2973 | */ | |
8c6e50b0 KS |
2974 | static void do_fault_around(struct vm_area_struct *vma, unsigned long address, |
2975 | pte_t *pte, pgoff_t pgoff, unsigned int flags) | |
2976 | { | |
aecd6f44 | 2977 | unsigned long start_addr, nr_pages, mask; |
8c6e50b0 KS |
2978 | pgoff_t max_pgoff; |
2979 | struct vm_fault vmf; | |
2980 | int off; | |
2981 | ||
4db0c3c2 | 2982 | nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT; |
aecd6f44 KS |
2983 | mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK; |
2984 | ||
2985 | start_addr = max(address & mask, vma->vm_start); | |
8c6e50b0 KS |
2986 | off = ((address - start_addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1); |
2987 | pte -= off; | |
2988 | pgoff -= off; | |
2989 | ||
2990 | /* | |
2991 | * max_pgoff is either end of page table or end of vma | |
850e9c69 | 2992 | * or fault_around_pages() from pgoff, depending what is nearest. |
8c6e50b0 KS |
2993 | */ |
2994 | max_pgoff = pgoff - ((start_addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) + | |
2995 | PTRS_PER_PTE - 1; | |
2996 | max_pgoff = min3(max_pgoff, vma_pages(vma) + vma->vm_pgoff - 1, | |
aecd6f44 | 2997 | pgoff + nr_pages - 1); |
8c6e50b0 KS |
2998 | |
2999 | /* Check if it makes any sense to call ->map_pages */ | |
3000 | while (!pte_none(*pte)) { | |
3001 | if (++pgoff > max_pgoff) | |
3002 | return; | |
3003 | start_addr += PAGE_SIZE; | |
3004 | if (start_addr >= vma->vm_end) | |
3005 | return; | |
3006 | pte++; | |
3007 | } | |
3008 | ||
3009 | vmf.virtual_address = (void __user *) start_addr; | |
3010 | vmf.pte = pte; | |
3011 | vmf.pgoff = pgoff; | |
3012 | vmf.max_pgoff = max_pgoff; | |
3013 | vmf.flags = flags; | |
c20cd45e | 3014 | vmf.gfp_mask = __get_fault_gfp_mask(vma); |
8c6e50b0 KS |
3015 | vma->vm_ops->map_pages(vma, &vmf); |
3016 | } | |
3017 | ||
e655fb29 KS |
3018 | static int do_read_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
3019 | unsigned long address, pmd_t *pmd, | |
3020 | pgoff_t pgoff, unsigned int flags, pte_t orig_pte) | |
3021 | { | |
3022 | struct page *fault_page; | |
3023 | spinlock_t *ptl; | |
3bb97794 | 3024 | pte_t *pte; |
8c6e50b0 KS |
3025 | int ret = 0; |
3026 | ||
3027 | /* | |
3028 | * Let's call ->map_pages() first and use ->fault() as fallback | |
3029 | * if page by the offset is not ready to be mapped (cold cache or | |
3030 | * something). | |
3031 | */ | |
9b4bdd2f | 3032 | if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) { |
8c6e50b0 | 3033 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); |
8c6e50b0 KS |
3034 | if (!pte_same(*pte, orig_pte)) |
3035 | goto unlock_out; | |
5c0a85fa KS |
3036 | do_fault_around(vma, address, pte, pgoff, flags); |
3037 | /* Check if the fault is handled by faultaround */ | |
3038 | if (!pte_same(*pte, orig_pte)) { | |
3039 | /* | |
3040 | * Faultaround produce old pte, but the pte we've | |
3041 | * handler fault for should be young. | |
3042 | */ | |
3043 | pte_t entry = pte_mkyoung(*pte); | |
3044 | if (ptep_set_access_flags(vma, address, pte, entry, 0)) | |
3045 | update_mmu_cache(vma, address, pte); | |
3046 | goto unlock_out; | |
3047 | } | |
8c6e50b0 KS |
3048 | pte_unmap_unlock(pte, ptl); |
3049 | } | |
e655fb29 | 3050 | |
2e4cdab0 | 3051 | ret = __do_fault(vma, address, pgoff, flags, NULL, &fault_page); |
e655fb29 KS |
3052 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3053 | return ret; | |
3054 | ||
3055 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); | |
3056 | if (unlikely(!pte_same(*pte, orig_pte))) { | |
3057 | pte_unmap_unlock(pte, ptl); | |
3058 | unlock_page(fault_page); | |
09cbfeaf | 3059 | put_page(fault_page); |
e655fb29 KS |
3060 | return ret; |
3061 | } | |
5c0a85fa | 3062 | do_set_pte(vma, address, fault_page, pte, false, false, false); |
e655fb29 | 3063 | unlock_page(fault_page); |
8c6e50b0 KS |
3064 | unlock_out: |
3065 | pte_unmap_unlock(pte, ptl); | |
e655fb29 KS |
3066 | return ret; |
3067 | } | |
3068 | ||
ec47c3b9 KS |
3069 | static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
3070 | unsigned long address, pmd_t *pmd, | |
3071 | pgoff_t pgoff, unsigned int flags, pte_t orig_pte) | |
3072 | { | |
3073 | struct page *fault_page, *new_page; | |
00501b53 | 3074 | struct mem_cgroup *memcg; |
ec47c3b9 | 3075 | spinlock_t *ptl; |
3bb97794 | 3076 | pte_t *pte; |
ec47c3b9 KS |
3077 | int ret; |
3078 | ||
3079 | if (unlikely(anon_vma_prepare(vma))) | |
3080 | return VM_FAULT_OOM; | |
3081 | ||
3082 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); | |
3083 | if (!new_page) | |
3084 | return VM_FAULT_OOM; | |
3085 | ||
f627c2f5 | 3086 | if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg, false)) { |
09cbfeaf | 3087 | put_page(new_page); |
ec47c3b9 KS |
3088 | return VM_FAULT_OOM; |
3089 | } | |
3090 | ||
2e4cdab0 | 3091 | ret = __do_fault(vma, address, pgoff, flags, new_page, &fault_page); |
ec47c3b9 KS |
3092 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3093 | goto uncharge_out; | |
3094 | ||
2e4cdab0 MW |
3095 | if (fault_page) |
3096 | copy_user_highpage(new_page, fault_page, address, vma); | |
ec47c3b9 KS |
3097 | __SetPageUptodate(new_page); |
3098 | ||
3099 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); | |
3100 | if (unlikely(!pte_same(*pte, orig_pte))) { | |
3101 | pte_unmap_unlock(pte, ptl); | |
2e4cdab0 MW |
3102 | if (fault_page) { |
3103 | unlock_page(fault_page); | |
09cbfeaf | 3104 | put_page(fault_page); |
2e4cdab0 MW |
3105 | } else { |
3106 | /* | |
3107 | * The fault handler has no page to lock, so it holds | |
0df9d41a | 3108 | * i_mmap_lock for read to protect against truncate. |
2e4cdab0 | 3109 | */ |
0df9d41a | 3110 | i_mmap_unlock_read(vma->vm_file->f_mapping); |
2e4cdab0 | 3111 | } |
ec47c3b9 KS |
3112 | goto uncharge_out; |
3113 | } | |
5c0a85fa | 3114 | do_set_pte(vma, address, new_page, pte, true, true, false); |
f627c2f5 | 3115 | mem_cgroup_commit_charge(new_page, memcg, false, false); |
00501b53 | 3116 | lru_cache_add_active_or_unevictable(new_page, vma); |
ec47c3b9 | 3117 | pte_unmap_unlock(pte, ptl); |
2e4cdab0 MW |
3118 | if (fault_page) { |
3119 | unlock_page(fault_page); | |
09cbfeaf | 3120 | put_page(fault_page); |
2e4cdab0 MW |
3121 | } else { |
3122 | /* | |
3123 | * The fault handler has no page to lock, so it holds | |
0df9d41a | 3124 | * i_mmap_lock for read to protect against truncate. |
2e4cdab0 | 3125 | */ |
0df9d41a | 3126 | i_mmap_unlock_read(vma->vm_file->f_mapping); |
2e4cdab0 | 3127 | } |
ec47c3b9 KS |
3128 | return ret; |
3129 | uncharge_out: | |
f627c2f5 | 3130 | mem_cgroup_cancel_charge(new_page, memcg, false); |
09cbfeaf | 3131 | put_page(new_page); |
ec47c3b9 KS |
3132 | return ret; |
3133 | } | |
3134 | ||
f0c6d4d2 | 3135 | static int do_shared_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
16abfa08 | 3136 | unsigned long address, pmd_t *pmd, |
54cb8821 | 3137 | pgoff_t pgoff, unsigned int flags, pte_t orig_pte) |
1da177e4 | 3138 | { |
f0c6d4d2 KS |
3139 | struct page *fault_page; |
3140 | struct address_space *mapping; | |
8f4e2101 | 3141 | spinlock_t *ptl; |
3bb97794 | 3142 | pte_t *pte; |
f0c6d4d2 | 3143 | int dirtied = 0; |
f0c6d4d2 | 3144 | int ret, tmp; |
1d65f86d | 3145 | |
2e4cdab0 | 3146 | ret = __do_fault(vma, address, pgoff, flags, NULL, &fault_page); |
7eae74af | 3147 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
f0c6d4d2 | 3148 | return ret; |
1da177e4 LT |
3149 | |
3150 | /* | |
f0c6d4d2 KS |
3151 | * Check if the backing address space wants to know that the page is |
3152 | * about to become writable | |
1da177e4 | 3153 | */ |
fb09a464 KS |
3154 | if (vma->vm_ops->page_mkwrite) { |
3155 | unlock_page(fault_page); | |
3156 | tmp = do_page_mkwrite(vma, fault_page, address); | |
3157 | if (unlikely(!tmp || | |
3158 | (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
09cbfeaf | 3159 | put_page(fault_page); |
fb09a464 | 3160 | return tmp; |
4294621f | 3161 | } |
fb09a464 KS |
3162 | } |
3163 | ||
f0c6d4d2 KS |
3164 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); |
3165 | if (unlikely(!pte_same(*pte, orig_pte))) { | |
3166 | pte_unmap_unlock(pte, ptl); | |
3167 | unlock_page(fault_page); | |
09cbfeaf | 3168 | put_page(fault_page); |
f0c6d4d2 | 3169 | return ret; |
1da177e4 | 3170 | } |
5c0a85fa | 3171 | do_set_pte(vma, address, fault_page, pte, true, false, false); |
f0c6d4d2 | 3172 | pte_unmap_unlock(pte, ptl); |
b827e496 | 3173 | |
f0c6d4d2 KS |
3174 | if (set_page_dirty(fault_page)) |
3175 | dirtied = 1; | |
d82fa87d AM |
3176 | /* |
3177 | * Take a local copy of the address_space - page.mapping may be zeroed | |
3178 | * by truncate after unlock_page(). The address_space itself remains | |
3179 | * pinned by vma->vm_file's reference. We rely on unlock_page()'s | |
3180 | * release semantics to prevent the compiler from undoing this copying. | |
3181 | */ | |
1c290f64 | 3182 | mapping = page_rmapping(fault_page); |
f0c6d4d2 KS |
3183 | unlock_page(fault_page); |
3184 | if ((dirtied || vma->vm_ops->page_mkwrite) && mapping) { | |
3185 | /* | |
3186 | * Some device drivers do not set page.mapping but still | |
3187 | * dirty their pages | |
3188 | */ | |
3189 | balance_dirty_pages_ratelimited(mapping); | |
d08b3851 | 3190 | } |
d00806b1 | 3191 | |
74ec6751 | 3192 | if (!vma->vm_ops->page_mkwrite) |
f0c6d4d2 | 3193 | file_update_time(vma->vm_file); |
b827e496 | 3194 | |
1d65f86d | 3195 | return ret; |
54cb8821 | 3196 | } |
d00806b1 | 3197 | |
9a95f3cf PC |
3198 | /* |
3199 | * We enter with non-exclusive mmap_sem (to exclude vma changes, | |
3200 | * but allow concurrent faults). | |
3201 | * The mmap_sem may have been released depending on flags and our | |
3202 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
3203 | */ | |
9b4bdd2f | 3204 | static int do_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
54cb8821 | 3205 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
30c9f3a9 | 3206 | unsigned int flags, pte_t orig_pte) |
54cb8821 | 3207 | { |
88193f7c | 3208 | pgoff_t pgoff = linear_page_index(vma, address); |
54cb8821 | 3209 | |
16abfa08 | 3210 | pte_unmap(page_table); |
6b7339f4 KS |
3211 | /* The VMA was not fully populated on mmap() or missing VM_DONTEXPAND */ |
3212 | if (!vma->vm_ops->fault) | |
3213 | return VM_FAULT_SIGBUS; | |
e655fb29 KS |
3214 | if (!(flags & FAULT_FLAG_WRITE)) |
3215 | return do_read_fault(mm, vma, address, pmd, pgoff, flags, | |
3216 | orig_pte); | |
ec47c3b9 KS |
3217 | if (!(vma->vm_flags & VM_SHARED)) |
3218 | return do_cow_fault(mm, vma, address, pmd, pgoff, flags, | |
3219 | orig_pte); | |
f0c6d4d2 | 3220 | return do_shared_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); |
54cb8821 NP |
3221 | } |
3222 | ||
b19a9939 | 3223 | static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, |
04bb2f94 RR |
3224 | unsigned long addr, int page_nid, |
3225 | int *flags) | |
9532fec1 MG |
3226 | { |
3227 | get_page(page); | |
3228 | ||
3229 | count_vm_numa_event(NUMA_HINT_FAULTS); | |
04bb2f94 | 3230 | if (page_nid == numa_node_id()) { |
9532fec1 | 3231 | count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); |
04bb2f94 RR |
3232 | *flags |= TNF_FAULT_LOCAL; |
3233 | } | |
9532fec1 MG |
3234 | |
3235 | return mpol_misplaced(page, vma, addr); | |
3236 | } | |
3237 | ||
b19a9939 | 3238 | static int do_numa_page(struct mm_struct *mm, struct vm_area_struct *vma, |
d10e63f2 MG |
3239 | unsigned long addr, pte_t pte, pte_t *ptep, pmd_t *pmd) |
3240 | { | |
4daae3b4 | 3241 | struct page *page = NULL; |
d10e63f2 | 3242 | spinlock_t *ptl; |
8191acbd | 3243 | int page_nid = -1; |
90572890 | 3244 | int last_cpupid; |
cbee9f88 | 3245 | int target_nid; |
b8593bfd | 3246 | bool migrated = false; |
b191f9b1 | 3247 | bool was_writable = pte_write(pte); |
6688cc05 | 3248 | int flags = 0; |
d10e63f2 | 3249 | |
c0e7cad9 MG |
3250 | /* A PROT_NONE fault should not end up here */ |
3251 | BUG_ON(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))); | |
3252 | ||
d10e63f2 MG |
3253 | /* |
3254 | * The "pte" at this point cannot be used safely without | |
3255 | * validation through pte_unmap_same(). It's of NUMA type but | |
3256 | * the pfn may be screwed if the read is non atomic. | |
3257 | * | |
4d942466 MG |
3258 | * We can safely just do a "set_pte_at()", because the old |
3259 | * page table entry is not accessible, so there would be no | |
3260 | * concurrent hardware modifications to the PTE. | |
d10e63f2 MG |
3261 | */ |
3262 | ptl = pte_lockptr(mm, pmd); | |
3263 | spin_lock(ptl); | |
4daae3b4 MG |
3264 | if (unlikely(!pte_same(*ptep, pte))) { |
3265 | pte_unmap_unlock(ptep, ptl); | |
3266 | goto out; | |
3267 | } | |
3268 | ||
4d942466 MG |
3269 | /* Make it present again */ |
3270 | pte = pte_modify(pte, vma->vm_page_prot); | |
3271 | pte = pte_mkyoung(pte); | |
b191f9b1 MG |
3272 | if (was_writable) |
3273 | pte = pte_mkwrite(pte); | |
d10e63f2 MG |
3274 | set_pte_at(mm, addr, ptep, pte); |
3275 | update_mmu_cache(vma, addr, ptep); | |
3276 | ||
3277 | page = vm_normal_page(vma, addr, pte); | |
3278 | if (!page) { | |
3279 | pte_unmap_unlock(ptep, ptl); | |
3280 | return 0; | |
3281 | } | |
3282 | ||
e81c4802 KS |
3283 | /* TODO: handle PTE-mapped THP */ |
3284 | if (PageCompound(page)) { | |
3285 | pte_unmap_unlock(ptep, ptl); | |
3286 | return 0; | |
3287 | } | |
3288 | ||
6688cc05 | 3289 | /* |
bea66fbd MG |
3290 | * Avoid grouping on RO pages in general. RO pages shouldn't hurt as |
3291 | * much anyway since they can be in shared cache state. This misses | |
3292 | * the case where a mapping is writable but the process never writes | |
3293 | * to it but pte_write gets cleared during protection updates and | |
3294 | * pte_dirty has unpredictable behaviour between PTE scan updates, | |
3295 | * background writeback, dirty balancing and application behaviour. | |
6688cc05 | 3296 | */ |
bea66fbd | 3297 | if (!(vma->vm_flags & VM_WRITE)) |
6688cc05 PZ |
3298 | flags |= TNF_NO_GROUP; |
3299 | ||
dabe1d99 RR |
3300 | /* |
3301 | * Flag if the page is shared between multiple address spaces. This | |
3302 | * is later used when determining whether to group tasks together | |
3303 | */ | |
3304 | if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) | |
3305 | flags |= TNF_SHARED; | |
3306 | ||
90572890 | 3307 | last_cpupid = page_cpupid_last(page); |
8191acbd | 3308 | page_nid = page_to_nid(page); |
04bb2f94 | 3309 | target_nid = numa_migrate_prep(page, vma, addr, page_nid, &flags); |
d10e63f2 | 3310 | pte_unmap_unlock(ptep, ptl); |
4daae3b4 | 3311 | if (target_nid == -1) { |
4daae3b4 MG |
3312 | put_page(page); |
3313 | goto out; | |
3314 | } | |
3315 | ||
3316 | /* Migrate to the requested node */ | |
1bc115d8 | 3317 | migrated = migrate_misplaced_page(page, vma, target_nid); |
6688cc05 | 3318 | if (migrated) { |
8191acbd | 3319 | page_nid = target_nid; |
6688cc05 | 3320 | flags |= TNF_MIGRATED; |
074c2381 MG |
3321 | } else |
3322 | flags |= TNF_MIGRATE_FAIL; | |
4daae3b4 MG |
3323 | |
3324 | out: | |
8191acbd | 3325 | if (page_nid != -1) |
6688cc05 | 3326 | task_numa_fault(last_cpupid, page_nid, 1, flags); |
d10e63f2 MG |
3327 | return 0; |
3328 | } | |
3329 | ||
b96375f7 MW |
3330 | static int create_huge_pmd(struct mm_struct *mm, struct vm_area_struct *vma, |
3331 | unsigned long address, pmd_t *pmd, unsigned int flags) | |
3332 | { | |
fb6dd5fa | 3333 | if (vma_is_anonymous(vma)) |
b96375f7 MW |
3334 | return do_huge_pmd_anonymous_page(mm, vma, address, pmd, flags); |
3335 | if (vma->vm_ops->pmd_fault) | |
3336 | return vma->vm_ops->pmd_fault(vma, address, pmd, flags); | |
3337 | return VM_FAULT_FALLBACK; | |
3338 | } | |
3339 | ||
3340 | static int wp_huge_pmd(struct mm_struct *mm, struct vm_area_struct *vma, | |
3341 | unsigned long address, pmd_t *pmd, pmd_t orig_pmd, | |
3342 | unsigned int flags) | |
3343 | { | |
fb6dd5fa | 3344 | if (vma_is_anonymous(vma)) |
b96375f7 MW |
3345 | return do_huge_pmd_wp_page(mm, vma, address, pmd, orig_pmd); |
3346 | if (vma->vm_ops->pmd_fault) | |
3347 | return vma->vm_ops->pmd_fault(vma, address, pmd, flags); | |
3348 | return VM_FAULT_FALLBACK; | |
3349 | } | |
3350 | ||
1da177e4 LT |
3351 | /* |
3352 | * These routines also need to handle stuff like marking pages dirty | |
3353 | * and/or accessed for architectures that don't do it in hardware (most | |
3354 | * RISC architectures). The early dirtying is also good on the i386. | |
3355 | * | |
3356 | * There is also a hook called "update_mmu_cache()" that architectures | |
3357 | * with external mmu caches can use to update those (ie the Sparc or | |
3358 | * PowerPC hashed page tables that act as extended TLBs). | |
3359 | * | |
c74df32c HD |
3360 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
3361 | * but allow concurrent faults), and pte mapped but not yet locked. | |
9a95f3cf PC |
3362 | * We return with pte unmapped and unlocked. |
3363 | * | |
3364 | * The mmap_sem may have been released depending on flags and our | |
3365 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 3366 | */ |
c0292554 | 3367 | static int handle_pte_fault(struct mm_struct *mm, |
71e3aac0 AA |
3368 | struct vm_area_struct *vma, unsigned long address, |
3369 | pte_t *pte, pmd_t *pmd, unsigned int flags) | |
1da177e4 LT |
3370 | { |
3371 | pte_t entry; | |
8f4e2101 | 3372 | spinlock_t *ptl; |
1da177e4 | 3373 | |
e37c6982 CB |
3374 | /* |
3375 | * some architectures can have larger ptes than wordsize, | |
3376 | * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and CONFIG_32BIT=y, | |
3377 | * so READ_ONCE or ACCESS_ONCE cannot guarantee atomic accesses. | |
3378 | * The code below just needs a consistent view for the ifs and | |
3379 | * we later double check anyway with the ptl lock held. So here | |
3380 | * a barrier will do. | |
3381 | */ | |
3382 | entry = *pte; | |
3383 | barrier(); | |
1da177e4 | 3384 | if (!pte_present(entry)) { |
65500d23 | 3385 | if (pte_none(entry)) { |
b5330628 ON |
3386 | if (vma_is_anonymous(vma)) |
3387 | return do_anonymous_page(mm, vma, address, | |
3388 | pte, pmd, flags); | |
3389 | else | |
6b7339f4 KS |
3390 | return do_fault(mm, vma, address, pte, pmd, |
3391 | flags, entry); | |
65500d23 | 3392 | } |
65500d23 | 3393 | return do_swap_page(mm, vma, address, |
30c9f3a9 | 3394 | pte, pmd, flags, entry); |
1da177e4 LT |
3395 | } |
3396 | ||
8a0516ed | 3397 | if (pte_protnone(entry)) |
d10e63f2 MG |
3398 | return do_numa_page(mm, vma, address, entry, pte, pmd); |
3399 | ||
4c21e2f2 | 3400 | ptl = pte_lockptr(mm, pmd); |
8f4e2101 HD |
3401 | spin_lock(ptl); |
3402 | if (unlikely(!pte_same(*pte, entry))) | |
3403 | goto unlock; | |
30c9f3a9 | 3404 | if (flags & FAULT_FLAG_WRITE) { |
1da177e4 | 3405 | if (!pte_write(entry)) |
8f4e2101 HD |
3406 | return do_wp_page(mm, vma, address, |
3407 | pte, pmd, ptl, entry); | |
1da177e4 LT |
3408 | entry = pte_mkdirty(entry); |
3409 | } | |
3410 | entry = pte_mkyoung(entry); | |
30c9f3a9 | 3411 | if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) { |
4b3073e1 | 3412 | update_mmu_cache(vma, address, pte); |
1a44e149 AA |
3413 | } else { |
3414 | /* | |
3415 | * This is needed only for protection faults but the arch code | |
3416 | * is not yet telling us if this is a protection fault or not. | |
3417 | * This still avoids useless tlb flushes for .text page faults | |
3418 | * with threads. | |
3419 | */ | |
30c9f3a9 | 3420 | if (flags & FAULT_FLAG_WRITE) |
61c77326 | 3421 | flush_tlb_fix_spurious_fault(vma, address); |
1a44e149 | 3422 | } |
8f4e2101 HD |
3423 | unlock: |
3424 | pte_unmap_unlock(pte, ptl); | |
83c54070 | 3425 | return 0; |
1da177e4 LT |
3426 | } |
3427 | ||
3428 | /* | |
3429 | * By the time we get here, we already hold the mm semaphore | |
9a95f3cf PC |
3430 | * |
3431 | * The mmap_sem may have been released depending on flags and our | |
3432 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 3433 | */ |
519e5247 JW |
3434 | static int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
3435 | unsigned long address, unsigned int flags) | |
1da177e4 LT |
3436 | { |
3437 | pgd_t *pgd; | |
3438 | pud_t *pud; | |
3439 | pmd_t *pmd; | |
3440 | pte_t *pte; | |
3441 | ||
1b2ee126 | 3442 | if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE, |
d61172b4 | 3443 | flags & FAULT_FLAG_INSTRUCTION, |
1b2ee126 | 3444 | flags & FAULT_FLAG_REMOTE)) |
33a709b2 DH |
3445 | return VM_FAULT_SIGSEGV; |
3446 | ||
ac9b9c66 | 3447 | if (unlikely(is_vm_hugetlb_page(vma))) |
30c9f3a9 | 3448 | return hugetlb_fault(mm, vma, address, flags); |
1da177e4 | 3449 | |
1da177e4 | 3450 | pgd = pgd_offset(mm, address); |
1da177e4 LT |
3451 | pud = pud_alloc(mm, pgd, address); |
3452 | if (!pud) | |
c74df32c | 3453 | return VM_FAULT_OOM; |
1da177e4 LT |
3454 | pmd = pmd_alloc(mm, pud, address); |
3455 | if (!pmd) | |
c74df32c | 3456 | return VM_FAULT_OOM; |
71e3aac0 | 3457 | if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) { |
b96375f7 | 3458 | int ret = create_huge_pmd(mm, vma, address, pmd, flags); |
c0292554 KS |
3459 | if (!(ret & VM_FAULT_FALLBACK)) |
3460 | return ret; | |
71e3aac0 AA |
3461 | } else { |
3462 | pmd_t orig_pmd = *pmd; | |
1f1d06c3 DR |
3463 | int ret; |
3464 | ||
71e3aac0 | 3465 | barrier(); |
5c7fb56e | 3466 | if (pmd_trans_huge(orig_pmd) || pmd_devmap(orig_pmd)) { |
a1dd450b WD |
3467 | unsigned int dirty = flags & FAULT_FLAG_WRITE; |
3468 | ||
8a0516ed | 3469 | if (pmd_protnone(orig_pmd)) |
4daae3b4 | 3470 | return do_huge_pmd_numa_page(mm, vma, address, |
d10e63f2 MG |
3471 | orig_pmd, pmd); |
3472 | ||
3d59eebc | 3473 | if (dirty && !pmd_write(orig_pmd)) { |
b96375f7 MW |
3474 | ret = wp_huge_pmd(mm, vma, address, pmd, |
3475 | orig_pmd, flags); | |
9845cbbd KS |
3476 | if (!(ret & VM_FAULT_FALLBACK)) |
3477 | return ret; | |
a1dd450b WD |
3478 | } else { |
3479 | huge_pmd_set_accessed(mm, vma, address, pmd, | |
3480 | orig_pmd, dirty); | |
9845cbbd | 3481 | return 0; |
1f1d06c3 | 3482 | } |
71e3aac0 AA |
3483 | } |
3484 | } | |
3485 | ||
3486 | /* | |
3ed3a4f0 | 3487 | * Use pte_alloc() instead of pte_alloc_map, because we can't |
71e3aac0 AA |
3488 | * run pte_offset_map on the pmd, if an huge pmd could |
3489 | * materialize from under us from a different thread. | |
3490 | */ | |
3ed3a4f0 | 3491 | if (unlikely(pte_alloc(mm, pmd, address))) |
c74df32c | 3492 | return VM_FAULT_OOM; |
ad33bb04 AA |
3493 | /* |
3494 | * If a huge pmd materialized under us just retry later. Use | |
3495 | * pmd_trans_unstable() instead of pmd_trans_huge() to ensure the pmd | |
3496 | * didn't become pmd_trans_huge under us and then back to pmd_none, as | |
3497 | * a result of MADV_DONTNEED running immediately after a huge pmd fault | |
3498 | * in a different thread of this mm, in turn leading to a misleading | |
3499 | * pmd_trans_huge() retval. All we have to ensure is that it is a | |
3500 | * regular pmd that we can walk with pte_offset_map() and we can do that | |
3501 | * through an atomic read in C, which is what pmd_trans_unstable() | |
3502 | * provides. | |
3503 | */ | |
3504 | if (unlikely(pmd_trans_unstable(pmd) || pmd_devmap(*pmd))) | |
71e3aac0 AA |
3505 | return 0; |
3506 | /* | |
3507 | * A regular pmd is established and it can't morph into a huge pmd | |
3508 | * from under us anymore at this point because we hold the mmap_sem | |
3509 | * read mode and khugepaged takes it in write mode. So now it's | |
3510 | * safe to run pte_offset_map(). | |
3511 | */ | |
3512 | pte = pte_offset_map(pmd, address); | |
1da177e4 | 3513 | |
30c9f3a9 | 3514 | return handle_pte_fault(mm, vma, address, pte, pmd, flags); |
1da177e4 LT |
3515 | } |
3516 | ||
9a95f3cf PC |
3517 | /* |
3518 | * By the time we get here, we already hold the mm semaphore | |
3519 | * | |
3520 | * The mmap_sem may have been released depending on flags and our | |
3521 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
3522 | */ | |
519e5247 JW |
3523 | int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
3524 | unsigned long address, unsigned int flags) | |
3525 | { | |
3526 | int ret; | |
3527 | ||
3528 | __set_current_state(TASK_RUNNING); | |
3529 | ||
3530 | count_vm_event(PGFAULT); | |
3531 | mem_cgroup_count_vm_event(mm, PGFAULT); | |
3532 | ||
3533 | /* do counter updates before entering really critical section. */ | |
3534 | check_sync_rss_stat(current); | |
3535 | ||
3536 | /* | |
3537 | * Enable the memcg OOM handling for faults triggered in user | |
3538 | * space. Kernel faults are handled more gracefully. | |
3539 | */ | |
3540 | if (flags & FAULT_FLAG_USER) | |
49426420 | 3541 | mem_cgroup_oom_enable(); |
519e5247 JW |
3542 | |
3543 | ret = __handle_mm_fault(mm, vma, address, flags); | |
3544 | ||
49426420 JW |
3545 | if (flags & FAULT_FLAG_USER) { |
3546 | mem_cgroup_oom_disable(); | |
3547 | /* | |
3548 | * The task may have entered a memcg OOM situation but | |
3549 | * if the allocation error was handled gracefully (no | |
3550 | * VM_FAULT_OOM), there is no need to kill anything. | |
3551 | * Just clean up the OOM state peacefully. | |
3552 | */ | |
3553 | if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) | |
3554 | mem_cgroup_oom_synchronize(false); | |
3555 | } | |
3812c8c8 | 3556 | |
519e5247 JW |
3557 | return ret; |
3558 | } | |
e1d6d01a | 3559 | EXPORT_SYMBOL_GPL(handle_mm_fault); |
519e5247 | 3560 | |
1da177e4 LT |
3561 | #ifndef __PAGETABLE_PUD_FOLDED |
3562 | /* | |
3563 | * Allocate page upper directory. | |
872fec16 | 3564 | * We've already handled the fast-path in-line. |
1da177e4 | 3565 | */ |
1bb3630e | 3566 | int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) |
1da177e4 | 3567 | { |
c74df32c HD |
3568 | pud_t *new = pud_alloc_one(mm, address); |
3569 | if (!new) | |
1bb3630e | 3570 | return -ENOMEM; |
1da177e4 | 3571 | |
362a61ad NP |
3572 | smp_wmb(); /* See comment in __pte_alloc */ |
3573 | ||
872fec16 | 3574 | spin_lock(&mm->page_table_lock); |
1bb3630e | 3575 | if (pgd_present(*pgd)) /* Another has populated it */ |
5e541973 | 3576 | pud_free(mm, new); |
1bb3630e HD |
3577 | else |
3578 | pgd_populate(mm, pgd, new); | |
c74df32c | 3579 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 3580 | return 0; |
1da177e4 LT |
3581 | } |
3582 | #endif /* __PAGETABLE_PUD_FOLDED */ | |
3583 | ||
3584 | #ifndef __PAGETABLE_PMD_FOLDED | |
3585 | /* | |
3586 | * Allocate page middle directory. | |
872fec16 | 3587 | * We've already handled the fast-path in-line. |
1da177e4 | 3588 | */ |
1bb3630e | 3589 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
1da177e4 | 3590 | { |
c74df32c HD |
3591 | pmd_t *new = pmd_alloc_one(mm, address); |
3592 | if (!new) | |
1bb3630e | 3593 | return -ENOMEM; |
1da177e4 | 3594 | |
362a61ad NP |
3595 | smp_wmb(); /* See comment in __pte_alloc */ |
3596 | ||
872fec16 | 3597 | spin_lock(&mm->page_table_lock); |
1da177e4 | 3598 | #ifndef __ARCH_HAS_4LEVEL_HACK |
dc6c9a35 KS |
3599 | if (!pud_present(*pud)) { |
3600 | mm_inc_nr_pmds(mm); | |
1bb3630e | 3601 | pud_populate(mm, pud, new); |
dc6c9a35 | 3602 | } else /* Another has populated it */ |
5e541973 | 3603 | pmd_free(mm, new); |
dc6c9a35 KS |
3604 | #else |
3605 | if (!pgd_present(*pud)) { | |
3606 | mm_inc_nr_pmds(mm); | |
1bb3630e | 3607 | pgd_populate(mm, pud, new); |
dc6c9a35 KS |
3608 | } else /* Another has populated it */ |
3609 | pmd_free(mm, new); | |
1da177e4 | 3610 | #endif /* __ARCH_HAS_4LEVEL_HACK */ |
c74df32c | 3611 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 3612 | return 0; |
e0f39591 | 3613 | } |
1da177e4 LT |
3614 | #endif /* __PAGETABLE_PMD_FOLDED */ |
3615 | ||
1b36ba81 | 3616 | static int __follow_pte(struct mm_struct *mm, unsigned long address, |
f8ad0f49 JW |
3617 | pte_t **ptepp, spinlock_t **ptlp) |
3618 | { | |
3619 | pgd_t *pgd; | |
3620 | pud_t *pud; | |
3621 | pmd_t *pmd; | |
3622 | pte_t *ptep; | |
3623 | ||
3624 | pgd = pgd_offset(mm, address); | |
3625 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
3626 | goto out; | |
3627 | ||
3628 | pud = pud_offset(pgd, address); | |
3629 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) | |
3630 | goto out; | |
3631 | ||
3632 | pmd = pmd_offset(pud, address); | |
f66055ab | 3633 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
f8ad0f49 JW |
3634 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) |
3635 | goto out; | |
3636 | ||
3637 | /* We cannot handle huge page PFN maps. Luckily they don't exist. */ | |
3638 | if (pmd_huge(*pmd)) | |
3639 | goto out; | |
3640 | ||
3641 | ptep = pte_offset_map_lock(mm, pmd, address, ptlp); | |
3642 | if (!ptep) | |
3643 | goto out; | |
3644 | if (!pte_present(*ptep)) | |
3645 | goto unlock; | |
3646 | *ptepp = ptep; | |
3647 | return 0; | |
3648 | unlock: | |
3649 | pte_unmap_unlock(ptep, *ptlp); | |
3650 | out: | |
3651 | return -EINVAL; | |
3652 | } | |
3653 | ||
1b36ba81 NK |
3654 | static inline int follow_pte(struct mm_struct *mm, unsigned long address, |
3655 | pte_t **ptepp, spinlock_t **ptlp) | |
3656 | { | |
3657 | int res; | |
3658 | ||
3659 | /* (void) is needed to make gcc happy */ | |
3660 | (void) __cond_lock(*ptlp, | |
3661 | !(res = __follow_pte(mm, address, ptepp, ptlp))); | |
3662 | return res; | |
3663 | } | |
3664 | ||
3b6748e2 JW |
3665 | /** |
3666 | * follow_pfn - look up PFN at a user virtual address | |
3667 | * @vma: memory mapping | |
3668 | * @address: user virtual address | |
3669 | * @pfn: location to store found PFN | |
3670 | * | |
3671 | * Only IO mappings and raw PFN mappings are allowed. | |
3672 | * | |
3673 | * Returns zero and the pfn at @pfn on success, -ve otherwise. | |
3674 | */ | |
3675 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
3676 | unsigned long *pfn) | |
3677 | { | |
3678 | int ret = -EINVAL; | |
3679 | spinlock_t *ptl; | |
3680 | pte_t *ptep; | |
3681 | ||
3682 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
3683 | return ret; | |
3684 | ||
3685 | ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); | |
3686 | if (ret) | |
3687 | return ret; | |
3688 | *pfn = pte_pfn(*ptep); | |
3689 | pte_unmap_unlock(ptep, ptl); | |
3690 | return 0; | |
3691 | } | |
3692 | EXPORT_SYMBOL(follow_pfn); | |
3693 | ||
28b2ee20 | 3694 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
d87fe660 | 3695 | int follow_phys(struct vm_area_struct *vma, |
3696 | unsigned long address, unsigned int flags, | |
3697 | unsigned long *prot, resource_size_t *phys) | |
28b2ee20 | 3698 | { |
03668a4d | 3699 | int ret = -EINVAL; |
28b2ee20 RR |
3700 | pte_t *ptep, pte; |
3701 | spinlock_t *ptl; | |
28b2ee20 | 3702 | |
d87fe660 | 3703 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
3704 | goto out; | |
28b2ee20 | 3705 | |
03668a4d | 3706 | if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) |
d87fe660 | 3707 | goto out; |
28b2ee20 | 3708 | pte = *ptep; |
03668a4d | 3709 | |
28b2ee20 RR |
3710 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
3711 | goto unlock; | |
28b2ee20 RR |
3712 | |
3713 | *prot = pgprot_val(pte_pgprot(pte)); | |
03668a4d | 3714 | *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; |
28b2ee20 | 3715 | |
03668a4d | 3716 | ret = 0; |
28b2ee20 RR |
3717 | unlock: |
3718 | pte_unmap_unlock(ptep, ptl); | |
3719 | out: | |
d87fe660 | 3720 | return ret; |
28b2ee20 RR |
3721 | } |
3722 | ||
3723 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, | |
3724 | void *buf, int len, int write) | |
3725 | { | |
3726 | resource_size_t phys_addr; | |
3727 | unsigned long prot = 0; | |
2bc7273b | 3728 | void __iomem *maddr; |
28b2ee20 RR |
3729 | int offset = addr & (PAGE_SIZE-1); |
3730 | ||
d87fe660 | 3731 | if (follow_phys(vma, addr, write, &prot, &phys_addr)) |
28b2ee20 RR |
3732 | return -EINVAL; |
3733 | ||
9cb12d7b | 3734 | maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); |
28b2ee20 RR |
3735 | if (write) |
3736 | memcpy_toio(maddr + offset, buf, len); | |
3737 | else | |
3738 | memcpy_fromio(buf, maddr + offset, len); | |
3739 | iounmap(maddr); | |
3740 | ||
3741 | return len; | |
3742 | } | |
5a73633e | 3743 | EXPORT_SYMBOL_GPL(generic_access_phys); |
28b2ee20 RR |
3744 | #endif |
3745 | ||
0ec76a11 | 3746 | /* |
206cb636 SW |
3747 | * Access another process' address space as given in mm. If non-NULL, use the |
3748 | * given task for page fault accounting. | |
0ec76a11 | 3749 | */ |
206cb636 SW |
3750 | static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, |
3751 | unsigned long addr, void *buf, int len, int write) | |
0ec76a11 | 3752 | { |
0ec76a11 | 3753 | struct vm_area_struct *vma; |
0ec76a11 DH |
3754 | void *old_buf = buf; |
3755 | ||
0ec76a11 | 3756 | down_read(&mm->mmap_sem); |
183ff22b | 3757 | /* ignore errors, just check how much was successfully transferred */ |
0ec76a11 DH |
3758 | while (len) { |
3759 | int bytes, ret, offset; | |
3760 | void *maddr; | |
28b2ee20 | 3761 | struct page *page = NULL; |
0ec76a11 | 3762 | |
1e987790 | 3763 | ret = get_user_pages_remote(tsk, mm, addr, 1, |
0ec76a11 | 3764 | write, 1, &page, &vma); |
28b2ee20 | 3765 | if (ret <= 0) { |
dbffcd03 RR |
3766 | #ifndef CONFIG_HAVE_IOREMAP_PROT |
3767 | break; | |
3768 | #else | |
28b2ee20 RR |
3769 | /* |
3770 | * Check if this is a VM_IO | VM_PFNMAP VMA, which | |
3771 | * we can access using slightly different code. | |
3772 | */ | |
28b2ee20 | 3773 | vma = find_vma(mm, addr); |
fe936dfc | 3774 | if (!vma || vma->vm_start > addr) |
28b2ee20 RR |
3775 | break; |
3776 | if (vma->vm_ops && vma->vm_ops->access) | |
3777 | ret = vma->vm_ops->access(vma, addr, buf, | |
3778 | len, write); | |
3779 | if (ret <= 0) | |
28b2ee20 RR |
3780 | break; |
3781 | bytes = ret; | |
dbffcd03 | 3782 | #endif |
0ec76a11 | 3783 | } else { |
28b2ee20 RR |
3784 | bytes = len; |
3785 | offset = addr & (PAGE_SIZE-1); | |
3786 | if (bytes > PAGE_SIZE-offset) | |
3787 | bytes = PAGE_SIZE-offset; | |
3788 | ||
3789 | maddr = kmap(page); | |
3790 | if (write) { | |
3791 | copy_to_user_page(vma, page, addr, | |
3792 | maddr + offset, buf, bytes); | |
3793 | set_page_dirty_lock(page); | |
3794 | } else { | |
3795 | copy_from_user_page(vma, page, addr, | |
3796 | buf, maddr + offset, bytes); | |
3797 | } | |
3798 | kunmap(page); | |
09cbfeaf | 3799 | put_page(page); |
0ec76a11 | 3800 | } |
0ec76a11 DH |
3801 | len -= bytes; |
3802 | buf += bytes; | |
3803 | addr += bytes; | |
3804 | } | |
3805 | up_read(&mm->mmap_sem); | |
0ec76a11 DH |
3806 | |
3807 | return buf - old_buf; | |
3808 | } | |
03252919 | 3809 | |
5ddd36b9 | 3810 | /** |
ae91dbfc | 3811 | * access_remote_vm - access another process' address space |
5ddd36b9 SW |
3812 | * @mm: the mm_struct of the target address space |
3813 | * @addr: start address to access | |
3814 | * @buf: source or destination buffer | |
3815 | * @len: number of bytes to transfer | |
3816 | * @write: whether the access is a write | |
3817 | * | |
3818 | * The caller must hold a reference on @mm. | |
3819 | */ | |
3820 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, | |
3821 | void *buf, int len, int write) | |
3822 | { | |
3823 | return __access_remote_vm(NULL, mm, addr, buf, len, write); | |
3824 | } | |
3825 | ||
206cb636 SW |
3826 | /* |
3827 | * Access another process' address space. | |
3828 | * Source/target buffer must be kernel space, | |
3829 | * Do not walk the page table directly, use get_user_pages | |
3830 | */ | |
3831 | int access_process_vm(struct task_struct *tsk, unsigned long addr, | |
3832 | void *buf, int len, int write) | |
3833 | { | |
3834 | struct mm_struct *mm; | |
3835 | int ret; | |
3836 | ||
3837 | mm = get_task_mm(tsk); | |
3838 | if (!mm) | |
3839 | return 0; | |
3840 | ||
3841 | ret = __access_remote_vm(tsk, mm, addr, buf, len, write); | |
3842 | mmput(mm); | |
3843 | ||
3844 | return ret; | |
3845 | } | |
3846 | ||
03252919 AK |
3847 | /* |
3848 | * Print the name of a VMA. | |
3849 | */ | |
3850 | void print_vma_addr(char *prefix, unsigned long ip) | |
3851 | { | |
3852 | struct mm_struct *mm = current->mm; | |
3853 | struct vm_area_struct *vma; | |
3854 | ||
e8bff74a IM |
3855 | /* |
3856 | * Do not print if we are in atomic | |
3857 | * contexts (in exception stacks, etc.): | |
3858 | */ | |
3859 | if (preempt_count()) | |
3860 | return; | |
3861 | ||
03252919 AK |
3862 | down_read(&mm->mmap_sem); |
3863 | vma = find_vma(mm, ip); | |
3864 | if (vma && vma->vm_file) { | |
3865 | struct file *f = vma->vm_file; | |
3866 | char *buf = (char *)__get_free_page(GFP_KERNEL); | |
3867 | if (buf) { | |
2fbc57c5 | 3868 | char *p; |
03252919 | 3869 | |
9bf39ab2 | 3870 | p = file_path(f, buf, PAGE_SIZE); |
03252919 AK |
3871 | if (IS_ERR(p)) |
3872 | p = "?"; | |
2fbc57c5 | 3873 | printk("%s%s[%lx+%lx]", prefix, kbasename(p), |
03252919 AK |
3874 | vma->vm_start, |
3875 | vma->vm_end - vma->vm_start); | |
3876 | free_page((unsigned long)buf); | |
3877 | } | |
3878 | } | |
51a07e50 | 3879 | up_read(&mm->mmap_sem); |
03252919 | 3880 | } |
3ee1afa3 | 3881 | |
662bbcb2 | 3882 | #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) |
9ec23531 | 3883 | void __might_fault(const char *file, int line) |
3ee1afa3 | 3884 | { |
95156f00 PZ |
3885 | /* |
3886 | * Some code (nfs/sunrpc) uses socket ops on kernel memory while | |
3887 | * holding the mmap_sem, this is safe because kernel memory doesn't | |
3888 | * get paged out, therefore we'll never actually fault, and the | |
3889 | * below annotations will generate false positives. | |
3890 | */ | |
3891 | if (segment_eq(get_fs(), KERNEL_DS)) | |
3892 | return; | |
9ec23531 | 3893 | if (pagefault_disabled()) |
662bbcb2 | 3894 | return; |
9ec23531 DH |
3895 | __might_sleep(file, line, 0); |
3896 | #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) | |
662bbcb2 | 3897 | if (current->mm) |
3ee1afa3 | 3898 | might_lock_read(¤t->mm->mmap_sem); |
9ec23531 | 3899 | #endif |
3ee1afa3 | 3900 | } |
9ec23531 | 3901 | EXPORT_SYMBOL(__might_fault); |
3ee1afa3 | 3902 | #endif |
47ad8475 AA |
3903 | |
3904 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) | |
3905 | static void clear_gigantic_page(struct page *page, | |
3906 | unsigned long addr, | |
3907 | unsigned int pages_per_huge_page) | |
3908 | { | |
3909 | int i; | |
3910 | struct page *p = page; | |
3911 | ||
3912 | might_sleep(); | |
3913 | for (i = 0; i < pages_per_huge_page; | |
3914 | i++, p = mem_map_next(p, page, i)) { | |
3915 | cond_resched(); | |
3916 | clear_user_highpage(p, addr + i * PAGE_SIZE); | |
3917 | } | |
3918 | } | |
3919 | void clear_huge_page(struct page *page, | |
3920 | unsigned long addr, unsigned int pages_per_huge_page) | |
3921 | { | |
3922 | int i; | |
3923 | ||
3924 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
3925 | clear_gigantic_page(page, addr, pages_per_huge_page); | |
3926 | return; | |
3927 | } | |
3928 | ||
3929 | might_sleep(); | |
3930 | for (i = 0; i < pages_per_huge_page; i++) { | |
3931 | cond_resched(); | |
3932 | clear_user_highpage(page + i, addr + i * PAGE_SIZE); | |
3933 | } | |
3934 | } | |
3935 | ||
3936 | static void copy_user_gigantic_page(struct page *dst, struct page *src, | |
3937 | unsigned long addr, | |
3938 | struct vm_area_struct *vma, | |
3939 | unsigned int pages_per_huge_page) | |
3940 | { | |
3941 | int i; | |
3942 | struct page *dst_base = dst; | |
3943 | struct page *src_base = src; | |
3944 | ||
3945 | for (i = 0; i < pages_per_huge_page; ) { | |
3946 | cond_resched(); | |
3947 | copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); | |
3948 | ||
3949 | i++; | |
3950 | dst = mem_map_next(dst, dst_base, i); | |
3951 | src = mem_map_next(src, src_base, i); | |
3952 | } | |
3953 | } | |
3954 | ||
3955 | void copy_user_huge_page(struct page *dst, struct page *src, | |
3956 | unsigned long addr, struct vm_area_struct *vma, | |
3957 | unsigned int pages_per_huge_page) | |
3958 | { | |
3959 | int i; | |
3960 | ||
3961 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
3962 | copy_user_gigantic_page(dst, src, addr, vma, | |
3963 | pages_per_huge_page); | |
3964 | return; | |
3965 | } | |
3966 | ||
3967 | might_sleep(); | |
3968 | for (i = 0; i < pages_per_huge_page; i++) { | |
3969 | cond_resched(); | |
3970 | copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma); | |
3971 | } | |
3972 | } | |
3973 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ | |
49076ec2 | 3974 | |
40b64acd | 3975 | #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS |
b35f1819 KS |
3976 | |
3977 | static struct kmem_cache *page_ptl_cachep; | |
3978 | ||
3979 | void __init ptlock_cache_init(void) | |
3980 | { | |
3981 | page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, | |
3982 | SLAB_PANIC, NULL); | |
3983 | } | |
3984 | ||
539edb58 | 3985 | bool ptlock_alloc(struct page *page) |
49076ec2 KS |
3986 | { |
3987 | spinlock_t *ptl; | |
3988 | ||
b35f1819 | 3989 | ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); |
49076ec2 KS |
3990 | if (!ptl) |
3991 | return false; | |
539edb58 | 3992 | page->ptl = ptl; |
49076ec2 KS |
3993 | return true; |
3994 | } | |
3995 | ||
539edb58 | 3996 | void ptlock_free(struct page *page) |
49076ec2 | 3997 | { |
b35f1819 | 3998 | kmem_cache_free(page_ptl_cachep, page->ptl); |
49076ec2 KS |
3999 | } |
4000 | #endif |