]>
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. | |
166f61b9 | 33 | * Idea by Alex Bligh (alex@cconcepts.co.uk) |
1da177e4 LT |
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> | |
6e84f315 | 43 | #include <linux/sched/mm.h> |
f7ccbae4 | 44 | #include <linux/sched/coredump.h> |
6a3827d7 | 45 | #include <linux/sched/numa_balancing.h> |
29930025 | 46 | #include <linux/sched/task.h> |
1da177e4 LT |
47 | #include <linux/hugetlb.h> |
48 | #include <linux/mman.h> | |
49 | #include <linux/swap.h> | |
50 | #include <linux/highmem.h> | |
51 | #include <linux/pagemap.h> | |
9a840895 | 52 | #include <linux/ksm.h> |
1da177e4 | 53 | #include <linux/rmap.h> |
b95f1b31 | 54 | #include <linux/export.h> |
0ff92245 | 55 | #include <linux/delayacct.h> |
1da177e4 | 56 | #include <linux/init.h> |
01c8f1c4 | 57 | #include <linux/pfn_t.h> |
edc79b2a | 58 | #include <linux/writeback.h> |
8a9f3ccd | 59 | #include <linux/memcontrol.h> |
cddb8a5c | 60 | #include <linux/mmu_notifier.h> |
3dc14741 HD |
61 | #include <linux/kallsyms.h> |
62 | #include <linux/swapops.h> | |
63 | #include <linux/elf.h> | |
5a0e3ad6 | 64 | #include <linux/gfp.h> |
4daae3b4 | 65 | #include <linux/migrate.h> |
2fbc57c5 | 66 | #include <linux/string.h> |
0abdd7a8 | 67 | #include <linux/dma-debug.h> |
1592eef0 | 68 | #include <linux/debugfs.h> |
6b251fc9 | 69 | #include <linux/userfaultfd_k.h> |
bc2466e4 | 70 | #include <linux/dax.h> |
1da177e4 | 71 | |
6952b61d | 72 | #include <asm/io.h> |
33a709b2 | 73 | #include <asm/mmu_context.h> |
1da177e4 | 74 | #include <asm/pgalloc.h> |
7c0f6ba6 | 75 | #include <linux/uaccess.h> |
1da177e4 LT |
76 | #include <asm/tlb.h> |
77 | #include <asm/tlbflush.h> | |
78 | #include <asm/pgtable.h> | |
79 | ||
42b77728 JB |
80 | #include "internal.h" |
81 | ||
90572890 PZ |
82 | #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS |
83 | #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid. | |
75980e97 PZ |
84 | #endif |
85 | ||
d41dee36 | 86 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
87 | /* use the per-pgdat data instead for discontigmem - mbligh */ |
88 | unsigned long max_mapnr; | |
1da177e4 | 89 | EXPORT_SYMBOL(max_mapnr); |
166f61b9 TH |
90 | |
91 | struct page *mem_map; | |
1da177e4 LT |
92 | EXPORT_SYMBOL(mem_map); |
93 | #endif | |
94 | ||
1da177e4 LT |
95 | /* |
96 | * A number of key systems in x86 including ioremap() rely on the assumption | |
97 | * that high_memory defines the upper bound on direct map memory, then end | |
98 | * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and | |
99 | * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL | |
100 | * and ZONE_HIGHMEM. | |
101 | */ | |
166f61b9 | 102 | void *high_memory; |
1da177e4 | 103 | EXPORT_SYMBOL(high_memory); |
1da177e4 | 104 | |
32a93233 IM |
105 | /* |
106 | * Randomize the address space (stacks, mmaps, brk, etc.). | |
107 | * | |
108 | * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization, | |
109 | * as ancient (libc5 based) binaries can segfault. ) | |
110 | */ | |
111 | int randomize_va_space __read_mostly = | |
112 | #ifdef CONFIG_COMPAT_BRK | |
113 | 1; | |
114 | #else | |
115 | 2; | |
116 | #endif | |
a62eaf15 AK |
117 | |
118 | static int __init disable_randmaps(char *s) | |
119 | { | |
120 | randomize_va_space = 0; | |
9b41046c | 121 | return 1; |
a62eaf15 AK |
122 | } |
123 | __setup("norandmaps", disable_randmaps); | |
124 | ||
62eede62 | 125 | unsigned long zero_pfn __read_mostly; |
0b70068e AB |
126 | EXPORT_SYMBOL(zero_pfn); |
127 | ||
166f61b9 TH |
128 | unsigned long highest_memmap_pfn __read_mostly; |
129 | ||
a13ea5b7 HD |
130 | /* |
131 | * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init() | |
132 | */ | |
133 | static int __init init_zero_pfn(void) | |
134 | { | |
135 | zero_pfn = page_to_pfn(ZERO_PAGE(0)); | |
136 | return 0; | |
137 | } | |
138 | core_initcall(init_zero_pfn); | |
a62eaf15 | 139 | |
d559db08 | 140 | |
34e55232 KH |
141 | #if defined(SPLIT_RSS_COUNTING) |
142 | ||
ea48cf78 | 143 | void sync_mm_rss(struct mm_struct *mm) |
34e55232 KH |
144 | { |
145 | int i; | |
146 | ||
147 | for (i = 0; i < NR_MM_COUNTERS; i++) { | |
05af2e10 DR |
148 | if (current->rss_stat.count[i]) { |
149 | add_mm_counter(mm, i, current->rss_stat.count[i]); | |
150 | current->rss_stat.count[i] = 0; | |
34e55232 KH |
151 | } |
152 | } | |
05af2e10 | 153 | current->rss_stat.events = 0; |
34e55232 KH |
154 | } |
155 | ||
156 | static void add_mm_counter_fast(struct mm_struct *mm, int member, int val) | |
157 | { | |
158 | struct task_struct *task = current; | |
159 | ||
160 | if (likely(task->mm == mm)) | |
161 | task->rss_stat.count[member] += val; | |
162 | else | |
163 | add_mm_counter(mm, member, val); | |
164 | } | |
165 | #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1) | |
166 | #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1) | |
167 | ||
168 | /* sync counter once per 64 page faults */ | |
169 | #define TASK_RSS_EVENTS_THRESH (64) | |
170 | static void check_sync_rss_stat(struct task_struct *task) | |
171 | { | |
172 | if (unlikely(task != current)) | |
173 | return; | |
174 | if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH)) | |
ea48cf78 | 175 | sync_mm_rss(task->mm); |
34e55232 | 176 | } |
9547d01b | 177 | #else /* SPLIT_RSS_COUNTING */ |
34e55232 KH |
178 | |
179 | #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member) | |
180 | #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member) | |
181 | ||
182 | static void check_sync_rss_stat(struct task_struct *task) | |
183 | { | |
184 | } | |
185 | ||
9547d01b PZ |
186 | #endif /* SPLIT_RSS_COUNTING */ |
187 | ||
188 | #ifdef HAVE_GENERIC_MMU_GATHER | |
189 | ||
ca1d6c7d | 190 | static bool tlb_next_batch(struct mmu_gather *tlb) |
9547d01b PZ |
191 | { |
192 | struct mmu_gather_batch *batch; | |
193 | ||
194 | batch = tlb->active; | |
195 | if (batch->next) { | |
196 | tlb->active = batch->next; | |
ca1d6c7d | 197 | return true; |
9547d01b PZ |
198 | } |
199 | ||
53a59fc6 | 200 | if (tlb->batch_count == MAX_GATHER_BATCH_COUNT) |
ca1d6c7d | 201 | return false; |
53a59fc6 | 202 | |
9547d01b PZ |
203 | batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0); |
204 | if (!batch) | |
ca1d6c7d | 205 | return false; |
9547d01b | 206 | |
53a59fc6 | 207 | tlb->batch_count++; |
9547d01b PZ |
208 | batch->next = NULL; |
209 | batch->nr = 0; | |
210 | batch->max = MAX_GATHER_BATCH; | |
211 | ||
212 | tlb->active->next = batch; | |
213 | tlb->active = batch; | |
214 | ||
ca1d6c7d | 215 | return true; |
9547d01b PZ |
216 | } |
217 | ||
56236a59 MK |
218 | void arch_tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, |
219 | 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 | |
e77b0852 | 235 | tlb->page_size = 0; |
fb7332a9 WD |
236 | |
237 | __tlb_reset_range(tlb); | |
9547d01b PZ |
238 | } |
239 | ||
1cf35d47 | 240 | static void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb) |
9547d01b | 241 | { |
721c21c1 WD |
242 | if (!tlb->end) |
243 | return; | |
244 | ||
9547d01b | 245 | tlb_flush(tlb); |
34ee645e | 246 | mmu_notifier_invalidate_range(tlb->mm, tlb->start, tlb->end); |
9547d01b PZ |
247 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE |
248 | tlb_table_flush(tlb); | |
34e55232 | 249 | #endif |
fb7332a9 | 250 | __tlb_reset_range(tlb); |
1cf35d47 LT |
251 | } |
252 | ||
253 | static void tlb_flush_mmu_free(struct mmu_gather *tlb) | |
254 | { | |
255 | struct mmu_gather_batch *batch; | |
34e55232 | 256 | |
721c21c1 | 257 | for (batch = &tlb->local; batch && batch->nr; batch = batch->next) { |
9547d01b PZ |
258 | free_pages_and_swap_cache(batch->pages, batch->nr); |
259 | batch->nr = 0; | |
260 | } | |
261 | tlb->active = &tlb->local; | |
262 | } | |
263 | ||
1cf35d47 LT |
264 | void tlb_flush_mmu(struct mmu_gather *tlb) |
265 | { | |
1cf35d47 LT |
266 | tlb_flush_mmu_tlbonly(tlb); |
267 | tlb_flush_mmu_free(tlb); | |
268 | } | |
269 | ||
9547d01b PZ |
270 | /* tlb_finish_mmu |
271 | * Called at the end of the shootdown operation to free up any resources | |
272 | * that were required. | |
273 | */ | |
56236a59 MK |
274 | void arch_tlb_finish_mmu(struct mmu_gather *tlb, |
275 | unsigned long start, unsigned long end) | |
9547d01b PZ |
276 | { |
277 | struct mmu_gather_batch *batch, *next; | |
278 | ||
279 | tlb_flush_mmu(tlb); | |
280 | ||
281 | /* keep the page table cache within bounds */ | |
282 | check_pgt_cache(); | |
283 | ||
284 | for (batch = tlb->local.next; batch; batch = next) { | |
285 | next = batch->next; | |
286 | free_pages((unsigned long)batch, 0); | |
287 | } | |
288 | tlb->local.next = NULL; | |
289 | } | |
290 | ||
291 | /* __tlb_remove_page | |
292 | * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while | |
293 | * handling the additional races in SMP caused by other CPUs caching valid | |
294 | * mappings in their TLBs. Returns the number of free page slots left. | |
295 | * When out of page slots we must call tlb_flush_mmu(). | |
e9d55e15 | 296 | *returns true if the caller should flush. |
9547d01b | 297 | */ |
e77b0852 | 298 | bool __tlb_remove_page_size(struct mmu_gather *tlb, struct page *page, int page_size) |
9547d01b PZ |
299 | { |
300 | struct mmu_gather_batch *batch; | |
301 | ||
fb7332a9 | 302 | VM_BUG_ON(!tlb->end); |
692a68c1 | 303 | VM_WARN_ON(tlb->page_size != page_size); |
e77b0852 | 304 | |
9547d01b | 305 | batch = tlb->active; |
692a68c1 AK |
306 | /* |
307 | * Add the page and check if we are full. If so | |
308 | * force a flush. | |
309 | */ | |
310 | batch->pages[batch->nr++] = page; | |
9547d01b PZ |
311 | if (batch->nr == batch->max) { |
312 | if (!tlb_next_batch(tlb)) | |
e9d55e15 | 313 | return true; |
0b43c3aa | 314 | batch = tlb->active; |
9547d01b | 315 | } |
309381fe | 316 | VM_BUG_ON_PAGE(batch->nr > batch->max, page); |
9547d01b | 317 | |
e9d55e15 | 318 | return false; |
9547d01b PZ |
319 | } |
320 | ||
321 | #endif /* HAVE_GENERIC_MMU_GATHER */ | |
322 | ||
26723911 PZ |
323 | #ifdef CONFIG_HAVE_RCU_TABLE_FREE |
324 | ||
325 | /* | |
326 | * See the comment near struct mmu_table_batch. | |
327 | */ | |
328 | ||
329 | static void tlb_remove_table_smp_sync(void *arg) | |
330 | { | |
331 | /* Simply deliver the interrupt */ | |
332 | } | |
333 | ||
334 | static void tlb_remove_table_one(void *table) | |
335 | { | |
336 | /* | |
337 | * This isn't an RCU grace period and hence the page-tables cannot be | |
338 | * assumed to be actually RCU-freed. | |
339 | * | |
340 | * It is however sufficient for software page-table walkers that rely on | |
341 | * IRQ disabling. See the comment near struct mmu_table_batch. | |
342 | */ | |
343 | smp_call_function(tlb_remove_table_smp_sync, NULL, 1); | |
344 | __tlb_remove_table(table); | |
345 | } | |
346 | ||
347 | static void tlb_remove_table_rcu(struct rcu_head *head) | |
348 | { | |
349 | struct mmu_table_batch *batch; | |
350 | int i; | |
351 | ||
352 | batch = container_of(head, struct mmu_table_batch, rcu); | |
353 | ||
354 | for (i = 0; i < batch->nr; i++) | |
355 | __tlb_remove_table(batch->tables[i]); | |
356 | ||
357 | free_page((unsigned long)batch); | |
358 | } | |
359 | ||
360 | void tlb_table_flush(struct mmu_gather *tlb) | |
361 | { | |
362 | struct mmu_table_batch **batch = &tlb->batch; | |
363 | ||
364 | if (*batch) { | |
365 | call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu); | |
366 | *batch = NULL; | |
367 | } | |
368 | } | |
369 | ||
370 | void tlb_remove_table(struct mmu_gather *tlb, void *table) | |
371 | { | |
372 | struct mmu_table_batch **batch = &tlb->batch; | |
373 | ||
26723911 PZ |
374 | /* |
375 | * When there's less then two users of this mm there cannot be a | |
376 | * concurrent page-table walk. | |
377 | */ | |
378 | if (atomic_read(&tlb->mm->mm_users) < 2) { | |
379 | __tlb_remove_table(table); | |
380 | return; | |
381 | } | |
382 | ||
383 | if (*batch == NULL) { | |
384 | *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN); | |
385 | if (*batch == NULL) { | |
386 | tlb_remove_table_one(table); | |
387 | return; | |
388 | } | |
389 | (*batch)->nr = 0; | |
390 | } | |
391 | (*batch)->tables[(*batch)->nr++] = table; | |
392 | if ((*batch)->nr == MAX_TABLE_BATCH) | |
393 | tlb_table_flush(tlb); | |
394 | } | |
395 | ||
9547d01b | 396 | #endif /* CONFIG_HAVE_RCU_TABLE_FREE */ |
26723911 | 397 | |
56236a59 MK |
398 | /* tlb_gather_mmu |
399 | * Called to initialize an (on-stack) mmu_gather structure for page-table | |
400 | * tear-down from @mm. The @fullmm argument is used when @mm is without | |
401 | * users and we're going to destroy the full address space (exit/execve). | |
402 | */ | |
403 | void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, | |
404 | unsigned long start, unsigned long end) | |
405 | { | |
406 | arch_tlb_gather_mmu(tlb, mm, start, end); | |
407 | } | |
408 | ||
409 | void tlb_finish_mmu(struct mmu_gather *tlb, | |
410 | unsigned long start, unsigned long end) | |
411 | { | |
412 | arch_tlb_finish_mmu(tlb, start, end); | |
413 | } | |
414 | ||
1da177e4 LT |
415 | /* |
416 | * Note: this doesn't free the actual pages themselves. That | |
417 | * has been handled earlier when unmapping all the memory regions. | |
418 | */ | |
9e1b32ca BH |
419 | static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, |
420 | unsigned long addr) | |
1da177e4 | 421 | { |
2f569afd | 422 | pgtable_t token = pmd_pgtable(*pmd); |
e0da382c | 423 | pmd_clear(pmd); |
9e1b32ca | 424 | pte_free_tlb(tlb, token, addr); |
e1f56c89 | 425 | atomic_long_dec(&tlb->mm->nr_ptes); |
1da177e4 LT |
426 | } |
427 | ||
e0da382c HD |
428 | static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, |
429 | unsigned long addr, unsigned long end, | |
430 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
431 | { |
432 | pmd_t *pmd; | |
433 | unsigned long next; | |
e0da382c | 434 | unsigned long start; |
1da177e4 | 435 | |
e0da382c | 436 | start = addr; |
1da177e4 | 437 | pmd = pmd_offset(pud, addr); |
1da177e4 LT |
438 | do { |
439 | next = pmd_addr_end(addr, end); | |
440 | if (pmd_none_or_clear_bad(pmd)) | |
441 | continue; | |
9e1b32ca | 442 | free_pte_range(tlb, pmd, addr); |
1da177e4 LT |
443 | } while (pmd++, addr = next, addr != end); |
444 | ||
e0da382c HD |
445 | start &= PUD_MASK; |
446 | if (start < floor) | |
447 | return; | |
448 | if (ceiling) { | |
449 | ceiling &= PUD_MASK; | |
450 | if (!ceiling) | |
451 | return; | |
1da177e4 | 452 | } |
e0da382c HD |
453 | if (end - 1 > ceiling - 1) |
454 | return; | |
455 | ||
456 | pmd = pmd_offset(pud, start); | |
457 | pud_clear(pud); | |
9e1b32ca | 458 | pmd_free_tlb(tlb, pmd, start); |
dc6c9a35 | 459 | mm_dec_nr_pmds(tlb->mm); |
1da177e4 LT |
460 | } |
461 | ||
c2febafc | 462 | static inline void free_pud_range(struct mmu_gather *tlb, p4d_t *p4d, |
e0da382c HD |
463 | unsigned long addr, unsigned long end, |
464 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
465 | { |
466 | pud_t *pud; | |
467 | unsigned long next; | |
e0da382c | 468 | unsigned long start; |
1da177e4 | 469 | |
e0da382c | 470 | start = addr; |
c2febafc | 471 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
472 | do { |
473 | next = pud_addr_end(addr, end); | |
474 | if (pud_none_or_clear_bad(pud)) | |
475 | continue; | |
e0da382c | 476 | free_pmd_range(tlb, pud, addr, next, floor, ceiling); |
1da177e4 LT |
477 | } while (pud++, addr = next, addr != end); |
478 | ||
c2febafc KS |
479 | start &= P4D_MASK; |
480 | if (start < floor) | |
481 | return; | |
482 | if (ceiling) { | |
483 | ceiling &= P4D_MASK; | |
484 | if (!ceiling) | |
485 | return; | |
486 | } | |
487 | if (end - 1 > ceiling - 1) | |
488 | return; | |
489 | ||
490 | pud = pud_offset(p4d, start); | |
491 | p4d_clear(p4d); | |
492 | pud_free_tlb(tlb, pud, start); | |
493 | } | |
494 | ||
495 | static inline void free_p4d_range(struct mmu_gather *tlb, pgd_t *pgd, | |
496 | unsigned long addr, unsigned long end, | |
497 | unsigned long floor, unsigned long ceiling) | |
498 | { | |
499 | p4d_t *p4d; | |
500 | unsigned long next; | |
501 | unsigned long start; | |
502 | ||
503 | start = addr; | |
504 | p4d = p4d_offset(pgd, addr); | |
505 | do { | |
506 | next = p4d_addr_end(addr, end); | |
507 | if (p4d_none_or_clear_bad(p4d)) | |
508 | continue; | |
509 | free_pud_range(tlb, p4d, addr, next, floor, ceiling); | |
510 | } while (p4d++, addr = next, addr != end); | |
511 | ||
e0da382c HD |
512 | start &= PGDIR_MASK; |
513 | if (start < floor) | |
514 | return; | |
515 | if (ceiling) { | |
516 | ceiling &= PGDIR_MASK; | |
517 | if (!ceiling) | |
518 | return; | |
1da177e4 | 519 | } |
e0da382c HD |
520 | if (end - 1 > ceiling - 1) |
521 | return; | |
522 | ||
c2febafc | 523 | p4d = p4d_offset(pgd, start); |
e0da382c | 524 | pgd_clear(pgd); |
c2febafc | 525 | p4d_free_tlb(tlb, p4d, start); |
1da177e4 LT |
526 | } |
527 | ||
528 | /* | |
e0da382c | 529 | * This function frees user-level page tables of a process. |
1da177e4 | 530 | */ |
42b77728 | 531 | void free_pgd_range(struct mmu_gather *tlb, |
e0da382c HD |
532 | unsigned long addr, unsigned long end, |
533 | unsigned long floor, unsigned long ceiling) | |
1da177e4 LT |
534 | { |
535 | pgd_t *pgd; | |
536 | unsigned long next; | |
e0da382c HD |
537 | |
538 | /* | |
539 | * The next few lines have given us lots of grief... | |
540 | * | |
541 | * Why are we testing PMD* at this top level? Because often | |
542 | * there will be no work to do at all, and we'd prefer not to | |
543 | * go all the way down to the bottom just to discover that. | |
544 | * | |
545 | * Why all these "- 1"s? Because 0 represents both the bottom | |
546 | * of the address space and the top of it (using -1 for the | |
547 | * top wouldn't help much: the masks would do the wrong thing). | |
548 | * The rule is that addr 0 and floor 0 refer to the bottom of | |
549 | * the address space, but end 0 and ceiling 0 refer to the top | |
550 | * Comparisons need to use "end - 1" and "ceiling - 1" (though | |
551 | * that end 0 case should be mythical). | |
552 | * | |
553 | * Wherever addr is brought up or ceiling brought down, we must | |
554 | * be careful to reject "the opposite 0" before it confuses the | |
555 | * subsequent tests. But what about where end is brought down | |
556 | * by PMD_SIZE below? no, end can't go down to 0 there. | |
557 | * | |
558 | * Whereas we round start (addr) and ceiling down, by different | |
559 | * masks at different levels, in order to test whether a table | |
560 | * now has no other vmas using it, so can be freed, we don't | |
561 | * bother to round floor or end up - the tests don't need that. | |
562 | */ | |
1da177e4 | 563 | |
e0da382c HD |
564 | addr &= PMD_MASK; |
565 | if (addr < floor) { | |
566 | addr += PMD_SIZE; | |
567 | if (!addr) | |
568 | return; | |
569 | } | |
570 | if (ceiling) { | |
571 | ceiling &= PMD_MASK; | |
572 | if (!ceiling) | |
573 | return; | |
574 | } | |
575 | if (end - 1 > ceiling - 1) | |
576 | end -= PMD_SIZE; | |
577 | if (addr > end - 1) | |
578 | return; | |
07e32661 AK |
579 | /* |
580 | * We add page table cache pages with PAGE_SIZE, | |
581 | * (see pte_free_tlb()), flush the tlb if we need | |
582 | */ | |
583 | tlb_remove_check_page_size_change(tlb, PAGE_SIZE); | |
42b77728 | 584 | pgd = pgd_offset(tlb->mm, addr); |
1da177e4 LT |
585 | do { |
586 | next = pgd_addr_end(addr, end); | |
587 | if (pgd_none_or_clear_bad(pgd)) | |
588 | continue; | |
c2febafc | 589 | free_p4d_range(tlb, pgd, addr, next, floor, ceiling); |
1da177e4 | 590 | } while (pgd++, addr = next, addr != end); |
e0da382c HD |
591 | } |
592 | ||
42b77728 | 593 | void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, |
3bf5ee95 | 594 | unsigned long floor, unsigned long ceiling) |
e0da382c HD |
595 | { |
596 | while (vma) { | |
597 | struct vm_area_struct *next = vma->vm_next; | |
598 | unsigned long addr = vma->vm_start; | |
599 | ||
8f4f8c16 | 600 | /* |
25d9e2d1 NP |
601 | * Hide vma from rmap and truncate_pagecache before freeing |
602 | * pgtables | |
8f4f8c16 | 603 | */ |
5beb4930 | 604 | unlink_anon_vmas(vma); |
8f4f8c16 HD |
605 | unlink_file_vma(vma); |
606 | ||
9da61aef | 607 | if (is_vm_hugetlb_page(vma)) { |
3bf5ee95 | 608 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
166f61b9 | 609 | floor, next ? next->vm_start : ceiling); |
3bf5ee95 HD |
610 | } else { |
611 | /* | |
612 | * Optimization: gather nearby vmas into one call down | |
613 | */ | |
614 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE | |
4866920b | 615 | && !is_vm_hugetlb_page(next)) { |
3bf5ee95 HD |
616 | vma = next; |
617 | next = vma->vm_next; | |
5beb4930 | 618 | unlink_anon_vmas(vma); |
8f4f8c16 | 619 | unlink_file_vma(vma); |
3bf5ee95 HD |
620 | } |
621 | free_pgd_range(tlb, addr, vma->vm_end, | |
166f61b9 | 622 | floor, next ? next->vm_start : ceiling); |
3bf5ee95 | 623 | } |
e0da382c HD |
624 | vma = next; |
625 | } | |
1da177e4 LT |
626 | } |
627 | ||
3ed3a4f0 | 628 | int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address) |
1da177e4 | 629 | { |
c4088ebd | 630 | spinlock_t *ptl; |
2f569afd | 631 | pgtable_t new = pte_alloc_one(mm, address); |
1bb3630e HD |
632 | if (!new) |
633 | return -ENOMEM; | |
634 | ||
362a61ad NP |
635 | /* |
636 | * Ensure all pte setup (eg. pte page lock and page clearing) are | |
637 | * visible before the pte is made visible to other CPUs by being | |
638 | * put into page tables. | |
639 | * | |
640 | * The other side of the story is the pointer chasing in the page | |
641 | * table walking code (when walking the page table without locking; | |
642 | * ie. most of the time). Fortunately, these data accesses consist | |
643 | * of a chain of data-dependent loads, meaning most CPUs (alpha | |
644 | * being the notable exception) will already guarantee loads are | |
645 | * seen in-order. See the alpha page table accessors for the | |
646 | * smp_read_barrier_depends() barriers in page table walking code. | |
647 | */ | |
648 | smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */ | |
649 | ||
c4088ebd | 650 | ptl = pmd_lock(mm, pmd); |
8ac1f832 | 651 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
e1f56c89 | 652 | atomic_long_inc(&mm->nr_ptes); |
1da177e4 | 653 | pmd_populate(mm, pmd, new); |
2f569afd | 654 | new = NULL; |
4b471e88 | 655 | } |
c4088ebd | 656 | spin_unlock(ptl); |
2f569afd MS |
657 | if (new) |
658 | pte_free(mm, new); | |
1bb3630e | 659 | return 0; |
1da177e4 LT |
660 | } |
661 | ||
1bb3630e | 662 | int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) |
1da177e4 | 663 | { |
1bb3630e HD |
664 | pte_t *new = pte_alloc_one_kernel(&init_mm, address); |
665 | if (!new) | |
666 | return -ENOMEM; | |
667 | ||
362a61ad NP |
668 | smp_wmb(); /* See comment in __pte_alloc */ |
669 | ||
1bb3630e | 670 | spin_lock(&init_mm.page_table_lock); |
8ac1f832 | 671 | if (likely(pmd_none(*pmd))) { /* Has another populated it ? */ |
1bb3630e | 672 | pmd_populate_kernel(&init_mm, pmd, new); |
2f569afd | 673 | new = NULL; |
4b471e88 | 674 | } |
1bb3630e | 675 | spin_unlock(&init_mm.page_table_lock); |
2f569afd MS |
676 | if (new) |
677 | pte_free_kernel(&init_mm, new); | |
1bb3630e | 678 | return 0; |
1da177e4 LT |
679 | } |
680 | ||
d559db08 KH |
681 | static inline void init_rss_vec(int *rss) |
682 | { | |
683 | memset(rss, 0, sizeof(int) * NR_MM_COUNTERS); | |
684 | } | |
685 | ||
686 | static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss) | |
ae859762 | 687 | { |
d559db08 KH |
688 | int i; |
689 | ||
34e55232 | 690 | if (current->mm == mm) |
05af2e10 | 691 | sync_mm_rss(mm); |
d559db08 KH |
692 | for (i = 0; i < NR_MM_COUNTERS; i++) |
693 | if (rss[i]) | |
694 | add_mm_counter(mm, i, rss[i]); | |
ae859762 HD |
695 | } |
696 | ||
b5810039 | 697 | /* |
6aab341e LT |
698 | * This function is called to print an error when a bad pte |
699 | * is found. For example, we might have a PFN-mapped pte in | |
700 | * a region that doesn't allow it. | |
b5810039 NP |
701 | * |
702 | * The calling function must still handle the error. | |
703 | */ | |
3dc14741 HD |
704 | static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, |
705 | pte_t pte, struct page *page) | |
b5810039 | 706 | { |
3dc14741 | 707 | pgd_t *pgd = pgd_offset(vma->vm_mm, addr); |
c2febafc KS |
708 | p4d_t *p4d = p4d_offset(pgd, addr); |
709 | pud_t *pud = pud_offset(p4d, addr); | |
3dc14741 HD |
710 | pmd_t *pmd = pmd_offset(pud, addr); |
711 | struct address_space *mapping; | |
712 | pgoff_t index; | |
d936cf9b HD |
713 | static unsigned long resume; |
714 | static unsigned long nr_shown; | |
715 | static unsigned long nr_unshown; | |
716 | ||
717 | /* | |
718 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
719 | * or allow a steady drip of one report per second. | |
720 | */ | |
721 | if (nr_shown == 60) { | |
722 | if (time_before(jiffies, resume)) { | |
723 | nr_unshown++; | |
724 | return; | |
725 | } | |
726 | if (nr_unshown) { | |
1170532b JP |
727 | pr_alert("BUG: Bad page map: %lu messages suppressed\n", |
728 | nr_unshown); | |
d936cf9b HD |
729 | nr_unshown = 0; |
730 | } | |
731 | nr_shown = 0; | |
732 | } | |
733 | if (nr_shown++ == 0) | |
734 | resume = jiffies + 60 * HZ; | |
3dc14741 HD |
735 | |
736 | mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; | |
737 | index = linear_page_index(vma, addr); | |
738 | ||
1170532b JP |
739 | pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", |
740 | current->comm, | |
741 | (long long)pte_val(pte), (long long)pmd_val(*pmd)); | |
718a3821 | 742 | if (page) |
f0b791a3 | 743 | dump_page(page, "bad pte"); |
1170532b JP |
744 | pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n", |
745 | (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index); | |
3dc14741 HD |
746 | /* |
747 | * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y | |
748 | */ | |
2682582a KK |
749 | pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n", |
750 | vma->vm_file, | |
751 | vma->vm_ops ? vma->vm_ops->fault : NULL, | |
752 | vma->vm_file ? vma->vm_file->f_op->mmap : NULL, | |
753 | mapping ? mapping->a_ops->readpage : NULL); | |
b5810039 | 754 | dump_stack(); |
373d4d09 | 755 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
b5810039 NP |
756 | } |
757 | ||
ee498ed7 | 758 | /* |
7e675137 | 759 | * vm_normal_page -- This function gets the "struct page" associated with a pte. |
6aab341e | 760 | * |
7e675137 NP |
761 | * "Special" mappings do not wish to be associated with a "struct page" (either |
762 | * it doesn't exist, or it exists but they don't want to touch it). In this | |
763 | * case, NULL is returned here. "Normal" mappings do have a struct page. | |
b379d790 | 764 | * |
7e675137 NP |
765 | * There are 2 broad cases. Firstly, an architecture may define a pte_special() |
766 | * pte bit, in which case this function is trivial. Secondly, an architecture | |
767 | * may not have a spare pte bit, which requires a more complicated scheme, | |
768 | * described below. | |
769 | * | |
770 | * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a | |
771 | * special mapping (even if there are underlying and valid "struct pages"). | |
772 | * COWed pages of a VM_PFNMAP are always normal. | |
6aab341e | 773 | * |
b379d790 JH |
774 | * The way we recognize COWed pages within VM_PFNMAP mappings is through the |
775 | * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit | |
7e675137 NP |
776 | * set, and the vm_pgoff will point to the first PFN mapped: thus every special |
777 | * mapping will always honor the rule | |
6aab341e LT |
778 | * |
779 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) | |
780 | * | |
7e675137 NP |
781 | * And for normal mappings this is false. |
782 | * | |
783 | * This restricts such mappings to be a linear translation from virtual address | |
784 | * to pfn. To get around this restriction, we allow arbitrary mappings so long | |
785 | * as the vma is not a COW mapping; in that case, we know that all ptes are | |
786 | * special (because none can have been COWed). | |
b379d790 | 787 | * |
b379d790 | 788 | * |
7e675137 | 789 | * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. |
b379d790 JH |
790 | * |
791 | * VM_MIXEDMAP mappings can likewise contain memory with or without "struct | |
792 | * page" backing, however the difference is that _all_ pages with a struct | |
793 | * page (that is, those where pfn_valid is true) are refcounted and considered | |
794 | * normal pages by the VM. The disadvantage is that pages are refcounted | |
795 | * (which can be slower and simply not an option for some PFNMAP users). The | |
796 | * advantage is that we don't have to follow the strict linearity rule of | |
797 | * PFNMAP mappings in order to support COWable mappings. | |
798 | * | |
ee498ed7 | 799 | */ |
7e675137 NP |
800 | #ifdef __HAVE_ARCH_PTE_SPECIAL |
801 | # define HAVE_PTE_SPECIAL 1 | |
802 | #else | |
803 | # define HAVE_PTE_SPECIAL 0 | |
804 | #endif | |
805 | struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, | |
806 | pte_t pte) | |
ee498ed7 | 807 | { |
22b31eec | 808 | unsigned long pfn = pte_pfn(pte); |
7e675137 NP |
809 | |
810 | if (HAVE_PTE_SPECIAL) { | |
b38af472 | 811 | if (likely(!pte_special(pte))) |
22b31eec | 812 | goto check_pfn; |
667a0a06 DV |
813 | if (vma->vm_ops && vma->vm_ops->find_special_page) |
814 | return vma->vm_ops->find_special_page(vma, addr); | |
a13ea5b7 HD |
815 | if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)) |
816 | return NULL; | |
62eede62 | 817 | if (!is_zero_pfn(pfn)) |
22b31eec | 818 | print_bad_pte(vma, addr, pte, NULL); |
7e675137 NP |
819 | return NULL; |
820 | } | |
821 | ||
822 | /* !HAVE_PTE_SPECIAL case follows: */ | |
823 | ||
b379d790 JH |
824 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { |
825 | if (vma->vm_flags & VM_MIXEDMAP) { | |
826 | if (!pfn_valid(pfn)) | |
827 | return NULL; | |
828 | goto out; | |
829 | } else { | |
7e675137 NP |
830 | unsigned long off; |
831 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
b379d790 JH |
832 | if (pfn == vma->vm_pgoff + off) |
833 | return NULL; | |
834 | if (!is_cow_mapping(vma->vm_flags)) | |
835 | return NULL; | |
836 | } | |
6aab341e LT |
837 | } |
838 | ||
b38af472 HD |
839 | if (is_zero_pfn(pfn)) |
840 | return NULL; | |
22b31eec HD |
841 | check_pfn: |
842 | if (unlikely(pfn > highest_memmap_pfn)) { | |
843 | print_bad_pte(vma, addr, pte, NULL); | |
844 | return NULL; | |
845 | } | |
6aab341e LT |
846 | |
847 | /* | |
7e675137 | 848 | * NOTE! We still have PageReserved() pages in the page tables. |
7e675137 | 849 | * eg. VDSO mappings can cause them to exist. |
6aab341e | 850 | */ |
b379d790 | 851 | out: |
6aab341e | 852 | return pfn_to_page(pfn); |
ee498ed7 HD |
853 | } |
854 | ||
28093f9f GS |
855 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
856 | struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr, | |
857 | pmd_t pmd) | |
858 | { | |
859 | unsigned long pfn = pmd_pfn(pmd); | |
860 | ||
861 | /* | |
862 | * There is no pmd_special() but there may be special pmds, e.g. | |
863 | * in a direct-access (dax) mapping, so let's just replicate the | |
864 | * !HAVE_PTE_SPECIAL case from vm_normal_page() here. | |
865 | */ | |
866 | if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) { | |
867 | if (vma->vm_flags & VM_MIXEDMAP) { | |
868 | if (!pfn_valid(pfn)) | |
869 | return NULL; | |
870 | goto out; | |
871 | } else { | |
872 | unsigned long off; | |
873 | off = (addr - vma->vm_start) >> PAGE_SHIFT; | |
874 | if (pfn == vma->vm_pgoff + off) | |
875 | return NULL; | |
876 | if (!is_cow_mapping(vma->vm_flags)) | |
877 | return NULL; | |
878 | } | |
879 | } | |
880 | ||
881 | if (is_zero_pfn(pfn)) | |
882 | return NULL; | |
883 | if (unlikely(pfn > highest_memmap_pfn)) | |
884 | return NULL; | |
885 | ||
886 | /* | |
887 | * NOTE! We still have PageReserved() pages in the page tables. | |
888 | * eg. VDSO mappings can cause them to exist. | |
889 | */ | |
890 | out: | |
891 | return pfn_to_page(pfn); | |
892 | } | |
893 | #endif | |
894 | ||
1da177e4 LT |
895 | /* |
896 | * copy one vm_area from one task to the other. Assumes the page tables | |
897 | * already present in the new task to be cleared in the whole range | |
898 | * covered by this vma. | |
1da177e4 LT |
899 | */ |
900 | ||
570a335b | 901 | static inline unsigned long |
1da177e4 | 902 | copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
b5810039 | 903 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, |
8c103762 | 904 | unsigned long addr, int *rss) |
1da177e4 | 905 | { |
b5810039 | 906 | unsigned long vm_flags = vma->vm_flags; |
1da177e4 LT |
907 | pte_t pte = *src_pte; |
908 | struct page *page; | |
1da177e4 LT |
909 | |
910 | /* pte contains position in swap or file, so copy. */ | |
911 | if (unlikely(!pte_present(pte))) { | |
0661a336 KS |
912 | swp_entry_t entry = pte_to_swp_entry(pte); |
913 | ||
914 | if (likely(!non_swap_entry(entry))) { | |
915 | if (swap_duplicate(entry) < 0) | |
916 | return entry.val; | |
917 | ||
918 | /* make sure dst_mm is on swapoff's mmlist. */ | |
919 | if (unlikely(list_empty(&dst_mm->mmlist))) { | |
920 | spin_lock(&mmlist_lock); | |
921 | if (list_empty(&dst_mm->mmlist)) | |
922 | list_add(&dst_mm->mmlist, | |
923 | &src_mm->mmlist); | |
924 | spin_unlock(&mmlist_lock); | |
925 | } | |
926 | rss[MM_SWAPENTS]++; | |
927 | } else if (is_migration_entry(entry)) { | |
928 | page = migration_entry_to_page(entry); | |
929 | ||
eca56ff9 | 930 | rss[mm_counter(page)]++; |
0661a336 KS |
931 | |
932 | if (is_write_migration_entry(entry) && | |
933 | is_cow_mapping(vm_flags)) { | |
934 | /* | |
935 | * COW mappings require pages in both | |
936 | * parent and child to be set to read. | |
937 | */ | |
938 | make_migration_entry_read(&entry); | |
939 | pte = swp_entry_to_pte(entry); | |
940 | if (pte_swp_soft_dirty(*src_pte)) | |
941 | pte = pte_swp_mksoft_dirty(pte); | |
942 | set_pte_at(src_mm, addr, src_pte, pte); | |
0697212a | 943 | } |
1da177e4 | 944 | } |
ae859762 | 945 | goto out_set_pte; |
1da177e4 LT |
946 | } |
947 | ||
1da177e4 LT |
948 | /* |
949 | * If it's a COW mapping, write protect it both | |
950 | * in the parent and the child | |
951 | */ | |
67121172 | 952 | if (is_cow_mapping(vm_flags)) { |
1da177e4 | 953 | ptep_set_wrprotect(src_mm, addr, src_pte); |
3dc90795 | 954 | pte = pte_wrprotect(pte); |
1da177e4 LT |
955 | } |
956 | ||
957 | /* | |
958 | * If it's a shared mapping, mark it clean in | |
959 | * the child | |
960 | */ | |
961 | if (vm_flags & VM_SHARED) | |
962 | pte = pte_mkclean(pte); | |
963 | pte = pte_mkold(pte); | |
6aab341e LT |
964 | |
965 | page = vm_normal_page(vma, addr, pte); | |
966 | if (page) { | |
967 | get_page(page); | |
53f9263b | 968 | page_dup_rmap(page, false); |
eca56ff9 | 969 | rss[mm_counter(page)]++; |
6aab341e | 970 | } |
ae859762 HD |
971 | |
972 | out_set_pte: | |
973 | set_pte_at(dst_mm, addr, dst_pte, pte); | |
570a335b | 974 | return 0; |
1da177e4 LT |
975 | } |
976 | ||
21bda264 | 977 | static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
71e3aac0 AA |
978 | pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, |
979 | unsigned long addr, unsigned long end) | |
1da177e4 | 980 | { |
c36987e2 | 981 | pte_t *orig_src_pte, *orig_dst_pte; |
1da177e4 | 982 | pte_t *src_pte, *dst_pte; |
c74df32c | 983 | spinlock_t *src_ptl, *dst_ptl; |
e040f218 | 984 | int progress = 0; |
d559db08 | 985 | int rss[NR_MM_COUNTERS]; |
570a335b | 986 | swp_entry_t entry = (swp_entry_t){0}; |
1da177e4 LT |
987 | |
988 | again: | |
d559db08 KH |
989 | init_rss_vec(rss); |
990 | ||
c74df32c | 991 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
1da177e4 LT |
992 | if (!dst_pte) |
993 | return -ENOMEM; | |
ece0e2b6 | 994 | src_pte = pte_offset_map(src_pmd, addr); |
4c21e2f2 | 995 | src_ptl = pte_lockptr(src_mm, src_pmd); |
f20dc5f7 | 996 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
c36987e2 DN |
997 | orig_src_pte = src_pte; |
998 | orig_dst_pte = dst_pte; | |
6606c3e0 | 999 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 1000 | |
1da177e4 LT |
1001 | do { |
1002 | /* | |
1003 | * We are holding two locks at this point - either of them | |
1004 | * could generate latencies in another task on another CPU. | |
1005 | */ | |
e040f218 HD |
1006 | if (progress >= 32) { |
1007 | progress = 0; | |
1008 | if (need_resched() || | |
95c354fe | 1009 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
e040f218 HD |
1010 | break; |
1011 | } | |
1da177e4 LT |
1012 | if (pte_none(*src_pte)) { |
1013 | progress++; | |
1014 | continue; | |
1015 | } | |
570a335b HD |
1016 | entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, |
1017 | vma, addr, rss); | |
1018 | if (entry.val) | |
1019 | break; | |
1da177e4 LT |
1020 | progress += 8; |
1021 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); | |
1da177e4 | 1022 | |
6606c3e0 | 1023 | arch_leave_lazy_mmu_mode(); |
c74df32c | 1024 | spin_unlock(src_ptl); |
ece0e2b6 | 1025 | pte_unmap(orig_src_pte); |
d559db08 | 1026 | add_mm_rss_vec(dst_mm, rss); |
c36987e2 | 1027 | pte_unmap_unlock(orig_dst_pte, dst_ptl); |
c74df32c | 1028 | cond_resched(); |
570a335b HD |
1029 | |
1030 | if (entry.val) { | |
1031 | if (add_swap_count_continuation(entry, GFP_KERNEL) < 0) | |
1032 | return -ENOMEM; | |
1033 | progress = 0; | |
1034 | } | |
1da177e4 LT |
1035 | if (addr != end) |
1036 | goto again; | |
1037 | return 0; | |
1038 | } | |
1039 | ||
1040 | static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
1041 | pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, | |
1042 | unsigned long addr, unsigned long end) | |
1043 | { | |
1044 | pmd_t *src_pmd, *dst_pmd; | |
1045 | unsigned long next; | |
1046 | ||
1047 | dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); | |
1048 | if (!dst_pmd) | |
1049 | return -ENOMEM; | |
1050 | src_pmd = pmd_offset(src_pud, addr); | |
1051 | do { | |
1052 | next = pmd_addr_end(addr, end); | |
5c7fb56e | 1053 | if (pmd_trans_huge(*src_pmd) || pmd_devmap(*src_pmd)) { |
71e3aac0 | 1054 | int err; |
a00cc7d9 | 1055 | VM_BUG_ON_VMA(next-addr != HPAGE_PMD_SIZE, vma); |
71e3aac0 AA |
1056 | err = copy_huge_pmd(dst_mm, src_mm, |
1057 | dst_pmd, src_pmd, addr, vma); | |
1058 | if (err == -ENOMEM) | |
1059 | return -ENOMEM; | |
1060 | if (!err) | |
1061 | continue; | |
1062 | /* fall through */ | |
1063 | } | |
1da177e4 LT |
1064 | if (pmd_none_or_clear_bad(src_pmd)) |
1065 | continue; | |
1066 | if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, | |
1067 | vma, addr, next)) | |
1068 | return -ENOMEM; | |
1069 | } while (dst_pmd++, src_pmd++, addr = next, addr != end); | |
1070 | return 0; | |
1071 | } | |
1072 | ||
1073 | static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, | |
c2febafc | 1074 | p4d_t *dst_p4d, p4d_t *src_p4d, struct vm_area_struct *vma, |
1da177e4 LT |
1075 | unsigned long addr, unsigned long end) |
1076 | { | |
1077 | pud_t *src_pud, *dst_pud; | |
1078 | unsigned long next; | |
1079 | ||
c2febafc | 1080 | dst_pud = pud_alloc(dst_mm, dst_p4d, addr); |
1da177e4 LT |
1081 | if (!dst_pud) |
1082 | return -ENOMEM; | |
c2febafc | 1083 | src_pud = pud_offset(src_p4d, addr); |
1da177e4 LT |
1084 | do { |
1085 | next = pud_addr_end(addr, end); | |
a00cc7d9 MW |
1086 | if (pud_trans_huge(*src_pud) || pud_devmap(*src_pud)) { |
1087 | int err; | |
1088 | ||
1089 | VM_BUG_ON_VMA(next-addr != HPAGE_PUD_SIZE, vma); | |
1090 | err = copy_huge_pud(dst_mm, src_mm, | |
1091 | dst_pud, src_pud, addr, vma); | |
1092 | if (err == -ENOMEM) | |
1093 | return -ENOMEM; | |
1094 | if (!err) | |
1095 | continue; | |
1096 | /* fall through */ | |
1097 | } | |
1da177e4 LT |
1098 | if (pud_none_or_clear_bad(src_pud)) |
1099 | continue; | |
1100 | if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, | |
1101 | vma, addr, next)) | |
1102 | return -ENOMEM; | |
1103 | } while (dst_pud++, src_pud++, addr = next, addr != end); | |
1104 | return 0; | |
1105 | } | |
1106 | ||
c2febafc KS |
1107 | static inline int copy_p4d_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
1108 | pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, | |
1109 | unsigned long addr, unsigned long end) | |
1110 | { | |
1111 | p4d_t *src_p4d, *dst_p4d; | |
1112 | unsigned long next; | |
1113 | ||
1114 | dst_p4d = p4d_alloc(dst_mm, dst_pgd, addr); | |
1115 | if (!dst_p4d) | |
1116 | return -ENOMEM; | |
1117 | src_p4d = p4d_offset(src_pgd, addr); | |
1118 | do { | |
1119 | next = p4d_addr_end(addr, end); | |
1120 | if (p4d_none_or_clear_bad(src_p4d)) | |
1121 | continue; | |
1122 | if (copy_pud_range(dst_mm, src_mm, dst_p4d, src_p4d, | |
1123 | vma, addr, next)) | |
1124 | return -ENOMEM; | |
1125 | } while (dst_p4d++, src_p4d++, addr = next, addr != end); | |
1126 | return 0; | |
1127 | } | |
1128 | ||
1da177e4 LT |
1129 | int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
1130 | struct vm_area_struct *vma) | |
1131 | { | |
1132 | pgd_t *src_pgd, *dst_pgd; | |
1133 | unsigned long next; | |
1134 | unsigned long addr = vma->vm_start; | |
1135 | unsigned long end = vma->vm_end; | |
2ec74c3e SG |
1136 | unsigned long mmun_start; /* For mmu_notifiers */ |
1137 | unsigned long mmun_end; /* For mmu_notifiers */ | |
1138 | bool is_cow; | |
cddb8a5c | 1139 | int ret; |
1da177e4 | 1140 | |
d992895b NP |
1141 | /* |
1142 | * Don't copy ptes where a page fault will fill them correctly. | |
1143 | * Fork becomes much lighter when there are big shared or private | |
1144 | * readonly mappings. The tradeoff is that copy_page_range is more | |
1145 | * efficient than faulting. | |
1146 | */ | |
0661a336 KS |
1147 | if (!(vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) && |
1148 | !vma->anon_vma) | |
1149 | return 0; | |
d992895b | 1150 | |
1da177e4 LT |
1151 | if (is_vm_hugetlb_page(vma)) |
1152 | return copy_hugetlb_page_range(dst_mm, src_mm, vma); | |
1153 | ||
b3b9c293 | 1154 | if (unlikely(vma->vm_flags & VM_PFNMAP)) { |
2ab64037 | 1155 | /* |
1156 | * We do not free on error cases below as remove_vma | |
1157 | * gets called on error from higher level routine | |
1158 | */ | |
5180da41 | 1159 | ret = track_pfn_copy(vma); |
2ab64037 | 1160 | if (ret) |
1161 | return ret; | |
1162 | } | |
1163 | ||
cddb8a5c AA |
1164 | /* |
1165 | * We need to invalidate the secondary MMU mappings only when | |
1166 | * there could be a permission downgrade on the ptes of the | |
1167 | * parent mm. And a permission downgrade will only happen if | |
1168 | * is_cow_mapping() returns true. | |
1169 | */ | |
2ec74c3e SG |
1170 | is_cow = is_cow_mapping(vma->vm_flags); |
1171 | mmun_start = addr; | |
1172 | mmun_end = end; | |
1173 | if (is_cow) | |
1174 | mmu_notifier_invalidate_range_start(src_mm, mmun_start, | |
1175 | mmun_end); | |
cddb8a5c AA |
1176 | |
1177 | ret = 0; | |
1da177e4 LT |
1178 | dst_pgd = pgd_offset(dst_mm, addr); |
1179 | src_pgd = pgd_offset(src_mm, addr); | |
1180 | do { | |
1181 | next = pgd_addr_end(addr, end); | |
1182 | if (pgd_none_or_clear_bad(src_pgd)) | |
1183 | continue; | |
c2febafc | 1184 | if (unlikely(copy_p4d_range(dst_mm, src_mm, dst_pgd, src_pgd, |
cddb8a5c AA |
1185 | vma, addr, next))) { |
1186 | ret = -ENOMEM; | |
1187 | break; | |
1188 | } | |
1da177e4 | 1189 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); |
cddb8a5c | 1190 | |
2ec74c3e SG |
1191 | if (is_cow) |
1192 | mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end); | |
cddb8a5c | 1193 | return ret; |
1da177e4 LT |
1194 | } |
1195 | ||
51c6f666 | 1196 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
b5810039 | 1197 | struct vm_area_struct *vma, pmd_t *pmd, |
1da177e4 | 1198 | unsigned long addr, unsigned long end, |
97a89413 | 1199 | struct zap_details *details) |
1da177e4 | 1200 | { |
b5810039 | 1201 | struct mm_struct *mm = tlb->mm; |
d16dfc55 | 1202 | int force_flush = 0; |
d559db08 | 1203 | int rss[NR_MM_COUNTERS]; |
97a89413 | 1204 | spinlock_t *ptl; |
5f1a1907 | 1205 | pte_t *start_pte; |
97a89413 | 1206 | pte_t *pte; |
8a5f14a2 | 1207 | swp_entry_t entry; |
d559db08 | 1208 | |
07e32661 | 1209 | tlb_remove_check_page_size_change(tlb, PAGE_SIZE); |
d16dfc55 | 1210 | again: |
e303297e | 1211 | init_rss_vec(rss); |
5f1a1907 SR |
1212 | start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
1213 | pte = start_pte; | |
3ea27719 | 1214 | flush_tlb_batched_pending(mm); |
6606c3e0 | 1215 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1216 | do { |
1217 | pte_t ptent = *pte; | |
166f61b9 | 1218 | if (pte_none(ptent)) |
1da177e4 | 1219 | continue; |
6f5e6b9e | 1220 | |
1da177e4 | 1221 | if (pte_present(ptent)) { |
ee498ed7 | 1222 | struct page *page; |
51c6f666 | 1223 | |
6aab341e | 1224 | page = vm_normal_page(vma, addr, ptent); |
1da177e4 LT |
1225 | if (unlikely(details) && page) { |
1226 | /* | |
1227 | * unmap_shared_mapping_pages() wants to | |
1228 | * invalidate cache without truncating: | |
1229 | * unmap shared but keep private pages. | |
1230 | */ | |
1231 | if (details->check_mapping && | |
800d8c63 | 1232 | details->check_mapping != page_rmapping(page)) |
1da177e4 | 1233 | continue; |
1da177e4 | 1234 | } |
b5810039 | 1235 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
a600388d | 1236 | tlb->fullmm); |
1da177e4 LT |
1237 | tlb_remove_tlb_entry(tlb, pte, addr); |
1238 | if (unlikely(!page)) | |
1239 | continue; | |
eca56ff9 JM |
1240 | |
1241 | if (!PageAnon(page)) { | |
1cf35d47 LT |
1242 | if (pte_dirty(ptent)) { |
1243 | force_flush = 1; | |
6237bcd9 | 1244 | set_page_dirty(page); |
1cf35d47 | 1245 | } |
4917e5d0 | 1246 | if (pte_young(ptent) && |
64363aad | 1247 | likely(!(vma->vm_flags & VM_SEQ_READ))) |
bf3f3bc5 | 1248 | mark_page_accessed(page); |
6237bcd9 | 1249 | } |
eca56ff9 | 1250 | rss[mm_counter(page)]--; |
d281ee61 | 1251 | page_remove_rmap(page, false); |
3dc14741 HD |
1252 | if (unlikely(page_mapcount(page) < 0)) |
1253 | print_bad_pte(vma, addr, ptent, page); | |
e9d55e15 | 1254 | if (unlikely(__tlb_remove_page(tlb, page))) { |
1cf35d47 | 1255 | force_flush = 1; |
ce9ec37b | 1256 | addr += PAGE_SIZE; |
d16dfc55 | 1257 | break; |
1cf35d47 | 1258 | } |
1da177e4 LT |
1259 | continue; |
1260 | } | |
3e8715fd KS |
1261 | /* If details->check_mapping, we leave swap entries. */ |
1262 | if (unlikely(details)) | |
1da177e4 | 1263 | continue; |
b084d435 | 1264 | |
8a5f14a2 KS |
1265 | entry = pte_to_swp_entry(ptent); |
1266 | if (!non_swap_entry(entry)) | |
1267 | rss[MM_SWAPENTS]--; | |
1268 | else if (is_migration_entry(entry)) { | |
1269 | struct page *page; | |
9f9f1acd | 1270 | |
8a5f14a2 | 1271 | page = migration_entry_to_page(entry); |
eca56ff9 | 1272 | rss[mm_counter(page)]--; |
b084d435 | 1273 | } |
8a5f14a2 KS |
1274 | if (unlikely(!free_swap_and_cache(entry))) |
1275 | print_bad_pte(vma, addr, ptent, NULL); | |
9888a1ca | 1276 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
97a89413 | 1277 | } while (pte++, addr += PAGE_SIZE, addr != end); |
ae859762 | 1278 | |
d559db08 | 1279 | add_mm_rss_vec(mm, rss); |
6606c3e0 | 1280 | arch_leave_lazy_mmu_mode(); |
51c6f666 | 1281 | |
1cf35d47 | 1282 | /* Do the actual TLB flush before dropping ptl */ |
fb7332a9 | 1283 | if (force_flush) |
1cf35d47 | 1284 | tlb_flush_mmu_tlbonly(tlb); |
1cf35d47 LT |
1285 | pte_unmap_unlock(start_pte, ptl); |
1286 | ||
1287 | /* | |
1288 | * If we forced a TLB flush (either due to running out of | |
1289 | * batch buffers or because we needed to flush dirty TLB | |
1290 | * entries before releasing the ptl), free the batched | |
1291 | * memory too. Restart if we didn't do everything. | |
1292 | */ | |
1293 | if (force_flush) { | |
1294 | force_flush = 0; | |
1295 | tlb_flush_mmu_free(tlb); | |
2b047252 | 1296 | if (addr != end) |
d16dfc55 PZ |
1297 | goto again; |
1298 | } | |
1299 | ||
51c6f666 | 1300 | return addr; |
1da177e4 LT |
1301 | } |
1302 | ||
51c6f666 | 1303 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
b5810039 | 1304 | struct vm_area_struct *vma, pud_t *pud, |
1da177e4 | 1305 | unsigned long addr, unsigned long end, |
97a89413 | 1306 | struct zap_details *details) |
1da177e4 LT |
1307 | { |
1308 | pmd_t *pmd; | |
1309 | unsigned long next; | |
1310 | ||
1311 | pmd = pmd_offset(pud, addr); | |
1312 | do { | |
1313 | next = pmd_addr_end(addr, end); | |
5c7fb56e | 1314 | if (pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { |
1a5a9906 | 1315 | if (next - addr != HPAGE_PMD_SIZE) { |
68428398 HD |
1316 | VM_BUG_ON_VMA(vma_is_anonymous(vma) && |
1317 | !rwsem_is_locked(&tlb->mm->mmap_sem), vma); | |
fd60775a | 1318 | __split_huge_pmd(vma, pmd, addr, false, NULL); |
f21760b1 | 1319 | } else if (zap_huge_pmd(tlb, vma, pmd, addr)) |
1a5a9906 | 1320 | goto next; |
71e3aac0 AA |
1321 | /* fall through */ |
1322 | } | |
1a5a9906 AA |
1323 | /* |
1324 | * Here there can be other concurrent MADV_DONTNEED or | |
1325 | * trans huge page faults running, and if the pmd is | |
1326 | * none or trans huge it can change under us. This is | |
1327 | * because MADV_DONTNEED holds the mmap_sem in read | |
1328 | * mode. | |
1329 | */ | |
1330 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) | |
1331 | goto next; | |
97a89413 | 1332 | next = zap_pte_range(tlb, vma, pmd, addr, next, details); |
1a5a9906 | 1333 | next: |
97a89413 PZ |
1334 | cond_resched(); |
1335 | } while (pmd++, addr = next, addr != end); | |
51c6f666 RH |
1336 | |
1337 | return addr; | |
1da177e4 LT |
1338 | } |
1339 | ||
51c6f666 | 1340 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
c2febafc | 1341 | struct vm_area_struct *vma, p4d_t *p4d, |
1da177e4 | 1342 | unsigned long addr, unsigned long end, |
97a89413 | 1343 | struct zap_details *details) |
1da177e4 LT |
1344 | { |
1345 | pud_t *pud; | |
1346 | unsigned long next; | |
1347 | ||
c2febafc | 1348 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
1349 | do { |
1350 | next = pud_addr_end(addr, end); | |
a00cc7d9 MW |
1351 | if (pud_trans_huge(*pud) || pud_devmap(*pud)) { |
1352 | if (next - addr != HPAGE_PUD_SIZE) { | |
1353 | VM_BUG_ON_VMA(!rwsem_is_locked(&tlb->mm->mmap_sem), vma); | |
1354 | split_huge_pud(vma, pud, addr); | |
1355 | } else if (zap_huge_pud(tlb, vma, pud, addr)) | |
1356 | goto next; | |
1357 | /* fall through */ | |
1358 | } | |
97a89413 | 1359 | if (pud_none_or_clear_bad(pud)) |
1da177e4 | 1360 | continue; |
97a89413 | 1361 | next = zap_pmd_range(tlb, vma, pud, addr, next, details); |
a00cc7d9 MW |
1362 | next: |
1363 | cond_resched(); | |
97a89413 | 1364 | } while (pud++, addr = next, addr != end); |
51c6f666 RH |
1365 | |
1366 | return addr; | |
1da177e4 LT |
1367 | } |
1368 | ||
c2febafc KS |
1369 | static inline unsigned long zap_p4d_range(struct mmu_gather *tlb, |
1370 | struct vm_area_struct *vma, pgd_t *pgd, | |
1371 | unsigned long addr, unsigned long end, | |
1372 | struct zap_details *details) | |
1373 | { | |
1374 | p4d_t *p4d; | |
1375 | unsigned long next; | |
1376 | ||
1377 | p4d = p4d_offset(pgd, addr); | |
1378 | do { | |
1379 | next = p4d_addr_end(addr, end); | |
1380 | if (p4d_none_or_clear_bad(p4d)) | |
1381 | continue; | |
1382 | next = zap_pud_range(tlb, vma, p4d, addr, next, details); | |
1383 | } while (p4d++, addr = next, addr != end); | |
1384 | ||
1385 | return addr; | |
1386 | } | |
1387 | ||
aac45363 | 1388 | void unmap_page_range(struct mmu_gather *tlb, |
038c7aa1 AV |
1389 | struct vm_area_struct *vma, |
1390 | unsigned long addr, unsigned long end, | |
1391 | struct zap_details *details) | |
1da177e4 LT |
1392 | { |
1393 | pgd_t *pgd; | |
1394 | unsigned long next; | |
1395 | ||
1da177e4 LT |
1396 | BUG_ON(addr >= end); |
1397 | tlb_start_vma(tlb, vma); | |
1398 | pgd = pgd_offset(vma->vm_mm, addr); | |
1399 | do { | |
1400 | next = pgd_addr_end(addr, end); | |
97a89413 | 1401 | if (pgd_none_or_clear_bad(pgd)) |
1da177e4 | 1402 | continue; |
c2febafc | 1403 | next = zap_p4d_range(tlb, vma, pgd, addr, next, details); |
97a89413 | 1404 | } while (pgd++, addr = next, addr != end); |
1da177e4 LT |
1405 | tlb_end_vma(tlb, vma); |
1406 | } | |
51c6f666 | 1407 | |
f5cc4eef AV |
1408 | |
1409 | static void unmap_single_vma(struct mmu_gather *tlb, | |
1410 | struct vm_area_struct *vma, unsigned long start_addr, | |
4f74d2c8 | 1411 | unsigned long end_addr, |
f5cc4eef AV |
1412 | struct zap_details *details) |
1413 | { | |
1414 | unsigned long start = max(vma->vm_start, start_addr); | |
1415 | unsigned long end; | |
1416 | ||
1417 | if (start >= vma->vm_end) | |
1418 | return; | |
1419 | end = min(vma->vm_end, end_addr); | |
1420 | if (end <= vma->vm_start) | |
1421 | return; | |
1422 | ||
cbc91f71 SD |
1423 | if (vma->vm_file) |
1424 | uprobe_munmap(vma, start, end); | |
1425 | ||
b3b9c293 | 1426 | if (unlikely(vma->vm_flags & VM_PFNMAP)) |
5180da41 | 1427 | untrack_pfn(vma, 0, 0); |
f5cc4eef AV |
1428 | |
1429 | if (start != end) { | |
1430 | if (unlikely(is_vm_hugetlb_page(vma))) { | |
1431 | /* | |
1432 | * It is undesirable to test vma->vm_file as it | |
1433 | * should be non-null for valid hugetlb area. | |
1434 | * However, vm_file will be NULL in the error | |
7aa6b4ad | 1435 | * cleanup path of mmap_region. When |
f5cc4eef | 1436 | * hugetlbfs ->mmap method fails, |
7aa6b4ad | 1437 | * mmap_region() nullifies vma->vm_file |
f5cc4eef AV |
1438 | * before calling this function to clean up. |
1439 | * Since no pte has actually been setup, it is | |
1440 | * safe to do nothing in this case. | |
1441 | */ | |
24669e58 | 1442 | if (vma->vm_file) { |
83cde9e8 | 1443 | i_mmap_lock_write(vma->vm_file->f_mapping); |
d833352a | 1444 | __unmap_hugepage_range_final(tlb, vma, start, end, NULL); |
83cde9e8 | 1445 | i_mmap_unlock_write(vma->vm_file->f_mapping); |
24669e58 | 1446 | } |
f5cc4eef AV |
1447 | } else |
1448 | unmap_page_range(tlb, vma, start, end, details); | |
1449 | } | |
1da177e4 LT |
1450 | } |
1451 | ||
1da177e4 LT |
1452 | /** |
1453 | * unmap_vmas - unmap a range of memory covered by a list of vma's | |
0164f69d | 1454 | * @tlb: address of the caller's struct mmu_gather |
1da177e4 LT |
1455 | * @vma: the starting vma |
1456 | * @start_addr: virtual address at which to start unmapping | |
1457 | * @end_addr: virtual address at which to end unmapping | |
1da177e4 | 1458 | * |
508034a3 | 1459 | * Unmap all pages in the vma list. |
1da177e4 | 1460 | * |
1da177e4 LT |
1461 | * Only addresses between `start' and `end' will be unmapped. |
1462 | * | |
1463 | * The VMA list must be sorted in ascending virtual address order. | |
1464 | * | |
1465 | * unmap_vmas() assumes that the caller will flush the whole unmapped address | |
1466 | * range after unmap_vmas() returns. So the only responsibility here is to | |
1467 | * ensure that any thus-far unmapped pages are flushed before unmap_vmas() | |
1468 | * drops the lock and schedules. | |
1469 | */ | |
6e8bb019 | 1470 | void unmap_vmas(struct mmu_gather *tlb, |
1da177e4 | 1471 | struct vm_area_struct *vma, unsigned long start_addr, |
4f74d2c8 | 1472 | unsigned long end_addr) |
1da177e4 | 1473 | { |
cddb8a5c | 1474 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 1475 | |
cddb8a5c | 1476 | mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); |
f5cc4eef | 1477 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) |
4f74d2c8 | 1478 | unmap_single_vma(tlb, vma, start_addr, end_addr, NULL); |
cddb8a5c | 1479 | mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); |
1da177e4 LT |
1480 | } |
1481 | ||
1482 | /** | |
1483 | * zap_page_range - remove user pages in a given range | |
1484 | * @vma: vm_area_struct holding the applicable pages | |
eb4546bb | 1485 | * @start: starting address of pages to zap |
1da177e4 | 1486 | * @size: number of bytes to zap |
f5cc4eef AV |
1487 | * |
1488 | * Caller must protect the VMA list | |
1da177e4 | 1489 | */ |
7e027b14 | 1490 | void zap_page_range(struct vm_area_struct *vma, unsigned long start, |
ecf1385d | 1491 | unsigned long size) |
1da177e4 LT |
1492 | { |
1493 | struct mm_struct *mm = vma->vm_mm; | |
d16dfc55 | 1494 | struct mmu_gather tlb; |
7e027b14 | 1495 | unsigned long end = start + size; |
1da177e4 | 1496 | |
1da177e4 | 1497 | lru_add_drain(); |
2b047252 | 1498 | tlb_gather_mmu(&tlb, mm, start, end); |
365e9c87 | 1499 | update_hiwater_rss(mm); |
7e027b14 LT |
1500 | mmu_notifier_invalidate_range_start(mm, start, end); |
1501 | for ( ; vma && vma->vm_start < end; vma = vma->vm_next) | |
ecf1385d | 1502 | unmap_single_vma(&tlb, vma, start, end, NULL); |
7e027b14 LT |
1503 | mmu_notifier_invalidate_range_end(mm, start, end); |
1504 | tlb_finish_mmu(&tlb, start, end); | |
1da177e4 LT |
1505 | } |
1506 | ||
f5cc4eef AV |
1507 | /** |
1508 | * zap_page_range_single - remove user pages in a given range | |
1509 | * @vma: vm_area_struct holding the applicable pages | |
1510 | * @address: starting address of pages to zap | |
1511 | * @size: number of bytes to zap | |
8a5f14a2 | 1512 | * @details: details of shared cache invalidation |
f5cc4eef AV |
1513 | * |
1514 | * The range must fit into one VMA. | |
1da177e4 | 1515 | */ |
f5cc4eef | 1516 | static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address, |
1da177e4 LT |
1517 | unsigned long size, struct zap_details *details) |
1518 | { | |
1519 | struct mm_struct *mm = vma->vm_mm; | |
d16dfc55 | 1520 | struct mmu_gather tlb; |
1da177e4 | 1521 | unsigned long end = address + size; |
1da177e4 | 1522 | |
1da177e4 | 1523 | lru_add_drain(); |
2b047252 | 1524 | tlb_gather_mmu(&tlb, mm, address, end); |
365e9c87 | 1525 | update_hiwater_rss(mm); |
f5cc4eef | 1526 | mmu_notifier_invalidate_range_start(mm, address, end); |
4f74d2c8 | 1527 | unmap_single_vma(&tlb, vma, address, end, details); |
f5cc4eef | 1528 | mmu_notifier_invalidate_range_end(mm, address, end); |
d16dfc55 | 1529 | tlb_finish_mmu(&tlb, address, end); |
1da177e4 LT |
1530 | } |
1531 | ||
c627f9cc JS |
1532 | /** |
1533 | * zap_vma_ptes - remove ptes mapping the vma | |
1534 | * @vma: vm_area_struct holding ptes to be zapped | |
1535 | * @address: starting address of pages to zap | |
1536 | * @size: number of bytes to zap | |
1537 | * | |
1538 | * This function only unmaps ptes assigned to VM_PFNMAP vmas. | |
1539 | * | |
1540 | * The entire address range must be fully contained within the vma. | |
1541 | * | |
1542 | * Returns 0 if successful. | |
1543 | */ | |
1544 | int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, | |
1545 | unsigned long size) | |
1546 | { | |
1547 | if (address < vma->vm_start || address + size > vma->vm_end || | |
1548 | !(vma->vm_flags & VM_PFNMAP)) | |
1549 | return -1; | |
f5cc4eef | 1550 | zap_page_range_single(vma, address, size, NULL); |
c627f9cc JS |
1551 | return 0; |
1552 | } | |
1553 | EXPORT_SYMBOL_GPL(zap_vma_ptes); | |
1554 | ||
25ca1d6c | 1555 | pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr, |
920c7a5d | 1556 | spinlock_t **ptl) |
c9cfcddf | 1557 | { |
c2febafc KS |
1558 | pgd_t *pgd; |
1559 | p4d_t *p4d; | |
1560 | pud_t *pud; | |
1561 | pmd_t *pmd; | |
1562 | ||
1563 | pgd = pgd_offset(mm, addr); | |
1564 | p4d = p4d_alloc(mm, pgd, addr); | |
1565 | if (!p4d) | |
1566 | return NULL; | |
1567 | pud = pud_alloc(mm, p4d, addr); | |
1568 | if (!pud) | |
1569 | return NULL; | |
1570 | pmd = pmd_alloc(mm, pud, addr); | |
1571 | if (!pmd) | |
1572 | return NULL; | |
1573 | ||
1574 | VM_BUG_ON(pmd_trans_huge(*pmd)); | |
1575 | return pte_alloc_map_lock(mm, pmd, addr, ptl); | |
c9cfcddf LT |
1576 | } |
1577 | ||
238f58d8 LT |
1578 | /* |
1579 | * This is the old fallback for page remapping. | |
1580 | * | |
1581 | * For historical reasons, it only allows reserved pages. Only | |
1582 | * old drivers should use this, and they needed to mark their | |
1583 | * pages reserved for the old functions anyway. | |
1584 | */ | |
423bad60 NP |
1585 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
1586 | struct page *page, pgprot_t prot) | |
238f58d8 | 1587 | { |
423bad60 | 1588 | struct mm_struct *mm = vma->vm_mm; |
238f58d8 | 1589 | int retval; |
c9cfcddf | 1590 | pte_t *pte; |
8a9f3ccd BS |
1591 | spinlock_t *ptl; |
1592 | ||
238f58d8 | 1593 | retval = -EINVAL; |
a145dd41 | 1594 | if (PageAnon(page)) |
5b4e655e | 1595 | goto out; |
238f58d8 LT |
1596 | retval = -ENOMEM; |
1597 | flush_dcache_page(page); | |
c9cfcddf | 1598 | pte = get_locked_pte(mm, addr, &ptl); |
238f58d8 | 1599 | if (!pte) |
5b4e655e | 1600 | goto out; |
238f58d8 LT |
1601 | retval = -EBUSY; |
1602 | if (!pte_none(*pte)) | |
1603 | goto out_unlock; | |
1604 | ||
1605 | /* Ok, finally just insert the thing.. */ | |
1606 | get_page(page); | |
eca56ff9 | 1607 | inc_mm_counter_fast(mm, mm_counter_file(page)); |
dd78fedd | 1608 | page_add_file_rmap(page, false); |
238f58d8 LT |
1609 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); |
1610 | ||
1611 | retval = 0; | |
8a9f3ccd BS |
1612 | pte_unmap_unlock(pte, ptl); |
1613 | return retval; | |
238f58d8 LT |
1614 | out_unlock: |
1615 | pte_unmap_unlock(pte, ptl); | |
1616 | out: | |
1617 | return retval; | |
1618 | } | |
1619 | ||
bfa5bf6d REB |
1620 | /** |
1621 | * vm_insert_page - insert single page into user vma | |
1622 | * @vma: user vma to map to | |
1623 | * @addr: target user address of this page | |
1624 | * @page: source kernel page | |
1625 | * | |
a145dd41 LT |
1626 | * This allows drivers to insert individual pages they've allocated |
1627 | * into a user vma. | |
1628 | * | |
1629 | * The page has to be a nice clean _individual_ kernel allocation. | |
1630 | * If you allocate a compound page, you need to have marked it as | |
1631 | * such (__GFP_COMP), or manually just split the page up yourself | |
8dfcc9ba | 1632 | * (see split_page()). |
a145dd41 LT |
1633 | * |
1634 | * NOTE! Traditionally this was done with "remap_pfn_range()" which | |
1635 | * took an arbitrary page protection parameter. This doesn't allow | |
1636 | * that. Your vma protection will have to be set up correctly, which | |
1637 | * means that if you want a shared writable mapping, you'd better | |
1638 | * ask for a shared writable mapping! | |
1639 | * | |
1640 | * The page does not need to be reserved. | |
4b6e1e37 KK |
1641 | * |
1642 | * Usually this function is called from f_op->mmap() handler | |
1643 | * under mm->mmap_sem write-lock, so it can change vma->vm_flags. | |
1644 | * Caller must set VM_MIXEDMAP on vma if it wants to call this | |
1645 | * function from other places, for example from page-fault handler. | |
a145dd41 | 1646 | */ |
423bad60 NP |
1647 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
1648 | struct page *page) | |
a145dd41 LT |
1649 | { |
1650 | if (addr < vma->vm_start || addr >= vma->vm_end) | |
1651 | return -EFAULT; | |
1652 | if (!page_count(page)) | |
1653 | return -EINVAL; | |
4b6e1e37 KK |
1654 | if (!(vma->vm_flags & VM_MIXEDMAP)) { |
1655 | BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem)); | |
1656 | BUG_ON(vma->vm_flags & VM_PFNMAP); | |
1657 | vma->vm_flags |= VM_MIXEDMAP; | |
1658 | } | |
423bad60 | 1659 | return insert_page(vma, addr, page, vma->vm_page_prot); |
a145dd41 | 1660 | } |
e3c3374f | 1661 | EXPORT_SYMBOL(vm_insert_page); |
a145dd41 | 1662 | |
423bad60 | 1663 | static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
01c8f1c4 | 1664 | pfn_t pfn, pgprot_t prot) |
423bad60 NP |
1665 | { |
1666 | struct mm_struct *mm = vma->vm_mm; | |
1667 | int retval; | |
1668 | pte_t *pte, entry; | |
1669 | spinlock_t *ptl; | |
1670 | ||
1671 | retval = -ENOMEM; | |
1672 | pte = get_locked_pte(mm, addr, &ptl); | |
1673 | if (!pte) | |
1674 | goto out; | |
1675 | retval = -EBUSY; | |
1676 | if (!pte_none(*pte)) | |
1677 | goto out_unlock; | |
1678 | ||
1679 | /* Ok, finally just insert the thing.. */ | |
01c8f1c4 DW |
1680 | if (pfn_t_devmap(pfn)) |
1681 | entry = pte_mkdevmap(pfn_t_pte(pfn, prot)); | |
1682 | else | |
1683 | entry = pte_mkspecial(pfn_t_pte(pfn, prot)); | |
423bad60 | 1684 | set_pte_at(mm, addr, pte, entry); |
4b3073e1 | 1685 | update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */ |
423bad60 NP |
1686 | |
1687 | retval = 0; | |
1688 | out_unlock: | |
1689 | pte_unmap_unlock(pte, ptl); | |
1690 | out: | |
1691 | return retval; | |
1692 | } | |
1693 | ||
e0dc0d8f NP |
1694 | /** |
1695 | * vm_insert_pfn - insert single pfn into user vma | |
1696 | * @vma: user vma to map to | |
1697 | * @addr: target user address of this page | |
1698 | * @pfn: source kernel pfn | |
1699 | * | |
c462f179 | 1700 | * Similar to vm_insert_page, this allows drivers to insert individual pages |
e0dc0d8f NP |
1701 | * they've allocated into a user vma. Same comments apply. |
1702 | * | |
1703 | * This function should only be called from a vm_ops->fault handler, and | |
1704 | * in that case the handler should return NULL. | |
0d71d10a NP |
1705 | * |
1706 | * vma cannot be a COW mapping. | |
1707 | * | |
1708 | * As this is called only for pages that do not currently exist, we | |
1709 | * do not need to flush old virtual caches or the TLB. | |
e0dc0d8f NP |
1710 | */ |
1711 | int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, | |
423bad60 | 1712 | unsigned long pfn) |
1745cbc5 AL |
1713 | { |
1714 | return vm_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot); | |
1715 | } | |
1716 | EXPORT_SYMBOL(vm_insert_pfn); | |
1717 | ||
1718 | /** | |
1719 | * vm_insert_pfn_prot - insert single pfn into user vma with specified pgprot | |
1720 | * @vma: user vma to map to | |
1721 | * @addr: target user address of this page | |
1722 | * @pfn: source kernel pfn | |
1723 | * @pgprot: pgprot flags for the inserted page | |
1724 | * | |
1725 | * This is exactly like vm_insert_pfn, except that it allows drivers to | |
1726 | * to override pgprot on a per-page basis. | |
1727 | * | |
1728 | * This only makes sense for IO mappings, and it makes no sense for | |
1729 | * cow mappings. In general, using multiple vmas is preferable; | |
1730 | * vm_insert_pfn_prot should only be used if using multiple VMAs is | |
1731 | * impractical. | |
1732 | */ | |
1733 | int vm_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr, | |
1734 | unsigned long pfn, pgprot_t pgprot) | |
e0dc0d8f | 1735 | { |
2ab64037 | 1736 | int ret; |
7e675137 NP |
1737 | /* |
1738 | * Technically, architectures with pte_special can avoid all these | |
1739 | * restrictions (same for remap_pfn_range). However we would like | |
1740 | * consistency in testing and feature parity among all, so we should | |
1741 | * try to keep these invariants in place for everybody. | |
1742 | */ | |
b379d790 JH |
1743 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); |
1744 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == | |
1745 | (VM_PFNMAP|VM_MIXEDMAP)); | |
1746 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); | |
1747 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); | |
e0dc0d8f | 1748 | |
423bad60 NP |
1749 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1750 | return -EFAULT; | |
308a047c BP |
1751 | |
1752 | track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV)); | |
2ab64037 | 1753 | |
01c8f1c4 | 1754 | ret = insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot); |
2ab64037 | 1755 | |
2ab64037 | 1756 | return ret; |
423bad60 | 1757 | } |
1745cbc5 | 1758 | EXPORT_SYMBOL(vm_insert_pfn_prot); |
e0dc0d8f | 1759 | |
423bad60 | 1760 | int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, |
01c8f1c4 | 1761 | pfn_t pfn) |
423bad60 | 1762 | { |
87744ab3 DW |
1763 | pgprot_t pgprot = vma->vm_page_prot; |
1764 | ||
423bad60 | 1765 | BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); |
e0dc0d8f | 1766 | |
423bad60 NP |
1767 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1768 | return -EFAULT; | |
308a047c BP |
1769 | |
1770 | track_pfn_insert(vma, &pgprot, pfn); | |
e0dc0d8f | 1771 | |
423bad60 NP |
1772 | /* |
1773 | * If we don't have pte special, then we have to use the pfn_valid() | |
1774 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* | |
1775 | * refcount the page if pfn_valid is true (hence insert_page rather | |
62eede62 HD |
1776 | * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP |
1777 | * without pte special, it would there be refcounted as a normal page. | |
423bad60 | 1778 | */ |
03fc2da6 | 1779 | if (!HAVE_PTE_SPECIAL && !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) { |
423bad60 NP |
1780 | struct page *page; |
1781 | ||
03fc2da6 DW |
1782 | /* |
1783 | * At this point we are committed to insert_page() | |
1784 | * regardless of whether the caller specified flags that | |
1785 | * result in pfn_t_has_page() == false. | |
1786 | */ | |
1787 | page = pfn_to_page(pfn_t_to_pfn(pfn)); | |
87744ab3 | 1788 | return insert_page(vma, addr, page, pgprot); |
423bad60 | 1789 | } |
87744ab3 | 1790 | return insert_pfn(vma, addr, pfn, pgprot); |
e0dc0d8f | 1791 | } |
423bad60 | 1792 | EXPORT_SYMBOL(vm_insert_mixed); |
e0dc0d8f | 1793 | |
1da177e4 LT |
1794 | /* |
1795 | * maps a range of physical memory into the requested pages. the old | |
1796 | * mappings are removed. any references to nonexistent pages results | |
1797 | * in null mappings (currently treated as "copy-on-access") | |
1798 | */ | |
1799 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, | |
1800 | unsigned long addr, unsigned long end, | |
1801 | unsigned long pfn, pgprot_t prot) | |
1802 | { | |
1803 | pte_t *pte; | |
c74df32c | 1804 | spinlock_t *ptl; |
1da177e4 | 1805 | |
c74df32c | 1806 | pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1da177e4 LT |
1807 | if (!pte) |
1808 | return -ENOMEM; | |
6606c3e0 | 1809 | arch_enter_lazy_mmu_mode(); |
1da177e4 LT |
1810 | do { |
1811 | BUG_ON(!pte_none(*pte)); | |
7e675137 | 1812 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
1da177e4 LT |
1813 | pfn++; |
1814 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 1815 | arch_leave_lazy_mmu_mode(); |
c74df32c | 1816 | pte_unmap_unlock(pte - 1, ptl); |
1da177e4 LT |
1817 | return 0; |
1818 | } | |
1819 | ||
1820 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, | |
1821 | unsigned long addr, unsigned long end, | |
1822 | unsigned long pfn, pgprot_t prot) | |
1823 | { | |
1824 | pmd_t *pmd; | |
1825 | unsigned long next; | |
1826 | ||
1827 | pfn -= addr >> PAGE_SHIFT; | |
1828 | pmd = pmd_alloc(mm, pud, addr); | |
1829 | if (!pmd) | |
1830 | return -ENOMEM; | |
f66055ab | 1831 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
1da177e4 LT |
1832 | do { |
1833 | next = pmd_addr_end(addr, end); | |
1834 | if (remap_pte_range(mm, pmd, addr, next, | |
1835 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1836 | return -ENOMEM; | |
1837 | } while (pmd++, addr = next, addr != end); | |
1838 | return 0; | |
1839 | } | |
1840 | ||
c2febafc | 1841 | static inline int remap_pud_range(struct mm_struct *mm, p4d_t *p4d, |
1da177e4 LT |
1842 | unsigned long addr, unsigned long end, |
1843 | unsigned long pfn, pgprot_t prot) | |
1844 | { | |
1845 | pud_t *pud; | |
1846 | unsigned long next; | |
1847 | ||
1848 | pfn -= addr >> PAGE_SHIFT; | |
c2febafc | 1849 | pud = pud_alloc(mm, p4d, addr); |
1da177e4 LT |
1850 | if (!pud) |
1851 | return -ENOMEM; | |
1852 | do { | |
1853 | next = pud_addr_end(addr, end); | |
1854 | if (remap_pmd_range(mm, pud, addr, next, | |
1855 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1856 | return -ENOMEM; | |
1857 | } while (pud++, addr = next, addr != end); | |
1858 | return 0; | |
1859 | } | |
1860 | ||
c2febafc KS |
1861 | static inline int remap_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
1862 | unsigned long addr, unsigned long end, | |
1863 | unsigned long pfn, pgprot_t prot) | |
1864 | { | |
1865 | p4d_t *p4d; | |
1866 | unsigned long next; | |
1867 | ||
1868 | pfn -= addr >> PAGE_SHIFT; | |
1869 | p4d = p4d_alloc(mm, pgd, addr); | |
1870 | if (!p4d) | |
1871 | return -ENOMEM; | |
1872 | do { | |
1873 | next = p4d_addr_end(addr, end); | |
1874 | if (remap_pud_range(mm, p4d, addr, next, | |
1875 | pfn + (addr >> PAGE_SHIFT), prot)) | |
1876 | return -ENOMEM; | |
1877 | } while (p4d++, addr = next, addr != end); | |
1878 | return 0; | |
1879 | } | |
1880 | ||
bfa5bf6d REB |
1881 | /** |
1882 | * remap_pfn_range - remap kernel memory to userspace | |
1883 | * @vma: user vma to map to | |
1884 | * @addr: target user address to start at | |
1885 | * @pfn: physical address of kernel memory | |
1886 | * @size: size of map area | |
1887 | * @prot: page protection flags for this mapping | |
1888 | * | |
1889 | * Note: this is only safe if the mm semaphore is held when called. | |
1890 | */ | |
1da177e4 LT |
1891 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
1892 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
1893 | { | |
1894 | pgd_t *pgd; | |
1895 | unsigned long next; | |
2d15cab8 | 1896 | unsigned long end = addr + PAGE_ALIGN(size); |
1da177e4 | 1897 | struct mm_struct *mm = vma->vm_mm; |
d5957d2f | 1898 | unsigned long remap_pfn = pfn; |
1da177e4 LT |
1899 | int err; |
1900 | ||
1901 | /* | |
1902 | * Physically remapped pages are special. Tell the | |
1903 | * rest of the world about it: | |
1904 | * VM_IO tells people not to look at these pages | |
1905 | * (accesses can have side effects). | |
6aab341e LT |
1906 | * VM_PFNMAP tells the core MM that the base pages are just |
1907 | * raw PFN mappings, and do not have a "struct page" associated | |
1908 | * with them. | |
314e51b9 KK |
1909 | * VM_DONTEXPAND |
1910 | * Disable vma merging and expanding with mremap(). | |
1911 | * VM_DONTDUMP | |
1912 | * Omit vma from core dump, even when VM_IO turned off. | |
fb155c16 LT |
1913 | * |
1914 | * There's a horrible special case to handle copy-on-write | |
1915 | * behaviour that some programs depend on. We mark the "original" | |
1916 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". | |
b3b9c293 | 1917 | * See vm_normal_page() for details. |
1da177e4 | 1918 | */ |
b3b9c293 KK |
1919 | if (is_cow_mapping(vma->vm_flags)) { |
1920 | if (addr != vma->vm_start || end != vma->vm_end) | |
1921 | return -EINVAL; | |
fb155c16 | 1922 | vma->vm_pgoff = pfn; |
b3b9c293 KK |
1923 | } |
1924 | ||
d5957d2f | 1925 | err = track_pfn_remap(vma, &prot, remap_pfn, addr, PAGE_ALIGN(size)); |
b3b9c293 | 1926 | if (err) |
3c8bb73a | 1927 | return -EINVAL; |
fb155c16 | 1928 | |
314e51b9 | 1929 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; |
1da177e4 LT |
1930 | |
1931 | BUG_ON(addr >= end); | |
1932 | pfn -= addr >> PAGE_SHIFT; | |
1933 | pgd = pgd_offset(mm, addr); | |
1934 | flush_cache_range(vma, addr, end); | |
1da177e4 LT |
1935 | do { |
1936 | next = pgd_addr_end(addr, end); | |
c2febafc | 1937 | err = remap_p4d_range(mm, pgd, addr, next, |
1da177e4 LT |
1938 | pfn + (addr >> PAGE_SHIFT), prot); |
1939 | if (err) | |
1940 | break; | |
1941 | } while (pgd++, addr = next, addr != end); | |
2ab64037 | 1942 | |
1943 | if (err) | |
d5957d2f | 1944 | untrack_pfn(vma, remap_pfn, PAGE_ALIGN(size)); |
2ab64037 | 1945 | |
1da177e4 LT |
1946 | return err; |
1947 | } | |
1948 | EXPORT_SYMBOL(remap_pfn_range); | |
1949 | ||
b4cbb197 LT |
1950 | /** |
1951 | * vm_iomap_memory - remap memory to userspace | |
1952 | * @vma: user vma to map to | |
1953 | * @start: start of area | |
1954 | * @len: size of area | |
1955 | * | |
1956 | * This is a simplified io_remap_pfn_range() for common driver use. The | |
1957 | * driver just needs to give us the physical memory range to be mapped, | |
1958 | * we'll figure out the rest from the vma information. | |
1959 | * | |
1960 | * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get | |
1961 | * whatever write-combining details or similar. | |
1962 | */ | |
1963 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) | |
1964 | { | |
1965 | unsigned long vm_len, pfn, pages; | |
1966 | ||
1967 | /* Check that the physical memory area passed in looks valid */ | |
1968 | if (start + len < start) | |
1969 | return -EINVAL; | |
1970 | /* | |
1971 | * You *really* shouldn't map things that aren't page-aligned, | |
1972 | * but we've historically allowed it because IO memory might | |
1973 | * just have smaller alignment. | |
1974 | */ | |
1975 | len += start & ~PAGE_MASK; | |
1976 | pfn = start >> PAGE_SHIFT; | |
1977 | pages = (len + ~PAGE_MASK) >> PAGE_SHIFT; | |
1978 | if (pfn + pages < pfn) | |
1979 | return -EINVAL; | |
1980 | ||
1981 | /* We start the mapping 'vm_pgoff' pages into the area */ | |
1982 | if (vma->vm_pgoff > pages) | |
1983 | return -EINVAL; | |
1984 | pfn += vma->vm_pgoff; | |
1985 | pages -= vma->vm_pgoff; | |
1986 | ||
1987 | /* Can we fit all of the mapping? */ | |
1988 | vm_len = vma->vm_end - vma->vm_start; | |
1989 | if (vm_len >> PAGE_SHIFT > pages) | |
1990 | return -EINVAL; | |
1991 | ||
1992 | /* Ok, let it rip */ | |
1993 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); | |
1994 | } | |
1995 | EXPORT_SYMBOL(vm_iomap_memory); | |
1996 | ||
aee16b3c JF |
1997 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
1998 | unsigned long addr, unsigned long end, | |
1999 | pte_fn_t fn, void *data) | |
2000 | { | |
2001 | pte_t *pte; | |
2002 | int err; | |
2f569afd | 2003 | pgtable_t token; |
94909914 | 2004 | spinlock_t *uninitialized_var(ptl); |
aee16b3c JF |
2005 | |
2006 | pte = (mm == &init_mm) ? | |
2007 | pte_alloc_kernel(pmd, addr) : | |
2008 | pte_alloc_map_lock(mm, pmd, addr, &ptl); | |
2009 | if (!pte) | |
2010 | return -ENOMEM; | |
2011 | ||
2012 | BUG_ON(pmd_huge(*pmd)); | |
2013 | ||
38e0edb1 JF |
2014 | arch_enter_lazy_mmu_mode(); |
2015 | ||
2f569afd | 2016 | token = pmd_pgtable(*pmd); |
aee16b3c JF |
2017 | |
2018 | do { | |
c36987e2 | 2019 | err = fn(pte++, token, addr, data); |
aee16b3c JF |
2020 | if (err) |
2021 | break; | |
c36987e2 | 2022 | } while (addr += PAGE_SIZE, addr != end); |
aee16b3c | 2023 | |
38e0edb1 JF |
2024 | arch_leave_lazy_mmu_mode(); |
2025 | ||
aee16b3c JF |
2026 | if (mm != &init_mm) |
2027 | pte_unmap_unlock(pte-1, ptl); | |
2028 | return err; | |
2029 | } | |
2030 | ||
2031 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, | |
2032 | unsigned long addr, unsigned long end, | |
2033 | pte_fn_t fn, void *data) | |
2034 | { | |
2035 | pmd_t *pmd; | |
2036 | unsigned long next; | |
2037 | int err; | |
2038 | ||
ceb86879 AK |
2039 | BUG_ON(pud_huge(*pud)); |
2040 | ||
aee16b3c JF |
2041 | pmd = pmd_alloc(mm, pud, addr); |
2042 | if (!pmd) | |
2043 | return -ENOMEM; | |
2044 | do { | |
2045 | next = pmd_addr_end(addr, end); | |
2046 | err = apply_to_pte_range(mm, pmd, addr, next, fn, data); | |
2047 | if (err) | |
2048 | break; | |
2049 | } while (pmd++, addr = next, addr != end); | |
2050 | return err; | |
2051 | } | |
2052 | ||
c2febafc | 2053 | static int apply_to_pud_range(struct mm_struct *mm, p4d_t *p4d, |
aee16b3c JF |
2054 | unsigned long addr, unsigned long end, |
2055 | pte_fn_t fn, void *data) | |
2056 | { | |
2057 | pud_t *pud; | |
2058 | unsigned long next; | |
2059 | int err; | |
2060 | ||
c2febafc | 2061 | pud = pud_alloc(mm, p4d, addr); |
aee16b3c JF |
2062 | if (!pud) |
2063 | return -ENOMEM; | |
2064 | do { | |
2065 | next = pud_addr_end(addr, end); | |
2066 | err = apply_to_pmd_range(mm, pud, addr, next, fn, data); | |
2067 | if (err) | |
2068 | break; | |
2069 | } while (pud++, addr = next, addr != end); | |
2070 | return err; | |
2071 | } | |
2072 | ||
c2febafc KS |
2073 | static int apply_to_p4d_range(struct mm_struct *mm, pgd_t *pgd, |
2074 | unsigned long addr, unsigned long end, | |
2075 | pte_fn_t fn, void *data) | |
2076 | { | |
2077 | p4d_t *p4d; | |
2078 | unsigned long next; | |
2079 | int err; | |
2080 | ||
2081 | p4d = p4d_alloc(mm, pgd, addr); | |
2082 | if (!p4d) | |
2083 | return -ENOMEM; | |
2084 | do { | |
2085 | next = p4d_addr_end(addr, end); | |
2086 | err = apply_to_pud_range(mm, p4d, addr, next, fn, data); | |
2087 | if (err) | |
2088 | break; | |
2089 | } while (p4d++, addr = next, addr != end); | |
2090 | return err; | |
2091 | } | |
2092 | ||
aee16b3c JF |
2093 | /* |
2094 | * Scan a region of virtual memory, filling in page tables as necessary | |
2095 | * and calling a provided function on each leaf page table. | |
2096 | */ | |
2097 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, | |
2098 | unsigned long size, pte_fn_t fn, void *data) | |
2099 | { | |
2100 | pgd_t *pgd; | |
2101 | unsigned long next; | |
57250a5b | 2102 | unsigned long end = addr + size; |
aee16b3c JF |
2103 | int err; |
2104 | ||
9cb65bc3 MP |
2105 | if (WARN_ON(addr >= end)) |
2106 | return -EINVAL; | |
2107 | ||
aee16b3c JF |
2108 | pgd = pgd_offset(mm, addr); |
2109 | do { | |
2110 | next = pgd_addr_end(addr, end); | |
c2febafc | 2111 | err = apply_to_p4d_range(mm, pgd, addr, next, fn, data); |
aee16b3c JF |
2112 | if (err) |
2113 | break; | |
2114 | } while (pgd++, addr = next, addr != end); | |
57250a5b | 2115 | |
aee16b3c JF |
2116 | return err; |
2117 | } | |
2118 | EXPORT_SYMBOL_GPL(apply_to_page_range); | |
2119 | ||
8f4e2101 | 2120 | /* |
9b4bdd2f KS |
2121 | * handle_pte_fault chooses page fault handler according to an entry which was |
2122 | * read non-atomically. Before making any commitment, on those architectures | |
2123 | * or configurations (e.g. i386 with PAE) which might give a mix of unmatched | |
2124 | * parts, do_swap_page must check under lock before unmapping the pte and | |
2125 | * proceeding (but do_wp_page is only called after already making such a check; | |
a335b2e1 | 2126 | * and do_anonymous_page can safely check later on). |
8f4e2101 | 2127 | */ |
4c21e2f2 | 2128 | static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, |
8f4e2101 HD |
2129 | pte_t *page_table, pte_t orig_pte) |
2130 | { | |
2131 | int same = 1; | |
2132 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) | |
2133 | if (sizeof(pte_t) > sizeof(unsigned long)) { | |
4c21e2f2 HD |
2134 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
2135 | spin_lock(ptl); | |
8f4e2101 | 2136 | same = pte_same(*page_table, orig_pte); |
4c21e2f2 | 2137 | spin_unlock(ptl); |
8f4e2101 HD |
2138 | } |
2139 | #endif | |
2140 | pte_unmap(page_table); | |
2141 | return same; | |
2142 | } | |
2143 | ||
9de455b2 | 2144 | static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) |
6aab341e | 2145 | { |
0abdd7a8 DW |
2146 | debug_dma_assert_idle(src); |
2147 | ||
6aab341e LT |
2148 | /* |
2149 | * If the source page was a PFN mapping, we don't have | |
2150 | * a "struct page" for it. We do a best-effort copy by | |
2151 | * just copying from the original user address. If that | |
2152 | * fails, we just zero-fill it. Live with it. | |
2153 | */ | |
2154 | if (unlikely(!src)) { | |
9b04c5fe | 2155 | void *kaddr = kmap_atomic(dst); |
5d2a2dbb LT |
2156 | void __user *uaddr = (void __user *)(va & PAGE_MASK); |
2157 | ||
2158 | /* | |
2159 | * This really shouldn't fail, because the page is there | |
2160 | * in the page tables. But it might just be unreadable, | |
2161 | * in which case we just give up and fill the result with | |
2162 | * zeroes. | |
2163 | */ | |
2164 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) | |
3ecb01df | 2165 | clear_page(kaddr); |
9b04c5fe | 2166 | kunmap_atomic(kaddr); |
c4ec7b0d | 2167 | flush_dcache_page(dst); |
0ed361de NP |
2168 | } else |
2169 | copy_user_highpage(dst, src, va, vma); | |
6aab341e LT |
2170 | } |
2171 | ||
c20cd45e MH |
2172 | static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma) |
2173 | { | |
2174 | struct file *vm_file = vma->vm_file; | |
2175 | ||
2176 | if (vm_file) | |
2177 | return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO; | |
2178 | ||
2179 | /* | |
2180 | * Special mappings (e.g. VDSO) do not have any file so fake | |
2181 | * a default GFP_KERNEL for them. | |
2182 | */ | |
2183 | return GFP_KERNEL; | |
2184 | } | |
2185 | ||
fb09a464 KS |
2186 | /* |
2187 | * Notify the address space that the page is about to become writable so that | |
2188 | * it can prohibit this or wait for the page to get into an appropriate state. | |
2189 | * | |
2190 | * We do this without the lock held, so that it can sleep if it needs to. | |
2191 | */ | |
38b8cb7f | 2192 | static int do_page_mkwrite(struct vm_fault *vmf) |
fb09a464 | 2193 | { |
fb09a464 | 2194 | int ret; |
38b8cb7f JK |
2195 | struct page *page = vmf->page; |
2196 | unsigned int old_flags = vmf->flags; | |
fb09a464 | 2197 | |
38b8cb7f | 2198 | vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; |
fb09a464 | 2199 | |
11bac800 | 2200 | ret = vmf->vma->vm_ops->page_mkwrite(vmf); |
38b8cb7f JK |
2201 | /* Restore original flags so that caller is not surprised */ |
2202 | vmf->flags = old_flags; | |
fb09a464 KS |
2203 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) |
2204 | return ret; | |
2205 | if (unlikely(!(ret & VM_FAULT_LOCKED))) { | |
2206 | lock_page(page); | |
2207 | if (!page->mapping) { | |
2208 | unlock_page(page); | |
2209 | return 0; /* retry */ | |
2210 | } | |
2211 | ret |= VM_FAULT_LOCKED; | |
2212 | } else | |
2213 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
2214 | return ret; | |
2215 | } | |
2216 | ||
97ba0c2b JK |
2217 | /* |
2218 | * Handle dirtying of a page in shared file mapping on a write fault. | |
2219 | * | |
2220 | * The function expects the page to be locked and unlocks it. | |
2221 | */ | |
2222 | static void fault_dirty_shared_page(struct vm_area_struct *vma, | |
2223 | struct page *page) | |
2224 | { | |
2225 | struct address_space *mapping; | |
2226 | bool dirtied; | |
2227 | bool page_mkwrite = vma->vm_ops && vma->vm_ops->page_mkwrite; | |
2228 | ||
2229 | dirtied = set_page_dirty(page); | |
2230 | VM_BUG_ON_PAGE(PageAnon(page), page); | |
2231 | /* | |
2232 | * Take a local copy of the address_space - page.mapping may be zeroed | |
2233 | * by truncate after unlock_page(). The address_space itself remains | |
2234 | * pinned by vma->vm_file's reference. We rely on unlock_page()'s | |
2235 | * release semantics to prevent the compiler from undoing this copying. | |
2236 | */ | |
2237 | mapping = page_rmapping(page); | |
2238 | unlock_page(page); | |
2239 | ||
2240 | if ((dirtied || page_mkwrite) && mapping) { | |
2241 | /* | |
2242 | * Some device drivers do not set page.mapping | |
2243 | * but still dirty their pages | |
2244 | */ | |
2245 | balance_dirty_pages_ratelimited(mapping); | |
2246 | } | |
2247 | ||
2248 | if (!page_mkwrite) | |
2249 | file_update_time(vma->vm_file); | |
2250 | } | |
2251 | ||
4e047f89 SR |
2252 | /* |
2253 | * Handle write page faults for pages that can be reused in the current vma | |
2254 | * | |
2255 | * This can happen either due to the mapping being with the VM_SHARED flag, | |
2256 | * or due to us being the last reference standing to the page. In either | |
2257 | * case, all we need to do here is to mark the page as writable and update | |
2258 | * any related book-keeping. | |
2259 | */ | |
997dd98d | 2260 | static inline void wp_page_reuse(struct vm_fault *vmf) |
82b0f8c3 | 2261 | __releases(vmf->ptl) |
4e047f89 | 2262 | { |
82b0f8c3 | 2263 | struct vm_area_struct *vma = vmf->vma; |
a41b70d6 | 2264 | struct page *page = vmf->page; |
4e047f89 SR |
2265 | pte_t entry; |
2266 | /* | |
2267 | * Clear the pages cpupid information as the existing | |
2268 | * information potentially belongs to a now completely | |
2269 | * unrelated process. | |
2270 | */ | |
2271 | if (page) | |
2272 | page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1); | |
2273 | ||
2994302b JK |
2274 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
2275 | entry = pte_mkyoung(vmf->orig_pte); | |
4e047f89 | 2276 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
82b0f8c3 JK |
2277 | if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1)) |
2278 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
2279 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
4e047f89 SR |
2280 | } |
2281 | ||
2f38ab2c SR |
2282 | /* |
2283 | * Handle the case of a page which we actually need to copy to a new page. | |
2284 | * | |
2285 | * Called with mmap_sem locked and the old page referenced, but | |
2286 | * without the ptl held. | |
2287 | * | |
2288 | * High level logic flow: | |
2289 | * | |
2290 | * - Allocate a page, copy the content of the old page to the new one. | |
2291 | * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc. | |
2292 | * - Take the PTL. If the pte changed, bail out and release the allocated page | |
2293 | * - If the pte is still the way we remember it, update the page table and all | |
2294 | * relevant references. This includes dropping the reference the page-table | |
2295 | * held to the old page, as well as updating the rmap. | |
2296 | * - In any case, unlock the PTL and drop the reference we took to the old page. | |
2297 | */ | |
a41b70d6 | 2298 | static int wp_page_copy(struct vm_fault *vmf) |
2f38ab2c | 2299 | { |
82b0f8c3 | 2300 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 2301 | struct mm_struct *mm = vma->vm_mm; |
a41b70d6 | 2302 | struct page *old_page = vmf->page; |
2f38ab2c | 2303 | struct page *new_page = NULL; |
2f38ab2c SR |
2304 | pte_t entry; |
2305 | int page_copied = 0; | |
82b0f8c3 | 2306 | const unsigned long mmun_start = vmf->address & PAGE_MASK; |
bae473a4 | 2307 | const unsigned long mmun_end = mmun_start + PAGE_SIZE; |
2f38ab2c SR |
2308 | struct mem_cgroup *memcg; |
2309 | ||
2310 | if (unlikely(anon_vma_prepare(vma))) | |
2311 | goto oom; | |
2312 | ||
2994302b | 2313 | if (is_zero_pfn(pte_pfn(vmf->orig_pte))) { |
82b0f8c3 JK |
2314 | new_page = alloc_zeroed_user_highpage_movable(vma, |
2315 | vmf->address); | |
2f38ab2c SR |
2316 | if (!new_page) |
2317 | goto oom; | |
2318 | } else { | |
bae473a4 | 2319 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, |
82b0f8c3 | 2320 | vmf->address); |
2f38ab2c SR |
2321 | if (!new_page) |
2322 | goto oom; | |
82b0f8c3 | 2323 | cow_user_page(new_page, old_page, vmf->address, vma); |
2f38ab2c | 2324 | } |
2f38ab2c | 2325 | |
f627c2f5 | 2326 | if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg, false)) |
2f38ab2c SR |
2327 | goto oom_free_new; |
2328 | ||
eb3c24f3 MG |
2329 | __SetPageUptodate(new_page); |
2330 | ||
2f38ab2c SR |
2331 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
2332 | ||
2333 | /* | |
2334 | * Re-check the pte - we dropped the lock | |
2335 | */ | |
82b0f8c3 | 2336 | vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl); |
2994302b | 2337 | if (likely(pte_same(*vmf->pte, vmf->orig_pte))) { |
2f38ab2c SR |
2338 | if (old_page) { |
2339 | if (!PageAnon(old_page)) { | |
eca56ff9 JM |
2340 | dec_mm_counter_fast(mm, |
2341 | mm_counter_file(old_page)); | |
2f38ab2c SR |
2342 | inc_mm_counter_fast(mm, MM_ANONPAGES); |
2343 | } | |
2344 | } else { | |
2345 | inc_mm_counter_fast(mm, MM_ANONPAGES); | |
2346 | } | |
2994302b | 2347 | flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte)); |
2f38ab2c SR |
2348 | entry = mk_pte(new_page, vma->vm_page_prot); |
2349 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
2350 | /* | |
2351 | * Clear the pte entry and flush it first, before updating the | |
2352 | * pte with the new entry. This will avoid a race condition | |
2353 | * seen in the presence of one thread doing SMC and another | |
2354 | * thread doing COW. | |
2355 | */ | |
82b0f8c3 JK |
2356 | ptep_clear_flush_notify(vma, vmf->address, vmf->pte); |
2357 | page_add_new_anon_rmap(new_page, vma, vmf->address, false); | |
f627c2f5 | 2358 | mem_cgroup_commit_charge(new_page, memcg, false, false); |
2f38ab2c SR |
2359 | lru_cache_add_active_or_unevictable(new_page, vma); |
2360 | /* | |
2361 | * We call the notify macro here because, when using secondary | |
2362 | * mmu page tables (such as kvm shadow page tables), we want the | |
2363 | * new page to be mapped directly into the secondary page table. | |
2364 | */ | |
82b0f8c3 JK |
2365 | set_pte_at_notify(mm, vmf->address, vmf->pte, entry); |
2366 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
2f38ab2c SR |
2367 | if (old_page) { |
2368 | /* | |
2369 | * Only after switching the pte to the new page may | |
2370 | * we remove the mapcount here. Otherwise another | |
2371 | * process may come and find the rmap count decremented | |
2372 | * before the pte is switched to the new page, and | |
2373 | * "reuse" the old page writing into it while our pte | |
2374 | * here still points into it and can be read by other | |
2375 | * threads. | |
2376 | * | |
2377 | * The critical issue is to order this | |
2378 | * page_remove_rmap with the ptp_clear_flush above. | |
2379 | * Those stores are ordered by (if nothing else,) | |
2380 | * the barrier present in the atomic_add_negative | |
2381 | * in page_remove_rmap. | |
2382 | * | |
2383 | * Then the TLB flush in ptep_clear_flush ensures that | |
2384 | * no process can access the old page before the | |
2385 | * decremented mapcount is visible. And the old page | |
2386 | * cannot be reused until after the decremented | |
2387 | * mapcount is visible. So transitively, TLBs to | |
2388 | * old page will be flushed before it can be reused. | |
2389 | */ | |
d281ee61 | 2390 | page_remove_rmap(old_page, false); |
2f38ab2c SR |
2391 | } |
2392 | ||
2393 | /* Free the old page.. */ | |
2394 | new_page = old_page; | |
2395 | page_copied = 1; | |
2396 | } else { | |
f627c2f5 | 2397 | mem_cgroup_cancel_charge(new_page, memcg, false); |
2f38ab2c SR |
2398 | } |
2399 | ||
2400 | if (new_page) | |
09cbfeaf | 2401 | put_page(new_page); |
2f38ab2c | 2402 | |
82b0f8c3 | 2403 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
2f38ab2c SR |
2404 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); |
2405 | if (old_page) { | |
2406 | /* | |
2407 | * Don't let another task, with possibly unlocked vma, | |
2408 | * keep the mlocked page. | |
2409 | */ | |
2410 | if (page_copied && (vma->vm_flags & VM_LOCKED)) { | |
2411 | lock_page(old_page); /* LRU manipulation */ | |
e90309c9 KS |
2412 | if (PageMlocked(old_page)) |
2413 | munlock_vma_page(old_page); | |
2f38ab2c SR |
2414 | unlock_page(old_page); |
2415 | } | |
09cbfeaf | 2416 | put_page(old_page); |
2f38ab2c SR |
2417 | } |
2418 | return page_copied ? VM_FAULT_WRITE : 0; | |
2419 | oom_free_new: | |
09cbfeaf | 2420 | put_page(new_page); |
2f38ab2c SR |
2421 | oom: |
2422 | if (old_page) | |
09cbfeaf | 2423 | put_page(old_page); |
2f38ab2c SR |
2424 | return VM_FAULT_OOM; |
2425 | } | |
2426 | ||
66a6197c JK |
2427 | /** |
2428 | * finish_mkwrite_fault - finish page fault for a shared mapping, making PTE | |
2429 | * writeable once the page is prepared | |
2430 | * | |
2431 | * @vmf: structure describing the fault | |
2432 | * | |
2433 | * This function handles all that is needed to finish a write page fault in a | |
2434 | * shared mapping due to PTE being read-only once the mapped page is prepared. | |
2435 | * It handles locking of PTE and modifying it. The function returns | |
2436 | * VM_FAULT_WRITE on success, 0 when PTE got changed before we acquired PTE | |
2437 | * lock. | |
2438 | * | |
2439 | * The function expects the page to be locked or other protection against | |
2440 | * concurrent faults / writeback (such as DAX radix tree locks). | |
2441 | */ | |
2442 | int finish_mkwrite_fault(struct vm_fault *vmf) | |
2443 | { | |
2444 | WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED)); | |
2445 | vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address, | |
2446 | &vmf->ptl); | |
2447 | /* | |
2448 | * We might have raced with another page fault while we released the | |
2449 | * pte_offset_map_lock. | |
2450 | */ | |
2451 | if (!pte_same(*vmf->pte, vmf->orig_pte)) { | |
2452 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
a19e2553 | 2453 | return VM_FAULT_NOPAGE; |
66a6197c JK |
2454 | } |
2455 | wp_page_reuse(vmf); | |
a19e2553 | 2456 | return 0; |
66a6197c JK |
2457 | } |
2458 | ||
dd906184 BH |
2459 | /* |
2460 | * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED | |
2461 | * mapping | |
2462 | */ | |
2994302b | 2463 | static int wp_pfn_shared(struct vm_fault *vmf) |
dd906184 | 2464 | { |
82b0f8c3 | 2465 | struct vm_area_struct *vma = vmf->vma; |
bae473a4 | 2466 | |
dd906184 | 2467 | if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) { |
dd906184 BH |
2468 | int ret; |
2469 | ||
82b0f8c3 | 2470 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
fe82221f | 2471 | vmf->flags |= FAULT_FLAG_MKWRITE; |
11bac800 | 2472 | ret = vma->vm_ops->pfn_mkwrite(vmf); |
2f89dc12 | 2473 | if (ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)) |
dd906184 | 2474 | return ret; |
66a6197c | 2475 | return finish_mkwrite_fault(vmf); |
dd906184 | 2476 | } |
997dd98d JK |
2477 | wp_page_reuse(vmf); |
2478 | return VM_FAULT_WRITE; | |
dd906184 BH |
2479 | } |
2480 | ||
a41b70d6 | 2481 | static int wp_page_shared(struct vm_fault *vmf) |
82b0f8c3 | 2482 | __releases(vmf->ptl) |
93e478d4 | 2483 | { |
82b0f8c3 | 2484 | struct vm_area_struct *vma = vmf->vma; |
93e478d4 | 2485 | |
a41b70d6 | 2486 | get_page(vmf->page); |
93e478d4 | 2487 | |
93e478d4 SR |
2488 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
2489 | int tmp; | |
2490 | ||
82b0f8c3 | 2491 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
38b8cb7f | 2492 | tmp = do_page_mkwrite(vmf); |
93e478d4 SR |
2493 | if (unlikely(!tmp || (tmp & |
2494 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
a41b70d6 | 2495 | put_page(vmf->page); |
93e478d4 SR |
2496 | return tmp; |
2497 | } | |
66a6197c | 2498 | tmp = finish_mkwrite_fault(vmf); |
a19e2553 | 2499 | if (unlikely(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { |
a41b70d6 | 2500 | unlock_page(vmf->page); |
a41b70d6 | 2501 | put_page(vmf->page); |
66a6197c | 2502 | return tmp; |
93e478d4 | 2503 | } |
66a6197c JK |
2504 | } else { |
2505 | wp_page_reuse(vmf); | |
997dd98d | 2506 | lock_page(vmf->page); |
93e478d4 | 2507 | } |
997dd98d JK |
2508 | fault_dirty_shared_page(vma, vmf->page); |
2509 | put_page(vmf->page); | |
93e478d4 | 2510 | |
997dd98d | 2511 | return VM_FAULT_WRITE; |
93e478d4 SR |
2512 | } |
2513 | ||
1da177e4 LT |
2514 | /* |
2515 | * This routine handles present pages, when users try to write | |
2516 | * to a shared page. It is done by copying the page to a new address | |
2517 | * and decrementing the shared-page counter for the old page. | |
2518 | * | |
1da177e4 LT |
2519 | * Note that this routine assumes that the protection checks have been |
2520 | * done by the caller (the low-level page fault routine in most cases). | |
2521 | * Thus we can safely just mark it writable once we've done any necessary | |
2522 | * COW. | |
2523 | * | |
2524 | * We also mark the page dirty at this point even though the page will | |
2525 | * change only once the write actually happens. This avoids a few races, | |
2526 | * and potentially makes it more efficient. | |
2527 | * | |
8f4e2101 HD |
2528 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2529 | * but allow concurrent faults), with pte both mapped and locked. | |
2530 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2531 | */ |
2994302b | 2532 | static int do_wp_page(struct vm_fault *vmf) |
82b0f8c3 | 2533 | __releases(vmf->ptl) |
1da177e4 | 2534 | { |
82b0f8c3 | 2535 | struct vm_area_struct *vma = vmf->vma; |
1da177e4 | 2536 | |
a41b70d6 JK |
2537 | vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte); |
2538 | if (!vmf->page) { | |
251b97f5 | 2539 | /* |
64e45507 PF |
2540 | * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a |
2541 | * VM_PFNMAP VMA. | |
251b97f5 PZ |
2542 | * |
2543 | * We should not cow pages in a shared writeable mapping. | |
dd906184 | 2544 | * Just mark the pages writable and/or call ops->pfn_mkwrite. |
251b97f5 PZ |
2545 | */ |
2546 | if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == | |
2547 | (VM_WRITE|VM_SHARED)) | |
2994302b | 2548 | return wp_pfn_shared(vmf); |
2f38ab2c | 2549 | |
82b0f8c3 | 2550 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2551 | return wp_page_copy(vmf); |
251b97f5 | 2552 | } |
1da177e4 | 2553 | |
d08b3851 | 2554 | /* |
ee6a6457 PZ |
2555 | * Take out anonymous pages first, anonymous shared vmas are |
2556 | * not dirty accountable. | |
d08b3851 | 2557 | */ |
a41b70d6 | 2558 | if (PageAnon(vmf->page) && !PageKsm(vmf->page)) { |
6d0a07ed | 2559 | int total_mapcount; |
a41b70d6 JK |
2560 | if (!trylock_page(vmf->page)) { |
2561 | get_page(vmf->page); | |
82b0f8c3 | 2562 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2563 | lock_page(vmf->page); |
82b0f8c3 JK |
2564 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
2565 | vmf->address, &vmf->ptl); | |
2994302b | 2566 | if (!pte_same(*vmf->pte, vmf->orig_pte)) { |
a41b70d6 | 2567 | unlock_page(vmf->page); |
82b0f8c3 | 2568 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2569 | put_page(vmf->page); |
28766805 | 2570 | return 0; |
ab967d86 | 2571 | } |
a41b70d6 | 2572 | put_page(vmf->page); |
ee6a6457 | 2573 | } |
a41b70d6 | 2574 | if (reuse_swap_page(vmf->page, &total_mapcount)) { |
6d0a07ed AA |
2575 | if (total_mapcount == 1) { |
2576 | /* | |
2577 | * The page is all ours. Move it to | |
2578 | * our anon_vma so the rmap code will | |
2579 | * not search our parent or siblings. | |
2580 | * Protected against the rmap code by | |
2581 | * the page lock. | |
2582 | */ | |
a41b70d6 | 2583 | page_move_anon_rmap(vmf->page, vma); |
6d0a07ed | 2584 | } |
a41b70d6 | 2585 | unlock_page(vmf->page); |
997dd98d JK |
2586 | wp_page_reuse(vmf); |
2587 | return VM_FAULT_WRITE; | |
b009c024 | 2588 | } |
a41b70d6 | 2589 | unlock_page(vmf->page); |
ee6a6457 | 2590 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
d08b3851 | 2591 | (VM_WRITE|VM_SHARED))) { |
a41b70d6 | 2592 | return wp_page_shared(vmf); |
1da177e4 | 2593 | } |
1da177e4 LT |
2594 | |
2595 | /* | |
2596 | * Ok, we need to copy. Oh, well.. | |
2597 | */ | |
a41b70d6 | 2598 | get_page(vmf->page); |
28766805 | 2599 | |
82b0f8c3 | 2600 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
a41b70d6 | 2601 | return wp_page_copy(vmf); |
1da177e4 LT |
2602 | } |
2603 | ||
97a89413 | 2604 | static void unmap_mapping_range_vma(struct vm_area_struct *vma, |
1da177e4 LT |
2605 | unsigned long start_addr, unsigned long end_addr, |
2606 | struct zap_details *details) | |
2607 | { | |
f5cc4eef | 2608 | zap_page_range_single(vma, start_addr, end_addr - start_addr, details); |
1da177e4 LT |
2609 | } |
2610 | ||
6b2dbba8 | 2611 | static inline void unmap_mapping_range_tree(struct rb_root *root, |
1da177e4 LT |
2612 | struct zap_details *details) |
2613 | { | |
2614 | struct vm_area_struct *vma; | |
1da177e4 LT |
2615 | pgoff_t vba, vea, zba, zea; |
2616 | ||
6b2dbba8 | 2617 | vma_interval_tree_foreach(vma, root, |
1da177e4 | 2618 | details->first_index, details->last_index) { |
1da177e4 LT |
2619 | |
2620 | vba = vma->vm_pgoff; | |
d6e93217 | 2621 | vea = vba + vma_pages(vma) - 1; |
1da177e4 LT |
2622 | zba = details->first_index; |
2623 | if (zba < vba) | |
2624 | zba = vba; | |
2625 | zea = details->last_index; | |
2626 | if (zea > vea) | |
2627 | zea = vea; | |
2628 | ||
97a89413 | 2629 | unmap_mapping_range_vma(vma, |
1da177e4 LT |
2630 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, |
2631 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, | |
97a89413 | 2632 | details); |
1da177e4 LT |
2633 | } |
2634 | } | |
2635 | ||
1da177e4 | 2636 | /** |
8a5f14a2 KS |
2637 | * unmap_mapping_range - unmap the portion of all mmaps in the specified |
2638 | * address_space corresponding to the specified page range in the underlying | |
2639 | * file. | |
2640 | * | |
3d41088f | 2641 | * @mapping: the address space containing mmaps to be unmapped. |
1da177e4 LT |
2642 | * @holebegin: byte in first page to unmap, relative to the start of |
2643 | * the underlying file. This will be rounded down to a PAGE_SIZE | |
25d9e2d1 | 2644 | * boundary. Note that this is different from truncate_pagecache(), which |
1da177e4 LT |
2645 | * must keep the partial page. In contrast, we must get rid of |
2646 | * partial pages. | |
2647 | * @holelen: size of prospective hole in bytes. This will be rounded | |
2648 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the | |
2649 | * end of the file. | |
2650 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; | |
2651 | * but 0 when invalidating pagecache, don't throw away private data. | |
2652 | */ | |
2653 | void unmap_mapping_range(struct address_space *mapping, | |
2654 | loff_t const holebegin, loff_t const holelen, int even_cows) | |
2655 | { | |
aac45363 | 2656 | struct zap_details details = { }; |
1da177e4 LT |
2657 | pgoff_t hba = holebegin >> PAGE_SHIFT; |
2658 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2659 | ||
2660 | /* Check for overflow. */ | |
2661 | if (sizeof(holelen) > sizeof(hlen)) { | |
2662 | long long holeend = | |
2663 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2664 | if (holeend & ~(long long)ULONG_MAX) | |
2665 | hlen = ULONG_MAX - hba + 1; | |
2666 | } | |
2667 | ||
166f61b9 | 2668 | details.check_mapping = even_cows ? NULL : mapping; |
1da177e4 LT |
2669 | details.first_index = hba; |
2670 | details.last_index = hba + hlen - 1; | |
2671 | if (details.last_index < details.first_index) | |
2672 | details.last_index = ULONG_MAX; | |
1da177e4 | 2673 | |
46c043ed | 2674 | i_mmap_lock_write(mapping); |
6b2dbba8 | 2675 | if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap))) |
1da177e4 | 2676 | unmap_mapping_range_tree(&mapping->i_mmap, &details); |
46c043ed | 2677 | i_mmap_unlock_write(mapping); |
1da177e4 LT |
2678 | } |
2679 | EXPORT_SYMBOL(unmap_mapping_range); | |
2680 | ||
1da177e4 | 2681 | /* |
8f4e2101 HD |
2682 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2683 | * but allow concurrent faults), and pte mapped but not yet locked. | |
9a95f3cf PC |
2684 | * We return with pte unmapped and unlocked. |
2685 | * | |
2686 | * We return with the mmap_sem locked or unlocked in the same cases | |
2687 | * as does filemap_fault(). | |
1da177e4 | 2688 | */ |
2994302b | 2689 | int do_swap_page(struct vm_fault *vmf) |
1da177e4 | 2690 | { |
82b0f8c3 | 2691 | struct vm_area_struct *vma = vmf->vma; |
56f31801 | 2692 | struct page *page, *swapcache; |
00501b53 | 2693 | struct mem_cgroup *memcg; |
65500d23 | 2694 | swp_entry_t entry; |
1da177e4 | 2695 | pte_t pte; |
d065bd81 | 2696 | int locked; |
ad8c2ee8 | 2697 | int exclusive = 0; |
83c54070 | 2698 | int ret = 0; |
1da177e4 | 2699 | |
2994302b | 2700 | if (!pte_unmap_same(vma->vm_mm, vmf->pmd, vmf->pte, vmf->orig_pte)) |
8f4e2101 | 2701 | goto out; |
65500d23 | 2702 | |
2994302b | 2703 | entry = pte_to_swp_entry(vmf->orig_pte); |
d1737fdb AK |
2704 | if (unlikely(non_swap_entry(entry))) { |
2705 | if (is_migration_entry(entry)) { | |
82b0f8c3 JK |
2706 | migration_entry_wait(vma->vm_mm, vmf->pmd, |
2707 | vmf->address); | |
d1737fdb AK |
2708 | } else if (is_hwpoison_entry(entry)) { |
2709 | ret = VM_FAULT_HWPOISON; | |
2710 | } else { | |
2994302b | 2711 | print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL); |
d99be1a8 | 2712 | ret = VM_FAULT_SIGBUS; |
d1737fdb | 2713 | } |
0697212a CL |
2714 | goto out; |
2715 | } | |
0ff92245 | 2716 | delayacct_set_flag(DELAYACCT_PF_SWAPIN); |
1da177e4 LT |
2717 | page = lookup_swap_cache(entry); |
2718 | if (!page) { | |
82b0f8c3 JK |
2719 | page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, vma, |
2720 | vmf->address); | |
1da177e4 LT |
2721 | if (!page) { |
2722 | /* | |
8f4e2101 HD |
2723 | * Back out if somebody else faulted in this pte |
2724 | * while we released the pte lock. | |
1da177e4 | 2725 | */ |
82b0f8c3 JK |
2726 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
2727 | vmf->address, &vmf->ptl); | |
2994302b | 2728 | if (likely(pte_same(*vmf->pte, vmf->orig_pte))) |
1da177e4 | 2729 | ret = VM_FAULT_OOM; |
0ff92245 | 2730 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
65500d23 | 2731 | goto unlock; |
1da177e4 LT |
2732 | } |
2733 | ||
2734 | /* Had to read the page from swap area: Major fault */ | |
2735 | ret = VM_FAULT_MAJOR; | |
f8891e5e | 2736 | count_vm_event(PGMAJFAULT); |
2262185c | 2737 | count_memcg_event_mm(vma->vm_mm, PGMAJFAULT); |
d1737fdb | 2738 | } else if (PageHWPoison(page)) { |
71f72525 WF |
2739 | /* |
2740 | * hwpoisoned dirty swapcache pages are kept for killing | |
2741 | * owner processes (which may be unknown at hwpoison time) | |
2742 | */ | |
d1737fdb AK |
2743 | ret = VM_FAULT_HWPOISON; |
2744 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); | |
56f31801 | 2745 | swapcache = page; |
4779cb31 | 2746 | goto out_release; |
1da177e4 LT |
2747 | } |
2748 | ||
56f31801 | 2749 | swapcache = page; |
82b0f8c3 | 2750 | locked = lock_page_or_retry(page, vma->vm_mm, vmf->flags); |
e709ffd6 | 2751 | |
073e587e | 2752 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
d065bd81 ML |
2753 | if (!locked) { |
2754 | ret |= VM_FAULT_RETRY; | |
2755 | goto out_release; | |
2756 | } | |
073e587e | 2757 | |
4969c119 | 2758 | /* |
31c4a3d3 HD |
2759 | * Make sure try_to_free_swap or reuse_swap_page or swapoff did not |
2760 | * release the swapcache from under us. The page pin, and pte_same | |
2761 | * test below, are not enough to exclude that. Even if it is still | |
2762 | * swapcache, we need to check that the page's swap has not changed. | |
4969c119 | 2763 | */ |
31c4a3d3 | 2764 | if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val)) |
4969c119 AA |
2765 | goto out_page; |
2766 | ||
82b0f8c3 | 2767 | page = ksm_might_need_to_copy(page, vma, vmf->address); |
cbf86cfe HD |
2768 | if (unlikely(!page)) { |
2769 | ret = VM_FAULT_OOM; | |
2770 | page = swapcache; | |
cbf86cfe | 2771 | goto out_page; |
5ad64688 HD |
2772 | } |
2773 | ||
bae473a4 KS |
2774 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, |
2775 | &memcg, false)) { | |
8a9f3ccd | 2776 | ret = VM_FAULT_OOM; |
bc43f75c | 2777 | goto out_page; |
8a9f3ccd BS |
2778 | } |
2779 | ||
1da177e4 | 2780 | /* |
8f4e2101 | 2781 | * Back out if somebody else already faulted in this pte. |
1da177e4 | 2782 | */ |
82b0f8c3 JK |
2783 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
2784 | &vmf->ptl); | |
2994302b | 2785 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) |
b8107480 | 2786 | goto out_nomap; |
b8107480 KK |
2787 | |
2788 | if (unlikely(!PageUptodate(page))) { | |
2789 | ret = VM_FAULT_SIGBUS; | |
2790 | goto out_nomap; | |
1da177e4 LT |
2791 | } |
2792 | ||
8c7c6e34 KH |
2793 | /* |
2794 | * The page isn't present yet, go ahead with the fault. | |
2795 | * | |
2796 | * Be careful about the sequence of operations here. | |
2797 | * To get its accounting right, reuse_swap_page() must be called | |
2798 | * while the page is counted on swap but not yet in mapcount i.e. | |
2799 | * before page_add_anon_rmap() and swap_free(); try_to_free_swap() | |
2800 | * must be called after the swap_free(), or it will never succeed. | |
8c7c6e34 | 2801 | */ |
1da177e4 | 2802 | |
bae473a4 KS |
2803 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
2804 | dec_mm_counter_fast(vma->vm_mm, MM_SWAPENTS); | |
1da177e4 | 2805 | pte = mk_pte(page, vma->vm_page_prot); |
82b0f8c3 | 2806 | if ((vmf->flags & FAULT_FLAG_WRITE) && reuse_swap_page(page, NULL)) { |
1da177e4 | 2807 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); |
82b0f8c3 | 2808 | vmf->flags &= ~FAULT_FLAG_WRITE; |
9a5b489b | 2809 | ret |= VM_FAULT_WRITE; |
d281ee61 | 2810 | exclusive = RMAP_EXCLUSIVE; |
1da177e4 | 2811 | } |
1da177e4 | 2812 | flush_icache_page(vma, page); |
2994302b | 2813 | if (pte_swp_soft_dirty(vmf->orig_pte)) |
179ef71c | 2814 | pte = pte_mksoft_dirty(pte); |
82b0f8c3 | 2815 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte); |
2994302b | 2816 | vmf->orig_pte = pte; |
00501b53 | 2817 | if (page == swapcache) { |
82b0f8c3 | 2818 | do_page_add_anon_rmap(page, vma, vmf->address, exclusive); |
f627c2f5 | 2819 | mem_cgroup_commit_charge(page, memcg, true, false); |
1a8018fb | 2820 | activate_page(page); |
00501b53 | 2821 | } else { /* ksm created a completely new copy */ |
82b0f8c3 | 2822 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
f627c2f5 | 2823 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 JW |
2824 | lru_cache_add_active_or_unevictable(page, vma); |
2825 | } | |
1da177e4 | 2826 | |
c475a8ab | 2827 | swap_free(entry); |
5ccc5aba VD |
2828 | if (mem_cgroup_swap_full(page) || |
2829 | (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) | |
a2c43eed | 2830 | try_to_free_swap(page); |
c475a8ab | 2831 | unlock_page(page); |
56f31801 | 2832 | if (page != swapcache) { |
4969c119 AA |
2833 | /* |
2834 | * Hold the lock to avoid the swap entry to be reused | |
2835 | * until we take the PT lock for the pte_same() check | |
2836 | * (to avoid false positives from pte_same). For | |
2837 | * further safety release the lock after the swap_free | |
2838 | * so that the swap count won't change under a | |
2839 | * parallel locked swapcache. | |
2840 | */ | |
2841 | unlock_page(swapcache); | |
09cbfeaf | 2842 | put_page(swapcache); |
4969c119 | 2843 | } |
c475a8ab | 2844 | |
82b0f8c3 | 2845 | if (vmf->flags & FAULT_FLAG_WRITE) { |
2994302b | 2846 | ret |= do_wp_page(vmf); |
61469f1d HD |
2847 | if (ret & VM_FAULT_ERROR) |
2848 | ret &= VM_FAULT_ERROR; | |
1da177e4 LT |
2849 | goto out; |
2850 | } | |
2851 | ||
2852 | /* No need to invalidate - it was non-present before */ | |
82b0f8c3 | 2853 | update_mmu_cache(vma, vmf->address, vmf->pte); |
65500d23 | 2854 | unlock: |
82b0f8c3 | 2855 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
1da177e4 LT |
2856 | out: |
2857 | return ret; | |
b8107480 | 2858 | out_nomap: |
f627c2f5 | 2859 | mem_cgroup_cancel_charge(page, memcg, false); |
82b0f8c3 | 2860 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
bc43f75c | 2861 | out_page: |
b8107480 | 2862 | unlock_page(page); |
4779cb31 | 2863 | out_release: |
09cbfeaf | 2864 | put_page(page); |
56f31801 | 2865 | if (page != swapcache) { |
4969c119 | 2866 | unlock_page(swapcache); |
09cbfeaf | 2867 | put_page(swapcache); |
4969c119 | 2868 | } |
65500d23 | 2869 | return ret; |
1da177e4 LT |
2870 | } |
2871 | ||
2872 | /* | |
8f4e2101 HD |
2873 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2874 | * but allow concurrent faults), and pte mapped but not yet locked. | |
2875 | * We return with mmap_sem still held, but pte unmapped and unlocked. | |
1da177e4 | 2876 | */ |
82b0f8c3 | 2877 | static int do_anonymous_page(struct vm_fault *vmf) |
1da177e4 | 2878 | { |
82b0f8c3 | 2879 | struct vm_area_struct *vma = vmf->vma; |
00501b53 | 2880 | struct mem_cgroup *memcg; |
8f4e2101 | 2881 | struct page *page; |
1da177e4 | 2882 | pte_t entry; |
1da177e4 | 2883 | |
6b7339f4 KS |
2884 | /* File mapping without ->vm_ops ? */ |
2885 | if (vma->vm_flags & VM_SHARED) | |
2886 | return VM_FAULT_SIGBUS; | |
2887 | ||
7267ec00 KS |
2888 | /* |
2889 | * Use pte_alloc() instead of pte_alloc_map(). We can't run | |
2890 | * pte_offset_map() on pmds where a huge pmd might be created | |
2891 | * from a different thread. | |
2892 | * | |
2893 | * pte_alloc_map() is safe to use under down_write(mmap_sem) or when | |
2894 | * parallel threads are excluded by other means. | |
2895 | * | |
2896 | * Here we only have down_read(mmap_sem). | |
2897 | */ | |
82b0f8c3 | 2898 | if (pte_alloc(vma->vm_mm, vmf->pmd, vmf->address)) |
7267ec00 KS |
2899 | return VM_FAULT_OOM; |
2900 | ||
2901 | /* See the comment in pte_alloc_one_map() */ | |
82b0f8c3 | 2902 | if (unlikely(pmd_trans_unstable(vmf->pmd))) |
7267ec00 KS |
2903 | return 0; |
2904 | ||
11ac5524 | 2905 | /* Use the zero-page for reads */ |
82b0f8c3 | 2906 | if (!(vmf->flags & FAULT_FLAG_WRITE) && |
bae473a4 | 2907 | !mm_forbids_zeropage(vma->vm_mm)) { |
82b0f8c3 | 2908 | entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address), |
62eede62 | 2909 | vma->vm_page_prot)); |
82b0f8c3 JK |
2910 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, |
2911 | vmf->address, &vmf->ptl); | |
2912 | if (!pte_none(*vmf->pte)) | |
a13ea5b7 | 2913 | goto unlock; |
6b251fc9 AA |
2914 | /* Deliver the page fault to userland, check inside PT lock */ |
2915 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 JK |
2916 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
2917 | return handle_userfault(vmf, VM_UFFD_MISSING); | |
6b251fc9 | 2918 | } |
a13ea5b7 HD |
2919 | goto setpte; |
2920 | } | |
2921 | ||
557ed1fa | 2922 | /* Allocate our own private page. */ |
557ed1fa NP |
2923 | if (unlikely(anon_vma_prepare(vma))) |
2924 | goto oom; | |
82b0f8c3 | 2925 | page = alloc_zeroed_user_highpage_movable(vma, vmf->address); |
557ed1fa NP |
2926 | if (!page) |
2927 | goto oom; | |
eb3c24f3 | 2928 | |
bae473a4 | 2929 | if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, &memcg, false)) |
eb3c24f3 MG |
2930 | goto oom_free_page; |
2931 | ||
52f37629 MK |
2932 | /* |
2933 | * The memory barrier inside __SetPageUptodate makes sure that | |
2934 | * preceeding stores to the page contents become visible before | |
2935 | * the set_pte_at() write. | |
2936 | */ | |
0ed361de | 2937 | __SetPageUptodate(page); |
8f4e2101 | 2938 | |
557ed1fa | 2939 | entry = mk_pte(page, vma->vm_page_prot); |
1ac0cb5d HD |
2940 | if (vma->vm_flags & VM_WRITE) |
2941 | entry = pte_mkwrite(pte_mkdirty(entry)); | |
1da177e4 | 2942 | |
82b0f8c3 JK |
2943 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
2944 | &vmf->ptl); | |
2945 | if (!pte_none(*vmf->pte)) | |
557ed1fa | 2946 | goto release; |
9ba69294 | 2947 | |
6b251fc9 AA |
2948 | /* Deliver the page fault to userland, check inside PT lock */ |
2949 | if (userfaultfd_missing(vma)) { | |
82b0f8c3 | 2950 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
f627c2f5 | 2951 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 2952 | put_page(page); |
82b0f8c3 | 2953 | return handle_userfault(vmf, VM_UFFD_MISSING); |
6b251fc9 AA |
2954 | } |
2955 | ||
bae473a4 | 2956 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
82b0f8c3 | 2957 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
f627c2f5 | 2958 | mem_cgroup_commit_charge(page, memcg, false, false); |
00501b53 | 2959 | lru_cache_add_active_or_unevictable(page, vma); |
a13ea5b7 | 2960 | setpte: |
82b0f8c3 | 2961 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); |
1da177e4 LT |
2962 | |
2963 | /* No need to invalidate - it was non-present before */ | |
82b0f8c3 | 2964 | update_mmu_cache(vma, vmf->address, vmf->pte); |
65500d23 | 2965 | unlock: |
82b0f8c3 | 2966 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
83c54070 | 2967 | return 0; |
8f4e2101 | 2968 | release: |
f627c2f5 | 2969 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 2970 | put_page(page); |
8f4e2101 | 2971 | goto unlock; |
8a9f3ccd | 2972 | oom_free_page: |
09cbfeaf | 2973 | put_page(page); |
65500d23 | 2974 | oom: |
1da177e4 LT |
2975 | return VM_FAULT_OOM; |
2976 | } | |
2977 | ||
9a95f3cf PC |
2978 | /* |
2979 | * The mmap_sem must have been held on entry, and may have been | |
2980 | * released depending on flags and vma->vm_ops->fault() return value. | |
2981 | * See filemap_fault() and __lock_page_retry(). | |
2982 | */ | |
936ca80d | 2983 | static int __do_fault(struct vm_fault *vmf) |
7eae74af | 2984 | { |
82b0f8c3 | 2985 | struct vm_area_struct *vma = vmf->vma; |
7eae74af KS |
2986 | int ret; |
2987 | ||
11bac800 | 2988 | ret = vma->vm_ops->fault(vmf); |
3917048d | 2989 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY | |
b1aa812b | 2990 | VM_FAULT_DONE_COW))) |
bc2466e4 | 2991 | return ret; |
7eae74af | 2992 | |
667240e0 | 2993 | if (unlikely(PageHWPoison(vmf->page))) { |
7eae74af | 2994 | if (ret & VM_FAULT_LOCKED) |
667240e0 JK |
2995 | unlock_page(vmf->page); |
2996 | put_page(vmf->page); | |
936ca80d | 2997 | vmf->page = NULL; |
7eae74af KS |
2998 | return VM_FAULT_HWPOISON; |
2999 | } | |
3000 | ||
3001 | if (unlikely(!(ret & VM_FAULT_LOCKED))) | |
667240e0 | 3002 | lock_page(vmf->page); |
7eae74af | 3003 | else |
667240e0 | 3004 | VM_BUG_ON_PAGE(!PageLocked(vmf->page), vmf->page); |
7eae74af | 3005 | |
7eae74af KS |
3006 | return ret; |
3007 | } | |
3008 | ||
d0f0931d RZ |
3009 | /* |
3010 | * The ordering of these checks is important for pmds with _PAGE_DEVMAP set. | |
3011 | * If we check pmd_trans_unstable() first we will trip the bad_pmd() check | |
3012 | * inside of pmd_none_or_trans_huge_or_clear_bad(). This will end up correctly | |
3013 | * returning 1 but not before it spams dmesg with the pmd_clear_bad() output. | |
3014 | */ | |
3015 | static int pmd_devmap_trans_unstable(pmd_t *pmd) | |
3016 | { | |
3017 | return pmd_devmap(*pmd) || pmd_trans_unstable(pmd); | |
3018 | } | |
3019 | ||
82b0f8c3 | 3020 | static int pte_alloc_one_map(struct vm_fault *vmf) |
7267ec00 | 3021 | { |
82b0f8c3 | 3022 | struct vm_area_struct *vma = vmf->vma; |
7267ec00 | 3023 | |
82b0f8c3 | 3024 | if (!pmd_none(*vmf->pmd)) |
7267ec00 | 3025 | goto map_pte; |
82b0f8c3 JK |
3026 | if (vmf->prealloc_pte) { |
3027 | vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); | |
3028 | if (unlikely(!pmd_none(*vmf->pmd))) { | |
3029 | spin_unlock(vmf->ptl); | |
7267ec00 KS |
3030 | goto map_pte; |
3031 | } | |
3032 | ||
3033 | atomic_long_inc(&vma->vm_mm->nr_ptes); | |
82b0f8c3 JK |
3034 | pmd_populate(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); |
3035 | spin_unlock(vmf->ptl); | |
7f2b6ce8 | 3036 | vmf->prealloc_pte = NULL; |
82b0f8c3 | 3037 | } else if (unlikely(pte_alloc(vma->vm_mm, vmf->pmd, vmf->address))) { |
7267ec00 KS |
3038 | return VM_FAULT_OOM; |
3039 | } | |
3040 | map_pte: | |
3041 | /* | |
3042 | * If a huge pmd materialized under us just retry later. Use | |
d0f0931d RZ |
3043 | * pmd_trans_unstable() via pmd_devmap_trans_unstable() instead of |
3044 | * pmd_trans_huge() to ensure the pmd didn't become pmd_trans_huge | |
3045 | * under us and then back to pmd_none, as a result of MADV_DONTNEED | |
3046 | * running immediately after a huge pmd fault in a different thread of | |
3047 | * this mm, in turn leading to a misleading pmd_trans_huge() retval. | |
3048 | * All we have to ensure is that it is a regular pmd that we can walk | |
3049 | * with pte_offset_map() and we can do that through an atomic read in | |
3050 | * C, which is what pmd_trans_unstable() provides. | |
7267ec00 | 3051 | */ |
d0f0931d | 3052 | if (pmd_devmap_trans_unstable(vmf->pmd)) |
7267ec00 KS |
3053 | return VM_FAULT_NOPAGE; |
3054 | ||
d0f0931d RZ |
3055 | /* |
3056 | * At this point we know that our vmf->pmd points to a page of ptes | |
3057 | * and it cannot become pmd_none(), pmd_devmap() or pmd_trans_huge() | |
3058 | * for the duration of the fault. If a racing MADV_DONTNEED runs and | |
3059 | * we zap the ptes pointed to by our vmf->pmd, the vmf->ptl will still | |
3060 | * be valid and we will re-check to make sure the vmf->pte isn't | |
3061 | * pte_none() under vmf->ptl protection when we return to | |
3062 | * alloc_set_pte(). | |
3063 | */ | |
82b0f8c3 JK |
3064 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, |
3065 | &vmf->ptl); | |
7267ec00 KS |
3066 | return 0; |
3067 | } | |
3068 | ||
e496cf3d | 3069 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
10102459 KS |
3070 | |
3071 | #define HPAGE_CACHE_INDEX_MASK (HPAGE_PMD_NR - 1) | |
3072 | static inline bool transhuge_vma_suitable(struct vm_area_struct *vma, | |
3073 | unsigned long haddr) | |
3074 | { | |
3075 | if (((vma->vm_start >> PAGE_SHIFT) & HPAGE_CACHE_INDEX_MASK) != | |
3076 | (vma->vm_pgoff & HPAGE_CACHE_INDEX_MASK)) | |
3077 | return false; | |
3078 | if (haddr < vma->vm_start || haddr + HPAGE_PMD_SIZE > vma->vm_end) | |
3079 | return false; | |
3080 | return true; | |
3081 | } | |
3082 | ||
82b0f8c3 | 3083 | static void deposit_prealloc_pte(struct vm_fault *vmf) |
953c66c2 | 3084 | { |
82b0f8c3 | 3085 | struct vm_area_struct *vma = vmf->vma; |
953c66c2 | 3086 | |
82b0f8c3 | 3087 | pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, vmf->prealloc_pte); |
953c66c2 AK |
3088 | /* |
3089 | * We are going to consume the prealloc table, | |
3090 | * count that as nr_ptes. | |
3091 | */ | |
3092 | atomic_long_inc(&vma->vm_mm->nr_ptes); | |
7f2b6ce8 | 3093 | vmf->prealloc_pte = NULL; |
953c66c2 AK |
3094 | } |
3095 | ||
82b0f8c3 | 3096 | static int do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 | 3097 | { |
82b0f8c3 JK |
3098 | struct vm_area_struct *vma = vmf->vma; |
3099 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
3100 | unsigned long haddr = vmf->address & HPAGE_PMD_MASK; | |
10102459 KS |
3101 | pmd_t entry; |
3102 | int i, ret; | |
3103 | ||
3104 | if (!transhuge_vma_suitable(vma, haddr)) | |
3105 | return VM_FAULT_FALLBACK; | |
3106 | ||
3107 | ret = VM_FAULT_FALLBACK; | |
3108 | page = compound_head(page); | |
3109 | ||
953c66c2 AK |
3110 | /* |
3111 | * Archs like ppc64 need additonal space to store information | |
3112 | * related to pte entry. Use the preallocated table for that. | |
3113 | */ | |
82b0f8c3 JK |
3114 | if (arch_needs_pgtable_deposit() && !vmf->prealloc_pte) { |
3115 | vmf->prealloc_pte = pte_alloc_one(vma->vm_mm, vmf->address); | |
3116 | if (!vmf->prealloc_pte) | |
953c66c2 AK |
3117 | return VM_FAULT_OOM; |
3118 | smp_wmb(); /* See comment in __pte_alloc() */ | |
3119 | } | |
3120 | ||
82b0f8c3 JK |
3121 | vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd); |
3122 | if (unlikely(!pmd_none(*vmf->pmd))) | |
10102459 KS |
3123 | goto out; |
3124 | ||
3125 | for (i = 0; i < HPAGE_PMD_NR; i++) | |
3126 | flush_icache_page(vma, page + i); | |
3127 | ||
3128 | entry = mk_huge_pmd(page, vma->vm_page_prot); | |
3129 | if (write) | |
3130 | entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma); | |
3131 | ||
3132 | add_mm_counter(vma->vm_mm, MM_FILEPAGES, HPAGE_PMD_NR); | |
3133 | page_add_file_rmap(page, true); | |
953c66c2 AK |
3134 | /* |
3135 | * deposit and withdraw with pmd lock held | |
3136 | */ | |
3137 | if (arch_needs_pgtable_deposit()) | |
82b0f8c3 | 3138 | deposit_prealloc_pte(vmf); |
10102459 | 3139 | |
82b0f8c3 | 3140 | set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry); |
10102459 | 3141 | |
82b0f8c3 | 3142 | update_mmu_cache_pmd(vma, haddr, vmf->pmd); |
10102459 KS |
3143 | |
3144 | /* fault is handled */ | |
3145 | ret = 0; | |
95ecedcd | 3146 | count_vm_event(THP_FILE_MAPPED); |
10102459 | 3147 | out: |
82b0f8c3 | 3148 | spin_unlock(vmf->ptl); |
10102459 KS |
3149 | return ret; |
3150 | } | |
3151 | #else | |
82b0f8c3 | 3152 | static int do_set_pmd(struct vm_fault *vmf, struct page *page) |
10102459 KS |
3153 | { |
3154 | BUILD_BUG(); | |
3155 | return 0; | |
3156 | } | |
3157 | #endif | |
3158 | ||
8c6e50b0 | 3159 | /** |
7267ec00 KS |
3160 | * alloc_set_pte - setup new PTE entry for given page and add reverse page |
3161 | * mapping. If needed, the fucntion allocates page table or use pre-allocated. | |
8c6e50b0 | 3162 | * |
82b0f8c3 | 3163 | * @vmf: fault environment |
7267ec00 | 3164 | * @memcg: memcg to charge page (only for private mappings) |
8c6e50b0 | 3165 | * @page: page to map |
8c6e50b0 | 3166 | * |
82b0f8c3 JK |
3167 | * Caller must take care of unlocking vmf->ptl, if vmf->pte is non-NULL on |
3168 | * return. | |
8c6e50b0 KS |
3169 | * |
3170 | * Target users are page handler itself and implementations of | |
3171 | * vm_ops->map_pages. | |
3172 | */ | |
82b0f8c3 | 3173 | int alloc_set_pte(struct vm_fault *vmf, struct mem_cgroup *memcg, |
7267ec00 | 3174 | struct page *page) |
3bb97794 | 3175 | { |
82b0f8c3 JK |
3176 | struct vm_area_struct *vma = vmf->vma; |
3177 | bool write = vmf->flags & FAULT_FLAG_WRITE; | |
3bb97794 | 3178 | pte_t entry; |
10102459 KS |
3179 | int ret; |
3180 | ||
82b0f8c3 | 3181 | if (pmd_none(*vmf->pmd) && PageTransCompound(page) && |
e496cf3d | 3182 | IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) { |
10102459 KS |
3183 | /* THP on COW? */ |
3184 | VM_BUG_ON_PAGE(memcg, page); | |
3185 | ||
82b0f8c3 | 3186 | ret = do_set_pmd(vmf, page); |
10102459 | 3187 | if (ret != VM_FAULT_FALLBACK) |
b0b9b3df | 3188 | return ret; |
10102459 | 3189 | } |
3bb97794 | 3190 | |
82b0f8c3 JK |
3191 | if (!vmf->pte) { |
3192 | ret = pte_alloc_one_map(vmf); | |
7267ec00 | 3193 | if (ret) |
b0b9b3df | 3194 | return ret; |
7267ec00 KS |
3195 | } |
3196 | ||
3197 | /* Re-check under ptl */ | |
b0b9b3df HD |
3198 | if (unlikely(!pte_none(*vmf->pte))) |
3199 | return VM_FAULT_NOPAGE; | |
7267ec00 | 3200 | |
3bb97794 KS |
3201 | flush_icache_page(vma, page); |
3202 | entry = mk_pte(page, vma->vm_page_prot); | |
3203 | if (write) | |
3204 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); | |
bae473a4 KS |
3205 | /* copy-on-write page */ |
3206 | if (write && !(vma->vm_flags & VM_SHARED)) { | |
3bb97794 | 3207 | inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); |
82b0f8c3 | 3208 | page_add_new_anon_rmap(page, vma, vmf->address, false); |
7267ec00 KS |
3209 | mem_cgroup_commit_charge(page, memcg, false, false); |
3210 | lru_cache_add_active_or_unevictable(page, vma); | |
3bb97794 | 3211 | } else { |
eca56ff9 | 3212 | inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page)); |
dd78fedd | 3213 | page_add_file_rmap(page, false); |
3bb97794 | 3214 | } |
82b0f8c3 | 3215 | set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry); |
3bb97794 KS |
3216 | |
3217 | /* no need to invalidate: a not-present page won't be cached */ | |
82b0f8c3 | 3218 | update_mmu_cache(vma, vmf->address, vmf->pte); |
7267ec00 | 3219 | |
b0b9b3df | 3220 | return 0; |
3bb97794 KS |
3221 | } |
3222 | ||
9118c0cb JK |
3223 | |
3224 | /** | |
3225 | * finish_fault - finish page fault once we have prepared the page to fault | |
3226 | * | |
3227 | * @vmf: structure describing the fault | |
3228 | * | |
3229 | * This function handles all that is needed to finish a page fault once the | |
3230 | * page to fault in is prepared. It handles locking of PTEs, inserts PTE for | |
3231 | * given page, adds reverse page mapping, handles memcg charges and LRU | |
3232 | * addition. The function returns 0 on success, VM_FAULT_ code in case of | |
3233 | * error. | |
3234 | * | |
3235 | * The function expects the page to be locked and on success it consumes a | |
3236 | * reference of a page being mapped (for the PTE which maps it). | |
3237 | */ | |
3238 | int finish_fault(struct vm_fault *vmf) | |
3239 | { | |
3240 | struct page *page; | |
3241 | int ret; | |
3242 | ||
3243 | /* Did we COW the page? */ | |
3244 | if ((vmf->flags & FAULT_FLAG_WRITE) && | |
3245 | !(vmf->vma->vm_flags & VM_SHARED)) | |
3246 | page = vmf->cow_page; | |
3247 | else | |
3248 | page = vmf->page; | |
3249 | ret = alloc_set_pte(vmf, vmf->memcg, page); | |
3250 | if (vmf->pte) | |
3251 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
3252 | return ret; | |
3253 | } | |
3254 | ||
3a91053a KS |
3255 | static unsigned long fault_around_bytes __read_mostly = |
3256 | rounddown_pow_of_two(65536); | |
a9b0f861 | 3257 | |
a9b0f861 KS |
3258 | #ifdef CONFIG_DEBUG_FS |
3259 | static int fault_around_bytes_get(void *data, u64 *val) | |
1592eef0 | 3260 | { |
a9b0f861 | 3261 | *val = fault_around_bytes; |
1592eef0 KS |
3262 | return 0; |
3263 | } | |
3264 | ||
b4903d6e AR |
3265 | /* |
3266 | * fault_around_pages() and fault_around_mask() expects fault_around_bytes | |
3267 | * rounded down to nearest page order. It's what do_fault_around() expects to | |
3268 | * see. | |
3269 | */ | |
a9b0f861 | 3270 | static int fault_around_bytes_set(void *data, u64 val) |
1592eef0 | 3271 | { |
a9b0f861 | 3272 | if (val / PAGE_SIZE > PTRS_PER_PTE) |
1592eef0 | 3273 | return -EINVAL; |
b4903d6e AR |
3274 | if (val > PAGE_SIZE) |
3275 | fault_around_bytes = rounddown_pow_of_two(val); | |
3276 | else | |
3277 | fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */ | |
1592eef0 KS |
3278 | return 0; |
3279 | } | |
0a1345f8 | 3280 | DEFINE_DEBUGFS_ATTRIBUTE(fault_around_bytes_fops, |
a9b0f861 | 3281 | fault_around_bytes_get, fault_around_bytes_set, "%llu\n"); |
1592eef0 KS |
3282 | |
3283 | static int __init fault_around_debugfs(void) | |
3284 | { | |
3285 | void *ret; | |
3286 | ||
0a1345f8 | 3287 | ret = debugfs_create_file_unsafe("fault_around_bytes", 0644, NULL, NULL, |
a9b0f861 | 3288 | &fault_around_bytes_fops); |
1592eef0 | 3289 | if (!ret) |
a9b0f861 | 3290 | pr_warn("Failed to create fault_around_bytes in debugfs"); |
1592eef0 KS |
3291 | return 0; |
3292 | } | |
3293 | late_initcall(fault_around_debugfs); | |
1592eef0 | 3294 | #endif |
8c6e50b0 | 3295 | |
1fdb412b KS |
3296 | /* |
3297 | * do_fault_around() tries to map few pages around the fault address. The hope | |
3298 | * is that the pages will be needed soon and this will lower the number of | |
3299 | * faults to handle. | |
3300 | * | |
3301 | * It uses vm_ops->map_pages() to map the pages, which skips the page if it's | |
3302 | * not ready to be mapped: not up-to-date, locked, etc. | |
3303 | * | |
3304 | * This function is called with the page table lock taken. In the split ptlock | |
3305 | * case the page table lock only protects only those entries which belong to | |
3306 | * the page table corresponding to the fault address. | |
3307 | * | |
3308 | * This function doesn't cross the VMA boundaries, in order to call map_pages() | |
3309 | * only once. | |
3310 | * | |
3311 | * fault_around_pages() defines how many pages we'll try to map. | |
3312 | * do_fault_around() expects it to return a power of two less than or equal to | |
3313 | * PTRS_PER_PTE. | |
3314 | * | |
3315 | * The virtual address of the area that we map is naturally aligned to the | |
3316 | * fault_around_pages() value (and therefore to page order). This way it's | |
3317 | * easier to guarantee that we don't cross page table boundaries. | |
3318 | */ | |
0721ec8b | 3319 | static int do_fault_around(struct vm_fault *vmf) |
8c6e50b0 | 3320 | { |
82b0f8c3 | 3321 | unsigned long address = vmf->address, nr_pages, mask; |
0721ec8b | 3322 | pgoff_t start_pgoff = vmf->pgoff; |
bae473a4 | 3323 | pgoff_t end_pgoff; |
7267ec00 | 3324 | int off, ret = 0; |
8c6e50b0 | 3325 | |
4db0c3c2 | 3326 | nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT; |
aecd6f44 KS |
3327 | mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK; |
3328 | ||
82b0f8c3 JK |
3329 | vmf->address = max(address & mask, vmf->vma->vm_start); |
3330 | off = ((address - vmf->address) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1); | |
bae473a4 | 3331 | start_pgoff -= off; |
8c6e50b0 KS |
3332 | |
3333 | /* | |
bae473a4 KS |
3334 | * end_pgoff is either end of page table or end of vma |
3335 | * or fault_around_pages() from start_pgoff, depending what is nearest. | |
8c6e50b0 | 3336 | */ |
bae473a4 | 3337 | end_pgoff = start_pgoff - |
82b0f8c3 | 3338 | ((vmf->address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) + |
8c6e50b0 | 3339 | PTRS_PER_PTE - 1; |
82b0f8c3 | 3340 | end_pgoff = min3(end_pgoff, vma_pages(vmf->vma) + vmf->vma->vm_pgoff - 1, |
bae473a4 | 3341 | start_pgoff + nr_pages - 1); |
8c6e50b0 | 3342 | |
82b0f8c3 JK |
3343 | if (pmd_none(*vmf->pmd)) { |
3344 | vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm, | |
3345 | vmf->address); | |
3346 | if (!vmf->prealloc_pte) | |
c5f88bd2 | 3347 | goto out; |
7267ec00 | 3348 | smp_wmb(); /* See comment in __pte_alloc() */ |
8c6e50b0 KS |
3349 | } |
3350 | ||
82b0f8c3 | 3351 | vmf->vma->vm_ops->map_pages(vmf, start_pgoff, end_pgoff); |
7267ec00 | 3352 | |
7267ec00 | 3353 | /* Huge page is mapped? Page fault is solved */ |
82b0f8c3 | 3354 | if (pmd_trans_huge(*vmf->pmd)) { |
7267ec00 KS |
3355 | ret = VM_FAULT_NOPAGE; |
3356 | goto out; | |
3357 | } | |
3358 | ||
3359 | /* ->map_pages() haven't done anything useful. Cold page cache? */ | |
82b0f8c3 | 3360 | if (!vmf->pte) |
7267ec00 KS |
3361 | goto out; |
3362 | ||
3363 | /* check if the page fault is solved */ | |
82b0f8c3 JK |
3364 | vmf->pte -= (vmf->address >> PAGE_SHIFT) - (address >> PAGE_SHIFT); |
3365 | if (!pte_none(*vmf->pte)) | |
7267ec00 | 3366 | ret = VM_FAULT_NOPAGE; |
82b0f8c3 | 3367 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
bae473a4 | 3368 | out: |
82b0f8c3 JK |
3369 | vmf->address = address; |
3370 | vmf->pte = NULL; | |
7267ec00 | 3371 | return ret; |
8c6e50b0 KS |
3372 | } |
3373 | ||
0721ec8b | 3374 | static int do_read_fault(struct vm_fault *vmf) |
e655fb29 | 3375 | { |
82b0f8c3 | 3376 | struct vm_area_struct *vma = vmf->vma; |
8c6e50b0 KS |
3377 | int ret = 0; |
3378 | ||
3379 | /* | |
3380 | * Let's call ->map_pages() first and use ->fault() as fallback | |
3381 | * if page by the offset is not ready to be mapped (cold cache or | |
3382 | * something). | |
3383 | */ | |
9b4bdd2f | 3384 | if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) { |
0721ec8b | 3385 | ret = do_fault_around(vmf); |
7267ec00 KS |
3386 | if (ret) |
3387 | return ret; | |
8c6e50b0 | 3388 | } |
e655fb29 | 3389 | |
936ca80d | 3390 | ret = __do_fault(vmf); |
e655fb29 KS |
3391 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3392 | return ret; | |
3393 | ||
9118c0cb | 3394 | ret |= finish_fault(vmf); |
936ca80d | 3395 | unlock_page(vmf->page); |
7267ec00 | 3396 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
936ca80d | 3397 | put_page(vmf->page); |
e655fb29 KS |
3398 | return ret; |
3399 | } | |
3400 | ||
0721ec8b | 3401 | static int do_cow_fault(struct vm_fault *vmf) |
ec47c3b9 | 3402 | { |
82b0f8c3 | 3403 | struct vm_area_struct *vma = vmf->vma; |
ec47c3b9 KS |
3404 | int ret; |
3405 | ||
3406 | if (unlikely(anon_vma_prepare(vma))) | |
3407 | return VM_FAULT_OOM; | |
3408 | ||
936ca80d JK |
3409 | vmf->cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vmf->address); |
3410 | if (!vmf->cow_page) | |
ec47c3b9 KS |
3411 | return VM_FAULT_OOM; |
3412 | ||
936ca80d | 3413 | if (mem_cgroup_try_charge(vmf->cow_page, vma->vm_mm, GFP_KERNEL, |
3917048d | 3414 | &vmf->memcg, false)) { |
936ca80d | 3415 | put_page(vmf->cow_page); |
ec47c3b9 KS |
3416 | return VM_FAULT_OOM; |
3417 | } | |
3418 | ||
936ca80d | 3419 | ret = __do_fault(vmf); |
ec47c3b9 KS |
3420 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3421 | goto uncharge_out; | |
3917048d JK |
3422 | if (ret & VM_FAULT_DONE_COW) |
3423 | return ret; | |
ec47c3b9 | 3424 | |
b1aa812b | 3425 | copy_user_highpage(vmf->cow_page, vmf->page, vmf->address, vma); |
936ca80d | 3426 | __SetPageUptodate(vmf->cow_page); |
ec47c3b9 | 3427 | |
9118c0cb | 3428 | ret |= finish_fault(vmf); |
b1aa812b JK |
3429 | unlock_page(vmf->page); |
3430 | put_page(vmf->page); | |
7267ec00 KS |
3431 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
3432 | goto uncharge_out; | |
ec47c3b9 KS |
3433 | return ret; |
3434 | uncharge_out: | |
3917048d | 3435 | mem_cgroup_cancel_charge(vmf->cow_page, vmf->memcg, false); |
936ca80d | 3436 | put_page(vmf->cow_page); |
ec47c3b9 KS |
3437 | return ret; |
3438 | } | |
3439 | ||
0721ec8b | 3440 | static int do_shared_fault(struct vm_fault *vmf) |
1da177e4 | 3441 | { |
82b0f8c3 | 3442 | struct vm_area_struct *vma = vmf->vma; |
f0c6d4d2 | 3443 | int ret, tmp; |
1d65f86d | 3444 | |
936ca80d | 3445 | ret = __do_fault(vmf); |
7eae74af | 3446 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY))) |
f0c6d4d2 | 3447 | return ret; |
1da177e4 LT |
3448 | |
3449 | /* | |
f0c6d4d2 KS |
3450 | * Check if the backing address space wants to know that the page is |
3451 | * about to become writable | |
1da177e4 | 3452 | */ |
fb09a464 | 3453 | if (vma->vm_ops->page_mkwrite) { |
936ca80d | 3454 | unlock_page(vmf->page); |
38b8cb7f | 3455 | tmp = do_page_mkwrite(vmf); |
fb09a464 KS |
3456 | if (unlikely(!tmp || |
3457 | (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) { | |
936ca80d | 3458 | put_page(vmf->page); |
fb09a464 | 3459 | return tmp; |
4294621f | 3460 | } |
fb09a464 KS |
3461 | } |
3462 | ||
9118c0cb | 3463 | ret |= finish_fault(vmf); |
7267ec00 KS |
3464 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | |
3465 | VM_FAULT_RETRY))) { | |
936ca80d JK |
3466 | unlock_page(vmf->page); |
3467 | put_page(vmf->page); | |
f0c6d4d2 | 3468 | return ret; |
1da177e4 | 3469 | } |
b827e496 | 3470 | |
97ba0c2b | 3471 | fault_dirty_shared_page(vma, vmf->page); |
1d65f86d | 3472 | return ret; |
54cb8821 | 3473 | } |
d00806b1 | 3474 | |
9a95f3cf PC |
3475 | /* |
3476 | * We enter with non-exclusive mmap_sem (to exclude vma changes, | |
3477 | * but allow concurrent faults). | |
3478 | * The mmap_sem may have been released depending on flags and our | |
3479 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
3480 | */ | |
82b0f8c3 | 3481 | static int do_fault(struct vm_fault *vmf) |
54cb8821 | 3482 | { |
82b0f8c3 | 3483 | struct vm_area_struct *vma = vmf->vma; |
b0b9b3df | 3484 | int ret; |
54cb8821 | 3485 | |
6b7339f4 KS |
3486 | /* The VMA was not fully populated on mmap() or missing VM_DONTEXPAND */ |
3487 | if (!vma->vm_ops->fault) | |
b0b9b3df HD |
3488 | ret = VM_FAULT_SIGBUS; |
3489 | else if (!(vmf->flags & FAULT_FLAG_WRITE)) | |
3490 | ret = do_read_fault(vmf); | |
3491 | else if (!(vma->vm_flags & VM_SHARED)) | |
3492 | ret = do_cow_fault(vmf); | |
3493 | else | |
3494 | ret = do_shared_fault(vmf); | |
3495 | ||
3496 | /* preallocated pagetable is unused: free it */ | |
3497 | if (vmf->prealloc_pte) { | |
3498 | pte_free(vma->vm_mm, vmf->prealloc_pte); | |
7f2b6ce8 | 3499 | vmf->prealloc_pte = NULL; |
b0b9b3df HD |
3500 | } |
3501 | return ret; | |
54cb8821 NP |
3502 | } |
3503 | ||
b19a9939 | 3504 | static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, |
04bb2f94 RR |
3505 | unsigned long addr, int page_nid, |
3506 | int *flags) | |
9532fec1 MG |
3507 | { |
3508 | get_page(page); | |
3509 | ||
3510 | count_vm_numa_event(NUMA_HINT_FAULTS); | |
04bb2f94 | 3511 | if (page_nid == numa_node_id()) { |
9532fec1 | 3512 | count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL); |
04bb2f94 RR |
3513 | *flags |= TNF_FAULT_LOCAL; |
3514 | } | |
9532fec1 MG |
3515 | |
3516 | return mpol_misplaced(page, vma, addr); | |
3517 | } | |
3518 | ||
2994302b | 3519 | static int do_numa_page(struct vm_fault *vmf) |
d10e63f2 | 3520 | { |
82b0f8c3 | 3521 | struct vm_area_struct *vma = vmf->vma; |
4daae3b4 | 3522 | struct page *page = NULL; |
8191acbd | 3523 | int page_nid = -1; |
90572890 | 3524 | int last_cpupid; |
cbee9f88 | 3525 | int target_nid; |
b8593bfd | 3526 | bool migrated = false; |
cee216a6 | 3527 | pte_t pte; |
288bc549 | 3528 | bool was_writable = pte_savedwrite(vmf->orig_pte); |
6688cc05 | 3529 | int flags = 0; |
d10e63f2 MG |
3530 | |
3531 | /* | |
166f61b9 TH |
3532 | * The "pte" at this point cannot be used safely without |
3533 | * validation through pte_unmap_same(). It's of NUMA type but | |
3534 | * the pfn may be screwed if the read is non atomic. | |
166f61b9 | 3535 | */ |
82b0f8c3 JK |
3536 | vmf->ptl = pte_lockptr(vma->vm_mm, vmf->pmd); |
3537 | spin_lock(vmf->ptl); | |
cee216a6 | 3538 | if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) { |
82b0f8c3 | 3539 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4daae3b4 MG |
3540 | goto out; |
3541 | } | |
3542 | ||
cee216a6 AK |
3543 | /* |
3544 | * Make it present again, Depending on how arch implementes non | |
3545 | * accessible ptes, some can allow access by kernel mode. | |
3546 | */ | |
3547 | pte = ptep_modify_prot_start(vma->vm_mm, vmf->address, vmf->pte); | |
4d942466 MG |
3548 | pte = pte_modify(pte, vma->vm_page_prot); |
3549 | pte = pte_mkyoung(pte); | |
b191f9b1 MG |
3550 | if (was_writable) |
3551 | pte = pte_mkwrite(pte); | |
cee216a6 | 3552 | ptep_modify_prot_commit(vma->vm_mm, vmf->address, vmf->pte, pte); |
82b0f8c3 | 3553 | update_mmu_cache(vma, vmf->address, vmf->pte); |
d10e63f2 | 3554 | |
82b0f8c3 | 3555 | page = vm_normal_page(vma, vmf->address, pte); |
d10e63f2 | 3556 | if (!page) { |
82b0f8c3 | 3557 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
d10e63f2 MG |
3558 | return 0; |
3559 | } | |
3560 | ||
e81c4802 KS |
3561 | /* TODO: handle PTE-mapped THP */ |
3562 | if (PageCompound(page)) { | |
82b0f8c3 | 3563 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
e81c4802 KS |
3564 | return 0; |
3565 | } | |
3566 | ||
6688cc05 | 3567 | /* |
bea66fbd MG |
3568 | * Avoid grouping on RO pages in general. RO pages shouldn't hurt as |
3569 | * much anyway since they can be in shared cache state. This misses | |
3570 | * the case where a mapping is writable but the process never writes | |
3571 | * to it but pte_write gets cleared during protection updates and | |
3572 | * pte_dirty has unpredictable behaviour between PTE scan updates, | |
3573 | * background writeback, dirty balancing and application behaviour. | |
6688cc05 | 3574 | */ |
d59dc7bc | 3575 | if (!pte_write(pte)) |
6688cc05 PZ |
3576 | flags |= TNF_NO_GROUP; |
3577 | ||
dabe1d99 RR |
3578 | /* |
3579 | * Flag if the page is shared between multiple address spaces. This | |
3580 | * is later used when determining whether to group tasks together | |
3581 | */ | |
3582 | if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED)) | |
3583 | flags |= TNF_SHARED; | |
3584 | ||
90572890 | 3585 | last_cpupid = page_cpupid_last(page); |
8191acbd | 3586 | page_nid = page_to_nid(page); |
82b0f8c3 | 3587 | target_nid = numa_migrate_prep(page, vma, vmf->address, page_nid, |
bae473a4 | 3588 | &flags); |
82b0f8c3 | 3589 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
4daae3b4 | 3590 | if (target_nid == -1) { |
4daae3b4 MG |
3591 | put_page(page); |
3592 | goto out; | |
3593 | } | |
3594 | ||
3595 | /* Migrate to the requested node */ | |
1bc115d8 | 3596 | migrated = migrate_misplaced_page(page, vma, target_nid); |
6688cc05 | 3597 | if (migrated) { |
8191acbd | 3598 | page_nid = target_nid; |
6688cc05 | 3599 | flags |= TNF_MIGRATED; |
074c2381 MG |
3600 | } else |
3601 | flags |= TNF_MIGRATE_FAIL; | |
4daae3b4 MG |
3602 | |
3603 | out: | |
8191acbd | 3604 | if (page_nid != -1) |
6688cc05 | 3605 | task_numa_fault(last_cpupid, page_nid, 1, flags); |
d10e63f2 MG |
3606 | return 0; |
3607 | } | |
3608 | ||
91a90140 | 3609 | static inline int create_huge_pmd(struct vm_fault *vmf) |
b96375f7 | 3610 | { |
f4200391 | 3611 | if (vma_is_anonymous(vmf->vma)) |
82b0f8c3 | 3612 | return do_huge_pmd_anonymous_page(vmf); |
a2d58167 | 3613 | if (vmf->vma->vm_ops->huge_fault) |
c791ace1 | 3614 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); |
b96375f7 MW |
3615 | return VM_FAULT_FALLBACK; |
3616 | } | |
3617 | ||
82b0f8c3 | 3618 | static int wp_huge_pmd(struct vm_fault *vmf, pmd_t orig_pmd) |
b96375f7 | 3619 | { |
82b0f8c3 JK |
3620 | if (vma_is_anonymous(vmf->vma)) |
3621 | return do_huge_pmd_wp_page(vmf, orig_pmd); | |
a2d58167 | 3622 | if (vmf->vma->vm_ops->huge_fault) |
c791ace1 | 3623 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD); |
af9e4d5f KS |
3624 | |
3625 | /* COW handled on pte level: split pmd */ | |
82b0f8c3 JK |
3626 | VM_BUG_ON_VMA(vmf->vma->vm_flags & VM_SHARED, vmf->vma); |
3627 | __split_huge_pmd(vmf->vma, vmf->pmd, vmf->address, false, NULL); | |
af9e4d5f | 3628 | |
b96375f7 MW |
3629 | return VM_FAULT_FALLBACK; |
3630 | } | |
3631 | ||
38e08854 LS |
3632 | static inline bool vma_is_accessible(struct vm_area_struct *vma) |
3633 | { | |
3634 | return vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE); | |
3635 | } | |
3636 | ||
a00cc7d9 MW |
3637 | static int create_huge_pud(struct vm_fault *vmf) |
3638 | { | |
3639 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
3640 | /* No support for anonymous transparent PUD pages yet */ | |
3641 | if (vma_is_anonymous(vmf->vma)) | |
3642 | return VM_FAULT_FALLBACK; | |
3643 | if (vmf->vma->vm_ops->huge_fault) | |
c791ace1 | 3644 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); |
a00cc7d9 MW |
3645 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
3646 | return VM_FAULT_FALLBACK; | |
3647 | } | |
3648 | ||
3649 | static int wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud) | |
3650 | { | |
3651 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
3652 | /* No support for anonymous transparent PUD pages yet */ | |
3653 | if (vma_is_anonymous(vmf->vma)) | |
3654 | return VM_FAULT_FALLBACK; | |
3655 | if (vmf->vma->vm_ops->huge_fault) | |
c791ace1 | 3656 | return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD); |
a00cc7d9 MW |
3657 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
3658 | return VM_FAULT_FALLBACK; | |
3659 | } | |
3660 | ||
1da177e4 LT |
3661 | /* |
3662 | * These routines also need to handle stuff like marking pages dirty | |
3663 | * and/or accessed for architectures that don't do it in hardware (most | |
3664 | * RISC architectures). The early dirtying is also good on the i386. | |
3665 | * | |
3666 | * There is also a hook called "update_mmu_cache()" that architectures | |
3667 | * with external mmu caches can use to update those (ie the Sparc or | |
3668 | * PowerPC hashed page tables that act as extended TLBs). | |
3669 | * | |
7267ec00 KS |
3670 | * We enter with non-exclusive mmap_sem (to exclude vma changes, but allow |
3671 | * concurrent faults). | |
9a95f3cf | 3672 | * |
7267ec00 KS |
3673 | * The mmap_sem may have been released depending on flags and our return value. |
3674 | * See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 3675 | */ |
82b0f8c3 | 3676 | static int handle_pte_fault(struct vm_fault *vmf) |
1da177e4 LT |
3677 | { |
3678 | pte_t entry; | |
3679 | ||
82b0f8c3 | 3680 | if (unlikely(pmd_none(*vmf->pmd))) { |
7267ec00 KS |
3681 | /* |
3682 | * Leave __pte_alloc() until later: because vm_ops->fault may | |
3683 | * want to allocate huge page, and if we expose page table | |
3684 | * for an instant, it will be difficult to retract from | |
3685 | * concurrent faults and from rmap lookups. | |
3686 | */ | |
82b0f8c3 | 3687 | vmf->pte = NULL; |
7267ec00 KS |
3688 | } else { |
3689 | /* See comment in pte_alloc_one_map() */ | |
d0f0931d | 3690 | if (pmd_devmap_trans_unstable(vmf->pmd)) |
7267ec00 KS |
3691 | return 0; |
3692 | /* | |
3693 | * A regular pmd is established and it can't morph into a huge | |
3694 | * pmd from under us anymore at this point because we hold the | |
3695 | * mmap_sem read mode and khugepaged takes it in write mode. | |
3696 | * So now it's safe to run pte_offset_map(). | |
3697 | */ | |
82b0f8c3 | 3698 | vmf->pte = pte_offset_map(vmf->pmd, vmf->address); |
2994302b | 3699 | vmf->orig_pte = *vmf->pte; |
7267ec00 KS |
3700 | |
3701 | /* | |
3702 | * some architectures can have larger ptes than wordsize, | |
3703 | * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and | |
3704 | * CONFIG_32BIT=y, so READ_ONCE or ACCESS_ONCE cannot guarantee | |
3705 | * atomic accesses. The code below just needs a consistent | |
3706 | * view for the ifs and we later double check anyway with the | |
3707 | * ptl lock held. So here a barrier will do. | |
3708 | */ | |
3709 | barrier(); | |
2994302b | 3710 | if (pte_none(vmf->orig_pte)) { |
82b0f8c3 JK |
3711 | pte_unmap(vmf->pte); |
3712 | vmf->pte = NULL; | |
65500d23 | 3713 | } |
1da177e4 LT |
3714 | } |
3715 | ||
82b0f8c3 JK |
3716 | if (!vmf->pte) { |
3717 | if (vma_is_anonymous(vmf->vma)) | |
3718 | return do_anonymous_page(vmf); | |
7267ec00 | 3719 | else |
82b0f8c3 | 3720 | return do_fault(vmf); |
7267ec00 KS |
3721 | } |
3722 | ||
2994302b JK |
3723 | if (!pte_present(vmf->orig_pte)) |
3724 | return do_swap_page(vmf); | |
7267ec00 | 3725 | |
2994302b JK |
3726 | if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma)) |
3727 | return do_numa_page(vmf); | |
d10e63f2 | 3728 | |
82b0f8c3 JK |
3729 | vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd); |
3730 | spin_lock(vmf->ptl); | |
2994302b | 3731 | entry = vmf->orig_pte; |
82b0f8c3 | 3732 | if (unlikely(!pte_same(*vmf->pte, entry))) |
8f4e2101 | 3733 | goto unlock; |
82b0f8c3 | 3734 | if (vmf->flags & FAULT_FLAG_WRITE) { |
1da177e4 | 3735 | if (!pte_write(entry)) |
2994302b | 3736 | return do_wp_page(vmf); |
1da177e4 LT |
3737 | entry = pte_mkdirty(entry); |
3738 | } | |
3739 | entry = pte_mkyoung(entry); | |
82b0f8c3 JK |
3740 | if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry, |
3741 | vmf->flags & FAULT_FLAG_WRITE)) { | |
3742 | update_mmu_cache(vmf->vma, vmf->address, vmf->pte); | |
1a44e149 AA |
3743 | } else { |
3744 | /* | |
3745 | * This is needed only for protection faults but the arch code | |
3746 | * is not yet telling us if this is a protection fault or not. | |
3747 | * This still avoids useless tlb flushes for .text page faults | |
3748 | * with threads. | |
3749 | */ | |
82b0f8c3 JK |
3750 | if (vmf->flags & FAULT_FLAG_WRITE) |
3751 | flush_tlb_fix_spurious_fault(vmf->vma, vmf->address); | |
1a44e149 | 3752 | } |
8f4e2101 | 3753 | unlock: |
82b0f8c3 | 3754 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
83c54070 | 3755 | return 0; |
1da177e4 LT |
3756 | } |
3757 | ||
3758 | /* | |
3759 | * By the time we get here, we already hold the mm semaphore | |
9a95f3cf PC |
3760 | * |
3761 | * The mmap_sem may have been released depending on flags and our | |
3762 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
1da177e4 | 3763 | */ |
dcddffd4 KS |
3764 | static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address, |
3765 | unsigned int flags) | |
1da177e4 | 3766 | { |
82b0f8c3 | 3767 | struct vm_fault vmf = { |
bae473a4 | 3768 | .vma = vma, |
1a29d85e | 3769 | .address = address & PAGE_MASK, |
bae473a4 | 3770 | .flags = flags, |
0721ec8b | 3771 | .pgoff = linear_page_index(vma, address), |
667240e0 | 3772 | .gfp_mask = __get_fault_gfp_mask(vma), |
bae473a4 | 3773 | }; |
dcddffd4 | 3774 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 3775 | pgd_t *pgd; |
c2febafc | 3776 | p4d_t *p4d; |
a2d58167 | 3777 | int ret; |
1da177e4 | 3778 | |
1da177e4 | 3779 | pgd = pgd_offset(mm, address); |
c2febafc KS |
3780 | p4d = p4d_alloc(mm, pgd, address); |
3781 | if (!p4d) | |
3782 | return VM_FAULT_OOM; | |
a00cc7d9 | 3783 | |
c2febafc | 3784 | vmf.pud = pud_alloc(mm, p4d, address); |
a00cc7d9 | 3785 | if (!vmf.pud) |
c74df32c | 3786 | return VM_FAULT_OOM; |
a00cc7d9 | 3787 | if (pud_none(*vmf.pud) && transparent_hugepage_enabled(vma)) { |
a00cc7d9 MW |
3788 | ret = create_huge_pud(&vmf); |
3789 | if (!(ret & VM_FAULT_FALLBACK)) | |
3790 | return ret; | |
3791 | } else { | |
3792 | pud_t orig_pud = *vmf.pud; | |
3793 | ||
3794 | barrier(); | |
3795 | if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) { | |
3796 | unsigned int dirty = flags & FAULT_FLAG_WRITE; | |
3797 | ||
a00cc7d9 MW |
3798 | /* NUMA case for anonymous PUDs would go here */ |
3799 | ||
3800 | if (dirty && !pud_write(orig_pud)) { | |
3801 | ret = wp_huge_pud(&vmf, orig_pud); | |
3802 | if (!(ret & VM_FAULT_FALLBACK)) | |
3803 | return ret; | |
3804 | } else { | |
3805 | huge_pud_set_accessed(&vmf, orig_pud); | |
3806 | return 0; | |
3807 | } | |
3808 | } | |
3809 | } | |
3810 | ||
3811 | vmf.pmd = pmd_alloc(mm, vmf.pud, address); | |
82b0f8c3 | 3812 | if (!vmf.pmd) |
c74df32c | 3813 | return VM_FAULT_OOM; |
82b0f8c3 | 3814 | if (pmd_none(*vmf.pmd) && transparent_hugepage_enabled(vma)) { |
a2d58167 | 3815 | ret = create_huge_pmd(&vmf); |
c0292554 KS |
3816 | if (!(ret & VM_FAULT_FALLBACK)) |
3817 | return ret; | |
71e3aac0 | 3818 | } else { |
82b0f8c3 | 3819 | pmd_t orig_pmd = *vmf.pmd; |
1f1d06c3 | 3820 | |
71e3aac0 | 3821 | barrier(); |
5c7fb56e | 3822 | if (pmd_trans_huge(orig_pmd) || pmd_devmap(orig_pmd)) { |
38e08854 | 3823 | if (pmd_protnone(orig_pmd) && vma_is_accessible(vma)) |
82b0f8c3 | 3824 | return do_huge_pmd_numa_page(&vmf, orig_pmd); |
d10e63f2 | 3825 | |
82b0f8c3 | 3826 | if ((vmf.flags & FAULT_FLAG_WRITE) && |
bae473a4 | 3827 | !pmd_write(orig_pmd)) { |
82b0f8c3 | 3828 | ret = wp_huge_pmd(&vmf, orig_pmd); |
9845cbbd KS |
3829 | if (!(ret & VM_FAULT_FALLBACK)) |
3830 | return ret; | |
a1dd450b | 3831 | } else { |
82b0f8c3 | 3832 | huge_pmd_set_accessed(&vmf, orig_pmd); |
9845cbbd | 3833 | return 0; |
1f1d06c3 | 3834 | } |
71e3aac0 AA |
3835 | } |
3836 | } | |
3837 | ||
82b0f8c3 | 3838 | return handle_pte_fault(&vmf); |
1da177e4 LT |
3839 | } |
3840 | ||
9a95f3cf PC |
3841 | /* |
3842 | * By the time we get here, we already hold the mm semaphore | |
3843 | * | |
3844 | * The mmap_sem may have been released depending on flags and our | |
3845 | * return value. See filemap_fault() and __lock_page_or_retry(). | |
3846 | */ | |
dcddffd4 KS |
3847 | int handle_mm_fault(struct vm_area_struct *vma, unsigned long address, |
3848 | unsigned int flags) | |
519e5247 JW |
3849 | { |
3850 | int ret; | |
3851 | ||
3852 | __set_current_state(TASK_RUNNING); | |
3853 | ||
3854 | count_vm_event(PGFAULT); | |
2262185c | 3855 | count_memcg_event_mm(vma->vm_mm, PGFAULT); |
519e5247 JW |
3856 | |
3857 | /* do counter updates before entering really critical section. */ | |
3858 | check_sync_rss_stat(current); | |
3859 | ||
3860 | /* | |
3861 | * Enable the memcg OOM handling for faults triggered in user | |
3862 | * space. Kernel faults are handled more gracefully. | |
3863 | */ | |
3864 | if (flags & FAULT_FLAG_USER) | |
49426420 | 3865 | mem_cgroup_oom_enable(); |
519e5247 | 3866 | |
bae473a4 KS |
3867 | if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE, |
3868 | flags & FAULT_FLAG_INSTRUCTION, | |
3869 | flags & FAULT_FLAG_REMOTE)) | |
3870 | return VM_FAULT_SIGSEGV; | |
3871 | ||
3872 | if (unlikely(is_vm_hugetlb_page(vma))) | |
3873 | ret = hugetlb_fault(vma->vm_mm, vma, address, flags); | |
3874 | else | |
3875 | ret = __handle_mm_fault(vma, address, flags); | |
519e5247 | 3876 | |
49426420 JW |
3877 | if (flags & FAULT_FLAG_USER) { |
3878 | mem_cgroup_oom_disable(); | |
166f61b9 TH |
3879 | /* |
3880 | * The task may have entered a memcg OOM situation but | |
3881 | * if the allocation error was handled gracefully (no | |
3882 | * VM_FAULT_OOM), there is no need to kill anything. | |
3883 | * Just clean up the OOM state peacefully. | |
3884 | */ | |
3885 | if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM)) | |
3886 | mem_cgroup_oom_synchronize(false); | |
49426420 | 3887 | } |
3812c8c8 | 3888 | |
3f70dc38 MH |
3889 | /* |
3890 | * This mm has been already reaped by the oom reaper and so the | |
3891 | * refault cannot be trusted in general. Anonymous refaults would | |
3892 | * lose data and give a zero page instead e.g. This is especially | |
3893 | * problem for use_mm() because regular tasks will just die and | |
3894 | * the corrupted data will not be visible anywhere while kthread | |
3895 | * will outlive the oom victim and potentially propagate the date | |
3896 | * further. | |
3897 | */ | |
3898 | if (unlikely((current->flags & PF_KTHREAD) && !(ret & VM_FAULT_ERROR) | |
3899 | && test_bit(MMF_UNSTABLE, &vma->vm_mm->flags))) | |
3900 | ret = VM_FAULT_SIGBUS; | |
3901 | ||
519e5247 JW |
3902 | return ret; |
3903 | } | |
e1d6d01a | 3904 | EXPORT_SYMBOL_GPL(handle_mm_fault); |
519e5247 | 3905 | |
90eceff1 KS |
3906 | #ifndef __PAGETABLE_P4D_FOLDED |
3907 | /* | |
3908 | * Allocate p4d page table. | |
3909 | * We've already handled the fast-path in-line. | |
3910 | */ | |
3911 | int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) | |
3912 | { | |
3913 | p4d_t *new = p4d_alloc_one(mm, address); | |
3914 | if (!new) | |
3915 | return -ENOMEM; | |
3916 | ||
3917 | smp_wmb(); /* See comment in __pte_alloc */ | |
3918 | ||
3919 | spin_lock(&mm->page_table_lock); | |
3920 | if (pgd_present(*pgd)) /* Another has populated it */ | |
3921 | p4d_free(mm, new); | |
3922 | else | |
3923 | pgd_populate(mm, pgd, new); | |
3924 | spin_unlock(&mm->page_table_lock); | |
3925 | return 0; | |
3926 | } | |
3927 | #endif /* __PAGETABLE_P4D_FOLDED */ | |
3928 | ||
1da177e4 LT |
3929 | #ifndef __PAGETABLE_PUD_FOLDED |
3930 | /* | |
3931 | * Allocate page upper directory. | |
872fec16 | 3932 | * We've already handled the fast-path in-line. |
1da177e4 | 3933 | */ |
c2febafc | 3934 | int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address) |
1da177e4 | 3935 | { |
c74df32c HD |
3936 | pud_t *new = pud_alloc_one(mm, address); |
3937 | if (!new) | |
1bb3630e | 3938 | return -ENOMEM; |
1da177e4 | 3939 | |
362a61ad NP |
3940 | smp_wmb(); /* See comment in __pte_alloc */ |
3941 | ||
872fec16 | 3942 | spin_lock(&mm->page_table_lock); |
c2febafc KS |
3943 | #ifndef __ARCH_HAS_5LEVEL_HACK |
3944 | if (p4d_present(*p4d)) /* Another has populated it */ | |
5e541973 | 3945 | pud_free(mm, new); |
1bb3630e | 3946 | else |
c2febafc KS |
3947 | p4d_populate(mm, p4d, new); |
3948 | #else | |
3949 | if (pgd_present(*p4d)) /* Another has populated it */ | |
5e541973 | 3950 | pud_free(mm, new); |
1bb3630e | 3951 | else |
c2febafc KS |
3952 | pgd_populate(mm, p4d, new); |
3953 | #endif /* __ARCH_HAS_5LEVEL_HACK */ | |
c74df32c | 3954 | spin_unlock(&mm->page_table_lock); |
1bb3630e | 3955 | return 0; |
1da177e4 LT |
3956 | } |
3957 | #endif /* __PAGETABLE_PUD_FOLDED */ | |
3958 | ||
3959 | #ifndef __PAGETABLE_PMD_FOLDED | |
3960 | /* | |
3961 | * Allocate page middle directory. | |
872fec16 | 3962 | * We've already handled the fast-path in-line. |
1da177e4 | 3963 | */ |
1bb3630e | 3964 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
1da177e4 | 3965 | { |
a00cc7d9 | 3966 | spinlock_t *ptl; |
c74df32c HD |
3967 | pmd_t *new = pmd_alloc_one(mm, address); |
3968 | if (!new) | |
1bb3630e | 3969 | return -ENOMEM; |
1da177e4 | 3970 | |
362a61ad NP |
3971 | smp_wmb(); /* See comment in __pte_alloc */ |
3972 | ||
a00cc7d9 | 3973 | ptl = pud_lock(mm, pud); |
1da177e4 | 3974 | #ifndef __ARCH_HAS_4LEVEL_HACK |
dc6c9a35 KS |
3975 | if (!pud_present(*pud)) { |
3976 | mm_inc_nr_pmds(mm); | |
1bb3630e | 3977 | pud_populate(mm, pud, new); |
dc6c9a35 | 3978 | } else /* Another has populated it */ |
5e541973 | 3979 | pmd_free(mm, new); |
dc6c9a35 KS |
3980 | #else |
3981 | if (!pgd_present(*pud)) { | |
3982 | mm_inc_nr_pmds(mm); | |
1bb3630e | 3983 | pgd_populate(mm, pud, new); |
dc6c9a35 KS |
3984 | } else /* Another has populated it */ |
3985 | pmd_free(mm, new); | |
1da177e4 | 3986 | #endif /* __ARCH_HAS_4LEVEL_HACK */ |
a00cc7d9 | 3987 | spin_unlock(ptl); |
1bb3630e | 3988 | return 0; |
e0f39591 | 3989 | } |
1da177e4 LT |
3990 | #endif /* __PAGETABLE_PMD_FOLDED */ |
3991 | ||
09796395 RZ |
3992 | static int __follow_pte_pmd(struct mm_struct *mm, unsigned long address, |
3993 | pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp) | |
f8ad0f49 JW |
3994 | { |
3995 | pgd_t *pgd; | |
c2febafc | 3996 | p4d_t *p4d; |
f8ad0f49 JW |
3997 | pud_t *pud; |
3998 | pmd_t *pmd; | |
3999 | pte_t *ptep; | |
4000 | ||
4001 | pgd = pgd_offset(mm, address); | |
4002 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
4003 | goto out; | |
4004 | ||
c2febafc KS |
4005 | p4d = p4d_offset(pgd, address); |
4006 | if (p4d_none(*p4d) || unlikely(p4d_bad(*p4d))) | |
4007 | goto out; | |
4008 | ||
4009 | pud = pud_offset(p4d, address); | |
f8ad0f49 JW |
4010 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) |
4011 | goto out; | |
4012 | ||
4013 | pmd = pmd_offset(pud, address); | |
f66055ab | 4014 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
f8ad0f49 | 4015 | |
09796395 RZ |
4016 | if (pmd_huge(*pmd)) { |
4017 | if (!pmdpp) | |
4018 | goto out; | |
4019 | ||
4020 | *ptlp = pmd_lock(mm, pmd); | |
4021 | if (pmd_huge(*pmd)) { | |
4022 | *pmdpp = pmd; | |
4023 | return 0; | |
4024 | } | |
4025 | spin_unlock(*ptlp); | |
4026 | } | |
4027 | ||
4028 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) | |
f8ad0f49 JW |
4029 | goto out; |
4030 | ||
4031 | ptep = pte_offset_map_lock(mm, pmd, address, ptlp); | |
f8ad0f49 JW |
4032 | if (!pte_present(*ptep)) |
4033 | goto unlock; | |
4034 | *ptepp = ptep; | |
4035 | return 0; | |
4036 | unlock: | |
4037 | pte_unmap_unlock(ptep, *ptlp); | |
4038 | out: | |
4039 | return -EINVAL; | |
4040 | } | |
4041 | ||
f729c8c9 RZ |
4042 | static inline int follow_pte(struct mm_struct *mm, unsigned long address, |
4043 | pte_t **ptepp, spinlock_t **ptlp) | |
1b36ba81 NK |
4044 | { |
4045 | int res; | |
4046 | ||
4047 | /* (void) is needed to make gcc happy */ | |
4048 | (void) __cond_lock(*ptlp, | |
09796395 RZ |
4049 | !(res = __follow_pte_pmd(mm, address, ptepp, NULL, |
4050 | ptlp))); | |
4051 | return res; | |
4052 | } | |
4053 | ||
4054 | int follow_pte_pmd(struct mm_struct *mm, unsigned long address, | |
4055 | pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp) | |
4056 | { | |
4057 | int res; | |
4058 | ||
4059 | /* (void) is needed to make gcc happy */ | |
4060 | (void) __cond_lock(*ptlp, | |
4061 | !(res = __follow_pte_pmd(mm, address, ptepp, pmdpp, | |
4062 | ptlp))); | |
1b36ba81 NK |
4063 | return res; |
4064 | } | |
09796395 | 4065 | EXPORT_SYMBOL(follow_pte_pmd); |
1b36ba81 | 4066 | |
3b6748e2 JW |
4067 | /** |
4068 | * follow_pfn - look up PFN at a user virtual address | |
4069 | * @vma: memory mapping | |
4070 | * @address: user virtual address | |
4071 | * @pfn: location to store found PFN | |
4072 | * | |
4073 | * Only IO mappings and raw PFN mappings are allowed. | |
4074 | * | |
4075 | * Returns zero and the pfn at @pfn on success, -ve otherwise. | |
4076 | */ | |
4077 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
4078 | unsigned long *pfn) | |
4079 | { | |
4080 | int ret = -EINVAL; | |
4081 | spinlock_t *ptl; | |
4082 | pte_t *ptep; | |
4083 | ||
4084 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
4085 | return ret; | |
4086 | ||
4087 | ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); | |
4088 | if (ret) | |
4089 | return ret; | |
4090 | *pfn = pte_pfn(*ptep); | |
4091 | pte_unmap_unlock(ptep, ptl); | |
4092 | return 0; | |
4093 | } | |
4094 | EXPORT_SYMBOL(follow_pfn); | |
4095 | ||
28b2ee20 | 4096 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
d87fe660 | 4097 | int follow_phys(struct vm_area_struct *vma, |
4098 | unsigned long address, unsigned int flags, | |
4099 | unsigned long *prot, resource_size_t *phys) | |
28b2ee20 | 4100 | { |
03668a4d | 4101 | int ret = -EINVAL; |
28b2ee20 RR |
4102 | pte_t *ptep, pte; |
4103 | spinlock_t *ptl; | |
28b2ee20 | 4104 | |
d87fe660 | 4105 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
4106 | goto out; | |
28b2ee20 | 4107 | |
03668a4d | 4108 | if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) |
d87fe660 | 4109 | goto out; |
28b2ee20 | 4110 | pte = *ptep; |
03668a4d | 4111 | |
28b2ee20 RR |
4112 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
4113 | goto unlock; | |
28b2ee20 RR |
4114 | |
4115 | *prot = pgprot_val(pte_pgprot(pte)); | |
03668a4d | 4116 | *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; |
28b2ee20 | 4117 | |
03668a4d | 4118 | ret = 0; |
28b2ee20 RR |
4119 | unlock: |
4120 | pte_unmap_unlock(ptep, ptl); | |
4121 | out: | |
d87fe660 | 4122 | return ret; |
28b2ee20 RR |
4123 | } |
4124 | ||
4125 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, | |
4126 | void *buf, int len, int write) | |
4127 | { | |
4128 | resource_size_t phys_addr; | |
4129 | unsigned long prot = 0; | |
2bc7273b | 4130 | void __iomem *maddr; |
28b2ee20 RR |
4131 | int offset = addr & (PAGE_SIZE-1); |
4132 | ||
d87fe660 | 4133 | if (follow_phys(vma, addr, write, &prot, &phys_addr)) |
28b2ee20 RR |
4134 | return -EINVAL; |
4135 | ||
9cb12d7b | 4136 | maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot); |
28b2ee20 RR |
4137 | if (write) |
4138 | memcpy_toio(maddr + offset, buf, len); | |
4139 | else | |
4140 | memcpy_fromio(buf, maddr + offset, len); | |
4141 | iounmap(maddr); | |
4142 | ||
4143 | return len; | |
4144 | } | |
5a73633e | 4145 | EXPORT_SYMBOL_GPL(generic_access_phys); |
28b2ee20 RR |
4146 | #endif |
4147 | ||
0ec76a11 | 4148 | /* |
206cb636 SW |
4149 | * Access another process' address space as given in mm. If non-NULL, use the |
4150 | * given task for page fault accounting. | |
0ec76a11 | 4151 | */ |
84d77d3f | 4152 | int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, |
442486ec | 4153 | unsigned long addr, void *buf, int len, unsigned int gup_flags) |
0ec76a11 | 4154 | { |
0ec76a11 | 4155 | struct vm_area_struct *vma; |
0ec76a11 | 4156 | void *old_buf = buf; |
442486ec | 4157 | int write = gup_flags & FOLL_WRITE; |
0ec76a11 | 4158 | |
0ec76a11 | 4159 | down_read(&mm->mmap_sem); |
183ff22b | 4160 | /* ignore errors, just check how much was successfully transferred */ |
0ec76a11 DH |
4161 | while (len) { |
4162 | int bytes, ret, offset; | |
4163 | void *maddr; | |
28b2ee20 | 4164 | struct page *page = NULL; |
0ec76a11 | 4165 | |
1e987790 | 4166 | ret = get_user_pages_remote(tsk, mm, addr, 1, |
5b56d49f | 4167 | gup_flags, &page, &vma, NULL); |
28b2ee20 | 4168 | if (ret <= 0) { |
dbffcd03 RR |
4169 | #ifndef CONFIG_HAVE_IOREMAP_PROT |
4170 | break; | |
4171 | #else | |
28b2ee20 RR |
4172 | /* |
4173 | * Check if this is a VM_IO | VM_PFNMAP VMA, which | |
4174 | * we can access using slightly different code. | |
4175 | */ | |
28b2ee20 | 4176 | vma = find_vma(mm, addr); |
fe936dfc | 4177 | if (!vma || vma->vm_start > addr) |
28b2ee20 RR |
4178 | break; |
4179 | if (vma->vm_ops && vma->vm_ops->access) | |
4180 | ret = vma->vm_ops->access(vma, addr, buf, | |
4181 | len, write); | |
4182 | if (ret <= 0) | |
28b2ee20 RR |
4183 | break; |
4184 | bytes = ret; | |
dbffcd03 | 4185 | #endif |
0ec76a11 | 4186 | } else { |
28b2ee20 RR |
4187 | bytes = len; |
4188 | offset = addr & (PAGE_SIZE-1); | |
4189 | if (bytes > PAGE_SIZE-offset) | |
4190 | bytes = PAGE_SIZE-offset; | |
4191 | ||
4192 | maddr = kmap(page); | |
4193 | if (write) { | |
4194 | copy_to_user_page(vma, page, addr, | |
4195 | maddr + offset, buf, bytes); | |
4196 | set_page_dirty_lock(page); | |
4197 | } else { | |
4198 | copy_from_user_page(vma, page, addr, | |
4199 | buf, maddr + offset, bytes); | |
4200 | } | |
4201 | kunmap(page); | |
09cbfeaf | 4202 | put_page(page); |
0ec76a11 | 4203 | } |
0ec76a11 DH |
4204 | len -= bytes; |
4205 | buf += bytes; | |
4206 | addr += bytes; | |
4207 | } | |
4208 | up_read(&mm->mmap_sem); | |
0ec76a11 DH |
4209 | |
4210 | return buf - old_buf; | |
4211 | } | |
03252919 | 4212 | |
5ddd36b9 | 4213 | /** |
ae91dbfc | 4214 | * access_remote_vm - access another process' address space |
5ddd36b9 SW |
4215 | * @mm: the mm_struct of the target address space |
4216 | * @addr: start address to access | |
4217 | * @buf: source or destination buffer | |
4218 | * @len: number of bytes to transfer | |
6347e8d5 | 4219 | * @gup_flags: flags modifying lookup behaviour |
5ddd36b9 SW |
4220 | * |
4221 | * The caller must hold a reference on @mm. | |
4222 | */ | |
4223 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, | |
6347e8d5 | 4224 | void *buf, int len, unsigned int gup_flags) |
5ddd36b9 | 4225 | { |
6347e8d5 | 4226 | return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags); |
5ddd36b9 SW |
4227 | } |
4228 | ||
206cb636 SW |
4229 | /* |
4230 | * Access another process' address space. | |
4231 | * Source/target buffer must be kernel space, | |
4232 | * Do not walk the page table directly, use get_user_pages | |
4233 | */ | |
4234 | int access_process_vm(struct task_struct *tsk, unsigned long addr, | |
f307ab6d | 4235 | void *buf, int len, unsigned int gup_flags) |
206cb636 SW |
4236 | { |
4237 | struct mm_struct *mm; | |
4238 | int ret; | |
4239 | ||
4240 | mm = get_task_mm(tsk); | |
4241 | if (!mm) | |
4242 | return 0; | |
4243 | ||
f307ab6d | 4244 | ret = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags); |
442486ec | 4245 | |
206cb636 SW |
4246 | mmput(mm); |
4247 | ||
4248 | return ret; | |
4249 | } | |
fcd35857 | 4250 | EXPORT_SYMBOL_GPL(access_process_vm); |
206cb636 | 4251 | |
03252919 AK |
4252 | /* |
4253 | * Print the name of a VMA. | |
4254 | */ | |
4255 | void print_vma_addr(char *prefix, unsigned long ip) | |
4256 | { | |
4257 | struct mm_struct *mm = current->mm; | |
4258 | struct vm_area_struct *vma; | |
4259 | ||
e8bff74a IM |
4260 | /* |
4261 | * Do not print if we are in atomic | |
4262 | * contexts (in exception stacks, etc.): | |
4263 | */ | |
4264 | if (preempt_count()) | |
4265 | return; | |
4266 | ||
03252919 AK |
4267 | down_read(&mm->mmap_sem); |
4268 | vma = find_vma(mm, ip); | |
4269 | if (vma && vma->vm_file) { | |
4270 | struct file *f = vma->vm_file; | |
4271 | char *buf = (char *)__get_free_page(GFP_KERNEL); | |
4272 | if (buf) { | |
2fbc57c5 | 4273 | char *p; |
03252919 | 4274 | |
9bf39ab2 | 4275 | p = file_path(f, buf, PAGE_SIZE); |
03252919 AK |
4276 | if (IS_ERR(p)) |
4277 | p = "?"; | |
2fbc57c5 | 4278 | printk("%s%s[%lx+%lx]", prefix, kbasename(p), |
03252919 AK |
4279 | vma->vm_start, |
4280 | vma->vm_end - vma->vm_start); | |
4281 | free_page((unsigned long)buf); | |
4282 | } | |
4283 | } | |
51a07e50 | 4284 | up_read(&mm->mmap_sem); |
03252919 | 4285 | } |
3ee1afa3 | 4286 | |
662bbcb2 | 4287 | #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP) |
9ec23531 | 4288 | void __might_fault(const char *file, int line) |
3ee1afa3 | 4289 | { |
95156f00 PZ |
4290 | /* |
4291 | * Some code (nfs/sunrpc) uses socket ops on kernel memory while | |
4292 | * holding the mmap_sem, this is safe because kernel memory doesn't | |
4293 | * get paged out, therefore we'll never actually fault, and the | |
4294 | * below annotations will generate false positives. | |
4295 | */ | |
db68ce10 | 4296 | if (uaccess_kernel()) |
95156f00 | 4297 | return; |
9ec23531 | 4298 | if (pagefault_disabled()) |
662bbcb2 | 4299 | return; |
9ec23531 DH |
4300 | __might_sleep(file, line, 0); |
4301 | #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) | |
662bbcb2 | 4302 | if (current->mm) |
3ee1afa3 | 4303 | might_lock_read(¤t->mm->mmap_sem); |
9ec23531 | 4304 | #endif |
3ee1afa3 | 4305 | } |
9ec23531 | 4306 | EXPORT_SYMBOL(__might_fault); |
3ee1afa3 | 4307 | #endif |
47ad8475 AA |
4308 | |
4309 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) | |
4310 | static void clear_gigantic_page(struct page *page, | |
4311 | unsigned long addr, | |
4312 | unsigned int pages_per_huge_page) | |
4313 | { | |
4314 | int i; | |
4315 | struct page *p = page; | |
4316 | ||
4317 | might_sleep(); | |
4318 | for (i = 0; i < pages_per_huge_page; | |
4319 | i++, p = mem_map_next(p, page, i)) { | |
4320 | cond_resched(); | |
4321 | clear_user_highpage(p, addr + i * PAGE_SIZE); | |
4322 | } | |
4323 | } | |
4324 | void clear_huge_page(struct page *page, | |
4325 | unsigned long addr, unsigned int pages_per_huge_page) | |
4326 | { | |
4327 | int i; | |
4328 | ||
4329 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
4330 | clear_gigantic_page(page, addr, pages_per_huge_page); | |
4331 | return; | |
4332 | } | |
4333 | ||
4334 | might_sleep(); | |
4335 | for (i = 0; i < pages_per_huge_page; i++) { | |
4336 | cond_resched(); | |
4337 | clear_user_highpage(page + i, addr + i * PAGE_SIZE); | |
4338 | } | |
4339 | } | |
4340 | ||
4341 | static void copy_user_gigantic_page(struct page *dst, struct page *src, | |
4342 | unsigned long addr, | |
4343 | struct vm_area_struct *vma, | |
4344 | unsigned int pages_per_huge_page) | |
4345 | { | |
4346 | int i; | |
4347 | struct page *dst_base = dst; | |
4348 | struct page *src_base = src; | |
4349 | ||
4350 | for (i = 0; i < pages_per_huge_page; ) { | |
4351 | cond_resched(); | |
4352 | copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma); | |
4353 | ||
4354 | i++; | |
4355 | dst = mem_map_next(dst, dst_base, i); | |
4356 | src = mem_map_next(src, src_base, i); | |
4357 | } | |
4358 | } | |
4359 | ||
4360 | void copy_user_huge_page(struct page *dst, struct page *src, | |
4361 | unsigned long addr, struct vm_area_struct *vma, | |
4362 | unsigned int pages_per_huge_page) | |
4363 | { | |
4364 | int i; | |
4365 | ||
4366 | if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) { | |
4367 | copy_user_gigantic_page(dst, src, addr, vma, | |
4368 | pages_per_huge_page); | |
4369 | return; | |
4370 | } | |
4371 | ||
4372 | might_sleep(); | |
4373 | for (i = 0; i < pages_per_huge_page; i++) { | |
4374 | cond_resched(); | |
4375 | copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma); | |
4376 | } | |
4377 | } | |
fa4d75c1 MK |
4378 | |
4379 | long copy_huge_page_from_user(struct page *dst_page, | |
4380 | const void __user *usr_src, | |
810a56b9 MK |
4381 | unsigned int pages_per_huge_page, |
4382 | bool allow_pagefault) | |
fa4d75c1 MK |
4383 | { |
4384 | void *src = (void *)usr_src; | |
4385 | void *page_kaddr; | |
4386 | unsigned long i, rc = 0; | |
4387 | unsigned long ret_val = pages_per_huge_page * PAGE_SIZE; | |
4388 | ||
4389 | for (i = 0; i < pages_per_huge_page; i++) { | |
810a56b9 MK |
4390 | if (allow_pagefault) |
4391 | page_kaddr = kmap(dst_page + i); | |
4392 | else | |
4393 | page_kaddr = kmap_atomic(dst_page + i); | |
fa4d75c1 MK |
4394 | rc = copy_from_user(page_kaddr, |
4395 | (const void __user *)(src + i * PAGE_SIZE), | |
4396 | PAGE_SIZE); | |
810a56b9 MK |
4397 | if (allow_pagefault) |
4398 | kunmap(dst_page + i); | |
4399 | else | |
4400 | kunmap_atomic(page_kaddr); | |
fa4d75c1 MK |
4401 | |
4402 | ret_val -= (PAGE_SIZE - rc); | |
4403 | if (rc) | |
4404 | break; | |
4405 | ||
4406 | cond_resched(); | |
4407 | } | |
4408 | return ret_val; | |
4409 | } | |
47ad8475 | 4410 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */ |
49076ec2 | 4411 | |
40b64acd | 4412 | #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS |
b35f1819 KS |
4413 | |
4414 | static struct kmem_cache *page_ptl_cachep; | |
4415 | ||
4416 | void __init ptlock_cache_init(void) | |
4417 | { | |
4418 | page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0, | |
4419 | SLAB_PANIC, NULL); | |
4420 | } | |
4421 | ||
539edb58 | 4422 | bool ptlock_alloc(struct page *page) |
49076ec2 KS |
4423 | { |
4424 | spinlock_t *ptl; | |
4425 | ||
b35f1819 | 4426 | ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL); |
49076ec2 KS |
4427 | if (!ptl) |
4428 | return false; | |
539edb58 | 4429 | page->ptl = ptl; |
49076ec2 KS |
4430 | return true; |
4431 | } | |
4432 | ||
539edb58 | 4433 | void ptlock_free(struct page *page) |
49076ec2 | 4434 | { |
b35f1819 | 4435 | kmem_cache_free(page_ptl_cachep, page->ptl); |
49076ec2 KS |
4436 | } |
4437 | #endif |