blob: d67fd9fcf1f2e11d8b77c513113475df125ad99d [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/memory.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 */
6
7/*
8 * demand-loading started 01.12.91 - seems it is high on the list of
9 * things wanted, and it should be easy to implement. - Linus
10 */
11
12/*
13 * Ok, demand-loading was easy, shared pages a little bit tricker. Shared
14 * pages started 02.12.91, seems to work. - Linus.
15 *
16 * Tested sharing by executing about 30 /bin/sh: under the old kernel it
17 * would have taken more than the 6M I have free, but it worked well as
18 * far as I could see.
19 *
20 * Also corrected some "invalidate()"s - I wasn't doing enough of them.
21 */
22
23/*
24 * Real VM (paging to/from disk) started 18.12.91. Much more work and
25 * thought has to go into this. Oh, well..
26 * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why.
27 * Found it. Everything seems to work now.
28 * 20.12.91 - Ok, making the swap-device changeable like the root.
29 */
30
31/*
32 * 05.04.94 - Multi-page memory management added for v1.1.
33 * Idea by Alex Bligh (alex@cconcepts.co.uk)
34 *
35 * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG
36 * (Gerhard.Wichert@pdb.siemens.de)
37 *
38 * Aug/Sep 2004 Changed to four level page tables (Andi Kleen)
39 */
40
41#include <linux/kernel_stat.h>
42#include <linux/mm.h>
43#include <linux/hugetlb.h>
44#include <linux/mman.h>
45#include <linux/swap.h>
46#include <linux/highmem.h>
47#include <linux/pagemap.h>
Hugh Dickins9a840892009-09-21 17:02:01 -070048#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include <linux/rmap.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040050#include <linux/export.h>
Shailabh Nagar0ff92242006-07-14 00:24:37 -070051#include <linux/delayacct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include <linux/init.h>
Peter Zijlstraedc79b22006-09-25 23:30:58 -070053#include <linux/writeback.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080054#include <linux/memcontrol.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070055#include <linux/mmu_notifier.h>
Hugh Dickins3dc14742009-01-06 14:40:08 -080056#include <linux/kallsyms.h>
57#include <linux/swapops.h>
58#include <linux/elf.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090059#include <linux/gfp.h>
Mel Gorman4daae3b2012-11-02 11:33:45 +000060#include <linux/migrate.h>
Andy Shevchenko2fbc57c2012-12-17 16:01:23 -080061#include <linux/string.h>
Dan Williams0abdd7a2014-01-21 15:48:12 -080062#include <linux/dma-debug.h>
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -070063#include <linux/debugfs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Alexey Dobriyan6952b612009-09-18 23:55:55 +040065#include <asm/io.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070066#include <asm/pgalloc.h>
67#include <asm/uaccess.h>
68#include <asm/tlb.h>
69#include <asm/tlbflush.h>
70#include <asm/pgtable.h>
71
Jan Beulich42b77722008-07-23 21:27:10 -070072#include "internal.h"
73
Peter Zijlstra90572892013-10-07 11:29:20 +010074#ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
75#warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid.
Peter Zijlstra75980e92013-02-22 16:34:32 -080076#endif
77
Andy Whitcroftd41dee32005-06-23 00:07:54 -070078#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -070079/* use the per-pgdat data instead for discontigmem - mbligh */
80unsigned long max_mapnr;
81struct page *mem_map;
82
83EXPORT_SYMBOL(max_mapnr);
84EXPORT_SYMBOL(mem_map);
85#endif
86
Linus Torvalds1da177e2005-04-16 15:20:36 -070087/*
88 * A number of key systems in x86 including ioremap() rely on the assumption
89 * that high_memory defines the upper bound on direct map memory, then end
90 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
91 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
92 * and ZONE_HIGHMEM.
93 */
94void * high_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -070095
Linus Torvalds1da177e2005-04-16 15:20:36 -070096EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
Ingo Molnar32a93232008-02-06 22:39:44 +010098/*
99 * Randomize the address space (stacks, mmaps, brk, etc.).
100 *
101 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
102 * as ancient (libc5 based) binaries can segfault. )
103 */
104int randomize_va_space __read_mostly =
105#ifdef CONFIG_COMPAT_BRK
106 1;
107#else
108 2;
109#endif
Andi Kleena62eaf12006-02-16 23:41:58 +0100110
111static int __init disable_randmaps(char *s)
112{
113 randomize_va_space = 0;
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800114 return 1;
Andi Kleena62eaf12006-02-16 23:41:58 +0100115}
116__setup("norandmaps", disable_randmaps);
117
Hugh Dickins62eede62009-09-21 17:03:34 -0700118unsigned long zero_pfn __read_mostly;
Hugh Dickins03f64622009-09-21 17:03:35 -0700119unsigned long highest_memmap_pfn __read_mostly;
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -0700120
121/*
122 * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init()
123 */
124static int __init init_zero_pfn(void)
125{
126 zero_pfn = page_to_pfn(ZERO_PAGE(0));
127 return 0;
128}
129core_initcall(init_zero_pfn);
Andi Kleena62eaf12006-02-16 23:41:58 +0100130
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800131
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800132#if defined(SPLIT_RSS_COUNTING)
133
David Rientjesea48cf72012-03-21 16:34:13 -0700134void sync_mm_rss(struct mm_struct *mm)
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800135{
136 int i;
137
138 for (i = 0; i < NR_MM_COUNTERS; i++) {
David Rientjes05af2e12012-03-21 16:34:13 -0700139 if (current->rss_stat.count[i]) {
140 add_mm_counter(mm, i, current->rss_stat.count[i]);
141 current->rss_stat.count[i] = 0;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800142 }
143 }
David Rientjes05af2e12012-03-21 16:34:13 -0700144 current->rss_stat.events = 0;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800145}
146
147static void add_mm_counter_fast(struct mm_struct *mm, int member, int val)
148{
149 struct task_struct *task = current;
150
151 if (likely(task->mm == mm))
152 task->rss_stat.count[member] += val;
153 else
154 add_mm_counter(mm, member, val);
155}
156#define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1)
157#define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1)
158
159/* sync counter once per 64 page faults */
160#define TASK_RSS_EVENTS_THRESH (64)
161static void check_sync_rss_stat(struct task_struct *task)
162{
163 if (unlikely(task != current))
164 return;
165 if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH))
David Rientjesea48cf72012-03-21 16:34:13 -0700166 sync_mm_rss(task->mm);
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800167}
Peter Zijlstra9547d012011-05-24 17:12:14 -0700168#else /* SPLIT_RSS_COUNTING */
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800169
170#define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member)
171#define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member)
172
173static void check_sync_rss_stat(struct task_struct *task)
174{
175}
176
Peter Zijlstra9547d012011-05-24 17:12:14 -0700177#endif /* SPLIT_RSS_COUNTING */
178
179#ifdef HAVE_GENERIC_MMU_GATHER
180
181static int tlb_next_batch(struct mmu_gather *tlb)
182{
183 struct mmu_gather_batch *batch;
184
185 batch = tlb->active;
186 if (batch->next) {
187 tlb->active = batch->next;
188 return 1;
189 }
190
Michal Hocko53a59fc2013-01-04 15:35:12 -0800191 if (tlb->batch_count == MAX_GATHER_BATCH_COUNT)
192 return 0;
193
Peter Zijlstra9547d012011-05-24 17:12:14 -0700194 batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0);
195 if (!batch)
196 return 0;
197
Michal Hocko53a59fc2013-01-04 15:35:12 -0800198 tlb->batch_count++;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700199 batch->next = NULL;
200 batch->nr = 0;
201 batch->max = MAX_GATHER_BATCH;
202
203 tlb->active->next = batch;
204 tlb->active = batch;
205
206 return 1;
207}
208
209/* tlb_gather_mmu
210 * Called to initialize an (on-stack) mmu_gather structure for page-table
211 * tear-down from @mm. The @fullmm argument is used when @mm is without
212 * users and we're going to destroy the full address space (exit/execve).
213 */
Linus Torvalds2b047252013-08-15 11:42:25 -0700214void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, unsigned long start, unsigned long end)
Peter Zijlstra9547d012011-05-24 17:12:14 -0700215{
216 tlb->mm = mm;
217
Linus Torvalds2b047252013-08-15 11:42:25 -0700218 /* Is it from 0 to ~0? */
219 tlb->fullmm = !(start | (end+1));
Dave Hansen1de14c32013-04-12 16:23:54 -0700220 tlb->need_flush_all = 0;
Linus Torvalds2b047252013-08-15 11:42:25 -0700221 tlb->start = start;
222 tlb->end = end;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700223 tlb->need_flush = 0;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700224 tlb->local.next = NULL;
225 tlb->local.nr = 0;
226 tlb->local.max = ARRAY_SIZE(tlb->__pages);
227 tlb->active = &tlb->local;
Michal Hocko53a59fc2013-01-04 15:35:12 -0800228 tlb->batch_count = 0;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700229
230#ifdef CONFIG_HAVE_RCU_TABLE_FREE
231 tlb->batch = NULL;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800232#endif
Peter Zijlstra9547d012011-05-24 17:12:14 -0700233}
234
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700235static void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb)
Peter Zijlstra9547d012011-05-24 17:12:14 -0700236{
Peter Zijlstra9547d012011-05-24 17:12:14 -0700237 tlb->need_flush = 0;
238 tlb_flush(tlb);
239#ifdef CONFIG_HAVE_RCU_TABLE_FREE
240 tlb_table_flush(tlb);
241#endif
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700242}
243
244static void tlb_flush_mmu_free(struct mmu_gather *tlb)
245{
246 struct mmu_gather_batch *batch;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700247
Peter Zijlstra9547d012011-05-24 17:12:14 -0700248 for (batch = &tlb->local; batch; batch = batch->next) {
249 free_pages_and_swap_cache(batch->pages, batch->nr);
250 batch->nr = 0;
251 }
252 tlb->active = &tlb->local;
253}
254
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700255void tlb_flush_mmu(struct mmu_gather *tlb)
256{
257 if (!tlb->need_flush)
258 return;
259 tlb_flush_mmu_tlbonly(tlb);
260 tlb_flush_mmu_free(tlb);
261}
262
Peter Zijlstra9547d012011-05-24 17:12:14 -0700263/* tlb_finish_mmu
264 * Called at the end of the shootdown operation to free up any resources
265 * that were required.
266 */
267void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end)
268{
269 struct mmu_gather_batch *batch, *next;
270
271 tlb_flush_mmu(tlb);
272
273 /* keep the page table cache within bounds */
274 check_pgt_cache();
275
276 for (batch = tlb->local.next; batch; batch = next) {
277 next = batch->next;
278 free_pages((unsigned long)batch, 0);
279 }
280 tlb->local.next = NULL;
281}
282
283/* __tlb_remove_page
284 * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while
285 * handling the additional races in SMP caused by other CPUs caching valid
286 * mappings in their TLBs. Returns the number of free page slots left.
287 * When out of page slots we must call tlb_flush_mmu().
288 */
289int __tlb_remove_page(struct mmu_gather *tlb, struct page *page)
290{
291 struct mmu_gather_batch *batch;
292
Shaohua Lif21760b2012-01-12 17:19:16 -0800293 VM_BUG_ON(!tlb->need_flush);
Peter Zijlstra9547d012011-05-24 17:12:14 -0700294
Peter Zijlstra9547d012011-05-24 17:12:14 -0700295 batch = tlb->active;
296 batch->pages[batch->nr++] = page;
297 if (batch->nr == batch->max) {
298 if (!tlb_next_batch(tlb))
299 return 0;
Shaohua Li0b43c3a2011-07-08 15:39:41 -0700300 batch = tlb->active;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700301 }
Sasha Levin309381fea2014-01-23 15:52:54 -0800302 VM_BUG_ON_PAGE(batch->nr > batch->max, page);
Peter Zijlstra9547d012011-05-24 17:12:14 -0700303
304 return batch->max - batch->nr;
305}
306
307#endif /* HAVE_GENERIC_MMU_GATHER */
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800308
Peter Zijlstra26723912011-05-24 17:12:00 -0700309#ifdef CONFIG_HAVE_RCU_TABLE_FREE
310
311/*
312 * See the comment near struct mmu_table_batch.
313 */
314
315static void tlb_remove_table_smp_sync(void *arg)
316{
317 /* Simply deliver the interrupt */
318}
319
320static void tlb_remove_table_one(void *table)
321{
322 /*
323 * This isn't an RCU grace period and hence the page-tables cannot be
324 * assumed to be actually RCU-freed.
325 *
326 * It is however sufficient for software page-table walkers that rely on
327 * IRQ disabling. See the comment near struct mmu_table_batch.
328 */
329 smp_call_function(tlb_remove_table_smp_sync, NULL, 1);
330 __tlb_remove_table(table);
331}
332
333static void tlb_remove_table_rcu(struct rcu_head *head)
334{
335 struct mmu_table_batch *batch;
336 int i;
337
338 batch = container_of(head, struct mmu_table_batch, rcu);
339
340 for (i = 0; i < batch->nr; i++)
341 __tlb_remove_table(batch->tables[i]);
342
343 free_page((unsigned long)batch);
344}
345
346void tlb_table_flush(struct mmu_gather *tlb)
347{
348 struct mmu_table_batch **batch = &tlb->batch;
349
350 if (*batch) {
351 call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu);
352 *batch = NULL;
353 }
354}
355
356void tlb_remove_table(struct mmu_gather *tlb, void *table)
357{
358 struct mmu_table_batch **batch = &tlb->batch;
359
360 tlb->need_flush = 1;
361
362 /*
363 * When there's less then two users of this mm there cannot be a
364 * concurrent page-table walk.
365 */
366 if (atomic_read(&tlb->mm->mm_users) < 2) {
367 __tlb_remove_table(table);
368 return;
369 }
370
371 if (*batch == NULL) {
372 *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN);
373 if (*batch == NULL) {
374 tlb_remove_table_one(table);
375 return;
376 }
377 (*batch)->nr = 0;
378 }
379 (*batch)->tables[(*batch)->nr++] = table;
380 if ((*batch)->nr == MAX_TABLE_BATCH)
381 tlb_table_flush(tlb);
382}
383
Peter Zijlstra9547d012011-05-24 17:12:14 -0700384#endif /* CONFIG_HAVE_RCU_TABLE_FREE */
Peter Zijlstra26723912011-05-24 17:12:00 -0700385
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 * Note: this doesn't free the actual pages themselves. That
388 * has been handled earlier when unmapping all the memory regions.
389 */
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000390static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
391 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800393 pgtable_t token = pmd_pgtable(*pmd);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700394 pmd_clear(pmd);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000395 pte_free_tlb(tlb, token, addr);
Kirill A. Shutemove1f56c82013-11-14 14:30:48 -0800396 atomic_long_dec(&tlb->mm->nr_ptes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397}
398
Hugh Dickinse0da3822005-04-19 13:29:15 -0700399static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
400 unsigned long addr, unsigned long end,
401 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402{
403 pmd_t *pmd;
404 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700405 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
Hugh Dickinse0da3822005-04-19 13:29:15 -0700407 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 do {
410 next = pmd_addr_end(addr, end);
411 if (pmd_none_or_clear_bad(pmd))
412 continue;
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000413 free_pte_range(tlb, pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 } while (pmd++, addr = next, addr != end);
415
Hugh Dickinse0da3822005-04-19 13:29:15 -0700416 start &= PUD_MASK;
417 if (start < floor)
418 return;
419 if (ceiling) {
420 ceiling &= PUD_MASK;
421 if (!ceiling)
422 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700424 if (end - 1 > ceiling - 1)
425 return;
426
427 pmd = pmd_offset(pud, start);
428 pud_clear(pud);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000429 pmd_free_tlb(tlb, pmd, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430}
431
Hugh Dickinse0da3822005-04-19 13:29:15 -0700432static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
433 unsigned long addr, unsigned long end,
434 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435{
436 pud_t *pud;
437 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700438 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439
Hugh Dickinse0da3822005-04-19 13:29:15 -0700440 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 pud = pud_offset(pgd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 do {
443 next = pud_addr_end(addr, end);
444 if (pud_none_or_clear_bad(pud))
445 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700446 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 } while (pud++, addr = next, addr != end);
448
Hugh Dickinse0da3822005-04-19 13:29:15 -0700449 start &= PGDIR_MASK;
450 if (start < floor)
451 return;
452 if (ceiling) {
453 ceiling &= PGDIR_MASK;
454 if (!ceiling)
455 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700457 if (end - 1 > ceiling - 1)
458 return;
459
460 pud = pud_offset(pgd, start);
461 pgd_clear(pgd);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000462 pud_free_tlb(tlb, pud, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463}
464
465/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700466 * This function frees user-level page tables of a process.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 */
Jan Beulich42b77722008-07-23 21:27:10 -0700468void free_pgd_range(struct mmu_gather *tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700469 unsigned long addr, unsigned long end,
470 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471{
472 pgd_t *pgd;
473 unsigned long next;
474
Hugh Dickinse0da3822005-04-19 13:29:15 -0700475 /*
476 * The next few lines have given us lots of grief...
477 *
478 * Why are we testing PMD* at this top level? Because often
479 * there will be no work to do at all, and we'd prefer not to
480 * go all the way down to the bottom just to discover that.
481 *
482 * Why all these "- 1"s? Because 0 represents both the bottom
483 * of the address space and the top of it (using -1 for the
484 * top wouldn't help much: the masks would do the wrong thing).
485 * The rule is that addr 0 and floor 0 refer to the bottom of
486 * the address space, but end 0 and ceiling 0 refer to the top
487 * Comparisons need to use "end - 1" and "ceiling - 1" (though
488 * that end 0 case should be mythical).
489 *
490 * Wherever addr is brought up or ceiling brought down, we must
491 * be careful to reject "the opposite 0" before it confuses the
492 * subsequent tests. But what about where end is brought down
493 * by PMD_SIZE below? no, end can't go down to 0 there.
494 *
495 * Whereas we round start (addr) and ceiling down, by different
496 * masks at different levels, in order to test whether a table
497 * now has no other vmas using it, so can be freed, we don't
498 * bother to round floor or end up - the tests don't need that.
499 */
500
501 addr &= PMD_MASK;
502 if (addr < floor) {
503 addr += PMD_SIZE;
504 if (!addr)
505 return;
506 }
507 if (ceiling) {
508 ceiling &= PMD_MASK;
509 if (!ceiling)
510 return;
511 }
512 if (end - 1 > ceiling - 1)
513 end -= PMD_SIZE;
514 if (addr > end - 1)
515 return;
516
Jan Beulich42b77722008-07-23 21:27:10 -0700517 pgd = pgd_offset(tlb->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 do {
519 next = pgd_addr_end(addr, end);
520 if (pgd_none_or_clear_bad(pgd))
521 continue;
Jan Beulich42b77722008-07-23 21:27:10 -0700522 free_pud_range(tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700524}
525
Jan Beulich42b77722008-07-23 21:27:10 -0700526void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700527 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700528{
529 while (vma) {
530 struct vm_area_struct *next = vma->vm_next;
531 unsigned long addr = vma->vm_start;
532
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700533 /*
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000534 * Hide vma from rmap and truncate_pagecache before freeing
535 * pgtables
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700536 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800537 unlink_anon_vmas(vma);
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700538 unlink_file_vma(vma);
539
David Gibson9da61ae2006-03-22 00:08:57 -0800540 if (is_vm_hugetlb_page(vma)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700541 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700542 floor, next? next->vm_start: ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700543 } else {
544 /*
545 * Optimization: gather nearby vmas into one call down
546 */
547 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
David Gibson48669202006-03-22 00:08:58 -0800548 && !is_vm_hugetlb_page(next)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700549 vma = next;
550 next = vma->vm_next;
Rik van Riel5beb4932010-03-05 13:42:07 -0800551 unlink_anon_vmas(vma);
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700552 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700553 }
554 free_pgd_range(tlb, addr, vma->vm_end,
555 floor, next? next->vm_start: ceiling);
556 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700557 vma = next;
558 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559}
560
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800561int __pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,
562 pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563{
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -0800564 spinlock_t *ptl;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800565 pgtable_t new = pte_alloc_one(mm, address);
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800566 int wait_split_huge_page;
Hugh Dickins1bb36302005-10-29 18:16:22 -0700567 if (!new)
568 return -ENOMEM;
569
Nick Piggin362a61a2008-05-14 06:37:36 +0200570 /*
571 * Ensure all pte setup (eg. pte page lock and page clearing) are
572 * visible before the pte is made visible to other CPUs by being
573 * put into page tables.
574 *
575 * The other side of the story is the pointer chasing in the page
576 * table walking code (when walking the page table without locking;
577 * ie. most of the time). Fortunately, these data accesses consist
578 * of a chain of data-dependent loads, meaning most CPUs (alpha
579 * being the notable exception) will already guarantee loads are
580 * seen in-order. See the alpha page table accessors for the
581 * smp_read_barrier_depends() barriers in page table walking code.
582 */
583 smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
584
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -0800585 ptl = pmd_lock(mm, pmd);
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800586 wait_split_huge_page = 0;
587 if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
Kirill A. Shutemove1f56c82013-11-14 14:30:48 -0800588 atomic_long_inc(&mm->nr_ptes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 pmd_populate(mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800590 new = NULL;
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800591 } else if (unlikely(pmd_trans_splitting(*pmd)))
592 wait_split_huge_page = 1;
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -0800593 spin_unlock(ptl);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800594 if (new)
595 pte_free(mm, new);
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800596 if (wait_split_huge_page)
597 wait_split_huge_page(vma->anon_vma, pmd);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700598 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599}
600
Hugh Dickins1bb36302005-10-29 18:16:22 -0700601int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700603 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
604 if (!new)
605 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606
Nick Piggin362a61a2008-05-14 06:37:36 +0200607 smp_wmb(); /* See comment in __pte_alloc */
608
Hugh Dickins1bb36302005-10-29 18:16:22 -0700609 spin_lock(&init_mm.page_table_lock);
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800610 if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
Hugh Dickins1bb36302005-10-29 18:16:22 -0700611 pmd_populate_kernel(&init_mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800612 new = NULL;
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800613 } else
614 VM_BUG_ON(pmd_trans_splitting(*pmd));
Hugh Dickins1bb36302005-10-29 18:16:22 -0700615 spin_unlock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800616 if (new)
617 pte_free_kernel(&init_mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700618 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619}
620
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800621static inline void init_rss_vec(int *rss)
Hugh Dickinsae859762005-10-29 18:16:05 -0700622{
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800623 memset(rss, 0, sizeof(int) * NR_MM_COUNTERS);
624}
625
626static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss)
627{
628 int i;
629
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800630 if (current->mm == mm)
David Rientjes05af2e12012-03-21 16:34:13 -0700631 sync_mm_rss(mm);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800632 for (i = 0; i < NR_MM_COUNTERS; i++)
633 if (rss[i])
634 add_mm_counter(mm, i, rss[i]);
Hugh Dickinsae859762005-10-29 18:16:05 -0700635}
636
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800638 * This function is called to print an error when a bad pte
639 * is found. For example, we might have a PFN-mapped pte in
640 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700641 *
642 * The calling function must still handle the error.
643 */
Hugh Dickins3dc14742009-01-06 14:40:08 -0800644static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
645 pte_t pte, struct page *page)
Nick Pigginb5810032005-10-29 18:16:12 -0700646{
Hugh Dickins3dc14742009-01-06 14:40:08 -0800647 pgd_t *pgd = pgd_offset(vma->vm_mm, addr);
648 pud_t *pud = pud_offset(pgd, addr);
649 pmd_t *pmd = pmd_offset(pud, addr);
650 struct address_space *mapping;
651 pgoff_t index;
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800652 static unsigned long resume;
653 static unsigned long nr_shown;
654 static unsigned long nr_unshown;
655
656 /*
657 * Allow a burst of 60 reports, then keep quiet for that minute;
658 * or allow a steady drip of one report per second.
659 */
660 if (nr_shown == 60) {
661 if (time_before(jiffies, resume)) {
662 nr_unshown++;
663 return;
664 }
665 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800666 printk(KERN_ALERT
667 "BUG: Bad page map: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800668 nr_unshown);
669 nr_unshown = 0;
670 }
671 nr_shown = 0;
672 }
673 if (nr_shown++ == 0)
674 resume = jiffies + 60 * HZ;
Hugh Dickins3dc14742009-01-06 14:40:08 -0800675
676 mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;
677 index = linear_page_index(vma, addr);
678
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800679 printk(KERN_ALERT
680 "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800681 current->comm,
682 (long long)pte_val(pte), (long long)pmd_val(*pmd));
Wu Fengguang718a3822010-03-10 15:20:43 -0800683 if (page)
Dave Hansenf0b791a2014-01-23 15:52:49 -0800684 dump_page(page, "bad pte");
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800685 printk(KERN_ALERT
Hugh Dickins3dc14742009-01-06 14:40:08 -0800686 "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
687 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
688 /*
689 * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y
690 */
691 if (vma->vm_ops)
Joe Perches071361d2012-12-12 10:19:12 -0800692 printk(KERN_ALERT "vma->vm_ops->fault: %pSR\n",
693 vma->vm_ops->fault);
Al Viro72c2d532013-09-22 16:27:52 -0400694 if (vma->vm_file)
Joe Perches071361d2012-12-12 10:19:12 -0800695 printk(KERN_ALERT "vma->vm_file->f_op->mmap: %pSR\n",
696 vma->vm_file->f_op->mmap);
Nick Pigginb5810032005-10-29 18:16:12 -0700697 dump_stack();
Rusty Russell373d4d02013-01-21 17:17:39 +1030698 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Nick Pigginb5810032005-10-29 18:16:12 -0700699}
700
701/*
Nick Piggin7e675132008-04-28 02:13:00 -0700702 * vm_normal_page -- This function gets the "struct page" associated with a pte.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800703 *
Nick Piggin7e675132008-04-28 02:13:00 -0700704 * "Special" mappings do not wish to be associated with a "struct page" (either
705 * it doesn't exist, or it exists but they don't want to touch it). In this
706 * case, NULL is returned here. "Normal" mappings do have a struct page.
Jared Hulbertb379d792008-04-28 02:12:58 -0700707 *
Nick Piggin7e675132008-04-28 02:13:00 -0700708 * There are 2 broad cases. Firstly, an architecture may define a pte_special()
709 * pte bit, in which case this function is trivial. Secondly, an architecture
710 * may not have a spare pte bit, which requires a more complicated scheme,
711 * described below.
712 *
713 * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
714 * special mapping (even if there are underlying and valid "struct pages").
715 * COWed pages of a VM_PFNMAP are always normal.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800716 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700717 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
718 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
Nick Piggin7e675132008-04-28 02:13:00 -0700719 * set, and the vm_pgoff will point to the first PFN mapped: thus every special
720 * mapping will always honor the rule
Linus Torvalds6aab3412005-11-28 14:34:23 -0800721 *
722 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
723 *
Nick Piggin7e675132008-04-28 02:13:00 -0700724 * And for normal mappings this is false.
Jared Hulbertb379d792008-04-28 02:12:58 -0700725 *
Nick Piggin7e675132008-04-28 02:13:00 -0700726 * This restricts such mappings to be a linear translation from virtual address
727 * to pfn. To get around this restriction, we allow arbitrary mappings so long
728 * as the vma is not a COW mapping; in that case, we know that all ptes are
729 * special (because none can have been COWed).
Jared Hulbertb379d792008-04-28 02:12:58 -0700730 *
731 *
Nick Piggin7e675132008-04-28 02:13:00 -0700732 * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
733 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700734 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
735 * page" backing, however the difference is that _all_ pages with a struct
736 * page (that is, those where pfn_valid is true) are refcounted and considered
737 * normal pages by the VM. The disadvantage is that pages are refcounted
738 * (which can be slower and simply not an option for some PFNMAP users). The
739 * advantage is that we don't have to follow the strict linearity rule of
740 * PFNMAP mappings in order to support COWable mappings.
741 *
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800742 */
Nick Piggin7e675132008-04-28 02:13:00 -0700743#ifdef __HAVE_ARCH_PTE_SPECIAL
744# define HAVE_PTE_SPECIAL 1
745#else
746# define HAVE_PTE_SPECIAL 0
747#endif
748struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
749 pte_t pte)
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800750{
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800751 unsigned long pfn = pte_pfn(pte);
Nick Piggin7e675132008-04-28 02:13:00 -0700752
753 if (HAVE_PTE_SPECIAL) {
Mel Gormanc46a7c82014-06-04 16:06:30 -0700754 if (likely(!pte_special(pte) || pte_numa(pte)))
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800755 goto check_pfn;
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -0700756 if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))
757 return NULL;
Hugh Dickins62eede62009-09-21 17:03:34 -0700758 if (!is_zero_pfn(pfn))
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800759 print_bad_pte(vma, addr, pte, NULL);
Nick Piggin7e675132008-04-28 02:13:00 -0700760 return NULL;
761 }
762
763 /* !HAVE_PTE_SPECIAL case follows: */
764
Jared Hulbertb379d792008-04-28 02:12:58 -0700765 if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
766 if (vma->vm_flags & VM_MIXEDMAP) {
767 if (!pfn_valid(pfn))
768 return NULL;
769 goto out;
770 } else {
Nick Piggin7e675132008-04-28 02:13:00 -0700771 unsigned long off;
772 off = (addr - vma->vm_start) >> PAGE_SHIFT;
Jared Hulbertb379d792008-04-28 02:12:58 -0700773 if (pfn == vma->vm_pgoff + off)
774 return NULL;
775 if (!is_cow_mapping(vma->vm_flags))
776 return NULL;
777 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800778 }
779
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800780check_pfn:
781 if (unlikely(pfn > highest_memmap_pfn)) {
782 print_bad_pte(vma, addr, pte, NULL);
783 return NULL;
784 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800785
Mel Gormanc46a7c82014-06-04 16:06:30 -0700786 if (is_zero_pfn(pfn))
787 return NULL;
788
Linus Torvalds6aab3412005-11-28 14:34:23 -0800789 /*
Nick Piggin7e675132008-04-28 02:13:00 -0700790 * NOTE! We still have PageReserved() pages in the page tables.
Nick Piggin7e675132008-04-28 02:13:00 -0700791 * eg. VDSO mappings can cause them to exist.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800792 */
Jared Hulbertb379d792008-04-28 02:12:58 -0700793out:
Linus Torvalds6aab3412005-11-28 14:34:23 -0800794 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800795}
796
797/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 * copy one vm_area from one task to the other. Assumes the page tables
799 * already present in the new task to be cleared in the whole range
800 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 */
802
Hugh Dickins570a335b2009-12-14 17:58:46 -0800803static inline unsigned long
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700805 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700806 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807{
Nick Pigginb5810032005-10-29 18:16:12 -0700808 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 pte_t pte = *src_pte;
810 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811
812 /* pte contains position in swap or file, so copy. */
813 if (unlikely(!pte_present(pte))) {
814 if (!pte_file(pte)) {
Christoph Lameter06972122006-06-23 02:03:35 -0700815 swp_entry_t entry = pte_to_swp_entry(pte);
816
Hugh Dickins570a335b2009-12-14 17:58:46 -0800817 if (swap_duplicate(entry) < 0)
818 return entry.val;
819
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 /* make sure dst_mm is on swapoff's mmlist. */
821 if (unlikely(list_empty(&dst_mm->mmlist))) {
822 spin_lock(&mmlist_lock);
Hugh Dickinsf412ac02005-10-29 18:16:41 -0700823 if (list_empty(&dst_mm->mmlist))
824 list_add(&dst_mm->mmlist,
825 &src_mm->mmlist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 spin_unlock(&mmlist_lock);
827 }
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -0800828 if (likely(!non_swap_entry(entry)))
829 rss[MM_SWAPENTS]++;
Konstantin Khlebnikov9f9f1ac2012-01-20 14:34:24 -0800830 else if (is_migration_entry(entry)) {
831 page = migration_entry_to_page(entry);
832
833 if (PageAnon(page))
834 rss[MM_ANONPAGES]++;
835 else
836 rss[MM_FILEPAGES]++;
837
838 if (is_write_migration_entry(entry) &&
839 is_cow_mapping(vm_flags)) {
840 /*
841 * COW mappings require pages in both
842 * parent and child to be set to read.
843 */
844 make_migration_entry_read(&entry);
845 pte = swp_entry_to_pte(entry);
Cyrill Gorcunovc3d16e12013-10-16 13:46:51 -0700846 if (pte_swp_soft_dirty(*src_pte))
847 pte = pte_swp_mksoft_dirty(pte);
Konstantin Khlebnikov9f9f1ac2012-01-20 14:34:24 -0800848 set_pte_at(src_mm, addr, src_pte, pte);
849 }
Christoph Lameter06972122006-06-23 02:03:35 -0700850 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700852 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 }
854
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 /*
856 * If it's a COW mapping, write protect it both
857 * in the parent and the child
858 */
Linus Torvalds67121172005-12-11 20:38:17 -0800859 if (is_cow_mapping(vm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 ptep_set_wrprotect(src_mm, addr, src_pte);
Zachary Amsden3dc90792006-09-30 23:29:30 -0700861 pte = pte_wrprotect(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 }
863
864 /*
865 * If it's a shared mapping, mark it clean in
866 * the child
867 */
868 if (vm_flags & VM_SHARED)
869 pte = pte_mkclean(pte);
870 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800871
872 page = vm_normal_page(vma, addr, pte);
873 if (page) {
874 get_page(page);
Hugh Dickins21333b22009-09-21 17:01:59 -0700875 page_dup_rmap(page);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800876 if (PageAnon(page))
877 rss[MM_ANONPAGES]++;
878 else
879 rss[MM_FILEPAGES]++;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800880 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700881
882out_set_pte:
883 set_pte_at(dst_mm, addr, dst_pte, pte);
Hugh Dickins570a335b2009-12-14 17:58:46 -0800884 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885}
886
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800887int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
888 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
889 unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890{
Daisuke Nishimurac36987e2009-10-26 16:50:23 -0700891 pte_t *orig_src_pte, *orig_dst_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700893 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -0700894 int progress = 0;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800895 int rss[NR_MM_COUNTERS];
Hugh Dickins570a335b2009-12-14 17:58:46 -0800896 swp_entry_t entry = (swp_entry_t){0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897
898again:
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800899 init_rss_vec(rss);
900
Hugh Dickinsc74df322005-10-29 18:16:23 -0700901 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 if (!dst_pte)
903 return -ENOMEM;
Peter Zijlstraece0e2b2010-10-26 14:21:52 -0700904 src_pte = pte_offset_map(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700905 src_ptl = pte_lockptr(src_mm, src_pmd);
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700906 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
Daisuke Nishimurac36987e2009-10-26 16:50:23 -0700907 orig_src_pte = src_pte;
908 orig_dst_pte = dst_pte;
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700909 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 do {
912 /*
913 * We are holding two locks at this point - either of them
914 * could generate latencies in another task on another CPU.
915 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700916 if (progress >= 32) {
917 progress = 0;
918 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +0100919 spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700920 break;
921 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 if (pte_none(*src_pte)) {
923 progress++;
924 continue;
925 }
Hugh Dickins570a335b2009-12-14 17:58:46 -0800926 entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte,
927 vma, addr, rss);
928 if (entry.val)
929 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 progress += 8;
931 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700933 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -0700934 spin_unlock(src_ptl);
Peter Zijlstraece0e2b2010-10-26 14:21:52 -0700935 pte_unmap(orig_src_pte);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800936 add_mm_rss_vec(dst_mm, rss);
Daisuke Nishimurac36987e2009-10-26 16:50:23 -0700937 pte_unmap_unlock(orig_dst_pte, dst_ptl);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700938 cond_resched();
Hugh Dickins570a335b2009-12-14 17:58:46 -0800939
940 if (entry.val) {
941 if (add_swap_count_continuation(entry, GFP_KERNEL) < 0)
942 return -ENOMEM;
943 progress = 0;
944 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 if (addr != end)
946 goto again;
947 return 0;
948}
949
950static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
951 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
952 unsigned long addr, unsigned long end)
953{
954 pmd_t *src_pmd, *dst_pmd;
955 unsigned long next;
956
957 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
958 if (!dst_pmd)
959 return -ENOMEM;
960 src_pmd = pmd_offset(src_pud, addr);
961 do {
962 next = pmd_addr_end(addr, end);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800963 if (pmd_trans_huge(*src_pmd)) {
964 int err;
Andrea Arcangeli14d1a552011-01-13 15:47:15 -0800965 VM_BUG_ON(next-addr != HPAGE_PMD_SIZE);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800966 err = copy_huge_pmd(dst_mm, src_mm,
967 dst_pmd, src_pmd, addr, vma);
968 if (err == -ENOMEM)
969 return -ENOMEM;
970 if (!err)
971 continue;
972 /* fall through */
973 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 if (pmd_none_or_clear_bad(src_pmd))
975 continue;
976 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
977 vma, addr, next))
978 return -ENOMEM;
979 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
980 return 0;
981}
982
983static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
984 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
985 unsigned long addr, unsigned long end)
986{
987 pud_t *src_pud, *dst_pud;
988 unsigned long next;
989
990 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
991 if (!dst_pud)
992 return -ENOMEM;
993 src_pud = pud_offset(src_pgd, addr);
994 do {
995 next = pud_addr_end(addr, end);
996 if (pud_none_or_clear_bad(src_pud))
997 continue;
998 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
999 vma, addr, next))
1000 return -ENOMEM;
1001 } while (dst_pud++, src_pud++, addr = next, addr != end);
1002 return 0;
1003}
1004
1005int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
1006 struct vm_area_struct *vma)
1007{
1008 pgd_t *src_pgd, *dst_pgd;
1009 unsigned long next;
1010 unsigned long addr = vma->vm_start;
1011 unsigned long end = vma->vm_end;
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001012 unsigned long mmun_start; /* For mmu_notifiers */
1013 unsigned long mmun_end; /* For mmu_notifiers */
1014 bool is_cow;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001015 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016
Nick Piggind9928952005-08-28 16:49:11 +10001017 /*
1018 * Don't copy ptes where a page fault will fill them correctly.
1019 * Fork becomes much lighter when there are big shared or private
1020 * readonly mappings. The tradeoff is that copy_page_range is more
1021 * efficient than faulting.
1022 */
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001023 if (!(vma->vm_flags & (VM_HUGETLB | VM_NONLINEAR |
1024 VM_PFNMAP | VM_MIXEDMAP))) {
Nick Piggind9928952005-08-28 16:49:11 +10001025 if (!vma->anon_vma)
1026 return 0;
1027 }
1028
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 if (is_vm_hugetlb_page(vma))
1030 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
1031
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07001032 if (unlikely(vma->vm_flags & VM_PFNMAP)) {
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001033 /*
1034 * We do not free on error cases below as remove_vma
1035 * gets called on error from higher level routine
1036 */
Suresh Siddha5180da42012-10-08 16:28:29 -07001037 ret = track_pfn_copy(vma);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001038 if (ret)
1039 return ret;
1040 }
1041
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001042 /*
1043 * We need to invalidate the secondary MMU mappings only when
1044 * there could be a permission downgrade on the ptes of the
1045 * parent mm. And a permission downgrade will only happen if
1046 * is_cow_mapping() returns true.
1047 */
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001048 is_cow = is_cow_mapping(vma->vm_flags);
1049 mmun_start = addr;
1050 mmun_end = end;
1051 if (is_cow)
1052 mmu_notifier_invalidate_range_start(src_mm, mmun_start,
1053 mmun_end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001054
1055 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 dst_pgd = pgd_offset(dst_mm, addr);
1057 src_pgd = pgd_offset(src_mm, addr);
1058 do {
1059 next = pgd_addr_end(addr, end);
1060 if (pgd_none_or_clear_bad(src_pgd))
1061 continue;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001062 if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
1063 vma, addr, next))) {
1064 ret = -ENOMEM;
1065 break;
1066 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001068
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001069 if (is_cow)
1070 mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001071 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072}
1073
Robin Holt51c6f662005-11-13 16:06:42 -08001074static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -07001075 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 unsigned long addr, unsigned long end,
Peter Zijlstra97a89412011-05-24 17:12:04 -07001077 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078{
Nick Pigginb5810032005-10-29 18:16:12 -07001079 struct mm_struct *mm = tlb->mm;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001080 int force_flush = 0;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001081 int rss[NR_MM_COUNTERS];
Peter Zijlstra97a89412011-05-24 17:12:04 -07001082 spinlock_t *ptl;
Steven Rostedt5f1a1902011-06-15 15:08:23 -07001083 pte_t *start_pte;
Peter Zijlstra97a89412011-05-24 17:12:04 -07001084 pte_t *pte;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001085
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001086again:
Peter Zijlstrae3032972011-05-24 17:12:01 -07001087 init_rss_vec(rss);
Steven Rostedt5f1a1902011-06-15 15:08:23 -07001088 start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
1089 pte = start_pte;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001090 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 do {
1092 pte_t ptent = *pte;
Robin Holt51c6f662005-11-13 16:06:42 -08001093 if (pte_none(ptent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 continue;
Robin Holt51c6f662005-11-13 16:06:42 -08001095 }
Hugh Dickins6f5e6b92006-03-16 23:04:09 -08001096
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -08001098 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -08001099
Linus Torvalds6aab3412005-11-28 14:34:23 -08001100 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 if (unlikely(details) && page) {
1102 /*
1103 * unmap_shared_mapping_pages() wants to
1104 * invalidate cache without truncating:
1105 * unmap shared but keep private pages.
1106 */
1107 if (details->check_mapping &&
1108 details->check_mapping != page->mapping)
1109 continue;
1110 /*
1111 * Each page->index must be checked when
1112 * invalidating or truncating nonlinear.
1113 */
1114 if (details->nonlinear_vma &&
1115 (page->index < details->first_index ||
1116 page->index > details->last_index))
1117 continue;
1118 }
Nick Pigginb5810032005-10-29 18:16:12 -07001119 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -07001120 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 tlb_remove_tlb_entry(tlb, pte, addr);
1122 if (unlikely(!page))
1123 continue;
1124 if (unlikely(details) && details->nonlinear_vma
1125 && linear_page_index(details->nonlinear_vma,
Cyrill Gorcunov41bb3472013-08-13 16:00:51 -07001126 addr) != page->index) {
1127 pte_t ptfile = pgoff_to_pte(page->index);
1128 if (pte_soft_dirty(ptent))
1129 pte_file_mksoft_dirty(ptfile);
1130 set_pte_at(mm, addr, pte, ptfile);
1131 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 if (PageAnon(page))
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001133 rss[MM_ANONPAGES]--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -07001134 else {
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001135 if (pte_dirty(ptent)) {
1136 force_flush = 1;
Hugh Dickins6237bcd2005-10-29 18:15:54 -07001137 set_page_dirty(page);
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001138 }
Johannes Weiner4917e5d2009-01-06 14:39:17 -08001139 if (pte_young(ptent) &&
Joe Perches64363aa2013-07-08 16:00:18 -07001140 likely(!(vma->vm_flags & VM_SEQ_READ)))
Nick Pigginbf3f3bc2009-01-06 14:38:55 -08001141 mark_page_accessed(page);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001142 rss[MM_FILEPAGES]--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -07001143 }
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001144 page_remove_rmap(page);
Hugh Dickins3dc14742009-01-06 14:40:08 -08001145 if (unlikely(page_mapcount(page) < 0))
1146 print_bad_pte(vma, addr, ptent, page);
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001147 if (unlikely(!__tlb_remove_page(tlb, page))) {
1148 force_flush = 1;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001149 break;
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001150 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 continue;
1152 }
1153 /*
1154 * If details->check_mapping, we leave swap entries;
1155 * if details->nonlinear_vma, we leave file entries.
1156 */
1157 if (unlikely(details))
1158 continue;
Hugh Dickins2509ef22009-01-06 14:40:10 -08001159 if (pte_file(ptent)) {
1160 if (unlikely(!(vma->vm_flags & VM_NONLINEAR)))
1161 print_bad_pte(vma, addr, ptent, NULL);
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001162 } else {
1163 swp_entry_t entry = pte_to_swp_entry(ptent);
1164
1165 if (!non_swap_entry(entry))
1166 rss[MM_SWAPENTS]--;
Konstantin Khlebnikov9f9f1ac2012-01-20 14:34:24 -08001167 else if (is_migration_entry(entry)) {
1168 struct page *page;
1169
1170 page = migration_entry_to_page(entry);
1171
1172 if (PageAnon(page))
1173 rss[MM_ANONPAGES]--;
1174 else
1175 rss[MM_FILEPAGES]--;
1176 }
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001177 if (unlikely(!free_swap_and_cache(entry)))
1178 print_bad_pte(vma, addr, ptent, NULL);
1179 }
Zachary Amsden9888a1c2006-09-30 23:29:31 -07001180 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
Peter Zijlstra97a89412011-05-24 17:12:04 -07001181 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickinsae859762005-10-29 18:16:05 -07001182
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001183 add_mm_rss_vec(mm, rss);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001184 arch_leave_lazy_mmu_mode();
Robin Holt51c6f662005-11-13 16:06:42 -08001185
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001186 /* Do the actual TLB flush before dropping ptl */
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001187 if (force_flush) {
Linus Torvalds2b047252013-08-15 11:42:25 -07001188 unsigned long old_end;
1189
Linus Torvalds2b047252013-08-15 11:42:25 -07001190 /*
1191 * Flush the TLB just for the previous segment,
1192 * then update the range to be the remaining
1193 * TLB range.
1194 */
1195 old_end = tlb->end;
Vineet Guptae6c495a2013-07-03 15:03:31 -07001196 tlb->end = addr;
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001197 tlb_flush_mmu_tlbonly(tlb);
Linus Torvalds2b047252013-08-15 11:42:25 -07001198 tlb->start = addr;
1199 tlb->end = old_end;
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001200 }
1201 pte_unmap_unlock(start_pte, ptl);
1202
1203 /*
1204 * If we forced a TLB flush (either due to running out of
1205 * batch buffers or because we needed to flush dirty TLB
1206 * entries before releasing the ptl), free the batched
1207 * memory too. Restart if we didn't do everything.
1208 */
1209 if (force_flush) {
1210 force_flush = 0;
1211 tlb_flush_mmu_free(tlb);
Linus Torvalds2b047252013-08-15 11:42:25 -07001212
1213 if (addr != end)
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001214 goto again;
1215 }
1216
Robin Holt51c6f662005-11-13 16:06:42 -08001217 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218}
1219
Robin Holt51c6f662005-11-13 16:06:42 -08001220static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -07001221 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 unsigned long addr, unsigned long end,
Peter Zijlstra97a89412011-05-24 17:12:04 -07001223 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224{
1225 pmd_t *pmd;
1226 unsigned long next;
1227
1228 pmd = pmd_offset(pud, addr);
1229 do {
1230 next = pmd_addr_end(addr, end);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001231 if (pmd_trans_huge(*pmd)) {
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001232 if (next - addr != HPAGE_PMD_SIZE) {
David Rientjese0897d72012-06-20 12:53:00 -07001233#ifdef CONFIG_DEBUG_VM
1234 if (!rwsem_is_locked(&tlb->mm->mmap_sem)) {
1235 pr_err("%s: mmap_sem is unlocked! addr=0x%lx end=0x%lx vma->vm_start=0x%lx vma->vm_end=0x%lx\n",
1236 __func__, addr, end,
1237 vma->vm_start,
1238 vma->vm_end);
1239 BUG();
1240 }
1241#endif
Kirill A. Shutemove1803772012-12-12 13:50:59 -08001242 split_huge_page_pmd(vma, addr, pmd);
Shaohua Lif21760b2012-01-12 17:19:16 -08001243 } else if (zap_huge_pmd(tlb, vma, pmd, addr))
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001244 goto next;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001245 /* fall through */
1246 }
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001247 /*
1248 * Here there can be other concurrent MADV_DONTNEED or
1249 * trans huge page faults running, and if the pmd is
1250 * none or trans huge it can change under us. This is
1251 * because MADV_DONTNEED holds the mmap_sem in read
1252 * mode.
1253 */
1254 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1255 goto next;
Peter Zijlstra97a89412011-05-24 17:12:04 -07001256 next = zap_pte_range(tlb, vma, pmd, addr, next, details);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001257next:
Peter Zijlstra97a89412011-05-24 17:12:04 -07001258 cond_resched();
1259 } while (pmd++, addr = next, addr != end);
Robin Holt51c6f662005-11-13 16:06:42 -08001260
1261 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262}
1263
Robin Holt51c6f662005-11-13 16:06:42 -08001264static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -07001265 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 unsigned long addr, unsigned long end,
Peter Zijlstra97a89412011-05-24 17:12:04 -07001267 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268{
1269 pud_t *pud;
1270 unsigned long next;
1271
1272 pud = pud_offset(pgd, addr);
1273 do {
1274 next = pud_addr_end(addr, end);
Peter Zijlstra97a89412011-05-24 17:12:04 -07001275 if (pud_none_or_clear_bad(pud))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 continue;
Peter Zijlstra97a89412011-05-24 17:12:04 -07001277 next = zap_pmd_range(tlb, vma, pud, addr, next, details);
1278 } while (pud++, addr = next, addr != end);
Robin Holt51c6f662005-11-13 16:06:42 -08001279
1280 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281}
1282
Al Viro038c7aa2012-03-05 13:25:09 -05001283static void unmap_page_range(struct mmu_gather *tlb,
1284 struct vm_area_struct *vma,
1285 unsigned long addr, unsigned long end,
1286 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287{
1288 pgd_t *pgd;
1289 unsigned long next;
1290
1291 if (details && !details->check_mapping && !details->nonlinear_vma)
1292 details = NULL;
1293
1294 BUG_ON(addr >= end);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08001295 mem_cgroup_uncharge_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 tlb_start_vma(tlb, vma);
1297 pgd = pgd_offset(vma->vm_mm, addr);
1298 do {
1299 next = pgd_addr_end(addr, end);
Peter Zijlstra97a89412011-05-24 17:12:04 -07001300 if (pgd_none_or_clear_bad(pgd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 continue;
Peter Zijlstra97a89412011-05-24 17:12:04 -07001302 next = zap_pud_range(tlb, vma, pgd, addr, next, details);
1303 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 tlb_end_vma(tlb, vma);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08001305 mem_cgroup_uncharge_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307
Al Virof5cc4ee2012-03-05 14:14:20 -05001308
1309static void unmap_single_vma(struct mmu_gather *tlb,
1310 struct vm_area_struct *vma, unsigned long start_addr,
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001311 unsigned long end_addr,
Al Virof5cc4ee2012-03-05 14:14:20 -05001312 struct zap_details *details)
1313{
1314 unsigned long start = max(vma->vm_start, start_addr);
1315 unsigned long end;
1316
1317 if (start >= vma->vm_end)
1318 return;
1319 end = min(vma->vm_end, end_addr);
1320 if (end <= vma->vm_start)
1321 return;
1322
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +05301323 if (vma->vm_file)
1324 uprobe_munmap(vma, start, end);
1325
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07001326 if (unlikely(vma->vm_flags & VM_PFNMAP))
Suresh Siddha5180da42012-10-08 16:28:29 -07001327 untrack_pfn(vma, 0, 0);
Al Virof5cc4ee2012-03-05 14:14:20 -05001328
1329 if (start != end) {
1330 if (unlikely(is_vm_hugetlb_page(vma))) {
1331 /*
1332 * It is undesirable to test vma->vm_file as it
1333 * should be non-null for valid hugetlb area.
1334 * However, vm_file will be NULL in the error
Davidlohr Bueso7aa6b4a2014-04-07 15:37:01 -07001335 * cleanup path of mmap_region. When
Al Virof5cc4ee2012-03-05 14:14:20 -05001336 * hugetlbfs ->mmap method fails,
Davidlohr Bueso7aa6b4a2014-04-07 15:37:01 -07001337 * mmap_region() nullifies vma->vm_file
Al Virof5cc4ee2012-03-05 14:14:20 -05001338 * before calling this function to clean up.
1339 * Since no pte has actually been setup, it is
1340 * safe to do nothing in this case.
1341 */
Aneesh Kumar K.V24669e52012-07-31 16:42:03 -07001342 if (vma->vm_file) {
1343 mutex_lock(&vma->vm_file->f_mapping->i_mmap_mutex);
Mel Gormand8333522012-07-31 16:46:20 -07001344 __unmap_hugepage_range_final(tlb, vma, start, end, NULL);
Aneesh Kumar K.V24669e52012-07-31 16:42:03 -07001345 mutex_unlock(&vma->vm_file->f_mapping->i_mmap_mutex);
1346 }
Al Virof5cc4ee2012-03-05 14:14:20 -05001347 } else
1348 unmap_page_range(tlb, vma, start, end, details);
1349 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350}
1351
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352/**
1353 * unmap_vmas - unmap a range of memory covered by a list of vma's
Randy Dunlap0164f692011-06-15 15:08:09 -07001354 * @tlb: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 * @vma: the starting vma
1356 * @start_addr: virtual address at which to start unmapping
1357 * @end_addr: virtual address at which to end unmapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 *
Hugh Dickins508034a2005-10-29 18:16:30 -07001359 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 * Only addresses between `start' and `end' will be unmapped.
1362 *
1363 * The VMA list must be sorted in ascending virtual address order.
1364 *
1365 * unmap_vmas() assumes that the caller will flush the whole unmapped address
1366 * range after unmap_vmas() returns. So the only responsibility here is to
1367 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
1368 * drops the lock and schedules.
1369 */
Al Viro6e8bb012012-03-05 13:41:15 -05001370void unmap_vmas(struct mmu_gather *tlb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 struct vm_area_struct *vma, unsigned long start_addr,
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001372 unsigned long end_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373{
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001374 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001376 mmu_notifier_invalidate_range_start(mm, start_addr, end_addr);
Al Virof5cc4ee2012-03-05 14:14:20 -05001377 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next)
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001378 unmap_single_vma(tlb, vma, start_addr, end_addr, NULL);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001379 mmu_notifier_invalidate_range_end(mm, start_addr, end_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380}
1381
1382/**
1383 * zap_page_range - remove user pages in a given range
1384 * @vma: vm_area_struct holding the applicable pages
Randy Dunlapeb4546b2012-06-20 12:53:02 -07001385 * @start: starting address of pages to zap
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 * @size: number of bytes to zap
1387 * @details: details of nonlinear truncation or shared cache invalidation
Al Virof5cc4ee2012-03-05 14:14:20 -05001388 *
1389 * Caller must protect the VMA list
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 */
Linus Torvalds7e027b12012-05-06 13:43:15 -07001391void zap_page_range(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 unsigned long size, struct zap_details *details)
1393{
1394 struct mm_struct *mm = vma->vm_mm;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001395 struct mmu_gather tlb;
Linus Torvalds7e027b12012-05-06 13:43:15 -07001396 unsigned long end = start + size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07001399 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001400 update_hiwater_rss(mm);
Linus Torvalds7e027b12012-05-06 13:43:15 -07001401 mmu_notifier_invalidate_range_start(mm, start, end);
1402 for ( ; vma && vma->vm_start < end; vma = vma->vm_next)
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001403 unmap_single_vma(&tlb, vma, start, end, details);
Linus Torvalds7e027b12012-05-06 13:43:15 -07001404 mmu_notifier_invalidate_range_end(mm, start, end);
1405 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406}
1407
Jack Steinerc627f9c2008-07-29 22:33:53 -07001408/**
Al Virof5cc4ee2012-03-05 14:14:20 -05001409 * zap_page_range_single - remove user pages in a given range
1410 * @vma: vm_area_struct holding the applicable pages
1411 * @address: starting address of pages to zap
1412 * @size: number of bytes to zap
1413 * @details: details of nonlinear truncation or shared cache invalidation
1414 *
1415 * The range must fit into one VMA.
1416 */
1417static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address,
1418 unsigned long size, struct zap_details *details)
1419{
1420 struct mm_struct *mm = vma->vm_mm;
1421 struct mmu_gather tlb;
1422 unsigned long end = address + size;
Al Virof5cc4ee2012-03-05 14:14:20 -05001423
1424 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07001425 tlb_gather_mmu(&tlb, mm, address, end);
Al Virof5cc4ee2012-03-05 14:14:20 -05001426 update_hiwater_rss(mm);
1427 mmu_notifier_invalidate_range_start(mm, address, end);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001428 unmap_single_vma(&tlb, vma, address, end, details);
Al Virof5cc4ee2012-03-05 14:14:20 -05001429 mmu_notifier_invalidate_range_end(mm, address, end);
1430 tlb_finish_mmu(&tlb, address, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431}
1432
Jack Steinerc627f9c2008-07-29 22:33:53 -07001433/**
1434 * zap_vma_ptes - remove ptes mapping the vma
1435 * @vma: vm_area_struct holding ptes to be zapped
1436 * @address: starting address of pages to zap
1437 * @size: number of bytes to zap
1438 *
1439 * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1440 *
1441 * The entire address range must be fully contained within the vma.
1442 *
1443 * Returns 0 if successful.
1444 */
1445int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
1446 unsigned long size)
1447{
1448 if (address < vma->vm_start || address + size > vma->vm_end ||
1449 !(vma->vm_flags & VM_PFNMAP))
1450 return -1;
Al Virof5cc4ee2012-03-05 14:14:20 -05001451 zap_page_range_single(vma, address, size, NULL);
Jack Steinerc627f9c2008-07-29 22:33:53 -07001452 return 0;
1453}
1454EXPORT_SYMBOL_GPL(zap_vma_ptes);
1455
Namhyung Kim25ca1d62010-10-26 14:21:59 -07001456pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr,
Harvey Harrison920c7a52008-02-04 22:29:26 -08001457 spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001458{
1459 pgd_t * pgd = pgd_offset(mm, addr);
1460 pud_t * pud = pud_alloc(mm, pgd, addr);
1461 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001462 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Andrea Arcangelif66055ab2011-01-13 15:46:54 -08001463 if (pmd) {
1464 VM_BUG_ON(pmd_trans_huge(*pmd));
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001465 return pte_alloc_map_lock(mm, pmd, addr, ptl);
Andrea Arcangelif66055ab2011-01-13 15:46:54 -08001466 }
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001467 }
1468 return NULL;
1469}
1470
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001472 * This is the old fallback for page remapping.
1473 *
1474 * For historical reasons, it only allows reserved pages. Only
1475 * old drivers should use this, and they needed to mark their
1476 * pages reserved for the old functions anyway.
1477 */
Nick Piggin423bad62008-04-28 02:13:01 -07001478static int insert_page(struct vm_area_struct *vma, unsigned long addr,
1479 struct page *page, pgprot_t prot)
Linus Torvalds238f58d2005-11-29 13:01:56 -08001480{
Nick Piggin423bad62008-04-28 02:13:01 -07001481 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001482 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001483 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001484 spinlock_t *ptl;
1485
Linus Torvalds238f58d2005-11-29 13:01:56 -08001486 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001487 if (PageAnon(page))
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001488 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001489 retval = -ENOMEM;
1490 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001491 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001492 if (!pte)
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001493 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001494 retval = -EBUSY;
1495 if (!pte_none(*pte))
1496 goto out_unlock;
1497
1498 /* Ok, finally just insert the thing.. */
1499 get_page(page);
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08001500 inc_mm_counter_fast(mm, MM_FILEPAGES);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001501 page_add_file_rmap(page);
1502 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1503
1504 retval = 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001505 pte_unmap_unlock(pte, ptl);
1506 return retval;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001507out_unlock:
1508 pte_unmap_unlock(pte, ptl);
1509out:
1510 return retval;
1511}
1512
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001513/**
1514 * vm_insert_page - insert single page into user vma
1515 * @vma: user vma to map to
1516 * @addr: target user address of this page
1517 * @page: source kernel page
1518 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001519 * This allows drivers to insert individual pages they've allocated
1520 * into a user vma.
1521 *
1522 * The page has to be a nice clean _individual_ kernel allocation.
1523 * If you allocate a compound page, you need to have marked it as
1524 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001525 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001526 *
1527 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1528 * took an arbitrary page protection parameter. This doesn't allow
1529 * that. Your vma protection will have to be set up correctly, which
1530 * means that if you want a shared writable mapping, you'd better
1531 * ask for a shared writable mapping!
1532 *
1533 * The page does not need to be reserved.
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001534 *
1535 * Usually this function is called from f_op->mmap() handler
1536 * under mm->mmap_sem write-lock, so it can change vma->vm_flags.
1537 * Caller must set VM_MIXEDMAP on vma if it wants to call this
1538 * function from other places, for example from page-fault handler.
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001539 */
Nick Piggin423bad62008-04-28 02:13:01 -07001540int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
1541 struct page *page)
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001542{
1543 if (addr < vma->vm_start || addr >= vma->vm_end)
1544 return -EFAULT;
1545 if (!page_count(page))
1546 return -EINVAL;
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001547 if (!(vma->vm_flags & VM_MIXEDMAP)) {
1548 BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem));
1549 BUG_ON(vma->vm_flags & VM_PFNMAP);
1550 vma->vm_flags |= VM_MIXEDMAP;
1551 }
Nick Piggin423bad62008-04-28 02:13:01 -07001552 return insert_page(vma, addr, page, vma->vm_page_prot);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001553}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001554EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001555
Nick Piggin423bad62008-04-28 02:13:01 -07001556static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1557 unsigned long pfn, pgprot_t prot)
1558{
1559 struct mm_struct *mm = vma->vm_mm;
1560 int retval;
1561 pte_t *pte, entry;
1562 spinlock_t *ptl;
1563
1564 retval = -ENOMEM;
1565 pte = get_locked_pte(mm, addr, &ptl);
1566 if (!pte)
1567 goto out;
1568 retval = -EBUSY;
1569 if (!pte_none(*pte))
1570 goto out_unlock;
1571
1572 /* Ok, finally just insert the thing.. */
1573 entry = pte_mkspecial(pfn_pte(pfn, prot));
1574 set_pte_at(mm, addr, pte, entry);
Russell King4b3073e2009-12-18 16:40:18 +00001575 update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */
Nick Piggin423bad62008-04-28 02:13:01 -07001576
1577 retval = 0;
1578out_unlock:
1579 pte_unmap_unlock(pte, ptl);
1580out:
1581 return retval;
1582}
1583
Nick Piggine0dc0d82007-02-12 00:51:36 -08001584/**
1585 * vm_insert_pfn - insert single pfn into user vma
1586 * @vma: user vma to map to
1587 * @addr: target user address of this page
1588 * @pfn: source kernel pfn
1589 *
Robert P. J. Dayc462f172012-10-08 16:33:43 -07001590 * Similar to vm_insert_page, this allows drivers to insert individual pages
Nick Piggine0dc0d82007-02-12 00:51:36 -08001591 * they've allocated into a user vma. Same comments apply.
1592 *
1593 * This function should only be called from a vm_ops->fault handler, and
1594 * in that case the handler should return NULL.
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001595 *
1596 * vma cannot be a COW mapping.
1597 *
1598 * As this is called only for pages that do not currently exist, we
1599 * do not need to flush old virtual caches or the TLB.
Nick Piggine0dc0d82007-02-12 00:51:36 -08001600 */
1601int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
Nick Piggin423bad62008-04-28 02:13:01 -07001602 unsigned long pfn)
Nick Piggine0dc0d82007-02-12 00:51:36 -08001603{
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001604 int ret;
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001605 pgprot_t pgprot = vma->vm_page_prot;
Nick Piggin7e675132008-04-28 02:13:00 -07001606 /*
1607 * Technically, architectures with pte_special can avoid all these
1608 * restrictions (same for remap_pfn_range). However we would like
1609 * consistency in testing and feature parity among all, so we should
1610 * try to keep these invariants in place for everybody.
1611 */
Jared Hulbertb379d792008-04-28 02:12:58 -07001612 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
1613 BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
1614 (VM_PFNMAP|VM_MIXEDMAP));
1615 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1616 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
Nick Piggine0dc0d82007-02-12 00:51:36 -08001617
Nick Piggin423bad62008-04-28 02:13:01 -07001618 if (addr < vma->vm_start || addr >= vma->vm_end)
1619 return -EFAULT;
Suresh Siddha5180da42012-10-08 16:28:29 -07001620 if (track_pfn_insert(vma, &pgprot, pfn))
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001621 return -EINVAL;
1622
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001623 ret = insert_pfn(vma, addr, pfn, pgprot);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001624
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001625 return ret;
Nick Piggine0dc0d82007-02-12 00:51:36 -08001626}
1627EXPORT_SYMBOL(vm_insert_pfn);
1628
Nick Piggin423bad62008-04-28 02:13:01 -07001629int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
1630 unsigned long pfn)
1631{
1632 BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
1633
1634 if (addr < vma->vm_start || addr >= vma->vm_end)
1635 return -EFAULT;
1636
1637 /*
1638 * If we don't have pte special, then we have to use the pfn_valid()
1639 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1640 * refcount the page if pfn_valid is true (hence insert_page rather
Hugh Dickins62eede62009-09-21 17:03:34 -07001641 * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP
1642 * without pte special, it would there be refcounted as a normal page.
Nick Piggin423bad62008-04-28 02:13:01 -07001643 */
1644 if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) {
1645 struct page *page;
1646
1647 page = pfn_to_page(pfn);
1648 return insert_page(vma, addr, page, vma->vm_page_prot);
1649 }
1650 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
1651}
1652EXPORT_SYMBOL(vm_insert_mixed);
1653
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001654/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 * maps a range of physical memory into the requested pages. the old
1656 * mappings are removed. any references to nonexistent pages results
1657 * in null mappings (currently treated as "copy-on-access")
1658 */
1659static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1660 unsigned long addr, unsigned long end,
1661 unsigned long pfn, pgprot_t prot)
1662{
1663 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001664 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665
Hugh Dickinsc74df322005-10-29 18:16:23 -07001666 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 if (!pte)
1668 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001669 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 do {
1671 BUG_ON(!pte_none(*pte));
Nick Piggin7e675132008-04-28 02:13:00 -07001672 set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 pfn++;
1674 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001675 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001676 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 return 0;
1678}
1679
1680static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1681 unsigned long addr, unsigned long end,
1682 unsigned long pfn, pgprot_t prot)
1683{
1684 pmd_t *pmd;
1685 unsigned long next;
1686
1687 pfn -= addr >> PAGE_SHIFT;
1688 pmd = pmd_alloc(mm, pud, addr);
1689 if (!pmd)
1690 return -ENOMEM;
Andrea Arcangelif66055ab2011-01-13 15:46:54 -08001691 VM_BUG_ON(pmd_trans_huge(*pmd));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 do {
1693 next = pmd_addr_end(addr, end);
1694 if (remap_pte_range(mm, pmd, addr, next,
1695 pfn + (addr >> PAGE_SHIFT), prot))
1696 return -ENOMEM;
1697 } while (pmd++, addr = next, addr != end);
1698 return 0;
1699}
1700
1701static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1702 unsigned long addr, unsigned long end,
1703 unsigned long pfn, pgprot_t prot)
1704{
1705 pud_t *pud;
1706 unsigned long next;
1707
1708 pfn -= addr >> PAGE_SHIFT;
1709 pud = pud_alloc(mm, pgd, addr);
1710 if (!pud)
1711 return -ENOMEM;
1712 do {
1713 next = pud_addr_end(addr, end);
1714 if (remap_pmd_range(mm, pud, addr, next,
1715 pfn + (addr >> PAGE_SHIFT), prot))
1716 return -ENOMEM;
1717 } while (pud++, addr = next, addr != end);
1718 return 0;
1719}
1720
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001721/**
1722 * remap_pfn_range - remap kernel memory to userspace
1723 * @vma: user vma to map to
1724 * @addr: target user address to start at
1725 * @pfn: physical address of kernel memory
1726 * @size: size of map area
1727 * @prot: page protection flags for this mapping
1728 *
1729 * Note: this is only safe if the mm semaphore is held when called.
1730 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1732 unsigned long pfn, unsigned long size, pgprot_t prot)
1733{
1734 pgd_t *pgd;
1735 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001736 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 struct mm_struct *mm = vma->vm_mm;
1738 int err;
1739
1740 /*
1741 * Physically remapped pages are special. Tell the
1742 * rest of the world about it:
1743 * VM_IO tells people not to look at these pages
1744 * (accesses can have side effects).
Linus Torvalds6aab3412005-11-28 14:34:23 -08001745 * VM_PFNMAP tells the core MM that the base pages are just
1746 * raw PFN mappings, and do not have a "struct page" associated
1747 * with them.
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -07001748 * VM_DONTEXPAND
1749 * Disable vma merging and expanding with mremap().
1750 * VM_DONTDUMP
1751 * Omit vma from core dump, even when VM_IO turned off.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001752 *
1753 * There's a horrible special case to handle copy-on-write
1754 * behaviour that some programs depend on. We mark the "original"
1755 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07001756 * See vm_normal_page() for details.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 */
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07001758 if (is_cow_mapping(vma->vm_flags)) {
1759 if (addr != vma->vm_start || end != vma->vm_end)
1760 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001761 vma->vm_pgoff = pfn;
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07001762 }
1763
1764 err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size));
1765 if (err)
venkatesh.pallipadi@intel.com3c8bb732008-12-18 11:41:27 -08001766 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001767
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -07001768 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
1770 BUG_ON(addr >= end);
1771 pfn -= addr >> PAGE_SHIFT;
1772 pgd = pgd_offset(mm, addr);
1773 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 do {
1775 next = pgd_addr_end(addr, end);
1776 err = remap_pud_range(mm, pgd, addr, next,
1777 pfn + (addr >> PAGE_SHIFT), prot);
1778 if (err)
1779 break;
1780 } while (pgd++, addr = next, addr != end);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001781
1782 if (err)
Suresh Siddha5180da42012-10-08 16:28:29 -07001783 untrack_pfn(vma, pfn, PAGE_ALIGN(size));
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001784
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 return err;
1786}
1787EXPORT_SYMBOL(remap_pfn_range);
1788
Linus Torvaldsb4cbb192013-04-16 13:45:37 -07001789/**
1790 * vm_iomap_memory - remap memory to userspace
1791 * @vma: user vma to map to
1792 * @start: start of area
1793 * @len: size of area
1794 *
1795 * This is a simplified io_remap_pfn_range() for common driver use. The
1796 * driver just needs to give us the physical memory range to be mapped,
1797 * we'll figure out the rest from the vma information.
1798 *
1799 * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get
1800 * whatever write-combining details or similar.
1801 */
1802int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1803{
1804 unsigned long vm_len, pfn, pages;
1805
1806 /* Check that the physical memory area passed in looks valid */
1807 if (start + len < start)
1808 return -EINVAL;
1809 /*
1810 * You *really* shouldn't map things that aren't page-aligned,
1811 * but we've historically allowed it because IO memory might
1812 * just have smaller alignment.
1813 */
1814 len += start & ~PAGE_MASK;
1815 pfn = start >> PAGE_SHIFT;
1816 pages = (len + ~PAGE_MASK) >> PAGE_SHIFT;
1817 if (pfn + pages < pfn)
1818 return -EINVAL;
1819
1820 /* We start the mapping 'vm_pgoff' pages into the area */
1821 if (vma->vm_pgoff > pages)
1822 return -EINVAL;
1823 pfn += vma->vm_pgoff;
1824 pages -= vma->vm_pgoff;
1825
1826 /* Can we fit all of the mapping? */
1827 vm_len = vma->vm_end - vma->vm_start;
1828 if (vm_len >> PAGE_SHIFT > pages)
1829 return -EINVAL;
1830
1831 /* Ok, let it rip */
1832 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1833}
1834EXPORT_SYMBOL(vm_iomap_memory);
1835
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001836static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1837 unsigned long addr, unsigned long end,
1838 pte_fn_t fn, void *data)
1839{
1840 pte_t *pte;
1841 int err;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001842 pgtable_t token;
Borislav Petkov94909912007-05-06 14:49:17 -07001843 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001844
1845 pte = (mm == &init_mm) ?
1846 pte_alloc_kernel(pmd, addr) :
1847 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1848 if (!pte)
1849 return -ENOMEM;
1850
1851 BUG_ON(pmd_huge(*pmd));
1852
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08001853 arch_enter_lazy_mmu_mode();
1854
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001855 token = pmd_pgtable(*pmd);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001856
1857 do {
Daisuke Nishimurac36987e2009-10-26 16:50:23 -07001858 err = fn(pte++, token, addr, data);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001859 if (err)
1860 break;
Daisuke Nishimurac36987e2009-10-26 16:50:23 -07001861 } while (addr += PAGE_SIZE, addr != end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001862
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08001863 arch_leave_lazy_mmu_mode();
1864
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001865 if (mm != &init_mm)
1866 pte_unmap_unlock(pte-1, ptl);
1867 return err;
1868}
1869
1870static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1871 unsigned long addr, unsigned long end,
1872 pte_fn_t fn, void *data)
1873{
1874 pmd_t *pmd;
1875 unsigned long next;
1876 int err;
1877
Andi Kleenceb86872008-07-23 21:27:50 -07001878 BUG_ON(pud_huge(*pud));
1879
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001880 pmd = pmd_alloc(mm, pud, addr);
1881 if (!pmd)
1882 return -ENOMEM;
1883 do {
1884 next = pmd_addr_end(addr, end);
1885 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1886 if (err)
1887 break;
1888 } while (pmd++, addr = next, addr != end);
1889 return err;
1890}
1891
1892static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1893 unsigned long addr, unsigned long end,
1894 pte_fn_t fn, void *data)
1895{
1896 pud_t *pud;
1897 unsigned long next;
1898 int err;
1899
1900 pud = pud_alloc(mm, pgd, addr);
1901 if (!pud)
1902 return -ENOMEM;
1903 do {
1904 next = pud_addr_end(addr, end);
1905 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1906 if (err)
1907 break;
1908 } while (pud++, addr = next, addr != end);
1909 return err;
1910}
1911
1912/*
1913 * Scan a region of virtual memory, filling in page tables as necessary
1914 * and calling a provided function on each leaf page table.
1915 */
1916int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1917 unsigned long size, pte_fn_t fn, void *data)
1918{
1919 pgd_t *pgd;
1920 unsigned long next;
Jeremy Fitzhardinge57250a52010-08-09 17:19:52 -07001921 unsigned long end = addr + size;
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001922 int err;
1923
1924 BUG_ON(addr >= end);
1925 pgd = pgd_offset(mm, addr);
1926 do {
1927 next = pgd_addr_end(addr, end);
1928 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1929 if (err)
1930 break;
1931 } while (pgd++, addr = next, addr != end);
Jeremy Fitzhardinge57250a52010-08-09 17:19:52 -07001932
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001933 return err;
1934}
1935EXPORT_SYMBOL_GPL(apply_to_page_range);
1936
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001938 * handle_pte_fault chooses page fault handler according to an entry
1939 * which was read non-atomically. Before making any commitment, on
1940 * those architectures or configurations (e.g. i386 with PAE) which
Ryota Ozakia335b2e2011-02-10 13:56:28 +09001941 * might give a mix of unmatched parts, do_swap_page and do_nonlinear_fault
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001942 * must check under lock before unmapping the pte and proceeding
1943 * (but do_wp_page is only called after already making such a check;
Ryota Ozakia335b2e2011-02-10 13:56:28 +09001944 * and do_anonymous_page can safely check later on).
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001945 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001946static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001947 pte_t *page_table, pte_t orig_pte)
1948{
1949 int same = 1;
1950#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1951 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001952 spinlock_t *ptl = pte_lockptr(mm, pmd);
1953 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001954 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001955 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001956 }
1957#endif
1958 pte_unmap(page_table);
1959 return same;
1960}
1961
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001962static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma)
Linus Torvalds6aab3412005-11-28 14:34:23 -08001963{
Dan Williams0abdd7a2014-01-21 15:48:12 -08001964 debug_dma_assert_idle(src);
1965
Linus Torvalds6aab3412005-11-28 14:34:23 -08001966 /*
1967 * If the source page was a PFN mapping, we don't have
1968 * a "struct page" for it. We do a best-effort copy by
1969 * just copying from the original user address. If that
1970 * fails, we just zero-fill it. Live with it.
1971 */
1972 if (unlikely(!src)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08001973 void *kaddr = kmap_atomic(dst);
Linus Torvalds5d2a2dbbc2005-11-29 14:07:55 -08001974 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1975
1976 /*
1977 * This really shouldn't fail, because the page is there
1978 * in the page tables. But it might just be unreadable,
1979 * in which case we just give up and fill the result with
1980 * zeroes.
1981 */
1982 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Jan Beulich3ecb01d2010-10-26 14:22:27 -07001983 clear_page(kaddr);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001984 kunmap_atomic(kaddr);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07001985 flush_dcache_page(dst);
Nick Piggin0ed361d2008-02-04 22:29:34 -08001986 } else
1987 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001988}
1989
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990/*
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07001991 * Notify the address space that the page is about to become writable so that
1992 * it can prohibit this or wait for the page to get into an appropriate state.
1993 *
1994 * We do this without the lock held, so that it can sleep if it needs to.
1995 */
1996static int do_page_mkwrite(struct vm_area_struct *vma, struct page *page,
1997 unsigned long address)
1998{
1999 struct vm_fault vmf;
2000 int ret;
2001
2002 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2003 vmf.pgoff = page->index;
2004 vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
2005 vmf.page = page;
2006
2007 ret = vma->vm_ops->page_mkwrite(vma, &vmf);
2008 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2009 return ret;
2010 if (unlikely(!(ret & VM_FAULT_LOCKED))) {
2011 lock_page(page);
2012 if (!page->mapping) {
2013 unlock_page(page);
2014 return 0; /* retry */
2015 }
2016 ret |= VM_FAULT_LOCKED;
2017 } else
2018 VM_BUG_ON_PAGE(!PageLocked(page), page);
2019 return ret;
2020}
2021
2022/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 * This routine handles present pages, when users try to write
2024 * to a shared page. It is done by copying the page to a new address
2025 * and decrementing the shared-page counter for the old page.
2026 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 * Note that this routine assumes that the protection checks have been
2028 * done by the caller (the low-level page fault routine in most cases).
2029 * Thus we can safely just mark it writable once we've done any necessary
2030 * COW.
2031 *
2032 * We also mark the page dirty at this point even though the page will
2033 * change only once the write actually happens. This avoids a few races,
2034 * and potentially makes it more efficient.
2035 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002036 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2037 * but allow concurrent faults), with pte both mapped and locked.
2038 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002040static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
2041 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002042 spinlock_t *ptl, pte_t orig_pte)
Namhyung Kime6219ec2010-10-26 14:22:00 -07002043 __releases(ptl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044{
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07002045 struct page *old_page, *new_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 pte_t entry;
Michel Lespinasseb009c022011-01-13 15:46:07 -08002047 int ret = 0;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002048 int page_mkwrite = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002049 struct page *dirty_page = NULL;
David Rientjes17569542012-11-16 14:14:48 -08002050 unsigned long mmun_start = 0; /* For mmu_notifiers */
2051 unsigned long mmun_end = 0; /* For mmu_notifiers */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052
Linus Torvalds6aab3412005-11-28 14:34:23 -08002053 old_page = vm_normal_page(vma, address, orig_pte);
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002054 if (!old_page) {
2055 /*
2056 * VM_MIXEDMAP !pfn_valid() case
2057 *
2058 * We should not cow pages in a shared writeable mapping.
2059 * Just mark the pages writable as we can't do any dirty
2060 * accounting on raw pfn maps.
2061 */
2062 if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
2063 (VM_WRITE|VM_SHARED))
2064 goto reuse;
Linus Torvalds6aab3412005-11-28 14:34:23 -08002065 goto gotten;
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002066 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002068 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002069 * Take out anonymous pages first, anonymous shared vmas are
2070 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002071 */
Hugh Dickins9a840892009-09-21 17:02:01 -07002072 if (PageAnon(old_page) && !PageKsm(old_page)) {
Hugh Dickinsab967d82009-01-06 14:39:33 -08002073 if (!trylock_page(old_page)) {
2074 page_cache_get(old_page);
2075 pte_unmap_unlock(page_table, ptl);
2076 lock_page(old_page);
2077 page_table = pte_offset_map_lock(mm, pmd, address,
2078 &ptl);
2079 if (!pte_same(*page_table, orig_pte)) {
2080 unlock_page(old_page);
Hugh Dickinsab967d82009-01-06 14:39:33 -08002081 goto unlock;
2082 }
2083 page_cache_release(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002084 }
Michel Lespinasseb009c022011-01-13 15:46:07 -08002085 if (reuse_swap_page(old_page)) {
Rik van Rielc44b6742010-03-05 13:42:09 -08002086 /*
2087 * The page is all ours. Move it to our anon_vma so
2088 * the rmap code will not search our parent or siblings.
2089 * Protected against the rmap code by the page lock.
2090 */
2091 page_move_anon_rmap(old_page, vma, address);
Michel Lespinasseb009c022011-01-13 15:46:07 -08002092 unlock_page(old_page);
2093 goto reuse;
2094 }
Hugh Dickinsab967d82009-01-06 14:39:33 -08002095 unlock_page(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002096 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002097 (VM_WRITE|VM_SHARED))) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002098 /*
2099 * Only catch write-faults on shared writable pages,
2100 * read-only shared pages can get COWed by
2101 * get_user_pages(.write=1, .force=1).
2102 */
David Howells9637a5e2006-06-23 02:03:43 -07002103 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
Nick Pigginc2ec1752009-03-31 15:23:21 -07002104 int tmp;
David Howells9637a5e2006-06-23 02:03:43 -07002105 page_cache_get(old_page);
2106 pte_unmap_unlock(page_table, ptl);
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07002107 tmp = do_page_mkwrite(vma, old_page, address);
2108 if (unlikely(!tmp || (tmp &
2109 (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
2110 page_cache_release(old_page);
2111 return tmp;
Nick Pigginc2ec1752009-03-31 15:23:21 -07002112 }
David Howells9637a5e2006-06-23 02:03:43 -07002113 /*
2114 * Since we dropped the lock we need to revalidate
2115 * the PTE as someone else may have changed it. If
2116 * they did, we just return, as we can count on the
2117 * MMU to tell us if they didn't also make it writable.
2118 */
2119 page_table = pte_offset_map_lock(mm, pmd, address,
2120 &ptl);
Nick Pigginb827e492009-04-30 15:08:16 -07002121 if (!pte_same(*page_table, orig_pte)) {
2122 unlock_page(old_page);
David Howells9637a5e2006-06-23 02:03:43 -07002123 goto unlock;
Nick Pigginb827e492009-04-30 15:08:16 -07002124 }
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002125
2126 page_mkwrite = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002128 dirty_page = old_page;
2129 get_page(dirty_page);
David Howells9637a5e2006-06-23 02:03:43 -07002130
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002131reuse:
Peter Zijlstra8c8a7432013-10-07 11:29:21 +01002132 /*
2133 * Clear the pages cpupid information as the existing
2134 * information potentially belongs to a now completely
2135 * unrelated process.
2136 */
2137 if (old_page)
2138 page_cpupid_xchg_last(old_page, (1 << LAST_CPUPID_SHIFT) - 1);
2139
David Howells9637a5e2006-06-23 02:03:43 -07002140 flush_cache_page(vma, address, pte_pfn(orig_pte));
2141 entry = pte_mkyoung(orig_pte);
2142 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
KAMEZAWA Hiroyuki954ffcb2007-10-16 01:25:44 -07002143 if (ptep_set_access_flags(vma, address, page_table, entry,1))
Russell King4b3073e2009-12-18 16:40:18 +00002144 update_mmu_cache(vma, address, page_table);
Michel Lespinasse72ddc8f2011-01-13 15:46:08 -08002145 pte_unmap_unlock(page_table, ptl);
David Howells9637a5e2006-06-23 02:03:43 -07002146 ret |= VM_FAULT_WRITE;
Michel Lespinasse72ddc8f2011-01-13 15:46:08 -08002147
2148 if (!dirty_page)
2149 return ret;
2150
2151 /*
2152 * Yes, Virginia, this is actually required to prevent a race
2153 * with clear_page_dirty_for_io() from clearing the page dirty
2154 * bit after it clear all dirty ptes, but before a racing
2155 * do_wp_page installs a dirty pte.
2156 *
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07002157 * do_shared_fault is protected similarly.
Michel Lespinasse72ddc8f2011-01-13 15:46:08 -08002158 */
2159 if (!page_mkwrite) {
2160 wait_on_page_locked(dirty_page);
Miklos Szeredied6d7c82014-04-07 15:37:51 -07002161 set_page_dirty_balance(dirty_page);
Jan Kara41c4d252012-06-12 16:20:28 +02002162 /* file_update_time outside page_lock */
2163 if (vma->vm_file)
2164 file_update_time(vma->vm_file);
Michel Lespinasse72ddc8f2011-01-13 15:46:08 -08002165 }
2166 put_page(dirty_page);
2167 if (page_mkwrite) {
2168 struct address_space *mapping = dirty_page->mapping;
2169
2170 set_page_dirty(dirty_page);
2171 unlock_page(dirty_page);
2172 page_cache_release(dirty_page);
2173 if (mapping) {
2174 /*
2175 * Some device drivers do not set page.mapping
2176 * but still dirty their pages
2177 */
2178 balance_dirty_pages_ratelimited(mapping);
2179 }
2180 }
2181
Michel Lespinasse72ddc8f2011-01-13 15:46:08 -08002182 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184
2185 /*
2186 * Ok, we need to copy. Oh, well..
2187 */
Nick Pigginb5810032005-10-29 18:16:12 -07002188 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08002189gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002190 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191
2192 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07002193 goto oom;
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002194
Hugh Dickins62eede62009-09-21 17:03:34 -07002195 if (is_zero_pfn(pte_pfn(orig_pte))) {
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002196 new_page = alloc_zeroed_user_highpage_movable(vma, address);
2197 if (!new_page)
2198 goto oom;
2199 } else {
2200 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
2201 if (!new_page)
2202 goto oom;
2203 cow_user_page(new_page, old_page, address, vma);
2204 }
2205 __SetPageUptodate(new_page);
2206
Michal Hockod715ae02014-04-07 15:37:46 -07002207 if (mem_cgroup_charge_anon(new_page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002208 goto oom_free_new;
2209
Haggai Eran6bdb9132012-10-08 16:33:35 -07002210 mmun_start = address & PAGE_MASK;
David Rientjes17569542012-11-16 14:14:48 -08002211 mmun_end = mmun_start + PAGE_SIZE;
Haggai Eran6bdb9132012-10-08 16:33:35 -07002212 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
2213
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 /*
2215 * Re-check the pte - we dropped the lock
2216 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002217 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07002218 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08002219 if (old_page) {
Hugh Dickins920fc352005-11-21 21:32:17 -08002220 if (!PageAnon(old_page)) {
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08002221 dec_mm_counter_fast(mm, MM_FILEPAGES);
2222 inc_mm_counter_fast(mm, MM_ANONPAGES);
Hugh Dickins920fc352005-11-21 21:32:17 -08002223 }
2224 } else
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08002225 inc_mm_counter_fast(mm, MM_ANONPAGES);
Ben Collinseca35132005-11-29 11:45:26 -08002226 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07002227 entry = mk_pte(new_page, vma->vm_page_prot);
2228 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Siddha, Suresh B4ce072f2006-09-29 01:58:42 -07002229 /*
2230 * Clear the pte entry and flush it first, before updating the
2231 * pte with the new entry. This will avoid a race condition
2232 * seen in the presence of one thread doing SMC and another
2233 * thread doing COW.
2234 */
Izik Eidus828502d2009-09-21 17:01:51 -07002235 ptep_clear_flush(vma, address, page_table);
Nick Piggin9617d952006-01-06 00:11:12 -08002236 page_add_new_anon_rmap(new_page, vma, address);
Izik Eidus828502d2009-09-21 17:01:51 -07002237 /*
2238 * We call the notify macro here because, when using secondary
2239 * mmu page tables (such as kvm shadow page tables), we want the
2240 * new page to be mapped directly into the secondary page table.
2241 */
2242 set_pte_at_notify(mm, address, page_table, entry);
Russell King4b3073e2009-12-18 16:40:18 +00002243 update_mmu_cache(vma, address, page_table);
Nick Piggin945754a2008-06-23 14:30:30 +02002244 if (old_page) {
2245 /*
2246 * Only after switching the pte to the new page may
2247 * we remove the mapcount here. Otherwise another
2248 * process may come and find the rmap count decremented
2249 * before the pte is switched to the new page, and
2250 * "reuse" the old page writing into it while our pte
2251 * here still points into it and can be read by other
2252 * threads.
2253 *
2254 * The critical issue is to order this
2255 * page_remove_rmap with the ptp_clear_flush above.
2256 * Those stores are ordered by (if nothing else,)
2257 * the barrier present in the atomic_add_negative
2258 * in page_remove_rmap.
2259 *
2260 * Then the TLB flush in ptep_clear_flush ensures that
2261 * no process can access the old page before the
2262 * decremented mapcount is visible. And the old page
2263 * cannot be reused until after the decremented
2264 * mapcount is visible. So transitively, TLBs to
2265 * old page will be flushed before it can be reused.
2266 */
Hugh Dickinsedc315f2009-01-06 14:40:11 -08002267 page_remove_rmap(old_page);
Nick Piggin945754a2008-06-23 14:30:30 +02002268 }
2269
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 /* Free the old page.. */
2271 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10002272 ret |= VM_FAULT_WRITE;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002273 } else
2274 mem_cgroup_uncharge_page(new_page);
2275
Haggai Eran6bdb9132012-10-08 16:33:35 -07002276 if (new_page)
2277 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002278unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002279 pte_unmap_unlock(page_table, ptl);
David Rientjes17569542012-11-16 14:14:48 -08002280 if (mmun_end > mmun_start)
Haggai Eran6bdb9132012-10-08 16:33:35 -07002281 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
Michel Lespinassee15f8c02011-02-10 15:01:32 -08002282 if (old_page) {
2283 /*
2284 * Don't let another task, with possibly unlocked vma,
2285 * keep the mlocked page.
2286 */
2287 if ((ret & VM_FAULT_WRITE) && (vma->vm_flags & VM_LOCKED)) {
2288 lock_page(old_page); /* LRU manipulation */
2289 munlock_vma_page(old_page);
2290 unlock_page(old_page);
2291 }
2292 page_cache_release(old_page);
2293 }
Nick Pigginf33ea7f2005-08-03 20:24:01 +10002294 return ret;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002295oom_free_new:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002296 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002297oom:
Dominik Dingel66521d52012-12-12 13:52:37 -08002298 if (old_page)
Hugh Dickins920fc352005-11-21 21:32:17 -08002299 page_cache_release(old_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 return VM_FAULT_OOM;
2301}
2302
Peter Zijlstra97a89412011-05-24 17:12:04 -07002303static void unmap_mapping_range_vma(struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 unsigned long start_addr, unsigned long end_addr,
2305 struct zap_details *details)
2306{
Al Virof5cc4ee2012-03-05 14:14:20 -05002307 zap_page_range_single(vma, start_addr, end_addr - start_addr, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308}
2309
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07002310static inline void unmap_mapping_range_tree(struct rb_root *root,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 struct zap_details *details)
2312{
2313 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 pgoff_t vba, vea, zba, zea;
2315
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07002316 vma_interval_tree_foreach(vma, root,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 details->first_index, details->last_index) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318
2319 vba = vma->vm_pgoff;
Libind6e93212013-07-03 15:01:26 -07002320 vea = vba + vma_pages(vma) - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
2322 zba = details->first_index;
2323 if (zba < vba)
2324 zba = vba;
2325 zea = details->last_index;
2326 if (zea > vea)
2327 zea = vea;
2328
Peter Zijlstra97a89412011-05-24 17:12:04 -07002329 unmap_mapping_range_vma(vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
2331 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
Peter Zijlstra97a89412011-05-24 17:12:04 -07002332 details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 }
2334}
2335
2336static inline void unmap_mapping_range_list(struct list_head *head,
2337 struct zap_details *details)
2338{
2339 struct vm_area_struct *vma;
2340
2341 /*
2342 * In nonlinear VMAs there is no correspondence between virtual address
2343 * offset and file offset. So we must perform an exhaustive search
2344 * across *all* the pages in each nonlinear VMA, not just the pages
2345 * whose virtual address lies outside the file truncation point.
2346 */
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07002347 list_for_each_entry(vma, head, shared.nonlinear) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 details->nonlinear_vma = vma;
Peter Zijlstra97a89412011-05-24 17:12:04 -07002349 unmap_mapping_range_vma(vma, vma->vm_start, vma->vm_end, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350 }
2351}
2352
2353/**
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002354 * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file.
Martin Waitz3d410882005-06-23 22:05:21 -07002355 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356 * @holebegin: byte in first page to unmap, relative to the start of
2357 * the underlying file. This will be rounded down to a PAGE_SIZE
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +10002358 * boundary. Note that this is different from truncate_pagecache(), which
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 * must keep the partial page. In contrast, we must get rid of
2360 * partial pages.
2361 * @holelen: size of prospective hole in bytes. This will be rounded
2362 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2363 * end of the file.
2364 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2365 * but 0 when invalidating pagecache, don't throw away private data.
2366 */
2367void unmap_mapping_range(struct address_space *mapping,
2368 loff_t const holebegin, loff_t const holelen, int even_cows)
2369{
2370 struct zap_details details;
2371 pgoff_t hba = holebegin >> PAGE_SHIFT;
2372 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2373
2374 /* Check for overflow. */
2375 if (sizeof(holelen) > sizeof(hlen)) {
2376 long long holeend =
2377 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2378 if (holeend & ~(long long)ULONG_MAX)
2379 hlen = ULONG_MAX - hba + 1;
2380 }
2381
2382 details.check_mapping = even_cows? NULL: mapping;
2383 details.nonlinear_vma = NULL;
2384 details.first_index = hba;
2385 details.last_index = hba + hlen - 1;
2386 if (details.last_index < details.first_index)
2387 details.last_index = ULONG_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388
Peter Zijlstra97a89412011-05-24 17:12:04 -07002389
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002390 mutex_lock(&mapping->i_mmap_mutex);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07002391 if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 unmap_mapping_range_tree(&mapping->i_mmap, &details);
2393 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
2394 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002395 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396}
2397EXPORT_SYMBOL(unmap_mapping_range);
2398
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002400 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2401 * but allow concurrent faults), and pte mapped but not yet locked.
2402 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002404static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2405 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002406 unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002408 spinlock_t *ptl;
Hugh Dickins56f31802013-02-22 16:36:10 -08002409 struct page *page, *swapcache;
Hugh Dickins65500d22005-10-29 18:15:59 -07002410 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 pte_t pte;
Michel Lespinassed065bd82010-10-26 14:21:57 -07002412 int locked;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07002413 struct mem_cgroup *ptr;
Rik van Rielad8c2ee2010-08-09 17:19:48 -07002414 int exclusive = 0;
Nick Piggin83c54072007-07-19 01:47:05 -07002415 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002417 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002418 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002419
2420 entry = pte_to_swp_entry(orig_pte);
Andi Kleend1737fd2009-09-16 11:50:06 +02002421 if (unlikely(non_swap_entry(entry))) {
2422 if (is_migration_entry(entry)) {
2423 migration_entry_wait(mm, pmd, address);
2424 } else if (is_hwpoison_entry(entry)) {
2425 ret = VM_FAULT_HWPOISON;
2426 } else {
2427 print_bad_pte(vma, address, orig_pte, NULL);
Hugh Dickinsd99be1a2009-12-14 17:59:04 -08002428 ret = VM_FAULT_SIGBUS;
Andi Kleend1737fd2009-09-16 11:50:06 +02002429 }
Christoph Lameter06972122006-06-23 02:03:35 -07002430 goto out;
2431 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002432 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433 page = lookup_swap_cache(entry);
2434 if (!page) {
Hugh Dickins02098fe2008-02-04 22:28:42 -08002435 page = swapin_readahead(entry,
2436 GFP_HIGHUSER_MOVABLE, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 if (!page) {
2438 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002439 * Back out if somebody else faulted in this pte
2440 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002442 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 if (likely(pte_same(*page_table, orig_pte)))
2444 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002445 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002446 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447 }
2448
2449 /* Had to read the page from swap area: Major fault */
2450 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002451 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07002452 mem_cgroup_count_vm_event(mm, PGMAJFAULT);
Andi Kleend1737fd2009-09-16 11:50:06 +02002453 } else if (PageHWPoison(page)) {
Wu Fengguang71f72522009-12-16 12:19:58 +01002454 /*
2455 * hwpoisoned dirty swapcache pages are kept for killing
2456 * owner processes (which may be unknown at hwpoison time)
2457 */
Andi Kleend1737fd2009-09-16 11:50:06 +02002458 ret = VM_FAULT_HWPOISON;
2459 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins56f31802013-02-22 16:36:10 -08002460 swapcache = page;
Andi Kleen4779cb32009-10-14 01:51:41 +02002461 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 }
2463
Hugh Dickins56f31802013-02-22 16:36:10 -08002464 swapcache = page;
Michel Lespinassed065bd82010-10-26 14:21:57 -07002465 locked = lock_page_or_retry(page, mm, flags);
Rik van Riele709ffd2012-05-29 15:06:18 -07002466
Balbir Singh20a10222007-11-14 17:00:33 -08002467 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Michel Lespinassed065bd82010-10-26 14:21:57 -07002468 if (!locked) {
2469 ret |= VM_FAULT_RETRY;
2470 goto out_release;
2471 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002473 /*
Hugh Dickins31c4a3d2010-09-19 19:40:22 -07002474 * Make sure try_to_free_swap or reuse_swap_page or swapoff did not
2475 * release the swapcache from under us. The page pin, and pte_same
2476 * test below, are not enough to exclude that. Even if it is still
2477 * swapcache, we need to check that the page's swap has not changed.
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002478 */
Hugh Dickins31c4a3d2010-09-19 19:40:22 -07002479 if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val))
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002480 goto out_page;
2481
Hugh Dickinscbf86cf2013-02-22 16:35:08 -08002482 page = ksm_might_need_to_copy(page, vma, address);
2483 if (unlikely(!page)) {
2484 ret = VM_FAULT_OOM;
2485 page = swapcache;
Hugh Dickinscbf86cf2013-02-22 16:35:08 -08002486 goto out_page;
Hugh Dickins5ad64682009-12-14 17:59:24 -08002487 }
2488
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002489 if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) {
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002490 ret = VM_FAULT_OOM;
Johannes Weinerbc43f752009-04-30 15:08:08 -07002491 goto out_page;
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002492 }
2493
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002495 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002497 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002498 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002499 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002500
2501 if (unlikely(!PageUptodate(page))) {
2502 ret = VM_FAULT_SIGBUS;
2503 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 }
2505
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002506 /*
2507 * The page isn't present yet, go ahead with the fault.
2508 *
2509 * Be careful about the sequence of operations here.
2510 * To get its accounting right, reuse_swap_page() must be called
2511 * while the page is counted on swap but not yet in mapcount i.e.
2512 * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
2513 * must be called after the swap_free(), or it will never succeed.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002514 * Because delete_from_swap_page() may be called by reuse_swap_page(),
2515 * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry
2516 * in page->private. In this case, a record in swap_cgroup is silently
2517 * discarded at swap_free().
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002518 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08002520 inc_mm_counter_fast(mm, MM_ANONPAGES);
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08002521 dec_mm_counter_fast(mm, MM_SWAPENTS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 pte = mk_pte(page, vma->vm_page_prot);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002523 if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002525 flags &= ~FAULT_FLAG_WRITE;
Andrea Arcangeli9a5b4892010-08-09 17:19:49 -07002526 ret |= VM_FAULT_WRITE;
Rik van Rielad8c2ee2010-08-09 17:19:48 -07002527 exclusive = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 flush_icache_page(vma, page);
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07002530 if (pte_swp_soft_dirty(orig_pte))
2531 pte = pte_mksoft_dirty(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 set_pte_at(mm, address, page_table, pte);
Hugh Dickins56f31802013-02-22 16:36:10 -08002533 if (page == swapcache)
Johannes Weineraf347702013-02-22 16:32:20 -08002534 do_page_add_anon_rmap(page, vma, address, exclusive);
Hugh Dickins56f31802013-02-22 16:36:10 -08002535 else /* ksm created a completely new copy */
2536 page_add_new_anon_rmap(page, vma, address);
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002537 /* It's better to call commit-charge after rmap is established */
2538 mem_cgroup_commit_charge_swapin(page, ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002540 swap_free(entry);
Nick Pigginb291f002008-10-18 20:26:44 -07002541 if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08002542 try_to_free_swap(page);
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002543 unlock_page(page);
Hugh Dickins56f31802013-02-22 16:36:10 -08002544 if (page != swapcache) {
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002545 /*
2546 * Hold the lock to avoid the swap entry to be reused
2547 * until we take the PT lock for the pte_same() check
2548 * (to avoid false positives from pte_same). For
2549 * further safety release the lock after the swap_free
2550 * so that the swap count won't change under a
2551 * parallel locked swapcache.
2552 */
2553 unlock_page(swapcache);
2554 page_cache_release(swapcache);
2555 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002556
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002557 if (flags & FAULT_FLAG_WRITE) {
Hugh Dickins61469f12008-03-04 14:29:04 -08002558 ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
2559 if (ret & VM_FAULT_ERROR)
2560 ret &= VM_FAULT_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 goto out;
2562 }
2563
2564 /* No need to invalidate - it was non-present before */
Russell King4b3073e2009-12-18 16:40:18 +00002565 update_mmu_cache(vma, address, page_table);
Hugh Dickins65500d22005-10-29 18:15:59 -07002566unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002567 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568out:
2569 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002570out_nomap:
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002571 mem_cgroup_cancel_charge_swapin(ptr);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002572 pte_unmap_unlock(page_table, ptl);
Johannes Weinerbc43f752009-04-30 15:08:08 -07002573out_page:
Kirill Korotaevb8107482005-05-16 21:53:50 -07002574 unlock_page(page);
Andi Kleen4779cb32009-10-14 01:51:41 +02002575out_release:
Kirill Korotaevb8107482005-05-16 21:53:50 -07002576 page_cache_release(page);
Hugh Dickins56f31802013-02-22 16:36:10 -08002577 if (page != swapcache) {
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002578 unlock_page(swapcache);
2579 page_cache_release(swapcache);
2580 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002581 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582}
2583
2584/*
Luck, Tony8ca3eb02010-08-24 11:44:18 -07002585 * This is like a special single-page "expand_{down|up}wards()",
2586 * except we must first make sure that 'address{-|+}PAGE_SIZE'
Linus Torvalds320b2b82010-08-12 17:54:33 -07002587 * doesn't hit another vma.
Linus Torvalds320b2b82010-08-12 17:54:33 -07002588 */
2589static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address)
2590{
2591 address &= PAGE_MASK;
2592 if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) {
Linus Torvalds0e8e50e2010-08-20 16:49:40 -07002593 struct vm_area_struct *prev = vma->vm_prev;
Linus Torvalds320b2b82010-08-12 17:54:33 -07002594
Linus Torvalds0e8e50e2010-08-20 16:49:40 -07002595 /*
2596 * Is there a mapping abutting this one below?
2597 *
2598 * That's only ok if it's the same stack mapping
2599 * that has gotten split..
2600 */
2601 if (prev && prev->vm_end == address)
2602 return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM;
2603
Michal Hockod05f3162011-05-24 17:11:44 -07002604 expand_downwards(vma, address - PAGE_SIZE);
Linus Torvalds320b2b82010-08-12 17:54:33 -07002605 }
Luck, Tony8ca3eb02010-08-24 11:44:18 -07002606 if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) {
2607 struct vm_area_struct *next = vma->vm_next;
2608
2609 /* As VM_GROWSDOWN but s/below/above/ */
2610 if (next && next->vm_start == address + PAGE_SIZE)
2611 return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM;
2612
2613 expand_upwards(vma, address + PAGE_SIZE);
2614 }
Linus Torvalds320b2b82010-08-12 17:54:33 -07002615 return 0;
2616}
2617
2618/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002619 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2620 * but allow concurrent faults), and pte mapped but not yet locked.
2621 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002623static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2624 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002625 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002627 struct page *page;
2628 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630
Linus Torvalds11ac5522010-08-14 11:44:56 -07002631 pte_unmap(page_table);
Linus Torvalds320b2b82010-08-12 17:54:33 -07002632
Linus Torvalds11ac5522010-08-14 11:44:56 -07002633 /* Check if we need to add a guard page to the stack */
2634 if (check_stack_guard_page(vma, address) < 0)
2635 return VM_FAULT_SIGBUS;
2636
2637 /* Use the zero-page for reads */
Hugh Dickins62eede62009-09-21 17:03:34 -07002638 if (!(flags & FAULT_FLAG_WRITE)) {
2639 entry = pte_mkspecial(pfn_pte(my_zero_pfn(address),
2640 vma->vm_page_prot));
Linus Torvalds11ac5522010-08-14 11:44:56 -07002641 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002642 if (!pte_none(*page_table))
2643 goto unlock;
2644 goto setpte;
2645 }
2646
Nick Piggin557ed1f2007-10-16 01:24:40 -07002647 /* Allocate our own private page. */
Nick Piggin557ed1f2007-10-16 01:24:40 -07002648 if (unlikely(anon_vma_prepare(vma)))
2649 goto oom;
2650 page = alloc_zeroed_user_highpage_movable(vma, address);
2651 if (!page)
2652 goto oom;
Minchan Kim52f37622013-04-29 15:08:15 -07002653 /*
2654 * The memory barrier inside __SetPageUptodate makes sure that
2655 * preceeding stores to the page contents become visible before
2656 * the set_pte_at() write.
2657 */
Nick Piggin0ed361d2008-02-04 22:29:34 -08002658 __SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659
Michal Hockod715ae02014-04-07 15:37:46 -07002660 if (mem_cgroup_charge_anon(page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002661 goto oom_free_page;
2662
Nick Piggin557ed1f2007-10-16 01:24:40 -07002663 entry = mk_pte(page, vma->vm_page_prot);
Hugh Dickins1ac0cb52009-09-21 17:03:29 -07002664 if (vma->vm_flags & VM_WRITE)
2665 entry = pte_mkwrite(pte_mkdirty(entry));
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002666
Nick Piggin557ed1f2007-10-16 01:24:40 -07002667 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Andrea Arcangeli1c2fb7a2009-09-21 17:02:22 -07002668 if (!pte_none(*page_table))
Nick Piggin557ed1f2007-10-16 01:24:40 -07002669 goto release;
Hugh Dickins9ba69292009-09-21 17:02:20 -07002670
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08002671 inc_mm_counter_fast(mm, MM_ANONPAGES);
Nick Piggin557ed1f2007-10-16 01:24:40 -07002672 page_add_new_anon_rmap(page, vma, address);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002673setpte:
Hugh Dickins65500d22005-10-29 18:15:59 -07002674 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675
2676 /* No need to invalidate - it was non-present before */
Russell King4b3073e2009-12-18 16:40:18 +00002677 update_mmu_cache(vma, address, page_table);
Hugh Dickins65500d22005-10-29 18:15:59 -07002678unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002679 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002680 return 0;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002681release:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002682 mem_cgroup_uncharge_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002683 page_cache_release(page);
2684 goto unlock;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002685oom_free_page:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002686 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002687oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 return VM_FAULT_OOM;
2689}
2690
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07002691static int __do_fault(struct vm_area_struct *vma, unsigned long address,
2692 pgoff_t pgoff, unsigned int flags, struct page **page)
2693{
2694 struct vm_fault vmf;
2695 int ret;
2696
2697 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2698 vmf.pgoff = pgoff;
2699 vmf.flags = flags;
2700 vmf.page = NULL;
2701
2702 ret = vma->vm_ops->fault(vma, &vmf);
2703 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
2704 return ret;
2705
2706 if (unlikely(PageHWPoison(vmf.page))) {
2707 if (ret & VM_FAULT_LOCKED)
2708 unlock_page(vmf.page);
2709 page_cache_release(vmf.page);
2710 return VM_FAULT_HWPOISON;
2711 }
2712
2713 if (unlikely(!(ret & VM_FAULT_LOCKED)))
2714 lock_page(vmf.page);
2715 else
2716 VM_BUG_ON_PAGE(!PageLocked(vmf.page), vmf.page);
2717
2718 *page = vmf.page;
2719 return ret;
2720}
2721
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002722/**
2723 * do_set_pte - setup new PTE entry for given page and add reverse page mapping.
2724 *
2725 * @vma: virtual memory area
2726 * @address: user virtual address
2727 * @page: page to map
2728 * @pte: pointer to target page table entry
2729 * @write: true, if new entry is writable
2730 * @anon: true, if it's anonymous page
2731 *
2732 * Caller must hold page table lock relevant for @pte.
2733 *
2734 * Target users are page handler itself and implementations of
2735 * vm_ops->map_pages.
2736 */
2737void do_set_pte(struct vm_area_struct *vma, unsigned long address,
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002738 struct page *page, pte_t *pte, bool write, bool anon)
2739{
2740 pte_t entry;
2741
2742 flush_icache_page(vma, page);
2743 entry = mk_pte(page, vma->vm_page_prot);
2744 if (write)
2745 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2746 else if (pte_file(*pte) && pte_file_soft_dirty(*pte))
2747 pte_mksoft_dirty(entry);
2748 if (anon) {
2749 inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES);
2750 page_add_new_anon_rmap(page, vma, address);
2751 } else {
2752 inc_mm_counter_fast(vma->vm_mm, MM_FILEPAGES);
2753 page_add_file_rmap(page);
2754 }
2755 set_pte_at(vma->vm_mm, address, pte, entry);
2756
2757 /* no need to invalidate: a not-present page won't be cached */
2758 update_mmu_cache(vma, address, pte);
2759}
2760
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002761static unsigned long fault_around_bytes = 65536;
2762
Kirill A. Shutemov1fdb4122014-06-04 16:10:55 -07002763/*
2764 * fault_around_pages() and fault_around_mask() round down fault_around_bytes
2765 * to nearest page order. It's what do_fault_around() expects to see.
2766 */
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002767static inline unsigned long fault_around_pages(void)
2768{
2769 return rounddown_pow_of_two(fault_around_bytes) / PAGE_SIZE;
2770}
2771
2772static inline unsigned long fault_around_mask(void)
2773{
2774 return ~(rounddown_pow_of_two(fault_around_bytes) - 1) & PAGE_MASK;
2775}
2776
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002777
2778#ifdef CONFIG_DEBUG_FS
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002779static int fault_around_bytes_get(void *data, u64 *val)
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002780{
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002781 *val = fault_around_bytes;
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002782 return 0;
2783}
2784
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002785static int fault_around_bytes_set(void *data, u64 val)
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002786{
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002787 if (val / PAGE_SIZE > PTRS_PER_PTE)
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002788 return -EINVAL;
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002789 fault_around_bytes = val;
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002790 return 0;
2791}
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002792DEFINE_SIMPLE_ATTRIBUTE(fault_around_bytes_fops,
2793 fault_around_bytes_get, fault_around_bytes_set, "%llu\n");
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002794
2795static int __init fault_around_debugfs(void)
2796{
2797 void *ret;
2798
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002799 ret = debugfs_create_file("fault_around_bytes", 0644, NULL, NULL,
2800 &fault_around_bytes_fops);
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002801 if (!ret)
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002802 pr_warn("Failed to create fault_around_bytes in debugfs");
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002803 return 0;
2804}
2805late_initcall(fault_around_debugfs);
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002806#endif
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002807
Kirill A. Shutemov1fdb4122014-06-04 16:10:55 -07002808/*
2809 * do_fault_around() tries to map few pages around the fault address. The hope
2810 * is that the pages will be needed soon and this will lower the number of
2811 * faults to handle.
2812 *
2813 * It uses vm_ops->map_pages() to map the pages, which skips the page if it's
2814 * not ready to be mapped: not up-to-date, locked, etc.
2815 *
2816 * This function is called with the page table lock taken. In the split ptlock
2817 * case the page table lock only protects only those entries which belong to
2818 * the page table corresponding to the fault address.
2819 *
2820 * This function doesn't cross the VMA boundaries, in order to call map_pages()
2821 * only once.
2822 *
2823 * fault_around_pages() defines how many pages we'll try to map.
2824 * do_fault_around() expects it to return a power of two less than or equal to
2825 * PTRS_PER_PTE.
2826 *
2827 * The virtual address of the area that we map is naturally aligned to the
2828 * fault_around_pages() value (and therefore to page order). This way it's
2829 * easier to guarantee that we don't cross page table boundaries.
2830 */
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002831static void do_fault_around(struct vm_area_struct *vma, unsigned long address,
2832 pte_t *pte, pgoff_t pgoff, unsigned int flags)
2833{
2834 unsigned long start_addr;
2835 pgoff_t max_pgoff;
2836 struct vm_fault vmf;
2837 int off;
2838
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002839 start_addr = max(address & fault_around_mask(), vma->vm_start);
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002840 off = ((address - start_addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
2841 pte -= off;
2842 pgoff -= off;
2843
2844 /*
2845 * max_pgoff is either end of page table or end of vma
Kirill A. Shutemov850e9c62014-06-04 16:10:45 -07002846 * or fault_around_pages() from pgoff, depending what is nearest.
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002847 */
2848 max_pgoff = pgoff - ((start_addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) +
2849 PTRS_PER_PTE - 1;
2850 max_pgoff = min3(max_pgoff, vma_pages(vma) + vma->vm_pgoff - 1,
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002851 pgoff + fault_around_pages() - 1);
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002852
2853 /* Check if it makes any sense to call ->map_pages */
2854 while (!pte_none(*pte)) {
2855 if (++pgoff > max_pgoff)
2856 return;
2857 start_addr += PAGE_SIZE;
2858 if (start_addr >= vma->vm_end)
2859 return;
2860 pte++;
2861 }
2862
2863 vmf.virtual_address = (void __user *) start_addr;
2864 vmf.pte = pte;
2865 vmf.pgoff = pgoff;
2866 vmf.max_pgoff = max_pgoff;
2867 vmf.flags = flags;
2868 vma->vm_ops->map_pages(vma, &vmf);
2869}
2870
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07002871static int do_read_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2872 unsigned long address, pmd_t *pmd,
2873 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
2874{
2875 struct page *fault_page;
2876 spinlock_t *ptl;
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002877 pte_t *pte;
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002878 int ret = 0;
2879
2880 /*
2881 * Let's call ->map_pages() first and use ->fault() as fallback
2882 * if page by the offset is not ready to be mapped (cold cache or
2883 * something).
2884 */
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002885 if (vma->vm_ops->map_pages && fault_around_pages() > 1) {
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002886 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
2887 do_fault_around(vma, address, pte, pgoff, flags);
2888 if (!pte_same(*pte, orig_pte))
2889 goto unlock_out;
2890 pte_unmap_unlock(pte, ptl);
2891 }
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07002892
2893 ret = __do_fault(vma, address, pgoff, flags, &fault_page);
2894 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
2895 return ret;
2896
2897 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
2898 if (unlikely(!pte_same(*pte, orig_pte))) {
2899 pte_unmap_unlock(pte, ptl);
2900 unlock_page(fault_page);
2901 page_cache_release(fault_page);
2902 return ret;
2903 }
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002904 do_set_pte(vma, address, fault_page, pte, false, false);
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07002905 unlock_page(fault_page);
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002906unlock_out:
2907 pte_unmap_unlock(pte, ptl);
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07002908 return ret;
2909}
2910
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07002911static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2912 unsigned long address, pmd_t *pmd,
2913 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
2914{
2915 struct page *fault_page, *new_page;
2916 spinlock_t *ptl;
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002917 pte_t *pte;
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07002918 int ret;
2919
2920 if (unlikely(anon_vma_prepare(vma)))
2921 return VM_FAULT_OOM;
2922
2923 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
2924 if (!new_page)
2925 return VM_FAULT_OOM;
2926
Michal Hockod715ae02014-04-07 15:37:46 -07002927 if (mem_cgroup_charge_anon(new_page, mm, GFP_KERNEL)) {
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07002928 page_cache_release(new_page);
2929 return VM_FAULT_OOM;
2930 }
2931
2932 ret = __do_fault(vma, address, pgoff, flags, &fault_page);
2933 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
2934 goto uncharge_out;
2935
2936 copy_user_highpage(new_page, fault_page, address, vma);
2937 __SetPageUptodate(new_page);
2938
2939 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
2940 if (unlikely(!pte_same(*pte, orig_pte))) {
2941 pte_unmap_unlock(pte, ptl);
2942 unlock_page(fault_page);
2943 page_cache_release(fault_page);
2944 goto uncharge_out;
2945 }
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002946 do_set_pte(vma, address, new_page, pte, true, true);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07002947 pte_unmap_unlock(pte, ptl);
2948 unlock_page(fault_page);
2949 page_cache_release(fault_page);
2950 return ret;
2951uncharge_out:
2952 mem_cgroup_uncharge_page(new_page);
2953 page_cache_release(new_page);
2954 return ret;
2955}
2956
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07002957static int do_shared_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002958 unsigned long address, pmd_t *pmd,
Nick Piggin54cb8822007-07-19 01:46:59 -07002959 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960{
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07002961 struct page *fault_page;
2962 struct address_space *mapping;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002963 spinlock_t *ptl;
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002964 pte_t *pte;
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07002965 int dirtied = 0;
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07002966 int ret, tmp;
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07002967
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07002968 ret = __do_fault(vma, address, pgoff, flags, &fault_page);
2969 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07002970 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002971
2972 /*
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07002973 * Check if the backing address space wants to know that the page is
2974 * about to become writable
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975 */
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07002976 if (vma->vm_ops->page_mkwrite) {
2977 unlock_page(fault_page);
2978 tmp = do_page_mkwrite(vma, fault_page, address);
2979 if (unlikely(!tmp ||
2980 (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
2981 page_cache_release(fault_page);
2982 return tmp;
2983 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984 }
2985
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07002986 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
2987 if (unlikely(!pte_same(*pte, orig_pte))) {
2988 pte_unmap_unlock(pte, ptl);
2989 unlock_page(fault_page);
2990 page_cache_release(fault_page);
2991 return ret;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002992 }
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002993 do_set_pte(vma, address, fault_page, pte, true, false);
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07002994 pte_unmap_unlock(pte, ptl);
2995
2996 if (set_page_dirty(fault_page))
2997 dirtied = 1;
2998 mapping = fault_page->mapping;
2999 unlock_page(fault_page);
3000 if ((dirtied || vma->vm_ops->page_mkwrite) && mapping) {
3001 /*
3002 * Some device drivers do not set page.mapping but still
3003 * dirty their pages
3004 */
3005 balance_dirty_pages_ratelimited(mapping);
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07003006 }
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003007
3008 /* file_update_time outside page_lock */
3009 if (vma->vm_file && !vma->vm_ops->page_mkwrite)
3010 file_update_time(vma->vm_file);
3011
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07003012 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07003013}
Nick Piggind00806b2007-07-19 01:46:57 -07003014
Nick Piggin54cb8822007-07-19 01:46:59 -07003015static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3016 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003017 unsigned int flags, pte_t orig_pte)
Nick Piggin54cb8822007-07-19 01:46:59 -07003018{
3019 pgoff_t pgoff = (((address & PAGE_MASK)
Dean Nelson0da7e012007-10-16 01:24:45 -07003020 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
Nick Piggin54cb8822007-07-19 01:46:59 -07003021
Hugh Dickins16abfa02007-10-04 16:56:06 +01003022 pte_unmap(page_table);
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07003023 if (!(flags & FAULT_FLAG_WRITE))
3024 return do_read_fault(mm, vma, address, pmd, pgoff, flags,
3025 orig_pte);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003026 if (!(vma->vm_flags & VM_SHARED))
3027 return do_cow_fault(mm, vma, address, pmd, pgoff, flags,
3028 orig_pte);
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003029 return do_shared_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Nick Piggin54cb8822007-07-19 01:46:59 -07003030}
3031
Jes Sorensenf4b81802006-09-27 01:50:10 -07003032/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033 * Fault of a previously existing named mapping. Repopulate the pte
3034 * from the encoded file_pte if possible. This enables swappable
3035 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003036 *
3037 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3038 * but allow concurrent faults), and pte mapped but not yet locked.
3039 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040 */
Nick Piggind0217ac2007-07-19 01:47:03 -07003041static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins65500d22005-10-29 18:15:59 -07003042 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003043 unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044{
Hugh Dickins65500d22005-10-29 18:15:59 -07003045 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003047 flags |= FAULT_FLAG_NONLINEAR;
3048
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07003049 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Nick Piggin83c54072007-07-19 01:47:05 -07003050 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051
Hugh Dickins2509ef22009-01-06 14:40:10 -08003052 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
Hugh Dickins65500d22005-10-29 18:15:59 -07003053 /*
3054 * Page table corrupted: show pte and kill process.
3055 */
Hugh Dickins3dc14742009-01-06 14:40:08 -08003056 print_bad_pte(vma, address, orig_pte, NULL);
Hugh Dickinsd99be1a2009-12-14 17:59:04 -08003057 return VM_FAULT_SIGBUS;
Hugh Dickins65500d22005-10-29 18:15:59 -07003058 }
Hugh Dickins65500d22005-10-29 18:15:59 -07003059
3060 pgoff = pte_to_pgoff(orig_pte);
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07003061 if (!(flags & FAULT_FLAG_WRITE))
3062 return do_read_fault(mm, vma, address, pmd, pgoff, flags,
3063 orig_pte);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003064 if (!(vma->vm_flags & VM_SHARED))
3065 return do_cow_fault(mm, vma, address, pmd, pgoff, flags,
3066 orig_pte);
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003067 return do_shared_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068}
3069
Rashika Kheriab19a9932014-04-03 14:48:02 -07003070static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
Rik van Riel04bb2f92013-10-07 11:29:36 +01003071 unsigned long addr, int page_nid,
3072 int *flags)
Mel Gorman9532fec2012-11-15 01:24:32 +00003073{
3074 get_page(page);
3075
3076 count_vm_numa_event(NUMA_HINT_FAULTS);
Rik van Riel04bb2f92013-10-07 11:29:36 +01003077 if (page_nid == numa_node_id()) {
Mel Gorman9532fec2012-11-15 01:24:32 +00003078 count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL);
Rik van Riel04bb2f92013-10-07 11:29:36 +01003079 *flags |= TNF_FAULT_LOCAL;
3080 }
Mel Gorman9532fec2012-11-15 01:24:32 +00003081
3082 return mpol_misplaced(page, vma, addr);
3083}
3084
Rashika Kheriab19a9932014-04-03 14:48:02 -07003085static int do_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,
Mel Gormand10e63f2012-10-25 14:16:31 +02003086 unsigned long addr, pte_t pte, pte_t *ptep, pmd_t *pmd)
3087{
Mel Gorman4daae3b2012-11-02 11:33:45 +00003088 struct page *page = NULL;
Mel Gormand10e63f2012-10-25 14:16:31 +02003089 spinlock_t *ptl;
Mel Gorman8191acb2013-10-07 11:28:45 +01003090 int page_nid = -1;
Peter Zijlstra90572892013-10-07 11:29:20 +01003091 int last_cpupid;
Peter Zijlstracbee9f82012-10-25 14:16:43 +02003092 int target_nid;
Mel Gormanb8593bf2012-11-21 01:18:23 +00003093 bool migrated = false;
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003094 int flags = 0;
Mel Gormand10e63f2012-10-25 14:16:31 +02003095
3096 /*
3097 * The "pte" at this point cannot be used safely without
3098 * validation through pte_unmap_same(). It's of NUMA type but
3099 * the pfn may be screwed if the read is non atomic.
3100 *
3101 * ptep_modify_prot_start is not called as this is clearing
3102 * the _PAGE_NUMA bit and it is not really expected that there
3103 * would be concurrent hardware modifications to the PTE.
3104 */
3105 ptl = pte_lockptr(mm, pmd);
3106 spin_lock(ptl);
Mel Gorman4daae3b2012-11-02 11:33:45 +00003107 if (unlikely(!pte_same(*ptep, pte))) {
3108 pte_unmap_unlock(ptep, ptl);
3109 goto out;
3110 }
3111
Mel Gormand10e63f2012-10-25 14:16:31 +02003112 pte = pte_mknonnuma(pte);
3113 set_pte_at(mm, addr, ptep, pte);
3114 update_mmu_cache(vma, addr, ptep);
3115
3116 page = vm_normal_page(vma, addr, pte);
3117 if (!page) {
3118 pte_unmap_unlock(ptep, ptl);
3119 return 0;
3120 }
Mel Gormana1a46182013-10-07 11:28:50 +01003121 BUG_ON(is_zero_pfn(page_to_pfn(page)));
Mel Gormand10e63f2012-10-25 14:16:31 +02003122
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003123 /*
3124 * Avoid grouping on DSO/COW pages in specific and RO pages
3125 * in general, RO pages shouldn't hurt as much anyway since
3126 * they can be in shared cache state.
3127 */
3128 if (!pte_write(pte))
3129 flags |= TNF_NO_GROUP;
3130
Rik van Rieldabe1d92013-10-07 11:29:34 +01003131 /*
3132 * Flag if the page is shared between multiple address spaces. This
3133 * is later used when determining whether to group tasks together
3134 */
3135 if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED))
3136 flags |= TNF_SHARED;
3137
Peter Zijlstra90572892013-10-07 11:29:20 +01003138 last_cpupid = page_cpupid_last(page);
Mel Gorman8191acb2013-10-07 11:28:45 +01003139 page_nid = page_to_nid(page);
Rik van Riel04bb2f92013-10-07 11:29:36 +01003140 target_nid = numa_migrate_prep(page, vma, addr, page_nid, &flags);
Mel Gormand10e63f2012-10-25 14:16:31 +02003141 pte_unmap_unlock(ptep, ptl);
Mel Gorman4daae3b2012-11-02 11:33:45 +00003142 if (target_nid == -1) {
Mel Gorman4daae3b2012-11-02 11:33:45 +00003143 put_page(page);
3144 goto out;
3145 }
3146
3147 /* Migrate to the requested node */
Mel Gorman1bc115d2013-10-07 11:29:05 +01003148 migrated = migrate_misplaced_page(page, vma, target_nid);
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003149 if (migrated) {
Mel Gorman8191acb2013-10-07 11:28:45 +01003150 page_nid = target_nid;
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003151 flags |= TNF_MIGRATED;
3152 }
Mel Gorman4daae3b2012-11-02 11:33:45 +00003153
3154out:
Mel Gorman8191acb2013-10-07 11:28:45 +01003155 if (page_nid != -1)
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003156 task_numa_fault(last_cpupid, page_nid, 1, flags);
Mel Gormand10e63f2012-10-25 14:16:31 +02003157 return 0;
3158}
3159
Linus Torvalds1da177e2005-04-16 15:20:36 -07003160/*
3161 * These routines also need to handle stuff like marking pages dirty
3162 * and/or accessed for architectures that don't do it in hardware (most
3163 * RISC architectures). The early dirtying is also good on the i386.
3164 *
3165 * There is also a hook called "update_mmu_cache()" that architectures
3166 * with external mmu caches can use to update those (ie the Sparc or
3167 * PowerPC hashed page tables that act as extended TLBs).
3168 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07003169 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3170 * but allow concurrent faults), and pte mapped but not yet locked.
3171 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003172 */
Kirill A. Shutemovc0292552013-09-12 15:14:05 -07003173static int handle_pte_fault(struct mm_struct *mm,
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003174 struct vm_area_struct *vma, unsigned long address,
3175 pte_t *pte, pmd_t *pmd, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003176{
3177 pte_t entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003178 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003179
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07003180 entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003181 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07003182 if (pte_none(entry)) {
Jes Sorensenf4b81802006-09-27 01:50:10 -07003183 if (vma->vm_ops) {
Nick Piggin3c18ddd2008-04-28 02:12:10 -07003184 if (likely(vma->vm_ops->fault))
Nick Piggin54cb8822007-07-19 01:46:59 -07003185 return do_linear_fault(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003186 pte, pmd, flags, entry);
Jes Sorensenf4b81802006-09-27 01:50:10 -07003187 }
3188 return do_anonymous_page(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003189 pte, pmd, flags);
Hugh Dickins65500d22005-10-29 18:15:59 -07003190 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191 if (pte_file(entry))
Nick Piggind0217ac2007-07-19 01:47:03 -07003192 return do_nonlinear_fault(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003193 pte, pmd, flags, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07003194 return do_swap_page(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003195 pte, pmd, flags, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003196 }
3197
Mel Gormand10e63f2012-10-25 14:16:31 +02003198 if (pte_numa(entry))
3199 return do_numa_page(mm, vma, address, entry, pte, pmd);
3200
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07003201 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003202 spin_lock(ptl);
3203 if (unlikely(!pte_same(*pte, entry)))
3204 goto unlock;
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003205 if (flags & FAULT_FLAG_WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003206 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003207 return do_wp_page(mm, vma, address,
3208 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003209 entry = pte_mkdirty(entry);
3210 }
3211 entry = pte_mkyoung(entry);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003212 if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) {
Russell King4b3073e2009-12-18 16:40:18 +00003213 update_mmu_cache(vma, address, pte);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07003214 } else {
3215 /*
3216 * This is needed only for protection faults but the arch code
3217 * is not yet telling us if this is a protection fault or not.
3218 * This still avoids useless tlb flushes for .text page faults
3219 * with threads.
3220 */
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003221 if (flags & FAULT_FLAG_WRITE)
Shaohua Li61c77322010-08-16 09:16:55 +08003222 flush_tlb_fix_spurious_fault(vma, address);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07003223 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003224unlock:
3225 pte_unmap_unlock(pte, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07003226 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003227}
3228
3229/*
3230 * By the time we get here, we already hold the mm semaphore
3231 */
Johannes Weiner519e5242013-09-12 15:13:42 -07003232static int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3233 unsigned long address, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003234{
3235 pgd_t *pgd;
3236 pud_t *pud;
3237 pmd_t *pmd;
3238 pte_t *pte;
3239
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01003240 if (unlikely(is_vm_hugetlb_page(vma)))
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003241 return hugetlb_fault(mm, vma, address, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003242
Linus Torvalds1da177e2005-04-16 15:20:36 -07003243 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003244 pud = pud_alloc(mm, pgd, address);
3245 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07003246 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003247 pmd = pmd_alloc(mm, pud, address);
3248 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07003249 return VM_FAULT_OOM;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003250 if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) {
Kirill A. Shutemovc0292552013-09-12 15:14:05 -07003251 int ret = VM_FAULT_FALLBACK;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003252 if (!vma->vm_ops)
Kirill A. Shutemovc0292552013-09-12 15:14:05 -07003253 ret = do_huge_pmd_anonymous_page(mm, vma, address,
3254 pmd, flags);
3255 if (!(ret & VM_FAULT_FALLBACK))
3256 return ret;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003257 } else {
3258 pmd_t orig_pmd = *pmd;
David Rientjes1f1d06c2012-05-29 15:06:23 -07003259 int ret;
3260
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003261 barrier();
3262 if (pmd_trans_huge(orig_pmd)) {
Will Deacona1dd4502012-12-11 16:01:27 -08003263 unsigned int dirty = flags & FAULT_FLAG_WRITE;
3264
Linus Torvaldse53289c2013-01-09 08:36:54 -08003265 /*
3266 * If the pmd is splitting, return and retry the
3267 * the fault. Alternative: wait until the split
3268 * is done, and goto retry.
3269 */
3270 if (pmd_trans_splitting(orig_pmd))
3271 return 0;
3272
Linus Torvalds3d59eeb2012-12-16 14:33:25 -08003273 if (pmd_numa(orig_pmd))
Mel Gorman4daae3b2012-11-02 11:33:45 +00003274 return do_huge_pmd_numa_page(mm, vma, address,
Mel Gormand10e63f2012-10-25 14:16:31 +02003275 orig_pmd, pmd);
3276
Linus Torvalds3d59eeb2012-12-16 14:33:25 -08003277 if (dirty && !pmd_write(orig_pmd)) {
David Rientjes1f1d06c2012-05-29 15:06:23 -07003278 ret = do_huge_pmd_wp_page(mm, vma, address, pmd,
3279 orig_pmd);
Kirill A. Shutemov9845cbb2014-02-25 15:01:42 -08003280 if (!(ret & VM_FAULT_FALLBACK))
3281 return ret;
Will Deacona1dd4502012-12-11 16:01:27 -08003282 } else {
3283 huge_pmd_set_accessed(mm, vma, address, pmd,
3284 orig_pmd, dirty);
Kirill A. Shutemov9845cbb2014-02-25 15:01:42 -08003285 return 0;
David Rientjes1f1d06c2012-05-29 15:06:23 -07003286 }
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003287 }
3288 }
3289
3290 /*
3291 * Use __pte_alloc instead of pte_alloc_map, because we can't
3292 * run pte_offset_map on the pmd, if an huge pmd could
3293 * materialize from under us from a different thread.
3294 */
Mel Gorman4fd01772011-10-12 21:06:51 +02003295 if (unlikely(pmd_none(*pmd)) &&
3296 unlikely(__pte_alloc(mm, vma, pmd, address)))
Hugh Dickinsc74df322005-10-29 18:16:23 -07003297 return VM_FAULT_OOM;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003298 /* if an huge pmd materialized from under us just retry later */
3299 if (unlikely(pmd_trans_huge(*pmd)))
3300 return 0;
3301 /*
3302 * A regular pmd is established and it can't morph into a huge pmd
3303 * from under us anymore at this point because we hold the mmap_sem
3304 * read mode and khugepaged takes it in write mode. So now it's
3305 * safe to run pte_offset_map().
3306 */
3307 pte = pte_offset_map(pmd, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003309 return handle_pte_fault(mm, vma, address, pte, pmd, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310}
3311
Johannes Weiner519e5242013-09-12 15:13:42 -07003312int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3313 unsigned long address, unsigned int flags)
3314{
3315 int ret;
3316
3317 __set_current_state(TASK_RUNNING);
3318
3319 count_vm_event(PGFAULT);
3320 mem_cgroup_count_vm_event(mm, PGFAULT);
3321
3322 /* do counter updates before entering really critical section. */
3323 check_sync_rss_stat(current);
3324
3325 /*
3326 * Enable the memcg OOM handling for faults triggered in user
3327 * space. Kernel faults are handled more gracefully.
3328 */
3329 if (flags & FAULT_FLAG_USER)
Johannes Weiner49426422013-10-16 13:46:59 -07003330 mem_cgroup_oom_enable();
Johannes Weiner519e5242013-09-12 15:13:42 -07003331
3332 ret = __handle_mm_fault(mm, vma, address, flags);
3333
Johannes Weiner49426422013-10-16 13:46:59 -07003334 if (flags & FAULT_FLAG_USER) {
3335 mem_cgroup_oom_disable();
3336 /*
3337 * The task may have entered a memcg OOM situation but
3338 * if the allocation error was handled gracefully (no
3339 * VM_FAULT_OOM), there is no need to kill anything.
3340 * Just clean up the OOM state peacefully.
3341 */
3342 if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM))
3343 mem_cgroup_oom_synchronize(false);
3344 }
Johannes Weiner3812c8c2013-09-12 15:13:44 -07003345
Johannes Weiner519e5242013-09-12 15:13:42 -07003346 return ret;
3347}
3348
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349#ifndef __PAGETABLE_PUD_FOLDED
3350/*
3351 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07003352 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003353 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07003354int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003355{
Hugh Dickinsc74df322005-10-29 18:16:23 -07003356 pud_t *new = pud_alloc_one(mm, address);
3357 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07003358 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003359
Nick Piggin362a61a2008-05-14 06:37:36 +02003360 smp_wmb(); /* See comment in __pte_alloc */
3361
Hugh Dickins872fec12005-10-29 18:16:21 -07003362 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003363 if (pgd_present(*pgd)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003364 pud_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003365 else
3366 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07003367 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003368 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003369}
3370#endif /* __PAGETABLE_PUD_FOLDED */
3371
3372#ifndef __PAGETABLE_PMD_FOLDED
3373/*
3374 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07003375 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003376 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07003377int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003378{
Hugh Dickinsc74df322005-10-29 18:16:23 -07003379 pmd_t *new = pmd_alloc_one(mm, address);
3380 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07003381 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003382
Nick Piggin362a61a2008-05-14 06:37:36 +02003383 smp_wmb(); /* See comment in __pte_alloc */
3384
Hugh Dickins872fec12005-10-29 18:16:21 -07003385 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003386#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07003387 if (pud_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003388 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003389 else
3390 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07003392 if (pgd_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003393 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003394 else
3395 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003396#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07003397 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003398 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003399}
3400#endif /* __PAGETABLE_PMD_FOLDED */
3401
Linus Torvalds1da177e2005-04-16 15:20:36 -07003402#if !defined(__HAVE_ARCH_GATE_AREA)
3403
3404#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07003405static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003406
3407static int __init gate_vma_init(void)
3408{
3409 gate_vma.vm_mm = NULL;
3410 gate_vma.vm_start = FIXADDR_USER_START;
3411 gate_vma.vm_end = FIXADDR_USER_END;
Roland McGrathb6558c42007-01-26 00:56:47 -08003412 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
3413 gate_vma.vm_page_prot = __P101;
Jason Baron909af762012-03-23 15:02:51 -07003414
Linus Torvalds1da177e2005-04-16 15:20:36 -07003415 return 0;
3416}
3417__initcall(gate_vma_init);
3418#endif
3419
Stephen Wilson31db58b2011-03-13 15:49:15 -04003420struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003421{
3422#ifdef AT_SYSINFO_EHDR
3423 return &gate_vma;
3424#else
3425 return NULL;
3426#endif
3427}
3428
Stephen Wilsoncae5d392011-03-13 15:49:17 -04003429int in_gate_area_no_mm(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003430{
3431#ifdef AT_SYSINFO_EHDR
3432 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
3433 return 1;
3434#endif
3435 return 0;
3436}
3437
3438#endif /* __HAVE_ARCH_GATE_AREA */
David Howells0ec76a12006-09-27 01:50:15 -07003439
Namhyung Kim1b36ba82010-10-26 14:22:00 -07003440static int __follow_pte(struct mm_struct *mm, unsigned long address,
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07003441 pte_t **ptepp, spinlock_t **ptlp)
3442{
3443 pgd_t *pgd;
3444 pud_t *pud;
3445 pmd_t *pmd;
3446 pte_t *ptep;
3447
3448 pgd = pgd_offset(mm, address);
3449 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
3450 goto out;
3451
3452 pud = pud_offset(pgd, address);
3453 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
3454 goto out;
3455
3456 pmd = pmd_offset(pud, address);
Andrea Arcangelif66055ab2011-01-13 15:46:54 -08003457 VM_BUG_ON(pmd_trans_huge(*pmd));
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07003458 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
3459 goto out;
3460
3461 /* We cannot handle huge page PFN maps. Luckily they don't exist. */
3462 if (pmd_huge(*pmd))
3463 goto out;
3464
3465 ptep = pte_offset_map_lock(mm, pmd, address, ptlp);
3466 if (!ptep)
3467 goto out;
3468 if (!pte_present(*ptep))
3469 goto unlock;
3470 *ptepp = ptep;
3471 return 0;
3472unlock:
3473 pte_unmap_unlock(ptep, *ptlp);
3474out:
3475 return -EINVAL;
3476}
3477
Namhyung Kim1b36ba82010-10-26 14:22:00 -07003478static inline int follow_pte(struct mm_struct *mm, unsigned long address,
3479 pte_t **ptepp, spinlock_t **ptlp)
3480{
3481 int res;
3482
3483 /* (void) is needed to make gcc happy */
3484 (void) __cond_lock(*ptlp,
3485 !(res = __follow_pte(mm, address, ptepp, ptlp)));
3486 return res;
3487}
3488
Johannes Weiner3b6748e2009-06-16 15:32:35 -07003489/**
3490 * follow_pfn - look up PFN at a user virtual address
3491 * @vma: memory mapping
3492 * @address: user virtual address
3493 * @pfn: location to store found PFN
3494 *
3495 * Only IO mappings and raw PFN mappings are allowed.
3496 *
3497 * Returns zero and the pfn at @pfn on success, -ve otherwise.
3498 */
3499int follow_pfn(struct vm_area_struct *vma, unsigned long address,
3500 unsigned long *pfn)
3501{
3502 int ret = -EINVAL;
3503 spinlock_t *ptl;
3504 pte_t *ptep;
3505
3506 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3507 return ret;
3508
3509 ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);
3510 if (ret)
3511 return ret;
3512 *pfn = pte_pfn(*ptep);
3513 pte_unmap_unlock(ptep, ptl);
3514 return 0;
3515}
3516EXPORT_SYMBOL(follow_pfn);
3517
Rik van Riel28b2ee22008-07-23 21:27:05 -07003518#ifdef CONFIG_HAVE_IOREMAP_PROT
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003519int follow_phys(struct vm_area_struct *vma,
3520 unsigned long address, unsigned int flags,
3521 unsigned long *prot, resource_size_t *phys)
Rik van Riel28b2ee22008-07-23 21:27:05 -07003522{
Johannes Weiner03668a42009-06-16 15:32:34 -07003523 int ret = -EINVAL;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003524 pte_t *ptep, pte;
3525 spinlock_t *ptl;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003526
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003527 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3528 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003529
Johannes Weiner03668a42009-06-16 15:32:34 -07003530 if (follow_pte(vma->vm_mm, address, &ptep, &ptl))
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003531 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003532 pte = *ptep;
Johannes Weiner03668a42009-06-16 15:32:34 -07003533
Rik van Riel28b2ee22008-07-23 21:27:05 -07003534 if ((flags & FOLL_WRITE) && !pte_write(pte))
3535 goto unlock;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003536
3537 *prot = pgprot_val(pte_pgprot(pte));
Johannes Weiner03668a42009-06-16 15:32:34 -07003538 *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003539
Johannes Weiner03668a42009-06-16 15:32:34 -07003540 ret = 0;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003541unlock:
3542 pte_unmap_unlock(ptep, ptl);
3543out:
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003544 return ret;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003545}
3546
3547int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
3548 void *buf, int len, int write)
3549{
3550 resource_size_t phys_addr;
3551 unsigned long prot = 0;
KOSAKI Motohiro2bc72732009-01-06 14:39:43 -08003552 void __iomem *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003553 int offset = addr & (PAGE_SIZE-1);
3554
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003555 if (follow_phys(vma, addr, write, &prot, &phys_addr))
Rik van Riel28b2ee22008-07-23 21:27:05 -07003556 return -EINVAL;
3557
3558 maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot);
3559 if (write)
3560 memcpy_toio(maddr + offset, buf, len);
3561 else
3562 memcpy_fromio(buf, maddr + offset, len);
3563 iounmap(maddr);
3564
3565 return len;
3566}
Uwe Kleine-König5a736332013-08-07 13:02:52 +02003567EXPORT_SYMBOL_GPL(generic_access_phys);
Rik van Riel28b2ee22008-07-23 21:27:05 -07003568#endif
3569
David Howells0ec76a12006-09-27 01:50:15 -07003570/*
Stephen Wilson206cb632011-03-13 15:49:19 -04003571 * Access another process' address space as given in mm. If non-NULL, use the
3572 * given task for page fault accounting.
David Howells0ec76a12006-09-27 01:50:15 -07003573 */
Stephen Wilson206cb632011-03-13 15:49:19 -04003574static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
3575 unsigned long addr, void *buf, int len, int write)
David Howells0ec76a12006-09-27 01:50:15 -07003576{
David Howells0ec76a12006-09-27 01:50:15 -07003577 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07003578 void *old_buf = buf;
3579
David Howells0ec76a12006-09-27 01:50:15 -07003580 down_read(&mm->mmap_sem);
Simon Arlott183ff222007-10-20 01:27:18 +02003581 /* ignore errors, just check how much was successfully transferred */
David Howells0ec76a12006-09-27 01:50:15 -07003582 while (len) {
3583 int bytes, ret, offset;
3584 void *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003585 struct page *page = NULL;
David Howells0ec76a12006-09-27 01:50:15 -07003586
3587 ret = get_user_pages(tsk, mm, addr, 1,
3588 write, 1, &page, &vma);
Rik van Riel28b2ee22008-07-23 21:27:05 -07003589 if (ret <= 0) {
3590 /*
3591 * Check if this is a VM_IO | VM_PFNMAP VMA, which
3592 * we can access using slightly different code.
3593 */
3594#ifdef CONFIG_HAVE_IOREMAP_PROT
3595 vma = find_vma(mm, addr);
Michael Ellermanfe936df2011-04-14 15:22:10 -07003596 if (!vma || vma->vm_start > addr)
Rik van Riel28b2ee22008-07-23 21:27:05 -07003597 break;
3598 if (vma->vm_ops && vma->vm_ops->access)
3599 ret = vma->vm_ops->access(vma, addr, buf,
3600 len, write);
3601 if (ret <= 0)
3602#endif
3603 break;
3604 bytes = ret;
David Howells0ec76a12006-09-27 01:50:15 -07003605 } else {
Rik van Riel28b2ee22008-07-23 21:27:05 -07003606 bytes = len;
3607 offset = addr & (PAGE_SIZE-1);
3608 if (bytes > PAGE_SIZE-offset)
3609 bytes = PAGE_SIZE-offset;
3610
3611 maddr = kmap(page);
3612 if (write) {
3613 copy_to_user_page(vma, page, addr,
3614 maddr + offset, buf, bytes);
3615 set_page_dirty_lock(page);
3616 } else {
3617 copy_from_user_page(vma, page, addr,
3618 buf, maddr + offset, bytes);
3619 }
3620 kunmap(page);
3621 page_cache_release(page);
David Howells0ec76a12006-09-27 01:50:15 -07003622 }
David Howells0ec76a12006-09-27 01:50:15 -07003623 len -= bytes;
3624 buf += bytes;
3625 addr += bytes;
3626 }
3627 up_read(&mm->mmap_sem);
David Howells0ec76a12006-09-27 01:50:15 -07003628
3629 return buf - old_buf;
3630}
Andi Kleen03252912008-01-30 13:33:18 +01003631
Stephen Wilson5ddd36b2011-03-13 15:49:20 -04003632/**
Randy Dunlapae91dbf2011-03-26 13:27:01 -07003633 * access_remote_vm - access another process' address space
Stephen Wilson5ddd36b2011-03-13 15:49:20 -04003634 * @mm: the mm_struct of the target address space
3635 * @addr: start address to access
3636 * @buf: source or destination buffer
3637 * @len: number of bytes to transfer
3638 * @write: whether the access is a write
3639 *
3640 * The caller must hold a reference on @mm.
3641 */
3642int access_remote_vm(struct mm_struct *mm, unsigned long addr,
3643 void *buf, int len, int write)
3644{
3645 return __access_remote_vm(NULL, mm, addr, buf, len, write);
3646}
3647
Andi Kleen03252912008-01-30 13:33:18 +01003648/*
Stephen Wilson206cb632011-03-13 15:49:19 -04003649 * Access another process' address space.
3650 * Source/target buffer must be kernel space,
3651 * Do not walk the page table directly, use get_user_pages
3652 */
3653int access_process_vm(struct task_struct *tsk, unsigned long addr,
3654 void *buf, int len, int write)
3655{
3656 struct mm_struct *mm;
3657 int ret;
3658
3659 mm = get_task_mm(tsk);
3660 if (!mm)
3661 return 0;
3662
3663 ret = __access_remote_vm(tsk, mm, addr, buf, len, write);
3664 mmput(mm);
3665
3666 return ret;
3667}
3668
Andi Kleen03252912008-01-30 13:33:18 +01003669/*
3670 * Print the name of a VMA.
3671 */
3672void print_vma_addr(char *prefix, unsigned long ip)
3673{
3674 struct mm_struct *mm = current->mm;
3675 struct vm_area_struct *vma;
3676
Ingo Molnare8bff742008-02-13 20:21:06 +01003677 /*
3678 * Do not print if we are in atomic
3679 * contexts (in exception stacks, etc.):
3680 */
3681 if (preempt_count())
3682 return;
3683
Andi Kleen03252912008-01-30 13:33:18 +01003684 down_read(&mm->mmap_sem);
3685 vma = find_vma(mm, ip);
3686 if (vma && vma->vm_file) {
3687 struct file *f = vma->vm_file;
3688 char *buf = (char *)__get_free_page(GFP_KERNEL);
3689 if (buf) {
Andy Shevchenko2fbc57c2012-12-17 16:01:23 -08003690 char *p;
Andi Kleen03252912008-01-30 13:33:18 +01003691
Jan Blunckcf28b482008-02-14 19:38:44 -08003692 p = d_path(&f->f_path, buf, PAGE_SIZE);
Andi Kleen03252912008-01-30 13:33:18 +01003693 if (IS_ERR(p))
3694 p = "?";
Andy Shevchenko2fbc57c2012-12-17 16:01:23 -08003695 printk("%s%s[%lx+%lx]", prefix, kbasename(p),
Andi Kleen03252912008-01-30 13:33:18 +01003696 vma->vm_start,
3697 vma->vm_end - vma->vm_start);
3698 free_page((unsigned long)buf);
3699 }
3700 }
Jeff Liu51a07e52012-07-31 16:43:18 -07003701 up_read(&mm->mmap_sem);
Andi Kleen03252912008-01-30 13:33:18 +01003702}
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003703
Michael S. Tsirkin662bbcb2013-05-26 17:32:23 +03003704#if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP)
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003705void might_fault(void)
3706{
Peter Zijlstra95156f02009-01-12 13:02:11 +01003707 /*
3708 * Some code (nfs/sunrpc) uses socket ops on kernel memory while
3709 * holding the mmap_sem, this is safe because kernel memory doesn't
3710 * get paged out, therefore we'll never actually fault, and the
3711 * below annotations will generate false positives.
3712 */
3713 if (segment_eq(get_fs(), KERNEL_DS))
3714 return;
3715
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003716 /*
3717 * it would be nicer only to annotate paths which are not under
3718 * pagefault_disable, however that requires a larger audit and
3719 * providing helpers like get_user_atomic.
3720 */
Michael S. Tsirkin662bbcb2013-05-26 17:32:23 +03003721 if (in_atomic())
3722 return;
3723
3724 __might_sleep(__FILE__, __LINE__, 0);
3725
3726 if (current->mm)
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003727 might_lock_read(&current->mm->mmap_sem);
3728}
3729EXPORT_SYMBOL(might_fault);
3730#endif
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08003731
3732#if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)
3733static void clear_gigantic_page(struct page *page,
3734 unsigned long addr,
3735 unsigned int pages_per_huge_page)
3736{
3737 int i;
3738 struct page *p = page;
3739
3740 might_sleep();
3741 for (i = 0; i < pages_per_huge_page;
3742 i++, p = mem_map_next(p, page, i)) {
3743 cond_resched();
3744 clear_user_highpage(p, addr + i * PAGE_SIZE);
3745 }
3746}
3747void clear_huge_page(struct page *page,
3748 unsigned long addr, unsigned int pages_per_huge_page)
3749{
3750 int i;
3751
3752 if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
3753 clear_gigantic_page(page, addr, pages_per_huge_page);
3754 return;
3755 }
3756
3757 might_sleep();
3758 for (i = 0; i < pages_per_huge_page; i++) {
3759 cond_resched();
3760 clear_user_highpage(page + i, addr + i * PAGE_SIZE);
3761 }
3762}
3763
3764static void copy_user_gigantic_page(struct page *dst, struct page *src,
3765 unsigned long addr,
3766 struct vm_area_struct *vma,
3767 unsigned int pages_per_huge_page)
3768{
3769 int i;
3770 struct page *dst_base = dst;
3771 struct page *src_base = src;
3772
3773 for (i = 0; i < pages_per_huge_page; ) {
3774 cond_resched();
3775 copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma);
3776
3777 i++;
3778 dst = mem_map_next(dst, dst_base, i);
3779 src = mem_map_next(src, src_base, i);
3780 }
3781}
3782
3783void copy_user_huge_page(struct page *dst, struct page *src,
3784 unsigned long addr, struct vm_area_struct *vma,
3785 unsigned int pages_per_huge_page)
3786{
3787 int i;
3788
3789 if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
3790 copy_user_gigantic_page(dst, src, addr, vma,
3791 pages_per_huge_page);
3792 return;
3793 }
3794
3795 might_sleep();
3796 for (i = 0; i < pages_per_huge_page; i++) {
3797 cond_resched();
3798 copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma);
3799 }
3800}
3801#endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003802
Olof Johansson40b64ac2013-12-20 14:28:05 -08003803#if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS
Kirill A. Shutemovb35f1812014-01-21 15:49:07 -08003804
3805static struct kmem_cache *page_ptl_cachep;
3806
3807void __init ptlock_cache_init(void)
3808{
3809 page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0,
3810 SLAB_PANIC, NULL);
3811}
3812
Peter Zijlstra539edb52013-11-14 14:31:52 -08003813bool ptlock_alloc(struct page *page)
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003814{
3815 spinlock_t *ptl;
3816
Kirill A. Shutemovb35f1812014-01-21 15:49:07 -08003817 ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL);
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003818 if (!ptl)
3819 return false;
Peter Zijlstra539edb52013-11-14 14:31:52 -08003820 page->ptl = ptl;
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003821 return true;
3822}
3823
Peter Zijlstra539edb52013-11-14 14:31:52 -08003824void ptlock_free(struct page *page)
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003825{
Kirill A. Shutemovb35f1812014-01-21 15:49:07 -08003826 kmem_cache_free(page_ptl_cachep, page->ptl);
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003827}
3828#endif