blob: c387430f06c319d52794152b7e100a66a10c9f68 [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>
Andrea Arcangeli6b251fc2015-09-04 15:46:20 -070064#include <linux/userfaultfd_k.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
Alexey Dobriyan6952b612009-09-18 23:55:55 +040066#include <asm/io.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070067#include <asm/pgalloc.h>
68#include <asm/uaccess.h>
69#include <asm/tlb.h>
70#include <asm/tlbflush.h>
71#include <asm/pgtable.h>
72
Jan Beulich42b77722008-07-23 21:27:10 -070073#include "internal.h"
74
Peter Zijlstra90572892013-10-07 11:29:20 +010075#ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
76#warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid.
Peter Zijlstra75980e92013-02-22 16:34:32 -080077#endif
78
Andy Whitcroftd41dee32005-06-23 00:07:54 -070079#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -070080/* use the per-pgdat data instead for discontigmem - mbligh */
81unsigned long max_mapnr;
82struct page *mem_map;
83
84EXPORT_SYMBOL(max_mapnr);
85EXPORT_SYMBOL(mem_map);
86#endif
87
Linus Torvalds1da177e2005-04-16 15:20:36 -070088/*
89 * A number of key systems in x86 including ioremap() rely on the assumption
90 * that high_memory defines the upper bound on direct map memory, then end
91 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
92 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
93 * and ZONE_HIGHMEM.
94 */
95void * high_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
Linus Torvalds1da177e2005-04-16 15:20:36 -070097EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -070098
Ingo Molnar32a93232008-02-06 22:39:44 +010099/*
100 * Randomize the address space (stacks, mmaps, brk, etc.).
101 *
102 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
103 * as ancient (libc5 based) binaries can segfault. )
104 */
105int randomize_va_space __read_mostly =
106#ifdef CONFIG_COMPAT_BRK
107 1;
108#else
109 2;
110#endif
Andi Kleena62eaf12006-02-16 23:41:58 +0100111
112static int __init disable_randmaps(char *s)
113{
114 randomize_va_space = 0;
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800115 return 1;
Andi Kleena62eaf12006-02-16 23:41:58 +0100116}
117__setup("norandmaps", disable_randmaps);
118
Hugh Dickins62eede62009-09-21 17:03:34 -0700119unsigned long zero_pfn __read_mostly;
Hugh Dickins03f64622009-09-21 17:03:35 -0700120unsigned long highest_memmap_pfn __read_mostly;
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -0700121
Ard Biesheuvel0b700682014-09-12 22:17:23 +0200122EXPORT_SYMBOL(zero_pfn);
123
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -0700124/*
125 * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init()
126 */
127static int __init init_zero_pfn(void)
128{
129 zero_pfn = page_to_pfn(ZERO_PAGE(0));
130 return 0;
131}
132core_initcall(init_zero_pfn);
Andi Kleena62eaf12006-02-16 23:41:58 +0100133
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800134
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800135#if defined(SPLIT_RSS_COUNTING)
136
David Rientjesea48cf72012-03-21 16:34:13 -0700137void sync_mm_rss(struct mm_struct *mm)
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800138{
139 int i;
140
141 for (i = 0; i < NR_MM_COUNTERS; i++) {
David Rientjes05af2e12012-03-21 16:34:13 -0700142 if (current->rss_stat.count[i]) {
143 add_mm_counter(mm, i, current->rss_stat.count[i]);
144 current->rss_stat.count[i] = 0;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800145 }
146 }
David Rientjes05af2e12012-03-21 16:34:13 -0700147 current->rss_stat.events = 0;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800148}
149
150static void add_mm_counter_fast(struct mm_struct *mm, int member, int val)
151{
152 struct task_struct *task = current;
153
154 if (likely(task->mm == mm))
155 task->rss_stat.count[member] += val;
156 else
157 add_mm_counter(mm, member, val);
158}
159#define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1)
160#define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1)
161
162/* sync counter once per 64 page faults */
163#define TASK_RSS_EVENTS_THRESH (64)
164static void check_sync_rss_stat(struct task_struct *task)
165{
166 if (unlikely(task != current))
167 return;
168 if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH))
David Rientjesea48cf72012-03-21 16:34:13 -0700169 sync_mm_rss(task->mm);
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800170}
Peter Zijlstra9547d012011-05-24 17:12:14 -0700171#else /* SPLIT_RSS_COUNTING */
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800172
173#define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member)
174#define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member)
175
176static void check_sync_rss_stat(struct task_struct *task)
177{
178}
179
Peter Zijlstra9547d012011-05-24 17:12:14 -0700180#endif /* SPLIT_RSS_COUNTING */
181
182#ifdef HAVE_GENERIC_MMU_GATHER
183
Nicholas Krauseca1d6c72015-09-04 15:48:22 -0700184static bool tlb_next_batch(struct mmu_gather *tlb)
Peter Zijlstra9547d012011-05-24 17:12:14 -0700185{
186 struct mmu_gather_batch *batch;
187
188 batch = tlb->active;
189 if (batch->next) {
190 tlb->active = batch->next;
Nicholas Krauseca1d6c72015-09-04 15:48:22 -0700191 return true;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700192 }
193
Michal Hocko53a59fc2013-01-04 15:35:12 -0800194 if (tlb->batch_count == MAX_GATHER_BATCH_COUNT)
Nicholas Krauseca1d6c72015-09-04 15:48:22 -0700195 return false;
Michal Hocko53a59fc2013-01-04 15:35:12 -0800196
Peter Zijlstra9547d012011-05-24 17:12:14 -0700197 batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0);
198 if (!batch)
Nicholas Krauseca1d6c72015-09-04 15:48:22 -0700199 return false;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700200
Michal Hocko53a59fc2013-01-04 15:35:12 -0800201 tlb->batch_count++;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700202 batch->next = NULL;
203 batch->nr = 0;
204 batch->max = MAX_GATHER_BATCH;
205
206 tlb->active->next = batch;
207 tlb->active = batch;
208
Nicholas Krauseca1d6c72015-09-04 15:48:22 -0700209 return true;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700210}
211
212/* tlb_gather_mmu
213 * Called to initialize an (on-stack) mmu_gather structure for page-table
214 * tear-down from @mm. The @fullmm argument is used when @mm is without
215 * users and we're going to destroy the full address space (exit/execve).
216 */
Linus Torvalds2b047252013-08-15 11:42:25 -0700217void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, unsigned long start, unsigned long end)
Peter Zijlstra9547d012011-05-24 17:12:14 -0700218{
219 tlb->mm = mm;
220
Linus Torvalds2b047252013-08-15 11:42:25 -0700221 /* Is it from 0 to ~0? */
222 tlb->fullmm = !(start | (end+1));
Dave Hansen1de14c32013-04-12 16:23:54 -0700223 tlb->need_flush_all = 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
Will Deaconfb7332a2014-10-29 10:03:09 +0000233
234 __tlb_reset_range(tlb);
Peter Zijlstra9547d012011-05-24 17:12:14 -0700235}
236
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700237static void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb)
Peter Zijlstra9547d012011-05-24 17:12:14 -0700238{
Will Deacon721c21c2015-01-12 19:10:55 +0000239 if (!tlb->end)
240 return;
241
Peter Zijlstra9547d012011-05-24 17:12:14 -0700242 tlb_flush(tlb);
Joerg Roedel34ee6452014-11-13 13:46:09 +1100243 mmu_notifier_invalidate_range(tlb->mm, tlb->start, tlb->end);
Peter Zijlstra9547d012011-05-24 17:12:14 -0700244#ifdef CONFIG_HAVE_RCU_TABLE_FREE
245 tlb_table_flush(tlb);
246#endif
Will Deaconfb7332a2014-10-29 10:03:09 +0000247 __tlb_reset_range(tlb);
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700248}
249
250static void tlb_flush_mmu_free(struct mmu_gather *tlb)
251{
252 struct mmu_gather_batch *batch;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700253
Will Deacon721c21c2015-01-12 19:10:55 +0000254 for (batch = &tlb->local; batch && batch->nr; batch = batch->next) {
Peter Zijlstra9547d012011-05-24 17:12:14 -0700255 free_pages_and_swap_cache(batch->pages, batch->nr);
256 batch->nr = 0;
257 }
258 tlb->active = &tlb->local;
259}
260
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700261void tlb_flush_mmu(struct mmu_gather *tlb)
262{
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700263 tlb_flush_mmu_tlbonly(tlb);
264 tlb_flush_mmu_free(tlb);
265}
266
Peter Zijlstra9547d012011-05-24 17:12:14 -0700267/* tlb_finish_mmu
268 * Called at the end of the shootdown operation to free up any resources
269 * that were required.
270 */
271void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end)
272{
273 struct mmu_gather_batch *batch, *next;
274
275 tlb_flush_mmu(tlb);
276
277 /* keep the page table cache within bounds */
278 check_pgt_cache();
279
280 for (batch = tlb->local.next; batch; batch = next) {
281 next = batch->next;
282 free_pages((unsigned long)batch, 0);
283 }
284 tlb->local.next = NULL;
285}
286
287/* __tlb_remove_page
288 * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while
289 * handling the additional races in SMP caused by other CPUs caching valid
290 * mappings in their TLBs. Returns the number of free page slots left.
291 * When out of page slots we must call tlb_flush_mmu().
292 */
293int __tlb_remove_page(struct mmu_gather *tlb, struct page *page)
294{
295 struct mmu_gather_batch *batch;
296
Will Deaconfb7332a2014-10-29 10:03:09 +0000297 VM_BUG_ON(!tlb->end);
Peter Zijlstra9547d012011-05-24 17:12:14 -0700298
Peter Zijlstra9547d012011-05-24 17:12:14 -0700299 batch = tlb->active;
300 batch->pages[batch->nr++] = page;
301 if (batch->nr == batch->max) {
302 if (!tlb_next_batch(tlb))
303 return 0;
Shaohua Li0b43c3a2011-07-08 15:39:41 -0700304 batch = tlb->active;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700305 }
Sasha Levin309381fea2014-01-23 15:52:54 -0800306 VM_BUG_ON_PAGE(batch->nr > batch->max, page);
Peter Zijlstra9547d012011-05-24 17:12:14 -0700307
308 return batch->max - batch->nr;
309}
310
311#endif /* HAVE_GENERIC_MMU_GATHER */
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800312
Peter Zijlstra26723912011-05-24 17:12:00 -0700313#ifdef CONFIG_HAVE_RCU_TABLE_FREE
314
315/*
316 * See the comment near struct mmu_table_batch.
317 */
318
319static void tlb_remove_table_smp_sync(void *arg)
320{
321 /* Simply deliver the interrupt */
322}
323
324static void tlb_remove_table_one(void *table)
325{
326 /*
327 * This isn't an RCU grace period and hence the page-tables cannot be
328 * assumed to be actually RCU-freed.
329 *
330 * It is however sufficient for software page-table walkers that rely on
331 * IRQ disabling. See the comment near struct mmu_table_batch.
332 */
333 smp_call_function(tlb_remove_table_smp_sync, NULL, 1);
334 __tlb_remove_table(table);
335}
336
337static void tlb_remove_table_rcu(struct rcu_head *head)
338{
339 struct mmu_table_batch *batch;
340 int i;
341
342 batch = container_of(head, struct mmu_table_batch, rcu);
343
344 for (i = 0; i < batch->nr; i++)
345 __tlb_remove_table(batch->tables[i]);
346
347 free_page((unsigned long)batch);
348}
349
350void tlb_table_flush(struct mmu_gather *tlb)
351{
352 struct mmu_table_batch **batch = &tlb->batch;
353
354 if (*batch) {
355 call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu);
356 *batch = NULL;
357 }
358}
359
360void tlb_remove_table(struct mmu_gather *tlb, void *table)
361{
362 struct mmu_table_batch **batch = &tlb->batch;
363
Peter Zijlstra26723912011-05-24 17:12:00 -0700364 /*
365 * When there's less then two users of this mm there cannot be a
366 * concurrent page-table walk.
367 */
368 if (atomic_read(&tlb->mm->mm_users) < 2) {
369 __tlb_remove_table(table);
370 return;
371 }
372
373 if (*batch == NULL) {
374 *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN);
375 if (*batch == NULL) {
376 tlb_remove_table_one(table);
377 return;
378 }
379 (*batch)->nr = 0;
380 }
381 (*batch)->tables[(*batch)->nr++] = table;
382 if ((*batch)->nr == MAX_TABLE_BATCH)
383 tlb_table_flush(tlb);
384}
385
Peter Zijlstra9547d012011-05-24 17:12:14 -0700386#endif /* CONFIG_HAVE_RCU_TABLE_FREE */
Peter Zijlstra26723912011-05-24 17:12:00 -0700387
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 * Note: this doesn't free the actual pages themselves. That
390 * has been handled earlier when unmapping all the memory regions.
391 */
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000392static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
393 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800395 pgtable_t token = pmd_pgtable(*pmd);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700396 pmd_clear(pmd);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000397 pte_free_tlb(tlb, token, addr);
Kirill A. Shutemove1f56c82013-11-14 14:30:48 -0800398 atomic_long_dec(&tlb->mm->nr_ptes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399}
400
Hugh Dickinse0da3822005-04-19 13:29:15 -0700401static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
402 unsigned long addr, unsigned long end,
403 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404{
405 pmd_t *pmd;
406 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700407 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408
Hugh Dickinse0da3822005-04-19 13:29:15 -0700409 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 do {
412 next = pmd_addr_end(addr, end);
413 if (pmd_none_or_clear_bad(pmd))
414 continue;
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000415 free_pte_range(tlb, pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 } while (pmd++, addr = next, addr != end);
417
Hugh Dickinse0da3822005-04-19 13:29:15 -0700418 start &= PUD_MASK;
419 if (start < floor)
420 return;
421 if (ceiling) {
422 ceiling &= PUD_MASK;
423 if (!ceiling)
424 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700426 if (end - 1 > ceiling - 1)
427 return;
428
429 pmd = pmd_offset(pud, start);
430 pud_clear(pud);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000431 pmd_free_tlb(tlb, pmd, start);
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -0800432 mm_dec_nr_pmds(tlb->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433}
434
Hugh Dickinse0da3822005-04-19 13:29:15 -0700435static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
436 unsigned long addr, unsigned long end,
437 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438{
439 pud_t *pud;
440 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700441 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442
Hugh Dickinse0da3822005-04-19 13:29:15 -0700443 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 pud = pud_offset(pgd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 do {
446 next = pud_addr_end(addr, end);
447 if (pud_none_or_clear_bad(pud))
448 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700449 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 } while (pud++, addr = next, addr != end);
451
Hugh Dickinse0da3822005-04-19 13:29:15 -0700452 start &= PGDIR_MASK;
453 if (start < floor)
454 return;
455 if (ceiling) {
456 ceiling &= PGDIR_MASK;
457 if (!ceiling)
458 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700460 if (end - 1 > ceiling - 1)
461 return;
462
463 pud = pud_offset(pgd, start);
464 pgd_clear(pgd);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000465 pud_free_tlb(tlb, pud, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466}
467
468/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700469 * This function frees user-level page tables of a process.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 */
Jan Beulich42b77722008-07-23 21:27:10 -0700471void free_pgd_range(struct mmu_gather *tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700472 unsigned long addr, unsigned long end,
473 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474{
475 pgd_t *pgd;
476 unsigned long next;
477
Hugh Dickinse0da3822005-04-19 13:29:15 -0700478 /*
479 * The next few lines have given us lots of grief...
480 *
481 * Why are we testing PMD* at this top level? Because often
482 * there will be no work to do at all, and we'd prefer not to
483 * go all the way down to the bottom just to discover that.
484 *
485 * Why all these "- 1"s? Because 0 represents both the bottom
486 * of the address space and the top of it (using -1 for the
487 * top wouldn't help much: the masks would do the wrong thing).
488 * The rule is that addr 0 and floor 0 refer to the bottom of
489 * the address space, but end 0 and ceiling 0 refer to the top
490 * Comparisons need to use "end - 1" and "ceiling - 1" (though
491 * that end 0 case should be mythical).
492 *
493 * Wherever addr is brought up or ceiling brought down, we must
494 * be careful to reject "the opposite 0" before it confuses the
495 * subsequent tests. But what about where end is brought down
496 * by PMD_SIZE below? no, end can't go down to 0 there.
497 *
498 * Whereas we round start (addr) and ceiling down, by different
499 * masks at different levels, in order to test whether a table
500 * now has no other vmas using it, so can be freed, we don't
501 * bother to round floor or end up - the tests don't need that.
502 */
503
504 addr &= PMD_MASK;
505 if (addr < floor) {
506 addr += PMD_SIZE;
507 if (!addr)
508 return;
509 }
510 if (ceiling) {
511 ceiling &= PMD_MASK;
512 if (!ceiling)
513 return;
514 }
515 if (end - 1 > ceiling - 1)
516 end -= PMD_SIZE;
517 if (addr > end - 1)
518 return;
519
Jan Beulich42b77722008-07-23 21:27:10 -0700520 pgd = pgd_offset(tlb->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 do {
522 next = pgd_addr_end(addr, end);
523 if (pgd_none_or_clear_bad(pgd))
524 continue;
Jan Beulich42b77722008-07-23 21:27:10 -0700525 free_pud_range(tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700527}
528
Jan Beulich42b77722008-07-23 21:27:10 -0700529void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700530 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700531{
532 while (vma) {
533 struct vm_area_struct *next = vma->vm_next;
534 unsigned long addr = vma->vm_start;
535
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700536 /*
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000537 * Hide vma from rmap and truncate_pagecache before freeing
538 * pgtables
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700539 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800540 unlink_anon_vmas(vma);
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700541 unlink_file_vma(vma);
542
David Gibson9da61ae2006-03-22 00:08:57 -0800543 if (is_vm_hugetlb_page(vma)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700544 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700545 floor, next? next->vm_start: ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700546 } else {
547 /*
548 * Optimization: gather nearby vmas into one call down
549 */
550 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
David Gibson48669202006-03-22 00:08:58 -0800551 && !is_vm_hugetlb_page(next)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700552 vma = next;
553 next = vma->vm_next;
Rik van Riel5beb4932010-03-05 13:42:07 -0800554 unlink_anon_vmas(vma);
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700555 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700556 }
557 free_pgd_range(tlb, addr, vma->vm_end,
558 floor, next? next->vm_start: ceiling);
559 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700560 vma = next;
561 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562}
563
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800564int __pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,
565 pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566{
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -0800567 spinlock_t *ptl;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800568 pgtable_t new = pte_alloc_one(mm, address);
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800569 int wait_split_huge_page;
Hugh Dickins1bb36302005-10-29 18:16:22 -0700570 if (!new)
571 return -ENOMEM;
572
Nick Piggin362a61a2008-05-14 06:37:36 +0200573 /*
574 * Ensure all pte setup (eg. pte page lock and page clearing) are
575 * visible before the pte is made visible to other CPUs by being
576 * put into page tables.
577 *
578 * The other side of the story is the pointer chasing in the page
579 * table walking code (when walking the page table without locking;
580 * ie. most of the time). Fortunately, these data accesses consist
581 * of a chain of data-dependent loads, meaning most CPUs (alpha
582 * being the notable exception) will already guarantee loads are
583 * seen in-order. See the alpha page table accessors for the
584 * smp_read_barrier_depends() barriers in page table walking code.
585 */
586 smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
587
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -0800588 ptl = pmd_lock(mm, pmd);
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800589 wait_split_huge_page = 0;
590 if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
Kirill A. Shutemove1f56c82013-11-14 14:30:48 -0800591 atomic_long_inc(&mm->nr_ptes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 pmd_populate(mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800593 new = NULL;
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800594 } else if (unlikely(pmd_trans_splitting(*pmd)))
595 wait_split_huge_page = 1;
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -0800596 spin_unlock(ptl);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800597 if (new)
598 pte_free(mm, new);
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800599 if (wait_split_huge_page)
600 wait_split_huge_page(vma->anon_vma, pmd);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700601 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602}
603
Hugh Dickins1bb36302005-10-29 18:16:22 -0700604int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700606 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
607 if (!new)
608 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
Nick Piggin362a61a2008-05-14 06:37:36 +0200610 smp_wmb(); /* See comment in __pte_alloc */
611
Hugh Dickins1bb36302005-10-29 18:16:22 -0700612 spin_lock(&init_mm.page_table_lock);
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800613 if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
Hugh Dickins1bb36302005-10-29 18:16:22 -0700614 pmd_populate_kernel(&init_mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800615 new = NULL;
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800616 } else
617 VM_BUG_ON(pmd_trans_splitting(*pmd));
Hugh Dickins1bb36302005-10-29 18:16:22 -0700618 spin_unlock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800619 if (new)
620 pte_free_kernel(&init_mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700621 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622}
623
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800624static inline void init_rss_vec(int *rss)
Hugh Dickinsae859762005-10-29 18:16:05 -0700625{
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800626 memset(rss, 0, sizeof(int) * NR_MM_COUNTERS);
627}
628
629static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss)
630{
631 int i;
632
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800633 if (current->mm == mm)
David Rientjes05af2e12012-03-21 16:34:13 -0700634 sync_mm_rss(mm);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800635 for (i = 0; i < NR_MM_COUNTERS; i++)
636 if (rss[i])
637 add_mm_counter(mm, i, rss[i]);
Hugh Dickinsae859762005-10-29 18:16:05 -0700638}
639
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800641 * This function is called to print an error when a bad pte
642 * is found. For example, we might have a PFN-mapped pte in
643 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700644 *
645 * The calling function must still handle the error.
646 */
Hugh Dickins3dc14742009-01-06 14:40:08 -0800647static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
648 pte_t pte, struct page *page)
Nick Pigginb5810032005-10-29 18:16:12 -0700649{
Hugh Dickins3dc14742009-01-06 14:40:08 -0800650 pgd_t *pgd = pgd_offset(vma->vm_mm, addr);
651 pud_t *pud = pud_offset(pgd, addr);
652 pmd_t *pmd = pmd_offset(pud, addr);
653 struct address_space *mapping;
654 pgoff_t index;
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800655 static unsigned long resume;
656 static unsigned long nr_shown;
657 static unsigned long nr_unshown;
658
659 /*
660 * Allow a burst of 60 reports, then keep quiet for that minute;
661 * or allow a steady drip of one report per second.
662 */
663 if (nr_shown == 60) {
664 if (time_before(jiffies, resume)) {
665 nr_unshown++;
666 return;
667 }
668 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800669 printk(KERN_ALERT
670 "BUG: Bad page map: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800671 nr_unshown);
672 nr_unshown = 0;
673 }
674 nr_shown = 0;
675 }
676 if (nr_shown++ == 0)
677 resume = jiffies + 60 * HZ;
Hugh Dickins3dc14742009-01-06 14:40:08 -0800678
679 mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;
680 index = linear_page_index(vma, addr);
681
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800682 printk(KERN_ALERT
683 "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800684 current->comm,
685 (long long)pte_val(pte), (long long)pmd_val(*pmd));
Wu Fengguang718a3822010-03-10 15:20:43 -0800686 if (page)
Dave Hansenf0b791a2014-01-23 15:52:49 -0800687 dump_page(page, "bad pte");
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800688 printk(KERN_ALERT
Hugh Dickins3dc14742009-01-06 14:40:08 -0800689 "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
690 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
691 /*
692 * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y
693 */
Konstantin Khlebnikov26825822015-04-15 16:15:08 -0700694 pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n",
695 vma->vm_file,
696 vma->vm_ops ? vma->vm_ops->fault : NULL,
697 vma->vm_file ? vma->vm_file->f_op->mmap : NULL,
698 mapping ? mapping->a_ops->readpage : NULL);
Nick Pigginb5810032005-10-29 18:16:12 -0700699 dump_stack();
Rusty Russell373d4d02013-01-21 17:17:39 +1030700 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Nick Pigginb5810032005-10-29 18:16:12 -0700701}
702
703/*
Nick Piggin7e675132008-04-28 02:13:00 -0700704 * vm_normal_page -- This function gets the "struct page" associated with a pte.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800705 *
Nick Piggin7e675132008-04-28 02:13:00 -0700706 * "Special" mappings do not wish to be associated with a "struct page" (either
707 * it doesn't exist, or it exists but they don't want to touch it). In this
708 * case, NULL is returned here. "Normal" mappings do have a struct page.
Jared Hulbertb379d792008-04-28 02:12:58 -0700709 *
Nick Piggin7e675132008-04-28 02:13:00 -0700710 * There are 2 broad cases. Firstly, an architecture may define a pte_special()
711 * pte bit, in which case this function is trivial. Secondly, an architecture
712 * may not have a spare pte bit, which requires a more complicated scheme,
713 * described below.
714 *
715 * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
716 * special mapping (even if there are underlying and valid "struct pages").
717 * COWed pages of a VM_PFNMAP are always normal.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800718 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700719 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
720 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
Nick Piggin7e675132008-04-28 02:13:00 -0700721 * set, and the vm_pgoff will point to the first PFN mapped: thus every special
722 * mapping will always honor the rule
Linus Torvalds6aab3412005-11-28 14:34:23 -0800723 *
724 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
725 *
Nick Piggin7e675132008-04-28 02:13:00 -0700726 * And for normal mappings this is false.
Jared Hulbertb379d792008-04-28 02:12:58 -0700727 *
Nick Piggin7e675132008-04-28 02:13:00 -0700728 * This restricts such mappings to be a linear translation from virtual address
729 * to pfn. To get around this restriction, we allow arbitrary mappings so long
730 * as the vma is not a COW mapping; in that case, we know that all ptes are
731 * special (because none can have been COWed).
Jared Hulbertb379d792008-04-28 02:12:58 -0700732 *
733 *
Nick Piggin7e675132008-04-28 02:13:00 -0700734 * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
735 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700736 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
737 * page" backing, however the difference is that _all_ pages with a struct
738 * page (that is, those where pfn_valid is true) are refcounted and considered
739 * normal pages by the VM. The disadvantage is that pages are refcounted
740 * (which can be slower and simply not an option for some PFNMAP users). The
741 * advantage is that we don't have to follow the strict linearity rule of
742 * PFNMAP mappings in order to support COWable mappings.
743 *
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800744 */
Nick Piggin7e675132008-04-28 02:13:00 -0700745#ifdef __HAVE_ARCH_PTE_SPECIAL
746# define HAVE_PTE_SPECIAL 1
747#else
748# define HAVE_PTE_SPECIAL 0
749#endif
750struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
751 pte_t pte)
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800752{
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800753 unsigned long pfn = pte_pfn(pte);
Nick Piggin7e675132008-04-28 02:13:00 -0700754
755 if (HAVE_PTE_SPECIAL) {
Hugh Dickinsb38af472014-08-29 15:18:44 -0700756 if (likely(!pte_special(pte)))
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800757 goto check_pfn;
David Vrabel667a0a02014-12-18 14:48:15 +0000758 if (vma->vm_ops && vma->vm_ops->find_special_page)
759 return vma->vm_ops->find_special_page(vma, addr);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -0700760 if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))
761 return NULL;
Hugh Dickins62eede62009-09-21 17:03:34 -0700762 if (!is_zero_pfn(pfn))
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800763 print_bad_pte(vma, addr, pte, NULL);
Nick Piggin7e675132008-04-28 02:13:00 -0700764 return NULL;
765 }
766
767 /* !HAVE_PTE_SPECIAL case follows: */
768
Jared Hulbertb379d792008-04-28 02:12:58 -0700769 if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
770 if (vma->vm_flags & VM_MIXEDMAP) {
771 if (!pfn_valid(pfn))
772 return NULL;
773 goto out;
774 } else {
Nick Piggin7e675132008-04-28 02:13:00 -0700775 unsigned long off;
776 off = (addr - vma->vm_start) >> PAGE_SHIFT;
Jared Hulbertb379d792008-04-28 02:12:58 -0700777 if (pfn == vma->vm_pgoff + off)
778 return NULL;
779 if (!is_cow_mapping(vma->vm_flags))
780 return NULL;
781 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800782 }
783
Hugh Dickinsb38af472014-08-29 15:18:44 -0700784 if (is_zero_pfn(pfn))
785 return NULL;
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800786check_pfn:
787 if (unlikely(pfn > highest_memmap_pfn)) {
788 print_bad_pte(vma, addr, pte, NULL);
789 return NULL;
790 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800791
792 /*
Nick Piggin7e675132008-04-28 02:13:00 -0700793 * NOTE! We still have PageReserved() pages in the page tables.
Nick Piggin7e675132008-04-28 02:13:00 -0700794 * eg. VDSO mappings can cause them to exist.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800795 */
Jared Hulbertb379d792008-04-28 02:12:58 -0700796out:
Linus Torvalds6aab3412005-11-28 14:34:23 -0800797 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800798}
799
800/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 * copy one vm_area from one task to the other. Assumes the page tables
802 * already present in the new task to be cleared in the whole range
803 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 */
805
Hugh Dickins570a335b2009-12-14 17:58:46 -0800806static inline unsigned long
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700808 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700809 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810{
Nick Pigginb5810032005-10-29 18:16:12 -0700811 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 pte_t pte = *src_pte;
813 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814
815 /* pte contains position in swap or file, so copy. */
816 if (unlikely(!pte_present(pte))) {
Kirill A. Shutemov0661a332015-02-10 14:10:04 -0800817 swp_entry_t entry = pte_to_swp_entry(pte);
Christoph Lameter06972122006-06-23 02:03:35 -0700818
Kirill A. Shutemov0661a332015-02-10 14:10:04 -0800819 if (likely(!non_swap_entry(entry))) {
820 if (swap_duplicate(entry) < 0)
821 return entry.val;
Hugh Dickins570a335b2009-12-14 17:58:46 -0800822
Kirill A. Shutemov0661a332015-02-10 14:10:04 -0800823 /* make sure dst_mm is on swapoff's mmlist. */
824 if (unlikely(list_empty(&dst_mm->mmlist))) {
825 spin_lock(&mmlist_lock);
826 if (list_empty(&dst_mm->mmlist))
827 list_add(&dst_mm->mmlist,
828 &src_mm->mmlist);
829 spin_unlock(&mmlist_lock);
830 }
831 rss[MM_SWAPENTS]++;
832 } else if (is_migration_entry(entry)) {
833 page = migration_entry_to_page(entry);
Konstantin Khlebnikov9f9f1ac2012-01-20 14:34:24 -0800834
Kirill A. Shutemov0661a332015-02-10 14:10:04 -0800835 if (PageAnon(page))
836 rss[MM_ANONPAGES]++;
837 else
838 rss[MM_FILEPAGES]++;
Konstantin Khlebnikov9f9f1ac2012-01-20 14:34:24 -0800839
Kirill A. Shutemov0661a332015-02-10 14:10:04 -0800840 if (is_write_migration_entry(entry) &&
841 is_cow_mapping(vm_flags)) {
842 /*
843 * COW mappings require pages in both
844 * parent and child to be set to read.
845 */
846 make_migration_entry_read(&entry);
847 pte = swp_entry_to_pte(entry);
848 if (pte_swp_soft_dirty(*src_pte))
849 pte = pte_swp_mksoft_dirty(pte);
850 set_pte_at(src_mm, addr, src_pte, pte);
Christoph Lameter06972122006-06-23 02:03:35 -0700851 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700853 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 }
855
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 /*
857 * If it's a COW mapping, write protect it both
858 * in the parent and the child
859 */
Linus Torvalds67121172005-12-11 20:38:17 -0800860 if (is_cow_mapping(vm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 ptep_set_wrprotect(src_mm, addr, src_pte);
Zachary Amsden3dc90792006-09-30 23:29:30 -0700862 pte = pte_wrprotect(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 }
864
865 /*
866 * If it's a shared mapping, mark it clean in
867 * the child
868 */
869 if (vm_flags & VM_SHARED)
870 pte = pte_mkclean(pte);
871 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800872
873 page = vm_normal_page(vma, addr, pte);
874 if (page) {
875 get_page(page);
Hugh Dickins21333b22009-09-21 17:01:59 -0700876 page_dup_rmap(page);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800877 if (PageAnon(page))
878 rss[MM_ANONPAGES]++;
879 else
880 rss[MM_FILEPAGES]++;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800881 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700882
883out_set_pte:
884 set_pte_at(dst_mm, addr, dst_pte, pte);
Hugh Dickins570a335b2009-12-14 17:58:46 -0800885 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886}
887
Jerome Marchand21bda262014-08-06 16:06:56 -0700888static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800889 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
890 unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891{
Daisuke Nishimurac36987e2009-10-26 16:50:23 -0700892 pte_t *orig_src_pte, *orig_dst_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700894 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -0700895 int progress = 0;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800896 int rss[NR_MM_COUNTERS];
Hugh Dickins570a335b2009-12-14 17:58:46 -0800897 swp_entry_t entry = (swp_entry_t){0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898
899again:
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800900 init_rss_vec(rss);
901
Hugh Dickinsc74df322005-10-29 18:16:23 -0700902 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 if (!dst_pte)
904 return -ENOMEM;
Peter Zijlstraece0e2b2010-10-26 14:21:52 -0700905 src_pte = pte_offset_map(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700906 src_ptl = pte_lockptr(src_mm, src_pmd);
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700907 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
Daisuke Nishimurac36987e2009-10-26 16:50:23 -0700908 orig_src_pte = src_pte;
909 orig_dst_pte = dst_pte;
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700910 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 do {
913 /*
914 * We are holding two locks at this point - either of them
915 * could generate latencies in another task on another CPU.
916 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700917 if (progress >= 32) {
918 progress = 0;
919 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +0100920 spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700921 break;
922 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 if (pte_none(*src_pte)) {
924 progress++;
925 continue;
926 }
Hugh Dickins570a335b2009-12-14 17:58:46 -0800927 entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte,
928 vma, addr, rss);
929 if (entry.val)
930 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 progress += 8;
932 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700934 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -0700935 spin_unlock(src_ptl);
Peter Zijlstraece0e2b2010-10-26 14:21:52 -0700936 pte_unmap(orig_src_pte);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800937 add_mm_rss_vec(dst_mm, rss);
Daisuke Nishimurac36987e2009-10-26 16:50:23 -0700938 pte_unmap_unlock(orig_dst_pte, dst_ptl);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700939 cond_resched();
Hugh Dickins570a335b2009-12-14 17:58:46 -0800940
941 if (entry.val) {
942 if (add_swap_count_continuation(entry, GFP_KERNEL) < 0)
943 return -ENOMEM;
944 progress = 0;
945 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 if (addr != end)
947 goto again;
948 return 0;
949}
950
951static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
952 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
953 unsigned long addr, unsigned long end)
954{
955 pmd_t *src_pmd, *dst_pmd;
956 unsigned long next;
957
958 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
959 if (!dst_pmd)
960 return -ENOMEM;
961 src_pmd = pmd_offset(src_pud, addr);
962 do {
963 next = pmd_addr_end(addr, end);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800964 if (pmd_trans_huge(*src_pmd)) {
965 int err;
Andrea Arcangeli14d1a552011-01-13 15:47:15 -0800966 VM_BUG_ON(next-addr != HPAGE_PMD_SIZE);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800967 err = copy_huge_pmd(dst_mm, src_mm,
968 dst_pmd, src_pmd, addr, vma);
969 if (err == -ENOMEM)
970 return -ENOMEM;
971 if (!err)
972 continue;
973 /* fall through */
974 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 if (pmd_none_or_clear_bad(src_pmd))
976 continue;
977 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
978 vma, addr, next))
979 return -ENOMEM;
980 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
981 return 0;
982}
983
984static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
985 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
986 unsigned long addr, unsigned long end)
987{
988 pud_t *src_pud, *dst_pud;
989 unsigned long next;
990
991 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
992 if (!dst_pud)
993 return -ENOMEM;
994 src_pud = pud_offset(src_pgd, addr);
995 do {
996 next = pud_addr_end(addr, end);
997 if (pud_none_or_clear_bad(src_pud))
998 continue;
999 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
1000 vma, addr, next))
1001 return -ENOMEM;
1002 } while (dst_pud++, src_pud++, addr = next, addr != end);
1003 return 0;
1004}
1005
1006int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
1007 struct vm_area_struct *vma)
1008{
1009 pgd_t *src_pgd, *dst_pgd;
1010 unsigned long next;
1011 unsigned long addr = vma->vm_start;
1012 unsigned long end = vma->vm_end;
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001013 unsigned long mmun_start; /* For mmu_notifiers */
1014 unsigned long mmun_end; /* For mmu_notifiers */
1015 bool is_cow;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001016 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
Nick Piggind9928952005-08-28 16:49:11 +10001018 /*
1019 * Don't copy ptes where a page fault will fill them correctly.
1020 * Fork becomes much lighter when there are big shared or private
1021 * readonly mappings. The tradeoff is that copy_page_range is more
1022 * efficient than faulting.
1023 */
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08001024 if (!(vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) &&
1025 !vma->anon_vma)
1026 return 0;
Nick Piggind9928952005-08-28 16:49:11 +10001027
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 if (is_vm_hugetlb_page(vma))
1029 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
1030
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07001031 if (unlikely(vma->vm_flags & VM_PFNMAP)) {
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001032 /*
1033 * We do not free on error cases below as remove_vma
1034 * gets called on error from higher level routine
1035 */
Suresh Siddha5180da42012-10-08 16:28:29 -07001036 ret = track_pfn_copy(vma);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001037 if (ret)
1038 return ret;
1039 }
1040
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001041 /*
1042 * We need to invalidate the secondary MMU mappings only when
1043 * there could be a permission downgrade on the ptes of the
1044 * parent mm. And a permission downgrade will only happen if
1045 * is_cow_mapping() returns true.
1046 */
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001047 is_cow = is_cow_mapping(vma->vm_flags);
1048 mmun_start = addr;
1049 mmun_end = end;
1050 if (is_cow)
1051 mmu_notifier_invalidate_range_start(src_mm, mmun_start,
1052 mmun_end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001053
1054 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 dst_pgd = pgd_offset(dst_mm, addr);
1056 src_pgd = pgd_offset(src_mm, addr);
1057 do {
1058 next = pgd_addr_end(addr, end);
1059 if (pgd_none_or_clear_bad(src_pgd))
1060 continue;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001061 if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
1062 vma, addr, next))) {
1063 ret = -ENOMEM;
1064 break;
1065 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001067
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001068 if (is_cow)
1069 mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001070 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071}
1072
Robin Holt51c6f662005-11-13 16:06:42 -08001073static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -07001074 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 unsigned long addr, unsigned long end,
Peter Zijlstra97a89412011-05-24 17:12:04 -07001076 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077{
Nick Pigginb5810032005-10-29 18:16:12 -07001078 struct mm_struct *mm = tlb->mm;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001079 int force_flush = 0;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001080 int rss[NR_MM_COUNTERS];
Peter Zijlstra97a89412011-05-24 17:12:04 -07001081 spinlock_t *ptl;
Steven Rostedt5f1a1902011-06-15 15:08:23 -07001082 pte_t *start_pte;
Peter Zijlstra97a89412011-05-24 17:12:04 -07001083 pte_t *pte;
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001084 swp_entry_t entry;
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;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 }
Nick Pigginb5810032005-10-29 18:16:12 -07001111 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -07001112 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 tlb_remove_tlb_entry(tlb, pte, addr);
1114 if (unlikely(!page))
1115 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 if (PageAnon(page))
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001117 rss[MM_ANONPAGES]--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -07001118 else {
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001119 if (pte_dirty(ptent)) {
1120 force_flush = 1;
Hugh Dickins6237bcd2005-10-29 18:15:54 -07001121 set_page_dirty(page);
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001122 }
Johannes Weiner4917e5d2009-01-06 14:39:17 -08001123 if (pte_young(ptent) &&
Joe Perches64363aa2013-07-08 16:00:18 -07001124 likely(!(vma->vm_flags & VM_SEQ_READ)))
Nick Pigginbf3f3bc2009-01-06 14:38:55 -08001125 mark_page_accessed(page);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001126 rss[MM_FILEPAGES]--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -07001127 }
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001128 page_remove_rmap(page);
Hugh Dickins3dc14742009-01-06 14:40:08 -08001129 if (unlikely(page_mapcount(page) < 0))
1130 print_bad_pte(vma, addr, ptent, page);
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001131 if (unlikely(!__tlb_remove_page(tlb, page))) {
1132 force_flush = 1;
Will Deaconce9ec372014-10-28 13:16:28 -07001133 addr += PAGE_SIZE;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001134 break;
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001135 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 continue;
1137 }
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001138 /* If details->check_mapping, we leave swap entries. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 if (unlikely(details))
1140 continue;
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001141
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001142 entry = pte_to_swp_entry(ptent);
1143 if (!non_swap_entry(entry))
1144 rss[MM_SWAPENTS]--;
1145 else if (is_migration_entry(entry)) {
1146 struct page *page;
Konstantin Khlebnikov9f9f1ac2012-01-20 14:34:24 -08001147
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001148 page = migration_entry_to_page(entry);
Konstantin Khlebnikov9f9f1ac2012-01-20 14:34:24 -08001149
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001150 if (PageAnon(page))
1151 rss[MM_ANONPAGES]--;
1152 else
1153 rss[MM_FILEPAGES]--;
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001154 }
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001155 if (unlikely(!free_swap_and_cache(entry)))
1156 print_bad_pte(vma, addr, ptent, NULL);
Zachary Amsden9888a1c2006-09-30 23:29:31 -07001157 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
Peter Zijlstra97a89412011-05-24 17:12:04 -07001158 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickinsae859762005-10-29 18:16:05 -07001159
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001160 add_mm_rss_vec(mm, rss);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001161 arch_leave_lazy_mmu_mode();
Robin Holt51c6f662005-11-13 16:06:42 -08001162
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001163 /* Do the actual TLB flush before dropping ptl */
Will Deaconfb7332a2014-10-29 10:03:09 +00001164 if (force_flush)
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001165 tlb_flush_mmu_tlbonly(tlb);
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001166 pte_unmap_unlock(start_pte, ptl);
1167
1168 /*
1169 * If we forced a TLB flush (either due to running out of
1170 * batch buffers or because we needed to flush dirty TLB
1171 * entries before releasing the ptl), free the batched
1172 * memory too. Restart if we didn't do everything.
1173 */
1174 if (force_flush) {
1175 force_flush = 0;
1176 tlb_flush_mmu_free(tlb);
Linus Torvalds2b047252013-08-15 11:42:25 -07001177
1178 if (addr != end)
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001179 goto again;
1180 }
1181
Robin Holt51c6f662005-11-13 16:06:42 -08001182 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183}
1184
Robin Holt51c6f662005-11-13 16:06:42 -08001185static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -07001186 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 unsigned long addr, unsigned long end,
Peter Zijlstra97a89412011-05-24 17:12:04 -07001188 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189{
1190 pmd_t *pmd;
1191 unsigned long next;
1192
1193 pmd = pmd_offset(pud, addr);
1194 do {
1195 next = pmd_addr_end(addr, end);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001196 if (pmd_trans_huge(*pmd)) {
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001197 if (next - addr != HPAGE_PMD_SIZE) {
David Rientjese0897d72012-06-20 12:53:00 -07001198#ifdef CONFIG_DEBUG_VM
1199 if (!rwsem_is_locked(&tlb->mm->mmap_sem)) {
1200 pr_err("%s: mmap_sem is unlocked! addr=0x%lx end=0x%lx vma->vm_start=0x%lx vma->vm_end=0x%lx\n",
1201 __func__, addr, end,
1202 vma->vm_start,
1203 vma->vm_end);
1204 BUG();
1205 }
1206#endif
Kirill A. Shutemove1803772012-12-12 13:50:59 -08001207 split_huge_page_pmd(vma, addr, pmd);
Shaohua Lif21760b2012-01-12 17:19:16 -08001208 } else if (zap_huge_pmd(tlb, vma, pmd, addr))
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001209 goto next;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001210 /* fall through */
1211 }
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001212 /*
1213 * Here there can be other concurrent MADV_DONTNEED or
1214 * trans huge page faults running, and if the pmd is
1215 * none or trans huge it can change under us. This is
1216 * because MADV_DONTNEED holds the mmap_sem in read
1217 * mode.
1218 */
1219 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1220 goto next;
Peter Zijlstra97a89412011-05-24 17:12:04 -07001221 next = zap_pte_range(tlb, vma, pmd, addr, next, details);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001222next:
Peter Zijlstra97a89412011-05-24 17:12:04 -07001223 cond_resched();
1224 } while (pmd++, addr = next, addr != end);
Robin Holt51c6f662005-11-13 16:06:42 -08001225
1226 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227}
1228
Robin Holt51c6f662005-11-13 16:06:42 -08001229static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -07001230 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 unsigned long addr, unsigned long end,
Peter Zijlstra97a89412011-05-24 17:12:04 -07001232 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233{
1234 pud_t *pud;
1235 unsigned long next;
1236
1237 pud = pud_offset(pgd, addr);
1238 do {
1239 next = pud_addr_end(addr, end);
Peter Zijlstra97a89412011-05-24 17:12:04 -07001240 if (pud_none_or_clear_bad(pud))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 continue;
Peter Zijlstra97a89412011-05-24 17:12:04 -07001242 next = zap_pmd_range(tlb, vma, pud, addr, next, details);
1243 } while (pud++, addr = next, addr != end);
Robin Holt51c6f662005-11-13 16:06:42 -08001244
1245 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246}
1247
Al Viro038c7aa2012-03-05 13:25:09 -05001248static void unmap_page_range(struct mmu_gather *tlb,
1249 struct vm_area_struct *vma,
1250 unsigned long addr, unsigned long end,
1251 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252{
1253 pgd_t *pgd;
1254 unsigned long next;
1255
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001256 if (details && !details->check_mapping)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 details = NULL;
1258
1259 BUG_ON(addr >= end);
1260 tlb_start_vma(tlb, vma);
1261 pgd = pgd_offset(vma->vm_mm, addr);
1262 do {
1263 next = pgd_addr_end(addr, end);
Peter Zijlstra97a89412011-05-24 17:12:04 -07001264 if (pgd_none_or_clear_bad(pgd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 continue;
Peter Zijlstra97a89412011-05-24 17:12:04 -07001266 next = zap_pud_range(tlb, vma, pgd, addr, next, details);
1267 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 tlb_end_vma(tlb, vma);
1269}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270
Al Virof5cc4ee2012-03-05 14:14:20 -05001271
1272static void unmap_single_vma(struct mmu_gather *tlb,
1273 struct vm_area_struct *vma, unsigned long start_addr,
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001274 unsigned long end_addr,
Al Virof5cc4ee2012-03-05 14:14:20 -05001275 struct zap_details *details)
1276{
1277 unsigned long start = max(vma->vm_start, start_addr);
1278 unsigned long end;
1279
1280 if (start >= vma->vm_end)
1281 return;
1282 end = min(vma->vm_end, end_addr);
1283 if (end <= vma->vm_start)
1284 return;
1285
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +05301286 if (vma->vm_file)
1287 uprobe_munmap(vma, start, end);
1288
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07001289 if (unlikely(vma->vm_flags & VM_PFNMAP))
Suresh Siddha5180da42012-10-08 16:28:29 -07001290 untrack_pfn(vma, 0, 0);
Al Virof5cc4ee2012-03-05 14:14:20 -05001291
1292 if (start != end) {
1293 if (unlikely(is_vm_hugetlb_page(vma))) {
1294 /*
1295 * It is undesirable to test vma->vm_file as it
1296 * should be non-null for valid hugetlb area.
1297 * However, vm_file will be NULL in the error
Davidlohr Bueso7aa6b4a2014-04-07 15:37:01 -07001298 * cleanup path of mmap_region. When
Al Virof5cc4ee2012-03-05 14:14:20 -05001299 * hugetlbfs ->mmap method fails,
Davidlohr Bueso7aa6b4a2014-04-07 15:37:01 -07001300 * mmap_region() nullifies vma->vm_file
Al Virof5cc4ee2012-03-05 14:14:20 -05001301 * before calling this function to clean up.
1302 * Since no pte has actually been setup, it is
1303 * safe to do nothing in this case.
1304 */
Aneesh Kumar K.V24669e52012-07-31 16:42:03 -07001305 if (vma->vm_file) {
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08001306 i_mmap_lock_write(vma->vm_file->f_mapping);
Mel Gormand8333522012-07-31 16:46:20 -07001307 __unmap_hugepage_range_final(tlb, vma, start, end, NULL);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08001308 i_mmap_unlock_write(vma->vm_file->f_mapping);
Aneesh Kumar K.V24669e52012-07-31 16:42:03 -07001309 }
Al Virof5cc4ee2012-03-05 14:14:20 -05001310 } else
1311 unmap_page_range(tlb, vma, start, end, details);
1312 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313}
1314
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315/**
1316 * unmap_vmas - unmap a range of memory covered by a list of vma's
Randy Dunlap0164f692011-06-15 15:08:09 -07001317 * @tlb: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 * @vma: the starting vma
1319 * @start_addr: virtual address at which to start unmapping
1320 * @end_addr: virtual address at which to end unmapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 *
Hugh Dickins508034a2005-10-29 18:16:30 -07001322 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 * Only addresses between `start' and `end' will be unmapped.
1325 *
1326 * The VMA list must be sorted in ascending virtual address order.
1327 *
1328 * unmap_vmas() assumes that the caller will flush the whole unmapped address
1329 * range after unmap_vmas() returns. So the only responsibility here is to
1330 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
1331 * drops the lock and schedules.
1332 */
Al Viro6e8bb012012-03-05 13:41:15 -05001333void unmap_vmas(struct mmu_gather *tlb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 struct vm_area_struct *vma, unsigned long start_addr,
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001335 unsigned long end_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336{
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001337 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001339 mmu_notifier_invalidate_range_start(mm, start_addr, end_addr);
Al Virof5cc4ee2012-03-05 14:14:20 -05001340 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next)
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001341 unmap_single_vma(tlb, vma, start_addr, end_addr, NULL);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001342 mmu_notifier_invalidate_range_end(mm, start_addr, end_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343}
1344
1345/**
1346 * zap_page_range - remove user pages in a given range
1347 * @vma: vm_area_struct holding the applicable pages
Randy Dunlapeb4546b2012-06-20 12:53:02 -07001348 * @start: starting address of pages to zap
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 * @size: number of bytes to zap
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001350 * @details: details of shared cache invalidation
Al Virof5cc4ee2012-03-05 14:14:20 -05001351 *
1352 * Caller must protect the VMA list
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 */
Linus Torvalds7e027b12012-05-06 13:43:15 -07001354void zap_page_range(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 unsigned long size, struct zap_details *details)
1356{
1357 struct mm_struct *mm = vma->vm_mm;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001358 struct mmu_gather tlb;
Linus Torvalds7e027b12012-05-06 13:43:15 -07001359 unsigned long end = start + size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07001362 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001363 update_hiwater_rss(mm);
Linus Torvalds7e027b12012-05-06 13:43:15 -07001364 mmu_notifier_invalidate_range_start(mm, start, end);
1365 for ( ; vma && vma->vm_start < end; vma = vma->vm_next)
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001366 unmap_single_vma(&tlb, vma, start, end, details);
Linus Torvalds7e027b12012-05-06 13:43:15 -07001367 mmu_notifier_invalidate_range_end(mm, start, end);
1368 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369}
1370
Jack Steinerc627f9c2008-07-29 22:33:53 -07001371/**
Al Virof5cc4ee2012-03-05 14:14:20 -05001372 * zap_page_range_single - remove user pages in a given range
1373 * @vma: vm_area_struct holding the applicable pages
1374 * @address: starting address of pages to zap
1375 * @size: number of bytes to zap
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001376 * @details: details of shared cache invalidation
Al Virof5cc4ee2012-03-05 14:14:20 -05001377 *
1378 * The range must fit into one VMA.
1379 */
1380static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address,
1381 unsigned long size, struct zap_details *details)
1382{
1383 struct mm_struct *mm = vma->vm_mm;
1384 struct mmu_gather tlb;
1385 unsigned long end = address + size;
Al Virof5cc4ee2012-03-05 14:14:20 -05001386
1387 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07001388 tlb_gather_mmu(&tlb, mm, address, end);
Al Virof5cc4ee2012-03-05 14:14:20 -05001389 update_hiwater_rss(mm);
1390 mmu_notifier_invalidate_range_start(mm, address, end);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001391 unmap_single_vma(&tlb, vma, address, end, details);
Al Virof5cc4ee2012-03-05 14:14:20 -05001392 mmu_notifier_invalidate_range_end(mm, address, end);
1393 tlb_finish_mmu(&tlb, address, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394}
1395
Jack Steinerc627f9c2008-07-29 22:33:53 -07001396/**
1397 * zap_vma_ptes - remove ptes mapping the vma
1398 * @vma: vm_area_struct holding ptes to be zapped
1399 * @address: starting address of pages to zap
1400 * @size: number of bytes to zap
1401 *
1402 * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1403 *
1404 * The entire address range must be fully contained within the vma.
1405 *
1406 * Returns 0 if successful.
1407 */
1408int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
1409 unsigned long size)
1410{
1411 if (address < vma->vm_start || address + size > vma->vm_end ||
1412 !(vma->vm_flags & VM_PFNMAP))
1413 return -1;
Al Virof5cc4ee2012-03-05 14:14:20 -05001414 zap_page_range_single(vma, address, size, NULL);
Jack Steinerc627f9c2008-07-29 22:33:53 -07001415 return 0;
1416}
1417EXPORT_SYMBOL_GPL(zap_vma_ptes);
1418
Namhyung Kim25ca1d62010-10-26 14:21:59 -07001419pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr,
Harvey Harrison920c7a52008-02-04 22:29:26 -08001420 spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001421{
1422 pgd_t * pgd = pgd_offset(mm, addr);
1423 pud_t * pud = pud_alloc(mm, pgd, addr);
1424 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001425 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Andrea Arcangelif66055ab2011-01-13 15:46:54 -08001426 if (pmd) {
1427 VM_BUG_ON(pmd_trans_huge(*pmd));
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001428 return pte_alloc_map_lock(mm, pmd, addr, ptl);
Andrea Arcangelif66055ab2011-01-13 15:46:54 -08001429 }
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001430 }
1431 return NULL;
1432}
1433
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001435 * This is the old fallback for page remapping.
1436 *
1437 * For historical reasons, it only allows reserved pages. Only
1438 * old drivers should use this, and they needed to mark their
1439 * pages reserved for the old functions anyway.
1440 */
Nick Piggin423bad62008-04-28 02:13:01 -07001441static int insert_page(struct vm_area_struct *vma, unsigned long addr,
1442 struct page *page, pgprot_t prot)
Linus Torvalds238f58d2005-11-29 13:01:56 -08001443{
Nick Piggin423bad62008-04-28 02:13:01 -07001444 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001445 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001446 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001447 spinlock_t *ptl;
1448
Linus Torvalds238f58d2005-11-29 13:01:56 -08001449 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001450 if (PageAnon(page))
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001451 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001452 retval = -ENOMEM;
1453 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001454 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001455 if (!pte)
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001456 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001457 retval = -EBUSY;
1458 if (!pte_none(*pte))
1459 goto out_unlock;
1460
1461 /* Ok, finally just insert the thing.. */
1462 get_page(page);
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08001463 inc_mm_counter_fast(mm, MM_FILEPAGES);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001464 page_add_file_rmap(page);
1465 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1466
1467 retval = 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001468 pte_unmap_unlock(pte, ptl);
1469 return retval;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001470out_unlock:
1471 pte_unmap_unlock(pte, ptl);
1472out:
1473 return retval;
1474}
1475
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001476/**
1477 * vm_insert_page - insert single page into user vma
1478 * @vma: user vma to map to
1479 * @addr: target user address of this page
1480 * @page: source kernel page
1481 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001482 * This allows drivers to insert individual pages they've allocated
1483 * into a user vma.
1484 *
1485 * The page has to be a nice clean _individual_ kernel allocation.
1486 * If you allocate a compound page, you need to have marked it as
1487 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001488 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001489 *
1490 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1491 * took an arbitrary page protection parameter. This doesn't allow
1492 * that. Your vma protection will have to be set up correctly, which
1493 * means that if you want a shared writable mapping, you'd better
1494 * ask for a shared writable mapping!
1495 *
1496 * The page does not need to be reserved.
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001497 *
1498 * Usually this function is called from f_op->mmap() handler
1499 * under mm->mmap_sem write-lock, so it can change vma->vm_flags.
1500 * Caller must set VM_MIXEDMAP on vma if it wants to call this
1501 * function from other places, for example from page-fault handler.
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001502 */
Nick Piggin423bad62008-04-28 02:13:01 -07001503int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
1504 struct page *page)
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001505{
1506 if (addr < vma->vm_start || addr >= vma->vm_end)
1507 return -EFAULT;
1508 if (!page_count(page))
1509 return -EINVAL;
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001510 if (!(vma->vm_flags & VM_MIXEDMAP)) {
1511 BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem));
1512 BUG_ON(vma->vm_flags & VM_PFNMAP);
1513 vma->vm_flags |= VM_MIXEDMAP;
1514 }
Nick Piggin423bad62008-04-28 02:13:01 -07001515 return insert_page(vma, addr, page, vma->vm_page_prot);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001516}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001517EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001518
Nick Piggin423bad62008-04-28 02:13:01 -07001519static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1520 unsigned long pfn, pgprot_t prot)
1521{
1522 struct mm_struct *mm = vma->vm_mm;
1523 int retval;
1524 pte_t *pte, entry;
1525 spinlock_t *ptl;
1526
1527 retval = -ENOMEM;
1528 pte = get_locked_pte(mm, addr, &ptl);
1529 if (!pte)
1530 goto out;
1531 retval = -EBUSY;
1532 if (!pte_none(*pte))
1533 goto out_unlock;
1534
1535 /* Ok, finally just insert the thing.. */
1536 entry = pte_mkspecial(pfn_pte(pfn, prot));
1537 set_pte_at(mm, addr, pte, entry);
Russell King4b3073e2009-12-18 16:40:18 +00001538 update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */
Nick Piggin423bad62008-04-28 02:13:01 -07001539
1540 retval = 0;
1541out_unlock:
1542 pte_unmap_unlock(pte, ptl);
1543out:
1544 return retval;
1545}
1546
Nick Piggine0dc0d82007-02-12 00:51:36 -08001547/**
1548 * vm_insert_pfn - insert single pfn into user vma
1549 * @vma: user vma to map to
1550 * @addr: target user address of this page
1551 * @pfn: source kernel pfn
1552 *
Robert P. J. Dayc462f172012-10-08 16:33:43 -07001553 * Similar to vm_insert_page, this allows drivers to insert individual pages
Nick Piggine0dc0d82007-02-12 00:51:36 -08001554 * they've allocated into a user vma. Same comments apply.
1555 *
1556 * This function should only be called from a vm_ops->fault handler, and
1557 * in that case the handler should return NULL.
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001558 *
1559 * vma cannot be a COW mapping.
1560 *
1561 * As this is called only for pages that do not currently exist, we
1562 * do not need to flush old virtual caches or the TLB.
Nick Piggine0dc0d82007-02-12 00:51:36 -08001563 */
1564int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
Nick Piggin423bad62008-04-28 02:13:01 -07001565 unsigned long pfn)
Nick Piggine0dc0d82007-02-12 00:51:36 -08001566{
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001567 int ret;
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001568 pgprot_t pgprot = vma->vm_page_prot;
Nick Piggin7e675132008-04-28 02:13:00 -07001569 /*
1570 * Technically, architectures with pte_special can avoid all these
1571 * restrictions (same for remap_pfn_range). However we would like
1572 * consistency in testing and feature parity among all, so we should
1573 * try to keep these invariants in place for everybody.
1574 */
Jared Hulbertb379d792008-04-28 02:12:58 -07001575 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
1576 BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
1577 (VM_PFNMAP|VM_MIXEDMAP));
1578 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1579 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
Nick Piggine0dc0d82007-02-12 00:51:36 -08001580
Nick Piggin423bad62008-04-28 02:13:01 -07001581 if (addr < vma->vm_start || addr >= vma->vm_end)
1582 return -EFAULT;
Suresh Siddha5180da42012-10-08 16:28:29 -07001583 if (track_pfn_insert(vma, &pgprot, pfn))
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001584 return -EINVAL;
1585
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001586 ret = insert_pfn(vma, addr, pfn, pgprot);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001587
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001588 return ret;
Nick Piggine0dc0d82007-02-12 00:51:36 -08001589}
1590EXPORT_SYMBOL(vm_insert_pfn);
1591
Nick Piggin423bad62008-04-28 02:13:01 -07001592int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
1593 unsigned long pfn)
1594{
1595 BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
1596
1597 if (addr < vma->vm_start || addr >= vma->vm_end)
1598 return -EFAULT;
1599
1600 /*
1601 * If we don't have pte special, then we have to use the pfn_valid()
1602 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1603 * refcount the page if pfn_valid is true (hence insert_page rather
Hugh Dickins62eede62009-09-21 17:03:34 -07001604 * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP
1605 * without pte special, it would there be refcounted as a normal page.
Nick Piggin423bad62008-04-28 02:13:01 -07001606 */
1607 if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) {
1608 struct page *page;
1609
1610 page = pfn_to_page(pfn);
1611 return insert_page(vma, addr, page, vma->vm_page_prot);
1612 }
1613 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
1614}
1615EXPORT_SYMBOL(vm_insert_mixed);
1616
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001617/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 * maps a range of physical memory into the requested pages. the old
1619 * mappings are removed. any references to nonexistent pages results
1620 * in null mappings (currently treated as "copy-on-access")
1621 */
1622static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1623 unsigned long addr, unsigned long end,
1624 unsigned long pfn, pgprot_t prot)
1625{
1626 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001627 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628
Hugh Dickinsc74df322005-10-29 18:16:23 -07001629 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 if (!pte)
1631 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001632 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 do {
1634 BUG_ON(!pte_none(*pte));
Nick Piggin7e675132008-04-28 02:13:00 -07001635 set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 pfn++;
1637 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001638 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001639 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 return 0;
1641}
1642
1643static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1644 unsigned long addr, unsigned long end,
1645 unsigned long pfn, pgprot_t prot)
1646{
1647 pmd_t *pmd;
1648 unsigned long next;
1649
1650 pfn -= addr >> PAGE_SHIFT;
1651 pmd = pmd_alloc(mm, pud, addr);
1652 if (!pmd)
1653 return -ENOMEM;
Andrea Arcangelif66055ab2011-01-13 15:46:54 -08001654 VM_BUG_ON(pmd_trans_huge(*pmd));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 do {
1656 next = pmd_addr_end(addr, end);
1657 if (remap_pte_range(mm, pmd, addr, next,
1658 pfn + (addr >> PAGE_SHIFT), prot))
1659 return -ENOMEM;
1660 } while (pmd++, addr = next, addr != end);
1661 return 0;
1662}
1663
1664static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1665 unsigned long addr, unsigned long end,
1666 unsigned long pfn, pgprot_t prot)
1667{
1668 pud_t *pud;
1669 unsigned long next;
1670
1671 pfn -= addr >> PAGE_SHIFT;
1672 pud = pud_alloc(mm, pgd, addr);
1673 if (!pud)
1674 return -ENOMEM;
1675 do {
1676 next = pud_addr_end(addr, end);
1677 if (remap_pmd_range(mm, pud, addr, next,
1678 pfn + (addr >> PAGE_SHIFT), prot))
1679 return -ENOMEM;
1680 } while (pud++, addr = next, addr != end);
1681 return 0;
1682}
1683
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001684/**
1685 * remap_pfn_range - remap kernel memory to userspace
1686 * @vma: user vma to map to
1687 * @addr: target user address to start at
1688 * @pfn: physical address of kernel memory
1689 * @size: size of map area
1690 * @prot: page protection flags for this mapping
1691 *
1692 * Note: this is only safe if the mm semaphore is held when called.
1693 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1695 unsigned long pfn, unsigned long size, pgprot_t prot)
1696{
1697 pgd_t *pgd;
1698 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001699 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 struct mm_struct *mm = vma->vm_mm;
1701 int err;
1702
1703 /*
1704 * Physically remapped pages are special. Tell the
1705 * rest of the world about it:
1706 * VM_IO tells people not to look at these pages
1707 * (accesses can have side effects).
Linus Torvalds6aab3412005-11-28 14:34:23 -08001708 * VM_PFNMAP tells the core MM that the base pages are just
1709 * raw PFN mappings, and do not have a "struct page" associated
1710 * with them.
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -07001711 * VM_DONTEXPAND
1712 * Disable vma merging and expanding with mremap().
1713 * VM_DONTDUMP
1714 * Omit vma from core dump, even when VM_IO turned off.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001715 *
1716 * There's a horrible special case to handle copy-on-write
1717 * behaviour that some programs depend on. We mark the "original"
1718 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07001719 * See vm_normal_page() for details.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 */
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07001721 if (is_cow_mapping(vma->vm_flags)) {
1722 if (addr != vma->vm_start || end != vma->vm_end)
1723 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001724 vma->vm_pgoff = pfn;
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07001725 }
1726
1727 err = track_pfn_remap(vma, &prot, pfn, addr, PAGE_ALIGN(size));
1728 if (err)
venkatesh.pallipadi@intel.com3c8bb732008-12-18 11:41:27 -08001729 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001730
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -07001731 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732
1733 BUG_ON(addr >= end);
1734 pfn -= addr >> PAGE_SHIFT;
1735 pgd = pgd_offset(mm, addr);
1736 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 do {
1738 next = pgd_addr_end(addr, end);
1739 err = remap_pud_range(mm, pgd, addr, next,
1740 pfn + (addr >> PAGE_SHIFT), prot);
1741 if (err)
1742 break;
1743 } while (pgd++, addr = next, addr != end);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001744
1745 if (err)
Suresh Siddha5180da42012-10-08 16:28:29 -07001746 untrack_pfn(vma, pfn, PAGE_ALIGN(size));
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001747
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 return err;
1749}
1750EXPORT_SYMBOL(remap_pfn_range);
1751
Linus Torvaldsb4cbb192013-04-16 13:45:37 -07001752/**
1753 * vm_iomap_memory - remap memory to userspace
1754 * @vma: user vma to map to
1755 * @start: start of area
1756 * @len: size of area
1757 *
1758 * This is a simplified io_remap_pfn_range() for common driver use. The
1759 * driver just needs to give us the physical memory range to be mapped,
1760 * we'll figure out the rest from the vma information.
1761 *
1762 * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get
1763 * whatever write-combining details or similar.
1764 */
1765int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1766{
1767 unsigned long vm_len, pfn, pages;
1768
1769 /* Check that the physical memory area passed in looks valid */
1770 if (start + len < start)
1771 return -EINVAL;
1772 /*
1773 * You *really* shouldn't map things that aren't page-aligned,
1774 * but we've historically allowed it because IO memory might
1775 * just have smaller alignment.
1776 */
1777 len += start & ~PAGE_MASK;
1778 pfn = start >> PAGE_SHIFT;
1779 pages = (len + ~PAGE_MASK) >> PAGE_SHIFT;
1780 if (pfn + pages < pfn)
1781 return -EINVAL;
1782
1783 /* We start the mapping 'vm_pgoff' pages into the area */
1784 if (vma->vm_pgoff > pages)
1785 return -EINVAL;
1786 pfn += vma->vm_pgoff;
1787 pages -= vma->vm_pgoff;
1788
1789 /* Can we fit all of the mapping? */
1790 vm_len = vma->vm_end - vma->vm_start;
1791 if (vm_len >> PAGE_SHIFT > pages)
1792 return -EINVAL;
1793
1794 /* Ok, let it rip */
1795 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1796}
1797EXPORT_SYMBOL(vm_iomap_memory);
1798
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001799static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1800 unsigned long addr, unsigned long end,
1801 pte_fn_t fn, void *data)
1802{
1803 pte_t *pte;
1804 int err;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001805 pgtable_t token;
Borislav Petkov94909912007-05-06 14:49:17 -07001806 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001807
1808 pte = (mm == &init_mm) ?
1809 pte_alloc_kernel(pmd, addr) :
1810 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1811 if (!pte)
1812 return -ENOMEM;
1813
1814 BUG_ON(pmd_huge(*pmd));
1815
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08001816 arch_enter_lazy_mmu_mode();
1817
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001818 token = pmd_pgtable(*pmd);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001819
1820 do {
Daisuke Nishimurac36987e2009-10-26 16:50:23 -07001821 err = fn(pte++, token, addr, data);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001822 if (err)
1823 break;
Daisuke Nishimurac36987e2009-10-26 16:50:23 -07001824 } while (addr += PAGE_SIZE, addr != end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001825
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08001826 arch_leave_lazy_mmu_mode();
1827
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001828 if (mm != &init_mm)
1829 pte_unmap_unlock(pte-1, ptl);
1830 return err;
1831}
1832
1833static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1834 unsigned long addr, unsigned long end,
1835 pte_fn_t fn, void *data)
1836{
1837 pmd_t *pmd;
1838 unsigned long next;
1839 int err;
1840
Andi Kleenceb86872008-07-23 21:27:50 -07001841 BUG_ON(pud_huge(*pud));
1842
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001843 pmd = pmd_alloc(mm, pud, addr);
1844 if (!pmd)
1845 return -ENOMEM;
1846 do {
1847 next = pmd_addr_end(addr, end);
1848 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1849 if (err)
1850 break;
1851 } while (pmd++, addr = next, addr != end);
1852 return err;
1853}
1854
1855static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1856 unsigned long addr, unsigned long end,
1857 pte_fn_t fn, void *data)
1858{
1859 pud_t *pud;
1860 unsigned long next;
1861 int err;
1862
1863 pud = pud_alloc(mm, pgd, addr);
1864 if (!pud)
1865 return -ENOMEM;
1866 do {
1867 next = pud_addr_end(addr, end);
1868 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1869 if (err)
1870 break;
1871 } while (pud++, addr = next, addr != end);
1872 return err;
1873}
1874
1875/*
1876 * Scan a region of virtual memory, filling in page tables as necessary
1877 * and calling a provided function on each leaf page table.
1878 */
1879int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1880 unsigned long size, pte_fn_t fn, void *data)
1881{
1882 pgd_t *pgd;
1883 unsigned long next;
Jeremy Fitzhardinge57250a52010-08-09 17:19:52 -07001884 unsigned long end = addr + size;
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001885 int err;
1886
1887 BUG_ON(addr >= end);
1888 pgd = pgd_offset(mm, addr);
1889 do {
1890 next = pgd_addr_end(addr, end);
1891 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1892 if (err)
1893 break;
1894 } while (pgd++, addr = next, addr != end);
Jeremy Fitzhardinge57250a52010-08-09 17:19:52 -07001895
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001896 return err;
1897}
1898EXPORT_SYMBOL_GPL(apply_to_page_range);
1899
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900/*
Kirill A. Shutemov9b4bdd22015-02-10 14:09:51 -08001901 * handle_pte_fault chooses page fault handler according to an entry which was
1902 * read non-atomically. Before making any commitment, on those architectures
1903 * or configurations (e.g. i386 with PAE) which might give a mix of unmatched
1904 * parts, do_swap_page must check under lock before unmapping the pte and
1905 * proceeding (but do_wp_page is only called after already making such a check;
Ryota Ozakia335b2e2011-02-10 13:56:28 +09001906 * and do_anonymous_page can safely check later on).
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001907 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001908static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001909 pte_t *page_table, pte_t orig_pte)
1910{
1911 int same = 1;
1912#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1913 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001914 spinlock_t *ptl = pte_lockptr(mm, pmd);
1915 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001916 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001917 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001918 }
1919#endif
1920 pte_unmap(page_table);
1921 return same;
1922}
1923
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001924static 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 -08001925{
Dan Williams0abdd7a2014-01-21 15:48:12 -08001926 debug_dma_assert_idle(src);
1927
Linus Torvalds6aab3412005-11-28 14:34:23 -08001928 /*
1929 * If the source page was a PFN mapping, we don't have
1930 * a "struct page" for it. We do a best-effort copy by
1931 * just copying from the original user address. If that
1932 * fails, we just zero-fill it. Live with it.
1933 */
1934 if (unlikely(!src)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08001935 void *kaddr = kmap_atomic(dst);
Linus Torvalds5d2a2dbbc2005-11-29 14:07:55 -08001936 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1937
1938 /*
1939 * This really shouldn't fail, because the page is there
1940 * in the page tables. But it might just be unreadable,
1941 * in which case we just give up and fill the result with
1942 * zeroes.
1943 */
1944 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Jan Beulich3ecb01d2010-10-26 14:22:27 -07001945 clear_page(kaddr);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001946 kunmap_atomic(kaddr);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07001947 flush_dcache_page(dst);
Nick Piggin0ed361d2008-02-04 22:29:34 -08001948 } else
1949 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001950}
1951
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952/*
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07001953 * Notify the address space that the page is about to become writable so that
1954 * it can prohibit this or wait for the page to get into an appropriate state.
1955 *
1956 * We do this without the lock held, so that it can sleep if it needs to.
1957 */
1958static int do_page_mkwrite(struct vm_area_struct *vma, struct page *page,
1959 unsigned long address)
1960{
1961 struct vm_fault vmf;
1962 int ret;
1963
1964 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
1965 vmf.pgoff = page->index;
1966 vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
1967 vmf.page = page;
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08001968 vmf.cow_page = NULL;
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07001969
1970 ret = vma->vm_ops->page_mkwrite(vma, &vmf);
1971 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
1972 return ret;
1973 if (unlikely(!(ret & VM_FAULT_LOCKED))) {
1974 lock_page(page);
1975 if (!page->mapping) {
1976 unlock_page(page);
1977 return 0; /* retry */
1978 }
1979 ret |= VM_FAULT_LOCKED;
1980 } else
1981 VM_BUG_ON_PAGE(!PageLocked(page), page);
1982 return ret;
1983}
1984
1985/*
Shachar Raindel4e047f82015-04-14 15:46:25 -07001986 * Handle write page faults for pages that can be reused in the current vma
1987 *
1988 * This can happen either due to the mapping being with the VM_SHARED flag,
1989 * or due to us being the last reference standing to the page. In either
1990 * case, all we need to do here is to mark the page as writable and update
1991 * any related book-keeping.
1992 */
1993static inline int wp_page_reuse(struct mm_struct *mm,
1994 struct vm_area_struct *vma, unsigned long address,
1995 pte_t *page_table, spinlock_t *ptl, pte_t orig_pte,
1996 struct page *page, int page_mkwrite,
1997 int dirty_shared)
1998 __releases(ptl)
1999{
2000 pte_t entry;
2001 /*
2002 * Clear the pages cpupid information as the existing
2003 * information potentially belongs to a now completely
2004 * unrelated process.
2005 */
2006 if (page)
2007 page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1);
2008
2009 flush_cache_page(vma, address, pte_pfn(orig_pte));
2010 entry = pte_mkyoung(orig_pte);
2011 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2012 if (ptep_set_access_flags(vma, address, page_table, entry, 1))
2013 update_mmu_cache(vma, address, page_table);
2014 pte_unmap_unlock(page_table, ptl);
2015
2016 if (dirty_shared) {
2017 struct address_space *mapping;
2018 int dirtied;
2019
2020 if (!page_mkwrite)
2021 lock_page(page);
2022
2023 dirtied = set_page_dirty(page);
2024 VM_BUG_ON_PAGE(PageAnon(page), page);
2025 mapping = page->mapping;
2026 unlock_page(page);
2027 page_cache_release(page);
2028
2029 if ((dirtied || page_mkwrite) && mapping) {
2030 /*
2031 * Some device drivers do not set page.mapping
2032 * but still dirty their pages
2033 */
2034 balance_dirty_pages_ratelimited(mapping);
2035 }
2036
2037 if (!page_mkwrite)
2038 file_update_time(vma->vm_file);
2039 }
2040
2041 return VM_FAULT_WRITE;
2042}
2043
2044/*
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002045 * Handle the case of a page which we actually need to copy to a new page.
2046 *
2047 * Called with mmap_sem locked and the old page referenced, but
2048 * without the ptl held.
2049 *
2050 * High level logic flow:
2051 *
2052 * - Allocate a page, copy the content of the old page to the new one.
2053 * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc.
2054 * - Take the PTL. If the pte changed, bail out and release the allocated page
2055 * - If the pte is still the way we remember it, update the page table and all
2056 * relevant references. This includes dropping the reference the page-table
2057 * held to the old page, as well as updating the rmap.
2058 * - In any case, unlock the PTL and drop the reference we took to the old page.
2059 */
2060static int wp_page_copy(struct mm_struct *mm, struct vm_area_struct *vma,
2061 unsigned long address, pte_t *page_table, pmd_t *pmd,
2062 pte_t orig_pte, struct page *old_page)
2063{
2064 struct page *new_page = NULL;
2065 spinlock_t *ptl = NULL;
2066 pte_t entry;
2067 int page_copied = 0;
2068 const unsigned long mmun_start = address & PAGE_MASK; /* For mmu_notifiers */
2069 const unsigned long mmun_end = mmun_start + PAGE_SIZE; /* For mmu_notifiers */
2070 struct mem_cgroup *memcg;
2071
2072 if (unlikely(anon_vma_prepare(vma)))
2073 goto oom;
2074
2075 if (is_zero_pfn(pte_pfn(orig_pte))) {
2076 new_page = alloc_zeroed_user_highpage_movable(vma, address);
2077 if (!new_page)
2078 goto oom;
2079 } else {
2080 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
2081 if (!new_page)
2082 goto oom;
2083 cow_user_page(new_page, old_page, address, vma);
2084 }
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002085
2086 if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg))
2087 goto oom_free_new;
2088
Mel Gormaneb3c24f2015-06-24 16:57:27 -07002089 __SetPageUptodate(new_page);
2090
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002091 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
2092
2093 /*
2094 * Re-check the pte - we dropped the lock
2095 */
2096 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2097 if (likely(pte_same(*page_table, orig_pte))) {
2098 if (old_page) {
2099 if (!PageAnon(old_page)) {
2100 dec_mm_counter_fast(mm, MM_FILEPAGES);
2101 inc_mm_counter_fast(mm, MM_ANONPAGES);
2102 }
2103 } else {
2104 inc_mm_counter_fast(mm, MM_ANONPAGES);
2105 }
2106 flush_cache_page(vma, address, pte_pfn(orig_pte));
2107 entry = mk_pte(new_page, vma->vm_page_prot);
2108 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2109 /*
2110 * Clear the pte entry and flush it first, before updating the
2111 * pte with the new entry. This will avoid a race condition
2112 * seen in the presence of one thread doing SMC and another
2113 * thread doing COW.
2114 */
2115 ptep_clear_flush_notify(vma, address, page_table);
2116 page_add_new_anon_rmap(new_page, vma, address);
2117 mem_cgroup_commit_charge(new_page, memcg, false);
2118 lru_cache_add_active_or_unevictable(new_page, vma);
2119 /*
2120 * We call the notify macro here because, when using secondary
2121 * mmu page tables (such as kvm shadow page tables), we want the
2122 * new page to be mapped directly into the secondary page table.
2123 */
2124 set_pte_at_notify(mm, address, page_table, entry);
2125 update_mmu_cache(vma, address, page_table);
2126 if (old_page) {
2127 /*
2128 * Only after switching the pte to the new page may
2129 * we remove the mapcount here. Otherwise another
2130 * process may come and find the rmap count decremented
2131 * before the pte is switched to the new page, and
2132 * "reuse" the old page writing into it while our pte
2133 * here still points into it and can be read by other
2134 * threads.
2135 *
2136 * The critical issue is to order this
2137 * page_remove_rmap with the ptp_clear_flush above.
2138 * Those stores are ordered by (if nothing else,)
2139 * the barrier present in the atomic_add_negative
2140 * in page_remove_rmap.
2141 *
2142 * Then the TLB flush in ptep_clear_flush ensures that
2143 * no process can access the old page before the
2144 * decremented mapcount is visible. And the old page
2145 * cannot be reused until after the decremented
2146 * mapcount is visible. So transitively, TLBs to
2147 * old page will be flushed before it can be reused.
2148 */
2149 page_remove_rmap(old_page);
2150 }
2151
2152 /* Free the old page.. */
2153 new_page = old_page;
2154 page_copied = 1;
2155 } else {
2156 mem_cgroup_cancel_charge(new_page, memcg);
2157 }
2158
2159 if (new_page)
2160 page_cache_release(new_page);
2161
2162 pte_unmap_unlock(page_table, ptl);
2163 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
2164 if (old_page) {
2165 /*
2166 * Don't let another task, with possibly unlocked vma,
2167 * keep the mlocked page.
2168 */
2169 if (page_copied && (vma->vm_flags & VM_LOCKED)) {
2170 lock_page(old_page); /* LRU manipulation */
2171 munlock_vma_page(old_page);
2172 unlock_page(old_page);
2173 }
2174 page_cache_release(old_page);
2175 }
2176 return page_copied ? VM_FAULT_WRITE : 0;
2177oom_free_new:
2178 page_cache_release(new_page);
2179oom:
2180 if (old_page)
2181 page_cache_release(old_page);
2182 return VM_FAULT_OOM;
2183}
2184
Boaz Harroshdd906182015-04-15 16:15:11 -07002185/*
2186 * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED
2187 * mapping
2188 */
2189static int wp_pfn_shared(struct mm_struct *mm,
2190 struct vm_area_struct *vma, unsigned long address,
2191 pte_t *page_table, spinlock_t *ptl, pte_t orig_pte,
2192 pmd_t *pmd)
2193{
2194 if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) {
2195 struct vm_fault vmf = {
2196 .page = NULL,
2197 .pgoff = linear_page_index(vma, address),
2198 .virtual_address = (void __user *)(address & PAGE_MASK),
2199 .flags = FAULT_FLAG_WRITE | FAULT_FLAG_MKWRITE,
2200 };
2201 int ret;
2202
2203 pte_unmap_unlock(page_table, ptl);
2204 ret = vma->vm_ops->pfn_mkwrite(vma, &vmf);
2205 if (ret & VM_FAULT_ERROR)
2206 return ret;
2207 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2208 /*
2209 * We might have raced with another page fault while we
2210 * released the pte_offset_map_lock.
2211 */
2212 if (!pte_same(*page_table, orig_pte)) {
2213 pte_unmap_unlock(page_table, ptl);
2214 return 0;
2215 }
2216 }
2217 return wp_page_reuse(mm, vma, address, page_table, ptl, orig_pte,
2218 NULL, 0, 0);
2219}
2220
Shachar Raindel93e478d2015-04-14 15:46:35 -07002221static int wp_page_shared(struct mm_struct *mm, struct vm_area_struct *vma,
2222 unsigned long address, pte_t *page_table,
2223 pmd_t *pmd, spinlock_t *ptl, pte_t orig_pte,
2224 struct page *old_page)
2225 __releases(ptl)
2226{
2227 int page_mkwrite = 0;
2228
2229 page_cache_get(old_page);
2230
2231 /*
2232 * Only catch write-faults on shared writable pages,
2233 * read-only shared pages can get COWed by
2234 * get_user_pages(.write=1, .force=1).
2235 */
2236 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
2237 int tmp;
2238
2239 pte_unmap_unlock(page_table, ptl);
2240 tmp = do_page_mkwrite(vma, old_page, address);
2241 if (unlikely(!tmp || (tmp &
2242 (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
2243 page_cache_release(old_page);
2244 return tmp;
2245 }
2246 /*
2247 * Since we dropped the lock we need to revalidate
2248 * the PTE as someone else may have changed it. If
2249 * they did, we just return, as we can count on the
2250 * MMU to tell us if they didn't also make it writable.
2251 */
2252 page_table = pte_offset_map_lock(mm, pmd, address,
2253 &ptl);
2254 if (!pte_same(*page_table, orig_pte)) {
2255 unlock_page(old_page);
2256 pte_unmap_unlock(page_table, ptl);
2257 page_cache_release(old_page);
2258 return 0;
2259 }
2260 page_mkwrite = 1;
2261 }
2262
2263 return wp_page_reuse(mm, vma, address, page_table, ptl,
2264 orig_pte, old_page, page_mkwrite, 1);
2265}
2266
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002267/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268 * This routine handles present pages, when users try to write
2269 * to a shared page. It is done by copying the page to a new address
2270 * and decrementing the shared-page counter for the old page.
2271 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 * Note that this routine assumes that the protection checks have been
2273 * done by the caller (the low-level page fault routine in most cases).
2274 * Thus we can safely just mark it writable once we've done any necessary
2275 * COW.
2276 *
2277 * We also mark the page dirty at this point even though the page will
2278 * change only once the write actually happens. This avoids a few races,
2279 * and potentially makes it more efficient.
2280 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002281 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2282 * but allow concurrent faults), with pte both mapped and locked.
2283 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002285static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
2286 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002287 spinlock_t *ptl, pte_t orig_pte)
Namhyung Kime6219ec2010-10-26 14:22:00 -07002288 __releases(ptl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289{
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002290 struct page *old_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291
Linus Torvalds6aab3412005-11-28 14:34:23 -08002292 old_page = vm_normal_page(vma, address, orig_pte);
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002293 if (!old_page) {
2294 /*
Peter Feiner64e45502014-10-13 15:55:46 -07002295 * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a
2296 * VM_PFNMAP VMA.
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002297 *
2298 * We should not cow pages in a shared writeable mapping.
Boaz Harroshdd906182015-04-15 16:15:11 -07002299 * Just mark the pages writable and/or call ops->pfn_mkwrite.
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002300 */
2301 if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
2302 (VM_WRITE|VM_SHARED))
Boaz Harroshdd906182015-04-15 16:15:11 -07002303 return wp_pfn_shared(mm, vma, address, page_table, ptl,
2304 orig_pte, pmd);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002305
2306 pte_unmap_unlock(page_table, ptl);
2307 return wp_page_copy(mm, vma, address, page_table, pmd,
2308 orig_pte, old_page);
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002309 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002311 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002312 * Take out anonymous pages first, anonymous shared vmas are
2313 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002314 */
Hugh Dickins9a840892009-09-21 17:02:01 -07002315 if (PageAnon(old_page) && !PageKsm(old_page)) {
Hugh Dickinsab967d82009-01-06 14:39:33 -08002316 if (!trylock_page(old_page)) {
2317 page_cache_get(old_page);
2318 pte_unmap_unlock(page_table, ptl);
2319 lock_page(old_page);
2320 page_table = pte_offset_map_lock(mm, pmd, address,
2321 &ptl);
2322 if (!pte_same(*page_table, orig_pte)) {
2323 unlock_page(old_page);
Shachar Raindel28766802015-04-14 15:46:29 -07002324 pte_unmap_unlock(page_table, ptl);
2325 page_cache_release(old_page);
2326 return 0;
Hugh Dickinsab967d82009-01-06 14:39:33 -08002327 }
2328 page_cache_release(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002329 }
Michel Lespinasseb009c022011-01-13 15:46:07 -08002330 if (reuse_swap_page(old_page)) {
Rik van Rielc44b6742010-03-05 13:42:09 -08002331 /*
2332 * The page is all ours. Move it to our anon_vma so
2333 * the rmap code will not search our parent or siblings.
2334 * Protected against the rmap code by the page lock.
2335 */
2336 page_move_anon_rmap(old_page, vma, address);
Michel Lespinasseb009c022011-01-13 15:46:07 -08002337 unlock_page(old_page);
Shachar Raindel4e047f82015-04-14 15:46:25 -07002338 return wp_page_reuse(mm, vma, address, page_table, ptl,
2339 orig_pte, old_page, 0, 0);
Michel Lespinasseb009c022011-01-13 15:46:07 -08002340 }
Hugh Dickinsab967d82009-01-06 14:39:33 -08002341 unlock_page(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002342 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002343 (VM_WRITE|VM_SHARED))) {
Shachar Raindel93e478d2015-04-14 15:46:35 -07002344 return wp_page_shared(mm, vma, address, page_table, pmd,
2345 ptl, orig_pte, old_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347
2348 /*
2349 * Ok, we need to copy. Oh, well..
2350 */
Nick Pigginb5810032005-10-29 18:16:12 -07002351 page_cache_get(old_page);
Shachar Raindel28766802015-04-14 15:46:29 -07002352
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002353 pte_unmap_unlock(page_table, ptl);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002354 return wp_page_copy(mm, vma, address, page_table, pmd,
2355 orig_pte, old_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356}
2357
Peter Zijlstra97a89412011-05-24 17:12:04 -07002358static void unmap_mapping_range_vma(struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 unsigned long start_addr, unsigned long end_addr,
2360 struct zap_details *details)
2361{
Al Virof5cc4ee2012-03-05 14:14:20 -05002362 zap_page_range_single(vma, start_addr, end_addr - start_addr, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363}
2364
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07002365static inline void unmap_mapping_range_tree(struct rb_root *root,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366 struct zap_details *details)
2367{
2368 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 pgoff_t vba, vea, zba, zea;
2370
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07002371 vma_interval_tree_foreach(vma, root,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 details->first_index, details->last_index) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373
2374 vba = vma->vm_pgoff;
Libind6e93212013-07-03 15:01:26 -07002375 vea = vba + vma_pages(vma) - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
2377 zba = details->first_index;
2378 if (zba < vba)
2379 zba = vba;
2380 zea = details->last_index;
2381 if (zea > vea)
2382 zea = vea;
2383
Peter Zijlstra97a89412011-05-24 17:12:04 -07002384 unmap_mapping_range_vma(vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
2386 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
Peter Zijlstra97a89412011-05-24 17:12:04 -07002387 details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 }
2389}
2390
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391/**
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08002392 * unmap_mapping_range - unmap the portion of all mmaps in the specified
2393 * address_space corresponding to the specified page range in the underlying
2394 * file.
2395 *
Martin Waitz3d410882005-06-23 22:05:21 -07002396 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397 * @holebegin: byte in first page to unmap, relative to the start of
2398 * the underlying file. This will be rounded down to a PAGE_SIZE
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +10002399 * boundary. Note that this is different from truncate_pagecache(), which
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 * must keep the partial page. In contrast, we must get rid of
2401 * partial pages.
2402 * @holelen: size of prospective hole in bytes. This will be rounded
2403 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2404 * end of the file.
2405 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2406 * but 0 when invalidating pagecache, don't throw away private data.
2407 */
2408void unmap_mapping_range(struct address_space *mapping,
2409 loff_t const holebegin, loff_t const holelen, int even_cows)
2410{
2411 struct zap_details details;
2412 pgoff_t hba = holebegin >> PAGE_SHIFT;
2413 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2414
2415 /* Check for overflow. */
2416 if (sizeof(holelen) > sizeof(hlen)) {
2417 long long holeend =
2418 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2419 if (holeend & ~(long long)ULONG_MAX)
2420 hlen = ULONG_MAX - hba + 1;
2421 }
2422
2423 details.check_mapping = even_cows? NULL: mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 details.first_index = hba;
2425 details.last_index = hba + hlen - 1;
2426 if (details.last_index < details.first_index)
2427 details.last_index = ULONG_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428
Ross Zwisler0f90cc62015-10-15 15:28:32 -07002429
2430 /* DAX uses i_mmap_lock to serialise file truncate vs page fault */
Kirill A. Shutemov46c043e2015-09-08 14:59:42 -07002431 i_mmap_lock_write(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07002432 if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433 unmap_mapping_range_tree(&mapping->i_mmap, &details);
Kirill A. Shutemov46c043e2015-09-08 14:59:42 -07002434 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435}
2436EXPORT_SYMBOL(unmap_mapping_range);
2437
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002439 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2440 * but allow concurrent faults), and pte mapped but not yet locked.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07002441 * We return with pte unmapped and unlocked.
2442 *
2443 * We return with the mmap_sem locked or unlocked in the same cases
2444 * as does filemap_fault().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002446static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2447 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002448 unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002450 spinlock_t *ptl;
Hugh Dickins56f31802013-02-22 16:36:10 -08002451 struct page *page, *swapcache;
Johannes Weiner00501b52014-08-08 14:19:20 -07002452 struct mem_cgroup *memcg;
Hugh Dickins65500d22005-10-29 18:15:59 -07002453 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 pte_t pte;
Michel Lespinassed065bd82010-10-26 14:21:57 -07002455 int locked;
Rik van Rielad8c2ee2010-08-09 17:19:48 -07002456 int exclusive = 0;
Nick Piggin83c54072007-07-19 01:47:05 -07002457 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002459 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002460 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002461
2462 entry = pte_to_swp_entry(orig_pte);
Andi Kleend1737fd2009-09-16 11:50:06 +02002463 if (unlikely(non_swap_entry(entry))) {
2464 if (is_migration_entry(entry)) {
2465 migration_entry_wait(mm, pmd, address);
2466 } else if (is_hwpoison_entry(entry)) {
2467 ret = VM_FAULT_HWPOISON;
2468 } else {
2469 print_bad_pte(vma, address, orig_pte, NULL);
Hugh Dickinsd99be1a2009-12-14 17:59:04 -08002470 ret = VM_FAULT_SIGBUS;
Andi Kleend1737fd2009-09-16 11:50:06 +02002471 }
Christoph Lameter06972122006-06-23 02:03:35 -07002472 goto out;
2473 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002474 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475 page = lookup_swap_cache(entry);
2476 if (!page) {
Hugh Dickins02098fe2008-02-04 22:28:42 -08002477 page = swapin_readahead(entry,
2478 GFP_HIGHUSER_MOVABLE, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 if (!page) {
2480 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002481 * Back out if somebody else faulted in this pte
2482 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002484 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 if (likely(pte_same(*page_table, orig_pte)))
2486 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002487 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002488 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 }
2490
2491 /* Had to read the page from swap area: Major fault */
2492 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002493 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07002494 mem_cgroup_count_vm_event(mm, PGMAJFAULT);
Andi Kleend1737fd2009-09-16 11:50:06 +02002495 } else if (PageHWPoison(page)) {
Wu Fengguang71f72522009-12-16 12:19:58 +01002496 /*
2497 * hwpoisoned dirty swapcache pages are kept for killing
2498 * owner processes (which may be unknown at hwpoison time)
2499 */
Andi Kleend1737fd2009-09-16 11:50:06 +02002500 ret = VM_FAULT_HWPOISON;
2501 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins56f31802013-02-22 16:36:10 -08002502 swapcache = page;
Andi Kleen4779cb32009-10-14 01:51:41 +02002503 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 }
2505
Hugh Dickins56f31802013-02-22 16:36:10 -08002506 swapcache = page;
Michel Lespinassed065bd82010-10-26 14:21:57 -07002507 locked = lock_page_or_retry(page, mm, flags);
Rik van Riele709ffd2012-05-29 15:06:18 -07002508
Balbir Singh20a10222007-11-14 17:00:33 -08002509 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Michel Lespinassed065bd82010-10-26 14:21:57 -07002510 if (!locked) {
2511 ret |= VM_FAULT_RETRY;
2512 goto out_release;
2513 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002515 /*
Hugh Dickins31c4a3d2010-09-19 19:40:22 -07002516 * Make sure try_to_free_swap or reuse_swap_page or swapoff did not
2517 * release the swapcache from under us. The page pin, and pte_same
2518 * test below, are not enough to exclude that. Even if it is still
2519 * swapcache, we need to check that the page's swap has not changed.
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002520 */
Hugh Dickins31c4a3d2010-09-19 19:40:22 -07002521 if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val))
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002522 goto out_page;
2523
Hugh Dickinscbf86cf2013-02-22 16:35:08 -08002524 page = ksm_might_need_to_copy(page, vma, address);
2525 if (unlikely(!page)) {
2526 ret = VM_FAULT_OOM;
2527 page = swapcache;
Hugh Dickinscbf86cf2013-02-22 16:35:08 -08002528 goto out_page;
Hugh Dickins5ad64682009-12-14 17:59:24 -08002529 }
2530
Johannes Weiner00501b52014-08-08 14:19:20 -07002531 if (mem_cgroup_try_charge(page, mm, GFP_KERNEL, &memcg)) {
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002532 ret = VM_FAULT_OOM;
Johannes Weinerbc43f752009-04-30 15:08:08 -07002533 goto out_page;
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002534 }
2535
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002537 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002539 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002540 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002541 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002542
2543 if (unlikely(!PageUptodate(page))) {
2544 ret = VM_FAULT_SIGBUS;
2545 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 }
2547
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002548 /*
2549 * The page isn't present yet, go ahead with the fault.
2550 *
2551 * Be careful about the sequence of operations here.
2552 * To get its accounting right, reuse_swap_page() must be called
2553 * while the page is counted on swap but not yet in mapcount i.e.
2554 * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
2555 * must be called after the swap_free(), or it will never succeed.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002556 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08002558 inc_mm_counter_fast(mm, MM_ANONPAGES);
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08002559 dec_mm_counter_fast(mm, MM_SWAPENTS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 pte = mk_pte(page, vma->vm_page_prot);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002561 if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002563 flags &= ~FAULT_FLAG_WRITE;
Andrea Arcangeli9a5b4892010-08-09 17:19:49 -07002564 ret |= VM_FAULT_WRITE;
Rik van Rielad8c2ee2010-08-09 17:19:48 -07002565 exclusive = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567 flush_icache_page(vma, page);
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07002568 if (pte_swp_soft_dirty(orig_pte))
2569 pte = pte_mksoft_dirty(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 set_pte_at(mm, address, page_table, pte);
Johannes Weiner00501b52014-08-08 14:19:20 -07002571 if (page == swapcache) {
Johannes Weineraf347702013-02-22 16:32:20 -08002572 do_page_add_anon_rmap(page, vma, address, exclusive);
Johannes Weiner00501b52014-08-08 14:19:20 -07002573 mem_cgroup_commit_charge(page, memcg, true);
2574 } else { /* ksm created a completely new copy */
Hugh Dickins56f31802013-02-22 16:36:10 -08002575 page_add_new_anon_rmap(page, vma, address);
Johannes Weiner00501b52014-08-08 14:19:20 -07002576 mem_cgroup_commit_charge(page, memcg, false);
2577 lru_cache_add_active_or_unevictable(page, vma);
2578 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002580 swap_free(entry);
Nick Pigginb291f002008-10-18 20:26:44 -07002581 if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08002582 try_to_free_swap(page);
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002583 unlock_page(page);
Hugh Dickins56f31802013-02-22 16:36:10 -08002584 if (page != swapcache) {
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002585 /*
2586 * Hold the lock to avoid the swap entry to be reused
2587 * until we take the PT lock for the pte_same() check
2588 * (to avoid false positives from pte_same). For
2589 * further safety release the lock after the swap_free
2590 * so that the swap count won't change under a
2591 * parallel locked swapcache.
2592 */
2593 unlock_page(swapcache);
2594 page_cache_release(swapcache);
2595 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002596
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002597 if (flags & FAULT_FLAG_WRITE) {
Hugh Dickins61469f12008-03-04 14:29:04 -08002598 ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
2599 if (ret & VM_FAULT_ERROR)
2600 ret &= VM_FAULT_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601 goto out;
2602 }
2603
2604 /* No need to invalidate - it was non-present before */
Russell King4b3073e2009-12-18 16:40:18 +00002605 update_mmu_cache(vma, address, page_table);
Hugh Dickins65500d22005-10-29 18:15:59 -07002606unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002607 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608out:
2609 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002610out_nomap:
Johannes Weiner00501b52014-08-08 14:19:20 -07002611 mem_cgroup_cancel_charge(page, memcg);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002612 pte_unmap_unlock(page_table, ptl);
Johannes Weinerbc43f752009-04-30 15:08:08 -07002613out_page:
Kirill Korotaevb8107482005-05-16 21:53:50 -07002614 unlock_page(page);
Andi Kleen4779cb32009-10-14 01:51:41 +02002615out_release:
Kirill Korotaevb8107482005-05-16 21:53:50 -07002616 page_cache_release(page);
Hugh Dickins56f31802013-02-22 16:36:10 -08002617 if (page != swapcache) {
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002618 unlock_page(swapcache);
2619 page_cache_release(swapcache);
2620 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002621 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622}
2623
2624/*
Luck, Tony8ca3eb02010-08-24 11:44:18 -07002625 * This is like a special single-page "expand_{down|up}wards()",
2626 * except we must first make sure that 'address{-|+}PAGE_SIZE'
Linus Torvalds320b2b82010-08-12 17:54:33 -07002627 * doesn't hit another vma.
Linus Torvalds320b2b82010-08-12 17:54:33 -07002628 */
2629static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address)
2630{
2631 address &= PAGE_MASK;
2632 if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) {
Linus Torvalds0e8e50e2010-08-20 16:49:40 -07002633 struct vm_area_struct *prev = vma->vm_prev;
Linus Torvalds320b2b82010-08-12 17:54:33 -07002634
Linus Torvalds0e8e50e2010-08-20 16:49:40 -07002635 /*
2636 * Is there a mapping abutting this one below?
2637 *
2638 * That's only ok if it's the same stack mapping
2639 * that has gotten split..
2640 */
2641 if (prev && prev->vm_end == address)
2642 return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM;
2643
Linus Torvaldsfee7e492015-01-06 13:00:05 -08002644 return expand_downwards(vma, address - PAGE_SIZE);
Linus Torvalds320b2b82010-08-12 17:54:33 -07002645 }
Luck, Tony8ca3eb02010-08-24 11:44:18 -07002646 if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) {
2647 struct vm_area_struct *next = vma->vm_next;
2648
2649 /* As VM_GROWSDOWN but s/below/above/ */
2650 if (next && next->vm_start == address + PAGE_SIZE)
2651 return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM;
2652
Linus Torvaldsfee7e492015-01-06 13:00:05 -08002653 return expand_upwards(vma, address + PAGE_SIZE);
Luck, Tony8ca3eb02010-08-24 11:44:18 -07002654 }
Linus Torvalds320b2b82010-08-12 17:54:33 -07002655 return 0;
2656}
2657
2658/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002659 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2660 * but allow concurrent faults), and pte mapped but not yet locked.
2661 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002663static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2664 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002665 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666{
Johannes Weiner00501b52014-08-08 14:19:20 -07002667 struct mem_cgroup *memcg;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002668 struct page *page;
2669 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671
Linus Torvalds11ac5522010-08-14 11:44:56 -07002672 pte_unmap(page_table);
Linus Torvalds320b2b82010-08-12 17:54:33 -07002673
Kirill A. Shutemov6b7339f2015-07-06 23:18:37 +03002674 /* File mapping without ->vm_ops ? */
2675 if (vma->vm_flags & VM_SHARED)
2676 return VM_FAULT_SIGBUS;
2677
Linus Torvalds11ac5522010-08-14 11:44:56 -07002678 /* Check if we need to add a guard page to the stack */
2679 if (check_stack_guard_page(vma, address) < 0)
Linus Torvalds9c145c562015-01-29 11:15:17 -08002680 return VM_FAULT_SIGSEGV;
Linus Torvalds11ac5522010-08-14 11:44:56 -07002681
2682 /* Use the zero-page for reads */
Dominik Dingel593befa2014-10-23 12:07:44 +02002683 if (!(flags & FAULT_FLAG_WRITE) && !mm_forbids_zeropage(mm)) {
Hugh Dickins62eede62009-09-21 17:03:34 -07002684 entry = pte_mkspecial(pfn_pte(my_zero_pfn(address),
2685 vma->vm_page_prot));
Linus Torvalds11ac5522010-08-14 11:44:56 -07002686 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002687 if (!pte_none(*page_table))
2688 goto unlock;
Andrea Arcangeli6b251fc2015-09-04 15:46:20 -07002689 /* Deliver the page fault to userland, check inside PT lock */
2690 if (userfaultfd_missing(vma)) {
2691 pte_unmap_unlock(page_table, ptl);
2692 return handle_userfault(vma, address, flags,
2693 VM_UFFD_MISSING);
2694 }
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002695 goto setpte;
2696 }
2697
Nick Piggin557ed1f2007-10-16 01:24:40 -07002698 /* Allocate our own private page. */
Nick Piggin557ed1f2007-10-16 01:24:40 -07002699 if (unlikely(anon_vma_prepare(vma)))
2700 goto oom;
2701 page = alloc_zeroed_user_highpage_movable(vma, address);
2702 if (!page)
2703 goto oom;
Mel Gormaneb3c24f2015-06-24 16:57:27 -07002704
2705 if (mem_cgroup_try_charge(page, mm, GFP_KERNEL, &memcg))
2706 goto oom_free_page;
2707
Minchan Kim52f37622013-04-29 15:08:15 -07002708 /*
2709 * The memory barrier inside __SetPageUptodate makes sure that
2710 * preceeding stores to the page contents become visible before
2711 * the set_pte_at() write.
2712 */
Nick Piggin0ed361d2008-02-04 22:29:34 -08002713 __SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714
Nick Piggin557ed1f2007-10-16 01:24:40 -07002715 entry = mk_pte(page, vma->vm_page_prot);
Hugh Dickins1ac0cb52009-09-21 17:03:29 -07002716 if (vma->vm_flags & VM_WRITE)
2717 entry = pte_mkwrite(pte_mkdirty(entry));
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002718
Nick Piggin557ed1f2007-10-16 01:24:40 -07002719 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Andrea Arcangeli1c2fb7a2009-09-21 17:02:22 -07002720 if (!pte_none(*page_table))
Nick Piggin557ed1f2007-10-16 01:24:40 -07002721 goto release;
Hugh Dickins9ba69292009-09-21 17:02:20 -07002722
Andrea Arcangeli6b251fc2015-09-04 15:46:20 -07002723 /* Deliver the page fault to userland, check inside PT lock */
2724 if (userfaultfd_missing(vma)) {
2725 pte_unmap_unlock(page_table, ptl);
2726 mem_cgroup_cancel_charge(page, memcg);
2727 page_cache_release(page);
2728 return handle_userfault(vma, address, flags,
2729 VM_UFFD_MISSING);
2730 }
2731
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08002732 inc_mm_counter_fast(mm, MM_ANONPAGES);
Nick Piggin557ed1f2007-10-16 01:24:40 -07002733 page_add_new_anon_rmap(page, vma, address);
Johannes Weiner00501b52014-08-08 14:19:20 -07002734 mem_cgroup_commit_charge(page, memcg, false);
2735 lru_cache_add_active_or_unevictable(page, vma);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002736setpte:
Hugh Dickins65500d22005-10-29 18:15:59 -07002737 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738
2739 /* No need to invalidate - it was non-present before */
Russell King4b3073e2009-12-18 16:40:18 +00002740 update_mmu_cache(vma, address, page_table);
Hugh Dickins65500d22005-10-29 18:15:59 -07002741unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002742 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002743 return 0;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002744release:
Johannes Weiner00501b52014-08-08 14:19:20 -07002745 mem_cgroup_cancel_charge(page, memcg);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002746 page_cache_release(page);
2747 goto unlock;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002748oom_free_page:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002749 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002750oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751 return VM_FAULT_OOM;
2752}
2753
Paul Cassella9a95f3c2014-08-06 16:07:24 -07002754/*
2755 * The mmap_sem must have been held on entry, and may have been
2756 * released depending on flags and vma->vm_ops->fault() return value.
2757 * See filemap_fault() and __lock_page_retry().
2758 */
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07002759static int __do_fault(struct vm_area_struct *vma, unsigned long address,
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08002760 pgoff_t pgoff, unsigned int flags,
2761 struct page *cow_page, struct page **page)
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07002762{
2763 struct vm_fault vmf;
2764 int ret;
2765
2766 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2767 vmf.pgoff = pgoff;
2768 vmf.flags = flags;
2769 vmf.page = NULL;
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08002770 vmf.cow_page = cow_page;
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07002771
2772 ret = vma->vm_ops->fault(vma, &vmf);
2773 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
2774 return ret;
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08002775 if (!vmf.page)
2776 goto out;
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07002777
2778 if (unlikely(PageHWPoison(vmf.page))) {
2779 if (ret & VM_FAULT_LOCKED)
2780 unlock_page(vmf.page);
2781 page_cache_release(vmf.page);
2782 return VM_FAULT_HWPOISON;
2783 }
2784
2785 if (unlikely(!(ret & VM_FAULT_LOCKED)))
2786 lock_page(vmf.page);
2787 else
2788 VM_BUG_ON_PAGE(!PageLocked(vmf.page), vmf.page);
2789
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08002790 out:
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07002791 *page = vmf.page;
2792 return ret;
2793}
2794
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002795/**
2796 * do_set_pte - setup new PTE entry for given page and add reverse page mapping.
2797 *
2798 * @vma: virtual memory area
2799 * @address: user virtual address
2800 * @page: page to map
2801 * @pte: pointer to target page table entry
2802 * @write: true, if new entry is writable
2803 * @anon: true, if it's anonymous page
2804 *
2805 * Caller must hold page table lock relevant for @pte.
2806 *
2807 * Target users are page handler itself and implementations of
2808 * vm_ops->map_pages.
2809 */
2810void do_set_pte(struct vm_area_struct *vma, unsigned long address,
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002811 struct page *page, pte_t *pte, bool write, bool anon)
2812{
2813 pte_t entry;
2814
2815 flush_icache_page(vma, page);
2816 entry = mk_pte(page, vma->vm_page_prot);
2817 if (write)
2818 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002819 if (anon) {
2820 inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES);
2821 page_add_new_anon_rmap(page, vma, address);
2822 } else {
2823 inc_mm_counter_fast(vma->vm_mm, MM_FILEPAGES);
2824 page_add_file_rmap(page);
2825 }
2826 set_pte_at(vma->vm_mm, address, pte, entry);
2827
2828 /* no need to invalidate: a not-present page won't be cached */
2829 update_mmu_cache(vma, address, pte);
2830}
2831
Kirill A. Shutemov3a910532014-08-06 16:08:07 -07002832static unsigned long fault_around_bytes __read_mostly =
2833 rounddown_pow_of_two(65536);
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002834
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002835#ifdef CONFIG_DEBUG_FS
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002836static int fault_around_bytes_get(void *data, u64 *val)
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002837{
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002838 *val = fault_around_bytes;
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002839 return 0;
2840}
2841
Andrey Ryabininb4903d62014-07-30 16:08:35 -07002842/*
2843 * fault_around_pages() and fault_around_mask() expects fault_around_bytes
2844 * rounded down to nearest page order. It's what do_fault_around() expects to
2845 * see.
2846 */
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002847static int fault_around_bytes_set(void *data, u64 val)
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002848{
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002849 if (val / PAGE_SIZE > PTRS_PER_PTE)
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002850 return -EINVAL;
Andrey Ryabininb4903d62014-07-30 16:08:35 -07002851 if (val > PAGE_SIZE)
2852 fault_around_bytes = rounddown_pow_of_two(val);
2853 else
2854 fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002855 return 0;
2856}
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002857DEFINE_SIMPLE_ATTRIBUTE(fault_around_bytes_fops,
2858 fault_around_bytes_get, fault_around_bytes_set, "%llu\n");
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002859
2860static int __init fault_around_debugfs(void)
2861{
2862 void *ret;
2863
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002864 ret = debugfs_create_file("fault_around_bytes", 0644, NULL, NULL,
2865 &fault_around_bytes_fops);
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002866 if (!ret)
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002867 pr_warn("Failed to create fault_around_bytes in debugfs");
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002868 return 0;
2869}
2870late_initcall(fault_around_debugfs);
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002871#endif
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002872
Kirill A. Shutemov1fdb4122014-06-04 16:10:55 -07002873/*
2874 * do_fault_around() tries to map few pages around the fault address. The hope
2875 * is that the pages will be needed soon and this will lower the number of
2876 * faults to handle.
2877 *
2878 * It uses vm_ops->map_pages() to map the pages, which skips the page if it's
2879 * not ready to be mapped: not up-to-date, locked, etc.
2880 *
2881 * This function is called with the page table lock taken. In the split ptlock
2882 * case the page table lock only protects only those entries which belong to
2883 * the page table corresponding to the fault address.
2884 *
2885 * This function doesn't cross the VMA boundaries, in order to call map_pages()
2886 * only once.
2887 *
2888 * fault_around_pages() defines how many pages we'll try to map.
2889 * do_fault_around() expects it to return a power of two less than or equal to
2890 * PTRS_PER_PTE.
2891 *
2892 * The virtual address of the area that we map is naturally aligned to the
2893 * fault_around_pages() value (and therefore to page order). This way it's
2894 * easier to guarantee that we don't cross page table boundaries.
2895 */
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002896static void do_fault_around(struct vm_area_struct *vma, unsigned long address,
2897 pte_t *pte, pgoff_t pgoff, unsigned int flags)
2898{
Kirill A. Shutemovaecd6f42014-08-06 16:08:05 -07002899 unsigned long start_addr, nr_pages, mask;
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002900 pgoff_t max_pgoff;
2901 struct vm_fault vmf;
2902 int off;
2903
Jason Low4db0c3c2015-04-15 16:14:08 -07002904 nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT;
Kirill A. Shutemovaecd6f42014-08-06 16:08:05 -07002905 mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK;
2906
2907 start_addr = max(address & mask, vma->vm_start);
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002908 off = ((address - start_addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
2909 pte -= off;
2910 pgoff -= off;
2911
2912 /*
2913 * max_pgoff is either end of page table or end of vma
Kirill A. Shutemov850e9c62014-06-04 16:10:45 -07002914 * or fault_around_pages() from pgoff, depending what is nearest.
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002915 */
2916 max_pgoff = pgoff - ((start_addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) +
2917 PTRS_PER_PTE - 1;
2918 max_pgoff = min3(max_pgoff, vma_pages(vma) + vma->vm_pgoff - 1,
Kirill A. Shutemovaecd6f42014-08-06 16:08:05 -07002919 pgoff + nr_pages - 1);
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002920
2921 /* Check if it makes any sense to call ->map_pages */
2922 while (!pte_none(*pte)) {
2923 if (++pgoff > max_pgoff)
2924 return;
2925 start_addr += PAGE_SIZE;
2926 if (start_addr >= vma->vm_end)
2927 return;
2928 pte++;
2929 }
2930
2931 vmf.virtual_address = (void __user *) start_addr;
2932 vmf.pte = pte;
2933 vmf.pgoff = pgoff;
2934 vmf.max_pgoff = max_pgoff;
2935 vmf.flags = flags;
2936 vma->vm_ops->map_pages(vma, &vmf);
2937}
2938
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07002939static int do_read_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2940 unsigned long address, pmd_t *pmd,
2941 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
2942{
2943 struct page *fault_page;
2944 spinlock_t *ptl;
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002945 pte_t *pte;
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002946 int ret = 0;
2947
2948 /*
2949 * Let's call ->map_pages() first and use ->fault() as fallback
2950 * if page by the offset is not ready to be mapped (cold cache or
2951 * something).
2952 */
Kirill A. Shutemov9b4bdd22015-02-10 14:09:51 -08002953 if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) {
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002954 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
2955 do_fault_around(vma, address, pte, pgoff, flags);
2956 if (!pte_same(*pte, orig_pte))
2957 goto unlock_out;
2958 pte_unmap_unlock(pte, ptl);
2959 }
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07002960
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08002961 ret = __do_fault(vma, address, pgoff, flags, NULL, &fault_page);
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07002962 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
2963 return ret;
2964
2965 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
2966 if (unlikely(!pte_same(*pte, orig_pte))) {
2967 pte_unmap_unlock(pte, ptl);
2968 unlock_page(fault_page);
2969 page_cache_release(fault_page);
2970 return ret;
2971 }
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002972 do_set_pte(vma, address, fault_page, pte, false, false);
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07002973 unlock_page(fault_page);
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002974unlock_out:
2975 pte_unmap_unlock(pte, ptl);
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07002976 return ret;
2977}
2978
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07002979static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2980 unsigned long address, pmd_t *pmd,
2981 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
2982{
2983 struct page *fault_page, *new_page;
Johannes Weiner00501b52014-08-08 14:19:20 -07002984 struct mem_cgroup *memcg;
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07002985 spinlock_t *ptl;
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002986 pte_t *pte;
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07002987 int ret;
2988
2989 if (unlikely(anon_vma_prepare(vma)))
2990 return VM_FAULT_OOM;
2991
2992 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
2993 if (!new_page)
2994 return VM_FAULT_OOM;
2995
Johannes Weiner00501b52014-08-08 14:19:20 -07002996 if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg)) {
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07002997 page_cache_release(new_page);
2998 return VM_FAULT_OOM;
2999 }
3000
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08003001 ret = __do_fault(vma, address, pgoff, flags, new_page, &fault_page);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003002 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
3003 goto uncharge_out;
3004
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08003005 if (fault_page)
3006 copy_user_highpage(new_page, fault_page, address, vma);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003007 __SetPageUptodate(new_page);
3008
3009 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
3010 if (unlikely(!pte_same(*pte, orig_pte))) {
3011 pte_unmap_unlock(pte, ptl);
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08003012 if (fault_page) {
3013 unlock_page(fault_page);
3014 page_cache_release(fault_page);
3015 } else {
3016 /*
3017 * The fault handler has no page to lock, so it holds
Yigal Korman0df9d412015-11-16 14:09:15 +02003018 * i_mmap_lock for read to protect against truncate.
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08003019 */
Yigal Korman0df9d412015-11-16 14:09:15 +02003020 i_mmap_unlock_read(vma->vm_file->f_mapping);
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08003021 }
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003022 goto uncharge_out;
3023 }
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003024 do_set_pte(vma, address, new_page, pte, true, true);
Johannes Weiner00501b52014-08-08 14:19:20 -07003025 mem_cgroup_commit_charge(new_page, memcg, false);
3026 lru_cache_add_active_or_unevictable(new_page, vma);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003027 pte_unmap_unlock(pte, ptl);
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08003028 if (fault_page) {
3029 unlock_page(fault_page);
3030 page_cache_release(fault_page);
3031 } else {
3032 /*
3033 * The fault handler has no page to lock, so it holds
Yigal Korman0df9d412015-11-16 14:09:15 +02003034 * i_mmap_lock for read to protect against truncate.
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08003035 */
Yigal Korman0df9d412015-11-16 14:09:15 +02003036 i_mmap_unlock_read(vma->vm_file->f_mapping);
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08003037 }
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003038 return ret;
3039uncharge_out:
Johannes Weiner00501b52014-08-08 14:19:20 -07003040 mem_cgroup_cancel_charge(new_page, memcg);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003041 page_cache_release(new_page);
3042 return ret;
3043}
3044
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003045static int do_shared_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins16abfa02007-10-04 16:56:06 +01003046 unsigned long address, pmd_t *pmd,
Nick Piggin54cb8822007-07-19 01:46:59 -07003047 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048{
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003049 struct page *fault_page;
3050 struct address_space *mapping;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003051 spinlock_t *ptl;
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003052 pte_t *pte;
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003053 int dirtied = 0;
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003054 int ret, tmp;
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07003055
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08003056 ret = __do_fault(vma, address, pgoff, flags, NULL, &fault_page);
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07003057 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003058 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003059
3060 /*
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003061 * Check if the backing address space wants to know that the page is
3062 * about to become writable
Linus Torvalds1da177e2005-04-16 15:20:36 -07003063 */
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07003064 if (vma->vm_ops->page_mkwrite) {
3065 unlock_page(fault_page);
3066 tmp = do_page_mkwrite(vma, fault_page, address);
3067 if (unlikely(!tmp ||
3068 (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
3069 page_cache_release(fault_page);
3070 return tmp;
3071 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003072 }
3073
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003074 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
3075 if (unlikely(!pte_same(*pte, orig_pte))) {
3076 pte_unmap_unlock(pte, ptl);
3077 unlock_page(fault_page);
3078 page_cache_release(fault_page);
3079 return ret;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07003080 }
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003081 do_set_pte(vma, address, fault_page, pte, true, false);
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003082 pte_unmap_unlock(pte, ptl);
3083
3084 if (set_page_dirty(fault_page))
3085 dirtied = 1;
Andrew Mortond82fa872014-12-18 16:17:29 -08003086 /*
3087 * Take a local copy of the address_space - page.mapping may be zeroed
3088 * by truncate after unlock_page(). The address_space itself remains
3089 * pinned by vma->vm_file's reference. We rely on unlock_page()'s
3090 * release semantics to prevent the compiler from undoing this copying.
3091 */
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003092 mapping = fault_page->mapping;
3093 unlock_page(fault_page);
3094 if ((dirtied || vma->vm_ops->page_mkwrite) && mapping) {
3095 /*
3096 * Some device drivers do not set page.mapping but still
3097 * dirty their pages
3098 */
3099 balance_dirty_pages_ratelimited(mapping);
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07003100 }
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003101
Johannes Weiner74ec6752015-02-10 14:11:28 -08003102 if (!vma->vm_ops->page_mkwrite)
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003103 file_update_time(vma->vm_file);
3104
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07003105 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07003106}
Nick Piggind00806b2007-07-19 01:46:57 -07003107
Paul Cassella9a95f3c2014-08-06 16:07:24 -07003108/*
3109 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3110 * but allow concurrent faults).
3111 * The mmap_sem may have been released depending on flags and our
3112 * return value. See filemap_fault() and __lock_page_or_retry().
3113 */
Kirill A. Shutemov9b4bdd22015-02-10 14:09:51 -08003114static int do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Nick Piggin54cb8822007-07-19 01:46:59 -07003115 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003116 unsigned int flags, pte_t orig_pte)
Nick Piggin54cb8822007-07-19 01:46:59 -07003117{
3118 pgoff_t pgoff = (((address & PAGE_MASK)
Dean Nelson0da7e012007-10-16 01:24:45 -07003119 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
Nick Piggin54cb8822007-07-19 01:46:59 -07003120
Hugh Dickins16abfa02007-10-04 16:56:06 +01003121 pte_unmap(page_table);
Kirill A. Shutemov6b7339f2015-07-06 23:18:37 +03003122 /* The VMA was not fully populated on mmap() or missing VM_DONTEXPAND */
3123 if (!vma->vm_ops->fault)
3124 return VM_FAULT_SIGBUS;
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07003125 if (!(flags & FAULT_FLAG_WRITE))
3126 return do_read_fault(mm, vma, address, pmd, pgoff, flags,
3127 orig_pte);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003128 if (!(vma->vm_flags & VM_SHARED))
3129 return do_cow_fault(mm, vma, address, pmd, pgoff, flags,
3130 orig_pte);
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003131 return do_shared_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Nick Piggin54cb8822007-07-19 01:46:59 -07003132}
3133
Rashika Kheriab19a9932014-04-03 14:48:02 -07003134static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
Rik van Riel04bb2f92013-10-07 11:29:36 +01003135 unsigned long addr, int page_nid,
3136 int *flags)
Mel Gorman9532fec2012-11-15 01:24:32 +00003137{
3138 get_page(page);
3139
3140 count_vm_numa_event(NUMA_HINT_FAULTS);
Rik van Riel04bb2f92013-10-07 11:29:36 +01003141 if (page_nid == numa_node_id()) {
Mel Gorman9532fec2012-11-15 01:24:32 +00003142 count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL);
Rik van Riel04bb2f92013-10-07 11:29:36 +01003143 *flags |= TNF_FAULT_LOCAL;
3144 }
Mel Gorman9532fec2012-11-15 01:24:32 +00003145
3146 return mpol_misplaced(page, vma, addr);
3147}
3148
Rashika Kheriab19a9932014-04-03 14:48:02 -07003149static int do_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,
Mel Gormand10e63f2012-10-25 14:16:31 +02003150 unsigned long addr, pte_t pte, pte_t *ptep, pmd_t *pmd)
3151{
Mel Gorman4daae3b2012-11-02 11:33:45 +00003152 struct page *page = NULL;
Mel Gormand10e63f2012-10-25 14:16:31 +02003153 spinlock_t *ptl;
Mel Gorman8191acb2013-10-07 11:28:45 +01003154 int page_nid = -1;
Peter Zijlstra90572892013-10-07 11:29:20 +01003155 int last_cpupid;
Peter Zijlstracbee9f82012-10-25 14:16:43 +02003156 int target_nid;
Mel Gormanb8593bf2012-11-21 01:18:23 +00003157 bool migrated = false;
Mel Gormanb191f9b2015-03-25 15:55:40 -07003158 bool was_writable = pte_write(pte);
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003159 int flags = 0;
Mel Gormand10e63f2012-10-25 14:16:31 +02003160
Mel Gormanc0e7cad2015-02-12 14:58:41 -08003161 /* A PROT_NONE fault should not end up here */
3162 BUG_ON(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)));
3163
Mel Gormand10e63f2012-10-25 14:16:31 +02003164 /*
3165 * The "pte" at this point cannot be used safely without
3166 * validation through pte_unmap_same(). It's of NUMA type but
3167 * the pfn may be screwed if the read is non atomic.
3168 *
Mel Gorman4d942462015-02-12 14:58:28 -08003169 * We can safely just do a "set_pte_at()", because the old
3170 * page table entry is not accessible, so there would be no
3171 * concurrent hardware modifications to the PTE.
Mel Gormand10e63f2012-10-25 14:16:31 +02003172 */
3173 ptl = pte_lockptr(mm, pmd);
3174 spin_lock(ptl);
Mel Gorman4daae3b2012-11-02 11:33:45 +00003175 if (unlikely(!pte_same(*ptep, pte))) {
3176 pte_unmap_unlock(ptep, ptl);
3177 goto out;
3178 }
3179
Mel Gorman4d942462015-02-12 14:58:28 -08003180 /* Make it present again */
3181 pte = pte_modify(pte, vma->vm_page_prot);
3182 pte = pte_mkyoung(pte);
Mel Gormanb191f9b2015-03-25 15:55:40 -07003183 if (was_writable)
3184 pte = pte_mkwrite(pte);
Mel Gormand10e63f2012-10-25 14:16:31 +02003185 set_pte_at(mm, addr, ptep, pte);
3186 update_mmu_cache(vma, addr, ptep);
3187
3188 page = vm_normal_page(vma, addr, pte);
3189 if (!page) {
3190 pte_unmap_unlock(ptep, ptl);
3191 return 0;
3192 }
3193
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003194 /*
Mel Gormanbea66fb2015-03-25 15:55:37 -07003195 * Avoid grouping on RO pages in general. RO pages shouldn't hurt as
3196 * much anyway since they can be in shared cache state. This misses
3197 * the case where a mapping is writable but the process never writes
3198 * to it but pte_write gets cleared during protection updates and
3199 * pte_dirty has unpredictable behaviour between PTE scan updates,
3200 * background writeback, dirty balancing and application behaviour.
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003201 */
Mel Gormanbea66fb2015-03-25 15:55:37 -07003202 if (!(vma->vm_flags & VM_WRITE))
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003203 flags |= TNF_NO_GROUP;
3204
Rik van Rieldabe1d92013-10-07 11:29:34 +01003205 /*
3206 * Flag if the page is shared between multiple address spaces. This
3207 * is later used when determining whether to group tasks together
3208 */
3209 if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED))
3210 flags |= TNF_SHARED;
3211
Peter Zijlstra90572892013-10-07 11:29:20 +01003212 last_cpupid = page_cpupid_last(page);
Mel Gorman8191acb2013-10-07 11:28:45 +01003213 page_nid = page_to_nid(page);
Rik van Riel04bb2f92013-10-07 11:29:36 +01003214 target_nid = numa_migrate_prep(page, vma, addr, page_nid, &flags);
Mel Gormand10e63f2012-10-25 14:16:31 +02003215 pte_unmap_unlock(ptep, ptl);
Mel Gorman4daae3b2012-11-02 11:33:45 +00003216 if (target_nid == -1) {
Mel Gorman4daae3b2012-11-02 11:33:45 +00003217 put_page(page);
3218 goto out;
3219 }
3220
3221 /* Migrate to the requested node */
Mel Gorman1bc115d2013-10-07 11:29:05 +01003222 migrated = migrate_misplaced_page(page, vma, target_nid);
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003223 if (migrated) {
Mel Gorman8191acb2013-10-07 11:28:45 +01003224 page_nid = target_nid;
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003225 flags |= TNF_MIGRATED;
Mel Gorman074c2382015-03-25 15:55:42 -07003226 } else
3227 flags |= TNF_MIGRATE_FAIL;
Mel Gorman4daae3b2012-11-02 11:33:45 +00003228
3229out:
Mel Gorman8191acb2013-10-07 11:28:45 +01003230 if (page_nid != -1)
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003231 task_numa_fault(last_cpupid, page_nid, 1, flags);
Mel Gormand10e63f2012-10-25 14:16:31 +02003232 return 0;
3233}
3234
Matthew Wilcoxb96375f2015-09-08 14:58:48 -07003235static int create_huge_pmd(struct mm_struct *mm, struct vm_area_struct *vma,
3236 unsigned long address, pmd_t *pmd, unsigned int flags)
3237{
Kirill A. Shutemovfb6dd5f2015-09-09 15:39:35 -07003238 if (vma_is_anonymous(vma))
Matthew Wilcoxb96375f2015-09-08 14:58:48 -07003239 return do_huge_pmd_anonymous_page(mm, vma, address, pmd, flags);
3240 if (vma->vm_ops->pmd_fault)
3241 return vma->vm_ops->pmd_fault(vma, address, pmd, flags);
3242 return VM_FAULT_FALLBACK;
3243}
3244
3245static int wp_huge_pmd(struct mm_struct *mm, struct vm_area_struct *vma,
3246 unsigned long address, pmd_t *pmd, pmd_t orig_pmd,
3247 unsigned int flags)
3248{
Kirill A. Shutemovfb6dd5f2015-09-09 15:39:35 -07003249 if (vma_is_anonymous(vma))
Matthew Wilcoxb96375f2015-09-08 14:58:48 -07003250 return do_huge_pmd_wp_page(mm, vma, address, pmd, orig_pmd);
3251 if (vma->vm_ops->pmd_fault)
3252 return vma->vm_ops->pmd_fault(vma, address, pmd, flags);
3253 return VM_FAULT_FALLBACK;
3254}
3255
Linus Torvalds1da177e2005-04-16 15:20:36 -07003256/*
3257 * These routines also need to handle stuff like marking pages dirty
3258 * and/or accessed for architectures that don't do it in hardware (most
3259 * RISC architectures). The early dirtying is also good on the i386.
3260 *
3261 * There is also a hook called "update_mmu_cache()" that architectures
3262 * with external mmu caches can use to update those (ie the Sparc or
3263 * PowerPC hashed page tables that act as extended TLBs).
3264 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07003265 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3266 * but allow concurrent faults), and pte mapped but not yet locked.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07003267 * We return with pte unmapped and unlocked.
3268 *
3269 * The mmap_sem may have been released depending on flags and our
3270 * return value. See filemap_fault() and __lock_page_or_retry().
Linus Torvalds1da177e2005-04-16 15:20:36 -07003271 */
Kirill A. Shutemovc0292552013-09-12 15:14:05 -07003272static int handle_pte_fault(struct mm_struct *mm,
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003273 struct vm_area_struct *vma, unsigned long address,
3274 pte_t *pte, pmd_t *pmd, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003275{
3276 pte_t entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003277 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278
Christian Borntraegere37c6982014-12-07 21:41:33 +01003279 /*
3280 * some architectures can have larger ptes than wordsize,
3281 * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and CONFIG_32BIT=y,
3282 * so READ_ONCE or ACCESS_ONCE cannot guarantee atomic accesses.
3283 * The code below just needs a consistent view for the ifs and
3284 * we later double check anyway with the ptl lock held. So here
3285 * a barrier will do.
3286 */
3287 entry = *pte;
3288 barrier();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003289 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07003290 if (pte_none(entry)) {
Oleg Nesterovb5330622015-09-08 14:58:28 -07003291 if (vma_is_anonymous(vma))
3292 return do_anonymous_page(mm, vma, address,
3293 pte, pmd, flags);
3294 else
Kirill A. Shutemov6b7339f2015-07-06 23:18:37 +03003295 return do_fault(mm, vma, address, pte, pmd,
3296 flags, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07003297 }
Hugh Dickins65500d22005-10-29 18:15:59 -07003298 return do_swap_page(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003299 pte, pmd, flags, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003300 }
3301
Mel Gorman8a0516e2015-02-12 14:58:22 -08003302 if (pte_protnone(entry))
Mel Gormand10e63f2012-10-25 14:16:31 +02003303 return do_numa_page(mm, vma, address, entry, pte, pmd);
3304
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07003305 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003306 spin_lock(ptl);
3307 if (unlikely(!pte_same(*pte, entry)))
3308 goto unlock;
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003309 if (flags & FAULT_FLAG_WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003311 return do_wp_page(mm, vma, address,
3312 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003313 entry = pte_mkdirty(entry);
3314 }
3315 entry = pte_mkyoung(entry);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003316 if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) {
Russell King4b3073e2009-12-18 16:40:18 +00003317 update_mmu_cache(vma, address, pte);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07003318 } else {
3319 /*
3320 * This is needed only for protection faults but the arch code
3321 * is not yet telling us if this is a protection fault or not.
3322 * This still avoids useless tlb flushes for .text page faults
3323 * with threads.
3324 */
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003325 if (flags & FAULT_FLAG_WRITE)
Shaohua Li61c77322010-08-16 09:16:55 +08003326 flush_tlb_fix_spurious_fault(vma, address);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07003327 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003328unlock:
3329 pte_unmap_unlock(pte, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07003330 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003331}
3332
3333/*
3334 * By the time we get here, we already hold the mm semaphore
Paul Cassella9a95f3c2014-08-06 16:07:24 -07003335 *
3336 * The mmap_sem may have been released depending on flags and our
3337 * return value. See filemap_fault() and __lock_page_or_retry().
Linus Torvalds1da177e2005-04-16 15:20:36 -07003338 */
Johannes Weiner519e5242013-09-12 15:13:42 -07003339static int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3340 unsigned long address, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003341{
3342 pgd_t *pgd;
3343 pud_t *pud;
3344 pmd_t *pmd;
3345 pte_t *pte;
3346
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01003347 if (unlikely(is_vm_hugetlb_page(vma)))
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003348 return hugetlb_fault(mm, vma, address, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349
Linus Torvalds1da177e2005-04-16 15:20:36 -07003350 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003351 pud = pud_alloc(mm, pgd, address);
3352 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07003353 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003354 pmd = pmd_alloc(mm, pud, address);
3355 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07003356 return VM_FAULT_OOM;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003357 if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) {
Matthew Wilcoxb96375f2015-09-08 14:58:48 -07003358 int ret = create_huge_pmd(mm, vma, address, pmd, flags);
Kirill A. Shutemovc0292552013-09-12 15:14:05 -07003359 if (!(ret & VM_FAULT_FALLBACK))
3360 return ret;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003361 } else {
3362 pmd_t orig_pmd = *pmd;
David Rientjes1f1d06c2012-05-29 15:06:23 -07003363 int ret;
3364
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003365 barrier();
3366 if (pmd_trans_huge(orig_pmd)) {
Will Deacona1dd4502012-12-11 16:01:27 -08003367 unsigned int dirty = flags & FAULT_FLAG_WRITE;
3368
Linus Torvaldse53289c2013-01-09 08:36:54 -08003369 /*
3370 * If the pmd is splitting, return and retry the
3371 * the fault. Alternative: wait until the split
3372 * is done, and goto retry.
3373 */
3374 if (pmd_trans_splitting(orig_pmd))
3375 return 0;
3376
Mel Gorman8a0516e2015-02-12 14:58:22 -08003377 if (pmd_protnone(orig_pmd))
Mel Gorman4daae3b2012-11-02 11:33:45 +00003378 return do_huge_pmd_numa_page(mm, vma, address,
Mel Gormand10e63f2012-10-25 14:16:31 +02003379 orig_pmd, pmd);
3380
Linus Torvalds3d59eeb2012-12-16 14:33:25 -08003381 if (dirty && !pmd_write(orig_pmd)) {
Matthew Wilcoxb96375f2015-09-08 14:58:48 -07003382 ret = wp_huge_pmd(mm, vma, address, pmd,
3383 orig_pmd, flags);
Kirill A. Shutemov9845cbb2014-02-25 15:01:42 -08003384 if (!(ret & VM_FAULT_FALLBACK))
3385 return ret;
Will Deacona1dd4502012-12-11 16:01:27 -08003386 } else {
3387 huge_pmd_set_accessed(mm, vma, address, pmd,
3388 orig_pmd, dirty);
Kirill A. Shutemov9845cbb2014-02-25 15:01:42 -08003389 return 0;
David Rientjes1f1d06c2012-05-29 15:06:23 -07003390 }
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003391 }
3392 }
3393
3394 /*
3395 * Use __pte_alloc instead of pte_alloc_map, because we can't
3396 * run pte_offset_map on the pmd, if an huge pmd could
3397 * materialize from under us from a different thread.
3398 */
Mel Gorman4fd01772011-10-12 21:06:51 +02003399 if (unlikely(pmd_none(*pmd)) &&
3400 unlikely(__pte_alloc(mm, vma, pmd, address)))
Hugh Dickinsc74df322005-10-29 18:16:23 -07003401 return VM_FAULT_OOM;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003402 /* if an huge pmd materialized from under us just retry later */
3403 if (unlikely(pmd_trans_huge(*pmd)))
3404 return 0;
3405 /*
3406 * A regular pmd is established and it can't morph into a huge pmd
3407 * from under us anymore at this point because we hold the mmap_sem
3408 * read mode and khugepaged takes it in write mode. So now it's
3409 * safe to run pte_offset_map().
3410 */
3411 pte = pte_offset_map(pmd, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003412
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003413 return handle_pte_fault(mm, vma, address, pte, pmd, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003414}
3415
Paul Cassella9a95f3c2014-08-06 16:07:24 -07003416/*
3417 * By the time we get here, we already hold the mm semaphore
3418 *
3419 * The mmap_sem may have been released depending on flags and our
3420 * return value. See filemap_fault() and __lock_page_or_retry().
3421 */
Johannes Weiner519e5242013-09-12 15:13:42 -07003422int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3423 unsigned long address, unsigned int flags)
3424{
3425 int ret;
3426
3427 __set_current_state(TASK_RUNNING);
3428
3429 count_vm_event(PGFAULT);
3430 mem_cgroup_count_vm_event(mm, PGFAULT);
3431
3432 /* do counter updates before entering really critical section. */
3433 check_sync_rss_stat(current);
3434
3435 /*
3436 * Enable the memcg OOM handling for faults triggered in user
3437 * space. Kernel faults are handled more gracefully.
3438 */
3439 if (flags & FAULT_FLAG_USER)
Johannes Weiner49426422013-10-16 13:46:59 -07003440 mem_cgroup_oom_enable();
Johannes Weiner519e5242013-09-12 15:13:42 -07003441
3442 ret = __handle_mm_fault(mm, vma, address, flags);
3443
Johannes Weiner49426422013-10-16 13:46:59 -07003444 if (flags & FAULT_FLAG_USER) {
3445 mem_cgroup_oom_disable();
3446 /*
3447 * The task may have entered a memcg OOM situation but
3448 * if the allocation error was handled gracefully (no
3449 * VM_FAULT_OOM), there is no need to kill anything.
3450 * Just clean up the OOM state peacefully.
3451 */
3452 if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM))
3453 mem_cgroup_oom_synchronize(false);
3454 }
Johannes Weiner3812c8c2013-09-12 15:13:44 -07003455
Johannes Weiner519e5242013-09-12 15:13:42 -07003456 return ret;
3457}
Jesse Barnese1d6d012014-12-12 16:55:27 -08003458EXPORT_SYMBOL_GPL(handle_mm_fault);
Johannes Weiner519e5242013-09-12 15:13:42 -07003459
Linus Torvalds1da177e2005-04-16 15:20:36 -07003460#ifndef __PAGETABLE_PUD_FOLDED
3461/*
3462 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07003463 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003464 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07003465int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003466{
Hugh Dickinsc74df322005-10-29 18:16:23 -07003467 pud_t *new = pud_alloc_one(mm, address);
3468 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07003469 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003470
Nick Piggin362a61a2008-05-14 06:37:36 +02003471 smp_wmb(); /* See comment in __pte_alloc */
3472
Hugh Dickins872fec12005-10-29 18:16:21 -07003473 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003474 if (pgd_present(*pgd)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003475 pud_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003476 else
3477 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07003478 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003479 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003480}
3481#endif /* __PAGETABLE_PUD_FOLDED */
3482
3483#ifndef __PAGETABLE_PMD_FOLDED
3484/*
3485 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07003486 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003487 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07003488int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003489{
Hugh Dickinsc74df322005-10-29 18:16:23 -07003490 pmd_t *new = pmd_alloc_one(mm, address);
3491 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07003492 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003493
Nick Piggin362a61a2008-05-14 06:37:36 +02003494 smp_wmb(); /* See comment in __pte_alloc */
3495
Hugh Dickins872fec12005-10-29 18:16:21 -07003496 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003497#ifndef __ARCH_HAS_4LEVEL_HACK
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -08003498 if (!pud_present(*pud)) {
3499 mm_inc_nr_pmds(mm);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003500 pud_populate(mm, pud, new);
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -08003501 } else /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003502 pmd_free(mm, new);
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -08003503#else
3504 if (!pgd_present(*pud)) {
3505 mm_inc_nr_pmds(mm);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003506 pgd_populate(mm, pud, new);
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -08003507 } else /* Another has populated it */
3508 pmd_free(mm, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003509#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07003510 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003511 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003512}
3513#endif /* __PAGETABLE_PMD_FOLDED */
3514
Namhyung Kim1b36ba82010-10-26 14:22:00 -07003515static int __follow_pte(struct mm_struct *mm, unsigned long address,
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07003516 pte_t **ptepp, spinlock_t **ptlp)
3517{
3518 pgd_t *pgd;
3519 pud_t *pud;
3520 pmd_t *pmd;
3521 pte_t *ptep;
3522
3523 pgd = pgd_offset(mm, address);
3524 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
3525 goto out;
3526
3527 pud = pud_offset(pgd, address);
3528 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
3529 goto out;
3530
3531 pmd = pmd_offset(pud, address);
Andrea Arcangelif66055ab2011-01-13 15:46:54 -08003532 VM_BUG_ON(pmd_trans_huge(*pmd));
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07003533 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
3534 goto out;
3535
3536 /* We cannot handle huge page PFN maps. Luckily they don't exist. */
3537 if (pmd_huge(*pmd))
3538 goto out;
3539
3540 ptep = pte_offset_map_lock(mm, pmd, address, ptlp);
3541 if (!ptep)
3542 goto out;
3543 if (!pte_present(*ptep))
3544 goto unlock;
3545 *ptepp = ptep;
3546 return 0;
3547unlock:
3548 pte_unmap_unlock(ptep, *ptlp);
3549out:
3550 return -EINVAL;
3551}
3552
Namhyung Kim1b36ba82010-10-26 14:22:00 -07003553static inline int follow_pte(struct mm_struct *mm, unsigned long address,
3554 pte_t **ptepp, spinlock_t **ptlp)
3555{
3556 int res;
3557
3558 /* (void) is needed to make gcc happy */
3559 (void) __cond_lock(*ptlp,
3560 !(res = __follow_pte(mm, address, ptepp, ptlp)));
3561 return res;
3562}
3563
Johannes Weiner3b6748e2009-06-16 15:32:35 -07003564/**
3565 * follow_pfn - look up PFN at a user virtual address
3566 * @vma: memory mapping
3567 * @address: user virtual address
3568 * @pfn: location to store found PFN
3569 *
3570 * Only IO mappings and raw PFN mappings are allowed.
3571 *
3572 * Returns zero and the pfn at @pfn on success, -ve otherwise.
3573 */
3574int follow_pfn(struct vm_area_struct *vma, unsigned long address,
3575 unsigned long *pfn)
3576{
3577 int ret = -EINVAL;
3578 spinlock_t *ptl;
3579 pte_t *ptep;
3580
3581 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3582 return ret;
3583
3584 ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);
3585 if (ret)
3586 return ret;
3587 *pfn = pte_pfn(*ptep);
3588 pte_unmap_unlock(ptep, ptl);
3589 return 0;
3590}
3591EXPORT_SYMBOL(follow_pfn);
3592
Rik van Riel28b2ee22008-07-23 21:27:05 -07003593#ifdef CONFIG_HAVE_IOREMAP_PROT
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003594int follow_phys(struct vm_area_struct *vma,
3595 unsigned long address, unsigned int flags,
3596 unsigned long *prot, resource_size_t *phys)
Rik van Riel28b2ee22008-07-23 21:27:05 -07003597{
Johannes Weiner03668a42009-06-16 15:32:34 -07003598 int ret = -EINVAL;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003599 pte_t *ptep, pte;
3600 spinlock_t *ptl;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003601
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003602 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3603 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003604
Johannes Weiner03668a42009-06-16 15:32:34 -07003605 if (follow_pte(vma->vm_mm, address, &ptep, &ptl))
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003606 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003607 pte = *ptep;
Johannes Weiner03668a42009-06-16 15:32:34 -07003608
Rik van Riel28b2ee22008-07-23 21:27:05 -07003609 if ((flags & FOLL_WRITE) && !pte_write(pte))
3610 goto unlock;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003611
3612 *prot = pgprot_val(pte_pgprot(pte));
Johannes Weiner03668a42009-06-16 15:32:34 -07003613 *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003614
Johannes Weiner03668a42009-06-16 15:32:34 -07003615 ret = 0;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003616unlock:
3617 pte_unmap_unlock(ptep, ptl);
3618out:
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003619 return ret;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003620}
3621
3622int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
3623 void *buf, int len, int write)
3624{
3625 resource_size_t phys_addr;
3626 unsigned long prot = 0;
KOSAKI Motohiro2bc72732009-01-06 14:39:43 -08003627 void __iomem *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003628 int offset = addr & (PAGE_SIZE-1);
3629
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003630 if (follow_phys(vma, addr, write, &prot, &phys_addr))
Rik van Riel28b2ee22008-07-23 21:27:05 -07003631 return -EINVAL;
3632
Grazvydas Ignotas9cb12d72015-02-12 15:00:19 -08003633 maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot);
Rik van Riel28b2ee22008-07-23 21:27:05 -07003634 if (write)
3635 memcpy_toio(maddr + offset, buf, len);
3636 else
3637 memcpy_fromio(buf, maddr + offset, len);
3638 iounmap(maddr);
3639
3640 return len;
3641}
Uwe Kleine-König5a736332013-08-07 13:02:52 +02003642EXPORT_SYMBOL_GPL(generic_access_phys);
Rik van Riel28b2ee22008-07-23 21:27:05 -07003643#endif
3644
David Howells0ec76a12006-09-27 01:50:15 -07003645/*
Stephen Wilson206cb632011-03-13 15:49:19 -04003646 * Access another process' address space as given in mm. If non-NULL, use the
3647 * given task for page fault accounting.
David Howells0ec76a12006-09-27 01:50:15 -07003648 */
Stephen Wilson206cb632011-03-13 15:49:19 -04003649static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
3650 unsigned long addr, void *buf, int len, int write)
David Howells0ec76a12006-09-27 01:50:15 -07003651{
David Howells0ec76a12006-09-27 01:50:15 -07003652 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07003653 void *old_buf = buf;
3654
David Howells0ec76a12006-09-27 01:50:15 -07003655 down_read(&mm->mmap_sem);
Simon Arlott183ff222007-10-20 01:27:18 +02003656 /* ignore errors, just check how much was successfully transferred */
David Howells0ec76a12006-09-27 01:50:15 -07003657 while (len) {
3658 int bytes, ret, offset;
3659 void *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003660 struct page *page = NULL;
David Howells0ec76a12006-09-27 01:50:15 -07003661
3662 ret = get_user_pages(tsk, mm, addr, 1,
3663 write, 1, &page, &vma);
Rik van Riel28b2ee22008-07-23 21:27:05 -07003664 if (ret <= 0) {
Rik van Rieldbffcd02014-08-06 16:08:12 -07003665#ifndef CONFIG_HAVE_IOREMAP_PROT
3666 break;
3667#else
Rik van Riel28b2ee22008-07-23 21:27:05 -07003668 /*
3669 * Check if this is a VM_IO | VM_PFNMAP VMA, which
3670 * we can access using slightly different code.
3671 */
Rik van Riel28b2ee22008-07-23 21:27:05 -07003672 vma = find_vma(mm, addr);
Michael Ellermanfe936df2011-04-14 15:22:10 -07003673 if (!vma || vma->vm_start > addr)
Rik van Riel28b2ee22008-07-23 21:27:05 -07003674 break;
3675 if (vma->vm_ops && vma->vm_ops->access)
3676 ret = vma->vm_ops->access(vma, addr, buf,
3677 len, write);
3678 if (ret <= 0)
Rik van Riel28b2ee22008-07-23 21:27:05 -07003679 break;
3680 bytes = ret;
Rik van Rieldbffcd02014-08-06 16:08:12 -07003681#endif
David Howells0ec76a12006-09-27 01:50:15 -07003682 } else {
Rik van Riel28b2ee22008-07-23 21:27:05 -07003683 bytes = len;
3684 offset = addr & (PAGE_SIZE-1);
3685 if (bytes > PAGE_SIZE-offset)
3686 bytes = PAGE_SIZE-offset;
3687
3688 maddr = kmap(page);
3689 if (write) {
3690 copy_to_user_page(vma, page, addr,
3691 maddr + offset, buf, bytes);
3692 set_page_dirty_lock(page);
3693 } else {
3694 copy_from_user_page(vma, page, addr,
3695 buf, maddr + offset, bytes);
3696 }
3697 kunmap(page);
3698 page_cache_release(page);
David Howells0ec76a12006-09-27 01:50:15 -07003699 }
David Howells0ec76a12006-09-27 01:50:15 -07003700 len -= bytes;
3701 buf += bytes;
3702 addr += bytes;
3703 }
3704 up_read(&mm->mmap_sem);
David Howells0ec76a12006-09-27 01:50:15 -07003705
3706 return buf - old_buf;
3707}
Andi Kleen03252912008-01-30 13:33:18 +01003708
Stephen Wilson5ddd36b2011-03-13 15:49:20 -04003709/**
Randy Dunlapae91dbf2011-03-26 13:27:01 -07003710 * access_remote_vm - access another process' address space
Stephen Wilson5ddd36b2011-03-13 15:49:20 -04003711 * @mm: the mm_struct of the target address space
3712 * @addr: start address to access
3713 * @buf: source or destination buffer
3714 * @len: number of bytes to transfer
3715 * @write: whether the access is a write
3716 *
3717 * The caller must hold a reference on @mm.
3718 */
3719int access_remote_vm(struct mm_struct *mm, unsigned long addr,
3720 void *buf, int len, int write)
3721{
3722 return __access_remote_vm(NULL, mm, addr, buf, len, write);
3723}
3724
Andi Kleen03252912008-01-30 13:33:18 +01003725/*
Stephen Wilson206cb632011-03-13 15:49:19 -04003726 * Access another process' address space.
3727 * Source/target buffer must be kernel space,
3728 * Do not walk the page table directly, use get_user_pages
3729 */
3730int access_process_vm(struct task_struct *tsk, unsigned long addr,
3731 void *buf, int len, int write)
3732{
3733 struct mm_struct *mm;
3734 int ret;
3735
3736 mm = get_task_mm(tsk);
3737 if (!mm)
3738 return 0;
3739
3740 ret = __access_remote_vm(tsk, mm, addr, buf, len, write);
3741 mmput(mm);
3742
3743 return ret;
3744}
3745
Andi Kleen03252912008-01-30 13:33:18 +01003746/*
3747 * Print the name of a VMA.
3748 */
3749void print_vma_addr(char *prefix, unsigned long ip)
3750{
3751 struct mm_struct *mm = current->mm;
3752 struct vm_area_struct *vma;
3753
Ingo Molnare8bff742008-02-13 20:21:06 +01003754 /*
3755 * Do not print if we are in atomic
3756 * contexts (in exception stacks, etc.):
3757 */
3758 if (preempt_count())
3759 return;
3760
Andi Kleen03252912008-01-30 13:33:18 +01003761 down_read(&mm->mmap_sem);
3762 vma = find_vma(mm, ip);
3763 if (vma && vma->vm_file) {
3764 struct file *f = vma->vm_file;
3765 char *buf = (char *)__get_free_page(GFP_KERNEL);
3766 if (buf) {
Andy Shevchenko2fbc57c2012-12-17 16:01:23 -08003767 char *p;
Andi Kleen03252912008-01-30 13:33:18 +01003768
Miklos Szeredi9bf39ab2015-06-19 10:29:13 +02003769 p = file_path(f, buf, PAGE_SIZE);
Andi Kleen03252912008-01-30 13:33:18 +01003770 if (IS_ERR(p))
3771 p = "?";
Andy Shevchenko2fbc57c2012-12-17 16:01:23 -08003772 printk("%s%s[%lx+%lx]", prefix, kbasename(p),
Andi Kleen03252912008-01-30 13:33:18 +01003773 vma->vm_start,
3774 vma->vm_end - vma->vm_start);
3775 free_page((unsigned long)buf);
3776 }
3777 }
Jeff Liu51a07e52012-07-31 16:43:18 -07003778 up_read(&mm->mmap_sem);
Andi Kleen03252912008-01-30 13:33:18 +01003779}
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003780
Michael S. Tsirkin662bbcb2013-05-26 17:32:23 +03003781#if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP)
David Hildenbrand9ec23532015-05-11 17:52:07 +02003782void __might_fault(const char *file, int line)
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003783{
Peter Zijlstra95156f02009-01-12 13:02:11 +01003784 /*
3785 * Some code (nfs/sunrpc) uses socket ops on kernel memory while
3786 * holding the mmap_sem, this is safe because kernel memory doesn't
3787 * get paged out, therefore we'll never actually fault, and the
3788 * below annotations will generate false positives.
3789 */
3790 if (segment_eq(get_fs(), KERNEL_DS))
3791 return;
David Hildenbrand9ec23532015-05-11 17:52:07 +02003792 if (pagefault_disabled())
Michael S. Tsirkin662bbcb2013-05-26 17:32:23 +03003793 return;
David Hildenbrand9ec23532015-05-11 17:52:07 +02003794 __might_sleep(file, line, 0);
3795#if defined(CONFIG_DEBUG_ATOMIC_SLEEP)
Michael S. Tsirkin662bbcb2013-05-26 17:32:23 +03003796 if (current->mm)
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003797 might_lock_read(&current->mm->mmap_sem);
David Hildenbrand9ec23532015-05-11 17:52:07 +02003798#endif
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003799}
David Hildenbrand9ec23532015-05-11 17:52:07 +02003800EXPORT_SYMBOL(__might_fault);
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003801#endif
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08003802
3803#if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)
3804static void clear_gigantic_page(struct page *page,
3805 unsigned long addr,
3806 unsigned int pages_per_huge_page)
3807{
3808 int i;
3809 struct page *p = page;
3810
3811 might_sleep();
3812 for (i = 0; i < pages_per_huge_page;
3813 i++, p = mem_map_next(p, page, i)) {
3814 cond_resched();
3815 clear_user_highpage(p, addr + i * PAGE_SIZE);
3816 }
3817}
3818void clear_huge_page(struct page *page,
3819 unsigned long addr, unsigned int pages_per_huge_page)
3820{
3821 int i;
3822
3823 if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
3824 clear_gigantic_page(page, addr, pages_per_huge_page);
3825 return;
3826 }
3827
3828 might_sleep();
3829 for (i = 0; i < pages_per_huge_page; i++) {
3830 cond_resched();
3831 clear_user_highpage(page + i, addr + i * PAGE_SIZE);
3832 }
3833}
3834
3835static void copy_user_gigantic_page(struct page *dst, struct page *src,
3836 unsigned long addr,
3837 struct vm_area_struct *vma,
3838 unsigned int pages_per_huge_page)
3839{
3840 int i;
3841 struct page *dst_base = dst;
3842 struct page *src_base = src;
3843
3844 for (i = 0; i < pages_per_huge_page; ) {
3845 cond_resched();
3846 copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma);
3847
3848 i++;
3849 dst = mem_map_next(dst, dst_base, i);
3850 src = mem_map_next(src, src_base, i);
3851 }
3852}
3853
3854void copy_user_huge_page(struct page *dst, struct page *src,
3855 unsigned long addr, struct vm_area_struct *vma,
3856 unsigned int pages_per_huge_page)
3857{
3858 int i;
3859
3860 if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
3861 copy_user_gigantic_page(dst, src, addr, vma,
3862 pages_per_huge_page);
3863 return;
3864 }
3865
3866 might_sleep();
3867 for (i = 0; i < pages_per_huge_page; i++) {
3868 cond_resched();
3869 copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma);
3870 }
3871}
3872#endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003873
Olof Johansson40b64ac2013-12-20 14:28:05 -08003874#if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS
Kirill A. Shutemovb35f1812014-01-21 15:49:07 -08003875
3876static struct kmem_cache *page_ptl_cachep;
3877
3878void __init ptlock_cache_init(void)
3879{
3880 page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0,
3881 SLAB_PANIC, NULL);
3882}
3883
Peter Zijlstra539edb52013-11-14 14:31:52 -08003884bool ptlock_alloc(struct page *page)
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003885{
3886 spinlock_t *ptl;
3887
Kirill A. Shutemovb35f1812014-01-21 15:49:07 -08003888 ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL);
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003889 if (!ptl)
3890 return false;
Peter Zijlstra539edb52013-11-14 14:31:52 -08003891 page->ptl = ptl;
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003892 return true;
3893}
3894
Peter Zijlstra539edb52013-11-14 14:31:52 -08003895void ptlock_free(struct page *page)
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003896{
Kirill A. Shutemovb35f1812014-01-21 15:49:07 -08003897 kmem_cache_free(page_ptl_cachep, page->ptl);
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003898}
3899#endif