blob: 8068893697bbdbb5d64f9d43508658d601a6932e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/memory.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 */
6
7/*
8 * demand-loading started 01.12.91 - seems it is high on the list of
9 * things wanted, and it should be easy to implement. - Linus
10 */
11
12/*
13 * Ok, demand-loading was easy, shared pages a little bit tricker. Shared
14 * pages started 02.12.91, seems to work. - Linus.
15 *
16 * Tested sharing by executing about 30 /bin/sh: under the old kernel it
17 * would have taken more than the 6M I have free, but it worked well as
18 * far as I could see.
19 *
20 * Also corrected some "invalidate()"s - I wasn't doing enough of them.
21 */
22
23/*
24 * Real VM (paging to/from disk) started 18.12.91. Much more work and
25 * thought has to go into this. Oh, well..
26 * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why.
27 * Found it. Everything seems to work now.
28 * 20.12.91 - Ok, making the swap-device changeable like the root.
29 */
30
31/*
32 * 05.04.94 - Multi-page memory management added for v1.1.
33 * Idea by Alex Bligh (alex@cconcepts.co.uk)
34 *
35 * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG
36 * (Gerhard.Wichert@pdb.siemens.de)
37 *
38 * Aug/Sep 2004 Changed to four level page tables (Andi Kleen)
39 */
40
41#include <linux/kernel_stat.h>
42#include <linux/mm.h>
43#include <linux/hugetlb.h>
44#include <linux/mman.h>
45#include <linux/swap.h>
46#include <linux/highmem.h>
47#include <linux/pagemap.h>
Hugh Dickins9a840892009-09-21 17:02:01 -070048#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include <linux/rmap.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040050#include <linux/export.h>
Shailabh Nagar0ff92242006-07-14 00:24:37 -070051#include <linux/delayacct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include <linux/init.h>
Peter Zijlstraedc79b22006-09-25 23:30:58 -070053#include <linux/writeback.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080054#include <linux/memcontrol.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070055#include <linux/mmu_notifier.h>
Hugh Dickins3dc14742009-01-06 14:40:08 -080056#include <linux/kallsyms.h>
57#include <linux/swapops.h>
58#include <linux/elf.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090059#include <linux/gfp.h>
Mel Gorman4daae3b2012-11-02 11:33:45 +000060#include <linux/migrate.h>
Andy Shevchenko2fbc57c2012-12-17 16:01:23 -080061#include <linux/string.h>
Dan Williams0abdd7a2014-01-21 15:48:12 -080062#include <linux/dma-debug.h>
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -070063#include <linux/debugfs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Alexey Dobriyan6952b612009-09-18 23:55:55 +040065#include <asm/io.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070066#include <asm/pgalloc.h>
67#include <asm/uaccess.h>
68#include <asm/tlb.h>
69#include <asm/tlbflush.h>
70#include <asm/pgtable.h>
71
Jan Beulich42b77722008-07-23 21:27:10 -070072#include "internal.h"
73
Peter Zijlstra90572892013-10-07 11:29:20 +010074#ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
75#warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid.
Peter Zijlstra75980e92013-02-22 16:34:32 -080076#endif
77
Andy Whitcroftd41dee32005-06-23 00:07:54 -070078#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -070079/* use the per-pgdat data instead for discontigmem - mbligh */
80unsigned long max_mapnr;
81struct page *mem_map;
82
83EXPORT_SYMBOL(max_mapnr);
84EXPORT_SYMBOL(mem_map);
85#endif
86
Linus Torvalds1da177e2005-04-16 15:20:36 -070087/*
88 * A number of key systems in x86 including ioremap() rely on the assumption
89 * that high_memory defines the upper bound on direct map memory, then end
90 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
91 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
92 * and ZONE_HIGHMEM.
93 */
94void * high_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -070095
Linus Torvalds1da177e2005-04-16 15:20:36 -070096EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
Ingo Molnar32a93232008-02-06 22:39:44 +010098/*
99 * Randomize the address space (stacks, mmaps, brk, etc.).
100 *
101 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
102 * as ancient (libc5 based) binaries can segfault. )
103 */
104int randomize_va_space __read_mostly =
105#ifdef CONFIG_COMPAT_BRK
106 1;
107#else
108 2;
109#endif
Andi Kleena62eaf12006-02-16 23:41:58 +0100110
111static int __init disable_randmaps(char *s)
112{
113 randomize_va_space = 0;
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800114 return 1;
Andi Kleena62eaf12006-02-16 23:41:58 +0100115}
116__setup("norandmaps", disable_randmaps);
117
Hugh Dickins62eede62009-09-21 17:03:34 -0700118unsigned long zero_pfn __read_mostly;
Hugh Dickins03f64622009-09-21 17:03:35 -0700119unsigned long highest_memmap_pfn __read_mostly;
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -0700120
Ard Biesheuvel0b700682014-09-12 22:17:23 +0200121EXPORT_SYMBOL(zero_pfn);
122
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -0700123/*
124 * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init()
125 */
126static int __init init_zero_pfn(void)
127{
128 zero_pfn = page_to_pfn(ZERO_PAGE(0));
129 return 0;
130}
131core_initcall(init_zero_pfn);
Andi Kleena62eaf12006-02-16 23:41:58 +0100132
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800133
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800134#if defined(SPLIT_RSS_COUNTING)
135
David Rientjesea48cf72012-03-21 16:34:13 -0700136void sync_mm_rss(struct mm_struct *mm)
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800137{
138 int i;
139
140 for (i = 0; i < NR_MM_COUNTERS; i++) {
David Rientjes05af2e12012-03-21 16:34:13 -0700141 if (current->rss_stat.count[i]) {
142 add_mm_counter(mm, i, current->rss_stat.count[i]);
143 current->rss_stat.count[i] = 0;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800144 }
145 }
David Rientjes05af2e12012-03-21 16:34:13 -0700146 current->rss_stat.events = 0;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800147}
148
149static void add_mm_counter_fast(struct mm_struct *mm, int member, int val)
150{
151 struct task_struct *task = current;
152
153 if (likely(task->mm == mm))
154 task->rss_stat.count[member] += val;
155 else
156 add_mm_counter(mm, member, val);
157}
158#define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1)
159#define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1)
160
161/* sync counter once per 64 page faults */
162#define TASK_RSS_EVENTS_THRESH (64)
163static void check_sync_rss_stat(struct task_struct *task)
164{
165 if (unlikely(task != current))
166 return;
167 if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH))
David Rientjesea48cf72012-03-21 16:34:13 -0700168 sync_mm_rss(task->mm);
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800169}
Peter Zijlstra9547d012011-05-24 17:12:14 -0700170#else /* SPLIT_RSS_COUNTING */
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800171
172#define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member)
173#define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member)
174
175static void check_sync_rss_stat(struct task_struct *task)
176{
177}
178
Peter Zijlstra9547d012011-05-24 17:12:14 -0700179#endif /* SPLIT_RSS_COUNTING */
180
181#ifdef HAVE_GENERIC_MMU_GATHER
182
183static int tlb_next_batch(struct mmu_gather *tlb)
184{
185 struct mmu_gather_batch *batch;
186
187 batch = tlb->active;
188 if (batch->next) {
189 tlb->active = batch->next;
190 return 1;
191 }
192
Michal Hocko53a59fc2013-01-04 15:35:12 -0800193 if (tlb->batch_count == MAX_GATHER_BATCH_COUNT)
194 return 0;
195
Peter Zijlstra9547d012011-05-24 17:12:14 -0700196 batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0);
197 if (!batch)
198 return 0;
199
Michal Hocko53a59fc2013-01-04 15:35:12 -0800200 tlb->batch_count++;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700201 batch->next = NULL;
202 batch->nr = 0;
203 batch->max = MAX_GATHER_BATCH;
204
205 tlb->active->next = batch;
206 tlb->active = batch;
207
208 return 1;
209}
210
211/* tlb_gather_mmu
212 * Called to initialize an (on-stack) mmu_gather structure for page-table
213 * tear-down from @mm. The @fullmm argument is used when @mm is without
214 * users and we're going to destroy the full address space (exit/execve).
215 */
Linus Torvalds2b047252013-08-15 11:42:25 -0700216void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, unsigned long start, unsigned long end)
Peter Zijlstra9547d012011-05-24 17:12:14 -0700217{
218 tlb->mm = mm;
219
Linus Torvalds2b047252013-08-15 11:42:25 -0700220 /* Is it from 0 to ~0? */
221 tlb->fullmm = !(start | (end+1));
Dave Hansen1de14c32013-04-12 16:23:54 -0700222 tlb->need_flush_all = 0;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700223 tlb->local.next = NULL;
224 tlb->local.nr = 0;
225 tlb->local.max = ARRAY_SIZE(tlb->__pages);
226 tlb->active = &tlb->local;
Michal Hocko53a59fc2013-01-04 15:35:12 -0800227 tlb->batch_count = 0;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700228
229#ifdef CONFIG_HAVE_RCU_TABLE_FREE
230 tlb->batch = NULL;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800231#endif
Will Deaconfb7332a2014-10-29 10:03:09 +0000232
233 __tlb_reset_range(tlb);
Peter Zijlstra9547d012011-05-24 17:12:14 -0700234}
235
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700236static void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb)
Peter Zijlstra9547d012011-05-24 17:12:14 -0700237{
Will Deacon721c21c2015-01-12 19:10:55 +0000238 if (!tlb->end)
239 return;
240
Peter Zijlstra9547d012011-05-24 17:12:14 -0700241 tlb_flush(tlb);
Joerg Roedel34ee6452014-11-13 13:46:09 +1100242 mmu_notifier_invalidate_range(tlb->mm, tlb->start, tlb->end);
Peter Zijlstra9547d012011-05-24 17:12:14 -0700243#ifdef CONFIG_HAVE_RCU_TABLE_FREE
244 tlb_table_flush(tlb);
245#endif
Will Deaconfb7332a2014-10-29 10:03:09 +0000246 __tlb_reset_range(tlb);
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700247}
248
249static void tlb_flush_mmu_free(struct mmu_gather *tlb)
250{
251 struct mmu_gather_batch *batch;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700252
Will Deacon721c21c2015-01-12 19:10:55 +0000253 for (batch = &tlb->local; batch && batch->nr; batch = batch->next) {
Peter Zijlstra9547d012011-05-24 17:12:14 -0700254 free_pages_and_swap_cache(batch->pages, batch->nr);
255 batch->nr = 0;
256 }
257 tlb->active = &tlb->local;
258}
259
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700260void tlb_flush_mmu(struct mmu_gather *tlb)
261{
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700262 tlb_flush_mmu_tlbonly(tlb);
263 tlb_flush_mmu_free(tlb);
264}
265
Peter Zijlstra9547d012011-05-24 17:12:14 -0700266/* tlb_finish_mmu
267 * Called at the end of the shootdown operation to free up any resources
268 * that were required.
269 */
270void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end)
271{
272 struct mmu_gather_batch *batch, *next;
273
274 tlb_flush_mmu(tlb);
275
276 /* keep the page table cache within bounds */
277 check_pgt_cache();
278
279 for (batch = tlb->local.next; batch; batch = next) {
280 next = batch->next;
281 free_pages((unsigned long)batch, 0);
282 }
283 tlb->local.next = NULL;
284}
285
286/* __tlb_remove_page
287 * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while
288 * handling the additional races in SMP caused by other CPUs caching valid
289 * mappings in their TLBs. Returns the number of free page slots left.
290 * When out of page slots we must call tlb_flush_mmu().
291 */
292int __tlb_remove_page(struct mmu_gather *tlb, struct page *page)
293{
294 struct mmu_gather_batch *batch;
295
Will Deaconfb7332a2014-10-29 10:03:09 +0000296 VM_BUG_ON(!tlb->end);
Peter Zijlstra9547d012011-05-24 17:12:14 -0700297
Peter Zijlstra9547d012011-05-24 17:12:14 -0700298 batch = tlb->active;
299 batch->pages[batch->nr++] = page;
300 if (batch->nr == batch->max) {
301 if (!tlb_next_batch(tlb))
302 return 0;
Shaohua Li0b43c3a2011-07-08 15:39:41 -0700303 batch = tlb->active;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700304 }
Sasha Levin309381fea2014-01-23 15:52:54 -0800305 VM_BUG_ON_PAGE(batch->nr > batch->max, page);
Peter Zijlstra9547d012011-05-24 17:12:14 -0700306
307 return batch->max - batch->nr;
308}
309
310#endif /* HAVE_GENERIC_MMU_GATHER */
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800311
Peter Zijlstra26723912011-05-24 17:12:00 -0700312#ifdef CONFIG_HAVE_RCU_TABLE_FREE
313
314/*
315 * See the comment near struct mmu_table_batch.
316 */
317
318static void tlb_remove_table_smp_sync(void *arg)
319{
320 /* Simply deliver the interrupt */
321}
322
323static void tlb_remove_table_one(void *table)
324{
325 /*
326 * This isn't an RCU grace period and hence the page-tables cannot be
327 * assumed to be actually RCU-freed.
328 *
329 * It is however sufficient for software page-table walkers that rely on
330 * IRQ disabling. See the comment near struct mmu_table_batch.
331 */
332 smp_call_function(tlb_remove_table_smp_sync, NULL, 1);
333 __tlb_remove_table(table);
334}
335
336static void tlb_remove_table_rcu(struct rcu_head *head)
337{
338 struct mmu_table_batch *batch;
339 int i;
340
341 batch = container_of(head, struct mmu_table_batch, rcu);
342
343 for (i = 0; i < batch->nr; i++)
344 __tlb_remove_table(batch->tables[i]);
345
346 free_page((unsigned long)batch);
347}
348
349void tlb_table_flush(struct mmu_gather *tlb)
350{
351 struct mmu_table_batch **batch = &tlb->batch;
352
353 if (*batch) {
354 call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu);
355 *batch = NULL;
356 }
357}
358
359void tlb_remove_table(struct mmu_gather *tlb, void *table)
360{
361 struct mmu_table_batch **batch = &tlb->batch;
362
Peter Zijlstra26723912011-05-24 17:12:00 -0700363 /*
364 * When there's less then two users of this mm there cannot be a
365 * concurrent page-table walk.
366 */
367 if (atomic_read(&tlb->mm->mm_users) < 2) {
368 __tlb_remove_table(table);
369 return;
370 }
371
372 if (*batch == NULL) {
373 *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN);
374 if (*batch == NULL) {
375 tlb_remove_table_one(table);
376 return;
377 }
378 (*batch)->nr = 0;
379 }
380 (*batch)->tables[(*batch)->nr++] = table;
381 if ((*batch)->nr == MAX_TABLE_BATCH)
382 tlb_table_flush(tlb);
383}
384
Peter Zijlstra9547d012011-05-24 17:12:14 -0700385#endif /* CONFIG_HAVE_RCU_TABLE_FREE */
Peter Zijlstra26723912011-05-24 17:12:00 -0700386
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 * Note: this doesn't free the actual pages themselves. That
389 * has been handled earlier when unmapping all the memory regions.
390 */
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000391static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
392 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800394 pgtable_t token = pmd_pgtable(*pmd);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700395 pmd_clear(pmd);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000396 pte_free_tlb(tlb, token, addr);
Kirill A. Shutemove1f56c82013-11-14 14:30:48 -0800397 atomic_long_dec(&tlb->mm->nr_ptes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398}
399
Hugh Dickinse0da3822005-04-19 13:29:15 -0700400static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
401 unsigned long addr, unsigned long end,
402 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403{
404 pmd_t *pmd;
405 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700406 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407
Hugh Dickinse0da3822005-04-19 13:29:15 -0700408 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 do {
411 next = pmd_addr_end(addr, end);
412 if (pmd_none_or_clear_bad(pmd))
413 continue;
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000414 free_pte_range(tlb, pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 } while (pmd++, addr = next, addr != end);
416
Hugh Dickinse0da3822005-04-19 13:29:15 -0700417 start &= PUD_MASK;
418 if (start < floor)
419 return;
420 if (ceiling) {
421 ceiling &= PUD_MASK;
422 if (!ceiling)
423 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700425 if (end - 1 > ceiling - 1)
426 return;
427
428 pmd = pmd_offset(pud, start);
429 pud_clear(pud);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000430 pmd_free_tlb(tlb, pmd, start);
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -0800431 mm_dec_nr_pmds(tlb->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432}
433
Hugh Dickinse0da3822005-04-19 13:29:15 -0700434static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
435 unsigned long addr, unsigned long end,
436 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437{
438 pud_t *pud;
439 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700440 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441
Hugh Dickinse0da3822005-04-19 13:29:15 -0700442 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 pud = pud_offset(pgd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 do {
445 next = pud_addr_end(addr, end);
446 if (pud_none_or_clear_bad(pud))
447 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700448 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 } while (pud++, addr = next, addr != end);
450
Hugh Dickinse0da3822005-04-19 13:29:15 -0700451 start &= PGDIR_MASK;
452 if (start < floor)
453 return;
454 if (ceiling) {
455 ceiling &= PGDIR_MASK;
456 if (!ceiling)
457 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700459 if (end - 1 > ceiling - 1)
460 return;
461
462 pud = pud_offset(pgd, start);
463 pgd_clear(pgd);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000464 pud_free_tlb(tlb, pud, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465}
466
467/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700468 * This function frees user-level page tables of a process.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 */
Jan Beulich42b77722008-07-23 21:27:10 -0700470void free_pgd_range(struct mmu_gather *tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700471 unsigned long addr, unsigned long end,
472 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473{
474 pgd_t *pgd;
475 unsigned long next;
476
Hugh Dickinse0da3822005-04-19 13:29:15 -0700477 /*
478 * The next few lines have given us lots of grief...
479 *
480 * Why are we testing PMD* at this top level? Because often
481 * there will be no work to do at all, and we'd prefer not to
482 * go all the way down to the bottom just to discover that.
483 *
484 * Why all these "- 1"s? Because 0 represents both the bottom
485 * of the address space and the top of it (using -1 for the
486 * top wouldn't help much: the masks would do the wrong thing).
487 * The rule is that addr 0 and floor 0 refer to the bottom of
488 * the address space, but end 0 and ceiling 0 refer to the top
489 * Comparisons need to use "end - 1" and "ceiling - 1" (though
490 * that end 0 case should be mythical).
491 *
492 * Wherever addr is brought up or ceiling brought down, we must
493 * be careful to reject "the opposite 0" before it confuses the
494 * subsequent tests. But what about where end is brought down
495 * by PMD_SIZE below? no, end can't go down to 0 there.
496 *
497 * Whereas we round start (addr) and ceiling down, by different
498 * masks at different levels, in order to test whether a table
499 * now has no other vmas using it, so can be freed, we don't
500 * bother to round floor or end up - the tests don't need that.
501 */
502
503 addr &= PMD_MASK;
504 if (addr < floor) {
505 addr += PMD_SIZE;
506 if (!addr)
507 return;
508 }
509 if (ceiling) {
510 ceiling &= PMD_MASK;
511 if (!ceiling)
512 return;
513 }
514 if (end - 1 > ceiling - 1)
515 end -= PMD_SIZE;
516 if (addr > end - 1)
517 return;
518
Jan Beulich42b77722008-07-23 21:27:10 -0700519 pgd = pgd_offset(tlb->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 do {
521 next = pgd_addr_end(addr, end);
522 if (pgd_none_or_clear_bad(pgd))
523 continue;
Jan Beulich42b77722008-07-23 21:27:10 -0700524 free_pud_range(tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700526}
527
Jan Beulich42b77722008-07-23 21:27:10 -0700528void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700529 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700530{
531 while (vma) {
532 struct vm_area_struct *next = vma->vm_next;
533 unsigned long addr = vma->vm_start;
534
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700535 /*
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000536 * Hide vma from rmap and truncate_pagecache before freeing
537 * pgtables
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700538 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800539 unlink_anon_vmas(vma);
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700540 unlink_file_vma(vma);
541
David Gibson9da61ae2006-03-22 00:08:57 -0800542 if (is_vm_hugetlb_page(vma)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700543 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700544 floor, next? next->vm_start: ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700545 } else {
546 /*
547 * Optimization: gather nearby vmas into one call down
548 */
549 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
David Gibson48669202006-03-22 00:08:58 -0800550 && !is_vm_hugetlb_page(next)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700551 vma = next;
552 next = vma->vm_next;
Rik van Riel5beb4932010-03-05 13:42:07 -0800553 unlink_anon_vmas(vma);
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700554 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700555 }
556 free_pgd_range(tlb, addr, vma->vm_end,
557 floor, next? next->vm_start: ceiling);
558 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700559 vma = next;
560 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561}
562
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800563int __pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,
564 pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565{
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -0800566 spinlock_t *ptl;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800567 pgtable_t new = pte_alloc_one(mm, address);
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800568 int wait_split_huge_page;
Hugh Dickins1bb36302005-10-29 18:16:22 -0700569 if (!new)
570 return -ENOMEM;
571
Nick Piggin362a61a2008-05-14 06:37:36 +0200572 /*
573 * Ensure all pte setup (eg. pte page lock and page clearing) are
574 * visible before the pte is made visible to other CPUs by being
575 * put into page tables.
576 *
577 * The other side of the story is the pointer chasing in the page
578 * table walking code (when walking the page table without locking;
579 * ie. most of the time). Fortunately, these data accesses consist
580 * of a chain of data-dependent loads, meaning most CPUs (alpha
581 * being the notable exception) will already guarantee loads are
582 * seen in-order. See the alpha page table accessors for the
583 * smp_read_barrier_depends() barriers in page table walking code.
584 */
585 smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
586
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -0800587 ptl = pmd_lock(mm, pmd);
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800588 wait_split_huge_page = 0;
589 if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
Kirill A. Shutemove1f56c82013-11-14 14:30:48 -0800590 atomic_long_inc(&mm->nr_ptes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 pmd_populate(mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800592 new = NULL;
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800593 } else if (unlikely(pmd_trans_splitting(*pmd)))
594 wait_split_huge_page = 1;
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -0800595 spin_unlock(ptl);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800596 if (new)
597 pte_free(mm, new);
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800598 if (wait_split_huge_page)
599 wait_split_huge_page(vma->anon_vma, pmd);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700600 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601}
602
Hugh Dickins1bb36302005-10-29 18:16:22 -0700603int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700605 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
606 if (!new)
607 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608
Nick Piggin362a61a2008-05-14 06:37:36 +0200609 smp_wmb(); /* See comment in __pte_alloc */
610
Hugh Dickins1bb36302005-10-29 18:16:22 -0700611 spin_lock(&init_mm.page_table_lock);
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800612 if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
Hugh Dickins1bb36302005-10-29 18:16:22 -0700613 pmd_populate_kernel(&init_mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800614 new = NULL;
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800615 } else
616 VM_BUG_ON(pmd_trans_splitting(*pmd));
Hugh Dickins1bb36302005-10-29 18:16:22 -0700617 spin_unlock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800618 if (new)
619 pte_free_kernel(&init_mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700620 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621}
622
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800623static inline void init_rss_vec(int *rss)
Hugh Dickinsae859762005-10-29 18:16:05 -0700624{
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800625 memset(rss, 0, sizeof(int) * NR_MM_COUNTERS);
626}
627
628static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss)
629{
630 int i;
631
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800632 if (current->mm == mm)
David Rientjes05af2e12012-03-21 16:34:13 -0700633 sync_mm_rss(mm);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800634 for (i = 0; i < NR_MM_COUNTERS; i++)
635 if (rss[i])
636 add_mm_counter(mm, i, rss[i]);
Hugh Dickinsae859762005-10-29 18:16:05 -0700637}
638
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800640 * This function is called to print an error when a bad pte
641 * is found. For example, we might have a PFN-mapped pte in
642 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700643 *
644 * The calling function must still handle the error.
645 */
Hugh Dickins3dc14742009-01-06 14:40:08 -0800646static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
647 pte_t pte, struct page *page)
Nick Pigginb5810032005-10-29 18:16:12 -0700648{
Hugh Dickins3dc14742009-01-06 14:40:08 -0800649 pgd_t *pgd = pgd_offset(vma->vm_mm, addr);
650 pud_t *pud = pud_offset(pgd, addr);
651 pmd_t *pmd = pmd_offset(pud, addr);
652 struct address_space *mapping;
653 pgoff_t index;
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800654 static unsigned long resume;
655 static unsigned long nr_shown;
656 static unsigned long nr_unshown;
657
658 /*
659 * Allow a burst of 60 reports, then keep quiet for that minute;
660 * or allow a steady drip of one report per second.
661 */
662 if (nr_shown == 60) {
663 if (time_before(jiffies, resume)) {
664 nr_unshown++;
665 return;
666 }
667 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800668 printk(KERN_ALERT
669 "BUG: Bad page map: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800670 nr_unshown);
671 nr_unshown = 0;
672 }
673 nr_shown = 0;
674 }
675 if (nr_shown++ == 0)
676 resume = jiffies + 60 * HZ;
Hugh Dickins3dc14742009-01-06 14:40:08 -0800677
678 mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;
679 index = linear_page_index(vma, addr);
680
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800681 printk(KERN_ALERT
682 "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800683 current->comm,
684 (long long)pte_val(pte), (long long)pmd_val(*pmd));
Wu Fengguang718a3822010-03-10 15:20:43 -0800685 if (page)
Dave Hansenf0b791a2014-01-23 15:52:49 -0800686 dump_page(page, "bad pte");
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800687 printk(KERN_ALERT
Hugh Dickins3dc14742009-01-06 14:40:08 -0800688 "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
689 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
690 /*
691 * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y
692 */
693 if (vma->vm_ops)
Joe Perches071361d2012-12-12 10:19:12 -0800694 printk(KERN_ALERT "vma->vm_ops->fault: %pSR\n",
695 vma->vm_ops->fault);
Al Viro72c2d532013-09-22 16:27:52 -0400696 if (vma->vm_file)
Joe Perches071361d2012-12-12 10:19:12 -0800697 printk(KERN_ALERT "vma->vm_file->f_op->mmap: %pSR\n",
698 vma->vm_file->f_op->mmap);
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 Piggin423bad602008-04-28 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 Piggin423bad602008-04-28 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 Piggin423bad602008-04-28 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 Piggin423bad602008-04-28 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 Piggin423bad602008-04-28 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 Piggin423bad602008-04-28 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 Piggin423bad602008-04-28 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 Piggin423bad602008-04-28 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 Piggin423bad602008-04-28 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 Piggin423bad602008-04-28 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/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 * This routine handles present pages, when users try to write
1987 * to a shared page. It is done by copying the page to a new address
1988 * and decrementing the shared-page counter for the old page.
1989 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 * Note that this routine assumes that the protection checks have been
1991 * done by the caller (the low-level page fault routine in most cases).
1992 * Thus we can safely just mark it writable once we've done any necessary
1993 * COW.
1994 *
1995 * We also mark the page dirty at this point even though the page will
1996 * change only once the write actually happens. This avoids a few races,
1997 * and potentially makes it more efficient.
1998 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001999 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2000 * but allow concurrent faults), with pte both mapped and locked.
2001 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002003static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
2004 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002005 spinlock_t *ptl, pte_t orig_pte)
Namhyung Kime6219ec2010-10-26 14:22:00 -07002006 __releases(ptl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007{
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07002008 struct page *old_page, *new_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 pte_t entry;
Michel Lespinasseb009c022011-01-13 15:46:07 -08002010 int ret = 0;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002011 int page_mkwrite = 0;
Johannes Weinerf38b4b32015-02-10 14:11:30 -08002012 bool dirty_shared = false;
David Rientjes17569542012-11-16 14:14:48 -08002013 unsigned long mmun_start = 0; /* For mmu_notifiers */
2014 unsigned long mmun_end = 0; /* For mmu_notifiers */
Johannes Weiner00501b52014-08-08 14:19:20 -07002015 struct mem_cgroup *memcg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016
Linus Torvalds6aab3412005-11-28 14:34:23 -08002017 old_page = vm_normal_page(vma, address, orig_pte);
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002018 if (!old_page) {
2019 /*
Peter Feiner64e45502014-10-13 15:55:46 -07002020 * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a
2021 * VM_PFNMAP VMA.
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002022 *
2023 * We should not cow pages in a shared writeable mapping.
2024 * Just mark the pages writable as we can't do any dirty
2025 * accounting on raw pfn maps.
2026 */
2027 if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
2028 (VM_WRITE|VM_SHARED))
2029 goto reuse;
Linus Torvalds6aab3412005-11-28 14:34:23 -08002030 goto gotten;
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002031 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002033 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002034 * Take out anonymous pages first, anonymous shared vmas are
2035 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002036 */
Hugh Dickins9a840892009-09-21 17:02:01 -07002037 if (PageAnon(old_page) && !PageKsm(old_page)) {
Hugh Dickinsab967d82009-01-06 14:39:33 -08002038 if (!trylock_page(old_page)) {
2039 page_cache_get(old_page);
2040 pte_unmap_unlock(page_table, ptl);
2041 lock_page(old_page);
2042 page_table = pte_offset_map_lock(mm, pmd, address,
2043 &ptl);
2044 if (!pte_same(*page_table, orig_pte)) {
2045 unlock_page(old_page);
Hugh Dickinsab967d82009-01-06 14:39:33 -08002046 goto unlock;
2047 }
2048 page_cache_release(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002049 }
Michel Lespinasseb009c022011-01-13 15:46:07 -08002050 if (reuse_swap_page(old_page)) {
Rik van Rielc44b6742010-03-05 13:42:09 -08002051 /*
2052 * The page is all ours. Move it to our anon_vma so
2053 * the rmap code will not search our parent or siblings.
2054 * Protected against the rmap code by the page lock.
2055 */
2056 page_move_anon_rmap(old_page, vma, address);
Michel Lespinasseb009c022011-01-13 15:46:07 -08002057 unlock_page(old_page);
2058 goto reuse;
2059 }
Hugh Dickinsab967d82009-01-06 14:39:33 -08002060 unlock_page(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002061 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002062 (VM_WRITE|VM_SHARED))) {
Johannes Weinerf38b4b32015-02-10 14:11:30 -08002063 page_cache_get(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002064 /*
2065 * Only catch write-faults on shared writable pages,
2066 * read-only shared pages can get COWed by
2067 * get_user_pages(.write=1, .force=1).
2068 */
David Howells9637a5e2006-06-23 02:03:43 -07002069 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
Nick Pigginc2ec1752009-03-31 15:23:21 -07002070 int tmp;
Johannes Weinerf38b4b32015-02-10 14:11:30 -08002071
David Howells9637a5e2006-06-23 02:03:43 -07002072 pte_unmap_unlock(page_table, ptl);
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07002073 tmp = do_page_mkwrite(vma, old_page, address);
2074 if (unlikely(!tmp || (tmp &
2075 (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
2076 page_cache_release(old_page);
2077 return tmp;
Nick Pigginc2ec1752009-03-31 15:23:21 -07002078 }
David Howells9637a5e2006-06-23 02:03:43 -07002079 /*
2080 * Since we dropped the lock we need to revalidate
2081 * the PTE as someone else may have changed it. If
2082 * they did, we just return, as we can count on the
2083 * MMU to tell us if they didn't also make it writable.
2084 */
2085 page_table = pte_offset_map_lock(mm, pmd, address,
2086 &ptl);
Nick Pigginb827e492009-04-30 15:08:16 -07002087 if (!pte_same(*page_table, orig_pte)) {
2088 unlock_page(old_page);
David Howells9637a5e2006-06-23 02:03:43 -07002089 goto unlock;
Nick Pigginb827e492009-04-30 15:08:16 -07002090 }
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002091 page_mkwrite = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 }
Johannes Weinerf38b4b32015-02-10 14:11:30 -08002093
2094 dirty_shared = true;
David Howells9637a5e2006-06-23 02:03:43 -07002095
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002096reuse:
Peter Zijlstra8c8a7432013-10-07 11:29:21 +01002097 /*
2098 * Clear the pages cpupid information as the existing
2099 * information potentially belongs to a now completely
2100 * unrelated process.
2101 */
2102 if (old_page)
2103 page_cpupid_xchg_last(old_page, (1 << LAST_CPUPID_SHIFT) - 1);
2104
David Howells9637a5e2006-06-23 02:03:43 -07002105 flush_cache_page(vma, address, pte_pfn(orig_pte));
2106 entry = pte_mkyoung(orig_pte);
2107 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
KAMEZAWA Hiroyuki954ffcb2007-10-16 01:25:44 -07002108 if (ptep_set_access_flags(vma, address, page_table, entry,1))
Russell King4b3073e2009-12-18 16:40:18 +00002109 update_mmu_cache(vma, address, page_table);
Michel Lespinasse72ddc8f2011-01-13 15:46:08 -08002110 pte_unmap_unlock(page_table, ptl);
David Howells9637a5e2006-06-23 02:03:43 -07002111 ret |= VM_FAULT_WRITE;
Michel Lespinasse72ddc8f2011-01-13 15:46:08 -08002112
Johannes Weinerf38b4b32015-02-10 14:11:30 -08002113 if (dirty_shared) {
Johannes Weiner2d6d7f92015-01-08 14:32:18 -08002114 struct address_space *mapping;
2115 int dirtied;
2116
Johannes Weinerf38b4b32015-02-10 14:11:30 -08002117 if (!page_mkwrite)
2118 lock_page(old_page);
Johannes Weiner2d6d7f92015-01-08 14:32:18 -08002119
Johannes Weinerf38b4b32015-02-10 14:11:30 -08002120 dirtied = set_page_dirty(old_page);
2121 VM_BUG_ON_PAGE(PageAnon(old_page), old_page);
2122 mapping = old_page->mapping;
2123 unlock_page(old_page);
2124 page_cache_release(old_page);
2125
2126 if ((dirtied || page_mkwrite) && mapping) {
Johannes Weiner2d6d7f92015-01-08 14:32:18 -08002127 /*
2128 * Some device drivers do not set page.mapping
2129 * but still dirty their pages
2130 */
2131 balance_dirty_pages_ratelimited(mapping);
2132 }
2133
Johannes Weinerf38b4b32015-02-10 14:11:30 -08002134 if (!page_mkwrite)
Jan Kara41c4d252012-06-12 16:20:28 +02002135 file_update_time(vma->vm_file);
Michel Lespinasse72ddc8f2011-01-13 15:46:08 -08002136 }
Michel Lespinasse72ddc8f2011-01-13 15:46:08 -08002137
Michel Lespinasse72ddc8f2011-01-13 15:46:08 -08002138 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140
2141 /*
2142 * Ok, we need to copy. Oh, well..
2143 */
Nick Pigginb5810032005-10-29 18:16:12 -07002144 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08002145gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002146 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147
2148 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07002149 goto oom;
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002150
Hugh Dickins62eede62009-09-21 17:03:34 -07002151 if (is_zero_pfn(pte_pfn(orig_pte))) {
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002152 new_page = alloc_zeroed_user_highpage_movable(vma, address);
2153 if (!new_page)
2154 goto oom;
2155 } else {
2156 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
2157 if (!new_page)
2158 goto oom;
2159 cow_user_page(new_page, old_page, address, vma);
2160 }
2161 __SetPageUptodate(new_page);
2162
Johannes Weiner00501b52014-08-08 14:19:20 -07002163 if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002164 goto oom_free_new;
2165
Haggai Eran6bdb9132012-10-08 16:33:35 -07002166 mmun_start = address & PAGE_MASK;
David Rientjes17569542012-11-16 14:14:48 -08002167 mmun_end = mmun_start + PAGE_SIZE;
Haggai Eran6bdb9132012-10-08 16:33:35 -07002168 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
2169
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 /*
2171 * Re-check the pte - we dropped the lock
2172 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002173 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07002174 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08002175 if (old_page) {
Hugh Dickins920fc352005-11-21 21:32:17 -08002176 if (!PageAnon(old_page)) {
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08002177 dec_mm_counter_fast(mm, MM_FILEPAGES);
2178 inc_mm_counter_fast(mm, MM_ANONPAGES);
Hugh Dickins920fc352005-11-21 21:32:17 -08002179 }
2180 } else
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08002181 inc_mm_counter_fast(mm, MM_ANONPAGES);
Ben Collinseca35132005-11-29 11:45:26 -08002182 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07002183 entry = mk_pte(new_page, vma->vm_page_prot);
2184 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Siddha, Suresh B4ce072f2006-09-29 01:58:42 -07002185 /*
2186 * Clear the pte entry and flush it first, before updating the
2187 * pte with the new entry. This will avoid a race condition
2188 * seen in the presence of one thread doing SMC and another
2189 * thread doing COW.
2190 */
Joerg Roedel34ee6452014-11-13 13:46:09 +11002191 ptep_clear_flush_notify(vma, address, page_table);
Nick Piggin9617d952006-01-06 00:11:12 -08002192 page_add_new_anon_rmap(new_page, vma, address);
Johannes Weiner00501b52014-08-08 14:19:20 -07002193 mem_cgroup_commit_charge(new_page, memcg, false);
2194 lru_cache_add_active_or_unevictable(new_page, vma);
Izik Eidus828502d2009-09-21 17:01:51 -07002195 /*
2196 * We call the notify macro here because, when using secondary
2197 * mmu page tables (such as kvm shadow page tables), we want the
2198 * new page to be mapped directly into the secondary page table.
2199 */
2200 set_pte_at_notify(mm, address, page_table, entry);
Russell King4b3073e2009-12-18 16:40:18 +00002201 update_mmu_cache(vma, address, page_table);
Nick Piggin945754a2008-06-23 14:30:30 +02002202 if (old_page) {
2203 /*
2204 * Only after switching the pte to the new page may
2205 * we remove the mapcount here. Otherwise another
2206 * process may come and find the rmap count decremented
2207 * before the pte is switched to the new page, and
2208 * "reuse" the old page writing into it while our pte
2209 * here still points into it and can be read by other
2210 * threads.
2211 *
2212 * The critical issue is to order this
2213 * page_remove_rmap with the ptp_clear_flush above.
2214 * Those stores are ordered by (if nothing else,)
2215 * the barrier present in the atomic_add_negative
2216 * in page_remove_rmap.
2217 *
2218 * Then the TLB flush in ptep_clear_flush ensures that
2219 * no process can access the old page before the
2220 * decremented mapcount is visible. And the old page
2221 * cannot be reused until after the decremented
2222 * mapcount is visible. So transitively, TLBs to
2223 * old page will be flushed before it can be reused.
2224 */
Hugh Dickinsedc315f2009-01-06 14:40:11 -08002225 page_remove_rmap(old_page);
Nick Piggin945754a2008-06-23 14:30:30 +02002226 }
2227
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 /* Free the old page.. */
2229 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10002230 ret |= VM_FAULT_WRITE;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002231 } else
Johannes Weiner00501b52014-08-08 14:19:20 -07002232 mem_cgroup_cancel_charge(new_page, memcg);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002233
Haggai Eran6bdb9132012-10-08 16:33:35 -07002234 if (new_page)
2235 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002236unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002237 pte_unmap_unlock(page_table, ptl);
David Rientjes17569542012-11-16 14:14:48 -08002238 if (mmun_end > mmun_start)
Haggai Eran6bdb9132012-10-08 16:33:35 -07002239 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
Michel Lespinassee15f8c02011-02-10 15:01:32 -08002240 if (old_page) {
2241 /*
2242 * Don't let another task, with possibly unlocked vma,
2243 * keep the mlocked page.
2244 */
2245 if ((ret & VM_FAULT_WRITE) && (vma->vm_flags & VM_LOCKED)) {
2246 lock_page(old_page); /* LRU manipulation */
2247 munlock_vma_page(old_page);
2248 unlock_page(old_page);
2249 }
2250 page_cache_release(old_page);
2251 }
Nick Pigginf33ea7f2005-08-03 20:24:01 +10002252 return ret;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002253oom_free_new:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002254 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002255oom:
Dominik Dingel66521d52012-12-12 13:52:37 -08002256 if (old_page)
Hugh Dickins920fc352005-11-21 21:32:17 -08002257 page_cache_release(old_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 return VM_FAULT_OOM;
2259}
2260
Peter Zijlstra97a89412011-05-24 17:12:04 -07002261static void unmap_mapping_range_vma(struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 unsigned long start_addr, unsigned long end_addr,
2263 struct zap_details *details)
2264{
Al Virof5cc4ee2012-03-05 14:14:20 -05002265 zap_page_range_single(vma, start_addr, end_addr - start_addr, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266}
2267
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07002268static inline void unmap_mapping_range_tree(struct rb_root *root,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 struct zap_details *details)
2270{
2271 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 pgoff_t vba, vea, zba, zea;
2273
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07002274 vma_interval_tree_foreach(vma, root,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 details->first_index, details->last_index) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276
2277 vba = vma->vm_pgoff;
Libind6e93212013-07-03 15:01:26 -07002278 vea = vba + vma_pages(vma) - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
2280 zba = details->first_index;
2281 if (zba < vba)
2282 zba = vba;
2283 zea = details->last_index;
2284 if (zea > vea)
2285 zea = vea;
2286
Peter Zijlstra97a89412011-05-24 17:12:04 -07002287 unmap_mapping_range_vma(vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
2289 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
Peter Zijlstra97a89412011-05-24 17:12:04 -07002290 details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 }
2292}
2293
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294/**
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08002295 * unmap_mapping_range - unmap the portion of all mmaps in the specified
2296 * address_space corresponding to the specified page range in the underlying
2297 * file.
2298 *
Martin Waitz3d410882005-06-23 22:05:21 -07002299 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 * @holebegin: byte in first page to unmap, relative to the start of
2301 * the underlying file. This will be rounded down to a PAGE_SIZE
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +10002302 * boundary. Note that this is different from truncate_pagecache(), which
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 * must keep the partial page. In contrast, we must get rid of
2304 * partial pages.
2305 * @holelen: size of prospective hole in bytes. This will be rounded
2306 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2307 * end of the file.
2308 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2309 * but 0 when invalidating pagecache, don't throw away private data.
2310 */
2311void unmap_mapping_range(struct address_space *mapping,
2312 loff_t const holebegin, loff_t const holelen, int even_cows)
2313{
2314 struct zap_details details;
2315 pgoff_t hba = holebegin >> PAGE_SHIFT;
2316 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2317
2318 /* Check for overflow. */
2319 if (sizeof(holelen) > sizeof(hlen)) {
2320 long long holeend =
2321 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2322 if (holeend & ~(long long)ULONG_MAX)
2323 hlen = ULONG_MAX - hba + 1;
2324 }
2325
2326 details.check_mapping = even_cows? NULL: mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 details.first_index = hba;
2328 details.last_index = hba + hlen - 1;
2329 if (details.last_index < details.first_index)
2330 details.last_index = ULONG_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331
Peter Zijlstra97a89412011-05-24 17:12:04 -07002332
Matthew Wilcox283307c2015-02-16 15:58:46 -08002333 /* DAX uses i_mmap_lock to serialise file truncate vs page fault */
Kirill A. Shutemov48ec8332014-12-22 21:01:54 +02002334 i_mmap_lock_write(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07002335 if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 unmap_mapping_range_tree(&mapping->i_mmap, &details);
Kirill A. Shutemov48ec8332014-12-22 21:01:54 +02002337 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338}
2339EXPORT_SYMBOL(unmap_mapping_range);
2340
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002342 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2343 * but allow concurrent faults), and pte mapped but not yet locked.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07002344 * We return with pte unmapped and unlocked.
2345 *
2346 * We return with the mmap_sem locked or unlocked in the same cases
2347 * as does filemap_fault().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002349static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2350 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002351 unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002353 spinlock_t *ptl;
Hugh Dickins56f31802013-02-22 16:36:10 -08002354 struct page *page, *swapcache;
Johannes Weiner00501b52014-08-08 14:19:20 -07002355 struct mem_cgroup *memcg;
Hugh Dickins65500d22005-10-29 18:15:59 -07002356 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 pte_t pte;
Michel Lespinassed065bd82010-10-26 14:21:57 -07002358 int locked;
Rik van Rielad8c2ee2010-08-09 17:19:48 -07002359 int exclusive = 0;
Nick Piggin83c54072007-07-19 01:47:05 -07002360 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002362 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002363 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002364
2365 entry = pte_to_swp_entry(orig_pte);
Andi Kleend1737fd2009-09-16 11:50:06 +02002366 if (unlikely(non_swap_entry(entry))) {
2367 if (is_migration_entry(entry)) {
2368 migration_entry_wait(mm, pmd, address);
2369 } else if (is_hwpoison_entry(entry)) {
2370 ret = VM_FAULT_HWPOISON;
2371 } else {
2372 print_bad_pte(vma, address, orig_pte, NULL);
Hugh Dickinsd99be1a2009-12-14 17:59:04 -08002373 ret = VM_FAULT_SIGBUS;
Andi Kleend1737fd2009-09-16 11:50:06 +02002374 }
Christoph Lameter06972122006-06-23 02:03:35 -07002375 goto out;
2376 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002377 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 page = lookup_swap_cache(entry);
2379 if (!page) {
Hugh Dickins02098fe2008-02-04 22:28:42 -08002380 page = swapin_readahead(entry,
2381 GFP_HIGHUSER_MOVABLE, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 if (!page) {
2383 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002384 * Back out if somebody else faulted in this pte
2385 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002387 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 if (likely(pte_same(*page_table, orig_pte)))
2389 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002390 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002391 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 }
2393
2394 /* Had to read the page from swap area: Major fault */
2395 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002396 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07002397 mem_cgroup_count_vm_event(mm, PGMAJFAULT);
Andi Kleend1737fd2009-09-16 11:50:06 +02002398 } else if (PageHWPoison(page)) {
Wu Fengguang71f72522009-12-16 12:19:58 +01002399 /*
2400 * hwpoisoned dirty swapcache pages are kept for killing
2401 * owner processes (which may be unknown at hwpoison time)
2402 */
Andi Kleend1737fd2009-09-16 11:50:06 +02002403 ret = VM_FAULT_HWPOISON;
2404 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins56f31802013-02-22 16:36:10 -08002405 swapcache = page;
Andi Kleen4779cb32009-10-14 01:51:41 +02002406 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 }
2408
Hugh Dickins56f31802013-02-22 16:36:10 -08002409 swapcache = page;
Michel Lespinassed065bd82010-10-26 14:21:57 -07002410 locked = lock_page_or_retry(page, mm, flags);
Rik van Riele709ffd2012-05-29 15:06:18 -07002411
Balbir Singh20a10222007-11-14 17:00:33 -08002412 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Michel Lespinassed065bd82010-10-26 14:21:57 -07002413 if (!locked) {
2414 ret |= VM_FAULT_RETRY;
2415 goto out_release;
2416 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002418 /*
Hugh Dickins31c4a3d2010-09-19 19:40:22 -07002419 * Make sure try_to_free_swap or reuse_swap_page or swapoff did not
2420 * release the swapcache from under us. The page pin, and pte_same
2421 * test below, are not enough to exclude that. Even if it is still
2422 * swapcache, we need to check that the page's swap has not changed.
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002423 */
Hugh Dickins31c4a3d2010-09-19 19:40:22 -07002424 if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val))
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002425 goto out_page;
2426
Hugh Dickinscbf86cf2013-02-22 16:35:08 -08002427 page = ksm_might_need_to_copy(page, vma, address);
2428 if (unlikely(!page)) {
2429 ret = VM_FAULT_OOM;
2430 page = swapcache;
Hugh Dickinscbf86cf2013-02-22 16:35:08 -08002431 goto out_page;
Hugh Dickins5ad64682009-12-14 17:59:24 -08002432 }
2433
Johannes Weiner00501b52014-08-08 14:19:20 -07002434 if (mem_cgroup_try_charge(page, mm, GFP_KERNEL, &memcg)) {
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002435 ret = VM_FAULT_OOM;
Johannes Weinerbc43f752009-04-30 15:08:08 -07002436 goto out_page;
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002437 }
2438
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002440 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002442 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002443 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002444 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002445
2446 if (unlikely(!PageUptodate(page))) {
2447 ret = VM_FAULT_SIGBUS;
2448 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449 }
2450
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002451 /*
2452 * The page isn't present yet, go ahead with the fault.
2453 *
2454 * Be careful about the sequence of operations here.
2455 * To get its accounting right, reuse_swap_page() must be called
2456 * while the page is counted on swap but not yet in mapcount i.e.
2457 * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
2458 * must be called after the swap_free(), or it will never succeed.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002459 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08002461 inc_mm_counter_fast(mm, MM_ANONPAGES);
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08002462 dec_mm_counter_fast(mm, MM_SWAPENTS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 pte = mk_pte(page, vma->vm_page_prot);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002464 if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002466 flags &= ~FAULT_FLAG_WRITE;
Andrea Arcangeli9a5b4892010-08-09 17:19:49 -07002467 ret |= VM_FAULT_WRITE;
Rik van Rielad8c2ee2010-08-09 17:19:48 -07002468 exclusive = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 flush_icache_page(vma, page);
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07002471 if (pte_swp_soft_dirty(orig_pte))
2472 pte = pte_mksoft_dirty(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473 set_pte_at(mm, address, page_table, pte);
Johannes Weiner00501b52014-08-08 14:19:20 -07002474 if (page == swapcache) {
Johannes Weineraf347702013-02-22 16:32:20 -08002475 do_page_add_anon_rmap(page, vma, address, exclusive);
Johannes Weiner00501b52014-08-08 14:19:20 -07002476 mem_cgroup_commit_charge(page, memcg, true);
2477 } else { /* ksm created a completely new copy */
Hugh Dickins56f31802013-02-22 16:36:10 -08002478 page_add_new_anon_rmap(page, vma, address);
Johannes Weiner00501b52014-08-08 14:19:20 -07002479 mem_cgroup_commit_charge(page, memcg, false);
2480 lru_cache_add_active_or_unevictable(page, vma);
2481 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002483 swap_free(entry);
Nick Pigginb291f002008-10-18 20:26:44 -07002484 if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08002485 try_to_free_swap(page);
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002486 unlock_page(page);
Hugh Dickins56f31802013-02-22 16:36:10 -08002487 if (page != swapcache) {
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002488 /*
2489 * Hold the lock to avoid the swap entry to be reused
2490 * until we take the PT lock for the pte_same() check
2491 * (to avoid false positives from pte_same). For
2492 * further safety release the lock after the swap_free
2493 * so that the swap count won't change under a
2494 * parallel locked swapcache.
2495 */
2496 unlock_page(swapcache);
2497 page_cache_release(swapcache);
2498 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002499
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002500 if (flags & FAULT_FLAG_WRITE) {
Hugh Dickins61469f12008-03-04 14:29:04 -08002501 ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
2502 if (ret & VM_FAULT_ERROR)
2503 ret &= VM_FAULT_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 goto out;
2505 }
2506
2507 /* No need to invalidate - it was non-present before */
Russell King4b3073e2009-12-18 16:40:18 +00002508 update_mmu_cache(vma, address, page_table);
Hugh Dickins65500d22005-10-29 18:15:59 -07002509unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002510 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511out:
2512 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002513out_nomap:
Johannes Weiner00501b52014-08-08 14:19:20 -07002514 mem_cgroup_cancel_charge(page, memcg);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002515 pte_unmap_unlock(page_table, ptl);
Johannes Weinerbc43f752009-04-30 15:08:08 -07002516out_page:
Kirill Korotaevb8107482005-05-16 21:53:50 -07002517 unlock_page(page);
Andi Kleen4779cb32009-10-14 01:51:41 +02002518out_release:
Kirill Korotaevb8107482005-05-16 21:53:50 -07002519 page_cache_release(page);
Hugh Dickins56f31802013-02-22 16:36:10 -08002520 if (page != swapcache) {
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002521 unlock_page(swapcache);
2522 page_cache_release(swapcache);
2523 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002524 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525}
2526
2527/*
Luck, Tony8ca3eb02010-08-24 11:44:18 -07002528 * This is like a special single-page "expand_{down|up}wards()",
2529 * except we must first make sure that 'address{-|+}PAGE_SIZE'
Linus Torvalds320b2b82010-08-12 17:54:33 -07002530 * doesn't hit another vma.
Linus Torvalds320b2b82010-08-12 17:54:33 -07002531 */
2532static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address)
2533{
2534 address &= PAGE_MASK;
2535 if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) {
Linus Torvalds0e8e50e2010-08-20 16:49:40 -07002536 struct vm_area_struct *prev = vma->vm_prev;
Linus Torvalds320b2b82010-08-12 17:54:33 -07002537
Linus Torvalds0e8e50e2010-08-20 16:49:40 -07002538 /*
2539 * Is there a mapping abutting this one below?
2540 *
2541 * That's only ok if it's the same stack mapping
2542 * that has gotten split..
2543 */
2544 if (prev && prev->vm_end == address)
2545 return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM;
2546
Linus Torvaldsfee7e492015-01-06 13:00:05 -08002547 return expand_downwards(vma, address - PAGE_SIZE);
Linus Torvalds320b2b82010-08-12 17:54:33 -07002548 }
Luck, Tony8ca3eb02010-08-24 11:44:18 -07002549 if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) {
2550 struct vm_area_struct *next = vma->vm_next;
2551
2552 /* As VM_GROWSDOWN but s/below/above/ */
2553 if (next && next->vm_start == address + PAGE_SIZE)
2554 return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM;
2555
Linus Torvaldsfee7e492015-01-06 13:00:05 -08002556 return expand_upwards(vma, address + PAGE_SIZE);
Luck, Tony8ca3eb02010-08-24 11:44:18 -07002557 }
Linus Torvalds320b2b82010-08-12 17:54:33 -07002558 return 0;
2559}
2560
2561/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002562 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2563 * but allow concurrent faults), and pte mapped but not yet locked.
2564 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002566static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2567 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002568 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569{
Johannes Weiner00501b52014-08-08 14:19:20 -07002570 struct mem_cgroup *memcg;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002571 struct page *page;
2572 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574
Linus Torvalds11ac5522010-08-14 11:44:56 -07002575 pte_unmap(page_table);
Linus Torvalds320b2b82010-08-12 17:54:33 -07002576
Linus Torvalds11ac5522010-08-14 11:44:56 -07002577 /* Check if we need to add a guard page to the stack */
2578 if (check_stack_guard_page(vma, address) < 0)
Linus Torvalds9c145c562015-01-29 11:15:17 -08002579 return VM_FAULT_SIGSEGV;
Linus Torvalds11ac5522010-08-14 11:44:56 -07002580
2581 /* Use the zero-page for reads */
Dominik Dingel593befa2014-10-23 12:07:44 +02002582 if (!(flags & FAULT_FLAG_WRITE) && !mm_forbids_zeropage(mm)) {
Hugh Dickins62eede62009-09-21 17:03:34 -07002583 entry = pte_mkspecial(pfn_pte(my_zero_pfn(address),
2584 vma->vm_page_prot));
Linus Torvalds11ac5522010-08-14 11:44:56 -07002585 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002586 if (!pte_none(*page_table))
2587 goto unlock;
2588 goto setpte;
2589 }
2590
Nick Piggin557ed1f2007-10-16 01:24:40 -07002591 /* Allocate our own private page. */
Nick Piggin557ed1f2007-10-16 01:24:40 -07002592 if (unlikely(anon_vma_prepare(vma)))
2593 goto oom;
2594 page = alloc_zeroed_user_highpage_movable(vma, address);
2595 if (!page)
2596 goto oom;
Minchan Kim52f37622013-04-29 15:08:15 -07002597 /*
2598 * The memory barrier inside __SetPageUptodate makes sure that
2599 * preceeding stores to the page contents become visible before
2600 * the set_pte_at() write.
2601 */
Nick Piggin0ed361d2008-02-04 22:29:34 -08002602 __SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603
Johannes Weiner00501b52014-08-08 14:19:20 -07002604 if (mem_cgroup_try_charge(page, mm, GFP_KERNEL, &memcg))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002605 goto oom_free_page;
2606
Nick Piggin557ed1f2007-10-16 01:24:40 -07002607 entry = mk_pte(page, vma->vm_page_prot);
Hugh Dickins1ac0cb52009-09-21 17:03:29 -07002608 if (vma->vm_flags & VM_WRITE)
2609 entry = pte_mkwrite(pte_mkdirty(entry));
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002610
Nick Piggin557ed1f2007-10-16 01:24:40 -07002611 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Andrea Arcangeli1c2fb7a2009-09-21 17:02:22 -07002612 if (!pte_none(*page_table))
Nick Piggin557ed1f2007-10-16 01:24:40 -07002613 goto release;
Hugh Dickins9ba69292009-09-21 17:02:20 -07002614
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08002615 inc_mm_counter_fast(mm, MM_ANONPAGES);
Nick Piggin557ed1f2007-10-16 01:24:40 -07002616 page_add_new_anon_rmap(page, vma, address);
Johannes Weiner00501b52014-08-08 14:19:20 -07002617 mem_cgroup_commit_charge(page, memcg, false);
2618 lru_cache_add_active_or_unevictable(page, vma);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002619setpte:
Hugh Dickins65500d22005-10-29 18:15:59 -07002620 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621
2622 /* No need to invalidate - it was non-present before */
Russell King4b3073e2009-12-18 16:40:18 +00002623 update_mmu_cache(vma, address, page_table);
Hugh Dickins65500d22005-10-29 18:15:59 -07002624unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002625 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002626 return 0;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002627release:
Johannes Weiner00501b52014-08-08 14:19:20 -07002628 mem_cgroup_cancel_charge(page, memcg);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002629 page_cache_release(page);
2630 goto unlock;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002631oom_free_page:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002632 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002633oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 return VM_FAULT_OOM;
2635}
2636
Paul Cassella9a95f3c2014-08-06 16:07:24 -07002637/*
2638 * The mmap_sem must have been held on entry, and may have been
2639 * released depending on flags and vma->vm_ops->fault() return value.
2640 * See filemap_fault() and __lock_page_retry().
2641 */
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07002642static int __do_fault(struct vm_area_struct *vma, unsigned long address,
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08002643 pgoff_t pgoff, unsigned int flags,
2644 struct page *cow_page, struct page **page)
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07002645{
2646 struct vm_fault vmf;
2647 int ret;
2648
2649 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2650 vmf.pgoff = pgoff;
2651 vmf.flags = flags;
2652 vmf.page = NULL;
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08002653 vmf.cow_page = cow_page;
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07002654
2655 ret = vma->vm_ops->fault(vma, &vmf);
2656 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
2657 return ret;
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08002658 if (!vmf.page)
2659 goto out;
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07002660
2661 if (unlikely(PageHWPoison(vmf.page))) {
2662 if (ret & VM_FAULT_LOCKED)
2663 unlock_page(vmf.page);
2664 page_cache_release(vmf.page);
2665 return VM_FAULT_HWPOISON;
2666 }
2667
2668 if (unlikely(!(ret & VM_FAULT_LOCKED)))
2669 lock_page(vmf.page);
2670 else
2671 VM_BUG_ON_PAGE(!PageLocked(vmf.page), vmf.page);
2672
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08002673 out:
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07002674 *page = vmf.page;
2675 return ret;
2676}
2677
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002678/**
2679 * do_set_pte - setup new PTE entry for given page and add reverse page mapping.
2680 *
2681 * @vma: virtual memory area
2682 * @address: user virtual address
2683 * @page: page to map
2684 * @pte: pointer to target page table entry
2685 * @write: true, if new entry is writable
2686 * @anon: true, if it's anonymous page
2687 *
2688 * Caller must hold page table lock relevant for @pte.
2689 *
2690 * Target users are page handler itself and implementations of
2691 * vm_ops->map_pages.
2692 */
2693void do_set_pte(struct vm_area_struct *vma, unsigned long address,
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002694 struct page *page, pte_t *pte, bool write, bool anon)
2695{
2696 pte_t entry;
2697
2698 flush_icache_page(vma, page);
2699 entry = mk_pte(page, vma->vm_page_prot);
2700 if (write)
2701 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002702 if (anon) {
2703 inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES);
2704 page_add_new_anon_rmap(page, vma, address);
2705 } else {
2706 inc_mm_counter_fast(vma->vm_mm, MM_FILEPAGES);
2707 page_add_file_rmap(page);
2708 }
2709 set_pte_at(vma->vm_mm, address, pte, entry);
2710
2711 /* no need to invalidate: a not-present page won't be cached */
2712 update_mmu_cache(vma, address, pte);
2713}
2714
Kirill A. Shutemov3a910532014-08-06 16:08:07 -07002715static unsigned long fault_around_bytes __read_mostly =
2716 rounddown_pow_of_two(65536);
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002717
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002718#ifdef CONFIG_DEBUG_FS
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002719static int fault_around_bytes_get(void *data, u64 *val)
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002720{
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002721 *val = fault_around_bytes;
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002722 return 0;
2723}
2724
Andrey Ryabininb4903d62014-07-30 16:08:35 -07002725/*
2726 * fault_around_pages() and fault_around_mask() expects fault_around_bytes
2727 * rounded down to nearest page order. It's what do_fault_around() expects to
2728 * see.
2729 */
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002730static int fault_around_bytes_set(void *data, u64 val)
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002731{
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002732 if (val / PAGE_SIZE > PTRS_PER_PTE)
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002733 return -EINVAL;
Andrey Ryabininb4903d62014-07-30 16:08:35 -07002734 if (val > PAGE_SIZE)
2735 fault_around_bytes = rounddown_pow_of_two(val);
2736 else
2737 fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002738 return 0;
2739}
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002740DEFINE_SIMPLE_ATTRIBUTE(fault_around_bytes_fops,
2741 fault_around_bytes_get, fault_around_bytes_set, "%llu\n");
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002742
2743static int __init fault_around_debugfs(void)
2744{
2745 void *ret;
2746
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002747 ret = debugfs_create_file("fault_around_bytes", 0644, NULL, NULL,
2748 &fault_around_bytes_fops);
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002749 if (!ret)
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002750 pr_warn("Failed to create fault_around_bytes in debugfs");
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002751 return 0;
2752}
2753late_initcall(fault_around_debugfs);
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002754#endif
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002755
Kirill A. Shutemov1fdb4122014-06-04 16:10:55 -07002756/*
2757 * do_fault_around() tries to map few pages around the fault address. The hope
2758 * is that the pages will be needed soon and this will lower the number of
2759 * faults to handle.
2760 *
2761 * It uses vm_ops->map_pages() to map the pages, which skips the page if it's
2762 * not ready to be mapped: not up-to-date, locked, etc.
2763 *
2764 * This function is called with the page table lock taken. In the split ptlock
2765 * case the page table lock only protects only those entries which belong to
2766 * the page table corresponding to the fault address.
2767 *
2768 * This function doesn't cross the VMA boundaries, in order to call map_pages()
2769 * only once.
2770 *
2771 * fault_around_pages() defines how many pages we'll try to map.
2772 * do_fault_around() expects it to return a power of two less than or equal to
2773 * PTRS_PER_PTE.
2774 *
2775 * The virtual address of the area that we map is naturally aligned to the
2776 * fault_around_pages() value (and therefore to page order). This way it's
2777 * easier to guarantee that we don't cross page table boundaries.
2778 */
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002779static void do_fault_around(struct vm_area_struct *vma, unsigned long address,
2780 pte_t *pte, pgoff_t pgoff, unsigned int flags)
2781{
Kirill A. Shutemovaecd6f42014-08-06 16:08:05 -07002782 unsigned long start_addr, nr_pages, mask;
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002783 pgoff_t max_pgoff;
2784 struct vm_fault vmf;
2785 int off;
2786
Kirill A. Shutemovaecd6f42014-08-06 16:08:05 -07002787 nr_pages = ACCESS_ONCE(fault_around_bytes) >> PAGE_SHIFT;
2788 mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK;
2789
2790 start_addr = max(address & mask, vma->vm_start);
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002791 off = ((address - start_addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
2792 pte -= off;
2793 pgoff -= off;
2794
2795 /*
2796 * max_pgoff is either end of page table or end of vma
Kirill A. Shutemov850e9c62014-06-04 16:10:45 -07002797 * or fault_around_pages() from pgoff, depending what is nearest.
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002798 */
2799 max_pgoff = pgoff - ((start_addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) +
2800 PTRS_PER_PTE - 1;
2801 max_pgoff = min3(max_pgoff, vma_pages(vma) + vma->vm_pgoff - 1,
Kirill A. Shutemovaecd6f42014-08-06 16:08:05 -07002802 pgoff + nr_pages - 1);
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002803
2804 /* Check if it makes any sense to call ->map_pages */
2805 while (!pte_none(*pte)) {
2806 if (++pgoff > max_pgoff)
2807 return;
2808 start_addr += PAGE_SIZE;
2809 if (start_addr >= vma->vm_end)
2810 return;
2811 pte++;
2812 }
2813
2814 vmf.virtual_address = (void __user *) start_addr;
2815 vmf.pte = pte;
2816 vmf.pgoff = pgoff;
2817 vmf.max_pgoff = max_pgoff;
2818 vmf.flags = flags;
2819 vma->vm_ops->map_pages(vma, &vmf);
2820}
2821
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07002822static int do_read_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2823 unsigned long address, pmd_t *pmd,
2824 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
2825{
2826 struct page *fault_page;
2827 spinlock_t *ptl;
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002828 pte_t *pte;
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002829 int ret = 0;
2830
2831 /*
2832 * Let's call ->map_pages() first and use ->fault() as fallback
2833 * if page by the offset is not ready to be mapped (cold cache or
2834 * something).
2835 */
Kirill A. Shutemov9b4bdd22015-02-10 14:09:51 -08002836 if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) {
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002837 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
2838 do_fault_around(vma, address, pte, pgoff, flags);
2839 if (!pte_same(*pte, orig_pte))
2840 goto unlock_out;
2841 pte_unmap_unlock(pte, ptl);
2842 }
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07002843
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08002844 ret = __do_fault(vma, address, pgoff, flags, NULL, &fault_page);
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07002845 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
2846 return ret;
2847
2848 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
2849 if (unlikely(!pte_same(*pte, orig_pte))) {
2850 pte_unmap_unlock(pte, ptl);
2851 unlock_page(fault_page);
2852 page_cache_release(fault_page);
2853 return ret;
2854 }
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002855 do_set_pte(vma, address, fault_page, pte, false, false);
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07002856 unlock_page(fault_page);
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002857unlock_out:
2858 pte_unmap_unlock(pte, ptl);
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07002859 return ret;
2860}
2861
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07002862static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2863 unsigned long address, pmd_t *pmd,
2864 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
2865{
2866 struct page *fault_page, *new_page;
Johannes Weiner00501b52014-08-08 14:19:20 -07002867 struct mem_cgroup *memcg;
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07002868 spinlock_t *ptl;
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002869 pte_t *pte;
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07002870 int ret;
2871
2872 if (unlikely(anon_vma_prepare(vma)))
2873 return VM_FAULT_OOM;
2874
2875 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
2876 if (!new_page)
2877 return VM_FAULT_OOM;
2878
Johannes Weiner00501b52014-08-08 14:19:20 -07002879 if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg)) {
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07002880 page_cache_release(new_page);
2881 return VM_FAULT_OOM;
2882 }
2883
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08002884 ret = __do_fault(vma, address, pgoff, flags, new_page, &fault_page);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07002885 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
2886 goto uncharge_out;
2887
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08002888 if (fault_page)
2889 copy_user_highpage(new_page, fault_page, address, vma);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07002890 __SetPageUptodate(new_page);
2891
2892 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
2893 if (unlikely(!pte_same(*pte, orig_pte))) {
2894 pte_unmap_unlock(pte, ptl);
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08002895 if (fault_page) {
2896 unlock_page(fault_page);
2897 page_cache_release(fault_page);
2898 } else {
2899 /*
2900 * The fault handler has no page to lock, so it holds
2901 * i_mmap_lock for read to protect against truncate.
2902 */
2903 i_mmap_unlock_read(vma->vm_file->f_mapping);
2904 }
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07002905 goto uncharge_out;
2906 }
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002907 do_set_pte(vma, address, new_page, pte, true, true);
Johannes Weiner00501b52014-08-08 14:19:20 -07002908 mem_cgroup_commit_charge(new_page, memcg, false);
2909 lru_cache_add_active_or_unevictable(new_page, vma);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07002910 pte_unmap_unlock(pte, ptl);
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08002911 if (fault_page) {
2912 unlock_page(fault_page);
2913 page_cache_release(fault_page);
2914 } else {
2915 /*
2916 * The fault handler has no page to lock, so it holds
2917 * i_mmap_lock for read to protect against truncate.
2918 */
2919 i_mmap_unlock_read(vma->vm_file->f_mapping);
2920 }
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07002921 return ret;
2922uncharge_out:
Johannes Weiner00501b52014-08-08 14:19:20 -07002923 mem_cgroup_cancel_charge(new_page, memcg);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07002924 page_cache_release(new_page);
2925 return ret;
2926}
2927
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07002928static int do_shared_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002929 unsigned long address, pmd_t *pmd,
Nick Piggin54cb8822007-07-19 01:46:59 -07002930 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931{
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07002932 struct page *fault_page;
2933 struct address_space *mapping;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002934 spinlock_t *ptl;
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002935 pte_t *pte;
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07002936 int dirtied = 0;
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07002937 int ret, tmp;
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07002938
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08002939 ret = __do_fault(vma, address, pgoff, flags, NULL, &fault_page);
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07002940 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07002941 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942
2943 /*
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07002944 * Check if the backing address space wants to know that the page is
2945 * about to become writable
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946 */
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07002947 if (vma->vm_ops->page_mkwrite) {
2948 unlock_page(fault_page);
2949 tmp = do_page_mkwrite(vma, fault_page, address);
2950 if (unlikely(!tmp ||
2951 (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
2952 page_cache_release(fault_page);
2953 return tmp;
2954 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955 }
2956
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07002957 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
2958 if (unlikely(!pte_same(*pte, orig_pte))) {
2959 pte_unmap_unlock(pte, ptl);
2960 unlock_page(fault_page);
2961 page_cache_release(fault_page);
2962 return ret;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002963 }
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002964 do_set_pte(vma, address, fault_page, pte, true, false);
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07002965 pte_unmap_unlock(pte, ptl);
2966
2967 if (set_page_dirty(fault_page))
2968 dirtied = 1;
Andrew Mortond82fa872014-12-18 16:17:29 -08002969 /*
2970 * Take a local copy of the address_space - page.mapping may be zeroed
2971 * by truncate after unlock_page(). The address_space itself remains
2972 * pinned by vma->vm_file's reference. We rely on unlock_page()'s
2973 * release semantics to prevent the compiler from undoing this copying.
2974 */
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07002975 mapping = fault_page->mapping;
2976 unlock_page(fault_page);
2977 if ((dirtied || vma->vm_ops->page_mkwrite) && mapping) {
2978 /*
2979 * Some device drivers do not set page.mapping but still
2980 * dirty their pages
2981 */
2982 balance_dirty_pages_ratelimited(mapping);
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07002983 }
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07002984
Johannes Weiner74ec6752015-02-10 14:11:28 -08002985 if (!vma->vm_ops->page_mkwrite)
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07002986 file_update_time(vma->vm_file);
2987
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07002988 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07002989}
Nick Piggind00806b2007-07-19 01:46:57 -07002990
Paul Cassella9a95f3c2014-08-06 16:07:24 -07002991/*
2992 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2993 * but allow concurrent faults).
2994 * The mmap_sem may have been released depending on flags and our
2995 * return value. See filemap_fault() and __lock_page_or_retry().
2996 */
Kirill A. Shutemov9b4bdd22015-02-10 14:09:51 -08002997static int do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Nick Piggin54cb8822007-07-19 01:46:59 -07002998 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002999 unsigned int flags, pte_t orig_pte)
Nick Piggin54cb8822007-07-19 01:46:59 -07003000{
3001 pgoff_t pgoff = (((address & PAGE_MASK)
Dean Nelson0da7e012007-10-16 01:24:45 -07003002 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
Nick Piggin54cb8822007-07-19 01:46:59 -07003003
Hugh Dickins16abfa02007-10-04 16:56:06 +01003004 pte_unmap(page_table);
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07003005 if (!(flags & FAULT_FLAG_WRITE))
3006 return do_read_fault(mm, vma, address, pmd, pgoff, flags,
3007 orig_pte);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003008 if (!(vma->vm_flags & VM_SHARED))
3009 return do_cow_fault(mm, vma, address, pmd, pgoff, flags,
3010 orig_pte);
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003011 return do_shared_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Nick Piggin54cb8822007-07-19 01:46:59 -07003012}
3013
Rashika Kheriab19a9932014-04-03 14:48:02 -07003014static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
Rik van Riel04bb2f92013-10-07 11:29:36 +01003015 unsigned long addr, int page_nid,
3016 int *flags)
Mel Gorman9532fec2012-11-15 01:24:32 +00003017{
3018 get_page(page);
3019
3020 count_vm_numa_event(NUMA_HINT_FAULTS);
Rik van Riel04bb2f92013-10-07 11:29:36 +01003021 if (page_nid == numa_node_id()) {
Mel Gorman9532fec2012-11-15 01:24:32 +00003022 count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL);
Rik van Riel04bb2f92013-10-07 11:29:36 +01003023 *flags |= TNF_FAULT_LOCAL;
3024 }
Mel Gorman9532fec2012-11-15 01:24:32 +00003025
3026 return mpol_misplaced(page, vma, addr);
3027}
3028
Rashika Kheriab19a9932014-04-03 14:48:02 -07003029static int do_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,
Mel Gormand10e63f2012-10-25 14:16:31 +02003030 unsigned long addr, pte_t pte, pte_t *ptep, pmd_t *pmd)
3031{
Mel Gorman4daae3b2012-11-02 11:33:45 +00003032 struct page *page = NULL;
Mel Gormand10e63f2012-10-25 14:16:31 +02003033 spinlock_t *ptl;
Mel Gorman8191acb2013-10-07 11:28:45 +01003034 int page_nid = -1;
Peter Zijlstra90572892013-10-07 11:29:20 +01003035 int last_cpupid;
Peter Zijlstracbee9f82012-10-25 14:16:43 +02003036 int target_nid;
Mel Gormanb8593bf2012-11-21 01:18:23 +00003037 bool migrated = false;
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003038 int flags = 0;
Mel Gormand10e63f2012-10-25 14:16:31 +02003039
Mel Gormanc0e7cad2015-02-12 14:58:41 -08003040 /* A PROT_NONE fault should not end up here */
3041 BUG_ON(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)));
3042
Mel Gormand10e63f2012-10-25 14:16:31 +02003043 /*
3044 * The "pte" at this point cannot be used safely without
3045 * validation through pte_unmap_same(). It's of NUMA type but
3046 * the pfn may be screwed if the read is non atomic.
3047 *
Mel Gorman4d942462015-02-12 14:58:28 -08003048 * We can safely just do a "set_pte_at()", because the old
3049 * page table entry is not accessible, so there would be no
3050 * concurrent hardware modifications to the PTE.
Mel Gormand10e63f2012-10-25 14:16:31 +02003051 */
3052 ptl = pte_lockptr(mm, pmd);
3053 spin_lock(ptl);
Mel Gorman4daae3b2012-11-02 11:33:45 +00003054 if (unlikely(!pte_same(*ptep, pte))) {
3055 pte_unmap_unlock(ptep, ptl);
3056 goto out;
3057 }
3058
Mel Gorman4d942462015-02-12 14:58:28 -08003059 /* Make it present again */
3060 pte = pte_modify(pte, vma->vm_page_prot);
3061 pte = pte_mkyoung(pte);
Mel Gormand10e63f2012-10-25 14:16:31 +02003062 set_pte_at(mm, addr, ptep, pte);
3063 update_mmu_cache(vma, addr, ptep);
3064
3065 page = vm_normal_page(vma, addr, pte);
3066 if (!page) {
3067 pte_unmap_unlock(ptep, ptl);
3068 return 0;
3069 }
3070
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003071 /*
3072 * Avoid grouping on DSO/COW pages in specific and RO pages
3073 * in general, RO pages shouldn't hurt as much anyway since
3074 * they can be in shared cache state.
3075 */
3076 if (!pte_write(pte))
3077 flags |= TNF_NO_GROUP;
3078
Rik van Rieldabe1d92013-10-07 11:29:34 +01003079 /*
3080 * Flag if the page is shared between multiple address spaces. This
3081 * is later used when determining whether to group tasks together
3082 */
3083 if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED))
3084 flags |= TNF_SHARED;
3085
Peter Zijlstra90572892013-10-07 11:29:20 +01003086 last_cpupid = page_cpupid_last(page);
Mel Gorman8191acb2013-10-07 11:28:45 +01003087 page_nid = page_to_nid(page);
Rik van Riel04bb2f92013-10-07 11:29:36 +01003088 target_nid = numa_migrate_prep(page, vma, addr, page_nid, &flags);
Mel Gormand10e63f2012-10-25 14:16:31 +02003089 pte_unmap_unlock(ptep, ptl);
Mel Gorman4daae3b2012-11-02 11:33:45 +00003090 if (target_nid == -1) {
Mel Gorman4daae3b2012-11-02 11:33:45 +00003091 put_page(page);
3092 goto out;
3093 }
3094
3095 /* Migrate to the requested node */
Mel Gorman1bc115d2013-10-07 11:29:05 +01003096 migrated = migrate_misplaced_page(page, vma, target_nid);
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003097 if (migrated) {
Mel Gorman8191acb2013-10-07 11:28:45 +01003098 page_nid = target_nid;
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003099 flags |= TNF_MIGRATED;
3100 }
Mel Gorman4daae3b2012-11-02 11:33:45 +00003101
3102out:
Mel Gorman8191acb2013-10-07 11:28:45 +01003103 if (page_nid != -1)
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003104 task_numa_fault(last_cpupid, page_nid, 1, flags);
Mel Gormand10e63f2012-10-25 14:16:31 +02003105 return 0;
3106}
3107
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108/*
3109 * These routines also need to handle stuff like marking pages dirty
3110 * and/or accessed for architectures that don't do it in hardware (most
3111 * RISC architectures). The early dirtying is also good on the i386.
3112 *
3113 * There is also a hook called "update_mmu_cache()" that architectures
3114 * with external mmu caches can use to update those (ie the Sparc or
3115 * PowerPC hashed page tables that act as extended TLBs).
3116 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07003117 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3118 * but allow concurrent faults), and pte mapped but not yet locked.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07003119 * We return with pte unmapped and unlocked.
3120 *
3121 * The mmap_sem may have been released depending on flags and our
3122 * return value. See filemap_fault() and __lock_page_or_retry().
Linus Torvalds1da177e2005-04-16 15:20:36 -07003123 */
Kirill A. Shutemovc0292552013-09-12 15:14:05 -07003124static int handle_pte_fault(struct mm_struct *mm,
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003125 struct vm_area_struct *vma, unsigned long address,
3126 pte_t *pte, pmd_t *pmd, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003127{
3128 pte_t entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003129 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130
Christian Borntraegere37c6982014-12-07 21:41:33 +01003131 /*
3132 * some architectures can have larger ptes than wordsize,
3133 * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and CONFIG_32BIT=y,
3134 * so READ_ONCE or ACCESS_ONCE cannot guarantee atomic accesses.
3135 * The code below just needs a consistent view for the ifs and
3136 * we later double check anyway with the ptl lock held. So here
3137 * a barrier will do.
3138 */
3139 entry = *pte;
3140 barrier();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003141 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07003142 if (pte_none(entry)) {
Jes Sorensenf4b81802006-09-27 01:50:10 -07003143 if (vma->vm_ops) {
Nick Piggin3c18ddd2008-04-28 02:12:10 -07003144 if (likely(vma->vm_ops->fault))
Kirill A. Shutemov9b4bdd22015-02-10 14:09:51 -08003145 return do_fault(mm, vma, address, pte,
3146 pmd, flags, entry);
Jes Sorensenf4b81802006-09-27 01:50:10 -07003147 }
3148 return do_anonymous_page(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003149 pte, pmd, flags);
Hugh Dickins65500d22005-10-29 18:15:59 -07003150 }
Hugh Dickins65500d22005-10-29 18:15:59 -07003151 return do_swap_page(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003152 pte, pmd, flags, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003153 }
3154
Mel Gorman8a0516e2015-02-12 14:58:22 -08003155 if (pte_protnone(entry))
Mel Gormand10e63f2012-10-25 14:16:31 +02003156 return do_numa_page(mm, vma, address, entry, pte, pmd);
3157
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07003158 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003159 spin_lock(ptl);
3160 if (unlikely(!pte_same(*pte, entry)))
3161 goto unlock;
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003162 if (flags & FAULT_FLAG_WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003164 return do_wp_page(mm, vma, address,
3165 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003166 entry = pte_mkdirty(entry);
3167 }
3168 entry = pte_mkyoung(entry);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003169 if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) {
Russell King4b3073e2009-12-18 16:40:18 +00003170 update_mmu_cache(vma, address, pte);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07003171 } else {
3172 /*
3173 * This is needed only for protection faults but the arch code
3174 * is not yet telling us if this is a protection fault or not.
3175 * This still avoids useless tlb flushes for .text page faults
3176 * with threads.
3177 */
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003178 if (flags & FAULT_FLAG_WRITE)
Shaohua Li61c77322010-08-16 09:16:55 +08003179 flush_tlb_fix_spurious_fault(vma, address);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07003180 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003181unlock:
3182 pte_unmap_unlock(pte, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07003183 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003184}
3185
3186/*
3187 * By the time we get here, we already hold the mm semaphore
Paul Cassella9a95f3c2014-08-06 16:07:24 -07003188 *
3189 * The mmap_sem may have been released depending on flags and our
3190 * return value. See filemap_fault() and __lock_page_or_retry().
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191 */
Johannes Weiner519e5242013-09-12 15:13:42 -07003192static int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3193 unsigned long address, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003194{
3195 pgd_t *pgd;
3196 pud_t *pud;
3197 pmd_t *pmd;
3198 pte_t *pte;
3199
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01003200 if (unlikely(is_vm_hugetlb_page(vma)))
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003201 return hugetlb_fault(mm, vma, address, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003202
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003204 pud = pud_alloc(mm, pgd, address);
3205 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07003206 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003207 pmd = pmd_alloc(mm, pud, address);
3208 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07003209 return VM_FAULT_OOM;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003210 if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) {
Kirill A. Shutemovc0292552013-09-12 15:14:05 -07003211 int ret = VM_FAULT_FALLBACK;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003212 if (!vma->vm_ops)
Kirill A. Shutemovc0292552013-09-12 15:14:05 -07003213 ret = do_huge_pmd_anonymous_page(mm, vma, address,
3214 pmd, flags);
3215 if (!(ret & VM_FAULT_FALLBACK))
3216 return ret;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003217 } else {
3218 pmd_t orig_pmd = *pmd;
David Rientjes1f1d06c2012-05-29 15:06:23 -07003219 int ret;
3220
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003221 barrier();
3222 if (pmd_trans_huge(orig_pmd)) {
Will Deacona1dd4502012-12-11 16:01:27 -08003223 unsigned int dirty = flags & FAULT_FLAG_WRITE;
3224
Linus Torvaldse53289c2013-01-09 08:36:54 -08003225 /*
3226 * If the pmd is splitting, return and retry the
3227 * the fault. Alternative: wait until the split
3228 * is done, and goto retry.
3229 */
3230 if (pmd_trans_splitting(orig_pmd))
3231 return 0;
3232
Mel Gorman8a0516e2015-02-12 14:58:22 -08003233 if (pmd_protnone(orig_pmd))
Mel Gorman4daae3b2012-11-02 11:33:45 +00003234 return do_huge_pmd_numa_page(mm, vma, address,
Mel Gormand10e63f2012-10-25 14:16:31 +02003235 orig_pmd, pmd);
3236
Linus Torvalds3d59eeb2012-12-16 14:33:25 -08003237 if (dirty && !pmd_write(orig_pmd)) {
David Rientjes1f1d06c2012-05-29 15:06:23 -07003238 ret = do_huge_pmd_wp_page(mm, vma, address, pmd,
3239 orig_pmd);
Kirill A. Shutemov9845cbb2014-02-25 15:01:42 -08003240 if (!(ret & VM_FAULT_FALLBACK))
3241 return ret;
Will Deacona1dd4502012-12-11 16:01:27 -08003242 } else {
3243 huge_pmd_set_accessed(mm, vma, address, pmd,
3244 orig_pmd, dirty);
Kirill A. Shutemov9845cbb2014-02-25 15:01:42 -08003245 return 0;
David Rientjes1f1d06c2012-05-29 15:06:23 -07003246 }
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003247 }
3248 }
3249
3250 /*
3251 * Use __pte_alloc instead of pte_alloc_map, because we can't
3252 * run pte_offset_map on the pmd, if an huge pmd could
3253 * materialize from under us from a different thread.
3254 */
Mel Gorman4fd01772011-10-12 21:06:51 +02003255 if (unlikely(pmd_none(*pmd)) &&
3256 unlikely(__pte_alloc(mm, vma, pmd, address)))
Hugh Dickinsc74df322005-10-29 18:16:23 -07003257 return VM_FAULT_OOM;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003258 /* if an huge pmd materialized from under us just retry later */
3259 if (unlikely(pmd_trans_huge(*pmd)))
3260 return 0;
3261 /*
3262 * A regular pmd is established and it can't morph into a huge pmd
3263 * from under us anymore at this point because we hold the mmap_sem
3264 * read mode and khugepaged takes it in write mode. So now it's
3265 * safe to run pte_offset_map().
3266 */
3267 pte = pte_offset_map(pmd, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003268
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003269 return handle_pte_fault(mm, vma, address, pte, pmd, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003270}
3271
Paul Cassella9a95f3c2014-08-06 16:07:24 -07003272/*
3273 * By the time we get here, we already hold the mm semaphore
3274 *
3275 * The mmap_sem may have been released depending on flags and our
3276 * return value. See filemap_fault() and __lock_page_or_retry().
3277 */
Johannes Weiner519e5242013-09-12 15:13:42 -07003278int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3279 unsigned long address, unsigned int flags)
3280{
3281 int ret;
3282
3283 __set_current_state(TASK_RUNNING);
3284
3285 count_vm_event(PGFAULT);
3286 mem_cgroup_count_vm_event(mm, PGFAULT);
3287
3288 /* do counter updates before entering really critical section. */
3289 check_sync_rss_stat(current);
3290
3291 /*
3292 * Enable the memcg OOM handling for faults triggered in user
3293 * space. Kernel faults are handled more gracefully.
3294 */
3295 if (flags & FAULT_FLAG_USER)
Johannes Weiner49426422013-10-16 13:46:59 -07003296 mem_cgroup_oom_enable();
Johannes Weiner519e5242013-09-12 15:13:42 -07003297
3298 ret = __handle_mm_fault(mm, vma, address, flags);
3299
Johannes Weiner49426422013-10-16 13:46:59 -07003300 if (flags & FAULT_FLAG_USER) {
3301 mem_cgroup_oom_disable();
3302 /*
3303 * The task may have entered a memcg OOM situation but
3304 * if the allocation error was handled gracefully (no
3305 * VM_FAULT_OOM), there is no need to kill anything.
3306 * Just clean up the OOM state peacefully.
3307 */
3308 if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM))
3309 mem_cgroup_oom_synchronize(false);
3310 }
Johannes Weiner3812c8c2013-09-12 15:13:44 -07003311
Johannes Weiner519e5242013-09-12 15:13:42 -07003312 return ret;
3313}
Jesse Barnese1d6d012014-12-12 16:55:27 -08003314EXPORT_SYMBOL_GPL(handle_mm_fault);
Johannes Weiner519e5242013-09-12 15:13:42 -07003315
Linus Torvalds1da177e2005-04-16 15:20:36 -07003316#ifndef __PAGETABLE_PUD_FOLDED
3317/*
3318 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07003319 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003320 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07003321int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003322{
Hugh Dickinsc74df322005-10-29 18:16:23 -07003323 pud_t *new = pud_alloc_one(mm, address);
3324 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07003325 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003326
Nick Piggin362a61a2008-05-14 06:37:36 +02003327 smp_wmb(); /* See comment in __pte_alloc */
3328
Hugh Dickins872fec12005-10-29 18:16:21 -07003329 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003330 if (pgd_present(*pgd)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003331 pud_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003332 else
3333 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07003334 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003335 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003336}
3337#endif /* __PAGETABLE_PUD_FOLDED */
3338
3339#ifndef __PAGETABLE_PMD_FOLDED
3340/*
3341 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07003342 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003343 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07003344int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003345{
Hugh Dickinsc74df322005-10-29 18:16:23 -07003346 pmd_t *new = pmd_alloc_one(mm, address);
3347 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07003348 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349
Nick Piggin362a61a2008-05-14 06:37:36 +02003350 smp_wmb(); /* See comment in __pte_alloc */
3351
Hugh Dickins872fec12005-10-29 18:16:21 -07003352 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003353#ifndef __ARCH_HAS_4LEVEL_HACK
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -08003354 if (!pud_present(*pud)) {
3355 mm_inc_nr_pmds(mm);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003356 pud_populate(mm, pud, new);
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -08003357 } else /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003358 pmd_free(mm, new);
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -08003359#else
3360 if (!pgd_present(*pud)) {
3361 mm_inc_nr_pmds(mm);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003362 pgd_populate(mm, pud, new);
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -08003363 } else /* Another has populated it */
3364 pmd_free(mm, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003365#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07003366 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003367 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003368}
3369#endif /* __PAGETABLE_PMD_FOLDED */
3370
Namhyung Kim1b36ba82010-10-26 14:22:00 -07003371static int __follow_pte(struct mm_struct *mm, unsigned long address,
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07003372 pte_t **ptepp, spinlock_t **ptlp)
3373{
3374 pgd_t *pgd;
3375 pud_t *pud;
3376 pmd_t *pmd;
3377 pte_t *ptep;
3378
3379 pgd = pgd_offset(mm, address);
3380 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
3381 goto out;
3382
3383 pud = pud_offset(pgd, address);
3384 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
3385 goto out;
3386
3387 pmd = pmd_offset(pud, address);
Andrea Arcangelif66055ab2011-01-13 15:46:54 -08003388 VM_BUG_ON(pmd_trans_huge(*pmd));
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07003389 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
3390 goto out;
3391
3392 /* We cannot handle huge page PFN maps. Luckily they don't exist. */
3393 if (pmd_huge(*pmd))
3394 goto out;
3395
3396 ptep = pte_offset_map_lock(mm, pmd, address, ptlp);
3397 if (!ptep)
3398 goto out;
3399 if (!pte_present(*ptep))
3400 goto unlock;
3401 *ptepp = ptep;
3402 return 0;
3403unlock:
3404 pte_unmap_unlock(ptep, *ptlp);
3405out:
3406 return -EINVAL;
3407}
3408
Namhyung Kim1b36ba82010-10-26 14:22:00 -07003409static inline int follow_pte(struct mm_struct *mm, unsigned long address,
3410 pte_t **ptepp, spinlock_t **ptlp)
3411{
3412 int res;
3413
3414 /* (void) is needed to make gcc happy */
3415 (void) __cond_lock(*ptlp,
3416 !(res = __follow_pte(mm, address, ptepp, ptlp)));
3417 return res;
3418}
3419
Johannes Weiner3b6748e2009-06-16 15:32:35 -07003420/**
3421 * follow_pfn - look up PFN at a user virtual address
3422 * @vma: memory mapping
3423 * @address: user virtual address
3424 * @pfn: location to store found PFN
3425 *
3426 * Only IO mappings and raw PFN mappings are allowed.
3427 *
3428 * Returns zero and the pfn at @pfn on success, -ve otherwise.
3429 */
3430int follow_pfn(struct vm_area_struct *vma, unsigned long address,
3431 unsigned long *pfn)
3432{
3433 int ret = -EINVAL;
3434 spinlock_t *ptl;
3435 pte_t *ptep;
3436
3437 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3438 return ret;
3439
3440 ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);
3441 if (ret)
3442 return ret;
3443 *pfn = pte_pfn(*ptep);
3444 pte_unmap_unlock(ptep, ptl);
3445 return 0;
3446}
3447EXPORT_SYMBOL(follow_pfn);
3448
Rik van Riel28b2ee22008-07-23 21:27:05 -07003449#ifdef CONFIG_HAVE_IOREMAP_PROT
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003450int follow_phys(struct vm_area_struct *vma,
3451 unsigned long address, unsigned int flags,
3452 unsigned long *prot, resource_size_t *phys)
Rik van Riel28b2ee22008-07-23 21:27:05 -07003453{
Johannes Weiner03668a42009-06-16 15:32:34 -07003454 int ret = -EINVAL;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003455 pte_t *ptep, pte;
3456 spinlock_t *ptl;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003457
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003458 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3459 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003460
Johannes Weiner03668a42009-06-16 15:32:34 -07003461 if (follow_pte(vma->vm_mm, address, &ptep, &ptl))
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003462 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003463 pte = *ptep;
Johannes Weiner03668a42009-06-16 15:32:34 -07003464
Rik van Riel28b2ee22008-07-23 21:27:05 -07003465 if ((flags & FOLL_WRITE) && !pte_write(pte))
3466 goto unlock;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003467
3468 *prot = pgprot_val(pte_pgprot(pte));
Johannes Weiner03668a42009-06-16 15:32:34 -07003469 *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003470
Johannes Weiner03668a42009-06-16 15:32:34 -07003471 ret = 0;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003472unlock:
3473 pte_unmap_unlock(ptep, ptl);
3474out:
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003475 return ret;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003476}
3477
3478int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
3479 void *buf, int len, int write)
3480{
3481 resource_size_t phys_addr;
3482 unsigned long prot = 0;
KOSAKI Motohiro2bc72732009-01-06 14:39:43 -08003483 void __iomem *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003484 int offset = addr & (PAGE_SIZE-1);
3485
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003486 if (follow_phys(vma, addr, write, &prot, &phys_addr))
Rik van Riel28b2ee22008-07-23 21:27:05 -07003487 return -EINVAL;
3488
Grazvydas Ignotas9cb12d72015-02-12 15:00:19 -08003489 maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot);
Rik van Riel28b2ee22008-07-23 21:27:05 -07003490 if (write)
3491 memcpy_toio(maddr + offset, buf, len);
3492 else
3493 memcpy_fromio(buf, maddr + offset, len);
3494 iounmap(maddr);
3495
3496 return len;
3497}
Uwe Kleine-König5a736332013-08-07 13:02:52 +02003498EXPORT_SYMBOL_GPL(generic_access_phys);
Rik van Riel28b2ee22008-07-23 21:27:05 -07003499#endif
3500
David Howells0ec76a12006-09-27 01:50:15 -07003501/*
Stephen Wilson206cb632011-03-13 15:49:19 -04003502 * Access another process' address space as given in mm. If non-NULL, use the
3503 * given task for page fault accounting.
David Howells0ec76a12006-09-27 01:50:15 -07003504 */
Stephen Wilson206cb632011-03-13 15:49:19 -04003505static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
3506 unsigned long addr, void *buf, int len, int write)
David Howells0ec76a12006-09-27 01:50:15 -07003507{
David Howells0ec76a12006-09-27 01:50:15 -07003508 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07003509 void *old_buf = buf;
3510
David Howells0ec76a12006-09-27 01:50:15 -07003511 down_read(&mm->mmap_sem);
Simon Arlott183ff222007-10-20 01:27:18 +02003512 /* ignore errors, just check how much was successfully transferred */
David Howells0ec76a12006-09-27 01:50:15 -07003513 while (len) {
3514 int bytes, ret, offset;
3515 void *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003516 struct page *page = NULL;
David Howells0ec76a12006-09-27 01:50:15 -07003517
3518 ret = get_user_pages(tsk, mm, addr, 1,
3519 write, 1, &page, &vma);
Rik van Riel28b2ee22008-07-23 21:27:05 -07003520 if (ret <= 0) {
Rik van Rieldbffcd02014-08-06 16:08:12 -07003521#ifndef CONFIG_HAVE_IOREMAP_PROT
3522 break;
3523#else
Rik van Riel28b2ee22008-07-23 21:27:05 -07003524 /*
3525 * Check if this is a VM_IO | VM_PFNMAP VMA, which
3526 * we can access using slightly different code.
3527 */
Rik van Riel28b2ee22008-07-23 21:27:05 -07003528 vma = find_vma(mm, addr);
Michael Ellermanfe936df2011-04-14 15:22:10 -07003529 if (!vma || vma->vm_start > addr)
Rik van Riel28b2ee22008-07-23 21:27:05 -07003530 break;
3531 if (vma->vm_ops && vma->vm_ops->access)
3532 ret = vma->vm_ops->access(vma, addr, buf,
3533 len, write);
3534 if (ret <= 0)
Rik van Riel28b2ee22008-07-23 21:27:05 -07003535 break;
3536 bytes = ret;
Rik van Rieldbffcd02014-08-06 16:08:12 -07003537#endif
David Howells0ec76a12006-09-27 01:50:15 -07003538 } else {
Rik van Riel28b2ee22008-07-23 21:27:05 -07003539 bytes = len;
3540 offset = addr & (PAGE_SIZE-1);
3541 if (bytes > PAGE_SIZE-offset)
3542 bytes = PAGE_SIZE-offset;
3543
3544 maddr = kmap(page);
3545 if (write) {
3546 copy_to_user_page(vma, page, addr,
3547 maddr + offset, buf, bytes);
3548 set_page_dirty_lock(page);
3549 } else {
3550 copy_from_user_page(vma, page, addr,
3551 buf, maddr + offset, bytes);
3552 }
3553 kunmap(page);
3554 page_cache_release(page);
David Howells0ec76a12006-09-27 01:50:15 -07003555 }
David Howells0ec76a12006-09-27 01:50:15 -07003556 len -= bytes;
3557 buf += bytes;
3558 addr += bytes;
3559 }
3560 up_read(&mm->mmap_sem);
David Howells0ec76a12006-09-27 01:50:15 -07003561
3562 return buf - old_buf;
3563}
Andi Kleen03252912008-01-30 13:33:18 +01003564
Stephen Wilson5ddd36b2011-03-13 15:49:20 -04003565/**
Randy Dunlapae91dbf2011-03-26 13:27:01 -07003566 * access_remote_vm - access another process' address space
Stephen Wilson5ddd36b2011-03-13 15:49:20 -04003567 * @mm: the mm_struct of the target address space
3568 * @addr: start address to access
3569 * @buf: source or destination buffer
3570 * @len: number of bytes to transfer
3571 * @write: whether the access is a write
3572 *
3573 * The caller must hold a reference on @mm.
3574 */
3575int access_remote_vm(struct mm_struct *mm, unsigned long addr,
3576 void *buf, int len, int write)
3577{
3578 return __access_remote_vm(NULL, mm, addr, buf, len, write);
3579}
3580
Andi Kleen03252912008-01-30 13:33:18 +01003581/*
Stephen Wilson206cb632011-03-13 15:49:19 -04003582 * Access another process' address space.
3583 * Source/target buffer must be kernel space,
3584 * Do not walk the page table directly, use get_user_pages
3585 */
3586int access_process_vm(struct task_struct *tsk, unsigned long addr,
3587 void *buf, int len, int write)
3588{
3589 struct mm_struct *mm;
3590 int ret;
3591
3592 mm = get_task_mm(tsk);
3593 if (!mm)
3594 return 0;
3595
3596 ret = __access_remote_vm(tsk, mm, addr, buf, len, write);
3597 mmput(mm);
3598
3599 return ret;
3600}
3601
Andi Kleen03252912008-01-30 13:33:18 +01003602/*
3603 * Print the name of a VMA.
3604 */
3605void print_vma_addr(char *prefix, unsigned long ip)
3606{
3607 struct mm_struct *mm = current->mm;
3608 struct vm_area_struct *vma;
3609
Ingo Molnare8bff742008-02-13 20:21:06 +01003610 /*
3611 * Do not print if we are in atomic
3612 * contexts (in exception stacks, etc.):
3613 */
3614 if (preempt_count())
3615 return;
3616
Andi Kleen03252912008-01-30 13:33:18 +01003617 down_read(&mm->mmap_sem);
3618 vma = find_vma(mm, ip);
3619 if (vma && vma->vm_file) {
3620 struct file *f = vma->vm_file;
3621 char *buf = (char *)__get_free_page(GFP_KERNEL);
3622 if (buf) {
Andy Shevchenko2fbc57c2012-12-17 16:01:23 -08003623 char *p;
Andi Kleen03252912008-01-30 13:33:18 +01003624
Jan Blunckcf28b482008-02-14 19:38:44 -08003625 p = d_path(&f->f_path, buf, PAGE_SIZE);
Andi Kleen03252912008-01-30 13:33:18 +01003626 if (IS_ERR(p))
3627 p = "?";
Andy Shevchenko2fbc57c2012-12-17 16:01:23 -08003628 printk("%s%s[%lx+%lx]", prefix, kbasename(p),
Andi Kleen03252912008-01-30 13:33:18 +01003629 vma->vm_start,
3630 vma->vm_end - vma->vm_start);
3631 free_page((unsigned long)buf);
3632 }
3633 }
Jeff Liu51a07e52012-07-31 16:43:18 -07003634 up_read(&mm->mmap_sem);
Andi Kleen03252912008-01-30 13:33:18 +01003635}
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003636
Michael S. Tsirkin662bbcb2013-05-26 17:32:23 +03003637#if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP)
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003638void might_fault(void)
3639{
Peter Zijlstra95156f02009-01-12 13:02:11 +01003640 /*
3641 * Some code (nfs/sunrpc) uses socket ops on kernel memory while
3642 * holding the mmap_sem, this is safe because kernel memory doesn't
3643 * get paged out, therefore we'll never actually fault, and the
3644 * below annotations will generate false positives.
3645 */
3646 if (segment_eq(get_fs(), KERNEL_DS))
3647 return;
3648
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003649 /*
3650 * it would be nicer only to annotate paths which are not under
3651 * pagefault_disable, however that requires a larger audit and
3652 * providing helpers like get_user_atomic.
3653 */
Michael S. Tsirkin662bbcb2013-05-26 17:32:23 +03003654 if (in_atomic())
3655 return;
3656
3657 __might_sleep(__FILE__, __LINE__, 0);
3658
3659 if (current->mm)
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003660 might_lock_read(&current->mm->mmap_sem);
3661}
3662EXPORT_SYMBOL(might_fault);
3663#endif
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08003664
3665#if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)
3666static void clear_gigantic_page(struct page *page,
3667 unsigned long addr,
3668 unsigned int pages_per_huge_page)
3669{
3670 int i;
3671 struct page *p = page;
3672
3673 might_sleep();
3674 for (i = 0; i < pages_per_huge_page;
3675 i++, p = mem_map_next(p, page, i)) {
3676 cond_resched();
3677 clear_user_highpage(p, addr + i * PAGE_SIZE);
3678 }
3679}
3680void clear_huge_page(struct page *page,
3681 unsigned long addr, unsigned int pages_per_huge_page)
3682{
3683 int i;
3684
3685 if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
3686 clear_gigantic_page(page, addr, pages_per_huge_page);
3687 return;
3688 }
3689
3690 might_sleep();
3691 for (i = 0; i < pages_per_huge_page; i++) {
3692 cond_resched();
3693 clear_user_highpage(page + i, addr + i * PAGE_SIZE);
3694 }
3695}
3696
3697static void copy_user_gigantic_page(struct page *dst, struct page *src,
3698 unsigned long addr,
3699 struct vm_area_struct *vma,
3700 unsigned int pages_per_huge_page)
3701{
3702 int i;
3703 struct page *dst_base = dst;
3704 struct page *src_base = src;
3705
3706 for (i = 0; i < pages_per_huge_page; ) {
3707 cond_resched();
3708 copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma);
3709
3710 i++;
3711 dst = mem_map_next(dst, dst_base, i);
3712 src = mem_map_next(src, src_base, i);
3713 }
3714}
3715
3716void copy_user_huge_page(struct page *dst, struct page *src,
3717 unsigned long addr, struct vm_area_struct *vma,
3718 unsigned int pages_per_huge_page)
3719{
3720 int i;
3721
3722 if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
3723 copy_user_gigantic_page(dst, src, addr, vma,
3724 pages_per_huge_page);
3725 return;
3726 }
3727
3728 might_sleep();
3729 for (i = 0; i < pages_per_huge_page; i++) {
3730 cond_resched();
3731 copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma);
3732 }
3733}
3734#endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003735
Olof Johansson40b64ac2013-12-20 14:28:05 -08003736#if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS
Kirill A. Shutemovb35f1812014-01-21 15:49:07 -08003737
3738static struct kmem_cache *page_ptl_cachep;
3739
3740void __init ptlock_cache_init(void)
3741{
3742 page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0,
3743 SLAB_PANIC, NULL);
3744}
3745
Peter Zijlstra539edb52013-11-14 14:31:52 -08003746bool ptlock_alloc(struct page *page)
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003747{
3748 spinlock_t *ptl;
3749
Kirill A. Shutemovb35f1812014-01-21 15:49:07 -08003750 ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL);
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003751 if (!ptl)
3752 return false;
Peter Zijlstra539edb52013-11-14 14:31:52 -08003753 page->ptl = ptl;
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003754 return true;
3755}
3756
Peter Zijlstra539edb52013-11-14 14:31:52 -08003757void ptlock_free(struct page *page)
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003758{
Kirill A. Shutemovb35f1812014-01-21 15:49:07 -08003759 kmem_cache_free(page_ptl_cachep, page->ptl);
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003760}
3761#endif