blob: 15322b73636b4272d6c84af6742a0822822301db [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>
Dan Williams01c8f1c2016-01-15 16:56:40 -080053#include <linux/pfn_t.h>
Peter Zijlstraedc79b22006-09-25 23:30:58 -070054#include <linux/writeback.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080055#include <linux/memcontrol.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070056#include <linux/mmu_notifier.h>
Hugh Dickins3dc14742009-01-06 14:40:08 -080057#include <linux/kallsyms.h>
58#include <linux/swapops.h>
59#include <linux/elf.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090060#include <linux/gfp.h>
Mel Gorman4daae3b2012-11-02 11:33:45 +000061#include <linux/migrate.h>
Andy Shevchenko2fbc57c2012-12-17 16:01:23 -080062#include <linux/string.h>
Dan Williams0abdd7a2014-01-21 15:48:12 -080063#include <linux/dma-debug.h>
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -070064#include <linux/debugfs.h>
Andrea Arcangeli6b251fc2015-09-04 15:46:20 -070065#include <linux/userfaultfd_k.h>
Jan Karabc2466e2016-05-12 18:29:19 +020066#include <linux/dax.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070067
Alexey Dobriyan6952b612009-09-18 23:55:55 +040068#include <asm/io.h>
Dave Hansen33a709b2016-02-12 13:02:19 -080069#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#include <asm/pgalloc.h>
71#include <asm/uaccess.h>
72#include <asm/tlb.h>
73#include <asm/tlbflush.h>
74#include <asm/pgtable.h>
75
Jan Beulich42b77722008-07-23 21:27:10 -070076#include "internal.h"
77
Peter Zijlstra90572892013-10-07 11:29:20 +010078#ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
79#warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid.
Peter Zijlstra75980e92013-02-22 16:34:32 -080080#endif
81
Andy Whitcroftd41dee32005-06-23 00:07:54 -070082#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -070083/* use the per-pgdat data instead for discontigmem - mbligh */
84unsigned long max_mapnr;
85struct page *mem_map;
86
87EXPORT_SYMBOL(max_mapnr);
88EXPORT_SYMBOL(mem_map);
89#endif
90
Linus Torvalds1da177e2005-04-16 15:20:36 -070091/*
92 * A number of key systems in x86 including ioremap() rely on the assumption
93 * that high_memory defines the upper bound on direct map memory, then end
94 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
95 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
96 * and ZONE_HIGHMEM.
97 */
98void * high_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
Ingo Molnar32a93232008-02-06 22:39:44 +0100102/*
103 * Randomize the address space (stacks, mmaps, brk, etc.).
104 *
105 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
106 * as ancient (libc5 based) binaries can segfault. )
107 */
108int randomize_va_space __read_mostly =
109#ifdef CONFIG_COMPAT_BRK
110 1;
111#else
112 2;
113#endif
Andi Kleena62eaf12006-02-16 23:41:58 +0100114
115static int __init disable_randmaps(char *s)
116{
117 randomize_va_space = 0;
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800118 return 1;
Andi Kleena62eaf12006-02-16 23:41:58 +0100119}
120__setup("norandmaps", disable_randmaps);
121
Hugh Dickins62eede62009-09-21 17:03:34 -0700122unsigned long zero_pfn __read_mostly;
Hugh Dickins03f64622009-09-21 17:03:35 -0700123unsigned long highest_memmap_pfn __read_mostly;
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -0700124
Ard Biesheuvel0b700682014-09-12 22:17:23 +0200125EXPORT_SYMBOL(zero_pfn);
126
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -0700127/*
128 * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init()
129 */
130static int __init init_zero_pfn(void)
131{
132 zero_pfn = page_to_pfn(ZERO_PAGE(0));
133 return 0;
134}
135core_initcall(init_zero_pfn);
Andi Kleena62eaf12006-02-16 23:41:58 +0100136
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800137
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800138#if defined(SPLIT_RSS_COUNTING)
139
David Rientjesea48cf72012-03-21 16:34:13 -0700140void sync_mm_rss(struct mm_struct *mm)
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800141{
142 int i;
143
144 for (i = 0; i < NR_MM_COUNTERS; i++) {
David Rientjes05af2e12012-03-21 16:34:13 -0700145 if (current->rss_stat.count[i]) {
146 add_mm_counter(mm, i, current->rss_stat.count[i]);
147 current->rss_stat.count[i] = 0;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800148 }
149 }
David Rientjes05af2e12012-03-21 16:34:13 -0700150 current->rss_stat.events = 0;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800151}
152
153static void add_mm_counter_fast(struct mm_struct *mm, int member, int val)
154{
155 struct task_struct *task = current;
156
157 if (likely(task->mm == mm))
158 task->rss_stat.count[member] += val;
159 else
160 add_mm_counter(mm, member, val);
161}
162#define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1)
163#define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1)
164
165/* sync counter once per 64 page faults */
166#define TASK_RSS_EVENTS_THRESH (64)
167static void check_sync_rss_stat(struct task_struct *task)
168{
169 if (unlikely(task != current))
170 return;
171 if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH))
David Rientjesea48cf72012-03-21 16:34:13 -0700172 sync_mm_rss(task->mm);
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800173}
Peter Zijlstra9547d012011-05-24 17:12:14 -0700174#else /* SPLIT_RSS_COUNTING */
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800175
176#define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member)
177#define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member)
178
179static void check_sync_rss_stat(struct task_struct *task)
180{
181}
182
Peter Zijlstra9547d012011-05-24 17:12:14 -0700183#endif /* SPLIT_RSS_COUNTING */
184
185#ifdef HAVE_GENERIC_MMU_GATHER
186
Nicholas Krauseca1d6c72015-09-04 15:48:22 -0700187static bool tlb_next_batch(struct mmu_gather *tlb)
Peter Zijlstra9547d012011-05-24 17:12:14 -0700188{
189 struct mmu_gather_batch *batch;
190
191 batch = tlb->active;
192 if (batch->next) {
193 tlb->active = batch->next;
Nicholas Krauseca1d6c72015-09-04 15:48:22 -0700194 return true;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700195 }
196
Michal Hocko53a59fc2013-01-04 15:35:12 -0800197 if (tlb->batch_count == MAX_GATHER_BATCH_COUNT)
Nicholas Krauseca1d6c72015-09-04 15:48:22 -0700198 return false;
Michal Hocko53a59fc2013-01-04 15:35:12 -0800199
Peter Zijlstra9547d012011-05-24 17:12:14 -0700200 batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0);
201 if (!batch)
Nicholas Krauseca1d6c72015-09-04 15:48:22 -0700202 return false;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700203
Michal Hocko53a59fc2013-01-04 15:35:12 -0800204 tlb->batch_count++;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700205 batch->next = NULL;
206 batch->nr = 0;
207 batch->max = MAX_GATHER_BATCH;
208
209 tlb->active->next = batch;
210 tlb->active = batch;
211
Nicholas Krauseca1d6c72015-09-04 15:48:22 -0700212 return true;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700213}
214
215/* tlb_gather_mmu
216 * Called to initialize an (on-stack) mmu_gather structure for page-table
217 * tear-down from @mm. The @fullmm argument is used when @mm is without
218 * users and we're going to destroy the full address space (exit/execve).
219 */
Linus Torvalds2b047252013-08-15 11:42:25 -0700220void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, unsigned long start, unsigned long end)
Peter Zijlstra9547d012011-05-24 17:12:14 -0700221{
222 tlb->mm = mm;
223
Linus Torvalds2b047252013-08-15 11:42:25 -0700224 /* Is it from 0 to ~0? */
225 tlb->fullmm = !(start | (end+1));
Dave Hansen1de14c32013-04-12 16:23:54 -0700226 tlb->need_flush_all = 0;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700227 tlb->local.next = NULL;
228 tlb->local.nr = 0;
229 tlb->local.max = ARRAY_SIZE(tlb->__pages);
230 tlb->active = &tlb->local;
Michal Hocko53a59fc2013-01-04 15:35:12 -0800231 tlb->batch_count = 0;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700232
233#ifdef CONFIG_HAVE_RCU_TABLE_FREE
234 tlb->batch = NULL;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800235#endif
Will Deaconfb7332a2014-10-29 10:03:09 +0000236
237 __tlb_reset_range(tlb);
Peter Zijlstra9547d012011-05-24 17:12:14 -0700238}
239
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700240static void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb)
Peter Zijlstra9547d012011-05-24 17:12:14 -0700241{
Will Deacon721c21c2015-01-12 19:10:55 +0000242 if (!tlb->end)
243 return;
244
Peter Zijlstra9547d012011-05-24 17:12:14 -0700245 tlb_flush(tlb);
Joerg Roedel34ee6452014-11-13 13:46:09 +1100246 mmu_notifier_invalidate_range(tlb->mm, tlb->start, tlb->end);
Peter Zijlstra9547d012011-05-24 17:12:14 -0700247#ifdef CONFIG_HAVE_RCU_TABLE_FREE
248 tlb_table_flush(tlb);
249#endif
Will Deaconfb7332a2014-10-29 10:03:09 +0000250 __tlb_reset_range(tlb);
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700251}
252
253static void tlb_flush_mmu_free(struct mmu_gather *tlb)
254{
255 struct mmu_gather_batch *batch;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700256
Will Deacon721c21c2015-01-12 19:10:55 +0000257 for (batch = &tlb->local; batch && batch->nr; batch = batch->next) {
Peter Zijlstra9547d012011-05-24 17:12:14 -0700258 free_pages_and_swap_cache(batch->pages, batch->nr);
259 batch->nr = 0;
260 }
261 tlb->active = &tlb->local;
262}
263
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700264void tlb_flush_mmu(struct mmu_gather *tlb)
265{
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700266 tlb_flush_mmu_tlbonly(tlb);
267 tlb_flush_mmu_free(tlb);
268}
269
Peter Zijlstra9547d012011-05-24 17:12:14 -0700270/* tlb_finish_mmu
271 * Called at the end of the shootdown operation to free up any resources
272 * that were required.
273 */
274void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end)
275{
276 struct mmu_gather_batch *batch, *next;
277
278 tlb_flush_mmu(tlb);
279
280 /* keep the page table cache within bounds */
281 check_pgt_cache();
282
283 for (batch = tlb->local.next; batch; batch = next) {
284 next = batch->next;
285 free_pages((unsigned long)batch, 0);
286 }
287 tlb->local.next = NULL;
288}
289
290/* __tlb_remove_page
291 * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while
292 * handling the additional races in SMP caused by other CPUs caching valid
293 * mappings in their TLBs. Returns the number of free page slots left.
294 * When out of page slots we must call tlb_flush_mmu().
295 */
296int __tlb_remove_page(struct mmu_gather *tlb, struct page *page)
297{
298 struct mmu_gather_batch *batch;
299
Will Deaconfb7332a2014-10-29 10:03:09 +0000300 VM_BUG_ON(!tlb->end);
Peter Zijlstra9547d012011-05-24 17:12:14 -0700301
Peter Zijlstra9547d012011-05-24 17:12:14 -0700302 batch = tlb->active;
303 batch->pages[batch->nr++] = page;
304 if (batch->nr == batch->max) {
305 if (!tlb_next_batch(tlb))
306 return 0;
Shaohua Li0b43c3a2011-07-08 15:39:41 -0700307 batch = tlb->active;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700308 }
Sasha Levin309381fea2014-01-23 15:52:54 -0800309 VM_BUG_ON_PAGE(batch->nr > batch->max, page);
Peter Zijlstra9547d012011-05-24 17:12:14 -0700310
311 return batch->max - batch->nr;
312}
313
314#endif /* HAVE_GENERIC_MMU_GATHER */
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800315
Peter Zijlstra26723912011-05-24 17:12:00 -0700316#ifdef CONFIG_HAVE_RCU_TABLE_FREE
317
318/*
319 * See the comment near struct mmu_table_batch.
320 */
321
322static void tlb_remove_table_smp_sync(void *arg)
323{
324 /* Simply deliver the interrupt */
325}
326
327static void tlb_remove_table_one(void *table)
328{
329 /*
330 * This isn't an RCU grace period and hence the page-tables cannot be
331 * assumed to be actually RCU-freed.
332 *
333 * It is however sufficient for software page-table walkers that rely on
334 * IRQ disabling. See the comment near struct mmu_table_batch.
335 */
336 smp_call_function(tlb_remove_table_smp_sync, NULL, 1);
337 __tlb_remove_table(table);
338}
339
340static void tlb_remove_table_rcu(struct rcu_head *head)
341{
342 struct mmu_table_batch *batch;
343 int i;
344
345 batch = container_of(head, struct mmu_table_batch, rcu);
346
347 for (i = 0; i < batch->nr; i++)
348 __tlb_remove_table(batch->tables[i]);
349
350 free_page((unsigned long)batch);
351}
352
353void tlb_table_flush(struct mmu_gather *tlb)
354{
355 struct mmu_table_batch **batch = &tlb->batch;
356
357 if (*batch) {
358 call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu);
359 *batch = NULL;
360 }
361}
362
363void tlb_remove_table(struct mmu_gather *tlb, void *table)
364{
365 struct mmu_table_batch **batch = &tlb->batch;
366
Peter Zijlstra26723912011-05-24 17:12:00 -0700367 /*
368 * When there's less then two users of this mm there cannot be a
369 * concurrent page-table walk.
370 */
371 if (atomic_read(&tlb->mm->mm_users) < 2) {
372 __tlb_remove_table(table);
373 return;
374 }
375
376 if (*batch == NULL) {
377 *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN);
378 if (*batch == NULL) {
379 tlb_remove_table_one(table);
380 return;
381 }
382 (*batch)->nr = 0;
383 }
384 (*batch)->tables[(*batch)->nr++] = table;
385 if ((*batch)->nr == MAX_TABLE_BATCH)
386 tlb_table_flush(tlb);
387}
388
Peter Zijlstra9547d012011-05-24 17:12:14 -0700389#endif /* CONFIG_HAVE_RCU_TABLE_FREE */
Peter Zijlstra26723912011-05-24 17:12:00 -0700390
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 * Note: this doesn't free the actual pages themselves. That
393 * has been handled earlier when unmapping all the memory regions.
394 */
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000395static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
396 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800398 pgtable_t token = pmd_pgtable(*pmd);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700399 pmd_clear(pmd);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000400 pte_free_tlb(tlb, token, addr);
Kirill A. Shutemove1f56c82013-11-14 14:30:48 -0800401 atomic_long_dec(&tlb->mm->nr_ptes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402}
403
Hugh Dickinse0da3822005-04-19 13:29:15 -0700404static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
405 unsigned long addr, unsigned long end,
406 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407{
408 pmd_t *pmd;
409 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700410 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
Hugh Dickinse0da3822005-04-19 13:29:15 -0700412 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 do {
415 next = pmd_addr_end(addr, end);
416 if (pmd_none_or_clear_bad(pmd))
417 continue;
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000418 free_pte_range(tlb, pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 } while (pmd++, addr = next, addr != end);
420
Hugh Dickinse0da3822005-04-19 13:29:15 -0700421 start &= PUD_MASK;
422 if (start < floor)
423 return;
424 if (ceiling) {
425 ceiling &= PUD_MASK;
426 if (!ceiling)
427 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700429 if (end - 1 > ceiling - 1)
430 return;
431
432 pmd = pmd_offset(pud, start);
433 pud_clear(pud);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000434 pmd_free_tlb(tlb, pmd, start);
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -0800435 mm_dec_nr_pmds(tlb->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436}
437
Hugh Dickinse0da3822005-04-19 13:29:15 -0700438static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
439 unsigned long addr, unsigned long end,
440 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441{
442 pud_t *pud;
443 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700444 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
Hugh Dickinse0da3822005-04-19 13:29:15 -0700446 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 pud = pud_offset(pgd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 do {
449 next = pud_addr_end(addr, end);
450 if (pud_none_or_clear_bad(pud))
451 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700452 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 } while (pud++, addr = next, addr != end);
454
Hugh Dickinse0da3822005-04-19 13:29:15 -0700455 start &= PGDIR_MASK;
456 if (start < floor)
457 return;
458 if (ceiling) {
459 ceiling &= PGDIR_MASK;
460 if (!ceiling)
461 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700463 if (end - 1 > ceiling - 1)
464 return;
465
466 pud = pud_offset(pgd, start);
467 pgd_clear(pgd);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000468 pud_free_tlb(tlb, pud, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469}
470
471/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700472 * This function frees user-level page tables of a process.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 */
Jan Beulich42b77722008-07-23 21:27:10 -0700474void free_pgd_range(struct mmu_gather *tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700475 unsigned long addr, unsigned long end,
476 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477{
478 pgd_t *pgd;
479 unsigned long next;
480
Hugh Dickinse0da3822005-04-19 13:29:15 -0700481 /*
482 * The next few lines have given us lots of grief...
483 *
484 * Why are we testing PMD* at this top level? Because often
485 * there will be no work to do at all, and we'd prefer not to
486 * go all the way down to the bottom just to discover that.
487 *
488 * Why all these "- 1"s? Because 0 represents both the bottom
489 * of the address space and the top of it (using -1 for the
490 * top wouldn't help much: the masks would do the wrong thing).
491 * The rule is that addr 0 and floor 0 refer to the bottom of
492 * the address space, but end 0 and ceiling 0 refer to the top
493 * Comparisons need to use "end - 1" and "ceiling - 1" (though
494 * that end 0 case should be mythical).
495 *
496 * Wherever addr is brought up or ceiling brought down, we must
497 * be careful to reject "the opposite 0" before it confuses the
498 * subsequent tests. But what about where end is brought down
499 * by PMD_SIZE below? no, end can't go down to 0 there.
500 *
501 * Whereas we round start (addr) and ceiling down, by different
502 * masks at different levels, in order to test whether a table
503 * now has no other vmas using it, so can be freed, we don't
504 * bother to round floor or end up - the tests don't need that.
505 */
506
507 addr &= PMD_MASK;
508 if (addr < floor) {
509 addr += PMD_SIZE;
510 if (!addr)
511 return;
512 }
513 if (ceiling) {
514 ceiling &= PMD_MASK;
515 if (!ceiling)
516 return;
517 }
518 if (end - 1 > ceiling - 1)
519 end -= PMD_SIZE;
520 if (addr > end - 1)
521 return;
522
Jan Beulich42b77722008-07-23 21:27:10 -0700523 pgd = pgd_offset(tlb->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 do {
525 next = pgd_addr_end(addr, end);
526 if (pgd_none_or_clear_bad(pgd))
527 continue;
Jan Beulich42b77722008-07-23 21:27:10 -0700528 free_pud_range(tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700530}
531
Jan Beulich42b77722008-07-23 21:27:10 -0700532void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700533 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700534{
535 while (vma) {
536 struct vm_area_struct *next = vma->vm_next;
537 unsigned long addr = vma->vm_start;
538
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700539 /*
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000540 * Hide vma from rmap and truncate_pagecache before freeing
541 * pgtables
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700542 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800543 unlink_anon_vmas(vma);
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700544 unlink_file_vma(vma);
545
David Gibson9da61ae2006-03-22 00:08:57 -0800546 if (is_vm_hugetlb_page(vma)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700547 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700548 floor, next? next->vm_start: ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700549 } else {
550 /*
551 * Optimization: gather nearby vmas into one call down
552 */
553 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
David Gibson48669202006-03-22 00:08:58 -0800554 && !is_vm_hugetlb_page(next)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700555 vma = next;
556 next = vma->vm_next;
Rik van Riel5beb4932010-03-05 13:42:07 -0800557 unlink_anon_vmas(vma);
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700558 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700559 }
560 free_pgd_range(tlb, addr, vma->vm_end,
561 floor, next? next->vm_start: ceiling);
562 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700563 vma = next;
564 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565}
566
Kirill A. Shutemov3ed3a4f2016-03-17 14:19:11 -0700567int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568{
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -0800569 spinlock_t *ptl;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800570 pgtable_t new = pte_alloc_one(mm, address);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700571 if (!new)
572 return -ENOMEM;
573
Nick Piggin362a61a2008-05-14 06:37:36 +0200574 /*
575 * Ensure all pte setup (eg. pte page lock and page clearing) are
576 * visible before the pte is made visible to other CPUs by being
577 * put into page tables.
578 *
579 * The other side of the story is the pointer chasing in the page
580 * table walking code (when walking the page table without locking;
581 * ie. most of the time). Fortunately, these data accesses consist
582 * of a chain of data-dependent loads, meaning most CPUs (alpha
583 * being the notable exception) will already guarantee loads are
584 * seen in-order. See the alpha page table accessors for the
585 * smp_read_barrier_depends() barriers in page table walking code.
586 */
587 smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
588
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -0800589 ptl = pmd_lock(mm, pmd);
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800590 if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
Kirill A. Shutemove1f56c82013-11-14 14:30:48 -0800591 atomic_long_inc(&mm->nr_ptes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 pmd_populate(mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800593 new = NULL;
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800594 }
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);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700598 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599}
600
Hugh Dickins1bb36302005-10-29 18:16:22 -0700601int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700603 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
604 if (!new)
605 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606
Nick Piggin362a61a2008-05-14 06:37:36 +0200607 smp_wmb(); /* See comment in __pte_alloc */
608
Hugh Dickins1bb36302005-10-29 18:16:22 -0700609 spin_lock(&init_mm.page_table_lock);
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800610 if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
Hugh Dickins1bb36302005-10-29 18:16:22 -0700611 pmd_populate_kernel(&init_mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800612 new = NULL;
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800613 }
Hugh Dickins1bb36302005-10-29 18:16:22 -0700614 spin_unlock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800615 if (new)
616 pte_free_kernel(&init_mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700617 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618}
619
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800620static inline void init_rss_vec(int *rss)
Hugh Dickinsae859762005-10-29 18:16:05 -0700621{
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800622 memset(rss, 0, sizeof(int) * NR_MM_COUNTERS);
623}
624
625static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss)
626{
627 int i;
628
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800629 if (current->mm == mm)
David Rientjes05af2e12012-03-21 16:34:13 -0700630 sync_mm_rss(mm);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800631 for (i = 0; i < NR_MM_COUNTERS; i++)
632 if (rss[i])
633 add_mm_counter(mm, i, rss[i]);
Hugh Dickinsae859762005-10-29 18:16:05 -0700634}
635
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800637 * This function is called to print an error when a bad pte
638 * is found. For example, we might have a PFN-mapped pte in
639 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700640 *
641 * The calling function must still handle the error.
642 */
Hugh Dickins3dc14742009-01-06 14:40:08 -0800643static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
644 pte_t pte, struct page *page)
Nick Pigginb5810032005-10-29 18:16:12 -0700645{
Hugh Dickins3dc14742009-01-06 14:40:08 -0800646 pgd_t *pgd = pgd_offset(vma->vm_mm, addr);
647 pud_t *pud = pud_offset(pgd, addr);
648 pmd_t *pmd = pmd_offset(pud, addr);
649 struct address_space *mapping;
650 pgoff_t index;
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800651 static unsigned long resume;
652 static unsigned long nr_shown;
653 static unsigned long nr_unshown;
654
655 /*
656 * Allow a burst of 60 reports, then keep quiet for that minute;
657 * or allow a steady drip of one report per second.
658 */
659 if (nr_shown == 60) {
660 if (time_before(jiffies, resume)) {
661 nr_unshown++;
662 return;
663 }
664 if (nr_unshown) {
Joe Perches11705322016-03-17 14:19:50 -0700665 pr_alert("BUG: Bad page map: %lu messages suppressed\n",
666 nr_unshown);
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800667 nr_unshown = 0;
668 }
669 nr_shown = 0;
670 }
671 if (nr_shown++ == 0)
672 resume = jiffies + 60 * HZ;
Hugh Dickins3dc14742009-01-06 14:40:08 -0800673
674 mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;
675 index = linear_page_index(vma, addr);
676
Joe Perches11705322016-03-17 14:19:50 -0700677 pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
678 current->comm,
679 (long long)pte_val(pte), (long long)pmd_val(*pmd));
Wu Fengguang718a3822010-03-10 15:20:43 -0800680 if (page)
Dave Hansenf0b791a2014-01-23 15:52:49 -0800681 dump_page(page, "bad pte");
Joe Perches11705322016-03-17 14:19:50 -0700682 pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
683 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
Hugh Dickins3dc14742009-01-06 14:40:08 -0800684 /*
685 * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y
686 */
Konstantin Khlebnikov26825822015-04-15 16:15:08 -0700687 pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n",
688 vma->vm_file,
689 vma->vm_ops ? vma->vm_ops->fault : NULL,
690 vma->vm_file ? vma->vm_file->f_op->mmap : NULL,
691 mapping ? mapping->a_ops->readpage : NULL);
Nick Pigginb5810032005-10-29 18:16:12 -0700692 dump_stack();
Rusty Russell373d4d02013-01-21 17:17:39 +1030693 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Nick Pigginb5810032005-10-29 18:16:12 -0700694}
695
696/*
Nick Piggin7e675132008-04-28 02:13:00 -0700697 * vm_normal_page -- This function gets the "struct page" associated with a pte.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800698 *
Nick Piggin7e675132008-04-28 02:13:00 -0700699 * "Special" mappings do not wish to be associated with a "struct page" (either
700 * it doesn't exist, or it exists but they don't want to touch it). In this
701 * case, NULL is returned here. "Normal" mappings do have a struct page.
Jared Hulbertb379d792008-04-28 02:12:58 -0700702 *
Nick Piggin7e675132008-04-28 02:13:00 -0700703 * There are 2 broad cases. Firstly, an architecture may define a pte_special()
704 * pte bit, in which case this function is trivial. Secondly, an architecture
705 * may not have a spare pte bit, which requires a more complicated scheme,
706 * described below.
707 *
708 * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
709 * special mapping (even if there are underlying and valid "struct pages").
710 * COWed pages of a VM_PFNMAP are always normal.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800711 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700712 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
713 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
Nick Piggin7e675132008-04-28 02:13:00 -0700714 * set, and the vm_pgoff will point to the first PFN mapped: thus every special
715 * mapping will always honor the rule
Linus Torvalds6aab3412005-11-28 14:34:23 -0800716 *
717 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
718 *
Nick Piggin7e675132008-04-28 02:13:00 -0700719 * And for normal mappings this is false.
Jared Hulbertb379d792008-04-28 02:12:58 -0700720 *
Nick Piggin7e675132008-04-28 02:13:00 -0700721 * This restricts such mappings to be a linear translation from virtual address
722 * to pfn. To get around this restriction, we allow arbitrary mappings so long
723 * as the vma is not a COW mapping; in that case, we know that all ptes are
724 * special (because none can have been COWed).
Jared Hulbertb379d792008-04-28 02:12:58 -0700725 *
726 *
Nick Piggin7e675132008-04-28 02:13:00 -0700727 * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
728 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700729 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
730 * page" backing, however the difference is that _all_ pages with a struct
731 * page (that is, those where pfn_valid is true) are refcounted and considered
732 * normal pages by the VM. The disadvantage is that pages are refcounted
733 * (which can be slower and simply not an option for some PFNMAP users). The
734 * advantage is that we don't have to follow the strict linearity rule of
735 * PFNMAP mappings in order to support COWable mappings.
736 *
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800737 */
Nick Piggin7e675132008-04-28 02:13:00 -0700738#ifdef __HAVE_ARCH_PTE_SPECIAL
739# define HAVE_PTE_SPECIAL 1
740#else
741# define HAVE_PTE_SPECIAL 0
742#endif
743struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
744 pte_t pte)
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800745{
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800746 unsigned long pfn = pte_pfn(pte);
Nick Piggin7e675132008-04-28 02:13:00 -0700747
748 if (HAVE_PTE_SPECIAL) {
Hugh Dickinsb38af472014-08-29 15:18:44 -0700749 if (likely(!pte_special(pte)))
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800750 goto check_pfn;
David Vrabel667a0a02014-12-18 14:48:15 +0000751 if (vma->vm_ops && vma->vm_ops->find_special_page)
752 return vma->vm_ops->find_special_page(vma, addr);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -0700753 if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))
754 return NULL;
Hugh Dickins62eede62009-09-21 17:03:34 -0700755 if (!is_zero_pfn(pfn))
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800756 print_bad_pte(vma, addr, pte, NULL);
Nick Piggin7e675132008-04-28 02:13:00 -0700757 return NULL;
758 }
759
760 /* !HAVE_PTE_SPECIAL case follows: */
761
Jared Hulbertb379d792008-04-28 02:12:58 -0700762 if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
763 if (vma->vm_flags & VM_MIXEDMAP) {
764 if (!pfn_valid(pfn))
765 return NULL;
766 goto out;
767 } else {
Nick Piggin7e675132008-04-28 02:13:00 -0700768 unsigned long off;
769 off = (addr - vma->vm_start) >> PAGE_SHIFT;
Jared Hulbertb379d792008-04-28 02:12:58 -0700770 if (pfn == vma->vm_pgoff + off)
771 return NULL;
772 if (!is_cow_mapping(vma->vm_flags))
773 return NULL;
774 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800775 }
776
Hugh Dickinsb38af472014-08-29 15:18:44 -0700777 if (is_zero_pfn(pfn))
778 return NULL;
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800779check_pfn:
780 if (unlikely(pfn > highest_memmap_pfn)) {
781 print_bad_pte(vma, addr, pte, NULL);
782 return NULL;
783 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800784
785 /*
Nick Piggin7e675132008-04-28 02:13:00 -0700786 * NOTE! We still have PageReserved() pages in the page tables.
Nick Piggin7e675132008-04-28 02:13:00 -0700787 * eg. VDSO mappings can cause them to exist.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800788 */
Jared Hulbertb379d792008-04-28 02:12:58 -0700789out:
Linus Torvalds6aab3412005-11-28 14:34:23 -0800790 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800791}
792
Gerald Schaefer28093f92016-04-28 16:18:35 -0700793#ifdef CONFIG_TRANSPARENT_HUGEPAGE
794struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr,
795 pmd_t pmd)
796{
797 unsigned long pfn = pmd_pfn(pmd);
798
799 /*
800 * There is no pmd_special() but there may be special pmds, e.g.
801 * in a direct-access (dax) mapping, so let's just replicate the
802 * !HAVE_PTE_SPECIAL case from vm_normal_page() here.
803 */
804 if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
805 if (vma->vm_flags & VM_MIXEDMAP) {
806 if (!pfn_valid(pfn))
807 return NULL;
808 goto out;
809 } else {
810 unsigned long off;
811 off = (addr - vma->vm_start) >> PAGE_SHIFT;
812 if (pfn == vma->vm_pgoff + off)
813 return NULL;
814 if (!is_cow_mapping(vma->vm_flags))
815 return NULL;
816 }
817 }
818
819 if (is_zero_pfn(pfn))
820 return NULL;
821 if (unlikely(pfn > highest_memmap_pfn))
822 return NULL;
823
824 /*
825 * NOTE! We still have PageReserved() pages in the page tables.
826 * eg. VDSO mappings can cause them to exist.
827 */
828out:
829 return pfn_to_page(pfn);
830}
831#endif
832
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800833/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 * copy one vm_area from one task to the other. Assumes the page tables
835 * already present in the new task to be cleared in the whole range
836 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 */
838
Hugh Dickins570a335b2009-12-14 17:58:46 -0800839static inline unsigned long
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700841 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700842 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843{
Nick Pigginb5810032005-10-29 18:16:12 -0700844 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 pte_t pte = *src_pte;
846 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847
848 /* pte contains position in swap or file, so copy. */
849 if (unlikely(!pte_present(pte))) {
Kirill A. Shutemov0661a332015-02-10 14:10:04 -0800850 swp_entry_t entry = pte_to_swp_entry(pte);
Christoph Lameter06972122006-06-23 02:03:35 -0700851
Kirill A. Shutemov0661a332015-02-10 14:10:04 -0800852 if (likely(!non_swap_entry(entry))) {
853 if (swap_duplicate(entry) < 0)
854 return entry.val;
Hugh Dickins570a335b2009-12-14 17:58:46 -0800855
Kirill A. Shutemov0661a332015-02-10 14:10:04 -0800856 /* make sure dst_mm is on swapoff's mmlist. */
857 if (unlikely(list_empty(&dst_mm->mmlist))) {
858 spin_lock(&mmlist_lock);
859 if (list_empty(&dst_mm->mmlist))
860 list_add(&dst_mm->mmlist,
861 &src_mm->mmlist);
862 spin_unlock(&mmlist_lock);
863 }
864 rss[MM_SWAPENTS]++;
865 } else if (is_migration_entry(entry)) {
866 page = migration_entry_to_page(entry);
Konstantin Khlebnikov9f9f1ac2012-01-20 14:34:24 -0800867
Jerome Marchandeca56ff2016-01-14 15:19:26 -0800868 rss[mm_counter(page)]++;
Konstantin Khlebnikov9f9f1ac2012-01-20 14:34:24 -0800869
Kirill A. Shutemov0661a332015-02-10 14:10:04 -0800870 if (is_write_migration_entry(entry) &&
871 is_cow_mapping(vm_flags)) {
872 /*
873 * COW mappings require pages in both
874 * parent and child to be set to read.
875 */
876 make_migration_entry_read(&entry);
877 pte = swp_entry_to_pte(entry);
878 if (pte_swp_soft_dirty(*src_pte))
879 pte = pte_swp_mksoft_dirty(pte);
880 set_pte_at(src_mm, addr, src_pte, pte);
Christoph Lameter06972122006-06-23 02:03:35 -0700881 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700883 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 }
885
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 /*
887 * If it's a COW mapping, write protect it both
888 * in the parent and the child
889 */
Linus Torvalds67121172005-12-11 20:38:17 -0800890 if (is_cow_mapping(vm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 ptep_set_wrprotect(src_mm, addr, src_pte);
Zachary Amsden3dc90792006-09-30 23:29:30 -0700892 pte = pte_wrprotect(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 }
894
895 /*
896 * If it's a shared mapping, mark it clean in
897 * the child
898 */
899 if (vm_flags & VM_SHARED)
900 pte = pte_mkclean(pte);
901 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800902
903 page = vm_normal_page(vma, addr, pte);
904 if (page) {
905 get_page(page);
Kirill A. Shutemov53f92632016-01-15 16:53:42 -0800906 page_dup_rmap(page, false);
Jerome Marchandeca56ff2016-01-14 15:19:26 -0800907 rss[mm_counter(page)]++;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800908 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700909
910out_set_pte:
911 set_pte_at(dst_mm, addr, dst_pte, pte);
Hugh Dickins570a335b2009-12-14 17:58:46 -0800912 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913}
914
Jerome Marchand21bda262014-08-06 16:06:56 -0700915static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800916 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
917 unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918{
Daisuke Nishimurac36987e2009-10-26 16:50:23 -0700919 pte_t *orig_src_pte, *orig_dst_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700921 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -0700922 int progress = 0;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800923 int rss[NR_MM_COUNTERS];
Hugh Dickins570a335b2009-12-14 17:58:46 -0800924 swp_entry_t entry = (swp_entry_t){0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925
926again:
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800927 init_rss_vec(rss);
928
Hugh Dickinsc74df322005-10-29 18:16:23 -0700929 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 if (!dst_pte)
931 return -ENOMEM;
Peter Zijlstraece0e2b2010-10-26 14:21:52 -0700932 src_pte = pte_offset_map(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700933 src_ptl = pte_lockptr(src_mm, src_pmd);
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700934 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
Daisuke Nishimurac36987e2009-10-26 16:50:23 -0700935 orig_src_pte = src_pte;
936 orig_dst_pte = dst_pte;
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700937 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 do {
940 /*
941 * We are holding two locks at this point - either of them
942 * could generate latencies in another task on another CPU.
943 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700944 if (progress >= 32) {
945 progress = 0;
946 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +0100947 spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700948 break;
949 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 if (pte_none(*src_pte)) {
951 progress++;
952 continue;
953 }
Hugh Dickins570a335b2009-12-14 17:58:46 -0800954 entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte,
955 vma, addr, rss);
956 if (entry.val)
957 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 progress += 8;
959 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700961 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -0700962 spin_unlock(src_ptl);
Peter Zijlstraece0e2b2010-10-26 14:21:52 -0700963 pte_unmap(orig_src_pte);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800964 add_mm_rss_vec(dst_mm, rss);
Daisuke Nishimurac36987e2009-10-26 16:50:23 -0700965 pte_unmap_unlock(orig_dst_pte, dst_ptl);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700966 cond_resched();
Hugh Dickins570a335b2009-12-14 17:58:46 -0800967
968 if (entry.val) {
969 if (add_swap_count_continuation(entry, GFP_KERNEL) < 0)
970 return -ENOMEM;
971 progress = 0;
972 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 if (addr != end)
974 goto again;
975 return 0;
976}
977
978static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
979 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
980 unsigned long addr, unsigned long end)
981{
982 pmd_t *src_pmd, *dst_pmd;
983 unsigned long next;
984
985 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
986 if (!dst_pmd)
987 return -ENOMEM;
988 src_pmd = pmd_offset(src_pud, addr);
989 do {
990 next = pmd_addr_end(addr, end);
Dan Williams5c7fb562016-01-15 16:56:52 -0800991 if (pmd_trans_huge(*src_pmd) || pmd_devmap(*src_pmd)) {
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800992 int err;
Andrea Arcangeli14d1a552011-01-13 15:47:15 -0800993 VM_BUG_ON(next-addr != HPAGE_PMD_SIZE);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800994 err = copy_huge_pmd(dst_mm, src_mm,
995 dst_pmd, src_pmd, addr, vma);
996 if (err == -ENOMEM)
997 return -ENOMEM;
998 if (!err)
999 continue;
1000 /* fall through */
1001 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 if (pmd_none_or_clear_bad(src_pmd))
1003 continue;
1004 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
1005 vma, addr, next))
1006 return -ENOMEM;
1007 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
1008 return 0;
1009}
1010
1011static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
1012 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
1013 unsigned long addr, unsigned long end)
1014{
1015 pud_t *src_pud, *dst_pud;
1016 unsigned long next;
1017
1018 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
1019 if (!dst_pud)
1020 return -ENOMEM;
1021 src_pud = pud_offset(src_pgd, addr);
1022 do {
1023 next = pud_addr_end(addr, end);
1024 if (pud_none_or_clear_bad(src_pud))
1025 continue;
1026 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
1027 vma, addr, next))
1028 return -ENOMEM;
1029 } while (dst_pud++, src_pud++, addr = next, addr != end);
1030 return 0;
1031}
1032
1033int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
1034 struct vm_area_struct *vma)
1035{
1036 pgd_t *src_pgd, *dst_pgd;
1037 unsigned long next;
1038 unsigned long addr = vma->vm_start;
1039 unsigned long end = vma->vm_end;
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001040 unsigned long mmun_start; /* For mmu_notifiers */
1041 unsigned long mmun_end; /* For mmu_notifiers */
1042 bool is_cow;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001043 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
Nick Piggind9928952005-08-28 16:49:11 +10001045 /*
1046 * Don't copy ptes where a page fault will fill them correctly.
1047 * Fork becomes much lighter when there are big shared or private
1048 * readonly mappings. The tradeoff is that copy_page_range is more
1049 * efficient than faulting.
1050 */
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08001051 if (!(vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) &&
1052 !vma->anon_vma)
1053 return 0;
Nick Piggind9928952005-08-28 16:49:11 +10001054
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 if (is_vm_hugetlb_page(vma))
1056 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
1057
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07001058 if (unlikely(vma->vm_flags & VM_PFNMAP)) {
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001059 /*
1060 * We do not free on error cases below as remove_vma
1061 * gets called on error from higher level routine
1062 */
Suresh Siddha5180da42012-10-08 16:28:29 -07001063 ret = track_pfn_copy(vma);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001064 if (ret)
1065 return ret;
1066 }
1067
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001068 /*
1069 * We need to invalidate the secondary MMU mappings only when
1070 * there could be a permission downgrade on the ptes of the
1071 * parent mm. And a permission downgrade will only happen if
1072 * is_cow_mapping() returns true.
1073 */
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001074 is_cow = is_cow_mapping(vma->vm_flags);
1075 mmun_start = addr;
1076 mmun_end = end;
1077 if (is_cow)
1078 mmu_notifier_invalidate_range_start(src_mm, mmun_start,
1079 mmun_end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001080
1081 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 dst_pgd = pgd_offset(dst_mm, addr);
1083 src_pgd = pgd_offset(src_mm, addr);
1084 do {
1085 next = pgd_addr_end(addr, end);
1086 if (pgd_none_or_clear_bad(src_pgd))
1087 continue;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001088 if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
1089 vma, addr, next))) {
1090 ret = -ENOMEM;
1091 break;
1092 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001094
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001095 if (is_cow)
1096 mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001097 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098}
1099
Robin Holt51c6f662005-11-13 16:06:42 -08001100static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -07001101 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 unsigned long addr, unsigned long end,
Peter Zijlstra97a89412011-05-24 17:12:04 -07001103 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104{
Nick Pigginb5810032005-10-29 18:16:12 -07001105 struct mm_struct *mm = tlb->mm;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001106 int force_flush = 0;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001107 int rss[NR_MM_COUNTERS];
Peter Zijlstra97a89412011-05-24 17:12:04 -07001108 spinlock_t *ptl;
Steven Rostedt5f1a1902011-06-15 15:08:23 -07001109 pte_t *start_pte;
Peter Zijlstra97a89412011-05-24 17:12:04 -07001110 pte_t *pte;
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001111 swp_entry_t entry;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001112
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001113again:
Peter Zijlstrae3032972011-05-24 17:12:01 -07001114 init_rss_vec(rss);
Steven Rostedt5f1a1902011-06-15 15:08:23 -07001115 start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
1116 pte = start_pte;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001117 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 do {
1119 pte_t ptent = *pte;
Robin Holt51c6f662005-11-13 16:06:42 -08001120 if (pte_none(ptent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 continue;
Robin Holt51c6f662005-11-13 16:06:42 -08001122 }
Hugh Dickins6f5e6b92006-03-16 23:04:09 -08001123
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -08001125 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -08001126
Linus Torvalds6aab3412005-11-28 14:34:23 -08001127 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 if (unlikely(details) && page) {
1129 /*
1130 * unmap_shared_mapping_pages() wants to
1131 * invalidate cache without truncating:
1132 * unmap shared but keep private pages.
1133 */
1134 if (details->check_mapping &&
1135 details->check_mapping != page->mapping)
1136 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 }
Nick Pigginb5810032005-10-29 18:16:12 -07001138 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -07001139 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 tlb_remove_tlb_entry(tlb, pte, addr);
1141 if (unlikely(!page))
1142 continue;
Jerome Marchandeca56ff2016-01-14 15:19:26 -08001143
1144 if (!PageAnon(page)) {
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001145 if (pte_dirty(ptent)) {
Michal Hockoaac45362016-03-25 14:20:24 -07001146 /*
1147 * oom_reaper cannot tear down dirty
1148 * pages
1149 */
1150 if (unlikely(details && details->ignore_dirty))
1151 continue;
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001152 force_flush = 1;
Hugh Dickins6237bcd2005-10-29 18:15:54 -07001153 set_page_dirty(page);
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001154 }
Johannes Weiner4917e5d2009-01-06 14:39:17 -08001155 if (pte_young(ptent) &&
Joe Perches64363aa2013-07-08 16:00:18 -07001156 likely(!(vma->vm_flags & VM_SEQ_READ)))
Nick Pigginbf3f3bc2009-01-06 14:38:55 -08001157 mark_page_accessed(page);
Hugh Dickins6237bcd2005-10-29 18:15:54 -07001158 }
Jerome Marchandeca56ff2016-01-14 15:19:26 -08001159 rss[mm_counter(page)]--;
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001160 page_remove_rmap(page, false);
Hugh Dickins3dc14742009-01-06 14:40:08 -08001161 if (unlikely(page_mapcount(page) < 0))
1162 print_bad_pte(vma, addr, ptent, page);
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001163 if (unlikely(!__tlb_remove_page(tlb, page))) {
1164 force_flush = 1;
Will Deaconce9ec372014-10-28 13:16:28 -07001165 addr += PAGE_SIZE;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001166 break;
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001167 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 continue;
1169 }
Michal Hockoaac45362016-03-25 14:20:24 -07001170 /* only check swap_entries if explicitly asked for in details */
1171 if (unlikely(details && !details->check_swap_entries))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 continue;
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001173
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001174 entry = pte_to_swp_entry(ptent);
1175 if (!non_swap_entry(entry))
1176 rss[MM_SWAPENTS]--;
1177 else if (is_migration_entry(entry)) {
1178 struct page *page;
Konstantin Khlebnikov9f9f1ac2012-01-20 14:34:24 -08001179
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001180 page = migration_entry_to_page(entry);
Jerome Marchandeca56ff2016-01-14 15:19:26 -08001181 rss[mm_counter(page)]--;
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001182 }
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001183 if (unlikely(!free_swap_and_cache(entry)))
1184 print_bad_pte(vma, addr, ptent, NULL);
Zachary Amsden9888a1c2006-09-30 23:29:31 -07001185 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
Peter Zijlstra97a89412011-05-24 17:12:04 -07001186 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickinsae859762005-10-29 18:16:05 -07001187
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001188 add_mm_rss_vec(mm, rss);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001189 arch_leave_lazy_mmu_mode();
Robin Holt51c6f662005-11-13 16:06:42 -08001190
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001191 /* Do the actual TLB flush before dropping ptl */
Will Deaconfb7332a2014-10-29 10:03:09 +00001192 if (force_flush)
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001193 tlb_flush_mmu_tlbonly(tlb);
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001194 pte_unmap_unlock(start_pte, ptl);
1195
1196 /*
1197 * If we forced a TLB flush (either due to running out of
1198 * batch buffers or because we needed to flush dirty TLB
1199 * entries before releasing the ptl), free the batched
1200 * memory too. Restart if we didn't do everything.
1201 */
1202 if (force_flush) {
1203 force_flush = 0;
1204 tlb_flush_mmu_free(tlb);
Linus Torvalds2b047252013-08-15 11:42:25 -07001205
1206 if (addr != end)
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001207 goto again;
1208 }
1209
Robin Holt51c6f662005-11-13 16:06:42 -08001210 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211}
1212
Robin Holt51c6f662005-11-13 16:06:42 -08001213static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -07001214 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 unsigned long addr, unsigned long end,
Peter Zijlstra97a89412011-05-24 17:12:04 -07001216 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217{
1218 pmd_t *pmd;
1219 unsigned long next;
1220
1221 pmd = pmd_offset(pud, addr);
1222 do {
1223 next = pmd_addr_end(addr, end);
Dan Williams5c7fb562016-01-15 16:56:52 -08001224 if (pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) {
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001225 if (next - addr != HPAGE_PMD_SIZE) {
Hugh Dickins68428392016-05-05 16:22:09 -07001226 VM_BUG_ON_VMA(vma_is_anonymous(vma) &&
1227 !rwsem_is_locked(&tlb->mm->mmap_sem), vma);
Kirill A. Shutemov78ddc532016-01-15 16:52:42 -08001228 split_huge_pmd(vma, pmd, addr);
Shaohua Lif21760b2012-01-12 17:19:16 -08001229 } else if (zap_huge_pmd(tlb, vma, pmd, addr))
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001230 goto next;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001231 /* fall through */
1232 }
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001233 /*
1234 * Here there can be other concurrent MADV_DONTNEED or
1235 * trans huge page faults running, and if the pmd is
1236 * none or trans huge it can change under us. This is
1237 * because MADV_DONTNEED holds the mmap_sem in read
1238 * mode.
1239 */
1240 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1241 goto next;
Peter Zijlstra97a89412011-05-24 17:12:04 -07001242 next = zap_pte_range(tlb, vma, pmd, addr, next, details);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001243next:
Peter Zijlstra97a89412011-05-24 17:12:04 -07001244 cond_resched();
1245 } while (pmd++, addr = next, addr != end);
Robin Holt51c6f662005-11-13 16:06:42 -08001246
1247 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248}
1249
Robin Holt51c6f662005-11-13 16:06:42 -08001250static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -07001251 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 unsigned long addr, unsigned long end,
Peter Zijlstra97a89412011-05-24 17:12:04 -07001253 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254{
1255 pud_t *pud;
1256 unsigned long next;
1257
1258 pud = pud_offset(pgd, addr);
1259 do {
1260 next = pud_addr_end(addr, end);
Peter Zijlstra97a89412011-05-24 17:12:04 -07001261 if (pud_none_or_clear_bad(pud))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 continue;
Peter Zijlstra97a89412011-05-24 17:12:04 -07001263 next = zap_pmd_range(tlb, vma, pud, addr, next, details);
1264 } while (pud++, addr = next, addr != end);
Robin Holt51c6f662005-11-13 16:06:42 -08001265
1266 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267}
1268
Michal Hockoaac45362016-03-25 14:20:24 -07001269void unmap_page_range(struct mmu_gather *tlb,
Al Viro038c7aa2012-03-05 13:25:09 -05001270 struct vm_area_struct *vma,
1271 unsigned long addr, unsigned long end,
1272 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273{
1274 pgd_t *pgd;
1275 unsigned long next;
1276
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 BUG_ON(addr >= end);
1278 tlb_start_vma(tlb, vma);
1279 pgd = pgd_offset(vma->vm_mm, addr);
1280 do {
1281 next = pgd_addr_end(addr, end);
Peter Zijlstra97a89412011-05-24 17:12:04 -07001282 if (pgd_none_or_clear_bad(pgd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 continue;
Peter Zijlstra97a89412011-05-24 17:12:04 -07001284 next = zap_pud_range(tlb, vma, pgd, addr, next, details);
1285 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 tlb_end_vma(tlb, vma);
1287}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288
Al Virof5cc4ee2012-03-05 14:14:20 -05001289
1290static void unmap_single_vma(struct mmu_gather *tlb,
1291 struct vm_area_struct *vma, unsigned long start_addr,
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001292 unsigned long end_addr,
Al Virof5cc4ee2012-03-05 14:14:20 -05001293 struct zap_details *details)
1294{
1295 unsigned long start = max(vma->vm_start, start_addr);
1296 unsigned long end;
1297
1298 if (start >= vma->vm_end)
1299 return;
1300 end = min(vma->vm_end, end_addr);
1301 if (end <= vma->vm_start)
1302 return;
1303
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +05301304 if (vma->vm_file)
1305 uprobe_munmap(vma, start, end);
1306
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07001307 if (unlikely(vma->vm_flags & VM_PFNMAP))
Suresh Siddha5180da42012-10-08 16:28:29 -07001308 untrack_pfn(vma, 0, 0);
Al Virof5cc4ee2012-03-05 14:14:20 -05001309
1310 if (start != end) {
1311 if (unlikely(is_vm_hugetlb_page(vma))) {
1312 /*
1313 * It is undesirable to test vma->vm_file as it
1314 * should be non-null for valid hugetlb area.
1315 * However, vm_file will be NULL in the error
Davidlohr Bueso7aa6b4a2014-04-07 15:37:01 -07001316 * cleanup path of mmap_region. When
Al Virof5cc4ee2012-03-05 14:14:20 -05001317 * hugetlbfs ->mmap method fails,
Davidlohr Bueso7aa6b4a2014-04-07 15:37:01 -07001318 * mmap_region() nullifies vma->vm_file
Al Virof5cc4ee2012-03-05 14:14:20 -05001319 * before calling this function to clean up.
1320 * Since no pte has actually been setup, it is
1321 * safe to do nothing in this case.
1322 */
Aneesh Kumar K.V24669e52012-07-31 16:42:03 -07001323 if (vma->vm_file) {
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08001324 i_mmap_lock_write(vma->vm_file->f_mapping);
Mel Gormand8333522012-07-31 16:46:20 -07001325 __unmap_hugepage_range_final(tlb, vma, start, end, NULL);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08001326 i_mmap_unlock_write(vma->vm_file->f_mapping);
Aneesh Kumar K.V24669e52012-07-31 16:42:03 -07001327 }
Al Virof5cc4ee2012-03-05 14:14:20 -05001328 } else
1329 unmap_page_range(tlb, vma, start, end, details);
1330 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331}
1332
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333/**
1334 * unmap_vmas - unmap a range of memory covered by a list of vma's
Randy Dunlap0164f692011-06-15 15:08:09 -07001335 * @tlb: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 * @vma: the starting vma
1337 * @start_addr: virtual address at which to start unmapping
1338 * @end_addr: virtual address at which to end unmapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 *
Hugh Dickins508034a2005-10-29 18:16:30 -07001340 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 * Only addresses between `start' and `end' will be unmapped.
1343 *
1344 * The VMA list must be sorted in ascending virtual address order.
1345 *
1346 * unmap_vmas() assumes that the caller will flush the whole unmapped address
1347 * range after unmap_vmas() returns. So the only responsibility here is to
1348 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
1349 * drops the lock and schedules.
1350 */
Al Viro6e8bb012012-03-05 13:41:15 -05001351void unmap_vmas(struct mmu_gather *tlb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 struct vm_area_struct *vma, unsigned long start_addr,
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001353 unsigned long end_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354{
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001355 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001357 mmu_notifier_invalidate_range_start(mm, start_addr, end_addr);
Al Virof5cc4ee2012-03-05 14:14:20 -05001358 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next)
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001359 unmap_single_vma(tlb, vma, start_addr, end_addr, NULL);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001360 mmu_notifier_invalidate_range_end(mm, start_addr, end_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361}
1362
1363/**
1364 * zap_page_range - remove user pages in a given range
1365 * @vma: vm_area_struct holding the applicable pages
Randy Dunlapeb4546b2012-06-20 12:53:02 -07001366 * @start: starting address of pages to zap
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 * @size: number of bytes to zap
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001368 * @details: details of shared cache invalidation
Al Virof5cc4ee2012-03-05 14:14:20 -05001369 *
1370 * Caller must protect the VMA list
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 */
Linus Torvalds7e027b12012-05-06 13:43:15 -07001372void zap_page_range(struct vm_area_struct *vma, unsigned long start,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 unsigned long size, struct zap_details *details)
1374{
1375 struct mm_struct *mm = vma->vm_mm;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001376 struct mmu_gather tlb;
Linus Torvalds7e027b12012-05-06 13:43:15 -07001377 unsigned long end = start + size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07001380 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001381 update_hiwater_rss(mm);
Linus Torvalds7e027b12012-05-06 13:43:15 -07001382 mmu_notifier_invalidate_range_start(mm, start, end);
1383 for ( ; vma && vma->vm_start < end; vma = vma->vm_next)
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001384 unmap_single_vma(&tlb, vma, start, end, details);
Linus Torvalds7e027b12012-05-06 13:43:15 -07001385 mmu_notifier_invalidate_range_end(mm, start, end);
1386 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387}
1388
Jack Steinerc627f9c2008-07-29 22:33:53 -07001389/**
Al Virof5cc4ee2012-03-05 14:14:20 -05001390 * zap_page_range_single - remove user pages in a given range
1391 * @vma: vm_area_struct holding the applicable pages
1392 * @address: starting address of pages to zap
1393 * @size: number of bytes to zap
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001394 * @details: details of shared cache invalidation
Al Virof5cc4ee2012-03-05 14:14:20 -05001395 *
1396 * The range must fit into one VMA.
1397 */
1398static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address,
1399 unsigned long size, struct zap_details *details)
1400{
1401 struct mm_struct *mm = vma->vm_mm;
1402 struct mmu_gather tlb;
1403 unsigned long end = address + size;
Al Virof5cc4ee2012-03-05 14:14:20 -05001404
1405 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07001406 tlb_gather_mmu(&tlb, mm, address, end);
Al Virof5cc4ee2012-03-05 14:14:20 -05001407 update_hiwater_rss(mm);
1408 mmu_notifier_invalidate_range_start(mm, address, end);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001409 unmap_single_vma(&tlb, vma, address, end, details);
Al Virof5cc4ee2012-03-05 14:14:20 -05001410 mmu_notifier_invalidate_range_end(mm, address, end);
1411 tlb_finish_mmu(&tlb, address, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412}
1413
Jack Steinerc627f9c2008-07-29 22:33:53 -07001414/**
1415 * zap_vma_ptes - remove ptes mapping the vma
1416 * @vma: vm_area_struct holding ptes to be zapped
1417 * @address: starting address of pages to zap
1418 * @size: number of bytes to zap
1419 *
1420 * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1421 *
1422 * The entire address range must be fully contained within the vma.
1423 *
1424 * Returns 0 if successful.
1425 */
1426int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
1427 unsigned long size)
1428{
1429 if (address < vma->vm_start || address + size > vma->vm_end ||
1430 !(vma->vm_flags & VM_PFNMAP))
1431 return -1;
Al Virof5cc4ee2012-03-05 14:14:20 -05001432 zap_page_range_single(vma, address, size, NULL);
Jack Steinerc627f9c2008-07-29 22:33:53 -07001433 return 0;
1434}
1435EXPORT_SYMBOL_GPL(zap_vma_ptes);
1436
Namhyung Kim25ca1d62010-10-26 14:21:59 -07001437pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr,
Harvey Harrison920c7a52008-02-04 22:29:26 -08001438 spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001439{
1440 pgd_t * pgd = pgd_offset(mm, addr);
1441 pud_t * pud = pud_alloc(mm, pgd, addr);
1442 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001443 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Andrea Arcangelif66055ab2011-01-13 15:46:54 -08001444 if (pmd) {
1445 VM_BUG_ON(pmd_trans_huge(*pmd));
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001446 return pte_alloc_map_lock(mm, pmd, addr, ptl);
Andrea Arcangelif66055ab2011-01-13 15:46:54 -08001447 }
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001448 }
1449 return NULL;
1450}
1451
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001453 * This is the old fallback for page remapping.
1454 *
1455 * For historical reasons, it only allows reserved pages. Only
1456 * old drivers should use this, and they needed to mark their
1457 * pages reserved for the old functions anyway.
1458 */
Nick Piggin423bad62008-04-28 02:13:01 -07001459static int insert_page(struct vm_area_struct *vma, unsigned long addr,
1460 struct page *page, pgprot_t prot)
Linus Torvalds238f58d2005-11-29 13:01:56 -08001461{
Nick Piggin423bad62008-04-28 02:13:01 -07001462 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001463 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001464 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001465 spinlock_t *ptl;
1466
Linus Torvalds238f58d2005-11-29 13:01:56 -08001467 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001468 if (PageAnon(page))
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001469 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001470 retval = -ENOMEM;
1471 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001472 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001473 if (!pte)
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001474 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001475 retval = -EBUSY;
1476 if (!pte_none(*pte))
1477 goto out_unlock;
1478
1479 /* Ok, finally just insert the thing.. */
1480 get_page(page);
Jerome Marchandeca56ff2016-01-14 15:19:26 -08001481 inc_mm_counter_fast(mm, mm_counter_file(page));
Linus Torvalds238f58d2005-11-29 13:01:56 -08001482 page_add_file_rmap(page);
1483 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1484
1485 retval = 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001486 pte_unmap_unlock(pte, ptl);
1487 return retval;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001488out_unlock:
1489 pte_unmap_unlock(pte, ptl);
1490out:
1491 return retval;
1492}
1493
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001494/**
1495 * vm_insert_page - insert single page into user vma
1496 * @vma: user vma to map to
1497 * @addr: target user address of this page
1498 * @page: source kernel page
1499 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001500 * This allows drivers to insert individual pages they've allocated
1501 * into a user vma.
1502 *
1503 * The page has to be a nice clean _individual_ kernel allocation.
1504 * If you allocate a compound page, you need to have marked it as
1505 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001506 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001507 *
1508 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1509 * took an arbitrary page protection parameter. This doesn't allow
1510 * that. Your vma protection will have to be set up correctly, which
1511 * means that if you want a shared writable mapping, you'd better
1512 * ask for a shared writable mapping!
1513 *
1514 * The page does not need to be reserved.
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001515 *
1516 * Usually this function is called from f_op->mmap() handler
1517 * under mm->mmap_sem write-lock, so it can change vma->vm_flags.
1518 * Caller must set VM_MIXEDMAP on vma if it wants to call this
1519 * function from other places, for example from page-fault handler.
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001520 */
Nick Piggin423bad62008-04-28 02:13:01 -07001521int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
1522 struct page *page)
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001523{
1524 if (addr < vma->vm_start || addr >= vma->vm_end)
1525 return -EFAULT;
1526 if (!page_count(page))
1527 return -EINVAL;
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001528 if (!(vma->vm_flags & VM_MIXEDMAP)) {
1529 BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem));
1530 BUG_ON(vma->vm_flags & VM_PFNMAP);
1531 vma->vm_flags |= VM_MIXEDMAP;
1532 }
Nick Piggin423bad62008-04-28 02:13:01 -07001533 return insert_page(vma, addr, page, vma->vm_page_prot);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001534}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001535EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001536
Nick Piggin423bad62008-04-28 02:13:01 -07001537static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
Dan Williams01c8f1c2016-01-15 16:56:40 -08001538 pfn_t pfn, pgprot_t prot)
Nick Piggin423bad62008-04-28 02:13:01 -07001539{
1540 struct mm_struct *mm = vma->vm_mm;
1541 int retval;
1542 pte_t *pte, entry;
1543 spinlock_t *ptl;
1544
1545 retval = -ENOMEM;
1546 pte = get_locked_pte(mm, addr, &ptl);
1547 if (!pte)
1548 goto out;
1549 retval = -EBUSY;
1550 if (!pte_none(*pte))
1551 goto out_unlock;
1552
1553 /* Ok, finally just insert the thing.. */
Dan Williams01c8f1c2016-01-15 16:56:40 -08001554 if (pfn_t_devmap(pfn))
1555 entry = pte_mkdevmap(pfn_t_pte(pfn, prot));
1556 else
1557 entry = pte_mkspecial(pfn_t_pte(pfn, prot));
Nick Piggin423bad62008-04-28 02:13:01 -07001558 set_pte_at(mm, addr, pte, entry);
Russell King4b3073e2009-12-18 16:40:18 +00001559 update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */
Nick Piggin423bad62008-04-28 02:13:01 -07001560
1561 retval = 0;
1562out_unlock:
1563 pte_unmap_unlock(pte, ptl);
1564out:
1565 return retval;
1566}
1567
Nick Piggine0dc0d82007-02-12 00:51:36 -08001568/**
1569 * vm_insert_pfn - insert single pfn into user vma
1570 * @vma: user vma to map to
1571 * @addr: target user address of this page
1572 * @pfn: source kernel pfn
1573 *
Robert P. J. Dayc462f172012-10-08 16:33:43 -07001574 * Similar to vm_insert_page, this allows drivers to insert individual pages
Nick Piggine0dc0d82007-02-12 00:51:36 -08001575 * they've allocated into a user vma. Same comments apply.
1576 *
1577 * This function should only be called from a vm_ops->fault handler, and
1578 * in that case the handler should return NULL.
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001579 *
1580 * vma cannot be a COW mapping.
1581 *
1582 * As this is called only for pages that do not currently exist, we
1583 * do not need to flush old virtual caches or the TLB.
Nick Piggine0dc0d82007-02-12 00:51:36 -08001584 */
1585int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
Nick Piggin423bad62008-04-28 02:13:01 -07001586 unsigned long pfn)
Nick Piggine0dc0d82007-02-12 00:51:36 -08001587{
Andy Lutomirski1745cbc2015-12-29 20:12:20 -08001588 return vm_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot);
1589}
1590EXPORT_SYMBOL(vm_insert_pfn);
1591
1592/**
1593 * vm_insert_pfn_prot - insert single pfn into user vma with specified pgprot
1594 * @vma: user vma to map to
1595 * @addr: target user address of this page
1596 * @pfn: source kernel pfn
1597 * @pgprot: pgprot flags for the inserted page
1598 *
1599 * This is exactly like vm_insert_pfn, except that it allows drivers to
1600 * to override pgprot on a per-page basis.
1601 *
1602 * This only makes sense for IO mappings, and it makes no sense for
1603 * cow mappings. In general, using multiple vmas is preferable;
1604 * vm_insert_pfn_prot should only be used if using multiple VMAs is
1605 * impractical.
1606 */
1607int vm_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr,
1608 unsigned long pfn, pgprot_t pgprot)
1609{
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001610 int ret;
Nick Piggin7e675132008-04-28 02:13:00 -07001611 /*
1612 * Technically, architectures with pte_special can avoid all these
1613 * restrictions (same for remap_pfn_range). However we would like
1614 * consistency in testing and feature parity among all, so we should
1615 * try to keep these invariants in place for everybody.
1616 */
Jared Hulbertb379d792008-04-28 02:12:58 -07001617 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
1618 BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
1619 (VM_PFNMAP|VM_MIXEDMAP));
1620 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1621 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
Nick Piggine0dc0d82007-02-12 00:51:36 -08001622
Nick Piggin423bad62008-04-28 02:13:01 -07001623 if (addr < vma->vm_start || addr >= vma->vm_end)
1624 return -EFAULT;
Dan Williamsf25748e32016-01-15 16:56:43 -08001625 if (track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV)))
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001626 return -EINVAL;
1627
Dan Williams01c8f1c2016-01-15 16:56:40 -08001628 ret = insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001629
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001630 return ret;
Nick Piggine0dc0d82007-02-12 00:51:36 -08001631}
Andy Lutomirski1745cbc2015-12-29 20:12:20 -08001632EXPORT_SYMBOL(vm_insert_pfn_prot);
Nick Piggine0dc0d82007-02-12 00:51:36 -08001633
Nick Piggin423bad62008-04-28 02:13:01 -07001634int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
Dan Williams01c8f1c2016-01-15 16:56:40 -08001635 pfn_t pfn)
Nick Piggin423bad62008-04-28 02:13:01 -07001636{
1637 BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
1638
1639 if (addr < vma->vm_start || addr >= vma->vm_end)
1640 return -EFAULT;
1641
1642 /*
1643 * If we don't have pte special, then we have to use the pfn_valid()
1644 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1645 * refcount the page if pfn_valid is true (hence insert_page rather
Hugh Dickins62eede62009-09-21 17:03:34 -07001646 * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP
1647 * without pte special, it would there be refcounted as a normal page.
Nick Piggin423bad62008-04-28 02:13:01 -07001648 */
Dan Williams03fc2da2016-01-26 09:48:05 -08001649 if (!HAVE_PTE_SPECIAL && !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) {
Nick Piggin423bad62008-04-28 02:13:01 -07001650 struct page *page;
1651
Dan Williams03fc2da2016-01-26 09:48:05 -08001652 /*
1653 * At this point we are committed to insert_page()
1654 * regardless of whether the caller specified flags that
1655 * result in pfn_t_has_page() == false.
1656 */
1657 page = pfn_to_page(pfn_t_to_pfn(pfn));
Nick Piggin423bad62008-04-28 02:13:01 -07001658 return insert_page(vma, addr, page, vma->vm_page_prot);
1659 }
1660 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
1661}
1662EXPORT_SYMBOL(vm_insert_mixed);
1663
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001664/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 * maps a range of physical memory into the requested pages. the old
1666 * mappings are removed. any references to nonexistent pages results
1667 * in null mappings (currently treated as "copy-on-access")
1668 */
1669static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1670 unsigned long addr, unsigned long end,
1671 unsigned long pfn, pgprot_t prot)
1672{
1673 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001674 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675
Hugh Dickinsc74df322005-10-29 18:16:23 -07001676 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 if (!pte)
1678 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001679 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 do {
1681 BUG_ON(!pte_none(*pte));
Nick Piggin7e675132008-04-28 02:13:00 -07001682 set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 pfn++;
1684 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001685 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001686 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 return 0;
1688}
1689
1690static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1691 unsigned long addr, unsigned long end,
1692 unsigned long pfn, pgprot_t prot)
1693{
1694 pmd_t *pmd;
1695 unsigned long next;
1696
1697 pfn -= addr >> PAGE_SHIFT;
1698 pmd = pmd_alloc(mm, pud, addr);
1699 if (!pmd)
1700 return -ENOMEM;
Andrea Arcangelif66055ab2011-01-13 15:46:54 -08001701 VM_BUG_ON(pmd_trans_huge(*pmd));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 do {
1703 next = pmd_addr_end(addr, end);
1704 if (remap_pte_range(mm, pmd, addr, next,
1705 pfn + (addr >> PAGE_SHIFT), prot))
1706 return -ENOMEM;
1707 } while (pmd++, addr = next, addr != end);
1708 return 0;
1709}
1710
1711static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1712 unsigned long addr, unsigned long end,
1713 unsigned long pfn, pgprot_t prot)
1714{
1715 pud_t *pud;
1716 unsigned long next;
1717
1718 pfn -= addr >> PAGE_SHIFT;
1719 pud = pud_alloc(mm, pgd, addr);
1720 if (!pud)
1721 return -ENOMEM;
1722 do {
1723 next = pud_addr_end(addr, end);
1724 if (remap_pmd_range(mm, pud, addr, next,
1725 pfn + (addr >> PAGE_SHIFT), prot))
1726 return -ENOMEM;
1727 } while (pud++, addr = next, addr != end);
1728 return 0;
1729}
1730
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001731/**
1732 * remap_pfn_range - remap kernel memory to userspace
1733 * @vma: user vma to map to
1734 * @addr: target user address to start at
1735 * @pfn: physical address of kernel memory
1736 * @size: size of map area
1737 * @prot: page protection flags for this mapping
1738 *
1739 * Note: this is only safe if the mm semaphore is held when called.
1740 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1742 unsigned long pfn, unsigned long size, pgprot_t prot)
1743{
1744 pgd_t *pgd;
1745 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001746 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 struct mm_struct *mm = vma->vm_mm;
Yongji Xied5957d22016-05-20 16:57:41 -07001748 unsigned long remap_pfn = pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 int err;
1750
1751 /*
1752 * Physically remapped pages are special. Tell the
1753 * rest of the world about it:
1754 * VM_IO tells people not to look at these pages
1755 * (accesses can have side effects).
Linus Torvalds6aab3412005-11-28 14:34:23 -08001756 * VM_PFNMAP tells the core MM that the base pages are just
1757 * raw PFN mappings, and do not have a "struct page" associated
1758 * with them.
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -07001759 * VM_DONTEXPAND
1760 * Disable vma merging and expanding with mremap().
1761 * VM_DONTDUMP
1762 * Omit vma from core dump, even when VM_IO turned off.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001763 *
1764 * There's a horrible special case to handle copy-on-write
1765 * behaviour that some programs depend on. We mark the "original"
1766 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07001767 * See vm_normal_page() for details.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 */
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07001769 if (is_cow_mapping(vma->vm_flags)) {
1770 if (addr != vma->vm_start || end != vma->vm_end)
1771 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001772 vma->vm_pgoff = pfn;
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07001773 }
1774
Yongji Xied5957d22016-05-20 16:57:41 -07001775 err = track_pfn_remap(vma, &prot, remap_pfn, addr, PAGE_ALIGN(size));
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07001776 if (err)
venkatesh.pallipadi@intel.com3c8bb732008-12-18 11:41:27 -08001777 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001778
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -07001779 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780
1781 BUG_ON(addr >= end);
1782 pfn -= addr >> PAGE_SHIFT;
1783 pgd = pgd_offset(mm, addr);
1784 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 do {
1786 next = pgd_addr_end(addr, end);
1787 err = remap_pud_range(mm, pgd, addr, next,
1788 pfn + (addr >> PAGE_SHIFT), prot);
1789 if (err)
1790 break;
1791 } while (pgd++, addr = next, addr != end);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001792
1793 if (err)
Yongji Xied5957d22016-05-20 16:57:41 -07001794 untrack_pfn(vma, remap_pfn, PAGE_ALIGN(size));
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001795
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 return err;
1797}
1798EXPORT_SYMBOL(remap_pfn_range);
1799
Linus Torvaldsb4cbb192013-04-16 13:45:37 -07001800/**
1801 * vm_iomap_memory - remap memory to userspace
1802 * @vma: user vma to map to
1803 * @start: start of area
1804 * @len: size of area
1805 *
1806 * This is a simplified io_remap_pfn_range() for common driver use. The
1807 * driver just needs to give us the physical memory range to be mapped,
1808 * we'll figure out the rest from the vma information.
1809 *
1810 * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get
1811 * whatever write-combining details or similar.
1812 */
1813int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1814{
1815 unsigned long vm_len, pfn, pages;
1816
1817 /* Check that the physical memory area passed in looks valid */
1818 if (start + len < start)
1819 return -EINVAL;
1820 /*
1821 * You *really* shouldn't map things that aren't page-aligned,
1822 * but we've historically allowed it because IO memory might
1823 * just have smaller alignment.
1824 */
1825 len += start & ~PAGE_MASK;
1826 pfn = start >> PAGE_SHIFT;
1827 pages = (len + ~PAGE_MASK) >> PAGE_SHIFT;
1828 if (pfn + pages < pfn)
1829 return -EINVAL;
1830
1831 /* We start the mapping 'vm_pgoff' pages into the area */
1832 if (vma->vm_pgoff > pages)
1833 return -EINVAL;
1834 pfn += vma->vm_pgoff;
1835 pages -= vma->vm_pgoff;
1836
1837 /* Can we fit all of the mapping? */
1838 vm_len = vma->vm_end - vma->vm_start;
1839 if (vm_len >> PAGE_SHIFT > pages)
1840 return -EINVAL;
1841
1842 /* Ok, let it rip */
1843 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1844}
1845EXPORT_SYMBOL(vm_iomap_memory);
1846
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001847static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1848 unsigned long addr, unsigned long end,
1849 pte_fn_t fn, void *data)
1850{
1851 pte_t *pte;
1852 int err;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001853 pgtable_t token;
Borislav Petkov94909912007-05-06 14:49:17 -07001854 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001855
1856 pte = (mm == &init_mm) ?
1857 pte_alloc_kernel(pmd, addr) :
1858 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1859 if (!pte)
1860 return -ENOMEM;
1861
1862 BUG_ON(pmd_huge(*pmd));
1863
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08001864 arch_enter_lazy_mmu_mode();
1865
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001866 token = pmd_pgtable(*pmd);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001867
1868 do {
Daisuke Nishimurac36987e2009-10-26 16:50:23 -07001869 err = fn(pte++, token, addr, data);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001870 if (err)
1871 break;
Daisuke Nishimurac36987e2009-10-26 16:50:23 -07001872 } while (addr += PAGE_SIZE, addr != end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001873
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08001874 arch_leave_lazy_mmu_mode();
1875
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001876 if (mm != &init_mm)
1877 pte_unmap_unlock(pte-1, ptl);
1878 return err;
1879}
1880
1881static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1882 unsigned long addr, unsigned long end,
1883 pte_fn_t fn, void *data)
1884{
1885 pmd_t *pmd;
1886 unsigned long next;
1887 int err;
1888
Andi Kleenceb86872008-07-23 21:27:50 -07001889 BUG_ON(pud_huge(*pud));
1890
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001891 pmd = pmd_alloc(mm, pud, addr);
1892 if (!pmd)
1893 return -ENOMEM;
1894 do {
1895 next = pmd_addr_end(addr, end);
1896 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1897 if (err)
1898 break;
1899 } while (pmd++, addr = next, addr != end);
1900 return err;
1901}
1902
1903static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1904 unsigned long addr, unsigned long end,
1905 pte_fn_t fn, void *data)
1906{
1907 pud_t *pud;
1908 unsigned long next;
1909 int err;
1910
1911 pud = pud_alloc(mm, pgd, addr);
1912 if (!pud)
1913 return -ENOMEM;
1914 do {
1915 next = pud_addr_end(addr, end);
1916 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1917 if (err)
1918 break;
1919 } while (pud++, addr = next, addr != end);
1920 return err;
1921}
1922
1923/*
1924 * Scan a region of virtual memory, filling in page tables as necessary
1925 * and calling a provided function on each leaf page table.
1926 */
1927int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1928 unsigned long size, pte_fn_t fn, void *data)
1929{
1930 pgd_t *pgd;
1931 unsigned long next;
Jeremy Fitzhardinge57250a52010-08-09 17:19:52 -07001932 unsigned long end = addr + size;
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001933 int err;
1934
Mika Penttilä9cb65bc2016-03-15 14:56:45 -07001935 if (WARN_ON(addr >= end))
1936 return -EINVAL;
1937
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001938 pgd = pgd_offset(mm, addr);
1939 do {
1940 next = pgd_addr_end(addr, end);
1941 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1942 if (err)
1943 break;
1944 } while (pgd++, addr = next, addr != end);
Jeremy Fitzhardinge57250a52010-08-09 17:19:52 -07001945
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001946 return err;
1947}
1948EXPORT_SYMBOL_GPL(apply_to_page_range);
1949
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950/*
Kirill A. Shutemov9b4bdd22015-02-10 14:09:51 -08001951 * handle_pte_fault chooses page fault handler according to an entry which was
1952 * read non-atomically. Before making any commitment, on those architectures
1953 * or configurations (e.g. i386 with PAE) which might give a mix of unmatched
1954 * parts, do_swap_page must check under lock before unmapping the pte and
1955 * proceeding (but do_wp_page is only called after already making such a check;
Ryota Ozakia335b2e2011-02-10 13:56:28 +09001956 * and do_anonymous_page can safely check later on).
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001957 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001958static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001959 pte_t *page_table, pte_t orig_pte)
1960{
1961 int same = 1;
1962#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1963 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001964 spinlock_t *ptl = pte_lockptr(mm, pmd);
1965 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001966 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001967 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001968 }
1969#endif
1970 pte_unmap(page_table);
1971 return same;
1972}
1973
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001974static 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 -08001975{
Dan Williams0abdd7a2014-01-21 15:48:12 -08001976 debug_dma_assert_idle(src);
1977
Linus Torvalds6aab3412005-11-28 14:34:23 -08001978 /*
1979 * If the source page was a PFN mapping, we don't have
1980 * a "struct page" for it. We do a best-effort copy by
1981 * just copying from the original user address. If that
1982 * fails, we just zero-fill it. Live with it.
1983 */
1984 if (unlikely(!src)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08001985 void *kaddr = kmap_atomic(dst);
Linus Torvalds5d2a2dbbc2005-11-29 14:07:55 -08001986 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1987
1988 /*
1989 * This really shouldn't fail, because the page is there
1990 * in the page tables. But it might just be unreadable,
1991 * in which case we just give up and fill the result with
1992 * zeroes.
1993 */
1994 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Jan Beulich3ecb01d2010-10-26 14:22:27 -07001995 clear_page(kaddr);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001996 kunmap_atomic(kaddr);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07001997 flush_dcache_page(dst);
Nick Piggin0ed361d2008-02-04 22:29:34 -08001998 } else
1999 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08002000}
2001
Michal Hockoc20cd452016-01-14 15:20:12 -08002002static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma)
2003{
2004 struct file *vm_file = vma->vm_file;
2005
2006 if (vm_file)
2007 return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO;
2008
2009 /*
2010 * Special mappings (e.g. VDSO) do not have any file so fake
2011 * a default GFP_KERNEL for them.
2012 */
2013 return GFP_KERNEL;
2014}
2015
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016/*
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07002017 * Notify the address space that the page is about to become writable so that
2018 * it can prohibit this or wait for the page to get into an appropriate state.
2019 *
2020 * We do this without the lock held, so that it can sleep if it needs to.
2021 */
2022static int do_page_mkwrite(struct vm_area_struct *vma, struct page *page,
2023 unsigned long address)
2024{
2025 struct vm_fault vmf;
2026 int ret;
2027
2028 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2029 vmf.pgoff = page->index;
2030 vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
Michal Hockoc20cd452016-01-14 15:20:12 -08002031 vmf.gfp_mask = __get_fault_gfp_mask(vma);
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07002032 vmf.page = page;
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08002033 vmf.cow_page = NULL;
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07002034
2035 ret = vma->vm_ops->page_mkwrite(vma, &vmf);
2036 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2037 return ret;
2038 if (unlikely(!(ret & VM_FAULT_LOCKED))) {
2039 lock_page(page);
2040 if (!page->mapping) {
2041 unlock_page(page);
2042 return 0; /* retry */
2043 }
2044 ret |= VM_FAULT_LOCKED;
2045 } else
2046 VM_BUG_ON_PAGE(!PageLocked(page), page);
2047 return ret;
2048}
2049
2050/*
Shachar Raindel4e047f82015-04-14 15:46:25 -07002051 * Handle write page faults for pages that can be reused in the current vma
2052 *
2053 * This can happen either due to the mapping being with the VM_SHARED flag,
2054 * or due to us being the last reference standing to the page. In either
2055 * case, all we need to do here is to mark the page as writable and update
2056 * any related book-keeping.
2057 */
2058static inline int wp_page_reuse(struct mm_struct *mm,
2059 struct vm_area_struct *vma, unsigned long address,
2060 pte_t *page_table, spinlock_t *ptl, pte_t orig_pte,
2061 struct page *page, int page_mkwrite,
2062 int dirty_shared)
2063 __releases(ptl)
2064{
2065 pte_t entry;
2066 /*
2067 * Clear the pages cpupid information as the existing
2068 * information potentially belongs to a now completely
2069 * unrelated process.
2070 */
2071 if (page)
2072 page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1);
2073
2074 flush_cache_page(vma, address, pte_pfn(orig_pte));
2075 entry = pte_mkyoung(orig_pte);
2076 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2077 if (ptep_set_access_flags(vma, address, page_table, entry, 1))
2078 update_mmu_cache(vma, address, page_table);
2079 pte_unmap_unlock(page_table, ptl);
2080
2081 if (dirty_shared) {
2082 struct address_space *mapping;
2083 int dirtied;
2084
2085 if (!page_mkwrite)
2086 lock_page(page);
2087
2088 dirtied = set_page_dirty(page);
2089 VM_BUG_ON_PAGE(PageAnon(page), page);
2090 mapping = page->mapping;
2091 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002092 put_page(page);
Shachar Raindel4e047f82015-04-14 15:46:25 -07002093
2094 if ((dirtied || page_mkwrite) && mapping) {
2095 /*
2096 * Some device drivers do not set page.mapping
2097 * but still dirty their pages
2098 */
2099 balance_dirty_pages_ratelimited(mapping);
2100 }
2101
2102 if (!page_mkwrite)
2103 file_update_time(vma->vm_file);
2104 }
2105
2106 return VM_FAULT_WRITE;
2107}
2108
2109/*
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002110 * Handle the case of a page which we actually need to copy to a new page.
2111 *
2112 * Called with mmap_sem locked and the old page referenced, but
2113 * without the ptl held.
2114 *
2115 * High level logic flow:
2116 *
2117 * - Allocate a page, copy the content of the old page to the new one.
2118 * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc.
2119 * - Take the PTL. If the pte changed, bail out and release the allocated page
2120 * - If the pte is still the way we remember it, update the page table and all
2121 * relevant references. This includes dropping the reference the page-table
2122 * held to the old page, as well as updating the rmap.
2123 * - In any case, unlock the PTL and drop the reference we took to the old page.
2124 */
2125static int wp_page_copy(struct mm_struct *mm, struct vm_area_struct *vma,
2126 unsigned long address, pte_t *page_table, pmd_t *pmd,
2127 pte_t orig_pte, struct page *old_page)
2128{
2129 struct page *new_page = NULL;
2130 spinlock_t *ptl = NULL;
2131 pte_t entry;
2132 int page_copied = 0;
2133 const unsigned long mmun_start = address & PAGE_MASK; /* For mmu_notifiers */
2134 const unsigned long mmun_end = mmun_start + PAGE_SIZE; /* For mmu_notifiers */
2135 struct mem_cgroup *memcg;
2136
2137 if (unlikely(anon_vma_prepare(vma)))
2138 goto oom;
2139
2140 if (is_zero_pfn(pte_pfn(orig_pte))) {
2141 new_page = alloc_zeroed_user_highpage_movable(vma, address);
2142 if (!new_page)
2143 goto oom;
2144 } else {
2145 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
2146 if (!new_page)
2147 goto oom;
2148 cow_user_page(new_page, old_page, address, vma);
2149 }
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002150
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08002151 if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg, false))
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002152 goto oom_free_new;
2153
Mel Gormaneb3c24f2015-06-24 16:57:27 -07002154 __SetPageUptodate(new_page);
2155
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002156 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
2157
2158 /*
2159 * Re-check the pte - we dropped the lock
2160 */
2161 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2162 if (likely(pte_same(*page_table, orig_pte))) {
2163 if (old_page) {
2164 if (!PageAnon(old_page)) {
Jerome Marchandeca56ff2016-01-14 15:19:26 -08002165 dec_mm_counter_fast(mm,
2166 mm_counter_file(old_page));
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002167 inc_mm_counter_fast(mm, MM_ANONPAGES);
2168 }
2169 } else {
2170 inc_mm_counter_fast(mm, MM_ANONPAGES);
2171 }
2172 flush_cache_page(vma, address, pte_pfn(orig_pte));
2173 entry = mk_pte(new_page, vma->vm_page_prot);
2174 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2175 /*
2176 * Clear the pte entry and flush it first, before updating the
2177 * pte with the new entry. This will avoid a race condition
2178 * seen in the presence of one thread doing SMC and another
2179 * thread doing COW.
2180 */
2181 ptep_clear_flush_notify(vma, address, page_table);
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08002182 page_add_new_anon_rmap(new_page, vma, address, false);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08002183 mem_cgroup_commit_charge(new_page, memcg, false, false);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002184 lru_cache_add_active_or_unevictable(new_page, vma);
2185 /*
2186 * We call the notify macro here because, when using secondary
2187 * mmu page tables (such as kvm shadow page tables), we want the
2188 * new page to be mapped directly into the secondary page table.
2189 */
2190 set_pte_at_notify(mm, address, page_table, entry);
2191 update_mmu_cache(vma, address, page_table);
2192 if (old_page) {
2193 /*
2194 * Only after switching the pte to the new page may
2195 * we remove the mapcount here. Otherwise another
2196 * process may come and find the rmap count decremented
2197 * before the pte is switched to the new page, and
2198 * "reuse" the old page writing into it while our pte
2199 * here still points into it and can be read by other
2200 * threads.
2201 *
2202 * The critical issue is to order this
2203 * page_remove_rmap with the ptp_clear_flush above.
2204 * Those stores are ordered by (if nothing else,)
2205 * the barrier present in the atomic_add_negative
2206 * in page_remove_rmap.
2207 *
2208 * Then the TLB flush in ptep_clear_flush ensures that
2209 * no process can access the old page before the
2210 * decremented mapcount is visible. And the old page
2211 * cannot be reused until after the decremented
2212 * mapcount is visible. So transitively, TLBs to
2213 * old page will be flushed before it can be reused.
2214 */
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08002215 page_remove_rmap(old_page, false);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002216 }
2217
2218 /* Free the old page.. */
2219 new_page = old_page;
2220 page_copied = 1;
2221 } else {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08002222 mem_cgroup_cancel_charge(new_page, memcg, false);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002223 }
2224
2225 if (new_page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002226 put_page(new_page);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002227
2228 pte_unmap_unlock(page_table, ptl);
2229 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
2230 if (old_page) {
2231 /*
2232 * Don't let another task, with possibly unlocked vma,
2233 * keep the mlocked page.
2234 */
2235 if (page_copied && (vma->vm_flags & VM_LOCKED)) {
2236 lock_page(old_page); /* LRU manipulation */
Kirill A. Shutemove90309c2016-01-15 16:54:33 -08002237 if (PageMlocked(old_page))
2238 munlock_vma_page(old_page);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002239 unlock_page(old_page);
2240 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002241 put_page(old_page);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002242 }
2243 return page_copied ? VM_FAULT_WRITE : 0;
2244oom_free_new:
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002245 put_page(new_page);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002246oom:
2247 if (old_page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002248 put_page(old_page);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002249 return VM_FAULT_OOM;
2250}
2251
Boaz Harroshdd906182015-04-15 16:15:11 -07002252/*
2253 * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED
2254 * mapping
2255 */
2256static int wp_pfn_shared(struct mm_struct *mm,
2257 struct vm_area_struct *vma, unsigned long address,
2258 pte_t *page_table, spinlock_t *ptl, pte_t orig_pte,
2259 pmd_t *pmd)
2260{
2261 if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) {
2262 struct vm_fault vmf = {
2263 .page = NULL,
2264 .pgoff = linear_page_index(vma, address),
2265 .virtual_address = (void __user *)(address & PAGE_MASK),
2266 .flags = FAULT_FLAG_WRITE | FAULT_FLAG_MKWRITE,
2267 };
2268 int ret;
2269
2270 pte_unmap_unlock(page_table, ptl);
2271 ret = vma->vm_ops->pfn_mkwrite(vma, &vmf);
2272 if (ret & VM_FAULT_ERROR)
2273 return ret;
2274 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2275 /*
2276 * We might have raced with another page fault while we
2277 * released the pte_offset_map_lock.
2278 */
2279 if (!pte_same(*page_table, orig_pte)) {
2280 pte_unmap_unlock(page_table, ptl);
2281 return 0;
2282 }
2283 }
2284 return wp_page_reuse(mm, vma, address, page_table, ptl, orig_pte,
2285 NULL, 0, 0);
2286}
2287
Shachar Raindel93e478d2015-04-14 15:46:35 -07002288static int wp_page_shared(struct mm_struct *mm, struct vm_area_struct *vma,
2289 unsigned long address, pte_t *page_table,
2290 pmd_t *pmd, spinlock_t *ptl, pte_t orig_pte,
2291 struct page *old_page)
2292 __releases(ptl)
2293{
2294 int page_mkwrite = 0;
2295
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002296 get_page(old_page);
Shachar Raindel93e478d2015-04-14 15:46:35 -07002297
Shachar Raindel93e478d2015-04-14 15:46:35 -07002298 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
2299 int tmp;
2300
2301 pte_unmap_unlock(page_table, ptl);
2302 tmp = do_page_mkwrite(vma, old_page, address);
2303 if (unlikely(!tmp || (tmp &
2304 (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002305 put_page(old_page);
Shachar Raindel93e478d2015-04-14 15:46:35 -07002306 return tmp;
2307 }
2308 /*
2309 * Since we dropped the lock we need to revalidate
2310 * the PTE as someone else may have changed it. If
2311 * they did, we just return, as we can count on the
2312 * MMU to tell us if they didn't also make it writable.
2313 */
2314 page_table = pte_offset_map_lock(mm, pmd, address,
2315 &ptl);
2316 if (!pte_same(*page_table, orig_pte)) {
2317 unlock_page(old_page);
2318 pte_unmap_unlock(page_table, ptl);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002319 put_page(old_page);
Shachar Raindel93e478d2015-04-14 15:46:35 -07002320 return 0;
2321 }
2322 page_mkwrite = 1;
2323 }
2324
2325 return wp_page_reuse(mm, vma, address, page_table, ptl,
2326 orig_pte, old_page, page_mkwrite, 1);
2327}
2328
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002329/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 * This routine handles present pages, when users try to write
2331 * to a shared page. It is done by copying the page to a new address
2332 * and decrementing the shared-page counter for the old page.
2333 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 * Note that this routine assumes that the protection checks have been
2335 * done by the caller (the low-level page fault routine in most cases).
2336 * Thus we can safely just mark it writable once we've done any necessary
2337 * COW.
2338 *
2339 * We also mark the page dirty at this point even though the page will
2340 * change only once the write actually happens. This avoids a few races,
2341 * and potentially makes it more efficient.
2342 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002343 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2344 * but allow concurrent faults), with pte both mapped and locked.
2345 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002347static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
2348 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002349 spinlock_t *ptl, pte_t orig_pte)
Namhyung Kime6219ec2010-10-26 14:22:00 -07002350 __releases(ptl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351{
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002352 struct page *old_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353
Linus Torvalds6aab3412005-11-28 14:34:23 -08002354 old_page = vm_normal_page(vma, address, orig_pte);
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002355 if (!old_page) {
2356 /*
Peter Feiner64e45502014-10-13 15:55:46 -07002357 * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a
2358 * VM_PFNMAP VMA.
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002359 *
2360 * We should not cow pages in a shared writeable mapping.
Boaz Harroshdd906182015-04-15 16:15:11 -07002361 * Just mark the pages writable and/or call ops->pfn_mkwrite.
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002362 */
2363 if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
2364 (VM_WRITE|VM_SHARED))
Boaz Harroshdd906182015-04-15 16:15:11 -07002365 return wp_pfn_shared(mm, vma, address, page_table, ptl,
2366 orig_pte, pmd);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002367
2368 pte_unmap_unlock(page_table, ptl);
2369 return wp_page_copy(mm, vma, address, page_table, pmd,
2370 orig_pte, old_page);
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002371 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002373 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002374 * Take out anonymous pages first, anonymous shared vmas are
2375 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002376 */
Hugh Dickins9a840892009-09-21 17:02:01 -07002377 if (PageAnon(old_page) && !PageKsm(old_page)) {
Andrea Arcangeli6d0a07e2016-05-12 15:42:25 -07002378 int total_mapcount;
Hugh Dickinsab967d82009-01-06 14:39:33 -08002379 if (!trylock_page(old_page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002380 get_page(old_page);
Hugh Dickinsab967d82009-01-06 14:39:33 -08002381 pte_unmap_unlock(page_table, ptl);
2382 lock_page(old_page);
2383 page_table = pte_offset_map_lock(mm, pmd, address,
2384 &ptl);
2385 if (!pte_same(*page_table, orig_pte)) {
2386 unlock_page(old_page);
Shachar Raindel28766802015-04-14 15:46:29 -07002387 pte_unmap_unlock(page_table, ptl);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002388 put_page(old_page);
Shachar Raindel28766802015-04-14 15:46:29 -07002389 return 0;
Hugh Dickinsab967d82009-01-06 14:39:33 -08002390 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002391 put_page(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002392 }
Andrea Arcangeli6d0a07e2016-05-12 15:42:25 -07002393 if (reuse_swap_page(old_page, &total_mapcount)) {
2394 if (total_mapcount == 1) {
2395 /*
2396 * The page is all ours. Move it to
2397 * our anon_vma so the rmap code will
2398 * not search our parent or siblings.
2399 * Protected against the rmap code by
2400 * the page lock.
2401 */
2402 page_move_anon_rmap(compound_head(old_page),
2403 vma, address);
2404 }
Michel Lespinasseb009c022011-01-13 15:46:07 -08002405 unlock_page(old_page);
Shachar Raindel4e047f82015-04-14 15:46:25 -07002406 return wp_page_reuse(mm, vma, address, page_table, ptl,
2407 orig_pte, old_page, 0, 0);
Michel Lespinasseb009c022011-01-13 15:46:07 -08002408 }
Hugh Dickinsab967d82009-01-06 14:39:33 -08002409 unlock_page(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002410 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002411 (VM_WRITE|VM_SHARED))) {
Shachar Raindel93e478d2015-04-14 15:46:35 -07002412 return wp_page_shared(mm, vma, address, page_table, pmd,
2413 ptl, orig_pte, old_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415
2416 /*
2417 * Ok, we need to copy. Oh, well..
2418 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002419 get_page(old_page);
Shachar Raindel28766802015-04-14 15:46:29 -07002420
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002421 pte_unmap_unlock(page_table, ptl);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002422 return wp_page_copy(mm, vma, address, page_table, pmd,
2423 orig_pte, old_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424}
2425
Peter Zijlstra97a89412011-05-24 17:12:04 -07002426static void unmap_mapping_range_vma(struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 unsigned long start_addr, unsigned long end_addr,
2428 struct zap_details *details)
2429{
Al Virof5cc4ee2012-03-05 14:14:20 -05002430 zap_page_range_single(vma, start_addr, end_addr - start_addr, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431}
2432
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07002433static inline void unmap_mapping_range_tree(struct rb_root *root,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434 struct zap_details *details)
2435{
2436 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 pgoff_t vba, vea, zba, zea;
2438
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07002439 vma_interval_tree_foreach(vma, root,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 details->first_index, details->last_index) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441
2442 vba = vma->vm_pgoff;
Libind6e93212013-07-03 15:01:26 -07002443 vea = vba + vma_pages(vma) - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 zba = details->first_index;
2445 if (zba < vba)
2446 zba = vba;
2447 zea = details->last_index;
2448 if (zea > vea)
2449 zea = vea;
2450
Peter Zijlstra97a89412011-05-24 17:12:04 -07002451 unmap_mapping_range_vma(vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
2453 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
Peter Zijlstra97a89412011-05-24 17:12:04 -07002454 details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455 }
2456}
2457
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458/**
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08002459 * unmap_mapping_range - unmap the portion of all mmaps in the specified
2460 * address_space corresponding to the specified page range in the underlying
2461 * file.
2462 *
Martin Waitz3d410882005-06-23 22:05:21 -07002463 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464 * @holebegin: byte in first page to unmap, relative to the start of
2465 * the underlying file. This will be rounded down to a PAGE_SIZE
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +10002466 * boundary. Note that this is different from truncate_pagecache(), which
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 * must keep the partial page. In contrast, we must get rid of
2468 * partial pages.
2469 * @holelen: size of prospective hole in bytes. This will be rounded
2470 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2471 * end of the file.
2472 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2473 * but 0 when invalidating pagecache, don't throw away private data.
2474 */
2475void unmap_mapping_range(struct address_space *mapping,
2476 loff_t const holebegin, loff_t const holelen, int even_cows)
2477{
Michal Hockoaac45362016-03-25 14:20:24 -07002478 struct zap_details details = { };
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 pgoff_t hba = holebegin >> PAGE_SHIFT;
2480 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2481
2482 /* Check for overflow. */
2483 if (sizeof(holelen) > sizeof(hlen)) {
2484 long long holeend =
2485 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2486 if (holeend & ~(long long)ULONG_MAX)
2487 hlen = ULONG_MAX - hba + 1;
2488 }
2489
2490 details.check_mapping = even_cows? NULL: mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 details.first_index = hba;
2492 details.last_index = hba + hlen - 1;
2493 if (details.last_index < details.first_index)
2494 details.last_index = ULONG_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495
Kirill A. Shutemov46c043e2015-09-08 14:59:42 -07002496 i_mmap_lock_write(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07002497 if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498 unmap_mapping_range_tree(&mapping->i_mmap, &details);
Kirill A. Shutemov46c043e2015-09-08 14:59:42 -07002499 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500}
2501EXPORT_SYMBOL(unmap_mapping_range);
2502
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002504 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2505 * but allow concurrent faults), and pte mapped but not yet locked.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07002506 * We return with pte unmapped and unlocked.
2507 *
2508 * We return with the mmap_sem locked or unlocked in the same cases
2509 * as does filemap_fault().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002511static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2512 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002513 unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002515 spinlock_t *ptl;
Hugh Dickins56f31802013-02-22 16:36:10 -08002516 struct page *page, *swapcache;
Johannes Weiner00501b52014-08-08 14:19:20 -07002517 struct mem_cgroup *memcg;
Hugh Dickins65500d22005-10-29 18:15:59 -07002518 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519 pte_t pte;
Michel Lespinassed065bd82010-10-26 14:21:57 -07002520 int locked;
Rik van Rielad8c2ee2010-08-09 17:19:48 -07002521 int exclusive = 0;
Nick Piggin83c54072007-07-19 01:47:05 -07002522 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002524 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002525 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002526
2527 entry = pte_to_swp_entry(orig_pte);
Andi Kleend1737fd2009-09-16 11:50:06 +02002528 if (unlikely(non_swap_entry(entry))) {
2529 if (is_migration_entry(entry)) {
2530 migration_entry_wait(mm, pmd, address);
2531 } else if (is_hwpoison_entry(entry)) {
2532 ret = VM_FAULT_HWPOISON;
2533 } else {
2534 print_bad_pte(vma, address, orig_pte, NULL);
Hugh Dickinsd99be1a2009-12-14 17:59:04 -08002535 ret = VM_FAULT_SIGBUS;
Andi Kleend1737fd2009-09-16 11:50:06 +02002536 }
Christoph Lameter06972122006-06-23 02:03:35 -07002537 goto out;
2538 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002539 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 page = lookup_swap_cache(entry);
2541 if (!page) {
Hugh Dickins02098fe2008-02-04 22:28:42 -08002542 page = swapin_readahead(entry,
2543 GFP_HIGHUSER_MOVABLE, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 if (!page) {
2545 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002546 * Back out if somebody else faulted in this pte
2547 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002549 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 if (likely(pte_same(*page_table, orig_pte)))
2551 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002552 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002553 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 }
2555
2556 /* Had to read the page from swap area: Major fault */
2557 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002558 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07002559 mem_cgroup_count_vm_event(mm, PGMAJFAULT);
Andi Kleend1737fd2009-09-16 11:50:06 +02002560 } else if (PageHWPoison(page)) {
Wu Fengguang71f72522009-12-16 12:19:58 +01002561 /*
2562 * hwpoisoned dirty swapcache pages are kept for killing
2563 * owner processes (which may be unknown at hwpoison time)
2564 */
Andi Kleend1737fd2009-09-16 11:50:06 +02002565 ret = VM_FAULT_HWPOISON;
2566 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins56f31802013-02-22 16:36:10 -08002567 swapcache = page;
Andi Kleen4779cb32009-10-14 01:51:41 +02002568 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569 }
2570
Hugh Dickins56f31802013-02-22 16:36:10 -08002571 swapcache = page;
Michel Lespinassed065bd82010-10-26 14:21:57 -07002572 locked = lock_page_or_retry(page, mm, flags);
Rik van Riele709ffd2012-05-29 15:06:18 -07002573
Balbir Singh20a10222007-11-14 17:00:33 -08002574 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Michel Lespinassed065bd82010-10-26 14:21:57 -07002575 if (!locked) {
2576 ret |= VM_FAULT_RETRY;
2577 goto out_release;
2578 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002580 /*
Hugh Dickins31c4a3d2010-09-19 19:40:22 -07002581 * Make sure try_to_free_swap or reuse_swap_page or swapoff did not
2582 * release the swapcache from under us. The page pin, and pte_same
2583 * test below, are not enough to exclude that. Even if it is still
2584 * swapcache, we need to check that the page's swap has not changed.
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002585 */
Hugh Dickins31c4a3d2010-09-19 19:40:22 -07002586 if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val))
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002587 goto out_page;
2588
Hugh Dickinscbf86cf2013-02-22 16:35:08 -08002589 page = ksm_might_need_to_copy(page, vma, address);
2590 if (unlikely(!page)) {
2591 ret = VM_FAULT_OOM;
2592 page = swapcache;
Hugh Dickinscbf86cf2013-02-22 16:35:08 -08002593 goto out_page;
Hugh Dickins5ad64682009-12-14 17:59:24 -08002594 }
2595
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08002596 if (mem_cgroup_try_charge(page, mm, GFP_KERNEL, &memcg, false)) {
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002597 ret = VM_FAULT_OOM;
Johannes Weinerbc43f752009-04-30 15:08:08 -07002598 goto out_page;
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002599 }
2600
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002602 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002604 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002605 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002606 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002607
2608 if (unlikely(!PageUptodate(page))) {
2609 ret = VM_FAULT_SIGBUS;
2610 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611 }
2612
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002613 /*
2614 * The page isn't present yet, go ahead with the fault.
2615 *
2616 * Be careful about the sequence of operations here.
2617 * To get its accounting right, reuse_swap_page() must be called
2618 * while the page is counted on swap but not yet in mapcount i.e.
2619 * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
2620 * must be called after the swap_free(), or it will never succeed.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002621 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08002623 inc_mm_counter_fast(mm, MM_ANONPAGES);
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08002624 dec_mm_counter_fast(mm, MM_SWAPENTS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 pte = mk_pte(page, vma->vm_page_prot);
Andrea Arcangeli6d0a07e2016-05-12 15:42:25 -07002626 if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002628 flags &= ~FAULT_FLAG_WRITE;
Andrea Arcangeli9a5b4892010-08-09 17:19:49 -07002629 ret |= VM_FAULT_WRITE;
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08002630 exclusive = RMAP_EXCLUSIVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 flush_icache_page(vma, page);
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07002633 if (pte_swp_soft_dirty(orig_pte))
2634 pte = pte_mksoft_dirty(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 set_pte_at(mm, address, page_table, pte);
Johannes Weiner00501b52014-08-08 14:19:20 -07002636 if (page == swapcache) {
Johannes Weineraf347702013-02-22 16:32:20 -08002637 do_page_add_anon_rmap(page, vma, address, exclusive);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08002638 mem_cgroup_commit_charge(page, memcg, true, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07002639 } else { /* ksm created a completely new copy */
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08002640 page_add_new_anon_rmap(page, vma, address, false);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08002641 mem_cgroup_commit_charge(page, memcg, false, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07002642 lru_cache_add_active_or_unevictable(page, vma);
2643 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002645 swap_free(entry);
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08002646 if (mem_cgroup_swap_full(page) ||
2647 (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08002648 try_to_free_swap(page);
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002649 unlock_page(page);
Hugh Dickins56f31802013-02-22 16:36:10 -08002650 if (page != swapcache) {
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002651 /*
2652 * Hold the lock to avoid the swap entry to be reused
2653 * until we take the PT lock for the pte_same() check
2654 * (to avoid false positives from pte_same). For
2655 * further safety release the lock after the swap_free
2656 * so that the swap count won't change under a
2657 * parallel locked swapcache.
2658 */
2659 unlock_page(swapcache);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002660 put_page(swapcache);
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002661 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002662
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002663 if (flags & FAULT_FLAG_WRITE) {
Hugh Dickins61469f12008-03-04 14:29:04 -08002664 ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
2665 if (ret & VM_FAULT_ERROR)
2666 ret &= VM_FAULT_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 goto out;
2668 }
2669
2670 /* No need to invalidate - it was non-present before */
Russell King4b3073e2009-12-18 16:40:18 +00002671 update_mmu_cache(vma, address, page_table);
Hugh Dickins65500d22005-10-29 18:15:59 -07002672unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002673 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674out:
2675 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002676out_nomap:
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08002677 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002678 pte_unmap_unlock(page_table, ptl);
Johannes Weinerbc43f752009-04-30 15:08:08 -07002679out_page:
Kirill Korotaevb8107482005-05-16 21:53:50 -07002680 unlock_page(page);
Andi Kleen4779cb32009-10-14 01:51:41 +02002681out_release:
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002682 put_page(page);
Hugh Dickins56f31802013-02-22 16:36:10 -08002683 if (page != swapcache) {
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002684 unlock_page(swapcache);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002685 put_page(swapcache);
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002686 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002687 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688}
2689
2690/*
Luck, Tony8ca3eb02010-08-24 11:44:18 -07002691 * This is like a special single-page "expand_{down|up}wards()",
2692 * except we must first make sure that 'address{-|+}PAGE_SIZE'
Linus Torvalds320b2b82010-08-12 17:54:33 -07002693 * doesn't hit another vma.
Linus Torvalds320b2b82010-08-12 17:54:33 -07002694 */
2695static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address)
2696{
2697 address &= PAGE_MASK;
2698 if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) {
Linus Torvalds0e8e50e2010-08-20 16:49:40 -07002699 struct vm_area_struct *prev = vma->vm_prev;
Linus Torvalds320b2b82010-08-12 17:54:33 -07002700
Linus Torvalds0e8e50e2010-08-20 16:49:40 -07002701 /*
2702 * Is there a mapping abutting this one below?
2703 *
2704 * That's only ok if it's the same stack mapping
2705 * that has gotten split..
2706 */
2707 if (prev && prev->vm_end == address)
2708 return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM;
2709
Linus Torvaldsfee7e492015-01-06 13:00:05 -08002710 return expand_downwards(vma, address - PAGE_SIZE);
Linus Torvalds320b2b82010-08-12 17:54:33 -07002711 }
Luck, Tony8ca3eb02010-08-24 11:44:18 -07002712 if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) {
2713 struct vm_area_struct *next = vma->vm_next;
2714
2715 /* As VM_GROWSDOWN but s/below/above/ */
2716 if (next && next->vm_start == address + PAGE_SIZE)
2717 return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM;
2718
Linus Torvaldsfee7e492015-01-06 13:00:05 -08002719 return expand_upwards(vma, address + PAGE_SIZE);
Luck, Tony8ca3eb02010-08-24 11:44:18 -07002720 }
Linus Torvalds320b2b82010-08-12 17:54:33 -07002721 return 0;
2722}
2723
2724/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002725 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2726 * but allow concurrent faults), and pte mapped but not yet locked.
2727 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002729static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2730 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002731 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732{
Johannes Weiner00501b52014-08-08 14:19:20 -07002733 struct mem_cgroup *memcg;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002734 struct page *page;
2735 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737
Linus Torvalds11ac5522010-08-14 11:44:56 -07002738 pte_unmap(page_table);
Linus Torvalds320b2b82010-08-12 17:54:33 -07002739
Kirill A. Shutemov6b7339f2015-07-06 23:18:37 +03002740 /* File mapping without ->vm_ops ? */
2741 if (vma->vm_flags & VM_SHARED)
2742 return VM_FAULT_SIGBUS;
2743
Linus Torvalds11ac5522010-08-14 11:44:56 -07002744 /* Check if we need to add a guard page to the stack */
2745 if (check_stack_guard_page(vma, address) < 0)
Linus Torvalds9c145c562015-01-29 11:15:17 -08002746 return VM_FAULT_SIGSEGV;
Linus Torvalds11ac5522010-08-14 11:44:56 -07002747
2748 /* Use the zero-page for reads */
Dominik Dingel593befa2014-10-23 12:07:44 +02002749 if (!(flags & FAULT_FLAG_WRITE) && !mm_forbids_zeropage(mm)) {
Hugh Dickins62eede62009-09-21 17:03:34 -07002750 entry = pte_mkspecial(pfn_pte(my_zero_pfn(address),
2751 vma->vm_page_prot));
Linus Torvalds11ac5522010-08-14 11:44:56 -07002752 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002753 if (!pte_none(*page_table))
2754 goto unlock;
Andrea Arcangeli6b251fc2015-09-04 15:46:20 -07002755 /* Deliver the page fault to userland, check inside PT lock */
2756 if (userfaultfd_missing(vma)) {
2757 pte_unmap_unlock(page_table, ptl);
2758 return handle_userfault(vma, address, flags,
2759 VM_UFFD_MISSING);
2760 }
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002761 goto setpte;
2762 }
2763
Nick Piggin557ed1f2007-10-16 01:24:40 -07002764 /* Allocate our own private page. */
Nick Piggin557ed1f2007-10-16 01:24:40 -07002765 if (unlikely(anon_vma_prepare(vma)))
2766 goto oom;
2767 page = alloc_zeroed_user_highpage_movable(vma, address);
2768 if (!page)
2769 goto oom;
Mel Gormaneb3c24f2015-06-24 16:57:27 -07002770
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08002771 if (mem_cgroup_try_charge(page, mm, GFP_KERNEL, &memcg, false))
Mel Gormaneb3c24f2015-06-24 16:57:27 -07002772 goto oom_free_page;
2773
Minchan Kim52f37622013-04-29 15:08:15 -07002774 /*
2775 * The memory barrier inside __SetPageUptodate makes sure that
2776 * preceeding stores to the page contents become visible before
2777 * the set_pte_at() write.
2778 */
Nick Piggin0ed361d2008-02-04 22:29:34 -08002779 __SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780
Nick Piggin557ed1f2007-10-16 01:24:40 -07002781 entry = mk_pte(page, vma->vm_page_prot);
Hugh Dickins1ac0cb52009-09-21 17:03:29 -07002782 if (vma->vm_flags & VM_WRITE)
2783 entry = pte_mkwrite(pte_mkdirty(entry));
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002784
Nick Piggin557ed1f2007-10-16 01:24:40 -07002785 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Andrea Arcangeli1c2fb7a2009-09-21 17:02:22 -07002786 if (!pte_none(*page_table))
Nick Piggin557ed1f2007-10-16 01:24:40 -07002787 goto release;
Hugh Dickins9ba69292009-09-21 17:02:20 -07002788
Andrea Arcangeli6b251fc2015-09-04 15:46:20 -07002789 /* Deliver the page fault to userland, check inside PT lock */
2790 if (userfaultfd_missing(vma)) {
2791 pte_unmap_unlock(page_table, ptl);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08002792 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002793 put_page(page);
Andrea Arcangeli6b251fc2015-09-04 15:46:20 -07002794 return handle_userfault(vma, address, flags,
2795 VM_UFFD_MISSING);
2796 }
2797
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08002798 inc_mm_counter_fast(mm, MM_ANONPAGES);
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08002799 page_add_new_anon_rmap(page, vma, address, false);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08002800 mem_cgroup_commit_charge(page, memcg, false, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07002801 lru_cache_add_active_or_unevictable(page, vma);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002802setpte:
Hugh Dickins65500d22005-10-29 18:15:59 -07002803 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804
2805 /* No need to invalidate - it was non-present before */
Russell King4b3073e2009-12-18 16:40:18 +00002806 update_mmu_cache(vma, address, page_table);
Hugh Dickins65500d22005-10-29 18:15:59 -07002807unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002808 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002809 return 0;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002810release:
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08002811 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002812 put_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002813 goto unlock;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002814oom_free_page:
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002815 put_page(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002816oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817 return VM_FAULT_OOM;
2818}
2819
Paul Cassella9a95f3c2014-08-06 16:07:24 -07002820/*
2821 * The mmap_sem must have been held on entry, and may have been
2822 * released depending on flags and vma->vm_ops->fault() return value.
2823 * See filemap_fault() and __lock_page_retry().
2824 */
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07002825static int __do_fault(struct vm_area_struct *vma, unsigned long address,
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08002826 pgoff_t pgoff, unsigned int flags,
Jan Karabc2466e2016-05-12 18:29:19 +02002827 struct page *cow_page, struct page **page,
2828 void **entry)
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07002829{
2830 struct vm_fault vmf;
2831 int ret;
2832
2833 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2834 vmf.pgoff = pgoff;
2835 vmf.flags = flags;
2836 vmf.page = NULL;
Michal Hockoc20cd452016-01-14 15:20:12 -08002837 vmf.gfp_mask = __get_fault_gfp_mask(vma);
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08002838 vmf.cow_page = cow_page;
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07002839
2840 ret = vma->vm_ops->fault(vma, &vmf);
2841 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
2842 return ret;
Jan Karabc2466e2016-05-12 18:29:19 +02002843 if (ret & VM_FAULT_DAX_LOCKED) {
2844 *entry = vmf.entry;
2845 return ret;
2846 }
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07002847
2848 if (unlikely(PageHWPoison(vmf.page))) {
2849 if (ret & VM_FAULT_LOCKED)
2850 unlock_page(vmf.page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002851 put_page(vmf.page);
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07002852 return VM_FAULT_HWPOISON;
2853 }
2854
2855 if (unlikely(!(ret & VM_FAULT_LOCKED)))
2856 lock_page(vmf.page);
2857 else
2858 VM_BUG_ON_PAGE(!PageLocked(vmf.page), vmf.page);
2859
2860 *page = vmf.page;
2861 return ret;
2862}
2863
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002864/**
2865 * do_set_pte - setup new PTE entry for given page and add reverse page mapping.
2866 *
2867 * @vma: virtual memory area
2868 * @address: user virtual address
2869 * @page: page to map
2870 * @pte: pointer to target page table entry
2871 * @write: true, if new entry is writable
2872 * @anon: true, if it's anonymous page
2873 *
2874 * Caller must hold page table lock relevant for @pte.
2875 *
2876 * Target users are page handler itself and implementations of
2877 * vm_ops->map_pages.
2878 */
2879void do_set_pte(struct vm_area_struct *vma, unsigned long address,
Kirill A. Shutemov5c0a85f2016-05-20 16:58:41 -07002880 struct page *page, pte_t *pte, bool write, bool anon, bool old)
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002881{
2882 pte_t entry;
2883
2884 flush_icache_page(vma, page);
2885 entry = mk_pte(page, vma->vm_page_prot);
2886 if (write)
2887 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Kirill A. Shutemov5c0a85f2016-05-20 16:58:41 -07002888 if (old)
2889 entry = pte_mkold(entry);
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002890 if (anon) {
2891 inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES);
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08002892 page_add_new_anon_rmap(page, vma, address, false);
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002893 } else {
Jerome Marchandeca56ff2016-01-14 15:19:26 -08002894 inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page));
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07002895 page_add_file_rmap(page);
2896 }
2897 set_pte_at(vma->vm_mm, address, pte, entry);
2898
2899 /* no need to invalidate: a not-present page won't be cached */
2900 update_mmu_cache(vma, address, pte);
2901}
2902
Minchan Kimd0834a62016-05-20 16:58:44 -07002903/*
2904 * If architecture emulates "accessed" or "young" bit without HW support,
2905 * there is no much gain with fault_around.
2906 */
Kirill A. Shutemov3a910532014-08-06 16:08:07 -07002907static unsigned long fault_around_bytes __read_mostly =
Minchan Kimd0834a62016-05-20 16:58:44 -07002908#ifndef __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
2909 PAGE_SIZE;
2910#else
Kirill A. Shutemov3a910532014-08-06 16:08:07 -07002911 rounddown_pow_of_two(65536);
Minchan Kimd0834a62016-05-20 16:58:44 -07002912#endif
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002913
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002914#ifdef CONFIG_DEBUG_FS
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002915static int fault_around_bytes_get(void *data, u64 *val)
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002916{
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002917 *val = fault_around_bytes;
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002918 return 0;
2919}
2920
Andrey Ryabininb4903d62014-07-30 16:08:35 -07002921/*
2922 * fault_around_pages() and fault_around_mask() expects fault_around_bytes
2923 * rounded down to nearest page order. It's what do_fault_around() expects to
2924 * see.
2925 */
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002926static int fault_around_bytes_set(void *data, u64 val)
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002927{
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002928 if (val / PAGE_SIZE > PTRS_PER_PTE)
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002929 return -EINVAL;
Andrey Ryabininb4903d62014-07-30 16:08:35 -07002930 if (val > PAGE_SIZE)
2931 fault_around_bytes = rounddown_pow_of_two(val);
2932 else
2933 fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002934 return 0;
2935}
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002936DEFINE_SIMPLE_ATTRIBUTE(fault_around_bytes_fops,
2937 fault_around_bytes_get, fault_around_bytes_set, "%llu\n");
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002938
2939static int __init fault_around_debugfs(void)
2940{
2941 void *ret;
2942
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002943 ret = debugfs_create_file("fault_around_bytes", 0644, NULL, NULL,
2944 &fault_around_bytes_fops);
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002945 if (!ret)
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07002946 pr_warn("Failed to create fault_around_bytes in debugfs");
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002947 return 0;
2948}
2949late_initcall(fault_around_debugfs);
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07002950#endif
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002951
Kirill A. Shutemov1fdb4122014-06-04 16:10:55 -07002952/*
2953 * do_fault_around() tries to map few pages around the fault address. The hope
2954 * is that the pages will be needed soon and this will lower the number of
2955 * faults to handle.
2956 *
2957 * It uses vm_ops->map_pages() to map the pages, which skips the page if it's
2958 * not ready to be mapped: not up-to-date, locked, etc.
2959 *
2960 * This function is called with the page table lock taken. In the split ptlock
2961 * case the page table lock only protects only those entries which belong to
2962 * the page table corresponding to the fault address.
2963 *
2964 * This function doesn't cross the VMA boundaries, in order to call map_pages()
2965 * only once.
2966 *
2967 * fault_around_pages() defines how many pages we'll try to map.
2968 * do_fault_around() expects it to return a power of two less than or equal to
2969 * PTRS_PER_PTE.
2970 *
2971 * The virtual address of the area that we map is naturally aligned to the
2972 * fault_around_pages() value (and therefore to page order). This way it's
2973 * easier to guarantee that we don't cross page table boundaries.
2974 */
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002975static void do_fault_around(struct vm_area_struct *vma, unsigned long address,
2976 pte_t *pte, pgoff_t pgoff, unsigned int flags)
2977{
Kirill A. Shutemovaecd6f42014-08-06 16:08:05 -07002978 unsigned long start_addr, nr_pages, mask;
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002979 pgoff_t max_pgoff;
2980 struct vm_fault vmf;
2981 int off;
2982
Jason Low4db0c3c2015-04-15 16:14:08 -07002983 nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT;
Kirill A. Shutemovaecd6f42014-08-06 16:08:05 -07002984 mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK;
2985
2986 start_addr = max(address & mask, vma->vm_start);
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002987 off = ((address - start_addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
2988 pte -= off;
2989 pgoff -= off;
2990
2991 /*
2992 * max_pgoff is either end of page table or end of vma
Kirill A. Shutemov850e9c62014-06-04 16:10:45 -07002993 * or fault_around_pages() from pgoff, depending what is nearest.
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002994 */
2995 max_pgoff = pgoff - ((start_addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) +
2996 PTRS_PER_PTE - 1;
2997 max_pgoff = min3(max_pgoff, vma_pages(vma) + vma->vm_pgoff - 1,
Kirill A. Shutemovaecd6f42014-08-06 16:08:05 -07002998 pgoff + nr_pages - 1);
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07002999
3000 /* Check if it makes any sense to call ->map_pages */
3001 while (!pte_none(*pte)) {
3002 if (++pgoff > max_pgoff)
3003 return;
3004 start_addr += PAGE_SIZE;
3005 if (start_addr >= vma->vm_end)
3006 return;
3007 pte++;
3008 }
3009
3010 vmf.virtual_address = (void __user *) start_addr;
3011 vmf.pte = pte;
3012 vmf.pgoff = pgoff;
3013 vmf.max_pgoff = max_pgoff;
3014 vmf.flags = flags;
Michal Hockoc20cd452016-01-14 15:20:12 -08003015 vmf.gfp_mask = __get_fault_gfp_mask(vma);
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003016 vma->vm_ops->map_pages(vma, &vmf);
3017}
3018
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07003019static int do_read_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3020 unsigned long address, pmd_t *pmd,
3021 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
3022{
3023 struct page *fault_page;
3024 spinlock_t *ptl;
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003025 pte_t *pte;
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003026 int ret = 0;
3027
3028 /*
3029 * Let's call ->map_pages() first and use ->fault() as fallback
3030 * if page by the offset is not ready to be mapped (cold cache or
3031 * something).
3032 */
Kirill A. Shutemov9b4bdd22015-02-10 14:09:51 -08003033 if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) {
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003034 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003035 if (!pte_same(*pte, orig_pte))
3036 goto unlock_out;
Kirill A. Shutemov5c0a85f2016-05-20 16:58:41 -07003037 do_fault_around(vma, address, pte, pgoff, flags);
3038 /* Check if the fault is handled by faultaround */
3039 if (!pte_same(*pte, orig_pte)) {
3040 /*
3041 * Faultaround produce old pte, but the pte we've
3042 * handler fault for should be young.
3043 */
3044 pte_t entry = pte_mkyoung(*pte);
3045 if (ptep_set_access_flags(vma, address, pte, entry, 0))
3046 update_mmu_cache(vma, address, pte);
3047 goto unlock_out;
3048 }
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003049 pte_unmap_unlock(pte, ptl);
3050 }
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07003051
Jan Karabc2466e2016-05-12 18:29:19 +02003052 ret = __do_fault(vma, address, pgoff, flags, NULL, &fault_page, NULL);
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07003053 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
3054 return ret;
3055
3056 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
3057 if (unlikely(!pte_same(*pte, orig_pte))) {
3058 pte_unmap_unlock(pte, ptl);
3059 unlock_page(fault_page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003060 put_page(fault_page);
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07003061 return ret;
3062 }
Kirill A. Shutemov5c0a85f2016-05-20 16:58:41 -07003063 do_set_pte(vma, address, fault_page, pte, false, false, false);
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07003064 unlock_page(fault_page);
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003065unlock_out:
3066 pte_unmap_unlock(pte, ptl);
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07003067 return ret;
3068}
3069
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003070static int do_cow_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3071 unsigned long address, pmd_t *pmd,
3072 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
3073{
3074 struct page *fault_page, *new_page;
Jan Karabc2466e2016-05-12 18:29:19 +02003075 void *fault_entry;
Johannes Weiner00501b52014-08-08 14:19:20 -07003076 struct mem_cgroup *memcg;
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003077 spinlock_t *ptl;
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003078 pte_t *pte;
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003079 int ret;
3080
3081 if (unlikely(anon_vma_prepare(vma)))
3082 return VM_FAULT_OOM;
3083
3084 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
3085 if (!new_page)
3086 return VM_FAULT_OOM;
3087
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08003088 if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg, false)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003089 put_page(new_page);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003090 return VM_FAULT_OOM;
3091 }
3092
Jan Karabc2466e2016-05-12 18:29:19 +02003093 ret = __do_fault(vma, address, pgoff, flags, new_page, &fault_page,
3094 &fault_entry);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003095 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
3096 goto uncharge_out;
3097
Jan Karabc2466e2016-05-12 18:29:19 +02003098 if (!(ret & VM_FAULT_DAX_LOCKED))
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08003099 copy_user_highpage(new_page, fault_page, address, vma);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003100 __SetPageUptodate(new_page);
3101
3102 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
3103 if (unlikely(!pte_same(*pte, orig_pte))) {
3104 pte_unmap_unlock(pte, ptl);
Jan Karabc2466e2016-05-12 18:29:19 +02003105 if (!(ret & VM_FAULT_DAX_LOCKED)) {
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08003106 unlock_page(fault_page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003107 put_page(fault_page);
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08003108 } else {
Jan Karabc2466e2016-05-12 18:29:19 +02003109 dax_unlock_mapping_entry(vma->vm_file->f_mapping,
3110 pgoff);
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08003111 }
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003112 goto uncharge_out;
3113 }
Kirill A. Shutemov5c0a85f2016-05-20 16:58:41 -07003114 do_set_pte(vma, address, new_page, pte, true, true, false);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08003115 mem_cgroup_commit_charge(new_page, memcg, false, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07003116 lru_cache_add_active_or_unevictable(new_page, vma);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003117 pte_unmap_unlock(pte, ptl);
Jan Karabc2466e2016-05-12 18:29:19 +02003118 if (!(ret & VM_FAULT_DAX_LOCKED)) {
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08003119 unlock_page(fault_page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003120 put_page(fault_page);
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08003121 } else {
Jan Karabc2466e2016-05-12 18:29:19 +02003122 dax_unlock_mapping_entry(vma->vm_file->f_mapping, pgoff);
Matthew Wilcox2e4cdab2015-02-16 15:58:50 -08003123 }
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003124 return ret;
3125uncharge_out:
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08003126 mem_cgroup_cancel_charge(new_page, memcg, false);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003127 put_page(new_page);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003128 return ret;
3129}
3130
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003131static int do_shared_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins16abfa02007-10-04 16:56:06 +01003132 unsigned long address, pmd_t *pmd,
Nick Piggin54cb8822007-07-19 01:46:59 -07003133 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003134{
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003135 struct page *fault_page;
3136 struct address_space *mapping;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003137 spinlock_t *ptl;
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003138 pte_t *pte;
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003139 int dirtied = 0;
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003140 int ret, tmp;
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07003141
Jan Karabc2466e2016-05-12 18:29:19 +02003142 ret = __do_fault(vma, address, pgoff, flags, NULL, &fault_page, NULL);
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07003143 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003144 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003145
3146 /*
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003147 * Check if the backing address space wants to know that the page is
3148 * about to become writable
Linus Torvalds1da177e2005-04-16 15:20:36 -07003149 */
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07003150 if (vma->vm_ops->page_mkwrite) {
3151 unlock_page(fault_page);
3152 tmp = do_page_mkwrite(vma, fault_page, address);
3153 if (unlikely(!tmp ||
3154 (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003155 put_page(fault_page);
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07003156 return tmp;
3157 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003158 }
3159
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003160 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
3161 if (unlikely(!pte_same(*pte, orig_pte))) {
3162 pte_unmap_unlock(pte, ptl);
3163 unlock_page(fault_page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003164 put_page(fault_page);
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003165 return ret;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07003166 }
Kirill A. Shutemov5c0a85f2016-05-20 16:58:41 -07003167 do_set_pte(vma, address, fault_page, pte, true, false, false);
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003168 pte_unmap_unlock(pte, ptl);
3169
3170 if (set_page_dirty(fault_page))
3171 dirtied = 1;
Andrew Mortond82fa872014-12-18 16:17:29 -08003172 /*
3173 * Take a local copy of the address_space - page.mapping may be zeroed
3174 * by truncate after unlock_page(). The address_space itself remains
3175 * pinned by vma->vm_file's reference. We rely on unlock_page()'s
3176 * release semantics to prevent the compiler from undoing this copying.
3177 */
Kirill A. Shutemov1c290f62016-01-15 16:52:07 -08003178 mapping = page_rmapping(fault_page);
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003179 unlock_page(fault_page);
3180 if ((dirtied || vma->vm_ops->page_mkwrite) && mapping) {
3181 /*
3182 * Some device drivers do not set page.mapping but still
3183 * dirty their pages
3184 */
3185 balance_dirty_pages_ratelimited(mapping);
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07003186 }
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003187
Johannes Weiner74ec6752015-02-10 14:11:28 -08003188 if (!vma->vm_ops->page_mkwrite)
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003189 file_update_time(vma->vm_file);
3190
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07003191 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07003192}
Nick Piggind00806b2007-07-19 01:46:57 -07003193
Paul Cassella9a95f3c2014-08-06 16:07:24 -07003194/*
3195 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3196 * but allow concurrent faults).
3197 * The mmap_sem may have been released depending on flags and our
3198 * return value. See filemap_fault() and __lock_page_or_retry().
3199 */
Kirill A. Shutemov9b4bdd22015-02-10 14:09:51 -08003200static int do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Nick Piggin54cb8822007-07-19 01:46:59 -07003201 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003202 unsigned int flags, pte_t orig_pte)
Nick Piggin54cb8822007-07-19 01:46:59 -07003203{
Matthew Wilcox88193f72016-03-15 14:57:28 -07003204 pgoff_t pgoff = linear_page_index(vma, address);
Nick Piggin54cb8822007-07-19 01:46:59 -07003205
Hugh Dickins16abfa02007-10-04 16:56:06 +01003206 pte_unmap(page_table);
Kirill A. Shutemov6b7339f2015-07-06 23:18:37 +03003207 /* The VMA was not fully populated on mmap() or missing VM_DONTEXPAND */
3208 if (!vma->vm_ops->fault)
3209 return VM_FAULT_SIGBUS;
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07003210 if (!(flags & FAULT_FLAG_WRITE))
3211 return do_read_fault(mm, vma, address, pmd, pgoff, flags,
3212 orig_pte);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003213 if (!(vma->vm_flags & VM_SHARED))
3214 return do_cow_fault(mm, vma, address, pmd, pgoff, flags,
3215 orig_pte);
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003216 return do_shared_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Nick Piggin54cb8822007-07-19 01:46:59 -07003217}
3218
Rashika Kheriab19a9932014-04-03 14:48:02 -07003219static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
Rik van Riel04bb2f92013-10-07 11:29:36 +01003220 unsigned long addr, int page_nid,
3221 int *flags)
Mel Gorman9532fec2012-11-15 01:24:32 +00003222{
3223 get_page(page);
3224
3225 count_vm_numa_event(NUMA_HINT_FAULTS);
Rik van Riel04bb2f92013-10-07 11:29:36 +01003226 if (page_nid == numa_node_id()) {
Mel Gorman9532fec2012-11-15 01:24:32 +00003227 count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL);
Rik van Riel04bb2f92013-10-07 11:29:36 +01003228 *flags |= TNF_FAULT_LOCAL;
3229 }
Mel Gorman9532fec2012-11-15 01:24:32 +00003230
3231 return mpol_misplaced(page, vma, addr);
3232}
3233
Rashika Kheriab19a9932014-04-03 14:48:02 -07003234static int do_numa_page(struct mm_struct *mm, struct vm_area_struct *vma,
Mel Gormand10e63f2012-10-25 14:16:31 +02003235 unsigned long addr, pte_t pte, pte_t *ptep, pmd_t *pmd)
3236{
Mel Gorman4daae3b2012-11-02 11:33:45 +00003237 struct page *page = NULL;
Mel Gormand10e63f2012-10-25 14:16:31 +02003238 spinlock_t *ptl;
Mel Gorman8191acb2013-10-07 11:28:45 +01003239 int page_nid = -1;
Peter Zijlstra90572892013-10-07 11:29:20 +01003240 int last_cpupid;
Peter Zijlstracbee9f82012-10-25 14:16:43 +02003241 int target_nid;
Mel Gormanb8593bf2012-11-21 01:18:23 +00003242 bool migrated = false;
Mel Gormanb191f9b2015-03-25 15:55:40 -07003243 bool was_writable = pte_write(pte);
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003244 int flags = 0;
Mel Gormand10e63f2012-10-25 14:16:31 +02003245
Mel Gormanc0e7cad2015-02-12 14:58:41 -08003246 /* A PROT_NONE fault should not end up here */
3247 BUG_ON(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)));
3248
Mel Gormand10e63f2012-10-25 14:16:31 +02003249 /*
3250 * The "pte" at this point cannot be used safely without
3251 * validation through pte_unmap_same(). It's of NUMA type but
3252 * the pfn may be screwed if the read is non atomic.
3253 *
Mel Gorman4d942462015-02-12 14:58:28 -08003254 * We can safely just do a "set_pte_at()", because the old
3255 * page table entry is not accessible, so there would be no
3256 * concurrent hardware modifications to the PTE.
Mel Gormand10e63f2012-10-25 14:16:31 +02003257 */
3258 ptl = pte_lockptr(mm, pmd);
3259 spin_lock(ptl);
Mel Gorman4daae3b2012-11-02 11:33:45 +00003260 if (unlikely(!pte_same(*ptep, pte))) {
3261 pte_unmap_unlock(ptep, ptl);
3262 goto out;
3263 }
3264
Mel Gorman4d942462015-02-12 14:58:28 -08003265 /* Make it present again */
3266 pte = pte_modify(pte, vma->vm_page_prot);
3267 pte = pte_mkyoung(pte);
Mel Gormanb191f9b2015-03-25 15:55:40 -07003268 if (was_writable)
3269 pte = pte_mkwrite(pte);
Mel Gormand10e63f2012-10-25 14:16:31 +02003270 set_pte_at(mm, addr, ptep, pte);
3271 update_mmu_cache(vma, addr, ptep);
3272
3273 page = vm_normal_page(vma, addr, pte);
3274 if (!page) {
3275 pte_unmap_unlock(ptep, ptl);
3276 return 0;
3277 }
3278
Kirill A. Shutemove81c4802016-01-15 16:53:49 -08003279 /* TODO: handle PTE-mapped THP */
3280 if (PageCompound(page)) {
3281 pte_unmap_unlock(ptep, ptl);
3282 return 0;
3283 }
3284
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003285 /*
Mel Gormanbea66fb2015-03-25 15:55:37 -07003286 * Avoid grouping on RO pages in general. RO pages shouldn't hurt as
3287 * much anyway since they can be in shared cache state. This misses
3288 * the case where a mapping is writable but the process never writes
3289 * to it but pte_write gets cleared during protection updates and
3290 * pte_dirty has unpredictable behaviour between PTE scan updates,
3291 * background writeback, dirty balancing and application behaviour.
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003292 */
Mel Gormanbea66fb2015-03-25 15:55:37 -07003293 if (!(vma->vm_flags & VM_WRITE))
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003294 flags |= TNF_NO_GROUP;
3295
Rik van Rieldabe1d92013-10-07 11:29:34 +01003296 /*
3297 * Flag if the page is shared between multiple address spaces. This
3298 * is later used when determining whether to group tasks together
3299 */
3300 if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED))
3301 flags |= TNF_SHARED;
3302
Peter Zijlstra90572892013-10-07 11:29:20 +01003303 last_cpupid = page_cpupid_last(page);
Mel Gorman8191acb2013-10-07 11:28:45 +01003304 page_nid = page_to_nid(page);
Rik van Riel04bb2f92013-10-07 11:29:36 +01003305 target_nid = numa_migrate_prep(page, vma, addr, page_nid, &flags);
Mel Gormand10e63f2012-10-25 14:16:31 +02003306 pte_unmap_unlock(ptep, ptl);
Mel Gorman4daae3b2012-11-02 11:33:45 +00003307 if (target_nid == -1) {
Mel Gorman4daae3b2012-11-02 11:33:45 +00003308 put_page(page);
3309 goto out;
3310 }
3311
3312 /* Migrate to the requested node */
Mel Gorman1bc115d2013-10-07 11:29:05 +01003313 migrated = migrate_misplaced_page(page, vma, target_nid);
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003314 if (migrated) {
Mel Gorman8191acb2013-10-07 11:28:45 +01003315 page_nid = target_nid;
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003316 flags |= TNF_MIGRATED;
Mel Gorman074c2382015-03-25 15:55:42 -07003317 } else
3318 flags |= TNF_MIGRATE_FAIL;
Mel Gorman4daae3b2012-11-02 11:33:45 +00003319
3320out:
Mel Gorman8191acb2013-10-07 11:28:45 +01003321 if (page_nid != -1)
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003322 task_numa_fault(last_cpupid, page_nid, 1, flags);
Mel Gormand10e63f2012-10-25 14:16:31 +02003323 return 0;
3324}
3325
Matthew Wilcoxb96375f2015-09-08 14:58:48 -07003326static int create_huge_pmd(struct mm_struct *mm, struct vm_area_struct *vma,
3327 unsigned long address, pmd_t *pmd, unsigned int flags)
3328{
Kirill A. Shutemovfb6dd5f2015-09-09 15:39:35 -07003329 if (vma_is_anonymous(vma))
Matthew Wilcoxb96375f2015-09-08 14:58:48 -07003330 return do_huge_pmd_anonymous_page(mm, vma, address, pmd, flags);
3331 if (vma->vm_ops->pmd_fault)
3332 return vma->vm_ops->pmd_fault(vma, address, pmd, flags);
3333 return VM_FAULT_FALLBACK;
3334}
3335
3336static int wp_huge_pmd(struct mm_struct *mm, struct vm_area_struct *vma,
3337 unsigned long address, pmd_t *pmd, pmd_t orig_pmd,
3338 unsigned int flags)
3339{
Kirill A. Shutemovfb6dd5f2015-09-09 15:39:35 -07003340 if (vma_is_anonymous(vma))
Matthew Wilcoxb96375f2015-09-08 14:58:48 -07003341 return do_huge_pmd_wp_page(mm, vma, address, pmd, orig_pmd);
3342 if (vma->vm_ops->pmd_fault)
3343 return vma->vm_ops->pmd_fault(vma, address, pmd, flags);
3344 return VM_FAULT_FALLBACK;
3345}
3346
Linus Torvalds1da177e2005-04-16 15:20:36 -07003347/*
3348 * These routines also need to handle stuff like marking pages dirty
3349 * and/or accessed for architectures that don't do it in hardware (most
3350 * RISC architectures). The early dirtying is also good on the i386.
3351 *
3352 * There is also a hook called "update_mmu_cache()" that architectures
3353 * with external mmu caches can use to update those (ie the Sparc or
3354 * PowerPC hashed page tables that act as extended TLBs).
3355 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07003356 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3357 * but allow concurrent faults), and pte mapped but not yet locked.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07003358 * We return with pte unmapped and unlocked.
3359 *
3360 * The mmap_sem may have been released depending on flags and our
3361 * return value. See filemap_fault() and __lock_page_or_retry().
Linus Torvalds1da177e2005-04-16 15:20:36 -07003362 */
Kirill A. Shutemovc0292552013-09-12 15:14:05 -07003363static int handle_pte_fault(struct mm_struct *mm,
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003364 struct vm_area_struct *vma, unsigned long address,
3365 pte_t *pte, pmd_t *pmd, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003366{
3367 pte_t entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003368 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003369
Christian Borntraegere37c6982014-12-07 21:41:33 +01003370 /*
3371 * some architectures can have larger ptes than wordsize,
3372 * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and CONFIG_32BIT=y,
3373 * so READ_ONCE or ACCESS_ONCE cannot guarantee atomic accesses.
3374 * The code below just needs a consistent view for the ifs and
3375 * we later double check anyway with the ptl lock held. So here
3376 * a barrier will do.
3377 */
3378 entry = *pte;
3379 barrier();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003380 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07003381 if (pte_none(entry)) {
Oleg Nesterovb5330622015-09-08 14:58:28 -07003382 if (vma_is_anonymous(vma))
3383 return do_anonymous_page(mm, vma, address,
3384 pte, pmd, flags);
3385 else
Kirill A. Shutemov6b7339f2015-07-06 23:18:37 +03003386 return do_fault(mm, vma, address, pte, pmd,
3387 flags, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07003388 }
Hugh Dickins65500d22005-10-29 18:15:59 -07003389 return do_swap_page(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003390 pte, pmd, flags, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391 }
3392
Mel Gorman8a0516e2015-02-12 14:58:22 -08003393 if (pte_protnone(entry))
Mel Gormand10e63f2012-10-25 14:16:31 +02003394 return do_numa_page(mm, vma, address, entry, pte, pmd);
3395
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07003396 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003397 spin_lock(ptl);
3398 if (unlikely(!pte_same(*pte, entry)))
3399 goto unlock;
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003400 if (flags & FAULT_FLAG_WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003401 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003402 return do_wp_page(mm, vma, address,
3403 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003404 entry = pte_mkdirty(entry);
3405 }
3406 entry = pte_mkyoung(entry);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003407 if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) {
Russell King4b3073e2009-12-18 16:40:18 +00003408 update_mmu_cache(vma, address, pte);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07003409 } else {
3410 /*
3411 * This is needed only for protection faults but the arch code
3412 * is not yet telling us if this is a protection fault or not.
3413 * This still avoids useless tlb flushes for .text page faults
3414 * with threads.
3415 */
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003416 if (flags & FAULT_FLAG_WRITE)
Shaohua Li61c77322010-08-16 09:16:55 +08003417 flush_tlb_fix_spurious_fault(vma, address);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07003418 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003419unlock:
3420 pte_unmap_unlock(pte, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07003421 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003422}
3423
3424/*
3425 * By the time we get here, we already hold the mm semaphore
Paul Cassella9a95f3c2014-08-06 16:07:24 -07003426 *
3427 * The mmap_sem may have been released depending on flags and our
3428 * return value. See filemap_fault() and __lock_page_or_retry().
Linus Torvalds1da177e2005-04-16 15:20:36 -07003429 */
Johannes Weiner519e5242013-09-12 15:13:42 -07003430static int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3431 unsigned long address, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003432{
3433 pgd_t *pgd;
3434 pud_t *pud;
3435 pmd_t *pmd;
3436 pte_t *pte;
3437
Dave Hansen1b2ee122016-02-12 13:02:21 -08003438 if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE,
Dave Hansend61172b2016-02-12 13:02:24 -08003439 flags & FAULT_FLAG_INSTRUCTION,
Dave Hansen1b2ee122016-02-12 13:02:21 -08003440 flags & FAULT_FLAG_REMOTE))
Dave Hansen33a709b2016-02-12 13:02:19 -08003441 return VM_FAULT_SIGSEGV;
3442
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01003443 if (unlikely(is_vm_hugetlb_page(vma)))
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003444 return hugetlb_fault(mm, vma, address, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003445
Linus Torvalds1da177e2005-04-16 15:20:36 -07003446 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003447 pud = pud_alloc(mm, pgd, address);
3448 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07003449 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003450 pmd = pmd_alloc(mm, pud, address);
3451 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07003452 return VM_FAULT_OOM;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003453 if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) {
Matthew Wilcoxb96375f2015-09-08 14:58:48 -07003454 int ret = create_huge_pmd(mm, vma, address, pmd, flags);
Kirill A. Shutemovc0292552013-09-12 15:14:05 -07003455 if (!(ret & VM_FAULT_FALLBACK))
3456 return ret;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003457 } else {
3458 pmd_t orig_pmd = *pmd;
David Rientjes1f1d06c2012-05-29 15:06:23 -07003459 int ret;
3460
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003461 barrier();
Dan Williams5c7fb562016-01-15 16:56:52 -08003462 if (pmd_trans_huge(orig_pmd) || pmd_devmap(orig_pmd)) {
Will Deacona1dd4502012-12-11 16:01:27 -08003463 unsigned int dirty = flags & FAULT_FLAG_WRITE;
3464
Mel Gorman8a0516e2015-02-12 14:58:22 -08003465 if (pmd_protnone(orig_pmd))
Mel Gorman4daae3b2012-11-02 11:33:45 +00003466 return do_huge_pmd_numa_page(mm, vma, address,
Mel Gormand10e63f2012-10-25 14:16:31 +02003467 orig_pmd, pmd);
3468
Linus Torvalds3d59eeb2012-12-16 14:33:25 -08003469 if (dirty && !pmd_write(orig_pmd)) {
Matthew Wilcoxb96375f2015-09-08 14:58:48 -07003470 ret = wp_huge_pmd(mm, vma, address, pmd,
3471 orig_pmd, flags);
Kirill A. Shutemov9845cbb2014-02-25 15:01:42 -08003472 if (!(ret & VM_FAULT_FALLBACK))
3473 return ret;
Will Deacona1dd4502012-12-11 16:01:27 -08003474 } else {
3475 huge_pmd_set_accessed(mm, vma, address, pmd,
3476 orig_pmd, dirty);
Kirill A. Shutemov9845cbb2014-02-25 15:01:42 -08003477 return 0;
David Rientjes1f1d06c2012-05-29 15:06:23 -07003478 }
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003479 }
3480 }
3481
3482 /*
Kirill A. Shutemov3ed3a4f2016-03-17 14:19:11 -07003483 * Use pte_alloc() instead of pte_alloc_map, because we can't
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003484 * run pte_offset_map on the pmd, if an huge pmd could
3485 * materialize from under us from a different thread.
3486 */
Kirill A. Shutemov3ed3a4f2016-03-17 14:19:11 -07003487 if (unlikely(pte_alloc(mm, pmd, address)))
Hugh Dickinsc74df322005-10-29 18:16:23 -07003488 return VM_FAULT_OOM;
Andrea Arcangeliad33bb02016-02-26 15:19:28 -08003489 /*
3490 * If a huge pmd materialized under us just retry later. Use
3491 * pmd_trans_unstable() instead of pmd_trans_huge() to ensure the pmd
3492 * didn't become pmd_trans_huge under us and then back to pmd_none, as
3493 * a result of MADV_DONTNEED running immediately after a huge pmd fault
3494 * in a different thread of this mm, in turn leading to a misleading
3495 * pmd_trans_huge() retval. All we have to ensure is that it is a
3496 * regular pmd that we can walk with pte_offset_map() and we can do that
3497 * through an atomic read in C, which is what pmd_trans_unstable()
3498 * provides.
3499 */
3500 if (unlikely(pmd_trans_unstable(pmd) || pmd_devmap(*pmd)))
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003501 return 0;
3502 /*
3503 * A regular pmd is established and it can't morph into a huge pmd
3504 * from under us anymore at this point because we hold the mmap_sem
3505 * read mode and khugepaged takes it in write mode. So now it's
3506 * safe to run pte_offset_map().
3507 */
3508 pte = pte_offset_map(pmd, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003509
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003510 return handle_pte_fault(mm, vma, address, pte, pmd, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003511}
3512
Paul Cassella9a95f3c2014-08-06 16:07:24 -07003513/*
3514 * By the time we get here, we already hold the mm semaphore
3515 *
3516 * The mmap_sem may have been released depending on flags and our
3517 * return value. See filemap_fault() and __lock_page_or_retry().
3518 */
Johannes Weiner519e5242013-09-12 15:13:42 -07003519int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3520 unsigned long address, unsigned int flags)
3521{
3522 int ret;
3523
3524 __set_current_state(TASK_RUNNING);
3525
3526 count_vm_event(PGFAULT);
3527 mem_cgroup_count_vm_event(mm, PGFAULT);
3528
3529 /* do counter updates before entering really critical section. */
3530 check_sync_rss_stat(current);
3531
3532 /*
3533 * Enable the memcg OOM handling for faults triggered in user
3534 * space. Kernel faults are handled more gracefully.
3535 */
3536 if (flags & FAULT_FLAG_USER)
Johannes Weiner49426422013-10-16 13:46:59 -07003537 mem_cgroup_oom_enable();
Johannes Weiner519e5242013-09-12 15:13:42 -07003538
3539 ret = __handle_mm_fault(mm, vma, address, flags);
3540
Johannes Weiner49426422013-10-16 13:46:59 -07003541 if (flags & FAULT_FLAG_USER) {
3542 mem_cgroup_oom_disable();
3543 /*
3544 * The task may have entered a memcg OOM situation but
3545 * if the allocation error was handled gracefully (no
3546 * VM_FAULT_OOM), there is no need to kill anything.
3547 * Just clean up the OOM state peacefully.
3548 */
3549 if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM))
3550 mem_cgroup_oom_synchronize(false);
3551 }
Johannes Weiner3812c8c2013-09-12 15:13:44 -07003552
Johannes Weiner519e5242013-09-12 15:13:42 -07003553 return ret;
3554}
Jesse Barnese1d6d012014-12-12 16:55:27 -08003555EXPORT_SYMBOL_GPL(handle_mm_fault);
Johannes Weiner519e5242013-09-12 15:13:42 -07003556
Linus Torvalds1da177e2005-04-16 15:20:36 -07003557#ifndef __PAGETABLE_PUD_FOLDED
3558/*
3559 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07003560 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003561 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07003562int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003563{
Hugh Dickinsc74df322005-10-29 18:16:23 -07003564 pud_t *new = pud_alloc_one(mm, address);
3565 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07003566 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003567
Nick Piggin362a61a2008-05-14 06:37:36 +02003568 smp_wmb(); /* See comment in __pte_alloc */
3569
Hugh Dickins872fec12005-10-29 18:16:21 -07003570 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003571 if (pgd_present(*pgd)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003572 pud_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003573 else
3574 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07003575 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003576 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003577}
3578#endif /* __PAGETABLE_PUD_FOLDED */
3579
3580#ifndef __PAGETABLE_PMD_FOLDED
3581/*
3582 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07003583 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003584 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07003585int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003586{
Hugh Dickinsc74df322005-10-29 18:16:23 -07003587 pmd_t *new = pmd_alloc_one(mm, address);
3588 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07003589 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003590
Nick Piggin362a61a2008-05-14 06:37:36 +02003591 smp_wmb(); /* See comment in __pte_alloc */
3592
Hugh Dickins872fec12005-10-29 18:16:21 -07003593 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003594#ifndef __ARCH_HAS_4LEVEL_HACK
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -08003595 if (!pud_present(*pud)) {
3596 mm_inc_nr_pmds(mm);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003597 pud_populate(mm, pud, new);
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -08003598 } else /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003599 pmd_free(mm, new);
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -08003600#else
3601 if (!pgd_present(*pud)) {
3602 mm_inc_nr_pmds(mm);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003603 pgd_populate(mm, pud, new);
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -08003604 } else /* Another has populated it */
3605 pmd_free(mm, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003606#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07003607 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003608 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003609}
3610#endif /* __PAGETABLE_PMD_FOLDED */
3611
Namhyung Kim1b36ba82010-10-26 14:22:00 -07003612static int __follow_pte(struct mm_struct *mm, unsigned long address,
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07003613 pte_t **ptepp, spinlock_t **ptlp)
3614{
3615 pgd_t *pgd;
3616 pud_t *pud;
3617 pmd_t *pmd;
3618 pte_t *ptep;
3619
3620 pgd = pgd_offset(mm, address);
3621 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
3622 goto out;
3623
3624 pud = pud_offset(pgd, address);
3625 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
3626 goto out;
3627
3628 pmd = pmd_offset(pud, address);
Andrea Arcangelif66055ab2011-01-13 15:46:54 -08003629 VM_BUG_ON(pmd_trans_huge(*pmd));
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07003630 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
3631 goto out;
3632
3633 /* We cannot handle huge page PFN maps. Luckily they don't exist. */
3634 if (pmd_huge(*pmd))
3635 goto out;
3636
3637 ptep = pte_offset_map_lock(mm, pmd, address, ptlp);
3638 if (!ptep)
3639 goto out;
3640 if (!pte_present(*ptep))
3641 goto unlock;
3642 *ptepp = ptep;
3643 return 0;
3644unlock:
3645 pte_unmap_unlock(ptep, *ptlp);
3646out:
3647 return -EINVAL;
3648}
3649
Namhyung Kim1b36ba82010-10-26 14:22:00 -07003650static inline int follow_pte(struct mm_struct *mm, unsigned long address,
3651 pte_t **ptepp, spinlock_t **ptlp)
3652{
3653 int res;
3654
3655 /* (void) is needed to make gcc happy */
3656 (void) __cond_lock(*ptlp,
3657 !(res = __follow_pte(mm, address, ptepp, ptlp)));
3658 return res;
3659}
3660
Johannes Weiner3b6748e2009-06-16 15:32:35 -07003661/**
3662 * follow_pfn - look up PFN at a user virtual address
3663 * @vma: memory mapping
3664 * @address: user virtual address
3665 * @pfn: location to store found PFN
3666 *
3667 * Only IO mappings and raw PFN mappings are allowed.
3668 *
3669 * Returns zero and the pfn at @pfn on success, -ve otherwise.
3670 */
3671int follow_pfn(struct vm_area_struct *vma, unsigned long address,
3672 unsigned long *pfn)
3673{
3674 int ret = -EINVAL;
3675 spinlock_t *ptl;
3676 pte_t *ptep;
3677
3678 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3679 return ret;
3680
3681 ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);
3682 if (ret)
3683 return ret;
3684 *pfn = pte_pfn(*ptep);
3685 pte_unmap_unlock(ptep, ptl);
3686 return 0;
3687}
3688EXPORT_SYMBOL(follow_pfn);
3689
Rik van Riel28b2ee22008-07-23 21:27:05 -07003690#ifdef CONFIG_HAVE_IOREMAP_PROT
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003691int follow_phys(struct vm_area_struct *vma,
3692 unsigned long address, unsigned int flags,
3693 unsigned long *prot, resource_size_t *phys)
Rik van Riel28b2ee22008-07-23 21:27:05 -07003694{
Johannes Weiner03668a42009-06-16 15:32:34 -07003695 int ret = -EINVAL;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003696 pte_t *ptep, pte;
3697 spinlock_t *ptl;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003698
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003699 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3700 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003701
Johannes Weiner03668a42009-06-16 15:32:34 -07003702 if (follow_pte(vma->vm_mm, address, &ptep, &ptl))
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003703 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003704 pte = *ptep;
Johannes Weiner03668a42009-06-16 15:32:34 -07003705
Rik van Riel28b2ee22008-07-23 21:27:05 -07003706 if ((flags & FOLL_WRITE) && !pte_write(pte))
3707 goto unlock;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003708
3709 *prot = pgprot_val(pte_pgprot(pte));
Johannes Weiner03668a42009-06-16 15:32:34 -07003710 *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003711
Johannes Weiner03668a42009-06-16 15:32:34 -07003712 ret = 0;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003713unlock:
3714 pte_unmap_unlock(ptep, ptl);
3715out:
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003716 return ret;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003717}
3718
3719int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
3720 void *buf, int len, int write)
3721{
3722 resource_size_t phys_addr;
3723 unsigned long prot = 0;
KOSAKI Motohiro2bc72732009-01-06 14:39:43 -08003724 void __iomem *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003725 int offset = addr & (PAGE_SIZE-1);
3726
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003727 if (follow_phys(vma, addr, write, &prot, &phys_addr))
Rik van Riel28b2ee22008-07-23 21:27:05 -07003728 return -EINVAL;
3729
Grazvydas Ignotas9cb12d72015-02-12 15:00:19 -08003730 maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot);
Rik van Riel28b2ee22008-07-23 21:27:05 -07003731 if (write)
3732 memcpy_toio(maddr + offset, buf, len);
3733 else
3734 memcpy_fromio(buf, maddr + offset, len);
3735 iounmap(maddr);
3736
3737 return len;
3738}
Uwe Kleine-König5a736332013-08-07 13:02:52 +02003739EXPORT_SYMBOL_GPL(generic_access_phys);
Rik van Riel28b2ee22008-07-23 21:27:05 -07003740#endif
3741
David Howells0ec76a12006-09-27 01:50:15 -07003742/*
Stephen Wilson206cb632011-03-13 15:49:19 -04003743 * Access another process' address space as given in mm. If non-NULL, use the
3744 * given task for page fault accounting.
David Howells0ec76a12006-09-27 01:50:15 -07003745 */
Stephen Wilson206cb632011-03-13 15:49:19 -04003746static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
3747 unsigned long addr, void *buf, int len, int write)
David Howells0ec76a12006-09-27 01:50:15 -07003748{
David Howells0ec76a12006-09-27 01:50:15 -07003749 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07003750 void *old_buf = buf;
3751
David Howells0ec76a12006-09-27 01:50:15 -07003752 down_read(&mm->mmap_sem);
Simon Arlott183ff222007-10-20 01:27:18 +02003753 /* ignore errors, just check how much was successfully transferred */
David Howells0ec76a12006-09-27 01:50:15 -07003754 while (len) {
3755 int bytes, ret, offset;
3756 void *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003757 struct page *page = NULL;
David Howells0ec76a12006-09-27 01:50:15 -07003758
Dave Hansen1e987792016-02-12 13:01:54 -08003759 ret = get_user_pages_remote(tsk, mm, addr, 1,
David Howells0ec76a12006-09-27 01:50:15 -07003760 write, 1, &page, &vma);
Rik van Riel28b2ee22008-07-23 21:27:05 -07003761 if (ret <= 0) {
Rik van Rieldbffcd02014-08-06 16:08:12 -07003762#ifndef CONFIG_HAVE_IOREMAP_PROT
3763 break;
3764#else
Rik van Riel28b2ee22008-07-23 21:27:05 -07003765 /*
3766 * Check if this is a VM_IO | VM_PFNMAP VMA, which
3767 * we can access using slightly different code.
3768 */
Rik van Riel28b2ee22008-07-23 21:27:05 -07003769 vma = find_vma(mm, addr);
Michael Ellermanfe936df2011-04-14 15:22:10 -07003770 if (!vma || vma->vm_start > addr)
Rik van Riel28b2ee22008-07-23 21:27:05 -07003771 break;
3772 if (vma->vm_ops && vma->vm_ops->access)
3773 ret = vma->vm_ops->access(vma, addr, buf,
3774 len, write);
3775 if (ret <= 0)
Rik van Riel28b2ee22008-07-23 21:27:05 -07003776 break;
3777 bytes = ret;
Rik van Rieldbffcd02014-08-06 16:08:12 -07003778#endif
David Howells0ec76a12006-09-27 01:50:15 -07003779 } else {
Rik van Riel28b2ee22008-07-23 21:27:05 -07003780 bytes = len;
3781 offset = addr & (PAGE_SIZE-1);
3782 if (bytes > PAGE_SIZE-offset)
3783 bytes = PAGE_SIZE-offset;
3784
3785 maddr = kmap(page);
3786 if (write) {
3787 copy_to_user_page(vma, page, addr,
3788 maddr + offset, buf, bytes);
3789 set_page_dirty_lock(page);
3790 } else {
3791 copy_from_user_page(vma, page, addr,
3792 buf, maddr + offset, bytes);
3793 }
3794 kunmap(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003795 put_page(page);
David Howells0ec76a12006-09-27 01:50:15 -07003796 }
David Howells0ec76a12006-09-27 01:50:15 -07003797 len -= bytes;
3798 buf += bytes;
3799 addr += bytes;
3800 }
3801 up_read(&mm->mmap_sem);
David Howells0ec76a12006-09-27 01:50:15 -07003802
3803 return buf - old_buf;
3804}
Andi Kleen03252912008-01-30 13:33:18 +01003805
Stephen Wilson5ddd36b2011-03-13 15:49:20 -04003806/**
Randy Dunlapae91dbf2011-03-26 13:27:01 -07003807 * access_remote_vm - access another process' address space
Stephen Wilson5ddd36b2011-03-13 15:49:20 -04003808 * @mm: the mm_struct of the target address space
3809 * @addr: start address to access
3810 * @buf: source or destination buffer
3811 * @len: number of bytes to transfer
3812 * @write: whether the access is a write
3813 *
3814 * The caller must hold a reference on @mm.
3815 */
3816int access_remote_vm(struct mm_struct *mm, unsigned long addr,
3817 void *buf, int len, int write)
3818{
3819 return __access_remote_vm(NULL, mm, addr, buf, len, write);
3820}
3821
Andi Kleen03252912008-01-30 13:33:18 +01003822/*
Stephen Wilson206cb632011-03-13 15:49:19 -04003823 * Access another process' address space.
3824 * Source/target buffer must be kernel space,
3825 * Do not walk the page table directly, use get_user_pages
3826 */
3827int access_process_vm(struct task_struct *tsk, unsigned long addr,
3828 void *buf, int len, int write)
3829{
3830 struct mm_struct *mm;
3831 int ret;
3832
3833 mm = get_task_mm(tsk);
3834 if (!mm)
3835 return 0;
3836
3837 ret = __access_remote_vm(tsk, mm, addr, buf, len, write);
3838 mmput(mm);
3839
3840 return ret;
3841}
3842
Andi Kleen03252912008-01-30 13:33:18 +01003843/*
3844 * Print the name of a VMA.
3845 */
3846void print_vma_addr(char *prefix, unsigned long ip)
3847{
3848 struct mm_struct *mm = current->mm;
3849 struct vm_area_struct *vma;
3850
Ingo Molnare8bff742008-02-13 20:21:06 +01003851 /*
3852 * Do not print if we are in atomic
3853 * contexts (in exception stacks, etc.):
3854 */
3855 if (preempt_count())
3856 return;
3857
Andi Kleen03252912008-01-30 13:33:18 +01003858 down_read(&mm->mmap_sem);
3859 vma = find_vma(mm, ip);
3860 if (vma && vma->vm_file) {
3861 struct file *f = vma->vm_file;
3862 char *buf = (char *)__get_free_page(GFP_KERNEL);
3863 if (buf) {
Andy Shevchenko2fbc57c2012-12-17 16:01:23 -08003864 char *p;
Andi Kleen03252912008-01-30 13:33:18 +01003865
Miklos Szeredi9bf39ab2015-06-19 10:29:13 +02003866 p = file_path(f, buf, PAGE_SIZE);
Andi Kleen03252912008-01-30 13:33:18 +01003867 if (IS_ERR(p))
3868 p = "?";
Andy Shevchenko2fbc57c2012-12-17 16:01:23 -08003869 printk("%s%s[%lx+%lx]", prefix, kbasename(p),
Andi Kleen03252912008-01-30 13:33:18 +01003870 vma->vm_start,
3871 vma->vm_end - vma->vm_start);
3872 free_page((unsigned long)buf);
3873 }
3874 }
Jeff Liu51a07e52012-07-31 16:43:18 -07003875 up_read(&mm->mmap_sem);
Andi Kleen03252912008-01-30 13:33:18 +01003876}
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003877
Michael S. Tsirkin662bbcb2013-05-26 17:32:23 +03003878#if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP)
David Hildenbrand9ec23532015-05-11 17:52:07 +02003879void __might_fault(const char *file, int line)
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003880{
Peter Zijlstra95156f02009-01-12 13:02:11 +01003881 /*
3882 * Some code (nfs/sunrpc) uses socket ops on kernel memory while
3883 * holding the mmap_sem, this is safe because kernel memory doesn't
3884 * get paged out, therefore we'll never actually fault, and the
3885 * below annotations will generate false positives.
3886 */
3887 if (segment_eq(get_fs(), KERNEL_DS))
3888 return;
David Hildenbrand9ec23532015-05-11 17:52:07 +02003889 if (pagefault_disabled())
Michael S. Tsirkin662bbcb2013-05-26 17:32:23 +03003890 return;
David Hildenbrand9ec23532015-05-11 17:52:07 +02003891 __might_sleep(file, line, 0);
3892#if defined(CONFIG_DEBUG_ATOMIC_SLEEP)
Michael S. Tsirkin662bbcb2013-05-26 17:32:23 +03003893 if (current->mm)
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003894 might_lock_read(&current->mm->mmap_sem);
David Hildenbrand9ec23532015-05-11 17:52:07 +02003895#endif
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003896}
David Hildenbrand9ec23532015-05-11 17:52:07 +02003897EXPORT_SYMBOL(__might_fault);
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003898#endif
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08003899
3900#if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)
3901static void clear_gigantic_page(struct page *page,
3902 unsigned long addr,
3903 unsigned int pages_per_huge_page)
3904{
3905 int i;
3906 struct page *p = page;
3907
3908 might_sleep();
3909 for (i = 0; i < pages_per_huge_page;
3910 i++, p = mem_map_next(p, page, i)) {
3911 cond_resched();
3912 clear_user_highpage(p, addr + i * PAGE_SIZE);
3913 }
3914}
3915void clear_huge_page(struct page *page,
3916 unsigned long addr, unsigned int pages_per_huge_page)
3917{
3918 int i;
3919
3920 if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
3921 clear_gigantic_page(page, addr, pages_per_huge_page);
3922 return;
3923 }
3924
3925 might_sleep();
3926 for (i = 0; i < pages_per_huge_page; i++) {
3927 cond_resched();
3928 clear_user_highpage(page + i, addr + i * PAGE_SIZE);
3929 }
3930}
3931
3932static void copy_user_gigantic_page(struct page *dst, struct page *src,
3933 unsigned long addr,
3934 struct vm_area_struct *vma,
3935 unsigned int pages_per_huge_page)
3936{
3937 int i;
3938 struct page *dst_base = dst;
3939 struct page *src_base = src;
3940
3941 for (i = 0; i < pages_per_huge_page; ) {
3942 cond_resched();
3943 copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma);
3944
3945 i++;
3946 dst = mem_map_next(dst, dst_base, i);
3947 src = mem_map_next(src, src_base, i);
3948 }
3949}
3950
3951void copy_user_huge_page(struct page *dst, struct page *src,
3952 unsigned long addr, struct vm_area_struct *vma,
3953 unsigned int pages_per_huge_page)
3954{
3955 int i;
3956
3957 if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
3958 copy_user_gigantic_page(dst, src, addr, vma,
3959 pages_per_huge_page);
3960 return;
3961 }
3962
3963 might_sleep();
3964 for (i = 0; i < pages_per_huge_page; i++) {
3965 cond_resched();
3966 copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma);
3967 }
3968}
3969#endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003970
Olof Johansson40b64ac2013-12-20 14:28:05 -08003971#if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS
Kirill A. Shutemovb35f1812014-01-21 15:49:07 -08003972
3973static struct kmem_cache *page_ptl_cachep;
3974
3975void __init ptlock_cache_init(void)
3976{
3977 page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0,
3978 SLAB_PANIC, NULL);
3979}
3980
Peter Zijlstra539edb52013-11-14 14:31:52 -08003981bool ptlock_alloc(struct page *page)
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003982{
3983 spinlock_t *ptl;
3984
Kirill A. Shutemovb35f1812014-01-21 15:49:07 -08003985 ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL);
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003986 if (!ptl)
3987 return false;
Peter Zijlstra539edb52013-11-14 14:31:52 -08003988 page->ptl = ptl;
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003989 return true;
3990}
3991
Peter Zijlstra539edb52013-11-14 14:31:52 -08003992void ptlock_free(struct page *page)
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003993{
Kirill A. Shutemovb35f1812014-01-21 15:49:07 -08003994 kmem_cache_free(page_ptl_cachep, page->ptl);
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08003995}
3996#endif