blob: 9b135a35f99c27bffddbcdd131097ded81e67ba2 [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.
Tobin C Harding166f61b2017-02-24 14:59:01 -080033 * Idea by Alex Bligh (alex@cconcepts.co.uk)
Linus Torvalds1da177e2005-04-16 15:20:36 -070034 *
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>
Ingo Molnar6e84f312017-02-08 18:51:29 +010043#include <linux/sched/mm.h>
Ingo Molnarf7ccbae2017-02-08 18:51:30 +010044#include <linux/sched/coredump.h>
Ingo Molnar6a3827d2017-02-08 18:51:31 +010045#include <linux/sched/numa_balancing.h>
Ingo Molnar29930022017-02-08 18:51:36 +010046#include <linux/sched/task.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include <linux/hugetlb.h>
48#include <linux/mman.h>
49#include <linux/swap.h>
50#include <linux/highmem.h>
51#include <linux/pagemap.h>
Jérôme Glisse5042db42017-09-08 16:11:43 -070052#include <linux/memremap.h>
Hugh Dickins9a840892009-09-21 17:02:01 -070053#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070054#include <linux/rmap.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040055#include <linux/export.h>
Shailabh Nagar0ff92242006-07-14 00:24:37 -070056#include <linux/delayacct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070057#include <linux/init.h>
Dan Williams01c8f1c2016-01-15 16:56:40 -080058#include <linux/pfn_t.h>
Peter Zijlstraedc79b22006-09-25 23:30:58 -070059#include <linux/writeback.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080060#include <linux/memcontrol.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070061#include <linux/mmu_notifier.h>
Hugh Dickins3dc14742009-01-06 14:40:08 -080062#include <linux/swapops.h>
63#include <linux/elf.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090064#include <linux/gfp.h>
Mel Gorman4daae3b2012-11-02 11:33:45 +000065#include <linux/migrate.h>
Andy Shevchenko2fbc57c2012-12-17 16:01:23 -080066#include <linux/string.h>
Dan Williams0abdd7a2014-01-21 15:48:12 -080067#include <linux/dma-debug.h>
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -070068#include <linux/debugfs.h>
Andrea Arcangeli6b251fc2015-09-04 15:46:20 -070069#include <linux/userfaultfd_k.h>
Jan Karabc2466e2016-05-12 18:29:19 +020070#include <linux/dax.h>
Michal Hocko6b31d592017-08-18 15:16:15 -070071#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
Alexey Dobriyan6952b612009-09-18 23:55:55 +040073#include <asm/io.h>
Dave Hansen33a709b2016-02-12 13:02:19 -080074#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070075#include <asm/pgalloc.h>
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080076#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070077#include <asm/tlb.h>
78#include <asm/tlbflush.h>
79#include <asm/pgtable.h>
80
Jan Beulich42b77722008-07-23 21:27:10 -070081#include "internal.h"
82
Arnd Bergmannaf27d942018-02-16 16:25:53 +010083#if defined(LAST_CPUPID_NOT_IN_PAGE_FLAGS) && !defined(CONFIG_COMPILE_TEST)
Peter Zijlstra90572892013-10-07 11:29:20 +010084#warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid.
Peter Zijlstra75980e92013-02-22 16:34:32 -080085#endif
86
Andy Whitcroftd41dee32005-06-23 00:07:54 -070087#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -070088/* use the per-pgdat data instead for discontigmem - mbligh */
89unsigned long max_mapnr;
Linus Torvalds1da177e2005-04-16 15:20:36 -070090EXPORT_SYMBOL(max_mapnr);
Tobin C Harding166f61b2017-02-24 14:59:01 -080091
92struct page *mem_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -070093EXPORT_SYMBOL(mem_map);
94#endif
95
Linus Torvalds1da177e2005-04-16 15:20:36 -070096/*
97 * A number of key systems in x86 including ioremap() rely on the assumption
98 * that high_memory defines the upper bound on direct map memory, then end
99 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
100 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
101 * and ZONE_HIGHMEM.
102 */
Tobin C Harding166f61b2017-02-24 14:59:01 -0800103void *high_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105
Ingo Molnar32a93232008-02-06 22:39:44 +0100106/*
107 * Randomize the address space (stacks, mmaps, brk, etc.).
108 *
109 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
110 * as ancient (libc5 based) binaries can segfault. )
111 */
112int randomize_va_space __read_mostly =
113#ifdef CONFIG_COMPAT_BRK
114 1;
115#else
116 2;
117#endif
Andi Kleena62eaf12006-02-16 23:41:58 +0100118
119static int __init disable_randmaps(char *s)
120{
121 randomize_va_space = 0;
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800122 return 1;
Andi Kleena62eaf12006-02-16 23:41:58 +0100123}
124__setup("norandmaps", disable_randmaps);
125
Hugh Dickins62eede62009-09-21 17:03:34 -0700126unsigned long zero_pfn __read_mostly;
Ard Biesheuvel0b700682014-09-12 22:17:23 +0200127EXPORT_SYMBOL(zero_pfn);
128
Tobin C Harding166f61b2017-02-24 14:59:01 -0800129unsigned long highest_memmap_pfn __read_mostly;
130
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -0700131/*
132 * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init()
133 */
134static int __init init_zero_pfn(void)
135{
136 zero_pfn = page_to_pfn(ZERO_PAGE(0));
137 return 0;
138}
139core_initcall(init_zero_pfn);
Andi Kleena62eaf12006-02-16 23:41:58 +0100140
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800141
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800142#if defined(SPLIT_RSS_COUNTING)
143
David Rientjesea48cf72012-03-21 16:34:13 -0700144void sync_mm_rss(struct mm_struct *mm)
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800145{
146 int i;
147
148 for (i = 0; i < NR_MM_COUNTERS; i++) {
David Rientjes05af2e12012-03-21 16:34:13 -0700149 if (current->rss_stat.count[i]) {
150 add_mm_counter(mm, i, current->rss_stat.count[i]);
151 current->rss_stat.count[i] = 0;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800152 }
153 }
David Rientjes05af2e12012-03-21 16:34:13 -0700154 current->rss_stat.events = 0;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800155}
156
157static void add_mm_counter_fast(struct mm_struct *mm, int member, int val)
158{
159 struct task_struct *task = current;
160
161 if (likely(task->mm == mm))
162 task->rss_stat.count[member] += val;
163 else
164 add_mm_counter(mm, member, val);
165}
166#define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1)
167#define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1)
168
169/* sync counter once per 64 page faults */
170#define TASK_RSS_EVENTS_THRESH (64)
171static void check_sync_rss_stat(struct task_struct *task)
172{
173 if (unlikely(task != current))
174 return;
175 if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH))
David Rientjesea48cf72012-03-21 16:34:13 -0700176 sync_mm_rss(task->mm);
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800177}
Peter Zijlstra9547d012011-05-24 17:12:14 -0700178#else /* SPLIT_RSS_COUNTING */
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800179
180#define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member)
181#define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member)
182
183static void check_sync_rss_stat(struct task_struct *task)
184{
185}
186
Peter Zijlstra9547d012011-05-24 17:12:14 -0700187#endif /* SPLIT_RSS_COUNTING */
188
189#ifdef HAVE_GENERIC_MMU_GATHER
190
Nicholas Krauseca1d6c72015-09-04 15:48:22 -0700191static bool tlb_next_batch(struct mmu_gather *tlb)
Peter Zijlstra9547d012011-05-24 17:12:14 -0700192{
193 struct mmu_gather_batch *batch;
194
195 batch = tlb->active;
196 if (batch->next) {
197 tlb->active = batch->next;
Nicholas Krauseca1d6c72015-09-04 15:48:22 -0700198 return true;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700199 }
200
Michal Hocko53a59fc2013-01-04 15:35:12 -0800201 if (tlb->batch_count == MAX_GATHER_BATCH_COUNT)
Nicholas Krauseca1d6c72015-09-04 15:48:22 -0700202 return false;
Michal Hocko53a59fc2013-01-04 15:35:12 -0800203
Peter Zijlstra9547d012011-05-24 17:12:14 -0700204 batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0);
205 if (!batch)
Nicholas Krauseca1d6c72015-09-04 15:48:22 -0700206 return false;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700207
Michal Hocko53a59fc2013-01-04 15:35:12 -0800208 tlb->batch_count++;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700209 batch->next = NULL;
210 batch->nr = 0;
211 batch->max = MAX_GATHER_BATCH;
212
213 tlb->active->next = batch;
214 tlb->active = batch;
215
Nicholas Krauseca1d6c72015-09-04 15:48:22 -0700216 return true;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700217}
218
Minchan Kim56236a52017-08-10 15:24:05 -0700219void arch_tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm,
220 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
Aneesh Kumar K.Ve77b0852016-07-26 15:24:12 -0700236 tlb->page_size = 0;
Will Deaconfb7332a2014-10-29 10:03:09 +0000237
238 __tlb_reset_range(tlb);
Peter Zijlstra9547d012011-05-24 17:12:14 -0700239}
240
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700241static void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb)
Peter Zijlstra9547d012011-05-24 17:12:14 -0700242{
Will Deacon721c21c2015-01-12 19:10:55 +0000243 if (!tlb->end)
244 return;
245
Peter Zijlstra9547d012011-05-24 17:12:14 -0700246 tlb_flush(tlb);
Joerg Roedel34ee6452014-11-13 13:46:09 +1100247 mmu_notifier_invalidate_range(tlb->mm, tlb->start, tlb->end);
Will Deaconfb7332a2014-10-29 10:03:09 +0000248 __tlb_reset_range(tlb);
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700249}
250
251static void tlb_flush_mmu_free(struct mmu_gather *tlb)
252{
253 struct mmu_gather_batch *batch;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700254
Nicholas Piggindb7ddef2018-08-23 18:47:08 +1000255#ifdef CONFIG_HAVE_RCU_TABLE_FREE
256 tlb_table_flush(tlb);
257#endif
Will Deacon721c21c2015-01-12 19:10:55 +0000258 for (batch = &tlb->local; batch && batch->nr; batch = batch->next) {
Peter Zijlstra9547d012011-05-24 17:12:14 -0700259 free_pages_and_swap_cache(batch->pages, batch->nr);
260 batch->nr = 0;
261 }
262 tlb->active = &tlb->local;
263}
264
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700265void tlb_flush_mmu(struct mmu_gather *tlb)
266{
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700267 tlb_flush_mmu_tlbonly(tlb);
268 tlb_flush_mmu_free(tlb);
269}
270
Peter Zijlstra9547d012011-05-24 17:12:14 -0700271/* tlb_finish_mmu
272 * Called at the end of the shootdown operation to free up any resources
273 * that were required.
274 */
Minchan Kim56236a52017-08-10 15:24:05 -0700275void arch_tlb_finish_mmu(struct mmu_gather *tlb,
Minchan Kim99baac22017-08-10 15:24:12 -0700276 unsigned long start, unsigned long end, bool force)
Peter Zijlstra9547d012011-05-24 17:12:14 -0700277{
278 struct mmu_gather_batch *batch, *next;
279
Minchan Kim99baac22017-08-10 15:24:12 -0700280 if (force)
281 __tlb_adjust_range(tlb, start, end - start);
282
Peter Zijlstra9547d012011-05-24 17:12:14 -0700283 tlb_flush_mmu(tlb);
284
285 /* keep the page table cache within bounds */
286 check_pgt_cache();
287
288 for (batch = tlb->local.next; batch; batch = next) {
289 next = batch->next;
290 free_pages((unsigned long)batch, 0);
291 }
292 tlb->local.next = NULL;
293}
294
295/* __tlb_remove_page
296 * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while
297 * handling the additional races in SMP caused by other CPUs caching valid
298 * mappings in their TLBs. Returns the number of free page slots left.
299 * When out of page slots we must call tlb_flush_mmu().
Aneesh Kumar K.Ve9d55e12016-07-26 15:24:09 -0700300 *returns true if the caller should flush.
Peter Zijlstra9547d012011-05-24 17:12:14 -0700301 */
Aneesh Kumar K.Ve77b0852016-07-26 15:24:12 -0700302bool __tlb_remove_page_size(struct mmu_gather *tlb, struct page *page, int page_size)
Peter Zijlstra9547d012011-05-24 17:12:14 -0700303{
304 struct mmu_gather_batch *batch;
305
Will Deaconfb7332a2014-10-29 10:03:09 +0000306 VM_BUG_ON(!tlb->end);
Aneesh Kumar K.V692a68c2016-12-12 16:42:43 -0800307 VM_WARN_ON(tlb->page_size != page_size);
Aneesh Kumar K.Ve77b0852016-07-26 15:24:12 -0700308
Peter Zijlstra9547d012011-05-24 17:12:14 -0700309 batch = tlb->active;
Aneesh Kumar K.V692a68c2016-12-12 16:42:43 -0800310 /*
311 * Add the page and check if we are full. If so
312 * force a flush.
313 */
314 batch->pages[batch->nr++] = page;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700315 if (batch->nr == batch->max) {
316 if (!tlb_next_batch(tlb))
Aneesh Kumar K.Ve9d55e12016-07-26 15:24:09 -0700317 return true;
Shaohua Li0b43c3a2011-07-08 15:39:41 -0700318 batch = tlb->active;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700319 }
Sasha Levin309381fea2014-01-23 15:52:54 -0800320 VM_BUG_ON_PAGE(batch->nr > batch->max, page);
Peter Zijlstra9547d012011-05-24 17:12:14 -0700321
Aneesh Kumar K.Ve9d55e12016-07-26 15:24:09 -0700322 return false;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700323}
324
325#endif /* HAVE_GENERIC_MMU_GATHER */
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800326
Peter Zijlstra26723912011-05-24 17:12:00 -0700327#ifdef CONFIG_HAVE_RCU_TABLE_FREE
328
Peter Zijlstra52a288c2018-08-22 17:30:13 +0200329/*
330 * See the comment near struct mmu_table_batch.
331 */
332
Peter Zijlstra26723912011-05-24 17:12:00 -0700333static void tlb_remove_table_smp_sync(void *arg)
334{
Peter Zijlstra52a288c2018-08-22 17:30:13 +0200335 /* Simply deliver the interrupt */
Peter Zijlstra26723912011-05-24 17:12:00 -0700336}
337
Peter Zijlstra52a288c2018-08-22 17:30:13 +0200338static void tlb_remove_table_one(void *table)
Peter Zijlstra26723912011-05-24 17:12:00 -0700339{
340 /*
341 * This isn't an RCU grace period and hence the page-tables cannot be
342 * assumed to be actually RCU-freed.
343 *
344 * It is however sufficient for software page-table walkers that rely on
345 * IRQ disabling. See the comment near struct mmu_table_batch.
346 */
Peter Zijlstra52a288c2018-08-22 17:30:13 +0200347 smp_call_function(tlb_remove_table_smp_sync, NULL, 1);
Peter Zijlstra26723912011-05-24 17:12:00 -0700348 __tlb_remove_table(table);
349}
350
351static void tlb_remove_table_rcu(struct rcu_head *head)
352{
353 struct mmu_table_batch *batch;
354 int i;
355
356 batch = container_of(head, struct mmu_table_batch, rcu);
357
358 for (i = 0; i < batch->nr; i++)
359 __tlb_remove_table(batch->tables[i]);
360
361 free_page((unsigned long)batch);
362}
363
364void tlb_table_flush(struct mmu_gather *tlb)
365{
366 struct mmu_table_batch **batch = &tlb->batch;
367
368 if (*batch) {
369 call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu);
370 *batch = NULL;
371 }
372}
373
374void tlb_remove_table(struct mmu_gather *tlb, void *table)
375{
376 struct mmu_table_batch **batch = &tlb->batch;
377
Peter Zijlstra26723912011-05-24 17:12:00 -0700378 /*
379 * When there's less then two users of this mm there cannot be a
380 * concurrent page-table walk.
381 */
382 if (atomic_read(&tlb->mm->mm_users) < 2) {
383 __tlb_remove_table(table);
384 return;
385 }
386
387 if (*batch == NULL) {
388 *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN);
389 if (*batch == NULL) {
Peter Zijlstra52a288c2018-08-22 17:30:13 +0200390 tlb_remove_table_one(table);
Peter Zijlstra26723912011-05-24 17:12:00 -0700391 return;
392 }
393 (*batch)->nr = 0;
394 }
395 (*batch)->tables[(*batch)->nr++] = table;
396 if ((*batch)->nr == MAX_TABLE_BATCH)
397 tlb_table_flush(tlb);
398}
399
Peter Zijlstra9547d012011-05-24 17:12:14 -0700400#endif /* CONFIG_HAVE_RCU_TABLE_FREE */
Peter Zijlstra26723912011-05-24 17:12:00 -0700401
Mike Rapoportef549e12018-01-31 16:17:17 -0800402/**
403 * tlb_gather_mmu - initialize an mmu_gather structure for page-table tear-down
404 * @tlb: the mmu_gather structure to initialize
405 * @mm: the mm_struct of the target address space
406 * @start: start of the region that will be removed from the page-table
407 * @end: end of the region that will be removed from the page-table
408 *
409 * Called to initialize an (on-stack) mmu_gather structure for page-table
410 * tear-down from @mm. The @start and @end are set to 0 and -1
411 * respectively when @mm is without users and we're going to destroy
412 * the full address space (exit/execve).
Minchan Kim56236a52017-08-10 15:24:05 -0700413 */
414void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm,
415 unsigned long start, unsigned long end)
416{
417 arch_tlb_gather_mmu(tlb, mm, start, end);
Minchan Kim99baac22017-08-10 15:24:12 -0700418 inc_tlb_flush_pending(tlb->mm);
Minchan Kim56236a52017-08-10 15:24:05 -0700419}
420
421void tlb_finish_mmu(struct mmu_gather *tlb,
422 unsigned long start, unsigned long end)
423{
Minchan Kim99baac22017-08-10 15:24:12 -0700424 /*
425 * If there are parallel threads are doing PTE changes on same range
426 * under non-exclusive lock(e.g., mmap_sem read-side) but defer TLB
427 * flush by batching, a thread has stable TLB entry can fail to flush
428 * the TLB by observing pte_none|!pte_dirty, for example so flush TLB
429 * forcefully if we detect parallel PTE batching threads.
430 */
431 bool force = mm_tlb_flush_nested(tlb->mm);
432
433 arch_tlb_finish_mmu(tlb, start, end, force);
434 dec_tlb_flush_pending(tlb->mm);
Minchan Kim56236a52017-08-10 15:24:05 -0700435}
436
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 * Note: this doesn't free the actual pages themselves. That
439 * has been handled earlier when unmapping all the memory regions.
440 */
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000441static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
442 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800444 pgtable_t token = pmd_pgtable(*pmd);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700445 pmd_clear(pmd);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000446 pte_free_tlb(tlb, token, addr);
Kirill A. Shutemovc4812902017-11-15 17:35:37 -0800447 mm_dec_nr_ptes(tlb->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448}
449
Hugh Dickinse0da3822005-04-19 13:29:15 -0700450static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
451 unsigned long addr, unsigned long end,
452 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
454 pmd_t *pmd;
455 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700456 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
Hugh Dickinse0da3822005-04-19 13:29:15 -0700458 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 do {
461 next = pmd_addr_end(addr, end);
462 if (pmd_none_or_clear_bad(pmd))
463 continue;
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000464 free_pte_range(tlb, pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 } while (pmd++, addr = next, addr != end);
466
Hugh Dickinse0da3822005-04-19 13:29:15 -0700467 start &= PUD_MASK;
468 if (start < floor)
469 return;
470 if (ceiling) {
471 ceiling &= PUD_MASK;
472 if (!ceiling)
473 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700475 if (end - 1 > ceiling - 1)
476 return;
477
478 pmd = pmd_offset(pud, start);
479 pud_clear(pud);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000480 pmd_free_tlb(tlb, pmd, start);
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -0800481 mm_dec_nr_pmds(tlb->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482}
483
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300484static inline void free_pud_range(struct mmu_gather *tlb, p4d_t *p4d,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700485 unsigned long addr, unsigned long end,
486 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487{
488 pud_t *pud;
489 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700490 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491
Hugh Dickinse0da3822005-04-19 13:29:15 -0700492 start = addr;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300493 pud = pud_offset(p4d, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 do {
495 next = pud_addr_end(addr, end);
496 if (pud_none_or_clear_bad(pud))
497 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700498 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 } while (pud++, addr = next, addr != end);
500
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300501 start &= P4D_MASK;
502 if (start < floor)
503 return;
504 if (ceiling) {
505 ceiling &= P4D_MASK;
506 if (!ceiling)
507 return;
508 }
509 if (end - 1 > ceiling - 1)
510 return;
511
512 pud = pud_offset(p4d, start);
513 p4d_clear(p4d);
514 pud_free_tlb(tlb, pud, start);
Kirill A. Shutemovb4e98d92017-11-15 17:35:33 -0800515 mm_dec_nr_puds(tlb->mm);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300516}
517
518static inline void free_p4d_range(struct mmu_gather *tlb, pgd_t *pgd,
519 unsigned long addr, unsigned long end,
520 unsigned long floor, unsigned long ceiling)
521{
522 p4d_t *p4d;
523 unsigned long next;
524 unsigned long start;
525
526 start = addr;
527 p4d = p4d_offset(pgd, addr);
528 do {
529 next = p4d_addr_end(addr, end);
530 if (p4d_none_or_clear_bad(p4d))
531 continue;
532 free_pud_range(tlb, p4d, addr, next, floor, ceiling);
533 } while (p4d++, addr = next, addr != end);
534
Hugh Dickinse0da3822005-04-19 13:29:15 -0700535 start &= PGDIR_MASK;
536 if (start < floor)
537 return;
538 if (ceiling) {
539 ceiling &= PGDIR_MASK;
540 if (!ceiling)
541 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700543 if (end - 1 > ceiling - 1)
544 return;
545
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300546 p4d = p4d_offset(pgd, start);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700547 pgd_clear(pgd);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300548 p4d_free_tlb(tlb, p4d, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549}
550
551/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700552 * This function frees user-level page tables of a process.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 */
Jan Beulich42b77722008-07-23 21:27:10 -0700554void free_pgd_range(struct mmu_gather *tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700555 unsigned long addr, unsigned long end,
556 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557{
558 pgd_t *pgd;
559 unsigned long next;
560
Hugh Dickinse0da3822005-04-19 13:29:15 -0700561 /*
562 * The next few lines have given us lots of grief...
563 *
564 * Why are we testing PMD* at this top level? Because often
565 * there will be no work to do at all, and we'd prefer not to
566 * go all the way down to the bottom just to discover that.
567 *
568 * Why all these "- 1"s? Because 0 represents both the bottom
569 * of the address space and the top of it (using -1 for the
570 * top wouldn't help much: the masks would do the wrong thing).
571 * The rule is that addr 0 and floor 0 refer to the bottom of
572 * the address space, but end 0 and ceiling 0 refer to the top
573 * Comparisons need to use "end - 1" and "ceiling - 1" (though
574 * that end 0 case should be mythical).
575 *
576 * Wherever addr is brought up or ceiling brought down, we must
577 * be careful to reject "the opposite 0" before it confuses the
578 * subsequent tests. But what about where end is brought down
579 * by PMD_SIZE below? no, end can't go down to 0 there.
580 *
581 * Whereas we round start (addr) and ceiling down, by different
582 * masks at different levels, in order to test whether a table
583 * now has no other vmas using it, so can be freed, we don't
584 * bother to round floor or end up - the tests don't need that.
585 */
586
587 addr &= PMD_MASK;
588 if (addr < floor) {
589 addr += PMD_SIZE;
590 if (!addr)
591 return;
592 }
593 if (ceiling) {
594 ceiling &= PMD_MASK;
595 if (!ceiling)
596 return;
597 }
598 if (end - 1 > ceiling - 1)
599 end -= PMD_SIZE;
600 if (addr > end - 1)
601 return;
Aneesh Kumar K.V07e32662016-12-12 16:42:40 -0800602 /*
603 * We add page table cache pages with PAGE_SIZE,
604 * (see pte_free_tlb()), flush the tlb if we need
605 */
606 tlb_remove_check_page_size_change(tlb, PAGE_SIZE);
Jan Beulich42b77722008-07-23 21:27:10 -0700607 pgd = pgd_offset(tlb->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 do {
609 next = pgd_addr_end(addr, end);
610 if (pgd_none_or_clear_bad(pgd))
611 continue;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300612 free_p4d_range(tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700614}
615
Jan Beulich42b77722008-07-23 21:27:10 -0700616void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700617 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700618{
619 while (vma) {
620 struct vm_area_struct *next = vma->vm_next;
621 unsigned long addr = vma->vm_start;
622
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700623 /*
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000624 * Hide vma from rmap and truncate_pagecache before freeing
625 * pgtables
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700626 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800627 unlink_anon_vmas(vma);
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700628 unlink_file_vma(vma);
629
David Gibson9da61ae2006-03-22 00:08:57 -0800630 if (is_vm_hugetlb_page(vma)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700631 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Tobin C Harding166f61b2017-02-24 14:59:01 -0800632 floor, next ? next->vm_start : ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700633 } else {
634 /*
635 * Optimization: gather nearby vmas into one call down
636 */
637 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
David Gibson48669202006-03-22 00:08:58 -0800638 && !is_vm_hugetlb_page(next)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700639 vma = next;
640 next = vma->vm_next;
Rik van Riel5beb4932010-03-05 13:42:07 -0800641 unlink_anon_vmas(vma);
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700642 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700643 }
644 free_pgd_range(tlb, addr, vma->vm_end,
Tobin C Harding166f61b2017-02-24 14:59:01 -0800645 floor, next ? next->vm_start : ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700646 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700647 vma = next;
648 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649}
650
Kirill A. Shutemov3ed3a4f2016-03-17 14:19:11 -0700651int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652{
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -0800653 spinlock_t *ptl;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800654 pgtable_t new = pte_alloc_one(mm, address);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700655 if (!new)
656 return -ENOMEM;
657
Nick Piggin362a61a2008-05-14 06:37:36 +0200658 /*
659 * Ensure all pte setup (eg. pte page lock and page clearing) are
660 * visible before the pte is made visible to other CPUs by being
661 * put into page tables.
662 *
663 * The other side of the story is the pointer chasing in the page
664 * table walking code (when walking the page table without locking;
665 * ie. most of the time). Fortunately, these data accesses consist
666 * of a chain of data-dependent loads, meaning most CPUs (alpha
667 * being the notable exception) will already guarantee loads are
668 * seen in-order. See the alpha page table accessors for the
669 * smp_read_barrier_depends() barriers in page table walking code.
670 */
671 smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
672
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -0800673 ptl = pmd_lock(mm, pmd);
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800674 if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
Kirill A. Shutemovc4812902017-11-15 17:35:37 -0800675 mm_inc_nr_ptes(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 pmd_populate(mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800677 new = NULL;
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800678 }
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -0800679 spin_unlock(ptl);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800680 if (new)
681 pte_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700682 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683}
684
Hugh Dickins1bb36302005-10-29 18:16:22 -0700685int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700687 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
688 if (!new)
689 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690
Nick Piggin362a61a2008-05-14 06:37:36 +0200691 smp_wmb(); /* See comment in __pte_alloc */
692
Hugh Dickins1bb36302005-10-29 18:16:22 -0700693 spin_lock(&init_mm.page_table_lock);
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800694 if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
Hugh Dickins1bb36302005-10-29 18:16:22 -0700695 pmd_populate_kernel(&init_mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800696 new = NULL;
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800697 }
Hugh Dickins1bb36302005-10-29 18:16:22 -0700698 spin_unlock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800699 if (new)
700 pte_free_kernel(&init_mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700701 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702}
703
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800704static inline void init_rss_vec(int *rss)
Hugh Dickinsae859762005-10-29 18:16:05 -0700705{
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800706 memset(rss, 0, sizeof(int) * NR_MM_COUNTERS);
707}
708
709static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss)
710{
711 int i;
712
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800713 if (current->mm == mm)
David Rientjes05af2e12012-03-21 16:34:13 -0700714 sync_mm_rss(mm);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800715 for (i = 0; i < NR_MM_COUNTERS; i++)
716 if (rss[i])
717 add_mm_counter(mm, i, rss[i]);
Hugh Dickinsae859762005-10-29 18:16:05 -0700718}
719
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800721 * This function is called to print an error when a bad pte
722 * is found. For example, we might have a PFN-mapped pte in
723 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700724 *
725 * The calling function must still handle the error.
726 */
Hugh Dickins3dc14742009-01-06 14:40:08 -0800727static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
728 pte_t pte, struct page *page)
Nick Pigginb5810032005-10-29 18:16:12 -0700729{
Hugh Dickins3dc14742009-01-06 14:40:08 -0800730 pgd_t *pgd = pgd_offset(vma->vm_mm, addr);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300731 p4d_t *p4d = p4d_offset(pgd, addr);
732 pud_t *pud = pud_offset(p4d, addr);
Hugh Dickins3dc14742009-01-06 14:40:08 -0800733 pmd_t *pmd = pmd_offset(pud, addr);
734 struct address_space *mapping;
735 pgoff_t index;
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800736 static unsigned long resume;
737 static unsigned long nr_shown;
738 static unsigned long nr_unshown;
739
740 /*
741 * Allow a burst of 60 reports, then keep quiet for that minute;
742 * or allow a steady drip of one report per second.
743 */
744 if (nr_shown == 60) {
745 if (time_before(jiffies, resume)) {
746 nr_unshown++;
747 return;
748 }
749 if (nr_unshown) {
Joe Perches11705322016-03-17 14:19:50 -0700750 pr_alert("BUG: Bad page map: %lu messages suppressed\n",
751 nr_unshown);
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800752 nr_unshown = 0;
753 }
754 nr_shown = 0;
755 }
756 if (nr_shown++ == 0)
757 resume = jiffies + 60 * HZ;
Hugh Dickins3dc14742009-01-06 14:40:08 -0800758
759 mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;
760 index = linear_page_index(vma, addr);
761
Joe Perches11705322016-03-17 14:19:50 -0700762 pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
763 current->comm,
764 (long long)pte_val(pte), (long long)pmd_val(*pmd));
Wu Fengguang718a3822010-03-10 15:20:43 -0800765 if (page)
Dave Hansenf0b791a2014-01-23 15:52:49 -0800766 dump_page(page, "bad pte");
Joe Perches11705322016-03-17 14:19:50 -0700767 pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
768 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
Konstantin Khlebnikov26825822015-04-15 16:15:08 -0700769 pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n",
770 vma->vm_file,
771 vma->vm_ops ? vma->vm_ops->fault : NULL,
772 vma->vm_file ? vma->vm_file->f_op->mmap : NULL,
773 mapping ? mapping->a_ops->readpage : NULL);
Nick Pigginb5810032005-10-29 18:16:12 -0700774 dump_stack();
Rusty Russell373d4d02013-01-21 17:17:39 +1030775 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Nick Pigginb5810032005-10-29 18:16:12 -0700776}
777
778/*
Nick Piggin7e675132008-04-28 02:13:00 -0700779 * vm_normal_page -- This function gets the "struct page" associated with a pte.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800780 *
Nick Piggin7e675132008-04-28 02:13:00 -0700781 * "Special" mappings do not wish to be associated with a "struct page" (either
782 * it doesn't exist, or it exists but they don't want to touch it). In this
783 * case, NULL is returned here. "Normal" mappings do have a struct page.
Jared Hulbertb379d792008-04-28 02:12:58 -0700784 *
Nick Piggin7e675132008-04-28 02:13:00 -0700785 * There are 2 broad cases. Firstly, an architecture may define a pte_special()
786 * pte bit, in which case this function is trivial. Secondly, an architecture
787 * may not have a spare pte bit, which requires a more complicated scheme,
788 * described below.
789 *
790 * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
791 * special mapping (even if there are underlying and valid "struct pages").
792 * COWed pages of a VM_PFNMAP are always normal.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800793 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700794 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
795 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
Nick Piggin7e675132008-04-28 02:13:00 -0700796 * set, and the vm_pgoff will point to the first PFN mapped: thus every special
797 * mapping will always honor the rule
Linus Torvalds6aab3412005-11-28 14:34:23 -0800798 *
799 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
800 *
Nick Piggin7e675132008-04-28 02:13:00 -0700801 * And for normal mappings this is false.
Jared Hulbertb379d792008-04-28 02:12:58 -0700802 *
Nick Piggin7e675132008-04-28 02:13:00 -0700803 * This restricts such mappings to be a linear translation from virtual address
804 * to pfn. To get around this restriction, we allow arbitrary mappings so long
805 * as the vma is not a COW mapping; in that case, we know that all ptes are
806 * special (because none can have been COWed).
Jared Hulbertb379d792008-04-28 02:12:58 -0700807 *
808 *
Nick Piggin7e675132008-04-28 02:13:00 -0700809 * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
810 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700811 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
812 * page" backing, however the difference is that _all_ pages with a struct
813 * page (that is, those where pfn_valid is true) are refcounted and considered
814 * normal pages by the VM. The disadvantage is that pages are refcounted
815 * (which can be slower and simply not an option for some PFNMAP users). The
816 * advantage is that we don't have to follow the strict linearity rule of
817 * PFNMAP mappings in order to support COWable mappings.
818 *
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800819 */
Jérôme Glissedf6ad692017-09-08 16:12:24 -0700820struct page *_vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
821 pte_t pte, bool with_public_device)
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800822{
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800823 unsigned long pfn = pte_pfn(pte);
Nick Piggin7e675132008-04-28 02:13:00 -0700824
Laurent Dufour00b3a332018-06-07 17:06:12 -0700825 if (IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL)) {
Hugh Dickinsb38af472014-08-29 15:18:44 -0700826 if (likely(!pte_special(pte)))
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800827 goto check_pfn;
David Vrabel667a0a02014-12-18 14:48:15 +0000828 if (vma->vm_ops && vma->vm_ops->find_special_page)
829 return vma->vm_ops->find_special_page(vma, addr);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -0700830 if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))
831 return NULL;
Jérôme Glissedf6ad692017-09-08 16:12:24 -0700832 if (is_zero_pfn(pfn))
833 return NULL;
834
835 /*
836 * Device public pages are special pages (they are ZONE_DEVICE
837 * pages but different from persistent memory). They behave
838 * allmost like normal pages. The difference is that they are
839 * not on the lru and thus should never be involve with any-
840 * thing that involve lru manipulation (mlock, numa balancing,
841 * ...).
842 *
843 * This is why we still want to return NULL for such page from
844 * vm_normal_page() so that we do not have to special case all
845 * call site of vm_normal_page().
846 */
Reza Arbab7d790d22017-10-03 16:15:35 -0700847 if (likely(pfn <= highest_memmap_pfn)) {
Jérôme Glissedf6ad692017-09-08 16:12:24 -0700848 struct page *page = pfn_to_page(pfn);
849
850 if (is_device_public_page(page)) {
851 if (with_public_device)
852 return page;
853 return NULL;
854 }
855 }
Dave Jiange1fb4a02018-08-17 15:43:40 -0700856
857 if (pte_devmap(pte))
858 return NULL;
859
Jérôme Glissedf6ad692017-09-08 16:12:24 -0700860 print_bad_pte(vma, addr, pte, NULL);
Nick Piggin7e675132008-04-28 02:13:00 -0700861 return NULL;
862 }
863
Laurent Dufour00b3a332018-06-07 17:06:12 -0700864 /* !CONFIG_ARCH_HAS_PTE_SPECIAL case follows: */
Nick Piggin7e675132008-04-28 02:13:00 -0700865
Jared Hulbertb379d792008-04-28 02:12:58 -0700866 if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
867 if (vma->vm_flags & VM_MIXEDMAP) {
868 if (!pfn_valid(pfn))
869 return NULL;
870 goto out;
871 } else {
Nick Piggin7e675132008-04-28 02:13:00 -0700872 unsigned long off;
873 off = (addr - vma->vm_start) >> PAGE_SHIFT;
Jared Hulbertb379d792008-04-28 02:12:58 -0700874 if (pfn == vma->vm_pgoff + off)
875 return NULL;
876 if (!is_cow_mapping(vma->vm_flags))
877 return NULL;
878 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800879 }
880
Hugh Dickinsb38af472014-08-29 15:18:44 -0700881 if (is_zero_pfn(pfn))
882 return NULL;
Laurent Dufour00b3a332018-06-07 17:06:12 -0700883
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800884check_pfn:
885 if (unlikely(pfn > highest_memmap_pfn)) {
886 print_bad_pte(vma, addr, pte, NULL);
887 return NULL;
888 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800889
890 /*
Nick Piggin7e675132008-04-28 02:13:00 -0700891 * NOTE! We still have PageReserved() pages in the page tables.
Nick Piggin7e675132008-04-28 02:13:00 -0700892 * eg. VDSO mappings can cause them to exist.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800893 */
Jared Hulbertb379d792008-04-28 02:12:58 -0700894out:
Linus Torvalds6aab3412005-11-28 14:34:23 -0800895 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800896}
897
Gerald Schaefer28093f92016-04-28 16:18:35 -0700898#ifdef CONFIG_TRANSPARENT_HUGEPAGE
899struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr,
900 pmd_t pmd)
901{
902 unsigned long pfn = pmd_pfn(pmd);
903
904 /*
905 * There is no pmd_special() but there may be special pmds, e.g.
906 * in a direct-access (dax) mapping, so let's just replicate the
Laurent Dufour00b3a332018-06-07 17:06:12 -0700907 * !CONFIG_ARCH_HAS_PTE_SPECIAL case from vm_normal_page() here.
Gerald Schaefer28093f92016-04-28 16:18:35 -0700908 */
909 if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
910 if (vma->vm_flags & VM_MIXEDMAP) {
911 if (!pfn_valid(pfn))
912 return NULL;
913 goto out;
914 } else {
915 unsigned long off;
916 off = (addr - vma->vm_start) >> PAGE_SHIFT;
917 if (pfn == vma->vm_pgoff + off)
918 return NULL;
919 if (!is_cow_mapping(vma->vm_flags))
920 return NULL;
921 }
922 }
923
Dave Jiange1fb4a02018-08-17 15:43:40 -0700924 if (pmd_devmap(pmd))
925 return NULL;
Gerald Schaefer28093f92016-04-28 16:18:35 -0700926 if (is_zero_pfn(pfn))
927 return NULL;
928 if (unlikely(pfn > highest_memmap_pfn))
929 return NULL;
930
931 /*
932 * NOTE! We still have PageReserved() pages in the page tables.
933 * eg. VDSO mappings can cause them to exist.
934 */
935out:
936 return pfn_to_page(pfn);
937}
938#endif
939
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800940/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 * copy one vm_area from one task to the other. Assumes the page tables
942 * already present in the new task to be cleared in the whole range
943 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 */
945
Hugh Dickins570a335b2009-12-14 17:58:46 -0800946static inline unsigned long
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700948 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700949 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950{
Nick Pigginb5810032005-10-29 18:16:12 -0700951 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 pte_t pte = *src_pte;
953 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954
955 /* pte contains position in swap or file, so copy. */
956 if (unlikely(!pte_present(pte))) {
Kirill A. Shutemov0661a332015-02-10 14:10:04 -0800957 swp_entry_t entry = pte_to_swp_entry(pte);
Christoph Lameter06972122006-06-23 02:03:35 -0700958
Kirill A. Shutemov0661a332015-02-10 14:10:04 -0800959 if (likely(!non_swap_entry(entry))) {
960 if (swap_duplicate(entry) < 0)
961 return entry.val;
Hugh Dickins570a335b2009-12-14 17:58:46 -0800962
Kirill A. Shutemov0661a332015-02-10 14:10:04 -0800963 /* make sure dst_mm is on swapoff's mmlist. */
964 if (unlikely(list_empty(&dst_mm->mmlist))) {
965 spin_lock(&mmlist_lock);
966 if (list_empty(&dst_mm->mmlist))
967 list_add(&dst_mm->mmlist,
968 &src_mm->mmlist);
969 spin_unlock(&mmlist_lock);
970 }
971 rss[MM_SWAPENTS]++;
972 } else if (is_migration_entry(entry)) {
973 page = migration_entry_to_page(entry);
Konstantin Khlebnikov9f9f1ac2012-01-20 14:34:24 -0800974
Jerome Marchandeca56ff2016-01-14 15:19:26 -0800975 rss[mm_counter(page)]++;
Konstantin Khlebnikov9f9f1ac2012-01-20 14:34:24 -0800976
Kirill A. Shutemov0661a332015-02-10 14:10:04 -0800977 if (is_write_migration_entry(entry) &&
978 is_cow_mapping(vm_flags)) {
979 /*
980 * COW mappings require pages in both
981 * parent and child to be set to read.
982 */
983 make_migration_entry_read(&entry);
984 pte = swp_entry_to_pte(entry);
985 if (pte_swp_soft_dirty(*src_pte))
986 pte = pte_swp_mksoft_dirty(pte);
987 set_pte_at(src_mm, addr, src_pte, pte);
Christoph Lameter06972122006-06-23 02:03:35 -0700988 }
Jérôme Glisse5042db42017-09-08 16:11:43 -0700989 } else if (is_device_private_entry(entry)) {
990 page = device_private_entry_to_page(entry);
991
992 /*
993 * Update rss count even for unaddressable pages, as
994 * they should treated just like normal pages in this
995 * respect.
996 *
997 * We will likely want to have some new rss counters
998 * for unaddressable pages, at some point. But for now
999 * keep things as they are.
1000 */
1001 get_page(page);
1002 rss[mm_counter(page)]++;
1003 page_dup_rmap(page, false);
1004
1005 /*
1006 * We do not preserve soft-dirty information, because so
1007 * far, checkpoint/restore is the only feature that
1008 * requires that. And checkpoint/restore does not work
1009 * when a device driver is involved (you cannot easily
1010 * save and restore device driver state).
1011 */
1012 if (is_write_device_private_entry(entry) &&
1013 is_cow_mapping(vm_flags)) {
1014 make_device_private_entry_read(&entry);
1015 pte = swp_entry_to_pte(entry);
1016 set_pte_at(src_mm, addr, src_pte, pte);
1017 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 }
Hugh Dickinsae859762005-10-29 18:16:05 -07001019 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 }
1021
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 /*
1023 * If it's a COW mapping, write protect it both
1024 * in the parent and the child
1025 */
Linus Torvalds67121172005-12-11 20:38:17 -08001026 if (is_cow_mapping(vm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 ptep_set_wrprotect(src_mm, addr, src_pte);
Zachary Amsden3dc90792006-09-30 23:29:30 -07001028 pte = pte_wrprotect(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 }
1030
1031 /*
1032 * If it's a shared mapping, mark it clean in
1033 * the child
1034 */
1035 if (vm_flags & VM_SHARED)
1036 pte = pte_mkclean(pte);
1037 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001038
1039 page = vm_normal_page(vma, addr, pte);
1040 if (page) {
1041 get_page(page);
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08001042 page_dup_rmap(page, false);
Jerome Marchandeca56ff2016-01-14 15:19:26 -08001043 rss[mm_counter(page)]++;
Jérôme Glissedf6ad692017-09-08 16:12:24 -07001044 } else if (pte_devmap(pte)) {
1045 page = pte_page(pte);
1046
1047 /*
1048 * Cache coherent device memory behave like regular page and
1049 * not like persistent memory page. For more informations see
1050 * MEMORY_DEVICE_CACHE_COHERENT in memory_hotplug.h
1051 */
1052 if (is_device_public_page(page)) {
1053 get_page(page);
1054 page_dup_rmap(page, false);
1055 rss[mm_counter(page)]++;
1056 }
Linus Torvalds6aab3412005-11-28 14:34:23 -08001057 }
Hugh Dickinsae859762005-10-29 18:16:05 -07001058
1059out_set_pte:
1060 set_pte_at(dst_mm, addr, dst_pte, pte);
Hugh Dickins570a335b2009-12-14 17:58:46 -08001061 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062}
1063
Jerome Marchand21bda262014-08-06 16:06:56 -07001064static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001065 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
1066 unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067{
Daisuke Nishimurac36987e2009-10-26 16:50:23 -07001068 pte_t *orig_src_pte, *orig_dst_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001070 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -07001071 int progress = 0;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001072 int rss[NR_MM_COUNTERS];
Hugh Dickins570a335b2009-12-14 17:58:46 -08001073 swp_entry_t entry = (swp_entry_t){0};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074
1075again:
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001076 init_rss_vec(rss);
1077
Hugh Dickinsc74df322005-10-29 18:16:23 -07001078 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 if (!dst_pte)
1080 return -ENOMEM;
Peter Zijlstraece0e2b2010-10-26 14:21:52 -07001081 src_pte = pte_offset_map(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001082 src_ptl = pte_lockptr(src_mm, src_pmd);
Ingo Molnarf20dc5f2006-07-03 00:25:08 -07001083 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
Daisuke Nishimurac36987e2009-10-26 16:50:23 -07001084 orig_src_pte = src_pte;
1085 orig_dst_pte = dst_pte;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001086 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 do {
1089 /*
1090 * We are holding two locks at this point - either of them
1091 * could generate latencies in another task on another CPU.
1092 */
Hugh Dickinse040f212005-10-29 18:15:53 -07001093 if (progress >= 32) {
1094 progress = 0;
1095 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +01001096 spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -07001097 break;
1098 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 if (pte_none(*src_pte)) {
1100 progress++;
1101 continue;
1102 }
Hugh Dickins570a335b2009-12-14 17:58:46 -08001103 entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte,
1104 vma, addr, rss);
1105 if (entry.val)
1106 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 progress += 8;
1108 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001110 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001111 spin_unlock(src_ptl);
Peter Zijlstraece0e2b2010-10-26 14:21:52 -07001112 pte_unmap(orig_src_pte);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001113 add_mm_rss_vec(dst_mm, rss);
Daisuke Nishimurac36987e2009-10-26 16:50:23 -07001114 pte_unmap_unlock(orig_dst_pte, dst_ptl);
Hugh Dickinsc74df322005-10-29 18:16:23 -07001115 cond_resched();
Hugh Dickins570a335b2009-12-14 17:58:46 -08001116
1117 if (entry.val) {
1118 if (add_swap_count_continuation(entry, GFP_KERNEL) < 0)
1119 return -ENOMEM;
1120 progress = 0;
1121 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 if (addr != end)
1123 goto again;
1124 return 0;
1125}
1126
1127static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
1128 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
1129 unsigned long addr, unsigned long end)
1130{
1131 pmd_t *src_pmd, *dst_pmd;
1132 unsigned long next;
1133
1134 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
1135 if (!dst_pmd)
1136 return -ENOMEM;
1137 src_pmd = pmd_offset(src_pud, addr);
1138 do {
1139 next = pmd_addr_end(addr, end);
Zi Yan84c3fc42017-09-08 16:11:01 -07001140 if (is_swap_pmd(*src_pmd) || pmd_trans_huge(*src_pmd)
1141 || pmd_devmap(*src_pmd)) {
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001142 int err;
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08001143 VM_BUG_ON_VMA(next-addr != HPAGE_PMD_SIZE, vma);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001144 err = copy_huge_pmd(dst_mm, src_mm,
1145 dst_pmd, src_pmd, addr, vma);
1146 if (err == -ENOMEM)
1147 return -ENOMEM;
1148 if (!err)
1149 continue;
1150 /* fall through */
1151 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 if (pmd_none_or_clear_bad(src_pmd))
1153 continue;
1154 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
1155 vma, addr, next))
1156 return -ENOMEM;
1157 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
1158 return 0;
1159}
1160
1161static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001162 p4d_t *dst_p4d, p4d_t *src_p4d, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 unsigned long addr, unsigned long end)
1164{
1165 pud_t *src_pud, *dst_pud;
1166 unsigned long next;
1167
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001168 dst_pud = pud_alloc(dst_mm, dst_p4d, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 if (!dst_pud)
1170 return -ENOMEM;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001171 src_pud = pud_offset(src_p4d, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 do {
1173 next = pud_addr_end(addr, end);
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08001174 if (pud_trans_huge(*src_pud) || pud_devmap(*src_pud)) {
1175 int err;
1176
1177 VM_BUG_ON_VMA(next-addr != HPAGE_PUD_SIZE, vma);
1178 err = copy_huge_pud(dst_mm, src_mm,
1179 dst_pud, src_pud, addr, vma);
1180 if (err == -ENOMEM)
1181 return -ENOMEM;
1182 if (!err)
1183 continue;
1184 /* fall through */
1185 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 if (pud_none_or_clear_bad(src_pud))
1187 continue;
1188 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
1189 vma, addr, next))
1190 return -ENOMEM;
1191 } while (dst_pud++, src_pud++, addr = next, addr != end);
1192 return 0;
1193}
1194
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001195static inline int copy_p4d_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
1196 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
1197 unsigned long addr, unsigned long end)
1198{
1199 p4d_t *src_p4d, *dst_p4d;
1200 unsigned long next;
1201
1202 dst_p4d = p4d_alloc(dst_mm, dst_pgd, addr);
1203 if (!dst_p4d)
1204 return -ENOMEM;
1205 src_p4d = p4d_offset(src_pgd, addr);
1206 do {
1207 next = p4d_addr_end(addr, end);
1208 if (p4d_none_or_clear_bad(src_p4d))
1209 continue;
1210 if (copy_pud_range(dst_mm, src_mm, dst_p4d, src_p4d,
1211 vma, addr, next))
1212 return -ENOMEM;
1213 } while (dst_p4d++, src_p4d++, addr = next, addr != end);
1214 return 0;
1215}
1216
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
1218 struct vm_area_struct *vma)
1219{
1220 pgd_t *src_pgd, *dst_pgd;
1221 unsigned long next;
1222 unsigned long addr = vma->vm_start;
1223 unsigned long end = vma->vm_end;
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001224 unsigned long mmun_start; /* For mmu_notifiers */
1225 unsigned long mmun_end; /* For mmu_notifiers */
1226 bool is_cow;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001227 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228
Nick Piggind9928952005-08-28 16:49:11 +10001229 /*
1230 * Don't copy ptes where a page fault will fill them correctly.
1231 * Fork becomes much lighter when there are big shared or private
1232 * readonly mappings. The tradeoff is that copy_page_range is more
1233 * efficient than faulting.
1234 */
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08001235 if (!(vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) &&
1236 !vma->anon_vma)
1237 return 0;
Nick Piggind9928952005-08-28 16:49:11 +10001238
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 if (is_vm_hugetlb_page(vma))
1240 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
1241
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07001242 if (unlikely(vma->vm_flags & VM_PFNMAP)) {
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001243 /*
1244 * We do not free on error cases below as remove_vma
1245 * gets called on error from higher level routine
1246 */
Suresh Siddha5180da42012-10-08 16:28:29 -07001247 ret = track_pfn_copy(vma);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001248 if (ret)
1249 return ret;
1250 }
1251
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001252 /*
1253 * We need to invalidate the secondary MMU mappings only when
1254 * there could be a permission downgrade on the ptes of the
1255 * parent mm. And a permission downgrade will only happen if
1256 * is_cow_mapping() returns true.
1257 */
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001258 is_cow = is_cow_mapping(vma->vm_flags);
1259 mmun_start = addr;
1260 mmun_end = end;
1261 if (is_cow)
1262 mmu_notifier_invalidate_range_start(src_mm, mmun_start,
1263 mmun_end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001264
1265 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 dst_pgd = pgd_offset(dst_mm, addr);
1267 src_pgd = pgd_offset(src_mm, addr);
1268 do {
1269 next = pgd_addr_end(addr, end);
1270 if (pgd_none_or_clear_bad(src_pgd))
1271 continue;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001272 if (unlikely(copy_p4d_range(dst_mm, src_mm, dst_pgd, src_pgd,
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001273 vma, addr, next))) {
1274 ret = -ENOMEM;
1275 break;
1276 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001278
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001279 if (is_cow)
1280 mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001281 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282}
1283
Robin Holt51c6f662005-11-13 16:06:42 -08001284static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -07001285 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 unsigned long addr, unsigned long end,
Peter Zijlstra97a89412011-05-24 17:12:04 -07001287 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288{
Nick Pigginb5810032005-10-29 18:16:12 -07001289 struct mm_struct *mm = tlb->mm;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001290 int force_flush = 0;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001291 int rss[NR_MM_COUNTERS];
Peter Zijlstra97a89412011-05-24 17:12:04 -07001292 spinlock_t *ptl;
Steven Rostedt5f1a1902011-06-15 15:08:23 -07001293 pte_t *start_pte;
Peter Zijlstra97a89412011-05-24 17:12:04 -07001294 pte_t *pte;
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001295 swp_entry_t entry;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001296
Aneesh Kumar K.V07e32662016-12-12 16:42:40 -08001297 tlb_remove_check_page_size_change(tlb, PAGE_SIZE);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001298again:
Peter Zijlstrae3032972011-05-24 17:12:01 -07001299 init_rss_vec(rss);
Steven Rostedt5f1a1902011-06-15 15:08:23 -07001300 start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
1301 pte = start_pte;
Mel Gorman3ea27712017-08-02 13:31:52 -07001302 flush_tlb_batched_pending(mm);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001303 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 do {
1305 pte_t ptent = *pte;
Tobin C Harding166f61b2017-02-24 14:59:01 -08001306 if (pte_none(ptent))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 continue;
Hugh Dickins6f5e6b92006-03-16 23:04:09 -08001308
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -08001310 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -08001311
Jérôme Glissedf6ad692017-09-08 16:12:24 -07001312 page = _vm_normal_page(vma, addr, ptent, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 if (unlikely(details) && page) {
1314 /*
1315 * unmap_shared_mapping_pages() wants to
1316 * invalidate cache without truncating:
1317 * unmap shared but keep private pages.
1318 */
1319 if (details->check_mapping &&
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001320 details->check_mapping != page_rmapping(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 }
Nick Pigginb5810032005-10-29 18:16:12 -07001323 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -07001324 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 tlb_remove_tlb_entry(tlb, pte, addr);
1326 if (unlikely(!page))
1327 continue;
Jerome Marchandeca56ff2016-01-14 15:19:26 -08001328
1329 if (!PageAnon(page)) {
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001330 if (pte_dirty(ptent)) {
1331 force_flush = 1;
Hugh Dickins6237bcd2005-10-29 18:15:54 -07001332 set_page_dirty(page);
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001333 }
Johannes Weiner4917e5d2009-01-06 14:39:17 -08001334 if (pte_young(ptent) &&
Joe Perches64363aa2013-07-08 16:00:18 -07001335 likely(!(vma->vm_flags & VM_SEQ_READ)))
Nick Pigginbf3f3bc2009-01-06 14:38:55 -08001336 mark_page_accessed(page);
Hugh Dickins6237bcd2005-10-29 18:15:54 -07001337 }
Jerome Marchandeca56ff2016-01-14 15:19:26 -08001338 rss[mm_counter(page)]--;
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001339 page_remove_rmap(page, false);
Hugh Dickins3dc14742009-01-06 14:40:08 -08001340 if (unlikely(page_mapcount(page) < 0))
1341 print_bad_pte(vma, addr, ptent, page);
Aneesh Kumar K.Ve9d55e12016-07-26 15:24:09 -07001342 if (unlikely(__tlb_remove_page(tlb, page))) {
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001343 force_flush = 1;
Will Deaconce9ec372014-10-28 13:16:28 -07001344 addr += PAGE_SIZE;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001345 break;
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001346 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 continue;
1348 }
Jérôme Glisse5042db42017-09-08 16:11:43 -07001349
1350 entry = pte_to_swp_entry(ptent);
1351 if (non_swap_entry(entry) && is_device_private_entry(entry)) {
1352 struct page *page = device_private_entry_to_page(entry);
1353
1354 if (unlikely(details && details->check_mapping)) {
1355 /*
1356 * unmap_shared_mapping_pages() wants to
1357 * invalidate cache without truncating:
1358 * unmap shared but keep private pages.
1359 */
1360 if (details->check_mapping !=
1361 page_rmapping(page))
1362 continue;
1363 }
1364
1365 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
1366 rss[mm_counter(page)]--;
1367 page_remove_rmap(page, false);
1368 put_page(page);
1369 continue;
1370 }
1371
Kirill A. Shutemov3e8715f2017-02-22 15:46:34 -08001372 /* If details->check_mapping, we leave swap entries. */
1373 if (unlikely(details))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 continue;
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001375
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001376 entry = pte_to_swp_entry(ptent);
1377 if (!non_swap_entry(entry))
1378 rss[MM_SWAPENTS]--;
1379 else if (is_migration_entry(entry)) {
1380 struct page *page;
Konstantin Khlebnikov9f9f1ac2012-01-20 14:34:24 -08001381
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001382 page = migration_entry_to_page(entry);
Jerome Marchandeca56ff2016-01-14 15:19:26 -08001383 rss[mm_counter(page)]--;
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001384 }
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001385 if (unlikely(!free_swap_and_cache(entry)))
1386 print_bad_pte(vma, addr, ptent, NULL);
Zachary Amsden9888a1c2006-09-30 23:29:31 -07001387 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
Peter Zijlstra97a89412011-05-24 17:12:04 -07001388 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickinsae859762005-10-29 18:16:05 -07001389
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001390 add_mm_rss_vec(mm, rss);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001391 arch_leave_lazy_mmu_mode();
Robin Holt51c6f662005-11-13 16:06:42 -08001392
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001393 /* Do the actual TLB flush before dropping ptl */
Will Deaconfb7332a2014-10-29 10:03:09 +00001394 if (force_flush)
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001395 tlb_flush_mmu_tlbonly(tlb);
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001396 pte_unmap_unlock(start_pte, ptl);
1397
1398 /*
1399 * If we forced a TLB flush (either due to running out of
1400 * batch buffers or because we needed to flush dirty TLB
1401 * entries before releasing the ptl), free the batched
1402 * memory too. Restart if we didn't do everything.
1403 */
1404 if (force_flush) {
1405 force_flush = 0;
1406 tlb_flush_mmu_free(tlb);
Linus Torvalds2b047252013-08-15 11:42:25 -07001407 if (addr != end)
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001408 goto again;
1409 }
1410
Robin Holt51c6f662005-11-13 16:06:42 -08001411 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412}
1413
Robin Holt51c6f662005-11-13 16:06:42 -08001414static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -07001415 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 unsigned long addr, unsigned long end,
Peter Zijlstra97a89412011-05-24 17:12:04 -07001417 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418{
1419 pmd_t *pmd;
1420 unsigned long next;
1421
1422 pmd = pmd_offset(pud, addr);
1423 do {
1424 next = pmd_addr_end(addr, end);
Zi Yan84c3fc42017-09-08 16:11:01 -07001425 if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) {
Hugh Dickins53406ed2018-08-01 11:31:52 -07001426 if (next - addr != HPAGE_PMD_SIZE)
David Rientjesfd607752016-12-12 16:42:20 -08001427 __split_huge_pmd(vma, pmd, addr, false, NULL);
Hugh Dickins53406ed2018-08-01 11:31:52 -07001428 else if (zap_huge_pmd(tlb, vma, pmd, addr))
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001429 goto next;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001430 /* fall through */
1431 }
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001432 /*
1433 * Here there can be other concurrent MADV_DONTNEED or
1434 * trans huge page faults running, and if the pmd is
1435 * none or trans huge it can change under us. This is
1436 * because MADV_DONTNEED holds the mmap_sem in read
1437 * mode.
1438 */
1439 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1440 goto next;
Peter Zijlstra97a89412011-05-24 17:12:04 -07001441 next = zap_pte_range(tlb, vma, pmd, addr, next, details);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001442next:
Peter Zijlstra97a89412011-05-24 17:12:04 -07001443 cond_resched();
1444 } while (pmd++, addr = next, addr != end);
Robin Holt51c6f662005-11-13 16:06:42 -08001445
1446 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447}
1448
Robin Holt51c6f662005-11-13 16:06:42 -08001449static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001450 struct vm_area_struct *vma, p4d_t *p4d,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 unsigned long addr, unsigned long end,
Peter Zijlstra97a89412011-05-24 17:12:04 -07001452 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453{
1454 pud_t *pud;
1455 unsigned long next;
1456
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001457 pud = pud_offset(p4d, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 do {
1459 next = pud_addr_end(addr, end);
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08001460 if (pud_trans_huge(*pud) || pud_devmap(*pud)) {
1461 if (next - addr != HPAGE_PUD_SIZE) {
1462 VM_BUG_ON_VMA(!rwsem_is_locked(&tlb->mm->mmap_sem), vma);
1463 split_huge_pud(vma, pud, addr);
1464 } else if (zap_huge_pud(tlb, vma, pud, addr))
1465 goto next;
1466 /* fall through */
1467 }
Peter Zijlstra97a89412011-05-24 17:12:04 -07001468 if (pud_none_or_clear_bad(pud))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 continue;
Peter Zijlstra97a89412011-05-24 17:12:04 -07001470 next = zap_pmd_range(tlb, vma, pud, addr, next, details);
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08001471next:
1472 cond_resched();
Peter Zijlstra97a89412011-05-24 17:12:04 -07001473 } while (pud++, addr = next, addr != end);
Robin Holt51c6f662005-11-13 16:06:42 -08001474
1475 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476}
1477
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001478static inline unsigned long zap_p4d_range(struct mmu_gather *tlb,
1479 struct vm_area_struct *vma, pgd_t *pgd,
1480 unsigned long addr, unsigned long end,
1481 struct zap_details *details)
1482{
1483 p4d_t *p4d;
1484 unsigned long next;
1485
1486 p4d = p4d_offset(pgd, addr);
1487 do {
1488 next = p4d_addr_end(addr, end);
1489 if (p4d_none_or_clear_bad(p4d))
1490 continue;
1491 next = zap_pud_range(tlb, vma, p4d, addr, next, details);
1492 } while (p4d++, addr = next, addr != end);
1493
1494 return addr;
1495}
1496
Michal Hockoaac45362016-03-25 14:20:24 -07001497void unmap_page_range(struct mmu_gather *tlb,
Al Viro038c7aa2012-03-05 13:25:09 -05001498 struct vm_area_struct *vma,
1499 unsigned long addr, unsigned long end,
1500 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501{
1502 pgd_t *pgd;
1503 unsigned long next;
1504
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 BUG_ON(addr >= end);
1506 tlb_start_vma(tlb, vma);
1507 pgd = pgd_offset(vma->vm_mm, addr);
1508 do {
1509 next = pgd_addr_end(addr, end);
Peter Zijlstra97a89412011-05-24 17:12:04 -07001510 if (pgd_none_or_clear_bad(pgd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 continue;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001512 next = zap_p4d_range(tlb, vma, pgd, addr, next, details);
Peter Zijlstra97a89412011-05-24 17:12:04 -07001513 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 tlb_end_vma(tlb, vma);
1515}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516
Al Virof5cc4ee2012-03-05 14:14:20 -05001517
1518static void unmap_single_vma(struct mmu_gather *tlb,
1519 struct vm_area_struct *vma, unsigned long start_addr,
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001520 unsigned long end_addr,
Al Virof5cc4ee2012-03-05 14:14:20 -05001521 struct zap_details *details)
1522{
1523 unsigned long start = max(vma->vm_start, start_addr);
1524 unsigned long end;
1525
1526 if (start >= vma->vm_end)
1527 return;
1528 end = min(vma->vm_end, end_addr);
1529 if (end <= vma->vm_start)
1530 return;
1531
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +05301532 if (vma->vm_file)
1533 uprobe_munmap(vma, start, end);
1534
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07001535 if (unlikely(vma->vm_flags & VM_PFNMAP))
Suresh Siddha5180da42012-10-08 16:28:29 -07001536 untrack_pfn(vma, 0, 0);
Al Virof5cc4ee2012-03-05 14:14:20 -05001537
1538 if (start != end) {
1539 if (unlikely(is_vm_hugetlb_page(vma))) {
1540 /*
1541 * It is undesirable to test vma->vm_file as it
1542 * should be non-null for valid hugetlb area.
1543 * However, vm_file will be NULL in the error
Davidlohr Bueso7aa6b4a2014-04-07 15:37:01 -07001544 * cleanup path of mmap_region. When
Al Virof5cc4ee2012-03-05 14:14:20 -05001545 * hugetlbfs ->mmap method fails,
Davidlohr Bueso7aa6b4a2014-04-07 15:37:01 -07001546 * mmap_region() nullifies vma->vm_file
Al Virof5cc4ee2012-03-05 14:14:20 -05001547 * before calling this function to clean up.
1548 * Since no pte has actually been setup, it is
1549 * safe to do nothing in this case.
1550 */
Aneesh Kumar K.V24669e52012-07-31 16:42:03 -07001551 if (vma->vm_file) {
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08001552 i_mmap_lock_write(vma->vm_file->f_mapping);
Mel Gormand8333522012-07-31 16:46:20 -07001553 __unmap_hugepage_range_final(tlb, vma, start, end, NULL);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08001554 i_mmap_unlock_write(vma->vm_file->f_mapping);
Aneesh Kumar K.V24669e52012-07-31 16:42:03 -07001555 }
Al Virof5cc4ee2012-03-05 14:14:20 -05001556 } else
1557 unmap_page_range(tlb, vma, start, end, details);
1558 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559}
1560
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561/**
1562 * unmap_vmas - unmap a range of memory covered by a list of vma's
Randy Dunlap0164f692011-06-15 15:08:09 -07001563 * @tlb: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 * @vma: the starting vma
1565 * @start_addr: virtual address at which to start unmapping
1566 * @end_addr: virtual address at which to end unmapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 *
Hugh Dickins508034a2005-10-29 18:16:30 -07001568 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 * Only addresses between `start' and `end' will be unmapped.
1571 *
1572 * The VMA list must be sorted in ascending virtual address order.
1573 *
1574 * unmap_vmas() assumes that the caller will flush the whole unmapped address
1575 * range after unmap_vmas() returns. So the only responsibility here is to
1576 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
1577 * drops the lock and schedules.
1578 */
Al Viro6e8bb012012-03-05 13:41:15 -05001579void unmap_vmas(struct mmu_gather *tlb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 struct vm_area_struct *vma, unsigned long start_addr,
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001581 unsigned long end_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582{
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001583 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001585 mmu_notifier_invalidate_range_start(mm, start_addr, end_addr);
Al Virof5cc4ee2012-03-05 14:14:20 -05001586 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next)
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001587 unmap_single_vma(tlb, vma, start_addr, end_addr, NULL);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001588 mmu_notifier_invalidate_range_end(mm, start_addr, end_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589}
1590
1591/**
1592 * zap_page_range - remove user pages in a given range
1593 * @vma: vm_area_struct holding the applicable pages
Randy Dunlapeb4546b2012-06-20 12:53:02 -07001594 * @start: starting address of pages to zap
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 * @size: number of bytes to zap
Al Virof5cc4ee2012-03-05 14:14:20 -05001596 *
1597 * Caller must protect the VMA list
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 */
Linus Torvalds7e027b12012-05-06 13:43:15 -07001599void zap_page_range(struct vm_area_struct *vma, unsigned long start,
Kirill A. Shutemovecf13852017-02-22 15:46:37 -08001600 unsigned long size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601{
1602 struct mm_struct *mm = vma->vm_mm;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001603 struct mmu_gather tlb;
Linus Torvalds7e027b12012-05-06 13:43:15 -07001604 unsigned long end = start + size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07001607 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001608 update_hiwater_rss(mm);
Linus Torvalds7e027b12012-05-06 13:43:15 -07001609 mmu_notifier_invalidate_range_start(mm, start, end);
Rik van Riel50c150f2018-08-17 15:48:53 -07001610 for ( ; vma && vma->vm_start < end; vma = vma->vm_next)
Kirill A. Shutemovecf13852017-02-22 15:46:37 -08001611 unmap_single_vma(&tlb, vma, start, end, NULL);
Linus Torvalds7e027b12012-05-06 13:43:15 -07001612 mmu_notifier_invalidate_range_end(mm, start, end);
1613 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614}
1615
Jack Steinerc627f9c2008-07-29 22:33:53 -07001616/**
Al Virof5cc4ee2012-03-05 14:14:20 -05001617 * zap_page_range_single - remove user pages in a given range
1618 * @vma: vm_area_struct holding the applicable pages
1619 * @address: starting address of pages to zap
1620 * @size: number of bytes to zap
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001621 * @details: details of shared cache invalidation
Al Virof5cc4ee2012-03-05 14:14:20 -05001622 *
1623 * The range must fit into one VMA.
1624 */
1625static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address,
1626 unsigned long size, struct zap_details *details)
1627{
1628 struct mm_struct *mm = vma->vm_mm;
1629 struct mmu_gather tlb;
1630 unsigned long end = address + size;
Al Virof5cc4ee2012-03-05 14:14:20 -05001631
1632 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07001633 tlb_gather_mmu(&tlb, mm, address, end);
Al Virof5cc4ee2012-03-05 14:14:20 -05001634 update_hiwater_rss(mm);
1635 mmu_notifier_invalidate_range_start(mm, address, end);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001636 unmap_single_vma(&tlb, vma, address, end, details);
Al Virof5cc4ee2012-03-05 14:14:20 -05001637 mmu_notifier_invalidate_range_end(mm, address, end);
1638 tlb_finish_mmu(&tlb, address, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639}
1640
Jack Steinerc627f9c2008-07-29 22:33:53 -07001641/**
1642 * zap_vma_ptes - remove ptes mapping the vma
1643 * @vma: vm_area_struct holding ptes to be zapped
1644 * @address: starting address of pages to zap
1645 * @size: number of bytes to zap
1646 *
1647 * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1648 *
1649 * The entire address range must be fully contained within the vma.
1650 *
Jack Steinerc627f9c2008-07-29 22:33:53 -07001651 */
Leon Romanovsky27d036e2018-05-29 15:14:07 +03001652void zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
Jack Steinerc627f9c2008-07-29 22:33:53 -07001653 unsigned long size)
1654{
1655 if (address < vma->vm_start || address + size > vma->vm_end ||
1656 !(vma->vm_flags & VM_PFNMAP))
Leon Romanovsky27d036e2018-05-29 15:14:07 +03001657 return;
1658
Al Virof5cc4ee2012-03-05 14:14:20 -05001659 zap_page_range_single(vma, address, size, NULL);
Jack Steinerc627f9c2008-07-29 22:33:53 -07001660}
1661EXPORT_SYMBOL_GPL(zap_vma_ptes);
1662
Namhyung Kim25ca1d62010-10-26 14:21:59 -07001663pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr,
Harvey Harrison920c7a52008-02-04 22:29:26 -08001664 spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001665{
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001666 pgd_t *pgd;
1667 p4d_t *p4d;
1668 pud_t *pud;
1669 pmd_t *pmd;
1670
1671 pgd = pgd_offset(mm, addr);
1672 p4d = p4d_alloc(mm, pgd, addr);
1673 if (!p4d)
1674 return NULL;
1675 pud = pud_alloc(mm, p4d, addr);
1676 if (!pud)
1677 return NULL;
1678 pmd = pmd_alloc(mm, pud, addr);
1679 if (!pmd)
1680 return NULL;
1681
1682 VM_BUG_ON(pmd_trans_huge(*pmd));
1683 return pte_alloc_map_lock(mm, pmd, addr, ptl);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001684}
1685
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001687 * This is the old fallback for page remapping.
1688 *
1689 * For historical reasons, it only allows reserved pages. Only
1690 * old drivers should use this, and they needed to mark their
1691 * pages reserved for the old functions anyway.
1692 */
Nick Piggin423bad602008-04-28 02:13:01 -07001693static int insert_page(struct vm_area_struct *vma, unsigned long addr,
1694 struct page *page, pgprot_t prot)
Linus Torvalds238f58d2005-11-29 13:01:56 -08001695{
Nick Piggin423bad602008-04-28 02:13:01 -07001696 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001697 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001698 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001699 spinlock_t *ptl;
1700
Linus Torvalds238f58d2005-11-29 13:01:56 -08001701 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001702 if (PageAnon(page))
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001703 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001704 retval = -ENOMEM;
1705 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001706 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001707 if (!pte)
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001708 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001709 retval = -EBUSY;
1710 if (!pte_none(*pte))
1711 goto out_unlock;
1712
1713 /* Ok, finally just insert the thing.. */
1714 get_page(page);
Jerome Marchandeca56ff2016-01-14 15:19:26 -08001715 inc_mm_counter_fast(mm, mm_counter_file(page));
Kirill A. Shutemovdd78fed2016-07-26 15:25:26 -07001716 page_add_file_rmap(page, false);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001717 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1718
1719 retval = 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001720 pte_unmap_unlock(pte, ptl);
1721 return retval;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001722out_unlock:
1723 pte_unmap_unlock(pte, ptl);
1724out:
1725 return retval;
1726}
1727
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001728/**
1729 * vm_insert_page - insert single page into user vma
1730 * @vma: user vma to map to
1731 * @addr: target user address of this page
1732 * @page: source kernel page
1733 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001734 * This allows drivers to insert individual pages they've allocated
1735 * into a user vma.
1736 *
1737 * The page has to be a nice clean _individual_ kernel allocation.
1738 * If you allocate a compound page, you need to have marked it as
1739 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001740 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001741 *
1742 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1743 * took an arbitrary page protection parameter. This doesn't allow
1744 * that. Your vma protection will have to be set up correctly, which
1745 * means that if you want a shared writable mapping, you'd better
1746 * ask for a shared writable mapping!
1747 *
1748 * The page does not need to be reserved.
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001749 *
1750 * Usually this function is called from f_op->mmap() handler
1751 * under mm->mmap_sem write-lock, so it can change vma->vm_flags.
1752 * Caller must set VM_MIXEDMAP on vma if it wants to call this
1753 * function from other places, for example from page-fault handler.
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001754 */
Nick Piggin423bad602008-04-28 02:13:01 -07001755int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
1756 struct page *page)
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001757{
1758 if (addr < vma->vm_start || addr >= vma->vm_end)
1759 return -EFAULT;
1760 if (!page_count(page))
1761 return -EINVAL;
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001762 if (!(vma->vm_flags & VM_MIXEDMAP)) {
1763 BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem));
1764 BUG_ON(vma->vm_flags & VM_PFNMAP);
1765 vma->vm_flags |= VM_MIXEDMAP;
1766 }
Nick Piggin423bad602008-04-28 02:13:01 -07001767 return insert_page(vma, addr, page, vma->vm_page_prot);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001768}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001769EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001770
Nick Piggin423bad602008-04-28 02:13:01 -07001771static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
Ross Zwislerb2770da62017-09-06 16:18:35 -07001772 pfn_t pfn, pgprot_t prot, bool mkwrite)
Nick Piggin423bad602008-04-28 02:13:01 -07001773{
1774 struct mm_struct *mm = vma->vm_mm;
1775 int retval;
1776 pte_t *pte, entry;
1777 spinlock_t *ptl;
1778
1779 retval = -ENOMEM;
1780 pte = get_locked_pte(mm, addr, &ptl);
1781 if (!pte)
1782 goto out;
1783 retval = -EBUSY;
Ross Zwislerb2770da62017-09-06 16:18:35 -07001784 if (!pte_none(*pte)) {
1785 if (mkwrite) {
1786 /*
1787 * For read faults on private mappings the PFN passed
1788 * in may not match the PFN we have mapped if the
1789 * mapped PFN is a writeable COW page. In the mkwrite
1790 * case we are creating a writable PTE for a shared
1791 * mapping and we expect the PFNs to match.
1792 */
1793 if (WARN_ON_ONCE(pte_pfn(*pte) != pfn_t_to_pfn(pfn)))
1794 goto out_unlock;
1795 entry = *pte;
1796 goto out_mkwrite;
1797 } else
1798 goto out_unlock;
1799 }
Nick Piggin423bad602008-04-28 02:13:01 -07001800
1801 /* Ok, finally just insert the thing.. */
Dan Williams01c8f1c2016-01-15 16:56:40 -08001802 if (pfn_t_devmap(pfn))
1803 entry = pte_mkdevmap(pfn_t_pte(pfn, prot));
1804 else
1805 entry = pte_mkspecial(pfn_t_pte(pfn, prot));
Ross Zwislerb2770da62017-09-06 16:18:35 -07001806
1807out_mkwrite:
1808 if (mkwrite) {
1809 entry = pte_mkyoung(entry);
1810 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1811 }
1812
Nick Piggin423bad602008-04-28 02:13:01 -07001813 set_pte_at(mm, addr, pte, entry);
Russell King4b3073e2009-12-18 16:40:18 +00001814 update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */
Nick Piggin423bad602008-04-28 02:13:01 -07001815
1816 retval = 0;
1817out_unlock:
1818 pte_unmap_unlock(pte, ptl);
1819out:
1820 return retval;
1821}
1822
Nick Piggine0dc0d82007-02-12 00:51:36 -08001823/**
1824 * vm_insert_pfn - insert single pfn into user vma
1825 * @vma: user vma to map to
1826 * @addr: target user address of this page
1827 * @pfn: source kernel pfn
1828 *
Robert P. J. Dayc462f172012-10-08 16:33:43 -07001829 * Similar to vm_insert_page, this allows drivers to insert individual pages
Nick Piggine0dc0d82007-02-12 00:51:36 -08001830 * they've allocated into a user vma. Same comments apply.
1831 *
1832 * This function should only be called from a vm_ops->fault handler, and
1833 * in that case the handler should return NULL.
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001834 *
1835 * vma cannot be a COW mapping.
1836 *
1837 * As this is called only for pages that do not currently exist, we
1838 * do not need to flush old virtual caches or the TLB.
Nick Piggine0dc0d82007-02-12 00:51:36 -08001839 */
1840int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
Nick Piggin423bad602008-04-28 02:13:01 -07001841 unsigned long pfn)
Nick Piggine0dc0d82007-02-12 00:51:36 -08001842{
Andy Lutomirski1745cbc2015-12-29 20:12:20 -08001843 return vm_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot);
1844}
1845EXPORT_SYMBOL(vm_insert_pfn);
1846
1847/**
1848 * vm_insert_pfn_prot - insert single pfn into user vma with specified pgprot
1849 * @vma: user vma to map to
1850 * @addr: target user address of this page
1851 * @pfn: source kernel pfn
1852 * @pgprot: pgprot flags for the inserted page
1853 *
1854 * This is exactly like vm_insert_pfn, except that it allows drivers to
1855 * to override pgprot on a per-page basis.
1856 *
1857 * This only makes sense for IO mappings, and it makes no sense for
1858 * cow mappings. In general, using multiple vmas is preferable;
1859 * vm_insert_pfn_prot should only be used if using multiple VMAs is
1860 * impractical.
1861 */
1862int vm_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr,
1863 unsigned long pfn, pgprot_t pgprot)
1864{
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001865 int ret;
Nick Piggin7e675132008-04-28 02:13:00 -07001866 /*
1867 * Technically, architectures with pte_special can avoid all these
1868 * restrictions (same for remap_pfn_range). However we would like
1869 * consistency in testing and feature parity among all, so we should
1870 * try to keep these invariants in place for everybody.
1871 */
Jared Hulbertb379d792008-04-28 02:12:58 -07001872 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
1873 BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
1874 (VM_PFNMAP|VM_MIXEDMAP));
1875 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1876 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
Nick Piggine0dc0d82007-02-12 00:51:36 -08001877
Nick Piggin423bad602008-04-28 02:13:01 -07001878 if (addr < vma->vm_start || addr >= vma->vm_end)
1879 return -EFAULT;
Borislav Petkov308a0472016-10-26 19:43:43 +02001880
Andi Kleen42e40892018-06-13 15:48:27 -07001881 if (!pfn_modify_allowed(pfn, pgprot))
1882 return -EACCES;
1883
Borislav Petkov308a0472016-10-26 19:43:43 +02001884 track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV));
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001885
Ross Zwislerb2770da62017-09-06 16:18:35 -07001886 ret = insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot,
1887 false);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001888
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001889 return ret;
Nick Piggine0dc0d82007-02-12 00:51:36 -08001890}
Andy Lutomirski1745cbc2015-12-29 20:12:20 -08001891EXPORT_SYMBOL(vm_insert_pfn_prot);
Nick Piggine0dc0d82007-02-12 00:51:36 -08001892
Dan Williams785a3fa2017-10-23 07:20:00 -07001893static bool vm_mixed_ok(struct vm_area_struct *vma, pfn_t pfn)
1894{
1895 /* these checks mirror the abort conditions in vm_normal_page */
1896 if (vma->vm_flags & VM_MIXEDMAP)
1897 return true;
1898 if (pfn_t_devmap(pfn))
1899 return true;
1900 if (pfn_t_special(pfn))
1901 return true;
1902 if (is_zero_pfn(pfn_t_to_pfn(pfn)))
1903 return true;
1904 return false;
1905}
1906
Ross Zwislerb2770da62017-09-06 16:18:35 -07001907static int __vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
1908 pfn_t pfn, bool mkwrite)
Nick Piggin423bad602008-04-28 02:13:01 -07001909{
Dan Williams87744ab2016-10-07 17:00:18 -07001910 pgprot_t pgprot = vma->vm_page_prot;
1911
Dan Williams785a3fa2017-10-23 07:20:00 -07001912 BUG_ON(!vm_mixed_ok(vma, pfn));
Nick Piggin423bad602008-04-28 02:13:01 -07001913
1914 if (addr < vma->vm_start || addr >= vma->vm_end)
1915 return -EFAULT;
Borislav Petkov308a0472016-10-26 19:43:43 +02001916
1917 track_pfn_insert(vma, &pgprot, pfn);
Nick Piggin423bad602008-04-28 02:13:01 -07001918
Andi Kleen42e40892018-06-13 15:48:27 -07001919 if (!pfn_modify_allowed(pfn_t_to_pfn(pfn), pgprot))
1920 return -EACCES;
1921
Nick Piggin423bad602008-04-28 02:13:01 -07001922 /*
1923 * If we don't have pte special, then we have to use the pfn_valid()
1924 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1925 * refcount the page if pfn_valid is true (hence insert_page rather
Hugh Dickins62eede62009-09-21 17:03:34 -07001926 * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP
1927 * without pte special, it would there be refcounted as a normal page.
Nick Piggin423bad602008-04-28 02:13:01 -07001928 */
Laurent Dufour00b3a332018-06-07 17:06:12 -07001929 if (!IS_ENABLED(CONFIG_ARCH_HAS_PTE_SPECIAL) &&
1930 !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) {
Nick Piggin423bad602008-04-28 02:13:01 -07001931 struct page *page;
1932
Dan Williams03fc2da2016-01-26 09:48:05 -08001933 /*
1934 * At this point we are committed to insert_page()
1935 * regardless of whether the caller specified flags that
1936 * result in pfn_t_has_page() == false.
1937 */
1938 page = pfn_to_page(pfn_t_to_pfn(pfn));
Dan Williams87744ab2016-10-07 17:00:18 -07001939 return insert_page(vma, addr, page, pgprot);
Nick Piggin423bad602008-04-28 02:13:01 -07001940 }
Ross Zwislerb2770da62017-09-06 16:18:35 -07001941 return insert_pfn(vma, addr, pfn, pgprot, mkwrite);
1942}
1943
1944int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
1945 pfn_t pfn)
1946{
1947 return __vm_insert_mixed(vma, addr, pfn, false);
1948
Nick Piggin423bad602008-04-28 02:13:01 -07001949}
1950EXPORT_SYMBOL(vm_insert_mixed);
1951
Souptick Joarderab77dab2018-06-07 17:04:29 -07001952/*
1953 * If the insertion of PTE failed because someone else already added a
1954 * different entry in the mean time, we treat that as success as we assume
1955 * the same entry was actually inserted.
1956 */
1957
1958vm_fault_t vmf_insert_mixed_mkwrite(struct vm_area_struct *vma,
1959 unsigned long addr, pfn_t pfn)
Ross Zwislerb2770da62017-09-06 16:18:35 -07001960{
Souptick Joarderab77dab2018-06-07 17:04:29 -07001961 int err;
1962
1963 err = __vm_insert_mixed(vma, addr, pfn, true);
1964 if (err == -ENOMEM)
1965 return VM_FAULT_OOM;
1966 if (err < 0 && err != -EBUSY)
1967 return VM_FAULT_SIGBUS;
1968 return VM_FAULT_NOPAGE;
Ross Zwislerb2770da62017-09-06 16:18:35 -07001969}
Souptick Joarderab77dab2018-06-07 17:04:29 -07001970EXPORT_SYMBOL(vmf_insert_mixed_mkwrite);
Ross Zwislerb2770da62017-09-06 16:18:35 -07001971
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001972/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 * maps a range of physical memory into the requested pages. the old
1974 * mappings are removed. any references to nonexistent pages results
1975 * in null mappings (currently treated as "copy-on-access")
1976 */
1977static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1978 unsigned long addr, unsigned long end,
1979 unsigned long pfn, pgprot_t prot)
1980{
1981 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001982 spinlock_t *ptl;
Andi Kleen42e40892018-06-13 15:48:27 -07001983 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984
Hugh Dickinsc74df322005-10-29 18:16:23 -07001985 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 if (!pte)
1987 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001988 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 do {
1990 BUG_ON(!pte_none(*pte));
Andi Kleen42e40892018-06-13 15:48:27 -07001991 if (!pfn_modify_allowed(pfn, prot)) {
1992 err = -EACCES;
1993 break;
1994 }
Nick Piggin7e675132008-04-28 02:13:00 -07001995 set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 pfn++;
1997 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001998 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001999 pte_unmap_unlock(pte - 1, ptl);
Andi Kleen42e40892018-06-13 15:48:27 -07002000 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001}
2002
2003static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
2004 unsigned long addr, unsigned long end,
2005 unsigned long pfn, pgprot_t prot)
2006{
2007 pmd_t *pmd;
2008 unsigned long next;
Andi Kleen42e40892018-06-13 15:48:27 -07002009 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010
2011 pfn -= addr >> PAGE_SHIFT;
2012 pmd = pmd_alloc(mm, pud, addr);
2013 if (!pmd)
2014 return -ENOMEM;
Andrea Arcangelif66055ab2011-01-13 15:46:54 -08002015 VM_BUG_ON(pmd_trans_huge(*pmd));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 do {
2017 next = pmd_addr_end(addr, end);
Andi Kleen42e40892018-06-13 15:48:27 -07002018 err = remap_pte_range(mm, pmd, addr, next,
2019 pfn + (addr >> PAGE_SHIFT), prot);
2020 if (err)
2021 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 } while (pmd++, addr = next, addr != end);
2023 return 0;
2024}
2025
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03002026static inline int remap_pud_range(struct mm_struct *mm, p4d_t *p4d,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 unsigned long addr, unsigned long end,
2028 unsigned long pfn, pgprot_t prot)
2029{
2030 pud_t *pud;
2031 unsigned long next;
Andi Kleen42e40892018-06-13 15:48:27 -07002032 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033
2034 pfn -= addr >> PAGE_SHIFT;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03002035 pud = pud_alloc(mm, p4d, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 if (!pud)
2037 return -ENOMEM;
2038 do {
2039 next = pud_addr_end(addr, end);
Andi Kleen42e40892018-06-13 15:48:27 -07002040 err = remap_pmd_range(mm, pud, addr, next,
2041 pfn + (addr >> PAGE_SHIFT), prot);
2042 if (err)
2043 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 } while (pud++, addr = next, addr != end);
2045 return 0;
2046}
2047
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03002048static inline int remap_p4d_range(struct mm_struct *mm, pgd_t *pgd,
2049 unsigned long addr, unsigned long end,
2050 unsigned long pfn, pgprot_t prot)
2051{
2052 p4d_t *p4d;
2053 unsigned long next;
Andi Kleen42e40892018-06-13 15:48:27 -07002054 int err;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03002055
2056 pfn -= addr >> PAGE_SHIFT;
2057 p4d = p4d_alloc(mm, pgd, addr);
2058 if (!p4d)
2059 return -ENOMEM;
2060 do {
2061 next = p4d_addr_end(addr, end);
Andi Kleen42e40892018-06-13 15:48:27 -07002062 err = remap_pud_range(mm, p4d, addr, next,
2063 pfn + (addr >> PAGE_SHIFT), prot);
2064 if (err)
2065 return err;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03002066 } while (p4d++, addr = next, addr != end);
2067 return 0;
2068}
2069
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07002070/**
2071 * remap_pfn_range - remap kernel memory to userspace
2072 * @vma: user vma to map to
2073 * @addr: target user address to start at
2074 * @pfn: physical address of kernel memory
2075 * @size: size of map area
2076 * @prot: page protection flags for this mapping
2077 *
2078 * Note: this is only safe if the mm semaphore is held when called.
2079 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
2081 unsigned long pfn, unsigned long size, pgprot_t prot)
2082{
2083 pgd_t *pgd;
2084 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07002085 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 struct mm_struct *mm = vma->vm_mm;
Yongji Xied5957d22016-05-20 16:57:41 -07002087 unsigned long remap_pfn = pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 int err;
2089
2090 /*
2091 * Physically remapped pages are special. Tell the
2092 * rest of the world about it:
2093 * VM_IO tells people not to look at these pages
2094 * (accesses can have side effects).
Linus Torvalds6aab3412005-11-28 14:34:23 -08002095 * VM_PFNMAP tells the core MM that the base pages are just
2096 * raw PFN mappings, and do not have a "struct page" associated
2097 * with them.
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -07002098 * VM_DONTEXPAND
2099 * Disable vma merging and expanding with mremap().
2100 * VM_DONTDUMP
2101 * Omit vma from core dump, even when VM_IO turned off.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08002102 *
2103 * There's a horrible special case to handle copy-on-write
2104 * behaviour that some programs depend on. We mark the "original"
2105 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07002106 * See vm_normal_page() for details.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 */
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07002108 if (is_cow_mapping(vma->vm_flags)) {
2109 if (addr != vma->vm_start || end != vma->vm_end)
2110 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08002111 vma->vm_pgoff = pfn;
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07002112 }
2113
Yongji Xied5957d22016-05-20 16:57:41 -07002114 err = track_pfn_remap(vma, &prot, remap_pfn, addr, PAGE_ALIGN(size));
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07002115 if (err)
venkatesh.pallipadi@intel.com3c8bb732008-12-18 11:41:27 -08002116 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08002117
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -07002118 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119
2120 BUG_ON(addr >= end);
2121 pfn -= addr >> PAGE_SHIFT;
2122 pgd = pgd_offset(mm, addr);
2123 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 do {
2125 next = pgd_addr_end(addr, end);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03002126 err = remap_p4d_range(mm, pgd, addr, next,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 pfn + (addr >> PAGE_SHIFT), prot);
2128 if (err)
2129 break;
2130 } while (pgd++, addr = next, addr != end);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08002131
2132 if (err)
Yongji Xied5957d22016-05-20 16:57:41 -07002133 untrack_pfn(vma, remap_pfn, PAGE_ALIGN(size));
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08002134
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 return err;
2136}
2137EXPORT_SYMBOL(remap_pfn_range);
2138
Linus Torvaldsb4cbb192013-04-16 13:45:37 -07002139/**
2140 * vm_iomap_memory - remap memory to userspace
2141 * @vma: user vma to map to
2142 * @start: start of area
2143 * @len: size of area
2144 *
2145 * This is a simplified io_remap_pfn_range() for common driver use. The
2146 * driver just needs to give us the physical memory range to be mapped,
2147 * we'll figure out the rest from the vma information.
2148 *
2149 * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get
2150 * whatever write-combining details or similar.
2151 */
2152int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
2153{
2154 unsigned long vm_len, pfn, pages;
2155
2156 /* Check that the physical memory area passed in looks valid */
2157 if (start + len < start)
2158 return -EINVAL;
2159 /*
2160 * You *really* shouldn't map things that aren't page-aligned,
2161 * but we've historically allowed it because IO memory might
2162 * just have smaller alignment.
2163 */
2164 len += start & ~PAGE_MASK;
2165 pfn = start >> PAGE_SHIFT;
2166 pages = (len + ~PAGE_MASK) >> PAGE_SHIFT;
2167 if (pfn + pages < pfn)
2168 return -EINVAL;
2169
2170 /* We start the mapping 'vm_pgoff' pages into the area */
2171 if (vma->vm_pgoff > pages)
2172 return -EINVAL;
2173 pfn += vma->vm_pgoff;
2174 pages -= vma->vm_pgoff;
2175
2176 /* Can we fit all of the mapping? */
2177 vm_len = vma->vm_end - vma->vm_start;
2178 if (vm_len >> PAGE_SHIFT > pages)
2179 return -EINVAL;
2180
2181 /* Ok, let it rip */
2182 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
2183}
2184EXPORT_SYMBOL(vm_iomap_memory);
2185
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002186static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
2187 unsigned long addr, unsigned long end,
2188 pte_fn_t fn, void *data)
2189{
2190 pte_t *pte;
2191 int err;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08002192 pgtable_t token;
Borislav Petkov94909912007-05-06 14:49:17 -07002193 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002194
2195 pte = (mm == &init_mm) ?
2196 pte_alloc_kernel(pmd, addr) :
2197 pte_alloc_map_lock(mm, pmd, addr, &ptl);
2198 if (!pte)
2199 return -ENOMEM;
2200
2201 BUG_ON(pmd_huge(*pmd));
2202
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08002203 arch_enter_lazy_mmu_mode();
2204
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08002205 token = pmd_pgtable(*pmd);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002206
2207 do {
Daisuke Nishimurac36987e2009-10-26 16:50:23 -07002208 err = fn(pte++, token, addr, data);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002209 if (err)
2210 break;
Daisuke Nishimurac36987e2009-10-26 16:50:23 -07002211 } while (addr += PAGE_SIZE, addr != end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002212
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08002213 arch_leave_lazy_mmu_mode();
2214
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002215 if (mm != &init_mm)
2216 pte_unmap_unlock(pte-1, ptl);
2217 return err;
2218}
2219
2220static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
2221 unsigned long addr, unsigned long end,
2222 pte_fn_t fn, void *data)
2223{
2224 pmd_t *pmd;
2225 unsigned long next;
2226 int err;
2227
Andi Kleenceb86872008-07-23 21:27:50 -07002228 BUG_ON(pud_huge(*pud));
2229
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002230 pmd = pmd_alloc(mm, pud, addr);
2231 if (!pmd)
2232 return -ENOMEM;
2233 do {
2234 next = pmd_addr_end(addr, end);
2235 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
2236 if (err)
2237 break;
2238 } while (pmd++, addr = next, addr != end);
2239 return err;
2240}
2241
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03002242static int apply_to_pud_range(struct mm_struct *mm, p4d_t *p4d,
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002243 unsigned long addr, unsigned long end,
2244 pte_fn_t fn, void *data)
2245{
2246 pud_t *pud;
2247 unsigned long next;
2248 int err;
2249
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03002250 pud = pud_alloc(mm, p4d, addr);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002251 if (!pud)
2252 return -ENOMEM;
2253 do {
2254 next = pud_addr_end(addr, end);
2255 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
2256 if (err)
2257 break;
2258 } while (pud++, addr = next, addr != end);
2259 return err;
2260}
2261
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03002262static int apply_to_p4d_range(struct mm_struct *mm, pgd_t *pgd,
2263 unsigned long addr, unsigned long end,
2264 pte_fn_t fn, void *data)
2265{
2266 p4d_t *p4d;
2267 unsigned long next;
2268 int err;
2269
2270 p4d = p4d_alloc(mm, pgd, addr);
2271 if (!p4d)
2272 return -ENOMEM;
2273 do {
2274 next = p4d_addr_end(addr, end);
2275 err = apply_to_pud_range(mm, p4d, addr, next, fn, data);
2276 if (err)
2277 break;
2278 } while (p4d++, addr = next, addr != end);
2279 return err;
2280}
2281
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002282/*
2283 * Scan a region of virtual memory, filling in page tables as necessary
2284 * and calling a provided function on each leaf page table.
2285 */
2286int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
2287 unsigned long size, pte_fn_t fn, void *data)
2288{
2289 pgd_t *pgd;
2290 unsigned long next;
Jeremy Fitzhardinge57250a52010-08-09 17:19:52 -07002291 unsigned long end = addr + size;
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002292 int err;
2293
Mika Penttilä9cb65bc32016-03-15 14:56:45 -07002294 if (WARN_ON(addr >= end))
2295 return -EINVAL;
2296
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002297 pgd = pgd_offset(mm, addr);
2298 do {
2299 next = pgd_addr_end(addr, end);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03002300 err = apply_to_p4d_range(mm, pgd, addr, next, fn, data);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002301 if (err)
2302 break;
2303 } while (pgd++, addr = next, addr != end);
Jeremy Fitzhardinge57250a52010-08-09 17:19:52 -07002304
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002305 return err;
2306}
2307EXPORT_SYMBOL_GPL(apply_to_page_range);
2308
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309/*
Kirill A. Shutemov9b4bdd22015-02-10 14:09:51 -08002310 * handle_pte_fault chooses page fault handler according to an entry which was
2311 * read non-atomically. Before making any commitment, on those architectures
2312 * or configurations (e.g. i386 with PAE) which might give a mix of unmatched
2313 * parts, do_swap_page must check under lock before unmapping the pte and
2314 * proceeding (but do_wp_page is only called after already making such a check;
Ryota Ozakia335b2e2011-02-10 13:56:28 +09002315 * and do_anonymous_page can safely check later on).
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002316 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002317static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002318 pte_t *page_table, pte_t orig_pte)
2319{
2320 int same = 1;
2321#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
2322 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002323 spinlock_t *ptl = pte_lockptr(mm, pmd);
2324 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002325 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002326 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002327 }
2328#endif
2329 pte_unmap(page_table);
2330 return same;
2331}
2332
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00002333static 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 -08002334{
Dan Williams0abdd7a2014-01-21 15:48:12 -08002335 debug_dma_assert_idle(src);
2336
Linus Torvalds6aab3412005-11-28 14:34:23 -08002337 /*
2338 * If the source page was a PFN mapping, we don't have
2339 * a "struct page" for it. We do a best-effort copy by
2340 * just copying from the original user address. If that
2341 * fails, we just zero-fill it. Live with it.
2342 */
2343 if (unlikely(!src)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002344 void *kaddr = kmap_atomic(dst);
Linus Torvalds5d2a2dbbc2005-11-29 14:07:55 -08002345 void __user *uaddr = (void __user *)(va & PAGE_MASK);
2346
2347 /*
2348 * This really shouldn't fail, because the page is there
2349 * in the page tables. But it might just be unreadable,
2350 * in which case we just give up and fill the result with
2351 * zeroes.
2352 */
2353 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Jan Beulich3ecb01d2010-10-26 14:22:27 -07002354 clear_page(kaddr);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002355 kunmap_atomic(kaddr);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07002356 flush_dcache_page(dst);
Nick Piggin0ed361d2008-02-04 22:29:34 -08002357 } else
2358 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08002359}
2360
Michal Hockoc20cd452016-01-14 15:20:12 -08002361static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma)
2362{
2363 struct file *vm_file = vma->vm_file;
2364
2365 if (vm_file)
2366 return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO;
2367
2368 /*
2369 * Special mappings (e.g. VDSO) do not have any file so fake
2370 * a default GFP_KERNEL for them.
2371 */
2372 return GFP_KERNEL;
2373}
2374
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375/*
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07002376 * Notify the address space that the page is about to become writable so that
2377 * it can prohibit this or wait for the page to get into an appropriate state.
2378 *
2379 * We do this without the lock held, so that it can sleep if it needs to.
2380 */
Jan Kara38b8cb72016-12-14 15:07:30 -08002381static int do_page_mkwrite(struct vm_fault *vmf)
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07002382{
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07002383 int ret;
Jan Kara38b8cb72016-12-14 15:07:30 -08002384 struct page *page = vmf->page;
2385 unsigned int old_flags = vmf->flags;
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07002386
Jan Kara38b8cb72016-12-14 15:07:30 -08002387 vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07002388
Dave Jiang11bac802017-02-24 14:56:41 -08002389 ret = vmf->vma->vm_ops->page_mkwrite(vmf);
Jan Kara38b8cb72016-12-14 15:07:30 -08002390 /* Restore original flags so that caller is not surprised */
2391 vmf->flags = old_flags;
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07002392 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2393 return ret;
2394 if (unlikely(!(ret & VM_FAULT_LOCKED))) {
2395 lock_page(page);
2396 if (!page->mapping) {
2397 unlock_page(page);
2398 return 0; /* retry */
2399 }
2400 ret |= VM_FAULT_LOCKED;
2401 } else
2402 VM_BUG_ON_PAGE(!PageLocked(page), page);
2403 return ret;
2404}
2405
2406/*
Jan Kara97ba0c22016-12-14 15:07:27 -08002407 * Handle dirtying of a page in shared file mapping on a write fault.
2408 *
2409 * The function expects the page to be locked and unlocks it.
2410 */
2411static void fault_dirty_shared_page(struct vm_area_struct *vma,
2412 struct page *page)
2413{
2414 struct address_space *mapping;
2415 bool dirtied;
2416 bool page_mkwrite = vma->vm_ops && vma->vm_ops->page_mkwrite;
2417
2418 dirtied = set_page_dirty(page);
2419 VM_BUG_ON_PAGE(PageAnon(page), page);
2420 /*
2421 * Take a local copy of the address_space - page.mapping may be zeroed
2422 * by truncate after unlock_page(). The address_space itself remains
2423 * pinned by vma->vm_file's reference. We rely on unlock_page()'s
2424 * release semantics to prevent the compiler from undoing this copying.
2425 */
2426 mapping = page_rmapping(page);
2427 unlock_page(page);
2428
2429 if ((dirtied || page_mkwrite) && mapping) {
2430 /*
2431 * Some device drivers do not set page.mapping
2432 * but still dirty their pages
2433 */
2434 balance_dirty_pages_ratelimited(mapping);
2435 }
2436
2437 if (!page_mkwrite)
2438 file_update_time(vma->vm_file);
2439}
2440
2441/*
Shachar Raindel4e047f82015-04-14 15:46:25 -07002442 * Handle write page faults for pages that can be reused in the current vma
2443 *
2444 * This can happen either due to the mapping being with the VM_SHARED flag,
2445 * or due to us being the last reference standing to the page. In either
2446 * case, all we need to do here is to mark the page as writable and update
2447 * any related book-keeping.
2448 */
Jan Kara997dd982016-12-14 15:07:36 -08002449static inline void wp_page_reuse(struct vm_fault *vmf)
Jan Kara82b0f8c2016-12-14 15:06:58 -08002450 __releases(vmf->ptl)
Shachar Raindel4e047f82015-04-14 15:46:25 -07002451{
Jan Kara82b0f8c2016-12-14 15:06:58 -08002452 struct vm_area_struct *vma = vmf->vma;
Jan Karaa41b70d2016-12-14 15:07:33 -08002453 struct page *page = vmf->page;
Shachar Raindel4e047f82015-04-14 15:46:25 -07002454 pte_t entry;
2455 /*
2456 * Clear the pages cpupid information as the existing
2457 * information potentially belongs to a now completely
2458 * unrelated process.
2459 */
2460 if (page)
2461 page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1);
2462
Jan Kara29943022016-12-14 15:07:16 -08002463 flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte));
2464 entry = pte_mkyoung(vmf->orig_pte);
Shachar Raindel4e047f82015-04-14 15:46:25 -07002465 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Jan Kara82b0f8c2016-12-14 15:06:58 -08002466 if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1))
2467 update_mmu_cache(vma, vmf->address, vmf->pte);
2468 pte_unmap_unlock(vmf->pte, vmf->ptl);
Shachar Raindel4e047f82015-04-14 15:46:25 -07002469}
2470
2471/*
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002472 * Handle the case of a page which we actually need to copy to a new page.
2473 *
2474 * Called with mmap_sem locked and the old page referenced, but
2475 * without the ptl held.
2476 *
2477 * High level logic flow:
2478 *
2479 * - Allocate a page, copy the content of the old page to the new one.
2480 * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc.
2481 * - Take the PTL. If the pte changed, bail out and release the allocated page
2482 * - If the pte is still the way we remember it, update the page table and all
2483 * relevant references. This includes dropping the reference the page-table
2484 * held to the old page, as well as updating the rmap.
2485 * - In any case, unlock the PTL and drop the reference we took to the old page.
2486 */
Jan Karaa41b70d2016-12-14 15:07:33 -08002487static int wp_page_copy(struct vm_fault *vmf)
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002488{
Jan Kara82b0f8c2016-12-14 15:06:58 -08002489 struct vm_area_struct *vma = vmf->vma;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002490 struct mm_struct *mm = vma->vm_mm;
Jan Karaa41b70d2016-12-14 15:07:33 -08002491 struct page *old_page = vmf->page;
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002492 struct page *new_page = NULL;
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002493 pte_t entry;
2494 int page_copied = 0;
Jan Kara82b0f8c2016-12-14 15:06:58 -08002495 const unsigned long mmun_start = vmf->address & PAGE_MASK;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002496 const unsigned long mmun_end = mmun_start + PAGE_SIZE;
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002497 struct mem_cgroup *memcg;
2498
2499 if (unlikely(anon_vma_prepare(vma)))
2500 goto oom;
2501
Jan Kara29943022016-12-14 15:07:16 -08002502 if (is_zero_pfn(pte_pfn(vmf->orig_pte))) {
Jan Kara82b0f8c2016-12-14 15:06:58 -08002503 new_page = alloc_zeroed_user_highpage_movable(vma,
2504 vmf->address);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002505 if (!new_page)
2506 goto oom;
2507 } else {
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002508 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma,
Jan Kara82b0f8c2016-12-14 15:06:58 -08002509 vmf->address);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002510 if (!new_page)
2511 goto oom;
Jan Kara82b0f8c2016-12-14 15:06:58 -08002512 cow_user_page(new_page, old_page, vmf->address, vma);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002513 }
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002514
Tejun Heo2cf85582018-07-03 11:14:56 -04002515 if (mem_cgroup_try_charge_delay(new_page, mm, GFP_KERNEL, &memcg, false))
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002516 goto oom_free_new;
2517
Mel Gormaneb3c24f2015-06-24 16:57:27 -07002518 __SetPageUptodate(new_page);
2519
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002520 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
2521
2522 /*
2523 * Re-check the pte - we dropped the lock
2524 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08002525 vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl);
Jan Kara29943022016-12-14 15:07:16 -08002526 if (likely(pte_same(*vmf->pte, vmf->orig_pte))) {
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002527 if (old_page) {
2528 if (!PageAnon(old_page)) {
Jerome Marchandeca56ff2016-01-14 15:19:26 -08002529 dec_mm_counter_fast(mm,
2530 mm_counter_file(old_page));
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002531 inc_mm_counter_fast(mm, MM_ANONPAGES);
2532 }
2533 } else {
2534 inc_mm_counter_fast(mm, MM_ANONPAGES);
2535 }
Jan Kara29943022016-12-14 15:07:16 -08002536 flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte));
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002537 entry = mk_pte(new_page, vma->vm_page_prot);
2538 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2539 /*
2540 * Clear the pte entry and flush it first, before updating the
2541 * pte with the new entry. This will avoid a race condition
2542 * seen in the presence of one thread doing SMC and another
2543 * thread doing COW.
2544 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08002545 ptep_clear_flush_notify(vma, vmf->address, vmf->pte);
2546 page_add_new_anon_rmap(new_page, vma, vmf->address, false);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08002547 mem_cgroup_commit_charge(new_page, memcg, false, false);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002548 lru_cache_add_active_or_unevictable(new_page, vma);
2549 /*
2550 * We call the notify macro here because, when using secondary
2551 * mmu page tables (such as kvm shadow page tables), we want the
2552 * new page to be mapped directly into the secondary page table.
2553 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08002554 set_pte_at_notify(mm, vmf->address, vmf->pte, entry);
2555 update_mmu_cache(vma, vmf->address, vmf->pte);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002556 if (old_page) {
2557 /*
2558 * Only after switching the pte to the new page may
2559 * we remove the mapcount here. Otherwise another
2560 * process may come and find the rmap count decremented
2561 * before the pte is switched to the new page, and
2562 * "reuse" the old page writing into it while our pte
2563 * here still points into it and can be read by other
2564 * threads.
2565 *
2566 * The critical issue is to order this
2567 * page_remove_rmap with the ptp_clear_flush above.
2568 * Those stores are ordered by (if nothing else,)
2569 * the barrier present in the atomic_add_negative
2570 * in page_remove_rmap.
2571 *
2572 * Then the TLB flush in ptep_clear_flush ensures that
2573 * no process can access the old page before the
2574 * decremented mapcount is visible. And the old page
2575 * cannot be reused until after the decremented
2576 * mapcount is visible. So transitively, TLBs to
2577 * old page will be flushed before it can be reused.
2578 */
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08002579 page_remove_rmap(old_page, false);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002580 }
2581
2582 /* Free the old page.. */
2583 new_page = old_page;
2584 page_copied = 1;
2585 } else {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08002586 mem_cgroup_cancel_charge(new_page, memcg, false);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002587 }
2588
2589 if (new_page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002590 put_page(new_page);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002591
Jan Kara82b0f8c2016-12-14 15:06:58 -08002592 pte_unmap_unlock(vmf->pte, vmf->ptl);
Jérôme Glisse4645b9f2017-11-15 17:34:11 -08002593 /*
2594 * No need to double call mmu_notifier->invalidate_range() callback as
2595 * the above ptep_clear_flush_notify() did already call it.
2596 */
2597 mmu_notifier_invalidate_range_only_end(mm, mmun_start, mmun_end);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002598 if (old_page) {
2599 /*
2600 * Don't let another task, with possibly unlocked vma,
2601 * keep the mlocked page.
2602 */
2603 if (page_copied && (vma->vm_flags & VM_LOCKED)) {
2604 lock_page(old_page); /* LRU manipulation */
Kirill A. Shutemove90309c2016-01-15 16:54:33 -08002605 if (PageMlocked(old_page))
2606 munlock_vma_page(old_page);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002607 unlock_page(old_page);
2608 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002609 put_page(old_page);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002610 }
2611 return page_copied ? VM_FAULT_WRITE : 0;
2612oom_free_new:
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002613 put_page(new_page);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002614oom:
2615 if (old_page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002616 put_page(old_page);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002617 return VM_FAULT_OOM;
2618}
2619
Jan Kara66a61972016-12-14 15:07:39 -08002620/**
2621 * finish_mkwrite_fault - finish page fault for a shared mapping, making PTE
2622 * writeable once the page is prepared
2623 *
2624 * @vmf: structure describing the fault
2625 *
2626 * This function handles all that is needed to finish a write page fault in a
2627 * shared mapping due to PTE being read-only once the mapped page is prepared.
2628 * It handles locking of PTE and modifying it. The function returns
2629 * VM_FAULT_WRITE on success, 0 when PTE got changed before we acquired PTE
2630 * lock.
2631 *
2632 * The function expects the page to be locked or other protection against
2633 * concurrent faults / writeback (such as DAX radix tree locks).
2634 */
2635int finish_mkwrite_fault(struct vm_fault *vmf)
2636{
2637 WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED));
2638 vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address,
2639 &vmf->ptl);
2640 /*
2641 * We might have raced with another page fault while we released the
2642 * pte_offset_map_lock.
2643 */
2644 if (!pte_same(*vmf->pte, vmf->orig_pte)) {
2645 pte_unmap_unlock(vmf->pte, vmf->ptl);
Jan Karaa19e2552016-12-14 15:07:42 -08002646 return VM_FAULT_NOPAGE;
Jan Kara66a61972016-12-14 15:07:39 -08002647 }
2648 wp_page_reuse(vmf);
Jan Karaa19e2552016-12-14 15:07:42 -08002649 return 0;
Jan Kara66a61972016-12-14 15:07:39 -08002650}
2651
Boaz Harroshdd906182015-04-15 16:15:11 -07002652/*
2653 * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED
2654 * mapping
2655 */
Jan Kara29943022016-12-14 15:07:16 -08002656static int wp_pfn_shared(struct vm_fault *vmf)
Boaz Harroshdd906182015-04-15 16:15:11 -07002657{
Jan Kara82b0f8c2016-12-14 15:06:58 -08002658 struct vm_area_struct *vma = vmf->vma;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002659
Boaz Harroshdd906182015-04-15 16:15:11 -07002660 if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) {
Boaz Harroshdd906182015-04-15 16:15:11 -07002661 int ret;
2662
Jan Kara82b0f8c2016-12-14 15:06:58 -08002663 pte_unmap_unlock(vmf->pte, vmf->ptl);
Jan Karafe822212016-12-14 15:07:13 -08002664 vmf->flags |= FAULT_FLAG_MKWRITE;
Dave Jiang11bac802017-02-24 14:56:41 -08002665 ret = vma->vm_ops->pfn_mkwrite(vmf);
Jan Kara2f89dc12016-12-14 15:07:50 -08002666 if (ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))
Boaz Harroshdd906182015-04-15 16:15:11 -07002667 return ret;
Jan Kara66a61972016-12-14 15:07:39 -08002668 return finish_mkwrite_fault(vmf);
Boaz Harroshdd906182015-04-15 16:15:11 -07002669 }
Jan Kara997dd982016-12-14 15:07:36 -08002670 wp_page_reuse(vmf);
2671 return VM_FAULT_WRITE;
Boaz Harroshdd906182015-04-15 16:15:11 -07002672}
2673
Jan Karaa41b70d2016-12-14 15:07:33 -08002674static int wp_page_shared(struct vm_fault *vmf)
Jan Kara82b0f8c2016-12-14 15:06:58 -08002675 __releases(vmf->ptl)
Shachar Raindel93e478d2015-04-14 15:46:35 -07002676{
Jan Kara82b0f8c2016-12-14 15:06:58 -08002677 struct vm_area_struct *vma = vmf->vma;
Shachar Raindel93e478d2015-04-14 15:46:35 -07002678
Jan Karaa41b70d2016-12-14 15:07:33 -08002679 get_page(vmf->page);
Shachar Raindel93e478d2015-04-14 15:46:35 -07002680
Shachar Raindel93e478d2015-04-14 15:46:35 -07002681 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
2682 int tmp;
2683
Jan Kara82b0f8c2016-12-14 15:06:58 -08002684 pte_unmap_unlock(vmf->pte, vmf->ptl);
Jan Kara38b8cb72016-12-14 15:07:30 -08002685 tmp = do_page_mkwrite(vmf);
Shachar Raindel93e478d2015-04-14 15:46:35 -07002686 if (unlikely(!tmp || (tmp &
2687 (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
Jan Karaa41b70d2016-12-14 15:07:33 -08002688 put_page(vmf->page);
Shachar Raindel93e478d2015-04-14 15:46:35 -07002689 return tmp;
2690 }
Jan Kara66a61972016-12-14 15:07:39 -08002691 tmp = finish_mkwrite_fault(vmf);
Jan Karaa19e2552016-12-14 15:07:42 -08002692 if (unlikely(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
Jan Karaa41b70d2016-12-14 15:07:33 -08002693 unlock_page(vmf->page);
Jan Karaa41b70d2016-12-14 15:07:33 -08002694 put_page(vmf->page);
Jan Kara66a61972016-12-14 15:07:39 -08002695 return tmp;
Shachar Raindel93e478d2015-04-14 15:46:35 -07002696 }
Jan Kara66a61972016-12-14 15:07:39 -08002697 } else {
2698 wp_page_reuse(vmf);
Jan Kara997dd982016-12-14 15:07:36 -08002699 lock_page(vmf->page);
Shachar Raindel93e478d2015-04-14 15:46:35 -07002700 }
Jan Kara997dd982016-12-14 15:07:36 -08002701 fault_dirty_shared_page(vma, vmf->page);
2702 put_page(vmf->page);
Shachar Raindel93e478d2015-04-14 15:46:35 -07002703
Jan Kara997dd982016-12-14 15:07:36 -08002704 return VM_FAULT_WRITE;
Shachar Raindel93e478d2015-04-14 15:46:35 -07002705}
2706
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002707/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708 * This routine handles present pages, when users try to write
2709 * to a shared page. It is done by copying the page to a new address
2710 * and decrementing the shared-page counter for the old page.
2711 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712 * Note that this routine assumes that the protection checks have been
2713 * done by the caller (the low-level page fault routine in most cases).
2714 * Thus we can safely just mark it writable once we've done any necessary
2715 * COW.
2716 *
2717 * We also mark the page dirty at this point even though the page will
2718 * change only once the write actually happens. This avoids a few races,
2719 * and potentially makes it more efficient.
2720 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002721 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2722 * but allow concurrent faults), with pte both mapped and locked.
2723 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724 */
Jan Kara29943022016-12-14 15:07:16 -08002725static int do_wp_page(struct vm_fault *vmf)
Jan Kara82b0f8c2016-12-14 15:06:58 -08002726 __releases(vmf->ptl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727{
Jan Kara82b0f8c2016-12-14 15:06:58 -08002728 struct vm_area_struct *vma = vmf->vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729
Jan Karaa41b70d2016-12-14 15:07:33 -08002730 vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte);
2731 if (!vmf->page) {
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002732 /*
Peter Feiner64e455072014-10-13 15:55:46 -07002733 * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a
2734 * VM_PFNMAP VMA.
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002735 *
2736 * We should not cow pages in a shared writeable mapping.
Boaz Harroshdd906182015-04-15 16:15:11 -07002737 * Just mark the pages writable and/or call ops->pfn_mkwrite.
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002738 */
2739 if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
2740 (VM_WRITE|VM_SHARED))
Jan Kara29943022016-12-14 15:07:16 -08002741 return wp_pfn_shared(vmf);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002742
Jan Kara82b0f8c2016-12-14 15:06:58 -08002743 pte_unmap_unlock(vmf->pte, vmf->ptl);
Jan Karaa41b70d2016-12-14 15:07:33 -08002744 return wp_page_copy(vmf);
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002745 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002747 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002748 * Take out anonymous pages first, anonymous shared vmas are
2749 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002750 */
Jan Karaa41b70d2016-12-14 15:07:33 -08002751 if (PageAnon(vmf->page) && !PageKsm(vmf->page)) {
Huang Yingba3c4ce2017-09-06 16:22:19 -07002752 int total_map_swapcount;
Jan Karaa41b70d2016-12-14 15:07:33 -08002753 if (!trylock_page(vmf->page)) {
2754 get_page(vmf->page);
Jan Kara82b0f8c2016-12-14 15:06:58 -08002755 pte_unmap_unlock(vmf->pte, vmf->ptl);
Jan Karaa41b70d2016-12-14 15:07:33 -08002756 lock_page(vmf->page);
Jan Kara82b0f8c2016-12-14 15:06:58 -08002757 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
2758 vmf->address, &vmf->ptl);
Jan Kara29943022016-12-14 15:07:16 -08002759 if (!pte_same(*vmf->pte, vmf->orig_pte)) {
Jan Karaa41b70d2016-12-14 15:07:33 -08002760 unlock_page(vmf->page);
Jan Kara82b0f8c2016-12-14 15:06:58 -08002761 pte_unmap_unlock(vmf->pte, vmf->ptl);
Jan Karaa41b70d2016-12-14 15:07:33 -08002762 put_page(vmf->page);
Shachar Raindel28766802015-04-14 15:46:29 -07002763 return 0;
Hugh Dickinsab967d82009-01-06 14:39:33 -08002764 }
Jan Karaa41b70d2016-12-14 15:07:33 -08002765 put_page(vmf->page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002766 }
Huang Yingba3c4ce2017-09-06 16:22:19 -07002767 if (reuse_swap_page(vmf->page, &total_map_swapcount)) {
2768 if (total_map_swapcount == 1) {
Andrea Arcangeli6d0a07e2016-05-12 15:42:25 -07002769 /*
2770 * The page is all ours. Move it to
2771 * our anon_vma so the rmap code will
2772 * not search our parent or siblings.
2773 * Protected against the rmap code by
2774 * the page lock.
2775 */
Jan Karaa41b70d2016-12-14 15:07:33 -08002776 page_move_anon_rmap(vmf->page, vma);
Andrea Arcangeli6d0a07e2016-05-12 15:42:25 -07002777 }
Jan Karaa41b70d2016-12-14 15:07:33 -08002778 unlock_page(vmf->page);
Jan Kara997dd982016-12-14 15:07:36 -08002779 wp_page_reuse(vmf);
2780 return VM_FAULT_WRITE;
Michel Lespinasseb009c022011-01-13 15:46:07 -08002781 }
Jan Karaa41b70d2016-12-14 15:07:33 -08002782 unlock_page(vmf->page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002783 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002784 (VM_WRITE|VM_SHARED))) {
Jan Karaa41b70d2016-12-14 15:07:33 -08002785 return wp_page_shared(vmf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787
2788 /*
2789 * Ok, we need to copy. Oh, well..
2790 */
Jan Karaa41b70d2016-12-14 15:07:33 -08002791 get_page(vmf->page);
Shachar Raindel28766802015-04-14 15:46:29 -07002792
Jan Kara82b0f8c2016-12-14 15:06:58 -08002793 pte_unmap_unlock(vmf->pte, vmf->ptl);
Jan Karaa41b70d2016-12-14 15:07:33 -08002794 return wp_page_copy(vmf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795}
2796
Peter Zijlstra97a89412011-05-24 17:12:04 -07002797static void unmap_mapping_range_vma(struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798 unsigned long start_addr, unsigned long end_addr,
2799 struct zap_details *details)
2800{
Al Virof5cc4ee2012-03-05 14:14:20 -05002801 zap_page_range_single(vma, start_addr, end_addr - start_addr, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802}
2803
Davidlohr Buesof808c132017-09-08 16:15:08 -07002804static inline void unmap_mapping_range_tree(struct rb_root_cached *root,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805 struct zap_details *details)
2806{
2807 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808 pgoff_t vba, vea, zba, zea;
2809
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07002810 vma_interval_tree_foreach(vma, root,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 details->first_index, details->last_index) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812
2813 vba = vma->vm_pgoff;
Libind6e93212013-07-03 15:01:26 -07002814 vea = vba + vma_pages(vma) - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 zba = details->first_index;
2816 if (zba < vba)
2817 zba = vba;
2818 zea = details->last_index;
2819 if (zea > vea)
2820 zea = vea;
2821
Peter Zijlstra97a89412011-05-24 17:12:04 -07002822 unmap_mapping_range_vma(vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
2824 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
Peter Zijlstra97a89412011-05-24 17:12:04 -07002825 details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826 }
2827}
2828
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829/**
Matthew Wilcox977fbdc2018-01-31 16:17:36 -08002830 * unmap_mapping_pages() - Unmap pages from processes.
2831 * @mapping: The address space containing pages to be unmapped.
2832 * @start: Index of first page to be unmapped.
2833 * @nr: Number of pages to be unmapped. 0 to unmap to end of file.
2834 * @even_cows: Whether to unmap even private COWed pages.
2835 *
2836 * Unmap the pages in this address space from any userspace process which
2837 * has them mmaped. Generally, you want to remove COWed pages as well when
2838 * a file is being truncated, but not when invalidating pages from the page
2839 * cache.
2840 */
2841void unmap_mapping_pages(struct address_space *mapping, pgoff_t start,
2842 pgoff_t nr, bool even_cows)
2843{
2844 struct zap_details details = { };
2845
2846 details.check_mapping = even_cows ? NULL : mapping;
2847 details.first_index = start;
2848 details.last_index = start + nr - 1;
2849 if (details.last_index < details.first_index)
2850 details.last_index = ULONG_MAX;
2851
2852 i_mmap_lock_write(mapping);
2853 if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)))
2854 unmap_mapping_range_tree(&mapping->i_mmap, &details);
2855 i_mmap_unlock_write(mapping);
2856}
2857
2858/**
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08002859 * unmap_mapping_range - unmap the portion of all mmaps in the specified
Matthew Wilcox977fbdc2018-01-31 16:17:36 -08002860 * address_space corresponding to the specified byte range in the underlying
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08002861 * file.
2862 *
Martin Waitz3d410882005-06-23 22:05:21 -07002863 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864 * @holebegin: byte in first page to unmap, relative to the start of
2865 * the underlying file. This will be rounded down to a PAGE_SIZE
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +10002866 * boundary. Note that this is different from truncate_pagecache(), which
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 * must keep the partial page. In contrast, we must get rid of
2868 * partial pages.
2869 * @holelen: size of prospective hole in bytes. This will be rounded
2870 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2871 * end of the file.
2872 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2873 * but 0 when invalidating pagecache, don't throw away private data.
2874 */
2875void unmap_mapping_range(struct address_space *mapping,
2876 loff_t const holebegin, loff_t const holelen, int even_cows)
2877{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 pgoff_t hba = holebegin >> PAGE_SHIFT;
2879 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2880
2881 /* Check for overflow. */
2882 if (sizeof(holelen) > sizeof(hlen)) {
2883 long long holeend =
2884 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2885 if (holeend & ~(long long)ULONG_MAX)
2886 hlen = ULONG_MAX - hba + 1;
2887 }
2888
Matthew Wilcox977fbdc2018-01-31 16:17:36 -08002889 unmap_mapping_pages(mapping, hba, hlen, even_cows);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890}
2891EXPORT_SYMBOL(unmap_mapping_range);
2892
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002894 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2895 * but allow concurrent faults), and pte mapped but not yet locked.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07002896 * We return with pte unmapped and unlocked.
2897 *
2898 * We return with the mmap_sem locked or unlocked in the same cases
2899 * as does filemap_fault().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002900 */
Jan Kara29943022016-12-14 15:07:16 -08002901int do_swap_page(struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902{
Jan Kara82b0f8c2016-12-14 15:06:58 -08002903 struct vm_area_struct *vma = vmf->vma;
Minchan Kimeaf649eb2018-04-05 16:23:39 -07002904 struct page *page = NULL, *swapcache;
Johannes Weiner00501b52014-08-08 14:19:20 -07002905 struct mem_cgroup *memcg;
Hugh Dickins65500d22005-10-29 18:15:59 -07002906 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 pte_t pte;
Michel Lespinassed065bd82010-10-26 14:21:57 -07002908 int locked;
Rik van Rielad8c2ee2010-08-09 17:19:48 -07002909 int exclusive = 0;
Nick Piggin83c54072007-07-19 01:47:05 -07002910 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911
Minchan Kimeaf649eb2018-04-05 16:23:39 -07002912 if (!pte_unmap_same(vma->vm_mm, vmf->pmd, vmf->pte, vmf->orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002913 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002914
Jan Kara29943022016-12-14 15:07:16 -08002915 entry = pte_to_swp_entry(vmf->orig_pte);
Andi Kleend1737fd2009-09-16 11:50:06 +02002916 if (unlikely(non_swap_entry(entry))) {
2917 if (is_migration_entry(entry)) {
Jan Kara82b0f8c2016-12-14 15:06:58 -08002918 migration_entry_wait(vma->vm_mm, vmf->pmd,
2919 vmf->address);
Jérôme Glisse5042db42017-09-08 16:11:43 -07002920 } else if (is_device_private_entry(entry)) {
2921 /*
2922 * For un-addressable device memory we call the pgmap
2923 * fault handler callback. The callback must migrate
2924 * the page back to some CPU accessible page.
2925 */
2926 ret = device_private_entry_fault(vma, vmf->address, entry,
2927 vmf->flags, vmf->pmd);
Andi Kleend1737fd2009-09-16 11:50:06 +02002928 } else if (is_hwpoison_entry(entry)) {
2929 ret = VM_FAULT_HWPOISON;
2930 } else {
Jan Kara29943022016-12-14 15:07:16 -08002931 print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL);
Hugh Dickinsd99be1a2009-12-14 17:59:04 -08002932 ret = VM_FAULT_SIGBUS;
Andi Kleend1737fd2009-09-16 11:50:06 +02002933 }
Christoph Lameter06972122006-06-23 02:03:35 -07002934 goto out;
2935 }
Minchan Kim0bcac062017-11-15 17:33:07 -08002936
2937
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002938 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Minchan Kimeaf649eb2018-04-05 16:23:39 -07002939 page = lookup_swap_cache(entry, vma, vmf->address);
2940 swapcache = page;
Minchan Kimf8020772018-01-18 16:33:50 -08002941
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942 if (!page) {
Minchan Kim0bcac062017-11-15 17:33:07 -08002943 struct swap_info_struct *si = swp_swap_info(entry);
2944
Minchan Kimaa8d22a2017-11-15 17:33:11 -08002945 if (si->flags & SWP_SYNCHRONOUS_IO &&
2946 __swap_count(si, entry) == 1) {
Minchan Kim0bcac062017-11-15 17:33:07 -08002947 /* skip swapcache */
Minchan Kime9e9b7e2018-04-05 16:23:42 -07002948 page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma,
2949 vmf->address);
Minchan Kim0bcac062017-11-15 17:33:07 -08002950 if (page) {
2951 __SetPageLocked(page);
2952 __SetPageSwapBacked(page);
2953 set_page_private(page, entry.val);
2954 lru_cache_add_anon(page);
2955 swap_readpage(page, true);
2956 }
Minchan Kimaa8d22a2017-11-15 17:33:11 -08002957 } else {
Minchan Kime9e9b7e2018-04-05 16:23:42 -07002958 page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE,
2959 vmf);
Minchan Kimaa8d22a2017-11-15 17:33:11 -08002960 swapcache = page;
Minchan Kim0bcac062017-11-15 17:33:07 -08002961 }
2962
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963 if (!page) {
2964 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002965 * Back out if somebody else faulted in this pte
2966 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08002968 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
2969 vmf->address, &vmf->ptl);
Jan Kara29943022016-12-14 15:07:16 -08002970 if (likely(pte_same(*vmf->pte, vmf->orig_pte)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002971 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002972 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002973 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002974 }
2975
2976 /* Had to read the page from swap area: Major fault */
2977 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002978 count_vm_event(PGMAJFAULT);
Roman Gushchin22621852017-07-06 15:40:25 -07002979 count_memcg_event_mm(vma->vm_mm, PGMAJFAULT);
Andi Kleend1737fd2009-09-16 11:50:06 +02002980 } else if (PageHWPoison(page)) {
Wu Fengguang71f72522009-12-16 12:19:58 +01002981 /*
2982 * hwpoisoned dirty swapcache pages are kept for killing
2983 * owner processes (which may be unknown at hwpoison time)
2984 */
Andi Kleend1737fd2009-09-16 11:50:06 +02002985 ret = VM_FAULT_HWPOISON;
2986 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Andi Kleen4779cb32009-10-14 01:51:41 +02002987 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988 }
2989
Jan Kara82b0f8c2016-12-14 15:06:58 -08002990 locked = lock_page_or_retry(page, vma->vm_mm, vmf->flags);
Rik van Riele709ffd2012-05-29 15:06:18 -07002991
Balbir Singh20a10222007-11-14 17:00:33 -08002992 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Michel Lespinassed065bd82010-10-26 14:21:57 -07002993 if (!locked) {
2994 ret |= VM_FAULT_RETRY;
2995 goto out_release;
2996 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002998 /*
Hugh Dickins31c4a3d2010-09-19 19:40:22 -07002999 * Make sure try_to_free_swap or reuse_swap_page or swapoff did not
3000 * release the swapcache from under us. The page pin, and pte_same
3001 * test below, are not enough to exclude that. Even if it is still
3002 * swapcache, we need to check that the page's swap has not changed.
Andrea Arcangeli4969c112010-09-09 16:37:52 -07003003 */
Minchan Kim0bcac062017-11-15 17:33:07 -08003004 if (unlikely((!PageSwapCache(page) ||
3005 page_private(page) != entry.val)) && swapcache)
Andrea Arcangeli4969c112010-09-09 16:37:52 -07003006 goto out_page;
3007
Jan Kara82b0f8c2016-12-14 15:06:58 -08003008 page = ksm_might_need_to_copy(page, vma, vmf->address);
Hugh Dickinscbf86cf2013-02-22 16:35:08 -08003009 if (unlikely(!page)) {
3010 ret = VM_FAULT_OOM;
3011 page = swapcache;
Hugh Dickinscbf86cf2013-02-22 16:35:08 -08003012 goto out_page;
Hugh Dickins5ad64682009-12-14 17:59:24 -08003013 }
3014
Tejun Heo2cf85582018-07-03 11:14:56 -04003015 if (mem_cgroup_try_charge_delay(page, vma->vm_mm, GFP_KERNEL,
3016 &memcg, false)) {
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07003017 ret = VM_FAULT_OOM;
Johannes Weinerbc43f752009-04-30 15:08:08 -07003018 goto out_page;
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07003019 }
3020
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003022 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003023 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003024 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address,
3025 &vmf->ptl);
Jan Kara29943022016-12-14 15:07:16 -08003026 if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07003027 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07003028
3029 if (unlikely(!PageUptodate(page))) {
3030 ret = VM_FAULT_SIGBUS;
3031 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003032 }
3033
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003034 /*
3035 * The page isn't present yet, go ahead with the fault.
3036 *
3037 * Be careful about the sequence of operations here.
3038 * To get its accounting right, reuse_swap_page() must be called
3039 * while the page is counted on swap but not yet in mapcount i.e.
3040 * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
3041 * must be called after the swap_free(), or it will never succeed.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003042 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003044 inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES);
3045 dec_mm_counter_fast(vma->vm_mm, MM_SWAPENTS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046 pte = mk_pte(page, vma->vm_page_prot);
Jan Kara82b0f8c2016-12-14 15:06:58 -08003047 if ((vmf->flags & FAULT_FLAG_WRITE) && reuse_swap_page(page, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
Jan Kara82b0f8c2016-12-14 15:06:58 -08003049 vmf->flags &= ~FAULT_FLAG_WRITE;
Andrea Arcangeli9a5b4892010-08-09 17:19:49 -07003050 ret |= VM_FAULT_WRITE;
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08003051 exclusive = RMAP_EXCLUSIVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003052 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003053 flush_icache_page(vma, page);
Jan Kara29943022016-12-14 15:07:16 -08003054 if (pte_swp_soft_dirty(vmf->orig_pte))
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07003055 pte = pte_mksoft_dirty(pte);
Jan Kara82b0f8c2016-12-14 15:06:58 -08003056 set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte);
Khalid Azizca827d52018-02-21 10:15:44 -07003057 arch_do_swap_page(vma->vm_mm, vma, vmf->address, pte, vmf->orig_pte);
Jan Kara29943022016-12-14 15:07:16 -08003058 vmf->orig_pte = pte;
Minchan Kim0bcac062017-11-15 17:33:07 -08003059
3060 /* ksm created a completely new copy */
3061 if (unlikely(page != swapcache && swapcache)) {
Jan Kara82b0f8c2016-12-14 15:06:58 -08003062 page_add_new_anon_rmap(page, vma, vmf->address, false);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08003063 mem_cgroup_commit_charge(page, memcg, false, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07003064 lru_cache_add_active_or_unevictable(page, vma);
Minchan Kim0bcac062017-11-15 17:33:07 -08003065 } else {
3066 do_page_add_anon_rmap(page, vma, vmf->address, exclusive);
3067 mem_cgroup_commit_charge(page, memcg, true, false);
3068 activate_page(page);
Johannes Weiner00501b52014-08-08 14:19:20 -07003069 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003070
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07003071 swap_free(entry);
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08003072 if (mem_cgroup_swap_full(page) ||
3073 (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08003074 try_to_free_swap(page);
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07003075 unlock_page(page);
Minchan Kim0bcac062017-11-15 17:33:07 -08003076 if (page != swapcache && swapcache) {
Andrea Arcangeli4969c112010-09-09 16:37:52 -07003077 /*
3078 * Hold the lock to avoid the swap entry to be reused
3079 * until we take the PT lock for the pte_same() check
3080 * (to avoid false positives from pte_same). For
3081 * further safety release the lock after the swap_free
3082 * so that the swap count won't change under a
3083 * parallel locked swapcache.
3084 */
3085 unlock_page(swapcache);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003086 put_page(swapcache);
Andrea Arcangeli4969c112010-09-09 16:37:52 -07003087 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07003088
Jan Kara82b0f8c2016-12-14 15:06:58 -08003089 if (vmf->flags & FAULT_FLAG_WRITE) {
Jan Kara29943022016-12-14 15:07:16 -08003090 ret |= do_wp_page(vmf);
Hugh Dickins61469f12008-03-04 14:29:04 -08003091 if (ret & VM_FAULT_ERROR)
3092 ret &= VM_FAULT_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093 goto out;
3094 }
3095
3096 /* No need to invalidate - it was non-present before */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003097 update_mmu_cache(vma, vmf->address, vmf->pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07003098unlock:
Jan Kara82b0f8c2016-12-14 15:06:58 -08003099 pte_unmap_unlock(vmf->pte, vmf->ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100out:
3101 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07003102out_nomap:
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08003103 mem_cgroup_cancel_charge(page, memcg, false);
Jan Kara82b0f8c2016-12-14 15:06:58 -08003104 pte_unmap_unlock(vmf->pte, vmf->ptl);
Johannes Weinerbc43f752009-04-30 15:08:08 -07003105out_page:
Kirill Korotaevb8107482005-05-16 21:53:50 -07003106 unlock_page(page);
Andi Kleen4779cb32009-10-14 01:51:41 +02003107out_release:
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003108 put_page(page);
Minchan Kim0bcac062017-11-15 17:33:07 -08003109 if (page != swapcache && swapcache) {
Andrea Arcangeli4969c112010-09-09 16:37:52 -07003110 unlock_page(swapcache);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003111 put_page(swapcache);
Andrea Arcangeli4969c112010-09-09 16:37:52 -07003112 }
Hugh Dickins65500d22005-10-29 18:15:59 -07003113 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003114}
3115
3116/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003117 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3118 * but allow concurrent faults), and pte mapped but not yet locked.
3119 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003121static int do_anonymous_page(struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003122{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003123 struct vm_area_struct *vma = vmf->vma;
Johannes Weiner00501b52014-08-08 14:19:20 -07003124 struct mem_cgroup *memcg;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003125 struct page *page;
Michal Hocko6b31d592017-08-18 15:16:15 -07003126 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003127 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003128
Kirill A. Shutemov6b7339f2015-07-06 23:18:37 +03003129 /* File mapping without ->vm_ops ? */
3130 if (vma->vm_flags & VM_SHARED)
3131 return VM_FAULT_SIGBUS;
3132
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003133 /*
3134 * Use pte_alloc() instead of pte_alloc_map(). We can't run
3135 * pte_offset_map() on pmds where a huge pmd might be created
3136 * from a different thread.
3137 *
3138 * pte_alloc_map() is safe to use under down_write(mmap_sem) or when
3139 * parallel threads are excluded by other means.
3140 *
3141 * Here we only have down_read(mmap_sem).
3142 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003143 if (pte_alloc(vma->vm_mm, vmf->pmd, vmf->address))
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003144 return VM_FAULT_OOM;
3145
3146 /* See the comment in pte_alloc_one_map() */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003147 if (unlikely(pmd_trans_unstable(vmf->pmd)))
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003148 return 0;
3149
Linus Torvalds11ac5522010-08-14 11:44:56 -07003150 /* Use the zero-page for reads */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003151 if (!(vmf->flags & FAULT_FLAG_WRITE) &&
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003152 !mm_forbids_zeropage(vma->vm_mm)) {
Jan Kara82b0f8c2016-12-14 15:06:58 -08003153 entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address),
Hugh Dickins62eede62009-09-21 17:03:34 -07003154 vma->vm_page_prot));
Jan Kara82b0f8c2016-12-14 15:06:58 -08003155 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
3156 vmf->address, &vmf->ptl);
3157 if (!pte_none(*vmf->pte))
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07003158 goto unlock;
Michal Hocko6b31d592017-08-18 15:16:15 -07003159 ret = check_stable_address_space(vma->vm_mm);
3160 if (ret)
3161 goto unlock;
Andrea Arcangeli6b251fc2015-09-04 15:46:20 -07003162 /* Deliver the page fault to userland, check inside PT lock */
3163 if (userfaultfd_missing(vma)) {
Jan Kara82b0f8c2016-12-14 15:06:58 -08003164 pte_unmap_unlock(vmf->pte, vmf->ptl);
3165 return handle_userfault(vmf, VM_UFFD_MISSING);
Andrea Arcangeli6b251fc2015-09-04 15:46:20 -07003166 }
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07003167 goto setpte;
3168 }
3169
Nick Piggin557ed1f2007-10-16 01:24:40 -07003170 /* Allocate our own private page. */
Nick Piggin557ed1f2007-10-16 01:24:40 -07003171 if (unlikely(anon_vma_prepare(vma)))
3172 goto oom;
Jan Kara82b0f8c2016-12-14 15:06:58 -08003173 page = alloc_zeroed_user_highpage_movable(vma, vmf->address);
Nick Piggin557ed1f2007-10-16 01:24:40 -07003174 if (!page)
3175 goto oom;
Mel Gormaneb3c24f2015-06-24 16:57:27 -07003176
Tejun Heo2cf85582018-07-03 11:14:56 -04003177 if (mem_cgroup_try_charge_delay(page, vma->vm_mm, GFP_KERNEL, &memcg,
3178 false))
Mel Gormaneb3c24f2015-06-24 16:57:27 -07003179 goto oom_free_page;
3180
Minchan Kim52f37622013-04-29 15:08:15 -07003181 /*
3182 * The memory barrier inside __SetPageUptodate makes sure that
3183 * preceeding stores to the page contents become visible before
3184 * the set_pte_at() write.
3185 */
Nick Piggin0ed361d2008-02-04 22:29:34 -08003186 __SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003187
Nick Piggin557ed1f2007-10-16 01:24:40 -07003188 entry = mk_pte(page, vma->vm_page_prot);
Hugh Dickins1ac0cb52009-09-21 17:03:29 -07003189 if (vma->vm_flags & VM_WRITE)
3190 entry = pte_mkwrite(pte_mkdirty(entry));
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003191
Jan Kara82b0f8c2016-12-14 15:06:58 -08003192 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address,
3193 &vmf->ptl);
3194 if (!pte_none(*vmf->pte))
Nick Piggin557ed1f2007-10-16 01:24:40 -07003195 goto release;
Hugh Dickins9ba69292009-09-21 17:02:20 -07003196
Michal Hocko6b31d592017-08-18 15:16:15 -07003197 ret = check_stable_address_space(vma->vm_mm);
3198 if (ret)
3199 goto release;
3200
Andrea Arcangeli6b251fc2015-09-04 15:46:20 -07003201 /* Deliver the page fault to userland, check inside PT lock */
3202 if (userfaultfd_missing(vma)) {
Jan Kara82b0f8c2016-12-14 15:06:58 -08003203 pte_unmap_unlock(vmf->pte, vmf->ptl);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08003204 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003205 put_page(page);
Jan Kara82b0f8c2016-12-14 15:06:58 -08003206 return handle_userfault(vmf, VM_UFFD_MISSING);
Andrea Arcangeli6b251fc2015-09-04 15:46:20 -07003207 }
3208
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003209 inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES);
Jan Kara82b0f8c2016-12-14 15:06:58 -08003210 page_add_new_anon_rmap(page, vma, vmf->address, false);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08003211 mem_cgroup_commit_charge(page, memcg, false, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07003212 lru_cache_add_active_or_unevictable(page, vma);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07003213setpte:
Jan Kara82b0f8c2016-12-14 15:06:58 -08003214 set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003215
3216 /* No need to invalidate - it was non-present before */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003217 update_mmu_cache(vma, vmf->address, vmf->pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07003218unlock:
Jan Kara82b0f8c2016-12-14 15:06:58 -08003219 pte_unmap_unlock(vmf->pte, vmf->ptl);
Michal Hocko6b31d592017-08-18 15:16:15 -07003220 return ret;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003221release:
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08003222 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003223 put_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003224 goto unlock;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08003225oom_free_page:
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003226 put_page(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07003227oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003228 return VM_FAULT_OOM;
3229}
3230
Paul Cassella9a95f3c2014-08-06 16:07:24 -07003231/*
3232 * The mmap_sem must have been held on entry, and may have been
3233 * released depending on flags and vma->vm_ops->fault() return value.
3234 * See filemap_fault() and __lock_page_retry().
3235 */
Jan Kara936ca802016-12-14 15:07:10 -08003236static int __do_fault(struct vm_fault *vmf)
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07003237{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003238 struct vm_area_struct *vma = vmf->vma;
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07003239 int ret;
3240
Dave Jiang11bac802017-02-24 14:56:41 -08003241 ret = vma->vm_ops->fault(vmf);
Jan Kara39170482016-12-14 15:07:18 -08003242 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY |
Jan Karab1aa8122016-12-14 15:07:24 -08003243 VM_FAULT_DONE_COW)))
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07003244 return ret;
3245
Jan Kara667240e2016-12-14 15:07:07 -08003246 if (unlikely(PageHWPoison(vmf->page))) {
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07003247 if (ret & VM_FAULT_LOCKED)
Jan Kara667240e2016-12-14 15:07:07 -08003248 unlock_page(vmf->page);
3249 put_page(vmf->page);
Jan Kara936ca802016-12-14 15:07:10 -08003250 vmf->page = NULL;
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07003251 return VM_FAULT_HWPOISON;
3252 }
3253
3254 if (unlikely(!(ret & VM_FAULT_LOCKED)))
Jan Kara667240e2016-12-14 15:07:07 -08003255 lock_page(vmf->page);
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07003256 else
Jan Kara667240e2016-12-14 15:07:07 -08003257 VM_BUG_ON_PAGE(!PageLocked(vmf->page), vmf->page);
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07003258
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07003259 return ret;
3260}
3261
Ross Zwislerd0f0931d2017-06-02 14:46:34 -07003262/*
3263 * The ordering of these checks is important for pmds with _PAGE_DEVMAP set.
3264 * If we check pmd_trans_unstable() first we will trip the bad_pmd() check
3265 * inside of pmd_none_or_trans_huge_or_clear_bad(). This will end up correctly
3266 * returning 1 but not before it spams dmesg with the pmd_clear_bad() output.
3267 */
3268static int pmd_devmap_trans_unstable(pmd_t *pmd)
3269{
3270 return pmd_devmap(*pmd) || pmd_trans_unstable(pmd);
3271}
3272
Jan Kara82b0f8c2016-12-14 15:06:58 -08003273static int pte_alloc_one_map(struct vm_fault *vmf)
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003274{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003275 struct vm_area_struct *vma = vmf->vma;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003276
Jan Kara82b0f8c2016-12-14 15:06:58 -08003277 if (!pmd_none(*vmf->pmd))
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003278 goto map_pte;
Jan Kara82b0f8c2016-12-14 15:06:58 -08003279 if (vmf->prealloc_pte) {
3280 vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd);
3281 if (unlikely(!pmd_none(*vmf->pmd))) {
3282 spin_unlock(vmf->ptl);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003283 goto map_pte;
3284 }
3285
Kirill A. Shutemovc4812902017-11-15 17:35:37 -08003286 mm_inc_nr_ptes(vma->vm_mm);
Jan Kara82b0f8c2016-12-14 15:06:58 -08003287 pmd_populate(vma->vm_mm, vmf->pmd, vmf->prealloc_pte);
3288 spin_unlock(vmf->ptl);
Tobin C Harding7f2b6ce2017-02-24 14:58:59 -08003289 vmf->prealloc_pte = NULL;
Jan Kara82b0f8c2016-12-14 15:06:58 -08003290 } else if (unlikely(pte_alloc(vma->vm_mm, vmf->pmd, vmf->address))) {
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003291 return VM_FAULT_OOM;
3292 }
3293map_pte:
3294 /*
3295 * If a huge pmd materialized under us just retry later. Use
Ross Zwislerd0f0931d2017-06-02 14:46:34 -07003296 * pmd_trans_unstable() via pmd_devmap_trans_unstable() instead of
3297 * pmd_trans_huge() to ensure the pmd didn't become pmd_trans_huge
3298 * under us and then back to pmd_none, as a result of MADV_DONTNEED
3299 * running immediately after a huge pmd fault in a different thread of
3300 * this mm, in turn leading to a misleading pmd_trans_huge() retval.
3301 * All we have to ensure is that it is a regular pmd that we can walk
3302 * with pte_offset_map() and we can do that through an atomic read in
3303 * C, which is what pmd_trans_unstable() provides.
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003304 */
Ross Zwislerd0f0931d2017-06-02 14:46:34 -07003305 if (pmd_devmap_trans_unstable(vmf->pmd))
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003306 return VM_FAULT_NOPAGE;
3307
Ross Zwislerd0f0931d2017-06-02 14:46:34 -07003308 /*
3309 * At this point we know that our vmf->pmd points to a page of ptes
3310 * and it cannot become pmd_none(), pmd_devmap() or pmd_trans_huge()
3311 * for the duration of the fault. If a racing MADV_DONTNEED runs and
3312 * we zap the ptes pointed to by our vmf->pmd, the vmf->ptl will still
3313 * be valid and we will re-check to make sure the vmf->pte isn't
3314 * pte_none() under vmf->ptl protection when we return to
3315 * alloc_set_pte().
3316 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003317 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address,
3318 &vmf->ptl);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003319 return 0;
3320}
3321
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003322#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003323
3324#define HPAGE_CACHE_INDEX_MASK (HPAGE_PMD_NR - 1)
3325static inline bool transhuge_vma_suitable(struct vm_area_struct *vma,
3326 unsigned long haddr)
3327{
3328 if (((vma->vm_start >> PAGE_SHIFT) & HPAGE_CACHE_INDEX_MASK) !=
3329 (vma->vm_pgoff & HPAGE_CACHE_INDEX_MASK))
3330 return false;
3331 if (haddr < vma->vm_start || haddr + HPAGE_PMD_SIZE > vma->vm_end)
3332 return false;
3333 return true;
3334}
3335
Jan Kara82b0f8c2016-12-14 15:06:58 -08003336static void deposit_prealloc_pte(struct vm_fault *vmf)
Aneesh Kumar K.V953c66c2016-12-12 16:44:32 -08003337{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003338 struct vm_area_struct *vma = vmf->vma;
Aneesh Kumar K.V953c66c2016-12-12 16:44:32 -08003339
Jan Kara82b0f8c2016-12-14 15:06:58 -08003340 pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, vmf->prealloc_pte);
Aneesh Kumar K.V953c66c2016-12-12 16:44:32 -08003341 /*
3342 * We are going to consume the prealloc table,
3343 * count that as nr_ptes.
3344 */
Kirill A. Shutemovc4812902017-11-15 17:35:37 -08003345 mm_inc_nr_ptes(vma->vm_mm);
Tobin C Harding7f2b6ce2017-02-24 14:58:59 -08003346 vmf->prealloc_pte = NULL;
Aneesh Kumar K.V953c66c2016-12-12 16:44:32 -08003347}
3348
Jan Kara82b0f8c2016-12-14 15:06:58 -08003349static int do_set_pmd(struct vm_fault *vmf, struct page *page)
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003350{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003351 struct vm_area_struct *vma = vmf->vma;
3352 bool write = vmf->flags & FAULT_FLAG_WRITE;
3353 unsigned long haddr = vmf->address & HPAGE_PMD_MASK;
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003354 pmd_t entry;
3355 int i, ret;
3356
3357 if (!transhuge_vma_suitable(vma, haddr))
3358 return VM_FAULT_FALLBACK;
3359
3360 ret = VM_FAULT_FALLBACK;
3361 page = compound_head(page);
3362
Aneesh Kumar K.V953c66c2016-12-12 16:44:32 -08003363 /*
3364 * Archs like ppc64 need additonal space to store information
3365 * related to pte entry. Use the preallocated table for that.
3366 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003367 if (arch_needs_pgtable_deposit() && !vmf->prealloc_pte) {
3368 vmf->prealloc_pte = pte_alloc_one(vma->vm_mm, vmf->address);
3369 if (!vmf->prealloc_pte)
Aneesh Kumar K.V953c66c2016-12-12 16:44:32 -08003370 return VM_FAULT_OOM;
3371 smp_wmb(); /* See comment in __pte_alloc() */
3372 }
3373
Jan Kara82b0f8c2016-12-14 15:06:58 -08003374 vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd);
3375 if (unlikely(!pmd_none(*vmf->pmd)))
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003376 goto out;
3377
3378 for (i = 0; i < HPAGE_PMD_NR; i++)
3379 flush_icache_page(vma, page + i);
3380
3381 entry = mk_huge_pmd(page, vma->vm_page_prot);
3382 if (write)
Linus Torvaldsf55e1012017-11-29 09:01:01 -08003383 entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma);
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003384
Yang Shifadae292018-08-17 15:44:55 -07003385 add_mm_counter(vma->vm_mm, mm_counter_file(page), HPAGE_PMD_NR);
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003386 page_add_file_rmap(page, true);
Aneesh Kumar K.V953c66c2016-12-12 16:44:32 -08003387 /*
3388 * deposit and withdraw with pmd lock held
3389 */
3390 if (arch_needs_pgtable_deposit())
Jan Kara82b0f8c2016-12-14 15:06:58 -08003391 deposit_prealloc_pte(vmf);
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003392
Jan Kara82b0f8c2016-12-14 15:06:58 -08003393 set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry);
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003394
Jan Kara82b0f8c2016-12-14 15:06:58 -08003395 update_mmu_cache_pmd(vma, haddr, vmf->pmd);
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003396
3397 /* fault is handled */
3398 ret = 0;
Kirill A. Shutemov95ecedc2016-07-26 15:25:31 -07003399 count_vm_event(THP_FILE_MAPPED);
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003400out:
Jan Kara82b0f8c2016-12-14 15:06:58 -08003401 spin_unlock(vmf->ptl);
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003402 return ret;
3403}
3404#else
Jan Kara82b0f8c2016-12-14 15:06:58 -08003405static int do_set_pmd(struct vm_fault *vmf, struct page *page)
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003406{
3407 BUILD_BUG();
3408 return 0;
3409}
3410#endif
3411
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003412/**
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003413 * alloc_set_pte - setup new PTE entry for given page and add reverse page
3414 * mapping. If needed, the fucntion allocates page table or use pre-allocated.
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003415 *
Jan Kara82b0f8c2016-12-14 15:06:58 -08003416 * @vmf: fault environment
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003417 * @memcg: memcg to charge page (only for private mappings)
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003418 * @page: page to map
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003419 *
Jan Kara82b0f8c2016-12-14 15:06:58 -08003420 * Caller must take care of unlocking vmf->ptl, if vmf->pte is non-NULL on
3421 * return.
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003422 *
3423 * Target users are page handler itself and implementations of
3424 * vm_ops->map_pages.
3425 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003426int alloc_set_pte(struct vm_fault *vmf, struct mem_cgroup *memcg,
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003427 struct page *page)
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003428{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003429 struct vm_area_struct *vma = vmf->vma;
3430 bool write = vmf->flags & FAULT_FLAG_WRITE;
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003431 pte_t entry;
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003432 int ret;
3433
Jan Kara82b0f8c2016-12-14 15:06:58 -08003434 if (pmd_none(*vmf->pmd) && PageTransCompound(page) &&
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003435 IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003436 /* THP on COW? */
3437 VM_BUG_ON_PAGE(memcg, page);
3438
Jan Kara82b0f8c2016-12-14 15:06:58 -08003439 ret = do_set_pmd(vmf, page);
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003440 if (ret != VM_FAULT_FALLBACK)
Hugh Dickinsb0b9b3d2017-01-07 15:37:31 -08003441 return ret;
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003442 }
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003443
Jan Kara82b0f8c2016-12-14 15:06:58 -08003444 if (!vmf->pte) {
3445 ret = pte_alloc_one_map(vmf);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003446 if (ret)
Hugh Dickinsb0b9b3d2017-01-07 15:37:31 -08003447 return ret;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003448 }
3449
3450 /* Re-check under ptl */
Hugh Dickinsb0b9b3d2017-01-07 15:37:31 -08003451 if (unlikely(!pte_none(*vmf->pte)))
3452 return VM_FAULT_NOPAGE;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003453
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003454 flush_icache_page(vma, page);
3455 entry = mk_pte(page, vma->vm_page_prot);
3456 if (write)
3457 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003458 /* copy-on-write page */
3459 if (write && !(vma->vm_flags & VM_SHARED)) {
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003460 inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES);
Jan Kara82b0f8c2016-12-14 15:06:58 -08003461 page_add_new_anon_rmap(page, vma, vmf->address, false);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003462 mem_cgroup_commit_charge(page, memcg, false, false);
3463 lru_cache_add_active_or_unevictable(page, vma);
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003464 } else {
Jerome Marchandeca56ff2016-01-14 15:19:26 -08003465 inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page));
Kirill A. Shutemovdd78fed2016-07-26 15:25:26 -07003466 page_add_file_rmap(page, false);
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003467 }
Jan Kara82b0f8c2016-12-14 15:06:58 -08003468 set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry);
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003469
3470 /* no need to invalidate: a not-present page won't be cached */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003471 update_mmu_cache(vma, vmf->address, vmf->pte);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003472
Hugh Dickinsb0b9b3d2017-01-07 15:37:31 -08003473 return 0;
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003474}
3475
Jan Kara9118c0c2016-12-14 15:07:21 -08003476
3477/**
3478 * finish_fault - finish page fault once we have prepared the page to fault
3479 *
3480 * @vmf: structure describing the fault
3481 *
3482 * This function handles all that is needed to finish a page fault once the
3483 * page to fault in is prepared. It handles locking of PTEs, inserts PTE for
3484 * given page, adds reverse page mapping, handles memcg charges and LRU
3485 * addition. The function returns 0 on success, VM_FAULT_ code in case of
3486 * error.
3487 *
3488 * The function expects the page to be locked and on success it consumes a
3489 * reference of a page being mapped (for the PTE which maps it).
3490 */
3491int finish_fault(struct vm_fault *vmf)
3492{
3493 struct page *page;
Michal Hocko6b31d592017-08-18 15:16:15 -07003494 int ret = 0;
Jan Kara9118c0c2016-12-14 15:07:21 -08003495
3496 /* Did we COW the page? */
3497 if ((vmf->flags & FAULT_FLAG_WRITE) &&
3498 !(vmf->vma->vm_flags & VM_SHARED))
3499 page = vmf->cow_page;
3500 else
3501 page = vmf->page;
Michal Hocko6b31d592017-08-18 15:16:15 -07003502
3503 /*
3504 * check even for read faults because we might have lost our CoWed
3505 * page
3506 */
3507 if (!(vmf->vma->vm_flags & VM_SHARED))
3508 ret = check_stable_address_space(vmf->vma->vm_mm);
3509 if (!ret)
3510 ret = alloc_set_pte(vmf, vmf->memcg, page);
Jan Kara9118c0c2016-12-14 15:07:21 -08003511 if (vmf->pte)
3512 pte_unmap_unlock(vmf->pte, vmf->ptl);
3513 return ret;
3514}
3515
Kirill A. Shutemov3a910532014-08-06 16:08:07 -07003516static unsigned long fault_around_bytes __read_mostly =
3517 rounddown_pow_of_two(65536);
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07003518
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07003519#ifdef CONFIG_DEBUG_FS
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07003520static int fault_around_bytes_get(void *data, u64 *val)
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07003521{
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07003522 *val = fault_around_bytes;
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07003523 return 0;
3524}
3525
Andrey Ryabininb4903d62014-07-30 16:08:35 -07003526/*
William Kucharskida391d62018-01-31 16:21:11 -08003527 * fault_around_bytes must be rounded down to the nearest page order as it's
3528 * what do_fault_around() expects to see.
Andrey Ryabininb4903d62014-07-30 16:08:35 -07003529 */
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07003530static int fault_around_bytes_set(void *data, u64 val)
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07003531{
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07003532 if (val / PAGE_SIZE > PTRS_PER_PTE)
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07003533 return -EINVAL;
Andrey Ryabininb4903d62014-07-30 16:08:35 -07003534 if (val > PAGE_SIZE)
3535 fault_around_bytes = rounddown_pow_of_two(val);
3536 else
3537 fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07003538 return 0;
3539}
Yevgen Pronenko0a1345f2017-07-10 15:47:17 -07003540DEFINE_DEBUGFS_ATTRIBUTE(fault_around_bytes_fops,
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07003541 fault_around_bytes_get, fault_around_bytes_set, "%llu\n");
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07003542
3543static int __init fault_around_debugfs(void)
3544{
3545 void *ret;
3546
Yevgen Pronenko0a1345f2017-07-10 15:47:17 -07003547 ret = debugfs_create_file_unsafe("fault_around_bytes", 0644, NULL, NULL,
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07003548 &fault_around_bytes_fops);
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07003549 if (!ret)
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07003550 pr_warn("Failed to create fault_around_bytes in debugfs");
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07003551 return 0;
3552}
3553late_initcall(fault_around_debugfs);
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07003554#endif
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003555
Kirill A. Shutemov1fdb4122014-06-04 16:10:55 -07003556/*
3557 * do_fault_around() tries to map few pages around the fault address. The hope
3558 * is that the pages will be needed soon and this will lower the number of
3559 * faults to handle.
3560 *
3561 * It uses vm_ops->map_pages() to map the pages, which skips the page if it's
3562 * not ready to be mapped: not up-to-date, locked, etc.
3563 *
3564 * This function is called with the page table lock taken. In the split ptlock
3565 * case the page table lock only protects only those entries which belong to
3566 * the page table corresponding to the fault address.
3567 *
3568 * This function doesn't cross the VMA boundaries, in order to call map_pages()
3569 * only once.
3570 *
William Kucharskida391d62018-01-31 16:21:11 -08003571 * fault_around_bytes defines how many bytes we'll try to map.
3572 * do_fault_around() expects it to be set to a power of two less than or equal
3573 * to PTRS_PER_PTE.
Kirill A. Shutemov1fdb4122014-06-04 16:10:55 -07003574 *
William Kucharskida391d62018-01-31 16:21:11 -08003575 * The virtual address of the area that we map is naturally aligned to
3576 * fault_around_bytes rounded down to the machine page size
3577 * (and therefore to page order). This way it's easier to guarantee
3578 * that we don't cross page table boundaries.
Kirill A. Shutemov1fdb4122014-06-04 16:10:55 -07003579 */
Jan Kara0721ec82016-12-14 15:07:04 -08003580static int do_fault_around(struct vm_fault *vmf)
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003581{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003582 unsigned long address = vmf->address, nr_pages, mask;
Jan Kara0721ec82016-12-14 15:07:04 -08003583 pgoff_t start_pgoff = vmf->pgoff;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003584 pgoff_t end_pgoff;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003585 int off, ret = 0;
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003586
Jason Low4db0c3c2015-04-15 16:14:08 -07003587 nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT;
Kirill A. Shutemovaecd6f42014-08-06 16:08:05 -07003588 mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK;
3589
Jan Kara82b0f8c2016-12-14 15:06:58 -08003590 vmf->address = max(address & mask, vmf->vma->vm_start);
3591 off = ((address - vmf->address) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003592 start_pgoff -= off;
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003593
3594 /*
William Kucharskida391d62018-01-31 16:21:11 -08003595 * end_pgoff is either the end of the page table, the end of
3596 * the vma or nr_pages from start_pgoff, depending what is nearest.
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003597 */
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003598 end_pgoff = start_pgoff -
Jan Kara82b0f8c2016-12-14 15:06:58 -08003599 ((vmf->address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) +
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003600 PTRS_PER_PTE - 1;
Jan Kara82b0f8c2016-12-14 15:06:58 -08003601 end_pgoff = min3(end_pgoff, vma_pages(vmf->vma) + vmf->vma->vm_pgoff - 1,
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003602 start_pgoff + nr_pages - 1);
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003603
Jan Kara82b0f8c2016-12-14 15:06:58 -08003604 if (pmd_none(*vmf->pmd)) {
3605 vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm,
3606 vmf->address);
3607 if (!vmf->prealloc_pte)
Vegard Nossumc5f88bd2016-08-02 14:02:22 -07003608 goto out;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003609 smp_wmb(); /* See comment in __pte_alloc() */
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003610 }
3611
Jan Kara82b0f8c2016-12-14 15:06:58 -08003612 vmf->vma->vm_ops->map_pages(vmf, start_pgoff, end_pgoff);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003613
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003614 /* Huge page is mapped? Page fault is solved */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003615 if (pmd_trans_huge(*vmf->pmd)) {
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003616 ret = VM_FAULT_NOPAGE;
3617 goto out;
3618 }
3619
3620 /* ->map_pages() haven't done anything useful. Cold page cache? */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003621 if (!vmf->pte)
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003622 goto out;
3623
3624 /* check if the page fault is solved */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003625 vmf->pte -= (vmf->address >> PAGE_SHIFT) - (address >> PAGE_SHIFT);
3626 if (!pte_none(*vmf->pte))
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003627 ret = VM_FAULT_NOPAGE;
Jan Kara82b0f8c2016-12-14 15:06:58 -08003628 pte_unmap_unlock(vmf->pte, vmf->ptl);
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003629out:
Jan Kara82b0f8c2016-12-14 15:06:58 -08003630 vmf->address = address;
3631 vmf->pte = NULL;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003632 return ret;
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003633}
3634
Jan Kara0721ec82016-12-14 15:07:04 -08003635static int do_read_fault(struct vm_fault *vmf)
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07003636{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003637 struct vm_area_struct *vma = vmf->vma;
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003638 int ret = 0;
3639
3640 /*
3641 * Let's call ->map_pages() first and use ->fault() as fallback
3642 * if page by the offset is not ready to be mapped (cold cache or
3643 * something).
3644 */
Kirill A. Shutemov9b4bdd22015-02-10 14:09:51 -08003645 if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) {
Jan Kara0721ec82016-12-14 15:07:04 -08003646 ret = do_fault_around(vmf);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003647 if (ret)
3648 return ret;
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003649 }
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07003650
Jan Kara936ca802016-12-14 15:07:10 -08003651 ret = __do_fault(vmf);
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07003652 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
3653 return ret;
3654
Jan Kara9118c0c2016-12-14 15:07:21 -08003655 ret |= finish_fault(vmf);
Jan Kara936ca802016-12-14 15:07:10 -08003656 unlock_page(vmf->page);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003657 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
Jan Kara936ca802016-12-14 15:07:10 -08003658 put_page(vmf->page);
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07003659 return ret;
3660}
3661
Jan Kara0721ec82016-12-14 15:07:04 -08003662static int do_cow_fault(struct vm_fault *vmf)
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003663{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003664 struct vm_area_struct *vma = vmf->vma;
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003665 int ret;
3666
3667 if (unlikely(anon_vma_prepare(vma)))
3668 return VM_FAULT_OOM;
3669
Jan Kara936ca802016-12-14 15:07:10 -08003670 vmf->cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vmf->address);
3671 if (!vmf->cow_page)
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003672 return VM_FAULT_OOM;
3673
Tejun Heo2cf85582018-07-03 11:14:56 -04003674 if (mem_cgroup_try_charge_delay(vmf->cow_page, vma->vm_mm, GFP_KERNEL,
Jan Kara39170482016-12-14 15:07:18 -08003675 &vmf->memcg, false)) {
Jan Kara936ca802016-12-14 15:07:10 -08003676 put_page(vmf->cow_page);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003677 return VM_FAULT_OOM;
3678 }
3679
Jan Kara936ca802016-12-14 15:07:10 -08003680 ret = __do_fault(vmf);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003681 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
3682 goto uncharge_out;
Jan Kara39170482016-12-14 15:07:18 -08003683 if (ret & VM_FAULT_DONE_COW)
3684 return ret;
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003685
Jan Karab1aa8122016-12-14 15:07:24 -08003686 copy_user_highpage(vmf->cow_page, vmf->page, vmf->address, vma);
Jan Kara936ca802016-12-14 15:07:10 -08003687 __SetPageUptodate(vmf->cow_page);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003688
Jan Kara9118c0c2016-12-14 15:07:21 -08003689 ret |= finish_fault(vmf);
Jan Karab1aa8122016-12-14 15:07:24 -08003690 unlock_page(vmf->page);
3691 put_page(vmf->page);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003692 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
3693 goto uncharge_out;
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003694 return ret;
3695uncharge_out:
Jan Kara39170482016-12-14 15:07:18 -08003696 mem_cgroup_cancel_charge(vmf->cow_page, vmf->memcg, false);
Jan Kara936ca802016-12-14 15:07:10 -08003697 put_page(vmf->cow_page);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003698 return ret;
3699}
3700
Jan Kara0721ec82016-12-14 15:07:04 -08003701static int do_shared_fault(struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003702{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003703 struct vm_area_struct *vma = vmf->vma;
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003704 int ret, tmp;
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07003705
Jan Kara936ca802016-12-14 15:07:10 -08003706 ret = __do_fault(vmf);
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07003707 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003708 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003709
3710 /*
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003711 * Check if the backing address space wants to know that the page is
3712 * about to become writable
Linus Torvalds1da177e2005-04-16 15:20:36 -07003713 */
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07003714 if (vma->vm_ops->page_mkwrite) {
Jan Kara936ca802016-12-14 15:07:10 -08003715 unlock_page(vmf->page);
Jan Kara38b8cb72016-12-14 15:07:30 -08003716 tmp = do_page_mkwrite(vmf);
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07003717 if (unlikely(!tmp ||
3718 (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
Jan Kara936ca802016-12-14 15:07:10 -08003719 put_page(vmf->page);
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07003720 return tmp;
3721 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003722 }
3723
Jan Kara9118c0c2016-12-14 15:07:21 -08003724 ret |= finish_fault(vmf);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003725 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE |
3726 VM_FAULT_RETRY))) {
Jan Kara936ca802016-12-14 15:07:10 -08003727 unlock_page(vmf->page);
3728 put_page(vmf->page);
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003729 return ret;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07003730 }
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003731
Jan Kara97ba0c22016-12-14 15:07:27 -08003732 fault_dirty_shared_page(vma, vmf->page);
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07003733 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07003734}
Nick Piggind00806b2007-07-19 01:46:57 -07003735
Paul Cassella9a95f3c2014-08-06 16:07:24 -07003736/*
3737 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3738 * but allow concurrent faults).
3739 * The mmap_sem may have been released depending on flags and our
3740 * return value. See filemap_fault() and __lock_page_or_retry().
3741 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003742static int do_fault(struct vm_fault *vmf)
Nick Piggin54cb8822007-07-19 01:46:59 -07003743{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003744 struct vm_area_struct *vma = vmf->vma;
Hugh Dickinsb0b9b3d2017-01-07 15:37:31 -08003745 int ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07003746
Kirill A. Shutemov6b7339f2015-07-06 23:18:37 +03003747 /* The VMA was not fully populated on mmap() or missing VM_DONTEXPAND */
3748 if (!vma->vm_ops->fault)
Hugh Dickinsb0b9b3d2017-01-07 15:37:31 -08003749 ret = VM_FAULT_SIGBUS;
3750 else if (!(vmf->flags & FAULT_FLAG_WRITE))
3751 ret = do_read_fault(vmf);
3752 else if (!(vma->vm_flags & VM_SHARED))
3753 ret = do_cow_fault(vmf);
3754 else
3755 ret = do_shared_fault(vmf);
3756
3757 /* preallocated pagetable is unused: free it */
3758 if (vmf->prealloc_pte) {
3759 pte_free(vma->vm_mm, vmf->prealloc_pte);
Tobin C Harding7f2b6ce2017-02-24 14:58:59 -08003760 vmf->prealloc_pte = NULL;
Hugh Dickinsb0b9b3d2017-01-07 15:37:31 -08003761 }
3762 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07003763}
3764
Rashika Kheriab19a9932014-04-03 14:48:02 -07003765static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
Rik van Riel04bb2f92013-10-07 11:29:36 +01003766 unsigned long addr, int page_nid,
3767 int *flags)
Mel Gorman9532fec2012-11-15 01:24:32 +00003768{
3769 get_page(page);
3770
3771 count_vm_numa_event(NUMA_HINT_FAULTS);
Rik van Riel04bb2f92013-10-07 11:29:36 +01003772 if (page_nid == numa_node_id()) {
Mel Gorman9532fec2012-11-15 01:24:32 +00003773 count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL);
Rik van Riel04bb2f92013-10-07 11:29:36 +01003774 *flags |= TNF_FAULT_LOCAL;
3775 }
Mel Gorman9532fec2012-11-15 01:24:32 +00003776
3777 return mpol_misplaced(page, vma, addr);
3778}
3779
Jan Kara29943022016-12-14 15:07:16 -08003780static int do_numa_page(struct vm_fault *vmf)
Mel Gormand10e63f2012-10-25 14:16:31 +02003781{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003782 struct vm_area_struct *vma = vmf->vma;
Mel Gorman4daae3b2012-11-02 11:33:45 +00003783 struct page *page = NULL;
Mel Gorman8191acb2013-10-07 11:28:45 +01003784 int page_nid = -1;
Peter Zijlstra90572892013-10-07 11:29:20 +01003785 int last_cpupid;
Peter Zijlstracbee9f82012-10-25 14:16:43 +02003786 int target_nid;
Mel Gormanb8593bf2012-11-21 01:18:23 +00003787 bool migrated = false;
Aneesh Kumar K.Vcee216a2017-02-24 14:59:13 -08003788 pte_t pte;
Aneesh Kumar K.V288bc542017-02-24 14:59:16 -08003789 bool was_writable = pte_savedwrite(vmf->orig_pte);
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003790 int flags = 0;
Mel Gormand10e63f2012-10-25 14:16:31 +02003791
3792 /*
Tobin C Harding166f61b2017-02-24 14:59:01 -08003793 * The "pte" at this point cannot be used safely without
3794 * validation through pte_unmap_same(). It's of NUMA type but
3795 * the pfn may be screwed if the read is non atomic.
Tobin C Harding166f61b2017-02-24 14:59:01 -08003796 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003797 vmf->ptl = pte_lockptr(vma->vm_mm, vmf->pmd);
3798 spin_lock(vmf->ptl);
Aneesh Kumar K.Vcee216a2017-02-24 14:59:13 -08003799 if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) {
Jan Kara82b0f8c2016-12-14 15:06:58 -08003800 pte_unmap_unlock(vmf->pte, vmf->ptl);
Mel Gorman4daae3b2012-11-02 11:33:45 +00003801 goto out;
3802 }
3803
Aneesh Kumar K.Vcee216a2017-02-24 14:59:13 -08003804 /*
3805 * Make it present again, Depending on how arch implementes non
3806 * accessible ptes, some can allow access by kernel mode.
3807 */
3808 pte = ptep_modify_prot_start(vma->vm_mm, vmf->address, vmf->pte);
Mel Gorman4d942462015-02-12 14:58:28 -08003809 pte = pte_modify(pte, vma->vm_page_prot);
3810 pte = pte_mkyoung(pte);
Mel Gormanb191f9b2015-03-25 15:55:40 -07003811 if (was_writable)
3812 pte = pte_mkwrite(pte);
Aneesh Kumar K.Vcee216a2017-02-24 14:59:13 -08003813 ptep_modify_prot_commit(vma->vm_mm, vmf->address, vmf->pte, pte);
Jan Kara82b0f8c2016-12-14 15:06:58 -08003814 update_mmu_cache(vma, vmf->address, vmf->pte);
Mel Gormand10e63f2012-10-25 14:16:31 +02003815
Jan Kara82b0f8c2016-12-14 15:06:58 -08003816 page = vm_normal_page(vma, vmf->address, pte);
Mel Gormand10e63f2012-10-25 14:16:31 +02003817 if (!page) {
Jan Kara82b0f8c2016-12-14 15:06:58 -08003818 pte_unmap_unlock(vmf->pte, vmf->ptl);
Mel Gormand10e63f2012-10-25 14:16:31 +02003819 return 0;
3820 }
3821
Kirill A. Shutemove81c4802016-01-15 16:53:49 -08003822 /* TODO: handle PTE-mapped THP */
3823 if (PageCompound(page)) {
Jan Kara82b0f8c2016-12-14 15:06:58 -08003824 pte_unmap_unlock(vmf->pte, vmf->ptl);
Kirill A. Shutemove81c4802016-01-15 16:53:49 -08003825 return 0;
3826 }
3827
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003828 /*
Mel Gormanbea66fb2015-03-25 15:55:37 -07003829 * Avoid grouping on RO pages in general. RO pages shouldn't hurt as
3830 * much anyway since they can be in shared cache state. This misses
3831 * the case where a mapping is writable but the process never writes
3832 * to it but pte_write gets cleared during protection updates and
3833 * pte_dirty has unpredictable behaviour between PTE scan updates,
3834 * background writeback, dirty balancing and application behaviour.
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003835 */
Rik van Rield59dc7b2016-09-08 21:30:53 -04003836 if (!pte_write(pte))
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003837 flags |= TNF_NO_GROUP;
3838
Rik van Rieldabe1d92013-10-07 11:29:34 +01003839 /*
3840 * Flag if the page is shared between multiple address spaces. This
3841 * is later used when determining whether to group tasks together
3842 */
3843 if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED))
3844 flags |= TNF_SHARED;
3845
Peter Zijlstra90572892013-10-07 11:29:20 +01003846 last_cpupid = page_cpupid_last(page);
Mel Gorman8191acb2013-10-07 11:28:45 +01003847 page_nid = page_to_nid(page);
Jan Kara82b0f8c2016-12-14 15:06:58 -08003848 target_nid = numa_migrate_prep(page, vma, vmf->address, page_nid,
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003849 &flags);
Jan Kara82b0f8c2016-12-14 15:06:58 -08003850 pte_unmap_unlock(vmf->pte, vmf->ptl);
Mel Gorman4daae3b2012-11-02 11:33:45 +00003851 if (target_nid == -1) {
Mel Gorman4daae3b2012-11-02 11:33:45 +00003852 put_page(page);
3853 goto out;
3854 }
3855
3856 /* Migrate to the requested node */
Mel Gorman1bc115d2013-10-07 11:29:05 +01003857 migrated = migrate_misplaced_page(page, vma, target_nid);
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003858 if (migrated) {
Mel Gorman8191acb2013-10-07 11:28:45 +01003859 page_nid = target_nid;
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003860 flags |= TNF_MIGRATED;
Mel Gorman074c2382015-03-25 15:55:42 -07003861 } else
3862 flags |= TNF_MIGRATE_FAIL;
Mel Gorman4daae3b2012-11-02 11:33:45 +00003863
3864out:
Mel Gorman8191acb2013-10-07 11:28:45 +01003865 if (page_nid != -1)
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003866 task_numa_fault(last_cpupid, page_nid, 1, flags);
Mel Gormand10e63f2012-10-25 14:16:31 +02003867 return 0;
3868}
3869
Geert Uytterhoeven91a90142017-07-12 14:33:08 -07003870static inline int create_huge_pmd(struct vm_fault *vmf)
Matthew Wilcoxb96375f2015-09-08 14:58:48 -07003871{
Dave Jiangf4200392017-02-22 15:40:06 -08003872 if (vma_is_anonymous(vmf->vma))
Jan Kara82b0f8c2016-12-14 15:06:58 -08003873 return do_huge_pmd_anonymous_page(vmf);
Dave Jianga2d58162017-02-24 14:56:59 -08003874 if (vmf->vma->vm_ops->huge_fault)
Dave Jiangc791ace2017-02-24 14:57:08 -08003875 return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD);
Matthew Wilcoxb96375f2015-09-08 14:58:48 -07003876 return VM_FAULT_FALLBACK;
3877}
3878
Geert Uytterhoeven183f24a2017-12-14 15:32:52 -08003879/* `inline' is required to avoid gcc 4.1.2 build error */
3880static inline int wp_huge_pmd(struct vm_fault *vmf, pmd_t orig_pmd)
Matthew Wilcoxb96375f2015-09-08 14:58:48 -07003881{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003882 if (vma_is_anonymous(vmf->vma))
3883 return do_huge_pmd_wp_page(vmf, orig_pmd);
Dave Jianga2d58162017-02-24 14:56:59 -08003884 if (vmf->vma->vm_ops->huge_fault)
Dave Jiangc791ace2017-02-24 14:57:08 -08003885 return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD);
Kirill A. Shutemovaf9e4d52016-07-26 15:25:40 -07003886
3887 /* COW handled on pte level: split pmd */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003888 VM_BUG_ON_VMA(vmf->vma->vm_flags & VM_SHARED, vmf->vma);
3889 __split_huge_pmd(vmf->vma, vmf->pmd, vmf->address, false, NULL);
Kirill A. Shutemovaf9e4d52016-07-26 15:25:40 -07003890
Matthew Wilcoxb96375f2015-09-08 14:58:48 -07003891 return VM_FAULT_FALLBACK;
3892}
3893
Lorenzo Stoakes38e08852016-09-11 23:54:25 +01003894static inline bool vma_is_accessible(struct vm_area_struct *vma)
3895{
3896 return vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE);
3897}
3898
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08003899static int create_huge_pud(struct vm_fault *vmf)
3900{
3901#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3902 /* No support for anonymous transparent PUD pages yet */
3903 if (vma_is_anonymous(vmf->vma))
3904 return VM_FAULT_FALLBACK;
3905 if (vmf->vma->vm_ops->huge_fault)
Dave Jiangc791ace2017-02-24 14:57:08 -08003906 return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD);
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08003907#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
3908 return VM_FAULT_FALLBACK;
3909}
3910
3911static int wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud)
3912{
3913#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3914 /* No support for anonymous transparent PUD pages yet */
3915 if (vma_is_anonymous(vmf->vma))
3916 return VM_FAULT_FALLBACK;
3917 if (vmf->vma->vm_ops->huge_fault)
Dave Jiangc791ace2017-02-24 14:57:08 -08003918 return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD);
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08003919#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
3920 return VM_FAULT_FALLBACK;
3921}
3922
Linus Torvalds1da177e2005-04-16 15:20:36 -07003923/*
3924 * These routines also need to handle stuff like marking pages dirty
3925 * and/or accessed for architectures that don't do it in hardware (most
3926 * RISC architectures). The early dirtying is also good on the i386.
3927 *
3928 * There is also a hook called "update_mmu_cache()" that architectures
3929 * with external mmu caches can use to update those (ie the Sparc or
3930 * PowerPC hashed page tables that act as extended TLBs).
3931 *
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003932 * We enter with non-exclusive mmap_sem (to exclude vma changes, but allow
3933 * concurrent faults).
Paul Cassella9a95f3c2014-08-06 16:07:24 -07003934 *
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003935 * The mmap_sem may have been released depending on flags and our return value.
3936 * See filemap_fault() and __lock_page_or_retry().
Linus Torvalds1da177e2005-04-16 15:20:36 -07003937 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003938static int handle_pte_fault(struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003939{
3940 pte_t entry;
3941
Jan Kara82b0f8c2016-12-14 15:06:58 -08003942 if (unlikely(pmd_none(*vmf->pmd))) {
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003943 /*
3944 * Leave __pte_alloc() until later: because vm_ops->fault may
3945 * want to allocate huge page, and if we expose page table
3946 * for an instant, it will be difficult to retract from
3947 * concurrent faults and from rmap lookups.
3948 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003949 vmf->pte = NULL;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003950 } else {
3951 /* See comment in pte_alloc_one_map() */
Ross Zwislerd0f0931d2017-06-02 14:46:34 -07003952 if (pmd_devmap_trans_unstable(vmf->pmd))
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003953 return 0;
3954 /*
3955 * A regular pmd is established and it can't morph into a huge
3956 * pmd from under us anymore at this point because we hold the
3957 * mmap_sem read mode and khugepaged takes it in write mode.
3958 * So now it's safe to run pte_offset_map().
3959 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003960 vmf->pte = pte_offset_map(vmf->pmd, vmf->address);
Jan Kara29943022016-12-14 15:07:16 -08003961 vmf->orig_pte = *vmf->pte;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003962
3963 /*
3964 * some architectures can have larger ptes than wordsize,
3965 * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and
Paul E. McKenneyb03a0fe2017-10-23 14:07:25 -07003966 * CONFIG_32BIT=y, so READ_ONCE cannot guarantee atomic
3967 * accesses. The code below just needs a consistent view
3968 * for the ifs and we later double check anyway with the
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003969 * ptl lock held. So here a barrier will do.
3970 */
3971 barrier();
Jan Kara29943022016-12-14 15:07:16 -08003972 if (pte_none(vmf->orig_pte)) {
Jan Kara82b0f8c2016-12-14 15:06:58 -08003973 pte_unmap(vmf->pte);
3974 vmf->pte = NULL;
Hugh Dickins65500d22005-10-29 18:15:59 -07003975 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003976 }
3977
Jan Kara82b0f8c2016-12-14 15:06:58 -08003978 if (!vmf->pte) {
3979 if (vma_is_anonymous(vmf->vma))
3980 return do_anonymous_page(vmf);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003981 else
Jan Kara82b0f8c2016-12-14 15:06:58 -08003982 return do_fault(vmf);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003983 }
3984
Jan Kara29943022016-12-14 15:07:16 -08003985 if (!pte_present(vmf->orig_pte))
3986 return do_swap_page(vmf);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003987
Jan Kara29943022016-12-14 15:07:16 -08003988 if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma))
3989 return do_numa_page(vmf);
Mel Gormand10e63f2012-10-25 14:16:31 +02003990
Jan Kara82b0f8c2016-12-14 15:06:58 -08003991 vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
3992 spin_lock(vmf->ptl);
Jan Kara29943022016-12-14 15:07:16 -08003993 entry = vmf->orig_pte;
Jan Kara82b0f8c2016-12-14 15:06:58 -08003994 if (unlikely(!pte_same(*vmf->pte, entry)))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003995 goto unlock;
Jan Kara82b0f8c2016-12-14 15:06:58 -08003996 if (vmf->flags & FAULT_FLAG_WRITE) {
Linus Torvaldsf6f37322017-12-15 18:53:22 -08003997 if (!pte_write(entry))
Jan Kara29943022016-12-14 15:07:16 -08003998 return do_wp_page(vmf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003999 entry = pte_mkdirty(entry);
4000 }
4001 entry = pte_mkyoung(entry);
Jan Kara82b0f8c2016-12-14 15:06:58 -08004002 if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry,
4003 vmf->flags & FAULT_FLAG_WRITE)) {
4004 update_mmu_cache(vmf->vma, vmf->address, vmf->pte);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07004005 } else {
4006 /*
4007 * This is needed only for protection faults but the arch code
4008 * is not yet telling us if this is a protection fault or not.
4009 * This still avoids useless tlb flushes for .text page faults
4010 * with threads.
4011 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08004012 if (vmf->flags & FAULT_FLAG_WRITE)
4013 flush_tlb_fix_spurious_fault(vmf->vma, vmf->address);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07004014 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07004015unlock:
Jan Kara82b0f8c2016-12-14 15:06:58 -08004016 pte_unmap_unlock(vmf->pte, vmf->ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07004017 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004018}
4019
4020/*
4021 * By the time we get here, we already hold the mm semaphore
Paul Cassella9a95f3c2014-08-06 16:07:24 -07004022 *
4023 * The mmap_sem may have been released depending on flags and our
4024 * return value. See filemap_fault() and __lock_page_or_retry().
Linus Torvalds1da177e2005-04-16 15:20:36 -07004025 */
Kirill A. Shutemovdcddffd2016-07-26 15:25:18 -07004026static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
4027 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004028{
Jan Kara82b0f8c2016-12-14 15:06:58 -08004029 struct vm_fault vmf = {
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07004030 .vma = vma,
Jan Kara1a29d852016-12-14 15:07:01 -08004031 .address = address & PAGE_MASK,
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07004032 .flags = flags,
Jan Kara0721ec82016-12-14 15:07:04 -08004033 .pgoff = linear_page_index(vma, address),
Jan Kara667240e2016-12-14 15:07:07 -08004034 .gfp_mask = __get_fault_gfp_mask(vma),
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07004035 };
Anshuman Khandualfde26be2017-09-08 16:12:45 -07004036 unsigned int dirty = flags & FAULT_FLAG_WRITE;
Kirill A. Shutemovdcddffd2016-07-26 15:25:18 -07004037 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004038 pgd_t *pgd;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03004039 p4d_t *p4d;
Dave Jianga2d58162017-02-24 14:56:59 -08004040 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004041
Linus Torvalds1da177e2005-04-16 15:20:36 -07004042 pgd = pgd_offset(mm, address);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03004043 p4d = p4d_alloc(mm, pgd, address);
4044 if (!p4d)
4045 return VM_FAULT_OOM;
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08004046
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03004047 vmf.pud = pud_alloc(mm, p4d, address);
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08004048 if (!vmf.pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07004049 return VM_FAULT_OOM;
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08004050 if (pud_none(*vmf.pud) && transparent_hugepage_enabled(vma)) {
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08004051 ret = create_huge_pud(&vmf);
4052 if (!(ret & VM_FAULT_FALLBACK))
4053 return ret;
4054 } else {
4055 pud_t orig_pud = *vmf.pud;
4056
4057 barrier();
4058 if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) {
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08004059
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08004060 /* NUMA case for anonymous PUDs would go here */
4061
Linus Torvaldsf6f37322017-12-15 18:53:22 -08004062 if (dirty && !pud_write(orig_pud)) {
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08004063 ret = wp_huge_pud(&vmf, orig_pud);
4064 if (!(ret & VM_FAULT_FALLBACK))
4065 return ret;
4066 } else {
4067 huge_pud_set_accessed(&vmf, orig_pud);
4068 return 0;
4069 }
4070 }
4071 }
4072
4073 vmf.pmd = pmd_alloc(mm, vmf.pud, address);
Jan Kara82b0f8c2016-12-14 15:06:58 -08004074 if (!vmf.pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07004075 return VM_FAULT_OOM;
Jan Kara82b0f8c2016-12-14 15:06:58 -08004076 if (pmd_none(*vmf.pmd) && transparent_hugepage_enabled(vma)) {
Dave Jianga2d58162017-02-24 14:56:59 -08004077 ret = create_huge_pmd(&vmf);
Kirill A. Shutemovc0292552013-09-12 15:14:05 -07004078 if (!(ret & VM_FAULT_FALLBACK))
4079 return ret;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08004080 } else {
Jan Kara82b0f8c2016-12-14 15:06:58 -08004081 pmd_t orig_pmd = *vmf.pmd;
David Rientjes1f1d06c2012-05-29 15:06:23 -07004082
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08004083 barrier();
Zi Yan84c3fc42017-09-08 16:11:01 -07004084 if (unlikely(is_swap_pmd(orig_pmd))) {
4085 VM_BUG_ON(thp_migration_supported() &&
4086 !is_pmd_migration_entry(orig_pmd));
4087 if (is_pmd_migration_entry(orig_pmd))
4088 pmd_migration_entry_wait(mm, vmf.pmd);
4089 return 0;
4090 }
Dan Williams5c7fb562016-01-15 16:56:52 -08004091 if (pmd_trans_huge(orig_pmd) || pmd_devmap(orig_pmd)) {
Lorenzo Stoakes38e08852016-09-11 23:54:25 +01004092 if (pmd_protnone(orig_pmd) && vma_is_accessible(vma))
Jan Kara82b0f8c2016-12-14 15:06:58 -08004093 return do_huge_pmd_numa_page(&vmf, orig_pmd);
Mel Gormand10e63f2012-10-25 14:16:31 +02004094
Linus Torvaldsf6f37322017-12-15 18:53:22 -08004095 if (dirty && !pmd_write(orig_pmd)) {
Jan Kara82b0f8c2016-12-14 15:06:58 -08004096 ret = wp_huge_pmd(&vmf, orig_pmd);
Kirill A. Shutemov9845cbb2014-02-25 15:01:42 -08004097 if (!(ret & VM_FAULT_FALLBACK))
4098 return ret;
Will Deacona1dd4502012-12-11 16:01:27 -08004099 } else {
Jan Kara82b0f8c2016-12-14 15:06:58 -08004100 huge_pmd_set_accessed(&vmf, orig_pmd);
Kirill A. Shutemov9845cbb2014-02-25 15:01:42 -08004101 return 0;
David Rientjes1f1d06c2012-05-29 15:06:23 -07004102 }
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08004103 }
4104 }
4105
Jan Kara82b0f8c2016-12-14 15:06:58 -08004106 return handle_pte_fault(&vmf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004107}
4108
Paul Cassella9a95f3c2014-08-06 16:07:24 -07004109/*
4110 * By the time we get here, we already hold the mm semaphore
4111 *
4112 * The mmap_sem may have been released depending on flags and our
4113 * return value. See filemap_fault() and __lock_page_or_retry().
4114 */
Kirill A. Shutemovdcddffd2016-07-26 15:25:18 -07004115int handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
4116 unsigned int flags)
Johannes Weiner519e5242013-09-12 15:13:42 -07004117{
4118 int ret;
4119
4120 __set_current_state(TASK_RUNNING);
4121
4122 count_vm_event(PGFAULT);
Roman Gushchin22621852017-07-06 15:40:25 -07004123 count_memcg_event_mm(vma->vm_mm, PGFAULT);
Johannes Weiner519e5242013-09-12 15:13:42 -07004124
4125 /* do counter updates before entering really critical section. */
4126 check_sync_rss_stat(current);
4127
Laurent Dufourde0c7992017-09-08 16:13:12 -07004128 if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE,
4129 flags & FAULT_FLAG_INSTRUCTION,
4130 flags & FAULT_FLAG_REMOTE))
4131 return VM_FAULT_SIGSEGV;
4132
Johannes Weiner519e5242013-09-12 15:13:42 -07004133 /*
4134 * Enable the memcg OOM handling for faults triggered in user
4135 * space. Kernel faults are handled more gracefully.
4136 */
4137 if (flags & FAULT_FLAG_USER)
Michal Hocko29ef6802018-08-17 15:47:11 -07004138 mem_cgroup_enter_user_fault();
Johannes Weiner519e5242013-09-12 15:13:42 -07004139
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07004140 if (unlikely(is_vm_hugetlb_page(vma)))
4141 ret = hugetlb_fault(vma->vm_mm, vma, address, flags);
4142 else
4143 ret = __handle_mm_fault(vma, address, flags);
Johannes Weiner519e5242013-09-12 15:13:42 -07004144
Johannes Weiner49426422013-10-16 13:46:59 -07004145 if (flags & FAULT_FLAG_USER) {
Michal Hocko29ef6802018-08-17 15:47:11 -07004146 mem_cgroup_exit_user_fault();
Tobin C Harding166f61b2017-02-24 14:59:01 -08004147 /*
4148 * The task may have entered a memcg OOM situation but
4149 * if the allocation error was handled gracefully (no
4150 * VM_FAULT_OOM), there is no need to kill anything.
4151 * Just clean up the OOM state peacefully.
4152 */
4153 if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM))
4154 mem_cgroup_oom_synchronize(false);
Johannes Weiner49426422013-10-16 13:46:59 -07004155 }
Johannes Weiner3812c8c2013-09-12 15:13:44 -07004156
Johannes Weiner519e5242013-09-12 15:13:42 -07004157 return ret;
4158}
Jesse Barnese1d6d012014-12-12 16:55:27 -08004159EXPORT_SYMBOL_GPL(handle_mm_fault);
Johannes Weiner519e5242013-09-12 15:13:42 -07004160
Kirill A. Shutemov90eceff2017-03-09 17:24:08 +03004161#ifndef __PAGETABLE_P4D_FOLDED
4162/*
4163 * Allocate p4d page table.
4164 * We've already handled the fast-path in-line.
4165 */
4166int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
4167{
4168 p4d_t *new = p4d_alloc_one(mm, address);
4169 if (!new)
4170 return -ENOMEM;
4171
4172 smp_wmb(); /* See comment in __pte_alloc */
4173
4174 spin_lock(&mm->page_table_lock);
4175 if (pgd_present(*pgd)) /* Another has populated it */
4176 p4d_free(mm, new);
4177 else
4178 pgd_populate(mm, pgd, new);
4179 spin_unlock(&mm->page_table_lock);
4180 return 0;
4181}
4182#endif /* __PAGETABLE_P4D_FOLDED */
4183
Linus Torvalds1da177e2005-04-16 15:20:36 -07004184#ifndef __PAGETABLE_PUD_FOLDED
4185/*
4186 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07004187 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004188 */
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03004189int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004190{
Hugh Dickinsc74df322005-10-29 18:16:23 -07004191 pud_t *new = pud_alloc_one(mm, address);
4192 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07004193 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004194
Nick Piggin362a61a2008-05-14 06:37:36 +02004195 smp_wmb(); /* See comment in __pte_alloc */
4196
Hugh Dickins872fec12005-10-29 18:16:21 -07004197 spin_lock(&mm->page_table_lock);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03004198#ifndef __ARCH_HAS_5LEVEL_HACK
Kirill A. Shutemovb4e98d92017-11-15 17:35:33 -08004199 if (!p4d_present(*p4d)) {
4200 mm_inc_nr_puds(mm);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03004201 p4d_populate(mm, p4d, new);
Kirill A. Shutemovb4e98d92017-11-15 17:35:33 -08004202 } else /* Another has populated it */
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03004203 pud_free(mm, new);
Kirill A. Shutemovb4e98d92017-11-15 17:35:33 -08004204#else
4205 if (!pgd_present(*p4d)) {
4206 mm_inc_nr_puds(mm);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03004207 pgd_populate(mm, p4d, new);
Kirill A. Shutemovb4e98d92017-11-15 17:35:33 -08004208 } else /* Another has populated it */
4209 pud_free(mm, new);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03004210#endif /* __ARCH_HAS_5LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07004211 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07004212 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004213}
4214#endif /* __PAGETABLE_PUD_FOLDED */
4215
4216#ifndef __PAGETABLE_PMD_FOLDED
4217/*
4218 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07004219 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004220 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07004221int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004222{
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08004223 spinlock_t *ptl;
Hugh Dickinsc74df322005-10-29 18:16:23 -07004224 pmd_t *new = pmd_alloc_one(mm, address);
4225 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07004226 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004227
Nick Piggin362a61a2008-05-14 06:37:36 +02004228 smp_wmb(); /* See comment in __pte_alloc */
4229
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08004230 ptl = pud_lock(mm, pud);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004231#ifndef __ARCH_HAS_4LEVEL_HACK
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -08004232 if (!pud_present(*pud)) {
4233 mm_inc_nr_pmds(mm);
Hugh Dickins1bb36302005-10-29 18:16:22 -07004234 pud_populate(mm, pud, new);
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -08004235 } else /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08004236 pmd_free(mm, new);
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -08004237#else
4238 if (!pgd_present(*pud)) {
4239 mm_inc_nr_pmds(mm);
Hugh Dickins1bb36302005-10-29 18:16:22 -07004240 pgd_populate(mm, pud, new);
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -08004241 } else /* Another has populated it */
4242 pmd_free(mm, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004243#endif /* __ARCH_HAS_4LEVEL_HACK */
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08004244 spin_unlock(ptl);
Hugh Dickins1bb36302005-10-29 18:16:22 -07004245 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004246}
4247#endif /* __PAGETABLE_PMD_FOLDED */
4248
Ross Zwisler09796392017-01-10 16:57:21 -08004249static int __follow_pte_pmd(struct mm_struct *mm, unsigned long address,
Jérôme Glissea4d1a882017-08-31 17:17:26 -04004250 unsigned long *start, unsigned long *end,
4251 pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp)
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07004252{
4253 pgd_t *pgd;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03004254 p4d_t *p4d;
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07004255 pud_t *pud;
4256 pmd_t *pmd;
4257 pte_t *ptep;
4258
4259 pgd = pgd_offset(mm, address);
4260 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
4261 goto out;
4262
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03004263 p4d = p4d_offset(pgd, address);
4264 if (p4d_none(*p4d) || unlikely(p4d_bad(*p4d)))
4265 goto out;
4266
4267 pud = pud_offset(p4d, address);
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07004268 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
4269 goto out;
4270
4271 pmd = pmd_offset(pud, address);
Andrea Arcangelif66055ab2011-01-13 15:46:54 -08004272 VM_BUG_ON(pmd_trans_huge(*pmd));
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07004273
Ross Zwisler09796392017-01-10 16:57:21 -08004274 if (pmd_huge(*pmd)) {
4275 if (!pmdpp)
4276 goto out;
4277
Jérôme Glissea4d1a882017-08-31 17:17:26 -04004278 if (start && end) {
4279 *start = address & PMD_MASK;
4280 *end = *start + PMD_SIZE;
4281 mmu_notifier_invalidate_range_start(mm, *start, *end);
4282 }
Ross Zwisler09796392017-01-10 16:57:21 -08004283 *ptlp = pmd_lock(mm, pmd);
4284 if (pmd_huge(*pmd)) {
4285 *pmdpp = pmd;
4286 return 0;
4287 }
4288 spin_unlock(*ptlp);
Jérôme Glissea4d1a882017-08-31 17:17:26 -04004289 if (start && end)
4290 mmu_notifier_invalidate_range_end(mm, *start, *end);
Ross Zwisler09796392017-01-10 16:57:21 -08004291 }
4292
4293 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07004294 goto out;
4295
Jérôme Glissea4d1a882017-08-31 17:17:26 -04004296 if (start && end) {
4297 *start = address & PAGE_MASK;
4298 *end = *start + PAGE_SIZE;
4299 mmu_notifier_invalidate_range_start(mm, *start, *end);
4300 }
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07004301 ptep = pte_offset_map_lock(mm, pmd, address, ptlp);
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07004302 if (!pte_present(*ptep))
4303 goto unlock;
4304 *ptepp = ptep;
4305 return 0;
4306unlock:
4307 pte_unmap_unlock(ptep, *ptlp);
Jérôme Glissea4d1a882017-08-31 17:17:26 -04004308 if (start && end)
4309 mmu_notifier_invalidate_range_end(mm, *start, *end);
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07004310out:
4311 return -EINVAL;
4312}
4313
Ross Zwislerf729c8c2017-01-10 16:57:24 -08004314static inline int follow_pte(struct mm_struct *mm, unsigned long address,
4315 pte_t **ptepp, spinlock_t **ptlp)
Namhyung Kim1b36ba82010-10-26 14:22:00 -07004316{
4317 int res;
4318
4319 /* (void) is needed to make gcc happy */
4320 (void) __cond_lock(*ptlp,
Jérôme Glissea4d1a882017-08-31 17:17:26 -04004321 !(res = __follow_pte_pmd(mm, address, NULL, NULL,
4322 ptepp, NULL, ptlp)));
Namhyung Kim1b36ba82010-10-26 14:22:00 -07004323 return res;
4324}
4325
Ross Zwisler09796392017-01-10 16:57:21 -08004326int follow_pte_pmd(struct mm_struct *mm, unsigned long address,
Jérôme Glissea4d1a882017-08-31 17:17:26 -04004327 unsigned long *start, unsigned long *end,
Ross Zwisler09796392017-01-10 16:57:21 -08004328 pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp)
4329{
4330 int res;
4331
4332 /* (void) is needed to make gcc happy */
4333 (void) __cond_lock(*ptlp,
Jérôme Glissea4d1a882017-08-31 17:17:26 -04004334 !(res = __follow_pte_pmd(mm, address, start, end,
4335 ptepp, pmdpp, ptlp)));
Ross Zwisler09796392017-01-10 16:57:21 -08004336 return res;
4337}
4338EXPORT_SYMBOL(follow_pte_pmd);
4339
Johannes Weiner3b6748e2009-06-16 15:32:35 -07004340/**
4341 * follow_pfn - look up PFN at a user virtual address
4342 * @vma: memory mapping
4343 * @address: user virtual address
4344 * @pfn: location to store found PFN
4345 *
4346 * Only IO mappings and raw PFN mappings are allowed.
4347 *
4348 * Returns zero and the pfn at @pfn on success, -ve otherwise.
4349 */
4350int follow_pfn(struct vm_area_struct *vma, unsigned long address,
4351 unsigned long *pfn)
4352{
4353 int ret = -EINVAL;
4354 spinlock_t *ptl;
4355 pte_t *ptep;
4356
4357 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
4358 return ret;
4359
4360 ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);
4361 if (ret)
4362 return ret;
4363 *pfn = pte_pfn(*ptep);
4364 pte_unmap_unlock(ptep, ptl);
4365 return 0;
4366}
4367EXPORT_SYMBOL(follow_pfn);
4368
Rik van Riel28b2ee22008-07-23 21:27:05 -07004369#ifdef CONFIG_HAVE_IOREMAP_PROT
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08004370int follow_phys(struct vm_area_struct *vma,
4371 unsigned long address, unsigned int flags,
4372 unsigned long *prot, resource_size_t *phys)
Rik van Riel28b2ee22008-07-23 21:27:05 -07004373{
Johannes Weiner03668a42009-06-16 15:32:34 -07004374 int ret = -EINVAL;
Rik van Riel28b2ee22008-07-23 21:27:05 -07004375 pte_t *ptep, pte;
4376 spinlock_t *ptl;
Rik van Riel28b2ee22008-07-23 21:27:05 -07004377
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08004378 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
4379 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07004380
Johannes Weiner03668a42009-06-16 15:32:34 -07004381 if (follow_pte(vma->vm_mm, address, &ptep, &ptl))
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08004382 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07004383 pte = *ptep;
Johannes Weiner03668a42009-06-16 15:32:34 -07004384
Linus Torvaldsf6f37322017-12-15 18:53:22 -08004385 if ((flags & FOLL_WRITE) && !pte_write(pte))
Rik van Riel28b2ee22008-07-23 21:27:05 -07004386 goto unlock;
Rik van Riel28b2ee22008-07-23 21:27:05 -07004387
4388 *prot = pgprot_val(pte_pgprot(pte));
Johannes Weiner03668a42009-06-16 15:32:34 -07004389 *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
Rik van Riel28b2ee22008-07-23 21:27:05 -07004390
Johannes Weiner03668a42009-06-16 15:32:34 -07004391 ret = 0;
Rik van Riel28b2ee22008-07-23 21:27:05 -07004392unlock:
4393 pte_unmap_unlock(ptep, ptl);
4394out:
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08004395 return ret;
Rik van Riel28b2ee22008-07-23 21:27:05 -07004396}
4397
4398int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
4399 void *buf, int len, int write)
4400{
4401 resource_size_t phys_addr;
4402 unsigned long prot = 0;
KOSAKI Motohiro2bc72732009-01-06 14:39:43 -08004403 void __iomem *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07004404 int offset = addr & (PAGE_SIZE-1);
4405
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08004406 if (follow_phys(vma, addr, write, &prot, &phys_addr))
Rik van Riel28b2ee22008-07-23 21:27:05 -07004407 return -EINVAL;
4408
Grazvydas Ignotas9cb12d72015-02-12 15:00:19 -08004409 maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot);
jie@chenjie6@huwei.com24eee1e2018-08-10 17:23:06 -07004410 if (!maddr)
4411 return -ENOMEM;
4412
Rik van Riel28b2ee22008-07-23 21:27:05 -07004413 if (write)
4414 memcpy_toio(maddr + offset, buf, len);
4415 else
4416 memcpy_fromio(buf, maddr + offset, len);
4417 iounmap(maddr);
4418
4419 return len;
4420}
Uwe Kleine-König5a736332013-08-07 13:02:52 +02004421EXPORT_SYMBOL_GPL(generic_access_phys);
Rik van Riel28b2ee22008-07-23 21:27:05 -07004422#endif
4423
David Howells0ec76a12006-09-27 01:50:15 -07004424/*
Stephen Wilson206cb632011-03-13 15:49:19 -04004425 * Access another process' address space as given in mm. If non-NULL, use the
4426 * given task for page fault accounting.
David Howells0ec76a12006-09-27 01:50:15 -07004427 */
Eric W. Biederman84d77d32016-11-22 12:06:50 -06004428int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
Lorenzo Stoakes442486e2016-10-13 01:20:18 +01004429 unsigned long addr, void *buf, int len, unsigned int gup_flags)
David Howells0ec76a12006-09-27 01:50:15 -07004430{
David Howells0ec76a12006-09-27 01:50:15 -07004431 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07004432 void *old_buf = buf;
Lorenzo Stoakes442486e2016-10-13 01:20:18 +01004433 int write = gup_flags & FOLL_WRITE;
David Howells0ec76a12006-09-27 01:50:15 -07004434
David Howells0ec76a12006-09-27 01:50:15 -07004435 down_read(&mm->mmap_sem);
Simon Arlott183ff222007-10-20 01:27:18 +02004436 /* ignore errors, just check how much was successfully transferred */
David Howells0ec76a12006-09-27 01:50:15 -07004437 while (len) {
4438 int bytes, ret, offset;
4439 void *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07004440 struct page *page = NULL;
David Howells0ec76a12006-09-27 01:50:15 -07004441
Dave Hansen1e987792016-02-12 13:01:54 -08004442 ret = get_user_pages_remote(tsk, mm, addr, 1,
Lorenzo Stoakes5b56d492016-12-14 15:06:52 -08004443 gup_flags, &page, &vma, NULL);
Rik van Riel28b2ee22008-07-23 21:27:05 -07004444 if (ret <= 0) {
Rik van Rieldbffcd02014-08-06 16:08:12 -07004445#ifndef CONFIG_HAVE_IOREMAP_PROT
4446 break;
4447#else
Rik van Riel28b2ee22008-07-23 21:27:05 -07004448 /*
4449 * Check if this is a VM_IO | VM_PFNMAP VMA, which
4450 * we can access using slightly different code.
4451 */
Rik van Riel28b2ee22008-07-23 21:27:05 -07004452 vma = find_vma(mm, addr);
Michael Ellermanfe936df2011-04-14 15:22:10 -07004453 if (!vma || vma->vm_start > addr)
Rik van Riel28b2ee22008-07-23 21:27:05 -07004454 break;
4455 if (vma->vm_ops && vma->vm_ops->access)
4456 ret = vma->vm_ops->access(vma, addr, buf,
4457 len, write);
4458 if (ret <= 0)
Rik van Riel28b2ee22008-07-23 21:27:05 -07004459 break;
4460 bytes = ret;
Rik van Rieldbffcd02014-08-06 16:08:12 -07004461#endif
David Howells0ec76a12006-09-27 01:50:15 -07004462 } else {
Rik van Riel28b2ee22008-07-23 21:27:05 -07004463 bytes = len;
4464 offset = addr & (PAGE_SIZE-1);
4465 if (bytes > PAGE_SIZE-offset)
4466 bytes = PAGE_SIZE-offset;
4467
4468 maddr = kmap(page);
4469 if (write) {
4470 copy_to_user_page(vma, page, addr,
4471 maddr + offset, buf, bytes);
4472 set_page_dirty_lock(page);
4473 } else {
4474 copy_from_user_page(vma, page, addr,
4475 buf, maddr + offset, bytes);
4476 }
4477 kunmap(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004478 put_page(page);
David Howells0ec76a12006-09-27 01:50:15 -07004479 }
David Howells0ec76a12006-09-27 01:50:15 -07004480 len -= bytes;
4481 buf += bytes;
4482 addr += bytes;
4483 }
4484 up_read(&mm->mmap_sem);
David Howells0ec76a12006-09-27 01:50:15 -07004485
4486 return buf - old_buf;
4487}
Andi Kleen03252912008-01-30 13:33:18 +01004488
Stephen Wilson5ddd36b2011-03-13 15:49:20 -04004489/**
Randy Dunlapae91dbf2011-03-26 13:27:01 -07004490 * access_remote_vm - access another process' address space
Stephen Wilson5ddd36b2011-03-13 15:49:20 -04004491 * @mm: the mm_struct of the target address space
4492 * @addr: start address to access
4493 * @buf: source or destination buffer
4494 * @len: number of bytes to transfer
Lorenzo Stoakes6347e8d2016-10-13 01:20:19 +01004495 * @gup_flags: flags modifying lookup behaviour
Stephen Wilson5ddd36b2011-03-13 15:49:20 -04004496 *
4497 * The caller must hold a reference on @mm.
4498 */
4499int access_remote_vm(struct mm_struct *mm, unsigned long addr,
Lorenzo Stoakes6347e8d2016-10-13 01:20:19 +01004500 void *buf, int len, unsigned int gup_flags)
Stephen Wilson5ddd36b2011-03-13 15:49:20 -04004501{
Lorenzo Stoakes6347e8d2016-10-13 01:20:19 +01004502 return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags);
Stephen Wilson5ddd36b2011-03-13 15:49:20 -04004503}
4504
Andi Kleen03252912008-01-30 13:33:18 +01004505/*
Stephen Wilson206cb632011-03-13 15:49:19 -04004506 * Access another process' address space.
4507 * Source/target buffer must be kernel space,
4508 * Do not walk the page table directly, use get_user_pages
4509 */
4510int access_process_vm(struct task_struct *tsk, unsigned long addr,
Lorenzo Stoakesf307ab62016-10-13 01:20:20 +01004511 void *buf, int len, unsigned int gup_flags)
Stephen Wilson206cb632011-03-13 15:49:19 -04004512{
4513 struct mm_struct *mm;
4514 int ret;
4515
4516 mm = get_task_mm(tsk);
4517 if (!mm)
4518 return 0;
4519
Lorenzo Stoakesf307ab62016-10-13 01:20:20 +01004520 ret = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags);
Lorenzo Stoakes442486e2016-10-13 01:20:18 +01004521
Stephen Wilson206cb632011-03-13 15:49:19 -04004522 mmput(mm);
4523
4524 return ret;
4525}
Catalin Marinasfcd35852016-11-01 14:43:25 -07004526EXPORT_SYMBOL_GPL(access_process_vm);
Stephen Wilson206cb632011-03-13 15:49:19 -04004527
Andi Kleen03252912008-01-30 13:33:18 +01004528/*
4529 * Print the name of a VMA.
4530 */
4531void print_vma_addr(char *prefix, unsigned long ip)
4532{
4533 struct mm_struct *mm = current->mm;
4534 struct vm_area_struct *vma;
4535
Ingo Molnare8bff742008-02-13 20:21:06 +01004536 /*
Michal Hocko0a7f6822017-11-15 17:38:59 -08004537 * we might be running from an atomic context so we cannot sleep
Ingo Molnare8bff742008-02-13 20:21:06 +01004538 */
Michal Hocko0a7f6822017-11-15 17:38:59 -08004539 if (!down_read_trylock(&mm->mmap_sem))
Ingo Molnare8bff742008-02-13 20:21:06 +01004540 return;
4541
Andi Kleen03252912008-01-30 13:33:18 +01004542 vma = find_vma(mm, ip);
4543 if (vma && vma->vm_file) {
4544 struct file *f = vma->vm_file;
Michal Hocko0a7f6822017-11-15 17:38:59 -08004545 char *buf = (char *)__get_free_page(GFP_NOWAIT);
Andi Kleen03252912008-01-30 13:33:18 +01004546 if (buf) {
Andy Shevchenko2fbc57c2012-12-17 16:01:23 -08004547 char *p;
Andi Kleen03252912008-01-30 13:33:18 +01004548
Miklos Szeredi9bf39ab2015-06-19 10:29:13 +02004549 p = file_path(f, buf, PAGE_SIZE);
Andi Kleen03252912008-01-30 13:33:18 +01004550 if (IS_ERR(p))
4551 p = "?";
Andy Shevchenko2fbc57c2012-12-17 16:01:23 -08004552 printk("%s%s[%lx+%lx]", prefix, kbasename(p),
Andi Kleen03252912008-01-30 13:33:18 +01004553 vma->vm_start,
4554 vma->vm_end - vma->vm_start);
4555 free_page((unsigned long)buf);
4556 }
4557 }
Jeff Liu51a07e52012-07-31 16:43:18 -07004558 up_read(&mm->mmap_sem);
Andi Kleen03252912008-01-30 13:33:18 +01004559}
Nick Piggin3ee1afa2008-09-10 13:37:17 +02004560
Michael S. Tsirkin662bbcb2013-05-26 17:32:23 +03004561#if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP)
David Hildenbrand9ec23532015-05-11 17:52:07 +02004562void __might_fault(const char *file, int line)
Nick Piggin3ee1afa2008-09-10 13:37:17 +02004563{
Peter Zijlstra95156f02009-01-12 13:02:11 +01004564 /*
4565 * Some code (nfs/sunrpc) uses socket ops on kernel memory while
4566 * holding the mmap_sem, this is safe because kernel memory doesn't
4567 * get paged out, therefore we'll never actually fault, and the
4568 * below annotations will generate false positives.
4569 */
Al Virodb68ce12017-03-20 21:08:07 -04004570 if (uaccess_kernel())
Peter Zijlstra95156f02009-01-12 13:02:11 +01004571 return;
David Hildenbrand9ec23532015-05-11 17:52:07 +02004572 if (pagefault_disabled())
Michael S. Tsirkin662bbcb2013-05-26 17:32:23 +03004573 return;
David Hildenbrand9ec23532015-05-11 17:52:07 +02004574 __might_sleep(file, line, 0);
4575#if defined(CONFIG_DEBUG_ATOMIC_SLEEP)
Michael S. Tsirkin662bbcb2013-05-26 17:32:23 +03004576 if (current->mm)
Nick Piggin3ee1afa2008-09-10 13:37:17 +02004577 might_lock_read(&current->mm->mmap_sem);
David Hildenbrand9ec23532015-05-11 17:52:07 +02004578#endif
Nick Piggin3ee1afa2008-09-10 13:37:17 +02004579}
David Hildenbrand9ec23532015-05-11 17:52:07 +02004580EXPORT_SYMBOL(__might_fault);
Nick Piggin3ee1afa2008-09-10 13:37:17 +02004581#endif
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08004582
4583#if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)
Huang Yingc6ddfb62018-08-17 15:45:46 -07004584/*
4585 * Process all subpages of the specified huge page with the specified
4586 * operation. The target subpage will be processed last to keep its
4587 * cache lines hot.
4588 */
4589static inline void process_huge_page(
4590 unsigned long addr_hint, unsigned int pages_per_huge_page,
4591 void (*process_subpage)(unsigned long addr, int idx, void *arg),
4592 void *arg)
4593{
4594 int i, n, base, l;
4595 unsigned long addr = addr_hint &
4596 ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1);
4597
4598 /* Process target subpage last to keep its cache lines hot */
4599 might_sleep();
4600 n = (addr_hint - addr) / PAGE_SIZE;
4601 if (2 * n <= pages_per_huge_page) {
4602 /* If target subpage in first half of huge page */
4603 base = 0;
4604 l = n;
4605 /* Process subpages at the end of huge page */
4606 for (i = pages_per_huge_page - 1; i >= 2 * n; i--) {
4607 cond_resched();
4608 process_subpage(addr + i * PAGE_SIZE, i, arg);
4609 }
4610 } else {
4611 /* If target subpage in second half of huge page */
4612 base = pages_per_huge_page - 2 * (pages_per_huge_page - n);
4613 l = pages_per_huge_page - n;
4614 /* Process subpages at the begin of huge page */
4615 for (i = 0; i < base; i++) {
4616 cond_resched();
4617 process_subpage(addr + i * PAGE_SIZE, i, arg);
4618 }
4619 }
4620 /*
4621 * Process remaining subpages in left-right-left-right pattern
4622 * towards the target subpage
4623 */
4624 for (i = 0; i < l; i++) {
4625 int left_idx = base + i;
4626 int right_idx = base + 2 * l - 1 - i;
4627
4628 cond_resched();
4629 process_subpage(addr + left_idx * PAGE_SIZE, left_idx, arg);
4630 cond_resched();
4631 process_subpage(addr + right_idx * PAGE_SIZE, right_idx, arg);
4632 }
4633}
4634
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08004635static void clear_gigantic_page(struct page *page,
4636 unsigned long addr,
4637 unsigned int pages_per_huge_page)
4638{
4639 int i;
4640 struct page *p = page;
4641
4642 might_sleep();
4643 for (i = 0; i < pages_per_huge_page;
4644 i++, p = mem_map_next(p, page, i)) {
4645 cond_resched();
4646 clear_user_highpage(p, addr + i * PAGE_SIZE);
4647 }
4648}
Huang Yingc6ddfb62018-08-17 15:45:46 -07004649
4650static void clear_subpage(unsigned long addr, int idx, void *arg)
4651{
4652 struct page *page = arg;
4653
4654 clear_user_highpage(page + idx, addr);
4655}
4656
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08004657void clear_huge_page(struct page *page,
Huang Yingc79b57e2017-09-06 16:25:04 -07004658 unsigned long addr_hint, unsigned int pages_per_huge_page)
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08004659{
Huang Yingc79b57e2017-09-06 16:25:04 -07004660 unsigned long addr = addr_hint &
4661 ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1);
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08004662
4663 if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
4664 clear_gigantic_page(page, addr, pages_per_huge_page);
4665 return;
4666 }
4667
Huang Yingc6ddfb62018-08-17 15:45:46 -07004668 process_huge_page(addr_hint, pages_per_huge_page, clear_subpage, page);
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08004669}
4670
4671static void copy_user_gigantic_page(struct page *dst, struct page *src,
4672 unsigned long addr,
4673 struct vm_area_struct *vma,
4674 unsigned int pages_per_huge_page)
4675{
4676 int i;
4677 struct page *dst_base = dst;
4678 struct page *src_base = src;
4679
4680 for (i = 0; i < pages_per_huge_page; ) {
4681 cond_resched();
4682 copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma);
4683
4684 i++;
4685 dst = mem_map_next(dst, dst_base, i);
4686 src = mem_map_next(src, src_base, i);
4687 }
4688}
4689
Huang Yingc9f4cd72018-08-17 15:45:49 -07004690struct copy_subpage_arg {
4691 struct page *dst;
4692 struct page *src;
4693 struct vm_area_struct *vma;
4694};
4695
4696static void copy_subpage(unsigned long addr, int idx, void *arg)
4697{
4698 struct copy_subpage_arg *copy_arg = arg;
4699
4700 copy_user_highpage(copy_arg->dst + idx, copy_arg->src + idx,
4701 addr, copy_arg->vma);
4702}
4703
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08004704void copy_user_huge_page(struct page *dst, struct page *src,
Huang Yingc9f4cd72018-08-17 15:45:49 -07004705 unsigned long addr_hint, struct vm_area_struct *vma,
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08004706 unsigned int pages_per_huge_page)
4707{
Huang Yingc9f4cd72018-08-17 15:45:49 -07004708 unsigned long addr = addr_hint &
4709 ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1);
4710 struct copy_subpage_arg arg = {
4711 .dst = dst,
4712 .src = src,
4713 .vma = vma,
4714 };
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08004715
4716 if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
4717 copy_user_gigantic_page(dst, src, addr, vma,
4718 pages_per_huge_page);
4719 return;
4720 }
4721
Huang Yingc9f4cd72018-08-17 15:45:49 -07004722 process_huge_page(addr_hint, pages_per_huge_page, copy_subpage, &arg);
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08004723}
Mike Kravetzfa4d75c2017-02-22 15:42:49 -08004724
4725long copy_huge_page_from_user(struct page *dst_page,
4726 const void __user *usr_src,
Mike Kravetz810a56b2017-02-22 15:42:58 -08004727 unsigned int pages_per_huge_page,
4728 bool allow_pagefault)
Mike Kravetzfa4d75c2017-02-22 15:42:49 -08004729{
4730 void *src = (void *)usr_src;
4731 void *page_kaddr;
4732 unsigned long i, rc = 0;
4733 unsigned long ret_val = pages_per_huge_page * PAGE_SIZE;
4734
4735 for (i = 0; i < pages_per_huge_page; i++) {
Mike Kravetz810a56b2017-02-22 15:42:58 -08004736 if (allow_pagefault)
4737 page_kaddr = kmap(dst_page + i);
4738 else
4739 page_kaddr = kmap_atomic(dst_page + i);
Mike Kravetzfa4d75c2017-02-22 15:42:49 -08004740 rc = copy_from_user(page_kaddr,
4741 (const void __user *)(src + i * PAGE_SIZE),
4742 PAGE_SIZE);
Mike Kravetz810a56b2017-02-22 15:42:58 -08004743 if (allow_pagefault)
4744 kunmap(dst_page + i);
4745 else
4746 kunmap_atomic(page_kaddr);
Mike Kravetzfa4d75c2017-02-22 15:42:49 -08004747
4748 ret_val -= (PAGE_SIZE - rc);
4749 if (rc)
4750 break;
4751
4752 cond_resched();
4753 }
4754 return ret_val;
4755}
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08004756#endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08004757
Olof Johansson40b64ac2013-12-20 14:28:05 -08004758#if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS
Kirill A. Shutemovb35f1812014-01-21 15:49:07 -08004759
4760static struct kmem_cache *page_ptl_cachep;
4761
4762void __init ptlock_cache_init(void)
4763{
4764 page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0,
4765 SLAB_PANIC, NULL);
4766}
4767
Peter Zijlstra539edb52013-11-14 14:31:52 -08004768bool ptlock_alloc(struct page *page)
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08004769{
4770 spinlock_t *ptl;
4771
Kirill A. Shutemovb35f1812014-01-21 15:49:07 -08004772 ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL);
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08004773 if (!ptl)
4774 return false;
Peter Zijlstra539edb52013-11-14 14:31:52 -08004775 page->ptl = ptl;
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08004776 return true;
4777}
4778
Peter Zijlstra539edb52013-11-14 14:31:52 -08004779void ptlock_free(struct page *page)
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08004780{
Kirill A. Shutemovb35f1812014-01-21 15:49:07 -08004781 kmem_cache_free(page_ptl_cachep, page->ptl);
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08004782}
4783#endif