blob: 079eeac0b009790ac6c707eb65f05562f9ddcb86 [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/kallsyms.h>
63#include <linux/swapops.h>
64#include <linux/elf.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090065#include <linux/gfp.h>
Mel Gorman4daae3b2012-11-02 11:33:45 +000066#include <linux/migrate.h>
Andy Shevchenko2fbc57c2012-12-17 16:01:23 -080067#include <linux/string.h>
Dan Williams0abdd7a2014-01-21 15:48:12 -080068#include <linux/dma-debug.h>
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -070069#include <linux/debugfs.h>
Andrea Arcangeli6b251fc2015-09-04 15:46:20 -070070#include <linux/userfaultfd_k.h>
Jan Karabc2466e2016-05-12 18:29:19 +020071#include <linux/dax.h>
Michal Hocko6b31d592017-08-18 15:16:15 -070072#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070073
Alexey Dobriyan6952b612009-09-18 23:55:55 +040074#include <asm/io.h>
Dave Hansen33a709b2016-02-12 13:02:19 -080075#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070076#include <asm/pgalloc.h>
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080077#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070078#include <asm/tlb.h>
79#include <asm/tlbflush.h>
80#include <asm/pgtable.h>
81
Jan Beulich42b77722008-07-23 21:27:10 -070082#include "internal.h"
83
Peter Zijlstra90572892013-10-07 11:29:20 +010084#ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
85#warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid.
Peter Zijlstra75980e92013-02-22 16:34:32 -080086#endif
87
Andy Whitcroftd41dee32005-06-23 00:07:54 -070088#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -070089/* use the per-pgdat data instead for discontigmem - mbligh */
90unsigned long max_mapnr;
Linus Torvalds1da177e2005-04-16 15:20:36 -070091EXPORT_SYMBOL(max_mapnr);
Tobin C Harding166f61b2017-02-24 14:59:01 -080092
93struct page *mem_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -070094EXPORT_SYMBOL(mem_map);
95#endif
96
Linus Torvalds1da177e2005-04-16 15:20:36 -070097/*
98 * A number of key systems in x86 including ioremap() rely on the assumption
99 * that high_memory defines the upper bound on direct map memory, then end
100 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
101 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
102 * and ZONE_HIGHMEM.
103 */
Tobin C Harding166f61b2017-02-24 14:59:01 -0800104void *high_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106
Ingo Molnar32a93232008-02-06 22:39:44 +0100107/*
108 * Randomize the address space (stacks, mmaps, brk, etc.).
109 *
110 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
111 * as ancient (libc5 based) binaries can segfault. )
112 */
113int randomize_va_space __read_mostly =
114#ifdef CONFIG_COMPAT_BRK
115 1;
116#else
117 2;
118#endif
Andi Kleena62eaf12006-02-16 23:41:58 +0100119
120static int __init disable_randmaps(char *s)
121{
122 randomize_va_space = 0;
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800123 return 1;
Andi Kleena62eaf12006-02-16 23:41:58 +0100124}
125__setup("norandmaps", disable_randmaps);
126
Hugh Dickins62eede62009-09-21 17:03:34 -0700127unsigned long zero_pfn __read_mostly;
Ard Biesheuvel0b700682014-09-12 22:17:23 +0200128EXPORT_SYMBOL(zero_pfn);
129
Tobin C Harding166f61b2017-02-24 14:59:01 -0800130unsigned long highest_memmap_pfn __read_mostly;
131
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -0700132/*
133 * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init()
134 */
135static int __init init_zero_pfn(void)
136{
137 zero_pfn = page_to_pfn(ZERO_PAGE(0));
138 return 0;
139}
140core_initcall(init_zero_pfn);
Andi Kleena62eaf12006-02-16 23:41:58 +0100141
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800142
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800143#if defined(SPLIT_RSS_COUNTING)
144
David Rientjesea48cf72012-03-21 16:34:13 -0700145void sync_mm_rss(struct mm_struct *mm)
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800146{
147 int i;
148
149 for (i = 0; i < NR_MM_COUNTERS; i++) {
David Rientjes05af2e12012-03-21 16:34:13 -0700150 if (current->rss_stat.count[i]) {
151 add_mm_counter(mm, i, current->rss_stat.count[i]);
152 current->rss_stat.count[i] = 0;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800153 }
154 }
David Rientjes05af2e12012-03-21 16:34:13 -0700155 current->rss_stat.events = 0;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800156}
157
158static void add_mm_counter_fast(struct mm_struct *mm, int member, int val)
159{
160 struct task_struct *task = current;
161
162 if (likely(task->mm == mm))
163 task->rss_stat.count[member] += val;
164 else
165 add_mm_counter(mm, member, val);
166}
167#define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1)
168#define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1)
169
170/* sync counter once per 64 page faults */
171#define TASK_RSS_EVENTS_THRESH (64)
172static void check_sync_rss_stat(struct task_struct *task)
173{
174 if (unlikely(task != current))
175 return;
176 if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH))
David Rientjesea48cf72012-03-21 16:34:13 -0700177 sync_mm_rss(task->mm);
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800178}
Peter Zijlstra9547d012011-05-24 17:12:14 -0700179#else /* SPLIT_RSS_COUNTING */
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800180
181#define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member)
182#define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member)
183
184static void check_sync_rss_stat(struct task_struct *task)
185{
186}
187
Peter Zijlstra9547d012011-05-24 17:12:14 -0700188#endif /* SPLIT_RSS_COUNTING */
189
190#ifdef HAVE_GENERIC_MMU_GATHER
191
Nicholas Krauseca1d6c72015-09-04 15:48:22 -0700192static bool tlb_next_batch(struct mmu_gather *tlb)
Peter Zijlstra9547d012011-05-24 17:12:14 -0700193{
194 struct mmu_gather_batch *batch;
195
196 batch = tlb->active;
197 if (batch->next) {
198 tlb->active = batch->next;
Nicholas Krauseca1d6c72015-09-04 15:48:22 -0700199 return true;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700200 }
201
Michal Hocko53a59fc2013-01-04 15:35:12 -0800202 if (tlb->batch_count == MAX_GATHER_BATCH_COUNT)
Nicholas Krauseca1d6c72015-09-04 15:48:22 -0700203 return false;
Michal Hocko53a59fc2013-01-04 15:35:12 -0800204
Peter Zijlstra9547d012011-05-24 17:12:14 -0700205 batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0);
206 if (!batch)
Nicholas Krauseca1d6c72015-09-04 15:48:22 -0700207 return false;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700208
Michal Hocko53a59fc2013-01-04 15:35:12 -0800209 tlb->batch_count++;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700210 batch->next = NULL;
211 batch->nr = 0;
212 batch->max = MAX_GATHER_BATCH;
213
214 tlb->active->next = batch;
215 tlb->active = batch;
216
Nicholas Krauseca1d6c72015-09-04 15:48:22 -0700217 return true;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700218}
219
Minchan Kim56236a52017-08-10 15:24:05 -0700220void arch_tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm,
221 unsigned long start, unsigned long end)
Peter Zijlstra9547d012011-05-24 17:12:14 -0700222{
223 tlb->mm = mm;
224
Linus Torvalds2b047252013-08-15 11:42:25 -0700225 /* Is it from 0 to ~0? */
226 tlb->fullmm = !(start | (end+1));
Dave Hansen1de14c32013-04-12 16:23:54 -0700227 tlb->need_flush_all = 0;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700228 tlb->local.next = NULL;
229 tlb->local.nr = 0;
230 tlb->local.max = ARRAY_SIZE(tlb->__pages);
231 tlb->active = &tlb->local;
Michal Hocko53a59fc2013-01-04 15:35:12 -0800232 tlb->batch_count = 0;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700233
234#ifdef CONFIG_HAVE_RCU_TABLE_FREE
235 tlb->batch = NULL;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800236#endif
Aneesh Kumar K.Ve77b0852016-07-26 15:24:12 -0700237 tlb->page_size = 0;
Will Deaconfb7332a2014-10-29 10:03:09 +0000238
239 __tlb_reset_range(tlb);
Peter Zijlstra9547d012011-05-24 17:12:14 -0700240}
241
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700242static void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb)
Peter Zijlstra9547d012011-05-24 17:12:14 -0700243{
Will Deacon721c21c2015-01-12 19:10:55 +0000244 if (!tlb->end)
245 return;
246
Peter Zijlstra9547d012011-05-24 17:12:14 -0700247 tlb_flush(tlb);
Joerg Roedel34ee6452014-11-13 13:46:09 +1100248 mmu_notifier_invalidate_range(tlb->mm, tlb->start, tlb->end);
Peter Zijlstra9547d012011-05-24 17:12:14 -0700249#ifdef CONFIG_HAVE_RCU_TABLE_FREE
250 tlb_table_flush(tlb);
251#endif
Will Deaconfb7332a2014-10-29 10:03:09 +0000252 __tlb_reset_range(tlb);
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700253}
254
255static void tlb_flush_mmu_free(struct mmu_gather *tlb)
256{
257 struct mmu_gather_batch *batch;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700258
Will Deacon721c21c2015-01-12 19:10:55 +0000259 for (batch = &tlb->local; batch && batch->nr; batch = batch->next) {
Peter Zijlstra9547d012011-05-24 17:12:14 -0700260 free_pages_and_swap_cache(batch->pages, batch->nr);
261 batch->nr = 0;
262 }
263 tlb->active = &tlb->local;
264}
265
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700266void tlb_flush_mmu(struct mmu_gather *tlb)
267{
Linus Torvalds1cf35d42014-04-25 16:05:40 -0700268 tlb_flush_mmu_tlbonly(tlb);
269 tlb_flush_mmu_free(tlb);
270}
271
Peter Zijlstra9547d012011-05-24 17:12:14 -0700272/* tlb_finish_mmu
273 * Called at the end of the shootdown operation to free up any resources
274 * that were required.
275 */
Minchan Kim56236a52017-08-10 15:24:05 -0700276void arch_tlb_finish_mmu(struct mmu_gather *tlb,
Minchan Kim99baac22017-08-10 15:24:12 -0700277 unsigned long start, unsigned long end, bool force)
Peter Zijlstra9547d012011-05-24 17:12:14 -0700278{
279 struct mmu_gather_batch *batch, *next;
280
Minchan Kim99baac22017-08-10 15:24:12 -0700281 if (force)
282 __tlb_adjust_range(tlb, start, end - start);
283
Peter Zijlstra9547d012011-05-24 17:12:14 -0700284 tlb_flush_mmu(tlb);
285
286 /* keep the page table cache within bounds */
287 check_pgt_cache();
288
289 for (batch = tlb->local.next; batch; batch = next) {
290 next = batch->next;
291 free_pages((unsigned long)batch, 0);
292 }
293 tlb->local.next = NULL;
294}
295
296/* __tlb_remove_page
297 * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while
298 * handling the additional races in SMP caused by other CPUs caching valid
299 * mappings in their TLBs. Returns the number of free page slots left.
300 * When out of page slots we must call tlb_flush_mmu().
Aneesh Kumar K.Ve9d55e12016-07-26 15:24:09 -0700301 *returns true if the caller should flush.
Peter Zijlstra9547d012011-05-24 17:12:14 -0700302 */
Aneesh Kumar K.Ve77b0852016-07-26 15:24:12 -0700303bool __tlb_remove_page_size(struct mmu_gather *tlb, struct page *page, int page_size)
Peter Zijlstra9547d012011-05-24 17:12:14 -0700304{
305 struct mmu_gather_batch *batch;
306
Will Deaconfb7332a2014-10-29 10:03:09 +0000307 VM_BUG_ON(!tlb->end);
Aneesh Kumar K.V692a68c2016-12-12 16:42:43 -0800308 VM_WARN_ON(tlb->page_size != page_size);
Aneesh Kumar K.Ve77b0852016-07-26 15:24:12 -0700309
Peter Zijlstra9547d012011-05-24 17:12:14 -0700310 batch = tlb->active;
Aneesh Kumar K.V692a68c2016-12-12 16:42:43 -0800311 /*
312 * Add the page and check if we are full. If so
313 * force a flush.
314 */
315 batch->pages[batch->nr++] = page;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700316 if (batch->nr == batch->max) {
317 if (!tlb_next_batch(tlb))
Aneesh Kumar K.Ve9d55e12016-07-26 15:24:09 -0700318 return true;
Shaohua Li0b43c3a2011-07-08 15:39:41 -0700319 batch = tlb->active;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700320 }
Sasha Levin309381fea2014-01-23 15:52:54 -0800321 VM_BUG_ON_PAGE(batch->nr > batch->max, page);
Peter Zijlstra9547d012011-05-24 17:12:14 -0700322
Aneesh Kumar K.Ve9d55e12016-07-26 15:24:09 -0700323 return false;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700324}
325
326#endif /* HAVE_GENERIC_MMU_GATHER */
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800327
Peter Zijlstra26723912011-05-24 17:12:00 -0700328#ifdef CONFIG_HAVE_RCU_TABLE_FREE
329
330/*
331 * See the comment near struct mmu_table_batch.
332 */
333
334static void tlb_remove_table_smp_sync(void *arg)
335{
336 /* Simply deliver the interrupt */
337}
338
339static void tlb_remove_table_one(void *table)
340{
341 /*
342 * This isn't an RCU grace period and hence the page-tables cannot be
343 * assumed to be actually RCU-freed.
344 *
345 * It is however sufficient for software page-table walkers that rely on
346 * IRQ disabling. See the comment near struct mmu_table_batch.
347 */
348 smp_call_function(tlb_remove_table_smp_sync, NULL, 1);
349 __tlb_remove_table(table);
350}
351
352static void tlb_remove_table_rcu(struct rcu_head *head)
353{
354 struct mmu_table_batch *batch;
355 int i;
356
357 batch = container_of(head, struct mmu_table_batch, rcu);
358
359 for (i = 0; i < batch->nr; i++)
360 __tlb_remove_table(batch->tables[i]);
361
362 free_page((unsigned long)batch);
363}
364
365void tlb_table_flush(struct mmu_gather *tlb)
366{
367 struct mmu_table_batch **batch = &tlb->batch;
368
369 if (*batch) {
370 call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu);
371 *batch = NULL;
372 }
373}
374
375void tlb_remove_table(struct mmu_gather *tlb, void *table)
376{
377 struct mmu_table_batch **batch = &tlb->batch;
378
Peter Zijlstra26723912011-05-24 17:12:00 -0700379 /*
380 * When there's less then two users of this mm there cannot be a
381 * concurrent page-table walk.
382 */
383 if (atomic_read(&tlb->mm->mm_users) < 2) {
384 __tlb_remove_table(table);
385 return;
386 }
387
388 if (*batch == NULL) {
389 *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN);
390 if (*batch == NULL) {
391 tlb_remove_table_one(table);
392 return;
393 }
394 (*batch)->nr = 0;
395 }
396 (*batch)->tables[(*batch)->nr++] = table;
397 if ((*batch)->nr == MAX_TABLE_BATCH)
398 tlb_table_flush(tlb);
399}
400
Peter Zijlstra9547d012011-05-24 17:12:14 -0700401#endif /* CONFIG_HAVE_RCU_TABLE_FREE */
Peter Zijlstra26723912011-05-24 17:12:00 -0700402
Minchan Kim56236a52017-08-10 15:24:05 -0700403/* tlb_gather_mmu
404 * Called to initialize an (on-stack) mmu_gather structure for page-table
405 * tear-down from @mm. The @fullmm argument is used when @mm is without
406 * users and we're going to destroy the full address space (exit/execve).
407 */
408void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm,
409 unsigned long start, unsigned long end)
410{
411 arch_tlb_gather_mmu(tlb, mm, start, end);
Minchan Kim99baac22017-08-10 15:24:12 -0700412 inc_tlb_flush_pending(tlb->mm);
Minchan Kim56236a52017-08-10 15:24:05 -0700413}
414
415void tlb_finish_mmu(struct mmu_gather *tlb,
416 unsigned long start, unsigned long end)
417{
Minchan Kim99baac22017-08-10 15:24:12 -0700418 /*
419 * If there are parallel threads are doing PTE changes on same range
420 * under non-exclusive lock(e.g., mmap_sem read-side) but defer TLB
421 * flush by batching, a thread has stable TLB entry can fail to flush
422 * the TLB by observing pte_none|!pte_dirty, for example so flush TLB
423 * forcefully if we detect parallel PTE batching threads.
424 */
425 bool force = mm_tlb_flush_nested(tlb->mm);
426
427 arch_tlb_finish_mmu(tlb, start, end, force);
428 dec_tlb_flush_pending(tlb->mm);
Minchan Kim56236a52017-08-10 15:24:05 -0700429}
430
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 * Note: this doesn't free the actual pages themselves. That
433 * has been handled earlier when unmapping all the memory regions.
434 */
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000435static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
436 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800438 pgtable_t token = pmd_pgtable(*pmd);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700439 pmd_clear(pmd);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000440 pte_free_tlb(tlb, token, addr);
Kirill A. Shutemove1f56c82013-11-14 14:30:48 -0800441 atomic_long_dec(&tlb->mm->nr_ptes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442}
443
Hugh Dickinse0da3822005-04-19 13:29:15 -0700444static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
445 unsigned long addr, unsigned long end,
446 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447{
448 pmd_t *pmd;
449 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700450 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
Hugh Dickinse0da3822005-04-19 13:29:15 -0700452 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 do {
455 next = pmd_addr_end(addr, end);
456 if (pmd_none_or_clear_bad(pmd))
457 continue;
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000458 free_pte_range(tlb, pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 } while (pmd++, addr = next, addr != end);
460
Hugh Dickinse0da3822005-04-19 13:29:15 -0700461 start &= PUD_MASK;
462 if (start < floor)
463 return;
464 if (ceiling) {
465 ceiling &= PUD_MASK;
466 if (!ceiling)
467 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700469 if (end - 1 > ceiling - 1)
470 return;
471
472 pmd = pmd_offset(pud, start);
473 pud_clear(pud);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000474 pmd_free_tlb(tlb, pmd, start);
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -0800475 mm_dec_nr_pmds(tlb->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476}
477
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300478static inline void free_pud_range(struct mmu_gather *tlb, p4d_t *p4d,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700479 unsigned long addr, unsigned long end,
480 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481{
482 pud_t *pud;
483 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700484 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485
Hugh Dickinse0da3822005-04-19 13:29:15 -0700486 start = addr;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300487 pud = pud_offset(p4d, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 do {
489 next = pud_addr_end(addr, end);
490 if (pud_none_or_clear_bad(pud))
491 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700492 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 } while (pud++, addr = next, addr != end);
494
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300495 start &= P4D_MASK;
496 if (start < floor)
497 return;
498 if (ceiling) {
499 ceiling &= P4D_MASK;
500 if (!ceiling)
501 return;
502 }
503 if (end - 1 > ceiling - 1)
504 return;
505
506 pud = pud_offset(p4d, start);
507 p4d_clear(p4d);
508 pud_free_tlb(tlb, pud, start);
509}
510
511static inline void free_p4d_range(struct mmu_gather *tlb, pgd_t *pgd,
512 unsigned long addr, unsigned long end,
513 unsigned long floor, unsigned long ceiling)
514{
515 p4d_t *p4d;
516 unsigned long next;
517 unsigned long start;
518
519 start = addr;
520 p4d = p4d_offset(pgd, addr);
521 do {
522 next = p4d_addr_end(addr, end);
523 if (p4d_none_or_clear_bad(p4d))
524 continue;
525 free_pud_range(tlb, p4d, addr, next, floor, ceiling);
526 } while (p4d++, addr = next, addr != end);
527
Hugh Dickinse0da3822005-04-19 13:29:15 -0700528 start &= PGDIR_MASK;
529 if (start < floor)
530 return;
531 if (ceiling) {
532 ceiling &= PGDIR_MASK;
533 if (!ceiling)
534 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700536 if (end - 1 > ceiling - 1)
537 return;
538
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300539 p4d = p4d_offset(pgd, start);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700540 pgd_clear(pgd);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300541 p4d_free_tlb(tlb, p4d, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542}
543
544/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700545 * This function frees user-level page tables of a process.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 */
Jan Beulich42b77722008-07-23 21:27:10 -0700547void free_pgd_range(struct mmu_gather *tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700548 unsigned long addr, unsigned long end,
549 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550{
551 pgd_t *pgd;
552 unsigned long next;
553
Hugh Dickinse0da3822005-04-19 13:29:15 -0700554 /*
555 * The next few lines have given us lots of grief...
556 *
557 * Why are we testing PMD* at this top level? Because often
558 * there will be no work to do at all, and we'd prefer not to
559 * go all the way down to the bottom just to discover that.
560 *
561 * Why all these "- 1"s? Because 0 represents both the bottom
562 * of the address space and the top of it (using -1 for the
563 * top wouldn't help much: the masks would do the wrong thing).
564 * The rule is that addr 0 and floor 0 refer to the bottom of
565 * the address space, but end 0 and ceiling 0 refer to the top
566 * Comparisons need to use "end - 1" and "ceiling - 1" (though
567 * that end 0 case should be mythical).
568 *
569 * Wherever addr is brought up or ceiling brought down, we must
570 * be careful to reject "the opposite 0" before it confuses the
571 * subsequent tests. But what about where end is brought down
572 * by PMD_SIZE below? no, end can't go down to 0 there.
573 *
574 * Whereas we round start (addr) and ceiling down, by different
575 * masks at different levels, in order to test whether a table
576 * now has no other vmas using it, so can be freed, we don't
577 * bother to round floor or end up - the tests don't need that.
578 */
579
580 addr &= PMD_MASK;
581 if (addr < floor) {
582 addr += PMD_SIZE;
583 if (!addr)
584 return;
585 }
586 if (ceiling) {
587 ceiling &= PMD_MASK;
588 if (!ceiling)
589 return;
590 }
591 if (end - 1 > ceiling - 1)
592 end -= PMD_SIZE;
593 if (addr > end - 1)
594 return;
Aneesh Kumar K.V07e32662016-12-12 16:42:40 -0800595 /*
596 * We add page table cache pages with PAGE_SIZE,
597 * (see pte_free_tlb()), flush the tlb if we need
598 */
599 tlb_remove_check_page_size_change(tlb, PAGE_SIZE);
Jan Beulich42b77722008-07-23 21:27:10 -0700600 pgd = pgd_offset(tlb->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 do {
602 next = pgd_addr_end(addr, end);
603 if (pgd_none_or_clear_bad(pgd))
604 continue;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300605 free_p4d_range(tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700607}
608
Jan Beulich42b77722008-07-23 21:27:10 -0700609void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700610 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700611{
612 while (vma) {
613 struct vm_area_struct *next = vma->vm_next;
614 unsigned long addr = vma->vm_start;
615
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700616 /*
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000617 * Hide vma from rmap and truncate_pagecache before freeing
618 * pgtables
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700619 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800620 unlink_anon_vmas(vma);
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700621 unlink_file_vma(vma);
622
David Gibson9da61ae2006-03-22 00:08:57 -0800623 if (is_vm_hugetlb_page(vma)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700624 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Tobin C Harding166f61b2017-02-24 14:59:01 -0800625 floor, next ? next->vm_start : ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700626 } else {
627 /*
628 * Optimization: gather nearby vmas into one call down
629 */
630 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
David Gibson48669202006-03-22 00:08:58 -0800631 && !is_vm_hugetlb_page(next)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700632 vma = next;
633 next = vma->vm_next;
Rik van Riel5beb4932010-03-05 13:42:07 -0800634 unlink_anon_vmas(vma);
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700635 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700636 }
637 free_pgd_range(tlb, addr, vma->vm_end,
Tobin C Harding166f61b2017-02-24 14:59:01 -0800638 floor, next ? next->vm_start : ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700639 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700640 vma = next;
641 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642}
643
Kirill A. Shutemov3ed3a4f2016-03-17 14:19:11 -0700644int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645{
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -0800646 spinlock_t *ptl;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800647 pgtable_t new = pte_alloc_one(mm, address);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700648 if (!new)
649 return -ENOMEM;
650
Nick Piggin362a61a2008-05-14 06:37:36 +0200651 /*
652 * Ensure all pte setup (eg. pte page lock and page clearing) are
653 * visible before the pte is made visible to other CPUs by being
654 * put into page tables.
655 *
656 * The other side of the story is the pointer chasing in the page
657 * table walking code (when walking the page table without locking;
658 * ie. most of the time). Fortunately, these data accesses consist
659 * of a chain of data-dependent loads, meaning most CPUs (alpha
660 * being the notable exception) will already guarantee loads are
661 * seen in-order. See the alpha page table accessors for the
662 * smp_read_barrier_depends() barriers in page table walking code.
663 */
664 smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
665
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -0800666 ptl = pmd_lock(mm, pmd);
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800667 if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
Kirill A. Shutemove1f56c82013-11-14 14:30:48 -0800668 atomic_long_inc(&mm->nr_ptes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 pmd_populate(mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800670 new = NULL;
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800671 }
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -0800672 spin_unlock(ptl);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800673 if (new)
674 pte_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700675 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676}
677
Hugh Dickins1bb36302005-10-29 18:16:22 -0700678int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700680 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
681 if (!new)
682 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683
Nick Piggin362a61a2008-05-14 06:37:36 +0200684 smp_wmb(); /* See comment in __pte_alloc */
685
Hugh Dickins1bb36302005-10-29 18:16:22 -0700686 spin_lock(&init_mm.page_table_lock);
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800687 if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
Hugh Dickins1bb36302005-10-29 18:16:22 -0700688 pmd_populate_kernel(&init_mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800689 new = NULL;
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800690 }
Hugh Dickins1bb36302005-10-29 18:16:22 -0700691 spin_unlock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800692 if (new)
693 pte_free_kernel(&init_mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700694 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695}
696
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800697static inline void init_rss_vec(int *rss)
Hugh Dickinsae859762005-10-29 18:16:05 -0700698{
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800699 memset(rss, 0, sizeof(int) * NR_MM_COUNTERS);
700}
701
702static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss)
703{
704 int i;
705
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800706 if (current->mm == mm)
David Rientjes05af2e12012-03-21 16:34:13 -0700707 sync_mm_rss(mm);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800708 for (i = 0; i < NR_MM_COUNTERS; i++)
709 if (rss[i])
710 add_mm_counter(mm, i, rss[i]);
Hugh Dickinsae859762005-10-29 18:16:05 -0700711}
712
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800714 * This function is called to print an error when a bad pte
715 * is found. For example, we might have a PFN-mapped pte in
716 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700717 *
718 * The calling function must still handle the error.
719 */
Hugh Dickins3dc14742009-01-06 14:40:08 -0800720static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
721 pte_t pte, struct page *page)
Nick Pigginb5810032005-10-29 18:16:12 -0700722{
Hugh Dickins3dc14742009-01-06 14:40:08 -0800723 pgd_t *pgd = pgd_offset(vma->vm_mm, addr);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300724 p4d_t *p4d = p4d_offset(pgd, addr);
725 pud_t *pud = pud_offset(p4d, addr);
Hugh Dickins3dc14742009-01-06 14:40:08 -0800726 pmd_t *pmd = pmd_offset(pud, addr);
727 struct address_space *mapping;
728 pgoff_t index;
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800729 static unsigned long resume;
730 static unsigned long nr_shown;
731 static unsigned long nr_unshown;
732
733 /*
734 * Allow a burst of 60 reports, then keep quiet for that minute;
735 * or allow a steady drip of one report per second.
736 */
737 if (nr_shown == 60) {
738 if (time_before(jiffies, resume)) {
739 nr_unshown++;
740 return;
741 }
742 if (nr_unshown) {
Joe Perches11705322016-03-17 14:19:50 -0700743 pr_alert("BUG: Bad page map: %lu messages suppressed\n",
744 nr_unshown);
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800745 nr_unshown = 0;
746 }
747 nr_shown = 0;
748 }
749 if (nr_shown++ == 0)
750 resume = jiffies + 60 * HZ;
Hugh Dickins3dc14742009-01-06 14:40:08 -0800751
752 mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;
753 index = linear_page_index(vma, addr);
754
Joe Perches11705322016-03-17 14:19:50 -0700755 pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
756 current->comm,
757 (long long)pte_val(pte), (long long)pmd_val(*pmd));
Wu Fengguang718a3822010-03-10 15:20:43 -0800758 if (page)
Dave Hansenf0b791a2014-01-23 15:52:49 -0800759 dump_page(page, "bad pte");
Joe Perches11705322016-03-17 14:19:50 -0700760 pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
761 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
Hugh Dickins3dc14742009-01-06 14:40:08 -0800762 /*
763 * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y
764 */
Konstantin Khlebnikov26825822015-04-15 16:15:08 -0700765 pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n",
766 vma->vm_file,
767 vma->vm_ops ? vma->vm_ops->fault : NULL,
768 vma->vm_file ? vma->vm_file->f_op->mmap : NULL,
769 mapping ? mapping->a_ops->readpage : NULL);
Nick Pigginb5810032005-10-29 18:16:12 -0700770 dump_stack();
Rusty Russell373d4d02013-01-21 17:17:39 +1030771 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Nick Pigginb5810032005-10-29 18:16:12 -0700772}
773
774/*
Nick Piggin7e675132008-04-28 02:13:00 -0700775 * vm_normal_page -- This function gets the "struct page" associated with a pte.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800776 *
Nick Piggin7e675132008-04-28 02:13:00 -0700777 * "Special" mappings do not wish to be associated with a "struct page" (either
778 * it doesn't exist, or it exists but they don't want to touch it). In this
779 * case, NULL is returned here. "Normal" mappings do have a struct page.
Jared Hulbertb379d792008-04-28 02:12:58 -0700780 *
Nick Piggin7e675132008-04-28 02:13:00 -0700781 * There are 2 broad cases. Firstly, an architecture may define a pte_special()
782 * pte bit, in which case this function is trivial. Secondly, an architecture
783 * may not have a spare pte bit, which requires a more complicated scheme,
784 * described below.
785 *
786 * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
787 * special mapping (even if there are underlying and valid "struct pages").
788 * COWed pages of a VM_PFNMAP are always normal.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800789 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700790 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
791 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
Nick Piggin7e675132008-04-28 02:13:00 -0700792 * set, and the vm_pgoff will point to the first PFN mapped: thus every special
793 * mapping will always honor the rule
Linus Torvalds6aab3412005-11-28 14:34:23 -0800794 *
795 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
796 *
Nick Piggin7e675132008-04-28 02:13:00 -0700797 * And for normal mappings this is false.
Jared Hulbertb379d792008-04-28 02:12:58 -0700798 *
Nick Piggin7e675132008-04-28 02:13:00 -0700799 * This restricts such mappings to be a linear translation from virtual address
800 * to pfn. To get around this restriction, we allow arbitrary mappings so long
801 * as the vma is not a COW mapping; in that case, we know that all ptes are
802 * special (because none can have been COWed).
Jared Hulbertb379d792008-04-28 02:12:58 -0700803 *
804 *
Nick Piggin7e675132008-04-28 02:13:00 -0700805 * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
806 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700807 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
808 * page" backing, however the difference is that _all_ pages with a struct
809 * page (that is, those where pfn_valid is true) are refcounted and considered
810 * normal pages by the VM. The disadvantage is that pages are refcounted
811 * (which can be slower and simply not an option for some PFNMAP users). The
812 * advantage is that we don't have to follow the strict linearity rule of
813 * PFNMAP mappings in order to support COWable mappings.
814 *
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800815 */
Nick Piggin7e675132008-04-28 02:13:00 -0700816#ifdef __HAVE_ARCH_PTE_SPECIAL
817# define HAVE_PTE_SPECIAL 1
818#else
819# define HAVE_PTE_SPECIAL 0
820#endif
821struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
822 pte_t pte)
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800823{
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800824 unsigned long pfn = pte_pfn(pte);
Nick Piggin7e675132008-04-28 02:13:00 -0700825
826 if (HAVE_PTE_SPECIAL) {
Hugh Dickinsb38af472014-08-29 15:18:44 -0700827 if (likely(!pte_special(pte)))
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800828 goto check_pfn;
David Vrabel667a0a02014-12-18 14:48:15 +0000829 if (vma->vm_ops && vma->vm_ops->find_special_page)
830 return vma->vm_ops->find_special_page(vma, addr);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -0700831 if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))
832 return NULL;
Hugh Dickins62eede62009-09-21 17:03:34 -0700833 if (!is_zero_pfn(pfn))
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800834 print_bad_pte(vma, addr, pte, NULL);
Nick Piggin7e675132008-04-28 02:13:00 -0700835 return NULL;
836 }
837
838 /* !HAVE_PTE_SPECIAL case follows: */
839
Jared Hulbertb379d792008-04-28 02:12:58 -0700840 if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
841 if (vma->vm_flags & VM_MIXEDMAP) {
842 if (!pfn_valid(pfn))
843 return NULL;
844 goto out;
845 } else {
Nick Piggin7e675132008-04-28 02:13:00 -0700846 unsigned long off;
847 off = (addr - vma->vm_start) >> PAGE_SHIFT;
Jared Hulbertb379d792008-04-28 02:12:58 -0700848 if (pfn == vma->vm_pgoff + off)
849 return NULL;
850 if (!is_cow_mapping(vma->vm_flags))
851 return NULL;
852 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800853 }
854
Hugh Dickinsb38af472014-08-29 15:18:44 -0700855 if (is_zero_pfn(pfn))
856 return NULL;
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800857check_pfn:
858 if (unlikely(pfn > highest_memmap_pfn)) {
859 print_bad_pte(vma, addr, pte, NULL);
860 return NULL;
861 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800862
863 /*
Nick Piggin7e675132008-04-28 02:13:00 -0700864 * NOTE! We still have PageReserved() pages in the page tables.
Nick Piggin7e675132008-04-28 02:13:00 -0700865 * eg. VDSO mappings can cause them to exist.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800866 */
Jared Hulbertb379d792008-04-28 02:12:58 -0700867out:
Linus Torvalds6aab3412005-11-28 14:34:23 -0800868 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800869}
870
Gerald Schaefer28093f92016-04-28 16:18:35 -0700871#ifdef CONFIG_TRANSPARENT_HUGEPAGE
872struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr,
873 pmd_t pmd)
874{
875 unsigned long pfn = pmd_pfn(pmd);
876
877 /*
878 * There is no pmd_special() but there may be special pmds, e.g.
879 * in a direct-access (dax) mapping, so let's just replicate the
880 * !HAVE_PTE_SPECIAL case from vm_normal_page() here.
881 */
882 if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
883 if (vma->vm_flags & VM_MIXEDMAP) {
884 if (!pfn_valid(pfn))
885 return NULL;
886 goto out;
887 } else {
888 unsigned long off;
889 off = (addr - vma->vm_start) >> PAGE_SHIFT;
890 if (pfn == vma->vm_pgoff + off)
891 return NULL;
892 if (!is_cow_mapping(vma->vm_flags))
893 return NULL;
894 }
895 }
896
897 if (is_zero_pfn(pfn))
898 return NULL;
899 if (unlikely(pfn > highest_memmap_pfn))
900 return NULL;
901
902 /*
903 * NOTE! We still have PageReserved() pages in the page tables.
904 * eg. VDSO mappings can cause them to exist.
905 */
906out:
907 return pfn_to_page(pfn);
908}
909#endif
910
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800911/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 * copy one vm_area from one task to the other. Assumes the page tables
913 * already present in the new task to be cleared in the whole range
914 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 */
916
Hugh Dickins570a335b2009-12-14 17:58:46 -0800917static inline unsigned long
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700919 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700920 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921{
Nick Pigginb5810032005-10-29 18:16:12 -0700922 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 pte_t pte = *src_pte;
924 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925
926 /* pte contains position in swap or file, so copy. */
927 if (unlikely(!pte_present(pte))) {
Kirill A. Shutemov0661a332015-02-10 14:10:04 -0800928 swp_entry_t entry = pte_to_swp_entry(pte);
Christoph Lameter06972122006-06-23 02:03:35 -0700929
Kirill A. Shutemov0661a332015-02-10 14:10:04 -0800930 if (likely(!non_swap_entry(entry))) {
931 if (swap_duplicate(entry) < 0)
932 return entry.val;
Hugh Dickins570a335b2009-12-14 17:58:46 -0800933
Kirill A. Shutemov0661a332015-02-10 14:10:04 -0800934 /* make sure dst_mm is on swapoff's mmlist. */
935 if (unlikely(list_empty(&dst_mm->mmlist))) {
936 spin_lock(&mmlist_lock);
937 if (list_empty(&dst_mm->mmlist))
938 list_add(&dst_mm->mmlist,
939 &src_mm->mmlist);
940 spin_unlock(&mmlist_lock);
941 }
942 rss[MM_SWAPENTS]++;
943 } else if (is_migration_entry(entry)) {
944 page = migration_entry_to_page(entry);
Konstantin Khlebnikov9f9f1ac2012-01-20 14:34:24 -0800945
Jerome Marchandeca56ff2016-01-14 15:19:26 -0800946 rss[mm_counter(page)]++;
Konstantin Khlebnikov9f9f1ac2012-01-20 14:34:24 -0800947
Kirill A. Shutemov0661a332015-02-10 14:10:04 -0800948 if (is_write_migration_entry(entry) &&
949 is_cow_mapping(vm_flags)) {
950 /*
951 * COW mappings require pages in both
952 * parent and child to be set to read.
953 */
954 make_migration_entry_read(&entry);
955 pte = swp_entry_to_pte(entry);
956 if (pte_swp_soft_dirty(*src_pte))
957 pte = pte_swp_mksoft_dirty(pte);
958 set_pte_at(src_mm, addr, src_pte, pte);
Christoph Lameter06972122006-06-23 02:03:35 -0700959 }
Jérôme Glisse5042db42017-09-08 16:11:43 -0700960 } else if (is_device_private_entry(entry)) {
961 page = device_private_entry_to_page(entry);
962
963 /*
964 * Update rss count even for unaddressable pages, as
965 * they should treated just like normal pages in this
966 * respect.
967 *
968 * We will likely want to have some new rss counters
969 * for unaddressable pages, at some point. But for now
970 * keep things as they are.
971 */
972 get_page(page);
973 rss[mm_counter(page)]++;
974 page_dup_rmap(page, false);
975
976 /*
977 * We do not preserve soft-dirty information, because so
978 * far, checkpoint/restore is the only feature that
979 * requires that. And checkpoint/restore does not work
980 * when a device driver is involved (you cannot easily
981 * save and restore device driver state).
982 */
983 if (is_write_device_private_entry(entry) &&
984 is_cow_mapping(vm_flags)) {
985 make_device_private_entry_read(&entry);
986 pte = swp_entry_to_pte(entry);
987 set_pte_at(src_mm, addr, src_pte, pte);
988 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700990 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 }
992
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 /*
994 * If it's a COW mapping, write protect it both
995 * in the parent and the child
996 */
Linus Torvalds67121172005-12-11 20:38:17 -0800997 if (is_cow_mapping(vm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 ptep_set_wrprotect(src_mm, addr, src_pte);
Zachary Amsden3dc90792006-09-30 23:29:30 -0700999 pte = pte_wrprotect(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 }
1001
1002 /*
1003 * If it's a shared mapping, mark it clean in
1004 * the child
1005 */
1006 if (vm_flags & VM_SHARED)
1007 pte = pte_mkclean(pte);
1008 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001009
1010 page = vm_normal_page(vma, addr, pte);
1011 if (page) {
1012 get_page(page);
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08001013 page_dup_rmap(page, false);
Jerome Marchandeca56ff2016-01-14 15:19:26 -08001014 rss[mm_counter(page)]++;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001015 }
Hugh Dickinsae859762005-10-29 18:16:05 -07001016
1017out_set_pte:
1018 set_pte_at(dst_mm, addr, dst_pte, pte);
Hugh Dickins570a335b2009-12-14 17:58:46 -08001019 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020}
1021
Jerome Marchand21bda262014-08-06 16:06:56 -07001022static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001023 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
1024 unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025{
Daisuke Nishimurac36987e2009-10-26 16:50:23 -07001026 pte_t *orig_src_pte, *orig_dst_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001028 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -07001029 int progress = 0;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001030 int rss[NR_MM_COUNTERS];
Hugh Dickins570a335b2009-12-14 17:58:46 -08001031 swp_entry_t entry = (swp_entry_t){0};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032
1033again:
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001034 init_rss_vec(rss);
1035
Hugh Dickinsc74df322005-10-29 18:16:23 -07001036 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 if (!dst_pte)
1038 return -ENOMEM;
Peter Zijlstraece0e2b2010-10-26 14:21:52 -07001039 src_pte = pte_offset_map(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001040 src_ptl = pte_lockptr(src_mm, src_pmd);
Ingo Molnarf20dc5f2006-07-03 00:25:08 -07001041 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
Daisuke Nishimurac36987e2009-10-26 16:50:23 -07001042 orig_src_pte = src_pte;
1043 orig_dst_pte = dst_pte;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001044 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 do {
1047 /*
1048 * We are holding two locks at this point - either of them
1049 * could generate latencies in another task on another CPU.
1050 */
Hugh Dickinse040f212005-10-29 18:15:53 -07001051 if (progress >= 32) {
1052 progress = 0;
1053 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +01001054 spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -07001055 break;
1056 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 if (pte_none(*src_pte)) {
1058 progress++;
1059 continue;
1060 }
Hugh Dickins570a335b2009-12-14 17:58:46 -08001061 entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte,
1062 vma, addr, rss);
1063 if (entry.val)
1064 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 progress += 8;
1066 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001068 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001069 spin_unlock(src_ptl);
Peter Zijlstraece0e2b2010-10-26 14:21:52 -07001070 pte_unmap(orig_src_pte);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001071 add_mm_rss_vec(dst_mm, rss);
Daisuke Nishimurac36987e2009-10-26 16:50:23 -07001072 pte_unmap_unlock(orig_dst_pte, dst_ptl);
Hugh Dickinsc74df322005-10-29 18:16:23 -07001073 cond_resched();
Hugh Dickins570a335b2009-12-14 17:58:46 -08001074
1075 if (entry.val) {
1076 if (add_swap_count_continuation(entry, GFP_KERNEL) < 0)
1077 return -ENOMEM;
1078 progress = 0;
1079 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 if (addr != end)
1081 goto again;
1082 return 0;
1083}
1084
1085static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
1086 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
1087 unsigned long addr, unsigned long end)
1088{
1089 pmd_t *src_pmd, *dst_pmd;
1090 unsigned long next;
1091
1092 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
1093 if (!dst_pmd)
1094 return -ENOMEM;
1095 src_pmd = pmd_offset(src_pud, addr);
1096 do {
1097 next = pmd_addr_end(addr, end);
Zi Yan84c3fc42017-09-08 16:11:01 -07001098 if (is_swap_pmd(*src_pmd) || pmd_trans_huge(*src_pmd)
1099 || pmd_devmap(*src_pmd)) {
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001100 int err;
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08001101 VM_BUG_ON_VMA(next-addr != HPAGE_PMD_SIZE, vma);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001102 err = copy_huge_pmd(dst_mm, src_mm,
1103 dst_pmd, src_pmd, addr, vma);
1104 if (err == -ENOMEM)
1105 return -ENOMEM;
1106 if (!err)
1107 continue;
1108 /* fall through */
1109 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 if (pmd_none_or_clear_bad(src_pmd))
1111 continue;
1112 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
1113 vma, addr, next))
1114 return -ENOMEM;
1115 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
1116 return 0;
1117}
1118
1119static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001120 p4d_t *dst_p4d, p4d_t *src_p4d, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 unsigned long addr, unsigned long end)
1122{
1123 pud_t *src_pud, *dst_pud;
1124 unsigned long next;
1125
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001126 dst_pud = pud_alloc(dst_mm, dst_p4d, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 if (!dst_pud)
1128 return -ENOMEM;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001129 src_pud = pud_offset(src_p4d, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 do {
1131 next = pud_addr_end(addr, end);
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08001132 if (pud_trans_huge(*src_pud) || pud_devmap(*src_pud)) {
1133 int err;
1134
1135 VM_BUG_ON_VMA(next-addr != HPAGE_PUD_SIZE, vma);
1136 err = copy_huge_pud(dst_mm, src_mm,
1137 dst_pud, src_pud, addr, vma);
1138 if (err == -ENOMEM)
1139 return -ENOMEM;
1140 if (!err)
1141 continue;
1142 /* fall through */
1143 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 if (pud_none_or_clear_bad(src_pud))
1145 continue;
1146 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
1147 vma, addr, next))
1148 return -ENOMEM;
1149 } while (dst_pud++, src_pud++, addr = next, addr != end);
1150 return 0;
1151}
1152
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001153static inline int copy_p4d_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
1154 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
1155 unsigned long addr, unsigned long end)
1156{
1157 p4d_t *src_p4d, *dst_p4d;
1158 unsigned long next;
1159
1160 dst_p4d = p4d_alloc(dst_mm, dst_pgd, addr);
1161 if (!dst_p4d)
1162 return -ENOMEM;
1163 src_p4d = p4d_offset(src_pgd, addr);
1164 do {
1165 next = p4d_addr_end(addr, end);
1166 if (p4d_none_or_clear_bad(src_p4d))
1167 continue;
1168 if (copy_pud_range(dst_mm, src_mm, dst_p4d, src_p4d,
1169 vma, addr, next))
1170 return -ENOMEM;
1171 } while (dst_p4d++, src_p4d++, addr = next, addr != end);
1172 return 0;
1173}
1174
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
1176 struct vm_area_struct *vma)
1177{
1178 pgd_t *src_pgd, *dst_pgd;
1179 unsigned long next;
1180 unsigned long addr = vma->vm_start;
1181 unsigned long end = vma->vm_end;
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001182 unsigned long mmun_start; /* For mmu_notifiers */
1183 unsigned long mmun_end; /* For mmu_notifiers */
1184 bool is_cow;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001185 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186
Nick Piggind9928952005-08-28 16:49:11 +10001187 /*
1188 * Don't copy ptes where a page fault will fill them correctly.
1189 * Fork becomes much lighter when there are big shared or private
1190 * readonly mappings. The tradeoff is that copy_page_range is more
1191 * efficient than faulting.
1192 */
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08001193 if (!(vma->vm_flags & (VM_HUGETLB | VM_PFNMAP | VM_MIXEDMAP)) &&
1194 !vma->anon_vma)
1195 return 0;
Nick Piggind9928952005-08-28 16:49:11 +10001196
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 if (is_vm_hugetlb_page(vma))
1198 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
1199
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07001200 if (unlikely(vma->vm_flags & VM_PFNMAP)) {
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001201 /*
1202 * We do not free on error cases below as remove_vma
1203 * gets called on error from higher level routine
1204 */
Suresh Siddha5180da42012-10-08 16:28:29 -07001205 ret = track_pfn_copy(vma);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001206 if (ret)
1207 return ret;
1208 }
1209
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001210 /*
1211 * We need to invalidate the secondary MMU mappings only when
1212 * there could be a permission downgrade on the ptes of the
1213 * parent mm. And a permission downgrade will only happen if
1214 * is_cow_mapping() returns true.
1215 */
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001216 is_cow = is_cow_mapping(vma->vm_flags);
1217 mmun_start = addr;
1218 mmun_end = end;
1219 if (is_cow)
1220 mmu_notifier_invalidate_range_start(src_mm, mmun_start,
1221 mmun_end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001222
1223 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 dst_pgd = pgd_offset(dst_mm, addr);
1225 src_pgd = pgd_offset(src_mm, addr);
1226 do {
1227 next = pgd_addr_end(addr, end);
1228 if (pgd_none_or_clear_bad(src_pgd))
1229 continue;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001230 if (unlikely(copy_p4d_range(dst_mm, src_mm, dst_pgd, src_pgd,
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001231 vma, addr, next))) {
1232 ret = -ENOMEM;
1233 break;
1234 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001236
Sagi Grimberg2ec74c32012-10-08 16:33:33 -07001237 if (is_cow)
1238 mmu_notifier_invalidate_range_end(src_mm, mmun_start, mmun_end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001239 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240}
1241
Robin Holt51c6f662005-11-13 16:06:42 -08001242static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -07001243 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 unsigned long addr, unsigned long end,
Peter Zijlstra97a89412011-05-24 17:12:04 -07001245 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246{
Nick Pigginb5810032005-10-29 18:16:12 -07001247 struct mm_struct *mm = tlb->mm;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001248 int force_flush = 0;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001249 int rss[NR_MM_COUNTERS];
Peter Zijlstra97a89412011-05-24 17:12:04 -07001250 spinlock_t *ptl;
Steven Rostedt5f1a1902011-06-15 15:08:23 -07001251 pte_t *start_pte;
Peter Zijlstra97a89412011-05-24 17:12:04 -07001252 pte_t *pte;
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001253 swp_entry_t entry;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001254
Aneesh Kumar K.V07e32662016-12-12 16:42:40 -08001255 tlb_remove_check_page_size_change(tlb, PAGE_SIZE);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001256again:
Peter Zijlstrae3032972011-05-24 17:12:01 -07001257 init_rss_vec(rss);
Steven Rostedt5f1a1902011-06-15 15:08:23 -07001258 start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
1259 pte = start_pte;
Mel Gorman3ea27712017-08-02 13:31:52 -07001260 flush_tlb_batched_pending(mm);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001261 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 do {
1263 pte_t ptent = *pte;
Tobin C Harding166f61b2017-02-24 14:59:01 -08001264 if (pte_none(ptent))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 continue;
Hugh Dickins6f5e6b92006-03-16 23:04:09 -08001266
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -08001268 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -08001269
Linus Torvalds6aab3412005-11-28 14:34:23 -08001270 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 if (unlikely(details) && page) {
1272 /*
1273 * unmap_shared_mapping_pages() wants to
1274 * invalidate cache without truncating:
1275 * unmap shared but keep private pages.
1276 */
1277 if (details->check_mapping &&
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001278 details->check_mapping != page_rmapping(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 }
Nick Pigginb5810032005-10-29 18:16:12 -07001281 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -07001282 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 tlb_remove_tlb_entry(tlb, pte, addr);
1284 if (unlikely(!page))
1285 continue;
Jerome Marchandeca56ff2016-01-14 15:19:26 -08001286
1287 if (!PageAnon(page)) {
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001288 if (pte_dirty(ptent)) {
1289 force_flush = 1;
Hugh Dickins6237bcd2005-10-29 18:15:54 -07001290 set_page_dirty(page);
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001291 }
Johannes Weiner4917e5d2009-01-06 14:39:17 -08001292 if (pte_young(ptent) &&
Joe Perches64363aa2013-07-08 16:00:18 -07001293 likely(!(vma->vm_flags & VM_SEQ_READ)))
Nick Pigginbf3f3bc2009-01-06 14:38:55 -08001294 mark_page_accessed(page);
Hugh Dickins6237bcd2005-10-29 18:15:54 -07001295 }
Jerome Marchandeca56ff2016-01-14 15:19:26 -08001296 rss[mm_counter(page)]--;
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08001297 page_remove_rmap(page, false);
Hugh Dickins3dc14742009-01-06 14:40:08 -08001298 if (unlikely(page_mapcount(page) < 0))
1299 print_bad_pte(vma, addr, ptent, page);
Aneesh Kumar K.Ve9d55e12016-07-26 15:24:09 -07001300 if (unlikely(__tlb_remove_page(tlb, page))) {
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001301 force_flush = 1;
Will Deaconce9ec372014-10-28 13:16:28 -07001302 addr += PAGE_SIZE;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001303 break;
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001304 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 continue;
1306 }
Jérôme Glisse5042db42017-09-08 16:11:43 -07001307
1308 entry = pte_to_swp_entry(ptent);
1309 if (non_swap_entry(entry) && is_device_private_entry(entry)) {
1310 struct page *page = device_private_entry_to_page(entry);
1311
1312 if (unlikely(details && details->check_mapping)) {
1313 /*
1314 * unmap_shared_mapping_pages() wants to
1315 * invalidate cache without truncating:
1316 * unmap shared but keep private pages.
1317 */
1318 if (details->check_mapping !=
1319 page_rmapping(page))
1320 continue;
1321 }
1322
1323 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
1324 rss[mm_counter(page)]--;
1325 page_remove_rmap(page, false);
1326 put_page(page);
1327 continue;
1328 }
1329
Kirill A. Shutemov3e8715f2017-02-22 15:46:34 -08001330 /* If details->check_mapping, we leave swap entries. */
1331 if (unlikely(details))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 continue;
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001333
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001334 entry = pte_to_swp_entry(ptent);
1335 if (!non_swap_entry(entry))
1336 rss[MM_SWAPENTS]--;
1337 else if (is_migration_entry(entry)) {
1338 struct page *page;
Konstantin Khlebnikov9f9f1ac2012-01-20 14:34:24 -08001339
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001340 page = migration_entry_to_page(entry);
Jerome Marchandeca56ff2016-01-14 15:19:26 -08001341 rss[mm_counter(page)]--;
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001342 }
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001343 if (unlikely(!free_swap_and_cache(entry)))
1344 print_bad_pte(vma, addr, ptent, NULL);
Zachary Amsden9888a1c2006-09-30 23:29:31 -07001345 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
Peter Zijlstra97a89412011-05-24 17:12:04 -07001346 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickinsae859762005-10-29 18:16:05 -07001347
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001348 add_mm_rss_vec(mm, rss);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001349 arch_leave_lazy_mmu_mode();
Robin Holt51c6f662005-11-13 16:06:42 -08001350
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001351 /* Do the actual TLB flush before dropping ptl */
Will Deaconfb7332a2014-10-29 10:03:09 +00001352 if (force_flush)
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001353 tlb_flush_mmu_tlbonly(tlb);
Linus Torvalds1cf35d42014-04-25 16:05:40 -07001354 pte_unmap_unlock(start_pte, ptl);
1355
1356 /*
1357 * If we forced a TLB flush (either due to running out of
1358 * batch buffers or because we needed to flush dirty TLB
1359 * entries before releasing the ptl), free the batched
1360 * memory too. Restart if we didn't do everything.
1361 */
1362 if (force_flush) {
1363 force_flush = 0;
1364 tlb_flush_mmu_free(tlb);
Linus Torvalds2b047252013-08-15 11:42:25 -07001365 if (addr != end)
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001366 goto again;
1367 }
1368
Robin Holt51c6f662005-11-13 16:06:42 -08001369 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370}
1371
Robin Holt51c6f662005-11-13 16:06:42 -08001372static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -07001373 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 unsigned long addr, unsigned long end,
Peter Zijlstra97a89412011-05-24 17:12:04 -07001375 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376{
1377 pmd_t *pmd;
1378 unsigned long next;
1379
1380 pmd = pmd_offset(pud, addr);
1381 do {
1382 next = pmd_addr_end(addr, end);
Zi Yan84c3fc42017-09-08 16:11:01 -07001383 if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) {
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001384 if (next - addr != HPAGE_PMD_SIZE) {
Hugh Dickins68428392016-05-05 16:22:09 -07001385 VM_BUG_ON_VMA(vma_is_anonymous(vma) &&
1386 !rwsem_is_locked(&tlb->mm->mmap_sem), vma);
David Rientjesfd607752016-12-12 16:42:20 -08001387 __split_huge_pmd(vma, pmd, addr, false, NULL);
Shaohua Lif21760b2012-01-12 17:19:16 -08001388 } else if (zap_huge_pmd(tlb, vma, pmd, addr))
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001389 goto next;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001390 /* fall through */
1391 }
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001392 /*
1393 * Here there can be other concurrent MADV_DONTNEED or
1394 * trans huge page faults running, and if the pmd is
1395 * none or trans huge it can change under us. This is
1396 * because MADV_DONTNEED holds the mmap_sem in read
1397 * mode.
1398 */
1399 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1400 goto next;
Peter Zijlstra97a89412011-05-24 17:12:04 -07001401 next = zap_pte_range(tlb, vma, pmd, addr, next, details);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001402next:
Peter Zijlstra97a89412011-05-24 17:12:04 -07001403 cond_resched();
1404 } while (pmd++, addr = next, addr != end);
Robin Holt51c6f662005-11-13 16:06:42 -08001405
1406 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407}
1408
Robin Holt51c6f662005-11-13 16:06:42 -08001409static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001410 struct vm_area_struct *vma, p4d_t *p4d,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 unsigned long addr, unsigned long end,
Peter Zijlstra97a89412011-05-24 17:12:04 -07001412 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413{
1414 pud_t *pud;
1415 unsigned long next;
1416
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001417 pud = pud_offset(p4d, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 do {
1419 next = pud_addr_end(addr, end);
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08001420 if (pud_trans_huge(*pud) || pud_devmap(*pud)) {
1421 if (next - addr != HPAGE_PUD_SIZE) {
1422 VM_BUG_ON_VMA(!rwsem_is_locked(&tlb->mm->mmap_sem), vma);
1423 split_huge_pud(vma, pud, addr);
1424 } else if (zap_huge_pud(tlb, vma, pud, addr))
1425 goto next;
1426 /* fall through */
1427 }
Peter Zijlstra97a89412011-05-24 17:12:04 -07001428 if (pud_none_or_clear_bad(pud))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 continue;
Peter Zijlstra97a89412011-05-24 17:12:04 -07001430 next = zap_pmd_range(tlb, vma, pud, addr, next, details);
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08001431next:
1432 cond_resched();
Peter Zijlstra97a89412011-05-24 17:12:04 -07001433 } while (pud++, addr = next, addr != end);
Robin Holt51c6f662005-11-13 16:06:42 -08001434
1435 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436}
1437
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001438static inline unsigned long zap_p4d_range(struct mmu_gather *tlb,
1439 struct vm_area_struct *vma, pgd_t *pgd,
1440 unsigned long addr, unsigned long end,
1441 struct zap_details *details)
1442{
1443 p4d_t *p4d;
1444 unsigned long next;
1445
1446 p4d = p4d_offset(pgd, addr);
1447 do {
1448 next = p4d_addr_end(addr, end);
1449 if (p4d_none_or_clear_bad(p4d))
1450 continue;
1451 next = zap_pud_range(tlb, vma, p4d, addr, next, details);
1452 } while (p4d++, addr = next, addr != end);
1453
1454 return addr;
1455}
1456
Michal Hockoaac45362016-03-25 14:20:24 -07001457void unmap_page_range(struct mmu_gather *tlb,
Al Viro038c7aa2012-03-05 13:25:09 -05001458 struct vm_area_struct *vma,
1459 unsigned long addr, unsigned long end,
1460 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461{
1462 pgd_t *pgd;
1463 unsigned long next;
1464
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 BUG_ON(addr >= end);
1466 tlb_start_vma(tlb, vma);
1467 pgd = pgd_offset(vma->vm_mm, addr);
1468 do {
1469 next = pgd_addr_end(addr, end);
Peter Zijlstra97a89412011-05-24 17:12:04 -07001470 if (pgd_none_or_clear_bad(pgd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 continue;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001472 next = zap_p4d_range(tlb, vma, pgd, addr, next, details);
Peter Zijlstra97a89412011-05-24 17:12:04 -07001473 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 tlb_end_vma(tlb, vma);
1475}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476
Al Virof5cc4ee2012-03-05 14:14:20 -05001477
1478static void unmap_single_vma(struct mmu_gather *tlb,
1479 struct vm_area_struct *vma, unsigned long start_addr,
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001480 unsigned long end_addr,
Al Virof5cc4ee2012-03-05 14:14:20 -05001481 struct zap_details *details)
1482{
1483 unsigned long start = max(vma->vm_start, start_addr);
1484 unsigned long end;
1485
1486 if (start >= vma->vm_end)
1487 return;
1488 end = min(vma->vm_end, end_addr);
1489 if (end <= vma->vm_start)
1490 return;
1491
Srikar Dronamrajucbc91f72012-04-11 16:05:27 +05301492 if (vma->vm_file)
1493 uprobe_munmap(vma, start, end);
1494
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07001495 if (unlikely(vma->vm_flags & VM_PFNMAP))
Suresh Siddha5180da42012-10-08 16:28:29 -07001496 untrack_pfn(vma, 0, 0);
Al Virof5cc4ee2012-03-05 14:14:20 -05001497
1498 if (start != end) {
1499 if (unlikely(is_vm_hugetlb_page(vma))) {
1500 /*
1501 * It is undesirable to test vma->vm_file as it
1502 * should be non-null for valid hugetlb area.
1503 * However, vm_file will be NULL in the error
Davidlohr Bueso7aa6b4a2014-04-07 15:37:01 -07001504 * cleanup path of mmap_region. When
Al Virof5cc4ee2012-03-05 14:14:20 -05001505 * hugetlbfs ->mmap method fails,
Davidlohr Bueso7aa6b4a2014-04-07 15:37:01 -07001506 * mmap_region() nullifies vma->vm_file
Al Virof5cc4ee2012-03-05 14:14:20 -05001507 * before calling this function to clean up.
1508 * Since no pte has actually been setup, it is
1509 * safe to do nothing in this case.
1510 */
Aneesh Kumar K.V24669e52012-07-31 16:42:03 -07001511 if (vma->vm_file) {
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08001512 i_mmap_lock_write(vma->vm_file->f_mapping);
Mel Gormand8333522012-07-31 16:46:20 -07001513 __unmap_hugepage_range_final(tlb, vma, start, end, NULL);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -08001514 i_mmap_unlock_write(vma->vm_file->f_mapping);
Aneesh Kumar K.V24669e52012-07-31 16:42:03 -07001515 }
Al Virof5cc4ee2012-03-05 14:14:20 -05001516 } else
1517 unmap_page_range(tlb, vma, start, end, details);
1518 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519}
1520
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521/**
1522 * unmap_vmas - unmap a range of memory covered by a list of vma's
Randy Dunlap0164f692011-06-15 15:08:09 -07001523 * @tlb: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 * @vma: the starting vma
1525 * @start_addr: virtual address at which to start unmapping
1526 * @end_addr: virtual address at which to end unmapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 *
Hugh Dickins508034a2005-10-29 18:16:30 -07001528 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 * Only addresses between `start' and `end' will be unmapped.
1531 *
1532 * The VMA list must be sorted in ascending virtual address order.
1533 *
1534 * unmap_vmas() assumes that the caller will flush the whole unmapped address
1535 * range after unmap_vmas() returns. So the only responsibility here is to
1536 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
1537 * drops the lock and schedules.
1538 */
Al Viro6e8bb012012-03-05 13:41:15 -05001539void unmap_vmas(struct mmu_gather *tlb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 struct vm_area_struct *vma, unsigned long start_addr,
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001541 unsigned long end_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542{
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001543 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001545 mmu_notifier_invalidate_range_start(mm, start_addr, end_addr);
Al Virof5cc4ee2012-03-05 14:14:20 -05001546 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next)
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001547 unmap_single_vma(tlb, vma, start_addr, end_addr, NULL);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001548 mmu_notifier_invalidate_range_end(mm, start_addr, end_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549}
1550
1551/**
1552 * zap_page_range - remove user pages in a given range
1553 * @vma: vm_area_struct holding the applicable pages
Randy Dunlapeb4546b2012-06-20 12:53:02 -07001554 * @start: starting address of pages to zap
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 * @size: number of bytes to zap
Al Virof5cc4ee2012-03-05 14:14:20 -05001556 *
1557 * Caller must protect the VMA list
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 */
Linus Torvalds7e027b12012-05-06 13:43:15 -07001559void zap_page_range(struct vm_area_struct *vma, unsigned long start,
Kirill A. Shutemovecf13852017-02-22 15:46:37 -08001560 unsigned long size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561{
1562 struct mm_struct *mm = vma->vm_mm;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001563 struct mmu_gather tlb;
Linus Torvalds7e027b12012-05-06 13:43:15 -07001564 unsigned long end = start + size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07001567 tlb_gather_mmu(&tlb, mm, start, end);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001568 update_hiwater_rss(mm);
Linus Torvalds7e027b12012-05-06 13:43:15 -07001569 mmu_notifier_invalidate_range_start(mm, start, end);
Mel Gorman46477062017-09-06 16:21:05 -07001570 for ( ; vma && vma->vm_start < end; vma = vma->vm_next) {
Kirill A. Shutemovecf13852017-02-22 15:46:37 -08001571 unmap_single_vma(&tlb, vma, start, end, NULL);
Mel Gorman46477062017-09-06 16:21:05 -07001572
1573 /*
1574 * zap_page_range does not specify whether mmap_sem should be
1575 * held for read or write. That allows parallel zap_page_range
1576 * operations to unmap a PTE and defer a flush meaning that
1577 * this call observes pte_none and fails to flush the TLB.
1578 * Rather than adding a complex API, ensure that no stale
1579 * TLB entries exist when this call returns.
1580 */
1581 flush_tlb_range(vma, start, end);
1582 }
1583
Linus Torvalds7e027b12012-05-06 13:43:15 -07001584 mmu_notifier_invalidate_range_end(mm, start, end);
1585 tlb_finish_mmu(&tlb, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586}
1587
Jack Steinerc627f9c2008-07-29 22:33:53 -07001588/**
Al Virof5cc4ee2012-03-05 14:14:20 -05001589 * zap_page_range_single - remove user pages in a given range
1590 * @vma: vm_area_struct holding the applicable pages
1591 * @address: starting address of pages to zap
1592 * @size: number of bytes to zap
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08001593 * @details: details of shared cache invalidation
Al Virof5cc4ee2012-03-05 14:14:20 -05001594 *
1595 * The range must fit into one VMA.
1596 */
1597static void zap_page_range_single(struct vm_area_struct *vma, unsigned long address,
1598 unsigned long size, struct zap_details *details)
1599{
1600 struct mm_struct *mm = vma->vm_mm;
1601 struct mmu_gather tlb;
1602 unsigned long end = address + size;
Al Virof5cc4ee2012-03-05 14:14:20 -05001603
1604 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -07001605 tlb_gather_mmu(&tlb, mm, address, end);
Al Virof5cc4ee2012-03-05 14:14:20 -05001606 update_hiwater_rss(mm);
1607 mmu_notifier_invalidate_range_start(mm, address, end);
Linus Torvalds4f74d2c2012-05-06 13:54:06 -07001608 unmap_single_vma(&tlb, vma, address, end, details);
Al Virof5cc4ee2012-03-05 14:14:20 -05001609 mmu_notifier_invalidate_range_end(mm, address, end);
1610 tlb_finish_mmu(&tlb, address, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611}
1612
Jack Steinerc627f9c2008-07-29 22:33:53 -07001613/**
1614 * zap_vma_ptes - remove ptes mapping the vma
1615 * @vma: vm_area_struct holding ptes to be zapped
1616 * @address: starting address of pages to zap
1617 * @size: number of bytes to zap
1618 *
1619 * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1620 *
1621 * The entire address range must be fully contained within the vma.
1622 *
1623 * Returns 0 if successful.
1624 */
1625int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
1626 unsigned long size)
1627{
1628 if (address < vma->vm_start || address + size > vma->vm_end ||
1629 !(vma->vm_flags & VM_PFNMAP))
1630 return -1;
Al Virof5cc4ee2012-03-05 14:14:20 -05001631 zap_page_range_single(vma, address, size, NULL);
Jack Steinerc627f9c2008-07-29 22:33:53 -07001632 return 0;
1633}
1634EXPORT_SYMBOL_GPL(zap_vma_ptes);
1635
Namhyung Kim25ca1d62010-10-26 14:21:59 -07001636pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr,
Harvey Harrison920c7a52008-02-04 22:29:26 -08001637 spinlock_t **ptl)
Linus Torvaldsc9cfcddf2005-11-29 14:03:14 -08001638{
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001639 pgd_t *pgd;
1640 p4d_t *p4d;
1641 pud_t *pud;
1642 pmd_t *pmd;
1643
1644 pgd = pgd_offset(mm, addr);
1645 p4d = p4d_alloc(mm, pgd, addr);
1646 if (!p4d)
1647 return NULL;
1648 pud = pud_alloc(mm, p4d, addr);
1649 if (!pud)
1650 return NULL;
1651 pmd = pmd_alloc(mm, pud, addr);
1652 if (!pmd)
1653 return NULL;
1654
1655 VM_BUG_ON(pmd_trans_huge(*pmd));
1656 return pte_alloc_map_lock(mm, pmd, addr, ptl);
Linus Torvaldsc9cfcddf2005-11-29 14:03:14 -08001657}
1658
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001660 * This is the old fallback for page remapping.
1661 *
1662 * For historical reasons, it only allows reserved pages. Only
1663 * old drivers should use this, and they needed to mark their
1664 * pages reserved for the old functions anyway.
1665 */
Nick Piggin423bad602008-04-28 02:13:01 -07001666static int insert_page(struct vm_area_struct *vma, unsigned long addr,
1667 struct page *page, pgprot_t prot)
Linus Torvalds238f58d2005-11-29 13:01:56 -08001668{
Nick Piggin423bad602008-04-28 02:13:01 -07001669 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001670 int retval;
Linus Torvaldsc9cfcddf2005-11-29 14:03:14 -08001671 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001672 spinlock_t *ptl;
1673
Linus Torvalds238f58d2005-11-29 13:01:56 -08001674 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001675 if (PageAnon(page))
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001676 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001677 retval = -ENOMEM;
1678 flush_dcache_page(page);
Linus Torvaldsc9cfcddf2005-11-29 14:03:14 -08001679 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001680 if (!pte)
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001681 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001682 retval = -EBUSY;
1683 if (!pte_none(*pte))
1684 goto out_unlock;
1685
1686 /* Ok, finally just insert the thing.. */
1687 get_page(page);
Jerome Marchandeca56ff2016-01-14 15:19:26 -08001688 inc_mm_counter_fast(mm, mm_counter_file(page));
Kirill A. Shutemovdd78fed2016-07-26 15:25:26 -07001689 page_add_file_rmap(page, false);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001690 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1691
1692 retval = 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001693 pte_unmap_unlock(pte, ptl);
1694 return retval;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001695out_unlock:
1696 pte_unmap_unlock(pte, ptl);
1697out:
1698 return retval;
1699}
1700
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001701/**
1702 * vm_insert_page - insert single page into user vma
1703 * @vma: user vma to map to
1704 * @addr: target user address of this page
1705 * @page: source kernel page
1706 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001707 * This allows drivers to insert individual pages they've allocated
1708 * into a user vma.
1709 *
1710 * The page has to be a nice clean _individual_ kernel allocation.
1711 * If you allocate a compound page, you need to have marked it as
1712 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001713 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001714 *
1715 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1716 * took an arbitrary page protection parameter. This doesn't allow
1717 * that. Your vma protection will have to be set up correctly, which
1718 * means that if you want a shared writable mapping, you'd better
1719 * ask for a shared writable mapping!
1720 *
1721 * The page does not need to be reserved.
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001722 *
1723 * Usually this function is called from f_op->mmap() handler
1724 * under mm->mmap_sem write-lock, so it can change vma->vm_flags.
1725 * Caller must set VM_MIXEDMAP on vma if it wants to call this
1726 * function from other places, for example from page-fault handler.
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001727 */
Nick Piggin423bad602008-04-28 02:13:01 -07001728int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
1729 struct page *page)
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001730{
1731 if (addr < vma->vm_start || addr >= vma->vm_end)
1732 return -EFAULT;
1733 if (!page_count(page))
1734 return -EINVAL;
Konstantin Khlebnikov4b6e1e32012-10-08 16:28:40 -07001735 if (!(vma->vm_flags & VM_MIXEDMAP)) {
1736 BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem));
1737 BUG_ON(vma->vm_flags & VM_PFNMAP);
1738 vma->vm_flags |= VM_MIXEDMAP;
1739 }
Nick Piggin423bad602008-04-28 02:13:01 -07001740 return insert_page(vma, addr, page, vma->vm_page_prot);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001741}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001742EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001743
Nick Piggin423bad602008-04-28 02:13:01 -07001744static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
Ross Zwislerb2770da62017-09-06 16:18:35 -07001745 pfn_t pfn, pgprot_t prot, bool mkwrite)
Nick Piggin423bad602008-04-28 02:13:01 -07001746{
1747 struct mm_struct *mm = vma->vm_mm;
1748 int retval;
1749 pte_t *pte, entry;
1750 spinlock_t *ptl;
1751
1752 retval = -ENOMEM;
1753 pte = get_locked_pte(mm, addr, &ptl);
1754 if (!pte)
1755 goto out;
1756 retval = -EBUSY;
Ross Zwislerb2770da62017-09-06 16:18:35 -07001757 if (!pte_none(*pte)) {
1758 if (mkwrite) {
1759 /*
1760 * For read faults on private mappings the PFN passed
1761 * in may not match the PFN we have mapped if the
1762 * mapped PFN is a writeable COW page. In the mkwrite
1763 * case we are creating a writable PTE for a shared
1764 * mapping and we expect the PFNs to match.
1765 */
1766 if (WARN_ON_ONCE(pte_pfn(*pte) != pfn_t_to_pfn(pfn)))
1767 goto out_unlock;
1768 entry = *pte;
1769 goto out_mkwrite;
1770 } else
1771 goto out_unlock;
1772 }
Nick Piggin423bad602008-04-28 02:13:01 -07001773
1774 /* Ok, finally just insert the thing.. */
Dan Williams01c8f1c2016-01-15 16:56:40 -08001775 if (pfn_t_devmap(pfn))
1776 entry = pte_mkdevmap(pfn_t_pte(pfn, prot));
1777 else
1778 entry = pte_mkspecial(pfn_t_pte(pfn, prot));
Ross Zwislerb2770da62017-09-06 16:18:35 -07001779
1780out_mkwrite:
1781 if (mkwrite) {
1782 entry = pte_mkyoung(entry);
1783 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1784 }
1785
Nick Piggin423bad602008-04-28 02:13:01 -07001786 set_pte_at(mm, addr, pte, entry);
Russell King4b3073e2009-12-18 16:40:18 +00001787 update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */
Nick Piggin423bad602008-04-28 02:13:01 -07001788
1789 retval = 0;
1790out_unlock:
1791 pte_unmap_unlock(pte, ptl);
1792out:
1793 return retval;
1794}
1795
Nick Piggine0dc0d82007-02-12 00:51:36 -08001796/**
1797 * vm_insert_pfn - insert single pfn into user vma
1798 * @vma: user vma to map to
1799 * @addr: target user address of this page
1800 * @pfn: source kernel pfn
1801 *
Robert P. J. Dayc462f172012-10-08 16:33:43 -07001802 * Similar to vm_insert_page, this allows drivers to insert individual pages
Nick Piggine0dc0d82007-02-12 00:51:36 -08001803 * they've allocated into a user vma. Same comments apply.
1804 *
1805 * This function should only be called from a vm_ops->fault handler, and
1806 * in that case the handler should return NULL.
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001807 *
1808 * vma cannot be a COW mapping.
1809 *
1810 * As this is called only for pages that do not currently exist, we
1811 * do not need to flush old virtual caches or the TLB.
Nick Piggine0dc0d82007-02-12 00:51:36 -08001812 */
1813int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
Nick Piggin423bad602008-04-28 02:13:01 -07001814 unsigned long pfn)
Nick Piggine0dc0d82007-02-12 00:51:36 -08001815{
Andy Lutomirski1745cbc2015-12-29 20:12:20 -08001816 return vm_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot);
1817}
1818EXPORT_SYMBOL(vm_insert_pfn);
1819
1820/**
1821 * vm_insert_pfn_prot - insert single pfn into user vma with specified pgprot
1822 * @vma: user vma to map to
1823 * @addr: target user address of this page
1824 * @pfn: source kernel pfn
1825 * @pgprot: pgprot flags for the inserted page
1826 *
1827 * This is exactly like vm_insert_pfn, except that it allows drivers to
1828 * to override pgprot on a per-page basis.
1829 *
1830 * This only makes sense for IO mappings, and it makes no sense for
1831 * cow mappings. In general, using multiple vmas is preferable;
1832 * vm_insert_pfn_prot should only be used if using multiple VMAs is
1833 * impractical.
1834 */
1835int vm_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr,
1836 unsigned long pfn, pgprot_t pgprot)
1837{
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001838 int ret;
Nick Piggin7e675132008-04-28 02:13:00 -07001839 /*
1840 * Technically, architectures with pte_special can avoid all these
1841 * restrictions (same for remap_pfn_range). However we would like
1842 * consistency in testing and feature parity among all, so we should
1843 * try to keep these invariants in place for everybody.
1844 */
Jared Hulbertb379d792008-04-28 02:12:58 -07001845 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
1846 BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
1847 (VM_PFNMAP|VM_MIXEDMAP));
1848 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1849 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
Nick Piggine0dc0d82007-02-12 00:51:36 -08001850
Nick Piggin423bad602008-04-28 02:13:01 -07001851 if (addr < vma->vm_start || addr >= vma->vm_end)
1852 return -EFAULT;
Borislav Petkov308a0472016-10-26 19:43:43 +02001853
1854 track_pfn_insert(vma, &pgprot, __pfn_to_pfn_t(pfn, PFN_DEV));
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001855
Ross Zwislerb2770da62017-09-06 16:18:35 -07001856 ret = insert_pfn(vma, addr, __pfn_to_pfn_t(pfn, PFN_DEV), pgprot,
1857 false);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001858
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001859 return ret;
Nick Piggine0dc0d82007-02-12 00:51:36 -08001860}
Andy Lutomirski1745cbc2015-12-29 20:12:20 -08001861EXPORT_SYMBOL(vm_insert_pfn_prot);
Nick Piggine0dc0d82007-02-12 00:51:36 -08001862
Ross Zwislerb2770da62017-09-06 16:18:35 -07001863static int __vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
1864 pfn_t pfn, bool mkwrite)
Nick Piggin423bad602008-04-28 02:13:01 -07001865{
Dan Williams87744ab2016-10-07 17:00:18 -07001866 pgprot_t pgprot = vma->vm_page_prot;
1867
Nick Piggin423bad602008-04-28 02:13:01 -07001868 BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
1869
1870 if (addr < vma->vm_start || addr >= vma->vm_end)
1871 return -EFAULT;
Borislav Petkov308a0472016-10-26 19:43:43 +02001872
1873 track_pfn_insert(vma, &pgprot, pfn);
Nick Piggin423bad602008-04-28 02:13:01 -07001874
1875 /*
1876 * If we don't have pte special, then we have to use the pfn_valid()
1877 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1878 * refcount the page if pfn_valid is true (hence insert_page rather
Hugh Dickins62eede62009-09-21 17:03:34 -07001879 * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP
1880 * without pte special, it would there be refcounted as a normal page.
Nick Piggin423bad602008-04-28 02:13:01 -07001881 */
Dan Williams03fc2da2016-01-26 09:48:05 -08001882 if (!HAVE_PTE_SPECIAL && !pfn_t_devmap(pfn) && pfn_t_valid(pfn)) {
Nick Piggin423bad602008-04-28 02:13:01 -07001883 struct page *page;
1884
Dan Williams03fc2da2016-01-26 09:48:05 -08001885 /*
1886 * At this point we are committed to insert_page()
1887 * regardless of whether the caller specified flags that
1888 * result in pfn_t_has_page() == false.
1889 */
1890 page = pfn_to_page(pfn_t_to_pfn(pfn));
Dan Williams87744ab2016-10-07 17:00:18 -07001891 return insert_page(vma, addr, page, pgprot);
Nick Piggin423bad602008-04-28 02:13:01 -07001892 }
Ross Zwislerb2770da62017-09-06 16:18:35 -07001893 return insert_pfn(vma, addr, pfn, pgprot, mkwrite);
1894}
1895
1896int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
1897 pfn_t pfn)
1898{
1899 return __vm_insert_mixed(vma, addr, pfn, false);
1900
Nick Piggin423bad602008-04-28 02:13:01 -07001901}
1902EXPORT_SYMBOL(vm_insert_mixed);
1903
Ross Zwislerb2770da62017-09-06 16:18:35 -07001904int vm_insert_mixed_mkwrite(struct vm_area_struct *vma, unsigned long addr,
1905 pfn_t pfn)
1906{
1907 return __vm_insert_mixed(vma, addr, pfn, true);
1908}
1909EXPORT_SYMBOL(vm_insert_mixed_mkwrite);
1910
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001911/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 * maps a range of physical memory into the requested pages. the old
1913 * mappings are removed. any references to nonexistent pages results
1914 * in null mappings (currently treated as "copy-on-access")
1915 */
1916static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1917 unsigned long addr, unsigned long end,
1918 unsigned long pfn, pgprot_t prot)
1919{
1920 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001921 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922
Hugh Dickinsc74df322005-10-29 18:16:23 -07001923 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 if (!pte)
1925 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001926 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927 do {
1928 BUG_ON(!pte_none(*pte));
Nick Piggin7e675132008-04-28 02:13:00 -07001929 set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 pfn++;
1931 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001932 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001933 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 return 0;
1935}
1936
1937static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1938 unsigned long addr, unsigned long end,
1939 unsigned long pfn, pgprot_t prot)
1940{
1941 pmd_t *pmd;
1942 unsigned long next;
1943
1944 pfn -= addr >> PAGE_SHIFT;
1945 pmd = pmd_alloc(mm, pud, addr);
1946 if (!pmd)
1947 return -ENOMEM;
Andrea Arcangelif66055ab2011-01-13 15:46:54 -08001948 VM_BUG_ON(pmd_trans_huge(*pmd));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 do {
1950 next = pmd_addr_end(addr, end);
1951 if (remap_pte_range(mm, pmd, addr, next,
1952 pfn + (addr >> PAGE_SHIFT), prot))
1953 return -ENOMEM;
1954 } while (pmd++, addr = next, addr != end);
1955 return 0;
1956}
1957
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001958static inline int remap_pud_range(struct mm_struct *mm, p4d_t *p4d,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 unsigned long addr, unsigned long end,
1960 unsigned long pfn, pgprot_t prot)
1961{
1962 pud_t *pud;
1963 unsigned long next;
1964
1965 pfn -= addr >> PAGE_SHIFT;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001966 pud = pud_alloc(mm, p4d, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 if (!pud)
1968 return -ENOMEM;
1969 do {
1970 next = pud_addr_end(addr, end);
1971 if (remap_pmd_range(mm, pud, addr, next,
1972 pfn + (addr >> PAGE_SHIFT), prot))
1973 return -ENOMEM;
1974 } while (pud++, addr = next, addr != end);
1975 return 0;
1976}
1977
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03001978static inline int remap_p4d_range(struct mm_struct *mm, pgd_t *pgd,
1979 unsigned long addr, unsigned long end,
1980 unsigned long pfn, pgprot_t prot)
1981{
1982 p4d_t *p4d;
1983 unsigned long next;
1984
1985 pfn -= addr >> PAGE_SHIFT;
1986 p4d = p4d_alloc(mm, pgd, addr);
1987 if (!p4d)
1988 return -ENOMEM;
1989 do {
1990 next = p4d_addr_end(addr, end);
1991 if (remap_pud_range(mm, p4d, addr, next,
1992 pfn + (addr >> PAGE_SHIFT), prot))
1993 return -ENOMEM;
1994 } while (p4d++, addr = next, addr != end);
1995 return 0;
1996}
1997
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001998/**
1999 * remap_pfn_range - remap kernel memory to userspace
2000 * @vma: user vma to map to
2001 * @addr: target user address to start at
2002 * @pfn: physical address of kernel memory
2003 * @size: size of map area
2004 * @prot: page protection flags for this mapping
2005 *
2006 * Note: this is only safe if the mm semaphore is held when called.
2007 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
2009 unsigned long pfn, unsigned long size, pgprot_t prot)
2010{
2011 pgd_t *pgd;
2012 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07002013 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 struct mm_struct *mm = vma->vm_mm;
Yongji Xied5957d22016-05-20 16:57:41 -07002015 unsigned long remap_pfn = pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 int err;
2017
2018 /*
2019 * Physically remapped pages are special. Tell the
2020 * rest of the world about it:
2021 * VM_IO tells people not to look at these pages
2022 * (accesses can have side effects).
Linus Torvalds6aab3412005-11-28 14:34:23 -08002023 * VM_PFNMAP tells the core MM that the base pages are just
2024 * raw PFN mappings, and do not have a "struct page" associated
2025 * with them.
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -07002026 * VM_DONTEXPAND
2027 * Disable vma merging and expanding with mremap().
2028 * VM_DONTDUMP
2029 * Omit vma from core dump, even when VM_IO turned off.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08002030 *
2031 * There's a horrible special case to handle copy-on-write
2032 * behaviour that some programs depend on. We mark the "original"
2033 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07002034 * See vm_normal_page() for details.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 */
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07002036 if (is_cow_mapping(vma->vm_flags)) {
2037 if (addr != vma->vm_start || end != vma->vm_end)
2038 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08002039 vma->vm_pgoff = pfn;
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07002040 }
2041
Yongji Xied5957d22016-05-20 16:57:41 -07002042 err = track_pfn_remap(vma, &prot, remap_pfn, addr, PAGE_ALIGN(size));
Konstantin Khlebnikovb3b9c292012-10-08 16:28:34 -07002043 if (err)
venkatesh.pallipadi@intel.com3c8bb732008-12-18 11:41:27 -08002044 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08002045
Konstantin Khlebnikov314e51b2012-10-08 16:29:02 -07002046 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047
2048 BUG_ON(addr >= end);
2049 pfn -= addr >> PAGE_SHIFT;
2050 pgd = pgd_offset(mm, addr);
2051 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 do {
2053 next = pgd_addr_end(addr, end);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03002054 err = remap_p4d_range(mm, pgd, addr, next,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 pfn + (addr >> PAGE_SHIFT), prot);
2056 if (err)
2057 break;
2058 } while (pgd++, addr = next, addr != end);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08002059
2060 if (err)
Yongji Xied5957d22016-05-20 16:57:41 -07002061 untrack_pfn(vma, remap_pfn, PAGE_ALIGN(size));
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08002062
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 return err;
2064}
2065EXPORT_SYMBOL(remap_pfn_range);
2066
Linus Torvaldsb4cbb192013-04-16 13:45:37 -07002067/**
2068 * vm_iomap_memory - remap memory to userspace
2069 * @vma: user vma to map to
2070 * @start: start of area
2071 * @len: size of area
2072 *
2073 * This is a simplified io_remap_pfn_range() for common driver use. The
2074 * driver just needs to give us the physical memory range to be mapped,
2075 * we'll figure out the rest from the vma information.
2076 *
2077 * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get
2078 * whatever write-combining details or similar.
2079 */
2080int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
2081{
2082 unsigned long vm_len, pfn, pages;
2083
2084 /* Check that the physical memory area passed in looks valid */
2085 if (start + len < start)
2086 return -EINVAL;
2087 /*
2088 * You *really* shouldn't map things that aren't page-aligned,
2089 * but we've historically allowed it because IO memory might
2090 * just have smaller alignment.
2091 */
2092 len += start & ~PAGE_MASK;
2093 pfn = start >> PAGE_SHIFT;
2094 pages = (len + ~PAGE_MASK) >> PAGE_SHIFT;
2095 if (pfn + pages < pfn)
2096 return -EINVAL;
2097
2098 /* We start the mapping 'vm_pgoff' pages into the area */
2099 if (vma->vm_pgoff > pages)
2100 return -EINVAL;
2101 pfn += vma->vm_pgoff;
2102 pages -= vma->vm_pgoff;
2103
2104 /* Can we fit all of the mapping? */
2105 vm_len = vma->vm_end - vma->vm_start;
2106 if (vm_len >> PAGE_SHIFT > pages)
2107 return -EINVAL;
2108
2109 /* Ok, let it rip */
2110 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
2111}
2112EXPORT_SYMBOL(vm_iomap_memory);
2113
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002114static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
2115 unsigned long addr, unsigned long end,
2116 pte_fn_t fn, void *data)
2117{
2118 pte_t *pte;
2119 int err;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08002120 pgtable_t token;
Borislav Petkov94909912007-05-06 14:49:17 -07002121 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002122
2123 pte = (mm == &init_mm) ?
2124 pte_alloc_kernel(pmd, addr) :
2125 pte_alloc_map_lock(mm, pmd, addr, &ptl);
2126 if (!pte)
2127 return -ENOMEM;
2128
2129 BUG_ON(pmd_huge(*pmd));
2130
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08002131 arch_enter_lazy_mmu_mode();
2132
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08002133 token = pmd_pgtable(*pmd);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002134
2135 do {
Daisuke Nishimurac36987e2009-10-26 16:50:23 -07002136 err = fn(pte++, token, addr, data);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002137 if (err)
2138 break;
Daisuke Nishimurac36987e2009-10-26 16:50:23 -07002139 } while (addr += PAGE_SIZE, addr != end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002140
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08002141 arch_leave_lazy_mmu_mode();
2142
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002143 if (mm != &init_mm)
2144 pte_unmap_unlock(pte-1, ptl);
2145 return err;
2146}
2147
2148static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
2149 unsigned long addr, unsigned long end,
2150 pte_fn_t fn, void *data)
2151{
2152 pmd_t *pmd;
2153 unsigned long next;
2154 int err;
2155
Andi Kleenceb86872008-07-23 21:27:50 -07002156 BUG_ON(pud_huge(*pud));
2157
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002158 pmd = pmd_alloc(mm, pud, addr);
2159 if (!pmd)
2160 return -ENOMEM;
2161 do {
2162 next = pmd_addr_end(addr, end);
2163 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
2164 if (err)
2165 break;
2166 } while (pmd++, addr = next, addr != end);
2167 return err;
2168}
2169
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03002170static int apply_to_pud_range(struct mm_struct *mm, p4d_t *p4d,
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002171 unsigned long addr, unsigned long end,
2172 pte_fn_t fn, void *data)
2173{
2174 pud_t *pud;
2175 unsigned long next;
2176 int err;
2177
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03002178 pud = pud_alloc(mm, p4d, addr);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002179 if (!pud)
2180 return -ENOMEM;
2181 do {
2182 next = pud_addr_end(addr, end);
2183 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
2184 if (err)
2185 break;
2186 } while (pud++, addr = next, addr != end);
2187 return err;
2188}
2189
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03002190static int apply_to_p4d_range(struct mm_struct *mm, pgd_t *pgd,
2191 unsigned long addr, unsigned long end,
2192 pte_fn_t fn, void *data)
2193{
2194 p4d_t *p4d;
2195 unsigned long next;
2196 int err;
2197
2198 p4d = p4d_alloc(mm, pgd, addr);
2199 if (!p4d)
2200 return -ENOMEM;
2201 do {
2202 next = p4d_addr_end(addr, end);
2203 err = apply_to_pud_range(mm, p4d, addr, next, fn, data);
2204 if (err)
2205 break;
2206 } while (p4d++, addr = next, addr != end);
2207 return err;
2208}
2209
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002210/*
2211 * Scan a region of virtual memory, filling in page tables as necessary
2212 * and calling a provided function on each leaf page table.
2213 */
2214int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
2215 unsigned long size, pte_fn_t fn, void *data)
2216{
2217 pgd_t *pgd;
2218 unsigned long next;
Jeremy Fitzhardinge57250a52010-08-09 17:19:52 -07002219 unsigned long end = addr + size;
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002220 int err;
2221
Mika Penttilä9cb65bc32016-03-15 14:56:45 -07002222 if (WARN_ON(addr >= end))
2223 return -EINVAL;
2224
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002225 pgd = pgd_offset(mm, addr);
2226 do {
2227 next = pgd_addr_end(addr, end);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03002228 err = apply_to_p4d_range(mm, pgd, addr, next, fn, data);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002229 if (err)
2230 break;
2231 } while (pgd++, addr = next, addr != end);
Jeremy Fitzhardinge57250a52010-08-09 17:19:52 -07002232
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002233 return err;
2234}
2235EXPORT_SYMBOL_GPL(apply_to_page_range);
2236
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237/*
Kirill A. Shutemov9b4bdd22015-02-10 14:09:51 -08002238 * handle_pte_fault chooses page fault handler according to an entry which was
2239 * read non-atomically. Before making any commitment, on those architectures
2240 * or configurations (e.g. i386 with PAE) which might give a mix of unmatched
2241 * parts, do_swap_page must check under lock before unmapping the pte and
2242 * proceeding (but do_wp_page is only called after already making such a check;
Ryota Ozakia335b2e2011-02-10 13:56:28 +09002243 * and do_anonymous_page can safely check later on).
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002244 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002245static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002246 pte_t *page_table, pte_t orig_pte)
2247{
2248 int same = 1;
2249#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
2250 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002251 spinlock_t *ptl = pte_lockptr(mm, pmd);
2252 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002253 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002254 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002255 }
2256#endif
2257 pte_unmap(page_table);
2258 return same;
2259}
2260
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00002261static 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 -08002262{
Dan Williams0abdd7a2014-01-21 15:48:12 -08002263 debug_dma_assert_idle(src);
2264
Linus Torvalds6aab3412005-11-28 14:34:23 -08002265 /*
2266 * If the source page was a PFN mapping, we don't have
2267 * a "struct page" for it. We do a best-effort copy by
2268 * just copying from the original user address. If that
2269 * fails, we just zero-fill it. Live with it.
2270 */
2271 if (unlikely(!src)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08002272 void *kaddr = kmap_atomic(dst);
Linus Torvalds5d2a2dbbc2005-11-29 14:07:55 -08002273 void __user *uaddr = (void __user *)(va & PAGE_MASK);
2274
2275 /*
2276 * This really shouldn't fail, because the page is there
2277 * in the page tables. But it might just be unreadable,
2278 * in which case we just give up and fill the result with
2279 * zeroes.
2280 */
2281 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Jan Beulich3ecb01d2010-10-26 14:22:27 -07002282 clear_page(kaddr);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002283 kunmap_atomic(kaddr);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07002284 flush_dcache_page(dst);
Nick Piggin0ed361d2008-02-04 22:29:34 -08002285 } else
2286 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08002287}
2288
Michal Hockoc20cd452016-01-14 15:20:12 -08002289static gfp_t __get_fault_gfp_mask(struct vm_area_struct *vma)
2290{
2291 struct file *vm_file = vma->vm_file;
2292
2293 if (vm_file)
2294 return mapping_gfp_mask(vm_file->f_mapping) | __GFP_FS | __GFP_IO;
2295
2296 /*
2297 * Special mappings (e.g. VDSO) do not have any file so fake
2298 * a default GFP_KERNEL for them.
2299 */
2300 return GFP_KERNEL;
2301}
2302
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303/*
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07002304 * Notify the address space that the page is about to become writable so that
2305 * it can prohibit this or wait for the page to get into an appropriate state.
2306 *
2307 * We do this without the lock held, so that it can sleep if it needs to.
2308 */
Jan Kara38b8cb72016-12-14 15:07:30 -08002309static int do_page_mkwrite(struct vm_fault *vmf)
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07002310{
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07002311 int ret;
Jan Kara38b8cb72016-12-14 15:07:30 -08002312 struct page *page = vmf->page;
2313 unsigned int old_flags = vmf->flags;
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07002314
Jan Kara38b8cb72016-12-14 15:07:30 -08002315 vmf->flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07002316
Dave Jiang11bac802017-02-24 14:56:41 -08002317 ret = vmf->vma->vm_ops->page_mkwrite(vmf);
Jan Kara38b8cb72016-12-14 15:07:30 -08002318 /* Restore original flags so that caller is not surprised */
2319 vmf->flags = old_flags;
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07002320 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2321 return ret;
2322 if (unlikely(!(ret & VM_FAULT_LOCKED))) {
2323 lock_page(page);
2324 if (!page->mapping) {
2325 unlock_page(page);
2326 return 0; /* retry */
2327 }
2328 ret |= VM_FAULT_LOCKED;
2329 } else
2330 VM_BUG_ON_PAGE(!PageLocked(page), page);
2331 return ret;
2332}
2333
2334/*
Jan Kara97ba0c22016-12-14 15:07:27 -08002335 * Handle dirtying of a page in shared file mapping on a write fault.
2336 *
2337 * The function expects the page to be locked and unlocks it.
2338 */
2339static void fault_dirty_shared_page(struct vm_area_struct *vma,
2340 struct page *page)
2341{
2342 struct address_space *mapping;
2343 bool dirtied;
2344 bool page_mkwrite = vma->vm_ops && vma->vm_ops->page_mkwrite;
2345
2346 dirtied = set_page_dirty(page);
2347 VM_BUG_ON_PAGE(PageAnon(page), page);
2348 /*
2349 * Take a local copy of the address_space - page.mapping may be zeroed
2350 * by truncate after unlock_page(). The address_space itself remains
2351 * pinned by vma->vm_file's reference. We rely on unlock_page()'s
2352 * release semantics to prevent the compiler from undoing this copying.
2353 */
2354 mapping = page_rmapping(page);
2355 unlock_page(page);
2356
2357 if ((dirtied || page_mkwrite) && mapping) {
2358 /*
2359 * Some device drivers do not set page.mapping
2360 * but still dirty their pages
2361 */
2362 balance_dirty_pages_ratelimited(mapping);
2363 }
2364
2365 if (!page_mkwrite)
2366 file_update_time(vma->vm_file);
2367}
2368
2369/*
Shachar Raindel4e047f82015-04-14 15:46:25 -07002370 * Handle write page faults for pages that can be reused in the current vma
2371 *
2372 * This can happen either due to the mapping being with the VM_SHARED flag,
2373 * or due to us being the last reference standing to the page. In either
2374 * case, all we need to do here is to mark the page as writable and update
2375 * any related book-keeping.
2376 */
Jan Kara997dd982016-12-14 15:07:36 -08002377static inline void wp_page_reuse(struct vm_fault *vmf)
Jan Kara82b0f8c2016-12-14 15:06:58 -08002378 __releases(vmf->ptl)
Shachar Raindel4e047f82015-04-14 15:46:25 -07002379{
Jan Kara82b0f8c2016-12-14 15:06:58 -08002380 struct vm_area_struct *vma = vmf->vma;
Jan Karaa41b70d2016-12-14 15:07:33 -08002381 struct page *page = vmf->page;
Shachar Raindel4e047f82015-04-14 15:46:25 -07002382 pte_t entry;
2383 /*
2384 * Clear the pages cpupid information as the existing
2385 * information potentially belongs to a now completely
2386 * unrelated process.
2387 */
2388 if (page)
2389 page_cpupid_xchg_last(page, (1 << LAST_CPUPID_SHIFT) - 1);
2390
Jan Kara29943022016-12-14 15:07:16 -08002391 flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte));
2392 entry = pte_mkyoung(vmf->orig_pte);
Shachar Raindel4e047f82015-04-14 15:46:25 -07002393 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Jan Kara82b0f8c2016-12-14 15:06:58 -08002394 if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1))
2395 update_mmu_cache(vma, vmf->address, vmf->pte);
2396 pte_unmap_unlock(vmf->pte, vmf->ptl);
Shachar Raindel4e047f82015-04-14 15:46:25 -07002397}
2398
2399/*
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002400 * Handle the case of a page which we actually need to copy to a new page.
2401 *
2402 * Called with mmap_sem locked and the old page referenced, but
2403 * without the ptl held.
2404 *
2405 * High level logic flow:
2406 *
2407 * - Allocate a page, copy the content of the old page to the new one.
2408 * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc.
2409 * - Take the PTL. If the pte changed, bail out and release the allocated page
2410 * - If the pte is still the way we remember it, update the page table and all
2411 * relevant references. This includes dropping the reference the page-table
2412 * held to the old page, as well as updating the rmap.
2413 * - In any case, unlock the PTL and drop the reference we took to the old page.
2414 */
Jan Karaa41b70d2016-12-14 15:07:33 -08002415static int wp_page_copy(struct vm_fault *vmf)
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002416{
Jan Kara82b0f8c2016-12-14 15:06:58 -08002417 struct vm_area_struct *vma = vmf->vma;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002418 struct mm_struct *mm = vma->vm_mm;
Jan Karaa41b70d2016-12-14 15:07:33 -08002419 struct page *old_page = vmf->page;
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002420 struct page *new_page = NULL;
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002421 pte_t entry;
2422 int page_copied = 0;
Jan Kara82b0f8c2016-12-14 15:06:58 -08002423 const unsigned long mmun_start = vmf->address & PAGE_MASK;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002424 const unsigned long mmun_end = mmun_start + PAGE_SIZE;
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002425 struct mem_cgroup *memcg;
2426
2427 if (unlikely(anon_vma_prepare(vma)))
2428 goto oom;
2429
Jan Kara29943022016-12-14 15:07:16 -08002430 if (is_zero_pfn(pte_pfn(vmf->orig_pte))) {
Jan Kara82b0f8c2016-12-14 15:06:58 -08002431 new_page = alloc_zeroed_user_highpage_movable(vma,
2432 vmf->address);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002433 if (!new_page)
2434 goto oom;
2435 } else {
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002436 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma,
Jan Kara82b0f8c2016-12-14 15:06:58 -08002437 vmf->address);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002438 if (!new_page)
2439 goto oom;
Jan Kara82b0f8c2016-12-14 15:06:58 -08002440 cow_user_page(new_page, old_page, vmf->address, vma);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002441 }
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002442
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08002443 if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg, false))
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002444 goto oom_free_new;
2445
Mel Gormaneb3c24f2015-06-24 16:57:27 -07002446 __SetPageUptodate(new_page);
2447
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002448 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
2449
2450 /*
2451 * Re-check the pte - we dropped the lock
2452 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08002453 vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl);
Jan Kara29943022016-12-14 15:07:16 -08002454 if (likely(pte_same(*vmf->pte, vmf->orig_pte))) {
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002455 if (old_page) {
2456 if (!PageAnon(old_page)) {
Jerome Marchandeca56ff2016-01-14 15:19:26 -08002457 dec_mm_counter_fast(mm,
2458 mm_counter_file(old_page));
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002459 inc_mm_counter_fast(mm, MM_ANONPAGES);
2460 }
2461 } else {
2462 inc_mm_counter_fast(mm, MM_ANONPAGES);
2463 }
Jan Kara29943022016-12-14 15:07:16 -08002464 flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte));
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002465 entry = mk_pte(new_page, vma->vm_page_prot);
2466 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2467 /*
2468 * Clear the pte entry and flush it first, before updating the
2469 * pte with the new entry. This will avoid a race condition
2470 * seen in the presence of one thread doing SMC and another
2471 * thread doing COW.
2472 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08002473 ptep_clear_flush_notify(vma, vmf->address, vmf->pte);
2474 page_add_new_anon_rmap(new_page, vma, vmf->address, false);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08002475 mem_cgroup_commit_charge(new_page, memcg, false, false);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002476 lru_cache_add_active_or_unevictable(new_page, vma);
2477 /*
2478 * We call the notify macro here because, when using secondary
2479 * mmu page tables (such as kvm shadow page tables), we want the
2480 * new page to be mapped directly into the secondary page table.
2481 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08002482 set_pte_at_notify(mm, vmf->address, vmf->pte, entry);
2483 update_mmu_cache(vma, vmf->address, vmf->pte);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002484 if (old_page) {
2485 /*
2486 * Only after switching the pte to the new page may
2487 * we remove the mapcount here. Otherwise another
2488 * process may come and find the rmap count decremented
2489 * before the pte is switched to the new page, and
2490 * "reuse" the old page writing into it while our pte
2491 * here still points into it and can be read by other
2492 * threads.
2493 *
2494 * The critical issue is to order this
2495 * page_remove_rmap with the ptp_clear_flush above.
2496 * Those stores are ordered by (if nothing else,)
2497 * the barrier present in the atomic_add_negative
2498 * in page_remove_rmap.
2499 *
2500 * Then the TLB flush in ptep_clear_flush ensures that
2501 * no process can access the old page before the
2502 * decremented mapcount is visible. And the old page
2503 * cannot be reused until after the decremented
2504 * mapcount is visible. So transitively, TLBs to
2505 * old page will be flushed before it can be reused.
2506 */
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08002507 page_remove_rmap(old_page, false);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002508 }
2509
2510 /* Free the old page.. */
2511 new_page = old_page;
2512 page_copied = 1;
2513 } else {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08002514 mem_cgroup_cancel_charge(new_page, memcg, false);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002515 }
2516
2517 if (new_page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002518 put_page(new_page);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002519
Jan Kara82b0f8c2016-12-14 15:06:58 -08002520 pte_unmap_unlock(vmf->pte, vmf->ptl);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002521 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
2522 if (old_page) {
2523 /*
2524 * Don't let another task, with possibly unlocked vma,
2525 * keep the mlocked page.
2526 */
2527 if (page_copied && (vma->vm_flags & VM_LOCKED)) {
2528 lock_page(old_page); /* LRU manipulation */
Kirill A. Shutemove90309c2016-01-15 16:54:33 -08002529 if (PageMlocked(old_page))
2530 munlock_vma_page(old_page);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002531 unlock_page(old_page);
2532 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002533 put_page(old_page);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002534 }
2535 return page_copied ? VM_FAULT_WRITE : 0;
2536oom_free_new:
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002537 put_page(new_page);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002538oom:
2539 if (old_page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002540 put_page(old_page);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002541 return VM_FAULT_OOM;
2542}
2543
Jan Kara66a61972016-12-14 15:07:39 -08002544/**
2545 * finish_mkwrite_fault - finish page fault for a shared mapping, making PTE
2546 * writeable once the page is prepared
2547 *
2548 * @vmf: structure describing the fault
2549 *
2550 * This function handles all that is needed to finish a write page fault in a
2551 * shared mapping due to PTE being read-only once the mapped page is prepared.
2552 * It handles locking of PTE and modifying it. The function returns
2553 * VM_FAULT_WRITE on success, 0 when PTE got changed before we acquired PTE
2554 * lock.
2555 *
2556 * The function expects the page to be locked or other protection against
2557 * concurrent faults / writeback (such as DAX radix tree locks).
2558 */
2559int finish_mkwrite_fault(struct vm_fault *vmf)
2560{
2561 WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED));
2562 vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address,
2563 &vmf->ptl);
2564 /*
2565 * We might have raced with another page fault while we released the
2566 * pte_offset_map_lock.
2567 */
2568 if (!pte_same(*vmf->pte, vmf->orig_pte)) {
2569 pte_unmap_unlock(vmf->pte, vmf->ptl);
Jan Karaa19e2552016-12-14 15:07:42 -08002570 return VM_FAULT_NOPAGE;
Jan Kara66a61972016-12-14 15:07:39 -08002571 }
2572 wp_page_reuse(vmf);
Jan Karaa19e2552016-12-14 15:07:42 -08002573 return 0;
Jan Kara66a61972016-12-14 15:07:39 -08002574}
2575
Boaz Harroshdd906182015-04-15 16:15:11 -07002576/*
2577 * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED
2578 * mapping
2579 */
Jan Kara29943022016-12-14 15:07:16 -08002580static int wp_pfn_shared(struct vm_fault *vmf)
Boaz Harroshdd906182015-04-15 16:15:11 -07002581{
Jan Kara82b0f8c2016-12-14 15:06:58 -08002582 struct vm_area_struct *vma = vmf->vma;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002583
Boaz Harroshdd906182015-04-15 16:15:11 -07002584 if (vma->vm_ops && vma->vm_ops->pfn_mkwrite) {
Boaz Harroshdd906182015-04-15 16:15:11 -07002585 int ret;
2586
Jan Kara82b0f8c2016-12-14 15:06:58 -08002587 pte_unmap_unlock(vmf->pte, vmf->ptl);
Jan Karafe822212016-12-14 15:07:13 -08002588 vmf->flags |= FAULT_FLAG_MKWRITE;
Dave Jiang11bac802017-02-24 14:56:41 -08002589 ret = vma->vm_ops->pfn_mkwrite(vmf);
Jan Kara2f89dc12016-12-14 15:07:50 -08002590 if (ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))
Boaz Harroshdd906182015-04-15 16:15:11 -07002591 return ret;
Jan Kara66a61972016-12-14 15:07:39 -08002592 return finish_mkwrite_fault(vmf);
Boaz Harroshdd906182015-04-15 16:15:11 -07002593 }
Jan Kara997dd982016-12-14 15:07:36 -08002594 wp_page_reuse(vmf);
2595 return VM_FAULT_WRITE;
Boaz Harroshdd906182015-04-15 16:15:11 -07002596}
2597
Jan Karaa41b70d2016-12-14 15:07:33 -08002598static int wp_page_shared(struct vm_fault *vmf)
Jan Kara82b0f8c2016-12-14 15:06:58 -08002599 __releases(vmf->ptl)
Shachar Raindel93e478d2015-04-14 15:46:35 -07002600{
Jan Kara82b0f8c2016-12-14 15:06:58 -08002601 struct vm_area_struct *vma = vmf->vma;
Shachar Raindel93e478d2015-04-14 15:46:35 -07002602
Jan Karaa41b70d2016-12-14 15:07:33 -08002603 get_page(vmf->page);
Shachar Raindel93e478d2015-04-14 15:46:35 -07002604
Shachar Raindel93e478d2015-04-14 15:46:35 -07002605 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
2606 int tmp;
2607
Jan Kara82b0f8c2016-12-14 15:06:58 -08002608 pte_unmap_unlock(vmf->pte, vmf->ptl);
Jan Kara38b8cb72016-12-14 15:07:30 -08002609 tmp = do_page_mkwrite(vmf);
Shachar Raindel93e478d2015-04-14 15:46:35 -07002610 if (unlikely(!tmp || (tmp &
2611 (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
Jan Karaa41b70d2016-12-14 15:07:33 -08002612 put_page(vmf->page);
Shachar Raindel93e478d2015-04-14 15:46:35 -07002613 return tmp;
2614 }
Jan Kara66a61972016-12-14 15:07:39 -08002615 tmp = finish_mkwrite_fault(vmf);
Jan Karaa19e2552016-12-14 15:07:42 -08002616 if (unlikely(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
Jan Karaa41b70d2016-12-14 15:07:33 -08002617 unlock_page(vmf->page);
Jan Karaa41b70d2016-12-14 15:07:33 -08002618 put_page(vmf->page);
Jan Kara66a61972016-12-14 15:07:39 -08002619 return tmp;
Shachar Raindel93e478d2015-04-14 15:46:35 -07002620 }
Jan Kara66a61972016-12-14 15:07:39 -08002621 } else {
2622 wp_page_reuse(vmf);
Jan Kara997dd982016-12-14 15:07:36 -08002623 lock_page(vmf->page);
Shachar Raindel93e478d2015-04-14 15:46:35 -07002624 }
Jan Kara997dd982016-12-14 15:07:36 -08002625 fault_dirty_shared_page(vma, vmf->page);
2626 put_page(vmf->page);
Shachar Raindel93e478d2015-04-14 15:46:35 -07002627
Jan Kara997dd982016-12-14 15:07:36 -08002628 return VM_FAULT_WRITE;
Shachar Raindel93e478d2015-04-14 15:46:35 -07002629}
2630
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002631/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 * This routine handles present pages, when users try to write
2633 * to a shared page. It is done by copying the page to a new address
2634 * and decrementing the shared-page counter for the old page.
2635 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 * Note that this routine assumes that the protection checks have been
2637 * done by the caller (the low-level page fault routine in most cases).
2638 * Thus we can safely just mark it writable once we've done any necessary
2639 * COW.
2640 *
2641 * We also mark the page dirty at this point even though the page will
2642 * change only once the write actually happens. This avoids a few races,
2643 * and potentially makes it more efficient.
2644 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002645 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2646 * but allow concurrent faults), with pte both mapped and locked.
2647 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 */
Jan Kara29943022016-12-14 15:07:16 -08002649static int do_wp_page(struct vm_fault *vmf)
Jan Kara82b0f8c2016-12-14 15:06:58 -08002650 __releases(vmf->ptl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651{
Jan Kara82b0f8c2016-12-14 15:06:58 -08002652 struct vm_area_struct *vma = vmf->vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653
Jan Karaa41b70d2016-12-14 15:07:33 -08002654 vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte);
2655 if (!vmf->page) {
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002656 /*
Peter Feiner64e455072014-10-13 15:55:46 -07002657 * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a
2658 * VM_PFNMAP VMA.
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002659 *
2660 * We should not cow pages in a shared writeable mapping.
Boaz Harroshdd906182015-04-15 16:15:11 -07002661 * Just mark the pages writable and/or call ops->pfn_mkwrite.
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002662 */
2663 if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
2664 (VM_WRITE|VM_SHARED))
Jan Kara29943022016-12-14 15:07:16 -08002665 return wp_pfn_shared(vmf);
Shachar Raindel2f38ab22015-04-14 15:46:32 -07002666
Jan Kara82b0f8c2016-12-14 15:06:58 -08002667 pte_unmap_unlock(vmf->pte, vmf->ptl);
Jan Karaa41b70d2016-12-14 15:07:33 -08002668 return wp_page_copy(vmf);
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002669 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002671 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002672 * Take out anonymous pages first, anonymous shared vmas are
2673 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002674 */
Jan Karaa41b70d2016-12-14 15:07:33 -08002675 if (PageAnon(vmf->page) && !PageKsm(vmf->page)) {
Huang Yingba3c4ce2017-09-06 16:22:19 -07002676 int total_map_swapcount;
Jan Karaa41b70d2016-12-14 15:07:33 -08002677 if (!trylock_page(vmf->page)) {
2678 get_page(vmf->page);
Jan Kara82b0f8c2016-12-14 15:06:58 -08002679 pte_unmap_unlock(vmf->pte, vmf->ptl);
Jan Karaa41b70d2016-12-14 15:07:33 -08002680 lock_page(vmf->page);
Jan Kara82b0f8c2016-12-14 15:06:58 -08002681 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
2682 vmf->address, &vmf->ptl);
Jan Kara29943022016-12-14 15:07:16 -08002683 if (!pte_same(*vmf->pte, vmf->orig_pte)) {
Jan Karaa41b70d2016-12-14 15:07:33 -08002684 unlock_page(vmf->page);
Jan Kara82b0f8c2016-12-14 15:06:58 -08002685 pte_unmap_unlock(vmf->pte, vmf->ptl);
Jan Karaa41b70d2016-12-14 15:07:33 -08002686 put_page(vmf->page);
Shachar Raindel28766802015-04-14 15:46:29 -07002687 return 0;
Hugh Dickinsab967d82009-01-06 14:39:33 -08002688 }
Jan Karaa41b70d2016-12-14 15:07:33 -08002689 put_page(vmf->page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002690 }
Huang Yingba3c4ce2017-09-06 16:22:19 -07002691 if (reuse_swap_page(vmf->page, &total_map_swapcount)) {
2692 if (total_map_swapcount == 1) {
Andrea Arcangeli6d0a07e2016-05-12 15:42:25 -07002693 /*
2694 * The page is all ours. Move it to
2695 * our anon_vma so the rmap code will
2696 * not search our parent or siblings.
2697 * Protected against the rmap code by
2698 * the page lock.
2699 */
Jan Karaa41b70d2016-12-14 15:07:33 -08002700 page_move_anon_rmap(vmf->page, vma);
Andrea Arcangeli6d0a07e2016-05-12 15:42:25 -07002701 }
Jan Karaa41b70d2016-12-14 15:07:33 -08002702 unlock_page(vmf->page);
Jan Kara997dd982016-12-14 15:07:36 -08002703 wp_page_reuse(vmf);
2704 return VM_FAULT_WRITE;
Michel Lespinasseb009c022011-01-13 15:46:07 -08002705 }
Jan Karaa41b70d2016-12-14 15:07:33 -08002706 unlock_page(vmf->page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002707 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002708 (VM_WRITE|VM_SHARED))) {
Jan Karaa41b70d2016-12-14 15:07:33 -08002709 return wp_page_shared(vmf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711
2712 /*
2713 * Ok, we need to copy. Oh, well..
2714 */
Jan Karaa41b70d2016-12-14 15:07:33 -08002715 get_page(vmf->page);
Shachar Raindel28766802015-04-14 15:46:29 -07002716
Jan Kara82b0f8c2016-12-14 15:06:58 -08002717 pte_unmap_unlock(vmf->pte, vmf->ptl);
Jan Karaa41b70d2016-12-14 15:07:33 -08002718 return wp_page_copy(vmf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719}
2720
Peter Zijlstra97a89412011-05-24 17:12:04 -07002721static void unmap_mapping_range_vma(struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722 unsigned long start_addr, unsigned long end_addr,
2723 struct zap_details *details)
2724{
Al Virof5cc4ee2012-03-05 14:14:20 -05002725 zap_page_range_single(vma, start_addr, end_addr - start_addr, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726}
2727
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07002728static inline void unmap_mapping_range_tree(struct rb_root *root,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729 struct zap_details *details)
2730{
2731 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 pgoff_t vba, vea, zba, zea;
2733
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07002734 vma_interval_tree_foreach(vma, root,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735 details->first_index, details->last_index) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736
2737 vba = vma->vm_pgoff;
Libind6e93212013-07-03 15:01:26 -07002738 vea = vba + vma_pages(vma) - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739 zba = details->first_index;
2740 if (zba < vba)
2741 zba = vba;
2742 zea = details->last_index;
2743 if (zea > vea)
2744 zea = vea;
2745
Peter Zijlstra97a89412011-05-24 17:12:04 -07002746 unmap_mapping_range_vma(vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
2748 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
Peter Zijlstra97a89412011-05-24 17:12:04 -07002749 details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750 }
2751}
2752
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753/**
Kirill A. Shutemov8a5f14a2015-02-10 14:09:49 -08002754 * unmap_mapping_range - unmap the portion of all mmaps in the specified
2755 * address_space corresponding to the specified page range in the underlying
2756 * file.
2757 *
Martin Waitz3d410882005-06-23 22:05:21 -07002758 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759 * @holebegin: byte in first page to unmap, relative to the start of
2760 * the underlying file. This will be rounded down to a PAGE_SIZE
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +10002761 * boundary. Note that this is different from truncate_pagecache(), which
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762 * must keep the partial page. In contrast, we must get rid of
2763 * partial pages.
2764 * @holelen: size of prospective hole in bytes. This will be rounded
2765 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2766 * end of the file.
2767 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2768 * but 0 when invalidating pagecache, don't throw away private data.
2769 */
2770void unmap_mapping_range(struct address_space *mapping,
2771 loff_t const holebegin, loff_t const holelen, int even_cows)
2772{
Michal Hockoaac45362016-03-25 14:20:24 -07002773 struct zap_details details = { };
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774 pgoff_t hba = holebegin >> PAGE_SHIFT;
2775 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2776
2777 /* Check for overflow. */
2778 if (sizeof(holelen) > sizeof(hlen)) {
2779 long long holeend =
2780 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2781 if (holeend & ~(long long)ULONG_MAX)
2782 hlen = ULONG_MAX - hba + 1;
2783 }
2784
Tobin C Harding166f61b2017-02-24 14:59:01 -08002785 details.check_mapping = even_cows ? NULL : mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 details.first_index = hba;
2787 details.last_index = hba + hlen - 1;
2788 if (details.last_index < details.first_index)
2789 details.last_index = ULONG_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790
Kirill A. Shutemov46c043e2015-09-08 14:59:42 -07002791 i_mmap_lock_write(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -07002792 if (unlikely(!RB_EMPTY_ROOT(&mapping->i_mmap)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793 unmap_mapping_range_tree(&mapping->i_mmap, &details);
Kirill A. Shutemov46c043e2015-09-08 14:59:42 -07002794 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795}
2796EXPORT_SYMBOL(unmap_mapping_range);
2797
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002799 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2800 * but allow concurrent faults), and pte mapped but not yet locked.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07002801 * We return with pte unmapped and unlocked.
2802 *
2803 * We return with the mmap_sem locked or unlocked in the same cases
2804 * as does filemap_fault().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805 */
Jan Kara29943022016-12-14 15:07:16 -08002806int do_swap_page(struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807{
Jan Kara82b0f8c2016-12-14 15:06:58 -08002808 struct vm_area_struct *vma = vmf->vma;
Huang Yingec560172017-09-06 16:24:36 -07002809 struct page *page = NULL, *swapcache;
Johannes Weiner00501b52014-08-08 14:19:20 -07002810 struct mem_cgroup *memcg;
Huang Yingec560172017-09-06 16:24:36 -07002811 struct vma_swap_readahead swap_ra;
Hugh Dickins65500d22005-10-29 18:15:59 -07002812 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813 pte_t pte;
Michel Lespinassed065bd82010-10-26 14:21:57 -07002814 int locked;
Rik van Rielad8c2ee2010-08-09 17:19:48 -07002815 int exclusive = 0;
Nick Piggin83c54072007-07-19 01:47:05 -07002816 int ret = 0;
Huang Yingec560172017-09-06 16:24:36 -07002817 bool vma_readahead = swap_use_vma_readahead();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818
Huang Yingec560172017-09-06 16:24:36 -07002819 if (vma_readahead)
2820 page = swap_readahead_detect(vmf, &swap_ra);
2821 if (!pte_unmap_same(vma->vm_mm, vmf->pmd, vmf->pte, vmf->orig_pte)) {
2822 if (page)
2823 put_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002824 goto out;
Huang Yingec560172017-09-06 16:24:36 -07002825 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002826
Jan Kara29943022016-12-14 15:07:16 -08002827 entry = pte_to_swp_entry(vmf->orig_pte);
Andi Kleend1737fd2009-09-16 11:50:06 +02002828 if (unlikely(non_swap_entry(entry))) {
2829 if (is_migration_entry(entry)) {
Jan Kara82b0f8c2016-12-14 15:06:58 -08002830 migration_entry_wait(vma->vm_mm, vmf->pmd,
2831 vmf->address);
Jérôme Glisse5042db42017-09-08 16:11:43 -07002832 } else if (is_device_private_entry(entry)) {
2833 /*
2834 * For un-addressable device memory we call the pgmap
2835 * fault handler callback. The callback must migrate
2836 * the page back to some CPU accessible page.
2837 */
2838 ret = device_private_entry_fault(vma, vmf->address, entry,
2839 vmf->flags, vmf->pmd);
Andi Kleend1737fd2009-09-16 11:50:06 +02002840 } else if (is_hwpoison_entry(entry)) {
2841 ret = VM_FAULT_HWPOISON;
2842 } else {
Jan Kara29943022016-12-14 15:07:16 -08002843 print_bad_pte(vma, vmf->address, vmf->orig_pte, NULL);
Hugh Dickinsd99be1a2009-12-14 17:59:04 -08002844 ret = VM_FAULT_SIGBUS;
Andi Kleend1737fd2009-09-16 11:50:06 +02002845 }
Christoph Lameter06972122006-06-23 02:03:35 -07002846 goto out;
2847 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002848 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Huang Yingec560172017-09-06 16:24:36 -07002849 if (!page)
2850 page = lookup_swap_cache(entry, vma_readahead ? vma : NULL,
2851 vmf->address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852 if (!page) {
Huang Yingec560172017-09-06 16:24:36 -07002853 if (vma_readahead)
2854 page = do_swap_page_readahead(entry,
2855 GFP_HIGHUSER_MOVABLE, vmf, &swap_ra);
2856 else
2857 page = swapin_readahead(entry,
2858 GFP_HIGHUSER_MOVABLE, vma, vmf->address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859 if (!page) {
2860 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002861 * Back out if somebody else faulted in this pte
2862 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08002864 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
2865 vmf->address, &vmf->ptl);
Jan Kara29943022016-12-14 15:07:16 -08002866 if (likely(pte_same(*vmf->pte, vmf->orig_pte)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002868 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002869 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870 }
2871
2872 /* Had to read the page from swap area: Major fault */
2873 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002874 count_vm_event(PGMAJFAULT);
Roman Gushchin22621852017-07-06 15:40:25 -07002875 count_memcg_event_mm(vma->vm_mm, PGMAJFAULT);
Andi Kleend1737fd2009-09-16 11:50:06 +02002876 } else if (PageHWPoison(page)) {
Wu Fengguang71f72522009-12-16 12:19:58 +01002877 /*
2878 * hwpoisoned dirty swapcache pages are kept for killing
2879 * owner processes (which may be unknown at hwpoison time)
2880 */
Andi Kleend1737fd2009-09-16 11:50:06 +02002881 ret = VM_FAULT_HWPOISON;
2882 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins56f31802013-02-22 16:36:10 -08002883 swapcache = page;
Andi Kleen4779cb32009-10-14 01:51:41 +02002884 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885 }
2886
Hugh Dickins56f31802013-02-22 16:36:10 -08002887 swapcache = page;
Jan Kara82b0f8c2016-12-14 15:06:58 -08002888 locked = lock_page_or_retry(page, vma->vm_mm, vmf->flags);
Rik van Riele709ffd2012-05-29 15:06:18 -07002889
Balbir Singh20a10222007-11-14 17:00:33 -08002890 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Michel Lespinassed065bd82010-10-26 14:21:57 -07002891 if (!locked) {
2892 ret |= VM_FAULT_RETRY;
2893 goto out_release;
2894 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002896 /*
Hugh Dickins31c4a3d2010-09-19 19:40:22 -07002897 * Make sure try_to_free_swap or reuse_swap_page or swapoff did not
2898 * release the swapcache from under us. The page pin, and pte_same
2899 * test below, are not enough to exclude that. Even if it is still
2900 * swapcache, we need to check that the page's swap has not changed.
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002901 */
Hugh Dickins31c4a3d2010-09-19 19:40:22 -07002902 if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val))
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002903 goto out_page;
2904
Jan Kara82b0f8c2016-12-14 15:06:58 -08002905 page = ksm_might_need_to_copy(page, vma, vmf->address);
Hugh Dickinscbf86cf2013-02-22 16:35:08 -08002906 if (unlikely(!page)) {
2907 ret = VM_FAULT_OOM;
2908 page = swapcache;
Hugh Dickinscbf86cf2013-02-22 16:35:08 -08002909 goto out_page;
Hugh Dickins5ad64682009-12-14 17:59:24 -08002910 }
2911
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002912 if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL,
2913 &memcg, false)) {
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002914 ret = VM_FAULT_OOM;
Johannes Weinerbc43f752009-04-30 15:08:08 -07002915 goto out_page;
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002916 }
2917
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002919 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002920 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08002921 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address,
2922 &vmf->ptl);
Jan Kara29943022016-12-14 15:07:16 -08002923 if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002924 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002925
2926 if (unlikely(!PageUptodate(page))) {
2927 ret = VM_FAULT_SIGBUS;
2928 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002929 }
2930
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002931 /*
2932 * The page isn't present yet, go ahead with the fault.
2933 *
2934 * Be careful about the sequence of operations here.
2935 * To get its accounting right, reuse_swap_page() must be called
2936 * while the page is counted on swap but not yet in mapcount i.e.
2937 * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
2938 * must be called after the swap_free(), or it will never succeed.
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002939 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002940
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002941 inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES);
2942 dec_mm_counter_fast(vma->vm_mm, MM_SWAPENTS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943 pte = mk_pte(page, vma->vm_page_prot);
Jan Kara82b0f8c2016-12-14 15:06:58 -08002944 if ((vmf->flags & FAULT_FLAG_WRITE) && reuse_swap_page(page, NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
Jan Kara82b0f8c2016-12-14 15:06:58 -08002946 vmf->flags &= ~FAULT_FLAG_WRITE;
Andrea Arcangeli9a5b4892010-08-09 17:19:49 -07002947 ret |= VM_FAULT_WRITE;
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08002948 exclusive = RMAP_EXCLUSIVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002950 flush_icache_page(vma, page);
Jan Kara29943022016-12-14 15:07:16 -08002951 if (pte_swp_soft_dirty(vmf->orig_pte))
Cyrill Gorcunov179ef712013-08-13 16:00:49 -07002952 pte = pte_mksoft_dirty(pte);
Jan Kara82b0f8c2016-12-14 15:06:58 -08002953 set_pte_at(vma->vm_mm, vmf->address, vmf->pte, pte);
Jan Kara29943022016-12-14 15:07:16 -08002954 vmf->orig_pte = pte;
Johannes Weiner00501b52014-08-08 14:19:20 -07002955 if (page == swapcache) {
Jan Kara82b0f8c2016-12-14 15:06:58 -08002956 do_page_add_anon_rmap(page, vma, vmf->address, exclusive);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08002957 mem_cgroup_commit_charge(page, memcg, true, false);
Minchan Kim1a8018f2016-08-02 14:02:25 -07002958 activate_page(page);
Johannes Weiner00501b52014-08-08 14:19:20 -07002959 } else { /* ksm created a completely new copy */
Jan Kara82b0f8c2016-12-14 15:06:58 -08002960 page_add_new_anon_rmap(page, vma, vmf->address, false);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08002961 mem_cgroup_commit_charge(page, memcg, false, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07002962 lru_cache_add_active_or_unevictable(page, vma);
2963 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002964
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002965 swap_free(entry);
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08002966 if (mem_cgroup_swap_full(page) ||
2967 (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08002968 try_to_free_swap(page);
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002969 unlock_page(page);
Hugh Dickins56f31802013-02-22 16:36:10 -08002970 if (page != swapcache) {
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002971 /*
2972 * Hold the lock to avoid the swap entry to be reused
2973 * until we take the PT lock for the pte_same() check
2974 * (to avoid false positives from pte_same). For
2975 * further safety release the lock after the swap_free
2976 * so that the swap count won't change under a
2977 * parallel locked swapcache.
2978 */
2979 unlock_page(swapcache);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002980 put_page(swapcache);
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002981 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002982
Jan Kara82b0f8c2016-12-14 15:06:58 -08002983 if (vmf->flags & FAULT_FLAG_WRITE) {
Jan Kara29943022016-12-14 15:07:16 -08002984 ret |= do_wp_page(vmf);
Hugh Dickins61469f12008-03-04 14:29:04 -08002985 if (ret & VM_FAULT_ERROR)
2986 ret &= VM_FAULT_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987 goto out;
2988 }
2989
2990 /* No need to invalidate - it was non-present before */
Jan Kara82b0f8c2016-12-14 15:06:58 -08002991 update_mmu_cache(vma, vmf->address, vmf->pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07002992unlock:
Jan Kara82b0f8c2016-12-14 15:06:58 -08002993 pte_unmap_unlock(vmf->pte, vmf->ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994out:
2995 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002996out_nomap:
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08002997 mem_cgroup_cancel_charge(page, memcg, false);
Jan Kara82b0f8c2016-12-14 15:06:58 -08002998 pte_unmap_unlock(vmf->pte, vmf->ptl);
Johannes Weinerbc43f752009-04-30 15:08:08 -07002999out_page:
Kirill Korotaevb8107482005-05-16 21:53:50 -07003000 unlock_page(page);
Andi Kleen4779cb32009-10-14 01:51:41 +02003001out_release:
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003002 put_page(page);
Hugh Dickins56f31802013-02-22 16:36:10 -08003003 if (page != swapcache) {
Andrea Arcangeli4969c112010-09-09 16:37:52 -07003004 unlock_page(swapcache);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003005 put_page(swapcache);
Andrea Arcangeli4969c112010-09-09 16:37:52 -07003006 }
Hugh Dickins65500d22005-10-29 18:15:59 -07003007 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008}
3009
3010/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003011 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3012 * but allow concurrent faults), and pte mapped but not yet locked.
3013 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003015static int do_anonymous_page(struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003016{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003017 struct vm_area_struct *vma = vmf->vma;
Johannes Weiner00501b52014-08-08 14:19:20 -07003018 struct mem_cgroup *memcg;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003019 struct page *page;
Michal Hocko6b31d592017-08-18 15:16:15 -07003020 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003022
Kirill A. Shutemov6b7339f2015-07-06 23:18:37 +03003023 /* File mapping without ->vm_ops ? */
3024 if (vma->vm_flags & VM_SHARED)
3025 return VM_FAULT_SIGBUS;
3026
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003027 /*
3028 * Use pte_alloc() instead of pte_alloc_map(). We can't run
3029 * pte_offset_map() on pmds where a huge pmd might be created
3030 * from a different thread.
3031 *
3032 * pte_alloc_map() is safe to use under down_write(mmap_sem) or when
3033 * parallel threads are excluded by other means.
3034 *
3035 * Here we only have down_read(mmap_sem).
3036 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003037 if (pte_alloc(vma->vm_mm, vmf->pmd, vmf->address))
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003038 return VM_FAULT_OOM;
3039
3040 /* See the comment in pte_alloc_one_map() */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003041 if (unlikely(pmd_trans_unstable(vmf->pmd)))
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003042 return 0;
3043
Linus Torvalds11ac5522010-08-14 11:44:56 -07003044 /* Use the zero-page for reads */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003045 if (!(vmf->flags & FAULT_FLAG_WRITE) &&
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003046 !mm_forbids_zeropage(vma->vm_mm)) {
Jan Kara82b0f8c2016-12-14 15:06:58 -08003047 entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address),
Hugh Dickins62eede62009-09-21 17:03:34 -07003048 vma->vm_page_prot));
Jan Kara82b0f8c2016-12-14 15:06:58 -08003049 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
3050 vmf->address, &vmf->ptl);
3051 if (!pte_none(*vmf->pte))
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07003052 goto unlock;
Michal Hocko6b31d592017-08-18 15:16:15 -07003053 ret = check_stable_address_space(vma->vm_mm);
3054 if (ret)
3055 goto unlock;
Andrea Arcangeli6b251fc2015-09-04 15:46:20 -07003056 /* Deliver the page fault to userland, check inside PT lock */
3057 if (userfaultfd_missing(vma)) {
Jan Kara82b0f8c2016-12-14 15:06:58 -08003058 pte_unmap_unlock(vmf->pte, vmf->ptl);
3059 return handle_userfault(vmf, VM_UFFD_MISSING);
Andrea Arcangeli6b251fc2015-09-04 15:46:20 -07003060 }
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07003061 goto setpte;
3062 }
3063
Nick Piggin557ed1f2007-10-16 01:24:40 -07003064 /* Allocate our own private page. */
Nick Piggin557ed1f2007-10-16 01:24:40 -07003065 if (unlikely(anon_vma_prepare(vma)))
3066 goto oom;
Jan Kara82b0f8c2016-12-14 15:06:58 -08003067 page = alloc_zeroed_user_highpage_movable(vma, vmf->address);
Nick Piggin557ed1f2007-10-16 01:24:40 -07003068 if (!page)
3069 goto oom;
Mel Gormaneb3c24f2015-06-24 16:57:27 -07003070
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003071 if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL, &memcg, false))
Mel Gormaneb3c24f2015-06-24 16:57:27 -07003072 goto oom_free_page;
3073
Minchan Kim52f37622013-04-29 15:08:15 -07003074 /*
3075 * The memory barrier inside __SetPageUptodate makes sure that
3076 * preceeding stores to the page contents become visible before
3077 * the set_pte_at() write.
3078 */
Nick Piggin0ed361d2008-02-04 22:29:34 -08003079 __SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080
Nick Piggin557ed1f2007-10-16 01:24:40 -07003081 entry = mk_pte(page, vma->vm_page_prot);
Hugh Dickins1ac0cb52009-09-21 17:03:29 -07003082 if (vma->vm_flags & VM_WRITE)
3083 entry = pte_mkwrite(pte_mkdirty(entry));
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003084
Jan Kara82b0f8c2016-12-14 15:06:58 -08003085 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address,
3086 &vmf->ptl);
3087 if (!pte_none(*vmf->pte))
Nick Piggin557ed1f2007-10-16 01:24:40 -07003088 goto release;
Hugh Dickins9ba69292009-09-21 17:02:20 -07003089
Michal Hocko6b31d592017-08-18 15:16:15 -07003090 ret = check_stable_address_space(vma->vm_mm);
3091 if (ret)
3092 goto release;
3093
Andrea Arcangeli6b251fc2015-09-04 15:46:20 -07003094 /* Deliver the page fault to userland, check inside PT lock */
3095 if (userfaultfd_missing(vma)) {
Jan Kara82b0f8c2016-12-14 15:06:58 -08003096 pte_unmap_unlock(vmf->pte, vmf->ptl);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08003097 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003098 put_page(page);
Jan Kara82b0f8c2016-12-14 15:06:58 -08003099 return handle_userfault(vmf, VM_UFFD_MISSING);
Andrea Arcangeli6b251fc2015-09-04 15:46:20 -07003100 }
3101
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003102 inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES);
Jan Kara82b0f8c2016-12-14 15:06:58 -08003103 page_add_new_anon_rmap(page, vma, vmf->address, false);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08003104 mem_cgroup_commit_charge(page, memcg, false, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07003105 lru_cache_add_active_or_unevictable(page, vma);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07003106setpte:
Jan Kara82b0f8c2016-12-14 15:06:58 -08003107 set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108
3109 /* No need to invalidate - it was non-present before */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003110 update_mmu_cache(vma, vmf->address, vmf->pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07003111unlock:
Jan Kara82b0f8c2016-12-14 15:06:58 -08003112 pte_unmap_unlock(vmf->pte, vmf->ptl);
Michal Hocko6b31d592017-08-18 15:16:15 -07003113 return ret;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003114release:
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08003115 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003116 put_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003117 goto unlock;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08003118oom_free_page:
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003119 put_page(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07003120oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003121 return VM_FAULT_OOM;
3122}
3123
Paul Cassella9a95f3c2014-08-06 16:07:24 -07003124/*
3125 * The mmap_sem must have been held on entry, and may have been
3126 * released depending on flags and vma->vm_ops->fault() return value.
3127 * See filemap_fault() and __lock_page_retry().
3128 */
Jan Kara936ca802016-12-14 15:07:10 -08003129static int __do_fault(struct vm_fault *vmf)
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07003130{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003131 struct vm_area_struct *vma = vmf->vma;
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07003132 int ret;
3133
Dave Jiang11bac802017-02-24 14:56:41 -08003134 ret = vma->vm_ops->fault(vmf);
Jan Kara39170482016-12-14 15:07:18 -08003135 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY |
Jan Karab1aa8122016-12-14 15:07:24 -08003136 VM_FAULT_DONE_COW)))
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07003137 return ret;
3138
Jan Kara667240e2016-12-14 15:07:07 -08003139 if (unlikely(PageHWPoison(vmf->page))) {
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07003140 if (ret & VM_FAULT_LOCKED)
Jan Kara667240e2016-12-14 15:07:07 -08003141 unlock_page(vmf->page);
3142 put_page(vmf->page);
Jan Kara936ca802016-12-14 15:07:10 -08003143 vmf->page = NULL;
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07003144 return VM_FAULT_HWPOISON;
3145 }
3146
3147 if (unlikely(!(ret & VM_FAULT_LOCKED)))
Jan Kara667240e2016-12-14 15:07:07 -08003148 lock_page(vmf->page);
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07003149 else
Jan Kara667240e2016-12-14 15:07:07 -08003150 VM_BUG_ON_PAGE(!PageLocked(vmf->page), vmf->page);
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07003151
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07003152 return ret;
3153}
3154
Ross Zwislerd0f0931d2017-06-02 14:46:34 -07003155/*
3156 * The ordering of these checks is important for pmds with _PAGE_DEVMAP set.
3157 * If we check pmd_trans_unstable() first we will trip the bad_pmd() check
3158 * inside of pmd_none_or_trans_huge_or_clear_bad(). This will end up correctly
3159 * returning 1 but not before it spams dmesg with the pmd_clear_bad() output.
3160 */
3161static int pmd_devmap_trans_unstable(pmd_t *pmd)
3162{
3163 return pmd_devmap(*pmd) || pmd_trans_unstable(pmd);
3164}
3165
Jan Kara82b0f8c2016-12-14 15:06:58 -08003166static int pte_alloc_one_map(struct vm_fault *vmf)
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003167{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003168 struct vm_area_struct *vma = vmf->vma;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003169
Jan Kara82b0f8c2016-12-14 15:06:58 -08003170 if (!pmd_none(*vmf->pmd))
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003171 goto map_pte;
Jan Kara82b0f8c2016-12-14 15:06:58 -08003172 if (vmf->prealloc_pte) {
3173 vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd);
3174 if (unlikely(!pmd_none(*vmf->pmd))) {
3175 spin_unlock(vmf->ptl);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003176 goto map_pte;
3177 }
3178
3179 atomic_long_inc(&vma->vm_mm->nr_ptes);
Jan Kara82b0f8c2016-12-14 15:06:58 -08003180 pmd_populate(vma->vm_mm, vmf->pmd, vmf->prealloc_pte);
3181 spin_unlock(vmf->ptl);
Tobin C Harding7f2b6ce2017-02-24 14:58:59 -08003182 vmf->prealloc_pte = NULL;
Jan Kara82b0f8c2016-12-14 15:06:58 -08003183 } else if (unlikely(pte_alloc(vma->vm_mm, vmf->pmd, vmf->address))) {
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003184 return VM_FAULT_OOM;
3185 }
3186map_pte:
3187 /*
3188 * If a huge pmd materialized under us just retry later. Use
Ross Zwislerd0f0931d2017-06-02 14:46:34 -07003189 * pmd_trans_unstable() via pmd_devmap_trans_unstable() instead of
3190 * pmd_trans_huge() to ensure the pmd didn't become pmd_trans_huge
3191 * under us and then back to pmd_none, as a result of MADV_DONTNEED
3192 * running immediately after a huge pmd fault in a different thread of
3193 * this mm, in turn leading to a misleading pmd_trans_huge() retval.
3194 * All we have to ensure is that it is a regular pmd that we can walk
3195 * with pte_offset_map() and we can do that through an atomic read in
3196 * C, which is what pmd_trans_unstable() provides.
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003197 */
Ross Zwislerd0f0931d2017-06-02 14:46:34 -07003198 if (pmd_devmap_trans_unstable(vmf->pmd))
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003199 return VM_FAULT_NOPAGE;
3200
Ross Zwislerd0f0931d2017-06-02 14:46:34 -07003201 /*
3202 * At this point we know that our vmf->pmd points to a page of ptes
3203 * and it cannot become pmd_none(), pmd_devmap() or pmd_trans_huge()
3204 * for the duration of the fault. If a racing MADV_DONTNEED runs and
3205 * we zap the ptes pointed to by our vmf->pmd, the vmf->ptl will still
3206 * be valid and we will re-check to make sure the vmf->pte isn't
3207 * pte_none() under vmf->ptl protection when we return to
3208 * alloc_set_pte().
3209 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003210 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address,
3211 &vmf->ptl);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003212 return 0;
3213}
3214
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003215#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003216
3217#define HPAGE_CACHE_INDEX_MASK (HPAGE_PMD_NR - 1)
3218static inline bool transhuge_vma_suitable(struct vm_area_struct *vma,
3219 unsigned long haddr)
3220{
3221 if (((vma->vm_start >> PAGE_SHIFT) & HPAGE_CACHE_INDEX_MASK) !=
3222 (vma->vm_pgoff & HPAGE_CACHE_INDEX_MASK))
3223 return false;
3224 if (haddr < vma->vm_start || haddr + HPAGE_PMD_SIZE > vma->vm_end)
3225 return false;
3226 return true;
3227}
3228
Jan Kara82b0f8c2016-12-14 15:06:58 -08003229static void deposit_prealloc_pte(struct vm_fault *vmf)
Aneesh Kumar K.V953c66c2016-12-12 16:44:32 -08003230{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003231 struct vm_area_struct *vma = vmf->vma;
Aneesh Kumar K.V953c66c2016-12-12 16:44:32 -08003232
Jan Kara82b0f8c2016-12-14 15:06:58 -08003233 pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, vmf->prealloc_pte);
Aneesh Kumar K.V953c66c2016-12-12 16:44:32 -08003234 /*
3235 * We are going to consume the prealloc table,
3236 * count that as nr_ptes.
3237 */
3238 atomic_long_inc(&vma->vm_mm->nr_ptes);
Tobin C Harding7f2b6ce2017-02-24 14:58:59 -08003239 vmf->prealloc_pte = NULL;
Aneesh Kumar K.V953c66c2016-12-12 16:44:32 -08003240}
3241
Jan Kara82b0f8c2016-12-14 15:06:58 -08003242static int do_set_pmd(struct vm_fault *vmf, struct page *page)
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003243{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003244 struct vm_area_struct *vma = vmf->vma;
3245 bool write = vmf->flags & FAULT_FLAG_WRITE;
3246 unsigned long haddr = vmf->address & HPAGE_PMD_MASK;
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003247 pmd_t entry;
3248 int i, ret;
3249
3250 if (!transhuge_vma_suitable(vma, haddr))
3251 return VM_FAULT_FALLBACK;
3252
3253 ret = VM_FAULT_FALLBACK;
3254 page = compound_head(page);
3255
Aneesh Kumar K.V953c66c2016-12-12 16:44:32 -08003256 /*
3257 * Archs like ppc64 need additonal space to store information
3258 * related to pte entry. Use the preallocated table for that.
3259 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003260 if (arch_needs_pgtable_deposit() && !vmf->prealloc_pte) {
3261 vmf->prealloc_pte = pte_alloc_one(vma->vm_mm, vmf->address);
3262 if (!vmf->prealloc_pte)
Aneesh Kumar K.V953c66c2016-12-12 16:44:32 -08003263 return VM_FAULT_OOM;
3264 smp_wmb(); /* See comment in __pte_alloc() */
3265 }
3266
Jan Kara82b0f8c2016-12-14 15:06:58 -08003267 vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd);
3268 if (unlikely(!pmd_none(*vmf->pmd)))
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003269 goto out;
3270
3271 for (i = 0; i < HPAGE_PMD_NR; i++)
3272 flush_icache_page(vma, page + i);
3273
3274 entry = mk_huge_pmd(page, vma->vm_page_prot);
3275 if (write)
3276 entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma);
3277
3278 add_mm_counter(vma->vm_mm, MM_FILEPAGES, HPAGE_PMD_NR);
3279 page_add_file_rmap(page, true);
Aneesh Kumar K.V953c66c2016-12-12 16:44:32 -08003280 /*
3281 * deposit and withdraw with pmd lock held
3282 */
3283 if (arch_needs_pgtable_deposit())
Jan Kara82b0f8c2016-12-14 15:06:58 -08003284 deposit_prealloc_pte(vmf);
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003285
Jan Kara82b0f8c2016-12-14 15:06:58 -08003286 set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry);
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003287
Jan Kara82b0f8c2016-12-14 15:06:58 -08003288 update_mmu_cache_pmd(vma, haddr, vmf->pmd);
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003289
3290 /* fault is handled */
3291 ret = 0;
Kirill A. Shutemov95ecedc2016-07-26 15:25:31 -07003292 count_vm_event(THP_FILE_MAPPED);
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003293out:
Jan Kara82b0f8c2016-12-14 15:06:58 -08003294 spin_unlock(vmf->ptl);
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003295 return ret;
3296}
3297#else
Jan Kara82b0f8c2016-12-14 15:06:58 -08003298static int do_set_pmd(struct vm_fault *vmf, struct page *page)
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003299{
3300 BUILD_BUG();
3301 return 0;
3302}
3303#endif
3304
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003305/**
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003306 * alloc_set_pte - setup new PTE entry for given page and add reverse page
3307 * mapping. If needed, the fucntion allocates page table or use pre-allocated.
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003308 *
Jan Kara82b0f8c2016-12-14 15:06:58 -08003309 * @vmf: fault environment
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003310 * @memcg: memcg to charge page (only for private mappings)
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003311 * @page: page to map
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003312 *
Jan Kara82b0f8c2016-12-14 15:06:58 -08003313 * Caller must take care of unlocking vmf->ptl, if vmf->pte is non-NULL on
3314 * return.
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003315 *
3316 * Target users are page handler itself and implementations of
3317 * vm_ops->map_pages.
3318 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003319int alloc_set_pte(struct vm_fault *vmf, struct mem_cgroup *memcg,
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003320 struct page *page)
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003321{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003322 struct vm_area_struct *vma = vmf->vma;
3323 bool write = vmf->flags & FAULT_FLAG_WRITE;
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003324 pte_t entry;
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003325 int ret;
3326
Jan Kara82b0f8c2016-12-14 15:06:58 -08003327 if (pmd_none(*vmf->pmd) && PageTransCompound(page) &&
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003328 IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003329 /* THP on COW? */
3330 VM_BUG_ON_PAGE(memcg, page);
3331
Jan Kara82b0f8c2016-12-14 15:06:58 -08003332 ret = do_set_pmd(vmf, page);
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003333 if (ret != VM_FAULT_FALLBACK)
Hugh Dickinsb0b9b3d2017-01-07 15:37:31 -08003334 return ret;
Kirill A. Shutemov10102452016-07-26 15:25:29 -07003335 }
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003336
Jan Kara82b0f8c2016-12-14 15:06:58 -08003337 if (!vmf->pte) {
3338 ret = pte_alloc_one_map(vmf);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003339 if (ret)
Hugh Dickinsb0b9b3d2017-01-07 15:37:31 -08003340 return ret;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003341 }
3342
3343 /* Re-check under ptl */
Hugh Dickinsb0b9b3d2017-01-07 15:37:31 -08003344 if (unlikely(!pte_none(*vmf->pte)))
3345 return VM_FAULT_NOPAGE;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003346
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003347 flush_icache_page(vma, page);
3348 entry = mk_pte(page, vma->vm_page_prot);
3349 if (write)
3350 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003351 /* copy-on-write page */
3352 if (write && !(vma->vm_flags & VM_SHARED)) {
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003353 inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES);
Jan Kara82b0f8c2016-12-14 15:06:58 -08003354 page_add_new_anon_rmap(page, vma, vmf->address, false);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003355 mem_cgroup_commit_charge(page, memcg, false, false);
3356 lru_cache_add_active_or_unevictable(page, vma);
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003357 } else {
Jerome Marchandeca56ff2016-01-14 15:19:26 -08003358 inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page));
Kirill A. Shutemovdd78fed2016-07-26 15:25:26 -07003359 page_add_file_rmap(page, false);
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003360 }
Jan Kara82b0f8c2016-12-14 15:06:58 -08003361 set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry);
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003362
3363 /* no need to invalidate: a not-present page won't be cached */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003364 update_mmu_cache(vma, vmf->address, vmf->pte);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003365
Hugh Dickinsb0b9b3d2017-01-07 15:37:31 -08003366 return 0;
Kirill A. Shutemov3bb97792014-04-03 14:48:16 -07003367}
3368
Jan Kara9118c0c2016-12-14 15:07:21 -08003369
3370/**
3371 * finish_fault - finish page fault once we have prepared the page to fault
3372 *
3373 * @vmf: structure describing the fault
3374 *
3375 * This function handles all that is needed to finish a page fault once the
3376 * page to fault in is prepared. It handles locking of PTEs, inserts PTE for
3377 * given page, adds reverse page mapping, handles memcg charges and LRU
3378 * addition. The function returns 0 on success, VM_FAULT_ code in case of
3379 * error.
3380 *
3381 * The function expects the page to be locked and on success it consumes a
3382 * reference of a page being mapped (for the PTE which maps it).
3383 */
3384int finish_fault(struct vm_fault *vmf)
3385{
3386 struct page *page;
Michal Hocko6b31d592017-08-18 15:16:15 -07003387 int ret = 0;
Jan Kara9118c0c2016-12-14 15:07:21 -08003388
3389 /* Did we COW the page? */
3390 if ((vmf->flags & FAULT_FLAG_WRITE) &&
3391 !(vmf->vma->vm_flags & VM_SHARED))
3392 page = vmf->cow_page;
3393 else
3394 page = vmf->page;
Michal Hocko6b31d592017-08-18 15:16:15 -07003395
3396 /*
3397 * check even for read faults because we might have lost our CoWed
3398 * page
3399 */
3400 if (!(vmf->vma->vm_flags & VM_SHARED))
3401 ret = check_stable_address_space(vmf->vma->vm_mm);
3402 if (!ret)
3403 ret = alloc_set_pte(vmf, vmf->memcg, page);
Jan Kara9118c0c2016-12-14 15:07:21 -08003404 if (vmf->pte)
3405 pte_unmap_unlock(vmf->pte, vmf->ptl);
3406 return ret;
3407}
3408
Kirill A. Shutemov3a910532014-08-06 16:08:07 -07003409static unsigned long fault_around_bytes __read_mostly =
3410 rounddown_pow_of_two(65536);
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07003411
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07003412#ifdef CONFIG_DEBUG_FS
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07003413static int fault_around_bytes_get(void *data, u64 *val)
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07003414{
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07003415 *val = fault_around_bytes;
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07003416 return 0;
3417}
3418
Andrey Ryabininb4903d62014-07-30 16:08:35 -07003419/*
3420 * fault_around_pages() and fault_around_mask() expects fault_around_bytes
3421 * rounded down to nearest page order. It's what do_fault_around() expects to
3422 * see.
3423 */
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07003424static int fault_around_bytes_set(void *data, u64 val)
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07003425{
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07003426 if (val / PAGE_SIZE > PTRS_PER_PTE)
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07003427 return -EINVAL;
Andrey Ryabininb4903d62014-07-30 16:08:35 -07003428 if (val > PAGE_SIZE)
3429 fault_around_bytes = rounddown_pow_of_two(val);
3430 else
3431 fault_around_bytes = PAGE_SIZE; /* rounddown_pow_of_two(0) is undefined */
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07003432 return 0;
3433}
Yevgen Pronenko0a1345f2017-07-10 15:47:17 -07003434DEFINE_DEBUGFS_ATTRIBUTE(fault_around_bytes_fops,
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07003435 fault_around_bytes_get, fault_around_bytes_set, "%llu\n");
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07003436
3437static int __init fault_around_debugfs(void)
3438{
3439 void *ret;
3440
Yevgen Pronenko0a1345f2017-07-10 15:47:17 -07003441 ret = debugfs_create_file_unsafe("fault_around_bytes", 0644, NULL, NULL,
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07003442 &fault_around_bytes_fops);
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07003443 if (!ret)
Kirill A. Shutemova9b0f862014-06-04 16:10:54 -07003444 pr_warn("Failed to create fault_around_bytes in debugfs");
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07003445 return 0;
3446}
3447late_initcall(fault_around_debugfs);
Kirill A. Shutemov1592eef2014-04-07 15:37:22 -07003448#endif
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003449
Kirill A. Shutemov1fdb4122014-06-04 16:10:55 -07003450/*
3451 * do_fault_around() tries to map few pages around the fault address. The hope
3452 * is that the pages will be needed soon and this will lower the number of
3453 * faults to handle.
3454 *
3455 * It uses vm_ops->map_pages() to map the pages, which skips the page if it's
3456 * not ready to be mapped: not up-to-date, locked, etc.
3457 *
3458 * This function is called with the page table lock taken. In the split ptlock
3459 * case the page table lock only protects only those entries which belong to
3460 * the page table corresponding to the fault address.
3461 *
3462 * This function doesn't cross the VMA boundaries, in order to call map_pages()
3463 * only once.
3464 *
3465 * fault_around_pages() defines how many pages we'll try to map.
3466 * do_fault_around() expects it to return a power of two less than or equal to
3467 * PTRS_PER_PTE.
3468 *
3469 * The virtual address of the area that we map is naturally aligned to the
3470 * fault_around_pages() value (and therefore to page order). This way it's
3471 * easier to guarantee that we don't cross page table boundaries.
3472 */
Jan Kara0721ec82016-12-14 15:07:04 -08003473static int do_fault_around(struct vm_fault *vmf)
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003474{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003475 unsigned long address = vmf->address, nr_pages, mask;
Jan Kara0721ec82016-12-14 15:07:04 -08003476 pgoff_t start_pgoff = vmf->pgoff;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003477 pgoff_t end_pgoff;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003478 int off, ret = 0;
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003479
Jason Low4db0c3c2015-04-15 16:14:08 -07003480 nr_pages = READ_ONCE(fault_around_bytes) >> PAGE_SHIFT;
Kirill A. Shutemovaecd6f42014-08-06 16:08:05 -07003481 mask = ~(nr_pages * PAGE_SIZE - 1) & PAGE_MASK;
3482
Jan Kara82b0f8c2016-12-14 15:06:58 -08003483 vmf->address = max(address & mask, vmf->vma->vm_start);
3484 off = ((address - vmf->address) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1);
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003485 start_pgoff -= off;
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003486
3487 /*
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003488 * end_pgoff is either end of page table or end of vma
3489 * or fault_around_pages() from start_pgoff, depending what is nearest.
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003490 */
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003491 end_pgoff = start_pgoff -
Jan Kara82b0f8c2016-12-14 15:06:58 -08003492 ((vmf->address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)) +
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003493 PTRS_PER_PTE - 1;
Jan Kara82b0f8c2016-12-14 15:06:58 -08003494 end_pgoff = min3(end_pgoff, vma_pages(vmf->vma) + vmf->vma->vm_pgoff - 1,
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003495 start_pgoff + nr_pages - 1);
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003496
Jan Kara82b0f8c2016-12-14 15:06:58 -08003497 if (pmd_none(*vmf->pmd)) {
3498 vmf->prealloc_pte = pte_alloc_one(vmf->vma->vm_mm,
3499 vmf->address);
3500 if (!vmf->prealloc_pte)
Vegard Nossumc5f88bd2016-08-02 14:02:22 -07003501 goto out;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003502 smp_wmb(); /* See comment in __pte_alloc() */
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003503 }
3504
Jan Kara82b0f8c2016-12-14 15:06:58 -08003505 vmf->vma->vm_ops->map_pages(vmf, start_pgoff, end_pgoff);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003506
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003507 /* Huge page is mapped? Page fault is solved */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003508 if (pmd_trans_huge(*vmf->pmd)) {
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003509 ret = VM_FAULT_NOPAGE;
3510 goto out;
3511 }
3512
3513 /* ->map_pages() haven't done anything useful. Cold page cache? */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003514 if (!vmf->pte)
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003515 goto out;
3516
3517 /* check if the page fault is solved */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003518 vmf->pte -= (vmf->address >> PAGE_SHIFT) - (address >> PAGE_SHIFT);
3519 if (!pte_none(*vmf->pte))
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003520 ret = VM_FAULT_NOPAGE;
Jan Kara82b0f8c2016-12-14 15:06:58 -08003521 pte_unmap_unlock(vmf->pte, vmf->ptl);
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003522out:
Jan Kara82b0f8c2016-12-14 15:06:58 -08003523 vmf->address = address;
3524 vmf->pte = NULL;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003525 return ret;
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003526}
3527
Jan Kara0721ec82016-12-14 15:07:04 -08003528static int do_read_fault(struct vm_fault *vmf)
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07003529{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003530 struct vm_area_struct *vma = vmf->vma;
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003531 int ret = 0;
3532
3533 /*
3534 * Let's call ->map_pages() first and use ->fault() as fallback
3535 * if page by the offset is not ready to be mapped (cold cache or
3536 * something).
3537 */
Kirill A. Shutemov9b4bdd22015-02-10 14:09:51 -08003538 if (vma->vm_ops->map_pages && fault_around_bytes >> PAGE_SHIFT > 1) {
Jan Kara0721ec82016-12-14 15:07:04 -08003539 ret = do_fault_around(vmf);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003540 if (ret)
3541 return ret;
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -07003542 }
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07003543
Jan Kara936ca802016-12-14 15:07:10 -08003544 ret = __do_fault(vmf);
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07003545 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
3546 return ret;
3547
Jan Kara9118c0c2016-12-14 15:07:21 -08003548 ret |= finish_fault(vmf);
Jan Kara936ca802016-12-14 15:07:10 -08003549 unlock_page(vmf->page);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003550 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
Jan Kara936ca802016-12-14 15:07:10 -08003551 put_page(vmf->page);
Kirill A. Shutemove655fb22014-04-03 14:48:11 -07003552 return ret;
3553}
3554
Jan Kara0721ec82016-12-14 15:07:04 -08003555static int do_cow_fault(struct vm_fault *vmf)
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003556{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003557 struct vm_area_struct *vma = vmf->vma;
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003558 int ret;
3559
3560 if (unlikely(anon_vma_prepare(vma)))
3561 return VM_FAULT_OOM;
3562
Jan Kara936ca802016-12-14 15:07:10 -08003563 vmf->cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vmf->address);
3564 if (!vmf->cow_page)
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003565 return VM_FAULT_OOM;
3566
Jan Kara936ca802016-12-14 15:07:10 -08003567 if (mem_cgroup_try_charge(vmf->cow_page, vma->vm_mm, GFP_KERNEL,
Jan Kara39170482016-12-14 15:07:18 -08003568 &vmf->memcg, false)) {
Jan Kara936ca802016-12-14 15:07:10 -08003569 put_page(vmf->cow_page);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003570 return VM_FAULT_OOM;
3571 }
3572
Jan Kara936ca802016-12-14 15:07:10 -08003573 ret = __do_fault(vmf);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003574 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
3575 goto uncharge_out;
Jan Kara39170482016-12-14 15:07:18 -08003576 if (ret & VM_FAULT_DONE_COW)
3577 return ret;
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003578
Jan Karab1aa8122016-12-14 15:07:24 -08003579 copy_user_highpage(vmf->cow_page, vmf->page, vmf->address, vma);
Jan Kara936ca802016-12-14 15:07:10 -08003580 __SetPageUptodate(vmf->cow_page);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003581
Jan Kara9118c0c2016-12-14 15:07:21 -08003582 ret |= finish_fault(vmf);
Jan Karab1aa8122016-12-14 15:07:24 -08003583 unlock_page(vmf->page);
3584 put_page(vmf->page);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003585 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
3586 goto uncharge_out;
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003587 return ret;
3588uncharge_out:
Jan Kara39170482016-12-14 15:07:18 -08003589 mem_cgroup_cancel_charge(vmf->cow_page, vmf->memcg, false);
Jan Kara936ca802016-12-14 15:07:10 -08003590 put_page(vmf->cow_page);
Kirill A. Shutemovec47c3b2014-04-03 14:48:12 -07003591 return ret;
3592}
3593
Jan Kara0721ec82016-12-14 15:07:04 -08003594static int do_shared_fault(struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003595{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003596 struct vm_area_struct *vma = vmf->vma;
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003597 int ret, tmp;
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07003598
Jan Kara936ca802016-12-14 15:07:10 -08003599 ret = __do_fault(vmf);
Kirill A. Shutemov7eae74a2014-04-03 14:48:10 -07003600 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003601 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003602
3603 /*
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003604 * Check if the backing address space wants to know that the page is
3605 * about to become writable
Linus Torvalds1da177e2005-04-16 15:20:36 -07003606 */
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07003607 if (vma->vm_ops->page_mkwrite) {
Jan Kara936ca802016-12-14 15:07:10 -08003608 unlock_page(vmf->page);
Jan Kara38b8cb72016-12-14 15:07:30 -08003609 tmp = do_page_mkwrite(vmf);
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07003610 if (unlikely(!tmp ||
3611 (tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
Jan Kara936ca802016-12-14 15:07:10 -08003612 put_page(vmf->page);
Kirill A. Shutemovfb09a462014-04-03 14:48:15 -07003613 return tmp;
3614 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003615 }
3616
Jan Kara9118c0c2016-12-14 15:07:21 -08003617 ret |= finish_fault(vmf);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003618 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE |
3619 VM_FAULT_RETRY))) {
Jan Kara936ca802016-12-14 15:07:10 -08003620 unlock_page(vmf->page);
3621 put_page(vmf->page);
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003622 return ret;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07003623 }
Kirill A. Shutemovf0c6d4d2014-04-03 14:48:13 -07003624
Jan Kara97ba0c22016-12-14 15:07:27 -08003625 fault_dirty_shared_page(vma, vmf->page);
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07003626 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07003627}
Nick Piggind00806b2007-07-19 01:46:57 -07003628
Paul Cassella9a95f3c2014-08-06 16:07:24 -07003629/*
3630 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3631 * but allow concurrent faults).
3632 * The mmap_sem may have been released depending on flags and our
3633 * return value. See filemap_fault() and __lock_page_or_retry().
3634 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003635static int do_fault(struct vm_fault *vmf)
Nick Piggin54cb8822007-07-19 01:46:59 -07003636{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003637 struct vm_area_struct *vma = vmf->vma;
Hugh Dickinsb0b9b3d2017-01-07 15:37:31 -08003638 int ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07003639
Kirill A. Shutemov6b7339f2015-07-06 23:18:37 +03003640 /* The VMA was not fully populated on mmap() or missing VM_DONTEXPAND */
3641 if (!vma->vm_ops->fault)
Hugh Dickinsb0b9b3d2017-01-07 15:37:31 -08003642 ret = VM_FAULT_SIGBUS;
3643 else if (!(vmf->flags & FAULT_FLAG_WRITE))
3644 ret = do_read_fault(vmf);
3645 else if (!(vma->vm_flags & VM_SHARED))
3646 ret = do_cow_fault(vmf);
3647 else
3648 ret = do_shared_fault(vmf);
3649
3650 /* preallocated pagetable is unused: free it */
3651 if (vmf->prealloc_pte) {
3652 pte_free(vma->vm_mm, vmf->prealloc_pte);
Tobin C Harding7f2b6ce2017-02-24 14:58:59 -08003653 vmf->prealloc_pte = NULL;
Hugh Dickinsb0b9b3d2017-01-07 15:37:31 -08003654 }
3655 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07003656}
3657
Rashika Kheriab19a9932014-04-03 14:48:02 -07003658static int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
Rik van Riel04bb2f92013-10-07 11:29:36 +01003659 unsigned long addr, int page_nid,
3660 int *flags)
Mel Gorman9532fec2012-11-15 01:24:32 +00003661{
3662 get_page(page);
3663
3664 count_vm_numa_event(NUMA_HINT_FAULTS);
Rik van Riel04bb2f92013-10-07 11:29:36 +01003665 if (page_nid == numa_node_id()) {
Mel Gorman9532fec2012-11-15 01:24:32 +00003666 count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL);
Rik van Riel04bb2f92013-10-07 11:29:36 +01003667 *flags |= TNF_FAULT_LOCAL;
3668 }
Mel Gorman9532fec2012-11-15 01:24:32 +00003669
3670 return mpol_misplaced(page, vma, addr);
3671}
3672
Jan Kara29943022016-12-14 15:07:16 -08003673static int do_numa_page(struct vm_fault *vmf)
Mel Gormand10e63f2012-10-25 14:16:31 +02003674{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003675 struct vm_area_struct *vma = vmf->vma;
Mel Gorman4daae3b2012-11-02 11:33:45 +00003676 struct page *page = NULL;
Mel Gorman8191acb2013-10-07 11:28:45 +01003677 int page_nid = -1;
Peter Zijlstra90572892013-10-07 11:29:20 +01003678 int last_cpupid;
Peter Zijlstracbee9f82012-10-25 14:16:43 +02003679 int target_nid;
Mel Gormanb8593bf2012-11-21 01:18:23 +00003680 bool migrated = false;
Aneesh Kumar K.Vcee216a2017-02-24 14:59:13 -08003681 pte_t pte;
Aneesh Kumar K.V288bc542017-02-24 14:59:16 -08003682 bool was_writable = pte_savedwrite(vmf->orig_pte);
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003683 int flags = 0;
Mel Gormand10e63f2012-10-25 14:16:31 +02003684
3685 /*
Tobin C Harding166f61b2017-02-24 14:59:01 -08003686 * The "pte" at this point cannot be used safely without
3687 * validation through pte_unmap_same(). It's of NUMA type but
3688 * the pfn may be screwed if the read is non atomic.
Tobin C Harding166f61b2017-02-24 14:59:01 -08003689 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003690 vmf->ptl = pte_lockptr(vma->vm_mm, vmf->pmd);
3691 spin_lock(vmf->ptl);
Aneesh Kumar K.Vcee216a2017-02-24 14:59:13 -08003692 if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) {
Jan Kara82b0f8c2016-12-14 15:06:58 -08003693 pte_unmap_unlock(vmf->pte, vmf->ptl);
Mel Gorman4daae3b2012-11-02 11:33:45 +00003694 goto out;
3695 }
3696
Aneesh Kumar K.Vcee216a2017-02-24 14:59:13 -08003697 /*
3698 * Make it present again, Depending on how arch implementes non
3699 * accessible ptes, some can allow access by kernel mode.
3700 */
3701 pte = ptep_modify_prot_start(vma->vm_mm, vmf->address, vmf->pte);
Mel Gorman4d942462015-02-12 14:58:28 -08003702 pte = pte_modify(pte, vma->vm_page_prot);
3703 pte = pte_mkyoung(pte);
Mel Gormanb191f9b2015-03-25 15:55:40 -07003704 if (was_writable)
3705 pte = pte_mkwrite(pte);
Aneesh Kumar K.Vcee216a2017-02-24 14:59:13 -08003706 ptep_modify_prot_commit(vma->vm_mm, vmf->address, vmf->pte, pte);
Jan Kara82b0f8c2016-12-14 15:06:58 -08003707 update_mmu_cache(vma, vmf->address, vmf->pte);
Mel Gormand10e63f2012-10-25 14:16:31 +02003708
Jan Kara82b0f8c2016-12-14 15:06:58 -08003709 page = vm_normal_page(vma, vmf->address, pte);
Mel Gormand10e63f2012-10-25 14:16:31 +02003710 if (!page) {
Jan Kara82b0f8c2016-12-14 15:06:58 -08003711 pte_unmap_unlock(vmf->pte, vmf->ptl);
Mel Gormand10e63f2012-10-25 14:16:31 +02003712 return 0;
3713 }
3714
Kirill A. Shutemove81c4802016-01-15 16:53:49 -08003715 /* TODO: handle PTE-mapped THP */
3716 if (PageCompound(page)) {
Jan Kara82b0f8c2016-12-14 15:06:58 -08003717 pte_unmap_unlock(vmf->pte, vmf->ptl);
Kirill A. Shutemove81c4802016-01-15 16:53:49 -08003718 return 0;
3719 }
3720
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003721 /*
Mel Gormanbea66fb2015-03-25 15:55:37 -07003722 * Avoid grouping on RO pages in general. RO pages shouldn't hurt as
3723 * much anyway since they can be in shared cache state. This misses
3724 * the case where a mapping is writable but the process never writes
3725 * to it but pte_write gets cleared during protection updates and
3726 * pte_dirty has unpredictable behaviour between PTE scan updates,
3727 * background writeback, dirty balancing and application behaviour.
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003728 */
Rik van Rield59dc7b2016-09-08 21:30:53 -04003729 if (!pte_write(pte))
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003730 flags |= TNF_NO_GROUP;
3731
Rik van Rieldabe1d92013-10-07 11:29:34 +01003732 /*
3733 * Flag if the page is shared between multiple address spaces. This
3734 * is later used when determining whether to group tasks together
3735 */
3736 if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED))
3737 flags |= TNF_SHARED;
3738
Peter Zijlstra90572892013-10-07 11:29:20 +01003739 last_cpupid = page_cpupid_last(page);
Mel Gorman8191acb2013-10-07 11:28:45 +01003740 page_nid = page_to_nid(page);
Jan Kara82b0f8c2016-12-14 15:06:58 -08003741 target_nid = numa_migrate_prep(page, vma, vmf->address, page_nid,
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003742 &flags);
Jan Kara82b0f8c2016-12-14 15:06:58 -08003743 pte_unmap_unlock(vmf->pte, vmf->ptl);
Mel Gorman4daae3b2012-11-02 11:33:45 +00003744 if (target_nid == -1) {
Mel Gorman4daae3b2012-11-02 11:33:45 +00003745 put_page(page);
3746 goto out;
3747 }
3748
3749 /* Migrate to the requested node */
Mel Gorman1bc115d2013-10-07 11:29:05 +01003750 migrated = migrate_misplaced_page(page, vma, target_nid);
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003751 if (migrated) {
Mel Gorman8191acb2013-10-07 11:28:45 +01003752 page_nid = target_nid;
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003753 flags |= TNF_MIGRATED;
Mel Gorman074c2382015-03-25 15:55:42 -07003754 } else
3755 flags |= TNF_MIGRATE_FAIL;
Mel Gorman4daae3b2012-11-02 11:33:45 +00003756
3757out:
Mel Gorman8191acb2013-10-07 11:28:45 +01003758 if (page_nid != -1)
Peter Zijlstra6688cc02013-10-07 11:29:24 +01003759 task_numa_fault(last_cpupid, page_nid, 1, flags);
Mel Gormand10e63f2012-10-25 14:16:31 +02003760 return 0;
3761}
3762
Geert Uytterhoeven91a90142017-07-12 14:33:08 -07003763static inline int create_huge_pmd(struct vm_fault *vmf)
Matthew Wilcoxb96375f2015-09-08 14:58:48 -07003764{
Dave Jiangf4200392017-02-22 15:40:06 -08003765 if (vma_is_anonymous(vmf->vma))
Jan Kara82b0f8c2016-12-14 15:06:58 -08003766 return do_huge_pmd_anonymous_page(vmf);
Dave Jianga2d58162017-02-24 14:56:59 -08003767 if (vmf->vma->vm_ops->huge_fault)
Dave Jiangc791ace2017-02-24 14:57:08 -08003768 return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD);
Matthew Wilcoxb96375f2015-09-08 14:58:48 -07003769 return VM_FAULT_FALLBACK;
3770}
3771
Jan Kara82b0f8c2016-12-14 15:06:58 -08003772static int wp_huge_pmd(struct vm_fault *vmf, pmd_t orig_pmd)
Matthew Wilcoxb96375f2015-09-08 14:58:48 -07003773{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003774 if (vma_is_anonymous(vmf->vma))
3775 return do_huge_pmd_wp_page(vmf, orig_pmd);
Dave Jianga2d58162017-02-24 14:56:59 -08003776 if (vmf->vma->vm_ops->huge_fault)
Dave Jiangc791ace2017-02-24 14:57:08 -08003777 return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD);
Kirill A. Shutemovaf9e4d52016-07-26 15:25:40 -07003778
3779 /* COW handled on pte level: split pmd */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003780 VM_BUG_ON_VMA(vmf->vma->vm_flags & VM_SHARED, vmf->vma);
3781 __split_huge_pmd(vmf->vma, vmf->pmd, vmf->address, false, NULL);
Kirill A. Shutemovaf9e4d52016-07-26 15:25:40 -07003782
Matthew Wilcoxb96375f2015-09-08 14:58:48 -07003783 return VM_FAULT_FALLBACK;
3784}
3785
Lorenzo Stoakes38e08852016-09-11 23:54:25 +01003786static inline bool vma_is_accessible(struct vm_area_struct *vma)
3787{
3788 return vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE);
3789}
3790
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08003791static int create_huge_pud(struct vm_fault *vmf)
3792{
3793#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3794 /* No support for anonymous transparent PUD pages yet */
3795 if (vma_is_anonymous(vmf->vma))
3796 return VM_FAULT_FALLBACK;
3797 if (vmf->vma->vm_ops->huge_fault)
Dave Jiangc791ace2017-02-24 14:57:08 -08003798 return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD);
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08003799#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
3800 return VM_FAULT_FALLBACK;
3801}
3802
3803static int wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud)
3804{
3805#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3806 /* No support for anonymous transparent PUD pages yet */
3807 if (vma_is_anonymous(vmf->vma))
3808 return VM_FAULT_FALLBACK;
3809 if (vmf->vma->vm_ops->huge_fault)
Dave Jiangc791ace2017-02-24 14:57:08 -08003810 return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PUD);
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08003811#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
3812 return VM_FAULT_FALLBACK;
3813}
3814
Linus Torvalds1da177e2005-04-16 15:20:36 -07003815/*
3816 * These routines also need to handle stuff like marking pages dirty
3817 * and/or accessed for architectures that don't do it in hardware (most
3818 * RISC architectures). The early dirtying is also good on the i386.
3819 *
3820 * There is also a hook called "update_mmu_cache()" that architectures
3821 * with external mmu caches can use to update those (ie the Sparc or
3822 * PowerPC hashed page tables that act as extended TLBs).
3823 *
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003824 * We enter with non-exclusive mmap_sem (to exclude vma changes, but allow
3825 * concurrent faults).
Paul Cassella9a95f3c2014-08-06 16:07:24 -07003826 *
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003827 * The mmap_sem may have been released depending on flags and our return value.
3828 * See filemap_fault() and __lock_page_or_retry().
Linus Torvalds1da177e2005-04-16 15:20:36 -07003829 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003830static int handle_pte_fault(struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003831{
3832 pte_t entry;
3833
Jan Kara82b0f8c2016-12-14 15:06:58 -08003834 if (unlikely(pmd_none(*vmf->pmd))) {
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003835 /*
3836 * Leave __pte_alloc() until later: because vm_ops->fault may
3837 * want to allocate huge page, and if we expose page table
3838 * for an instant, it will be difficult to retract from
3839 * concurrent faults and from rmap lookups.
3840 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003841 vmf->pte = NULL;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003842 } else {
3843 /* See comment in pte_alloc_one_map() */
Ross Zwislerd0f0931d2017-06-02 14:46:34 -07003844 if (pmd_devmap_trans_unstable(vmf->pmd))
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003845 return 0;
3846 /*
3847 * A regular pmd is established and it can't morph into a huge
3848 * pmd from under us anymore at this point because we hold the
3849 * mmap_sem read mode and khugepaged takes it in write mode.
3850 * So now it's safe to run pte_offset_map().
3851 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003852 vmf->pte = pte_offset_map(vmf->pmd, vmf->address);
Jan Kara29943022016-12-14 15:07:16 -08003853 vmf->orig_pte = *vmf->pte;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003854
3855 /*
3856 * some architectures can have larger ptes than wordsize,
3857 * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and
3858 * CONFIG_32BIT=y, so READ_ONCE or ACCESS_ONCE cannot guarantee
3859 * atomic accesses. The code below just needs a consistent
3860 * view for the ifs and we later double check anyway with the
3861 * ptl lock held. So here a barrier will do.
3862 */
3863 barrier();
Jan Kara29943022016-12-14 15:07:16 -08003864 if (pte_none(vmf->orig_pte)) {
Jan Kara82b0f8c2016-12-14 15:06:58 -08003865 pte_unmap(vmf->pte);
3866 vmf->pte = NULL;
Hugh Dickins65500d22005-10-29 18:15:59 -07003867 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003868 }
3869
Jan Kara82b0f8c2016-12-14 15:06:58 -08003870 if (!vmf->pte) {
3871 if (vma_is_anonymous(vmf->vma))
3872 return do_anonymous_page(vmf);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003873 else
Jan Kara82b0f8c2016-12-14 15:06:58 -08003874 return do_fault(vmf);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003875 }
3876
Jan Kara29943022016-12-14 15:07:16 -08003877 if (!pte_present(vmf->orig_pte))
3878 return do_swap_page(vmf);
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003879
Jan Kara29943022016-12-14 15:07:16 -08003880 if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma))
3881 return do_numa_page(vmf);
Mel Gormand10e63f2012-10-25 14:16:31 +02003882
Jan Kara82b0f8c2016-12-14 15:06:58 -08003883 vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
3884 spin_lock(vmf->ptl);
Jan Kara29943022016-12-14 15:07:16 -08003885 entry = vmf->orig_pte;
Jan Kara82b0f8c2016-12-14 15:06:58 -08003886 if (unlikely(!pte_same(*vmf->pte, entry)))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003887 goto unlock;
Jan Kara82b0f8c2016-12-14 15:06:58 -08003888 if (vmf->flags & FAULT_FLAG_WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003889 if (!pte_write(entry))
Jan Kara29943022016-12-14 15:07:16 -08003890 return do_wp_page(vmf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003891 entry = pte_mkdirty(entry);
3892 }
3893 entry = pte_mkyoung(entry);
Jan Kara82b0f8c2016-12-14 15:06:58 -08003894 if (ptep_set_access_flags(vmf->vma, vmf->address, vmf->pte, entry,
3895 vmf->flags & FAULT_FLAG_WRITE)) {
3896 update_mmu_cache(vmf->vma, vmf->address, vmf->pte);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07003897 } else {
3898 /*
3899 * This is needed only for protection faults but the arch code
3900 * is not yet telling us if this is a protection fault or not.
3901 * This still avoids useless tlb flushes for .text page faults
3902 * with threads.
3903 */
Jan Kara82b0f8c2016-12-14 15:06:58 -08003904 if (vmf->flags & FAULT_FLAG_WRITE)
3905 flush_tlb_fix_spurious_fault(vmf->vma, vmf->address);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07003906 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003907unlock:
Jan Kara82b0f8c2016-12-14 15:06:58 -08003908 pte_unmap_unlock(vmf->pte, vmf->ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07003909 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003910}
3911
3912/*
3913 * By the time we get here, we already hold the mm semaphore
Paul Cassella9a95f3c2014-08-06 16:07:24 -07003914 *
3915 * The mmap_sem may have been released depending on flags and our
3916 * return value. See filemap_fault() and __lock_page_or_retry().
Linus Torvalds1da177e2005-04-16 15:20:36 -07003917 */
Kirill A. Shutemovdcddffd2016-07-26 15:25:18 -07003918static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
3919 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003920{
Jan Kara82b0f8c2016-12-14 15:06:58 -08003921 struct vm_fault vmf = {
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003922 .vma = vma,
Jan Kara1a29d852016-12-14 15:07:01 -08003923 .address = address & PAGE_MASK,
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003924 .flags = flags,
Jan Kara0721ec82016-12-14 15:07:04 -08003925 .pgoff = linear_page_index(vma, address),
Jan Kara667240e2016-12-14 15:07:07 -08003926 .gfp_mask = __get_fault_gfp_mask(vma),
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003927 };
Kirill A. Shutemovdcddffd2016-07-26 15:25:18 -07003928 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003929 pgd_t *pgd;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03003930 p4d_t *p4d;
Dave Jianga2d58162017-02-24 14:56:59 -08003931 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003932
Linus Torvalds1da177e2005-04-16 15:20:36 -07003933 pgd = pgd_offset(mm, address);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03003934 p4d = p4d_alloc(mm, pgd, address);
3935 if (!p4d)
3936 return VM_FAULT_OOM;
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08003937
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03003938 vmf.pud = pud_alloc(mm, p4d, address);
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08003939 if (!vmf.pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07003940 return VM_FAULT_OOM;
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08003941 if (pud_none(*vmf.pud) && transparent_hugepage_enabled(vma)) {
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08003942 ret = create_huge_pud(&vmf);
3943 if (!(ret & VM_FAULT_FALLBACK))
3944 return ret;
3945 } else {
3946 pud_t orig_pud = *vmf.pud;
3947
3948 barrier();
3949 if (pud_trans_huge(orig_pud) || pud_devmap(orig_pud)) {
3950 unsigned int dirty = flags & FAULT_FLAG_WRITE;
3951
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08003952 /* NUMA case for anonymous PUDs would go here */
3953
3954 if (dirty && !pud_write(orig_pud)) {
3955 ret = wp_huge_pud(&vmf, orig_pud);
3956 if (!(ret & VM_FAULT_FALLBACK))
3957 return ret;
3958 } else {
3959 huge_pud_set_accessed(&vmf, orig_pud);
3960 return 0;
3961 }
3962 }
3963 }
3964
3965 vmf.pmd = pmd_alloc(mm, vmf.pud, address);
Jan Kara82b0f8c2016-12-14 15:06:58 -08003966 if (!vmf.pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07003967 return VM_FAULT_OOM;
Jan Kara82b0f8c2016-12-14 15:06:58 -08003968 if (pmd_none(*vmf.pmd) && transparent_hugepage_enabled(vma)) {
Dave Jianga2d58162017-02-24 14:56:59 -08003969 ret = create_huge_pmd(&vmf);
Kirill A. Shutemovc0292552013-09-12 15:14:05 -07003970 if (!(ret & VM_FAULT_FALLBACK))
3971 return ret;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003972 } else {
Jan Kara82b0f8c2016-12-14 15:06:58 -08003973 pmd_t orig_pmd = *vmf.pmd;
David Rientjes1f1d06c2012-05-29 15:06:23 -07003974
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003975 barrier();
Zi Yan84c3fc42017-09-08 16:11:01 -07003976 if (unlikely(is_swap_pmd(orig_pmd))) {
3977 VM_BUG_ON(thp_migration_supported() &&
3978 !is_pmd_migration_entry(orig_pmd));
3979 if (is_pmd_migration_entry(orig_pmd))
3980 pmd_migration_entry_wait(mm, vmf.pmd);
3981 return 0;
3982 }
Dan Williams5c7fb562016-01-15 16:56:52 -08003983 if (pmd_trans_huge(orig_pmd) || pmd_devmap(orig_pmd)) {
Lorenzo Stoakes38e08852016-09-11 23:54:25 +01003984 if (pmd_protnone(orig_pmd) && vma_is_accessible(vma))
Jan Kara82b0f8c2016-12-14 15:06:58 -08003985 return do_huge_pmd_numa_page(&vmf, orig_pmd);
Mel Gormand10e63f2012-10-25 14:16:31 +02003986
Jan Kara82b0f8c2016-12-14 15:06:58 -08003987 if ((vmf.flags & FAULT_FLAG_WRITE) &&
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003988 !pmd_write(orig_pmd)) {
Jan Kara82b0f8c2016-12-14 15:06:58 -08003989 ret = wp_huge_pmd(&vmf, orig_pmd);
Kirill A. Shutemov9845cbb2014-02-25 15:01:42 -08003990 if (!(ret & VM_FAULT_FALLBACK))
3991 return ret;
Will Deacona1dd4502012-12-11 16:01:27 -08003992 } else {
Jan Kara82b0f8c2016-12-14 15:06:58 -08003993 huge_pmd_set_accessed(&vmf, orig_pmd);
Kirill A. Shutemov9845cbb2014-02-25 15:01:42 -08003994 return 0;
David Rientjes1f1d06c2012-05-29 15:06:23 -07003995 }
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003996 }
3997 }
3998
Jan Kara82b0f8c2016-12-14 15:06:58 -08003999 return handle_pte_fault(&vmf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004000}
4001
Paul Cassella9a95f3c2014-08-06 16:07:24 -07004002/*
4003 * By the time we get here, we already hold the mm semaphore
4004 *
4005 * The mmap_sem may have been released depending on flags and our
4006 * return value. See filemap_fault() and __lock_page_or_retry().
4007 */
Kirill A. Shutemovdcddffd2016-07-26 15:25:18 -07004008int handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
4009 unsigned int flags)
Johannes Weiner519e5242013-09-12 15:13:42 -07004010{
4011 int ret;
4012
4013 __set_current_state(TASK_RUNNING);
4014
4015 count_vm_event(PGFAULT);
Roman Gushchin22621852017-07-06 15:40:25 -07004016 count_memcg_event_mm(vma->vm_mm, PGFAULT);
Johannes Weiner519e5242013-09-12 15:13:42 -07004017
4018 /* do counter updates before entering really critical section. */
4019 check_sync_rss_stat(current);
4020
4021 /*
4022 * Enable the memcg OOM handling for faults triggered in user
4023 * space. Kernel faults are handled more gracefully.
4024 */
4025 if (flags & FAULT_FLAG_USER)
Johannes Weiner49426422013-10-16 13:46:59 -07004026 mem_cgroup_oom_enable();
Johannes Weiner519e5242013-09-12 15:13:42 -07004027
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07004028 if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE,
4029 flags & FAULT_FLAG_INSTRUCTION,
4030 flags & FAULT_FLAG_REMOTE))
4031 return VM_FAULT_SIGSEGV;
4032
4033 if (unlikely(is_vm_hugetlb_page(vma)))
4034 ret = hugetlb_fault(vma->vm_mm, vma, address, flags);
4035 else
4036 ret = __handle_mm_fault(vma, address, flags);
Johannes Weiner519e5242013-09-12 15:13:42 -07004037
Johannes Weiner49426422013-10-16 13:46:59 -07004038 if (flags & FAULT_FLAG_USER) {
4039 mem_cgroup_oom_disable();
Tobin C Harding166f61b2017-02-24 14:59:01 -08004040 /*
4041 * The task may have entered a memcg OOM situation but
4042 * if the allocation error was handled gracefully (no
4043 * VM_FAULT_OOM), there is no need to kill anything.
4044 * Just clean up the OOM state peacefully.
4045 */
4046 if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM))
4047 mem_cgroup_oom_synchronize(false);
Johannes Weiner49426422013-10-16 13:46:59 -07004048 }
Johannes Weiner3812c8c2013-09-12 15:13:44 -07004049
Johannes Weiner519e5242013-09-12 15:13:42 -07004050 return ret;
4051}
Jesse Barnese1d6d012014-12-12 16:55:27 -08004052EXPORT_SYMBOL_GPL(handle_mm_fault);
Johannes Weiner519e5242013-09-12 15:13:42 -07004053
Kirill A. Shutemov90eceff2017-03-09 17:24:08 +03004054#ifndef __PAGETABLE_P4D_FOLDED
4055/*
4056 * Allocate p4d page table.
4057 * We've already handled the fast-path in-line.
4058 */
4059int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
4060{
4061 p4d_t *new = p4d_alloc_one(mm, address);
4062 if (!new)
4063 return -ENOMEM;
4064
4065 smp_wmb(); /* See comment in __pte_alloc */
4066
4067 spin_lock(&mm->page_table_lock);
4068 if (pgd_present(*pgd)) /* Another has populated it */
4069 p4d_free(mm, new);
4070 else
4071 pgd_populate(mm, pgd, new);
4072 spin_unlock(&mm->page_table_lock);
4073 return 0;
4074}
4075#endif /* __PAGETABLE_P4D_FOLDED */
4076
Linus Torvalds1da177e2005-04-16 15:20:36 -07004077#ifndef __PAGETABLE_PUD_FOLDED
4078/*
4079 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07004080 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004081 */
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03004082int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004083{
Hugh Dickinsc74df322005-10-29 18:16:23 -07004084 pud_t *new = pud_alloc_one(mm, address);
4085 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07004086 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004087
Nick Piggin362a61a2008-05-14 06:37:36 +02004088 smp_wmb(); /* See comment in __pte_alloc */
4089
Hugh Dickins872fec12005-10-29 18:16:21 -07004090 spin_lock(&mm->page_table_lock);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03004091#ifndef __ARCH_HAS_5LEVEL_HACK
4092 if (p4d_present(*p4d)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08004093 pud_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07004094 else
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03004095 p4d_populate(mm, p4d, new);
4096#else
4097 if (pgd_present(*p4d)) /* Another has populated it */
4098 pud_free(mm, new);
4099 else
4100 pgd_populate(mm, p4d, new);
4101#endif /* __ARCH_HAS_5LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07004102 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07004103 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004104}
4105#endif /* __PAGETABLE_PUD_FOLDED */
4106
4107#ifndef __PAGETABLE_PMD_FOLDED
4108/*
4109 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07004110 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004111 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07004112int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004113{
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08004114 spinlock_t *ptl;
Hugh Dickinsc74df322005-10-29 18:16:23 -07004115 pmd_t *new = pmd_alloc_one(mm, address);
4116 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07004117 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004118
Nick Piggin362a61a2008-05-14 06:37:36 +02004119 smp_wmb(); /* See comment in __pte_alloc */
4120
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08004121 ptl = pud_lock(mm, pud);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004122#ifndef __ARCH_HAS_4LEVEL_HACK
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -08004123 if (!pud_present(*pud)) {
4124 mm_inc_nr_pmds(mm);
Hugh Dickins1bb36302005-10-29 18:16:22 -07004125 pud_populate(mm, pud, new);
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -08004126 } else /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08004127 pmd_free(mm, new);
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -08004128#else
4129 if (!pgd_present(*pud)) {
4130 mm_inc_nr_pmds(mm);
Hugh Dickins1bb36302005-10-29 18:16:22 -07004131 pgd_populate(mm, pud, new);
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -08004132 } else /* Another has populated it */
4133 pmd_free(mm, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004134#endif /* __ARCH_HAS_4LEVEL_HACK */
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -08004135 spin_unlock(ptl);
Hugh Dickins1bb36302005-10-29 18:16:22 -07004136 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004137}
4138#endif /* __PAGETABLE_PMD_FOLDED */
4139
Ross Zwisler09796392017-01-10 16:57:21 -08004140static int __follow_pte_pmd(struct mm_struct *mm, unsigned long address,
Jérôme Glissea4d1a882017-08-31 17:17:26 -04004141 unsigned long *start, unsigned long *end,
4142 pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp)
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07004143{
4144 pgd_t *pgd;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03004145 p4d_t *p4d;
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07004146 pud_t *pud;
4147 pmd_t *pmd;
4148 pte_t *ptep;
4149
4150 pgd = pgd_offset(mm, address);
4151 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
4152 goto out;
4153
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +03004154 p4d = p4d_offset(pgd, address);
4155 if (p4d_none(*p4d) || unlikely(p4d_bad(*p4d)))
4156 goto out;
4157
4158 pud = pud_offset(p4d, address);
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07004159 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
4160 goto out;
4161
4162 pmd = pmd_offset(pud, address);
Andrea Arcangelif66055ab2011-01-13 15:46:54 -08004163 VM_BUG_ON(pmd_trans_huge(*pmd));
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07004164
Ross Zwisler09796392017-01-10 16:57:21 -08004165 if (pmd_huge(*pmd)) {
4166 if (!pmdpp)
4167 goto out;
4168
Jérôme Glissea4d1a882017-08-31 17:17:26 -04004169 if (start && end) {
4170 *start = address & PMD_MASK;
4171 *end = *start + PMD_SIZE;
4172 mmu_notifier_invalidate_range_start(mm, *start, *end);
4173 }
Ross Zwisler09796392017-01-10 16:57:21 -08004174 *ptlp = pmd_lock(mm, pmd);
4175 if (pmd_huge(*pmd)) {
4176 *pmdpp = pmd;
4177 return 0;
4178 }
4179 spin_unlock(*ptlp);
Jérôme Glissea4d1a882017-08-31 17:17:26 -04004180 if (start && end)
4181 mmu_notifier_invalidate_range_end(mm, *start, *end);
Ross Zwisler09796392017-01-10 16:57:21 -08004182 }
4183
4184 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07004185 goto out;
4186
Jérôme Glissea4d1a882017-08-31 17:17:26 -04004187 if (start && end) {
4188 *start = address & PAGE_MASK;
4189 *end = *start + PAGE_SIZE;
4190 mmu_notifier_invalidate_range_start(mm, *start, *end);
4191 }
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07004192 ptep = pte_offset_map_lock(mm, pmd, address, ptlp);
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07004193 if (!pte_present(*ptep))
4194 goto unlock;
4195 *ptepp = ptep;
4196 return 0;
4197unlock:
4198 pte_unmap_unlock(ptep, *ptlp);
Jérôme Glissea4d1a882017-08-31 17:17:26 -04004199 if (start && end)
4200 mmu_notifier_invalidate_range_end(mm, *start, *end);
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07004201out:
4202 return -EINVAL;
4203}
4204
Ross Zwislerf729c8c2017-01-10 16:57:24 -08004205static inline int follow_pte(struct mm_struct *mm, unsigned long address,
4206 pte_t **ptepp, spinlock_t **ptlp)
Namhyung Kim1b36ba82010-10-26 14:22:00 -07004207{
4208 int res;
4209
4210 /* (void) is needed to make gcc happy */
4211 (void) __cond_lock(*ptlp,
Jérôme Glissea4d1a882017-08-31 17:17:26 -04004212 !(res = __follow_pte_pmd(mm, address, NULL, NULL,
4213 ptepp, NULL, ptlp)));
Namhyung Kim1b36ba82010-10-26 14:22:00 -07004214 return res;
4215}
4216
Ross Zwisler09796392017-01-10 16:57:21 -08004217int follow_pte_pmd(struct mm_struct *mm, unsigned long address,
Jérôme Glissea4d1a882017-08-31 17:17:26 -04004218 unsigned long *start, unsigned long *end,
Ross Zwisler09796392017-01-10 16:57:21 -08004219 pte_t **ptepp, pmd_t **pmdpp, spinlock_t **ptlp)
4220{
4221 int res;
4222
4223 /* (void) is needed to make gcc happy */
4224 (void) __cond_lock(*ptlp,
Jérôme Glissea4d1a882017-08-31 17:17:26 -04004225 !(res = __follow_pte_pmd(mm, address, start, end,
4226 ptepp, pmdpp, ptlp)));
Ross Zwisler09796392017-01-10 16:57:21 -08004227 return res;
4228}
4229EXPORT_SYMBOL(follow_pte_pmd);
4230
Johannes Weiner3b6748e2009-06-16 15:32:35 -07004231/**
4232 * follow_pfn - look up PFN at a user virtual address
4233 * @vma: memory mapping
4234 * @address: user virtual address
4235 * @pfn: location to store found PFN
4236 *
4237 * Only IO mappings and raw PFN mappings are allowed.
4238 *
4239 * Returns zero and the pfn at @pfn on success, -ve otherwise.
4240 */
4241int follow_pfn(struct vm_area_struct *vma, unsigned long address,
4242 unsigned long *pfn)
4243{
4244 int ret = -EINVAL;
4245 spinlock_t *ptl;
4246 pte_t *ptep;
4247
4248 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
4249 return ret;
4250
4251 ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);
4252 if (ret)
4253 return ret;
4254 *pfn = pte_pfn(*ptep);
4255 pte_unmap_unlock(ptep, ptl);
4256 return 0;
4257}
4258EXPORT_SYMBOL(follow_pfn);
4259
Rik van Riel28b2ee22008-07-23 21:27:05 -07004260#ifdef CONFIG_HAVE_IOREMAP_PROT
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08004261int follow_phys(struct vm_area_struct *vma,
4262 unsigned long address, unsigned int flags,
4263 unsigned long *prot, resource_size_t *phys)
Rik van Riel28b2ee22008-07-23 21:27:05 -07004264{
Johannes Weiner03668a42009-06-16 15:32:34 -07004265 int ret = -EINVAL;
Rik van Riel28b2ee22008-07-23 21:27:05 -07004266 pte_t *ptep, pte;
4267 spinlock_t *ptl;
Rik van Riel28b2ee22008-07-23 21:27:05 -07004268
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08004269 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
4270 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07004271
Johannes Weiner03668a42009-06-16 15:32:34 -07004272 if (follow_pte(vma->vm_mm, address, &ptep, &ptl))
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08004273 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07004274 pte = *ptep;
Johannes Weiner03668a42009-06-16 15:32:34 -07004275
Rik van Riel28b2ee22008-07-23 21:27:05 -07004276 if ((flags & FOLL_WRITE) && !pte_write(pte))
4277 goto unlock;
Rik van Riel28b2ee22008-07-23 21:27:05 -07004278
4279 *prot = pgprot_val(pte_pgprot(pte));
Johannes Weiner03668a42009-06-16 15:32:34 -07004280 *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
Rik van Riel28b2ee22008-07-23 21:27:05 -07004281
Johannes Weiner03668a42009-06-16 15:32:34 -07004282 ret = 0;
Rik van Riel28b2ee22008-07-23 21:27:05 -07004283unlock:
4284 pte_unmap_unlock(ptep, ptl);
4285out:
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08004286 return ret;
Rik van Riel28b2ee22008-07-23 21:27:05 -07004287}
4288
4289int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
4290 void *buf, int len, int write)
4291{
4292 resource_size_t phys_addr;
4293 unsigned long prot = 0;
KOSAKI Motohiro2bc72732009-01-06 14:39:43 -08004294 void __iomem *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07004295 int offset = addr & (PAGE_SIZE-1);
4296
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08004297 if (follow_phys(vma, addr, write, &prot, &phys_addr))
Rik van Riel28b2ee22008-07-23 21:27:05 -07004298 return -EINVAL;
4299
Grazvydas Ignotas9cb12d72015-02-12 15:00:19 -08004300 maddr = ioremap_prot(phys_addr, PAGE_ALIGN(len + offset), prot);
Rik van Riel28b2ee22008-07-23 21:27:05 -07004301 if (write)
4302 memcpy_toio(maddr + offset, buf, len);
4303 else
4304 memcpy_fromio(buf, maddr + offset, len);
4305 iounmap(maddr);
4306
4307 return len;
4308}
Uwe Kleine-König5a736332013-08-07 13:02:52 +02004309EXPORT_SYMBOL_GPL(generic_access_phys);
Rik van Riel28b2ee22008-07-23 21:27:05 -07004310#endif
4311
David Howells0ec76a12006-09-27 01:50:15 -07004312/*
Stephen Wilson206cb632011-03-13 15:49:19 -04004313 * Access another process' address space as given in mm. If non-NULL, use the
4314 * given task for page fault accounting.
David Howells0ec76a12006-09-27 01:50:15 -07004315 */
Eric W. Biederman84d77d32016-11-22 12:06:50 -06004316int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
Lorenzo Stoakes442486e2016-10-13 01:20:18 +01004317 unsigned long addr, void *buf, int len, unsigned int gup_flags)
David Howells0ec76a12006-09-27 01:50:15 -07004318{
David Howells0ec76a12006-09-27 01:50:15 -07004319 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07004320 void *old_buf = buf;
Lorenzo Stoakes442486e2016-10-13 01:20:18 +01004321 int write = gup_flags & FOLL_WRITE;
David Howells0ec76a12006-09-27 01:50:15 -07004322
David Howells0ec76a12006-09-27 01:50:15 -07004323 down_read(&mm->mmap_sem);
Simon Arlott183ff222007-10-20 01:27:18 +02004324 /* ignore errors, just check how much was successfully transferred */
David Howells0ec76a12006-09-27 01:50:15 -07004325 while (len) {
4326 int bytes, ret, offset;
4327 void *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07004328 struct page *page = NULL;
David Howells0ec76a12006-09-27 01:50:15 -07004329
Dave Hansen1e987792016-02-12 13:01:54 -08004330 ret = get_user_pages_remote(tsk, mm, addr, 1,
Lorenzo Stoakes5b56d492016-12-14 15:06:52 -08004331 gup_flags, &page, &vma, NULL);
Rik van Riel28b2ee22008-07-23 21:27:05 -07004332 if (ret <= 0) {
Rik van Rieldbffcd02014-08-06 16:08:12 -07004333#ifndef CONFIG_HAVE_IOREMAP_PROT
4334 break;
4335#else
Rik van Riel28b2ee22008-07-23 21:27:05 -07004336 /*
4337 * Check if this is a VM_IO | VM_PFNMAP VMA, which
4338 * we can access using slightly different code.
4339 */
Rik van Riel28b2ee22008-07-23 21:27:05 -07004340 vma = find_vma(mm, addr);
Michael Ellermanfe936df2011-04-14 15:22:10 -07004341 if (!vma || vma->vm_start > addr)
Rik van Riel28b2ee22008-07-23 21:27:05 -07004342 break;
4343 if (vma->vm_ops && vma->vm_ops->access)
4344 ret = vma->vm_ops->access(vma, addr, buf,
4345 len, write);
4346 if (ret <= 0)
Rik van Riel28b2ee22008-07-23 21:27:05 -07004347 break;
4348 bytes = ret;
Rik van Rieldbffcd02014-08-06 16:08:12 -07004349#endif
David Howells0ec76a12006-09-27 01:50:15 -07004350 } else {
Rik van Riel28b2ee22008-07-23 21:27:05 -07004351 bytes = len;
4352 offset = addr & (PAGE_SIZE-1);
4353 if (bytes > PAGE_SIZE-offset)
4354 bytes = PAGE_SIZE-offset;
4355
4356 maddr = kmap(page);
4357 if (write) {
4358 copy_to_user_page(vma, page, addr,
4359 maddr + offset, buf, bytes);
4360 set_page_dirty_lock(page);
4361 } else {
4362 copy_from_user_page(vma, page, addr,
4363 buf, maddr + offset, bytes);
4364 }
4365 kunmap(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004366 put_page(page);
David Howells0ec76a12006-09-27 01:50:15 -07004367 }
David Howells0ec76a12006-09-27 01:50:15 -07004368 len -= bytes;
4369 buf += bytes;
4370 addr += bytes;
4371 }
4372 up_read(&mm->mmap_sem);
David Howells0ec76a12006-09-27 01:50:15 -07004373
4374 return buf - old_buf;
4375}
Andi Kleen03252912008-01-30 13:33:18 +01004376
Stephen Wilson5ddd36b2011-03-13 15:49:20 -04004377/**
Randy Dunlapae91dbf2011-03-26 13:27:01 -07004378 * access_remote_vm - access another process' address space
Stephen Wilson5ddd36b2011-03-13 15:49:20 -04004379 * @mm: the mm_struct of the target address space
4380 * @addr: start address to access
4381 * @buf: source or destination buffer
4382 * @len: number of bytes to transfer
Lorenzo Stoakes6347e8d2016-10-13 01:20:19 +01004383 * @gup_flags: flags modifying lookup behaviour
Stephen Wilson5ddd36b2011-03-13 15:49:20 -04004384 *
4385 * The caller must hold a reference on @mm.
4386 */
4387int access_remote_vm(struct mm_struct *mm, unsigned long addr,
Lorenzo Stoakes6347e8d2016-10-13 01:20:19 +01004388 void *buf, int len, unsigned int gup_flags)
Stephen Wilson5ddd36b2011-03-13 15:49:20 -04004389{
Lorenzo Stoakes6347e8d2016-10-13 01:20:19 +01004390 return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags);
Stephen Wilson5ddd36b2011-03-13 15:49:20 -04004391}
4392
Andi Kleen03252912008-01-30 13:33:18 +01004393/*
Stephen Wilson206cb632011-03-13 15:49:19 -04004394 * Access another process' address space.
4395 * Source/target buffer must be kernel space,
4396 * Do not walk the page table directly, use get_user_pages
4397 */
4398int access_process_vm(struct task_struct *tsk, unsigned long addr,
Lorenzo Stoakesf307ab62016-10-13 01:20:20 +01004399 void *buf, int len, unsigned int gup_flags)
Stephen Wilson206cb632011-03-13 15:49:19 -04004400{
4401 struct mm_struct *mm;
4402 int ret;
4403
4404 mm = get_task_mm(tsk);
4405 if (!mm)
4406 return 0;
4407
Lorenzo Stoakesf307ab62016-10-13 01:20:20 +01004408 ret = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags);
Lorenzo Stoakes442486e2016-10-13 01:20:18 +01004409
Stephen Wilson206cb632011-03-13 15:49:19 -04004410 mmput(mm);
4411
4412 return ret;
4413}
Catalin Marinasfcd35852016-11-01 14:43:25 -07004414EXPORT_SYMBOL_GPL(access_process_vm);
Stephen Wilson206cb632011-03-13 15:49:19 -04004415
Andi Kleen03252912008-01-30 13:33:18 +01004416/*
4417 * Print the name of a VMA.
4418 */
4419void print_vma_addr(char *prefix, unsigned long ip)
4420{
4421 struct mm_struct *mm = current->mm;
4422 struct vm_area_struct *vma;
4423
Ingo Molnare8bff742008-02-13 20:21:06 +01004424 /*
4425 * Do not print if we are in atomic
4426 * contexts (in exception stacks, etc.):
4427 */
4428 if (preempt_count())
4429 return;
4430
Andi Kleen03252912008-01-30 13:33:18 +01004431 down_read(&mm->mmap_sem);
4432 vma = find_vma(mm, ip);
4433 if (vma && vma->vm_file) {
4434 struct file *f = vma->vm_file;
4435 char *buf = (char *)__get_free_page(GFP_KERNEL);
4436 if (buf) {
Andy Shevchenko2fbc57c2012-12-17 16:01:23 -08004437 char *p;
Andi Kleen03252912008-01-30 13:33:18 +01004438
Miklos Szeredi9bf39ab2015-06-19 10:29:13 +02004439 p = file_path(f, buf, PAGE_SIZE);
Andi Kleen03252912008-01-30 13:33:18 +01004440 if (IS_ERR(p))
4441 p = "?";
Andy Shevchenko2fbc57c2012-12-17 16:01:23 -08004442 printk("%s%s[%lx+%lx]", prefix, kbasename(p),
Andi Kleen03252912008-01-30 13:33:18 +01004443 vma->vm_start,
4444 vma->vm_end - vma->vm_start);
4445 free_page((unsigned long)buf);
4446 }
4447 }
Jeff Liu51a07e52012-07-31 16:43:18 -07004448 up_read(&mm->mmap_sem);
Andi Kleen03252912008-01-30 13:33:18 +01004449}
Nick Piggin3ee1afa2008-09-10 13:37:17 +02004450
Michael S. Tsirkin662bbcb2013-05-26 17:32:23 +03004451#if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP)
David Hildenbrand9ec23532015-05-11 17:52:07 +02004452void __might_fault(const char *file, int line)
Nick Piggin3ee1afa2008-09-10 13:37:17 +02004453{
Peter Zijlstra95156f02009-01-12 13:02:11 +01004454 /*
4455 * Some code (nfs/sunrpc) uses socket ops on kernel memory while
4456 * holding the mmap_sem, this is safe because kernel memory doesn't
4457 * get paged out, therefore we'll never actually fault, and the
4458 * below annotations will generate false positives.
4459 */
Al Virodb68ce12017-03-20 21:08:07 -04004460 if (uaccess_kernel())
Peter Zijlstra95156f02009-01-12 13:02:11 +01004461 return;
David Hildenbrand9ec23532015-05-11 17:52:07 +02004462 if (pagefault_disabled())
Michael S. Tsirkin662bbcb2013-05-26 17:32:23 +03004463 return;
David Hildenbrand9ec23532015-05-11 17:52:07 +02004464 __might_sleep(file, line, 0);
4465#if defined(CONFIG_DEBUG_ATOMIC_SLEEP)
Michael S. Tsirkin662bbcb2013-05-26 17:32:23 +03004466 if (current->mm)
Nick Piggin3ee1afa2008-09-10 13:37:17 +02004467 might_lock_read(&current->mm->mmap_sem);
David Hildenbrand9ec23532015-05-11 17:52:07 +02004468#endif
Nick Piggin3ee1afa2008-09-10 13:37:17 +02004469}
David Hildenbrand9ec23532015-05-11 17:52:07 +02004470EXPORT_SYMBOL(__might_fault);
Nick Piggin3ee1afa2008-09-10 13:37:17 +02004471#endif
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08004472
4473#if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)
4474static void clear_gigantic_page(struct page *page,
4475 unsigned long addr,
4476 unsigned int pages_per_huge_page)
4477{
4478 int i;
4479 struct page *p = page;
4480
4481 might_sleep();
4482 for (i = 0; i < pages_per_huge_page;
4483 i++, p = mem_map_next(p, page, i)) {
4484 cond_resched();
4485 clear_user_highpage(p, addr + i * PAGE_SIZE);
4486 }
4487}
4488void clear_huge_page(struct page *page,
Huang Yingc79b57e2017-09-06 16:25:04 -07004489 unsigned long addr_hint, unsigned int pages_per_huge_page)
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08004490{
Huang Yingc79b57e2017-09-06 16:25:04 -07004491 int i, n, base, l;
4492 unsigned long addr = addr_hint &
4493 ~(((unsigned long)pages_per_huge_page << PAGE_SHIFT) - 1);
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08004494
4495 if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
4496 clear_gigantic_page(page, addr, pages_per_huge_page);
4497 return;
4498 }
4499
Huang Yingc79b57e2017-09-06 16:25:04 -07004500 /* Clear sub-page to access last to keep its cache lines hot */
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08004501 might_sleep();
Huang Yingc79b57e2017-09-06 16:25:04 -07004502 n = (addr_hint - addr) / PAGE_SIZE;
4503 if (2 * n <= pages_per_huge_page) {
4504 /* If sub-page to access in first half of huge page */
4505 base = 0;
4506 l = n;
4507 /* Clear sub-pages at the end of huge page */
4508 for (i = pages_per_huge_page - 1; i >= 2 * n; i--) {
4509 cond_resched();
4510 clear_user_highpage(page + i, addr + i * PAGE_SIZE);
4511 }
4512 } else {
4513 /* If sub-page to access in second half of huge page */
4514 base = pages_per_huge_page - 2 * (pages_per_huge_page - n);
4515 l = pages_per_huge_page - n;
4516 /* Clear sub-pages at the begin of huge page */
4517 for (i = 0; i < base; i++) {
4518 cond_resched();
4519 clear_user_highpage(page + i, addr + i * PAGE_SIZE);
4520 }
4521 }
4522 /*
4523 * Clear remaining sub-pages in left-right-left-right pattern
4524 * towards the sub-page to access
4525 */
4526 for (i = 0; i < l; i++) {
4527 int left_idx = base + i;
4528 int right_idx = base + 2 * l - 1 - i;
4529
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08004530 cond_resched();
Huang Yingc79b57e2017-09-06 16:25:04 -07004531 clear_user_highpage(page + left_idx,
4532 addr + left_idx * PAGE_SIZE);
4533 cond_resched();
4534 clear_user_highpage(page + right_idx,
4535 addr + right_idx * PAGE_SIZE);
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08004536 }
4537}
4538
4539static void copy_user_gigantic_page(struct page *dst, struct page *src,
4540 unsigned long addr,
4541 struct vm_area_struct *vma,
4542 unsigned int pages_per_huge_page)
4543{
4544 int i;
4545 struct page *dst_base = dst;
4546 struct page *src_base = src;
4547
4548 for (i = 0; i < pages_per_huge_page; ) {
4549 cond_resched();
4550 copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma);
4551
4552 i++;
4553 dst = mem_map_next(dst, dst_base, i);
4554 src = mem_map_next(src, src_base, i);
4555 }
4556}
4557
4558void copy_user_huge_page(struct page *dst, struct page *src,
4559 unsigned long addr, struct vm_area_struct *vma,
4560 unsigned int pages_per_huge_page)
4561{
4562 int i;
4563
4564 if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
4565 copy_user_gigantic_page(dst, src, addr, vma,
4566 pages_per_huge_page);
4567 return;
4568 }
4569
4570 might_sleep();
4571 for (i = 0; i < pages_per_huge_page; i++) {
4572 cond_resched();
4573 copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma);
4574 }
4575}
Mike Kravetzfa4d75c2017-02-22 15:42:49 -08004576
4577long copy_huge_page_from_user(struct page *dst_page,
4578 const void __user *usr_src,
Mike Kravetz810a56b2017-02-22 15:42:58 -08004579 unsigned int pages_per_huge_page,
4580 bool allow_pagefault)
Mike Kravetzfa4d75c2017-02-22 15:42:49 -08004581{
4582 void *src = (void *)usr_src;
4583 void *page_kaddr;
4584 unsigned long i, rc = 0;
4585 unsigned long ret_val = pages_per_huge_page * PAGE_SIZE;
4586
4587 for (i = 0; i < pages_per_huge_page; i++) {
Mike Kravetz810a56b2017-02-22 15:42:58 -08004588 if (allow_pagefault)
4589 page_kaddr = kmap(dst_page + i);
4590 else
4591 page_kaddr = kmap_atomic(dst_page + i);
Mike Kravetzfa4d75c2017-02-22 15:42:49 -08004592 rc = copy_from_user(page_kaddr,
4593 (const void __user *)(src + i * PAGE_SIZE),
4594 PAGE_SIZE);
Mike Kravetz810a56b2017-02-22 15:42:58 -08004595 if (allow_pagefault)
4596 kunmap(dst_page + i);
4597 else
4598 kunmap_atomic(page_kaddr);
Mike Kravetzfa4d75c2017-02-22 15:42:49 -08004599
4600 ret_val -= (PAGE_SIZE - rc);
4601 if (rc)
4602 break;
4603
4604 cond_resched();
4605 }
4606 return ret_val;
4607}
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08004608#endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08004609
Olof Johansson40b64ac2013-12-20 14:28:05 -08004610#if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS
Kirill A. Shutemovb35f1812014-01-21 15:49:07 -08004611
4612static struct kmem_cache *page_ptl_cachep;
4613
4614void __init ptlock_cache_init(void)
4615{
4616 page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0,
4617 SLAB_PANIC, NULL);
4618}
4619
Peter Zijlstra539edb52013-11-14 14:31:52 -08004620bool ptlock_alloc(struct page *page)
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08004621{
4622 spinlock_t *ptl;
4623
Kirill A. Shutemovb35f1812014-01-21 15:49:07 -08004624 ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL);
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08004625 if (!ptl)
4626 return false;
Peter Zijlstra539edb52013-11-14 14:31:52 -08004627 page->ptl = ptl;
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08004628 return true;
4629}
4630
Peter Zijlstra539edb52013-11-14 14:31:52 -08004631void ptlock_free(struct page *page)
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08004632{
Kirill A. Shutemovb35f1812014-01-21 15:49:07 -08004633 kmem_cache_free(page_ptl_cachep, page->ptl);
Kirill A. Shutemov49076ec2013-11-14 14:31:51 -08004634}
4635#endif