blob: fa2f04e0337c437e739e75a683ff5b80a070a95a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/memory.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 */
6
7/*
8 * demand-loading started 01.12.91 - seems it is high on the list of
9 * things wanted, and it should be easy to implement. - Linus
10 */
11
12/*
13 * Ok, demand-loading was easy, shared pages a little bit tricker. Shared
14 * pages started 02.12.91, seems to work. - Linus.
15 *
16 * Tested sharing by executing about 30 /bin/sh: under the old kernel it
17 * would have taken more than the 6M I have free, but it worked well as
18 * far as I could see.
19 *
20 * Also corrected some "invalidate()"s - I wasn't doing enough of them.
21 */
22
23/*
24 * Real VM (paging to/from disk) started 18.12.91. Much more work and
25 * thought has to go into this. Oh, well..
26 * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why.
27 * Found it. Everything seems to work now.
28 * 20.12.91 - Ok, making the swap-device changeable like the root.
29 */
30
31/*
32 * 05.04.94 - Multi-page memory management added for v1.1.
33 * Idea by Alex Bligh (alex@cconcepts.co.uk)
34 *
35 * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG
36 * (Gerhard.Wichert@pdb.siemens.de)
37 *
38 * Aug/Sep 2004 Changed to four level page tables (Andi Kleen)
39 */
40
41#include <linux/kernel_stat.h>
42#include <linux/mm.h>
43#include <linux/hugetlb.h>
44#include <linux/mman.h>
45#include <linux/swap.h>
46#include <linux/highmem.h>
47#include <linux/pagemap.h>
Hugh Dickins9a840892009-09-21 17:02:01 -070048#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include <linux/rmap.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040050#include <linux/export.h>
Shailabh Nagar0ff92242006-07-14 00:24:37 -070051#include <linux/delayacct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include <linux/init.h>
Peter Zijlstraedc79b22006-09-25 23:30:58 -070053#include <linux/writeback.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080054#include <linux/memcontrol.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070055#include <linux/mmu_notifier.h>
Hugh Dickins3dc14742009-01-06 14:40:08 -080056#include <linux/kallsyms.h>
57#include <linux/swapops.h>
58#include <linux/elf.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090059#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Alexey Dobriyan6952b612009-09-18 23:55:55 +040061#include <asm/io.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070062#include <asm/pgalloc.h>
63#include <asm/uaccess.h>
64#include <asm/tlb.h>
65#include <asm/tlbflush.h>
66#include <asm/pgtable.h>
67
Jan Beulich42b77722008-07-23 21:27:10 -070068#include "internal.h"
69
Andy Whitcroftd41dee32005-06-23 00:07:54 -070070#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -070071/* use the per-pgdat data instead for discontigmem - mbligh */
72unsigned long max_mapnr;
73struct page *mem_map;
74
75EXPORT_SYMBOL(max_mapnr);
76EXPORT_SYMBOL(mem_map);
77#endif
78
79unsigned long num_physpages;
80/*
81 * A number of key systems in x86 including ioremap() rely on the assumption
82 * that high_memory defines the upper bound on direct map memory, then end
83 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
84 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
85 * and ZONE_HIGHMEM.
86 */
87void * high_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
89EXPORT_SYMBOL(num_physpages);
90EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -070091
Ingo Molnar32a93232008-02-06 22:39:44 +010092/*
93 * Randomize the address space (stacks, mmaps, brk, etc.).
94 *
95 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
96 * as ancient (libc5 based) binaries can segfault. )
97 */
98int randomize_va_space __read_mostly =
99#ifdef CONFIG_COMPAT_BRK
100 1;
101#else
102 2;
103#endif
Andi Kleena62eaf12006-02-16 23:41:58 +0100104
105static int __init disable_randmaps(char *s)
106{
107 randomize_va_space = 0;
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800108 return 1;
Andi Kleena62eaf12006-02-16 23:41:58 +0100109}
110__setup("norandmaps", disable_randmaps);
111
Hugh Dickins62eede62009-09-21 17:03:34 -0700112unsigned long zero_pfn __read_mostly;
Hugh Dickins03f64622009-09-21 17:03:35 -0700113unsigned long highest_memmap_pfn __read_mostly;
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -0700114
115/*
116 * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init()
117 */
118static int __init init_zero_pfn(void)
119{
120 zero_pfn = page_to_pfn(ZERO_PAGE(0));
121 return 0;
122}
123core_initcall(init_zero_pfn);
Andi Kleena62eaf12006-02-16 23:41:58 +0100124
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800125
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800126#if defined(SPLIT_RSS_COUNTING)
127
KAMEZAWA Hiroyukia3a2e762010-04-06 14:34:42 -0700128static void __sync_task_rss_stat(struct task_struct *task, struct mm_struct *mm)
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800129{
130 int i;
131
132 for (i = 0; i < NR_MM_COUNTERS; i++) {
133 if (task->rss_stat.count[i]) {
134 add_mm_counter(mm, i, task->rss_stat.count[i]);
135 task->rss_stat.count[i] = 0;
136 }
137 }
138 task->rss_stat.events = 0;
139}
140
141static void add_mm_counter_fast(struct mm_struct *mm, int member, int val)
142{
143 struct task_struct *task = current;
144
145 if (likely(task->mm == mm))
146 task->rss_stat.count[member] += val;
147 else
148 add_mm_counter(mm, member, val);
149}
150#define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1)
151#define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1)
152
153/* sync counter once per 64 page faults */
154#define TASK_RSS_EVENTS_THRESH (64)
155static void check_sync_rss_stat(struct task_struct *task)
156{
157 if (unlikely(task != current))
158 return;
159 if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH))
160 __sync_task_rss_stat(task, task->mm);
161}
162
163unsigned long get_mm_counter(struct mm_struct *mm, int member)
164{
165 long val = 0;
166
167 /*
168 * Don't use task->mm here...for avoiding to use task_get_mm()..
169 * The caller must guarantee task->mm is not invalid.
170 */
171 val = atomic_long_read(&mm->rss_stat.count[member]);
172 /*
173 * counter is updated in asynchronous manner and may go to minus.
174 * But it's never be expected number for users.
175 */
176 if (val < 0)
177 return 0;
178 return (unsigned long)val;
179}
180
181void sync_mm_rss(struct task_struct *task, struct mm_struct *mm)
182{
183 __sync_task_rss_stat(task, mm);
184}
Peter Zijlstra9547d012011-05-24 17:12:14 -0700185#else /* SPLIT_RSS_COUNTING */
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800186
187#define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member)
188#define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member)
189
190static void check_sync_rss_stat(struct task_struct *task)
191{
192}
193
Peter Zijlstra9547d012011-05-24 17:12:14 -0700194#endif /* SPLIT_RSS_COUNTING */
195
196#ifdef HAVE_GENERIC_MMU_GATHER
197
198static int tlb_next_batch(struct mmu_gather *tlb)
199{
200 struct mmu_gather_batch *batch;
201
202 batch = tlb->active;
203 if (batch->next) {
204 tlb->active = batch->next;
205 return 1;
206 }
207
208 batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0);
209 if (!batch)
210 return 0;
211
212 batch->next = NULL;
213 batch->nr = 0;
214 batch->max = MAX_GATHER_BATCH;
215
216 tlb->active->next = batch;
217 tlb->active = batch;
218
219 return 1;
220}
221
222/* tlb_gather_mmu
223 * Called to initialize an (on-stack) mmu_gather structure for page-table
224 * tear-down from @mm. The @fullmm argument is used when @mm is without
225 * users and we're going to destroy the full address space (exit/execve).
226 */
227void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, bool fullmm)
228{
229 tlb->mm = mm;
230
231 tlb->fullmm = fullmm;
232 tlb->need_flush = 0;
233 tlb->fast_mode = (num_possible_cpus() == 1);
234 tlb->local.next = NULL;
235 tlb->local.nr = 0;
236 tlb->local.max = ARRAY_SIZE(tlb->__pages);
237 tlb->active = &tlb->local;
238
239#ifdef CONFIG_HAVE_RCU_TABLE_FREE
240 tlb->batch = NULL;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800241#endif
Peter Zijlstra9547d012011-05-24 17:12:14 -0700242}
243
244void tlb_flush_mmu(struct mmu_gather *tlb)
245{
246 struct mmu_gather_batch *batch;
247
248 if (!tlb->need_flush)
249 return;
250 tlb->need_flush = 0;
251 tlb_flush(tlb);
252#ifdef CONFIG_HAVE_RCU_TABLE_FREE
253 tlb_table_flush(tlb);
254#endif
255
256 if (tlb_fast_mode(tlb))
257 return;
258
259 for (batch = &tlb->local; batch; batch = batch->next) {
260 free_pages_and_swap_cache(batch->pages, batch->nr);
261 batch->nr = 0;
262 }
263 tlb->active = &tlb->local;
264}
265
266/* tlb_finish_mmu
267 * Called at the end of the shootdown operation to free up any resources
268 * that were required.
269 */
270void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end)
271{
272 struct mmu_gather_batch *batch, *next;
273
274 tlb_flush_mmu(tlb);
275
276 /* keep the page table cache within bounds */
277 check_pgt_cache();
278
279 for (batch = tlb->local.next; batch; batch = next) {
280 next = batch->next;
281 free_pages((unsigned long)batch, 0);
282 }
283 tlb->local.next = NULL;
284}
285
286/* __tlb_remove_page
287 * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while
288 * handling the additional races in SMP caused by other CPUs caching valid
289 * mappings in their TLBs. Returns the number of free page slots left.
290 * When out of page slots we must call tlb_flush_mmu().
291 */
292int __tlb_remove_page(struct mmu_gather *tlb, struct page *page)
293{
294 struct mmu_gather_batch *batch;
295
Shaohua Lif21760b2012-01-12 17:19:16 -0800296 VM_BUG_ON(!tlb->need_flush);
Peter Zijlstra9547d012011-05-24 17:12:14 -0700297
298 if (tlb_fast_mode(tlb)) {
299 free_page_and_swap_cache(page);
300 return 1; /* avoid calling tlb_flush_mmu() */
301 }
302
303 batch = tlb->active;
304 batch->pages[batch->nr++] = page;
305 if (batch->nr == batch->max) {
306 if (!tlb_next_batch(tlb))
307 return 0;
Shaohua Li0b43c3a2011-07-08 15:39:41 -0700308 batch = tlb->active;
Peter Zijlstra9547d012011-05-24 17:12:14 -0700309 }
310 VM_BUG_ON(batch->nr > batch->max);
311
312 return batch->max - batch->nr;
313}
314
315#endif /* HAVE_GENERIC_MMU_GATHER */
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800316
Peter Zijlstra26723912011-05-24 17:12:00 -0700317#ifdef CONFIG_HAVE_RCU_TABLE_FREE
318
319/*
320 * See the comment near struct mmu_table_batch.
321 */
322
323static void tlb_remove_table_smp_sync(void *arg)
324{
325 /* Simply deliver the interrupt */
326}
327
328static void tlb_remove_table_one(void *table)
329{
330 /*
331 * This isn't an RCU grace period and hence the page-tables cannot be
332 * assumed to be actually RCU-freed.
333 *
334 * It is however sufficient for software page-table walkers that rely on
335 * IRQ disabling. See the comment near struct mmu_table_batch.
336 */
337 smp_call_function(tlb_remove_table_smp_sync, NULL, 1);
338 __tlb_remove_table(table);
339}
340
341static void tlb_remove_table_rcu(struct rcu_head *head)
342{
343 struct mmu_table_batch *batch;
344 int i;
345
346 batch = container_of(head, struct mmu_table_batch, rcu);
347
348 for (i = 0; i < batch->nr; i++)
349 __tlb_remove_table(batch->tables[i]);
350
351 free_page((unsigned long)batch);
352}
353
354void tlb_table_flush(struct mmu_gather *tlb)
355{
356 struct mmu_table_batch **batch = &tlb->batch;
357
358 if (*batch) {
359 call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu);
360 *batch = NULL;
361 }
362}
363
364void tlb_remove_table(struct mmu_gather *tlb, void *table)
365{
366 struct mmu_table_batch **batch = &tlb->batch;
367
368 tlb->need_flush = 1;
369
370 /*
371 * When there's less then two users of this mm there cannot be a
372 * concurrent page-table walk.
373 */
374 if (atomic_read(&tlb->mm->mm_users) < 2) {
375 __tlb_remove_table(table);
376 return;
377 }
378
379 if (*batch == NULL) {
380 *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN);
381 if (*batch == NULL) {
382 tlb_remove_table_one(table);
383 return;
384 }
385 (*batch)->nr = 0;
386 }
387 (*batch)->tables[(*batch)->nr++] = table;
388 if ((*batch)->nr == MAX_TABLE_BATCH)
389 tlb_table_flush(tlb);
390}
391
Peter Zijlstra9547d012011-05-24 17:12:14 -0700392#endif /* CONFIG_HAVE_RCU_TABLE_FREE */
Peter Zijlstra26723912011-05-24 17:12:00 -0700393
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394/*
395 * If a p?d_bad entry is found while walking page tables, report
396 * the error, before resetting entry to p?d_none. Usually (but
397 * very seldom) called out from the p?d_none_or_clear_bad macros.
398 */
399
400void pgd_clear_bad(pgd_t *pgd)
401{
402 pgd_ERROR(*pgd);
403 pgd_clear(pgd);
404}
405
406void pud_clear_bad(pud_t *pud)
407{
408 pud_ERROR(*pud);
409 pud_clear(pud);
410}
411
412void pmd_clear_bad(pmd_t *pmd)
413{
414 pmd_ERROR(*pmd);
415 pmd_clear(pmd);
416}
417
418/*
419 * Note: this doesn't free the actual pages themselves. That
420 * has been handled earlier when unmapping all the memory regions.
421 */
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000422static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
423 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800425 pgtable_t token = pmd_pgtable(*pmd);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700426 pmd_clear(pmd);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000427 pte_free_tlb(tlb, token, addr);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700428 tlb->mm->nr_ptes--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429}
430
Hugh Dickinse0da3822005-04-19 13:29:15 -0700431static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
432 unsigned long addr, unsigned long end,
433 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434{
435 pmd_t *pmd;
436 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700437 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
Hugh Dickinse0da3822005-04-19 13:29:15 -0700439 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 do {
442 next = pmd_addr_end(addr, end);
443 if (pmd_none_or_clear_bad(pmd))
444 continue;
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000445 free_pte_range(tlb, pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 } while (pmd++, addr = next, addr != end);
447
Hugh Dickinse0da3822005-04-19 13:29:15 -0700448 start &= PUD_MASK;
449 if (start < floor)
450 return;
451 if (ceiling) {
452 ceiling &= PUD_MASK;
453 if (!ceiling)
454 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700456 if (end - 1 > ceiling - 1)
457 return;
458
459 pmd = pmd_offset(pud, start);
460 pud_clear(pud);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000461 pmd_free_tlb(tlb, pmd, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462}
463
Hugh Dickinse0da3822005-04-19 13:29:15 -0700464static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
465 unsigned long addr, unsigned long end,
466 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467{
468 pud_t *pud;
469 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700470 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
Hugh Dickinse0da3822005-04-19 13:29:15 -0700472 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 pud = pud_offset(pgd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 do {
475 next = pud_addr_end(addr, end);
476 if (pud_none_or_clear_bad(pud))
477 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700478 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 } while (pud++, addr = next, addr != end);
480
Hugh Dickinse0da3822005-04-19 13:29:15 -0700481 start &= PGDIR_MASK;
482 if (start < floor)
483 return;
484 if (ceiling) {
485 ceiling &= PGDIR_MASK;
486 if (!ceiling)
487 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700489 if (end - 1 > ceiling - 1)
490 return;
491
492 pud = pud_offset(pgd, start);
493 pgd_clear(pgd);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000494 pud_free_tlb(tlb, pud, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495}
496
497/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700498 * This function frees user-level page tables of a process.
499 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 * Must be called with pagetable lock held.
501 */
Jan Beulich42b77722008-07-23 21:27:10 -0700502void free_pgd_range(struct mmu_gather *tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700503 unsigned long addr, unsigned long end,
504 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505{
506 pgd_t *pgd;
507 unsigned long next;
508
Hugh Dickinse0da3822005-04-19 13:29:15 -0700509 /*
510 * The next few lines have given us lots of grief...
511 *
512 * Why are we testing PMD* at this top level? Because often
513 * there will be no work to do at all, and we'd prefer not to
514 * go all the way down to the bottom just to discover that.
515 *
516 * Why all these "- 1"s? Because 0 represents both the bottom
517 * of the address space and the top of it (using -1 for the
518 * top wouldn't help much: the masks would do the wrong thing).
519 * The rule is that addr 0 and floor 0 refer to the bottom of
520 * the address space, but end 0 and ceiling 0 refer to the top
521 * Comparisons need to use "end - 1" and "ceiling - 1" (though
522 * that end 0 case should be mythical).
523 *
524 * Wherever addr is brought up or ceiling brought down, we must
525 * be careful to reject "the opposite 0" before it confuses the
526 * subsequent tests. But what about where end is brought down
527 * by PMD_SIZE below? no, end can't go down to 0 there.
528 *
529 * Whereas we round start (addr) and ceiling down, by different
530 * masks at different levels, in order to test whether a table
531 * now has no other vmas using it, so can be freed, we don't
532 * bother to round floor or end up - the tests don't need that.
533 */
534
535 addr &= PMD_MASK;
536 if (addr < floor) {
537 addr += PMD_SIZE;
538 if (!addr)
539 return;
540 }
541 if (ceiling) {
542 ceiling &= PMD_MASK;
543 if (!ceiling)
544 return;
545 }
546 if (end - 1 > ceiling - 1)
547 end -= PMD_SIZE;
548 if (addr > end - 1)
549 return;
550
Jan Beulich42b77722008-07-23 21:27:10 -0700551 pgd = pgd_offset(tlb->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 do {
553 next = pgd_addr_end(addr, end);
554 if (pgd_none_or_clear_bad(pgd))
555 continue;
Jan Beulich42b77722008-07-23 21:27:10 -0700556 free_pud_range(tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700558}
559
Jan Beulich42b77722008-07-23 21:27:10 -0700560void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700561 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700562{
563 while (vma) {
564 struct vm_area_struct *next = vma->vm_next;
565 unsigned long addr = vma->vm_start;
566
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700567 /*
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000568 * Hide vma from rmap and truncate_pagecache before freeing
569 * pgtables
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700570 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800571 unlink_anon_vmas(vma);
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700572 unlink_file_vma(vma);
573
David Gibson9da61ae2006-03-22 00:08:57 -0800574 if (is_vm_hugetlb_page(vma)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700575 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700576 floor, next? next->vm_start: ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700577 } else {
578 /*
579 * Optimization: gather nearby vmas into one call down
580 */
581 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
David Gibson48669202006-03-22 00:08:58 -0800582 && !is_vm_hugetlb_page(next)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700583 vma = next;
584 next = vma->vm_next;
Rik van Riel5beb4932010-03-05 13:42:07 -0800585 unlink_anon_vmas(vma);
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700586 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700587 }
588 free_pgd_range(tlb, addr, vma->vm_end,
589 floor, next? next->vm_start: ceiling);
590 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700591 vma = next;
592 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593}
594
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800595int __pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,
596 pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800598 pgtable_t new = pte_alloc_one(mm, address);
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800599 int wait_split_huge_page;
Hugh Dickins1bb36302005-10-29 18:16:22 -0700600 if (!new)
601 return -ENOMEM;
602
Nick Piggin362a61a2008-05-14 06:37:36 +0200603 /*
604 * Ensure all pte setup (eg. pte page lock and page clearing) are
605 * visible before the pte is made visible to other CPUs by being
606 * put into page tables.
607 *
608 * The other side of the story is the pointer chasing in the page
609 * table walking code (when walking the page table without locking;
610 * ie. most of the time). Fortunately, these data accesses consist
611 * of a chain of data-dependent loads, meaning most CPUs (alpha
612 * being the notable exception) will already guarantee loads are
613 * seen in-order. See the alpha page table accessors for the
614 * smp_read_barrier_depends() barriers in page table walking code.
615 */
616 smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
617
Hugh Dickinsc74df322005-10-29 18:16:23 -0700618 spin_lock(&mm->page_table_lock);
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800619 wait_split_huge_page = 0;
620 if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 mm->nr_ptes++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 pmd_populate(mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800623 new = NULL;
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800624 } else if (unlikely(pmd_trans_splitting(*pmd)))
625 wait_split_huge_page = 1;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700626 spin_unlock(&mm->page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800627 if (new)
628 pte_free(mm, new);
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800629 if (wait_split_huge_page)
630 wait_split_huge_page(vma->anon_vma, pmd);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700631 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632}
633
Hugh Dickins1bb36302005-10-29 18:16:22 -0700634int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700636 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
637 if (!new)
638 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639
Nick Piggin362a61a2008-05-14 06:37:36 +0200640 smp_wmb(); /* See comment in __pte_alloc */
641
Hugh Dickins1bb36302005-10-29 18:16:22 -0700642 spin_lock(&init_mm.page_table_lock);
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800643 if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
Hugh Dickins1bb36302005-10-29 18:16:22 -0700644 pmd_populate_kernel(&init_mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800645 new = NULL;
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800646 } else
647 VM_BUG_ON(pmd_trans_splitting(*pmd));
Hugh Dickins1bb36302005-10-29 18:16:22 -0700648 spin_unlock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800649 if (new)
650 pte_free_kernel(&init_mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700651 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652}
653
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800654static inline void init_rss_vec(int *rss)
Hugh Dickinsae859762005-10-29 18:16:05 -0700655{
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800656 memset(rss, 0, sizeof(int) * NR_MM_COUNTERS);
657}
658
659static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss)
660{
661 int i;
662
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800663 if (current->mm == mm)
664 sync_mm_rss(current, mm);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800665 for (i = 0; i < NR_MM_COUNTERS; i++)
666 if (rss[i])
667 add_mm_counter(mm, i, rss[i]);
Hugh Dickinsae859762005-10-29 18:16:05 -0700668}
669
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800671 * This function is called to print an error when a bad pte
672 * is found. For example, we might have a PFN-mapped pte in
673 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700674 *
675 * The calling function must still handle the error.
676 */
Hugh Dickins3dc14742009-01-06 14:40:08 -0800677static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
678 pte_t pte, struct page *page)
Nick Pigginb5810032005-10-29 18:16:12 -0700679{
Hugh Dickins3dc14742009-01-06 14:40:08 -0800680 pgd_t *pgd = pgd_offset(vma->vm_mm, addr);
681 pud_t *pud = pud_offset(pgd, addr);
682 pmd_t *pmd = pmd_offset(pud, addr);
683 struct address_space *mapping;
684 pgoff_t index;
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800685 static unsigned long resume;
686 static unsigned long nr_shown;
687 static unsigned long nr_unshown;
688
689 /*
690 * Allow a burst of 60 reports, then keep quiet for that minute;
691 * or allow a steady drip of one report per second.
692 */
693 if (nr_shown == 60) {
694 if (time_before(jiffies, resume)) {
695 nr_unshown++;
696 return;
697 }
698 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800699 printk(KERN_ALERT
700 "BUG: Bad page map: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800701 nr_unshown);
702 nr_unshown = 0;
703 }
704 nr_shown = 0;
705 }
706 if (nr_shown++ == 0)
707 resume = jiffies + 60 * HZ;
Hugh Dickins3dc14742009-01-06 14:40:08 -0800708
709 mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;
710 index = linear_page_index(vma, addr);
711
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800712 printk(KERN_ALERT
713 "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800714 current->comm,
715 (long long)pte_val(pte), (long long)pmd_val(*pmd));
Wu Fengguang718a3822010-03-10 15:20:43 -0800716 if (page)
717 dump_page(page);
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800718 printk(KERN_ALERT
Hugh Dickins3dc14742009-01-06 14:40:08 -0800719 "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
720 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
721 /*
722 * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y
723 */
724 if (vma->vm_ops)
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800725 print_symbol(KERN_ALERT "vma->vm_ops->fault: %s\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800726 (unsigned long)vma->vm_ops->fault);
727 if (vma->vm_file && vma->vm_file->f_op)
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800728 print_symbol(KERN_ALERT "vma->vm_file->f_op->mmap: %s\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800729 (unsigned long)vma->vm_file->f_op->mmap);
Nick Pigginb5810032005-10-29 18:16:12 -0700730 dump_stack();
Hugh Dickins3dc14742009-01-06 14:40:08 -0800731 add_taint(TAINT_BAD_PAGE);
Nick Pigginb5810032005-10-29 18:16:12 -0700732}
733
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900734static inline int is_cow_mapping(vm_flags_t flags)
Linus Torvalds67121172005-12-11 20:38:17 -0800735{
736 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
737}
738
Hugh Dickins62eede62009-09-21 17:03:34 -0700739#ifndef is_zero_pfn
740static inline int is_zero_pfn(unsigned long pfn)
741{
742 return pfn == zero_pfn;
743}
744#endif
745
746#ifndef my_zero_pfn
747static inline unsigned long my_zero_pfn(unsigned long addr)
748{
749 return zero_pfn;
750}
751#endif
752
Nick Pigginb5810032005-10-29 18:16:12 -0700753/*
Nick Piggin7e675132008-04-28 02:13:00 -0700754 * vm_normal_page -- This function gets the "struct page" associated with a pte.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800755 *
Nick Piggin7e675132008-04-28 02:13:00 -0700756 * "Special" mappings do not wish to be associated with a "struct page" (either
757 * it doesn't exist, or it exists but they don't want to touch it). In this
758 * case, NULL is returned here. "Normal" mappings do have a struct page.
Jared Hulbertb379d792008-04-28 02:12:58 -0700759 *
Nick Piggin7e675132008-04-28 02:13:00 -0700760 * There are 2 broad cases. Firstly, an architecture may define a pte_special()
761 * pte bit, in which case this function is trivial. Secondly, an architecture
762 * may not have a spare pte bit, which requires a more complicated scheme,
763 * described below.
764 *
765 * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
766 * special mapping (even if there are underlying and valid "struct pages").
767 * COWed pages of a VM_PFNMAP are always normal.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800768 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700769 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
770 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
Nick Piggin7e675132008-04-28 02:13:00 -0700771 * set, and the vm_pgoff will point to the first PFN mapped: thus every special
772 * mapping will always honor the rule
Linus Torvalds6aab3412005-11-28 14:34:23 -0800773 *
774 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
775 *
Nick Piggin7e675132008-04-28 02:13:00 -0700776 * And for normal mappings this is false.
Jared Hulbertb379d792008-04-28 02:12:58 -0700777 *
Nick Piggin7e675132008-04-28 02:13:00 -0700778 * This restricts such mappings to be a linear translation from virtual address
779 * to pfn. To get around this restriction, we allow arbitrary mappings so long
780 * as the vma is not a COW mapping; in that case, we know that all ptes are
781 * special (because none can have been COWed).
Jared Hulbertb379d792008-04-28 02:12:58 -0700782 *
783 *
Nick Piggin7e675132008-04-28 02:13:00 -0700784 * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
785 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700786 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
787 * page" backing, however the difference is that _all_ pages with a struct
788 * page (that is, those where pfn_valid is true) are refcounted and considered
789 * normal pages by the VM. The disadvantage is that pages are refcounted
790 * (which can be slower and simply not an option for some PFNMAP users). The
791 * advantage is that we don't have to follow the strict linearity rule of
792 * PFNMAP mappings in order to support COWable mappings.
793 *
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800794 */
Nick Piggin7e675132008-04-28 02:13:00 -0700795#ifdef __HAVE_ARCH_PTE_SPECIAL
796# define HAVE_PTE_SPECIAL 1
797#else
798# define HAVE_PTE_SPECIAL 0
799#endif
800struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
801 pte_t pte)
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800802{
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800803 unsigned long pfn = pte_pfn(pte);
Nick Piggin7e675132008-04-28 02:13:00 -0700804
805 if (HAVE_PTE_SPECIAL) {
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800806 if (likely(!pte_special(pte)))
807 goto check_pfn;
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -0700808 if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))
809 return NULL;
Hugh Dickins62eede62009-09-21 17:03:34 -0700810 if (!is_zero_pfn(pfn))
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800811 print_bad_pte(vma, addr, pte, NULL);
Nick Piggin7e675132008-04-28 02:13:00 -0700812 return NULL;
813 }
814
815 /* !HAVE_PTE_SPECIAL case follows: */
816
Jared Hulbertb379d792008-04-28 02:12:58 -0700817 if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
818 if (vma->vm_flags & VM_MIXEDMAP) {
819 if (!pfn_valid(pfn))
820 return NULL;
821 goto out;
822 } else {
Nick Piggin7e675132008-04-28 02:13:00 -0700823 unsigned long off;
824 off = (addr - vma->vm_start) >> PAGE_SHIFT;
Jared Hulbertb379d792008-04-28 02:12:58 -0700825 if (pfn == vma->vm_pgoff + off)
826 return NULL;
827 if (!is_cow_mapping(vma->vm_flags))
828 return NULL;
829 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800830 }
831
Hugh Dickins62eede62009-09-21 17:03:34 -0700832 if (is_zero_pfn(pfn))
833 return NULL;
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800834check_pfn:
835 if (unlikely(pfn > highest_memmap_pfn)) {
836 print_bad_pte(vma, addr, pte, NULL);
837 return NULL;
838 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800839
840 /*
Nick Piggin7e675132008-04-28 02:13:00 -0700841 * NOTE! We still have PageReserved() pages in the page tables.
Nick Piggin7e675132008-04-28 02:13:00 -0700842 * eg. VDSO mappings can cause them to exist.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800843 */
Jared Hulbertb379d792008-04-28 02:12:58 -0700844out:
Linus Torvalds6aab3412005-11-28 14:34:23 -0800845 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800846}
847
848/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 * copy one vm_area from one task to the other. Assumes the page tables
850 * already present in the new task to be cleared in the whole range
851 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 */
853
Hugh Dickins570a3352009-12-14 17:58:46 -0800854static inline unsigned long
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700856 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700857 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858{
Nick Pigginb5810032005-10-29 18:16:12 -0700859 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 pte_t pte = *src_pte;
861 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862
863 /* pte contains position in swap or file, so copy. */
864 if (unlikely(!pte_present(pte))) {
865 if (!pte_file(pte)) {
Christoph Lameter06972122006-06-23 02:03:35 -0700866 swp_entry_t entry = pte_to_swp_entry(pte);
867
Hugh Dickins570a3352009-12-14 17:58:46 -0800868 if (swap_duplicate(entry) < 0)
869 return entry.val;
870
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 /* make sure dst_mm is on swapoff's mmlist. */
872 if (unlikely(list_empty(&dst_mm->mmlist))) {
873 spin_lock(&mmlist_lock);
Hugh Dickinsf412ac02005-10-29 18:16:41 -0700874 if (list_empty(&dst_mm->mmlist))
875 list_add(&dst_mm->mmlist,
876 &src_mm->mmlist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 spin_unlock(&mmlist_lock);
878 }
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -0800879 if (likely(!non_swap_entry(entry)))
880 rss[MM_SWAPENTS]++;
Konstantin Khlebnikov9f9f1ac2012-01-20 14:34:24 -0800881 else if (is_migration_entry(entry)) {
882 page = migration_entry_to_page(entry);
883
884 if (PageAnon(page))
885 rss[MM_ANONPAGES]++;
886 else
887 rss[MM_FILEPAGES]++;
888
889 if (is_write_migration_entry(entry) &&
890 is_cow_mapping(vm_flags)) {
891 /*
892 * COW mappings require pages in both
893 * parent and child to be set to read.
894 */
895 make_migration_entry_read(&entry);
896 pte = swp_entry_to_pte(entry);
897 set_pte_at(src_mm, addr, src_pte, pte);
898 }
Christoph Lameter06972122006-06-23 02:03:35 -0700899 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700901 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 }
903
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 /*
905 * If it's a COW mapping, write protect it both
906 * in the parent and the child
907 */
Linus Torvalds67121172005-12-11 20:38:17 -0800908 if (is_cow_mapping(vm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 ptep_set_wrprotect(src_mm, addr, src_pte);
Zachary Amsden3dc90792006-09-30 23:29:30 -0700910 pte = pte_wrprotect(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 }
912
913 /*
914 * If it's a shared mapping, mark it clean in
915 * the child
916 */
917 if (vm_flags & VM_SHARED)
918 pte = pte_mkclean(pte);
919 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800920
921 page = vm_normal_page(vma, addr, pte);
922 if (page) {
923 get_page(page);
Hugh Dickins21333b22009-09-21 17:01:59 -0700924 page_dup_rmap(page);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800925 if (PageAnon(page))
926 rss[MM_ANONPAGES]++;
927 else
928 rss[MM_FILEPAGES]++;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800929 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700930
931out_set_pte:
932 set_pte_at(dst_mm, addr, dst_pte, pte);
Hugh Dickins570a3352009-12-14 17:58:46 -0800933 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934}
935
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -0800936int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
937 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
938 unsigned long addr, unsigned long end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939{
Daisuke Nishimurac36987e2009-10-26 16:50:23 -0700940 pte_t *orig_src_pte, *orig_dst_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700942 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -0700943 int progress = 0;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800944 int rss[NR_MM_COUNTERS];
Hugh Dickins570a3352009-12-14 17:58:46 -0800945 swp_entry_t entry = (swp_entry_t){0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946
947again:
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800948 init_rss_vec(rss);
949
Hugh Dickinsc74df322005-10-29 18:16:23 -0700950 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 if (!dst_pte)
952 return -ENOMEM;
Peter Zijlstraece0e2b2010-10-26 14:21:52 -0700953 src_pte = pte_offset_map(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700954 src_ptl = pte_lockptr(src_mm, src_pmd);
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700955 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
Daisuke Nishimurac36987e2009-10-26 16:50:23 -0700956 orig_src_pte = src_pte;
957 orig_dst_pte = dst_pte;
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700958 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 do {
961 /*
962 * We are holding two locks at this point - either of them
963 * could generate latencies in another task on another CPU.
964 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700965 if (progress >= 32) {
966 progress = 0;
967 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +0100968 spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700969 break;
970 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 if (pte_none(*src_pte)) {
972 progress++;
973 continue;
974 }
Hugh Dickins570a3352009-12-14 17:58:46 -0800975 entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte,
976 vma, addr, rss);
977 if (entry.val)
978 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 progress += 8;
980 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700982 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -0700983 spin_unlock(src_ptl);
Peter Zijlstraece0e2b2010-10-26 14:21:52 -0700984 pte_unmap(orig_src_pte);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800985 add_mm_rss_vec(dst_mm, rss);
Daisuke Nishimurac36987e2009-10-26 16:50:23 -0700986 pte_unmap_unlock(orig_dst_pte, dst_ptl);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700987 cond_resched();
Hugh Dickins570a3352009-12-14 17:58:46 -0800988
989 if (entry.val) {
990 if (add_swap_count_continuation(entry, GFP_KERNEL) < 0)
991 return -ENOMEM;
992 progress = 0;
993 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 if (addr != end)
995 goto again;
996 return 0;
997}
998
999static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
1000 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
1001 unsigned long addr, unsigned long end)
1002{
1003 pmd_t *src_pmd, *dst_pmd;
1004 unsigned long next;
1005
1006 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
1007 if (!dst_pmd)
1008 return -ENOMEM;
1009 src_pmd = pmd_offset(src_pud, addr);
1010 do {
1011 next = pmd_addr_end(addr, end);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001012 if (pmd_trans_huge(*src_pmd)) {
1013 int err;
Andrea Arcangeli14d1a552011-01-13 15:47:15 -08001014 VM_BUG_ON(next-addr != HPAGE_PMD_SIZE);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001015 err = copy_huge_pmd(dst_mm, src_mm,
1016 dst_pmd, src_pmd, addr, vma);
1017 if (err == -ENOMEM)
1018 return -ENOMEM;
1019 if (!err)
1020 continue;
1021 /* fall through */
1022 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 if (pmd_none_or_clear_bad(src_pmd))
1024 continue;
1025 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
1026 vma, addr, next))
1027 return -ENOMEM;
1028 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
1029 return 0;
1030}
1031
1032static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
1033 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
1034 unsigned long addr, unsigned long end)
1035{
1036 pud_t *src_pud, *dst_pud;
1037 unsigned long next;
1038
1039 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
1040 if (!dst_pud)
1041 return -ENOMEM;
1042 src_pud = pud_offset(src_pgd, addr);
1043 do {
1044 next = pud_addr_end(addr, end);
1045 if (pud_none_or_clear_bad(src_pud))
1046 continue;
1047 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
1048 vma, addr, next))
1049 return -ENOMEM;
1050 } while (dst_pud++, src_pud++, addr = next, addr != end);
1051 return 0;
1052}
1053
1054int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
1055 struct vm_area_struct *vma)
1056{
1057 pgd_t *src_pgd, *dst_pgd;
1058 unsigned long next;
1059 unsigned long addr = vma->vm_start;
1060 unsigned long end = vma->vm_end;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001061 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062
Nick Piggind9928952005-08-28 16:49:11 +10001063 /*
1064 * Don't copy ptes where a page fault will fill them correctly.
1065 * Fork becomes much lighter when there are big shared or private
1066 * readonly mappings. The tradeoff is that copy_page_range is more
1067 * efficient than faulting.
1068 */
Linus Torvalds4d7672b2005-12-16 10:21:23 -08001069 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
Nick Piggind9928952005-08-28 16:49:11 +10001070 if (!vma->anon_vma)
1071 return 0;
1072 }
1073
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 if (is_vm_hugetlb_page(vma))
1075 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
1076
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -08001077 if (unlikely(is_pfn_mapping(vma))) {
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001078 /*
1079 * We do not free on error cases below as remove_vma
1080 * gets called on error from higher level routine
1081 */
1082 ret = track_pfn_vma_copy(vma);
1083 if (ret)
1084 return ret;
1085 }
1086
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001087 /*
1088 * We need to invalidate the secondary MMU mappings only when
1089 * there could be a permission downgrade on the ptes of the
1090 * parent mm. And a permission downgrade will only happen if
1091 * is_cow_mapping() returns true.
1092 */
1093 if (is_cow_mapping(vma->vm_flags))
1094 mmu_notifier_invalidate_range_start(src_mm, addr, end);
1095
1096 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 dst_pgd = pgd_offset(dst_mm, addr);
1098 src_pgd = pgd_offset(src_mm, addr);
1099 do {
1100 next = pgd_addr_end(addr, end);
1101 if (pgd_none_or_clear_bad(src_pgd))
1102 continue;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001103 if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
1104 vma, addr, next))) {
1105 ret = -ENOMEM;
1106 break;
1107 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001109
1110 if (is_cow_mapping(vma->vm_flags))
1111 mmu_notifier_invalidate_range_end(src_mm,
1112 vma->vm_start, end);
1113 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114}
1115
Robin Holt51c6f662005-11-13 16:06:42 -08001116static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -07001117 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 unsigned long addr, unsigned long end,
Peter Zijlstra97a89412011-05-24 17:12:04 -07001119 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120{
Nick Pigginb5810032005-10-29 18:16:12 -07001121 struct mm_struct *mm = tlb->mm;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001122 int force_flush = 0;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001123 int rss[NR_MM_COUNTERS];
Peter Zijlstra97a89412011-05-24 17:12:04 -07001124 spinlock_t *ptl;
Steven Rostedt5f1a1902011-06-15 15:08:23 -07001125 pte_t *start_pte;
Peter Zijlstra97a89412011-05-24 17:12:04 -07001126 pte_t *pte;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001127
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001128again:
Peter Zijlstrae3032972011-05-24 17:12:01 -07001129 init_rss_vec(rss);
Steven Rostedt5f1a1902011-06-15 15:08:23 -07001130 start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
1131 pte = start_pte;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001132 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 do {
1134 pte_t ptent = *pte;
Robin Holt51c6f662005-11-13 16:06:42 -08001135 if (pte_none(ptent)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 continue;
Robin Holt51c6f662005-11-13 16:06:42 -08001137 }
Hugh Dickins6f5e6b92006-03-16 23:04:09 -08001138
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -08001140 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -08001141
Linus Torvalds6aab3412005-11-28 14:34:23 -08001142 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 if (unlikely(details) && page) {
1144 /*
1145 * unmap_shared_mapping_pages() wants to
1146 * invalidate cache without truncating:
1147 * unmap shared but keep private pages.
1148 */
1149 if (details->check_mapping &&
1150 details->check_mapping != page->mapping)
1151 continue;
1152 /*
1153 * Each page->index must be checked when
1154 * invalidating or truncating nonlinear.
1155 */
1156 if (details->nonlinear_vma &&
1157 (page->index < details->first_index ||
1158 page->index > details->last_index))
1159 continue;
1160 }
Nick Pigginb5810032005-10-29 18:16:12 -07001161 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -07001162 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 tlb_remove_tlb_entry(tlb, pte, addr);
1164 if (unlikely(!page))
1165 continue;
1166 if (unlikely(details) && details->nonlinear_vma
1167 && linear_page_index(details->nonlinear_vma,
1168 addr) != page->index)
Nick Pigginb5810032005-10-29 18:16:12 -07001169 set_pte_at(mm, addr, pte,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 pgoff_to_pte(page->index));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 if (PageAnon(page))
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001172 rss[MM_ANONPAGES]--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -07001173 else {
1174 if (pte_dirty(ptent))
1175 set_page_dirty(page);
Johannes Weiner4917e5d2009-01-06 14:39:17 -08001176 if (pte_young(ptent) &&
1177 likely(!VM_SequentialReadHint(vma)))
Nick Pigginbf3f3bc2009-01-06 14:38:55 -08001178 mark_page_accessed(page);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001179 rss[MM_FILEPAGES]--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -07001180 }
Hugh Dickinsedc315f2009-01-06 14:40:11 -08001181 page_remove_rmap(page);
Hugh Dickins3dc14742009-01-06 14:40:08 -08001182 if (unlikely(page_mapcount(page) < 0))
1183 print_bad_pte(vma, addr, ptent, page);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001184 force_flush = !__tlb_remove_page(tlb, page);
1185 if (force_flush)
1186 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 continue;
1188 }
1189 /*
1190 * If details->check_mapping, we leave swap entries;
1191 * if details->nonlinear_vma, we leave file entries.
1192 */
1193 if (unlikely(details))
1194 continue;
Hugh Dickins2509ef22009-01-06 14:40:10 -08001195 if (pte_file(ptent)) {
1196 if (unlikely(!(vma->vm_flags & VM_NONLINEAR)))
1197 print_bad_pte(vma, addr, ptent, NULL);
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001198 } else {
1199 swp_entry_t entry = pte_to_swp_entry(ptent);
1200
1201 if (!non_swap_entry(entry))
1202 rss[MM_SWAPENTS]--;
Konstantin Khlebnikov9f9f1ac2012-01-20 14:34:24 -08001203 else if (is_migration_entry(entry)) {
1204 struct page *page;
1205
1206 page = migration_entry_to_page(entry);
1207
1208 if (PageAnon(page))
1209 rss[MM_ANONPAGES]--;
1210 else
1211 rss[MM_FILEPAGES]--;
1212 }
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08001213 if (unlikely(!free_swap_and_cache(entry)))
1214 print_bad_pte(vma, addr, ptent, NULL);
1215 }
Zachary Amsden9888a1c2006-09-30 23:29:31 -07001216 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
Peter Zijlstra97a89412011-05-24 17:12:04 -07001217 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickinsae859762005-10-29 18:16:05 -07001218
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -08001219 add_mm_rss_vec(mm, rss);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001220 arch_leave_lazy_mmu_mode();
Steven Rostedt5f1a1902011-06-15 15:08:23 -07001221 pte_unmap_unlock(start_pte, ptl);
Robin Holt51c6f662005-11-13 16:06:42 -08001222
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001223 /*
1224 * mmu_gather ran out of room to batch pages, we break out of
1225 * the PTE lock to avoid doing the potential expensive TLB invalidate
1226 * and page-free while holding it.
1227 */
1228 if (force_flush) {
1229 force_flush = 0;
1230 tlb_flush_mmu(tlb);
1231 if (addr != end)
1232 goto again;
1233 }
1234
Robin Holt51c6f662005-11-13 16:06:42 -08001235 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236}
1237
Robin Holt51c6f662005-11-13 16:06:42 -08001238static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -07001239 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 unsigned long addr, unsigned long end,
Peter Zijlstra97a89412011-05-24 17:12:04 -07001241 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242{
1243 pmd_t *pmd;
1244 unsigned long next;
1245
1246 pmd = pmd_offset(pud, addr);
1247 do {
1248 next = pmd_addr_end(addr, end);
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001249 if (pmd_trans_huge(*pmd)) {
Andrea Arcangeli14d1a552011-01-13 15:47:15 -08001250 if (next-addr != HPAGE_PMD_SIZE) {
1251 VM_BUG_ON(!rwsem_is_locked(&tlb->mm->mmap_sem));
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001252 split_huge_page_pmd(vma->vm_mm, pmd);
Shaohua Lif21760b2012-01-12 17:19:16 -08001253 } else if (zap_huge_pmd(tlb, vma, pmd, addr))
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001254 continue;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001255 /* fall through */
1256 }
Peter Zijlstra97a89412011-05-24 17:12:04 -07001257 if (pmd_none_or_clear_bad(pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 continue;
Peter Zijlstra97a89412011-05-24 17:12:04 -07001259 next = zap_pte_range(tlb, vma, pmd, addr, next, details);
1260 cond_resched();
1261 } while (pmd++, addr = next, addr != end);
Robin Holt51c6f662005-11-13 16:06:42 -08001262
1263 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264}
1265
Robin Holt51c6f662005-11-13 16:06:42 -08001266static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -07001267 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 unsigned long addr, unsigned long end,
Peter Zijlstra97a89412011-05-24 17:12:04 -07001269 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270{
1271 pud_t *pud;
1272 unsigned long next;
1273
1274 pud = pud_offset(pgd, addr);
1275 do {
1276 next = pud_addr_end(addr, end);
Peter Zijlstra97a89412011-05-24 17:12:04 -07001277 if (pud_none_or_clear_bad(pud))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 continue;
Peter Zijlstra97a89412011-05-24 17:12:04 -07001279 next = zap_pmd_range(tlb, vma, pud, addr, next, details);
1280 } while (pud++, addr = next, addr != end);
Robin Holt51c6f662005-11-13 16:06:42 -08001281
1282 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283}
1284
Robin Holt51c6f662005-11-13 16:06:42 -08001285static unsigned long unmap_page_range(struct mmu_gather *tlb,
1286 struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 unsigned long addr, unsigned long end,
Peter Zijlstra97a89412011-05-24 17:12:04 -07001288 struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289{
1290 pgd_t *pgd;
1291 unsigned long next;
1292
1293 if (details && !details->check_mapping && !details->nonlinear_vma)
1294 details = NULL;
1295
1296 BUG_ON(addr >= end);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08001297 mem_cgroup_uncharge_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 tlb_start_vma(tlb, vma);
1299 pgd = pgd_offset(vma->vm_mm, addr);
1300 do {
1301 next = pgd_addr_end(addr, end);
Peter Zijlstra97a89412011-05-24 17:12:04 -07001302 if (pgd_none_or_clear_bad(pgd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 continue;
Peter Zijlstra97a89412011-05-24 17:12:04 -07001304 next = zap_pud_range(tlb, vma, pgd, addr, next, details);
1305 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 tlb_end_vma(tlb, vma);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08001307 mem_cgroup_uncharge_end();
Robin Holt51c6f662005-11-13 16:06:42 -08001308
1309 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310}
1311
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312/**
1313 * unmap_vmas - unmap a range of memory covered by a list of vma's
Randy Dunlap0164f692011-06-15 15:08:09 -07001314 * @tlb: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 * @vma: the starting vma
1316 * @start_addr: virtual address at which to start unmapping
1317 * @end_addr: virtual address at which to end unmapping
1318 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
1319 * @details: details of nonlinear truncation or shared cache invalidation
1320 *
Hugh Dickinsee39b372005-04-19 13:29:15 -07001321 * Returns the end address of the unmapping (restart addr if interrupted).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 *
Hugh Dickins508034a2005-10-29 18:16:30 -07001323 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 * Only addresses between `start' and `end' will be unmapped.
1326 *
1327 * The VMA list must be sorted in ascending virtual address order.
1328 *
1329 * unmap_vmas() assumes that the caller will flush the whole unmapped address
1330 * range after unmap_vmas() returns. So the only responsibility here is to
1331 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
1332 * drops the lock and schedules.
1333 */
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001334unsigned long unmap_vmas(struct mmu_gather *tlb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 struct vm_area_struct *vma, unsigned long start_addr,
1336 unsigned long end_addr, unsigned long *nr_accounted,
1337 struct zap_details *details)
1338{
Hugh Dickinsee39b372005-04-19 13:29:15 -07001339 unsigned long start = start_addr;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001340 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001342 mmu_notifier_invalidate_range_start(mm, start_addr, end_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 unsigned long end;
1345
1346 start = max(vma->vm_start, start_addr);
1347 if (start >= vma->vm_end)
1348 continue;
1349 end = min(vma->vm_end, end_addr);
1350 if (end <= vma->vm_start)
1351 continue;
1352
1353 if (vma->vm_flags & VM_ACCOUNT)
1354 *nr_accounted += (end - start) >> PAGE_SHIFT;
1355
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -08001356 if (unlikely(is_pfn_mapping(vma)))
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001357 untrack_pfn_vma(vma, 0, 0);
1358
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 while (start != end) {
Robin Holt51c6f662005-11-13 16:06:42 -08001360 if (unlikely(is_vm_hugetlb_page(vma))) {
Andi Kleena137e1c2008-07-23 21:27:43 -07001361 /*
1362 * It is undesirable to test vma->vm_file as it
1363 * should be non-null for valid hugetlb area.
1364 * However, vm_file will be NULL in the error
1365 * cleanup path of do_mmap_pgoff. When
1366 * hugetlbfs ->mmap method fails,
1367 * do_mmap_pgoff() nullifies vma->vm_file
1368 * before calling this function to clean up.
1369 * Since no pte has actually been setup, it is
1370 * safe to do nothing in this case.
1371 */
Peter Zijlstra97a89412011-05-24 17:12:04 -07001372 if (vma->vm_file)
Andi Kleena137e1c2008-07-23 21:27:43 -07001373 unmap_hugepage_range(vma, start, end, NULL);
Andi Kleena137e1c2008-07-23 21:27:43 -07001374
Robin Holt51c6f662005-11-13 16:06:42 -08001375 start = end;
1376 } else
Peter Zijlstra97a89412011-05-24 17:12:04 -07001377 start = unmap_page_range(tlb, vma, start, end, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 }
1379 }
Peter Zijlstra97a89412011-05-24 17:12:04 -07001380
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001381 mmu_notifier_invalidate_range_end(mm, start_addr, end_addr);
Hugh Dickinsee39b372005-04-19 13:29:15 -07001382 return start; /* which is now the end (or restart) address */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383}
1384
1385/**
1386 * zap_page_range - remove user pages in a given range
1387 * @vma: vm_area_struct holding the applicable pages
1388 * @address: starting address of pages to zap
1389 * @size: number of bytes to zap
1390 * @details: details of nonlinear truncation or shared cache invalidation
1391 */
Hugh Dickinsee39b372005-04-19 13:29:15 -07001392unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 unsigned long size, struct zap_details *details)
1394{
1395 struct mm_struct *mm = vma->vm_mm;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001396 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 unsigned long end = address + size;
1398 unsigned long nr_accounted = 0;
1399
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 lru_add_drain();
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001401 tlb_gather_mmu(&tlb, mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001402 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -07001403 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
Peter Zijlstrad16dfc52011-05-24 17:11:45 -07001404 tlb_finish_mmu(&tlb, address, end);
Hugh Dickinsee39b372005-04-19 13:29:15 -07001405 return end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406}
1407
Jack Steinerc627f9c2008-07-29 22:33:53 -07001408/**
1409 * zap_vma_ptes - remove ptes mapping the vma
1410 * @vma: vm_area_struct holding ptes to be zapped
1411 * @address: starting address of pages to zap
1412 * @size: number of bytes to zap
1413 *
1414 * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1415 *
1416 * The entire address range must be fully contained within the vma.
1417 *
1418 * Returns 0 if successful.
1419 */
1420int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
1421 unsigned long size)
1422{
1423 if (address < vma->vm_start || address + size > vma->vm_end ||
1424 !(vma->vm_flags & VM_PFNMAP))
1425 return -1;
1426 zap_page_range(vma, address, size, NULL);
1427 return 0;
1428}
1429EXPORT_SYMBOL_GPL(zap_vma_ptes);
1430
Johannes Weiner142762b2010-05-24 14:32:39 -07001431/**
1432 * follow_page - look up a page descriptor from a user-virtual address
1433 * @vma: vm_area_struct mapping @address
1434 * @address: virtual address to look up
1435 * @flags: flags modifying lookup behaviour
1436 *
1437 * @flags can have FOLL_ flags set, defined in <linux/mm.h>
1438 *
1439 * Returns the mapped (struct page *), %NULL if no mapping exists, or
1440 * an error pointer if there is a mapping to something not represented
1441 * by a page descriptor (see also vm_normal_page()).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 */
Linus Torvalds6aab3412005-11-28 14:34:23 -08001443struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001444 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445{
1446 pgd_t *pgd;
1447 pud_t *pud;
1448 pmd_t *pmd;
1449 pte_t *ptep, pte;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001450 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 struct page *page;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001452 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001454 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
1455 if (!IS_ERR(page)) {
1456 BUG_ON(flags & FOLL_GET);
1457 goto out;
1458 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001460 page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 pgd = pgd_offset(mm, address);
1462 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001463 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464
1465 pud = pud_offset(pgd, address);
Andi Kleenceb86872008-07-23 21:27:50 -07001466 if (pud_none(*pud))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001467 goto no_page_table;
Hugh Dickins8a076512011-01-13 15:46:52 -08001468 if (pud_huge(*pud) && vma->vm_flags & VM_HUGETLB) {
Andi Kleenceb86872008-07-23 21:27:50 -07001469 BUG_ON(flags & FOLL_GET);
1470 page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE);
1471 goto out;
1472 }
1473 if (unlikely(pud_bad(*pud)))
1474 goto no_page_table;
1475
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 pmd = pmd_offset(pud, address);
Hugh Dickinsaeed5fc2008-05-06 20:49:23 +01001477 if (pmd_none(*pmd))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001478 goto no_page_table;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001479 if (pmd_huge(*pmd) && vma->vm_flags & VM_HUGETLB) {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001480 BUG_ON(flags & FOLL_GET);
1481 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
1482 goto out;
1483 }
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001484 if (pmd_trans_huge(*pmd)) {
Andrea Arcangeli500d65d2011-01-13 15:46:55 -08001485 if (flags & FOLL_SPLIT) {
1486 split_huge_page_pmd(mm, pmd);
1487 goto split_fallthrough;
1488 }
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08001489 spin_lock(&mm->page_table_lock);
1490 if (likely(pmd_trans_huge(*pmd))) {
1491 if (unlikely(pmd_trans_splitting(*pmd))) {
1492 spin_unlock(&mm->page_table_lock);
1493 wait_split_huge_page(vma->anon_vma, pmd);
1494 } else {
1495 page = follow_trans_huge_pmd(mm, address,
1496 pmd, flags);
1497 spin_unlock(&mm->page_table_lock);
1498 goto out;
1499 }
1500 } else
1501 spin_unlock(&mm->page_table_lock);
1502 /* fall through */
1503 }
Andrea Arcangeli500d65d2011-01-13 15:46:55 -08001504split_fallthrough:
Hugh Dickinsaeed5fc2008-05-06 20:49:23 +01001505 if (unlikely(pmd_bad(*pmd)))
1506 goto no_page_table;
1507
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001508 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509
1510 pte = *ptep;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001511 if (!pte_present(pte))
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001512 goto no_page;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001513 if ((flags & FOLL_WRITE) && !pte_write(pte))
1514 goto unlock;
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07001515
Linus Torvalds6aab3412005-11-28 14:34:23 -08001516 page = vm_normal_page(vma, address, pte);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07001517 if (unlikely(!page)) {
1518 if ((flags & FOLL_DUMP) ||
Hugh Dickins62eede62009-09-21 17:03:34 -07001519 !is_zero_pfn(pte_pfn(pte)))
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07001520 goto bad_page;
1521 page = pte_page(pte);
1522 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001523
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001524 if (flags & FOLL_GET)
Andrea Arcangeli70b50f92011-11-02 13:36:59 -07001525 get_page_foll(page);
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001526 if (flags & FOLL_TOUCH) {
1527 if ((flags & FOLL_WRITE) &&
1528 !pte_dirty(pte) && !PageDirty(page))
1529 set_page_dirty(page);
KOSAKI Motohirobd775c42009-03-31 15:19:37 -07001530 /*
1531 * pte_mkyoung() would be more correct here, but atomic care
1532 * is needed to avoid losing the dirty bit: it is easier to use
1533 * mark_page_accessed().
1534 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001535 mark_page_accessed(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 }
Linus Torvaldsa1fde082011-05-04 21:30:28 -07001537 if ((flags & FOLL_MLOCK) && (vma->vm_flags & VM_LOCKED)) {
Michel Lespinasse110d74a2011-01-13 15:46:11 -08001538 /*
1539 * The preliminary mapping check is mainly to avoid the
1540 * pointless overhead of lock_page on the ZERO_PAGE
1541 * which might bounce very badly if there is contention.
1542 *
1543 * If the page is already locked, we don't need to
1544 * handle it now - vmscan will handle it later if and
1545 * when it attempts to reclaim the page.
1546 */
1547 if (page->mapping && trylock_page(page)) {
1548 lru_add_drain(); /* push cached pages to LRU */
1549 /*
1550 * Because we lock page here and migration is
1551 * blocked by the pte's page reference, we need
1552 * only check for file-cache page truncation.
1553 */
1554 if (page->mapping)
1555 mlock_vma_page(page);
1556 unlock_page(page);
1557 }
1558 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001559unlock:
1560 pte_unmap_unlock(ptep, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561out:
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001562 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001564bad_page:
1565 pte_unmap_unlock(ptep, ptl);
1566 return ERR_PTR(-EFAULT);
1567
1568no_page:
1569 pte_unmap_unlock(ptep, ptl);
1570 if (!pte_none(pte))
1571 return page;
Hugh Dickins8e4b9a62009-09-21 17:03:26 -07001572
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001573no_page_table:
1574 /*
1575 * When core dumping an enormous anonymous area that nobody
Hugh Dickins8e4b9a62009-09-21 17:03:26 -07001576 * has touched so far, we don't want to allocate unnecessary pages or
1577 * page tables. Return error instead of NULL to skip handle_mm_fault,
1578 * then get_dump_page() will return NULL to leave a hole in the dump.
1579 * But we can only make this optimization where a hole would surely
1580 * be zero-filled if handle_mm_fault() actually did handle it.
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001581 */
Hugh Dickins8e4b9a62009-09-21 17:03:26 -07001582 if ((flags & FOLL_DUMP) &&
1583 (!vma->vm_ops || !vma->vm_ops->fault))
1584 return ERR_PTR(-EFAULT);
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001585 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586}
1587
Linus Torvalds95042f92011-04-12 14:15:51 -07001588static inline int stack_guard_page(struct vm_area_struct *vma, unsigned long addr)
1589{
Mikulas Patockaa09a79f2011-05-09 13:01:09 +02001590 return stack_guard_page_start(vma, addr) ||
1591 stack_guard_page_end(vma, addr+PAGE_SIZE);
Linus Torvalds95042f92011-04-12 14:15:51 -07001592}
1593
Huang Ying0014bd92011-01-30 11:15:47 +08001594/**
1595 * __get_user_pages() - pin user pages in memory
1596 * @tsk: task_struct of target task
1597 * @mm: mm_struct of target mm
1598 * @start: starting user address
1599 * @nr_pages: number of pages from start to pin
1600 * @gup_flags: flags modifying pin behaviour
1601 * @pages: array that receives pointers to the pages pinned.
1602 * Should be at least nr_pages long. Or NULL, if caller
1603 * only intends to ensure the pages are faulted in.
1604 * @vmas: array of pointers to vmas corresponding to each page.
1605 * Or NULL if the caller does not require them.
1606 * @nonblocking: whether waiting for disk IO or mmap_sem contention
1607 *
1608 * Returns number of pages pinned. This may be fewer than the number
1609 * requested. If nr_pages is 0 or negative, returns 0. If no pages
1610 * were pinned, returns -errno. Each page returned must be released
1611 * with a put_page() call when it is finished with. vmas will only
1612 * remain valid while mmap_sem is held.
1613 *
1614 * Must be called with mmap_sem held for read or write.
1615 *
1616 * __get_user_pages walks a process's page tables and takes a reference to
1617 * each struct page that each user address corresponds to at a given
1618 * instant. That is, it takes the page that would be accessed if a user
1619 * thread accesses the given user virtual address at that instant.
1620 *
1621 * This does not guarantee that the page exists in the user mappings when
1622 * __get_user_pages returns, and there may even be a completely different
1623 * page there in some cases (eg. if mmapped pagecache has been invalidated
1624 * and subsequently re faulted). However it does guarantee that the page
1625 * won't be freed completely. And mostly callers simply care that the page
1626 * contains data that was valid *at some point in time*. Typically, an IO
1627 * or similar operation cannot guarantee anything stronger anyway because
1628 * locks can't be held over the syscall boundary.
1629 *
1630 * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If
1631 * the page is written to, set_page_dirty (or set_page_dirty_lock, as
1632 * appropriate) must be called after the page is finished with, and
1633 * before put_page is called.
1634 *
1635 * If @nonblocking != NULL, __get_user_pages will not wait for disk IO
1636 * or mmap_sem contention, and if waiting is needed to pin all pages,
1637 * *@nonblocking will be set to 0.
1638 *
1639 * In most cases, get_user_pages or get_user_pages_fast should be used
1640 * instead of __get_user_pages. __get_user_pages should be used only if
1641 * you need some special @gup_flags.
1642 */
Nick Pigginb291f002008-10-18 20:26:44 -07001643int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
Hugh Dickins58fa8792009-09-21 17:03:31 -07001644 unsigned long start, int nr_pages, unsigned int gup_flags,
Michel Lespinasse53a77062011-01-13 15:46:14 -08001645 struct page **pages, struct vm_area_struct **vmas,
1646 int *nonblocking)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647{
1648 int i;
Hugh Dickins58fa8792009-09-21 17:03:31 -07001649 unsigned long vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650
Peter Zijlstra9d737772009-06-25 11:58:55 +02001651 if (nr_pages <= 0)
Jonathan Corbet900cf082008-02-11 16:17:33 -07001652 return 0;
Hugh Dickins58fa8792009-09-21 17:03:31 -07001653
1654 VM_BUG_ON(!!pages != !!(gup_flags & FOLL_GET));
1655
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 /*
1657 * Require read or write permissions.
Hugh Dickins58fa8792009-09-21 17:03:31 -07001658 * If FOLL_FORCE is set, we only require the "MAY" flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 */
Hugh Dickins58fa8792009-09-21 17:03:31 -07001660 vm_flags = (gup_flags & FOLL_WRITE) ?
1661 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
1662 vm_flags &= (gup_flags & FOLL_FORCE) ?
1663 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 i = 0;
1665
1666 do {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001667 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668
1669 vma = find_extend_vma(mm, start);
Stephen Wilsone7f22e22011-03-13 15:49:18 -04001670 if (!vma && in_gate_area(mm, start)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 unsigned long pg = start & PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 pgd_t *pgd;
1673 pud_t *pud;
1674 pmd_t *pmd;
1675 pte_t *pte;
Nick Pigginb291f002008-10-18 20:26:44 -07001676
1677 /* user gate pages are read-only */
Hugh Dickins58fa8792009-09-21 17:03:31 -07001678 if (gup_flags & FOLL_WRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 return i ? : -EFAULT;
1680 if (pg > TASK_SIZE)
1681 pgd = pgd_offset_k(pg);
1682 else
1683 pgd = pgd_offset_gate(mm, pg);
1684 BUG_ON(pgd_none(*pgd));
1685 pud = pud_offset(pgd, pg);
1686 BUG_ON(pud_none(*pud));
1687 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001688 if (pmd_none(*pmd))
1689 return i ? : -EFAULT;
Andrea Arcangelif66055a2011-01-13 15:46:54 -08001690 VM_BUG_ON(pmd_trans_huge(*pmd));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001692 if (pte_none(*pte)) {
1693 pte_unmap(pte);
1694 return i ? : -EFAULT;
1695 }
Linus Torvalds95042f92011-04-12 14:15:51 -07001696 vma = get_gate_vma(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 if (pages) {
Hugh Dickinsde512572010-07-30 10:58:26 -07001698 struct page *page;
1699
Linus Torvalds95042f92011-04-12 14:15:51 -07001700 page = vm_normal_page(vma, start, *pte);
Hugh Dickinsde512572010-07-30 10:58:26 -07001701 if (!page) {
1702 if (!(gup_flags & FOLL_DUMP) &&
1703 is_zero_pfn(pte_pfn(*pte)))
1704 page = pte_page(*pte);
1705 else {
1706 pte_unmap(pte);
1707 return i ? : -EFAULT;
1708 }
1709 }
Linus Torvalds6aab3412005-11-28 14:34:23 -08001710 pages[i] = page;
Hugh Dickinsde512572010-07-30 10:58:26 -07001711 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 }
1713 pte_unmap(pte);
Linus Torvalds95042f92011-04-12 14:15:51 -07001714 goto next_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 }
1716
Nick Pigginb291f002008-10-18 20:26:44 -07001717 if (!vma ||
1718 (vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
Hugh Dickins1c3aff12009-09-21 17:03:24 -07001719 !(vm_flags & vma->vm_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 return i ? : -EFAULT;
1721
Hugh Dickins2a15efc2009-09-21 17:03:27 -07001722 if (is_vm_hugetlb_page(vma)) {
1723 i = follow_hugetlb_page(mm, vma, pages, vmas,
Hugh Dickins58fa8792009-09-21 17:03:31 -07001724 &start, &nr_pages, i, gup_flags);
Hugh Dickins2a15efc2009-09-21 17:03:27 -07001725 continue;
1726 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001727
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 do {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001729 struct page *page;
Hugh Dickins58fa8792009-09-21 17:03:31 -07001730 unsigned int foll_flags = gup_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731
Ethan Solomita462e00c2007-07-15 23:38:16 -07001732 /*
Ying Han47792802009-01-06 14:40:18 -08001733 * If we have a pending SIGKILL, don't keep faulting
Hugh Dickins1c3aff12009-09-21 17:03:24 -07001734 * pages and potentially allocating memory.
Ethan Solomita462e00c2007-07-15 23:38:16 -07001735 */
Hugh Dickins1c3aff12009-09-21 17:03:24 -07001736 if (unlikely(fatal_signal_pending(current)))
Ying Han47792802009-01-06 14:40:18 -08001737 return i ? i : -ERESTARTSYS;
Ethan Solomita462e00c2007-07-15 23:38:16 -07001738
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001739 cond_resched();
Linus Torvalds6aab3412005-11-28 14:34:23 -08001740 while (!(page = follow_page(vma, start, foll_flags))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001741 int ret;
Michel Lespinasse53a77062011-01-13 15:46:14 -08001742 unsigned int fault_flags = 0;
1743
Mikulas Patockaa09a79f2011-05-09 13:01:09 +02001744 /* For mlock, just skip the stack guard page. */
1745 if (foll_flags & FOLL_MLOCK) {
1746 if (stack_guard_page(vma, start))
1747 goto next_page;
1748 }
Michel Lespinasse53a77062011-01-13 15:46:14 -08001749 if (foll_flags & FOLL_WRITE)
1750 fault_flags |= FAULT_FLAG_WRITE;
1751 if (nonblocking)
1752 fault_flags |= FAULT_FLAG_ALLOW_RETRY;
Gleb Natapov318b2752011-03-22 16:30:51 -07001753 if (foll_flags & FOLL_NOWAIT)
1754 fault_flags |= (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT);
Linus Torvaldsd06063c2009-04-10 09:01:23 -07001755
Hugh Dickinsd26ed652009-06-23 13:52:49 +01001756 ret = handle_mm_fault(mm, vma, start,
Michel Lespinasse53a77062011-01-13 15:46:14 -08001757 fault_flags);
Hugh Dickinsd26ed652009-06-23 13:52:49 +01001758
Nick Piggin83c54072007-07-19 01:47:05 -07001759 if (ret & VM_FAULT_ERROR) {
1760 if (ret & VM_FAULT_OOM)
1761 return i ? i : -ENOMEM;
Huang Ying69ebb832011-01-30 11:15:48 +08001762 if (ret & (VM_FAULT_HWPOISON |
1763 VM_FAULT_HWPOISON_LARGE)) {
1764 if (i)
1765 return i;
1766 else if (gup_flags & FOLL_HWPOISON)
1767 return -EHWPOISON;
1768 else
1769 return -EFAULT;
1770 }
1771 if (ret & VM_FAULT_SIGBUS)
Nick Piggin83c54072007-07-19 01:47:05 -07001772 return i ? i : -EFAULT;
1773 BUG();
1774 }
Stephen Wilsone7f22e22011-03-13 15:49:18 -04001775
1776 if (tsk) {
1777 if (ret & VM_FAULT_MAJOR)
1778 tsk->maj_flt++;
1779 else
1780 tsk->min_flt++;
1781 }
Nick Piggin83c54072007-07-19 01:47:05 -07001782
Michel Lespinasse53a77062011-01-13 15:46:14 -08001783 if (ret & VM_FAULT_RETRY) {
Gleb Natapov318b2752011-03-22 16:30:51 -07001784 if (nonblocking)
1785 *nonblocking = 0;
Michel Lespinasse53a77062011-01-13 15:46:14 -08001786 return i;
1787 }
1788
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001789 /*
Nick Piggin83c54072007-07-19 01:47:05 -07001790 * The VM_FAULT_WRITE bit tells us that
1791 * do_wp_page has broken COW when necessary,
1792 * even if maybe_mkwrite decided not to set
1793 * pte_write. We can thus safely do subsequent
Hugh Dickins878b63a2009-01-06 14:39:32 -08001794 * page lookups as if they were reads. But only
1795 * do so when looping for pte_write is futile:
1796 * in some cases userspace may also be wanting
1797 * to write to the gotten user page, which a
1798 * read fault here might prevent (a readonly
1799 * page might get reCOWed by userspace write).
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001800 */
Hugh Dickins878b63a2009-01-06 14:39:32 -08001801 if ((ret & VM_FAULT_WRITE) &&
1802 !(vma->vm_flags & VM_WRITE))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001803 foll_flags &= ~FOLL_WRITE;
Nick Piggin83c54072007-07-19 01:47:05 -07001804
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07001805 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 }
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001807 if (IS_ERR(page))
1808 return i ? i : PTR_ERR(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001810 pages[i] = page;
James Bottomley03beb072006-03-26 01:36:57 -08001811
Russell Kinga6f36be2006-12-30 22:24:19 +00001812 flush_anon_page(vma, page, start);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001813 flush_dcache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 }
Linus Torvalds95042f92011-04-12 14:15:51 -07001815next_page:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 if (vmas)
1817 vmas[i] = vma;
1818 i++;
1819 start += PAGE_SIZE;
Peter Zijlstra9d737772009-06-25 11:58:55 +02001820 nr_pages--;
1821 } while (nr_pages && start < vma->vm_end);
1822 } while (nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 return i;
1824}
Huang Ying0014bd92011-01-30 11:15:47 +08001825EXPORT_SYMBOL(__get_user_pages);
Nick Pigginb291f002008-10-18 20:26:44 -07001826
Benjamin Herrenschmidt2efaca92011-07-25 17:12:32 -07001827/*
1828 * fixup_user_fault() - manually resolve a user page fault
1829 * @tsk: the task_struct to use for page fault accounting, or
1830 * NULL if faults are not to be recorded.
1831 * @mm: mm_struct of target mm
1832 * @address: user address
1833 * @fault_flags:flags to pass down to handle_mm_fault()
1834 *
1835 * This is meant to be called in the specific scenario where for locking reasons
1836 * we try to access user memory in atomic context (within a pagefault_disable()
1837 * section), this returns -EFAULT, and we want to resolve the user fault before
1838 * trying again.
1839 *
1840 * Typically this is meant to be used by the futex code.
1841 *
1842 * The main difference with get_user_pages() is that this function will
1843 * unconditionally call handle_mm_fault() which will in turn perform all the
1844 * necessary SW fixup of the dirty and young bits in the PTE, while
1845 * handle_mm_fault() only guarantees to update these in the struct page.
1846 *
1847 * This is important for some architectures where those bits also gate the
1848 * access permission to the page because they are maintained in software. On
1849 * such architectures, gup() will not be enough to make a subsequent access
1850 * succeed.
1851 *
1852 * This should be called with the mm_sem held for read.
1853 */
1854int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm,
1855 unsigned long address, unsigned int fault_flags)
1856{
1857 struct vm_area_struct *vma;
1858 int ret;
1859
1860 vma = find_extend_vma(mm, address);
1861 if (!vma || address < vma->vm_start)
1862 return -EFAULT;
1863
1864 ret = handle_mm_fault(mm, vma, address, fault_flags);
1865 if (ret & VM_FAULT_ERROR) {
1866 if (ret & VM_FAULT_OOM)
1867 return -ENOMEM;
1868 if (ret & (VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE))
1869 return -EHWPOISON;
1870 if (ret & VM_FAULT_SIGBUS)
1871 return -EFAULT;
1872 BUG();
1873 }
1874 if (tsk) {
1875 if (ret & VM_FAULT_MAJOR)
1876 tsk->maj_flt++;
1877 else
1878 tsk->min_flt++;
1879 }
1880 return 0;
1881}
1882
1883/*
Nick Piggind2bf6be2009-06-16 15:31:39 -07001884 * get_user_pages() - pin user pages in memory
Stephen Wilsone7f22e22011-03-13 15:49:18 -04001885 * @tsk: the task_struct to use for page fault accounting, or
1886 * NULL if faults are not to be recorded.
Nick Piggind2bf6be2009-06-16 15:31:39 -07001887 * @mm: mm_struct of target mm
1888 * @start: starting user address
Peter Zijlstra9d737772009-06-25 11:58:55 +02001889 * @nr_pages: number of pages from start to pin
Nick Piggind2bf6be2009-06-16 15:31:39 -07001890 * @write: whether pages will be written to by the caller
1891 * @force: whether to force write access even if user mapping is
1892 * readonly. This will result in the page being COWed even
1893 * in MAP_SHARED mappings. You do not want this.
1894 * @pages: array that receives pointers to the pages pinned.
1895 * Should be at least nr_pages long. Or NULL, if caller
1896 * only intends to ensure the pages are faulted in.
1897 * @vmas: array of pointers to vmas corresponding to each page.
1898 * Or NULL if the caller does not require them.
1899 *
1900 * Returns number of pages pinned. This may be fewer than the number
Peter Zijlstra9d737772009-06-25 11:58:55 +02001901 * requested. If nr_pages is 0 or negative, returns 0. If no pages
Nick Piggind2bf6be2009-06-16 15:31:39 -07001902 * were pinned, returns -errno. Each page returned must be released
1903 * with a put_page() call when it is finished with. vmas will only
1904 * remain valid while mmap_sem is held.
1905 *
1906 * Must be called with mmap_sem held for read or write.
1907 *
1908 * get_user_pages walks a process's page tables and takes a reference to
1909 * each struct page that each user address corresponds to at a given
1910 * instant. That is, it takes the page that would be accessed if a user
1911 * thread accesses the given user virtual address at that instant.
1912 *
1913 * This does not guarantee that the page exists in the user mappings when
1914 * get_user_pages returns, and there may even be a completely different
1915 * page there in some cases (eg. if mmapped pagecache has been invalidated
1916 * and subsequently re faulted). However it does guarantee that the page
1917 * won't be freed completely. And mostly callers simply care that the page
1918 * contains data that was valid *at some point in time*. Typically, an IO
1919 * or similar operation cannot guarantee anything stronger anyway because
1920 * locks can't be held over the syscall boundary.
1921 *
1922 * If write=0, the page must not be written to. If the page is written to,
1923 * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called
1924 * after the page is finished with, and before put_page is called.
1925 *
1926 * get_user_pages is typically used for fewer-copy IO operations, to get a
1927 * handle on the memory by some means other than accesses via the user virtual
1928 * addresses. The pages may be submitted for DMA to devices or accessed via
1929 * their kernel linear mapping (via the kmap APIs). Care should be taken to
1930 * use the correct cache flushing APIs.
1931 *
1932 * See also get_user_pages_fast, for performance critical applications.
1933 */
Nick Pigginb291f002008-10-18 20:26:44 -07001934int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
Peter Zijlstra9d737772009-06-25 11:58:55 +02001935 unsigned long start, int nr_pages, int write, int force,
Nick Pigginb291f002008-10-18 20:26:44 -07001936 struct page **pages, struct vm_area_struct **vmas)
1937{
Hugh Dickins58fa8792009-09-21 17:03:31 -07001938 int flags = FOLL_TOUCH;
Nick Pigginb291f002008-10-18 20:26:44 -07001939
Hugh Dickins58fa8792009-09-21 17:03:31 -07001940 if (pages)
1941 flags |= FOLL_GET;
Nick Pigginb291f002008-10-18 20:26:44 -07001942 if (write)
Hugh Dickins58fa8792009-09-21 17:03:31 -07001943 flags |= FOLL_WRITE;
Nick Pigginb291f002008-10-18 20:26:44 -07001944 if (force)
Hugh Dickins58fa8792009-09-21 17:03:31 -07001945 flags |= FOLL_FORCE;
Nick Pigginb291f002008-10-18 20:26:44 -07001946
Michel Lespinasse53a77062011-01-13 15:46:14 -08001947 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
1948 NULL);
Nick Pigginb291f002008-10-18 20:26:44 -07001949}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950EXPORT_SYMBOL(get_user_pages);
1951
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07001952/**
1953 * get_dump_page() - pin user page in memory while writing it to core dump
1954 * @addr: user address
1955 *
1956 * Returns struct page pointer of user page pinned for dump,
1957 * to be freed afterwards by page_cache_release() or put_page().
1958 *
1959 * Returns NULL on any kind of failure - a hole must then be inserted into
1960 * the corefile, to preserve alignment with its headers; and also returns
1961 * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found -
1962 * allowing a hole to be left in the corefile to save diskspace.
1963 *
1964 * Called without mmap_sem, but after all other threads have been killed.
1965 */
1966#ifdef CONFIG_ELF_CORE
1967struct page *get_dump_page(unsigned long addr)
1968{
1969 struct vm_area_struct *vma;
1970 struct page *page;
1971
1972 if (__get_user_pages(current, current->mm, addr, 1,
Michel Lespinasse53a77062011-01-13 15:46:14 -08001973 FOLL_FORCE | FOLL_DUMP | FOLL_GET, &page, &vma,
1974 NULL) < 1)
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07001975 return NULL;
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07001976 flush_cache_page(vma, addr, page_to_pfn(page));
1977 return page;
1978}
1979#endif /* CONFIG_ELF_CORE */
1980
Namhyung Kim25ca1d62010-10-26 14:21:59 -07001981pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr,
Harvey Harrison920c7a52008-02-04 22:29:26 -08001982 spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001983{
1984 pgd_t * pgd = pgd_offset(mm, addr);
1985 pud_t * pud = pud_alloc(mm, pgd, addr);
1986 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001987 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Andrea Arcangelif66055a2011-01-13 15:46:54 -08001988 if (pmd) {
1989 VM_BUG_ON(pmd_trans_huge(*pmd));
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001990 return pte_alloc_map_lock(mm, pmd, addr, ptl);
Andrea Arcangelif66055a2011-01-13 15:46:54 -08001991 }
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001992 }
1993 return NULL;
1994}
1995
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001997 * This is the old fallback for page remapping.
1998 *
1999 * For historical reasons, it only allows reserved pages. Only
2000 * old drivers should use this, and they needed to mark their
2001 * pages reserved for the old functions anyway.
2002 */
Nick Piggin423bad62008-04-28 02:13:01 -07002003static int insert_page(struct vm_area_struct *vma, unsigned long addr,
2004 struct page *page, pgprot_t prot)
Linus Torvalds238f58d2005-11-29 13:01:56 -08002005{
Nick Piggin423bad62008-04-28 02:13:01 -07002006 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds238f58d2005-11-29 13:01:56 -08002007 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08002008 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002009 spinlock_t *ptl;
2010
Linus Torvalds238f58d2005-11-29 13:01:56 -08002011 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08002012 if (PageAnon(page))
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002013 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08002014 retval = -ENOMEM;
2015 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08002016 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08002017 if (!pte)
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002018 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08002019 retval = -EBUSY;
2020 if (!pte_none(*pte))
2021 goto out_unlock;
2022
2023 /* Ok, finally just insert the thing.. */
2024 get_page(page);
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08002025 inc_mm_counter_fast(mm, MM_FILEPAGES);
Linus Torvalds238f58d2005-11-29 13:01:56 -08002026 page_add_file_rmap(page);
2027 set_pte_at(mm, addr, pte, mk_pte(page, prot));
2028
2029 retval = 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002030 pte_unmap_unlock(pte, ptl);
2031 return retval;
Linus Torvalds238f58d2005-11-29 13:01:56 -08002032out_unlock:
2033 pte_unmap_unlock(pte, ptl);
2034out:
2035 return retval;
2036}
2037
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07002038/**
2039 * vm_insert_page - insert single page into user vma
2040 * @vma: user vma to map to
2041 * @addr: target user address of this page
2042 * @page: source kernel page
2043 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08002044 * This allows drivers to insert individual pages they've allocated
2045 * into a user vma.
2046 *
2047 * The page has to be a nice clean _individual_ kernel allocation.
2048 * If you allocate a compound page, you need to have marked it as
2049 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08002050 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08002051 *
2052 * NOTE! Traditionally this was done with "remap_pfn_range()" which
2053 * took an arbitrary page protection parameter. This doesn't allow
2054 * that. Your vma protection will have to be set up correctly, which
2055 * means that if you want a shared writable mapping, you'd better
2056 * ask for a shared writable mapping!
2057 *
2058 * The page does not need to be reserved.
2059 */
Nick Piggin423bad62008-04-28 02:13:01 -07002060int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
2061 struct page *page)
Linus Torvaldsa145dd42005-11-30 09:35:19 -08002062{
2063 if (addr < vma->vm_start || addr >= vma->vm_end)
2064 return -EFAULT;
2065 if (!page_count(page))
2066 return -EINVAL;
Linus Torvalds4d7672b2005-12-16 10:21:23 -08002067 vma->vm_flags |= VM_INSERTPAGE;
Nick Piggin423bad62008-04-28 02:13:01 -07002068 return insert_page(vma, addr, page, vma->vm_page_prot);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08002069}
Linus Torvaldse3c33742005-12-03 20:48:11 -08002070EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08002071
Nick Piggin423bad62008-04-28 02:13:01 -07002072static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
2073 unsigned long pfn, pgprot_t prot)
2074{
2075 struct mm_struct *mm = vma->vm_mm;
2076 int retval;
2077 pte_t *pte, entry;
2078 spinlock_t *ptl;
2079
2080 retval = -ENOMEM;
2081 pte = get_locked_pte(mm, addr, &ptl);
2082 if (!pte)
2083 goto out;
2084 retval = -EBUSY;
2085 if (!pte_none(*pte))
2086 goto out_unlock;
2087
2088 /* Ok, finally just insert the thing.. */
2089 entry = pte_mkspecial(pfn_pte(pfn, prot));
2090 set_pte_at(mm, addr, pte, entry);
Russell King4b3073e2009-12-18 16:40:18 +00002091 update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */
Nick Piggin423bad62008-04-28 02:13:01 -07002092
2093 retval = 0;
2094out_unlock:
2095 pte_unmap_unlock(pte, ptl);
2096out:
2097 return retval;
2098}
2099
Nick Piggine0dc0d82007-02-12 00:51:36 -08002100/**
2101 * vm_insert_pfn - insert single pfn into user vma
2102 * @vma: user vma to map to
2103 * @addr: target user address of this page
2104 * @pfn: source kernel pfn
2105 *
2106 * Similar to vm_inert_page, this allows drivers to insert individual pages
2107 * they've allocated into a user vma. Same comments apply.
2108 *
2109 * This function should only be called from a vm_ops->fault handler, and
2110 * in that case the handler should return NULL.
Nick Piggin0d71d10a2008-07-23 21:27:05 -07002111 *
2112 * vma cannot be a COW mapping.
2113 *
2114 * As this is called only for pages that do not currently exist, we
2115 * do not need to flush old virtual caches or the TLB.
Nick Piggine0dc0d82007-02-12 00:51:36 -08002116 */
2117int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
Nick Piggin423bad62008-04-28 02:13:01 -07002118 unsigned long pfn)
Nick Piggine0dc0d82007-02-12 00:51:36 -08002119{
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08002120 int ret;
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08002121 pgprot_t pgprot = vma->vm_page_prot;
Nick Piggin7e675132008-04-28 02:13:00 -07002122 /*
2123 * Technically, architectures with pte_special can avoid all these
2124 * restrictions (same for remap_pfn_range). However we would like
2125 * consistency in testing and feature parity among all, so we should
2126 * try to keep these invariants in place for everybody.
2127 */
Jared Hulbertb379d792008-04-28 02:12:58 -07002128 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
2129 BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
2130 (VM_PFNMAP|VM_MIXEDMAP));
2131 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
2132 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
Nick Piggine0dc0d82007-02-12 00:51:36 -08002133
Nick Piggin423bad62008-04-28 02:13:01 -07002134 if (addr < vma->vm_start || addr >= vma->vm_end)
2135 return -EFAULT;
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08002136 if (track_pfn_vma_new(vma, &pgprot, pfn, PAGE_SIZE))
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08002137 return -EINVAL;
2138
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08002139 ret = insert_pfn(vma, addr, pfn, pgprot);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08002140
2141 if (ret)
2142 untrack_pfn_vma(vma, pfn, PAGE_SIZE);
2143
2144 return ret;
Nick Piggine0dc0d82007-02-12 00:51:36 -08002145}
2146EXPORT_SYMBOL(vm_insert_pfn);
2147
Nick Piggin423bad62008-04-28 02:13:01 -07002148int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
2149 unsigned long pfn)
2150{
2151 BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
2152
2153 if (addr < vma->vm_start || addr >= vma->vm_end)
2154 return -EFAULT;
2155
2156 /*
2157 * If we don't have pte special, then we have to use the pfn_valid()
2158 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
2159 * refcount the page if pfn_valid is true (hence insert_page rather
Hugh Dickins62eede62009-09-21 17:03:34 -07002160 * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP
2161 * without pte special, it would there be refcounted as a normal page.
Nick Piggin423bad62008-04-28 02:13:01 -07002162 */
2163 if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) {
2164 struct page *page;
2165
2166 page = pfn_to_page(pfn);
2167 return insert_page(vma, addr, page, vma->vm_page_prot);
2168 }
2169 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
2170}
2171EXPORT_SYMBOL(vm_insert_mixed);
2172
Linus Torvaldsa145dd42005-11-30 09:35:19 -08002173/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 * maps a range of physical memory into the requested pages. the old
2175 * mappings are removed. any references to nonexistent pages results
2176 * in null mappings (currently treated as "copy-on-access")
2177 */
2178static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
2179 unsigned long addr, unsigned long end,
2180 unsigned long pfn, pgprot_t prot)
2181{
2182 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07002183 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184
Hugh Dickinsc74df322005-10-29 18:16:23 -07002185 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 if (!pte)
2187 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07002188 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 do {
2190 BUG_ON(!pte_none(*pte));
Nick Piggin7e675132008-04-28 02:13:00 -07002191 set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 pfn++;
2193 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07002194 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07002195 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 return 0;
2197}
2198
2199static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
2200 unsigned long addr, unsigned long end,
2201 unsigned long pfn, pgprot_t prot)
2202{
2203 pmd_t *pmd;
2204 unsigned long next;
2205
2206 pfn -= addr >> PAGE_SHIFT;
2207 pmd = pmd_alloc(mm, pud, addr);
2208 if (!pmd)
2209 return -ENOMEM;
Andrea Arcangelif66055a2011-01-13 15:46:54 -08002210 VM_BUG_ON(pmd_trans_huge(*pmd));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 do {
2212 next = pmd_addr_end(addr, end);
2213 if (remap_pte_range(mm, pmd, addr, next,
2214 pfn + (addr >> PAGE_SHIFT), prot))
2215 return -ENOMEM;
2216 } while (pmd++, addr = next, addr != end);
2217 return 0;
2218}
2219
2220static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
2221 unsigned long addr, unsigned long end,
2222 unsigned long pfn, pgprot_t prot)
2223{
2224 pud_t *pud;
2225 unsigned long next;
2226
2227 pfn -= addr >> PAGE_SHIFT;
2228 pud = pud_alloc(mm, pgd, addr);
2229 if (!pud)
2230 return -ENOMEM;
2231 do {
2232 next = pud_addr_end(addr, end);
2233 if (remap_pmd_range(mm, pud, addr, next,
2234 pfn + (addr >> PAGE_SHIFT), prot))
2235 return -ENOMEM;
2236 } while (pud++, addr = next, addr != end);
2237 return 0;
2238}
2239
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07002240/**
2241 * remap_pfn_range - remap kernel memory to userspace
2242 * @vma: user vma to map to
2243 * @addr: target user address to start at
2244 * @pfn: physical address of kernel memory
2245 * @size: size of map area
2246 * @prot: page protection flags for this mapping
2247 *
2248 * Note: this is only safe if the mm semaphore is held when called.
2249 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
2251 unsigned long pfn, unsigned long size, pgprot_t prot)
2252{
2253 pgd_t *pgd;
2254 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07002255 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 struct mm_struct *mm = vma->vm_mm;
2257 int err;
2258
2259 /*
2260 * Physically remapped pages are special. Tell the
2261 * rest of the world about it:
2262 * VM_IO tells people not to look at these pages
2263 * (accesses can have side effects).
Hugh Dickins0b14c172005-11-21 21:32:15 -08002264 * VM_RESERVED is specified all over the place, because
2265 * in 2.4 it kept swapout's vma scan off this vma; but
2266 * in 2.6 the LRU scan won't even find its pages, so this
2267 * flag means no more than count its pages in reserved_vm,
2268 * and omit it from core dump, even when VM_IO turned off.
Linus Torvalds6aab3412005-11-28 14:34:23 -08002269 * VM_PFNMAP tells the core MM that the base pages are just
2270 * raw PFN mappings, and do not have a "struct page" associated
2271 * with them.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08002272 *
2273 * There's a horrible special case to handle copy-on-write
2274 * behaviour that some programs depend on. We mark the "original"
2275 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 */
Pallipadi, Venkatesh4bb9c5c2009-03-12 17:45:27 -07002277 if (addr == vma->vm_start && end == vma->vm_end) {
Linus Torvaldsfb155c12005-12-11 19:46:02 -08002278 vma->vm_pgoff = pfn;
Pallipadi, Venkatesh895791d2009-03-13 16:35:44 -07002279 vma->vm_flags |= VM_PFN_AT_MMAP;
Pallipadi, Venkatesh4bb9c5c2009-03-12 17:45:27 -07002280 } else if (is_cow_mapping(vma->vm_flags))
venkatesh.pallipadi@intel.com3c8bb732008-12-18 11:41:27 -08002281 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08002282
Linus Torvalds6aab3412005-11-28 14:34:23 -08002283 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08002285 err = track_pfn_vma_new(vma, &prot, pfn, PAGE_ALIGN(size));
venkatesh.pallipadi@intel.coma3670612009-01-09 16:13:09 -08002286 if (err) {
2287 /*
2288 * To indicate that track_pfn related cleanup is not
2289 * needed from higher level routine calling unmap_vmas
2290 */
2291 vma->vm_flags &= ~(VM_IO | VM_RESERVED | VM_PFNMAP);
Pallipadi, Venkatesh895791d2009-03-13 16:35:44 -07002292 vma->vm_flags &= ~VM_PFN_AT_MMAP;
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08002293 return -EINVAL;
venkatesh.pallipadi@intel.coma3670612009-01-09 16:13:09 -08002294 }
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08002295
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 BUG_ON(addr >= end);
2297 pfn -= addr >> PAGE_SHIFT;
2298 pgd = pgd_offset(mm, addr);
2299 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 do {
2301 next = pgd_addr_end(addr, end);
2302 err = remap_pud_range(mm, pgd, addr, next,
2303 pfn + (addr >> PAGE_SHIFT), prot);
2304 if (err)
2305 break;
2306 } while (pgd++, addr = next, addr != end);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08002307
2308 if (err)
2309 untrack_pfn_vma(vma, pfn, PAGE_ALIGN(size));
2310
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 return err;
2312}
2313EXPORT_SYMBOL(remap_pfn_range);
2314
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002315static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
2316 unsigned long addr, unsigned long end,
2317 pte_fn_t fn, void *data)
2318{
2319 pte_t *pte;
2320 int err;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08002321 pgtable_t token;
Borislav Petkov94909912007-05-06 14:49:17 -07002322 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002323
2324 pte = (mm == &init_mm) ?
2325 pte_alloc_kernel(pmd, addr) :
2326 pte_alloc_map_lock(mm, pmd, addr, &ptl);
2327 if (!pte)
2328 return -ENOMEM;
2329
2330 BUG_ON(pmd_huge(*pmd));
2331
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08002332 arch_enter_lazy_mmu_mode();
2333
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08002334 token = pmd_pgtable(*pmd);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002335
2336 do {
Daisuke Nishimurac36987e2009-10-26 16:50:23 -07002337 err = fn(pte++, token, addr, data);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002338 if (err)
2339 break;
Daisuke Nishimurac36987e2009-10-26 16:50:23 -07002340 } while (addr += PAGE_SIZE, addr != end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002341
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08002342 arch_leave_lazy_mmu_mode();
2343
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002344 if (mm != &init_mm)
2345 pte_unmap_unlock(pte-1, ptl);
2346 return err;
2347}
2348
2349static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
2350 unsigned long addr, unsigned long end,
2351 pte_fn_t fn, void *data)
2352{
2353 pmd_t *pmd;
2354 unsigned long next;
2355 int err;
2356
Andi Kleenceb86872008-07-23 21:27:50 -07002357 BUG_ON(pud_huge(*pud));
2358
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002359 pmd = pmd_alloc(mm, pud, addr);
2360 if (!pmd)
2361 return -ENOMEM;
2362 do {
2363 next = pmd_addr_end(addr, end);
2364 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
2365 if (err)
2366 break;
2367 } while (pmd++, addr = next, addr != end);
2368 return err;
2369}
2370
2371static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
2372 unsigned long addr, unsigned long end,
2373 pte_fn_t fn, void *data)
2374{
2375 pud_t *pud;
2376 unsigned long next;
2377 int err;
2378
2379 pud = pud_alloc(mm, pgd, addr);
2380 if (!pud)
2381 return -ENOMEM;
2382 do {
2383 next = pud_addr_end(addr, end);
2384 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
2385 if (err)
2386 break;
2387 } while (pud++, addr = next, addr != end);
2388 return err;
2389}
2390
2391/*
2392 * Scan a region of virtual memory, filling in page tables as necessary
2393 * and calling a provided function on each leaf page table.
2394 */
2395int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
2396 unsigned long size, pte_fn_t fn, void *data)
2397{
2398 pgd_t *pgd;
2399 unsigned long next;
Jeremy Fitzhardinge57250a52010-08-09 17:19:52 -07002400 unsigned long end = addr + size;
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002401 int err;
2402
2403 BUG_ON(addr >= end);
2404 pgd = pgd_offset(mm, addr);
2405 do {
2406 next = pgd_addr_end(addr, end);
2407 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
2408 if (err)
2409 break;
2410 } while (pgd++, addr = next, addr != end);
Jeremy Fitzhardinge57250a52010-08-09 17:19:52 -07002411
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002412 return err;
2413}
2414EXPORT_SYMBOL_GPL(apply_to_page_range);
2415
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002417 * handle_pte_fault chooses page fault handler according to an entry
2418 * which was read non-atomically. Before making any commitment, on
2419 * those architectures or configurations (e.g. i386 with PAE) which
Ryota Ozakia335b2e2011-02-10 13:56:28 +09002420 * might give a mix of unmatched parts, do_swap_page and do_nonlinear_fault
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002421 * must check under lock before unmapping the pte and proceeding
2422 * (but do_wp_page is only called after already making such a check;
Ryota Ozakia335b2e2011-02-10 13:56:28 +09002423 * and do_anonymous_page can safely check later on).
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002424 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002425static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002426 pte_t *page_table, pte_t orig_pte)
2427{
2428 int same = 1;
2429#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
2430 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002431 spinlock_t *ptl = pte_lockptr(mm, pmd);
2432 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002433 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002434 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002435 }
2436#endif
2437 pte_unmap(page_table);
2438 return same;
2439}
2440
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00002441static 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 -08002442{
2443 /*
2444 * If the source page was a PFN mapping, we don't have
2445 * a "struct page" for it. We do a best-effort copy by
2446 * just copying from the original user address. If that
2447 * fails, we just zero-fill it. Live with it.
2448 */
2449 if (unlikely(!src)) {
2450 void *kaddr = kmap_atomic(dst, KM_USER0);
Linus Torvalds5d2a2db2005-11-29 14:07:55 -08002451 void __user *uaddr = (void __user *)(va & PAGE_MASK);
2452
2453 /*
2454 * This really shouldn't fail, because the page is there
2455 * in the page tables. But it might just be unreadable,
2456 * in which case we just give up and fill the result with
2457 * zeroes.
2458 */
2459 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Jan Beulich3ecb01d2010-10-26 14:22:27 -07002460 clear_page(kaddr);
Linus Torvalds6aab3412005-11-28 14:34:23 -08002461 kunmap_atomic(kaddr, KM_USER0);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07002462 flush_dcache_page(dst);
Nick Piggin0ed361d2008-02-04 22:29:34 -08002463 } else
2464 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08002465}
2466
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468 * This routine handles present pages, when users try to write
2469 * to a shared page. It is done by copying the page to a new address
2470 * and decrementing the shared-page counter for the old page.
2471 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 * Note that this routine assumes that the protection checks have been
2473 * done by the caller (the low-level page fault routine in most cases).
2474 * Thus we can safely just mark it writable once we've done any necessary
2475 * COW.
2476 *
2477 * We also mark the page dirty at this point even though the page will
2478 * change only once the write actually happens. This avoids a few races,
2479 * and potentially makes it more efficient.
2480 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002481 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2482 * but allow concurrent faults), with pte both mapped and locked.
2483 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002485static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
2486 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002487 spinlock_t *ptl, pte_t orig_pte)
Namhyung Kime6219ec2010-10-26 14:22:00 -07002488 __releases(ptl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489{
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00002490 struct page *old_page, *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 pte_t entry;
Michel Lespinasseb009c022011-01-13 15:46:07 -08002492 int ret = 0;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002493 int page_mkwrite = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002494 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495
Linus Torvalds6aab3412005-11-28 14:34:23 -08002496 old_page = vm_normal_page(vma, address, orig_pte);
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002497 if (!old_page) {
2498 /*
2499 * VM_MIXEDMAP !pfn_valid() case
2500 *
2501 * We should not cow pages in a shared writeable mapping.
2502 * Just mark the pages writable as we can't do any dirty
2503 * accounting on raw pfn maps.
2504 */
2505 if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
2506 (VM_WRITE|VM_SHARED))
2507 goto reuse;
Linus Torvalds6aab3412005-11-28 14:34:23 -08002508 goto gotten;
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002509 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002511 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002512 * Take out anonymous pages first, anonymous shared vmas are
2513 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002514 */
Hugh Dickins9a840892009-09-21 17:02:01 -07002515 if (PageAnon(old_page) && !PageKsm(old_page)) {
Hugh Dickinsab967d82009-01-06 14:39:33 -08002516 if (!trylock_page(old_page)) {
2517 page_cache_get(old_page);
2518 pte_unmap_unlock(page_table, ptl);
2519 lock_page(old_page);
2520 page_table = pte_offset_map_lock(mm, pmd, address,
2521 &ptl);
2522 if (!pte_same(*page_table, orig_pte)) {
2523 unlock_page(old_page);
Hugh Dickinsab967d82009-01-06 14:39:33 -08002524 goto unlock;
2525 }
2526 page_cache_release(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002527 }
Michel Lespinasseb009c022011-01-13 15:46:07 -08002528 if (reuse_swap_page(old_page)) {
Rik van Rielc44b6742010-03-05 13:42:09 -08002529 /*
2530 * The page is all ours. Move it to our anon_vma so
2531 * the rmap code will not search our parent or siblings.
2532 * Protected against the rmap code by the page lock.
2533 */
2534 page_move_anon_rmap(old_page, vma, address);
Michel Lespinasseb009c022011-01-13 15:46:07 -08002535 unlock_page(old_page);
2536 goto reuse;
2537 }
Hugh Dickinsab967d82009-01-06 14:39:33 -08002538 unlock_page(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002539 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002540 (VM_WRITE|VM_SHARED))) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002541 /*
2542 * Only catch write-faults on shared writable pages,
2543 * read-only shared pages can get COWed by
2544 * get_user_pages(.write=1, .force=1).
2545 */
David Howells9637a5e2006-06-23 02:03:43 -07002546 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
Nick Pigginc2ec1752009-03-31 15:23:21 -07002547 struct vm_fault vmf;
2548 int tmp;
2549
2550 vmf.virtual_address = (void __user *)(address &
2551 PAGE_MASK);
2552 vmf.pgoff = old_page->index;
2553 vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
2554 vmf.page = old_page;
2555
David Howells9637a5e2006-06-23 02:03:43 -07002556 /*
2557 * Notify the address space that the page is about to
2558 * become writable so that it can prohibit this or wait
2559 * for the page to get into an appropriate state.
2560 *
2561 * We do this without the lock held, so that it can
2562 * sleep if it needs to.
2563 */
2564 page_cache_get(old_page);
2565 pte_unmap_unlock(page_table, ptl);
2566
Nick Pigginc2ec1752009-03-31 15:23:21 -07002567 tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
2568 if (unlikely(tmp &
2569 (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
2570 ret = tmp;
David Howells9637a5e2006-06-23 02:03:43 -07002571 goto unwritable_page;
Nick Pigginc2ec1752009-03-31 15:23:21 -07002572 }
Nick Pigginb827e492009-04-30 15:08:16 -07002573 if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
2574 lock_page(old_page);
2575 if (!old_page->mapping) {
2576 ret = 0; /* retry the fault */
2577 unlock_page(old_page);
2578 goto unwritable_page;
2579 }
2580 } else
2581 VM_BUG_ON(!PageLocked(old_page));
David Howells9637a5e2006-06-23 02:03:43 -07002582
David Howells9637a5e2006-06-23 02:03:43 -07002583 /*
2584 * Since we dropped the lock we need to revalidate
2585 * the PTE as someone else may have changed it. If
2586 * they did, we just return, as we can count on the
2587 * MMU to tell us if they didn't also make it writable.
2588 */
2589 page_table = pte_offset_map_lock(mm, pmd, address,
2590 &ptl);
Nick Pigginb827e492009-04-30 15:08:16 -07002591 if (!pte_same(*page_table, orig_pte)) {
2592 unlock_page(old_page);
David Howells9637a5e2006-06-23 02:03:43 -07002593 goto unlock;
Nick Pigginb827e492009-04-30 15:08:16 -07002594 }
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002595
2596 page_mkwrite = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002598 dirty_page = old_page;
2599 get_page(dirty_page);
David Howells9637a5e2006-06-23 02:03:43 -07002600
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002601reuse:
David Howells9637a5e2006-06-23 02:03:43 -07002602 flush_cache_page(vma, address, pte_pfn(orig_pte));
2603 entry = pte_mkyoung(orig_pte);
2604 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
KAMEZAWA Hiroyuki954ffcb2007-10-16 01:25:44 -07002605 if (ptep_set_access_flags(vma, address, page_table, entry,1))
Russell King4b3073e2009-12-18 16:40:18 +00002606 update_mmu_cache(vma, address, page_table);
Michel Lespinasse72ddc8f2011-01-13 15:46:08 -08002607 pte_unmap_unlock(page_table, ptl);
David Howells9637a5e2006-06-23 02:03:43 -07002608 ret |= VM_FAULT_WRITE;
Michel Lespinasse72ddc8f2011-01-13 15:46:08 -08002609
2610 if (!dirty_page)
2611 return ret;
2612
2613 /*
2614 * Yes, Virginia, this is actually required to prevent a race
2615 * with clear_page_dirty_for_io() from clearing the page dirty
2616 * bit after it clear all dirty ptes, but before a racing
2617 * do_wp_page installs a dirty pte.
2618 *
Ryota Ozakia335b2e2011-02-10 13:56:28 +09002619 * __do_fault is protected similarly.
Michel Lespinasse72ddc8f2011-01-13 15:46:08 -08002620 */
2621 if (!page_mkwrite) {
2622 wait_on_page_locked(dirty_page);
2623 set_page_dirty_balance(dirty_page, page_mkwrite);
2624 }
2625 put_page(dirty_page);
2626 if (page_mkwrite) {
2627 struct address_space *mapping = dirty_page->mapping;
2628
2629 set_page_dirty(dirty_page);
2630 unlock_page(dirty_page);
2631 page_cache_release(dirty_page);
2632 if (mapping) {
2633 /*
2634 * Some device drivers do not set page.mapping
2635 * but still dirty their pages
2636 */
2637 balance_dirty_pages_ratelimited(mapping);
2638 }
2639 }
2640
2641 /* file_update_time outside page_lock */
2642 if (vma->vm_file)
2643 file_update_time(vma->vm_file);
2644
2645 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647
2648 /*
2649 * Ok, we need to copy. Oh, well..
2650 */
Nick Pigginb5810032005-10-29 18:16:12 -07002651 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08002652gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002653 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654
2655 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07002656 goto oom;
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002657
Hugh Dickins62eede62009-09-21 17:03:34 -07002658 if (is_zero_pfn(pte_pfn(orig_pte))) {
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002659 new_page = alloc_zeroed_user_highpage_movable(vma, address);
2660 if (!new_page)
2661 goto oom;
2662 } else {
2663 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
2664 if (!new_page)
2665 goto oom;
2666 cow_user_page(new_page, old_page, address, vma);
2667 }
2668 __SetPageUptodate(new_page);
2669
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002670 if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002671 goto oom_free_new;
2672
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673 /*
2674 * Re-check the pte - we dropped the lock
2675 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002676 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07002677 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08002678 if (old_page) {
Hugh Dickins920fc352005-11-21 21:32:17 -08002679 if (!PageAnon(old_page)) {
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08002680 dec_mm_counter_fast(mm, MM_FILEPAGES);
2681 inc_mm_counter_fast(mm, MM_ANONPAGES);
Hugh Dickins920fc352005-11-21 21:32:17 -08002682 }
2683 } else
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08002684 inc_mm_counter_fast(mm, MM_ANONPAGES);
Ben Collinseca35132005-11-29 11:45:26 -08002685 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07002686 entry = mk_pte(new_page, vma->vm_page_prot);
2687 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Siddha, Suresh B4ce072f2006-09-29 01:58:42 -07002688 /*
2689 * Clear the pte entry and flush it first, before updating the
2690 * pte with the new entry. This will avoid a race condition
2691 * seen in the presence of one thread doing SMC and another
2692 * thread doing COW.
2693 */
Izik Eidus828502d2009-09-21 17:01:51 -07002694 ptep_clear_flush(vma, address, page_table);
Nick Piggin9617d952006-01-06 00:11:12 -08002695 page_add_new_anon_rmap(new_page, vma, address);
Izik Eidus828502d2009-09-21 17:01:51 -07002696 /*
2697 * We call the notify macro here because, when using secondary
2698 * mmu page tables (such as kvm shadow page tables), we want the
2699 * new page to be mapped directly into the secondary page table.
2700 */
2701 set_pte_at_notify(mm, address, page_table, entry);
Russell King4b3073e2009-12-18 16:40:18 +00002702 update_mmu_cache(vma, address, page_table);
Nick Piggin945754a2008-06-23 14:30:30 +02002703 if (old_page) {
2704 /*
2705 * Only after switching the pte to the new page may
2706 * we remove the mapcount here. Otherwise another
2707 * process may come and find the rmap count decremented
2708 * before the pte is switched to the new page, and
2709 * "reuse" the old page writing into it while our pte
2710 * here still points into it and can be read by other
2711 * threads.
2712 *
2713 * The critical issue is to order this
2714 * page_remove_rmap with the ptp_clear_flush above.
2715 * Those stores are ordered by (if nothing else,)
2716 * the barrier present in the atomic_add_negative
2717 * in page_remove_rmap.
2718 *
2719 * Then the TLB flush in ptep_clear_flush ensures that
2720 * no process can access the old page before the
2721 * decremented mapcount is visible. And the old page
2722 * cannot be reused until after the decremented
2723 * mapcount is visible. So transitively, TLBs to
2724 * old page will be flushed before it can be reused.
2725 */
Hugh Dickinsedc315f2009-01-06 14:40:11 -08002726 page_remove_rmap(old_page);
Nick Piggin945754a2008-06-23 14:30:30 +02002727 }
2728
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729 /* Free the old page.. */
2730 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10002731 ret |= VM_FAULT_WRITE;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002732 } else
2733 mem_cgroup_uncharge_page(new_page);
2734
Hugh Dickins920fc352005-11-21 21:32:17 -08002735 if (new_page)
2736 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002737unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002738 pte_unmap_unlock(page_table, ptl);
Michel Lespinassee15f8c02011-02-10 15:01:32 -08002739 if (old_page) {
2740 /*
2741 * Don't let another task, with possibly unlocked vma,
2742 * keep the mlocked page.
2743 */
2744 if ((ret & VM_FAULT_WRITE) && (vma->vm_flags & VM_LOCKED)) {
2745 lock_page(old_page); /* LRU manipulation */
2746 munlock_vma_page(old_page);
2747 unlock_page(old_page);
2748 }
2749 page_cache_release(old_page);
2750 }
Nick Pigginf33ea7f2005-08-03 20:24:01 +10002751 return ret;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002752oom_free_new:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002753 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002754oom:
Nick Pigginb827e492009-04-30 15:08:16 -07002755 if (old_page) {
2756 if (page_mkwrite) {
2757 unlock_page(old_page);
2758 page_cache_release(old_page);
2759 }
Hugh Dickins920fc352005-11-21 21:32:17 -08002760 page_cache_release(old_page);
Nick Pigginb827e492009-04-30 15:08:16 -07002761 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762 return VM_FAULT_OOM;
David Howells9637a5e2006-06-23 02:03:43 -07002763
2764unwritable_page:
2765 page_cache_release(old_page);
Nick Pigginc2ec1752009-03-31 15:23:21 -07002766 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002767}
2768
Peter Zijlstra97a89412011-05-24 17:12:04 -07002769static void unmap_mapping_range_vma(struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770 unsigned long start_addr, unsigned long end_addr,
2771 struct zap_details *details)
2772{
Peter Zijlstra97a89412011-05-24 17:12:04 -07002773 zap_page_range(vma, start_addr, end_addr - start_addr, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774}
2775
2776static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
2777 struct zap_details *details)
2778{
2779 struct vm_area_struct *vma;
2780 struct prio_tree_iter iter;
2781 pgoff_t vba, vea, zba, zea;
2782
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783 vma_prio_tree_foreach(vma, &iter, root,
2784 details->first_index, details->last_index) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785
2786 vba = vma->vm_pgoff;
2787 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
2788 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
2789 zba = details->first_index;
2790 if (zba < vba)
2791 zba = vba;
2792 zea = details->last_index;
2793 if (zea > vea)
2794 zea = vea;
2795
Peter Zijlstra97a89412011-05-24 17:12:04 -07002796 unmap_mapping_range_vma(vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
2798 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
Peter Zijlstra97a89412011-05-24 17:12:04 -07002799 details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 }
2801}
2802
2803static inline void unmap_mapping_range_list(struct list_head *head,
2804 struct zap_details *details)
2805{
2806 struct vm_area_struct *vma;
2807
2808 /*
2809 * In nonlinear VMAs there is no correspondence between virtual address
2810 * offset and file offset. So we must perform an exhaustive search
2811 * across *all* the pages in each nonlinear VMA, not just the pages
2812 * whose virtual address lies outside the file truncation point.
2813 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814 list_for_each_entry(vma, head, shared.vm_set.list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 details->nonlinear_vma = vma;
Peter Zijlstra97a89412011-05-24 17:12:04 -07002816 unmap_mapping_range_vma(vma, vma->vm_start, vma->vm_end, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817 }
2818}
2819
2820/**
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002821 * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file.
Martin Waitz3d410882005-06-23 22:05:21 -07002822 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823 * @holebegin: byte in first page to unmap, relative to the start of
2824 * the underlying file. This will be rounded down to a PAGE_SIZE
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +10002825 * boundary. Note that this is different from truncate_pagecache(), which
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826 * must keep the partial page. In contrast, we must get rid of
2827 * partial pages.
2828 * @holelen: size of prospective hole in bytes. This will be rounded
2829 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2830 * end of the file.
2831 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2832 * but 0 when invalidating pagecache, don't throw away private data.
2833 */
2834void unmap_mapping_range(struct address_space *mapping,
2835 loff_t const holebegin, loff_t const holelen, int even_cows)
2836{
2837 struct zap_details details;
2838 pgoff_t hba = holebegin >> PAGE_SHIFT;
2839 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2840
2841 /* Check for overflow. */
2842 if (sizeof(holelen) > sizeof(hlen)) {
2843 long long holeend =
2844 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2845 if (holeend & ~(long long)ULONG_MAX)
2846 hlen = ULONG_MAX - hba + 1;
2847 }
2848
2849 details.check_mapping = even_cows? NULL: mapping;
2850 details.nonlinear_vma = NULL;
2851 details.first_index = hba;
2852 details.last_index = hba + hlen - 1;
2853 if (details.last_index < details.first_index)
2854 details.last_index = ULONG_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855
Peter Zijlstra97a89412011-05-24 17:12:04 -07002856
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002857 mutex_lock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
2859 unmap_mapping_range_tree(&mapping->i_mmap, &details);
2860 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
2861 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
Peter Zijlstra3d48ae42011-05-24 17:12:06 -07002862 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863}
2864EXPORT_SYMBOL(unmap_mapping_range);
2865
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002867 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2868 * but allow concurrent faults), and pte mapped but not yet locked.
2869 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002871static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2872 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002873 unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002875 spinlock_t *ptl;
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002876 struct page *page, *swapcache = NULL;
Hugh Dickins65500d22005-10-29 18:15:59 -07002877 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 pte_t pte;
Michel Lespinassed065bd82010-10-26 14:21:57 -07002879 int locked;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -07002880 struct mem_cgroup *ptr;
Rik van Rielad8c2ee2010-08-09 17:19:48 -07002881 int exclusive = 0;
Nick Piggin83c54072007-07-19 01:47:05 -07002882 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002884 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002885 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002886
2887 entry = pte_to_swp_entry(orig_pte);
Andi Kleend1737fd2009-09-16 11:50:06 +02002888 if (unlikely(non_swap_entry(entry))) {
2889 if (is_migration_entry(entry)) {
2890 migration_entry_wait(mm, pmd, address);
2891 } else if (is_hwpoison_entry(entry)) {
2892 ret = VM_FAULT_HWPOISON;
2893 } else {
2894 print_bad_pte(vma, address, orig_pte, NULL);
Hugh Dickinsd99be1a2009-12-14 17:59:04 -08002895 ret = VM_FAULT_SIGBUS;
Andi Kleend1737fd2009-09-16 11:50:06 +02002896 }
Christoph Lameter06972122006-06-23 02:03:35 -07002897 goto out;
2898 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002899 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002900 page = lookup_swap_cache(entry);
2901 if (!page) {
Hugh Dickinsa5c9b692009-06-23 12:36:58 -07002902 grab_swap_token(mm); /* Contend for token _before_ read-in */
Hugh Dickins02098fe2008-02-04 22:28:42 -08002903 page = swapin_readahead(entry,
2904 GFP_HIGHUSER_MOVABLE, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002905 if (!page) {
2906 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002907 * Back out if somebody else faulted in this pte
2908 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002910 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911 if (likely(pte_same(*page_table, orig_pte)))
2912 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002913 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002914 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915 }
2916
2917 /* Had to read the page from swap area: Major fault */
2918 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002919 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07002920 mem_cgroup_count_vm_event(mm, PGMAJFAULT);
Andi Kleend1737fd2009-09-16 11:50:06 +02002921 } else if (PageHWPoison(page)) {
Wu Fengguang71f72522009-12-16 12:19:58 +01002922 /*
2923 * hwpoisoned dirty swapcache pages are kept for killing
2924 * owner processes (which may be unknown at hwpoison time)
2925 */
Andi Kleend1737fd2009-09-16 11:50:06 +02002926 ret = VM_FAULT_HWPOISON;
2927 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Andi Kleen4779cb32009-10-14 01:51:41 +02002928 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002929 }
2930
Michel Lespinassed065bd82010-10-26 14:21:57 -07002931 locked = lock_page_or_retry(page, mm, flags);
Balbir Singh20a10222007-11-14 17:00:33 -08002932 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Michel Lespinassed065bd82010-10-26 14:21:57 -07002933 if (!locked) {
2934 ret |= VM_FAULT_RETRY;
2935 goto out_release;
2936 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002938 /*
Hugh Dickins31c4a3d2010-09-19 19:40:22 -07002939 * Make sure try_to_free_swap or reuse_swap_page or swapoff did not
2940 * release the swapcache from under us. The page pin, and pte_same
2941 * test below, are not enough to exclude that. Even if it is still
2942 * swapcache, we need to check that the page's swap has not changed.
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002943 */
Hugh Dickins31c4a3d2010-09-19 19:40:22 -07002944 if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val))
Andrea Arcangeli4969c112010-09-09 16:37:52 -07002945 goto out_page;
2946
2947 if (ksm_might_need_to_copy(page, vma, address)) {
2948 swapcache = page;
2949 page = ksm_does_need_to_copy(page, vma, address);
2950
2951 if (unlikely(!page)) {
2952 ret = VM_FAULT_OOM;
2953 page = swapcache;
2954 swapcache = NULL;
2955 goto out_page;
2956 }
Hugh Dickins5ad64682009-12-14 17:59:24 -08002957 }
2958
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002959 if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) {
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002960 ret = VM_FAULT_OOM;
Johannes Weinerbc43f752009-04-30 15:08:08 -07002961 goto out_page;
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002962 }
2963
Linus Torvalds1da177e2005-04-16 15:20:36 -07002964 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002965 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002966 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002967 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002968 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002969 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002970
2971 if (unlikely(!PageUptodate(page))) {
2972 ret = VM_FAULT_SIGBUS;
2973 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002974 }
2975
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002976 /*
2977 * The page isn't present yet, go ahead with the fault.
2978 *
2979 * Be careful about the sequence of operations here.
2980 * To get its accounting right, reuse_swap_page() must be called
2981 * while the page is counted on swap but not yet in mapcount i.e.
2982 * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
2983 * must be called after the swap_free(), or it will never succeed.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002984 * Because delete_from_swap_page() may be called by reuse_swap_page(),
2985 * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry
2986 * in page->private. In this case, a record in swap_cgroup is silently
2987 * discarded at swap_free().
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002988 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08002990 inc_mm_counter_fast(mm, MM_ANONPAGES);
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08002991 dec_mm_counter_fast(mm, MM_SWAPENTS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992 pte = mk_pte(page, vma->vm_page_prot);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002993 if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002995 flags &= ~FAULT_FLAG_WRITE;
Andrea Arcangeli9a5b4892010-08-09 17:19:49 -07002996 ret |= VM_FAULT_WRITE;
Rik van Rielad8c2ee2010-08-09 17:19:48 -07002997 exclusive = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999 flush_icache_page(vma, page);
3000 set_pte_at(mm, address, page_table, pte);
Rik van Rielad8c2ee2010-08-09 17:19:48 -07003001 do_page_add_anon_rmap(page, vma, address, exclusive);
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08003002 /* It's better to call commit-charge after rmap is established */
3003 mem_cgroup_commit_charge_swapin(page, ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07003005 swap_free(entry);
Nick Pigginb291f002008-10-18 20:26:44 -07003006 if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08003007 try_to_free_swap(page);
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07003008 unlock_page(page);
Andrea Arcangeli4969c112010-09-09 16:37:52 -07003009 if (swapcache) {
3010 /*
3011 * Hold the lock to avoid the swap entry to be reused
3012 * until we take the PT lock for the pte_same() check
3013 * (to avoid false positives from pte_same). For
3014 * further safety release the lock after the swap_free
3015 * so that the swap count won't change under a
3016 * parallel locked swapcache.
3017 */
3018 unlock_page(swapcache);
3019 page_cache_release(swapcache);
3020 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07003021
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003022 if (flags & FAULT_FLAG_WRITE) {
Hugh Dickins61469f12008-03-04 14:29:04 -08003023 ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
3024 if (ret & VM_FAULT_ERROR)
3025 ret &= VM_FAULT_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026 goto out;
3027 }
3028
3029 /* No need to invalidate - it was non-present before */
Russell King4b3073e2009-12-18 16:40:18 +00003030 update_mmu_cache(vma, address, page_table);
Hugh Dickins65500d22005-10-29 18:15:59 -07003031unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003032 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033out:
3034 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07003035out_nomap:
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08003036 mem_cgroup_cancel_charge_swapin(ptr);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003037 pte_unmap_unlock(page_table, ptl);
Johannes Weinerbc43f752009-04-30 15:08:08 -07003038out_page:
Kirill Korotaevb8107482005-05-16 21:53:50 -07003039 unlock_page(page);
Andi Kleen4779cb32009-10-14 01:51:41 +02003040out_release:
Kirill Korotaevb8107482005-05-16 21:53:50 -07003041 page_cache_release(page);
Andrea Arcangeli4969c112010-09-09 16:37:52 -07003042 if (swapcache) {
3043 unlock_page(swapcache);
3044 page_cache_release(swapcache);
3045 }
Hugh Dickins65500d22005-10-29 18:15:59 -07003046 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003047}
3048
3049/*
Luck, Tony8ca3eb02010-08-24 11:44:18 -07003050 * This is like a special single-page "expand_{down|up}wards()",
3051 * except we must first make sure that 'address{-|+}PAGE_SIZE'
Linus Torvalds320b2b82010-08-12 17:54:33 -07003052 * doesn't hit another vma.
Linus Torvalds320b2b82010-08-12 17:54:33 -07003053 */
3054static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address)
3055{
3056 address &= PAGE_MASK;
3057 if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) {
Linus Torvalds0e8e50e2010-08-20 16:49:40 -07003058 struct vm_area_struct *prev = vma->vm_prev;
Linus Torvalds320b2b82010-08-12 17:54:33 -07003059
Linus Torvalds0e8e50e2010-08-20 16:49:40 -07003060 /*
3061 * Is there a mapping abutting this one below?
3062 *
3063 * That's only ok if it's the same stack mapping
3064 * that has gotten split..
3065 */
3066 if (prev && prev->vm_end == address)
3067 return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM;
3068
Michal Hockod05f3162011-05-24 17:11:44 -07003069 expand_downwards(vma, address - PAGE_SIZE);
Linus Torvalds320b2b82010-08-12 17:54:33 -07003070 }
Luck, Tony8ca3eb02010-08-24 11:44:18 -07003071 if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) {
3072 struct vm_area_struct *next = vma->vm_next;
3073
3074 /* As VM_GROWSDOWN but s/below/above/ */
3075 if (next && next->vm_start == address + PAGE_SIZE)
3076 return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM;
3077
3078 expand_upwards(vma, address + PAGE_SIZE);
3079 }
Linus Torvalds320b2b82010-08-12 17:54:33 -07003080 return 0;
3081}
3082
3083/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003084 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3085 * but allow concurrent faults), and pte mapped but not yet locked.
3086 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003087 */
Hugh Dickins65500d22005-10-29 18:15:59 -07003088static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
3089 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003090 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003092 struct page *page;
3093 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003094 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095
Linus Torvalds11ac5522010-08-14 11:44:56 -07003096 pte_unmap(page_table);
Linus Torvalds320b2b82010-08-12 17:54:33 -07003097
Linus Torvalds11ac5522010-08-14 11:44:56 -07003098 /* Check if we need to add a guard page to the stack */
3099 if (check_stack_guard_page(vma, address) < 0)
3100 return VM_FAULT_SIGBUS;
3101
3102 /* Use the zero-page for reads */
Hugh Dickins62eede62009-09-21 17:03:34 -07003103 if (!(flags & FAULT_FLAG_WRITE)) {
3104 entry = pte_mkspecial(pfn_pte(my_zero_pfn(address),
3105 vma->vm_page_prot));
Linus Torvalds11ac5522010-08-14 11:44:56 -07003106 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07003107 if (!pte_none(*page_table))
3108 goto unlock;
3109 goto setpte;
3110 }
3111
Nick Piggin557ed1f2007-10-16 01:24:40 -07003112 /* Allocate our own private page. */
Nick Piggin557ed1f2007-10-16 01:24:40 -07003113 if (unlikely(anon_vma_prepare(vma)))
3114 goto oom;
3115 page = alloc_zeroed_user_highpage_movable(vma, address);
3116 if (!page)
3117 goto oom;
Nick Piggin0ed361d2008-02-04 22:29:34 -08003118 __SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003119
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08003120 if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08003121 goto oom_free_page;
3122
Nick Piggin557ed1f2007-10-16 01:24:40 -07003123 entry = mk_pte(page, vma->vm_page_prot);
Hugh Dickins1ac0cb52009-09-21 17:03:29 -07003124 if (vma->vm_flags & VM_WRITE)
3125 entry = pte_mkwrite(pte_mkdirty(entry));
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003126
Nick Piggin557ed1f2007-10-16 01:24:40 -07003127 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Andrea Arcangeli1c2fb7a2009-09-21 17:02:22 -07003128 if (!pte_none(*page_table))
Nick Piggin557ed1f2007-10-16 01:24:40 -07003129 goto release;
Hugh Dickins9ba69292009-09-21 17:02:20 -07003130
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08003131 inc_mm_counter_fast(mm, MM_ANONPAGES);
Nick Piggin557ed1f2007-10-16 01:24:40 -07003132 page_add_new_anon_rmap(page, vma, address);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07003133setpte:
Hugh Dickins65500d22005-10-29 18:15:59 -07003134 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003135
3136 /* No need to invalidate - it was non-present before */
Russell King4b3073e2009-12-18 16:40:18 +00003137 update_mmu_cache(vma, address, page_table);
Hugh Dickins65500d22005-10-29 18:15:59 -07003138unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003139 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07003140 return 0;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003141release:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08003142 mem_cgroup_uncharge_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003143 page_cache_release(page);
3144 goto unlock;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08003145oom_free_page:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08003146 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07003147oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003148 return VM_FAULT_OOM;
3149}
3150
3151/*
Nick Piggin54cb8822007-07-19 01:46:59 -07003152 * __do_fault() tries to create a new page mapping. It aggressively
Linus Torvalds1da177e2005-04-16 15:20:36 -07003153 * tries to share with existing pages, but makes a separate copy if
Nick Piggin54cb8822007-07-19 01:46:59 -07003154 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
3155 * the next page fault.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003156 *
3157 * As this is called only for pages that do not currently exist, we
3158 * do not need to flush old virtual caches or the TLB.
3159 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003160 * We enter with non-exclusive mmap_sem (to exclude vma changes,
Hugh Dickins16abfa02007-10-04 16:56:06 +01003161 * but allow concurrent faults), and pte neither mapped nor locked.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003162 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163 */
Nick Piggin54cb8822007-07-19 01:46:59 -07003164static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins16abfa02007-10-04 16:56:06 +01003165 unsigned long address, pmd_t *pmd,
Nick Piggin54cb8822007-07-19 01:46:59 -07003166 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167{
Hugh Dickins16abfa02007-10-04 16:56:06 +01003168 pte_t *page_table;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003169 spinlock_t *ptl;
Nick Piggind0217ac2007-07-19 01:47:03 -07003170 struct page *page;
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07003171 struct page *cow_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003172 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003173 int anon = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07003174 struct page *dirty_page = NULL;
Nick Piggind0217ac2007-07-19 01:47:03 -07003175 struct vm_fault vmf;
3176 int ret;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02003177 int page_mkwrite = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07003178
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07003179 /*
3180 * If we do COW later, allocate page befor taking lock_page()
3181 * on the file cache page. This will reduce lock holding time.
3182 */
3183 if ((flags & FAULT_FLAG_WRITE) && !(vma->vm_flags & VM_SHARED)) {
3184
3185 if (unlikely(anon_vma_prepare(vma)))
3186 return VM_FAULT_OOM;
3187
3188 cow_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
3189 if (!cow_page)
3190 return VM_FAULT_OOM;
3191
3192 if (mem_cgroup_newpage_charge(cow_page, mm, GFP_KERNEL)) {
3193 page_cache_release(cow_page);
3194 return VM_FAULT_OOM;
3195 }
3196 } else
3197 cow_page = NULL;
3198
Nick Piggind0217ac2007-07-19 01:47:03 -07003199 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
3200 vmf.pgoff = pgoff;
3201 vmf.flags = flags;
3202 vmf.page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203
Nick Piggin3c18ddd2008-04-28 02:12:10 -07003204 ret = vma->vm_ops->fault(vma, &vmf);
Michel Lespinassed065bd82010-10-26 14:21:57 -07003205 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE |
3206 VM_FAULT_RETRY)))
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07003207 goto uncharge_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003208
Andi Kleena3b947e2009-09-16 11:50:08 +02003209 if (unlikely(PageHWPoison(vmf.page))) {
3210 if (ret & VM_FAULT_LOCKED)
3211 unlock_page(vmf.page);
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07003212 ret = VM_FAULT_HWPOISON;
3213 goto uncharge_out;
Andi Kleena3b947e2009-09-16 11:50:08 +02003214 }
3215
Nick Piggind00806b2007-07-19 01:46:57 -07003216 /*
Nick Piggind0217ac2007-07-19 01:47:03 -07003217 * For consistency in subsequent calls, make the faulted page always
Nick Piggind00806b2007-07-19 01:46:57 -07003218 * locked.
3219 */
Nick Piggin83c54072007-07-19 01:47:05 -07003220 if (unlikely(!(ret & VM_FAULT_LOCKED)))
Nick Piggind0217ac2007-07-19 01:47:03 -07003221 lock_page(vmf.page);
Nick Piggin54cb8822007-07-19 01:46:59 -07003222 else
Nick Piggind0217ac2007-07-19 01:47:03 -07003223 VM_BUG_ON(!PageLocked(vmf.page));
Nick Piggind00806b2007-07-19 01:46:57 -07003224
Linus Torvalds1da177e2005-04-16 15:20:36 -07003225 /*
3226 * Should we do an early C-O-W break?
3227 */
Nick Piggind0217ac2007-07-19 01:47:03 -07003228 page = vmf.page;
Nick Piggin54cb8822007-07-19 01:46:59 -07003229 if (flags & FAULT_FLAG_WRITE) {
David Howells9637a5e2006-06-23 02:03:43 -07003230 if (!(vma->vm_flags & VM_SHARED)) {
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07003231 page = cow_page;
Nick Piggin54cb8822007-07-19 01:46:59 -07003232 anon = 1;
Nick Piggind0217ac2007-07-19 01:47:03 -07003233 copy_user_highpage(page, vmf.page, address, vma);
Nick Piggin0ed361d2008-02-04 22:29:34 -08003234 __SetPageUptodate(page);
David Howells9637a5e2006-06-23 02:03:43 -07003235 } else {
Nick Piggin54cb8822007-07-19 01:46:59 -07003236 /*
3237 * If the page will be shareable, see if the backing
David Howells9637a5e2006-06-23 02:03:43 -07003238 * address space wants to know that the page is about
Nick Piggin54cb8822007-07-19 01:46:59 -07003239 * to become writable
3240 */
Mark Fasheh69676142007-07-19 01:47:00 -07003241 if (vma->vm_ops->page_mkwrite) {
Nick Pigginc2ec1752009-03-31 15:23:21 -07003242 int tmp;
3243
Mark Fasheh69676142007-07-19 01:47:00 -07003244 unlock_page(page);
Nick Pigginb827e492009-04-30 15:08:16 -07003245 vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
Nick Pigginc2ec1752009-03-31 15:23:21 -07003246 tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
3247 if (unlikely(tmp &
3248 (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
3249 ret = tmp;
Nick Pigginb827e492009-04-30 15:08:16 -07003250 goto unwritable_page;
Mark Fasheh69676142007-07-19 01:47:00 -07003251 }
Nick Pigginb827e492009-04-30 15:08:16 -07003252 if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
3253 lock_page(page);
3254 if (!page->mapping) {
3255 ret = 0; /* retry the fault */
3256 unlock_page(page);
3257 goto unwritable_page;
3258 }
3259 } else
3260 VM_BUG_ON(!PageLocked(page));
Peter Zijlstraa200ee12007-10-08 18:54:37 +02003261 page_mkwrite = 1;
David Howells9637a5e2006-06-23 02:03:43 -07003262 }
3263 }
Nick Piggin54cb8822007-07-19 01:46:59 -07003264
Linus Torvalds1da177e2005-04-16 15:20:36 -07003265 }
3266
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003267 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003268
3269 /*
3270 * This silly early PAGE_DIRTY setting removes a race
3271 * due to the bad i386 page protection. But it's valid
3272 * for other architectures too.
3273 *
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003274 * Note that if FAULT_FLAG_WRITE is set, we either now have
Linus Torvalds1da177e2005-04-16 15:20:36 -07003275 * an exclusive copy of the page, or this is a shared mapping,
3276 * so we can make it writable and dirty to avoid having to
3277 * handle that later.
3278 */
3279 /* Only go through if we didn't race with anybody else... */
Andrea Arcangeli1c2fb7a2009-09-21 17:02:22 -07003280 if (likely(pte_same(*page_table, orig_pte))) {
Nick Piggind00806b2007-07-19 01:46:57 -07003281 flush_icache_page(vma, page);
3282 entry = mk_pte(page, vma->vm_page_prot);
Nick Piggin54cb8822007-07-19 01:46:59 -07003283 if (flags & FAULT_FLAG_WRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003284 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003285 if (anon) {
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08003286 inc_mm_counter_fast(mm, MM_ANONPAGES);
Lee Schermerhorn64d65192008-10-18 20:26:52 -07003287 page_add_new_anon_rmap(page, vma, address);
Hugh Dickinsf57e88a2005-11-21 21:32:19 -08003288 } else {
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08003289 inc_mm_counter_fast(mm, MM_FILEPAGES);
Nick Piggind00806b2007-07-19 01:46:57 -07003290 page_add_file_rmap(page);
Nick Piggin54cb8822007-07-19 01:46:59 -07003291 if (flags & FAULT_FLAG_WRITE) {
Nick Piggind00806b2007-07-19 01:46:57 -07003292 dirty_page = page;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07003293 get_page(dirty_page);
3294 }
Hugh Dickins42946212005-10-29 18:16:05 -07003295 }
Lee Schermerhorn64d65192008-10-18 20:26:52 -07003296 set_pte_at(mm, address, page_table, entry);
Nick Piggind00806b2007-07-19 01:46:57 -07003297
3298 /* no need to invalidate: a not-present page won't be cached */
Russell King4b3073e2009-12-18 16:40:18 +00003299 update_mmu_cache(vma, address, page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003300 } else {
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07003301 if (cow_page)
3302 mem_cgroup_uncharge_page(cow_page);
Nick Piggind00806b2007-07-19 01:46:57 -07003303 if (anon)
3304 page_cache_release(page);
3305 else
Nick Piggin54cb8822007-07-19 01:46:59 -07003306 anon = 1; /* no anon but release faulted_page */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003307 }
3308
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003309 pte_unmap_unlock(page_table, ptl);
Nick Piggind00806b2007-07-19 01:46:57 -07003310
Nick Pigginb827e492009-04-30 15:08:16 -07003311 if (dirty_page) {
3312 struct address_space *mapping = page->mapping;
3313
3314 if (set_page_dirty(dirty_page))
3315 page_mkwrite = 1;
3316 unlock_page(dirty_page);
3317 put_page(dirty_page);
3318 if (page_mkwrite && mapping) {
3319 /*
3320 * Some device drivers do not set page.mapping but still
3321 * dirty their pages
3322 */
3323 balance_dirty_pages_ratelimited(mapping);
3324 }
3325
3326 /* file_update_time outside page_lock */
Anton Salikhmetov8f7b3d12008-01-23 02:21:18 +03003327 if (vma->vm_file)
3328 file_update_time(vma->vm_file);
Nick Pigginb827e492009-04-30 15:08:16 -07003329 } else {
3330 unlock_page(vmf.page);
3331 if (anon)
3332 page_cache_release(vmf.page);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07003333 }
Nick Piggind00806b2007-07-19 01:46:57 -07003334
Nick Piggin83c54072007-07-19 01:47:05 -07003335 return ret;
Nick Pigginb827e492009-04-30 15:08:16 -07003336
3337unwritable_page:
3338 page_cache_release(page);
3339 return ret;
KAMEZAWA Hiroyuki1d65f862011-07-25 17:12:27 -07003340uncharge_out:
3341 /* fs's fault handler get error */
3342 if (cow_page) {
3343 mem_cgroup_uncharge_page(cow_page);
3344 page_cache_release(cow_page);
3345 }
3346 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07003347}
Nick Piggind00806b2007-07-19 01:46:57 -07003348
Nick Piggin54cb8822007-07-19 01:46:59 -07003349static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3350 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003351 unsigned int flags, pte_t orig_pte)
Nick Piggin54cb8822007-07-19 01:46:59 -07003352{
3353 pgoff_t pgoff = (((address & PAGE_MASK)
Dean Nelson0da7e012007-10-16 01:24:45 -07003354 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
Nick Piggin54cb8822007-07-19 01:46:59 -07003355
Hugh Dickins16abfa02007-10-04 16:56:06 +01003356 pte_unmap(page_table);
3357 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Nick Piggin54cb8822007-07-19 01:46:59 -07003358}
3359
Jes Sorensenf4b81802006-09-27 01:50:10 -07003360/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003361 * Fault of a previously existing named mapping. Repopulate the pte
3362 * from the encoded file_pte if possible. This enables swappable
3363 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003364 *
3365 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3366 * but allow concurrent faults), and pte mapped but not yet locked.
3367 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003368 */
Nick Piggind0217ac2007-07-19 01:47:03 -07003369static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins65500d22005-10-29 18:15:59 -07003370 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003371 unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003372{
Hugh Dickins65500d22005-10-29 18:15:59 -07003373 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003374
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003375 flags |= FAULT_FLAG_NONLINEAR;
3376
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07003377 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Nick Piggin83c54072007-07-19 01:47:05 -07003378 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379
Hugh Dickins2509ef22009-01-06 14:40:10 -08003380 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
Hugh Dickins65500d22005-10-29 18:15:59 -07003381 /*
3382 * Page table corrupted: show pte and kill process.
3383 */
Hugh Dickins3dc14742009-01-06 14:40:08 -08003384 print_bad_pte(vma, address, orig_pte, NULL);
Hugh Dickinsd99be1a2009-12-14 17:59:04 -08003385 return VM_FAULT_SIGBUS;
Hugh Dickins65500d22005-10-29 18:15:59 -07003386 }
Hugh Dickins65500d22005-10-29 18:15:59 -07003387
3388 pgoff = pte_to_pgoff(orig_pte);
Hugh Dickins16abfa02007-10-04 16:56:06 +01003389 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003390}
3391
3392/*
3393 * These routines also need to handle stuff like marking pages dirty
3394 * and/or accessed for architectures that don't do it in hardware (most
3395 * RISC architectures). The early dirtying is also good on the i386.
3396 *
3397 * There is also a hook called "update_mmu_cache()" that architectures
3398 * with external mmu caches can use to update those (ie the Sparc or
3399 * PowerPC hashed page tables that act as extended TLBs).
3400 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07003401 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3402 * but allow concurrent faults), and pte mapped but not yet locked.
3403 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003404 */
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003405int handle_pte_fault(struct mm_struct *mm,
3406 struct vm_area_struct *vma, unsigned long address,
3407 pte_t *pte, pmd_t *pmd, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003408{
3409 pte_t entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003410 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003411
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07003412 entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003413 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07003414 if (pte_none(entry)) {
Jes Sorensenf4b81802006-09-27 01:50:10 -07003415 if (vma->vm_ops) {
Nick Piggin3c18ddd2008-04-28 02:12:10 -07003416 if (likely(vma->vm_ops->fault))
Nick Piggin54cb8822007-07-19 01:46:59 -07003417 return do_linear_fault(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003418 pte, pmd, flags, entry);
Jes Sorensenf4b81802006-09-27 01:50:10 -07003419 }
3420 return do_anonymous_page(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003421 pte, pmd, flags);
Hugh Dickins65500d22005-10-29 18:15:59 -07003422 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003423 if (pte_file(entry))
Nick Piggind0217ac2007-07-19 01:47:03 -07003424 return do_nonlinear_fault(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003425 pte, pmd, flags, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07003426 return do_swap_page(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003427 pte, pmd, flags, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003428 }
3429
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07003430 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003431 spin_lock(ptl);
3432 if (unlikely(!pte_same(*pte, entry)))
3433 goto unlock;
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003434 if (flags & FAULT_FLAG_WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003435 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003436 return do_wp_page(mm, vma, address,
3437 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003438 entry = pte_mkdirty(entry);
3439 }
3440 entry = pte_mkyoung(entry);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003441 if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) {
Russell King4b3073e2009-12-18 16:40:18 +00003442 update_mmu_cache(vma, address, pte);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07003443 } else {
3444 /*
3445 * This is needed only for protection faults but the arch code
3446 * is not yet telling us if this is a protection fault or not.
3447 * This still avoids useless tlb flushes for .text page faults
3448 * with threads.
3449 */
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003450 if (flags & FAULT_FLAG_WRITE)
Shaohua Li61c77322010-08-16 09:16:55 +08003451 flush_tlb_fix_spurious_fault(vma, address);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07003452 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003453unlock:
3454 pte_unmap_unlock(pte, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07003455 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003456}
3457
3458/*
3459 * By the time we get here, we already hold the mm semaphore
3460 */
Nick Piggin83c54072007-07-19 01:47:05 -07003461int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvaldsd06063c2009-04-10 09:01:23 -07003462 unsigned long address, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003463{
3464 pgd_t *pgd;
3465 pud_t *pud;
3466 pmd_t *pmd;
3467 pte_t *pte;
3468
3469 __set_current_state(TASK_RUNNING);
3470
Christoph Lameterf8891e52006-06-30 01:55:45 -07003471 count_vm_event(PGFAULT);
Ying Han456f9982011-05-26 16:25:38 -07003472 mem_cgroup_count_vm_event(mm, PGFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003473
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08003474 /* do counter updates before entering really critical section. */
3475 check_sync_rss_stat(current);
3476
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01003477 if (unlikely(is_vm_hugetlb_page(vma)))
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003478 return hugetlb_fault(mm, vma, address, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003479
Linus Torvalds1da177e2005-04-16 15:20:36 -07003480 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003481 pud = pud_alloc(mm, pgd, address);
3482 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07003483 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003484 pmd = pmd_alloc(mm, pud, address);
3485 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07003486 return VM_FAULT_OOM;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003487 if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) {
3488 if (!vma->vm_ops)
3489 return do_huge_pmd_anonymous_page(mm, vma, address,
3490 pmd, flags);
3491 } else {
3492 pmd_t orig_pmd = *pmd;
3493 barrier();
3494 if (pmd_trans_huge(orig_pmd)) {
3495 if (flags & FAULT_FLAG_WRITE &&
3496 !pmd_write(orig_pmd) &&
3497 !pmd_trans_splitting(orig_pmd))
3498 return do_huge_pmd_wp_page(mm, vma, address,
3499 pmd, orig_pmd);
3500 return 0;
3501 }
3502 }
3503
3504 /*
3505 * Use __pte_alloc instead of pte_alloc_map, because we can't
3506 * run pte_offset_map on the pmd, if an huge pmd could
3507 * materialize from under us from a different thread.
3508 */
Mel Gormancc036382011-04-27 15:26:56 -07003509 if (unlikely(pmd_none(*pmd)) && __pte_alloc(mm, vma, pmd, address))
Hugh Dickinsc74df322005-10-29 18:16:23 -07003510 return VM_FAULT_OOM;
Andrea Arcangeli71e3aac2011-01-13 15:46:52 -08003511 /* if an huge pmd materialized from under us just retry later */
3512 if (unlikely(pmd_trans_huge(*pmd)))
3513 return 0;
3514 /*
3515 * A regular pmd is established and it can't morph into a huge pmd
3516 * from under us anymore at this point because we hold the mmap_sem
3517 * read mode and khugepaged takes it in write mode. So now it's
3518 * safe to run pte_offset_map().
3519 */
3520 pte = pte_offset_map(pmd, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003521
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003522 return handle_pte_fault(mm, vma, address, pte, pmd, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003523}
3524
3525#ifndef __PAGETABLE_PUD_FOLDED
3526/*
3527 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07003528 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003529 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07003530int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003531{
Hugh Dickinsc74df322005-10-29 18:16:23 -07003532 pud_t *new = pud_alloc_one(mm, address);
3533 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07003534 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003535
Nick Piggin362a61a2008-05-14 06:37:36 +02003536 smp_wmb(); /* See comment in __pte_alloc */
3537
Hugh Dickins872fec12005-10-29 18:16:21 -07003538 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003539 if (pgd_present(*pgd)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003540 pud_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003541 else
3542 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07003543 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003544 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003545}
3546#endif /* __PAGETABLE_PUD_FOLDED */
3547
3548#ifndef __PAGETABLE_PMD_FOLDED
3549/*
3550 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07003551 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003552 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07003553int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003554{
Hugh Dickinsc74df322005-10-29 18:16:23 -07003555 pmd_t *new = pmd_alloc_one(mm, address);
3556 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07003557 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003558
Nick Piggin362a61a2008-05-14 06:37:36 +02003559 smp_wmb(); /* See comment in __pte_alloc */
3560
Hugh Dickins872fec12005-10-29 18:16:21 -07003561 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07003563 if (pud_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003564 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003565 else
3566 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003567#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07003568 if (pgd_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003569 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003570 else
3571 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003572#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07003573 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003574 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003575}
3576#endif /* __PAGETABLE_PMD_FOLDED */
3577
3578int make_pages_present(unsigned long addr, unsigned long end)
3579{
3580 int ret, len, write;
3581 struct vm_area_struct * vma;
3582
3583 vma = find_vma(current->mm, addr);
3584 if (!vma)
KOSAKI Motohiroa4770972008-08-04 13:41:14 -07003585 return -ENOMEM;
Michel Lespinasse5ecfda02011-01-13 15:46:09 -08003586 /*
3587 * We want to touch writable mappings with a write fault in order
3588 * to break COW, except for shared mappings because these don't COW
3589 * and we would not want to dirty them for nothing.
3590 */
3591 write = (vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE;
Eric Sesterhenn5bcb28b2006-03-26 18:30:52 +02003592 BUG_ON(addr >= end);
3593 BUG_ON(end > vma->vm_end);
Rolf Eike Beer68e116a2007-07-15 23:38:03 -07003594 len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003595 ret = get_user_pages(current, current->mm, addr,
3596 len, write, 0, NULL, NULL);
Lee Schermerhornc11d69d2008-10-18 20:26:56 -07003597 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003598 return ret;
Lee Schermerhorn9978ad52008-10-18 20:26:56 -07003599 return ret == len ? 0 : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003600}
3601
Linus Torvalds1da177e2005-04-16 15:20:36 -07003602#if !defined(__HAVE_ARCH_GATE_AREA)
3603
3604#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07003605static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003606
3607static int __init gate_vma_init(void)
3608{
3609 gate_vma.vm_mm = NULL;
3610 gate_vma.vm_start = FIXADDR_USER_START;
3611 gate_vma.vm_end = FIXADDR_USER_END;
Roland McGrathb6558c42007-01-26 00:56:47 -08003612 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
3613 gate_vma.vm_page_prot = __P101;
Roland McGrathf47aef52007-01-26 00:56:49 -08003614 /*
3615 * Make sure the vDSO gets into every core dump.
3616 * Dumping its contents makes post-mortem fully interpretable later
3617 * without matching up the same kernel and hardware config to see
3618 * what PC values meant.
3619 */
3620 gate_vma.vm_flags |= VM_ALWAYSDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003621 return 0;
3622}
3623__initcall(gate_vma_init);
3624#endif
3625
Stephen Wilson31db58b2011-03-13 15:49:15 -04003626struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003627{
3628#ifdef AT_SYSINFO_EHDR
3629 return &gate_vma;
3630#else
3631 return NULL;
3632#endif
3633}
3634
Stephen Wilsoncae5d392011-03-13 15:49:17 -04003635int in_gate_area_no_mm(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003636{
3637#ifdef AT_SYSINFO_EHDR
3638 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
3639 return 1;
3640#endif
3641 return 0;
3642}
3643
3644#endif /* __HAVE_ARCH_GATE_AREA */
David Howells0ec76a12006-09-27 01:50:15 -07003645
Namhyung Kim1b36ba82010-10-26 14:22:00 -07003646static int __follow_pte(struct mm_struct *mm, unsigned long address,
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07003647 pte_t **ptepp, spinlock_t **ptlp)
3648{
3649 pgd_t *pgd;
3650 pud_t *pud;
3651 pmd_t *pmd;
3652 pte_t *ptep;
3653
3654 pgd = pgd_offset(mm, address);
3655 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
3656 goto out;
3657
3658 pud = pud_offset(pgd, address);
3659 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
3660 goto out;
3661
3662 pmd = pmd_offset(pud, address);
Andrea Arcangelif66055a2011-01-13 15:46:54 -08003663 VM_BUG_ON(pmd_trans_huge(*pmd));
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07003664 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
3665 goto out;
3666
3667 /* We cannot handle huge page PFN maps. Luckily they don't exist. */
3668 if (pmd_huge(*pmd))
3669 goto out;
3670
3671 ptep = pte_offset_map_lock(mm, pmd, address, ptlp);
3672 if (!ptep)
3673 goto out;
3674 if (!pte_present(*ptep))
3675 goto unlock;
3676 *ptepp = ptep;
3677 return 0;
3678unlock:
3679 pte_unmap_unlock(ptep, *ptlp);
3680out:
3681 return -EINVAL;
3682}
3683
Namhyung Kim1b36ba82010-10-26 14:22:00 -07003684static inline int follow_pte(struct mm_struct *mm, unsigned long address,
3685 pte_t **ptepp, spinlock_t **ptlp)
3686{
3687 int res;
3688
3689 /* (void) is needed to make gcc happy */
3690 (void) __cond_lock(*ptlp,
3691 !(res = __follow_pte(mm, address, ptepp, ptlp)));
3692 return res;
3693}
3694
Johannes Weiner3b6748e2009-06-16 15:32:35 -07003695/**
3696 * follow_pfn - look up PFN at a user virtual address
3697 * @vma: memory mapping
3698 * @address: user virtual address
3699 * @pfn: location to store found PFN
3700 *
3701 * Only IO mappings and raw PFN mappings are allowed.
3702 *
3703 * Returns zero and the pfn at @pfn on success, -ve otherwise.
3704 */
3705int follow_pfn(struct vm_area_struct *vma, unsigned long address,
3706 unsigned long *pfn)
3707{
3708 int ret = -EINVAL;
3709 spinlock_t *ptl;
3710 pte_t *ptep;
3711
3712 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3713 return ret;
3714
3715 ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);
3716 if (ret)
3717 return ret;
3718 *pfn = pte_pfn(*ptep);
3719 pte_unmap_unlock(ptep, ptl);
3720 return 0;
3721}
3722EXPORT_SYMBOL(follow_pfn);
3723
Rik van Riel28b2ee22008-07-23 21:27:05 -07003724#ifdef CONFIG_HAVE_IOREMAP_PROT
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003725int follow_phys(struct vm_area_struct *vma,
3726 unsigned long address, unsigned int flags,
3727 unsigned long *prot, resource_size_t *phys)
Rik van Riel28b2ee22008-07-23 21:27:05 -07003728{
Johannes Weiner03668a42009-06-16 15:32:34 -07003729 int ret = -EINVAL;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003730 pte_t *ptep, pte;
3731 spinlock_t *ptl;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003732
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003733 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3734 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003735
Johannes Weiner03668a42009-06-16 15:32:34 -07003736 if (follow_pte(vma->vm_mm, address, &ptep, &ptl))
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003737 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003738 pte = *ptep;
Johannes Weiner03668a42009-06-16 15:32:34 -07003739
Rik van Riel28b2ee22008-07-23 21:27:05 -07003740 if ((flags & FOLL_WRITE) && !pte_write(pte))
3741 goto unlock;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003742
3743 *prot = pgprot_val(pte_pgprot(pte));
Johannes Weiner03668a42009-06-16 15:32:34 -07003744 *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003745
Johannes Weiner03668a42009-06-16 15:32:34 -07003746 ret = 0;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003747unlock:
3748 pte_unmap_unlock(ptep, ptl);
3749out:
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003750 return ret;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003751}
3752
3753int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
3754 void *buf, int len, int write)
3755{
3756 resource_size_t phys_addr;
3757 unsigned long prot = 0;
KOSAKI Motohiro2bc72732009-01-06 14:39:43 -08003758 void __iomem *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003759 int offset = addr & (PAGE_SIZE-1);
3760
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003761 if (follow_phys(vma, addr, write, &prot, &phys_addr))
Rik van Riel28b2ee22008-07-23 21:27:05 -07003762 return -EINVAL;
3763
3764 maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot);
3765 if (write)
3766 memcpy_toio(maddr + offset, buf, len);
3767 else
3768 memcpy_fromio(buf, maddr + offset, len);
3769 iounmap(maddr);
3770
3771 return len;
3772}
3773#endif
3774
David Howells0ec76a12006-09-27 01:50:15 -07003775/*
Stephen Wilson206cb632011-03-13 15:49:19 -04003776 * Access another process' address space as given in mm. If non-NULL, use the
3777 * given task for page fault accounting.
David Howells0ec76a12006-09-27 01:50:15 -07003778 */
Stephen Wilson206cb632011-03-13 15:49:19 -04003779static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
3780 unsigned long addr, void *buf, int len, int write)
David Howells0ec76a12006-09-27 01:50:15 -07003781{
David Howells0ec76a12006-09-27 01:50:15 -07003782 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07003783 void *old_buf = buf;
3784
David Howells0ec76a12006-09-27 01:50:15 -07003785 down_read(&mm->mmap_sem);
Simon Arlott183ff222007-10-20 01:27:18 +02003786 /* ignore errors, just check how much was successfully transferred */
David Howells0ec76a12006-09-27 01:50:15 -07003787 while (len) {
3788 int bytes, ret, offset;
3789 void *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003790 struct page *page = NULL;
David Howells0ec76a12006-09-27 01:50:15 -07003791
3792 ret = get_user_pages(tsk, mm, addr, 1,
3793 write, 1, &page, &vma);
Rik van Riel28b2ee22008-07-23 21:27:05 -07003794 if (ret <= 0) {
3795 /*
3796 * Check if this is a VM_IO | VM_PFNMAP VMA, which
3797 * we can access using slightly different code.
3798 */
3799#ifdef CONFIG_HAVE_IOREMAP_PROT
3800 vma = find_vma(mm, addr);
Michael Ellermanfe936df2011-04-14 15:22:10 -07003801 if (!vma || vma->vm_start > addr)
Rik van Riel28b2ee22008-07-23 21:27:05 -07003802 break;
3803 if (vma->vm_ops && vma->vm_ops->access)
3804 ret = vma->vm_ops->access(vma, addr, buf,
3805 len, write);
3806 if (ret <= 0)
3807#endif
3808 break;
3809 bytes = ret;
David Howells0ec76a12006-09-27 01:50:15 -07003810 } else {
Rik van Riel28b2ee22008-07-23 21:27:05 -07003811 bytes = len;
3812 offset = addr & (PAGE_SIZE-1);
3813 if (bytes > PAGE_SIZE-offset)
3814 bytes = PAGE_SIZE-offset;
3815
3816 maddr = kmap(page);
3817 if (write) {
3818 copy_to_user_page(vma, page, addr,
3819 maddr + offset, buf, bytes);
3820 set_page_dirty_lock(page);
3821 } else {
3822 copy_from_user_page(vma, page, addr,
3823 buf, maddr + offset, bytes);
3824 }
3825 kunmap(page);
3826 page_cache_release(page);
David Howells0ec76a12006-09-27 01:50:15 -07003827 }
David Howells0ec76a12006-09-27 01:50:15 -07003828 len -= bytes;
3829 buf += bytes;
3830 addr += bytes;
3831 }
3832 up_read(&mm->mmap_sem);
David Howells0ec76a12006-09-27 01:50:15 -07003833
3834 return buf - old_buf;
3835}
Andi Kleen03252912008-01-30 13:33:18 +01003836
Stephen Wilson5ddd36b2011-03-13 15:49:20 -04003837/**
Randy Dunlapae91dbf2011-03-26 13:27:01 -07003838 * access_remote_vm - access another process' address space
Stephen Wilson5ddd36b2011-03-13 15:49:20 -04003839 * @mm: the mm_struct of the target address space
3840 * @addr: start address to access
3841 * @buf: source or destination buffer
3842 * @len: number of bytes to transfer
3843 * @write: whether the access is a write
3844 *
3845 * The caller must hold a reference on @mm.
3846 */
3847int access_remote_vm(struct mm_struct *mm, unsigned long addr,
3848 void *buf, int len, int write)
3849{
3850 return __access_remote_vm(NULL, mm, addr, buf, len, write);
3851}
3852
Andi Kleen03252912008-01-30 13:33:18 +01003853/*
Stephen Wilson206cb632011-03-13 15:49:19 -04003854 * Access another process' address space.
3855 * Source/target buffer must be kernel space,
3856 * Do not walk the page table directly, use get_user_pages
3857 */
3858int access_process_vm(struct task_struct *tsk, unsigned long addr,
3859 void *buf, int len, int write)
3860{
3861 struct mm_struct *mm;
3862 int ret;
3863
3864 mm = get_task_mm(tsk);
3865 if (!mm)
3866 return 0;
3867
3868 ret = __access_remote_vm(tsk, mm, addr, buf, len, write);
3869 mmput(mm);
3870
3871 return ret;
3872}
3873
Andi Kleen03252912008-01-30 13:33:18 +01003874/*
3875 * Print the name of a VMA.
3876 */
3877void print_vma_addr(char *prefix, unsigned long ip)
3878{
3879 struct mm_struct *mm = current->mm;
3880 struct vm_area_struct *vma;
3881
Ingo Molnare8bff742008-02-13 20:21:06 +01003882 /*
3883 * Do not print if we are in atomic
3884 * contexts (in exception stacks, etc.):
3885 */
3886 if (preempt_count())
3887 return;
3888
Andi Kleen03252912008-01-30 13:33:18 +01003889 down_read(&mm->mmap_sem);
3890 vma = find_vma(mm, ip);
3891 if (vma && vma->vm_file) {
3892 struct file *f = vma->vm_file;
3893 char *buf = (char *)__get_free_page(GFP_KERNEL);
3894 if (buf) {
3895 char *p, *s;
3896
Jan Blunckcf28b482008-02-14 19:38:44 -08003897 p = d_path(&f->f_path, buf, PAGE_SIZE);
Andi Kleen03252912008-01-30 13:33:18 +01003898 if (IS_ERR(p))
3899 p = "?";
3900 s = strrchr(p, '/');
3901 if (s)
3902 p = s+1;
3903 printk("%s%s[%lx+%lx]", prefix, p,
3904 vma->vm_start,
3905 vma->vm_end - vma->vm_start);
3906 free_page((unsigned long)buf);
3907 }
3908 }
3909 up_read(&current->mm->mmap_sem);
3910}
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003911
3912#ifdef CONFIG_PROVE_LOCKING
3913void might_fault(void)
3914{
Peter Zijlstra95156f02009-01-12 13:02:11 +01003915 /*
3916 * Some code (nfs/sunrpc) uses socket ops on kernel memory while
3917 * holding the mmap_sem, this is safe because kernel memory doesn't
3918 * get paged out, therefore we'll never actually fault, and the
3919 * below annotations will generate false positives.
3920 */
3921 if (segment_eq(get_fs(), KERNEL_DS))
3922 return;
3923
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003924 might_sleep();
3925 /*
3926 * it would be nicer only to annotate paths which are not under
3927 * pagefault_disable, however that requires a larger audit and
3928 * providing helpers like get_user_atomic.
3929 */
3930 if (!in_atomic() && current->mm)
3931 might_lock_read(&current->mm->mmap_sem);
3932}
3933EXPORT_SYMBOL(might_fault);
3934#endif
Andrea Arcangeli47ad8472011-01-13 15:46:47 -08003935
3936#if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)
3937static void clear_gigantic_page(struct page *page,
3938 unsigned long addr,
3939 unsigned int pages_per_huge_page)
3940{
3941 int i;
3942 struct page *p = page;
3943
3944 might_sleep();
3945 for (i = 0; i < pages_per_huge_page;
3946 i++, p = mem_map_next(p, page, i)) {
3947 cond_resched();
3948 clear_user_highpage(p, addr + i * PAGE_SIZE);
3949 }
3950}
3951void clear_huge_page(struct page *page,
3952 unsigned long addr, unsigned int pages_per_huge_page)
3953{
3954 int i;
3955
3956 if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
3957 clear_gigantic_page(page, addr, pages_per_huge_page);
3958 return;
3959 }
3960
3961 might_sleep();
3962 for (i = 0; i < pages_per_huge_page; i++) {
3963 cond_resched();
3964 clear_user_highpage(page + i, addr + i * PAGE_SIZE);
3965 }
3966}
3967
3968static void copy_user_gigantic_page(struct page *dst, struct page *src,
3969 unsigned long addr,
3970 struct vm_area_struct *vma,
3971 unsigned int pages_per_huge_page)
3972{
3973 int i;
3974 struct page *dst_base = dst;
3975 struct page *src_base = src;
3976
3977 for (i = 0; i < pages_per_huge_page; ) {
3978 cond_resched();
3979 copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma);
3980
3981 i++;
3982 dst = mem_map_next(dst, dst_base, i);
3983 src = mem_map_next(src, src_base, i);
3984 }
3985}
3986
3987void copy_user_huge_page(struct page *dst, struct page *src,
3988 unsigned long addr, struct vm_area_struct *vma,
3989 unsigned int pages_per_huge_page)
3990{
3991 int i;
3992
3993 if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
3994 copy_user_gigantic_page(dst, src, addr, vma,
3995 pages_per_huge_page);
3996 return;
3997 }
3998
3999 might_sleep();
4000 for (i = 0; i < pages_per_huge_page; i++) {
4001 cond_resched();
4002 copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma);
4003 }
4004}
4005#endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */