blob: bde42c6d3633f15cfe51b1eae8e3e56833e86955 [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>
50#include <linux/module.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}
185#else
186
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
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800194#endif
195
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196/*
197 * If a p?d_bad entry is found while walking page tables, report
198 * the error, before resetting entry to p?d_none. Usually (but
199 * very seldom) called out from the p?d_none_or_clear_bad macros.
200 */
201
202void pgd_clear_bad(pgd_t *pgd)
203{
204 pgd_ERROR(*pgd);
205 pgd_clear(pgd);
206}
207
208void pud_clear_bad(pud_t *pud)
209{
210 pud_ERROR(*pud);
211 pud_clear(pud);
212}
213
214void pmd_clear_bad(pmd_t *pmd)
215{
216 pmd_ERROR(*pmd);
217 pmd_clear(pmd);
218}
219
220/*
221 * Note: this doesn't free the actual pages themselves. That
222 * has been handled earlier when unmapping all the memory regions.
223 */
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000224static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
225 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800227 pgtable_t token = pmd_pgtable(*pmd);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700228 pmd_clear(pmd);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000229 pte_free_tlb(tlb, token, addr);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700230 tlb->mm->nr_ptes--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231}
232
Hugh Dickinse0da3822005-04-19 13:29:15 -0700233static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
234 unsigned long addr, unsigned long end,
235 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236{
237 pmd_t *pmd;
238 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700239 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240
Hugh Dickinse0da3822005-04-19 13:29:15 -0700241 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 do {
244 next = pmd_addr_end(addr, end);
245 if (pmd_none_or_clear_bad(pmd))
246 continue;
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000247 free_pte_range(tlb, pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 } while (pmd++, addr = next, addr != end);
249
Hugh Dickinse0da3822005-04-19 13:29:15 -0700250 start &= PUD_MASK;
251 if (start < floor)
252 return;
253 if (ceiling) {
254 ceiling &= PUD_MASK;
255 if (!ceiling)
256 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700258 if (end - 1 > ceiling - 1)
259 return;
260
261 pmd = pmd_offset(pud, start);
262 pud_clear(pud);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000263 pmd_free_tlb(tlb, pmd, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264}
265
Hugh Dickinse0da3822005-04-19 13:29:15 -0700266static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
267 unsigned long addr, unsigned long end,
268 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269{
270 pud_t *pud;
271 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700272 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273
Hugh Dickinse0da3822005-04-19 13:29:15 -0700274 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 pud = pud_offset(pgd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 do {
277 next = pud_addr_end(addr, end);
278 if (pud_none_or_clear_bad(pud))
279 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700280 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 } while (pud++, addr = next, addr != end);
282
Hugh Dickinse0da3822005-04-19 13:29:15 -0700283 start &= PGDIR_MASK;
284 if (start < floor)
285 return;
286 if (ceiling) {
287 ceiling &= PGDIR_MASK;
288 if (!ceiling)
289 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700291 if (end - 1 > ceiling - 1)
292 return;
293
294 pud = pud_offset(pgd, start);
295 pgd_clear(pgd);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000296 pud_free_tlb(tlb, pud, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297}
298
299/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700300 * This function frees user-level page tables of a process.
301 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 * Must be called with pagetable lock held.
303 */
Jan Beulich42b77722008-07-23 21:27:10 -0700304void free_pgd_range(struct mmu_gather *tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700305 unsigned long addr, unsigned long end,
306 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307{
308 pgd_t *pgd;
309 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700310 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311
Hugh Dickinse0da3822005-04-19 13:29:15 -0700312 /*
313 * The next few lines have given us lots of grief...
314 *
315 * Why are we testing PMD* at this top level? Because often
316 * there will be no work to do at all, and we'd prefer not to
317 * go all the way down to the bottom just to discover that.
318 *
319 * Why all these "- 1"s? Because 0 represents both the bottom
320 * of the address space and the top of it (using -1 for the
321 * top wouldn't help much: the masks would do the wrong thing).
322 * The rule is that addr 0 and floor 0 refer to the bottom of
323 * the address space, but end 0 and ceiling 0 refer to the top
324 * Comparisons need to use "end - 1" and "ceiling - 1" (though
325 * that end 0 case should be mythical).
326 *
327 * Wherever addr is brought up or ceiling brought down, we must
328 * be careful to reject "the opposite 0" before it confuses the
329 * subsequent tests. But what about where end is brought down
330 * by PMD_SIZE below? no, end can't go down to 0 there.
331 *
332 * Whereas we round start (addr) and ceiling down, by different
333 * masks at different levels, in order to test whether a table
334 * now has no other vmas using it, so can be freed, we don't
335 * bother to round floor or end up - the tests don't need that.
336 */
337
338 addr &= PMD_MASK;
339 if (addr < floor) {
340 addr += PMD_SIZE;
341 if (!addr)
342 return;
343 }
344 if (ceiling) {
345 ceiling &= PMD_MASK;
346 if (!ceiling)
347 return;
348 }
349 if (end - 1 > ceiling - 1)
350 end -= PMD_SIZE;
351 if (addr > end - 1)
352 return;
353
354 start = addr;
Jan Beulich42b77722008-07-23 21:27:10 -0700355 pgd = pgd_offset(tlb->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 do {
357 next = pgd_addr_end(addr, end);
358 if (pgd_none_or_clear_bad(pgd))
359 continue;
Jan Beulich42b77722008-07-23 21:27:10 -0700360 free_pud_range(tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700362}
363
Jan Beulich42b77722008-07-23 21:27:10 -0700364void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700365 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700366{
367 while (vma) {
368 struct vm_area_struct *next = vma->vm_next;
369 unsigned long addr = vma->vm_start;
370
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700371 /*
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000372 * Hide vma from rmap and truncate_pagecache before freeing
373 * pgtables
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700374 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800375 unlink_anon_vmas(vma);
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700376 unlink_file_vma(vma);
377
David Gibson9da61ae2006-03-22 00:08:57 -0800378 if (is_vm_hugetlb_page(vma)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700379 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700380 floor, next? next->vm_start: ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700381 } else {
382 /*
383 * Optimization: gather nearby vmas into one call down
384 */
385 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
David Gibson48669202006-03-22 00:08:58 -0800386 && !is_vm_hugetlb_page(next)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700387 vma = next;
388 next = vma->vm_next;
Rik van Riel5beb4932010-03-05 13:42:07 -0800389 unlink_anon_vmas(vma);
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700390 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700391 }
392 free_pgd_range(tlb, addr, vma->vm_end,
393 floor, next? next->vm_start: ceiling);
394 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700395 vma = next;
396 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397}
398
Hugh Dickins1bb36302005-10-29 18:16:22 -0700399int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800401 pgtable_t new = pte_alloc_one(mm, address);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700402 if (!new)
403 return -ENOMEM;
404
Nick Piggin362a61a2008-05-14 06:37:36 +0200405 /*
406 * Ensure all pte setup (eg. pte page lock and page clearing) are
407 * visible before the pte is made visible to other CPUs by being
408 * put into page tables.
409 *
410 * The other side of the story is the pointer chasing in the page
411 * table walking code (when walking the page table without locking;
412 * ie. most of the time). Fortunately, these data accesses consist
413 * of a chain of data-dependent loads, meaning most CPUs (alpha
414 * being the notable exception) will already guarantee loads are
415 * seen in-order. See the alpha page table accessors for the
416 * smp_read_barrier_depends() barriers in page table walking code.
417 */
418 smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
419
Hugh Dickinsc74df322005-10-29 18:16:23 -0700420 spin_lock(&mm->page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800421 if (!pmd_present(*pmd)) { /* Has another populated it ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 mm->nr_ptes++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 pmd_populate(mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800424 new = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 }
Hugh Dickinsc74df322005-10-29 18:16:23 -0700426 spin_unlock(&mm->page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800427 if (new)
428 pte_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700429 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430}
431
Hugh Dickins1bb36302005-10-29 18:16:22 -0700432int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700434 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
435 if (!new)
436 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437
Nick Piggin362a61a2008-05-14 06:37:36 +0200438 smp_wmb(); /* See comment in __pte_alloc */
439
Hugh Dickins1bb36302005-10-29 18:16:22 -0700440 spin_lock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800441 if (!pmd_present(*pmd)) { /* Has another populated it ? */
Hugh Dickins1bb36302005-10-29 18:16:22 -0700442 pmd_populate_kernel(&init_mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800443 new = NULL;
444 }
Hugh Dickins1bb36302005-10-29 18:16:22 -0700445 spin_unlock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800446 if (new)
447 pte_free_kernel(&init_mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700448 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449}
450
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800451static inline void init_rss_vec(int *rss)
Hugh Dickinsae859762005-10-29 18:16:05 -0700452{
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800453 memset(rss, 0, sizeof(int) * NR_MM_COUNTERS);
454}
455
456static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss)
457{
458 int i;
459
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800460 if (current->mm == mm)
461 sync_mm_rss(current, mm);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800462 for (i = 0; i < NR_MM_COUNTERS; i++)
463 if (rss[i])
464 add_mm_counter(mm, i, rss[i]);
Hugh Dickinsae859762005-10-29 18:16:05 -0700465}
466
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800468 * This function is called to print an error when a bad pte
469 * is found. For example, we might have a PFN-mapped pte in
470 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700471 *
472 * The calling function must still handle the error.
473 */
Hugh Dickins3dc14742009-01-06 14:40:08 -0800474static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
475 pte_t pte, struct page *page)
Nick Pigginb5810032005-10-29 18:16:12 -0700476{
Hugh Dickins3dc14742009-01-06 14:40:08 -0800477 pgd_t *pgd = pgd_offset(vma->vm_mm, addr);
478 pud_t *pud = pud_offset(pgd, addr);
479 pmd_t *pmd = pmd_offset(pud, addr);
480 struct address_space *mapping;
481 pgoff_t index;
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800482 static unsigned long resume;
483 static unsigned long nr_shown;
484 static unsigned long nr_unshown;
485
486 /*
487 * Allow a burst of 60 reports, then keep quiet for that minute;
488 * or allow a steady drip of one report per second.
489 */
490 if (nr_shown == 60) {
491 if (time_before(jiffies, resume)) {
492 nr_unshown++;
493 return;
494 }
495 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800496 printk(KERN_ALERT
497 "BUG: Bad page map: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800498 nr_unshown);
499 nr_unshown = 0;
500 }
501 nr_shown = 0;
502 }
503 if (nr_shown++ == 0)
504 resume = jiffies + 60 * HZ;
Hugh Dickins3dc14742009-01-06 14:40:08 -0800505
506 mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;
507 index = linear_page_index(vma, addr);
508
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800509 printk(KERN_ALERT
510 "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800511 current->comm,
512 (long long)pte_val(pte), (long long)pmd_val(*pmd));
Wu Fengguang718a3822010-03-10 15:20:43 -0800513 if (page)
514 dump_page(page);
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800515 printk(KERN_ALERT
Hugh Dickins3dc14742009-01-06 14:40:08 -0800516 "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
517 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
518 /*
519 * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y
520 */
521 if (vma->vm_ops)
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800522 print_symbol(KERN_ALERT "vma->vm_ops->fault: %s\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800523 (unsigned long)vma->vm_ops->fault);
524 if (vma->vm_file && vma->vm_file->f_op)
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800525 print_symbol(KERN_ALERT "vma->vm_file->f_op->mmap: %s\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800526 (unsigned long)vma->vm_file->f_op->mmap);
Nick Pigginb5810032005-10-29 18:16:12 -0700527 dump_stack();
Hugh Dickins3dc14742009-01-06 14:40:08 -0800528 add_taint(TAINT_BAD_PAGE);
Nick Pigginb5810032005-10-29 18:16:12 -0700529}
530
Linus Torvalds67121172005-12-11 20:38:17 -0800531static inline int is_cow_mapping(unsigned int flags)
532{
533 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
534}
535
Hugh Dickins62eede62009-09-21 17:03:34 -0700536#ifndef is_zero_pfn
537static inline int is_zero_pfn(unsigned long pfn)
538{
539 return pfn == zero_pfn;
540}
541#endif
542
543#ifndef my_zero_pfn
544static inline unsigned long my_zero_pfn(unsigned long addr)
545{
546 return zero_pfn;
547}
548#endif
549
Nick Pigginb5810032005-10-29 18:16:12 -0700550/*
Nick Piggin7e675132008-04-28 02:13:00 -0700551 * vm_normal_page -- This function gets the "struct page" associated with a pte.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800552 *
Nick Piggin7e675132008-04-28 02:13:00 -0700553 * "Special" mappings do not wish to be associated with a "struct page" (either
554 * it doesn't exist, or it exists but they don't want to touch it). In this
555 * case, NULL is returned here. "Normal" mappings do have a struct page.
Jared Hulbertb379d792008-04-28 02:12:58 -0700556 *
Nick Piggin7e675132008-04-28 02:13:00 -0700557 * There are 2 broad cases. Firstly, an architecture may define a pte_special()
558 * pte bit, in which case this function is trivial. Secondly, an architecture
559 * may not have a spare pte bit, which requires a more complicated scheme,
560 * described below.
561 *
562 * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
563 * special mapping (even if there are underlying and valid "struct pages").
564 * COWed pages of a VM_PFNMAP are always normal.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800565 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700566 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
567 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
Nick Piggin7e675132008-04-28 02:13:00 -0700568 * set, and the vm_pgoff will point to the first PFN mapped: thus every special
569 * mapping will always honor the rule
Linus Torvalds6aab3412005-11-28 14:34:23 -0800570 *
571 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
572 *
Nick Piggin7e675132008-04-28 02:13:00 -0700573 * And for normal mappings this is false.
Jared Hulbertb379d792008-04-28 02:12:58 -0700574 *
Nick Piggin7e675132008-04-28 02:13:00 -0700575 * This restricts such mappings to be a linear translation from virtual address
576 * to pfn. To get around this restriction, we allow arbitrary mappings so long
577 * as the vma is not a COW mapping; in that case, we know that all ptes are
578 * special (because none can have been COWed).
Jared Hulbertb379d792008-04-28 02:12:58 -0700579 *
580 *
Nick Piggin7e675132008-04-28 02:13:00 -0700581 * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
582 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700583 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
584 * page" backing, however the difference is that _all_ pages with a struct
585 * page (that is, those where pfn_valid is true) are refcounted and considered
586 * normal pages by the VM. The disadvantage is that pages are refcounted
587 * (which can be slower and simply not an option for some PFNMAP users). The
588 * advantage is that we don't have to follow the strict linearity rule of
589 * PFNMAP mappings in order to support COWable mappings.
590 *
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800591 */
Nick Piggin7e675132008-04-28 02:13:00 -0700592#ifdef __HAVE_ARCH_PTE_SPECIAL
593# define HAVE_PTE_SPECIAL 1
594#else
595# define HAVE_PTE_SPECIAL 0
596#endif
597struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
598 pte_t pte)
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800599{
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800600 unsigned long pfn = pte_pfn(pte);
Nick Piggin7e675132008-04-28 02:13:00 -0700601
602 if (HAVE_PTE_SPECIAL) {
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800603 if (likely(!pte_special(pte)))
604 goto check_pfn;
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -0700605 if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))
606 return NULL;
Hugh Dickins62eede62009-09-21 17:03:34 -0700607 if (!is_zero_pfn(pfn))
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800608 print_bad_pte(vma, addr, pte, NULL);
Nick Piggin7e675132008-04-28 02:13:00 -0700609 return NULL;
610 }
611
612 /* !HAVE_PTE_SPECIAL case follows: */
613
Jared Hulbertb379d792008-04-28 02:12:58 -0700614 if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
615 if (vma->vm_flags & VM_MIXEDMAP) {
616 if (!pfn_valid(pfn))
617 return NULL;
618 goto out;
619 } else {
Nick Piggin7e675132008-04-28 02:13:00 -0700620 unsigned long off;
621 off = (addr - vma->vm_start) >> PAGE_SHIFT;
Jared Hulbertb379d792008-04-28 02:12:58 -0700622 if (pfn == vma->vm_pgoff + off)
623 return NULL;
624 if (!is_cow_mapping(vma->vm_flags))
625 return NULL;
626 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800627 }
628
Hugh Dickins62eede62009-09-21 17:03:34 -0700629 if (is_zero_pfn(pfn))
630 return NULL;
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800631check_pfn:
632 if (unlikely(pfn > highest_memmap_pfn)) {
633 print_bad_pte(vma, addr, pte, NULL);
634 return NULL;
635 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800636
637 /*
Nick Piggin7e675132008-04-28 02:13:00 -0700638 * NOTE! We still have PageReserved() pages in the page tables.
Nick Piggin7e675132008-04-28 02:13:00 -0700639 * eg. VDSO mappings can cause them to exist.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800640 */
Jared Hulbertb379d792008-04-28 02:12:58 -0700641out:
Linus Torvalds6aab3412005-11-28 14:34:23 -0800642 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800643}
644
645/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 * copy one vm_area from one task to the other. Assumes the page tables
647 * already present in the new task to be cleared in the whole range
648 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 */
650
Hugh Dickins570a335b2009-12-14 17:58:46 -0800651static inline unsigned long
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700653 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700654 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655{
Nick Pigginb5810032005-10-29 18:16:12 -0700656 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 pte_t pte = *src_pte;
658 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
660 /* pte contains position in swap or file, so copy. */
661 if (unlikely(!pte_present(pte))) {
662 if (!pte_file(pte)) {
Christoph Lameter06972122006-06-23 02:03:35 -0700663 swp_entry_t entry = pte_to_swp_entry(pte);
664
Hugh Dickins570a335b2009-12-14 17:58:46 -0800665 if (swap_duplicate(entry) < 0)
666 return entry.val;
667
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 /* make sure dst_mm is on swapoff's mmlist. */
669 if (unlikely(list_empty(&dst_mm->mmlist))) {
670 spin_lock(&mmlist_lock);
Hugh Dickinsf412ac02005-10-29 18:16:41 -0700671 if (list_empty(&dst_mm->mmlist))
672 list_add(&dst_mm->mmlist,
673 &src_mm->mmlist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 spin_unlock(&mmlist_lock);
675 }
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -0800676 if (likely(!non_swap_entry(entry)))
677 rss[MM_SWAPENTS]++;
678 else if (is_write_migration_entry(entry) &&
Christoph Lameter06972122006-06-23 02:03:35 -0700679 is_cow_mapping(vm_flags)) {
680 /*
681 * COW mappings require pages in both parent
682 * and child to be set to read.
683 */
684 make_migration_entry_read(&entry);
685 pte = swp_entry_to_pte(entry);
686 set_pte_at(src_mm, addr, src_pte, pte);
687 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700689 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 }
691
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 /*
693 * If it's a COW mapping, write protect it both
694 * in the parent and the child
695 */
Linus Torvalds67121172005-12-11 20:38:17 -0800696 if (is_cow_mapping(vm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 ptep_set_wrprotect(src_mm, addr, src_pte);
Zachary Amsden3dc90792006-09-30 23:29:30 -0700698 pte = pte_wrprotect(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 }
700
701 /*
702 * If it's a shared mapping, mark it clean in
703 * the child
704 */
705 if (vm_flags & VM_SHARED)
706 pte = pte_mkclean(pte);
707 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800708
709 page = vm_normal_page(vma, addr, pte);
710 if (page) {
711 get_page(page);
Hugh Dickins21333b22009-09-21 17:01:59 -0700712 page_dup_rmap(page);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800713 if (PageAnon(page))
714 rss[MM_ANONPAGES]++;
715 else
716 rss[MM_FILEPAGES]++;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800717 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700718
719out_set_pte:
720 set_pte_at(dst_mm, addr, dst_pte, pte);
Hugh Dickins570a335b2009-12-14 17:58:46 -0800721 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722}
723
724static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
725 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
726 unsigned long addr, unsigned long end)
727{
Daisuke Nishimurac36987e2009-10-26 16:50:23 -0700728 pte_t *orig_src_pte, *orig_dst_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700730 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -0700731 int progress = 0;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800732 int rss[NR_MM_COUNTERS];
Hugh Dickins570a335b2009-12-14 17:58:46 -0800733 swp_entry_t entry = (swp_entry_t){0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734
735again:
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800736 init_rss_vec(rss);
737
Hugh Dickinsc74df322005-10-29 18:16:23 -0700738 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 if (!dst_pte)
740 return -ENOMEM;
741 src_pte = pte_offset_map_nested(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700742 src_ptl = pte_lockptr(src_mm, src_pmd);
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700743 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
Daisuke Nishimurac36987e2009-10-26 16:50:23 -0700744 orig_src_pte = src_pte;
745 orig_dst_pte = dst_pte;
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700746 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 do {
749 /*
750 * We are holding two locks at this point - either of them
751 * could generate latencies in another task on another CPU.
752 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700753 if (progress >= 32) {
754 progress = 0;
755 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +0100756 spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700757 break;
758 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 if (pte_none(*src_pte)) {
760 progress++;
761 continue;
762 }
Hugh Dickins570a335b2009-12-14 17:58:46 -0800763 entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte,
764 vma, addr, rss);
765 if (entry.val)
766 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 progress += 8;
768 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700770 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -0700771 spin_unlock(src_ptl);
Daisuke Nishimurac36987e2009-10-26 16:50:23 -0700772 pte_unmap_nested(orig_src_pte);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800773 add_mm_rss_vec(dst_mm, rss);
Daisuke Nishimurac36987e2009-10-26 16:50:23 -0700774 pte_unmap_unlock(orig_dst_pte, dst_ptl);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700775 cond_resched();
Hugh Dickins570a335b2009-12-14 17:58:46 -0800776
777 if (entry.val) {
778 if (add_swap_count_continuation(entry, GFP_KERNEL) < 0)
779 return -ENOMEM;
780 progress = 0;
781 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 if (addr != end)
783 goto again;
784 return 0;
785}
786
787static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
788 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
789 unsigned long addr, unsigned long end)
790{
791 pmd_t *src_pmd, *dst_pmd;
792 unsigned long next;
793
794 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
795 if (!dst_pmd)
796 return -ENOMEM;
797 src_pmd = pmd_offset(src_pud, addr);
798 do {
799 next = pmd_addr_end(addr, end);
800 if (pmd_none_or_clear_bad(src_pmd))
801 continue;
802 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
803 vma, addr, next))
804 return -ENOMEM;
805 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
806 return 0;
807}
808
809static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
810 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
811 unsigned long addr, unsigned long end)
812{
813 pud_t *src_pud, *dst_pud;
814 unsigned long next;
815
816 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
817 if (!dst_pud)
818 return -ENOMEM;
819 src_pud = pud_offset(src_pgd, addr);
820 do {
821 next = pud_addr_end(addr, end);
822 if (pud_none_or_clear_bad(src_pud))
823 continue;
824 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
825 vma, addr, next))
826 return -ENOMEM;
827 } while (dst_pud++, src_pud++, addr = next, addr != end);
828 return 0;
829}
830
831int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
832 struct vm_area_struct *vma)
833{
834 pgd_t *src_pgd, *dst_pgd;
835 unsigned long next;
836 unsigned long addr = vma->vm_start;
837 unsigned long end = vma->vm_end;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700838 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839
Nick Piggind9928952005-08-28 16:49:11 +1000840 /*
841 * Don't copy ptes where a page fault will fill them correctly.
842 * Fork becomes much lighter when there are big shared or private
843 * readonly mappings. The tradeoff is that copy_page_range is more
844 * efficient than faulting.
845 */
Linus Torvalds4d7672b2005-12-16 10:21:23 -0800846 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
Nick Piggind9928952005-08-28 16:49:11 +1000847 if (!vma->anon_vma)
848 return 0;
849 }
850
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 if (is_vm_hugetlb_page(vma))
852 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
853
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800854 if (unlikely(is_pfn_mapping(vma))) {
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -0800855 /*
856 * We do not free on error cases below as remove_vma
857 * gets called on error from higher level routine
858 */
859 ret = track_pfn_vma_copy(vma);
860 if (ret)
861 return ret;
862 }
863
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700864 /*
865 * We need to invalidate the secondary MMU mappings only when
866 * there could be a permission downgrade on the ptes of the
867 * parent mm. And a permission downgrade will only happen if
868 * is_cow_mapping() returns true.
869 */
870 if (is_cow_mapping(vma->vm_flags))
871 mmu_notifier_invalidate_range_start(src_mm, addr, end);
872
873 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 dst_pgd = pgd_offset(dst_mm, addr);
875 src_pgd = pgd_offset(src_mm, addr);
876 do {
877 next = pgd_addr_end(addr, end);
878 if (pgd_none_or_clear_bad(src_pgd))
879 continue;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700880 if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
881 vma, addr, next))) {
882 ret = -ENOMEM;
883 break;
884 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700886
887 if (is_cow_mapping(vma->vm_flags))
888 mmu_notifier_invalidate_range_end(src_mm,
889 vma->vm_start, end);
890 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891}
892
Robin Holt51c6f662005-11-13 16:06:42 -0800893static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700894 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800896 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897{
Nick Pigginb5810032005-10-29 18:16:12 -0700898 struct mm_struct *mm = tlb->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 pte_t *pte;
Hugh Dickins508034a2005-10-29 18:16:30 -0700900 spinlock_t *ptl;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800901 int rss[NR_MM_COUNTERS];
902
903 init_rss_vec(rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904
Hugh Dickins508034a2005-10-29 18:16:30 -0700905 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700906 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 do {
908 pte_t ptent = *pte;
Robin Holt51c6f662005-11-13 16:06:42 -0800909 if (pte_none(ptent)) {
910 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800912 }
Hugh Dickins6f5e6b92006-03-16 23:04:09 -0800913
914 (*zap_work) -= PAGE_SIZE;
915
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800917 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -0800918
Linus Torvalds6aab3412005-11-28 14:34:23 -0800919 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 if (unlikely(details) && page) {
921 /*
922 * unmap_shared_mapping_pages() wants to
923 * invalidate cache without truncating:
924 * unmap shared but keep private pages.
925 */
926 if (details->check_mapping &&
927 details->check_mapping != page->mapping)
928 continue;
929 /*
930 * Each page->index must be checked when
931 * invalidating or truncating nonlinear.
932 */
933 if (details->nonlinear_vma &&
934 (page->index < details->first_index ||
935 page->index > details->last_index))
936 continue;
937 }
Nick Pigginb5810032005-10-29 18:16:12 -0700938 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -0700939 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 tlb_remove_tlb_entry(tlb, pte, addr);
941 if (unlikely(!page))
942 continue;
943 if (unlikely(details) && details->nonlinear_vma
944 && linear_page_index(details->nonlinear_vma,
945 addr) != page->index)
Nick Pigginb5810032005-10-29 18:16:12 -0700946 set_pte_at(mm, addr, pte,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 pgoff_to_pte(page->index));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 if (PageAnon(page))
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800949 rss[MM_ANONPAGES]--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700950 else {
951 if (pte_dirty(ptent))
952 set_page_dirty(page);
Johannes Weiner4917e5d2009-01-06 14:39:17 -0800953 if (pte_young(ptent) &&
954 likely(!VM_SequentialReadHint(vma)))
Nick Pigginbf3f3bc2009-01-06 14:38:55 -0800955 mark_page_accessed(page);
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800956 rss[MM_FILEPAGES]--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700957 }
Hugh Dickinsedc315f2009-01-06 14:40:11 -0800958 page_remove_rmap(page);
Hugh Dickins3dc14742009-01-06 14:40:08 -0800959 if (unlikely(page_mapcount(page) < 0))
960 print_bad_pte(vma, addr, ptent, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 tlb_remove_page(tlb, page);
962 continue;
963 }
964 /*
965 * If details->check_mapping, we leave swap entries;
966 * if details->nonlinear_vma, we leave file entries.
967 */
968 if (unlikely(details))
969 continue;
Hugh Dickins2509ef22009-01-06 14:40:10 -0800970 if (pte_file(ptent)) {
971 if (unlikely(!(vma->vm_flags & VM_NONLINEAR)))
972 print_bad_pte(vma, addr, ptent, NULL);
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -0800973 } else {
974 swp_entry_t entry = pte_to_swp_entry(ptent);
975
976 if (!non_swap_entry(entry))
977 rss[MM_SWAPENTS]--;
978 if (unlikely(!free_swap_and_cache(entry)))
979 print_bad_pte(vma, addr, ptent, NULL);
980 }
Zachary Amsden9888a1c2006-09-30 23:29:31 -0700981 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
Robin Holt51c6f662005-11-13 16:06:42 -0800982 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
Hugh Dickinsae859762005-10-29 18:16:05 -0700983
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -0800984 add_mm_rss_vec(mm, rss);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700985 arch_leave_lazy_mmu_mode();
Hugh Dickins508034a2005-10-29 18:16:30 -0700986 pte_unmap_unlock(pte - 1, ptl);
Robin Holt51c6f662005-11-13 16:06:42 -0800987
988 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989}
990
Robin Holt51c6f662005-11-13 16:06:42 -0800991static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700992 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800994 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995{
996 pmd_t *pmd;
997 unsigned long next;
998
999 pmd = pmd_offset(pud, addr);
1000 do {
1001 next = pmd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -08001002 if (pmd_none_or_clear_bad(pmd)) {
1003 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 continue;
Robin Holt51c6f662005-11-13 16:06:42 -08001005 }
1006 next = zap_pte_range(tlb, vma, pmd, addr, next,
1007 zap_work, details);
1008 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
1009
1010 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011}
1012
Robin Holt51c6f662005-11-13 16:06:42 -08001013static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -07001014 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -08001016 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017{
1018 pud_t *pud;
1019 unsigned long next;
1020
1021 pud = pud_offset(pgd, addr);
1022 do {
1023 next = pud_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -08001024 if (pud_none_or_clear_bad(pud)) {
1025 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 continue;
Robin Holt51c6f662005-11-13 16:06:42 -08001027 }
1028 next = zap_pmd_range(tlb, vma, pud, addr, next,
1029 zap_work, details);
1030 } while (pud++, addr = next, (addr != end && *zap_work > 0));
1031
1032 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033}
1034
Robin Holt51c6f662005-11-13 16:06:42 -08001035static unsigned long unmap_page_range(struct mmu_gather *tlb,
1036 struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -08001038 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039{
1040 pgd_t *pgd;
1041 unsigned long next;
1042
1043 if (details && !details->check_mapping && !details->nonlinear_vma)
1044 details = NULL;
1045
1046 BUG_ON(addr >= end);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08001047 mem_cgroup_uncharge_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 tlb_start_vma(tlb, vma);
1049 pgd = pgd_offset(vma->vm_mm, addr);
1050 do {
1051 next = pgd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -08001052 if (pgd_none_or_clear_bad(pgd)) {
1053 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 continue;
Robin Holt51c6f662005-11-13 16:06:42 -08001055 }
1056 next = zap_pud_range(tlb, vma, pgd, addr, next,
1057 zap_work, details);
1058 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 tlb_end_vma(tlb, vma);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -08001060 mem_cgroup_uncharge_end();
Robin Holt51c6f662005-11-13 16:06:42 -08001061
1062 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063}
1064
1065#ifdef CONFIG_PREEMPT
1066# define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
1067#else
1068/* No preempt: go for improved straight-line efficiency */
1069# define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
1070#endif
1071
1072/**
1073 * unmap_vmas - unmap a range of memory covered by a list of vma's
1074 * @tlbp: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 * @vma: the starting vma
1076 * @start_addr: virtual address at which to start unmapping
1077 * @end_addr: virtual address at which to end unmapping
1078 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
1079 * @details: details of nonlinear truncation or shared cache invalidation
1080 *
Hugh Dickinsee39b372005-04-19 13:29:15 -07001081 * Returns the end address of the unmapping (restart addr if interrupted).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 *
Hugh Dickins508034a2005-10-29 18:16:30 -07001083 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 *
Hugh Dickins508034a2005-10-29 18:16:30 -07001085 * We aim to not hold locks for too long (for scheduling latency reasons).
1086 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 * return the ending mmu_gather to the caller.
1088 *
1089 * Only addresses between `start' and `end' will be unmapped.
1090 *
1091 * The VMA list must be sorted in ascending virtual address order.
1092 *
1093 * unmap_vmas() assumes that the caller will flush the whole unmapped address
1094 * range after unmap_vmas() returns. So the only responsibility here is to
1095 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
1096 * drops the lock and schedules.
1097 */
Hugh Dickins508034a2005-10-29 18:16:30 -07001098unsigned long unmap_vmas(struct mmu_gather **tlbp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 struct vm_area_struct *vma, unsigned long start_addr,
1100 unsigned long end_addr, unsigned long *nr_accounted,
1101 struct zap_details *details)
1102{
Robin Holt51c6f662005-11-13 16:06:42 -08001103 long zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
1105 int tlb_start_valid = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -07001106 unsigned long start = start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -07001108 int fullmm = (*tlbp)->fullmm;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001109 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001111 mmu_notifier_invalidate_range_start(mm, start_addr, end_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 unsigned long end;
1114
1115 start = max(vma->vm_start, start_addr);
1116 if (start >= vma->vm_end)
1117 continue;
1118 end = min(vma->vm_end, end_addr);
1119 if (end <= vma->vm_start)
1120 continue;
1121
1122 if (vma->vm_flags & VM_ACCOUNT)
1123 *nr_accounted += (end - start) >> PAGE_SHIFT;
1124
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -08001125 if (unlikely(is_pfn_mapping(vma)))
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001126 untrack_pfn_vma(vma, 0, 0);
1127
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 while (start != end) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 if (!tlb_start_valid) {
1130 tlb_start = start;
1131 tlb_start_valid = 1;
1132 }
1133
Robin Holt51c6f662005-11-13 16:06:42 -08001134 if (unlikely(is_vm_hugetlb_page(vma))) {
Andi Kleena137e1c2008-07-23 21:27:43 -07001135 /*
1136 * It is undesirable to test vma->vm_file as it
1137 * should be non-null for valid hugetlb area.
1138 * However, vm_file will be NULL in the error
1139 * cleanup path of do_mmap_pgoff. When
1140 * hugetlbfs ->mmap method fails,
1141 * do_mmap_pgoff() nullifies vma->vm_file
1142 * before calling this function to clean up.
1143 * Since no pte has actually been setup, it is
1144 * safe to do nothing in this case.
1145 */
1146 if (vma->vm_file) {
1147 unmap_hugepage_range(vma, start, end, NULL);
1148 zap_work -= (end - start) /
Andi Kleena5516432008-07-23 21:27:41 -07001149 pages_per_huge_page(hstate_vma(vma));
Andi Kleena137e1c2008-07-23 21:27:43 -07001150 }
1151
Robin Holt51c6f662005-11-13 16:06:42 -08001152 start = end;
1153 } else
1154 start = unmap_page_range(*tlbp, vma,
1155 start, end, &zap_work, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156
Robin Holt51c6f662005-11-13 16:06:42 -08001157 if (zap_work > 0) {
1158 BUG_ON(start != end);
1159 break;
1160 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161
1162 tlb_finish_mmu(*tlbp, tlb_start, start);
1163
1164 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +01001165 (i_mmap_lock && spin_needbreak(i_mmap_lock))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 if (i_mmap_lock) {
Hugh Dickins508034a2005-10-29 18:16:30 -07001167 *tlbp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 goto out;
1169 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 }
1172
Hugh Dickins508034a2005-10-29 18:16:30 -07001173 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 tlb_start_valid = 0;
Robin Holt51c6f662005-11-13 16:06:42 -08001175 zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 }
1177 }
1178out:
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001179 mmu_notifier_invalidate_range_end(mm, start_addr, end_addr);
Hugh Dickinsee39b372005-04-19 13:29:15 -07001180 return start; /* which is now the end (or restart) address */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181}
1182
1183/**
1184 * zap_page_range - remove user pages in a given range
1185 * @vma: vm_area_struct holding the applicable pages
1186 * @address: starting address of pages to zap
1187 * @size: number of bytes to zap
1188 * @details: details of nonlinear truncation or shared cache invalidation
1189 */
Hugh Dickinsee39b372005-04-19 13:29:15 -07001190unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 unsigned long size, struct zap_details *details)
1192{
1193 struct mm_struct *mm = vma->vm_mm;
1194 struct mmu_gather *tlb;
1195 unsigned long end = address + size;
1196 unsigned long nr_accounted = 0;
1197
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001200 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -07001201 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
1202 if (tlb)
1203 tlb_finish_mmu(tlb, address, end);
Hugh Dickinsee39b372005-04-19 13:29:15 -07001204 return end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205}
1206
Jack Steinerc627f9c2008-07-29 22:33:53 -07001207/**
1208 * zap_vma_ptes - remove ptes mapping the vma
1209 * @vma: vm_area_struct holding ptes to be zapped
1210 * @address: starting address of pages to zap
1211 * @size: number of bytes to zap
1212 *
1213 * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1214 *
1215 * The entire address range must be fully contained within the vma.
1216 *
1217 * Returns 0 if successful.
1218 */
1219int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
1220 unsigned long size)
1221{
1222 if (address < vma->vm_start || address + size > vma->vm_end ||
1223 !(vma->vm_flags & VM_PFNMAP))
1224 return -1;
1225 zap_page_range(vma, address, size, NULL);
1226 return 0;
1227}
1228EXPORT_SYMBOL_GPL(zap_vma_ptes);
1229
Johannes Weiner142762b2010-05-24 14:32:39 -07001230/**
1231 * follow_page - look up a page descriptor from a user-virtual address
1232 * @vma: vm_area_struct mapping @address
1233 * @address: virtual address to look up
1234 * @flags: flags modifying lookup behaviour
1235 *
1236 * @flags can have FOLL_ flags set, defined in <linux/mm.h>
1237 *
1238 * Returns the mapped (struct page *), %NULL if no mapping exists, or
1239 * an error pointer if there is a mapping to something not represented
1240 * by a page descriptor (see also vm_normal_page()).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 */
Linus Torvalds6aab3412005-11-28 14:34:23 -08001242struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001243 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244{
1245 pgd_t *pgd;
1246 pud_t *pud;
1247 pmd_t *pmd;
1248 pte_t *ptep, pte;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001249 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 struct page *page;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001251 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001253 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
1254 if (!IS_ERR(page)) {
1255 BUG_ON(flags & FOLL_GET);
1256 goto out;
1257 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001259 page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 pgd = pgd_offset(mm, address);
1261 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001262 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263
1264 pud = pud_offset(pgd, address);
Andi Kleenceb86872008-07-23 21:27:50 -07001265 if (pud_none(*pud))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001266 goto no_page_table;
Andi Kleenceb86872008-07-23 21:27:50 -07001267 if (pud_huge(*pud)) {
1268 BUG_ON(flags & FOLL_GET);
1269 page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE);
1270 goto out;
1271 }
1272 if (unlikely(pud_bad(*pud)))
1273 goto no_page_table;
1274
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 pmd = pmd_offset(pud, address);
Hugh Dickinsaeed5fc2008-05-06 20:49:23 +01001276 if (pmd_none(*pmd))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001277 goto no_page_table;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001278 if (pmd_huge(*pmd)) {
1279 BUG_ON(flags & FOLL_GET);
1280 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
1281 goto out;
1282 }
Hugh Dickinsaeed5fc2008-05-06 20:49:23 +01001283 if (unlikely(pmd_bad(*pmd)))
1284 goto no_page_table;
1285
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001286 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287
1288 pte = *ptep;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001289 if (!pte_present(pte))
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001290 goto no_page;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001291 if ((flags & FOLL_WRITE) && !pte_write(pte))
1292 goto unlock;
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07001293
Linus Torvalds6aab3412005-11-28 14:34:23 -08001294 page = vm_normal_page(vma, address, pte);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07001295 if (unlikely(!page)) {
1296 if ((flags & FOLL_DUMP) ||
Hugh Dickins62eede62009-09-21 17:03:34 -07001297 !is_zero_pfn(pte_pfn(pte)))
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07001298 goto bad_page;
1299 page = pte_page(pte);
1300 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001301
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001302 if (flags & FOLL_GET)
1303 get_page(page);
1304 if (flags & FOLL_TOUCH) {
1305 if ((flags & FOLL_WRITE) &&
1306 !pte_dirty(pte) && !PageDirty(page))
1307 set_page_dirty(page);
KOSAKI Motohirobd775c42009-03-31 15:19:37 -07001308 /*
1309 * pte_mkyoung() would be more correct here, but atomic care
1310 * is needed to avoid losing the dirty bit: it is easier to use
1311 * mark_page_accessed().
1312 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001313 mark_page_accessed(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001315unlock:
1316 pte_unmap_unlock(ptep, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317out:
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001318 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001320bad_page:
1321 pte_unmap_unlock(ptep, ptl);
1322 return ERR_PTR(-EFAULT);
1323
1324no_page:
1325 pte_unmap_unlock(ptep, ptl);
1326 if (!pte_none(pte))
1327 return page;
Hugh Dickins8e4b9a62009-09-21 17:03:26 -07001328
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001329no_page_table:
1330 /*
1331 * When core dumping an enormous anonymous area that nobody
Hugh Dickins8e4b9a62009-09-21 17:03:26 -07001332 * has touched so far, we don't want to allocate unnecessary pages or
1333 * page tables. Return error instead of NULL to skip handle_mm_fault,
1334 * then get_dump_page() will return NULL to leave a hole in the dump.
1335 * But we can only make this optimization where a hole would surely
1336 * be zero-filled if handle_mm_fault() actually did handle it.
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001337 */
Hugh Dickins8e4b9a62009-09-21 17:03:26 -07001338 if ((flags & FOLL_DUMP) &&
1339 (!vma->vm_ops || !vma->vm_ops->fault))
1340 return ERR_PTR(-EFAULT);
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001341 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342}
1343
Nick Pigginb291f002008-10-18 20:26:44 -07001344int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
Hugh Dickins58fa8792009-09-21 17:03:31 -07001345 unsigned long start, int nr_pages, unsigned int gup_flags,
Peter Zijlstra9d737772009-06-25 11:58:55 +02001346 struct page **pages, struct vm_area_struct **vmas)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347{
1348 int i;
Hugh Dickins58fa8792009-09-21 17:03:31 -07001349 unsigned long vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
Peter Zijlstra9d737772009-06-25 11:58:55 +02001351 if (nr_pages <= 0)
Jonathan Corbet900cf082008-02-11 16:17:33 -07001352 return 0;
Hugh Dickins58fa8792009-09-21 17:03:31 -07001353
1354 VM_BUG_ON(!!pages != !!(gup_flags & FOLL_GET));
1355
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 /*
1357 * Require read or write permissions.
Hugh Dickins58fa8792009-09-21 17:03:31 -07001358 * If FOLL_FORCE is set, we only require the "MAY" flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 */
Hugh Dickins58fa8792009-09-21 17:03:31 -07001360 vm_flags = (gup_flags & FOLL_WRITE) ?
1361 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
1362 vm_flags &= (gup_flags & FOLL_FORCE) ?
1363 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 i = 0;
1365
1366 do {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001367 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368
1369 vma = find_extend_vma(mm, start);
1370 if (!vma && in_gate_area(tsk, start)) {
1371 unsigned long pg = start & PAGE_MASK;
1372 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
1373 pgd_t *pgd;
1374 pud_t *pud;
1375 pmd_t *pmd;
1376 pte_t *pte;
Nick Pigginb291f002008-10-18 20:26:44 -07001377
1378 /* user gate pages are read-only */
Hugh Dickins58fa8792009-09-21 17:03:31 -07001379 if (gup_flags & FOLL_WRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 return i ? : -EFAULT;
1381 if (pg > TASK_SIZE)
1382 pgd = pgd_offset_k(pg);
1383 else
1384 pgd = pgd_offset_gate(mm, pg);
1385 BUG_ON(pgd_none(*pgd));
1386 pud = pud_offset(pgd, pg);
1387 BUG_ON(pud_none(*pud));
1388 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001389 if (pmd_none(*pmd))
1390 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001392 if (pte_none(*pte)) {
1393 pte_unmap(pte);
1394 return i ? : -EFAULT;
1395 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 if (pages) {
Hugh Dickinsde51257a2010-07-30 10:58:26 -07001397 struct page *page;
1398
1399 page = vm_normal_page(gate_vma, start, *pte);
1400 if (!page) {
1401 if (!(gup_flags & FOLL_DUMP) &&
1402 is_zero_pfn(pte_pfn(*pte)))
1403 page = pte_page(*pte);
1404 else {
1405 pte_unmap(pte);
1406 return i ? : -EFAULT;
1407 }
1408 }
Linus Torvalds6aab3412005-11-28 14:34:23 -08001409 pages[i] = page;
Hugh Dickinsde51257a2010-07-30 10:58:26 -07001410 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 }
1412 pte_unmap(pte);
1413 if (vmas)
1414 vmas[i] = gate_vma;
1415 i++;
1416 start += PAGE_SIZE;
Peter Zijlstra9d737772009-06-25 11:58:55 +02001417 nr_pages--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 continue;
1419 }
1420
Nick Pigginb291f002008-10-18 20:26:44 -07001421 if (!vma ||
1422 (vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
Hugh Dickins1c3aff12009-09-21 17:03:24 -07001423 !(vm_flags & vma->vm_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 return i ? : -EFAULT;
1425
Hugh Dickins2a15efc2009-09-21 17:03:27 -07001426 if (is_vm_hugetlb_page(vma)) {
1427 i = follow_hugetlb_page(mm, vma, pages, vmas,
Hugh Dickins58fa8792009-09-21 17:03:31 -07001428 &start, &nr_pages, i, gup_flags);
Hugh Dickins2a15efc2009-09-21 17:03:27 -07001429 continue;
1430 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001431
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 do {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001433 struct page *page;
Hugh Dickins58fa8792009-09-21 17:03:31 -07001434 unsigned int foll_flags = gup_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
Ethan Solomita462e00c2007-07-15 23:38:16 -07001436 /*
Ying Han47792802009-01-06 14:40:18 -08001437 * If we have a pending SIGKILL, don't keep faulting
Hugh Dickins1c3aff12009-09-21 17:03:24 -07001438 * pages and potentially allocating memory.
Ethan Solomita462e00c2007-07-15 23:38:16 -07001439 */
Hugh Dickins1c3aff12009-09-21 17:03:24 -07001440 if (unlikely(fatal_signal_pending(current)))
Ying Han47792802009-01-06 14:40:18 -08001441 return i ? i : -ERESTARTSYS;
Ethan Solomita462e00c2007-07-15 23:38:16 -07001442
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001443 cond_resched();
Linus Torvalds6aab3412005-11-28 14:34:23 -08001444 while (!(page = follow_page(vma, start, foll_flags))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001445 int ret;
Linus Torvaldsd06063c2009-04-10 09:01:23 -07001446
Hugh Dickinsd26ed652009-06-23 13:52:49 +01001447 ret = handle_mm_fault(mm, vma, start,
1448 (foll_flags & FOLL_WRITE) ?
1449 FAULT_FLAG_WRITE : 0);
1450
Nick Piggin83c54072007-07-19 01:47:05 -07001451 if (ret & VM_FAULT_ERROR) {
1452 if (ret & VM_FAULT_OOM)
1453 return i ? i : -ENOMEM;
Andi Kleend1737fd2009-09-16 11:50:06 +02001454 if (ret &
1455 (VM_FAULT_HWPOISON|VM_FAULT_SIGBUS))
Nick Piggin83c54072007-07-19 01:47:05 -07001456 return i ? i : -EFAULT;
1457 BUG();
1458 }
1459 if (ret & VM_FAULT_MAJOR)
1460 tsk->maj_flt++;
1461 else
1462 tsk->min_flt++;
1463
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001464 /*
Nick Piggin83c54072007-07-19 01:47:05 -07001465 * The VM_FAULT_WRITE bit tells us that
1466 * do_wp_page has broken COW when necessary,
1467 * even if maybe_mkwrite decided not to set
1468 * pte_write. We can thus safely do subsequent
Hugh Dickins878b63a2009-01-06 14:39:32 -08001469 * page lookups as if they were reads. But only
1470 * do so when looping for pte_write is futile:
1471 * in some cases userspace may also be wanting
1472 * to write to the gotten user page, which a
1473 * read fault here might prevent (a readonly
1474 * page might get reCOWed by userspace write).
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001475 */
Hugh Dickins878b63a2009-01-06 14:39:32 -08001476 if ((ret & VM_FAULT_WRITE) &&
1477 !(vma->vm_flags & VM_WRITE))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001478 foll_flags &= ~FOLL_WRITE;
Nick Piggin83c54072007-07-19 01:47:05 -07001479
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07001480 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 }
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001482 if (IS_ERR(page))
1483 return i ? i : PTR_ERR(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001485 pages[i] = page;
James Bottomley03beb072006-03-26 01:36:57 -08001486
Russell Kinga6f36be2006-12-30 22:24:19 +00001487 flush_anon_page(vma, page, start);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001488 flush_dcache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 }
1490 if (vmas)
1491 vmas[i] = vma;
1492 i++;
1493 start += PAGE_SIZE;
Peter Zijlstra9d737772009-06-25 11:58:55 +02001494 nr_pages--;
1495 } while (nr_pages && start < vma->vm_end);
1496 } while (nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 return i;
1498}
Nick Pigginb291f002008-10-18 20:26:44 -07001499
Nick Piggind2bf6be2009-06-16 15:31:39 -07001500/**
1501 * get_user_pages() - pin user pages in memory
1502 * @tsk: task_struct of target task
1503 * @mm: mm_struct of target mm
1504 * @start: starting user address
Peter Zijlstra9d737772009-06-25 11:58:55 +02001505 * @nr_pages: number of pages from start to pin
Nick Piggind2bf6be2009-06-16 15:31:39 -07001506 * @write: whether pages will be written to by the caller
1507 * @force: whether to force write access even if user mapping is
1508 * readonly. This will result in the page being COWed even
1509 * in MAP_SHARED mappings. You do not want this.
1510 * @pages: array that receives pointers to the pages pinned.
1511 * Should be at least nr_pages long. Or NULL, if caller
1512 * only intends to ensure the pages are faulted in.
1513 * @vmas: array of pointers to vmas corresponding to each page.
1514 * Or NULL if the caller does not require them.
1515 *
1516 * Returns number of pages pinned. This may be fewer than the number
Peter Zijlstra9d737772009-06-25 11:58:55 +02001517 * requested. If nr_pages is 0 or negative, returns 0. If no pages
Nick Piggind2bf6be2009-06-16 15:31:39 -07001518 * were pinned, returns -errno. Each page returned must be released
1519 * with a put_page() call when it is finished with. vmas will only
1520 * remain valid while mmap_sem is held.
1521 *
1522 * Must be called with mmap_sem held for read or write.
1523 *
1524 * get_user_pages walks a process's page tables and takes a reference to
1525 * each struct page that each user address corresponds to at a given
1526 * instant. That is, it takes the page that would be accessed if a user
1527 * thread accesses the given user virtual address at that instant.
1528 *
1529 * This does not guarantee that the page exists in the user mappings when
1530 * get_user_pages returns, and there may even be a completely different
1531 * page there in some cases (eg. if mmapped pagecache has been invalidated
1532 * and subsequently re faulted). However it does guarantee that the page
1533 * won't be freed completely. And mostly callers simply care that the page
1534 * contains data that was valid *at some point in time*. Typically, an IO
1535 * or similar operation cannot guarantee anything stronger anyway because
1536 * locks can't be held over the syscall boundary.
1537 *
1538 * If write=0, the page must not be written to. If the page is written to,
1539 * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called
1540 * after the page is finished with, and before put_page is called.
1541 *
1542 * get_user_pages is typically used for fewer-copy IO operations, to get a
1543 * handle on the memory by some means other than accesses via the user virtual
1544 * addresses. The pages may be submitted for DMA to devices or accessed via
1545 * their kernel linear mapping (via the kmap APIs). Care should be taken to
1546 * use the correct cache flushing APIs.
1547 *
1548 * See also get_user_pages_fast, for performance critical applications.
1549 */
Nick Pigginb291f002008-10-18 20:26:44 -07001550int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
Peter Zijlstra9d737772009-06-25 11:58:55 +02001551 unsigned long start, int nr_pages, int write, int force,
Nick Pigginb291f002008-10-18 20:26:44 -07001552 struct page **pages, struct vm_area_struct **vmas)
1553{
Hugh Dickins58fa8792009-09-21 17:03:31 -07001554 int flags = FOLL_TOUCH;
Nick Pigginb291f002008-10-18 20:26:44 -07001555
Hugh Dickins58fa8792009-09-21 17:03:31 -07001556 if (pages)
1557 flags |= FOLL_GET;
Nick Pigginb291f002008-10-18 20:26:44 -07001558 if (write)
Hugh Dickins58fa8792009-09-21 17:03:31 -07001559 flags |= FOLL_WRITE;
Nick Pigginb291f002008-10-18 20:26:44 -07001560 if (force)
Hugh Dickins58fa8792009-09-21 17:03:31 -07001561 flags |= FOLL_FORCE;
Nick Pigginb291f002008-10-18 20:26:44 -07001562
Peter Zijlstra9d737772009-06-25 11:58:55 +02001563 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas);
Nick Pigginb291f002008-10-18 20:26:44 -07001564}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565EXPORT_SYMBOL(get_user_pages);
1566
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07001567/**
1568 * get_dump_page() - pin user page in memory while writing it to core dump
1569 * @addr: user address
1570 *
1571 * Returns struct page pointer of user page pinned for dump,
1572 * to be freed afterwards by page_cache_release() or put_page().
1573 *
1574 * Returns NULL on any kind of failure - a hole must then be inserted into
1575 * the corefile, to preserve alignment with its headers; and also returns
1576 * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found -
1577 * allowing a hole to be left in the corefile to save diskspace.
1578 *
1579 * Called without mmap_sem, but after all other threads have been killed.
1580 */
1581#ifdef CONFIG_ELF_CORE
1582struct page *get_dump_page(unsigned long addr)
1583{
1584 struct vm_area_struct *vma;
1585 struct page *page;
1586
1587 if (__get_user_pages(current, current->mm, addr, 1,
Hugh Dickins58fa8792009-09-21 17:03:31 -07001588 FOLL_FORCE | FOLL_DUMP | FOLL_GET, &page, &vma) < 1)
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07001589 return NULL;
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07001590 flush_cache_page(vma, addr, page_to_pfn(page));
1591 return page;
1592}
1593#endif /* CONFIG_ELF_CORE */
1594
Harvey Harrison920c7a52008-02-04 22:29:26 -08001595pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr,
1596 spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001597{
1598 pgd_t * pgd = pgd_offset(mm, addr);
1599 pud_t * pud = pud_alloc(mm, pgd, addr);
1600 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001601 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001602 if (pmd)
1603 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1604 }
1605 return NULL;
1606}
1607
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001609 * This is the old fallback for page remapping.
1610 *
1611 * For historical reasons, it only allows reserved pages. Only
1612 * old drivers should use this, and they needed to mark their
1613 * pages reserved for the old functions anyway.
1614 */
Nick Piggin423bad602008-04-28 02:13:01 -07001615static int insert_page(struct vm_area_struct *vma, unsigned long addr,
1616 struct page *page, pgprot_t prot)
Linus Torvalds238f58d2005-11-29 13:01:56 -08001617{
Nick Piggin423bad602008-04-28 02:13:01 -07001618 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001619 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001620 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001621 spinlock_t *ptl;
1622
Linus Torvalds238f58d2005-11-29 13:01:56 -08001623 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001624 if (PageAnon(page))
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001625 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001626 retval = -ENOMEM;
1627 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001628 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001629 if (!pte)
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001630 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001631 retval = -EBUSY;
1632 if (!pte_none(*pte))
1633 goto out_unlock;
1634
1635 /* Ok, finally just insert the thing.. */
1636 get_page(page);
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08001637 inc_mm_counter_fast(mm, MM_FILEPAGES);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001638 page_add_file_rmap(page);
1639 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1640
1641 retval = 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001642 pte_unmap_unlock(pte, ptl);
1643 return retval;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001644out_unlock:
1645 pte_unmap_unlock(pte, ptl);
1646out:
1647 return retval;
1648}
1649
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001650/**
1651 * vm_insert_page - insert single page into user vma
1652 * @vma: user vma to map to
1653 * @addr: target user address of this page
1654 * @page: source kernel page
1655 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001656 * This allows drivers to insert individual pages they've allocated
1657 * into a user vma.
1658 *
1659 * The page has to be a nice clean _individual_ kernel allocation.
1660 * If you allocate a compound page, you need to have marked it as
1661 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001662 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001663 *
1664 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1665 * took an arbitrary page protection parameter. This doesn't allow
1666 * that. Your vma protection will have to be set up correctly, which
1667 * means that if you want a shared writable mapping, you'd better
1668 * ask for a shared writable mapping!
1669 *
1670 * The page does not need to be reserved.
1671 */
Nick Piggin423bad602008-04-28 02:13:01 -07001672int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
1673 struct page *page)
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001674{
1675 if (addr < vma->vm_start || addr >= vma->vm_end)
1676 return -EFAULT;
1677 if (!page_count(page))
1678 return -EINVAL;
Linus Torvalds4d7672b2005-12-16 10:21:23 -08001679 vma->vm_flags |= VM_INSERTPAGE;
Nick Piggin423bad602008-04-28 02:13:01 -07001680 return insert_page(vma, addr, page, vma->vm_page_prot);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001681}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001682EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001683
Nick Piggin423bad602008-04-28 02:13:01 -07001684static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1685 unsigned long pfn, pgprot_t prot)
1686{
1687 struct mm_struct *mm = vma->vm_mm;
1688 int retval;
1689 pte_t *pte, entry;
1690 spinlock_t *ptl;
1691
1692 retval = -ENOMEM;
1693 pte = get_locked_pte(mm, addr, &ptl);
1694 if (!pte)
1695 goto out;
1696 retval = -EBUSY;
1697 if (!pte_none(*pte))
1698 goto out_unlock;
1699
1700 /* Ok, finally just insert the thing.. */
1701 entry = pte_mkspecial(pfn_pte(pfn, prot));
1702 set_pte_at(mm, addr, pte, entry);
Russell King4b3073e2009-12-18 16:40:18 +00001703 update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */
Nick Piggin423bad602008-04-28 02:13:01 -07001704
1705 retval = 0;
1706out_unlock:
1707 pte_unmap_unlock(pte, ptl);
1708out:
1709 return retval;
1710}
1711
Nick Piggine0dc0d82007-02-12 00:51:36 -08001712/**
1713 * vm_insert_pfn - insert single pfn into user vma
1714 * @vma: user vma to map to
1715 * @addr: target user address of this page
1716 * @pfn: source kernel pfn
1717 *
1718 * Similar to vm_inert_page, this allows drivers to insert individual pages
1719 * they've allocated into a user vma. Same comments apply.
1720 *
1721 * This function should only be called from a vm_ops->fault handler, and
1722 * in that case the handler should return NULL.
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001723 *
1724 * vma cannot be a COW mapping.
1725 *
1726 * As this is called only for pages that do not currently exist, we
1727 * do not need to flush old virtual caches or the TLB.
Nick Piggine0dc0d82007-02-12 00:51:36 -08001728 */
1729int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
Nick Piggin423bad602008-04-28 02:13:01 -07001730 unsigned long pfn)
Nick Piggine0dc0d82007-02-12 00:51:36 -08001731{
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001732 int ret;
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001733 pgprot_t pgprot = vma->vm_page_prot;
Nick Piggin7e675132008-04-28 02:13:00 -07001734 /*
1735 * Technically, architectures with pte_special can avoid all these
1736 * restrictions (same for remap_pfn_range). However we would like
1737 * consistency in testing and feature parity among all, so we should
1738 * try to keep these invariants in place for everybody.
1739 */
Jared Hulbertb379d792008-04-28 02:12:58 -07001740 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
1741 BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
1742 (VM_PFNMAP|VM_MIXEDMAP));
1743 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1744 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
Nick Piggine0dc0d82007-02-12 00:51:36 -08001745
Nick Piggin423bad602008-04-28 02:13:01 -07001746 if (addr < vma->vm_start || addr >= vma->vm_end)
1747 return -EFAULT;
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001748 if (track_pfn_vma_new(vma, &pgprot, pfn, PAGE_SIZE))
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001749 return -EINVAL;
1750
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001751 ret = insert_pfn(vma, addr, pfn, pgprot);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001752
1753 if (ret)
1754 untrack_pfn_vma(vma, pfn, PAGE_SIZE);
1755
1756 return ret;
Nick Piggine0dc0d82007-02-12 00:51:36 -08001757}
1758EXPORT_SYMBOL(vm_insert_pfn);
1759
Nick Piggin423bad602008-04-28 02:13:01 -07001760int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
1761 unsigned long pfn)
1762{
1763 BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
1764
1765 if (addr < vma->vm_start || addr >= vma->vm_end)
1766 return -EFAULT;
1767
1768 /*
1769 * If we don't have pte special, then we have to use the pfn_valid()
1770 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1771 * refcount the page if pfn_valid is true (hence insert_page rather
Hugh Dickins62eede62009-09-21 17:03:34 -07001772 * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP
1773 * without pte special, it would there be refcounted as a normal page.
Nick Piggin423bad602008-04-28 02:13:01 -07001774 */
1775 if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) {
1776 struct page *page;
1777
1778 page = pfn_to_page(pfn);
1779 return insert_page(vma, addr, page, vma->vm_page_prot);
1780 }
1781 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
1782}
1783EXPORT_SYMBOL(vm_insert_mixed);
1784
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001785/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 * maps a range of physical memory into the requested pages. the old
1787 * mappings are removed. any references to nonexistent pages results
1788 * in null mappings (currently treated as "copy-on-access")
1789 */
1790static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1791 unsigned long addr, unsigned long end,
1792 unsigned long pfn, pgprot_t prot)
1793{
1794 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001795 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796
Hugh Dickinsc74df322005-10-29 18:16:23 -07001797 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 if (!pte)
1799 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001800 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 do {
1802 BUG_ON(!pte_none(*pte));
Nick Piggin7e675132008-04-28 02:13:00 -07001803 set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 pfn++;
1805 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001806 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001807 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 return 0;
1809}
1810
1811static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1812 unsigned long addr, unsigned long end,
1813 unsigned long pfn, pgprot_t prot)
1814{
1815 pmd_t *pmd;
1816 unsigned long next;
1817
1818 pfn -= addr >> PAGE_SHIFT;
1819 pmd = pmd_alloc(mm, pud, addr);
1820 if (!pmd)
1821 return -ENOMEM;
1822 do {
1823 next = pmd_addr_end(addr, end);
1824 if (remap_pte_range(mm, pmd, addr, next,
1825 pfn + (addr >> PAGE_SHIFT), prot))
1826 return -ENOMEM;
1827 } while (pmd++, addr = next, addr != end);
1828 return 0;
1829}
1830
1831static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1832 unsigned long addr, unsigned long end,
1833 unsigned long pfn, pgprot_t prot)
1834{
1835 pud_t *pud;
1836 unsigned long next;
1837
1838 pfn -= addr >> PAGE_SHIFT;
1839 pud = pud_alloc(mm, pgd, addr);
1840 if (!pud)
1841 return -ENOMEM;
1842 do {
1843 next = pud_addr_end(addr, end);
1844 if (remap_pmd_range(mm, pud, addr, next,
1845 pfn + (addr >> PAGE_SHIFT), prot))
1846 return -ENOMEM;
1847 } while (pud++, addr = next, addr != end);
1848 return 0;
1849}
1850
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001851/**
1852 * remap_pfn_range - remap kernel memory to userspace
1853 * @vma: user vma to map to
1854 * @addr: target user address to start at
1855 * @pfn: physical address of kernel memory
1856 * @size: size of map area
1857 * @prot: page protection flags for this mapping
1858 *
1859 * Note: this is only safe if the mm semaphore is held when called.
1860 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1862 unsigned long pfn, unsigned long size, pgprot_t prot)
1863{
1864 pgd_t *pgd;
1865 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001866 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 struct mm_struct *mm = vma->vm_mm;
1868 int err;
1869
1870 /*
1871 * Physically remapped pages are special. Tell the
1872 * rest of the world about it:
1873 * VM_IO tells people not to look at these pages
1874 * (accesses can have side effects).
Hugh Dickins0b14c172005-11-21 21:32:15 -08001875 * VM_RESERVED is specified all over the place, because
1876 * in 2.4 it kept swapout's vma scan off this vma; but
1877 * in 2.6 the LRU scan won't even find its pages, so this
1878 * flag means no more than count its pages in reserved_vm,
1879 * and omit it from core dump, even when VM_IO turned off.
Linus Torvalds6aab3412005-11-28 14:34:23 -08001880 * VM_PFNMAP tells the core MM that the base pages are just
1881 * raw PFN mappings, and do not have a "struct page" associated
1882 * with them.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001883 *
1884 * There's a horrible special case to handle copy-on-write
1885 * behaviour that some programs depend on. We mark the "original"
1886 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 */
Pallipadi, Venkatesh4bb9c5c2009-03-12 17:45:27 -07001888 if (addr == vma->vm_start && end == vma->vm_end) {
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001889 vma->vm_pgoff = pfn;
Pallipadi, Venkatesh895791d2009-03-13 16:35:44 -07001890 vma->vm_flags |= VM_PFN_AT_MMAP;
Pallipadi, Venkatesh4bb9c5c2009-03-12 17:45:27 -07001891 } else if (is_cow_mapping(vma->vm_flags))
venkatesh.pallipadi@intel.com3c8bb732008-12-18 11:41:27 -08001892 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001893
Linus Torvalds6aab3412005-11-28 14:34:23 -08001894 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001896 err = track_pfn_vma_new(vma, &prot, pfn, PAGE_ALIGN(size));
venkatesh.pallipadi@intel.coma3670612009-01-09 16:13:09 -08001897 if (err) {
1898 /*
1899 * To indicate that track_pfn related cleanup is not
1900 * needed from higher level routine calling unmap_vmas
1901 */
1902 vma->vm_flags &= ~(VM_IO | VM_RESERVED | VM_PFNMAP);
Pallipadi, Venkatesh895791d2009-03-13 16:35:44 -07001903 vma->vm_flags &= ~VM_PFN_AT_MMAP;
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001904 return -EINVAL;
venkatesh.pallipadi@intel.coma3670612009-01-09 16:13:09 -08001905 }
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001906
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 BUG_ON(addr >= end);
1908 pfn -= addr >> PAGE_SHIFT;
1909 pgd = pgd_offset(mm, addr);
1910 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 do {
1912 next = pgd_addr_end(addr, end);
1913 err = remap_pud_range(mm, pgd, addr, next,
1914 pfn + (addr >> PAGE_SHIFT), prot);
1915 if (err)
1916 break;
1917 } while (pgd++, addr = next, addr != end);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001918
1919 if (err)
1920 untrack_pfn_vma(vma, pfn, PAGE_ALIGN(size));
1921
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 return err;
1923}
1924EXPORT_SYMBOL(remap_pfn_range);
1925
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001926static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1927 unsigned long addr, unsigned long end,
1928 pte_fn_t fn, void *data)
1929{
1930 pte_t *pte;
1931 int err;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001932 pgtable_t token;
Borislav Petkov94909912007-05-06 14:49:17 -07001933 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001934
1935 pte = (mm == &init_mm) ?
1936 pte_alloc_kernel(pmd, addr) :
1937 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1938 if (!pte)
1939 return -ENOMEM;
1940
1941 BUG_ON(pmd_huge(*pmd));
1942
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08001943 arch_enter_lazy_mmu_mode();
1944
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001945 token = pmd_pgtable(*pmd);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001946
1947 do {
Daisuke Nishimurac36987e2009-10-26 16:50:23 -07001948 err = fn(pte++, token, addr, data);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001949 if (err)
1950 break;
Daisuke Nishimurac36987e2009-10-26 16:50:23 -07001951 } while (addr += PAGE_SIZE, addr != end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001952
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08001953 arch_leave_lazy_mmu_mode();
1954
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001955 if (mm != &init_mm)
1956 pte_unmap_unlock(pte-1, ptl);
1957 return err;
1958}
1959
1960static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1961 unsigned long addr, unsigned long end,
1962 pte_fn_t fn, void *data)
1963{
1964 pmd_t *pmd;
1965 unsigned long next;
1966 int err;
1967
Andi Kleenceb86872008-07-23 21:27:50 -07001968 BUG_ON(pud_huge(*pud));
1969
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001970 pmd = pmd_alloc(mm, pud, addr);
1971 if (!pmd)
1972 return -ENOMEM;
1973 do {
1974 next = pmd_addr_end(addr, end);
1975 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1976 if (err)
1977 break;
1978 } while (pmd++, addr = next, addr != end);
1979 return err;
1980}
1981
1982static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1983 unsigned long addr, unsigned long end,
1984 pte_fn_t fn, void *data)
1985{
1986 pud_t *pud;
1987 unsigned long next;
1988 int err;
1989
1990 pud = pud_alloc(mm, pgd, addr);
1991 if (!pud)
1992 return -ENOMEM;
1993 do {
1994 next = pud_addr_end(addr, end);
1995 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1996 if (err)
1997 break;
1998 } while (pud++, addr = next, addr != end);
1999 return err;
2000}
2001
2002/*
2003 * Scan a region of virtual memory, filling in page tables as necessary
2004 * and calling a provided function on each leaf page table.
2005 */
2006int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
2007 unsigned long size, pte_fn_t fn, void *data)
2008{
2009 pgd_t *pgd;
2010 unsigned long next;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002011 unsigned long start = addr, end = addr + size;
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002012 int err;
2013
2014 BUG_ON(addr >= end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002015 mmu_notifier_invalidate_range_start(mm, start, end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002016 pgd = pgd_offset(mm, addr);
2017 do {
2018 next = pgd_addr_end(addr, end);
2019 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
2020 if (err)
2021 break;
2022 } while (pgd++, addr = next, addr != end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002023 mmu_notifier_invalidate_range_end(mm, start, end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07002024 return err;
2025}
2026EXPORT_SYMBOL_GPL(apply_to_page_range);
2027
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002029 * handle_pte_fault chooses page fault handler according to an entry
2030 * which was read non-atomically. Before making any commitment, on
2031 * those architectures or configurations (e.g. i386 with PAE) which
2032 * might give a mix of unmatched parts, do_swap_page and do_file_page
2033 * must check under lock before unmapping the pte and proceeding
2034 * (but do_wp_page is only called after already making such a check;
2035 * and do_anonymous_page and do_no_page can safely check later on).
2036 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002037static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002038 pte_t *page_table, pte_t orig_pte)
2039{
2040 int same = 1;
2041#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
2042 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002043 spinlock_t *ptl = pte_lockptr(mm, pmd);
2044 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002045 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002046 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002047 }
2048#endif
2049 pte_unmap(page_table);
2050 return same;
2051}
2052
2053/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
2055 * servicing faults for write access. In the normal case, do always want
2056 * pte_mkwrite. But get_user_pages can cause write faults for mappings
2057 * that do not have writing enabled, when used by access_process_vm.
2058 */
2059static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
2060{
2061 if (likely(vma->vm_flags & VM_WRITE))
2062 pte = pte_mkwrite(pte);
2063 return pte;
2064}
2065
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00002066static 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 -08002067{
2068 /*
2069 * If the source page was a PFN mapping, we don't have
2070 * a "struct page" for it. We do a best-effort copy by
2071 * just copying from the original user address. If that
2072 * fails, we just zero-fill it. Live with it.
2073 */
2074 if (unlikely(!src)) {
2075 void *kaddr = kmap_atomic(dst, KM_USER0);
Linus Torvalds5d2a2dbbc2005-11-29 14:07:55 -08002076 void __user *uaddr = (void __user *)(va & PAGE_MASK);
2077
2078 /*
2079 * This really shouldn't fail, because the page is there
2080 * in the page tables. But it might just be unreadable,
2081 * in which case we just give up and fill the result with
2082 * zeroes.
2083 */
2084 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Linus Torvalds6aab3412005-11-28 14:34:23 -08002085 memset(kaddr, 0, PAGE_SIZE);
2086 kunmap_atomic(kaddr, KM_USER0);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07002087 flush_dcache_page(dst);
Nick Piggin0ed361d2008-02-04 22:29:34 -08002088 } else
2089 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08002090}
2091
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 * This routine handles present pages, when users try to write
2094 * to a shared page. It is done by copying the page to a new address
2095 * and decrementing the shared-page counter for the old page.
2096 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 * Note that this routine assumes that the protection checks have been
2098 * done by the caller (the low-level page fault routine in most cases).
2099 * Thus we can safely just mark it writable once we've done any necessary
2100 * COW.
2101 *
2102 * We also mark the page dirty at this point even though the page will
2103 * change only once the write actually happens. This avoids a few races,
2104 * and potentially makes it more efficient.
2105 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002106 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2107 * but allow concurrent faults), with pte both mapped and locked.
2108 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002110static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
2111 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002112 spinlock_t *ptl, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113{
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00002114 struct page *old_page, *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 pte_t entry;
Nick Piggin83c54072007-07-19 01:47:05 -07002116 int reuse = 0, ret = 0;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002117 int page_mkwrite = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002118 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119
Linus Torvalds6aab3412005-11-28 14:34:23 -08002120 old_page = vm_normal_page(vma, address, orig_pte);
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002121 if (!old_page) {
2122 /*
2123 * VM_MIXEDMAP !pfn_valid() case
2124 *
2125 * We should not cow pages in a shared writeable mapping.
2126 * Just mark the pages writable as we can't do any dirty
2127 * accounting on raw pfn maps.
2128 */
2129 if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
2130 (VM_WRITE|VM_SHARED))
2131 goto reuse;
Linus Torvalds6aab3412005-11-28 14:34:23 -08002132 goto gotten;
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002133 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002135 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002136 * Take out anonymous pages first, anonymous shared vmas are
2137 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002138 */
Hugh Dickins9a840892009-09-21 17:02:01 -07002139 if (PageAnon(old_page) && !PageKsm(old_page)) {
Hugh Dickinsab967d82009-01-06 14:39:33 -08002140 if (!trylock_page(old_page)) {
2141 page_cache_get(old_page);
2142 pte_unmap_unlock(page_table, ptl);
2143 lock_page(old_page);
2144 page_table = pte_offset_map_lock(mm, pmd, address,
2145 &ptl);
2146 if (!pte_same(*page_table, orig_pte)) {
2147 unlock_page(old_page);
2148 page_cache_release(old_page);
2149 goto unlock;
2150 }
2151 page_cache_release(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002152 }
Hugh Dickins7b1fe592009-01-06 14:39:34 -08002153 reuse = reuse_swap_page(old_page);
Rik van Rielc44b6742010-03-05 13:42:09 -08002154 if (reuse)
2155 /*
2156 * The page is all ours. Move it to our anon_vma so
2157 * the rmap code will not search our parent or siblings.
2158 * Protected against the rmap code by the page lock.
2159 */
2160 page_move_anon_rmap(old_page, vma, address);
Hugh Dickinsab967d82009-01-06 14:39:33 -08002161 unlock_page(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002162 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002163 (VM_WRITE|VM_SHARED))) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002164 /*
2165 * Only catch write-faults on shared writable pages,
2166 * read-only shared pages can get COWed by
2167 * get_user_pages(.write=1, .force=1).
2168 */
David Howells9637a5e2006-06-23 02:03:43 -07002169 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
Nick Pigginc2ec1752009-03-31 15:23:21 -07002170 struct vm_fault vmf;
2171 int tmp;
2172
2173 vmf.virtual_address = (void __user *)(address &
2174 PAGE_MASK);
2175 vmf.pgoff = old_page->index;
2176 vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
2177 vmf.page = old_page;
2178
David Howells9637a5e2006-06-23 02:03:43 -07002179 /*
2180 * Notify the address space that the page is about to
2181 * become writable so that it can prohibit this or wait
2182 * for the page to get into an appropriate state.
2183 *
2184 * We do this without the lock held, so that it can
2185 * sleep if it needs to.
2186 */
2187 page_cache_get(old_page);
2188 pte_unmap_unlock(page_table, ptl);
2189
Nick Pigginc2ec1752009-03-31 15:23:21 -07002190 tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
2191 if (unlikely(tmp &
2192 (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
2193 ret = tmp;
David Howells9637a5e2006-06-23 02:03:43 -07002194 goto unwritable_page;
Nick Pigginc2ec1752009-03-31 15:23:21 -07002195 }
Nick Pigginb827e492009-04-30 15:08:16 -07002196 if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
2197 lock_page(old_page);
2198 if (!old_page->mapping) {
2199 ret = 0; /* retry the fault */
2200 unlock_page(old_page);
2201 goto unwritable_page;
2202 }
2203 } else
2204 VM_BUG_ON(!PageLocked(old_page));
David Howells9637a5e2006-06-23 02:03:43 -07002205
David Howells9637a5e2006-06-23 02:03:43 -07002206 /*
2207 * Since we dropped the lock we need to revalidate
2208 * the PTE as someone else may have changed it. If
2209 * they did, we just return, as we can count on the
2210 * MMU to tell us if they didn't also make it writable.
2211 */
2212 page_table = pte_offset_map_lock(mm, pmd, address,
2213 &ptl);
Nick Pigginb827e492009-04-30 15:08:16 -07002214 if (!pte_same(*page_table, orig_pte)) {
2215 unlock_page(old_page);
2216 page_cache_release(old_page);
David Howells9637a5e2006-06-23 02:03:43 -07002217 goto unlock;
Nick Pigginb827e492009-04-30 15:08:16 -07002218 }
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002219
2220 page_mkwrite = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002222 dirty_page = old_page;
2223 get_page(dirty_page);
David Howells9637a5e2006-06-23 02:03:43 -07002224 reuse = 1;
David Howells9637a5e2006-06-23 02:03:43 -07002225 }
2226
2227 if (reuse) {
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002228reuse:
David Howells9637a5e2006-06-23 02:03:43 -07002229 flush_cache_page(vma, address, pte_pfn(orig_pte));
2230 entry = pte_mkyoung(orig_pte);
2231 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
KAMEZAWA Hiroyuki954ffcb2007-10-16 01:25:44 -07002232 if (ptep_set_access_flags(vma, address, page_table, entry,1))
Russell King4b3073e2009-12-18 16:40:18 +00002233 update_mmu_cache(vma, address, page_table);
David Howells9637a5e2006-06-23 02:03:43 -07002234 ret |= VM_FAULT_WRITE;
2235 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237
2238 /*
2239 * Ok, we need to copy. Oh, well..
2240 */
Nick Pigginb5810032005-10-29 18:16:12 -07002241 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08002242gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002243 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244
2245 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07002246 goto oom;
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002247
Hugh Dickins62eede62009-09-21 17:03:34 -07002248 if (is_zero_pfn(pte_pfn(orig_pte))) {
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002249 new_page = alloc_zeroed_user_highpage_movable(vma, address);
2250 if (!new_page)
2251 goto oom;
2252 } else {
2253 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
2254 if (!new_page)
2255 goto oom;
2256 cow_user_page(new_page, old_page, address, vma);
2257 }
2258 __SetPageUptodate(new_page);
2259
Nick Pigginb291f002008-10-18 20:26:44 -07002260 /*
2261 * Don't let another task, with possibly unlocked vma,
2262 * keep the mlocked page.
2263 */
Carsten Otteab926612009-02-04 15:12:16 -08002264 if ((vma->vm_flags & VM_LOCKED) && old_page) {
Nick Pigginb291f002008-10-18 20:26:44 -07002265 lock_page(old_page); /* for LRU manipulation */
2266 clear_page_mlock(old_page);
2267 unlock_page(old_page);
2268 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002269
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002270 if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002271 goto oom_free_new;
2272
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 /*
2274 * Re-check the pte - we dropped the lock
2275 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002276 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07002277 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08002278 if (old_page) {
Hugh Dickins920fc352005-11-21 21:32:17 -08002279 if (!PageAnon(old_page)) {
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08002280 dec_mm_counter_fast(mm, MM_FILEPAGES);
2281 inc_mm_counter_fast(mm, MM_ANONPAGES);
Hugh Dickins920fc352005-11-21 21:32:17 -08002282 }
2283 } else
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08002284 inc_mm_counter_fast(mm, MM_ANONPAGES);
Ben Collinseca35132005-11-29 11:45:26 -08002285 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07002286 entry = mk_pte(new_page, vma->vm_page_prot);
2287 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Siddha, Suresh B4ce072f2006-09-29 01:58:42 -07002288 /*
2289 * Clear the pte entry and flush it first, before updating the
2290 * pte with the new entry. This will avoid a race condition
2291 * seen in the presence of one thread doing SMC and another
2292 * thread doing COW.
2293 */
Izik Eidus828502d2009-09-21 17:01:51 -07002294 ptep_clear_flush(vma, address, page_table);
Nick Piggin9617d952006-01-06 00:11:12 -08002295 page_add_new_anon_rmap(new_page, vma, address);
Izik Eidus828502d2009-09-21 17:01:51 -07002296 /*
2297 * We call the notify macro here because, when using secondary
2298 * mmu page tables (such as kvm shadow page tables), we want the
2299 * new page to be mapped directly into the secondary page table.
2300 */
2301 set_pte_at_notify(mm, address, page_table, entry);
Russell King4b3073e2009-12-18 16:40:18 +00002302 update_mmu_cache(vma, address, page_table);
Nick Piggin945754a2008-06-23 14:30:30 +02002303 if (old_page) {
2304 /*
2305 * Only after switching the pte to the new page may
2306 * we remove the mapcount here. Otherwise another
2307 * process may come and find the rmap count decremented
2308 * before the pte is switched to the new page, and
2309 * "reuse" the old page writing into it while our pte
2310 * here still points into it and can be read by other
2311 * threads.
2312 *
2313 * The critical issue is to order this
2314 * page_remove_rmap with the ptp_clear_flush above.
2315 * Those stores are ordered by (if nothing else,)
2316 * the barrier present in the atomic_add_negative
2317 * in page_remove_rmap.
2318 *
2319 * Then the TLB flush in ptep_clear_flush ensures that
2320 * no process can access the old page before the
2321 * decremented mapcount is visible. And the old page
2322 * cannot be reused until after the decremented
2323 * mapcount is visible. So transitively, TLBs to
2324 * old page will be flushed before it can be reused.
2325 */
Hugh Dickinsedc315f2009-01-06 14:40:11 -08002326 page_remove_rmap(old_page);
Nick Piggin945754a2008-06-23 14:30:30 +02002327 }
2328
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 /* Free the old page.. */
2330 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10002331 ret |= VM_FAULT_WRITE;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002332 } else
2333 mem_cgroup_uncharge_page(new_page);
2334
Hugh Dickins920fc352005-11-21 21:32:17 -08002335 if (new_page)
2336 page_cache_release(new_page);
2337 if (old_page)
2338 page_cache_release(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002339unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002340 pte_unmap_unlock(page_table, ptl);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002341 if (dirty_page) {
Nick Piggin79352892007-07-19 01:47:22 -07002342 /*
2343 * Yes, Virginia, this is actually required to prevent a race
2344 * with clear_page_dirty_for_io() from clearing the page dirty
2345 * bit after it clear all dirty ptes, but before a racing
2346 * do_wp_page installs a dirty pte.
2347 *
2348 * do_no_page is protected similarly.
2349 */
Nick Pigginb827e492009-04-30 15:08:16 -07002350 if (!page_mkwrite) {
2351 wait_on_page_locked(dirty_page);
2352 set_page_dirty_balance(dirty_page, page_mkwrite);
2353 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002354 put_page(dirty_page);
Nick Pigginb827e492009-04-30 15:08:16 -07002355 if (page_mkwrite) {
2356 struct address_space *mapping = dirty_page->mapping;
2357
2358 set_page_dirty(dirty_page);
2359 unlock_page(dirty_page);
2360 page_cache_release(dirty_page);
2361 if (mapping) {
2362 /*
2363 * Some device drivers do not set page.mapping
2364 * but still dirty their pages
2365 */
2366 balance_dirty_pages_ratelimited(mapping);
2367 }
2368 }
2369
2370 /* file_update_time outside page_lock */
2371 if (vma->vm_file)
2372 file_update_time(vma->vm_file);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002373 }
Nick Pigginf33ea7f2005-08-03 20:24:01 +10002374 return ret;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002375oom_free_new:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002376 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002377oom:
Nick Pigginb827e492009-04-30 15:08:16 -07002378 if (old_page) {
2379 if (page_mkwrite) {
2380 unlock_page(old_page);
2381 page_cache_release(old_page);
2382 }
Hugh Dickins920fc352005-11-21 21:32:17 -08002383 page_cache_release(old_page);
Nick Pigginb827e492009-04-30 15:08:16 -07002384 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 return VM_FAULT_OOM;
David Howells9637a5e2006-06-23 02:03:43 -07002386
2387unwritable_page:
2388 page_cache_release(old_page);
Nick Pigginc2ec1752009-03-31 15:23:21 -07002389 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390}
2391
2392/*
2393 * Helper functions for unmap_mapping_range().
2394 *
2395 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
2396 *
2397 * We have to restart searching the prio_tree whenever we drop the lock,
2398 * since the iterator is only valid while the lock is held, and anyway
2399 * a later vma might be split and reinserted earlier while lock dropped.
2400 *
2401 * The list of nonlinear vmas could be handled more efficiently, using
2402 * a placeholder, but handle it in the same way until a need is shown.
2403 * It is important to search the prio_tree before nonlinear list: a vma
2404 * may become nonlinear and be shifted from prio_tree to nonlinear list
2405 * while the lock is dropped; but never shifted from list to prio_tree.
2406 *
2407 * In order to make forward progress despite restarting the search,
2408 * vm_truncate_count is used to mark a vma as now dealt with, so we can
2409 * quickly skip it next time around. Since the prio_tree search only
2410 * shows us those vmas affected by unmapping the range in question, we
2411 * can't efficiently keep all vmas in step with mapping->truncate_count:
2412 * so instead reset them all whenever it wraps back to 0 (then go to 1).
2413 * mapping->truncate_count and vma->vm_truncate_count are protected by
2414 * i_mmap_lock.
2415 *
2416 * In order to make forward progress despite repeatedly restarting some
Hugh Dickinsee39b372005-04-19 13:29:15 -07002417 * large vma, note the restart_addr from unmap_vmas when it breaks out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 * and restart from that address when we reach that vma again. It might
2419 * have been split or merged, shrunk or extended, but never shifted: so
2420 * restart_addr remains valid so long as it remains in the vma's range.
2421 * unmap_mapping_range forces truncate_count to leap over page-aligned
2422 * values so we can save vma's restart_addr in its truncate_count field.
2423 */
2424#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
2425
2426static void reset_vma_truncate_counts(struct address_space *mapping)
2427{
2428 struct vm_area_struct *vma;
2429 struct prio_tree_iter iter;
2430
2431 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
2432 vma->vm_truncate_count = 0;
2433 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
2434 vma->vm_truncate_count = 0;
2435}
2436
2437static int unmap_mapping_range_vma(struct vm_area_struct *vma,
2438 unsigned long start_addr, unsigned long end_addr,
2439 struct zap_details *details)
2440{
2441 unsigned long restart_addr;
2442 int need_break;
2443
Nick Piggind00806b2007-07-19 01:46:57 -07002444 /*
2445 * files that support invalidating or truncating portions of the
Nick Piggind0217ac2007-07-19 01:47:03 -07002446 * file from under mmaped areas must have their ->fault function
Nick Piggin83c54072007-07-19 01:47:05 -07002447 * return a locked page (and set VM_FAULT_LOCKED in the return).
2448 * This provides synchronisation against concurrent unmapping here.
Nick Piggind00806b2007-07-19 01:46:57 -07002449 */
Nick Piggind00806b2007-07-19 01:46:57 -07002450
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451again:
2452 restart_addr = vma->vm_truncate_count;
2453 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
2454 start_addr = restart_addr;
2455 if (start_addr >= end_addr) {
2456 /* Top of vma has been split off since last time */
2457 vma->vm_truncate_count = details->truncate_count;
2458 return 0;
2459 }
2460 }
2461
Hugh Dickinsee39b372005-04-19 13:29:15 -07002462 restart_addr = zap_page_range(vma, start_addr,
2463 end_addr - start_addr, details);
Nick Piggin95c354f2008-01-30 13:31:20 +01002464 need_break = need_resched() || spin_needbreak(details->i_mmap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465
Hugh Dickinsee39b372005-04-19 13:29:15 -07002466 if (restart_addr >= end_addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 /* We have now completed this vma: mark it so */
2468 vma->vm_truncate_count = details->truncate_count;
2469 if (!need_break)
2470 return 0;
2471 } else {
2472 /* Note restart_addr in vma's truncate_count field */
Hugh Dickinsee39b372005-04-19 13:29:15 -07002473 vma->vm_truncate_count = restart_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 if (!need_break)
2475 goto again;
2476 }
2477
2478 spin_unlock(details->i_mmap_lock);
2479 cond_resched();
2480 spin_lock(details->i_mmap_lock);
2481 return -EINTR;
2482}
2483
2484static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
2485 struct zap_details *details)
2486{
2487 struct vm_area_struct *vma;
2488 struct prio_tree_iter iter;
2489 pgoff_t vba, vea, zba, zea;
2490
2491restart:
2492 vma_prio_tree_foreach(vma, &iter, root,
2493 details->first_index, details->last_index) {
2494 /* Skip quickly over those we have already dealt with */
2495 if (vma->vm_truncate_count == details->truncate_count)
2496 continue;
2497
2498 vba = vma->vm_pgoff;
2499 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
2500 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
2501 zba = details->first_index;
2502 if (zba < vba)
2503 zba = vba;
2504 zea = details->last_index;
2505 if (zea > vea)
2506 zea = vea;
2507
2508 if (unmap_mapping_range_vma(vma,
2509 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
2510 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
2511 details) < 0)
2512 goto restart;
2513 }
2514}
2515
2516static inline void unmap_mapping_range_list(struct list_head *head,
2517 struct zap_details *details)
2518{
2519 struct vm_area_struct *vma;
2520
2521 /*
2522 * In nonlinear VMAs there is no correspondence between virtual address
2523 * offset and file offset. So we must perform an exhaustive search
2524 * across *all* the pages in each nonlinear VMA, not just the pages
2525 * whose virtual address lies outside the file truncation point.
2526 */
2527restart:
2528 list_for_each_entry(vma, head, shared.vm_set.list) {
2529 /* Skip quickly over those we have already dealt with */
2530 if (vma->vm_truncate_count == details->truncate_count)
2531 continue;
2532 details->nonlinear_vma = vma;
2533 if (unmap_mapping_range_vma(vma, vma->vm_start,
2534 vma->vm_end, details) < 0)
2535 goto restart;
2536 }
2537}
2538
2539/**
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002540 * 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 -07002541 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 * @holebegin: byte in first page to unmap, relative to the start of
2543 * the underlying file. This will be rounded down to a PAGE_SIZE
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +10002544 * boundary. Note that this is different from truncate_pagecache(), which
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 * must keep the partial page. In contrast, we must get rid of
2546 * partial pages.
2547 * @holelen: size of prospective hole in bytes. This will be rounded
2548 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2549 * end of the file.
2550 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2551 * but 0 when invalidating pagecache, don't throw away private data.
2552 */
2553void unmap_mapping_range(struct address_space *mapping,
2554 loff_t const holebegin, loff_t const holelen, int even_cows)
2555{
2556 struct zap_details details;
2557 pgoff_t hba = holebegin >> PAGE_SHIFT;
2558 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2559
2560 /* Check for overflow. */
2561 if (sizeof(holelen) > sizeof(hlen)) {
2562 long long holeend =
2563 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2564 if (holeend & ~(long long)ULONG_MAX)
2565 hlen = ULONG_MAX - hba + 1;
2566 }
2567
2568 details.check_mapping = even_cows? NULL: mapping;
2569 details.nonlinear_vma = NULL;
2570 details.first_index = hba;
2571 details.last_index = hba + hlen - 1;
2572 if (details.last_index < details.first_index)
2573 details.last_index = ULONG_MAX;
2574 details.i_mmap_lock = &mapping->i_mmap_lock;
2575
2576 spin_lock(&mapping->i_mmap_lock);
2577
Nick Piggind00806b2007-07-19 01:46:57 -07002578 /* Protect against endless unmapping loops */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 mapping->truncate_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580 if (unlikely(is_restart_addr(mapping->truncate_count))) {
2581 if (mapping->truncate_count == 0)
2582 reset_vma_truncate_counts(mapping);
2583 mapping->truncate_count++;
2584 }
2585 details.truncate_count = mapping->truncate_count;
2586
2587 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
2588 unmap_mapping_range_tree(&mapping->i_mmap, &details);
2589 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
2590 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
2591 spin_unlock(&mapping->i_mmap_lock);
2592}
2593EXPORT_SYMBOL(unmap_mapping_range);
2594
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002595int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
2596{
2597 struct address_space *mapping = inode->i_mapping;
2598
2599 /*
2600 * If the underlying filesystem is not going to provide
2601 * a way to truncate a range of blocks (punch a hole) -
2602 * we should return failure right now.
2603 */
Al Viroacfa4382008-12-04 10:06:33 -05002604 if (!inode->i_op->truncate_range)
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002605 return -ENOSYS;
2606
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002607 mutex_lock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002608 down_write(&inode->i_alloc_sem);
2609 unmap_mapping_range(mapping, offset, (end - offset), 1);
2610 truncate_inode_pages_range(mapping, offset, end);
Nick Piggind00806b2007-07-19 01:46:57 -07002611 unmap_mapping_range(mapping, offset, (end - offset), 1);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002612 inode->i_op->truncate_range(inode, offset, end);
2613 up_write(&inode->i_alloc_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002614 mutex_unlock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002615
2616 return 0;
2617}
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002618
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002620 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2621 * but allow concurrent faults), and pte mapped but not yet locked.
2622 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002624static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2625 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002626 unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002628 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629 struct page *page;
Hugh Dickins65500d22005-10-29 18:15:59 -07002630 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 pte_t pte;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002632 struct mem_cgroup *ptr = NULL;
Nick Piggin83c54072007-07-19 01:47:05 -07002633 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002635 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002636 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002637
2638 entry = pte_to_swp_entry(orig_pte);
Andi Kleend1737fd2009-09-16 11:50:06 +02002639 if (unlikely(non_swap_entry(entry))) {
2640 if (is_migration_entry(entry)) {
2641 migration_entry_wait(mm, pmd, address);
2642 } else if (is_hwpoison_entry(entry)) {
2643 ret = VM_FAULT_HWPOISON;
2644 } else {
2645 print_bad_pte(vma, address, orig_pte, NULL);
Hugh Dickinsd99be1a2009-12-14 17:59:04 -08002646 ret = VM_FAULT_SIGBUS;
Andi Kleend1737fd2009-09-16 11:50:06 +02002647 }
Christoph Lameter06972122006-06-23 02:03:35 -07002648 goto out;
2649 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002650 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 page = lookup_swap_cache(entry);
2652 if (!page) {
Hugh Dickinsa5c9b692009-06-23 12:36:58 -07002653 grab_swap_token(mm); /* Contend for token _before_ read-in */
Hugh Dickins02098fe2008-02-04 22:28:42 -08002654 page = swapin_readahead(entry,
2655 GFP_HIGHUSER_MOVABLE, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656 if (!page) {
2657 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002658 * Back out if somebody else faulted in this pte
2659 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002661 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 if (likely(pte_same(*page_table, orig_pte)))
2663 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002664 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002665 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 }
2667
2668 /* Had to read the page from swap area: Major fault */
2669 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002670 count_vm_event(PGMAJFAULT);
Andi Kleend1737fd2009-09-16 11:50:06 +02002671 } else if (PageHWPoison(page)) {
Wu Fengguang71f72522009-12-16 12:19:58 +01002672 /*
2673 * hwpoisoned dirty swapcache pages are kept for killing
2674 * owner processes (which may be unknown at hwpoison time)
2675 */
Andi Kleend1737fd2009-09-16 11:50:06 +02002676 ret = VM_FAULT_HWPOISON;
2677 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Andi Kleen4779cb32009-10-14 01:51:41 +02002678 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679 }
2680
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681 lock_page(page);
Balbir Singh20a10222007-11-14 17:00:33 -08002682 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683
Hugh Dickins5ad64682009-12-14 17:59:24 -08002684 page = ksm_might_need_to_copy(page, vma, address);
2685 if (!page) {
2686 ret = VM_FAULT_OOM;
2687 goto out;
2688 }
2689
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002690 if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) {
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002691 ret = VM_FAULT_OOM;
Johannes Weinerbc43f752009-04-30 15:08:08 -07002692 goto out_page;
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002693 }
2694
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002696 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002698 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002699 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002700 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002701
2702 if (unlikely(!PageUptodate(page))) {
2703 ret = VM_FAULT_SIGBUS;
2704 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 }
2706
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002707 /*
2708 * The page isn't present yet, go ahead with the fault.
2709 *
2710 * Be careful about the sequence of operations here.
2711 * To get its accounting right, reuse_swap_page() must be called
2712 * while the page is counted on swap but not yet in mapcount i.e.
2713 * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
2714 * must be called after the swap_free(), or it will never succeed.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002715 * Because delete_from_swap_page() may be called by reuse_swap_page(),
2716 * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry
2717 * in page->private. In this case, a record in swap_cgroup is silently
2718 * discarded at swap_free().
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002719 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08002721 inc_mm_counter_fast(mm, MM_ANONPAGES);
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -08002722 dec_mm_counter_fast(mm, MM_SWAPENTS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723 pte = mk_pte(page, vma->vm_page_prot);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002724 if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002726 flags &= ~FAULT_FLAG_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728 flush_icache_page(vma, page);
2729 set_pte_at(mm, address, page_table, pte);
2730 page_add_anon_rmap(page, vma, address);
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002731 /* It's better to call commit-charge after rmap is established */
2732 mem_cgroup_commit_charge_swapin(page, ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002734 swap_free(entry);
Nick Pigginb291f002008-10-18 20:26:44 -07002735 if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08002736 try_to_free_swap(page);
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002737 unlock_page(page);
2738
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002739 if (flags & FAULT_FLAG_WRITE) {
Hugh Dickins61469f12008-03-04 14:29:04 -08002740 ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
2741 if (ret & VM_FAULT_ERROR)
2742 ret &= VM_FAULT_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743 goto out;
2744 }
2745
2746 /* No need to invalidate - it was non-present before */
Russell King4b3073e2009-12-18 16:40:18 +00002747 update_mmu_cache(vma, address, page_table);
Hugh Dickins65500d22005-10-29 18:15:59 -07002748unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002749 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750out:
2751 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002752out_nomap:
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002753 mem_cgroup_cancel_charge_swapin(ptr);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002754 pte_unmap_unlock(page_table, ptl);
Johannes Weinerbc43f752009-04-30 15:08:08 -07002755out_page:
Kirill Korotaevb8107482005-05-16 21:53:50 -07002756 unlock_page(page);
Andi Kleen4779cb32009-10-14 01:51:41 +02002757out_release:
Kirill Korotaevb8107482005-05-16 21:53:50 -07002758 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002759 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760}
2761
2762/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002763 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2764 * but allow concurrent faults), and pte mapped but not yet locked.
2765 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002767static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2768 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002769 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002771 struct page *page;
2772 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774
Hugh Dickins62eede62009-09-21 17:03:34 -07002775 if (!(flags & FAULT_FLAG_WRITE)) {
2776 entry = pte_mkspecial(pfn_pte(my_zero_pfn(address),
2777 vma->vm_page_prot));
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002778 ptl = pte_lockptr(mm, pmd);
2779 spin_lock(ptl);
2780 if (!pte_none(*page_table))
2781 goto unlock;
2782 goto setpte;
2783 }
2784
Nick Piggin557ed1f2007-10-16 01:24:40 -07002785 /* Allocate our own private page. */
2786 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787
Nick Piggin557ed1f2007-10-16 01:24:40 -07002788 if (unlikely(anon_vma_prepare(vma)))
2789 goto oom;
2790 page = alloc_zeroed_user_highpage_movable(vma, address);
2791 if (!page)
2792 goto oom;
Nick Piggin0ed361d2008-02-04 22:29:34 -08002793 __SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002795 if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002796 goto oom_free_page;
2797
Nick Piggin557ed1f2007-10-16 01:24:40 -07002798 entry = mk_pte(page, vma->vm_page_prot);
Hugh Dickins1ac0cb52009-09-21 17:03:29 -07002799 if (vma->vm_flags & VM_WRITE)
2800 entry = pte_mkwrite(pte_mkdirty(entry));
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002801
Nick Piggin557ed1f2007-10-16 01:24:40 -07002802 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Andrea Arcangeli1c2fb7a2009-09-21 17:02:22 -07002803 if (!pte_none(*page_table))
Nick Piggin557ed1f2007-10-16 01:24:40 -07002804 goto release;
Hugh Dickins9ba69292009-09-21 17:02:20 -07002805
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08002806 inc_mm_counter_fast(mm, MM_ANONPAGES);
Nick Piggin557ed1f2007-10-16 01:24:40 -07002807 page_add_new_anon_rmap(page, vma, address);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002808setpte:
Hugh Dickins65500d22005-10-29 18:15:59 -07002809 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810
2811 /* No need to invalidate - it was non-present before */
Russell King4b3073e2009-12-18 16:40:18 +00002812 update_mmu_cache(vma, address, page_table);
Hugh Dickins65500d22005-10-29 18:15:59 -07002813unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002814 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002815 return 0;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002816release:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002817 mem_cgroup_uncharge_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002818 page_cache_release(page);
2819 goto unlock;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002820oom_free_page:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002821 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002822oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823 return VM_FAULT_OOM;
2824}
2825
2826/*
Nick Piggin54cb8822007-07-19 01:46:59 -07002827 * __do_fault() tries to create a new page mapping. It aggressively
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 * tries to share with existing pages, but makes a separate copy if
Nick Piggin54cb8822007-07-19 01:46:59 -07002829 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2830 * the next page fault.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 *
2832 * As this is called only for pages that do not currently exist, we
2833 * do not need to flush old virtual caches or the TLB.
2834 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002835 * We enter with non-exclusive mmap_sem (to exclude vma changes,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002836 * but allow concurrent faults), and pte neither mapped nor locked.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002837 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 */
Nick Piggin54cb8822007-07-19 01:46:59 -07002839static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002840 unsigned long address, pmd_t *pmd,
Nick Piggin54cb8822007-07-19 01:46:59 -07002841 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842{
Hugh Dickins16abfa02007-10-04 16:56:06 +01002843 pte_t *page_table;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002844 spinlock_t *ptl;
Nick Piggind0217ac2007-07-19 01:47:03 -07002845 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847 int anon = 0;
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002848 int charged = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002849 struct page *dirty_page = NULL;
Nick Piggind0217ac2007-07-19 01:47:03 -07002850 struct vm_fault vmf;
2851 int ret;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002852 int page_mkwrite = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07002853
Nick Piggind0217ac2007-07-19 01:47:03 -07002854 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2855 vmf.pgoff = pgoff;
2856 vmf.flags = flags;
2857 vmf.page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858
Nick Piggin3c18ddd2008-04-28 02:12:10 -07002859 ret = vma->vm_ops->fault(vma, &vmf);
2860 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2861 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862
Andi Kleena3b947e2009-09-16 11:50:08 +02002863 if (unlikely(PageHWPoison(vmf.page))) {
2864 if (ret & VM_FAULT_LOCKED)
2865 unlock_page(vmf.page);
2866 return VM_FAULT_HWPOISON;
2867 }
2868
Nick Piggind00806b2007-07-19 01:46:57 -07002869 /*
Nick Piggind0217ac2007-07-19 01:47:03 -07002870 * For consistency in subsequent calls, make the faulted page always
Nick Piggind00806b2007-07-19 01:46:57 -07002871 * locked.
2872 */
Nick Piggin83c54072007-07-19 01:47:05 -07002873 if (unlikely(!(ret & VM_FAULT_LOCKED)))
Nick Piggind0217ac2007-07-19 01:47:03 -07002874 lock_page(vmf.page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002875 else
Nick Piggind0217ac2007-07-19 01:47:03 -07002876 VM_BUG_ON(!PageLocked(vmf.page));
Nick Piggind00806b2007-07-19 01:46:57 -07002877
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 /*
2879 * Should we do an early C-O-W break?
2880 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002881 page = vmf.page;
Nick Piggin54cb8822007-07-19 01:46:59 -07002882 if (flags & FAULT_FLAG_WRITE) {
David Howells9637a5e2006-06-23 02:03:43 -07002883 if (!(vma->vm_flags & VM_SHARED)) {
Nick Piggin54cb8822007-07-19 01:46:59 -07002884 anon = 1;
Nick Piggind00806b2007-07-19 01:46:57 -07002885 if (unlikely(anon_vma_prepare(vma))) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002886 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002887 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002888 }
Nick Piggin83c54072007-07-19 01:47:05 -07002889 page = alloc_page_vma(GFP_HIGHUSER_MOVABLE,
2890 vma, address);
Nick Piggind00806b2007-07-19 01:46:57 -07002891 if (!page) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002892 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002893 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002894 }
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002895 if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) {
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002896 ret = VM_FAULT_OOM;
2897 page_cache_release(page);
2898 goto out;
2899 }
2900 charged = 1;
Nick Pigginb291f002008-10-18 20:26:44 -07002901 /*
2902 * Don't let another task, with possibly unlocked vma,
2903 * keep the mlocked page.
2904 */
2905 if (vma->vm_flags & VM_LOCKED)
2906 clear_page_mlock(vmf.page);
Nick Piggind0217ac2007-07-19 01:47:03 -07002907 copy_user_highpage(page, vmf.page, address, vma);
Nick Piggin0ed361d2008-02-04 22:29:34 -08002908 __SetPageUptodate(page);
David Howells9637a5e2006-06-23 02:03:43 -07002909 } else {
Nick Piggin54cb8822007-07-19 01:46:59 -07002910 /*
2911 * If the page will be shareable, see if the backing
David Howells9637a5e2006-06-23 02:03:43 -07002912 * address space wants to know that the page is about
Nick Piggin54cb8822007-07-19 01:46:59 -07002913 * to become writable
2914 */
Mark Fasheh69676142007-07-19 01:47:00 -07002915 if (vma->vm_ops->page_mkwrite) {
Nick Pigginc2ec1752009-03-31 15:23:21 -07002916 int tmp;
2917
Mark Fasheh69676142007-07-19 01:47:00 -07002918 unlock_page(page);
Nick Pigginb827e492009-04-30 15:08:16 -07002919 vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
Nick Pigginc2ec1752009-03-31 15:23:21 -07002920 tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
2921 if (unlikely(tmp &
2922 (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
2923 ret = tmp;
Nick Pigginb827e492009-04-30 15:08:16 -07002924 goto unwritable_page;
Mark Fasheh69676142007-07-19 01:47:00 -07002925 }
Nick Pigginb827e492009-04-30 15:08:16 -07002926 if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
2927 lock_page(page);
2928 if (!page->mapping) {
2929 ret = 0; /* retry the fault */
2930 unlock_page(page);
2931 goto unwritable_page;
2932 }
2933 } else
2934 VM_BUG_ON(!PageLocked(page));
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002935 page_mkwrite = 1;
David Howells9637a5e2006-06-23 02:03:43 -07002936 }
2937 }
Nick Piggin54cb8822007-07-19 01:46:59 -07002938
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939 }
2940
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002941 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942
2943 /*
2944 * This silly early PAGE_DIRTY setting removes a race
2945 * due to the bad i386 page protection. But it's valid
2946 * for other architectures too.
2947 *
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002948 * Note that if FAULT_FLAG_WRITE is set, we either now have
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 * an exclusive copy of the page, or this is a shared mapping,
2950 * so we can make it writable and dirty to avoid having to
2951 * handle that later.
2952 */
2953 /* Only go through if we didn't race with anybody else... */
Andrea Arcangeli1c2fb7a2009-09-21 17:02:22 -07002954 if (likely(pte_same(*page_table, orig_pte))) {
Nick Piggind00806b2007-07-19 01:46:57 -07002955 flush_icache_page(vma, page);
2956 entry = mk_pte(page, vma->vm_page_prot);
Nick Piggin54cb8822007-07-19 01:46:59 -07002957 if (flags & FAULT_FLAG_WRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002958 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002959 if (anon) {
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08002960 inc_mm_counter_fast(mm, MM_ANONPAGES);
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002961 page_add_new_anon_rmap(page, vma, address);
Hugh Dickinsf57e88a2005-11-21 21:32:19 -08002962 } else {
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08002963 inc_mm_counter_fast(mm, MM_FILEPAGES);
Nick Piggind00806b2007-07-19 01:46:57 -07002964 page_add_file_rmap(page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002965 if (flags & FAULT_FLAG_WRITE) {
Nick Piggind00806b2007-07-19 01:46:57 -07002966 dirty_page = page;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002967 get_page(dirty_page);
2968 }
Hugh Dickins42946212005-10-29 18:16:05 -07002969 }
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002970 set_pte_at(mm, address, page_table, entry);
Nick Piggind00806b2007-07-19 01:46:57 -07002971
2972 /* no need to invalidate: a not-present page won't be cached */
Russell King4b3073e2009-12-18 16:40:18 +00002973 update_mmu_cache(vma, address, page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002974 } else {
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002975 if (charged)
2976 mem_cgroup_uncharge_page(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002977 if (anon)
2978 page_cache_release(page);
2979 else
Nick Piggin54cb8822007-07-19 01:46:59 -07002980 anon = 1; /* no anon but release faulted_page */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981 }
2982
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002983 pte_unmap_unlock(page_table, ptl);
Nick Piggind00806b2007-07-19 01:46:57 -07002984
2985out:
Nick Pigginb827e492009-04-30 15:08:16 -07002986 if (dirty_page) {
2987 struct address_space *mapping = page->mapping;
2988
2989 if (set_page_dirty(dirty_page))
2990 page_mkwrite = 1;
2991 unlock_page(dirty_page);
2992 put_page(dirty_page);
2993 if (page_mkwrite && mapping) {
2994 /*
2995 * Some device drivers do not set page.mapping but still
2996 * dirty their pages
2997 */
2998 balance_dirty_pages_ratelimited(mapping);
2999 }
3000
3001 /* file_update_time outside page_lock */
Anton Salikhmetov8f7b3d12008-01-23 02:21:18 +03003002 if (vma->vm_file)
3003 file_update_time(vma->vm_file);
Nick Pigginb827e492009-04-30 15:08:16 -07003004 } else {
3005 unlock_page(vmf.page);
3006 if (anon)
3007 page_cache_release(vmf.page);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07003008 }
Nick Piggind00806b2007-07-19 01:46:57 -07003009
Nick Piggin83c54072007-07-19 01:47:05 -07003010 return ret;
Nick Pigginb827e492009-04-30 15:08:16 -07003011
3012unwritable_page:
3013 page_cache_release(page);
3014 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07003015}
Nick Piggind00806b2007-07-19 01:46:57 -07003016
Nick Piggin54cb8822007-07-19 01:46:59 -07003017static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3018 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003019 unsigned int flags, pte_t orig_pte)
Nick Piggin54cb8822007-07-19 01:46:59 -07003020{
3021 pgoff_t pgoff = (((address & PAGE_MASK)
Dean Nelson0da7e012007-10-16 01:24:45 -07003022 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
Nick Piggin54cb8822007-07-19 01:46:59 -07003023
Hugh Dickins16abfa02007-10-04 16:56:06 +01003024 pte_unmap(page_table);
3025 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Nick Piggin54cb8822007-07-19 01:46:59 -07003026}
3027
Jes Sorensenf4b81802006-09-27 01:50:10 -07003028/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029 * Fault of a previously existing named mapping. Repopulate the pte
3030 * from the encoded file_pte if possible. This enables swappable
3031 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003032 *
3033 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3034 * but allow concurrent faults), and pte mapped but not yet locked.
3035 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036 */
Nick Piggind0217ac2007-07-19 01:47:03 -07003037static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins65500d22005-10-29 18:15:59 -07003038 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003039 unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040{
Hugh Dickins65500d22005-10-29 18:15:59 -07003041 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003043 flags |= FAULT_FLAG_NONLINEAR;
3044
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07003045 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Nick Piggin83c54072007-07-19 01:47:05 -07003046 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003047
Hugh Dickins2509ef22009-01-06 14:40:10 -08003048 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
Hugh Dickins65500d22005-10-29 18:15:59 -07003049 /*
3050 * Page table corrupted: show pte and kill process.
3051 */
Hugh Dickins3dc14742009-01-06 14:40:08 -08003052 print_bad_pte(vma, address, orig_pte, NULL);
Hugh Dickinsd99be1a2009-12-14 17:59:04 -08003053 return VM_FAULT_SIGBUS;
Hugh Dickins65500d22005-10-29 18:15:59 -07003054 }
Hugh Dickins65500d22005-10-29 18:15:59 -07003055
3056 pgoff = pte_to_pgoff(orig_pte);
Hugh Dickins16abfa02007-10-04 16:56:06 +01003057 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058}
3059
3060/*
3061 * These routines also need to handle stuff like marking pages dirty
3062 * and/or accessed for architectures that don't do it in hardware (most
3063 * RISC architectures). The early dirtying is also good on the i386.
3064 *
3065 * There is also a hook called "update_mmu_cache()" that architectures
3066 * with external mmu caches can use to update those (ie the Sparc or
3067 * PowerPC hashed page tables that act as extended TLBs).
3068 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07003069 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3070 * but allow concurrent faults), and pte mapped but not yet locked.
3071 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003072 */
3073static inline int handle_pte_fault(struct mm_struct *mm,
Hugh Dickins65500d22005-10-29 18:15:59 -07003074 struct vm_area_struct *vma, unsigned long address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003075 pte_t *pte, pmd_t *pmd, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003076{
3077 pte_t entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003078 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07003080 entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003081 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07003082 if (pte_none(entry)) {
Jes Sorensenf4b81802006-09-27 01:50:10 -07003083 if (vma->vm_ops) {
Nick Piggin3c18ddd2008-04-28 02:12:10 -07003084 if (likely(vma->vm_ops->fault))
Nick Piggin54cb8822007-07-19 01:46:59 -07003085 return do_linear_fault(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003086 pte, pmd, flags, entry);
Jes Sorensenf4b81802006-09-27 01:50:10 -07003087 }
3088 return do_anonymous_page(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003089 pte, pmd, flags);
Hugh Dickins65500d22005-10-29 18:15:59 -07003090 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091 if (pte_file(entry))
Nick Piggind0217ac2007-07-19 01:47:03 -07003092 return do_nonlinear_fault(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003093 pte, pmd, flags, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07003094 return do_swap_page(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003095 pte, pmd, flags, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096 }
3097
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07003098 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003099 spin_lock(ptl);
3100 if (unlikely(!pte_same(*pte, entry)))
3101 goto unlock;
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003102 if (flags & FAULT_FLAG_WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003103 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003104 return do_wp_page(mm, vma, address,
3105 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106 entry = pte_mkdirty(entry);
3107 }
3108 entry = pte_mkyoung(entry);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003109 if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) {
Russell King4b3073e2009-12-18 16:40:18 +00003110 update_mmu_cache(vma, address, pte);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07003111 } else {
3112 /*
3113 * This is needed only for protection faults but the arch code
3114 * is not yet telling us if this is a protection fault or not.
3115 * This still avoids useless tlb flushes for .text page faults
3116 * with threads.
3117 */
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003118 if (flags & FAULT_FLAG_WRITE)
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07003119 flush_tlb_page(vma, address);
3120 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07003121unlock:
3122 pte_unmap_unlock(pte, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07003123 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003124}
3125
3126/*
3127 * By the time we get here, we already hold the mm semaphore
3128 */
Nick Piggin83c54072007-07-19 01:47:05 -07003129int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvaldsd06063c2009-04-10 09:01:23 -07003130 unsigned long address, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003131{
3132 pgd_t *pgd;
3133 pud_t *pud;
3134 pmd_t *pmd;
3135 pte_t *pte;
3136
3137 __set_current_state(TASK_RUNNING);
3138
Christoph Lameterf8891e52006-06-30 01:55:45 -07003139 count_vm_event(PGFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003140
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -08003141 /* do counter updates before entering really critical section. */
3142 check_sync_rss_stat(current);
3143
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01003144 if (unlikely(is_vm_hugetlb_page(vma)))
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003145 return hugetlb_fault(mm, vma, address, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003146
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003148 pud = pud_alloc(mm, pgd, address);
3149 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07003150 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003151 pmd = pmd_alloc(mm, pud, address);
3152 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07003153 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154 pte = pte_alloc_map(mm, pmd, address);
3155 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -07003156 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003157
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003158 return handle_pte_fault(mm, vma, address, pte, pmd, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003159}
3160
3161#ifndef __PAGETABLE_PUD_FOLDED
3162/*
3163 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07003164 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003165 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07003166int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167{
Hugh Dickinsc74df322005-10-29 18:16:23 -07003168 pud_t *new = pud_alloc_one(mm, address);
3169 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07003170 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003171
Nick Piggin362a61a2008-05-14 06:37:36 +02003172 smp_wmb(); /* See comment in __pte_alloc */
3173
Hugh Dickins872fec12005-10-29 18:16:21 -07003174 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003175 if (pgd_present(*pgd)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003176 pud_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003177 else
3178 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07003179 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003180 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003181}
3182#endif /* __PAGETABLE_PUD_FOLDED */
3183
3184#ifndef __PAGETABLE_PMD_FOLDED
3185/*
3186 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07003187 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003188 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07003189int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003190{
Hugh Dickinsc74df322005-10-29 18:16:23 -07003191 pmd_t *new = pmd_alloc_one(mm, address);
3192 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07003193 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003194
Nick Piggin362a61a2008-05-14 06:37:36 +02003195 smp_wmb(); /* See comment in __pte_alloc */
3196
Hugh Dickins872fec12005-10-29 18:16:21 -07003197 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003198#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07003199 if (pud_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003200 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003201 else
3202 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07003204 if (pgd_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003205 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003206 else
3207 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003208#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07003209 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003210 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003211}
3212#endif /* __PAGETABLE_PMD_FOLDED */
3213
3214int make_pages_present(unsigned long addr, unsigned long end)
3215{
3216 int ret, len, write;
3217 struct vm_area_struct * vma;
3218
3219 vma = find_vma(current->mm, addr);
3220 if (!vma)
KOSAKI Motohiroa4770972008-08-04 13:41:14 -07003221 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003222 write = (vma->vm_flags & VM_WRITE) != 0;
Eric Sesterhenn5bcb28b2006-03-26 18:30:52 +02003223 BUG_ON(addr >= end);
3224 BUG_ON(end > vma->vm_end);
Rolf Eike Beer68e116a2007-07-15 23:38:03 -07003225 len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003226 ret = get_user_pages(current, current->mm, addr,
3227 len, write, 0, NULL, NULL);
Lee Schermerhornc11d69d2008-10-18 20:26:56 -07003228 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003229 return ret;
Lee Schermerhorn9978ad52008-10-18 20:26:56 -07003230 return ret == len ? 0 : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003231}
3232
Linus Torvalds1da177e2005-04-16 15:20:36 -07003233#if !defined(__HAVE_ARCH_GATE_AREA)
3234
3235#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07003236static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003237
3238static int __init gate_vma_init(void)
3239{
3240 gate_vma.vm_mm = NULL;
3241 gate_vma.vm_start = FIXADDR_USER_START;
3242 gate_vma.vm_end = FIXADDR_USER_END;
Roland McGrathb6558c42007-01-26 00:56:47 -08003243 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
3244 gate_vma.vm_page_prot = __P101;
Roland McGrathf47aef52007-01-26 00:56:49 -08003245 /*
3246 * Make sure the vDSO gets into every core dump.
3247 * Dumping its contents makes post-mortem fully interpretable later
3248 * without matching up the same kernel and hardware config to see
3249 * what PC values meant.
3250 */
3251 gate_vma.vm_flags |= VM_ALWAYSDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003252 return 0;
3253}
3254__initcall(gate_vma_init);
3255#endif
3256
3257struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
3258{
3259#ifdef AT_SYSINFO_EHDR
3260 return &gate_vma;
3261#else
3262 return NULL;
3263#endif
3264}
3265
3266int in_gate_area_no_task(unsigned long addr)
3267{
3268#ifdef AT_SYSINFO_EHDR
3269 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
3270 return 1;
3271#endif
3272 return 0;
3273}
3274
3275#endif /* __HAVE_ARCH_GATE_AREA */
David Howells0ec76a12006-09-27 01:50:15 -07003276
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07003277static int follow_pte(struct mm_struct *mm, unsigned long address,
3278 pte_t **ptepp, spinlock_t **ptlp)
3279{
3280 pgd_t *pgd;
3281 pud_t *pud;
3282 pmd_t *pmd;
3283 pte_t *ptep;
3284
3285 pgd = pgd_offset(mm, address);
3286 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
3287 goto out;
3288
3289 pud = pud_offset(pgd, address);
3290 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
3291 goto out;
3292
3293 pmd = pmd_offset(pud, address);
3294 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
3295 goto out;
3296
3297 /* We cannot handle huge page PFN maps. Luckily they don't exist. */
3298 if (pmd_huge(*pmd))
3299 goto out;
3300
3301 ptep = pte_offset_map_lock(mm, pmd, address, ptlp);
3302 if (!ptep)
3303 goto out;
3304 if (!pte_present(*ptep))
3305 goto unlock;
3306 *ptepp = ptep;
3307 return 0;
3308unlock:
3309 pte_unmap_unlock(ptep, *ptlp);
3310out:
3311 return -EINVAL;
3312}
3313
Johannes Weiner3b6748e2009-06-16 15:32:35 -07003314/**
3315 * follow_pfn - look up PFN at a user virtual address
3316 * @vma: memory mapping
3317 * @address: user virtual address
3318 * @pfn: location to store found PFN
3319 *
3320 * Only IO mappings and raw PFN mappings are allowed.
3321 *
3322 * Returns zero and the pfn at @pfn on success, -ve otherwise.
3323 */
3324int follow_pfn(struct vm_area_struct *vma, unsigned long address,
3325 unsigned long *pfn)
3326{
3327 int ret = -EINVAL;
3328 spinlock_t *ptl;
3329 pte_t *ptep;
3330
3331 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3332 return ret;
3333
3334 ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);
3335 if (ret)
3336 return ret;
3337 *pfn = pte_pfn(*ptep);
3338 pte_unmap_unlock(ptep, ptl);
3339 return 0;
3340}
3341EXPORT_SYMBOL(follow_pfn);
3342
Rik van Riel28b2ee22008-07-23 21:27:05 -07003343#ifdef CONFIG_HAVE_IOREMAP_PROT
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003344int follow_phys(struct vm_area_struct *vma,
3345 unsigned long address, unsigned int flags,
3346 unsigned long *prot, resource_size_t *phys)
Rik van Riel28b2ee22008-07-23 21:27:05 -07003347{
Johannes Weiner03668a42009-06-16 15:32:34 -07003348 int ret = -EINVAL;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003349 pte_t *ptep, pte;
3350 spinlock_t *ptl;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003351
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003352 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3353 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003354
Johannes Weiner03668a42009-06-16 15:32:34 -07003355 if (follow_pte(vma->vm_mm, address, &ptep, &ptl))
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003356 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003357 pte = *ptep;
Johannes Weiner03668a42009-06-16 15:32:34 -07003358
Rik van Riel28b2ee22008-07-23 21:27:05 -07003359 if ((flags & FOLL_WRITE) && !pte_write(pte))
3360 goto unlock;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003361
3362 *prot = pgprot_val(pte_pgprot(pte));
Johannes Weiner03668a42009-06-16 15:32:34 -07003363 *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003364
Johannes Weiner03668a42009-06-16 15:32:34 -07003365 ret = 0;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003366unlock:
3367 pte_unmap_unlock(ptep, ptl);
3368out:
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003369 return ret;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003370}
3371
3372int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
3373 void *buf, int len, int write)
3374{
3375 resource_size_t phys_addr;
3376 unsigned long prot = 0;
KOSAKI Motohiro2bc72732009-01-06 14:39:43 -08003377 void __iomem *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003378 int offset = addr & (PAGE_SIZE-1);
3379
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003380 if (follow_phys(vma, addr, write, &prot, &phys_addr))
Rik van Riel28b2ee22008-07-23 21:27:05 -07003381 return -EINVAL;
3382
3383 maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot);
3384 if (write)
3385 memcpy_toio(maddr + offset, buf, len);
3386 else
3387 memcpy_fromio(buf, maddr + offset, len);
3388 iounmap(maddr);
3389
3390 return len;
3391}
3392#endif
3393
David Howells0ec76a12006-09-27 01:50:15 -07003394/*
3395 * Access another process' address space.
3396 * Source/target buffer must be kernel space,
3397 * Do not walk the page table directly, use get_user_pages
3398 */
3399int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
3400{
3401 struct mm_struct *mm;
3402 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07003403 void *old_buf = buf;
3404
3405 mm = get_task_mm(tsk);
3406 if (!mm)
3407 return 0;
3408
3409 down_read(&mm->mmap_sem);
Simon Arlott183ff222007-10-20 01:27:18 +02003410 /* ignore errors, just check how much was successfully transferred */
David Howells0ec76a12006-09-27 01:50:15 -07003411 while (len) {
3412 int bytes, ret, offset;
3413 void *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003414 struct page *page = NULL;
David Howells0ec76a12006-09-27 01:50:15 -07003415
3416 ret = get_user_pages(tsk, mm, addr, 1,
3417 write, 1, &page, &vma);
Rik van Riel28b2ee22008-07-23 21:27:05 -07003418 if (ret <= 0) {
3419 /*
3420 * Check if this is a VM_IO | VM_PFNMAP VMA, which
3421 * we can access using slightly different code.
3422 */
3423#ifdef CONFIG_HAVE_IOREMAP_PROT
3424 vma = find_vma(mm, addr);
3425 if (!vma)
3426 break;
3427 if (vma->vm_ops && vma->vm_ops->access)
3428 ret = vma->vm_ops->access(vma, addr, buf,
3429 len, write);
3430 if (ret <= 0)
3431#endif
3432 break;
3433 bytes = ret;
David Howells0ec76a12006-09-27 01:50:15 -07003434 } else {
Rik van Riel28b2ee22008-07-23 21:27:05 -07003435 bytes = len;
3436 offset = addr & (PAGE_SIZE-1);
3437 if (bytes > PAGE_SIZE-offset)
3438 bytes = PAGE_SIZE-offset;
3439
3440 maddr = kmap(page);
3441 if (write) {
3442 copy_to_user_page(vma, page, addr,
3443 maddr + offset, buf, bytes);
3444 set_page_dirty_lock(page);
3445 } else {
3446 copy_from_user_page(vma, page, addr,
3447 buf, maddr + offset, bytes);
3448 }
3449 kunmap(page);
3450 page_cache_release(page);
David Howells0ec76a12006-09-27 01:50:15 -07003451 }
David Howells0ec76a12006-09-27 01:50:15 -07003452 len -= bytes;
3453 buf += bytes;
3454 addr += bytes;
3455 }
3456 up_read(&mm->mmap_sem);
3457 mmput(mm);
3458
3459 return buf - old_buf;
3460}
Andi Kleen03252912008-01-30 13:33:18 +01003461
3462/*
3463 * Print the name of a VMA.
3464 */
3465void print_vma_addr(char *prefix, unsigned long ip)
3466{
3467 struct mm_struct *mm = current->mm;
3468 struct vm_area_struct *vma;
3469
Ingo Molnare8bff742008-02-13 20:21:06 +01003470 /*
3471 * Do not print if we are in atomic
3472 * contexts (in exception stacks, etc.):
3473 */
3474 if (preempt_count())
3475 return;
3476
Andi Kleen03252912008-01-30 13:33:18 +01003477 down_read(&mm->mmap_sem);
3478 vma = find_vma(mm, ip);
3479 if (vma && vma->vm_file) {
3480 struct file *f = vma->vm_file;
3481 char *buf = (char *)__get_free_page(GFP_KERNEL);
3482 if (buf) {
3483 char *p, *s;
3484
Jan Blunckcf28b482008-02-14 19:38:44 -08003485 p = d_path(&f->f_path, buf, PAGE_SIZE);
Andi Kleen03252912008-01-30 13:33:18 +01003486 if (IS_ERR(p))
3487 p = "?";
3488 s = strrchr(p, '/');
3489 if (s)
3490 p = s+1;
3491 printk("%s%s[%lx+%lx]", prefix, p,
3492 vma->vm_start,
3493 vma->vm_end - vma->vm_start);
3494 free_page((unsigned long)buf);
3495 }
3496 }
3497 up_read(&current->mm->mmap_sem);
3498}
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003499
3500#ifdef CONFIG_PROVE_LOCKING
3501void might_fault(void)
3502{
Peter Zijlstra95156f02009-01-12 13:02:11 +01003503 /*
3504 * Some code (nfs/sunrpc) uses socket ops on kernel memory while
3505 * holding the mmap_sem, this is safe because kernel memory doesn't
3506 * get paged out, therefore we'll never actually fault, and the
3507 * below annotations will generate false positives.
3508 */
3509 if (segment_eq(get_fs(), KERNEL_DS))
3510 return;
3511
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003512 might_sleep();
3513 /*
3514 * it would be nicer only to annotate paths which are not under
3515 * pagefault_disable, however that requires a larger audit and
3516 * providing helpers like get_user_atomic.
3517 */
3518 if (!in_atomic() && current->mm)
3519 might_lock_read(&current->mm->mmap_sem);
3520}
3521EXPORT_SYMBOL(might_fault);
3522#endif