blob: 6ab19dd4a1990f3f463c1db3eee4f19ce4f5a3de [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
Alexey Dobriyan6952b612009-09-18 23:55:55 +040060#include <asm/io.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070061#include <asm/pgalloc.h>
62#include <asm/uaccess.h>
63#include <asm/tlb.h>
64#include <asm/tlbflush.h>
65#include <asm/pgtable.h>
66
Jan Beulich42b77722008-07-23 21:27:10 -070067#include "internal.h"
68
Andy Whitcroftd41dee32005-06-23 00:07:54 -070069#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -070070/* use the per-pgdat data instead for discontigmem - mbligh */
71unsigned long max_mapnr;
72struct page *mem_map;
73
74EXPORT_SYMBOL(max_mapnr);
75EXPORT_SYMBOL(mem_map);
76#endif
77
78unsigned long num_physpages;
79/*
80 * A number of key systems in x86 including ioremap() rely on the assumption
81 * that high_memory defines the upper bound on direct map memory, then end
82 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
83 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
84 * and ZONE_HIGHMEM.
85 */
86void * high_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -070087
88EXPORT_SYMBOL(num_physpages);
89EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Ingo Molnar32a93232008-02-06 22:39:44 +010091/*
92 * Randomize the address space (stacks, mmaps, brk, etc.).
93 *
94 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
95 * as ancient (libc5 based) binaries can segfault. )
96 */
97int randomize_va_space __read_mostly =
98#ifdef CONFIG_COMPAT_BRK
99 1;
100#else
101 2;
102#endif
Andi Kleena62eaf12006-02-16 23:41:58 +0100103
104static int __init disable_randmaps(char *s)
105{
106 randomize_va_space = 0;
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800107 return 1;
Andi Kleena62eaf12006-02-16 23:41:58 +0100108}
109__setup("norandmaps", disable_randmaps);
110
Hugh Dickins62eede62009-09-21 17:03:34 -0700111unsigned long zero_pfn __read_mostly;
Hugh Dickins03f64622009-09-21 17:03:35 -0700112unsigned long highest_memmap_pfn __read_mostly;
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -0700113
114/*
115 * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init()
116 */
117static int __init init_zero_pfn(void)
118{
119 zero_pfn = page_to_pfn(ZERO_PAGE(0));
120 return 0;
121}
122core_initcall(init_zero_pfn);
Andi Kleena62eaf12006-02-16 23:41:58 +0100123
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124/*
125 * If a p?d_bad entry is found while walking page tables, report
126 * the error, before resetting entry to p?d_none. Usually (but
127 * very seldom) called out from the p?d_none_or_clear_bad macros.
128 */
129
130void pgd_clear_bad(pgd_t *pgd)
131{
132 pgd_ERROR(*pgd);
133 pgd_clear(pgd);
134}
135
136void pud_clear_bad(pud_t *pud)
137{
138 pud_ERROR(*pud);
139 pud_clear(pud);
140}
141
142void pmd_clear_bad(pmd_t *pmd)
143{
144 pmd_ERROR(*pmd);
145 pmd_clear(pmd);
146}
147
148/*
149 * Note: this doesn't free the actual pages themselves. That
150 * has been handled earlier when unmapping all the memory regions.
151 */
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000152static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
153 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800155 pgtable_t token = pmd_pgtable(*pmd);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700156 pmd_clear(pmd);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000157 pte_free_tlb(tlb, token, addr);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700158 tlb->mm->nr_ptes--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159}
160
Hugh Dickinse0da3822005-04-19 13:29:15 -0700161static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
162 unsigned long addr, unsigned long end,
163 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164{
165 pmd_t *pmd;
166 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700167 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168
Hugh Dickinse0da3822005-04-19 13:29:15 -0700169 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 do {
172 next = pmd_addr_end(addr, end);
173 if (pmd_none_or_clear_bad(pmd))
174 continue;
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000175 free_pte_range(tlb, pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 } while (pmd++, addr = next, addr != end);
177
Hugh Dickinse0da3822005-04-19 13:29:15 -0700178 start &= PUD_MASK;
179 if (start < floor)
180 return;
181 if (ceiling) {
182 ceiling &= PUD_MASK;
183 if (!ceiling)
184 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700186 if (end - 1 > ceiling - 1)
187 return;
188
189 pmd = pmd_offset(pud, start);
190 pud_clear(pud);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000191 pmd_free_tlb(tlb, pmd, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192}
193
Hugh Dickinse0da3822005-04-19 13:29:15 -0700194static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
195 unsigned long addr, unsigned long end,
196 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197{
198 pud_t *pud;
199 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700200 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201
Hugh Dickinse0da3822005-04-19 13:29:15 -0700202 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 pud = pud_offset(pgd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 do {
205 next = pud_addr_end(addr, end);
206 if (pud_none_or_clear_bad(pud))
207 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700208 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 } while (pud++, addr = next, addr != end);
210
Hugh Dickinse0da3822005-04-19 13:29:15 -0700211 start &= PGDIR_MASK;
212 if (start < floor)
213 return;
214 if (ceiling) {
215 ceiling &= PGDIR_MASK;
216 if (!ceiling)
217 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700219 if (end - 1 > ceiling - 1)
220 return;
221
222 pud = pud_offset(pgd, start);
223 pgd_clear(pgd);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000224 pud_free_tlb(tlb, pud, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225}
226
227/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700228 * This function frees user-level page tables of a process.
229 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 * Must be called with pagetable lock held.
231 */
Jan Beulich42b77722008-07-23 21:27:10 -0700232void free_pgd_range(struct mmu_gather *tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700233 unsigned long addr, unsigned long end,
234 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235{
236 pgd_t *pgd;
237 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700238 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
Hugh Dickinse0da3822005-04-19 13:29:15 -0700240 /*
241 * The next few lines have given us lots of grief...
242 *
243 * Why are we testing PMD* at this top level? Because often
244 * there will be no work to do at all, and we'd prefer not to
245 * go all the way down to the bottom just to discover that.
246 *
247 * Why all these "- 1"s? Because 0 represents both the bottom
248 * of the address space and the top of it (using -1 for the
249 * top wouldn't help much: the masks would do the wrong thing).
250 * The rule is that addr 0 and floor 0 refer to the bottom of
251 * the address space, but end 0 and ceiling 0 refer to the top
252 * Comparisons need to use "end - 1" and "ceiling - 1" (though
253 * that end 0 case should be mythical).
254 *
255 * Wherever addr is brought up or ceiling brought down, we must
256 * be careful to reject "the opposite 0" before it confuses the
257 * subsequent tests. But what about where end is brought down
258 * by PMD_SIZE below? no, end can't go down to 0 there.
259 *
260 * Whereas we round start (addr) and ceiling down, by different
261 * masks at different levels, in order to test whether a table
262 * now has no other vmas using it, so can be freed, we don't
263 * bother to round floor or end up - the tests don't need that.
264 */
265
266 addr &= PMD_MASK;
267 if (addr < floor) {
268 addr += PMD_SIZE;
269 if (!addr)
270 return;
271 }
272 if (ceiling) {
273 ceiling &= PMD_MASK;
274 if (!ceiling)
275 return;
276 }
277 if (end - 1 > ceiling - 1)
278 end -= PMD_SIZE;
279 if (addr > end - 1)
280 return;
281
282 start = addr;
Jan Beulich42b77722008-07-23 21:27:10 -0700283 pgd = pgd_offset(tlb->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 do {
285 next = pgd_addr_end(addr, end);
286 if (pgd_none_or_clear_bad(pgd))
287 continue;
Jan Beulich42b77722008-07-23 21:27:10 -0700288 free_pud_range(tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700290}
291
Jan Beulich42b77722008-07-23 21:27:10 -0700292void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700293 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700294{
295 while (vma) {
296 struct vm_area_struct *next = vma->vm_next;
297 unsigned long addr = vma->vm_start;
298
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700299 /*
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000300 * Hide vma from rmap and truncate_pagecache before freeing
301 * pgtables
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700302 */
303 anon_vma_unlink(vma);
304 unlink_file_vma(vma);
305
David Gibson9da61ae2006-03-22 00:08:57 -0800306 if (is_vm_hugetlb_page(vma)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700307 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700308 floor, next? next->vm_start: ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700309 } else {
310 /*
311 * Optimization: gather nearby vmas into one call down
312 */
313 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
David Gibson48669202006-03-22 00:08:58 -0800314 && !is_vm_hugetlb_page(next)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700315 vma = next;
316 next = vma->vm_next;
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700317 anon_vma_unlink(vma);
318 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700319 }
320 free_pgd_range(tlb, addr, vma->vm_end,
321 floor, next? next->vm_start: ceiling);
322 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700323 vma = next;
324 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325}
326
Hugh Dickins1bb36302005-10-29 18:16:22 -0700327int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800329 pgtable_t new = pte_alloc_one(mm, address);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700330 if (!new)
331 return -ENOMEM;
332
Nick Piggin362a61a2008-05-14 06:37:36 +0200333 /*
334 * Ensure all pte setup (eg. pte page lock and page clearing) are
335 * visible before the pte is made visible to other CPUs by being
336 * put into page tables.
337 *
338 * The other side of the story is the pointer chasing in the page
339 * table walking code (when walking the page table without locking;
340 * ie. most of the time). Fortunately, these data accesses consist
341 * of a chain of data-dependent loads, meaning most CPUs (alpha
342 * being the notable exception) will already guarantee loads are
343 * seen in-order. See the alpha page table accessors for the
344 * smp_read_barrier_depends() barriers in page table walking code.
345 */
346 smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
347
Hugh Dickinsc74df322005-10-29 18:16:23 -0700348 spin_lock(&mm->page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800349 if (!pmd_present(*pmd)) { /* Has another populated it ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 mm->nr_ptes++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 pmd_populate(mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800352 new = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 }
Hugh Dickinsc74df322005-10-29 18:16:23 -0700354 spin_unlock(&mm->page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800355 if (new)
356 pte_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700357 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358}
359
Hugh Dickins1bb36302005-10-29 18:16:22 -0700360int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700362 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
363 if (!new)
364 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365
Nick Piggin362a61a2008-05-14 06:37:36 +0200366 smp_wmb(); /* See comment in __pte_alloc */
367
Hugh Dickins1bb36302005-10-29 18:16:22 -0700368 spin_lock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800369 if (!pmd_present(*pmd)) { /* Has another populated it ? */
Hugh Dickins1bb36302005-10-29 18:16:22 -0700370 pmd_populate_kernel(&init_mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800371 new = NULL;
372 }
Hugh Dickins1bb36302005-10-29 18:16:22 -0700373 spin_unlock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800374 if (new)
375 pte_free_kernel(&init_mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700376 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377}
378
Hugh Dickinsae859762005-10-29 18:16:05 -0700379static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
380{
381 if (file_rss)
382 add_mm_counter(mm, file_rss, file_rss);
383 if (anon_rss)
384 add_mm_counter(mm, anon_rss, anon_rss);
385}
386
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800388 * This function is called to print an error when a bad pte
389 * is found. For example, we might have a PFN-mapped pte in
390 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700391 *
392 * The calling function must still handle the error.
393 */
Hugh Dickins3dc14742009-01-06 14:40:08 -0800394static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
395 pte_t pte, struct page *page)
Nick Pigginb5810032005-10-29 18:16:12 -0700396{
Hugh Dickins3dc14742009-01-06 14:40:08 -0800397 pgd_t *pgd = pgd_offset(vma->vm_mm, addr);
398 pud_t *pud = pud_offset(pgd, addr);
399 pmd_t *pmd = pmd_offset(pud, addr);
400 struct address_space *mapping;
401 pgoff_t index;
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800402 static unsigned long resume;
403 static unsigned long nr_shown;
404 static unsigned long nr_unshown;
405
406 /*
407 * Allow a burst of 60 reports, then keep quiet for that minute;
408 * or allow a steady drip of one report per second.
409 */
410 if (nr_shown == 60) {
411 if (time_before(jiffies, resume)) {
412 nr_unshown++;
413 return;
414 }
415 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800416 printk(KERN_ALERT
417 "BUG: Bad page map: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800418 nr_unshown);
419 nr_unshown = 0;
420 }
421 nr_shown = 0;
422 }
423 if (nr_shown++ == 0)
424 resume = jiffies + 60 * HZ;
Hugh Dickins3dc14742009-01-06 14:40:08 -0800425
426 mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;
427 index = linear_page_index(vma, addr);
428
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800429 printk(KERN_ALERT
430 "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800431 current->comm,
432 (long long)pte_val(pte), (long long)pmd_val(*pmd));
433 if (page) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800434 printk(KERN_ALERT
Hugh Dickins3dc14742009-01-06 14:40:08 -0800435 "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
436 page, (void *)page->flags, page_count(page),
437 page_mapcount(page), page->mapping, page->index);
438 }
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800439 printk(KERN_ALERT
Hugh Dickins3dc14742009-01-06 14:40:08 -0800440 "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
441 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
442 /*
443 * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y
444 */
445 if (vma->vm_ops)
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800446 print_symbol(KERN_ALERT "vma->vm_ops->fault: %s\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800447 (unsigned long)vma->vm_ops->fault);
448 if (vma->vm_file && vma->vm_file->f_op)
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800449 print_symbol(KERN_ALERT "vma->vm_file->f_op->mmap: %s\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800450 (unsigned long)vma->vm_file->f_op->mmap);
Nick Pigginb5810032005-10-29 18:16:12 -0700451 dump_stack();
Hugh Dickins3dc14742009-01-06 14:40:08 -0800452 add_taint(TAINT_BAD_PAGE);
Nick Pigginb5810032005-10-29 18:16:12 -0700453}
454
Linus Torvalds67121172005-12-11 20:38:17 -0800455static inline int is_cow_mapping(unsigned int flags)
456{
457 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
458}
459
Hugh Dickins62eede62009-09-21 17:03:34 -0700460#ifndef is_zero_pfn
461static inline int is_zero_pfn(unsigned long pfn)
462{
463 return pfn == zero_pfn;
464}
465#endif
466
467#ifndef my_zero_pfn
468static inline unsigned long my_zero_pfn(unsigned long addr)
469{
470 return zero_pfn;
471}
472#endif
473
Nick Pigginb5810032005-10-29 18:16:12 -0700474/*
Nick Piggin7e675132008-04-28 02:13:00 -0700475 * vm_normal_page -- This function gets the "struct page" associated with a pte.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800476 *
Nick Piggin7e675132008-04-28 02:13:00 -0700477 * "Special" mappings do not wish to be associated with a "struct page" (either
478 * it doesn't exist, or it exists but they don't want to touch it). In this
479 * case, NULL is returned here. "Normal" mappings do have a struct page.
Jared Hulbertb379d792008-04-28 02:12:58 -0700480 *
Nick Piggin7e675132008-04-28 02:13:00 -0700481 * There are 2 broad cases. Firstly, an architecture may define a pte_special()
482 * pte bit, in which case this function is trivial. Secondly, an architecture
483 * may not have a spare pte bit, which requires a more complicated scheme,
484 * described below.
485 *
486 * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
487 * special mapping (even if there are underlying and valid "struct pages").
488 * COWed pages of a VM_PFNMAP are always normal.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800489 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700490 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
491 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
Nick Piggin7e675132008-04-28 02:13:00 -0700492 * set, and the vm_pgoff will point to the first PFN mapped: thus every special
493 * mapping will always honor the rule
Linus Torvalds6aab3412005-11-28 14:34:23 -0800494 *
495 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
496 *
Nick Piggin7e675132008-04-28 02:13:00 -0700497 * And for normal mappings this is false.
Jared Hulbertb379d792008-04-28 02:12:58 -0700498 *
Nick Piggin7e675132008-04-28 02:13:00 -0700499 * This restricts such mappings to be a linear translation from virtual address
500 * to pfn. To get around this restriction, we allow arbitrary mappings so long
501 * as the vma is not a COW mapping; in that case, we know that all ptes are
502 * special (because none can have been COWed).
Jared Hulbertb379d792008-04-28 02:12:58 -0700503 *
504 *
Nick Piggin7e675132008-04-28 02:13:00 -0700505 * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
506 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700507 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
508 * page" backing, however the difference is that _all_ pages with a struct
509 * page (that is, those where pfn_valid is true) are refcounted and considered
510 * normal pages by the VM. The disadvantage is that pages are refcounted
511 * (which can be slower and simply not an option for some PFNMAP users). The
512 * advantage is that we don't have to follow the strict linearity rule of
513 * PFNMAP mappings in order to support COWable mappings.
514 *
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800515 */
Nick Piggin7e675132008-04-28 02:13:00 -0700516#ifdef __HAVE_ARCH_PTE_SPECIAL
517# define HAVE_PTE_SPECIAL 1
518#else
519# define HAVE_PTE_SPECIAL 0
520#endif
521struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
522 pte_t pte)
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800523{
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800524 unsigned long pfn = pte_pfn(pte);
Nick Piggin7e675132008-04-28 02:13:00 -0700525
526 if (HAVE_PTE_SPECIAL) {
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800527 if (likely(!pte_special(pte)))
528 goto check_pfn;
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -0700529 if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))
530 return NULL;
Hugh Dickins62eede62009-09-21 17:03:34 -0700531 if (!is_zero_pfn(pfn))
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800532 print_bad_pte(vma, addr, pte, NULL);
Nick Piggin7e675132008-04-28 02:13:00 -0700533 return NULL;
534 }
535
536 /* !HAVE_PTE_SPECIAL case follows: */
537
Jared Hulbertb379d792008-04-28 02:12:58 -0700538 if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
539 if (vma->vm_flags & VM_MIXEDMAP) {
540 if (!pfn_valid(pfn))
541 return NULL;
542 goto out;
543 } else {
Nick Piggin7e675132008-04-28 02:13:00 -0700544 unsigned long off;
545 off = (addr - vma->vm_start) >> PAGE_SHIFT;
Jared Hulbertb379d792008-04-28 02:12:58 -0700546 if (pfn == vma->vm_pgoff + off)
547 return NULL;
548 if (!is_cow_mapping(vma->vm_flags))
549 return NULL;
550 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800551 }
552
Hugh Dickins62eede62009-09-21 17:03:34 -0700553 if (is_zero_pfn(pfn))
554 return NULL;
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800555check_pfn:
556 if (unlikely(pfn > highest_memmap_pfn)) {
557 print_bad_pte(vma, addr, pte, NULL);
558 return NULL;
559 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800560
561 /*
Nick Piggin7e675132008-04-28 02:13:00 -0700562 * NOTE! We still have PageReserved() pages in the page tables.
Nick Piggin7e675132008-04-28 02:13:00 -0700563 * eg. VDSO mappings can cause them to exist.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800564 */
Jared Hulbertb379d792008-04-28 02:12:58 -0700565out:
Linus Torvalds6aab3412005-11-28 14:34:23 -0800566 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800567}
568
569/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 * copy one vm_area from one task to the other. Assumes the page tables
571 * already present in the new task to be cleared in the whole range
572 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 */
574
Hugh Dickins8c103762005-10-29 18:16:13 -0700575static inline void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700577 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700578 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579{
Nick Pigginb5810032005-10-29 18:16:12 -0700580 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 pte_t pte = *src_pte;
582 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
584 /* pte contains position in swap or file, so copy. */
585 if (unlikely(!pte_present(pte))) {
586 if (!pte_file(pte)) {
Christoph Lameter06972122006-06-23 02:03:35 -0700587 swp_entry_t entry = pte_to_swp_entry(pte);
588
589 swap_duplicate(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 /* make sure dst_mm is on swapoff's mmlist. */
591 if (unlikely(list_empty(&dst_mm->mmlist))) {
592 spin_lock(&mmlist_lock);
Hugh Dickinsf412ac02005-10-29 18:16:41 -0700593 if (list_empty(&dst_mm->mmlist))
594 list_add(&dst_mm->mmlist,
595 &src_mm->mmlist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 spin_unlock(&mmlist_lock);
597 }
Christoph Lameter06972122006-06-23 02:03:35 -0700598 if (is_write_migration_entry(entry) &&
599 is_cow_mapping(vm_flags)) {
600 /*
601 * COW mappings require pages in both parent
602 * and child to be set to read.
603 */
604 make_migration_entry_read(&entry);
605 pte = swp_entry_to_pte(entry);
606 set_pte_at(src_mm, addr, src_pte, pte);
607 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700609 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 }
611
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 /*
613 * If it's a COW mapping, write protect it both
614 * in the parent and the child
615 */
Linus Torvalds67121172005-12-11 20:38:17 -0800616 if (is_cow_mapping(vm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 ptep_set_wrprotect(src_mm, addr, src_pte);
Zachary Amsden3dc90792006-09-30 23:29:30 -0700618 pte = pte_wrprotect(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 }
620
621 /*
622 * If it's a shared mapping, mark it clean in
623 * the child
624 */
625 if (vm_flags & VM_SHARED)
626 pte = pte_mkclean(pte);
627 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800628
629 page = vm_normal_page(vma, addr, pte);
630 if (page) {
631 get_page(page);
Hugh Dickins21333b22009-09-21 17:01:59 -0700632 page_dup_rmap(page);
Johannes Weinerb7c46d12009-09-21 17:02:56 -0700633 rss[PageAnon(page)]++;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800634 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700635
636out_set_pte:
637 set_pte_at(dst_mm, addr, dst_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638}
639
640static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
641 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
642 unsigned long addr, unsigned long end)
643{
Daisuke Nishimurac36987e2009-10-26 16:50:23 -0700644 pte_t *orig_src_pte, *orig_dst_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700646 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -0700647 int progress = 0;
Hugh Dickins8c103762005-10-29 18:16:13 -0700648 int rss[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649
650again:
Hugh Dickinsae859762005-10-29 18:16:05 -0700651 rss[1] = rss[0] = 0;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700652 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 if (!dst_pte)
654 return -ENOMEM;
655 src_pte = pte_offset_map_nested(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700656 src_ptl = pte_lockptr(src_mm, src_pmd);
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700657 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
Daisuke Nishimurac36987e2009-10-26 16:50:23 -0700658 orig_src_pte = src_pte;
659 orig_dst_pte = dst_pte;
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700660 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 do {
663 /*
664 * We are holding two locks at this point - either of them
665 * could generate latencies in another task on another CPU.
666 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700667 if (progress >= 32) {
668 progress = 0;
669 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +0100670 spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700671 break;
672 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 if (pte_none(*src_pte)) {
674 progress++;
675 continue;
676 }
Hugh Dickins8c103762005-10-29 18:16:13 -0700677 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 progress += 8;
679 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700681 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -0700682 spin_unlock(src_ptl);
Daisuke Nishimurac36987e2009-10-26 16:50:23 -0700683 pte_unmap_nested(orig_src_pte);
Hugh Dickinsae859762005-10-29 18:16:05 -0700684 add_mm_rss(dst_mm, rss[0], rss[1]);
Daisuke Nishimurac36987e2009-10-26 16:50:23 -0700685 pte_unmap_unlock(orig_dst_pte, dst_ptl);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700686 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 if (addr != end)
688 goto again;
689 return 0;
690}
691
692static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
693 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
694 unsigned long addr, unsigned long end)
695{
696 pmd_t *src_pmd, *dst_pmd;
697 unsigned long next;
698
699 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
700 if (!dst_pmd)
701 return -ENOMEM;
702 src_pmd = pmd_offset(src_pud, addr);
703 do {
704 next = pmd_addr_end(addr, end);
705 if (pmd_none_or_clear_bad(src_pmd))
706 continue;
707 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
708 vma, addr, next))
709 return -ENOMEM;
710 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
711 return 0;
712}
713
714static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
715 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
716 unsigned long addr, unsigned long end)
717{
718 pud_t *src_pud, *dst_pud;
719 unsigned long next;
720
721 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
722 if (!dst_pud)
723 return -ENOMEM;
724 src_pud = pud_offset(src_pgd, addr);
725 do {
726 next = pud_addr_end(addr, end);
727 if (pud_none_or_clear_bad(src_pud))
728 continue;
729 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
730 vma, addr, next))
731 return -ENOMEM;
732 } while (dst_pud++, src_pud++, addr = next, addr != end);
733 return 0;
734}
735
736int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
737 struct vm_area_struct *vma)
738{
739 pgd_t *src_pgd, *dst_pgd;
740 unsigned long next;
741 unsigned long addr = vma->vm_start;
742 unsigned long end = vma->vm_end;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700743 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744
Nick Piggind9928952005-08-28 16:49:11 +1000745 /*
746 * Don't copy ptes where a page fault will fill them correctly.
747 * Fork becomes much lighter when there are big shared or private
748 * readonly mappings. The tradeoff is that copy_page_range is more
749 * efficient than faulting.
750 */
Linus Torvalds4d7672b2005-12-16 10:21:23 -0800751 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
Nick Piggind9928952005-08-28 16:49:11 +1000752 if (!vma->anon_vma)
753 return 0;
754 }
755
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 if (is_vm_hugetlb_page(vma))
757 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
758
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800759 if (unlikely(is_pfn_mapping(vma))) {
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -0800760 /*
761 * We do not free on error cases below as remove_vma
762 * gets called on error from higher level routine
763 */
764 ret = track_pfn_vma_copy(vma);
765 if (ret)
766 return ret;
767 }
768
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700769 /*
770 * We need to invalidate the secondary MMU mappings only when
771 * there could be a permission downgrade on the ptes of the
772 * parent mm. And a permission downgrade will only happen if
773 * is_cow_mapping() returns true.
774 */
775 if (is_cow_mapping(vma->vm_flags))
776 mmu_notifier_invalidate_range_start(src_mm, addr, end);
777
778 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 dst_pgd = pgd_offset(dst_mm, addr);
780 src_pgd = pgd_offset(src_mm, addr);
781 do {
782 next = pgd_addr_end(addr, end);
783 if (pgd_none_or_clear_bad(src_pgd))
784 continue;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700785 if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
786 vma, addr, next))) {
787 ret = -ENOMEM;
788 break;
789 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700791
792 if (is_cow_mapping(vma->vm_flags))
793 mmu_notifier_invalidate_range_end(src_mm,
794 vma->vm_start, end);
795 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796}
797
Robin Holt51c6f662005-11-13 16:06:42 -0800798static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700799 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800801 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802{
Nick Pigginb5810032005-10-29 18:16:12 -0700803 struct mm_struct *mm = tlb->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 pte_t *pte;
Hugh Dickins508034a2005-10-29 18:16:30 -0700805 spinlock_t *ptl;
Hugh Dickinsae859762005-10-29 18:16:05 -0700806 int file_rss = 0;
807 int anon_rss = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808
Hugh Dickins508034a2005-10-29 18:16:30 -0700809 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700810 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 do {
812 pte_t ptent = *pte;
Robin Holt51c6f662005-11-13 16:06:42 -0800813 if (pte_none(ptent)) {
814 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800816 }
Hugh Dickins6f5e6b92006-03-16 23:04:09 -0800817
818 (*zap_work) -= PAGE_SIZE;
819
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800821 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -0800822
Linus Torvalds6aab3412005-11-28 14:34:23 -0800823 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 if (unlikely(details) && page) {
825 /*
826 * unmap_shared_mapping_pages() wants to
827 * invalidate cache without truncating:
828 * unmap shared but keep private pages.
829 */
830 if (details->check_mapping &&
831 details->check_mapping != page->mapping)
832 continue;
833 /*
834 * Each page->index must be checked when
835 * invalidating or truncating nonlinear.
836 */
837 if (details->nonlinear_vma &&
838 (page->index < details->first_index ||
839 page->index > details->last_index))
840 continue;
841 }
Nick Pigginb5810032005-10-29 18:16:12 -0700842 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -0700843 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 tlb_remove_tlb_entry(tlb, pte, addr);
845 if (unlikely(!page))
846 continue;
847 if (unlikely(details) && details->nonlinear_vma
848 && linear_page_index(details->nonlinear_vma,
849 addr) != page->index)
Nick Pigginb5810032005-10-29 18:16:12 -0700850 set_pte_at(mm, addr, pte,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 pgoff_to_pte(page->index));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 if (PageAnon(page))
Hugh Dickins86d912f2005-10-29 18:16:14 -0700853 anon_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700854 else {
855 if (pte_dirty(ptent))
856 set_page_dirty(page);
Johannes Weiner4917e5d2009-01-06 14:39:17 -0800857 if (pte_young(ptent) &&
858 likely(!VM_SequentialReadHint(vma)))
Nick Pigginbf3f3bc2009-01-06 14:38:55 -0800859 mark_page_accessed(page);
Hugh Dickins86d912f2005-10-29 18:16:14 -0700860 file_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700861 }
Hugh Dickinsedc315f2009-01-06 14:40:11 -0800862 page_remove_rmap(page);
Hugh Dickins3dc14742009-01-06 14:40:08 -0800863 if (unlikely(page_mapcount(page) < 0))
864 print_bad_pte(vma, addr, ptent, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 tlb_remove_page(tlb, page);
866 continue;
867 }
868 /*
869 * If details->check_mapping, we leave swap entries;
870 * if details->nonlinear_vma, we leave file entries.
871 */
872 if (unlikely(details))
873 continue;
Hugh Dickins2509ef22009-01-06 14:40:10 -0800874 if (pte_file(ptent)) {
875 if (unlikely(!(vma->vm_flags & VM_NONLINEAR)))
876 print_bad_pte(vma, addr, ptent, NULL);
877 } else if
878 (unlikely(!free_swap_and_cache(pte_to_swp_entry(ptent))))
879 print_bad_pte(vma, addr, ptent, NULL);
Zachary Amsden9888a1c2006-09-30 23:29:31 -0700880 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
Robin Holt51c6f662005-11-13 16:06:42 -0800881 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
Hugh Dickinsae859762005-10-29 18:16:05 -0700882
Hugh Dickins86d912f2005-10-29 18:16:14 -0700883 add_mm_rss(mm, file_rss, anon_rss);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700884 arch_leave_lazy_mmu_mode();
Hugh Dickins508034a2005-10-29 18:16:30 -0700885 pte_unmap_unlock(pte - 1, ptl);
Robin Holt51c6f662005-11-13 16:06:42 -0800886
887 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888}
889
Robin Holt51c6f662005-11-13 16:06:42 -0800890static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700891 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800893 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894{
895 pmd_t *pmd;
896 unsigned long next;
897
898 pmd = pmd_offset(pud, addr);
899 do {
900 next = pmd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800901 if (pmd_none_or_clear_bad(pmd)) {
902 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800904 }
905 next = zap_pte_range(tlb, vma, pmd, addr, next,
906 zap_work, details);
907 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
908
909 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910}
911
Robin Holt51c6f662005-11-13 16:06:42 -0800912static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700913 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800915 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916{
917 pud_t *pud;
918 unsigned long next;
919
920 pud = pud_offset(pgd, addr);
921 do {
922 next = pud_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800923 if (pud_none_or_clear_bad(pud)) {
924 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800926 }
927 next = zap_pmd_range(tlb, vma, pud, addr, next,
928 zap_work, details);
929 } while (pud++, addr = next, (addr != end && *zap_work > 0));
930
931 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932}
933
Robin Holt51c6f662005-11-13 16:06:42 -0800934static unsigned long unmap_page_range(struct mmu_gather *tlb,
935 struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800937 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938{
939 pgd_t *pgd;
940 unsigned long next;
941
942 if (details && !details->check_mapping && !details->nonlinear_vma)
943 details = NULL;
944
945 BUG_ON(addr >= end);
946 tlb_start_vma(tlb, vma);
947 pgd = pgd_offset(vma->vm_mm, addr);
948 do {
949 next = pgd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800950 if (pgd_none_or_clear_bad(pgd)) {
951 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800953 }
954 next = zap_pud_range(tlb, vma, pgd, addr, next,
955 zap_work, details);
956 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 tlb_end_vma(tlb, vma);
Robin Holt51c6f662005-11-13 16:06:42 -0800958
959 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960}
961
962#ifdef CONFIG_PREEMPT
963# define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
964#else
965/* No preempt: go for improved straight-line efficiency */
966# define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
967#endif
968
969/**
970 * unmap_vmas - unmap a range of memory covered by a list of vma's
971 * @tlbp: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 * @vma: the starting vma
973 * @start_addr: virtual address at which to start unmapping
974 * @end_addr: virtual address at which to end unmapping
975 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
976 * @details: details of nonlinear truncation or shared cache invalidation
977 *
Hugh Dickinsee39b372005-04-19 13:29:15 -0700978 * Returns the end address of the unmapping (restart addr if interrupted).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700980 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700982 * We aim to not hold locks for too long (for scheduling latency reasons).
983 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 * return the ending mmu_gather to the caller.
985 *
986 * Only addresses between `start' and `end' will be unmapped.
987 *
988 * The VMA list must be sorted in ascending virtual address order.
989 *
990 * unmap_vmas() assumes that the caller will flush the whole unmapped address
991 * range after unmap_vmas() returns. So the only responsibility here is to
992 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
993 * drops the lock and schedules.
994 */
Hugh Dickins508034a2005-10-29 18:16:30 -0700995unsigned long unmap_vmas(struct mmu_gather **tlbp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 struct vm_area_struct *vma, unsigned long start_addr,
997 unsigned long end_addr, unsigned long *nr_accounted,
998 struct zap_details *details)
999{
Robin Holt51c6f662005-11-13 16:06:42 -08001000 long zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
1002 int tlb_start_valid = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -07001003 unsigned long start = start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -07001005 int fullmm = (*tlbp)->fullmm;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001006 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001008 mmu_notifier_invalidate_range_start(mm, start_addr, end_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 unsigned long end;
1011
1012 start = max(vma->vm_start, start_addr);
1013 if (start >= vma->vm_end)
1014 continue;
1015 end = min(vma->vm_end, end_addr);
1016 if (end <= vma->vm_start)
1017 continue;
1018
1019 if (vma->vm_flags & VM_ACCOUNT)
1020 *nr_accounted += (end - start) >> PAGE_SHIFT;
1021
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -08001022 if (unlikely(is_pfn_mapping(vma)))
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001023 untrack_pfn_vma(vma, 0, 0);
1024
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 while (start != end) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 if (!tlb_start_valid) {
1027 tlb_start = start;
1028 tlb_start_valid = 1;
1029 }
1030
Robin Holt51c6f662005-11-13 16:06:42 -08001031 if (unlikely(is_vm_hugetlb_page(vma))) {
Andi Kleena137e1c2008-07-23 21:27:43 -07001032 /*
1033 * It is undesirable to test vma->vm_file as it
1034 * should be non-null for valid hugetlb area.
1035 * However, vm_file will be NULL in the error
1036 * cleanup path of do_mmap_pgoff. When
1037 * hugetlbfs ->mmap method fails,
1038 * do_mmap_pgoff() nullifies vma->vm_file
1039 * before calling this function to clean up.
1040 * Since no pte has actually been setup, it is
1041 * safe to do nothing in this case.
1042 */
1043 if (vma->vm_file) {
1044 unmap_hugepage_range(vma, start, end, NULL);
1045 zap_work -= (end - start) /
Andi Kleena5516432008-07-23 21:27:41 -07001046 pages_per_huge_page(hstate_vma(vma));
Andi Kleena137e1c2008-07-23 21:27:43 -07001047 }
1048
Robin Holt51c6f662005-11-13 16:06:42 -08001049 start = end;
1050 } else
1051 start = unmap_page_range(*tlbp, vma,
1052 start, end, &zap_work, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053
Robin Holt51c6f662005-11-13 16:06:42 -08001054 if (zap_work > 0) {
1055 BUG_ON(start != end);
1056 break;
1057 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058
1059 tlb_finish_mmu(*tlbp, tlb_start, start);
1060
1061 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +01001062 (i_mmap_lock && spin_needbreak(i_mmap_lock))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 if (i_mmap_lock) {
Hugh Dickins508034a2005-10-29 18:16:30 -07001064 *tlbp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 goto out;
1066 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 }
1069
Hugh Dickins508034a2005-10-29 18:16:30 -07001070 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 tlb_start_valid = 0;
Robin Holt51c6f662005-11-13 16:06:42 -08001072 zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 }
1074 }
1075out:
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001076 mmu_notifier_invalidate_range_end(mm, start_addr, end_addr);
Hugh Dickinsee39b372005-04-19 13:29:15 -07001077 return start; /* which is now the end (or restart) address */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078}
1079
1080/**
1081 * zap_page_range - remove user pages in a given range
1082 * @vma: vm_area_struct holding the applicable pages
1083 * @address: starting address of pages to zap
1084 * @size: number of bytes to zap
1085 * @details: details of nonlinear truncation or shared cache invalidation
1086 */
Hugh Dickinsee39b372005-04-19 13:29:15 -07001087unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 unsigned long size, struct zap_details *details)
1089{
1090 struct mm_struct *mm = vma->vm_mm;
1091 struct mmu_gather *tlb;
1092 unsigned long end = address + size;
1093 unsigned long nr_accounted = 0;
1094
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001097 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -07001098 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
1099 if (tlb)
1100 tlb_finish_mmu(tlb, address, end);
Hugh Dickinsee39b372005-04-19 13:29:15 -07001101 return end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102}
1103
Jack Steinerc627f9c2008-07-29 22:33:53 -07001104/**
1105 * zap_vma_ptes - remove ptes mapping the vma
1106 * @vma: vm_area_struct holding ptes to be zapped
1107 * @address: starting address of pages to zap
1108 * @size: number of bytes to zap
1109 *
1110 * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1111 *
1112 * The entire address range must be fully contained within the vma.
1113 *
1114 * Returns 0 if successful.
1115 */
1116int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
1117 unsigned long size)
1118{
1119 if (address < vma->vm_start || address + size > vma->vm_end ||
1120 !(vma->vm_flags & VM_PFNMAP))
1121 return -1;
1122 zap_page_range(vma, address, size, NULL);
1123 return 0;
1124}
1125EXPORT_SYMBOL_GPL(zap_vma_ptes);
1126
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127/*
1128 * Do a quick page-table lookup for a single page.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 */
Linus Torvalds6aab3412005-11-28 14:34:23 -08001130struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001131 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132{
1133 pgd_t *pgd;
1134 pud_t *pud;
1135 pmd_t *pmd;
1136 pte_t *ptep, pte;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001137 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 struct page *page;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001139 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001141 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
1142 if (!IS_ERR(page)) {
1143 BUG_ON(flags & FOLL_GET);
1144 goto out;
1145 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001147 page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 pgd = pgd_offset(mm, address);
1149 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001150 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151
1152 pud = pud_offset(pgd, address);
Andi Kleenceb86872008-07-23 21:27:50 -07001153 if (pud_none(*pud))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001154 goto no_page_table;
Andi Kleenceb86872008-07-23 21:27:50 -07001155 if (pud_huge(*pud)) {
1156 BUG_ON(flags & FOLL_GET);
1157 page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE);
1158 goto out;
1159 }
1160 if (unlikely(pud_bad(*pud)))
1161 goto no_page_table;
1162
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 pmd = pmd_offset(pud, address);
Hugh Dickinsaeed5fc2008-05-06 20:49:23 +01001164 if (pmd_none(*pmd))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001165 goto no_page_table;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001166 if (pmd_huge(*pmd)) {
1167 BUG_ON(flags & FOLL_GET);
1168 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
1169 goto out;
1170 }
Hugh Dickinsaeed5fc2008-05-06 20:49:23 +01001171 if (unlikely(pmd_bad(*pmd)))
1172 goto no_page_table;
1173
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001174 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175
1176 pte = *ptep;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001177 if (!pte_present(pte))
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001178 goto no_page;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001179 if ((flags & FOLL_WRITE) && !pte_write(pte))
1180 goto unlock;
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07001181
Linus Torvalds6aab3412005-11-28 14:34:23 -08001182 page = vm_normal_page(vma, address, pte);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07001183 if (unlikely(!page)) {
1184 if ((flags & FOLL_DUMP) ||
Hugh Dickins62eede62009-09-21 17:03:34 -07001185 !is_zero_pfn(pte_pfn(pte)))
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07001186 goto bad_page;
1187 page = pte_page(pte);
1188 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001189
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001190 if (flags & FOLL_GET)
1191 get_page(page);
1192 if (flags & FOLL_TOUCH) {
1193 if ((flags & FOLL_WRITE) &&
1194 !pte_dirty(pte) && !PageDirty(page))
1195 set_page_dirty(page);
KOSAKI Motohirobd775c42009-03-31 15:19:37 -07001196 /*
1197 * pte_mkyoung() would be more correct here, but atomic care
1198 * is needed to avoid losing the dirty bit: it is easier to use
1199 * mark_page_accessed().
1200 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001201 mark_page_accessed(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001203unlock:
1204 pte_unmap_unlock(ptep, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205out:
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001206 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001208bad_page:
1209 pte_unmap_unlock(ptep, ptl);
1210 return ERR_PTR(-EFAULT);
1211
1212no_page:
1213 pte_unmap_unlock(ptep, ptl);
1214 if (!pte_none(pte))
1215 return page;
Hugh Dickins8e4b9a602009-09-21 17:03:26 -07001216
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001217no_page_table:
1218 /*
1219 * When core dumping an enormous anonymous area that nobody
Hugh Dickins8e4b9a602009-09-21 17:03:26 -07001220 * has touched so far, we don't want to allocate unnecessary pages or
1221 * page tables. Return error instead of NULL to skip handle_mm_fault,
1222 * then get_dump_page() will return NULL to leave a hole in the dump.
1223 * But we can only make this optimization where a hole would surely
1224 * be zero-filled if handle_mm_fault() actually did handle it.
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001225 */
Hugh Dickins8e4b9a602009-09-21 17:03:26 -07001226 if ((flags & FOLL_DUMP) &&
1227 (!vma->vm_ops || !vma->vm_ops->fault))
1228 return ERR_PTR(-EFAULT);
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001229 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230}
1231
Nick Pigginb291f002008-10-18 20:26:44 -07001232int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
Hugh Dickins58fa8792009-09-21 17:03:31 -07001233 unsigned long start, int nr_pages, unsigned int gup_flags,
Peter Zijlstra9d737772009-06-25 11:58:55 +02001234 struct page **pages, struct vm_area_struct **vmas)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235{
1236 int i;
Hugh Dickins58fa8792009-09-21 17:03:31 -07001237 unsigned long vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238
Peter Zijlstra9d737772009-06-25 11:58:55 +02001239 if (nr_pages <= 0)
Jonathan Corbet900cf082008-02-11 16:17:33 -07001240 return 0;
Hugh Dickins58fa8792009-09-21 17:03:31 -07001241
1242 VM_BUG_ON(!!pages != !!(gup_flags & FOLL_GET));
1243
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 /*
1245 * Require read or write permissions.
Hugh Dickins58fa8792009-09-21 17:03:31 -07001246 * If FOLL_FORCE is set, we only require the "MAY" flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 */
Hugh Dickins58fa8792009-09-21 17:03:31 -07001248 vm_flags = (gup_flags & FOLL_WRITE) ?
1249 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
1250 vm_flags &= (gup_flags & FOLL_FORCE) ?
1251 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 i = 0;
1253
1254 do {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001255 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256
1257 vma = find_extend_vma(mm, start);
1258 if (!vma && in_gate_area(tsk, start)) {
1259 unsigned long pg = start & PAGE_MASK;
1260 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
1261 pgd_t *pgd;
1262 pud_t *pud;
1263 pmd_t *pmd;
1264 pte_t *pte;
Nick Pigginb291f002008-10-18 20:26:44 -07001265
1266 /* user gate pages are read-only */
Hugh Dickins58fa8792009-09-21 17:03:31 -07001267 if (gup_flags & FOLL_WRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 return i ? : -EFAULT;
1269 if (pg > TASK_SIZE)
1270 pgd = pgd_offset_k(pg);
1271 else
1272 pgd = pgd_offset_gate(mm, pg);
1273 BUG_ON(pgd_none(*pgd));
1274 pud = pud_offset(pgd, pg);
1275 BUG_ON(pud_none(*pud));
1276 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001277 if (pmd_none(*pmd))
1278 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001280 if (pte_none(*pte)) {
1281 pte_unmap(pte);
1282 return i ? : -EFAULT;
1283 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 if (pages) {
Nick Pigginfa2a4552005-11-29 18:43:17 +11001285 struct page *page = vm_normal_page(gate_vma, start, *pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001286 pages[i] = page;
1287 if (page)
1288 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 }
1290 pte_unmap(pte);
1291 if (vmas)
1292 vmas[i] = gate_vma;
1293 i++;
1294 start += PAGE_SIZE;
Peter Zijlstra9d737772009-06-25 11:58:55 +02001295 nr_pages--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 continue;
1297 }
1298
Nick Pigginb291f002008-10-18 20:26:44 -07001299 if (!vma ||
1300 (vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
Hugh Dickins1c3aff12009-09-21 17:03:24 -07001301 !(vm_flags & vma->vm_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 return i ? : -EFAULT;
1303
Hugh Dickins2a15efc2009-09-21 17:03:27 -07001304 if (is_vm_hugetlb_page(vma)) {
1305 i = follow_hugetlb_page(mm, vma, pages, vmas,
Hugh Dickins58fa8792009-09-21 17:03:31 -07001306 &start, &nr_pages, i, gup_flags);
Hugh Dickins2a15efc2009-09-21 17:03:27 -07001307 continue;
1308 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001309
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 do {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001311 struct page *page;
Hugh Dickins58fa8792009-09-21 17:03:31 -07001312 unsigned int foll_flags = gup_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313
Ethan Solomita462e00c2007-07-15 23:38:16 -07001314 /*
Ying Han47792802009-01-06 14:40:18 -08001315 * If we have a pending SIGKILL, don't keep faulting
Hugh Dickins1c3aff12009-09-21 17:03:24 -07001316 * pages and potentially allocating memory.
Ethan Solomita462e00c2007-07-15 23:38:16 -07001317 */
Hugh Dickins1c3aff12009-09-21 17:03:24 -07001318 if (unlikely(fatal_signal_pending(current)))
Ying Han47792802009-01-06 14:40:18 -08001319 return i ? i : -ERESTARTSYS;
Ethan Solomita462e00c2007-07-15 23:38:16 -07001320
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001321 cond_resched();
Linus Torvalds6aab3412005-11-28 14:34:23 -08001322 while (!(page = follow_page(vma, start, foll_flags))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001323 int ret;
Linus Torvaldsd06063c2009-04-10 09:01:23 -07001324
Hugh Dickinsd26ed652009-06-23 13:52:49 +01001325 ret = handle_mm_fault(mm, vma, start,
1326 (foll_flags & FOLL_WRITE) ?
1327 FAULT_FLAG_WRITE : 0);
1328
Nick Piggin83c54072007-07-19 01:47:05 -07001329 if (ret & VM_FAULT_ERROR) {
1330 if (ret & VM_FAULT_OOM)
1331 return i ? i : -ENOMEM;
Andi Kleend1737fd2009-09-16 11:50:06 +02001332 if (ret &
1333 (VM_FAULT_HWPOISON|VM_FAULT_SIGBUS))
Nick Piggin83c54072007-07-19 01:47:05 -07001334 return i ? i : -EFAULT;
1335 BUG();
1336 }
1337 if (ret & VM_FAULT_MAJOR)
1338 tsk->maj_flt++;
1339 else
1340 tsk->min_flt++;
1341
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001342 /*
Nick Piggin83c54072007-07-19 01:47:05 -07001343 * The VM_FAULT_WRITE bit tells us that
1344 * do_wp_page has broken COW when necessary,
1345 * even if maybe_mkwrite decided not to set
1346 * pte_write. We can thus safely do subsequent
Hugh Dickins878b63a2009-01-06 14:39:32 -08001347 * page lookups as if they were reads. But only
1348 * do so when looping for pte_write is futile:
1349 * in some cases userspace may also be wanting
1350 * to write to the gotten user page, which a
1351 * read fault here might prevent (a readonly
1352 * page might get reCOWed by userspace write).
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001353 */
Hugh Dickins878b63a2009-01-06 14:39:32 -08001354 if ((ret & VM_FAULT_WRITE) &&
1355 !(vma->vm_flags & VM_WRITE))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001356 foll_flags &= ~FOLL_WRITE;
Nick Piggin83c54072007-07-19 01:47:05 -07001357
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07001358 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 }
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001360 if (IS_ERR(page))
1361 return i ? i : PTR_ERR(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001363 pages[i] = page;
James Bottomley03beb072006-03-26 01:36:57 -08001364
Russell Kinga6f36be2006-12-30 22:24:19 +00001365 flush_anon_page(vma, page, start);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001366 flush_dcache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 }
1368 if (vmas)
1369 vmas[i] = vma;
1370 i++;
1371 start += PAGE_SIZE;
Peter Zijlstra9d737772009-06-25 11:58:55 +02001372 nr_pages--;
1373 } while (nr_pages && start < vma->vm_end);
1374 } while (nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 return i;
1376}
Nick Pigginb291f002008-10-18 20:26:44 -07001377
Nick Piggind2bf6be2009-06-16 15:31:39 -07001378/**
1379 * get_user_pages() - pin user pages in memory
1380 * @tsk: task_struct of target task
1381 * @mm: mm_struct of target mm
1382 * @start: starting user address
Peter Zijlstra9d737772009-06-25 11:58:55 +02001383 * @nr_pages: number of pages from start to pin
Nick Piggind2bf6be2009-06-16 15:31:39 -07001384 * @write: whether pages will be written to by the caller
1385 * @force: whether to force write access even if user mapping is
1386 * readonly. This will result in the page being COWed even
1387 * in MAP_SHARED mappings. You do not want this.
1388 * @pages: array that receives pointers to the pages pinned.
1389 * Should be at least nr_pages long. Or NULL, if caller
1390 * only intends to ensure the pages are faulted in.
1391 * @vmas: array of pointers to vmas corresponding to each page.
1392 * Or NULL if the caller does not require them.
1393 *
1394 * Returns number of pages pinned. This may be fewer than the number
Peter Zijlstra9d737772009-06-25 11:58:55 +02001395 * requested. If nr_pages is 0 or negative, returns 0. If no pages
Nick Piggind2bf6be2009-06-16 15:31:39 -07001396 * were pinned, returns -errno. Each page returned must be released
1397 * with a put_page() call when it is finished with. vmas will only
1398 * remain valid while mmap_sem is held.
1399 *
1400 * Must be called with mmap_sem held for read or write.
1401 *
1402 * get_user_pages walks a process's page tables and takes a reference to
1403 * each struct page that each user address corresponds to at a given
1404 * instant. That is, it takes the page that would be accessed if a user
1405 * thread accesses the given user virtual address at that instant.
1406 *
1407 * This does not guarantee that the page exists in the user mappings when
1408 * get_user_pages returns, and there may even be a completely different
1409 * page there in some cases (eg. if mmapped pagecache has been invalidated
1410 * and subsequently re faulted). However it does guarantee that the page
1411 * won't be freed completely. And mostly callers simply care that the page
1412 * contains data that was valid *at some point in time*. Typically, an IO
1413 * or similar operation cannot guarantee anything stronger anyway because
1414 * locks can't be held over the syscall boundary.
1415 *
1416 * If write=0, the page must not be written to. If the page is written to,
1417 * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called
1418 * after the page is finished with, and before put_page is called.
1419 *
1420 * get_user_pages is typically used for fewer-copy IO operations, to get a
1421 * handle on the memory by some means other than accesses via the user virtual
1422 * addresses. The pages may be submitted for DMA to devices or accessed via
1423 * their kernel linear mapping (via the kmap APIs). Care should be taken to
1424 * use the correct cache flushing APIs.
1425 *
1426 * See also get_user_pages_fast, for performance critical applications.
1427 */
Nick Pigginb291f002008-10-18 20:26:44 -07001428int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
Peter Zijlstra9d737772009-06-25 11:58:55 +02001429 unsigned long start, int nr_pages, int write, int force,
Nick Pigginb291f002008-10-18 20:26:44 -07001430 struct page **pages, struct vm_area_struct **vmas)
1431{
Hugh Dickins58fa8792009-09-21 17:03:31 -07001432 int flags = FOLL_TOUCH;
Nick Pigginb291f002008-10-18 20:26:44 -07001433
Hugh Dickins58fa8792009-09-21 17:03:31 -07001434 if (pages)
1435 flags |= FOLL_GET;
Nick Pigginb291f002008-10-18 20:26:44 -07001436 if (write)
Hugh Dickins58fa8792009-09-21 17:03:31 -07001437 flags |= FOLL_WRITE;
Nick Pigginb291f002008-10-18 20:26:44 -07001438 if (force)
Hugh Dickins58fa8792009-09-21 17:03:31 -07001439 flags |= FOLL_FORCE;
Nick Pigginb291f002008-10-18 20:26:44 -07001440
Peter Zijlstra9d737772009-06-25 11:58:55 +02001441 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas);
Nick Pigginb291f002008-10-18 20:26:44 -07001442}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443EXPORT_SYMBOL(get_user_pages);
1444
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07001445/**
1446 * get_dump_page() - pin user page in memory while writing it to core dump
1447 * @addr: user address
1448 *
1449 * Returns struct page pointer of user page pinned for dump,
1450 * to be freed afterwards by page_cache_release() or put_page().
1451 *
1452 * Returns NULL on any kind of failure - a hole must then be inserted into
1453 * the corefile, to preserve alignment with its headers; and also returns
1454 * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found -
1455 * allowing a hole to be left in the corefile to save diskspace.
1456 *
1457 * Called without mmap_sem, but after all other threads have been killed.
1458 */
1459#ifdef CONFIG_ELF_CORE
1460struct page *get_dump_page(unsigned long addr)
1461{
1462 struct vm_area_struct *vma;
1463 struct page *page;
1464
1465 if (__get_user_pages(current, current->mm, addr, 1,
Hugh Dickins58fa8792009-09-21 17:03:31 -07001466 FOLL_FORCE | FOLL_DUMP | FOLL_GET, &page, &vma) < 1)
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07001467 return NULL;
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07001468 flush_cache_page(vma, addr, page_to_pfn(page));
1469 return page;
1470}
1471#endif /* CONFIG_ELF_CORE */
1472
Harvey Harrison920c7a52008-02-04 22:29:26 -08001473pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr,
1474 spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001475{
1476 pgd_t * pgd = pgd_offset(mm, addr);
1477 pud_t * pud = pud_alloc(mm, pgd, addr);
1478 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001479 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001480 if (pmd)
1481 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1482 }
1483 return NULL;
1484}
1485
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001487 * This is the old fallback for page remapping.
1488 *
1489 * For historical reasons, it only allows reserved pages. Only
1490 * old drivers should use this, and they needed to mark their
1491 * pages reserved for the old functions anyway.
1492 */
Nick Piggin423bad62008-04-28 02:13:01 -07001493static int insert_page(struct vm_area_struct *vma, unsigned long addr,
1494 struct page *page, pgprot_t prot)
Linus Torvalds238f58d2005-11-29 13:01:56 -08001495{
Nick Piggin423bad62008-04-28 02:13:01 -07001496 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001497 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001498 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001499 spinlock_t *ptl;
1500
Linus Torvalds238f58d2005-11-29 13:01:56 -08001501 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001502 if (PageAnon(page))
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001503 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001504 retval = -ENOMEM;
1505 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001506 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001507 if (!pte)
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001508 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001509 retval = -EBUSY;
1510 if (!pte_none(*pte))
1511 goto out_unlock;
1512
1513 /* Ok, finally just insert the thing.. */
1514 get_page(page);
1515 inc_mm_counter(mm, file_rss);
1516 page_add_file_rmap(page);
1517 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1518
1519 retval = 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001520 pte_unmap_unlock(pte, ptl);
1521 return retval;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001522out_unlock:
1523 pte_unmap_unlock(pte, ptl);
1524out:
1525 return retval;
1526}
1527
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001528/**
1529 * vm_insert_page - insert single page into user vma
1530 * @vma: user vma to map to
1531 * @addr: target user address of this page
1532 * @page: source kernel page
1533 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001534 * This allows drivers to insert individual pages they've allocated
1535 * into a user vma.
1536 *
1537 * The page has to be a nice clean _individual_ kernel allocation.
1538 * If you allocate a compound page, you need to have marked it as
1539 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001540 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001541 *
1542 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1543 * took an arbitrary page protection parameter. This doesn't allow
1544 * that. Your vma protection will have to be set up correctly, which
1545 * means that if you want a shared writable mapping, you'd better
1546 * ask for a shared writable mapping!
1547 *
1548 * The page does not need to be reserved.
1549 */
Nick Piggin423bad62008-04-28 02:13:01 -07001550int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
1551 struct page *page)
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001552{
1553 if (addr < vma->vm_start || addr >= vma->vm_end)
1554 return -EFAULT;
1555 if (!page_count(page))
1556 return -EINVAL;
Linus Torvalds4d7672b2005-12-16 10:21:23 -08001557 vma->vm_flags |= VM_INSERTPAGE;
Nick Piggin423bad62008-04-28 02:13:01 -07001558 return insert_page(vma, addr, page, vma->vm_page_prot);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001559}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001560EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001561
Nick Piggin423bad62008-04-28 02:13:01 -07001562static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1563 unsigned long pfn, pgprot_t prot)
1564{
1565 struct mm_struct *mm = vma->vm_mm;
1566 int retval;
1567 pte_t *pte, entry;
1568 spinlock_t *ptl;
1569
1570 retval = -ENOMEM;
1571 pte = get_locked_pte(mm, addr, &ptl);
1572 if (!pte)
1573 goto out;
1574 retval = -EBUSY;
1575 if (!pte_none(*pte))
1576 goto out_unlock;
1577
1578 /* Ok, finally just insert the thing.. */
1579 entry = pte_mkspecial(pfn_pte(pfn, prot));
1580 set_pte_at(mm, addr, pte, entry);
1581 update_mmu_cache(vma, addr, entry); /* XXX: why not for insert_page? */
1582
1583 retval = 0;
1584out_unlock:
1585 pte_unmap_unlock(pte, ptl);
1586out:
1587 return retval;
1588}
1589
Nick Piggine0dc0d82007-02-12 00:51:36 -08001590/**
1591 * vm_insert_pfn - insert single pfn into user vma
1592 * @vma: user vma to map to
1593 * @addr: target user address of this page
1594 * @pfn: source kernel pfn
1595 *
1596 * Similar to vm_inert_page, this allows drivers to insert individual pages
1597 * they've allocated into a user vma. Same comments apply.
1598 *
1599 * This function should only be called from a vm_ops->fault handler, and
1600 * in that case the handler should return NULL.
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001601 *
1602 * vma cannot be a COW mapping.
1603 *
1604 * As this is called only for pages that do not currently exist, we
1605 * do not need to flush old virtual caches or the TLB.
Nick Piggine0dc0d82007-02-12 00:51:36 -08001606 */
1607int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
Nick Piggin423bad62008-04-28 02:13:01 -07001608 unsigned long pfn)
Nick Piggine0dc0d82007-02-12 00:51:36 -08001609{
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001610 int ret;
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001611 pgprot_t pgprot = vma->vm_page_prot;
Nick Piggin7e675132008-04-28 02:13:00 -07001612 /*
1613 * Technically, architectures with pte_special can avoid all these
1614 * restrictions (same for remap_pfn_range). However we would like
1615 * consistency in testing and feature parity among all, so we should
1616 * try to keep these invariants in place for everybody.
1617 */
Jared Hulbertb379d792008-04-28 02:12:58 -07001618 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
1619 BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
1620 (VM_PFNMAP|VM_MIXEDMAP));
1621 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1622 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
Nick Piggine0dc0d82007-02-12 00:51:36 -08001623
Nick Piggin423bad62008-04-28 02:13:01 -07001624 if (addr < vma->vm_start || addr >= vma->vm_end)
1625 return -EFAULT;
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001626 if (track_pfn_vma_new(vma, &pgprot, pfn, PAGE_SIZE))
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001627 return -EINVAL;
1628
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001629 ret = insert_pfn(vma, addr, pfn, pgprot);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001630
1631 if (ret)
1632 untrack_pfn_vma(vma, pfn, PAGE_SIZE);
1633
1634 return ret;
Nick Piggine0dc0d82007-02-12 00:51:36 -08001635}
1636EXPORT_SYMBOL(vm_insert_pfn);
1637
Nick Piggin423bad62008-04-28 02:13:01 -07001638int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
1639 unsigned long pfn)
1640{
1641 BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
1642
1643 if (addr < vma->vm_start || addr >= vma->vm_end)
1644 return -EFAULT;
1645
1646 /*
1647 * If we don't have pte special, then we have to use the pfn_valid()
1648 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1649 * refcount the page if pfn_valid is true (hence insert_page rather
Hugh Dickins62eede62009-09-21 17:03:34 -07001650 * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP
1651 * without pte special, it would there be refcounted as a normal page.
Nick Piggin423bad62008-04-28 02:13:01 -07001652 */
1653 if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) {
1654 struct page *page;
1655
1656 page = pfn_to_page(pfn);
1657 return insert_page(vma, addr, page, vma->vm_page_prot);
1658 }
1659 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
1660}
1661EXPORT_SYMBOL(vm_insert_mixed);
1662
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001663/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 * maps a range of physical memory into the requested pages. the old
1665 * mappings are removed. any references to nonexistent pages results
1666 * in null mappings (currently treated as "copy-on-access")
1667 */
1668static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1669 unsigned long addr, unsigned long end,
1670 unsigned long pfn, pgprot_t prot)
1671{
1672 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001673 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674
Hugh Dickinsc74df322005-10-29 18:16:23 -07001675 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 if (!pte)
1677 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001678 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 do {
1680 BUG_ON(!pte_none(*pte));
Nick Piggin7e675132008-04-28 02:13:00 -07001681 set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 pfn++;
1683 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001684 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001685 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 return 0;
1687}
1688
1689static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1690 unsigned long addr, unsigned long end,
1691 unsigned long pfn, pgprot_t prot)
1692{
1693 pmd_t *pmd;
1694 unsigned long next;
1695
1696 pfn -= addr >> PAGE_SHIFT;
1697 pmd = pmd_alloc(mm, pud, addr);
1698 if (!pmd)
1699 return -ENOMEM;
1700 do {
1701 next = pmd_addr_end(addr, end);
1702 if (remap_pte_range(mm, pmd, addr, next,
1703 pfn + (addr >> PAGE_SHIFT), prot))
1704 return -ENOMEM;
1705 } while (pmd++, addr = next, addr != end);
1706 return 0;
1707}
1708
1709static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1710 unsigned long addr, unsigned long end,
1711 unsigned long pfn, pgprot_t prot)
1712{
1713 pud_t *pud;
1714 unsigned long next;
1715
1716 pfn -= addr >> PAGE_SHIFT;
1717 pud = pud_alloc(mm, pgd, addr);
1718 if (!pud)
1719 return -ENOMEM;
1720 do {
1721 next = pud_addr_end(addr, end);
1722 if (remap_pmd_range(mm, pud, addr, next,
1723 pfn + (addr >> PAGE_SHIFT), prot))
1724 return -ENOMEM;
1725 } while (pud++, addr = next, addr != end);
1726 return 0;
1727}
1728
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001729/**
1730 * remap_pfn_range - remap kernel memory to userspace
1731 * @vma: user vma to map to
1732 * @addr: target user address to start at
1733 * @pfn: physical address of kernel memory
1734 * @size: size of map area
1735 * @prot: page protection flags for this mapping
1736 *
1737 * Note: this is only safe if the mm semaphore is held when called.
1738 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1740 unsigned long pfn, unsigned long size, pgprot_t prot)
1741{
1742 pgd_t *pgd;
1743 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001744 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 struct mm_struct *mm = vma->vm_mm;
1746 int err;
1747
1748 /*
1749 * Physically remapped pages are special. Tell the
1750 * rest of the world about it:
1751 * VM_IO tells people not to look at these pages
1752 * (accesses can have side effects).
Hugh Dickins0b14c172005-11-21 21:32:15 -08001753 * VM_RESERVED is specified all over the place, because
1754 * in 2.4 it kept swapout's vma scan off this vma; but
1755 * in 2.6 the LRU scan won't even find its pages, so this
1756 * flag means no more than count its pages in reserved_vm,
1757 * and omit it from core dump, even when VM_IO turned off.
Linus Torvalds6aab3412005-11-28 14:34:23 -08001758 * VM_PFNMAP tells the core MM that the base pages are just
1759 * raw PFN mappings, and do not have a "struct page" associated
1760 * with them.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001761 *
1762 * There's a horrible special case to handle copy-on-write
1763 * behaviour that some programs depend on. We mark the "original"
1764 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 */
Pallipadi, Venkatesh4bb9c5c2009-03-12 17:45:27 -07001766 if (addr == vma->vm_start && end == vma->vm_end) {
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001767 vma->vm_pgoff = pfn;
Pallipadi, Venkatesh895791d2009-03-13 16:35:44 -07001768 vma->vm_flags |= VM_PFN_AT_MMAP;
Pallipadi, Venkatesh4bb9c5c2009-03-12 17:45:27 -07001769 } else if (is_cow_mapping(vma->vm_flags))
venkatesh.pallipadi@intel.com3c8bb732008-12-18 11:41:27 -08001770 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001771
Linus Torvalds6aab3412005-11-28 14:34:23 -08001772 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001774 err = track_pfn_vma_new(vma, &prot, pfn, PAGE_ALIGN(size));
venkatesh.pallipadi@intel.coma3670612009-01-09 16:13:09 -08001775 if (err) {
1776 /*
1777 * To indicate that track_pfn related cleanup is not
1778 * needed from higher level routine calling unmap_vmas
1779 */
1780 vma->vm_flags &= ~(VM_IO | VM_RESERVED | VM_PFNMAP);
Pallipadi, Venkatesh895791d2009-03-13 16:35:44 -07001781 vma->vm_flags &= ~VM_PFN_AT_MMAP;
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001782 return -EINVAL;
venkatesh.pallipadi@intel.coma3670612009-01-09 16:13:09 -08001783 }
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001784
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 BUG_ON(addr >= end);
1786 pfn -= addr >> PAGE_SHIFT;
1787 pgd = pgd_offset(mm, addr);
1788 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 do {
1790 next = pgd_addr_end(addr, end);
1791 err = remap_pud_range(mm, pgd, addr, next,
1792 pfn + (addr >> PAGE_SHIFT), prot);
1793 if (err)
1794 break;
1795 } while (pgd++, addr = next, addr != end);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001796
1797 if (err)
1798 untrack_pfn_vma(vma, pfn, PAGE_ALIGN(size));
1799
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 return err;
1801}
1802EXPORT_SYMBOL(remap_pfn_range);
1803
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001804static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1805 unsigned long addr, unsigned long end,
1806 pte_fn_t fn, void *data)
1807{
1808 pte_t *pte;
1809 int err;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001810 pgtable_t token;
Borislav Petkov94909912007-05-06 14:49:17 -07001811 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001812
1813 pte = (mm == &init_mm) ?
1814 pte_alloc_kernel(pmd, addr) :
1815 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1816 if (!pte)
1817 return -ENOMEM;
1818
1819 BUG_ON(pmd_huge(*pmd));
1820
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08001821 arch_enter_lazy_mmu_mode();
1822
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001823 token = pmd_pgtable(*pmd);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001824
1825 do {
Daisuke Nishimurac36987e2009-10-26 16:50:23 -07001826 err = fn(pte++, token, addr, data);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001827 if (err)
1828 break;
Daisuke Nishimurac36987e2009-10-26 16:50:23 -07001829 } while (addr += PAGE_SIZE, addr != end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001830
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08001831 arch_leave_lazy_mmu_mode();
1832
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001833 if (mm != &init_mm)
1834 pte_unmap_unlock(pte-1, ptl);
1835 return err;
1836}
1837
1838static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1839 unsigned long addr, unsigned long end,
1840 pte_fn_t fn, void *data)
1841{
1842 pmd_t *pmd;
1843 unsigned long next;
1844 int err;
1845
Andi Kleenceb86872008-07-23 21:27:50 -07001846 BUG_ON(pud_huge(*pud));
1847
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001848 pmd = pmd_alloc(mm, pud, addr);
1849 if (!pmd)
1850 return -ENOMEM;
1851 do {
1852 next = pmd_addr_end(addr, end);
1853 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1854 if (err)
1855 break;
1856 } while (pmd++, addr = next, addr != end);
1857 return err;
1858}
1859
1860static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1861 unsigned long addr, unsigned long end,
1862 pte_fn_t fn, void *data)
1863{
1864 pud_t *pud;
1865 unsigned long next;
1866 int err;
1867
1868 pud = pud_alloc(mm, pgd, addr);
1869 if (!pud)
1870 return -ENOMEM;
1871 do {
1872 next = pud_addr_end(addr, end);
1873 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1874 if (err)
1875 break;
1876 } while (pud++, addr = next, addr != end);
1877 return err;
1878}
1879
1880/*
1881 * Scan a region of virtual memory, filling in page tables as necessary
1882 * and calling a provided function on each leaf page table.
1883 */
1884int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1885 unsigned long size, pte_fn_t fn, void *data)
1886{
1887 pgd_t *pgd;
1888 unsigned long next;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001889 unsigned long start = addr, end = addr + size;
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001890 int err;
1891
1892 BUG_ON(addr >= end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001893 mmu_notifier_invalidate_range_start(mm, start, end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001894 pgd = pgd_offset(mm, addr);
1895 do {
1896 next = pgd_addr_end(addr, end);
1897 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1898 if (err)
1899 break;
1900 } while (pgd++, addr = next, addr != end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001901 mmu_notifier_invalidate_range_end(mm, start, end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001902 return err;
1903}
1904EXPORT_SYMBOL_GPL(apply_to_page_range);
1905
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001907 * handle_pte_fault chooses page fault handler according to an entry
1908 * which was read non-atomically. Before making any commitment, on
1909 * those architectures or configurations (e.g. i386 with PAE) which
1910 * might give a mix of unmatched parts, do_swap_page and do_file_page
1911 * must check under lock before unmapping the pte and proceeding
1912 * (but do_wp_page is only called after already making such a check;
1913 * and do_anonymous_page and do_no_page can safely check later on).
1914 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001915static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001916 pte_t *page_table, pte_t orig_pte)
1917{
1918 int same = 1;
1919#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1920 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001921 spinlock_t *ptl = pte_lockptr(mm, pmd);
1922 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001923 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001924 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001925 }
1926#endif
1927 pte_unmap(page_table);
1928 return same;
1929}
1930
1931/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1933 * servicing faults for write access. In the normal case, do always want
1934 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1935 * that do not have writing enabled, when used by access_process_vm.
1936 */
1937static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1938{
1939 if (likely(vma->vm_flags & VM_WRITE))
1940 pte = pte_mkwrite(pte);
1941 return pte;
1942}
1943
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001944static 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 -08001945{
1946 /*
1947 * If the source page was a PFN mapping, we don't have
1948 * a "struct page" for it. We do a best-effort copy by
1949 * just copying from the original user address. If that
1950 * fails, we just zero-fill it. Live with it.
1951 */
1952 if (unlikely(!src)) {
1953 void *kaddr = kmap_atomic(dst, KM_USER0);
Linus Torvalds5d2a2dbbc2005-11-29 14:07:55 -08001954 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1955
1956 /*
1957 * This really shouldn't fail, because the page is there
1958 * in the page tables. But it might just be unreadable,
1959 * in which case we just give up and fill the result with
1960 * zeroes.
1961 */
1962 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Linus Torvalds6aab3412005-11-28 14:34:23 -08001963 memset(kaddr, 0, PAGE_SIZE);
1964 kunmap_atomic(kaddr, KM_USER0);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07001965 flush_dcache_page(dst);
Nick Piggin0ed361d2008-02-04 22:29:34 -08001966 } else
1967 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001968}
1969
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 * This routine handles present pages, when users try to write
1972 * to a shared page. It is done by copying the page to a new address
1973 * and decrementing the shared-page counter for the old page.
1974 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 * Note that this routine assumes that the protection checks have been
1976 * done by the caller (the low-level page fault routine in most cases).
1977 * Thus we can safely just mark it writable once we've done any necessary
1978 * COW.
1979 *
1980 * We also mark the page dirty at this point even though the page will
1981 * change only once the write actually happens. This avoids a few races,
1982 * and potentially makes it more efficient.
1983 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001984 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1985 * but allow concurrent faults), with pte both mapped and locked.
1986 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001988static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1989 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001990 spinlock_t *ptl, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991{
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001992 struct page *old_page, *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 pte_t entry;
Nick Piggin83c54072007-07-19 01:47:05 -07001994 int reuse = 0, ret = 0;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001995 int page_mkwrite = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001996 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997
Linus Torvalds6aab3412005-11-28 14:34:23 -08001998 old_page = vm_normal_page(vma, address, orig_pte);
Peter Zijlstra251b97f2008-07-04 09:59:24 -07001999 if (!old_page) {
2000 /*
2001 * VM_MIXEDMAP !pfn_valid() case
2002 *
2003 * We should not cow pages in a shared writeable mapping.
2004 * Just mark the pages writable as we can't do any dirty
2005 * accounting on raw pfn maps.
2006 */
2007 if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
2008 (VM_WRITE|VM_SHARED))
2009 goto reuse;
Linus Torvalds6aab3412005-11-28 14:34:23 -08002010 goto gotten;
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002011 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002013 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002014 * Take out anonymous pages first, anonymous shared vmas are
2015 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002016 */
Hugh Dickins9a840892009-09-21 17:02:01 -07002017 if (PageAnon(old_page) && !PageKsm(old_page)) {
Hugh Dickinsab967d82009-01-06 14:39:33 -08002018 if (!trylock_page(old_page)) {
2019 page_cache_get(old_page);
2020 pte_unmap_unlock(page_table, ptl);
2021 lock_page(old_page);
2022 page_table = pte_offset_map_lock(mm, pmd, address,
2023 &ptl);
2024 if (!pte_same(*page_table, orig_pte)) {
2025 unlock_page(old_page);
2026 page_cache_release(old_page);
2027 goto unlock;
2028 }
2029 page_cache_release(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002030 }
Hugh Dickins7b1fe592009-01-06 14:39:34 -08002031 reuse = reuse_swap_page(old_page);
Hugh Dickinsab967d82009-01-06 14:39:33 -08002032 unlock_page(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002033 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002034 (VM_WRITE|VM_SHARED))) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002035 /*
2036 * Only catch write-faults on shared writable pages,
2037 * read-only shared pages can get COWed by
2038 * get_user_pages(.write=1, .force=1).
2039 */
David Howells9637a5e2006-06-23 02:03:43 -07002040 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
Nick Pigginc2ec1752009-03-31 15:23:21 -07002041 struct vm_fault vmf;
2042 int tmp;
2043
2044 vmf.virtual_address = (void __user *)(address &
2045 PAGE_MASK);
2046 vmf.pgoff = old_page->index;
2047 vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
2048 vmf.page = old_page;
2049
David Howells9637a5e2006-06-23 02:03:43 -07002050 /*
2051 * Notify the address space that the page is about to
2052 * become writable so that it can prohibit this or wait
2053 * for the page to get into an appropriate state.
2054 *
2055 * We do this without the lock held, so that it can
2056 * sleep if it needs to.
2057 */
2058 page_cache_get(old_page);
2059 pte_unmap_unlock(page_table, ptl);
2060
Nick Pigginc2ec1752009-03-31 15:23:21 -07002061 tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
2062 if (unlikely(tmp &
2063 (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
2064 ret = tmp;
David Howells9637a5e2006-06-23 02:03:43 -07002065 goto unwritable_page;
Nick Pigginc2ec1752009-03-31 15:23:21 -07002066 }
Nick Pigginb827e492009-04-30 15:08:16 -07002067 if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
2068 lock_page(old_page);
2069 if (!old_page->mapping) {
2070 ret = 0; /* retry the fault */
2071 unlock_page(old_page);
2072 goto unwritable_page;
2073 }
2074 } else
2075 VM_BUG_ON(!PageLocked(old_page));
David Howells9637a5e2006-06-23 02:03:43 -07002076
David Howells9637a5e2006-06-23 02:03:43 -07002077 /*
2078 * Since we dropped the lock we need to revalidate
2079 * the PTE as someone else may have changed it. If
2080 * they did, we just return, as we can count on the
2081 * MMU to tell us if they didn't also make it writable.
2082 */
2083 page_table = pte_offset_map_lock(mm, pmd, address,
2084 &ptl);
Nick Pigginb827e492009-04-30 15:08:16 -07002085 if (!pte_same(*page_table, orig_pte)) {
2086 unlock_page(old_page);
2087 page_cache_release(old_page);
David Howells9637a5e2006-06-23 02:03:43 -07002088 goto unlock;
Nick Pigginb827e492009-04-30 15:08:16 -07002089 }
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002090
2091 page_mkwrite = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002093 dirty_page = old_page;
2094 get_page(dirty_page);
David Howells9637a5e2006-06-23 02:03:43 -07002095 reuse = 1;
David Howells9637a5e2006-06-23 02:03:43 -07002096 }
2097
2098 if (reuse) {
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002099reuse:
David Howells9637a5e2006-06-23 02:03:43 -07002100 flush_cache_page(vma, address, pte_pfn(orig_pte));
2101 entry = pte_mkyoung(orig_pte);
2102 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
KAMEZAWA Hiroyuki954ffcb2007-10-16 01:25:44 -07002103 if (ptep_set_access_flags(vma, address, page_table, entry,1))
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002104 update_mmu_cache(vma, address, entry);
David Howells9637a5e2006-06-23 02:03:43 -07002105 ret |= VM_FAULT_WRITE;
2106 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108
2109 /*
2110 * Ok, we need to copy. Oh, well..
2111 */
Nick Pigginb5810032005-10-29 18:16:12 -07002112 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08002113gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002114 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115
2116 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07002117 goto oom;
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002118
Hugh Dickins62eede62009-09-21 17:03:34 -07002119 if (is_zero_pfn(pte_pfn(orig_pte))) {
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002120 new_page = alloc_zeroed_user_highpage_movable(vma, address);
2121 if (!new_page)
2122 goto oom;
2123 } else {
2124 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
2125 if (!new_page)
2126 goto oom;
2127 cow_user_page(new_page, old_page, address, vma);
2128 }
2129 __SetPageUptodate(new_page);
2130
Nick Pigginb291f002008-10-18 20:26:44 -07002131 /*
2132 * Don't let another task, with possibly unlocked vma,
2133 * keep the mlocked page.
2134 */
Carsten Otteab926612009-02-04 15:12:16 -08002135 if ((vma->vm_flags & VM_LOCKED) && old_page) {
Nick Pigginb291f002008-10-18 20:26:44 -07002136 lock_page(old_page); /* for LRU manipulation */
2137 clear_page_mlock(old_page);
2138 unlock_page(old_page);
2139 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002140
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002141 if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002142 goto oom_free_new;
2143
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 /*
2145 * Re-check the pte - we dropped the lock
2146 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002147 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07002148 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08002149 if (old_page) {
Hugh Dickins920fc352005-11-21 21:32:17 -08002150 if (!PageAnon(old_page)) {
2151 dec_mm_counter(mm, file_rss);
2152 inc_mm_counter(mm, anon_rss);
2153 }
2154 } else
Hugh Dickins42946212005-10-29 18:16:05 -07002155 inc_mm_counter(mm, anon_rss);
Ben Collinseca35132005-11-29 11:45:26 -08002156 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07002157 entry = mk_pte(new_page, vma->vm_page_prot);
2158 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Siddha, Suresh B4ce072f2006-09-29 01:58:42 -07002159 /*
2160 * Clear the pte entry and flush it first, before updating the
2161 * pte with the new entry. This will avoid a race condition
2162 * seen in the presence of one thread doing SMC and another
2163 * thread doing COW.
2164 */
Izik Eidus828502d2009-09-21 17:01:51 -07002165 ptep_clear_flush(vma, address, page_table);
Nick Piggin9617d952006-01-06 00:11:12 -08002166 page_add_new_anon_rmap(new_page, vma, address);
Izik Eidus828502d2009-09-21 17:01:51 -07002167 /*
2168 * We call the notify macro here because, when using secondary
2169 * mmu page tables (such as kvm shadow page tables), we want the
2170 * new page to be mapped directly into the secondary page table.
2171 */
2172 set_pte_at_notify(mm, address, page_table, entry);
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002173 update_mmu_cache(vma, address, entry);
Nick Piggin945754a2008-06-23 14:30:30 +02002174 if (old_page) {
2175 /*
2176 * Only after switching the pte to the new page may
2177 * we remove the mapcount here. Otherwise another
2178 * process may come and find the rmap count decremented
2179 * before the pte is switched to the new page, and
2180 * "reuse" the old page writing into it while our pte
2181 * here still points into it and can be read by other
2182 * threads.
2183 *
2184 * The critical issue is to order this
2185 * page_remove_rmap with the ptp_clear_flush above.
2186 * Those stores are ordered by (if nothing else,)
2187 * the barrier present in the atomic_add_negative
2188 * in page_remove_rmap.
2189 *
2190 * Then the TLB flush in ptep_clear_flush ensures that
2191 * no process can access the old page before the
2192 * decremented mapcount is visible. And the old page
2193 * cannot be reused until after the decremented
2194 * mapcount is visible. So transitively, TLBs to
2195 * old page will be flushed before it can be reused.
2196 */
Hugh Dickinsedc315f2009-01-06 14:40:11 -08002197 page_remove_rmap(old_page);
Nick Piggin945754a2008-06-23 14:30:30 +02002198 }
2199
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 /* Free the old page.. */
2201 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10002202 ret |= VM_FAULT_WRITE;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002203 } else
2204 mem_cgroup_uncharge_page(new_page);
2205
Hugh Dickins920fc352005-11-21 21:32:17 -08002206 if (new_page)
2207 page_cache_release(new_page);
2208 if (old_page)
2209 page_cache_release(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002210unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002211 pte_unmap_unlock(page_table, ptl);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002212 if (dirty_page) {
Nick Piggin79352892007-07-19 01:47:22 -07002213 /*
2214 * Yes, Virginia, this is actually required to prevent a race
2215 * with clear_page_dirty_for_io() from clearing the page dirty
2216 * bit after it clear all dirty ptes, but before a racing
2217 * do_wp_page installs a dirty pte.
2218 *
2219 * do_no_page is protected similarly.
2220 */
Nick Pigginb827e492009-04-30 15:08:16 -07002221 if (!page_mkwrite) {
2222 wait_on_page_locked(dirty_page);
2223 set_page_dirty_balance(dirty_page, page_mkwrite);
2224 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002225 put_page(dirty_page);
Nick Pigginb827e492009-04-30 15:08:16 -07002226 if (page_mkwrite) {
2227 struct address_space *mapping = dirty_page->mapping;
2228
2229 set_page_dirty(dirty_page);
2230 unlock_page(dirty_page);
2231 page_cache_release(dirty_page);
2232 if (mapping) {
2233 /*
2234 * Some device drivers do not set page.mapping
2235 * but still dirty their pages
2236 */
2237 balance_dirty_pages_ratelimited(mapping);
2238 }
2239 }
2240
2241 /* file_update_time outside page_lock */
2242 if (vma->vm_file)
2243 file_update_time(vma->vm_file);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002244 }
Nick Pigginf33ea7f2005-08-03 20:24:01 +10002245 return ret;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002246oom_free_new:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002247 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002248oom:
Nick Pigginb827e492009-04-30 15:08:16 -07002249 if (old_page) {
2250 if (page_mkwrite) {
2251 unlock_page(old_page);
2252 page_cache_release(old_page);
2253 }
Hugh Dickins920fc352005-11-21 21:32:17 -08002254 page_cache_release(old_page);
Nick Pigginb827e492009-04-30 15:08:16 -07002255 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 return VM_FAULT_OOM;
David Howells9637a5e2006-06-23 02:03:43 -07002257
2258unwritable_page:
2259 page_cache_release(old_page);
Nick Pigginc2ec1752009-03-31 15:23:21 -07002260 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261}
2262
2263/*
2264 * Helper functions for unmap_mapping_range().
2265 *
2266 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
2267 *
2268 * We have to restart searching the prio_tree whenever we drop the lock,
2269 * since the iterator is only valid while the lock is held, and anyway
2270 * a later vma might be split and reinserted earlier while lock dropped.
2271 *
2272 * The list of nonlinear vmas could be handled more efficiently, using
2273 * a placeholder, but handle it in the same way until a need is shown.
2274 * It is important to search the prio_tree before nonlinear list: a vma
2275 * may become nonlinear and be shifted from prio_tree to nonlinear list
2276 * while the lock is dropped; but never shifted from list to prio_tree.
2277 *
2278 * In order to make forward progress despite restarting the search,
2279 * vm_truncate_count is used to mark a vma as now dealt with, so we can
2280 * quickly skip it next time around. Since the prio_tree search only
2281 * shows us those vmas affected by unmapping the range in question, we
2282 * can't efficiently keep all vmas in step with mapping->truncate_count:
2283 * so instead reset them all whenever it wraps back to 0 (then go to 1).
2284 * mapping->truncate_count and vma->vm_truncate_count are protected by
2285 * i_mmap_lock.
2286 *
2287 * In order to make forward progress despite repeatedly restarting some
Hugh Dickinsee39b372005-04-19 13:29:15 -07002288 * large vma, note the restart_addr from unmap_vmas when it breaks out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 * and restart from that address when we reach that vma again. It might
2290 * have been split or merged, shrunk or extended, but never shifted: so
2291 * restart_addr remains valid so long as it remains in the vma's range.
2292 * unmap_mapping_range forces truncate_count to leap over page-aligned
2293 * values so we can save vma's restart_addr in its truncate_count field.
2294 */
2295#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
2296
2297static void reset_vma_truncate_counts(struct address_space *mapping)
2298{
2299 struct vm_area_struct *vma;
2300 struct prio_tree_iter iter;
2301
2302 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
2303 vma->vm_truncate_count = 0;
2304 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
2305 vma->vm_truncate_count = 0;
2306}
2307
2308static int unmap_mapping_range_vma(struct vm_area_struct *vma,
2309 unsigned long start_addr, unsigned long end_addr,
2310 struct zap_details *details)
2311{
2312 unsigned long restart_addr;
2313 int need_break;
2314
Nick Piggind00806b2007-07-19 01:46:57 -07002315 /*
2316 * files that support invalidating or truncating portions of the
Nick Piggind0217ac2007-07-19 01:47:03 -07002317 * file from under mmaped areas must have their ->fault function
Nick Piggin83c54072007-07-19 01:47:05 -07002318 * return a locked page (and set VM_FAULT_LOCKED in the return).
2319 * This provides synchronisation against concurrent unmapping here.
Nick Piggind00806b2007-07-19 01:46:57 -07002320 */
Nick Piggind00806b2007-07-19 01:46:57 -07002321
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322again:
2323 restart_addr = vma->vm_truncate_count;
2324 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
2325 start_addr = restart_addr;
2326 if (start_addr >= end_addr) {
2327 /* Top of vma has been split off since last time */
2328 vma->vm_truncate_count = details->truncate_count;
2329 return 0;
2330 }
2331 }
2332
Hugh Dickinsee39b372005-04-19 13:29:15 -07002333 restart_addr = zap_page_range(vma, start_addr,
2334 end_addr - start_addr, details);
Nick Piggin95c354f2008-01-30 13:31:20 +01002335 need_break = need_resched() || spin_needbreak(details->i_mmap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336
Hugh Dickinsee39b372005-04-19 13:29:15 -07002337 if (restart_addr >= end_addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 /* We have now completed this vma: mark it so */
2339 vma->vm_truncate_count = details->truncate_count;
2340 if (!need_break)
2341 return 0;
2342 } else {
2343 /* Note restart_addr in vma's truncate_count field */
Hugh Dickinsee39b372005-04-19 13:29:15 -07002344 vma->vm_truncate_count = restart_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 if (!need_break)
2346 goto again;
2347 }
2348
2349 spin_unlock(details->i_mmap_lock);
2350 cond_resched();
2351 spin_lock(details->i_mmap_lock);
2352 return -EINTR;
2353}
2354
2355static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
2356 struct zap_details *details)
2357{
2358 struct vm_area_struct *vma;
2359 struct prio_tree_iter iter;
2360 pgoff_t vba, vea, zba, zea;
2361
2362restart:
2363 vma_prio_tree_foreach(vma, &iter, root,
2364 details->first_index, details->last_index) {
2365 /* Skip quickly over those we have already dealt with */
2366 if (vma->vm_truncate_count == details->truncate_count)
2367 continue;
2368
2369 vba = vma->vm_pgoff;
2370 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
2371 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
2372 zba = details->first_index;
2373 if (zba < vba)
2374 zba = vba;
2375 zea = details->last_index;
2376 if (zea > vea)
2377 zea = vea;
2378
2379 if (unmap_mapping_range_vma(vma,
2380 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
2381 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
2382 details) < 0)
2383 goto restart;
2384 }
2385}
2386
2387static inline void unmap_mapping_range_list(struct list_head *head,
2388 struct zap_details *details)
2389{
2390 struct vm_area_struct *vma;
2391
2392 /*
2393 * In nonlinear VMAs there is no correspondence between virtual address
2394 * offset and file offset. So we must perform an exhaustive search
2395 * across *all* the pages in each nonlinear VMA, not just the pages
2396 * whose virtual address lies outside the file truncation point.
2397 */
2398restart:
2399 list_for_each_entry(vma, head, shared.vm_set.list) {
2400 /* Skip quickly over those we have already dealt with */
2401 if (vma->vm_truncate_count == details->truncate_count)
2402 continue;
2403 details->nonlinear_vma = vma;
2404 if (unmap_mapping_range_vma(vma, vma->vm_start,
2405 vma->vm_end, details) < 0)
2406 goto restart;
2407 }
2408}
2409
2410/**
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002411 * 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 -07002412 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 * @holebegin: byte in first page to unmap, relative to the start of
2414 * the underlying file. This will be rounded down to a PAGE_SIZE
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +10002415 * boundary. Note that this is different from truncate_pagecache(), which
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 * must keep the partial page. In contrast, we must get rid of
2417 * partial pages.
2418 * @holelen: size of prospective hole in bytes. This will be rounded
2419 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2420 * end of the file.
2421 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2422 * but 0 when invalidating pagecache, don't throw away private data.
2423 */
2424void unmap_mapping_range(struct address_space *mapping,
2425 loff_t const holebegin, loff_t const holelen, int even_cows)
2426{
2427 struct zap_details details;
2428 pgoff_t hba = holebegin >> PAGE_SHIFT;
2429 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2430
2431 /* Check for overflow. */
2432 if (sizeof(holelen) > sizeof(hlen)) {
2433 long long holeend =
2434 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2435 if (holeend & ~(long long)ULONG_MAX)
2436 hlen = ULONG_MAX - hba + 1;
2437 }
2438
2439 details.check_mapping = even_cows? NULL: mapping;
2440 details.nonlinear_vma = NULL;
2441 details.first_index = hba;
2442 details.last_index = hba + hlen - 1;
2443 if (details.last_index < details.first_index)
2444 details.last_index = ULONG_MAX;
2445 details.i_mmap_lock = &mapping->i_mmap_lock;
2446
2447 spin_lock(&mapping->i_mmap_lock);
2448
Nick Piggind00806b2007-07-19 01:46:57 -07002449 /* Protect against endless unmapping loops */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450 mapping->truncate_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451 if (unlikely(is_restart_addr(mapping->truncate_count))) {
2452 if (mapping->truncate_count == 0)
2453 reset_vma_truncate_counts(mapping);
2454 mapping->truncate_count++;
2455 }
2456 details.truncate_count = mapping->truncate_count;
2457
2458 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
2459 unmap_mapping_range_tree(&mapping->i_mmap, &details);
2460 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
2461 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
2462 spin_unlock(&mapping->i_mmap_lock);
2463}
2464EXPORT_SYMBOL(unmap_mapping_range);
2465
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002466int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
2467{
2468 struct address_space *mapping = inode->i_mapping;
2469
2470 /*
2471 * If the underlying filesystem is not going to provide
2472 * a way to truncate a range of blocks (punch a hole) -
2473 * we should return failure right now.
2474 */
Al Viroacfa4382008-12-04 10:06:33 -05002475 if (!inode->i_op->truncate_range)
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002476 return -ENOSYS;
2477
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002478 mutex_lock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002479 down_write(&inode->i_alloc_sem);
2480 unmap_mapping_range(mapping, offset, (end - offset), 1);
2481 truncate_inode_pages_range(mapping, offset, end);
Nick Piggind00806b2007-07-19 01:46:57 -07002482 unmap_mapping_range(mapping, offset, (end - offset), 1);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002483 inode->i_op->truncate_range(inode, offset, end);
2484 up_write(&inode->i_alloc_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002485 mutex_unlock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002486
2487 return 0;
2488}
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002489
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002491 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2492 * but allow concurrent faults), and pte mapped but not yet locked.
2493 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002495static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2496 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002497 unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002499 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500 struct page *page;
Hugh Dickins65500d22005-10-29 18:15:59 -07002501 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 pte_t pte;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002503 struct mem_cgroup *ptr = NULL;
Nick Piggin83c54072007-07-19 01:47:05 -07002504 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002506 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002507 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002508
2509 entry = pte_to_swp_entry(orig_pte);
Andi Kleend1737fd2009-09-16 11:50:06 +02002510 if (unlikely(non_swap_entry(entry))) {
2511 if (is_migration_entry(entry)) {
2512 migration_entry_wait(mm, pmd, address);
2513 } else if (is_hwpoison_entry(entry)) {
2514 ret = VM_FAULT_HWPOISON;
2515 } else {
2516 print_bad_pte(vma, address, orig_pte, NULL);
2517 ret = VM_FAULT_OOM;
2518 }
Christoph Lameter06972122006-06-23 02:03:35 -07002519 goto out;
2520 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002521 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 page = lookup_swap_cache(entry);
2523 if (!page) {
Hugh Dickinsa5c9b692009-06-23 12:36:58 -07002524 grab_swap_token(mm); /* Contend for token _before_ read-in */
Hugh Dickins02098fe2008-02-04 22:28:42 -08002525 page = swapin_readahead(entry,
2526 GFP_HIGHUSER_MOVABLE, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 if (!page) {
2528 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002529 * Back out if somebody else faulted in this pte
2530 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002532 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 if (likely(pte_same(*page_table, orig_pte)))
2534 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002535 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002536 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 }
2538
2539 /* Had to read the page from swap area: Major fault */
2540 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002541 count_vm_event(PGMAJFAULT);
Andi Kleend1737fd2009-09-16 11:50:06 +02002542 } else if (PageHWPoison(page)) {
2543 ret = VM_FAULT_HWPOISON;
2544 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Andi Kleen4779cb32009-10-14 01:51:41 +02002545 goto out_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 }
2547
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 lock_page(page);
Balbir Singh20a10222007-11-14 17:00:33 -08002549 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002551 if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) {
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002552 ret = VM_FAULT_OOM;
Johannes Weinerbc43f752009-04-30 15:08:08 -07002553 goto out_page;
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002554 }
2555
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002557 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002559 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002560 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002561 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002562
2563 if (unlikely(!PageUptodate(page))) {
2564 ret = VM_FAULT_SIGBUS;
2565 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566 }
2567
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002568 /*
2569 * The page isn't present yet, go ahead with the fault.
2570 *
2571 * Be careful about the sequence of operations here.
2572 * To get its accounting right, reuse_swap_page() must be called
2573 * while the page is counted on swap but not yet in mapcount i.e.
2574 * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
2575 * must be called after the swap_free(), or it will never succeed.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002576 * Because delete_from_swap_page() may be called by reuse_swap_page(),
2577 * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry
2578 * in page->private. In this case, a record in swap_cgroup is silently
2579 * discarded at swap_free().
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002580 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581
Hugh Dickins42946212005-10-29 18:16:05 -07002582 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 pte = mk_pte(page, vma->vm_page_prot);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002584 if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002586 flags &= ~FAULT_FLAG_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588 flush_icache_page(vma, page);
2589 set_pte_at(mm, address, page_table, pte);
2590 page_add_anon_rmap(page, vma, address);
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002591 /* It's better to call commit-charge after rmap is established */
2592 mem_cgroup_commit_charge_swapin(page, ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002594 swap_free(entry);
Nick Pigginb291f002008-10-18 20:26:44 -07002595 if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08002596 try_to_free_swap(page);
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002597 unlock_page(page);
2598
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002599 if (flags & FAULT_FLAG_WRITE) {
Hugh Dickins61469f12008-03-04 14:29:04 -08002600 ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
2601 if (ret & VM_FAULT_ERROR)
2602 ret &= VM_FAULT_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 goto out;
2604 }
2605
2606 /* No need to invalidate - it was non-present before */
2607 update_mmu_cache(vma, address, pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07002608unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002609 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610out:
2611 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002612out_nomap:
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002613 mem_cgroup_cancel_charge_swapin(ptr);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002614 pte_unmap_unlock(page_table, ptl);
Johannes Weinerbc43f752009-04-30 15:08:08 -07002615out_page:
Kirill Korotaevb8107482005-05-16 21:53:50 -07002616 unlock_page(page);
Andi Kleen4779cb32009-10-14 01:51:41 +02002617out_release:
Kirill Korotaevb8107482005-05-16 21:53:50 -07002618 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002619 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620}
2621
2622/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002623 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2624 * but allow concurrent faults), and pte mapped but not yet locked.
2625 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002627static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2628 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002629 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002631 struct page *page;
2632 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634
Hugh Dickins62eede62009-09-21 17:03:34 -07002635 if (!(flags & FAULT_FLAG_WRITE)) {
2636 entry = pte_mkspecial(pfn_pte(my_zero_pfn(address),
2637 vma->vm_page_prot));
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002638 ptl = pte_lockptr(mm, pmd);
2639 spin_lock(ptl);
2640 if (!pte_none(*page_table))
2641 goto unlock;
2642 goto setpte;
2643 }
2644
Nick Piggin557ed1f2007-10-16 01:24:40 -07002645 /* Allocate our own private page. */
2646 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647
Nick Piggin557ed1f2007-10-16 01:24:40 -07002648 if (unlikely(anon_vma_prepare(vma)))
2649 goto oom;
2650 page = alloc_zeroed_user_highpage_movable(vma, address);
2651 if (!page)
2652 goto oom;
Nick Piggin0ed361d2008-02-04 22:29:34 -08002653 __SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002655 if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002656 goto oom_free_page;
2657
Nick Piggin557ed1f2007-10-16 01:24:40 -07002658 entry = mk_pte(page, vma->vm_page_prot);
Hugh Dickins1ac0cb52009-09-21 17:03:29 -07002659 if (vma->vm_flags & VM_WRITE)
2660 entry = pte_mkwrite(pte_mkdirty(entry));
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002661
Nick Piggin557ed1f2007-10-16 01:24:40 -07002662 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Andrea Arcangeli1c2fb7a2009-09-21 17:02:22 -07002663 if (!pte_none(*page_table))
Nick Piggin557ed1f2007-10-16 01:24:40 -07002664 goto release;
Hugh Dickins9ba69292009-09-21 17:02:20 -07002665
Nick Piggin557ed1f2007-10-16 01:24:40 -07002666 inc_mm_counter(mm, anon_rss);
Nick Piggin557ed1f2007-10-16 01:24:40 -07002667 page_add_new_anon_rmap(page, vma, address);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002668setpte:
Hugh Dickins65500d22005-10-29 18:15:59 -07002669 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670
2671 /* No need to invalidate - it was non-present before */
Hugh Dickins65500d22005-10-29 18:15:59 -07002672 update_mmu_cache(vma, address, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002673unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002674 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002675 return 0;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002676release:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002677 mem_cgroup_uncharge_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002678 page_cache_release(page);
2679 goto unlock;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002680oom_free_page:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002681 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002682oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683 return VM_FAULT_OOM;
2684}
2685
2686/*
Nick Piggin54cb8822007-07-19 01:46:59 -07002687 * __do_fault() tries to create a new page mapping. It aggressively
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 * tries to share with existing pages, but makes a separate copy if
Nick Piggin54cb8822007-07-19 01:46:59 -07002689 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2690 * the next page fault.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691 *
2692 * As this is called only for pages that do not currently exist, we
2693 * do not need to flush old virtual caches or the TLB.
2694 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002695 * We enter with non-exclusive mmap_sem (to exclude vma changes,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002696 * but allow concurrent faults), and pte neither mapped nor locked.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002697 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698 */
Nick Piggin54cb8822007-07-19 01:46:59 -07002699static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002700 unsigned long address, pmd_t *pmd,
Nick Piggin54cb8822007-07-19 01:46:59 -07002701 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702{
Hugh Dickins16abfa02007-10-04 16:56:06 +01002703 pte_t *page_table;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002704 spinlock_t *ptl;
Nick Piggind0217ac2007-07-19 01:47:03 -07002705 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707 int anon = 0;
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002708 int charged = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002709 struct page *dirty_page = NULL;
Nick Piggind0217ac2007-07-19 01:47:03 -07002710 struct vm_fault vmf;
2711 int ret;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002712 int page_mkwrite = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07002713
Nick Piggind0217ac2007-07-19 01:47:03 -07002714 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2715 vmf.pgoff = pgoff;
2716 vmf.flags = flags;
2717 vmf.page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718
Nick Piggin3c18ddd2008-04-28 02:12:10 -07002719 ret = vma->vm_ops->fault(vma, &vmf);
2720 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2721 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722
Andi Kleena3b947e2009-09-16 11:50:08 +02002723 if (unlikely(PageHWPoison(vmf.page))) {
2724 if (ret & VM_FAULT_LOCKED)
2725 unlock_page(vmf.page);
2726 return VM_FAULT_HWPOISON;
2727 }
2728
Nick Piggind00806b2007-07-19 01:46:57 -07002729 /*
Nick Piggind0217ac2007-07-19 01:47:03 -07002730 * For consistency in subsequent calls, make the faulted page always
Nick Piggind00806b2007-07-19 01:46:57 -07002731 * locked.
2732 */
Nick Piggin83c54072007-07-19 01:47:05 -07002733 if (unlikely(!(ret & VM_FAULT_LOCKED)))
Nick Piggind0217ac2007-07-19 01:47:03 -07002734 lock_page(vmf.page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002735 else
Nick Piggind0217ac2007-07-19 01:47:03 -07002736 VM_BUG_ON(!PageLocked(vmf.page));
Nick Piggind00806b2007-07-19 01:46:57 -07002737
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738 /*
2739 * Should we do an early C-O-W break?
2740 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002741 page = vmf.page;
Nick Piggin54cb8822007-07-19 01:46:59 -07002742 if (flags & FAULT_FLAG_WRITE) {
David Howells9637a5e2006-06-23 02:03:43 -07002743 if (!(vma->vm_flags & VM_SHARED)) {
Nick Piggin54cb8822007-07-19 01:46:59 -07002744 anon = 1;
Nick Piggind00806b2007-07-19 01:46:57 -07002745 if (unlikely(anon_vma_prepare(vma))) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002746 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002747 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002748 }
Nick Piggin83c54072007-07-19 01:47:05 -07002749 page = alloc_page_vma(GFP_HIGHUSER_MOVABLE,
2750 vma, address);
Nick Piggind00806b2007-07-19 01:46:57 -07002751 if (!page) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002752 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002753 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002754 }
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002755 if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) {
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002756 ret = VM_FAULT_OOM;
2757 page_cache_release(page);
2758 goto out;
2759 }
2760 charged = 1;
Nick Pigginb291f002008-10-18 20:26:44 -07002761 /*
2762 * Don't let another task, with possibly unlocked vma,
2763 * keep the mlocked page.
2764 */
2765 if (vma->vm_flags & VM_LOCKED)
2766 clear_page_mlock(vmf.page);
Nick Piggind0217ac2007-07-19 01:47:03 -07002767 copy_user_highpage(page, vmf.page, address, vma);
Nick Piggin0ed361d2008-02-04 22:29:34 -08002768 __SetPageUptodate(page);
David Howells9637a5e2006-06-23 02:03:43 -07002769 } else {
Nick Piggin54cb8822007-07-19 01:46:59 -07002770 /*
2771 * If the page will be shareable, see if the backing
David Howells9637a5e2006-06-23 02:03:43 -07002772 * address space wants to know that the page is about
Nick Piggin54cb8822007-07-19 01:46:59 -07002773 * to become writable
2774 */
Mark Fasheh69676142007-07-19 01:47:00 -07002775 if (vma->vm_ops->page_mkwrite) {
Nick Pigginc2ec1752009-03-31 15:23:21 -07002776 int tmp;
2777
Mark Fasheh69676142007-07-19 01:47:00 -07002778 unlock_page(page);
Nick Pigginb827e492009-04-30 15:08:16 -07002779 vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
Nick Pigginc2ec1752009-03-31 15:23:21 -07002780 tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
2781 if (unlikely(tmp &
2782 (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
2783 ret = tmp;
Nick Pigginb827e492009-04-30 15:08:16 -07002784 goto unwritable_page;
Mark Fasheh69676142007-07-19 01:47:00 -07002785 }
Nick Pigginb827e492009-04-30 15:08:16 -07002786 if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
2787 lock_page(page);
2788 if (!page->mapping) {
2789 ret = 0; /* retry the fault */
2790 unlock_page(page);
2791 goto unwritable_page;
2792 }
2793 } else
2794 VM_BUG_ON(!PageLocked(page));
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002795 page_mkwrite = 1;
David Howells9637a5e2006-06-23 02:03:43 -07002796 }
2797 }
Nick Piggin54cb8822007-07-19 01:46:59 -07002798
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799 }
2800
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002801 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802
2803 /*
2804 * This silly early PAGE_DIRTY setting removes a race
2805 * due to the bad i386 page protection. But it's valid
2806 * for other architectures too.
2807 *
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002808 * Note that if FAULT_FLAG_WRITE is set, we either now have
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809 * an exclusive copy of the page, or this is a shared mapping,
2810 * so we can make it writable and dirty to avoid having to
2811 * handle that later.
2812 */
2813 /* Only go through if we didn't race with anybody else... */
Andrea Arcangeli1c2fb7a2009-09-21 17:02:22 -07002814 if (likely(pte_same(*page_table, orig_pte))) {
Nick Piggind00806b2007-07-19 01:46:57 -07002815 flush_icache_page(vma, page);
2816 entry = mk_pte(page, vma->vm_page_prot);
Nick Piggin54cb8822007-07-19 01:46:59 -07002817 if (flags & FAULT_FLAG_WRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819 if (anon) {
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002820 inc_mm_counter(mm, anon_rss);
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002821 page_add_new_anon_rmap(page, vma, address);
Hugh Dickinsf57e88a2005-11-21 21:32:19 -08002822 } else {
Hugh Dickins42946212005-10-29 18:16:05 -07002823 inc_mm_counter(mm, file_rss);
Nick Piggind00806b2007-07-19 01:46:57 -07002824 page_add_file_rmap(page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002825 if (flags & FAULT_FLAG_WRITE) {
Nick Piggind00806b2007-07-19 01:46:57 -07002826 dirty_page = page;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002827 get_page(dirty_page);
2828 }
Hugh Dickins42946212005-10-29 18:16:05 -07002829 }
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002830 set_pte_at(mm, address, page_table, entry);
Nick Piggind00806b2007-07-19 01:46:57 -07002831
2832 /* no need to invalidate: a not-present page won't be cached */
2833 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834 } else {
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002835 if (charged)
2836 mem_cgroup_uncharge_page(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002837 if (anon)
2838 page_cache_release(page);
2839 else
Nick Piggin54cb8822007-07-19 01:46:59 -07002840 anon = 1; /* no anon but release faulted_page */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841 }
2842
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002843 pte_unmap_unlock(page_table, ptl);
Nick Piggind00806b2007-07-19 01:46:57 -07002844
2845out:
Nick Pigginb827e492009-04-30 15:08:16 -07002846 if (dirty_page) {
2847 struct address_space *mapping = page->mapping;
2848
2849 if (set_page_dirty(dirty_page))
2850 page_mkwrite = 1;
2851 unlock_page(dirty_page);
2852 put_page(dirty_page);
2853 if (page_mkwrite && mapping) {
2854 /*
2855 * Some device drivers do not set page.mapping but still
2856 * dirty their pages
2857 */
2858 balance_dirty_pages_ratelimited(mapping);
2859 }
2860
2861 /* file_update_time outside page_lock */
Anton Salikhmetov8f7b3d12008-01-23 02:21:18 +03002862 if (vma->vm_file)
2863 file_update_time(vma->vm_file);
Nick Pigginb827e492009-04-30 15:08:16 -07002864 } else {
2865 unlock_page(vmf.page);
2866 if (anon)
2867 page_cache_release(vmf.page);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002868 }
Nick Piggind00806b2007-07-19 01:46:57 -07002869
Nick Piggin83c54072007-07-19 01:47:05 -07002870 return ret;
Nick Pigginb827e492009-04-30 15:08:16 -07002871
2872unwritable_page:
2873 page_cache_release(page);
2874 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07002875}
Nick Piggind00806b2007-07-19 01:46:57 -07002876
Nick Piggin54cb8822007-07-19 01:46:59 -07002877static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2878 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002879 unsigned int flags, pte_t orig_pte)
Nick Piggin54cb8822007-07-19 01:46:59 -07002880{
2881 pgoff_t pgoff = (((address & PAGE_MASK)
Dean Nelson0da7e012007-10-16 01:24:45 -07002882 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
Nick Piggin54cb8822007-07-19 01:46:59 -07002883
Hugh Dickins16abfa02007-10-04 16:56:06 +01002884 pte_unmap(page_table);
2885 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Nick Piggin54cb8822007-07-19 01:46:59 -07002886}
2887
Jes Sorensenf4b81802006-09-27 01:50:10 -07002888/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889 * Fault of a previously existing named mapping. Repopulate the pte
2890 * from the encoded file_pte if possible. This enables swappable
2891 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002892 *
2893 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2894 * but allow concurrent faults), and pte mapped but not yet locked.
2895 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002897static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins65500d22005-10-29 18:15:59 -07002898 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002899 unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002900{
Hugh Dickins65500d22005-10-29 18:15:59 -07002901 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002903 flags |= FAULT_FLAG_NONLINEAR;
2904
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002905 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Nick Piggin83c54072007-07-19 01:47:05 -07002906 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907
Hugh Dickins2509ef22009-01-06 14:40:10 -08002908 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002909 /*
2910 * Page table corrupted: show pte and kill process.
2911 */
Hugh Dickins3dc14742009-01-06 14:40:08 -08002912 print_bad_pte(vma, address, orig_pte, NULL);
Hugh Dickins65500d22005-10-29 18:15:59 -07002913 return VM_FAULT_OOM;
2914 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002915
2916 pgoff = pte_to_pgoff(orig_pte);
Hugh Dickins16abfa02007-10-04 16:56:06 +01002917 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918}
2919
2920/*
2921 * These routines also need to handle stuff like marking pages dirty
2922 * and/or accessed for architectures that don't do it in hardware (most
2923 * RISC architectures). The early dirtying is also good on the i386.
2924 *
2925 * There is also a hook called "update_mmu_cache()" that architectures
2926 * with external mmu caches can use to update those (ie the Sparc or
2927 * PowerPC hashed page tables that act as extended TLBs).
2928 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07002929 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2930 * but allow concurrent faults), and pte mapped but not yet locked.
2931 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932 */
2933static inline int handle_pte_fault(struct mm_struct *mm,
Hugh Dickins65500d22005-10-29 18:15:59 -07002934 struct vm_area_struct *vma, unsigned long address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002935 pte_t *pte, pmd_t *pmd, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002936{
2937 pte_t entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002938 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002940 entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002941 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002942 if (pte_none(entry)) {
Jes Sorensenf4b81802006-09-27 01:50:10 -07002943 if (vma->vm_ops) {
Nick Piggin3c18ddd2008-04-28 02:12:10 -07002944 if (likely(vma->vm_ops->fault))
Nick Piggin54cb8822007-07-19 01:46:59 -07002945 return do_linear_fault(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002946 pte, pmd, flags, entry);
Jes Sorensenf4b81802006-09-27 01:50:10 -07002947 }
2948 return do_anonymous_page(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002949 pte, pmd, flags);
Hugh Dickins65500d22005-10-29 18:15:59 -07002950 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002951 if (pte_file(entry))
Nick Piggind0217ac2007-07-19 01:47:03 -07002952 return do_nonlinear_fault(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002953 pte, pmd, flags, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002954 return do_swap_page(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002955 pte, pmd, flags, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956 }
2957
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002958 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002959 spin_lock(ptl);
2960 if (unlikely(!pte_same(*pte, entry)))
2961 goto unlock;
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002962 if (flags & FAULT_FLAG_WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002964 return do_wp_page(mm, vma, address,
2965 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002966 entry = pte_mkdirty(entry);
2967 }
2968 entry = pte_mkyoung(entry);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002969 if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) {
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002970 update_mmu_cache(vma, address, entry);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002971 } else {
2972 /*
2973 * This is needed only for protection faults but the arch code
2974 * is not yet telling us if this is a protection fault or not.
2975 * This still avoids useless tlb flushes for .text page faults
2976 * with threads.
2977 */
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002978 if (flags & FAULT_FLAG_WRITE)
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002979 flush_tlb_page(vma, address);
2980 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002981unlock:
2982 pte_unmap_unlock(pte, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002983 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984}
2985
2986/*
2987 * By the time we get here, we already hold the mm semaphore
2988 */
Nick Piggin83c54072007-07-19 01:47:05 -07002989int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvaldsd06063c2009-04-10 09:01:23 -07002990 unsigned long address, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991{
2992 pgd_t *pgd;
2993 pud_t *pud;
2994 pmd_t *pmd;
2995 pte_t *pte;
2996
2997 __set_current_state(TASK_RUNNING);
2998
Christoph Lameterf8891e52006-06-30 01:55:45 -07002999 count_vm_event(PGFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01003001 if (unlikely(is_vm_hugetlb_page(vma)))
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003002 return hugetlb_fault(mm, vma, address, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003003
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005 pud = pud_alloc(mm, pgd, address);
3006 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07003007 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008 pmd = pmd_alloc(mm, pud, address);
3009 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07003010 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011 pte = pte_alloc_map(mm, pmd, address);
3012 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -07003013 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003015 return handle_pte_fault(mm, vma, address, pte, pmd, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003016}
3017
3018#ifndef __PAGETABLE_PUD_FOLDED
3019/*
3020 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07003021 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003022 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07003023int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024{
Hugh Dickinsc74df322005-10-29 18:16:23 -07003025 pud_t *new = pud_alloc_one(mm, address);
3026 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07003027 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028
Nick Piggin362a61a2008-05-14 06:37:36 +02003029 smp_wmb(); /* See comment in __pte_alloc */
3030
Hugh Dickins872fec12005-10-29 18:16:21 -07003031 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003032 if (pgd_present(*pgd)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003033 pud_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003034 else
3035 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07003036 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003037 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038}
3039#endif /* __PAGETABLE_PUD_FOLDED */
3040
3041#ifndef __PAGETABLE_PMD_FOLDED
3042/*
3043 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07003044 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07003046int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003047{
Hugh Dickinsc74df322005-10-29 18:16:23 -07003048 pmd_t *new = pmd_alloc_one(mm, address);
3049 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07003050 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051
Nick Piggin362a61a2008-05-14 06:37:36 +02003052 smp_wmb(); /* See comment in __pte_alloc */
3053
Hugh Dickins872fec12005-10-29 18:16:21 -07003054 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003055#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07003056 if (pud_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003057 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003058 else
3059 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07003061 if (pgd_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003062 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003063 else
3064 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07003066 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003067 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068}
3069#endif /* __PAGETABLE_PMD_FOLDED */
3070
3071int make_pages_present(unsigned long addr, unsigned long end)
3072{
3073 int ret, len, write;
3074 struct vm_area_struct * vma;
3075
3076 vma = find_vma(current->mm, addr);
3077 if (!vma)
KOSAKI Motohiroa4770972008-08-04 13:41:14 -07003078 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079 write = (vma->vm_flags & VM_WRITE) != 0;
Eric Sesterhenn5bcb28b2006-03-26 18:30:52 +02003080 BUG_ON(addr >= end);
3081 BUG_ON(end > vma->vm_end);
Rolf Eike Beer68e116a2007-07-15 23:38:03 -07003082 len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083 ret = get_user_pages(current, current->mm, addr,
3084 len, write, 0, NULL, NULL);
Lee Schermerhornc11d69d2008-10-18 20:26:56 -07003085 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086 return ret;
Lee Schermerhorn9978ad52008-10-18 20:26:56 -07003087 return ret == len ? 0 : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003088}
3089
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090#if !defined(__HAVE_ARCH_GATE_AREA)
3091
3092#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07003093static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003094
3095static int __init gate_vma_init(void)
3096{
3097 gate_vma.vm_mm = NULL;
3098 gate_vma.vm_start = FIXADDR_USER_START;
3099 gate_vma.vm_end = FIXADDR_USER_END;
Roland McGrathb6558c42007-01-26 00:56:47 -08003100 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
3101 gate_vma.vm_page_prot = __P101;
Roland McGrathf47aef52007-01-26 00:56:49 -08003102 /*
3103 * Make sure the vDSO gets into every core dump.
3104 * Dumping its contents makes post-mortem fully interpretable later
3105 * without matching up the same kernel and hardware config to see
3106 * what PC values meant.
3107 */
3108 gate_vma.vm_flags |= VM_ALWAYSDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109 return 0;
3110}
3111__initcall(gate_vma_init);
3112#endif
3113
3114struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
3115{
3116#ifdef AT_SYSINFO_EHDR
3117 return &gate_vma;
3118#else
3119 return NULL;
3120#endif
3121}
3122
3123int in_gate_area_no_task(unsigned long addr)
3124{
3125#ifdef AT_SYSINFO_EHDR
3126 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
3127 return 1;
3128#endif
3129 return 0;
3130}
3131
3132#endif /* __HAVE_ARCH_GATE_AREA */
David Howells0ec76a12006-09-27 01:50:15 -07003133
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07003134static int follow_pte(struct mm_struct *mm, unsigned long address,
3135 pte_t **ptepp, spinlock_t **ptlp)
3136{
3137 pgd_t *pgd;
3138 pud_t *pud;
3139 pmd_t *pmd;
3140 pte_t *ptep;
3141
3142 pgd = pgd_offset(mm, address);
3143 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
3144 goto out;
3145
3146 pud = pud_offset(pgd, address);
3147 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
3148 goto out;
3149
3150 pmd = pmd_offset(pud, address);
3151 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
3152 goto out;
3153
3154 /* We cannot handle huge page PFN maps. Luckily they don't exist. */
3155 if (pmd_huge(*pmd))
3156 goto out;
3157
3158 ptep = pte_offset_map_lock(mm, pmd, address, ptlp);
3159 if (!ptep)
3160 goto out;
3161 if (!pte_present(*ptep))
3162 goto unlock;
3163 *ptepp = ptep;
3164 return 0;
3165unlock:
3166 pte_unmap_unlock(ptep, *ptlp);
3167out:
3168 return -EINVAL;
3169}
3170
Johannes Weiner3b6748e2009-06-16 15:32:35 -07003171/**
3172 * follow_pfn - look up PFN at a user virtual address
3173 * @vma: memory mapping
3174 * @address: user virtual address
3175 * @pfn: location to store found PFN
3176 *
3177 * Only IO mappings and raw PFN mappings are allowed.
3178 *
3179 * Returns zero and the pfn at @pfn on success, -ve otherwise.
3180 */
3181int follow_pfn(struct vm_area_struct *vma, unsigned long address,
3182 unsigned long *pfn)
3183{
3184 int ret = -EINVAL;
3185 spinlock_t *ptl;
3186 pte_t *ptep;
3187
3188 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3189 return ret;
3190
3191 ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);
3192 if (ret)
3193 return ret;
3194 *pfn = pte_pfn(*ptep);
3195 pte_unmap_unlock(ptep, ptl);
3196 return 0;
3197}
3198EXPORT_SYMBOL(follow_pfn);
3199
Rik van Riel28b2ee22008-07-23 21:27:05 -07003200#ifdef CONFIG_HAVE_IOREMAP_PROT
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003201int follow_phys(struct vm_area_struct *vma,
3202 unsigned long address, unsigned int flags,
3203 unsigned long *prot, resource_size_t *phys)
Rik van Riel28b2ee22008-07-23 21:27:05 -07003204{
Johannes Weiner03668a42009-06-16 15:32:34 -07003205 int ret = -EINVAL;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003206 pte_t *ptep, pte;
3207 spinlock_t *ptl;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003208
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003209 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3210 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003211
Johannes Weiner03668a42009-06-16 15:32:34 -07003212 if (follow_pte(vma->vm_mm, address, &ptep, &ptl))
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003213 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003214 pte = *ptep;
Johannes Weiner03668a42009-06-16 15:32:34 -07003215
Rik van Riel28b2ee22008-07-23 21:27:05 -07003216 if ((flags & FOLL_WRITE) && !pte_write(pte))
3217 goto unlock;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003218
3219 *prot = pgprot_val(pte_pgprot(pte));
Johannes Weiner03668a42009-06-16 15:32:34 -07003220 *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003221
Johannes Weiner03668a42009-06-16 15:32:34 -07003222 ret = 0;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003223unlock:
3224 pte_unmap_unlock(ptep, ptl);
3225out:
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003226 return ret;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003227}
3228
3229int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
3230 void *buf, int len, int write)
3231{
3232 resource_size_t phys_addr;
3233 unsigned long prot = 0;
KOSAKI Motohiro2bc72732009-01-06 14:39:43 -08003234 void __iomem *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003235 int offset = addr & (PAGE_SIZE-1);
3236
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003237 if (follow_phys(vma, addr, write, &prot, &phys_addr))
Rik van Riel28b2ee22008-07-23 21:27:05 -07003238 return -EINVAL;
3239
3240 maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot);
3241 if (write)
3242 memcpy_toio(maddr + offset, buf, len);
3243 else
3244 memcpy_fromio(buf, maddr + offset, len);
3245 iounmap(maddr);
3246
3247 return len;
3248}
3249#endif
3250
David Howells0ec76a12006-09-27 01:50:15 -07003251/*
3252 * Access another process' address space.
3253 * Source/target buffer must be kernel space,
3254 * Do not walk the page table directly, use get_user_pages
3255 */
3256int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
3257{
3258 struct mm_struct *mm;
3259 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07003260 void *old_buf = buf;
3261
3262 mm = get_task_mm(tsk);
3263 if (!mm)
3264 return 0;
3265
3266 down_read(&mm->mmap_sem);
Simon Arlott183ff222007-10-20 01:27:18 +02003267 /* ignore errors, just check how much was successfully transferred */
David Howells0ec76a12006-09-27 01:50:15 -07003268 while (len) {
3269 int bytes, ret, offset;
3270 void *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003271 struct page *page = NULL;
David Howells0ec76a12006-09-27 01:50:15 -07003272
3273 ret = get_user_pages(tsk, mm, addr, 1,
3274 write, 1, &page, &vma);
Rik van Riel28b2ee22008-07-23 21:27:05 -07003275 if (ret <= 0) {
3276 /*
3277 * Check if this is a VM_IO | VM_PFNMAP VMA, which
3278 * we can access using slightly different code.
3279 */
3280#ifdef CONFIG_HAVE_IOREMAP_PROT
3281 vma = find_vma(mm, addr);
3282 if (!vma)
3283 break;
3284 if (vma->vm_ops && vma->vm_ops->access)
3285 ret = vma->vm_ops->access(vma, addr, buf,
3286 len, write);
3287 if (ret <= 0)
3288#endif
3289 break;
3290 bytes = ret;
David Howells0ec76a12006-09-27 01:50:15 -07003291 } else {
Rik van Riel28b2ee22008-07-23 21:27:05 -07003292 bytes = len;
3293 offset = addr & (PAGE_SIZE-1);
3294 if (bytes > PAGE_SIZE-offset)
3295 bytes = PAGE_SIZE-offset;
3296
3297 maddr = kmap(page);
3298 if (write) {
3299 copy_to_user_page(vma, page, addr,
3300 maddr + offset, buf, bytes);
3301 set_page_dirty_lock(page);
3302 } else {
3303 copy_from_user_page(vma, page, addr,
3304 buf, maddr + offset, bytes);
3305 }
3306 kunmap(page);
3307 page_cache_release(page);
David Howells0ec76a12006-09-27 01:50:15 -07003308 }
David Howells0ec76a12006-09-27 01:50:15 -07003309 len -= bytes;
3310 buf += bytes;
3311 addr += bytes;
3312 }
3313 up_read(&mm->mmap_sem);
3314 mmput(mm);
3315
3316 return buf - old_buf;
3317}
Andi Kleen03252912008-01-30 13:33:18 +01003318
3319/*
3320 * Print the name of a VMA.
3321 */
3322void print_vma_addr(char *prefix, unsigned long ip)
3323{
3324 struct mm_struct *mm = current->mm;
3325 struct vm_area_struct *vma;
3326
Ingo Molnare8bff742008-02-13 20:21:06 +01003327 /*
3328 * Do not print if we are in atomic
3329 * contexts (in exception stacks, etc.):
3330 */
3331 if (preempt_count())
3332 return;
3333
Andi Kleen03252912008-01-30 13:33:18 +01003334 down_read(&mm->mmap_sem);
3335 vma = find_vma(mm, ip);
3336 if (vma && vma->vm_file) {
3337 struct file *f = vma->vm_file;
3338 char *buf = (char *)__get_free_page(GFP_KERNEL);
3339 if (buf) {
3340 char *p, *s;
3341
Jan Blunckcf28b482008-02-14 19:38:44 -08003342 p = d_path(&f->f_path, buf, PAGE_SIZE);
Andi Kleen03252912008-01-30 13:33:18 +01003343 if (IS_ERR(p))
3344 p = "?";
3345 s = strrchr(p, '/');
3346 if (s)
3347 p = s+1;
3348 printk("%s%s[%lx+%lx]", prefix, p,
3349 vma->vm_start,
3350 vma->vm_end - vma->vm_start);
3351 free_page((unsigned long)buf);
3352 }
3353 }
3354 up_read(&current->mm->mmap_sem);
3355}
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003356
3357#ifdef CONFIG_PROVE_LOCKING
3358void might_fault(void)
3359{
Peter Zijlstra95156f02009-01-12 13:02:11 +01003360 /*
3361 * Some code (nfs/sunrpc) uses socket ops on kernel memory while
3362 * holding the mmap_sem, this is safe because kernel memory doesn't
3363 * get paged out, therefore we'll never actually fault, and the
3364 * below annotations will generate false positives.
3365 */
3366 if (segment_eq(get_fs(), KERNEL_DS))
3367 return;
3368
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003369 might_sleep();
3370 /*
3371 * it would be nicer only to annotate paths which are not under
3372 * pagefault_disable, however that requires a larger audit and
3373 * providing helpers like get_user_atomic.
3374 */
3375 if (!in_atomic() && current->mm)
3376 might_lock_read(&current->mm->mmap_sem);
3377}
3378EXPORT_SYMBOL(might_fault);
3379#endif