blob: 7e91b5f9f690e4ae9d7c3e58bc1530c995311089 [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{
644 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700645 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -0700646 int progress = 0;
Hugh Dickins8c103762005-10-29 18:16:13 -0700647 int rss[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
649again:
Hugh Dickinsae859762005-10-29 18:16:05 -0700650 rss[1] = rss[0] = 0;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700651 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 if (!dst_pte)
653 return -ENOMEM;
654 src_pte = pte_offset_map_nested(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700655 src_ptl = pte_lockptr(src_mm, src_pmd);
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700656 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700657 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 do {
660 /*
661 * We are holding two locks at this point - either of them
662 * could generate latencies in another task on another CPU.
663 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700664 if (progress >= 32) {
665 progress = 0;
666 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +0100667 spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700668 break;
669 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 if (pte_none(*src_pte)) {
671 progress++;
672 continue;
673 }
Hugh Dickins8c103762005-10-29 18:16:13 -0700674 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 progress += 8;
676 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700678 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -0700679 spin_unlock(src_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 pte_unmap_nested(src_pte - 1);
Hugh Dickinsae859762005-10-29 18:16:05 -0700681 add_mm_rss(dst_mm, rss[0], rss[1]);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700682 pte_unmap_unlock(dst_pte - 1, dst_ptl);
683 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 if (addr != end)
685 goto again;
686 return 0;
687}
688
689static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
690 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
691 unsigned long addr, unsigned long end)
692{
693 pmd_t *src_pmd, *dst_pmd;
694 unsigned long next;
695
696 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
697 if (!dst_pmd)
698 return -ENOMEM;
699 src_pmd = pmd_offset(src_pud, addr);
700 do {
701 next = pmd_addr_end(addr, end);
702 if (pmd_none_or_clear_bad(src_pmd))
703 continue;
704 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
705 vma, addr, next))
706 return -ENOMEM;
707 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
708 return 0;
709}
710
711static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
712 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
713 unsigned long addr, unsigned long end)
714{
715 pud_t *src_pud, *dst_pud;
716 unsigned long next;
717
718 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
719 if (!dst_pud)
720 return -ENOMEM;
721 src_pud = pud_offset(src_pgd, addr);
722 do {
723 next = pud_addr_end(addr, end);
724 if (pud_none_or_clear_bad(src_pud))
725 continue;
726 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
727 vma, addr, next))
728 return -ENOMEM;
729 } while (dst_pud++, src_pud++, addr = next, addr != end);
730 return 0;
731}
732
733int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
734 struct vm_area_struct *vma)
735{
736 pgd_t *src_pgd, *dst_pgd;
737 unsigned long next;
738 unsigned long addr = vma->vm_start;
739 unsigned long end = vma->vm_end;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700740 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741
Nick Piggind9928952005-08-28 16:49:11 +1000742 /*
743 * Don't copy ptes where a page fault will fill them correctly.
744 * Fork becomes much lighter when there are big shared or private
745 * readonly mappings. The tradeoff is that copy_page_range is more
746 * efficient than faulting.
747 */
Linus Torvalds4d7672b2005-12-16 10:21:23 -0800748 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
Nick Piggind9928952005-08-28 16:49:11 +1000749 if (!vma->anon_vma)
750 return 0;
751 }
752
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 if (is_vm_hugetlb_page(vma))
754 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
755
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800756 if (unlikely(is_pfn_mapping(vma))) {
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -0800757 /*
758 * We do not free on error cases below as remove_vma
759 * gets called on error from higher level routine
760 */
761 ret = track_pfn_vma_copy(vma);
762 if (ret)
763 return ret;
764 }
765
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700766 /*
767 * We need to invalidate the secondary MMU mappings only when
768 * there could be a permission downgrade on the ptes of the
769 * parent mm. And a permission downgrade will only happen if
770 * is_cow_mapping() returns true.
771 */
772 if (is_cow_mapping(vma->vm_flags))
773 mmu_notifier_invalidate_range_start(src_mm, addr, end);
774
775 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 dst_pgd = pgd_offset(dst_mm, addr);
777 src_pgd = pgd_offset(src_mm, addr);
778 do {
779 next = pgd_addr_end(addr, end);
780 if (pgd_none_or_clear_bad(src_pgd))
781 continue;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700782 if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
783 vma, addr, next))) {
784 ret = -ENOMEM;
785 break;
786 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700788
789 if (is_cow_mapping(vma->vm_flags))
790 mmu_notifier_invalidate_range_end(src_mm,
791 vma->vm_start, end);
792 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793}
794
Robin Holt51c6f662005-11-13 16:06:42 -0800795static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700796 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800798 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799{
Nick Pigginb5810032005-10-29 18:16:12 -0700800 struct mm_struct *mm = tlb->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 pte_t *pte;
Hugh Dickins508034a2005-10-29 18:16:30 -0700802 spinlock_t *ptl;
Hugh Dickinsae859762005-10-29 18:16:05 -0700803 int file_rss = 0;
804 int anon_rss = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805
Hugh Dickins508034a2005-10-29 18:16:30 -0700806 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700807 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 do {
809 pte_t ptent = *pte;
Robin Holt51c6f662005-11-13 16:06:42 -0800810 if (pte_none(ptent)) {
811 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800813 }
Hugh Dickins6f5e6b92006-03-16 23:04:09 -0800814
815 (*zap_work) -= PAGE_SIZE;
816
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800818 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -0800819
Linus Torvalds6aab3412005-11-28 14:34:23 -0800820 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 if (unlikely(details) && page) {
822 /*
823 * unmap_shared_mapping_pages() wants to
824 * invalidate cache without truncating:
825 * unmap shared but keep private pages.
826 */
827 if (details->check_mapping &&
828 details->check_mapping != page->mapping)
829 continue;
830 /*
831 * Each page->index must be checked when
832 * invalidating or truncating nonlinear.
833 */
834 if (details->nonlinear_vma &&
835 (page->index < details->first_index ||
836 page->index > details->last_index))
837 continue;
838 }
Nick Pigginb5810032005-10-29 18:16:12 -0700839 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -0700840 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 tlb_remove_tlb_entry(tlb, pte, addr);
842 if (unlikely(!page))
843 continue;
844 if (unlikely(details) && details->nonlinear_vma
845 && linear_page_index(details->nonlinear_vma,
846 addr) != page->index)
Nick Pigginb5810032005-10-29 18:16:12 -0700847 set_pte_at(mm, addr, pte,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 pgoff_to_pte(page->index));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 if (PageAnon(page))
Hugh Dickins86d912f2005-10-29 18:16:14 -0700850 anon_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700851 else {
852 if (pte_dirty(ptent))
853 set_page_dirty(page);
Johannes Weiner4917e5d2009-01-06 14:39:17 -0800854 if (pte_young(ptent) &&
855 likely(!VM_SequentialReadHint(vma)))
Nick Pigginbf3f3bc2009-01-06 14:38:55 -0800856 mark_page_accessed(page);
Hugh Dickins86d912f2005-10-29 18:16:14 -0700857 file_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700858 }
Hugh Dickinsedc315f2009-01-06 14:40:11 -0800859 page_remove_rmap(page);
Hugh Dickins3dc14742009-01-06 14:40:08 -0800860 if (unlikely(page_mapcount(page) < 0))
861 print_bad_pte(vma, addr, ptent, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 tlb_remove_page(tlb, page);
863 continue;
864 }
865 /*
866 * If details->check_mapping, we leave swap entries;
867 * if details->nonlinear_vma, we leave file entries.
868 */
869 if (unlikely(details))
870 continue;
Hugh Dickins2509ef22009-01-06 14:40:10 -0800871 if (pte_file(ptent)) {
872 if (unlikely(!(vma->vm_flags & VM_NONLINEAR)))
873 print_bad_pte(vma, addr, ptent, NULL);
874 } else if
875 (unlikely(!free_swap_and_cache(pte_to_swp_entry(ptent))))
876 print_bad_pte(vma, addr, ptent, NULL);
Zachary Amsden9888a1c2006-09-30 23:29:31 -0700877 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
Robin Holt51c6f662005-11-13 16:06:42 -0800878 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
Hugh Dickinsae859762005-10-29 18:16:05 -0700879
Hugh Dickins86d912f2005-10-29 18:16:14 -0700880 add_mm_rss(mm, file_rss, anon_rss);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700881 arch_leave_lazy_mmu_mode();
Hugh Dickins508034a2005-10-29 18:16:30 -0700882 pte_unmap_unlock(pte - 1, ptl);
Robin Holt51c6f662005-11-13 16:06:42 -0800883
884 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885}
886
Robin Holt51c6f662005-11-13 16:06:42 -0800887static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700888 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800890 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891{
892 pmd_t *pmd;
893 unsigned long next;
894
895 pmd = pmd_offset(pud, addr);
896 do {
897 next = pmd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800898 if (pmd_none_or_clear_bad(pmd)) {
899 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800901 }
902 next = zap_pte_range(tlb, vma, pmd, addr, next,
903 zap_work, details);
904 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
905
906 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907}
908
Robin Holt51c6f662005-11-13 16:06:42 -0800909static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700910 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800912 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913{
914 pud_t *pud;
915 unsigned long next;
916
917 pud = pud_offset(pgd, addr);
918 do {
919 next = pud_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800920 if (pud_none_or_clear_bad(pud)) {
921 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800923 }
924 next = zap_pmd_range(tlb, vma, pud, addr, next,
925 zap_work, details);
926 } while (pud++, addr = next, (addr != end && *zap_work > 0));
927
928 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929}
930
Robin Holt51c6f662005-11-13 16:06:42 -0800931static unsigned long unmap_page_range(struct mmu_gather *tlb,
932 struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800934 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935{
936 pgd_t *pgd;
937 unsigned long next;
938
939 if (details && !details->check_mapping && !details->nonlinear_vma)
940 details = NULL;
941
942 BUG_ON(addr >= end);
943 tlb_start_vma(tlb, vma);
944 pgd = pgd_offset(vma->vm_mm, addr);
945 do {
946 next = pgd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800947 if (pgd_none_or_clear_bad(pgd)) {
948 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800950 }
951 next = zap_pud_range(tlb, vma, pgd, addr, next,
952 zap_work, details);
953 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 tlb_end_vma(tlb, vma);
Robin Holt51c6f662005-11-13 16:06:42 -0800955
956 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957}
958
959#ifdef CONFIG_PREEMPT
960# define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
961#else
962/* No preempt: go for improved straight-line efficiency */
963# define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
964#endif
965
966/**
967 * unmap_vmas - unmap a range of memory covered by a list of vma's
968 * @tlbp: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 * @vma: the starting vma
970 * @start_addr: virtual address at which to start unmapping
971 * @end_addr: virtual address at which to end unmapping
972 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
973 * @details: details of nonlinear truncation or shared cache invalidation
974 *
Hugh Dickinsee39b372005-04-19 13:29:15 -0700975 * Returns the end address of the unmapping (restart addr if interrupted).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700977 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700979 * We aim to not hold locks for too long (for scheduling latency reasons).
980 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 * return the ending mmu_gather to the caller.
982 *
983 * Only addresses between `start' and `end' will be unmapped.
984 *
985 * The VMA list must be sorted in ascending virtual address order.
986 *
987 * unmap_vmas() assumes that the caller will flush the whole unmapped address
988 * range after unmap_vmas() returns. So the only responsibility here is to
989 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
990 * drops the lock and schedules.
991 */
Hugh Dickins508034a2005-10-29 18:16:30 -0700992unsigned long unmap_vmas(struct mmu_gather **tlbp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 struct vm_area_struct *vma, unsigned long start_addr,
994 unsigned long end_addr, unsigned long *nr_accounted,
995 struct zap_details *details)
996{
Robin Holt51c6f662005-11-13 16:06:42 -0800997 long zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
999 int tlb_start_valid = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -07001000 unsigned long start = start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -07001002 int fullmm = (*tlbp)->fullmm;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001003 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001005 mmu_notifier_invalidate_range_start(mm, start_addr, end_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 unsigned long end;
1008
1009 start = max(vma->vm_start, start_addr);
1010 if (start >= vma->vm_end)
1011 continue;
1012 end = min(vma->vm_end, end_addr);
1013 if (end <= vma->vm_start)
1014 continue;
1015
1016 if (vma->vm_flags & VM_ACCOUNT)
1017 *nr_accounted += (end - start) >> PAGE_SHIFT;
1018
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -08001019 if (unlikely(is_pfn_mapping(vma)))
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001020 untrack_pfn_vma(vma, 0, 0);
1021
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 while (start != end) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 if (!tlb_start_valid) {
1024 tlb_start = start;
1025 tlb_start_valid = 1;
1026 }
1027
Robin Holt51c6f662005-11-13 16:06:42 -08001028 if (unlikely(is_vm_hugetlb_page(vma))) {
Andi Kleena137e1c2008-07-23 21:27:43 -07001029 /*
1030 * It is undesirable to test vma->vm_file as it
1031 * should be non-null for valid hugetlb area.
1032 * However, vm_file will be NULL in the error
1033 * cleanup path of do_mmap_pgoff. When
1034 * hugetlbfs ->mmap method fails,
1035 * do_mmap_pgoff() nullifies vma->vm_file
1036 * before calling this function to clean up.
1037 * Since no pte has actually been setup, it is
1038 * safe to do nothing in this case.
1039 */
1040 if (vma->vm_file) {
1041 unmap_hugepage_range(vma, start, end, NULL);
1042 zap_work -= (end - start) /
Andi Kleena5516432008-07-23 21:27:41 -07001043 pages_per_huge_page(hstate_vma(vma));
Andi Kleena137e1c2008-07-23 21:27:43 -07001044 }
1045
Robin Holt51c6f662005-11-13 16:06:42 -08001046 start = end;
1047 } else
1048 start = unmap_page_range(*tlbp, vma,
1049 start, end, &zap_work, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050
Robin Holt51c6f662005-11-13 16:06:42 -08001051 if (zap_work > 0) {
1052 BUG_ON(start != end);
1053 break;
1054 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055
1056 tlb_finish_mmu(*tlbp, tlb_start, start);
1057
1058 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +01001059 (i_mmap_lock && spin_needbreak(i_mmap_lock))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 if (i_mmap_lock) {
Hugh Dickins508034a2005-10-29 18:16:30 -07001061 *tlbp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 goto out;
1063 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 }
1066
Hugh Dickins508034a2005-10-29 18:16:30 -07001067 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 tlb_start_valid = 0;
Robin Holt51c6f662005-11-13 16:06:42 -08001069 zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 }
1071 }
1072out:
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001073 mmu_notifier_invalidate_range_end(mm, start_addr, end_addr);
Hugh Dickinsee39b372005-04-19 13:29:15 -07001074 return start; /* which is now the end (or restart) address */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075}
1076
1077/**
1078 * zap_page_range - remove user pages in a given range
1079 * @vma: vm_area_struct holding the applicable pages
1080 * @address: starting address of pages to zap
1081 * @size: number of bytes to zap
1082 * @details: details of nonlinear truncation or shared cache invalidation
1083 */
Hugh Dickinsee39b372005-04-19 13:29:15 -07001084unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 unsigned long size, struct zap_details *details)
1086{
1087 struct mm_struct *mm = vma->vm_mm;
1088 struct mmu_gather *tlb;
1089 unsigned long end = address + size;
1090 unsigned long nr_accounted = 0;
1091
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001094 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -07001095 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
1096 if (tlb)
1097 tlb_finish_mmu(tlb, address, end);
Hugh Dickinsee39b372005-04-19 13:29:15 -07001098 return end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099}
1100
Jack Steinerc627f9c2008-07-29 22:33:53 -07001101/**
1102 * zap_vma_ptes - remove ptes mapping the vma
1103 * @vma: vm_area_struct holding ptes to be zapped
1104 * @address: starting address of pages to zap
1105 * @size: number of bytes to zap
1106 *
1107 * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1108 *
1109 * The entire address range must be fully contained within the vma.
1110 *
1111 * Returns 0 if successful.
1112 */
1113int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
1114 unsigned long size)
1115{
1116 if (address < vma->vm_start || address + size > vma->vm_end ||
1117 !(vma->vm_flags & VM_PFNMAP))
1118 return -1;
1119 zap_page_range(vma, address, size, NULL);
1120 return 0;
1121}
1122EXPORT_SYMBOL_GPL(zap_vma_ptes);
1123
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124/*
1125 * Do a quick page-table lookup for a single page.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 */
Linus Torvalds6aab3412005-11-28 14:34:23 -08001127struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001128 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129{
1130 pgd_t *pgd;
1131 pud_t *pud;
1132 pmd_t *pmd;
1133 pte_t *ptep, pte;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001134 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 struct page *page;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001136 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001138 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
1139 if (!IS_ERR(page)) {
1140 BUG_ON(flags & FOLL_GET);
1141 goto out;
1142 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001144 page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 pgd = pgd_offset(mm, address);
1146 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001147 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148
1149 pud = pud_offset(pgd, address);
Andi Kleenceb86872008-07-23 21:27:50 -07001150 if (pud_none(*pud))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001151 goto no_page_table;
Andi Kleenceb86872008-07-23 21:27:50 -07001152 if (pud_huge(*pud)) {
1153 BUG_ON(flags & FOLL_GET);
1154 page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE);
1155 goto out;
1156 }
1157 if (unlikely(pud_bad(*pud)))
1158 goto no_page_table;
1159
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 pmd = pmd_offset(pud, address);
Hugh Dickinsaeed5fc2008-05-06 20:49:23 +01001161 if (pmd_none(*pmd))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001162 goto no_page_table;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001163 if (pmd_huge(*pmd)) {
1164 BUG_ON(flags & FOLL_GET);
1165 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
1166 goto out;
1167 }
Hugh Dickinsaeed5fc2008-05-06 20:49:23 +01001168 if (unlikely(pmd_bad(*pmd)))
1169 goto no_page_table;
1170
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001171 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172
1173 pte = *ptep;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001174 if (!pte_present(pte))
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001175 goto no_page;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001176 if ((flags & FOLL_WRITE) && !pte_write(pte))
1177 goto unlock;
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07001178
Linus Torvalds6aab3412005-11-28 14:34:23 -08001179 page = vm_normal_page(vma, address, pte);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07001180 if (unlikely(!page)) {
1181 if ((flags & FOLL_DUMP) ||
Hugh Dickins62eede62009-09-21 17:03:34 -07001182 !is_zero_pfn(pte_pfn(pte)))
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07001183 goto bad_page;
1184 page = pte_page(pte);
1185 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001186
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001187 if (flags & FOLL_GET)
1188 get_page(page);
1189 if (flags & FOLL_TOUCH) {
1190 if ((flags & FOLL_WRITE) &&
1191 !pte_dirty(pte) && !PageDirty(page))
1192 set_page_dirty(page);
KOSAKI Motohirobd775c42009-03-31 15:19:37 -07001193 /*
1194 * pte_mkyoung() would be more correct here, but atomic care
1195 * is needed to avoid losing the dirty bit: it is easier to use
1196 * mark_page_accessed().
1197 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001198 mark_page_accessed(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001200unlock:
1201 pte_unmap_unlock(ptep, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202out:
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001203 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001205bad_page:
1206 pte_unmap_unlock(ptep, ptl);
1207 return ERR_PTR(-EFAULT);
1208
1209no_page:
1210 pte_unmap_unlock(ptep, ptl);
1211 if (!pte_none(pte))
1212 return page;
Hugh Dickins8e4b9a602009-09-21 17:03:26 -07001213
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001214no_page_table:
1215 /*
1216 * When core dumping an enormous anonymous area that nobody
Hugh Dickins8e4b9a602009-09-21 17:03:26 -07001217 * has touched so far, we don't want to allocate unnecessary pages or
1218 * page tables. Return error instead of NULL to skip handle_mm_fault,
1219 * then get_dump_page() will return NULL to leave a hole in the dump.
1220 * But we can only make this optimization where a hole would surely
1221 * be zero-filled if handle_mm_fault() actually did handle it.
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001222 */
Hugh Dickins8e4b9a602009-09-21 17:03:26 -07001223 if ((flags & FOLL_DUMP) &&
1224 (!vma->vm_ops || !vma->vm_ops->fault))
1225 return ERR_PTR(-EFAULT);
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001226 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227}
1228
Nick Pigginb291f002008-10-18 20:26:44 -07001229int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
Hugh Dickins58fa8792009-09-21 17:03:31 -07001230 unsigned long start, int nr_pages, unsigned int gup_flags,
Peter Zijlstra9d737772009-06-25 11:58:55 +02001231 struct page **pages, struct vm_area_struct **vmas)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232{
1233 int i;
Hugh Dickins58fa8792009-09-21 17:03:31 -07001234 unsigned long vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235
Peter Zijlstra9d737772009-06-25 11:58:55 +02001236 if (nr_pages <= 0)
Jonathan Corbet900cf082008-02-11 16:17:33 -07001237 return 0;
Hugh Dickins58fa8792009-09-21 17:03:31 -07001238
1239 VM_BUG_ON(!!pages != !!(gup_flags & FOLL_GET));
1240
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 /*
1242 * Require read or write permissions.
Hugh Dickins58fa8792009-09-21 17:03:31 -07001243 * If FOLL_FORCE is set, we only require the "MAY" flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 */
Hugh Dickins58fa8792009-09-21 17:03:31 -07001245 vm_flags = (gup_flags & FOLL_WRITE) ?
1246 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
1247 vm_flags &= (gup_flags & FOLL_FORCE) ?
1248 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 i = 0;
1250
1251 do {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001252 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
1254 vma = find_extend_vma(mm, start);
1255 if (!vma && in_gate_area(tsk, start)) {
1256 unsigned long pg = start & PAGE_MASK;
1257 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
1258 pgd_t *pgd;
1259 pud_t *pud;
1260 pmd_t *pmd;
1261 pte_t *pte;
Nick Pigginb291f002008-10-18 20:26:44 -07001262
1263 /* user gate pages are read-only */
Hugh Dickins58fa8792009-09-21 17:03:31 -07001264 if (gup_flags & FOLL_WRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 return i ? : -EFAULT;
1266 if (pg > TASK_SIZE)
1267 pgd = pgd_offset_k(pg);
1268 else
1269 pgd = pgd_offset_gate(mm, pg);
1270 BUG_ON(pgd_none(*pgd));
1271 pud = pud_offset(pgd, pg);
1272 BUG_ON(pud_none(*pud));
1273 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001274 if (pmd_none(*pmd))
1275 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001277 if (pte_none(*pte)) {
1278 pte_unmap(pte);
1279 return i ? : -EFAULT;
1280 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 if (pages) {
Nick Pigginfa2a4552005-11-29 18:43:17 +11001282 struct page *page = vm_normal_page(gate_vma, start, *pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001283 pages[i] = page;
1284 if (page)
1285 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 }
1287 pte_unmap(pte);
1288 if (vmas)
1289 vmas[i] = gate_vma;
1290 i++;
1291 start += PAGE_SIZE;
Peter Zijlstra9d737772009-06-25 11:58:55 +02001292 nr_pages--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 continue;
1294 }
1295
Nick Pigginb291f002008-10-18 20:26:44 -07001296 if (!vma ||
1297 (vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
Hugh Dickins1c3aff12009-09-21 17:03:24 -07001298 !(vm_flags & vma->vm_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 return i ? : -EFAULT;
1300
Hugh Dickins2a15efc2009-09-21 17:03:27 -07001301 if (is_vm_hugetlb_page(vma)) {
1302 i = follow_hugetlb_page(mm, vma, pages, vmas,
Hugh Dickins58fa8792009-09-21 17:03:31 -07001303 &start, &nr_pages, i, gup_flags);
Hugh Dickins2a15efc2009-09-21 17:03:27 -07001304 continue;
1305 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001306
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 do {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001308 struct page *page;
Hugh Dickins58fa8792009-09-21 17:03:31 -07001309 unsigned int foll_flags = gup_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310
Ethan Solomita462e00c2007-07-15 23:38:16 -07001311 /*
Ying Han47792802009-01-06 14:40:18 -08001312 * If we have a pending SIGKILL, don't keep faulting
Hugh Dickins1c3aff12009-09-21 17:03:24 -07001313 * pages and potentially allocating memory.
Ethan Solomita462e00c2007-07-15 23:38:16 -07001314 */
Hugh Dickins1c3aff12009-09-21 17:03:24 -07001315 if (unlikely(fatal_signal_pending(current)))
Ying Han47792802009-01-06 14:40:18 -08001316 return i ? i : -ERESTARTSYS;
Ethan Solomita462e00c2007-07-15 23:38:16 -07001317
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001318 cond_resched();
Linus Torvalds6aab3412005-11-28 14:34:23 -08001319 while (!(page = follow_page(vma, start, foll_flags))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001320 int ret;
Linus Torvaldsd06063c2009-04-10 09:01:23 -07001321
Hugh Dickinsd26ed652009-06-23 13:52:49 +01001322 ret = handle_mm_fault(mm, vma, start,
1323 (foll_flags & FOLL_WRITE) ?
1324 FAULT_FLAG_WRITE : 0);
1325
Nick Piggin83c54072007-07-19 01:47:05 -07001326 if (ret & VM_FAULT_ERROR) {
1327 if (ret & VM_FAULT_OOM)
1328 return i ? i : -ENOMEM;
Andi Kleend1737fd2009-09-16 11:50:06 +02001329 if (ret &
1330 (VM_FAULT_HWPOISON|VM_FAULT_SIGBUS))
Nick Piggin83c54072007-07-19 01:47:05 -07001331 return i ? i : -EFAULT;
1332 BUG();
1333 }
1334 if (ret & VM_FAULT_MAJOR)
1335 tsk->maj_flt++;
1336 else
1337 tsk->min_flt++;
1338
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001339 /*
Nick Piggin83c54072007-07-19 01:47:05 -07001340 * The VM_FAULT_WRITE bit tells us that
1341 * do_wp_page has broken COW when necessary,
1342 * even if maybe_mkwrite decided not to set
1343 * pte_write. We can thus safely do subsequent
Hugh Dickins878b63a2009-01-06 14:39:32 -08001344 * page lookups as if they were reads. But only
1345 * do so when looping for pte_write is futile:
1346 * in some cases userspace may also be wanting
1347 * to write to the gotten user page, which a
1348 * read fault here might prevent (a readonly
1349 * page might get reCOWed by userspace write).
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001350 */
Hugh Dickins878b63a2009-01-06 14:39:32 -08001351 if ((ret & VM_FAULT_WRITE) &&
1352 !(vma->vm_flags & VM_WRITE))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001353 foll_flags &= ~FOLL_WRITE;
Nick Piggin83c54072007-07-19 01:47:05 -07001354
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07001355 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 }
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001357 if (IS_ERR(page))
1358 return i ? i : PTR_ERR(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001360 pages[i] = page;
James Bottomley03beb072006-03-26 01:36:57 -08001361
Russell Kinga6f36be2006-12-30 22:24:19 +00001362 flush_anon_page(vma, page, start);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001363 flush_dcache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 }
1365 if (vmas)
1366 vmas[i] = vma;
1367 i++;
1368 start += PAGE_SIZE;
Peter Zijlstra9d737772009-06-25 11:58:55 +02001369 nr_pages--;
1370 } while (nr_pages && start < vma->vm_end);
1371 } while (nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 return i;
1373}
Nick Pigginb291f002008-10-18 20:26:44 -07001374
Nick Piggind2bf6be2009-06-16 15:31:39 -07001375/**
1376 * get_user_pages() - pin user pages in memory
1377 * @tsk: task_struct of target task
1378 * @mm: mm_struct of target mm
1379 * @start: starting user address
Peter Zijlstra9d737772009-06-25 11:58:55 +02001380 * @nr_pages: number of pages from start to pin
Nick Piggind2bf6be2009-06-16 15:31:39 -07001381 * @write: whether pages will be written to by the caller
1382 * @force: whether to force write access even if user mapping is
1383 * readonly. This will result in the page being COWed even
1384 * in MAP_SHARED mappings. You do not want this.
1385 * @pages: array that receives pointers to the pages pinned.
1386 * Should be at least nr_pages long. Or NULL, if caller
1387 * only intends to ensure the pages are faulted in.
1388 * @vmas: array of pointers to vmas corresponding to each page.
1389 * Or NULL if the caller does not require them.
1390 *
1391 * Returns number of pages pinned. This may be fewer than the number
Peter Zijlstra9d737772009-06-25 11:58:55 +02001392 * requested. If nr_pages is 0 or negative, returns 0. If no pages
Nick Piggind2bf6be2009-06-16 15:31:39 -07001393 * were pinned, returns -errno. Each page returned must be released
1394 * with a put_page() call when it is finished with. vmas will only
1395 * remain valid while mmap_sem is held.
1396 *
1397 * Must be called with mmap_sem held for read or write.
1398 *
1399 * get_user_pages walks a process's page tables and takes a reference to
1400 * each struct page that each user address corresponds to at a given
1401 * instant. That is, it takes the page that would be accessed if a user
1402 * thread accesses the given user virtual address at that instant.
1403 *
1404 * This does not guarantee that the page exists in the user mappings when
1405 * get_user_pages returns, and there may even be a completely different
1406 * page there in some cases (eg. if mmapped pagecache has been invalidated
1407 * and subsequently re faulted). However it does guarantee that the page
1408 * won't be freed completely. And mostly callers simply care that the page
1409 * contains data that was valid *at some point in time*. Typically, an IO
1410 * or similar operation cannot guarantee anything stronger anyway because
1411 * locks can't be held over the syscall boundary.
1412 *
1413 * If write=0, the page must not be written to. If the page is written to,
1414 * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called
1415 * after the page is finished with, and before put_page is called.
1416 *
1417 * get_user_pages is typically used for fewer-copy IO operations, to get a
1418 * handle on the memory by some means other than accesses via the user virtual
1419 * addresses. The pages may be submitted for DMA to devices or accessed via
1420 * their kernel linear mapping (via the kmap APIs). Care should be taken to
1421 * use the correct cache flushing APIs.
1422 *
1423 * See also get_user_pages_fast, for performance critical applications.
1424 */
Nick Pigginb291f002008-10-18 20:26:44 -07001425int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
Peter Zijlstra9d737772009-06-25 11:58:55 +02001426 unsigned long start, int nr_pages, int write, int force,
Nick Pigginb291f002008-10-18 20:26:44 -07001427 struct page **pages, struct vm_area_struct **vmas)
1428{
Hugh Dickins58fa8792009-09-21 17:03:31 -07001429 int flags = FOLL_TOUCH;
Nick Pigginb291f002008-10-18 20:26:44 -07001430
Hugh Dickins58fa8792009-09-21 17:03:31 -07001431 if (pages)
1432 flags |= FOLL_GET;
Nick Pigginb291f002008-10-18 20:26:44 -07001433 if (write)
Hugh Dickins58fa8792009-09-21 17:03:31 -07001434 flags |= FOLL_WRITE;
Nick Pigginb291f002008-10-18 20:26:44 -07001435 if (force)
Hugh Dickins58fa8792009-09-21 17:03:31 -07001436 flags |= FOLL_FORCE;
Nick Pigginb291f002008-10-18 20:26:44 -07001437
Peter Zijlstra9d737772009-06-25 11:58:55 +02001438 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas);
Nick Pigginb291f002008-10-18 20:26:44 -07001439}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440EXPORT_SYMBOL(get_user_pages);
1441
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07001442/**
1443 * get_dump_page() - pin user page in memory while writing it to core dump
1444 * @addr: user address
1445 *
1446 * Returns struct page pointer of user page pinned for dump,
1447 * to be freed afterwards by page_cache_release() or put_page().
1448 *
1449 * Returns NULL on any kind of failure - a hole must then be inserted into
1450 * the corefile, to preserve alignment with its headers; and also returns
1451 * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found -
1452 * allowing a hole to be left in the corefile to save diskspace.
1453 *
1454 * Called without mmap_sem, but after all other threads have been killed.
1455 */
1456#ifdef CONFIG_ELF_CORE
1457struct page *get_dump_page(unsigned long addr)
1458{
1459 struct vm_area_struct *vma;
1460 struct page *page;
1461
1462 if (__get_user_pages(current, current->mm, addr, 1,
Hugh Dickins58fa8792009-09-21 17:03:31 -07001463 FOLL_FORCE | FOLL_DUMP | FOLL_GET, &page, &vma) < 1)
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07001464 return NULL;
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07001465 flush_cache_page(vma, addr, page_to_pfn(page));
1466 return page;
1467}
1468#endif /* CONFIG_ELF_CORE */
1469
Harvey Harrison920c7a52008-02-04 22:29:26 -08001470pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr,
1471 spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001472{
1473 pgd_t * pgd = pgd_offset(mm, addr);
1474 pud_t * pud = pud_alloc(mm, pgd, addr);
1475 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001476 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001477 if (pmd)
1478 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1479 }
1480 return NULL;
1481}
1482
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001484 * This is the old fallback for page remapping.
1485 *
1486 * For historical reasons, it only allows reserved pages. Only
1487 * old drivers should use this, and they needed to mark their
1488 * pages reserved for the old functions anyway.
1489 */
Nick Piggin423bad62008-04-28 02:13:01 -07001490static int insert_page(struct vm_area_struct *vma, unsigned long addr,
1491 struct page *page, pgprot_t prot)
Linus Torvalds238f58d2005-11-29 13:01:56 -08001492{
Nick Piggin423bad62008-04-28 02:13:01 -07001493 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001494 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001495 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001496 spinlock_t *ptl;
1497
Linus Torvalds238f58d2005-11-29 13:01:56 -08001498 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001499 if (PageAnon(page))
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001500 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001501 retval = -ENOMEM;
1502 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001503 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001504 if (!pte)
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001505 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001506 retval = -EBUSY;
1507 if (!pte_none(*pte))
1508 goto out_unlock;
1509
1510 /* Ok, finally just insert the thing.. */
1511 get_page(page);
1512 inc_mm_counter(mm, file_rss);
1513 page_add_file_rmap(page);
1514 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1515
1516 retval = 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001517 pte_unmap_unlock(pte, ptl);
1518 return retval;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001519out_unlock:
1520 pte_unmap_unlock(pte, ptl);
1521out:
1522 return retval;
1523}
1524
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001525/**
1526 * vm_insert_page - insert single page into user vma
1527 * @vma: user vma to map to
1528 * @addr: target user address of this page
1529 * @page: source kernel page
1530 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001531 * This allows drivers to insert individual pages they've allocated
1532 * into a user vma.
1533 *
1534 * The page has to be a nice clean _individual_ kernel allocation.
1535 * If you allocate a compound page, you need to have marked it as
1536 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001537 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001538 *
1539 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1540 * took an arbitrary page protection parameter. This doesn't allow
1541 * that. Your vma protection will have to be set up correctly, which
1542 * means that if you want a shared writable mapping, you'd better
1543 * ask for a shared writable mapping!
1544 *
1545 * The page does not need to be reserved.
1546 */
Nick Piggin423bad62008-04-28 02:13:01 -07001547int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
1548 struct page *page)
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001549{
1550 if (addr < vma->vm_start || addr >= vma->vm_end)
1551 return -EFAULT;
1552 if (!page_count(page))
1553 return -EINVAL;
Linus Torvalds4d7672b2005-12-16 10:21:23 -08001554 vma->vm_flags |= VM_INSERTPAGE;
Nick Piggin423bad62008-04-28 02:13:01 -07001555 return insert_page(vma, addr, page, vma->vm_page_prot);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001556}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001557EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001558
Nick Piggin423bad62008-04-28 02:13:01 -07001559static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1560 unsigned long pfn, pgprot_t prot)
1561{
1562 struct mm_struct *mm = vma->vm_mm;
1563 int retval;
1564 pte_t *pte, entry;
1565 spinlock_t *ptl;
1566
1567 retval = -ENOMEM;
1568 pte = get_locked_pte(mm, addr, &ptl);
1569 if (!pte)
1570 goto out;
1571 retval = -EBUSY;
1572 if (!pte_none(*pte))
1573 goto out_unlock;
1574
1575 /* Ok, finally just insert the thing.. */
1576 entry = pte_mkspecial(pfn_pte(pfn, prot));
1577 set_pte_at(mm, addr, pte, entry);
1578 update_mmu_cache(vma, addr, entry); /* XXX: why not for insert_page? */
1579
1580 retval = 0;
1581out_unlock:
1582 pte_unmap_unlock(pte, ptl);
1583out:
1584 return retval;
1585}
1586
Nick Piggine0dc0d82007-02-12 00:51:36 -08001587/**
1588 * vm_insert_pfn - insert single pfn into user vma
1589 * @vma: user vma to map to
1590 * @addr: target user address of this page
1591 * @pfn: source kernel pfn
1592 *
1593 * Similar to vm_inert_page, this allows drivers to insert individual pages
1594 * they've allocated into a user vma. Same comments apply.
1595 *
1596 * This function should only be called from a vm_ops->fault handler, and
1597 * in that case the handler should return NULL.
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001598 *
1599 * vma cannot be a COW mapping.
1600 *
1601 * As this is called only for pages that do not currently exist, we
1602 * do not need to flush old virtual caches or the TLB.
Nick Piggine0dc0d82007-02-12 00:51:36 -08001603 */
1604int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
Nick Piggin423bad62008-04-28 02:13:01 -07001605 unsigned long pfn)
Nick Piggine0dc0d82007-02-12 00:51:36 -08001606{
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001607 int ret;
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001608 pgprot_t pgprot = vma->vm_page_prot;
Nick Piggin7e675132008-04-28 02:13:00 -07001609 /*
1610 * Technically, architectures with pte_special can avoid all these
1611 * restrictions (same for remap_pfn_range). However we would like
1612 * consistency in testing and feature parity among all, so we should
1613 * try to keep these invariants in place for everybody.
1614 */
Jared Hulbertb379d792008-04-28 02:12:58 -07001615 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
1616 BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
1617 (VM_PFNMAP|VM_MIXEDMAP));
1618 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1619 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
Nick Piggine0dc0d82007-02-12 00:51:36 -08001620
Nick Piggin423bad62008-04-28 02:13:01 -07001621 if (addr < vma->vm_start || addr >= vma->vm_end)
1622 return -EFAULT;
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001623 if (track_pfn_vma_new(vma, &pgprot, pfn, PAGE_SIZE))
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001624 return -EINVAL;
1625
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001626 ret = insert_pfn(vma, addr, pfn, pgprot);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001627
1628 if (ret)
1629 untrack_pfn_vma(vma, pfn, PAGE_SIZE);
1630
1631 return ret;
Nick Piggine0dc0d82007-02-12 00:51:36 -08001632}
1633EXPORT_SYMBOL(vm_insert_pfn);
1634
Nick Piggin423bad62008-04-28 02:13:01 -07001635int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
1636 unsigned long pfn)
1637{
1638 BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
1639
1640 if (addr < vma->vm_start || addr >= vma->vm_end)
1641 return -EFAULT;
1642
1643 /*
1644 * If we don't have pte special, then we have to use the pfn_valid()
1645 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1646 * refcount the page if pfn_valid is true (hence insert_page rather
Hugh Dickins62eede62009-09-21 17:03:34 -07001647 * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP
1648 * without pte special, it would there be refcounted as a normal page.
Nick Piggin423bad62008-04-28 02:13:01 -07001649 */
1650 if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) {
1651 struct page *page;
1652
1653 page = pfn_to_page(pfn);
1654 return insert_page(vma, addr, page, vma->vm_page_prot);
1655 }
1656 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
1657}
1658EXPORT_SYMBOL(vm_insert_mixed);
1659
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001660/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 * maps a range of physical memory into the requested pages. the old
1662 * mappings are removed. any references to nonexistent pages results
1663 * in null mappings (currently treated as "copy-on-access")
1664 */
1665static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1666 unsigned long addr, unsigned long end,
1667 unsigned long pfn, pgprot_t prot)
1668{
1669 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001670 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671
Hugh Dickinsc74df322005-10-29 18:16:23 -07001672 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 if (!pte)
1674 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001675 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 do {
1677 BUG_ON(!pte_none(*pte));
Nick Piggin7e675132008-04-28 02:13:00 -07001678 set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 pfn++;
1680 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001681 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001682 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 return 0;
1684}
1685
1686static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1687 unsigned long addr, unsigned long end,
1688 unsigned long pfn, pgprot_t prot)
1689{
1690 pmd_t *pmd;
1691 unsigned long next;
1692
1693 pfn -= addr >> PAGE_SHIFT;
1694 pmd = pmd_alloc(mm, pud, addr);
1695 if (!pmd)
1696 return -ENOMEM;
1697 do {
1698 next = pmd_addr_end(addr, end);
1699 if (remap_pte_range(mm, pmd, addr, next,
1700 pfn + (addr >> PAGE_SHIFT), prot))
1701 return -ENOMEM;
1702 } while (pmd++, addr = next, addr != end);
1703 return 0;
1704}
1705
1706static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1707 unsigned long addr, unsigned long end,
1708 unsigned long pfn, pgprot_t prot)
1709{
1710 pud_t *pud;
1711 unsigned long next;
1712
1713 pfn -= addr >> PAGE_SHIFT;
1714 pud = pud_alloc(mm, pgd, addr);
1715 if (!pud)
1716 return -ENOMEM;
1717 do {
1718 next = pud_addr_end(addr, end);
1719 if (remap_pmd_range(mm, pud, addr, next,
1720 pfn + (addr >> PAGE_SHIFT), prot))
1721 return -ENOMEM;
1722 } while (pud++, addr = next, addr != end);
1723 return 0;
1724}
1725
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001726/**
1727 * remap_pfn_range - remap kernel memory to userspace
1728 * @vma: user vma to map to
1729 * @addr: target user address to start at
1730 * @pfn: physical address of kernel memory
1731 * @size: size of map area
1732 * @prot: page protection flags for this mapping
1733 *
1734 * Note: this is only safe if the mm semaphore is held when called.
1735 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1737 unsigned long pfn, unsigned long size, pgprot_t prot)
1738{
1739 pgd_t *pgd;
1740 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001741 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 struct mm_struct *mm = vma->vm_mm;
1743 int err;
1744
1745 /*
1746 * Physically remapped pages are special. Tell the
1747 * rest of the world about it:
1748 * VM_IO tells people not to look at these pages
1749 * (accesses can have side effects).
Hugh Dickins0b14c172005-11-21 21:32:15 -08001750 * VM_RESERVED is specified all over the place, because
1751 * in 2.4 it kept swapout's vma scan off this vma; but
1752 * in 2.6 the LRU scan won't even find its pages, so this
1753 * flag means no more than count its pages in reserved_vm,
1754 * and omit it from core dump, even when VM_IO turned off.
Linus Torvalds6aab3412005-11-28 14:34:23 -08001755 * VM_PFNMAP tells the core MM that the base pages are just
1756 * raw PFN mappings, and do not have a "struct page" associated
1757 * with them.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001758 *
1759 * There's a horrible special case to handle copy-on-write
1760 * behaviour that some programs depend on. We mark the "original"
1761 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 */
Pallipadi, Venkatesh4bb9c5c2009-03-12 17:45:27 -07001763 if (addr == vma->vm_start && end == vma->vm_end) {
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001764 vma->vm_pgoff = pfn;
Pallipadi, Venkatesh895791d2009-03-13 16:35:44 -07001765 vma->vm_flags |= VM_PFN_AT_MMAP;
Pallipadi, Venkatesh4bb9c5c2009-03-12 17:45:27 -07001766 } else if (is_cow_mapping(vma->vm_flags))
venkatesh.pallipadi@intel.com3c8bb732008-12-18 11:41:27 -08001767 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001768
Linus Torvalds6aab3412005-11-28 14:34:23 -08001769 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001771 err = track_pfn_vma_new(vma, &prot, pfn, PAGE_ALIGN(size));
venkatesh.pallipadi@intel.coma3670612009-01-09 16:13:09 -08001772 if (err) {
1773 /*
1774 * To indicate that track_pfn related cleanup is not
1775 * needed from higher level routine calling unmap_vmas
1776 */
1777 vma->vm_flags &= ~(VM_IO | VM_RESERVED | VM_PFNMAP);
Pallipadi, Venkatesh895791d2009-03-13 16:35:44 -07001778 vma->vm_flags &= ~VM_PFN_AT_MMAP;
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001779 return -EINVAL;
venkatesh.pallipadi@intel.coma3670612009-01-09 16:13:09 -08001780 }
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001781
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 BUG_ON(addr >= end);
1783 pfn -= addr >> PAGE_SHIFT;
1784 pgd = pgd_offset(mm, addr);
1785 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 do {
1787 next = pgd_addr_end(addr, end);
1788 err = remap_pud_range(mm, pgd, addr, next,
1789 pfn + (addr >> PAGE_SHIFT), prot);
1790 if (err)
1791 break;
1792 } while (pgd++, addr = next, addr != end);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001793
1794 if (err)
1795 untrack_pfn_vma(vma, pfn, PAGE_ALIGN(size));
1796
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 return err;
1798}
1799EXPORT_SYMBOL(remap_pfn_range);
1800
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001801static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1802 unsigned long addr, unsigned long end,
1803 pte_fn_t fn, void *data)
1804{
1805 pte_t *pte;
1806 int err;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001807 pgtable_t token;
Borislav Petkov94909912007-05-06 14:49:17 -07001808 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001809
1810 pte = (mm == &init_mm) ?
1811 pte_alloc_kernel(pmd, addr) :
1812 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1813 if (!pte)
1814 return -ENOMEM;
1815
1816 BUG_ON(pmd_huge(*pmd));
1817
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08001818 arch_enter_lazy_mmu_mode();
1819
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001820 token = pmd_pgtable(*pmd);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001821
1822 do {
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001823 err = fn(pte, token, addr, data);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001824 if (err)
1825 break;
1826 } while (pte++, addr += PAGE_SIZE, addr != end);
1827
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08001828 arch_leave_lazy_mmu_mode();
1829
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001830 if (mm != &init_mm)
1831 pte_unmap_unlock(pte-1, ptl);
1832 return err;
1833}
1834
1835static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1836 unsigned long addr, unsigned long end,
1837 pte_fn_t fn, void *data)
1838{
1839 pmd_t *pmd;
1840 unsigned long next;
1841 int err;
1842
Andi Kleenceb86872008-07-23 21:27:50 -07001843 BUG_ON(pud_huge(*pud));
1844
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001845 pmd = pmd_alloc(mm, pud, addr);
1846 if (!pmd)
1847 return -ENOMEM;
1848 do {
1849 next = pmd_addr_end(addr, end);
1850 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1851 if (err)
1852 break;
1853 } while (pmd++, addr = next, addr != end);
1854 return err;
1855}
1856
1857static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1858 unsigned long addr, unsigned long end,
1859 pte_fn_t fn, void *data)
1860{
1861 pud_t *pud;
1862 unsigned long next;
1863 int err;
1864
1865 pud = pud_alloc(mm, pgd, addr);
1866 if (!pud)
1867 return -ENOMEM;
1868 do {
1869 next = pud_addr_end(addr, end);
1870 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1871 if (err)
1872 break;
1873 } while (pud++, addr = next, addr != end);
1874 return err;
1875}
1876
1877/*
1878 * Scan a region of virtual memory, filling in page tables as necessary
1879 * and calling a provided function on each leaf page table.
1880 */
1881int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1882 unsigned long size, pte_fn_t fn, void *data)
1883{
1884 pgd_t *pgd;
1885 unsigned long next;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001886 unsigned long start = addr, end = addr + size;
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001887 int err;
1888
1889 BUG_ON(addr >= end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001890 mmu_notifier_invalidate_range_start(mm, start, end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001891 pgd = pgd_offset(mm, addr);
1892 do {
1893 next = pgd_addr_end(addr, end);
1894 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1895 if (err)
1896 break;
1897 } while (pgd++, addr = next, addr != end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001898 mmu_notifier_invalidate_range_end(mm, start, end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001899 return err;
1900}
1901EXPORT_SYMBOL_GPL(apply_to_page_range);
1902
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001904 * handle_pte_fault chooses page fault handler according to an entry
1905 * which was read non-atomically. Before making any commitment, on
1906 * those architectures or configurations (e.g. i386 with PAE) which
1907 * might give a mix of unmatched parts, do_swap_page and do_file_page
1908 * must check under lock before unmapping the pte and proceeding
1909 * (but do_wp_page is only called after already making such a check;
1910 * and do_anonymous_page and do_no_page can safely check later on).
1911 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001912static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001913 pte_t *page_table, pte_t orig_pte)
1914{
1915 int same = 1;
1916#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1917 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001918 spinlock_t *ptl = pte_lockptr(mm, pmd);
1919 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001920 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001921 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001922 }
1923#endif
1924 pte_unmap(page_table);
1925 return same;
1926}
1927
1928/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1930 * servicing faults for write access. In the normal case, do always want
1931 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1932 * that do not have writing enabled, when used by access_process_vm.
1933 */
1934static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1935{
1936 if (likely(vma->vm_flags & VM_WRITE))
1937 pte = pte_mkwrite(pte);
1938 return pte;
1939}
1940
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001941static 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 -08001942{
1943 /*
1944 * If the source page was a PFN mapping, we don't have
1945 * a "struct page" for it. We do a best-effort copy by
1946 * just copying from the original user address. If that
1947 * fails, we just zero-fill it. Live with it.
1948 */
1949 if (unlikely(!src)) {
1950 void *kaddr = kmap_atomic(dst, KM_USER0);
Linus Torvalds5d2a2dbbc2005-11-29 14:07:55 -08001951 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1952
1953 /*
1954 * This really shouldn't fail, because the page is there
1955 * in the page tables. But it might just be unreadable,
1956 * in which case we just give up and fill the result with
1957 * zeroes.
1958 */
1959 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Linus Torvalds6aab3412005-11-28 14:34:23 -08001960 memset(kaddr, 0, PAGE_SIZE);
1961 kunmap_atomic(kaddr, KM_USER0);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07001962 flush_dcache_page(dst);
Nick Piggin0ed361d2008-02-04 22:29:34 -08001963 } else
1964 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001965}
1966
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 * This routine handles present pages, when users try to write
1969 * to a shared page. It is done by copying the page to a new address
1970 * and decrementing the shared-page counter for the old page.
1971 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 * Note that this routine assumes that the protection checks have been
1973 * done by the caller (the low-level page fault routine in most cases).
1974 * Thus we can safely just mark it writable once we've done any necessary
1975 * COW.
1976 *
1977 * We also mark the page dirty at this point even though the page will
1978 * change only once the write actually happens. This avoids a few races,
1979 * and potentially makes it more efficient.
1980 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001981 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1982 * but allow concurrent faults), with pte both mapped and locked.
1983 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001985static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1986 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001987 spinlock_t *ptl, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988{
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001989 struct page *old_page, *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 pte_t entry;
Nick Piggin83c54072007-07-19 01:47:05 -07001991 int reuse = 0, ret = 0;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001992 int page_mkwrite = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001993 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994
Linus Torvalds6aab3412005-11-28 14:34:23 -08001995 old_page = vm_normal_page(vma, address, orig_pte);
Peter Zijlstra251b97f2008-07-04 09:59:24 -07001996 if (!old_page) {
1997 /*
1998 * VM_MIXEDMAP !pfn_valid() case
1999 *
2000 * We should not cow pages in a shared writeable mapping.
2001 * Just mark the pages writable as we can't do any dirty
2002 * accounting on raw pfn maps.
2003 */
2004 if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
2005 (VM_WRITE|VM_SHARED))
2006 goto reuse;
Linus Torvalds6aab3412005-11-28 14:34:23 -08002007 goto gotten;
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002008 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002010 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002011 * Take out anonymous pages first, anonymous shared vmas are
2012 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002013 */
Hugh Dickins9a840892009-09-21 17:02:01 -07002014 if (PageAnon(old_page) && !PageKsm(old_page)) {
Hugh Dickinsab967d82009-01-06 14:39:33 -08002015 if (!trylock_page(old_page)) {
2016 page_cache_get(old_page);
2017 pte_unmap_unlock(page_table, ptl);
2018 lock_page(old_page);
2019 page_table = pte_offset_map_lock(mm, pmd, address,
2020 &ptl);
2021 if (!pte_same(*page_table, orig_pte)) {
2022 unlock_page(old_page);
2023 page_cache_release(old_page);
2024 goto unlock;
2025 }
2026 page_cache_release(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002027 }
Hugh Dickins7b1fe592009-01-06 14:39:34 -08002028 reuse = reuse_swap_page(old_page);
Hugh Dickinsab967d82009-01-06 14:39:33 -08002029 unlock_page(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002030 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002031 (VM_WRITE|VM_SHARED))) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07002032 /*
2033 * Only catch write-faults on shared writable pages,
2034 * read-only shared pages can get COWed by
2035 * get_user_pages(.write=1, .force=1).
2036 */
David Howells9637a5e2006-06-23 02:03:43 -07002037 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
Nick Pigginc2ec1752009-03-31 15:23:21 -07002038 struct vm_fault vmf;
2039 int tmp;
2040
2041 vmf.virtual_address = (void __user *)(address &
2042 PAGE_MASK);
2043 vmf.pgoff = old_page->index;
2044 vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
2045 vmf.page = old_page;
2046
David Howells9637a5e2006-06-23 02:03:43 -07002047 /*
2048 * Notify the address space that the page is about to
2049 * become writable so that it can prohibit this or wait
2050 * for the page to get into an appropriate state.
2051 *
2052 * We do this without the lock held, so that it can
2053 * sleep if it needs to.
2054 */
2055 page_cache_get(old_page);
2056 pte_unmap_unlock(page_table, ptl);
2057
Nick Pigginc2ec1752009-03-31 15:23:21 -07002058 tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
2059 if (unlikely(tmp &
2060 (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
2061 ret = tmp;
David Howells9637a5e2006-06-23 02:03:43 -07002062 goto unwritable_page;
Nick Pigginc2ec1752009-03-31 15:23:21 -07002063 }
Nick Pigginb827e492009-04-30 15:08:16 -07002064 if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
2065 lock_page(old_page);
2066 if (!old_page->mapping) {
2067 ret = 0; /* retry the fault */
2068 unlock_page(old_page);
2069 goto unwritable_page;
2070 }
2071 } else
2072 VM_BUG_ON(!PageLocked(old_page));
David Howells9637a5e2006-06-23 02:03:43 -07002073
David Howells9637a5e2006-06-23 02:03:43 -07002074 /*
2075 * Since we dropped the lock we need to revalidate
2076 * the PTE as someone else may have changed it. If
2077 * they did, we just return, as we can count on the
2078 * MMU to tell us if they didn't also make it writable.
2079 */
2080 page_table = pte_offset_map_lock(mm, pmd, address,
2081 &ptl);
Nick Pigginb827e492009-04-30 15:08:16 -07002082 if (!pte_same(*page_table, orig_pte)) {
2083 unlock_page(old_page);
2084 page_cache_release(old_page);
David Howells9637a5e2006-06-23 02:03:43 -07002085 goto unlock;
Nick Pigginb827e492009-04-30 15:08:16 -07002086 }
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002087
2088 page_mkwrite = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002090 dirty_page = old_page;
2091 get_page(dirty_page);
David Howells9637a5e2006-06-23 02:03:43 -07002092 reuse = 1;
David Howells9637a5e2006-06-23 02:03:43 -07002093 }
2094
2095 if (reuse) {
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002096reuse:
David Howells9637a5e2006-06-23 02:03:43 -07002097 flush_cache_page(vma, address, pte_pfn(orig_pte));
2098 entry = pte_mkyoung(orig_pte);
2099 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
KAMEZAWA Hiroyuki954ffcb2007-10-16 01:25:44 -07002100 if (ptep_set_access_flags(vma, address, page_table, entry,1))
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002101 update_mmu_cache(vma, address, entry);
David Howells9637a5e2006-06-23 02:03:43 -07002102 ret |= VM_FAULT_WRITE;
2103 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105
2106 /*
2107 * Ok, we need to copy. Oh, well..
2108 */
Nick Pigginb5810032005-10-29 18:16:12 -07002109 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08002110gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002111 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112
2113 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07002114 goto oom;
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002115
Hugh Dickins62eede62009-09-21 17:03:34 -07002116 if (is_zero_pfn(pte_pfn(orig_pte))) {
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002117 new_page = alloc_zeroed_user_highpage_movable(vma, address);
2118 if (!new_page)
2119 goto oom;
2120 } else {
2121 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
2122 if (!new_page)
2123 goto oom;
2124 cow_user_page(new_page, old_page, address, vma);
2125 }
2126 __SetPageUptodate(new_page);
2127
Nick Pigginb291f002008-10-18 20:26:44 -07002128 /*
2129 * Don't let another task, with possibly unlocked vma,
2130 * keep the mlocked page.
2131 */
Carsten Otteab926612009-02-04 15:12:16 -08002132 if ((vma->vm_flags & VM_LOCKED) && old_page) {
Nick Pigginb291f002008-10-18 20:26:44 -07002133 lock_page(old_page); /* for LRU manipulation */
2134 clear_page_mlock(old_page);
2135 unlock_page(old_page);
2136 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002137
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002138 if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002139 goto oom_free_new;
2140
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 /*
2142 * Re-check the pte - we dropped the lock
2143 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002144 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07002145 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08002146 if (old_page) {
Hugh Dickins920fc352005-11-21 21:32:17 -08002147 if (!PageAnon(old_page)) {
2148 dec_mm_counter(mm, file_rss);
2149 inc_mm_counter(mm, anon_rss);
2150 }
2151 } else
Hugh Dickins42946212005-10-29 18:16:05 -07002152 inc_mm_counter(mm, anon_rss);
Ben Collinseca35132005-11-29 11:45:26 -08002153 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07002154 entry = mk_pte(new_page, vma->vm_page_prot);
2155 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Siddha, Suresh B4ce072f2006-09-29 01:58:42 -07002156 /*
2157 * Clear the pte entry and flush it first, before updating the
2158 * pte with the new entry. This will avoid a race condition
2159 * seen in the presence of one thread doing SMC and another
2160 * thread doing COW.
2161 */
Izik Eidus828502d2009-09-21 17:01:51 -07002162 ptep_clear_flush(vma, address, page_table);
Nick Piggin9617d952006-01-06 00:11:12 -08002163 page_add_new_anon_rmap(new_page, vma, address);
Izik Eidus828502d2009-09-21 17:01:51 -07002164 /*
2165 * We call the notify macro here because, when using secondary
2166 * mmu page tables (such as kvm shadow page tables), we want the
2167 * new page to be mapped directly into the secondary page table.
2168 */
2169 set_pte_at_notify(mm, address, page_table, entry);
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002170 update_mmu_cache(vma, address, entry);
Nick Piggin945754a2008-06-23 14:30:30 +02002171 if (old_page) {
2172 /*
2173 * Only after switching the pte to the new page may
2174 * we remove the mapcount here. Otherwise another
2175 * process may come and find the rmap count decremented
2176 * before the pte is switched to the new page, and
2177 * "reuse" the old page writing into it while our pte
2178 * here still points into it and can be read by other
2179 * threads.
2180 *
2181 * The critical issue is to order this
2182 * page_remove_rmap with the ptp_clear_flush above.
2183 * Those stores are ordered by (if nothing else,)
2184 * the barrier present in the atomic_add_negative
2185 * in page_remove_rmap.
2186 *
2187 * Then the TLB flush in ptep_clear_flush ensures that
2188 * no process can access the old page before the
2189 * decremented mapcount is visible. And the old page
2190 * cannot be reused until after the decremented
2191 * mapcount is visible. So transitively, TLBs to
2192 * old page will be flushed before it can be reused.
2193 */
Hugh Dickinsedc315f2009-01-06 14:40:11 -08002194 page_remove_rmap(old_page);
Nick Piggin945754a2008-06-23 14:30:30 +02002195 }
2196
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 /* Free the old page.. */
2198 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10002199 ret |= VM_FAULT_WRITE;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002200 } else
2201 mem_cgroup_uncharge_page(new_page);
2202
Hugh Dickins920fc352005-11-21 21:32:17 -08002203 if (new_page)
2204 page_cache_release(new_page);
2205 if (old_page)
2206 page_cache_release(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002207unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002208 pte_unmap_unlock(page_table, ptl);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002209 if (dirty_page) {
Nick Piggin79352892007-07-19 01:47:22 -07002210 /*
2211 * Yes, Virginia, this is actually required to prevent a race
2212 * with clear_page_dirty_for_io() from clearing the page dirty
2213 * bit after it clear all dirty ptes, but before a racing
2214 * do_wp_page installs a dirty pte.
2215 *
2216 * do_no_page is protected similarly.
2217 */
Nick Pigginb827e492009-04-30 15:08:16 -07002218 if (!page_mkwrite) {
2219 wait_on_page_locked(dirty_page);
2220 set_page_dirty_balance(dirty_page, page_mkwrite);
2221 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002222 put_page(dirty_page);
Nick Pigginb827e492009-04-30 15:08:16 -07002223 if (page_mkwrite) {
2224 struct address_space *mapping = dirty_page->mapping;
2225
2226 set_page_dirty(dirty_page);
2227 unlock_page(dirty_page);
2228 page_cache_release(dirty_page);
2229 if (mapping) {
2230 /*
2231 * Some device drivers do not set page.mapping
2232 * but still dirty their pages
2233 */
2234 balance_dirty_pages_ratelimited(mapping);
2235 }
2236 }
2237
2238 /* file_update_time outside page_lock */
2239 if (vma->vm_file)
2240 file_update_time(vma->vm_file);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002241 }
Nick Pigginf33ea7f2005-08-03 20:24:01 +10002242 return ret;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002243oom_free_new:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002244 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002245oom:
Nick Pigginb827e492009-04-30 15:08:16 -07002246 if (old_page) {
2247 if (page_mkwrite) {
2248 unlock_page(old_page);
2249 page_cache_release(old_page);
2250 }
Hugh Dickins920fc352005-11-21 21:32:17 -08002251 page_cache_release(old_page);
Nick Pigginb827e492009-04-30 15:08:16 -07002252 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 return VM_FAULT_OOM;
David Howells9637a5e2006-06-23 02:03:43 -07002254
2255unwritable_page:
2256 page_cache_release(old_page);
Nick Pigginc2ec1752009-03-31 15:23:21 -07002257 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258}
2259
2260/*
2261 * Helper functions for unmap_mapping_range().
2262 *
2263 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
2264 *
2265 * We have to restart searching the prio_tree whenever we drop the lock,
2266 * since the iterator is only valid while the lock is held, and anyway
2267 * a later vma might be split and reinserted earlier while lock dropped.
2268 *
2269 * The list of nonlinear vmas could be handled more efficiently, using
2270 * a placeholder, but handle it in the same way until a need is shown.
2271 * It is important to search the prio_tree before nonlinear list: a vma
2272 * may become nonlinear and be shifted from prio_tree to nonlinear list
2273 * while the lock is dropped; but never shifted from list to prio_tree.
2274 *
2275 * In order to make forward progress despite restarting the search,
2276 * vm_truncate_count is used to mark a vma as now dealt with, so we can
2277 * quickly skip it next time around. Since the prio_tree search only
2278 * shows us those vmas affected by unmapping the range in question, we
2279 * can't efficiently keep all vmas in step with mapping->truncate_count:
2280 * so instead reset them all whenever it wraps back to 0 (then go to 1).
2281 * mapping->truncate_count and vma->vm_truncate_count are protected by
2282 * i_mmap_lock.
2283 *
2284 * In order to make forward progress despite repeatedly restarting some
Hugh Dickinsee39b372005-04-19 13:29:15 -07002285 * large vma, note the restart_addr from unmap_vmas when it breaks out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 * and restart from that address when we reach that vma again. It might
2287 * have been split or merged, shrunk or extended, but never shifted: so
2288 * restart_addr remains valid so long as it remains in the vma's range.
2289 * unmap_mapping_range forces truncate_count to leap over page-aligned
2290 * values so we can save vma's restart_addr in its truncate_count field.
2291 */
2292#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
2293
2294static void reset_vma_truncate_counts(struct address_space *mapping)
2295{
2296 struct vm_area_struct *vma;
2297 struct prio_tree_iter iter;
2298
2299 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
2300 vma->vm_truncate_count = 0;
2301 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
2302 vma->vm_truncate_count = 0;
2303}
2304
2305static int unmap_mapping_range_vma(struct vm_area_struct *vma,
2306 unsigned long start_addr, unsigned long end_addr,
2307 struct zap_details *details)
2308{
2309 unsigned long restart_addr;
2310 int need_break;
2311
Nick Piggind00806b2007-07-19 01:46:57 -07002312 /*
2313 * files that support invalidating or truncating portions of the
Nick Piggind0217ac2007-07-19 01:47:03 -07002314 * file from under mmaped areas must have their ->fault function
Nick Piggin83c54072007-07-19 01:47:05 -07002315 * return a locked page (and set VM_FAULT_LOCKED in the return).
2316 * This provides synchronisation against concurrent unmapping here.
Nick Piggind00806b2007-07-19 01:46:57 -07002317 */
Nick Piggind00806b2007-07-19 01:46:57 -07002318
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319again:
2320 restart_addr = vma->vm_truncate_count;
2321 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
2322 start_addr = restart_addr;
2323 if (start_addr >= end_addr) {
2324 /* Top of vma has been split off since last time */
2325 vma->vm_truncate_count = details->truncate_count;
2326 return 0;
2327 }
2328 }
2329
Hugh Dickinsee39b372005-04-19 13:29:15 -07002330 restart_addr = zap_page_range(vma, start_addr,
2331 end_addr - start_addr, details);
Nick Piggin95c354f2008-01-30 13:31:20 +01002332 need_break = need_resched() || spin_needbreak(details->i_mmap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333
Hugh Dickinsee39b372005-04-19 13:29:15 -07002334 if (restart_addr >= end_addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 /* We have now completed this vma: mark it so */
2336 vma->vm_truncate_count = details->truncate_count;
2337 if (!need_break)
2338 return 0;
2339 } else {
2340 /* Note restart_addr in vma's truncate_count field */
Hugh Dickinsee39b372005-04-19 13:29:15 -07002341 vma->vm_truncate_count = restart_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 if (!need_break)
2343 goto again;
2344 }
2345
2346 spin_unlock(details->i_mmap_lock);
2347 cond_resched();
2348 spin_lock(details->i_mmap_lock);
2349 return -EINTR;
2350}
2351
2352static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
2353 struct zap_details *details)
2354{
2355 struct vm_area_struct *vma;
2356 struct prio_tree_iter iter;
2357 pgoff_t vba, vea, zba, zea;
2358
2359restart:
2360 vma_prio_tree_foreach(vma, &iter, root,
2361 details->first_index, details->last_index) {
2362 /* Skip quickly over those we have already dealt with */
2363 if (vma->vm_truncate_count == details->truncate_count)
2364 continue;
2365
2366 vba = vma->vm_pgoff;
2367 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
2368 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
2369 zba = details->first_index;
2370 if (zba < vba)
2371 zba = vba;
2372 zea = details->last_index;
2373 if (zea > vea)
2374 zea = vea;
2375
2376 if (unmap_mapping_range_vma(vma,
2377 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
2378 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
2379 details) < 0)
2380 goto restart;
2381 }
2382}
2383
2384static inline void unmap_mapping_range_list(struct list_head *head,
2385 struct zap_details *details)
2386{
2387 struct vm_area_struct *vma;
2388
2389 /*
2390 * In nonlinear VMAs there is no correspondence between virtual address
2391 * offset and file offset. So we must perform an exhaustive search
2392 * across *all* the pages in each nonlinear VMA, not just the pages
2393 * whose virtual address lies outside the file truncation point.
2394 */
2395restart:
2396 list_for_each_entry(vma, head, shared.vm_set.list) {
2397 /* Skip quickly over those we have already dealt with */
2398 if (vma->vm_truncate_count == details->truncate_count)
2399 continue;
2400 details->nonlinear_vma = vma;
2401 if (unmap_mapping_range_vma(vma, vma->vm_start,
2402 vma->vm_end, details) < 0)
2403 goto restart;
2404 }
2405}
2406
2407/**
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002408 * 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 -07002409 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 * @holebegin: byte in first page to unmap, relative to the start of
2411 * the underlying file. This will be rounded down to a PAGE_SIZE
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +10002412 * boundary. Note that this is different from truncate_pagecache(), which
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 * must keep the partial page. In contrast, we must get rid of
2414 * partial pages.
2415 * @holelen: size of prospective hole in bytes. This will be rounded
2416 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2417 * end of the file.
2418 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2419 * but 0 when invalidating pagecache, don't throw away private data.
2420 */
2421void unmap_mapping_range(struct address_space *mapping,
2422 loff_t const holebegin, loff_t const holelen, int even_cows)
2423{
2424 struct zap_details details;
2425 pgoff_t hba = holebegin >> PAGE_SHIFT;
2426 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2427
2428 /* Check for overflow. */
2429 if (sizeof(holelen) > sizeof(hlen)) {
2430 long long holeend =
2431 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2432 if (holeend & ~(long long)ULONG_MAX)
2433 hlen = ULONG_MAX - hba + 1;
2434 }
2435
2436 details.check_mapping = even_cows? NULL: mapping;
2437 details.nonlinear_vma = NULL;
2438 details.first_index = hba;
2439 details.last_index = hba + hlen - 1;
2440 if (details.last_index < details.first_index)
2441 details.last_index = ULONG_MAX;
2442 details.i_mmap_lock = &mapping->i_mmap_lock;
2443
2444 spin_lock(&mapping->i_mmap_lock);
2445
Nick Piggind00806b2007-07-19 01:46:57 -07002446 /* Protect against endless unmapping loops */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447 mapping->truncate_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448 if (unlikely(is_restart_addr(mapping->truncate_count))) {
2449 if (mapping->truncate_count == 0)
2450 reset_vma_truncate_counts(mapping);
2451 mapping->truncate_count++;
2452 }
2453 details.truncate_count = mapping->truncate_count;
2454
2455 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
2456 unmap_mapping_range_tree(&mapping->i_mmap, &details);
2457 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
2458 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
2459 spin_unlock(&mapping->i_mmap_lock);
2460}
2461EXPORT_SYMBOL(unmap_mapping_range);
2462
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002463int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
2464{
2465 struct address_space *mapping = inode->i_mapping;
2466
2467 /*
2468 * If the underlying filesystem is not going to provide
2469 * a way to truncate a range of blocks (punch a hole) -
2470 * we should return failure right now.
2471 */
Al Viroacfa4382008-12-04 10:06:33 -05002472 if (!inode->i_op->truncate_range)
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002473 return -ENOSYS;
2474
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002475 mutex_lock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002476 down_write(&inode->i_alloc_sem);
2477 unmap_mapping_range(mapping, offset, (end - offset), 1);
2478 truncate_inode_pages_range(mapping, offset, end);
Nick Piggind00806b2007-07-19 01:46:57 -07002479 unmap_mapping_range(mapping, offset, (end - offset), 1);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002480 inode->i_op->truncate_range(inode, offset, end);
2481 up_write(&inode->i_alloc_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002482 mutex_unlock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002483
2484 return 0;
2485}
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002486
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002488 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2489 * but allow concurrent faults), and pte mapped but not yet locked.
2490 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002492static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2493 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002494 unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002496 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 struct page *page;
Hugh Dickins65500d22005-10-29 18:15:59 -07002498 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499 pte_t pte;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002500 struct mem_cgroup *ptr = NULL;
Nick Piggin83c54072007-07-19 01:47:05 -07002501 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002503 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002504 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002505
2506 entry = pte_to_swp_entry(orig_pte);
Andi Kleend1737fd2009-09-16 11:50:06 +02002507 if (unlikely(non_swap_entry(entry))) {
2508 if (is_migration_entry(entry)) {
2509 migration_entry_wait(mm, pmd, address);
2510 } else if (is_hwpoison_entry(entry)) {
2511 ret = VM_FAULT_HWPOISON;
2512 } else {
2513 print_bad_pte(vma, address, orig_pte, NULL);
2514 ret = VM_FAULT_OOM;
2515 }
Christoph Lameter06972122006-06-23 02:03:35 -07002516 goto out;
2517 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002518 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519 page = lookup_swap_cache(entry);
2520 if (!page) {
Hugh Dickinsa5c9b692009-06-23 12:36:58 -07002521 grab_swap_token(mm); /* Contend for token _before_ read-in */
Hugh Dickins02098fe2008-02-04 22:28:42 -08002522 page = swapin_readahead(entry,
2523 GFP_HIGHUSER_MOVABLE, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 if (!page) {
2525 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002526 * Back out if somebody else faulted in this pte
2527 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002529 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 if (likely(pte_same(*page_table, orig_pte)))
2531 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002532 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002533 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 }
2535
2536 /* Had to read the page from swap area: Major fault */
2537 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002538 count_vm_event(PGMAJFAULT);
Andi Kleend1737fd2009-09-16 11:50:06 +02002539 } else if (PageHWPoison(page)) {
2540 ret = VM_FAULT_HWPOISON;
2541 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2542 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 }
2544
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 lock_page(page);
Balbir Singh20a10222007-11-14 17:00:33 -08002546 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002548 if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) {
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002549 ret = VM_FAULT_OOM;
Johannes Weinerbc43f752009-04-30 15:08:08 -07002550 goto out_page;
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002551 }
2552
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002554 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002556 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002557 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002558 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002559
2560 if (unlikely(!PageUptodate(page))) {
2561 ret = VM_FAULT_SIGBUS;
2562 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 }
2564
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002565 /*
2566 * The page isn't present yet, go ahead with the fault.
2567 *
2568 * Be careful about the sequence of operations here.
2569 * To get its accounting right, reuse_swap_page() must be called
2570 * while the page is counted on swap but not yet in mapcount i.e.
2571 * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
2572 * must be called after the swap_free(), or it will never succeed.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002573 * Because delete_from_swap_page() may be called by reuse_swap_page(),
2574 * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry
2575 * in page->private. In this case, a record in swap_cgroup is silently
2576 * discarded at swap_free().
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002577 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578
Hugh Dickins42946212005-10-29 18:16:05 -07002579 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580 pte = mk_pte(page, vma->vm_page_prot);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002581 if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002583 flags &= ~FAULT_FLAG_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 flush_icache_page(vma, page);
2586 set_pte_at(mm, address, page_table, pte);
2587 page_add_anon_rmap(page, vma, address);
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002588 /* It's better to call commit-charge after rmap is established */
2589 mem_cgroup_commit_charge_swapin(page, ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002591 swap_free(entry);
Nick Pigginb291f002008-10-18 20:26:44 -07002592 if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08002593 try_to_free_swap(page);
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002594 unlock_page(page);
2595
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002596 if (flags & FAULT_FLAG_WRITE) {
Hugh Dickins61469f12008-03-04 14:29:04 -08002597 ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
2598 if (ret & VM_FAULT_ERROR)
2599 ret &= VM_FAULT_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 goto out;
2601 }
2602
2603 /* No need to invalidate - it was non-present before */
2604 update_mmu_cache(vma, address, pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07002605unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002606 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607out:
2608 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002609out_nomap:
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002610 mem_cgroup_cancel_charge_swapin(ptr);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002611 pte_unmap_unlock(page_table, ptl);
Johannes Weinerbc43f752009-04-30 15:08:08 -07002612out_page:
Kirill Korotaevb8107482005-05-16 21:53:50 -07002613 unlock_page(page);
2614 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002615 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616}
2617
2618/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002619 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2620 * but allow concurrent faults), and pte mapped but not yet locked.
2621 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002623static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2624 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002625 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002627 struct page *page;
2628 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630
Hugh Dickins62eede62009-09-21 17:03:34 -07002631 if (!(flags & FAULT_FLAG_WRITE)) {
2632 entry = pte_mkspecial(pfn_pte(my_zero_pfn(address),
2633 vma->vm_page_prot));
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002634 ptl = pte_lockptr(mm, pmd);
2635 spin_lock(ptl);
2636 if (!pte_none(*page_table))
2637 goto unlock;
2638 goto setpte;
2639 }
2640
Nick Piggin557ed1f2007-10-16 01:24:40 -07002641 /* Allocate our own private page. */
2642 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643
Nick Piggin557ed1f2007-10-16 01:24:40 -07002644 if (unlikely(anon_vma_prepare(vma)))
2645 goto oom;
2646 page = alloc_zeroed_user_highpage_movable(vma, address);
2647 if (!page)
2648 goto oom;
Nick Piggin0ed361d2008-02-04 22:29:34 -08002649 __SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002651 if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002652 goto oom_free_page;
2653
Nick Piggin557ed1f2007-10-16 01:24:40 -07002654 entry = mk_pte(page, vma->vm_page_prot);
Hugh Dickins1ac0cb52009-09-21 17:03:29 -07002655 if (vma->vm_flags & VM_WRITE)
2656 entry = pte_mkwrite(pte_mkdirty(entry));
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002657
Nick Piggin557ed1f2007-10-16 01:24:40 -07002658 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Andrea Arcangeli1c2fb7a2009-09-21 17:02:22 -07002659 if (!pte_none(*page_table))
Nick Piggin557ed1f2007-10-16 01:24:40 -07002660 goto release;
Hugh Dickins9ba69292009-09-21 17:02:20 -07002661
Nick Piggin557ed1f2007-10-16 01:24:40 -07002662 inc_mm_counter(mm, anon_rss);
Nick Piggin557ed1f2007-10-16 01:24:40 -07002663 page_add_new_anon_rmap(page, vma, address);
Hugh Dickinsa13ea5b2009-09-21 17:03:30 -07002664setpte:
Hugh Dickins65500d22005-10-29 18:15:59 -07002665 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666
2667 /* No need to invalidate - it was non-present before */
Hugh Dickins65500d22005-10-29 18:15:59 -07002668 update_mmu_cache(vma, address, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002669unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002670 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002671 return 0;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002672release:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002673 mem_cgroup_uncharge_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002674 page_cache_release(page);
2675 goto unlock;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002676oom_free_page:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002677 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002678oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679 return VM_FAULT_OOM;
2680}
2681
2682/*
Nick Piggin54cb8822007-07-19 01:46:59 -07002683 * __do_fault() tries to create a new page mapping. It aggressively
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684 * tries to share with existing pages, but makes a separate copy if
Nick Piggin54cb8822007-07-19 01:46:59 -07002685 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2686 * the next page fault.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687 *
2688 * As this is called only for pages that do not currently exist, we
2689 * do not need to flush old virtual caches or the TLB.
2690 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002691 * We enter with non-exclusive mmap_sem (to exclude vma changes,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002692 * but allow concurrent faults), and pte neither mapped nor locked.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002693 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694 */
Nick Piggin54cb8822007-07-19 01:46:59 -07002695static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002696 unsigned long address, pmd_t *pmd,
Nick Piggin54cb8822007-07-19 01:46:59 -07002697 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698{
Hugh Dickins16abfa02007-10-04 16:56:06 +01002699 pte_t *page_table;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002700 spinlock_t *ptl;
Nick Piggind0217ac2007-07-19 01:47:03 -07002701 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703 int anon = 0;
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002704 int charged = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002705 struct page *dirty_page = NULL;
Nick Piggind0217ac2007-07-19 01:47:03 -07002706 struct vm_fault vmf;
2707 int ret;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002708 int page_mkwrite = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07002709
Nick Piggind0217ac2007-07-19 01:47:03 -07002710 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2711 vmf.pgoff = pgoff;
2712 vmf.flags = flags;
2713 vmf.page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714
Nick Piggin3c18ddd2008-04-28 02:12:10 -07002715 ret = vma->vm_ops->fault(vma, &vmf);
2716 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2717 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718
Andi Kleena3b947e2009-09-16 11:50:08 +02002719 if (unlikely(PageHWPoison(vmf.page))) {
2720 if (ret & VM_FAULT_LOCKED)
2721 unlock_page(vmf.page);
2722 return VM_FAULT_HWPOISON;
2723 }
2724
Nick Piggind00806b2007-07-19 01:46:57 -07002725 /*
Nick Piggind0217ac2007-07-19 01:47:03 -07002726 * For consistency in subsequent calls, make the faulted page always
Nick Piggind00806b2007-07-19 01:46:57 -07002727 * locked.
2728 */
Nick Piggin83c54072007-07-19 01:47:05 -07002729 if (unlikely(!(ret & VM_FAULT_LOCKED)))
Nick Piggind0217ac2007-07-19 01:47:03 -07002730 lock_page(vmf.page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002731 else
Nick Piggind0217ac2007-07-19 01:47:03 -07002732 VM_BUG_ON(!PageLocked(vmf.page));
Nick Piggind00806b2007-07-19 01:46:57 -07002733
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734 /*
2735 * Should we do an early C-O-W break?
2736 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002737 page = vmf.page;
Nick Piggin54cb8822007-07-19 01:46:59 -07002738 if (flags & FAULT_FLAG_WRITE) {
David Howells9637a5e2006-06-23 02:03:43 -07002739 if (!(vma->vm_flags & VM_SHARED)) {
Nick Piggin54cb8822007-07-19 01:46:59 -07002740 anon = 1;
Nick Piggind00806b2007-07-19 01:46:57 -07002741 if (unlikely(anon_vma_prepare(vma))) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002742 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002743 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002744 }
Nick Piggin83c54072007-07-19 01:47:05 -07002745 page = alloc_page_vma(GFP_HIGHUSER_MOVABLE,
2746 vma, address);
Nick Piggind00806b2007-07-19 01:46:57 -07002747 if (!page) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002748 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002749 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002750 }
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002751 if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) {
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002752 ret = VM_FAULT_OOM;
2753 page_cache_release(page);
2754 goto out;
2755 }
2756 charged = 1;
Nick Pigginb291f002008-10-18 20:26:44 -07002757 /*
2758 * Don't let another task, with possibly unlocked vma,
2759 * keep the mlocked page.
2760 */
2761 if (vma->vm_flags & VM_LOCKED)
2762 clear_page_mlock(vmf.page);
Nick Piggind0217ac2007-07-19 01:47:03 -07002763 copy_user_highpage(page, vmf.page, address, vma);
Nick Piggin0ed361d2008-02-04 22:29:34 -08002764 __SetPageUptodate(page);
David Howells9637a5e2006-06-23 02:03:43 -07002765 } else {
Nick Piggin54cb8822007-07-19 01:46:59 -07002766 /*
2767 * If the page will be shareable, see if the backing
David Howells9637a5e2006-06-23 02:03:43 -07002768 * address space wants to know that the page is about
Nick Piggin54cb8822007-07-19 01:46:59 -07002769 * to become writable
2770 */
Mark Fasheh69676142007-07-19 01:47:00 -07002771 if (vma->vm_ops->page_mkwrite) {
Nick Pigginc2ec1752009-03-31 15:23:21 -07002772 int tmp;
2773
Mark Fasheh69676142007-07-19 01:47:00 -07002774 unlock_page(page);
Nick Pigginb827e492009-04-30 15:08:16 -07002775 vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
Nick Pigginc2ec1752009-03-31 15:23:21 -07002776 tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
2777 if (unlikely(tmp &
2778 (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
2779 ret = tmp;
Nick Pigginb827e492009-04-30 15:08:16 -07002780 goto unwritable_page;
Mark Fasheh69676142007-07-19 01:47:00 -07002781 }
Nick Pigginb827e492009-04-30 15:08:16 -07002782 if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
2783 lock_page(page);
2784 if (!page->mapping) {
2785 ret = 0; /* retry the fault */
2786 unlock_page(page);
2787 goto unwritable_page;
2788 }
2789 } else
2790 VM_BUG_ON(!PageLocked(page));
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002791 page_mkwrite = 1;
David Howells9637a5e2006-06-23 02:03:43 -07002792 }
2793 }
Nick Piggin54cb8822007-07-19 01:46:59 -07002794
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795 }
2796
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002797 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798
2799 /*
2800 * This silly early PAGE_DIRTY setting removes a race
2801 * due to the bad i386 page protection. But it's valid
2802 * for other architectures too.
2803 *
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002804 * Note that if FAULT_FLAG_WRITE is set, we either now have
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805 * an exclusive copy of the page, or this is a shared mapping,
2806 * so we can make it writable and dirty to avoid having to
2807 * handle that later.
2808 */
2809 /* Only go through if we didn't race with anybody else... */
Andrea Arcangeli1c2fb7a2009-09-21 17:02:22 -07002810 if (likely(pte_same(*page_table, orig_pte))) {
Nick Piggind00806b2007-07-19 01:46:57 -07002811 flush_icache_page(vma, page);
2812 entry = mk_pte(page, vma->vm_page_prot);
Nick Piggin54cb8822007-07-19 01:46:59 -07002813 if (flags & FAULT_FLAG_WRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 if (anon) {
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002816 inc_mm_counter(mm, anon_rss);
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002817 page_add_new_anon_rmap(page, vma, address);
Hugh Dickinsf57e88a2005-11-21 21:32:19 -08002818 } else {
Hugh Dickins42946212005-10-29 18:16:05 -07002819 inc_mm_counter(mm, file_rss);
Nick Piggind00806b2007-07-19 01:46:57 -07002820 page_add_file_rmap(page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002821 if (flags & FAULT_FLAG_WRITE) {
Nick Piggind00806b2007-07-19 01:46:57 -07002822 dirty_page = page;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002823 get_page(dirty_page);
2824 }
Hugh Dickins42946212005-10-29 18:16:05 -07002825 }
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002826 set_pte_at(mm, address, page_table, entry);
Nick Piggind00806b2007-07-19 01:46:57 -07002827
2828 /* no need to invalidate: a not-present page won't be cached */
2829 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 } else {
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002831 if (charged)
2832 mem_cgroup_uncharge_page(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002833 if (anon)
2834 page_cache_release(page);
2835 else
Nick Piggin54cb8822007-07-19 01:46:59 -07002836 anon = 1; /* no anon but release faulted_page */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 }
2838
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002839 pte_unmap_unlock(page_table, ptl);
Nick Piggind00806b2007-07-19 01:46:57 -07002840
2841out:
Nick Pigginb827e492009-04-30 15:08:16 -07002842 if (dirty_page) {
2843 struct address_space *mapping = page->mapping;
2844
2845 if (set_page_dirty(dirty_page))
2846 page_mkwrite = 1;
2847 unlock_page(dirty_page);
2848 put_page(dirty_page);
2849 if (page_mkwrite && mapping) {
2850 /*
2851 * Some device drivers do not set page.mapping but still
2852 * dirty their pages
2853 */
2854 balance_dirty_pages_ratelimited(mapping);
2855 }
2856
2857 /* file_update_time outside page_lock */
Anton Salikhmetov8f7b3d12008-01-23 02:21:18 +03002858 if (vma->vm_file)
2859 file_update_time(vma->vm_file);
Nick Pigginb827e492009-04-30 15:08:16 -07002860 } else {
2861 unlock_page(vmf.page);
2862 if (anon)
2863 page_cache_release(vmf.page);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002864 }
Nick Piggind00806b2007-07-19 01:46:57 -07002865
Nick Piggin83c54072007-07-19 01:47:05 -07002866 return ret;
Nick Pigginb827e492009-04-30 15:08:16 -07002867
2868unwritable_page:
2869 page_cache_release(page);
2870 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07002871}
Nick Piggind00806b2007-07-19 01:46:57 -07002872
Nick Piggin54cb8822007-07-19 01:46:59 -07002873static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2874 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002875 unsigned int flags, pte_t orig_pte)
Nick Piggin54cb8822007-07-19 01:46:59 -07002876{
2877 pgoff_t pgoff = (((address & PAGE_MASK)
Dean Nelson0da7e012007-10-16 01:24:45 -07002878 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
Nick Piggin54cb8822007-07-19 01:46:59 -07002879
Hugh Dickins16abfa02007-10-04 16:56:06 +01002880 pte_unmap(page_table);
2881 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Nick Piggin54cb8822007-07-19 01:46:59 -07002882}
2883
Jes Sorensenf4b81802006-09-27 01:50:10 -07002884/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885 * Fault of a previously existing named mapping. Repopulate the pte
2886 * from the encoded file_pte if possible. This enables swappable
2887 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002888 *
2889 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2890 * but allow concurrent faults), and pte mapped but not yet locked.
2891 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002893static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins65500d22005-10-29 18:15:59 -07002894 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002895 unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896{
Hugh Dickins65500d22005-10-29 18:15:59 -07002897 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002899 flags |= FAULT_FLAG_NONLINEAR;
2900
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002901 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Nick Piggin83c54072007-07-19 01:47:05 -07002902 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002903
Hugh Dickins2509ef22009-01-06 14:40:10 -08002904 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002905 /*
2906 * Page table corrupted: show pte and kill process.
2907 */
Hugh Dickins3dc14742009-01-06 14:40:08 -08002908 print_bad_pte(vma, address, orig_pte, NULL);
Hugh Dickins65500d22005-10-29 18:15:59 -07002909 return VM_FAULT_OOM;
2910 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002911
2912 pgoff = pte_to_pgoff(orig_pte);
Hugh Dickins16abfa02007-10-04 16:56:06 +01002913 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914}
2915
2916/*
2917 * These routines also need to handle stuff like marking pages dirty
2918 * and/or accessed for architectures that don't do it in hardware (most
2919 * RISC architectures). The early dirtying is also good on the i386.
2920 *
2921 * There is also a hook called "update_mmu_cache()" that architectures
2922 * with external mmu caches can use to update those (ie the Sparc or
2923 * PowerPC hashed page tables that act as extended TLBs).
2924 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07002925 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2926 * but allow concurrent faults), and pte mapped but not yet locked.
2927 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928 */
2929static inline int handle_pte_fault(struct mm_struct *mm,
Hugh Dickins65500d22005-10-29 18:15:59 -07002930 struct vm_area_struct *vma, unsigned long address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002931 pte_t *pte, pmd_t *pmd, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932{
2933 pte_t entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002934 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002935
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002936 entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002938 if (pte_none(entry)) {
Jes Sorensenf4b81802006-09-27 01:50:10 -07002939 if (vma->vm_ops) {
Nick Piggin3c18ddd2008-04-28 02:12:10 -07002940 if (likely(vma->vm_ops->fault))
Nick Piggin54cb8822007-07-19 01:46:59 -07002941 return do_linear_fault(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002942 pte, pmd, flags, entry);
Jes Sorensenf4b81802006-09-27 01:50:10 -07002943 }
2944 return do_anonymous_page(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002945 pte, pmd, flags);
Hugh Dickins65500d22005-10-29 18:15:59 -07002946 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002947 if (pte_file(entry))
Nick Piggind0217ac2007-07-19 01:47:03 -07002948 return do_nonlinear_fault(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002949 pte, pmd, flags, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002950 return do_swap_page(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002951 pte, pmd, flags, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002952 }
2953
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002954 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002955 spin_lock(ptl);
2956 if (unlikely(!pte_same(*pte, entry)))
2957 goto unlock;
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002958 if (flags & FAULT_FLAG_WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002959 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002960 return do_wp_page(mm, vma, address,
2961 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962 entry = pte_mkdirty(entry);
2963 }
2964 entry = pte_mkyoung(entry);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002965 if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) {
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002966 update_mmu_cache(vma, address, entry);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002967 } else {
2968 /*
2969 * This is needed only for protection faults but the arch code
2970 * is not yet telling us if this is a protection fault or not.
2971 * This still avoids useless tlb flushes for .text page faults
2972 * with threads.
2973 */
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002974 if (flags & FAULT_FLAG_WRITE)
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002975 flush_tlb_page(vma, address);
2976 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002977unlock:
2978 pte_unmap_unlock(pte, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002979 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980}
2981
2982/*
2983 * By the time we get here, we already hold the mm semaphore
2984 */
Nick Piggin83c54072007-07-19 01:47:05 -07002985int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvaldsd06063c2009-04-10 09:01:23 -07002986 unsigned long address, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987{
2988 pgd_t *pgd;
2989 pud_t *pud;
2990 pmd_t *pmd;
2991 pte_t *pte;
2992
2993 __set_current_state(TASK_RUNNING);
2994
Christoph Lameterf8891e52006-06-30 01:55:45 -07002995 count_vm_event(PGFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01002997 if (unlikely(is_vm_hugetlb_page(vma)))
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002998 return hugetlb_fault(mm, vma, address, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001 pud = pud_alloc(mm, pgd, address);
3002 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07003003 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 pmd = pmd_alloc(mm, pud, address);
3005 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07003006 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007 pte = pte_alloc_map(mm, pmd, address);
3008 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -07003009 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07003011 return handle_pte_fault(mm, vma, address, pte, pmd, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003012}
3013
3014#ifndef __PAGETABLE_PUD_FOLDED
3015/*
3016 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07003017 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07003019int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020{
Hugh Dickinsc74df322005-10-29 18:16:23 -07003021 pud_t *new = pud_alloc_one(mm, address);
3022 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07003023 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024
Nick Piggin362a61a2008-05-14 06:37:36 +02003025 smp_wmb(); /* See comment in __pte_alloc */
3026
Hugh Dickins872fec12005-10-29 18:16:21 -07003027 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003028 if (pgd_present(*pgd)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003029 pud_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003030 else
3031 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07003032 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003033 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003034}
3035#endif /* __PAGETABLE_PUD_FOLDED */
3036
3037#ifndef __PAGETABLE_PMD_FOLDED
3038/*
3039 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07003040 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07003042int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043{
Hugh Dickinsc74df322005-10-29 18:16:23 -07003044 pmd_t *new = pmd_alloc_one(mm, address);
3045 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07003046 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003047
Nick Piggin362a61a2008-05-14 06:37:36 +02003048 smp_wmb(); /* See comment in __pte_alloc */
3049
Hugh Dickins872fec12005-10-29 18:16:21 -07003050 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07003052 if (pud_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003053 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003054 else
3055 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07003057 if (pgd_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003058 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003059 else
3060 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07003062 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003063 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064}
3065#endif /* __PAGETABLE_PMD_FOLDED */
3066
3067int make_pages_present(unsigned long addr, unsigned long end)
3068{
3069 int ret, len, write;
3070 struct vm_area_struct * vma;
3071
3072 vma = find_vma(current->mm, addr);
3073 if (!vma)
KOSAKI Motohiroa4770972008-08-04 13:41:14 -07003074 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075 write = (vma->vm_flags & VM_WRITE) != 0;
Eric Sesterhenn5bcb28b2006-03-26 18:30:52 +02003076 BUG_ON(addr >= end);
3077 BUG_ON(end > vma->vm_end);
Rolf Eike Beer68e116a2007-07-15 23:38:03 -07003078 len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079 ret = get_user_pages(current, current->mm, addr,
3080 len, write, 0, NULL, NULL);
Lee Schermerhornc11d69d2008-10-18 20:26:56 -07003081 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082 return ret;
Lee Schermerhorn9978ad52008-10-18 20:26:56 -07003083 return ret == len ? 0 : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003084}
3085
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086#if !defined(__HAVE_ARCH_GATE_AREA)
3087
3088#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07003089static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090
3091static int __init gate_vma_init(void)
3092{
3093 gate_vma.vm_mm = NULL;
3094 gate_vma.vm_start = FIXADDR_USER_START;
3095 gate_vma.vm_end = FIXADDR_USER_END;
Roland McGrathb6558c42007-01-26 00:56:47 -08003096 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
3097 gate_vma.vm_page_prot = __P101;
Roland McGrathf47aef52007-01-26 00:56:49 -08003098 /*
3099 * Make sure the vDSO gets into every core dump.
3100 * Dumping its contents makes post-mortem fully interpretable later
3101 * without matching up the same kernel and hardware config to see
3102 * what PC values meant.
3103 */
3104 gate_vma.vm_flags |= VM_ALWAYSDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003105 return 0;
3106}
3107__initcall(gate_vma_init);
3108#endif
3109
3110struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
3111{
3112#ifdef AT_SYSINFO_EHDR
3113 return &gate_vma;
3114#else
3115 return NULL;
3116#endif
3117}
3118
3119int in_gate_area_no_task(unsigned long addr)
3120{
3121#ifdef AT_SYSINFO_EHDR
3122 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
3123 return 1;
3124#endif
3125 return 0;
3126}
3127
3128#endif /* __HAVE_ARCH_GATE_AREA */
David Howells0ec76a12006-09-27 01:50:15 -07003129
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07003130static int follow_pte(struct mm_struct *mm, unsigned long address,
3131 pte_t **ptepp, spinlock_t **ptlp)
3132{
3133 pgd_t *pgd;
3134 pud_t *pud;
3135 pmd_t *pmd;
3136 pte_t *ptep;
3137
3138 pgd = pgd_offset(mm, address);
3139 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
3140 goto out;
3141
3142 pud = pud_offset(pgd, address);
3143 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
3144 goto out;
3145
3146 pmd = pmd_offset(pud, address);
3147 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
3148 goto out;
3149
3150 /* We cannot handle huge page PFN maps. Luckily they don't exist. */
3151 if (pmd_huge(*pmd))
3152 goto out;
3153
3154 ptep = pte_offset_map_lock(mm, pmd, address, ptlp);
3155 if (!ptep)
3156 goto out;
3157 if (!pte_present(*ptep))
3158 goto unlock;
3159 *ptepp = ptep;
3160 return 0;
3161unlock:
3162 pte_unmap_unlock(ptep, *ptlp);
3163out:
3164 return -EINVAL;
3165}
3166
Johannes Weiner3b6748e2009-06-16 15:32:35 -07003167/**
3168 * follow_pfn - look up PFN at a user virtual address
3169 * @vma: memory mapping
3170 * @address: user virtual address
3171 * @pfn: location to store found PFN
3172 *
3173 * Only IO mappings and raw PFN mappings are allowed.
3174 *
3175 * Returns zero and the pfn at @pfn on success, -ve otherwise.
3176 */
3177int follow_pfn(struct vm_area_struct *vma, unsigned long address,
3178 unsigned long *pfn)
3179{
3180 int ret = -EINVAL;
3181 spinlock_t *ptl;
3182 pte_t *ptep;
3183
3184 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3185 return ret;
3186
3187 ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);
3188 if (ret)
3189 return ret;
3190 *pfn = pte_pfn(*ptep);
3191 pte_unmap_unlock(ptep, ptl);
3192 return 0;
3193}
3194EXPORT_SYMBOL(follow_pfn);
3195
Rik van Riel28b2ee22008-07-23 21:27:05 -07003196#ifdef CONFIG_HAVE_IOREMAP_PROT
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003197int follow_phys(struct vm_area_struct *vma,
3198 unsigned long address, unsigned int flags,
3199 unsigned long *prot, resource_size_t *phys)
Rik van Riel28b2ee22008-07-23 21:27:05 -07003200{
Johannes Weiner03668a42009-06-16 15:32:34 -07003201 int ret = -EINVAL;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003202 pte_t *ptep, pte;
3203 spinlock_t *ptl;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003204
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003205 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3206 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003207
Johannes Weiner03668a42009-06-16 15:32:34 -07003208 if (follow_pte(vma->vm_mm, address, &ptep, &ptl))
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003209 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003210 pte = *ptep;
Johannes Weiner03668a42009-06-16 15:32:34 -07003211
Rik van Riel28b2ee22008-07-23 21:27:05 -07003212 if ((flags & FOLL_WRITE) && !pte_write(pte))
3213 goto unlock;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003214
3215 *prot = pgprot_val(pte_pgprot(pte));
Johannes Weiner03668a42009-06-16 15:32:34 -07003216 *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003217
Johannes Weiner03668a42009-06-16 15:32:34 -07003218 ret = 0;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003219unlock:
3220 pte_unmap_unlock(ptep, ptl);
3221out:
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003222 return ret;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003223}
3224
3225int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
3226 void *buf, int len, int write)
3227{
3228 resource_size_t phys_addr;
3229 unsigned long prot = 0;
KOSAKI Motohiro2bc72732009-01-06 14:39:43 -08003230 void __iomem *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003231 int offset = addr & (PAGE_SIZE-1);
3232
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003233 if (follow_phys(vma, addr, write, &prot, &phys_addr))
Rik van Riel28b2ee22008-07-23 21:27:05 -07003234 return -EINVAL;
3235
3236 maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot);
3237 if (write)
3238 memcpy_toio(maddr + offset, buf, len);
3239 else
3240 memcpy_fromio(buf, maddr + offset, len);
3241 iounmap(maddr);
3242
3243 return len;
3244}
3245#endif
3246
David Howells0ec76a12006-09-27 01:50:15 -07003247/*
3248 * Access another process' address space.
3249 * Source/target buffer must be kernel space,
3250 * Do not walk the page table directly, use get_user_pages
3251 */
3252int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
3253{
3254 struct mm_struct *mm;
3255 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07003256 void *old_buf = buf;
3257
3258 mm = get_task_mm(tsk);
3259 if (!mm)
3260 return 0;
3261
3262 down_read(&mm->mmap_sem);
Simon Arlott183ff222007-10-20 01:27:18 +02003263 /* ignore errors, just check how much was successfully transferred */
David Howells0ec76a12006-09-27 01:50:15 -07003264 while (len) {
3265 int bytes, ret, offset;
3266 void *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003267 struct page *page = NULL;
David Howells0ec76a12006-09-27 01:50:15 -07003268
3269 ret = get_user_pages(tsk, mm, addr, 1,
3270 write, 1, &page, &vma);
Rik van Riel28b2ee22008-07-23 21:27:05 -07003271 if (ret <= 0) {
3272 /*
3273 * Check if this is a VM_IO | VM_PFNMAP VMA, which
3274 * we can access using slightly different code.
3275 */
3276#ifdef CONFIG_HAVE_IOREMAP_PROT
3277 vma = find_vma(mm, addr);
3278 if (!vma)
3279 break;
3280 if (vma->vm_ops && vma->vm_ops->access)
3281 ret = vma->vm_ops->access(vma, addr, buf,
3282 len, write);
3283 if (ret <= 0)
3284#endif
3285 break;
3286 bytes = ret;
David Howells0ec76a12006-09-27 01:50:15 -07003287 } else {
Rik van Riel28b2ee22008-07-23 21:27:05 -07003288 bytes = len;
3289 offset = addr & (PAGE_SIZE-1);
3290 if (bytes > PAGE_SIZE-offset)
3291 bytes = PAGE_SIZE-offset;
3292
3293 maddr = kmap(page);
3294 if (write) {
3295 copy_to_user_page(vma, page, addr,
3296 maddr + offset, buf, bytes);
3297 set_page_dirty_lock(page);
3298 } else {
3299 copy_from_user_page(vma, page, addr,
3300 buf, maddr + offset, bytes);
3301 }
3302 kunmap(page);
3303 page_cache_release(page);
David Howells0ec76a12006-09-27 01:50:15 -07003304 }
David Howells0ec76a12006-09-27 01:50:15 -07003305 len -= bytes;
3306 buf += bytes;
3307 addr += bytes;
3308 }
3309 up_read(&mm->mmap_sem);
3310 mmput(mm);
3311
3312 return buf - old_buf;
3313}
Andi Kleen03252912008-01-30 13:33:18 +01003314
3315/*
3316 * Print the name of a VMA.
3317 */
3318void print_vma_addr(char *prefix, unsigned long ip)
3319{
3320 struct mm_struct *mm = current->mm;
3321 struct vm_area_struct *vma;
3322
Ingo Molnare8bff742008-02-13 20:21:06 +01003323 /*
3324 * Do not print if we are in atomic
3325 * contexts (in exception stacks, etc.):
3326 */
3327 if (preempt_count())
3328 return;
3329
Andi Kleen03252912008-01-30 13:33:18 +01003330 down_read(&mm->mmap_sem);
3331 vma = find_vma(mm, ip);
3332 if (vma && vma->vm_file) {
3333 struct file *f = vma->vm_file;
3334 char *buf = (char *)__get_free_page(GFP_KERNEL);
3335 if (buf) {
3336 char *p, *s;
3337
Jan Blunckcf28b482008-02-14 19:38:44 -08003338 p = d_path(&f->f_path, buf, PAGE_SIZE);
Andi Kleen03252912008-01-30 13:33:18 +01003339 if (IS_ERR(p))
3340 p = "?";
3341 s = strrchr(p, '/');
3342 if (s)
3343 p = s+1;
3344 printk("%s%s[%lx+%lx]", prefix, p,
3345 vma->vm_start,
3346 vma->vm_end - vma->vm_start);
3347 free_page((unsigned long)buf);
3348 }
3349 }
3350 up_read(&current->mm->mmap_sem);
3351}
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003352
3353#ifdef CONFIG_PROVE_LOCKING
3354void might_fault(void)
3355{
Peter Zijlstra95156f02009-01-12 13:02:11 +01003356 /*
3357 * Some code (nfs/sunrpc) uses socket ops on kernel memory while
3358 * holding the mmap_sem, this is safe because kernel memory doesn't
3359 * get paged out, therefore we'll never actually fault, and the
3360 * below annotations will generate false positives.
3361 */
3362 if (segment_eq(get_fs(), KERNEL_DS))
3363 return;
3364
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003365 might_sleep();
3366 /*
3367 * it would be nicer only to annotate paths which are not under
3368 * pagefault_disable, however that requires a larger audit and
3369 * providing helpers like get_user_atomic.
3370 */
3371 if (!in_atomic() && current->mm)
3372 might_lock_read(&current->mm->mmap_sem);
3373}
3374EXPORT_SYMBOL(might_fault);
3375#endif