blob: e8f63d9961ea58f5c9643c111859911bc91dc675 [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>
48#include <linux/rmap.h>
49#include <linux/module.h>
Shailabh Nagar0ff92242006-07-14 00:24:37 -070050#include <linux/delayacct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include <linux/init.h>
Peter Zijlstraedc79b22006-09-25 23:30:58 -070052#include <linux/writeback.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080053#include <linux/memcontrol.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070054#include <linux/mmu_notifier.h>
Hugh Dickins3dc14742009-01-06 14:40:08 -080055#include <linux/kallsyms.h>
56#include <linux/swapops.h>
57#include <linux/elf.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Alexey Dobriyan6952b612009-09-18 23:55:55 +040059#include <asm/io.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070060#include <asm/pgalloc.h>
61#include <asm/uaccess.h>
62#include <asm/tlb.h>
63#include <asm/tlbflush.h>
64#include <asm/pgtable.h>
65
Jan Beulich42b77722008-07-23 21:27:10 -070066#include "internal.h"
67
Andy Whitcroftd41dee32005-06-23 00:07:54 -070068#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -070069/* use the per-pgdat data instead for discontigmem - mbligh */
70unsigned long max_mapnr;
71struct page *mem_map;
72
73EXPORT_SYMBOL(max_mapnr);
74EXPORT_SYMBOL(mem_map);
75#endif
76
77unsigned long num_physpages;
78/*
79 * A number of key systems in x86 including ioremap() rely on the assumption
80 * that high_memory defines the upper bound on direct map memory, then end
81 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
82 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
83 * and ZONE_HIGHMEM.
84 */
85void * high_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
87EXPORT_SYMBOL(num_physpages);
88EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
Ingo Molnar32a93232008-02-06 22:39:44 +010090/*
91 * Randomize the address space (stacks, mmaps, brk, etc.).
92 *
93 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
94 * as ancient (libc5 based) binaries can segfault. )
95 */
96int randomize_va_space __read_mostly =
97#ifdef CONFIG_COMPAT_BRK
98 1;
99#else
100 2;
101#endif
Andi Kleena62eaf12006-02-16 23:41:58 +0100102
103static int __init disable_randmaps(char *s)
104{
105 randomize_va_space = 0;
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800106 return 1;
Andi Kleena62eaf12006-02-16 23:41:58 +0100107}
108__setup("norandmaps", disable_randmaps);
109
110
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111/*
112 * If a p?d_bad entry is found while walking page tables, report
113 * the error, before resetting entry to p?d_none. Usually (but
114 * very seldom) called out from the p?d_none_or_clear_bad macros.
115 */
116
117void pgd_clear_bad(pgd_t *pgd)
118{
119 pgd_ERROR(*pgd);
120 pgd_clear(pgd);
121}
122
123void pud_clear_bad(pud_t *pud)
124{
125 pud_ERROR(*pud);
126 pud_clear(pud);
127}
128
129void pmd_clear_bad(pmd_t *pmd)
130{
131 pmd_ERROR(*pmd);
132 pmd_clear(pmd);
133}
134
135/*
136 * Note: this doesn't free the actual pages themselves. That
137 * has been handled earlier when unmapping all the memory regions.
138 */
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000139static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
140 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800142 pgtable_t token = pmd_pgtable(*pmd);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700143 pmd_clear(pmd);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000144 pte_free_tlb(tlb, token, addr);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700145 tlb->mm->nr_ptes--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146}
147
Hugh Dickinse0da3822005-04-19 13:29:15 -0700148static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
149 unsigned long addr, unsigned long end,
150 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151{
152 pmd_t *pmd;
153 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700154 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155
Hugh Dickinse0da3822005-04-19 13:29:15 -0700156 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 do {
159 next = pmd_addr_end(addr, end);
160 if (pmd_none_or_clear_bad(pmd))
161 continue;
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000162 free_pte_range(tlb, pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 } while (pmd++, addr = next, addr != end);
164
Hugh Dickinse0da3822005-04-19 13:29:15 -0700165 start &= PUD_MASK;
166 if (start < floor)
167 return;
168 if (ceiling) {
169 ceiling &= PUD_MASK;
170 if (!ceiling)
171 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700173 if (end - 1 > ceiling - 1)
174 return;
175
176 pmd = pmd_offset(pud, start);
177 pud_clear(pud);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000178 pmd_free_tlb(tlb, pmd, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179}
180
Hugh Dickinse0da3822005-04-19 13:29:15 -0700181static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
182 unsigned long addr, unsigned long end,
183 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184{
185 pud_t *pud;
186 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700187 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
Hugh Dickinse0da3822005-04-19 13:29:15 -0700189 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 pud = pud_offset(pgd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 do {
192 next = pud_addr_end(addr, end);
193 if (pud_none_or_clear_bad(pud))
194 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700195 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 } while (pud++, addr = next, addr != end);
197
Hugh Dickinse0da3822005-04-19 13:29:15 -0700198 start &= PGDIR_MASK;
199 if (start < floor)
200 return;
201 if (ceiling) {
202 ceiling &= PGDIR_MASK;
203 if (!ceiling)
204 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700206 if (end - 1 > ceiling - 1)
207 return;
208
209 pud = pud_offset(pgd, start);
210 pgd_clear(pgd);
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000211 pud_free_tlb(tlb, pud, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212}
213
214/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700215 * This function frees user-level page tables of a process.
216 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 * Must be called with pagetable lock held.
218 */
Jan Beulich42b77722008-07-23 21:27:10 -0700219void free_pgd_range(struct mmu_gather *tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700220 unsigned long addr, unsigned long end,
221 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222{
223 pgd_t *pgd;
224 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700225 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226
Hugh Dickinse0da3822005-04-19 13:29:15 -0700227 /*
228 * The next few lines have given us lots of grief...
229 *
230 * Why are we testing PMD* at this top level? Because often
231 * there will be no work to do at all, and we'd prefer not to
232 * go all the way down to the bottom just to discover that.
233 *
234 * Why all these "- 1"s? Because 0 represents both the bottom
235 * of the address space and the top of it (using -1 for the
236 * top wouldn't help much: the masks would do the wrong thing).
237 * The rule is that addr 0 and floor 0 refer to the bottom of
238 * the address space, but end 0 and ceiling 0 refer to the top
239 * Comparisons need to use "end - 1" and "ceiling - 1" (though
240 * that end 0 case should be mythical).
241 *
242 * Wherever addr is brought up or ceiling brought down, we must
243 * be careful to reject "the opposite 0" before it confuses the
244 * subsequent tests. But what about where end is brought down
245 * by PMD_SIZE below? no, end can't go down to 0 there.
246 *
247 * Whereas we round start (addr) and ceiling down, by different
248 * masks at different levels, in order to test whether a table
249 * now has no other vmas using it, so can be freed, we don't
250 * bother to round floor or end up - the tests don't need that.
251 */
252
253 addr &= PMD_MASK;
254 if (addr < floor) {
255 addr += PMD_SIZE;
256 if (!addr)
257 return;
258 }
259 if (ceiling) {
260 ceiling &= PMD_MASK;
261 if (!ceiling)
262 return;
263 }
264 if (end - 1 > ceiling - 1)
265 end -= PMD_SIZE;
266 if (addr > end - 1)
267 return;
268
269 start = addr;
Jan Beulich42b77722008-07-23 21:27:10 -0700270 pgd = pgd_offset(tlb->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 do {
272 next = pgd_addr_end(addr, end);
273 if (pgd_none_or_clear_bad(pgd))
274 continue;
Jan Beulich42b77722008-07-23 21:27:10 -0700275 free_pud_range(tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700277}
278
Jan Beulich42b77722008-07-23 21:27:10 -0700279void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700280 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700281{
282 while (vma) {
283 struct vm_area_struct *next = vma->vm_next;
284 unsigned long addr = vma->vm_start;
285
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700286 /*
287 * Hide vma from rmap and vmtruncate before freeing pgtables
288 */
289 anon_vma_unlink(vma);
290 unlink_file_vma(vma);
291
David Gibson9da61ae2006-03-22 00:08:57 -0800292 if (is_vm_hugetlb_page(vma)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700293 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700294 floor, next? next->vm_start: ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700295 } else {
296 /*
297 * Optimization: gather nearby vmas into one call down
298 */
299 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
David Gibson48669202006-03-22 00:08:58 -0800300 && !is_vm_hugetlb_page(next)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700301 vma = next;
302 next = vma->vm_next;
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700303 anon_vma_unlink(vma);
304 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700305 }
306 free_pgd_range(tlb, addr, vma->vm_end,
307 floor, next? next->vm_start: ceiling);
308 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700309 vma = next;
310 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311}
312
Hugh Dickins1bb36302005-10-29 18:16:22 -0700313int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800315 pgtable_t new = pte_alloc_one(mm, address);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700316 if (!new)
317 return -ENOMEM;
318
Nick Piggin362a61a2008-05-14 06:37:36 +0200319 /*
320 * Ensure all pte setup (eg. pte page lock and page clearing) are
321 * visible before the pte is made visible to other CPUs by being
322 * put into page tables.
323 *
324 * The other side of the story is the pointer chasing in the page
325 * table walking code (when walking the page table without locking;
326 * ie. most of the time). Fortunately, these data accesses consist
327 * of a chain of data-dependent loads, meaning most CPUs (alpha
328 * being the notable exception) will already guarantee loads are
329 * seen in-order. See the alpha page table accessors for the
330 * smp_read_barrier_depends() barriers in page table walking code.
331 */
332 smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
333
Hugh Dickinsc74df322005-10-29 18:16:23 -0700334 spin_lock(&mm->page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800335 if (!pmd_present(*pmd)) { /* Has another populated it ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 mm->nr_ptes++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 pmd_populate(mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800338 new = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 }
Hugh Dickinsc74df322005-10-29 18:16:23 -0700340 spin_unlock(&mm->page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800341 if (new)
342 pte_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700343 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344}
345
Hugh Dickins1bb36302005-10-29 18:16:22 -0700346int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700348 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
349 if (!new)
350 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351
Nick Piggin362a61a2008-05-14 06:37:36 +0200352 smp_wmb(); /* See comment in __pte_alloc */
353
Hugh Dickins1bb36302005-10-29 18:16:22 -0700354 spin_lock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800355 if (!pmd_present(*pmd)) { /* Has another populated it ? */
Hugh Dickins1bb36302005-10-29 18:16:22 -0700356 pmd_populate_kernel(&init_mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800357 new = NULL;
358 }
Hugh Dickins1bb36302005-10-29 18:16:22 -0700359 spin_unlock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800360 if (new)
361 pte_free_kernel(&init_mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700362 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363}
364
Hugh Dickinsae859762005-10-29 18:16:05 -0700365static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
366{
367 if (file_rss)
368 add_mm_counter(mm, file_rss, file_rss);
369 if (anon_rss)
370 add_mm_counter(mm, anon_rss, anon_rss);
371}
372
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800374 * This function is called to print an error when a bad pte
375 * is found. For example, we might have a PFN-mapped pte in
376 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700377 *
378 * The calling function must still handle the error.
379 */
Hugh Dickins3dc14742009-01-06 14:40:08 -0800380static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
381 pte_t pte, struct page *page)
Nick Pigginb5810032005-10-29 18:16:12 -0700382{
Hugh Dickins3dc14742009-01-06 14:40:08 -0800383 pgd_t *pgd = pgd_offset(vma->vm_mm, addr);
384 pud_t *pud = pud_offset(pgd, addr);
385 pmd_t *pmd = pmd_offset(pud, addr);
386 struct address_space *mapping;
387 pgoff_t index;
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800388 static unsigned long resume;
389 static unsigned long nr_shown;
390 static unsigned long nr_unshown;
391
392 /*
393 * Allow a burst of 60 reports, then keep quiet for that minute;
394 * or allow a steady drip of one report per second.
395 */
396 if (nr_shown == 60) {
397 if (time_before(jiffies, resume)) {
398 nr_unshown++;
399 return;
400 }
401 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800402 printk(KERN_ALERT
403 "BUG: Bad page map: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800404 nr_unshown);
405 nr_unshown = 0;
406 }
407 nr_shown = 0;
408 }
409 if (nr_shown++ == 0)
410 resume = jiffies + 60 * HZ;
Hugh Dickins3dc14742009-01-06 14:40:08 -0800411
412 mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;
413 index = linear_page_index(vma, addr);
414
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800415 printk(KERN_ALERT
416 "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800417 current->comm,
418 (long long)pte_val(pte), (long long)pmd_val(*pmd));
419 if (page) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800420 printk(KERN_ALERT
Hugh Dickins3dc14742009-01-06 14:40:08 -0800421 "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
422 page, (void *)page->flags, page_count(page),
423 page_mapcount(page), page->mapping, page->index);
424 }
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800425 printk(KERN_ALERT
Hugh Dickins3dc14742009-01-06 14:40:08 -0800426 "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
427 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
428 /*
429 * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y
430 */
431 if (vma->vm_ops)
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800432 print_symbol(KERN_ALERT "vma->vm_ops->fault: %s\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800433 (unsigned long)vma->vm_ops->fault);
434 if (vma->vm_file && vma->vm_file->f_op)
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800435 print_symbol(KERN_ALERT "vma->vm_file->f_op->mmap: %s\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800436 (unsigned long)vma->vm_file->f_op->mmap);
Nick Pigginb5810032005-10-29 18:16:12 -0700437 dump_stack();
Hugh Dickins3dc14742009-01-06 14:40:08 -0800438 add_taint(TAINT_BAD_PAGE);
Nick Pigginb5810032005-10-29 18:16:12 -0700439}
440
Linus Torvalds67121172005-12-11 20:38:17 -0800441static inline int is_cow_mapping(unsigned int flags)
442{
443 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
444}
445
Nick Pigginb5810032005-10-29 18:16:12 -0700446/*
Nick Piggin7e675132008-04-28 02:13:00 -0700447 * vm_normal_page -- This function gets the "struct page" associated with a pte.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800448 *
Nick Piggin7e675132008-04-28 02:13:00 -0700449 * "Special" mappings do not wish to be associated with a "struct page" (either
450 * it doesn't exist, or it exists but they don't want to touch it). In this
451 * case, NULL is returned here. "Normal" mappings do have a struct page.
Jared Hulbertb379d792008-04-28 02:12:58 -0700452 *
Nick Piggin7e675132008-04-28 02:13:00 -0700453 * There are 2 broad cases. Firstly, an architecture may define a pte_special()
454 * pte bit, in which case this function is trivial. Secondly, an architecture
455 * may not have a spare pte bit, which requires a more complicated scheme,
456 * described below.
457 *
458 * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
459 * special mapping (even if there are underlying and valid "struct pages").
460 * COWed pages of a VM_PFNMAP are always normal.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800461 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700462 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
463 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
Nick Piggin7e675132008-04-28 02:13:00 -0700464 * set, and the vm_pgoff will point to the first PFN mapped: thus every special
465 * mapping will always honor the rule
Linus Torvalds6aab3412005-11-28 14:34:23 -0800466 *
467 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
468 *
Nick Piggin7e675132008-04-28 02:13:00 -0700469 * And for normal mappings this is false.
Jared Hulbertb379d792008-04-28 02:12:58 -0700470 *
Nick Piggin7e675132008-04-28 02:13:00 -0700471 * This restricts such mappings to be a linear translation from virtual address
472 * to pfn. To get around this restriction, we allow arbitrary mappings so long
473 * as the vma is not a COW mapping; in that case, we know that all ptes are
474 * special (because none can have been COWed).
Jared Hulbertb379d792008-04-28 02:12:58 -0700475 *
476 *
Nick Piggin7e675132008-04-28 02:13:00 -0700477 * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
478 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700479 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
480 * page" backing, however the difference is that _all_ pages with a struct
481 * page (that is, those where pfn_valid is true) are refcounted and considered
482 * normal pages by the VM. The disadvantage is that pages are refcounted
483 * (which can be slower and simply not an option for some PFNMAP users). The
484 * advantage is that we don't have to follow the strict linearity rule of
485 * PFNMAP mappings in order to support COWable mappings.
486 *
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800487 */
Nick Piggin7e675132008-04-28 02:13:00 -0700488#ifdef __HAVE_ARCH_PTE_SPECIAL
489# define HAVE_PTE_SPECIAL 1
490#else
491# define HAVE_PTE_SPECIAL 0
492#endif
493struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
494 pte_t pte)
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800495{
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800496 unsigned long pfn = pte_pfn(pte);
Nick Piggin7e675132008-04-28 02:13:00 -0700497
498 if (HAVE_PTE_SPECIAL) {
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800499 if (likely(!pte_special(pte)))
500 goto check_pfn;
501 if (!(vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)))
502 print_bad_pte(vma, addr, pte, NULL);
Nick Piggin7e675132008-04-28 02:13:00 -0700503 return NULL;
504 }
505
506 /* !HAVE_PTE_SPECIAL case follows: */
507
Jared Hulbertb379d792008-04-28 02:12:58 -0700508 if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
509 if (vma->vm_flags & VM_MIXEDMAP) {
510 if (!pfn_valid(pfn))
511 return NULL;
512 goto out;
513 } else {
Nick Piggin7e675132008-04-28 02:13:00 -0700514 unsigned long off;
515 off = (addr - vma->vm_start) >> PAGE_SHIFT;
Jared Hulbertb379d792008-04-28 02:12:58 -0700516 if (pfn == vma->vm_pgoff + off)
517 return NULL;
518 if (!is_cow_mapping(vma->vm_flags))
519 return NULL;
520 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800521 }
522
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800523check_pfn:
524 if (unlikely(pfn > highest_memmap_pfn)) {
525 print_bad_pte(vma, addr, pte, NULL);
526 return NULL;
527 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800528
529 /*
Nick Piggin7e675132008-04-28 02:13:00 -0700530 * NOTE! We still have PageReserved() pages in the page tables.
Nick Piggin7e675132008-04-28 02:13:00 -0700531 * eg. VDSO mappings can cause them to exist.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800532 */
Jared Hulbertb379d792008-04-28 02:12:58 -0700533out:
Linus Torvalds6aab3412005-11-28 14:34:23 -0800534 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800535}
536
537/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 * copy one vm_area from one task to the other. Assumes the page tables
539 * already present in the new task to be cleared in the whole range
540 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 */
542
Hugh Dickins8c103762005-10-29 18:16:13 -0700543static inline void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700545 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700546 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547{
Nick Pigginb5810032005-10-29 18:16:12 -0700548 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 pte_t pte = *src_pte;
550 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
552 /* pte contains position in swap or file, so copy. */
553 if (unlikely(!pte_present(pte))) {
554 if (!pte_file(pte)) {
Christoph Lameter06972122006-06-23 02:03:35 -0700555 swp_entry_t entry = pte_to_swp_entry(pte);
556
557 swap_duplicate(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 /* make sure dst_mm is on swapoff's mmlist. */
559 if (unlikely(list_empty(&dst_mm->mmlist))) {
560 spin_lock(&mmlist_lock);
Hugh Dickinsf412ac02005-10-29 18:16:41 -0700561 if (list_empty(&dst_mm->mmlist))
562 list_add(&dst_mm->mmlist,
563 &src_mm->mmlist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 spin_unlock(&mmlist_lock);
565 }
Christoph Lameter06972122006-06-23 02:03:35 -0700566 if (is_write_migration_entry(entry) &&
567 is_cow_mapping(vm_flags)) {
568 /*
569 * COW mappings require pages in both parent
570 * and child to be set to read.
571 */
572 make_migration_entry_read(&entry);
573 pte = swp_entry_to_pte(entry);
574 set_pte_at(src_mm, addr, src_pte, pte);
575 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700577 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 }
579
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 /*
581 * If it's a COW mapping, write protect it both
582 * in the parent and the child
583 */
Linus Torvalds67121172005-12-11 20:38:17 -0800584 if (is_cow_mapping(vm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 ptep_set_wrprotect(src_mm, addr, src_pte);
Zachary Amsden3dc90792006-09-30 23:29:30 -0700586 pte = pte_wrprotect(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 }
588
589 /*
590 * If it's a shared mapping, mark it clean in
591 * the child
592 */
593 if (vm_flags & VM_SHARED)
594 pte = pte_mkclean(pte);
595 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800596
597 page = vm_normal_page(vma, addr, pte);
598 if (page) {
599 get_page(page);
Nick Pigginc97a9e12007-05-16 22:11:21 -0700600 page_dup_rmap(page, vma, addr);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800601 rss[!!PageAnon(page)]++;
602 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700603
604out_set_pte:
605 set_pte_at(dst_mm, addr, dst_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606}
607
608static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
609 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
610 unsigned long addr, unsigned long end)
611{
612 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700613 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -0700614 int progress = 0;
Hugh Dickins8c103762005-10-29 18:16:13 -0700615 int rss[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
617again:
Hugh Dickinsae859762005-10-29 18:16:05 -0700618 rss[1] = rss[0] = 0;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700619 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 if (!dst_pte)
621 return -ENOMEM;
622 src_pte = pte_offset_map_nested(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700623 src_ptl = pte_lockptr(src_mm, src_pmd);
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700624 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700625 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 do {
628 /*
629 * We are holding two locks at this point - either of them
630 * could generate latencies in another task on another CPU.
631 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700632 if (progress >= 32) {
633 progress = 0;
634 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +0100635 spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700636 break;
637 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 if (pte_none(*src_pte)) {
639 progress++;
640 continue;
641 }
Hugh Dickins8c103762005-10-29 18:16:13 -0700642 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 progress += 8;
644 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700646 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -0700647 spin_unlock(src_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 pte_unmap_nested(src_pte - 1);
Hugh Dickinsae859762005-10-29 18:16:05 -0700649 add_mm_rss(dst_mm, rss[0], rss[1]);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700650 pte_unmap_unlock(dst_pte - 1, dst_ptl);
651 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 if (addr != end)
653 goto again;
654 return 0;
655}
656
657static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
658 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
659 unsigned long addr, unsigned long end)
660{
661 pmd_t *src_pmd, *dst_pmd;
662 unsigned long next;
663
664 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
665 if (!dst_pmd)
666 return -ENOMEM;
667 src_pmd = pmd_offset(src_pud, addr);
668 do {
669 next = pmd_addr_end(addr, end);
670 if (pmd_none_or_clear_bad(src_pmd))
671 continue;
672 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
673 vma, addr, next))
674 return -ENOMEM;
675 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
676 return 0;
677}
678
679static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
680 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
681 unsigned long addr, unsigned long end)
682{
683 pud_t *src_pud, *dst_pud;
684 unsigned long next;
685
686 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
687 if (!dst_pud)
688 return -ENOMEM;
689 src_pud = pud_offset(src_pgd, addr);
690 do {
691 next = pud_addr_end(addr, end);
692 if (pud_none_or_clear_bad(src_pud))
693 continue;
694 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
695 vma, addr, next))
696 return -ENOMEM;
697 } while (dst_pud++, src_pud++, addr = next, addr != end);
698 return 0;
699}
700
701int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
702 struct vm_area_struct *vma)
703{
704 pgd_t *src_pgd, *dst_pgd;
705 unsigned long next;
706 unsigned long addr = vma->vm_start;
707 unsigned long end = vma->vm_end;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700708 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709
Nick Piggind9928952005-08-28 16:49:11 +1000710 /*
711 * Don't copy ptes where a page fault will fill them correctly.
712 * Fork becomes much lighter when there are big shared or private
713 * readonly mappings. The tradeoff is that copy_page_range is more
714 * efficient than faulting.
715 */
Linus Torvalds4d7672b2005-12-16 10:21:23 -0800716 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
Nick Piggind9928952005-08-28 16:49:11 +1000717 if (!vma->anon_vma)
718 return 0;
719 }
720
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 if (is_vm_hugetlb_page(vma))
722 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
723
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800724 if (unlikely(is_pfn_mapping(vma))) {
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -0800725 /*
726 * We do not free on error cases below as remove_vma
727 * gets called on error from higher level routine
728 */
729 ret = track_pfn_vma_copy(vma);
730 if (ret)
731 return ret;
732 }
733
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700734 /*
735 * We need to invalidate the secondary MMU mappings only when
736 * there could be a permission downgrade on the ptes of the
737 * parent mm. And a permission downgrade will only happen if
738 * is_cow_mapping() returns true.
739 */
740 if (is_cow_mapping(vma->vm_flags))
741 mmu_notifier_invalidate_range_start(src_mm, addr, end);
742
743 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 dst_pgd = pgd_offset(dst_mm, addr);
745 src_pgd = pgd_offset(src_mm, addr);
746 do {
747 next = pgd_addr_end(addr, end);
748 if (pgd_none_or_clear_bad(src_pgd))
749 continue;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700750 if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
751 vma, addr, next))) {
752 ret = -ENOMEM;
753 break;
754 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700756
757 if (is_cow_mapping(vma->vm_flags))
758 mmu_notifier_invalidate_range_end(src_mm,
759 vma->vm_start, end);
760 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761}
762
Robin Holt51c6f662005-11-13 16:06:42 -0800763static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700764 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800766 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767{
Nick Pigginb5810032005-10-29 18:16:12 -0700768 struct mm_struct *mm = tlb->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 pte_t *pte;
Hugh Dickins508034a2005-10-29 18:16:30 -0700770 spinlock_t *ptl;
Hugh Dickinsae859762005-10-29 18:16:05 -0700771 int file_rss = 0;
772 int anon_rss = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773
Hugh Dickins508034a2005-10-29 18:16:30 -0700774 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700775 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 do {
777 pte_t ptent = *pte;
Robin Holt51c6f662005-11-13 16:06:42 -0800778 if (pte_none(ptent)) {
779 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800781 }
Hugh Dickins6f5e6b92006-03-16 23:04:09 -0800782
783 (*zap_work) -= PAGE_SIZE;
784
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800786 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -0800787
Linus Torvalds6aab3412005-11-28 14:34:23 -0800788 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 if (unlikely(details) && page) {
790 /*
791 * unmap_shared_mapping_pages() wants to
792 * invalidate cache without truncating:
793 * unmap shared but keep private pages.
794 */
795 if (details->check_mapping &&
796 details->check_mapping != page->mapping)
797 continue;
798 /*
799 * Each page->index must be checked when
800 * invalidating or truncating nonlinear.
801 */
802 if (details->nonlinear_vma &&
803 (page->index < details->first_index ||
804 page->index > details->last_index))
805 continue;
806 }
Nick Pigginb5810032005-10-29 18:16:12 -0700807 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -0700808 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 tlb_remove_tlb_entry(tlb, pte, addr);
810 if (unlikely(!page))
811 continue;
812 if (unlikely(details) && details->nonlinear_vma
813 && linear_page_index(details->nonlinear_vma,
814 addr) != page->index)
Nick Pigginb5810032005-10-29 18:16:12 -0700815 set_pte_at(mm, addr, pte,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 pgoff_to_pte(page->index));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 if (PageAnon(page))
Hugh Dickins86d912f2005-10-29 18:16:14 -0700818 anon_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700819 else {
820 if (pte_dirty(ptent))
821 set_page_dirty(page);
Johannes Weiner4917e5d2009-01-06 14:39:17 -0800822 if (pte_young(ptent) &&
823 likely(!VM_SequentialReadHint(vma)))
Nick Pigginbf3f3bc2009-01-06 14:38:55 -0800824 mark_page_accessed(page);
Hugh Dickins86d912f2005-10-29 18:16:14 -0700825 file_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700826 }
Hugh Dickinsedc315f2009-01-06 14:40:11 -0800827 page_remove_rmap(page);
Hugh Dickins3dc14742009-01-06 14:40:08 -0800828 if (unlikely(page_mapcount(page) < 0))
829 print_bad_pte(vma, addr, ptent, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 tlb_remove_page(tlb, page);
831 continue;
832 }
833 /*
834 * If details->check_mapping, we leave swap entries;
835 * if details->nonlinear_vma, we leave file entries.
836 */
837 if (unlikely(details))
838 continue;
Hugh Dickins2509ef22009-01-06 14:40:10 -0800839 if (pte_file(ptent)) {
840 if (unlikely(!(vma->vm_flags & VM_NONLINEAR)))
841 print_bad_pte(vma, addr, ptent, NULL);
842 } else if
843 (unlikely(!free_swap_and_cache(pte_to_swp_entry(ptent))))
844 print_bad_pte(vma, addr, ptent, NULL);
Zachary Amsden9888a1c2006-09-30 23:29:31 -0700845 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
Robin Holt51c6f662005-11-13 16:06:42 -0800846 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
Hugh Dickinsae859762005-10-29 18:16:05 -0700847
Hugh Dickins86d912f2005-10-29 18:16:14 -0700848 add_mm_rss(mm, file_rss, anon_rss);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700849 arch_leave_lazy_mmu_mode();
Hugh Dickins508034a2005-10-29 18:16:30 -0700850 pte_unmap_unlock(pte - 1, ptl);
Robin Holt51c6f662005-11-13 16:06:42 -0800851
852 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853}
854
Robin Holt51c6f662005-11-13 16:06:42 -0800855static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700856 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800858 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859{
860 pmd_t *pmd;
861 unsigned long next;
862
863 pmd = pmd_offset(pud, addr);
864 do {
865 next = pmd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800866 if (pmd_none_or_clear_bad(pmd)) {
867 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800869 }
870 next = zap_pte_range(tlb, vma, pmd, addr, next,
871 zap_work, details);
872 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
873
874 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875}
876
Robin Holt51c6f662005-11-13 16:06:42 -0800877static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700878 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800880 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881{
882 pud_t *pud;
883 unsigned long next;
884
885 pud = pud_offset(pgd, addr);
886 do {
887 next = pud_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800888 if (pud_none_or_clear_bad(pud)) {
889 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800891 }
892 next = zap_pmd_range(tlb, vma, pud, addr, next,
893 zap_work, details);
894 } while (pud++, addr = next, (addr != end && *zap_work > 0));
895
896 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897}
898
Robin Holt51c6f662005-11-13 16:06:42 -0800899static unsigned long unmap_page_range(struct mmu_gather *tlb,
900 struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800902 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903{
904 pgd_t *pgd;
905 unsigned long next;
906
907 if (details && !details->check_mapping && !details->nonlinear_vma)
908 details = NULL;
909
910 BUG_ON(addr >= end);
911 tlb_start_vma(tlb, vma);
912 pgd = pgd_offset(vma->vm_mm, addr);
913 do {
914 next = pgd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800915 if (pgd_none_or_clear_bad(pgd)) {
916 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800918 }
919 next = zap_pud_range(tlb, vma, pgd, addr, next,
920 zap_work, details);
921 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 tlb_end_vma(tlb, vma);
Robin Holt51c6f662005-11-13 16:06:42 -0800923
924 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925}
926
927#ifdef CONFIG_PREEMPT
928# define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
929#else
930/* No preempt: go for improved straight-line efficiency */
931# define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
932#endif
933
934/**
935 * unmap_vmas - unmap a range of memory covered by a list of vma's
936 * @tlbp: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 * @vma: the starting vma
938 * @start_addr: virtual address at which to start unmapping
939 * @end_addr: virtual address at which to end unmapping
940 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
941 * @details: details of nonlinear truncation or shared cache invalidation
942 *
Hugh Dickinsee39b372005-04-19 13:29:15 -0700943 * Returns the end address of the unmapping (restart addr if interrupted).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700945 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700947 * We aim to not hold locks for too long (for scheduling latency reasons).
948 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 * return the ending mmu_gather to the caller.
950 *
951 * Only addresses between `start' and `end' will be unmapped.
952 *
953 * The VMA list must be sorted in ascending virtual address order.
954 *
955 * unmap_vmas() assumes that the caller will flush the whole unmapped address
956 * range after unmap_vmas() returns. So the only responsibility here is to
957 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
958 * drops the lock and schedules.
959 */
Hugh Dickins508034a2005-10-29 18:16:30 -0700960unsigned long unmap_vmas(struct mmu_gather **tlbp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 struct vm_area_struct *vma, unsigned long start_addr,
962 unsigned long end_addr, unsigned long *nr_accounted,
963 struct zap_details *details)
964{
Robin Holt51c6f662005-11-13 16:06:42 -0800965 long zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
967 int tlb_start_valid = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -0700968 unsigned long start = start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700970 int fullmm = (*tlbp)->fullmm;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700971 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700973 mmu_notifier_invalidate_range_start(mm, start_addr, end_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 unsigned long end;
976
977 start = max(vma->vm_start, start_addr);
978 if (start >= vma->vm_end)
979 continue;
980 end = min(vma->vm_end, end_addr);
981 if (end <= vma->vm_start)
982 continue;
983
984 if (vma->vm_flags & VM_ACCOUNT)
985 *nr_accounted += (end - start) >> PAGE_SHIFT;
986
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800987 if (unlikely(is_pfn_mapping(vma)))
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -0800988 untrack_pfn_vma(vma, 0, 0);
989
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 while (start != end) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 if (!tlb_start_valid) {
992 tlb_start = start;
993 tlb_start_valid = 1;
994 }
995
Robin Holt51c6f662005-11-13 16:06:42 -0800996 if (unlikely(is_vm_hugetlb_page(vma))) {
Andi Kleena137e1c2008-07-23 21:27:43 -0700997 /*
998 * It is undesirable to test vma->vm_file as it
999 * should be non-null for valid hugetlb area.
1000 * However, vm_file will be NULL in the error
1001 * cleanup path of do_mmap_pgoff. When
1002 * hugetlbfs ->mmap method fails,
1003 * do_mmap_pgoff() nullifies vma->vm_file
1004 * before calling this function to clean up.
1005 * Since no pte has actually been setup, it is
1006 * safe to do nothing in this case.
1007 */
1008 if (vma->vm_file) {
1009 unmap_hugepage_range(vma, start, end, NULL);
1010 zap_work -= (end - start) /
Andi Kleena5516432008-07-23 21:27:41 -07001011 pages_per_huge_page(hstate_vma(vma));
Andi Kleena137e1c2008-07-23 21:27:43 -07001012 }
1013
Robin Holt51c6f662005-11-13 16:06:42 -08001014 start = end;
1015 } else
1016 start = unmap_page_range(*tlbp, vma,
1017 start, end, &zap_work, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018
Robin Holt51c6f662005-11-13 16:06:42 -08001019 if (zap_work > 0) {
1020 BUG_ON(start != end);
1021 break;
1022 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023
1024 tlb_finish_mmu(*tlbp, tlb_start, start);
1025
1026 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +01001027 (i_mmap_lock && spin_needbreak(i_mmap_lock))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 if (i_mmap_lock) {
Hugh Dickins508034a2005-10-29 18:16:30 -07001029 *tlbp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 goto out;
1031 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 }
1034
Hugh Dickins508034a2005-10-29 18:16:30 -07001035 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 tlb_start_valid = 0;
Robin Holt51c6f662005-11-13 16:06:42 -08001037 zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 }
1039 }
1040out:
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001041 mmu_notifier_invalidate_range_end(mm, start_addr, end_addr);
Hugh Dickinsee39b372005-04-19 13:29:15 -07001042 return start; /* which is now the end (or restart) address */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043}
1044
1045/**
1046 * zap_page_range - remove user pages in a given range
1047 * @vma: vm_area_struct holding the applicable pages
1048 * @address: starting address of pages to zap
1049 * @size: number of bytes to zap
1050 * @details: details of nonlinear truncation or shared cache invalidation
1051 */
Hugh Dickinsee39b372005-04-19 13:29:15 -07001052unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 unsigned long size, struct zap_details *details)
1054{
1055 struct mm_struct *mm = vma->vm_mm;
1056 struct mmu_gather *tlb;
1057 unsigned long end = address + size;
1058 unsigned long nr_accounted = 0;
1059
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001062 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -07001063 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
1064 if (tlb)
1065 tlb_finish_mmu(tlb, address, end);
Hugh Dickinsee39b372005-04-19 13:29:15 -07001066 return end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067}
1068
Jack Steinerc627f9c2008-07-29 22:33:53 -07001069/**
1070 * zap_vma_ptes - remove ptes mapping the vma
1071 * @vma: vm_area_struct holding ptes to be zapped
1072 * @address: starting address of pages to zap
1073 * @size: number of bytes to zap
1074 *
1075 * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1076 *
1077 * The entire address range must be fully contained within the vma.
1078 *
1079 * Returns 0 if successful.
1080 */
1081int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
1082 unsigned long size)
1083{
1084 if (address < vma->vm_start || address + size > vma->vm_end ||
1085 !(vma->vm_flags & VM_PFNMAP))
1086 return -1;
1087 zap_page_range(vma, address, size, NULL);
1088 return 0;
1089}
1090EXPORT_SYMBOL_GPL(zap_vma_ptes);
1091
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092/*
1093 * Do a quick page-table lookup for a single page.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 */
Linus Torvalds6aab3412005-11-28 14:34:23 -08001095struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001096 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097{
1098 pgd_t *pgd;
1099 pud_t *pud;
1100 pmd_t *pmd;
1101 pte_t *ptep, pte;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001102 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 struct page *page;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001104 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001106 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
1107 if (!IS_ERR(page)) {
1108 BUG_ON(flags & FOLL_GET);
1109 goto out;
1110 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001112 page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 pgd = pgd_offset(mm, address);
1114 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001115 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116
1117 pud = pud_offset(pgd, address);
Andi Kleenceb86872008-07-23 21:27:50 -07001118 if (pud_none(*pud))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001119 goto no_page_table;
Andi Kleenceb86872008-07-23 21:27:50 -07001120 if (pud_huge(*pud)) {
1121 BUG_ON(flags & FOLL_GET);
1122 page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE);
1123 goto out;
1124 }
1125 if (unlikely(pud_bad(*pud)))
1126 goto no_page_table;
1127
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 pmd = pmd_offset(pud, address);
Hugh Dickinsaeed5fc2008-05-06 20:49:23 +01001129 if (pmd_none(*pmd))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001130 goto no_page_table;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001131 if (pmd_huge(*pmd)) {
1132 BUG_ON(flags & FOLL_GET);
1133 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
1134 goto out;
1135 }
Hugh Dickinsaeed5fc2008-05-06 20:49:23 +01001136 if (unlikely(pmd_bad(*pmd)))
1137 goto no_page_table;
1138
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001139 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140
1141 pte = *ptep;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001142 if (!pte_present(pte))
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001143 goto no_page;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001144 if ((flags & FOLL_WRITE) && !pte_write(pte))
1145 goto unlock;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001146 page = vm_normal_page(vma, address, pte);
1147 if (unlikely(!page))
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001148 goto bad_page;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001149
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001150 if (flags & FOLL_GET)
1151 get_page(page);
1152 if (flags & FOLL_TOUCH) {
1153 if ((flags & FOLL_WRITE) &&
1154 !pte_dirty(pte) && !PageDirty(page))
1155 set_page_dirty(page);
KOSAKI Motohirobd775c42009-03-31 15:19:37 -07001156 /*
1157 * pte_mkyoung() would be more correct here, but atomic care
1158 * is needed to avoid losing the dirty bit: it is easier to use
1159 * mark_page_accessed().
1160 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001161 mark_page_accessed(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001163unlock:
1164 pte_unmap_unlock(ptep, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165out:
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001166 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001168bad_page:
1169 pte_unmap_unlock(ptep, ptl);
1170 return ERR_PTR(-EFAULT);
1171
1172no_page:
1173 pte_unmap_unlock(ptep, ptl);
1174 if (!pte_none(pte))
1175 return page;
1176 /* Fall through to ZERO_PAGE handling */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001177no_page_table:
1178 /*
1179 * When core dumping an enormous anonymous area that nobody
1180 * has touched so far, we don't want to allocate page tables.
1181 */
1182 if (flags & FOLL_ANON) {
Nick Piggin557ed1f2007-10-16 01:24:40 -07001183 page = ZERO_PAGE(0);
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001184 if (flags & FOLL_GET)
1185 get_page(page);
1186 BUG_ON(flags & FOLL_WRITE);
1187 }
1188 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189}
1190
Linus Torvalds672ca282008-06-23 11:21:37 -07001191/* Can we do the FOLL_ANON optimization? */
1192static inline int use_zero_page(struct vm_area_struct *vma)
1193{
1194 /*
1195 * We don't want to optimize FOLL_ANON for make_pages_present()
1196 * when it tries to page in a VM_LOCKED region. As to VM_SHARED,
1197 * we want to get the page from the page tables to make sure
1198 * that we serialize and update with any other user of that
1199 * mapping.
1200 */
1201 if (vma->vm_flags & (VM_LOCKED | VM_SHARED))
1202 return 0;
1203 /*
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001204 * And if we have a fault routine, it's not an anonymous region.
Linus Torvalds672ca282008-06-23 11:21:37 -07001205 */
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001206 return !vma->vm_ops || !vma->vm_ops->fault;
Linus Torvalds672ca282008-06-23 11:21:37 -07001207}
1208
Nick Pigginb291f002008-10-18 20:26:44 -07001209
1210
1211int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
Peter Zijlstra9d737772009-06-25 11:58:55 +02001212 unsigned long start, int nr_pages, int flags,
1213 struct page **pages, struct vm_area_struct **vmas)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214{
1215 int i;
Nick Pigginb291f002008-10-18 20:26:44 -07001216 unsigned int vm_flags = 0;
1217 int write = !!(flags & GUP_FLAGS_WRITE);
1218 int force = !!(flags & GUP_FLAGS_FORCE);
1219 int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS);
Ying Han47792802009-01-06 14:40:18 -08001220 int ignore_sigkill = !!(flags & GUP_FLAGS_IGNORE_SIGKILL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221
Peter Zijlstra9d737772009-06-25 11:58:55 +02001222 if (nr_pages <= 0)
Jonathan Corbet900cf082008-02-11 16:17:33 -07001223 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 /*
1225 * Require read or write permissions.
1226 * If 'force' is set, we only require the "MAY" flags.
1227 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001228 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
1229 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 i = 0;
1231
1232 do {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001233 struct vm_area_struct *vma;
1234 unsigned int foll_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235
1236 vma = find_extend_vma(mm, start);
1237 if (!vma && in_gate_area(tsk, start)) {
1238 unsigned long pg = start & PAGE_MASK;
1239 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
1240 pgd_t *pgd;
1241 pud_t *pud;
1242 pmd_t *pmd;
1243 pte_t *pte;
Nick Pigginb291f002008-10-18 20:26:44 -07001244
1245 /* user gate pages are read-only */
1246 if (!ignore && write)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 return i ? : -EFAULT;
1248 if (pg > TASK_SIZE)
1249 pgd = pgd_offset_k(pg);
1250 else
1251 pgd = pgd_offset_gate(mm, pg);
1252 BUG_ON(pgd_none(*pgd));
1253 pud = pud_offset(pgd, pg);
1254 BUG_ON(pud_none(*pud));
1255 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001256 if (pmd_none(*pmd))
1257 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001259 if (pte_none(*pte)) {
1260 pte_unmap(pte);
1261 return i ? : -EFAULT;
1262 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 if (pages) {
Nick Pigginfa2a4552005-11-29 18:43:17 +11001264 struct page *page = vm_normal_page(gate_vma, start, *pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001265 pages[i] = page;
1266 if (page)
1267 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 }
1269 pte_unmap(pte);
1270 if (vmas)
1271 vmas[i] = gate_vma;
1272 i++;
1273 start += PAGE_SIZE;
Peter Zijlstra9d737772009-06-25 11:58:55 +02001274 nr_pages--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 continue;
1276 }
1277
Nick Pigginb291f002008-10-18 20:26:44 -07001278 if (!vma ||
1279 (vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
1280 (!ignore && !(vm_flags & vma->vm_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 return i ? : -EFAULT;
1282
1283 if (is_vm_hugetlb_page(vma)) {
1284 i = follow_hugetlb_page(mm, vma, pages, vmas,
Peter Zijlstra9d737772009-06-25 11:58:55 +02001285 &start, &nr_pages, i, write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 continue;
1287 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001288
1289 foll_flags = FOLL_TOUCH;
1290 if (pages)
1291 foll_flags |= FOLL_GET;
Linus Torvalds672ca282008-06-23 11:21:37 -07001292 if (!write && use_zero_page(vma))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001293 foll_flags |= FOLL_ANON;
1294
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 do {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001296 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297
Ethan Solomita462e00c2007-07-15 23:38:16 -07001298 /*
Ying Han47792802009-01-06 14:40:18 -08001299 * If we have a pending SIGKILL, don't keep faulting
1300 * pages and potentially allocating memory, unless
1301 * current is handling munlock--e.g., on exit. In
1302 * that case, we are not allocating memory. Rather,
1303 * we're only unlocking already resident/mapped pages.
Ethan Solomita462e00c2007-07-15 23:38:16 -07001304 */
Ying Han47792802009-01-06 14:40:18 -08001305 if (unlikely(!ignore_sigkill &&
1306 fatal_signal_pending(current)))
1307 return i ? i : -ERESTARTSYS;
Ethan Solomita462e00c2007-07-15 23:38:16 -07001308
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001309 if (write)
1310 foll_flags |= FOLL_WRITE;
1311
1312 cond_resched();
Linus Torvalds6aab3412005-11-28 14:34:23 -08001313 while (!(page = follow_page(vma, start, foll_flags))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001314 int ret;
Linus Torvaldsd06063c2009-04-10 09:01:23 -07001315
Hugh Dickinsd26ed652009-06-23 13:52:49 +01001316 ret = handle_mm_fault(mm, vma, start,
1317 (foll_flags & FOLL_WRITE) ?
1318 FAULT_FLAG_WRITE : 0);
1319
Nick Piggin83c54072007-07-19 01:47:05 -07001320 if (ret & VM_FAULT_ERROR) {
1321 if (ret & VM_FAULT_OOM)
1322 return i ? i : -ENOMEM;
1323 else if (ret & VM_FAULT_SIGBUS)
1324 return i ? i : -EFAULT;
1325 BUG();
1326 }
1327 if (ret & VM_FAULT_MAJOR)
1328 tsk->maj_flt++;
1329 else
1330 tsk->min_flt++;
1331
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001332 /*
Nick Piggin83c54072007-07-19 01:47:05 -07001333 * The VM_FAULT_WRITE bit tells us that
1334 * do_wp_page has broken COW when necessary,
1335 * even if maybe_mkwrite decided not to set
1336 * pte_write. We can thus safely do subsequent
Hugh Dickins878b63a2009-01-06 14:39:32 -08001337 * page lookups as if they were reads. But only
1338 * do so when looping for pte_write is futile:
1339 * in some cases userspace may also be wanting
1340 * to write to the gotten user page, which a
1341 * read fault here might prevent (a readonly
1342 * page might get reCOWed by userspace write).
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001343 */
Hugh Dickins878b63a2009-01-06 14:39:32 -08001344 if ((ret & VM_FAULT_WRITE) &&
1345 !(vma->vm_flags & VM_WRITE))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001346 foll_flags &= ~FOLL_WRITE;
Nick Piggin83c54072007-07-19 01:47:05 -07001347
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07001348 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 }
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001350 if (IS_ERR(page))
1351 return i ? i : PTR_ERR(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001353 pages[i] = page;
James Bottomley03beb072006-03-26 01:36:57 -08001354
Russell Kinga6f36be2006-12-30 22:24:19 +00001355 flush_anon_page(vma, page, start);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001356 flush_dcache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 }
1358 if (vmas)
1359 vmas[i] = vma;
1360 i++;
1361 start += PAGE_SIZE;
Peter Zijlstra9d737772009-06-25 11:58:55 +02001362 nr_pages--;
1363 } while (nr_pages && start < vma->vm_end);
1364 } while (nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 return i;
1366}
Nick Pigginb291f002008-10-18 20:26:44 -07001367
Nick Piggind2bf6be2009-06-16 15:31:39 -07001368/**
1369 * get_user_pages() - pin user pages in memory
1370 * @tsk: task_struct of target task
1371 * @mm: mm_struct of target mm
1372 * @start: starting user address
Peter Zijlstra9d737772009-06-25 11:58:55 +02001373 * @nr_pages: number of pages from start to pin
Nick Piggind2bf6be2009-06-16 15:31:39 -07001374 * @write: whether pages will be written to by the caller
1375 * @force: whether to force write access even if user mapping is
1376 * readonly. This will result in the page being COWed even
1377 * in MAP_SHARED mappings. You do not want this.
1378 * @pages: array that receives pointers to the pages pinned.
1379 * Should be at least nr_pages long. Or NULL, if caller
1380 * only intends to ensure the pages are faulted in.
1381 * @vmas: array of pointers to vmas corresponding to each page.
1382 * Or NULL if the caller does not require them.
1383 *
1384 * Returns number of pages pinned. This may be fewer than the number
Peter Zijlstra9d737772009-06-25 11:58:55 +02001385 * requested. If nr_pages is 0 or negative, returns 0. If no pages
Nick Piggind2bf6be2009-06-16 15:31:39 -07001386 * were pinned, returns -errno. Each page returned must be released
1387 * with a put_page() call when it is finished with. vmas will only
1388 * remain valid while mmap_sem is held.
1389 *
1390 * Must be called with mmap_sem held for read or write.
1391 *
1392 * get_user_pages walks a process's page tables and takes a reference to
1393 * each struct page that each user address corresponds to at a given
1394 * instant. That is, it takes the page that would be accessed if a user
1395 * thread accesses the given user virtual address at that instant.
1396 *
1397 * This does not guarantee that the page exists in the user mappings when
1398 * get_user_pages returns, and there may even be a completely different
1399 * page there in some cases (eg. if mmapped pagecache has been invalidated
1400 * and subsequently re faulted). However it does guarantee that the page
1401 * won't be freed completely. And mostly callers simply care that the page
1402 * contains data that was valid *at some point in time*. Typically, an IO
1403 * or similar operation cannot guarantee anything stronger anyway because
1404 * locks can't be held over the syscall boundary.
1405 *
1406 * If write=0, the page must not be written to. If the page is written to,
1407 * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called
1408 * after the page is finished with, and before put_page is called.
1409 *
1410 * get_user_pages is typically used for fewer-copy IO operations, to get a
1411 * handle on the memory by some means other than accesses via the user virtual
1412 * addresses. The pages may be submitted for DMA to devices or accessed via
1413 * their kernel linear mapping (via the kmap APIs). Care should be taken to
1414 * use the correct cache flushing APIs.
1415 *
1416 * See also get_user_pages_fast, for performance critical applications.
1417 */
Nick Pigginb291f002008-10-18 20:26:44 -07001418int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
Peter Zijlstra9d737772009-06-25 11:58:55 +02001419 unsigned long start, int nr_pages, int write, int force,
Nick Pigginb291f002008-10-18 20:26:44 -07001420 struct page **pages, struct vm_area_struct **vmas)
1421{
1422 int flags = 0;
1423
1424 if (write)
1425 flags |= GUP_FLAGS_WRITE;
1426 if (force)
1427 flags |= GUP_FLAGS_FORCE;
1428
Peter Zijlstra9d737772009-06-25 11:58:55 +02001429 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas);
Nick Pigginb291f002008-10-18 20:26:44 -07001430}
1431
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432EXPORT_SYMBOL(get_user_pages);
1433
Harvey Harrison920c7a52008-02-04 22:29:26 -08001434pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr,
1435 spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001436{
1437 pgd_t * pgd = pgd_offset(mm, addr);
1438 pud_t * pud = pud_alloc(mm, pgd, addr);
1439 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001440 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001441 if (pmd)
1442 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1443 }
1444 return NULL;
1445}
1446
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001448 * This is the old fallback for page remapping.
1449 *
1450 * For historical reasons, it only allows reserved pages. Only
1451 * old drivers should use this, and they needed to mark their
1452 * pages reserved for the old functions anyway.
1453 */
Nick Piggin423bad62008-04-28 02:13:01 -07001454static int insert_page(struct vm_area_struct *vma, unsigned long addr,
1455 struct page *page, pgprot_t prot)
Linus Torvalds238f58d2005-11-29 13:01:56 -08001456{
Nick Piggin423bad62008-04-28 02:13:01 -07001457 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001458 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001459 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001460 spinlock_t *ptl;
1461
Linus Torvalds238f58d2005-11-29 13:01:56 -08001462 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001463 if (PageAnon(page))
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001464 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001465 retval = -ENOMEM;
1466 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001467 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001468 if (!pte)
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001469 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001470 retval = -EBUSY;
1471 if (!pte_none(*pte))
1472 goto out_unlock;
1473
1474 /* Ok, finally just insert the thing.. */
1475 get_page(page);
1476 inc_mm_counter(mm, file_rss);
1477 page_add_file_rmap(page);
1478 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1479
1480 retval = 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001481 pte_unmap_unlock(pte, ptl);
1482 return retval;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001483out_unlock:
1484 pte_unmap_unlock(pte, ptl);
1485out:
1486 return retval;
1487}
1488
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001489/**
1490 * vm_insert_page - insert single page into user vma
1491 * @vma: user vma to map to
1492 * @addr: target user address of this page
1493 * @page: source kernel page
1494 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001495 * This allows drivers to insert individual pages they've allocated
1496 * into a user vma.
1497 *
1498 * The page has to be a nice clean _individual_ kernel allocation.
1499 * If you allocate a compound page, you need to have marked it as
1500 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001501 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001502 *
1503 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1504 * took an arbitrary page protection parameter. This doesn't allow
1505 * that. Your vma protection will have to be set up correctly, which
1506 * means that if you want a shared writable mapping, you'd better
1507 * ask for a shared writable mapping!
1508 *
1509 * The page does not need to be reserved.
1510 */
Nick Piggin423bad62008-04-28 02:13:01 -07001511int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
1512 struct page *page)
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001513{
1514 if (addr < vma->vm_start || addr >= vma->vm_end)
1515 return -EFAULT;
1516 if (!page_count(page))
1517 return -EINVAL;
Linus Torvalds4d7672b2005-12-16 10:21:23 -08001518 vma->vm_flags |= VM_INSERTPAGE;
Nick Piggin423bad62008-04-28 02:13:01 -07001519 return insert_page(vma, addr, page, vma->vm_page_prot);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001520}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001521EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001522
Nick Piggin423bad62008-04-28 02:13:01 -07001523static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1524 unsigned long pfn, pgprot_t prot)
1525{
1526 struct mm_struct *mm = vma->vm_mm;
1527 int retval;
1528 pte_t *pte, entry;
1529 spinlock_t *ptl;
1530
1531 retval = -ENOMEM;
1532 pte = get_locked_pte(mm, addr, &ptl);
1533 if (!pte)
1534 goto out;
1535 retval = -EBUSY;
1536 if (!pte_none(*pte))
1537 goto out_unlock;
1538
1539 /* Ok, finally just insert the thing.. */
1540 entry = pte_mkspecial(pfn_pte(pfn, prot));
1541 set_pte_at(mm, addr, pte, entry);
1542 update_mmu_cache(vma, addr, entry); /* XXX: why not for insert_page? */
1543
1544 retval = 0;
1545out_unlock:
1546 pte_unmap_unlock(pte, ptl);
1547out:
1548 return retval;
1549}
1550
Nick Piggine0dc0d82007-02-12 00:51:36 -08001551/**
1552 * vm_insert_pfn - insert single pfn into user vma
1553 * @vma: user vma to map to
1554 * @addr: target user address of this page
1555 * @pfn: source kernel pfn
1556 *
1557 * Similar to vm_inert_page, this allows drivers to insert individual pages
1558 * they've allocated into a user vma. Same comments apply.
1559 *
1560 * This function should only be called from a vm_ops->fault handler, and
1561 * in that case the handler should return NULL.
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001562 *
1563 * vma cannot be a COW mapping.
1564 *
1565 * As this is called only for pages that do not currently exist, we
1566 * do not need to flush old virtual caches or the TLB.
Nick Piggine0dc0d82007-02-12 00:51:36 -08001567 */
1568int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
Nick Piggin423bad62008-04-28 02:13:01 -07001569 unsigned long pfn)
Nick Piggine0dc0d82007-02-12 00:51:36 -08001570{
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001571 int ret;
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001572 pgprot_t pgprot = vma->vm_page_prot;
Nick Piggin7e675132008-04-28 02:13:00 -07001573 /*
1574 * Technically, architectures with pte_special can avoid all these
1575 * restrictions (same for remap_pfn_range). However we would like
1576 * consistency in testing and feature parity among all, so we should
1577 * try to keep these invariants in place for everybody.
1578 */
Jared Hulbertb379d792008-04-28 02:12:58 -07001579 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
1580 BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
1581 (VM_PFNMAP|VM_MIXEDMAP));
1582 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1583 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
Nick Piggine0dc0d82007-02-12 00:51:36 -08001584
Nick Piggin423bad62008-04-28 02:13:01 -07001585 if (addr < vma->vm_start || addr >= vma->vm_end)
1586 return -EFAULT;
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001587 if (track_pfn_vma_new(vma, &pgprot, pfn, PAGE_SIZE))
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001588 return -EINVAL;
1589
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001590 ret = insert_pfn(vma, addr, pfn, pgprot);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001591
1592 if (ret)
1593 untrack_pfn_vma(vma, pfn, PAGE_SIZE);
1594
1595 return ret;
Nick Piggine0dc0d82007-02-12 00:51:36 -08001596}
1597EXPORT_SYMBOL(vm_insert_pfn);
1598
Nick Piggin423bad62008-04-28 02:13:01 -07001599int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
1600 unsigned long pfn)
1601{
1602 BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
1603
1604 if (addr < vma->vm_start || addr >= vma->vm_end)
1605 return -EFAULT;
1606
1607 /*
1608 * If we don't have pte special, then we have to use the pfn_valid()
1609 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1610 * refcount the page if pfn_valid is true (hence insert_page rather
1611 * than insert_pfn).
1612 */
1613 if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) {
1614 struct page *page;
1615
1616 page = pfn_to_page(pfn);
1617 return insert_page(vma, addr, page, vma->vm_page_prot);
1618 }
1619 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
1620}
1621EXPORT_SYMBOL(vm_insert_mixed);
1622
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001623/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 * maps a range of physical memory into the requested pages. the old
1625 * mappings are removed. any references to nonexistent pages results
1626 * in null mappings (currently treated as "copy-on-access")
1627 */
1628static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1629 unsigned long addr, unsigned long end,
1630 unsigned long pfn, pgprot_t prot)
1631{
1632 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001633 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634
Hugh Dickinsc74df322005-10-29 18:16:23 -07001635 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 if (!pte)
1637 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001638 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 do {
1640 BUG_ON(!pte_none(*pte));
Nick Piggin7e675132008-04-28 02:13:00 -07001641 set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 pfn++;
1643 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001644 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001645 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 return 0;
1647}
1648
1649static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1650 unsigned long addr, unsigned long end,
1651 unsigned long pfn, pgprot_t prot)
1652{
1653 pmd_t *pmd;
1654 unsigned long next;
1655
1656 pfn -= addr >> PAGE_SHIFT;
1657 pmd = pmd_alloc(mm, pud, addr);
1658 if (!pmd)
1659 return -ENOMEM;
1660 do {
1661 next = pmd_addr_end(addr, end);
1662 if (remap_pte_range(mm, pmd, addr, next,
1663 pfn + (addr >> PAGE_SHIFT), prot))
1664 return -ENOMEM;
1665 } while (pmd++, addr = next, addr != end);
1666 return 0;
1667}
1668
1669static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1670 unsigned long addr, unsigned long end,
1671 unsigned long pfn, pgprot_t prot)
1672{
1673 pud_t *pud;
1674 unsigned long next;
1675
1676 pfn -= addr >> PAGE_SHIFT;
1677 pud = pud_alloc(mm, pgd, addr);
1678 if (!pud)
1679 return -ENOMEM;
1680 do {
1681 next = pud_addr_end(addr, end);
1682 if (remap_pmd_range(mm, pud, addr, next,
1683 pfn + (addr >> PAGE_SHIFT), prot))
1684 return -ENOMEM;
1685 } while (pud++, addr = next, addr != end);
1686 return 0;
1687}
1688
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001689/**
1690 * remap_pfn_range - remap kernel memory to userspace
1691 * @vma: user vma to map to
1692 * @addr: target user address to start at
1693 * @pfn: physical address of kernel memory
1694 * @size: size of map area
1695 * @prot: page protection flags for this mapping
1696 *
1697 * Note: this is only safe if the mm semaphore is held when called.
1698 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1700 unsigned long pfn, unsigned long size, pgprot_t prot)
1701{
1702 pgd_t *pgd;
1703 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001704 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 struct mm_struct *mm = vma->vm_mm;
1706 int err;
1707
1708 /*
1709 * Physically remapped pages are special. Tell the
1710 * rest of the world about it:
1711 * VM_IO tells people not to look at these pages
1712 * (accesses can have side effects).
Hugh Dickins0b14c172005-11-21 21:32:15 -08001713 * VM_RESERVED is specified all over the place, because
1714 * in 2.4 it kept swapout's vma scan off this vma; but
1715 * in 2.6 the LRU scan won't even find its pages, so this
1716 * flag means no more than count its pages in reserved_vm,
1717 * and omit it from core dump, even when VM_IO turned off.
Linus Torvalds6aab3412005-11-28 14:34:23 -08001718 * VM_PFNMAP tells the core MM that the base pages are just
1719 * raw PFN mappings, and do not have a "struct page" associated
1720 * with them.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001721 *
1722 * There's a horrible special case to handle copy-on-write
1723 * behaviour that some programs depend on. We mark the "original"
1724 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 */
Pallipadi, Venkatesh4bb9c5c2009-03-12 17:45:27 -07001726 if (addr == vma->vm_start && end == vma->vm_end) {
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001727 vma->vm_pgoff = pfn;
Pallipadi, Venkatesh895791d2009-03-13 16:35:44 -07001728 vma->vm_flags |= VM_PFN_AT_MMAP;
Pallipadi, Venkatesh4bb9c5c2009-03-12 17:45:27 -07001729 } else if (is_cow_mapping(vma->vm_flags))
venkatesh.pallipadi@intel.com3c8bb732008-12-18 11:41:27 -08001730 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001731
Linus Torvalds6aab3412005-11-28 14:34:23 -08001732 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001734 err = track_pfn_vma_new(vma, &prot, pfn, PAGE_ALIGN(size));
venkatesh.pallipadi@intel.coma3670612009-01-09 16:13:09 -08001735 if (err) {
1736 /*
1737 * To indicate that track_pfn related cleanup is not
1738 * needed from higher level routine calling unmap_vmas
1739 */
1740 vma->vm_flags &= ~(VM_IO | VM_RESERVED | VM_PFNMAP);
Pallipadi, Venkatesh895791d2009-03-13 16:35:44 -07001741 vma->vm_flags &= ~VM_PFN_AT_MMAP;
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001742 return -EINVAL;
venkatesh.pallipadi@intel.coma3670612009-01-09 16:13:09 -08001743 }
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001744
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 BUG_ON(addr >= end);
1746 pfn -= addr >> PAGE_SHIFT;
1747 pgd = pgd_offset(mm, addr);
1748 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 do {
1750 next = pgd_addr_end(addr, end);
1751 err = remap_pud_range(mm, pgd, addr, next,
1752 pfn + (addr >> PAGE_SHIFT), prot);
1753 if (err)
1754 break;
1755 } while (pgd++, addr = next, addr != end);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001756
1757 if (err)
1758 untrack_pfn_vma(vma, pfn, PAGE_ALIGN(size));
1759
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 return err;
1761}
1762EXPORT_SYMBOL(remap_pfn_range);
1763
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001764static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1765 unsigned long addr, unsigned long end,
1766 pte_fn_t fn, void *data)
1767{
1768 pte_t *pte;
1769 int err;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001770 pgtable_t token;
Borislav Petkov94909912007-05-06 14:49:17 -07001771 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001772
1773 pte = (mm == &init_mm) ?
1774 pte_alloc_kernel(pmd, addr) :
1775 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1776 if (!pte)
1777 return -ENOMEM;
1778
1779 BUG_ON(pmd_huge(*pmd));
1780
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08001781 arch_enter_lazy_mmu_mode();
1782
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001783 token = pmd_pgtable(*pmd);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001784
1785 do {
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001786 err = fn(pte, token, addr, data);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001787 if (err)
1788 break;
1789 } while (pte++, addr += PAGE_SIZE, addr != end);
1790
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08001791 arch_leave_lazy_mmu_mode();
1792
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001793 if (mm != &init_mm)
1794 pte_unmap_unlock(pte-1, ptl);
1795 return err;
1796}
1797
1798static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1799 unsigned long addr, unsigned long end,
1800 pte_fn_t fn, void *data)
1801{
1802 pmd_t *pmd;
1803 unsigned long next;
1804 int err;
1805
Andi Kleenceb86872008-07-23 21:27:50 -07001806 BUG_ON(pud_huge(*pud));
1807
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001808 pmd = pmd_alloc(mm, pud, addr);
1809 if (!pmd)
1810 return -ENOMEM;
1811 do {
1812 next = pmd_addr_end(addr, end);
1813 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1814 if (err)
1815 break;
1816 } while (pmd++, addr = next, addr != end);
1817 return err;
1818}
1819
1820static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1821 unsigned long addr, unsigned long end,
1822 pte_fn_t fn, void *data)
1823{
1824 pud_t *pud;
1825 unsigned long next;
1826 int err;
1827
1828 pud = pud_alloc(mm, pgd, addr);
1829 if (!pud)
1830 return -ENOMEM;
1831 do {
1832 next = pud_addr_end(addr, end);
1833 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1834 if (err)
1835 break;
1836 } while (pud++, addr = next, addr != end);
1837 return err;
1838}
1839
1840/*
1841 * Scan a region of virtual memory, filling in page tables as necessary
1842 * and calling a provided function on each leaf page table.
1843 */
1844int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1845 unsigned long size, pte_fn_t fn, void *data)
1846{
1847 pgd_t *pgd;
1848 unsigned long next;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001849 unsigned long start = addr, end = addr + size;
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001850 int err;
1851
1852 BUG_ON(addr >= end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001853 mmu_notifier_invalidate_range_start(mm, start, end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001854 pgd = pgd_offset(mm, addr);
1855 do {
1856 next = pgd_addr_end(addr, end);
1857 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1858 if (err)
1859 break;
1860 } while (pgd++, addr = next, addr != end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001861 mmu_notifier_invalidate_range_end(mm, start, end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001862 return err;
1863}
1864EXPORT_SYMBOL_GPL(apply_to_page_range);
1865
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001867 * handle_pte_fault chooses page fault handler according to an entry
1868 * which was read non-atomically. Before making any commitment, on
1869 * those architectures or configurations (e.g. i386 with PAE) which
1870 * might give a mix of unmatched parts, do_swap_page and do_file_page
1871 * must check under lock before unmapping the pte and proceeding
1872 * (but do_wp_page is only called after already making such a check;
1873 * and do_anonymous_page and do_no_page can safely check later on).
1874 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001875static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001876 pte_t *page_table, pte_t orig_pte)
1877{
1878 int same = 1;
1879#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1880 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001881 spinlock_t *ptl = pte_lockptr(mm, pmd);
1882 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001883 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001884 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001885 }
1886#endif
1887 pte_unmap(page_table);
1888 return same;
1889}
1890
1891/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1893 * servicing faults for write access. In the normal case, do always want
1894 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1895 * that do not have writing enabled, when used by access_process_vm.
1896 */
1897static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1898{
1899 if (likely(vma->vm_flags & VM_WRITE))
1900 pte = pte_mkwrite(pte);
1901 return pte;
1902}
1903
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001904static 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 -08001905{
1906 /*
1907 * If the source page was a PFN mapping, we don't have
1908 * a "struct page" for it. We do a best-effort copy by
1909 * just copying from the original user address. If that
1910 * fails, we just zero-fill it. Live with it.
1911 */
1912 if (unlikely(!src)) {
1913 void *kaddr = kmap_atomic(dst, KM_USER0);
Linus Torvalds5d2a2db2005-11-29 14:07:55 -08001914 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1915
1916 /*
1917 * This really shouldn't fail, because the page is there
1918 * in the page tables. But it might just be unreadable,
1919 * in which case we just give up and fill the result with
1920 * zeroes.
1921 */
1922 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Linus Torvalds6aab3412005-11-28 14:34:23 -08001923 memset(kaddr, 0, PAGE_SIZE);
1924 kunmap_atomic(kaddr, KM_USER0);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07001925 flush_dcache_page(dst);
Nick Piggin0ed361d2008-02-04 22:29:34 -08001926 } else
1927 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001928}
1929
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 * This routine handles present pages, when users try to write
1932 * to a shared page. It is done by copying the page to a new address
1933 * and decrementing the shared-page counter for the old page.
1934 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 * Note that this routine assumes that the protection checks have been
1936 * done by the caller (the low-level page fault routine in most cases).
1937 * Thus we can safely just mark it writable once we've done any necessary
1938 * COW.
1939 *
1940 * We also mark the page dirty at this point even though the page will
1941 * change only once the write actually happens. This avoids a few races,
1942 * and potentially makes it more efficient.
1943 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001944 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1945 * but allow concurrent faults), with pte both mapped and locked.
1946 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001948static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1949 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001950 spinlock_t *ptl, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951{
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001952 struct page *old_page, *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 pte_t entry;
Nick Piggin83c54072007-07-19 01:47:05 -07001954 int reuse = 0, ret = 0;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001955 int page_mkwrite = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001956 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957
Linus Torvalds6aab3412005-11-28 14:34:23 -08001958 old_page = vm_normal_page(vma, address, orig_pte);
Peter Zijlstra251b97f2008-07-04 09:59:24 -07001959 if (!old_page) {
1960 /*
1961 * VM_MIXEDMAP !pfn_valid() case
1962 *
1963 * We should not cow pages in a shared writeable mapping.
1964 * Just mark the pages writable as we can't do any dirty
1965 * accounting on raw pfn maps.
1966 */
1967 if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
1968 (VM_WRITE|VM_SHARED))
1969 goto reuse;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001970 goto gotten;
Peter Zijlstra251b97f2008-07-04 09:59:24 -07001971 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001973 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001974 * Take out anonymous pages first, anonymous shared vmas are
1975 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001976 */
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001977 if (PageAnon(old_page)) {
Hugh Dickinsab967d82009-01-06 14:39:33 -08001978 if (!trylock_page(old_page)) {
1979 page_cache_get(old_page);
1980 pte_unmap_unlock(page_table, ptl);
1981 lock_page(old_page);
1982 page_table = pte_offset_map_lock(mm, pmd, address,
1983 &ptl);
1984 if (!pte_same(*page_table, orig_pte)) {
1985 unlock_page(old_page);
1986 page_cache_release(old_page);
1987 goto unlock;
1988 }
1989 page_cache_release(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001990 }
Hugh Dickins7b1fe592009-01-06 14:39:34 -08001991 reuse = reuse_swap_page(old_page);
Hugh Dickinsab967d82009-01-06 14:39:33 -08001992 unlock_page(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001993 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001994 (VM_WRITE|VM_SHARED))) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001995 /*
1996 * Only catch write-faults on shared writable pages,
1997 * read-only shared pages can get COWed by
1998 * get_user_pages(.write=1, .force=1).
1999 */
David Howells9637a5e2006-06-23 02:03:43 -07002000 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
Nick Pigginc2ec1752009-03-31 15:23:21 -07002001 struct vm_fault vmf;
2002 int tmp;
2003
2004 vmf.virtual_address = (void __user *)(address &
2005 PAGE_MASK);
2006 vmf.pgoff = old_page->index;
2007 vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
2008 vmf.page = old_page;
2009
David Howells9637a5e2006-06-23 02:03:43 -07002010 /*
2011 * Notify the address space that the page is about to
2012 * become writable so that it can prohibit this or wait
2013 * for the page to get into an appropriate state.
2014 *
2015 * We do this without the lock held, so that it can
2016 * sleep if it needs to.
2017 */
2018 page_cache_get(old_page);
2019 pte_unmap_unlock(page_table, ptl);
2020
Nick Pigginc2ec1752009-03-31 15:23:21 -07002021 tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
2022 if (unlikely(tmp &
2023 (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
2024 ret = tmp;
David Howells9637a5e2006-06-23 02:03:43 -07002025 goto unwritable_page;
Nick Pigginc2ec1752009-03-31 15:23:21 -07002026 }
Nick Pigginb827e492009-04-30 15:08:16 -07002027 if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
2028 lock_page(old_page);
2029 if (!old_page->mapping) {
2030 ret = 0; /* retry the fault */
2031 unlock_page(old_page);
2032 goto unwritable_page;
2033 }
2034 } else
2035 VM_BUG_ON(!PageLocked(old_page));
David Howells9637a5e2006-06-23 02:03:43 -07002036
David Howells9637a5e2006-06-23 02:03:43 -07002037 /*
2038 * Since we dropped the lock we need to revalidate
2039 * the PTE as someone else may have changed it. If
2040 * they did, we just return, as we can count on the
2041 * MMU to tell us if they didn't also make it writable.
2042 */
2043 page_table = pte_offset_map_lock(mm, pmd, address,
2044 &ptl);
Nick Pigginb827e492009-04-30 15:08:16 -07002045 if (!pte_same(*page_table, orig_pte)) {
2046 unlock_page(old_page);
2047 page_cache_release(old_page);
David Howells9637a5e2006-06-23 02:03:43 -07002048 goto unlock;
Nick Pigginb827e492009-04-30 15:08:16 -07002049 }
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002050
2051 page_mkwrite = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002053 dirty_page = old_page;
2054 get_page(dirty_page);
David Howells9637a5e2006-06-23 02:03:43 -07002055 reuse = 1;
David Howells9637a5e2006-06-23 02:03:43 -07002056 }
2057
2058 if (reuse) {
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002059reuse:
David Howells9637a5e2006-06-23 02:03:43 -07002060 flush_cache_page(vma, address, pte_pfn(orig_pte));
2061 entry = pte_mkyoung(orig_pte);
2062 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
KAMEZAWA Hiroyuki954ffcb2007-10-16 01:25:44 -07002063 if (ptep_set_access_flags(vma, address, page_table, entry,1))
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002064 update_mmu_cache(vma, address, entry);
David Howells9637a5e2006-06-23 02:03:43 -07002065 ret |= VM_FAULT_WRITE;
2066 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068
2069 /*
2070 * Ok, we need to copy. Oh, well..
2071 */
Nick Pigginb5810032005-10-29 18:16:12 -07002072 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08002073gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002074 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075
2076 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07002077 goto oom;
Nick Piggin557ed1f2007-10-16 01:24:40 -07002078 VM_BUG_ON(old_page == ZERO_PAGE(0));
2079 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
2080 if (!new_page)
2081 goto oom;
Nick Pigginb291f002008-10-18 20:26:44 -07002082 /*
2083 * Don't let another task, with possibly unlocked vma,
2084 * keep the mlocked page.
2085 */
Carsten Otteab926612009-02-04 15:12:16 -08002086 if ((vma->vm_flags & VM_LOCKED) && old_page) {
Nick Pigginb291f002008-10-18 20:26:44 -07002087 lock_page(old_page); /* for LRU manipulation */
2088 clear_page_mlock(old_page);
2089 unlock_page(old_page);
2090 }
Nick Piggin557ed1f2007-10-16 01:24:40 -07002091 cow_user_page(new_page, old_page, address, vma);
Nick Piggin0ed361d2008-02-04 22:29:34 -08002092 __SetPageUptodate(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002093
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002094 if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002095 goto oom_free_new;
2096
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 /*
2098 * Re-check the pte - we dropped the lock
2099 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002100 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07002101 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08002102 if (old_page) {
Hugh Dickins920fc352005-11-21 21:32:17 -08002103 if (!PageAnon(old_page)) {
2104 dec_mm_counter(mm, file_rss);
2105 inc_mm_counter(mm, anon_rss);
2106 }
2107 } else
Hugh Dickins42946212005-10-29 18:16:05 -07002108 inc_mm_counter(mm, anon_rss);
Ben Collinseca35132005-11-29 11:45:26 -08002109 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07002110 entry = mk_pte(new_page, vma->vm_page_prot);
2111 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Siddha, Suresh B4ce072f2006-09-29 01:58:42 -07002112 /*
2113 * Clear the pte entry and flush it first, before updating the
2114 * pte with the new entry. This will avoid a race condition
2115 * seen in the presence of one thread doing SMC and another
2116 * thread doing COW.
2117 */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002118 ptep_clear_flush_notify(vma, address, page_table);
Nick Piggin9617d952006-01-06 00:11:12 -08002119 page_add_new_anon_rmap(new_page, vma, address);
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002120 set_pte_at(mm, address, page_table, entry);
2121 update_mmu_cache(vma, address, entry);
Nick Piggin945754a2008-06-23 14:30:30 +02002122 if (old_page) {
2123 /*
2124 * Only after switching the pte to the new page may
2125 * we remove the mapcount here. Otherwise another
2126 * process may come and find the rmap count decremented
2127 * before the pte is switched to the new page, and
2128 * "reuse" the old page writing into it while our pte
2129 * here still points into it and can be read by other
2130 * threads.
2131 *
2132 * The critical issue is to order this
2133 * page_remove_rmap with the ptp_clear_flush above.
2134 * Those stores are ordered by (if nothing else,)
2135 * the barrier present in the atomic_add_negative
2136 * in page_remove_rmap.
2137 *
2138 * Then the TLB flush in ptep_clear_flush ensures that
2139 * no process can access the old page before the
2140 * decremented mapcount is visible. And the old page
2141 * cannot be reused until after the decremented
2142 * mapcount is visible. So transitively, TLBs to
2143 * old page will be flushed before it can be reused.
2144 */
Hugh Dickinsedc315f2009-01-06 14:40:11 -08002145 page_remove_rmap(old_page);
Nick Piggin945754a2008-06-23 14:30:30 +02002146 }
2147
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 /* Free the old page.. */
2149 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10002150 ret |= VM_FAULT_WRITE;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002151 } else
2152 mem_cgroup_uncharge_page(new_page);
2153
Hugh Dickins920fc352005-11-21 21:32:17 -08002154 if (new_page)
2155 page_cache_release(new_page);
2156 if (old_page)
2157 page_cache_release(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002158unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002159 pte_unmap_unlock(page_table, ptl);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002160 if (dirty_page) {
Nick Piggin79352892007-07-19 01:47:22 -07002161 /*
2162 * Yes, Virginia, this is actually required to prevent a race
2163 * with clear_page_dirty_for_io() from clearing the page dirty
2164 * bit after it clear all dirty ptes, but before a racing
2165 * do_wp_page installs a dirty pte.
2166 *
2167 * do_no_page is protected similarly.
2168 */
Nick Pigginb827e492009-04-30 15:08:16 -07002169 if (!page_mkwrite) {
2170 wait_on_page_locked(dirty_page);
2171 set_page_dirty_balance(dirty_page, page_mkwrite);
2172 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002173 put_page(dirty_page);
Nick Pigginb827e492009-04-30 15:08:16 -07002174 if (page_mkwrite) {
2175 struct address_space *mapping = dirty_page->mapping;
2176
2177 set_page_dirty(dirty_page);
2178 unlock_page(dirty_page);
2179 page_cache_release(dirty_page);
2180 if (mapping) {
2181 /*
2182 * Some device drivers do not set page.mapping
2183 * but still dirty their pages
2184 */
2185 balance_dirty_pages_ratelimited(mapping);
2186 }
2187 }
2188
2189 /* file_update_time outside page_lock */
2190 if (vma->vm_file)
2191 file_update_time(vma->vm_file);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002192 }
Nick Pigginf33ea7f2005-08-03 20:24:01 +10002193 return ret;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002194oom_free_new:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002195 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002196oom:
Nick Pigginb827e492009-04-30 15:08:16 -07002197 if (old_page) {
2198 if (page_mkwrite) {
2199 unlock_page(old_page);
2200 page_cache_release(old_page);
2201 }
Hugh Dickins920fc352005-11-21 21:32:17 -08002202 page_cache_release(old_page);
Nick Pigginb827e492009-04-30 15:08:16 -07002203 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 return VM_FAULT_OOM;
David Howells9637a5e2006-06-23 02:03:43 -07002205
2206unwritable_page:
2207 page_cache_release(old_page);
Nick Pigginc2ec1752009-03-31 15:23:21 -07002208 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209}
2210
2211/*
2212 * Helper functions for unmap_mapping_range().
2213 *
2214 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
2215 *
2216 * We have to restart searching the prio_tree whenever we drop the lock,
2217 * since the iterator is only valid while the lock is held, and anyway
2218 * a later vma might be split and reinserted earlier while lock dropped.
2219 *
2220 * The list of nonlinear vmas could be handled more efficiently, using
2221 * a placeholder, but handle it in the same way until a need is shown.
2222 * It is important to search the prio_tree before nonlinear list: a vma
2223 * may become nonlinear and be shifted from prio_tree to nonlinear list
2224 * while the lock is dropped; but never shifted from list to prio_tree.
2225 *
2226 * In order to make forward progress despite restarting the search,
2227 * vm_truncate_count is used to mark a vma as now dealt with, so we can
2228 * quickly skip it next time around. Since the prio_tree search only
2229 * shows us those vmas affected by unmapping the range in question, we
2230 * can't efficiently keep all vmas in step with mapping->truncate_count:
2231 * so instead reset them all whenever it wraps back to 0 (then go to 1).
2232 * mapping->truncate_count and vma->vm_truncate_count are protected by
2233 * i_mmap_lock.
2234 *
2235 * In order to make forward progress despite repeatedly restarting some
Hugh Dickinsee39b372005-04-19 13:29:15 -07002236 * large vma, note the restart_addr from unmap_vmas when it breaks out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 * and restart from that address when we reach that vma again. It might
2238 * have been split or merged, shrunk or extended, but never shifted: so
2239 * restart_addr remains valid so long as it remains in the vma's range.
2240 * unmap_mapping_range forces truncate_count to leap over page-aligned
2241 * values so we can save vma's restart_addr in its truncate_count field.
2242 */
2243#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
2244
2245static void reset_vma_truncate_counts(struct address_space *mapping)
2246{
2247 struct vm_area_struct *vma;
2248 struct prio_tree_iter iter;
2249
2250 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
2251 vma->vm_truncate_count = 0;
2252 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
2253 vma->vm_truncate_count = 0;
2254}
2255
2256static int unmap_mapping_range_vma(struct vm_area_struct *vma,
2257 unsigned long start_addr, unsigned long end_addr,
2258 struct zap_details *details)
2259{
2260 unsigned long restart_addr;
2261 int need_break;
2262
Nick Piggind00806b2007-07-19 01:46:57 -07002263 /*
2264 * files that support invalidating or truncating portions of the
Nick Piggind0217ac2007-07-19 01:47:03 -07002265 * file from under mmaped areas must have their ->fault function
Nick Piggin83c54072007-07-19 01:47:05 -07002266 * return a locked page (and set VM_FAULT_LOCKED in the return).
2267 * This provides synchronisation against concurrent unmapping here.
Nick Piggind00806b2007-07-19 01:46:57 -07002268 */
Nick Piggind00806b2007-07-19 01:46:57 -07002269
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270again:
2271 restart_addr = vma->vm_truncate_count;
2272 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
2273 start_addr = restart_addr;
2274 if (start_addr >= end_addr) {
2275 /* Top of vma has been split off since last time */
2276 vma->vm_truncate_count = details->truncate_count;
2277 return 0;
2278 }
2279 }
2280
Hugh Dickinsee39b372005-04-19 13:29:15 -07002281 restart_addr = zap_page_range(vma, start_addr,
2282 end_addr - start_addr, details);
Nick Piggin95c354f2008-01-30 13:31:20 +01002283 need_break = need_resched() || spin_needbreak(details->i_mmap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284
Hugh Dickinsee39b372005-04-19 13:29:15 -07002285 if (restart_addr >= end_addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 /* We have now completed this vma: mark it so */
2287 vma->vm_truncate_count = details->truncate_count;
2288 if (!need_break)
2289 return 0;
2290 } else {
2291 /* Note restart_addr in vma's truncate_count field */
Hugh Dickinsee39b372005-04-19 13:29:15 -07002292 vma->vm_truncate_count = restart_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 if (!need_break)
2294 goto again;
2295 }
2296
2297 spin_unlock(details->i_mmap_lock);
2298 cond_resched();
2299 spin_lock(details->i_mmap_lock);
2300 return -EINTR;
2301}
2302
2303static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
2304 struct zap_details *details)
2305{
2306 struct vm_area_struct *vma;
2307 struct prio_tree_iter iter;
2308 pgoff_t vba, vea, zba, zea;
2309
2310restart:
2311 vma_prio_tree_foreach(vma, &iter, root,
2312 details->first_index, details->last_index) {
2313 /* Skip quickly over those we have already dealt with */
2314 if (vma->vm_truncate_count == details->truncate_count)
2315 continue;
2316
2317 vba = vma->vm_pgoff;
2318 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
2319 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
2320 zba = details->first_index;
2321 if (zba < vba)
2322 zba = vba;
2323 zea = details->last_index;
2324 if (zea > vea)
2325 zea = vea;
2326
2327 if (unmap_mapping_range_vma(vma,
2328 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
2329 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
2330 details) < 0)
2331 goto restart;
2332 }
2333}
2334
2335static inline void unmap_mapping_range_list(struct list_head *head,
2336 struct zap_details *details)
2337{
2338 struct vm_area_struct *vma;
2339
2340 /*
2341 * In nonlinear VMAs there is no correspondence between virtual address
2342 * offset and file offset. So we must perform an exhaustive search
2343 * across *all* the pages in each nonlinear VMA, not just the pages
2344 * whose virtual address lies outside the file truncation point.
2345 */
2346restart:
2347 list_for_each_entry(vma, head, shared.vm_set.list) {
2348 /* Skip quickly over those we have already dealt with */
2349 if (vma->vm_truncate_count == details->truncate_count)
2350 continue;
2351 details->nonlinear_vma = vma;
2352 if (unmap_mapping_range_vma(vma, vma->vm_start,
2353 vma->vm_end, details) < 0)
2354 goto restart;
2355 }
2356}
2357
2358/**
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002359 * 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 -07002360 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 * @holebegin: byte in first page to unmap, relative to the start of
2362 * the underlying file. This will be rounded down to a PAGE_SIZE
2363 * boundary. Note that this is different from vmtruncate(), which
2364 * must keep the partial page. In contrast, we must get rid of
2365 * partial pages.
2366 * @holelen: size of prospective hole in bytes. This will be rounded
2367 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2368 * end of the file.
2369 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2370 * but 0 when invalidating pagecache, don't throw away private data.
2371 */
2372void unmap_mapping_range(struct address_space *mapping,
2373 loff_t const holebegin, loff_t const holelen, int even_cows)
2374{
2375 struct zap_details details;
2376 pgoff_t hba = holebegin >> PAGE_SHIFT;
2377 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2378
2379 /* Check for overflow. */
2380 if (sizeof(holelen) > sizeof(hlen)) {
2381 long long holeend =
2382 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2383 if (holeend & ~(long long)ULONG_MAX)
2384 hlen = ULONG_MAX - hba + 1;
2385 }
2386
2387 details.check_mapping = even_cows? NULL: mapping;
2388 details.nonlinear_vma = NULL;
2389 details.first_index = hba;
2390 details.last_index = hba + hlen - 1;
2391 if (details.last_index < details.first_index)
2392 details.last_index = ULONG_MAX;
2393 details.i_mmap_lock = &mapping->i_mmap_lock;
2394
2395 spin_lock(&mapping->i_mmap_lock);
2396
Nick Piggind00806b2007-07-19 01:46:57 -07002397 /* Protect against endless unmapping loops */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398 mapping->truncate_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 if (unlikely(is_restart_addr(mapping->truncate_count))) {
2400 if (mapping->truncate_count == 0)
2401 reset_vma_truncate_counts(mapping);
2402 mapping->truncate_count++;
2403 }
2404 details.truncate_count = mapping->truncate_count;
2405
2406 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
2407 unmap_mapping_range_tree(&mapping->i_mmap, &details);
2408 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
2409 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
2410 spin_unlock(&mapping->i_mmap_lock);
2411}
2412EXPORT_SYMBOL(unmap_mapping_range);
2413
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07002414/**
2415 * vmtruncate - unmap mappings "freed" by truncate() syscall
2416 * @inode: inode of the file used
2417 * @offset: file offset to start truncating
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 *
2419 * NOTE! We have to be ready to update the memory sharing
2420 * between the file and the memory map for a potential last
2421 * incomplete page. Ugly, but necessary.
2422 */
2423int vmtruncate(struct inode * inode, loff_t offset)
2424{
Christoph Hellwig61d50482008-02-04 22:28:56 -08002425 if (inode->i_size < offset) {
2426 unsigned long limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427
Christoph Hellwig61d50482008-02-04 22:28:56 -08002428 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
2429 if (limit != RLIM_INFINITY && offset > limit)
2430 goto out_sig;
2431 if (offset > inode->i_sb->s_maxbytes)
2432 goto out_big;
2433 i_size_write(inode, offset);
2434 } else {
2435 struct address_space *mapping = inode->i_mapping;
Nick Piggind00806b2007-07-19 01:46:57 -07002436
Christoph Hellwig61d50482008-02-04 22:28:56 -08002437 /*
2438 * truncation of in-use swapfiles is disallowed - it would
2439 * cause subsequent swapout to scribble on the now-freed
2440 * blocks.
2441 */
2442 if (IS_SWAPFILE(inode))
2443 return -ETXTBSY;
2444 i_size_write(inode, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445
Christoph Hellwig61d50482008-02-04 22:28:56 -08002446 /*
2447 * unmap_mapping_range is called twice, first simply for
2448 * efficiency so that truncate_inode_pages does fewer
2449 * single-page unmaps. However after this first call, and
2450 * before truncate_inode_pages finishes, it is possible for
2451 * private pages to be COWed, which remain after
2452 * truncate_inode_pages finishes, hence the second
2453 * unmap_mapping_range call must be made for correctness.
2454 */
2455 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2456 truncate_inode_pages(mapping, offset);
2457 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2458 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459
Al Viroacfa4382008-12-04 10:06:33 -05002460 if (inode->i_op->truncate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 inode->i_op->truncate(inode);
2462 return 0;
Christoph Hellwig61d50482008-02-04 22:28:56 -08002463
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464out_sig:
2465 send_sig(SIGXFSZ, current, 0);
2466out_big:
2467 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469EXPORT_SYMBOL(vmtruncate);
2470
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002471int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
2472{
2473 struct address_space *mapping = inode->i_mapping;
2474
2475 /*
2476 * If the underlying filesystem is not going to provide
2477 * a way to truncate a range of blocks (punch a hole) -
2478 * we should return failure right now.
2479 */
Al Viroacfa4382008-12-04 10:06:33 -05002480 if (!inode->i_op->truncate_range)
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002481 return -ENOSYS;
2482
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002483 mutex_lock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002484 down_write(&inode->i_alloc_sem);
2485 unmap_mapping_range(mapping, offset, (end - offset), 1);
2486 truncate_inode_pages_range(mapping, offset, end);
Nick Piggind00806b2007-07-19 01:46:57 -07002487 unmap_mapping_range(mapping, offset, (end - offset), 1);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002488 inode->i_op->truncate_range(inode, offset, end);
2489 up_write(&inode->i_alloc_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002490 mutex_unlock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002491
2492 return 0;
2493}
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002494
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002496 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2497 * but allow concurrent faults), and pte mapped but not yet locked.
2498 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002500static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2501 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002502 unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002504 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505 struct page *page;
Hugh Dickins65500d22005-10-29 18:15:59 -07002506 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 pte_t pte;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002508 struct mem_cgroup *ptr = NULL;
Nick Piggin83c54072007-07-19 01:47:05 -07002509 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002511 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002512 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002513
2514 entry = pte_to_swp_entry(orig_pte);
Christoph Lameter06972122006-06-23 02:03:35 -07002515 if (is_migration_entry(entry)) {
2516 migration_entry_wait(mm, pmd, address);
2517 goto out;
2518 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002519 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520 page = lookup_swap_cache(entry);
2521 if (!page) {
Hugh Dickinsa5c9b692009-06-23 12:36:58 -07002522 grab_swap_token(mm); /* Contend for token _before_ read-in */
Hugh Dickins02098fe2008-02-04 22:28:42 -08002523 page = swapin_readahead(entry,
2524 GFP_HIGHUSER_MOVABLE, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 if (!page) {
2526 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002527 * Back out if somebody else faulted in this pte
2528 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002530 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 if (likely(pte_same(*page_table, orig_pte)))
2532 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002533 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002534 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535 }
2536
2537 /* Had to read the page from swap area: Major fault */
2538 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002539 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 }
2541
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 lock_page(page);
Balbir Singh20a10222007-11-14 17:00:33 -08002543 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002545 if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) {
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002546 ret = VM_FAULT_OOM;
Johannes Weinerbc43f752009-04-30 15:08:08 -07002547 goto out_page;
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002548 }
2549
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002551 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002553 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002554 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002555 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002556
2557 if (unlikely(!PageUptodate(page))) {
2558 ret = VM_FAULT_SIGBUS;
2559 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 }
2561
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002562 /*
2563 * The page isn't present yet, go ahead with the fault.
2564 *
2565 * Be careful about the sequence of operations here.
2566 * To get its accounting right, reuse_swap_page() must be called
2567 * while the page is counted on swap but not yet in mapcount i.e.
2568 * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
2569 * must be called after the swap_free(), or it will never succeed.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002570 * Because delete_from_swap_page() may be called by reuse_swap_page(),
2571 * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry
2572 * in page->private. In this case, a record in swap_cgroup is silently
2573 * discarded at swap_free().
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002574 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575
Hugh Dickins42946212005-10-29 18:16:05 -07002576 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577 pte = mk_pte(page, vma->vm_page_prot);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002578 if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002580 flags &= ~FAULT_FLAG_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 flush_icache_page(vma, page);
2583 set_pte_at(mm, address, page_table, pte);
2584 page_add_anon_rmap(page, vma, address);
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002585 /* It's better to call commit-charge after rmap is established */
2586 mem_cgroup_commit_charge_swapin(page, ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002588 swap_free(entry);
Nick Pigginb291f002008-10-18 20:26:44 -07002589 if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08002590 try_to_free_swap(page);
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002591 unlock_page(page);
2592
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002593 if (flags & FAULT_FLAG_WRITE) {
Hugh Dickins61469f12008-03-04 14:29:04 -08002594 ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
2595 if (ret & VM_FAULT_ERROR)
2596 ret &= VM_FAULT_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 goto out;
2598 }
2599
2600 /* No need to invalidate - it was non-present before */
2601 update_mmu_cache(vma, address, pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07002602unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002603 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604out:
2605 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002606out_nomap:
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002607 mem_cgroup_cancel_charge_swapin(ptr);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002608 pte_unmap_unlock(page_table, ptl);
Johannes Weinerbc43f752009-04-30 15:08:08 -07002609out_page:
Kirill Korotaevb8107482005-05-16 21:53:50 -07002610 unlock_page(page);
2611 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002612 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613}
2614
2615/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002616 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2617 * but allow concurrent faults), and pte mapped but not yet locked.
2618 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002620static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2621 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002622 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002624 struct page *page;
2625 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627
Nick Piggin557ed1f2007-10-16 01:24:40 -07002628 /* Allocate our own private page. */
2629 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630
Nick Piggin557ed1f2007-10-16 01:24:40 -07002631 if (unlikely(anon_vma_prepare(vma)))
2632 goto oom;
2633 page = alloc_zeroed_user_highpage_movable(vma, address);
2634 if (!page)
2635 goto oom;
Nick Piggin0ed361d2008-02-04 22:29:34 -08002636 __SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002638 if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002639 goto oom_free_page;
2640
Nick Piggin557ed1f2007-10-16 01:24:40 -07002641 entry = mk_pte(page, vma->vm_page_prot);
2642 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002643
Nick Piggin557ed1f2007-10-16 01:24:40 -07002644 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2645 if (!pte_none(*page_table))
2646 goto release;
2647 inc_mm_counter(mm, anon_rss);
Nick Piggin557ed1f2007-10-16 01:24:40 -07002648 page_add_new_anon_rmap(page, vma, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002649 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650
2651 /* No need to invalidate - it was non-present before */
Hugh Dickins65500d22005-10-29 18:15:59 -07002652 update_mmu_cache(vma, address, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002653unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002654 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002655 return 0;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002656release:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002657 mem_cgroup_uncharge_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002658 page_cache_release(page);
2659 goto unlock;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002660oom_free_page:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002661 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002662oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663 return VM_FAULT_OOM;
2664}
2665
2666/*
Nick Piggin54cb8822007-07-19 01:46:59 -07002667 * __do_fault() tries to create a new page mapping. It aggressively
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668 * tries to share with existing pages, but makes a separate copy if
Nick Piggin54cb8822007-07-19 01:46:59 -07002669 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2670 * the next page fault.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 *
2672 * As this is called only for pages that do not currently exist, we
2673 * do not need to flush old virtual caches or the TLB.
2674 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002675 * We enter with non-exclusive mmap_sem (to exclude vma changes,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002676 * but allow concurrent faults), and pte neither mapped nor locked.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002677 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 */
Nick Piggin54cb8822007-07-19 01:46:59 -07002679static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002680 unsigned long address, pmd_t *pmd,
Nick Piggin54cb8822007-07-19 01:46:59 -07002681 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682{
Hugh Dickins16abfa02007-10-04 16:56:06 +01002683 pte_t *page_table;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002684 spinlock_t *ptl;
Nick Piggind0217ac2007-07-19 01:47:03 -07002685 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687 int anon = 0;
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002688 int charged = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002689 struct page *dirty_page = NULL;
Nick Piggind0217ac2007-07-19 01:47:03 -07002690 struct vm_fault vmf;
2691 int ret;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002692 int page_mkwrite = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07002693
Nick Piggind0217ac2007-07-19 01:47:03 -07002694 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2695 vmf.pgoff = pgoff;
2696 vmf.flags = flags;
2697 vmf.page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698
Nick Piggin3c18ddd2008-04-28 02:12:10 -07002699 ret = vma->vm_ops->fault(vma, &vmf);
2700 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2701 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702
Nick Piggind00806b2007-07-19 01:46:57 -07002703 /*
Nick Piggind0217ac2007-07-19 01:47:03 -07002704 * For consistency in subsequent calls, make the faulted page always
Nick Piggind00806b2007-07-19 01:46:57 -07002705 * locked.
2706 */
Nick Piggin83c54072007-07-19 01:47:05 -07002707 if (unlikely(!(ret & VM_FAULT_LOCKED)))
Nick Piggind0217ac2007-07-19 01:47:03 -07002708 lock_page(vmf.page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002709 else
Nick Piggind0217ac2007-07-19 01:47:03 -07002710 VM_BUG_ON(!PageLocked(vmf.page));
Nick Piggind00806b2007-07-19 01:46:57 -07002711
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712 /*
2713 * Should we do an early C-O-W break?
2714 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002715 page = vmf.page;
Nick Piggin54cb8822007-07-19 01:46:59 -07002716 if (flags & FAULT_FLAG_WRITE) {
David Howells9637a5e2006-06-23 02:03:43 -07002717 if (!(vma->vm_flags & VM_SHARED)) {
Nick Piggin54cb8822007-07-19 01:46:59 -07002718 anon = 1;
Nick Piggind00806b2007-07-19 01:46:57 -07002719 if (unlikely(anon_vma_prepare(vma))) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002720 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002721 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002722 }
Nick Piggin83c54072007-07-19 01:47:05 -07002723 page = alloc_page_vma(GFP_HIGHUSER_MOVABLE,
2724 vma, address);
Nick Piggind00806b2007-07-19 01:46:57 -07002725 if (!page) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002726 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002727 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002728 }
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002729 if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) {
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002730 ret = VM_FAULT_OOM;
2731 page_cache_release(page);
2732 goto out;
2733 }
2734 charged = 1;
Nick Pigginb291f002008-10-18 20:26:44 -07002735 /*
2736 * Don't let another task, with possibly unlocked vma,
2737 * keep the mlocked page.
2738 */
2739 if (vma->vm_flags & VM_LOCKED)
2740 clear_page_mlock(vmf.page);
Nick Piggind0217ac2007-07-19 01:47:03 -07002741 copy_user_highpage(page, vmf.page, address, vma);
Nick Piggin0ed361d2008-02-04 22:29:34 -08002742 __SetPageUptodate(page);
David Howells9637a5e2006-06-23 02:03:43 -07002743 } else {
Nick Piggin54cb8822007-07-19 01:46:59 -07002744 /*
2745 * If the page will be shareable, see if the backing
David Howells9637a5e2006-06-23 02:03:43 -07002746 * address space wants to know that the page is about
Nick Piggin54cb8822007-07-19 01:46:59 -07002747 * to become writable
2748 */
Mark Fasheh69676142007-07-19 01:47:00 -07002749 if (vma->vm_ops->page_mkwrite) {
Nick Pigginc2ec1752009-03-31 15:23:21 -07002750 int tmp;
2751
Mark Fasheh69676142007-07-19 01:47:00 -07002752 unlock_page(page);
Nick Pigginb827e492009-04-30 15:08:16 -07002753 vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
Nick Pigginc2ec1752009-03-31 15:23:21 -07002754 tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
2755 if (unlikely(tmp &
2756 (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
2757 ret = tmp;
Nick Pigginb827e492009-04-30 15:08:16 -07002758 goto unwritable_page;
Mark Fasheh69676142007-07-19 01:47:00 -07002759 }
Nick Pigginb827e492009-04-30 15:08:16 -07002760 if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
2761 lock_page(page);
2762 if (!page->mapping) {
2763 ret = 0; /* retry the fault */
2764 unlock_page(page);
2765 goto unwritable_page;
2766 }
2767 } else
2768 VM_BUG_ON(!PageLocked(page));
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002769 page_mkwrite = 1;
David Howells9637a5e2006-06-23 02:03:43 -07002770 }
2771 }
Nick Piggin54cb8822007-07-19 01:46:59 -07002772
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773 }
2774
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002775 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776
2777 /*
2778 * This silly early PAGE_DIRTY setting removes a race
2779 * due to the bad i386 page protection. But it's valid
2780 * for other architectures too.
2781 *
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002782 * Note that if FAULT_FLAG_WRITE is set, we either now have
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783 * an exclusive copy of the page, or this is a shared mapping,
2784 * so we can make it writable and dirty to avoid having to
2785 * handle that later.
2786 */
2787 /* Only go through if we didn't race with anybody else... */
Nick Piggin54cb8822007-07-19 01:46:59 -07002788 if (likely(pte_same(*page_table, orig_pte))) {
Nick Piggind00806b2007-07-19 01:46:57 -07002789 flush_icache_page(vma, page);
2790 entry = mk_pte(page, vma->vm_page_prot);
Nick Piggin54cb8822007-07-19 01:46:59 -07002791 if (flags & FAULT_FLAG_WRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793 if (anon) {
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002794 inc_mm_counter(mm, anon_rss);
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002795 page_add_new_anon_rmap(page, vma, address);
Hugh Dickinsf57e88a2005-11-21 21:32:19 -08002796 } else {
Hugh Dickins42946212005-10-29 18:16:05 -07002797 inc_mm_counter(mm, file_rss);
Nick Piggind00806b2007-07-19 01:46:57 -07002798 page_add_file_rmap(page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002799 if (flags & FAULT_FLAG_WRITE) {
Nick Piggind00806b2007-07-19 01:46:57 -07002800 dirty_page = page;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002801 get_page(dirty_page);
2802 }
Hugh Dickins42946212005-10-29 18:16:05 -07002803 }
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002804 set_pte_at(mm, address, page_table, entry);
Nick Piggind00806b2007-07-19 01:46:57 -07002805
2806 /* no need to invalidate: a not-present page won't be cached */
2807 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808 } else {
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002809 if (charged)
2810 mem_cgroup_uncharge_page(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002811 if (anon)
2812 page_cache_release(page);
2813 else
Nick Piggin54cb8822007-07-19 01:46:59 -07002814 anon = 1; /* no anon but release faulted_page */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 }
2816
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002817 pte_unmap_unlock(page_table, ptl);
Nick Piggind00806b2007-07-19 01:46:57 -07002818
2819out:
Nick Pigginb827e492009-04-30 15:08:16 -07002820 if (dirty_page) {
2821 struct address_space *mapping = page->mapping;
2822
2823 if (set_page_dirty(dirty_page))
2824 page_mkwrite = 1;
2825 unlock_page(dirty_page);
2826 put_page(dirty_page);
2827 if (page_mkwrite && mapping) {
2828 /*
2829 * Some device drivers do not set page.mapping but still
2830 * dirty their pages
2831 */
2832 balance_dirty_pages_ratelimited(mapping);
2833 }
2834
2835 /* file_update_time outside page_lock */
Anton Salikhmetov8f7b3d12008-01-23 02:21:18 +03002836 if (vma->vm_file)
2837 file_update_time(vma->vm_file);
Nick Pigginb827e492009-04-30 15:08:16 -07002838 } else {
2839 unlock_page(vmf.page);
2840 if (anon)
2841 page_cache_release(vmf.page);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002842 }
Nick Piggind00806b2007-07-19 01:46:57 -07002843
Nick Piggin83c54072007-07-19 01:47:05 -07002844 return ret;
Nick Pigginb827e492009-04-30 15:08:16 -07002845
2846unwritable_page:
2847 page_cache_release(page);
2848 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07002849}
Nick Piggind00806b2007-07-19 01:46:57 -07002850
Nick Piggin54cb8822007-07-19 01:46:59 -07002851static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2852 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002853 unsigned int flags, pte_t orig_pte)
Nick Piggin54cb8822007-07-19 01:46:59 -07002854{
2855 pgoff_t pgoff = (((address & PAGE_MASK)
Dean Nelson0da7e012007-10-16 01:24:45 -07002856 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
Nick Piggin54cb8822007-07-19 01:46:59 -07002857
Hugh Dickins16abfa02007-10-04 16:56:06 +01002858 pte_unmap(page_table);
2859 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Nick Piggin54cb8822007-07-19 01:46:59 -07002860}
2861
Jes Sorensenf4b81802006-09-27 01:50:10 -07002862/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863 * Fault of a previously existing named mapping. Repopulate the pte
2864 * from the encoded file_pte if possible. This enables swappable
2865 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002866 *
2867 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2868 * but allow concurrent faults), and pte mapped but not yet locked.
2869 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002871static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins65500d22005-10-29 18:15:59 -07002872 unsigned long address, pte_t *page_table, pmd_t *pmd,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002873 unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874{
Hugh Dickins65500d22005-10-29 18:15:59 -07002875 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002877 flags |= FAULT_FLAG_NONLINEAR;
2878
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002879 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Nick Piggin83c54072007-07-19 01:47:05 -07002880 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881
Hugh Dickins2509ef22009-01-06 14:40:10 -08002882 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002883 /*
2884 * Page table corrupted: show pte and kill process.
2885 */
Hugh Dickins3dc14742009-01-06 14:40:08 -08002886 print_bad_pte(vma, address, orig_pte, NULL);
Hugh Dickins65500d22005-10-29 18:15:59 -07002887 return VM_FAULT_OOM;
2888 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002889
2890 pgoff = pte_to_pgoff(orig_pte);
Hugh Dickins16abfa02007-10-04 16:56:06 +01002891 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892}
2893
2894/*
2895 * These routines also need to handle stuff like marking pages dirty
2896 * and/or accessed for architectures that don't do it in hardware (most
2897 * RISC architectures). The early dirtying is also good on the i386.
2898 *
2899 * There is also a hook called "update_mmu_cache()" that architectures
2900 * with external mmu caches can use to update those (ie the Sparc or
2901 * PowerPC hashed page tables that act as extended TLBs).
2902 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07002903 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2904 * but allow concurrent faults), and pte mapped but not yet locked.
2905 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906 */
2907static inline int handle_pte_fault(struct mm_struct *mm,
Hugh Dickins65500d22005-10-29 18:15:59 -07002908 struct vm_area_struct *vma, unsigned long address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002909 pte_t *pte, pmd_t *pmd, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002910{
2911 pte_t entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002912 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002914 entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002916 if (pte_none(entry)) {
Jes Sorensenf4b81802006-09-27 01:50:10 -07002917 if (vma->vm_ops) {
Nick Piggin3c18ddd2008-04-28 02:12:10 -07002918 if (likely(vma->vm_ops->fault))
Nick Piggin54cb8822007-07-19 01:46:59 -07002919 return do_linear_fault(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002920 pte, pmd, flags, entry);
Jes Sorensenf4b81802006-09-27 01:50:10 -07002921 }
2922 return do_anonymous_page(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002923 pte, pmd, flags);
Hugh Dickins65500d22005-10-29 18:15:59 -07002924 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925 if (pte_file(entry))
Nick Piggind0217ac2007-07-19 01:47:03 -07002926 return do_nonlinear_fault(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002927 pte, pmd, flags, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002928 return do_swap_page(mm, vma, address,
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002929 pte, pmd, flags, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002930 }
2931
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002932 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002933 spin_lock(ptl);
2934 if (unlikely(!pte_same(*pte, entry)))
2935 goto unlock;
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002936 if (flags & FAULT_FLAG_WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002938 return do_wp_page(mm, vma, address,
2939 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002940 entry = pte_mkdirty(entry);
2941 }
2942 entry = pte_mkyoung(entry);
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002943 if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) {
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002944 update_mmu_cache(vma, address, entry);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002945 } else {
2946 /*
2947 * This is needed only for protection faults but the arch code
2948 * is not yet telling us if this is a protection fault or not.
2949 * This still avoids useless tlb flushes for .text page faults
2950 * with threads.
2951 */
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002952 if (flags & FAULT_FLAG_WRITE)
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002953 flush_tlb_page(vma, address);
2954 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002955unlock:
2956 pte_unmap_unlock(pte, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002957 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002958}
2959
2960/*
2961 * By the time we get here, we already hold the mm semaphore
2962 */
Nick Piggin83c54072007-07-19 01:47:05 -07002963int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvaldsd06063c2009-04-10 09:01:23 -07002964 unsigned long address, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965{
2966 pgd_t *pgd;
2967 pud_t *pud;
2968 pmd_t *pmd;
2969 pte_t *pte;
2970
2971 __set_current_state(TASK_RUNNING);
2972
Christoph Lameterf8891e52006-06-30 01:55:45 -07002973 count_vm_event(PGFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002974
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01002975 if (unlikely(is_vm_hugetlb_page(vma)))
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002976 return hugetlb_fault(mm, vma, address, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977
Linus Torvalds1da177e2005-04-16 15:20:36 -07002978 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979 pud = pud_alloc(mm, pgd, address);
2980 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002981 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982 pmd = pmd_alloc(mm, pud, address);
2983 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002984 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985 pte = pte_alloc_map(mm, pmd, address);
2986 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002987 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988
Linus Torvalds30c9f3a2009-04-10 08:43:11 -07002989 return handle_pte_fault(mm, vma, address, pte, pmd, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990}
2991
2992#ifndef __PAGETABLE_PUD_FOLDED
2993/*
2994 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002995 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002997int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002999 pud_t *new = pud_alloc_one(mm, address);
3000 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07003001 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002
Nick Piggin362a61a2008-05-14 06:37:36 +02003003 smp_wmb(); /* See comment in __pte_alloc */
3004
Hugh Dickins872fec12005-10-29 18:16:21 -07003005 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003006 if (pgd_present(*pgd)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003007 pud_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003008 else
3009 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07003010 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003011 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003012}
3013#endif /* __PAGETABLE_PUD_FOLDED */
3014
3015#ifndef __PAGETABLE_PMD_FOLDED
3016/*
3017 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07003018 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003019 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07003020int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021{
Hugh Dickinsc74df322005-10-29 18:16:23 -07003022 pmd_t *new = pmd_alloc_one(mm, address);
3023 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07003024 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025
Nick Piggin362a61a2008-05-14 06:37:36 +02003026 smp_wmb(); /* See comment in __pte_alloc */
3027
Hugh Dickins872fec12005-10-29 18:16:21 -07003028 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07003030 if (pud_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003031 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003032 else
3033 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003034#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07003035 if (pgd_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08003036 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003037 else
3038 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003039#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07003040 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07003041 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042}
3043#endif /* __PAGETABLE_PMD_FOLDED */
3044
3045int make_pages_present(unsigned long addr, unsigned long end)
3046{
3047 int ret, len, write;
3048 struct vm_area_struct * vma;
3049
3050 vma = find_vma(current->mm, addr);
3051 if (!vma)
KOSAKI Motohiroa4770972008-08-04 13:41:14 -07003052 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003053 write = (vma->vm_flags & VM_WRITE) != 0;
Eric Sesterhenn5bcb28b2006-03-26 18:30:52 +02003054 BUG_ON(addr >= end);
3055 BUG_ON(end > vma->vm_end);
Rolf Eike Beer68e116a2007-07-15 23:38:03 -07003056 len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057 ret = get_user_pages(current, current->mm, addr,
3058 len, write, 0, NULL, NULL);
Lee Schermerhornc11d69d2008-10-18 20:26:56 -07003059 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060 return ret;
Lee Schermerhorn9978ad52008-10-18 20:26:56 -07003061 return ret == len ? 0 : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003062}
3063
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064#if !defined(__HAVE_ARCH_GATE_AREA)
3065
3066#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07003067static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068
3069static int __init gate_vma_init(void)
3070{
3071 gate_vma.vm_mm = NULL;
3072 gate_vma.vm_start = FIXADDR_USER_START;
3073 gate_vma.vm_end = FIXADDR_USER_END;
Roland McGrathb6558c42007-01-26 00:56:47 -08003074 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
3075 gate_vma.vm_page_prot = __P101;
Roland McGrathf47aef52007-01-26 00:56:49 -08003076 /*
3077 * Make sure the vDSO gets into every core dump.
3078 * Dumping its contents makes post-mortem fully interpretable later
3079 * without matching up the same kernel and hardware config to see
3080 * what PC values meant.
3081 */
3082 gate_vma.vm_flags |= VM_ALWAYSDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083 return 0;
3084}
3085__initcall(gate_vma_init);
3086#endif
3087
3088struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
3089{
3090#ifdef AT_SYSINFO_EHDR
3091 return &gate_vma;
3092#else
3093 return NULL;
3094#endif
3095}
3096
3097int in_gate_area_no_task(unsigned long addr)
3098{
3099#ifdef AT_SYSINFO_EHDR
3100 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
3101 return 1;
3102#endif
3103 return 0;
3104}
3105
3106#endif /* __HAVE_ARCH_GATE_AREA */
David Howells0ec76a12006-09-27 01:50:15 -07003107
Johannes Weinerf8ad0f492009-06-16 15:32:33 -07003108static int follow_pte(struct mm_struct *mm, unsigned long address,
3109 pte_t **ptepp, spinlock_t **ptlp)
3110{
3111 pgd_t *pgd;
3112 pud_t *pud;
3113 pmd_t *pmd;
3114 pte_t *ptep;
3115
3116 pgd = pgd_offset(mm, address);
3117 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
3118 goto out;
3119
3120 pud = pud_offset(pgd, address);
3121 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
3122 goto out;
3123
3124 pmd = pmd_offset(pud, address);
3125 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
3126 goto out;
3127
3128 /* We cannot handle huge page PFN maps. Luckily they don't exist. */
3129 if (pmd_huge(*pmd))
3130 goto out;
3131
3132 ptep = pte_offset_map_lock(mm, pmd, address, ptlp);
3133 if (!ptep)
3134 goto out;
3135 if (!pte_present(*ptep))
3136 goto unlock;
3137 *ptepp = ptep;
3138 return 0;
3139unlock:
3140 pte_unmap_unlock(ptep, *ptlp);
3141out:
3142 return -EINVAL;
3143}
3144
Johannes Weiner3b6748e2009-06-16 15:32:35 -07003145/**
3146 * follow_pfn - look up PFN at a user virtual address
3147 * @vma: memory mapping
3148 * @address: user virtual address
3149 * @pfn: location to store found PFN
3150 *
3151 * Only IO mappings and raw PFN mappings are allowed.
3152 *
3153 * Returns zero and the pfn at @pfn on success, -ve otherwise.
3154 */
3155int follow_pfn(struct vm_area_struct *vma, unsigned long address,
3156 unsigned long *pfn)
3157{
3158 int ret = -EINVAL;
3159 spinlock_t *ptl;
3160 pte_t *ptep;
3161
3162 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3163 return ret;
3164
3165 ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);
3166 if (ret)
3167 return ret;
3168 *pfn = pte_pfn(*ptep);
3169 pte_unmap_unlock(ptep, ptl);
3170 return 0;
3171}
3172EXPORT_SYMBOL(follow_pfn);
3173
Rik van Riel28b2ee22008-07-23 21:27:05 -07003174#ifdef CONFIG_HAVE_IOREMAP_PROT
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003175int follow_phys(struct vm_area_struct *vma,
3176 unsigned long address, unsigned int flags,
3177 unsigned long *prot, resource_size_t *phys)
Rik van Riel28b2ee22008-07-23 21:27:05 -07003178{
Johannes Weiner03668a42009-06-16 15:32:34 -07003179 int ret = -EINVAL;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003180 pte_t *ptep, pte;
3181 spinlock_t *ptl;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003182
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003183 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3184 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003185
Johannes Weiner03668a42009-06-16 15:32:34 -07003186 if (follow_pte(vma->vm_mm, address, &ptep, &ptl))
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003187 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003188 pte = *ptep;
Johannes Weiner03668a42009-06-16 15:32:34 -07003189
Rik van Riel28b2ee22008-07-23 21:27:05 -07003190 if ((flags & FOLL_WRITE) && !pte_write(pte))
3191 goto unlock;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003192
3193 *prot = pgprot_val(pte_pgprot(pte));
Johannes Weiner03668a42009-06-16 15:32:34 -07003194 *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003195
Johannes Weiner03668a42009-06-16 15:32:34 -07003196 ret = 0;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003197unlock:
3198 pte_unmap_unlock(ptep, ptl);
3199out:
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003200 return ret;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003201}
3202
3203int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
3204 void *buf, int len, int write)
3205{
3206 resource_size_t phys_addr;
3207 unsigned long prot = 0;
KOSAKI Motohiro2bc72732009-01-06 14:39:43 -08003208 void __iomem *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003209 int offset = addr & (PAGE_SIZE-1);
3210
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003211 if (follow_phys(vma, addr, write, &prot, &phys_addr))
Rik van Riel28b2ee22008-07-23 21:27:05 -07003212 return -EINVAL;
3213
3214 maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot);
3215 if (write)
3216 memcpy_toio(maddr + offset, buf, len);
3217 else
3218 memcpy_fromio(buf, maddr + offset, len);
3219 iounmap(maddr);
3220
3221 return len;
3222}
3223#endif
3224
David Howells0ec76a12006-09-27 01:50:15 -07003225/*
3226 * Access another process' address space.
3227 * Source/target buffer must be kernel space,
3228 * Do not walk the page table directly, use get_user_pages
3229 */
3230int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
3231{
3232 struct mm_struct *mm;
3233 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07003234 void *old_buf = buf;
3235
3236 mm = get_task_mm(tsk);
3237 if (!mm)
3238 return 0;
3239
3240 down_read(&mm->mmap_sem);
Simon Arlott183ff222007-10-20 01:27:18 +02003241 /* ignore errors, just check how much was successfully transferred */
David Howells0ec76a12006-09-27 01:50:15 -07003242 while (len) {
3243 int bytes, ret, offset;
3244 void *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003245 struct page *page = NULL;
David Howells0ec76a12006-09-27 01:50:15 -07003246
3247 ret = get_user_pages(tsk, mm, addr, 1,
3248 write, 1, &page, &vma);
Rik van Riel28b2ee22008-07-23 21:27:05 -07003249 if (ret <= 0) {
3250 /*
3251 * Check if this is a VM_IO | VM_PFNMAP VMA, which
3252 * we can access using slightly different code.
3253 */
3254#ifdef CONFIG_HAVE_IOREMAP_PROT
3255 vma = find_vma(mm, addr);
3256 if (!vma)
3257 break;
3258 if (vma->vm_ops && vma->vm_ops->access)
3259 ret = vma->vm_ops->access(vma, addr, buf,
3260 len, write);
3261 if (ret <= 0)
3262#endif
3263 break;
3264 bytes = ret;
David Howells0ec76a12006-09-27 01:50:15 -07003265 } else {
Rik van Riel28b2ee22008-07-23 21:27:05 -07003266 bytes = len;
3267 offset = addr & (PAGE_SIZE-1);
3268 if (bytes > PAGE_SIZE-offset)
3269 bytes = PAGE_SIZE-offset;
3270
3271 maddr = kmap(page);
3272 if (write) {
3273 copy_to_user_page(vma, page, addr,
3274 maddr + offset, buf, bytes);
3275 set_page_dirty_lock(page);
3276 } else {
3277 copy_from_user_page(vma, page, addr,
3278 buf, maddr + offset, bytes);
3279 }
3280 kunmap(page);
3281 page_cache_release(page);
David Howells0ec76a12006-09-27 01:50:15 -07003282 }
David Howells0ec76a12006-09-27 01:50:15 -07003283 len -= bytes;
3284 buf += bytes;
3285 addr += bytes;
3286 }
3287 up_read(&mm->mmap_sem);
3288 mmput(mm);
3289
3290 return buf - old_buf;
3291}
Andi Kleen03252912008-01-30 13:33:18 +01003292
3293/*
3294 * Print the name of a VMA.
3295 */
3296void print_vma_addr(char *prefix, unsigned long ip)
3297{
3298 struct mm_struct *mm = current->mm;
3299 struct vm_area_struct *vma;
3300
Ingo Molnare8bff742008-02-13 20:21:06 +01003301 /*
3302 * Do not print if we are in atomic
3303 * contexts (in exception stacks, etc.):
3304 */
3305 if (preempt_count())
3306 return;
3307
Andi Kleen03252912008-01-30 13:33:18 +01003308 down_read(&mm->mmap_sem);
3309 vma = find_vma(mm, ip);
3310 if (vma && vma->vm_file) {
3311 struct file *f = vma->vm_file;
3312 char *buf = (char *)__get_free_page(GFP_KERNEL);
3313 if (buf) {
3314 char *p, *s;
3315
Jan Blunckcf28b482008-02-14 19:38:44 -08003316 p = d_path(&f->f_path, buf, PAGE_SIZE);
Andi Kleen03252912008-01-30 13:33:18 +01003317 if (IS_ERR(p))
3318 p = "?";
3319 s = strrchr(p, '/');
3320 if (s)
3321 p = s+1;
3322 printk("%s%s[%lx+%lx]", prefix, p,
3323 vma->vm_start,
3324 vma->vm_end - vma->vm_start);
3325 free_page((unsigned long)buf);
3326 }
3327 }
3328 up_read(&current->mm->mmap_sem);
3329}
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003330
3331#ifdef CONFIG_PROVE_LOCKING
3332void might_fault(void)
3333{
Peter Zijlstra95156f02009-01-12 13:02:11 +01003334 /*
3335 * Some code (nfs/sunrpc) uses socket ops on kernel memory while
3336 * holding the mmap_sem, this is safe because kernel memory doesn't
3337 * get paged out, therefore we'll never actually fault, and the
3338 * below annotations will generate false positives.
3339 */
3340 if (segment_eq(get_fs(), KERNEL_DS))
3341 return;
3342
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003343 might_sleep();
3344 /*
3345 * it would be nicer only to annotate paths which are not under
3346 * pagefault_disable, however that requires a larger audit and
3347 * providing helpers like get_user_atomic.
3348 */
3349 if (!in_atomic() && current->mm)
3350 might_lock_read(&current->mm->mmap_sem);
3351}
3352EXPORT_SYMBOL(might_fault);
3353#endif