blob: 4126dd16778c36595af938988c9d8ec144d0f8aa [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
59#include <asm/pgalloc.h>
60#include <asm/uaccess.h>
61#include <asm/tlb.h>
62#include <asm/tlbflush.h>
63#include <asm/pgtable.h>
64
Jan Beulich42b77722008-07-23 21:27:10 -070065#include "internal.h"
66
Andy Whitcroftd41dee32005-06-23 00:07:54 -070067#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -070068/* use the per-pgdat data instead for discontigmem - mbligh */
69unsigned long max_mapnr;
70struct page *mem_map;
71
72EXPORT_SYMBOL(max_mapnr);
73EXPORT_SYMBOL(mem_map);
74#endif
75
76unsigned long num_physpages;
77/*
78 * A number of key systems in x86 including ioremap() rely on the assumption
79 * that high_memory defines the upper bound on direct map memory, then end
80 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
81 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
82 * and ZONE_HIGHMEM.
83 */
84void * high_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
86EXPORT_SYMBOL(num_physpages);
87EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Ingo Molnar32a93232008-02-06 22:39:44 +010089/*
90 * Randomize the address space (stacks, mmaps, brk, etc.).
91 *
92 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
93 * as ancient (libc5 based) binaries can segfault. )
94 */
95int randomize_va_space __read_mostly =
96#ifdef CONFIG_COMPAT_BRK
97 1;
98#else
99 2;
100#endif
Andi Kleena62eaf12006-02-16 23:41:58 +0100101
102static int __init disable_randmaps(char *s)
103{
104 randomize_va_space = 0;
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800105 return 1;
Andi Kleena62eaf12006-02-16 23:41:58 +0100106}
107__setup("norandmaps", disable_randmaps);
108
109
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110/*
111 * If a p?d_bad entry is found while walking page tables, report
112 * the error, before resetting entry to p?d_none. Usually (but
113 * very seldom) called out from the p?d_none_or_clear_bad macros.
114 */
115
116void pgd_clear_bad(pgd_t *pgd)
117{
118 pgd_ERROR(*pgd);
119 pgd_clear(pgd);
120}
121
122void pud_clear_bad(pud_t *pud)
123{
124 pud_ERROR(*pud);
125 pud_clear(pud);
126}
127
128void pmd_clear_bad(pmd_t *pmd)
129{
130 pmd_ERROR(*pmd);
131 pmd_clear(pmd);
132}
133
134/*
135 * Note: this doesn't free the actual pages themselves. That
136 * has been handled earlier when unmapping all the memory regions.
137 */
Hugh Dickinse0da3822005-04-19 13:29:15 -0700138static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800140 pgtable_t token = pmd_pgtable(*pmd);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700141 pmd_clear(pmd);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800142 pte_free_tlb(tlb, token);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700143 tlb->mm->nr_ptes--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144}
145
Hugh Dickinse0da3822005-04-19 13:29:15 -0700146static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
147 unsigned long addr, unsigned long end,
148 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149{
150 pmd_t *pmd;
151 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700152 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
Hugh Dickinse0da3822005-04-19 13:29:15 -0700154 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 do {
157 next = pmd_addr_end(addr, end);
158 if (pmd_none_or_clear_bad(pmd))
159 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700160 free_pte_range(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 } while (pmd++, addr = next, addr != end);
162
Hugh Dickinse0da3822005-04-19 13:29:15 -0700163 start &= PUD_MASK;
164 if (start < floor)
165 return;
166 if (ceiling) {
167 ceiling &= PUD_MASK;
168 if (!ceiling)
169 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700171 if (end - 1 > ceiling - 1)
172 return;
173
174 pmd = pmd_offset(pud, start);
175 pud_clear(pud);
176 pmd_free_tlb(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177}
178
Hugh Dickinse0da3822005-04-19 13:29:15 -0700179static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
180 unsigned long addr, unsigned long end,
181 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182{
183 pud_t *pud;
184 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700185 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
Hugh Dickinse0da3822005-04-19 13:29:15 -0700187 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 pud = pud_offset(pgd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 do {
190 next = pud_addr_end(addr, end);
191 if (pud_none_or_clear_bad(pud))
192 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700193 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 } while (pud++, addr = next, addr != end);
195
Hugh Dickinse0da3822005-04-19 13:29:15 -0700196 start &= PGDIR_MASK;
197 if (start < floor)
198 return;
199 if (ceiling) {
200 ceiling &= PGDIR_MASK;
201 if (!ceiling)
202 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700204 if (end - 1 > ceiling - 1)
205 return;
206
207 pud = pud_offset(pgd, start);
208 pgd_clear(pgd);
209 pud_free_tlb(tlb, pud);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210}
211
212/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700213 * This function frees user-level page tables of a process.
214 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 * Must be called with pagetable lock held.
216 */
Jan Beulich42b77722008-07-23 21:27:10 -0700217void free_pgd_range(struct mmu_gather *tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700218 unsigned long addr, unsigned long end,
219 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220{
221 pgd_t *pgd;
222 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700223 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224
Hugh Dickinse0da3822005-04-19 13:29:15 -0700225 /*
226 * The next few lines have given us lots of grief...
227 *
228 * Why are we testing PMD* at this top level? Because often
229 * there will be no work to do at all, and we'd prefer not to
230 * go all the way down to the bottom just to discover that.
231 *
232 * Why all these "- 1"s? Because 0 represents both the bottom
233 * of the address space and the top of it (using -1 for the
234 * top wouldn't help much: the masks would do the wrong thing).
235 * The rule is that addr 0 and floor 0 refer to the bottom of
236 * the address space, but end 0 and ceiling 0 refer to the top
237 * Comparisons need to use "end - 1" and "ceiling - 1" (though
238 * that end 0 case should be mythical).
239 *
240 * Wherever addr is brought up or ceiling brought down, we must
241 * be careful to reject "the opposite 0" before it confuses the
242 * subsequent tests. But what about where end is brought down
243 * by PMD_SIZE below? no, end can't go down to 0 there.
244 *
245 * Whereas we round start (addr) and ceiling down, by different
246 * masks at different levels, in order to test whether a table
247 * now has no other vmas using it, so can be freed, we don't
248 * bother to round floor or end up - the tests don't need that.
249 */
250
251 addr &= PMD_MASK;
252 if (addr < floor) {
253 addr += PMD_SIZE;
254 if (!addr)
255 return;
256 }
257 if (ceiling) {
258 ceiling &= PMD_MASK;
259 if (!ceiling)
260 return;
261 }
262 if (end - 1 > ceiling - 1)
263 end -= PMD_SIZE;
264 if (addr > end - 1)
265 return;
266
267 start = addr;
Jan Beulich42b77722008-07-23 21:27:10 -0700268 pgd = pgd_offset(tlb->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 do {
270 next = pgd_addr_end(addr, end);
271 if (pgd_none_or_clear_bad(pgd))
272 continue;
Jan Beulich42b77722008-07-23 21:27:10 -0700273 free_pud_range(tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700275}
276
Jan Beulich42b77722008-07-23 21:27:10 -0700277void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700278 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700279{
280 while (vma) {
281 struct vm_area_struct *next = vma->vm_next;
282 unsigned long addr = vma->vm_start;
283
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700284 /*
285 * Hide vma from rmap and vmtruncate before freeing pgtables
286 */
287 anon_vma_unlink(vma);
288 unlink_file_vma(vma);
289
David Gibson9da61ae2006-03-22 00:08:57 -0800290 if (is_vm_hugetlb_page(vma)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700291 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700292 floor, next? next->vm_start: ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700293 } else {
294 /*
295 * Optimization: gather nearby vmas into one call down
296 */
297 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
David Gibson48669202006-03-22 00:08:58 -0800298 && !is_vm_hugetlb_page(next)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700299 vma = next;
300 next = vma->vm_next;
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700301 anon_vma_unlink(vma);
302 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700303 }
304 free_pgd_range(tlb, addr, vma->vm_end,
305 floor, next? next->vm_start: ceiling);
306 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700307 vma = next;
308 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309}
310
Hugh Dickins1bb36302005-10-29 18:16:22 -0700311int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800313 pgtable_t new = pte_alloc_one(mm, address);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700314 if (!new)
315 return -ENOMEM;
316
Nick Piggin362a61a2008-05-14 06:37:36 +0200317 /*
318 * Ensure all pte setup (eg. pte page lock and page clearing) are
319 * visible before the pte is made visible to other CPUs by being
320 * put into page tables.
321 *
322 * The other side of the story is the pointer chasing in the page
323 * table walking code (when walking the page table without locking;
324 * ie. most of the time). Fortunately, these data accesses consist
325 * of a chain of data-dependent loads, meaning most CPUs (alpha
326 * being the notable exception) will already guarantee loads are
327 * seen in-order. See the alpha page table accessors for the
328 * smp_read_barrier_depends() barriers in page table walking code.
329 */
330 smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
331
Hugh Dickinsc74df322005-10-29 18:16:23 -0700332 spin_lock(&mm->page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800333 if (!pmd_present(*pmd)) { /* Has another populated it ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 mm->nr_ptes++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 pmd_populate(mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800336 new = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 }
Hugh Dickinsc74df322005-10-29 18:16:23 -0700338 spin_unlock(&mm->page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800339 if (new)
340 pte_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700341 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342}
343
Hugh Dickins1bb36302005-10-29 18:16:22 -0700344int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700346 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
347 if (!new)
348 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349
Nick Piggin362a61a2008-05-14 06:37:36 +0200350 smp_wmb(); /* See comment in __pte_alloc */
351
Hugh Dickins1bb36302005-10-29 18:16:22 -0700352 spin_lock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800353 if (!pmd_present(*pmd)) { /* Has another populated it ? */
Hugh Dickins1bb36302005-10-29 18:16:22 -0700354 pmd_populate_kernel(&init_mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800355 new = NULL;
356 }
Hugh Dickins1bb36302005-10-29 18:16:22 -0700357 spin_unlock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800358 if (new)
359 pte_free_kernel(&init_mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700360 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361}
362
Hugh Dickinsae859762005-10-29 18:16:05 -0700363static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
364{
365 if (file_rss)
366 add_mm_counter(mm, file_rss, file_rss);
367 if (anon_rss)
368 add_mm_counter(mm, anon_rss, anon_rss);
369}
370
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800372 * This function is called to print an error when a bad pte
373 * is found. For example, we might have a PFN-mapped pte in
374 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700375 *
376 * The calling function must still handle the error.
377 */
Hugh Dickins3dc14742009-01-06 14:40:08 -0800378static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
379 pte_t pte, struct page *page)
Nick Pigginb5810032005-10-29 18:16:12 -0700380{
Hugh Dickins3dc14742009-01-06 14:40:08 -0800381 pgd_t *pgd = pgd_offset(vma->vm_mm, addr);
382 pud_t *pud = pud_offset(pgd, addr);
383 pmd_t *pmd = pmd_offset(pud, addr);
384 struct address_space *mapping;
385 pgoff_t index;
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800386 static unsigned long resume;
387 static unsigned long nr_shown;
388 static unsigned long nr_unshown;
389
390 /*
391 * Allow a burst of 60 reports, then keep quiet for that minute;
392 * or allow a steady drip of one report per second.
393 */
394 if (nr_shown == 60) {
395 if (time_before(jiffies, resume)) {
396 nr_unshown++;
397 return;
398 }
399 if (nr_unshown) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800400 printk(KERN_ALERT
401 "BUG: Bad page map: %lu messages suppressed\n",
Hugh Dickinsd936cf92009-01-06 14:40:12 -0800402 nr_unshown);
403 nr_unshown = 0;
404 }
405 nr_shown = 0;
406 }
407 if (nr_shown++ == 0)
408 resume = jiffies + 60 * HZ;
Hugh Dickins3dc14742009-01-06 14:40:08 -0800409
410 mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;
411 index = linear_page_index(vma, addr);
412
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800413 printk(KERN_ALERT
414 "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800415 current->comm,
416 (long long)pte_val(pte), (long long)pmd_val(*pmd));
417 if (page) {
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800418 printk(KERN_ALERT
Hugh Dickins3dc14742009-01-06 14:40:08 -0800419 "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
420 page, (void *)page->flags, page_count(page),
421 page_mapcount(page), page->mapping, page->index);
422 }
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800423 printk(KERN_ALERT
Hugh Dickins3dc14742009-01-06 14:40:08 -0800424 "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
425 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
426 /*
427 * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y
428 */
429 if (vma->vm_ops)
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800430 print_symbol(KERN_ALERT "vma->vm_ops->fault: %s\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800431 (unsigned long)vma->vm_ops->fault);
432 if (vma->vm_file && vma->vm_file->f_op)
Hugh Dickins1e9e6362009-01-06 14:40:13 -0800433 print_symbol(KERN_ALERT "vma->vm_file->f_op->mmap: %s\n",
Hugh Dickins3dc14742009-01-06 14:40:08 -0800434 (unsigned long)vma->vm_file->f_op->mmap);
Nick Pigginb5810032005-10-29 18:16:12 -0700435 dump_stack();
Hugh Dickins3dc14742009-01-06 14:40:08 -0800436 add_taint(TAINT_BAD_PAGE);
Nick Pigginb5810032005-10-29 18:16:12 -0700437}
438
Linus Torvalds67121172005-12-11 20:38:17 -0800439static inline int is_cow_mapping(unsigned int flags)
440{
441 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
442}
443
Nick Pigginb5810032005-10-29 18:16:12 -0700444/*
Nick Piggin7e675132008-04-28 02:13:00 -0700445 * vm_normal_page -- This function gets the "struct page" associated with a pte.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800446 *
Nick Piggin7e675132008-04-28 02:13:00 -0700447 * "Special" mappings do not wish to be associated with a "struct page" (either
448 * it doesn't exist, or it exists but they don't want to touch it). In this
449 * case, NULL is returned here. "Normal" mappings do have a struct page.
Jared Hulbertb379d792008-04-28 02:12:58 -0700450 *
Nick Piggin7e675132008-04-28 02:13:00 -0700451 * There are 2 broad cases. Firstly, an architecture may define a pte_special()
452 * pte bit, in which case this function is trivial. Secondly, an architecture
453 * may not have a spare pte bit, which requires a more complicated scheme,
454 * described below.
455 *
456 * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
457 * special mapping (even if there are underlying and valid "struct pages").
458 * COWed pages of a VM_PFNMAP are always normal.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800459 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700460 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
461 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
Nick Piggin7e675132008-04-28 02:13:00 -0700462 * set, and the vm_pgoff will point to the first PFN mapped: thus every special
463 * mapping will always honor the rule
Linus Torvalds6aab3412005-11-28 14:34:23 -0800464 *
465 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
466 *
Nick Piggin7e675132008-04-28 02:13:00 -0700467 * And for normal mappings this is false.
Jared Hulbertb379d792008-04-28 02:12:58 -0700468 *
Nick Piggin7e675132008-04-28 02:13:00 -0700469 * This restricts such mappings to be a linear translation from virtual address
470 * to pfn. To get around this restriction, we allow arbitrary mappings so long
471 * as the vma is not a COW mapping; in that case, we know that all ptes are
472 * special (because none can have been COWed).
Jared Hulbertb379d792008-04-28 02:12:58 -0700473 *
474 *
Nick Piggin7e675132008-04-28 02:13:00 -0700475 * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
476 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700477 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
478 * page" backing, however the difference is that _all_ pages with a struct
479 * page (that is, those where pfn_valid is true) are refcounted and considered
480 * normal pages by the VM. The disadvantage is that pages are refcounted
481 * (which can be slower and simply not an option for some PFNMAP users). The
482 * advantage is that we don't have to follow the strict linearity rule of
483 * PFNMAP mappings in order to support COWable mappings.
484 *
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800485 */
Nick Piggin7e675132008-04-28 02:13:00 -0700486#ifdef __HAVE_ARCH_PTE_SPECIAL
487# define HAVE_PTE_SPECIAL 1
488#else
489# define HAVE_PTE_SPECIAL 0
490#endif
491struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
492 pte_t pte)
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800493{
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800494 unsigned long pfn = pte_pfn(pte);
Nick Piggin7e675132008-04-28 02:13:00 -0700495
496 if (HAVE_PTE_SPECIAL) {
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800497 if (likely(!pte_special(pte)))
498 goto check_pfn;
499 if (!(vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)))
500 print_bad_pte(vma, addr, pte, NULL);
Nick Piggin7e675132008-04-28 02:13:00 -0700501 return NULL;
502 }
503
504 /* !HAVE_PTE_SPECIAL case follows: */
505
Jared Hulbertb379d792008-04-28 02:12:58 -0700506 if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
507 if (vma->vm_flags & VM_MIXEDMAP) {
508 if (!pfn_valid(pfn))
509 return NULL;
510 goto out;
511 } else {
Nick Piggin7e675132008-04-28 02:13:00 -0700512 unsigned long off;
513 off = (addr - vma->vm_start) >> PAGE_SHIFT;
Jared Hulbertb379d792008-04-28 02:12:58 -0700514 if (pfn == vma->vm_pgoff + off)
515 return NULL;
516 if (!is_cow_mapping(vma->vm_flags))
517 return NULL;
518 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800519 }
520
Hugh Dickins22b31ee2009-01-06 14:40:09 -0800521check_pfn:
522 if (unlikely(pfn > highest_memmap_pfn)) {
523 print_bad_pte(vma, addr, pte, NULL);
524 return NULL;
525 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800526
527 /*
Nick Piggin7e675132008-04-28 02:13:00 -0700528 * NOTE! We still have PageReserved() pages in the page tables.
Nick Piggin7e675132008-04-28 02:13:00 -0700529 * eg. VDSO mappings can cause them to exist.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800530 */
Jared Hulbertb379d792008-04-28 02:12:58 -0700531out:
Linus Torvalds6aab3412005-11-28 14:34:23 -0800532 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800533}
534
535/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 * copy one vm_area from one task to the other. Assumes the page tables
537 * already present in the new task to be cleared in the whole range
538 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 */
540
Hugh Dickins8c103762005-10-29 18:16:13 -0700541static inline void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700543 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700544 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545{
Nick Pigginb5810032005-10-29 18:16:12 -0700546 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 pte_t pte = *src_pte;
548 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
550 /* pte contains position in swap or file, so copy. */
551 if (unlikely(!pte_present(pte))) {
552 if (!pte_file(pte)) {
Christoph Lameter06972122006-06-23 02:03:35 -0700553 swp_entry_t entry = pte_to_swp_entry(pte);
554
555 swap_duplicate(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 /* make sure dst_mm is on swapoff's mmlist. */
557 if (unlikely(list_empty(&dst_mm->mmlist))) {
558 spin_lock(&mmlist_lock);
Hugh Dickinsf412ac02005-10-29 18:16:41 -0700559 if (list_empty(&dst_mm->mmlist))
560 list_add(&dst_mm->mmlist,
561 &src_mm->mmlist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 spin_unlock(&mmlist_lock);
563 }
Christoph Lameter06972122006-06-23 02:03:35 -0700564 if (is_write_migration_entry(entry) &&
565 is_cow_mapping(vm_flags)) {
566 /*
567 * COW mappings require pages in both parent
568 * and child to be set to read.
569 */
570 make_migration_entry_read(&entry);
571 pte = swp_entry_to_pte(entry);
572 set_pte_at(src_mm, addr, src_pte, pte);
573 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700575 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 }
577
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 /*
579 * If it's a COW mapping, write protect it both
580 * in the parent and the child
581 */
Linus Torvalds67121172005-12-11 20:38:17 -0800582 if (is_cow_mapping(vm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 ptep_set_wrprotect(src_mm, addr, src_pte);
Zachary Amsden3dc90792006-09-30 23:29:30 -0700584 pte = pte_wrprotect(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 }
586
587 /*
588 * If it's a shared mapping, mark it clean in
589 * the child
590 */
591 if (vm_flags & VM_SHARED)
592 pte = pte_mkclean(pte);
593 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800594
595 page = vm_normal_page(vma, addr, pte);
596 if (page) {
597 get_page(page);
Nick Pigginc97a9e12007-05-16 22:11:21 -0700598 page_dup_rmap(page, vma, addr);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800599 rss[!!PageAnon(page)]++;
600 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700601
602out_set_pte:
603 set_pte_at(dst_mm, addr, dst_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604}
605
606static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
607 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
608 unsigned long addr, unsigned long end)
609{
610 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700611 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -0700612 int progress = 0;
Hugh Dickins8c103762005-10-29 18:16:13 -0700613 int rss[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
615again:
Hugh Dickinsae859762005-10-29 18:16:05 -0700616 rss[1] = rss[0] = 0;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700617 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 if (!dst_pte)
619 return -ENOMEM;
620 src_pte = pte_offset_map_nested(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700621 src_ptl = pte_lockptr(src_mm, src_pmd);
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700622 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700623 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 do {
626 /*
627 * We are holding two locks at this point - either of them
628 * could generate latencies in another task on another CPU.
629 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700630 if (progress >= 32) {
631 progress = 0;
632 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +0100633 spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700634 break;
635 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 if (pte_none(*src_pte)) {
637 progress++;
638 continue;
639 }
Hugh Dickins8c103762005-10-29 18:16:13 -0700640 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 progress += 8;
642 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700644 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -0700645 spin_unlock(src_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 pte_unmap_nested(src_pte - 1);
Hugh Dickinsae859762005-10-29 18:16:05 -0700647 add_mm_rss(dst_mm, rss[0], rss[1]);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700648 pte_unmap_unlock(dst_pte - 1, dst_ptl);
649 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 if (addr != end)
651 goto again;
652 return 0;
653}
654
655static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
656 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
657 unsigned long addr, unsigned long end)
658{
659 pmd_t *src_pmd, *dst_pmd;
660 unsigned long next;
661
662 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
663 if (!dst_pmd)
664 return -ENOMEM;
665 src_pmd = pmd_offset(src_pud, addr);
666 do {
667 next = pmd_addr_end(addr, end);
668 if (pmd_none_or_clear_bad(src_pmd))
669 continue;
670 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
671 vma, addr, next))
672 return -ENOMEM;
673 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
674 return 0;
675}
676
677static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
678 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
679 unsigned long addr, unsigned long end)
680{
681 pud_t *src_pud, *dst_pud;
682 unsigned long next;
683
684 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
685 if (!dst_pud)
686 return -ENOMEM;
687 src_pud = pud_offset(src_pgd, addr);
688 do {
689 next = pud_addr_end(addr, end);
690 if (pud_none_or_clear_bad(src_pud))
691 continue;
692 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
693 vma, addr, next))
694 return -ENOMEM;
695 } while (dst_pud++, src_pud++, addr = next, addr != end);
696 return 0;
697}
698
699int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
700 struct vm_area_struct *vma)
701{
702 pgd_t *src_pgd, *dst_pgd;
703 unsigned long next;
704 unsigned long addr = vma->vm_start;
705 unsigned long end = vma->vm_end;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700706 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707
Nick Piggind9928952005-08-28 16:49:11 +1000708 /*
709 * Don't copy ptes where a page fault will fill them correctly.
710 * Fork becomes much lighter when there are big shared or private
711 * readonly mappings. The tradeoff is that copy_page_range is more
712 * efficient than faulting.
713 */
Linus Torvalds4d7672b2005-12-16 10:21:23 -0800714 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
Nick Piggind9928952005-08-28 16:49:11 +1000715 if (!vma->anon_vma)
716 return 0;
717 }
718
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 if (is_vm_hugetlb_page(vma))
720 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
721
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800722 if (unlikely(is_pfn_mapping(vma))) {
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -0800723 /*
724 * We do not free on error cases below as remove_vma
725 * gets called on error from higher level routine
726 */
727 ret = track_pfn_vma_copy(vma);
728 if (ret)
729 return ret;
730 }
731
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700732 /*
733 * We need to invalidate the secondary MMU mappings only when
734 * there could be a permission downgrade on the ptes of the
735 * parent mm. And a permission downgrade will only happen if
736 * is_cow_mapping() returns true.
737 */
738 if (is_cow_mapping(vma->vm_flags))
739 mmu_notifier_invalidate_range_start(src_mm, addr, end);
740
741 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 dst_pgd = pgd_offset(dst_mm, addr);
743 src_pgd = pgd_offset(src_mm, addr);
744 do {
745 next = pgd_addr_end(addr, end);
746 if (pgd_none_or_clear_bad(src_pgd))
747 continue;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700748 if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
749 vma, addr, next))) {
750 ret = -ENOMEM;
751 break;
752 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700754
755 if (is_cow_mapping(vma->vm_flags))
756 mmu_notifier_invalidate_range_end(src_mm,
757 vma->vm_start, end);
758 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759}
760
Robin Holt51c6f662005-11-13 16:06:42 -0800761static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700762 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800764 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765{
Nick Pigginb5810032005-10-29 18:16:12 -0700766 struct mm_struct *mm = tlb->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 pte_t *pte;
Hugh Dickins508034a2005-10-29 18:16:30 -0700768 spinlock_t *ptl;
Hugh Dickinsae859762005-10-29 18:16:05 -0700769 int file_rss = 0;
770 int anon_rss = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771
Hugh Dickins508034a2005-10-29 18:16:30 -0700772 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700773 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 do {
775 pte_t ptent = *pte;
Robin Holt51c6f662005-11-13 16:06:42 -0800776 if (pte_none(ptent)) {
777 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800779 }
Hugh Dickins6f5e6b92006-03-16 23:04:09 -0800780
781 (*zap_work) -= PAGE_SIZE;
782
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800784 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -0800785
Linus Torvalds6aab3412005-11-28 14:34:23 -0800786 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 if (unlikely(details) && page) {
788 /*
789 * unmap_shared_mapping_pages() wants to
790 * invalidate cache without truncating:
791 * unmap shared but keep private pages.
792 */
793 if (details->check_mapping &&
794 details->check_mapping != page->mapping)
795 continue;
796 /*
797 * Each page->index must be checked when
798 * invalidating or truncating nonlinear.
799 */
800 if (details->nonlinear_vma &&
801 (page->index < details->first_index ||
802 page->index > details->last_index))
803 continue;
804 }
Nick Pigginb5810032005-10-29 18:16:12 -0700805 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -0700806 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 tlb_remove_tlb_entry(tlb, pte, addr);
808 if (unlikely(!page))
809 continue;
810 if (unlikely(details) && details->nonlinear_vma
811 && linear_page_index(details->nonlinear_vma,
812 addr) != page->index)
Nick Pigginb5810032005-10-29 18:16:12 -0700813 set_pte_at(mm, addr, pte,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 pgoff_to_pte(page->index));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 if (PageAnon(page))
Hugh Dickins86d912f2005-10-29 18:16:14 -0700816 anon_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700817 else {
818 if (pte_dirty(ptent))
819 set_page_dirty(page);
Johannes Weiner4917e5d2009-01-06 14:39:17 -0800820 if (pte_young(ptent) &&
821 likely(!VM_SequentialReadHint(vma)))
Nick Pigginbf3f3bc2009-01-06 14:38:55 -0800822 mark_page_accessed(page);
Hugh Dickins86d912f2005-10-29 18:16:14 -0700823 file_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700824 }
Hugh Dickinsedc315f2009-01-06 14:40:11 -0800825 page_remove_rmap(page);
Hugh Dickins3dc14742009-01-06 14:40:08 -0800826 if (unlikely(page_mapcount(page) < 0))
827 print_bad_pte(vma, addr, ptent, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 tlb_remove_page(tlb, page);
829 continue;
830 }
831 /*
832 * If details->check_mapping, we leave swap entries;
833 * if details->nonlinear_vma, we leave file entries.
834 */
835 if (unlikely(details))
836 continue;
Hugh Dickins2509ef22009-01-06 14:40:10 -0800837 if (pte_file(ptent)) {
838 if (unlikely(!(vma->vm_flags & VM_NONLINEAR)))
839 print_bad_pte(vma, addr, ptent, NULL);
840 } else if
841 (unlikely(!free_swap_and_cache(pte_to_swp_entry(ptent))))
842 print_bad_pte(vma, addr, ptent, NULL);
Zachary Amsden9888a1c2006-09-30 23:29:31 -0700843 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
Robin Holt51c6f662005-11-13 16:06:42 -0800844 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
Hugh Dickinsae859762005-10-29 18:16:05 -0700845
Hugh Dickins86d912f2005-10-29 18:16:14 -0700846 add_mm_rss(mm, file_rss, anon_rss);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700847 arch_leave_lazy_mmu_mode();
Hugh Dickins508034a2005-10-29 18:16:30 -0700848 pte_unmap_unlock(pte - 1, ptl);
Robin Holt51c6f662005-11-13 16:06:42 -0800849
850 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851}
852
Robin Holt51c6f662005-11-13 16:06:42 -0800853static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700854 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800856 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857{
858 pmd_t *pmd;
859 unsigned long next;
860
861 pmd = pmd_offset(pud, addr);
862 do {
863 next = pmd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800864 if (pmd_none_or_clear_bad(pmd)) {
865 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800867 }
868 next = zap_pte_range(tlb, vma, pmd, addr, next,
869 zap_work, details);
870 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
871
872 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873}
874
Robin Holt51c6f662005-11-13 16:06:42 -0800875static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700876 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800878 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879{
880 pud_t *pud;
881 unsigned long next;
882
883 pud = pud_offset(pgd, addr);
884 do {
885 next = pud_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800886 if (pud_none_or_clear_bad(pud)) {
887 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800889 }
890 next = zap_pmd_range(tlb, vma, pud, addr, next,
891 zap_work, details);
892 } while (pud++, addr = next, (addr != end && *zap_work > 0));
893
894 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895}
896
Robin Holt51c6f662005-11-13 16:06:42 -0800897static unsigned long unmap_page_range(struct mmu_gather *tlb,
898 struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800900 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901{
902 pgd_t *pgd;
903 unsigned long next;
904
905 if (details && !details->check_mapping && !details->nonlinear_vma)
906 details = NULL;
907
908 BUG_ON(addr >= end);
909 tlb_start_vma(tlb, vma);
910 pgd = pgd_offset(vma->vm_mm, addr);
911 do {
912 next = pgd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800913 if (pgd_none_or_clear_bad(pgd)) {
914 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800916 }
917 next = zap_pud_range(tlb, vma, pgd, addr, next,
918 zap_work, details);
919 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 tlb_end_vma(tlb, vma);
Robin Holt51c6f662005-11-13 16:06:42 -0800921
922 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923}
924
925#ifdef CONFIG_PREEMPT
926# define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
927#else
928/* No preempt: go for improved straight-line efficiency */
929# define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
930#endif
931
932/**
933 * unmap_vmas - unmap a range of memory covered by a list of vma's
934 * @tlbp: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 * @vma: the starting vma
936 * @start_addr: virtual address at which to start unmapping
937 * @end_addr: virtual address at which to end unmapping
938 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
939 * @details: details of nonlinear truncation or shared cache invalidation
940 *
Hugh Dickinsee39b372005-04-19 13:29:15 -0700941 * Returns the end address of the unmapping (restart addr if interrupted).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700943 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700945 * We aim to not hold locks for too long (for scheduling latency reasons).
946 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 * return the ending mmu_gather to the caller.
948 *
949 * Only addresses between `start' and `end' will be unmapped.
950 *
951 * The VMA list must be sorted in ascending virtual address order.
952 *
953 * unmap_vmas() assumes that the caller will flush the whole unmapped address
954 * range after unmap_vmas() returns. So the only responsibility here is to
955 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
956 * drops the lock and schedules.
957 */
Hugh Dickins508034a2005-10-29 18:16:30 -0700958unsigned long unmap_vmas(struct mmu_gather **tlbp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 struct vm_area_struct *vma, unsigned long start_addr,
960 unsigned long end_addr, unsigned long *nr_accounted,
961 struct zap_details *details)
962{
Robin Holt51c6f662005-11-13 16:06:42 -0800963 long zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
965 int tlb_start_valid = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -0700966 unsigned long start = start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700968 int fullmm = (*tlbp)->fullmm;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700969 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700971 mmu_notifier_invalidate_range_start(mm, start_addr, end_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 unsigned long end;
974
975 start = max(vma->vm_start, start_addr);
976 if (start >= vma->vm_end)
977 continue;
978 end = min(vma->vm_end, end_addr);
979 if (end <= vma->vm_start)
980 continue;
981
982 if (vma->vm_flags & VM_ACCOUNT)
983 *nr_accounted += (end - start) >> PAGE_SHIFT;
984
venkatesh.pallipadi@intel.com34801ba2008-12-19 13:47:29 -0800985 if (unlikely(is_pfn_mapping(vma)))
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -0800986 untrack_pfn_vma(vma, 0, 0);
987
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 while (start != end) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 if (!tlb_start_valid) {
990 tlb_start = start;
991 tlb_start_valid = 1;
992 }
993
Robin Holt51c6f662005-11-13 16:06:42 -0800994 if (unlikely(is_vm_hugetlb_page(vma))) {
Andi Kleena137e1c2008-07-23 21:27:43 -0700995 /*
996 * It is undesirable to test vma->vm_file as it
997 * should be non-null for valid hugetlb area.
998 * However, vm_file will be NULL in the error
999 * cleanup path of do_mmap_pgoff. When
1000 * hugetlbfs ->mmap method fails,
1001 * do_mmap_pgoff() nullifies vma->vm_file
1002 * before calling this function to clean up.
1003 * Since no pte has actually been setup, it is
1004 * safe to do nothing in this case.
1005 */
1006 if (vma->vm_file) {
1007 unmap_hugepage_range(vma, start, end, NULL);
1008 zap_work -= (end - start) /
Andi Kleena5516432008-07-23 21:27:41 -07001009 pages_per_huge_page(hstate_vma(vma));
Andi Kleena137e1c2008-07-23 21:27:43 -07001010 }
1011
Robin Holt51c6f662005-11-13 16:06:42 -08001012 start = end;
1013 } else
1014 start = unmap_page_range(*tlbp, vma,
1015 start, end, &zap_work, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016
Robin Holt51c6f662005-11-13 16:06:42 -08001017 if (zap_work > 0) {
1018 BUG_ON(start != end);
1019 break;
1020 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021
1022 tlb_finish_mmu(*tlbp, tlb_start, start);
1023
1024 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +01001025 (i_mmap_lock && spin_needbreak(i_mmap_lock))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 if (i_mmap_lock) {
Hugh Dickins508034a2005-10-29 18:16:30 -07001027 *tlbp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 goto out;
1029 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 }
1032
Hugh Dickins508034a2005-10-29 18:16:30 -07001033 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 tlb_start_valid = 0;
Robin Holt51c6f662005-11-13 16:06:42 -08001035 zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 }
1037 }
1038out:
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001039 mmu_notifier_invalidate_range_end(mm, start_addr, end_addr);
Hugh Dickinsee39b372005-04-19 13:29:15 -07001040 return start; /* which is now the end (or restart) address */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041}
1042
1043/**
1044 * zap_page_range - remove user pages in a given range
1045 * @vma: vm_area_struct holding the applicable pages
1046 * @address: starting address of pages to zap
1047 * @size: number of bytes to zap
1048 * @details: details of nonlinear truncation or shared cache invalidation
1049 */
Hugh Dickinsee39b372005-04-19 13:29:15 -07001050unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 unsigned long size, struct zap_details *details)
1052{
1053 struct mm_struct *mm = vma->vm_mm;
1054 struct mmu_gather *tlb;
1055 unsigned long end = address + size;
1056 unsigned long nr_accounted = 0;
1057
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -07001060 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -07001061 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
1062 if (tlb)
1063 tlb_finish_mmu(tlb, address, end);
Hugh Dickinsee39b372005-04-19 13:29:15 -07001064 return end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065}
1066
Jack Steinerc627f9c2008-07-29 22:33:53 -07001067/**
1068 * zap_vma_ptes - remove ptes mapping the vma
1069 * @vma: vm_area_struct holding ptes to be zapped
1070 * @address: starting address of pages to zap
1071 * @size: number of bytes to zap
1072 *
1073 * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1074 *
1075 * The entire address range must be fully contained within the vma.
1076 *
1077 * Returns 0 if successful.
1078 */
1079int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
1080 unsigned long size)
1081{
1082 if (address < vma->vm_start || address + size > vma->vm_end ||
1083 !(vma->vm_flags & VM_PFNMAP))
1084 return -1;
1085 zap_page_range(vma, address, size, NULL);
1086 return 0;
1087}
1088EXPORT_SYMBOL_GPL(zap_vma_ptes);
1089
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090/*
1091 * Do a quick page-table lookup for a single page.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 */
Linus Torvalds6aab3412005-11-28 14:34:23 -08001093struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001094 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095{
1096 pgd_t *pgd;
1097 pud_t *pud;
1098 pmd_t *pmd;
1099 pte_t *ptep, pte;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001100 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 struct page *page;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001102 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001104 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
1105 if (!IS_ERR(page)) {
1106 BUG_ON(flags & FOLL_GET);
1107 goto out;
1108 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001110 page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 pgd = pgd_offset(mm, address);
1112 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001113 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114
1115 pud = pud_offset(pgd, address);
Andi Kleenceb86872008-07-23 21:27:50 -07001116 if (pud_none(*pud))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001117 goto no_page_table;
Andi Kleenceb86872008-07-23 21:27:50 -07001118 if (pud_huge(*pud)) {
1119 BUG_ON(flags & FOLL_GET);
1120 page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE);
1121 goto out;
1122 }
1123 if (unlikely(pud_bad(*pud)))
1124 goto no_page_table;
1125
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 pmd = pmd_offset(pud, address);
Hugh Dickinsaeed5fc2008-05-06 20:49:23 +01001127 if (pmd_none(*pmd))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001128 goto no_page_table;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001129 if (pmd_huge(*pmd)) {
1130 BUG_ON(flags & FOLL_GET);
1131 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
1132 goto out;
1133 }
Hugh Dickinsaeed5fc2008-05-06 20:49:23 +01001134 if (unlikely(pmd_bad(*pmd)))
1135 goto no_page_table;
1136
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001137 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
1139 pte = *ptep;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001140 if (!pte_present(pte))
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001141 goto no_page;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001142 if ((flags & FOLL_WRITE) && !pte_write(pte))
1143 goto unlock;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001144 page = vm_normal_page(vma, address, pte);
1145 if (unlikely(!page))
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001146 goto bad_page;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001147
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001148 if (flags & FOLL_GET)
1149 get_page(page);
1150 if (flags & FOLL_TOUCH) {
1151 if ((flags & FOLL_WRITE) &&
1152 !pte_dirty(pte) && !PageDirty(page))
1153 set_page_dirty(page);
KOSAKI Motohirobd775c42009-03-31 15:19:37 -07001154 /*
1155 * pte_mkyoung() would be more correct here, but atomic care
1156 * is needed to avoid losing the dirty bit: it is easier to use
1157 * mark_page_accessed().
1158 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001159 mark_page_accessed(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001161unlock:
1162 pte_unmap_unlock(ptep, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163out:
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001164 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001166bad_page:
1167 pte_unmap_unlock(ptep, ptl);
1168 return ERR_PTR(-EFAULT);
1169
1170no_page:
1171 pte_unmap_unlock(ptep, ptl);
1172 if (!pte_none(pte))
1173 return page;
1174 /* Fall through to ZERO_PAGE handling */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001175no_page_table:
1176 /*
1177 * When core dumping an enormous anonymous area that nobody
1178 * has touched so far, we don't want to allocate page tables.
1179 */
1180 if (flags & FOLL_ANON) {
Nick Piggin557ed1f2007-10-16 01:24:40 -07001181 page = ZERO_PAGE(0);
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001182 if (flags & FOLL_GET)
1183 get_page(page);
1184 BUG_ON(flags & FOLL_WRITE);
1185 }
1186 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187}
1188
Linus Torvalds672ca282008-06-23 11:21:37 -07001189/* Can we do the FOLL_ANON optimization? */
1190static inline int use_zero_page(struct vm_area_struct *vma)
1191{
1192 /*
1193 * We don't want to optimize FOLL_ANON for make_pages_present()
1194 * when it tries to page in a VM_LOCKED region. As to VM_SHARED,
1195 * we want to get the page from the page tables to make sure
1196 * that we serialize and update with any other user of that
1197 * mapping.
1198 */
1199 if (vma->vm_flags & (VM_LOCKED | VM_SHARED))
1200 return 0;
1201 /*
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001202 * And if we have a fault routine, it's not an anonymous region.
Linus Torvalds672ca282008-06-23 11:21:37 -07001203 */
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001204 return !vma->vm_ops || !vma->vm_ops->fault;
Linus Torvalds672ca282008-06-23 11:21:37 -07001205}
1206
Nick Pigginb291f002008-10-18 20:26:44 -07001207
1208
1209int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
1210 unsigned long start, int len, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 struct page **pages, struct vm_area_struct **vmas)
1212{
1213 int i;
Nick Pigginb291f002008-10-18 20:26:44 -07001214 unsigned int vm_flags = 0;
1215 int write = !!(flags & GUP_FLAGS_WRITE);
1216 int force = !!(flags & GUP_FLAGS_FORCE);
1217 int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS);
Ying Han47792802009-01-06 14:40:18 -08001218 int ignore_sigkill = !!(flags & GUP_FLAGS_IGNORE_SIGKILL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219
Jonathan Corbet900cf082008-02-11 16:17:33 -07001220 if (len <= 0)
1221 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 /*
1223 * Require read or write permissions.
1224 * If 'force' is set, we only require the "MAY" flags.
1225 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001226 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
1227 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 i = 0;
1229
1230 do {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001231 struct vm_area_struct *vma;
1232 unsigned int foll_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
1234 vma = find_extend_vma(mm, start);
1235 if (!vma && in_gate_area(tsk, start)) {
1236 unsigned long pg = start & PAGE_MASK;
1237 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
1238 pgd_t *pgd;
1239 pud_t *pud;
1240 pmd_t *pmd;
1241 pte_t *pte;
Nick Pigginb291f002008-10-18 20:26:44 -07001242
1243 /* user gate pages are read-only */
1244 if (!ignore && write)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 return i ? : -EFAULT;
1246 if (pg > TASK_SIZE)
1247 pgd = pgd_offset_k(pg);
1248 else
1249 pgd = pgd_offset_gate(mm, pg);
1250 BUG_ON(pgd_none(*pgd));
1251 pud = pud_offset(pgd, pg);
1252 BUG_ON(pud_none(*pud));
1253 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001254 if (pmd_none(*pmd))
1255 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001257 if (pte_none(*pte)) {
1258 pte_unmap(pte);
1259 return i ? : -EFAULT;
1260 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 if (pages) {
Nick Pigginfa2a4552005-11-29 18:43:17 +11001262 struct page *page = vm_normal_page(gate_vma, start, *pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001263 pages[i] = page;
1264 if (page)
1265 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 }
1267 pte_unmap(pte);
1268 if (vmas)
1269 vmas[i] = gate_vma;
1270 i++;
1271 start += PAGE_SIZE;
1272 len--;
1273 continue;
1274 }
1275
Nick Pigginb291f002008-10-18 20:26:44 -07001276 if (!vma ||
1277 (vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
1278 (!ignore && !(vm_flags & vma->vm_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 return i ? : -EFAULT;
1280
1281 if (is_vm_hugetlb_page(vma)) {
1282 i = follow_hugetlb_page(mm, vma, pages, vmas,
Adam Litke5b23dbe2007-11-14 16:59:33 -08001283 &start, &len, i, write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 continue;
1285 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001286
1287 foll_flags = FOLL_TOUCH;
1288 if (pages)
1289 foll_flags |= FOLL_GET;
Linus Torvalds672ca282008-06-23 11:21:37 -07001290 if (!write && use_zero_page(vma))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001291 foll_flags |= FOLL_ANON;
1292
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 do {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001294 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295
Ethan Solomita462e00c2007-07-15 23:38:16 -07001296 /*
Ying Han47792802009-01-06 14:40:18 -08001297 * If we have a pending SIGKILL, don't keep faulting
1298 * pages and potentially allocating memory, unless
1299 * current is handling munlock--e.g., on exit. In
1300 * that case, we are not allocating memory. Rather,
1301 * we're only unlocking already resident/mapped pages.
Ethan Solomita462e00c2007-07-15 23:38:16 -07001302 */
Ying Han47792802009-01-06 14:40:18 -08001303 if (unlikely(!ignore_sigkill &&
1304 fatal_signal_pending(current)))
1305 return i ? i : -ERESTARTSYS;
Ethan Solomita462e00c2007-07-15 23:38:16 -07001306
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001307 if (write)
1308 foll_flags |= FOLL_WRITE;
1309
1310 cond_resched();
Linus Torvalds6aab3412005-11-28 14:34:23 -08001311 while (!(page = follow_page(vma, start, foll_flags))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001312 int ret;
Nick Piggin83c54072007-07-19 01:47:05 -07001313 ret = handle_mm_fault(mm, vma, start,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001314 foll_flags & FOLL_WRITE);
Nick Piggin83c54072007-07-19 01:47:05 -07001315 if (ret & VM_FAULT_ERROR) {
1316 if (ret & VM_FAULT_OOM)
1317 return i ? i : -ENOMEM;
1318 else if (ret & VM_FAULT_SIGBUS)
1319 return i ? i : -EFAULT;
1320 BUG();
1321 }
1322 if (ret & VM_FAULT_MAJOR)
1323 tsk->maj_flt++;
1324 else
1325 tsk->min_flt++;
1326
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001327 /*
Nick Piggin83c54072007-07-19 01:47:05 -07001328 * The VM_FAULT_WRITE bit tells us that
1329 * do_wp_page has broken COW when necessary,
1330 * even if maybe_mkwrite decided not to set
1331 * pte_write. We can thus safely do subsequent
Hugh Dickins878b63a2009-01-06 14:39:32 -08001332 * page lookups as if they were reads. But only
1333 * do so when looping for pte_write is futile:
1334 * in some cases userspace may also be wanting
1335 * to write to the gotten user page, which a
1336 * read fault here might prevent (a readonly
1337 * page might get reCOWed by userspace write).
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001338 */
Hugh Dickins878b63a2009-01-06 14:39:32 -08001339 if ((ret & VM_FAULT_WRITE) &&
1340 !(vma->vm_flags & VM_WRITE))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001341 foll_flags &= ~FOLL_WRITE;
Nick Piggin83c54072007-07-19 01:47:05 -07001342
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07001343 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 }
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001345 if (IS_ERR(page))
1346 return i ? i : PTR_ERR(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001348 pages[i] = page;
James Bottomley03beb072006-03-26 01:36:57 -08001349
Russell Kinga6f36be2006-12-30 22:24:19 +00001350 flush_anon_page(vma, page, start);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001351 flush_dcache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 }
1353 if (vmas)
1354 vmas[i] = vma;
1355 i++;
1356 start += PAGE_SIZE;
1357 len--;
Hugh Dickins08ef4722005-06-21 17:15:10 -07001358 } while (len && start < vma->vm_end);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001359 } while (len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 return i;
1361}
Nick Pigginb291f002008-10-18 20:26:44 -07001362
1363int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
1364 unsigned long start, int len, int write, int force,
1365 struct page **pages, struct vm_area_struct **vmas)
1366{
1367 int flags = 0;
1368
1369 if (write)
1370 flags |= GUP_FLAGS_WRITE;
1371 if (force)
1372 flags |= GUP_FLAGS_FORCE;
1373
1374 return __get_user_pages(tsk, mm,
1375 start, len, flags,
1376 pages, vmas);
1377}
1378
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379EXPORT_SYMBOL(get_user_pages);
1380
Harvey Harrison920c7a52008-02-04 22:29:26 -08001381pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr,
1382 spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001383{
1384 pgd_t * pgd = pgd_offset(mm, addr);
1385 pud_t * pud = pud_alloc(mm, pgd, addr);
1386 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001387 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001388 if (pmd)
1389 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1390 }
1391 return NULL;
1392}
1393
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001395 * This is the old fallback for page remapping.
1396 *
1397 * For historical reasons, it only allows reserved pages. Only
1398 * old drivers should use this, and they needed to mark their
1399 * pages reserved for the old functions anyway.
1400 */
Nick Piggin423bad62008-04-28 02:13:01 -07001401static int insert_page(struct vm_area_struct *vma, unsigned long addr,
1402 struct page *page, pgprot_t prot)
Linus Torvalds238f58d2005-11-29 13:01:56 -08001403{
Nick Piggin423bad62008-04-28 02:13:01 -07001404 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001405 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001406 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001407 spinlock_t *ptl;
1408
Linus Torvalds238f58d2005-11-29 13:01:56 -08001409 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001410 if (PageAnon(page))
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001411 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001412 retval = -ENOMEM;
1413 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001414 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001415 if (!pte)
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001416 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001417 retval = -EBUSY;
1418 if (!pte_none(*pte))
1419 goto out_unlock;
1420
1421 /* Ok, finally just insert the thing.. */
1422 get_page(page);
1423 inc_mm_counter(mm, file_rss);
1424 page_add_file_rmap(page);
1425 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1426
1427 retval = 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001428 pte_unmap_unlock(pte, ptl);
1429 return retval;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001430out_unlock:
1431 pte_unmap_unlock(pte, ptl);
1432out:
1433 return retval;
1434}
1435
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001436/**
1437 * vm_insert_page - insert single page into user vma
1438 * @vma: user vma to map to
1439 * @addr: target user address of this page
1440 * @page: source kernel page
1441 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001442 * This allows drivers to insert individual pages they've allocated
1443 * into a user vma.
1444 *
1445 * The page has to be a nice clean _individual_ kernel allocation.
1446 * If you allocate a compound page, you need to have marked it as
1447 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001448 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001449 *
1450 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1451 * took an arbitrary page protection parameter. This doesn't allow
1452 * that. Your vma protection will have to be set up correctly, which
1453 * means that if you want a shared writable mapping, you'd better
1454 * ask for a shared writable mapping!
1455 *
1456 * The page does not need to be reserved.
1457 */
Nick Piggin423bad62008-04-28 02:13:01 -07001458int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
1459 struct page *page)
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001460{
1461 if (addr < vma->vm_start || addr >= vma->vm_end)
1462 return -EFAULT;
1463 if (!page_count(page))
1464 return -EINVAL;
Linus Torvalds4d7672b2005-12-16 10:21:23 -08001465 vma->vm_flags |= VM_INSERTPAGE;
Nick Piggin423bad62008-04-28 02:13:01 -07001466 return insert_page(vma, addr, page, vma->vm_page_prot);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001467}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001468EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001469
Nick Piggin423bad62008-04-28 02:13:01 -07001470static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1471 unsigned long pfn, pgprot_t prot)
1472{
1473 struct mm_struct *mm = vma->vm_mm;
1474 int retval;
1475 pte_t *pte, entry;
1476 spinlock_t *ptl;
1477
1478 retval = -ENOMEM;
1479 pte = get_locked_pte(mm, addr, &ptl);
1480 if (!pte)
1481 goto out;
1482 retval = -EBUSY;
1483 if (!pte_none(*pte))
1484 goto out_unlock;
1485
1486 /* Ok, finally just insert the thing.. */
1487 entry = pte_mkspecial(pfn_pte(pfn, prot));
1488 set_pte_at(mm, addr, pte, entry);
1489 update_mmu_cache(vma, addr, entry); /* XXX: why not for insert_page? */
1490
1491 retval = 0;
1492out_unlock:
1493 pte_unmap_unlock(pte, ptl);
1494out:
1495 return retval;
1496}
1497
Nick Piggine0dc0d82007-02-12 00:51:36 -08001498/**
1499 * vm_insert_pfn - insert single pfn into user vma
1500 * @vma: user vma to map to
1501 * @addr: target user address of this page
1502 * @pfn: source kernel pfn
1503 *
1504 * Similar to vm_inert_page, this allows drivers to insert individual pages
1505 * they've allocated into a user vma. Same comments apply.
1506 *
1507 * This function should only be called from a vm_ops->fault handler, and
1508 * in that case the handler should return NULL.
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001509 *
1510 * vma cannot be a COW mapping.
1511 *
1512 * As this is called only for pages that do not currently exist, we
1513 * do not need to flush old virtual caches or the TLB.
Nick Piggine0dc0d82007-02-12 00:51:36 -08001514 */
1515int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
Nick Piggin423bad62008-04-28 02:13:01 -07001516 unsigned long pfn)
Nick Piggine0dc0d82007-02-12 00:51:36 -08001517{
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001518 int ret;
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001519 pgprot_t pgprot = vma->vm_page_prot;
Nick Piggin7e675132008-04-28 02:13:00 -07001520 /*
1521 * Technically, architectures with pte_special can avoid all these
1522 * restrictions (same for remap_pfn_range). However we would like
1523 * consistency in testing and feature parity among all, so we should
1524 * try to keep these invariants in place for everybody.
1525 */
Jared Hulbertb379d792008-04-28 02:12:58 -07001526 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
1527 BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
1528 (VM_PFNMAP|VM_MIXEDMAP));
1529 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1530 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
Nick Piggine0dc0d82007-02-12 00:51:36 -08001531
Nick Piggin423bad62008-04-28 02:13:01 -07001532 if (addr < vma->vm_start || addr >= vma->vm_end)
1533 return -EFAULT;
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001534 if (track_pfn_vma_new(vma, &pgprot, pfn, PAGE_SIZE))
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001535 return -EINVAL;
1536
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001537 ret = insert_pfn(vma, addr, pfn, pgprot);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001538
1539 if (ret)
1540 untrack_pfn_vma(vma, pfn, PAGE_SIZE);
1541
1542 return ret;
Nick Piggine0dc0d82007-02-12 00:51:36 -08001543}
1544EXPORT_SYMBOL(vm_insert_pfn);
1545
Nick Piggin423bad62008-04-28 02:13:01 -07001546int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
1547 unsigned long pfn)
1548{
1549 BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
1550
1551 if (addr < vma->vm_start || addr >= vma->vm_end)
1552 return -EFAULT;
1553
1554 /*
1555 * If we don't have pte special, then we have to use the pfn_valid()
1556 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1557 * refcount the page if pfn_valid is true (hence insert_page rather
1558 * than insert_pfn).
1559 */
1560 if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) {
1561 struct page *page;
1562
1563 page = pfn_to_page(pfn);
1564 return insert_page(vma, addr, page, vma->vm_page_prot);
1565 }
1566 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
1567}
1568EXPORT_SYMBOL(vm_insert_mixed);
1569
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001570/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 * maps a range of physical memory into the requested pages. the old
1572 * mappings are removed. any references to nonexistent pages results
1573 * in null mappings (currently treated as "copy-on-access")
1574 */
1575static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1576 unsigned long addr, unsigned long end,
1577 unsigned long pfn, pgprot_t prot)
1578{
1579 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001580 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581
Hugh Dickinsc74df322005-10-29 18:16:23 -07001582 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 if (!pte)
1584 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001585 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 do {
1587 BUG_ON(!pte_none(*pte));
Nick Piggin7e675132008-04-28 02:13:00 -07001588 set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 pfn++;
1590 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001591 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001592 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 return 0;
1594}
1595
1596static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1597 unsigned long addr, unsigned long end,
1598 unsigned long pfn, pgprot_t prot)
1599{
1600 pmd_t *pmd;
1601 unsigned long next;
1602
1603 pfn -= addr >> PAGE_SHIFT;
1604 pmd = pmd_alloc(mm, pud, addr);
1605 if (!pmd)
1606 return -ENOMEM;
1607 do {
1608 next = pmd_addr_end(addr, end);
1609 if (remap_pte_range(mm, pmd, addr, next,
1610 pfn + (addr >> PAGE_SHIFT), prot))
1611 return -ENOMEM;
1612 } while (pmd++, addr = next, addr != end);
1613 return 0;
1614}
1615
1616static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1617 unsigned long addr, unsigned long end,
1618 unsigned long pfn, pgprot_t prot)
1619{
1620 pud_t *pud;
1621 unsigned long next;
1622
1623 pfn -= addr >> PAGE_SHIFT;
1624 pud = pud_alloc(mm, pgd, addr);
1625 if (!pud)
1626 return -ENOMEM;
1627 do {
1628 next = pud_addr_end(addr, end);
1629 if (remap_pmd_range(mm, pud, addr, next,
1630 pfn + (addr >> PAGE_SHIFT), prot))
1631 return -ENOMEM;
1632 } while (pud++, addr = next, addr != end);
1633 return 0;
1634}
1635
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001636/**
1637 * remap_pfn_range - remap kernel memory to userspace
1638 * @vma: user vma to map to
1639 * @addr: target user address to start at
1640 * @pfn: physical address of kernel memory
1641 * @size: size of map area
1642 * @prot: page protection flags for this mapping
1643 *
1644 * Note: this is only safe if the mm semaphore is held when called.
1645 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1647 unsigned long pfn, unsigned long size, pgprot_t prot)
1648{
1649 pgd_t *pgd;
1650 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001651 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 struct mm_struct *mm = vma->vm_mm;
1653 int err;
1654
1655 /*
1656 * Physically remapped pages are special. Tell the
1657 * rest of the world about it:
1658 * VM_IO tells people not to look at these pages
1659 * (accesses can have side effects).
Hugh Dickins0b14c172005-11-21 21:32:15 -08001660 * VM_RESERVED is specified all over the place, because
1661 * in 2.4 it kept swapout's vma scan off this vma; but
1662 * in 2.6 the LRU scan won't even find its pages, so this
1663 * flag means no more than count its pages in reserved_vm,
1664 * and omit it from core dump, even when VM_IO turned off.
Linus Torvalds6aab3412005-11-28 14:34:23 -08001665 * VM_PFNMAP tells the core MM that the base pages are just
1666 * raw PFN mappings, and do not have a "struct page" associated
1667 * with them.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001668 *
1669 * There's a horrible special case to handle copy-on-write
1670 * behaviour that some programs depend on. We mark the "original"
1671 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 */
Pallipadi, Venkatesh4bb9c5c2009-03-12 17:45:27 -07001673 if (addr == vma->vm_start && end == vma->vm_end) {
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001674 vma->vm_pgoff = pfn;
Pallipadi, Venkatesh895791d2009-03-13 16:35:44 -07001675 vma->vm_flags |= VM_PFN_AT_MMAP;
Pallipadi, Venkatesh4bb9c5c2009-03-12 17:45:27 -07001676 } else if (is_cow_mapping(vma->vm_flags))
venkatesh.pallipadi@intel.com3c8bb732008-12-18 11:41:27 -08001677 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001678
Linus Torvalds6aab3412005-11-28 14:34:23 -08001679 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001681 err = track_pfn_vma_new(vma, &prot, pfn, PAGE_ALIGN(size));
venkatesh.pallipadi@intel.coma3670612009-01-09 16:13:09 -08001682 if (err) {
1683 /*
1684 * To indicate that track_pfn related cleanup is not
1685 * needed from higher level routine calling unmap_vmas
1686 */
1687 vma->vm_flags &= ~(VM_IO | VM_RESERVED | VM_PFNMAP);
Pallipadi, Venkatesh895791d2009-03-13 16:35:44 -07001688 vma->vm_flags &= ~VM_PFN_AT_MMAP;
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001689 return -EINVAL;
venkatesh.pallipadi@intel.coma3670612009-01-09 16:13:09 -08001690 }
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001691
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 BUG_ON(addr >= end);
1693 pfn -= addr >> PAGE_SHIFT;
1694 pgd = pgd_offset(mm, addr);
1695 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 do {
1697 next = pgd_addr_end(addr, end);
1698 err = remap_pud_range(mm, pgd, addr, next,
1699 pfn + (addr >> PAGE_SHIFT), prot);
1700 if (err)
1701 break;
1702 } while (pgd++, addr = next, addr != end);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001703
1704 if (err)
1705 untrack_pfn_vma(vma, pfn, PAGE_ALIGN(size));
1706
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 return err;
1708}
1709EXPORT_SYMBOL(remap_pfn_range);
1710
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001711static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1712 unsigned long addr, unsigned long end,
1713 pte_fn_t fn, void *data)
1714{
1715 pte_t *pte;
1716 int err;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001717 pgtable_t token;
Borislav Petkov94909912007-05-06 14:49:17 -07001718 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001719
1720 pte = (mm == &init_mm) ?
1721 pte_alloc_kernel(pmd, addr) :
1722 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1723 if (!pte)
1724 return -ENOMEM;
1725
1726 BUG_ON(pmd_huge(*pmd));
1727
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08001728 arch_enter_lazy_mmu_mode();
1729
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001730 token = pmd_pgtable(*pmd);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001731
1732 do {
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001733 err = fn(pte, token, addr, data);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001734 if (err)
1735 break;
1736 } while (pte++, addr += PAGE_SIZE, addr != end);
1737
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08001738 arch_leave_lazy_mmu_mode();
1739
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001740 if (mm != &init_mm)
1741 pte_unmap_unlock(pte-1, ptl);
1742 return err;
1743}
1744
1745static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1746 unsigned long addr, unsigned long end,
1747 pte_fn_t fn, void *data)
1748{
1749 pmd_t *pmd;
1750 unsigned long next;
1751 int err;
1752
Andi Kleenceb86872008-07-23 21:27:50 -07001753 BUG_ON(pud_huge(*pud));
1754
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001755 pmd = pmd_alloc(mm, pud, addr);
1756 if (!pmd)
1757 return -ENOMEM;
1758 do {
1759 next = pmd_addr_end(addr, end);
1760 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1761 if (err)
1762 break;
1763 } while (pmd++, addr = next, addr != end);
1764 return err;
1765}
1766
1767static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1768 unsigned long addr, unsigned long end,
1769 pte_fn_t fn, void *data)
1770{
1771 pud_t *pud;
1772 unsigned long next;
1773 int err;
1774
1775 pud = pud_alloc(mm, pgd, addr);
1776 if (!pud)
1777 return -ENOMEM;
1778 do {
1779 next = pud_addr_end(addr, end);
1780 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1781 if (err)
1782 break;
1783 } while (pud++, addr = next, addr != end);
1784 return err;
1785}
1786
1787/*
1788 * Scan a region of virtual memory, filling in page tables as necessary
1789 * and calling a provided function on each leaf page table.
1790 */
1791int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1792 unsigned long size, pte_fn_t fn, void *data)
1793{
1794 pgd_t *pgd;
1795 unsigned long next;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001796 unsigned long start = addr, end = addr + size;
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001797 int err;
1798
1799 BUG_ON(addr >= end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001800 mmu_notifier_invalidate_range_start(mm, start, end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001801 pgd = pgd_offset(mm, addr);
1802 do {
1803 next = pgd_addr_end(addr, end);
1804 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1805 if (err)
1806 break;
1807 } while (pgd++, addr = next, addr != end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001808 mmu_notifier_invalidate_range_end(mm, start, end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001809 return err;
1810}
1811EXPORT_SYMBOL_GPL(apply_to_page_range);
1812
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001814 * handle_pte_fault chooses page fault handler according to an entry
1815 * which was read non-atomically. Before making any commitment, on
1816 * those architectures or configurations (e.g. i386 with PAE) which
1817 * might give a mix of unmatched parts, do_swap_page and do_file_page
1818 * must check under lock before unmapping the pte and proceeding
1819 * (but do_wp_page is only called after already making such a check;
1820 * and do_anonymous_page and do_no_page can safely check later on).
1821 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001822static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001823 pte_t *page_table, pte_t orig_pte)
1824{
1825 int same = 1;
1826#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1827 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001828 spinlock_t *ptl = pte_lockptr(mm, pmd);
1829 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001830 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001831 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001832 }
1833#endif
1834 pte_unmap(page_table);
1835 return same;
1836}
1837
1838/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1840 * servicing faults for write access. In the normal case, do always want
1841 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1842 * that do not have writing enabled, when used by access_process_vm.
1843 */
1844static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1845{
1846 if (likely(vma->vm_flags & VM_WRITE))
1847 pte = pte_mkwrite(pte);
1848 return pte;
1849}
1850
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001851static 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 -08001852{
1853 /*
1854 * If the source page was a PFN mapping, we don't have
1855 * a "struct page" for it. We do a best-effort copy by
1856 * just copying from the original user address. If that
1857 * fails, we just zero-fill it. Live with it.
1858 */
1859 if (unlikely(!src)) {
1860 void *kaddr = kmap_atomic(dst, KM_USER0);
Linus Torvalds5d2a2dbbc2005-11-29 14:07:55 -08001861 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1862
1863 /*
1864 * This really shouldn't fail, because the page is there
1865 * in the page tables. But it might just be unreadable,
1866 * in which case we just give up and fill the result with
1867 * zeroes.
1868 */
1869 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Linus Torvalds6aab3412005-11-28 14:34:23 -08001870 memset(kaddr, 0, PAGE_SIZE);
1871 kunmap_atomic(kaddr, KM_USER0);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07001872 flush_dcache_page(dst);
Nick Piggin0ed361d2008-02-04 22:29:34 -08001873 } else
1874 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001875}
1876
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 * This routine handles present pages, when users try to write
1879 * to a shared page. It is done by copying the page to a new address
1880 * and decrementing the shared-page counter for the old page.
1881 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 * Note that this routine assumes that the protection checks have been
1883 * done by the caller (the low-level page fault routine in most cases).
1884 * Thus we can safely just mark it writable once we've done any necessary
1885 * COW.
1886 *
1887 * We also mark the page dirty at this point even though the page will
1888 * change only once the write actually happens. This avoids a few races,
1889 * and potentially makes it more efficient.
1890 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001891 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1892 * but allow concurrent faults), with pte both mapped and locked.
1893 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001895static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1896 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001897 spinlock_t *ptl, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898{
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001899 struct page *old_page, *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 pte_t entry;
Nick Piggin83c54072007-07-19 01:47:05 -07001901 int reuse = 0, ret = 0;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001902 int page_mkwrite = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001903 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904
Linus Torvalds6aab3412005-11-28 14:34:23 -08001905 old_page = vm_normal_page(vma, address, orig_pte);
Peter Zijlstra251b97f2008-07-04 09:59:24 -07001906 if (!old_page) {
1907 /*
1908 * VM_MIXEDMAP !pfn_valid() case
1909 *
1910 * We should not cow pages in a shared writeable mapping.
1911 * Just mark the pages writable as we can't do any dirty
1912 * accounting on raw pfn maps.
1913 */
1914 if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
1915 (VM_WRITE|VM_SHARED))
1916 goto reuse;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001917 goto gotten;
Peter Zijlstra251b97f2008-07-04 09:59:24 -07001918 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001920 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001921 * Take out anonymous pages first, anonymous shared vmas are
1922 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001923 */
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001924 if (PageAnon(old_page)) {
Hugh Dickinsab967d82009-01-06 14:39:33 -08001925 if (!trylock_page(old_page)) {
1926 page_cache_get(old_page);
1927 pte_unmap_unlock(page_table, ptl);
1928 lock_page(old_page);
1929 page_table = pte_offset_map_lock(mm, pmd, address,
1930 &ptl);
1931 if (!pte_same(*page_table, orig_pte)) {
1932 unlock_page(old_page);
1933 page_cache_release(old_page);
1934 goto unlock;
1935 }
1936 page_cache_release(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001937 }
Hugh Dickins7b1fe592009-01-06 14:39:34 -08001938 reuse = reuse_swap_page(old_page);
Hugh Dickinsab967d82009-01-06 14:39:33 -08001939 unlock_page(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001940 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001941 (VM_WRITE|VM_SHARED))) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001942 /*
1943 * Only catch write-faults on shared writable pages,
1944 * read-only shared pages can get COWed by
1945 * get_user_pages(.write=1, .force=1).
1946 */
David Howells9637a5e2006-06-23 02:03:43 -07001947 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
Nick Pigginc2ec1752009-03-31 15:23:21 -07001948 struct vm_fault vmf;
1949 int tmp;
1950
1951 vmf.virtual_address = (void __user *)(address &
1952 PAGE_MASK);
1953 vmf.pgoff = old_page->index;
1954 vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
1955 vmf.page = old_page;
1956
David Howells9637a5e2006-06-23 02:03:43 -07001957 /*
1958 * Notify the address space that the page is about to
1959 * become writable so that it can prohibit this or wait
1960 * for the page to get into an appropriate state.
1961 *
1962 * We do this without the lock held, so that it can
1963 * sleep if it needs to.
1964 */
1965 page_cache_get(old_page);
1966 pte_unmap_unlock(page_table, ptl);
1967
Nick Pigginc2ec1752009-03-31 15:23:21 -07001968 tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
1969 if (unlikely(tmp &
1970 (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
1971 ret = tmp;
David Howells9637a5e2006-06-23 02:03:43 -07001972 goto unwritable_page;
Nick Pigginc2ec1752009-03-31 15:23:21 -07001973 }
Nick Pigginb827e492009-04-30 15:08:16 -07001974 if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
1975 lock_page(old_page);
1976 if (!old_page->mapping) {
1977 ret = 0; /* retry the fault */
1978 unlock_page(old_page);
1979 goto unwritable_page;
1980 }
1981 } else
1982 VM_BUG_ON(!PageLocked(old_page));
David Howells9637a5e2006-06-23 02:03:43 -07001983
David Howells9637a5e2006-06-23 02:03:43 -07001984 /*
1985 * Since we dropped the lock we need to revalidate
1986 * the PTE as someone else may have changed it. If
1987 * they did, we just return, as we can count on the
1988 * MMU to tell us if they didn't also make it writable.
1989 */
1990 page_table = pte_offset_map_lock(mm, pmd, address,
1991 &ptl);
Nick Pigginb827e492009-04-30 15:08:16 -07001992 if (!pte_same(*page_table, orig_pte)) {
1993 unlock_page(old_page);
1994 page_cache_release(old_page);
David Howells9637a5e2006-06-23 02:03:43 -07001995 goto unlock;
Nick Pigginb827e492009-04-30 15:08:16 -07001996 }
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001997
1998 page_mkwrite = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002000 dirty_page = old_page;
2001 get_page(dirty_page);
David Howells9637a5e2006-06-23 02:03:43 -07002002 reuse = 1;
David Howells9637a5e2006-06-23 02:03:43 -07002003 }
2004
2005 if (reuse) {
Peter Zijlstra251b97f2008-07-04 09:59:24 -07002006reuse:
David Howells9637a5e2006-06-23 02:03:43 -07002007 flush_cache_page(vma, address, pte_pfn(orig_pte));
2008 entry = pte_mkyoung(orig_pte);
2009 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
KAMEZAWA Hiroyuki954ffcb2007-10-16 01:25:44 -07002010 if (ptep_set_access_flags(vma, address, page_table, entry,1))
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002011 update_mmu_cache(vma, address, entry);
David Howells9637a5e2006-06-23 02:03:43 -07002012 ret |= VM_FAULT_WRITE;
2013 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015
2016 /*
2017 * Ok, we need to copy. Oh, well..
2018 */
Nick Pigginb5810032005-10-29 18:16:12 -07002019 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08002020gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002021 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022
2023 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07002024 goto oom;
Nick Piggin557ed1f2007-10-16 01:24:40 -07002025 VM_BUG_ON(old_page == ZERO_PAGE(0));
2026 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
2027 if (!new_page)
2028 goto oom;
Nick Pigginb291f002008-10-18 20:26:44 -07002029 /*
2030 * Don't let another task, with possibly unlocked vma,
2031 * keep the mlocked page.
2032 */
Carsten Otteab926612009-02-04 15:12:16 -08002033 if ((vma->vm_flags & VM_LOCKED) && old_page) {
Nick Pigginb291f002008-10-18 20:26:44 -07002034 lock_page(old_page); /* for LRU manipulation */
2035 clear_page_mlock(old_page);
2036 unlock_page(old_page);
2037 }
Nick Piggin557ed1f2007-10-16 01:24:40 -07002038 cow_user_page(new_page, old_page, address, vma);
Nick Piggin0ed361d2008-02-04 22:29:34 -08002039 __SetPageUptodate(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002040
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002041 if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002042 goto oom_free_new;
2043
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 /*
2045 * Re-check the pte - we dropped the lock
2046 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002047 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07002048 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08002049 if (old_page) {
Hugh Dickins920fc352005-11-21 21:32:17 -08002050 if (!PageAnon(old_page)) {
2051 dec_mm_counter(mm, file_rss);
2052 inc_mm_counter(mm, anon_rss);
2053 }
2054 } else
Hugh Dickins42946212005-10-29 18:16:05 -07002055 inc_mm_counter(mm, anon_rss);
Ben Collinseca35132005-11-29 11:45:26 -08002056 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07002057 entry = mk_pte(new_page, vma->vm_page_prot);
2058 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Siddha, Suresh B4ce072f2006-09-29 01:58:42 -07002059 /*
2060 * Clear the pte entry and flush it first, before updating the
2061 * pte with the new entry. This will avoid a race condition
2062 * seen in the presence of one thread doing SMC and another
2063 * thread doing COW.
2064 */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002065 ptep_clear_flush_notify(vma, address, page_table);
Nick Piggin9617d952006-01-06 00:11:12 -08002066 page_add_new_anon_rmap(new_page, vma, address);
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002067 set_pte_at(mm, address, page_table, entry);
2068 update_mmu_cache(vma, address, entry);
Nick Piggin945754a2008-06-23 14:30:30 +02002069 if (old_page) {
2070 /*
2071 * Only after switching the pte to the new page may
2072 * we remove the mapcount here. Otherwise another
2073 * process may come and find the rmap count decremented
2074 * before the pte is switched to the new page, and
2075 * "reuse" the old page writing into it while our pte
2076 * here still points into it and can be read by other
2077 * threads.
2078 *
2079 * The critical issue is to order this
2080 * page_remove_rmap with the ptp_clear_flush above.
2081 * Those stores are ordered by (if nothing else,)
2082 * the barrier present in the atomic_add_negative
2083 * in page_remove_rmap.
2084 *
2085 * Then the TLB flush in ptep_clear_flush ensures that
2086 * no process can access the old page before the
2087 * decremented mapcount is visible. And the old page
2088 * cannot be reused until after the decremented
2089 * mapcount is visible. So transitively, TLBs to
2090 * old page will be flushed before it can be reused.
2091 */
Hugh Dickinsedc315f2009-01-06 14:40:11 -08002092 page_remove_rmap(old_page);
Nick Piggin945754a2008-06-23 14:30:30 +02002093 }
2094
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 /* Free the old page.. */
2096 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10002097 ret |= VM_FAULT_WRITE;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002098 } else
2099 mem_cgroup_uncharge_page(new_page);
2100
Hugh Dickins920fc352005-11-21 21:32:17 -08002101 if (new_page)
2102 page_cache_release(new_page);
2103 if (old_page)
2104 page_cache_release(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002105unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002106 pte_unmap_unlock(page_table, ptl);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002107 if (dirty_page) {
Nick Piggin79352892007-07-19 01:47:22 -07002108 /*
2109 * Yes, Virginia, this is actually required to prevent a race
2110 * with clear_page_dirty_for_io() from clearing the page dirty
2111 * bit after it clear all dirty ptes, but before a racing
2112 * do_wp_page installs a dirty pte.
2113 *
2114 * do_no_page is protected similarly.
2115 */
Nick Pigginb827e492009-04-30 15:08:16 -07002116 if (!page_mkwrite) {
2117 wait_on_page_locked(dirty_page);
2118 set_page_dirty_balance(dirty_page, page_mkwrite);
2119 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002120 put_page(dirty_page);
Nick Pigginb827e492009-04-30 15:08:16 -07002121 if (page_mkwrite) {
2122 struct address_space *mapping = dirty_page->mapping;
2123
2124 set_page_dirty(dirty_page);
2125 unlock_page(dirty_page);
2126 page_cache_release(dirty_page);
2127 if (mapping) {
2128 /*
2129 * Some device drivers do not set page.mapping
2130 * but still dirty their pages
2131 */
2132 balance_dirty_pages_ratelimited(mapping);
2133 }
2134 }
2135
2136 /* file_update_time outside page_lock */
2137 if (vma->vm_file)
2138 file_update_time(vma->vm_file);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002139 }
Nick Pigginf33ea7f2005-08-03 20:24:01 +10002140 return ret;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002141oom_free_new:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002142 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002143oom:
Nick Pigginb827e492009-04-30 15:08:16 -07002144 if (old_page) {
2145 if (page_mkwrite) {
2146 unlock_page(old_page);
2147 page_cache_release(old_page);
2148 }
Hugh Dickins920fc352005-11-21 21:32:17 -08002149 page_cache_release(old_page);
Nick Pigginb827e492009-04-30 15:08:16 -07002150 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 return VM_FAULT_OOM;
David Howells9637a5e2006-06-23 02:03:43 -07002152
2153unwritable_page:
2154 page_cache_release(old_page);
Nick Pigginc2ec1752009-03-31 15:23:21 -07002155 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156}
2157
2158/*
2159 * Helper functions for unmap_mapping_range().
2160 *
2161 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
2162 *
2163 * We have to restart searching the prio_tree whenever we drop the lock,
2164 * since the iterator is only valid while the lock is held, and anyway
2165 * a later vma might be split and reinserted earlier while lock dropped.
2166 *
2167 * The list of nonlinear vmas could be handled more efficiently, using
2168 * a placeholder, but handle it in the same way until a need is shown.
2169 * It is important to search the prio_tree before nonlinear list: a vma
2170 * may become nonlinear and be shifted from prio_tree to nonlinear list
2171 * while the lock is dropped; but never shifted from list to prio_tree.
2172 *
2173 * In order to make forward progress despite restarting the search,
2174 * vm_truncate_count is used to mark a vma as now dealt with, so we can
2175 * quickly skip it next time around. Since the prio_tree search only
2176 * shows us those vmas affected by unmapping the range in question, we
2177 * can't efficiently keep all vmas in step with mapping->truncate_count:
2178 * so instead reset them all whenever it wraps back to 0 (then go to 1).
2179 * mapping->truncate_count and vma->vm_truncate_count are protected by
2180 * i_mmap_lock.
2181 *
2182 * In order to make forward progress despite repeatedly restarting some
Hugh Dickinsee39b372005-04-19 13:29:15 -07002183 * large vma, note the restart_addr from unmap_vmas when it breaks out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184 * and restart from that address when we reach that vma again. It might
2185 * have been split or merged, shrunk or extended, but never shifted: so
2186 * restart_addr remains valid so long as it remains in the vma's range.
2187 * unmap_mapping_range forces truncate_count to leap over page-aligned
2188 * values so we can save vma's restart_addr in its truncate_count field.
2189 */
2190#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
2191
2192static void reset_vma_truncate_counts(struct address_space *mapping)
2193{
2194 struct vm_area_struct *vma;
2195 struct prio_tree_iter iter;
2196
2197 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
2198 vma->vm_truncate_count = 0;
2199 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
2200 vma->vm_truncate_count = 0;
2201}
2202
2203static int unmap_mapping_range_vma(struct vm_area_struct *vma,
2204 unsigned long start_addr, unsigned long end_addr,
2205 struct zap_details *details)
2206{
2207 unsigned long restart_addr;
2208 int need_break;
2209
Nick Piggind00806b2007-07-19 01:46:57 -07002210 /*
2211 * files that support invalidating or truncating portions of the
Nick Piggind0217ac2007-07-19 01:47:03 -07002212 * file from under mmaped areas must have their ->fault function
Nick Piggin83c54072007-07-19 01:47:05 -07002213 * return a locked page (and set VM_FAULT_LOCKED in the return).
2214 * This provides synchronisation against concurrent unmapping here.
Nick Piggind00806b2007-07-19 01:46:57 -07002215 */
Nick Piggind00806b2007-07-19 01:46:57 -07002216
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217again:
2218 restart_addr = vma->vm_truncate_count;
2219 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
2220 start_addr = restart_addr;
2221 if (start_addr >= end_addr) {
2222 /* Top of vma has been split off since last time */
2223 vma->vm_truncate_count = details->truncate_count;
2224 return 0;
2225 }
2226 }
2227
Hugh Dickinsee39b372005-04-19 13:29:15 -07002228 restart_addr = zap_page_range(vma, start_addr,
2229 end_addr - start_addr, details);
Nick Piggin95c354f2008-01-30 13:31:20 +01002230 need_break = need_resched() || spin_needbreak(details->i_mmap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231
Hugh Dickinsee39b372005-04-19 13:29:15 -07002232 if (restart_addr >= end_addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 /* We have now completed this vma: mark it so */
2234 vma->vm_truncate_count = details->truncate_count;
2235 if (!need_break)
2236 return 0;
2237 } else {
2238 /* Note restart_addr in vma's truncate_count field */
Hugh Dickinsee39b372005-04-19 13:29:15 -07002239 vma->vm_truncate_count = restart_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 if (!need_break)
2241 goto again;
2242 }
2243
2244 spin_unlock(details->i_mmap_lock);
2245 cond_resched();
2246 spin_lock(details->i_mmap_lock);
2247 return -EINTR;
2248}
2249
2250static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
2251 struct zap_details *details)
2252{
2253 struct vm_area_struct *vma;
2254 struct prio_tree_iter iter;
2255 pgoff_t vba, vea, zba, zea;
2256
2257restart:
2258 vma_prio_tree_foreach(vma, &iter, root,
2259 details->first_index, details->last_index) {
2260 /* Skip quickly over those we have already dealt with */
2261 if (vma->vm_truncate_count == details->truncate_count)
2262 continue;
2263
2264 vba = vma->vm_pgoff;
2265 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
2266 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
2267 zba = details->first_index;
2268 if (zba < vba)
2269 zba = vba;
2270 zea = details->last_index;
2271 if (zea > vea)
2272 zea = vea;
2273
2274 if (unmap_mapping_range_vma(vma,
2275 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
2276 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
2277 details) < 0)
2278 goto restart;
2279 }
2280}
2281
2282static inline void unmap_mapping_range_list(struct list_head *head,
2283 struct zap_details *details)
2284{
2285 struct vm_area_struct *vma;
2286
2287 /*
2288 * In nonlinear VMAs there is no correspondence between virtual address
2289 * offset and file offset. So we must perform an exhaustive search
2290 * across *all* the pages in each nonlinear VMA, not just the pages
2291 * whose virtual address lies outside the file truncation point.
2292 */
2293restart:
2294 list_for_each_entry(vma, head, shared.vm_set.list) {
2295 /* Skip quickly over those we have already dealt with */
2296 if (vma->vm_truncate_count == details->truncate_count)
2297 continue;
2298 details->nonlinear_vma = vma;
2299 if (unmap_mapping_range_vma(vma, vma->vm_start,
2300 vma->vm_end, details) < 0)
2301 goto restart;
2302 }
2303}
2304
2305/**
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002306 * 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 -07002307 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 * @holebegin: byte in first page to unmap, relative to the start of
2309 * the underlying file. This will be rounded down to a PAGE_SIZE
2310 * boundary. Note that this is different from vmtruncate(), which
2311 * must keep the partial page. In contrast, we must get rid of
2312 * partial pages.
2313 * @holelen: size of prospective hole in bytes. This will be rounded
2314 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2315 * end of the file.
2316 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2317 * but 0 when invalidating pagecache, don't throw away private data.
2318 */
2319void unmap_mapping_range(struct address_space *mapping,
2320 loff_t const holebegin, loff_t const holelen, int even_cows)
2321{
2322 struct zap_details details;
2323 pgoff_t hba = holebegin >> PAGE_SHIFT;
2324 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2325
2326 /* Check for overflow. */
2327 if (sizeof(holelen) > sizeof(hlen)) {
2328 long long holeend =
2329 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2330 if (holeend & ~(long long)ULONG_MAX)
2331 hlen = ULONG_MAX - hba + 1;
2332 }
2333
2334 details.check_mapping = even_cows? NULL: mapping;
2335 details.nonlinear_vma = NULL;
2336 details.first_index = hba;
2337 details.last_index = hba + hlen - 1;
2338 if (details.last_index < details.first_index)
2339 details.last_index = ULONG_MAX;
2340 details.i_mmap_lock = &mapping->i_mmap_lock;
2341
2342 spin_lock(&mapping->i_mmap_lock);
2343
Nick Piggind00806b2007-07-19 01:46:57 -07002344 /* Protect against endless unmapping loops */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 mapping->truncate_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 if (unlikely(is_restart_addr(mapping->truncate_count))) {
2347 if (mapping->truncate_count == 0)
2348 reset_vma_truncate_counts(mapping);
2349 mapping->truncate_count++;
2350 }
2351 details.truncate_count = mapping->truncate_count;
2352
2353 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
2354 unmap_mapping_range_tree(&mapping->i_mmap, &details);
2355 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
2356 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
2357 spin_unlock(&mapping->i_mmap_lock);
2358}
2359EXPORT_SYMBOL(unmap_mapping_range);
2360
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07002361/**
2362 * vmtruncate - unmap mappings "freed" by truncate() syscall
2363 * @inode: inode of the file used
2364 * @offset: file offset to start truncating
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365 *
2366 * NOTE! We have to be ready to update the memory sharing
2367 * between the file and the memory map for a potential last
2368 * incomplete page. Ugly, but necessary.
2369 */
2370int vmtruncate(struct inode * inode, loff_t offset)
2371{
Christoph Hellwig61d50482008-02-04 22:28:56 -08002372 if (inode->i_size < offset) {
2373 unsigned long limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374
Christoph Hellwig61d50482008-02-04 22:28:56 -08002375 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
2376 if (limit != RLIM_INFINITY && offset > limit)
2377 goto out_sig;
2378 if (offset > inode->i_sb->s_maxbytes)
2379 goto out_big;
2380 i_size_write(inode, offset);
2381 } else {
2382 struct address_space *mapping = inode->i_mapping;
Nick Piggind00806b2007-07-19 01:46:57 -07002383
Christoph Hellwig61d50482008-02-04 22:28:56 -08002384 /*
2385 * truncation of in-use swapfiles is disallowed - it would
2386 * cause subsequent swapout to scribble on the now-freed
2387 * blocks.
2388 */
2389 if (IS_SWAPFILE(inode))
2390 return -ETXTBSY;
2391 i_size_write(inode, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392
Christoph Hellwig61d50482008-02-04 22:28:56 -08002393 /*
2394 * unmap_mapping_range is called twice, first simply for
2395 * efficiency so that truncate_inode_pages does fewer
2396 * single-page unmaps. However after this first call, and
2397 * before truncate_inode_pages finishes, it is possible for
2398 * private pages to be COWed, which remain after
2399 * truncate_inode_pages finishes, hence the second
2400 * unmap_mapping_range call must be made for correctness.
2401 */
2402 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2403 truncate_inode_pages(mapping, offset);
2404 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2405 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406
Al Viroacfa4382008-12-04 10:06:33 -05002407 if (inode->i_op->truncate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 inode->i_op->truncate(inode);
2409 return 0;
Christoph Hellwig61d50482008-02-04 22:28:56 -08002410
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411out_sig:
2412 send_sig(SIGXFSZ, current, 0);
2413out_big:
2414 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416EXPORT_SYMBOL(vmtruncate);
2417
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002418int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
2419{
2420 struct address_space *mapping = inode->i_mapping;
2421
2422 /*
2423 * If the underlying filesystem is not going to provide
2424 * a way to truncate a range of blocks (punch a hole) -
2425 * we should return failure right now.
2426 */
Al Viroacfa4382008-12-04 10:06:33 -05002427 if (!inode->i_op->truncate_range)
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002428 return -ENOSYS;
2429
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002430 mutex_lock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002431 down_write(&inode->i_alloc_sem);
2432 unmap_mapping_range(mapping, offset, (end - offset), 1);
2433 truncate_inode_pages_range(mapping, offset, end);
Nick Piggind00806b2007-07-19 01:46:57 -07002434 unmap_mapping_range(mapping, offset, (end - offset), 1);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002435 inode->i_op->truncate_range(inode, offset, end);
2436 up_write(&inode->i_alloc_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002437 mutex_unlock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002438
2439 return 0;
2440}
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002441
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002443 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2444 * but allow concurrent faults), and pte mapped but not yet locked.
2445 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002447static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2448 unsigned long address, pte_t *page_table, pmd_t *pmd,
2449 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002451 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452 struct page *page;
Hugh Dickins65500d22005-10-29 18:15:59 -07002453 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 pte_t pte;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002455 struct mem_cgroup *ptr = NULL;
Nick Piggin83c54072007-07-19 01:47:05 -07002456 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002458 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002459 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002460
2461 entry = pte_to_swp_entry(orig_pte);
Christoph Lameter06972122006-06-23 02:03:35 -07002462 if (is_migration_entry(entry)) {
2463 migration_entry_wait(mm, pmd, address);
2464 goto out;
2465 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002466 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 page = lookup_swap_cache(entry);
2468 if (!page) {
Ashwin Chaugule098fe652006-12-06 20:31:54 -08002469 grab_swap_token(); /* Contend for token _before_ read-in */
Hugh Dickins02098fe2008-02-04 22:28:42 -08002470 page = swapin_readahead(entry,
2471 GFP_HIGHUSER_MOVABLE, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 if (!page) {
2473 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002474 * Back out if somebody else faulted in this pte
2475 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002477 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 if (likely(pte_same(*page_table, orig_pte)))
2479 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002480 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002481 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 }
2483
2484 /* Had to read the page from swap area: Major fault */
2485 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002486 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487 }
2488
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 lock_page(page);
Balbir Singh20a10222007-11-14 17:00:33 -08002490 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002492 if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) {
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002493 ret = VM_FAULT_OOM;
Johannes Weinerbc43f752009-04-30 15:08:08 -07002494 goto out_page;
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002495 }
2496
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002498 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002500 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002501 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002502 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002503
2504 if (unlikely(!PageUptodate(page))) {
2505 ret = VM_FAULT_SIGBUS;
2506 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 }
2508
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002509 /*
2510 * The page isn't present yet, go ahead with the fault.
2511 *
2512 * Be careful about the sequence of operations here.
2513 * To get its accounting right, reuse_swap_page() must be called
2514 * while the page is counted on swap but not yet in mapcount i.e.
2515 * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
2516 * must be called after the swap_free(), or it will never succeed.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002517 * Because delete_from_swap_page() may be called by reuse_swap_page(),
2518 * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry
2519 * in page->private. In this case, a record in swap_cgroup is silently
2520 * discarded at swap_free().
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002521 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522
Hugh Dickins42946212005-10-29 18:16:05 -07002523 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 pte = mk_pte(page, vma->vm_page_prot);
Hugh Dickins7b1fe592009-01-06 14:39:34 -08002525 if (write_access && reuse_swap_page(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2527 write_access = 0;
2528 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 flush_icache_page(vma, page);
2530 set_pte_at(mm, address, page_table, pte);
2531 page_add_anon_rmap(page, vma, address);
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002532 /* It's better to call commit-charge after rmap is established */
2533 mem_cgroup_commit_charge_swapin(page, ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002535 swap_free(entry);
Nick Pigginb291f002008-10-18 20:26:44 -07002536 if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08002537 try_to_free_swap(page);
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002538 unlock_page(page);
2539
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 if (write_access) {
Hugh Dickins61469f12008-03-04 14:29:04 -08002541 ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
2542 if (ret & VM_FAULT_ERROR)
2543 ret &= VM_FAULT_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 goto out;
2545 }
2546
2547 /* No need to invalidate - it was non-present before */
2548 update_mmu_cache(vma, address, pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07002549unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002550 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551out:
2552 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002553out_nomap:
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002554 mem_cgroup_cancel_charge_swapin(ptr);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002555 pte_unmap_unlock(page_table, ptl);
Johannes Weinerbc43f752009-04-30 15:08:08 -07002556out_page:
Kirill Korotaevb8107482005-05-16 21:53:50 -07002557 unlock_page(page);
2558 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002559 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560}
2561
2562/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002563 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2564 * but allow concurrent faults), and pte mapped but not yet locked.
2565 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002567static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2568 unsigned long address, pte_t *page_table, pmd_t *pmd,
2569 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002571 struct page *page;
2572 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574
Nick Piggin557ed1f2007-10-16 01:24:40 -07002575 /* Allocate our own private page. */
2576 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577
Nick Piggin557ed1f2007-10-16 01:24:40 -07002578 if (unlikely(anon_vma_prepare(vma)))
2579 goto oom;
2580 page = alloc_zeroed_user_highpage_movable(vma, address);
2581 if (!page)
2582 goto oom;
Nick Piggin0ed361d2008-02-04 22:29:34 -08002583 __SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002585 if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002586 goto oom_free_page;
2587
Nick Piggin557ed1f2007-10-16 01:24:40 -07002588 entry = mk_pte(page, vma->vm_page_prot);
2589 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002590
Nick Piggin557ed1f2007-10-16 01:24:40 -07002591 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2592 if (!pte_none(*page_table))
2593 goto release;
2594 inc_mm_counter(mm, anon_rss);
Nick Piggin557ed1f2007-10-16 01:24:40 -07002595 page_add_new_anon_rmap(page, vma, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002596 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597
2598 /* No need to invalidate - it was non-present before */
Hugh Dickins65500d22005-10-29 18:15:59 -07002599 update_mmu_cache(vma, address, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002600unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002601 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002602 return 0;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002603release:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002604 mem_cgroup_uncharge_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002605 page_cache_release(page);
2606 goto unlock;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002607oom_free_page:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002608 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002609oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610 return VM_FAULT_OOM;
2611}
2612
2613/*
Nick Piggin54cb8822007-07-19 01:46:59 -07002614 * __do_fault() tries to create a new page mapping. It aggressively
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 * tries to share with existing pages, but makes a separate copy if
Nick Piggin54cb8822007-07-19 01:46:59 -07002616 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2617 * the next page fault.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 *
2619 * As this is called only for pages that do not currently exist, we
2620 * do not need to flush old virtual caches or the TLB.
2621 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002622 * We enter with non-exclusive mmap_sem (to exclude vma changes,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002623 * but allow concurrent faults), and pte neither mapped nor locked.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002624 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 */
Nick Piggin54cb8822007-07-19 01:46:59 -07002626static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002627 unsigned long address, pmd_t *pmd,
Nick Piggin54cb8822007-07-19 01:46:59 -07002628 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629{
Hugh Dickins16abfa02007-10-04 16:56:06 +01002630 pte_t *page_table;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002631 spinlock_t *ptl;
Nick Piggind0217ac2007-07-19 01:47:03 -07002632 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 int anon = 0;
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002635 int charged = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002636 struct page *dirty_page = NULL;
Nick Piggind0217ac2007-07-19 01:47:03 -07002637 struct vm_fault vmf;
2638 int ret;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002639 int page_mkwrite = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07002640
Nick Piggind0217ac2007-07-19 01:47:03 -07002641 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2642 vmf.pgoff = pgoff;
2643 vmf.flags = flags;
2644 vmf.page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645
Nick Piggin3c18ddd2008-04-28 02:12:10 -07002646 ret = vma->vm_ops->fault(vma, &vmf);
2647 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2648 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649
Nick Piggind00806b2007-07-19 01:46:57 -07002650 /*
Nick Piggind0217ac2007-07-19 01:47:03 -07002651 * For consistency in subsequent calls, make the faulted page always
Nick Piggind00806b2007-07-19 01:46:57 -07002652 * locked.
2653 */
Nick Piggin83c54072007-07-19 01:47:05 -07002654 if (unlikely(!(ret & VM_FAULT_LOCKED)))
Nick Piggind0217ac2007-07-19 01:47:03 -07002655 lock_page(vmf.page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002656 else
Nick Piggind0217ac2007-07-19 01:47:03 -07002657 VM_BUG_ON(!PageLocked(vmf.page));
Nick Piggind00806b2007-07-19 01:46:57 -07002658
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659 /*
2660 * Should we do an early C-O-W break?
2661 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002662 page = vmf.page;
Nick Piggin54cb8822007-07-19 01:46:59 -07002663 if (flags & FAULT_FLAG_WRITE) {
David Howells9637a5e2006-06-23 02:03:43 -07002664 if (!(vma->vm_flags & VM_SHARED)) {
Nick Piggin54cb8822007-07-19 01:46:59 -07002665 anon = 1;
Nick Piggind00806b2007-07-19 01:46:57 -07002666 if (unlikely(anon_vma_prepare(vma))) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002667 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002668 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002669 }
Nick Piggin83c54072007-07-19 01:47:05 -07002670 page = alloc_page_vma(GFP_HIGHUSER_MOVABLE,
2671 vma, address);
Nick Piggind00806b2007-07-19 01:46:57 -07002672 if (!page) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002673 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002674 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002675 }
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002676 if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) {
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002677 ret = VM_FAULT_OOM;
2678 page_cache_release(page);
2679 goto out;
2680 }
2681 charged = 1;
Nick Pigginb291f002008-10-18 20:26:44 -07002682 /*
2683 * Don't let another task, with possibly unlocked vma,
2684 * keep the mlocked page.
2685 */
2686 if (vma->vm_flags & VM_LOCKED)
2687 clear_page_mlock(vmf.page);
Nick Piggind0217ac2007-07-19 01:47:03 -07002688 copy_user_highpage(page, vmf.page, address, vma);
Nick Piggin0ed361d2008-02-04 22:29:34 -08002689 __SetPageUptodate(page);
David Howells9637a5e2006-06-23 02:03:43 -07002690 } else {
Nick Piggin54cb8822007-07-19 01:46:59 -07002691 /*
2692 * If the page will be shareable, see if the backing
David Howells9637a5e2006-06-23 02:03:43 -07002693 * address space wants to know that the page is about
Nick Piggin54cb8822007-07-19 01:46:59 -07002694 * to become writable
2695 */
Mark Fasheh69676142007-07-19 01:47:00 -07002696 if (vma->vm_ops->page_mkwrite) {
Nick Pigginc2ec1752009-03-31 15:23:21 -07002697 int tmp;
2698
Mark Fasheh69676142007-07-19 01:47:00 -07002699 unlock_page(page);
Nick Pigginb827e492009-04-30 15:08:16 -07002700 vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
Nick Pigginc2ec1752009-03-31 15:23:21 -07002701 tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
2702 if (unlikely(tmp &
2703 (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
2704 ret = tmp;
Nick Pigginb827e492009-04-30 15:08:16 -07002705 goto unwritable_page;
Mark Fasheh69676142007-07-19 01:47:00 -07002706 }
Nick Pigginb827e492009-04-30 15:08:16 -07002707 if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
2708 lock_page(page);
2709 if (!page->mapping) {
2710 ret = 0; /* retry the fault */
2711 unlock_page(page);
2712 goto unwritable_page;
2713 }
2714 } else
2715 VM_BUG_ON(!PageLocked(page));
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002716 page_mkwrite = 1;
David Howells9637a5e2006-06-23 02:03:43 -07002717 }
2718 }
Nick Piggin54cb8822007-07-19 01:46:59 -07002719
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 }
2721
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002722 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723
2724 /*
2725 * This silly early PAGE_DIRTY setting removes a race
2726 * due to the bad i386 page protection. But it's valid
2727 * for other architectures too.
2728 *
2729 * Note that if write_access is true, we either now have
2730 * an exclusive copy of the page, or this is a shared mapping,
2731 * so we can make it writable and dirty to avoid having to
2732 * handle that later.
2733 */
2734 /* Only go through if we didn't race with anybody else... */
Nick Piggin54cb8822007-07-19 01:46:59 -07002735 if (likely(pte_same(*page_table, orig_pte))) {
Nick Piggind00806b2007-07-19 01:46:57 -07002736 flush_icache_page(vma, page);
2737 entry = mk_pte(page, vma->vm_page_prot);
Nick Piggin54cb8822007-07-19 01:46:59 -07002738 if (flags & FAULT_FLAG_WRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740 if (anon) {
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002741 inc_mm_counter(mm, anon_rss);
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002742 page_add_new_anon_rmap(page, vma, address);
Hugh Dickinsf57e88a2005-11-21 21:32:19 -08002743 } else {
Hugh Dickins42946212005-10-29 18:16:05 -07002744 inc_mm_counter(mm, file_rss);
Nick Piggind00806b2007-07-19 01:46:57 -07002745 page_add_file_rmap(page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002746 if (flags & FAULT_FLAG_WRITE) {
Nick Piggind00806b2007-07-19 01:46:57 -07002747 dirty_page = page;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002748 get_page(dirty_page);
2749 }
Hugh Dickins42946212005-10-29 18:16:05 -07002750 }
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002751 set_pte_at(mm, address, page_table, entry);
Nick Piggind00806b2007-07-19 01:46:57 -07002752
2753 /* no need to invalidate: a not-present page won't be cached */
2754 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755 } else {
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002756 if (charged)
2757 mem_cgroup_uncharge_page(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002758 if (anon)
2759 page_cache_release(page);
2760 else
Nick Piggin54cb8822007-07-19 01:46:59 -07002761 anon = 1; /* no anon but release faulted_page */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762 }
2763
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002764 pte_unmap_unlock(page_table, ptl);
Nick Piggind00806b2007-07-19 01:46:57 -07002765
2766out:
Nick Pigginb827e492009-04-30 15:08:16 -07002767 if (dirty_page) {
2768 struct address_space *mapping = page->mapping;
2769
2770 if (set_page_dirty(dirty_page))
2771 page_mkwrite = 1;
2772 unlock_page(dirty_page);
2773 put_page(dirty_page);
2774 if (page_mkwrite && mapping) {
2775 /*
2776 * Some device drivers do not set page.mapping but still
2777 * dirty their pages
2778 */
2779 balance_dirty_pages_ratelimited(mapping);
2780 }
2781
2782 /* file_update_time outside page_lock */
Anton Salikhmetov8f7b3d12008-01-23 02:21:18 +03002783 if (vma->vm_file)
2784 file_update_time(vma->vm_file);
Nick Pigginb827e492009-04-30 15:08:16 -07002785 } else {
2786 unlock_page(vmf.page);
2787 if (anon)
2788 page_cache_release(vmf.page);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002789 }
Nick Piggind00806b2007-07-19 01:46:57 -07002790
Nick Piggin83c54072007-07-19 01:47:05 -07002791 return ret;
Nick Pigginb827e492009-04-30 15:08:16 -07002792
2793unwritable_page:
2794 page_cache_release(page);
2795 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07002796}
Nick Piggind00806b2007-07-19 01:46:57 -07002797
Nick Piggin54cb8822007-07-19 01:46:59 -07002798static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2799 unsigned long address, pte_t *page_table, pmd_t *pmd,
2800 int write_access, pte_t orig_pte)
2801{
2802 pgoff_t pgoff = (((address & PAGE_MASK)
Dean Nelson0da7e012007-10-16 01:24:45 -07002803 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
Nick Piggin54cb8822007-07-19 01:46:59 -07002804 unsigned int flags = (write_access ? FAULT_FLAG_WRITE : 0);
2805
Hugh Dickins16abfa02007-10-04 16:56:06 +01002806 pte_unmap(page_table);
2807 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Nick Piggin54cb8822007-07-19 01:46:59 -07002808}
2809
Jes Sorensenf4b81802006-09-27 01:50:10 -07002810/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 * Fault of a previously existing named mapping. Repopulate the pte
2812 * from the encoded file_pte if possible. This enables swappable
2813 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002814 *
2815 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2816 * but allow concurrent faults), and pte mapped but not yet locked.
2817 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002819static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins65500d22005-10-29 18:15:59 -07002820 unsigned long address, pte_t *page_table, pmd_t *pmd,
2821 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822{
Nick Piggind0217ac2007-07-19 01:47:03 -07002823 unsigned int flags = FAULT_FLAG_NONLINEAR |
2824 (write_access ? FAULT_FLAG_WRITE : 0);
Hugh Dickins65500d22005-10-29 18:15:59 -07002825 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002827 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Nick Piggin83c54072007-07-19 01:47:05 -07002828 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829
Hugh Dickins2509ef22009-01-06 14:40:10 -08002830 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002831 /*
2832 * Page table corrupted: show pte and kill process.
2833 */
Hugh Dickins3dc14742009-01-06 14:40:08 -08002834 print_bad_pte(vma, address, orig_pte, NULL);
Hugh Dickins65500d22005-10-29 18:15:59 -07002835 return VM_FAULT_OOM;
2836 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002837
2838 pgoff = pte_to_pgoff(orig_pte);
Hugh Dickins16abfa02007-10-04 16:56:06 +01002839 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840}
2841
2842/*
2843 * These routines also need to handle stuff like marking pages dirty
2844 * and/or accessed for architectures that don't do it in hardware (most
2845 * RISC architectures). The early dirtying is also good on the i386.
2846 *
2847 * There is also a hook called "update_mmu_cache()" that architectures
2848 * with external mmu caches can use to update those (ie the Sparc or
2849 * PowerPC hashed page tables that act as extended TLBs).
2850 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07002851 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2852 * but allow concurrent faults), and pte mapped but not yet locked.
2853 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854 */
2855static inline int handle_pte_fault(struct mm_struct *mm,
Hugh Dickins65500d22005-10-29 18:15:59 -07002856 struct vm_area_struct *vma, unsigned long address,
2857 pte_t *pte, pmd_t *pmd, int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858{
2859 pte_t entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002860 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002862 entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002864 if (pte_none(entry)) {
Jes Sorensenf4b81802006-09-27 01:50:10 -07002865 if (vma->vm_ops) {
Nick Piggin3c18ddd2008-04-28 02:12:10 -07002866 if (likely(vma->vm_ops->fault))
Nick Piggin54cb8822007-07-19 01:46:59 -07002867 return do_linear_fault(mm, vma, address,
2868 pte, pmd, write_access, entry);
Jes Sorensenf4b81802006-09-27 01:50:10 -07002869 }
2870 return do_anonymous_page(mm, vma, address,
2871 pte, pmd, write_access);
Hugh Dickins65500d22005-10-29 18:15:59 -07002872 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873 if (pte_file(entry))
Nick Piggind0217ac2007-07-19 01:47:03 -07002874 return do_nonlinear_fault(mm, vma, address,
Hugh Dickins65500d22005-10-29 18:15:59 -07002875 pte, pmd, write_access, entry);
2876 return do_swap_page(mm, vma, address,
2877 pte, pmd, write_access, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 }
2879
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002880 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002881 spin_lock(ptl);
2882 if (unlikely(!pte_same(*pte, entry)))
2883 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884 if (write_access) {
2885 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002886 return do_wp_page(mm, vma, address,
2887 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888 entry = pte_mkdirty(entry);
2889 }
2890 entry = pte_mkyoung(entry);
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002891 if (ptep_set_access_flags(vma, address, pte, entry, write_access)) {
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002892 update_mmu_cache(vma, address, entry);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002893 } else {
2894 /*
2895 * This is needed only for protection faults but the arch code
2896 * is not yet telling us if this is a protection fault or not.
2897 * This still avoids useless tlb flushes for .text page faults
2898 * with threads.
2899 */
2900 if (write_access)
2901 flush_tlb_page(vma, address);
2902 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002903unlock:
2904 pte_unmap_unlock(pte, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002905 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906}
2907
2908/*
2909 * By the time we get here, we already hold the mm semaphore
2910 */
Nick Piggin83c54072007-07-19 01:47:05 -07002911int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912 unsigned long address, int write_access)
2913{
2914 pgd_t *pgd;
2915 pud_t *pud;
2916 pmd_t *pmd;
2917 pte_t *pte;
2918
2919 __set_current_state(TASK_RUNNING);
2920
Christoph Lameterf8891e52006-06-30 01:55:45 -07002921 count_vm_event(PGFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002922
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01002923 if (unlikely(is_vm_hugetlb_page(vma)))
2924 return hugetlb_fault(mm, vma, address, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927 pud = pud_alloc(mm, pgd, address);
2928 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002929 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002930 pmd = pmd_alloc(mm, pud, address);
2931 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002932 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933 pte = pte_alloc_map(mm, pmd, address);
2934 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002935 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002936
Hugh Dickinsc74df322005-10-29 18:16:23 -07002937 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938}
2939
2940#ifndef __PAGETABLE_PUD_FOLDED
2941/*
2942 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002943 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002944 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002945int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002947 pud_t *new = pud_alloc_one(mm, address);
2948 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002949 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002950
Nick Piggin362a61a2008-05-14 06:37:36 +02002951 smp_wmb(); /* See comment in __pte_alloc */
2952
Hugh Dickins872fec12005-10-29 18:16:21 -07002953 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002954 if (pgd_present(*pgd)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002955 pud_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002956 else
2957 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07002958 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002959 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960}
2961#endif /* __PAGETABLE_PUD_FOLDED */
2962
2963#ifndef __PAGETABLE_PMD_FOLDED
2964/*
2965 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002966 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002968int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002970 pmd_t *new = pmd_alloc_one(mm, address);
2971 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002972 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973
Nick Piggin362a61a2008-05-14 06:37:36 +02002974 smp_wmb(); /* See comment in __pte_alloc */
2975
Hugh Dickins872fec12005-10-29 18:16:21 -07002976 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07002978 if (pud_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002979 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002980 else
2981 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07002983 if (pgd_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002984 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002985 else
2986 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07002988 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002989 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990}
2991#endif /* __PAGETABLE_PMD_FOLDED */
2992
2993int make_pages_present(unsigned long addr, unsigned long end)
2994{
2995 int ret, len, write;
2996 struct vm_area_struct * vma;
2997
2998 vma = find_vma(current->mm, addr);
2999 if (!vma)
KOSAKI Motohiroa4770972008-08-04 13:41:14 -07003000 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001 write = (vma->vm_flags & VM_WRITE) != 0;
Eric Sesterhenn5bcb28b2006-03-26 18:30:52 +02003002 BUG_ON(addr >= end);
3003 BUG_ON(end > vma->vm_end);
Rolf Eike Beer68e116a2007-07-15 23:38:03 -07003004 len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005 ret = get_user_pages(current, current->mm, addr,
3006 len, write, 0, NULL, NULL);
Lee Schermerhornc11d69d2008-10-18 20:26:56 -07003007 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008 return ret;
Lee Schermerhorn9978ad52008-10-18 20:26:56 -07003009 return ret == len ? 0 : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010}
3011
Linus Torvalds1da177e2005-04-16 15:20:36 -07003012#if !defined(__HAVE_ARCH_GATE_AREA)
3013
3014#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07003015static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003016
3017static int __init gate_vma_init(void)
3018{
3019 gate_vma.vm_mm = NULL;
3020 gate_vma.vm_start = FIXADDR_USER_START;
3021 gate_vma.vm_end = FIXADDR_USER_END;
Roland McGrathb6558c42007-01-26 00:56:47 -08003022 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
3023 gate_vma.vm_page_prot = __P101;
Roland McGrathf47aef52007-01-26 00:56:49 -08003024 /*
3025 * Make sure the vDSO gets into every core dump.
3026 * Dumping its contents makes post-mortem fully interpretable later
3027 * without matching up the same kernel and hardware config to see
3028 * what PC values meant.
3029 */
3030 gate_vma.vm_flags |= VM_ALWAYSDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003031 return 0;
3032}
3033__initcall(gate_vma_init);
3034#endif
3035
3036struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
3037{
3038#ifdef AT_SYSINFO_EHDR
3039 return &gate_vma;
3040#else
3041 return NULL;
3042#endif
3043}
3044
3045int in_gate_area_no_task(unsigned long addr)
3046{
3047#ifdef AT_SYSINFO_EHDR
3048 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
3049 return 1;
3050#endif
3051 return 0;
3052}
3053
3054#endif /* __HAVE_ARCH_GATE_AREA */
David Howells0ec76a12006-09-27 01:50:15 -07003055
Rik van Riel28b2ee22008-07-23 21:27:05 -07003056#ifdef CONFIG_HAVE_IOREMAP_PROT
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003057int follow_phys(struct vm_area_struct *vma,
3058 unsigned long address, unsigned int flags,
3059 unsigned long *prot, resource_size_t *phys)
Rik van Riel28b2ee22008-07-23 21:27:05 -07003060{
3061 pgd_t *pgd;
3062 pud_t *pud;
3063 pmd_t *pmd;
3064 pte_t *ptep, pte;
3065 spinlock_t *ptl;
3066 resource_size_t phys_addr = 0;
3067 struct mm_struct *mm = vma->vm_mm;
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003068 int ret = -EINVAL;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003069
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003070 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3071 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003072
3073 pgd = pgd_offset(mm, address);
3074 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003075 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003076
3077 pud = pud_offset(pgd, address);
3078 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003079 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003080
3081 pmd = pmd_offset(pud, address);
3082 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003083 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003084
3085 /* We cannot handle huge page PFN maps. Luckily they don't exist. */
3086 if (pmd_huge(*pmd))
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003087 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003088
3089 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
3090 if (!ptep)
3091 goto out;
3092
3093 pte = *ptep;
3094 if (!pte_present(pte))
3095 goto unlock;
3096 if ((flags & FOLL_WRITE) && !pte_write(pte))
3097 goto unlock;
3098 phys_addr = pte_pfn(pte);
3099 phys_addr <<= PAGE_SHIFT; /* Shift here to avoid overflow on PAE */
3100
3101 *prot = pgprot_val(pte_pgprot(pte));
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003102 *phys = phys_addr;
3103 ret = 0;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003104
3105unlock:
3106 pte_unmap_unlock(ptep, ptl);
3107out:
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003108 return ret;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003109}
3110
3111int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
3112 void *buf, int len, int write)
3113{
3114 resource_size_t phys_addr;
3115 unsigned long prot = 0;
KOSAKI Motohiro2bc72732009-01-06 14:39:43 -08003116 void __iomem *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003117 int offset = addr & (PAGE_SIZE-1);
3118
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003119 if (follow_phys(vma, addr, write, &prot, &phys_addr))
Rik van Riel28b2ee22008-07-23 21:27:05 -07003120 return -EINVAL;
3121
3122 maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot);
3123 if (write)
3124 memcpy_toio(maddr + offset, buf, len);
3125 else
3126 memcpy_fromio(buf, maddr + offset, len);
3127 iounmap(maddr);
3128
3129 return len;
3130}
3131#endif
3132
David Howells0ec76a12006-09-27 01:50:15 -07003133/*
3134 * Access another process' address space.
3135 * Source/target buffer must be kernel space,
3136 * Do not walk the page table directly, use get_user_pages
3137 */
3138int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
3139{
3140 struct mm_struct *mm;
3141 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07003142 void *old_buf = buf;
3143
3144 mm = get_task_mm(tsk);
3145 if (!mm)
3146 return 0;
3147
3148 down_read(&mm->mmap_sem);
Simon Arlott183ff222007-10-20 01:27:18 +02003149 /* ignore errors, just check how much was successfully transferred */
David Howells0ec76a12006-09-27 01:50:15 -07003150 while (len) {
3151 int bytes, ret, offset;
3152 void *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003153 struct page *page = NULL;
David Howells0ec76a12006-09-27 01:50:15 -07003154
3155 ret = get_user_pages(tsk, mm, addr, 1,
3156 write, 1, &page, &vma);
Rik van Riel28b2ee22008-07-23 21:27:05 -07003157 if (ret <= 0) {
3158 /*
3159 * Check if this is a VM_IO | VM_PFNMAP VMA, which
3160 * we can access using slightly different code.
3161 */
3162#ifdef CONFIG_HAVE_IOREMAP_PROT
3163 vma = find_vma(mm, addr);
3164 if (!vma)
3165 break;
3166 if (vma->vm_ops && vma->vm_ops->access)
3167 ret = vma->vm_ops->access(vma, addr, buf,
3168 len, write);
3169 if (ret <= 0)
3170#endif
3171 break;
3172 bytes = ret;
David Howells0ec76a12006-09-27 01:50:15 -07003173 } else {
Rik van Riel28b2ee22008-07-23 21:27:05 -07003174 bytes = len;
3175 offset = addr & (PAGE_SIZE-1);
3176 if (bytes > PAGE_SIZE-offset)
3177 bytes = PAGE_SIZE-offset;
3178
3179 maddr = kmap(page);
3180 if (write) {
3181 copy_to_user_page(vma, page, addr,
3182 maddr + offset, buf, bytes);
3183 set_page_dirty_lock(page);
3184 } else {
3185 copy_from_user_page(vma, page, addr,
3186 buf, maddr + offset, bytes);
3187 }
3188 kunmap(page);
3189 page_cache_release(page);
David Howells0ec76a12006-09-27 01:50:15 -07003190 }
David Howells0ec76a12006-09-27 01:50:15 -07003191 len -= bytes;
3192 buf += bytes;
3193 addr += bytes;
3194 }
3195 up_read(&mm->mmap_sem);
3196 mmput(mm);
3197
3198 return buf - old_buf;
3199}
Andi Kleen03252912008-01-30 13:33:18 +01003200
3201/*
3202 * Print the name of a VMA.
3203 */
3204void print_vma_addr(char *prefix, unsigned long ip)
3205{
3206 struct mm_struct *mm = current->mm;
3207 struct vm_area_struct *vma;
3208
Ingo Molnare8bff742008-02-13 20:21:06 +01003209 /*
3210 * Do not print if we are in atomic
3211 * contexts (in exception stacks, etc.):
3212 */
3213 if (preempt_count())
3214 return;
3215
Andi Kleen03252912008-01-30 13:33:18 +01003216 down_read(&mm->mmap_sem);
3217 vma = find_vma(mm, ip);
3218 if (vma && vma->vm_file) {
3219 struct file *f = vma->vm_file;
3220 char *buf = (char *)__get_free_page(GFP_KERNEL);
3221 if (buf) {
3222 char *p, *s;
3223
Jan Blunckcf28b482008-02-14 19:38:44 -08003224 p = d_path(&f->f_path, buf, PAGE_SIZE);
Andi Kleen03252912008-01-30 13:33:18 +01003225 if (IS_ERR(p))
3226 p = "?";
3227 s = strrchr(p, '/');
3228 if (s)
3229 p = s+1;
3230 printk("%s%s[%lx+%lx]", prefix, p,
3231 vma->vm_start,
3232 vma->vm_end - vma->vm_start);
3233 free_page((unsigned long)buf);
3234 }
3235 }
3236 up_read(&current->mm->mmap_sem);
3237}
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003238
3239#ifdef CONFIG_PROVE_LOCKING
3240void might_fault(void)
3241{
Peter Zijlstra95156f02009-01-12 13:02:11 +01003242 /*
3243 * Some code (nfs/sunrpc) uses socket ops on kernel memory while
3244 * holding the mmap_sem, this is safe because kernel memory doesn't
3245 * get paged out, therefore we'll never actually fault, and the
3246 * below annotations will generate false positives.
3247 */
3248 if (segment_eq(get_fs(), KERNEL_DS))
3249 return;
3250
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003251 might_sleep();
3252 /*
3253 * it would be nicer only to annotate paths which are not under
3254 * pagefault_disable, however that requires a larger audit and
3255 * providing helpers like get_user_atomic.
3256 */
3257 if (!in_atomic() && current->mm)
3258 might_lock_read(&current->mm->mmap_sem);
3259}
3260EXPORT_SYMBOL(might_fault);
3261#endif