blob: baa999e87cd21d2f9e67c8e3b1093bb6831513fb [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);
1154 mark_page_accessed(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001156unlock:
1157 pte_unmap_unlock(ptep, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158out:
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001159 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001161bad_page:
1162 pte_unmap_unlock(ptep, ptl);
1163 return ERR_PTR(-EFAULT);
1164
1165no_page:
1166 pte_unmap_unlock(ptep, ptl);
1167 if (!pte_none(pte))
1168 return page;
1169 /* Fall through to ZERO_PAGE handling */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001170no_page_table:
1171 /*
1172 * When core dumping an enormous anonymous area that nobody
1173 * has touched so far, we don't want to allocate page tables.
1174 */
1175 if (flags & FOLL_ANON) {
Nick Piggin557ed1f2007-10-16 01:24:40 -07001176 page = ZERO_PAGE(0);
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001177 if (flags & FOLL_GET)
1178 get_page(page);
1179 BUG_ON(flags & FOLL_WRITE);
1180 }
1181 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182}
1183
Linus Torvalds672ca282008-06-23 11:21:37 -07001184/* Can we do the FOLL_ANON optimization? */
1185static inline int use_zero_page(struct vm_area_struct *vma)
1186{
1187 /*
1188 * We don't want to optimize FOLL_ANON for make_pages_present()
1189 * when it tries to page in a VM_LOCKED region. As to VM_SHARED,
1190 * we want to get the page from the page tables to make sure
1191 * that we serialize and update with any other user of that
1192 * mapping.
1193 */
1194 if (vma->vm_flags & (VM_LOCKED | VM_SHARED))
1195 return 0;
1196 /*
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001197 * And if we have a fault routine, it's not an anonymous region.
Linus Torvalds672ca282008-06-23 11:21:37 -07001198 */
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001199 return !vma->vm_ops || !vma->vm_ops->fault;
Linus Torvalds672ca282008-06-23 11:21:37 -07001200}
1201
Nick Pigginb291f002008-10-18 20:26:44 -07001202
1203
1204int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
1205 unsigned long start, int len, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 struct page **pages, struct vm_area_struct **vmas)
1207{
1208 int i;
Nick Pigginb291f002008-10-18 20:26:44 -07001209 unsigned int vm_flags = 0;
1210 int write = !!(flags & GUP_FLAGS_WRITE);
1211 int force = !!(flags & GUP_FLAGS_FORCE);
1212 int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS);
Ying Han47792802009-01-06 14:40:18 -08001213 int ignore_sigkill = !!(flags & GUP_FLAGS_IGNORE_SIGKILL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214
Jonathan Corbet900cf082008-02-11 16:17:33 -07001215 if (len <= 0)
1216 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 /*
1218 * Require read or write permissions.
1219 * If 'force' is set, we only require the "MAY" flags.
1220 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001221 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
1222 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 i = 0;
1224
1225 do {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001226 struct vm_area_struct *vma;
1227 unsigned int foll_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228
1229 vma = find_extend_vma(mm, start);
1230 if (!vma && in_gate_area(tsk, start)) {
1231 unsigned long pg = start & PAGE_MASK;
1232 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
1233 pgd_t *pgd;
1234 pud_t *pud;
1235 pmd_t *pmd;
1236 pte_t *pte;
Nick Pigginb291f002008-10-18 20:26:44 -07001237
1238 /* user gate pages are read-only */
1239 if (!ignore && write)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 return i ? : -EFAULT;
1241 if (pg > TASK_SIZE)
1242 pgd = pgd_offset_k(pg);
1243 else
1244 pgd = pgd_offset_gate(mm, pg);
1245 BUG_ON(pgd_none(*pgd));
1246 pud = pud_offset(pgd, pg);
1247 BUG_ON(pud_none(*pud));
1248 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001249 if (pmd_none(*pmd))
1250 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001252 if (pte_none(*pte)) {
1253 pte_unmap(pte);
1254 return i ? : -EFAULT;
1255 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 if (pages) {
Nick Pigginfa2a4552005-11-29 18:43:17 +11001257 struct page *page = vm_normal_page(gate_vma, start, *pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001258 pages[i] = page;
1259 if (page)
1260 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 }
1262 pte_unmap(pte);
1263 if (vmas)
1264 vmas[i] = gate_vma;
1265 i++;
1266 start += PAGE_SIZE;
1267 len--;
1268 continue;
1269 }
1270
Nick Pigginb291f002008-10-18 20:26:44 -07001271 if (!vma ||
1272 (vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
1273 (!ignore && !(vm_flags & vma->vm_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 return i ? : -EFAULT;
1275
1276 if (is_vm_hugetlb_page(vma)) {
1277 i = follow_hugetlb_page(mm, vma, pages, vmas,
Adam Litke5b23dbe2007-11-14 16:59:33 -08001278 &start, &len, i, write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 continue;
1280 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001281
1282 foll_flags = FOLL_TOUCH;
1283 if (pages)
1284 foll_flags |= FOLL_GET;
Linus Torvalds672ca282008-06-23 11:21:37 -07001285 if (!write && use_zero_page(vma))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001286 foll_flags |= FOLL_ANON;
1287
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 do {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001289 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290
Ethan Solomita462e00c2007-07-15 23:38:16 -07001291 /*
Ying Han47792802009-01-06 14:40:18 -08001292 * If we have a pending SIGKILL, don't keep faulting
1293 * pages and potentially allocating memory, unless
1294 * current is handling munlock--e.g., on exit. In
1295 * that case, we are not allocating memory. Rather,
1296 * we're only unlocking already resident/mapped pages.
Ethan Solomita462e00c2007-07-15 23:38:16 -07001297 */
Ying Han47792802009-01-06 14:40:18 -08001298 if (unlikely(!ignore_sigkill &&
1299 fatal_signal_pending(current)))
1300 return i ? i : -ERESTARTSYS;
Ethan Solomita462e00c2007-07-15 23:38:16 -07001301
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001302 if (write)
1303 foll_flags |= FOLL_WRITE;
1304
1305 cond_resched();
Linus Torvalds6aab3412005-11-28 14:34:23 -08001306 while (!(page = follow_page(vma, start, foll_flags))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001307 int ret;
Nick Piggin83c54072007-07-19 01:47:05 -07001308 ret = handle_mm_fault(mm, vma, start,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001309 foll_flags & FOLL_WRITE);
Nick Piggin83c54072007-07-19 01:47:05 -07001310 if (ret & VM_FAULT_ERROR) {
1311 if (ret & VM_FAULT_OOM)
1312 return i ? i : -ENOMEM;
1313 else if (ret & VM_FAULT_SIGBUS)
1314 return i ? i : -EFAULT;
1315 BUG();
1316 }
1317 if (ret & VM_FAULT_MAJOR)
1318 tsk->maj_flt++;
1319 else
1320 tsk->min_flt++;
1321
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001322 /*
Nick Piggin83c54072007-07-19 01:47:05 -07001323 * The VM_FAULT_WRITE bit tells us that
1324 * do_wp_page has broken COW when necessary,
1325 * even if maybe_mkwrite decided not to set
1326 * pte_write. We can thus safely do subsequent
Hugh Dickins878b63a2009-01-06 14:39:32 -08001327 * page lookups as if they were reads. But only
1328 * do so when looping for pte_write is futile:
1329 * in some cases userspace may also be wanting
1330 * to write to the gotten user page, which a
1331 * read fault here might prevent (a readonly
1332 * page might get reCOWed by userspace write).
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001333 */
Hugh Dickins878b63a2009-01-06 14:39:32 -08001334 if ((ret & VM_FAULT_WRITE) &&
1335 !(vma->vm_flags & VM_WRITE))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001336 foll_flags &= ~FOLL_WRITE;
Nick Piggin83c54072007-07-19 01:47:05 -07001337
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07001338 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 }
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001340 if (IS_ERR(page))
1341 return i ? i : PTR_ERR(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001343 pages[i] = page;
James Bottomley03beb072006-03-26 01:36:57 -08001344
Russell Kinga6f36be2006-12-30 22:24:19 +00001345 flush_anon_page(vma, page, start);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001346 flush_dcache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 }
1348 if (vmas)
1349 vmas[i] = vma;
1350 i++;
1351 start += PAGE_SIZE;
1352 len--;
Hugh Dickins08ef4722005-06-21 17:15:10 -07001353 } while (len && start < vma->vm_end);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001354 } while (len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 return i;
1356}
Nick Pigginb291f002008-10-18 20:26:44 -07001357
1358int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
1359 unsigned long start, int len, int write, int force,
1360 struct page **pages, struct vm_area_struct **vmas)
1361{
1362 int flags = 0;
1363
1364 if (write)
1365 flags |= GUP_FLAGS_WRITE;
1366 if (force)
1367 flags |= GUP_FLAGS_FORCE;
1368
1369 return __get_user_pages(tsk, mm,
1370 start, len, flags,
1371 pages, vmas);
1372}
1373
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374EXPORT_SYMBOL(get_user_pages);
1375
Harvey Harrison920c7a52008-02-04 22:29:26 -08001376pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr,
1377 spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001378{
1379 pgd_t * pgd = pgd_offset(mm, addr);
1380 pud_t * pud = pud_alloc(mm, pgd, addr);
1381 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001382 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001383 if (pmd)
1384 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1385 }
1386 return NULL;
1387}
1388
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001390 * This is the old fallback for page remapping.
1391 *
1392 * For historical reasons, it only allows reserved pages. Only
1393 * old drivers should use this, and they needed to mark their
1394 * pages reserved for the old functions anyway.
1395 */
Nick Piggin423bad62008-04-28 02:13:01 -07001396static int insert_page(struct vm_area_struct *vma, unsigned long addr,
1397 struct page *page, pgprot_t prot)
Linus Torvalds238f58d2005-11-29 13:01:56 -08001398{
Nick Piggin423bad62008-04-28 02:13:01 -07001399 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001400 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001401 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001402 spinlock_t *ptl;
1403
Linus Torvalds238f58d2005-11-29 13:01:56 -08001404 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001405 if (PageAnon(page))
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001406 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001407 retval = -ENOMEM;
1408 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001409 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001410 if (!pte)
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001411 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001412 retval = -EBUSY;
1413 if (!pte_none(*pte))
1414 goto out_unlock;
1415
1416 /* Ok, finally just insert the thing.. */
1417 get_page(page);
1418 inc_mm_counter(mm, file_rss);
1419 page_add_file_rmap(page);
1420 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1421
1422 retval = 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001423 pte_unmap_unlock(pte, ptl);
1424 return retval;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001425out_unlock:
1426 pte_unmap_unlock(pte, ptl);
1427out:
1428 return retval;
1429}
1430
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001431/**
1432 * vm_insert_page - insert single page into user vma
1433 * @vma: user vma to map to
1434 * @addr: target user address of this page
1435 * @page: source kernel page
1436 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001437 * This allows drivers to insert individual pages they've allocated
1438 * into a user vma.
1439 *
1440 * The page has to be a nice clean _individual_ kernel allocation.
1441 * If you allocate a compound page, you need to have marked it as
1442 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001443 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001444 *
1445 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1446 * took an arbitrary page protection parameter. This doesn't allow
1447 * that. Your vma protection will have to be set up correctly, which
1448 * means that if you want a shared writable mapping, you'd better
1449 * ask for a shared writable mapping!
1450 *
1451 * The page does not need to be reserved.
1452 */
Nick Piggin423bad62008-04-28 02:13:01 -07001453int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
1454 struct page *page)
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001455{
1456 if (addr < vma->vm_start || addr >= vma->vm_end)
1457 return -EFAULT;
1458 if (!page_count(page))
1459 return -EINVAL;
Linus Torvalds4d7672b2005-12-16 10:21:23 -08001460 vma->vm_flags |= VM_INSERTPAGE;
Nick Piggin423bad62008-04-28 02:13:01 -07001461 return insert_page(vma, addr, page, vma->vm_page_prot);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001462}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001463EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001464
Nick Piggin423bad62008-04-28 02:13:01 -07001465static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1466 unsigned long pfn, pgprot_t prot)
1467{
1468 struct mm_struct *mm = vma->vm_mm;
1469 int retval;
1470 pte_t *pte, entry;
1471 spinlock_t *ptl;
1472
1473 retval = -ENOMEM;
1474 pte = get_locked_pte(mm, addr, &ptl);
1475 if (!pte)
1476 goto out;
1477 retval = -EBUSY;
1478 if (!pte_none(*pte))
1479 goto out_unlock;
1480
1481 /* Ok, finally just insert the thing.. */
1482 entry = pte_mkspecial(pfn_pte(pfn, prot));
1483 set_pte_at(mm, addr, pte, entry);
1484 update_mmu_cache(vma, addr, entry); /* XXX: why not for insert_page? */
1485
1486 retval = 0;
1487out_unlock:
1488 pte_unmap_unlock(pte, ptl);
1489out:
1490 return retval;
1491}
1492
Nick Piggine0dc0d82007-02-12 00:51:36 -08001493/**
1494 * vm_insert_pfn - insert single pfn into user vma
1495 * @vma: user vma to map to
1496 * @addr: target user address of this page
1497 * @pfn: source kernel pfn
1498 *
1499 * Similar to vm_inert_page, this allows drivers to insert individual pages
1500 * they've allocated into a user vma. Same comments apply.
1501 *
1502 * This function should only be called from a vm_ops->fault handler, and
1503 * in that case the handler should return NULL.
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001504 *
1505 * vma cannot be a COW mapping.
1506 *
1507 * As this is called only for pages that do not currently exist, we
1508 * do not need to flush old virtual caches or the TLB.
Nick Piggine0dc0d82007-02-12 00:51:36 -08001509 */
1510int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
Nick Piggin423bad62008-04-28 02:13:01 -07001511 unsigned long pfn)
Nick Piggine0dc0d82007-02-12 00:51:36 -08001512{
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001513 int ret;
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001514 pgprot_t pgprot = vma->vm_page_prot;
Nick Piggin7e675132008-04-28 02:13:00 -07001515 /*
1516 * Technically, architectures with pte_special can avoid all these
1517 * restrictions (same for remap_pfn_range). However we would like
1518 * consistency in testing and feature parity among all, so we should
1519 * try to keep these invariants in place for everybody.
1520 */
Jared Hulbertb379d792008-04-28 02:12:58 -07001521 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
1522 BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
1523 (VM_PFNMAP|VM_MIXEDMAP));
1524 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1525 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
Nick Piggine0dc0d82007-02-12 00:51:36 -08001526
Nick Piggin423bad62008-04-28 02:13:01 -07001527 if (addr < vma->vm_start || addr >= vma->vm_end)
1528 return -EFAULT;
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001529 if (track_pfn_vma_new(vma, &pgprot, pfn, PAGE_SIZE))
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001530 return -EINVAL;
1531
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001532 ret = insert_pfn(vma, addr, pfn, pgprot);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001533
1534 if (ret)
1535 untrack_pfn_vma(vma, pfn, PAGE_SIZE);
1536
1537 return ret;
Nick Piggine0dc0d82007-02-12 00:51:36 -08001538}
1539EXPORT_SYMBOL(vm_insert_pfn);
1540
Nick Piggin423bad62008-04-28 02:13:01 -07001541int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
1542 unsigned long pfn)
1543{
1544 BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
1545
1546 if (addr < vma->vm_start || addr >= vma->vm_end)
1547 return -EFAULT;
1548
1549 /*
1550 * If we don't have pte special, then we have to use the pfn_valid()
1551 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1552 * refcount the page if pfn_valid is true (hence insert_page rather
1553 * than insert_pfn).
1554 */
1555 if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) {
1556 struct page *page;
1557
1558 page = pfn_to_page(pfn);
1559 return insert_page(vma, addr, page, vma->vm_page_prot);
1560 }
1561 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
1562}
1563EXPORT_SYMBOL(vm_insert_mixed);
1564
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001565/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 * maps a range of physical memory into the requested pages. the old
1567 * mappings are removed. any references to nonexistent pages results
1568 * in null mappings (currently treated as "copy-on-access")
1569 */
1570static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1571 unsigned long addr, unsigned long end,
1572 unsigned long pfn, pgprot_t prot)
1573{
1574 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001575 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576
Hugh Dickinsc74df322005-10-29 18:16:23 -07001577 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 if (!pte)
1579 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001580 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 do {
1582 BUG_ON(!pte_none(*pte));
Nick Piggin7e675132008-04-28 02:13:00 -07001583 set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 pfn++;
1585 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001586 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001587 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 return 0;
1589}
1590
1591static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1592 unsigned long addr, unsigned long end,
1593 unsigned long pfn, pgprot_t prot)
1594{
1595 pmd_t *pmd;
1596 unsigned long next;
1597
1598 pfn -= addr >> PAGE_SHIFT;
1599 pmd = pmd_alloc(mm, pud, addr);
1600 if (!pmd)
1601 return -ENOMEM;
1602 do {
1603 next = pmd_addr_end(addr, end);
1604 if (remap_pte_range(mm, pmd, addr, next,
1605 pfn + (addr >> PAGE_SHIFT), prot))
1606 return -ENOMEM;
1607 } while (pmd++, addr = next, addr != end);
1608 return 0;
1609}
1610
1611static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1612 unsigned long addr, unsigned long end,
1613 unsigned long pfn, pgprot_t prot)
1614{
1615 pud_t *pud;
1616 unsigned long next;
1617
1618 pfn -= addr >> PAGE_SHIFT;
1619 pud = pud_alloc(mm, pgd, addr);
1620 if (!pud)
1621 return -ENOMEM;
1622 do {
1623 next = pud_addr_end(addr, end);
1624 if (remap_pmd_range(mm, pud, addr, next,
1625 pfn + (addr >> PAGE_SHIFT), prot))
1626 return -ENOMEM;
1627 } while (pud++, addr = next, addr != end);
1628 return 0;
1629}
1630
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001631/**
1632 * remap_pfn_range - remap kernel memory to userspace
1633 * @vma: user vma to map to
1634 * @addr: target user address to start at
1635 * @pfn: physical address of kernel memory
1636 * @size: size of map area
1637 * @prot: page protection flags for this mapping
1638 *
1639 * Note: this is only safe if the mm semaphore is held when called.
1640 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1642 unsigned long pfn, unsigned long size, pgprot_t prot)
1643{
1644 pgd_t *pgd;
1645 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001646 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 struct mm_struct *mm = vma->vm_mm;
1648 int err;
1649
1650 /*
1651 * Physically remapped pages are special. Tell the
1652 * rest of the world about it:
1653 * VM_IO tells people not to look at these pages
1654 * (accesses can have side effects).
Hugh Dickins0b14c172005-11-21 21:32:15 -08001655 * VM_RESERVED is specified all over the place, because
1656 * in 2.4 it kept swapout's vma scan off this vma; but
1657 * in 2.6 the LRU scan won't even find its pages, so this
1658 * flag means no more than count its pages in reserved_vm,
1659 * and omit it from core dump, even when VM_IO turned off.
Linus Torvalds6aab3412005-11-28 14:34:23 -08001660 * VM_PFNMAP tells the core MM that the base pages are just
1661 * raw PFN mappings, and do not have a "struct page" associated
1662 * with them.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001663 *
1664 * There's a horrible special case to handle copy-on-write
1665 * behaviour that some programs depend on. We mark the "original"
1666 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 */
venkatesh.pallipadi@intel.com3c8bb732008-12-18 11:41:27 -08001668 if (addr == vma->vm_start && end == vma->vm_end)
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001669 vma->vm_pgoff = pfn;
venkatesh.pallipadi@intel.com3c8bb732008-12-18 11:41:27 -08001670 else if (is_cow_mapping(vma->vm_flags))
1671 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001672
Linus Torvalds6aab3412005-11-28 14:34:23 -08001673 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -08001675 err = track_pfn_vma_new(vma, &prot, pfn, PAGE_ALIGN(size));
venkatesh.pallipadi@intel.coma3670612009-01-09 16:13:09 -08001676 if (err) {
1677 /*
1678 * To indicate that track_pfn related cleanup is not
1679 * needed from higher level routine calling unmap_vmas
1680 */
1681 vma->vm_flags &= ~(VM_IO | VM_RESERVED | VM_PFNMAP);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001682 return -EINVAL;
venkatesh.pallipadi@intel.coma3670612009-01-09 16:13:09 -08001683 }
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001684
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 BUG_ON(addr >= end);
1686 pfn -= addr >> PAGE_SHIFT;
1687 pgd = pgd_offset(mm, addr);
1688 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 do {
1690 next = pgd_addr_end(addr, end);
1691 err = remap_pud_range(mm, pgd, addr, next,
1692 pfn + (addr >> PAGE_SHIFT), prot);
1693 if (err)
1694 break;
1695 } while (pgd++, addr = next, addr != end);
venkatesh.pallipadi@intel.com2ab64032008-12-18 11:41:29 -08001696
1697 if (err)
1698 untrack_pfn_vma(vma, pfn, PAGE_ALIGN(size));
1699
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 return err;
1701}
1702EXPORT_SYMBOL(remap_pfn_range);
1703
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001704static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1705 unsigned long addr, unsigned long end,
1706 pte_fn_t fn, void *data)
1707{
1708 pte_t *pte;
1709 int err;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001710 pgtable_t token;
Borislav Petkov94909912007-05-06 14:49:17 -07001711 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001712
1713 pte = (mm == &init_mm) ?
1714 pte_alloc_kernel(pmd, addr) :
1715 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1716 if (!pte)
1717 return -ENOMEM;
1718
1719 BUG_ON(pmd_huge(*pmd));
1720
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08001721 arch_enter_lazy_mmu_mode();
1722
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001723 token = pmd_pgtable(*pmd);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001724
1725 do {
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001726 err = fn(pte, token, addr, data);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001727 if (err)
1728 break;
1729 } while (pte++, addr += PAGE_SIZE, addr != end);
1730
Jeremy Fitzhardinge38e0edb2009-01-06 14:39:21 -08001731 arch_leave_lazy_mmu_mode();
1732
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001733 if (mm != &init_mm)
1734 pte_unmap_unlock(pte-1, ptl);
1735 return err;
1736}
1737
1738static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1739 unsigned long addr, unsigned long end,
1740 pte_fn_t fn, void *data)
1741{
1742 pmd_t *pmd;
1743 unsigned long next;
1744 int err;
1745
Andi Kleenceb86872008-07-23 21:27:50 -07001746 BUG_ON(pud_huge(*pud));
1747
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001748 pmd = pmd_alloc(mm, pud, addr);
1749 if (!pmd)
1750 return -ENOMEM;
1751 do {
1752 next = pmd_addr_end(addr, end);
1753 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1754 if (err)
1755 break;
1756 } while (pmd++, addr = next, addr != end);
1757 return err;
1758}
1759
1760static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1761 unsigned long addr, unsigned long end,
1762 pte_fn_t fn, void *data)
1763{
1764 pud_t *pud;
1765 unsigned long next;
1766 int err;
1767
1768 pud = pud_alloc(mm, pgd, addr);
1769 if (!pud)
1770 return -ENOMEM;
1771 do {
1772 next = pud_addr_end(addr, end);
1773 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1774 if (err)
1775 break;
1776 } while (pud++, addr = next, addr != end);
1777 return err;
1778}
1779
1780/*
1781 * Scan a region of virtual memory, filling in page tables as necessary
1782 * and calling a provided function on each leaf page table.
1783 */
1784int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1785 unsigned long size, pte_fn_t fn, void *data)
1786{
1787 pgd_t *pgd;
1788 unsigned long next;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001789 unsigned long start = addr, end = addr + size;
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001790 int err;
1791
1792 BUG_ON(addr >= end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001793 mmu_notifier_invalidate_range_start(mm, start, end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001794 pgd = pgd_offset(mm, addr);
1795 do {
1796 next = pgd_addr_end(addr, end);
1797 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1798 if (err)
1799 break;
1800 } while (pgd++, addr = next, addr != end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001801 mmu_notifier_invalidate_range_end(mm, start, end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001802 return err;
1803}
1804EXPORT_SYMBOL_GPL(apply_to_page_range);
1805
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001807 * handle_pte_fault chooses page fault handler according to an entry
1808 * which was read non-atomically. Before making any commitment, on
1809 * those architectures or configurations (e.g. i386 with PAE) which
1810 * might give a mix of unmatched parts, do_swap_page and do_file_page
1811 * must check under lock before unmapping the pte and proceeding
1812 * (but do_wp_page is only called after already making such a check;
1813 * and do_anonymous_page and do_no_page can safely check later on).
1814 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001815static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001816 pte_t *page_table, pte_t orig_pte)
1817{
1818 int same = 1;
1819#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1820 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001821 spinlock_t *ptl = pte_lockptr(mm, pmd);
1822 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001823 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001824 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001825 }
1826#endif
1827 pte_unmap(page_table);
1828 return same;
1829}
1830
1831/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1833 * servicing faults for write access. In the normal case, do always want
1834 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1835 * that do not have writing enabled, when used by access_process_vm.
1836 */
1837static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1838{
1839 if (likely(vma->vm_flags & VM_WRITE))
1840 pte = pte_mkwrite(pte);
1841 return pte;
1842}
1843
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001844static 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 -08001845{
1846 /*
1847 * If the source page was a PFN mapping, we don't have
1848 * a "struct page" for it. We do a best-effort copy by
1849 * just copying from the original user address. If that
1850 * fails, we just zero-fill it. Live with it.
1851 */
1852 if (unlikely(!src)) {
1853 void *kaddr = kmap_atomic(dst, KM_USER0);
Linus Torvalds5d2a2db2005-11-29 14:07:55 -08001854 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1855
1856 /*
1857 * This really shouldn't fail, because the page is there
1858 * in the page tables. But it might just be unreadable,
1859 * in which case we just give up and fill the result with
1860 * zeroes.
1861 */
1862 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Linus Torvalds6aab3412005-11-28 14:34:23 -08001863 memset(kaddr, 0, PAGE_SIZE);
1864 kunmap_atomic(kaddr, KM_USER0);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07001865 flush_dcache_page(dst);
Nick Piggin0ed361d2008-02-04 22:29:34 -08001866 } else
1867 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001868}
1869
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 * This routine handles present pages, when users try to write
1872 * to a shared page. It is done by copying the page to a new address
1873 * and decrementing the shared-page counter for the old page.
1874 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 * Note that this routine assumes that the protection checks have been
1876 * done by the caller (the low-level page fault routine in most cases).
1877 * Thus we can safely just mark it writable once we've done any necessary
1878 * COW.
1879 *
1880 * We also mark the page dirty at this point even though the page will
1881 * change only once the write actually happens. This avoids a few races,
1882 * and potentially makes it more efficient.
1883 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001884 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1885 * but allow concurrent faults), with pte both mapped and locked.
1886 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001888static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1889 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001890 spinlock_t *ptl, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891{
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001892 struct page *old_page, *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 pte_t entry;
Nick Piggin83c54072007-07-19 01:47:05 -07001894 int reuse = 0, ret = 0;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001895 int page_mkwrite = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001896 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897
Linus Torvalds6aab3412005-11-28 14:34:23 -08001898 old_page = vm_normal_page(vma, address, orig_pte);
Peter Zijlstra251b97f2008-07-04 09:59:24 -07001899 if (!old_page) {
1900 /*
1901 * VM_MIXEDMAP !pfn_valid() case
1902 *
1903 * We should not cow pages in a shared writeable mapping.
1904 * Just mark the pages writable as we can't do any dirty
1905 * accounting on raw pfn maps.
1906 */
1907 if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
1908 (VM_WRITE|VM_SHARED))
1909 goto reuse;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001910 goto gotten;
Peter Zijlstra251b97f2008-07-04 09:59:24 -07001911 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001913 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001914 * Take out anonymous pages first, anonymous shared vmas are
1915 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001916 */
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001917 if (PageAnon(old_page)) {
Hugh Dickinsab967d82009-01-06 14:39:33 -08001918 if (!trylock_page(old_page)) {
1919 page_cache_get(old_page);
1920 pte_unmap_unlock(page_table, ptl);
1921 lock_page(old_page);
1922 page_table = pte_offset_map_lock(mm, pmd, address,
1923 &ptl);
1924 if (!pte_same(*page_table, orig_pte)) {
1925 unlock_page(old_page);
1926 page_cache_release(old_page);
1927 goto unlock;
1928 }
1929 page_cache_release(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001930 }
Hugh Dickins7b1fe592009-01-06 14:39:34 -08001931 reuse = reuse_swap_page(old_page);
Hugh Dickinsab967d82009-01-06 14:39:33 -08001932 unlock_page(old_page);
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001933 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001934 (VM_WRITE|VM_SHARED))) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001935 /*
1936 * Only catch write-faults on shared writable pages,
1937 * read-only shared pages can get COWed by
1938 * get_user_pages(.write=1, .force=1).
1939 */
David Howells9637a5e2006-06-23 02:03:43 -07001940 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
1941 /*
1942 * Notify the address space that the page is about to
1943 * become writable so that it can prohibit this or wait
1944 * for the page to get into an appropriate state.
1945 *
1946 * We do this without the lock held, so that it can
1947 * sleep if it needs to.
1948 */
1949 page_cache_get(old_page);
1950 pte_unmap_unlock(page_table, ptl);
1951
1952 if (vma->vm_ops->page_mkwrite(vma, old_page) < 0)
1953 goto unwritable_page;
1954
David Howells9637a5e2006-06-23 02:03:43 -07001955 /*
1956 * Since we dropped the lock we need to revalidate
1957 * the PTE as someone else may have changed it. If
1958 * they did, we just return, as we can count on the
1959 * MMU to tell us if they didn't also make it writable.
1960 */
1961 page_table = pte_offset_map_lock(mm, pmd, address,
1962 &ptl);
Hugh Dickinsc3704ce2007-02-10 01:43:00 -08001963 page_cache_release(old_page);
David Howells9637a5e2006-06-23 02:03:43 -07001964 if (!pte_same(*page_table, orig_pte))
1965 goto unlock;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001966
1967 page_mkwrite = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001969 dirty_page = old_page;
1970 get_page(dirty_page);
David Howells9637a5e2006-06-23 02:03:43 -07001971 reuse = 1;
David Howells9637a5e2006-06-23 02:03:43 -07001972 }
1973
1974 if (reuse) {
Peter Zijlstra251b97f2008-07-04 09:59:24 -07001975reuse:
David Howells9637a5e2006-06-23 02:03:43 -07001976 flush_cache_page(vma, address, pte_pfn(orig_pte));
1977 entry = pte_mkyoung(orig_pte);
1978 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
KAMEZAWA Hiroyuki954ffcb2007-10-16 01:25:44 -07001979 if (ptep_set_access_flags(vma, address, page_table, entry,1))
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07001980 update_mmu_cache(vma, address, entry);
David Howells9637a5e2006-06-23 02:03:43 -07001981 ret |= VM_FAULT_WRITE;
1982 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984
1985 /*
1986 * Ok, we need to copy. Oh, well..
1987 */
Nick Pigginb5810032005-10-29 18:16:12 -07001988 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08001989gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001990 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991
1992 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07001993 goto oom;
Nick Piggin557ed1f2007-10-16 01:24:40 -07001994 VM_BUG_ON(old_page == ZERO_PAGE(0));
1995 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
1996 if (!new_page)
1997 goto oom;
Nick Pigginb291f002008-10-18 20:26:44 -07001998 /*
1999 * Don't let another task, with possibly unlocked vma,
2000 * keep the mlocked page.
2001 */
Carsten Otteab926612009-02-04 15:12:16 -08002002 if ((vma->vm_flags & VM_LOCKED) && old_page) {
Nick Pigginb291f002008-10-18 20:26:44 -07002003 lock_page(old_page); /* for LRU manipulation */
2004 clear_page_mlock(old_page);
2005 unlock_page(old_page);
2006 }
Nick Piggin557ed1f2007-10-16 01:24:40 -07002007 cow_user_page(new_page, old_page, address, vma);
Nick Piggin0ed361d2008-02-04 22:29:34 -08002008 __SetPageUptodate(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002009
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002010 if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002011 goto oom_free_new;
2012
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 /*
2014 * Re-check the pte - we dropped the lock
2015 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002016 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07002017 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08002018 if (old_page) {
Hugh Dickins920fc352005-11-21 21:32:17 -08002019 if (!PageAnon(old_page)) {
2020 dec_mm_counter(mm, file_rss);
2021 inc_mm_counter(mm, anon_rss);
2022 }
2023 } else
Hugh Dickins42946212005-10-29 18:16:05 -07002024 inc_mm_counter(mm, anon_rss);
Ben Collinseca35132005-11-29 11:45:26 -08002025 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07002026 entry = mk_pte(new_page, vma->vm_page_prot);
2027 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Siddha, Suresh B4ce072f2006-09-29 01:58:42 -07002028 /*
2029 * Clear the pte entry and flush it first, before updating the
2030 * pte with the new entry. This will avoid a race condition
2031 * seen in the presence of one thread doing SMC and another
2032 * thread doing COW.
2033 */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07002034 ptep_clear_flush_notify(vma, address, page_table);
Nick Piggin9617d952006-01-06 00:11:12 -08002035 page_add_new_anon_rmap(new_page, vma, address);
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002036 set_pte_at(mm, address, page_table, entry);
2037 update_mmu_cache(vma, address, entry);
Nick Piggin945754a2008-06-23 14:30:30 +02002038 if (old_page) {
2039 /*
2040 * Only after switching the pte to the new page may
2041 * we remove the mapcount here. Otherwise another
2042 * process may come and find the rmap count decremented
2043 * before the pte is switched to the new page, and
2044 * "reuse" the old page writing into it while our pte
2045 * here still points into it and can be read by other
2046 * threads.
2047 *
2048 * The critical issue is to order this
2049 * page_remove_rmap with the ptp_clear_flush above.
2050 * Those stores are ordered by (if nothing else,)
2051 * the barrier present in the atomic_add_negative
2052 * in page_remove_rmap.
2053 *
2054 * Then the TLB flush in ptep_clear_flush ensures that
2055 * no process can access the old page before the
2056 * decremented mapcount is visible. And the old page
2057 * cannot be reused until after the decremented
2058 * mapcount is visible. So transitively, TLBs to
2059 * old page will be flushed before it can be reused.
2060 */
Hugh Dickinsedc315f2009-01-06 14:40:11 -08002061 page_remove_rmap(old_page);
Nick Piggin945754a2008-06-23 14:30:30 +02002062 }
2063
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 /* Free the old page.. */
2065 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10002066 ret |= VM_FAULT_WRITE;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002067 } else
2068 mem_cgroup_uncharge_page(new_page);
2069
Hugh Dickins920fc352005-11-21 21:32:17 -08002070 if (new_page)
2071 page_cache_release(new_page);
2072 if (old_page)
2073 page_cache_release(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002074unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002075 pte_unmap_unlock(page_table, ptl);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002076 if (dirty_page) {
Anton Salikhmetov8f7b3d12008-01-23 02:21:18 +03002077 if (vma->vm_file)
2078 file_update_time(vma->vm_file);
2079
Nick Piggin79352892007-07-19 01:47:22 -07002080 /*
2081 * Yes, Virginia, this is actually required to prevent a race
2082 * with clear_page_dirty_for_io() from clearing the page dirty
2083 * bit after it clear all dirty ptes, but before a racing
2084 * do_wp_page installs a dirty pte.
2085 *
2086 * do_no_page is protected similarly.
2087 */
2088 wait_on_page_locked(dirty_page);
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002089 set_page_dirty_balance(dirty_page, page_mkwrite);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002090 put_page(dirty_page);
2091 }
Nick Pigginf33ea7f2005-08-03 20:24:01 +10002092 return ret;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002093oom_free_new:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002094 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002095oom:
Hugh Dickins920fc352005-11-21 21:32:17 -08002096 if (old_page)
2097 page_cache_release(old_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 return VM_FAULT_OOM;
David Howells9637a5e2006-06-23 02:03:43 -07002099
2100unwritable_page:
2101 page_cache_release(old_page);
2102 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103}
2104
2105/*
2106 * Helper functions for unmap_mapping_range().
2107 *
2108 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
2109 *
2110 * We have to restart searching the prio_tree whenever we drop the lock,
2111 * since the iterator is only valid while the lock is held, and anyway
2112 * a later vma might be split and reinserted earlier while lock dropped.
2113 *
2114 * The list of nonlinear vmas could be handled more efficiently, using
2115 * a placeholder, but handle it in the same way until a need is shown.
2116 * It is important to search the prio_tree before nonlinear list: a vma
2117 * may become nonlinear and be shifted from prio_tree to nonlinear list
2118 * while the lock is dropped; but never shifted from list to prio_tree.
2119 *
2120 * In order to make forward progress despite restarting the search,
2121 * vm_truncate_count is used to mark a vma as now dealt with, so we can
2122 * quickly skip it next time around. Since the prio_tree search only
2123 * shows us those vmas affected by unmapping the range in question, we
2124 * can't efficiently keep all vmas in step with mapping->truncate_count:
2125 * so instead reset them all whenever it wraps back to 0 (then go to 1).
2126 * mapping->truncate_count and vma->vm_truncate_count are protected by
2127 * i_mmap_lock.
2128 *
2129 * In order to make forward progress despite repeatedly restarting some
Hugh Dickinsee39b372005-04-19 13:29:15 -07002130 * large vma, note the restart_addr from unmap_vmas when it breaks out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 * and restart from that address when we reach that vma again. It might
2132 * have been split or merged, shrunk or extended, but never shifted: so
2133 * restart_addr remains valid so long as it remains in the vma's range.
2134 * unmap_mapping_range forces truncate_count to leap over page-aligned
2135 * values so we can save vma's restart_addr in its truncate_count field.
2136 */
2137#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
2138
2139static void reset_vma_truncate_counts(struct address_space *mapping)
2140{
2141 struct vm_area_struct *vma;
2142 struct prio_tree_iter iter;
2143
2144 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
2145 vma->vm_truncate_count = 0;
2146 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
2147 vma->vm_truncate_count = 0;
2148}
2149
2150static int unmap_mapping_range_vma(struct vm_area_struct *vma,
2151 unsigned long start_addr, unsigned long end_addr,
2152 struct zap_details *details)
2153{
2154 unsigned long restart_addr;
2155 int need_break;
2156
Nick Piggind00806b2007-07-19 01:46:57 -07002157 /*
2158 * files that support invalidating or truncating portions of the
Nick Piggind0217ac2007-07-19 01:47:03 -07002159 * file from under mmaped areas must have their ->fault function
Nick Piggin83c54072007-07-19 01:47:05 -07002160 * return a locked page (and set VM_FAULT_LOCKED in the return).
2161 * This provides synchronisation against concurrent unmapping here.
Nick Piggind00806b2007-07-19 01:46:57 -07002162 */
Nick Piggind00806b2007-07-19 01:46:57 -07002163
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164again:
2165 restart_addr = vma->vm_truncate_count;
2166 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
2167 start_addr = restart_addr;
2168 if (start_addr >= end_addr) {
2169 /* Top of vma has been split off since last time */
2170 vma->vm_truncate_count = details->truncate_count;
2171 return 0;
2172 }
2173 }
2174
Hugh Dickinsee39b372005-04-19 13:29:15 -07002175 restart_addr = zap_page_range(vma, start_addr,
2176 end_addr - start_addr, details);
Nick Piggin95c354f2008-01-30 13:31:20 +01002177 need_break = need_resched() || spin_needbreak(details->i_mmap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178
Hugh Dickinsee39b372005-04-19 13:29:15 -07002179 if (restart_addr >= end_addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 /* We have now completed this vma: mark it so */
2181 vma->vm_truncate_count = details->truncate_count;
2182 if (!need_break)
2183 return 0;
2184 } else {
2185 /* Note restart_addr in vma's truncate_count field */
Hugh Dickinsee39b372005-04-19 13:29:15 -07002186 vma->vm_truncate_count = restart_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 if (!need_break)
2188 goto again;
2189 }
2190
2191 spin_unlock(details->i_mmap_lock);
2192 cond_resched();
2193 spin_lock(details->i_mmap_lock);
2194 return -EINTR;
2195}
2196
2197static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
2198 struct zap_details *details)
2199{
2200 struct vm_area_struct *vma;
2201 struct prio_tree_iter iter;
2202 pgoff_t vba, vea, zba, zea;
2203
2204restart:
2205 vma_prio_tree_foreach(vma, &iter, root,
2206 details->first_index, details->last_index) {
2207 /* Skip quickly over those we have already dealt with */
2208 if (vma->vm_truncate_count == details->truncate_count)
2209 continue;
2210
2211 vba = vma->vm_pgoff;
2212 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
2213 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
2214 zba = details->first_index;
2215 if (zba < vba)
2216 zba = vba;
2217 zea = details->last_index;
2218 if (zea > vea)
2219 zea = vea;
2220
2221 if (unmap_mapping_range_vma(vma,
2222 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
2223 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
2224 details) < 0)
2225 goto restart;
2226 }
2227}
2228
2229static inline void unmap_mapping_range_list(struct list_head *head,
2230 struct zap_details *details)
2231{
2232 struct vm_area_struct *vma;
2233
2234 /*
2235 * In nonlinear VMAs there is no correspondence between virtual address
2236 * offset and file offset. So we must perform an exhaustive search
2237 * across *all* the pages in each nonlinear VMA, not just the pages
2238 * whose virtual address lies outside the file truncation point.
2239 */
2240restart:
2241 list_for_each_entry(vma, head, shared.vm_set.list) {
2242 /* Skip quickly over those we have already dealt with */
2243 if (vma->vm_truncate_count == details->truncate_count)
2244 continue;
2245 details->nonlinear_vma = vma;
2246 if (unmap_mapping_range_vma(vma, vma->vm_start,
2247 vma->vm_end, details) < 0)
2248 goto restart;
2249 }
2250}
2251
2252/**
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002253 * 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 -07002254 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 * @holebegin: byte in first page to unmap, relative to the start of
2256 * the underlying file. This will be rounded down to a PAGE_SIZE
2257 * boundary. Note that this is different from vmtruncate(), which
2258 * must keep the partial page. In contrast, we must get rid of
2259 * partial pages.
2260 * @holelen: size of prospective hole in bytes. This will be rounded
2261 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2262 * end of the file.
2263 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2264 * but 0 when invalidating pagecache, don't throw away private data.
2265 */
2266void unmap_mapping_range(struct address_space *mapping,
2267 loff_t const holebegin, loff_t const holelen, int even_cows)
2268{
2269 struct zap_details details;
2270 pgoff_t hba = holebegin >> PAGE_SHIFT;
2271 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2272
2273 /* Check for overflow. */
2274 if (sizeof(holelen) > sizeof(hlen)) {
2275 long long holeend =
2276 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2277 if (holeend & ~(long long)ULONG_MAX)
2278 hlen = ULONG_MAX - hba + 1;
2279 }
2280
2281 details.check_mapping = even_cows? NULL: mapping;
2282 details.nonlinear_vma = NULL;
2283 details.first_index = hba;
2284 details.last_index = hba + hlen - 1;
2285 if (details.last_index < details.first_index)
2286 details.last_index = ULONG_MAX;
2287 details.i_mmap_lock = &mapping->i_mmap_lock;
2288
2289 spin_lock(&mapping->i_mmap_lock);
2290
Nick Piggind00806b2007-07-19 01:46:57 -07002291 /* Protect against endless unmapping loops */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 mapping->truncate_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 if (unlikely(is_restart_addr(mapping->truncate_count))) {
2294 if (mapping->truncate_count == 0)
2295 reset_vma_truncate_counts(mapping);
2296 mapping->truncate_count++;
2297 }
2298 details.truncate_count = mapping->truncate_count;
2299
2300 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
2301 unmap_mapping_range_tree(&mapping->i_mmap, &details);
2302 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
2303 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
2304 spin_unlock(&mapping->i_mmap_lock);
2305}
2306EXPORT_SYMBOL(unmap_mapping_range);
2307
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07002308/**
2309 * vmtruncate - unmap mappings "freed" by truncate() syscall
2310 * @inode: inode of the file used
2311 * @offset: file offset to start truncating
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 *
2313 * NOTE! We have to be ready to update the memory sharing
2314 * between the file and the memory map for a potential last
2315 * incomplete page. Ugly, but necessary.
2316 */
2317int vmtruncate(struct inode * inode, loff_t offset)
2318{
Christoph Hellwig61d50482008-02-04 22:28:56 -08002319 if (inode->i_size < offset) {
2320 unsigned long limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321
Christoph Hellwig61d50482008-02-04 22:28:56 -08002322 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
2323 if (limit != RLIM_INFINITY && offset > limit)
2324 goto out_sig;
2325 if (offset > inode->i_sb->s_maxbytes)
2326 goto out_big;
2327 i_size_write(inode, offset);
2328 } else {
2329 struct address_space *mapping = inode->i_mapping;
Nick Piggind00806b2007-07-19 01:46:57 -07002330
Christoph Hellwig61d50482008-02-04 22:28:56 -08002331 /*
2332 * truncation of in-use swapfiles is disallowed - it would
2333 * cause subsequent swapout to scribble on the now-freed
2334 * blocks.
2335 */
2336 if (IS_SWAPFILE(inode))
2337 return -ETXTBSY;
2338 i_size_write(inode, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339
Christoph Hellwig61d50482008-02-04 22:28:56 -08002340 /*
2341 * unmap_mapping_range is called twice, first simply for
2342 * efficiency so that truncate_inode_pages does fewer
2343 * single-page unmaps. However after this first call, and
2344 * before truncate_inode_pages finishes, it is possible for
2345 * private pages to be COWed, which remain after
2346 * truncate_inode_pages finishes, hence the second
2347 * unmap_mapping_range call must be made for correctness.
2348 */
2349 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2350 truncate_inode_pages(mapping, offset);
2351 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2352 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353
Al Viroacfa4382008-12-04 10:06:33 -05002354 if (inode->i_op->truncate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 inode->i_op->truncate(inode);
2356 return 0;
Christoph Hellwig61d50482008-02-04 22:28:56 -08002357
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358out_sig:
2359 send_sig(SIGXFSZ, current, 0);
2360out_big:
2361 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363EXPORT_SYMBOL(vmtruncate);
2364
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002365int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
2366{
2367 struct address_space *mapping = inode->i_mapping;
2368
2369 /*
2370 * If the underlying filesystem is not going to provide
2371 * a way to truncate a range of blocks (punch a hole) -
2372 * we should return failure right now.
2373 */
Al Viroacfa4382008-12-04 10:06:33 -05002374 if (!inode->i_op->truncate_range)
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002375 return -ENOSYS;
2376
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002377 mutex_lock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002378 down_write(&inode->i_alloc_sem);
2379 unmap_mapping_range(mapping, offset, (end - offset), 1);
2380 truncate_inode_pages_range(mapping, offset, end);
Nick Piggind00806b2007-07-19 01:46:57 -07002381 unmap_mapping_range(mapping, offset, (end - offset), 1);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002382 inode->i_op->truncate_range(inode, offset, end);
2383 up_write(&inode->i_alloc_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002384 mutex_unlock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002385
2386 return 0;
2387}
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002388
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002390 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2391 * but allow concurrent faults), and pte mapped but not yet locked.
2392 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002394static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2395 unsigned long address, pte_t *page_table, pmd_t *pmd,
2396 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002398 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 struct page *page;
Hugh Dickins65500d22005-10-29 18:15:59 -07002400 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 pte_t pte;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002402 struct mem_cgroup *ptr = NULL;
Nick Piggin83c54072007-07-19 01:47:05 -07002403 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002405 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002406 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002407
2408 entry = pte_to_swp_entry(orig_pte);
Christoph Lameter06972122006-06-23 02:03:35 -07002409 if (is_migration_entry(entry)) {
2410 migration_entry_wait(mm, pmd, address);
2411 goto out;
2412 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002413 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414 page = lookup_swap_cache(entry);
2415 if (!page) {
Ashwin Chaugule098fe652006-12-06 20:31:54 -08002416 grab_swap_token(); /* Contend for token _before_ read-in */
Hugh Dickins02098fe2008-02-04 22:28:42 -08002417 page = swapin_readahead(entry,
2418 GFP_HIGHUSER_MOVABLE, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419 if (!page) {
2420 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002421 * Back out if somebody else faulted in this pte
2422 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002424 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425 if (likely(pte_same(*page_table, orig_pte)))
2426 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002427 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002428 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 }
2430
2431 /* Had to read the page from swap area: Major fault */
2432 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002433 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434 }
2435
2436 mark_page_accessed(page);
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002437
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 lock_page(page);
Balbir Singh20a10222007-11-14 17:00:33 -08002439 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002441 if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) {
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002442 ret = VM_FAULT_OOM;
2443 unlock_page(page);
2444 goto out;
2445 }
2446
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002448 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002450 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002451 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002452 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002453
2454 if (unlikely(!PageUptodate(page))) {
2455 ret = VM_FAULT_SIGBUS;
2456 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 }
2458
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002459 /*
2460 * The page isn't present yet, go ahead with the fault.
2461 *
2462 * Be careful about the sequence of operations here.
2463 * To get its accounting right, reuse_swap_page() must be called
2464 * while the page is counted on swap but not yet in mapcount i.e.
2465 * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
2466 * must be called after the swap_free(), or it will never succeed.
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002467 * Because delete_from_swap_page() may be called by reuse_swap_page(),
2468 * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry
2469 * in page->private. In this case, a record in swap_cgroup is silently
2470 * discarded at swap_free().
KAMEZAWA Hiroyuki8c7c6e32009-01-07 18:08:00 -08002471 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472
Hugh Dickins42946212005-10-29 18:16:05 -07002473 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 pte = mk_pte(page, vma->vm_page_prot);
Hugh Dickins7b1fe592009-01-06 14:39:34 -08002475 if (write_access && reuse_swap_page(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2477 write_access = 0;
2478 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 flush_icache_page(vma, page);
2480 set_pte_at(mm, address, page_table, pte);
2481 page_add_anon_rmap(page, vma, address);
KAMEZAWA Hiroyuki03f3c432009-01-07 18:08:31 -08002482 /* It's better to call commit-charge after rmap is established */
2483 mem_cgroup_commit_charge_swapin(page, ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002485 swap_free(entry);
Nick Pigginb291f002008-10-18 20:26:44 -07002486 if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -08002487 try_to_free_swap(page);
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002488 unlock_page(page);
2489
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490 if (write_access) {
Hugh Dickins61469f12008-03-04 14:29:04 -08002491 ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
2492 if (ret & VM_FAULT_ERROR)
2493 ret &= VM_FAULT_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 goto out;
2495 }
2496
2497 /* No need to invalidate - it was non-present before */
2498 update_mmu_cache(vma, address, pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07002499unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002500 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501out:
2502 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002503out_nomap:
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002504 mem_cgroup_cancel_charge_swapin(ptr);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002505 pte_unmap_unlock(page_table, ptl);
Kirill Korotaevb8107482005-05-16 21:53:50 -07002506 unlock_page(page);
2507 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002508 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509}
2510
2511/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002512 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2513 * but allow concurrent faults), and pte mapped but not yet locked.
2514 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002516static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2517 unsigned long address, pte_t *page_table, pmd_t *pmd,
2518 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002520 struct page *page;
2521 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523
Nick Piggin557ed1f2007-10-16 01:24:40 -07002524 /* Allocate our own private page. */
2525 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526
Nick Piggin557ed1f2007-10-16 01:24:40 -07002527 if (unlikely(anon_vma_prepare(vma)))
2528 goto oom;
2529 page = alloc_zeroed_user_highpage_movable(vma, address);
2530 if (!page)
2531 goto oom;
Nick Piggin0ed361d2008-02-04 22:29:34 -08002532 __SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002534 if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002535 goto oom_free_page;
2536
Nick Piggin557ed1f2007-10-16 01:24:40 -07002537 entry = mk_pte(page, vma->vm_page_prot);
2538 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002539
Nick Piggin557ed1f2007-10-16 01:24:40 -07002540 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2541 if (!pte_none(*page_table))
2542 goto release;
2543 inc_mm_counter(mm, anon_rss);
Nick Piggin557ed1f2007-10-16 01:24:40 -07002544 page_add_new_anon_rmap(page, vma, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002545 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546
2547 /* No need to invalidate - it was non-present before */
Hugh Dickins65500d22005-10-29 18:15:59 -07002548 update_mmu_cache(vma, address, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002549unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002550 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002551 return 0;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002552release:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002553 mem_cgroup_uncharge_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002554 page_cache_release(page);
2555 goto unlock;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002556oom_free_page:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002557 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002558oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 return VM_FAULT_OOM;
2560}
2561
2562/*
Nick Piggin54cb8822007-07-19 01:46:59 -07002563 * __do_fault() tries to create a new page mapping. It aggressively
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 * tries to share with existing pages, but makes a separate copy if
Nick Piggin54cb8822007-07-19 01:46:59 -07002565 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2566 * the next page fault.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567 *
2568 * As this is called only for pages that do not currently exist, we
2569 * do not need to flush old virtual caches or the TLB.
2570 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002571 * We enter with non-exclusive mmap_sem (to exclude vma changes,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002572 * but allow concurrent faults), and pte neither mapped nor locked.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002573 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574 */
Nick Piggin54cb8822007-07-19 01:46:59 -07002575static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002576 unsigned long address, pmd_t *pmd,
Nick Piggin54cb8822007-07-19 01:46:59 -07002577 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578{
Hugh Dickins16abfa02007-10-04 16:56:06 +01002579 pte_t *page_table;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002580 spinlock_t *ptl;
Nick Piggind0217ac2007-07-19 01:47:03 -07002581 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 int anon = 0;
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002584 int charged = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002585 struct page *dirty_page = NULL;
Nick Piggind0217ac2007-07-19 01:47:03 -07002586 struct vm_fault vmf;
2587 int ret;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002588 int page_mkwrite = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07002589
Nick Piggind0217ac2007-07-19 01:47:03 -07002590 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2591 vmf.pgoff = pgoff;
2592 vmf.flags = flags;
2593 vmf.page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594
Nick Piggin3c18ddd2008-04-28 02:12:10 -07002595 ret = vma->vm_ops->fault(vma, &vmf);
2596 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2597 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598
Nick Piggind00806b2007-07-19 01:46:57 -07002599 /*
Nick Piggind0217ac2007-07-19 01:47:03 -07002600 * For consistency in subsequent calls, make the faulted page always
Nick Piggind00806b2007-07-19 01:46:57 -07002601 * locked.
2602 */
Nick Piggin83c54072007-07-19 01:47:05 -07002603 if (unlikely(!(ret & VM_FAULT_LOCKED)))
Nick Piggind0217ac2007-07-19 01:47:03 -07002604 lock_page(vmf.page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002605 else
Nick Piggind0217ac2007-07-19 01:47:03 -07002606 VM_BUG_ON(!PageLocked(vmf.page));
Nick Piggind00806b2007-07-19 01:46:57 -07002607
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 /*
2609 * Should we do an early C-O-W break?
2610 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002611 page = vmf.page;
Nick Piggin54cb8822007-07-19 01:46:59 -07002612 if (flags & FAULT_FLAG_WRITE) {
David Howells9637a5e2006-06-23 02:03:43 -07002613 if (!(vma->vm_flags & VM_SHARED)) {
Nick Piggin54cb8822007-07-19 01:46:59 -07002614 anon = 1;
Nick Piggind00806b2007-07-19 01:46:57 -07002615 if (unlikely(anon_vma_prepare(vma))) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002616 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002617 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002618 }
Nick Piggin83c54072007-07-19 01:47:05 -07002619 page = alloc_page_vma(GFP_HIGHUSER_MOVABLE,
2620 vma, address);
Nick Piggind00806b2007-07-19 01:46:57 -07002621 if (!page) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002622 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002623 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002624 }
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -08002625 if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) {
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002626 ret = VM_FAULT_OOM;
2627 page_cache_release(page);
2628 goto out;
2629 }
2630 charged = 1;
Nick Pigginb291f002008-10-18 20:26:44 -07002631 /*
2632 * Don't let another task, with possibly unlocked vma,
2633 * keep the mlocked page.
2634 */
2635 if (vma->vm_flags & VM_LOCKED)
2636 clear_page_mlock(vmf.page);
Nick Piggind0217ac2007-07-19 01:47:03 -07002637 copy_user_highpage(page, vmf.page, address, vma);
Nick Piggin0ed361d2008-02-04 22:29:34 -08002638 __SetPageUptodate(page);
David Howells9637a5e2006-06-23 02:03:43 -07002639 } else {
Nick Piggin54cb8822007-07-19 01:46:59 -07002640 /*
2641 * If the page will be shareable, see if the backing
David Howells9637a5e2006-06-23 02:03:43 -07002642 * address space wants to know that the page is about
Nick Piggin54cb8822007-07-19 01:46:59 -07002643 * to become writable
2644 */
Mark Fasheh69676142007-07-19 01:47:00 -07002645 if (vma->vm_ops->page_mkwrite) {
2646 unlock_page(page);
2647 if (vma->vm_ops->page_mkwrite(vma, page) < 0) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002648 ret = VM_FAULT_SIGBUS;
2649 anon = 1; /* no anon but release vmf.page */
Mark Fasheh69676142007-07-19 01:47:00 -07002650 goto out_unlocked;
2651 }
2652 lock_page(page);
Nick Piggind0217ac2007-07-19 01:47:03 -07002653 /*
2654 * XXX: this is not quite right (racy vs
2655 * invalidate) to unlock and relock the page
2656 * like this, however a better fix requires
2657 * reworking page_mkwrite locking API, which
2658 * is better done later.
2659 */
2660 if (!page->mapping) {
Nick Piggin83c54072007-07-19 01:47:05 -07002661 ret = 0;
Nick Piggind0217ac2007-07-19 01:47:03 -07002662 anon = 1; /* no anon but release vmf.page */
2663 goto out;
2664 }
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002665 page_mkwrite = 1;
David Howells9637a5e2006-06-23 02:03:43 -07002666 }
2667 }
Nick Piggin54cb8822007-07-19 01:46:59 -07002668
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 }
2670
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002671 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672
2673 /*
2674 * This silly early PAGE_DIRTY setting removes a race
2675 * due to the bad i386 page protection. But it's valid
2676 * for other architectures too.
2677 *
2678 * Note that if write_access is true, we either now have
2679 * an exclusive copy of the page, or this is a shared mapping,
2680 * so we can make it writable and dirty to avoid having to
2681 * handle that later.
2682 */
2683 /* Only go through if we didn't race with anybody else... */
Nick Piggin54cb8822007-07-19 01:46:59 -07002684 if (likely(pte_same(*page_table, orig_pte))) {
Nick Piggind00806b2007-07-19 01:46:57 -07002685 flush_icache_page(vma, page);
2686 entry = mk_pte(page, vma->vm_page_prot);
Nick Piggin54cb8822007-07-19 01:46:59 -07002687 if (flags & FAULT_FLAG_WRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689 if (anon) {
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002690 inc_mm_counter(mm, anon_rss);
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002691 page_add_new_anon_rmap(page, vma, address);
Hugh Dickinsf57e88a2005-11-21 21:32:19 -08002692 } else {
Hugh Dickins42946212005-10-29 18:16:05 -07002693 inc_mm_counter(mm, file_rss);
Nick Piggind00806b2007-07-19 01:46:57 -07002694 page_add_file_rmap(page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002695 if (flags & FAULT_FLAG_WRITE) {
Nick Piggind00806b2007-07-19 01:46:57 -07002696 dirty_page = page;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002697 get_page(dirty_page);
2698 }
Hugh Dickins42946212005-10-29 18:16:05 -07002699 }
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002700 set_pte_at(mm, address, page_table, entry);
Nick Piggind00806b2007-07-19 01:46:57 -07002701
2702 /* no need to invalidate: a not-present page won't be cached */
2703 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704 } else {
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002705 if (charged)
2706 mem_cgroup_uncharge_page(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002707 if (anon)
2708 page_cache_release(page);
2709 else
Nick Piggin54cb8822007-07-19 01:46:59 -07002710 anon = 1; /* no anon but release faulted_page */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711 }
2712
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002713 pte_unmap_unlock(page_table, ptl);
Nick Piggind00806b2007-07-19 01:46:57 -07002714
2715out:
Nick Piggind0217ac2007-07-19 01:47:03 -07002716 unlock_page(vmf.page);
Mark Fasheh69676142007-07-19 01:47:00 -07002717out_unlocked:
Nick Piggind00806b2007-07-19 01:46:57 -07002718 if (anon)
Nick Piggind0217ac2007-07-19 01:47:03 -07002719 page_cache_release(vmf.page);
Nick Piggind00806b2007-07-19 01:46:57 -07002720 else if (dirty_page) {
Anton Salikhmetov8f7b3d12008-01-23 02:21:18 +03002721 if (vma->vm_file)
2722 file_update_time(vma->vm_file);
2723
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002724 set_page_dirty_balance(dirty_page, page_mkwrite);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002725 put_page(dirty_page);
2726 }
Nick Piggind00806b2007-07-19 01:46:57 -07002727
Nick Piggin83c54072007-07-19 01:47:05 -07002728 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07002729}
Nick Piggind00806b2007-07-19 01:46:57 -07002730
Nick Piggin54cb8822007-07-19 01:46:59 -07002731static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2732 unsigned long address, pte_t *page_table, pmd_t *pmd,
2733 int write_access, pte_t orig_pte)
2734{
2735 pgoff_t pgoff = (((address & PAGE_MASK)
Dean Nelson0da7e012007-10-16 01:24:45 -07002736 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
Nick Piggin54cb8822007-07-19 01:46:59 -07002737 unsigned int flags = (write_access ? FAULT_FLAG_WRITE : 0);
2738
Hugh Dickins16abfa02007-10-04 16:56:06 +01002739 pte_unmap(page_table);
2740 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Nick Piggin54cb8822007-07-19 01:46:59 -07002741}
2742
Jes Sorensenf4b81802006-09-27 01:50:10 -07002743/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 * Fault of a previously existing named mapping. Repopulate the pte
2745 * from the encoded file_pte if possible. This enables swappable
2746 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002747 *
2748 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2749 * but allow concurrent faults), and pte mapped but not yet locked.
2750 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002752static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins65500d22005-10-29 18:15:59 -07002753 unsigned long address, pte_t *page_table, pmd_t *pmd,
2754 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755{
Nick Piggind0217ac2007-07-19 01:47:03 -07002756 unsigned int flags = FAULT_FLAG_NONLINEAR |
2757 (write_access ? FAULT_FLAG_WRITE : 0);
Hugh Dickins65500d22005-10-29 18:15:59 -07002758 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002760 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Nick Piggin83c54072007-07-19 01:47:05 -07002761 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762
Hugh Dickins2509ef22009-01-06 14:40:10 -08002763 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002764 /*
2765 * Page table corrupted: show pte and kill process.
2766 */
Hugh Dickins3dc14742009-01-06 14:40:08 -08002767 print_bad_pte(vma, address, orig_pte, NULL);
Hugh Dickins65500d22005-10-29 18:15:59 -07002768 return VM_FAULT_OOM;
2769 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002770
2771 pgoff = pte_to_pgoff(orig_pte);
Hugh Dickins16abfa02007-10-04 16:56:06 +01002772 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773}
2774
2775/*
2776 * These routines also need to handle stuff like marking pages dirty
2777 * and/or accessed for architectures that don't do it in hardware (most
2778 * RISC architectures). The early dirtying is also good on the i386.
2779 *
2780 * There is also a hook called "update_mmu_cache()" that architectures
2781 * with external mmu caches can use to update those (ie the Sparc or
2782 * PowerPC hashed page tables that act as extended TLBs).
2783 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07002784 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2785 * but allow concurrent faults), and pte mapped but not yet locked.
2786 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787 */
2788static inline int handle_pte_fault(struct mm_struct *mm,
Hugh Dickins65500d22005-10-29 18:15:59 -07002789 struct vm_area_struct *vma, unsigned long address,
2790 pte_t *pte, pmd_t *pmd, int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791{
2792 pte_t entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002793 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002795 entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002797 if (pte_none(entry)) {
Jes Sorensenf4b81802006-09-27 01:50:10 -07002798 if (vma->vm_ops) {
Nick Piggin3c18ddd2008-04-28 02:12:10 -07002799 if (likely(vma->vm_ops->fault))
Nick Piggin54cb8822007-07-19 01:46:59 -07002800 return do_linear_fault(mm, vma, address,
2801 pte, pmd, write_access, entry);
Jes Sorensenf4b81802006-09-27 01:50:10 -07002802 }
2803 return do_anonymous_page(mm, vma, address,
2804 pte, pmd, write_access);
Hugh Dickins65500d22005-10-29 18:15:59 -07002805 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806 if (pte_file(entry))
Nick Piggind0217ac2007-07-19 01:47:03 -07002807 return do_nonlinear_fault(mm, vma, address,
Hugh Dickins65500d22005-10-29 18:15:59 -07002808 pte, pmd, write_access, entry);
2809 return do_swap_page(mm, vma, address,
2810 pte, pmd, write_access, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 }
2812
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002813 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002814 spin_lock(ptl);
2815 if (unlikely(!pte_same(*pte, entry)))
2816 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817 if (write_access) {
2818 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002819 return do_wp_page(mm, vma, address,
2820 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 entry = pte_mkdirty(entry);
2822 }
2823 entry = pte_mkyoung(entry);
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002824 if (ptep_set_access_flags(vma, address, pte, entry, write_access)) {
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002825 update_mmu_cache(vma, address, entry);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002826 } else {
2827 /*
2828 * This is needed only for protection faults but the arch code
2829 * is not yet telling us if this is a protection fault or not.
2830 * This still avoids useless tlb flushes for .text page faults
2831 * with threads.
2832 */
2833 if (write_access)
2834 flush_tlb_page(vma, address);
2835 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002836unlock:
2837 pte_unmap_unlock(pte, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002838 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839}
2840
2841/*
2842 * By the time we get here, we already hold the mm semaphore
2843 */
Nick Piggin83c54072007-07-19 01:47:05 -07002844int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845 unsigned long address, int write_access)
2846{
2847 pgd_t *pgd;
2848 pud_t *pud;
2849 pmd_t *pmd;
2850 pte_t *pte;
2851
2852 __set_current_state(TASK_RUNNING);
2853
Christoph Lameterf8891e52006-06-30 01:55:45 -07002854 count_vm_event(PGFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01002856 if (unlikely(is_vm_hugetlb_page(vma)))
2857 return hugetlb_fault(mm, vma, address, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860 pud = pud_alloc(mm, pgd, address);
2861 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002862 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863 pmd = pmd_alloc(mm, pud, address);
2864 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002865 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866 pte = pte_alloc_map(mm, pmd, address);
2867 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002868 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002869
Hugh Dickinsc74df322005-10-29 18:16:23 -07002870 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871}
2872
2873#ifndef __PAGETABLE_PUD_FOLDED
2874/*
2875 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002876 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002878int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002880 pud_t *new = pud_alloc_one(mm, address);
2881 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002882 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883
Nick Piggin362a61a2008-05-14 06:37:36 +02002884 smp_wmb(); /* See comment in __pte_alloc */
2885
Hugh Dickins872fec12005-10-29 18:16:21 -07002886 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002887 if (pgd_present(*pgd)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002888 pud_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002889 else
2890 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07002891 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002892 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893}
2894#endif /* __PAGETABLE_PUD_FOLDED */
2895
2896#ifndef __PAGETABLE_PMD_FOLDED
2897/*
2898 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002899 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002900 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002901int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002903 pmd_t *new = pmd_alloc_one(mm, address);
2904 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002905 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906
Nick Piggin362a61a2008-05-14 06:37:36 +02002907 smp_wmb(); /* See comment in __pte_alloc */
2908
Hugh Dickins872fec12005-10-29 18:16:21 -07002909 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002910#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07002911 if (pud_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002912 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002913 else
2914 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07002916 if (pgd_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002917 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002918 else
2919 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002920#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07002921 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002922 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923}
2924#endif /* __PAGETABLE_PMD_FOLDED */
2925
2926int make_pages_present(unsigned long addr, unsigned long end)
2927{
2928 int ret, len, write;
2929 struct vm_area_struct * vma;
2930
2931 vma = find_vma(current->mm, addr);
2932 if (!vma)
KOSAKI Motohiroa4770972008-08-04 13:41:14 -07002933 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 write = (vma->vm_flags & VM_WRITE) != 0;
Eric Sesterhenn5bcb28b2006-03-26 18:30:52 +02002935 BUG_ON(addr >= end);
2936 BUG_ON(end > vma->vm_end);
Rolf Eike Beer68e116a2007-07-15 23:38:03 -07002937 len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938 ret = get_user_pages(current, current->mm, addr,
2939 len, write, 0, NULL, NULL);
Lee Schermerhornc11d69d2008-10-18 20:26:56 -07002940 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002941 return ret;
Lee Schermerhorn9978ad52008-10-18 20:26:56 -07002942 return ret == len ? 0 : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943}
2944
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945#if !defined(__HAVE_ARCH_GATE_AREA)
2946
2947#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07002948static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949
2950static int __init gate_vma_init(void)
2951{
2952 gate_vma.vm_mm = NULL;
2953 gate_vma.vm_start = FIXADDR_USER_START;
2954 gate_vma.vm_end = FIXADDR_USER_END;
Roland McGrathb6558c42007-01-26 00:56:47 -08002955 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
2956 gate_vma.vm_page_prot = __P101;
Roland McGrathf47aef52007-01-26 00:56:49 -08002957 /*
2958 * Make sure the vDSO gets into every core dump.
2959 * Dumping its contents makes post-mortem fully interpretable later
2960 * without matching up the same kernel and hardware config to see
2961 * what PC values meant.
2962 */
2963 gate_vma.vm_flags |= VM_ALWAYSDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002964 return 0;
2965}
2966__initcall(gate_vma_init);
2967#endif
2968
2969struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2970{
2971#ifdef AT_SYSINFO_EHDR
2972 return &gate_vma;
2973#else
2974 return NULL;
2975#endif
2976}
2977
2978int in_gate_area_no_task(unsigned long addr)
2979{
2980#ifdef AT_SYSINFO_EHDR
2981 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2982 return 1;
2983#endif
2984 return 0;
2985}
2986
2987#endif /* __HAVE_ARCH_GATE_AREA */
David Howells0ec76a12006-09-27 01:50:15 -07002988
Rik van Riel28b2ee22008-07-23 21:27:05 -07002989#ifdef CONFIG_HAVE_IOREMAP_PROT
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08002990int follow_phys(struct vm_area_struct *vma,
2991 unsigned long address, unsigned int flags,
2992 unsigned long *prot, resource_size_t *phys)
Rik van Riel28b2ee22008-07-23 21:27:05 -07002993{
2994 pgd_t *pgd;
2995 pud_t *pud;
2996 pmd_t *pmd;
2997 pte_t *ptep, pte;
2998 spinlock_t *ptl;
2999 resource_size_t phys_addr = 0;
3000 struct mm_struct *mm = vma->vm_mm;
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003001 int ret = -EINVAL;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003002
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003003 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3004 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003005
3006 pgd = pgd_offset(mm, address);
3007 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003008 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003009
3010 pud = pud_offset(pgd, address);
3011 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003012 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003013
3014 pmd = pmd_offset(pud, address);
3015 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003016 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003017
3018 /* We cannot handle huge page PFN maps. Luckily they don't exist. */
3019 if (pmd_huge(*pmd))
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003020 goto out;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003021
3022 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
3023 if (!ptep)
3024 goto out;
3025
3026 pte = *ptep;
3027 if (!pte_present(pte))
3028 goto unlock;
3029 if ((flags & FOLL_WRITE) && !pte_write(pte))
3030 goto unlock;
3031 phys_addr = pte_pfn(pte);
3032 phys_addr <<= PAGE_SHIFT; /* Shift here to avoid overflow on PAE */
3033
3034 *prot = pgprot_val(pte_pgprot(pte));
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003035 *phys = phys_addr;
3036 ret = 0;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003037
3038unlock:
3039 pte_unmap_unlock(ptep, ptl);
3040out:
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003041 return ret;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003042}
3043
3044int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
3045 void *buf, int len, int write)
3046{
3047 resource_size_t phys_addr;
3048 unsigned long prot = 0;
KOSAKI Motohiro2bc72732009-01-06 14:39:43 -08003049 void __iomem *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003050 int offset = addr & (PAGE_SIZE-1);
3051
venkatesh.pallipadi@intel.comd87fe662008-12-19 13:47:27 -08003052 if (follow_phys(vma, addr, write, &prot, &phys_addr))
Rik van Riel28b2ee22008-07-23 21:27:05 -07003053 return -EINVAL;
3054
3055 maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot);
3056 if (write)
3057 memcpy_toio(maddr + offset, buf, len);
3058 else
3059 memcpy_fromio(buf, maddr + offset, len);
3060 iounmap(maddr);
3061
3062 return len;
3063}
3064#endif
3065
David Howells0ec76a12006-09-27 01:50:15 -07003066/*
3067 * Access another process' address space.
3068 * Source/target buffer must be kernel space,
3069 * Do not walk the page table directly, use get_user_pages
3070 */
3071int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
3072{
3073 struct mm_struct *mm;
3074 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07003075 void *old_buf = buf;
3076
3077 mm = get_task_mm(tsk);
3078 if (!mm)
3079 return 0;
3080
3081 down_read(&mm->mmap_sem);
Simon Arlott183ff222007-10-20 01:27:18 +02003082 /* ignore errors, just check how much was successfully transferred */
David Howells0ec76a12006-09-27 01:50:15 -07003083 while (len) {
3084 int bytes, ret, offset;
3085 void *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07003086 struct page *page = NULL;
David Howells0ec76a12006-09-27 01:50:15 -07003087
3088 ret = get_user_pages(tsk, mm, addr, 1,
3089 write, 1, &page, &vma);
Rik van Riel28b2ee22008-07-23 21:27:05 -07003090 if (ret <= 0) {
3091 /*
3092 * Check if this is a VM_IO | VM_PFNMAP VMA, which
3093 * we can access using slightly different code.
3094 */
3095#ifdef CONFIG_HAVE_IOREMAP_PROT
3096 vma = find_vma(mm, addr);
3097 if (!vma)
3098 break;
3099 if (vma->vm_ops && vma->vm_ops->access)
3100 ret = vma->vm_ops->access(vma, addr, buf,
3101 len, write);
3102 if (ret <= 0)
3103#endif
3104 break;
3105 bytes = ret;
David Howells0ec76a12006-09-27 01:50:15 -07003106 } else {
Rik van Riel28b2ee22008-07-23 21:27:05 -07003107 bytes = len;
3108 offset = addr & (PAGE_SIZE-1);
3109 if (bytes > PAGE_SIZE-offset)
3110 bytes = PAGE_SIZE-offset;
3111
3112 maddr = kmap(page);
3113 if (write) {
3114 copy_to_user_page(vma, page, addr,
3115 maddr + offset, buf, bytes);
3116 set_page_dirty_lock(page);
3117 } else {
3118 copy_from_user_page(vma, page, addr,
3119 buf, maddr + offset, bytes);
3120 }
3121 kunmap(page);
3122 page_cache_release(page);
David Howells0ec76a12006-09-27 01:50:15 -07003123 }
David Howells0ec76a12006-09-27 01:50:15 -07003124 len -= bytes;
3125 buf += bytes;
3126 addr += bytes;
3127 }
3128 up_read(&mm->mmap_sem);
3129 mmput(mm);
3130
3131 return buf - old_buf;
3132}
Andi Kleen03252912008-01-30 13:33:18 +01003133
3134/*
3135 * Print the name of a VMA.
3136 */
3137void print_vma_addr(char *prefix, unsigned long ip)
3138{
3139 struct mm_struct *mm = current->mm;
3140 struct vm_area_struct *vma;
3141
Ingo Molnare8bff742008-02-13 20:21:06 +01003142 /*
3143 * Do not print if we are in atomic
3144 * contexts (in exception stacks, etc.):
3145 */
3146 if (preempt_count())
3147 return;
3148
Andi Kleen03252912008-01-30 13:33:18 +01003149 down_read(&mm->mmap_sem);
3150 vma = find_vma(mm, ip);
3151 if (vma && vma->vm_file) {
3152 struct file *f = vma->vm_file;
3153 char *buf = (char *)__get_free_page(GFP_KERNEL);
3154 if (buf) {
3155 char *p, *s;
3156
Jan Blunckcf28b482008-02-14 19:38:44 -08003157 p = d_path(&f->f_path, buf, PAGE_SIZE);
Andi Kleen03252912008-01-30 13:33:18 +01003158 if (IS_ERR(p))
3159 p = "?";
3160 s = strrchr(p, '/');
3161 if (s)
3162 p = s+1;
3163 printk("%s%s[%lx+%lx]", prefix, p,
3164 vma->vm_start,
3165 vma->vm_end - vma->vm_start);
3166 free_page((unsigned long)buf);
3167 }
3168 }
3169 up_read(&current->mm->mmap_sem);
3170}
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003171
3172#ifdef CONFIG_PROVE_LOCKING
3173void might_fault(void)
3174{
Peter Zijlstra95156f02009-01-12 13:02:11 +01003175 /*
3176 * Some code (nfs/sunrpc) uses socket ops on kernel memory while
3177 * holding the mmap_sem, this is safe because kernel memory doesn't
3178 * get paged out, therefore we'll never actually fault, and the
3179 * below annotations will generate false positives.
3180 */
3181 if (segment_eq(get_fs(), KERNEL_DS))
3182 return;
3183
Nick Piggin3ee1afa2008-09-10 13:37:17 +02003184 might_sleep();
3185 /*
3186 * it would be nicer only to annotate paths which are not under
3187 * pagefault_disable, however that requires a larger audit and
3188 * providing helpers like get_user_atomic.
3189 */
3190 if (!in_atomic() && current->mm)
3191 might_lock_read(&current->mm->mmap_sem);
3192}
3193EXPORT_SYMBOL(might_fault);
3194#endif