blob: 6dd1cd88bfb67411767b28e0cdc11a35b52fc92a [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
54#include <asm/pgalloc.h>
55#include <asm/uaccess.h>
56#include <asm/tlb.h>
57#include <asm/tlbflush.h>
58#include <asm/pgtable.h>
59
60#include <linux/swapops.h>
61#include <linux/elf.h>
62
Andy Whitcroftd41dee32005-06-23 00:07:54 -070063#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -070064/* use the per-pgdat data instead for discontigmem - mbligh */
65unsigned long max_mapnr;
66struct page *mem_map;
67
68EXPORT_SYMBOL(max_mapnr);
69EXPORT_SYMBOL(mem_map);
70#endif
71
72unsigned long num_physpages;
73/*
74 * A number of key systems in x86 including ioremap() rely on the assumption
75 * that high_memory defines the upper bound on direct map memory, then end
76 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
77 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
78 * and ZONE_HIGHMEM.
79 */
80void * high_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
82EXPORT_SYMBOL(num_physpages);
83EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Andi Kleena62eaf12006-02-16 23:41:58 +010085int randomize_va_space __read_mostly = 1;
86
87static int __init disable_randmaps(char *s)
88{
89 randomize_va_space = 0;
OGAWA Hirofumi9b410462006-03-31 02:30:33 -080090 return 1;
Andi Kleena62eaf12006-02-16 23:41:58 +010091}
92__setup("norandmaps", disable_randmaps);
93
94
Linus Torvalds1da177e2005-04-16 15:20:36 -070095/*
96 * If a p?d_bad entry is found while walking page tables, report
97 * the error, before resetting entry to p?d_none. Usually (but
98 * very seldom) called out from the p?d_none_or_clear_bad macros.
99 */
100
101void pgd_clear_bad(pgd_t *pgd)
102{
103 pgd_ERROR(*pgd);
104 pgd_clear(pgd);
105}
106
107void pud_clear_bad(pud_t *pud)
108{
109 pud_ERROR(*pud);
110 pud_clear(pud);
111}
112
113void pmd_clear_bad(pmd_t *pmd)
114{
115 pmd_ERROR(*pmd);
116 pmd_clear(pmd);
117}
118
119/*
120 * Note: this doesn't free the actual pages themselves. That
121 * has been handled earlier when unmapping all the memory regions.
122 */
Hugh Dickinse0da3822005-04-19 13:29:15 -0700123static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124{
Hugh Dickinse0da3822005-04-19 13:29:15 -0700125 struct page *page = pmd_page(*pmd);
126 pmd_clear(pmd);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700127 pte_lock_deinit(page);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700128 pte_free_tlb(tlb, page);
Christoph Lameterdf849a12006-06-30 01:55:38 -0700129 dec_zone_page_state(page, NR_PAGETABLE);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700130 tlb->mm->nr_ptes--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131}
132
Hugh Dickinse0da3822005-04-19 13:29:15 -0700133static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
134 unsigned long addr, unsigned long end,
135 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136{
137 pmd_t *pmd;
138 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700139 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140
Hugh Dickinse0da3822005-04-19 13:29:15 -0700141 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 do {
144 next = pmd_addr_end(addr, end);
145 if (pmd_none_or_clear_bad(pmd))
146 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700147 free_pte_range(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 } while (pmd++, addr = next, addr != end);
149
Hugh Dickinse0da3822005-04-19 13:29:15 -0700150 start &= PUD_MASK;
151 if (start < floor)
152 return;
153 if (ceiling) {
154 ceiling &= PUD_MASK;
155 if (!ceiling)
156 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700158 if (end - 1 > ceiling - 1)
159 return;
160
161 pmd = pmd_offset(pud, start);
162 pud_clear(pud);
163 pmd_free_tlb(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164}
165
Hugh Dickinse0da3822005-04-19 13:29:15 -0700166static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
167 unsigned long addr, unsigned long end,
168 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169{
170 pud_t *pud;
171 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700172 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173
Hugh Dickinse0da3822005-04-19 13:29:15 -0700174 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 pud = pud_offset(pgd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 do {
177 next = pud_addr_end(addr, end);
178 if (pud_none_or_clear_bad(pud))
179 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700180 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 } while (pud++, addr = next, addr != end);
182
Hugh Dickinse0da3822005-04-19 13:29:15 -0700183 start &= PGDIR_MASK;
184 if (start < floor)
185 return;
186 if (ceiling) {
187 ceiling &= PGDIR_MASK;
188 if (!ceiling)
189 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700191 if (end - 1 > ceiling - 1)
192 return;
193
194 pud = pud_offset(pgd, start);
195 pgd_clear(pgd);
196 pud_free_tlb(tlb, pud);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197}
198
199/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700200 * This function frees user-level page tables of a process.
201 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 * Must be called with pagetable lock held.
203 */
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700204void free_pgd_range(struct mmu_gather **tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700205 unsigned long addr, unsigned long end,
206 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207{
208 pgd_t *pgd;
209 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700210 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
Hugh Dickinse0da3822005-04-19 13:29:15 -0700212 /*
213 * The next few lines have given us lots of grief...
214 *
215 * Why are we testing PMD* at this top level? Because often
216 * there will be no work to do at all, and we'd prefer not to
217 * go all the way down to the bottom just to discover that.
218 *
219 * Why all these "- 1"s? Because 0 represents both the bottom
220 * of the address space and the top of it (using -1 for the
221 * top wouldn't help much: the masks would do the wrong thing).
222 * The rule is that addr 0 and floor 0 refer to the bottom of
223 * the address space, but end 0 and ceiling 0 refer to the top
224 * Comparisons need to use "end - 1" and "ceiling - 1" (though
225 * that end 0 case should be mythical).
226 *
227 * Wherever addr is brought up or ceiling brought down, we must
228 * be careful to reject "the opposite 0" before it confuses the
229 * subsequent tests. But what about where end is brought down
230 * by PMD_SIZE below? no, end can't go down to 0 there.
231 *
232 * Whereas we round start (addr) and ceiling down, by different
233 * masks at different levels, in order to test whether a table
234 * now has no other vmas using it, so can be freed, we don't
235 * bother to round floor or end up - the tests don't need that.
236 */
237
238 addr &= PMD_MASK;
239 if (addr < floor) {
240 addr += PMD_SIZE;
241 if (!addr)
242 return;
243 }
244 if (ceiling) {
245 ceiling &= PMD_MASK;
246 if (!ceiling)
247 return;
248 }
249 if (end - 1 > ceiling - 1)
250 end -= PMD_SIZE;
251 if (addr > end - 1)
252 return;
253
254 start = addr;
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700255 pgd = pgd_offset((*tlb)->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 do {
257 next = pgd_addr_end(addr, end);
258 if (pgd_none_or_clear_bad(pgd))
259 continue;
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700260 free_pud_range(*tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700262}
263
264void free_pgtables(struct mmu_gather **tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700265 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700266{
267 while (vma) {
268 struct vm_area_struct *next = vma->vm_next;
269 unsigned long addr = vma->vm_start;
270
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700271 /*
272 * Hide vma from rmap and vmtruncate before freeing pgtables
273 */
274 anon_vma_unlink(vma);
275 unlink_file_vma(vma);
276
David Gibson9da61ae2006-03-22 00:08:57 -0800277 if (is_vm_hugetlb_page(vma)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700278 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700279 floor, next? next->vm_start: ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700280 } else {
281 /*
282 * Optimization: gather nearby vmas into one call down
283 */
284 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
David Gibson48669202006-03-22 00:08:58 -0800285 && !is_vm_hugetlb_page(next)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700286 vma = next;
287 next = vma->vm_next;
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700288 anon_vma_unlink(vma);
289 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700290 }
291 free_pgd_range(tlb, addr, vma->vm_end,
292 floor, next? next->vm_start: ceiling);
293 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700294 vma = next;
295 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296}
297
Hugh Dickins1bb36302005-10-29 18:16:22 -0700298int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299{
Hugh Dickinsc74df322005-10-29 18:16:23 -0700300 struct page *new = pte_alloc_one(mm, address);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700301 if (!new)
302 return -ENOMEM;
303
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700304 pte_lock_init(new);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700305 spin_lock(&mm->page_table_lock);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700306 if (pmd_present(*pmd)) { /* Another has populated it */
307 pte_lock_deinit(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700308 pte_free(new);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700309 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 mm->nr_ptes++;
Christoph Lameterdf849a12006-06-30 01:55:38 -0700311 inc_zone_page_state(new, NR_PAGETABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 pmd_populate(mm, pmd, new);
313 }
Hugh Dickinsc74df322005-10-29 18:16:23 -0700314 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700315 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316}
317
Hugh Dickins1bb36302005-10-29 18:16:22 -0700318int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700320 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
321 if (!new)
322 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323
Hugh Dickins1bb36302005-10-29 18:16:22 -0700324 spin_lock(&init_mm.page_table_lock);
325 if (pmd_present(*pmd)) /* Another has populated it */
326 pte_free_kernel(new);
327 else
328 pmd_populate_kernel(&init_mm, pmd, new);
329 spin_unlock(&init_mm.page_table_lock);
330 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331}
332
Hugh Dickinsae859762005-10-29 18:16:05 -0700333static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
334{
335 if (file_rss)
336 add_mm_counter(mm, file_rss, file_rss);
337 if (anon_rss)
338 add_mm_counter(mm, anon_rss, anon_rss);
339}
340
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800342 * This function is called to print an error when a bad pte
343 * is found. For example, we might have a PFN-mapped pte in
344 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700345 *
346 * The calling function must still handle the error.
347 */
348void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr)
349{
350 printk(KERN_ERR "Bad pte = %08llx, process = %s, "
351 "vm_flags = %lx, vaddr = %lx\n",
352 (long long)pte_val(pte),
353 (vma->vm_mm == current->mm ? current->comm : "???"),
354 vma->vm_flags, vaddr);
355 dump_stack();
356}
357
Linus Torvalds67121172005-12-11 20:38:17 -0800358static inline int is_cow_mapping(unsigned int flags)
359{
360 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
361}
362
Nick Pigginb5810032005-10-29 18:16:12 -0700363/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800364 * This function gets the "struct page" associated with a pte.
365 *
366 * NOTE! Some mappings do not have "struct pages". A raw PFN mapping
367 * will have each page table entry just pointing to a raw page frame
368 * number, and as far as the VM layer is concerned, those do not have
369 * pages associated with them - even if the PFN might point to memory
370 * that otherwise is perfectly fine and has a "struct page".
371 *
372 * The way we recognize those mappings is through the rules set up
373 * by "remap_pfn_range()": the vma will have the VM_PFNMAP bit set,
374 * and the vm_pgoff will point to the first PFN mapped: thus every
375 * page that is a raw mapping will always honor the rule
376 *
377 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
378 *
379 * and if that isn't true, the page has been COW'ed (in which case it
380 * _does_ have a "struct page" associated with it even if it is in a
381 * VM_PFNMAP range).
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800382 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800383struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800384{
Linus Torvalds6aab3412005-11-28 14:34:23 -0800385 unsigned long pfn = pte_pfn(pte);
386
Nick Pigginb7ab7952006-03-22 00:08:42 -0800387 if (unlikely(vma->vm_flags & VM_PFNMAP)) {
Linus Torvalds6aab3412005-11-28 14:34:23 -0800388 unsigned long off = (addr - vma->vm_start) >> PAGE_SHIFT;
389 if (pfn == vma->vm_pgoff + off)
390 return NULL;
Linus Torvalds67121172005-12-11 20:38:17 -0800391 if (!is_cow_mapping(vma->vm_flags))
Linus Torvaldsfb155c12005-12-11 19:46:02 -0800392 return NULL;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800393 }
394
Carsten Otte97231982008-01-17 15:21:17 -0800395#ifdef CONFIG_DEBUG_VM
Nick Piggin315ab192006-03-25 16:20:22 +0100396 /*
397 * Add some anal sanity checks for now. Eventually,
398 * we should just do "return pfn_to_page(pfn)", but
399 * in the meantime we check that we get a valid pfn,
400 * and that the resulting page looks ok.
401 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800402 if (unlikely(!pfn_valid(pfn))) {
403 print_bad_pte(vma, pte, addr);
404 return NULL;
405 }
Carsten Otte97231982008-01-17 15:21:17 -0800406#endif
Linus Torvalds6aab3412005-11-28 14:34:23 -0800407
408 /*
409 * NOTE! We still have PageReserved() pages in the page
410 * tables.
411 *
412 * The PAGE_ZERO() pages and various VDSO mappings can
413 * cause them to exist.
414 */
415 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800416}
417
418/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 * copy one vm_area from one task to the other. Assumes the page tables
420 * already present in the new task to be cleared in the whole range
421 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 */
423
Hugh Dickins8c103762005-10-29 18:16:13 -0700424static inline void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700426 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700427 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428{
Nick Pigginb5810032005-10-29 18:16:12 -0700429 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 pte_t pte = *src_pte;
431 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432
433 /* pte contains position in swap or file, so copy. */
434 if (unlikely(!pte_present(pte))) {
435 if (!pte_file(pte)) {
Christoph Lameter06972122006-06-23 02:03:35 -0700436 swp_entry_t entry = pte_to_swp_entry(pte);
437
438 swap_duplicate(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 /* make sure dst_mm is on swapoff's mmlist. */
440 if (unlikely(list_empty(&dst_mm->mmlist))) {
441 spin_lock(&mmlist_lock);
Hugh Dickinsf412ac02005-10-29 18:16:41 -0700442 if (list_empty(&dst_mm->mmlist))
443 list_add(&dst_mm->mmlist,
444 &src_mm->mmlist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 spin_unlock(&mmlist_lock);
446 }
Christoph Lameter06972122006-06-23 02:03:35 -0700447 if (is_write_migration_entry(entry) &&
448 is_cow_mapping(vm_flags)) {
449 /*
450 * COW mappings require pages in both parent
451 * and child to be set to read.
452 */
453 make_migration_entry_read(&entry);
454 pte = swp_entry_to_pte(entry);
455 set_pte_at(src_mm, addr, src_pte, pte);
456 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700458 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 }
460
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 /*
462 * If it's a COW mapping, write protect it both
463 * in the parent and the child
464 */
Linus Torvalds67121172005-12-11 20:38:17 -0800465 if (is_cow_mapping(vm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 ptep_set_wrprotect(src_mm, addr, src_pte);
Zachary Amsden3dc90792006-09-30 23:29:30 -0700467 pte = pte_wrprotect(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 }
469
470 /*
471 * If it's a shared mapping, mark it clean in
472 * the child
473 */
474 if (vm_flags & VM_SHARED)
475 pte = pte_mkclean(pte);
476 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800477
478 page = vm_normal_page(vma, addr, pte);
479 if (page) {
480 get_page(page);
Nick Pigginc97a9e12007-05-16 22:11:21 -0700481 page_dup_rmap(page, vma, addr);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800482 rss[!!PageAnon(page)]++;
483 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700484
485out_set_pte:
486 set_pte_at(dst_mm, addr, dst_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487}
488
489static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
490 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
491 unsigned long addr, unsigned long end)
492{
493 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700494 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -0700495 int progress = 0;
Hugh Dickins8c103762005-10-29 18:16:13 -0700496 int rss[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
498again:
Hugh Dickinsae859762005-10-29 18:16:05 -0700499 rss[1] = rss[0] = 0;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700500 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 if (!dst_pte)
502 return -ENOMEM;
503 src_pte = pte_offset_map_nested(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700504 src_ptl = pte_lockptr(src_mm, src_pmd);
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700505 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700506 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 do {
509 /*
510 * We are holding two locks at this point - either of them
511 * could generate latencies in another task on another CPU.
512 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700513 if (progress >= 32) {
514 progress = 0;
515 if (need_resched() ||
Hugh Dickinsc74df322005-10-29 18:16:23 -0700516 need_lockbreak(src_ptl) ||
517 need_lockbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700518 break;
519 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 if (pte_none(*src_pte)) {
521 progress++;
522 continue;
523 }
Hugh Dickins8c103762005-10-29 18:16:13 -0700524 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 progress += 8;
526 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700528 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -0700529 spin_unlock(src_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 pte_unmap_nested(src_pte - 1);
Hugh Dickinsae859762005-10-29 18:16:05 -0700531 add_mm_rss(dst_mm, rss[0], rss[1]);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700532 pte_unmap_unlock(dst_pte - 1, dst_ptl);
533 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 if (addr != end)
535 goto again;
536 return 0;
537}
538
539static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
540 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
541 unsigned long addr, unsigned long end)
542{
543 pmd_t *src_pmd, *dst_pmd;
544 unsigned long next;
545
546 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
547 if (!dst_pmd)
548 return -ENOMEM;
549 src_pmd = pmd_offset(src_pud, addr);
550 do {
551 next = pmd_addr_end(addr, end);
552 if (pmd_none_or_clear_bad(src_pmd))
553 continue;
554 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
555 vma, addr, next))
556 return -ENOMEM;
557 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
558 return 0;
559}
560
561static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
562 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
563 unsigned long addr, unsigned long end)
564{
565 pud_t *src_pud, *dst_pud;
566 unsigned long next;
567
568 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
569 if (!dst_pud)
570 return -ENOMEM;
571 src_pud = pud_offset(src_pgd, addr);
572 do {
573 next = pud_addr_end(addr, end);
574 if (pud_none_or_clear_bad(src_pud))
575 continue;
576 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
577 vma, addr, next))
578 return -ENOMEM;
579 } while (dst_pud++, src_pud++, addr = next, addr != end);
580 return 0;
581}
582
583int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
584 struct vm_area_struct *vma)
585{
586 pgd_t *src_pgd, *dst_pgd;
587 unsigned long next;
588 unsigned long addr = vma->vm_start;
589 unsigned long end = vma->vm_end;
590
Nick Piggind9928952005-08-28 16:49:11 +1000591 /*
592 * Don't copy ptes where a page fault will fill them correctly.
593 * Fork becomes much lighter when there are big shared or private
594 * readonly mappings. The tradeoff is that copy_page_range is more
595 * efficient than faulting.
596 */
Linus Torvalds4d7672b2005-12-16 10:21:23 -0800597 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
Nick Piggind9928952005-08-28 16:49:11 +1000598 if (!vma->anon_vma)
599 return 0;
600 }
601
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 if (is_vm_hugetlb_page(vma))
603 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
604
605 dst_pgd = pgd_offset(dst_mm, addr);
606 src_pgd = pgd_offset(src_mm, addr);
607 do {
608 next = pgd_addr_end(addr, end);
609 if (pgd_none_or_clear_bad(src_pgd))
610 continue;
611 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
612 vma, addr, next))
613 return -ENOMEM;
614 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
615 return 0;
616}
617
Robin Holt51c6f662005-11-13 16:06:42 -0800618static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700619 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800621 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622{
Nick Pigginb5810032005-10-29 18:16:12 -0700623 struct mm_struct *mm = tlb->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 pte_t *pte;
Hugh Dickins508034a2005-10-29 18:16:30 -0700625 spinlock_t *ptl;
Hugh Dickinsae859762005-10-29 18:16:05 -0700626 int file_rss = 0;
627 int anon_rss = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
Hugh Dickins508034a2005-10-29 18:16:30 -0700629 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700630 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 do {
632 pte_t ptent = *pte;
Robin Holt51c6f662005-11-13 16:06:42 -0800633 if (pte_none(ptent)) {
634 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800636 }
Hugh Dickins6f5e6b92006-03-16 23:04:09 -0800637
638 (*zap_work) -= PAGE_SIZE;
639
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800641 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -0800642
Linus Torvalds6aab3412005-11-28 14:34:23 -0800643 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 if (unlikely(details) && page) {
645 /*
646 * unmap_shared_mapping_pages() wants to
647 * invalidate cache without truncating:
648 * unmap shared but keep private pages.
649 */
650 if (details->check_mapping &&
651 details->check_mapping != page->mapping)
652 continue;
653 /*
654 * Each page->index must be checked when
655 * invalidating or truncating nonlinear.
656 */
657 if (details->nonlinear_vma &&
658 (page->index < details->first_index ||
659 page->index > details->last_index))
660 continue;
661 }
Nick Pigginb5810032005-10-29 18:16:12 -0700662 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -0700663 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 tlb_remove_tlb_entry(tlb, pte, addr);
665 if (unlikely(!page))
666 continue;
667 if (unlikely(details) && details->nonlinear_vma
668 && linear_page_index(details->nonlinear_vma,
669 addr) != page->index)
Nick Pigginb5810032005-10-29 18:16:12 -0700670 set_pte_at(mm, addr, pte,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 pgoff_to_pte(page->index));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 if (PageAnon(page))
Hugh Dickins86d912f2005-10-29 18:16:14 -0700673 anon_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700674 else {
675 if (pte_dirty(ptent))
676 set_page_dirty(page);
677 if (pte_young(ptent))
Ken Chendaa88c82007-02-10 01:43:18 -0800678 SetPageReferenced(page);
Hugh Dickins86d912f2005-10-29 18:16:14 -0700679 file_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700680 }
Nick Piggin7de6b802006-12-22 01:09:33 -0800681 page_remove_rmap(page, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 tlb_remove_page(tlb, page);
683 continue;
684 }
685 /*
686 * If details->check_mapping, we leave swap entries;
687 * if details->nonlinear_vma, we leave file entries.
688 */
689 if (unlikely(details))
690 continue;
691 if (!pte_file(ptent))
692 free_swap_and_cache(pte_to_swp_entry(ptent));
Zachary Amsden9888a1c2006-09-30 23:29:31 -0700693 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
Robin Holt51c6f662005-11-13 16:06:42 -0800694 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
Hugh Dickinsae859762005-10-29 18:16:05 -0700695
Hugh Dickins86d912f2005-10-29 18:16:14 -0700696 add_mm_rss(mm, file_rss, anon_rss);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700697 arch_leave_lazy_mmu_mode();
Hugh Dickins508034a2005-10-29 18:16:30 -0700698 pte_unmap_unlock(pte - 1, ptl);
Robin Holt51c6f662005-11-13 16:06:42 -0800699
700 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701}
702
Robin Holt51c6f662005-11-13 16:06:42 -0800703static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700704 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800706 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707{
708 pmd_t *pmd;
709 unsigned long next;
710
711 pmd = pmd_offset(pud, addr);
712 do {
713 next = pmd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800714 if (pmd_none_or_clear_bad(pmd)) {
715 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800717 }
718 next = zap_pte_range(tlb, vma, pmd, addr, next,
719 zap_work, details);
720 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
721
722 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723}
724
Robin Holt51c6f662005-11-13 16:06:42 -0800725static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700726 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800728 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729{
730 pud_t *pud;
731 unsigned long next;
732
733 pud = pud_offset(pgd, addr);
734 do {
735 next = pud_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800736 if (pud_none_or_clear_bad(pud)) {
737 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800739 }
740 next = zap_pmd_range(tlb, vma, pud, addr, next,
741 zap_work, details);
742 } while (pud++, addr = next, (addr != end && *zap_work > 0));
743
744 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745}
746
Robin Holt51c6f662005-11-13 16:06:42 -0800747static unsigned long unmap_page_range(struct mmu_gather *tlb,
748 struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800750 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751{
752 pgd_t *pgd;
753 unsigned long next;
754
755 if (details && !details->check_mapping && !details->nonlinear_vma)
756 details = NULL;
757
758 BUG_ON(addr >= end);
759 tlb_start_vma(tlb, vma);
760 pgd = pgd_offset(vma->vm_mm, addr);
761 do {
762 next = pgd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800763 if (pgd_none_or_clear_bad(pgd)) {
764 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800766 }
767 next = zap_pud_range(tlb, vma, pgd, addr, next,
768 zap_work, details);
769 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 tlb_end_vma(tlb, vma);
Robin Holt51c6f662005-11-13 16:06:42 -0800771
772 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773}
774
775#ifdef CONFIG_PREEMPT
776# define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
777#else
778/* No preempt: go for improved straight-line efficiency */
779# define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
780#endif
781
782/**
783 * unmap_vmas - unmap a range of memory covered by a list of vma's
784 * @tlbp: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 * @vma: the starting vma
786 * @start_addr: virtual address at which to start unmapping
787 * @end_addr: virtual address at which to end unmapping
788 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
789 * @details: details of nonlinear truncation or shared cache invalidation
790 *
Hugh Dickinsee39b372005-04-19 13:29:15 -0700791 * Returns the end address of the unmapping (restart addr if interrupted).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700793 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700795 * We aim to not hold locks for too long (for scheduling latency reasons).
796 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 * return the ending mmu_gather to the caller.
798 *
799 * Only addresses between `start' and `end' will be unmapped.
800 *
801 * The VMA list must be sorted in ascending virtual address order.
802 *
803 * unmap_vmas() assumes that the caller will flush the whole unmapped address
804 * range after unmap_vmas() returns. So the only responsibility here is to
805 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
806 * drops the lock and schedules.
807 */
Hugh Dickins508034a2005-10-29 18:16:30 -0700808unsigned long unmap_vmas(struct mmu_gather **tlbp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 struct vm_area_struct *vma, unsigned long start_addr,
810 unsigned long end_addr, unsigned long *nr_accounted,
811 struct zap_details *details)
812{
Robin Holt51c6f662005-11-13 16:06:42 -0800813 long zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
815 int tlb_start_valid = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -0700816 unsigned long start = start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700818 int fullmm = (*tlbp)->fullmm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
820 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 unsigned long end;
822
823 start = max(vma->vm_start, start_addr);
824 if (start >= vma->vm_end)
825 continue;
826 end = min(vma->vm_end, end_addr);
827 if (end <= vma->vm_start)
828 continue;
829
830 if (vma->vm_flags & VM_ACCOUNT)
831 *nr_accounted += (end - start) >> PAGE_SHIFT;
832
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 while (start != end) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 if (!tlb_start_valid) {
835 tlb_start = start;
836 tlb_start_valid = 1;
837 }
838
Robin Holt51c6f662005-11-13 16:06:42 -0800839 if (unlikely(is_vm_hugetlb_page(vma))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 unmap_hugepage_range(vma, start, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800841 zap_work -= (end - start) /
842 (HPAGE_SIZE / PAGE_SIZE);
843 start = end;
844 } else
845 start = unmap_page_range(*tlbp, vma,
846 start, end, &zap_work, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847
Robin Holt51c6f662005-11-13 16:06:42 -0800848 if (zap_work > 0) {
849 BUG_ON(start != end);
850 break;
851 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852
853 tlb_finish_mmu(*tlbp, tlb_start, start);
854
855 if (need_resched() ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 (i_mmap_lock && need_lockbreak(i_mmap_lock))) {
857 if (i_mmap_lock) {
Hugh Dickins508034a2005-10-29 18:16:30 -0700858 *tlbp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 goto out;
860 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 }
863
Hugh Dickins508034a2005-10-29 18:16:30 -0700864 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 tlb_start_valid = 0;
Robin Holt51c6f662005-11-13 16:06:42 -0800866 zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 }
868 }
869out:
Hugh Dickinsee39b372005-04-19 13:29:15 -0700870 return start; /* which is now the end (or restart) address */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871}
872
873/**
874 * zap_page_range - remove user pages in a given range
875 * @vma: vm_area_struct holding the applicable pages
876 * @address: starting address of pages to zap
877 * @size: number of bytes to zap
878 * @details: details of nonlinear truncation or shared cache invalidation
879 */
Hugh Dickinsee39b372005-04-19 13:29:15 -0700880unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 unsigned long size, struct zap_details *details)
882{
883 struct mm_struct *mm = vma->vm_mm;
884 struct mmu_gather *tlb;
885 unsigned long end = address + size;
886 unsigned long nr_accounted = 0;
887
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -0700890 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -0700891 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
892 if (tlb)
893 tlb_finish_mmu(tlb, address, end);
Hugh Dickinsee39b372005-04-19 13:29:15 -0700894 return end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895}
896
897/*
898 * Do a quick page-table lookup for a single page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800900struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700901 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902{
903 pgd_t *pgd;
904 pud_t *pud;
905 pmd_t *pmd;
906 pte_t *ptep, pte;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700907 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 struct page *page;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800909 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700911 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
912 if (!IS_ERR(page)) {
913 BUG_ON(flags & FOLL_GET);
914 goto out;
915 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700917 page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 pgd = pgd_offset(mm, address);
919 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700920 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921
922 pud = pud_offset(pgd, address);
923 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700924 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925
926 pmd = pmd_offset(pud, address);
927 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700928 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700930 if (pmd_huge(*pmd)) {
931 BUG_ON(flags & FOLL_GET);
932 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
933 goto out;
934 }
935
936 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 if (!ptep)
938 goto out;
939
940 pte = *ptep;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700941 if (!pte_present(pte))
942 goto unlock;
943 if ((flags & FOLL_WRITE) && !pte_write(pte))
944 goto unlock;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800945 page = vm_normal_page(vma, address, pte);
946 if (unlikely(!page))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700947 goto unlock;
948
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700949 if (flags & FOLL_GET)
950 get_page(page);
951 if (flags & FOLL_TOUCH) {
952 if ((flags & FOLL_WRITE) &&
953 !pte_dirty(pte) && !PageDirty(page))
954 set_page_dirty(page);
955 mark_page_accessed(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700957unlock:
958 pte_unmap_unlock(ptep, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959out:
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700960 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700962no_page_table:
963 /*
964 * When core dumping an enormous anonymous area that nobody
965 * has touched so far, we don't want to allocate page tables.
966 */
967 if (flags & FOLL_ANON) {
Nick Piggin557ed1f2007-10-16 01:24:40 -0700968 page = ZERO_PAGE(0);
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700969 if (flags & FOLL_GET)
970 get_page(page);
971 BUG_ON(flags & FOLL_WRITE);
972 }
973 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974}
975
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
977 unsigned long start, int len, int write, int force,
978 struct page **pages, struct vm_area_struct **vmas)
979{
980 int i;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700981 unsigned int vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982
983 /*
984 * Require read or write permissions.
985 * If 'force' is set, we only require the "MAY" flags.
986 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700987 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
988 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 i = 0;
990
991 do {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700992 struct vm_area_struct *vma;
993 unsigned int foll_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
995 vma = find_extend_vma(mm, start);
996 if (!vma && in_gate_area(tsk, start)) {
997 unsigned long pg = start & PAGE_MASK;
998 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
999 pgd_t *pgd;
1000 pud_t *pud;
1001 pmd_t *pmd;
1002 pte_t *pte;
1003 if (write) /* user gate pages are read-only */
1004 return i ? : -EFAULT;
1005 if (pg > TASK_SIZE)
1006 pgd = pgd_offset_k(pg);
1007 else
1008 pgd = pgd_offset_gate(mm, pg);
1009 BUG_ON(pgd_none(*pgd));
1010 pud = pud_offset(pgd, pg);
1011 BUG_ON(pud_none(*pud));
1012 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001013 if (pmd_none(*pmd))
1014 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001016 if (pte_none(*pte)) {
1017 pte_unmap(pte);
1018 return i ? : -EFAULT;
1019 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 if (pages) {
Nick Pigginfa2a4552005-11-29 18:43:17 +11001021 struct page *page = vm_normal_page(gate_vma, start, *pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001022 pages[i] = page;
1023 if (page)
1024 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 }
1026 pte_unmap(pte);
1027 if (vmas)
1028 vmas[i] = gate_vma;
1029 i++;
1030 start += PAGE_SIZE;
1031 len--;
1032 continue;
1033 }
1034
Linus Torvalds1ff80382005-12-12 16:24:33 -08001035 if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001036 || !(vm_flags & vma->vm_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 return i ? : -EFAULT;
1038
1039 if (is_vm_hugetlb_page(vma)) {
1040 i = follow_hugetlb_page(mm, vma, pages, vmas,
Adam Litke5b23dbe2007-11-14 16:59:33 -08001041 &start, &len, i, write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 continue;
1043 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001044
1045 foll_flags = FOLL_TOUCH;
1046 if (pages)
1047 foll_flags |= FOLL_GET;
1048 if (!write && !(vma->vm_flags & VM_LOCKED) &&
Nick Piggin54cb8822007-07-19 01:46:59 -07001049 (!vma->vm_ops || (!vma->vm_ops->nopage &&
1050 !vma->vm_ops->fault)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001051 foll_flags |= FOLL_ANON;
1052
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 do {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001054 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055
Ethan Solomita462e00c2007-07-15 23:38:16 -07001056 /*
1057 * If tsk is ooming, cut off its access to large memory
1058 * allocations. It has a pending SIGKILL, but it can't
1059 * be processed until returning to user space.
1060 */
1061 if (unlikely(test_tsk_thread_flag(tsk, TIF_MEMDIE)))
1062 return -ENOMEM;
1063
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001064 if (write)
1065 foll_flags |= FOLL_WRITE;
1066
1067 cond_resched();
Linus Torvalds6aab3412005-11-28 14:34:23 -08001068 while (!(page = follow_page(vma, start, foll_flags))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001069 int ret;
Nick Piggin83c54072007-07-19 01:47:05 -07001070 ret = handle_mm_fault(mm, vma, start,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001071 foll_flags & FOLL_WRITE);
Nick Piggin83c54072007-07-19 01:47:05 -07001072 if (ret & VM_FAULT_ERROR) {
1073 if (ret & VM_FAULT_OOM)
1074 return i ? i : -ENOMEM;
1075 else if (ret & VM_FAULT_SIGBUS)
1076 return i ? i : -EFAULT;
1077 BUG();
1078 }
1079 if (ret & VM_FAULT_MAJOR)
1080 tsk->maj_flt++;
1081 else
1082 tsk->min_flt++;
1083
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001084 /*
Nick Piggin83c54072007-07-19 01:47:05 -07001085 * The VM_FAULT_WRITE bit tells us that
1086 * do_wp_page has broken COW when necessary,
1087 * even if maybe_mkwrite decided not to set
1088 * pte_write. We can thus safely do subsequent
1089 * page lookups as if they were reads.
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001090 */
1091 if (ret & VM_FAULT_WRITE)
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001092 foll_flags &= ~FOLL_WRITE;
Nick Piggin83c54072007-07-19 01:47:05 -07001093
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07001094 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 }
1096 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001097 pages[i] = page;
James Bottomley03beb072006-03-26 01:36:57 -08001098
Russell Kinga6f36be2006-12-30 22:24:19 +00001099 flush_anon_page(vma, page, start);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001100 flush_dcache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 }
1102 if (vmas)
1103 vmas[i] = vma;
1104 i++;
1105 start += PAGE_SIZE;
1106 len--;
Hugh Dickins08ef4722005-06-21 17:15:10 -07001107 } while (len && start < vma->vm_end);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001108 } while (len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 return i;
1110}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111EXPORT_SYMBOL(get_user_pages);
1112
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001113pte_t * fastcall get_locked_pte(struct mm_struct *mm, unsigned long addr, spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001114{
1115 pgd_t * pgd = pgd_offset(mm, addr);
1116 pud_t * pud = pud_alloc(mm, pgd, addr);
1117 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001118 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001119 if (pmd)
1120 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1121 }
1122 return NULL;
1123}
1124
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001126 * This is the old fallback for page remapping.
1127 *
1128 * For historical reasons, it only allows reserved pages. Only
1129 * old drivers should use this, and they needed to mark their
1130 * pages reserved for the old functions anyway.
1131 */
1132static int insert_page(struct mm_struct *mm, unsigned long addr, struct page *page, pgprot_t prot)
1133{
1134 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001135 pte_t *pte;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001136 spinlock_t *ptl;
1137
1138 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001139 if (PageAnon(page))
Linus Torvalds238f58d2005-11-29 13:01:56 -08001140 goto out;
1141 retval = -ENOMEM;
1142 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001143 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001144 if (!pte)
1145 goto out;
1146 retval = -EBUSY;
1147 if (!pte_none(*pte))
1148 goto out_unlock;
1149
1150 /* Ok, finally just insert the thing.. */
1151 get_page(page);
1152 inc_mm_counter(mm, file_rss);
1153 page_add_file_rmap(page);
1154 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1155
1156 retval = 0;
1157out_unlock:
1158 pte_unmap_unlock(pte, ptl);
1159out:
1160 return retval;
1161}
1162
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001163/**
1164 * vm_insert_page - insert single page into user vma
1165 * @vma: user vma to map to
1166 * @addr: target user address of this page
1167 * @page: source kernel page
1168 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001169 * This allows drivers to insert individual pages they've allocated
1170 * into a user vma.
1171 *
1172 * The page has to be a nice clean _individual_ kernel allocation.
1173 * If you allocate a compound page, you need to have marked it as
1174 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001175 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001176 *
1177 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1178 * took an arbitrary page protection parameter. This doesn't allow
1179 * that. Your vma protection will have to be set up correctly, which
1180 * means that if you want a shared writable mapping, you'd better
1181 * ask for a shared writable mapping!
1182 *
1183 * The page does not need to be reserved.
1184 */
1185int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, struct page *page)
1186{
1187 if (addr < vma->vm_start || addr >= vma->vm_end)
1188 return -EFAULT;
1189 if (!page_count(page))
1190 return -EINVAL;
Linus Torvalds4d7672b2005-12-16 10:21:23 -08001191 vma->vm_flags |= VM_INSERTPAGE;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001192 return insert_page(vma->vm_mm, addr, page, vma->vm_page_prot);
1193}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001194EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001195
Nick Piggine0dc0d82007-02-12 00:51:36 -08001196/**
1197 * vm_insert_pfn - insert single pfn into user vma
1198 * @vma: user vma to map to
1199 * @addr: target user address of this page
1200 * @pfn: source kernel pfn
1201 *
1202 * Similar to vm_inert_page, this allows drivers to insert individual pages
1203 * they've allocated into a user vma. Same comments apply.
1204 *
1205 * This function should only be called from a vm_ops->fault handler, and
1206 * in that case the handler should return NULL.
1207 */
1208int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1209 unsigned long pfn)
1210{
1211 struct mm_struct *mm = vma->vm_mm;
1212 int retval;
1213 pte_t *pte, entry;
1214 spinlock_t *ptl;
1215
1216 BUG_ON(!(vma->vm_flags & VM_PFNMAP));
1217 BUG_ON(is_cow_mapping(vma->vm_flags));
1218
1219 retval = -ENOMEM;
1220 pte = get_locked_pte(mm, addr, &ptl);
1221 if (!pte)
1222 goto out;
1223 retval = -EBUSY;
1224 if (!pte_none(*pte))
1225 goto out_unlock;
1226
1227 /* Ok, finally just insert the thing.. */
1228 entry = pfn_pte(pfn, vma->vm_page_prot);
1229 set_pte_at(mm, addr, pte, entry);
1230 update_mmu_cache(vma, addr, entry);
1231
1232 retval = 0;
1233out_unlock:
1234 pte_unmap_unlock(pte, ptl);
1235
1236out:
1237 return retval;
1238}
1239EXPORT_SYMBOL(vm_insert_pfn);
1240
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001241/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 * maps a range of physical memory into the requested pages. the old
1243 * mappings are removed. any references to nonexistent pages results
1244 * in null mappings (currently treated as "copy-on-access")
1245 */
1246static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1247 unsigned long addr, unsigned long end,
1248 unsigned long pfn, pgprot_t prot)
1249{
1250 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001251 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252
Hugh Dickinsc74df322005-10-29 18:16:23 -07001253 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 if (!pte)
1255 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001256 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 do {
1258 BUG_ON(!pte_none(*pte));
Nick Pigginb5810032005-10-29 18:16:12 -07001259 set_pte_at(mm, addr, pte, pfn_pte(pfn, prot));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 pfn++;
1261 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001262 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001263 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 return 0;
1265}
1266
1267static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1268 unsigned long addr, unsigned long end,
1269 unsigned long pfn, pgprot_t prot)
1270{
1271 pmd_t *pmd;
1272 unsigned long next;
1273
1274 pfn -= addr >> PAGE_SHIFT;
1275 pmd = pmd_alloc(mm, pud, addr);
1276 if (!pmd)
1277 return -ENOMEM;
1278 do {
1279 next = pmd_addr_end(addr, end);
1280 if (remap_pte_range(mm, pmd, addr, next,
1281 pfn + (addr >> PAGE_SHIFT), prot))
1282 return -ENOMEM;
1283 } while (pmd++, addr = next, addr != end);
1284 return 0;
1285}
1286
1287static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1288 unsigned long addr, unsigned long end,
1289 unsigned long pfn, pgprot_t prot)
1290{
1291 pud_t *pud;
1292 unsigned long next;
1293
1294 pfn -= addr >> PAGE_SHIFT;
1295 pud = pud_alloc(mm, pgd, addr);
1296 if (!pud)
1297 return -ENOMEM;
1298 do {
1299 next = pud_addr_end(addr, end);
1300 if (remap_pmd_range(mm, pud, addr, next,
1301 pfn + (addr >> PAGE_SHIFT), prot))
1302 return -ENOMEM;
1303 } while (pud++, addr = next, addr != end);
1304 return 0;
1305}
1306
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001307/**
1308 * remap_pfn_range - remap kernel memory to userspace
1309 * @vma: user vma to map to
1310 * @addr: target user address to start at
1311 * @pfn: physical address of kernel memory
1312 * @size: size of map area
1313 * @prot: page protection flags for this mapping
1314 *
1315 * Note: this is only safe if the mm semaphore is held when called.
1316 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1318 unsigned long pfn, unsigned long size, pgprot_t prot)
1319{
1320 pgd_t *pgd;
1321 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001322 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 struct mm_struct *mm = vma->vm_mm;
1324 int err;
1325
1326 /*
1327 * Physically remapped pages are special. Tell the
1328 * rest of the world about it:
1329 * VM_IO tells people not to look at these pages
1330 * (accesses can have side effects).
Hugh Dickins0b14c172005-11-21 21:32:15 -08001331 * VM_RESERVED is specified all over the place, because
1332 * in 2.4 it kept swapout's vma scan off this vma; but
1333 * in 2.6 the LRU scan won't even find its pages, so this
1334 * flag means no more than count its pages in reserved_vm,
1335 * and omit it from core dump, even when VM_IO turned off.
Linus Torvalds6aab3412005-11-28 14:34:23 -08001336 * VM_PFNMAP tells the core MM that the base pages are just
1337 * raw PFN mappings, and do not have a "struct page" associated
1338 * with them.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001339 *
1340 * There's a horrible special case to handle copy-on-write
1341 * behaviour that some programs depend on. We mark the "original"
1342 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 */
Linus Torvalds67121172005-12-11 20:38:17 -08001344 if (is_cow_mapping(vma->vm_flags)) {
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001345 if (addr != vma->vm_start || end != vma->vm_end)
Linus Torvalds7fc7e2e2005-12-11 19:57:52 -08001346 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001347 vma->vm_pgoff = pfn;
1348 }
1349
Linus Torvalds6aab3412005-11-28 14:34:23 -08001350 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351
1352 BUG_ON(addr >= end);
1353 pfn -= addr >> PAGE_SHIFT;
1354 pgd = pgd_offset(mm, addr);
1355 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 do {
1357 next = pgd_addr_end(addr, end);
1358 err = remap_pud_range(mm, pgd, addr, next,
1359 pfn + (addr >> PAGE_SHIFT), prot);
1360 if (err)
1361 break;
1362 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 return err;
1364}
1365EXPORT_SYMBOL(remap_pfn_range);
1366
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001367static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1368 unsigned long addr, unsigned long end,
1369 pte_fn_t fn, void *data)
1370{
1371 pte_t *pte;
1372 int err;
1373 struct page *pmd_page;
Borislav Petkov94909912007-05-06 14:49:17 -07001374 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001375
1376 pte = (mm == &init_mm) ?
1377 pte_alloc_kernel(pmd, addr) :
1378 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1379 if (!pte)
1380 return -ENOMEM;
1381
1382 BUG_ON(pmd_huge(*pmd));
1383
1384 pmd_page = pmd_page(*pmd);
1385
1386 do {
1387 err = fn(pte, pmd_page, addr, data);
1388 if (err)
1389 break;
1390 } while (pte++, addr += PAGE_SIZE, addr != end);
1391
1392 if (mm != &init_mm)
1393 pte_unmap_unlock(pte-1, ptl);
1394 return err;
1395}
1396
1397static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1398 unsigned long addr, unsigned long end,
1399 pte_fn_t fn, void *data)
1400{
1401 pmd_t *pmd;
1402 unsigned long next;
1403 int err;
1404
1405 pmd = pmd_alloc(mm, pud, addr);
1406 if (!pmd)
1407 return -ENOMEM;
1408 do {
1409 next = pmd_addr_end(addr, end);
1410 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1411 if (err)
1412 break;
1413 } while (pmd++, addr = next, addr != end);
1414 return err;
1415}
1416
1417static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1418 unsigned long addr, unsigned long end,
1419 pte_fn_t fn, void *data)
1420{
1421 pud_t *pud;
1422 unsigned long next;
1423 int err;
1424
1425 pud = pud_alloc(mm, pgd, addr);
1426 if (!pud)
1427 return -ENOMEM;
1428 do {
1429 next = pud_addr_end(addr, end);
1430 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1431 if (err)
1432 break;
1433 } while (pud++, addr = next, addr != end);
1434 return err;
1435}
1436
1437/*
1438 * Scan a region of virtual memory, filling in page tables as necessary
1439 * and calling a provided function on each leaf page table.
1440 */
1441int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1442 unsigned long size, pte_fn_t fn, void *data)
1443{
1444 pgd_t *pgd;
1445 unsigned long next;
1446 unsigned long end = addr + size;
1447 int err;
1448
1449 BUG_ON(addr >= end);
1450 pgd = pgd_offset(mm, addr);
1451 do {
1452 next = pgd_addr_end(addr, end);
1453 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1454 if (err)
1455 break;
1456 } while (pgd++, addr = next, addr != end);
1457 return err;
1458}
1459EXPORT_SYMBOL_GPL(apply_to_page_range);
1460
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001462 * handle_pte_fault chooses page fault handler according to an entry
1463 * which was read non-atomically. Before making any commitment, on
1464 * those architectures or configurations (e.g. i386 with PAE) which
1465 * might give a mix of unmatched parts, do_swap_page and do_file_page
1466 * must check under lock before unmapping the pte and proceeding
1467 * (but do_wp_page is only called after already making such a check;
1468 * and do_anonymous_page and do_no_page can safely check later on).
1469 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001470static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001471 pte_t *page_table, pte_t orig_pte)
1472{
1473 int same = 1;
1474#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1475 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001476 spinlock_t *ptl = pte_lockptr(mm, pmd);
1477 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001478 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001479 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001480 }
1481#endif
1482 pte_unmap(page_table);
1483 return same;
1484}
1485
1486/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1488 * servicing faults for write access. In the normal case, do always want
1489 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1490 * that do not have writing enabled, when used by access_process_vm.
1491 */
1492static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1493{
1494 if (likely(vma->vm_flags & VM_WRITE))
1495 pte = pte_mkwrite(pte);
1496 return pte;
1497}
1498
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001499static 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 -08001500{
1501 /*
1502 * If the source page was a PFN mapping, we don't have
1503 * a "struct page" for it. We do a best-effort copy by
1504 * just copying from the original user address. If that
1505 * fails, we just zero-fill it. Live with it.
1506 */
1507 if (unlikely(!src)) {
1508 void *kaddr = kmap_atomic(dst, KM_USER0);
Linus Torvalds5d2a2db2005-11-29 14:07:55 -08001509 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1510
1511 /*
1512 * This really shouldn't fail, because the page is there
1513 * in the page tables. But it might just be unreadable,
1514 * in which case we just give up and fill the result with
1515 * zeroes.
1516 */
1517 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Linus Torvalds6aab3412005-11-28 14:34:23 -08001518 memset(kaddr, 0, PAGE_SIZE);
1519 kunmap_atomic(kaddr, KM_USER0);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07001520 flush_dcache_page(dst);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001521 return;
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001522
Linus Torvalds6aab3412005-11-28 14:34:23 -08001523 }
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001524 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001525}
1526
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 * This routine handles present pages, when users try to write
1529 * to a shared page. It is done by copying the page to a new address
1530 * and decrementing the shared-page counter for the old page.
1531 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 * Note that this routine assumes that the protection checks have been
1533 * done by the caller (the low-level page fault routine in most cases).
1534 * Thus we can safely just mark it writable once we've done any necessary
1535 * COW.
1536 *
1537 * We also mark the page dirty at this point even though the page will
1538 * change only once the write actually happens. This avoids a few races,
1539 * and potentially makes it more efficient.
1540 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001541 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1542 * but allow concurrent faults), with pte both mapped and locked.
1543 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001545static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1546 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001547 spinlock_t *ptl, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548{
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001549 struct page *old_page, *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 pte_t entry;
Nick Piggin83c54072007-07-19 01:47:05 -07001551 int reuse = 0, ret = 0;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001552 int page_mkwrite = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001553 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554
Linus Torvalds6aab3412005-11-28 14:34:23 -08001555 old_page = vm_normal_page(vma, address, orig_pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001556 if (!old_page)
1557 goto gotten;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001559 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001560 * Take out anonymous pages first, anonymous shared vmas are
1561 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001562 */
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001563 if (PageAnon(old_page)) {
1564 if (!TestSetPageLocked(old_page)) {
1565 reuse = can_share_swap_page(old_page);
1566 unlock_page(old_page);
1567 }
1568 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001569 (VM_WRITE|VM_SHARED))) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001570 /*
1571 * Only catch write-faults on shared writable pages,
1572 * read-only shared pages can get COWed by
1573 * get_user_pages(.write=1, .force=1).
1574 */
David Howells9637a5e2006-06-23 02:03:43 -07001575 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
1576 /*
1577 * Notify the address space that the page is about to
1578 * become writable so that it can prohibit this or wait
1579 * for the page to get into an appropriate state.
1580 *
1581 * We do this without the lock held, so that it can
1582 * sleep if it needs to.
1583 */
1584 page_cache_get(old_page);
1585 pte_unmap_unlock(page_table, ptl);
1586
1587 if (vma->vm_ops->page_mkwrite(vma, old_page) < 0)
1588 goto unwritable_page;
1589
David Howells9637a5e2006-06-23 02:03:43 -07001590 /*
1591 * Since we dropped the lock we need to revalidate
1592 * the PTE as someone else may have changed it. If
1593 * they did, we just return, as we can count on the
1594 * MMU to tell us if they didn't also make it writable.
1595 */
1596 page_table = pte_offset_map_lock(mm, pmd, address,
1597 &ptl);
Hugh Dickinsc3704ce2007-02-10 01:43:00 -08001598 page_cache_release(old_page);
David Howells9637a5e2006-06-23 02:03:43 -07001599 if (!pte_same(*page_table, orig_pte))
1600 goto unlock;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001601
1602 page_mkwrite = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001604 dirty_page = old_page;
1605 get_page(dirty_page);
David Howells9637a5e2006-06-23 02:03:43 -07001606 reuse = 1;
David Howells9637a5e2006-06-23 02:03:43 -07001607 }
1608
1609 if (reuse) {
1610 flush_cache_page(vma, address, pte_pfn(orig_pte));
1611 entry = pte_mkyoung(orig_pte);
1612 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
KAMEZAWA Hiroyuki954ffcb2007-10-16 01:25:44 -07001613 if (ptep_set_access_flags(vma, address, page_table, entry,1))
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07001614 update_mmu_cache(vma, address, entry);
David Howells9637a5e2006-06-23 02:03:43 -07001615 ret |= VM_FAULT_WRITE;
1616 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618
1619 /*
1620 * Ok, we need to copy. Oh, well..
1621 */
Nick Pigginb5810032005-10-29 18:16:12 -07001622 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08001623gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001624 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625
1626 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07001627 goto oom;
Nick Piggin557ed1f2007-10-16 01:24:40 -07001628 VM_BUG_ON(old_page == ZERO_PAGE(0));
1629 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
1630 if (!new_page)
1631 goto oom;
1632 cow_user_page(new_page, old_page, address, vma);
Hugh Dickins65500d22005-10-29 18:15:59 -07001633
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 /*
1635 * Re-check the pte - we dropped the lock
1636 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001637 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07001638 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08001639 if (old_page) {
Nick Piggin7de6b802006-12-22 01:09:33 -08001640 page_remove_rmap(old_page, vma);
Hugh Dickins920fc352005-11-21 21:32:17 -08001641 if (!PageAnon(old_page)) {
1642 dec_mm_counter(mm, file_rss);
1643 inc_mm_counter(mm, anon_rss);
1644 }
1645 } else
Hugh Dickins42946212005-10-29 18:16:05 -07001646 inc_mm_counter(mm, anon_rss);
Ben Collinseca35132005-11-29 11:45:26 -08001647 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07001648 entry = mk_pte(new_page, vma->vm_page_prot);
1649 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Siddha, Suresh B4ce072f2006-09-29 01:58:42 -07001650 /*
1651 * Clear the pte entry and flush it first, before updating the
1652 * pte with the new entry. This will avoid a race condition
1653 * seen in the presence of one thread doing SMC and another
1654 * thread doing COW.
1655 */
1656 ptep_clear_flush(vma, address, page_table);
1657 set_pte_at(mm, address, page_table, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07001658 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 lru_cache_add_active(new_page);
Nick Piggin9617d952006-01-06 00:11:12 -08001660 page_add_new_anon_rmap(new_page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661
1662 /* Free the old page.. */
1663 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001664 ret |= VM_FAULT_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 }
Hugh Dickins920fc352005-11-21 21:32:17 -08001666 if (new_page)
1667 page_cache_release(new_page);
1668 if (old_page)
1669 page_cache_release(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001670unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001671 pte_unmap_unlock(page_table, ptl);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001672 if (dirty_page) {
Nick Piggin79352892007-07-19 01:47:22 -07001673 /*
1674 * Yes, Virginia, this is actually required to prevent a race
1675 * with clear_page_dirty_for_io() from clearing the page dirty
1676 * bit after it clear all dirty ptes, but before a racing
1677 * do_wp_page installs a dirty pte.
1678 *
1679 * do_no_page is protected similarly.
1680 */
1681 wait_on_page_locked(dirty_page);
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001682 set_page_dirty_balance(dirty_page, page_mkwrite);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001683 put_page(dirty_page);
1684 }
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001685 return ret;
Hugh Dickins65500d22005-10-29 18:15:59 -07001686oom:
Hugh Dickins920fc352005-11-21 21:32:17 -08001687 if (old_page)
1688 page_cache_release(old_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 return VM_FAULT_OOM;
David Howells9637a5e2006-06-23 02:03:43 -07001690
1691unwritable_page:
1692 page_cache_release(old_page);
1693 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694}
1695
1696/*
1697 * Helper functions for unmap_mapping_range().
1698 *
1699 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1700 *
1701 * We have to restart searching the prio_tree whenever we drop the lock,
1702 * since the iterator is only valid while the lock is held, and anyway
1703 * a later vma might be split and reinserted earlier while lock dropped.
1704 *
1705 * The list of nonlinear vmas could be handled more efficiently, using
1706 * a placeholder, but handle it in the same way until a need is shown.
1707 * It is important to search the prio_tree before nonlinear list: a vma
1708 * may become nonlinear and be shifted from prio_tree to nonlinear list
1709 * while the lock is dropped; but never shifted from list to prio_tree.
1710 *
1711 * In order to make forward progress despite restarting the search,
1712 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1713 * quickly skip it next time around. Since the prio_tree search only
1714 * shows us those vmas affected by unmapping the range in question, we
1715 * can't efficiently keep all vmas in step with mapping->truncate_count:
1716 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1717 * mapping->truncate_count and vma->vm_truncate_count are protected by
1718 * i_mmap_lock.
1719 *
1720 * In order to make forward progress despite repeatedly restarting some
Hugh Dickinsee39b372005-04-19 13:29:15 -07001721 * large vma, note the restart_addr from unmap_vmas when it breaks out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 * and restart from that address when we reach that vma again. It might
1723 * have been split or merged, shrunk or extended, but never shifted: so
1724 * restart_addr remains valid so long as it remains in the vma's range.
1725 * unmap_mapping_range forces truncate_count to leap over page-aligned
1726 * values so we can save vma's restart_addr in its truncate_count field.
1727 */
1728#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1729
1730static void reset_vma_truncate_counts(struct address_space *mapping)
1731{
1732 struct vm_area_struct *vma;
1733 struct prio_tree_iter iter;
1734
1735 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1736 vma->vm_truncate_count = 0;
1737 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1738 vma->vm_truncate_count = 0;
1739}
1740
1741static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1742 unsigned long start_addr, unsigned long end_addr,
1743 struct zap_details *details)
1744{
1745 unsigned long restart_addr;
1746 int need_break;
1747
Nick Piggind00806b2007-07-19 01:46:57 -07001748 /*
1749 * files that support invalidating or truncating portions of the
Nick Piggind0217ac2007-07-19 01:47:03 -07001750 * file from under mmaped areas must have their ->fault function
Nick Piggin83c54072007-07-19 01:47:05 -07001751 * return a locked page (and set VM_FAULT_LOCKED in the return).
1752 * This provides synchronisation against concurrent unmapping here.
Nick Piggind00806b2007-07-19 01:46:57 -07001753 */
Nick Piggind00806b2007-07-19 01:46:57 -07001754
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755again:
1756 restart_addr = vma->vm_truncate_count;
1757 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1758 start_addr = restart_addr;
1759 if (start_addr >= end_addr) {
1760 /* Top of vma has been split off since last time */
1761 vma->vm_truncate_count = details->truncate_count;
1762 return 0;
1763 }
1764 }
1765
Hugh Dickinsee39b372005-04-19 13:29:15 -07001766 restart_addr = zap_page_range(vma, start_addr,
1767 end_addr - start_addr, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 need_break = need_resched() ||
1769 need_lockbreak(details->i_mmap_lock);
1770
Hugh Dickinsee39b372005-04-19 13:29:15 -07001771 if (restart_addr >= end_addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 /* We have now completed this vma: mark it so */
1773 vma->vm_truncate_count = details->truncate_count;
1774 if (!need_break)
1775 return 0;
1776 } else {
1777 /* Note restart_addr in vma's truncate_count field */
Hugh Dickinsee39b372005-04-19 13:29:15 -07001778 vma->vm_truncate_count = restart_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 if (!need_break)
1780 goto again;
1781 }
1782
1783 spin_unlock(details->i_mmap_lock);
1784 cond_resched();
1785 spin_lock(details->i_mmap_lock);
1786 return -EINTR;
1787}
1788
1789static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1790 struct zap_details *details)
1791{
1792 struct vm_area_struct *vma;
1793 struct prio_tree_iter iter;
1794 pgoff_t vba, vea, zba, zea;
1795
1796restart:
1797 vma_prio_tree_foreach(vma, &iter, root,
1798 details->first_index, details->last_index) {
1799 /* Skip quickly over those we have already dealt with */
1800 if (vma->vm_truncate_count == details->truncate_count)
1801 continue;
1802
1803 vba = vma->vm_pgoff;
1804 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1805 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1806 zba = details->first_index;
1807 if (zba < vba)
1808 zba = vba;
1809 zea = details->last_index;
1810 if (zea > vea)
1811 zea = vea;
1812
1813 if (unmap_mapping_range_vma(vma,
1814 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1815 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1816 details) < 0)
1817 goto restart;
1818 }
1819}
1820
1821static inline void unmap_mapping_range_list(struct list_head *head,
1822 struct zap_details *details)
1823{
1824 struct vm_area_struct *vma;
1825
1826 /*
1827 * In nonlinear VMAs there is no correspondence between virtual address
1828 * offset and file offset. So we must perform an exhaustive search
1829 * across *all* the pages in each nonlinear VMA, not just the pages
1830 * whose virtual address lies outside the file truncation point.
1831 */
1832restart:
1833 list_for_each_entry(vma, head, shared.vm_set.list) {
1834 /* Skip quickly over those we have already dealt with */
1835 if (vma->vm_truncate_count == details->truncate_count)
1836 continue;
1837 details->nonlinear_vma = vma;
1838 if (unmap_mapping_range_vma(vma, vma->vm_start,
1839 vma->vm_end, details) < 0)
1840 goto restart;
1841 }
1842}
1843
1844/**
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001845 * 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 -07001846 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 * @holebegin: byte in first page to unmap, relative to the start of
1848 * the underlying file. This will be rounded down to a PAGE_SIZE
1849 * boundary. Note that this is different from vmtruncate(), which
1850 * must keep the partial page. In contrast, we must get rid of
1851 * partial pages.
1852 * @holelen: size of prospective hole in bytes. This will be rounded
1853 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1854 * end of the file.
1855 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1856 * but 0 when invalidating pagecache, don't throw away private data.
1857 */
1858void unmap_mapping_range(struct address_space *mapping,
1859 loff_t const holebegin, loff_t const holelen, int even_cows)
1860{
1861 struct zap_details details;
1862 pgoff_t hba = holebegin >> PAGE_SHIFT;
1863 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1864
1865 /* Check for overflow. */
1866 if (sizeof(holelen) > sizeof(hlen)) {
1867 long long holeend =
1868 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1869 if (holeend & ~(long long)ULONG_MAX)
1870 hlen = ULONG_MAX - hba + 1;
1871 }
1872
1873 details.check_mapping = even_cows? NULL: mapping;
1874 details.nonlinear_vma = NULL;
1875 details.first_index = hba;
1876 details.last_index = hba + hlen - 1;
1877 if (details.last_index < details.first_index)
1878 details.last_index = ULONG_MAX;
1879 details.i_mmap_lock = &mapping->i_mmap_lock;
1880
1881 spin_lock(&mapping->i_mmap_lock);
1882
Nick Piggind00806b2007-07-19 01:46:57 -07001883 /* Protect against endless unmapping loops */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 mapping->truncate_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 if (unlikely(is_restart_addr(mapping->truncate_count))) {
1886 if (mapping->truncate_count == 0)
1887 reset_vma_truncate_counts(mapping);
1888 mapping->truncate_count++;
1889 }
1890 details.truncate_count = mapping->truncate_count;
1891
1892 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
1893 unmap_mapping_range_tree(&mapping->i_mmap, &details);
1894 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
1895 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
1896 spin_unlock(&mapping->i_mmap_lock);
1897}
1898EXPORT_SYMBOL(unmap_mapping_range);
1899
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001900/**
1901 * vmtruncate - unmap mappings "freed" by truncate() syscall
1902 * @inode: inode of the file used
1903 * @offset: file offset to start truncating
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 *
1905 * NOTE! We have to be ready to update the memory sharing
1906 * between the file and the memory map for a potential last
1907 * incomplete page. Ugly, but necessary.
1908 */
1909int vmtruncate(struct inode * inode, loff_t offset)
1910{
1911 struct address_space *mapping = inode->i_mapping;
1912 unsigned long limit;
1913
1914 if (inode->i_size < offset)
1915 goto do_expand;
1916 /*
1917 * truncation of in-use swapfiles is disallowed - it would cause
1918 * subsequent swapout to scribble on the now-freed blocks.
1919 */
1920 if (IS_SWAPFILE(inode))
1921 goto out_busy;
1922 i_size_write(inode, offset);
Nick Piggind00806b2007-07-19 01:46:57 -07001923
1924 /*
1925 * unmap_mapping_range is called twice, first simply for efficiency
1926 * so that truncate_inode_pages does fewer single-page unmaps. However
1927 * after this first call, and before truncate_inode_pages finishes,
1928 * it is possible for private pages to be COWed, which remain after
1929 * truncate_inode_pages finishes, hence the second unmap_mapping_range
1930 * call must be made for correctness.
1931 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
1933 truncate_inode_pages(mapping, offset);
Nick Piggind00806b2007-07-19 01:46:57 -07001934 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 goto out_truncate;
1936
1937do_expand:
1938 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1939 if (limit != RLIM_INFINITY && offset > limit)
1940 goto out_sig;
1941 if (offset > inode->i_sb->s_maxbytes)
1942 goto out_big;
1943 i_size_write(inode, offset);
1944
1945out_truncate:
1946 if (inode->i_op && inode->i_op->truncate)
1947 inode->i_op->truncate(inode);
1948 return 0;
1949out_sig:
1950 send_sig(SIGXFSZ, current, 0);
1951out_big:
1952 return -EFBIG;
1953out_busy:
1954 return -ETXTBSY;
1955}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956EXPORT_SYMBOL(vmtruncate);
1957
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001958int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
1959{
1960 struct address_space *mapping = inode->i_mapping;
1961
1962 /*
1963 * If the underlying filesystem is not going to provide
1964 * a way to truncate a range of blocks (punch a hole) -
1965 * we should return failure right now.
1966 */
1967 if (!inode->i_op || !inode->i_op->truncate_range)
1968 return -ENOSYS;
1969
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001970 mutex_lock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001971 down_write(&inode->i_alloc_sem);
1972 unmap_mapping_range(mapping, offset, (end - offset), 1);
1973 truncate_inode_pages_range(mapping, offset, end);
Nick Piggind00806b2007-07-19 01:46:57 -07001974 unmap_mapping_range(mapping, offset, (end - offset), 1);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001975 inode->i_op->truncate_range(inode, offset, end);
1976 up_write(&inode->i_alloc_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001977 mutex_unlock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001978
1979 return 0;
1980}
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001981
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001982/**
1983 * swapin_readahead - swap in pages in hope we need them soon
1984 * @entry: swap entry of this memory
1985 * @addr: address to start
1986 * @vma: user vma this addresses belong to
1987 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 * Primitive swap readahead code. We simply read an aligned block of
1989 * (1 << page_cluster) entries in the swap area. This method is chosen
1990 * because it doesn't cost us any seek time. We also make sure to queue
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001991 * the 'original' request together with the readahead ones...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 *
1993 * This has been extended to use the NUMA policies from the mm triggering
1994 * the readahead.
1995 *
1996 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
1997 */
1998void swapin_readahead(swp_entry_t entry, unsigned long addr,struct vm_area_struct *vma)
1999{
2000#ifdef CONFIG_NUMA
2001 struct vm_area_struct *next_vma = vma ? vma->vm_next : NULL;
2002#endif
2003 int i, num;
2004 struct page *new_page;
2005 unsigned long offset;
2006
2007 /*
2008 * Get the number of handles we should do readahead io to.
2009 */
2010 num = valid_swaphandles(entry, &offset);
2011 for (i = 0; i < num; offset++, i++) {
2012 /* Ok, do the async read-ahead now */
2013 new_page = read_swap_cache_async(swp_entry(swp_type(entry),
2014 offset), vma, addr);
2015 if (!new_page)
2016 break;
2017 page_cache_release(new_page);
2018#ifdef CONFIG_NUMA
2019 /*
2020 * Find the next applicable VMA for the NUMA policy.
2021 */
2022 addr += PAGE_SIZE;
2023 if (addr == 0)
2024 vma = NULL;
2025 if (vma) {
2026 if (addr >= vma->vm_end) {
2027 vma = next_vma;
2028 next_vma = vma ? vma->vm_next : NULL;
2029 }
2030 if (vma && addr < vma->vm_start)
2031 vma = NULL;
2032 } else {
2033 if (next_vma && addr >= next_vma->vm_start) {
2034 vma = next_vma;
2035 next_vma = vma->vm_next;
2036 }
2037 }
2038#endif
2039 }
2040 lru_add_drain(); /* Push any new pages onto the LRU now */
2041}
2042
2043/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002044 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2045 * but allow concurrent faults), and pte mapped but not yet locked.
2046 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002048static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2049 unsigned long address, pte_t *page_table, pmd_t *pmd,
2050 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002052 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 struct page *page;
Hugh Dickins65500d22005-10-29 18:15:59 -07002054 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 pte_t pte;
Nick Piggin83c54072007-07-19 01:47:05 -07002056 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002058 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002059 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002060
2061 entry = pte_to_swp_entry(orig_pte);
Christoph Lameter06972122006-06-23 02:03:35 -07002062 if (is_migration_entry(entry)) {
2063 migration_entry_wait(mm, pmd, address);
2064 goto out;
2065 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002066 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 page = lookup_swap_cache(entry);
2068 if (!page) {
Ashwin Chaugule098fe652006-12-06 20:31:54 -08002069 grab_swap_token(); /* Contend for token _before_ read-in */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 swapin_readahead(entry, address, vma);
2071 page = read_swap_cache_async(entry, vma, address);
2072 if (!page) {
2073 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002074 * Back out if somebody else faulted in this pte
2075 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002077 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 if (likely(pte_same(*page_table, orig_pte)))
2079 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002080 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002081 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 }
2083
2084 /* Had to read the page from swap area: Major fault */
2085 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002086 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 }
2088
2089 mark_page_accessed(page);
2090 lock_page(page);
Balbir Singh20a10222007-11-14 17:00:33 -08002091 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092
2093 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002094 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002096 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002097 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002098 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002099
2100 if (unlikely(!PageUptodate(page))) {
2101 ret = VM_FAULT_SIGBUS;
2102 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 }
2104
2105 /* The page isn't present yet, go ahead with the fault. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106
Hugh Dickins42946212005-10-29 18:16:05 -07002107 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 pte = mk_pte(page, vma->vm_page_prot);
2109 if (write_access && can_share_swap_page(page)) {
2110 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2111 write_access = 0;
2112 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113
2114 flush_icache_page(vma, page);
2115 set_pte_at(mm, address, page_table, pte);
2116 page_add_anon_rmap(page, vma, address);
2117
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002118 swap_free(entry);
2119 if (vm_swap_full())
2120 remove_exclusive_swap_page(page);
2121 unlock_page(page);
2122
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 if (write_access) {
Nick Piggin83c54072007-07-19 01:47:05 -07002124 /* XXX: We could OR the do_wp_page code with this one? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 if (do_wp_page(mm, vma, address,
Nick Piggin83c54072007-07-19 01:47:05 -07002126 page_table, pmd, ptl, pte) & VM_FAULT_OOM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 ret = VM_FAULT_OOM;
2128 goto out;
2129 }
2130
2131 /* No need to invalidate - it was non-present before */
2132 update_mmu_cache(vma, address, pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07002133unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002134 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135out:
2136 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002137out_nomap:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002138 pte_unmap_unlock(page_table, ptl);
Kirill Korotaevb8107482005-05-16 21:53:50 -07002139 unlock_page(page);
2140 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002141 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142}
2143
2144/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002145 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2146 * but allow concurrent faults), and pte mapped but not yet locked.
2147 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002149static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2150 unsigned long address, pte_t *page_table, pmd_t *pmd,
2151 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002153 struct page *page;
2154 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156
Nick Piggin557ed1f2007-10-16 01:24:40 -07002157 /* Allocate our own private page. */
2158 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159
Nick Piggin557ed1f2007-10-16 01:24:40 -07002160 if (unlikely(anon_vma_prepare(vma)))
2161 goto oom;
2162 page = alloc_zeroed_user_highpage_movable(vma, address);
2163 if (!page)
2164 goto oom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165
Nick Piggin557ed1f2007-10-16 01:24:40 -07002166 entry = mk_pte(page, vma->vm_page_prot);
2167 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002168
Nick Piggin557ed1f2007-10-16 01:24:40 -07002169 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2170 if (!pte_none(*page_table))
2171 goto release;
2172 inc_mm_counter(mm, anon_rss);
2173 lru_cache_add_active(page);
2174 page_add_new_anon_rmap(page, vma, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002175 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176
2177 /* No need to invalidate - it was non-present before */
Hugh Dickins65500d22005-10-29 18:15:59 -07002178 update_mmu_cache(vma, address, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002179unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002180 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002181 return 0;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002182release:
2183 page_cache_release(page);
2184 goto unlock;
Hugh Dickins65500d22005-10-29 18:15:59 -07002185oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 return VM_FAULT_OOM;
2187}
2188
2189/*
Nick Piggin54cb8822007-07-19 01:46:59 -07002190 * __do_fault() tries to create a new page mapping. It aggressively
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 * tries to share with existing pages, but makes a separate copy if
Nick Piggin54cb8822007-07-19 01:46:59 -07002192 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2193 * the next page fault.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 *
2195 * As this is called only for pages that do not currently exist, we
2196 * do not need to flush old virtual caches or the TLB.
2197 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002198 * We enter with non-exclusive mmap_sem (to exclude vma changes,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002199 * but allow concurrent faults), and pte neither mapped nor locked.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002200 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 */
Nick Piggin54cb8822007-07-19 01:46:59 -07002202static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002203 unsigned long address, pmd_t *pmd,
Nick Piggin54cb8822007-07-19 01:46:59 -07002204 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205{
Hugh Dickins16abfa02007-10-04 16:56:06 +01002206 pte_t *page_table;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002207 spinlock_t *ptl;
Nick Piggind0217ac2007-07-19 01:47:03 -07002208 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210 int anon = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002211 struct page *dirty_page = NULL;
Nick Piggind0217ac2007-07-19 01:47:03 -07002212 struct vm_fault vmf;
2213 int ret;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002214 int page_mkwrite = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07002215
Nick Piggind0217ac2007-07-19 01:47:03 -07002216 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2217 vmf.pgoff = pgoff;
2218 vmf.flags = flags;
2219 vmf.page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220
Hugh Dickins325f04d2005-11-29 16:55:48 +00002221 BUG_ON(vma->vm_flags & VM_PFNMAP);
2222
Nick Piggin54cb8822007-07-19 01:46:59 -07002223 if (likely(vma->vm_ops->fault)) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002224 ret = vma->vm_ops->fault(vma, &vmf);
Nick Piggin83c54072007-07-19 01:47:05 -07002225 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2226 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07002227 } else {
2228 /* Legacy ->nopage path */
Nick Piggin83c54072007-07-19 01:47:05 -07002229 ret = 0;
Nick Piggind0217ac2007-07-19 01:47:03 -07002230 vmf.page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret);
Nick Piggin54cb8822007-07-19 01:46:59 -07002231 /* no page was available -- either SIGBUS or OOM */
Nick Piggind0217ac2007-07-19 01:47:03 -07002232 if (unlikely(vmf.page == NOPAGE_SIGBUS))
Nick Piggin54cb8822007-07-19 01:46:59 -07002233 return VM_FAULT_SIGBUS;
Nick Piggind0217ac2007-07-19 01:47:03 -07002234 else if (unlikely(vmf.page == NOPAGE_OOM))
Nick Piggin54cb8822007-07-19 01:46:59 -07002235 return VM_FAULT_OOM;
2236 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237
Nick Piggind00806b2007-07-19 01:46:57 -07002238 /*
Nick Piggind0217ac2007-07-19 01:47:03 -07002239 * For consistency in subsequent calls, make the faulted page always
Nick Piggind00806b2007-07-19 01:46:57 -07002240 * locked.
2241 */
Nick Piggin83c54072007-07-19 01:47:05 -07002242 if (unlikely(!(ret & VM_FAULT_LOCKED)))
Nick Piggind0217ac2007-07-19 01:47:03 -07002243 lock_page(vmf.page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002244 else
Nick Piggind0217ac2007-07-19 01:47:03 -07002245 VM_BUG_ON(!PageLocked(vmf.page));
Nick Piggind00806b2007-07-19 01:46:57 -07002246
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 /*
2248 * Should we do an early C-O-W break?
2249 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002250 page = vmf.page;
Nick Piggin54cb8822007-07-19 01:46:59 -07002251 if (flags & FAULT_FLAG_WRITE) {
David Howells9637a5e2006-06-23 02:03:43 -07002252 if (!(vma->vm_flags & VM_SHARED)) {
Nick Piggin54cb8822007-07-19 01:46:59 -07002253 anon = 1;
Nick Piggind00806b2007-07-19 01:46:57 -07002254 if (unlikely(anon_vma_prepare(vma))) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002255 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002256 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002257 }
Nick Piggin83c54072007-07-19 01:47:05 -07002258 page = alloc_page_vma(GFP_HIGHUSER_MOVABLE,
2259 vma, address);
Nick Piggind00806b2007-07-19 01:46:57 -07002260 if (!page) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002261 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002262 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002263 }
Nick Piggind0217ac2007-07-19 01:47:03 -07002264 copy_user_highpage(page, vmf.page, address, vma);
David Howells9637a5e2006-06-23 02:03:43 -07002265 } else {
Nick Piggin54cb8822007-07-19 01:46:59 -07002266 /*
2267 * If the page will be shareable, see if the backing
David Howells9637a5e2006-06-23 02:03:43 -07002268 * address space wants to know that the page is about
Nick Piggin54cb8822007-07-19 01:46:59 -07002269 * to become writable
2270 */
Mark Fasheh69676142007-07-19 01:47:00 -07002271 if (vma->vm_ops->page_mkwrite) {
2272 unlock_page(page);
2273 if (vma->vm_ops->page_mkwrite(vma, page) < 0) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002274 ret = VM_FAULT_SIGBUS;
2275 anon = 1; /* no anon but release vmf.page */
Mark Fasheh69676142007-07-19 01:47:00 -07002276 goto out_unlocked;
2277 }
2278 lock_page(page);
Nick Piggind0217ac2007-07-19 01:47:03 -07002279 /*
2280 * XXX: this is not quite right (racy vs
2281 * invalidate) to unlock and relock the page
2282 * like this, however a better fix requires
2283 * reworking page_mkwrite locking API, which
2284 * is better done later.
2285 */
2286 if (!page->mapping) {
Nick Piggin83c54072007-07-19 01:47:05 -07002287 ret = 0;
Nick Piggind0217ac2007-07-19 01:47:03 -07002288 anon = 1; /* no anon but release vmf.page */
2289 goto out;
2290 }
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002291 page_mkwrite = 1;
David Howells9637a5e2006-06-23 02:03:43 -07002292 }
2293 }
Nick Piggin54cb8822007-07-19 01:46:59 -07002294
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 }
2296
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002297 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298
2299 /*
2300 * This silly early PAGE_DIRTY setting removes a race
2301 * due to the bad i386 page protection. But it's valid
2302 * for other architectures too.
2303 *
2304 * Note that if write_access is true, we either now have
2305 * an exclusive copy of the page, or this is a shared mapping,
2306 * so we can make it writable and dirty to avoid having to
2307 * handle that later.
2308 */
2309 /* Only go through if we didn't race with anybody else... */
Nick Piggin54cb8822007-07-19 01:46:59 -07002310 if (likely(pte_same(*page_table, orig_pte))) {
Nick Piggind00806b2007-07-19 01:46:57 -07002311 flush_icache_page(vma, page);
2312 entry = mk_pte(page, vma->vm_page_prot);
Nick Piggin54cb8822007-07-19 01:46:59 -07002313 if (flags & FAULT_FLAG_WRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2315 set_pte_at(mm, address, page_table, entry);
2316 if (anon) {
Nick Piggind00806b2007-07-19 01:46:57 -07002317 inc_mm_counter(mm, anon_rss);
2318 lru_cache_add_active(page);
2319 page_add_new_anon_rmap(page, vma, address);
Hugh Dickinsf57e88a2005-11-21 21:32:19 -08002320 } else {
Hugh Dickins42946212005-10-29 18:16:05 -07002321 inc_mm_counter(mm, file_rss);
Nick Piggind00806b2007-07-19 01:46:57 -07002322 page_add_file_rmap(page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002323 if (flags & FAULT_FLAG_WRITE) {
Nick Piggind00806b2007-07-19 01:46:57 -07002324 dirty_page = page;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002325 get_page(dirty_page);
2326 }
Hugh Dickins42946212005-10-29 18:16:05 -07002327 }
Nick Piggind00806b2007-07-19 01:46:57 -07002328
2329 /* no need to invalidate: a not-present page won't be cached */
2330 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 } else {
Nick Piggind00806b2007-07-19 01:46:57 -07002332 if (anon)
2333 page_cache_release(page);
2334 else
Nick Piggin54cb8822007-07-19 01:46:59 -07002335 anon = 1; /* no anon but release faulted_page */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 }
2337
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002338 pte_unmap_unlock(page_table, ptl);
Nick Piggind00806b2007-07-19 01:46:57 -07002339
2340out:
Nick Piggind0217ac2007-07-19 01:47:03 -07002341 unlock_page(vmf.page);
Mark Fasheh69676142007-07-19 01:47:00 -07002342out_unlocked:
Nick Piggind00806b2007-07-19 01:46:57 -07002343 if (anon)
Nick Piggind0217ac2007-07-19 01:47:03 -07002344 page_cache_release(vmf.page);
Nick Piggind00806b2007-07-19 01:46:57 -07002345 else if (dirty_page) {
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002346 set_page_dirty_balance(dirty_page, page_mkwrite);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002347 put_page(dirty_page);
2348 }
Nick Piggind00806b2007-07-19 01:46:57 -07002349
Nick Piggin83c54072007-07-19 01:47:05 -07002350 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07002351}
Nick Piggind00806b2007-07-19 01:46:57 -07002352
Nick Piggin54cb8822007-07-19 01:46:59 -07002353static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2354 unsigned long address, pte_t *page_table, pmd_t *pmd,
2355 int write_access, pte_t orig_pte)
2356{
2357 pgoff_t pgoff = (((address & PAGE_MASK)
Dean Nelson0da7e012007-10-16 01:24:45 -07002358 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
Nick Piggin54cb8822007-07-19 01:46:59 -07002359 unsigned int flags = (write_access ? FAULT_FLAG_WRITE : 0);
2360
Hugh Dickins16abfa02007-10-04 16:56:06 +01002361 pte_unmap(page_table);
2362 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Nick Piggin54cb8822007-07-19 01:46:59 -07002363}
2364
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365
2366/*
Jes Sorensenf4b81802006-09-27 01:50:10 -07002367 * do_no_pfn() tries to create a new page mapping for a page without
2368 * a struct_page backing it
2369 *
2370 * As this is called only for pages that do not currently exist, we
2371 * do not need to flush old virtual caches or the TLB.
2372 *
2373 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2374 * but allow concurrent faults), and pte mapped but not yet locked.
2375 * We return with mmap_sem still held, but pte unmapped and unlocked.
2376 *
2377 * It is expected that the ->nopfn handler always returns the same pfn
2378 * for a given virtual mapping.
2379 *
2380 * Mark this `noinline' to prevent it from bloating the main pagefault code.
2381 */
2382static noinline int do_no_pfn(struct mm_struct *mm, struct vm_area_struct *vma,
2383 unsigned long address, pte_t *page_table, pmd_t *pmd,
2384 int write_access)
2385{
2386 spinlock_t *ptl;
2387 pte_t entry;
2388 unsigned long pfn;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002389
2390 pte_unmap(page_table);
2391 BUG_ON(!(vma->vm_flags & VM_PFNMAP));
2392 BUG_ON(is_cow_mapping(vma->vm_flags));
2393
2394 pfn = vma->vm_ops->nopfn(vma, address & PAGE_MASK);
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002395 if (unlikely(pfn == NOPFN_OOM))
Jes Sorensenf4b81802006-09-27 01:50:10 -07002396 return VM_FAULT_OOM;
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002397 else if (unlikely(pfn == NOPFN_SIGBUS))
Jes Sorensenf4b81802006-09-27 01:50:10 -07002398 return VM_FAULT_SIGBUS;
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002399 else if (unlikely(pfn == NOPFN_REFAULT))
Nick Piggin83c54072007-07-19 01:47:05 -07002400 return 0;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002401
2402 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2403
2404 /* Only go through if we didn't race with anybody else... */
2405 if (pte_none(*page_table)) {
2406 entry = pfn_pte(pfn, vma->vm_page_prot);
2407 if (write_access)
2408 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2409 set_pte_at(mm, address, page_table, entry);
2410 }
2411 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002412 return 0;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002413}
2414
2415/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 * Fault of a previously existing named mapping. Repopulate the pte
2417 * from the encoded file_pte if possible. This enables swappable
2418 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002419 *
2420 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2421 * but allow concurrent faults), and pte mapped but not yet locked.
2422 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002424static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins65500d22005-10-29 18:15:59 -07002425 unsigned long address, pte_t *page_table, pmd_t *pmd,
2426 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427{
Nick Piggind0217ac2007-07-19 01:47:03 -07002428 unsigned int flags = FAULT_FLAG_NONLINEAR |
2429 (write_access ? FAULT_FLAG_WRITE : 0);
Hugh Dickins65500d22005-10-29 18:15:59 -07002430 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002432 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Nick Piggin83c54072007-07-19 01:47:05 -07002433 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434
Nick Piggind0217ac2007-07-19 01:47:03 -07002435 if (unlikely(!(vma->vm_flags & VM_NONLINEAR) ||
2436 !(vma->vm_flags & VM_CAN_NONLINEAR))) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002437 /*
2438 * Page table corrupted: show pte and kill process.
2439 */
Nick Pigginb5810032005-10-29 18:16:12 -07002440 print_bad_pte(vma, orig_pte, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002441 return VM_FAULT_OOM;
2442 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002443
2444 pgoff = pte_to_pgoff(orig_pte);
Hugh Dickins16abfa02007-10-04 16:56:06 +01002445 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446}
2447
2448/*
2449 * These routines also need to handle stuff like marking pages dirty
2450 * and/or accessed for architectures that don't do it in hardware (most
2451 * RISC architectures). The early dirtying is also good on the i386.
2452 *
2453 * There is also a hook called "update_mmu_cache()" that architectures
2454 * with external mmu caches can use to update those (ie the Sparc or
2455 * PowerPC hashed page tables that act as extended TLBs).
2456 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07002457 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2458 * but allow concurrent faults), and pte mapped but not yet locked.
2459 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460 */
2461static inline int handle_pte_fault(struct mm_struct *mm,
Hugh Dickins65500d22005-10-29 18:15:59 -07002462 struct vm_area_struct *vma, unsigned long address,
2463 pte_t *pte, pmd_t *pmd, int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464{
2465 pte_t entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002466 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002468 entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002470 if (pte_none(entry)) {
Jes Sorensenf4b81802006-09-27 01:50:10 -07002471 if (vma->vm_ops) {
Nick Piggin54cb8822007-07-19 01:46:59 -07002472 if (vma->vm_ops->fault || vma->vm_ops->nopage)
2473 return do_linear_fault(mm, vma, address,
2474 pte, pmd, write_access, entry);
Jes Sorensenf4b81802006-09-27 01:50:10 -07002475 if (unlikely(vma->vm_ops->nopfn))
2476 return do_no_pfn(mm, vma, address, pte,
2477 pmd, write_access);
2478 }
2479 return do_anonymous_page(mm, vma, address,
2480 pte, pmd, write_access);
Hugh Dickins65500d22005-10-29 18:15:59 -07002481 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 if (pte_file(entry))
Nick Piggind0217ac2007-07-19 01:47:03 -07002483 return do_nonlinear_fault(mm, vma, address,
Hugh Dickins65500d22005-10-29 18:15:59 -07002484 pte, pmd, write_access, entry);
2485 return do_swap_page(mm, vma, address,
2486 pte, pmd, write_access, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487 }
2488
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002489 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002490 spin_lock(ptl);
2491 if (unlikely(!pte_same(*pte, entry)))
2492 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493 if (write_access) {
2494 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002495 return do_wp_page(mm, vma, address,
2496 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 entry = pte_mkdirty(entry);
2498 }
2499 entry = pte_mkyoung(entry);
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002500 if (ptep_set_access_flags(vma, address, pte, entry, write_access)) {
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002501 update_mmu_cache(vma, address, entry);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002502 } else {
2503 /*
2504 * This is needed only for protection faults but the arch code
2505 * is not yet telling us if this is a protection fault or not.
2506 * This still avoids useless tlb flushes for .text page faults
2507 * with threads.
2508 */
2509 if (write_access)
2510 flush_tlb_page(vma, address);
2511 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002512unlock:
2513 pte_unmap_unlock(pte, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002514 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515}
2516
2517/*
2518 * By the time we get here, we already hold the mm semaphore
2519 */
Nick Piggin83c54072007-07-19 01:47:05 -07002520int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521 unsigned long address, int write_access)
2522{
2523 pgd_t *pgd;
2524 pud_t *pud;
2525 pmd_t *pmd;
2526 pte_t *pte;
2527
2528 __set_current_state(TASK_RUNNING);
2529
Christoph Lameterf8891e52006-06-30 01:55:45 -07002530 count_vm_event(PGFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01002532 if (unlikely(is_vm_hugetlb_page(vma)))
2533 return hugetlb_fault(mm, vma, address, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536 pud = pud_alloc(mm, pgd, address);
2537 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002538 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539 pmd = pmd_alloc(mm, pud, address);
2540 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002541 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 pte = pte_alloc_map(mm, pmd, address);
2543 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002544 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545
Hugh Dickinsc74df322005-10-29 18:16:23 -07002546 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547}
2548
2549#ifndef __PAGETABLE_PUD_FOLDED
2550/*
2551 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002552 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002554int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002556 pud_t *new = pud_alloc_one(mm, address);
2557 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002558 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559
Hugh Dickins872fec12005-10-29 18:16:21 -07002560 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002561 if (pgd_present(*pgd)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562 pud_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002563 else
2564 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07002565 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002566 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567}
2568#endif /* __PAGETABLE_PUD_FOLDED */
2569
2570#ifndef __PAGETABLE_PMD_FOLDED
2571/*
2572 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002573 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002575int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002577 pmd_t *new = pmd_alloc_one(mm, address);
2578 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002579 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580
Hugh Dickins872fec12005-10-29 18:16:21 -07002581 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07002583 if (pud_present(*pud)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 pmd_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002585 else
2586 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07002588 if (pgd_present(*pud)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589 pmd_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002590 else
2591 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07002593 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002594 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595}
2596#endif /* __PAGETABLE_PMD_FOLDED */
2597
2598int make_pages_present(unsigned long addr, unsigned long end)
2599{
2600 int ret, len, write;
2601 struct vm_area_struct * vma;
2602
2603 vma = find_vma(current->mm, addr);
2604 if (!vma)
2605 return -1;
2606 write = (vma->vm_flags & VM_WRITE) != 0;
Eric Sesterhenn5bcb28b2006-03-26 18:30:52 +02002607 BUG_ON(addr >= end);
2608 BUG_ON(end > vma->vm_end);
Rolf Eike Beer68e116a2007-07-15 23:38:03 -07002609 len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610 ret = get_user_pages(current, current->mm, addr,
2611 len, write, 0, NULL, NULL);
2612 if (ret < 0)
2613 return ret;
2614 return ret == len ? 0 : -1;
2615}
2616
2617/*
2618 * Map a vmalloc()-space virtual address to the physical page.
2619 */
2620struct page * vmalloc_to_page(void * vmalloc_addr)
2621{
2622 unsigned long addr = (unsigned long) vmalloc_addr;
2623 struct page *page = NULL;
2624 pgd_t *pgd = pgd_offset_k(addr);
2625 pud_t *pud;
2626 pmd_t *pmd;
2627 pte_t *ptep, pte;
2628
2629 if (!pgd_none(*pgd)) {
2630 pud = pud_offset(pgd, addr);
2631 if (!pud_none(*pud)) {
2632 pmd = pmd_offset(pud, addr);
2633 if (!pmd_none(*pmd)) {
2634 ptep = pte_offset_map(pmd, addr);
2635 pte = *ptep;
2636 if (pte_present(pte))
2637 page = pte_page(pte);
2638 pte_unmap(ptep);
2639 }
2640 }
2641 }
2642 return page;
2643}
2644
2645EXPORT_SYMBOL(vmalloc_to_page);
2646
2647/*
2648 * Map a vmalloc()-space virtual address to the physical page frame number.
2649 */
2650unsigned long vmalloc_to_pfn(void * vmalloc_addr)
2651{
2652 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
2653}
2654
2655EXPORT_SYMBOL(vmalloc_to_pfn);
2656
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657#if !defined(__HAVE_ARCH_GATE_AREA)
2658
2659#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07002660static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661
2662static int __init gate_vma_init(void)
2663{
2664 gate_vma.vm_mm = NULL;
2665 gate_vma.vm_start = FIXADDR_USER_START;
2666 gate_vma.vm_end = FIXADDR_USER_END;
Roland McGrathb6558c42007-01-26 00:56:47 -08002667 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
2668 gate_vma.vm_page_prot = __P101;
Roland McGrathf47aef52007-01-26 00:56:49 -08002669 /*
2670 * Make sure the vDSO gets into every core dump.
2671 * Dumping its contents makes post-mortem fully interpretable later
2672 * without matching up the same kernel and hardware config to see
2673 * what PC values meant.
2674 */
2675 gate_vma.vm_flags |= VM_ALWAYSDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 return 0;
2677}
2678__initcall(gate_vma_init);
2679#endif
2680
2681struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2682{
2683#ifdef AT_SYSINFO_EHDR
2684 return &gate_vma;
2685#else
2686 return NULL;
2687#endif
2688}
2689
2690int in_gate_area_no_task(unsigned long addr)
2691{
2692#ifdef AT_SYSINFO_EHDR
2693 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2694 return 1;
2695#endif
2696 return 0;
2697}
2698
2699#endif /* __HAVE_ARCH_GATE_AREA */
David Howells0ec76a12006-09-27 01:50:15 -07002700
2701/*
2702 * Access another process' address space.
2703 * Source/target buffer must be kernel space,
2704 * Do not walk the page table directly, use get_user_pages
2705 */
2706int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2707{
2708 struct mm_struct *mm;
2709 struct vm_area_struct *vma;
2710 struct page *page;
2711 void *old_buf = buf;
2712
2713 mm = get_task_mm(tsk);
2714 if (!mm)
2715 return 0;
2716
2717 down_read(&mm->mmap_sem);
Simon Arlott183ff222007-10-20 01:27:18 +02002718 /* ignore errors, just check how much was successfully transferred */
David Howells0ec76a12006-09-27 01:50:15 -07002719 while (len) {
2720 int bytes, ret, offset;
2721 void *maddr;
2722
2723 ret = get_user_pages(tsk, mm, addr, 1,
2724 write, 1, &page, &vma);
2725 if (ret <= 0)
2726 break;
2727
2728 bytes = len;
2729 offset = addr & (PAGE_SIZE-1);
2730 if (bytes > PAGE_SIZE-offset)
2731 bytes = PAGE_SIZE-offset;
2732
2733 maddr = kmap(page);
2734 if (write) {
2735 copy_to_user_page(vma, page, addr,
2736 maddr + offset, buf, bytes);
2737 set_page_dirty_lock(page);
2738 } else {
2739 copy_from_user_page(vma, page, addr,
2740 buf, maddr + offset, bytes);
2741 }
2742 kunmap(page);
2743 page_cache_release(page);
2744 len -= bytes;
2745 buf += bytes;
2746 addr += bytes;
2747 }
2748 up_read(&mm->mmap_sem);
2749 mmput(mm);
2750
2751 return buf - old_buf;
2752}