blob: f82b359b27452dd1f7e682197479fa1391d19a46 [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
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700263 if (!(*tlb)->fullmm)
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700264 flush_tlb_pgtables((*tlb)->mm, start, end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700265}
266
267void free_pgtables(struct mmu_gather **tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700268 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700269{
270 while (vma) {
271 struct vm_area_struct *next = vma->vm_next;
272 unsigned long addr = vma->vm_start;
273
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700274 /*
275 * Hide vma from rmap and vmtruncate before freeing pgtables
276 */
277 anon_vma_unlink(vma);
278 unlink_file_vma(vma);
279
David Gibson9da61ae2006-03-22 00:08:57 -0800280 if (is_vm_hugetlb_page(vma)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700281 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700282 floor, next? next->vm_start: ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700283 } else {
284 /*
285 * Optimization: gather nearby vmas into one call down
286 */
287 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
David Gibson48669202006-03-22 00:08:58 -0800288 && !is_vm_hugetlb_page(next)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700289 vma = next;
290 next = vma->vm_next;
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700291 anon_vma_unlink(vma);
292 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700293 }
294 free_pgd_range(tlb, addr, vma->vm_end,
295 floor, next? next->vm_start: ceiling);
296 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700297 vma = next;
298 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299}
300
Hugh Dickins1bb36302005-10-29 18:16:22 -0700301int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302{
Hugh Dickinsc74df322005-10-29 18:16:23 -0700303 struct page *new = pte_alloc_one(mm, address);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700304 if (!new)
305 return -ENOMEM;
306
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700307 pte_lock_init(new);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700308 spin_lock(&mm->page_table_lock);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700309 if (pmd_present(*pmd)) { /* Another has populated it */
310 pte_lock_deinit(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700311 pte_free(new);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700312 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 mm->nr_ptes++;
Christoph Lameterdf849a12006-06-30 01:55:38 -0700314 inc_zone_page_state(new, NR_PAGETABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 pmd_populate(mm, pmd, new);
316 }
Hugh Dickinsc74df322005-10-29 18:16:23 -0700317 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700318 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319}
320
Hugh Dickins1bb36302005-10-29 18:16:22 -0700321int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700323 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
324 if (!new)
325 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326
Hugh Dickins1bb36302005-10-29 18:16:22 -0700327 spin_lock(&init_mm.page_table_lock);
328 if (pmd_present(*pmd)) /* Another has populated it */
329 pte_free_kernel(new);
330 else
331 pmd_populate_kernel(&init_mm, pmd, new);
332 spin_unlock(&init_mm.page_table_lock);
333 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334}
335
Hugh Dickinsae859762005-10-29 18:16:05 -0700336static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
337{
338 if (file_rss)
339 add_mm_counter(mm, file_rss, file_rss);
340 if (anon_rss)
341 add_mm_counter(mm, anon_rss, anon_rss);
342}
343
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800345 * This function is called to print an error when a bad pte
346 * is found. For example, we might have a PFN-mapped pte in
347 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700348 *
349 * The calling function must still handle the error.
350 */
351void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr)
352{
353 printk(KERN_ERR "Bad pte = %08llx, process = %s, "
354 "vm_flags = %lx, vaddr = %lx\n",
355 (long long)pte_val(pte),
356 (vma->vm_mm == current->mm ? current->comm : "???"),
357 vma->vm_flags, vaddr);
358 dump_stack();
359}
360
Linus Torvalds67121172005-12-11 20:38:17 -0800361static inline int is_cow_mapping(unsigned int flags)
362{
363 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
364}
365
Nick Pigginb5810032005-10-29 18:16:12 -0700366/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800367 * This function gets the "struct page" associated with a pte.
368 *
369 * NOTE! Some mappings do not have "struct pages". A raw PFN mapping
370 * will have each page table entry just pointing to a raw page frame
371 * number, and as far as the VM layer is concerned, those do not have
372 * pages associated with them - even if the PFN might point to memory
373 * that otherwise is perfectly fine and has a "struct page".
374 *
375 * The way we recognize those mappings is through the rules set up
376 * by "remap_pfn_range()": the vma will have the VM_PFNMAP bit set,
377 * and the vm_pgoff will point to the first PFN mapped: thus every
378 * page that is a raw mapping will always honor the rule
379 *
380 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
381 *
382 * and if that isn't true, the page has been COW'ed (in which case it
383 * _does_ have a "struct page" associated with it even if it is in a
384 * VM_PFNMAP range).
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800385 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800386struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800387{
Linus Torvalds6aab3412005-11-28 14:34:23 -0800388 unsigned long pfn = pte_pfn(pte);
389
Nick Pigginb7ab7952006-03-22 00:08:42 -0800390 if (unlikely(vma->vm_flags & VM_PFNMAP)) {
Linus Torvalds6aab3412005-11-28 14:34:23 -0800391 unsigned long off = (addr - vma->vm_start) >> PAGE_SHIFT;
392 if (pfn == vma->vm_pgoff + off)
393 return NULL;
Linus Torvalds67121172005-12-11 20:38:17 -0800394 if (!is_cow_mapping(vma->vm_flags))
Linus Torvaldsfb155c12005-12-11 19:46:02 -0800395 return NULL;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800396 }
397
Nick Piggin315ab192006-03-25 16:20:22 +0100398 /*
399 * Add some anal sanity checks for now. Eventually,
400 * we should just do "return pfn_to_page(pfn)", but
401 * in the meantime we check that we get a valid pfn,
402 * and that the resulting page looks ok.
403 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800404 if (unlikely(!pfn_valid(pfn))) {
405 print_bad_pte(vma, pte, addr);
406 return NULL;
407 }
408
409 /*
410 * NOTE! We still have PageReserved() pages in the page
411 * tables.
412 *
413 * The PAGE_ZERO() pages and various VDSO mappings can
414 * cause them to exist.
415 */
416 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800417}
418
419/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 * copy one vm_area from one task to the other. Assumes the page tables
421 * already present in the new task to be cleared in the whole range
422 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 */
424
Hugh Dickins8c103762005-10-29 18:16:13 -0700425static inline void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700427 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700428 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429{
Nick Pigginb5810032005-10-29 18:16:12 -0700430 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 pte_t pte = *src_pte;
432 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433
434 /* pte contains position in swap or file, so copy. */
435 if (unlikely(!pte_present(pte))) {
436 if (!pte_file(pte)) {
Christoph Lameter06972122006-06-23 02:03:35 -0700437 swp_entry_t entry = pte_to_swp_entry(pte);
438
439 swap_duplicate(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 /* make sure dst_mm is on swapoff's mmlist. */
441 if (unlikely(list_empty(&dst_mm->mmlist))) {
442 spin_lock(&mmlist_lock);
Hugh Dickinsf412ac02005-10-29 18:16:41 -0700443 if (list_empty(&dst_mm->mmlist))
444 list_add(&dst_mm->mmlist,
445 &src_mm->mmlist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 spin_unlock(&mmlist_lock);
447 }
Christoph Lameter06972122006-06-23 02:03:35 -0700448 if (is_write_migration_entry(entry) &&
449 is_cow_mapping(vm_flags)) {
450 /*
451 * COW mappings require pages in both parent
452 * and child to be set to read.
453 */
454 make_migration_entry_read(&entry);
455 pte = swp_entry_to_pte(entry);
456 set_pte_at(src_mm, addr, src_pte, pte);
457 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700459 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 }
461
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 /*
463 * If it's a COW mapping, write protect it both
464 * in the parent and the child
465 */
Linus Torvalds67121172005-12-11 20:38:17 -0800466 if (is_cow_mapping(vm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 ptep_set_wrprotect(src_mm, addr, src_pte);
Zachary Amsden3dc90792006-09-30 23:29:30 -0700468 pte = pte_wrprotect(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 }
470
471 /*
472 * If it's a shared mapping, mark it clean in
473 * the child
474 */
475 if (vm_flags & VM_SHARED)
476 pte = pte_mkclean(pte);
477 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800478
479 page = vm_normal_page(vma, addr, pte);
480 if (page) {
481 get_page(page);
Nick Pigginc97a9e12007-05-16 22:11:21 -0700482 page_dup_rmap(page, vma, addr);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800483 rss[!!PageAnon(page)]++;
484 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700485
486out_set_pte:
487 set_pte_at(dst_mm, addr, dst_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488}
489
490static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
491 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
492 unsigned long addr, unsigned long end)
493{
494 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700495 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -0700496 int progress = 0;
Hugh Dickins8c103762005-10-29 18:16:13 -0700497 int rss[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498
499again:
Hugh Dickinsae859762005-10-29 18:16:05 -0700500 rss[1] = rss[0] = 0;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700501 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 if (!dst_pte)
503 return -ENOMEM;
504 src_pte = pte_offset_map_nested(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700505 src_ptl = pte_lockptr(src_mm, src_pmd);
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700506 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700507 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 do {
510 /*
511 * We are holding two locks at this point - either of them
512 * could generate latencies in another task on another CPU.
513 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700514 if (progress >= 32) {
515 progress = 0;
516 if (need_resched() ||
Hugh Dickinsc74df322005-10-29 18:16:23 -0700517 need_lockbreak(src_ptl) ||
518 need_lockbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700519 break;
520 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 if (pte_none(*src_pte)) {
522 progress++;
523 continue;
524 }
Hugh Dickins8c103762005-10-29 18:16:13 -0700525 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 progress += 8;
527 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700529 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -0700530 spin_unlock(src_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 pte_unmap_nested(src_pte - 1);
Hugh Dickinsae859762005-10-29 18:16:05 -0700532 add_mm_rss(dst_mm, rss[0], rss[1]);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700533 pte_unmap_unlock(dst_pte - 1, dst_ptl);
534 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 if (addr != end)
536 goto again;
537 return 0;
538}
539
540static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
541 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
542 unsigned long addr, unsigned long end)
543{
544 pmd_t *src_pmd, *dst_pmd;
545 unsigned long next;
546
547 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
548 if (!dst_pmd)
549 return -ENOMEM;
550 src_pmd = pmd_offset(src_pud, addr);
551 do {
552 next = pmd_addr_end(addr, end);
553 if (pmd_none_or_clear_bad(src_pmd))
554 continue;
555 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
556 vma, addr, next))
557 return -ENOMEM;
558 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
559 return 0;
560}
561
562static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
563 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
564 unsigned long addr, unsigned long end)
565{
566 pud_t *src_pud, *dst_pud;
567 unsigned long next;
568
569 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
570 if (!dst_pud)
571 return -ENOMEM;
572 src_pud = pud_offset(src_pgd, addr);
573 do {
574 next = pud_addr_end(addr, end);
575 if (pud_none_or_clear_bad(src_pud))
576 continue;
577 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
578 vma, addr, next))
579 return -ENOMEM;
580 } while (dst_pud++, src_pud++, addr = next, addr != end);
581 return 0;
582}
583
584int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
585 struct vm_area_struct *vma)
586{
587 pgd_t *src_pgd, *dst_pgd;
588 unsigned long next;
589 unsigned long addr = vma->vm_start;
590 unsigned long end = vma->vm_end;
591
Nick Piggind9928952005-08-28 16:49:11 +1000592 /*
593 * Don't copy ptes where a page fault will fill them correctly.
594 * Fork becomes much lighter when there are big shared or private
595 * readonly mappings. The tradeoff is that copy_page_range is more
596 * efficient than faulting.
597 */
Linus Torvalds4d7672b2005-12-16 10:21:23 -0800598 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
Nick Piggind9928952005-08-28 16:49:11 +1000599 if (!vma->anon_vma)
600 return 0;
601 }
602
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 if (is_vm_hugetlb_page(vma))
604 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
605
606 dst_pgd = pgd_offset(dst_mm, addr);
607 src_pgd = pgd_offset(src_mm, addr);
608 do {
609 next = pgd_addr_end(addr, end);
610 if (pgd_none_or_clear_bad(src_pgd))
611 continue;
612 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
613 vma, addr, next))
614 return -ENOMEM;
615 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
616 return 0;
617}
618
Robin Holt51c6f662005-11-13 16:06:42 -0800619static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700620 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800622 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623{
Nick Pigginb5810032005-10-29 18:16:12 -0700624 struct mm_struct *mm = tlb->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 pte_t *pte;
Hugh Dickins508034a2005-10-29 18:16:30 -0700626 spinlock_t *ptl;
Hugh Dickinsae859762005-10-29 18:16:05 -0700627 int file_rss = 0;
628 int anon_rss = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629
Hugh Dickins508034a2005-10-29 18:16:30 -0700630 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700631 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 do {
633 pte_t ptent = *pte;
Robin Holt51c6f662005-11-13 16:06:42 -0800634 if (pte_none(ptent)) {
635 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800637 }
Hugh Dickins6f5e6b92006-03-16 23:04:09 -0800638
639 (*zap_work) -= PAGE_SIZE;
640
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800642 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -0800643
Linus Torvalds6aab3412005-11-28 14:34:23 -0800644 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 if (unlikely(details) && page) {
646 /*
647 * unmap_shared_mapping_pages() wants to
648 * invalidate cache without truncating:
649 * unmap shared but keep private pages.
650 */
651 if (details->check_mapping &&
652 details->check_mapping != page->mapping)
653 continue;
654 /*
655 * Each page->index must be checked when
656 * invalidating or truncating nonlinear.
657 */
658 if (details->nonlinear_vma &&
659 (page->index < details->first_index ||
660 page->index > details->last_index))
661 continue;
662 }
Nick Pigginb5810032005-10-29 18:16:12 -0700663 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -0700664 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 tlb_remove_tlb_entry(tlb, pte, addr);
666 if (unlikely(!page))
667 continue;
668 if (unlikely(details) && details->nonlinear_vma
669 && linear_page_index(details->nonlinear_vma,
670 addr) != page->index)
Nick Pigginb5810032005-10-29 18:16:12 -0700671 set_pte_at(mm, addr, pte,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 pgoff_to_pte(page->index));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 if (PageAnon(page))
Hugh Dickins86d912f2005-10-29 18:16:14 -0700674 anon_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700675 else {
676 if (pte_dirty(ptent))
677 set_page_dirty(page);
678 if (pte_young(ptent))
Ken Chendaa88c82007-02-10 01:43:18 -0800679 SetPageReferenced(page);
Hugh Dickins86d912f2005-10-29 18:16:14 -0700680 file_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700681 }
Nick Piggin7de6b802006-12-22 01:09:33 -0800682 page_remove_rmap(page, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 tlb_remove_page(tlb, page);
684 continue;
685 }
686 /*
687 * If details->check_mapping, we leave swap entries;
688 * if details->nonlinear_vma, we leave file entries.
689 */
690 if (unlikely(details))
691 continue;
692 if (!pte_file(ptent))
693 free_swap_and_cache(pte_to_swp_entry(ptent));
Zachary Amsden9888a1c2006-09-30 23:29:31 -0700694 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
Robin Holt51c6f662005-11-13 16:06:42 -0800695 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
Hugh Dickinsae859762005-10-29 18:16:05 -0700696
Hugh Dickins86d912f2005-10-29 18:16:14 -0700697 add_mm_rss(mm, file_rss, anon_rss);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700698 arch_leave_lazy_mmu_mode();
Hugh Dickins508034a2005-10-29 18:16:30 -0700699 pte_unmap_unlock(pte - 1, ptl);
Robin Holt51c6f662005-11-13 16:06:42 -0800700
701 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702}
703
Robin Holt51c6f662005-11-13 16:06:42 -0800704static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700705 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800707 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708{
709 pmd_t *pmd;
710 unsigned long next;
711
712 pmd = pmd_offset(pud, addr);
713 do {
714 next = pmd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800715 if (pmd_none_or_clear_bad(pmd)) {
716 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800718 }
719 next = zap_pte_range(tlb, vma, pmd, addr, next,
720 zap_work, details);
721 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
722
723 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724}
725
Robin Holt51c6f662005-11-13 16:06:42 -0800726static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700727 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800729 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730{
731 pud_t *pud;
732 unsigned long next;
733
734 pud = pud_offset(pgd, addr);
735 do {
736 next = pud_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800737 if (pud_none_or_clear_bad(pud)) {
738 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800740 }
741 next = zap_pmd_range(tlb, vma, pud, addr, next,
742 zap_work, details);
743 } while (pud++, addr = next, (addr != end && *zap_work > 0));
744
745 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746}
747
Robin Holt51c6f662005-11-13 16:06:42 -0800748static unsigned long unmap_page_range(struct mmu_gather *tlb,
749 struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800751 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752{
753 pgd_t *pgd;
754 unsigned long next;
755
756 if (details && !details->check_mapping && !details->nonlinear_vma)
757 details = NULL;
758
759 BUG_ON(addr >= end);
760 tlb_start_vma(tlb, vma);
761 pgd = pgd_offset(vma->vm_mm, addr);
762 do {
763 next = pgd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800764 if (pgd_none_or_clear_bad(pgd)) {
765 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800767 }
768 next = zap_pud_range(tlb, vma, pgd, addr, next,
769 zap_work, details);
770 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 tlb_end_vma(tlb, vma);
Robin Holt51c6f662005-11-13 16:06:42 -0800772
773 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774}
775
776#ifdef CONFIG_PREEMPT
777# define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
778#else
779/* No preempt: go for improved straight-line efficiency */
780# define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
781#endif
782
783/**
784 * unmap_vmas - unmap a range of memory covered by a list of vma's
785 * @tlbp: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 * @vma: the starting vma
787 * @start_addr: virtual address at which to start unmapping
788 * @end_addr: virtual address at which to end unmapping
789 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
790 * @details: details of nonlinear truncation or shared cache invalidation
791 *
Hugh Dickinsee39b372005-04-19 13:29:15 -0700792 * Returns the end address of the unmapping (restart addr if interrupted).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700794 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700796 * We aim to not hold locks for too long (for scheduling latency reasons).
797 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 * return the ending mmu_gather to the caller.
799 *
800 * Only addresses between `start' and `end' will be unmapped.
801 *
802 * The VMA list must be sorted in ascending virtual address order.
803 *
804 * unmap_vmas() assumes that the caller will flush the whole unmapped address
805 * range after unmap_vmas() returns. So the only responsibility here is to
806 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
807 * drops the lock and schedules.
808 */
Hugh Dickins508034a2005-10-29 18:16:30 -0700809unsigned long unmap_vmas(struct mmu_gather **tlbp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 struct vm_area_struct *vma, unsigned long start_addr,
811 unsigned long end_addr, unsigned long *nr_accounted,
812 struct zap_details *details)
813{
Robin Holt51c6f662005-11-13 16:06:42 -0800814 long zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
816 int tlb_start_valid = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -0700817 unsigned long start = start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700819 int fullmm = (*tlbp)->fullmm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
821 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 unsigned long end;
823
824 start = max(vma->vm_start, start_addr);
825 if (start >= vma->vm_end)
826 continue;
827 end = min(vma->vm_end, end_addr);
828 if (end <= vma->vm_start)
829 continue;
830
831 if (vma->vm_flags & VM_ACCOUNT)
832 *nr_accounted += (end - start) >> PAGE_SHIFT;
833
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 while (start != end) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 if (!tlb_start_valid) {
836 tlb_start = start;
837 tlb_start_valid = 1;
838 }
839
Robin Holt51c6f662005-11-13 16:06:42 -0800840 if (unlikely(is_vm_hugetlb_page(vma))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 unmap_hugepage_range(vma, start, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800842 zap_work -= (end - start) /
843 (HPAGE_SIZE / PAGE_SIZE);
844 start = end;
845 } else
846 start = unmap_page_range(*tlbp, vma,
847 start, end, &zap_work, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848
Robin Holt51c6f662005-11-13 16:06:42 -0800849 if (zap_work > 0) {
850 BUG_ON(start != end);
851 break;
852 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853
854 tlb_finish_mmu(*tlbp, tlb_start, start);
855
856 if (need_resched() ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 (i_mmap_lock && need_lockbreak(i_mmap_lock))) {
858 if (i_mmap_lock) {
Hugh Dickins508034a2005-10-29 18:16:30 -0700859 *tlbp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 goto out;
861 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 }
864
Hugh Dickins508034a2005-10-29 18:16:30 -0700865 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 tlb_start_valid = 0;
Robin Holt51c6f662005-11-13 16:06:42 -0800867 zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 }
869 }
870out:
Hugh Dickinsee39b372005-04-19 13:29:15 -0700871 return start; /* which is now the end (or restart) address */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872}
873
874/**
875 * zap_page_range - remove user pages in a given range
876 * @vma: vm_area_struct holding the applicable pages
877 * @address: starting address of pages to zap
878 * @size: number of bytes to zap
879 * @details: details of nonlinear truncation or shared cache invalidation
880 */
Hugh Dickinsee39b372005-04-19 13:29:15 -0700881unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 unsigned long size, struct zap_details *details)
883{
884 struct mm_struct *mm = vma->vm_mm;
885 struct mmu_gather *tlb;
886 unsigned long end = address + size;
887 unsigned long nr_accounted = 0;
888
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -0700891 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -0700892 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
893 if (tlb)
894 tlb_finish_mmu(tlb, address, end);
Hugh Dickinsee39b372005-04-19 13:29:15 -0700895 return end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896}
897
898/*
899 * Do a quick page-table lookup for a single page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800901struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700902 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903{
904 pgd_t *pgd;
905 pud_t *pud;
906 pmd_t *pmd;
907 pte_t *ptep, pte;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700908 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 struct page *page;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800910 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700912 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
913 if (!IS_ERR(page)) {
914 BUG_ON(flags & FOLL_GET);
915 goto out;
916 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700918 page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 pgd = pgd_offset(mm, address);
920 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700921 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922
923 pud = pud_offset(pgd, address);
924 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700925 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926
927 pmd = pmd_offset(pud, address);
928 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700929 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700931 if (pmd_huge(*pmd)) {
932 BUG_ON(flags & FOLL_GET);
933 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
934 goto out;
935 }
936
937 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 if (!ptep)
939 goto out;
940
941 pte = *ptep;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700942 if (!pte_present(pte))
943 goto unlock;
944 if ((flags & FOLL_WRITE) && !pte_write(pte))
945 goto unlock;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800946 page = vm_normal_page(vma, address, pte);
947 if (unlikely(!page))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700948 goto unlock;
949
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700950 if (flags & FOLL_GET)
951 get_page(page);
952 if (flags & FOLL_TOUCH) {
953 if ((flags & FOLL_WRITE) &&
954 !pte_dirty(pte) && !PageDirty(page))
955 set_page_dirty(page);
956 mark_page_accessed(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700958unlock:
959 pte_unmap_unlock(ptep, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960out:
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700961 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700963no_page_table:
964 /*
965 * When core dumping an enormous anonymous area that nobody
966 * has touched so far, we don't want to allocate page tables.
967 */
968 if (flags & FOLL_ANON) {
969 page = ZERO_PAGE(address);
970 if (flags & FOLL_GET)
971 get_page(page);
972 BUG_ON(flags & FOLL_WRITE);
973 }
974 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975}
976
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
978 unsigned long start, int len, int write, int force,
979 struct page **pages, struct vm_area_struct **vmas)
980{
981 int i;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700982 unsigned int vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983
984 /*
985 * Require read or write permissions.
986 * If 'force' is set, we only require the "MAY" flags.
987 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700988 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
989 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 i = 0;
991
992 do {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700993 struct vm_area_struct *vma;
994 unsigned int foll_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995
996 vma = find_extend_vma(mm, start);
997 if (!vma && in_gate_area(tsk, start)) {
998 unsigned long pg = start & PAGE_MASK;
999 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
1000 pgd_t *pgd;
1001 pud_t *pud;
1002 pmd_t *pmd;
1003 pte_t *pte;
1004 if (write) /* user gate pages are read-only */
1005 return i ? : -EFAULT;
1006 if (pg > TASK_SIZE)
1007 pgd = pgd_offset_k(pg);
1008 else
1009 pgd = pgd_offset_gate(mm, pg);
1010 BUG_ON(pgd_none(*pgd));
1011 pud = pud_offset(pgd, pg);
1012 BUG_ON(pud_none(*pud));
1013 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001014 if (pmd_none(*pmd))
1015 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001017 if (pte_none(*pte)) {
1018 pte_unmap(pte);
1019 return i ? : -EFAULT;
1020 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 if (pages) {
Nick Pigginfa2a4552005-11-29 18:43:17 +11001022 struct page *page = vm_normal_page(gate_vma, start, *pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001023 pages[i] = page;
1024 if (page)
1025 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 }
1027 pte_unmap(pte);
1028 if (vmas)
1029 vmas[i] = gate_vma;
1030 i++;
1031 start += PAGE_SIZE;
1032 len--;
1033 continue;
1034 }
1035
Linus Torvalds1ff80382005-12-12 16:24:33 -08001036 if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001037 || !(vm_flags & vma->vm_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 return i ? : -EFAULT;
1039
1040 if (is_vm_hugetlb_page(vma)) {
1041 i = follow_hugetlb_page(mm, vma, pages, vmas,
1042 &start, &len, i);
1043 continue;
1044 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001045
1046 foll_flags = FOLL_TOUCH;
1047 if (pages)
1048 foll_flags |= FOLL_GET;
1049 if (!write && !(vma->vm_flags & VM_LOCKED) &&
Nick Piggin54cb8822007-07-19 01:46:59 -07001050 (!vma->vm_ops || (!vma->vm_ops->nopage &&
1051 !vma->vm_ops->fault)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001052 foll_flags |= FOLL_ANON;
1053
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 do {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001055 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056
Ethan Solomita462e00c2007-07-15 23:38:16 -07001057 /*
1058 * If tsk is ooming, cut off its access to large memory
1059 * allocations. It has a pending SIGKILL, but it can't
1060 * be processed until returning to user space.
1061 */
1062 if (unlikely(test_tsk_thread_flag(tsk, TIF_MEMDIE)))
1063 return -ENOMEM;
1064
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001065 if (write)
1066 foll_flags |= FOLL_WRITE;
1067
1068 cond_resched();
Linus Torvalds6aab3412005-11-28 14:34:23 -08001069 while (!(page = follow_page(vma, start, foll_flags))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001070 int ret;
Nick Piggin83c54072007-07-19 01:47:05 -07001071 ret = handle_mm_fault(mm, vma, start,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001072 foll_flags & FOLL_WRITE);
Nick Piggin83c54072007-07-19 01:47:05 -07001073 if (ret & VM_FAULT_ERROR) {
1074 if (ret & VM_FAULT_OOM)
1075 return i ? i : -ENOMEM;
1076 else if (ret & VM_FAULT_SIGBUS)
1077 return i ? i : -EFAULT;
1078 BUG();
1079 }
1080 if (ret & VM_FAULT_MAJOR)
1081 tsk->maj_flt++;
1082 else
1083 tsk->min_flt++;
1084
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001085 /*
Nick Piggin83c54072007-07-19 01:47:05 -07001086 * The VM_FAULT_WRITE bit tells us that
1087 * do_wp_page has broken COW when necessary,
1088 * even if maybe_mkwrite decided not to set
1089 * pte_write. We can thus safely do subsequent
1090 * page lookups as if they were reads.
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001091 */
1092 if (ret & VM_FAULT_WRITE)
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001093 foll_flags &= ~FOLL_WRITE;
Nick Piggin83c54072007-07-19 01:47:05 -07001094
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07001095 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 }
1097 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001098 pages[i] = page;
James Bottomley03beb072006-03-26 01:36:57 -08001099
Russell Kinga6f36be2006-12-30 22:24:19 +00001100 flush_anon_page(vma, page, start);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001101 flush_dcache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 }
1103 if (vmas)
1104 vmas[i] = vma;
1105 i++;
1106 start += PAGE_SIZE;
1107 len--;
Hugh Dickins08ef4722005-06-21 17:15:10 -07001108 } while (len && start < vma->vm_end);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001109 } while (len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 return i;
1111}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112EXPORT_SYMBOL(get_user_pages);
1113
1114static int zeromap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1115 unsigned long addr, unsigned long end, pgprot_t prot)
1116{
1117 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001118 spinlock_t *ptl;
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001119 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
Hugh Dickinsc74df322005-10-29 18:16:23 -07001121 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 if (!pte)
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001123 return -EAGAIN;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001124 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 do {
Nick Pigginb5810032005-10-29 18:16:12 -07001126 struct page *page = ZERO_PAGE(addr);
1127 pte_t zero_pte = pte_wrprotect(mk_pte(page, prot));
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001128
1129 if (unlikely(!pte_none(*pte))) {
1130 err = -EEXIST;
1131 pte++;
1132 break;
1133 }
Nick Pigginb5810032005-10-29 18:16:12 -07001134 page_cache_get(page);
1135 page_add_file_rmap(page);
1136 inc_mm_counter(mm, file_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 set_pte_at(mm, addr, pte, zero_pte);
1138 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001139 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001140 pte_unmap_unlock(pte - 1, ptl);
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001141 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142}
1143
1144static inline int zeromap_pmd_range(struct mm_struct *mm, pud_t *pud,
1145 unsigned long addr, unsigned long end, pgprot_t prot)
1146{
1147 pmd_t *pmd;
1148 unsigned long next;
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001149 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150
1151 pmd = pmd_alloc(mm, pud, addr);
1152 if (!pmd)
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001153 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 do {
1155 next = pmd_addr_end(addr, end);
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001156 err = zeromap_pte_range(mm, pmd, addr, next, prot);
1157 if (err)
1158 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 } while (pmd++, addr = next, addr != end);
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001160 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161}
1162
1163static inline int zeromap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1164 unsigned long addr, unsigned long end, pgprot_t prot)
1165{
1166 pud_t *pud;
1167 unsigned long next;
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001168 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169
1170 pud = pud_alloc(mm, pgd, addr);
1171 if (!pud)
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001172 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 do {
1174 next = pud_addr_end(addr, end);
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001175 err = zeromap_pmd_range(mm, pud, addr, next, prot);
1176 if (err)
1177 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 } while (pud++, addr = next, addr != end);
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001179 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180}
1181
1182int zeromap_page_range(struct vm_area_struct *vma,
1183 unsigned long addr, unsigned long size, pgprot_t prot)
1184{
1185 pgd_t *pgd;
1186 unsigned long next;
1187 unsigned long end = addr + size;
1188 struct mm_struct *mm = vma->vm_mm;
1189 int err;
1190
1191 BUG_ON(addr >= end);
1192 pgd = pgd_offset(mm, addr);
1193 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 do {
1195 next = pgd_addr_end(addr, end);
1196 err = zeromap_pud_range(mm, pgd, addr, next, prot);
1197 if (err)
1198 break;
1199 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 return err;
1201}
1202
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001203pte_t * fastcall get_locked_pte(struct mm_struct *mm, unsigned long addr, spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001204{
1205 pgd_t * pgd = pgd_offset(mm, addr);
1206 pud_t * pud = pud_alloc(mm, pgd, addr);
1207 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001208 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001209 if (pmd)
1210 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1211 }
1212 return NULL;
1213}
1214
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001216 * This is the old fallback for page remapping.
1217 *
1218 * For historical reasons, it only allows reserved pages. Only
1219 * old drivers should use this, and they needed to mark their
1220 * pages reserved for the old functions anyway.
1221 */
1222static int insert_page(struct mm_struct *mm, unsigned long addr, struct page *page, pgprot_t prot)
1223{
1224 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001225 pte_t *pte;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001226 spinlock_t *ptl;
1227
1228 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001229 if (PageAnon(page))
Linus Torvalds238f58d2005-11-29 13:01:56 -08001230 goto out;
1231 retval = -ENOMEM;
1232 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001233 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001234 if (!pte)
1235 goto out;
1236 retval = -EBUSY;
1237 if (!pte_none(*pte))
1238 goto out_unlock;
1239
1240 /* Ok, finally just insert the thing.. */
1241 get_page(page);
1242 inc_mm_counter(mm, file_rss);
1243 page_add_file_rmap(page);
1244 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1245
1246 retval = 0;
1247out_unlock:
1248 pte_unmap_unlock(pte, ptl);
1249out:
1250 return retval;
1251}
1252
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001253/**
1254 * vm_insert_page - insert single page into user vma
1255 * @vma: user vma to map to
1256 * @addr: target user address of this page
1257 * @page: source kernel page
1258 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001259 * This allows drivers to insert individual pages they've allocated
1260 * into a user vma.
1261 *
1262 * The page has to be a nice clean _individual_ kernel allocation.
1263 * If you allocate a compound page, you need to have marked it as
1264 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001265 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001266 *
1267 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1268 * took an arbitrary page protection parameter. This doesn't allow
1269 * that. Your vma protection will have to be set up correctly, which
1270 * means that if you want a shared writable mapping, you'd better
1271 * ask for a shared writable mapping!
1272 *
1273 * The page does not need to be reserved.
1274 */
1275int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, struct page *page)
1276{
1277 if (addr < vma->vm_start || addr >= vma->vm_end)
1278 return -EFAULT;
1279 if (!page_count(page))
1280 return -EINVAL;
Linus Torvalds4d7672b2005-12-16 10:21:23 -08001281 vma->vm_flags |= VM_INSERTPAGE;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001282 return insert_page(vma->vm_mm, addr, page, vma->vm_page_prot);
1283}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001284EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001285
Nick Piggine0dc0d82007-02-12 00:51:36 -08001286/**
1287 * vm_insert_pfn - insert single pfn into user vma
1288 * @vma: user vma to map to
1289 * @addr: target user address of this page
1290 * @pfn: source kernel pfn
1291 *
1292 * Similar to vm_inert_page, this allows drivers to insert individual pages
1293 * they've allocated into a user vma. Same comments apply.
1294 *
1295 * This function should only be called from a vm_ops->fault handler, and
1296 * in that case the handler should return NULL.
1297 */
1298int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1299 unsigned long pfn)
1300{
1301 struct mm_struct *mm = vma->vm_mm;
1302 int retval;
1303 pte_t *pte, entry;
1304 spinlock_t *ptl;
1305
1306 BUG_ON(!(vma->vm_flags & VM_PFNMAP));
1307 BUG_ON(is_cow_mapping(vma->vm_flags));
1308
1309 retval = -ENOMEM;
1310 pte = get_locked_pte(mm, addr, &ptl);
1311 if (!pte)
1312 goto out;
1313 retval = -EBUSY;
1314 if (!pte_none(*pte))
1315 goto out_unlock;
1316
1317 /* Ok, finally just insert the thing.. */
1318 entry = pfn_pte(pfn, vma->vm_page_prot);
1319 set_pte_at(mm, addr, pte, entry);
1320 update_mmu_cache(vma, addr, entry);
1321
1322 retval = 0;
1323out_unlock:
1324 pte_unmap_unlock(pte, ptl);
1325
1326out:
1327 return retval;
1328}
1329EXPORT_SYMBOL(vm_insert_pfn);
1330
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001331/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 * maps a range of physical memory into the requested pages. the old
1333 * mappings are removed. any references to nonexistent pages results
1334 * in null mappings (currently treated as "copy-on-access")
1335 */
1336static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1337 unsigned long addr, unsigned long end,
1338 unsigned long pfn, pgprot_t prot)
1339{
1340 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001341 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342
Hugh Dickinsc74df322005-10-29 18:16:23 -07001343 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 if (!pte)
1345 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001346 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 do {
1348 BUG_ON(!pte_none(*pte));
Nick Pigginb5810032005-10-29 18:16:12 -07001349 set_pte_at(mm, addr, pte, pfn_pte(pfn, prot));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 pfn++;
1351 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001352 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001353 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 return 0;
1355}
1356
1357static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1358 unsigned long addr, unsigned long end,
1359 unsigned long pfn, pgprot_t prot)
1360{
1361 pmd_t *pmd;
1362 unsigned long next;
1363
1364 pfn -= addr >> PAGE_SHIFT;
1365 pmd = pmd_alloc(mm, pud, addr);
1366 if (!pmd)
1367 return -ENOMEM;
1368 do {
1369 next = pmd_addr_end(addr, end);
1370 if (remap_pte_range(mm, pmd, addr, next,
1371 pfn + (addr >> PAGE_SHIFT), prot))
1372 return -ENOMEM;
1373 } while (pmd++, addr = next, addr != end);
1374 return 0;
1375}
1376
1377static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1378 unsigned long addr, unsigned long end,
1379 unsigned long pfn, pgprot_t prot)
1380{
1381 pud_t *pud;
1382 unsigned long next;
1383
1384 pfn -= addr >> PAGE_SHIFT;
1385 pud = pud_alloc(mm, pgd, addr);
1386 if (!pud)
1387 return -ENOMEM;
1388 do {
1389 next = pud_addr_end(addr, end);
1390 if (remap_pmd_range(mm, pud, addr, next,
1391 pfn + (addr >> PAGE_SHIFT), prot))
1392 return -ENOMEM;
1393 } while (pud++, addr = next, addr != end);
1394 return 0;
1395}
1396
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001397/**
1398 * remap_pfn_range - remap kernel memory to userspace
1399 * @vma: user vma to map to
1400 * @addr: target user address to start at
1401 * @pfn: physical address of kernel memory
1402 * @size: size of map area
1403 * @prot: page protection flags for this mapping
1404 *
1405 * Note: this is only safe if the mm semaphore is held when called.
1406 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1408 unsigned long pfn, unsigned long size, pgprot_t prot)
1409{
1410 pgd_t *pgd;
1411 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001412 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 struct mm_struct *mm = vma->vm_mm;
1414 int err;
1415
1416 /*
1417 * Physically remapped pages are special. Tell the
1418 * rest of the world about it:
1419 * VM_IO tells people not to look at these pages
1420 * (accesses can have side effects).
Hugh Dickins0b14c172005-11-21 21:32:15 -08001421 * VM_RESERVED is specified all over the place, because
1422 * in 2.4 it kept swapout's vma scan off this vma; but
1423 * in 2.6 the LRU scan won't even find its pages, so this
1424 * flag means no more than count its pages in reserved_vm,
1425 * and omit it from core dump, even when VM_IO turned off.
Linus Torvalds6aab3412005-11-28 14:34:23 -08001426 * VM_PFNMAP tells the core MM that the base pages are just
1427 * raw PFN mappings, and do not have a "struct page" associated
1428 * with them.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001429 *
1430 * There's a horrible special case to handle copy-on-write
1431 * behaviour that some programs depend on. We mark the "original"
1432 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 */
Linus Torvalds67121172005-12-11 20:38:17 -08001434 if (is_cow_mapping(vma->vm_flags)) {
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001435 if (addr != vma->vm_start || end != vma->vm_end)
Linus Torvalds7fc7e2e2005-12-11 19:57:52 -08001436 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001437 vma->vm_pgoff = pfn;
1438 }
1439
Linus Torvalds6aab3412005-11-28 14:34:23 -08001440 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441
1442 BUG_ON(addr >= end);
1443 pfn -= addr >> PAGE_SHIFT;
1444 pgd = pgd_offset(mm, addr);
1445 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 do {
1447 next = pgd_addr_end(addr, end);
1448 err = remap_pud_range(mm, pgd, addr, next,
1449 pfn + (addr >> PAGE_SHIFT), prot);
1450 if (err)
1451 break;
1452 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 return err;
1454}
1455EXPORT_SYMBOL(remap_pfn_range);
1456
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001457static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1458 unsigned long addr, unsigned long end,
1459 pte_fn_t fn, void *data)
1460{
1461 pte_t *pte;
1462 int err;
1463 struct page *pmd_page;
Borislav Petkov94909912007-05-06 14:49:17 -07001464 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001465
1466 pte = (mm == &init_mm) ?
1467 pte_alloc_kernel(pmd, addr) :
1468 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1469 if (!pte)
1470 return -ENOMEM;
1471
1472 BUG_ON(pmd_huge(*pmd));
1473
1474 pmd_page = pmd_page(*pmd);
1475
1476 do {
1477 err = fn(pte, pmd_page, addr, data);
1478 if (err)
1479 break;
1480 } while (pte++, addr += PAGE_SIZE, addr != end);
1481
1482 if (mm != &init_mm)
1483 pte_unmap_unlock(pte-1, ptl);
1484 return err;
1485}
1486
1487static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1488 unsigned long addr, unsigned long end,
1489 pte_fn_t fn, void *data)
1490{
1491 pmd_t *pmd;
1492 unsigned long next;
1493 int err;
1494
1495 pmd = pmd_alloc(mm, pud, addr);
1496 if (!pmd)
1497 return -ENOMEM;
1498 do {
1499 next = pmd_addr_end(addr, end);
1500 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1501 if (err)
1502 break;
1503 } while (pmd++, addr = next, addr != end);
1504 return err;
1505}
1506
1507static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1508 unsigned long addr, unsigned long end,
1509 pte_fn_t fn, void *data)
1510{
1511 pud_t *pud;
1512 unsigned long next;
1513 int err;
1514
1515 pud = pud_alloc(mm, pgd, addr);
1516 if (!pud)
1517 return -ENOMEM;
1518 do {
1519 next = pud_addr_end(addr, end);
1520 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1521 if (err)
1522 break;
1523 } while (pud++, addr = next, addr != end);
1524 return err;
1525}
1526
1527/*
1528 * Scan a region of virtual memory, filling in page tables as necessary
1529 * and calling a provided function on each leaf page table.
1530 */
1531int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1532 unsigned long size, pte_fn_t fn, void *data)
1533{
1534 pgd_t *pgd;
1535 unsigned long next;
1536 unsigned long end = addr + size;
1537 int err;
1538
1539 BUG_ON(addr >= end);
1540 pgd = pgd_offset(mm, addr);
1541 do {
1542 next = pgd_addr_end(addr, end);
1543 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1544 if (err)
1545 break;
1546 } while (pgd++, addr = next, addr != end);
1547 return err;
1548}
1549EXPORT_SYMBOL_GPL(apply_to_page_range);
1550
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001552 * handle_pte_fault chooses page fault handler according to an entry
1553 * which was read non-atomically. Before making any commitment, on
1554 * those architectures or configurations (e.g. i386 with PAE) which
1555 * might give a mix of unmatched parts, do_swap_page and do_file_page
1556 * must check under lock before unmapping the pte and proceeding
1557 * (but do_wp_page is only called after already making such a check;
1558 * and do_anonymous_page and do_no_page can safely check later on).
1559 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001560static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001561 pte_t *page_table, pte_t orig_pte)
1562{
1563 int same = 1;
1564#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1565 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001566 spinlock_t *ptl = pte_lockptr(mm, pmd);
1567 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001568 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001569 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001570 }
1571#endif
1572 pte_unmap(page_table);
1573 return same;
1574}
1575
1576/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1578 * servicing faults for write access. In the normal case, do always want
1579 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1580 * that do not have writing enabled, when used by access_process_vm.
1581 */
1582static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1583{
1584 if (likely(vma->vm_flags & VM_WRITE))
1585 pte = pte_mkwrite(pte);
1586 return pte;
1587}
1588
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001589static 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 -08001590{
1591 /*
1592 * If the source page was a PFN mapping, we don't have
1593 * a "struct page" for it. We do a best-effort copy by
1594 * just copying from the original user address. If that
1595 * fails, we just zero-fill it. Live with it.
1596 */
1597 if (unlikely(!src)) {
1598 void *kaddr = kmap_atomic(dst, KM_USER0);
Linus Torvalds5d2a2db2005-11-29 14:07:55 -08001599 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1600
1601 /*
1602 * This really shouldn't fail, because the page is there
1603 * in the page tables. But it might just be unreadable,
1604 * in which case we just give up and fill the result with
1605 * zeroes.
1606 */
1607 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Linus Torvalds6aab3412005-11-28 14:34:23 -08001608 memset(kaddr, 0, PAGE_SIZE);
1609 kunmap_atomic(kaddr, KM_USER0);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07001610 flush_dcache_page(dst);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001611 return;
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001612
Linus Torvalds6aab3412005-11-28 14:34:23 -08001613 }
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001614 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001615}
1616
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 * This routine handles present pages, when users try to write
1619 * to a shared page. It is done by copying the page to a new address
1620 * and decrementing the shared-page counter for the old page.
1621 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 * Note that this routine assumes that the protection checks have been
1623 * done by the caller (the low-level page fault routine in most cases).
1624 * Thus we can safely just mark it writable once we've done any necessary
1625 * COW.
1626 *
1627 * We also mark the page dirty at this point even though the page will
1628 * change only once the write actually happens. This avoids a few races,
1629 * and potentially makes it more efficient.
1630 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001631 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1632 * but allow concurrent faults), with pte both mapped and locked.
1633 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001635static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1636 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001637 spinlock_t *ptl, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638{
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001639 struct page *old_page, *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 pte_t entry;
Nick Piggin83c54072007-07-19 01:47:05 -07001641 int reuse = 0, ret = 0;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001642 int page_mkwrite = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001643 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644
Linus Torvalds6aab3412005-11-28 14:34:23 -08001645 old_page = vm_normal_page(vma, address, orig_pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001646 if (!old_page)
1647 goto gotten;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001649 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001650 * Take out anonymous pages first, anonymous shared vmas are
1651 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001652 */
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001653 if (PageAnon(old_page)) {
1654 if (!TestSetPageLocked(old_page)) {
1655 reuse = can_share_swap_page(old_page);
1656 unlock_page(old_page);
1657 }
1658 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001659 (VM_WRITE|VM_SHARED))) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001660 /*
1661 * Only catch write-faults on shared writable pages,
1662 * read-only shared pages can get COWed by
1663 * get_user_pages(.write=1, .force=1).
1664 */
David Howells9637a5e2006-06-23 02:03:43 -07001665 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
1666 /*
1667 * Notify the address space that the page is about to
1668 * become writable so that it can prohibit this or wait
1669 * for the page to get into an appropriate state.
1670 *
1671 * We do this without the lock held, so that it can
1672 * sleep if it needs to.
1673 */
1674 page_cache_get(old_page);
1675 pte_unmap_unlock(page_table, ptl);
1676
1677 if (vma->vm_ops->page_mkwrite(vma, old_page) < 0)
1678 goto unwritable_page;
1679
David Howells9637a5e2006-06-23 02:03:43 -07001680 /*
1681 * Since we dropped the lock we need to revalidate
1682 * the PTE as someone else may have changed it. If
1683 * they did, we just return, as we can count on the
1684 * MMU to tell us if they didn't also make it writable.
1685 */
1686 page_table = pte_offset_map_lock(mm, pmd, address,
1687 &ptl);
Hugh Dickinsc3704ce2007-02-10 01:43:00 -08001688 page_cache_release(old_page);
David Howells9637a5e2006-06-23 02:03:43 -07001689 if (!pte_same(*page_table, orig_pte))
1690 goto unlock;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001691
1692 page_mkwrite = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001694 dirty_page = old_page;
1695 get_page(dirty_page);
David Howells9637a5e2006-06-23 02:03:43 -07001696 reuse = 1;
David Howells9637a5e2006-06-23 02:03:43 -07001697 }
1698
1699 if (reuse) {
1700 flush_cache_page(vma, address, pte_pfn(orig_pte));
1701 entry = pte_mkyoung(orig_pte);
1702 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07001703 if (ptep_set_access_flags(vma, address, page_table, entry,1)) {
1704 update_mmu_cache(vma, address, entry);
1705 lazy_mmu_prot_update(entry);
1706 }
David Howells9637a5e2006-06-23 02:03:43 -07001707 ret |= VM_FAULT_WRITE;
1708 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710
1711 /*
1712 * Ok, we need to copy. Oh, well..
1713 */
Nick Pigginb5810032005-10-29 18:16:12 -07001714 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08001715gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001716 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717
1718 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07001719 goto oom;
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001720 if (old_page == ZERO_PAGE(address)) {
Mel Gorman769848c2007-07-17 04:03:05 -07001721 new_page = alloc_zeroed_user_highpage_movable(vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 if (!new_page)
Hugh Dickins65500d22005-10-29 18:15:59 -07001723 goto oom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 } else {
Mel Gorman769848c2007-07-17 04:03:05 -07001725 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 if (!new_page)
Hugh Dickins65500d22005-10-29 18:15:59 -07001727 goto oom;
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001728 cow_user_page(new_page, old_page, address, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 }
Hugh Dickins65500d22005-10-29 18:15:59 -07001730
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 /*
1732 * Re-check the pte - we dropped the lock
1733 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001734 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07001735 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08001736 if (old_page) {
Nick Piggin7de6b802006-12-22 01:09:33 -08001737 page_remove_rmap(old_page, vma);
Hugh Dickins920fc352005-11-21 21:32:17 -08001738 if (!PageAnon(old_page)) {
1739 dec_mm_counter(mm, file_rss);
1740 inc_mm_counter(mm, anon_rss);
1741 }
1742 } else
Hugh Dickins42946212005-10-29 18:16:05 -07001743 inc_mm_counter(mm, anon_rss);
Ben Collinseca35132005-11-29 11:45:26 -08001744 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07001745 entry = mk_pte(new_page, vma->vm_page_prot);
1746 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Anil Keshavamurthyc38c8db2006-07-14 00:23:57 -07001747 lazy_mmu_prot_update(entry);
Siddha, Suresh B4ce072f2006-09-29 01:58:42 -07001748 /*
1749 * Clear the pte entry and flush it first, before updating the
1750 * pte with the new entry. This will avoid a race condition
1751 * seen in the presence of one thread doing SMC and another
1752 * thread doing COW.
1753 */
1754 ptep_clear_flush(vma, address, page_table);
1755 set_pte_at(mm, address, page_table, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07001756 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 lru_cache_add_active(new_page);
Nick Piggin9617d952006-01-06 00:11:12 -08001758 page_add_new_anon_rmap(new_page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759
1760 /* Free the old page.. */
1761 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001762 ret |= VM_FAULT_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 }
Hugh Dickins920fc352005-11-21 21:32:17 -08001764 if (new_page)
1765 page_cache_release(new_page);
1766 if (old_page)
1767 page_cache_release(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001768unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001769 pte_unmap_unlock(page_table, ptl);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001770 if (dirty_page) {
Nick Piggin79352892007-07-19 01:47:22 -07001771 /*
1772 * Yes, Virginia, this is actually required to prevent a race
1773 * with clear_page_dirty_for_io() from clearing the page dirty
1774 * bit after it clear all dirty ptes, but before a racing
1775 * do_wp_page installs a dirty pte.
1776 *
1777 * do_no_page is protected similarly.
1778 */
1779 wait_on_page_locked(dirty_page);
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001780 set_page_dirty_balance(dirty_page, page_mkwrite);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001781 put_page(dirty_page);
1782 }
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001783 return ret;
Hugh Dickins65500d22005-10-29 18:15:59 -07001784oom:
Hugh Dickins920fc352005-11-21 21:32:17 -08001785 if (old_page)
1786 page_cache_release(old_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 return VM_FAULT_OOM;
David Howells9637a5e2006-06-23 02:03:43 -07001788
1789unwritable_page:
1790 page_cache_release(old_page);
1791 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792}
1793
1794/*
1795 * Helper functions for unmap_mapping_range().
1796 *
1797 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1798 *
1799 * We have to restart searching the prio_tree whenever we drop the lock,
1800 * since the iterator is only valid while the lock is held, and anyway
1801 * a later vma might be split and reinserted earlier while lock dropped.
1802 *
1803 * The list of nonlinear vmas could be handled more efficiently, using
1804 * a placeholder, but handle it in the same way until a need is shown.
1805 * It is important to search the prio_tree before nonlinear list: a vma
1806 * may become nonlinear and be shifted from prio_tree to nonlinear list
1807 * while the lock is dropped; but never shifted from list to prio_tree.
1808 *
1809 * In order to make forward progress despite restarting the search,
1810 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1811 * quickly skip it next time around. Since the prio_tree search only
1812 * shows us those vmas affected by unmapping the range in question, we
1813 * can't efficiently keep all vmas in step with mapping->truncate_count:
1814 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1815 * mapping->truncate_count and vma->vm_truncate_count are protected by
1816 * i_mmap_lock.
1817 *
1818 * In order to make forward progress despite repeatedly restarting some
Hugh Dickinsee39b372005-04-19 13:29:15 -07001819 * large vma, note the restart_addr from unmap_vmas when it breaks out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 * and restart from that address when we reach that vma again. It might
1821 * have been split or merged, shrunk or extended, but never shifted: so
1822 * restart_addr remains valid so long as it remains in the vma's range.
1823 * unmap_mapping_range forces truncate_count to leap over page-aligned
1824 * values so we can save vma's restart_addr in its truncate_count field.
1825 */
1826#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1827
1828static void reset_vma_truncate_counts(struct address_space *mapping)
1829{
1830 struct vm_area_struct *vma;
1831 struct prio_tree_iter iter;
1832
1833 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1834 vma->vm_truncate_count = 0;
1835 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1836 vma->vm_truncate_count = 0;
1837}
1838
1839static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1840 unsigned long start_addr, unsigned long end_addr,
1841 struct zap_details *details)
1842{
1843 unsigned long restart_addr;
1844 int need_break;
1845
Nick Piggind00806b2007-07-19 01:46:57 -07001846 /*
1847 * files that support invalidating or truncating portions of the
Nick Piggind0217ac2007-07-19 01:47:03 -07001848 * file from under mmaped areas must have their ->fault function
Nick Piggin83c54072007-07-19 01:47:05 -07001849 * return a locked page (and set VM_FAULT_LOCKED in the return).
1850 * This provides synchronisation against concurrent unmapping here.
Nick Piggind00806b2007-07-19 01:46:57 -07001851 */
Nick Piggind00806b2007-07-19 01:46:57 -07001852
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853again:
1854 restart_addr = vma->vm_truncate_count;
1855 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1856 start_addr = restart_addr;
1857 if (start_addr >= end_addr) {
1858 /* Top of vma has been split off since last time */
1859 vma->vm_truncate_count = details->truncate_count;
1860 return 0;
1861 }
1862 }
1863
Hugh Dickinsee39b372005-04-19 13:29:15 -07001864 restart_addr = zap_page_range(vma, start_addr,
1865 end_addr - start_addr, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 need_break = need_resched() ||
1867 need_lockbreak(details->i_mmap_lock);
1868
Hugh Dickinsee39b372005-04-19 13:29:15 -07001869 if (restart_addr >= end_addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 /* We have now completed this vma: mark it so */
1871 vma->vm_truncate_count = details->truncate_count;
1872 if (!need_break)
1873 return 0;
1874 } else {
1875 /* Note restart_addr in vma's truncate_count field */
Hugh Dickinsee39b372005-04-19 13:29:15 -07001876 vma->vm_truncate_count = restart_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 if (!need_break)
1878 goto again;
1879 }
1880
1881 spin_unlock(details->i_mmap_lock);
1882 cond_resched();
1883 spin_lock(details->i_mmap_lock);
1884 return -EINTR;
1885}
1886
1887static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1888 struct zap_details *details)
1889{
1890 struct vm_area_struct *vma;
1891 struct prio_tree_iter iter;
1892 pgoff_t vba, vea, zba, zea;
1893
1894restart:
1895 vma_prio_tree_foreach(vma, &iter, root,
1896 details->first_index, details->last_index) {
1897 /* Skip quickly over those we have already dealt with */
1898 if (vma->vm_truncate_count == details->truncate_count)
1899 continue;
1900
1901 vba = vma->vm_pgoff;
1902 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1903 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1904 zba = details->first_index;
1905 if (zba < vba)
1906 zba = vba;
1907 zea = details->last_index;
1908 if (zea > vea)
1909 zea = vea;
1910
1911 if (unmap_mapping_range_vma(vma,
1912 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1913 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1914 details) < 0)
1915 goto restart;
1916 }
1917}
1918
1919static inline void unmap_mapping_range_list(struct list_head *head,
1920 struct zap_details *details)
1921{
1922 struct vm_area_struct *vma;
1923
1924 /*
1925 * In nonlinear VMAs there is no correspondence between virtual address
1926 * offset and file offset. So we must perform an exhaustive search
1927 * across *all* the pages in each nonlinear VMA, not just the pages
1928 * whose virtual address lies outside the file truncation point.
1929 */
1930restart:
1931 list_for_each_entry(vma, head, shared.vm_set.list) {
1932 /* Skip quickly over those we have already dealt with */
1933 if (vma->vm_truncate_count == details->truncate_count)
1934 continue;
1935 details->nonlinear_vma = vma;
1936 if (unmap_mapping_range_vma(vma, vma->vm_start,
1937 vma->vm_end, details) < 0)
1938 goto restart;
1939 }
1940}
1941
1942/**
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001943 * 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 -07001944 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 * @holebegin: byte in first page to unmap, relative to the start of
1946 * the underlying file. This will be rounded down to a PAGE_SIZE
1947 * boundary. Note that this is different from vmtruncate(), which
1948 * must keep the partial page. In contrast, we must get rid of
1949 * partial pages.
1950 * @holelen: size of prospective hole in bytes. This will be rounded
1951 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1952 * end of the file.
1953 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1954 * but 0 when invalidating pagecache, don't throw away private data.
1955 */
1956void unmap_mapping_range(struct address_space *mapping,
1957 loff_t const holebegin, loff_t const holelen, int even_cows)
1958{
1959 struct zap_details details;
1960 pgoff_t hba = holebegin >> PAGE_SHIFT;
1961 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1962
1963 /* Check for overflow. */
1964 if (sizeof(holelen) > sizeof(hlen)) {
1965 long long holeend =
1966 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1967 if (holeend & ~(long long)ULONG_MAX)
1968 hlen = ULONG_MAX - hba + 1;
1969 }
1970
1971 details.check_mapping = even_cows? NULL: mapping;
1972 details.nonlinear_vma = NULL;
1973 details.first_index = hba;
1974 details.last_index = hba + hlen - 1;
1975 if (details.last_index < details.first_index)
1976 details.last_index = ULONG_MAX;
1977 details.i_mmap_lock = &mapping->i_mmap_lock;
1978
1979 spin_lock(&mapping->i_mmap_lock);
1980
Nick Piggind00806b2007-07-19 01:46:57 -07001981 /* Protect against endless unmapping loops */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 mapping->truncate_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 if (unlikely(is_restart_addr(mapping->truncate_count))) {
1984 if (mapping->truncate_count == 0)
1985 reset_vma_truncate_counts(mapping);
1986 mapping->truncate_count++;
1987 }
1988 details.truncate_count = mapping->truncate_count;
1989
1990 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
1991 unmap_mapping_range_tree(&mapping->i_mmap, &details);
1992 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
1993 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
1994 spin_unlock(&mapping->i_mmap_lock);
1995}
1996EXPORT_SYMBOL(unmap_mapping_range);
1997
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001998/**
1999 * vmtruncate - unmap mappings "freed" by truncate() syscall
2000 * @inode: inode of the file used
2001 * @offset: file offset to start truncating
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 *
2003 * NOTE! We have to be ready to update the memory sharing
2004 * between the file and the memory map for a potential last
2005 * incomplete page. Ugly, but necessary.
2006 */
2007int vmtruncate(struct inode * inode, loff_t offset)
2008{
2009 struct address_space *mapping = inode->i_mapping;
2010 unsigned long limit;
2011
2012 if (inode->i_size < offset)
2013 goto do_expand;
2014 /*
2015 * truncation of in-use swapfiles is disallowed - it would cause
2016 * subsequent swapout to scribble on the now-freed blocks.
2017 */
2018 if (IS_SWAPFILE(inode))
2019 goto out_busy;
2020 i_size_write(inode, offset);
Nick Piggind00806b2007-07-19 01:46:57 -07002021
2022 /*
2023 * unmap_mapping_range is called twice, first simply for efficiency
2024 * so that truncate_inode_pages does fewer single-page unmaps. However
2025 * after this first call, and before truncate_inode_pages finishes,
2026 * it is possible for private pages to be COWed, which remain after
2027 * truncate_inode_pages finishes, hence the second unmap_mapping_range
2028 * call must be made for correctness.
2029 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2031 truncate_inode_pages(mapping, offset);
Nick Piggind00806b2007-07-19 01:46:57 -07002032 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 goto out_truncate;
2034
2035do_expand:
2036 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
2037 if (limit != RLIM_INFINITY && offset > limit)
2038 goto out_sig;
2039 if (offset > inode->i_sb->s_maxbytes)
2040 goto out_big;
2041 i_size_write(inode, offset);
2042
2043out_truncate:
2044 if (inode->i_op && inode->i_op->truncate)
2045 inode->i_op->truncate(inode);
2046 return 0;
2047out_sig:
2048 send_sig(SIGXFSZ, current, 0);
2049out_big:
2050 return -EFBIG;
2051out_busy:
2052 return -ETXTBSY;
2053}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054EXPORT_SYMBOL(vmtruncate);
2055
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002056int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
2057{
2058 struct address_space *mapping = inode->i_mapping;
2059
2060 /*
2061 * If the underlying filesystem is not going to provide
2062 * a way to truncate a range of blocks (punch a hole) -
2063 * we should return failure right now.
2064 */
2065 if (!inode->i_op || !inode->i_op->truncate_range)
2066 return -ENOSYS;
2067
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002068 mutex_lock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002069 down_write(&inode->i_alloc_sem);
2070 unmap_mapping_range(mapping, offset, (end - offset), 1);
2071 truncate_inode_pages_range(mapping, offset, end);
Nick Piggind00806b2007-07-19 01:46:57 -07002072 unmap_mapping_range(mapping, offset, (end - offset), 1);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002073 inode->i_op->truncate_range(inode, offset, end);
2074 up_write(&inode->i_alloc_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002075 mutex_unlock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002076
2077 return 0;
2078}
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002079
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07002080/**
2081 * swapin_readahead - swap in pages in hope we need them soon
2082 * @entry: swap entry of this memory
2083 * @addr: address to start
2084 * @vma: user vma this addresses belong to
2085 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 * Primitive swap readahead code. We simply read an aligned block of
2087 * (1 << page_cluster) entries in the swap area. This method is chosen
2088 * because it doesn't cost us any seek time. We also make sure to queue
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07002089 * the 'original' request together with the readahead ones...
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 *
2091 * This has been extended to use the NUMA policies from the mm triggering
2092 * the readahead.
2093 *
2094 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
2095 */
2096void swapin_readahead(swp_entry_t entry, unsigned long addr,struct vm_area_struct *vma)
2097{
2098#ifdef CONFIG_NUMA
2099 struct vm_area_struct *next_vma = vma ? vma->vm_next : NULL;
2100#endif
2101 int i, num;
2102 struct page *new_page;
2103 unsigned long offset;
2104
2105 /*
2106 * Get the number of handles we should do readahead io to.
2107 */
2108 num = valid_swaphandles(entry, &offset);
2109 for (i = 0; i < num; offset++, i++) {
2110 /* Ok, do the async read-ahead now */
2111 new_page = read_swap_cache_async(swp_entry(swp_type(entry),
2112 offset), vma, addr);
2113 if (!new_page)
2114 break;
2115 page_cache_release(new_page);
2116#ifdef CONFIG_NUMA
2117 /*
2118 * Find the next applicable VMA for the NUMA policy.
2119 */
2120 addr += PAGE_SIZE;
2121 if (addr == 0)
2122 vma = NULL;
2123 if (vma) {
2124 if (addr >= vma->vm_end) {
2125 vma = next_vma;
2126 next_vma = vma ? vma->vm_next : NULL;
2127 }
2128 if (vma && addr < vma->vm_start)
2129 vma = NULL;
2130 } else {
2131 if (next_vma && addr >= next_vma->vm_start) {
2132 vma = next_vma;
2133 next_vma = vma->vm_next;
2134 }
2135 }
2136#endif
2137 }
2138 lru_add_drain(); /* Push any new pages onto the LRU now */
2139}
2140
2141/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002142 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2143 * but allow concurrent faults), and pte mapped but not yet locked.
2144 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002146static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2147 unsigned long address, pte_t *page_table, pmd_t *pmd,
2148 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002150 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 struct page *page;
Hugh Dickins65500d22005-10-29 18:15:59 -07002152 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 pte_t pte;
Nick Piggin83c54072007-07-19 01:47:05 -07002154 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002156 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002157 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002158
2159 entry = pte_to_swp_entry(orig_pte);
Christoph Lameter06972122006-06-23 02:03:35 -07002160 if (is_migration_entry(entry)) {
2161 migration_entry_wait(mm, pmd, address);
2162 goto out;
2163 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002164 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 page = lookup_swap_cache(entry);
2166 if (!page) {
Ashwin Chaugule098fe652006-12-06 20:31:54 -08002167 grab_swap_token(); /* Contend for token _before_ read-in */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 swapin_readahead(entry, address, vma);
2169 page = read_swap_cache_async(entry, vma, address);
2170 if (!page) {
2171 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002172 * Back out if somebody else faulted in this pte
2173 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002175 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 if (likely(pte_same(*page_table, orig_pte)))
2177 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002178 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002179 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 }
2181
2182 /* Had to read the page from swap area: Major fault */
2183 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002184 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 }
2186
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002187 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 mark_page_accessed(page);
2189 lock_page(page);
2190
2191 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002192 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002194 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002195 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002196 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002197
2198 if (unlikely(!PageUptodate(page))) {
2199 ret = VM_FAULT_SIGBUS;
2200 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 }
2202
2203 /* The page isn't present yet, go ahead with the fault. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204
Hugh Dickins42946212005-10-29 18:16:05 -07002205 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 pte = mk_pte(page, vma->vm_page_prot);
2207 if (write_access && can_share_swap_page(page)) {
2208 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2209 write_access = 0;
2210 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211
2212 flush_icache_page(vma, page);
2213 set_pte_at(mm, address, page_table, pte);
2214 page_add_anon_rmap(page, vma, address);
2215
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002216 swap_free(entry);
2217 if (vm_swap_full())
2218 remove_exclusive_swap_page(page);
2219 unlock_page(page);
2220
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 if (write_access) {
Nick Piggin83c54072007-07-19 01:47:05 -07002222 /* XXX: We could OR the do_wp_page code with this one? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 if (do_wp_page(mm, vma, address,
Nick Piggin83c54072007-07-19 01:47:05 -07002224 page_table, pmd, ptl, pte) & VM_FAULT_OOM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 ret = VM_FAULT_OOM;
2226 goto out;
2227 }
2228
2229 /* No need to invalidate - it was non-present before */
2230 update_mmu_cache(vma, address, pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07002231unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002232 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233out:
2234 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002235out_nomap:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002236 pte_unmap_unlock(page_table, ptl);
Kirill Korotaevb8107482005-05-16 21:53:50 -07002237 unlock_page(page);
2238 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002239 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240}
2241
2242/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002243 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2244 * but allow concurrent faults), and pte mapped but not yet locked.
2245 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002247static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2248 unsigned long address, pte_t *page_table, pmd_t *pmd,
2249 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002251 struct page *page;
2252 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254
Linus Torvalds6aab3412005-11-28 14:34:23 -08002255 if (write_access) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 /* Allocate our own private page. */
2257 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258
2259 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07002260 goto oom;
Mel Gorman769848c2007-07-17 04:03:05 -07002261 page = alloc_zeroed_user_highpage_movable(vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 if (!page)
Hugh Dickins65500d22005-10-29 18:15:59 -07002263 goto oom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264
Hugh Dickins65500d22005-10-29 18:15:59 -07002265 entry = mk_pte(page, vma->vm_page_prot);
2266 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002267
2268 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2269 if (!pte_none(*page_table))
2270 goto release;
2271 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 lru_cache_add_active(page);
Nick Piggin9617d952006-01-06 00:11:12 -08002273 page_add_new_anon_rmap(page, vma, address);
Nick Pigginb5810032005-10-29 18:16:12 -07002274 } else {
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002275 /* Map the ZERO_PAGE - vm_page_prot is readonly */
2276 page = ZERO_PAGE(address);
2277 page_cache_get(page);
2278 entry = mk_pte(page, vma->vm_page_prot);
2279
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002280 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002281 spin_lock(ptl);
2282 if (!pte_none(*page_table))
2283 goto release;
Nick Pigginb5810032005-10-29 18:16:12 -07002284 inc_mm_counter(mm, file_rss);
2285 page_add_file_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 }
2287
Hugh Dickins65500d22005-10-29 18:15:59 -07002288 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289
2290 /* No need to invalidate - it was non-present before */
Hugh Dickins65500d22005-10-29 18:15:59 -07002291 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 lazy_mmu_prot_update(entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002293unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002294 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002295 return 0;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002296release:
2297 page_cache_release(page);
2298 goto unlock;
Hugh Dickins65500d22005-10-29 18:15:59 -07002299oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 return VM_FAULT_OOM;
2301}
2302
2303/*
Nick Piggin54cb8822007-07-19 01:46:59 -07002304 * __do_fault() tries to create a new page mapping. It aggressively
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305 * tries to share with existing pages, but makes a separate copy if
Nick Piggin54cb8822007-07-19 01:46:59 -07002306 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2307 * the next page fault.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 *
2309 * As this is called only for pages that do not currently exist, we
2310 * do not need to flush old virtual caches or the TLB.
2311 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002312 * We enter with non-exclusive mmap_sem (to exclude vma changes,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002313 * but allow concurrent faults), and pte neither mapped nor locked.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002314 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 */
Nick Piggin54cb8822007-07-19 01:46:59 -07002316static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002317 unsigned long address, pmd_t *pmd,
Nick Piggin54cb8822007-07-19 01:46:59 -07002318 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319{
Hugh Dickins16abfa02007-10-04 16:56:06 +01002320 pte_t *page_table;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002321 spinlock_t *ptl;
Nick Piggind0217ac2007-07-19 01:47:03 -07002322 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 int anon = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002325 struct page *dirty_page = NULL;
Nick Piggind0217ac2007-07-19 01:47:03 -07002326 struct vm_fault vmf;
2327 int ret;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002328 int page_mkwrite = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07002329
Nick Piggind0217ac2007-07-19 01:47:03 -07002330 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2331 vmf.pgoff = pgoff;
2332 vmf.flags = flags;
2333 vmf.page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334
Hugh Dickins325f04d2005-11-29 16:55:48 +00002335 BUG_ON(vma->vm_flags & VM_PFNMAP);
2336
Nick Piggin54cb8822007-07-19 01:46:59 -07002337 if (likely(vma->vm_ops->fault)) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002338 ret = vma->vm_ops->fault(vma, &vmf);
Nick Piggin83c54072007-07-19 01:47:05 -07002339 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2340 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07002341 } else {
2342 /* Legacy ->nopage path */
Nick Piggin83c54072007-07-19 01:47:05 -07002343 ret = 0;
Nick Piggind0217ac2007-07-19 01:47:03 -07002344 vmf.page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret);
Nick Piggin54cb8822007-07-19 01:46:59 -07002345 /* no page was available -- either SIGBUS or OOM */
Nick Piggind0217ac2007-07-19 01:47:03 -07002346 if (unlikely(vmf.page == NOPAGE_SIGBUS))
Nick Piggin54cb8822007-07-19 01:46:59 -07002347 return VM_FAULT_SIGBUS;
Nick Piggind0217ac2007-07-19 01:47:03 -07002348 else if (unlikely(vmf.page == NOPAGE_OOM))
Nick Piggin54cb8822007-07-19 01:46:59 -07002349 return VM_FAULT_OOM;
2350 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351
Nick Piggind00806b2007-07-19 01:46:57 -07002352 /*
Nick Piggind0217ac2007-07-19 01:47:03 -07002353 * For consistency in subsequent calls, make the faulted page always
Nick Piggind00806b2007-07-19 01:46:57 -07002354 * locked.
2355 */
Nick Piggin83c54072007-07-19 01:47:05 -07002356 if (unlikely(!(ret & VM_FAULT_LOCKED)))
Nick Piggind0217ac2007-07-19 01:47:03 -07002357 lock_page(vmf.page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002358 else
Nick Piggind0217ac2007-07-19 01:47:03 -07002359 VM_BUG_ON(!PageLocked(vmf.page));
Nick Piggind00806b2007-07-19 01:46:57 -07002360
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 /*
2362 * Should we do an early C-O-W break?
2363 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002364 page = vmf.page;
Nick Piggin54cb8822007-07-19 01:46:59 -07002365 if (flags & FAULT_FLAG_WRITE) {
David Howells9637a5e2006-06-23 02:03:43 -07002366 if (!(vma->vm_flags & VM_SHARED)) {
Nick Piggin54cb8822007-07-19 01:46:59 -07002367 anon = 1;
Nick Piggind00806b2007-07-19 01:46:57 -07002368 if (unlikely(anon_vma_prepare(vma))) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002369 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002370 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002371 }
Nick Piggin83c54072007-07-19 01:47:05 -07002372 page = alloc_page_vma(GFP_HIGHUSER_MOVABLE,
2373 vma, address);
Nick Piggind00806b2007-07-19 01:46:57 -07002374 if (!page) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002375 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002376 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002377 }
Nick Piggind0217ac2007-07-19 01:47:03 -07002378 copy_user_highpage(page, vmf.page, address, vma);
David Howells9637a5e2006-06-23 02:03:43 -07002379 } else {
Nick Piggin54cb8822007-07-19 01:46:59 -07002380 /*
2381 * If the page will be shareable, see if the backing
David Howells9637a5e2006-06-23 02:03:43 -07002382 * address space wants to know that the page is about
Nick Piggin54cb8822007-07-19 01:46:59 -07002383 * to become writable
2384 */
Mark Fasheh69676142007-07-19 01:47:00 -07002385 if (vma->vm_ops->page_mkwrite) {
2386 unlock_page(page);
2387 if (vma->vm_ops->page_mkwrite(vma, page) < 0) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002388 ret = VM_FAULT_SIGBUS;
2389 anon = 1; /* no anon but release vmf.page */
Mark Fasheh69676142007-07-19 01:47:00 -07002390 goto out_unlocked;
2391 }
2392 lock_page(page);
Nick Piggind0217ac2007-07-19 01:47:03 -07002393 /*
2394 * XXX: this is not quite right (racy vs
2395 * invalidate) to unlock and relock the page
2396 * like this, however a better fix requires
2397 * reworking page_mkwrite locking API, which
2398 * is better done later.
2399 */
2400 if (!page->mapping) {
Nick Piggin83c54072007-07-19 01:47:05 -07002401 ret = 0;
Nick Piggind0217ac2007-07-19 01:47:03 -07002402 anon = 1; /* no anon but release vmf.page */
2403 goto out;
2404 }
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002405 page_mkwrite = 1;
David Howells9637a5e2006-06-23 02:03:43 -07002406 }
2407 }
Nick Piggin54cb8822007-07-19 01:46:59 -07002408
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 }
2410
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002411 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412
2413 /*
2414 * This silly early PAGE_DIRTY setting removes a race
2415 * due to the bad i386 page protection. But it's valid
2416 * for other architectures too.
2417 *
2418 * Note that if write_access is true, we either now have
2419 * an exclusive copy of the page, or this is a shared mapping,
2420 * so we can make it writable and dirty to avoid having to
2421 * handle that later.
2422 */
2423 /* Only go through if we didn't race with anybody else... */
Nick Piggin54cb8822007-07-19 01:46:59 -07002424 if (likely(pte_same(*page_table, orig_pte))) {
Nick Piggind00806b2007-07-19 01:46:57 -07002425 flush_icache_page(vma, page);
2426 entry = mk_pte(page, vma->vm_page_prot);
Nick Piggin54cb8822007-07-19 01:46:59 -07002427 if (flags & FAULT_FLAG_WRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2429 set_pte_at(mm, address, page_table, entry);
2430 if (anon) {
Nick Piggind00806b2007-07-19 01:46:57 -07002431 inc_mm_counter(mm, anon_rss);
2432 lru_cache_add_active(page);
2433 page_add_new_anon_rmap(page, vma, address);
Hugh Dickinsf57e88a2005-11-21 21:32:19 -08002434 } else {
Hugh Dickins42946212005-10-29 18:16:05 -07002435 inc_mm_counter(mm, file_rss);
Nick Piggind00806b2007-07-19 01:46:57 -07002436 page_add_file_rmap(page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002437 if (flags & FAULT_FLAG_WRITE) {
Nick Piggind00806b2007-07-19 01:46:57 -07002438 dirty_page = page;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002439 get_page(dirty_page);
2440 }
Hugh Dickins42946212005-10-29 18:16:05 -07002441 }
Nick Piggind00806b2007-07-19 01:46:57 -07002442
2443 /* no need to invalidate: a not-present page won't be cached */
2444 update_mmu_cache(vma, address, entry);
2445 lazy_mmu_prot_update(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 } else {
Nick Piggind00806b2007-07-19 01:46:57 -07002447 if (anon)
2448 page_cache_release(page);
2449 else
Nick Piggin54cb8822007-07-19 01:46:59 -07002450 anon = 1; /* no anon but release faulted_page */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451 }
2452
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002453 pte_unmap_unlock(page_table, ptl);
Nick Piggind00806b2007-07-19 01:46:57 -07002454
2455out:
Nick Piggind0217ac2007-07-19 01:47:03 -07002456 unlock_page(vmf.page);
Mark Fasheh69676142007-07-19 01:47:00 -07002457out_unlocked:
Nick Piggind00806b2007-07-19 01:46:57 -07002458 if (anon)
Nick Piggind0217ac2007-07-19 01:47:03 -07002459 page_cache_release(vmf.page);
Nick Piggind00806b2007-07-19 01:46:57 -07002460 else if (dirty_page) {
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002461 set_page_dirty_balance(dirty_page, page_mkwrite);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002462 put_page(dirty_page);
2463 }
Nick Piggind00806b2007-07-19 01:46:57 -07002464
Nick Piggin83c54072007-07-19 01:47:05 -07002465 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07002466}
Nick Piggind00806b2007-07-19 01:46:57 -07002467
Nick Piggin54cb8822007-07-19 01:46:59 -07002468static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2469 unsigned long address, pte_t *page_table, pmd_t *pmd,
2470 int write_access, pte_t orig_pte)
2471{
2472 pgoff_t pgoff = (((address & PAGE_MASK)
2473 - vma->vm_start) >> PAGE_CACHE_SHIFT) + vma->vm_pgoff;
2474 unsigned int flags = (write_access ? FAULT_FLAG_WRITE : 0);
2475
Hugh Dickins16abfa02007-10-04 16:56:06 +01002476 pte_unmap(page_table);
2477 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Nick Piggin54cb8822007-07-19 01:46:59 -07002478}
2479
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480
2481/*
Jes Sorensenf4b81802006-09-27 01:50:10 -07002482 * do_no_pfn() tries to create a new page mapping for a page without
2483 * a struct_page backing it
2484 *
2485 * As this is called only for pages that do not currently exist, we
2486 * do not need to flush old virtual caches or the TLB.
2487 *
2488 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2489 * but allow concurrent faults), and pte mapped but not yet locked.
2490 * We return with mmap_sem still held, but pte unmapped and unlocked.
2491 *
2492 * It is expected that the ->nopfn handler always returns the same pfn
2493 * for a given virtual mapping.
2494 *
2495 * Mark this `noinline' to prevent it from bloating the main pagefault code.
2496 */
2497static noinline int do_no_pfn(struct mm_struct *mm, struct vm_area_struct *vma,
2498 unsigned long address, pte_t *page_table, pmd_t *pmd,
2499 int write_access)
2500{
2501 spinlock_t *ptl;
2502 pte_t entry;
2503 unsigned long pfn;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002504
2505 pte_unmap(page_table);
2506 BUG_ON(!(vma->vm_flags & VM_PFNMAP));
2507 BUG_ON(is_cow_mapping(vma->vm_flags));
2508
2509 pfn = vma->vm_ops->nopfn(vma, address & PAGE_MASK);
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002510 if (unlikely(pfn == NOPFN_OOM))
Jes Sorensenf4b81802006-09-27 01:50:10 -07002511 return VM_FAULT_OOM;
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002512 else if (unlikely(pfn == NOPFN_SIGBUS))
Jes Sorensenf4b81802006-09-27 01:50:10 -07002513 return VM_FAULT_SIGBUS;
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002514 else if (unlikely(pfn == NOPFN_REFAULT))
Nick Piggin83c54072007-07-19 01:47:05 -07002515 return 0;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002516
2517 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2518
2519 /* Only go through if we didn't race with anybody else... */
2520 if (pte_none(*page_table)) {
2521 entry = pfn_pte(pfn, vma->vm_page_prot);
2522 if (write_access)
2523 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2524 set_pte_at(mm, address, page_table, entry);
2525 }
2526 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002527 return 0;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002528}
2529
2530/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 * Fault of a previously existing named mapping. Repopulate the pte
2532 * from the encoded file_pte if possible. This enables swappable
2533 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002534 *
2535 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2536 * but allow concurrent faults), and pte mapped but not yet locked.
2537 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002539static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins65500d22005-10-29 18:15:59 -07002540 unsigned long address, pte_t *page_table, pmd_t *pmd,
2541 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542{
Nick Piggind0217ac2007-07-19 01:47:03 -07002543 unsigned int flags = FAULT_FLAG_NONLINEAR |
2544 (write_access ? FAULT_FLAG_WRITE : 0);
Hugh Dickins65500d22005-10-29 18:15:59 -07002545 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002547 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Nick Piggin83c54072007-07-19 01:47:05 -07002548 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549
Nick Piggind0217ac2007-07-19 01:47:03 -07002550 if (unlikely(!(vma->vm_flags & VM_NONLINEAR) ||
2551 !(vma->vm_flags & VM_CAN_NONLINEAR))) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002552 /*
2553 * Page table corrupted: show pte and kill process.
2554 */
Nick Pigginb5810032005-10-29 18:16:12 -07002555 print_bad_pte(vma, orig_pte, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002556 return VM_FAULT_OOM;
2557 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002558
2559 pgoff = pte_to_pgoff(orig_pte);
Hugh Dickins16abfa02007-10-04 16:56:06 +01002560 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561}
2562
2563/*
2564 * These routines also need to handle stuff like marking pages dirty
2565 * and/or accessed for architectures that don't do it in hardware (most
2566 * RISC architectures). The early dirtying is also good on the i386.
2567 *
2568 * There is also a hook called "update_mmu_cache()" that architectures
2569 * with external mmu caches can use to update those (ie the Sparc or
2570 * PowerPC hashed page tables that act as extended TLBs).
2571 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07002572 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2573 * but allow concurrent faults), and pte mapped but not yet locked.
2574 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 */
2576static inline int handle_pte_fault(struct mm_struct *mm,
Hugh Dickins65500d22005-10-29 18:15:59 -07002577 struct vm_area_struct *vma, unsigned long address,
2578 pte_t *pte, pmd_t *pmd, int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579{
2580 pte_t entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002581 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002583 entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002585 if (pte_none(entry)) {
Jes Sorensenf4b81802006-09-27 01:50:10 -07002586 if (vma->vm_ops) {
Nick Piggin54cb8822007-07-19 01:46:59 -07002587 if (vma->vm_ops->fault || vma->vm_ops->nopage)
2588 return do_linear_fault(mm, vma, address,
2589 pte, pmd, write_access, entry);
Jes Sorensenf4b81802006-09-27 01:50:10 -07002590 if (unlikely(vma->vm_ops->nopfn))
2591 return do_no_pfn(mm, vma, address, pte,
2592 pmd, write_access);
2593 }
2594 return do_anonymous_page(mm, vma, address,
2595 pte, pmd, write_access);
Hugh Dickins65500d22005-10-29 18:15:59 -07002596 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 if (pte_file(entry))
Nick Piggind0217ac2007-07-19 01:47:03 -07002598 return do_nonlinear_fault(mm, vma, address,
Hugh Dickins65500d22005-10-29 18:15:59 -07002599 pte, pmd, write_access, entry);
2600 return do_swap_page(mm, vma, address,
2601 pte, pmd, write_access, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602 }
2603
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002604 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002605 spin_lock(ptl);
2606 if (unlikely(!pte_same(*pte, entry)))
2607 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 if (write_access) {
2609 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002610 return do_wp_page(mm, vma, address,
2611 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 entry = pte_mkdirty(entry);
2613 }
2614 entry = pte_mkyoung(entry);
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002615 if (ptep_set_access_flags(vma, address, pte, entry, write_access)) {
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002616 update_mmu_cache(vma, address, entry);
2617 lazy_mmu_prot_update(entry);
2618 } else {
2619 /*
2620 * This is needed only for protection faults but the arch code
2621 * is not yet telling us if this is a protection fault or not.
2622 * This still avoids useless tlb flushes for .text page faults
2623 * with threads.
2624 */
2625 if (write_access)
2626 flush_tlb_page(vma, address);
2627 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002628unlock:
2629 pte_unmap_unlock(pte, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002630 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631}
2632
2633/*
2634 * By the time we get here, we already hold the mm semaphore
2635 */
Nick Piggin83c54072007-07-19 01:47:05 -07002636int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637 unsigned long address, int write_access)
2638{
2639 pgd_t *pgd;
2640 pud_t *pud;
2641 pmd_t *pmd;
2642 pte_t *pte;
2643
2644 __set_current_state(TASK_RUNNING);
2645
Christoph Lameterf8891e52006-06-30 01:55:45 -07002646 count_vm_event(PGFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01002648 if (unlikely(is_vm_hugetlb_page(vma)))
2649 return hugetlb_fault(mm, vma, address, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652 pud = pud_alloc(mm, pgd, address);
2653 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002654 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655 pmd = pmd_alloc(mm, pud, address);
2656 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002657 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 pte = pte_alloc_map(mm, pmd, address);
2659 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002660 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661
Hugh Dickinsc74df322005-10-29 18:16:23 -07002662 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663}
2664
2665#ifndef __PAGETABLE_PUD_FOLDED
2666/*
2667 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002668 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002670int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002672 pud_t *new = pud_alloc_one(mm, address);
2673 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002674 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675
Hugh Dickins872fec12005-10-29 18:16:21 -07002676 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002677 if (pgd_present(*pgd)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 pud_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002679 else
2680 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07002681 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002682 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683}
2684#endif /* __PAGETABLE_PUD_FOLDED */
2685
2686#ifndef __PAGETABLE_PMD_FOLDED
2687/*
2688 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002689 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002691int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002693 pmd_t *new = pmd_alloc_one(mm, address);
2694 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002695 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696
Hugh Dickins872fec12005-10-29 18:16:21 -07002697 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07002699 if (pud_present(*pud)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 pmd_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002701 else
2702 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07002704 if (pgd_present(*pud)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 pmd_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002706 else
2707 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07002709 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002710 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711}
2712#endif /* __PAGETABLE_PMD_FOLDED */
2713
2714int make_pages_present(unsigned long addr, unsigned long end)
2715{
2716 int ret, len, write;
2717 struct vm_area_struct * vma;
2718
2719 vma = find_vma(current->mm, addr);
2720 if (!vma)
2721 return -1;
2722 write = (vma->vm_flags & VM_WRITE) != 0;
Eric Sesterhenn5bcb28b2006-03-26 18:30:52 +02002723 BUG_ON(addr >= end);
2724 BUG_ON(end > vma->vm_end);
Rolf Eike Beer68e116a2007-07-15 23:38:03 -07002725 len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726 ret = get_user_pages(current, current->mm, addr,
2727 len, write, 0, NULL, NULL);
2728 if (ret < 0)
2729 return ret;
2730 return ret == len ? 0 : -1;
2731}
2732
2733/*
2734 * Map a vmalloc()-space virtual address to the physical page.
2735 */
2736struct page * vmalloc_to_page(void * vmalloc_addr)
2737{
2738 unsigned long addr = (unsigned long) vmalloc_addr;
2739 struct page *page = NULL;
2740 pgd_t *pgd = pgd_offset_k(addr);
2741 pud_t *pud;
2742 pmd_t *pmd;
2743 pte_t *ptep, pte;
2744
2745 if (!pgd_none(*pgd)) {
2746 pud = pud_offset(pgd, addr);
2747 if (!pud_none(*pud)) {
2748 pmd = pmd_offset(pud, addr);
2749 if (!pmd_none(*pmd)) {
2750 ptep = pte_offset_map(pmd, addr);
2751 pte = *ptep;
2752 if (pte_present(pte))
2753 page = pte_page(pte);
2754 pte_unmap(ptep);
2755 }
2756 }
2757 }
2758 return page;
2759}
2760
2761EXPORT_SYMBOL(vmalloc_to_page);
2762
2763/*
2764 * Map a vmalloc()-space virtual address to the physical page frame number.
2765 */
2766unsigned long vmalloc_to_pfn(void * vmalloc_addr)
2767{
2768 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
2769}
2770
2771EXPORT_SYMBOL(vmalloc_to_pfn);
2772
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773#if !defined(__HAVE_ARCH_GATE_AREA)
2774
2775#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07002776static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777
2778static int __init gate_vma_init(void)
2779{
2780 gate_vma.vm_mm = NULL;
2781 gate_vma.vm_start = FIXADDR_USER_START;
2782 gate_vma.vm_end = FIXADDR_USER_END;
Roland McGrathb6558c42007-01-26 00:56:47 -08002783 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
2784 gate_vma.vm_page_prot = __P101;
Roland McGrathf47aef52007-01-26 00:56:49 -08002785 /*
2786 * Make sure the vDSO gets into every core dump.
2787 * Dumping its contents makes post-mortem fully interpretable later
2788 * without matching up the same kernel and hardware config to see
2789 * what PC values meant.
2790 */
2791 gate_vma.vm_flags |= VM_ALWAYSDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792 return 0;
2793}
2794__initcall(gate_vma_init);
2795#endif
2796
2797struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2798{
2799#ifdef AT_SYSINFO_EHDR
2800 return &gate_vma;
2801#else
2802 return NULL;
2803#endif
2804}
2805
2806int in_gate_area_no_task(unsigned long addr)
2807{
2808#ifdef AT_SYSINFO_EHDR
2809 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2810 return 1;
2811#endif
2812 return 0;
2813}
2814
2815#endif /* __HAVE_ARCH_GATE_AREA */
David Howells0ec76a12006-09-27 01:50:15 -07002816
2817/*
2818 * Access another process' address space.
2819 * Source/target buffer must be kernel space,
2820 * Do not walk the page table directly, use get_user_pages
2821 */
2822int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2823{
2824 struct mm_struct *mm;
2825 struct vm_area_struct *vma;
2826 struct page *page;
2827 void *old_buf = buf;
2828
2829 mm = get_task_mm(tsk);
2830 if (!mm)
2831 return 0;
2832
2833 down_read(&mm->mmap_sem);
2834 /* ignore errors, just check how much was sucessfully transfered */
2835 while (len) {
2836 int bytes, ret, offset;
2837 void *maddr;
2838
2839 ret = get_user_pages(tsk, mm, addr, 1,
2840 write, 1, &page, &vma);
2841 if (ret <= 0)
2842 break;
2843
2844 bytes = len;
2845 offset = addr & (PAGE_SIZE-1);
2846 if (bytes > PAGE_SIZE-offset)
2847 bytes = PAGE_SIZE-offset;
2848
2849 maddr = kmap(page);
2850 if (write) {
2851 copy_to_user_page(vma, page, addr,
2852 maddr + offset, buf, bytes);
2853 set_page_dirty_lock(page);
2854 } else {
2855 copy_from_user_page(vma, page, addr,
2856 buf, maddr + offset, bytes);
2857 }
2858 kunmap(page);
2859 page_cache_release(page);
2860 len -= bytes;
2861 buf += bytes;
2862 addr += bytes;
2863 }
2864 up_read(&mm->mmap_sem);
2865 mmput(mm);
2866
2867 return buf - old_buf;
2868}
Rusty Russell5992b6d2007-07-19 01:49:21 -07002869EXPORT_SYMBOL_GPL(access_process_vm);