blob: bc137751da7f39d14c618497d2e88b067553044e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/memory.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 */
6
7/*
8 * demand-loading started 01.12.91 - seems it is high on the list of
9 * things wanted, and it should be easy to implement. - Linus
10 */
11
12/*
13 * Ok, demand-loading was easy, shared pages a little bit tricker. Shared
14 * pages started 02.12.91, seems to work. - Linus.
15 *
16 * Tested sharing by executing about 30 /bin/sh: under the old kernel it
17 * would have taken more than the 6M I have free, but it worked well as
18 * far as I could see.
19 *
20 * Also corrected some "invalidate()"s - I wasn't doing enough of them.
21 */
22
23/*
24 * Real VM (paging to/from disk) started 18.12.91. Much more work and
25 * thought has to go into this. Oh, well..
26 * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why.
27 * Found it. Everything seems to work now.
28 * 20.12.91 - Ok, making the swap-device changeable like the root.
29 */
30
31/*
32 * 05.04.94 - Multi-page memory management added for v1.1.
33 * Idea by Alex Bligh (alex@cconcepts.co.uk)
34 *
35 * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG
36 * (Gerhard.Wichert@pdb.siemens.de)
37 *
38 * Aug/Sep 2004 Changed to four level page tables (Andi Kleen)
39 */
40
41#include <linux/kernel_stat.h>
42#include <linux/mm.h>
43#include <linux/hugetlb.h>
44#include <linux/mman.h>
45#include <linux/swap.h>
46#include <linux/highmem.h>
47#include <linux/pagemap.h>
48#include <linux/rmap.h>
49#include <linux/module.h>
Shailabh Nagar0ff92242006-07-14 00:24:37 -070050#include <linux/delayacct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include <linux/init.h>
Peter Zijlstraedc79b22006-09-25 23:30:58 -070052#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
54#include <asm/pgalloc.h>
55#include <asm/uaccess.h>
56#include <asm/tlb.h>
57#include <asm/tlbflush.h>
58#include <asm/pgtable.h>
59
60#include <linux/swapops.h>
61#include <linux/elf.h>
62
Andy Whitcroftd41dee32005-06-23 00:07:54 -070063#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -070064/* use the per-pgdat data instead for discontigmem - mbligh */
65unsigned long max_mapnr;
66struct page *mem_map;
67
68EXPORT_SYMBOL(max_mapnr);
69EXPORT_SYMBOL(mem_map);
70#endif
71
72unsigned long num_physpages;
73/*
74 * A number of key systems in x86 including ioremap() rely on the assumption
75 * that high_memory defines the upper bound on direct map memory, then end
76 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
77 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
78 * and ZONE_HIGHMEM.
79 */
80void * high_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
82EXPORT_SYMBOL(num_physpages);
83EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Andi Kleena62eaf12006-02-16 23:41:58 +010085int randomize_va_space __read_mostly = 1;
86
87static int __init disable_randmaps(char *s)
88{
89 randomize_va_space = 0;
OGAWA Hirofumi9b410462006-03-31 02:30:33 -080090 return 1;
Andi Kleena62eaf12006-02-16 23:41:58 +010091}
92__setup("norandmaps", disable_randmaps);
93
94
Linus Torvalds1da177e2005-04-16 15:20:36 -070095/*
96 * If a p?d_bad entry is found while walking page tables, report
97 * the error, before resetting entry to p?d_none. Usually (but
98 * very seldom) called out from the p?d_none_or_clear_bad macros.
99 */
100
101void pgd_clear_bad(pgd_t *pgd)
102{
103 pgd_ERROR(*pgd);
104 pgd_clear(pgd);
105}
106
107void pud_clear_bad(pud_t *pud)
108{
109 pud_ERROR(*pud);
110 pud_clear(pud);
111}
112
113void pmd_clear_bad(pmd_t *pmd)
114{
115 pmd_ERROR(*pmd);
116 pmd_clear(pmd);
117}
118
119/*
120 * Note: this doesn't free the actual pages themselves. That
121 * has been handled earlier when unmapping all the memory regions.
122 */
Hugh Dickinse0da3822005-04-19 13:29:15 -0700123static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124{
Hugh Dickinse0da3822005-04-19 13:29:15 -0700125 struct page *page = pmd_page(*pmd);
126 pmd_clear(pmd);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700127 pte_lock_deinit(page);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700128 pte_free_tlb(tlb, page);
Christoph Lameterdf849a12006-06-30 01:55:38 -0700129 dec_zone_page_state(page, NR_PAGETABLE);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700130 tlb->mm->nr_ptes--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131}
132
Hugh Dickinse0da3822005-04-19 13:29:15 -0700133static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
134 unsigned long addr, unsigned long end,
135 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136{
137 pmd_t *pmd;
138 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700139 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140
Hugh Dickinse0da3822005-04-19 13:29:15 -0700141 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 do {
144 next = pmd_addr_end(addr, end);
145 if (pmd_none_or_clear_bad(pmd))
146 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700147 free_pte_range(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 } while (pmd++, addr = next, addr != end);
149
Hugh Dickinse0da3822005-04-19 13:29:15 -0700150 start &= PUD_MASK;
151 if (start < floor)
152 return;
153 if (ceiling) {
154 ceiling &= PUD_MASK;
155 if (!ceiling)
156 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700158 if (end - 1 > ceiling - 1)
159 return;
160
161 pmd = pmd_offset(pud, start);
162 pud_clear(pud);
163 pmd_free_tlb(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164}
165
Hugh Dickinse0da3822005-04-19 13:29:15 -0700166static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
167 unsigned long addr, unsigned long end,
168 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169{
170 pud_t *pud;
171 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700172 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173
Hugh Dickinse0da3822005-04-19 13:29:15 -0700174 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 pud = pud_offset(pgd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 do {
177 next = pud_addr_end(addr, end);
178 if (pud_none_or_clear_bad(pud))
179 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700180 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 } while (pud++, addr = next, addr != end);
182
Hugh Dickinse0da3822005-04-19 13:29:15 -0700183 start &= PGDIR_MASK;
184 if (start < floor)
185 return;
186 if (ceiling) {
187 ceiling &= PGDIR_MASK;
188 if (!ceiling)
189 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700191 if (end - 1 > ceiling - 1)
192 return;
193
194 pud = pud_offset(pgd, start);
195 pgd_clear(pgd);
196 pud_free_tlb(tlb, pud);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197}
198
199/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700200 * This function frees user-level page tables of a process.
201 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 * Must be called with pagetable lock held.
203 */
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700204void free_pgd_range(struct mmu_gather **tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700205 unsigned long addr, unsigned long end,
206 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207{
208 pgd_t *pgd;
209 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700210 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
Hugh Dickinse0da3822005-04-19 13:29:15 -0700212 /*
213 * The next few lines have given us lots of grief...
214 *
215 * Why are we testing PMD* at this top level? Because often
216 * there will be no work to do at all, and we'd prefer not to
217 * go all the way down to the bottom just to discover that.
218 *
219 * Why all these "- 1"s? Because 0 represents both the bottom
220 * of the address space and the top of it (using -1 for the
221 * top wouldn't help much: the masks would do the wrong thing).
222 * The rule is that addr 0 and floor 0 refer to the bottom of
223 * the address space, but end 0 and ceiling 0 refer to the top
224 * Comparisons need to use "end - 1" and "ceiling - 1" (though
225 * that end 0 case should be mythical).
226 *
227 * Wherever addr is brought up or ceiling brought down, we must
228 * be careful to reject "the opposite 0" before it confuses the
229 * subsequent tests. But what about where end is brought down
230 * by PMD_SIZE below? no, end can't go down to 0 there.
231 *
232 * Whereas we round start (addr) and ceiling down, by different
233 * masks at different levels, in order to test whether a table
234 * now has no other vmas using it, so can be freed, we don't
235 * bother to round floor or end up - the tests don't need that.
236 */
237
238 addr &= PMD_MASK;
239 if (addr < floor) {
240 addr += PMD_SIZE;
241 if (!addr)
242 return;
243 }
244 if (ceiling) {
245 ceiling &= PMD_MASK;
246 if (!ceiling)
247 return;
248 }
249 if (end - 1 > ceiling - 1)
250 end -= PMD_SIZE;
251 if (addr > end - 1)
252 return;
253
254 start = addr;
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700255 pgd = pgd_offset((*tlb)->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 do {
257 next = pgd_addr_end(addr, end);
258 if (pgd_none_or_clear_bad(pgd))
259 continue;
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700260 free_pud_range(*tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700262}
263
264void free_pgtables(struct mmu_gather **tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700265 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700266{
267 while (vma) {
268 struct vm_area_struct *next = vma->vm_next;
269 unsigned long addr = vma->vm_start;
270
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700271 /*
272 * Hide vma from rmap and vmtruncate before freeing pgtables
273 */
274 anon_vma_unlink(vma);
275 unlink_file_vma(vma);
276
David Gibson9da61ae2006-03-22 00:08:57 -0800277 if (is_vm_hugetlb_page(vma)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700278 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700279 floor, next? next->vm_start: ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700280 } else {
281 /*
282 * Optimization: gather nearby vmas into one call down
283 */
284 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
David Gibson48669202006-03-22 00:08:58 -0800285 && !is_vm_hugetlb_page(next)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700286 vma = next;
287 next = vma->vm_next;
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700288 anon_vma_unlink(vma);
289 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700290 }
291 free_pgd_range(tlb, addr, vma->vm_end,
292 floor, next? next->vm_start: ceiling);
293 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700294 vma = next;
295 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296}
297
Hugh Dickins1bb36302005-10-29 18:16:22 -0700298int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299{
Hugh Dickinsc74df322005-10-29 18:16:23 -0700300 struct page *new = pte_alloc_one(mm, address);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700301 if (!new)
302 return -ENOMEM;
303
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700304 pte_lock_init(new);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700305 spin_lock(&mm->page_table_lock);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700306 if (pmd_present(*pmd)) { /* Another has populated it */
307 pte_lock_deinit(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700308 pte_free(new);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700309 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 mm->nr_ptes++;
Christoph Lameterdf849a12006-06-30 01:55:38 -0700311 inc_zone_page_state(new, NR_PAGETABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 pmd_populate(mm, pmd, new);
313 }
Hugh Dickinsc74df322005-10-29 18:16:23 -0700314 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700315 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316}
317
Hugh Dickins1bb36302005-10-29 18:16:22 -0700318int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700320 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
321 if (!new)
322 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323
Hugh Dickins1bb36302005-10-29 18:16:22 -0700324 spin_lock(&init_mm.page_table_lock);
325 if (pmd_present(*pmd)) /* Another has populated it */
326 pte_free_kernel(new);
327 else
328 pmd_populate_kernel(&init_mm, pmd, new);
329 spin_unlock(&init_mm.page_table_lock);
330 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331}
332
Hugh Dickinsae859762005-10-29 18:16:05 -0700333static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
334{
335 if (file_rss)
336 add_mm_counter(mm, file_rss, file_rss);
337 if (anon_rss)
338 add_mm_counter(mm, anon_rss, anon_rss);
339}
340
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800342 * This function is called to print an error when a bad pte
343 * is found. For example, we might have a PFN-mapped pte in
344 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700345 *
346 * The calling function must still handle the error.
347 */
348void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr)
349{
350 printk(KERN_ERR "Bad pte = %08llx, process = %s, "
351 "vm_flags = %lx, vaddr = %lx\n",
352 (long long)pte_val(pte),
353 (vma->vm_mm == current->mm ? current->comm : "???"),
354 vma->vm_flags, vaddr);
355 dump_stack();
356}
357
Linus Torvalds67121172005-12-11 20:38:17 -0800358static inline int is_cow_mapping(unsigned int flags)
359{
360 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
361}
362
Nick Pigginb5810032005-10-29 18:16:12 -0700363/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800364 * This function gets the "struct page" associated with a pte.
365 *
366 * NOTE! Some mappings do not have "struct pages". A raw PFN mapping
367 * will have each page table entry just pointing to a raw page frame
368 * number, and as far as the VM layer is concerned, those do not have
369 * pages associated with them - even if the PFN might point to memory
370 * that otherwise is perfectly fine and has a "struct page".
371 *
372 * The way we recognize those mappings is through the rules set up
373 * by "remap_pfn_range()": the vma will have the VM_PFNMAP bit set,
374 * and the vm_pgoff will point to the first PFN mapped: thus every
375 * page that is a raw mapping will always honor the rule
376 *
377 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
378 *
379 * and if that isn't true, the page has been COW'ed (in which case it
380 * _does_ have a "struct page" associated with it even if it is in a
381 * VM_PFNMAP range).
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800382 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800383struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800384{
Linus Torvalds6aab3412005-11-28 14:34:23 -0800385 unsigned long pfn = pte_pfn(pte);
386
Nick Pigginb7ab7952006-03-22 00:08:42 -0800387 if (unlikely(vma->vm_flags & VM_PFNMAP)) {
Linus Torvalds6aab3412005-11-28 14:34:23 -0800388 unsigned long off = (addr - vma->vm_start) >> PAGE_SHIFT;
389 if (pfn == vma->vm_pgoff + off)
390 return NULL;
Linus Torvalds67121172005-12-11 20:38:17 -0800391 if (!is_cow_mapping(vma->vm_flags))
Linus Torvaldsfb155c12005-12-11 19:46:02 -0800392 return NULL;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800393 }
394
Carsten Otte97231982008-01-17 15:21:17 -0800395#ifdef CONFIG_DEBUG_VM
Nick Piggin315ab192006-03-25 16:20:22 +0100396 /*
397 * Add some anal sanity checks for now. Eventually,
398 * we should just do "return pfn_to_page(pfn)", but
399 * in the meantime we check that we get a valid pfn,
400 * and that the resulting page looks ok.
401 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800402 if (unlikely(!pfn_valid(pfn))) {
403 print_bad_pte(vma, pte, addr);
404 return NULL;
405 }
Carsten Otte97231982008-01-17 15:21:17 -0800406#endif
Linus Torvalds6aab3412005-11-28 14:34:23 -0800407
408 /*
409 * NOTE! We still have PageReserved() pages in the page
410 * tables.
411 *
412 * The PAGE_ZERO() pages and various VDSO mappings can
413 * cause them to exist.
414 */
415 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800416}
417
418/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 * copy one vm_area from one task to the other. Assumes the page tables
420 * already present in the new task to be cleared in the whole range
421 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 */
423
Hugh Dickins8c103762005-10-29 18:16:13 -0700424static inline void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700426 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700427 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428{
Nick Pigginb5810032005-10-29 18:16:12 -0700429 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 pte_t pte = *src_pte;
431 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432
433 /* pte contains position in swap or file, so copy. */
434 if (unlikely(!pte_present(pte))) {
435 if (!pte_file(pte)) {
Christoph Lameter06972122006-06-23 02:03:35 -0700436 swp_entry_t entry = pte_to_swp_entry(pte);
437
438 swap_duplicate(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 /* make sure dst_mm is on swapoff's mmlist. */
440 if (unlikely(list_empty(&dst_mm->mmlist))) {
441 spin_lock(&mmlist_lock);
Hugh Dickinsf412ac02005-10-29 18:16:41 -0700442 if (list_empty(&dst_mm->mmlist))
443 list_add(&dst_mm->mmlist,
444 &src_mm->mmlist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 spin_unlock(&mmlist_lock);
446 }
Christoph Lameter06972122006-06-23 02:03:35 -0700447 if (is_write_migration_entry(entry) &&
448 is_cow_mapping(vm_flags)) {
449 /*
450 * COW mappings require pages in both parent
451 * and child to be set to read.
452 */
453 make_migration_entry_read(&entry);
454 pte = swp_entry_to_pte(entry);
455 set_pte_at(src_mm, addr, src_pte, pte);
456 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700458 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 }
460
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 /*
462 * If it's a COW mapping, write protect it both
463 * in the parent and the child
464 */
Linus Torvalds67121172005-12-11 20:38:17 -0800465 if (is_cow_mapping(vm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 ptep_set_wrprotect(src_mm, addr, src_pte);
Zachary Amsden3dc90792006-09-30 23:29:30 -0700467 pte = pte_wrprotect(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 }
469
470 /*
471 * If it's a shared mapping, mark it clean in
472 * the child
473 */
474 if (vm_flags & VM_SHARED)
475 pte = pte_mkclean(pte);
476 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800477
478 page = vm_normal_page(vma, addr, pte);
479 if (page) {
480 get_page(page);
Nick Pigginc97a9e12007-05-16 22:11:21 -0700481 page_dup_rmap(page, vma, addr);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800482 rss[!!PageAnon(page)]++;
483 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700484
485out_set_pte:
486 set_pte_at(dst_mm, addr, dst_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487}
488
489static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
490 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
491 unsigned long addr, unsigned long end)
492{
493 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700494 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -0700495 int progress = 0;
Hugh Dickins8c103762005-10-29 18:16:13 -0700496 int rss[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
498again:
Hugh Dickinsae859762005-10-29 18:16:05 -0700499 rss[1] = rss[0] = 0;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700500 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 if (!dst_pte)
502 return -ENOMEM;
503 src_pte = pte_offset_map_nested(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700504 src_ptl = pte_lockptr(src_mm, src_pmd);
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700505 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700506 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 do {
509 /*
510 * We are holding two locks at this point - either of them
511 * could generate latencies in another task on another CPU.
512 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700513 if (progress >= 32) {
514 progress = 0;
515 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +0100516 spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700517 break;
518 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 if (pte_none(*src_pte)) {
520 progress++;
521 continue;
522 }
Hugh Dickins8c103762005-10-29 18:16:13 -0700523 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 progress += 8;
525 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700527 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -0700528 spin_unlock(src_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 pte_unmap_nested(src_pte - 1);
Hugh Dickinsae859762005-10-29 18:16:05 -0700530 add_mm_rss(dst_mm, rss[0], rss[1]);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700531 pte_unmap_unlock(dst_pte - 1, dst_ptl);
532 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 if (addr != end)
534 goto again;
535 return 0;
536}
537
538static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
539 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
540 unsigned long addr, unsigned long end)
541{
542 pmd_t *src_pmd, *dst_pmd;
543 unsigned long next;
544
545 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
546 if (!dst_pmd)
547 return -ENOMEM;
548 src_pmd = pmd_offset(src_pud, addr);
549 do {
550 next = pmd_addr_end(addr, end);
551 if (pmd_none_or_clear_bad(src_pmd))
552 continue;
553 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
554 vma, addr, next))
555 return -ENOMEM;
556 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
557 return 0;
558}
559
560static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
561 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
562 unsigned long addr, unsigned long end)
563{
564 pud_t *src_pud, *dst_pud;
565 unsigned long next;
566
567 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
568 if (!dst_pud)
569 return -ENOMEM;
570 src_pud = pud_offset(src_pgd, addr);
571 do {
572 next = pud_addr_end(addr, end);
573 if (pud_none_or_clear_bad(src_pud))
574 continue;
575 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
576 vma, addr, next))
577 return -ENOMEM;
578 } while (dst_pud++, src_pud++, addr = next, addr != end);
579 return 0;
580}
581
582int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
583 struct vm_area_struct *vma)
584{
585 pgd_t *src_pgd, *dst_pgd;
586 unsigned long next;
587 unsigned long addr = vma->vm_start;
588 unsigned long end = vma->vm_end;
589
Nick Piggind9928952005-08-28 16:49:11 +1000590 /*
591 * Don't copy ptes where a page fault will fill them correctly.
592 * Fork becomes much lighter when there are big shared or private
593 * readonly mappings. The tradeoff is that copy_page_range is more
594 * efficient than faulting.
595 */
Linus Torvalds4d7672b2005-12-16 10:21:23 -0800596 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
Nick Piggind9928952005-08-28 16:49:11 +1000597 if (!vma->anon_vma)
598 return 0;
599 }
600
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 if (is_vm_hugetlb_page(vma))
602 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
603
604 dst_pgd = pgd_offset(dst_mm, addr);
605 src_pgd = pgd_offset(src_mm, addr);
606 do {
607 next = pgd_addr_end(addr, end);
608 if (pgd_none_or_clear_bad(src_pgd))
609 continue;
610 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
611 vma, addr, next))
612 return -ENOMEM;
613 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
614 return 0;
615}
616
Robin Holt51c6f662005-11-13 16:06:42 -0800617static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700618 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800620 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621{
Nick Pigginb5810032005-10-29 18:16:12 -0700622 struct mm_struct *mm = tlb->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 pte_t *pte;
Hugh Dickins508034a2005-10-29 18:16:30 -0700624 spinlock_t *ptl;
Hugh Dickinsae859762005-10-29 18:16:05 -0700625 int file_rss = 0;
626 int anon_rss = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
Hugh Dickins508034a2005-10-29 18:16:30 -0700628 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700629 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 do {
631 pte_t ptent = *pte;
Robin Holt51c6f662005-11-13 16:06:42 -0800632 if (pte_none(ptent)) {
633 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800635 }
Hugh Dickins6f5e6b92006-03-16 23:04:09 -0800636
637 (*zap_work) -= PAGE_SIZE;
638
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800640 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -0800641
Linus Torvalds6aab3412005-11-28 14:34:23 -0800642 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 if (unlikely(details) && page) {
644 /*
645 * unmap_shared_mapping_pages() wants to
646 * invalidate cache without truncating:
647 * unmap shared but keep private pages.
648 */
649 if (details->check_mapping &&
650 details->check_mapping != page->mapping)
651 continue;
652 /*
653 * Each page->index must be checked when
654 * invalidating or truncating nonlinear.
655 */
656 if (details->nonlinear_vma &&
657 (page->index < details->first_index ||
658 page->index > details->last_index))
659 continue;
660 }
Nick Pigginb5810032005-10-29 18:16:12 -0700661 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -0700662 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 tlb_remove_tlb_entry(tlb, pte, addr);
664 if (unlikely(!page))
665 continue;
666 if (unlikely(details) && details->nonlinear_vma
667 && linear_page_index(details->nonlinear_vma,
668 addr) != page->index)
Nick Pigginb5810032005-10-29 18:16:12 -0700669 set_pte_at(mm, addr, pte,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 pgoff_to_pte(page->index));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 if (PageAnon(page))
Hugh Dickins86d912f2005-10-29 18:16:14 -0700672 anon_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700673 else {
674 if (pte_dirty(ptent))
675 set_page_dirty(page);
676 if (pte_young(ptent))
Ken Chendaa88c82007-02-10 01:43:18 -0800677 SetPageReferenced(page);
Hugh Dickins86d912f2005-10-29 18:16:14 -0700678 file_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700679 }
Nick Piggin7de6b802006-12-22 01:09:33 -0800680 page_remove_rmap(page, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 tlb_remove_page(tlb, page);
682 continue;
683 }
684 /*
685 * If details->check_mapping, we leave swap entries;
686 * if details->nonlinear_vma, we leave file entries.
687 */
688 if (unlikely(details))
689 continue;
690 if (!pte_file(ptent))
691 free_swap_and_cache(pte_to_swp_entry(ptent));
Zachary Amsden9888a1c2006-09-30 23:29:31 -0700692 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
Robin Holt51c6f662005-11-13 16:06:42 -0800693 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
Hugh Dickinsae859762005-10-29 18:16:05 -0700694
Hugh Dickins86d912f2005-10-29 18:16:14 -0700695 add_mm_rss(mm, file_rss, anon_rss);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700696 arch_leave_lazy_mmu_mode();
Hugh Dickins508034a2005-10-29 18:16:30 -0700697 pte_unmap_unlock(pte - 1, ptl);
Robin Holt51c6f662005-11-13 16:06:42 -0800698
699 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700}
701
Robin Holt51c6f662005-11-13 16:06:42 -0800702static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700703 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800705 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706{
707 pmd_t *pmd;
708 unsigned long next;
709
710 pmd = pmd_offset(pud, addr);
711 do {
712 next = pmd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800713 if (pmd_none_or_clear_bad(pmd)) {
714 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800716 }
717 next = zap_pte_range(tlb, vma, pmd, addr, next,
718 zap_work, details);
719 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
720
721 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722}
723
Robin Holt51c6f662005-11-13 16:06:42 -0800724static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700725 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800727 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728{
729 pud_t *pud;
730 unsigned long next;
731
732 pud = pud_offset(pgd, addr);
733 do {
734 next = pud_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800735 if (pud_none_or_clear_bad(pud)) {
736 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800738 }
739 next = zap_pmd_range(tlb, vma, pud, addr, next,
740 zap_work, details);
741 } while (pud++, addr = next, (addr != end && *zap_work > 0));
742
743 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744}
745
Robin Holt51c6f662005-11-13 16:06:42 -0800746static unsigned long unmap_page_range(struct mmu_gather *tlb,
747 struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800749 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750{
751 pgd_t *pgd;
752 unsigned long next;
753
754 if (details && !details->check_mapping && !details->nonlinear_vma)
755 details = NULL;
756
757 BUG_ON(addr >= end);
758 tlb_start_vma(tlb, vma);
759 pgd = pgd_offset(vma->vm_mm, addr);
760 do {
761 next = pgd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800762 if (pgd_none_or_clear_bad(pgd)) {
763 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800765 }
766 next = zap_pud_range(tlb, vma, pgd, addr, next,
767 zap_work, details);
768 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 tlb_end_vma(tlb, vma);
Robin Holt51c6f662005-11-13 16:06:42 -0800770
771 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772}
773
774#ifdef CONFIG_PREEMPT
775# define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
776#else
777/* No preempt: go for improved straight-line efficiency */
778# define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
779#endif
780
781/**
782 * unmap_vmas - unmap a range of memory covered by a list of vma's
783 * @tlbp: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 * @vma: the starting vma
785 * @start_addr: virtual address at which to start unmapping
786 * @end_addr: virtual address at which to end unmapping
787 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
788 * @details: details of nonlinear truncation or shared cache invalidation
789 *
Hugh Dickinsee39b372005-04-19 13:29:15 -0700790 * Returns the end address of the unmapping (restart addr if interrupted).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700792 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700794 * We aim to not hold locks for too long (for scheduling latency reasons).
795 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 * return the ending mmu_gather to the caller.
797 *
798 * Only addresses between `start' and `end' will be unmapped.
799 *
800 * The VMA list must be sorted in ascending virtual address order.
801 *
802 * unmap_vmas() assumes that the caller will flush the whole unmapped address
803 * range after unmap_vmas() returns. So the only responsibility here is to
804 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
805 * drops the lock and schedules.
806 */
Hugh Dickins508034a2005-10-29 18:16:30 -0700807unsigned long unmap_vmas(struct mmu_gather **tlbp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 struct vm_area_struct *vma, unsigned long start_addr,
809 unsigned long end_addr, unsigned long *nr_accounted,
810 struct zap_details *details)
811{
Robin Holt51c6f662005-11-13 16:06:42 -0800812 long zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
814 int tlb_start_valid = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -0700815 unsigned long start = start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700817 int fullmm = (*tlbp)->fullmm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
819 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 unsigned long end;
821
822 start = max(vma->vm_start, start_addr);
823 if (start >= vma->vm_end)
824 continue;
825 end = min(vma->vm_end, end_addr);
826 if (end <= vma->vm_start)
827 continue;
828
829 if (vma->vm_flags & VM_ACCOUNT)
830 *nr_accounted += (end - start) >> PAGE_SHIFT;
831
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 while (start != end) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 if (!tlb_start_valid) {
834 tlb_start = start;
835 tlb_start_valid = 1;
836 }
837
Robin Holt51c6f662005-11-13 16:06:42 -0800838 if (unlikely(is_vm_hugetlb_page(vma))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 unmap_hugepage_range(vma, start, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800840 zap_work -= (end - start) /
841 (HPAGE_SIZE / PAGE_SIZE);
842 start = end;
843 } else
844 start = unmap_page_range(*tlbp, vma,
845 start, end, &zap_work, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846
Robin Holt51c6f662005-11-13 16:06:42 -0800847 if (zap_work > 0) {
848 BUG_ON(start != end);
849 break;
850 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851
852 tlb_finish_mmu(*tlbp, tlb_start, start);
853
854 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +0100855 (i_mmap_lock && spin_needbreak(i_mmap_lock))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 if (i_mmap_lock) {
Hugh Dickins508034a2005-10-29 18:16:30 -0700857 *tlbp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 goto out;
859 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 }
862
Hugh Dickins508034a2005-10-29 18:16:30 -0700863 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 tlb_start_valid = 0;
Robin Holt51c6f662005-11-13 16:06:42 -0800865 zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 }
867 }
868out:
Hugh Dickinsee39b372005-04-19 13:29:15 -0700869 return start; /* which is now the end (or restart) address */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870}
871
872/**
873 * zap_page_range - remove user pages in a given range
874 * @vma: vm_area_struct holding the applicable pages
875 * @address: starting address of pages to zap
876 * @size: number of bytes to zap
877 * @details: details of nonlinear truncation or shared cache invalidation
878 */
Hugh Dickinsee39b372005-04-19 13:29:15 -0700879unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 unsigned long size, struct zap_details *details)
881{
882 struct mm_struct *mm = vma->vm_mm;
883 struct mmu_gather *tlb;
884 unsigned long end = address + size;
885 unsigned long nr_accounted = 0;
886
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -0700889 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -0700890 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
891 if (tlb)
892 tlb_finish_mmu(tlb, address, end);
Hugh Dickinsee39b372005-04-19 13:29:15 -0700893 return end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894}
895
896/*
897 * Do a quick page-table lookup for a single page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800899struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700900 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901{
902 pgd_t *pgd;
903 pud_t *pud;
904 pmd_t *pmd;
905 pte_t *ptep, pte;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700906 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 struct page *page;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800908 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700910 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
911 if (!IS_ERR(page)) {
912 BUG_ON(flags & FOLL_GET);
913 goto out;
914 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700916 page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 pgd = pgd_offset(mm, address);
918 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700919 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920
921 pud = pud_offset(pgd, address);
922 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700923 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924
925 pmd = pmd_offset(pud, address);
926 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700927 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700929 if (pmd_huge(*pmd)) {
930 BUG_ON(flags & FOLL_GET);
931 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
932 goto out;
933 }
934
935 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 if (!ptep)
937 goto out;
938
939 pte = *ptep;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700940 if (!pte_present(pte))
941 goto unlock;
942 if ((flags & FOLL_WRITE) && !pte_write(pte))
943 goto unlock;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800944 page = vm_normal_page(vma, address, pte);
945 if (unlikely(!page))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700946 goto unlock;
947
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700948 if (flags & FOLL_GET)
949 get_page(page);
950 if (flags & FOLL_TOUCH) {
951 if ((flags & FOLL_WRITE) &&
952 !pte_dirty(pte) && !PageDirty(page))
953 set_page_dirty(page);
954 mark_page_accessed(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700956unlock:
957 pte_unmap_unlock(ptep, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958out:
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700959 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700961no_page_table:
962 /*
963 * When core dumping an enormous anonymous area that nobody
964 * has touched so far, we don't want to allocate page tables.
965 */
966 if (flags & FOLL_ANON) {
Nick Piggin557ed1f2007-10-16 01:24:40 -0700967 page = ZERO_PAGE(0);
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700968 if (flags & FOLL_GET)
969 get_page(page);
970 BUG_ON(flags & FOLL_WRITE);
971 }
972 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973}
974
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
976 unsigned long start, int len, int write, int force,
977 struct page **pages, struct vm_area_struct **vmas)
978{
979 int i;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700980 unsigned int vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981
982 /*
983 * Require read or write permissions.
984 * If 'force' is set, we only require the "MAY" flags.
985 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700986 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
987 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 i = 0;
989
990 do {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700991 struct vm_area_struct *vma;
992 unsigned int foll_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
994 vma = find_extend_vma(mm, start);
995 if (!vma && in_gate_area(tsk, start)) {
996 unsigned long pg = start & PAGE_MASK;
997 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
998 pgd_t *pgd;
999 pud_t *pud;
1000 pmd_t *pmd;
1001 pte_t *pte;
1002 if (write) /* user gate pages are read-only */
1003 return i ? : -EFAULT;
1004 if (pg > TASK_SIZE)
1005 pgd = pgd_offset_k(pg);
1006 else
1007 pgd = pgd_offset_gate(mm, pg);
1008 BUG_ON(pgd_none(*pgd));
1009 pud = pud_offset(pgd, pg);
1010 BUG_ON(pud_none(*pud));
1011 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001012 if (pmd_none(*pmd))
1013 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001015 if (pte_none(*pte)) {
1016 pte_unmap(pte);
1017 return i ? : -EFAULT;
1018 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 if (pages) {
Nick Pigginfa2a4552005-11-29 18:43:17 +11001020 struct page *page = vm_normal_page(gate_vma, start, *pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001021 pages[i] = page;
1022 if (page)
1023 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 }
1025 pte_unmap(pte);
1026 if (vmas)
1027 vmas[i] = gate_vma;
1028 i++;
1029 start += PAGE_SIZE;
1030 len--;
1031 continue;
1032 }
1033
Linus Torvalds1ff80382005-12-12 16:24:33 -08001034 if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001035 || !(vm_flags & vma->vm_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 return i ? : -EFAULT;
1037
1038 if (is_vm_hugetlb_page(vma)) {
1039 i = follow_hugetlb_page(mm, vma, pages, vmas,
Adam Litke5b23dbe2007-11-14 16:59:33 -08001040 &start, &len, i, write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 continue;
1042 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001043
1044 foll_flags = FOLL_TOUCH;
1045 if (pages)
1046 foll_flags |= FOLL_GET;
1047 if (!write && !(vma->vm_flags & VM_LOCKED) &&
Nick Piggin54cb8822007-07-19 01:46:59 -07001048 (!vma->vm_ops || (!vma->vm_ops->nopage &&
1049 !vma->vm_ops->fault)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001050 foll_flags |= FOLL_ANON;
1051
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 do {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001053 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054
Ethan Solomita462e00c2007-07-15 23:38:16 -07001055 /*
1056 * If tsk is ooming, cut off its access to large memory
1057 * allocations. It has a pending SIGKILL, but it can't
1058 * be processed until returning to user space.
1059 */
1060 if (unlikely(test_tsk_thread_flag(tsk, TIF_MEMDIE)))
1061 return -ENOMEM;
1062
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001063 if (write)
1064 foll_flags |= FOLL_WRITE;
1065
1066 cond_resched();
Linus Torvalds6aab3412005-11-28 14:34:23 -08001067 while (!(page = follow_page(vma, start, foll_flags))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001068 int ret;
Nick Piggin83c54072007-07-19 01:47:05 -07001069 ret = handle_mm_fault(mm, vma, start,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001070 foll_flags & FOLL_WRITE);
Nick Piggin83c54072007-07-19 01:47:05 -07001071 if (ret & VM_FAULT_ERROR) {
1072 if (ret & VM_FAULT_OOM)
1073 return i ? i : -ENOMEM;
1074 else if (ret & VM_FAULT_SIGBUS)
1075 return i ? i : -EFAULT;
1076 BUG();
1077 }
1078 if (ret & VM_FAULT_MAJOR)
1079 tsk->maj_flt++;
1080 else
1081 tsk->min_flt++;
1082
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001083 /*
Nick Piggin83c54072007-07-19 01:47:05 -07001084 * The VM_FAULT_WRITE bit tells us that
1085 * do_wp_page has broken COW when necessary,
1086 * even if maybe_mkwrite decided not to set
1087 * pte_write. We can thus safely do subsequent
1088 * page lookups as if they were reads.
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001089 */
1090 if (ret & VM_FAULT_WRITE)
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001091 foll_flags &= ~FOLL_WRITE;
Nick Piggin83c54072007-07-19 01:47:05 -07001092
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07001093 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 }
1095 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001096 pages[i] = page;
James Bottomley03beb072006-03-26 01:36:57 -08001097
Russell Kinga6f36be2006-12-30 22:24:19 +00001098 flush_anon_page(vma, page, start);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001099 flush_dcache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 }
1101 if (vmas)
1102 vmas[i] = vma;
1103 i++;
1104 start += PAGE_SIZE;
1105 len--;
Hugh Dickins08ef4722005-06-21 17:15:10 -07001106 } while (len && start < vma->vm_end);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001107 } while (len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 return i;
1109}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110EXPORT_SYMBOL(get_user_pages);
1111
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001112pte_t * fastcall get_locked_pte(struct mm_struct *mm, unsigned long addr, spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001113{
1114 pgd_t * pgd = pgd_offset(mm, addr);
1115 pud_t * pud = pud_alloc(mm, pgd, addr);
1116 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001117 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001118 if (pmd)
1119 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1120 }
1121 return NULL;
1122}
1123
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001125 * This is the old fallback for page remapping.
1126 *
1127 * For historical reasons, it only allows reserved pages. Only
1128 * old drivers should use this, and they needed to mark their
1129 * pages reserved for the old functions anyway.
1130 */
1131static int insert_page(struct mm_struct *mm, unsigned long addr, struct page *page, pgprot_t prot)
1132{
1133 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001134 pte_t *pte;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001135 spinlock_t *ptl;
1136
1137 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001138 if (PageAnon(page))
Linus Torvalds238f58d2005-11-29 13:01:56 -08001139 goto out;
1140 retval = -ENOMEM;
1141 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001142 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001143 if (!pte)
1144 goto out;
1145 retval = -EBUSY;
1146 if (!pte_none(*pte))
1147 goto out_unlock;
1148
1149 /* Ok, finally just insert the thing.. */
1150 get_page(page);
1151 inc_mm_counter(mm, file_rss);
1152 page_add_file_rmap(page);
1153 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1154
1155 retval = 0;
1156out_unlock:
1157 pte_unmap_unlock(pte, ptl);
1158out:
1159 return retval;
1160}
1161
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001162/**
1163 * vm_insert_page - insert single page into user vma
1164 * @vma: user vma to map to
1165 * @addr: target user address of this page
1166 * @page: source kernel page
1167 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001168 * This allows drivers to insert individual pages they've allocated
1169 * into a user vma.
1170 *
1171 * The page has to be a nice clean _individual_ kernel allocation.
1172 * If you allocate a compound page, you need to have marked it as
1173 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001174 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001175 *
1176 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1177 * took an arbitrary page protection parameter. This doesn't allow
1178 * that. Your vma protection will have to be set up correctly, which
1179 * means that if you want a shared writable mapping, you'd better
1180 * ask for a shared writable mapping!
1181 *
1182 * The page does not need to be reserved.
1183 */
1184int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, struct page *page)
1185{
1186 if (addr < vma->vm_start || addr >= vma->vm_end)
1187 return -EFAULT;
1188 if (!page_count(page))
1189 return -EINVAL;
Linus Torvalds4d7672b2005-12-16 10:21:23 -08001190 vma->vm_flags |= VM_INSERTPAGE;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001191 return insert_page(vma->vm_mm, addr, page, vma->vm_page_prot);
1192}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001193EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001194
Nick Piggine0dc0d82007-02-12 00:51:36 -08001195/**
1196 * vm_insert_pfn - insert single pfn into user vma
1197 * @vma: user vma to map to
1198 * @addr: target user address of this page
1199 * @pfn: source kernel pfn
1200 *
1201 * Similar to vm_inert_page, this allows drivers to insert individual pages
1202 * they've allocated into a user vma. Same comments apply.
1203 *
1204 * This function should only be called from a vm_ops->fault handler, and
1205 * in that case the handler should return NULL.
1206 */
1207int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1208 unsigned long pfn)
1209{
1210 struct mm_struct *mm = vma->vm_mm;
1211 int retval;
1212 pte_t *pte, entry;
1213 spinlock_t *ptl;
1214
1215 BUG_ON(!(vma->vm_flags & VM_PFNMAP));
1216 BUG_ON(is_cow_mapping(vma->vm_flags));
1217
1218 retval = -ENOMEM;
1219 pte = get_locked_pte(mm, addr, &ptl);
1220 if (!pte)
1221 goto out;
1222 retval = -EBUSY;
1223 if (!pte_none(*pte))
1224 goto out_unlock;
1225
1226 /* Ok, finally just insert the thing.. */
1227 entry = pfn_pte(pfn, vma->vm_page_prot);
1228 set_pte_at(mm, addr, pte, entry);
1229 update_mmu_cache(vma, addr, entry);
1230
1231 retval = 0;
1232out_unlock:
1233 pte_unmap_unlock(pte, ptl);
1234
1235out:
1236 return retval;
1237}
1238EXPORT_SYMBOL(vm_insert_pfn);
1239
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001240/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 * maps a range of physical memory into the requested pages. the old
1242 * mappings are removed. any references to nonexistent pages results
1243 * in null mappings (currently treated as "copy-on-access")
1244 */
1245static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1246 unsigned long addr, unsigned long end,
1247 unsigned long pfn, pgprot_t prot)
1248{
1249 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001250 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251
Hugh Dickinsc74df322005-10-29 18:16:23 -07001252 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 if (!pte)
1254 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001255 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 do {
1257 BUG_ON(!pte_none(*pte));
Nick Pigginb5810032005-10-29 18:16:12 -07001258 set_pte_at(mm, addr, pte, pfn_pte(pfn, prot));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 pfn++;
1260 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001261 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001262 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 return 0;
1264}
1265
1266static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1267 unsigned long addr, unsigned long end,
1268 unsigned long pfn, pgprot_t prot)
1269{
1270 pmd_t *pmd;
1271 unsigned long next;
1272
1273 pfn -= addr >> PAGE_SHIFT;
1274 pmd = pmd_alloc(mm, pud, addr);
1275 if (!pmd)
1276 return -ENOMEM;
1277 do {
1278 next = pmd_addr_end(addr, end);
1279 if (remap_pte_range(mm, pmd, addr, next,
1280 pfn + (addr >> PAGE_SHIFT), prot))
1281 return -ENOMEM;
1282 } while (pmd++, addr = next, addr != end);
1283 return 0;
1284}
1285
1286static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1287 unsigned long addr, unsigned long end,
1288 unsigned long pfn, pgprot_t prot)
1289{
1290 pud_t *pud;
1291 unsigned long next;
1292
1293 pfn -= addr >> PAGE_SHIFT;
1294 pud = pud_alloc(mm, pgd, addr);
1295 if (!pud)
1296 return -ENOMEM;
1297 do {
1298 next = pud_addr_end(addr, end);
1299 if (remap_pmd_range(mm, pud, addr, next,
1300 pfn + (addr >> PAGE_SHIFT), prot))
1301 return -ENOMEM;
1302 } while (pud++, addr = next, addr != end);
1303 return 0;
1304}
1305
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001306/**
1307 * remap_pfn_range - remap kernel memory to userspace
1308 * @vma: user vma to map to
1309 * @addr: target user address to start at
1310 * @pfn: physical address of kernel memory
1311 * @size: size of map area
1312 * @prot: page protection flags for this mapping
1313 *
1314 * Note: this is only safe if the mm semaphore is held when called.
1315 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1317 unsigned long pfn, unsigned long size, pgprot_t prot)
1318{
1319 pgd_t *pgd;
1320 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001321 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 struct mm_struct *mm = vma->vm_mm;
1323 int err;
1324
1325 /*
1326 * Physically remapped pages are special. Tell the
1327 * rest of the world about it:
1328 * VM_IO tells people not to look at these pages
1329 * (accesses can have side effects).
Hugh Dickins0b14c172005-11-21 21:32:15 -08001330 * VM_RESERVED is specified all over the place, because
1331 * in 2.4 it kept swapout's vma scan off this vma; but
1332 * in 2.6 the LRU scan won't even find its pages, so this
1333 * flag means no more than count its pages in reserved_vm,
1334 * and omit it from core dump, even when VM_IO turned off.
Linus Torvalds6aab3412005-11-28 14:34:23 -08001335 * VM_PFNMAP tells the core MM that the base pages are just
1336 * raw PFN mappings, and do not have a "struct page" associated
1337 * with them.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001338 *
1339 * There's a horrible special case to handle copy-on-write
1340 * behaviour that some programs depend on. We mark the "original"
1341 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 */
Linus Torvalds67121172005-12-11 20:38:17 -08001343 if (is_cow_mapping(vma->vm_flags)) {
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001344 if (addr != vma->vm_start || end != vma->vm_end)
Linus Torvalds7fc7e2e2005-12-11 19:57:52 -08001345 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001346 vma->vm_pgoff = pfn;
1347 }
1348
Linus Torvalds6aab3412005-11-28 14:34:23 -08001349 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
1351 BUG_ON(addr >= end);
1352 pfn -= addr >> PAGE_SHIFT;
1353 pgd = pgd_offset(mm, addr);
1354 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 do {
1356 next = pgd_addr_end(addr, end);
1357 err = remap_pud_range(mm, pgd, addr, next,
1358 pfn + (addr >> PAGE_SHIFT), prot);
1359 if (err)
1360 break;
1361 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 return err;
1363}
1364EXPORT_SYMBOL(remap_pfn_range);
1365
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001366static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1367 unsigned long addr, unsigned long end,
1368 pte_fn_t fn, void *data)
1369{
1370 pte_t *pte;
1371 int err;
1372 struct page *pmd_page;
Borislav Petkov94909912007-05-06 14:49:17 -07001373 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001374
1375 pte = (mm == &init_mm) ?
1376 pte_alloc_kernel(pmd, addr) :
1377 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1378 if (!pte)
1379 return -ENOMEM;
1380
1381 BUG_ON(pmd_huge(*pmd));
1382
1383 pmd_page = pmd_page(*pmd);
1384
1385 do {
1386 err = fn(pte, pmd_page, addr, data);
1387 if (err)
1388 break;
1389 } while (pte++, addr += PAGE_SIZE, addr != end);
1390
1391 if (mm != &init_mm)
1392 pte_unmap_unlock(pte-1, ptl);
1393 return err;
1394}
1395
1396static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1397 unsigned long addr, unsigned long end,
1398 pte_fn_t fn, void *data)
1399{
1400 pmd_t *pmd;
1401 unsigned long next;
1402 int err;
1403
1404 pmd = pmd_alloc(mm, pud, addr);
1405 if (!pmd)
1406 return -ENOMEM;
1407 do {
1408 next = pmd_addr_end(addr, end);
1409 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1410 if (err)
1411 break;
1412 } while (pmd++, addr = next, addr != end);
1413 return err;
1414}
1415
1416static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1417 unsigned long addr, unsigned long end,
1418 pte_fn_t fn, void *data)
1419{
1420 pud_t *pud;
1421 unsigned long next;
1422 int err;
1423
1424 pud = pud_alloc(mm, pgd, addr);
1425 if (!pud)
1426 return -ENOMEM;
1427 do {
1428 next = pud_addr_end(addr, end);
1429 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1430 if (err)
1431 break;
1432 } while (pud++, addr = next, addr != end);
1433 return err;
1434}
1435
1436/*
1437 * Scan a region of virtual memory, filling in page tables as necessary
1438 * and calling a provided function on each leaf page table.
1439 */
1440int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1441 unsigned long size, pte_fn_t fn, void *data)
1442{
1443 pgd_t *pgd;
1444 unsigned long next;
1445 unsigned long end = addr + size;
1446 int err;
1447
1448 BUG_ON(addr >= end);
1449 pgd = pgd_offset(mm, addr);
1450 do {
1451 next = pgd_addr_end(addr, end);
1452 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1453 if (err)
1454 break;
1455 } while (pgd++, addr = next, addr != end);
1456 return err;
1457}
1458EXPORT_SYMBOL_GPL(apply_to_page_range);
1459
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001461 * handle_pte_fault chooses page fault handler according to an entry
1462 * which was read non-atomically. Before making any commitment, on
1463 * those architectures or configurations (e.g. i386 with PAE) which
1464 * might give a mix of unmatched parts, do_swap_page and do_file_page
1465 * must check under lock before unmapping the pte and proceeding
1466 * (but do_wp_page is only called after already making such a check;
1467 * and do_anonymous_page and do_no_page can safely check later on).
1468 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001469static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001470 pte_t *page_table, pte_t orig_pte)
1471{
1472 int same = 1;
1473#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1474 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001475 spinlock_t *ptl = pte_lockptr(mm, pmd);
1476 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001477 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001478 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001479 }
1480#endif
1481 pte_unmap(page_table);
1482 return same;
1483}
1484
1485/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1487 * servicing faults for write access. In the normal case, do always want
1488 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1489 * that do not have writing enabled, when used by access_process_vm.
1490 */
1491static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1492{
1493 if (likely(vma->vm_flags & VM_WRITE))
1494 pte = pte_mkwrite(pte);
1495 return pte;
1496}
1497
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001498static 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 -08001499{
1500 /*
1501 * If the source page was a PFN mapping, we don't have
1502 * a "struct page" for it. We do a best-effort copy by
1503 * just copying from the original user address. If that
1504 * fails, we just zero-fill it. Live with it.
1505 */
1506 if (unlikely(!src)) {
1507 void *kaddr = kmap_atomic(dst, KM_USER0);
Linus Torvalds5d2a2db2005-11-29 14:07:55 -08001508 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1509
1510 /*
1511 * This really shouldn't fail, because the page is there
1512 * in the page tables. But it might just be unreadable,
1513 * in which case we just give up and fill the result with
1514 * zeroes.
1515 */
1516 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Linus Torvalds6aab3412005-11-28 14:34:23 -08001517 memset(kaddr, 0, PAGE_SIZE);
1518 kunmap_atomic(kaddr, KM_USER0);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07001519 flush_dcache_page(dst);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001520 return;
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001521
Linus Torvalds6aab3412005-11-28 14:34:23 -08001522 }
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001523 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001524}
1525
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 * This routine handles present pages, when users try to write
1528 * to a shared page. It is done by copying the page to a new address
1529 * and decrementing the shared-page counter for the old page.
1530 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 * Note that this routine assumes that the protection checks have been
1532 * done by the caller (the low-level page fault routine in most cases).
1533 * Thus we can safely just mark it writable once we've done any necessary
1534 * COW.
1535 *
1536 * We also mark the page dirty at this point even though the page will
1537 * change only once the write actually happens. This avoids a few races,
1538 * and potentially makes it more efficient.
1539 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001540 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1541 * but allow concurrent faults), with pte both mapped and locked.
1542 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001544static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1545 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001546 spinlock_t *ptl, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547{
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001548 struct page *old_page, *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 pte_t entry;
Nick Piggin83c54072007-07-19 01:47:05 -07001550 int reuse = 0, ret = 0;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001551 int page_mkwrite = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001552 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553
Linus Torvalds6aab3412005-11-28 14:34:23 -08001554 old_page = vm_normal_page(vma, address, orig_pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001555 if (!old_page)
1556 goto gotten;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001558 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001559 * Take out anonymous pages first, anonymous shared vmas are
1560 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001561 */
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001562 if (PageAnon(old_page)) {
1563 if (!TestSetPageLocked(old_page)) {
1564 reuse = can_share_swap_page(old_page);
1565 unlock_page(old_page);
1566 }
1567 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001568 (VM_WRITE|VM_SHARED))) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001569 /*
1570 * Only catch write-faults on shared writable pages,
1571 * read-only shared pages can get COWed by
1572 * get_user_pages(.write=1, .force=1).
1573 */
David Howells9637a5e2006-06-23 02:03:43 -07001574 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
1575 /*
1576 * Notify the address space that the page is about to
1577 * become writable so that it can prohibit this or wait
1578 * for the page to get into an appropriate state.
1579 *
1580 * We do this without the lock held, so that it can
1581 * sleep if it needs to.
1582 */
1583 page_cache_get(old_page);
1584 pte_unmap_unlock(page_table, ptl);
1585
1586 if (vma->vm_ops->page_mkwrite(vma, old_page) < 0)
1587 goto unwritable_page;
1588
David Howells9637a5e2006-06-23 02:03:43 -07001589 /*
1590 * Since we dropped the lock we need to revalidate
1591 * the PTE as someone else may have changed it. If
1592 * they did, we just return, as we can count on the
1593 * MMU to tell us if they didn't also make it writable.
1594 */
1595 page_table = pte_offset_map_lock(mm, pmd, address,
1596 &ptl);
Hugh Dickinsc3704ce2007-02-10 01:43:00 -08001597 page_cache_release(old_page);
David Howells9637a5e2006-06-23 02:03:43 -07001598 if (!pte_same(*page_table, orig_pte))
1599 goto unlock;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001600
1601 page_mkwrite = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001603 dirty_page = old_page;
1604 get_page(dirty_page);
David Howells9637a5e2006-06-23 02:03:43 -07001605 reuse = 1;
David Howells9637a5e2006-06-23 02:03:43 -07001606 }
1607
1608 if (reuse) {
1609 flush_cache_page(vma, address, pte_pfn(orig_pte));
1610 entry = pte_mkyoung(orig_pte);
1611 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
KAMEZAWA Hiroyuki954ffcb2007-10-16 01:25:44 -07001612 if (ptep_set_access_flags(vma, address, page_table, entry,1))
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07001613 update_mmu_cache(vma, address, entry);
David Howells9637a5e2006-06-23 02:03:43 -07001614 ret |= VM_FAULT_WRITE;
1615 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617
1618 /*
1619 * Ok, we need to copy. Oh, well..
1620 */
Nick Pigginb5810032005-10-29 18:16:12 -07001621 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08001622gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001623 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624
1625 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07001626 goto oom;
Nick Piggin557ed1f2007-10-16 01:24:40 -07001627 VM_BUG_ON(old_page == ZERO_PAGE(0));
1628 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
1629 if (!new_page)
1630 goto oom;
1631 cow_user_page(new_page, old_page, address, vma);
Hugh Dickins65500d22005-10-29 18:15:59 -07001632
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 /*
1634 * Re-check the pte - we dropped the lock
1635 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001636 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07001637 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08001638 if (old_page) {
Nick Piggin7de6b802006-12-22 01:09:33 -08001639 page_remove_rmap(old_page, vma);
Hugh Dickins920fc352005-11-21 21:32:17 -08001640 if (!PageAnon(old_page)) {
1641 dec_mm_counter(mm, file_rss);
1642 inc_mm_counter(mm, anon_rss);
1643 }
1644 } else
Hugh Dickins42946212005-10-29 18:16:05 -07001645 inc_mm_counter(mm, anon_rss);
Ben Collinseca35132005-11-29 11:45:26 -08001646 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07001647 entry = mk_pte(new_page, vma->vm_page_prot);
1648 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Siddha, Suresh B4ce072f2006-09-29 01:58:42 -07001649 /*
1650 * Clear the pte entry and flush it first, before updating the
1651 * pte with the new entry. This will avoid a race condition
1652 * seen in the presence of one thread doing SMC and another
1653 * thread doing COW.
1654 */
1655 ptep_clear_flush(vma, address, page_table);
1656 set_pte_at(mm, address, page_table, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07001657 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 lru_cache_add_active(new_page);
Nick Piggin9617d952006-01-06 00:11:12 -08001659 page_add_new_anon_rmap(new_page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660
1661 /* Free the old page.. */
1662 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001663 ret |= VM_FAULT_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 }
Hugh Dickins920fc352005-11-21 21:32:17 -08001665 if (new_page)
1666 page_cache_release(new_page);
1667 if (old_page)
1668 page_cache_release(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001669unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001670 pte_unmap_unlock(page_table, ptl);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001671 if (dirty_page) {
Anton Salikhmetov8f7b3d12008-01-23 02:21:18 +03001672 if (vma->vm_file)
1673 file_update_time(vma->vm_file);
1674
Nick Piggin79352892007-07-19 01:47:22 -07001675 /*
1676 * Yes, Virginia, this is actually required to prevent a race
1677 * with clear_page_dirty_for_io() from clearing the page dirty
1678 * bit after it clear all dirty ptes, but before a racing
1679 * do_wp_page installs a dirty pte.
1680 *
1681 * do_no_page is protected similarly.
1682 */
1683 wait_on_page_locked(dirty_page);
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001684 set_page_dirty_balance(dirty_page, page_mkwrite);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001685 put_page(dirty_page);
1686 }
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001687 return ret;
Hugh Dickins65500d22005-10-29 18:15:59 -07001688oom:
Hugh Dickins920fc352005-11-21 21:32:17 -08001689 if (old_page)
1690 page_cache_release(old_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 return VM_FAULT_OOM;
David Howells9637a5e2006-06-23 02:03:43 -07001692
1693unwritable_page:
1694 page_cache_release(old_page);
1695 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696}
1697
1698/*
1699 * Helper functions for unmap_mapping_range().
1700 *
1701 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1702 *
1703 * We have to restart searching the prio_tree whenever we drop the lock,
1704 * since the iterator is only valid while the lock is held, and anyway
1705 * a later vma might be split and reinserted earlier while lock dropped.
1706 *
1707 * The list of nonlinear vmas could be handled more efficiently, using
1708 * a placeholder, but handle it in the same way until a need is shown.
1709 * It is important to search the prio_tree before nonlinear list: a vma
1710 * may become nonlinear and be shifted from prio_tree to nonlinear list
1711 * while the lock is dropped; but never shifted from list to prio_tree.
1712 *
1713 * In order to make forward progress despite restarting the search,
1714 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1715 * quickly skip it next time around. Since the prio_tree search only
1716 * shows us those vmas affected by unmapping the range in question, we
1717 * can't efficiently keep all vmas in step with mapping->truncate_count:
1718 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1719 * mapping->truncate_count and vma->vm_truncate_count are protected by
1720 * i_mmap_lock.
1721 *
1722 * In order to make forward progress despite repeatedly restarting some
Hugh Dickinsee39b372005-04-19 13:29:15 -07001723 * large vma, note the restart_addr from unmap_vmas when it breaks out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 * and restart from that address when we reach that vma again. It might
1725 * have been split or merged, shrunk or extended, but never shifted: so
1726 * restart_addr remains valid so long as it remains in the vma's range.
1727 * unmap_mapping_range forces truncate_count to leap over page-aligned
1728 * values so we can save vma's restart_addr in its truncate_count field.
1729 */
1730#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1731
1732static void reset_vma_truncate_counts(struct address_space *mapping)
1733{
1734 struct vm_area_struct *vma;
1735 struct prio_tree_iter iter;
1736
1737 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1738 vma->vm_truncate_count = 0;
1739 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1740 vma->vm_truncate_count = 0;
1741}
1742
1743static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1744 unsigned long start_addr, unsigned long end_addr,
1745 struct zap_details *details)
1746{
1747 unsigned long restart_addr;
1748 int need_break;
1749
Nick Piggind00806b2007-07-19 01:46:57 -07001750 /*
1751 * files that support invalidating or truncating portions of the
Nick Piggind0217ac2007-07-19 01:47:03 -07001752 * file from under mmaped areas must have their ->fault function
Nick Piggin83c54072007-07-19 01:47:05 -07001753 * return a locked page (and set VM_FAULT_LOCKED in the return).
1754 * This provides synchronisation against concurrent unmapping here.
Nick Piggind00806b2007-07-19 01:46:57 -07001755 */
Nick Piggind00806b2007-07-19 01:46:57 -07001756
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757again:
1758 restart_addr = vma->vm_truncate_count;
1759 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1760 start_addr = restart_addr;
1761 if (start_addr >= end_addr) {
1762 /* Top of vma has been split off since last time */
1763 vma->vm_truncate_count = details->truncate_count;
1764 return 0;
1765 }
1766 }
1767
Hugh Dickinsee39b372005-04-19 13:29:15 -07001768 restart_addr = zap_page_range(vma, start_addr,
1769 end_addr - start_addr, details);
Nick Piggin95c354f2008-01-30 13:31:20 +01001770 need_break = need_resched() || spin_needbreak(details->i_mmap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771
Hugh Dickinsee39b372005-04-19 13:29:15 -07001772 if (restart_addr >= end_addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 /* We have now completed this vma: mark it so */
1774 vma->vm_truncate_count = details->truncate_count;
1775 if (!need_break)
1776 return 0;
1777 } else {
1778 /* Note restart_addr in vma's truncate_count field */
Hugh Dickinsee39b372005-04-19 13:29:15 -07001779 vma->vm_truncate_count = restart_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 if (!need_break)
1781 goto again;
1782 }
1783
1784 spin_unlock(details->i_mmap_lock);
1785 cond_resched();
1786 spin_lock(details->i_mmap_lock);
1787 return -EINTR;
1788}
1789
1790static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1791 struct zap_details *details)
1792{
1793 struct vm_area_struct *vma;
1794 struct prio_tree_iter iter;
1795 pgoff_t vba, vea, zba, zea;
1796
1797restart:
1798 vma_prio_tree_foreach(vma, &iter, root,
1799 details->first_index, details->last_index) {
1800 /* Skip quickly over those we have already dealt with */
1801 if (vma->vm_truncate_count == details->truncate_count)
1802 continue;
1803
1804 vba = vma->vm_pgoff;
1805 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1806 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1807 zba = details->first_index;
1808 if (zba < vba)
1809 zba = vba;
1810 zea = details->last_index;
1811 if (zea > vea)
1812 zea = vea;
1813
1814 if (unmap_mapping_range_vma(vma,
1815 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1816 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1817 details) < 0)
1818 goto restart;
1819 }
1820}
1821
1822static inline void unmap_mapping_range_list(struct list_head *head,
1823 struct zap_details *details)
1824{
1825 struct vm_area_struct *vma;
1826
1827 /*
1828 * In nonlinear VMAs there is no correspondence between virtual address
1829 * offset and file offset. So we must perform an exhaustive search
1830 * across *all* the pages in each nonlinear VMA, not just the pages
1831 * whose virtual address lies outside the file truncation point.
1832 */
1833restart:
1834 list_for_each_entry(vma, head, shared.vm_set.list) {
1835 /* Skip quickly over those we have already dealt with */
1836 if (vma->vm_truncate_count == details->truncate_count)
1837 continue;
1838 details->nonlinear_vma = vma;
1839 if (unmap_mapping_range_vma(vma, vma->vm_start,
1840 vma->vm_end, details) < 0)
1841 goto restart;
1842 }
1843}
1844
1845/**
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001846 * 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 -07001847 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 * @holebegin: byte in first page to unmap, relative to the start of
1849 * the underlying file. This will be rounded down to a PAGE_SIZE
1850 * boundary. Note that this is different from vmtruncate(), which
1851 * must keep the partial page. In contrast, we must get rid of
1852 * partial pages.
1853 * @holelen: size of prospective hole in bytes. This will be rounded
1854 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1855 * end of the file.
1856 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1857 * but 0 when invalidating pagecache, don't throw away private data.
1858 */
1859void unmap_mapping_range(struct address_space *mapping,
1860 loff_t const holebegin, loff_t const holelen, int even_cows)
1861{
1862 struct zap_details details;
1863 pgoff_t hba = holebegin >> PAGE_SHIFT;
1864 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1865
1866 /* Check for overflow. */
1867 if (sizeof(holelen) > sizeof(hlen)) {
1868 long long holeend =
1869 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1870 if (holeend & ~(long long)ULONG_MAX)
1871 hlen = ULONG_MAX - hba + 1;
1872 }
1873
1874 details.check_mapping = even_cows? NULL: mapping;
1875 details.nonlinear_vma = NULL;
1876 details.first_index = hba;
1877 details.last_index = hba + hlen - 1;
1878 if (details.last_index < details.first_index)
1879 details.last_index = ULONG_MAX;
1880 details.i_mmap_lock = &mapping->i_mmap_lock;
1881
1882 spin_lock(&mapping->i_mmap_lock);
1883
Nick Piggind00806b2007-07-19 01:46:57 -07001884 /* Protect against endless unmapping loops */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 mapping->truncate_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 if (unlikely(is_restart_addr(mapping->truncate_count))) {
1887 if (mapping->truncate_count == 0)
1888 reset_vma_truncate_counts(mapping);
1889 mapping->truncate_count++;
1890 }
1891 details.truncate_count = mapping->truncate_count;
1892
1893 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
1894 unmap_mapping_range_tree(&mapping->i_mmap, &details);
1895 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
1896 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
1897 spin_unlock(&mapping->i_mmap_lock);
1898}
1899EXPORT_SYMBOL(unmap_mapping_range);
1900
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001901/**
1902 * vmtruncate - unmap mappings "freed" by truncate() syscall
1903 * @inode: inode of the file used
1904 * @offset: file offset to start truncating
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 *
1906 * NOTE! We have to be ready to update the memory sharing
1907 * between the file and the memory map for a potential last
1908 * incomplete page. Ugly, but necessary.
1909 */
1910int vmtruncate(struct inode * inode, loff_t offset)
1911{
1912 struct address_space *mapping = inode->i_mapping;
1913 unsigned long limit;
1914
1915 if (inode->i_size < offset)
1916 goto do_expand;
1917 /*
1918 * truncation of in-use swapfiles is disallowed - it would cause
1919 * subsequent swapout to scribble on the now-freed blocks.
1920 */
1921 if (IS_SWAPFILE(inode))
1922 goto out_busy;
1923 i_size_write(inode, offset);
Nick Piggind00806b2007-07-19 01:46:57 -07001924
1925 /*
1926 * unmap_mapping_range is called twice, first simply for efficiency
1927 * so that truncate_inode_pages does fewer single-page unmaps. However
1928 * after this first call, and before truncate_inode_pages finishes,
1929 * it is possible for private pages to be COWed, which remain after
1930 * truncate_inode_pages finishes, hence the second unmap_mapping_range
1931 * call must be made for correctness.
1932 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
1934 truncate_inode_pages(mapping, offset);
Nick Piggind00806b2007-07-19 01:46:57 -07001935 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 goto out_truncate;
1937
1938do_expand:
1939 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1940 if (limit != RLIM_INFINITY && offset > limit)
1941 goto out_sig;
1942 if (offset > inode->i_sb->s_maxbytes)
1943 goto out_big;
1944 i_size_write(inode, offset);
1945
1946out_truncate:
1947 if (inode->i_op && inode->i_op->truncate)
1948 inode->i_op->truncate(inode);
1949 return 0;
1950out_sig:
1951 send_sig(SIGXFSZ, current, 0);
1952out_big:
1953 return -EFBIG;
1954out_busy:
1955 return -ETXTBSY;
1956}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957EXPORT_SYMBOL(vmtruncate);
1958
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001959int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
1960{
1961 struct address_space *mapping = inode->i_mapping;
1962
1963 /*
1964 * If the underlying filesystem is not going to provide
1965 * a way to truncate a range of blocks (punch a hole) -
1966 * we should return failure right now.
1967 */
1968 if (!inode->i_op || !inode->i_op->truncate_range)
1969 return -ENOSYS;
1970
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001971 mutex_lock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001972 down_write(&inode->i_alloc_sem);
1973 unmap_mapping_range(mapping, offset, (end - offset), 1);
1974 truncate_inode_pages_range(mapping, offset, end);
Nick Piggind00806b2007-07-19 01:46:57 -07001975 unmap_mapping_range(mapping, offset, (end - offset), 1);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001976 inode->i_op->truncate_range(inode, offset, end);
1977 up_write(&inode->i_alloc_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001978 mutex_unlock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001979
1980 return 0;
1981}
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001982
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001984 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1985 * but allow concurrent faults), and pte mapped but not yet locked.
1986 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001988static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
1989 unsigned long address, pte_t *page_table, pmd_t *pmd,
1990 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001992 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 struct page *page;
Hugh Dickins65500d22005-10-29 18:15:59 -07001994 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 pte_t pte;
Nick Piggin83c54072007-07-19 01:47:05 -07001996 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001998 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001999 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002000
2001 entry = pte_to_swp_entry(orig_pte);
Christoph Lameter06972122006-06-23 02:03:35 -07002002 if (is_migration_entry(entry)) {
2003 migration_entry_wait(mm, pmd, address);
2004 goto out;
2005 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002006 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 page = lookup_swap_cache(entry);
2008 if (!page) {
Ashwin Chaugule098fe652006-12-06 20:31:54 -08002009 grab_swap_token(); /* Contend for token _before_ read-in */
Hugh Dickins02098fe2008-02-04 22:28:42 -08002010 page = swapin_readahead(entry,
2011 GFP_HIGHUSER_MOVABLE, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 if (!page) {
2013 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002014 * Back out if somebody else faulted in this pte
2015 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002017 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 if (likely(pte_same(*page_table, orig_pte)))
2019 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002020 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002021 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 }
2023
2024 /* Had to read the page from swap area: Major fault */
2025 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002026 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 }
2028
2029 mark_page_accessed(page);
2030 lock_page(page);
Balbir Singh20a10222007-11-14 17:00:33 -08002031 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032
2033 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002034 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002036 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002037 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002038 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002039
2040 if (unlikely(!PageUptodate(page))) {
2041 ret = VM_FAULT_SIGBUS;
2042 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 }
2044
2045 /* The page isn't present yet, go ahead with the fault. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046
Hugh Dickins42946212005-10-29 18:16:05 -07002047 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 pte = mk_pte(page, vma->vm_page_prot);
2049 if (write_access && can_share_swap_page(page)) {
2050 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2051 write_access = 0;
2052 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053
2054 flush_icache_page(vma, page);
2055 set_pte_at(mm, address, page_table, pte);
2056 page_add_anon_rmap(page, vma, address);
2057
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002058 swap_free(entry);
2059 if (vm_swap_full())
2060 remove_exclusive_swap_page(page);
2061 unlock_page(page);
2062
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 if (write_access) {
Nick Piggin83c54072007-07-19 01:47:05 -07002064 /* XXX: We could OR the do_wp_page code with this one? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 if (do_wp_page(mm, vma, address,
Nick Piggin83c54072007-07-19 01:47:05 -07002066 page_table, pmd, ptl, pte) & VM_FAULT_OOM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 ret = VM_FAULT_OOM;
2068 goto out;
2069 }
2070
2071 /* No need to invalidate - it was non-present before */
2072 update_mmu_cache(vma, address, pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07002073unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002074 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075out:
2076 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002077out_nomap:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002078 pte_unmap_unlock(page_table, ptl);
Kirill Korotaevb8107482005-05-16 21:53:50 -07002079 unlock_page(page);
2080 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002081 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082}
2083
2084/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002085 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2086 * but allow concurrent faults), and pte mapped but not yet locked.
2087 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002089static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2090 unsigned long address, pte_t *page_table, pmd_t *pmd,
2091 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002093 struct page *page;
2094 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096
Nick Piggin557ed1f2007-10-16 01:24:40 -07002097 /* Allocate our own private page. */
2098 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099
Nick Piggin557ed1f2007-10-16 01:24:40 -07002100 if (unlikely(anon_vma_prepare(vma)))
2101 goto oom;
2102 page = alloc_zeroed_user_highpage_movable(vma, address);
2103 if (!page)
2104 goto oom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105
Nick Piggin557ed1f2007-10-16 01:24:40 -07002106 entry = mk_pte(page, vma->vm_page_prot);
2107 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002108
Nick Piggin557ed1f2007-10-16 01:24:40 -07002109 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2110 if (!pte_none(*page_table))
2111 goto release;
2112 inc_mm_counter(mm, anon_rss);
2113 lru_cache_add_active(page);
2114 page_add_new_anon_rmap(page, vma, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002115 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116
2117 /* No need to invalidate - it was non-present before */
Hugh Dickins65500d22005-10-29 18:15:59 -07002118 update_mmu_cache(vma, address, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002119unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002120 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002121 return 0;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002122release:
2123 page_cache_release(page);
2124 goto unlock;
Hugh Dickins65500d22005-10-29 18:15:59 -07002125oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 return VM_FAULT_OOM;
2127}
2128
2129/*
Nick Piggin54cb8822007-07-19 01:46:59 -07002130 * __do_fault() tries to create a new page mapping. It aggressively
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 * tries to share with existing pages, but makes a separate copy if
Nick Piggin54cb8822007-07-19 01:46:59 -07002132 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2133 * the next page fault.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 *
2135 * As this is called only for pages that do not currently exist, we
2136 * do not need to flush old virtual caches or the TLB.
2137 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002138 * We enter with non-exclusive mmap_sem (to exclude vma changes,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002139 * but allow concurrent faults), and pte neither mapped nor locked.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002140 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 */
Nick Piggin54cb8822007-07-19 01:46:59 -07002142static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002143 unsigned long address, pmd_t *pmd,
Nick Piggin54cb8822007-07-19 01:46:59 -07002144 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145{
Hugh Dickins16abfa02007-10-04 16:56:06 +01002146 pte_t *page_table;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002147 spinlock_t *ptl;
Nick Piggind0217ac2007-07-19 01:47:03 -07002148 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 int anon = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002151 struct page *dirty_page = NULL;
Nick Piggind0217ac2007-07-19 01:47:03 -07002152 struct vm_fault vmf;
2153 int ret;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002154 int page_mkwrite = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07002155
Nick Piggind0217ac2007-07-19 01:47:03 -07002156 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2157 vmf.pgoff = pgoff;
2158 vmf.flags = flags;
2159 vmf.page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160
Hugh Dickins325f04d2005-11-29 16:55:48 +00002161 BUG_ON(vma->vm_flags & VM_PFNMAP);
2162
Nick Piggin54cb8822007-07-19 01:46:59 -07002163 if (likely(vma->vm_ops->fault)) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002164 ret = vma->vm_ops->fault(vma, &vmf);
Nick Piggin83c54072007-07-19 01:47:05 -07002165 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2166 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07002167 } else {
2168 /* Legacy ->nopage path */
Nick Piggin83c54072007-07-19 01:47:05 -07002169 ret = 0;
Nick Piggind0217ac2007-07-19 01:47:03 -07002170 vmf.page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret);
Nick Piggin54cb8822007-07-19 01:46:59 -07002171 /* no page was available -- either SIGBUS or OOM */
Nick Piggind0217ac2007-07-19 01:47:03 -07002172 if (unlikely(vmf.page == NOPAGE_SIGBUS))
Nick Piggin54cb8822007-07-19 01:46:59 -07002173 return VM_FAULT_SIGBUS;
Nick Piggind0217ac2007-07-19 01:47:03 -07002174 else if (unlikely(vmf.page == NOPAGE_OOM))
Nick Piggin54cb8822007-07-19 01:46:59 -07002175 return VM_FAULT_OOM;
2176 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177
Nick Piggind00806b2007-07-19 01:46:57 -07002178 /*
Nick Piggind0217ac2007-07-19 01:47:03 -07002179 * For consistency in subsequent calls, make the faulted page always
Nick Piggind00806b2007-07-19 01:46:57 -07002180 * locked.
2181 */
Nick Piggin83c54072007-07-19 01:47:05 -07002182 if (unlikely(!(ret & VM_FAULT_LOCKED)))
Nick Piggind0217ac2007-07-19 01:47:03 -07002183 lock_page(vmf.page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002184 else
Nick Piggind0217ac2007-07-19 01:47:03 -07002185 VM_BUG_ON(!PageLocked(vmf.page));
Nick Piggind00806b2007-07-19 01:46:57 -07002186
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 /*
2188 * Should we do an early C-O-W break?
2189 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002190 page = vmf.page;
Nick Piggin54cb8822007-07-19 01:46:59 -07002191 if (flags & FAULT_FLAG_WRITE) {
David Howells9637a5e2006-06-23 02:03:43 -07002192 if (!(vma->vm_flags & VM_SHARED)) {
Nick Piggin54cb8822007-07-19 01:46:59 -07002193 anon = 1;
Nick Piggind00806b2007-07-19 01:46:57 -07002194 if (unlikely(anon_vma_prepare(vma))) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002195 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002196 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002197 }
Nick Piggin83c54072007-07-19 01:47:05 -07002198 page = alloc_page_vma(GFP_HIGHUSER_MOVABLE,
2199 vma, address);
Nick Piggind00806b2007-07-19 01:46:57 -07002200 if (!page) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002201 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002202 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002203 }
Nick Piggind0217ac2007-07-19 01:47:03 -07002204 copy_user_highpage(page, vmf.page, address, vma);
David Howells9637a5e2006-06-23 02:03:43 -07002205 } else {
Nick Piggin54cb8822007-07-19 01:46:59 -07002206 /*
2207 * If the page will be shareable, see if the backing
David Howells9637a5e2006-06-23 02:03:43 -07002208 * address space wants to know that the page is about
Nick Piggin54cb8822007-07-19 01:46:59 -07002209 * to become writable
2210 */
Mark Fasheh69676142007-07-19 01:47:00 -07002211 if (vma->vm_ops->page_mkwrite) {
2212 unlock_page(page);
2213 if (vma->vm_ops->page_mkwrite(vma, page) < 0) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002214 ret = VM_FAULT_SIGBUS;
2215 anon = 1; /* no anon but release vmf.page */
Mark Fasheh69676142007-07-19 01:47:00 -07002216 goto out_unlocked;
2217 }
2218 lock_page(page);
Nick Piggind0217ac2007-07-19 01:47:03 -07002219 /*
2220 * XXX: this is not quite right (racy vs
2221 * invalidate) to unlock and relock the page
2222 * like this, however a better fix requires
2223 * reworking page_mkwrite locking API, which
2224 * is better done later.
2225 */
2226 if (!page->mapping) {
Nick Piggin83c54072007-07-19 01:47:05 -07002227 ret = 0;
Nick Piggind0217ac2007-07-19 01:47:03 -07002228 anon = 1; /* no anon but release vmf.page */
2229 goto out;
2230 }
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002231 page_mkwrite = 1;
David Howells9637a5e2006-06-23 02:03:43 -07002232 }
2233 }
Nick Piggin54cb8822007-07-19 01:46:59 -07002234
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 }
2236
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002237 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238
2239 /*
2240 * This silly early PAGE_DIRTY setting removes a race
2241 * due to the bad i386 page protection. But it's valid
2242 * for other architectures too.
2243 *
2244 * Note that if write_access is true, we either now have
2245 * an exclusive copy of the page, or this is a shared mapping,
2246 * so we can make it writable and dirty to avoid having to
2247 * handle that later.
2248 */
2249 /* Only go through if we didn't race with anybody else... */
Nick Piggin54cb8822007-07-19 01:46:59 -07002250 if (likely(pte_same(*page_table, orig_pte))) {
Nick Piggind00806b2007-07-19 01:46:57 -07002251 flush_icache_page(vma, page);
2252 entry = mk_pte(page, vma->vm_page_prot);
Nick Piggin54cb8822007-07-19 01:46:59 -07002253 if (flags & FAULT_FLAG_WRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2255 set_pte_at(mm, address, page_table, entry);
2256 if (anon) {
Nick Piggind00806b2007-07-19 01:46:57 -07002257 inc_mm_counter(mm, anon_rss);
2258 lru_cache_add_active(page);
2259 page_add_new_anon_rmap(page, vma, address);
Hugh Dickinsf57e88a2005-11-21 21:32:19 -08002260 } else {
Hugh Dickins42946212005-10-29 18:16:05 -07002261 inc_mm_counter(mm, file_rss);
Nick Piggind00806b2007-07-19 01:46:57 -07002262 page_add_file_rmap(page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002263 if (flags & FAULT_FLAG_WRITE) {
Nick Piggind00806b2007-07-19 01:46:57 -07002264 dirty_page = page;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002265 get_page(dirty_page);
2266 }
Hugh Dickins42946212005-10-29 18:16:05 -07002267 }
Nick Piggind00806b2007-07-19 01:46:57 -07002268
2269 /* no need to invalidate: a not-present page won't be cached */
2270 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271 } else {
Nick Piggind00806b2007-07-19 01:46:57 -07002272 if (anon)
2273 page_cache_release(page);
2274 else
Nick Piggin54cb8822007-07-19 01:46:59 -07002275 anon = 1; /* no anon but release faulted_page */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 }
2277
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002278 pte_unmap_unlock(page_table, ptl);
Nick Piggind00806b2007-07-19 01:46:57 -07002279
2280out:
Nick Piggind0217ac2007-07-19 01:47:03 -07002281 unlock_page(vmf.page);
Mark Fasheh69676142007-07-19 01:47:00 -07002282out_unlocked:
Nick Piggind00806b2007-07-19 01:46:57 -07002283 if (anon)
Nick Piggind0217ac2007-07-19 01:47:03 -07002284 page_cache_release(vmf.page);
Nick Piggind00806b2007-07-19 01:46:57 -07002285 else if (dirty_page) {
Anton Salikhmetov8f7b3d12008-01-23 02:21:18 +03002286 if (vma->vm_file)
2287 file_update_time(vma->vm_file);
2288
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002289 set_page_dirty_balance(dirty_page, page_mkwrite);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002290 put_page(dirty_page);
2291 }
Nick Piggind00806b2007-07-19 01:46:57 -07002292
Nick Piggin83c54072007-07-19 01:47:05 -07002293 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07002294}
Nick Piggind00806b2007-07-19 01:46:57 -07002295
Nick Piggin54cb8822007-07-19 01:46:59 -07002296static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2297 unsigned long address, pte_t *page_table, pmd_t *pmd,
2298 int write_access, pte_t orig_pte)
2299{
2300 pgoff_t pgoff = (((address & PAGE_MASK)
Dean Nelson0da7e012007-10-16 01:24:45 -07002301 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
Nick Piggin54cb8822007-07-19 01:46:59 -07002302 unsigned int flags = (write_access ? FAULT_FLAG_WRITE : 0);
2303
Hugh Dickins16abfa02007-10-04 16:56:06 +01002304 pte_unmap(page_table);
2305 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Nick Piggin54cb8822007-07-19 01:46:59 -07002306}
2307
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308
2309/*
Jes Sorensenf4b81802006-09-27 01:50:10 -07002310 * do_no_pfn() tries to create a new page mapping for a page without
2311 * a struct_page backing it
2312 *
2313 * As this is called only for pages that do not currently exist, we
2314 * do not need to flush old virtual caches or the TLB.
2315 *
2316 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2317 * but allow concurrent faults), and pte mapped but not yet locked.
2318 * We return with mmap_sem still held, but pte unmapped and unlocked.
2319 *
2320 * It is expected that the ->nopfn handler always returns the same pfn
2321 * for a given virtual mapping.
2322 *
2323 * Mark this `noinline' to prevent it from bloating the main pagefault code.
2324 */
2325static noinline int do_no_pfn(struct mm_struct *mm, struct vm_area_struct *vma,
2326 unsigned long address, pte_t *page_table, pmd_t *pmd,
2327 int write_access)
2328{
2329 spinlock_t *ptl;
2330 pte_t entry;
2331 unsigned long pfn;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002332
2333 pte_unmap(page_table);
2334 BUG_ON(!(vma->vm_flags & VM_PFNMAP));
2335 BUG_ON(is_cow_mapping(vma->vm_flags));
2336
2337 pfn = vma->vm_ops->nopfn(vma, address & PAGE_MASK);
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002338 if (unlikely(pfn == NOPFN_OOM))
Jes Sorensenf4b81802006-09-27 01:50:10 -07002339 return VM_FAULT_OOM;
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002340 else if (unlikely(pfn == NOPFN_SIGBUS))
Jes Sorensenf4b81802006-09-27 01:50:10 -07002341 return VM_FAULT_SIGBUS;
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002342 else if (unlikely(pfn == NOPFN_REFAULT))
Nick Piggin83c54072007-07-19 01:47:05 -07002343 return 0;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002344
2345 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2346
2347 /* Only go through if we didn't race with anybody else... */
2348 if (pte_none(*page_table)) {
2349 entry = pfn_pte(pfn, vma->vm_page_prot);
2350 if (write_access)
2351 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2352 set_pte_at(mm, address, page_table, entry);
2353 }
2354 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002355 return 0;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002356}
2357
2358/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 * Fault of a previously existing named mapping. Repopulate the pte
2360 * from the encoded file_pte if possible. This enables swappable
2361 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002362 *
2363 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2364 * but allow concurrent faults), and pte mapped but not yet locked.
2365 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002367static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins65500d22005-10-29 18:15:59 -07002368 unsigned long address, pte_t *page_table, pmd_t *pmd,
2369 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370{
Nick Piggind0217ac2007-07-19 01:47:03 -07002371 unsigned int flags = FAULT_FLAG_NONLINEAR |
2372 (write_access ? FAULT_FLAG_WRITE : 0);
Hugh Dickins65500d22005-10-29 18:15:59 -07002373 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002375 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Nick Piggin83c54072007-07-19 01:47:05 -07002376 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377
Nick Piggind0217ac2007-07-19 01:47:03 -07002378 if (unlikely(!(vma->vm_flags & VM_NONLINEAR) ||
2379 !(vma->vm_flags & VM_CAN_NONLINEAR))) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002380 /*
2381 * Page table corrupted: show pte and kill process.
2382 */
Nick Pigginb5810032005-10-29 18:16:12 -07002383 print_bad_pte(vma, orig_pte, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002384 return VM_FAULT_OOM;
2385 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002386
2387 pgoff = pte_to_pgoff(orig_pte);
Hugh Dickins16abfa02007-10-04 16:56:06 +01002388 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389}
2390
2391/*
2392 * These routines also need to handle stuff like marking pages dirty
2393 * and/or accessed for architectures that don't do it in hardware (most
2394 * RISC architectures). The early dirtying is also good on the i386.
2395 *
2396 * There is also a hook called "update_mmu_cache()" that architectures
2397 * with external mmu caches can use to update those (ie the Sparc or
2398 * PowerPC hashed page tables that act as extended TLBs).
2399 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07002400 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2401 * but allow concurrent faults), and pte mapped but not yet locked.
2402 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 */
2404static inline int handle_pte_fault(struct mm_struct *mm,
Hugh Dickins65500d22005-10-29 18:15:59 -07002405 struct vm_area_struct *vma, unsigned long address,
2406 pte_t *pte, pmd_t *pmd, int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407{
2408 pte_t entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002409 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002411 entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002413 if (pte_none(entry)) {
Jes Sorensenf4b81802006-09-27 01:50:10 -07002414 if (vma->vm_ops) {
Nick Piggin54cb8822007-07-19 01:46:59 -07002415 if (vma->vm_ops->fault || vma->vm_ops->nopage)
2416 return do_linear_fault(mm, vma, address,
2417 pte, pmd, write_access, entry);
Jes Sorensenf4b81802006-09-27 01:50:10 -07002418 if (unlikely(vma->vm_ops->nopfn))
2419 return do_no_pfn(mm, vma, address, pte,
2420 pmd, write_access);
2421 }
2422 return do_anonymous_page(mm, vma, address,
2423 pte, pmd, write_access);
Hugh Dickins65500d22005-10-29 18:15:59 -07002424 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425 if (pte_file(entry))
Nick Piggind0217ac2007-07-19 01:47:03 -07002426 return do_nonlinear_fault(mm, vma, address,
Hugh Dickins65500d22005-10-29 18:15:59 -07002427 pte, pmd, write_access, entry);
2428 return do_swap_page(mm, vma, address,
2429 pte, pmd, write_access, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430 }
2431
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002432 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002433 spin_lock(ptl);
2434 if (unlikely(!pte_same(*pte, entry)))
2435 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 if (write_access) {
2437 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002438 return do_wp_page(mm, vma, address,
2439 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 entry = pte_mkdirty(entry);
2441 }
2442 entry = pte_mkyoung(entry);
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002443 if (ptep_set_access_flags(vma, address, pte, entry, write_access)) {
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002444 update_mmu_cache(vma, address, entry);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002445 } else {
2446 /*
2447 * This is needed only for protection faults but the arch code
2448 * is not yet telling us if this is a protection fault or not.
2449 * This still avoids useless tlb flushes for .text page faults
2450 * with threads.
2451 */
2452 if (write_access)
2453 flush_tlb_page(vma, address);
2454 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002455unlock:
2456 pte_unmap_unlock(pte, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002457 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458}
2459
2460/*
2461 * By the time we get here, we already hold the mm semaphore
2462 */
Nick Piggin83c54072007-07-19 01:47:05 -07002463int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464 unsigned long address, int write_access)
2465{
2466 pgd_t *pgd;
2467 pud_t *pud;
2468 pmd_t *pmd;
2469 pte_t *pte;
2470
2471 __set_current_state(TASK_RUNNING);
2472
Christoph Lameterf8891e52006-06-30 01:55:45 -07002473 count_vm_event(PGFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01002475 if (unlikely(is_vm_hugetlb_page(vma)))
2476 return hugetlb_fault(mm, vma, address, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 pud = pud_alloc(mm, pgd, address);
2480 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002481 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 pmd = pmd_alloc(mm, pud, address);
2483 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002484 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 pte = pte_alloc_map(mm, pmd, address);
2486 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002487 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488
Hugh Dickinsc74df322005-10-29 18:16:23 -07002489 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490}
2491
2492#ifndef __PAGETABLE_PUD_FOLDED
2493/*
2494 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002495 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002497int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002499 pud_t *new = pud_alloc_one(mm, address);
2500 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002501 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502
Hugh Dickins872fec12005-10-29 18:16:21 -07002503 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002504 if (pgd_present(*pgd)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505 pud_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002506 else
2507 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07002508 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002509 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510}
2511#endif /* __PAGETABLE_PUD_FOLDED */
2512
2513#ifndef __PAGETABLE_PMD_FOLDED
2514/*
2515 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002516 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002518int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002520 pmd_t *new = pmd_alloc_one(mm, address);
2521 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002522 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523
Hugh Dickins872fec12005-10-29 18:16:21 -07002524 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07002526 if (pud_present(*pud)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 pmd_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002528 else
2529 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07002531 if (pgd_present(*pud)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 pmd_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002533 else
2534 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07002536 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002537 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538}
2539#endif /* __PAGETABLE_PMD_FOLDED */
2540
2541int make_pages_present(unsigned long addr, unsigned long end)
2542{
2543 int ret, len, write;
2544 struct vm_area_struct * vma;
2545
2546 vma = find_vma(current->mm, addr);
2547 if (!vma)
2548 return -1;
2549 write = (vma->vm_flags & VM_WRITE) != 0;
Eric Sesterhenn5bcb28b2006-03-26 18:30:52 +02002550 BUG_ON(addr >= end);
2551 BUG_ON(end > vma->vm_end);
Rolf Eike Beer68e116a2007-07-15 23:38:03 -07002552 len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 ret = get_user_pages(current, current->mm, addr,
2554 len, write, 0, NULL, NULL);
2555 if (ret < 0)
2556 return ret;
2557 return ret == len ? 0 : -1;
2558}
2559
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560#if !defined(__HAVE_ARCH_GATE_AREA)
2561
2562#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07002563static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564
2565static int __init gate_vma_init(void)
2566{
2567 gate_vma.vm_mm = NULL;
2568 gate_vma.vm_start = FIXADDR_USER_START;
2569 gate_vma.vm_end = FIXADDR_USER_END;
Roland McGrathb6558c42007-01-26 00:56:47 -08002570 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
2571 gate_vma.vm_page_prot = __P101;
Roland McGrathf47aef52007-01-26 00:56:49 -08002572 /*
2573 * Make sure the vDSO gets into every core dump.
2574 * Dumping its contents makes post-mortem fully interpretable later
2575 * without matching up the same kernel and hardware config to see
2576 * what PC values meant.
2577 */
2578 gate_vma.vm_flags |= VM_ALWAYSDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 return 0;
2580}
2581__initcall(gate_vma_init);
2582#endif
2583
2584struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2585{
2586#ifdef AT_SYSINFO_EHDR
2587 return &gate_vma;
2588#else
2589 return NULL;
2590#endif
2591}
2592
2593int in_gate_area_no_task(unsigned long addr)
2594{
2595#ifdef AT_SYSINFO_EHDR
2596 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2597 return 1;
2598#endif
2599 return 0;
2600}
2601
2602#endif /* __HAVE_ARCH_GATE_AREA */
David Howells0ec76a12006-09-27 01:50:15 -07002603
2604/*
2605 * Access another process' address space.
2606 * Source/target buffer must be kernel space,
2607 * Do not walk the page table directly, use get_user_pages
2608 */
2609int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2610{
2611 struct mm_struct *mm;
2612 struct vm_area_struct *vma;
2613 struct page *page;
2614 void *old_buf = buf;
2615
2616 mm = get_task_mm(tsk);
2617 if (!mm)
2618 return 0;
2619
2620 down_read(&mm->mmap_sem);
Simon Arlott183ff222007-10-20 01:27:18 +02002621 /* ignore errors, just check how much was successfully transferred */
David Howells0ec76a12006-09-27 01:50:15 -07002622 while (len) {
2623 int bytes, ret, offset;
2624 void *maddr;
2625
2626 ret = get_user_pages(tsk, mm, addr, 1,
2627 write, 1, &page, &vma);
2628 if (ret <= 0)
2629 break;
2630
2631 bytes = len;
2632 offset = addr & (PAGE_SIZE-1);
2633 if (bytes > PAGE_SIZE-offset)
2634 bytes = PAGE_SIZE-offset;
2635
2636 maddr = kmap(page);
2637 if (write) {
2638 copy_to_user_page(vma, page, addr,
2639 maddr + offset, buf, bytes);
2640 set_page_dirty_lock(page);
2641 } else {
2642 copy_from_user_page(vma, page, addr,
2643 buf, maddr + offset, bytes);
2644 }
2645 kunmap(page);
2646 page_cache_release(page);
2647 len -= bytes;
2648 buf += bytes;
2649 addr += bytes;
2650 }
2651 up_read(&mm->mmap_sem);
2652 mmput(mm);
2653
2654 return buf - old_buf;
2655}
Andi Kleen03252912008-01-30 13:33:18 +01002656
2657/*
2658 * Print the name of a VMA.
2659 */
2660void print_vma_addr(char *prefix, unsigned long ip)
2661{
2662 struct mm_struct *mm = current->mm;
2663 struct vm_area_struct *vma;
2664
2665 down_read(&mm->mmap_sem);
2666 vma = find_vma(mm, ip);
2667 if (vma && vma->vm_file) {
2668 struct file *f = vma->vm_file;
2669 char *buf = (char *)__get_free_page(GFP_KERNEL);
2670 if (buf) {
2671 char *p, *s;
2672
2673 p = d_path(f->f_dentry, f->f_vfsmnt, buf, PAGE_SIZE);
2674 if (IS_ERR(p))
2675 p = "?";
2676 s = strrchr(p, '/');
2677 if (s)
2678 p = s+1;
2679 printk("%s%s[%lx+%lx]", prefix, p,
2680 vma->vm_start,
2681 vma->vm_end - vma->vm_start);
2682 free_page((unsigned long)buf);
2683 }
2684 }
2685 up_read(&current->mm->mmap_sem);
2686}