blob: eefd5b68bc42d76768e773a43ae310806c9fa17c [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
Nick Piggin315ab192006-03-25 16:20:22 +0100395 /*
396 * Add some anal sanity checks for now. Eventually,
397 * we should just do "return pfn_to_page(pfn)", but
398 * in the meantime we check that we get a valid pfn,
399 * and that the resulting page looks ok.
400 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800401 if (unlikely(!pfn_valid(pfn))) {
402 print_bad_pte(vma, pte, addr);
403 return NULL;
404 }
405
406 /*
407 * NOTE! We still have PageReserved() pages in the page
408 * tables.
409 *
410 * The PAGE_ZERO() pages and various VDSO mappings can
411 * cause them to exist.
412 */
413 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800414}
415
416/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 * copy one vm_area from one task to the other. Assumes the page tables
418 * already present in the new task to be cleared in the whole range
419 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 */
421
Hugh Dickins8c103762005-10-29 18:16:13 -0700422static inline void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700424 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700425 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426{
Nick Pigginb5810032005-10-29 18:16:12 -0700427 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 pte_t pte = *src_pte;
429 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430
431 /* pte contains position in swap or file, so copy. */
432 if (unlikely(!pte_present(pte))) {
433 if (!pte_file(pte)) {
Christoph Lameter06972122006-06-23 02:03:35 -0700434 swp_entry_t entry = pte_to_swp_entry(pte);
435
436 swap_duplicate(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 /* make sure dst_mm is on swapoff's mmlist. */
438 if (unlikely(list_empty(&dst_mm->mmlist))) {
439 spin_lock(&mmlist_lock);
Hugh Dickinsf412ac02005-10-29 18:16:41 -0700440 if (list_empty(&dst_mm->mmlist))
441 list_add(&dst_mm->mmlist,
442 &src_mm->mmlist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 spin_unlock(&mmlist_lock);
444 }
Christoph Lameter06972122006-06-23 02:03:35 -0700445 if (is_write_migration_entry(entry) &&
446 is_cow_mapping(vm_flags)) {
447 /*
448 * COW mappings require pages in both parent
449 * and child to be set to read.
450 */
451 make_migration_entry_read(&entry);
452 pte = swp_entry_to_pte(entry);
453 set_pte_at(src_mm, addr, src_pte, pte);
454 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700456 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 }
458
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 /*
460 * If it's a COW mapping, write protect it both
461 * in the parent and the child
462 */
Linus Torvalds67121172005-12-11 20:38:17 -0800463 if (is_cow_mapping(vm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 ptep_set_wrprotect(src_mm, addr, src_pte);
Zachary Amsden3dc90792006-09-30 23:29:30 -0700465 pte = pte_wrprotect(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 }
467
468 /*
469 * If it's a shared mapping, mark it clean in
470 * the child
471 */
472 if (vm_flags & VM_SHARED)
473 pte = pte_mkclean(pte);
474 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800475
476 page = vm_normal_page(vma, addr, pte);
477 if (page) {
478 get_page(page);
Nick Pigginc97a9e12007-05-16 22:11:21 -0700479 page_dup_rmap(page, vma, addr);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800480 rss[!!PageAnon(page)]++;
481 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700482
483out_set_pte:
484 set_pte_at(dst_mm, addr, dst_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485}
486
487static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
488 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
489 unsigned long addr, unsigned long end)
490{
491 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700492 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -0700493 int progress = 0;
Hugh Dickins8c103762005-10-29 18:16:13 -0700494 int rss[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
496again:
Hugh Dickinsae859762005-10-29 18:16:05 -0700497 rss[1] = rss[0] = 0;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700498 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 if (!dst_pte)
500 return -ENOMEM;
501 src_pte = pte_offset_map_nested(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700502 src_ptl = pte_lockptr(src_mm, src_pmd);
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700503 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700504 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 do {
507 /*
508 * We are holding two locks at this point - either of them
509 * could generate latencies in another task on another CPU.
510 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700511 if (progress >= 32) {
512 progress = 0;
513 if (need_resched() ||
Hugh Dickinsc74df322005-10-29 18:16:23 -0700514 need_lockbreak(src_ptl) ||
515 need_lockbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700516 break;
517 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 if (pte_none(*src_pte)) {
519 progress++;
520 continue;
521 }
Hugh Dickins8c103762005-10-29 18:16:13 -0700522 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 progress += 8;
524 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700526 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -0700527 spin_unlock(src_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 pte_unmap_nested(src_pte - 1);
Hugh Dickinsae859762005-10-29 18:16:05 -0700529 add_mm_rss(dst_mm, rss[0], rss[1]);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700530 pte_unmap_unlock(dst_pte - 1, dst_ptl);
531 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 if (addr != end)
533 goto again;
534 return 0;
535}
536
537static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
538 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
539 unsigned long addr, unsigned long end)
540{
541 pmd_t *src_pmd, *dst_pmd;
542 unsigned long next;
543
544 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
545 if (!dst_pmd)
546 return -ENOMEM;
547 src_pmd = pmd_offset(src_pud, addr);
548 do {
549 next = pmd_addr_end(addr, end);
550 if (pmd_none_or_clear_bad(src_pmd))
551 continue;
552 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
553 vma, addr, next))
554 return -ENOMEM;
555 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
556 return 0;
557}
558
559static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
560 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
561 unsigned long addr, unsigned long end)
562{
563 pud_t *src_pud, *dst_pud;
564 unsigned long next;
565
566 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
567 if (!dst_pud)
568 return -ENOMEM;
569 src_pud = pud_offset(src_pgd, addr);
570 do {
571 next = pud_addr_end(addr, end);
572 if (pud_none_or_clear_bad(src_pud))
573 continue;
574 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
575 vma, addr, next))
576 return -ENOMEM;
577 } while (dst_pud++, src_pud++, addr = next, addr != end);
578 return 0;
579}
580
581int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
582 struct vm_area_struct *vma)
583{
584 pgd_t *src_pgd, *dst_pgd;
585 unsigned long next;
586 unsigned long addr = vma->vm_start;
587 unsigned long end = vma->vm_end;
588
Nick Piggind9928952005-08-28 16:49:11 +1000589 /*
590 * Don't copy ptes where a page fault will fill them correctly.
591 * Fork becomes much lighter when there are big shared or private
592 * readonly mappings. The tradeoff is that copy_page_range is more
593 * efficient than faulting.
594 */
Linus Torvalds4d7672b2005-12-16 10:21:23 -0800595 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
Nick Piggind9928952005-08-28 16:49:11 +1000596 if (!vma->anon_vma)
597 return 0;
598 }
599
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 if (is_vm_hugetlb_page(vma))
601 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
602
603 dst_pgd = pgd_offset(dst_mm, addr);
604 src_pgd = pgd_offset(src_mm, addr);
605 do {
606 next = pgd_addr_end(addr, end);
607 if (pgd_none_or_clear_bad(src_pgd))
608 continue;
609 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
610 vma, addr, next))
611 return -ENOMEM;
612 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
613 return 0;
614}
615
Robin Holt51c6f662005-11-13 16:06:42 -0800616static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700617 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800619 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620{
Nick Pigginb5810032005-10-29 18:16:12 -0700621 struct mm_struct *mm = tlb->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 pte_t *pte;
Hugh Dickins508034a2005-10-29 18:16:30 -0700623 spinlock_t *ptl;
Hugh Dickinsae859762005-10-29 18:16:05 -0700624 int file_rss = 0;
625 int anon_rss = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626
Hugh Dickins508034a2005-10-29 18:16:30 -0700627 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700628 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 do {
630 pte_t ptent = *pte;
Robin Holt51c6f662005-11-13 16:06:42 -0800631 if (pte_none(ptent)) {
632 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800634 }
Hugh Dickins6f5e6b92006-03-16 23:04:09 -0800635
636 (*zap_work) -= PAGE_SIZE;
637
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800639 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -0800640
Linus Torvalds6aab3412005-11-28 14:34:23 -0800641 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 if (unlikely(details) && page) {
643 /*
644 * unmap_shared_mapping_pages() wants to
645 * invalidate cache without truncating:
646 * unmap shared but keep private pages.
647 */
648 if (details->check_mapping &&
649 details->check_mapping != page->mapping)
650 continue;
651 /*
652 * Each page->index must be checked when
653 * invalidating or truncating nonlinear.
654 */
655 if (details->nonlinear_vma &&
656 (page->index < details->first_index ||
657 page->index > details->last_index))
658 continue;
659 }
Nick Pigginb5810032005-10-29 18:16:12 -0700660 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -0700661 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 tlb_remove_tlb_entry(tlb, pte, addr);
663 if (unlikely(!page))
664 continue;
665 if (unlikely(details) && details->nonlinear_vma
666 && linear_page_index(details->nonlinear_vma,
667 addr) != page->index)
Nick Pigginb5810032005-10-29 18:16:12 -0700668 set_pte_at(mm, addr, pte,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 pgoff_to_pte(page->index));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 if (PageAnon(page))
Hugh Dickins86d912f2005-10-29 18:16:14 -0700671 anon_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700672 else {
673 if (pte_dirty(ptent))
674 set_page_dirty(page);
675 if (pte_young(ptent))
Ken Chendaa88c82007-02-10 01:43:18 -0800676 SetPageReferenced(page);
Hugh Dickins86d912f2005-10-29 18:16:14 -0700677 file_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700678 }
Nick Piggin7de6b802006-12-22 01:09:33 -0800679 page_remove_rmap(page, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 tlb_remove_page(tlb, page);
681 continue;
682 }
683 /*
684 * If details->check_mapping, we leave swap entries;
685 * if details->nonlinear_vma, we leave file entries.
686 */
687 if (unlikely(details))
688 continue;
689 if (!pte_file(ptent))
690 free_swap_and_cache(pte_to_swp_entry(ptent));
Zachary Amsden9888a1c2006-09-30 23:29:31 -0700691 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
Robin Holt51c6f662005-11-13 16:06:42 -0800692 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
Hugh Dickinsae859762005-10-29 18:16:05 -0700693
Hugh Dickins86d912f2005-10-29 18:16:14 -0700694 add_mm_rss(mm, file_rss, anon_rss);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700695 arch_leave_lazy_mmu_mode();
Hugh Dickins508034a2005-10-29 18:16:30 -0700696 pte_unmap_unlock(pte - 1, ptl);
Robin Holt51c6f662005-11-13 16:06:42 -0800697
698 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699}
700
Robin Holt51c6f662005-11-13 16:06:42 -0800701static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700702 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800704 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705{
706 pmd_t *pmd;
707 unsigned long next;
708
709 pmd = pmd_offset(pud, addr);
710 do {
711 next = pmd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800712 if (pmd_none_or_clear_bad(pmd)) {
713 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800715 }
716 next = zap_pte_range(tlb, vma, pmd, addr, next,
717 zap_work, details);
718 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
719
720 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721}
722
Robin Holt51c6f662005-11-13 16:06:42 -0800723static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700724 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800726 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727{
728 pud_t *pud;
729 unsigned long next;
730
731 pud = pud_offset(pgd, addr);
732 do {
733 next = pud_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800734 if (pud_none_or_clear_bad(pud)) {
735 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800737 }
738 next = zap_pmd_range(tlb, vma, pud, addr, next,
739 zap_work, details);
740 } while (pud++, addr = next, (addr != end && *zap_work > 0));
741
742 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743}
744
Robin Holt51c6f662005-11-13 16:06:42 -0800745static unsigned long unmap_page_range(struct mmu_gather *tlb,
746 struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800748 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749{
750 pgd_t *pgd;
751 unsigned long next;
752
753 if (details && !details->check_mapping && !details->nonlinear_vma)
754 details = NULL;
755
756 BUG_ON(addr >= end);
757 tlb_start_vma(tlb, vma);
758 pgd = pgd_offset(vma->vm_mm, addr);
759 do {
760 next = pgd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800761 if (pgd_none_or_clear_bad(pgd)) {
762 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800764 }
765 next = zap_pud_range(tlb, vma, pgd, addr, next,
766 zap_work, details);
767 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 tlb_end_vma(tlb, vma);
Robin Holt51c6f662005-11-13 16:06:42 -0800769
770 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771}
772
773#ifdef CONFIG_PREEMPT
774# define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
775#else
776/* No preempt: go for improved straight-line efficiency */
777# define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
778#endif
779
780/**
781 * unmap_vmas - unmap a range of memory covered by a list of vma's
782 * @tlbp: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 * @vma: the starting vma
784 * @start_addr: virtual address at which to start unmapping
785 * @end_addr: virtual address at which to end unmapping
786 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
787 * @details: details of nonlinear truncation or shared cache invalidation
788 *
Hugh Dickinsee39b372005-04-19 13:29:15 -0700789 * Returns the end address of the unmapping (restart addr if interrupted).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700791 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700793 * We aim to not hold locks for too long (for scheduling latency reasons).
794 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 * return the ending mmu_gather to the caller.
796 *
797 * Only addresses between `start' and `end' will be unmapped.
798 *
799 * The VMA list must be sorted in ascending virtual address order.
800 *
801 * unmap_vmas() assumes that the caller will flush the whole unmapped address
802 * range after unmap_vmas() returns. So the only responsibility here is to
803 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
804 * drops the lock and schedules.
805 */
Hugh Dickins508034a2005-10-29 18:16:30 -0700806unsigned long unmap_vmas(struct mmu_gather **tlbp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 struct vm_area_struct *vma, unsigned long start_addr,
808 unsigned long end_addr, unsigned long *nr_accounted,
809 struct zap_details *details)
810{
Robin Holt51c6f662005-11-13 16:06:42 -0800811 long zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
813 int tlb_start_valid = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -0700814 unsigned long start = start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700816 int fullmm = (*tlbp)->fullmm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817
818 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 unsigned long end;
820
821 start = max(vma->vm_start, start_addr);
822 if (start >= vma->vm_end)
823 continue;
824 end = min(vma->vm_end, end_addr);
825 if (end <= vma->vm_start)
826 continue;
827
828 if (vma->vm_flags & VM_ACCOUNT)
829 *nr_accounted += (end - start) >> PAGE_SHIFT;
830
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 while (start != end) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 if (!tlb_start_valid) {
833 tlb_start = start;
834 tlb_start_valid = 1;
835 }
836
Robin Holt51c6f662005-11-13 16:06:42 -0800837 if (unlikely(is_vm_hugetlb_page(vma))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 unmap_hugepage_range(vma, start, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800839 zap_work -= (end - start) /
840 (HPAGE_SIZE / PAGE_SIZE);
841 start = end;
842 } else
843 start = unmap_page_range(*tlbp, vma,
844 start, end, &zap_work, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845
Robin Holt51c6f662005-11-13 16:06:42 -0800846 if (zap_work > 0) {
847 BUG_ON(start != end);
848 break;
849 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850
851 tlb_finish_mmu(*tlbp, tlb_start, start);
852
853 if (need_resched() ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 (i_mmap_lock && need_lockbreak(i_mmap_lock))) {
855 if (i_mmap_lock) {
Hugh Dickins508034a2005-10-29 18:16:30 -0700856 *tlbp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 goto out;
858 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 }
861
Hugh Dickins508034a2005-10-29 18:16:30 -0700862 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 tlb_start_valid = 0;
Robin Holt51c6f662005-11-13 16:06:42 -0800864 zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 }
866 }
867out:
Hugh Dickinsee39b372005-04-19 13:29:15 -0700868 return start; /* which is now the end (or restart) address */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869}
870
871/**
872 * zap_page_range - remove user pages in a given range
873 * @vma: vm_area_struct holding the applicable pages
874 * @address: starting address of pages to zap
875 * @size: number of bytes to zap
876 * @details: details of nonlinear truncation or shared cache invalidation
877 */
Hugh Dickinsee39b372005-04-19 13:29:15 -0700878unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 unsigned long size, struct zap_details *details)
880{
881 struct mm_struct *mm = vma->vm_mm;
882 struct mmu_gather *tlb;
883 unsigned long end = address + size;
884 unsigned long nr_accounted = 0;
885
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -0700888 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -0700889 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
890 if (tlb)
891 tlb_finish_mmu(tlb, address, end);
Hugh Dickinsee39b372005-04-19 13:29:15 -0700892 return end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893}
894
895/*
896 * Do a quick page-table lookup for a single page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800898struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700899 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900{
901 pgd_t *pgd;
902 pud_t *pud;
903 pmd_t *pmd;
904 pte_t *ptep, pte;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700905 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 struct page *page;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800907 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700909 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
910 if (!IS_ERR(page)) {
911 BUG_ON(flags & FOLL_GET);
912 goto out;
913 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700915 page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 pgd = pgd_offset(mm, address);
917 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700918 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
920 pud = pud_offset(pgd, address);
921 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700922 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923
924 pmd = pmd_offset(pud, address);
925 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700926 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700928 if (pmd_huge(*pmd)) {
929 BUG_ON(flags & FOLL_GET);
930 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
931 goto out;
932 }
933
934 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 if (!ptep)
936 goto out;
937
938 pte = *ptep;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700939 if (!pte_present(pte))
940 goto unlock;
941 if ((flags & FOLL_WRITE) && !pte_write(pte))
942 goto unlock;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800943 page = vm_normal_page(vma, address, pte);
944 if (unlikely(!page))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700945 goto unlock;
946
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700947 if (flags & FOLL_GET)
948 get_page(page);
949 if (flags & FOLL_TOUCH) {
950 if ((flags & FOLL_WRITE) &&
951 !pte_dirty(pte) && !PageDirty(page))
952 set_page_dirty(page);
953 mark_page_accessed(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700955unlock:
956 pte_unmap_unlock(ptep, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957out:
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700958 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700960no_page_table:
961 /*
962 * When core dumping an enormous anonymous area that nobody
963 * has touched so far, we don't want to allocate page tables.
964 */
965 if (flags & FOLL_ANON) {
Nick Piggin557ed1f2007-10-16 01:24:40 -0700966 page = ZERO_PAGE(0);
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700967 if (flags & FOLL_GET)
968 get_page(page);
969 BUG_ON(flags & FOLL_WRITE);
970 }
971 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972}
973
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
975 unsigned long start, int len, int write, int force,
976 struct page **pages, struct vm_area_struct **vmas)
977{
978 int i;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700979 unsigned int vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980
981 /*
982 * Require read or write permissions.
983 * If 'force' is set, we only require the "MAY" flags.
984 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700985 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
986 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 i = 0;
988
989 do {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700990 struct vm_area_struct *vma;
991 unsigned int foll_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
993 vma = find_extend_vma(mm, start);
994 if (!vma && in_gate_area(tsk, start)) {
995 unsigned long pg = start & PAGE_MASK;
996 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
997 pgd_t *pgd;
998 pud_t *pud;
999 pmd_t *pmd;
1000 pte_t *pte;
1001 if (write) /* user gate pages are read-only */
1002 return i ? : -EFAULT;
1003 if (pg > TASK_SIZE)
1004 pgd = pgd_offset_k(pg);
1005 else
1006 pgd = pgd_offset_gate(mm, pg);
1007 BUG_ON(pgd_none(*pgd));
1008 pud = pud_offset(pgd, pg);
1009 BUG_ON(pud_none(*pud));
1010 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001011 if (pmd_none(*pmd))
1012 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001014 if (pte_none(*pte)) {
1015 pte_unmap(pte);
1016 return i ? : -EFAULT;
1017 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 if (pages) {
Nick Pigginfa2a4552005-11-29 18:43:17 +11001019 struct page *page = vm_normal_page(gate_vma, start, *pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001020 pages[i] = page;
1021 if (page)
1022 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 }
1024 pte_unmap(pte);
1025 if (vmas)
1026 vmas[i] = gate_vma;
1027 i++;
1028 start += PAGE_SIZE;
1029 len--;
1030 continue;
1031 }
1032
Linus Torvalds1ff80382005-12-12 16:24:33 -08001033 if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001034 || !(vm_flags & vma->vm_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 return i ? : -EFAULT;
1036
1037 if (is_vm_hugetlb_page(vma)) {
1038 i = follow_hugetlb_page(mm, vma, pages, vmas,
1039 &start, &len, i);
1040 continue;
1041 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001042
1043 foll_flags = FOLL_TOUCH;
1044 if (pages)
1045 foll_flags |= FOLL_GET;
1046 if (!write && !(vma->vm_flags & VM_LOCKED) &&
Nick Piggin54cb8822007-07-19 01:46:59 -07001047 (!vma->vm_ops || (!vma->vm_ops->nopage &&
1048 !vma->vm_ops->fault)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001049 foll_flags |= FOLL_ANON;
1050
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 do {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001052 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053
Ethan Solomita462e00c2007-07-15 23:38:16 -07001054 /*
1055 * If tsk is ooming, cut off its access to large memory
1056 * allocations. It has a pending SIGKILL, but it can't
1057 * be processed until returning to user space.
1058 */
1059 if (unlikely(test_tsk_thread_flag(tsk, TIF_MEMDIE)))
1060 return -ENOMEM;
1061
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001062 if (write)
1063 foll_flags |= FOLL_WRITE;
1064
1065 cond_resched();
Linus Torvalds6aab3412005-11-28 14:34:23 -08001066 while (!(page = follow_page(vma, start, foll_flags))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001067 int ret;
Nick Piggin83c54072007-07-19 01:47:05 -07001068 ret = handle_mm_fault(mm, vma, start,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001069 foll_flags & FOLL_WRITE);
Nick Piggin83c54072007-07-19 01:47:05 -07001070 if (ret & VM_FAULT_ERROR) {
1071 if (ret & VM_FAULT_OOM)
1072 return i ? i : -ENOMEM;
1073 else if (ret & VM_FAULT_SIGBUS)
1074 return i ? i : -EFAULT;
1075 BUG();
1076 }
1077 if (ret & VM_FAULT_MAJOR)
1078 tsk->maj_flt++;
1079 else
1080 tsk->min_flt++;
1081
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001082 /*
Nick Piggin83c54072007-07-19 01:47:05 -07001083 * The VM_FAULT_WRITE bit tells us that
1084 * do_wp_page has broken COW when necessary,
1085 * even if maybe_mkwrite decided not to set
1086 * pte_write. We can thus safely do subsequent
1087 * page lookups as if they were reads.
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001088 */
1089 if (ret & VM_FAULT_WRITE)
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001090 foll_flags &= ~FOLL_WRITE;
Nick Piggin83c54072007-07-19 01:47:05 -07001091
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07001092 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 }
1094 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001095 pages[i] = page;
James Bottomley03beb072006-03-26 01:36:57 -08001096
Russell Kinga6f36be2006-12-30 22:24:19 +00001097 flush_anon_page(vma, page, start);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001098 flush_dcache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 }
1100 if (vmas)
1101 vmas[i] = vma;
1102 i++;
1103 start += PAGE_SIZE;
1104 len--;
Hugh Dickins08ef4722005-06-21 17:15:10 -07001105 } while (len && start < vma->vm_end);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001106 } while (len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 return i;
1108}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109EXPORT_SYMBOL(get_user_pages);
1110
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001111pte_t * fastcall get_locked_pte(struct mm_struct *mm, unsigned long addr, spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001112{
1113 pgd_t * pgd = pgd_offset(mm, addr);
1114 pud_t * pud = pud_alloc(mm, pgd, addr);
1115 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001116 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001117 if (pmd)
1118 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1119 }
1120 return NULL;
1121}
1122
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001124 * This is the old fallback for page remapping.
1125 *
1126 * For historical reasons, it only allows reserved pages. Only
1127 * old drivers should use this, and they needed to mark their
1128 * pages reserved for the old functions anyway.
1129 */
1130static int insert_page(struct mm_struct *mm, unsigned long addr, struct page *page, pgprot_t prot)
1131{
1132 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001133 pte_t *pte;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001134 spinlock_t *ptl;
1135
1136 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001137 if (PageAnon(page))
Linus Torvalds238f58d2005-11-29 13:01:56 -08001138 goto out;
1139 retval = -ENOMEM;
1140 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001141 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001142 if (!pte)
1143 goto out;
1144 retval = -EBUSY;
1145 if (!pte_none(*pte))
1146 goto out_unlock;
1147
1148 /* Ok, finally just insert the thing.. */
1149 get_page(page);
1150 inc_mm_counter(mm, file_rss);
1151 page_add_file_rmap(page);
1152 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1153
1154 retval = 0;
1155out_unlock:
1156 pte_unmap_unlock(pte, ptl);
1157out:
1158 return retval;
1159}
1160
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001161/**
1162 * vm_insert_page - insert single page into user vma
1163 * @vma: user vma to map to
1164 * @addr: target user address of this page
1165 * @page: source kernel page
1166 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001167 * This allows drivers to insert individual pages they've allocated
1168 * into a user vma.
1169 *
1170 * The page has to be a nice clean _individual_ kernel allocation.
1171 * If you allocate a compound page, you need to have marked it as
1172 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001173 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001174 *
1175 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1176 * took an arbitrary page protection parameter. This doesn't allow
1177 * that. Your vma protection will have to be set up correctly, which
1178 * means that if you want a shared writable mapping, you'd better
1179 * ask for a shared writable mapping!
1180 *
1181 * The page does not need to be reserved.
1182 */
1183int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, struct page *page)
1184{
1185 if (addr < vma->vm_start || addr >= vma->vm_end)
1186 return -EFAULT;
1187 if (!page_count(page))
1188 return -EINVAL;
Linus Torvalds4d7672b2005-12-16 10:21:23 -08001189 vma->vm_flags |= VM_INSERTPAGE;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001190 return insert_page(vma->vm_mm, addr, page, vma->vm_page_prot);
1191}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001192EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001193
Nick Piggine0dc0d82007-02-12 00:51:36 -08001194/**
1195 * vm_insert_pfn - insert single pfn into user vma
1196 * @vma: user vma to map to
1197 * @addr: target user address of this page
1198 * @pfn: source kernel pfn
1199 *
1200 * Similar to vm_inert_page, this allows drivers to insert individual pages
1201 * they've allocated into a user vma. Same comments apply.
1202 *
1203 * This function should only be called from a vm_ops->fault handler, and
1204 * in that case the handler should return NULL.
1205 */
1206int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1207 unsigned long pfn)
1208{
1209 struct mm_struct *mm = vma->vm_mm;
1210 int retval;
1211 pte_t *pte, entry;
1212 spinlock_t *ptl;
1213
1214 BUG_ON(!(vma->vm_flags & VM_PFNMAP));
1215 BUG_ON(is_cow_mapping(vma->vm_flags));
1216
1217 retval = -ENOMEM;
1218 pte = get_locked_pte(mm, addr, &ptl);
1219 if (!pte)
1220 goto out;
1221 retval = -EBUSY;
1222 if (!pte_none(*pte))
1223 goto out_unlock;
1224
1225 /* Ok, finally just insert the thing.. */
1226 entry = pfn_pte(pfn, vma->vm_page_prot);
1227 set_pte_at(mm, addr, pte, entry);
1228 update_mmu_cache(vma, addr, entry);
1229
1230 retval = 0;
1231out_unlock:
1232 pte_unmap_unlock(pte, ptl);
1233
1234out:
1235 return retval;
1236}
1237EXPORT_SYMBOL(vm_insert_pfn);
1238
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001239/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 * maps a range of physical memory into the requested pages. the old
1241 * mappings are removed. any references to nonexistent pages results
1242 * in null mappings (currently treated as "copy-on-access")
1243 */
1244static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1245 unsigned long addr, unsigned long end,
1246 unsigned long pfn, pgprot_t prot)
1247{
1248 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001249 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250
Hugh Dickinsc74df322005-10-29 18:16:23 -07001251 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 if (!pte)
1253 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001254 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 do {
1256 BUG_ON(!pte_none(*pte));
Nick Pigginb5810032005-10-29 18:16:12 -07001257 set_pte_at(mm, addr, pte, pfn_pte(pfn, prot));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 pfn++;
1259 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001260 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001261 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 return 0;
1263}
1264
1265static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1266 unsigned long addr, unsigned long end,
1267 unsigned long pfn, pgprot_t prot)
1268{
1269 pmd_t *pmd;
1270 unsigned long next;
1271
1272 pfn -= addr >> PAGE_SHIFT;
1273 pmd = pmd_alloc(mm, pud, addr);
1274 if (!pmd)
1275 return -ENOMEM;
1276 do {
1277 next = pmd_addr_end(addr, end);
1278 if (remap_pte_range(mm, pmd, addr, next,
1279 pfn + (addr >> PAGE_SHIFT), prot))
1280 return -ENOMEM;
1281 } while (pmd++, addr = next, addr != end);
1282 return 0;
1283}
1284
1285static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1286 unsigned long addr, unsigned long end,
1287 unsigned long pfn, pgprot_t prot)
1288{
1289 pud_t *pud;
1290 unsigned long next;
1291
1292 pfn -= addr >> PAGE_SHIFT;
1293 pud = pud_alloc(mm, pgd, addr);
1294 if (!pud)
1295 return -ENOMEM;
1296 do {
1297 next = pud_addr_end(addr, end);
1298 if (remap_pmd_range(mm, pud, addr, next,
1299 pfn + (addr >> PAGE_SHIFT), prot))
1300 return -ENOMEM;
1301 } while (pud++, addr = next, addr != end);
1302 return 0;
1303}
1304
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001305/**
1306 * remap_pfn_range - remap kernel memory to userspace
1307 * @vma: user vma to map to
1308 * @addr: target user address to start at
1309 * @pfn: physical address of kernel memory
1310 * @size: size of map area
1311 * @prot: page protection flags for this mapping
1312 *
1313 * Note: this is only safe if the mm semaphore is held when called.
1314 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1316 unsigned long pfn, unsigned long size, pgprot_t prot)
1317{
1318 pgd_t *pgd;
1319 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001320 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 struct mm_struct *mm = vma->vm_mm;
1322 int err;
1323
1324 /*
1325 * Physically remapped pages are special. Tell the
1326 * rest of the world about it:
1327 * VM_IO tells people not to look at these pages
1328 * (accesses can have side effects).
Hugh Dickins0b14c172005-11-21 21:32:15 -08001329 * VM_RESERVED is specified all over the place, because
1330 * in 2.4 it kept swapout's vma scan off this vma; but
1331 * in 2.6 the LRU scan won't even find its pages, so this
1332 * flag means no more than count its pages in reserved_vm,
1333 * and omit it from core dump, even when VM_IO turned off.
Linus Torvalds6aab3412005-11-28 14:34:23 -08001334 * VM_PFNMAP tells the core MM that the base pages are just
1335 * raw PFN mappings, and do not have a "struct page" associated
1336 * with them.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001337 *
1338 * There's a horrible special case to handle copy-on-write
1339 * behaviour that some programs depend on. We mark the "original"
1340 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 */
Linus Torvalds67121172005-12-11 20:38:17 -08001342 if (is_cow_mapping(vma->vm_flags)) {
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001343 if (addr != vma->vm_start || end != vma->vm_end)
Linus Torvalds7fc7e2e2005-12-11 19:57:52 -08001344 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001345 vma->vm_pgoff = pfn;
1346 }
1347
Linus Torvalds6aab3412005-11-28 14:34:23 -08001348 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349
1350 BUG_ON(addr >= end);
1351 pfn -= addr >> PAGE_SHIFT;
1352 pgd = pgd_offset(mm, addr);
1353 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 do {
1355 next = pgd_addr_end(addr, end);
1356 err = remap_pud_range(mm, pgd, addr, next,
1357 pfn + (addr >> PAGE_SHIFT), prot);
1358 if (err)
1359 break;
1360 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 return err;
1362}
1363EXPORT_SYMBOL(remap_pfn_range);
1364
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001365static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1366 unsigned long addr, unsigned long end,
1367 pte_fn_t fn, void *data)
1368{
1369 pte_t *pte;
1370 int err;
1371 struct page *pmd_page;
Borislav Petkov94909912007-05-06 14:49:17 -07001372 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001373
1374 pte = (mm == &init_mm) ?
1375 pte_alloc_kernel(pmd, addr) :
1376 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1377 if (!pte)
1378 return -ENOMEM;
1379
1380 BUG_ON(pmd_huge(*pmd));
1381
1382 pmd_page = pmd_page(*pmd);
1383
1384 do {
1385 err = fn(pte, pmd_page, addr, data);
1386 if (err)
1387 break;
1388 } while (pte++, addr += PAGE_SIZE, addr != end);
1389
1390 if (mm != &init_mm)
1391 pte_unmap_unlock(pte-1, ptl);
1392 return err;
1393}
1394
1395static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1396 unsigned long addr, unsigned long end,
1397 pte_fn_t fn, void *data)
1398{
1399 pmd_t *pmd;
1400 unsigned long next;
1401 int err;
1402
1403 pmd = pmd_alloc(mm, pud, addr);
1404 if (!pmd)
1405 return -ENOMEM;
1406 do {
1407 next = pmd_addr_end(addr, end);
1408 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1409 if (err)
1410 break;
1411 } while (pmd++, addr = next, addr != end);
1412 return err;
1413}
1414
1415static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1416 unsigned long addr, unsigned long end,
1417 pte_fn_t fn, void *data)
1418{
1419 pud_t *pud;
1420 unsigned long next;
1421 int err;
1422
1423 pud = pud_alloc(mm, pgd, addr);
1424 if (!pud)
1425 return -ENOMEM;
1426 do {
1427 next = pud_addr_end(addr, end);
1428 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1429 if (err)
1430 break;
1431 } while (pud++, addr = next, addr != end);
1432 return err;
1433}
1434
1435/*
1436 * Scan a region of virtual memory, filling in page tables as necessary
1437 * and calling a provided function on each leaf page table.
1438 */
1439int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1440 unsigned long size, pte_fn_t fn, void *data)
1441{
1442 pgd_t *pgd;
1443 unsigned long next;
1444 unsigned long end = addr + size;
1445 int err;
1446
1447 BUG_ON(addr >= end);
1448 pgd = pgd_offset(mm, addr);
1449 do {
1450 next = pgd_addr_end(addr, end);
1451 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1452 if (err)
1453 break;
1454 } while (pgd++, addr = next, addr != end);
1455 return err;
1456}
1457EXPORT_SYMBOL_GPL(apply_to_page_range);
1458
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001460 * handle_pte_fault chooses page fault handler according to an entry
1461 * which was read non-atomically. Before making any commitment, on
1462 * those architectures or configurations (e.g. i386 with PAE) which
1463 * might give a mix of unmatched parts, do_swap_page and do_file_page
1464 * must check under lock before unmapping the pte and proceeding
1465 * (but do_wp_page is only called after already making such a check;
1466 * and do_anonymous_page and do_no_page can safely check later on).
1467 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001468static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001469 pte_t *page_table, pte_t orig_pte)
1470{
1471 int same = 1;
1472#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1473 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001474 spinlock_t *ptl = pte_lockptr(mm, pmd);
1475 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001476 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001477 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001478 }
1479#endif
1480 pte_unmap(page_table);
1481 return same;
1482}
1483
1484/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1486 * servicing faults for write access. In the normal case, do always want
1487 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1488 * that do not have writing enabled, when used by access_process_vm.
1489 */
1490static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1491{
1492 if (likely(vma->vm_flags & VM_WRITE))
1493 pte = pte_mkwrite(pte);
1494 return pte;
1495}
1496
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001497static 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 -08001498{
1499 /*
1500 * If the source page was a PFN mapping, we don't have
1501 * a "struct page" for it. We do a best-effort copy by
1502 * just copying from the original user address. If that
1503 * fails, we just zero-fill it. Live with it.
1504 */
1505 if (unlikely(!src)) {
1506 void *kaddr = kmap_atomic(dst, KM_USER0);
Linus Torvalds5d2a2db2005-11-29 14:07:55 -08001507 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1508
1509 /*
1510 * This really shouldn't fail, because the page is there
1511 * in the page tables. But it might just be unreadable,
1512 * in which case we just give up and fill the result with
1513 * zeroes.
1514 */
1515 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Linus Torvalds6aab3412005-11-28 14:34:23 -08001516 memset(kaddr, 0, PAGE_SIZE);
1517 kunmap_atomic(kaddr, KM_USER0);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07001518 flush_dcache_page(dst);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001519 return;
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001520
Linus Torvalds6aab3412005-11-28 14:34:23 -08001521 }
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001522 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001523}
1524
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 * This routine handles present pages, when users try to write
1527 * to a shared page. It is done by copying the page to a new address
1528 * and decrementing the shared-page counter for the old page.
1529 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 * Note that this routine assumes that the protection checks have been
1531 * done by the caller (the low-level page fault routine in most cases).
1532 * Thus we can safely just mark it writable once we've done any necessary
1533 * COW.
1534 *
1535 * We also mark the page dirty at this point even though the page will
1536 * change only once the write actually happens. This avoids a few races,
1537 * and potentially makes it more efficient.
1538 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001539 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1540 * but allow concurrent faults), with pte both mapped and locked.
1541 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001543static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1544 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001545 spinlock_t *ptl, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546{
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001547 struct page *old_page, *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 pte_t entry;
Nick Piggin83c54072007-07-19 01:47:05 -07001549 int reuse = 0, ret = 0;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001550 int page_mkwrite = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001551 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552
Linus Torvalds6aab3412005-11-28 14:34:23 -08001553 old_page = vm_normal_page(vma, address, orig_pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001554 if (!old_page)
1555 goto gotten;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001557 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001558 * Take out anonymous pages first, anonymous shared vmas are
1559 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001560 */
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001561 if (PageAnon(old_page)) {
1562 if (!TestSetPageLocked(old_page)) {
1563 reuse = can_share_swap_page(old_page);
1564 unlock_page(old_page);
1565 }
1566 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001567 (VM_WRITE|VM_SHARED))) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001568 /*
1569 * Only catch write-faults on shared writable pages,
1570 * read-only shared pages can get COWed by
1571 * get_user_pages(.write=1, .force=1).
1572 */
David Howells9637a5e2006-06-23 02:03:43 -07001573 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
1574 /*
1575 * Notify the address space that the page is about to
1576 * become writable so that it can prohibit this or wait
1577 * for the page to get into an appropriate state.
1578 *
1579 * We do this without the lock held, so that it can
1580 * sleep if it needs to.
1581 */
1582 page_cache_get(old_page);
1583 pte_unmap_unlock(page_table, ptl);
1584
1585 if (vma->vm_ops->page_mkwrite(vma, old_page) < 0)
1586 goto unwritable_page;
1587
David Howells9637a5e2006-06-23 02:03:43 -07001588 /*
1589 * Since we dropped the lock we need to revalidate
1590 * the PTE as someone else may have changed it. If
1591 * they did, we just return, as we can count on the
1592 * MMU to tell us if they didn't also make it writable.
1593 */
1594 page_table = pte_offset_map_lock(mm, pmd, address,
1595 &ptl);
Hugh Dickinsc3704ce2007-02-10 01:43:00 -08001596 page_cache_release(old_page);
David Howells9637a5e2006-06-23 02:03:43 -07001597 if (!pte_same(*page_table, orig_pte))
1598 goto unlock;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001599
1600 page_mkwrite = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001602 dirty_page = old_page;
1603 get_page(dirty_page);
David Howells9637a5e2006-06-23 02:03:43 -07001604 reuse = 1;
David Howells9637a5e2006-06-23 02:03:43 -07001605 }
1606
1607 if (reuse) {
1608 flush_cache_page(vma, address, pte_pfn(orig_pte));
1609 entry = pte_mkyoung(orig_pte);
1610 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
KAMEZAWA Hiroyuki954ffcb2007-10-16 01:25:44 -07001611 if (ptep_set_access_flags(vma, address, page_table, entry,1))
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07001612 update_mmu_cache(vma, address, entry);
David Howells9637a5e2006-06-23 02:03:43 -07001613 ret |= VM_FAULT_WRITE;
1614 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616
1617 /*
1618 * Ok, we need to copy. Oh, well..
1619 */
Nick Pigginb5810032005-10-29 18:16:12 -07001620 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08001621gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001622 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623
1624 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07001625 goto oom;
Nick Piggin557ed1f2007-10-16 01:24:40 -07001626 VM_BUG_ON(old_page == ZERO_PAGE(0));
1627 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
1628 if (!new_page)
1629 goto oom;
1630 cow_user_page(new_page, old_page, address, vma);
Hugh Dickins65500d22005-10-29 18:15:59 -07001631
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 /*
1633 * Re-check the pte - we dropped the lock
1634 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001635 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07001636 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08001637 if (old_page) {
Nick Piggin7de6b802006-12-22 01:09:33 -08001638 page_remove_rmap(old_page, vma);
Hugh Dickins920fc352005-11-21 21:32:17 -08001639 if (!PageAnon(old_page)) {
1640 dec_mm_counter(mm, file_rss);
1641 inc_mm_counter(mm, anon_rss);
1642 }
1643 } else
Hugh Dickins42946212005-10-29 18:16:05 -07001644 inc_mm_counter(mm, anon_rss);
Ben Collinseca35132005-11-29 11:45:26 -08001645 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07001646 entry = mk_pte(new_page, vma->vm_page_prot);
1647 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Siddha, Suresh B4ce072f2006-09-29 01:58:42 -07001648 /*
1649 * Clear the pte entry and flush it first, before updating the
1650 * pte with the new entry. This will avoid a race condition
1651 * seen in the presence of one thread doing SMC and another
1652 * thread doing COW.
1653 */
1654 ptep_clear_flush(vma, address, page_table);
1655 set_pte_at(mm, address, page_table, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07001656 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 lru_cache_add_active(new_page);
Nick Piggin9617d952006-01-06 00:11:12 -08001658 page_add_new_anon_rmap(new_page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659
1660 /* Free the old page.. */
1661 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001662 ret |= VM_FAULT_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 }
Hugh Dickins920fc352005-11-21 21:32:17 -08001664 if (new_page)
1665 page_cache_release(new_page);
1666 if (old_page)
1667 page_cache_release(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001668unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001669 pte_unmap_unlock(page_table, ptl);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001670 if (dirty_page) {
Nick Piggin79352892007-07-19 01:47:22 -07001671 /*
1672 * Yes, Virginia, this is actually required to prevent a race
1673 * with clear_page_dirty_for_io() from clearing the page dirty
1674 * bit after it clear all dirty ptes, but before a racing
1675 * do_wp_page installs a dirty pte.
1676 *
1677 * do_no_page is protected similarly.
1678 */
1679 wait_on_page_locked(dirty_page);
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001680 set_page_dirty_balance(dirty_page, page_mkwrite);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001681 put_page(dirty_page);
1682 }
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001683 return ret;
Hugh Dickins65500d22005-10-29 18:15:59 -07001684oom:
Hugh Dickins920fc352005-11-21 21:32:17 -08001685 if (old_page)
1686 page_cache_release(old_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 return VM_FAULT_OOM;
David Howells9637a5e2006-06-23 02:03:43 -07001688
1689unwritable_page:
1690 page_cache_release(old_page);
1691 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692}
1693
1694/*
1695 * Helper functions for unmap_mapping_range().
1696 *
1697 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1698 *
1699 * We have to restart searching the prio_tree whenever we drop the lock,
1700 * since the iterator is only valid while the lock is held, and anyway
1701 * a later vma might be split and reinserted earlier while lock dropped.
1702 *
1703 * The list of nonlinear vmas could be handled more efficiently, using
1704 * a placeholder, but handle it in the same way until a need is shown.
1705 * It is important to search the prio_tree before nonlinear list: a vma
1706 * may become nonlinear and be shifted from prio_tree to nonlinear list
1707 * while the lock is dropped; but never shifted from list to prio_tree.
1708 *
1709 * In order to make forward progress despite restarting the search,
1710 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1711 * quickly skip it next time around. Since the prio_tree search only
1712 * shows us those vmas affected by unmapping the range in question, we
1713 * can't efficiently keep all vmas in step with mapping->truncate_count:
1714 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1715 * mapping->truncate_count and vma->vm_truncate_count are protected by
1716 * i_mmap_lock.
1717 *
1718 * In order to make forward progress despite repeatedly restarting some
Hugh Dickinsee39b372005-04-19 13:29:15 -07001719 * large vma, note the restart_addr from unmap_vmas when it breaks out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 * and restart from that address when we reach that vma again. It might
1721 * have been split or merged, shrunk or extended, but never shifted: so
1722 * restart_addr remains valid so long as it remains in the vma's range.
1723 * unmap_mapping_range forces truncate_count to leap over page-aligned
1724 * values so we can save vma's restart_addr in its truncate_count field.
1725 */
1726#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1727
1728static void reset_vma_truncate_counts(struct address_space *mapping)
1729{
1730 struct vm_area_struct *vma;
1731 struct prio_tree_iter iter;
1732
1733 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1734 vma->vm_truncate_count = 0;
1735 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1736 vma->vm_truncate_count = 0;
1737}
1738
1739static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1740 unsigned long start_addr, unsigned long end_addr,
1741 struct zap_details *details)
1742{
1743 unsigned long restart_addr;
1744 int need_break;
1745
Nick Piggind00806b2007-07-19 01:46:57 -07001746 /*
1747 * files that support invalidating or truncating portions of the
Nick Piggind0217ac2007-07-19 01:47:03 -07001748 * file from under mmaped areas must have their ->fault function
Nick Piggin83c54072007-07-19 01:47:05 -07001749 * return a locked page (and set VM_FAULT_LOCKED in the return).
1750 * This provides synchronisation against concurrent unmapping here.
Nick Piggind00806b2007-07-19 01:46:57 -07001751 */
Nick Piggind00806b2007-07-19 01:46:57 -07001752
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753again:
1754 restart_addr = vma->vm_truncate_count;
1755 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1756 start_addr = restart_addr;
1757 if (start_addr >= end_addr) {
1758 /* Top of vma has been split off since last time */
1759 vma->vm_truncate_count = details->truncate_count;
1760 return 0;
1761 }
1762 }
1763
Hugh Dickinsee39b372005-04-19 13:29:15 -07001764 restart_addr = zap_page_range(vma, start_addr,
1765 end_addr - start_addr, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 need_break = need_resched() ||
1767 need_lockbreak(details->i_mmap_lock);
1768
Hugh Dickinsee39b372005-04-19 13:29:15 -07001769 if (restart_addr >= end_addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 /* We have now completed this vma: mark it so */
1771 vma->vm_truncate_count = details->truncate_count;
1772 if (!need_break)
1773 return 0;
1774 } else {
1775 /* Note restart_addr in vma's truncate_count field */
Hugh Dickinsee39b372005-04-19 13:29:15 -07001776 vma->vm_truncate_count = restart_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 if (!need_break)
1778 goto again;
1779 }
1780
1781 spin_unlock(details->i_mmap_lock);
1782 cond_resched();
1783 spin_lock(details->i_mmap_lock);
1784 return -EINTR;
1785}
1786
1787static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1788 struct zap_details *details)
1789{
1790 struct vm_area_struct *vma;
1791 struct prio_tree_iter iter;
1792 pgoff_t vba, vea, zba, zea;
1793
1794restart:
1795 vma_prio_tree_foreach(vma, &iter, root,
1796 details->first_index, details->last_index) {
1797 /* Skip quickly over those we have already dealt with */
1798 if (vma->vm_truncate_count == details->truncate_count)
1799 continue;
1800
1801 vba = vma->vm_pgoff;
1802 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1803 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1804 zba = details->first_index;
1805 if (zba < vba)
1806 zba = vba;
1807 zea = details->last_index;
1808 if (zea > vea)
1809 zea = vea;
1810
1811 if (unmap_mapping_range_vma(vma,
1812 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1813 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1814 details) < 0)
1815 goto restart;
1816 }
1817}
1818
1819static inline void unmap_mapping_range_list(struct list_head *head,
1820 struct zap_details *details)
1821{
1822 struct vm_area_struct *vma;
1823
1824 /*
1825 * In nonlinear VMAs there is no correspondence between virtual address
1826 * offset and file offset. So we must perform an exhaustive search
1827 * across *all* the pages in each nonlinear VMA, not just the pages
1828 * whose virtual address lies outside the file truncation point.
1829 */
1830restart:
1831 list_for_each_entry(vma, head, shared.vm_set.list) {
1832 /* Skip quickly over those we have already dealt with */
1833 if (vma->vm_truncate_count == details->truncate_count)
1834 continue;
1835 details->nonlinear_vma = vma;
1836 if (unmap_mapping_range_vma(vma, vma->vm_start,
1837 vma->vm_end, details) < 0)
1838 goto restart;
1839 }
1840}
1841
1842/**
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001843 * 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 -07001844 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 * @holebegin: byte in first page to unmap, relative to the start of
1846 * the underlying file. This will be rounded down to a PAGE_SIZE
1847 * boundary. Note that this is different from vmtruncate(), which
1848 * must keep the partial page. In contrast, we must get rid of
1849 * partial pages.
1850 * @holelen: size of prospective hole in bytes. This will be rounded
1851 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1852 * end of the file.
1853 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1854 * but 0 when invalidating pagecache, don't throw away private data.
1855 */
1856void unmap_mapping_range(struct address_space *mapping,
1857 loff_t const holebegin, loff_t const holelen, int even_cows)
1858{
1859 struct zap_details details;
1860 pgoff_t hba = holebegin >> PAGE_SHIFT;
1861 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1862
1863 /* Check for overflow. */
1864 if (sizeof(holelen) > sizeof(hlen)) {
1865 long long holeend =
1866 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1867 if (holeend & ~(long long)ULONG_MAX)
1868 hlen = ULONG_MAX - hba + 1;
1869 }
1870
1871 details.check_mapping = even_cows? NULL: mapping;
1872 details.nonlinear_vma = NULL;
1873 details.first_index = hba;
1874 details.last_index = hba + hlen - 1;
1875 if (details.last_index < details.first_index)
1876 details.last_index = ULONG_MAX;
1877 details.i_mmap_lock = &mapping->i_mmap_lock;
1878
1879 spin_lock(&mapping->i_mmap_lock);
1880
Nick Piggind00806b2007-07-19 01:46:57 -07001881 /* Protect against endless unmapping loops */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 mapping->truncate_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 if (unlikely(is_restart_addr(mapping->truncate_count))) {
1884 if (mapping->truncate_count == 0)
1885 reset_vma_truncate_counts(mapping);
1886 mapping->truncate_count++;
1887 }
1888 details.truncate_count = mapping->truncate_count;
1889
1890 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
1891 unmap_mapping_range_tree(&mapping->i_mmap, &details);
1892 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
1893 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
1894 spin_unlock(&mapping->i_mmap_lock);
1895}
1896EXPORT_SYMBOL(unmap_mapping_range);
1897
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001898/**
1899 * vmtruncate - unmap mappings "freed" by truncate() syscall
1900 * @inode: inode of the file used
1901 * @offset: file offset to start truncating
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 *
1903 * NOTE! We have to be ready to update the memory sharing
1904 * between the file and the memory map for a potential last
1905 * incomplete page. Ugly, but necessary.
1906 */
1907int vmtruncate(struct inode * inode, loff_t offset)
1908{
1909 struct address_space *mapping = inode->i_mapping;
1910 unsigned long limit;
1911
1912 if (inode->i_size < offset)
1913 goto do_expand;
1914 /*
1915 * truncation of in-use swapfiles is disallowed - it would cause
1916 * subsequent swapout to scribble on the now-freed blocks.
1917 */
1918 if (IS_SWAPFILE(inode))
1919 goto out_busy;
1920 i_size_write(inode, offset);
Nick Piggind00806b2007-07-19 01:46:57 -07001921
1922 /*
1923 * unmap_mapping_range is called twice, first simply for efficiency
1924 * so that truncate_inode_pages does fewer single-page unmaps. However
1925 * after this first call, and before truncate_inode_pages finishes,
1926 * it is possible for private pages to be COWed, which remain after
1927 * truncate_inode_pages finishes, hence the second unmap_mapping_range
1928 * call must be made for correctness.
1929 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
1931 truncate_inode_pages(mapping, offset);
Nick Piggind00806b2007-07-19 01:46:57 -07001932 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 goto out_truncate;
1934
1935do_expand:
1936 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1937 if (limit != RLIM_INFINITY && offset > limit)
1938 goto out_sig;
1939 if (offset > inode->i_sb->s_maxbytes)
1940 goto out_big;
1941 i_size_write(inode, offset);
1942
1943out_truncate:
1944 if (inode->i_op && inode->i_op->truncate)
1945 inode->i_op->truncate(inode);
1946 return 0;
1947out_sig:
1948 send_sig(SIGXFSZ, current, 0);
1949out_big:
1950 return -EFBIG;
1951out_busy:
1952 return -ETXTBSY;
1953}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954EXPORT_SYMBOL(vmtruncate);
1955
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001956int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
1957{
1958 struct address_space *mapping = inode->i_mapping;
1959
1960 /*
1961 * If the underlying filesystem is not going to provide
1962 * a way to truncate a range of blocks (punch a hole) -
1963 * we should return failure right now.
1964 */
1965 if (!inode->i_op || !inode->i_op->truncate_range)
1966 return -ENOSYS;
1967
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001968 mutex_lock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001969 down_write(&inode->i_alloc_sem);
1970 unmap_mapping_range(mapping, offset, (end - offset), 1);
1971 truncate_inode_pages_range(mapping, offset, end);
Nick Piggind00806b2007-07-19 01:46:57 -07001972 unmap_mapping_range(mapping, offset, (end - offset), 1);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001973 inode->i_op->truncate_range(inode, offset, end);
1974 up_write(&inode->i_alloc_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001975 mutex_unlock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001976
1977 return 0;
1978}
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001979
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001980/**
1981 * swapin_readahead - swap in pages in hope we need them soon
1982 * @entry: swap entry of this memory
1983 * @addr: address to start
1984 * @vma: user vma this addresses belong to
1985 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 * Primitive swap readahead code. We simply read an aligned block of
1987 * (1 << page_cluster) entries in the swap area. This method is chosen
1988 * because it doesn't cost us any seek time. We also make sure to queue
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001989 * the 'original' request together with the readahead ones...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 *
1991 * This has been extended to use the NUMA policies from the mm triggering
1992 * the readahead.
1993 *
1994 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
1995 */
1996void swapin_readahead(swp_entry_t entry, unsigned long addr,struct vm_area_struct *vma)
1997{
1998#ifdef CONFIG_NUMA
1999 struct vm_area_struct *next_vma = vma ? vma->vm_next : NULL;
2000#endif
2001 int i, num;
2002 struct page *new_page;
2003 unsigned long offset;
2004
2005 /*
2006 * Get the number of handles we should do readahead io to.
2007 */
2008 num = valid_swaphandles(entry, &offset);
2009 for (i = 0; i < num; offset++, i++) {
2010 /* Ok, do the async read-ahead now */
2011 new_page = read_swap_cache_async(swp_entry(swp_type(entry),
2012 offset), vma, addr);
2013 if (!new_page)
2014 break;
2015 page_cache_release(new_page);
2016#ifdef CONFIG_NUMA
2017 /*
2018 * Find the next applicable VMA for the NUMA policy.
2019 */
2020 addr += PAGE_SIZE;
2021 if (addr == 0)
2022 vma = NULL;
2023 if (vma) {
2024 if (addr >= vma->vm_end) {
2025 vma = next_vma;
2026 next_vma = vma ? vma->vm_next : NULL;
2027 }
2028 if (vma && addr < vma->vm_start)
2029 vma = NULL;
2030 } else {
2031 if (next_vma && addr >= next_vma->vm_start) {
2032 vma = next_vma;
2033 next_vma = vma->vm_next;
2034 }
2035 }
2036#endif
2037 }
2038 lru_add_drain(); /* Push any new pages onto the LRU now */
2039}
2040
2041/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002042 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2043 * but allow concurrent faults), and pte mapped but not yet locked.
2044 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002046static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2047 unsigned long address, pte_t *page_table, pmd_t *pmd,
2048 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002050 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 struct page *page;
Hugh Dickins65500d22005-10-29 18:15:59 -07002052 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 pte_t pte;
Nick Piggin83c54072007-07-19 01:47:05 -07002054 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002056 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002057 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002058
2059 entry = pte_to_swp_entry(orig_pte);
Christoph Lameter06972122006-06-23 02:03:35 -07002060 if (is_migration_entry(entry)) {
2061 migration_entry_wait(mm, pmd, address);
2062 goto out;
2063 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002064 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 page = lookup_swap_cache(entry);
2066 if (!page) {
Ashwin Chaugule098fe652006-12-06 20:31:54 -08002067 grab_swap_token(); /* Contend for token _before_ read-in */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 swapin_readahead(entry, address, vma);
2069 page = read_swap_cache_async(entry, vma, address);
2070 if (!page) {
2071 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002072 * Back out if somebody else faulted in this pte
2073 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002075 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 if (likely(pte_same(*page_table, orig_pte)))
2077 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002078 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002079 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 }
2081
2082 /* Had to read the page from swap area: Major fault */
2083 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002084 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 }
2086
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002087 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 mark_page_accessed(page);
2089 lock_page(page);
2090
2091 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002092 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002094 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002095 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002096 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002097
2098 if (unlikely(!PageUptodate(page))) {
2099 ret = VM_FAULT_SIGBUS;
2100 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 }
2102
2103 /* The page isn't present yet, go ahead with the fault. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104
Hugh Dickins42946212005-10-29 18:16:05 -07002105 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 pte = mk_pte(page, vma->vm_page_prot);
2107 if (write_access && can_share_swap_page(page)) {
2108 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2109 write_access = 0;
2110 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111
2112 flush_icache_page(vma, page);
2113 set_pte_at(mm, address, page_table, pte);
2114 page_add_anon_rmap(page, vma, address);
2115
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002116 swap_free(entry);
2117 if (vm_swap_full())
2118 remove_exclusive_swap_page(page);
2119 unlock_page(page);
2120
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 if (write_access) {
Nick Piggin83c54072007-07-19 01:47:05 -07002122 /* XXX: We could OR the do_wp_page code with this one? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 if (do_wp_page(mm, vma, address,
Nick Piggin83c54072007-07-19 01:47:05 -07002124 page_table, pmd, ptl, pte) & VM_FAULT_OOM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 ret = VM_FAULT_OOM;
2126 goto out;
2127 }
2128
2129 /* No need to invalidate - it was non-present before */
2130 update_mmu_cache(vma, address, pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07002131unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002132 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133out:
2134 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002135out_nomap:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002136 pte_unmap_unlock(page_table, ptl);
Kirill Korotaevb8107482005-05-16 21:53:50 -07002137 unlock_page(page);
2138 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002139 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140}
2141
2142/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002143 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2144 * but allow concurrent faults), and pte mapped but not yet locked.
2145 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002147static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2148 unsigned long address, pte_t *page_table, pmd_t *pmd,
2149 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002151 struct page *page;
2152 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154
Nick Piggin557ed1f2007-10-16 01:24:40 -07002155 /* Allocate our own private page. */
2156 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157
Nick Piggin557ed1f2007-10-16 01:24:40 -07002158 if (unlikely(anon_vma_prepare(vma)))
2159 goto oom;
2160 page = alloc_zeroed_user_highpage_movable(vma, address);
2161 if (!page)
2162 goto oom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163
Nick Piggin557ed1f2007-10-16 01:24:40 -07002164 entry = mk_pte(page, vma->vm_page_prot);
2165 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002166
Nick Piggin557ed1f2007-10-16 01:24:40 -07002167 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2168 if (!pte_none(*page_table))
2169 goto release;
2170 inc_mm_counter(mm, anon_rss);
2171 lru_cache_add_active(page);
2172 page_add_new_anon_rmap(page, vma, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002173 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174
2175 /* No need to invalidate - it was non-present before */
Hugh Dickins65500d22005-10-29 18:15:59 -07002176 update_mmu_cache(vma, address, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002177unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002178 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002179 return 0;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002180release:
2181 page_cache_release(page);
2182 goto unlock;
Hugh Dickins65500d22005-10-29 18:15:59 -07002183oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184 return VM_FAULT_OOM;
2185}
2186
2187/*
Nick Piggin54cb8822007-07-19 01:46:59 -07002188 * __do_fault() tries to create a new page mapping. It aggressively
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 * tries to share with existing pages, but makes a separate copy if
Nick Piggin54cb8822007-07-19 01:46:59 -07002190 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2191 * the next page fault.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 *
2193 * As this is called only for pages that do not currently exist, we
2194 * do not need to flush old virtual caches or the TLB.
2195 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002196 * We enter with non-exclusive mmap_sem (to exclude vma changes,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002197 * but allow concurrent faults), and pte neither mapped nor locked.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002198 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 */
Nick Piggin54cb8822007-07-19 01:46:59 -07002200static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002201 unsigned long address, pmd_t *pmd,
Nick Piggin54cb8822007-07-19 01:46:59 -07002202 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203{
Hugh Dickins16abfa02007-10-04 16:56:06 +01002204 pte_t *page_table;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002205 spinlock_t *ptl;
Nick Piggind0217ac2007-07-19 01:47:03 -07002206 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 int anon = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002209 struct page *dirty_page = NULL;
Nick Piggind0217ac2007-07-19 01:47:03 -07002210 struct vm_fault vmf;
2211 int ret;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002212 int page_mkwrite = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07002213
Nick Piggind0217ac2007-07-19 01:47:03 -07002214 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2215 vmf.pgoff = pgoff;
2216 vmf.flags = flags;
2217 vmf.page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218
Hugh Dickins325f04d2005-11-29 16:55:48 +00002219 BUG_ON(vma->vm_flags & VM_PFNMAP);
2220
Nick Piggin54cb8822007-07-19 01:46:59 -07002221 if (likely(vma->vm_ops->fault)) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002222 ret = vma->vm_ops->fault(vma, &vmf);
Nick Piggin83c54072007-07-19 01:47:05 -07002223 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2224 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07002225 } else {
2226 /* Legacy ->nopage path */
Nick Piggin83c54072007-07-19 01:47:05 -07002227 ret = 0;
Nick Piggind0217ac2007-07-19 01:47:03 -07002228 vmf.page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret);
Nick Piggin54cb8822007-07-19 01:46:59 -07002229 /* no page was available -- either SIGBUS or OOM */
Nick Piggind0217ac2007-07-19 01:47:03 -07002230 if (unlikely(vmf.page == NOPAGE_SIGBUS))
Nick Piggin54cb8822007-07-19 01:46:59 -07002231 return VM_FAULT_SIGBUS;
Nick Piggind0217ac2007-07-19 01:47:03 -07002232 else if (unlikely(vmf.page == NOPAGE_OOM))
Nick Piggin54cb8822007-07-19 01:46:59 -07002233 return VM_FAULT_OOM;
2234 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235
Nick Piggind00806b2007-07-19 01:46:57 -07002236 /*
Nick Piggind0217ac2007-07-19 01:47:03 -07002237 * For consistency in subsequent calls, make the faulted page always
Nick Piggind00806b2007-07-19 01:46:57 -07002238 * locked.
2239 */
Nick Piggin83c54072007-07-19 01:47:05 -07002240 if (unlikely(!(ret & VM_FAULT_LOCKED)))
Nick Piggind0217ac2007-07-19 01:47:03 -07002241 lock_page(vmf.page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002242 else
Nick Piggind0217ac2007-07-19 01:47:03 -07002243 VM_BUG_ON(!PageLocked(vmf.page));
Nick Piggind00806b2007-07-19 01:46:57 -07002244
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 /*
2246 * Should we do an early C-O-W break?
2247 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002248 page = vmf.page;
Nick Piggin54cb8822007-07-19 01:46:59 -07002249 if (flags & FAULT_FLAG_WRITE) {
David Howells9637a5e2006-06-23 02:03:43 -07002250 if (!(vma->vm_flags & VM_SHARED)) {
Nick Piggin54cb8822007-07-19 01:46:59 -07002251 anon = 1;
Nick Piggind00806b2007-07-19 01:46:57 -07002252 if (unlikely(anon_vma_prepare(vma))) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002253 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002254 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002255 }
Nick Piggin83c54072007-07-19 01:47:05 -07002256 page = alloc_page_vma(GFP_HIGHUSER_MOVABLE,
2257 vma, address);
Nick Piggind00806b2007-07-19 01:46:57 -07002258 if (!page) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002259 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002260 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002261 }
Nick Piggind0217ac2007-07-19 01:47:03 -07002262 copy_user_highpage(page, vmf.page, address, vma);
David Howells9637a5e2006-06-23 02:03:43 -07002263 } else {
Nick Piggin54cb8822007-07-19 01:46:59 -07002264 /*
2265 * If the page will be shareable, see if the backing
David Howells9637a5e2006-06-23 02:03:43 -07002266 * address space wants to know that the page is about
Nick Piggin54cb8822007-07-19 01:46:59 -07002267 * to become writable
2268 */
Mark Fasheh69676142007-07-19 01:47:00 -07002269 if (vma->vm_ops->page_mkwrite) {
2270 unlock_page(page);
2271 if (vma->vm_ops->page_mkwrite(vma, page) < 0) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002272 ret = VM_FAULT_SIGBUS;
2273 anon = 1; /* no anon but release vmf.page */
Mark Fasheh69676142007-07-19 01:47:00 -07002274 goto out_unlocked;
2275 }
2276 lock_page(page);
Nick Piggind0217ac2007-07-19 01:47:03 -07002277 /*
2278 * XXX: this is not quite right (racy vs
2279 * invalidate) to unlock and relock the page
2280 * like this, however a better fix requires
2281 * reworking page_mkwrite locking API, which
2282 * is better done later.
2283 */
2284 if (!page->mapping) {
Nick Piggin83c54072007-07-19 01:47:05 -07002285 ret = 0;
Nick Piggind0217ac2007-07-19 01:47:03 -07002286 anon = 1; /* no anon but release vmf.page */
2287 goto out;
2288 }
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002289 page_mkwrite = 1;
David Howells9637a5e2006-06-23 02:03:43 -07002290 }
2291 }
Nick Piggin54cb8822007-07-19 01:46:59 -07002292
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 }
2294
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002295 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296
2297 /*
2298 * This silly early PAGE_DIRTY setting removes a race
2299 * due to the bad i386 page protection. But it's valid
2300 * for other architectures too.
2301 *
2302 * Note that if write_access is true, we either now have
2303 * an exclusive copy of the page, or this is a shared mapping,
2304 * so we can make it writable and dirty to avoid having to
2305 * handle that later.
2306 */
2307 /* Only go through if we didn't race with anybody else... */
Nick Piggin54cb8822007-07-19 01:46:59 -07002308 if (likely(pte_same(*page_table, orig_pte))) {
Nick Piggind00806b2007-07-19 01:46:57 -07002309 flush_icache_page(vma, page);
2310 entry = mk_pte(page, vma->vm_page_prot);
Nick Piggin54cb8822007-07-19 01:46:59 -07002311 if (flags & FAULT_FLAG_WRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2313 set_pte_at(mm, address, page_table, entry);
2314 if (anon) {
Nick Piggind00806b2007-07-19 01:46:57 -07002315 inc_mm_counter(mm, anon_rss);
2316 lru_cache_add_active(page);
2317 page_add_new_anon_rmap(page, vma, address);
Hugh Dickinsf57e88a2005-11-21 21:32:19 -08002318 } else {
Hugh Dickins42946212005-10-29 18:16:05 -07002319 inc_mm_counter(mm, file_rss);
Nick Piggind00806b2007-07-19 01:46:57 -07002320 page_add_file_rmap(page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002321 if (flags & FAULT_FLAG_WRITE) {
Nick Piggind00806b2007-07-19 01:46:57 -07002322 dirty_page = page;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002323 get_page(dirty_page);
2324 }
Hugh Dickins42946212005-10-29 18:16:05 -07002325 }
Nick Piggind00806b2007-07-19 01:46:57 -07002326
2327 /* no need to invalidate: a not-present page won't be cached */
2328 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 } else {
Nick Piggind00806b2007-07-19 01:46:57 -07002330 if (anon)
2331 page_cache_release(page);
2332 else
Nick Piggin54cb8822007-07-19 01:46:59 -07002333 anon = 1; /* no anon but release faulted_page */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 }
2335
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002336 pte_unmap_unlock(page_table, ptl);
Nick Piggind00806b2007-07-19 01:46:57 -07002337
2338out:
Nick Piggind0217ac2007-07-19 01:47:03 -07002339 unlock_page(vmf.page);
Mark Fasheh69676142007-07-19 01:47:00 -07002340out_unlocked:
Nick Piggind00806b2007-07-19 01:46:57 -07002341 if (anon)
Nick Piggind0217ac2007-07-19 01:47:03 -07002342 page_cache_release(vmf.page);
Nick Piggind00806b2007-07-19 01:46:57 -07002343 else if (dirty_page) {
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002344 set_page_dirty_balance(dirty_page, page_mkwrite);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002345 put_page(dirty_page);
2346 }
Nick Piggind00806b2007-07-19 01:46:57 -07002347
Nick Piggin83c54072007-07-19 01:47:05 -07002348 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07002349}
Nick Piggind00806b2007-07-19 01:46:57 -07002350
Nick Piggin54cb8822007-07-19 01:46:59 -07002351static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2352 unsigned long address, pte_t *page_table, pmd_t *pmd,
2353 int write_access, pte_t orig_pte)
2354{
2355 pgoff_t pgoff = (((address & PAGE_MASK)
Dean Nelson0da7e012007-10-16 01:24:45 -07002356 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
Nick Piggin54cb8822007-07-19 01:46:59 -07002357 unsigned int flags = (write_access ? FAULT_FLAG_WRITE : 0);
2358
Hugh Dickins16abfa02007-10-04 16:56:06 +01002359 pte_unmap(page_table);
2360 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Nick Piggin54cb8822007-07-19 01:46:59 -07002361}
2362
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363
2364/*
Jes Sorensenf4b81802006-09-27 01:50:10 -07002365 * do_no_pfn() tries to create a new page mapping for a page without
2366 * a struct_page backing it
2367 *
2368 * As this is called only for pages that do not currently exist, we
2369 * do not need to flush old virtual caches or the TLB.
2370 *
2371 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2372 * but allow concurrent faults), and pte mapped but not yet locked.
2373 * We return with mmap_sem still held, but pte unmapped and unlocked.
2374 *
2375 * It is expected that the ->nopfn handler always returns the same pfn
2376 * for a given virtual mapping.
2377 *
2378 * Mark this `noinline' to prevent it from bloating the main pagefault code.
2379 */
2380static noinline int do_no_pfn(struct mm_struct *mm, struct vm_area_struct *vma,
2381 unsigned long address, pte_t *page_table, pmd_t *pmd,
2382 int write_access)
2383{
2384 spinlock_t *ptl;
2385 pte_t entry;
2386 unsigned long pfn;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002387
2388 pte_unmap(page_table);
2389 BUG_ON(!(vma->vm_flags & VM_PFNMAP));
2390 BUG_ON(is_cow_mapping(vma->vm_flags));
2391
2392 pfn = vma->vm_ops->nopfn(vma, address & PAGE_MASK);
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002393 if (unlikely(pfn == NOPFN_OOM))
Jes Sorensenf4b81802006-09-27 01:50:10 -07002394 return VM_FAULT_OOM;
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002395 else if (unlikely(pfn == NOPFN_SIGBUS))
Jes Sorensenf4b81802006-09-27 01:50:10 -07002396 return VM_FAULT_SIGBUS;
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002397 else if (unlikely(pfn == NOPFN_REFAULT))
Nick Piggin83c54072007-07-19 01:47:05 -07002398 return 0;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002399
2400 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2401
2402 /* Only go through if we didn't race with anybody else... */
2403 if (pte_none(*page_table)) {
2404 entry = pfn_pte(pfn, vma->vm_page_prot);
2405 if (write_access)
2406 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2407 set_pte_at(mm, address, page_table, entry);
2408 }
2409 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002410 return 0;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002411}
2412
2413/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414 * Fault of a previously existing named mapping. Repopulate the pte
2415 * from the encoded file_pte if possible. This enables swappable
2416 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002417 *
2418 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2419 * but allow concurrent faults), and pte mapped but not yet locked.
2420 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002422static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins65500d22005-10-29 18:15:59 -07002423 unsigned long address, pte_t *page_table, pmd_t *pmd,
2424 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425{
Nick Piggind0217ac2007-07-19 01:47:03 -07002426 unsigned int flags = FAULT_FLAG_NONLINEAR |
2427 (write_access ? FAULT_FLAG_WRITE : 0);
Hugh Dickins65500d22005-10-29 18:15:59 -07002428 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002430 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Nick Piggin83c54072007-07-19 01:47:05 -07002431 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432
Nick Piggind0217ac2007-07-19 01:47:03 -07002433 if (unlikely(!(vma->vm_flags & VM_NONLINEAR) ||
2434 !(vma->vm_flags & VM_CAN_NONLINEAR))) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002435 /*
2436 * Page table corrupted: show pte and kill process.
2437 */
Nick Pigginb5810032005-10-29 18:16:12 -07002438 print_bad_pte(vma, orig_pte, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002439 return VM_FAULT_OOM;
2440 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002441
2442 pgoff = pte_to_pgoff(orig_pte);
Hugh Dickins16abfa02007-10-04 16:56:06 +01002443 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444}
2445
2446/*
2447 * These routines also need to handle stuff like marking pages dirty
2448 * and/or accessed for architectures that don't do it in hardware (most
2449 * RISC architectures). The early dirtying is also good on the i386.
2450 *
2451 * There is also a hook called "update_mmu_cache()" that architectures
2452 * with external mmu caches can use to update those (ie the Sparc or
2453 * PowerPC hashed page tables that act as extended TLBs).
2454 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07002455 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2456 * but allow concurrent faults), and pte mapped but not yet locked.
2457 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458 */
2459static inline int handle_pte_fault(struct mm_struct *mm,
Hugh Dickins65500d22005-10-29 18:15:59 -07002460 struct vm_area_struct *vma, unsigned long address,
2461 pte_t *pte, pmd_t *pmd, int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462{
2463 pte_t entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002464 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002466 entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002468 if (pte_none(entry)) {
Jes Sorensenf4b81802006-09-27 01:50:10 -07002469 if (vma->vm_ops) {
Nick Piggin54cb8822007-07-19 01:46:59 -07002470 if (vma->vm_ops->fault || vma->vm_ops->nopage)
2471 return do_linear_fault(mm, vma, address,
2472 pte, pmd, write_access, entry);
Jes Sorensenf4b81802006-09-27 01:50:10 -07002473 if (unlikely(vma->vm_ops->nopfn))
2474 return do_no_pfn(mm, vma, address, pte,
2475 pmd, write_access);
2476 }
2477 return do_anonymous_page(mm, vma, address,
2478 pte, pmd, write_access);
Hugh Dickins65500d22005-10-29 18:15:59 -07002479 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480 if (pte_file(entry))
Nick Piggind0217ac2007-07-19 01:47:03 -07002481 return do_nonlinear_fault(mm, vma, address,
Hugh Dickins65500d22005-10-29 18:15:59 -07002482 pte, pmd, write_access, entry);
2483 return do_swap_page(mm, vma, address,
2484 pte, pmd, write_access, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 }
2486
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002487 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002488 spin_lock(ptl);
2489 if (unlikely(!pte_same(*pte, entry)))
2490 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 if (write_access) {
2492 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002493 return do_wp_page(mm, vma, address,
2494 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 entry = pte_mkdirty(entry);
2496 }
2497 entry = pte_mkyoung(entry);
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002498 if (ptep_set_access_flags(vma, address, pte, entry, write_access)) {
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002499 update_mmu_cache(vma, address, entry);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002500 } else {
2501 /*
2502 * This is needed only for protection faults but the arch code
2503 * is not yet telling us if this is a protection fault or not.
2504 * This still avoids useless tlb flushes for .text page faults
2505 * with threads.
2506 */
2507 if (write_access)
2508 flush_tlb_page(vma, address);
2509 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002510unlock:
2511 pte_unmap_unlock(pte, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002512 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513}
2514
2515/*
2516 * By the time we get here, we already hold the mm semaphore
2517 */
Nick Piggin83c54072007-07-19 01:47:05 -07002518int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519 unsigned long address, int write_access)
2520{
2521 pgd_t *pgd;
2522 pud_t *pud;
2523 pmd_t *pmd;
2524 pte_t *pte;
2525
2526 __set_current_state(TASK_RUNNING);
2527
Christoph Lameterf8891e52006-06-30 01:55:45 -07002528 count_vm_event(PGFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01002530 if (unlikely(is_vm_hugetlb_page(vma)))
2531 return hugetlb_fault(mm, vma, address, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 pud = pud_alloc(mm, pgd, address);
2535 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002536 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 pmd = pmd_alloc(mm, pud, address);
2538 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002539 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 pte = pte_alloc_map(mm, pmd, address);
2541 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002542 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543
Hugh Dickinsc74df322005-10-29 18:16:23 -07002544 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545}
2546
2547#ifndef __PAGETABLE_PUD_FOLDED
2548/*
2549 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002550 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002552int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002554 pud_t *new = pud_alloc_one(mm, address);
2555 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002556 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557
Hugh Dickins872fec12005-10-29 18:16:21 -07002558 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002559 if (pgd_present(*pgd)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 pud_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002561 else
2562 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07002563 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002564 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565}
2566#endif /* __PAGETABLE_PUD_FOLDED */
2567
2568#ifndef __PAGETABLE_PMD_FOLDED
2569/*
2570 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002571 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002573int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002575 pmd_t *new = pmd_alloc_one(mm, address);
2576 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002577 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578
Hugh Dickins872fec12005-10-29 18:16:21 -07002579 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07002581 if (pud_present(*pud)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 pmd_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002583 else
2584 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07002586 if (pgd_present(*pud)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 pmd_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002588 else
2589 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07002591 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002592 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593}
2594#endif /* __PAGETABLE_PMD_FOLDED */
2595
2596int make_pages_present(unsigned long addr, unsigned long end)
2597{
2598 int ret, len, write;
2599 struct vm_area_struct * vma;
2600
2601 vma = find_vma(current->mm, addr);
2602 if (!vma)
2603 return -1;
2604 write = (vma->vm_flags & VM_WRITE) != 0;
Eric Sesterhenn5bcb28b2006-03-26 18:30:52 +02002605 BUG_ON(addr >= end);
2606 BUG_ON(end > vma->vm_end);
Rolf Eike Beer68e116a2007-07-15 23:38:03 -07002607 len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 ret = get_user_pages(current, current->mm, addr,
2609 len, write, 0, NULL, NULL);
2610 if (ret < 0)
2611 return ret;
2612 return ret == len ? 0 : -1;
2613}
2614
2615/*
2616 * Map a vmalloc()-space virtual address to the physical page.
2617 */
2618struct page * vmalloc_to_page(void * vmalloc_addr)
2619{
2620 unsigned long addr = (unsigned long) vmalloc_addr;
2621 struct page *page = NULL;
2622 pgd_t *pgd = pgd_offset_k(addr);
2623 pud_t *pud;
2624 pmd_t *pmd;
2625 pte_t *ptep, pte;
2626
2627 if (!pgd_none(*pgd)) {
2628 pud = pud_offset(pgd, addr);
2629 if (!pud_none(*pud)) {
2630 pmd = pmd_offset(pud, addr);
2631 if (!pmd_none(*pmd)) {
2632 ptep = pte_offset_map(pmd, addr);
2633 pte = *ptep;
2634 if (pte_present(pte))
2635 page = pte_page(pte);
2636 pte_unmap(ptep);
2637 }
2638 }
2639 }
2640 return page;
2641}
2642
2643EXPORT_SYMBOL(vmalloc_to_page);
2644
2645/*
2646 * Map a vmalloc()-space virtual address to the physical page frame number.
2647 */
2648unsigned long vmalloc_to_pfn(void * vmalloc_addr)
2649{
2650 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
2651}
2652
2653EXPORT_SYMBOL(vmalloc_to_pfn);
2654
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655#if !defined(__HAVE_ARCH_GATE_AREA)
2656
2657#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07002658static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659
2660static int __init gate_vma_init(void)
2661{
2662 gate_vma.vm_mm = NULL;
2663 gate_vma.vm_start = FIXADDR_USER_START;
2664 gate_vma.vm_end = FIXADDR_USER_END;
Roland McGrathb6558c42007-01-26 00:56:47 -08002665 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
2666 gate_vma.vm_page_prot = __P101;
Roland McGrathf47aef52007-01-26 00:56:49 -08002667 /*
2668 * Make sure the vDSO gets into every core dump.
2669 * Dumping its contents makes post-mortem fully interpretable later
2670 * without matching up the same kernel and hardware config to see
2671 * what PC values meant.
2672 */
2673 gate_vma.vm_flags |= VM_ALWAYSDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 return 0;
2675}
2676__initcall(gate_vma_init);
2677#endif
2678
2679struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2680{
2681#ifdef AT_SYSINFO_EHDR
2682 return &gate_vma;
2683#else
2684 return NULL;
2685#endif
2686}
2687
2688int in_gate_area_no_task(unsigned long addr)
2689{
2690#ifdef AT_SYSINFO_EHDR
2691 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2692 return 1;
2693#endif
2694 return 0;
2695}
2696
2697#endif /* __HAVE_ARCH_GATE_AREA */
David Howells0ec76a12006-09-27 01:50:15 -07002698
2699/*
2700 * Access another process' address space.
2701 * Source/target buffer must be kernel space,
2702 * Do not walk the page table directly, use get_user_pages
2703 */
2704int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2705{
2706 struct mm_struct *mm;
2707 struct vm_area_struct *vma;
2708 struct page *page;
2709 void *old_buf = buf;
2710
2711 mm = get_task_mm(tsk);
2712 if (!mm)
2713 return 0;
2714
2715 down_read(&mm->mmap_sem);
Simon Arlott183ff222007-10-20 01:27:18 +02002716 /* ignore errors, just check how much was successfully transferred */
David Howells0ec76a12006-09-27 01:50:15 -07002717 while (len) {
2718 int bytes, ret, offset;
2719 void *maddr;
2720
2721 ret = get_user_pages(tsk, mm, addr, 1,
2722 write, 1, &page, &vma);
2723 if (ret <= 0)
2724 break;
2725
2726 bytes = len;
2727 offset = addr & (PAGE_SIZE-1);
2728 if (bytes > PAGE_SIZE-offset)
2729 bytes = PAGE_SIZE-offset;
2730
2731 maddr = kmap(page);
2732 if (write) {
2733 copy_to_user_page(vma, page, addr,
2734 maddr + offset, buf, bytes);
2735 set_page_dirty_lock(page);
2736 } else {
2737 copy_from_user_page(vma, page, addr,
2738 buf, maddr + offset, bytes);
2739 }
2740 kunmap(page);
2741 page_cache_release(page);
2742 len -= bytes;
2743 buf += bytes;
2744 addr += bytes;
2745 }
2746 up_read(&mm->mmap_sem);
2747 mmput(mm);
2748
2749 return buf - old_buf;
2750}
Rusty Russell5992b6d2007-07-19 01:49:21 -07002751EXPORT_SYMBOL_GPL(access_process_vm);