blob: f64cbf9baa3633fc1755f675ec2b46c920c4ecef [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;
81unsigned long vmalloc_earlyreserve;
82
83EXPORT_SYMBOL(num_physpages);
84EXPORT_SYMBOL(high_memory);
85EXPORT_SYMBOL(vmalloc_earlyreserve);
86
Andi Kleena62eaf12006-02-16 23:41:58 +010087int randomize_va_space __read_mostly = 1;
88
89static int __init disable_randmaps(char *s)
90{
91 randomize_va_space = 0;
OGAWA Hirofumi9b410462006-03-31 02:30:33 -080092 return 1;
Andi Kleena62eaf12006-02-16 23:41:58 +010093}
94__setup("norandmaps", disable_randmaps);
95
96
Linus Torvalds1da177e2005-04-16 15:20:36 -070097/*
98 * If a p?d_bad entry is found while walking page tables, report
99 * the error, before resetting entry to p?d_none. Usually (but
100 * very seldom) called out from the p?d_none_or_clear_bad macros.
101 */
102
103void pgd_clear_bad(pgd_t *pgd)
104{
105 pgd_ERROR(*pgd);
106 pgd_clear(pgd);
107}
108
109void pud_clear_bad(pud_t *pud)
110{
111 pud_ERROR(*pud);
112 pud_clear(pud);
113}
114
115void pmd_clear_bad(pmd_t *pmd)
116{
117 pmd_ERROR(*pmd);
118 pmd_clear(pmd);
119}
120
121/*
122 * Note: this doesn't free the actual pages themselves. That
123 * has been handled earlier when unmapping all the memory regions.
124 */
Hugh Dickinse0da3822005-04-19 13:29:15 -0700125static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126{
Hugh Dickinse0da3822005-04-19 13:29:15 -0700127 struct page *page = pmd_page(*pmd);
128 pmd_clear(pmd);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700129 pte_lock_deinit(page);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700130 pte_free_tlb(tlb, page);
Christoph Lameterdf849a12006-06-30 01:55:38 -0700131 dec_zone_page_state(page, NR_PAGETABLE);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700132 tlb->mm->nr_ptes--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133}
134
Hugh Dickinse0da3822005-04-19 13:29:15 -0700135static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
136 unsigned long addr, unsigned long end,
137 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138{
139 pmd_t *pmd;
140 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700141 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142
Hugh Dickinse0da3822005-04-19 13:29:15 -0700143 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 do {
146 next = pmd_addr_end(addr, end);
147 if (pmd_none_or_clear_bad(pmd))
148 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700149 free_pte_range(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 } while (pmd++, addr = next, addr != end);
151
Hugh Dickinse0da3822005-04-19 13:29:15 -0700152 start &= PUD_MASK;
153 if (start < floor)
154 return;
155 if (ceiling) {
156 ceiling &= PUD_MASK;
157 if (!ceiling)
158 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700160 if (end - 1 > ceiling - 1)
161 return;
162
163 pmd = pmd_offset(pud, start);
164 pud_clear(pud);
165 pmd_free_tlb(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166}
167
Hugh Dickinse0da3822005-04-19 13:29:15 -0700168static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
169 unsigned long addr, unsigned long end,
170 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171{
172 pud_t *pud;
173 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700174 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175
Hugh Dickinse0da3822005-04-19 13:29:15 -0700176 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 pud = pud_offset(pgd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 do {
179 next = pud_addr_end(addr, end);
180 if (pud_none_or_clear_bad(pud))
181 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700182 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 } while (pud++, addr = next, addr != end);
184
Hugh Dickinse0da3822005-04-19 13:29:15 -0700185 start &= PGDIR_MASK;
186 if (start < floor)
187 return;
188 if (ceiling) {
189 ceiling &= PGDIR_MASK;
190 if (!ceiling)
191 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700193 if (end - 1 > ceiling - 1)
194 return;
195
196 pud = pud_offset(pgd, start);
197 pgd_clear(pgd);
198 pud_free_tlb(tlb, pud);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199}
200
201/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700202 * This function frees user-level page tables of a process.
203 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 * Must be called with pagetable lock held.
205 */
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700206void free_pgd_range(struct mmu_gather **tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700207 unsigned long addr, unsigned long end,
208 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209{
210 pgd_t *pgd;
211 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700212 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213
Hugh Dickinse0da3822005-04-19 13:29:15 -0700214 /*
215 * The next few lines have given us lots of grief...
216 *
217 * Why are we testing PMD* at this top level? Because often
218 * there will be no work to do at all, and we'd prefer not to
219 * go all the way down to the bottom just to discover that.
220 *
221 * Why all these "- 1"s? Because 0 represents both the bottom
222 * of the address space and the top of it (using -1 for the
223 * top wouldn't help much: the masks would do the wrong thing).
224 * The rule is that addr 0 and floor 0 refer to the bottom of
225 * the address space, but end 0 and ceiling 0 refer to the top
226 * Comparisons need to use "end - 1" and "ceiling - 1" (though
227 * that end 0 case should be mythical).
228 *
229 * Wherever addr is brought up or ceiling brought down, we must
230 * be careful to reject "the opposite 0" before it confuses the
231 * subsequent tests. But what about where end is brought down
232 * by PMD_SIZE below? no, end can't go down to 0 there.
233 *
234 * Whereas we round start (addr) and ceiling down, by different
235 * masks at different levels, in order to test whether a table
236 * now has no other vmas using it, so can be freed, we don't
237 * bother to round floor or end up - the tests don't need that.
238 */
239
240 addr &= PMD_MASK;
241 if (addr < floor) {
242 addr += PMD_SIZE;
243 if (!addr)
244 return;
245 }
246 if (ceiling) {
247 ceiling &= PMD_MASK;
248 if (!ceiling)
249 return;
250 }
251 if (end - 1 > ceiling - 1)
252 end -= PMD_SIZE;
253 if (addr > end - 1)
254 return;
255
256 start = addr;
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700257 pgd = pgd_offset((*tlb)->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 do {
259 next = pgd_addr_end(addr, end);
260 if (pgd_none_or_clear_bad(pgd))
261 continue;
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700262 free_pud_range(*tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700264
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700265 if (!(*tlb)->fullmm)
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700266 flush_tlb_pgtables((*tlb)->mm, start, end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700267}
268
269void free_pgtables(struct mmu_gather **tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700270 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700271{
272 while (vma) {
273 struct vm_area_struct *next = vma->vm_next;
274 unsigned long addr = vma->vm_start;
275
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700276 /*
277 * Hide vma from rmap and vmtruncate before freeing pgtables
278 */
279 anon_vma_unlink(vma);
280 unlink_file_vma(vma);
281
David Gibson9da61ae2006-03-22 00:08:57 -0800282 if (is_vm_hugetlb_page(vma)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700283 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700284 floor, next? next->vm_start: ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700285 } else {
286 /*
287 * Optimization: gather nearby vmas into one call down
288 */
289 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
David Gibson48669202006-03-22 00:08:58 -0800290 && !is_vm_hugetlb_page(next)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700291 vma = next;
292 next = vma->vm_next;
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700293 anon_vma_unlink(vma);
294 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700295 }
296 free_pgd_range(tlb, addr, vma->vm_end,
297 floor, next? next->vm_start: ceiling);
298 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700299 vma = next;
300 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301}
302
Hugh Dickins1bb36302005-10-29 18:16:22 -0700303int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304{
Hugh Dickinsc74df322005-10-29 18:16:23 -0700305 struct page *new = pte_alloc_one(mm, address);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700306 if (!new)
307 return -ENOMEM;
308
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700309 pte_lock_init(new);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700310 spin_lock(&mm->page_table_lock);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700311 if (pmd_present(*pmd)) { /* Another has populated it */
312 pte_lock_deinit(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700313 pte_free(new);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700314 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 mm->nr_ptes++;
Christoph Lameterdf849a12006-06-30 01:55:38 -0700316 inc_zone_page_state(new, NR_PAGETABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 pmd_populate(mm, pmd, new);
318 }
Hugh Dickinsc74df322005-10-29 18:16:23 -0700319 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700320 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321}
322
Hugh Dickins1bb36302005-10-29 18:16:22 -0700323int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700325 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
326 if (!new)
327 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328
Hugh Dickins1bb36302005-10-29 18:16:22 -0700329 spin_lock(&init_mm.page_table_lock);
330 if (pmd_present(*pmd)) /* Another has populated it */
331 pte_free_kernel(new);
332 else
333 pmd_populate_kernel(&init_mm, pmd, new);
334 spin_unlock(&init_mm.page_table_lock);
335 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336}
337
Hugh Dickinsae859762005-10-29 18:16:05 -0700338static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
339{
340 if (file_rss)
341 add_mm_counter(mm, file_rss, file_rss);
342 if (anon_rss)
343 add_mm_counter(mm, anon_rss, anon_rss);
344}
345
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800347 * This function is called to print an error when a bad pte
348 * is found. For example, we might have a PFN-mapped pte in
349 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700350 *
351 * The calling function must still handle the error.
352 */
353void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr)
354{
355 printk(KERN_ERR "Bad pte = %08llx, process = %s, "
356 "vm_flags = %lx, vaddr = %lx\n",
357 (long long)pte_val(pte),
358 (vma->vm_mm == current->mm ? current->comm : "???"),
359 vma->vm_flags, vaddr);
360 dump_stack();
361}
362
Linus Torvalds67121172005-12-11 20:38:17 -0800363static inline int is_cow_mapping(unsigned int flags)
364{
365 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
366}
367
Nick Pigginb5810032005-10-29 18:16:12 -0700368/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800369 * This function gets the "struct page" associated with a pte.
370 *
371 * NOTE! Some mappings do not have "struct pages". A raw PFN mapping
372 * will have each page table entry just pointing to a raw page frame
373 * number, and as far as the VM layer is concerned, those do not have
374 * pages associated with them - even if the PFN might point to memory
375 * that otherwise is perfectly fine and has a "struct page".
376 *
377 * The way we recognize those mappings is through the rules set up
378 * by "remap_pfn_range()": the vma will have the VM_PFNMAP bit set,
379 * and the vm_pgoff will point to the first PFN mapped: thus every
380 * page that is a raw mapping will always honor the rule
381 *
382 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
383 *
384 * and if that isn't true, the page has been COW'ed (in which case it
385 * _does_ have a "struct page" associated with it even if it is in a
386 * VM_PFNMAP range).
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800387 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800388struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800389{
Linus Torvalds6aab3412005-11-28 14:34:23 -0800390 unsigned long pfn = pte_pfn(pte);
391
Nick Pigginb7ab7952006-03-22 00:08:42 -0800392 if (unlikely(vma->vm_flags & VM_PFNMAP)) {
Linus Torvalds6aab3412005-11-28 14:34:23 -0800393 unsigned long off = (addr - vma->vm_start) >> PAGE_SHIFT;
394 if (pfn == vma->vm_pgoff + off)
395 return NULL;
Linus Torvalds67121172005-12-11 20:38:17 -0800396 if (!is_cow_mapping(vma->vm_flags))
Linus Torvaldsfb155c12005-12-11 19:46:02 -0800397 return NULL;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800398 }
399
Nick Piggin315ab192006-03-25 16:20:22 +0100400 /*
401 * Add some anal sanity checks for now. Eventually,
402 * we should just do "return pfn_to_page(pfn)", but
403 * in the meantime we check that we get a valid pfn,
404 * and that the resulting page looks ok.
405 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800406 if (unlikely(!pfn_valid(pfn))) {
407 print_bad_pte(vma, pte, addr);
408 return NULL;
409 }
410
411 /*
412 * NOTE! We still have PageReserved() pages in the page
413 * tables.
414 *
415 * The PAGE_ZERO() pages and various VDSO mappings can
416 * cause them to exist.
417 */
418 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800419}
420
421/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 * copy one vm_area from one task to the other. Assumes the page tables
423 * already present in the new task to be cleared in the whole range
424 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 */
426
Hugh Dickins8c103762005-10-29 18:16:13 -0700427static inline void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700429 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700430 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431{
Nick Pigginb5810032005-10-29 18:16:12 -0700432 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 pte_t pte = *src_pte;
434 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
436 /* pte contains position in swap or file, so copy. */
437 if (unlikely(!pte_present(pte))) {
438 if (!pte_file(pte)) {
Christoph Lameter06972122006-06-23 02:03:35 -0700439 swp_entry_t entry = pte_to_swp_entry(pte);
440
441 swap_duplicate(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 /* make sure dst_mm is on swapoff's mmlist. */
443 if (unlikely(list_empty(&dst_mm->mmlist))) {
444 spin_lock(&mmlist_lock);
Hugh Dickinsf412ac02005-10-29 18:16:41 -0700445 if (list_empty(&dst_mm->mmlist))
446 list_add(&dst_mm->mmlist,
447 &src_mm->mmlist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 spin_unlock(&mmlist_lock);
449 }
Christoph Lameter06972122006-06-23 02:03:35 -0700450 if (is_write_migration_entry(entry) &&
451 is_cow_mapping(vm_flags)) {
452 /*
453 * COW mappings require pages in both parent
454 * and child to be set to read.
455 */
456 make_migration_entry_read(&entry);
457 pte = swp_entry_to_pte(entry);
458 set_pte_at(src_mm, addr, src_pte, pte);
459 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700461 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 }
463
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 /*
465 * If it's a COW mapping, write protect it both
466 * in the parent and the child
467 */
Linus Torvalds67121172005-12-11 20:38:17 -0800468 if (is_cow_mapping(vm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 ptep_set_wrprotect(src_mm, addr, src_pte);
Zachary Amsden3dc90792006-09-30 23:29:30 -0700470 pte = pte_wrprotect(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 }
472
473 /*
474 * If it's a shared mapping, mark it clean in
475 * the child
476 */
477 if (vm_flags & VM_SHARED)
478 pte = pte_mkclean(pte);
479 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800480
481 page = vm_normal_page(vma, addr, pte);
482 if (page) {
483 get_page(page);
Nick Pigginc97a9e12007-05-16 22:11:21 -0700484 page_dup_rmap(page, vma, addr);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800485 rss[!!PageAnon(page)]++;
486 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700487
488out_set_pte:
489 set_pte_at(dst_mm, addr, dst_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490}
491
492static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
493 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
494 unsigned long addr, unsigned long end)
495{
496 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700497 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -0700498 int progress = 0;
Hugh Dickins8c103762005-10-29 18:16:13 -0700499 int rss[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500
501again:
Hugh Dickinsae859762005-10-29 18:16:05 -0700502 rss[1] = rss[0] = 0;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700503 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 if (!dst_pte)
505 return -ENOMEM;
506 src_pte = pte_offset_map_nested(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700507 src_ptl = pte_lockptr(src_mm, src_pmd);
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700508 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700509 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 do {
512 /*
513 * We are holding two locks at this point - either of them
514 * could generate latencies in another task on another CPU.
515 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700516 if (progress >= 32) {
517 progress = 0;
518 if (need_resched() ||
Hugh Dickinsc74df322005-10-29 18:16:23 -0700519 need_lockbreak(src_ptl) ||
520 need_lockbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700521 break;
522 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 if (pte_none(*src_pte)) {
524 progress++;
525 continue;
526 }
Hugh Dickins8c103762005-10-29 18:16:13 -0700527 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 progress += 8;
529 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700531 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -0700532 spin_unlock(src_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 pte_unmap_nested(src_pte - 1);
Hugh Dickinsae859762005-10-29 18:16:05 -0700534 add_mm_rss(dst_mm, rss[0], rss[1]);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700535 pte_unmap_unlock(dst_pte - 1, dst_ptl);
536 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 if (addr != end)
538 goto again;
539 return 0;
540}
541
542static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
543 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
544 unsigned long addr, unsigned long end)
545{
546 pmd_t *src_pmd, *dst_pmd;
547 unsigned long next;
548
549 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
550 if (!dst_pmd)
551 return -ENOMEM;
552 src_pmd = pmd_offset(src_pud, addr);
553 do {
554 next = pmd_addr_end(addr, end);
555 if (pmd_none_or_clear_bad(src_pmd))
556 continue;
557 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
558 vma, addr, next))
559 return -ENOMEM;
560 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
561 return 0;
562}
563
564static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
565 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
566 unsigned long addr, unsigned long end)
567{
568 pud_t *src_pud, *dst_pud;
569 unsigned long next;
570
571 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
572 if (!dst_pud)
573 return -ENOMEM;
574 src_pud = pud_offset(src_pgd, addr);
575 do {
576 next = pud_addr_end(addr, end);
577 if (pud_none_or_clear_bad(src_pud))
578 continue;
579 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
580 vma, addr, next))
581 return -ENOMEM;
582 } while (dst_pud++, src_pud++, addr = next, addr != end);
583 return 0;
584}
585
586int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
587 struct vm_area_struct *vma)
588{
589 pgd_t *src_pgd, *dst_pgd;
590 unsigned long next;
591 unsigned long addr = vma->vm_start;
592 unsigned long end = vma->vm_end;
593
Nick Piggind9928952005-08-28 16:49:11 +1000594 /*
595 * Don't copy ptes where a page fault will fill them correctly.
596 * Fork becomes much lighter when there are big shared or private
597 * readonly mappings. The tradeoff is that copy_page_range is more
598 * efficient than faulting.
599 */
Linus Torvalds4d7672b2005-12-16 10:21:23 -0800600 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
Nick Piggind9928952005-08-28 16:49:11 +1000601 if (!vma->anon_vma)
602 return 0;
603 }
604
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 if (is_vm_hugetlb_page(vma))
606 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
607
608 dst_pgd = pgd_offset(dst_mm, addr);
609 src_pgd = pgd_offset(src_mm, addr);
610 do {
611 next = pgd_addr_end(addr, end);
612 if (pgd_none_or_clear_bad(src_pgd))
613 continue;
614 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
615 vma, addr, next))
616 return -ENOMEM;
617 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
618 return 0;
619}
620
Robin Holt51c6f662005-11-13 16:06:42 -0800621static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700622 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800624 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625{
Nick Pigginb5810032005-10-29 18:16:12 -0700626 struct mm_struct *mm = tlb->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 pte_t *pte;
Hugh Dickins508034a2005-10-29 18:16:30 -0700628 spinlock_t *ptl;
Hugh Dickinsae859762005-10-29 18:16:05 -0700629 int file_rss = 0;
630 int anon_rss = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
Hugh Dickins508034a2005-10-29 18:16:30 -0700632 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700633 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 do {
635 pte_t ptent = *pte;
Robin Holt51c6f662005-11-13 16:06:42 -0800636 if (pte_none(ptent)) {
637 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800639 }
Hugh Dickins6f5e6b92006-03-16 23:04:09 -0800640
641 (*zap_work) -= PAGE_SIZE;
642
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800644 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -0800645
Linus Torvalds6aab3412005-11-28 14:34:23 -0800646 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 if (unlikely(details) && page) {
648 /*
649 * unmap_shared_mapping_pages() wants to
650 * invalidate cache without truncating:
651 * unmap shared but keep private pages.
652 */
653 if (details->check_mapping &&
654 details->check_mapping != page->mapping)
655 continue;
656 /*
657 * Each page->index must be checked when
658 * invalidating or truncating nonlinear.
659 */
660 if (details->nonlinear_vma &&
661 (page->index < details->first_index ||
662 page->index > details->last_index))
663 continue;
664 }
Nick Pigginb5810032005-10-29 18:16:12 -0700665 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -0700666 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 tlb_remove_tlb_entry(tlb, pte, addr);
668 if (unlikely(!page))
669 continue;
670 if (unlikely(details) && details->nonlinear_vma
671 && linear_page_index(details->nonlinear_vma,
672 addr) != page->index)
Nick Pigginb5810032005-10-29 18:16:12 -0700673 set_pte_at(mm, addr, pte,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 pgoff_to_pte(page->index));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 if (PageAnon(page))
Hugh Dickins86d912f2005-10-29 18:16:14 -0700676 anon_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700677 else {
678 if (pte_dirty(ptent))
679 set_page_dirty(page);
680 if (pte_young(ptent))
Ken Chendaa88c82007-02-10 01:43:18 -0800681 SetPageReferenced(page);
Hugh Dickins86d912f2005-10-29 18:16:14 -0700682 file_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700683 }
Nick Piggin7de6b802006-12-22 01:09:33 -0800684 page_remove_rmap(page, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 tlb_remove_page(tlb, page);
686 continue;
687 }
688 /*
689 * If details->check_mapping, we leave swap entries;
690 * if details->nonlinear_vma, we leave file entries.
691 */
692 if (unlikely(details))
693 continue;
694 if (!pte_file(ptent))
695 free_swap_and_cache(pte_to_swp_entry(ptent));
Zachary Amsden9888a1c2006-09-30 23:29:31 -0700696 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
Robin Holt51c6f662005-11-13 16:06:42 -0800697 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
Hugh Dickinsae859762005-10-29 18:16:05 -0700698
Hugh Dickins86d912f2005-10-29 18:16:14 -0700699 add_mm_rss(mm, file_rss, anon_rss);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700700 arch_leave_lazy_mmu_mode();
Hugh Dickins508034a2005-10-29 18:16:30 -0700701 pte_unmap_unlock(pte - 1, ptl);
Robin Holt51c6f662005-11-13 16:06:42 -0800702
703 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704}
705
Robin Holt51c6f662005-11-13 16:06:42 -0800706static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700707 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800709 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710{
711 pmd_t *pmd;
712 unsigned long next;
713
714 pmd = pmd_offset(pud, addr);
715 do {
716 next = pmd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800717 if (pmd_none_or_clear_bad(pmd)) {
718 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800720 }
721 next = zap_pte_range(tlb, vma, pmd, addr, next,
722 zap_work, details);
723 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
724
725 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726}
727
Robin Holt51c6f662005-11-13 16:06:42 -0800728static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700729 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800731 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732{
733 pud_t *pud;
734 unsigned long next;
735
736 pud = pud_offset(pgd, addr);
737 do {
738 next = pud_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800739 if (pud_none_or_clear_bad(pud)) {
740 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800742 }
743 next = zap_pmd_range(tlb, vma, pud, addr, next,
744 zap_work, details);
745 } while (pud++, addr = next, (addr != end && *zap_work > 0));
746
747 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748}
749
Robin Holt51c6f662005-11-13 16:06:42 -0800750static unsigned long unmap_page_range(struct mmu_gather *tlb,
751 struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800753 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754{
755 pgd_t *pgd;
756 unsigned long next;
757
758 if (details && !details->check_mapping && !details->nonlinear_vma)
759 details = NULL;
760
761 BUG_ON(addr >= end);
762 tlb_start_vma(tlb, vma);
763 pgd = pgd_offset(vma->vm_mm, addr);
764 do {
765 next = pgd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800766 if (pgd_none_or_clear_bad(pgd)) {
767 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800769 }
770 next = zap_pud_range(tlb, vma, pgd, addr, next,
771 zap_work, details);
772 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 tlb_end_vma(tlb, vma);
Robin Holt51c6f662005-11-13 16:06:42 -0800774
775 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776}
777
778#ifdef CONFIG_PREEMPT
779# define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
780#else
781/* No preempt: go for improved straight-line efficiency */
782# define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
783#endif
784
785/**
786 * unmap_vmas - unmap a range of memory covered by a list of vma's
787 * @tlbp: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 * @vma: the starting vma
789 * @start_addr: virtual address at which to start unmapping
790 * @end_addr: virtual address at which to end unmapping
791 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
792 * @details: details of nonlinear truncation or shared cache invalidation
793 *
Hugh Dickinsee39b372005-04-19 13:29:15 -0700794 * Returns the end address of the unmapping (restart addr if interrupted).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700796 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700798 * We aim to not hold locks for too long (for scheduling latency reasons).
799 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 * return the ending mmu_gather to the caller.
801 *
802 * Only addresses between `start' and `end' will be unmapped.
803 *
804 * The VMA list must be sorted in ascending virtual address order.
805 *
806 * unmap_vmas() assumes that the caller will flush the whole unmapped address
807 * range after unmap_vmas() returns. So the only responsibility here is to
808 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
809 * drops the lock and schedules.
810 */
Hugh Dickins508034a2005-10-29 18:16:30 -0700811unsigned long unmap_vmas(struct mmu_gather **tlbp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 struct vm_area_struct *vma, unsigned long start_addr,
813 unsigned long end_addr, unsigned long *nr_accounted,
814 struct zap_details *details)
815{
Robin Holt51c6f662005-11-13 16:06:42 -0800816 long zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
818 int tlb_start_valid = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -0700819 unsigned long start = start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700821 int fullmm = (*tlbp)->fullmm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822
823 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 unsigned long end;
825
826 start = max(vma->vm_start, start_addr);
827 if (start >= vma->vm_end)
828 continue;
829 end = min(vma->vm_end, end_addr);
830 if (end <= vma->vm_start)
831 continue;
832
833 if (vma->vm_flags & VM_ACCOUNT)
834 *nr_accounted += (end - start) >> PAGE_SHIFT;
835
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 while (start != end) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 if (!tlb_start_valid) {
838 tlb_start = start;
839 tlb_start_valid = 1;
840 }
841
Robin Holt51c6f662005-11-13 16:06:42 -0800842 if (unlikely(is_vm_hugetlb_page(vma))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 unmap_hugepage_range(vma, start, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800844 zap_work -= (end - start) /
845 (HPAGE_SIZE / PAGE_SIZE);
846 start = end;
847 } else
848 start = unmap_page_range(*tlbp, vma,
849 start, end, &zap_work, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850
Robin Holt51c6f662005-11-13 16:06:42 -0800851 if (zap_work > 0) {
852 BUG_ON(start != end);
853 break;
854 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855
856 tlb_finish_mmu(*tlbp, tlb_start, start);
857
858 if (need_resched() ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 (i_mmap_lock && need_lockbreak(i_mmap_lock))) {
860 if (i_mmap_lock) {
Hugh Dickins508034a2005-10-29 18:16:30 -0700861 *tlbp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 goto out;
863 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 }
866
Hugh Dickins508034a2005-10-29 18:16:30 -0700867 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 tlb_start_valid = 0;
Robin Holt51c6f662005-11-13 16:06:42 -0800869 zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 }
871 }
872out:
Hugh Dickinsee39b372005-04-19 13:29:15 -0700873 return start; /* which is now the end (or restart) address */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874}
875
876/**
877 * zap_page_range - remove user pages in a given range
878 * @vma: vm_area_struct holding the applicable pages
879 * @address: starting address of pages to zap
880 * @size: number of bytes to zap
881 * @details: details of nonlinear truncation or shared cache invalidation
882 */
Hugh Dickinsee39b372005-04-19 13:29:15 -0700883unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 unsigned long size, struct zap_details *details)
885{
886 struct mm_struct *mm = vma->vm_mm;
887 struct mmu_gather *tlb;
888 unsigned long end = address + size;
889 unsigned long nr_accounted = 0;
890
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -0700893 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -0700894 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
895 if (tlb)
896 tlb_finish_mmu(tlb, address, end);
Hugh Dickinsee39b372005-04-19 13:29:15 -0700897 return end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898}
899
900/*
901 * Do a quick page-table lookup for a single page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800903struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700904 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905{
906 pgd_t *pgd;
907 pud_t *pud;
908 pmd_t *pmd;
909 pte_t *ptep, pte;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700910 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 struct page *page;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800912 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700914 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
915 if (!IS_ERR(page)) {
916 BUG_ON(flags & FOLL_GET);
917 goto out;
918 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700920 page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 pgd = pgd_offset(mm, address);
922 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700923 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924
925 pud = pud_offset(pgd, address);
926 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700927 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928
929 pmd = pmd_offset(pud, address);
930 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700931 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700933 if (pmd_huge(*pmd)) {
934 BUG_ON(flags & FOLL_GET);
935 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
936 goto out;
937 }
938
939 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 if (!ptep)
941 goto out;
942
943 pte = *ptep;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700944 if (!pte_present(pte))
945 goto unlock;
946 if ((flags & FOLL_WRITE) && !pte_write(pte))
947 goto unlock;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800948 page = vm_normal_page(vma, address, pte);
949 if (unlikely(!page))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700950 goto unlock;
951
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700952 if (flags & FOLL_GET)
953 get_page(page);
954 if (flags & FOLL_TOUCH) {
955 if ((flags & FOLL_WRITE) &&
956 !pte_dirty(pte) && !PageDirty(page))
957 set_page_dirty(page);
958 mark_page_accessed(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700960unlock:
961 pte_unmap_unlock(ptep, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962out:
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700963 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700965no_page_table:
966 /*
967 * When core dumping an enormous anonymous area that nobody
968 * has touched so far, we don't want to allocate page tables.
969 */
970 if (flags & FOLL_ANON) {
971 page = ZERO_PAGE(address);
972 if (flags & FOLL_GET)
973 get_page(page);
974 BUG_ON(flags & FOLL_WRITE);
975 }
976 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977}
978
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
980 unsigned long start, int len, int write, int force,
981 struct page **pages, struct vm_area_struct **vmas)
982{
983 int i;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700984 unsigned int vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985
986 /*
987 * Require read or write permissions.
988 * If 'force' is set, we only require the "MAY" flags.
989 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700990 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
991 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 i = 0;
993
994 do {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700995 struct vm_area_struct *vma;
996 unsigned int foll_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
998 vma = find_extend_vma(mm, start);
999 if (!vma && in_gate_area(tsk, start)) {
1000 unsigned long pg = start & PAGE_MASK;
1001 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
1002 pgd_t *pgd;
1003 pud_t *pud;
1004 pmd_t *pmd;
1005 pte_t *pte;
1006 if (write) /* user gate pages are read-only */
1007 return i ? : -EFAULT;
1008 if (pg > TASK_SIZE)
1009 pgd = pgd_offset_k(pg);
1010 else
1011 pgd = pgd_offset_gate(mm, pg);
1012 BUG_ON(pgd_none(*pgd));
1013 pud = pud_offset(pgd, pg);
1014 BUG_ON(pud_none(*pud));
1015 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001016 if (pmd_none(*pmd))
1017 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001019 if (pte_none(*pte)) {
1020 pte_unmap(pte);
1021 return i ? : -EFAULT;
1022 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 if (pages) {
Nick Pigginfa2a4552005-11-29 18:43:17 +11001024 struct page *page = vm_normal_page(gate_vma, start, *pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001025 pages[i] = page;
1026 if (page)
1027 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 }
1029 pte_unmap(pte);
1030 if (vmas)
1031 vmas[i] = gate_vma;
1032 i++;
1033 start += PAGE_SIZE;
1034 len--;
1035 continue;
1036 }
1037
Linus Torvalds1ff80382005-12-12 16:24:33 -08001038 if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001039 || !(vm_flags & vma->vm_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 return i ? : -EFAULT;
1041
1042 if (is_vm_hugetlb_page(vma)) {
1043 i = follow_hugetlb_page(mm, vma, pages, vmas,
1044 &start, &len, i);
1045 continue;
1046 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001047
1048 foll_flags = FOLL_TOUCH;
1049 if (pages)
1050 foll_flags |= FOLL_GET;
1051 if (!write && !(vma->vm_flags & VM_LOCKED) &&
1052 (!vma->vm_ops || !vma->vm_ops->nopage))
1053 foll_flags |= FOLL_ANON;
1054
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 do {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001056 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001058 if (write)
1059 foll_flags |= FOLL_WRITE;
1060
1061 cond_resched();
Linus Torvalds6aab3412005-11-28 14:34:23 -08001062 while (!(page = follow_page(vma, start, foll_flags))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001063 int ret;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001064 ret = __handle_mm_fault(mm, vma, start,
1065 foll_flags & FOLL_WRITE);
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001066 /*
1067 * The VM_FAULT_WRITE bit tells us that do_wp_page has
1068 * broken COW when necessary, even if maybe_mkwrite
1069 * decided not to set pte_write. We can thus safely do
1070 * subsequent page lookups as if they were reads.
1071 */
1072 if (ret & VM_FAULT_WRITE)
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001073 foll_flags &= ~FOLL_WRITE;
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001074
1075 switch (ret & ~VM_FAULT_WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 case VM_FAULT_MINOR:
1077 tsk->min_flt++;
1078 break;
1079 case VM_FAULT_MAJOR:
1080 tsk->maj_flt++;
1081 break;
1082 case VM_FAULT_SIGBUS:
1083 return i ? i : -EFAULT;
1084 case VM_FAULT_OOM:
1085 return i ? i : -ENOMEM;
1086 default:
1087 BUG();
1088 }
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07001089 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 }
1091 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001092 pages[i] = page;
James Bottomley03beb072006-03-26 01:36:57 -08001093
Russell Kinga6f36be2006-12-30 22:24:19 +00001094 flush_anon_page(vma, page, start);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001095 flush_dcache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 }
1097 if (vmas)
1098 vmas[i] = vma;
1099 i++;
1100 start += PAGE_SIZE;
1101 len--;
Hugh Dickins08ef4722005-06-21 17:15:10 -07001102 } while (len && start < vma->vm_end);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001103 } while (len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 return i;
1105}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106EXPORT_SYMBOL(get_user_pages);
1107
1108static int zeromap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1109 unsigned long addr, unsigned long end, pgprot_t prot)
1110{
1111 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001112 spinlock_t *ptl;
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001113 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114
Hugh Dickinsc74df322005-10-29 18:16:23 -07001115 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 if (!pte)
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001117 return -EAGAIN;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001118 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 do {
Nick Pigginb5810032005-10-29 18:16:12 -07001120 struct page *page = ZERO_PAGE(addr);
1121 pte_t zero_pte = pte_wrprotect(mk_pte(page, prot));
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001122
1123 if (unlikely(!pte_none(*pte))) {
1124 err = -EEXIST;
1125 pte++;
1126 break;
1127 }
Nick Pigginb5810032005-10-29 18:16:12 -07001128 page_cache_get(page);
1129 page_add_file_rmap(page);
1130 inc_mm_counter(mm, file_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 set_pte_at(mm, addr, pte, zero_pte);
1132 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001133 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001134 pte_unmap_unlock(pte - 1, ptl);
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001135 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136}
1137
1138static inline int zeromap_pmd_range(struct mm_struct *mm, pud_t *pud,
1139 unsigned long addr, unsigned long end, pgprot_t prot)
1140{
1141 pmd_t *pmd;
1142 unsigned long next;
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001143 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144
1145 pmd = pmd_alloc(mm, pud, addr);
1146 if (!pmd)
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001147 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 do {
1149 next = pmd_addr_end(addr, end);
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001150 err = zeromap_pte_range(mm, pmd, addr, next, prot);
1151 if (err)
1152 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 } while (pmd++, addr = next, addr != end);
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001154 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155}
1156
1157static inline int zeromap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1158 unsigned long addr, unsigned long end, pgprot_t prot)
1159{
1160 pud_t *pud;
1161 unsigned long next;
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001162 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163
1164 pud = pud_alloc(mm, pgd, addr);
1165 if (!pud)
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001166 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 do {
1168 next = pud_addr_end(addr, end);
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001169 err = zeromap_pmd_range(mm, pud, addr, next, prot);
1170 if (err)
1171 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 } while (pud++, addr = next, addr != end);
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -08001173 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174}
1175
1176int zeromap_page_range(struct vm_area_struct *vma,
1177 unsigned long addr, unsigned long size, pgprot_t prot)
1178{
1179 pgd_t *pgd;
1180 unsigned long next;
1181 unsigned long end = addr + size;
1182 struct mm_struct *mm = vma->vm_mm;
1183 int err;
1184
1185 BUG_ON(addr >= end);
1186 pgd = pgd_offset(mm, addr);
1187 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 do {
1189 next = pgd_addr_end(addr, end);
1190 err = zeromap_pud_range(mm, pgd, addr, next, prot);
1191 if (err)
1192 break;
1193 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 return err;
1195}
1196
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001197pte_t * fastcall get_locked_pte(struct mm_struct *mm, unsigned long addr, spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001198{
1199 pgd_t * pgd = pgd_offset(mm, addr);
1200 pud_t * pud = pud_alloc(mm, pgd, addr);
1201 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001202 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001203 if (pmd)
1204 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1205 }
1206 return NULL;
1207}
1208
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001210 * This is the old fallback for page remapping.
1211 *
1212 * For historical reasons, it only allows reserved pages. Only
1213 * old drivers should use this, and they needed to mark their
1214 * pages reserved for the old functions anyway.
1215 */
1216static int insert_page(struct mm_struct *mm, unsigned long addr, struct page *page, pgprot_t prot)
1217{
1218 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001219 pte_t *pte;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001220 spinlock_t *ptl;
1221
1222 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001223 if (PageAnon(page))
Linus Torvalds238f58d2005-11-29 13:01:56 -08001224 goto out;
1225 retval = -ENOMEM;
1226 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001227 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001228 if (!pte)
1229 goto out;
1230 retval = -EBUSY;
1231 if (!pte_none(*pte))
1232 goto out_unlock;
1233
1234 /* Ok, finally just insert the thing.. */
1235 get_page(page);
1236 inc_mm_counter(mm, file_rss);
1237 page_add_file_rmap(page);
1238 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1239
1240 retval = 0;
1241out_unlock:
1242 pte_unmap_unlock(pte, ptl);
1243out:
1244 return retval;
1245}
1246
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001247/**
1248 * vm_insert_page - insert single page into user vma
1249 * @vma: user vma to map to
1250 * @addr: target user address of this page
1251 * @page: source kernel page
1252 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001253 * This allows drivers to insert individual pages they've allocated
1254 * into a user vma.
1255 *
1256 * The page has to be a nice clean _individual_ kernel allocation.
1257 * If you allocate a compound page, you need to have marked it as
1258 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001259 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001260 *
1261 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1262 * took an arbitrary page protection parameter. This doesn't allow
1263 * that. Your vma protection will have to be set up correctly, which
1264 * means that if you want a shared writable mapping, you'd better
1265 * ask for a shared writable mapping!
1266 *
1267 * The page does not need to be reserved.
1268 */
1269int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, struct page *page)
1270{
1271 if (addr < vma->vm_start || addr >= vma->vm_end)
1272 return -EFAULT;
1273 if (!page_count(page))
1274 return -EINVAL;
Linus Torvalds4d7672b2005-12-16 10:21:23 -08001275 vma->vm_flags |= VM_INSERTPAGE;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001276 return insert_page(vma->vm_mm, addr, page, vma->vm_page_prot);
1277}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001278EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001279
Nick Piggine0dc0d82007-02-12 00:51:36 -08001280/**
1281 * vm_insert_pfn - insert single pfn into user vma
1282 * @vma: user vma to map to
1283 * @addr: target user address of this page
1284 * @pfn: source kernel pfn
1285 *
1286 * Similar to vm_inert_page, this allows drivers to insert individual pages
1287 * they've allocated into a user vma. Same comments apply.
1288 *
1289 * This function should only be called from a vm_ops->fault handler, and
1290 * in that case the handler should return NULL.
1291 */
1292int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1293 unsigned long pfn)
1294{
1295 struct mm_struct *mm = vma->vm_mm;
1296 int retval;
1297 pte_t *pte, entry;
1298 spinlock_t *ptl;
1299
1300 BUG_ON(!(vma->vm_flags & VM_PFNMAP));
1301 BUG_ON(is_cow_mapping(vma->vm_flags));
1302
1303 retval = -ENOMEM;
1304 pte = get_locked_pte(mm, addr, &ptl);
1305 if (!pte)
1306 goto out;
1307 retval = -EBUSY;
1308 if (!pte_none(*pte))
1309 goto out_unlock;
1310
1311 /* Ok, finally just insert the thing.. */
1312 entry = pfn_pte(pfn, vma->vm_page_prot);
1313 set_pte_at(mm, addr, pte, entry);
1314 update_mmu_cache(vma, addr, entry);
1315
1316 retval = 0;
1317out_unlock:
1318 pte_unmap_unlock(pte, ptl);
1319
1320out:
1321 return retval;
1322}
1323EXPORT_SYMBOL(vm_insert_pfn);
1324
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001325/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 * maps a range of physical memory into the requested pages. the old
1327 * mappings are removed. any references to nonexistent pages results
1328 * in null mappings (currently treated as "copy-on-access")
1329 */
1330static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1331 unsigned long addr, unsigned long end,
1332 unsigned long pfn, pgprot_t prot)
1333{
1334 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001335 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336
Hugh Dickinsc74df322005-10-29 18:16:23 -07001337 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 if (!pte)
1339 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001340 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 do {
1342 BUG_ON(!pte_none(*pte));
Nick Pigginb5810032005-10-29 18:16:12 -07001343 set_pte_at(mm, addr, pte, pfn_pte(pfn, prot));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 pfn++;
1345 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001346 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001347 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 return 0;
1349}
1350
1351static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1352 unsigned long addr, unsigned long end,
1353 unsigned long pfn, pgprot_t prot)
1354{
1355 pmd_t *pmd;
1356 unsigned long next;
1357
1358 pfn -= addr >> PAGE_SHIFT;
1359 pmd = pmd_alloc(mm, pud, addr);
1360 if (!pmd)
1361 return -ENOMEM;
1362 do {
1363 next = pmd_addr_end(addr, end);
1364 if (remap_pte_range(mm, pmd, addr, next,
1365 pfn + (addr >> PAGE_SHIFT), prot))
1366 return -ENOMEM;
1367 } while (pmd++, addr = next, addr != end);
1368 return 0;
1369}
1370
1371static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1372 unsigned long addr, unsigned long end,
1373 unsigned long pfn, pgprot_t prot)
1374{
1375 pud_t *pud;
1376 unsigned long next;
1377
1378 pfn -= addr >> PAGE_SHIFT;
1379 pud = pud_alloc(mm, pgd, addr);
1380 if (!pud)
1381 return -ENOMEM;
1382 do {
1383 next = pud_addr_end(addr, end);
1384 if (remap_pmd_range(mm, pud, addr, next,
1385 pfn + (addr >> PAGE_SHIFT), prot))
1386 return -ENOMEM;
1387 } while (pud++, addr = next, addr != end);
1388 return 0;
1389}
1390
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001391/**
1392 * remap_pfn_range - remap kernel memory to userspace
1393 * @vma: user vma to map to
1394 * @addr: target user address to start at
1395 * @pfn: physical address of kernel memory
1396 * @size: size of map area
1397 * @prot: page protection flags for this mapping
1398 *
1399 * Note: this is only safe if the mm semaphore is held when called.
1400 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1402 unsigned long pfn, unsigned long size, pgprot_t prot)
1403{
1404 pgd_t *pgd;
1405 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001406 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 struct mm_struct *mm = vma->vm_mm;
1408 int err;
1409
1410 /*
1411 * Physically remapped pages are special. Tell the
1412 * rest of the world about it:
1413 * VM_IO tells people not to look at these pages
1414 * (accesses can have side effects).
Hugh Dickins0b14c172005-11-21 21:32:15 -08001415 * VM_RESERVED is specified all over the place, because
1416 * in 2.4 it kept swapout's vma scan off this vma; but
1417 * in 2.6 the LRU scan won't even find its pages, so this
1418 * flag means no more than count its pages in reserved_vm,
1419 * and omit it from core dump, even when VM_IO turned off.
Linus Torvalds6aab3412005-11-28 14:34:23 -08001420 * VM_PFNMAP tells the core MM that the base pages are just
1421 * raw PFN mappings, and do not have a "struct page" associated
1422 * with them.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001423 *
1424 * There's a horrible special case to handle copy-on-write
1425 * behaviour that some programs depend on. We mark the "original"
1426 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 */
Linus Torvalds67121172005-12-11 20:38:17 -08001428 if (is_cow_mapping(vma->vm_flags)) {
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001429 if (addr != vma->vm_start || end != vma->vm_end)
Linus Torvalds7fc7e2e2005-12-11 19:57:52 -08001430 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001431 vma->vm_pgoff = pfn;
1432 }
1433
Linus Torvalds6aab3412005-11-28 14:34:23 -08001434 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
1436 BUG_ON(addr >= end);
1437 pfn -= addr >> PAGE_SHIFT;
1438 pgd = pgd_offset(mm, addr);
1439 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 do {
1441 next = pgd_addr_end(addr, end);
1442 err = remap_pud_range(mm, pgd, addr, next,
1443 pfn + (addr >> PAGE_SHIFT), prot);
1444 if (err)
1445 break;
1446 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 return err;
1448}
1449EXPORT_SYMBOL(remap_pfn_range);
1450
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001451static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1452 unsigned long addr, unsigned long end,
1453 pte_fn_t fn, void *data)
1454{
1455 pte_t *pte;
1456 int err;
1457 struct page *pmd_page;
Borislav Petkov94909912007-05-06 14:49:17 -07001458 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001459
1460 pte = (mm == &init_mm) ?
1461 pte_alloc_kernel(pmd, addr) :
1462 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1463 if (!pte)
1464 return -ENOMEM;
1465
1466 BUG_ON(pmd_huge(*pmd));
1467
1468 pmd_page = pmd_page(*pmd);
1469
1470 do {
1471 err = fn(pte, pmd_page, addr, data);
1472 if (err)
1473 break;
1474 } while (pte++, addr += PAGE_SIZE, addr != end);
1475
1476 if (mm != &init_mm)
1477 pte_unmap_unlock(pte-1, ptl);
1478 return err;
1479}
1480
1481static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1482 unsigned long addr, unsigned long end,
1483 pte_fn_t fn, void *data)
1484{
1485 pmd_t *pmd;
1486 unsigned long next;
1487 int err;
1488
1489 pmd = pmd_alloc(mm, pud, addr);
1490 if (!pmd)
1491 return -ENOMEM;
1492 do {
1493 next = pmd_addr_end(addr, end);
1494 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1495 if (err)
1496 break;
1497 } while (pmd++, addr = next, addr != end);
1498 return err;
1499}
1500
1501static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1502 unsigned long addr, unsigned long end,
1503 pte_fn_t fn, void *data)
1504{
1505 pud_t *pud;
1506 unsigned long next;
1507 int err;
1508
1509 pud = pud_alloc(mm, pgd, addr);
1510 if (!pud)
1511 return -ENOMEM;
1512 do {
1513 next = pud_addr_end(addr, end);
1514 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1515 if (err)
1516 break;
1517 } while (pud++, addr = next, addr != end);
1518 return err;
1519}
1520
1521/*
1522 * Scan a region of virtual memory, filling in page tables as necessary
1523 * and calling a provided function on each leaf page table.
1524 */
1525int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1526 unsigned long size, pte_fn_t fn, void *data)
1527{
1528 pgd_t *pgd;
1529 unsigned long next;
1530 unsigned long end = addr + size;
1531 int err;
1532
1533 BUG_ON(addr >= end);
1534 pgd = pgd_offset(mm, addr);
1535 do {
1536 next = pgd_addr_end(addr, end);
1537 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1538 if (err)
1539 break;
1540 } while (pgd++, addr = next, addr != end);
1541 return err;
1542}
1543EXPORT_SYMBOL_GPL(apply_to_page_range);
1544
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001546 * handle_pte_fault chooses page fault handler according to an entry
1547 * which was read non-atomically. Before making any commitment, on
1548 * those architectures or configurations (e.g. i386 with PAE) which
1549 * might give a mix of unmatched parts, do_swap_page and do_file_page
1550 * must check under lock before unmapping the pte and proceeding
1551 * (but do_wp_page is only called after already making such a check;
1552 * and do_anonymous_page and do_no_page can safely check later on).
1553 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001554static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001555 pte_t *page_table, pte_t orig_pte)
1556{
1557 int same = 1;
1558#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1559 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001560 spinlock_t *ptl = pte_lockptr(mm, pmd);
1561 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001562 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001563 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001564 }
1565#endif
1566 pte_unmap(page_table);
1567 return same;
1568}
1569
1570/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1572 * servicing faults for write access. In the normal case, do always want
1573 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1574 * that do not have writing enabled, when used by access_process_vm.
1575 */
1576static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1577{
1578 if (likely(vma->vm_flags & VM_WRITE))
1579 pte = pte_mkwrite(pte);
1580 return pte;
1581}
1582
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001583static 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 -08001584{
1585 /*
1586 * If the source page was a PFN mapping, we don't have
1587 * a "struct page" for it. We do a best-effort copy by
1588 * just copying from the original user address. If that
1589 * fails, we just zero-fill it. Live with it.
1590 */
1591 if (unlikely(!src)) {
1592 void *kaddr = kmap_atomic(dst, KM_USER0);
Linus Torvalds5d2a2dbbc2005-11-29 14:07:55 -08001593 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1594
1595 /*
1596 * This really shouldn't fail, because the page is there
1597 * in the page tables. But it might just be unreadable,
1598 * in which case we just give up and fill the result with
1599 * zeroes.
1600 */
1601 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Linus Torvalds6aab3412005-11-28 14:34:23 -08001602 memset(kaddr, 0, PAGE_SIZE);
1603 kunmap_atomic(kaddr, KM_USER0);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07001604 flush_dcache_page(dst);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001605 return;
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001606
Linus Torvalds6aab3412005-11-28 14:34:23 -08001607 }
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001608 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001609}
1610
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 * This routine handles present pages, when users try to write
1613 * to a shared page. It is done by copying the page to a new address
1614 * and decrementing the shared-page counter for the old page.
1615 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 * Note that this routine assumes that the protection checks have been
1617 * done by the caller (the low-level page fault routine in most cases).
1618 * Thus we can safely just mark it writable once we've done any necessary
1619 * COW.
1620 *
1621 * We also mark the page dirty at this point even though the page will
1622 * change only once the write actually happens. This avoids a few races,
1623 * and potentially makes it more efficient.
1624 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001625 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1626 * but allow concurrent faults), with pte both mapped and locked.
1627 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001629static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1630 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001631 spinlock_t *ptl, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632{
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001633 struct page *old_page, *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 pte_t entry;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001635 int reuse = 0, ret = VM_FAULT_MINOR;
1636 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637
Linus Torvalds6aab3412005-11-28 14:34:23 -08001638 old_page = vm_normal_page(vma, address, orig_pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001639 if (!old_page)
1640 goto gotten;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001642 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001643 * Take out anonymous pages first, anonymous shared vmas are
1644 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001645 */
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001646 if (PageAnon(old_page)) {
1647 if (!TestSetPageLocked(old_page)) {
1648 reuse = can_share_swap_page(old_page);
1649 unlock_page(old_page);
1650 }
1651 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001652 (VM_WRITE|VM_SHARED))) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001653 /*
1654 * Only catch write-faults on shared writable pages,
1655 * read-only shared pages can get COWed by
1656 * get_user_pages(.write=1, .force=1).
1657 */
David Howells9637a5e2006-06-23 02:03:43 -07001658 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
1659 /*
1660 * Notify the address space that the page is about to
1661 * become writable so that it can prohibit this or wait
1662 * for the page to get into an appropriate state.
1663 *
1664 * We do this without the lock held, so that it can
1665 * sleep if it needs to.
1666 */
1667 page_cache_get(old_page);
1668 pte_unmap_unlock(page_table, ptl);
1669
1670 if (vma->vm_ops->page_mkwrite(vma, old_page) < 0)
1671 goto unwritable_page;
1672
David Howells9637a5e2006-06-23 02:03:43 -07001673 /*
1674 * Since we dropped the lock we need to revalidate
1675 * the PTE as someone else may have changed it. If
1676 * they did, we just return, as we can count on the
1677 * MMU to tell us if they didn't also make it writable.
1678 */
1679 page_table = pte_offset_map_lock(mm, pmd, address,
1680 &ptl);
Hugh Dickinsc3704ce2007-02-10 01:43:00 -08001681 page_cache_release(old_page);
David Howells9637a5e2006-06-23 02:03:43 -07001682 if (!pte_same(*page_table, orig_pte))
1683 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001685 dirty_page = old_page;
1686 get_page(dirty_page);
David Howells9637a5e2006-06-23 02:03:43 -07001687 reuse = 1;
David Howells9637a5e2006-06-23 02:03:43 -07001688 }
1689
1690 if (reuse) {
1691 flush_cache_page(vma, address, pte_pfn(orig_pte));
1692 entry = pte_mkyoung(orig_pte);
1693 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07001694 if (ptep_set_access_flags(vma, address, page_table, entry,1)) {
1695 update_mmu_cache(vma, address, entry);
1696 lazy_mmu_prot_update(entry);
1697 }
David Howells9637a5e2006-06-23 02:03:43 -07001698 ret |= VM_FAULT_WRITE;
1699 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701
1702 /*
1703 * Ok, we need to copy. Oh, well..
1704 */
Nick Pigginb5810032005-10-29 18:16:12 -07001705 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08001706gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001707 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708
1709 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07001710 goto oom;
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001711 if (old_page == ZERO_PAGE(address)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 new_page = alloc_zeroed_user_highpage(vma, address);
1713 if (!new_page)
Hugh Dickins65500d22005-10-29 18:15:59 -07001714 goto oom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 } else {
1716 new_page = alloc_page_vma(GFP_HIGHUSER, vma, address);
1717 if (!new_page)
Hugh Dickins65500d22005-10-29 18:15:59 -07001718 goto oom;
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001719 cow_user_page(new_page, old_page, address, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 }
Hugh Dickins65500d22005-10-29 18:15:59 -07001721
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 /*
1723 * Re-check the pte - we dropped the lock
1724 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001725 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07001726 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08001727 if (old_page) {
Nick Piggin7de6b802006-12-22 01:09:33 -08001728 page_remove_rmap(old_page, vma);
Hugh Dickins920fc352005-11-21 21:32:17 -08001729 if (!PageAnon(old_page)) {
1730 dec_mm_counter(mm, file_rss);
1731 inc_mm_counter(mm, anon_rss);
1732 }
1733 } else
Hugh Dickins42946212005-10-29 18:16:05 -07001734 inc_mm_counter(mm, anon_rss);
Ben Collinseca35132005-11-29 11:45:26 -08001735 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07001736 entry = mk_pte(new_page, vma->vm_page_prot);
1737 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Anil Keshavamurthyc38c8db2006-07-14 00:23:57 -07001738 lazy_mmu_prot_update(entry);
Siddha, Suresh B4ce072f2006-09-29 01:58:42 -07001739 /*
1740 * Clear the pte entry and flush it first, before updating the
1741 * pte with the new entry. This will avoid a race condition
1742 * seen in the presence of one thread doing SMC and another
1743 * thread doing COW.
1744 */
1745 ptep_clear_flush(vma, address, page_table);
1746 set_pte_at(mm, address, page_table, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07001747 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 lru_cache_add_active(new_page);
Nick Piggin9617d952006-01-06 00:11:12 -08001749 page_add_new_anon_rmap(new_page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750
1751 /* Free the old page.. */
1752 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001753 ret |= VM_FAULT_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 }
Hugh Dickins920fc352005-11-21 21:32:17 -08001755 if (new_page)
1756 page_cache_release(new_page);
1757 if (old_page)
1758 page_cache_release(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001759unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001760 pte_unmap_unlock(page_table, ptl);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001761 if (dirty_page) {
Peter Zijlstraedc79b22006-09-25 23:30:58 -07001762 set_page_dirty_balance(dirty_page);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001763 put_page(dirty_page);
1764 }
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001765 return ret;
Hugh Dickins65500d22005-10-29 18:15:59 -07001766oom:
Hugh Dickins920fc352005-11-21 21:32:17 -08001767 if (old_page)
1768 page_cache_release(old_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 return VM_FAULT_OOM;
David Howells9637a5e2006-06-23 02:03:43 -07001770
1771unwritable_page:
1772 page_cache_release(old_page);
1773 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774}
1775
1776/*
1777 * Helper functions for unmap_mapping_range().
1778 *
1779 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1780 *
1781 * We have to restart searching the prio_tree whenever we drop the lock,
1782 * since the iterator is only valid while the lock is held, and anyway
1783 * a later vma might be split and reinserted earlier while lock dropped.
1784 *
1785 * The list of nonlinear vmas could be handled more efficiently, using
1786 * a placeholder, but handle it in the same way until a need is shown.
1787 * It is important to search the prio_tree before nonlinear list: a vma
1788 * may become nonlinear and be shifted from prio_tree to nonlinear list
1789 * while the lock is dropped; but never shifted from list to prio_tree.
1790 *
1791 * In order to make forward progress despite restarting the search,
1792 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1793 * quickly skip it next time around. Since the prio_tree search only
1794 * shows us those vmas affected by unmapping the range in question, we
1795 * can't efficiently keep all vmas in step with mapping->truncate_count:
1796 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1797 * mapping->truncate_count and vma->vm_truncate_count are protected by
1798 * i_mmap_lock.
1799 *
1800 * In order to make forward progress despite repeatedly restarting some
Hugh Dickinsee39b372005-04-19 13:29:15 -07001801 * large vma, note the restart_addr from unmap_vmas when it breaks out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 * and restart from that address when we reach that vma again. It might
1803 * have been split or merged, shrunk or extended, but never shifted: so
1804 * restart_addr remains valid so long as it remains in the vma's range.
1805 * unmap_mapping_range forces truncate_count to leap over page-aligned
1806 * values so we can save vma's restart_addr in its truncate_count field.
1807 */
1808#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1809
1810static void reset_vma_truncate_counts(struct address_space *mapping)
1811{
1812 struct vm_area_struct *vma;
1813 struct prio_tree_iter iter;
1814
1815 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1816 vma->vm_truncate_count = 0;
1817 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1818 vma->vm_truncate_count = 0;
1819}
1820
1821static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1822 unsigned long start_addr, unsigned long end_addr,
1823 struct zap_details *details)
1824{
1825 unsigned long restart_addr;
1826 int need_break;
1827
1828again:
1829 restart_addr = vma->vm_truncate_count;
1830 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1831 start_addr = restart_addr;
1832 if (start_addr >= end_addr) {
1833 /* Top of vma has been split off since last time */
1834 vma->vm_truncate_count = details->truncate_count;
1835 return 0;
1836 }
1837 }
1838
Hugh Dickinsee39b372005-04-19 13:29:15 -07001839 restart_addr = zap_page_range(vma, start_addr,
1840 end_addr - start_addr, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 need_break = need_resched() ||
1842 need_lockbreak(details->i_mmap_lock);
1843
Hugh Dickinsee39b372005-04-19 13:29:15 -07001844 if (restart_addr >= end_addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 /* We have now completed this vma: mark it so */
1846 vma->vm_truncate_count = details->truncate_count;
1847 if (!need_break)
1848 return 0;
1849 } else {
1850 /* Note restart_addr in vma's truncate_count field */
Hugh Dickinsee39b372005-04-19 13:29:15 -07001851 vma->vm_truncate_count = restart_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 if (!need_break)
1853 goto again;
1854 }
1855
1856 spin_unlock(details->i_mmap_lock);
1857 cond_resched();
1858 spin_lock(details->i_mmap_lock);
1859 return -EINTR;
1860}
1861
1862static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1863 struct zap_details *details)
1864{
1865 struct vm_area_struct *vma;
1866 struct prio_tree_iter iter;
1867 pgoff_t vba, vea, zba, zea;
1868
1869restart:
1870 vma_prio_tree_foreach(vma, &iter, root,
1871 details->first_index, details->last_index) {
1872 /* Skip quickly over those we have already dealt with */
1873 if (vma->vm_truncate_count == details->truncate_count)
1874 continue;
1875
1876 vba = vma->vm_pgoff;
1877 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1878 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1879 zba = details->first_index;
1880 if (zba < vba)
1881 zba = vba;
1882 zea = details->last_index;
1883 if (zea > vea)
1884 zea = vea;
1885
1886 if (unmap_mapping_range_vma(vma,
1887 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1888 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1889 details) < 0)
1890 goto restart;
1891 }
1892}
1893
1894static inline void unmap_mapping_range_list(struct list_head *head,
1895 struct zap_details *details)
1896{
1897 struct vm_area_struct *vma;
1898
1899 /*
1900 * In nonlinear VMAs there is no correspondence between virtual address
1901 * offset and file offset. So we must perform an exhaustive search
1902 * across *all* the pages in each nonlinear VMA, not just the pages
1903 * whose virtual address lies outside the file truncation point.
1904 */
1905restart:
1906 list_for_each_entry(vma, head, shared.vm_set.list) {
1907 /* Skip quickly over those we have already dealt with */
1908 if (vma->vm_truncate_count == details->truncate_count)
1909 continue;
1910 details->nonlinear_vma = vma;
1911 if (unmap_mapping_range_vma(vma, vma->vm_start,
1912 vma->vm_end, details) < 0)
1913 goto restart;
1914 }
1915}
1916
1917/**
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001918 * 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 -07001919 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 * @holebegin: byte in first page to unmap, relative to the start of
1921 * the underlying file. This will be rounded down to a PAGE_SIZE
1922 * boundary. Note that this is different from vmtruncate(), which
1923 * must keep the partial page. In contrast, we must get rid of
1924 * partial pages.
1925 * @holelen: size of prospective hole in bytes. This will be rounded
1926 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1927 * end of the file.
1928 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1929 * but 0 when invalidating pagecache, don't throw away private data.
1930 */
1931void unmap_mapping_range(struct address_space *mapping,
1932 loff_t const holebegin, loff_t const holelen, int even_cows)
1933{
1934 struct zap_details details;
1935 pgoff_t hba = holebegin >> PAGE_SHIFT;
1936 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1937
1938 /* Check for overflow. */
1939 if (sizeof(holelen) > sizeof(hlen)) {
1940 long long holeend =
1941 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1942 if (holeend & ~(long long)ULONG_MAX)
1943 hlen = ULONG_MAX - hba + 1;
1944 }
1945
1946 details.check_mapping = even_cows? NULL: mapping;
1947 details.nonlinear_vma = NULL;
1948 details.first_index = hba;
1949 details.last_index = hba + hlen - 1;
1950 if (details.last_index < details.first_index)
1951 details.last_index = ULONG_MAX;
1952 details.i_mmap_lock = &mapping->i_mmap_lock;
1953
1954 spin_lock(&mapping->i_mmap_lock);
1955
1956 /* serialize i_size write against truncate_count write */
1957 smp_wmb();
1958 /* Protect against page faults, and endless unmapping loops */
1959 mapping->truncate_count++;
1960 /*
1961 * For archs where spin_lock has inclusive semantics like ia64
1962 * this smp_mb() will prevent to read pagetable contents
1963 * before the truncate_count increment is visible to
1964 * other cpus.
1965 */
1966 smp_mb();
1967 if (unlikely(is_restart_addr(mapping->truncate_count))) {
1968 if (mapping->truncate_count == 0)
1969 reset_vma_truncate_counts(mapping);
1970 mapping->truncate_count++;
1971 }
1972 details.truncate_count = mapping->truncate_count;
1973
1974 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
1975 unmap_mapping_range_tree(&mapping->i_mmap, &details);
1976 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
1977 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
1978 spin_unlock(&mapping->i_mmap_lock);
1979}
1980EXPORT_SYMBOL(unmap_mapping_range);
1981
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001982/**
1983 * vmtruncate - unmap mappings "freed" by truncate() syscall
1984 * @inode: inode of the file used
1985 * @offset: file offset to start truncating
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 *
1987 * NOTE! We have to be ready to update the memory sharing
1988 * between the file and the memory map for a potential last
1989 * incomplete page. Ugly, but necessary.
1990 */
1991int vmtruncate(struct inode * inode, loff_t offset)
1992{
1993 struct address_space *mapping = inode->i_mapping;
1994 unsigned long limit;
1995
1996 if (inode->i_size < offset)
1997 goto do_expand;
1998 /*
1999 * truncation of in-use swapfiles is disallowed - it would cause
2000 * subsequent swapout to scribble on the now-freed blocks.
2001 */
2002 if (IS_SWAPFILE(inode))
2003 goto out_busy;
2004 i_size_write(inode, offset);
2005 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2006 truncate_inode_pages(mapping, offset);
2007 goto out_truncate;
2008
2009do_expand:
2010 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
2011 if (limit != RLIM_INFINITY && offset > limit)
2012 goto out_sig;
2013 if (offset > inode->i_sb->s_maxbytes)
2014 goto out_big;
2015 i_size_write(inode, offset);
2016
2017out_truncate:
2018 if (inode->i_op && inode->i_op->truncate)
2019 inode->i_op->truncate(inode);
2020 return 0;
2021out_sig:
2022 send_sig(SIGXFSZ, current, 0);
2023out_big:
2024 return -EFBIG;
2025out_busy:
2026 return -ETXTBSY;
2027}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028EXPORT_SYMBOL(vmtruncate);
2029
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002030int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
2031{
2032 struct address_space *mapping = inode->i_mapping;
2033
2034 /*
2035 * If the underlying filesystem is not going to provide
2036 * a way to truncate a range of blocks (punch a hole) -
2037 * we should return failure right now.
2038 */
2039 if (!inode->i_op || !inode->i_op->truncate_range)
2040 return -ENOSYS;
2041
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002042 mutex_lock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002043 down_write(&inode->i_alloc_sem);
2044 unmap_mapping_range(mapping, offset, (end - offset), 1);
2045 truncate_inode_pages_range(mapping, offset, end);
2046 inode->i_op->truncate_range(inode, offset, end);
2047 up_write(&inode->i_alloc_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002048 mutex_unlock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002049
2050 return 0;
2051}
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002052
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07002053/**
2054 * swapin_readahead - swap in pages in hope we need them soon
2055 * @entry: swap entry of this memory
2056 * @addr: address to start
2057 * @vma: user vma this addresses belong to
2058 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 * Primitive swap readahead code. We simply read an aligned block of
2060 * (1 << page_cluster) entries in the swap area. This method is chosen
2061 * because it doesn't cost us any seek time. We also make sure to queue
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07002062 * the 'original' request together with the readahead ones...
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 *
2064 * This has been extended to use the NUMA policies from the mm triggering
2065 * the readahead.
2066 *
2067 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
2068 */
2069void swapin_readahead(swp_entry_t entry, unsigned long addr,struct vm_area_struct *vma)
2070{
2071#ifdef CONFIG_NUMA
2072 struct vm_area_struct *next_vma = vma ? vma->vm_next : NULL;
2073#endif
2074 int i, num;
2075 struct page *new_page;
2076 unsigned long offset;
2077
2078 /*
2079 * Get the number of handles we should do readahead io to.
2080 */
2081 num = valid_swaphandles(entry, &offset);
2082 for (i = 0; i < num; offset++, i++) {
2083 /* Ok, do the async read-ahead now */
2084 new_page = read_swap_cache_async(swp_entry(swp_type(entry),
2085 offset), vma, addr);
2086 if (!new_page)
2087 break;
2088 page_cache_release(new_page);
2089#ifdef CONFIG_NUMA
2090 /*
2091 * Find the next applicable VMA for the NUMA policy.
2092 */
2093 addr += PAGE_SIZE;
2094 if (addr == 0)
2095 vma = NULL;
2096 if (vma) {
2097 if (addr >= vma->vm_end) {
2098 vma = next_vma;
2099 next_vma = vma ? vma->vm_next : NULL;
2100 }
2101 if (vma && addr < vma->vm_start)
2102 vma = NULL;
2103 } else {
2104 if (next_vma && addr >= next_vma->vm_start) {
2105 vma = next_vma;
2106 next_vma = vma->vm_next;
2107 }
2108 }
2109#endif
2110 }
2111 lru_add_drain(); /* Push any new pages onto the LRU now */
2112}
2113
2114/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002115 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2116 * but allow concurrent faults), and pte mapped but not yet locked.
2117 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002119static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2120 unsigned long address, pte_t *page_table, pmd_t *pmd,
2121 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002123 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 struct page *page;
Hugh Dickins65500d22005-10-29 18:15:59 -07002125 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 pte_t pte;
2127 int ret = VM_FAULT_MINOR;
2128
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002129 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002130 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002131
2132 entry = pte_to_swp_entry(orig_pte);
Christoph Lameter06972122006-06-23 02:03:35 -07002133 if (is_migration_entry(entry)) {
2134 migration_entry_wait(mm, pmd, address);
2135 goto out;
2136 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002137 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 page = lookup_swap_cache(entry);
2139 if (!page) {
Ashwin Chaugule098fe652006-12-06 20:31:54 -08002140 grab_swap_token(); /* Contend for token _before_ read-in */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 swapin_readahead(entry, address, vma);
2142 page = read_swap_cache_async(entry, vma, address);
2143 if (!page) {
2144 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002145 * Back out if somebody else faulted in this pte
2146 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002148 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 if (likely(pte_same(*page_table, orig_pte)))
2150 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002151 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002152 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 }
2154
2155 /* Had to read the page from swap area: Major fault */
2156 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002157 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 }
2159
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002160 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 mark_page_accessed(page);
2162 lock_page(page);
2163
2164 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002165 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002167 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002168 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002169 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002170
2171 if (unlikely(!PageUptodate(page))) {
2172 ret = VM_FAULT_SIGBUS;
2173 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 }
2175
2176 /* The page isn't present yet, go ahead with the fault. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177
Hugh Dickins42946212005-10-29 18:16:05 -07002178 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 pte = mk_pte(page, vma->vm_page_prot);
2180 if (write_access && can_share_swap_page(page)) {
2181 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2182 write_access = 0;
2183 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184
2185 flush_icache_page(vma, page);
2186 set_pte_at(mm, address, page_table, pte);
2187 page_add_anon_rmap(page, vma, address);
2188
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002189 swap_free(entry);
2190 if (vm_swap_full())
2191 remove_exclusive_swap_page(page);
2192 unlock_page(page);
2193
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 if (write_access) {
2195 if (do_wp_page(mm, vma, address,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002196 page_table, pmd, ptl, pte) == VM_FAULT_OOM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 ret = VM_FAULT_OOM;
2198 goto out;
2199 }
2200
2201 /* No need to invalidate - it was non-present before */
2202 update_mmu_cache(vma, address, pte);
2203 lazy_mmu_prot_update(pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07002204unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002205 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206out:
2207 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002208out_nomap:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002209 pte_unmap_unlock(page_table, ptl);
Kirill Korotaevb8107482005-05-16 21:53:50 -07002210 unlock_page(page);
2211 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002212 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213}
2214
2215/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002216 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2217 * but allow concurrent faults), and pte mapped but not yet locked.
2218 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002220static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2221 unsigned long address, pte_t *page_table, pmd_t *pmd,
2222 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002224 struct page *page;
2225 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227
Linus Torvalds6aab3412005-11-28 14:34:23 -08002228 if (write_access) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 /* Allocate our own private page. */
2230 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231
2232 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07002233 goto oom;
2234 page = alloc_zeroed_user_highpage(vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 if (!page)
Hugh Dickins65500d22005-10-29 18:15:59 -07002236 goto oom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237
Hugh Dickins65500d22005-10-29 18:15:59 -07002238 entry = mk_pte(page, vma->vm_page_prot);
2239 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002240
2241 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2242 if (!pte_none(*page_table))
2243 goto release;
2244 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 lru_cache_add_active(page);
Nick Piggin9617d952006-01-06 00:11:12 -08002246 page_add_new_anon_rmap(page, vma, address);
Nick Pigginb5810032005-10-29 18:16:12 -07002247 } else {
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002248 /* Map the ZERO_PAGE - vm_page_prot is readonly */
2249 page = ZERO_PAGE(address);
2250 page_cache_get(page);
2251 entry = mk_pte(page, vma->vm_page_prot);
2252
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002253 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002254 spin_lock(ptl);
2255 if (!pte_none(*page_table))
2256 goto release;
Nick Pigginb5810032005-10-29 18:16:12 -07002257 inc_mm_counter(mm, file_rss);
2258 page_add_file_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 }
2260
Hugh Dickins65500d22005-10-29 18:15:59 -07002261 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262
2263 /* No need to invalidate - it was non-present before */
Hugh Dickins65500d22005-10-29 18:15:59 -07002264 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 lazy_mmu_prot_update(entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002266unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002267 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268 return VM_FAULT_MINOR;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002269release:
2270 page_cache_release(page);
2271 goto unlock;
Hugh Dickins65500d22005-10-29 18:15:59 -07002272oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 return VM_FAULT_OOM;
2274}
2275
2276/*
2277 * do_no_page() tries to create a new page mapping. It aggressively
2278 * tries to share with existing pages, but makes a separate copy if
2279 * the "write_access" parameter is true in order to avoid the next
2280 * page fault.
2281 *
2282 * As this is called only for pages that do not currently exist, we
2283 * do not need to flush old virtual caches or the TLB.
2284 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002285 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2286 * but allow concurrent faults), and pte mapped but not yet locked.
2287 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002289static int do_no_page(struct mm_struct *mm, struct vm_area_struct *vma,
2290 unsigned long address, pte_t *page_table, pmd_t *pmd,
2291 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002293 spinlock_t *ptl;
Hugh Dickins65500d22005-10-29 18:15:59 -07002294 struct page *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 struct address_space *mapping = NULL;
2296 pte_t entry;
2297 unsigned int sequence = 0;
2298 int ret = VM_FAULT_MINOR;
2299 int anon = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002300 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 pte_unmap(page_table);
Hugh Dickins325f04d2005-11-29 16:55:48 +00002303 BUG_ON(vma->vm_flags & VM_PFNMAP);
2304
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305 if (vma->vm_file) {
2306 mapping = vma->vm_file->f_mapping;
2307 sequence = mapping->truncate_count;
2308 smp_rmb(); /* serializes i_size against truncate_count */
2309 }
2310retry:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 new_page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret);
2312 /*
2313 * No smp_rmb is needed here as long as there's a full
2314 * spin_lock/unlock sequence inside the ->nopage callback
2315 * (for the pagecache lookup) that acts as an implicit
2316 * smp_mb() and prevents the i_size read to happen
2317 * after the next truncate_count read.
2318 */
2319
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07002320 /* no page was available -- either SIGBUS, OOM or REFAULT */
2321 if (unlikely(new_page == NOPAGE_SIGBUS))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 return VM_FAULT_SIGBUS;
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07002323 else if (unlikely(new_page == NOPAGE_OOM))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 return VM_FAULT_OOM;
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07002325 else if (unlikely(new_page == NOPAGE_REFAULT))
2326 return VM_FAULT_MINOR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327
2328 /*
2329 * Should we do an early C-O-W break?
2330 */
David Howells9637a5e2006-06-23 02:03:43 -07002331 if (write_access) {
2332 if (!(vma->vm_flags & VM_SHARED)) {
2333 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334
David Howells9637a5e2006-06-23 02:03:43 -07002335 if (unlikely(anon_vma_prepare(vma)))
2336 goto oom;
2337 page = alloc_page_vma(GFP_HIGHUSER, vma, address);
2338 if (!page)
2339 goto oom;
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00002340 copy_user_highpage(page, new_page, address, vma);
David Howells9637a5e2006-06-23 02:03:43 -07002341 page_cache_release(new_page);
2342 new_page = page;
2343 anon = 1;
2344
2345 } else {
2346 /* if the page will be shareable, see if the backing
2347 * address space wants to know that the page is about
2348 * to become writable */
2349 if (vma->vm_ops->page_mkwrite &&
2350 vma->vm_ops->page_mkwrite(vma, new_page) < 0
2351 ) {
2352 page_cache_release(new_page);
2353 return VM_FAULT_SIGBUS;
2354 }
2355 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356 }
2357
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002358 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 /*
2360 * For a file-backed vma, someone could have truncated or otherwise
2361 * invalidated this page. If unmap_mapping_range got called,
2362 * retry getting the page.
2363 */
2364 if (mapping && unlikely(sequence != mapping->truncate_count)) {
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002365 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002367 cond_resched();
2368 sequence = mapping->truncate_count;
2369 smp_rmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 goto retry;
2371 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372
2373 /*
2374 * This silly early PAGE_DIRTY setting removes a race
2375 * due to the bad i386 page protection. But it's valid
2376 * for other architectures too.
2377 *
2378 * Note that if write_access is true, we either now have
2379 * an exclusive copy of the page, or this is a shared mapping,
2380 * so we can make it writable and dirty to avoid having to
2381 * handle that later.
2382 */
2383 /* Only go through if we didn't race with anybody else... */
2384 if (pte_none(*page_table)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 flush_icache_page(vma, new_page);
2386 entry = mk_pte(new_page, vma->vm_page_prot);
2387 if (write_access)
2388 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2389 set_pte_at(mm, address, page_table, entry);
2390 if (anon) {
Hugh Dickins42946212005-10-29 18:16:05 -07002391 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 lru_cache_add_active(new_page);
Nick Piggin9617d952006-01-06 00:11:12 -08002393 page_add_new_anon_rmap(new_page, vma, address);
Hugh Dickinsf57e88a2005-11-21 21:32:19 -08002394 } else {
Hugh Dickins42946212005-10-29 18:16:05 -07002395 inc_mm_counter(mm, file_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 page_add_file_rmap(new_page);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002397 if (write_access) {
2398 dirty_page = new_page;
2399 get_page(dirty_page);
2400 }
Hugh Dickins42946212005-10-29 18:16:05 -07002401 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402 } else {
2403 /* One of our sibling threads was faster, back out. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002405 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 }
2407
2408 /* no need to invalidate: a not-present page shouldn't be cached */
2409 update_mmu_cache(vma, address, entry);
2410 lazy_mmu_prot_update(entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002411unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002412 pte_unmap_unlock(page_table, ptl);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002413 if (dirty_page) {
Peter Zijlstraedc79b22006-09-25 23:30:58 -07002414 set_page_dirty_balance(dirty_page);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002415 put_page(dirty_page);
2416 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417 return ret;
2418oom:
2419 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002420 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421}
2422
2423/*
Jes Sorensenf4b81802006-09-27 01:50:10 -07002424 * do_no_pfn() tries to create a new page mapping for a page without
2425 * a struct_page backing it
2426 *
2427 * As this is called only for pages that do not currently exist, we
2428 * do not need to flush old virtual caches or the TLB.
2429 *
2430 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2431 * but allow concurrent faults), and pte mapped but not yet locked.
2432 * We return with mmap_sem still held, but pte unmapped and unlocked.
2433 *
2434 * It is expected that the ->nopfn handler always returns the same pfn
2435 * for a given virtual mapping.
2436 *
2437 * Mark this `noinline' to prevent it from bloating the main pagefault code.
2438 */
2439static noinline int do_no_pfn(struct mm_struct *mm, struct vm_area_struct *vma,
2440 unsigned long address, pte_t *page_table, pmd_t *pmd,
2441 int write_access)
2442{
2443 spinlock_t *ptl;
2444 pte_t entry;
2445 unsigned long pfn;
2446 int ret = VM_FAULT_MINOR;
2447
2448 pte_unmap(page_table);
2449 BUG_ON(!(vma->vm_flags & VM_PFNMAP));
2450 BUG_ON(is_cow_mapping(vma->vm_flags));
2451
2452 pfn = vma->vm_ops->nopfn(vma, address & PAGE_MASK);
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002453 if (unlikely(pfn == NOPFN_OOM))
Jes Sorensenf4b81802006-09-27 01:50:10 -07002454 return VM_FAULT_OOM;
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002455 else if (unlikely(pfn == NOPFN_SIGBUS))
Jes Sorensenf4b81802006-09-27 01:50:10 -07002456 return VM_FAULT_SIGBUS;
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002457 else if (unlikely(pfn == NOPFN_REFAULT))
2458 return VM_FAULT_MINOR;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002459
2460 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2461
2462 /* Only go through if we didn't race with anybody else... */
2463 if (pte_none(*page_table)) {
2464 entry = pfn_pte(pfn, vma->vm_page_prot);
2465 if (write_access)
2466 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2467 set_pte_at(mm, address, page_table, entry);
2468 }
2469 pte_unmap_unlock(page_table, ptl);
2470 return ret;
2471}
2472
2473/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 * Fault of a previously existing named mapping. Repopulate the pte
2475 * from the encoded file_pte if possible. This enables swappable
2476 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002477 *
2478 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2479 * but allow concurrent faults), and pte mapped but not yet locked.
2480 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002482static int do_file_page(struct mm_struct *mm, struct vm_area_struct *vma,
2483 unsigned long address, pte_t *page_table, pmd_t *pmd,
2484 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485{
Hugh Dickins65500d22005-10-29 18:15:59 -07002486 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487 int err;
2488
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002489 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002490 return VM_FAULT_MINOR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491
Hugh Dickins65500d22005-10-29 18:15:59 -07002492 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
2493 /*
2494 * Page table corrupted: show pte and kill process.
2495 */
Nick Pigginb5810032005-10-29 18:16:12 -07002496 print_bad_pte(vma, orig_pte, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002497 return VM_FAULT_OOM;
2498 }
2499 /* We can then assume vm->vm_ops && vma->vm_ops->populate */
2500
2501 pgoff = pte_to_pgoff(orig_pte);
2502 err = vma->vm_ops->populate(vma, address & PAGE_MASK, PAGE_SIZE,
2503 vma->vm_page_prot, pgoff, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 if (err == -ENOMEM)
2505 return VM_FAULT_OOM;
2506 if (err)
2507 return VM_FAULT_SIGBUS;
2508 return VM_FAULT_MAJOR;
2509}
2510
2511/*
2512 * These routines also need to handle stuff like marking pages dirty
2513 * and/or accessed for architectures that don't do it in hardware (most
2514 * RISC architectures). The early dirtying is also good on the i386.
2515 *
2516 * There is also a hook called "update_mmu_cache()" that architectures
2517 * with external mmu caches can use to update those (ie the Sparc or
2518 * PowerPC hashed page tables that act as extended TLBs).
2519 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07002520 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2521 * but allow concurrent faults), and pte mapped but not yet locked.
2522 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523 */
2524static inline int handle_pte_fault(struct mm_struct *mm,
Hugh Dickins65500d22005-10-29 18:15:59 -07002525 struct vm_area_struct *vma, unsigned long address,
2526 pte_t *pte, pmd_t *pmd, int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527{
2528 pte_t entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002529 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002531 entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002533 if (pte_none(entry)) {
Jes Sorensenf4b81802006-09-27 01:50:10 -07002534 if (vma->vm_ops) {
2535 if (vma->vm_ops->nopage)
2536 return do_no_page(mm, vma, address,
2537 pte, pmd,
2538 write_access);
2539 if (unlikely(vma->vm_ops->nopfn))
2540 return do_no_pfn(mm, vma, address, pte,
2541 pmd, write_access);
2542 }
2543 return do_anonymous_page(mm, vma, address,
2544 pte, pmd, write_access);
Hugh Dickins65500d22005-10-29 18:15:59 -07002545 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 if (pte_file(entry))
Hugh Dickins65500d22005-10-29 18:15:59 -07002547 return do_file_page(mm, vma, address,
2548 pte, pmd, write_access, entry);
2549 return do_swap_page(mm, vma, address,
2550 pte, pmd, write_access, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 }
2552
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002553 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002554 spin_lock(ptl);
2555 if (unlikely(!pte_same(*pte, entry)))
2556 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 if (write_access) {
2558 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002559 return do_wp_page(mm, vma, address,
2560 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 entry = pte_mkdirty(entry);
2562 }
2563 entry = pte_mkyoung(entry);
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002564 if (ptep_set_access_flags(vma, address, pte, entry, write_access)) {
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002565 update_mmu_cache(vma, address, entry);
2566 lazy_mmu_prot_update(entry);
2567 } else {
2568 /*
2569 * This is needed only for protection faults but the arch code
2570 * is not yet telling us if this is a protection fault or not.
2571 * This still avoids useless tlb flushes for .text page faults
2572 * with threads.
2573 */
2574 if (write_access)
2575 flush_tlb_page(vma, address);
2576 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002577unlock:
2578 pte_unmap_unlock(pte, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 return VM_FAULT_MINOR;
2580}
2581
2582/*
2583 * By the time we get here, we already hold the mm semaphore
2584 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002585int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 unsigned long address, int write_access)
2587{
2588 pgd_t *pgd;
2589 pud_t *pud;
2590 pmd_t *pmd;
2591 pte_t *pte;
2592
2593 __set_current_state(TASK_RUNNING);
2594
Christoph Lameterf8891e52006-06-30 01:55:45 -07002595 count_vm_event(PGFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01002597 if (unlikely(is_vm_hugetlb_page(vma)))
2598 return hugetlb_fault(mm, vma, address, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601 pud = pud_alloc(mm, pgd, address);
2602 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002603 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604 pmd = pmd_alloc(mm, pud, address);
2605 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002606 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 pte = pte_alloc_map(mm, pmd, address);
2608 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002609 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610
Hugh Dickinsc74df322005-10-29 18:16:23 -07002611 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612}
2613
Arnd Bergmann67207b92005-11-15 15:53:48 -05002614EXPORT_SYMBOL_GPL(__handle_mm_fault);
2615
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616#ifndef __PAGETABLE_PUD_FOLDED
2617/*
2618 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002619 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002621int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002623 pud_t *new = pud_alloc_one(mm, address);
2624 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002625 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626
Hugh Dickins872fec12005-10-29 18:16:21 -07002627 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002628 if (pgd_present(*pgd)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629 pud_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002630 else
2631 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07002632 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002633 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634}
2635#endif /* __PAGETABLE_PUD_FOLDED */
2636
2637#ifndef __PAGETABLE_PMD_FOLDED
2638/*
2639 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002640 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002642int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002644 pmd_t *new = pmd_alloc_one(mm, address);
2645 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002646 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647
Hugh Dickins872fec12005-10-29 18:16:21 -07002648 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07002650 if (pud_present(*pud)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 pmd_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002652 else
2653 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07002655 if (pgd_present(*pud)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656 pmd_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002657 else
2658 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07002660 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002661 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662}
2663#endif /* __PAGETABLE_PMD_FOLDED */
2664
2665int make_pages_present(unsigned long addr, unsigned long end)
2666{
2667 int ret, len, write;
2668 struct vm_area_struct * vma;
2669
2670 vma = find_vma(current->mm, addr);
2671 if (!vma)
2672 return -1;
2673 write = (vma->vm_flags & VM_WRITE) != 0;
Eric Sesterhenn5bcb28b2006-03-26 18:30:52 +02002674 BUG_ON(addr >= end);
2675 BUG_ON(end > vma->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 len = (end+PAGE_SIZE-1)/PAGE_SIZE-addr/PAGE_SIZE;
2677 ret = get_user_pages(current, current->mm, addr,
2678 len, write, 0, NULL, NULL);
2679 if (ret < 0)
2680 return ret;
2681 return ret == len ? 0 : -1;
2682}
2683
2684/*
2685 * Map a vmalloc()-space virtual address to the physical page.
2686 */
2687struct page * vmalloc_to_page(void * vmalloc_addr)
2688{
2689 unsigned long addr = (unsigned long) vmalloc_addr;
2690 struct page *page = NULL;
2691 pgd_t *pgd = pgd_offset_k(addr);
2692 pud_t *pud;
2693 pmd_t *pmd;
2694 pte_t *ptep, pte;
2695
2696 if (!pgd_none(*pgd)) {
2697 pud = pud_offset(pgd, addr);
2698 if (!pud_none(*pud)) {
2699 pmd = pmd_offset(pud, addr);
2700 if (!pmd_none(*pmd)) {
2701 ptep = pte_offset_map(pmd, addr);
2702 pte = *ptep;
2703 if (pte_present(pte))
2704 page = pte_page(pte);
2705 pte_unmap(ptep);
2706 }
2707 }
2708 }
2709 return page;
2710}
2711
2712EXPORT_SYMBOL(vmalloc_to_page);
2713
2714/*
2715 * Map a vmalloc()-space virtual address to the physical page frame number.
2716 */
2717unsigned long vmalloc_to_pfn(void * vmalloc_addr)
2718{
2719 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
2720}
2721
2722EXPORT_SYMBOL(vmalloc_to_pfn);
2723
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724#if !defined(__HAVE_ARCH_GATE_AREA)
2725
2726#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07002727static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728
2729static int __init gate_vma_init(void)
2730{
2731 gate_vma.vm_mm = NULL;
2732 gate_vma.vm_start = FIXADDR_USER_START;
2733 gate_vma.vm_end = FIXADDR_USER_END;
Roland McGrathb6558c42007-01-26 00:56:47 -08002734 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
2735 gate_vma.vm_page_prot = __P101;
Roland McGrathf47aef52007-01-26 00:56:49 -08002736 /*
2737 * Make sure the vDSO gets into every core dump.
2738 * Dumping its contents makes post-mortem fully interpretable later
2739 * without matching up the same kernel and hardware config to see
2740 * what PC values meant.
2741 */
2742 gate_vma.vm_flags |= VM_ALWAYSDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743 return 0;
2744}
2745__initcall(gate_vma_init);
2746#endif
2747
2748struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2749{
2750#ifdef AT_SYSINFO_EHDR
2751 return &gate_vma;
2752#else
2753 return NULL;
2754#endif
2755}
2756
2757int in_gate_area_no_task(unsigned long addr)
2758{
2759#ifdef AT_SYSINFO_EHDR
2760 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2761 return 1;
2762#endif
2763 return 0;
2764}
2765
2766#endif /* __HAVE_ARCH_GATE_AREA */
David Howells0ec76a12006-09-27 01:50:15 -07002767
2768/*
2769 * Access another process' address space.
2770 * Source/target buffer must be kernel space,
2771 * Do not walk the page table directly, use get_user_pages
2772 */
2773int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2774{
2775 struct mm_struct *mm;
2776 struct vm_area_struct *vma;
2777 struct page *page;
2778 void *old_buf = buf;
2779
2780 mm = get_task_mm(tsk);
2781 if (!mm)
2782 return 0;
2783
2784 down_read(&mm->mmap_sem);
2785 /* ignore errors, just check how much was sucessfully transfered */
2786 while (len) {
2787 int bytes, ret, offset;
2788 void *maddr;
2789
2790 ret = get_user_pages(tsk, mm, addr, 1,
2791 write, 1, &page, &vma);
2792 if (ret <= 0)
2793 break;
2794
2795 bytes = len;
2796 offset = addr & (PAGE_SIZE-1);
2797 if (bytes > PAGE_SIZE-offset)
2798 bytes = PAGE_SIZE-offset;
2799
2800 maddr = kmap(page);
2801 if (write) {
2802 copy_to_user_page(vma, page, addr,
2803 maddr + offset, buf, bytes);
2804 set_page_dirty_lock(page);
2805 } else {
2806 copy_from_user_page(vma, page, addr,
2807 buf, maddr + offset, bytes);
2808 }
2809 kunmap(page);
2810 page_cache_release(page);
2811 len -= bytes;
2812 buf += bytes;
2813 addr += bytes;
2814 }
2815 up_read(&mm->mmap_sem);
2816 mmput(mm);
2817
2818 return buf - old_buf;
2819}