blob: 92a3ebd8d7951daff767f409836c8490cf283028 [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);
470 pte = *src_pte;
471 }
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);
484 page_dup_rmap(page);
485 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);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 do {
511 /*
512 * We are holding two locks at this point - either of them
513 * could generate latencies in another task on another CPU.
514 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700515 if (progress >= 32) {
516 progress = 0;
517 if (need_resched() ||
Hugh Dickinsc74df322005-10-29 18:16:23 -0700518 need_lockbreak(src_ptl) ||
519 need_lockbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700520 break;
521 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 if (pte_none(*src_pte)) {
523 progress++;
524 continue;
525 }
Hugh Dickins8c103762005-10-29 18:16:13 -0700526 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 progress += 8;
528 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529
Hugh Dickinsc74df322005-10-29 18:16:23 -0700530 spin_unlock(src_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 pte_unmap_nested(src_pte - 1);
Hugh Dickinsae859762005-10-29 18:16:05 -0700532 add_mm_rss(dst_mm, rss[0], rss[1]);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700533 pte_unmap_unlock(dst_pte - 1, dst_ptl);
534 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 if (addr != end)
536 goto again;
537 return 0;
538}
539
540static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
541 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
542 unsigned long addr, unsigned long end)
543{
544 pmd_t *src_pmd, *dst_pmd;
545 unsigned long next;
546
547 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
548 if (!dst_pmd)
549 return -ENOMEM;
550 src_pmd = pmd_offset(src_pud, addr);
551 do {
552 next = pmd_addr_end(addr, end);
553 if (pmd_none_or_clear_bad(src_pmd))
554 continue;
555 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
556 vma, addr, next))
557 return -ENOMEM;
558 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
559 return 0;
560}
561
562static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
563 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
564 unsigned long addr, unsigned long end)
565{
566 pud_t *src_pud, *dst_pud;
567 unsigned long next;
568
569 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
570 if (!dst_pud)
571 return -ENOMEM;
572 src_pud = pud_offset(src_pgd, addr);
573 do {
574 next = pud_addr_end(addr, end);
575 if (pud_none_or_clear_bad(src_pud))
576 continue;
577 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
578 vma, addr, next))
579 return -ENOMEM;
580 } while (dst_pud++, src_pud++, addr = next, addr != end);
581 return 0;
582}
583
584int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
585 struct vm_area_struct *vma)
586{
587 pgd_t *src_pgd, *dst_pgd;
588 unsigned long next;
589 unsigned long addr = vma->vm_start;
590 unsigned long end = vma->vm_end;
591
Nick Piggind9928952005-08-28 16:49:11 +1000592 /*
593 * Don't copy ptes where a page fault will fill them correctly.
594 * Fork becomes much lighter when there are big shared or private
595 * readonly mappings. The tradeoff is that copy_page_range is more
596 * efficient than faulting.
597 */
Linus Torvalds4d7672b2005-12-16 10:21:23 -0800598 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
Nick Piggind9928952005-08-28 16:49:11 +1000599 if (!vma->anon_vma)
600 return 0;
601 }
602
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 if (is_vm_hugetlb_page(vma))
604 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
605
606 dst_pgd = pgd_offset(dst_mm, addr);
607 src_pgd = pgd_offset(src_mm, addr);
608 do {
609 next = pgd_addr_end(addr, end);
610 if (pgd_none_or_clear_bad(src_pgd))
611 continue;
612 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
613 vma, addr, next))
614 return -ENOMEM;
615 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
616 return 0;
617}
618
Robin Holt51c6f662005-11-13 16:06:42 -0800619static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700620 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800622 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623{
Nick Pigginb5810032005-10-29 18:16:12 -0700624 struct mm_struct *mm = tlb->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 pte_t *pte;
Hugh Dickins508034a2005-10-29 18:16:30 -0700626 spinlock_t *ptl;
Hugh Dickinsae859762005-10-29 18:16:05 -0700627 int file_rss = 0;
628 int anon_rss = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629
Hugh Dickins508034a2005-10-29 18:16:30 -0700630 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 do {
632 pte_t ptent = *pte;
Robin Holt51c6f662005-11-13 16:06:42 -0800633 if (pte_none(ptent)) {
634 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800636 }
Hugh Dickins6f5e6b92006-03-16 23:04:09 -0800637
638 (*zap_work) -= PAGE_SIZE;
639
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800641 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -0800642
Linus Torvalds6aab3412005-11-28 14:34:23 -0800643 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 if (unlikely(details) && page) {
645 /*
646 * unmap_shared_mapping_pages() wants to
647 * invalidate cache without truncating:
648 * unmap shared but keep private pages.
649 */
650 if (details->check_mapping &&
651 details->check_mapping != page->mapping)
652 continue;
653 /*
654 * Each page->index must be checked when
655 * invalidating or truncating nonlinear.
656 */
657 if (details->nonlinear_vma &&
658 (page->index < details->first_index ||
659 page->index > details->last_index))
660 continue;
661 }
Nick Pigginb5810032005-10-29 18:16:12 -0700662 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -0700663 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 tlb_remove_tlb_entry(tlb, pte, addr);
665 if (unlikely(!page))
666 continue;
667 if (unlikely(details) && details->nonlinear_vma
668 && linear_page_index(details->nonlinear_vma,
669 addr) != page->index)
Nick Pigginb5810032005-10-29 18:16:12 -0700670 set_pte_at(mm, addr, pte,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 pgoff_to_pte(page->index));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 if (PageAnon(page))
Hugh Dickins86d912f2005-10-29 18:16:14 -0700673 anon_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700674 else {
675 if (pte_dirty(ptent))
676 set_page_dirty(page);
677 if (pte_young(ptent))
678 mark_page_accessed(page);
Hugh Dickins86d912f2005-10-29 18:16:14 -0700679 file_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700680 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 page_remove_rmap(page);
682 tlb_remove_page(tlb, page);
683 continue;
684 }
685 /*
686 * If details->check_mapping, we leave swap entries;
687 * if details->nonlinear_vma, we leave file entries.
688 */
689 if (unlikely(details))
690 continue;
691 if (!pte_file(ptent))
692 free_swap_and_cache(pte_to_swp_entry(ptent));
Nick Pigginb5810032005-10-29 18:16:12 -0700693 pte_clear_full(mm, addr, pte, tlb->fullmm);
Robin Holt51c6f662005-11-13 16:06:42 -0800694 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
Hugh Dickinsae859762005-10-29 18:16:05 -0700695
Hugh Dickins86d912f2005-10-29 18:16:14 -0700696 add_mm_rss(mm, file_rss, anon_rss);
Hugh Dickins508034a2005-10-29 18:16:30 -0700697 pte_unmap_unlock(pte - 1, ptl);
Robin Holt51c6f662005-11-13 16:06:42 -0800698
699 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700}
701
Robin Holt51c6f662005-11-13 16:06:42 -0800702static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700703 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800705 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706{
707 pmd_t *pmd;
708 unsigned long next;
709
710 pmd = pmd_offset(pud, addr);
711 do {
712 next = pmd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800713 if (pmd_none_or_clear_bad(pmd)) {
714 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800716 }
717 next = zap_pte_range(tlb, vma, pmd, addr, next,
718 zap_work, details);
719 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
720
721 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722}
723
Robin Holt51c6f662005-11-13 16:06:42 -0800724static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700725 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800727 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728{
729 pud_t *pud;
730 unsigned long next;
731
732 pud = pud_offset(pgd, addr);
733 do {
734 next = pud_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800735 if (pud_none_or_clear_bad(pud)) {
736 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800738 }
739 next = zap_pmd_range(tlb, vma, pud, addr, next,
740 zap_work, details);
741 } while (pud++, addr = next, (addr != end && *zap_work > 0));
742
743 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744}
745
Robin Holt51c6f662005-11-13 16:06:42 -0800746static unsigned long unmap_page_range(struct mmu_gather *tlb,
747 struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800749 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750{
751 pgd_t *pgd;
752 unsigned long next;
753
754 if (details && !details->check_mapping && !details->nonlinear_vma)
755 details = NULL;
756
757 BUG_ON(addr >= end);
758 tlb_start_vma(tlb, vma);
759 pgd = pgd_offset(vma->vm_mm, addr);
760 do {
761 next = pgd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800762 if (pgd_none_or_clear_bad(pgd)) {
763 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800765 }
766 next = zap_pud_range(tlb, vma, pgd, addr, next,
767 zap_work, details);
768 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 tlb_end_vma(tlb, vma);
Robin Holt51c6f662005-11-13 16:06:42 -0800770
771 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772}
773
774#ifdef CONFIG_PREEMPT
775# define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
776#else
777/* No preempt: go for improved straight-line efficiency */
778# define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
779#endif
780
781/**
782 * unmap_vmas - unmap a range of memory covered by a list of vma's
783 * @tlbp: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 * @vma: the starting vma
785 * @start_addr: virtual address at which to start unmapping
786 * @end_addr: virtual address at which to end unmapping
787 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
788 * @details: details of nonlinear truncation or shared cache invalidation
789 *
Hugh Dickinsee39b372005-04-19 13:29:15 -0700790 * Returns the end address of the unmapping (restart addr if interrupted).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700792 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700794 * We aim to not hold locks for too long (for scheduling latency reasons).
795 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 * return the ending mmu_gather to the caller.
797 *
798 * Only addresses between `start' and `end' will be unmapped.
799 *
800 * The VMA list must be sorted in ascending virtual address order.
801 *
802 * unmap_vmas() assumes that the caller will flush the whole unmapped address
803 * range after unmap_vmas() returns. So the only responsibility here is to
804 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
805 * drops the lock and schedules.
806 */
Hugh Dickins508034a2005-10-29 18:16:30 -0700807unsigned long unmap_vmas(struct mmu_gather **tlbp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 struct vm_area_struct *vma, unsigned long start_addr,
809 unsigned long end_addr, unsigned long *nr_accounted,
810 struct zap_details *details)
811{
Robin Holt51c6f662005-11-13 16:06:42 -0800812 long zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
814 int tlb_start_valid = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -0700815 unsigned long start = start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700817 int fullmm = (*tlbp)->fullmm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
819 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 unsigned long end;
821
822 start = max(vma->vm_start, start_addr);
823 if (start >= vma->vm_end)
824 continue;
825 end = min(vma->vm_end, end_addr);
826 if (end <= vma->vm_start)
827 continue;
828
829 if (vma->vm_flags & VM_ACCOUNT)
830 *nr_accounted += (end - start) >> PAGE_SHIFT;
831
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 while (start != end) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 if (!tlb_start_valid) {
834 tlb_start = start;
835 tlb_start_valid = 1;
836 }
837
Robin Holt51c6f662005-11-13 16:06:42 -0800838 if (unlikely(is_vm_hugetlb_page(vma))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 unmap_hugepage_range(vma, start, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800840 zap_work -= (end - start) /
841 (HPAGE_SIZE / PAGE_SIZE);
842 start = end;
843 } else
844 start = unmap_page_range(*tlbp, vma,
845 start, end, &zap_work, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846
Robin Holt51c6f662005-11-13 16:06:42 -0800847 if (zap_work > 0) {
848 BUG_ON(start != end);
849 break;
850 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851
852 tlb_finish_mmu(*tlbp, tlb_start, start);
853
854 if (need_resched() ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 (i_mmap_lock && need_lockbreak(i_mmap_lock))) {
856 if (i_mmap_lock) {
Hugh Dickins508034a2005-10-29 18:16:30 -0700857 *tlbp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 goto out;
859 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 }
862
Hugh Dickins508034a2005-10-29 18:16:30 -0700863 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 tlb_start_valid = 0;
Robin Holt51c6f662005-11-13 16:06:42 -0800865 zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 }
867 }
868out:
Hugh Dickinsee39b372005-04-19 13:29:15 -0700869 return start; /* which is now the end (or restart) address */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870}
871
872/**
873 * zap_page_range - remove user pages in a given range
874 * @vma: vm_area_struct holding the applicable pages
875 * @address: starting address of pages to zap
876 * @size: number of bytes to zap
877 * @details: details of nonlinear truncation or shared cache invalidation
878 */
Hugh Dickinsee39b372005-04-19 13:29:15 -0700879unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 unsigned long size, struct zap_details *details)
881{
882 struct mm_struct *mm = vma->vm_mm;
883 struct mmu_gather *tlb;
884 unsigned long end = address + size;
885 unsigned long nr_accounted = 0;
886
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -0700889 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -0700890 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
891 if (tlb)
892 tlb_finish_mmu(tlb, address, end);
Hugh Dickinsee39b372005-04-19 13:29:15 -0700893 return end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894}
895
896/*
897 * Do a quick page-table lookup for a single page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800899struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700900 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901{
902 pgd_t *pgd;
903 pud_t *pud;
904 pmd_t *pmd;
905 pte_t *ptep, pte;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700906 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 struct page *page;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800908 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700910 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
911 if (!IS_ERR(page)) {
912 BUG_ON(flags & FOLL_GET);
913 goto out;
914 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700916 page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 pgd = pgd_offset(mm, address);
918 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700919 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920
921 pud = pud_offset(pgd, address);
922 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700923 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924
925 pmd = pmd_offset(pud, address);
926 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700927 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700929 if (pmd_huge(*pmd)) {
930 BUG_ON(flags & FOLL_GET);
931 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
932 goto out;
933 }
934
935 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 if (!ptep)
937 goto out;
938
939 pte = *ptep;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700940 if (!pte_present(pte))
941 goto unlock;
942 if ((flags & FOLL_WRITE) && !pte_write(pte))
943 goto unlock;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800944 page = vm_normal_page(vma, address, pte);
945 if (unlikely(!page))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700946 goto unlock;
947
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700948 if (flags & FOLL_GET)
949 get_page(page);
950 if (flags & FOLL_TOUCH) {
951 if ((flags & FOLL_WRITE) &&
952 !pte_dirty(pte) && !PageDirty(page))
953 set_page_dirty(page);
954 mark_page_accessed(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700956unlock:
957 pte_unmap_unlock(ptep, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958out:
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700959 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700961no_page_table:
962 /*
963 * When core dumping an enormous anonymous area that nobody
964 * has touched so far, we don't want to allocate page tables.
965 */
966 if (flags & FOLL_ANON) {
967 page = ZERO_PAGE(address);
968 if (flags & FOLL_GET)
969 get_page(page);
970 BUG_ON(flags & FOLL_WRITE);
971 }
972 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973}
974
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
976 unsigned long start, int len, int write, int force,
977 struct page **pages, struct vm_area_struct **vmas)
978{
979 int i;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700980 unsigned int vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981
982 /*
983 * Require read or write permissions.
984 * If 'force' is set, we only require the "MAY" flags.
985 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700986 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
987 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 i = 0;
989
990 do {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700991 struct vm_area_struct *vma;
992 unsigned int foll_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
994 vma = find_extend_vma(mm, start);
995 if (!vma && in_gate_area(tsk, start)) {
996 unsigned long pg = start & PAGE_MASK;
997 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
998 pgd_t *pgd;
999 pud_t *pud;
1000 pmd_t *pmd;
1001 pte_t *pte;
1002 if (write) /* user gate pages are read-only */
1003 return i ? : -EFAULT;
1004 if (pg > TASK_SIZE)
1005 pgd = pgd_offset_k(pg);
1006 else
1007 pgd = pgd_offset_gate(mm, pg);
1008 BUG_ON(pgd_none(*pgd));
1009 pud = pud_offset(pgd, pg);
1010 BUG_ON(pud_none(*pud));
1011 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001012 if (pmd_none(*pmd))
1013 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001015 if (pte_none(*pte)) {
1016 pte_unmap(pte);
1017 return i ? : -EFAULT;
1018 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 if (pages) {
Nick Pigginfa2a4552005-11-29 18:43:17 +11001020 struct page *page = vm_normal_page(gate_vma, start, *pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001021 pages[i] = page;
1022 if (page)
1023 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 }
1025 pte_unmap(pte);
1026 if (vmas)
1027 vmas[i] = gate_vma;
1028 i++;
1029 start += PAGE_SIZE;
1030 len--;
1031 continue;
1032 }
1033
Linus Torvalds1ff80382005-12-12 16:24:33 -08001034 if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001035 || !(vm_flags & vma->vm_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 return i ? : -EFAULT;
1037
1038 if (is_vm_hugetlb_page(vma)) {
1039 i = follow_hugetlb_page(mm, vma, pages, vmas,
1040 &start, &len, i);
1041 continue;
1042 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001043
1044 foll_flags = FOLL_TOUCH;
1045 if (pages)
1046 foll_flags |= FOLL_GET;
1047 if (!write && !(vma->vm_flags & VM_LOCKED) &&
1048 (!vma->vm_ops || !vma->vm_ops->nopage))
1049 foll_flags |= FOLL_ANON;
1050
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 do {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001052 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001054 if (write)
1055 foll_flags |= FOLL_WRITE;
1056
1057 cond_resched();
Linus Torvalds6aab3412005-11-28 14:34:23 -08001058 while (!(page = follow_page(vma, start, foll_flags))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001059 int ret;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001060 ret = __handle_mm_fault(mm, vma, start,
1061 foll_flags & FOLL_WRITE);
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001062 /*
1063 * The VM_FAULT_WRITE bit tells us that do_wp_page has
1064 * broken COW when necessary, even if maybe_mkwrite
1065 * decided not to set pte_write. We can thus safely do
1066 * subsequent page lookups as if they were reads.
1067 */
1068 if (ret & VM_FAULT_WRITE)
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001069 foll_flags &= ~FOLL_WRITE;
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001070
1071 switch (ret & ~VM_FAULT_WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 case VM_FAULT_MINOR:
1073 tsk->min_flt++;
1074 break;
1075 case VM_FAULT_MAJOR:
1076 tsk->maj_flt++;
1077 break;
1078 case VM_FAULT_SIGBUS:
1079 return i ? i : -EFAULT;
1080 case VM_FAULT_OOM:
1081 return i ? i : -ENOMEM;
1082 default:
1083 BUG();
1084 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 }
1086 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001087 pages[i] = page;
James Bottomley03beb072006-03-26 01:36:57 -08001088
1089 flush_anon_page(page, start);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001090 flush_dcache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 }
1092 if (vmas)
1093 vmas[i] = vma;
1094 i++;
1095 start += PAGE_SIZE;
1096 len--;
Hugh Dickins08ef4722005-06-21 17:15:10 -07001097 } while (len && start < vma->vm_end);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001098 } while (len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 return i;
1100}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101EXPORT_SYMBOL(get_user_pages);
1102
1103static int zeromap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1104 unsigned long addr, unsigned long end, pgprot_t prot)
1105{
1106 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001107 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108
Hugh Dickinsc74df322005-10-29 18:16:23 -07001109 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 if (!pte)
1111 return -ENOMEM;
1112 do {
Nick Pigginb5810032005-10-29 18:16:12 -07001113 struct page *page = ZERO_PAGE(addr);
1114 pte_t zero_pte = pte_wrprotect(mk_pte(page, prot));
1115 page_cache_get(page);
1116 page_add_file_rmap(page);
1117 inc_mm_counter(mm, file_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 BUG_ON(!pte_none(*pte));
1119 set_pte_at(mm, addr, pte, zero_pte);
1120 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickinsc74df322005-10-29 18:16:23 -07001121 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 return 0;
1123}
1124
1125static inline int zeromap_pmd_range(struct mm_struct *mm, pud_t *pud,
1126 unsigned long addr, unsigned long end, pgprot_t prot)
1127{
1128 pmd_t *pmd;
1129 unsigned long next;
1130
1131 pmd = pmd_alloc(mm, pud, addr);
1132 if (!pmd)
1133 return -ENOMEM;
1134 do {
1135 next = pmd_addr_end(addr, end);
1136 if (zeromap_pte_range(mm, pmd, addr, next, prot))
1137 return -ENOMEM;
1138 } while (pmd++, addr = next, addr != end);
1139 return 0;
1140}
1141
1142static inline int zeromap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1143 unsigned long addr, unsigned long end, pgprot_t prot)
1144{
1145 pud_t *pud;
1146 unsigned long next;
1147
1148 pud = pud_alloc(mm, pgd, addr);
1149 if (!pud)
1150 return -ENOMEM;
1151 do {
1152 next = pud_addr_end(addr, end);
1153 if (zeromap_pmd_range(mm, pud, addr, next, prot))
1154 return -ENOMEM;
1155 } while (pud++, addr = next, addr != end);
1156 return 0;
1157}
1158
1159int zeromap_page_range(struct vm_area_struct *vma,
1160 unsigned long addr, unsigned long size, pgprot_t prot)
1161{
1162 pgd_t *pgd;
1163 unsigned long next;
1164 unsigned long end = addr + size;
1165 struct mm_struct *mm = vma->vm_mm;
1166 int err;
1167
1168 BUG_ON(addr >= end);
1169 pgd = pgd_offset(mm, addr);
1170 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 do {
1172 next = pgd_addr_end(addr, end);
1173 err = zeromap_pud_range(mm, pgd, addr, next, prot);
1174 if (err)
1175 break;
1176 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 return err;
1178}
1179
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001180pte_t * fastcall get_locked_pte(struct mm_struct *mm, unsigned long addr, spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001181{
1182 pgd_t * pgd = pgd_offset(mm, addr);
1183 pud_t * pud = pud_alloc(mm, pgd, addr);
1184 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001185 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001186 if (pmd)
1187 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1188 }
1189 return NULL;
1190}
1191
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001193 * This is the old fallback for page remapping.
1194 *
1195 * For historical reasons, it only allows reserved pages. Only
1196 * old drivers should use this, and they needed to mark their
1197 * pages reserved for the old functions anyway.
1198 */
1199static int insert_page(struct mm_struct *mm, unsigned long addr, struct page *page, pgprot_t prot)
1200{
1201 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001202 pte_t *pte;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001203 spinlock_t *ptl;
1204
1205 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001206 if (PageAnon(page))
Linus Torvalds238f58d2005-11-29 13:01:56 -08001207 goto out;
1208 retval = -ENOMEM;
1209 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001210 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001211 if (!pte)
1212 goto out;
1213 retval = -EBUSY;
1214 if (!pte_none(*pte))
1215 goto out_unlock;
1216
1217 /* Ok, finally just insert the thing.. */
1218 get_page(page);
1219 inc_mm_counter(mm, file_rss);
1220 page_add_file_rmap(page);
1221 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1222
1223 retval = 0;
1224out_unlock:
1225 pte_unmap_unlock(pte, ptl);
1226out:
1227 return retval;
1228}
1229
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001230/**
1231 * vm_insert_page - insert single page into user vma
1232 * @vma: user vma to map to
1233 * @addr: target user address of this page
1234 * @page: source kernel page
1235 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001236 * This allows drivers to insert individual pages they've allocated
1237 * into a user vma.
1238 *
1239 * The page has to be a nice clean _individual_ kernel allocation.
1240 * If you allocate a compound page, you need to have marked it as
1241 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001242 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001243 *
1244 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1245 * took an arbitrary page protection parameter. This doesn't allow
1246 * that. Your vma protection will have to be set up correctly, which
1247 * means that if you want a shared writable mapping, you'd better
1248 * ask for a shared writable mapping!
1249 *
1250 * The page does not need to be reserved.
1251 */
1252int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, struct page *page)
1253{
1254 if (addr < vma->vm_start || addr >= vma->vm_end)
1255 return -EFAULT;
1256 if (!page_count(page))
1257 return -EINVAL;
Linus Torvalds4d7672b2005-12-16 10:21:23 -08001258 vma->vm_flags |= VM_INSERTPAGE;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001259 return insert_page(vma->vm_mm, addr, page, vma->vm_page_prot);
1260}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001261EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001262
1263/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 * maps a range of physical memory into the requested pages. the old
1265 * mappings are removed. any references to nonexistent pages results
1266 * in null mappings (currently treated as "copy-on-access")
1267 */
1268static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1269 unsigned long addr, unsigned long end,
1270 unsigned long pfn, pgprot_t prot)
1271{
1272 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001273 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274
Hugh Dickinsc74df322005-10-29 18:16:23 -07001275 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 if (!pte)
1277 return -ENOMEM;
1278 do {
1279 BUG_ON(!pte_none(*pte));
Nick Pigginb5810032005-10-29 18:16:12 -07001280 set_pte_at(mm, addr, pte, pfn_pte(pfn, prot));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 pfn++;
1282 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickinsc74df322005-10-29 18:16:23 -07001283 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 return 0;
1285}
1286
1287static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1288 unsigned long addr, unsigned long end,
1289 unsigned long pfn, pgprot_t prot)
1290{
1291 pmd_t *pmd;
1292 unsigned long next;
1293
1294 pfn -= addr >> PAGE_SHIFT;
1295 pmd = pmd_alloc(mm, pud, addr);
1296 if (!pmd)
1297 return -ENOMEM;
1298 do {
1299 next = pmd_addr_end(addr, end);
1300 if (remap_pte_range(mm, pmd, addr, next,
1301 pfn + (addr >> PAGE_SHIFT), prot))
1302 return -ENOMEM;
1303 } while (pmd++, addr = next, addr != end);
1304 return 0;
1305}
1306
1307static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1308 unsigned long addr, unsigned long end,
1309 unsigned long pfn, pgprot_t prot)
1310{
1311 pud_t *pud;
1312 unsigned long next;
1313
1314 pfn -= addr >> PAGE_SHIFT;
1315 pud = pud_alloc(mm, pgd, addr);
1316 if (!pud)
1317 return -ENOMEM;
1318 do {
1319 next = pud_addr_end(addr, end);
1320 if (remap_pmd_range(mm, pud, addr, next,
1321 pfn + (addr >> PAGE_SHIFT), prot))
1322 return -ENOMEM;
1323 } while (pud++, addr = next, addr != end);
1324 return 0;
1325}
1326
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001327/**
1328 * remap_pfn_range - remap kernel memory to userspace
1329 * @vma: user vma to map to
1330 * @addr: target user address to start at
1331 * @pfn: physical address of kernel memory
1332 * @size: size of map area
1333 * @prot: page protection flags for this mapping
1334 *
1335 * Note: this is only safe if the mm semaphore is held when called.
1336 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1338 unsigned long pfn, unsigned long size, pgprot_t prot)
1339{
1340 pgd_t *pgd;
1341 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001342 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 struct mm_struct *mm = vma->vm_mm;
1344 int err;
1345
1346 /*
1347 * Physically remapped pages are special. Tell the
1348 * rest of the world about it:
1349 * VM_IO tells people not to look at these pages
1350 * (accesses can have side effects).
Hugh Dickins0b14c172005-11-21 21:32:15 -08001351 * VM_RESERVED is specified all over the place, because
1352 * in 2.4 it kept swapout's vma scan off this vma; but
1353 * in 2.6 the LRU scan won't even find its pages, so this
1354 * flag means no more than count its pages in reserved_vm,
1355 * and omit it from core dump, even when VM_IO turned off.
Linus Torvalds6aab3412005-11-28 14:34:23 -08001356 * VM_PFNMAP tells the core MM that the base pages are just
1357 * raw PFN mappings, and do not have a "struct page" associated
1358 * with them.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001359 *
1360 * There's a horrible special case to handle copy-on-write
1361 * behaviour that some programs depend on. We mark the "original"
1362 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 */
Linus Torvalds67121172005-12-11 20:38:17 -08001364 if (is_cow_mapping(vma->vm_flags)) {
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001365 if (addr != vma->vm_start || end != vma->vm_end)
Linus Torvalds7fc7e2e2005-12-11 19:57:52 -08001366 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001367 vma->vm_pgoff = pfn;
1368 }
1369
Linus Torvalds6aab3412005-11-28 14:34:23 -08001370 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371
1372 BUG_ON(addr >= end);
1373 pfn -= addr >> PAGE_SHIFT;
1374 pgd = pgd_offset(mm, addr);
1375 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 do {
1377 next = pgd_addr_end(addr, end);
1378 err = remap_pud_range(mm, pgd, addr, next,
1379 pfn + (addr >> PAGE_SHIFT), prot);
1380 if (err)
1381 break;
1382 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 return err;
1384}
1385EXPORT_SYMBOL(remap_pfn_range);
1386
1387/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001388 * handle_pte_fault chooses page fault handler according to an entry
1389 * which was read non-atomically. Before making any commitment, on
1390 * those architectures or configurations (e.g. i386 with PAE) which
1391 * might give a mix of unmatched parts, do_swap_page and do_file_page
1392 * must check under lock before unmapping the pte and proceeding
1393 * (but do_wp_page is only called after already making such a check;
1394 * and do_anonymous_page and do_no_page can safely check later on).
1395 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001396static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001397 pte_t *page_table, pte_t orig_pte)
1398{
1399 int same = 1;
1400#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1401 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001402 spinlock_t *ptl = pte_lockptr(mm, pmd);
1403 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001404 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001405 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001406 }
1407#endif
1408 pte_unmap(page_table);
1409 return same;
1410}
1411
1412/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1414 * servicing faults for write access. In the normal case, do always want
1415 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1416 * that do not have writing enabled, when used by access_process_vm.
1417 */
1418static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1419{
1420 if (likely(vma->vm_flags & VM_WRITE))
1421 pte = pte_mkwrite(pte);
1422 return pte;
1423}
1424
Linus Torvalds6aab3412005-11-28 14:34:23 -08001425static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va)
1426{
1427 /*
1428 * If the source page was a PFN mapping, we don't have
1429 * a "struct page" for it. We do a best-effort copy by
1430 * just copying from the original user address. If that
1431 * fails, we just zero-fill it. Live with it.
1432 */
1433 if (unlikely(!src)) {
1434 void *kaddr = kmap_atomic(dst, KM_USER0);
Linus Torvalds5d2a2dbbc2005-11-29 14:07:55 -08001435 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1436
1437 /*
1438 * This really shouldn't fail, because the page is there
1439 * in the page tables. But it might just be unreadable,
1440 * in which case we just give up and fill the result with
1441 * zeroes.
1442 */
1443 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Linus Torvalds6aab3412005-11-28 14:34:23 -08001444 memset(kaddr, 0, PAGE_SIZE);
1445 kunmap_atomic(kaddr, KM_USER0);
1446 return;
1447
1448 }
1449 copy_user_highpage(dst, src, va);
1450}
1451
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 * This routine handles present pages, when users try to write
1454 * to a shared page. It is done by copying the page to a new address
1455 * and decrementing the shared-page counter for the old page.
1456 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 * Note that this routine assumes that the protection checks have been
1458 * done by the caller (the low-level page fault routine in most cases).
1459 * Thus we can safely just mark it writable once we've done any necessary
1460 * COW.
1461 *
1462 * We also mark the page dirty at this point even though the page will
1463 * change only once the write actually happens. This avoids a few races,
1464 * and potentially makes it more efficient.
1465 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001466 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1467 * but allow concurrent faults), with pte both mapped and locked.
1468 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001470static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1471 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001472 spinlock_t *ptl, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473{
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001474 struct page *old_page, *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 pte_t entry;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001476 int reuse = 0, ret = VM_FAULT_MINOR;
1477 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478
Linus Torvalds6aab3412005-11-28 14:34:23 -08001479 old_page = vm_normal_page(vma, address, orig_pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001480 if (!old_page)
1481 goto gotten;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001483 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001484 * Take out anonymous pages first, anonymous shared vmas are
1485 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001486 */
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001487 if (PageAnon(old_page)) {
1488 if (!TestSetPageLocked(old_page)) {
1489 reuse = can_share_swap_page(old_page);
1490 unlock_page(old_page);
1491 }
1492 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001493 (VM_WRITE|VM_SHARED))) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001494 /*
1495 * Only catch write-faults on shared writable pages,
1496 * read-only shared pages can get COWed by
1497 * get_user_pages(.write=1, .force=1).
1498 */
David Howells9637a5e2006-06-23 02:03:43 -07001499 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
1500 /*
1501 * Notify the address space that the page is about to
1502 * become writable so that it can prohibit this or wait
1503 * for the page to get into an appropriate state.
1504 *
1505 * We do this without the lock held, so that it can
1506 * sleep if it needs to.
1507 */
1508 page_cache_get(old_page);
1509 pte_unmap_unlock(page_table, ptl);
1510
1511 if (vma->vm_ops->page_mkwrite(vma, old_page) < 0)
1512 goto unwritable_page;
1513
1514 page_cache_release(old_page);
1515
1516 /*
1517 * Since we dropped the lock we need to revalidate
1518 * the PTE as someone else may have changed it. If
1519 * they did, we just return, as we can count on the
1520 * MMU to tell us if they didn't also make it writable.
1521 */
1522 page_table = pte_offset_map_lock(mm, pmd, address,
1523 &ptl);
1524 if (!pte_same(*page_table, orig_pte))
1525 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001527 dirty_page = old_page;
1528 get_page(dirty_page);
David Howells9637a5e2006-06-23 02:03:43 -07001529 reuse = 1;
David Howells9637a5e2006-06-23 02:03:43 -07001530 }
1531
1532 if (reuse) {
1533 flush_cache_page(vma, address, pte_pfn(orig_pte));
1534 entry = pte_mkyoung(orig_pte);
1535 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1536 ptep_set_access_flags(vma, address, page_table, entry, 1);
1537 update_mmu_cache(vma, address, entry);
1538 lazy_mmu_prot_update(entry);
1539 ret |= VM_FAULT_WRITE;
1540 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542
1543 /*
1544 * Ok, we need to copy. Oh, well..
1545 */
Nick Pigginb5810032005-10-29 18:16:12 -07001546 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08001547gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001548 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549
1550 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07001551 goto oom;
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001552 if (old_page == ZERO_PAGE(address)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 new_page = alloc_zeroed_user_highpage(vma, address);
1554 if (!new_page)
Hugh Dickins65500d22005-10-29 18:15:59 -07001555 goto oom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 } else {
1557 new_page = alloc_page_vma(GFP_HIGHUSER, vma, address);
1558 if (!new_page)
Hugh Dickins65500d22005-10-29 18:15:59 -07001559 goto oom;
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001560 cow_user_page(new_page, old_page, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 }
Hugh Dickins65500d22005-10-29 18:15:59 -07001562
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 /*
1564 * Re-check the pte - we dropped the lock
1565 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001566 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07001567 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08001568 if (old_page) {
1569 page_remove_rmap(old_page);
1570 if (!PageAnon(old_page)) {
1571 dec_mm_counter(mm, file_rss);
1572 inc_mm_counter(mm, anon_rss);
1573 }
1574 } else
Hugh Dickins42946212005-10-29 18:16:05 -07001575 inc_mm_counter(mm, anon_rss);
Ben Collinseca35132005-11-29 11:45:26 -08001576 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07001577 entry = mk_pte(new_page, vma->vm_page_prot);
1578 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Anil Keshavamurthyc38c8db2006-07-14 00:23:57 -07001579 lazy_mmu_prot_update(entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07001580 ptep_establish(vma, address, page_table, entry);
1581 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 lru_cache_add_active(new_page);
Nick Piggin9617d952006-01-06 00:11:12 -08001583 page_add_new_anon_rmap(new_page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584
1585 /* Free the old page.. */
1586 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001587 ret |= VM_FAULT_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 }
Hugh Dickins920fc352005-11-21 21:32:17 -08001589 if (new_page)
1590 page_cache_release(new_page);
1591 if (old_page)
1592 page_cache_release(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001593unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001594 pte_unmap_unlock(page_table, ptl);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001595 if (dirty_page) {
Peter Zijlstraedc79b22006-09-25 23:30:58 -07001596 set_page_dirty_balance(dirty_page);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001597 put_page(dirty_page);
1598 }
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001599 return ret;
Hugh Dickins65500d22005-10-29 18:15:59 -07001600oom:
Hugh Dickins920fc352005-11-21 21:32:17 -08001601 if (old_page)
1602 page_cache_release(old_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 return VM_FAULT_OOM;
David Howells9637a5e2006-06-23 02:03:43 -07001604
1605unwritable_page:
1606 page_cache_release(old_page);
1607 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608}
1609
1610/*
1611 * Helper functions for unmap_mapping_range().
1612 *
1613 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1614 *
1615 * We have to restart searching the prio_tree whenever we drop the lock,
1616 * since the iterator is only valid while the lock is held, and anyway
1617 * a later vma might be split and reinserted earlier while lock dropped.
1618 *
1619 * The list of nonlinear vmas could be handled more efficiently, using
1620 * a placeholder, but handle it in the same way until a need is shown.
1621 * It is important to search the prio_tree before nonlinear list: a vma
1622 * may become nonlinear and be shifted from prio_tree to nonlinear list
1623 * while the lock is dropped; but never shifted from list to prio_tree.
1624 *
1625 * In order to make forward progress despite restarting the search,
1626 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1627 * quickly skip it next time around. Since the prio_tree search only
1628 * shows us those vmas affected by unmapping the range in question, we
1629 * can't efficiently keep all vmas in step with mapping->truncate_count:
1630 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1631 * mapping->truncate_count and vma->vm_truncate_count are protected by
1632 * i_mmap_lock.
1633 *
1634 * In order to make forward progress despite repeatedly restarting some
Hugh Dickinsee39b372005-04-19 13:29:15 -07001635 * large vma, note the restart_addr from unmap_vmas when it breaks out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 * and restart from that address when we reach that vma again. It might
1637 * have been split or merged, shrunk or extended, but never shifted: so
1638 * restart_addr remains valid so long as it remains in the vma's range.
1639 * unmap_mapping_range forces truncate_count to leap over page-aligned
1640 * values so we can save vma's restart_addr in its truncate_count field.
1641 */
1642#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1643
1644static void reset_vma_truncate_counts(struct address_space *mapping)
1645{
1646 struct vm_area_struct *vma;
1647 struct prio_tree_iter iter;
1648
1649 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1650 vma->vm_truncate_count = 0;
1651 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1652 vma->vm_truncate_count = 0;
1653}
1654
1655static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1656 unsigned long start_addr, unsigned long end_addr,
1657 struct zap_details *details)
1658{
1659 unsigned long restart_addr;
1660 int need_break;
1661
1662again:
1663 restart_addr = vma->vm_truncate_count;
1664 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1665 start_addr = restart_addr;
1666 if (start_addr >= end_addr) {
1667 /* Top of vma has been split off since last time */
1668 vma->vm_truncate_count = details->truncate_count;
1669 return 0;
1670 }
1671 }
1672
Hugh Dickinsee39b372005-04-19 13:29:15 -07001673 restart_addr = zap_page_range(vma, start_addr,
1674 end_addr - start_addr, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 need_break = need_resched() ||
1676 need_lockbreak(details->i_mmap_lock);
1677
Hugh Dickinsee39b372005-04-19 13:29:15 -07001678 if (restart_addr >= end_addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 /* We have now completed this vma: mark it so */
1680 vma->vm_truncate_count = details->truncate_count;
1681 if (!need_break)
1682 return 0;
1683 } else {
1684 /* Note restart_addr in vma's truncate_count field */
Hugh Dickinsee39b372005-04-19 13:29:15 -07001685 vma->vm_truncate_count = restart_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 if (!need_break)
1687 goto again;
1688 }
1689
1690 spin_unlock(details->i_mmap_lock);
1691 cond_resched();
1692 spin_lock(details->i_mmap_lock);
1693 return -EINTR;
1694}
1695
1696static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1697 struct zap_details *details)
1698{
1699 struct vm_area_struct *vma;
1700 struct prio_tree_iter iter;
1701 pgoff_t vba, vea, zba, zea;
1702
1703restart:
1704 vma_prio_tree_foreach(vma, &iter, root,
1705 details->first_index, details->last_index) {
1706 /* Skip quickly over those we have already dealt with */
1707 if (vma->vm_truncate_count == details->truncate_count)
1708 continue;
1709
1710 vba = vma->vm_pgoff;
1711 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1712 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1713 zba = details->first_index;
1714 if (zba < vba)
1715 zba = vba;
1716 zea = details->last_index;
1717 if (zea > vea)
1718 zea = vea;
1719
1720 if (unmap_mapping_range_vma(vma,
1721 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1722 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1723 details) < 0)
1724 goto restart;
1725 }
1726}
1727
1728static inline void unmap_mapping_range_list(struct list_head *head,
1729 struct zap_details *details)
1730{
1731 struct vm_area_struct *vma;
1732
1733 /*
1734 * In nonlinear VMAs there is no correspondence between virtual address
1735 * offset and file offset. So we must perform an exhaustive search
1736 * across *all* the pages in each nonlinear VMA, not just the pages
1737 * whose virtual address lies outside the file truncation point.
1738 */
1739restart:
1740 list_for_each_entry(vma, head, shared.vm_set.list) {
1741 /* Skip quickly over those we have already dealt with */
1742 if (vma->vm_truncate_count == details->truncate_count)
1743 continue;
1744 details->nonlinear_vma = vma;
1745 if (unmap_mapping_range_vma(vma, vma->vm_start,
1746 vma->vm_end, details) < 0)
1747 goto restart;
1748 }
1749}
1750
1751/**
1752 * unmap_mapping_range - unmap the portion of all mmaps
1753 * in the specified address_space corresponding to the specified
1754 * page range in the underlying file.
Martin Waitz3d410882005-06-23 22:05:21 -07001755 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 * @holebegin: byte in first page to unmap, relative to the start of
1757 * the underlying file. This will be rounded down to a PAGE_SIZE
1758 * boundary. Note that this is different from vmtruncate(), which
1759 * must keep the partial page. In contrast, we must get rid of
1760 * partial pages.
1761 * @holelen: size of prospective hole in bytes. This will be rounded
1762 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1763 * end of the file.
1764 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1765 * but 0 when invalidating pagecache, don't throw away private data.
1766 */
1767void unmap_mapping_range(struct address_space *mapping,
1768 loff_t const holebegin, loff_t const holelen, int even_cows)
1769{
1770 struct zap_details details;
1771 pgoff_t hba = holebegin >> PAGE_SHIFT;
1772 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1773
1774 /* Check for overflow. */
1775 if (sizeof(holelen) > sizeof(hlen)) {
1776 long long holeend =
1777 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1778 if (holeend & ~(long long)ULONG_MAX)
1779 hlen = ULONG_MAX - hba + 1;
1780 }
1781
1782 details.check_mapping = even_cows? NULL: mapping;
1783 details.nonlinear_vma = NULL;
1784 details.first_index = hba;
1785 details.last_index = hba + hlen - 1;
1786 if (details.last_index < details.first_index)
1787 details.last_index = ULONG_MAX;
1788 details.i_mmap_lock = &mapping->i_mmap_lock;
1789
1790 spin_lock(&mapping->i_mmap_lock);
1791
1792 /* serialize i_size write against truncate_count write */
1793 smp_wmb();
1794 /* Protect against page faults, and endless unmapping loops */
1795 mapping->truncate_count++;
1796 /*
1797 * For archs where spin_lock has inclusive semantics like ia64
1798 * this smp_mb() will prevent to read pagetable contents
1799 * before the truncate_count increment is visible to
1800 * other cpus.
1801 */
1802 smp_mb();
1803 if (unlikely(is_restart_addr(mapping->truncate_count))) {
1804 if (mapping->truncate_count == 0)
1805 reset_vma_truncate_counts(mapping);
1806 mapping->truncate_count++;
1807 }
1808 details.truncate_count = mapping->truncate_count;
1809
1810 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
1811 unmap_mapping_range_tree(&mapping->i_mmap, &details);
1812 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
1813 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
1814 spin_unlock(&mapping->i_mmap_lock);
1815}
1816EXPORT_SYMBOL(unmap_mapping_range);
1817
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001818/**
1819 * vmtruncate - unmap mappings "freed" by truncate() syscall
1820 * @inode: inode of the file used
1821 * @offset: file offset to start truncating
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 *
1823 * NOTE! We have to be ready to update the memory sharing
1824 * between the file and the memory map for a potential last
1825 * incomplete page. Ugly, but necessary.
1826 */
1827int vmtruncate(struct inode * inode, loff_t offset)
1828{
1829 struct address_space *mapping = inode->i_mapping;
1830 unsigned long limit;
1831
1832 if (inode->i_size < offset)
1833 goto do_expand;
1834 /*
1835 * truncation of in-use swapfiles is disallowed - it would cause
1836 * subsequent swapout to scribble on the now-freed blocks.
1837 */
1838 if (IS_SWAPFILE(inode))
1839 goto out_busy;
1840 i_size_write(inode, offset);
1841 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
1842 truncate_inode_pages(mapping, offset);
1843 goto out_truncate;
1844
1845do_expand:
1846 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1847 if (limit != RLIM_INFINITY && offset > limit)
1848 goto out_sig;
1849 if (offset > inode->i_sb->s_maxbytes)
1850 goto out_big;
1851 i_size_write(inode, offset);
1852
1853out_truncate:
1854 if (inode->i_op && inode->i_op->truncate)
1855 inode->i_op->truncate(inode);
1856 return 0;
1857out_sig:
1858 send_sig(SIGXFSZ, current, 0);
1859out_big:
1860 return -EFBIG;
1861out_busy:
1862 return -ETXTBSY;
1863}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864EXPORT_SYMBOL(vmtruncate);
1865
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001866int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
1867{
1868 struct address_space *mapping = inode->i_mapping;
1869
1870 /*
1871 * If the underlying filesystem is not going to provide
1872 * a way to truncate a range of blocks (punch a hole) -
1873 * we should return failure right now.
1874 */
1875 if (!inode->i_op || !inode->i_op->truncate_range)
1876 return -ENOSYS;
1877
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001878 mutex_lock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001879 down_write(&inode->i_alloc_sem);
1880 unmap_mapping_range(mapping, offset, (end - offset), 1);
1881 truncate_inode_pages_range(mapping, offset, end);
1882 inode->i_op->truncate_range(inode, offset, end);
1883 up_write(&inode->i_alloc_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001884 mutex_unlock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001885
1886 return 0;
1887}
Adrian Bunk26fc5232006-07-10 04:44:22 -07001888EXPORT_UNUSED_SYMBOL(vmtruncate_range); /* June 2006 */
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001889
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001890/**
1891 * swapin_readahead - swap in pages in hope we need them soon
1892 * @entry: swap entry of this memory
1893 * @addr: address to start
1894 * @vma: user vma this addresses belong to
1895 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 * Primitive swap readahead code. We simply read an aligned block of
1897 * (1 << page_cluster) entries in the swap area. This method is chosen
1898 * because it doesn't cost us any seek time. We also make sure to queue
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001899 * the 'original' request together with the readahead ones...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 *
1901 * This has been extended to use the NUMA policies from the mm triggering
1902 * the readahead.
1903 *
1904 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
1905 */
1906void swapin_readahead(swp_entry_t entry, unsigned long addr,struct vm_area_struct *vma)
1907{
1908#ifdef CONFIG_NUMA
1909 struct vm_area_struct *next_vma = vma ? vma->vm_next : NULL;
1910#endif
1911 int i, num;
1912 struct page *new_page;
1913 unsigned long offset;
1914
1915 /*
1916 * Get the number of handles we should do readahead io to.
1917 */
1918 num = valid_swaphandles(entry, &offset);
1919 for (i = 0; i < num; offset++, i++) {
1920 /* Ok, do the async read-ahead now */
1921 new_page = read_swap_cache_async(swp_entry(swp_type(entry),
1922 offset), vma, addr);
1923 if (!new_page)
1924 break;
1925 page_cache_release(new_page);
1926#ifdef CONFIG_NUMA
1927 /*
1928 * Find the next applicable VMA for the NUMA policy.
1929 */
1930 addr += PAGE_SIZE;
1931 if (addr == 0)
1932 vma = NULL;
1933 if (vma) {
1934 if (addr >= vma->vm_end) {
1935 vma = next_vma;
1936 next_vma = vma ? vma->vm_next : NULL;
1937 }
1938 if (vma && addr < vma->vm_start)
1939 vma = NULL;
1940 } else {
1941 if (next_vma && addr >= next_vma->vm_start) {
1942 vma = next_vma;
1943 next_vma = vma->vm_next;
1944 }
1945 }
1946#endif
1947 }
1948 lru_add_drain(); /* Push any new pages onto the LRU now */
1949}
1950
1951/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001952 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1953 * but allow concurrent faults), and pte mapped but not yet locked.
1954 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001956static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
1957 unsigned long address, pte_t *page_table, pmd_t *pmd,
1958 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001960 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 struct page *page;
Hugh Dickins65500d22005-10-29 18:15:59 -07001962 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 pte_t pte;
1964 int ret = VM_FAULT_MINOR;
1965
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001966 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001967 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07001968
1969 entry = pte_to_swp_entry(orig_pte);
Christoph Lameter06972122006-06-23 02:03:35 -07001970 if (is_migration_entry(entry)) {
1971 migration_entry_wait(mm, pmd, address);
1972 goto out;
1973 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07001974 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 page = lookup_swap_cache(entry);
1976 if (!page) {
1977 swapin_readahead(entry, address, vma);
1978 page = read_swap_cache_async(entry, vma, address);
1979 if (!page) {
1980 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001981 * Back out if somebody else faulted in this pte
1982 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001984 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 if (likely(pte_same(*page_table, orig_pte)))
1986 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07001987 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07001988 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 }
1990
1991 /* Had to read the page from swap area: Major fault */
1992 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001993 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 grab_swap_token();
1995 }
1996
Shailabh Nagar0ff92242006-07-14 00:24:37 -07001997 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 mark_page_accessed(page);
1999 lock_page(page);
2000
2001 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002002 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002004 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002005 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002006 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002007
2008 if (unlikely(!PageUptodate(page))) {
2009 ret = VM_FAULT_SIGBUS;
2010 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 }
2012
2013 /* The page isn't present yet, go ahead with the fault. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014
Hugh Dickins42946212005-10-29 18:16:05 -07002015 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 pte = mk_pte(page, vma->vm_page_prot);
2017 if (write_access && can_share_swap_page(page)) {
2018 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2019 write_access = 0;
2020 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021
2022 flush_icache_page(vma, page);
2023 set_pte_at(mm, address, page_table, pte);
2024 page_add_anon_rmap(page, vma, address);
2025
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002026 swap_free(entry);
2027 if (vm_swap_full())
2028 remove_exclusive_swap_page(page);
2029 unlock_page(page);
2030
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 if (write_access) {
2032 if (do_wp_page(mm, vma, address,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002033 page_table, pmd, ptl, pte) == VM_FAULT_OOM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 ret = VM_FAULT_OOM;
2035 goto out;
2036 }
2037
2038 /* No need to invalidate - it was non-present before */
2039 update_mmu_cache(vma, address, pte);
2040 lazy_mmu_prot_update(pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07002041unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002042 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043out:
2044 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002045out_nomap:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002046 pte_unmap_unlock(page_table, ptl);
Kirill Korotaevb8107482005-05-16 21:53:50 -07002047 unlock_page(page);
2048 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002049 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050}
2051
2052/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002053 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2054 * but allow concurrent faults), and pte mapped but not yet locked.
2055 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002057static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2058 unsigned long address, pte_t *page_table, pmd_t *pmd,
2059 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002061 struct page *page;
2062 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064
Linus Torvalds6aab3412005-11-28 14:34:23 -08002065 if (write_access) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 /* Allocate our own private page. */
2067 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068
2069 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07002070 goto oom;
2071 page = alloc_zeroed_user_highpage(vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 if (!page)
Hugh Dickins65500d22005-10-29 18:15:59 -07002073 goto oom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074
Hugh Dickins65500d22005-10-29 18:15:59 -07002075 entry = mk_pte(page, vma->vm_page_prot);
2076 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002077
2078 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2079 if (!pte_none(*page_table))
2080 goto release;
2081 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 lru_cache_add_active(page);
Nick Piggin9617d952006-01-06 00:11:12 -08002083 page_add_new_anon_rmap(page, vma, address);
Nick Pigginb5810032005-10-29 18:16:12 -07002084 } else {
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002085 /* Map the ZERO_PAGE - vm_page_prot is readonly */
2086 page = ZERO_PAGE(address);
2087 page_cache_get(page);
2088 entry = mk_pte(page, vma->vm_page_prot);
2089
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002090 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002091 spin_lock(ptl);
2092 if (!pte_none(*page_table))
2093 goto release;
Nick Pigginb5810032005-10-29 18:16:12 -07002094 inc_mm_counter(mm, file_rss);
2095 page_add_file_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 }
2097
Hugh Dickins65500d22005-10-29 18:15:59 -07002098 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099
2100 /* No need to invalidate - it was non-present before */
Hugh Dickins65500d22005-10-29 18:15:59 -07002101 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 lazy_mmu_prot_update(entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002103unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002104 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 return VM_FAULT_MINOR;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002106release:
2107 page_cache_release(page);
2108 goto unlock;
Hugh Dickins65500d22005-10-29 18:15:59 -07002109oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 return VM_FAULT_OOM;
2111}
2112
2113/*
2114 * do_no_page() tries to create a new page mapping. It aggressively
2115 * tries to share with existing pages, but makes a separate copy if
2116 * the "write_access" parameter is true in order to avoid the next
2117 * page fault.
2118 *
2119 * As this is called only for pages that do not currently exist, we
2120 * do not need to flush old virtual caches or the TLB.
2121 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002122 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2123 * but allow concurrent faults), and pte mapped but not yet locked.
2124 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002126static int do_no_page(struct mm_struct *mm, struct vm_area_struct *vma,
2127 unsigned long address, pte_t *page_table, pmd_t *pmd,
2128 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002130 spinlock_t *ptl;
Hugh Dickins65500d22005-10-29 18:15:59 -07002131 struct page *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 struct address_space *mapping = NULL;
2133 pte_t entry;
2134 unsigned int sequence = 0;
2135 int ret = VM_FAULT_MINOR;
2136 int anon = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002137 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 pte_unmap(page_table);
Hugh Dickins325f04d2005-11-29 16:55:48 +00002140 BUG_ON(vma->vm_flags & VM_PFNMAP);
2141
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 if (vma->vm_file) {
2143 mapping = vma->vm_file->f_mapping;
2144 sequence = mapping->truncate_count;
2145 smp_rmb(); /* serializes i_size against truncate_count */
2146 }
2147retry:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 new_page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret);
2149 /*
2150 * No smp_rmb is needed here as long as there's a full
2151 * spin_lock/unlock sequence inside the ->nopage callback
2152 * (for the pagecache lookup) that acts as an implicit
2153 * smp_mb() and prevents the i_size read to happen
2154 * after the next truncate_count read.
2155 */
2156
2157 /* no page was available -- either SIGBUS or OOM */
2158 if (new_page == NOPAGE_SIGBUS)
2159 return VM_FAULT_SIGBUS;
2160 if (new_page == NOPAGE_OOM)
2161 return VM_FAULT_OOM;
2162
2163 /*
2164 * Should we do an early C-O-W break?
2165 */
David Howells9637a5e2006-06-23 02:03:43 -07002166 if (write_access) {
2167 if (!(vma->vm_flags & VM_SHARED)) {
2168 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169
David Howells9637a5e2006-06-23 02:03:43 -07002170 if (unlikely(anon_vma_prepare(vma)))
2171 goto oom;
2172 page = alloc_page_vma(GFP_HIGHUSER, vma, address);
2173 if (!page)
2174 goto oom;
2175 copy_user_highpage(page, new_page, address);
2176 page_cache_release(new_page);
2177 new_page = page;
2178 anon = 1;
2179
2180 } else {
2181 /* if the page will be shareable, see if the backing
2182 * address space wants to know that the page is about
2183 * to become writable */
2184 if (vma->vm_ops->page_mkwrite &&
2185 vma->vm_ops->page_mkwrite(vma, new_page) < 0
2186 ) {
2187 page_cache_release(new_page);
2188 return VM_FAULT_SIGBUS;
2189 }
2190 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 }
2192
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002193 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 /*
2195 * For a file-backed vma, someone could have truncated or otherwise
2196 * invalidated this page. If unmap_mapping_range got called,
2197 * retry getting the page.
2198 */
2199 if (mapping && unlikely(sequence != mapping->truncate_count)) {
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002200 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002202 cond_resched();
2203 sequence = mapping->truncate_count;
2204 smp_rmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 goto retry;
2206 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207
2208 /*
2209 * This silly early PAGE_DIRTY setting removes a race
2210 * due to the bad i386 page protection. But it's valid
2211 * for other architectures too.
2212 *
2213 * Note that if write_access is true, we either now have
2214 * an exclusive copy of the page, or this is a shared mapping,
2215 * so we can make it writable and dirty to avoid having to
2216 * handle that later.
2217 */
2218 /* Only go through if we didn't race with anybody else... */
2219 if (pte_none(*page_table)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 flush_icache_page(vma, new_page);
2221 entry = mk_pte(new_page, vma->vm_page_prot);
2222 if (write_access)
2223 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2224 set_pte_at(mm, address, page_table, entry);
2225 if (anon) {
Hugh Dickins42946212005-10-29 18:16:05 -07002226 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 lru_cache_add_active(new_page);
Nick Piggin9617d952006-01-06 00:11:12 -08002228 page_add_new_anon_rmap(new_page, vma, address);
Hugh Dickinsf57e88a2005-11-21 21:32:19 -08002229 } else {
Hugh Dickins42946212005-10-29 18:16:05 -07002230 inc_mm_counter(mm, file_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 page_add_file_rmap(new_page);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002232 if (write_access) {
2233 dirty_page = new_page;
2234 get_page(dirty_page);
2235 }
Hugh Dickins42946212005-10-29 18:16:05 -07002236 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 } else {
2238 /* One of our sibling threads was faster, back out. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002240 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 }
2242
2243 /* no need to invalidate: a not-present page shouldn't be cached */
2244 update_mmu_cache(vma, address, entry);
2245 lazy_mmu_prot_update(entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002246unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002247 pte_unmap_unlock(page_table, ptl);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002248 if (dirty_page) {
Peter Zijlstraedc79b22006-09-25 23:30:58 -07002249 set_page_dirty_balance(dirty_page);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002250 put_page(dirty_page);
2251 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 return ret;
2253oom:
2254 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002255 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256}
2257
2258/*
2259 * Fault of a previously existing named mapping. Repopulate the pte
2260 * from the encoded file_pte if possible. This enables swappable
2261 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002262 *
2263 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2264 * but allow concurrent faults), and pte mapped but not yet locked.
2265 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002267static int do_file_page(struct mm_struct *mm, struct vm_area_struct *vma,
2268 unsigned long address, pte_t *page_table, pmd_t *pmd,
2269 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270{
Hugh Dickins65500d22005-10-29 18:15:59 -07002271 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 int err;
2273
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002274 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002275 return VM_FAULT_MINOR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276
Hugh Dickins65500d22005-10-29 18:15:59 -07002277 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
2278 /*
2279 * Page table corrupted: show pte and kill process.
2280 */
Nick Pigginb5810032005-10-29 18:16:12 -07002281 print_bad_pte(vma, orig_pte, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002282 return VM_FAULT_OOM;
2283 }
2284 /* We can then assume vm->vm_ops && vma->vm_ops->populate */
2285
2286 pgoff = pte_to_pgoff(orig_pte);
2287 err = vma->vm_ops->populate(vma, address & PAGE_MASK, PAGE_SIZE,
2288 vma->vm_page_prot, pgoff, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 if (err == -ENOMEM)
2290 return VM_FAULT_OOM;
2291 if (err)
2292 return VM_FAULT_SIGBUS;
2293 return VM_FAULT_MAJOR;
2294}
2295
2296/*
2297 * These routines also need to handle stuff like marking pages dirty
2298 * and/or accessed for architectures that don't do it in hardware (most
2299 * RISC architectures). The early dirtying is also good on the i386.
2300 *
2301 * There is also a hook called "update_mmu_cache()" that architectures
2302 * with external mmu caches can use to update those (ie the Sparc or
2303 * PowerPC hashed page tables that act as extended TLBs).
2304 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07002305 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2306 * but allow concurrent faults), and pte mapped but not yet locked.
2307 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 */
2309static inline int handle_pte_fault(struct mm_struct *mm,
Hugh Dickins65500d22005-10-29 18:15:59 -07002310 struct vm_area_struct *vma, unsigned long address,
2311 pte_t *pte, pmd_t *pmd, int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312{
2313 pte_t entry;
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002314 pte_t old_entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002315 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002317 old_entry = entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002319 if (pte_none(entry)) {
2320 if (!vma->vm_ops || !vma->vm_ops->nopage)
2321 return do_anonymous_page(mm, vma, address,
2322 pte, pmd, write_access);
2323 return do_no_page(mm, vma, address,
2324 pte, pmd, write_access);
2325 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 if (pte_file(entry))
Hugh Dickins65500d22005-10-29 18:15:59 -07002327 return do_file_page(mm, vma, address,
2328 pte, pmd, write_access, entry);
2329 return do_swap_page(mm, vma, address,
2330 pte, pmd, write_access, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 }
2332
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002333 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002334 spin_lock(ptl);
2335 if (unlikely(!pte_same(*pte, entry)))
2336 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 if (write_access) {
2338 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002339 return do_wp_page(mm, vma, address,
2340 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 entry = pte_mkdirty(entry);
2342 }
2343 entry = pte_mkyoung(entry);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002344 if (!pte_same(old_entry, entry)) {
2345 ptep_set_access_flags(vma, address, pte, entry, write_access);
2346 update_mmu_cache(vma, address, entry);
2347 lazy_mmu_prot_update(entry);
2348 } else {
2349 /*
2350 * This is needed only for protection faults but the arch code
2351 * is not yet telling us if this is a protection fault or not.
2352 * This still avoids useless tlb flushes for .text page faults
2353 * with threads.
2354 */
2355 if (write_access)
2356 flush_tlb_page(vma, address);
2357 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002358unlock:
2359 pte_unmap_unlock(pte, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 return VM_FAULT_MINOR;
2361}
2362
2363/*
2364 * By the time we get here, we already hold the mm semaphore
2365 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002366int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367 unsigned long address, int write_access)
2368{
2369 pgd_t *pgd;
2370 pud_t *pud;
2371 pmd_t *pmd;
2372 pte_t *pte;
2373
2374 __set_current_state(TASK_RUNNING);
2375
Christoph Lameterf8891e52006-06-30 01:55:45 -07002376 count_vm_event(PGFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01002378 if (unlikely(is_vm_hugetlb_page(vma)))
2379 return hugetlb_fault(mm, vma, address, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 pud = pud_alloc(mm, pgd, address);
2383 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002384 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 pmd = pmd_alloc(mm, pud, address);
2386 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002387 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 pte = pte_alloc_map(mm, pmd, address);
2389 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002390 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391
Hugh Dickinsc74df322005-10-29 18:16:23 -07002392 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393}
2394
Arnd Bergmann67207b92005-11-15 15:53:48 -05002395EXPORT_SYMBOL_GPL(__handle_mm_fault);
2396
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397#ifndef __PAGETABLE_PUD_FOLDED
2398/*
2399 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002400 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002402int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002404 pud_t *new = pud_alloc_one(mm, address);
2405 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002406 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407
Hugh Dickins872fec12005-10-29 18:16:21 -07002408 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002409 if (pgd_present(*pgd)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 pud_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002411 else
2412 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07002413 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002414 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415}
Alan Sterne0f39592005-11-28 13:43:44 -08002416#else
2417/* Workaround for gcc 2.96 */
2418int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
2419{
2420 return 0;
2421}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422#endif /* __PAGETABLE_PUD_FOLDED */
2423
2424#ifndef __PAGETABLE_PMD_FOLDED
2425/*
2426 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002427 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002429int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002431 pmd_t *new = pmd_alloc_one(mm, address);
2432 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002433 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434
Hugh Dickins872fec12005-10-29 18:16:21 -07002435 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07002437 if (pud_present(*pud)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 pmd_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002439 else
2440 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07002442 if (pgd_present(*pud)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 pmd_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002444 else
2445 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07002447 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002448 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449}
Alan Sterne0f39592005-11-28 13:43:44 -08002450#else
2451/* Workaround for gcc 2.96 */
2452int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
2453{
2454 return 0;
2455}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456#endif /* __PAGETABLE_PMD_FOLDED */
2457
2458int make_pages_present(unsigned long addr, unsigned long end)
2459{
2460 int ret, len, write;
2461 struct vm_area_struct * vma;
2462
2463 vma = find_vma(current->mm, addr);
2464 if (!vma)
2465 return -1;
2466 write = (vma->vm_flags & VM_WRITE) != 0;
Eric Sesterhenn5bcb28b2006-03-26 18:30:52 +02002467 BUG_ON(addr >= end);
2468 BUG_ON(end > vma->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 len = (end+PAGE_SIZE-1)/PAGE_SIZE-addr/PAGE_SIZE;
2470 ret = get_user_pages(current, current->mm, addr,
2471 len, write, 0, NULL, NULL);
2472 if (ret < 0)
2473 return ret;
2474 return ret == len ? 0 : -1;
2475}
2476
2477/*
2478 * Map a vmalloc()-space virtual address to the physical page.
2479 */
2480struct page * vmalloc_to_page(void * vmalloc_addr)
2481{
2482 unsigned long addr = (unsigned long) vmalloc_addr;
2483 struct page *page = NULL;
2484 pgd_t *pgd = pgd_offset_k(addr);
2485 pud_t *pud;
2486 pmd_t *pmd;
2487 pte_t *ptep, pte;
2488
2489 if (!pgd_none(*pgd)) {
2490 pud = pud_offset(pgd, addr);
2491 if (!pud_none(*pud)) {
2492 pmd = pmd_offset(pud, addr);
2493 if (!pmd_none(*pmd)) {
2494 ptep = pte_offset_map(pmd, addr);
2495 pte = *ptep;
2496 if (pte_present(pte))
2497 page = pte_page(pte);
2498 pte_unmap(ptep);
2499 }
2500 }
2501 }
2502 return page;
2503}
2504
2505EXPORT_SYMBOL(vmalloc_to_page);
2506
2507/*
2508 * Map a vmalloc()-space virtual address to the physical page frame number.
2509 */
2510unsigned long vmalloc_to_pfn(void * vmalloc_addr)
2511{
2512 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
2513}
2514
2515EXPORT_SYMBOL(vmalloc_to_pfn);
2516
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517#if !defined(__HAVE_ARCH_GATE_AREA)
2518
2519#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07002520static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521
2522static int __init gate_vma_init(void)
2523{
2524 gate_vma.vm_mm = NULL;
2525 gate_vma.vm_start = FIXADDR_USER_START;
2526 gate_vma.vm_end = FIXADDR_USER_END;
2527 gate_vma.vm_page_prot = PAGE_READONLY;
Hugh Dickins0b14c172005-11-21 21:32:15 -08002528 gate_vma.vm_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 return 0;
2530}
2531__initcall(gate_vma_init);
2532#endif
2533
2534struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2535{
2536#ifdef AT_SYSINFO_EHDR
2537 return &gate_vma;
2538#else
2539 return NULL;
2540#endif
2541}
2542
2543int in_gate_area_no_task(unsigned long addr)
2544{
2545#ifdef AT_SYSINFO_EHDR
2546 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2547 return 1;
2548#endif
2549 return 0;
2550}
2551
2552#endif /* __HAVE_ARCH_GATE_AREA */