blob: 156861fcac436e4716537c7e5dff565dded43224 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/memory.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 */
6
7/*
8 * demand-loading started 01.12.91 - seems it is high on the list of
9 * things wanted, and it should be easy to implement. - Linus
10 */
11
12/*
13 * Ok, demand-loading was easy, shared pages a little bit tricker. Shared
14 * pages started 02.12.91, seems to work. - Linus.
15 *
16 * Tested sharing by executing about 30 /bin/sh: under the old kernel it
17 * would have taken more than the 6M I have free, but it worked well as
18 * far as I could see.
19 *
20 * Also corrected some "invalidate()"s - I wasn't doing enough of them.
21 */
22
23/*
24 * Real VM (paging to/from disk) started 18.12.91. Much more work and
25 * thought has to go into this. Oh, well..
26 * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why.
27 * Found it. Everything seems to work now.
28 * 20.12.91 - Ok, making the swap-device changeable like the root.
29 */
30
31/*
32 * 05.04.94 - Multi-page memory management added for v1.1.
33 * Idea by Alex Bligh (alex@cconcepts.co.uk)
34 *
35 * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG
36 * (Gerhard.Wichert@pdb.siemens.de)
37 *
38 * Aug/Sep 2004 Changed to four level page tables (Andi Kleen)
39 */
40
41#include <linux/kernel_stat.h>
42#include <linux/mm.h>
43#include <linux/hugetlb.h>
44#include <linux/mman.h>
45#include <linux/swap.h>
46#include <linux/highmem.h>
47#include <linux/pagemap.h>
48#include <linux/rmap.h>
49#include <linux/module.h>
Shailabh Nagar0ff92242006-07-14 00:24:37 -070050#include <linux/delayacct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include <linux/init.h>
Peter Zijlstraedc79b22006-09-25 23:30:58 -070052#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
54#include <asm/pgalloc.h>
55#include <asm/uaccess.h>
56#include <asm/tlb.h>
57#include <asm/tlbflush.h>
58#include <asm/pgtable.h>
59
60#include <linux/swapops.h>
61#include <linux/elf.h>
62
Andy Whitcroftd41dee32005-06-23 00:07:54 -070063#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -070064/* use the per-pgdat data instead for discontigmem - mbligh */
65unsigned long max_mapnr;
66struct page *mem_map;
67
68EXPORT_SYMBOL(max_mapnr);
69EXPORT_SYMBOL(mem_map);
70#endif
71
72unsigned long num_physpages;
73/*
74 * A number of key systems in x86 including ioremap() rely on the assumption
75 * that high_memory defines the upper bound on direct map memory, then end
76 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
77 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
78 * and ZONE_HIGHMEM.
79 */
80void * high_memory;
81unsigned long vmalloc_earlyreserve;
82
83EXPORT_SYMBOL(num_physpages);
84EXPORT_SYMBOL(high_memory);
85EXPORT_SYMBOL(vmalloc_earlyreserve);
86
Andi Kleena62eaf12006-02-16 23:41:58 +010087int randomize_va_space __read_mostly = 1;
88
89static int __init disable_randmaps(char *s)
90{
91 randomize_va_space = 0;
OGAWA Hirofumi9b410462006-03-31 02:30:33 -080092 return 1;
Andi Kleena62eaf12006-02-16 23:41:58 +010093}
94__setup("norandmaps", disable_randmaps);
95
96
Linus Torvalds1da177e2005-04-16 15:20:36 -070097/*
98 * If a p?d_bad entry is found while walking page tables, report
99 * the error, before resetting entry to p?d_none. Usually (but
100 * very seldom) called out from the p?d_none_or_clear_bad macros.
101 */
102
103void pgd_clear_bad(pgd_t *pgd)
104{
105 pgd_ERROR(*pgd);
106 pgd_clear(pgd);
107}
108
109void pud_clear_bad(pud_t *pud)
110{
111 pud_ERROR(*pud);
112 pud_clear(pud);
113}
114
115void pmd_clear_bad(pmd_t *pmd)
116{
117 pmd_ERROR(*pmd);
118 pmd_clear(pmd);
119}
120
121/*
122 * Note: this doesn't free the actual pages themselves. That
123 * has been handled earlier when unmapping all the memory regions.
124 */
Hugh Dickinse0da3822005-04-19 13:29:15 -0700125static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126{
Hugh Dickinse0da3822005-04-19 13:29:15 -0700127 struct page *page = pmd_page(*pmd);
128 pmd_clear(pmd);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700129 pte_lock_deinit(page);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700130 pte_free_tlb(tlb, page);
Christoph Lameterdf849a12006-06-30 01:55:38 -0700131 dec_zone_page_state(page, NR_PAGETABLE);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700132 tlb->mm->nr_ptes--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133}
134
Hugh Dickinse0da3822005-04-19 13:29:15 -0700135static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
136 unsigned long addr, unsigned long end,
137 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138{
139 pmd_t *pmd;
140 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700141 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142
Hugh Dickinse0da3822005-04-19 13:29:15 -0700143 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 do {
146 next = pmd_addr_end(addr, end);
147 if (pmd_none_or_clear_bad(pmd))
148 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700149 free_pte_range(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 } while (pmd++, addr = next, addr != end);
151
Hugh Dickinse0da3822005-04-19 13:29:15 -0700152 start &= PUD_MASK;
153 if (start < floor)
154 return;
155 if (ceiling) {
156 ceiling &= PUD_MASK;
157 if (!ceiling)
158 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700160 if (end - 1 > ceiling - 1)
161 return;
162
163 pmd = pmd_offset(pud, start);
164 pud_clear(pud);
165 pmd_free_tlb(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166}
167
Hugh Dickinse0da3822005-04-19 13:29:15 -0700168static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
169 unsigned long addr, unsigned long end,
170 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171{
172 pud_t *pud;
173 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700174 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175
Hugh Dickinse0da3822005-04-19 13:29:15 -0700176 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 pud = pud_offset(pgd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 do {
179 next = pud_addr_end(addr, end);
180 if (pud_none_or_clear_bad(pud))
181 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700182 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 } while (pud++, addr = next, addr != end);
184
Hugh Dickinse0da3822005-04-19 13:29:15 -0700185 start &= PGDIR_MASK;
186 if (start < floor)
187 return;
188 if (ceiling) {
189 ceiling &= PGDIR_MASK;
190 if (!ceiling)
191 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700193 if (end - 1 > ceiling - 1)
194 return;
195
196 pud = pud_offset(pgd, start);
197 pgd_clear(pgd);
198 pud_free_tlb(tlb, pud);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199}
200
201/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700202 * This function frees user-level page tables of a process.
203 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 * Must be called with pagetable lock held.
205 */
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700206void free_pgd_range(struct mmu_gather **tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700207 unsigned long addr, unsigned long end,
208 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209{
210 pgd_t *pgd;
211 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700212 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213
Hugh Dickinse0da3822005-04-19 13:29:15 -0700214 /*
215 * The next few lines have given us lots of grief...
216 *
217 * Why are we testing PMD* at this top level? Because often
218 * there will be no work to do at all, and we'd prefer not to
219 * go all the way down to the bottom just to discover that.
220 *
221 * Why all these "- 1"s? Because 0 represents both the bottom
222 * of the address space and the top of it (using -1 for the
223 * top wouldn't help much: the masks would do the wrong thing).
224 * The rule is that addr 0 and floor 0 refer to the bottom of
225 * the address space, but end 0 and ceiling 0 refer to the top
226 * Comparisons need to use "end - 1" and "ceiling - 1" (though
227 * that end 0 case should be mythical).
228 *
229 * Wherever addr is brought up or ceiling brought down, we must
230 * be careful to reject "the opposite 0" before it confuses the
231 * subsequent tests. But what about where end is brought down
232 * by PMD_SIZE below? no, end can't go down to 0 there.
233 *
234 * Whereas we round start (addr) and ceiling down, by different
235 * masks at different levels, in order to test whether a table
236 * now has no other vmas using it, so can be freed, we don't
237 * bother to round floor or end up - the tests don't need that.
238 */
239
240 addr &= PMD_MASK;
241 if (addr < floor) {
242 addr += PMD_SIZE;
243 if (!addr)
244 return;
245 }
246 if (ceiling) {
247 ceiling &= PMD_MASK;
248 if (!ceiling)
249 return;
250 }
251 if (end - 1 > ceiling - 1)
252 end -= PMD_SIZE;
253 if (addr > end - 1)
254 return;
255
256 start = addr;
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700257 pgd = pgd_offset((*tlb)->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 do {
259 next = pgd_addr_end(addr, end);
260 if (pgd_none_or_clear_bad(pgd))
261 continue;
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700262 free_pud_range(*tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700264
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700265 if (!(*tlb)->fullmm)
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700266 flush_tlb_pgtables((*tlb)->mm, start, end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700267}
268
269void free_pgtables(struct mmu_gather **tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700270 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700271{
272 while (vma) {
273 struct vm_area_struct *next = vma->vm_next;
274 unsigned long addr = vma->vm_start;
275
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700276 /*
277 * Hide vma from rmap and vmtruncate before freeing pgtables
278 */
279 anon_vma_unlink(vma);
280 unlink_file_vma(vma);
281
David Gibson9da61ae2006-03-22 00:08:57 -0800282 if (is_vm_hugetlb_page(vma)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700283 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700284 floor, next? next->vm_start: ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700285 } else {
286 /*
287 * Optimization: gather nearby vmas into one call down
288 */
289 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
David Gibson48669202006-03-22 00:08:58 -0800290 && !is_vm_hugetlb_page(next)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700291 vma = next;
292 next = vma->vm_next;
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700293 anon_vma_unlink(vma);
294 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700295 }
296 free_pgd_range(tlb, addr, vma->vm_end,
297 floor, next? next->vm_start: ceiling);
298 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700299 vma = next;
300 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301}
302
Hugh Dickins1bb36302005-10-29 18:16:22 -0700303int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304{
Hugh Dickinsc74df322005-10-29 18:16:23 -0700305 struct page *new = pte_alloc_one(mm, address);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700306 if (!new)
307 return -ENOMEM;
308
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700309 pte_lock_init(new);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700310 spin_lock(&mm->page_table_lock);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700311 if (pmd_present(*pmd)) { /* Another has populated it */
312 pte_lock_deinit(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700313 pte_free(new);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700314 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 mm->nr_ptes++;
Christoph Lameterdf849a12006-06-30 01:55:38 -0700316 inc_zone_page_state(new, NR_PAGETABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 pmd_populate(mm, pmd, new);
318 }
Hugh Dickinsc74df322005-10-29 18:16:23 -0700319 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700320 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321}
322
Hugh Dickins1bb36302005-10-29 18:16:22 -0700323int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700325 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
326 if (!new)
327 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328
Hugh Dickins1bb36302005-10-29 18:16:22 -0700329 spin_lock(&init_mm.page_table_lock);
330 if (pmd_present(*pmd)) /* Another has populated it */
331 pte_free_kernel(new);
332 else
333 pmd_populate_kernel(&init_mm, pmd, new);
334 spin_unlock(&init_mm.page_table_lock);
335 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336}
337
Hugh Dickinsae859762005-10-29 18:16:05 -0700338static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
339{
340 if (file_rss)
341 add_mm_counter(mm, file_rss, file_rss);
342 if (anon_rss)
343 add_mm_counter(mm, anon_rss, anon_rss);
344}
345
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800347 * This function is called to print an error when a bad pte
348 * is found. For example, we might have a PFN-mapped pte in
349 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700350 *
351 * The calling function must still handle the error.
352 */
353void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr)
354{
355 printk(KERN_ERR "Bad pte = %08llx, process = %s, "
356 "vm_flags = %lx, vaddr = %lx\n",
357 (long long)pte_val(pte),
358 (vma->vm_mm == current->mm ? current->comm : "???"),
359 vma->vm_flags, vaddr);
360 dump_stack();
361}
362
Linus Torvalds67121172005-12-11 20:38:17 -0800363static inline int is_cow_mapping(unsigned int flags)
364{
365 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
366}
367
Nick Pigginb5810032005-10-29 18:16:12 -0700368/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800369 * This function gets the "struct page" associated with a pte.
370 *
371 * NOTE! Some mappings do not have "struct pages". A raw PFN mapping
372 * will have each page table entry just pointing to a raw page frame
373 * number, and as far as the VM layer is concerned, those do not have
374 * pages associated with them - even if the PFN might point to memory
375 * that otherwise is perfectly fine and has a "struct page".
376 *
377 * The way we recognize those mappings is through the rules set up
378 * by "remap_pfn_range()": the vma will have the VM_PFNMAP bit set,
379 * and the vm_pgoff will point to the first PFN mapped: thus every
380 * page that is a raw mapping will always honor the rule
381 *
382 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
383 *
384 * and if that isn't true, the page has been COW'ed (in which case it
385 * _does_ have a "struct page" associated with it even if it is in a
386 * VM_PFNMAP range).
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800387 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800388struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800389{
Linus Torvalds6aab3412005-11-28 14:34:23 -0800390 unsigned long pfn = pte_pfn(pte);
391
Nick Pigginb7ab7952006-03-22 00:08:42 -0800392 if (unlikely(vma->vm_flags & VM_PFNMAP)) {
Linus Torvalds6aab3412005-11-28 14:34:23 -0800393 unsigned long off = (addr - vma->vm_start) >> PAGE_SHIFT;
394 if (pfn == vma->vm_pgoff + off)
395 return NULL;
Linus Torvalds67121172005-12-11 20:38:17 -0800396 if (!is_cow_mapping(vma->vm_flags))
Linus Torvaldsfb155c12005-12-11 19:46:02 -0800397 return NULL;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800398 }
399
Nick Piggin315ab192006-03-25 16:20:22 +0100400 /*
401 * Add some anal sanity checks for now. Eventually,
402 * we should just do "return pfn_to_page(pfn)", but
403 * in the meantime we check that we get a valid pfn,
404 * and that the resulting page looks ok.
405 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800406 if (unlikely(!pfn_valid(pfn))) {
407 print_bad_pte(vma, pte, addr);
408 return NULL;
409 }
410
411 /*
412 * NOTE! We still have PageReserved() pages in the page
413 * tables.
414 *
415 * The PAGE_ZERO() pages and various VDSO mappings can
416 * cause them to exist.
417 */
418 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800419}
420
421/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 * copy one vm_area from one task to the other. Assumes the page tables
423 * already present in the new task to be cleared in the whole range
424 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 */
426
Hugh Dickins8c103762005-10-29 18:16:13 -0700427static inline void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700429 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700430 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431{
Nick Pigginb5810032005-10-29 18:16:12 -0700432 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 pte_t pte = *src_pte;
434 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
436 /* pte contains position in swap or file, so copy. */
437 if (unlikely(!pte_present(pte))) {
438 if (!pte_file(pte)) {
Christoph Lameter06972122006-06-23 02:03:35 -0700439 swp_entry_t entry = pte_to_swp_entry(pte);
440
441 swap_duplicate(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 /* make sure dst_mm is on swapoff's mmlist. */
443 if (unlikely(list_empty(&dst_mm->mmlist))) {
444 spin_lock(&mmlist_lock);
Hugh Dickinsf412ac02005-10-29 18:16:41 -0700445 if (list_empty(&dst_mm->mmlist))
446 list_add(&dst_mm->mmlist,
447 &src_mm->mmlist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 spin_unlock(&mmlist_lock);
449 }
Christoph Lameter06972122006-06-23 02:03:35 -0700450 if (is_write_migration_entry(entry) &&
451 is_cow_mapping(vm_flags)) {
452 /*
453 * COW mappings require pages in both parent
454 * and child to be set to read.
455 */
456 make_migration_entry_read(&entry);
457 pte = swp_entry_to_pte(entry);
458 set_pte_at(src_mm, addr, src_pte, pte);
459 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700461 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 }
463
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 /*
465 * If it's a COW mapping, write protect it both
466 * in the parent and the child
467 */
Linus Torvalds67121172005-12-11 20:38:17 -0800468 if (is_cow_mapping(vm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 ptep_set_wrprotect(src_mm, addr, src_pte);
Zachary Amsden3dc90792006-09-30 23:29:30 -0700470 pte = pte_wrprotect(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 }
472
473 /*
474 * If it's a shared mapping, mark it clean in
475 * the child
476 */
477 if (vm_flags & VM_SHARED)
478 pte = pte_mkclean(pte);
479 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800480
481 page = vm_normal_page(vma, addr, pte);
482 if (page) {
483 get_page(page);
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);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700509 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 do {
512 /*
513 * We are holding two locks at this point - either of them
514 * could generate latencies in another task on another CPU.
515 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700516 if (progress >= 32) {
517 progress = 0;
518 if (need_resched() ||
Hugh Dickinsc74df322005-10-29 18:16:23 -0700519 need_lockbreak(src_ptl) ||
520 need_lockbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700521 break;
522 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 if (pte_none(*src_pte)) {
524 progress++;
525 continue;
526 }
Hugh Dickins8c103762005-10-29 18:16:13 -0700527 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 progress += 8;
529 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700531 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -0700532 spin_unlock(src_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 pte_unmap_nested(src_pte - 1);
Hugh Dickinsae859762005-10-29 18:16:05 -0700534 add_mm_rss(dst_mm, rss[0], rss[1]);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700535 pte_unmap_unlock(dst_pte - 1, dst_ptl);
536 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 if (addr != end)
538 goto again;
539 return 0;
540}
541
542static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
543 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
544 unsigned long addr, unsigned long end)
545{
546 pmd_t *src_pmd, *dst_pmd;
547 unsigned long next;
548
549 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
550 if (!dst_pmd)
551 return -ENOMEM;
552 src_pmd = pmd_offset(src_pud, addr);
553 do {
554 next = pmd_addr_end(addr, end);
555 if (pmd_none_or_clear_bad(src_pmd))
556 continue;
557 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
558 vma, addr, next))
559 return -ENOMEM;
560 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
561 return 0;
562}
563
564static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
565 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
566 unsigned long addr, unsigned long end)
567{
568 pud_t *src_pud, *dst_pud;
569 unsigned long next;
570
571 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
572 if (!dst_pud)
573 return -ENOMEM;
574 src_pud = pud_offset(src_pgd, addr);
575 do {
576 next = pud_addr_end(addr, end);
577 if (pud_none_or_clear_bad(src_pud))
578 continue;
579 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
580 vma, addr, next))
581 return -ENOMEM;
582 } while (dst_pud++, src_pud++, addr = next, addr != end);
583 return 0;
584}
585
586int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
587 struct vm_area_struct *vma)
588{
589 pgd_t *src_pgd, *dst_pgd;
590 unsigned long next;
591 unsigned long addr = vma->vm_start;
592 unsigned long end = vma->vm_end;
593
Nick Piggind9928952005-08-28 16:49:11 +1000594 /*
595 * Don't copy ptes where a page fault will fill them correctly.
596 * Fork becomes much lighter when there are big shared or private
597 * readonly mappings. The tradeoff is that copy_page_range is more
598 * efficient than faulting.
599 */
Linus Torvalds4d7672b2005-12-16 10:21:23 -0800600 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
Nick Piggind9928952005-08-28 16:49:11 +1000601 if (!vma->anon_vma)
602 return 0;
603 }
604
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 if (is_vm_hugetlb_page(vma))
606 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
607
608 dst_pgd = pgd_offset(dst_mm, addr);
609 src_pgd = pgd_offset(src_mm, addr);
610 do {
611 next = pgd_addr_end(addr, end);
612 if (pgd_none_or_clear_bad(src_pgd))
613 continue;
614 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
615 vma, addr, next))
616 return -ENOMEM;
617 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
618 return 0;
619}
620
Robin Holt51c6f662005-11-13 16:06:42 -0800621static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700622 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800624 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625{
Nick Pigginb5810032005-10-29 18:16:12 -0700626 struct mm_struct *mm = tlb->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 pte_t *pte;
Hugh Dickins508034a2005-10-29 18:16:30 -0700628 spinlock_t *ptl;
Hugh Dickinsae859762005-10-29 18:16:05 -0700629 int file_rss = 0;
630 int anon_rss = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
Hugh Dickins508034a2005-10-29 18:16:30 -0700632 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700633 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 do {
635 pte_t ptent = *pte;
Robin Holt51c6f662005-11-13 16:06:42 -0800636 if (pte_none(ptent)) {
637 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800639 }
Hugh Dickins6f5e6b92006-03-16 23:04:09 -0800640
641 (*zap_work) -= PAGE_SIZE;
642
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800644 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -0800645
Linus Torvalds6aab3412005-11-28 14:34:23 -0800646 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 if (unlikely(details) && page) {
648 /*
649 * unmap_shared_mapping_pages() wants to
650 * invalidate cache without truncating:
651 * unmap shared but keep private pages.
652 */
653 if (details->check_mapping &&
654 details->check_mapping != page->mapping)
655 continue;
656 /*
657 * Each page->index must be checked when
658 * invalidating or truncating nonlinear.
659 */
660 if (details->nonlinear_vma &&
661 (page->index < details->first_index ||
662 page->index > details->last_index))
663 continue;
664 }
Nick Pigginb5810032005-10-29 18:16:12 -0700665 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -0700666 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 tlb_remove_tlb_entry(tlb, pte, addr);
668 if (unlikely(!page))
669 continue;
670 if (unlikely(details) && details->nonlinear_vma
671 && linear_page_index(details->nonlinear_vma,
672 addr) != page->index)
Nick Pigginb5810032005-10-29 18:16:12 -0700673 set_pte_at(mm, addr, pte,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 pgoff_to_pte(page->index));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 if (PageAnon(page))
Hugh Dickins86d912f2005-10-29 18:16:14 -0700676 anon_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700677 else {
678 if (pte_dirty(ptent))
679 set_page_dirty(page);
680 if (pte_young(ptent))
681 mark_page_accessed(page);
Hugh Dickins86d912f2005-10-29 18:16:14 -0700682 file_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700683 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 page_remove_rmap(page);
685 tlb_remove_page(tlb, page);
686 continue;
687 }
688 /*
689 * If details->check_mapping, we leave swap entries;
690 * if details->nonlinear_vma, we leave file entries.
691 */
692 if (unlikely(details))
693 continue;
694 if (!pte_file(ptent))
695 free_swap_and_cache(pte_to_swp_entry(ptent));
Zachary Amsden9888a1c2006-09-30 23:29:31 -0700696 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
Robin Holt51c6f662005-11-13 16:06:42 -0800697 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
Hugh Dickinsae859762005-10-29 18:16:05 -0700698
Hugh Dickins86d912f2005-10-29 18:16:14 -0700699 add_mm_rss(mm, file_rss, anon_rss);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700700 arch_leave_lazy_mmu_mode();
Hugh Dickins508034a2005-10-29 18:16:30 -0700701 pte_unmap_unlock(pte - 1, ptl);
Robin Holt51c6f662005-11-13 16:06:42 -0800702
703 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704}
705
Robin Holt51c6f662005-11-13 16:06:42 -0800706static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700707 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800709 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710{
711 pmd_t *pmd;
712 unsigned long next;
713
714 pmd = pmd_offset(pud, addr);
715 do {
716 next = pmd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800717 if (pmd_none_or_clear_bad(pmd)) {
718 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800720 }
721 next = zap_pte_range(tlb, vma, pmd, addr, next,
722 zap_work, details);
723 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
724
725 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726}
727
Robin Holt51c6f662005-11-13 16:06:42 -0800728static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700729 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800731 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732{
733 pud_t *pud;
734 unsigned long next;
735
736 pud = pud_offset(pgd, addr);
737 do {
738 next = pud_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800739 if (pud_none_or_clear_bad(pud)) {
740 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800742 }
743 next = zap_pmd_range(tlb, vma, pud, addr, next,
744 zap_work, details);
745 } while (pud++, addr = next, (addr != end && *zap_work > 0));
746
747 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748}
749
Robin Holt51c6f662005-11-13 16:06:42 -0800750static unsigned long unmap_page_range(struct mmu_gather *tlb,
751 struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800753 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754{
755 pgd_t *pgd;
756 unsigned long next;
757
758 if (details && !details->check_mapping && !details->nonlinear_vma)
759 details = NULL;
760
761 BUG_ON(addr >= end);
762 tlb_start_vma(tlb, vma);
763 pgd = pgd_offset(vma->vm_mm, addr);
764 do {
765 next = pgd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800766 if (pgd_none_or_clear_bad(pgd)) {
767 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800769 }
770 next = zap_pud_range(tlb, vma, pgd, addr, next,
771 zap_work, details);
772 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 tlb_end_vma(tlb, vma);
Robin Holt51c6f662005-11-13 16:06:42 -0800774
775 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776}
777
778#ifdef CONFIG_PREEMPT
779# define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
780#else
781/* No preempt: go for improved straight-line efficiency */
782# define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
783#endif
784
785/**
786 * unmap_vmas - unmap a range of memory covered by a list of vma's
787 * @tlbp: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 * @vma: the starting vma
789 * @start_addr: virtual address at which to start unmapping
790 * @end_addr: virtual address at which to end unmapping
791 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
792 * @details: details of nonlinear truncation or shared cache invalidation
793 *
Hugh Dickinsee39b372005-04-19 13:29:15 -0700794 * Returns the end address of the unmapping (restart addr if interrupted).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700796 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700798 * We aim to not hold locks for too long (for scheduling latency reasons).
799 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 * return the ending mmu_gather to the caller.
801 *
802 * Only addresses between `start' and `end' will be unmapped.
803 *
804 * The VMA list must be sorted in ascending virtual address order.
805 *
806 * unmap_vmas() assumes that the caller will flush the whole unmapped address
807 * range after unmap_vmas() returns. So the only responsibility here is to
808 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
809 * drops the lock and schedules.
810 */
Hugh Dickins508034a2005-10-29 18:16:30 -0700811unsigned long unmap_vmas(struct mmu_gather **tlbp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 struct vm_area_struct *vma, unsigned long start_addr,
813 unsigned long end_addr, unsigned long *nr_accounted,
814 struct zap_details *details)
815{
Robin Holt51c6f662005-11-13 16:06:42 -0800816 long zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
818 int tlb_start_valid = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -0700819 unsigned long start = start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700821 int fullmm = (*tlbp)->fullmm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822
823 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 unsigned long end;
825
826 start = max(vma->vm_start, start_addr);
827 if (start >= vma->vm_end)
828 continue;
829 end = min(vma->vm_end, end_addr);
830 if (end <= vma->vm_start)
831 continue;
832
833 if (vma->vm_flags & VM_ACCOUNT)
834 *nr_accounted += (end - start) >> PAGE_SHIFT;
835
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 while (start != end) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 if (!tlb_start_valid) {
838 tlb_start = start;
839 tlb_start_valid = 1;
840 }
841
Robin Holt51c6f662005-11-13 16:06:42 -0800842 if (unlikely(is_vm_hugetlb_page(vma))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 unmap_hugepage_range(vma, start, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800844 zap_work -= (end - start) /
845 (HPAGE_SIZE / PAGE_SIZE);
846 start = end;
847 } else
848 start = unmap_page_range(*tlbp, vma,
849 start, end, &zap_work, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850
Robin Holt51c6f662005-11-13 16:06:42 -0800851 if (zap_work > 0) {
852 BUG_ON(start != end);
853 break;
854 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855
856 tlb_finish_mmu(*tlbp, tlb_start, start);
857
858 if (need_resched() ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 (i_mmap_lock && need_lockbreak(i_mmap_lock))) {
860 if (i_mmap_lock) {
Hugh Dickins508034a2005-10-29 18:16:30 -0700861 *tlbp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 goto out;
863 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 }
866
Hugh Dickins508034a2005-10-29 18:16:30 -0700867 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 tlb_start_valid = 0;
Robin Holt51c6f662005-11-13 16:06:42 -0800869 zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 }
871 }
872out:
Hugh Dickinsee39b372005-04-19 13:29:15 -0700873 return start; /* which is now the end (or restart) address */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874}
875
876/**
877 * zap_page_range - remove user pages in a given range
878 * @vma: vm_area_struct holding the applicable pages
879 * @address: starting address of pages to zap
880 * @size: number of bytes to zap
881 * @details: details of nonlinear truncation or shared cache invalidation
882 */
Hugh Dickinsee39b372005-04-19 13:29:15 -0700883unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 unsigned long size, struct zap_details *details)
885{
886 struct mm_struct *mm = vma->vm_mm;
887 struct mmu_gather *tlb;
888 unsigned long end = address + size;
889 unsigned long nr_accounted = 0;
890
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -0700893 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -0700894 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
895 if (tlb)
896 tlb_finish_mmu(tlb, address, end);
Hugh Dickinsee39b372005-04-19 13:29:15 -0700897 return end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898}
899
900/*
901 * Do a quick page-table lookup for a single page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800903struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700904 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905{
906 pgd_t *pgd;
907 pud_t *pud;
908 pmd_t *pmd;
909 pte_t *ptep, pte;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700910 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 struct page *page;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800912 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700914 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
915 if (!IS_ERR(page)) {
916 BUG_ON(flags & FOLL_GET);
917 goto out;
918 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700920 page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 pgd = pgd_offset(mm, address);
922 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700923 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924
925 pud = pud_offset(pgd, address);
926 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700927 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928
929 pmd = pmd_offset(pud, address);
930 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700931 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700933 if (pmd_huge(*pmd)) {
934 BUG_ON(flags & FOLL_GET);
935 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
936 goto out;
937 }
938
939 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 if (!ptep)
941 goto out;
942
943 pte = *ptep;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700944 if (!pte_present(pte))
945 goto unlock;
946 if ((flags & FOLL_WRITE) && !pte_write(pte))
947 goto unlock;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800948 page = vm_normal_page(vma, address, pte);
949 if (unlikely(!page))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700950 goto unlock;
951
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700952 if (flags & FOLL_GET)
953 get_page(page);
954 if (flags & FOLL_TOUCH) {
955 if ((flags & FOLL_WRITE) &&
956 !pte_dirty(pte) && !PageDirty(page))
957 set_page_dirty(page);
958 mark_page_accessed(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700960unlock:
961 pte_unmap_unlock(ptep, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962out:
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700963 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700965no_page_table:
966 /*
967 * When core dumping an enormous anonymous area that nobody
968 * has touched so far, we don't want to allocate page tables.
969 */
970 if (flags & FOLL_ANON) {
971 page = ZERO_PAGE(address);
972 if (flags & FOLL_GET)
973 get_page(page);
974 BUG_ON(flags & FOLL_WRITE);
975 }
976 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977}
978
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
980 unsigned long start, int len, int write, int force,
981 struct page **pages, struct vm_area_struct **vmas)
982{
983 int i;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700984 unsigned int vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985
986 /*
987 * Require read or write permissions.
988 * If 'force' is set, we only require the "MAY" flags.
989 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700990 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
991 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 i = 0;
993
994 do {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700995 struct vm_area_struct *vma;
996 unsigned int foll_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
998 vma = find_extend_vma(mm, start);
999 if (!vma && in_gate_area(tsk, start)) {
1000 unsigned long pg = start & PAGE_MASK;
1001 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
1002 pgd_t *pgd;
1003 pud_t *pud;
1004 pmd_t *pmd;
1005 pte_t *pte;
1006 if (write) /* user gate pages are read-only */
1007 return i ? : -EFAULT;
1008 if (pg > TASK_SIZE)
1009 pgd = pgd_offset_k(pg);
1010 else
1011 pgd = pgd_offset_gate(mm, pg);
1012 BUG_ON(pgd_none(*pgd));
1013 pud = pud_offset(pgd, pg);
1014 BUG_ON(pud_none(*pud));
1015 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001016 if (pmd_none(*pmd))
1017 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001019 if (pte_none(*pte)) {
1020 pte_unmap(pte);
1021 return i ? : -EFAULT;
1022 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 if (pages) {
Nick Pigginfa2a4552005-11-29 18:43:17 +11001024 struct page *page = vm_normal_page(gate_vma, start, *pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001025 pages[i] = page;
1026 if (page)
1027 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 }
1029 pte_unmap(pte);
1030 if (vmas)
1031 vmas[i] = gate_vma;
1032 i++;
1033 start += PAGE_SIZE;
1034 len--;
1035 continue;
1036 }
1037
Linus Torvalds1ff80382005-12-12 16:24:33 -08001038 if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001039 || !(vm_flags & vma->vm_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 return i ? : -EFAULT;
1041
1042 if (is_vm_hugetlb_page(vma)) {
1043 i = follow_hugetlb_page(mm, vma, pages, vmas,
1044 &start, &len, i);
1045 continue;
1046 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001047
1048 foll_flags = FOLL_TOUCH;
1049 if (pages)
1050 foll_flags |= FOLL_GET;
1051 if (!write && !(vma->vm_flags & VM_LOCKED) &&
1052 (!vma->vm_ops || !vma->vm_ops->nopage))
1053 foll_flags |= FOLL_ANON;
1054
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 do {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001056 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001058 if (write)
1059 foll_flags |= FOLL_WRITE;
1060
1061 cond_resched();
Linus Torvalds6aab3412005-11-28 14:34:23 -08001062 while (!(page = follow_page(vma, start, foll_flags))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001063 int ret;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001064 ret = __handle_mm_fault(mm, vma, start,
1065 foll_flags & FOLL_WRITE);
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001066 /*
1067 * The VM_FAULT_WRITE bit tells us that do_wp_page has
1068 * broken COW when necessary, even if maybe_mkwrite
1069 * decided not to set pte_write. We can thus safely do
1070 * subsequent page lookups as if they were reads.
1071 */
1072 if (ret & VM_FAULT_WRITE)
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001073 foll_flags &= ~FOLL_WRITE;
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001074
1075 switch (ret & ~VM_FAULT_WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 case VM_FAULT_MINOR:
1077 tsk->min_flt++;
1078 break;
1079 case VM_FAULT_MAJOR:
1080 tsk->maj_flt++;
1081 break;
1082 case VM_FAULT_SIGBUS:
1083 return i ? i : -EFAULT;
1084 case VM_FAULT_OOM:
1085 return i ? i : -ENOMEM;
1086 default:
1087 BUG();
1088 }
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07001089 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 }
1091 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001092 pages[i] = page;
James Bottomley03beb072006-03-26 01:36:57 -08001093
1094 flush_anon_page(page, start);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001095 flush_dcache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 }
1097 if (vmas)
1098 vmas[i] = vma;
1099 i++;
1100 start += PAGE_SIZE;
1101 len--;
Hugh Dickins08ef4722005-06-21 17:15:10 -07001102 } while (len && start < vma->vm_end);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001103 } while (len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 return i;
1105}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106EXPORT_SYMBOL(get_user_pages);
1107
1108static int zeromap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1109 unsigned long addr, unsigned long end, pgprot_t prot)
1110{
1111 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001112 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113
Hugh Dickinsc74df322005-10-29 18:16:23 -07001114 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 if (!pte)
1116 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001117 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 do {
Nick Pigginb5810032005-10-29 18:16:12 -07001119 struct page *page = ZERO_PAGE(addr);
1120 pte_t zero_pte = pte_wrprotect(mk_pte(page, prot));
1121 page_cache_get(page);
1122 page_add_file_rmap(page);
1123 inc_mm_counter(mm, file_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 BUG_ON(!pte_none(*pte));
1125 set_pte_at(mm, addr, pte, zero_pte);
1126 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001127 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001128 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 return 0;
1130}
1131
1132static inline int zeromap_pmd_range(struct mm_struct *mm, pud_t *pud,
1133 unsigned long addr, unsigned long end, pgprot_t prot)
1134{
1135 pmd_t *pmd;
1136 unsigned long next;
1137
1138 pmd = pmd_alloc(mm, pud, addr);
1139 if (!pmd)
1140 return -ENOMEM;
1141 do {
1142 next = pmd_addr_end(addr, end);
1143 if (zeromap_pte_range(mm, pmd, addr, next, prot))
1144 return -ENOMEM;
1145 } while (pmd++, addr = next, addr != end);
1146 return 0;
1147}
1148
1149static inline int zeromap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1150 unsigned long addr, unsigned long end, pgprot_t prot)
1151{
1152 pud_t *pud;
1153 unsigned long next;
1154
1155 pud = pud_alloc(mm, pgd, addr);
1156 if (!pud)
1157 return -ENOMEM;
1158 do {
1159 next = pud_addr_end(addr, end);
1160 if (zeromap_pmd_range(mm, pud, addr, next, prot))
1161 return -ENOMEM;
1162 } while (pud++, addr = next, addr != end);
1163 return 0;
1164}
1165
1166int zeromap_page_range(struct vm_area_struct *vma,
1167 unsigned long addr, unsigned long size, pgprot_t prot)
1168{
1169 pgd_t *pgd;
1170 unsigned long next;
1171 unsigned long end = addr + size;
1172 struct mm_struct *mm = vma->vm_mm;
1173 int err;
1174
1175 BUG_ON(addr >= end);
1176 pgd = pgd_offset(mm, addr);
1177 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 do {
1179 next = pgd_addr_end(addr, end);
1180 err = zeromap_pud_range(mm, pgd, addr, next, prot);
1181 if (err)
1182 break;
1183 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 return err;
1185}
1186
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001187pte_t * fastcall get_locked_pte(struct mm_struct *mm, unsigned long addr, spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001188{
1189 pgd_t * pgd = pgd_offset(mm, addr);
1190 pud_t * pud = pud_alloc(mm, pgd, addr);
1191 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001192 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001193 if (pmd)
1194 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1195 }
1196 return NULL;
1197}
1198
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001200 * This is the old fallback for page remapping.
1201 *
1202 * For historical reasons, it only allows reserved pages. Only
1203 * old drivers should use this, and they needed to mark their
1204 * pages reserved for the old functions anyway.
1205 */
1206static int insert_page(struct mm_struct *mm, unsigned long addr, struct page *page, pgprot_t prot)
1207{
1208 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001209 pte_t *pte;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001210 spinlock_t *ptl;
1211
1212 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001213 if (PageAnon(page))
Linus Torvalds238f58d2005-11-29 13:01:56 -08001214 goto out;
1215 retval = -ENOMEM;
1216 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001217 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001218 if (!pte)
1219 goto out;
1220 retval = -EBUSY;
1221 if (!pte_none(*pte))
1222 goto out_unlock;
1223
1224 /* Ok, finally just insert the thing.. */
1225 get_page(page);
1226 inc_mm_counter(mm, file_rss);
1227 page_add_file_rmap(page);
1228 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1229
1230 retval = 0;
1231out_unlock:
1232 pte_unmap_unlock(pte, ptl);
1233out:
1234 return retval;
1235}
1236
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001237/**
1238 * vm_insert_page - insert single page into user vma
1239 * @vma: user vma to map to
1240 * @addr: target user address of this page
1241 * @page: source kernel page
1242 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001243 * This allows drivers to insert individual pages they've allocated
1244 * into a user vma.
1245 *
1246 * The page has to be a nice clean _individual_ kernel allocation.
1247 * If you allocate a compound page, you need to have marked it as
1248 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001249 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001250 *
1251 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1252 * took an arbitrary page protection parameter. This doesn't allow
1253 * that. Your vma protection will have to be set up correctly, which
1254 * means that if you want a shared writable mapping, you'd better
1255 * ask for a shared writable mapping!
1256 *
1257 * The page does not need to be reserved.
1258 */
1259int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, struct page *page)
1260{
1261 if (addr < vma->vm_start || addr >= vma->vm_end)
1262 return -EFAULT;
1263 if (!page_count(page))
1264 return -EINVAL;
Linus Torvalds4d7672b2005-12-16 10:21:23 -08001265 vma->vm_flags |= VM_INSERTPAGE;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001266 return insert_page(vma->vm_mm, addr, page, vma->vm_page_prot);
1267}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001268EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001269
1270/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 * maps a range of physical memory into the requested pages. the old
1272 * mappings are removed. any references to nonexistent pages results
1273 * in null mappings (currently treated as "copy-on-access")
1274 */
1275static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1276 unsigned long addr, unsigned long end,
1277 unsigned long pfn, pgprot_t prot)
1278{
1279 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001280 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281
Hugh Dickinsc74df322005-10-29 18:16:23 -07001282 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 if (!pte)
1284 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001285 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 do {
1287 BUG_ON(!pte_none(*pte));
Nick Pigginb5810032005-10-29 18:16:12 -07001288 set_pte_at(mm, addr, pte, pfn_pte(pfn, prot));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 pfn++;
1290 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001291 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001292 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 return 0;
1294}
1295
1296static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1297 unsigned long addr, unsigned long end,
1298 unsigned long pfn, pgprot_t prot)
1299{
1300 pmd_t *pmd;
1301 unsigned long next;
1302
1303 pfn -= addr >> PAGE_SHIFT;
1304 pmd = pmd_alloc(mm, pud, addr);
1305 if (!pmd)
1306 return -ENOMEM;
1307 do {
1308 next = pmd_addr_end(addr, end);
1309 if (remap_pte_range(mm, pmd, addr, next,
1310 pfn + (addr >> PAGE_SHIFT), prot))
1311 return -ENOMEM;
1312 } while (pmd++, addr = next, addr != end);
1313 return 0;
1314}
1315
1316static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1317 unsigned long addr, unsigned long end,
1318 unsigned long pfn, pgprot_t prot)
1319{
1320 pud_t *pud;
1321 unsigned long next;
1322
1323 pfn -= addr >> PAGE_SHIFT;
1324 pud = pud_alloc(mm, pgd, addr);
1325 if (!pud)
1326 return -ENOMEM;
1327 do {
1328 next = pud_addr_end(addr, end);
1329 if (remap_pmd_range(mm, pud, addr, next,
1330 pfn + (addr >> PAGE_SHIFT), prot))
1331 return -ENOMEM;
1332 } while (pud++, addr = next, addr != end);
1333 return 0;
1334}
1335
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001336/**
1337 * remap_pfn_range - remap kernel memory to userspace
1338 * @vma: user vma to map to
1339 * @addr: target user address to start at
1340 * @pfn: physical address of kernel memory
1341 * @size: size of map area
1342 * @prot: page protection flags for this mapping
1343 *
1344 * Note: this is only safe if the mm semaphore is held when called.
1345 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1347 unsigned long pfn, unsigned long size, pgprot_t prot)
1348{
1349 pgd_t *pgd;
1350 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001351 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 struct mm_struct *mm = vma->vm_mm;
1353 int err;
1354
1355 /*
1356 * Physically remapped pages are special. Tell the
1357 * rest of the world about it:
1358 * VM_IO tells people not to look at these pages
1359 * (accesses can have side effects).
Hugh Dickins0b14c172005-11-21 21:32:15 -08001360 * VM_RESERVED is specified all over the place, because
1361 * in 2.4 it kept swapout's vma scan off this vma; but
1362 * in 2.6 the LRU scan won't even find its pages, so this
1363 * flag means no more than count its pages in reserved_vm,
1364 * and omit it from core dump, even when VM_IO turned off.
Linus Torvalds6aab3412005-11-28 14:34:23 -08001365 * VM_PFNMAP tells the core MM that the base pages are just
1366 * raw PFN mappings, and do not have a "struct page" associated
1367 * with them.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001368 *
1369 * There's a horrible special case to handle copy-on-write
1370 * behaviour that some programs depend on. We mark the "original"
1371 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 */
Linus Torvalds67121172005-12-11 20:38:17 -08001373 if (is_cow_mapping(vma->vm_flags)) {
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001374 if (addr != vma->vm_start || end != vma->vm_end)
Linus Torvalds7fc7e2e2005-12-11 19:57:52 -08001375 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001376 vma->vm_pgoff = pfn;
1377 }
1378
Linus Torvalds6aab3412005-11-28 14:34:23 -08001379 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380
1381 BUG_ON(addr >= end);
1382 pfn -= addr >> PAGE_SHIFT;
1383 pgd = pgd_offset(mm, addr);
1384 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 do {
1386 next = pgd_addr_end(addr, end);
1387 err = remap_pud_range(mm, pgd, addr, next,
1388 pfn + (addr >> PAGE_SHIFT), prot);
1389 if (err)
1390 break;
1391 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 return err;
1393}
1394EXPORT_SYMBOL(remap_pfn_range);
1395
1396/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001397 * handle_pte_fault chooses page fault handler according to an entry
1398 * which was read non-atomically. Before making any commitment, on
1399 * those architectures or configurations (e.g. i386 with PAE) which
1400 * might give a mix of unmatched parts, do_swap_page and do_file_page
1401 * must check under lock before unmapping the pte and proceeding
1402 * (but do_wp_page is only called after already making such a check;
1403 * and do_anonymous_page and do_no_page can safely check later on).
1404 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001405static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001406 pte_t *page_table, pte_t orig_pte)
1407{
1408 int same = 1;
1409#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1410 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001411 spinlock_t *ptl = pte_lockptr(mm, pmd);
1412 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001413 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001414 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001415 }
1416#endif
1417 pte_unmap(page_table);
1418 return same;
1419}
1420
1421/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1423 * servicing faults for write access. In the normal case, do always want
1424 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1425 * that do not have writing enabled, when used by access_process_vm.
1426 */
1427static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1428{
1429 if (likely(vma->vm_flags & VM_WRITE))
1430 pte = pte_mkwrite(pte);
1431 return pte;
1432}
1433
Linus Torvalds6aab3412005-11-28 14:34:23 -08001434static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va)
1435{
1436 /*
1437 * If the source page was a PFN mapping, we don't have
1438 * a "struct page" for it. We do a best-effort copy by
1439 * just copying from the original user address. If that
1440 * fails, we just zero-fill it. Live with it.
1441 */
1442 if (unlikely(!src)) {
1443 void *kaddr = kmap_atomic(dst, KM_USER0);
Linus Torvalds5d2a2dbbc2005-11-29 14:07:55 -08001444 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1445
1446 /*
1447 * This really shouldn't fail, because the page is there
1448 * in the page tables. But it might just be unreadable,
1449 * in which case we just give up and fill the result with
1450 * zeroes.
1451 */
1452 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Linus Torvalds6aab3412005-11-28 14:34:23 -08001453 memset(kaddr, 0, PAGE_SIZE);
1454 kunmap_atomic(kaddr, KM_USER0);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07001455 flush_dcache_page(dst);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001456 return;
1457
1458 }
1459 copy_user_highpage(dst, src, va);
1460}
1461
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 * This routine handles present pages, when users try to write
1464 * to a shared page. It is done by copying the page to a new address
1465 * and decrementing the shared-page counter for the old page.
1466 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 * Note that this routine assumes that the protection checks have been
1468 * done by the caller (the low-level page fault routine in most cases).
1469 * Thus we can safely just mark it writable once we've done any necessary
1470 * COW.
1471 *
1472 * We also mark the page dirty at this point even though the page will
1473 * change only once the write actually happens. This avoids a few races,
1474 * and potentially makes it more efficient.
1475 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001476 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1477 * but allow concurrent faults), with pte both mapped and locked.
1478 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001480static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1481 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001482 spinlock_t *ptl, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483{
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001484 struct page *old_page, *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 pte_t entry;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001486 int reuse = 0, ret = VM_FAULT_MINOR;
1487 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488
Linus Torvalds6aab3412005-11-28 14:34:23 -08001489 old_page = vm_normal_page(vma, address, orig_pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001490 if (!old_page)
1491 goto gotten;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001493 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001494 * Take out anonymous pages first, anonymous shared vmas are
1495 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001496 */
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001497 if (PageAnon(old_page)) {
1498 if (!TestSetPageLocked(old_page)) {
1499 reuse = can_share_swap_page(old_page);
1500 unlock_page(old_page);
1501 }
1502 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001503 (VM_WRITE|VM_SHARED))) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001504 /*
1505 * Only catch write-faults on shared writable pages,
1506 * read-only shared pages can get COWed by
1507 * get_user_pages(.write=1, .force=1).
1508 */
David Howells9637a5e2006-06-23 02:03:43 -07001509 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
1510 /*
1511 * Notify the address space that the page is about to
1512 * become writable so that it can prohibit this or wait
1513 * for the page to get into an appropriate state.
1514 *
1515 * We do this without the lock held, so that it can
1516 * sleep if it needs to.
1517 */
1518 page_cache_get(old_page);
1519 pte_unmap_unlock(page_table, ptl);
1520
1521 if (vma->vm_ops->page_mkwrite(vma, old_page) < 0)
1522 goto unwritable_page;
1523
1524 page_cache_release(old_page);
1525
1526 /*
1527 * Since we dropped the lock we need to revalidate
1528 * the PTE as someone else may have changed it. If
1529 * they did, we just return, as we can count on the
1530 * MMU to tell us if they didn't also make it writable.
1531 */
1532 page_table = pte_offset_map_lock(mm, pmd, address,
1533 &ptl);
1534 if (!pte_same(*page_table, orig_pte))
1535 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001537 dirty_page = old_page;
1538 get_page(dirty_page);
David Howells9637a5e2006-06-23 02:03:43 -07001539 reuse = 1;
David Howells9637a5e2006-06-23 02:03:43 -07001540 }
1541
1542 if (reuse) {
1543 flush_cache_page(vma, address, pte_pfn(orig_pte));
1544 entry = pte_mkyoung(orig_pte);
1545 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1546 ptep_set_access_flags(vma, address, page_table, entry, 1);
1547 update_mmu_cache(vma, address, entry);
1548 lazy_mmu_prot_update(entry);
1549 ret |= VM_FAULT_WRITE;
1550 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552
1553 /*
1554 * Ok, we need to copy. Oh, well..
1555 */
Nick Pigginb5810032005-10-29 18:16:12 -07001556 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08001557gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001558 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559
1560 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07001561 goto oom;
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001562 if (old_page == ZERO_PAGE(address)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 new_page = alloc_zeroed_user_highpage(vma, address);
1564 if (!new_page)
Hugh Dickins65500d22005-10-29 18:15:59 -07001565 goto oom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 } else {
1567 new_page = alloc_page_vma(GFP_HIGHUSER, vma, address);
1568 if (!new_page)
Hugh Dickins65500d22005-10-29 18:15:59 -07001569 goto oom;
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001570 cow_user_page(new_page, old_page, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 }
Hugh Dickins65500d22005-10-29 18:15:59 -07001572
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 /*
1574 * Re-check the pte - we dropped the lock
1575 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001576 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07001577 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08001578 if (old_page) {
1579 page_remove_rmap(old_page);
1580 if (!PageAnon(old_page)) {
1581 dec_mm_counter(mm, file_rss);
1582 inc_mm_counter(mm, anon_rss);
1583 }
1584 } else
Hugh Dickins42946212005-10-29 18:16:05 -07001585 inc_mm_counter(mm, anon_rss);
Ben Collinseca35132005-11-29 11:45:26 -08001586 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07001587 entry = mk_pte(new_page, vma->vm_page_prot);
1588 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Anil Keshavamurthyc38c8db2006-07-14 00:23:57 -07001589 lazy_mmu_prot_update(entry);
Siddha, Suresh B4ce072f2006-09-29 01:58:42 -07001590 /*
1591 * Clear the pte entry and flush it first, before updating the
1592 * pte with the new entry. This will avoid a race condition
1593 * seen in the presence of one thread doing SMC and another
1594 * thread doing COW.
1595 */
1596 ptep_clear_flush(vma, address, page_table);
1597 set_pte_at(mm, address, page_table, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07001598 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 lru_cache_add_active(new_page);
Nick Piggin9617d952006-01-06 00:11:12 -08001600 page_add_new_anon_rmap(new_page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601
1602 /* Free the old page.. */
1603 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001604 ret |= VM_FAULT_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 }
Hugh Dickins920fc352005-11-21 21:32:17 -08001606 if (new_page)
1607 page_cache_release(new_page);
1608 if (old_page)
1609 page_cache_release(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001610unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001611 pte_unmap_unlock(page_table, ptl);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001612 if (dirty_page) {
Peter Zijlstraedc79b22006-09-25 23:30:58 -07001613 set_page_dirty_balance(dirty_page);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001614 put_page(dirty_page);
1615 }
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001616 return ret;
Hugh Dickins65500d22005-10-29 18:15:59 -07001617oom:
Hugh Dickins920fc352005-11-21 21:32:17 -08001618 if (old_page)
1619 page_cache_release(old_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 return VM_FAULT_OOM;
David Howells9637a5e2006-06-23 02:03:43 -07001621
1622unwritable_page:
1623 page_cache_release(old_page);
1624 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625}
1626
1627/*
1628 * Helper functions for unmap_mapping_range().
1629 *
1630 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1631 *
1632 * We have to restart searching the prio_tree whenever we drop the lock,
1633 * since the iterator is only valid while the lock is held, and anyway
1634 * a later vma might be split and reinserted earlier while lock dropped.
1635 *
1636 * The list of nonlinear vmas could be handled more efficiently, using
1637 * a placeholder, but handle it in the same way until a need is shown.
1638 * It is important to search the prio_tree before nonlinear list: a vma
1639 * may become nonlinear and be shifted from prio_tree to nonlinear list
1640 * while the lock is dropped; but never shifted from list to prio_tree.
1641 *
1642 * In order to make forward progress despite restarting the search,
1643 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1644 * quickly skip it next time around. Since the prio_tree search only
1645 * shows us those vmas affected by unmapping the range in question, we
1646 * can't efficiently keep all vmas in step with mapping->truncate_count:
1647 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1648 * mapping->truncate_count and vma->vm_truncate_count are protected by
1649 * i_mmap_lock.
1650 *
1651 * In order to make forward progress despite repeatedly restarting some
Hugh Dickinsee39b372005-04-19 13:29:15 -07001652 * large vma, note the restart_addr from unmap_vmas when it breaks out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 * and restart from that address when we reach that vma again. It might
1654 * have been split or merged, shrunk or extended, but never shifted: so
1655 * restart_addr remains valid so long as it remains in the vma's range.
1656 * unmap_mapping_range forces truncate_count to leap over page-aligned
1657 * values so we can save vma's restart_addr in its truncate_count field.
1658 */
1659#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1660
1661static void reset_vma_truncate_counts(struct address_space *mapping)
1662{
1663 struct vm_area_struct *vma;
1664 struct prio_tree_iter iter;
1665
1666 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1667 vma->vm_truncate_count = 0;
1668 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1669 vma->vm_truncate_count = 0;
1670}
1671
1672static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1673 unsigned long start_addr, unsigned long end_addr,
1674 struct zap_details *details)
1675{
1676 unsigned long restart_addr;
1677 int need_break;
1678
1679again:
1680 restart_addr = vma->vm_truncate_count;
1681 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1682 start_addr = restart_addr;
1683 if (start_addr >= end_addr) {
1684 /* Top of vma has been split off since last time */
1685 vma->vm_truncate_count = details->truncate_count;
1686 return 0;
1687 }
1688 }
1689
Hugh Dickinsee39b372005-04-19 13:29:15 -07001690 restart_addr = zap_page_range(vma, start_addr,
1691 end_addr - start_addr, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 need_break = need_resched() ||
1693 need_lockbreak(details->i_mmap_lock);
1694
Hugh Dickinsee39b372005-04-19 13:29:15 -07001695 if (restart_addr >= end_addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 /* We have now completed this vma: mark it so */
1697 vma->vm_truncate_count = details->truncate_count;
1698 if (!need_break)
1699 return 0;
1700 } else {
1701 /* Note restart_addr in vma's truncate_count field */
Hugh Dickinsee39b372005-04-19 13:29:15 -07001702 vma->vm_truncate_count = restart_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 if (!need_break)
1704 goto again;
1705 }
1706
1707 spin_unlock(details->i_mmap_lock);
1708 cond_resched();
1709 spin_lock(details->i_mmap_lock);
1710 return -EINTR;
1711}
1712
1713static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1714 struct zap_details *details)
1715{
1716 struct vm_area_struct *vma;
1717 struct prio_tree_iter iter;
1718 pgoff_t vba, vea, zba, zea;
1719
1720restart:
1721 vma_prio_tree_foreach(vma, &iter, root,
1722 details->first_index, details->last_index) {
1723 /* Skip quickly over those we have already dealt with */
1724 if (vma->vm_truncate_count == details->truncate_count)
1725 continue;
1726
1727 vba = vma->vm_pgoff;
1728 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1729 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1730 zba = details->first_index;
1731 if (zba < vba)
1732 zba = vba;
1733 zea = details->last_index;
1734 if (zea > vea)
1735 zea = vea;
1736
1737 if (unmap_mapping_range_vma(vma,
1738 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1739 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1740 details) < 0)
1741 goto restart;
1742 }
1743}
1744
1745static inline void unmap_mapping_range_list(struct list_head *head,
1746 struct zap_details *details)
1747{
1748 struct vm_area_struct *vma;
1749
1750 /*
1751 * In nonlinear VMAs there is no correspondence between virtual address
1752 * offset and file offset. So we must perform an exhaustive search
1753 * across *all* the pages in each nonlinear VMA, not just the pages
1754 * whose virtual address lies outside the file truncation point.
1755 */
1756restart:
1757 list_for_each_entry(vma, head, shared.vm_set.list) {
1758 /* Skip quickly over those we have already dealt with */
1759 if (vma->vm_truncate_count == details->truncate_count)
1760 continue;
1761 details->nonlinear_vma = vma;
1762 if (unmap_mapping_range_vma(vma, vma->vm_start,
1763 vma->vm_end, details) < 0)
1764 goto restart;
1765 }
1766}
1767
1768/**
1769 * unmap_mapping_range - unmap the portion of all mmaps
1770 * in the specified address_space corresponding to the specified
1771 * page range in the underlying file.
Martin Waitz3d410882005-06-23 22:05:21 -07001772 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 * @holebegin: byte in first page to unmap, relative to the start of
1774 * the underlying file. This will be rounded down to a PAGE_SIZE
1775 * boundary. Note that this is different from vmtruncate(), which
1776 * must keep the partial page. In contrast, we must get rid of
1777 * partial pages.
1778 * @holelen: size of prospective hole in bytes. This will be rounded
1779 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1780 * end of the file.
1781 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1782 * but 0 when invalidating pagecache, don't throw away private data.
1783 */
1784void unmap_mapping_range(struct address_space *mapping,
1785 loff_t const holebegin, loff_t const holelen, int even_cows)
1786{
1787 struct zap_details details;
1788 pgoff_t hba = holebegin >> PAGE_SHIFT;
1789 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1790
1791 /* Check for overflow. */
1792 if (sizeof(holelen) > sizeof(hlen)) {
1793 long long holeend =
1794 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1795 if (holeend & ~(long long)ULONG_MAX)
1796 hlen = ULONG_MAX - hba + 1;
1797 }
1798
1799 details.check_mapping = even_cows? NULL: mapping;
1800 details.nonlinear_vma = NULL;
1801 details.first_index = hba;
1802 details.last_index = hba + hlen - 1;
1803 if (details.last_index < details.first_index)
1804 details.last_index = ULONG_MAX;
1805 details.i_mmap_lock = &mapping->i_mmap_lock;
1806
1807 spin_lock(&mapping->i_mmap_lock);
1808
1809 /* serialize i_size write against truncate_count write */
1810 smp_wmb();
1811 /* Protect against page faults, and endless unmapping loops */
1812 mapping->truncate_count++;
1813 /*
1814 * For archs where spin_lock has inclusive semantics like ia64
1815 * this smp_mb() will prevent to read pagetable contents
1816 * before the truncate_count increment is visible to
1817 * other cpus.
1818 */
1819 smp_mb();
1820 if (unlikely(is_restart_addr(mapping->truncate_count))) {
1821 if (mapping->truncate_count == 0)
1822 reset_vma_truncate_counts(mapping);
1823 mapping->truncate_count++;
1824 }
1825 details.truncate_count = mapping->truncate_count;
1826
1827 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
1828 unmap_mapping_range_tree(&mapping->i_mmap, &details);
1829 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
1830 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
1831 spin_unlock(&mapping->i_mmap_lock);
1832}
1833EXPORT_SYMBOL(unmap_mapping_range);
1834
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001835/**
1836 * vmtruncate - unmap mappings "freed" by truncate() syscall
1837 * @inode: inode of the file used
1838 * @offset: file offset to start truncating
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 *
1840 * NOTE! We have to be ready to update the memory sharing
1841 * between the file and the memory map for a potential last
1842 * incomplete page. Ugly, but necessary.
1843 */
1844int vmtruncate(struct inode * inode, loff_t offset)
1845{
1846 struct address_space *mapping = inode->i_mapping;
1847 unsigned long limit;
1848
1849 if (inode->i_size < offset)
1850 goto do_expand;
1851 /*
1852 * truncation of in-use swapfiles is disallowed - it would cause
1853 * subsequent swapout to scribble on the now-freed blocks.
1854 */
1855 if (IS_SWAPFILE(inode))
1856 goto out_busy;
1857 i_size_write(inode, offset);
1858 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
1859 truncate_inode_pages(mapping, offset);
1860 goto out_truncate;
1861
1862do_expand:
1863 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1864 if (limit != RLIM_INFINITY && offset > limit)
1865 goto out_sig;
1866 if (offset > inode->i_sb->s_maxbytes)
1867 goto out_big;
1868 i_size_write(inode, offset);
1869
1870out_truncate:
1871 if (inode->i_op && inode->i_op->truncate)
1872 inode->i_op->truncate(inode);
1873 return 0;
1874out_sig:
1875 send_sig(SIGXFSZ, current, 0);
1876out_big:
1877 return -EFBIG;
1878out_busy:
1879 return -ETXTBSY;
1880}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881EXPORT_SYMBOL(vmtruncate);
1882
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001883int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
1884{
1885 struct address_space *mapping = inode->i_mapping;
1886
1887 /*
1888 * If the underlying filesystem is not going to provide
1889 * a way to truncate a range of blocks (punch a hole) -
1890 * we should return failure right now.
1891 */
1892 if (!inode->i_op || !inode->i_op->truncate_range)
1893 return -ENOSYS;
1894
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001895 mutex_lock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001896 down_write(&inode->i_alloc_sem);
1897 unmap_mapping_range(mapping, offset, (end - offset), 1);
1898 truncate_inode_pages_range(mapping, offset, end);
1899 inode->i_op->truncate_range(inode, offset, end);
1900 up_write(&inode->i_alloc_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001901 mutex_unlock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001902
1903 return 0;
1904}
Adrian Bunk26fc5232006-07-10 04:44:22 -07001905EXPORT_UNUSED_SYMBOL(vmtruncate_range); /* June 2006 */
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001906
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001907/**
1908 * swapin_readahead - swap in pages in hope we need them soon
1909 * @entry: swap entry of this memory
1910 * @addr: address to start
1911 * @vma: user vma this addresses belong to
1912 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 * Primitive swap readahead code. We simply read an aligned block of
1914 * (1 << page_cluster) entries in the swap area. This method is chosen
1915 * because it doesn't cost us any seek time. We also make sure to queue
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001916 * the 'original' request together with the readahead ones...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 *
1918 * This has been extended to use the NUMA policies from the mm triggering
1919 * the readahead.
1920 *
1921 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
1922 */
1923void swapin_readahead(swp_entry_t entry, unsigned long addr,struct vm_area_struct *vma)
1924{
1925#ifdef CONFIG_NUMA
1926 struct vm_area_struct *next_vma = vma ? vma->vm_next : NULL;
1927#endif
1928 int i, num;
1929 struct page *new_page;
1930 unsigned long offset;
1931
1932 /*
1933 * Get the number of handles we should do readahead io to.
1934 */
1935 num = valid_swaphandles(entry, &offset);
1936 for (i = 0; i < num; offset++, i++) {
1937 /* Ok, do the async read-ahead now */
1938 new_page = read_swap_cache_async(swp_entry(swp_type(entry),
1939 offset), vma, addr);
1940 if (!new_page)
1941 break;
1942 page_cache_release(new_page);
1943#ifdef CONFIG_NUMA
1944 /*
1945 * Find the next applicable VMA for the NUMA policy.
1946 */
1947 addr += PAGE_SIZE;
1948 if (addr == 0)
1949 vma = NULL;
1950 if (vma) {
1951 if (addr >= vma->vm_end) {
1952 vma = next_vma;
1953 next_vma = vma ? vma->vm_next : NULL;
1954 }
1955 if (vma && addr < vma->vm_start)
1956 vma = NULL;
1957 } else {
1958 if (next_vma && addr >= next_vma->vm_start) {
1959 vma = next_vma;
1960 next_vma = vma->vm_next;
1961 }
1962 }
1963#endif
1964 }
1965 lru_add_drain(); /* Push any new pages onto the LRU now */
1966}
1967
1968/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001969 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1970 * but allow concurrent faults), and pte mapped but not yet locked.
1971 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001973static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
1974 unsigned long address, pte_t *page_table, pmd_t *pmd,
1975 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001977 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 struct page *page;
Hugh Dickins65500d22005-10-29 18:15:59 -07001979 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 pte_t pte;
1981 int ret = VM_FAULT_MINOR;
1982
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001983 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001984 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07001985
1986 entry = pte_to_swp_entry(orig_pte);
Christoph Lameter06972122006-06-23 02:03:35 -07001987 if (is_migration_entry(entry)) {
1988 migration_entry_wait(mm, pmd, address);
1989 goto out;
1990 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07001991 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 page = lookup_swap_cache(entry);
1993 if (!page) {
1994 swapin_readahead(entry, address, vma);
1995 page = read_swap_cache_async(entry, vma, address);
1996 if (!page) {
1997 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001998 * Back out if somebody else faulted in this pte
1999 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002001 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 if (likely(pte_same(*page_table, orig_pte)))
2003 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002004 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002005 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006 }
2007
2008 /* Had to read the page from swap area: Major fault */
2009 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002010 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 grab_swap_token();
2012 }
2013
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002014 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 mark_page_accessed(page);
2016 lock_page(page);
2017
2018 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002019 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002021 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002022 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002023 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002024
2025 if (unlikely(!PageUptodate(page))) {
2026 ret = VM_FAULT_SIGBUS;
2027 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 }
2029
2030 /* The page isn't present yet, go ahead with the fault. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031
Hugh Dickins42946212005-10-29 18:16:05 -07002032 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 pte = mk_pte(page, vma->vm_page_prot);
2034 if (write_access && can_share_swap_page(page)) {
2035 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2036 write_access = 0;
2037 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038
2039 flush_icache_page(vma, page);
2040 set_pte_at(mm, address, page_table, pte);
2041 page_add_anon_rmap(page, vma, address);
2042
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002043 swap_free(entry);
2044 if (vm_swap_full())
2045 remove_exclusive_swap_page(page);
2046 unlock_page(page);
2047
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 if (write_access) {
2049 if (do_wp_page(mm, vma, address,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002050 page_table, pmd, ptl, pte) == VM_FAULT_OOM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 ret = VM_FAULT_OOM;
2052 goto out;
2053 }
2054
2055 /* No need to invalidate - it was non-present before */
2056 update_mmu_cache(vma, address, pte);
2057 lazy_mmu_prot_update(pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07002058unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002059 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060out:
2061 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002062out_nomap:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002063 pte_unmap_unlock(page_table, ptl);
Kirill Korotaevb8107482005-05-16 21:53:50 -07002064 unlock_page(page);
2065 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002066 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067}
2068
2069/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002070 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2071 * but allow concurrent faults), and pte mapped but not yet locked.
2072 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002074static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2075 unsigned long address, pte_t *page_table, pmd_t *pmd,
2076 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002078 struct page *page;
2079 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081
Linus Torvalds6aab3412005-11-28 14:34:23 -08002082 if (write_access) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 /* Allocate our own private page. */
2084 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085
2086 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07002087 goto oom;
2088 page = alloc_zeroed_user_highpage(vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 if (!page)
Hugh Dickins65500d22005-10-29 18:15:59 -07002090 goto oom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091
Hugh Dickins65500d22005-10-29 18:15:59 -07002092 entry = mk_pte(page, vma->vm_page_prot);
2093 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002094
2095 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2096 if (!pte_none(*page_table))
2097 goto release;
2098 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 lru_cache_add_active(page);
Nick Piggin9617d952006-01-06 00:11:12 -08002100 page_add_new_anon_rmap(page, vma, address);
Nick Pigginb5810032005-10-29 18:16:12 -07002101 } else {
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002102 /* Map the ZERO_PAGE - vm_page_prot is readonly */
2103 page = ZERO_PAGE(address);
2104 page_cache_get(page);
2105 entry = mk_pte(page, vma->vm_page_prot);
2106
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002107 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002108 spin_lock(ptl);
2109 if (!pte_none(*page_table))
2110 goto release;
Nick Pigginb5810032005-10-29 18:16:12 -07002111 inc_mm_counter(mm, file_rss);
2112 page_add_file_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 }
2114
Hugh Dickins65500d22005-10-29 18:15:59 -07002115 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116
2117 /* No need to invalidate - it was non-present before */
Hugh Dickins65500d22005-10-29 18:15:59 -07002118 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 lazy_mmu_prot_update(entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002120unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002121 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 return VM_FAULT_MINOR;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002123release:
2124 page_cache_release(page);
2125 goto unlock;
Hugh Dickins65500d22005-10-29 18:15:59 -07002126oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 return VM_FAULT_OOM;
2128}
2129
2130/*
2131 * do_no_page() tries to create a new page mapping. It aggressively
2132 * tries to share with existing pages, but makes a separate copy if
2133 * the "write_access" parameter is true in order to avoid the next
2134 * page fault.
2135 *
2136 * As this is called only for pages that do not currently exist, we
2137 * do not need to flush old virtual caches or the TLB.
2138 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002139 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2140 * but allow concurrent faults), and pte mapped but not yet locked.
2141 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002143static int do_no_page(struct mm_struct *mm, struct vm_area_struct *vma,
2144 unsigned long address, pte_t *page_table, pmd_t *pmd,
2145 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002147 spinlock_t *ptl;
Hugh Dickins65500d22005-10-29 18:15:59 -07002148 struct page *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 struct address_space *mapping = NULL;
2150 pte_t entry;
2151 unsigned int sequence = 0;
2152 int ret = VM_FAULT_MINOR;
2153 int anon = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002154 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 pte_unmap(page_table);
Hugh Dickins325f04d2005-11-29 16:55:48 +00002157 BUG_ON(vma->vm_flags & VM_PFNMAP);
2158
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 if (vma->vm_file) {
2160 mapping = vma->vm_file->f_mapping;
2161 sequence = mapping->truncate_count;
2162 smp_rmb(); /* serializes i_size against truncate_count */
2163 }
2164retry:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 new_page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret);
2166 /*
2167 * No smp_rmb is needed here as long as there's a full
2168 * spin_lock/unlock sequence inside the ->nopage callback
2169 * (for the pagecache lookup) that acts as an implicit
2170 * smp_mb() and prevents the i_size read to happen
2171 * after the next truncate_count read.
2172 */
2173
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07002174 /* no page was available -- either SIGBUS, OOM or REFAULT */
2175 if (unlikely(new_page == NOPAGE_SIGBUS))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 return VM_FAULT_SIGBUS;
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07002177 else if (unlikely(new_page == NOPAGE_OOM))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 return VM_FAULT_OOM;
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07002179 else if (unlikely(new_page == NOPAGE_REFAULT))
2180 return VM_FAULT_MINOR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181
2182 /*
2183 * Should we do an early C-O-W break?
2184 */
David Howells9637a5e2006-06-23 02:03:43 -07002185 if (write_access) {
2186 if (!(vma->vm_flags & VM_SHARED)) {
2187 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188
David Howells9637a5e2006-06-23 02:03:43 -07002189 if (unlikely(anon_vma_prepare(vma)))
2190 goto oom;
2191 page = alloc_page_vma(GFP_HIGHUSER, vma, address);
2192 if (!page)
2193 goto oom;
2194 copy_user_highpage(page, new_page, address);
2195 page_cache_release(new_page);
2196 new_page = page;
2197 anon = 1;
2198
2199 } else {
2200 /* if the page will be shareable, see if the backing
2201 * address space wants to know that the page is about
2202 * to become writable */
2203 if (vma->vm_ops->page_mkwrite &&
2204 vma->vm_ops->page_mkwrite(vma, new_page) < 0
2205 ) {
2206 page_cache_release(new_page);
2207 return VM_FAULT_SIGBUS;
2208 }
2209 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210 }
2211
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002212 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 /*
2214 * For a file-backed vma, someone could have truncated or otherwise
2215 * invalidated this page. If unmap_mapping_range got called,
2216 * retry getting the page.
2217 */
2218 if (mapping && unlikely(sequence != mapping->truncate_count)) {
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002219 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002221 cond_resched();
2222 sequence = mapping->truncate_count;
2223 smp_rmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 goto retry;
2225 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226
2227 /*
2228 * This silly early PAGE_DIRTY setting removes a race
2229 * due to the bad i386 page protection. But it's valid
2230 * for other architectures too.
2231 *
2232 * Note that if write_access is true, we either now have
2233 * an exclusive copy of the page, or this is a shared mapping,
2234 * so we can make it writable and dirty to avoid having to
2235 * handle that later.
2236 */
2237 /* Only go through if we didn't race with anybody else... */
2238 if (pte_none(*page_table)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 flush_icache_page(vma, new_page);
2240 entry = mk_pte(new_page, vma->vm_page_prot);
2241 if (write_access)
2242 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2243 set_pte_at(mm, address, page_table, entry);
2244 if (anon) {
Hugh Dickins42946212005-10-29 18:16:05 -07002245 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 lru_cache_add_active(new_page);
Nick Piggin9617d952006-01-06 00:11:12 -08002247 page_add_new_anon_rmap(new_page, vma, address);
Hugh Dickinsf57e88a2005-11-21 21:32:19 -08002248 } else {
Hugh Dickins42946212005-10-29 18:16:05 -07002249 inc_mm_counter(mm, file_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 page_add_file_rmap(new_page);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002251 if (write_access) {
2252 dirty_page = new_page;
2253 get_page(dirty_page);
2254 }
Hugh Dickins42946212005-10-29 18:16:05 -07002255 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 } else {
2257 /* One of our sibling threads was faster, back out. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002259 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 }
2261
2262 /* no need to invalidate: a not-present page shouldn't be cached */
2263 update_mmu_cache(vma, address, entry);
2264 lazy_mmu_prot_update(entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002265unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002266 pte_unmap_unlock(page_table, ptl);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002267 if (dirty_page) {
Peter Zijlstraedc79b22006-09-25 23:30:58 -07002268 set_page_dirty_balance(dirty_page);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002269 put_page(dirty_page);
2270 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271 return ret;
2272oom:
2273 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002274 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275}
2276
2277/*
Jes Sorensenf4b81802006-09-27 01:50:10 -07002278 * do_no_pfn() tries to create a new page mapping for a page without
2279 * a struct_page backing it
2280 *
2281 * As this is called only for pages that do not currently exist, we
2282 * do not need to flush old virtual caches or the TLB.
2283 *
2284 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2285 * but allow concurrent faults), and pte mapped but not yet locked.
2286 * We return with mmap_sem still held, but pte unmapped and unlocked.
2287 *
2288 * It is expected that the ->nopfn handler always returns the same pfn
2289 * for a given virtual mapping.
2290 *
2291 * Mark this `noinline' to prevent it from bloating the main pagefault code.
2292 */
2293static noinline int do_no_pfn(struct mm_struct *mm, struct vm_area_struct *vma,
2294 unsigned long address, pte_t *page_table, pmd_t *pmd,
2295 int write_access)
2296{
2297 spinlock_t *ptl;
2298 pte_t entry;
2299 unsigned long pfn;
2300 int ret = VM_FAULT_MINOR;
2301
2302 pte_unmap(page_table);
2303 BUG_ON(!(vma->vm_flags & VM_PFNMAP));
2304 BUG_ON(is_cow_mapping(vma->vm_flags));
2305
2306 pfn = vma->vm_ops->nopfn(vma, address & PAGE_MASK);
2307 if (pfn == NOPFN_OOM)
2308 return VM_FAULT_OOM;
2309 if (pfn == NOPFN_SIGBUS)
2310 return VM_FAULT_SIGBUS;
2311
2312 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2313
2314 /* Only go through if we didn't race with anybody else... */
2315 if (pte_none(*page_table)) {
2316 entry = pfn_pte(pfn, vma->vm_page_prot);
2317 if (write_access)
2318 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2319 set_pte_at(mm, address, page_table, entry);
2320 }
2321 pte_unmap_unlock(page_table, ptl);
2322 return ret;
2323}
2324
2325/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 * Fault of a previously existing named mapping. Repopulate the pte
2327 * from the encoded file_pte if possible. This enables swappable
2328 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002329 *
2330 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2331 * but allow concurrent faults), and pte mapped but not yet locked.
2332 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002334static int do_file_page(struct mm_struct *mm, struct vm_area_struct *vma,
2335 unsigned long address, pte_t *page_table, pmd_t *pmd,
2336 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337{
Hugh Dickins65500d22005-10-29 18:15:59 -07002338 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 int err;
2340
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002341 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002342 return VM_FAULT_MINOR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343
Hugh Dickins65500d22005-10-29 18:15:59 -07002344 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
2345 /*
2346 * Page table corrupted: show pte and kill process.
2347 */
Nick Pigginb5810032005-10-29 18:16:12 -07002348 print_bad_pte(vma, orig_pte, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002349 return VM_FAULT_OOM;
2350 }
2351 /* We can then assume vm->vm_ops && vma->vm_ops->populate */
2352
2353 pgoff = pte_to_pgoff(orig_pte);
2354 err = vma->vm_ops->populate(vma, address & PAGE_MASK, PAGE_SIZE,
2355 vma->vm_page_prot, pgoff, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356 if (err == -ENOMEM)
2357 return VM_FAULT_OOM;
2358 if (err)
2359 return VM_FAULT_SIGBUS;
2360 return VM_FAULT_MAJOR;
2361}
2362
2363/*
2364 * These routines also need to handle stuff like marking pages dirty
2365 * and/or accessed for architectures that don't do it in hardware (most
2366 * RISC architectures). The early dirtying is also good on the i386.
2367 *
2368 * There is also a hook called "update_mmu_cache()" that architectures
2369 * with external mmu caches can use to update those (ie the Sparc or
2370 * PowerPC hashed page tables that act as extended TLBs).
2371 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07002372 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2373 * but allow concurrent faults), and pte mapped but not yet locked.
2374 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 */
2376static inline int handle_pte_fault(struct mm_struct *mm,
Hugh Dickins65500d22005-10-29 18:15:59 -07002377 struct vm_area_struct *vma, unsigned long address,
2378 pte_t *pte, pmd_t *pmd, int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379{
2380 pte_t entry;
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002381 pte_t old_entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002382 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002384 old_entry = entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002386 if (pte_none(entry)) {
Jes Sorensenf4b81802006-09-27 01:50:10 -07002387 if (vma->vm_ops) {
2388 if (vma->vm_ops->nopage)
2389 return do_no_page(mm, vma, address,
2390 pte, pmd,
2391 write_access);
2392 if (unlikely(vma->vm_ops->nopfn))
2393 return do_no_pfn(mm, vma, address, pte,
2394 pmd, write_access);
2395 }
2396 return do_anonymous_page(mm, vma, address,
2397 pte, pmd, write_access);
Hugh Dickins65500d22005-10-29 18:15:59 -07002398 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 if (pte_file(entry))
Hugh Dickins65500d22005-10-29 18:15:59 -07002400 return do_file_page(mm, vma, address,
2401 pte, pmd, write_access, entry);
2402 return do_swap_page(mm, vma, address,
2403 pte, pmd, write_access, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 }
2405
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002406 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002407 spin_lock(ptl);
2408 if (unlikely(!pte_same(*pte, entry)))
2409 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 if (write_access) {
2411 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002412 return do_wp_page(mm, vma, address,
2413 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414 entry = pte_mkdirty(entry);
2415 }
2416 entry = pte_mkyoung(entry);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002417 if (!pte_same(old_entry, entry)) {
2418 ptep_set_access_flags(vma, address, pte, entry, write_access);
2419 update_mmu_cache(vma, address, entry);
2420 lazy_mmu_prot_update(entry);
2421 } else {
2422 /*
2423 * This is needed only for protection faults but the arch code
2424 * is not yet telling us if this is a protection fault or not.
2425 * This still avoids useless tlb flushes for .text page faults
2426 * with threads.
2427 */
2428 if (write_access)
2429 flush_tlb_page(vma, address);
2430 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002431unlock:
2432 pte_unmap_unlock(pte, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433 return VM_FAULT_MINOR;
2434}
2435
2436/*
2437 * By the time we get here, we already hold the mm semaphore
2438 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002439int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 unsigned long address, int write_access)
2441{
2442 pgd_t *pgd;
2443 pud_t *pud;
2444 pmd_t *pmd;
2445 pte_t *pte;
2446
2447 __set_current_state(TASK_RUNNING);
2448
Christoph Lameterf8891e52006-06-30 01:55:45 -07002449 count_vm_event(PGFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01002451 if (unlikely(is_vm_hugetlb_page(vma)))
2452 return hugetlb_fault(mm, vma, address, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455 pud = pud_alloc(mm, pgd, address);
2456 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002457 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458 pmd = pmd_alloc(mm, pud, address);
2459 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002460 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 pte = pte_alloc_map(mm, pmd, address);
2462 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002463 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464
Hugh Dickinsc74df322005-10-29 18:16:23 -07002465 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466}
2467
Arnd Bergmann67207b92005-11-15 15:53:48 -05002468EXPORT_SYMBOL_GPL(__handle_mm_fault);
2469
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470#ifndef __PAGETABLE_PUD_FOLDED
2471/*
2472 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002473 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002475int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002477 pud_t *new = pud_alloc_one(mm, address);
2478 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002479 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480
Hugh Dickins872fec12005-10-29 18:16:21 -07002481 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002482 if (pgd_present(*pgd)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 pud_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002484 else
2485 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07002486 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002487 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488}
Alan Sterne0f39592005-11-28 13:43:44 -08002489#else
2490/* Workaround for gcc 2.96 */
2491int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
2492{
2493 return 0;
2494}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495#endif /* __PAGETABLE_PUD_FOLDED */
2496
2497#ifndef __PAGETABLE_PMD_FOLDED
2498/*
2499 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002500 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002502int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002504 pmd_t *new = pmd_alloc_one(mm, address);
2505 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002506 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507
Hugh Dickins872fec12005-10-29 18:16:21 -07002508 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07002510 if (pud_present(*pud)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 pmd_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002512 else
2513 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07002515 if (pgd_present(*pud)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516 pmd_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002517 else
2518 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07002520 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002521 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522}
Alan Sterne0f39592005-11-28 13:43:44 -08002523#else
2524/* Workaround for gcc 2.96 */
2525int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
2526{
2527 return 0;
2528}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529#endif /* __PAGETABLE_PMD_FOLDED */
2530
2531int make_pages_present(unsigned long addr, unsigned long end)
2532{
2533 int ret, len, write;
2534 struct vm_area_struct * vma;
2535
2536 vma = find_vma(current->mm, addr);
2537 if (!vma)
2538 return -1;
2539 write = (vma->vm_flags & VM_WRITE) != 0;
Eric Sesterhenn5bcb28b2006-03-26 18:30:52 +02002540 BUG_ON(addr >= end);
2541 BUG_ON(end > vma->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 len = (end+PAGE_SIZE-1)/PAGE_SIZE-addr/PAGE_SIZE;
2543 ret = get_user_pages(current, current->mm, addr,
2544 len, write, 0, NULL, NULL);
2545 if (ret < 0)
2546 return ret;
2547 return ret == len ? 0 : -1;
2548}
2549
2550/*
2551 * Map a vmalloc()-space virtual address to the physical page.
2552 */
2553struct page * vmalloc_to_page(void * vmalloc_addr)
2554{
2555 unsigned long addr = (unsigned long) vmalloc_addr;
2556 struct page *page = NULL;
2557 pgd_t *pgd = pgd_offset_k(addr);
2558 pud_t *pud;
2559 pmd_t *pmd;
2560 pte_t *ptep, pte;
2561
2562 if (!pgd_none(*pgd)) {
2563 pud = pud_offset(pgd, addr);
2564 if (!pud_none(*pud)) {
2565 pmd = pmd_offset(pud, addr);
2566 if (!pmd_none(*pmd)) {
2567 ptep = pte_offset_map(pmd, addr);
2568 pte = *ptep;
2569 if (pte_present(pte))
2570 page = pte_page(pte);
2571 pte_unmap(ptep);
2572 }
2573 }
2574 }
2575 return page;
2576}
2577
2578EXPORT_SYMBOL(vmalloc_to_page);
2579
2580/*
2581 * Map a vmalloc()-space virtual address to the physical page frame number.
2582 */
2583unsigned long vmalloc_to_pfn(void * vmalloc_addr)
2584{
2585 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
2586}
2587
2588EXPORT_SYMBOL(vmalloc_to_pfn);
2589
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590#if !defined(__HAVE_ARCH_GATE_AREA)
2591
2592#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07002593static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594
2595static int __init gate_vma_init(void)
2596{
2597 gate_vma.vm_mm = NULL;
2598 gate_vma.vm_start = FIXADDR_USER_START;
2599 gate_vma.vm_end = FIXADDR_USER_END;
2600 gate_vma.vm_page_prot = PAGE_READONLY;
Hugh Dickins0b14c172005-11-21 21:32:15 -08002601 gate_vma.vm_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602 return 0;
2603}
2604__initcall(gate_vma_init);
2605#endif
2606
2607struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2608{
2609#ifdef AT_SYSINFO_EHDR
2610 return &gate_vma;
2611#else
2612 return NULL;
2613#endif
2614}
2615
2616int in_gate_area_no_task(unsigned long addr)
2617{
2618#ifdef AT_SYSINFO_EHDR
2619 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2620 return 1;
2621#endif
2622 return 0;
2623}
2624
2625#endif /* __HAVE_ARCH_GATE_AREA */
David Howells0ec76a12006-09-27 01:50:15 -07002626
2627/*
2628 * Access another process' address space.
2629 * Source/target buffer must be kernel space,
2630 * Do not walk the page table directly, use get_user_pages
2631 */
2632int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2633{
2634 struct mm_struct *mm;
2635 struct vm_area_struct *vma;
2636 struct page *page;
2637 void *old_buf = buf;
2638
2639 mm = get_task_mm(tsk);
2640 if (!mm)
2641 return 0;
2642
2643 down_read(&mm->mmap_sem);
2644 /* ignore errors, just check how much was sucessfully transfered */
2645 while (len) {
2646 int bytes, ret, offset;
2647 void *maddr;
2648
2649 ret = get_user_pages(tsk, mm, addr, 1,
2650 write, 1, &page, &vma);
2651 if (ret <= 0)
2652 break;
2653
2654 bytes = len;
2655 offset = addr & (PAGE_SIZE-1);
2656 if (bytes > PAGE_SIZE-offset)
2657 bytes = PAGE_SIZE-offset;
2658
2659 maddr = kmap(page);
2660 if (write) {
2661 copy_to_user_page(vma, page, addr,
2662 maddr + offset, buf, bytes);
2663 set_page_dirty_lock(page);
2664 } else {
2665 copy_from_user_page(vma, page, addr,
2666 buf, maddr + offset, bytes);
2667 }
2668 kunmap(page);
2669 page_cache_release(page);
2670 len -= bytes;
2671 buf += bytes;
2672 addr += bytes;
2673 }
2674 up_read(&mm->mmap_sem);
2675 mmput(mm);
2676
2677 return buf - old_buf;
2678}