blob: 153a54b2013ca9927edf1dda4e047a82a952c05e [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>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080053#include <linux/memcontrol.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
55#include <asm/pgalloc.h>
56#include <asm/uaccess.h>
57#include <asm/tlb.h>
58#include <asm/tlbflush.h>
59#include <asm/pgtable.h>
60
61#include <linux/swapops.h>
62#include <linux/elf.h>
63
Andy Whitcroftd41dee32005-06-23 00:07:54 -070064#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -070065/* use the per-pgdat data instead for discontigmem - mbligh */
66unsigned long max_mapnr;
67struct page *mem_map;
68
69EXPORT_SYMBOL(max_mapnr);
70EXPORT_SYMBOL(mem_map);
71#endif
72
73unsigned long num_physpages;
74/*
75 * A number of key systems in x86 including ioremap() rely on the assumption
76 * that high_memory defines the upper bound on direct map memory, then end
77 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
78 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
79 * and ZONE_HIGHMEM.
80 */
81void * high_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
83EXPORT_SYMBOL(num_physpages);
84EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
Ingo Molnar32a93232008-02-06 22:39:44 +010086/*
87 * Randomize the address space (stacks, mmaps, brk, etc.).
88 *
89 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
90 * as ancient (libc5 based) binaries can segfault. )
91 */
92int randomize_va_space __read_mostly =
93#ifdef CONFIG_COMPAT_BRK
94 1;
95#else
96 2;
97#endif
Andi Kleena62eaf12006-02-16 23:41:58 +010098
99static int __init disable_randmaps(char *s)
100{
101 randomize_va_space = 0;
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800102 return 1;
Andi Kleena62eaf12006-02-16 23:41:58 +0100103}
104__setup("norandmaps", disable_randmaps);
105
106
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107/*
108 * If a p?d_bad entry is found while walking page tables, report
109 * the error, before resetting entry to p?d_none. Usually (but
110 * very seldom) called out from the p?d_none_or_clear_bad macros.
111 */
112
113void pgd_clear_bad(pgd_t *pgd)
114{
115 pgd_ERROR(*pgd);
116 pgd_clear(pgd);
117}
118
119void pud_clear_bad(pud_t *pud)
120{
121 pud_ERROR(*pud);
122 pud_clear(pud);
123}
124
125void pmd_clear_bad(pmd_t *pmd)
126{
127 pmd_ERROR(*pmd);
128 pmd_clear(pmd);
129}
130
131/*
132 * Note: this doesn't free the actual pages themselves. That
133 * has been handled earlier when unmapping all the memory regions.
134 */
Hugh Dickinse0da3822005-04-19 13:29:15 -0700135static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136{
Hugh Dickinse0da3822005-04-19 13:29:15 -0700137 struct page *page = pmd_page(*pmd);
138 pmd_clear(pmd);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700139 pte_lock_deinit(page);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700140 pte_free_tlb(tlb, page);
Christoph Lameterdf849a12006-06-30 01:55:38 -0700141 dec_zone_page_state(page, NR_PAGETABLE);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700142 tlb->mm->nr_ptes--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143}
144
Hugh Dickinse0da3822005-04-19 13:29:15 -0700145static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
146 unsigned long addr, unsigned long end,
147 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148{
149 pmd_t *pmd;
150 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700151 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
Hugh Dickinse0da3822005-04-19 13:29:15 -0700153 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 do {
156 next = pmd_addr_end(addr, end);
157 if (pmd_none_or_clear_bad(pmd))
158 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700159 free_pte_range(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 } while (pmd++, addr = next, addr != end);
161
Hugh Dickinse0da3822005-04-19 13:29:15 -0700162 start &= PUD_MASK;
163 if (start < floor)
164 return;
165 if (ceiling) {
166 ceiling &= PUD_MASK;
167 if (!ceiling)
168 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700170 if (end - 1 > ceiling - 1)
171 return;
172
173 pmd = pmd_offset(pud, start);
174 pud_clear(pud);
175 pmd_free_tlb(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176}
177
Hugh Dickinse0da3822005-04-19 13:29:15 -0700178static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
179 unsigned long addr, unsigned long end,
180 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181{
182 pud_t *pud;
183 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700184 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
Hugh Dickinse0da3822005-04-19 13:29:15 -0700186 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 pud = pud_offset(pgd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 do {
189 next = pud_addr_end(addr, end);
190 if (pud_none_or_clear_bad(pud))
191 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700192 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 } while (pud++, addr = next, addr != end);
194
Hugh Dickinse0da3822005-04-19 13:29:15 -0700195 start &= PGDIR_MASK;
196 if (start < floor)
197 return;
198 if (ceiling) {
199 ceiling &= PGDIR_MASK;
200 if (!ceiling)
201 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700203 if (end - 1 > ceiling - 1)
204 return;
205
206 pud = pud_offset(pgd, start);
207 pgd_clear(pgd);
208 pud_free_tlb(tlb, pud);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209}
210
211/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700212 * This function frees user-level page tables of a process.
213 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 * Must be called with pagetable lock held.
215 */
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700216void free_pgd_range(struct mmu_gather **tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700217 unsigned long addr, unsigned long end,
218 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219{
220 pgd_t *pgd;
221 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700222 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223
Hugh Dickinse0da3822005-04-19 13:29:15 -0700224 /*
225 * The next few lines have given us lots of grief...
226 *
227 * Why are we testing PMD* at this top level? Because often
228 * there will be no work to do at all, and we'd prefer not to
229 * go all the way down to the bottom just to discover that.
230 *
231 * Why all these "- 1"s? Because 0 represents both the bottom
232 * of the address space and the top of it (using -1 for the
233 * top wouldn't help much: the masks would do the wrong thing).
234 * The rule is that addr 0 and floor 0 refer to the bottom of
235 * the address space, but end 0 and ceiling 0 refer to the top
236 * Comparisons need to use "end - 1" and "ceiling - 1" (though
237 * that end 0 case should be mythical).
238 *
239 * Wherever addr is brought up or ceiling brought down, we must
240 * be careful to reject "the opposite 0" before it confuses the
241 * subsequent tests. But what about where end is brought down
242 * by PMD_SIZE below? no, end can't go down to 0 there.
243 *
244 * Whereas we round start (addr) and ceiling down, by different
245 * masks at different levels, in order to test whether a table
246 * now has no other vmas using it, so can be freed, we don't
247 * bother to round floor or end up - the tests don't need that.
248 */
249
250 addr &= PMD_MASK;
251 if (addr < floor) {
252 addr += PMD_SIZE;
253 if (!addr)
254 return;
255 }
256 if (ceiling) {
257 ceiling &= PMD_MASK;
258 if (!ceiling)
259 return;
260 }
261 if (end - 1 > ceiling - 1)
262 end -= PMD_SIZE;
263 if (addr > end - 1)
264 return;
265
266 start = addr;
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700267 pgd = pgd_offset((*tlb)->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 do {
269 next = pgd_addr_end(addr, end);
270 if (pgd_none_or_clear_bad(pgd))
271 continue;
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700272 free_pud_range(*tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700274}
275
276void free_pgtables(struct mmu_gather **tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700277 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700278{
279 while (vma) {
280 struct vm_area_struct *next = vma->vm_next;
281 unsigned long addr = vma->vm_start;
282
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700283 /*
284 * Hide vma from rmap and vmtruncate before freeing pgtables
285 */
286 anon_vma_unlink(vma);
287 unlink_file_vma(vma);
288
David Gibson9da61ae2006-03-22 00:08:57 -0800289 if (is_vm_hugetlb_page(vma)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700290 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700291 floor, next? next->vm_start: ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700292 } else {
293 /*
294 * Optimization: gather nearby vmas into one call down
295 */
296 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
David Gibson48669202006-03-22 00:08:58 -0800297 && !is_vm_hugetlb_page(next)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700298 vma = next;
299 next = vma->vm_next;
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700300 anon_vma_unlink(vma);
301 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700302 }
303 free_pgd_range(tlb, addr, vma->vm_end,
304 floor, next? next->vm_start: ceiling);
305 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700306 vma = next;
307 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308}
309
Hugh Dickins1bb36302005-10-29 18:16:22 -0700310int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311{
Hugh Dickinsc74df322005-10-29 18:16:23 -0700312 struct page *new = pte_alloc_one(mm, address);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700313 if (!new)
314 return -ENOMEM;
315
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700316 pte_lock_init(new);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700317 spin_lock(&mm->page_table_lock);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700318 if (pmd_present(*pmd)) { /* Another has populated it */
319 pte_lock_deinit(new);
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -0800320 pte_free(mm, new);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700321 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 mm->nr_ptes++;
Christoph Lameterdf849a12006-06-30 01:55:38 -0700323 inc_zone_page_state(new, NR_PAGETABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 pmd_populate(mm, pmd, new);
325 }
Hugh Dickinsc74df322005-10-29 18:16:23 -0700326 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700327 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328}
329
Hugh Dickins1bb36302005-10-29 18:16:22 -0700330int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700332 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
333 if (!new)
334 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335
Hugh Dickins1bb36302005-10-29 18:16:22 -0700336 spin_lock(&init_mm.page_table_lock);
337 if (pmd_present(*pmd)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -0800338 pte_free_kernel(&init_mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700339 else
340 pmd_populate_kernel(&init_mm, pmd, new);
341 spin_unlock(&init_mm.page_table_lock);
342 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343}
344
Hugh Dickinsae859762005-10-29 18:16:05 -0700345static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
346{
347 if (file_rss)
348 add_mm_counter(mm, file_rss, file_rss);
349 if (anon_rss)
350 add_mm_counter(mm, anon_rss, anon_rss);
351}
352
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800354 * This function is called to print an error when a bad pte
355 * is found. For example, we might have a PFN-mapped pte in
356 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700357 *
358 * The calling function must still handle the error.
359 */
360void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr)
361{
362 printk(KERN_ERR "Bad pte = %08llx, process = %s, "
363 "vm_flags = %lx, vaddr = %lx\n",
364 (long long)pte_val(pte),
365 (vma->vm_mm == current->mm ? current->comm : "???"),
366 vma->vm_flags, vaddr);
367 dump_stack();
368}
369
Linus Torvalds67121172005-12-11 20:38:17 -0800370static inline int is_cow_mapping(unsigned int flags)
371{
372 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
373}
374
Nick Pigginb5810032005-10-29 18:16:12 -0700375/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800376 * This function gets the "struct page" associated with a pte.
377 *
378 * NOTE! Some mappings do not have "struct pages". A raw PFN mapping
379 * will have each page table entry just pointing to a raw page frame
380 * number, and as far as the VM layer is concerned, those do not have
381 * pages associated with them - even if the PFN might point to memory
382 * that otherwise is perfectly fine and has a "struct page".
383 *
384 * The way we recognize those mappings is through the rules set up
385 * by "remap_pfn_range()": the vma will have the VM_PFNMAP bit set,
386 * and the vm_pgoff will point to the first PFN mapped: thus every
387 * page that is a raw mapping will always honor the rule
388 *
389 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
390 *
391 * and if that isn't true, the page has been COW'ed (in which case it
392 * _does_ have a "struct page" associated with it even if it is in a
393 * VM_PFNMAP range).
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800394 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800395struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800396{
Linus Torvalds6aab3412005-11-28 14:34:23 -0800397 unsigned long pfn = pte_pfn(pte);
398
Nick Pigginb7ab7952006-03-22 00:08:42 -0800399 if (unlikely(vma->vm_flags & VM_PFNMAP)) {
Linus Torvalds6aab3412005-11-28 14:34:23 -0800400 unsigned long off = (addr - vma->vm_start) >> PAGE_SHIFT;
401 if (pfn == vma->vm_pgoff + off)
402 return NULL;
Linus Torvalds67121172005-12-11 20:38:17 -0800403 if (!is_cow_mapping(vma->vm_flags))
Linus Torvaldsfb155c12005-12-11 19:46:02 -0800404 return NULL;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800405 }
406
Carsten Otte97231982008-01-17 15:21:17 -0800407#ifdef CONFIG_DEBUG_VM
Nick Piggin315ab192006-03-25 16:20:22 +0100408 /*
409 * Add some anal sanity checks for now. Eventually,
410 * we should just do "return pfn_to_page(pfn)", but
411 * in the meantime we check that we get a valid pfn,
412 * and that the resulting page looks ok.
413 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800414 if (unlikely(!pfn_valid(pfn))) {
415 print_bad_pte(vma, pte, addr);
416 return NULL;
417 }
Carsten Otte97231982008-01-17 15:21:17 -0800418#endif
Linus Torvalds6aab3412005-11-28 14:34:23 -0800419
420 /*
421 * NOTE! We still have PageReserved() pages in the page
422 * tables.
423 *
424 * The PAGE_ZERO() pages and various VDSO mappings can
425 * cause them to exist.
426 */
427 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800428}
429
430/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 * copy one vm_area from one task to the other. Assumes the page tables
432 * already present in the new task to be cleared in the whole range
433 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 */
435
Hugh Dickins8c103762005-10-29 18:16:13 -0700436static inline void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700438 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700439 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440{
Nick Pigginb5810032005-10-29 18:16:12 -0700441 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 pte_t pte = *src_pte;
443 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
445 /* pte contains position in swap or file, so copy. */
446 if (unlikely(!pte_present(pte))) {
447 if (!pte_file(pte)) {
Christoph Lameter06972122006-06-23 02:03:35 -0700448 swp_entry_t entry = pte_to_swp_entry(pte);
449
450 swap_duplicate(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 /* make sure dst_mm is on swapoff's mmlist. */
452 if (unlikely(list_empty(&dst_mm->mmlist))) {
453 spin_lock(&mmlist_lock);
Hugh Dickinsf412ac02005-10-29 18:16:41 -0700454 if (list_empty(&dst_mm->mmlist))
455 list_add(&dst_mm->mmlist,
456 &src_mm->mmlist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 spin_unlock(&mmlist_lock);
458 }
Christoph Lameter06972122006-06-23 02:03:35 -0700459 if (is_write_migration_entry(entry) &&
460 is_cow_mapping(vm_flags)) {
461 /*
462 * COW mappings require pages in both parent
463 * and child to be set to read.
464 */
465 make_migration_entry_read(&entry);
466 pte = swp_entry_to_pte(entry);
467 set_pte_at(src_mm, addr, src_pte, pte);
468 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700470 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 }
472
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 /*
474 * If it's a COW mapping, write protect it both
475 * in the parent and the child
476 */
Linus Torvalds67121172005-12-11 20:38:17 -0800477 if (is_cow_mapping(vm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 ptep_set_wrprotect(src_mm, addr, src_pte);
Zachary Amsden3dc90792006-09-30 23:29:30 -0700479 pte = pte_wrprotect(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 }
481
482 /*
483 * If it's a shared mapping, mark it clean in
484 * the child
485 */
486 if (vm_flags & VM_SHARED)
487 pte = pte_mkclean(pte);
488 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800489
490 page = vm_normal_page(vma, addr, pte);
491 if (page) {
492 get_page(page);
Nick Pigginc97a9e12007-05-16 22:11:21 -0700493 page_dup_rmap(page, vma, addr);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800494 rss[!!PageAnon(page)]++;
495 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700496
497out_set_pte:
498 set_pte_at(dst_mm, addr, dst_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499}
500
501static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
502 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
503 unsigned long addr, unsigned long end)
504{
505 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700506 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -0700507 int progress = 0;
Hugh Dickins8c103762005-10-29 18:16:13 -0700508 int rss[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
510again:
Hugh Dickinsae859762005-10-29 18:16:05 -0700511 rss[1] = rss[0] = 0;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700512 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 if (!dst_pte)
514 return -ENOMEM;
515 src_pte = pte_offset_map_nested(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700516 src_ptl = pte_lockptr(src_mm, src_pmd);
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700517 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700518 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 do {
521 /*
522 * We are holding two locks at this point - either of them
523 * could generate latencies in another task on another CPU.
524 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700525 if (progress >= 32) {
526 progress = 0;
527 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +0100528 spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700529 break;
530 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 if (pte_none(*src_pte)) {
532 progress++;
533 continue;
534 }
Hugh Dickins8c103762005-10-29 18:16:13 -0700535 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 progress += 8;
537 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700539 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -0700540 spin_unlock(src_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 pte_unmap_nested(src_pte - 1);
Hugh Dickinsae859762005-10-29 18:16:05 -0700542 add_mm_rss(dst_mm, rss[0], rss[1]);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700543 pte_unmap_unlock(dst_pte - 1, dst_ptl);
544 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 if (addr != end)
546 goto again;
547 return 0;
548}
549
550static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
551 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
552 unsigned long addr, unsigned long end)
553{
554 pmd_t *src_pmd, *dst_pmd;
555 unsigned long next;
556
557 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
558 if (!dst_pmd)
559 return -ENOMEM;
560 src_pmd = pmd_offset(src_pud, addr);
561 do {
562 next = pmd_addr_end(addr, end);
563 if (pmd_none_or_clear_bad(src_pmd))
564 continue;
565 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
566 vma, addr, next))
567 return -ENOMEM;
568 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
569 return 0;
570}
571
572static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
573 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
574 unsigned long addr, unsigned long end)
575{
576 pud_t *src_pud, *dst_pud;
577 unsigned long next;
578
579 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
580 if (!dst_pud)
581 return -ENOMEM;
582 src_pud = pud_offset(src_pgd, addr);
583 do {
584 next = pud_addr_end(addr, end);
585 if (pud_none_or_clear_bad(src_pud))
586 continue;
587 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
588 vma, addr, next))
589 return -ENOMEM;
590 } while (dst_pud++, src_pud++, addr = next, addr != end);
591 return 0;
592}
593
594int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
595 struct vm_area_struct *vma)
596{
597 pgd_t *src_pgd, *dst_pgd;
598 unsigned long next;
599 unsigned long addr = vma->vm_start;
600 unsigned long end = vma->vm_end;
601
Nick Piggind9928952005-08-28 16:49:11 +1000602 /*
603 * Don't copy ptes where a page fault will fill them correctly.
604 * Fork becomes much lighter when there are big shared or private
605 * readonly mappings. The tradeoff is that copy_page_range is more
606 * efficient than faulting.
607 */
Linus Torvalds4d7672b2005-12-16 10:21:23 -0800608 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
Nick Piggind9928952005-08-28 16:49:11 +1000609 if (!vma->anon_vma)
610 return 0;
611 }
612
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 if (is_vm_hugetlb_page(vma))
614 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
615
616 dst_pgd = pgd_offset(dst_mm, addr);
617 src_pgd = pgd_offset(src_mm, addr);
618 do {
619 next = pgd_addr_end(addr, end);
620 if (pgd_none_or_clear_bad(src_pgd))
621 continue;
622 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
623 vma, addr, next))
624 return -ENOMEM;
625 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
626 return 0;
627}
628
Robin Holt51c6f662005-11-13 16:06:42 -0800629static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700630 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800632 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633{
Nick Pigginb5810032005-10-29 18:16:12 -0700634 struct mm_struct *mm = tlb->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 pte_t *pte;
Hugh Dickins508034a2005-10-29 18:16:30 -0700636 spinlock_t *ptl;
Hugh Dickinsae859762005-10-29 18:16:05 -0700637 int file_rss = 0;
638 int anon_rss = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639
Hugh Dickins508034a2005-10-29 18:16:30 -0700640 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700641 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 do {
643 pte_t ptent = *pte;
Robin Holt51c6f662005-11-13 16:06:42 -0800644 if (pte_none(ptent)) {
645 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800647 }
Hugh Dickins6f5e6b92006-03-16 23:04:09 -0800648
649 (*zap_work) -= PAGE_SIZE;
650
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800652 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -0800653
Linus Torvalds6aab3412005-11-28 14:34:23 -0800654 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 if (unlikely(details) && page) {
656 /*
657 * unmap_shared_mapping_pages() wants to
658 * invalidate cache without truncating:
659 * unmap shared but keep private pages.
660 */
661 if (details->check_mapping &&
662 details->check_mapping != page->mapping)
663 continue;
664 /*
665 * Each page->index must be checked when
666 * invalidating or truncating nonlinear.
667 */
668 if (details->nonlinear_vma &&
669 (page->index < details->first_index ||
670 page->index > details->last_index))
671 continue;
672 }
Nick Pigginb5810032005-10-29 18:16:12 -0700673 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -0700674 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 tlb_remove_tlb_entry(tlb, pte, addr);
676 if (unlikely(!page))
677 continue;
678 if (unlikely(details) && details->nonlinear_vma
679 && linear_page_index(details->nonlinear_vma,
680 addr) != page->index)
Nick Pigginb5810032005-10-29 18:16:12 -0700681 set_pte_at(mm, addr, pte,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 pgoff_to_pte(page->index));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 if (PageAnon(page))
Hugh Dickins86d912f2005-10-29 18:16:14 -0700684 anon_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700685 else {
686 if (pte_dirty(ptent))
687 set_page_dirty(page);
688 if (pte_young(ptent))
Ken Chendaa88c82007-02-10 01:43:18 -0800689 SetPageReferenced(page);
Hugh Dickins86d912f2005-10-29 18:16:14 -0700690 file_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700691 }
Nick Piggin7de6b802006-12-22 01:09:33 -0800692 page_remove_rmap(page, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 tlb_remove_page(tlb, page);
694 continue;
695 }
696 /*
697 * If details->check_mapping, we leave swap entries;
698 * if details->nonlinear_vma, we leave file entries.
699 */
700 if (unlikely(details))
701 continue;
702 if (!pte_file(ptent))
703 free_swap_and_cache(pte_to_swp_entry(ptent));
Zachary Amsden9888a1c2006-09-30 23:29:31 -0700704 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
Robin Holt51c6f662005-11-13 16:06:42 -0800705 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
Hugh Dickinsae859762005-10-29 18:16:05 -0700706
Hugh Dickins86d912f2005-10-29 18:16:14 -0700707 add_mm_rss(mm, file_rss, anon_rss);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700708 arch_leave_lazy_mmu_mode();
Hugh Dickins508034a2005-10-29 18:16:30 -0700709 pte_unmap_unlock(pte - 1, ptl);
Robin Holt51c6f662005-11-13 16:06:42 -0800710
711 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712}
713
Robin Holt51c6f662005-11-13 16:06:42 -0800714static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700715 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800717 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718{
719 pmd_t *pmd;
720 unsigned long next;
721
722 pmd = pmd_offset(pud, addr);
723 do {
724 next = pmd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800725 if (pmd_none_or_clear_bad(pmd)) {
726 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800728 }
729 next = zap_pte_range(tlb, vma, pmd, addr, next,
730 zap_work, details);
731 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
732
733 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734}
735
Robin Holt51c6f662005-11-13 16:06:42 -0800736static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700737 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800739 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740{
741 pud_t *pud;
742 unsigned long next;
743
744 pud = pud_offset(pgd, addr);
745 do {
746 next = pud_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800747 if (pud_none_or_clear_bad(pud)) {
748 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800750 }
751 next = zap_pmd_range(tlb, vma, pud, addr, next,
752 zap_work, details);
753 } while (pud++, addr = next, (addr != end && *zap_work > 0));
754
755 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756}
757
Robin Holt51c6f662005-11-13 16:06:42 -0800758static unsigned long unmap_page_range(struct mmu_gather *tlb,
759 struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800761 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762{
763 pgd_t *pgd;
764 unsigned long next;
765
766 if (details && !details->check_mapping && !details->nonlinear_vma)
767 details = NULL;
768
769 BUG_ON(addr >= end);
770 tlb_start_vma(tlb, vma);
771 pgd = pgd_offset(vma->vm_mm, addr);
772 do {
773 next = pgd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800774 if (pgd_none_or_clear_bad(pgd)) {
775 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800777 }
778 next = zap_pud_range(tlb, vma, pgd, addr, next,
779 zap_work, details);
780 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 tlb_end_vma(tlb, vma);
Robin Holt51c6f662005-11-13 16:06:42 -0800782
783 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784}
785
786#ifdef CONFIG_PREEMPT
787# define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
788#else
789/* No preempt: go for improved straight-line efficiency */
790# define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
791#endif
792
793/**
794 * unmap_vmas - unmap a range of memory covered by a list of vma's
795 * @tlbp: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 * @vma: the starting vma
797 * @start_addr: virtual address at which to start unmapping
798 * @end_addr: virtual address at which to end unmapping
799 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
800 * @details: details of nonlinear truncation or shared cache invalidation
801 *
Hugh Dickinsee39b372005-04-19 13:29:15 -0700802 * Returns the end address of the unmapping (restart addr if interrupted).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700804 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700806 * We aim to not hold locks for too long (for scheduling latency reasons).
807 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 * return the ending mmu_gather to the caller.
809 *
810 * Only addresses between `start' and `end' will be unmapped.
811 *
812 * The VMA list must be sorted in ascending virtual address order.
813 *
814 * unmap_vmas() assumes that the caller will flush the whole unmapped address
815 * range after unmap_vmas() returns. So the only responsibility here is to
816 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
817 * drops the lock and schedules.
818 */
Hugh Dickins508034a2005-10-29 18:16:30 -0700819unsigned long unmap_vmas(struct mmu_gather **tlbp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 struct vm_area_struct *vma, unsigned long start_addr,
821 unsigned long end_addr, unsigned long *nr_accounted,
822 struct zap_details *details)
823{
Robin Holt51c6f662005-11-13 16:06:42 -0800824 long zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
826 int tlb_start_valid = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -0700827 unsigned long start = start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700829 int fullmm = (*tlbp)->fullmm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830
831 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 unsigned long end;
833
834 start = max(vma->vm_start, start_addr);
835 if (start >= vma->vm_end)
836 continue;
837 end = min(vma->vm_end, end_addr);
838 if (end <= vma->vm_start)
839 continue;
840
841 if (vma->vm_flags & VM_ACCOUNT)
842 *nr_accounted += (end - start) >> PAGE_SHIFT;
843
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 while (start != end) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 if (!tlb_start_valid) {
846 tlb_start = start;
847 tlb_start_valid = 1;
848 }
849
Robin Holt51c6f662005-11-13 16:06:42 -0800850 if (unlikely(is_vm_hugetlb_page(vma))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 unmap_hugepage_range(vma, start, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800852 zap_work -= (end - start) /
853 (HPAGE_SIZE / PAGE_SIZE);
854 start = end;
855 } else
856 start = unmap_page_range(*tlbp, vma,
857 start, end, &zap_work, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858
Robin Holt51c6f662005-11-13 16:06:42 -0800859 if (zap_work > 0) {
860 BUG_ON(start != end);
861 break;
862 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
864 tlb_finish_mmu(*tlbp, tlb_start, start);
865
866 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +0100867 (i_mmap_lock && spin_needbreak(i_mmap_lock))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 if (i_mmap_lock) {
Hugh Dickins508034a2005-10-29 18:16:30 -0700869 *tlbp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 goto out;
871 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 }
874
Hugh Dickins508034a2005-10-29 18:16:30 -0700875 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 tlb_start_valid = 0;
Robin Holt51c6f662005-11-13 16:06:42 -0800877 zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 }
879 }
880out:
Hugh Dickinsee39b372005-04-19 13:29:15 -0700881 return start; /* which is now the end (or restart) address */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882}
883
884/**
885 * zap_page_range - remove user pages in a given range
886 * @vma: vm_area_struct holding the applicable pages
887 * @address: starting address of pages to zap
888 * @size: number of bytes to zap
889 * @details: details of nonlinear truncation or shared cache invalidation
890 */
Hugh Dickinsee39b372005-04-19 13:29:15 -0700891unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 unsigned long size, struct zap_details *details)
893{
894 struct mm_struct *mm = vma->vm_mm;
895 struct mmu_gather *tlb;
896 unsigned long end = address + size;
897 unsigned long nr_accounted = 0;
898
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -0700901 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -0700902 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
903 if (tlb)
904 tlb_finish_mmu(tlb, address, end);
Hugh Dickinsee39b372005-04-19 13:29:15 -0700905 return end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906}
907
908/*
909 * Do a quick page-table lookup for a single page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800911struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700912 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913{
914 pgd_t *pgd;
915 pud_t *pud;
916 pmd_t *pmd;
917 pte_t *ptep, pte;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700918 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 struct page *page;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800920 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700922 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
923 if (!IS_ERR(page)) {
924 BUG_ON(flags & FOLL_GET);
925 goto out;
926 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700928 page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 pgd = pgd_offset(mm, address);
930 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700931 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932
933 pud = pud_offset(pgd, address);
934 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700935 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936
937 pmd = pmd_offset(pud, address);
938 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700939 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700941 if (pmd_huge(*pmd)) {
942 BUG_ON(flags & FOLL_GET);
943 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
944 goto out;
945 }
946
947 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 if (!ptep)
949 goto out;
950
951 pte = *ptep;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700952 if (!pte_present(pte))
953 goto unlock;
954 if ((flags & FOLL_WRITE) && !pte_write(pte))
955 goto unlock;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800956 page = vm_normal_page(vma, address, pte);
957 if (unlikely(!page))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700958 goto unlock;
959
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700960 if (flags & FOLL_GET)
961 get_page(page);
962 if (flags & FOLL_TOUCH) {
963 if ((flags & FOLL_WRITE) &&
964 !pte_dirty(pte) && !PageDirty(page))
965 set_page_dirty(page);
966 mark_page_accessed(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700968unlock:
969 pte_unmap_unlock(ptep, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970out:
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700971 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700973no_page_table:
974 /*
975 * When core dumping an enormous anonymous area that nobody
976 * has touched so far, we don't want to allocate page tables.
977 */
978 if (flags & FOLL_ANON) {
Nick Piggin557ed1f2007-10-16 01:24:40 -0700979 page = ZERO_PAGE(0);
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700980 if (flags & FOLL_GET)
981 get_page(page);
982 BUG_ON(flags & FOLL_WRITE);
983 }
984 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985}
986
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
988 unsigned long start, int len, int write, int force,
989 struct page **pages, struct vm_area_struct **vmas)
990{
991 int i;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700992 unsigned int vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
994 /*
995 * Require read or write permissions.
996 * If 'force' is set, we only require the "MAY" flags.
997 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700998 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
999 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 i = 0;
1001
1002 do {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001003 struct vm_area_struct *vma;
1004 unsigned int foll_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005
1006 vma = find_extend_vma(mm, start);
1007 if (!vma && in_gate_area(tsk, start)) {
1008 unsigned long pg = start & PAGE_MASK;
1009 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
1010 pgd_t *pgd;
1011 pud_t *pud;
1012 pmd_t *pmd;
1013 pte_t *pte;
1014 if (write) /* user gate pages are read-only */
1015 return i ? : -EFAULT;
1016 if (pg > TASK_SIZE)
1017 pgd = pgd_offset_k(pg);
1018 else
1019 pgd = pgd_offset_gate(mm, pg);
1020 BUG_ON(pgd_none(*pgd));
1021 pud = pud_offset(pgd, pg);
1022 BUG_ON(pud_none(*pud));
1023 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001024 if (pmd_none(*pmd))
1025 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001027 if (pte_none(*pte)) {
1028 pte_unmap(pte);
1029 return i ? : -EFAULT;
1030 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 if (pages) {
Nick Pigginfa2a4552005-11-29 18:43:17 +11001032 struct page *page = vm_normal_page(gate_vma, start, *pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001033 pages[i] = page;
1034 if (page)
1035 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 }
1037 pte_unmap(pte);
1038 if (vmas)
1039 vmas[i] = gate_vma;
1040 i++;
1041 start += PAGE_SIZE;
1042 len--;
1043 continue;
1044 }
1045
Linus Torvalds1ff80382005-12-12 16:24:33 -08001046 if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001047 || !(vm_flags & vma->vm_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 return i ? : -EFAULT;
1049
1050 if (is_vm_hugetlb_page(vma)) {
1051 i = follow_hugetlb_page(mm, vma, pages, vmas,
Adam Litke5b23dbe2007-11-14 16:59:33 -08001052 &start, &len, i, write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 continue;
1054 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001055
1056 foll_flags = FOLL_TOUCH;
1057 if (pages)
1058 foll_flags |= FOLL_GET;
1059 if (!write && !(vma->vm_flags & VM_LOCKED) &&
Nick Piggin54cb8822007-07-19 01:46:59 -07001060 (!vma->vm_ops || (!vma->vm_ops->nopage &&
1061 !vma->vm_ops->fault)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001062 foll_flags |= FOLL_ANON;
1063
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 do {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001065 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066
Ethan Solomita462e00c2007-07-15 23:38:16 -07001067 /*
1068 * If tsk is ooming, cut off its access to large memory
1069 * allocations. It has a pending SIGKILL, but it can't
1070 * be processed until returning to user space.
1071 */
1072 if (unlikely(test_tsk_thread_flag(tsk, TIF_MEMDIE)))
1073 return -ENOMEM;
1074
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001075 if (write)
1076 foll_flags |= FOLL_WRITE;
1077
1078 cond_resched();
Linus Torvalds6aab3412005-11-28 14:34:23 -08001079 while (!(page = follow_page(vma, start, foll_flags))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001080 int ret;
Nick Piggin83c54072007-07-19 01:47:05 -07001081 ret = handle_mm_fault(mm, vma, start,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001082 foll_flags & FOLL_WRITE);
Nick Piggin83c54072007-07-19 01:47:05 -07001083 if (ret & VM_FAULT_ERROR) {
1084 if (ret & VM_FAULT_OOM)
1085 return i ? i : -ENOMEM;
1086 else if (ret & VM_FAULT_SIGBUS)
1087 return i ? i : -EFAULT;
1088 BUG();
1089 }
1090 if (ret & VM_FAULT_MAJOR)
1091 tsk->maj_flt++;
1092 else
1093 tsk->min_flt++;
1094
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001095 /*
Nick Piggin83c54072007-07-19 01:47:05 -07001096 * The VM_FAULT_WRITE bit tells us that
1097 * do_wp_page has broken COW when necessary,
1098 * even if maybe_mkwrite decided not to set
1099 * pte_write. We can thus safely do subsequent
1100 * page lookups as if they were reads.
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001101 */
1102 if (ret & VM_FAULT_WRITE)
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001103 foll_flags &= ~FOLL_WRITE;
Nick Piggin83c54072007-07-19 01:47:05 -07001104
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07001105 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 }
1107 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001108 pages[i] = page;
James Bottomley03beb072006-03-26 01:36:57 -08001109
Russell Kinga6f36be2006-12-30 22:24:19 +00001110 flush_anon_page(vma, page, start);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001111 flush_dcache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 }
1113 if (vmas)
1114 vmas[i] = vma;
1115 i++;
1116 start += PAGE_SIZE;
1117 len--;
Hugh Dickins08ef4722005-06-21 17:15:10 -07001118 } while (len && start < vma->vm_end);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001119 } while (len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 return i;
1121}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122EXPORT_SYMBOL(get_user_pages);
1123
Harvey Harrison920c7a52008-02-04 22:29:26 -08001124pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr,
1125 spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001126{
1127 pgd_t * pgd = pgd_offset(mm, addr);
1128 pud_t * pud = pud_alloc(mm, pgd, addr);
1129 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001130 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001131 if (pmd)
1132 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1133 }
1134 return NULL;
1135}
1136
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001138 * This is the old fallback for page remapping.
1139 *
1140 * For historical reasons, it only allows reserved pages. Only
1141 * old drivers should use this, and they needed to mark their
1142 * pages reserved for the old functions anyway.
1143 */
1144static int insert_page(struct mm_struct *mm, unsigned long addr, struct page *page, pgprot_t prot)
1145{
1146 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001147 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001148 spinlock_t *ptl;
1149
Balbir Singhe1a1cd52008-02-07 00:14:02 -08001150 retval = mem_cgroup_charge(page, mm, GFP_KERNEL);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001151 if (retval)
1152 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001153
1154 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001155 if (PageAnon(page))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001156 goto out_uncharge;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001157 retval = -ENOMEM;
1158 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001159 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001160 if (!pte)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001161 goto out_uncharge;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001162 retval = -EBUSY;
1163 if (!pte_none(*pte))
1164 goto out_unlock;
1165
1166 /* Ok, finally just insert the thing.. */
1167 get_page(page);
1168 inc_mm_counter(mm, file_rss);
1169 page_add_file_rmap(page);
1170 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1171
1172 retval = 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001173 pte_unmap_unlock(pte, ptl);
1174 return retval;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001175out_unlock:
1176 pte_unmap_unlock(pte, ptl);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001177out_uncharge:
1178 mem_cgroup_uncharge_page(page);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001179out:
1180 return retval;
1181}
1182
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001183/**
1184 * vm_insert_page - insert single page into user vma
1185 * @vma: user vma to map to
1186 * @addr: target user address of this page
1187 * @page: source kernel page
1188 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001189 * This allows drivers to insert individual pages they've allocated
1190 * into a user vma.
1191 *
1192 * The page has to be a nice clean _individual_ kernel allocation.
1193 * If you allocate a compound page, you need to have marked it as
1194 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001195 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001196 *
1197 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1198 * took an arbitrary page protection parameter. This doesn't allow
1199 * that. Your vma protection will have to be set up correctly, which
1200 * means that if you want a shared writable mapping, you'd better
1201 * ask for a shared writable mapping!
1202 *
1203 * The page does not need to be reserved.
1204 */
1205int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, struct page *page)
1206{
1207 if (addr < vma->vm_start || addr >= vma->vm_end)
1208 return -EFAULT;
1209 if (!page_count(page))
1210 return -EINVAL;
Linus Torvalds4d7672b2005-12-16 10:21:23 -08001211 vma->vm_flags |= VM_INSERTPAGE;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001212 return insert_page(vma->vm_mm, addr, page, vma->vm_page_prot);
1213}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001214EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001215
Nick Piggine0dc0d82007-02-12 00:51:36 -08001216/**
1217 * vm_insert_pfn - insert single pfn into user vma
1218 * @vma: user vma to map to
1219 * @addr: target user address of this page
1220 * @pfn: source kernel pfn
1221 *
1222 * Similar to vm_inert_page, this allows drivers to insert individual pages
1223 * they've allocated into a user vma. Same comments apply.
1224 *
1225 * This function should only be called from a vm_ops->fault handler, and
1226 * in that case the handler should return NULL.
1227 */
1228int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1229 unsigned long pfn)
1230{
1231 struct mm_struct *mm = vma->vm_mm;
1232 int retval;
1233 pte_t *pte, entry;
1234 spinlock_t *ptl;
1235
1236 BUG_ON(!(vma->vm_flags & VM_PFNMAP));
1237 BUG_ON(is_cow_mapping(vma->vm_flags));
1238
1239 retval = -ENOMEM;
1240 pte = get_locked_pte(mm, addr, &ptl);
1241 if (!pte)
1242 goto out;
1243 retval = -EBUSY;
1244 if (!pte_none(*pte))
1245 goto out_unlock;
1246
1247 /* Ok, finally just insert the thing.. */
1248 entry = pfn_pte(pfn, vma->vm_page_prot);
1249 set_pte_at(mm, addr, pte, entry);
1250 update_mmu_cache(vma, addr, entry);
1251
1252 retval = 0;
1253out_unlock:
1254 pte_unmap_unlock(pte, ptl);
1255
1256out:
1257 return retval;
1258}
1259EXPORT_SYMBOL(vm_insert_pfn);
1260
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001261/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 * maps a range of physical memory into the requested pages. the old
1263 * mappings are removed. any references to nonexistent pages results
1264 * in null mappings (currently treated as "copy-on-access")
1265 */
1266static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1267 unsigned long addr, unsigned long end,
1268 unsigned long pfn, pgprot_t prot)
1269{
1270 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001271 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272
Hugh Dickinsc74df322005-10-29 18:16:23 -07001273 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 if (!pte)
1275 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001276 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 do {
1278 BUG_ON(!pte_none(*pte));
Nick Pigginb5810032005-10-29 18:16:12 -07001279 set_pte_at(mm, addr, pte, pfn_pte(pfn, prot));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 pfn++;
1281 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001282 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001283 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 return 0;
1285}
1286
1287static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1288 unsigned long addr, unsigned long end,
1289 unsigned long pfn, pgprot_t prot)
1290{
1291 pmd_t *pmd;
1292 unsigned long next;
1293
1294 pfn -= addr >> PAGE_SHIFT;
1295 pmd = pmd_alloc(mm, pud, addr);
1296 if (!pmd)
1297 return -ENOMEM;
1298 do {
1299 next = pmd_addr_end(addr, end);
1300 if (remap_pte_range(mm, pmd, addr, next,
1301 pfn + (addr >> PAGE_SHIFT), prot))
1302 return -ENOMEM;
1303 } while (pmd++, addr = next, addr != end);
1304 return 0;
1305}
1306
1307static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1308 unsigned long addr, unsigned long end,
1309 unsigned long pfn, pgprot_t prot)
1310{
1311 pud_t *pud;
1312 unsigned long next;
1313
1314 pfn -= addr >> PAGE_SHIFT;
1315 pud = pud_alloc(mm, pgd, addr);
1316 if (!pud)
1317 return -ENOMEM;
1318 do {
1319 next = pud_addr_end(addr, end);
1320 if (remap_pmd_range(mm, pud, addr, next,
1321 pfn + (addr >> PAGE_SHIFT), prot))
1322 return -ENOMEM;
1323 } while (pud++, addr = next, addr != end);
1324 return 0;
1325}
1326
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001327/**
1328 * remap_pfn_range - remap kernel memory to userspace
1329 * @vma: user vma to map to
1330 * @addr: target user address to start at
1331 * @pfn: physical address of kernel memory
1332 * @size: size of map area
1333 * @prot: page protection flags for this mapping
1334 *
1335 * Note: this is only safe if the mm semaphore is held when called.
1336 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1338 unsigned long pfn, unsigned long size, pgprot_t prot)
1339{
1340 pgd_t *pgd;
1341 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001342 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 struct mm_struct *mm = vma->vm_mm;
1344 int err;
1345
1346 /*
1347 * Physically remapped pages are special. Tell the
1348 * rest of the world about it:
1349 * VM_IO tells people not to look at these pages
1350 * (accesses can have side effects).
Hugh Dickins0b14c172005-11-21 21:32:15 -08001351 * VM_RESERVED is specified all over the place, because
1352 * in 2.4 it kept swapout's vma scan off this vma; but
1353 * in 2.6 the LRU scan won't even find its pages, so this
1354 * flag means no more than count its pages in reserved_vm,
1355 * and omit it from core dump, even when VM_IO turned off.
Linus Torvalds6aab3412005-11-28 14:34:23 -08001356 * VM_PFNMAP tells the core MM that the base pages are just
1357 * raw PFN mappings, and do not have a "struct page" associated
1358 * with them.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001359 *
1360 * There's a horrible special case to handle copy-on-write
1361 * behaviour that some programs depend on. We mark the "original"
1362 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 */
Linus Torvalds67121172005-12-11 20:38:17 -08001364 if (is_cow_mapping(vma->vm_flags)) {
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001365 if (addr != vma->vm_start || end != vma->vm_end)
Linus Torvalds7fc7e2e2005-12-11 19:57:52 -08001366 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001367 vma->vm_pgoff = pfn;
1368 }
1369
Linus Torvalds6aab3412005-11-28 14:34:23 -08001370 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371
1372 BUG_ON(addr >= end);
1373 pfn -= addr >> PAGE_SHIFT;
1374 pgd = pgd_offset(mm, addr);
1375 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 do {
1377 next = pgd_addr_end(addr, end);
1378 err = remap_pud_range(mm, pgd, addr, next,
1379 pfn + (addr >> PAGE_SHIFT), prot);
1380 if (err)
1381 break;
1382 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 return err;
1384}
1385EXPORT_SYMBOL(remap_pfn_range);
1386
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001387static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1388 unsigned long addr, unsigned long end,
1389 pte_fn_t fn, void *data)
1390{
1391 pte_t *pte;
1392 int err;
1393 struct page *pmd_page;
Borislav Petkov94909912007-05-06 14:49:17 -07001394 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001395
1396 pte = (mm == &init_mm) ?
1397 pte_alloc_kernel(pmd, addr) :
1398 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1399 if (!pte)
1400 return -ENOMEM;
1401
1402 BUG_ON(pmd_huge(*pmd));
1403
1404 pmd_page = pmd_page(*pmd);
1405
1406 do {
1407 err = fn(pte, pmd_page, addr, data);
1408 if (err)
1409 break;
1410 } while (pte++, addr += PAGE_SIZE, addr != end);
1411
1412 if (mm != &init_mm)
1413 pte_unmap_unlock(pte-1, ptl);
1414 return err;
1415}
1416
1417static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1418 unsigned long addr, unsigned long end,
1419 pte_fn_t fn, void *data)
1420{
1421 pmd_t *pmd;
1422 unsigned long next;
1423 int err;
1424
1425 pmd = pmd_alloc(mm, pud, addr);
1426 if (!pmd)
1427 return -ENOMEM;
1428 do {
1429 next = pmd_addr_end(addr, end);
1430 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1431 if (err)
1432 break;
1433 } while (pmd++, addr = next, addr != end);
1434 return err;
1435}
1436
1437static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1438 unsigned long addr, unsigned long end,
1439 pte_fn_t fn, void *data)
1440{
1441 pud_t *pud;
1442 unsigned long next;
1443 int err;
1444
1445 pud = pud_alloc(mm, pgd, addr);
1446 if (!pud)
1447 return -ENOMEM;
1448 do {
1449 next = pud_addr_end(addr, end);
1450 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1451 if (err)
1452 break;
1453 } while (pud++, addr = next, addr != end);
1454 return err;
1455}
1456
1457/*
1458 * Scan a region of virtual memory, filling in page tables as necessary
1459 * and calling a provided function on each leaf page table.
1460 */
1461int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1462 unsigned long size, pte_fn_t fn, void *data)
1463{
1464 pgd_t *pgd;
1465 unsigned long next;
1466 unsigned long end = addr + size;
1467 int err;
1468
1469 BUG_ON(addr >= end);
1470 pgd = pgd_offset(mm, addr);
1471 do {
1472 next = pgd_addr_end(addr, end);
1473 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1474 if (err)
1475 break;
1476 } while (pgd++, addr = next, addr != end);
1477 return err;
1478}
1479EXPORT_SYMBOL_GPL(apply_to_page_range);
1480
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001482 * handle_pte_fault chooses page fault handler according to an entry
1483 * which was read non-atomically. Before making any commitment, on
1484 * those architectures or configurations (e.g. i386 with PAE) which
1485 * might give a mix of unmatched parts, do_swap_page and do_file_page
1486 * must check under lock before unmapping the pte and proceeding
1487 * (but do_wp_page is only called after already making such a check;
1488 * and do_anonymous_page and do_no_page can safely check later on).
1489 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001490static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001491 pte_t *page_table, pte_t orig_pte)
1492{
1493 int same = 1;
1494#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1495 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001496 spinlock_t *ptl = pte_lockptr(mm, pmd);
1497 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001498 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001499 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001500 }
1501#endif
1502 pte_unmap(page_table);
1503 return same;
1504}
1505
1506/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1508 * servicing faults for write access. In the normal case, do always want
1509 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1510 * that do not have writing enabled, when used by access_process_vm.
1511 */
1512static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1513{
1514 if (likely(vma->vm_flags & VM_WRITE))
1515 pte = pte_mkwrite(pte);
1516 return pte;
1517}
1518
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001519static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma)
Linus Torvalds6aab3412005-11-28 14:34:23 -08001520{
1521 /*
1522 * If the source page was a PFN mapping, we don't have
1523 * a "struct page" for it. We do a best-effort copy by
1524 * just copying from the original user address. If that
1525 * fails, we just zero-fill it. Live with it.
1526 */
1527 if (unlikely(!src)) {
1528 void *kaddr = kmap_atomic(dst, KM_USER0);
Linus Torvalds5d2a2db2005-11-29 14:07:55 -08001529 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1530
1531 /*
1532 * This really shouldn't fail, because the page is there
1533 * in the page tables. But it might just be unreadable,
1534 * in which case we just give up and fill the result with
1535 * zeroes.
1536 */
1537 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Linus Torvalds6aab3412005-11-28 14:34:23 -08001538 memset(kaddr, 0, PAGE_SIZE);
1539 kunmap_atomic(kaddr, KM_USER0);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07001540 flush_dcache_page(dst);
Nick Piggin0ed361d2008-02-04 22:29:34 -08001541 } else
1542 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001543}
1544
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 * This routine handles present pages, when users try to write
1547 * to a shared page. It is done by copying the page to a new address
1548 * and decrementing the shared-page counter for the old page.
1549 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 * Note that this routine assumes that the protection checks have been
1551 * done by the caller (the low-level page fault routine in most cases).
1552 * Thus we can safely just mark it writable once we've done any necessary
1553 * COW.
1554 *
1555 * We also mark the page dirty at this point even though the page will
1556 * change only once the write actually happens. This avoids a few races,
1557 * and potentially makes it more efficient.
1558 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001559 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1560 * but allow concurrent faults), with pte both mapped and locked.
1561 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001563static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1564 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001565 spinlock_t *ptl, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566{
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001567 struct page *old_page, *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 pte_t entry;
Nick Piggin83c54072007-07-19 01:47:05 -07001569 int reuse = 0, ret = 0;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001570 int page_mkwrite = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001571 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572
Linus Torvalds6aab3412005-11-28 14:34:23 -08001573 old_page = vm_normal_page(vma, address, orig_pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001574 if (!old_page)
1575 goto gotten;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001577 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001578 * Take out anonymous pages first, anonymous shared vmas are
1579 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001580 */
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001581 if (PageAnon(old_page)) {
1582 if (!TestSetPageLocked(old_page)) {
1583 reuse = can_share_swap_page(old_page);
1584 unlock_page(old_page);
1585 }
1586 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001587 (VM_WRITE|VM_SHARED))) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001588 /*
1589 * Only catch write-faults on shared writable pages,
1590 * read-only shared pages can get COWed by
1591 * get_user_pages(.write=1, .force=1).
1592 */
David Howells9637a5e2006-06-23 02:03:43 -07001593 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
1594 /*
1595 * Notify the address space that the page is about to
1596 * become writable so that it can prohibit this or wait
1597 * for the page to get into an appropriate state.
1598 *
1599 * We do this without the lock held, so that it can
1600 * sleep if it needs to.
1601 */
1602 page_cache_get(old_page);
1603 pte_unmap_unlock(page_table, ptl);
1604
1605 if (vma->vm_ops->page_mkwrite(vma, old_page) < 0)
1606 goto unwritable_page;
1607
David Howells9637a5e2006-06-23 02:03:43 -07001608 /*
1609 * Since we dropped the lock we need to revalidate
1610 * the PTE as someone else may have changed it. If
1611 * they did, we just return, as we can count on the
1612 * MMU to tell us if they didn't also make it writable.
1613 */
1614 page_table = pte_offset_map_lock(mm, pmd, address,
1615 &ptl);
Hugh Dickinsc3704ce2007-02-10 01:43:00 -08001616 page_cache_release(old_page);
David Howells9637a5e2006-06-23 02:03:43 -07001617 if (!pte_same(*page_table, orig_pte))
1618 goto unlock;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001619
1620 page_mkwrite = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001622 dirty_page = old_page;
1623 get_page(dirty_page);
David Howells9637a5e2006-06-23 02:03:43 -07001624 reuse = 1;
David Howells9637a5e2006-06-23 02:03:43 -07001625 }
1626
1627 if (reuse) {
1628 flush_cache_page(vma, address, pte_pfn(orig_pte));
1629 entry = pte_mkyoung(orig_pte);
1630 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
KAMEZAWA Hiroyuki954ffcb2007-10-16 01:25:44 -07001631 if (ptep_set_access_flags(vma, address, page_table, entry,1))
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07001632 update_mmu_cache(vma, address, entry);
David Howells9637a5e2006-06-23 02:03:43 -07001633 ret |= VM_FAULT_WRITE;
1634 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636
1637 /*
1638 * Ok, we need to copy. Oh, well..
1639 */
Nick Pigginb5810032005-10-29 18:16:12 -07001640 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08001641gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001642 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643
1644 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07001645 goto oom;
Nick Piggin557ed1f2007-10-16 01:24:40 -07001646 VM_BUG_ON(old_page == ZERO_PAGE(0));
1647 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
1648 if (!new_page)
1649 goto oom;
1650 cow_user_page(new_page, old_page, address, vma);
Nick Piggin0ed361d2008-02-04 22:29:34 -08001651 __SetPageUptodate(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001652
Balbir Singhe1a1cd52008-02-07 00:14:02 -08001653 if (mem_cgroup_charge(new_page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001654 goto oom_free_new;
1655
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 /*
1657 * Re-check the pte - we dropped the lock
1658 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001659 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07001660 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08001661 if (old_page) {
Nick Piggin7de6b802006-12-22 01:09:33 -08001662 page_remove_rmap(old_page, vma);
Hugh Dickins920fc352005-11-21 21:32:17 -08001663 if (!PageAnon(old_page)) {
1664 dec_mm_counter(mm, file_rss);
1665 inc_mm_counter(mm, anon_rss);
1666 }
1667 } else
Hugh Dickins42946212005-10-29 18:16:05 -07001668 inc_mm_counter(mm, anon_rss);
Ben Collinseca35132005-11-29 11:45:26 -08001669 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07001670 entry = mk_pte(new_page, vma->vm_page_prot);
1671 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Siddha, Suresh B4ce072f2006-09-29 01:58:42 -07001672 /*
1673 * Clear the pte entry and flush it first, before updating the
1674 * pte with the new entry. This will avoid a race condition
1675 * seen in the presence of one thread doing SMC and another
1676 * thread doing COW.
1677 */
1678 ptep_clear_flush(vma, address, page_table);
1679 set_pte_at(mm, address, page_table, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07001680 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 lru_cache_add_active(new_page);
Nick Piggin9617d952006-01-06 00:11:12 -08001682 page_add_new_anon_rmap(new_page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683
1684 /* Free the old page.. */
1685 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001686 ret |= VM_FAULT_WRITE;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001687 } else
1688 mem_cgroup_uncharge_page(new_page);
1689
Hugh Dickins920fc352005-11-21 21:32:17 -08001690 if (new_page)
1691 page_cache_release(new_page);
1692 if (old_page)
1693 page_cache_release(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001694unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001695 pte_unmap_unlock(page_table, ptl);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001696 if (dirty_page) {
Anton Salikhmetov8f7b3d12008-01-23 02:21:18 +03001697 if (vma->vm_file)
1698 file_update_time(vma->vm_file);
1699
Nick Piggin79352892007-07-19 01:47:22 -07001700 /*
1701 * Yes, Virginia, this is actually required to prevent a race
1702 * with clear_page_dirty_for_io() from clearing the page dirty
1703 * bit after it clear all dirty ptes, but before a racing
1704 * do_wp_page installs a dirty pte.
1705 *
1706 * do_no_page is protected similarly.
1707 */
1708 wait_on_page_locked(dirty_page);
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001709 set_page_dirty_balance(dirty_page, page_mkwrite);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001710 put_page(dirty_page);
1711 }
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001712 return ret;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001713oom_free_new:
1714 __free_page(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001715oom:
Hugh Dickins920fc352005-11-21 21:32:17 -08001716 if (old_page)
1717 page_cache_release(old_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 return VM_FAULT_OOM;
David Howells9637a5e2006-06-23 02:03:43 -07001719
1720unwritable_page:
1721 page_cache_release(old_page);
1722 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723}
1724
1725/*
1726 * Helper functions for unmap_mapping_range().
1727 *
1728 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1729 *
1730 * We have to restart searching the prio_tree whenever we drop the lock,
1731 * since the iterator is only valid while the lock is held, and anyway
1732 * a later vma might be split and reinserted earlier while lock dropped.
1733 *
1734 * The list of nonlinear vmas could be handled more efficiently, using
1735 * a placeholder, but handle it in the same way until a need is shown.
1736 * It is important to search the prio_tree before nonlinear list: a vma
1737 * may become nonlinear and be shifted from prio_tree to nonlinear list
1738 * while the lock is dropped; but never shifted from list to prio_tree.
1739 *
1740 * In order to make forward progress despite restarting the search,
1741 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1742 * quickly skip it next time around. Since the prio_tree search only
1743 * shows us those vmas affected by unmapping the range in question, we
1744 * can't efficiently keep all vmas in step with mapping->truncate_count:
1745 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1746 * mapping->truncate_count and vma->vm_truncate_count are protected by
1747 * i_mmap_lock.
1748 *
1749 * In order to make forward progress despite repeatedly restarting some
Hugh Dickinsee39b372005-04-19 13:29:15 -07001750 * large vma, note the restart_addr from unmap_vmas when it breaks out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 * and restart from that address when we reach that vma again. It might
1752 * have been split or merged, shrunk or extended, but never shifted: so
1753 * restart_addr remains valid so long as it remains in the vma's range.
1754 * unmap_mapping_range forces truncate_count to leap over page-aligned
1755 * values so we can save vma's restart_addr in its truncate_count field.
1756 */
1757#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1758
1759static void reset_vma_truncate_counts(struct address_space *mapping)
1760{
1761 struct vm_area_struct *vma;
1762 struct prio_tree_iter iter;
1763
1764 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1765 vma->vm_truncate_count = 0;
1766 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1767 vma->vm_truncate_count = 0;
1768}
1769
1770static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1771 unsigned long start_addr, unsigned long end_addr,
1772 struct zap_details *details)
1773{
1774 unsigned long restart_addr;
1775 int need_break;
1776
Nick Piggind00806b2007-07-19 01:46:57 -07001777 /*
1778 * files that support invalidating or truncating portions of the
Nick Piggind0217ac2007-07-19 01:47:03 -07001779 * file from under mmaped areas must have their ->fault function
Nick Piggin83c54072007-07-19 01:47:05 -07001780 * return a locked page (and set VM_FAULT_LOCKED in the return).
1781 * This provides synchronisation against concurrent unmapping here.
Nick Piggind00806b2007-07-19 01:46:57 -07001782 */
Nick Piggind00806b2007-07-19 01:46:57 -07001783
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784again:
1785 restart_addr = vma->vm_truncate_count;
1786 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1787 start_addr = restart_addr;
1788 if (start_addr >= end_addr) {
1789 /* Top of vma has been split off since last time */
1790 vma->vm_truncate_count = details->truncate_count;
1791 return 0;
1792 }
1793 }
1794
Hugh Dickinsee39b372005-04-19 13:29:15 -07001795 restart_addr = zap_page_range(vma, start_addr,
1796 end_addr - start_addr, details);
Nick Piggin95c354f2008-01-30 13:31:20 +01001797 need_break = need_resched() || spin_needbreak(details->i_mmap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798
Hugh Dickinsee39b372005-04-19 13:29:15 -07001799 if (restart_addr >= end_addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 /* We have now completed this vma: mark it so */
1801 vma->vm_truncate_count = details->truncate_count;
1802 if (!need_break)
1803 return 0;
1804 } else {
1805 /* Note restart_addr in vma's truncate_count field */
Hugh Dickinsee39b372005-04-19 13:29:15 -07001806 vma->vm_truncate_count = restart_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 if (!need_break)
1808 goto again;
1809 }
1810
1811 spin_unlock(details->i_mmap_lock);
1812 cond_resched();
1813 spin_lock(details->i_mmap_lock);
1814 return -EINTR;
1815}
1816
1817static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1818 struct zap_details *details)
1819{
1820 struct vm_area_struct *vma;
1821 struct prio_tree_iter iter;
1822 pgoff_t vba, vea, zba, zea;
1823
1824restart:
1825 vma_prio_tree_foreach(vma, &iter, root,
1826 details->first_index, details->last_index) {
1827 /* Skip quickly over those we have already dealt with */
1828 if (vma->vm_truncate_count == details->truncate_count)
1829 continue;
1830
1831 vba = vma->vm_pgoff;
1832 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1833 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1834 zba = details->first_index;
1835 if (zba < vba)
1836 zba = vba;
1837 zea = details->last_index;
1838 if (zea > vea)
1839 zea = vea;
1840
1841 if (unmap_mapping_range_vma(vma,
1842 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1843 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1844 details) < 0)
1845 goto restart;
1846 }
1847}
1848
1849static inline void unmap_mapping_range_list(struct list_head *head,
1850 struct zap_details *details)
1851{
1852 struct vm_area_struct *vma;
1853
1854 /*
1855 * In nonlinear VMAs there is no correspondence between virtual address
1856 * offset and file offset. So we must perform an exhaustive search
1857 * across *all* the pages in each nonlinear VMA, not just the pages
1858 * whose virtual address lies outside the file truncation point.
1859 */
1860restart:
1861 list_for_each_entry(vma, head, shared.vm_set.list) {
1862 /* Skip quickly over those we have already dealt with */
1863 if (vma->vm_truncate_count == details->truncate_count)
1864 continue;
1865 details->nonlinear_vma = vma;
1866 if (unmap_mapping_range_vma(vma, vma->vm_start,
1867 vma->vm_end, details) < 0)
1868 goto restart;
1869 }
1870}
1871
1872/**
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001873 * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file.
Martin Waitz3d410882005-06-23 22:05:21 -07001874 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 * @holebegin: byte in first page to unmap, relative to the start of
1876 * the underlying file. This will be rounded down to a PAGE_SIZE
1877 * boundary. Note that this is different from vmtruncate(), which
1878 * must keep the partial page. In contrast, we must get rid of
1879 * partial pages.
1880 * @holelen: size of prospective hole in bytes. This will be rounded
1881 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1882 * end of the file.
1883 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1884 * but 0 when invalidating pagecache, don't throw away private data.
1885 */
1886void unmap_mapping_range(struct address_space *mapping,
1887 loff_t const holebegin, loff_t const holelen, int even_cows)
1888{
1889 struct zap_details details;
1890 pgoff_t hba = holebegin >> PAGE_SHIFT;
1891 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1892
1893 /* Check for overflow. */
1894 if (sizeof(holelen) > sizeof(hlen)) {
1895 long long holeend =
1896 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1897 if (holeend & ~(long long)ULONG_MAX)
1898 hlen = ULONG_MAX - hba + 1;
1899 }
1900
1901 details.check_mapping = even_cows? NULL: mapping;
1902 details.nonlinear_vma = NULL;
1903 details.first_index = hba;
1904 details.last_index = hba + hlen - 1;
1905 if (details.last_index < details.first_index)
1906 details.last_index = ULONG_MAX;
1907 details.i_mmap_lock = &mapping->i_mmap_lock;
1908
1909 spin_lock(&mapping->i_mmap_lock);
1910
Nick Piggind00806b2007-07-19 01:46:57 -07001911 /* Protect against endless unmapping loops */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 mapping->truncate_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 if (unlikely(is_restart_addr(mapping->truncate_count))) {
1914 if (mapping->truncate_count == 0)
1915 reset_vma_truncate_counts(mapping);
1916 mapping->truncate_count++;
1917 }
1918 details.truncate_count = mapping->truncate_count;
1919
1920 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
1921 unmap_mapping_range_tree(&mapping->i_mmap, &details);
1922 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
1923 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
1924 spin_unlock(&mapping->i_mmap_lock);
1925}
1926EXPORT_SYMBOL(unmap_mapping_range);
1927
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001928/**
1929 * vmtruncate - unmap mappings "freed" by truncate() syscall
1930 * @inode: inode of the file used
1931 * @offset: file offset to start truncating
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 *
1933 * NOTE! We have to be ready to update the memory sharing
1934 * between the file and the memory map for a potential last
1935 * incomplete page. Ugly, but necessary.
1936 */
1937int vmtruncate(struct inode * inode, loff_t offset)
1938{
Christoph Hellwig61d50482008-02-04 22:28:56 -08001939 if (inode->i_size < offset) {
1940 unsigned long limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941
Christoph Hellwig61d50482008-02-04 22:28:56 -08001942 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1943 if (limit != RLIM_INFINITY && offset > limit)
1944 goto out_sig;
1945 if (offset > inode->i_sb->s_maxbytes)
1946 goto out_big;
1947 i_size_write(inode, offset);
1948 } else {
1949 struct address_space *mapping = inode->i_mapping;
Nick Piggind00806b2007-07-19 01:46:57 -07001950
Christoph Hellwig61d50482008-02-04 22:28:56 -08001951 /*
1952 * truncation of in-use swapfiles is disallowed - it would
1953 * cause subsequent swapout to scribble on the now-freed
1954 * blocks.
1955 */
1956 if (IS_SWAPFILE(inode))
1957 return -ETXTBSY;
1958 i_size_write(inode, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959
Christoph Hellwig61d50482008-02-04 22:28:56 -08001960 /*
1961 * unmap_mapping_range is called twice, first simply for
1962 * efficiency so that truncate_inode_pages does fewer
1963 * single-page unmaps. However after this first call, and
1964 * before truncate_inode_pages finishes, it is possible for
1965 * private pages to be COWed, which remain after
1966 * truncate_inode_pages finishes, hence the second
1967 * unmap_mapping_range call must be made for correctness.
1968 */
1969 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
1970 truncate_inode_pages(mapping, offset);
1971 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
1972 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 if (inode->i_op && inode->i_op->truncate)
1975 inode->i_op->truncate(inode);
1976 return 0;
Christoph Hellwig61d50482008-02-04 22:28:56 -08001977
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978out_sig:
1979 send_sig(SIGXFSZ, current, 0);
1980out_big:
1981 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983EXPORT_SYMBOL(vmtruncate);
1984
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001985int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
1986{
1987 struct address_space *mapping = inode->i_mapping;
1988
1989 /*
1990 * If the underlying filesystem is not going to provide
1991 * a way to truncate a range of blocks (punch a hole) -
1992 * we should return failure right now.
1993 */
1994 if (!inode->i_op || !inode->i_op->truncate_range)
1995 return -ENOSYS;
1996
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001997 mutex_lock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001998 down_write(&inode->i_alloc_sem);
1999 unmap_mapping_range(mapping, offset, (end - offset), 1);
2000 truncate_inode_pages_range(mapping, offset, end);
Nick Piggind00806b2007-07-19 01:46:57 -07002001 unmap_mapping_range(mapping, offset, (end - offset), 1);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002002 inode->i_op->truncate_range(inode, offset, end);
2003 up_write(&inode->i_alloc_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002004 mutex_unlock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002005
2006 return 0;
2007}
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002008
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002010 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2011 * but allow concurrent faults), and pte mapped but not yet locked.
2012 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002014static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2015 unsigned long address, pte_t *page_table, pmd_t *pmd,
2016 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002018 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 struct page *page;
Hugh Dickins65500d22005-10-29 18:15:59 -07002020 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 pte_t pte;
Nick Piggin83c54072007-07-19 01:47:05 -07002022 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002024 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002025 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002026
2027 entry = pte_to_swp_entry(orig_pte);
Christoph Lameter06972122006-06-23 02:03:35 -07002028 if (is_migration_entry(entry)) {
2029 migration_entry_wait(mm, pmd, address);
2030 goto out;
2031 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002032 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 page = lookup_swap_cache(entry);
2034 if (!page) {
Ashwin Chaugule098fe652006-12-06 20:31:54 -08002035 grab_swap_token(); /* Contend for token _before_ read-in */
Hugh Dickins02098fe2008-02-04 22:28:42 -08002036 page = swapin_readahead(entry,
2037 GFP_HIGHUSER_MOVABLE, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 if (!page) {
2039 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002040 * Back out if somebody else faulted in this pte
2041 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002043 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 if (likely(pte_same(*page_table, orig_pte)))
2045 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002046 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002047 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 }
2049
2050 /* Had to read the page from swap area: Major fault */
2051 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002052 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 }
2054
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002055 if (mem_cgroup_charge(page, mm, GFP_KERNEL)) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002056 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2057 ret = VM_FAULT_OOM;
2058 goto out;
2059 }
2060
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 mark_page_accessed(page);
2062 lock_page(page);
Balbir Singh20a10222007-11-14 17:00:33 -08002063 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064
2065 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002066 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002068 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002069 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002070 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002071
2072 if (unlikely(!PageUptodate(page))) {
2073 ret = VM_FAULT_SIGBUS;
2074 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 }
2076
2077 /* The page isn't present yet, go ahead with the fault. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078
Hugh Dickins42946212005-10-29 18:16:05 -07002079 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 pte = mk_pte(page, vma->vm_page_prot);
2081 if (write_access && can_share_swap_page(page)) {
2082 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2083 write_access = 0;
2084 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085
2086 flush_icache_page(vma, page);
2087 set_pte_at(mm, address, page_table, pte);
2088 page_add_anon_rmap(page, vma, address);
2089
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002090 swap_free(entry);
2091 if (vm_swap_full())
2092 remove_exclusive_swap_page(page);
2093 unlock_page(page);
2094
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 if (write_access) {
Nick Piggin83c54072007-07-19 01:47:05 -07002096 /* XXX: We could OR the do_wp_page code with this one? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 if (do_wp_page(mm, vma, address,
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002098 page_table, pmd, ptl, pte) & VM_FAULT_OOM) {
2099 mem_cgroup_uncharge_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 ret = VM_FAULT_OOM;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002101 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 goto out;
2103 }
2104
2105 /* No need to invalidate - it was non-present before */
2106 update_mmu_cache(vma, address, pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07002107unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002108 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109out:
2110 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002111out_nomap:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002112 mem_cgroup_uncharge_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002113 pte_unmap_unlock(page_table, ptl);
Kirill Korotaevb8107482005-05-16 21:53:50 -07002114 unlock_page(page);
2115 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002116 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117}
2118
2119/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002120 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2121 * but allow concurrent faults), and pte mapped but not yet locked.
2122 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002124static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2125 unsigned long address, pte_t *page_table, pmd_t *pmd,
2126 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002128 struct page *page;
2129 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131
Nick Piggin557ed1f2007-10-16 01:24:40 -07002132 /* Allocate our own private page. */
2133 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134
Nick Piggin557ed1f2007-10-16 01:24:40 -07002135 if (unlikely(anon_vma_prepare(vma)))
2136 goto oom;
2137 page = alloc_zeroed_user_highpage_movable(vma, address);
2138 if (!page)
2139 goto oom;
Nick Piggin0ed361d2008-02-04 22:29:34 -08002140 __SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002142 if (mem_cgroup_charge(page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002143 goto oom_free_page;
2144
Nick Piggin557ed1f2007-10-16 01:24:40 -07002145 entry = mk_pte(page, vma->vm_page_prot);
2146 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002147
Nick Piggin557ed1f2007-10-16 01:24:40 -07002148 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2149 if (!pte_none(*page_table))
2150 goto release;
2151 inc_mm_counter(mm, anon_rss);
2152 lru_cache_add_active(page);
2153 page_add_new_anon_rmap(page, vma, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002154 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155
2156 /* No need to invalidate - it was non-present before */
Hugh Dickins65500d22005-10-29 18:15:59 -07002157 update_mmu_cache(vma, address, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002158unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002159 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002160 return 0;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002161release:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002162 mem_cgroup_uncharge_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002163 page_cache_release(page);
2164 goto unlock;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002165oom_free_page:
2166 __free_page(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002167oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 return VM_FAULT_OOM;
2169}
2170
2171/*
Nick Piggin54cb8822007-07-19 01:46:59 -07002172 * __do_fault() tries to create a new page mapping. It aggressively
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 * tries to share with existing pages, but makes a separate copy if
Nick Piggin54cb8822007-07-19 01:46:59 -07002174 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2175 * the next page fault.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 *
2177 * As this is called only for pages that do not currently exist, we
2178 * do not need to flush old virtual caches or the TLB.
2179 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002180 * We enter with non-exclusive mmap_sem (to exclude vma changes,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002181 * but allow concurrent faults), and pte neither mapped nor locked.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002182 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 */
Nick Piggin54cb8822007-07-19 01:46:59 -07002184static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002185 unsigned long address, pmd_t *pmd,
Nick Piggin54cb8822007-07-19 01:46:59 -07002186 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187{
Hugh Dickins16abfa02007-10-04 16:56:06 +01002188 pte_t *page_table;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002189 spinlock_t *ptl;
Nick Piggind0217ac2007-07-19 01:47:03 -07002190 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 int anon = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002193 struct page *dirty_page = NULL;
Nick Piggind0217ac2007-07-19 01:47:03 -07002194 struct vm_fault vmf;
2195 int ret;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002196 int page_mkwrite = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07002197
Nick Piggind0217ac2007-07-19 01:47:03 -07002198 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2199 vmf.pgoff = pgoff;
2200 vmf.flags = flags;
2201 vmf.page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202
Hugh Dickins325f04d2005-11-29 16:55:48 +00002203 BUG_ON(vma->vm_flags & VM_PFNMAP);
2204
Nick Piggin54cb8822007-07-19 01:46:59 -07002205 if (likely(vma->vm_ops->fault)) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002206 ret = vma->vm_ops->fault(vma, &vmf);
Nick Piggin83c54072007-07-19 01:47:05 -07002207 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2208 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07002209 } else {
2210 /* Legacy ->nopage path */
Nick Piggin83c54072007-07-19 01:47:05 -07002211 ret = 0;
Nick Piggind0217ac2007-07-19 01:47:03 -07002212 vmf.page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret);
Nick Piggin54cb8822007-07-19 01:46:59 -07002213 /* no page was available -- either SIGBUS or OOM */
Nick Piggind0217ac2007-07-19 01:47:03 -07002214 if (unlikely(vmf.page == NOPAGE_SIGBUS))
Nick Piggin54cb8822007-07-19 01:46:59 -07002215 return VM_FAULT_SIGBUS;
Nick Piggind0217ac2007-07-19 01:47:03 -07002216 else if (unlikely(vmf.page == NOPAGE_OOM))
Nick Piggin54cb8822007-07-19 01:46:59 -07002217 return VM_FAULT_OOM;
2218 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219
Nick Piggind00806b2007-07-19 01:46:57 -07002220 /*
Nick Piggind0217ac2007-07-19 01:47:03 -07002221 * For consistency in subsequent calls, make the faulted page always
Nick Piggind00806b2007-07-19 01:46:57 -07002222 * locked.
2223 */
Nick Piggin83c54072007-07-19 01:47:05 -07002224 if (unlikely(!(ret & VM_FAULT_LOCKED)))
Nick Piggind0217ac2007-07-19 01:47:03 -07002225 lock_page(vmf.page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002226 else
Nick Piggind0217ac2007-07-19 01:47:03 -07002227 VM_BUG_ON(!PageLocked(vmf.page));
Nick Piggind00806b2007-07-19 01:46:57 -07002228
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 /*
2230 * Should we do an early C-O-W break?
2231 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002232 page = vmf.page;
Nick Piggin54cb8822007-07-19 01:46:59 -07002233 if (flags & FAULT_FLAG_WRITE) {
David Howells9637a5e2006-06-23 02:03:43 -07002234 if (!(vma->vm_flags & VM_SHARED)) {
Nick Piggin54cb8822007-07-19 01:46:59 -07002235 anon = 1;
Nick Piggind00806b2007-07-19 01:46:57 -07002236 if (unlikely(anon_vma_prepare(vma))) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002237 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002238 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002239 }
Nick Piggin83c54072007-07-19 01:47:05 -07002240 page = alloc_page_vma(GFP_HIGHUSER_MOVABLE,
2241 vma, address);
Nick Piggind00806b2007-07-19 01:46:57 -07002242 if (!page) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002243 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002244 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002245 }
Nick Piggind0217ac2007-07-19 01:47:03 -07002246 copy_user_highpage(page, vmf.page, address, vma);
Nick Piggin0ed361d2008-02-04 22:29:34 -08002247 __SetPageUptodate(page);
David Howells9637a5e2006-06-23 02:03:43 -07002248 } else {
Nick Piggin54cb8822007-07-19 01:46:59 -07002249 /*
2250 * If the page will be shareable, see if the backing
David Howells9637a5e2006-06-23 02:03:43 -07002251 * address space wants to know that the page is about
Nick Piggin54cb8822007-07-19 01:46:59 -07002252 * to become writable
2253 */
Mark Fasheh69676142007-07-19 01:47:00 -07002254 if (vma->vm_ops->page_mkwrite) {
2255 unlock_page(page);
2256 if (vma->vm_ops->page_mkwrite(vma, page) < 0) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002257 ret = VM_FAULT_SIGBUS;
2258 anon = 1; /* no anon but release vmf.page */
Mark Fasheh69676142007-07-19 01:47:00 -07002259 goto out_unlocked;
2260 }
2261 lock_page(page);
Nick Piggind0217ac2007-07-19 01:47:03 -07002262 /*
2263 * XXX: this is not quite right (racy vs
2264 * invalidate) to unlock and relock the page
2265 * like this, however a better fix requires
2266 * reworking page_mkwrite locking API, which
2267 * is better done later.
2268 */
2269 if (!page->mapping) {
Nick Piggin83c54072007-07-19 01:47:05 -07002270 ret = 0;
Nick Piggind0217ac2007-07-19 01:47:03 -07002271 anon = 1; /* no anon but release vmf.page */
2272 goto out;
2273 }
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002274 page_mkwrite = 1;
David Howells9637a5e2006-06-23 02:03:43 -07002275 }
2276 }
Nick Piggin54cb8822007-07-19 01:46:59 -07002277
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 }
2279
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002280 if (mem_cgroup_charge(page, mm, GFP_KERNEL)) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002281 ret = VM_FAULT_OOM;
2282 goto out;
2283 }
2284
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002285 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286
2287 /*
2288 * This silly early PAGE_DIRTY setting removes a race
2289 * due to the bad i386 page protection. But it's valid
2290 * for other architectures too.
2291 *
2292 * Note that if write_access is true, we either now have
2293 * an exclusive copy of the page, or this is a shared mapping,
2294 * so we can make it writable and dirty to avoid having to
2295 * handle that later.
2296 */
2297 /* Only go through if we didn't race with anybody else... */
Nick Piggin54cb8822007-07-19 01:46:59 -07002298 if (likely(pte_same(*page_table, orig_pte))) {
Nick Piggind00806b2007-07-19 01:46:57 -07002299 flush_icache_page(vma, page);
2300 entry = mk_pte(page, vma->vm_page_prot);
Nick Piggin54cb8822007-07-19 01:46:59 -07002301 if (flags & FAULT_FLAG_WRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2303 set_pte_at(mm, address, page_table, entry);
2304 if (anon) {
Nick Piggind00806b2007-07-19 01:46:57 -07002305 inc_mm_counter(mm, anon_rss);
2306 lru_cache_add_active(page);
2307 page_add_new_anon_rmap(page, vma, address);
Hugh Dickinsf57e88a2005-11-21 21:32:19 -08002308 } else {
Hugh Dickins42946212005-10-29 18:16:05 -07002309 inc_mm_counter(mm, file_rss);
Nick Piggind00806b2007-07-19 01:46:57 -07002310 page_add_file_rmap(page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002311 if (flags & FAULT_FLAG_WRITE) {
Nick Piggind00806b2007-07-19 01:46:57 -07002312 dirty_page = page;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002313 get_page(dirty_page);
2314 }
Hugh Dickins42946212005-10-29 18:16:05 -07002315 }
Nick Piggind00806b2007-07-19 01:46:57 -07002316
2317 /* no need to invalidate: a not-present page won't be cached */
2318 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 } else {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002320 mem_cgroup_uncharge_page(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002321 if (anon)
2322 page_cache_release(page);
2323 else
Nick Piggin54cb8822007-07-19 01:46:59 -07002324 anon = 1; /* no anon but release faulted_page */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 }
2326
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002327 pte_unmap_unlock(page_table, ptl);
Nick Piggind00806b2007-07-19 01:46:57 -07002328
2329out:
Nick Piggind0217ac2007-07-19 01:47:03 -07002330 unlock_page(vmf.page);
Mark Fasheh69676142007-07-19 01:47:00 -07002331out_unlocked:
Nick Piggind00806b2007-07-19 01:46:57 -07002332 if (anon)
Nick Piggind0217ac2007-07-19 01:47:03 -07002333 page_cache_release(vmf.page);
Nick Piggind00806b2007-07-19 01:46:57 -07002334 else if (dirty_page) {
Anton Salikhmetov8f7b3d12008-01-23 02:21:18 +03002335 if (vma->vm_file)
2336 file_update_time(vma->vm_file);
2337
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002338 set_page_dirty_balance(dirty_page, page_mkwrite);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002339 put_page(dirty_page);
2340 }
Nick Piggind00806b2007-07-19 01:46:57 -07002341
Nick Piggin83c54072007-07-19 01:47:05 -07002342 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07002343}
Nick Piggind00806b2007-07-19 01:46:57 -07002344
Nick Piggin54cb8822007-07-19 01:46:59 -07002345static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2346 unsigned long address, pte_t *page_table, pmd_t *pmd,
2347 int write_access, pte_t orig_pte)
2348{
2349 pgoff_t pgoff = (((address & PAGE_MASK)
Dean Nelson0da7e012007-10-16 01:24:45 -07002350 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
Nick Piggin54cb8822007-07-19 01:46:59 -07002351 unsigned int flags = (write_access ? FAULT_FLAG_WRITE : 0);
2352
Hugh Dickins16abfa02007-10-04 16:56:06 +01002353 pte_unmap(page_table);
2354 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Nick Piggin54cb8822007-07-19 01:46:59 -07002355}
2356
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357
2358/*
Jes Sorensenf4b81802006-09-27 01:50:10 -07002359 * do_no_pfn() tries to create a new page mapping for a page without
2360 * a struct_page backing it
2361 *
2362 * As this is called only for pages that do not currently exist, we
2363 * do not need to flush old virtual caches or the TLB.
2364 *
2365 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2366 * but allow concurrent faults), and pte mapped but not yet locked.
2367 * We return with mmap_sem still held, but pte unmapped and unlocked.
2368 *
2369 * It is expected that the ->nopfn handler always returns the same pfn
2370 * for a given virtual mapping.
2371 *
2372 * Mark this `noinline' to prevent it from bloating the main pagefault code.
2373 */
2374static noinline int do_no_pfn(struct mm_struct *mm, struct vm_area_struct *vma,
2375 unsigned long address, pte_t *page_table, pmd_t *pmd,
2376 int write_access)
2377{
2378 spinlock_t *ptl;
2379 pte_t entry;
2380 unsigned long pfn;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002381
2382 pte_unmap(page_table);
2383 BUG_ON(!(vma->vm_flags & VM_PFNMAP));
2384 BUG_ON(is_cow_mapping(vma->vm_flags));
2385
2386 pfn = vma->vm_ops->nopfn(vma, address & PAGE_MASK);
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002387 if (unlikely(pfn == NOPFN_OOM))
Jes Sorensenf4b81802006-09-27 01:50:10 -07002388 return VM_FAULT_OOM;
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002389 else if (unlikely(pfn == NOPFN_SIGBUS))
Jes Sorensenf4b81802006-09-27 01:50:10 -07002390 return VM_FAULT_SIGBUS;
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002391 else if (unlikely(pfn == NOPFN_REFAULT))
Nick Piggin83c54072007-07-19 01:47:05 -07002392 return 0;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002393
2394 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2395
2396 /* Only go through if we didn't race with anybody else... */
2397 if (pte_none(*page_table)) {
2398 entry = pfn_pte(pfn, vma->vm_page_prot);
2399 if (write_access)
2400 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2401 set_pte_at(mm, address, page_table, entry);
2402 }
2403 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002404 return 0;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002405}
2406
2407/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 * Fault of a previously existing named mapping. Repopulate the pte
2409 * from the encoded file_pte if possible. This enables swappable
2410 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002411 *
2412 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2413 * but allow concurrent faults), and pte mapped but not yet locked.
2414 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002416static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins65500d22005-10-29 18:15:59 -07002417 unsigned long address, pte_t *page_table, pmd_t *pmd,
2418 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419{
Nick Piggind0217ac2007-07-19 01:47:03 -07002420 unsigned int flags = FAULT_FLAG_NONLINEAR |
2421 (write_access ? FAULT_FLAG_WRITE : 0);
Hugh Dickins65500d22005-10-29 18:15:59 -07002422 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002424 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Nick Piggin83c54072007-07-19 01:47:05 -07002425 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426
Nick Piggind0217ac2007-07-19 01:47:03 -07002427 if (unlikely(!(vma->vm_flags & VM_NONLINEAR) ||
2428 !(vma->vm_flags & VM_CAN_NONLINEAR))) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002429 /*
2430 * Page table corrupted: show pte and kill process.
2431 */
Nick Pigginb5810032005-10-29 18:16:12 -07002432 print_bad_pte(vma, orig_pte, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002433 return VM_FAULT_OOM;
2434 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002435
2436 pgoff = pte_to_pgoff(orig_pte);
Hugh Dickins16abfa02007-10-04 16:56:06 +01002437 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438}
2439
2440/*
2441 * These routines also need to handle stuff like marking pages dirty
2442 * and/or accessed for architectures that don't do it in hardware (most
2443 * RISC architectures). The early dirtying is also good on the i386.
2444 *
2445 * There is also a hook called "update_mmu_cache()" that architectures
2446 * with external mmu caches can use to update those (ie the Sparc or
2447 * PowerPC hashed page tables that act as extended TLBs).
2448 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07002449 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2450 * but allow concurrent faults), and pte mapped but not yet locked.
2451 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452 */
2453static inline int handle_pte_fault(struct mm_struct *mm,
Hugh Dickins65500d22005-10-29 18:15:59 -07002454 struct vm_area_struct *vma, unsigned long address,
2455 pte_t *pte, pmd_t *pmd, int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456{
2457 pte_t entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002458 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002460 entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002462 if (pte_none(entry)) {
Jes Sorensenf4b81802006-09-27 01:50:10 -07002463 if (vma->vm_ops) {
Nick Piggin54cb8822007-07-19 01:46:59 -07002464 if (vma->vm_ops->fault || vma->vm_ops->nopage)
2465 return do_linear_fault(mm, vma, address,
2466 pte, pmd, write_access, entry);
Jes Sorensenf4b81802006-09-27 01:50:10 -07002467 if (unlikely(vma->vm_ops->nopfn))
2468 return do_no_pfn(mm, vma, address, pte,
2469 pmd, write_access);
2470 }
2471 return do_anonymous_page(mm, vma, address,
2472 pte, pmd, write_access);
Hugh Dickins65500d22005-10-29 18:15:59 -07002473 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 if (pte_file(entry))
Nick Piggind0217ac2007-07-19 01:47:03 -07002475 return do_nonlinear_fault(mm, vma, address,
Hugh Dickins65500d22005-10-29 18:15:59 -07002476 pte, pmd, write_access, entry);
2477 return do_swap_page(mm, vma, address,
2478 pte, pmd, write_access, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 }
2480
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002481 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002482 spin_lock(ptl);
2483 if (unlikely(!pte_same(*pte, entry)))
2484 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 if (write_access) {
2486 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002487 return do_wp_page(mm, vma, address,
2488 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 entry = pte_mkdirty(entry);
2490 }
2491 entry = pte_mkyoung(entry);
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002492 if (ptep_set_access_flags(vma, address, pte, entry, write_access)) {
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002493 update_mmu_cache(vma, address, entry);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002494 } else {
2495 /*
2496 * This is needed only for protection faults but the arch code
2497 * is not yet telling us if this is a protection fault or not.
2498 * This still avoids useless tlb flushes for .text page faults
2499 * with threads.
2500 */
2501 if (write_access)
2502 flush_tlb_page(vma, address);
2503 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002504unlock:
2505 pte_unmap_unlock(pte, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002506 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507}
2508
2509/*
2510 * By the time we get here, we already hold the mm semaphore
2511 */
Nick Piggin83c54072007-07-19 01:47:05 -07002512int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513 unsigned long address, int write_access)
2514{
2515 pgd_t *pgd;
2516 pud_t *pud;
2517 pmd_t *pmd;
2518 pte_t *pte;
2519
2520 __set_current_state(TASK_RUNNING);
2521
Christoph Lameterf8891e52006-06-30 01:55:45 -07002522 count_vm_event(PGFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01002524 if (unlikely(is_vm_hugetlb_page(vma)))
2525 return hugetlb_fault(mm, vma, address, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 pud = pud_alloc(mm, pgd, address);
2529 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002530 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 pmd = pmd_alloc(mm, pud, address);
2532 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002533 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 pte = pte_alloc_map(mm, pmd, address);
2535 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002536 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537
Hugh Dickinsc74df322005-10-29 18:16:23 -07002538 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539}
2540
2541#ifndef __PAGETABLE_PUD_FOLDED
2542/*
2543 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002544 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002546int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002548 pud_t *new = pud_alloc_one(mm, address);
2549 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002550 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551
Hugh Dickins872fec12005-10-29 18:16:21 -07002552 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002553 if (pgd_present(*pgd)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002554 pud_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002555 else
2556 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07002557 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002558 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559}
2560#endif /* __PAGETABLE_PUD_FOLDED */
2561
2562#ifndef __PAGETABLE_PMD_FOLDED
2563/*
2564 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002565 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002567int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002569 pmd_t *new = pmd_alloc_one(mm, address);
2570 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002571 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572
Hugh Dickins872fec12005-10-29 18:16:21 -07002573 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07002575 if (pud_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002576 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002577 else
2578 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07002580 if (pgd_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002581 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002582 else
2583 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07002585 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002586 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587}
2588#endif /* __PAGETABLE_PMD_FOLDED */
2589
2590int make_pages_present(unsigned long addr, unsigned long end)
2591{
2592 int ret, len, write;
2593 struct vm_area_struct * vma;
2594
2595 vma = find_vma(current->mm, addr);
2596 if (!vma)
2597 return -1;
2598 write = (vma->vm_flags & VM_WRITE) != 0;
Eric Sesterhenn5bcb28b2006-03-26 18:30:52 +02002599 BUG_ON(addr >= end);
2600 BUG_ON(end > vma->vm_end);
Rolf Eike Beer68e116a2007-07-15 23:38:03 -07002601 len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602 ret = get_user_pages(current, current->mm, addr,
2603 len, write, 0, NULL, NULL);
2604 if (ret < 0)
2605 return ret;
2606 return ret == len ? 0 : -1;
2607}
2608
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609#if !defined(__HAVE_ARCH_GATE_AREA)
2610
2611#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07002612static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613
2614static int __init gate_vma_init(void)
2615{
2616 gate_vma.vm_mm = NULL;
2617 gate_vma.vm_start = FIXADDR_USER_START;
2618 gate_vma.vm_end = FIXADDR_USER_END;
Roland McGrathb6558c42007-01-26 00:56:47 -08002619 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
2620 gate_vma.vm_page_prot = __P101;
Roland McGrathf47aef52007-01-26 00:56:49 -08002621 /*
2622 * Make sure the vDSO gets into every core dump.
2623 * Dumping its contents makes post-mortem fully interpretable later
2624 * without matching up the same kernel and hardware config to see
2625 * what PC values meant.
2626 */
2627 gate_vma.vm_flags |= VM_ALWAYSDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 return 0;
2629}
2630__initcall(gate_vma_init);
2631#endif
2632
2633struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2634{
2635#ifdef AT_SYSINFO_EHDR
2636 return &gate_vma;
2637#else
2638 return NULL;
2639#endif
2640}
2641
2642int in_gate_area_no_task(unsigned long addr)
2643{
2644#ifdef AT_SYSINFO_EHDR
2645 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2646 return 1;
2647#endif
2648 return 0;
2649}
2650
2651#endif /* __HAVE_ARCH_GATE_AREA */
David Howells0ec76a12006-09-27 01:50:15 -07002652
2653/*
2654 * Access another process' address space.
2655 * Source/target buffer must be kernel space,
2656 * Do not walk the page table directly, use get_user_pages
2657 */
2658int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2659{
2660 struct mm_struct *mm;
2661 struct vm_area_struct *vma;
2662 struct page *page;
2663 void *old_buf = buf;
2664
2665 mm = get_task_mm(tsk);
2666 if (!mm)
2667 return 0;
2668
2669 down_read(&mm->mmap_sem);
Simon Arlott183ff222007-10-20 01:27:18 +02002670 /* ignore errors, just check how much was successfully transferred */
David Howells0ec76a12006-09-27 01:50:15 -07002671 while (len) {
2672 int bytes, ret, offset;
2673 void *maddr;
2674
2675 ret = get_user_pages(tsk, mm, addr, 1,
2676 write, 1, &page, &vma);
2677 if (ret <= 0)
2678 break;
2679
2680 bytes = len;
2681 offset = addr & (PAGE_SIZE-1);
2682 if (bytes > PAGE_SIZE-offset)
2683 bytes = PAGE_SIZE-offset;
2684
2685 maddr = kmap(page);
2686 if (write) {
2687 copy_to_user_page(vma, page, addr,
2688 maddr + offset, buf, bytes);
2689 set_page_dirty_lock(page);
2690 } else {
2691 copy_from_user_page(vma, page, addr,
2692 buf, maddr + offset, bytes);
2693 }
2694 kunmap(page);
2695 page_cache_release(page);
2696 len -= bytes;
2697 buf += bytes;
2698 addr += bytes;
2699 }
2700 up_read(&mm->mmap_sem);
2701 mmput(mm);
2702
2703 return buf - old_buf;
2704}
Andi Kleen03252912008-01-30 13:33:18 +01002705
2706/*
2707 * Print the name of a VMA.
2708 */
2709void print_vma_addr(char *prefix, unsigned long ip)
2710{
2711 struct mm_struct *mm = current->mm;
2712 struct vm_area_struct *vma;
2713
2714 down_read(&mm->mmap_sem);
2715 vma = find_vma(mm, ip);
2716 if (vma && vma->vm_file) {
2717 struct file *f = vma->vm_file;
2718 char *buf = (char *)__get_free_page(GFP_KERNEL);
2719 if (buf) {
2720 char *p, *s;
2721
2722 p = d_path(f->f_dentry, f->f_vfsmnt, buf, PAGE_SIZE);
2723 if (IS_ERR(p))
2724 p = "?";
2725 s = strrchr(p, '/');
2726 if (s)
2727 p = s+1;
2728 printk("%s%s[%lx+%lx]", prefix, p,
2729 vma->vm_start,
2730 vma->vm_end - vma->vm_start);
2731 free_page((unsigned long)buf);
2732 }
2733 }
2734 up_read(&current->mm->mmap_sem);
2735}