blob: 74839b3a39997ffeabf203d983453f42720dc88a [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>
50#include <linux/init.h>
51
52#include <asm/pgalloc.h>
53#include <asm/uaccess.h>
54#include <asm/tlb.h>
55#include <asm/tlbflush.h>
56#include <asm/pgtable.h>
57
58#include <linux/swapops.h>
59#include <linux/elf.h>
60
Andy Whitcroftd41dee32005-06-23 00:07:54 -070061#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -070062/* use the per-pgdat data instead for discontigmem - mbligh */
63unsigned long max_mapnr;
64struct page *mem_map;
65
66EXPORT_SYMBOL(max_mapnr);
67EXPORT_SYMBOL(mem_map);
68#endif
69
70unsigned long num_physpages;
71/*
72 * A number of key systems in x86 including ioremap() rely on the assumption
73 * that high_memory defines the upper bound on direct map memory, then end
74 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
75 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
76 * and ZONE_HIGHMEM.
77 */
78void * high_memory;
79unsigned long vmalloc_earlyreserve;
80
81EXPORT_SYMBOL(num_physpages);
82EXPORT_SYMBOL(high_memory);
83EXPORT_SYMBOL(vmalloc_earlyreserve);
84
85/*
86 * If a p?d_bad entry is found while walking page tables, report
87 * the error, before resetting entry to p?d_none. Usually (but
88 * very seldom) called out from the p?d_none_or_clear_bad macros.
89 */
90
91void pgd_clear_bad(pgd_t *pgd)
92{
93 pgd_ERROR(*pgd);
94 pgd_clear(pgd);
95}
96
97void pud_clear_bad(pud_t *pud)
98{
99 pud_ERROR(*pud);
100 pud_clear(pud);
101}
102
103void pmd_clear_bad(pmd_t *pmd)
104{
105 pmd_ERROR(*pmd);
106 pmd_clear(pmd);
107}
108
109/*
110 * Note: this doesn't free the actual pages themselves. That
111 * has been handled earlier when unmapping all the memory regions.
112 */
Hugh Dickinse0da3822005-04-19 13:29:15 -0700113static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114{
Hugh Dickinse0da3822005-04-19 13:29:15 -0700115 struct page *page = pmd_page(*pmd);
116 pmd_clear(pmd);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700117 pte_lock_deinit(page);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700118 pte_free_tlb(tlb, page);
119 dec_page_state(nr_page_table_pages);
120 tlb->mm->nr_ptes--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121}
122
Hugh Dickinse0da3822005-04-19 13:29:15 -0700123static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
124 unsigned long addr, unsigned long end,
125 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126{
127 pmd_t *pmd;
128 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700129 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130
Hugh Dickinse0da3822005-04-19 13:29:15 -0700131 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 do {
134 next = pmd_addr_end(addr, end);
135 if (pmd_none_or_clear_bad(pmd))
136 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700137 free_pte_range(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 } while (pmd++, addr = next, addr != end);
139
Hugh Dickinse0da3822005-04-19 13:29:15 -0700140 start &= PUD_MASK;
141 if (start < floor)
142 return;
143 if (ceiling) {
144 ceiling &= PUD_MASK;
145 if (!ceiling)
146 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700148 if (end - 1 > ceiling - 1)
149 return;
150
151 pmd = pmd_offset(pud, start);
152 pud_clear(pud);
153 pmd_free_tlb(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154}
155
Hugh Dickinse0da3822005-04-19 13:29:15 -0700156static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
157 unsigned long addr, unsigned long end,
158 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159{
160 pud_t *pud;
161 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700162 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163
Hugh Dickinse0da3822005-04-19 13:29:15 -0700164 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 pud = pud_offset(pgd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 do {
167 next = pud_addr_end(addr, end);
168 if (pud_none_or_clear_bad(pud))
169 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700170 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 } while (pud++, addr = next, addr != end);
172
Hugh Dickinse0da3822005-04-19 13:29:15 -0700173 start &= PGDIR_MASK;
174 if (start < floor)
175 return;
176 if (ceiling) {
177 ceiling &= PGDIR_MASK;
178 if (!ceiling)
179 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700181 if (end - 1 > ceiling - 1)
182 return;
183
184 pud = pud_offset(pgd, start);
185 pgd_clear(pgd);
186 pud_free_tlb(tlb, pud);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187}
188
189/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700190 * This function frees user-level page tables of a process.
191 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 * Must be called with pagetable lock held.
193 */
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700194void free_pgd_range(struct mmu_gather **tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700195 unsigned long addr, unsigned long end,
196 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197{
198 pgd_t *pgd;
199 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700200 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201
Hugh Dickinse0da3822005-04-19 13:29:15 -0700202 /*
203 * The next few lines have given us lots of grief...
204 *
205 * Why are we testing PMD* at this top level? Because often
206 * there will be no work to do at all, and we'd prefer not to
207 * go all the way down to the bottom just to discover that.
208 *
209 * Why all these "- 1"s? Because 0 represents both the bottom
210 * of the address space and the top of it (using -1 for the
211 * top wouldn't help much: the masks would do the wrong thing).
212 * The rule is that addr 0 and floor 0 refer to the bottom of
213 * the address space, but end 0 and ceiling 0 refer to the top
214 * Comparisons need to use "end - 1" and "ceiling - 1" (though
215 * that end 0 case should be mythical).
216 *
217 * Wherever addr is brought up or ceiling brought down, we must
218 * be careful to reject "the opposite 0" before it confuses the
219 * subsequent tests. But what about where end is brought down
220 * by PMD_SIZE below? no, end can't go down to 0 there.
221 *
222 * Whereas we round start (addr) and ceiling down, by different
223 * masks at different levels, in order to test whether a table
224 * now has no other vmas using it, so can be freed, we don't
225 * bother to round floor or end up - the tests don't need that.
226 */
227
228 addr &= PMD_MASK;
229 if (addr < floor) {
230 addr += PMD_SIZE;
231 if (!addr)
232 return;
233 }
234 if (ceiling) {
235 ceiling &= PMD_MASK;
236 if (!ceiling)
237 return;
238 }
239 if (end - 1 > ceiling - 1)
240 end -= PMD_SIZE;
241 if (addr > end - 1)
242 return;
243
244 start = addr;
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700245 pgd = pgd_offset((*tlb)->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 do {
247 next = pgd_addr_end(addr, end);
248 if (pgd_none_or_clear_bad(pgd))
249 continue;
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700250 free_pud_range(*tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700252
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700253 if (!(*tlb)->fullmm)
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700254 flush_tlb_pgtables((*tlb)->mm, start, end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700255}
256
257void free_pgtables(struct mmu_gather **tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700258 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700259{
260 while (vma) {
261 struct vm_area_struct *next = vma->vm_next;
262 unsigned long addr = vma->vm_start;
263
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700264 /*
265 * Hide vma from rmap and vmtruncate before freeing pgtables
266 */
267 anon_vma_unlink(vma);
268 unlink_file_vma(vma);
269
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700270 if (is_hugepage_only_range(vma->vm_mm, addr, HPAGE_SIZE)) {
271 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700272 floor, next? next->vm_start: ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700273 } else {
274 /*
275 * Optimization: gather nearby vmas into one call down
276 */
277 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
278 && !is_hugepage_only_range(vma->vm_mm, next->vm_start,
279 HPAGE_SIZE)) {
280 vma = next;
281 next = vma->vm_next;
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700282 anon_vma_unlink(vma);
283 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700284 }
285 free_pgd_range(tlb, addr, vma->vm_end,
286 floor, next? next->vm_start: ceiling);
287 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700288 vma = next;
289 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290}
291
Hugh Dickins1bb36302005-10-29 18:16:22 -0700292int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293{
Hugh Dickinsc74df322005-10-29 18:16:23 -0700294 struct page *new = pte_alloc_one(mm, address);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700295 if (!new)
296 return -ENOMEM;
297
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700298 pte_lock_init(new);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700299 spin_lock(&mm->page_table_lock);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700300 if (pmd_present(*pmd)) { /* Another has populated it */
301 pte_lock_deinit(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700302 pte_free(new);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700303 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 mm->nr_ptes++;
305 inc_page_state(nr_page_table_pages);
306 pmd_populate(mm, pmd, new);
307 }
Hugh Dickinsc74df322005-10-29 18:16:23 -0700308 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700309 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310}
311
Hugh Dickins1bb36302005-10-29 18:16:22 -0700312int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700314 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
315 if (!new)
316 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317
Hugh Dickins1bb36302005-10-29 18:16:22 -0700318 spin_lock(&init_mm.page_table_lock);
319 if (pmd_present(*pmd)) /* Another has populated it */
320 pte_free_kernel(new);
321 else
322 pmd_populate_kernel(&init_mm, pmd, new);
323 spin_unlock(&init_mm.page_table_lock);
324 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325}
326
Hugh Dickinsae859762005-10-29 18:16:05 -0700327static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
328{
329 if (file_rss)
330 add_mm_counter(mm, file_rss, file_rss);
331 if (anon_rss)
332 add_mm_counter(mm, anon_rss, anon_rss);
333}
334
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800336 * This function is called to print an error when a bad pte
337 * is found. For example, we might have a PFN-mapped pte in
338 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700339 *
340 * The calling function must still handle the error.
341 */
342void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr)
343{
344 printk(KERN_ERR "Bad pte = %08llx, process = %s, "
345 "vm_flags = %lx, vaddr = %lx\n",
346 (long long)pte_val(pte),
347 (vma->vm_mm == current->mm ? current->comm : "???"),
348 vma->vm_flags, vaddr);
349 dump_stack();
350}
351
352/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800353 * This function gets the "struct page" associated with a pte.
354 *
355 * NOTE! Some mappings do not have "struct pages". A raw PFN mapping
356 * will have each page table entry just pointing to a raw page frame
357 * number, and as far as the VM layer is concerned, those do not have
358 * pages associated with them - even if the PFN might point to memory
359 * that otherwise is perfectly fine and has a "struct page".
360 *
361 * The way we recognize those mappings is through the rules set up
362 * by "remap_pfn_range()": the vma will have the VM_PFNMAP bit set,
363 * and the vm_pgoff will point to the first PFN mapped: thus every
364 * page that is a raw mapping will always honor the rule
365 *
366 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
367 *
368 * and if that isn't true, the page has been COW'ed (in which case it
369 * _does_ have a "struct page" associated with it even if it is in a
370 * VM_PFNMAP range).
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800371 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800372struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800373{
Linus Torvalds6aab3412005-11-28 14:34:23 -0800374 unsigned long pfn = pte_pfn(pte);
375
376 if (vma->vm_flags & VM_PFNMAP) {
377 unsigned long off = (addr - vma->vm_start) >> PAGE_SHIFT;
378 if (pfn == vma->vm_pgoff + off)
379 return NULL;
380 }
381
382 /*
383 * Add some anal sanity checks for now. Eventually,
384 * we should just do "return pfn_to_page(pfn)", but
385 * in the meantime we check that we get a valid pfn,
386 * and that the resulting page looks ok.
387 *
388 * Remove this test eventually!
389 */
390 if (unlikely(!pfn_valid(pfn))) {
391 print_bad_pte(vma, pte, addr);
392 return NULL;
393 }
394
395 /*
396 * NOTE! We still have PageReserved() pages in the page
397 * tables.
398 *
399 * The PAGE_ZERO() pages and various VDSO mappings can
400 * cause them to exist.
401 */
402 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800403}
404
405/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 * copy one vm_area from one task to the other. Assumes the page tables
407 * already present in the new task to be cleared in the whole range
408 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 */
410
Hugh Dickins8c103762005-10-29 18:16:13 -0700411static inline void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700413 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700414 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415{
Nick Pigginb5810032005-10-29 18:16:12 -0700416 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 pte_t pte = *src_pte;
418 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419
420 /* pte contains position in swap or file, so copy. */
421 if (unlikely(!pte_present(pte))) {
422 if (!pte_file(pte)) {
423 swap_duplicate(pte_to_swp_entry(pte));
424 /* make sure dst_mm is on swapoff's mmlist. */
425 if (unlikely(list_empty(&dst_mm->mmlist))) {
426 spin_lock(&mmlist_lock);
Hugh Dickinsf412ac02005-10-29 18:16:41 -0700427 if (list_empty(&dst_mm->mmlist))
428 list_add(&dst_mm->mmlist,
429 &src_mm->mmlist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 spin_unlock(&mmlist_lock);
431 }
432 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700433 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 }
435
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 /*
437 * If it's a COW mapping, write protect it both
438 * in the parent and the child
439 */
440 if ((vm_flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE) {
441 ptep_set_wrprotect(src_mm, addr, src_pte);
442 pte = *src_pte;
443 }
444
445 /*
446 * If it's a shared mapping, mark it clean in
447 * the child
448 */
449 if (vm_flags & VM_SHARED)
450 pte = pte_mkclean(pte);
451 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800452
453 page = vm_normal_page(vma, addr, pte);
454 if (page) {
455 get_page(page);
456 page_dup_rmap(page);
457 rss[!!PageAnon(page)]++;
458 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700459
460out_set_pte:
461 set_pte_at(dst_mm, addr, dst_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462}
463
464static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
465 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
466 unsigned long addr, unsigned long end)
467{
468 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700469 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -0700470 int progress = 0;
Hugh Dickins8c103762005-10-29 18:16:13 -0700471 int rss[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
473again:
Hugh Dickinsae859762005-10-29 18:16:05 -0700474 rss[1] = rss[0] = 0;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700475 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 if (!dst_pte)
477 return -ENOMEM;
478 src_pte = pte_offset_map_nested(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700479 src_ptl = pte_lockptr(src_mm, src_pmd);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700480 spin_lock(src_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 do {
483 /*
484 * We are holding two locks at this point - either of them
485 * could generate latencies in another task on another CPU.
486 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700487 if (progress >= 32) {
488 progress = 0;
489 if (need_resched() ||
Hugh Dickinsc74df322005-10-29 18:16:23 -0700490 need_lockbreak(src_ptl) ||
491 need_lockbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700492 break;
493 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 if (pte_none(*src_pte)) {
495 progress++;
496 continue;
497 }
Hugh Dickins8c103762005-10-29 18:16:13 -0700498 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 progress += 8;
500 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501
Hugh Dickinsc74df322005-10-29 18:16:23 -0700502 spin_unlock(src_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 pte_unmap_nested(src_pte - 1);
Hugh Dickinsae859762005-10-29 18:16:05 -0700504 add_mm_rss(dst_mm, rss[0], rss[1]);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700505 pte_unmap_unlock(dst_pte - 1, dst_ptl);
506 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 if (addr != end)
508 goto again;
509 return 0;
510}
511
512static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
513 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
514 unsigned long addr, unsigned long end)
515{
516 pmd_t *src_pmd, *dst_pmd;
517 unsigned long next;
518
519 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
520 if (!dst_pmd)
521 return -ENOMEM;
522 src_pmd = pmd_offset(src_pud, addr);
523 do {
524 next = pmd_addr_end(addr, end);
525 if (pmd_none_or_clear_bad(src_pmd))
526 continue;
527 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
528 vma, addr, next))
529 return -ENOMEM;
530 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
531 return 0;
532}
533
534static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
535 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
536 unsigned long addr, unsigned long end)
537{
538 pud_t *src_pud, *dst_pud;
539 unsigned long next;
540
541 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
542 if (!dst_pud)
543 return -ENOMEM;
544 src_pud = pud_offset(src_pgd, addr);
545 do {
546 next = pud_addr_end(addr, end);
547 if (pud_none_or_clear_bad(src_pud))
548 continue;
549 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
550 vma, addr, next))
551 return -ENOMEM;
552 } while (dst_pud++, src_pud++, addr = next, addr != end);
553 return 0;
554}
555
556int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
557 struct vm_area_struct *vma)
558{
559 pgd_t *src_pgd, *dst_pgd;
560 unsigned long next;
561 unsigned long addr = vma->vm_start;
562 unsigned long end = vma->vm_end;
563
Nick Piggind9928952005-08-28 16:49:11 +1000564 /*
565 * Don't copy ptes where a page fault will fill them correctly.
566 * Fork becomes much lighter when there are big shared or private
567 * readonly mappings. The tradeoff is that copy_page_range is more
568 * efficient than faulting.
569 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800570 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP))) {
Nick Piggind9928952005-08-28 16:49:11 +1000571 if (!vma->anon_vma)
572 return 0;
573 }
574
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 if (is_vm_hugetlb_page(vma))
576 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
577
578 dst_pgd = pgd_offset(dst_mm, addr);
579 src_pgd = pgd_offset(src_mm, addr);
580 do {
581 next = pgd_addr_end(addr, end);
582 if (pgd_none_or_clear_bad(src_pgd))
583 continue;
584 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
585 vma, addr, next))
586 return -ENOMEM;
587 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
588 return 0;
589}
590
Robin Holt51c6f662005-11-13 16:06:42 -0800591static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700592 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800594 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595{
Nick Pigginb5810032005-10-29 18:16:12 -0700596 struct mm_struct *mm = tlb->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 pte_t *pte;
Hugh Dickins508034a2005-10-29 18:16:30 -0700598 spinlock_t *ptl;
Hugh Dickinsae859762005-10-29 18:16:05 -0700599 int file_rss = 0;
600 int anon_rss = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601
Hugh Dickins508034a2005-10-29 18:16:30 -0700602 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 do {
604 pte_t ptent = *pte;
Robin Holt51c6f662005-11-13 16:06:42 -0800605 if (pte_none(ptent)) {
606 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800608 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800610 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -0800611
612 (*zap_work) -= PAGE_SIZE;
613
Linus Torvalds6aab3412005-11-28 14:34:23 -0800614 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 if (unlikely(details) && page) {
616 /*
617 * unmap_shared_mapping_pages() wants to
618 * invalidate cache without truncating:
619 * unmap shared but keep private pages.
620 */
621 if (details->check_mapping &&
622 details->check_mapping != page->mapping)
623 continue;
624 /*
625 * Each page->index must be checked when
626 * invalidating or truncating nonlinear.
627 */
628 if (details->nonlinear_vma &&
629 (page->index < details->first_index ||
630 page->index > details->last_index))
631 continue;
632 }
Nick Pigginb5810032005-10-29 18:16:12 -0700633 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -0700634 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 tlb_remove_tlb_entry(tlb, pte, addr);
636 if (unlikely(!page))
637 continue;
638 if (unlikely(details) && details->nonlinear_vma
639 && linear_page_index(details->nonlinear_vma,
640 addr) != page->index)
Nick Pigginb5810032005-10-29 18:16:12 -0700641 set_pte_at(mm, addr, pte,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 pgoff_to_pte(page->index));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 if (PageAnon(page))
Hugh Dickins86d912f2005-10-29 18:16:14 -0700644 anon_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700645 else {
646 if (pte_dirty(ptent))
647 set_page_dirty(page);
648 if (pte_young(ptent))
649 mark_page_accessed(page);
Hugh Dickins86d912f2005-10-29 18:16:14 -0700650 file_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700651 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 page_remove_rmap(page);
653 tlb_remove_page(tlb, page);
654 continue;
655 }
656 /*
657 * If details->check_mapping, we leave swap entries;
658 * if details->nonlinear_vma, we leave file entries.
659 */
660 if (unlikely(details))
661 continue;
662 if (!pte_file(ptent))
663 free_swap_and_cache(pte_to_swp_entry(ptent));
Nick Pigginb5810032005-10-29 18:16:12 -0700664 pte_clear_full(mm, addr, pte, tlb->fullmm);
Robin Holt51c6f662005-11-13 16:06:42 -0800665 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
Hugh Dickinsae859762005-10-29 18:16:05 -0700666
Hugh Dickins86d912f2005-10-29 18:16:14 -0700667 add_mm_rss(mm, file_rss, anon_rss);
Hugh Dickins508034a2005-10-29 18:16:30 -0700668 pte_unmap_unlock(pte - 1, ptl);
Robin Holt51c6f662005-11-13 16:06:42 -0800669
670 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671}
672
Robin Holt51c6f662005-11-13 16:06:42 -0800673static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700674 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800676 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677{
678 pmd_t *pmd;
679 unsigned long next;
680
681 pmd = pmd_offset(pud, addr);
682 do {
683 next = pmd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800684 if (pmd_none_or_clear_bad(pmd)) {
685 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800687 }
688 next = zap_pte_range(tlb, vma, pmd, addr, next,
689 zap_work, details);
690 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
691
692 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693}
694
Robin Holt51c6f662005-11-13 16:06:42 -0800695static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700696 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800698 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699{
700 pud_t *pud;
701 unsigned long next;
702
703 pud = pud_offset(pgd, addr);
704 do {
705 next = pud_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800706 if (pud_none_or_clear_bad(pud)) {
707 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800709 }
710 next = zap_pmd_range(tlb, vma, pud, addr, next,
711 zap_work, details);
712 } while (pud++, addr = next, (addr != end && *zap_work > 0));
713
714 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715}
716
Robin Holt51c6f662005-11-13 16:06:42 -0800717static unsigned long unmap_page_range(struct mmu_gather *tlb,
718 struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800720 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721{
722 pgd_t *pgd;
723 unsigned long next;
724
725 if (details && !details->check_mapping && !details->nonlinear_vma)
726 details = NULL;
727
728 BUG_ON(addr >= end);
729 tlb_start_vma(tlb, vma);
730 pgd = pgd_offset(vma->vm_mm, addr);
731 do {
732 next = pgd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800733 if (pgd_none_or_clear_bad(pgd)) {
734 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800736 }
737 next = zap_pud_range(tlb, vma, pgd, addr, next,
738 zap_work, details);
739 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 tlb_end_vma(tlb, vma);
Robin Holt51c6f662005-11-13 16:06:42 -0800741
742 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743}
744
745#ifdef CONFIG_PREEMPT
746# define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
747#else
748/* No preempt: go for improved straight-line efficiency */
749# define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
750#endif
751
752/**
753 * unmap_vmas - unmap a range of memory covered by a list of vma's
754 * @tlbp: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 * @vma: the starting vma
756 * @start_addr: virtual address at which to start unmapping
757 * @end_addr: virtual address at which to end unmapping
758 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
759 * @details: details of nonlinear truncation or shared cache invalidation
760 *
Hugh Dickinsee39b372005-04-19 13:29:15 -0700761 * Returns the end address of the unmapping (restart addr if interrupted).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700763 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700765 * We aim to not hold locks for too long (for scheduling latency reasons).
766 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 * return the ending mmu_gather to the caller.
768 *
769 * Only addresses between `start' and `end' will be unmapped.
770 *
771 * The VMA list must be sorted in ascending virtual address order.
772 *
773 * unmap_vmas() assumes that the caller will flush the whole unmapped address
774 * range after unmap_vmas() returns. So the only responsibility here is to
775 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
776 * drops the lock and schedules.
777 */
Hugh Dickins508034a2005-10-29 18:16:30 -0700778unsigned long unmap_vmas(struct mmu_gather **tlbp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 struct vm_area_struct *vma, unsigned long start_addr,
780 unsigned long end_addr, unsigned long *nr_accounted,
781 struct zap_details *details)
782{
Robin Holt51c6f662005-11-13 16:06:42 -0800783 long zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
785 int tlb_start_valid = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -0700786 unsigned long start = start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700788 int fullmm = (*tlbp)->fullmm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789
790 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 unsigned long end;
792
793 start = max(vma->vm_start, start_addr);
794 if (start >= vma->vm_end)
795 continue;
796 end = min(vma->vm_end, end_addr);
797 if (end <= vma->vm_start)
798 continue;
799
800 if (vma->vm_flags & VM_ACCOUNT)
801 *nr_accounted += (end - start) >> PAGE_SHIFT;
802
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 while (start != end) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 if (!tlb_start_valid) {
805 tlb_start = start;
806 tlb_start_valid = 1;
807 }
808
Robin Holt51c6f662005-11-13 16:06:42 -0800809 if (unlikely(is_vm_hugetlb_page(vma))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 unmap_hugepage_range(vma, start, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800811 zap_work -= (end - start) /
812 (HPAGE_SIZE / PAGE_SIZE);
813 start = end;
814 } else
815 start = unmap_page_range(*tlbp, vma,
816 start, end, &zap_work, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817
Robin Holt51c6f662005-11-13 16:06:42 -0800818 if (zap_work > 0) {
819 BUG_ON(start != end);
820 break;
821 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822
823 tlb_finish_mmu(*tlbp, tlb_start, start);
824
825 if (need_resched() ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 (i_mmap_lock && need_lockbreak(i_mmap_lock))) {
827 if (i_mmap_lock) {
Hugh Dickins508034a2005-10-29 18:16:30 -0700828 *tlbp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 goto out;
830 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 }
833
Hugh Dickins508034a2005-10-29 18:16:30 -0700834 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 tlb_start_valid = 0;
Robin Holt51c6f662005-11-13 16:06:42 -0800836 zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 }
838 }
839out:
Hugh Dickinsee39b372005-04-19 13:29:15 -0700840 return start; /* which is now the end (or restart) address */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841}
842
843/**
844 * zap_page_range - remove user pages in a given range
845 * @vma: vm_area_struct holding the applicable pages
846 * @address: starting address of pages to zap
847 * @size: number of bytes to zap
848 * @details: details of nonlinear truncation or shared cache invalidation
849 */
Hugh Dickinsee39b372005-04-19 13:29:15 -0700850unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 unsigned long size, struct zap_details *details)
852{
853 struct mm_struct *mm = vma->vm_mm;
854 struct mmu_gather *tlb;
855 unsigned long end = address + size;
856 unsigned long nr_accounted = 0;
857
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -0700860 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -0700861 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
862 if (tlb)
863 tlb_finish_mmu(tlb, address, end);
Hugh Dickinsee39b372005-04-19 13:29:15 -0700864 return end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865}
866
867/*
868 * Do a quick page-table lookup for a single page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800870struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700871 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872{
873 pgd_t *pgd;
874 pud_t *pud;
875 pmd_t *pmd;
876 pte_t *ptep, pte;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700877 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 struct page *page;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800879 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700881 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
882 if (!IS_ERR(page)) {
883 BUG_ON(flags & FOLL_GET);
884 goto out;
885 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700887 page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 pgd = pgd_offset(mm, address);
889 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700890 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891
892 pud = pud_offset(pgd, address);
893 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700894 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895
896 pmd = pmd_offset(pud, address);
897 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700898 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700900 if (pmd_huge(*pmd)) {
901 BUG_ON(flags & FOLL_GET);
902 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
903 goto out;
904 }
905
906 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 if (!ptep)
908 goto out;
909
910 pte = *ptep;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700911 if (!pte_present(pte))
912 goto unlock;
913 if ((flags & FOLL_WRITE) && !pte_write(pte))
914 goto unlock;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800915 page = vm_normal_page(vma, address, pte);
916 if (unlikely(!page))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700917 goto unlock;
918
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700919 if (flags & FOLL_GET)
920 get_page(page);
921 if (flags & FOLL_TOUCH) {
922 if ((flags & FOLL_WRITE) &&
923 !pte_dirty(pte) && !PageDirty(page))
924 set_page_dirty(page);
925 mark_page_accessed(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700927unlock:
928 pte_unmap_unlock(ptep, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929out:
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700930 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700932no_page_table:
933 /*
934 * When core dumping an enormous anonymous area that nobody
935 * has touched so far, we don't want to allocate page tables.
936 */
937 if (flags & FOLL_ANON) {
938 page = ZERO_PAGE(address);
939 if (flags & FOLL_GET)
940 get_page(page);
941 BUG_ON(flags & FOLL_WRITE);
942 }
943 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944}
945
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
947 unsigned long start, int len, int write, int force,
948 struct page **pages, struct vm_area_struct **vmas)
949{
950 int i;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700951 unsigned int vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952
953 /*
954 * Require read or write permissions.
955 * If 'force' is set, we only require the "MAY" flags.
956 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700957 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
958 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 i = 0;
960
961 do {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700962 struct vm_area_struct *vma;
963 unsigned int foll_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964
965 vma = find_extend_vma(mm, start);
966 if (!vma && in_gate_area(tsk, start)) {
967 unsigned long pg = start & PAGE_MASK;
968 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
969 pgd_t *pgd;
970 pud_t *pud;
971 pmd_t *pmd;
972 pte_t *pte;
973 if (write) /* user gate pages are read-only */
974 return i ? : -EFAULT;
975 if (pg > TASK_SIZE)
976 pgd = pgd_offset_k(pg);
977 else
978 pgd = pgd_offset_gate(mm, pg);
979 BUG_ON(pgd_none(*pgd));
980 pud = pud_offset(pgd, pg);
981 BUG_ON(pud_none(*pud));
982 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -0700983 if (pmd_none(*pmd))
984 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -0700986 if (pte_none(*pte)) {
987 pte_unmap(pte);
988 return i ? : -EFAULT;
989 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 if (pages) {
Nick Pigginfa2a4552005-11-29 18:43:17 +1100991 struct page *page = vm_normal_page(gate_vma, start, *pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800992 pages[i] = page;
993 if (page)
994 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 }
996 pte_unmap(pte);
997 if (vmas)
998 vmas[i] = gate_vma;
999 i++;
1000 start += PAGE_SIZE;
1001 len--;
1002 continue;
1003 }
1004
Hugh Dickinsed5297a2005-11-21 21:32:11 -08001005 if (!vma || (vma->vm_flags & VM_IO)
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001006 || !(vm_flags & vma->vm_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 return i ? : -EFAULT;
1008
1009 if (is_vm_hugetlb_page(vma)) {
1010 i = follow_hugetlb_page(mm, vma, pages, vmas,
1011 &start, &len, i);
1012 continue;
1013 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001014
1015 foll_flags = FOLL_TOUCH;
1016 if (pages)
1017 foll_flags |= FOLL_GET;
1018 if (!write && !(vma->vm_flags & VM_LOCKED) &&
1019 (!vma->vm_ops || !vma->vm_ops->nopage))
1020 foll_flags |= FOLL_ANON;
1021
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 do {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001023 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001025 if (write)
1026 foll_flags |= FOLL_WRITE;
1027
1028 cond_resched();
Linus Torvalds6aab3412005-11-28 14:34:23 -08001029 while (!(page = follow_page(vma, start, foll_flags))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001030 int ret;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001031 ret = __handle_mm_fault(mm, vma, start,
1032 foll_flags & FOLL_WRITE);
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001033 /*
1034 * The VM_FAULT_WRITE bit tells us that do_wp_page has
1035 * broken COW when necessary, even if maybe_mkwrite
1036 * decided not to set pte_write. We can thus safely do
1037 * subsequent page lookups as if they were reads.
1038 */
1039 if (ret & VM_FAULT_WRITE)
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001040 foll_flags &= ~FOLL_WRITE;
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001041
1042 switch (ret & ~VM_FAULT_WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 case VM_FAULT_MINOR:
1044 tsk->min_flt++;
1045 break;
1046 case VM_FAULT_MAJOR:
1047 tsk->maj_flt++;
1048 break;
1049 case VM_FAULT_SIGBUS:
1050 return i ? i : -EFAULT;
1051 case VM_FAULT_OOM:
1052 return i ? i : -ENOMEM;
1053 default:
1054 BUG();
1055 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 }
1057 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001058 pages[i] = page;
1059 flush_dcache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 }
1061 if (vmas)
1062 vmas[i] = vma;
1063 i++;
1064 start += PAGE_SIZE;
1065 len--;
Hugh Dickins08ef4722005-06-21 17:15:10 -07001066 } while (len && start < vma->vm_end);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001067 } while (len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 return i;
1069}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070EXPORT_SYMBOL(get_user_pages);
1071
1072static int zeromap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1073 unsigned long addr, unsigned long end, pgprot_t prot)
1074{
1075 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001076 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077
Hugh Dickinsc74df322005-10-29 18:16:23 -07001078 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 if (!pte)
1080 return -ENOMEM;
1081 do {
Nick Pigginb5810032005-10-29 18:16:12 -07001082 struct page *page = ZERO_PAGE(addr);
1083 pte_t zero_pte = pte_wrprotect(mk_pte(page, prot));
1084 page_cache_get(page);
1085 page_add_file_rmap(page);
1086 inc_mm_counter(mm, file_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 BUG_ON(!pte_none(*pte));
1088 set_pte_at(mm, addr, pte, zero_pte);
1089 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickinsc74df322005-10-29 18:16:23 -07001090 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 return 0;
1092}
1093
1094static inline int zeromap_pmd_range(struct mm_struct *mm, pud_t *pud,
1095 unsigned long addr, unsigned long end, pgprot_t prot)
1096{
1097 pmd_t *pmd;
1098 unsigned long next;
1099
1100 pmd = pmd_alloc(mm, pud, addr);
1101 if (!pmd)
1102 return -ENOMEM;
1103 do {
1104 next = pmd_addr_end(addr, end);
1105 if (zeromap_pte_range(mm, pmd, addr, next, prot))
1106 return -ENOMEM;
1107 } while (pmd++, addr = next, addr != end);
1108 return 0;
1109}
1110
1111static inline int zeromap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1112 unsigned long addr, unsigned long end, pgprot_t prot)
1113{
1114 pud_t *pud;
1115 unsigned long next;
1116
1117 pud = pud_alloc(mm, pgd, addr);
1118 if (!pud)
1119 return -ENOMEM;
1120 do {
1121 next = pud_addr_end(addr, end);
1122 if (zeromap_pmd_range(mm, pud, addr, next, prot))
1123 return -ENOMEM;
1124 } while (pud++, addr = next, addr != end);
1125 return 0;
1126}
1127
1128int zeromap_page_range(struct vm_area_struct *vma,
1129 unsigned long addr, unsigned long size, pgprot_t prot)
1130{
1131 pgd_t *pgd;
1132 unsigned long next;
1133 unsigned long end = addr + size;
1134 struct mm_struct *mm = vma->vm_mm;
1135 int err;
1136
1137 BUG_ON(addr >= end);
1138 pgd = pgd_offset(mm, addr);
1139 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 do {
1141 next = pgd_addr_end(addr, end);
1142 err = zeromap_pud_range(mm, pgd, addr, next, prot);
1143 if (err)
1144 break;
1145 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 return err;
1147}
1148
1149/*
1150 * maps a range of physical memory into the requested pages. the old
1151 * mappings are removed. any references to nonexistent pages results
1152 * in null mappings (currently treated as "copy-on-access")
1153 */
1154static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1155 unsigned long addr, unsigned long end,
1156 unsigned long pfn, pgprot_t prot)
1157{
1158 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001159 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160
Hugh Dickinsc74df322005-10-29 18:16:23 -07001161 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 if (!pte)
1163 return -ENOMEM;
1164 do {
1165 BUG_ON(!pte_none(*pte));
Nick Pigginb5810032005-10-29 18:16:12 -07001166 set_pte_at(mm, addr, pte, pfn_pte(pfn, prot));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 pfn++;
1168 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickinsc74df322005-10-29 18:16:23 -07001169 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 return 0;
1171}
1172
1173static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1174 unsigned long addr, unsigned long end,
1175 unsigned long pfn, pgprot_t prot)
1176{
1177 pmd_t *pmd;
1178 unsigned long next;
1179
1180 pfn -= addr >> PAGE_SHIFT;
1181 pmd = pmd_alloc(mm, pud, addr);
1182 if (!pmd)
1183 return -ENOMEM;
1184 do {
1185 next = pmd_addr_end(addr, end);
1186 if (remap_pte_range(mm, pmd, addr, next,
1187 pfn + (addr >> PAGE_SHIFT), prot))
1188 return -ENOMEM;
1189 } while (pmd++, addr = next, addr != end);
1190 return 0;
1191}
1192
1193static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1194 unsigned long addr, unsigned long end,
1195 unsigned long pfn, pgprot_t prot)
1196{
1197 pud_t *pud;
1198 unsigned long next;
1199
1200 pfn -= addr >> PAGE_SHIFT;
1201 pud = pud_alloc(mm, pgd, addr);
1202 if (!pud)
1203 return -ENOMEM;
1204 do {
1205 next = pud_addr_end(addr, end);
1206 if (remap_pmd_range(mm, pud, addr, next,
1207 pfn + (addr >> PAGE_SHIFT), prot))
1208 return -ENOMEM;
1209 } while (pud++, addr = next, addr != end);
1210 return 0;
1211}
1212
1213/* Note: this is only safe if the mm semaphore is held when called. */
1214int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1215 unsigned long pfn, unsigned long size, pgprot_t prot)
1216{
1217 pgd_t *pgd;
1218 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001219 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 struct mm_struct *mm = vma->vm_mm;
1221 int err;
1222
1223 /*
1224 * Physically remapped pages are special. Tell the
1225 * rest of the world about it:
1226 * VM_IO tells people not to look at these pages
1227 * (accesses can have side effects).
Hugh Dickins0b14c172005-11-21 21:32:15 -08001228 * VM_RESERVED is specified all over the place, because
1229 * in 2.4 it kept swapout's vma scan off this vma; but
1230 * in 2.6 the LRU scan won't even find its pages, so this
1231 * flag means no more than count its pages in reserved_vm,
1232 * and omit it from core dump, even when VM_IO turned off.
Linus Torvalds6aab3412005-11-28 14:34:23 -08001233 * VM_PFNMAP tells the core MM that the base pages are just
1234 * raw PFN mappings, and do not have a "struct page" associated
1235 * with them.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 */
Linus Torvalds6aab3412005-11-28 14:34:23 -08001237 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
1238 vma->vm_pgoff = pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239
1240 BUG_ON(addr >= end);
1241 pfn -= addr >> PAGE_SHIFT;
1242 pgd = pgd_offset(mm, addr);
1243 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 do {
1245 next = pgd_addr_end(addr, end);
1246 err = remap_pud_range(mm, pgd, addr, next,
1247 pfn + (addr >> PAGE_SHIFT), prot);
1248 if (err)
1249 break;
1250 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 return err;
1252}
1253EXPORT_SYMBOL(remap_pfn_range);
1254
1255/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001256 * handle_pte_fault chooses page fault handler according to an entry
1257 * which was read non-atomically. Before making any commitment, on
1258 * those architectures or configurations (e.g. i386 with PAE) which
1259 * might give a mix of unmatched parts, do_swap_page and do_file_page
1260 * must check under lock before unmapping the pte and proceeding
1261 * (but do_wp_page is only called after already making such a check;
1262 * and do_anonymous_page and do_no_page can safely check later on).
1263 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001264static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001265 pte_t *page_table, pte_t orig_pte)
1266{
1267 int same = 1;
1268#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1269 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001270 spinlock_t *ptl = pte_lockptr(mm, pmd);
1271 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001272 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001273 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001274 }
1275#endif
1276 pte_unmap(page_table);
1277 return same;
1278}
1279
1280/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1282 * servicing faults for write access. In the normal case, do always want
1283 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1284 * that do not have writing enabled, when used by access_process_vm.
1285 */
1286static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1287{
1288 if (likely(vma->vm_flags & VM_WRITE))
1289 pte = pte_mkwrite(pte);
1290 return pte;
1291}
1292
Linus Torvalds6aab3412005-11-28 14:34:23 -08001293static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va)
1294{
1295 /*
1296 * If the source page was a PFN mapping, we don't have
1297 * a "struct page" for it. We do a best-effort copy by
1298 * just copying from the original user address. If that
1299 * fails, we just zero-fill it. Live with it.
1300 */
1301 if (unlikely(!src)) {
1302 void *kaddr = kmap_atomic(dst, KM_USER0);
1303 unsigned long left = __copy_from_user_inatomic(kaddr, (void __user *)va, PAGE_SIZE);
1304 if (left)
1305 memset(kaddr, 0, PAGE_SIZE);
1306 kunmap_atomic(kaddr, KM_USER0);
1307 return;
1308
1309 }
1310 copy_user_highpage(dst, src, va);
1311}
1312
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 * This routine handles present pages, when users try to write
1315 * to a shared page. It is done by copying the page to a new address
1316 * and decrementing the shared-page counter for the old page.
1317 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 * Note that this routine assumes that the protection checks have been
1319 * done by the caller (the low-level page fault routine in most cases).
1320 * Thus we can safely just mark it writable once we've done any necessary
1321 * COW.
1322 *
1323 * We also mark the page dirty at this point even though the page will
1324 * change only once the write actually happens. This avoids a few races,
1325 * and potentially makes it more efficient.
1326 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001327 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1328 * but allow concurrent faults), with pte both mapped and locked.
1329 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001331static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1332 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001333 spinlock_t *ptl, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334{
Hugh Dickins920fc352005-11-21 21:32:17 -08001335 struct page *old_page, *src_page, *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 pte_t entry;
Hugh Dickins65500d22005-10-29 18:15:59 -07001337 int ret = VM_FAULT_MINOR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
Linus Torvalds6aab3412005-11-28 14:34:23 -08001339 old_page = vm_normal_page(vma, address, orig_pte);
Hugh Dickins920fc352005-11-21 21:32:17 -08001340 src_page = old_page;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001341 if (!old_page)
1342 goto gotten;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343
Hugh Dickinsd296e9c2005-06-21 17:15:11 -07001344 if (PageAnon(old_page) && !TestSetPageLocked(old_page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 int reuse = can_share_swap_page(old_page);
1346 unlock_page(old_page);
1347 if (reuse) {
Ben Collinseca35132005-11-29 11:45:26 -08001348 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07001349 entry = pte_mkyoung(orig_pte);
1350 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 ptep_set_access_flags(vma, address, page_table, entry, 1);
1352 update_mmu_cache(vma, address, entry);
1353 lazy_mmu_prot_update(entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07001354 ret |= VM_FAULT_WRITE;
1355 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 }
1357 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358
1359 /*
1360 * Ok, we need to copy. Oh, well..
1361 */
Nick Pigginb5810032005-10-29 18:16:12 -07001362 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08001363gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001364 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365
1366 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07001367 goto oom;
Hugh Dickins920fc352005-11-21 21:32:17 -08001368 if (src_page == ZERO_PAGE(address)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 new_page = alloc_zeroed_user_highpage(vma, address);
1370 if (!new_page)
Hugh Dickins65500d22005-10-29 18:15:59 -07001371 goto oom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 } else {
1373 new_page = alloc_page_vma(GFP_HIGHUSER, vma, address);
1374 if (!new_page)
Hugh Dickins65500d22005-10-29 18:15:59 -07001375 goto oom;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001376 cow_user_page(new_page, src_page, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 }
Hugh Dickins65500d22005-10-29 18:15:59 -07001378
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 /*
1380 * Re-check the pte - we dropped the lock
1381 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001382 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07001383 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08001384 if (old_page) {
1385 page_remove_rmap(old_page);
1386 if (!PageAnon(old_page)) {
1387 dec_mm_counter(mm, file_rss);
1388 inc_mm_counter(mm, anon_rss);
1389 }
1390 } else
Hugh Dickins42946212005-10-29 18:16:05 -07001391 inc_mm_counter(mm, anon_rss);
Ben Collinseca35132005-11-29 11:45:26 -08001392 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07001393 entry = mk_pte(new_page, vma->vm_page_prot);
1394 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1395 ptep_establish(vma, address, page_table, entry);
1396 update_mmu_cache(vma, address, entry);
1397 lazy_mmu_prot_update(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 lru_cache_add_active(new_page);
1399 page_add_anon_rmap(new_page, vma, address);
1400
1401 /* Free the old page.. */
1402 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001403 ret |= VM_FAULT_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 }
Hugh Dickins920fc352005-11-21 21:32:17 -08001405 if (new_page)
1406 page_cache_release(new_page);
1407 if (old_page)
1408 page_cache_release(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001409unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001410 pte_unmap_unlock(page_table, ptl);
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001411 return ret;
Hugh Dickins65500d22005-10-29 18:15:59 -07001412oom:
Hugh Dickins920fc352005-11-21 21:32:17 -08001413 if (old_page)
1414 page_cache_release(old_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 return VM_FAULT_OOM;
1416}
1417
1418/*
1419 * Helper functions for unmap_mapping_range().
1420 *
1421 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1422 *
1423 * We have to restart searching the prio_tree whenever we drop the lock,
1424 * since the iterator is only valid while the lock is held, and anyway
1425 * a later vma might be split and reinserted earlier while lock dropped.
1426 *
1427 * The list of nonlinear vmas could be handled more efficiently, using
1428 * a placeholder, but handle it in the same way until a need is shown.
1429 * It is important to search the prio_tree before nonlinear list: a vma
1430 * may become nonlinear and be shifted from prio_tree to nonlinear list
1431 * while the lock is dropped; but never shifted from list to prio_tree.
1432 *
1433 * In order to make forward progress despite restarting the search,
1434 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1435 * quickly skip it next time around. Since the prio_tree search only
1436 * shows us those vmas affected by unmapping the range in question, we
1437 * can't efficiently keep all vmas in step with mapping->truncate_count:
1438 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1439 * mapping->truncate_count and vma->vm_truncate_count are protected by
1440 * i_mmap_lock.
1441 *
1442 * In order to make forward progress despite repeatedly restarting some
Hugh Dickinsee39b372005-04-19 13:29:15 -07001443 * large vma, note the restart_addr from unmap_vmas when it breaks out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 * and restart from that address when we reach that vma again. It might
1445 * have been split or merged, shrunk or extended, but never shifted: so
1446 * restart_addr remains valid so long as it remains in the vma's range.
1447 * unmap_mapping_range forces truncate_count to leap over page-aligned
1448 * values so we can save vma's restart_addr in its truncate_count field.
1449 */
1450#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1451
1452static void reset_vma_truncate_counts(struct address_space *mapping)
1453{
1454 struct vm_area_struct *vma;
1455 struct prio_tree_iter iter;
1456
1457 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1458 vma->vm_truncate_count = 0;
1459 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1460 vma->vm_truncate_count = 0;
1461}
1462
1463static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1464 unsigned long start_addr, unsigned long end_addr,
1465 struct zap_details *details)
1466{
1467 unsigned long restart_addr;
1468 int need_break;
1469
1470again:
1471 restart_addr = vma->vm_truncate_count;
1472 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1473 start_addr = restart_addr;
1474 if (start_addr >= end_addr) {
1475 /* Top of vma has been split off since last time */
1476 vma->vm_truncate_count = details->truncate_count;
1477 return 0;
1478 }
1479 }
1480
Hugh Dickinsee39b372005-04-19 13:29:15 -07001481 restart_addr = zap_page_range(vma, start_addr,
1482 end_addr - start_addr, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 need_break = need_resched() ||
1484 need_lockbreak(details->i_mmap_lock);
1485
Hugh Dickinsee39b372005-04-19 13:29:15 -07001486 if (restart_addr >= end_addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 /* We have now completed this vma: mark it so */
1488 vma->vm_truncate_count = details->truncate_count;
1489 if (!need_break)
1490 return 0;
1491 } else {
1492 /* Note restart_addr in vma's truncate_count field */
Hugh Dickinsee39b372005-04-19 13:29:15 -07001493 vma->vm_truncate_count = restart_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 if (!need_break)
1495 goto again;
1496 }
1497
1498 spin_unlock(details->i_mmap_lock);
1499 cond_resched();
1500 spin_lock(details->i_mmap_lock);
1501 return -EINTR;
1502}
1503
1504static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1505 struct zap_details *details)
1506{
1507 struct vm_area_struct *vma;
1508 struct prio_tree_iter iter;
1509 pgoff_t vba, vea, zba, zea;
1510
1511restart:
1512 vma_prio_tree_foreach(vma, &iter, root,
1513 details->first_index, details->last_index) {
1514 /* Skip quickly over those we have already dealt with */
1515 if (vma->vm_truncate_count == details->truncate_count)
1516 continue;
1517
1518 vba = vma->vm_pgoff;
1519 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1520 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1521 zba = details->first_index;
1522 if (zba < vba)
1523 zba = vba;
1524 zea = details->last_index;
1525 if (zea > vea)
1526 zea = vea;
1527
1528 if (unmap_mapping_range_vma(vma,
1529 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1530 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1531 details) < 0)
1532 goto restart;
1533 }
1534}
1535
1536static inline void unmap_mapping_range_list(struct list_head *head,
1537 struct zap_details *details)
1538{
1539 struct vm_area_struct *vma;
1540
1541 /*
1542 * In nonlinear VMAs there is no correspondence between virtual address
1543 * offset and file offset. So we must perform an exhaustive search
1544 * across *all* the pages in each nonlinear VMA, not just the pages
1545 * whose virtual address lies outside the file truncation point.
1546 */
1547restart:
1548 list_for_each_entry(vma, head, shared.vm_set.list) {
1549 /* Skip quickly over those we have already dealt with */
1550 if (vma->vm_truncate_count == details->truncate_count)
1551 continue;
1552 details->nonlinear_vma = vma;
1553 if (unmap_mapping_range_vma(vma, vma->vm_start,
1554 vma->vm_end, details) < 0)
1555 goto restart;
1556 }
1557}
1558
1559/**
1560 * unmap_mapping_range - unmap the portion of all mmaps
1561 * in the specified address_space corresponding to the specified
1562 * page range in the underlying file.
Martin Waitz3d410882005-06-23 22:05:21 -07001563 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 * @holebegin: byte in first page to unmap, relative to the start of
1565 * the underlying file. This will be rounded down to a PAGE_SIZE
1566 * boundary. Note that this is different from vmtruncate(), which
1567 * must keep the partial page. In contrast, we must get rid of
1568 * partial pages.
1569 * @holelen: size of prospective hole in bytes. This will be rounded
1570 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1571 * end of the file.
1572 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1573 * but 0 when invalidating pagecache, don't throw away private data.
1574 */
1575void unmap_mapping_range(struct address_space *mapping,
1576 loff_t const holebegin, loff_t const holelen, int even_cows)
1577{
1578 struct zap_details details;
1579 pgoff_t hba = holebegin >> PAGE_SHIFT;
1580 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1581
1582 /* Check for overflow. */
1583 if (sizeof(holelen) > sizeof(hlen)) {
1584 long long holeend =
1585 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1586 if (holeend & ~(long long)ULONG_MAX)
1587 hlen = ULONG_MAX - hba + 1;
1588 }
1589
1590 details.check_mapping = even_cows? NULL: mapping;
1591 details.nonlinear_vma = NULL;
1592 details.first_index = hba;
1593 details.last_index = hba + hlen - 1;
1594 if (details.last_index < details.first_index)
1595 details.last_index = ULONG_MAX;
1596 details.i_mmap_lock = &mapping->i_mmap_lock;
1597
1598 spin_lock(&mapping->i_mmap_lock);
1599
1600 /* serialize i_size write against truncate_count write */
1601 smp_wmb();
1602 /* Protect against page faults, and endless unmapping loops */
1603 mapping->truncate_count++;
1604 /*
1605 * For archs where spin_lock has inclusive semantics like ia64
1606 * this smp_mb() will prevent to read pagetable contents
1607 * before the truncate_count increment is visible to
1608 * other cpus.
1609 */
1610 smp_mb();
1611 if (unlikely(is_restart_addr(mapping->truncate_count))) {
1612 if (mapping->truncate_count == 0)
1613 reset_vma_truncate_counts(mapping);
1614 mapping->truncate_count++;
1615 }
1616 details.truncate_count = mapping->truncate_count;
1617
1618 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
1619 unmap_mapping_range_tree(&mapping->i_mmap, &details);
1620 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
1621 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
1622 spin_unlock(&mapping->i_mmap_lock);
1623}
1624EXPORT_SYMBOL(unmap_mapping_range);
1625
1626/*
1627 * Handle all mappings that got truncated by a "truncate()"
1628 * system call.
1629 *
1630 * NOTE! We have to be ready to update the memory sharing
1631 * between the file and the memory map for a potential last
1632 * incomplete page. Ugly, but necessary.
1633 */
1634int vmtruncate(struct inode * inode, loff_t offset)
1635{
1636 struct address_space *mapping = inode->i_mapping;
1637 unsigned long limit;
1638
1639 if (inode->i_size < offset)
1640 goto do_expand;
1641 /*
1642 * truncation of in-use swapfiles is disallowed - it would cause
1643 * subsequent swapout to scribble on the now-freed blocks.
1644 */
1645 if (IS_SWAPFILE(inode))
1646 goto out_busy;
1647 i_size_write(inode, offset);
1648 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
1649 truncate_inode_pages(mapping, offset);
1650 goto out_truncate;
1651
1652do_expand:
1653 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1654 if (limit != RLIM_INFINITY && offset > limit)
1655 goto out_sig;
1656 if (offset > inode->i_sb->s_maxbytes)
1657 goto out_big;
1658 i_size_write(inode, offset);
1659
1660out_truncate:
1661 if (inode->i_op && inode->i_op->truncate)
1662 inode->i_op->truncate(inode);
1663 return 0;
1664out_sig:
1665 send_sig(SIGXFSZ, current, 0);
1666out_big:
1667 return -EFBIG;
1668out_busy:
1669 return -ETXTBSY;
1670}
1671
1672EXPORT_SYMBOL(vmtruncate);
1673
1674/*
1675 * Primitive swap readahead code. We simply read an aligned block of
1676 * (1 << page_cluster) entries in the swap area. This method is chosen
1677 * because it doesn't cost us any seek time. We also make sure to queue
1678 * the 'original' request together with the readahead ones...
1679 *
1680 * This has been extended to use the NUMA policies from the mm triggering
1681 * the readahead.
1682 *
1683 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
1684 */
1685void swapin_readahead(swp_entry_t entry, unsigned long addr,struct vm_area_struct *vma)
1686{
1687#ifdef CONFIG_NUMA
1688 struct vm_area_struct *next_vma = vma ? vma->vm_next : NULL;
1689#endif
1690 int i, num;
1691 struct page *new_page;
1692 unsigned long offset;
1693
1694 /*
1695 * Get the number of handles we should do readahead io to.
1696 */
1697 num = valid_swaphandles(entry, &offset);
1698 for (i = 0; i < num; offset++, i++) {
1699 /* Ok, do the async read-ahead now */
1700 new_page = read_swap_cache_async(swp_entry(swp_type(entry),
1701 offset), vma, addr);
1702 if (!new_page)
1703 break;
1704 page_cache_release(new_page);
1705#ifdef CONFIG_NUMA
1706 /*
1707 * Find the next applicable VMA for the NUMA policy.
1708 */
1709 addr += PAGE_SIZE;
1710 if (addr == 0)
1711 vma = NULL;
1712 if (vma) {
1713 if (addr >= vma->vm_end) {
1714 vma = next_vma;
1715 next_vma = vma ? vma->vm_next : NULL;
1716 }
1717 if (vma && addr < vma->vm_start)
1718 vma = NULL;
1719 } else {
1720 if (next_vma && addr >= next_vma->vm_start) {
1721 vma = next_vma;
1722 next_vma = vma->vm_next;
1723 }
1724 }
1725#endif
1726 }
1727 lru_add_drain(); /* Push any new pages onto the LRU now */
1728}
1729
1730/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001731 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1732 * but allow concurrent faults), and pte mapped but not yet locked.
1733 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001735static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
1736 unsigned long address, pte_t *page_table, pmd_t *pmd,
1737 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001739 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 struct page *page;
Hugh Dickins65500d22005-10-29 18:15:59 -07001741 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 pte_t pte;
1743 int ret = VM_FAULT_MINOR;
1744
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001745 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001746 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07001747
1748 entry = pte_to_swp_entry(orig_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 page = lookup_swap_cache(entry);
1750 if (!page) {
1751 swapin_readahead(entry, address, vma);
1752 page = read_swap_cache_async(entry, vma, address);
1753 if (!page) {
1754 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001755 * Back out if somebody else faulted in this pte
1756 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001758 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 if (likely(pte_same(*page_table, orig_pte)))
1760 ret = VM_FAULT_OOM;
Hugh Dickins65500d22005-10-29 18:15:59 -07001761 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 }
1763
1764 /* Had to read the page from swap area: Major fault */
1765 ret = VM_FAULT_MAJOR;
1766 inc_page_state(pgmajfault);
1767 grab_swap_token();
1768 }
1769
1770 mark_page_accessed(page);
1771 lock_page(page);
1772
1773 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001774 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001776 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07001777 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07001778 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07001779
1780 if (unlikely(!PageUptodate(page))) {
1781 ret = VM_FAULT_SIGBUS;
1782 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 }
1784
1785 /* The page isn't present yet, go ahead with the fault. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786
Hugh Dickins42946212005-10-29 18:16:05 -07001787 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 pte = mk_pte(page, vma->vm_page_prot);
1789 if (write_access && can_share_swap_page(page)) {
1790 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
1791 write_access = 0;
1792 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793
1794 flush_icache_page(vma, page);
1795 set_pte_at(mm, address, page_table, pte);
1796 page_add_anon_rmap(page, vma, address);
1797
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001798 swap_free(entry);
1799 if (vm_swap_full())
1800 remove_exclusive_swap_page(page);
1801 unlock_page(page);
1802
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 if (write_access) {
1804 if (do_wp_page(mm, vma, address,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001805 page_table, pmd, ptl, pte) == VM_FAULT_OOM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 ret = VM_FAULT_OOM;
1807 goto out;
1808 }
1809
1810 /* No need to invalidate - it was non-present before */
1811 update_mmu_cache(vma, address, pte);
1812 lazy_mmu_prot_update(pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07001813unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001814 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815out:
1816 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07001817out_nomap:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001818 pte_unmap_unlock(page_table, ptl);
Kirill Korotaevb8107482005-05-16 21:53:50 -07001819 unlock_page(page);
1820 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001821 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822}
1823
1824/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001825 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1826 * but allow concurrent faults), and pte mapped but not yet locked.
1827 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001829static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
1830 unsigned long address, pte_t *page_table, pmd_t *pmd,
1831 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001833 struct page *page;
1834 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836
Linus Torvalds6aab3412005-11-28 14:34:23 -08001837 if (write_access) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 /* Allocate our own private page. */
1839 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840
1841 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07001842 goto oom;
1843 page = alloc_zeroed_user_highpage(vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 if (!page)
Hugh Dickins65500d22005-10-29 18:15:59 -07001845 goto oom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846
Hugh Dickins65500d22005-10-29 18:15:59 -07001847 entry = mk_pte(page, vma->vm_page_prot);
1848 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001849
1850 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
1851 if (!pte_none(*page_table))
1852 goto release;
1853 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 lru_cache_add_active(page);
1855 SetPageReferenced(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001856 page_add_anon_rmap(page, vma, address);
Nick Pigginb5810032005-10-29 18:16:12 -07001857 } else {
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001858 /* Map the ZERO_PAGE - vm_page_prot is readonly */
1859 page = ZERO_PAGE(address);
1860 page_cache_get(page);
1861 entry = mk_pte(page, vma->vm_page_prot);
1862
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001863 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001864 spin_lock(ptl);
1865 if (!pte_none(*page_table))
1866 goto release;
Nick Pigginb5810032005-10-29 18:16:12 -07001867 inc_mm_counter(mm, file_rss);
1868 page_add_file_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 }
1870
Hugh Dickins65500d22005-10-29 18:15:59 -07001871 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872
1873 /* No need to invalidate - it was non-present before */
Hugh Dickins65500d22005-10-29 18:15:59 -07001874 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 lazy_mmu_prot_update(entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07001876unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001877 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 return VM_FAULT_MINOR;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001879release:
1880 page_cache_release(page);
1881 goto unlock;
Hugh Dickins65500d22005-10-29 18:15:59 -07001882oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 return VM_FAULT_OOM;
1884}
1885
1886/*
1887 * do_no_page() tries to create a new page mapping. It aggressively
1888 * tries to share with existing pages, but makes a separate copy if
1889 * the "write_access" parameter is true in order to avoid the next
1890 * page fault.
1891 *
1892 * As this is called only for pages that do not currently exist, we
1893 * do not need to flush old virtual caches or the TLB.
1894 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001895 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1896 * but allow concurrent faults), and pte mapped but not yet locked.
1897 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001899static int do_no_page(struct mm_struct *mm, struct vm_area_struct *vma,
1900 unsigned long address, pte_t *page_table, pmd_t *pmd,
1901 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001903 spinlock_t *ptl;
Hugh Dickins65500d22005-10-29 18:15:59 -07001904 struct page *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 struct address_space *mapping = NULL;
1906 pte_t entry;
1907 unsigned int sequence = 0;
1908 int ret = VM_FAULT_MINOR;
1909 int anon = 0;
1910
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 if (vma->vm_file) {
1913 mapping = vma->vm_file->f_mapping;
1914 sequence = mapping->truncate_count;
1915 smp_rmb(); /* serializes i_size against truncate_count */
1916 }
1917retry:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 new_page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret);
1919 /*
1920 * No smp_rmb is needed here as long as there's a full
1921 * spin_lock/unlock sequence inside the ->nopage callback
1922 * (for the pagecache lookup) that acts as an implicit
1923 * smp_mb() and prevents the i_size read to happen
1924 * after the next truncate_count read.
1925 */
1926
1927 /* no page was available -- either SIGBUS or OOM */
1928 if (new_page == NOPAGE_SIGBUS)
1929 return VM_FAULT_SIGBUS;
1930 if (new_page == NOPAGE_OOM)
1931 return VM_FAULT_OOM;
1932
1933 /*
1934 * Should we do an early C-O-W break?
1935 */
1936 if (write_access && !(vma->vm_flags & VM_SHARED)) {
1937 struct page *page;
1938
1939 if (unlikely(anon_vma_prepare(vma)))
1940 goto oom;
1941 page = alloc_page_vma(GFP_HIGHUSER, vma, address);
1942 if (!page)
1943 goto oom;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001944 cow_user_page(page, new_page, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 page_cache_release(new_page);
1946 new_page = page;
1947 anon = 1;
1948 }
1949
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001950 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 /*
1952 * For a file-backed vma, someone could have truncated or otherwise
1953 * invalidated this page. If unmap_mapping_range got called,
1954 * retry getting the page.
1955 */
1956 if (mapping && unlikely(sequence != mapping->truncate_count)) {
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001957 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001959 cond_resched();
1960 sequence = mapping->truncate_count;
1961 smp_rmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 goto retry;
1963 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964
1965 /*
1966 * This silly early PAGE_DIRTY setting removes a race
1967 * due to the bad i386 page protection. But it's valid
1968 * for other architectures too.
1969 *
1970 * Note that if write_access is true, we either now have
1971 * an exclusive copy of the page, or this is a shared mapping,
1972 * so we can make it writable and dirty to avoid having to
1973 * handle that later.
1974 */
1975 /* Only go through if we didn't race with anybody else... */
1976 if (pte_none(*page_table)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 flush_icache_page(vma, new_page);
1978 entry = mk_pte(new_page, vma->vm_page_prot);
1979 if (write_access)
1980 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1981 set_pte_at(mm, address, page_table, entry);
1982 if (anon) {
Hugh Dickins42946212005-10-29 18:16:05 -07001983 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 lru_cache_add_active(new_page);
1985 page_add_anon_rmap(new_page, vma, address);
Hugh Dickinsf57e88a2005-11-21 21:32:19 -08001986 } else {
Hugh Dickins42946212005-10-29 18:16:05 -07001987 inc_mm_counter(mm, file_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 page_add_file_rmap(new_page);
Hugh Dickins42946212005-10-29 18:16:05 -07001989 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 } else {
1991 /* One of our sibling threads was faster, back out. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001993 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 }
1995
1996 /* no need to invalidate: a not-present page shouldn't be cached */
1997 update_mmu_cache(vma, address, entry);
1998 lazy_mmu_prot_update(entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07001999unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002000 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 return ret;
2002oom:
2003 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002004 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005}
2006
2007/*
2008 * Fault of a previously existing named mapping. Repopulate the pte
2009 * from the encoded file_pte if possible. This enables swappable
2010 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002011 *
2012 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2013 * but allow concurrent faults), and pte mapped but not yet locked.
2014 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002016static int do_file_page(struct mm_struct *mm, struct vm_area_struct *vma,
2017 unsigned long address, pte_t *page_table, pmd_t *pmd,
2018 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019{
Hugh Dickins65500d22005-10-29 18:15:59 -07002020 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 int err;
2022
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002023 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002024 return VM_FAULT_MINOR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025
Hugh Dickins65500d22005-10-29 18:15:59 -07002026 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
2027 /*
2028 * Page table corrupted: show pte and kill process.
2029 */
Nick Pigginb5810032005-10-29 18:16:12 -07002030 print_bad_pte(vma, orig_pte, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002031 return VM_FAULT_OOM;
2032 }
2033 /* We can then assume vm->vm_ops && vma->vm_ops->populate */
2034
2035 pgoff = pte_to_pgoff(orig_pte);
2036 err = vma->vm_ops->populate(vma, address & PAGE_MASK, PAGE_SIZE,
2037 vma->vm_page_prot, pgoff, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 if (err == -ENOMEM)
2039 return VM_FAULT_OOM;
2040 if (err)
2041 return VM_FAULT_SIGBUS;
2042 return VM_FAULT_MAJOR;
2043}
2044
2045/*
2046 * These routines also need to handle stuff like marking pages dirty
2047 * and/or accessed for architectures that don't do it in hardware (most
2048 * RISC architectures). The early dirtying is also good on the i386.
2049 *
2050 * There is also a hook called "update_mmu_cache()" that architectures
2051 * with external mmu caches can use to update those (ie the Sparc or
2052 * PowerPC hashed page tables that act as extended TLBs).
2053 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07002054 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2055 * but allow concurrent faults), and pte mapped but not yet locked.
2056 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 */
2058static inline int handle_pte_fault(struct mm_struct *mm,
Hugh Dickins65500d22005-10-29 18:15:59 -07002059 struct vm_area_struct *vma, unsigned long address,
2060 pte_t *pte, pmd_t *pmd, int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061{
2062 pte_t entry;
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002063 pte_t old_entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002064 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002066 old_entry = entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002068 if (pte_none(entry)) {
2069 if (!vma->vm_ops || !vma->vm_ops->nopage)
2070 return do_anonymous_page(mm, vma, address,
2071 pte, pmd, write_access);
2072 return do_no_page(mm, vma, address,
2073 pte, pmd, write_access);
2074 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 if (pte_file(entry))
Hugh Dickins65500d22005-10-29 18:15:59 -07002076 return do_file_page(mm, vma, address,
2077 pte, pmd, write_access, entry);
2078 return do_swap_page(mm, vma, address,
2079 pte, pmd, write_access, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 }
2081
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002082 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002083 spin_lock(ptl);
2084 if (unlikely(!pte_same(*pte, entry)))
2085 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 if (write_access) {
2087 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002088 return do_wp_page(mm, vma, address,
2089 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 entry = pte_mkdirty(entry);
2091 }
2092 entry = pte_mkyoung(entry);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002093 if (!pte_same(old_entry, entry)) {
2094 ptep_set_access_flags(vma, address, pte, entry, write_access);
2095 update_mmu_cache(vma, address, entry);
2096 lazy_mmu_prot_update(entry);
2097 } else {
2098 /*
2099 * This is needed only for protection faults but the arch code
2100 * is not yet telling us if this is a protection fault or not.
2101 * This still avoids useless tlb flushes for .text page faults
2102 * with threads.
2103 */
2104 if (write_access)
2105 flush_tlb_page(vma, address);
2106 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002107unlock:
2108 pte_unmap_unlock(pte, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 return VM_FAULT_MINOR;
2110}
2111
2112/*
2113 * By the time we get here, we already hold the mm semaphore
2114 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002115int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 unsigned long address, int write_access)
2117{
2118 pgd_t *pgd;
2119 pud_t *pud;
2120 pmd_t *pmd;
2121 pte_t *pte;
2122
2123 __set_current_state(TASK_RUNNING);
2124
2125 inc_page_state(pgfault);
2126
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01002127 if (unlikely(is_vm_hugetlb_page(vma)))
2128 return hugetlb_fault(mm, vma, address, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 pud = pud_alloc(mm, pgd, address);
2132 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002133 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 pmd = pmd_alloc(mm, pud, address);
2135 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002136 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 pte = pte_alloc_map(mm, pmd, address);
2138 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002139 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140
Hugh Dickinsc74df322005-10-29 18:16:23 -07002141 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142}
2143
2144#ifndef __PAGETABLE_PUD_FOLDED
2145/*
2146 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002147 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002149int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002151 pud_t *new = pud_alloc_one(mm, address);
2152 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002153 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154
Hugh Dickins872fec12005-10-29 18:16:21 -07002155 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002156 if (pgd_present(*pgd)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 pud_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002158 else
2159 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07002160 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002161 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162}
Alan Sterne0f39592005-11-28 13:43:44 -08002163#else
2164/* Workaround for gcc 2.96 */
2165int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
2166{
2167 return 0;
2168}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169#endif /* __PAGETABLE_PUD_FOLDED */
2170
2171#ifndef __PAGETABLE_PMD_FOLDED
2172/*
2173 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002174 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002176int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002178 pmd_t *new = pmd_alloc_one(mm, address);
2179 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002180 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181
Hugh Dickins872fec12005-10-29 18:16:21 -07002182 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07002184 if (pud_present(*pud)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 pmd_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002186 else
2187 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07002189 if (pgd_present(*pud)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190 pmd_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002191 else
2192 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07002194 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002195 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196}
Alan Sterne0f39592005-11-28 13:43:44 -08002197#else
2198/* Workaround for gcc 2.96 */
2199int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
2200{
2201 return 0;
2202}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203#endif /* __PAGETABLE_PMD_FOLDED */
2204
2205int make_pages_present(unsigned long addr, unsigned long end)
2206{
2207 int ret, len, write;
2208 struct vm_area_struct * vma;
2209
2210 vma = find_vma(current->mm, addr);
2211 if (!vma)
2212 return -1;
2213 write = (vma->vm_flags & VM_WRITE) != 0;
2214 if (addr >= end)
2215 BUG();
2216 if (end > vma->vm_end)
2217 BUG();
2218 len = (end+PAGE_SIZE-1)/PAGE_SIZE-addr/PAGE_SIZE;
2219 ret = get_user_pages(current, current->mm, addr,
2220 len, write, 0, NULL, NULL);
2221 if (ret < 0)
2222 return ret;
2223 return ret == len ? 0 : -1;
2224}
2225
2226/*
2227 * Map a vmalloc()-space virtual address to the physical page.
2228 */
2229struct page * vmalloc_to_page(void * vmalloc_addr)
2230{
2231 unsigned long addr = (unsigned long) vmalloc_addr;
2232 struct page *page = NULL;
2233 pgd_t *pgd = pgd_offset_k(addr);
2234 pud_t *pud;
2235 pmd_t *pmd;
2236 pte_t *ptep, pte;
2237
2238 if (!pgd_none(*pgd)) {
2239 pud = pud_offset(pgd, addr);
2240 if (!pud_none(*pud)) {
2241 pmd = pmd_offset(pud, addr);
2242 if (!pmd_none(*pmd)) {
2243 ptep = pte_offset_map(pmd, addr);
2244 pte = *ptep;
2245 if (pte_present(pte))
2246 page = pte_page(pte);
2247 pte_unmap(ptep);
2248 }
2249 }
2250 }
2251 return page;
2252}
2253
2254EXPORT_SYMBOL(vmalloc_to_page);
2255
2256/*
2257 * Map a vmalloc()-space virtual address to the physical page frame number.
2258 */
2259unsigned long vmalloc_to_pfn(void * vmalloc_addr)
2260{
2261 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
2262}
2263
2264EXPORT_SYMBOL(vmalloc_to_pfn);
2265
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266#if !defined(__HAVE_ARCH_GATE_AREA)
2267
2268#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07002269static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270
2271static int __init gate_vma_init(void)
2272{
2273 gate_vma.vm_mm = NULL;
2274 gate_vma.vm_start = FIXADDR_USER_START;
2275 gate_vma.vm_end = FIXADDR_USER_END;
2276 gate_vma.vm_page_prot = PAGE_READONLY;
Hugh Dickins0b14c172005-11-21 21:32:15 -08002277 gate_vma.vm_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 return 0;
2279}
2280__initcall(gate_vma_init);
2281#endif
2282
2283struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2284{
2285#ifdef AT_SYSINFO_EHDR
2286 return &gate_vma;
2287#else
2288 return NULL;
2289#endif
2290}
2291
2292int in_gate_area_no_task(unsigned long addr)
2293{
2294#ifdef AT_SYSINFO_EHDR
2295 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2296 return 1;
2297#endif
2298 return 0;
2299}
2300
2301#endif /* __HAVE_ARCH_GATE_AREA */