blob: 85e80a57db29e2412ef766268996ec005c3e16c8 [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
Andi Kleena62eaf12006-02-16 23:41:58 +010085int randomize_va_space __read_mostly = 1;
86
87static int __init disable_randmaps(char *s)
88{
89 randomize_va_space = 0;
90 return 0;
91}
92__setup("norandmaps", disable_randmaps);
93
94
Linus Torvalds1da177e2005-04-16 15:20:36 -070095/*
96 * If a p?d_bad entry is found while walking page tables, report
97 * the error, before resetting entry to p?d_none. Usually (but
98 * very seldom) called out from the p?d_none_or_clear_bad macros.
99 */
100
101void pgd_clear_bad(pgd_t *pgd)
102{
103 pgd_ERROR(*pgd);
104 pgd_clear(pgd);
105}
106
107void pud_clear_bad(pud_t *pud)
108{
109 pud_ERROR(*pud);
110 pud_clear(pud);
111}
112
113void pmd_clear_bad(pmd_t *pmd)
114{
115 pmd_ERROR(*pmd);
116 pmd_clear(pmd);
117}
118
119/*
120 * Note: this doesn't free the actual pages themselves. That
121 * has been handled earlier when unmapping all the memory regions.
122 */
Hugh Dickinse0da3822005-04-19 13:29:15 -0700123static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124{
Hugh Dickinse0da3822005-04-19 13:29:15 -0700125 struct page *page = pmd_page(*pmd);
126 pmd_clear(pmd);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700127 pte_lock_deinit(page);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700128 pte_free_tlb(tlb, page);
129 dec_page_state(nr_page_table_pages);
130 tlb->mm->nr_ptes--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131}
132
Hugh Dickinse0da3822005-04-19 13:29:15 -0700133static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
134 unsigned long addr, unsigned long end,
135 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136{
137 pmd_t *pmd;
138 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700139 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140
Hugh Dickinse0da3822005-04-19 13:29:15 -0700141 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 do {
144 next = pmd_addr_end(addr, end);
145 if (pmd_none_or_clear_bad(pmd))
146 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700147 free_pte_range(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 } while (pmd++, addr = next, addr != end);
149
Hugh Dickinse0da3822005-04-19 13:29:15 -0700150 start &= PUD_MASK;
151 if (start < floor)
152 return;
153 if (ceiling) {
154 ceiling &= PUD_MASK;
155 if (!ceiling)
156 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700158 if (end - 1 > ceiling - 1)
159 return;
160
161 pmd = pmd_offset(pud, start);
162 pud_clear(pud);
163 pmd_free_tlb(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164}
165
Hugh Dickinse0da3822005-04-19 13:29:15 -0700166static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
167 unsigned long addr, unsigned long end,
168 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169{
170 pud_t *pud;
171 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700172 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173
Hugh Dickinse0da3822005-04-19 13:29:15 -0700174 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 pud = pud_offset(pgd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 do {
177 next = pud_addr_end(addr, end);
178 if (pud_none_or_clear_bad(pud))
179 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700180 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 } while (pud++, addr = next, addr != end);
182
Hugh Dickinse0da3822005-04-19 13:29:15 -0700183 start &= PGDIR_MASK;
184 if (start < floor)
185 return;
186 if (ceiling) {
187 ceiling &= PGDIR_MASK;
188 if (!ceiling)
189 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700191 if (end - 1 > ceiling - 1)
192 return;
193
194 pud = pud_offset(pgd, start);
195 pgd_clear(pgd);
196 pud_free_tlb(tlb, pud);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197}
198
199/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700200 * This function frees user-level page tables of a process.
201 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 * Must be called with pagetable lock held.
203 */
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700204void free_pgd_range(struct mmu_gather **tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700205 unsigned long addr, unsigned long end,
206 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207{
208 pgd_t *pgd;
209 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700210 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
Hugh Dickinse0da3822005-04-19 13:29:15 -0700212 /*
213 * The next few lines have given us lots of grief...
214 *
215 * Why are we testing PMD* at this top level? Because often
216 * there will be no work to do at all, and we'd prefer not to
217 * go all the way down to the bottom just to discover that.
218 *
219 * Why all these "- 1"s? Because 0 represents both the bottom
220 * of the address space and the top of it (using -1 for the
221 * top wouldn't help much: the masks would do the wrong thing).
222 * The rule is that addr 0 and floor 0 refer to the bottom of
223 * the address space, but end 0 and ceiling 0 refer to the top
224 * Comparisons need to use "end - 1" and "ceiling - 1" (though
225 * that end 0 case should be mythical).
226 *
227 * Wherever addr is brought up or ceiling brought down, we must
228 * be careful to reject "the opposite 0" before it confuses the
229 * subsequent tests. But what about where end is brought down
230 * by PMD_SIZE below? no, end can't go down to 0 there.
231 *
232 * Whereas we round start (addr) and ceiling down, by different
233 * masks at different levels, in order to test whether a table
234 * now has no other vmas using it, so can be freed, we don't
235 * bother to round floor or end up - the tests don't need that.
236 */
237
238 addr &= PMD_MASK;
239 if (addr < floor) {
240 addr += PMD_SIZE;
241 if (!addr)
242 return;
243 }
244 if (ceiling) {
245 ceiling &= PMD_MASK;
246 if (!ceiling)
247 return;
248 }
249 if (end - 1 > ceiling - 1)
250 end -= PMD_SIZE;
251 if (addr > end - 1)
252 return;
253
254 start = addr;
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700255 pgd = pgd_offset((*tlb)->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 do {
257 next = pgd_addr_end(addr, end);
258 if (pgd_none_or_clear_bad(pgd))
259 continue;
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700260 free_pud_range(*tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700262
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700263 if (!(*tlb)->fullmm)
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700264 flush_tlb_pgtables((*tlb)->mm, start, end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700265}
266
267void free_pgtables(struct mmu_gather **tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700268 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700269{
270 while (vma) {
271 struct vm_area_struct *next = vma->vm_next;
272 unsigned long addr = vma->vm_start;
273
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700274 /*
275 * Hide vma from rmap and vmtruncate before freeing pgtables
276 */
277 anon_vma_unlink(vma);
278 unlink_file_vma(vma);
279
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700280 if (is_hugepage_only_range(vma->vm_mm, addr, HPAGE_SIZE)) {
281 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700282 floor, next? next->vm_start: ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700283 } else {
284 /*
285 * Optimization: gather nearby vmas into one call down
286 */
287 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
288 && !is_hugepage_only_range(vma->vm_mm, next->vm_start,
289 HPAGE_SIZE)) {
290 vma = next;
291 next = vma->vm_next;
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700292 anon_vma_unlink(vma);
293 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700294 }
295 free_pgd_range(tlb, addr, vma->vm_end,
296 floor, next? next->vm_start: ceiling);
297 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700298 vma = next;
299 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300}
301
Hugh Dickins1bb36302005-10-29 18:16:22 -0700302int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303{
Hugh Dickinsc74df322005-10-29 18:16:23 -0700304 struct page *new = pte_alloc_one(mm, address);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700305 if (!new)
306 return -ENOMEM;
307
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700308 pte_lock_init(new);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700309 spin_lock(&mm->page_table_lock);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700310 if (pmd_present(*pmd)) { /* Another has populated it */
311 pte_lock_deinit(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700312 pte_free(new);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700313 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 mm->nr_ptes++;
315 inc_page_state(nr_page_table_pages);
316 pmd_populate(mm, pmd, new);
317 }
Hugh Dickinsc74df322005-10-29 18:16:23 -0700318 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700319 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320}
321
Hugh Dickins1bb36302005-10-29 18:16:22 -0700322int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700324 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
325 if (!new)
326 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327
Hugh Dickins1bb36302005-10-29 18:16:22 -0700328 spin_lock(&init_mm.page_table_lock);
329 if (pmd_present(*pmd)) /* Another has populated it */
330 pte_free_kernel(new);
331 else
332 pmd_populate_kernel(&init_mm, pmd, new);
333 spin_unlock(&init_mm.page_table_lock);
334 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335}
336
Hugh Dickinsae859762005-10-29 18:16:05 -0700337static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
338{
339 if (file_rss)
340 add_mm_counter(mm, file_rss, file_rss);
341 if (anon_rss)
342 add_mm_counter(mm, anon_rss, anon_rss);
343}
344
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800346 * This function is called to print an error when a bad pte
347 * is found. For example, we might have a PFN-mapped pte in
348 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700349 *
350 * The calling function must still handle the error.
351 */
352void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr)
353{
354 printk(KERN_ERR "Bad pte = %08llx, process = %s, "
355 "vm_flags = %lx, vaddr = %lx\n",
356 (long long)pte_val(pte),
357 (vma->vm_mm == current->mm ? current->comm : "???"),
358 vma->vm_flags, vaddr);
359 dump_stack();
360}
361
Linus Torvalds67121172005-12-11 20:38:17 -0800362static inline int is_cow_mapping(unsigned int flags)
363{
364 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
365}
366
Nick Pigginb5810032005-10-29 18:16:12 -0700367/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800368 * This function gets the "struct page" associated with a pte.
369 *
370 * NOTE! Some mappings do not have "struct pages". A raw PFN mapping
371 * will have each page table entry just pointing to a raw page frame
372 * number, and as far as the VM layer is concerned, those do not have
373 * pages associated with them - even if the PFN might point to memory
374 * that otherwise is perfectly fine and has a "struct page".
375 *
376 * The way we recognize those mappings is through the rules set up
377 * by "remap_pfn_range()": the vma will have the VM_PFNMAP bit set,
378 * and the vm_pgoff will point to the first PFN mapped: thus every
379 * page that is a raw mapping will always honor the rule
380 *
381 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
382 *
383 * and if that isn't true, the page has been COW'ed (in which case it
384 * _does_ have a "struct page" associated with it even if it is in a
385 * VM_PFNMAP range).
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800386 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800387struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800388{
Linus Torvalds6aab3412005-11-28 14:34:23 -0800389 unsigned long pfn = pte_pfn(pte);
390
391 if (vma->vm_flags & VM_PFNMAP) {
392 unsigned long off = (addr - vma->vm_start) >> PAGE_SHIFT;
393 if (pfn == vma->vm_pgoff + off)
394 return NULL;
Linus Torvalds67121172005-12-11 20:38:17 -0800395 if (!is_cow_mapping(vma->vm_flags))
Linus Torvaldsfb155c12005-12-11 19:46:02 -0800396 return NULL;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800397 }
398
399 /*
400 * Add some anal sanity checks for now. Eventually,
401 * we should just do "return pfn_to_page(pfn)", but
402 * in the meantime we check that we get a valid pfn,
403 * and that the resulting page looks ok.
404 *
405 * Remove this test eventually!
406 */
407 if (unlikely(!pfn_valid(pfn))) {
408 print_bad_pte(vma, pte, addr);
409 return NULL;
410 }
411
412 /*
413 * NOTE! We still have PageReserved() pages in the page
414 * tables.
415 *
416 * The PAGE_ZERO() pages and various VDSO mappings can
417 * cause them to exist.
418 */
419 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800420}
421
422/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 * copy one vm_area from one task to the other. Assumes the page tables
424 * already present in the new task to be cleared in the whole range
425 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 */
427
Hugh Dickins8c103762005-10-29 18:16:13 -0700428static inline void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700430 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700431 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432{
Nick Pigginb5810032005-10-29 18:16:12 -0700433 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 pte_t pte = *src_pte;
435 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436
437 /* pte contains position in swap or file, so copy. */
438 if (unlikely(!pte_present(pte))) {
439 if (!pte_file(pte)) {
440 swap_duplicate(pte_to_swp_entry(pte));
441 /* make sure dst_mm is on swapoff's mmlist. */
442 if (unlikely(list_empty(&dst_mm->mmlist))) {
443 spin_lock(&mmlist_lock);
Hugh Dickinsf412ac02005-10-29 18:16:41 -0700444 if (list_empty(&dst_mm->mmlist))
445 list_add(&dst_mm->mmlist,
446 &src_mm->mmlist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 spin_unlock(&mmlist_lock);
448 }
449 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700450 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 }
452
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 /*
454 * If it's a COW mapping, write protect it both
455 * in the parent and the child
456 */
Linus Torvalds67121172005-12-11 20:38:17 -0800457 if (is_cow_mapping(vm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 ptep_set_wrprotect(src_mm, addr, src_pte);
459 pte = *src_pte;
460 }
461
462 /*
463 * If it's a shared mapping, mark it clean in
464 * the child
465 */
466 if (vm_flags & VM_SHARED)
467 pte = pte_mkclean(pte);
468 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800469
470 page = vm_normal_page(vma, addr, pte);
471 if (page) {
472 get_page(page);
473 page_dup_rmap(page);
474 rss[!!PageAnon(page)]++;
475 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700476
477out_set_pte:
478 set_pte_at(dst_mm, addr, dst_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479}
480
481static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
482 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
483 unsigned long addr, unsigned long end)
484{
485 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700486 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -0700487 int progress = 0;
Hugh Dickins8c103762005-10-29 18:16:13 -0700488 int rss[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489
490again:
Hugh Dickinsae859762005-10-29 18:16:05 -0700491 rss[1] = rss[0] = 0;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700492 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 if (!dst_pte)
494 return -ENOMEM;
495 src_pte = pte_offset_map_nested(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700496 src_ptl = pte_lockptr(src_mm, src_pmd);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700497 spin_lock(src_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 do {
500 /*
501 * We are holding two locks at this point - either of them
502 * could generate latencies in another task on another CPU.
503 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700504 if (progress >= 32) {
505 progress = 0;
506 if (need_resched() ||
Hugh Dickinsc74df322005-10-29 18:16:23 -0700507 need_lockbreak(src_ptl) ||
508 need_lockbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700509 break;
510 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 if (pte_none(*src_pte)) {
512 progress++;
513 continue;
514 }
Hugh Dickins8c103762005-10-29 18:16:13 -0700515 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 progress += 8;
517 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518
Hugh Dickinsc74df322005-10-29 18:16:23 -0700519 spin_unlock(src_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 pte_unmap_nested(src_pte - 1);
Hugh Dickinsae859762005-10-29 18:16:05 -0700521 add_mm_rss(dst_mm, rss[0], rss[1]);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700522 pte_unmap_unlock(dst_pte - 1, dst_ptl);
523 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 if (addr != end)
525 goto again;
526 return 0;
527}
528
529static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
530 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
531 unsigned long addr, unsigned long end)
532{
533 pmd_t *src_pmd, *dst_pmd;
534 unsigned long next;
535
536 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
537 if (!dst_pmd)
538 return -ENOMEM;
539 src_pmd = pmd_offset(src_pud, addr);
540 do {
541 next = pmd_addr_end(addr, end);
542 if (pmd_none_or_clear_bad(src_pmd))
543 continue;
544 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
545 vma, addr, next))
546 return -ENOMEM;
547 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
548 return 0;
549}
550
551static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
552 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
553 unsigned long addr, unsigned long end)
554{
555 pud_t *src_pud, *dst_pud;
556 unsigned long next;
557
558 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
559 if (!dst_pud)
560 return -ENOMEM;
561 src_pud = pud_offset(src_pgd, addr);
562 do {
563 next = pud_addr_end(addr, end);
564 if (pud_none_or_clear_bad(src_pud))
565 continue;
566 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
567 vma, addr, next))
568 return -ENOMEM;
569 } while (dst_pud++, src_pud++, addr = next, addr != end);
570 return 0;
571}
572
573int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
574 struct vm_area_struct *vma)
575{
576 pgd_t *src_pgd, *dst_pgd;
577 unsigned long next;
578 unsigned long addr = vma->vm_start;
579 unsigned long end = vma->vm_end;
580
Nick Piggind9928952005-08-28 16:49:11 +1000581 /*
582 * Don't copy ptes where a page fault will fill them correctly.
583 * Fork becomes much lighter when there are big shared or private
584 * readonly mappings. The tradeoff is that copy_page_range is more
585 * efficient than faulting.
586 */
Linus Torvalds4d7672b2005-12-16 10:21:23 -0800587 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
Nick Piggind9928952005-08-28 16:49:11 +1000588 if (!vma->anon_vma)
589 return 0;
590 }
591
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 if (is_vm_hugetlb_page(vma))
593 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
594
595 dst_pgd = pgd_offset(dst_mm, addr);
596 src_pgd = pgd_offset(src_mm, addr);
597 do {
598 next = pgd_addr_end(addr, end);
599 if (pgd_none_or_clear_bad(src_pgd))
600 continue;
601 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
602 vma, addr, next))
603 return -ENOMEM;
604 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
605 return 0;
606}
607
Robin Holt51c6f662005-11-13 16:06:42 -0800608static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700609 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800611 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612{
Nick Pigginb5810032005-10-29 18:16:12 -0700613 struct mm_struct *mm = tlb->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 pte_t *pte;
Hugh Dickins508034a2005-10-29 18:16:30 -0700615 spinlock_t *ptl;
Hugh Dickinsae859762005-10-29 18:16:05 -0700616 int file_rss = 0;
617 int anon_rss = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
Hugh Dickins508034a2005-10-29 18:16:30 -0700619 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 do {
621 pte_t ptent = *pte;
Robin Holt51c6f662005-11-13 16:06:42 -0800622 if (pte_none(ptent)) {
623 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800625 }
Hugh Dickins6f5e6b92006-03-16 23:04:09 -0800626
627 (*zap_work) -= PAGE_SIZE;
628
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800630 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -0800631
Linus Torvalds6aab3412005-11-28 14:34:23 -0800632 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 if (unlikely(details) && page) {
634 /*
635 * unmap_shared_mapping_pages() wants to
636 * invalidate cache without truncating:
637 * unmap shared but keep private pages.
638 */
639 if (details->check_mapping &&
640 details->check_mapping != page->mapping)
641 continue;
642 /*
643 * Each page->index must be checked when
644 * invalidating or truncating nonlinear.
645 */
646 if (details->nonlinear_vma &&
647 (page->index < details->first_index ||
648 page->index > details->last_index))
649 continue;
650 }
Nick Pigginb5810032005-10-29 18:16:12 -0700651 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -0700652 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 tlb_remove_tlb_entry(tlb, pte, addr);
654 if (unlikely(!page))
655 continue;
656 if (unlikely(details) && details->nonlinear_vma
657 && linear_page_index(details->nonlinear_vma,
658 addr) != page->index)
Nick Pigginb5810032005-10-29 18:16:12 -0700659 set_pte_at(mm, addr, pte,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 pgoff_to_pte(page->index));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 if (PageAnon(page))
Hugh Dickins86d912f2005-10-29 18:16:14 -0700662 anon_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700663 else {
664 if (pte_dirty(ptent))
665 set_page_dirty(page);
666 if (pte_young(ptent))
667 mark_page_accessed(page);
Hugh Dickins86d912f2005-10-29 18:16:14 -0700668 file_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700669 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 page_remove_rmap(page);
671 tlb_remove_page(tlb, page);
672 continue;
673 }
674 /*
675 * If details->check_mapping, we leave swap entries;
676 * if details->nonlinear_vma, we leave file entries.
677 */
678 if (unlikely(details))
679 continue;
680 if (!pte_file(ptent))
681 free_swap_and_cache(pte_to_swp_entry(ptent));
Nick Pigginb5810032005-10-29 18:16:12 -0700682 pte_clear_full(mm, addr, pte, tlb->fullmm);
Robin Holt51c6f662005-11-13 16:06:42 -0800683 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
Hugh Dickinsae859762005-10-29 18:16:05 -0700684
Hugh Dickins86d912f2005-10-29 18:16:14 -0700685 add_mm_rss(mm, file_rss, anon_rss);
Hugh Dickins508034a2005-10-29 18:16:30 -0700686 pte_unmap_unlock(pte - 1, ptl);
Robin Holt51c6f662005-11-13 16:06:42 -0800687
688 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689}
690
Robin Holt51c6f662005-11-13 16:06:42 -0800691static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700692 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800694 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695{
696 pmd_t *pmd;
697 unsigned long next;
698
699 pmd = pmd_offset(pud, addr);
700 do {
701 next = pmd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800702 if (pmd_none_or_clear_bad(pmd)) {
703 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800705 }
706 next = zap_pte_range(tlb, vma, pmd, addr, next,
707 zap_work, details);
708 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
709
710 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711}
712
Robin Holt51c6f662005-11-13 16:06:42 -0800713static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700714 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800716 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717{
718 pud_t *pud;
719 unsigned long next;
720
721 pud = pud_offset(pgd, addr);
722 do {
723 next = pud_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800724 if (pud_none_or_clear_bad(pud)) {
725 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800727 }
728 next = zap_pmd_range(tlb, vma, pud, addr, next,
729 zap_work, details);
730 } while (pud++, addr = next, (addr != end && *zap_work > 0));
731
732 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733}
734
Robin Holt51c6f662005-11-13 16:06:42 -0800735static unsigned long unmap_page_range(struct mmu_gather *tlb,
736 struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800738 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739{
740 pgd_t *pgd;
741 unsigned long next;
742
743 if (details && !details->check_mapping && !details->nonlinear_vma)
744 details = NULL;
745
746 BUG_ON(addr >= end);
747 tlb_start_vma(tlb, vma);
748 pgd = pgd_offset(vma->vm_mm, addr);
749 do {
750 next = pgd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800751 if (pgd_none_or_clear_bad(pgd)) {
752 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800754 }
755 next = zap_pud_range(tlb, vma, pgd, addr, next,
756 zap_work, details);
757 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 tlb_end_vma(tlb, vma);
Robin Holt51c6f662005-11-13 16:06:42 -0800759
760 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761}
762
763#ifdef CONFIG_PREEMPT
764# define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
765#else
766/* No preempt: go for improved straight-line efficiency */
767# define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
768#endif
769
770/**
771 * unmap_vmas - unmap a range of memory covered by a list of vma's
772 * @tlbp: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 * @vma: the starting vma
774 * @start_addr: virtual address at which to start unmapping
775 * @end_addr: virtual address at which to end unmapping
776 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
777 * @details: details of nonlinear truncation or shared cache invalidation
778 *
Hugh Dickinsee39b372005-04-19 13:29:15 -0700779 * Returns the end address of the unmapping (restart addr if interrupted).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700781 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700783 * We aim to not hold locks for too long (for scheduling latency reasons).
784 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 * return the ending mmu_gather to the caller.
786 *
787 * Only addresses between `start' and `end' will be unmapped.
788 *
789 * The VMA list must be sorted in ascending virtual address order.
790 *
791 * unmap_vmas() assumes that the caller will flush the whole unmapped address
792 * range after unmap_vmas() returns. So the only responsibility here is to
793 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
794 * drops the lock and schedules.
795 */
Hugh Dickins508034a2005-10-29 18:16:30 -0700796unsigned long unmap_vmas(struct mmu_gather **tlbp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 struct vm_area_struct *vma, unsigned long start_addr,
798 unsigned long end_addr, unsigned long *nr_accounted,
799 struct zap_details *details)
800{
Robin Holt51c6f662005-11-13 16:06:42 -0800801 long zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
803 int tlb_start_valid = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -0700804 unsigned long start = start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700806 int fullmm = (*tlbp)->fullmm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807
808 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 unsigned long end;
810
811 start = max(vma->vm_start, start_addr);
812 if (start >= vma->vm_end)
813 continue;
814 end = min(vma->vm_end, end_addr);
815 if (end <= vma->vm_start)
816 continue;
817
818 if (vma->vm_flags & VM_ACCOUNT)
819 *nr_accounted += (end - start) >> PAGE_SHIFT;
820
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 while (start != end) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 if (!tlb_start_valid) {
823 tlb_start = start;
824 tlb_start_valid = 1;
825 }
826
Robin Holt51c6f662005-11-13 16:06:42 -0800827 if (unlikely(is_vm_hugetlb_page(vma))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 unmap_hugepage_range(vma, start, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800829 zap_work -= (end - start) /
830 (HPAGE_SIZE / PAGE_SIZE);
831 start = end;
832 } else
833 start = unmap_page_range(*tlbp, vma,
834 start, end, &zap_work, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835
Robin Holt51c6f662005-11-13 16:06:42 -0800836 if (zap_work > 0) {
837 BUG_ON(start != end);
838 break;
839 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
841 tlb_finish_mmu(*tlbp, tlb_start, start);
842
843 if (need_resched() ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 (i_mmap_lock && need_lockbreak(i_mmap_lock))) {
845 if (i_mmap_lock) {
Hugh Dickins508034a2005-10-29 18:16:30 -0700846 *tlbp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 goto out;
848 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 }
851
Hugh Dickins508034a2005-10-29 18:16:30 -0700852 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 tlb_start_valid = 0;
Robin Holt51c6f662005-11-13 16:06:42 -0800854 zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 }
856 }
857out:
Hugh Dickinsee39b372005-04-19 13:29:15 -0700858 return start; /* which is now the end (or restart) address */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859}
860
861/**
862 * zap_page_range - remove user pages in a given range
863 * @vma: vm_area_struct holding the applicable pages
864 * @address: starting address of pages to zap
865 * @size: number of bytes to zap
866 * @details: details of nonlinear truncation or shared cache invalidation
867 */
Hugh Dickinsee39b372005-04-19 13:29:15 -0700868unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 unsigned long size, struct zap_details *details)
870{
871 struct mm_struct *mm = vma->vm_mm;
872 struct mmu_gather *tlb;
873 unsigned long end = address + size;
874 unsigned long nr_accounted = 0;
875
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -0700878 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -0700879 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
880 if (tlb)
881 tlb_finish_mmu(tlb, address, end);
Hugh Dickinsee39b372005-04-19 13:29:15 -0700882 return end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883}
884
885/*
886 * Do a quick page-table lookup for a single page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800888struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700889 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890{
891 pgd_t *pgd;
892 pud_t *pud;
893 pmd_t *pmd;
894 pte_t *ptep, pte;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700895 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 struct page *page;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800897 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700899 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
900 if (!IS_ERR(page)) {
901 BUG_ON(flags & FOLL_GET);
902 goto out;
903 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700905 page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 pgd = pgd_offset(mm, address);
907 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700908 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909
910 pud = pud_offset(pgd, address);
911 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700912 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913
914 pmd = pmd_offset(pud, address);
915 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700916 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700918 if (pmd_huge(*pmd)) {
919 BUG_ON(flags & FOLL_GET);
920 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
921 goto out;
922 }
923
924 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 if (!ptep)
926 goto out;
927
928 pte = *ptep;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700929 if (!pte_present(pte))
930 goto unlock;
931 if ((flags & FOLL_WRITE) && !pte_write(pte))
932 goto unlock;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800933 page = vm_normal_page(vma, address, pte);
934 if (unlikely(!page))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700935 goto unlock;
936
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700937 if (flags & FOLL_GET)
938 get_page(page);
939 if (flags & FOLL_TOUCH) {
940 if ((flags & FOLL_WRITE) &&
941 !pte_dirty(pte) && !PageDirty(page))
942 set_page_dirty(page);
943 mark_page_accessed(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700945unlock:
946 pte_unmap_unlock(ptep, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947out:
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700948 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700950no_page_table:
951 /*
952 * When core dumping an enormous anonymous area that nobody
953 * has touched so far, we don't want to allocate page tables.
954 */
955 if (flags & FOLL_ANON) {
956 page = ZERO_PAGE(address);
957 if (flags & FOLL_GET)
958 get_page(page);
959 BUG_ON(flags & FOLL_WRITE);
960 }
961 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962}
963
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
965 unsigned long start, int len, int write, int force,
966 struct page **pages, struct vm_area_struct **vmas)
967{
968 int i;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700969 unsigned int vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970
971 /*
972 * Require read or write permissions.
973 * If 'force' is set, we only require the "MAY" flags.
974 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700975 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
976 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 i = 0;
978
979 do {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700980 struct vm_area_struct *vma;
981 unsigned int foll_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982
983 vma = find_extend_vma(mm, start);
984 if (!vma && in_gate_area(tsk, start)) {
985 unsigned long pg = start & PAGE_MASK;
986 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
987 pgd_t *pgd;
988 pud_t *pud;
989 pmd_t *pmd;
990 pte_t *pte;
991 if (write) /* user gate pages are read-only */
992 return i ? : -EFAULT;
993 if (pg > TASK_SIZE)
994 pgd = pgd_offset_k(pg);
995 else
996 pgd = pgd_offset_gate(mm, pg);
997 BUG_ON(pgd_none(*pgd));
998 pud = pud_offset(pgd, pg);
999 BUG_ON(pud_none(*pud));
1000 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001001 if (pmd_none(*pmd))
1002 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001004 if (pte_none(*pte)) {
1005 pte_unmap(pte);
1006 return i ? : -EFAULT;
1007 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 if (pages) {
Nick Pigginfa2a4552005-11-29 18:43:17 +11001009 struct page *page = vm_normal_page(gate_vma, start, *pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001010 pages[i] = page;
1011 if (page)
1012 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 }
1014 pte_unmap(pte);
1015 if (vmas)
1016 vmas[i] = gate_vma;
1017 i++;
1018 start += PAGE_SIZE;
1019 len--;
1020 continue;
1021 }
1022
Linus Torvalds1ff80382005-12-12 16:24:33 -08001023 if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001024 || !(vm_flags & vma->vm_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 return i ? : -EFAULT;
1026
1027 if (is_vm_hugetlb_page(vma)) {
1028 i = follow_hugetlb_page(mm, vma, pages, vmas,
1029 &start, &len, i);
1030 continue;
1031 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001032
1033 foll_flags = FOLL_TOUCH;
1034 if (pages)
1035 foll_flags |= FOLL_GET;
1036 if (!write && !(vma->vm_flags & VM_LOCKED) &&
1037 (!vma->vm_ops || !vma->vm_ops->nopage))
1038 foll_flags |= FOLL_ANON;
1039
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 do {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001041 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001043 if (write)
1044 foll_flags |= FOLL_WRITE;
1045
1046 cond_resched();
Linus Torvalds6aab3412005-11-28 14:34:23 -08001047 while (!(page = follow_page(vma, start, foll_flags))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001048 int ret;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001049 ret = __handle_mm_fault(mm, vma, start,
1050 foll_flags & FOLL_WRITE);
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001051 /*
1052 * The VM_FAULT_WRITE bit tells us that do_wp_page has
1053 * broken COW when necessary, even if maybe_mkwrite
1054 * decided not to set pte_write. We can thus safely do
1055 * subsequent page lookups as if they were reads.
1056 */
1057 if (ret & VM_FAULT_WRITE)
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001058 foll_flags &= ~FOLL_WRITE;
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001059
1060 switch (ret & ~VM_FAULT_WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 case VM_FAULT_MINOR:
1062 tsk->min_flt++;
1063 break;
1064 case VM_FAULT_MAJOR:
1065 tsk->maj_flt++;
1066 break;
1067 case VM_FAULT_SIGBUS:
1068 return i ? i : -EFAULT;
1069 case VM_FAULT_OOM:
1070 return i ? i : -ENOMEM;
1071 default:
1072 BUG();
1073 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 }
1075 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001076 pages[i] = page;
1077 flush_dcache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 }
1079 if (vmas)
1080 vmas[i] = vma;
1081 i++;
1082 start += PAGE_SIZE;
1083 len--;
Hugh Dickins08ef4722005-06-21 17:15:10 -07001084 } while (len && start < vma->vm_end);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001085 } while (len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 return i;
1087}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088EXPORT_SYMBOL(get_user_pages);
1089
1090static int zeromap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1091 unsigned long addr, unsigned long end, pgprot_t prot)
1092{
1093 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001094 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
Hugh Dickinsc74df322005-10-29 18:16:23 -07001096 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 if (!pte)
1098 return -ENOMEM;
1099 do {
Nick Pigginb5810032005-10-29 18:16:12 -07001100 struct page *page = ZERO_PAGE(addr);
1101 pte_t zero_pte = pte_wrprotect(mk_pte(page, prot));
1102 page_cache_get(page);
1103 page_add_file_rmap(page);
1104 inc_mm_counter(mm, file_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 BUG_ON(!pte_none(*pte));
1106 set_pte_at(mm, addr, pte, zero_pte);
1107 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickinsc74df322005-10-29 18:16:23 -07001108 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 return 0;
1110}
1111
1112static inline int zeromap_pmd_range(struct mm_struct *mm, pud_t *pud,
1113 unsigned long addr, unsigned long end, pgprot_t prot)
1114{
1115 pmd_t *pmd;
1116 unsigned long next;
1117
1118 pmd = pmd_alloc(mm, pud, addr);
1119 if (!pmd)
1120 return -ENOMEM;
1121 do {
1122 next = pmd_addr_end(addr, end);
1123 if (zeromap_pte_range(mm, pmd, addr, next, prot))
1124 return -ENOMEM;
1125 } while (pmd++, addr = next, addr != end);
1126 return 0;
1127}
1128
1129static inline int zeromap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1130 unsigned long addr, unsigned long end, pgprot_t prot)
1131{
1132 pud_t *pud;
1133 unsigned long next;
1134
1135 pud = pud_alloc(mm, pgd, addr);
1136 if (!pud)
1137 return -ENOMEM;
1138 do {
1139 next = pud_addr_end(addr, end);
1140 if (zeromap_pmd_range(mm, pud, addr, next, prot))
1141 return -ENOMEM;
1142 } while (pud++, addr = next, addr != end);
1143 return 0;
1144}
1145
1146int zeromap_page_range(struct vm_area_struct *vma,
1147 unsigned long addr, unsigned long size, pgprot_t prot)
1148{
1149 pgd_t *pgd;
1150 unsigned long next;
1151 unsigned long end = addr + size;
1152 struct mm_struct *mm = vma->vm_mm;
1153 int err;
1154
1155 BUG_ON(addr >= end);
1156 pgd = pgd_offset(mm, addr);
1157 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 do {
1159 next = pgd_addr_end(addr, end);
1160 err = zeromap_pud_range(mm, pgd, addr, next, prot);
1161 if (err)
1162 break;
1163 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 return err;
1165}
1166
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001167pte_t * fastcall get_locked_pte(struct mm_struct *mm, unsigned long addr, spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001168{
1169 pgd_t * pgd = pgd_offset(mm, addr);
1170 pud_t * pud = pud_alloc(mm, pgd, addr);
1171 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001172 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001173 if (pmd)
1174 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1175 }
1176 return NULL;
1177}
1178
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001180 * This is the old fallback for page remapping.
1181 *
1182 * For historical reasons, it only allows reserved pages. Only
1183 * old drivers should use this, and they needed to mark their
1184 * pages reserved for the old functions anyway.
1185 */
1186static int insert_page(struct mm_struct *mm, unsigned long addr, struct page *page, pgprot_t prot)
1187{
1188 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001189 pte_t *pte;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001190 spinlock_t *ptl;
1191
1192 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001193 if (PageAnon(page))
Linus Torvalds238f58d2005-11-29 13:01:56 -08001194 goto out;
1195 retval = -ENOMEM;
1196 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001197 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001198 if (!pte)
1199 goto out;
1200 retval = -EBUSY;
1201 if (!pte_none(*pte))
1202 goto out_unlock;
1203
1204 /* Ok, finally just insert the thing.. */
1205 get_page(page);
1206 inc_mm_counter(mm, file_rss);
1207 page_add_file_rmap(page);
1208 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1209
1210 retval = 0;
1211out_unlock:
1212 pte_unmap_unlock(pte, ptl);
1213out:
1214 return retval;
1215}
1216
1217/*
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001218 * This allows drivers to insert individual pages they've allocated
1219 * into a user vma.
1220 *
1221 * The page has to be a nice clean _individual_ kernel allocation.
1222 * If you allocate a compound page, you need to have marked it as
1223 * such (__GFP_COMP), or manually just split the page up yourself
1224 * (which is mainly an issue of doing "set_page_count(page, 1)" for
1225 * each sub-page, and then freeing them one by one when you free
1226 * them rather than freeing it as a compound page).
1227 *
1228 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1229 * took an arbitrary page protection parameter. This doesn't allow
1230 * that. Your vma protection will have to be set up correctly, which
1231 * means that if you want a shared writable mapping, you'd better
1232 * ask for a shared writable mapping!
1233 *
1234 * The page does not need to be reserved.
1235 */
1236int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, struct page *page)
1237{
1238 if (addr < vma->vm_start || addr >= vma->vm_end)
1239 return -EFAULT;
1240 if (!page_count(page))
1241 return -EINVAL;
Linus Torvalds4d7672b2005-12-16 10:21:23 -08001242 vma->vm_flags |= VM_INSERTPAGE;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001243 return insert_page(vma->vm_mm, addr, page, vma->vm_page_prot);
1244}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001245EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001246
1247/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 * maps a range of physical memory into the requested pages. the old
1249 * mappings are removed. any references to nonexistent pages results
1250 * in null mappings (currently treated as "copy-on-access")
1251 */
1252static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1253 unsigned long addr, unsigned long end,
1254 unsigned long pfn, pgprot_t prot)
1255{
1256 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001257 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258
Hugh Dickinsc74df322005-10-29 18:16:23 -07001259 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 if (!pte)
1261 return -ENOMEM;
1262 do {
1263 BUG_ON(!pte_none(*pte));
Nick Pigginb5810032005-10-29 18:16:12 -07001264 set_pte_at(mm, addr, pte, pfn_pte(pfn, prot));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 pfn++;
1266 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickinsc74df322005-10-29 18:16:23 -07001267 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 return 0;
1269}
1270
1271static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1272 unsigned long addr, unsigned long end,
1273 unsigned long pfn, pgprot_t prot)
1274{
1275 pmd_t *pmd;
1276 unsigned long next;
1277
1278 pfn -= addr >> PAGE_SHIFT;
1279 pmd = pmd_alloc(mm, pud, addr);
1280 if (!pmd)
1281 return -ENOMEM;
1282 do {
1283 next = pmd_addr_end(addr, end);
1284 if (remap_pte_range(mm, pmd, addr, next,
1285 pfn + (addr >> PAGE_SHIFT), prot))
1286 return -ENOMEM;
1287 } while (pmd++, addr = next, addr != end);
1288 return 0;
1289}
1290
1291static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1292 unsigned long addr, unsigned long end,
1293 unsigned long pfn, pgprot_t prot)
1294{
1295 pud_t *pud;
1296 unsigned long next;
1297
1298 pfn -= addr >> PAGE_SHIFT;
1299 pud = pud_alloc(mm, pgd, addr);
1300 if (!pud)
1301 return -ENOMEM;
1302 do {
1303 next = pud_addr_end(addr, end);
1304 if (remap_pmd_range(mm, pud, addr, next,
1305 pfn + (addr >> PAGE_SHIFT), prot))
1306 return -ENOMEM;
1307 } while (pud++, addr = next, addr != end);
1308 return 0;
1309}
1310
1311/* Note: this is only safe if the mm semaphore is held when called. */
1312int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1313 unsigned long pfn, unsigned long size, pgprot_t prot)
1314{
1315 pgd_t *pgd;
1316 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001317 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 struct mm_struct *mm = vma->vm_mm;
1319 int err;
1320
1321 /*
1322 * Physically remapped pages are special. Tell the
1323 * rest of the world about it:
1324 * VM_IO tells people not to look at these pages
1325 * (accesses can have side effects).
Hugh Dickins0b14c172005-11-21 21:32:15 -08001326 * VM_RESERVED is specified all over the place, because
1327 * in 2.4 it kept swapout's vma scan off this vma; but
1328 * in 2.6 the LRU scan won't even find its pages, so this
1329 * flag means no more than count its pages in reserved_vm,
1330 * and omit it from core dump, even when VM_IO turned off.
Linus Torvalds6aab3412005-11-28 14:34:23 -08001331 * VM_PFNMAP tells the core MM that the base pages are just
1332 * raw PFN mappings, and do not have a "struct page" associated
1333 * with them.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001334 *
1335 * There's a horrible special case to handle copy-on-write
1336 * behaviour that some programs depend on. We mark the "original"
1337 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 */
Linus Torvalds67121172005-12-11 20:38:17 -08001339 if (is_cow_mapping(vma->vm_flags)) {
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001340 if (addr != vma->vm_start || end != vma->vm_end)
Linus Torvalds7fc7e2e2005-12-11 19:57:52 -08001341 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001342 vma->vm_pgoff = pfn;
1343 }
1344
Linus Torvalds6aab3412005-11-28 14:34:23 -08001345 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346
1347 BUG_ON(addr >= end);
1348 pfn -= addr >> PAGE_SHIFT;
1349 pgd = pgd_offset(mm, addr);
1350 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 do {
1352 next = pgd_addr_end(addr, end);
1353 err = remap_pud_range(mm, pgd, addr, next,
1354 pfn + (addr >> PAGE_SHIFT), prot);
1355 if (err)
1356 break;
1357 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 return err;
1359}
1360EXPORT_SYMBOL(remap_pfn_range);
1361
1362/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001363 * handle_pte_fault chooses page fault handler according to an entry
1364 * which was read non-atomically. Before making any commitment, on
1365 * those architectures or configurations (e.g. i386 with PAE) which
1366 * might give a mix of unmatched parts, do_swap_page and do_file_page
1367 * must check under lock before unmapping the pte and proceeding
1368 * (but do_wp_page is only called after already making such a check;
1369 * and do_anonymous_page and do_no_page can safely check later on).
1370 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001371static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001372 pte_t *page_table, pte_t orig_pte)
1373{
1374 int same = 1;
1375#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1376 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001377 spinlock_t *ptl = pte_lockptr(mm, pmd);
1378 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001379 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001380 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001381 }
1382#endif
1383 pte_unmap(page_table);
1384 return same;
1385}
1386
1387/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1389 * servicing faults for write access. In the normal case, do always want
1390 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1391 * that do not have writing enabled, when used by access_process_vm.
1392 */
1393static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1394{
1395 if (likely(vma->vm_flags & VM_WRITE))
1396 pte = pte_mkwrite(pte);
1397 return pte;
1398}
1399
Linus Torvalds6aab3412005-11-28 14:34:23 -08001400static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va)
1401{
1402 /*
1403 * If the source page was a PFN mapping, we don't have
1404 * a "struct page" for it. We do a best-effort copy by
1405 * just copying from the original user address. If that
1406 * fails, we just zero-fill it. Live with it.
1407 */
1408 if (unlikely(!src)) {
1409 void *kaddr = kmap_atomic(dst, KM_USER0);
Linus Torvalds5d2a2dbbc2005-11-29 14:07:55 -08001410 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1411
1412 /*
1413 * This really shouldn't fail, because the page is there
1414 * in the page tables. But it might just be unreadable,
1415 * in which case we just give up and fill the result with
1416 * zeroes.
1417 */
1418 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Linus Torvalds6aab3412005-11-28 14:34:23 -08001419 memset(kaddr, 0, PAGE_SIZE);
1420 kunmap_atomic(kaddr, KM_USER0);
1421 return;
1422
1423 }
1424 copy_user_highpage(dst, src, va);
1425}
1426
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 * This routine handles present pages, when users try to write
1429 * to a shared page. It is done by copying the page to a new address
1430 * and decrementing the shared-page counter for the old page.
1431 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 * Note that this routine assumes that the protection checks have been
1433 * done by the caller (the low-level page fault routine in most cases).
1434 * Thus we can safely just mark it writable once we've done any necessary
1435 * COW.
1436 *
1437 * We also mark the page dirty at this point even though the page will
1438 * change only once the write actually happens. This avoids a few races,
1439 * and potentially makes it more efficient.
1440 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001441 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1442 * but allow concurrent faults), with pte both mapped and locked.
1443 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001445static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1446 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001447 spinlock_t *ptl, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448{
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001449 struct page *old_page, *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 pte_t entry;
Hugh Dickins65500d22005-10-29 18:15:59 -07001451 int ret = VM_FAULT_MINOR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452
Linus Torvalds6aab3412005-11-28 14:34:23 -08001453 old_page = vm_normal_page(vma, address, orig_pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001454 if (!old_page)
1455 goto gotten;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456
Hugh Dickinsd296e9c2005-06-21 17:15:11 -07001457 if (PageAnon(old_page) && !TestSetPageLocked(old_page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 int reuse = can_share_swap_page(old_page);
1459 unlock_page(old_page);
1460 if (reuse) {
Ben Collinseca35132005-11-29 11:45:26 -08001461 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07001462 entry = pte_mkyoung(orig_pte);
1463 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 ptep_set_access_flags(vma, address, page_table, entry, 1);
1465 update_mmu_cache(vma, address, entry);
1466 lazy_mmu_prot_update(entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07001467 ret |= VM_FAULT_WRITE;
1468 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 }
1470 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471
1472 /*
1473 * Ok, we need to copy. Oh, well..
1474 */
Nick Pigginb5810032005-10-29 18:16:12 -07001475 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08001476gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001477 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478
1479 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07001480 goto oom;
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001481 if (old_page == ZERO_PAGE(address)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 new_page = alloc_zeroed_user_highpage(vma, address);
1483 if (!new_page)
Hugh Dickins65500d22005-10-29 18:15:59 -07001484 goto oom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 } else {
1486 new_page = alloc_page_vma(GFP_HIGHUSER, vma, address);
1487 if (!new_page)
Hugh Dickins65500d22005-10-29 18:15:59 -07001488 goto oom;
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001489 cow_user_page(new_page, old_page, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 }
Hugh Dickins65500d22005-10-29 18:15:59 -07001491
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 /*
1493 * Re-check the pte - we dropped the lock
1494 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001495 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07001496 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08001497 if (old_page) {
1498 page_remove_rmap(old_page);
1499 if (!PageAnon(old_page)) {
1500 dec_mm_counter(mm, file_rss);
1501 inc_mm_counter(mm, anon_rss);
1502 }
1503 } else
Hugh Dickins42946212005-10-29 18:16:05 -07001504 inc_mm_counter(mm, anon_rss);
Ben Collinseca35132005-11-29 11:45:26 -08001505 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07001506 entry = mk_pte(new_page, vma->vm_page_prot);
1507 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1508 ptep_establish(vma, address, page_table, entry);
1509 update_mmu_cache(vma, address, entry);
1510 lazy_mmu_prot_update(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 lru_cache_add_active(new_page);
Nick Piggin9617d952006-01-06 00:11:12 -08001512 page_add_new_anon_rmap(new_page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513
1514 /* Free the old page.. */
1515 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001516 ret |= VM_FAULT_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 }
Hugh Dickins920fc352005-11-21 21:32:17 -08001518 if (new_page)
1519 page_cache_release(new_page);
1520 if (old_page)
1521 page_cache_release(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001522unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001523 pte_unmap_unlock(page_table, ptl);
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001524 return ret;
Hugh Dickins65500d22005-10-29 18:15:59 -07001525oom:
Hugh Dickins920fc352005-11-21 21:32:17 -08001526 if (old_page)
1527 page_cache_release(old_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 return VM_FAULT_OOM;
1529}
1530
1531/*
1532 * Helper functions for unmap_mapping_range().
1533 *
1534 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1535 *
1536 * We have to restart searching the prio_tree whenever we drop the lock,
1537 * since the iterator is only valid while the lock is held, and anyway
1538 * a later vma might be split and reinserted earlier while lock dropped.
1539 *
1540 * The list of nonlinear vmas could be handled more efficiently, using
1541 * a placeholder, but handle it in the same way until a need is shown.
1542 * It is important to search the prio_tree before nonlinear list: a vma
1543 * may become nonlinear and be shifted from prio_tree to nonlinear list
1544 * while the lock is dropped; but never shifted from list to prio_tree.
1545 *
1546 * In order to make forward progress despite restarting the search,
1547 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1548 * quickly skip it next time around. Since the prio_tree search only
1549 * shows us those vmas affected by unmapping the range in question, we
1550 * can't efficiently keep all vmas in step with mapping->truncate_count:
1551 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1552 * mapping->truncate_count and vma->vm_truncate_count are protected by
1553 * i_mmap_lock.
1554 *
1555 * In order to make forward progress despite repeatedly restarting some
Hugh Dickinsee39b372005-04-19 13:29:15 -07001556 * large vma, note the restart_addr from unmap_vmas when it breaks out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 * and restart from that address when we reach that vma again. It might
1558 * have been split or merged, shrunk or extended, but never shifted: so
1559 * restart_addr remains valid so long as it remains in the vma's range.
1560 * unmap_mapping_range forces truncate_count to leap over page-aligned
1561 * values so we can save vma's restart_addr in its truncate_count field.
1562 */
1563#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1564
1565static void reset_vma_truncate_counts(struct address_space *mapping)
1566{
1567 struct vm_area_struct *vma;
1568 struct prio_tree_iter iter;
1569
1570 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1571 vma->vm_truncate_count = 0;
1572 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1573 vma->vm_truncate_count = 0;
1574}
1575
1576static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1577 unsigned long start_addr, unsigned long end_addr,
1578 struct zap_details *details)
1579{
1580 unsigned long restart_addr;
1581 int need_break;
1582
1583again:
1584 restart_addr = vma->vm_truncate_count;
1585 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1586 start_addr = restart_addr;
1587 if (start_addr >= end_addr) {
1588 /* Top of vma has been split off since last time */
1589 vma->vm_truncate_count = details->truncate_count;
1590 return 0;
1591 }
1592 }
1593
Hugh Dickinsee39b372005-04-19 13:29:15 -07001594 restart_addr = zap_page_range(vma, start_addr,
1595 end_addr - start_addr, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 need_break = need_resched() ||
1597 need_lockbreak(details->i_mmap_lock);
1598
Hugh Dickinsee39b372005-04-19 13:29:15 -07001599 if (restart_addr >= end_addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 /* We have now completed this vma: mark it so */
1601 vma->vm_truncate_count = details->truncate_count;
1602 if (!need_break)
1603 return 0;
1604 } else {
1605 /* Note restart_addr in vma's truncate_count field */
Hugh Dickinsee39b372005-04-19 13:29:15 -07001606 vma->vm_truncate_count = restart_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 if (!need_break)
1608 goto again;
1609 }
1610
1611 spin_unlock(details->i_mmap_lock);
1612 cond_resched();
1613 spin_lock(details->i_mmap_lock);
1614 return -EINTR;
1615}
1616
1617static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1618 struct zap_details *details)
1619{
1620 struct vm_area_struct *vma;
1621 struct prio_tree_iter iter;
1622 pgoff_t vba, vea, zba, zea;
1623
1624restart:
1625 vma_prio_tree_foreach(vma, &iter, root,
1626 details->first_index, details->last_index) {
1627 /* Skip quickly over those we have already dealt with */
1628 if (vma->vm_truncate_count == details->truncate_count)
1629 continue;
1630
1631 vba = vma->vm_pgoff;
1632 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1633 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1634 zba = details->first_index;
1635 if (zba < vba)
1636 zba = vba;
1637 zea = details->last_index;
1638 if (zea > vea)
1639 zea = vea;
1640
1641 if (unmap_mapping_range_vma(vma,
1642 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1643 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1644 details) < 0)
1645 goto restart;
1646 }
1647}
1648
1649static inline void unmap_mapping_range_list(struct list_head *head,
1650 struct zap_details *details)
1651{
1652 struct vm_area_struct *vma;
1653
1654 /*
1655 * In nonlinear VMAs there is no correspondence between virtual address
1656 * offset and file offset. So we must perform an exhaustive search
1657 * across *all* the pages in each nonlinear VMA, not just the pages
1658 * whose virtual address lies outside the file truncation point.
1659 */
1660restart:
1661 list_for_each_entry(vma, head, shared.vm_set.list) {
1662 /* Skip quickly over those we have already dealt with */
1663 if (vma->vm_truncate_count == details->truncate_count)
1664 continue;
1665 details->nonlinear_vma = vma;
1666 if (unmap_mapping_range_vma(vma, vma->vm_start,
1667 vma->vm_end, details) < 0)
1668 goto restart;
1669 }
1670}
1671
1672/**
1673 * unmap_mapping_range - unmap the portion of all mmaps
1674 * in the specified address_space corresponding to the specified
1675 * page range in the underlying file.
Martin Waitz3d410882005-06-23 22:05:21 -07001676 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 * @holebegin: byte in first page to unmap, relative to the start of
1678 * the underlying file. This will be rounded down to a PAGE_SIZE
1679 * boundary. Note that this is different from vmtruncate(), which
1680 * must keep the partial page. In contrast, we must get rid of
1681 * partial pages.
1682 * @holelen: size of prospective hole in bytes. This will be rounded
1683 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1684 * end of the file.
1685 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1686 * but 0 when invalidating pagecache, don't throw away private data.
1687 */
1688void unmap_mapping_range(struct address_space *mapping,
1689 loff_t const holebegin, loff_t const holelen, int even_cows)
1690{
1691 struct zap_details details;
1692 pgoff_t hba = holebegin >> PAGE_SHIFT;
1693 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1694
1695 /* Check for overflow. */
1696 if (sizeof(holelen) > sizeof(hlen)) {
1697 long long holeend =
1698 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1699 if (holeend & ~(long long)ULONG_MAX)
1700 hlen = ULONG_MAX - hba + 1;
1701 }
1702
1703 details.check_mapping = even_cows? NULL: mapping;
1704 details.nonlinear_vma = NULL;
1705 details.first_index = hba;
1706 details.last_index = hba + hlen - 1;
1707 if (details.last_index < details.first_index)
1708 details.last_index = ULONG_MAX;
1709 details.i_mmap_lock = &mapping->i_mmap_lock;
1710
1711 spin_lock(&mapping->i_mmap_lock);
1712
1713 /* serialize i_size write against truncate_count write */
1714 smp_wmb();
1715 /* Protect against page faults, and endless unmapping loops */
1716 mapping->truncate_count++;
1717 /*
1718 * For archs where spin_lock has inclusive semantics like ia64
1719 * this smp_mb() will prevent to read pagetable contents
1720 * before the truncate_count increment is visible to
1721 * other cpus.
1722 */
1723 smp_mb();
1724 if (unlikely(is_restart_addr(mapping->truncate_count))) {
1725 if (mapping->truncate_count == 0)
1726 reset_vma_truncate_counts(mapping);
1727 mapping->truncate_count++;
1728 }
1729 details.truncate_count = mapping->truncate_count;
1730
1731 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
1732 unmap_mapping_range_tree(&mapping->i_mmap, &details);
1733 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
1734 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
1735 spin_unlock(&mapping->i_mmap_lock);
1736}
1737EXPORT_SYMBOL(unmap_mapping_range);
1738
1739/*
1740 * Handle all mappings that got truncated by a "truncate()"
1741 * system call.
1742 *
1743 * NOTE! We have to be ready to update the memory sharing
1744 * between the file and the memory map for a potential last
1745 * incomplete page. Ugly, but necessary.
1746 */
1747int vmtruncate(struct inode * inode, loff_t offset)
1748{
1749 struct address_space *mapping = inode->i_mapping;
1750 unsigned long limit;
1751
1752 if (inode->i_size < offset)
1753 goto do_expand;
1754 /*
1755 * truncation of in-use swapfiles is disallowed - it would cause
1756 * subsequent swapout to scribble on the now-freed blocks.
1757 */
1758 if (IS_SWAPFILE(inode))
1759 goto out_busy;
1760 i_size_write(inode, offset);
1761 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
1762 truncate_inode_pages(mapping, offset);
1763 goto out_truncate;
1764
1765do_expand:
1766 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1767 if (limit != RLIM_INFINITY && offset > limit)
1768 goto out_sig;
1769 if (offset > inode->i_sb->s_maxbytes)
1770 goto out_big;
1771 i_size_write(inode, offset);
1772
1773out_truncate:
1774 if (inode->i_op && inode->i_op->truncate)
1775 inode->i_op->truncate(inode);
1776 return 0;
1777out_sig:
1778 send_sig(SIGXFSZ, current, 0);
1779out_big:
1780 return -EFBIG;
1781out_busy:
1782 return -ETXTBSY;
1783}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784EXPORT_SYMBOL(vmtruncate);
1785
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001786int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
1787{
1788 struct address_space *mapping = inode->i_mapping;
1789
1790 /*
1791 * If the underlying filesystem is not going to provide
1792 * a way to truncate a range of blocks (punch a hole) -
1793 * we should return failure right now.
1794 */
1795 if (!inode->i_op || !inode->i_op->truncate_range)
1796 return -ENOSYS;
1797
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001798 mutex_lock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001799 down_write(&inode->i_alloc_sem);
1800 unmap_mapping_range(mapping, offset, (end - offset), 1);
1801 truncate_inode_pages_range(mapping, offset, end);
1802 inode->i_op->truncate_range(inode, offset, end);
1803 up_write(&inode->i_alloc_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001804 mutex_unlock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08001805
1806 return 0;
1807}
1808EXPORT_SYMBOL(vmtruncate_range);
1809
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810/*
1811 * Primitive swap readahead code. We simply read an aligned block of
1812 * (1 << page_cluster) entries in the swap area. This method is chosen
1813 * because it doesn't cost us any seek time. We also make sure to queue
1814 * the 'original' request together with the readahead ones...
1815 *
1816 * This has been extended to use the NUMA policies from the mm triggering
1817 * the readahead.
1818 *
1819 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
1820 */
1821void swapin_readahead(swp_entry_t entry, unsigned long addr,struct vm_area_struct *vma)
1822{
1823#ifdef CONFIG_NUMA
1824 struct vm_area_struct *next_vma = vma ? vma->vm_next : NULL;
1825#endif
1826 int i, num;
1827 struct page *new_page;
1828 unsigned long offset;
1829
1830 /*
1831 * Get the number of handles we should do readahead io to.
1832 */
1833 num = valid_swaphandles(entry, &offset);
1834 for (i = 0; i < num; offset++, i++) {
1835 /* Ok, do the async read-ahead now */
1836 new_page = read_swap_cache_async(swp_entry(swp_type(entry),
1837 offset), vma, addr);
1838 if (!new_page)
1839 break;
1840 page_cache_release(new_page);
1841#ifdef CONFIG_NUMA
1842 /*
1843 * Find the next applicable VMA for the NUMA policy.
1844 */
1845 addr += PAGE_SIZE;
1846 if (addr == 0)
1847 vma = NULL;
1848 if (vma) {
1849 if (addr >= vma->vm_end) {
1850 vma = next_vma;
1851 next_vma = vma ? vma->vm_next : NULL;
1852 }
1853 if (vma && addr < vma->vm_start)
1854 vma = NULL;
1855 } else {
1856 if (next_vma && addr >= next_vma->vm_start) {
1857 vma = next_vma;
1858 next_vma = vma->vm_next;
1859 }
1860 }
1861#endif
1862 }
1863 lru_add_drain(); /* Push any new pages onto the LRU now */
1864}
1865
1866/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001867 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1868 * but allow concurrent faults), and pte mapped but not yet locked.
1869 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001871static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
1872 unsigned long address, pte_t *page_table, pmd_t *pmd,
1873 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001875 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 struct page *page;
Hugh Dickins65500d22005-10-29 18:15:59 -07001877 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 pte_t pte;
1879 int ret = VM_FAULT_MINOR;
1880
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001881 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001882 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07001883
1884 entry = pte_to_swp_entry(orig_pte);
Christoph Lameterb16664e2006-02-01 03:05:36 -08001885again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 page = lookup_swap_cache(entry);
1887 if (!page) {
1888 swapin_readahead(entry, address, vma);
1889 page = read_swap_cache_async(entry, vma, address);
1890 if (!page) {
1891 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001892 * Back out if somebody else faulted in this pte
1893 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001895 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 if (likely(pte_same(*page_table, orig_pte)))
1897 ret = VM_FAULT_OOM;
Hugh Dickins65500d22005-10-29 18:15:59 -07001898 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 }
1900
1901 /* Had to read the page from swap area: Major fault */
1902 ret = VM_FAULT_MAJOR;
1903 inc_page_state(pgmajfault);
1904 grab_swap_token();
1905 }
1906
1907 mark_page_accessed(page);
1908 lock_page(page);
Christoph Lameterb16664e2006-02-01 03:05:36 -08001909 if (!PageSwapCache(page)) {
1910 /* Page migration has occured */
1911 unlock_page(page);
1912 page_cache_release(page);
1913 goto again;
1914 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915
1916 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001917 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001919 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07001920 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07001921 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07001922
1923 if (unlikely(!PageUptodate(page))) {
1924 ret = VM_FAULT_SIGBUS;
1925 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 }
1927
1928 /* The page isn't present yet, go ahead with the fault. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929
Hugh Dickins42946212005-10-29 18:16:05 -07001930 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 pte = mk_pte(page, vma->vm_page_prot);
1932 if (write_access && can_share_swap_page(page)) {
1933 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
1934 write_access = 0;
1935 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936
1937 flush_icache_page(vma, page);
1938 set_pte_at(mm, address, page_table, pte);
1939 page_add_anon_rmap(page, vma, address);
1940
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001941 swap_free(entry);
1942 if (vm_swap_full())
1943 remove_exclusive_swap_page(page);
1944 unlock_page(page);
1945
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 if (write_access) {
1947 if (do_wp_page(mm, vma, address,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001948 page_table, pmd, ptl, pte) == VM_FAULT_OOM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 ret = VM_FAULT_OOM;
1950 goto out;
1951 }
1952
1953 /* No need to invalidate - it was non-present before */
1954 update_mmu_cache(vma, address, pte);
1955 lazy_mmu_prot_update(pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07001956unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001957 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958out:
1959 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07001960out_nomap:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001961 pte_unmap_unlock(page_table, ptl);
Kirill Korotaevb8107482005-05-16 21:53:50 -07001962 unlock_page(page);
1963 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001964 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965}
1966
1967/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001968 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1969 * but allow concurrent faults), and pte mapped but not yet locked.
1970 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001972static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
1973 unsigned long address, pte_t *page_table, pmd_t *pmd,
1974 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001976 struct page *page;
1977 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979
Linus Torvalds6aab3412005-11-28 14:34:23 -08001980 if (write_access) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 /* Allocate our own private page. */
1982 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983
1984 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07001985 goto oom;
1986 page = alloc_zeroed_user_highpage(vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 if (!page)
Hugh Dickins65500d22005-10-29 18:15:59 -07001988 goto oom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989
Hugh Dickins65500d22005-10-29 18:15:59 -07001990 entry = mk_pte(page, vma->vm_page_prot);
1991 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001992
1993 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
1994 if (!pte_none(*page_table))
1995 goto release;
1996 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 lru_cache_add_active(page);
Nick Piggin9617d952006-01-06 00:11:12 -08001998 page_add_new_anon_rmap(page, vma, address);
Nick Pigginb5810032005-10-29 18:16:12 -07001999 } else {
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002000 /* Map the ZERO_PAGE - vm_page_prot is readonly */
2001 page = ZERO_PAGE(address);
2002 page_cache_get(page);
2003 entry = mk_pte(page, vma->vm_page_prot);
2004
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002005 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002006 spin_lock(ptl);
2007 if (!pte_none(*page_table))
2008 goto release;
Nick Pigginb5810032005-10-29 18:16:12 -07002009 inc_mm_counter(mm, file_rss);
2010 page_add_file_rmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 }
2012
Hugh Dickins65500d22005-10-29 18:15:59 -07002013 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014
2015 /* No need to invalidate - it was non-present before */
Hugh Dickins65500d22005-10-29 18:15:59 -07002016 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 lazy_mmu_prot_update(entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002018unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002019 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 return VM_FAULT_MINOR;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002021release:
2022 page_cache_release(page);
2023 goto unlock;
Hugh Dickins65500d22005-10-29 18:15:59 -07002024oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 return VM_FAULT_OOM;
2026}
2027
2028/*
2029 * do_no_page() tries to create a new page mapping. It aggressively
2030 * tries to share with existing pages, but makes a separate copy if
2031 * the "write_access" parameter is true in order to avoid the next
2032 * page fault.
2033 *
2034 * As this is called only for pages that do not currently exist, we
2035 * do not need to flush old virtual caches or the TLB.
2036 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002037 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2038 * but allow concurrent faults), and pte mapped but not yet locked.
2039 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002041static int do_no_page(struct mm_struct *mm, struct vm_area_struct *vma,
2042 unsigned long address, pte_t *page_table, pmd_t *pmd,
2043 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002045 spinlock_t *ptl;
Hugh Dickins65500d22005-10-29 18:15:59 -07002046 struct page *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 struct address_space *mapping = NULL;
2048 pte_t entry;
2049 unsigned int sequence = 0;
2050 int ret = VM_FAULT_MINOR;
2051 int anon = 0;
2052
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 pte_unmap(page_table);
Hugh Dickins325f04d2005-11-29 16:55:48 +00002054 BUG_ON(vma->vm_flags & VM_PFNMAP);
2055
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 if (vma->vm_file) {
2057 mapping = vma->vm_file->f_mapping;
2058 sequence = mapping->truncate_count;
2059 smp_rmb(); /* serializes i_size against truncate_count */
2060 }
2061retry:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 new_page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret);
2063 /*
2064 * No smp_rmb is needed here as long as there's a full
2065 * spin_lock/unlock sequence inside the ->nopage callback
2066 * (for the pagecache lookup) that acts as an implicit
2067 * smp_mb() and prevents the i_size read to happen
2068 * after the next truncate_count read.
2069 */
2070
2071 /* no page was available -- either SIGBUS or OOM */
2072 if (new_page == NOPAGE_SIGBUS)
2073 return VM_FAULT_SIGBUS;
2074 if (new_page == NOPAGE_OOM)
2075 return VM_FAULT_OOM;
2076
2077 /*
2078 * Should we do an early C-O-W break?
2079 */
2080 if (write_access && !(vma->vm_flags & VM_SHARED)) {
2081 struct page *page;
2082
2083 if (unlikely(anon_vma_prepare(vma)))
2084 goto oom;
2085 page = alloc_page_vma(GFP_HIGHUSER, vma, address);
2086 if (!page)
2087 goto oom;
Hugh Dickins325f04d2005-11-29 16:55:48 +00002088 copy_user_highpage(page, new_page, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 page_cache_release(new_page);
2090 new_page = page;
2091 anon = 1;
2092 }
2093
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002094 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 /*
2096 * For a file-backed vma, someone could have truncated or otherwise
2097 * invalidated this page. If unmap_mapping_range got called,
2098 * retry getting the page.
2099 */
2100 if (mapping && unlikely(sequence != mapping->truncate_count)) {
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002101 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002103 cond_resched();
2104 sequence = mapping->truncate_count;
2105 smp_rmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 goto retry;
2107 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108
2109 /*
2110 * This silly early PAGE_DIRTY setting removes a race
2111 * due to the bad i386 page protection. But it's valid
2112 * for other architectures too.
2113 *
2114 * Note that if write_access is true, we either now have
2115 * an exclusive copy of the page, or this is a shared mapping,
2116 * so we can make it writable and dirty to avoid having to
2117 * handle that later.
2118 */
2119 /* Only go through if we didn't race with anybody else... */
2120 if (pte_none(*page_table)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 flush_icache_page(vma, new_page);
2122 entry = mk_pte(new_page, vma->vm_page_prot);
2123 if (write_access)
2124 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2125 set_pte_at(mm, address, page_table, entry);
2126 if (anon) {
Hugh Dickins42946212005-10-29 18:16:05 -07002127 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 lru_cache_add_active(new_page);
Nick Piggin9617d952006-01-06 00:11:12 -08002129 page_add_new_anon_rmap(new_page, vma, address);
Hugh Dickinsf57e88a2005-11-21 21:32:19 -08002130 } else {
Hugh Dickins42946212005-10-29 18:16:05 -07002131 inc_mm_counter(mm, file_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 page_add_file_rmap(new_page);
Hugh Dickins42946212005-10-29 18:16:05 -07002133 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 } else {
2135 /* One of our sibling threads was faster, back out. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002137 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 }
2139
2140 /* no need to invalidate: a not-present page shouldn't be cached */
2141 update_mmu_cache(vma, address, entry);
2142 lazy_mmu_prot_update(entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002143unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002144 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 return ret;
2146oom:
2147 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002148 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149}
2150
2151/*
2152 * Fault of a previously existing named mapping. Repopulate the pte
2153 * from the encoded file_pte if possible. This enables swappable
2154 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002155 *
2156 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2157 * but allow concurrent faults), and pte mapped but not yet locked.
2158 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002160static int do_file_page(struct mm_struct *mm, struct vm_area_struct *vma,
2161 unsigned long address, pte_t *page_table, pmd_t *pmd,
2162 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163{
Hugh Dickins65500d22005-10-29 18:15:59 -07002164 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 int err;
2166
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002167 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002168 return VM_FAULT_MINOR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169
Hugh Dickins65500d22005-10-29 18:15:59 -07002170 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
2171 /*
2172 * Page table corrupted: show pte and kill process.
2173 */
Nick Pigginb5810032005-10-29 18:16:12 -07002174 print_bad_pte(vma, orig_pte, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002175 return VM_FAULT_OOM;
2176 }
2177 /* We can then assume vm->vm_ops && vma->vm_ops->populate */
2178
2179 pgoff = pte_to_pgoff(orig_pte);
2180 err = vma->vm_ops->populate(vma, address & PAGE_MASK, PAGE_SIZE,
2181 vma->vm_page_prot, pgoff, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 if (err == -ENOMEM)
2183 return VM_FAULT_OOM;
2184 if (err)
2185 return VM_FAULT_SIGBUS;
2186 return VM_FAULT_MAJOR;
2187}
2188
2189/*
2190 * These routines also need to handle stuff like marking pages dirty
2191 * and/or accessed for architectures that don't do it in hardware (most
2192 * RISC architectures). The early dirtying is also good on the i386.
2193 *
2194 * There is also a hook called "update_mmu_cache()" that architectures
2195 * with external mmu caches can use to update those (ie the Sparc or
2196 * PowerPC hashed page tables that act as extended TLBs).
2197 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07002198 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2199 * but allow concurrent faults), and pte mapped but not yet locked.
2200 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 */
2202static inline int handle_pte_fault(struct mm_struct *mm,
Hugh Dickins65500d22005-10-29 18:15:59 -07002203 struct vm_area_struct *vma, unsigned long address,
2204 pte_t *pte, pmd_t *pmd, int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205{
2206 pte_t entry;
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002207 pte_t old_entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002208 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002210 old_entry = entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002212 if (pte_none(entry)) {
2213 if (!vma->vm_ops || !vma->vm_ops->nopage)
2214 return do_anonymous_page(mm, vma, address,
2215 pte, pmd, write_access);
2216 return do_no_page(mm, vma, address,
2217 pte, pmd, write_access);
2218 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 if (pte_file(entry))
Hugh Dickins65500d22005-10-29 18:15:59 -07002220 return do_file_page(mm, vma, address,
2221 pte, pmd, write_access, entry);
2222 return do_swap_page(mm, vma, address,
2223 pte, pmd, write_access, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 }
2225
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002226 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002227 spin_lock(ptl);
2228 if (unlikely(!pte_same(*pte, entry)))
2229 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230 if (write_access) {
2231 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002232 return do_wp_page(mm, vma, address,
2233 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 entry = pte_mkdirty(entry);
2235 }
2236 entry = pte_mkyoung(entry);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002237 if (!pte_same(old_entry, entry)) {
2238 ptep_set_access_flags(vma, address, pte, entry, write_access);
2239 update_mmu_cache(vma, address, entry);
2240 lazy_mmu_prot_update(entry);
2241 } else {
2242 /*
2243 * This is needed only for protection faults but the arch code
2244 * is not yet telling us if this is a protection fault or not.
2245 * This still avoids useless tlb flushes for .text page faults
2246 * with threads.
2247 */
2248 if (write_access)
2249 flush_tlb_page(vma, address);
2250 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002251unlock:
2252 pte_unmap_unlock(pte, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 return VM_FAULT_MINOR;
2254}
2255
2256/*
2257 * By the time we get here, we already hold the mm semaphore
2258 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002259int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 unsigned long address, int write_access)
2261{
2262 pgd_t *pgd;
2263 pud_t *pud;
2264 pmd_t *pmd;
2265 pte_t *pte;
2266
2267 __set_current_state(TASK_RUNNING);
2268
2269 inc_page_state(pgfault);
2270
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01002271 if (unlikely(is_vm_hugetlb_page(vma)))
2272 return hugetlb_fault(mm, vma, address, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 pud = pud_alloc(mm, pgd, address);
2276 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002277 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 pmd = pmd_alloc(mm, pud, address);
2279 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002280 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 pte = pte_alloc_map(mm, pmd, address);
2282 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002283 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284
Hugh Dickinsc74df322005-10-29 18:16:23 -07002285 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286}
2287
Arnd Bergmann67207b92005-11-15 15:53:48 -05002288EXPORT_SYMBOL_GPL(__handle_mm_fault);
2289
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290#ifndef __PAGETABLE_PUD_FOLDED
2291/*
2292 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002293 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002295int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002297 pud_t *new = pud_alloc_one(mm, address);
2298 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002299 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300
Hugh Dickins872fec12005-10-29 18:16:21 -07002301 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002302 if (pgd_present(*pgd)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 pud_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002304 else
2305 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07002306 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002307 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308}
Alan Sterne0f39592005-11-28 13:43:44 -08002309#else
2310/* Workaround for gcc 2.96 */
2311int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
2312{
2313 return 0;
2314}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315#endif /* __PAGETABLE_PUD_FOLDED */
2316
2317#ifndef __PAGETABLE_PMD_FOLDED
2318/*
2319 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002320 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002322int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002324 pmd_t *new = pmd_alloc_one(mm, address);
2325 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002326 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327
Hugh Dickins872fec12005-10-29 18:16:21 -07002328 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07002330 if (pud_present(*pud)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 pmd_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002332 else
2333 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07002335 if (pgd_present(*pud)) /* Another has populated it */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 pmd_free(new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002337 else
2338 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07002340 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002341 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342}
Alan Sterne0f39592005-11-28 13:43:44 -08002343#else
2344/* Workaround for gcc 2.96 */
2345int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
2346{
2347 return 0;
2348}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349#endif /* __PAGETABLE_PMD_FOLDED */
2350
2351int make_pages_present(unsigned long addr, unsigned long end)
2352{
2353 int ret, len, write;
2354 struct vm_area_struct * vma;
2355
2356 vma = find_vma(current->mm, addr);
2357 if (!vma)
2358 return -1;
2359 write = (vma->vm_flags & VM_WRITE) != 0;
2360 if (addr >= end)
2361 BUG();
2362 if (end > vma->vm_end)
2363 BUG();
2364 len = (end+PAGE_SIZE-1)/PAGE_SIZE-addr/PAGE_SIZE;
2365 ret = get_user_pages(current, current->mm, addr,
2366 len, write, 0, NULL, NULL);
2367 if (ret < 0)
2368 return ret;
2369 return ret == len ? 0 : -1;
2370}
2371
2372/*
2373 * Map a vmalloc()-space virtual address to the physical page.
2374 */
2375struct page * vmalloc_to_page(void * vmalloc_addr)
2376{
2377 unsigned long addr = (unsigned long) vmalloc_addr;
2378 struct page *page = NULL;
2379 pgd_t *pgd = pgd_offset_k(addr);
2380 pud_t *pud;
2381 pmd_t *pmd;
2382 pte_t *ptep, pte;
2383
2384 if (!pgd_none(*pgd)) {
2385 pud = pud_offset(pgd, addr);
2386 if (!pud_none(*pud)) {
2387 pmd = pmd_offset(pud, addr);
2388 if (!pmd_none(*pmd)) {
2389 ptep = pte_offset_map(pmd, addr);
2390 pte = *ptep;
2391 if (pte_present(pte))
2392 page = pte_page(pte);
2393 pte_unmap(ptep);
2394 }
2395 }
2396 }
2397 return page;
2398}
2399
2400EXPORT_SYMBOL(vmalloc_to_page);
2401
2402/*
2403 * Map a vmalloc()-space virtual address to the physical page frame number.
2404 */
2405unsigned long vmalloc_to_pfn(void * vmalloc_addr)
2406{
2407 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
2408}
2409
2410EXPORT_SYMBOL(vmalloc_to_pfn);
2411
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412#if !defined(__HAVE_ARCH_GATE_AREA)
2413
2414#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07002415static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416
2417static int __init gate_vma_init(void)
2418{
2419 gate_vma.vm_mm = NULL;
2420 gate_vma.vm_start = FIXADDR_USER_START;
2421 gate_vma.vm_end = FIXADDR_USER_END;
2422 gate_vma.vm_page_prot = PAGE_READONLY;
Hugh Dickins0b14c172005-11-21 21:32:15 -08002423 gate_vma.vm_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 return 0;
2425}
2426__initcall(gate_vma_init);
2427#endif
2428
2429struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2430{
2431#ifdef AT_SYSINFO_EHDR
2432 return &gate_vma;
2433#else
2434 return NULL;
2435#endif
2436}
2437
2438int in_gate_area_no_task(unsigned long addr)
2439{
2440#ifdef AT_SYSINFO_EHDR
2441 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2442 return 1;
2443#endif
2444 return 0;
2445}
2446
2447#endif /* __HAVE_ARCH_GATE_AREA */