blob: 95a4553c75f708d5d680fe429518f92d605c6381 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/memory.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 */
6
7/*
8 * demand-loading started 01.12.91 - seems it is high on the list of
9 * things wanted, and it should be easy to implement. - Linus
10 */
11
12/*
13 * Ok, demand-loading was easy, shared pages a little bit tricker. Shared
14 * pages started 02.12.91, seems to work. - Linus.
15 *
16 * Tested sharing by executing about 30 /bin/sh: under the old kernel it
17 * would have taken more than the 6M I have free, but it worked well as
18 * far as I could see.
19 *
20 * Also corrected some "invalidate()"s - I wasn't doing enough of them.
21 */
22
23/*
24 * Real VM (paging to/from disk) started 18.12.91. Much more work and
25 * thought has to go into this. Oh, well..
26 * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why.
27 * Found it. Everything seems to work now.
28 * 20.12.91 - Ok, making the swap-device changeable like the root.
29 */
30
31/*
32 * 05.04.94 - Multi-page memory management added for v1.1.
33 * Idea by Alex Bligh (alex@cconcepts.co.uk)
34 *
35 * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG
36 * (Gerhard.Wichert@pdb.siemens.de)
37 *
38 * Aug/Sep 2004 Changed to four level page tables (Andi Kleen)
39 */
40
41#include <linux/kernel_stat.h>
42#include <linux/mm.h>
43#include <linux/hugetlb.h>
44#include <linux/mman.h>
45#include <linux/swap.h>
46#include <linux/highmem.h>
47#include <linux/pagemap.h>
48#include <linux/rmap.h>
49#include <linux/module.h>
50#include <linux/init.h>
51
52#include <asm/pgalloc.h>
53#include <asm/uaccess.h>
54#include <asm/tlb.h>
55#include <asm/tlbflush.h>
56#include <asm/pgtable.h>
57
58#include <linux/swapops.h>
59#include <linux/elf.h>
60
Andy Whitcroftd41dee32005-06-23 00:07:54 -070061#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -070062/* use the per-pgdat data instead for discontigmem - mbligh */
63unsigned long max_mapnr;
64struct page *mem_map;
65
66EXPORT_SYMBOL(max_mapnr);
67EXPORT_SYMBOL(mem_map);
68#endif
69
70unsigned long num_physpages;
71/*
72 * A number of key systems in x86 including ioremap() rely on the assumption
73 * that high_memory defines the upper bound on direct map memory, then end
74 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
75 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
76 * and ZONE_HIGHMEM.
77 */
78void * high_memory;
79unsigned long vmalloc_earlyreserve;
80
81EXPORT_SYMBOL(num_physpages);
82EXPORT_SYMBOL(high_memory);
83EXPORT_SYMBOL(vmalloc_earlyreserve);
84
85/*
86 * If a p?d_bad entry is found while walking page tables, report
87 * the error, before resetting entry to p?d_none. Usually (but
88 * very seldom) called out from the p?d_none_or_clear_bad macros.
89 */
90
91void pgd_clear_bad(pgd_t *pgd)
92{
93 pgd_ERROR(*pgd);
94 pgd_clear(pgd);
95}
96
97void pud_clear_bad(pud_t *pud)
98{
99 pud_ERROR(*pud);
100 pud_clear(pud);
101}
102
103void pmd_clear_bad(pmd_t *pmd)
104{
105 pmd_ERROR(*pmd);
106 pmd_clear(pmd);
107}
108
109/*
110 * Note: this doesn't free the actual pages themselves. That
111 * has been handled earlier when unmapping all the memory regions.
112 */
Hugh Dickinse0da3822005-04-19 13:29:15 -0700113static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114{
Hugh Dickinse0da3822005-04-19 13:29:15 -0700115 struct page *page = pmd_page(*pmd);
116 pmd_clear(pmd);
117 pte_free_tlb(tlb, page);
118 dec_page_state(nr_page_table_pages);
119 tlb->mm->nr_ptes--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120}
121
Hugh Dickinse0da3822005-04-19 13:29:15 -0700122static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
123 unsigned long addr, unsigned long end,
124 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125{
126 pmd_t *pmd;
127 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700128 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129
Hugh Dickinse0da3822005-04-19 13:29:15 -0700130 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 do {
133 next = pmd_addr_end(addr, end);
134 if (pmd_none_or_clear_bad(pmd))
135 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700136 free_pte_range(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 } while (pmd++, addr = next, addr != end);
138
Hugh Dickinse0da3822005-04-19 13:29:15 -0700139 start &= PUD_MASK;
140 if (start < floor)
141 return;
142 if (ceiling) {
143 ceiling &= PUD_MASK;
144 if (!ceiling)
145 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700147 if (end - 1 > ceiling - 1)
148 return;
149
150 pmd = pmd_offset(pud, start);
151 pud_clear(pud);
152 pmd_free_tlb(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153}
154
Hugh Dickinse0da3822005-04-19 13:29:15 -0700155static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
156 unsigned long addr, unsigned long end,
157 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158{
159 pud_t *pud;
160 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700161 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
Hugh Dickinse0da3822005-04-19 13:29:15 -0700163 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 pud = pud_offset(pgd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 do {
166 next = pud_addr_end(addr, end);
167 if (pud_none_or_clear_bad(pud))
168 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700169 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 } while (pud++, addr = next, addr != end);
171
Hugh Dickinse0da3822005-04-19 13:29:15 -0700172 start &= PGDIR_MASK;
173 if (start < floor)
174 return;
175 if (ceiling) {
176 ceiling &= PGDIR_MASK;
177 if (!ceiling)
178 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700180 if (end - 1 > ceiling - 1)
181 return;
182
183 pud = pud_offset(pgd, start);
184 pgd_clear(pgd);
185 pud_free_tlb(tlb, pud);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186}
187
188/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700189 * This function frees user-level page tables of a process.
190 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 * Must be called with pagetable lock held.
192 */
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700193void free_pgd_range(struct mmu_gather **tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700194 unsigned long addr, unsigned long end,
195 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196{
197 pgd_t *pgd;
198 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700199 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200
Hugh Dickinse0da3822005-04-19 13:29:15 -0700201 /*
202 * The next few lines have given us lots of grief...
203 *
204 * Why are we testing PMD* at this top level? Because often
205 * there will be no work to do at all, and we'd prefer not to
206 * go all the way down to the bottom just to discover that.
207 *
208 * Why all these "- 1"s? Because 0 represents both the bottom
209 * of the address space and the top of it (using -1 for the
210 * top wouldn't help much: the masks would do the wrong thing).
211 * The rule is that addr 0 and floor 0 refer to the bottom of
212 * the address space, but end 0 and ceiling 0 refer to the top
213 * Comparisons need to use "end - 1" and "ceiling - 1" (though
214 * that end 0 case should be mythical).
215 *
216 * Wherever addr is brought up or ceiling brought down, we must
217 * be careful to reject "the opposite 0" before it confuses the
218 * subsequent tests. But what about where end is brought down
219 * by PMD_SIZE below? no, end can't go down to 0 there.
220 *
221 * Whereas we round start (addr) and ceiling down, by different
222 * masks at different levels, in order to test whether a table
223 * now has no other vmas using it, so can be freed, we don't
224 * bother to round floor or end up - the tests don't need that.
225 */
226
227 addr &= PMD_MASK;
228 if (addr < floor) {
229 addr += PMD_SIZE;
230 if (!addr)
231 return;
232 }
233 if (ceiling) {
234 ceiling &= PMD_MASK;
235 if (!ceiling)
236 return;
237 }
238 if (end - 1 > ceiling - 1)
239 end -= PMD_SIZE;
240 if (addr > end - 1)
241 return;
242
243 start = addr;
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700244 pgd = pgd_offset((*tlb)->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 do {
246 next = pgd_addr_end(addr, end);
247 if (pgd_none_or_clear_bad(pgd))
248 continue;
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700249 free_pud_range(*tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700251
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700252 if (!(*tlb)->fullmm)
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700253 flush_tlb_pgtables((*tlb)->mm, start, end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700254}
255
256void free_pgtables(struct mmu_gather **tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700257 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700258{
259 while (vma) {
260 struct vm_area_struct *next = vma->vm_next;
261 unsigned long addr = vma->vm_start;
262
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700263 if (is_hugepage_only_range(vma->vm_mm, addr, HPAGE_SIZE)) {
264 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700265 floor, next? next->vm_start: ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700266 } else {
267 /*
268 * Optimization: gather nearby vmas into one call down
269 */
270 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
271 && !is_hugepage_only_range(vma->vm_mm, next->vm_start,
272 HPAGE_SIZE)) {
273 vma = next;
274 next = vma->vm_next;
275 }
276 free_pgd_range(tlb, addr, vma->vm_end,
277 floor, next? next->vm_start: ceiling);
278 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700279 vma = next;
280 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281}
282
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700283pte_t fastcall *pte_alloc_map(struct mm_struct *mm, pmd_t *pmd,
284 unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285{
286 if (!pmd_present(*pmd)) {
287 struct page *new;
288
289 spin_unlock(&mm->page_table_lock);
290 new = pte_alloc_one(mm, address);
291 spin_lock(&mm->page_table_lock);
292 if (!new)
293 return NULL;
294 /*
295 * Because we dropped the lock, we should re-check the
296 * entry, as somebody else could have populated it..
297 */
298 if (pmd_present(*pmd)) {
299 pte_free(new);
300 goto out;
301 }
302 mm->nr_ptes++;
303 inc_page_state(nr_page_table_pages);
304 pmd_populate(mm, pmd, new);
305 }
306out:
307 return pte_offset_map(pmd, address);
308}
309
Hugh Dickins872fec12005-10-29 18:16:21 -0700310pte_t fastcall * pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311{
312 if (!pmd_present(*pmd)) {
313 pte_t *new;
314
Hugh Dickins872fec12005-10-29 18:16:21 -0700315 new = pte_alloc_one_kernel(&init_mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 if (!new)
317 return NULL;
318
Hugh Dickins872fec12005-10-29 18:16:21 -0700319 spin_lock(&init_mm.page_table_lock);
320 if (pmd_present(*pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 pte_free_kernel(new);
Hugh Dickins872fec12005-10-29 18:16:21 -0700322 else
323 pmd_populate_kernel(&init_mm, pmd, new);
324 spin_unlock(&init_mm.page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 return pte_offset_kernel(pmd, address);
327}
328
Hugh Dickinsae859762005-10-29 18:16:05 -0700329static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
330{
331 if (file_rss)
332 add_mm_counter(mm, file_rss, file_rss);
333 if (anon_rss)
334 add_mm_counter(mm, anon_rss, anon_rss);
335}
336
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337/*
Nick Pigginb5810032005-10-29 18:16:12 -0700338 * This function is called to print an error when a pte in a
339 * !VM_RESERVED region is found pointing to an invalid pfn (which
340 * is an error.
341 *
342 * The calling function must still handle the error.
343 */
344void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr)
345{
346 printk(KERN_ERR "Bad pte = %08llx, process = %s, "
347 "vm_flags = %lx, vaddr = %lx\n",
348 (long long)pte_val(pte),
349 (vma->vm_mm == current->mm ? current->comm : "???"),
350 vma->vm_flags, vaddr);
351 dump_stack();
352}
353
354/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 * copy one vm_area from one task to the other. Assumes the page tables
356 * already present in the new task to be cleared in the whole range
357 * covered by this vma.
358 *
359 * dst->page_table_lock is held on entry and exit,
360 * but may be dropped within p[mg]d_alloc() and pte_alloc_map().
361 */
362
Hugh Dickins8c103762005-10-29 18:16:13 -0700363static inline void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700365 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700366 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367{
Nick Pigginb5810032005-10-29 18:16:12 -0700368 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 pte_t pte = *src_pte;
370 struct page *page;
371 unsigned long pfn;
372
373 /* pte contains position in swap or file, so copy. */
374 if (unlikely(!pte_present(pte))) {
375 if (!pte_file(pte)) {
376 swap_duplicate(pte_to_swp_entry(pte));
377 /* make sure dst_mm is on swapoff's mmlist. */
378 if (unlikely(list_empty(&dst_mm->mmlist))) {
379 spin_lock(&mmlist_lock);
380 list_add(&dst_mm->mmlist, &src_mm->mmlist);
381 spin_unlock(&mmlist_lock);
382 }
383 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700384 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 }
386
Nick Pigginb5810032005-10-29 18:16:12 -0700387 /* If the region is VM_RESERVED, the mapping is not
388 * mapped via rmap - duplicate the pte as is.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 */
Nick Pigginb5810032005-10-29 18:16:12 -0700390 if (vm_flags & VM_RESERVED)
Hugh Dickinsae859762005-10-29 18:16:05 -0700391 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392
Nick Pigginb5810032005-10-29 18:16:12 -0700393 pfn = pte_pfn(pte);
394 /* If the pte points outside of valid memory but
395 * the region is not VM_RESERVED, we have a problem.
396 */
397 if (unlikely(!pfn_valid(pfn))) {
398 print_bad_pte(vma, pte, addr);
399 goto out_set_pte; /* try to do something sane */
400 }
401
402 page = pfn_to_page(pfn);
403
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 /*
405 * If it's a COW mapping, write protect it both
406 * in the parent and the child
407 */
408 if ((vm_flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE) {
409 ptep_set_wrprotect(src_mm, addr, src_pte);
410 pte = *src_pte;
411 }
412
413 /*
414 * If it's a shared mapping, mark it clean in
415 * the child
416 */
417 if (vm_flags & VM_SHARED)
418 pte = pte_mkclean(pte);
419 pte = pte_mkold(pte);
420 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 page_dup_rmap(page);
Hugh Dickins8c103762005-10-29 18:16:13 -0700422 rss[!!PageAnon(page)]++;
Hugh Dickinsae859762005-10-29 18:16:05 -0700423
424out_set_pte:
425 set_pte_at(dst_mm, addr, dst_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426}
427
428static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
429 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
430 unsigned long addr, unsigned long end)
431{
432 pte_t *src_pte, *dst_pte;
Hugh Dickinse040f212005-10-29 18:15:53 -0700433 int progress = 0;
Hugh Dickins8c103762005-10-29 18:16:13 -0700434 int rss[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
436again:
Hugh Dickinsae859762005-10-29 18:16:05 -0700437 rss[1] = rss[0] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 dst_pte = pte_alloc_map(dst_mm, dst_pmd, addr);
439 if (!dst_pte)
440 return -ENOMEM;
441 src_pte = pte_offset_map_nested(src_pmd, addr);
442
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 spin_lock(&src_mm->page_table_lock);
444 do {
445 /*
446 * We are holding two locks at this point - either of them
447 * could generate latencies in another task on another CPU.
448 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700449 if (progress >= 32) {
450 progress = 0;
451 if (need_resched() ||
452 need_lockbreak(&src_mm->page_table_lock) ||
453 need_lockbreak(&dst_mm->page_table_lock))
454 break;
455 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 if (pte_none(*src_pte)) {
457 progress++;
458 continue;
459 }
Hugh Dickins8c103762005-10-29 18:16:13 -0700460 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 progress += 8;
462 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
463 spin_unlock(&src_mm->page_table_lock);
464
465 pte_unmap_nested(src_pte - 1);
466 pte_unmap(dst_pte - 1);
Hugh Dickinsae859762005-10-29 18:16:05 -0700467 add_mm_rss(dst_mm, rss[0], rss[1]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 cond_resched_lock(&dst_mm->page_table_lock);
469 if (addr != end)
470 goto again;
471 return 0;
472}
473
474static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
475 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
476 unsigned long addr, unsigned long end)
477{
478 pmd_t *src_pmd, *dst_pmd;
479 unsigned long next;
480
481 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
482 if (!dst_pmd)
483 return -ENOMEM;
484 src_pmd = pmd_offset(src_pud, addr);
485 do {
486 next = pmd_addr_end(addr, end);
487 if (pmd_none_or_clear_bad(src_pmd))
488 continue;
489 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
490 vma, addr, next))
491 return -ENOMEM;
492 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
493 return 0;
494}
495
496static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
497 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
498 unsigned long addr, unsigned long end)
499{
500 pud_t *src_pud, *dst_pud;
501 unsigned long next;
502
503 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
504 if (!dst_pud)
505 return -ENOMEM;
506 src_pud = pud_offset(src_pgd, addr);
507 do {
508 next = pud_addr_end(addr, end);
509 if (pud_none_or_clear_bad(src_pud))
510 continue;
511 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
512 vma, addr, next))
513 return -ENOMEM;
514 } while (dst_pud++, src_pud++, addr = next, addr != end);
515 return 0;
516}
517
518int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
519 struct vm_area_struct *vma)
520{
521 pgd_t *src_pgd, *dst_pgd;
522 unsigned long next;
523 unsigned long addr = vma->vm_start;
524 unsigned long end = vma->vm_end;
525
Nick Piggind9928952005-08-28 16:49:11 +1000526 /*
527 * Don't copy ptes where a page fault will fill them correctly.
528 * Fork becomes much lighter when there are big shared or private
529 * readonly mappings. The tradeoff is that copy_page_range is more
530 * efficient than faulting.
531 */
532 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_RESERVED))) {
533 if (!vma->anon_vma)
534 return 0;
535 }
536
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 if (is_vm_hugetlb_page(vma))
538 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
539
540 dst_pgd = pgd_offset(dst_mm, addr);
541 src_pgd = pgd_offset(src_mm, addr);
542 do {
543 next = pgd_addr_end(addr, end);
544 if (pgd_none_or_clear_bad(src_pgd))
545 continue;
546 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
547 vma, addr, next))
548 return -ENOMEM;
549 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
550 return 0;
551}
552
Nick Pigginb5810032005-10-29 18:16:12 -0700553static void zap_pte_range(struct mmu_gather *tlb,
554 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 unsigned long addr, unsigned long end,
556 struct zap_details *details)
557{
Nick Pigginb5810032005-10-29 18:16:12 -0700558 struct mm_struct *mm = tlb->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 pte_t *pte;
Hugh Dickinsae859762005-10-29 18:16:05 -0700560 int file_rss = 0;
561 int anon_rss = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562
563 pte = pte_offset_map(pmd, addr);
564 do {
565 pte_t ptent = *pte;
566 if (pte_none(ptent))
567 continue;
568 if (pte_present(ptent)) {
569 struct page *page = NULL;
Nick Pigginb5810032005-10-29 18:16:12 -0700570 if (!(vma->vm_flags & VM_RESERVED)) {
571 unsigned long pfn = pte_pfn(ptent);
572 if (unlikely(!pfn_valid(pfn)))
573 print_bad_pte(vma, ptent, addr);
574 else
575 page = pfn_to_page(pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 }
577 if (unlikely(details) && page) {
578 /*
579 * unmap_shared_mapping_pages() wants to
580 * invalidate cache without truncating:
581 * unmap shared but keep private pages.
582 */
583 if (details->check_mapping &&
584 details->check_mapping != page->mapping)
585 continue;
586 /*
587 * Each page->index must be checked when
588 * invalidating or truncating nonlinear.
589 */
590 if (details->nonlinear_vma &&
591 (page->index < details->first_index ||
592 page->index > details->last_index))
593 continue;
594 }
Nick Pigginb5810032005-10-29 18:16:12 -0700595 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -0700596 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 tlb_remove_tlb_entry(tlb, pte, addr);
598 if (unlikely(!page))
599 continue;
600 if (unlikely(details) && details->nonlinear_vma
601 && linear_page_index(details->nonlinear_vma,
602 addr) != page->index)
Nick Pigginb5810032005-10-29 18:16:12 -0700603 set_pte_at(mm, addr, pte,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 pgoff_to_pte(page->index));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 if (PageAnon(page))
Hugh Dickins86d912f2005-10-29 18:16:14 -0700606 anon_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700607 else {
608 if (pte_dirty(ptent))
609 set_page_dirty(page);
610 if (pte_young(ptent))
611 mark_page_accessed(page);
Hugh Dickins86d912f2005-10-29 18:16:14 -0700612 file_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700613 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 page_remove_rmap(page);
615 tlb_remove_page(tlb, page);
616 continue;
617 }
618 /*
619 * If details->check_mapping, we leave swap entries;
620 * if details->nonlinear_vma, we leave file entries.
621 */
622 if (unlikely(details))
623 continue;
624 if (!pte_file(ptent))
625 free_swap_and_cache(pte_to_swp_entry(ptent));
Nick Pigginb5810032005-10-29 18:16:12 -0700626 pte_clear_full(mm, addr, pte, tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickinsae859762005-10-29 18:16:05 -0700628
Hugh Dickins86d912f2005-10-29 18:16:14 -0700629 add_mm_rss(mm, file_rss, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 pte_unmap(pte - 1);
631}
632
Nick Pigginb5810032005-10-29 18:16:12 -0700633static inline void zap_pmd_range(struct mmu_gather *tlb,
634 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 unsigned long addr, unsigned long end,
636 struct zap_details *details)
637{
638 pmd_t *pmd;
639 unsigned long next;
640
641 pmd = pmd_offset(pud, addr);
642 do {
643 next = pmd_addr_end(addr, end);
644 if (pmd_none_or_clear_bad(pmd))
645 continue;
Nick Pigginb5810032005-10-29 18:16:12 -0700646 zap_pte_range(tlb, vma, pmd, addr, next, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 } while (pmd++, addr = next, addr != end);
648}
649
Nick Pigginb5810032005-10-29 18:16:12 -0700650static inline void zap_pud_range(struct mmu_gather *tlb,
651 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 unsigned long addr, unsigned long end,
653 struct zap_details *details)
654{
655 pud_t *pud;
656 unsigned long next;
657
658 pud = pud_offset(pgd, addr);
659 do {
660 next = pud_addr_end(addr, end);
661 if (pud_none_or_clear_bad(pud))
662 continue;
Nick Pigginb5810032005-10-29 18:16:12 -0700663 zap_pmd_range(tlb, vma, pud, addr, next, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 } while (pud++, addr = next, addr != end);
665}
666
667static void unmap_page_range(struct mmu_gather *tlb, struct vm_area_struct *vma,
668 unsigned long addr, unsigned long end,
669 struct zap_details *details)
670{
671 pgd_t *pgd;
672 unsigned long next;
673
674 if (details && !details->check_mapping && !details->nonlinear_vma)
675 details = NULL;
676
677 BUG_ON(addr >= end);
678 tlb_start_vma(tlb, vma);
679 pgd = pgd_offset(vma->vm_mm, addr);
680 do {
681 next = pgd_addr_end(addr, end);
682 if (pgd_none_or_clear_bad(pgd))
683 continue;
Nick Pigginb5810032005-10-29 18:16:12 -0700684 zap_pud_range(tlb, vma, pgd, addr, next, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 } while (pgd++, addr = next, addr != end);
686 tlb_end_vma(tlb, vma);
687}
688
689#ifdef CONFIG_PREEMPT
690# define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
691#else
692/* No preempt: go for improved straight-line efficiency */
693# define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
694#endif
695
696/**
697 * unmap_vmas - unmap a range of memory covered by a list of vma's
698 * @tlbp: address of the caller's struct mmu_gather
699 * @mm: the controlling mm_struct
700 * @vma: the starting vma
701 * @start_addr: virtual address at which to start unmapping
702 * @end_addr: virtual address at which to end unmapping
703 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
704 * @details: details of nonlinear truncation or shared cache invalidation
705 *
Hugh Dickinsee39b372005-04-19 13:29:15 -0700706 * Returns the end address of the unmapping (restart addr if interrupted).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 *
708 * Unmap all pages in the vma list. Called under page_table_lock.
709 *
710 * We aim to not hold page_table_lock for too long (for scheduling latency
711 * reasons). So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
712 * return the ending mmu_gather to the caller.
713 *
714 * Only addresses between `start' and `end' will be unmapped.
715 *
716 * The VMA list must be sorted in ascending virtual address order.
717 *
718 * unmap_vmas() assumes that the caller will flush the whole unmapped address
719 * range after unmap_vmas() returns. So the only responsibility here is to
720 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
721 * drops the lock and schedules.
722 */
Hugh Dickinsee39b372005-04-19 13:29:15 -0700723unsigned long unmap_vmas(struct mmu_gather **tlbp, struct mm_struct *mm,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 struct vm_area_struct *vma, unsigned long start_addr,
725 unsigned long end_addr, unsigned long *nr_accounted,
726 struct zap_details *details)
727{
728 unsigned long zap_bytes = ZAP_BLOCK_SIZE;
729 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
730 int tlb_start_valid = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -0700731 unsigned long start = start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700733 int fullmm = (*tlbp)->fullmm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734
735 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 unsigned long end;
737
738 start = max(vma->vm_start, start_addr);
739 if (start >= vma->vm_end)
740 continue;
741 end = min(vma->vm_end, end_addr);
742 if (end <= vma->vm_start)
743 continue;
744
745 if (vma->vm_flags & VM_ACCOUNT)
746 *nr_accounted += (end - start) >> PAGE_SHIFT;
747
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 while (start != end) {
749 unsigned long block;
750
751 if (!tlb_start_valid) {
752 tlb_start = start;
753 tlb_start_valid = 1;
754 }
755
756 if (is_vm_hugetlb_page(vma)) {
757 block = end - start;
758 unmap_hugepage_range(vma, start, end);
759 } else {
760 block = min(zap_bytes, end - start);
761 unmap_page_range(*tlbp, vma, start,
762 start + block, details);
763 }
764
765 start += block;
766 zap_bytes -= block;
767 if ((long)zap_bytes > 0)
768 continue;
769
770 tlb_finish_mmu(*tlbp, tlb_start, start);
771
772 if (need_resched() ||
773 need_lockbreak(&mm->page_table_lock) ||
774 (i_mmap_lock && need_lockbreak(i_mmap_lock))) {
775 if (i_mmap_lock) {
776 /* must reset count of rss freed */
777 *tlbp = tlb_gather_mmu(mm, fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 goto out;
779 }
780 spin_unlock(&mm->page_table_lock);
781 cond_resched();
782 spin_lock(&mm->page_table_lock);
783 }
784
785 *tlbp = tlb_gather_mmu(mm, fullmm);
786 tlb_start_valid = 0;
787 zap_bytes = ZAP_BLOCK_SIZE;
788 }
789 }
790out:
Hugh Dickinsee39b372005-04-19 13:29:15 -0700791 return start; /* which is now the end (or restart) address */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792}
793
794/**
795 * zap_page_range - remove user pages in a given range
796 * @vma: vm_area_struct holding the applicable pages
797 * @address: starting address of pages to zap
798 * @size: number of bytes to zap
799 * @details: details of nonlinear truncation or shared cache invalidation
800 */
Hugh Dickinsee39b372005-04-19 13:29:15 -0700801unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 unsigned long size, struct zap_details *details)
803{
804 struct mm_struct *mm = vma->vm_mm;
805 struct mmu_gather *tlb;
806 unsigned long end = address + size;
807 unsigned long nr_accounted = 0;
808
809 if (is_vm_hugetlb_page(vma)) {
810 zap_hugepage_range(vma, address, size);
Hugh Dickinsee39b372005-04-19 13:29:15 -0700811 return end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 }
813
814 lru_add_drain();
815 spin_lock(&mm->page_table_lock);
816 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -0700817 update_hiwater_rss(mm);
Hugh Dickinsee39b372005-04-19 13:29:15 -0700818 end = unmap_vmas(&tlb, mm, vma, address, end, &nr_accounted, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 tlb_finish_mmu(tlb, address, end);
820 spin_unlock(&mm->page_table_lock);
Hugh Dickinsee39b372005-04-19 13:29:15 -0700821 return end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822}
823
824/*
825 * Do a quick page-table lookup for a single page.
826 * mm->page_table_lock must be held.
827 */
Andrew Morton1aaf18f2005-07-27 11:43:54 -0700828static struct page *__follow_page(struct mm_struct *mm, unsigned long address,
829 int read, int write, int accessed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830{
831 pgd_t *pgd;
832 pud_t *pud;
833 pmd_t *pmd;
834 pte_t *ptep, pte;
835 unsigned long pfn;
836 struct page *page;
837
838 page = follow_huge_addr(mm, address, write);
839 if (! IS_ERR(page))
840 return page;
841
842 pgd = pgd_offset(mm, address);
843 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
844 goto out;
845
846 pud = pud_offset(pgd, address);
847 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
848 goto out;
849
850 pmd = pmd_offset(pud, address);
851 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
852 goto out;
853 if (pmd_huge(*pmd))
854 return follow_huge_pmd(mm, address, pmd, write);
855
856 ptep = pte_offset_map(pmd, address);
857 if (!ptep)
858 goto out;
859
860 pte = *ptep;
861 pte_unmap(ptep);
862 if (pte_present(pte)) {
Nick Pigginf33ea7f2005-08-03 20:24:01 +1000863 if (write && !pte_write(pte))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 goto out;
865 if (read && !pte_read(pte))
866 goto out;
867 pfn = pte_pfn(pte);
868 if (pfn_valid(pfn)) {
869 page = pfn_to_page(pfn);
Nick Pigginf33ea7f2005-08-03 20:24:01 +1000870 if (accessed) {
871 if (write && !pte_dirty(pte) &&!PageDirty(page))
872 set_page_dirty(page);
Andrew Morton1aaf18f2005-07-27 11:43:54 -0700873 mark_page_accessed(page);
Nick Pigginf33ea7f2005-08-03 20:24:01 +1000874 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 return page;
876 }
877 }
878
879out:
880 return NULL;
881}
882
Andrew Morton1aaf18f2005-07-27 11:43:54 -0700883inline struct page *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884follow_page(struct mm_struct *mm, unsigned long address, int write)
885{
Andrew Morton1aaf18f2005-07-27 11:43:54 -0700886 return __follow_page(mm, address, 0, write, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887}
888
Andrew Morton1aaf18f2005-07-27 11:43:54 -0700889/*
890 * check_user_page_readable() can be called frm niterrupt context by oprofile,
891 * so we need to avoid taking any non-irq-safe locks
892 */
893int check_user_page_readable(struct mm_struct *mm, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894{
Andrew Morton1aaf18f2005-07-27 11:43:54 -0700895 return __follow_page(mm, address, 1, 0, 0) != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897EXPORT_SYMBOL(check_user_page_readable);
898
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899static inline int
900untouched_anonymous_page(struct mm_struct* mm, struct vm_area_struct *vma,
901 unsigned long address)
902{
903 pgd_t *pgd;
904 pud_t *pud;
905 pmd_t *pmd;
906
907 /* Check if the vma is for an anonymous mapping. */
908 if (vma->vm_ops && vma->vm_ops->nopage)
909 return 0;
910
911 /* Check if page directory entry exists. */
912 pgd = pgd_offset(mm, address);
913 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
914 return 1;
915
916 pud = pud_offset(pgd, address);
917 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
918 return 1;
919
920 /* Check if page middle directory entry exists. */
921 pmd = pmd_offset(pud, address);
922 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
923 return 1;
924
925 /* There is a pte slot for 'address' in 'mm'. */
926 return 0;
927}
928
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
930 unsigned long start, int len, int write, int force,
931 struct page **pages, struct vm_area_struct **vmas)
932{
933 int i;
934 unsigned int flags;
935
936 /*
937 * Require read or write permissions.
938 * If 'force' is set, we only require the "MAY" flags.
939 */
940 flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
941 flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
942 i = 0;
943
944 do {
945 struct vm_area_struct * vma;
946
947 vma = find_extend_vma(mm, start);
948 if (!vma && in_gate_area(tsk, start)) {
949 unsigned long pg = start & PAGE_MASK;
950 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
951 pgd_t *pgd;
952 pud_t *pud;
953 pmd_t *pmd;
954 pte_t *pte;
955 if (write) /* user gate pages are read-only */
956 return i ? : -EFAULT;
957 if (pg > TASK_SIZE)
958 pgd = pgd_offset_k(pg);
959 else
960 pgd = pgd_offset_gate(mm, pg);
961 BUG_ON(pgd_none(*pgd));
962 pud = pud_offset(pgd, pg);
963 BUG_ON(pud_none(*pud));
964 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -0700965 if (pmd_none(*pmd))
966 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -0700968 if (pte_none(*pte)) {
969 pte_unmap(pte);
970 return i ? : -EFAULT;
971 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 if (pages) {
973 pages[i] = pte_page(*pte);
974 get_page(pages[i]);
975 }
976 pte_unmap(pte);
977 if (vmas)
978 vmas[i] = gate_vma;
979 i++;
980 start += PAGE_SIZE;
981 len--;
982 continue;
983 }
984
Nick Pigginb5810032005-10-29 18:16:12 -0700985 if (!vma || (vma->vm_flags & (VM_IO | VM_RESERVED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 || !(flags & vma->vm_flags))
987 return i ? : -EFAULT;
988
989 if (is_vm_hugetlb_page(vma)) {
990 i = follow_hugetlb_page(mm, vma, pages, vmas,
991 &start, &len, i);
992 continue;
993 }
994 spin_lock(&mm->page_table_lock);
995 do {
Nick Pigginf33ea7f2005-08-03 20:24:01 +1000996 int write_access = write;
Hugh Dickins08ef4722005-06-21 17:15:10 -0700997 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998
999 cond_resched_lock(&mm->page_table_lock);
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001000 while (!(page = follow_page(mm, start, write_access))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001001 int ret;
1002
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 /*
1004 * Shortcut for anonymous pages. We don't want
1005 * to force the creation of pages tables for
Hugh Dickins08ef4722005-06-21 17:15:10 -07001006 * insanely big anonymously mapped areas that
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 * nobody touched so far. This is important
1008 * for doing a core dump for these mappings.
1009 */
Linus Torvalds4ceb5db2005-08-01 11:14:49 -07001010 if (!write && untouched_anonymous_page(mm,vma,start)) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001011 page = ZERO_PAGE(start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 break;
1013 }
1014 spin_unlock(&mm->page_table_lock);
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001015 ret = __handle_mm_fault(mm, vma, start, write_access);
1016
1017 /*
1018 * The VM_FAULT_WRITE bit tells us that do_wp_page has
1019 * broken COW when necessary, even if maybe_mkwrite
1020 * decided not to set pte_write. We can thus safely do
1021 * subsequent page lookups as if they were reads.
1022 */
1023 if (ret & VM_FAULT_WRITE)
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001024 write_access = 0;
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001025
1026 switch (ret & ~VM_FAULT_WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 case VM_FAULT_MINOR:
1028 tsk->min_flt++;
1029 break;
1030 case VM_FAULT_MAJOR:
1031 tsk->maj_flt++;
1032 break;
1033 case VM_FAULT_SIGBUS:
1034 return i ? i : -EFAULT;
1035 case VM_FAULT_OOM:
1036 return i ? i : -ENOMEM;
1037 default:
1038 BUG();
1039 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 spin_lock(&mm->page_table_lock);
1041 }
1042 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001043 pages[i] = page;
1044 flush_dcache_page(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001045 page_cache_get(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 }
1047 if (vmas)
1048 vmas[i] = vma;
1049 i++;
1050 start += PAGE_SIZE;
1051 len--;
Hugh Dickins08ef4722005-06-21 17:15:10 -07001052 } while (len && start < vma->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 spin_unlock(&mm->page_table_lock);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001054 } while (len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 return i;
1056}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057EXPORT_SYMBOL(get_user_pages);
1058
1059static int zeromap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1060 unsigned long addr, unsigned long end, pgprot_t prot)
1061{
1062 pte_t *pte;
1063
1064 pte = pte_alloc_map(mm, pmd, addr);
1065 if (!pte)
1066 return -ENOMEM;
1067 do {
Nick Pigginb5810032005-10-29 18:16:12 -07001068 struct page *page = ZERO_PAGE(addr);
1069 pte_t zero_pte = pte_wrprotect(mk_pte(page, prot));
1070 page_cache_get(page);
1071 page_add_file_rmap(page);
1072 inc_mm_counter(mm, file_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 BUG_ON(!pte_none(*pte));
1074 set_pte_at(mm, addr, pte, zero_pte);
1075 } while (pte++, addr += PAGE_SIZE, addr != end);
1076 pte_unmap(pte - 1);
1077 return 0;
1078}
1079
1080static inline int zeromap_pmd_range(struct mm_struct *mm, pud_t *pud,
1081 unsigned long addr, unsigned long end, pgprot_t prot)
1082{
1083 pmd_t *pmd;
1084 unsigned long next;
1085
1086 pmd = pmd_alloc(mm, pud, addr);
1087 if (!pmd)
1088 return -ENOMEM;
1089 do {
1090 next = pmd_addr_end(addr, end);
1091 if (zeromap_pte_range(mm, pmd, addr, next, prot))
1092 return -ENOMEM;
1093 } while (pmd++, addr = next, addr != end);
1094 return 0;
1095}
1096
1097static inline int zeromap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1098 unsigned long addr, unsigned long end, pgprot_t prot)
1099{
1100 pud_t *pud;
1101 unsigned long next;
1102
1103 pud = pud_alloc(mm, pgd, addr);
1104 if (!pud)
1105 return -ENOMEM;
1106 do {
1107 next = pud_addr_end(addr, end);
1108 if (zeromap_pmd_range(mm, pud, addr, next, prot))
1109 return -ENOMEM;
1110 } while (pud++, addr = next, addr != end);
1111 return 0;
1112}
1113
1114int zeromap_page_range(struct vm_area_struct *vma,
1115 unsigned long addr, unsigned long size, pgprot_t prot)
1116{
1117 pgd_t *pgd;
1118 unsigned long next;
1119 unsigned long end = addr + size;
1120 struct mm_struct *mm = vma->vm_mm;
1121 int err;
1122
1123 BUG_ON(addr >= end);
1124 pgd = pgd_offset(mm, addr);
1125 flush_cache_range(vma, addr, end);
1126 spin_lock(&mm->page_table_lock);
1127 do {
1128 next = pgd_addr_end(addr, end);
1129 err = zeromap_pud_range(mm, pgd, addr, next, prot);
1130 if (err)
1131 break;
1132 } while (pgd++, addr = next, addr != end);
1133 spin_unlock(&mm->page_table_lock);
1134 return err;
1135}
1136
1137/*
1138 * maps a range of physical memory into the requested pages. the old
1139 * mappings are removed. any references to nonexistent pages results
1140 * in null mappings (currently treated as "copy-on-access")
1141 */
1142static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1143 unsigned long addr, unsigned long end,
1144 unsigned long pfn, pgprot_t prot)
1145{
1146 pte_t *pte;
1147
1148 pte = pte_alloc_map(mm, pmd, addr);
1149 if (!pte)
1150 return -ENOMEM;
1151 do {
1152 BUG_ON(!pte_none(*pte));
Nick Pigginb5810032005-10-29 18:16:12 -07001153 set_pte_at(mm, addr, pte, pfn_pte(pfn, prot));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 pfn++;
1155 } while (pte++, addr += PAGE_SIZE, addr != end);
1156 pte_unmap(pte - 1);
1157 return 0;
1158}
1159
1160static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1161 unsigned long addr, unsigned long end,
1162 unsigned long pfn, pgprot_t prot)
1163{
1164 pmd_t *pmd;
1165 unsigned long next;
1166
1167 pfn -= addr >> PAGE_SHIFT;
1168 pmd = pmd_alloc(mm, pud, addr);
1169 if (!pmd)
1170 return -ENOMEM;
1171 do {
1172 next = pmd_addr_end(addr, end);
1173 if (remap_pte_range(mm, pmd, addr, next,
1174 pfn + (addr >> PAGE_SHIFT), prot))
1175 return -ENOMEM;
1176 } while (pmd++, addr = next, addr != end);
1177 return 0;
1178}
1179
1180static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1181 unsigned long addr, unsigned long end,
1182 unsigned long pfn, pgprot_t prot)
1183{
1184 pud_t *pud;
1185 unsigned long next;
1186
1187 pfn -= addr >> PAGE_SHIFT;
1188 pud = pud_alloc(mm, pgd, addr);
1189 if (!pud)
1190 return -ENOMEM;
1191 do {
1192 next = pud_addr_end(addr, end);
1193 if (remap_pmd_range(mm, pud, addr, next,
1194 pfn + (addr >> PAGE_SHIFT), prot))
1195 return -ENOMEM;
1196 } while (pud++, addr = next, addr != end);
1197 return 0;
1198}
1199
1200/* Note: this is only safe if the mm semaphore is held when called. */
1201int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1202 unsigned long pfn, unsigned long size, pgprot_t prot)
1203{
1204 pgd_t *pgd;
1205 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001206 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 struct mm_struct *mm = vma->vm_mm;
1208 int err;
1209
1210 /*
1211 * Physically remapped pages are special. Tell the
1212 * rest of the world about it:
1213 * VM_IO tells people not to look at these pages
1214 * (accesses can have side effects).
Nick Pigginb5810032005-10-29 18:16:12 -07001215 * VM_RESERVED tells the core MM not to "manage" these pages
1216 * (e.g. refcount, mapcount, try to swap them out).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 */
1218 vma->vm_flags |= VM_IO | VM_RESERVED;
1219
1220 BUG_ON(addr >= end);
1221 pfn -= addr >> PAGE_SHIFT;
1222 pgd = pgd_offset(mm, addr);
1223 flush_cache_range(vma, addr, end);
1224 spin_lock(&mm->page_table_lock);
1225 do {
1226 next = pgd_addr_end(addr, end);
1227 err = remap_pud_range(mm, pgd, addr, next,
1228 pfn + (addr >> PAGE_SHIFT), prot);
1229 if (err)
1230 break;
1231 } while (pgd++, addr = next, addr != end);
1232 spin_unlock(&mm->page_table_lock);
1233 return err;
1234}
1235EXPORT_SYMBOL(remap_pfn_range);
1236
1237/*
1238 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1239 * servicing faults for write access. In the normal case, do always want
1240 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1241 * that do not have writing enabled, when used by access_process_vm.
1242 */
1243static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1244{
1245 if (likely(vma->vm_flags & VM_WRITE))
1246 pte = pte_mkwrite(pte);
1247 return pte;
1248}
1249
1250/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 * This routine handles present pages, when users try to write
1252 * to a shared page. It is done by copying the page to a new address
1253 * and decrementing the shared-page counter for the old page.
1254 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 * Note that this routine assumes that the protection checks have been
1256 * done by the caller (the low-level page fault routine in most cases).
1257 * Thus we can safely just mark it writable once we've done any necessary
1258 * COW.
1259 *
1260 * We also mark the page dirty at this point even though the page will
1261 * change only once the write actually happens. This avoids a few races,
1262 * and potentially makes it more efficient.
1263 *
1264 * We hold the mm semaphore and the page_table_lock on entry and exit
1265 * with the page_table_lock released.
1266 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001267static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1268 unsigned long address, pte_t *page_table, pmd_t *pmd,
1269 pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270{
1271 struct page *old_page, *new_page;
Hugh Dickins65500d22005-10-29 18:15:59 -07001272 unsigned long pfn = pte_pfn(orig_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 pte_t entry;
Hugh Dickins65500d22005-10-29 18:15:59 -07001274 int ret = VM_FAULT_MINOR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
Nick Pigginb5810032005-10-29 18:16:12 -07001276 BUG_ON(vma->vm_flags & VM_RESERVED);
1277
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 if (unlikely(!pfn_valid(pfn))) {
1279 /*
Hugh Dickins65500d22005-10-29 18:15:59 -07001280 * Page table corrupted: show pte and kill process.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 */
Nick Pigginb5810032005-10-29 18:16:12 -07001282 print_bad_pte(vma, orig_pte, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07001283 ret = VM_FAULT_OOM;
1284 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 }
1286 old_page = pfn_to_page(pfn);
1287
Hugh Dickinsd296e9c2005-06-21 17:15:11 -07001288 if (PageAnon(old_page) && !TestSetPageLocked(old_page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 int reuse = can_share_swap_page(old_page);
1290 unlock_page(old_page);
1291 if (reuse) {
1292 flush_cache_page(vma, address, pfn);
Hugh Dickins65500d22005-10-29 18:15:59 -07001293 entry = pte_mkyoung(orig_pte);
1294 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 ptep_set_access_flags(vma, address, page_table, entry, 1);
1296 update_mmu_cache(vma, address, entry);
1297 lazy_mmu_prot_update(entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07001298 ret |= VM_FAULT_WRITE;
1299 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 }
1301 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302
1303 /*
1304 * Ok, we need to copy. Oh, well..
1305 */
Nick Pigginb5810032005-10-29 18:16:12 -07001306 page_cache_get(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001307 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 spin_unlock(&mm->page_table_lock);
1309
1310 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07001311 goto oom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 if (old_page == ZERO_PAGE(address)) {
1313 new_page = alloc_zeroed_user_highpage(vma, address);
1314 if (!new_page)
Hugh Dickins65500d22005-10-29 18:15:59 -07001315 goto oom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 } else {
1317 new_page = alloc_page_vma(GFP_HIGHUSER, vma, address);
1318 if (!new_page)
Hugh Dickins65500d22005-10-29 18:15:59 -07001319 goto oom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 copy_user_highpage(new_page, old_page, address);
1321 }
Hugh Dickins65500d22005-10-29 18:15:59 -07001322
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 /*
1324 * Re-check the pte - we dropped the lock
1325 */
1326 spin_lock(&mm->page_table_lock);
1327 page_table = pte_offset_map(pmd, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07001328 if (likely(pte_same(*page_table, orig_pte))) {
Nick Pigginb5810032005-10-29 18:16:12 -07001329 page_remove_rmap(old_page);
1330 if (!PageAnon(old_page)) {
Hugh Dickins42946212005-10-29 18:16:05 -07001331 inc_mm_counter(mm, anon_rss);
Nick Pigginb5810032005-10-29 18:16:12 -07001332 dec_mm_counter(mm, file_rss);
Hugh Dickins42946212005-10-29 18:16:05 -07001333 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 flush_cache_page(vma, address, pfn);
Hugh Dickins65500d22005-10-29 18:15:59 -07001335 entry = mk_pte(new_page, vma->vm_page_prot);
1336 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1337 ptep_establish(vma, address, page_table, entry);
1338 update_mmu_cache(vma, address, entry);
1339 lazy_mmu_prot_update(entry);
1340
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 lru_cache_add_active(new_page);
1342 page_add_anon_rmap(new_page, vma, address);
1343
1344 /* Free the old page.. */
1345 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001346 ret |= VM_FAULT_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 page_cache_release(new_page);
1349 page_cache_release(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001350unlock:
1351 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 spin_unlock(&mm->page_table_lock);
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001353 return ret;
Hugh Dickins65500d22005-10-29 18:15:59 -07001354oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 page_cache_release(old_page);
1356 return VM_FAULT_OOM;
1357}
1358
1359/*
1360 * Helper functions for unmap_mapping_range().
1361 *
1362 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1363 *
1364 * We have to restart searching the prio_tree whenever we drop the lock,
1365 * since the iterator is only valid while the lock is held, and anyway
1366 * a later vma might be split and reinserted earlier while lock dropped.
1367 *
1368 * The list of nonlinear vmas could be handled more efficiently, using
1369 * a placeholder, but handle it in the same way until a need is shown.
1370 * It is important to search the prio_tree before nonlinear list: a vma
1371 * may become nonlinear and be shifted from prio_tree to nonlinear list
1372 * while the lock is dropped; but never shifted from list to prio_tree.
1373 *
1374 * In order to make forward progress despite restarting the search,
1375 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1376 * quickly skip it next time around. Since the prio_tree search only
1377 * shows us those vmas affected by unmapping the range in question, we
1378 * can't efficiently keep all vmas in step with mapping->truncate_count:
1379 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1380 * mapping->truncate_count and vma->vm_truncate_count are protected by
1381 * i_mmap_lock.
1382 *
1383 * In order to make forward progress despite repeatedly restarting some
Hugh Dickinsee39b372005-04-19 13:29:15 -07001384 * large vma, note the restart_addr from unmap_vmas when it breaks out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 * and restart from that address when we reach that vma again. It might
1386 * have been split or merged, shrunk or extended, but never shifted: so
1387 * restart_addr remains valid so long as it remains in the vma's range.
1388 * unmap_mapping_range forces truncate_count to leap over page-aligned
1389 * values so we can save vma's restart_addr in its truncate_count field.
1390 */
1391#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1392
1393static void reset_vma_truncate_counts(struct address_space *mapping)
1394{
1395 struct vm_area_struct *vma;
1396 struct prio_tree_iter iter;
1397
1398 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1399 vma->vm_truncate_count = 0;
1400 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1401 vma->vm_truncate_count = 0;
1402}
1403
1404static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1405 unsigned long start_addr, unsigned long end_addr,
1406 struct zap_details *details)
1407{
1408 unsigned long restart_addr;
1409 int need_break;
1410
1411again:
1412 restart_addr = vma->vm_truncate_count;
1413 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1414 start_addr = restart_addr;
1415 if (start_addr >= end_addr) {
1416 /* Top of vma has been split off since last time */
1417 vma->vm_truncate_count = details->truncate_count;
1418 return 0;
1419 }
1420 }
1421
Hugh Dickinsee39b372005-04-19 13:29:15 -07001422 restart_addr = zap_page_range(vma, start_addr,
1423 end_addr - start_addr, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424
1425 /*
1426 * We cannot rely on the break test in unmap_vmas:
1427 * on the one hand, we don't want to restart our loop
1428 * just because that broke out for the page_table_lock;
1429 * on the other hand, it does no test when vma is small.
1430 */
1431 need_break = need_resched() ||
1432 need_lockbreak(details->i_mmap_lock);
1433
Hugh Dickinsee39b372005-04-19 13:29:15 -07001434 if (restart_addr >= end_addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 /* We have now completed this vma: mark it so */
1436 vma->vm_truncate_count = details->truncate_count;
1437 if (!need_break)
1438 return 0;
1439 } else {
1440 /* Note restart_addr in vma's truncate_count field */
Hugh Dickinsee39b372005-04-19 13:29:15 -07001441 vma->vm_truncate_count = restart_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 if (!need_break)
1443 goto again;
1444 }
1445
1446 spin_unlock(details->i_mmap_lock);
1447 cond_resched();
1448 spin_lock(details->i_mmap_lock);
1449 return -EINTR;
1450}
1451
1452static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1453 struct zap_details *details)
1454{
1455 struct vm_area_struct *vma;
1456 struct prio_tree_iter iter;
1457 pgoff_t vba, vea, zba, zea;
1458
1459restart:
1460 vma_prio_tree_foreach(vma, &iter, root,
1461 details->first_index, details->last_index) {
1462 /* Skip quickly over those we have already dealt with */
1463 if (vma->vm_truncate_count == details->truncate_count)
1464 continue;
1465
1466 vba = vma->vm_pgoff;
1467 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1468 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1469 zba = details->first_index;
1470 if (zba < vba)
1471 zba = vba;
1472 zea = details->last_index;
1473 if (zea > vea)
1474 zea = vea;
1475
1476 if (unmap_mapping_range_vma(vma,
1477 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1478 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1479 details) < 0)
1480 goto restart;
1481 }
1482}
1483
1484static inline void unmap_mapping_range_list(struct list_head *head,
1485 struct zap_details *details)
1486{
1487 struct vm_area_struct *vma;
1488
1489 /*
1490 * In nonlinear VMAs there is no correspondence between virtual address
1491 * offset and file offset. So we must perform an exhaustive search
1492 * across *all* the pages in each nonlinear VMA, not just the pages
1493 * whose virtual address lies outside the file truncation point.
1494 */
1495restart:
1496 list_for_each_entry(vma, head, shared.vm_set.list) {
1497 /* Skip quickly over those we have already dealt with */
1498 if (vma->vm_truncate_count == details->truncate_count)
1499 continue;
1500 details->nonlinear_vma = vma;
1501 if (unmap_mapping_range_vma(vma, vma->vm_start,
1502 vma->vm_end, details) < 0)
1503 goto restart;
1504 }
1505}
1506
1507/**
1508 * unmap_mapping_range - unmap the portion of all mmaps
1509 * in the specified address_space corresponding to the specified
1510 * page range in the underlying file.
Martin Waitz3d410882005-06-23 22:05:21 -07001511 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 * @holebegin: byte in first page to unmap, relative to the start of
1513 * the underlying file. This will be rounded down to a PAGE_SIZE
1514 * boundary. Note that this is different from vmtruncate(), which
1515 * must keep the partial page. In contrast, we must get rid of
1516 * partial pages.
1517 * @holelen: size of prospective hole in bytes. This will be rounded
1518 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1519 * end of the file.
1520 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1521 * but 0 when invalidating pagecache, don't throw away private data.
1522 */
1523void unmap_mapping_range(struct address_space *mapping,
1524 loff_t const holebegin, loff_t const holelen, int even_cows)
1525{
1526 struct zap_details details;
1527 pgoff_t hba = holebegin >> PAGE_SHIFT;
1528 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1529
1530 /* Check for overflow. */
1531 if (sizeof(holelen) > sizeof(hlen)) {
1532 long long holeend =
1533 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1534 if (holeend & ~(long long)ULONG_MAX)
1535 hlen = ULONG_MAX - hba + 1;
1536 }
1537
1538 details.check_mapping = even_cows? NULL: mapping;
1539 details.nonlinear_vma = NULL;
1540 details.first_index = hba;
1541 details.last_index = hba + hlen - 1;
1542 if (details.last_index < details.first_index)
1543 details.last_index = ULONG_MAX;
1544 details.i_mmap_lock = &mapping->i_mmap_lock;
1545
1546 spin_lock(&mapping->i_mmap_lock);
1547
1548 /* serialize i_size write against truncate_count write */
1549 smp_wmb();
1550 /* Protect against page faults, and endless unmapping loops */
1551 mapping->truncate_count++;
1552 /*
1553 * For archs where spin_lock has inclusive semantics like ia64
1554 * this smp_mb() will prevent to read pagetable contents
1555 * before the truncate_count increment is visible to
1556 * other cpus.
1557 */
1558 smp_mb();
1559 if (unlikely(is_restart_addr(mapping->truncate_count))) {
1560 if (mapping->truncate_count == 0)
1561 reset_vma_truncate_counts(mapping);
1562 mapping->truncate_count++;
1563 }
1564 details.truncate_count = mapping->truncate_count;
1565
1566 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
1567 unmap_mapping_range_tree(&mapping->i_mmap, &details);
1568 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
1569 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
1570 spin_unlock(&mapping->i_mmap_lock);
1571}
1572EXPORT_SYMBOL(unmap_mapping_range);
1573
1574/*
1575 * Handle all mappings that got truncated by a "truncate()"
1576 * system call.
1577 *
1578 * NOTE! We have to be ready to update the memory sharing
1579 * between the file and the memory map for a potential last
1580 * incomplete page. Ugly, but necessary.
1581 */
1582int vmtruncate(struct inode * inode, loff_t offset)
1583{
1584 struct address_space *mapping = inode->i_mapping;
1585 unsigned long limit;
1586
1587 if (inode->i_size < offset)
1588 goto do_expand;
1589 /*
1590 * truncation of in-use swapfiles is disallowed - it would cause
1591 * subsequent swapout to scribble on the now-freed blocks.
1592 */
1593 if (IS_SWAPFILE(inode))
1594 goto out_busy;
1595 i_size_write(inode, offset);
1596 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
1597 truncate_inode_pages(mapping, offset);
1598 goto out_truncate;
1599
1600do_expand:
1601 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1602 if (limit != RLIM_INFINITY && offset > limit)
1603 goto out_sig;
1604 if (offset > inode->i_sb->s_maxbytes)
1605 goto out_big;
1606 i_size_write(inode, offset);
1607
1608out_truncate:
1609 if (inode->i_op && inode->i_op->truncate)
1610 inode->i_op->truncate(inode);
1611 return 0;
1612out_sig:
1613 send_sig(SIGXFSZ, current, 0);
1614out_big:
1615 return -EFBIG;
1616out_busy:
1617 return -ETXTBSY;
1618}
1619
1620EXPORT_SYMBOL(vmtruncate);
1621
1622/*
1623 * Primitive swap readahead code. We simply read an aligned block of
1624 * (1 << page_cluster) entries in the swap area. This method is chosen
1625 * because it doesn't cost us any seek time. We also make sure to queue
1626 * the 'original' request together with the readahead ones...
1627 *
1628 * This has been extended to use the NUMA policies from the mm triggering
1629 * the readahead.
1630 *
1631 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
1632 */
1633void swapin_readahead(swp_entry_t entry, unsigned long addr,struct vm_area_struct *vma)
1634{
1635#ifdef CONFIG_NUMA
1636 struct vm_area_struct *next_vma = vma ? vma->vm_next : NULL;
1637#endif
1638 int i, num;
1639 struct page *new_page;
1640 unsigned long offset;
1641
1642 /*
1643 * Get the number of handles we should do readahead io to.
1644 */
1645 num = valid_swaphandles(entry, &offset);
1646 for (i = 0; i < num; offset++, i++) {
1647 /* Ok, do the async read-ahead now */
1648 new_page = read_swap_cache_async(swp_entry(swp_type(entry),
1649 offset), vma, addr);
1650 if (!new_page)
1651 break;
1652 page_cache_release(new_page);
1653#ifdef CONFIG_NUMA
1654 /*
1655 * Find the next applicable VMA for the NUMA policy.
1656 */
1657 addr += PAGE_SIZE;
1658 if (addr == 0)
1659 vma = NULL;
1660 if (vma) {
1661 if (addr >= vma->vm_end) {
1662 vma = next_vma;
1663 next_vma = vma ? vma->vm_next : NULL;
1664 }
1665 if (vma && addr < vma->vm_start)
1666 vma = NULL;
1667 } else {
1668 if (next_vma && addr >= next_vma->vm_start) {
1669 vma = next_vma;
1670 next_vma = vma->vm_next;
1671 }
1672 }
1673#endif
1674 }
1675 lru_add_drain(); /* Push any new pages onto the LRU now */
1676}
1677
1678/*
1679 * We hold the mm semaphore and the page_table_lock on entry and
1680 * should release the pagetable lock on exit..
1681 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001682static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
1683 unsigned long address, pte_t *page_table, pmd_t *pmd,
1684 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685{
1686 struct page *page;
Hugh Dickins65500d22005-10-29 18:15:59 -07001687 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 pte_t pte;
1689 int ret = VM_FAULT_MINOR;
1690
1691 pte_unmap(page_table);
1692 spin_unlock(&mm->page_table_lock);
Hugh Dickins65500d22005-10-29 18:15:59 -07001693
1694 entry = pte_to_swp_entry(orig_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 page = lookup_swap_cache(entry);
1696 if (!page) {
1697 swapin_readahead(entry, address, vma);
1698 page = read_swap_cache_async(entry, vma, address);
1699 if (!page) {
1700 /*
1701 * Back out if somebody else faulted in this pte while
1702 * we released the page table lock.
1703 */
1704 spin_lock(&mm->page_table_lock);
1705 page_table = pte_offset_map(pmd, address);
1706 if (likely(pte_same(*page_table, orig_pte)))
1707 ret = VM_FAULT_OOM;
Hugh Dickins65500d22005-10-29 18:15:59 -07001708 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 }
1710
1711 /* Had to read the page from swap area: Major fault */
1712 ret = VM_FAULT_MAJOR;
1713 inc_page_state(pgmajfault);
1714 grab_swap_token();
1715 }
1716
1717 mark_page_accessed(page);
1718 lock_page(page);
1719
1720 /*
1721 * Back out if somebody else faulted in this pte while we
1722 * released the page table lock.
1723 */
1724 spin_lock(&mm->page_table_lock);
1725 page_table = pte_offset_map(pmd, address);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07001726 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07001727 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07001728
1729 if (unlikely(!PageUptodate(page))) {
1730 ret = VM_FAULT_SIGBUS;
1731 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 }
1733
1734 /* The page isn't present yet, go ahead with the fault. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735
Hugh Dickins42946212005-10-29 18:16:05 -07001736 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 pte = mk_pte(page, vma->vm_page_prot);
1738 if (write_access && can_share_swap_page(page)) {
1739 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
1740 write_access = 0;
1741 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742
1743 flush_icache_page(vma, page);
1744 set_pte_at(mm, address, page_table, pte);
1745 page_add_anon_rmap(page, vma, address);
1746
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07001747 swap_free(entry);
1748 if (vm_swap_full())
1749 remove_exclusive_swap_page(page);
1750 unlock_page(page);
1751
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 if (write_access) {
1753 if (do_wp_page(mm, vma, address,
1754 page_table, pmd, pte) == VM_FAULT_OOM)
1755 ret = VM_FAULT_OOM;
1756 goto out;
1757 }
1758
1759 /* No need to invalidate - it was non-present before */
1760 update_mmu_cache(vma, address, pte);
1761 lazy_mmu_prot_update(pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07001762unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 pte_unmap(page_table);
1764 spin_unlock(&mm->page_table_lock);
1765out:
1766 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07001767out_nomap:
1768 pte_unmap(page_table);
1769 spin_unlock(&mm->page_table_lock);
1770 unlock_page(page);
1771 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001772 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773}
1774
1775/*
1776 * We are called with the MM semaphore and page_table_lock
1777 * spinlock held to protect against concurrent faults in
1778 * multithreaded programs.
1779 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001780static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
1781 unsigned long address, pte_t *page_table, pmd_t *pmd,
1782 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783{
Nick Pigginb5810032005-10-29 18:16:12 -07001784 struct page *page = ZERO_PAGE(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786
Hugh Dickins72866f62005-10-29 18:15:55 -07001787 /* Mapping of ZERO_PAGE - vm_page_prot is readonly */
Nick Pigginb5810032005-10-29 18:16:12 -07001788 entry = mk_pte(page, vma->vm_page_prot);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 if (write_access) {
1791 /* Allocate our own private page. */
1792 pte_unmap(page_table);
1793 spin_unlock(&mm->page_table_lock);
1794
1795 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07001796 goto oom;
1797 page = alloc_zeroed_user_highpage(vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 if (!page)
Hugh Dickins65500d22005-10-29 18:15:59 -07001799 goto oom;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800
1801 spin_lock(&mm->page_table_lock);
Hugh Dickins65500d22005-10-29 18:15:59 -07001802 page_table = pte_offset_map(pmd, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803
1804 if (!pte_none(*page_table)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001806 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 }
Hugh Dickins42946212005-10-29 18:16:05 -07001808 inc_mm_counter(mm, anon_rss);
Hugh Dickins65500d22005-10-29 18:15:59 -07001809 entry = mk_pte(page, vma->vm_page_prot);
1810 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 lru_cache_add_active(page);
1812 SetPageReferenced(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001813 page_add_anon_rmap(page, vma, address);
Nick Pigginb5810032005-10-29 18:16:12 -07001814 } else {
1815 inc_mm_counter(mm, file_rss);
1816 page_add_file_rmap(page);
1817 page_cache_get(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 }
1819
Hugh Dickins65500d22005-10-29 18:15:59 -07001820 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821
1822 /* No need to invalidate - it was non-present before */
Hugh Dickins65500d22005-10-29 18:15:59 -07001823 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 lazy_mmu_prot_update(entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07001825unlock:
1826 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 spin_unlock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 return VM_FAULT_MINOR;
Hugh Dickins65500d22005-10-29 18:15:59 -07001829oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830 return VM_FAULT_OOM;
1831}
1832
1833/*
1834 * do_no_page() tries to create a new page mapping. It aggressively
1835 * tries to share with existing pages, but makes a separate copy if
1836 * the "write_access" parameter is true in order to avoid the next
1837 * page fault.
1838 *
1839 * As this is called only for pages that do not currently exist, we
1840 * do not need to flush old virtual caches or the TLB.
1841 *
1842 * This is called with the MM semaphore held and the page table
1843 * spinlock held. Exit with the spinlock released.
1844 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001845static int do_no_page(struct mm_struct *mm, struct vm_area_struct *vma,
1846 unsigned long address, pte_t *page_table, pmd_t *pmd,
1847 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848{
Hugh Dickins65500d22005-10-29 18:15:59 -07001849 struct page *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 struct address_space *mapping = NULL;
1851 pte_t entry;
1852 unsigned int sequence = 0;
1853 int ret = VM_FAULT_MINOR;
1854 int anon = 0;
1855
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 pte_unmap(page_table);
1857 spin_unlock(&mm->page_table_lock);
1858
1859 if (vma->vm_file) {
1860 mapping = vma->vm_file->f_mapping;
1861 sequence = mapping->truncate_count;
1862 smp_rmb(); /* serializes i_size against truncate_count */
1863 }
1864retry:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 new_page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret);
1866 /*
1867 * No smp_rmb is needed here as long as there's a full
1868 * spin_lock/unlock sequence inside the ->nopage callback
1869 * (for the pagecache lookup) that acts as an implicit
1870 * smp_mb() and prevents the i_size read to happen
1871 * after the next truncate_count read.
1872 */
1873
1874 /* no page was available -- either SIGBUS or OOM */
1875 if (new_page == NOPAGE_SIGBUS)
1876 return VM_FAULT_SIGBUS;
1877 if (new_page == NOPAGE_OOM)
1878 return VM_FAULT_OOM;
1879
1880 /*
1881 * Should we do an early C-O-W break?
1882 */
1883 if (write_access && !(vma->vm_flags & VM_SHARED)) {
1884 struct page *page;
1885
1886 if (unlikely(anon_vma_prepare(vma)))
1887 goto oom;
1888 page = alloc_page_vma(GFP_HIGHUSER, vma, address);
1889 if (!page)
1890 goto oom;
1891 copy_user_highpage(page, new_page, address);
1892 page_cache_release(new_page);
1893 new_page = page;
1894 anon = 1;
1895 }
1896
1897 spin_lock(&mm->page_table_lock);
1898 /*
1899 * For a file-backed vma, someone could have truncated or otherwise
1900 * invalidated this page. If unmap_mapping_range got called,
1901 * retry getting the page.
1902 */
1903 if (mapping && unlikely(sequence != mapping->truncate_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 spin_unlock(&mm->page_table_lock);
1905 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001906 cond_resched();
1907 sequence = mapping->truncate_count;
1908 smp_rmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 goto retry;
1910 }
1911 page_table = pte_offset_map(pmd, address);
1912
1913 /*
1914 * This silly early PAGE_DIRTY setting removes a race
1915 * due to the bad i386 page protection. But it's valid
1916 * for other architectures too.
1917 *
1918 * Note that if write_access is true, we either now have
1919 * an exclusive copy of the page, or this is a shared mapping,
1920 * so we can make it writable and dirty to avoid having to
1921 * handle that later.
1922 */
1923 /* Only go through if we didn't race with anybody else... */
1924 if (pte_none(*page_table)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 flush_icache_page(vma, new_page);
1926 entry = mk_pte(new_page, vma->vm_page_prot);
1927 if (write_access)
1928 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1929 set_pte_at(mm, address, page_table, entry);
1930 if (anon) {
Hugh Dickins42946212005-10-29 18:16:05 -07001931 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 lru_cache_add_active(new_page);
1933 page_add_anon_rmap(new_page, vma, address);
Nick Pigginb5810032005-10-29 18:16:12 -07001934 } else if (!(vma->vm_flags & VM_RESERVED)) {
Hugh Dickins42946212005-10-29 18:16:05 -07001935 inc_mm_counter(mm, file_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 page_add_file_rmap(new_page);
Hugh Dickins42946212005-10-29 18:16:05 -07001937 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 } else {
1939 /* One of our sibling threads was faster, back out. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001941 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 }
1943
1944 /* no need to invalidate: a not-present page shouldn't be cached */
1945 update_mmu_cache(vma, address, entry);
1946 lazy_mmu_prot_update(entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07001947unlock:
1948 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 spin_unlock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 return ret;
1951oom:
1952 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001953 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954}
1955
1956/*
1957 * Fault of a previously existing named mapping. Repopulate the pte
1958 * from the encoded file_pte if possible. This enables swappable
1959 * nonlinear vmas.
1960 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001961static int do_file_page(struct mm_struct *mm, struct vm_area_struct *vma,
1962 unsigned long address, pte_t *page_table, pmd_t *pmd,
1963 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964{
Hugh Dickins65500d22005-10-29 18:15:59 -07001965 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 int err;
1967
Hugh Dickins65500d22005-10-29 18:15:59 -07001968 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 spin_unlock(&mm->page_table_lock);
1970
Hugh Dickins65500d22005-10-29 18:15:59 -07001971 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
1972 /*
1973 * Page table corrupted: show pte and kill process.
1974 */
Nick Pigginb5810032005-10-29 18:16:12 -07001975 print_bad_pte(vma, orig_pte, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07001976 return VM_FAULT_OOM;
1977 }
1978 /* We can then assume vm->vm_ops && vma->vm_ops->populate */
1979
1980 pgoff = pte_to_pgoff(orig_pte);
1981 err = vma->vm_ops->populate(vma, address & PAGE_MASK, PAGE_SIZE,
1982 vma->vm_page_prot, pgoff, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 if (err == -ENOMEM)
1984 return VM_FAULT_OOM;
1985 if (err)
1986 return VM_FAULT_SIGBUS;
1987 return VM_FAULT_MAJOR;
1988}
1989
1990/*
1991 * These routines also need to handle stuff like marking pages dirty
1992 * and/or accessed for architectures that don't do it in hardware (most
1993 * RISC architectures). The early dirtying is also good on the i386.
1994 *
1995 * There is also a hook called "update_mmu_cache()" that architectures
1996 * with external mmu caches can use to update those (ie the Sparc or
1997 * PowerPC hashed page tables that act as extended TLBs).
1998 *
1999 * Note the "page_table_lock". It is to protect against kswapd removing
2000 * pages from under us. Note that kswapd only ever _removes_ pages, never
2001 * adds them. As such, once we have noticed that the page is not present,
2002 * we can drop the lock early.
2003 *
2004 * The adding of pages is protected by the MM semaphore (which we hold),
2005 * so we don't need to worry about a page being suddenly been added into
2006 * our VM.
2007 *
2008 * We enter with the pagetable spinlock held, we are supposed to
2009 * release it when done.
2010 */
2011static inline int handle_pte_fault(struct mm_struct *mm,
Hugh Dickins65500d22005-10-29 18:15:59 -07002012 struct vm_area_struct *vma, unsigned long address,
2013 pte_t *pte, pmd_t *pmd, int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014{
2015 pte_t entry;
2016
2017 entry = *pte;
2018 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002019 if (pte_none(entry)) {
2020 if (!vma->vm_ops || !vma->vm_ops->nopage)
2021 return do_anonymous_page(mm, vma, address,
2022 pte, pmd, write_access);
2023 return do_no_page(mm, vma, address,
2024 pte, pmd, write_access);
2025 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 if (pte_file(entry))
Hugh Dickins65500d22005-10-29 18:15:59 -07002027 return do_file_page(mm, vma, address,
2028 pte, pmd, write_access, entry);
2029 return do_swap_page(mm, vma, address,
2030 pte, pmd, write_access, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 }
2032
2033 if (write_access) {
2034 if (!pte_write(entry))
2035 return do_wp_page(mm, vma, address, pte, pmd, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 entry = pte_mkdirty(entry);
2037 }
2038 entry = pte_mkyoung(entry);
2039 ptep_set_access_flags(vma, address, pte, entry, write_access);
2040 update_mmu_cache(vma, address, entry);
2041 lazy_mmu_prot_update(entry);
2042 pte_unmap(pte);
2043 spin_unlock(&mm->page_table_lock);
2044 return VM_FAULT_MINOR;
2045}
2046
2047/*
2048 * By the time we get here, we already hold the mm semaphore
2049 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002050int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 unsigned long address, int write_access)
2052{
2053 pgd_t *pgd;
2054 pud_t *pud;
2055 pmd_t *pmd;
2056 pte_t *pte;
2057
2058 __set_current_state(TASK_RUNNING);
2059
2060 inc_page_state(pgfault);
2061
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01002062 if (unlikely(is_vm_hugetlb_page(vma)))
2063 return hugetlb_fault(mm, vma, address, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064
2065 /*
2066 * We need the page table lock to synchronize with kswapd
2067 * and the SMP-safe atomic PTE updates.
2068 */
2069 pgd = pgd_offset(mm, address);
2070 spin_lock(&mm->page_table_lock);
2071
2072 pud = pud_alloc(mm, pgd, address);
2073 if (!pud)
2074 goto oom;
2075
2076 pmd = pmd_alloc(mm, pud, address);
2077 if (!pmd)
2078 goto oom;
2079
2080 pte = pte_alloc_map(mm, pmd, address);
2081 if (!pte)
2082 goto oom;
2083
Hugh Dickins65500d22005-10-29 18:15:59 -07002084 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085
2086 oom:
2087 spin_unlock(&mm->page_table_lock);
2088 return VM_FAULT_OOM;
2089}
2090
2091#ifndef __PAGETABLE_PUD_FOLDED
2092/*
2093 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002094 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 */
2096pud_t fastcall *__pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
2097{
2098 pud_t *new;
2099
Hugh Dickins872fec12005-10-29 18:16:21 -07002100 if (mm != &init_mm) /* Temporary bridging hack */
2101 spin_unlock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 new = pud_alloc_one(mm, address);
Hugh Dickins872fec12005-10-29 18:16:21 -07002103 if (!new) {
2104 if (mm != &init_mm) /* Temporary bridging hack */
2105 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 return NULL;
Hugh Dickins872fec12005-10-29 18:16:21 -07002107 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108
Hugh Dickins872fec12005-10-29 18:16:21 -07002109 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 if (pgd_present(*pgd)) {
2111 pud_free(new);
2112 goto out;
2113 }
2114 pgd_populate(mm, pgd, new);
2115 out:
Hugh Dickins872fec12005-10-29 18:16:21 -07002116 if (mm == &init_mm) /* Temporary bridging hack */
2117 spin_unlock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 return pud_offset(pgd, address);
2119}
2120#endif /* __PAGETABLE_PUD_FOLDED */
2121
2122#ifndef __PAGETABLE_PMD_FOLDED
2123/*
2124 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002125 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 */
2127pmd_t fastcall *__pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
2128{
2129 pmd_t *new;
2130
Hugh Dickins872fec12005-10-29 18:16:21 -07002131 if (mm != &init_mm) /* Temporary bridging hack */
2132 spin_unlock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 new = pmd_alloc_one(mm, address);
Hugh Dickins872fec12005-10-29 18:16:21 -07002134 if (!new) {
2135 if (mm != &init_mm) /* Temporary bridging hack */
2136 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 return NULL;
Hugh Dickins872fec12005-10-29 18:16:21 -07002138 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139
Hugh Dickins872fec12005-10-29 18:16:21 -07002140 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141#ifndef __ARCH_HAS_4LEVEL_HACK
2142 if (pud_present(*pud)) {
2143 pmd_free(new);
2144 goto out;
2145 }
2146 pud_populate(mm, pud, new);
2147#else
2148 if (pgd_present(*pud)) {
2149 pmd_free(new);
2150 goto out;
2151 }
2152 pgd_populate(mm, pud, new);
2153#endif /* __ARCH_HAS_4LEVEL_HACK */
2154
2155 out:
Hugh Dickins872fec12005-10-29 18:16:21 -07002156 if (mm == &init_mm) /* Temporary bridging hack */
2157 spin_unlock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 return pmd_offset(pud, address);
2159}
2160#endif /* __PAGETABLE_PMD_FOLDED */
2161
2162int make_pages_present(unsigned long addr, unsigned long end)
2163{
2164 int ret, len, write;
2165 struct vm_area_struct * vma;
2166
2167 vma = find_vma(current->mm, addr);
2168 if (!vma)
2169 return -1;
2170 write = (vma->vm_flags & VM_WRITE) != 0;
2171 if (addr >= end)
2172 BUG();
2173 if (end > vma->vm_end)
2174 BUG();
2175 len = (end+PAGE_SIZE-1)/PAGE_SIZE-addr/PAGE_SIZE;
2176 ret = get_user_pages(current, current->mm, addr,
2177 len, write, 0, NULL, NULL);
2178 if (ret < 0)
2179 return ret;
2180 return ret == len ? 0 : -1;
2181}
2182
2183/*
2184 * Map a vmalloc()-space virtual address to the physical page.
2185 */
2186struct page * vmalloc_to_page(void * vmalloc_addr)
2187{
2188 unsigned long addr = (unsigned long) vmalloc_addr;
2189 struct page *page = NULL;
2190 pgd_t *pgd = pgd_offset_k(addr);
2191 pud_t *pud;
2192 pmd_t *pmd;
2193 pte_t *ptep, pte;
2194
2195 if (!pgd_none(*pgd)) {
2196 pud = pud_offset(pgd, addr);
2197 if (!pud_none(*pud)) {
2198 pmd = pmd_offset(pud, addr);
2199 if (!pmd_none(*pmd)) {
2200 ptep = pte_offset_map(pmd, addr);
2201 pte = *ptep;
2202 if (pte_present(pte))
2203 page = pte_page(pte);
2204 pte_unmap(ptep);
2205 }
2206 }
2207 }
2208 return page;
2209}
2210
2211EXPORT_SYMBOL(vmalloc_to_page);
2212
2213/*
2214 * Map a vmalloc()-space virtual address to the physical page frame number.
2215 */
2216unsigned long vmalloc_to_pfn(void * vmalloc_addr)
2217{
2218 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
2219}
2220
2221EXPORT_SYMBOL(vmalloc_to_pfn);
2222
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223#if !defined(__HAVE_ARCH_GATE_AREA)
2224
2225#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07002226static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227
2228static int __init gate_vma_init(void)
2229{
2230 gate_vma.vm_mm = NULL;
2231 gate_vma.vm_start = FIXADDR_USER_START;
2232 gate_vma.vm_end = FIXADDR_USER_END;
2233 gate_vma.vm_page_prot = PAGE_READONLY;
Nick Pigginb5810032005-10-29 18:16:12 -07002234 gate_vma.vm_flags = VM_RESERVED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 return 0;
2236}
2237__initcall(gate_vma_init);
2238#endif
2239
2240struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2241{
2242#ifdef AT_SYSINFO_EHDR
2243 return &gate_vma;
2244#else
2245 return NULL;
2246#endif
2247}
2248
2249int in_gate_area_no_task(unsigned long addr)
2250{
2251#ifdef AT_SYSINFO_EHDR
2252 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2253 return 1;
2254#endif
2255 return 0;
2256}
2257
2258#endif /* __HAVE_ARCH_GATE_AREA */