blob: 48c122d42ed743dcc90178170b4aff380ec3ad8c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/memory.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 */
6
7/*
8 * demand-loading started 01.12.91 - seems it is high on the list of
9 * things wanted, and it should be easy to implement. - Linus
10 */
11
12/*
13 * Ok, demand-loading was easy, shared pages a little bit tricker. Shared
14 * pages started 02.12.91, seems to work. - Linus.
15 *
16 * Tested sharing by executing about 30 /bin/sh: under the old kernel it
17 * would have taken more than the 6M I have free, but it worked well as
18 * far as I could see.
19 *
20 * Also corrected some "invalidate()"s - I wasn't doing enough of them.
21 */
22
23/*
24 * Real VM (paging to/from disk) started 18.12.91. Much more work and
25 * thought has to go into this. Oh, well..
26 * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why.
27 * Found it. Everything seems to work now.
28 * 20.12.91 - Ok, making the swap-device changeable like the root.
29 */
30
31/*
32 * 05.04.94 - Multi-page memory management added for v1.1.
33 * Idea by Alex Bligh (alex@cconcepts.co.uk)
34 *
35 * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG
36 * (Gerhard.Wichert@pdb.siemens.de)
37 *
38 * Aug/Sep 2004 Changed to four level page tables (Andi Kleen)
39 */
40
41#include <linux/kernel_stat.h>
42#include <linux/mm.h>
43#include <linux/hugetlb.h>
44#include <linux/mman.h>
45#include <linux/swap.h>
46#include <linux/highmem.h>
47#include <linux/pagemap.h>
48#include <linux/rmap.h>
49#include <linux/module.h>
Shailabh Nagar0ff92242006-07-14 00:24:37 -070050#include <linux/delayacct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include <linux/init.h>
Peter Zijlstraedc79b22006-09-25 23:30:58 -070052#include <linux/writeback.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080053#include <linux/memcontrol.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
55#include <asm/pgalloc.h>
56#include <asm/uaccess.h>
57#include <asm/tlb.h>
58#include <asm/tlbflush.h>
59#include <asm/pgtable.h>
60
61#include <linux/swapops.h>
62#include <linux/elf.h>
63
Andy Whitcroftd41dee32005-06-23 00:07:54 -070064#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -070065/* use the per-pgdat data instead for discontigmem - mbligh */
66unsigned long max_mapnr;
67struct page *mem_map;
68
69EXPORT_SYMBOL(max_mapnr);
70EXPORT_SYMBOL(mem_map);
71#endif
72
73unsigned long num_physpages;
74/*
75 * A number of key systems in x86 including ioremap() rely on the assumption
76 * that high_memory defines the upper bound on direct map memory, then end
77 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
78 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
79 * and ZONE_HIGHMEM.
80 */
81void * high_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
83EXPORT_SYMBOL(num_physpages);
84EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
Ingo Molnar32a93232008-02-06 22:39:44 +010086/*
87 * Randomize the address space (stacks, mmaps, brk, etc.).
88 *
89 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
90 * as ancient (libc5 based) binaries can segfault. )
91 */
92int randomize_va_space __read_mostly =
93#ifdef CONFIG_COMPAT_BRK
94 1;
95#else
96 2;
97#endif
Andi Kleena62eaf12006-02-16 23:41:58 +010098
99static int __init disable_randmaps(char *s)
100{
101 randomize_va_space = 0;
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800102 return 1;
Andi Kleena62eaf12006-02-16 23:41:58 +0100103}
104__setup("norandmaps", disable_randmaps);
105
106
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107/*
108 * If a p?d_bad entry is found while walking page tables, report
109 * the error, before resetting entry to p?d_none. Usually (but
110 * very seldom) called out from the p?d_none_or_clear_bad macros.
111 */
112
113void pgd_clear_bad(pgd_t *pgd)
114{
115 pgd_ERROR(*pgd);
116 pgd_clear(pgd);
117}
118
119void pud_clear_bad(pud_t *pud)
120{
121 pud_ERROR(*pud);
122 pud_clear(pud);
123}
124
125void pmd_clear_bad(pmd_t *pmd)
126{
127 pmd_ERROR(*pmd);
128 pmd_clear(pmd);
129}
130
131/*
132 * Note: this doesn't free the actual pages themselves. That
133 * has been handled earlier when unmapping all the memory regions.
134 */
Hugh Dickinse0da3822005-04-19 13:29:15 -0700135static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800137 pgtable_t token = pmd_pgtable(*pmd);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700138 pmd_clear(pmd);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800139 pte_free_tlb(tlb, token);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700140 tlb->mm->nr_ptes--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141}
142
Hugh Dickinse0da3822005-04-19 13:29:15 -0700143static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
144 unsigned long addr, unsigned long end,
145 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146{
147 pmd_t *pmd;
148 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700149 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150
Hugh Dickinse0da3822005-04-19 13:29:15 -0700151 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 do {
154 next = pmd_addr_end(addr, end);
155 if (pmd_none_or_clear_bad(pmd))
156 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700157 free_pte_range(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 } while (pmd++, addr = next, addr != end);
159
Hugh Dickinse0da3822005-04-19 13:29:15 -0700160 start &= PUD_MASK;
161 if (start < floor)
162 return;
163 if (ceiling) {
164 ceiling &= PUD_MASK;
165 if (!ceiling)
166 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700168 if (end - 1 > ceiling - 1)
169 return;
170
171 pmd = pmd_offset(pud, start);
172 pud_clear(pud);
173 pmd_free_tlb(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174}
175
Hugh Dickinse0da3822005-04-19 13:29:15 -0700176static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
177 unsigned long addr, unsigned long end,
178 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179{
180 pud_t *pud;
181 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700182 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183
Hugh Dickinse0da3822005-04-19 13:29:15 -0700184 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 pud = pud_offset(pgd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 do {
187 next = pud_addr_end(addr, end);
188 if (pud_none_or_clear_bad(pud))
189 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700190 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 } while (pud++, addr = next, addr != end);
192
Hugh Dickinse0da3822005-04-19 13:29:15 -0700193 start &= PGDIR_MASK;
194 if (start < floor)
195 return;
196 if (ceiling) {
197 ceiling &= PGDIR_MASK;
198 if (!ceiling)
199 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700201 if (end - 1 > ceiling - 1)
202 return;
203
204 pud = pud_offset(pgd, start);
205 pgd_clear(pgd);
206 pud_free_tlb(tlb, pud);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207}
208
209/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700210 * This function frees user-level page tables of a process.
211 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 * Must be called with pagetable lock held.
213 */
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700214void free_pgd_range(struct mmu_gather **tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700215 unsigned long addr, unsigned long end,
216 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217{
218 pgd_t *pgd;
219 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700220 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221
Hugh Dickinse0da3822005-04-19 13:29:15 -0700222 /*
223 * The next few lines have given us lots of grief...
224 *
225 * Why are we testing PMD* at this top level? Because often
226 * there will be no work to do at all, and we'd prefer not to
227 * go all the way down to the bottom just to discover that.
228 *
229 * Why all these "- 1"s? Because 0 represents both the bottom
230 * of the address space and the top of it (using -1 for the
231 * top wouldn't help much: the masks would do the wrong thing).
232 * The rule is that addr 0 and floor 0 refer to the bottom of
233 * the address space, but end 0 and ceiling 0 refer to the top
234 * Comparisons need to use "end - 1" and "ceiling - 1" (though
235 * that end 0 case should be mythical).
236 *
237 * Wherever addr is brought up or ceiling brought down, we must
238 * be careful to reject "the opposite 0" before it confuses the
239 * subsequent tests. But what about where end is brought down
240 * by PMD_SIZE below? no, end can't go down to 0 there.
241 *
242 * Whereas we round start (addr) and ceiling down, by different
243 * masks at different levels, in order to test whether a table
244 * now has no other vmas using it, so can be freed, we don't
245 * bother to round floor or end up - the tests don't need that.
246 */
247
248 addr &= PMD_MASK;
249 if (addr < floor) {
250 addr += PMD_SIZE;
251 if (!addr)
252 return;
253 }
254 if (ceiling) {
255 ceiling &= PMD_MASK;
256 if (!ceiling)
257 return;
258 }
259 if (end - 1 > ceiling - 1)
260 end -= PMD_SIZE;
261 if (addr > end - 1)
262 return;
263
264 start = addr;
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700265 pgd = pgd_offset((*tlb)->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 do {
267 next = pgd_addr_end(addr, end);
268 if (pgd_none_or_clear_bad(pgd))
269 continue;
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700270 free_pud_range(*tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700272}
273
274void free_pgtables(struct mmu_gather **tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700275 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700276{
277 while (vma) {
278 struct vm_area_struct *next = vma->vm_next;
279 unsigned long addr = vma->vm_start;
280
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700281 /*
282 * Hide vma from rmap and vmtruncate before freeing pgtables
283 */
284 anon_vma_unlink(vma);
285 unlink_file_vma(vma);
286
David Gibson9da61ae2006-03-22 00:08:57 -0800287 if (is_vm_hugetlb_page(vma)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700288 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700289 floor, next? next->vm_start: ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700290 } else {
291 /*
292 * Optimization: gather nearby vmas into one call down
293 */
294 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
David Gibson48669202006-03-22 00:08:58 -0800295 && !is_vm_hugetlb_page(next)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700296 vma = next;
297 next = vma->vm_next;
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700298 anon_vma_unlink(vma);
299 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700300 }
301 free_pgd_range(tlb, addr, vma->vm_end,
302 floor, next? next->vm_start: ceiling);
303 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700304 vma = next;
305 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306}
307
Hugh Dickins1bb36302005-10-29 18:16:22 -0700308int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800310 pgtable_t new = pte_alloc_one(mm, address);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700311 if (!new)
312 return -ENOMEM;
313
Hugh Dickinsc74df322005-10-29 18:16:23 -0700314 spin_lock(&mm->page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800315 if (!pmd_present(*pmd)) { /* Has another populated it ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 mm->nr_ptes++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 pmd_populate(mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800318 new = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 }
Hugh Dickinsc74df322005-10-29 18:16:23 -0700320 spin_unlock(&mm->page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800321 if (new)
322 pte_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700323 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324}
325
Hugh Dickins1bb36302005-10-29 18:16:22 -0700326int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700328 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
329 if (!new)
330 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331
Hugh Dickins1bb36302005-10-29 18:16:22 -0700332 spin_lock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800333 if (!pmd_present(*pmd)) { /* Has another populated it ? */
Hugh Dickins1bb36302005-10-29 18:16:22 -0700334 pmd_populate_kernel(&init_mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800335 new = NULL;
336 }
Hugh Dickins1bb36302005-10-29 18:16:22 -0700337 spin_unlock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800338 if (new)
339 pte_free_kernel(&init_mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700340 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341}
342
Hugh Dickinsae859762005-10-29 18:16:05 -0700343static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
344{
345 if (file_rss)
346 add_mm_counter(mm, file_rss, file_rss);
347 if (anon_rss)
348 add_mm_counter(mm, anon_rss, anon_rss);
349}
350
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800352 * This function is called to print an error when a bad pte
353 * is found. For example, we might have a PFN-mapped pte in
354 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700355 *
356 * The calling function must still handle the error.
357 */
358void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr)
359{
360 printk(KERN_ERR "Bad pte = %08llx, process = %s, "
361 "vm_flags = %lx, vaddr = %lx\n",
362 (long long)pte_val(pte),
363 (vma->vm_mm == current->mm ? current->comm : "???"),
364 vma->vm_flags, vaddr);
365 dump_stack();
366}
367
Linus Torvalds67121172005-12-11 20:38:17 -0800368static inline int is_cow_mapping(unsigned int flags)
369{
370 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
371}
372
Nick Pigginb5810032005-10-29 18:16:12 -0700373/*
Nick Piggin7e675132008-04-28 02:13:00 -0700374 * vm_normal_page -- This function gets the "struct page" associated with a pte.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800375 *
Nick Piggin7e675132008-04-28 02:13:00 -0700376 * "Special" mappings do not wish to be associated with a "struct page" (either
377 * it doesn't exist, or it exists but they don't want to touch it). In this
378 * case, NULL is returned here. "Normal" mappings do have a struct page.
Jared Hulbertb379d792008-04-28 02:12:58 -0700379 *
Nick Piggin7e675132008-04-28 02:13:00 -0700380 * There are 2 broad cases. Firstly, an architecture may define a pte_special()
381 * pte bit, in which case this function is trivial. Secondly, an architecture
382 * may not have a spare pte bit, which requires a more complicated scheme,
383 * described below.
384 *
385 * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
386 * special mapping (even if there are underlying and valid "struct pages").
387 * COWed pages of a VM_PFNMAP are always normal.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800388 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700389 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
390 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
Nick Piggin7e675132008-04-28 02:13:00 -0700391 * set, and the vm_pgoff will point to the first PFN mapped: thus every special
392 * mapping will always honor the rule
Linus Torvalds6aab3412005-11-28 14:34:23 -0800393 *
394 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
395 *
Nick Piggin7e675132008-04-28 02:13:00 -0700396 * And for normal mappings this is false.
Jared Hulbertb379d792008-04-28 02:12:58 -0700397 *
Nick Piggin7e675132008-04-28 02:13:00 -0700398 * This restricts such mappings to be a linear translation from virtual address
399 * to pfn. To get around this restriction, we allow arbitrary mappings so long
400 * as the vma is not a COW mapping; in that case, we know that all ptes are
401 * special (because none can have been COWed).
Jared Hulbertb379d792008-04-28 02:12:58 -0700402 *
403 *
Nick Piggin7e675132008-04-28 02:13:00 -0700404 * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
405 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700406 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
407 * page" backing, however the difference is that _all_ pages with a struct
408 * page (that is, those where pfn_valid is true) are refcounted and considered
409 * normal pages by the VM. The disadvantage is that pages are refcounted
410 * (which can be slower and simply not an option for some PFNMAP users). The
411 * advantage is that we don't have to follow the strict linearity rule of
412 * PFNMAP mappings in order to support COWable mappings.
413 *
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800414 */
Nick Piggin7e675132008-04-28 02:13:00 -0700415#ifdef __HAVE_ARCH_PTE_SPECIAL
416# define HAVE_PTE_SPECIAL 1
417#else
418# define HAVE_PTE_SPECIAL 0
419#endif
420struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
421 pte_t pte)
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800422{
Nick Piggin7e675132008-04-28 02:13:00 -0700423 unsigned long pfn;
424
425 if (HAVE_PTE_SPECIAL) {
426 if (likely(!pte_special(pte))) {
427 VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
428 return pte_page(pte);
429 }
430 VM_BUG_ON(!(vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)));
431 return NULL;
432 }
433
434 /* !HAVE_PTE_SPECIAL case follows: */
435
436 pfn = pte_pfn(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800437
Jared Hulbertb379d792008-04-28 02:12:58 -0700438 if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
439 if (vma->vm_flags & VM_MIXEDMAP) {
440 if (!pfn_valid(pfn))
441 return NULL;
442 goto out;
443 } else {
Nick Piggin7e675132008-04-28 02:13:00 -0700444 unsigned long off;
445 off = (addr - vma->vm_start) >> PAGE_SHIFT;
Jared Hulbertb379d792008-04-28 02:12:58 -0700446 if (pfn == vma->vm_pgoff + off)
447 return NULL;
448 if (!is_cow_mapping(vma->vm_flags))
449 return NULL;
450 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800451 }
452
Nick Piggin7e675132008-04-28 02:13:00 -0700453 VM_BUG_ON(!pfn_valid(pfn));
Linus Torvalds6aab3412005-11-28 14:34:23 -0800454
455 /*
Nick Piggin7e675132008-04-28 02:13:00 -0700456 * NOTE! We still have PageReserved() pages in the page tables.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800457 *
Nick Piggin7e675132008-04-28 02:13:00 -0700458 * eg. VDSO mappings can cause them to exist.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800459 */
Jared Hulbertb379d792008-04-28 02:12:58 -0700460out:
Linus Torvalds6aab3412005-11-28 14:34:23 -0800461 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800462}
463
464/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 * copy one vm_area from one task to the other. Assumes the page tables
466 * already present in the new task to be cleared in the whole range
467 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 */
469
Hugh Dickins8c103762005-10-29 18:16:13 -0700470static inline void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700472 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700473 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474{
Nick Pigginb5810032005-10-29 18:16:12 -0700475 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 pte_t pte = *src_pte;
477 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478
479 /* pte contains position in swap or file, so copy. */
480 if (unlikely(!pte_present(pte))) {
481 if (!pte_file(pte)) {
Christoph Lameter06972122006-06-23 02:03:35 -0700482 swp_entry_t entry = pte_to_swp_entry(pte);
483
484 swap_duplicate(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 /* make sure dst_mm is on swapoff's mmlist. */
486 if (unlikely(list_empty(&dst_mm->mmlist))) {
487 spin_lock(&mmlist_lock);
Hugh Dickinsf412ac02005-10-29 18:16:41 -0700488 if (list_empty(&dst_mm->mmlist))
489 list_add(&dst_mm->mmlist,
490 &src_mm->mmlist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 spin_unlock(&mmlist_lock);
492 }
Christoph Lameter06972122006-06-23 02:03:35 -0700493 if (is_write_migration_entry(entry) &&
494 is_cow_mapping(vm_flags)) {
495 /*
496 * COW mappings require pages in both parent
497 * and child to be set to read.
498 */
499 make_migration_entry_read(&entry);
500 pte = swp_entry_to_pte(entry);
501 set_pte_at(src_mm, addr, src_pte, pte);
502 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700504 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 }
506
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 /*
508 * If it's a COW mapping, write protect it both
509 * in the parent and the child
510 */
Linus Torvalds67121172005-12-11 20:38:17 -0800511 if (is_cow_mapping(vm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 ptep_set_wrprotect(src_mm, addr, src_pte);
Zachary Amsden3dc90792006-09-30 23:29:30 -0700513 pte = pte_wrprotect(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 }
515
516 /*
517 * If it's a shared mapping, mark it clean in
518 * the child
519 */
520 if (vm_flags & VM_SHARED)
521 pte = pte_mkclean(pte);
522 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800523
524 page = vm_normal_page(vma, addr, pte);
525 if (page) {
526 get_page(page);
Nick Pigginc97a9e12007-05-16 22:11:21 -0700527 page_dup_rmap(page, vma, addr);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800528 rss[!!PageAnon(page)]++;
529 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700530
531out_set_pte:
532 set_pte_at(dst_mm, addr, dst_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533}
534
535static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
536 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
537 unsigned long addr, unsigned long end)
538{
539 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700540 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -0700541 int progress = 0;
Hugh Dickins8c103762005-10-29 18:16:13 -0700542 int rss[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543
544again:
Hugh Dickinsae859762005-10-29 18:16:05 -0700545 rss[1] = rss[0] = 0;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700546 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 if (!dst_pte)
548 return -ENOMEM;
549 src_pte = pte_offset_map_nested(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700550 src_ptl = pte_lockptr(src_mm, src_pmd);
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700551 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700552 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 do {
555 /*
556 * We are holding two locks at this point - either of them
557 * could generate latencies in another task on another CPU.
558 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700559 if (progress >= 32) {
560 progress = 0;
561 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +0100562 spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700563 break;
564 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 if (pte_none(*src_pte)) {
566 progress++;
567 continue;
568 }
Hugh Dickins8c103762005-10-29 18:16:13 -0700569 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 progress += 8;
571 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700573 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -0700574 spin_unlock(src_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 pte_unmap_nested(src_pte - 1);
Hugh Dickinsae859762005-10-29 18:16:05 -0700576 add_mm_rss(dst_mm, rss[0], rss[1]);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700577 pte_unmap_unlock(dst_pte - 1, dst_ptl);
578 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 if (addr != end)
580 goto again;
581 return 0;
582}
583
584static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
585 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
586 unsigned long addr, unsigned long end)
587{
588 pmd_t *src_pmd, *dst_pmd;
589 unsigned long next;
590
591 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
592 if (!dst_pmd)
593 return -ENOMEM;
594 src_pmd = pmd_offset(src_pud, addr);
595 do {
596 next = pmd_addr_end(addr, end);
597 if (pmd_none_or_clear_bad(src_pmd))
598 continue;
599 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
600 vma, addr, next))
601 return -ENOMEM;
602 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
603 return 0;
604}
605
606static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
607 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
608 unsigned long addr, unsigned long end)
609{
610 pud_t *src_pud, *dst_pud;
611 unsigned long next;
612
613 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
614 if (!dst_pud)
615 return -ENOMEM;
616 src_pud = pud_offset(src_pgd, addr);
617 do {
618 next = pud_addr_end(addr, end);
619 if (pud_none_or_clear_bad(src_pud))
620 continue;
621 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
622 vma, addr, next))
623 return -ENOMEM;
624 } while (dst_pud++, src_pud++, addr = next, addr != end);
625 return 0;
626}
627
628int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
629 struct vm_area_struct *vma)
630{
631 pgd_t *src_pgd, *dst_pgd;
632 unsigned long next;
633 unsigned long addr = vma->vm_start;
634 unsigned long end = vma->vm_end;
635
Nick Piggind9928952005-08-28 16:49:11 +1000636 /*
637 * Don't copy ptes where a page fault will fill them correctly.
638 * Fork becomes much lighter when there are big shared or private
639 * readonly mappings. The tradeoff is that copy_page_range is more
640 * efficient than faulting.
641 */
Linus Torvalds4d7672b2005-12-16 10:21:23 -0800642 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
Nick Piggind9928952005-08-28 16:49:11 +1000643 if (!vma->anon_vma)
644 return 0;
645 }
646
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 if (is_vm_hugetlb_page(vma))
648 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
649
650 dst_pgd = pgd_offset(dst_mm, addr);
651 src_pgd = pgd_offset(src_mm, addr);
652 do {
653 next = pgd_addr_end(addr, end);
654 if (pgd_none_or_clear_bad(src_pgd))
655 continue;
656 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
657 vma, addr, next))
658 return -ENOMEM;
659 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
660 return 0;
661}
662
Robin Holt51c6f662005-11-13 16:06:42 -0800663static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700664 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800666 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667{
Nick Pigginb5810032005-10-29 18:16:12 -0700668 struct mm_struct *mm = tlb->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 pte_t *pte;
Hugh Dickins508034a2005-10-29 18:16:30 -0700670 spinlock_t *ptl;
Hugh Dickinsae859762005-10-29 18:16:05 -0700671 int file_rss = 0;
672 int anon_rss = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673
Hugh Dickins508034a2005-10-29 18:16:30 -0700674 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700675 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 do {
677 pte_t ptent = *pte;
Robin Holt51c6f662005-11-13 16:06:42 -0800678 if (pte_none(ptent)) {
679 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800681 }
Hugh Dickins6f5e6b92006-03-16 23:04:09 -0800682
683 (*zap_work) -= PAGE_SIZE;
684
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800686 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -0800687
Linus Torvalds6aab3412005-11-28 14:34:23 -0800688 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 if (unlikely(details) && page) {
690 /*
691 * unmap_shared_mapping_pages() wants to
692 * invalidate cache without truncating:
693 * unmap shared but keep private pages.
694 */
695 if (details->check_mapping &&
696 details->check_mapping != page->mapping)
697 continue;
698 /*
699 * Each page->index must be checked when
700 * invalidating or truncating nonlinear.
701 */
702 if (details->nonlinear_vma &&
703 (page->index < details->first_index ||
704 page->index > details->last_index))
705 continue;
706 }
Nick Pigginb5810032005-10-29 18:16:12 -0700707 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -0700708 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 tlb_remove_tlb_entry(tlb, pte, addr);
710 if (unlikely(!page))
711 continue;
712 if (unlikely(details) && details->nonlinear_vma
713 && linear_page_index(details->nonlinear_vma,
714 addr) != page->index)
Nick Pigginb5810032005-10-29 18:16:12 -0700715 set_pte_at(mm, addr, pte,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 pgoff_to_pte(page->index));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 if (PageAnon(page))
Hugh Dickins86d912f2005-10-29 18:16:14 -0700718 anon_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700719 else {
720 if (pte_dirty(ptent))
721 set_page_dirty(page);
722 if (pte_young(ptent))
Ken Chendaa88c82007-02-10 01:43:18 -0800723 SetPageReferenced(page);
Hugh Dickins86d912f2005-10-29 18:16:14 -0700724 file_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700725 }
Nick Piggin7de6b802006-12-22 01:09:33 -0800726 page_remove_rmap(page, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 tlb_remove_page(tlb, page);
728 continue;
729 }
730 /*
731 * If details->check_mapping, we leave swap entries;
732 * if details->nonlinear_vma, we leave file entries.
733 */
734 if (unlikely(details))
735 continue;
736 if (!pte_file(ptent))
737 free_swap_and_cache(pte_to_swp_entry(ptent));
Zachary Amsden9888a1c2006-09-30 23:29:31 -0700738 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
Robin Holt51c6f662005-11-13 16:06:42 -0800739 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
Hugh Dickinsae859762005-10-29 18:16:05 -0700740
Hugh Dickins86d912f2005-10-29 18:16:14 -0700741 add_mm_rss(mm, file_rss, anon_rss);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700742 arch_leave_lazy_mmu_mode();
Hugh Dickins508034a2005-10-29 18:16:30 -0700743 pte_unmap_unlock(pte - 1, ptl);
Robin Holt51c6f662005-11-13 16:06:42 -0800744
745 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746}
747
Robin Holt51c6f662005-11-13 16:06:42 -0800748static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700749 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800751 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752{
753 pmd_t *pmd;
754 unsigned long next;
755
756 pmd = pmd_offset(pud, addr);
757 do {
758 next = pmd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800759 if (pmd_none_or_clear_bad(pmd)) {
760 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800762 }
763 next = zap_pte_range(tlb, vma, pmd, addr, next,
764 zap_work, details);
765 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
766
767 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768}
769
Robin Holt51c6f662005-11-13 16:06:42 -0800770static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700771 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800773 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774{
775 pud_t *pud;
776 unsigned long next;
777
778 pud = pud_offset(pgd, addr);
779 do {
780 next = pud_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800781 if (pud_none_or_clear_bad(pud)) {
782 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800784 }
785 next = zap_pmd_range(tlb, vma, pud, addr, next,
786 zap_work, details);
787 } while (pud++, addr = next, (addr != end && *zap_work > 0));
788
789 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790}
791
Robin Holt51c6f662005-11-13 16:06:42 -0800792static unsigned long unmap_page_range(struct mmu_gather *tlb,
793 struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800795 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796{
797 pgd_t *pgd;
798 unsigned long next;
799
800 if (details && !details->check_mapping && !details->nonlinear_vma)
801 details = NULL;
802
803 BUG_ON(addr >= end);
804 tlb_start_vma(tlb, vma);
805 pgd = pgd_offset(vma->vm_mm, addr);
806 do {
807 next = pgd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800808 if (pgd_none_or_clear_bad(pgd)) {
809 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800811 }
812 next = zap_pud_range(tlb, vma, pgd, addr, next,
813 zap_work, details);
814 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 tlb_end_vma(tlb, vma);
Robin Holt51c6f662005-11-13 16:06:42 -0800816
817 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818}
819
820#ifdef CONFIG_PREEMPT
821# define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
822#else
823/* No preempt: go for improved straight-line efficiency */
824# define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
825#endif
826
827/**
828 * unmap_vmas - unmap a range of memory covered by a list of vma's
829 * @tlbp: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 * @vma: the starting vma
831 * @start_addr: virtual address at which to start unmapping
832 * @end_addr: virtual address at which to end unmapping
833 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
834 * @details: details of nonlinear truncation or shared cache invalidation
835 *
Hugh Dickinsee39b372005-04-19 13:29:15 -0700836 * Returns the end address of the unmapping (restart addr if interrupted).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700838 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700840 * We aim to not hold locks for too long (for scheduling latency reasons).
841 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 * return the ending mmu_gather to the caller.
843 *
844 * Only addresses between `start' and `end' will be unmapped.
845 *
846 * The VMA list must be sorted in ascending virtual address order.
847 *
848 * unmap_vmas() assumes that the caller will flush the whole unmapped address
849 * range after unmap_vmas() returns. So the only responsibility here is to
850 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
851 * drops the lock and schedules.
852 */
Hugh Dickins508034a2005-10-29 18:16:30 -0700853unsigned long unmap_vmas(struct mmu_gather **tlbp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 struct vm_area_struct *vma, unsigned long start_addr,
855 unsigned long end_addr, unsigned long *nr_accounted,
856 struct zap_details *details)
857{
Robin Holt51c6f662005-11-13 16:06:42 -0800858 long zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
860 int tlb_start_valid = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -0700861 unsigned long start = start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700863 int fullmm = (*tlbp)->fullmm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864
865 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 unsigned long end;
867
868 start = max(vma->vm_start, start_addr);
869 if (start >= vma->vm_end)
870 continue;
871 end = min(vma->vm_end, end_addr);
872 if (end <= vma->vm_start)
873 continue;
874
875 if (vma->vm_flags & VM_ACCOUNT)
876 *nr_accounted += (end - start) >> PAGE_SHIFT;
877
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 while (start != end) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 if (!tlb_start_valid) {
880 tlb_start = start;
881 tlb_start_valid = 1;
882 }
883
Robin Holt51c6f662005-11-13 16:06:42 -0800884 if (unlikely(is_vm_hugetlb_page(vma))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 unmap_hugepage_range(vma, start, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800886 zap_work -= (end - start) /
887 (HPAGE_SIZE / PAGE_SIZE);
888 start = end;
889 } else
890 start = unmap_page_range(*tlbp, vma,
891 start, end, &zap_work, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892
Robin Holt51c6f662005-11-13 16:06:42 -0800893 if (zap_work > 0) {
894 BUG_ON(start != end);
895 break;
896 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897
898 tlb_finish_mmu(*tlbp, tlb_start, start);
899
900 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +0100901 (i_mmap_lock && spin_needbreak(i_mmap_lock))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 if (i_mmap_lock) {
Hugh Dickins508034a2005-10-29 18:16:30 -0700903 *tlbp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 goto out;
905 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 }
908
Hugh Dickins508034a2005-10-29 18:16:30 -0700909 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 tlb_start_valid = 0;
Robin Holt51c6f662005-11-13 16:06:42 -0800911 zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 }
913 }
914out:
Hugh Dickinsee39b372005-04-19 13:29:15 -0700915 return start; /* which is now the end (or restart) address */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916}
917
918/**
919 * zap_page_range - remove user pages in a given range
920 * @vma: vm_area_struct holding the applicable pages
921 * @address: starting address of pages to zap
922 * @size: number of bytes to zap
923 * @details: details of nonlinear truncation or shared cache invalidation
924 */
Hugh Dickinsee39b372005-04-19 13:29:15 -0700925unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 unsigned long size, struct zap_details *details)
927{
928 struct mm_struct *mm = vma->vm_mm;
929 struct mmu_gather *tlb;
930 unsigned long end = address + size;
931 unsigned long nr_accounted = 0;
932
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -0700935 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -0700936 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
937 if (tlb)
938 tlb_finish_mmu(tlb, address, end);
Hugh Dickinsee39b372005-04-19 13:29:15 -0700939 return end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940}
941
942/*
943 * Do a quick page-table lookup for a single page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800945struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700946 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947{
948 pgd_t *pgd;
949 pud_t *pud;
950 pmd_t *pmd;
951 pte_t *ptep, pte;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700952 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 struct page *page;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800954 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700956 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
957 if (!IS_ERR(page)) {
958 BUG_ON(flags & FOLL_GET);
959 goto out;
960 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700962 page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 pgd = pgd_offset(mm, address);
964 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700965 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966
967 pud = pud_offset(pgd, address);
968 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700969 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970
971 pmd = pmd_offset(pud, address);
Hugh Dickinsaeed5fc2008-05-06 20:49:23 +0100972 if (pmd_none(*pmd))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700973 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700975 if (pmd_huge(*pmd)) {
976 BUG_ON(flags & FOLL_GET);
977 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
978 goto out;
979 }
980
Hugh Dickinsaeed5fc2008-05-06 20:49:23 +0100981 if (unlikely(pmd_bad(*pmd)))
982 goto no_page_table;
983
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700984 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 if (!ptep)
986 goto out;
987
988 pte = *ptep;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700989 if (!pte_present(pte))
990 goto unlock;
991 if ((flags & FOLL_WRITE) && !pte_write(pte))
992 goto unlock;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800993 page = vm_normal_page(vma, address, pte);
994 if (unlikely(!page))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700995 goto unlock;
996
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700997 if (flags & FOLL_GET)
998 get_page(page);
999 if (flags & FOLL_TOUCH) {
1000 if ((flags & FOLL_WRITE) &&
1001 !pte_dirty(pte) && !PageDirty(page))
1002 set_page_dirty(page);
1003 mark_page_accessed(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001005unlock:
1006 pte_unmap_unlock(ptep, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007out:
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001008 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001010no_page_table:
1011 /*
1012 * When core dumping an enormous anonymous area that nobody
1013 * has touched so far, we don't want to allocate page tables.
1014 */
1015 if (flags & FOLL_ANON) {
Nick Piggin557ed1f2007-10-16 01:24:40 -07001016 page = ZERO_PAGE(0);
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001017 if (flags & FOLL_GET)
1018 get_page(page);
1019 BUG_ON(flags & FOLL_WRITE);
1020 }
1021 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022}
1023
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
1025 unsigned long start, int len, int write, int force,
1026 struct page **pages, struct vm_area_struct **vmas)
1027{
1028 int i;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001029 unsigned int vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
Jonathan Corbet900cf082008-02-11 16:17:33 -07001031 if (len <= 0)
1032 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 /*
1034 * Require read or write permissions.
1035 * If 'force' is set, we only require the "MAY" flags.
1036 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001037 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
1038 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 i = 0;
1040
1041 do {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001042 struct vm_area_struct *vma;
1043 unsigned int foll_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
1045 vma = find_extend_vma(mm, start);
1046 if (!vma && in_gate_area(tsk, start)) {
1047 unsigned long pg = start & PAGE_MASK;
1048 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
1049 pgd_t *pgd;
1050 pud_t *pud;
1051 pmd_t *pmd;
1052 pte_t *pte;
1053 if (write) /* user gate pages are read-only */
1054 return i ? : -EFAULT;
1055 if (pg > TASK_SIZE)
1056 pgd = pgd_offset_k(pg);
1057 else
1058 pgd = pgd_offset_gate(mm, pg);
1059 BUG_ON(pgd_none(*pgd));
1060 pud = pud_offset(pgd, pg);
1061 BUG_ON(pud_none(*pud));
1062 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001063 if (pmd_none(*pmd))
1064 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001066 if (pte_none(*pte)) {
1067 pte_unmap(pte);
1068 return i ? : -EFAULT;
1069 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 if (pages) {
Nick Pigginfa2a4552005-11-29 18:43:17 +11001071 struct page *page = vm_normal_page(gate_vma, start, *pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001072 pages[i] = page;
1073 if (page)
1074 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 }
1076 pte_unmap(pte);
1077 if (vmas)
1078 vmas[i] = gate_vma;
1079 i++;
1080 start += PAGE_SIZE;
1081 len--;
1082 continue;
1083 }
1084
Linus Torvalds1ff80382005-12-12 16:24:33 -08001085 if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001086 || !(vm_flags & vma->vm_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 return i ? : -EFAULT;
1088
1089 if (is_vm_hugetlb_page(vma)) {
1090 i = follow_hugetlb_page(mm, vma, pages, vmas,
Adam Litke5b23dbe2007-11-14 16:59:33 -08001091 &start, &len, i, write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 continue;
1093 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001094
1095 foll_flags = FOLL_TOUCH;
1096 if (pages)
1097 foll_flags |= FOLL_GET;
1098 if (!write && !(vma->vm_flags & VM_LOCKED) &&
Nick Piggin3c18ddd2008-04-28 02:12:10 -07001099 (!vma->vm_ops || !vma->vm_ops->fault))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001100 foll_flags |= FOLL_ANON;
1101
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 do {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001103 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104
Ethan Solomita462e00c2007-07-15 23:38:16 -07001105 /*
1106 * If tsk is ooming, cut off its access to large memory
1107 * allocations. It has a pending SIGKILL, but it can't
1108 * be processed until returning to user space.
1109 */
1110 if (unlikely(test_tsk_thread_flag(tsk, TIF_MEMDIE)))
1111 return -ENOMEM;
1112
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001113 if (write)
1114 foll_flags |= FOLL_WRITE;
1115
1116 cond_resched();
Linus Torvalds6aab3412005-11-28 14:34:23 -08001117 while (!(page = follow_page(vma, start, foll_flags))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001118 int ret;
Nick Piggin83c54072007-07-19 01:47:05 -07001119 ret = handle_mm_fault(mm, vma, start,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001120 foll_flags & FOLL_WRITE);
Nick Piggin83c54072007-07-19 01:47:05 -07001121 if (ret & VM_FAULT_ERROR) {
1122 if (ret & VM_FAULT_OOM)
1123 return i ? i : -ENOMEM;
1124 else if (ret & VM_FAULT_SIGBUS)
1125 return i ? i : -EFAULT;
1126 BUG();
1127 }
1128 if (ret & VM_FAULT_MAJOR)
1129 tsk->maj_flt++;
1130 else
1131 tsk->min_flt++;
1132
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001133 /*
Nick Piggin83c54072007-07-19 01:47:05 -07001134 * The VM_FAULT_WRITE bit tells us that
1135 * do_wp_page has broken COW when necessary,
1136 * even if maybe_mkwrite decided not to set
1137 * pte_write. We can thus safely do subsequent
1138 * page lookups as if they were reads.
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001139 */
1140 if (ret & VM_FAULT_WRITE)
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001141 foll_flags &= ~FOLL_WRITE;
Nick Piggin83c54072007-07-19 01:47:05 -07001142
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07001143 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 }
1145 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001146 pages[i] = page;
James Bottomley03beb072006-03-26 01:36:57 -08001147
Russell Kinga6f36be2006-12-30 22:24:19 +00001148 flush_anon_page(vma, page, start);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001149 flush_dcache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 }
1151 if (vmas)
1152 vmas[i] = vma;
1153 i++;
1154 start += PAGE_SIZE;
1155 len--;
Hugh Dickins08ef4722005-06-21 17:15:10 -07001156 } while (len && start < vma->vm_end);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001157 } while (len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 return i;
1159}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160EXPORT_SYMBOL(get_user_pages);
1161
Harvey Harrison920c7a52008-02-04 22:29:26 -08001162pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr,
1163 spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001164{
1165 pgd_t * pgd = pgd_offset(mm, addr);
1166 pud_t * pud = pud_alloc(mm, pgd, addr);
1167 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001168 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001169 if (pmd)
1170 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1171 }
1172 return NULL;
1173}
1174
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001176 * This is the old fallback for page remapping.
1177 *
1178 * For historical reasons, it only allows reserved pages. Only
1179 * old drivers should use this, and they needed to mark their
1180 * pages reserved for the old functions anyway.
1181 */
Nick Piggin423bad62008-04-28 02:13:01 -07001182static int insert_page(struct vm_area_struct *vma, unsigned long addr,
1183 struct page *page, pgprot_t prot)
Linus Torvalds238f58d2005-11-29 13:01:56 -08001184{
Nick Piggin423bad62008-04-28 02:13:01 -07001185 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001186 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001187 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001188 spinlock_t *ptl;
1189
Balbir Singhe1a1cd52008-02-07 00:14:02 -08001190 retval = mem_cgroup_charge(page, mm, GFP_KERNEL);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001191 if (retval)
1192 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001193
1194 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001195 if (PageAnon(page))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001196 goto out_uncharge;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001197 retval = -ENOMEM;
1198 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001199 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001200 if (!pte)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001201 goto out_uncharge;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001202 retval = -EBUSY;
1203 if (!pte_none(*pte))
1204 goto out_unlock;
1205
1206 /* Ok, finally just insert the thing.. */
1207 get_page(page);
1208 inc_mm_counter(mm, file_rss);
1209 page_add_file_rmap(page);
1210 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1211
1212 retval = 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001213 pte_unmap_unlock(pte, ptl);
1214 return retval;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001215out_unlock:
1216 pte_unmap_unlock(pte, ptl);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001217out_uncharge:
1218 mem_cgroup_uncharge_page(page);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001219out:
1220 return retval;
1221}
1222
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001223/**
1224 * vm_insert_page - insert single page into user vma
1225 * @vma: user vma to map to
1226 * @addr: target user address of this page
1227 * @page: source kernel page
1228 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001229 * This allows drivers to insert individual pages they've allocated
1230 * into a user vma.
1231 *
1232 * The page has to be a nice clean _individual_ kernel allocation.
1233 * If you allocate a compound page, you need to have marked it as
1234 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001235 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001236 *
1237 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1238 * took an arbitrary page protection parameter. This doesn't allow
1239 * that. Your vma protection will have to be set up correctly, which
1240 * means that if you want a shared writable mapping, you'd better
1241 * ask for a shared writable mapping!
1242 *
1243 * The page does not need to be reserved.
1244 */
Nick Piggin423bad62008-04-28 02:13:01 -07001245int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
1246 struct page *page)
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001247{
1248 if (addr < vma->vm_start || addr >= vma->vm_end)
1249 return -EFAULT;
1250 if (!page_count(page))
1251 return -EINVAL;
Linus Torvalds4d7672b2005-12-16 10:21:23 -08001252 vma->vm_flags |= VM_INSERTPAGE;
Nick Piggin423bad62008-04-28 02:13:01 -07001253 return insert_page(vma, addr, page, vma->vm_page_prot);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001254}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001255EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001256
Nick Piggin423bad62008-04-28 02:13:01 -07001257static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1258 unsigned long pfn, pgprot_t prot)
1259{
1260 struct mm_struct *mm = vma->vm_mm;
1261 int retval;
1262 pte_t *pte, entry;
1263 spinlock_t *ptl;
1264
1265 retval = -ENOMEM;
1266 pte = get_locked_pte(mm, addr, &ptl);
1267 if (!pte)
1268 goto out;
1269 retval = -EBUSY;
1270 if (!pte_none(*pte))
1271 goto out_unlock;
1272
1273 /* Ok, finally just insert the thing.. */
1274 entry = pte_mkspecial(pfn_pte(pfn, prot));
1275 set_pte_at(mm, addr, pte, entry);
1276 update_mmu_cache(vma, addr, entry); /* XXX: why not for insert_page? */
1277
1278 retval = 0;
1279out_unlock:
1280 pte_unmap_unlock(pte, ptl);
1281out:
1282 return retval;
1283}
1284
Nick Piggine0dc0d82007-02-12 00:51:36 -08001285/**
1286 * vm_insert_pfn - insert single pfn into user vma
1287 * @vma: user vma to map to
1288 * @addr: target user address of this page
1289 * @pfn: source kernel pfn
1290 *
1291 * Similar to vm_inert_page, this allows drivers to insert individual pages
1292 * they've allocated into a user vma. Same comments apply.
1293 *
1294 * This function should only be called from a vm_ops->fault handler, and
1295 * in that case the handler should return NULL.
1296 */
1297int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
Nick Piggin423bad62008-04-28 02:13:01 -07001298 unsigned long pfn)
Nick Piggine0dc0d82007-02-12 00:51:36 -08001299{
Nick Piggin7e675132008-04-28 02:13:00 -07001300 /*
1301 * Technically, architectures with pte_special can avoid all these
1302 * restrictions (same for remap_pfn_range). However we would like
1303 * consistency in testing and feature parity among all, so we should
1304 * try to keep these invariants in place for everybody.
1305 */
Jared Hulbertb379d792008-04-28 02:12:58 -07001306 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
1307 BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
1308 (VM_PFNMAP|VM_MIXEDMAP));
1309 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1310 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
Nick Piggine0dc0d82007-02-12 00:51:36 -08001311
Nick Piggin423bad62008-04-28 02:13:01 -07001312 if (addr < vma->vm_start || addr >= vma->vm_end)
1313 return -EFAULT;
1314 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
Nick Piggine0dc0d82007-02-12 00:51:36 -08001315}
1316EXPORT_SYMBOL(vm_insert_pfn);
1317
Nick Piggin423bad62008-04-28 02:13:01 -07001318int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
1319 unsigned long pfn)
1320{
1321 BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
1322
1323 if (addr < vma->vm_start || addr >= vma->vm_end)
1324 return -EFAULT;
1325
1326 /*
1327 * If we don't have pte special, then we have to use the pfn_valid()
1328 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1329 * refcount the page if pfn_valid is true (hence insert_page rather
1330 * than insert_pfn).
1331 */
1332 if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) {
1333 struct page *page;
1334
1335 page = pfn_to_page(pfn);
1336 return insert_page(vma, addr, page, vma->vm_page_prot);
1337 }
1338 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
1339}
1340EXPORT_SYMBOL(vm_insert_mixed);
1341
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001342/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 * maps a range of physical memory into the requested pages. the old
1344 * mappings are removed. any references to nonexistent pages results
1345 * in null mappings (currently treated as "copy-on-access")
1346 */
1347static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1348 unsigned long addr, unsigned long end,
1349 unsigned long pfn, pgprot_t prot)
1350{
1351 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001352 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353
Hugh Dickinsc74df322005-10-29 18:16:23 -07001354 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 if (!pte)
1356 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001357 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 do {
1359 BUG_ON(!pte_none(*pte));
Nick Piggin7e675132008-04-28 02:13:00 -07001360 set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 pfn++;
1362 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001363 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001364 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 return 0;
1366}
1367
1368static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1369 unsigned long addr, unsigned long end,
1370 unsigned long pfn, pgprot_t prot)
1371{
1372 pmd_t *pmd;
1373 unsigned long next;
1374
1375 pfn -= addr >> PAGE_SHIFT;
1376 pmd = pmd_alloc(mm, pud, addr);
1377 if (!pmd)
1378 return -ENOMEM;
1379 do {
1380 next = pmd_addr_end(addr, end);
1381 if (remap_pte_range(mm, pmd, addr, next,
1382 pfn + (addr >> PAGE_SHIFT), prot))
1383 return -ENOMEM;
1384 } while (pmd++, addr = next, addr != end);
1385 return 0;
1386}
1387
1388static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1389 unsigned long addr, unsigned long end,
1390 unsigned long pfn, pgprot_t prot)
1391{
1392 pud_t *pud;
1393 unsigned long next;
1394
1395 pfn -= addr >> PAGE_SHIFT;
1396 pud = pud_alloc(mm, pgd, addr);
1397 if (!pud)
1398 return -ENOMEM;
1399 do {
1400 next = pud_addr_end(addr, end);
1401 if (remap_pmd_range(mm, pud, addr, next,
1402 pfn + (addr >> PAGE_SHIFT), prot))
1403 return -ENOMEM;
1404 } while (pud++, addr = next, addr != end);
1405 return 0;
1406}
1407
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001408/**
1409 * remap_pfn_range - remap kernel memory to userspace
1410 * @vma: user vma to map to
1411 * @addr: target user address to start at
1412 * @pfn: physical address of kernel memory
1413 * @size: size of map area
1414 * @prot: page protection flags for this mapping
1415 *
1416 * Note: this is only safe if the mm semaphore is held when called.
1417 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1419 unsigned long pfn, unsigned long size, pgprot_t prot)
1420{
1421 pgd_t *pgd;
1422 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001423 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 struct mm_struct *mm = vma->vm_mm;
1425 int err;
1426
1427 /*
1428 * Physically remapped pages are special. Tell the
1429 * rest of the world about it:
1430 * VM_IO tells people not to look at these pages
1431 * (accesses can have side effects).
Hugh Dickins0b14c172005-11-21 21:32:15 -08001432 * VM_RESERVED is specified all over the place, because
1433 * in 2.4 it kept swapout's vma scan off this vma; but
1434 * in 2.6 the LRU scan won't even find its pages, so this
1435 * flag means no more than count its pages in reserved_vm,
1436 * and omit it from core dump, even when VM_IO turned off.
Linus Torvalds6aab3412005-11-28 14:34:23 -08001437 * VM_PFNMAP tells the core MM that the base pages are just
1438 * raw PFN mappings, and do not have a "struct page" associated
1439 * with them.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001440 *
1441 * There's a horrible special case to handle copy-on-write
1442 * behaviour that some programs depend on. We mark the "original"
1443 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 */
Linus Torvalds67121172005-12-11 20:38:17 -08001445 if (is_cow_mapping(vma->vm_flags)) {
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001446 if (addr != vma->vm_start || end != vma->vm_end)
Linus Torvalds7fc7e2e2005-12-11 19:57:52 -08001447 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001448 vma->vm_pgoff = pfn;
1449 }
1450
Linus Torvalds6aab3412005-11-28 14:34:23 -08001451 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452
1453 BUG_ON(addr >= end);
1454 pfn -= addr >> PAGE_SHIFT;
1455 pgd = pgd_offset(mm, addr);
1456 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 do {
1458 next = pgd_addr_end(addr, end);
1459 err = remap_pud_range(mm, pgd, addr, next,
1460 pfn + (addr >> PAGE_SHIFT), prot);
1461 if (err)
1462 break;
1463 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 return err;
1465}
1466EXPORT_SYMBOL(remap_pfn_range);
1467
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001468static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1469 unsigned long addr, unsigned long end,
1470 pte_fn_t fn, void *data)
1471{
1472 pte_t *pte;
1473 int err;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001474 pgtable_t token;
Borislav Petkov94909912007-05-06 14:49:17 -07001475 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001476
1477 pte = (mm == &init_mm) ?
1478 pte_alloc_kernel(pmd, addr) :
1479 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1480 if (!pte)
1481 return -ENOMEM;
1482
1483 BUG_ON(pmd_huge(*pmd));
1484
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001485 token = pmd_pgtable(*pmd);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001486
1487 do {
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001488 err = fn(pte, token, addr, data);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001489 if (err)
1490 break;
1491 } while (pte++, addr += PAGE_SIZE, addr != end);
1492
1493 if (mm != &init_mm)
1494 pte_unmap_unlock(pte-1, ptl);
1495 return err;
1496}
1497
1498static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1499 unsigned long addr, unsigned long end,
1500 pte_fn_t fn, void *data)
1501{
1502 pmd_t *pmd;
1503 unsigned long next;
1504 int err;
1505
1506 pmd = pmd_alloc(mm, pud, addr);
1507 if (!pmd)
1508 return -ENOMEM;
1509 do {
1510 next = pmd_addr_end(addr, end);
1511 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1512 if (err)
1513 break;
1514 } while (pmd++, addr = next, addr != end);
1515 return err;
1516}
1517
1518static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1519 unsigned long addr, unsigned long end,
1520 pte_fn_t fn, void *data)
1521{
1522 pud_t *pud;
1523 unsigned long next;
1524 int err;
1525
1526 pud = pud_alloc(mm, pgd, addr);
1527 if (!pud)
1528 return -ENOMEM;
1529 do {
1530 next = pud_addr_end(addr, end);
1531 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1532 if (err)
1533 break;
1534 } while (pud++, addr = next, addr != end);
1535 return err;
1536}
1537
1538/*
1539 * Scan a region of virtual memory, filling in page tables as necessary
1540 * and calling a provided function on each leaf page table.
1541 */
1542int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1543 unsigned long size, pte_fn_t fn, void *data)
1544{
1545 pgd_t *pgd;
1546 unsigned long next;
1547 unsigned long end = addr + size;
1548 int err;
1549
1550 BUG_ON(addr >= end);
1551 pgd = pgd_offset(mm, addr);
1552 do {
1553 next = pgd_addr_end(addr, end);
1554 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1555 if (err)
1556 break;
1557 } while (pgd++, addr = next, addr != end);
1558 return err;
1559}
1560EXPORT_SYMBOL_GPL(apply_to_page_range);
1561
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001563 * handle_pte_fault chooses page fault handler according to an entry
1564 * which was read non-atomically. Before making any commitment, on
1565 * those architectures or configurations (e.g. i386 with PAE) which
1566 * might give a mix of unmatched parts, do_swap_page and do_file_page
1567 * must check under lock before unmapping the pte and proceeding
1568 * (but do_wp_page is only called after already making such a check;
1569 * and do_anonymous_page and do_no_page can safely check later on).
1570 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001571static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001572 pte_t *page_table, pte_t orig_pte)
1573{
1574 int same = 1;
1575#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1576 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001577 spinlock_t *ptl = pte_lockptr(mm, pmd);
1578 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001579 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001580 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001581 }
1582#endif
1583 pte_unmap(page_table);
1584 return same;
1585}
1586
1587/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1589 * servicing faults for write access. In the normal case, do always want
1590 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1591 * that do not have writing enabled, when used by access_process_vm.
1592 */
1593static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1594{
1595 if (likely(vma->vm_flags & VM_WRITE))
1596 pte = pte_mkwrite(pte);
1597 return pte;
1598}
1599
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001600static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma)
Linus Torvalds6aab3412005-11-28 14:34:23 -08001601{
1602 /*
1603 * If the source page was a PFN mapping, we don't have
1604 * a "struct page" for it. We do a best-effort copy by
1605 * just copying from the original user address. If that
1606 * fails, we just zero-fill it. Live with it.
1607 */
1608 if (unlikely(!src)) {
1609 void *kaddr = kmap_atomic(dst, KM_USER0);
Linus Torvalds5d2a2dbbc2005-11-29 14:07:55 -08001610 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1611
1612 /*
1613 * This really shouldn't fail, because the page is there
1614 * in the page tables. But it might just be unreadable,
1615 * in which case we just give up and fill the result with
1616 * zeroes.
1617 */
1618 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Linus Torvalds6aab3412005-11-28 14:34:23 -08001619 memset(kaddr, 0, PAGE_SIZE);
1620 kunmap_atomic(kaddr, KM_USER0);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07001621 flush_dcache_page(dst);
Nick Piggin0ed361d2008-02-04 22:29:34 -08001622 } else
1623 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001624}
1625
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 * This routine handles present pages, when users try to write
1628 * to a shared page. It is done by copying the page to a new address
1629 * and decrementing the shared-page counter for the old page.
1630 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 * Note that this routine assumes that the protection checks have been
1632 * done by the caller (the low-level page fault routine in most cases).
1633 * Thus we can safely just mark it writable once we've done any necessary
1634 * COW.
1635 *
1636 * We also mark the page dirty at this point even though the page will
1637 * change only once the write actually happens. This avoids a few races,
1638 * and potentially makes it more efficient.
1639 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001640 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1641 * but allow concurrent faults), with pte both mapped and locked.
1642 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001644static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1645 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001646 spinlock_t *ptl, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647{
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001648 struct page *old_page, *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 pte_t entry;
Nick Piggin83c54072007-07-19 01:47:05 -07001650 int reuse = 0, ret = 0;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001651 int page_mkwrite = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001652 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653
Linus Torvalds6aab3412005-11-28 14:34:23 -08001654 old_page = vm_normal_page(vma, address, orig_pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001655 if (!old_page)
1656 goto gotten;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001658 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001659 * Take out anonymous pages first, anonymous shared vmas are
1660 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001661 */
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001662 if (PageAnon(old_page)) {
1663 if (!TestSetPageLocked(old_page)) {
1664 reuse = can_share_swap_page(old_page);
1665 unlock_page(old_page);
1666 }
1667 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001668 (VM_WRITE|VM_SHARED))) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001669 /*
1670 * Only catch write-faults on shared writable pages,
1671 * read-only shared pages can get COWed by
1672 * get_user_pages(.write=1, .force=1).
1673 */
David Howells9637a5e2006-06-23 02:03:43 -07001674 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
1675 /*
1676 * Notify the address space that the page is about to
1677 * become writable so that it can prohibit this or wait
1678 * for the page to get into an appropriate state.
1679 *
1680 * We do this without the lock held, so that it can
1681 * sleep if it needs to.
1682 */
1683 page_cache_get(old_page);
1684 pte_unmap_unlock(page_table, ptl);
1685
1686 if (vma->vm_ops->page_mkwrite(vma, old_page) < 0)
1687 goto unwritable_page;
1688
David Howells9637a5e2006-06-23 02:03:43 -07001689 /*
1690 * Since we dropped the lock we need to revalidate
1691 * the PTE as someone else may have changed it. If
1692 * they did, we just return, as we can count on the
1693 * MMU to tell us if they didn't also make it writable.
1694 */
1695 page_table = pte_offset_map_lock(mm, pmd, address,
1696 &ptl);
Hugh Dickinsc3704ce2007-02-10 01:43:00 -08001697 page_cache_release(old_page);
David Howells9637a5e2006-06-23 02:03:43 -07001698 if (!pte_same(*page_table, orig_pte))
1699 goto unlock;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001700
1701 page_mkwrite = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001703 dirty_page = old_page;
1704 get_page(dirty_page);
David Howells9637a5e2006-06-23 02:03:43 -07001705 reuse = 1;
David Howells9637a5e2006-06-23 02:03:43 -07001706 }
1707
1708 if (reuse) {
1709 flush_cache_page(vma, address, pte_pfn(orig_pte));
1710 entry = pte_mkyoung(orig_pte);
1711 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
KAMEZAWA Hiroyuki954ffcb2007-10-16 01:25:44 -07001712 if (ptep_set_access_flags(vma, address, page_table, entry,1))
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07001713 update_mmu_cache(vma, address, entry);
David Howells9637a5e2006-06-23 02:03:43 -07001714 ret |= VM_FAULT_WRITE;
1715 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717
1718 /*
1719 * Ok, we need to copy. Oh, well..
1720 */
Nick Pigginb5810032005-10-29 18:16:12 -07001721 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08001722gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001723 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724
1725 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07001726 goto oom;
Nick Piggin557ed1f2007-10-16 01:24:40 -07001727 VM_BUG_ON(old_page == ZERO_PAGE(0));
1728 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
1729 if (!new_page)
1730 goto oom;
1731 cow_user_page(new_page, old_page, address, vma);
Nick Piggin0ed361d2008-02-04 22:29:34 -08001732 __SetPageUptodate(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001733
Balbir Singhe1a1cd52008-02-07 00:14:02 -08001734 if (mem_cgroup_charge(new_page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001735 goto oom_free_new;
1736
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 /*
1738 * Re-check the pte - we dropped the lock
1739 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001740 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07001741 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08001742 if (old_page) {
Nick Piggin7de6b802006-12-22 01:09:33 -08001743 page_remove_rmap(old_page, vma);
Hugh Dickins920fc352005-11-21 21:32:17 -08001744 if (!PageAnon(old_page)) {
1745 dec_mm_counter(mm, file_rss);
1746 inc_mm_counter(mm, anon_rss);
1747 }
1748 } else
Hugh Dickins42946212005-10-29 18:16:05 -07001749 inc_mm_counter(mm, anon_rss);
Ben Collinseca35132005-11-29 11:45:26 -08001750 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07001751 entry = mk_pte(new_page, vma->vm_page_prot);
1752 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Siddha, Suresh B4ce072f2006-09-29 01:58:42 -07001753 /*
1754 * Clear the pte entry and flush it first, before updating the
1755 * pte with the new entry. This will avoid a race condition
1756 * seen in the presence of one thread doing SMC and another
1757 * thread doing COW.
1758 */
1759 ptep_clear_flush(vma, address, page_table);
1760 set_pte_at(mm, address, page_table, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07001761 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 lru_cache_add_active(new_page);
Nick Piggin9617d952006-01-06 00:11:12 -08001763 page_add_new_anon_rmap(new_page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764
1765 /* Free the old page.. */
1766 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001767 ret |= VM_FAULT_WRITE;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001768 } else
1769 mem_cgroup_uncharge_page(new_page);
1770
Hugh Dickins920fc352005-11-21 21:32:17 -08001771 if (new_page)
1772 page_cache_release(new_page);
1773 if (old_page)
1774 page_cache_release(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001775unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001776 pte_unmap_unlock(page_table, ptl);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001777 if (dirty_page) {
Anton Salikhmetov8f7b3d12008-01-23 02:21:18 +03001778 if (vma->vm_file)
1779 file_update_time(vma->vm_file);
1780
Nick Piggin79352892007-07-19 01:47:22 -07001781 /*
1782 * Yes, Virginia, this is actually required to prevent a race
1783 * with clear_page_dirty_for_io() from clearing the page dirty
1784 * bit after it clear all dirty ptes, but before a racing
1785 * do_wp_page installs a dirty pte.
1786 *
1787 * do_no_page is protected similarly.
1788 */
1789 wait_on_page_locked(dirty_page);
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001790 set_page_dirty_balance(dirty_page, page_mkwrite);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001791 put_page(dirty_page);
1792 }
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001793 return ret;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001794oom_free_new:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08001795 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001796oom:
Hugh Dickins920fc352005-11-21 21:32:17 -08001797 if (old_page)
1798 page_cache_release(old_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 return VM_FAULT_OOM;
David Howells9637a5e2006-06-23 02:03:43 -07001800
1801unwritable_page:
1802 page_cache_release(old_page);
1803 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804}
1805
1806/*
1807 * Helper functions for unmap_mapping_range().
1808 *
1809 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1810 *
1811 * We have to restart searching the prio_tree whenever we drop the lock,
1812 * since the iterator is only valid while the lock is held, and anyway
1813 * a later vma might be split and reinserted earlier while lock dropped.
1814 *
1815 * The list of nonlinear vmas could be handled more efficiently, using
1816 * a placeholder, but handle it in the same way until a need is shown.
1817 * It is important to search the prio_tree before nonlinear list: a vma
1818 * may become nonlinear and be shifted from prio_tree to nonlinear list
1819 * while the lock is dropped; but never shifted from list to prio_tree.
1820 *
1821 * In order to make forward progress despite restarting the search,
1822 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1823 * quickly skip it next time around. Since the prio_tree search only
1824 * shows us those vmas affected by unmapping the range in question, we
1825 * can't efficiently keep all vmas in step with mapping->truncate_count:
1826 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1827 * mapping->truncate_count and vma->vm_truncate_count are protected by
1828 * i_mmap_lock.
1829 *
1830 * In order to make forward progress despite repeatedly restarting some
Hugh Dickinsee39b372005-04-19 13:29:15 -07001831 * large vma, note the restart_addr from unmap_vmas when it breaks out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 * and restart from that address when we reach that vma again. It might
1833 * have been split or merged, shrunk or extended, but never shifted: so
1834 * restart_addr remains valid so long as it remains in the vma's range.
1835 * unmap_mapping_range forces truncate_count to leap over page-aligned
1836 * values so we can save vma's restart_addr in its truncate_count field.
1837 */
1838#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1839
1840static void reset_vma_truncate_counts(struct address_space *mapping)
1841{
1842 struct vm_area_struct *vma;
1843 struct prio_tree_iter iter;
1844
1845 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1846 vma->vm_truncate_count = 0;
1847 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1848 vma->vm_truncate_count = 0;
1849}
1850
1851static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1852 unsigned long start_addr, unsigned long end_addr,
1853 struct zap_details *details)
1854{
1855 unsigned long restart_addr;
1856 int need_break;
1857
Nick Piggind00806b2007-07-19 01:46:57 -07001858 /*
1859 * files that support invalidating or truncating portions of the
Nick Piggind0217ac2007-07-19 01:47:03 -07001860 * file from under mmaped areas must have their ->fault function
Nick Piggin83c54072007-07-19 01:47:05 -07001861 * return a locked page (and set VM_FAULT_LOCKED in the return).
1862 * This provides synchronisation against concurrent unmapping here.
Nick Piggind00806b2007-07-19 01:46:57 -07001863 */
Nick Piggind00806b2007-07-19 01:46:57 -07001864
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865again:
1866 restart_addr = vma->vm_truncate_count;
1867 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1868 start_addr = restart_addr;
1869 if (start_addr >= end_addr) {
1870 /* Top of vma has been split off since last time */
1871 vma->vm_truncate_count = details->truncate_count;
1872 return 0;
1873 }
1874 }
1875
Hugh Dickinsee39b372005-04-19 13:29:15 -07001876 restart_addr = zap_page_range(vma, start_addr,
1877 end_addr - start_addr, details);
Nick Piggin95c354f2008-01-30 13:31:20 +01001878 need_break = need_resched() || spin_needbreak(details->i_mmap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879
Hugh Dickinsee39b372005-04-19 13:29:15 -07001880 if (restart_addr >= end_addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 /* We have now completed this vma: mark it so */
1882 vma->vm_truncate_count = details->truncate_count;
1883 if (!need_break)
1884 return 0;
1885 } else {
1886 /* Note restart_addr in vma's truncate_count field */
Hugh Dickinsee39b372005-04-19 13:29:15 -07001887 vma->vm_truncate_count = restart_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 if (!need_break)
1889 goto again;
1890 }
1891
1892 spin_unlock(details->i_mmap_lock);
1893 cond_resched();
1894 spin_lock(details->i_mmap_lock);
1895 return -EINTR;
1896}
1897
1898static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1899 struct zap_details *details)
1900{
1901 struct vm_area_struct *vma;
1902 struct prio_tree_iter iter;
1903 pgoff_t vba, vea, zba, zea;
1904
1905restart:
1906 vma_prio_tree_foreach(vma, &iter, root,
1907 details->first_index, details->last_index) {
1908 /* Skip quickly over those we have already dealt with */
1909 if (vma->vm_truncate_count == details->truncate_count)
1910 continue;
1911
1912 vba = vma->vm_pgoff;
1913 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1914 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1915 zba = details->first_index;
1916 if (zba < vba)
1917 zba = vba;
1918 zea = details->last_index;
1919 if (zea > vea)
1920 zea = vea;
1921
1922 if (unmap_mapping_range_vma(vma,
1923 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1924 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1925 details) < 0)
1926 goto restart;
1927 }
1928}
1929
1930static inline void unmap_mapping_range_list(struct list_head *head,
1931 struct zap_details *details)
1932{
1933 struct vm_area_struct *vma;
1934
1935 /*
1936 * In nonlinear VMAs there is no correspondence between virtual address
1937 * offset and file offset. So we must perform an exhaustive search
1938 * across *all* the pages in each nonlinear VMA, not just the pages
1939 * whose virtual address lies outside the file truncation point.
1940 */
1941restart:
1942 list_for_each_entry(vma, head, shared.vm_set.list) {
1943 /* Skip quickly over those we have already dealt with */
1944 if (vma->vm_truncate_count == details->truncate_count)
1945 continue;
1946 details->nonlinear_vma = vma;
1947 if (unmap_mapping_range_vma(vma, vma->vm_start,
1948 vma->vm_end, details) < 0)
1949 goto restart;
1950 }
1951}
1952
1953/**
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001954 * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file.
Martin Waitz3d410882005-06-23 22:05:21 -07001955 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 * @holebegin: byte in first page to unmap, relative to the start of
1957 * the underlying file. This will be rounded down to a PAGE_SIZE
1958 * boundary. Note that this is different from vmtruncate(), which
1959 * must keep the partial page. In contrast, we must get rid of
1960 * partial pages.
1961 * @holelen: size of prospective hole in bytes. This will be rounded
1962 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1963 * end of the file.
1964 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1965 * but 0 when invalidating pagecache, don't throw away private data.
1966 */
1967void unmap_mapping_range(struct address_space *mapping,
1968 loff_t const holebegin, loff_t const holelen, int even_cows)
1969{
1970 struct zap_details details;
1971 pgoff_t hba = holebegin >> PAGE_SHIFT;
1972 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1973
1974 /* Check for overflow. */
1975 if (sizeof(holelen) > sizeof(hlen)) {
1976 long long holeend =
1977 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1978 if (holeend & ~(long long)ULONG_MAX)
1979 hlen = ULONG_MAX - hba + 1;
1980 }
1981
1982 details.check_mapping = even_cows? NULL: mapping;
1983 details.nonlinear_vma = NULL;
1984 details.first_index = hba;
1985 details.last_index = hba + hlen - 1;
1986 if (details.last_index < details.first_index)
1987 details.last_index = ULONG_MAX;
1988 details.i_mmap_lock = &mapping->i_mmap_lock;
1989
1990 spin_lock(&mapping->i_mmap_lock);
1991
Nick Piggind00806b2007-07-19 01:46:57 -07001992 /* Protect against endless unmapping loops */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 mapping->truncate_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 if (unlikely(is_restart_addr(mapping->truncate_count))) {
1995 if (mapping->truncate_count == 0)
1996 reset_vma_truncate_counts(mapping);
1997 mapping->truncate_count++;
1998 }
1999 details.truncate_count = mapping->truncate_count;
2000
2001 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
2002 unmap_mapping_range_tree(&mapping->i_mmap, &details);
2003 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
2004 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
2005 spin_unlock(&mapping->i_mmap_lock);
2006}
2007EXPORT_SYMBOL(unmap_mapping_range);
2008
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07002009/**
2010 * vmtruncate - unmap mappings "freed" by truncate() syscall
2011 * @inode: inode of the file used
2012 * @offset: file offset to start truncating
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 *
2014 * NOTE! We have to be ready to update the memory sharing
2015 * between the file and the memory map for a potential last
2016 * incomplete page. Ugly, but necessary.
2017 */
2018int vmtruncate(struct inode * inode, loff_t offset)
2019{
Christoph Hellwig61d50482008-02-04 22:28:56 -08002020 if (inode->i_size < offset) {
2021 unsigned long limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022
Christoph Hellwig61d50482008-02-04 22:28:56 -08002023 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
2024 if (limit != RLIM_INFINITY && offset > limit)
2025 goto out_sig;
2026 if (offset > inode->i_sb->s_maxbytes)
2027 goto out_big;
2028 i_size_write(inode, offset);
2029 } else {
2030 struct address_space *mapping = inode->i_mapping;
Nick Piggind00806b2007-07-19 01:46:57 -07002031
Christoph Hellwig61d50482008-02-04 22:28:56 -08002032 /*
2033 * truncation of in-use swapfiles is disallowed - it would
2034 * cause subsequent swapout to scribble on the now-freed
2035 * blocks.
2036 */
2037 if (IS_SWAPFILE(inode))
2038 return -ETXTBSY;
2039 i_size_write(inode, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040
Christoph Hellwig61d50482008-02-04 22:28:56 -08002041 /*
2042 * unmap_mapping_range is called twice, first simply for
2043 * efficiency so that truncate_inode_pages does fewer
2044 * single-page unmaps. However after this first call, and
2045 * before truncate_inode_pages finishes, it is possible for
2046 * private pages to be COWed, which remain after
2047 * truncate_inode_pages finishes, hence the second
2048 * unmap_mapping_range call must be made for correctness.
2049 */
2050 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2051 truncate_inode_pages(mapping, offset);
2052 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2053 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 if (inode->i_op && inode->i_op->truncate)
2056 inode->i_op->truncate(inode);
2057 return 0;
Christoph Hellwig61d50482008-02-04 22:28:56 -08002058
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059out_sig:
2060 send_sig(SIGXFSZ, current, 0);
2061out_big:
2062 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064EXPORT_SYMBOL(vmtruncate);
2065
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002066int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
2067{
2068 struct address_space *mapping = inode->i_mapping;
2069
2070 /*
2071 * If the underlying filesystem is not going to provide
2072 * a way to truncate a range of blocks (punch a hole) -
2073 * we should return failure right now.
2074 */
2075 if (!inode->i_op || !inode->i_op->truncate_range)
2076 return -ENOSYS;
2077
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002078 mutex_lock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002079 down_write(&inode->i_alloc_sem);
2080 unmap_mapping_range(mapping, offset, (end - offset), 1);
2081 truncate_inode_pages_range(mapping, offset, end);
Nick Piggind00806b2007-07-19 01:46:57 -07002082 unmap_mapping_range(mapping, offset, (end - offset), 1);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002083 inode->i_op->truncate_range(inode, offset, end);
2084 up_write(&inode->i_alloc_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002085 mutex_unlock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002086
2087 return 0;
2088}
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002089
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002091 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2092 * but allow concurrent faults), and pte mapped but not yet locked.
2093 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002095static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2096 unsigned long address, pte_t *page_table, pmd_t *pmd,
2097 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002099 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 struct page *page;
Hugh Dickins65500d22005-10-29 18:15:59 -07002101 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 pte_t pte;
Nick Piggin83c54072007-07-19 01:47:05 -07002103 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002105 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002106 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002107
2108 entry = pte_to_swp_entry(orig_pte);
Christoph Lameter06972122006-06-23 02:03:35 -07002109 if (is_migration_entry(entry)) {
2110 migration_entry_wait(mm, pmd, address);
2111 goto out;
2112 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002113 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 page = lookup_swap_cache(entry);
2115 if (!page) {
Ashwin Chaugule098fe652006-12-06 20:31:54 -08002116 grab_swap_token(); /* Contend for token _before_ read-in */
Hugh Dickins02098fe2008-02-04 22:28:42 -08002117 page = swapin_readahead(entry,
2118 GFP_HIGHUSER_MOVABLE, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 if (!page) {
2120 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002121 * Back out if somebody else faulted in this pte
2122 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002124 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 if (likely(pte_same(*page_table, orig_pte)))
2126 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002127 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002128 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 }
2130
2131 /* Had to read the page from swap area: Major fault */
2132 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002133 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 }
2135
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002136 if (mem_cgroup_charge(page, mm, GFP_KERNEL)) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002137 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2138 ret = VM_FAULT_OOM;
2139 goto out;
2140 }
2141
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 mark_page_accessed(page);
2143 lock_page(page);
Balbir Singh20a10222007-11-14 17:00:33 -08002144 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145
2146 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002147 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002149 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002150 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002151 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002152
2153 if (unlikely(!PageUptodate(page))) {
2154 ret = VM_FAULT_SIGBUS;
2155 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 }
2157
2158 /* The page isn't present yet, go ahead with the fault. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159
Hugh Dickins42946212005-10-29 18:16:05 -07002160 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 pte = mk_pte(page, vma->vm_page_prot);
2162 if (write_access && can_share_swap_page(page)) {
2163 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2164 write_access = 0;
2165 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166
2167 flush_icache_page(vma, page);
2168 set_pte_at(mm, address, page_table, pte);
2169 page_add_anon_rmap(page, vma, address);
2170
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002171 swap_free(entry);
2172 if (vm_swap_full())
2173 remove_exclusive_swap_page(page);
2174 unlock_page(page);
2175
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 if (write_access) {
Hugh Dickins61469f12008-03-04 14:29:04 -08002177 ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
2178 if (ret & VM_FAULT_ERROR)
2179 ret &= VM_FAULT_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 goto out;
2181 }
2182
2183 /* No need to invalidate - it was non-present before */
2184 update_mmu_cache(vma, address, pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07002185unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002186 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187out:
2188 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002189out_nomap:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002190 mem_cgroup_uncharge_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002191 pte_unmap_unlock(page_table, ptl);
Kirill Korotaevb8107482005-05-16 21:53:50 -07002192 unlock_page(page);
2193 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002194 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195}
2196
2197/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -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 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002202static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2203 unsigned long address, pte_t *page_table, pmd_t *pmd,
2204 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002206 struct page *page;
2207 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209
Nick Piggin557ed1f2007-10-16 01:24:40 -07002210 /* Allocate our own private page. */
2211 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212
Nick Piggin557ed1f2007-10-16 01:24:40 -07002213 if (unlikely(anon_vma_prepare(vma)))
2214 goto oom;
2215 page = alloc_zeroed_user_highpage_movable(vma, address);
2216 if (!page)
2217 goto oom;
Nick Piggin0ed361d2008-02-04 22:29:34 -08002218 __SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002220 if (mem_cgroup_charge(page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002221 goto oom_free_page;
2222
Nick Piggin557ed1f2007-10-16 01:24:40 -07002223 entry = mk_pte(page, vma->vm_page_prot);
2224 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002225
Nick Piggin557ed1f2007-10-16 01:24:40 -07002226 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2227 if (!pte_none(*page_table))
2228 goto release;
2229 inc_mm_counter(mm, anon_rss);
2230 lru_cache_add_active(page);
2231 page_add_new_anon_rmap(page, vma, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002232 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233
2234 /* No need to invalidate - it was non-present before */
Hugh Dickins65500d22005-10-29 18:15:59 -07002235 update_mmu_cache(vma, address, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002236unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002237 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002238 return 0;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002239release:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002240 mem_cgroup_uncharge_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002241 page_cache_release(page);
2242 goto unlock;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002243oom_free_page:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002244 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002245oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 return VM_FAULT_OOM;
2247}
2248
2249/*
Nick Piggin54cb8822007-07-19 01:46:59 -07002250 * __do_fault() tries to create a new page mapping. It aggressively
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 * tries to share with existing pages, but makes a separate copy if
Nick Piggin54cb8822007-07-19 01:46:59 -07002252 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2253 * the next page fault.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 *
2255 * As this is called only for pages that do not currently exist, we
2256 * do not need to flush old virtual caches or the TLB.
2257 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002258 * We enter with non-exclusive mmap_sem (to exclude vma changes,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002259 * but allow concurrent faults), and pte neither mapped nor locked.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002260 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 */
Nick Piggin54cb8822007-07-19 01:46:59 -07002262static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002263 unsigned long address, pmd_t *pmd,
Nick Piggin54cb8822007-07-19 01:46:59 -07002264 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265{
Hugh Dickins16abfa02007-10-04 16:56:06 +01002266 pte_t *page_table;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002267 spinlock_t *ptl;
Nick Piggind0217ac2007-07-19 01:47:03 -07002268 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 int anon = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002271 struct page *dirty_page = NULL;
Nick Piggind0217ac2007-07-19 01:47:03 -07002272 struct vm_fault vmf;
2273 int ret;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002274 int page_mkwrite = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07002275
Nick Piggind0217ac2007-07-19 01:47:03 -07002276 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2277 vmf.pgoff = pgoff;
2278 vmf.flags = flags;
2279 vmf.page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280
Hugh Dickins325f04d2005-11-29 16:55:48 +00002281 BUG_ON(vma->vm_flags & VM_PFNMAP);
2282
Nick Piggin3c18ddd2008-04-28 02:12:10 -07002283 ret = vma->vm_ops->fault(vma, &vmf);
2284 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2285 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286
Nick Piggind00806b2007-07-19 01:46:57 -07002287 /*
Nick Piggind0217ac2007-07-19 01:47:03 -07002288 * For consistency in subsequent calls, make the faulted page always
Nick Piggind00806b2007-07-19 01:46:57 -07002289 * locked.
2290 */
Nick Piggin83c54072007-07-19 01:47:05 -07002291 if (unlikely(!(ret & VM_FAULT_LOCKED)))
Nick Piggind0217ac2007-07-19 01:47:03 -07002292 lock_page(vmf.page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002293 else
Nick Piggind0217ac2007-07-19 01:47:03 -07002294 VM_BUG_ON(!PageLocked(vmf.page));
Nick Piggind00806b2007-07-19 01:46:57 -07002295
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 /*
2297 * Should we do an early C-O-W break?
2298 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002299 page = vmf.page;
Nick Piggin54cb8822007-07-19 01:46:59 -07002300 if (flags & FAULT_FLAG_WRITE) {
David Howells9637a5e2006-06-23 02:03:43 -07002301 if (!(vma->vm_flags & VM_SHARED)) {
Nick Piggin54cb8822007-07-19 01:46:59 -07002302 anon = 1;
Nick Piggind00806b2007-07-19 01:46:57 -07002303 if (unlikely(anon_vma_prepare(vma))) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002304 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002305 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002306 }
Nick Piggin83c54072007-07-19 01:47:05 -07002307 page = alloc_page_vma(GFP_HIGHUSER_MOVABLE,
2308 vma, address);
Nick Piggind00806b2007-07-19 01:46:57 -07002309 if (!page) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002310 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002311 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002312 }
Nick Piggind0217ac2007-07-19 01:47:03 -07002313 copy_user_highpage(page, vmf.page, address, vma);
Nick Piggin0ed361d2008-02-04 22:29:34 -08002314 __SetPageUptodate(page);
David Howells9637a5e2006-06-23 02:03:43 -07002315 } else {
Nick Piggin54cb8822007-07-19 01:46:59 -07002316 /*
2317 * If the page will be shareable, see if the backing
David Howells9637a5e2006-06-23 02:03:43 -07002318 * address space wants to know that the page is about
Nick Piggin54cb8822007-07-19 01:46:59 -07002319 * to become writable
2320 */
Mark Fasheh69676142007-07-19 01:47:00 -07002321 if (vma->vm_ops->page_mkwrite) {
2322 unlock_page(page);
2323 if (vma->vm_ops->page_mkwrite(vma, page) < 0) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002324 ret = VM_FAULT_SIGBUS;
2325 anon = 1; /* no anon but release vmf.page */
Mark Fasheh69676142007-07-19 01:47:00 -07002326 goto out_unlocked;
2327 }
2328 lock_page(page);
Nick Piggind0217ac2007-07-19 01:47:03 -07002329 /*
2330 * XXX: this is not quite right (racy vs
2331 * invalidate) to unlock and relock the page
2332 * like this, however a better fix requires
2333 * reworking page_mkwrite locking API, which
2334 * is better done later.
2335 */
2336 if (!page->mapping) {
Nick Piggin83c54072007-07-19 01:47:05 -07002337 ret = 0;
Nick Piggind0217ac2007-07-19 01:47:03 -07002338 anon = 1; /* no anon but release vmf.page */
2339 goto out;
2340 }
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002341 page_mkwrite = 1;
David Howells9637a5e2006-06-23 02:03:43 -07002342 }
2343 }
Nick Piggin54cb8822007-07-19 01:46:59 -07002344
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 }
2346
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002347 if (mem_cgroup_charge(page, mm, GFP_KERNEL)) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002348 ret = VM_FAULT_OOM;
2349 goto out;
2350 }
2351
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002352 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353
2354 /*
2355 * This silly early PAGE_DIRTY setting removes a race
2356 * due to the bad i386 page protection. But it's valid
2357 * for other architectures too.
2358 *
2359 * Note that if write_access is true, we either now have
2360 * an exclusive copy of the page, or this is a shared mapping,
2361 * so we can make it writable and dirty to avoid having to
2362 * handle that later.
2363 */
2364 /* Only go through if we didn't race with anybody else... */
Nick Piggin54cb8822007-07-19 01:46:59 -07002365 if (likely(pte_same(*page_table, orig_pte))) {
Nick Piggind00806b2007-07-19 01:46:57 -07002366 flush_icache_page(vma, page);
2367 entry = mk_pte(page, vma->vm_page_prot);
Nick Piggin54cb8822007-07-19 01:46:59 -07002368 if (flags & FAULT_FLAG_WRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2370 set_pte_at(mm, address, page_table, entry);
2371 if (anon) {
Nick Piggind00806b2007-07-19 01:46:57 -07002372 inc_mm_counter(mm, anon_rss);
2373 lru_cache_add_active(page);
2374 page_add_new_anon_rmap(page, vma, address);
Hugh Dickinsf57e88a2005-11-21 21:32:19 -08002375 } else {
Hugh Dickins42946212005-10-29 18:16:05 -07002376 inc_mm_counter(mm, file_rss);
Nick Piggind00806b2007-07-19 01:46:57 -07002377 page_add_file_rmap(page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002378 if (flags & FAULT_FLAG_WRITE) {
Nick Piggind00806b2007-07-19 01:46:57 -07002379 dirty_page = page;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002380 get_page(dirty_page);
2381 }
Hugh Dickins42946212005-10-29 18:16:05 -07002382 }
Nick Piggind00806b2007-07-19 01:46:57 -07002383
2384 /* no need to invalidate: a not-present page won't be cached */
2385 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 } else {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002387 mem_cgroup_uncharge_page(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002388 if (anon)
2389 page_cache_release(page);
2390 else
Nick Piggin54cb8822007-07-19 01:46:59 -07002391 anon = 1; /* no anon but release faulted_page */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 }
2393
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002394 pte_unmap_unlock(page_table, ptl);
Nick Piggind00806b2007-07-19 01:46:57 -07002395
2396out:
Nick Piggind0217ac2007-07-19 01:47:03 -07002397 unlock_page(vmf.page);
Mark Fasheh69676142007-07-19 01:47:00 -07002398out_unlocked:
Nick Piggind00806b2007-07-19 01:46:57 -07002399 if (anon)
Nick Piggind0217ac2007-07-19 01:47:03 -07002400 page_cache_release(vmf.page);
Nick Piggind00806b2007-07-19 01:46:57 -07002401 else if (dirty_page) {
Anton Salikhmetov8f7b3d12008-01-23 02:21:18 +03002402 if (vma->vm_file)
2403 file_update_time(vma->vm_file);
2404
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002405 set_page_dirty_balance(dirty_page, page_mkwrite);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002406 put_page(dirty_page);
2407 }
Nick Piggind00806b2007-07-19 01:46:57 -07002408
Nick Piggin83c54072007-07-19 01:47:05 -07002409 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07002410}
Nick Piggind00806b2007-07-19 01:46:57 -07002411
Nick Piggin54cb8822007-07-19 01:46:59 -07002412static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2413 unsigned long address, pte_t *page_table, pmd_t *pmd,
2414 int write_access, pte_t orig_pte)
2415{
2416 pgoff_t pgoff = (((address & PAGE_MASK)
Dean Nelson0da7e012007-10-16 01:24:45 -07002417 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
Nick Piggin54cb8822007-07-19 01:46:59 -07002418 unsigned int flags = (write_access ? FAULT_FLAG_WRITE : 0);
2419
Hugh Dickins16abfa02007-10-04 16:56:06 +01002420 pte_unmap(page_table);
2421 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Nick Piggin54cb8822007-07-19 01:46:59 -07002422}
2423
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424
2425/*
Jes Sorensenf4b81802006-09-27 01:50:10 -07002426 * do_no_pfn() tries to create a new page mapping for a page without
2427 * a struct_page backing it
2428 *
2429 * As this is called only for pages that do not currently exist, we
2430 * do not need to flush old virtual caches or the TLB.
2431 *
2432 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2433 * but allow concurrent faults), and pte mapped but not yet locked.
2434 * We return with mmap_sem still held, but pte unmapped and unlocked.
2435 *
2436 * It is expected that the ->nopfn handler always returns the same pfn
2437 * for a given virtual mapping.
2438 *
2439 * Mark this `noinline' to prevent it from bloating the main pagefault code.
2440 */
2441static noinline int do_no_pfn(struct mm_struct *mm, struct vm_area_struct *vma,
2442 unsigned long address, pte_t *page_table, pmd_t *pmd,
2443 int write_access)
2444{
2445 spinlock_t *ptl;
2446 pte_t entry;
2447 unsigned long pfn;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002448
2449 pte_unmap(page_table);
Jared Hulbertb379d792008-04-28 02:12:58 -07002450 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
2451 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
Jes Sorensenf4b81802006-09-27 01:50:10 -07002452
2453 pfn = vma->vm_ops->nopfn(vma, address & PAGE_MASK);
Jared Hulbertb379d792008-04-28 02:12:58 -07002454
2455 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
2456
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002457 if (unlikely(pfn == NOPFN_OOM))
Jes Sorensenf4b81802006-09-27 01:50:10 -07002458 return VM_FAULT_OOM;
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002459 else if (unlikely(pfn == NOPFN_SIGBUS))
Jes Sorensenf4b81802006-09-27 01:50:10 -07002460 return VM_FAULT_SIGBUS;
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002461 else if (unlikely(pfn == NOPFN_REFAULT))
Nick Piggin83c54072007-07-19 01:47:05 -07002462 return 0;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002463
2464 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2465
2466 /* Only go through if we didn't race with anybody else... */
2467 if (pte_none(*page_table)) {
2468 entry = pfn_pte(pfn, vma->vm_page_prot);
2469 if (write_access)
2470 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2471 set_pte_at(mm, address, page_table, entry);
2472 }
2473 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002474 return 0;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002475}
2476
2477/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 * Fault of a previously existing named mapping. Repopulate the pte
2479 * from the encoded file_pte if possible. This enables swappable
2480 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002481 *
2482 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2483 * but allow concurrent faults), and pte mapped but not yet locked.
2484 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002486static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins65500d22005-10-29 18:15:59 -07002487 unsigned long address, pte_t *page_table, pmd_t *pmd,
2488 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489{
Nick Piggind0217ac2007-07-19 01:47:03 -07002490 unsigned int flags = FAULT_FLAG_NONLINEAR |
2491 (write_access ? FAULT_FLAG_WRITE : 0);
Hugh Dickins65500d22005-10-29 18:15:59 -07002492 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002494 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Nick Piggin83c54072007-07-19 01:47:05 -07002495 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496
Nick Piggind0217ac2007-07-19 01:47:03 -07002497 if (unlikely(!(vma->vm_flags & VM_NONLINEAR) ||
2498 !(vma->vm_flags & VM_CAN_NONLINEAR))) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002499 /*
2500 * Page table corrupted: show pte and kill process.
2501 */
Nick Pigginb5810032005-10-29 18:16:12 -07002502 print_bad_pte(vma, orig_pte, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002503 return VM_FAULT_OOM;
2504 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002505
2506 pgoff = pte_to_pgoff(orig_pte);
Hugh Dickins16abfa02007-10-04 16:56:06 +01002507 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508}
2509
2510/*
2511 * These routines also need to handle stuff like marking pages dirty
2512 * and/or accessed for architectures that don't do it in hardware (most
2513 * RISC architectures). The early dirtying is also good on the i386.
2514 *
2515 * There is also a hook called "update_mmu_cache()" that architectures
2516 * with external mmu caches can use to update those (ie the Sparc or
2517 * PowerPC hashed page tables that act as extended TLBs).
2518 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07002519 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2520 * but allow concurrent faults), and pte mapped but not yet locked.
2521 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 */
2523static inline int handle_pte_fault(struct mm_struct *mm,
Hugh Dickins65500d22005-10-29 18:15:59 -07002524 struct vm_area_struct *vma, unsigned long address,
2525 pte_t *pte, pmd_t *pmd, int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526{
2527 pte_t entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002528 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002530 entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002532 if (pte_none(entry)) {
Jes Sorensenf4b81802006-09-27 01:50:10 -07002533 if (vma->vm_ops) {
Nick Piggin3c18ddd2008-04-28 02:12:10 -07002534 if (likely(vma->vm_ops->fault))
Nick Piggin54cb8822007-07-19 01:46:59 -07002535 return do_linear_fault(mm, vma, address,
2536 pte, pmd, write_access, entry);
Jes Sorensenf4b81802006-09-27 01:50:10 -07002537 if (unlikely(vma->vm_ops->nopfn))
2538 return do_no_pfn(mm, vma, address, pte,
2539 pmd, write_access);
2540 }
2541 return do_anonymous_page(mm, vma, address,
2542 pte, pmd, write_access);
Hugh Dickins65500d22005-10-29 18:15:59 -07002543 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 if (pte_file(entry))
Nick Piggind0217ac2007-07-19 01:47:03 -07002545 return do_nonlinear_fault(mm, vma, address,
Hugh Dickins65500d22005-10-29 18:15:59 -07002546 pte, pmd, write_access, entry);
2547 return do_swap_page(mm, vma, address,
2548 pte, pmd, write_access, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 }
2550
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002551 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002552 spin_lock(ptl);
2553 if (unlikely(!pte_same(*pte, entry)))
2554 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 if (write_access) {
2556 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002557 return do_wp_page(mm, vma, address,
2558 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 entry = pte_mkdirty(entry);
2560 }
2561 entry = pte_mkyoung(entry);
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002562 if (ptep_set_access_flags(vma, address, pte, entry, write_access)) {
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002563 update_mmu_cache(vma, address, entry);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002564 } else {
2565 /*
2566 * This is needed only for protection faults but the arch code
2567 * is not yet telling us if this is a protection fault or not.
2568 * This still avoids useless tlb flushes for .text page faults
2569 * with threads.
2570 */
2571 if (write_access)
2572 flush_tlb_page(vma, address);
2573 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002574unlock:
2575 pte_unmap_unlock(pte, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002576 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577}
2578
2579/*
2580 * By the time we get here, we already hold the mm semaphore
2581 */
Nick Piggin83c54072007-07-19 01:47:05 -07002582int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 unsigned long address, int write_access)
2584{
2585 pgd_t *pgd;
2586 pud_t *pud;
2587 pmd_t *pmd;
2588 pte_t *pte;
2589
2590 __set_current_state(TASK_RUNNING);
2591
Christoph Lameterf8891e52006-06-30 01:55:45 -07002592 count_vm_event(PGFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01002594 if (unlikely(is_vm_hugetlb_page(vma)))
2595 return hugetlb_fault(mm, vma, address, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598 pud = pud_alloc(mm, pgd, address);
2599 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002600 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601 pmd = pmd_alloc(mm, pud, address);
2602 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002603 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604 pte = pte_alloc_map(mm, pmd, address);
2605 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002606 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607
Hugh Dickinsc74df322005-10-29 18:16:23 -07002608 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609}
2610
2611#ifndef __PAGETABLE_PUD_FOLDED
2612/*
2613 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002614 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002616int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002618 pud_t *new = pud_alloc_one(mm, address);
2619 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002620 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621
Hugh Dickins872fec12005-10-29 18:16:21 -07002622 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002623 if (pgd_present(*pgd)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002624 pud_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002625 else
2626 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07002627 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002628 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629}
2630#endif /* __PAGETABLE_PUD_FOLDED */
2631
2632#ifndef __PAGETABLE_PMD_FOLDED
2633/*
2634 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002635 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002637int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002639 pmd_t *new = pmd_alloc_one(mm, address);
2640 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002641 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642
Hugh Dickins872fec12005-10-29 18:16:21 -07002643 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07002645 if (pud_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002646 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002647 else
2648 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07002650 if (pgd_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002651 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002652 else
2653 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07002655 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002656 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657}
2658#endif /* __PAGETABLE_PMD_FOLDED */
2659
2660int make_pages_present(unsigned long addr, unsigned long end)
2661{
2662 int ret, len, write;
2663 struct vm_area_struct * vma;
2664
2665 vma = find_vma(current->mm, addr);
2666 if (!vma)
2667 return -1;
2668 write = (vma->vm_flags & VM_WRITE) != 0;
Eric Sesterhenn5bcb28b2006-03-26 18:30:52 +02002669 BUG_ON(addr >= end);
2670 BUG_ON(end > vma->vm_end);
Rolf Eike Beer68e116a2007-07-15 23:38:03 -07002671 len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672 ret = get_user_pages(current, current->mm, addr,
2673 len, write, 0, NULL, NULL);
2674 if (ret < 0)
2675 return ret;
2676 return ret == len ? 0 : -1;
2677}
2678
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679#if !defined(__HAVE_ARCH_GATE_AREA)
2680
2681#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07002682static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683
2684static int __init gate_vma_init(void)
2685{
2686 gate_vma.vm_mm = NULL;
2687 gate_vma.vm_start = FIXADDR_USER_START;
2688 gate_vma.vm_end = FIXADDR_USER_END;
Roland McGrathb6558c42007-01-26 00:56:47 -08002689 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
2690 gate_vma.vm_page_prot = __P101;
Roland McGrathf47aef52007-01-26 00:56:49 -08002691 /*
2692 * Make sure the vDSO gets into every core dump.
2693 * Dumping its contents makes post-mortem fully interpretable later
2694 * without matching up the same kernel and hardware config to see
2695 * what PC values meant.
2696 */
2697 gate_vma.vm_flags |= VM_ALWAYSDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698 return 0;
2699}
2700__initcall(gate_vma_init);
2701#endif
2702
2703struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2704{
2705#ifdef AT_SYSINFO_EHDR
2706 return &gate_vma;
2707#else
2708 return NULL;
2709#endif
2710}
2711
2712int in_gate_area_no_task(unsigned long addr)
2713{
2714#ifdef AT_SYSINFO_EHDR
2715 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2716 return 1;
2717#endif
2718 return 0;
2719}
2720
2721#endif /* __HAVE_ARCH_GATE_AREA */
David Howells0ec76a12006-09-27 01:50:15 -07002722
2723/*
2724 * Access another process' address space.
2725 * Source/target buffer must be kernel space,
2726 * Do not walk the page table directly, use get_user_pages
2727 */
2728int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2729{
2730 struct mm_struct *mm;
2731 struct vm_area_struct *vma;
2732 struct page *page;
2733 void *old_buf = buf;
2734
2735 mm = get_task_mm(tsk);
2736 if (!mm)
2737 return 0;
2738
2739 down_read(&mm->mmap_sem);
Simon Arlott183ff222007-10-20 01:27:18 +02002740 /* ignore errors, just check how much was successfully transferred */
David Howells0ec76a12006-09-27 01:50:15 -07002741 while (len) {
2742 int bytes, ret, offset;
2743 void *maddr;
2744
2745 ret = get_user_pages(tsk, mm, addr, 1,
2746 write, 1, &page, &vma);
2747 if (ret <= 0)
2748 break;
2749
2750 bytes = len;
2751 offset = addr & (PAGE_SIZE-1);
2752 if (bytes > PAGE_SIZE-offset)
2753 bytes = PAGE_SIZE-offset;
2754
2755 maddr = kmap(page);
2756 if (write) {
2757 copy_to_user_page(vma, page, addr,
2758 maddr + offset, buf, bytes);
2759 set_page_dirty_lock(page);
2760 } else {
2761 copy_from_user_page(vma, page, addr,
2762 buf, maddr + offset, bytes);
2763 }
2764 kunmap(page);
2765 page_cache_release(page);
2766 len -= bytes;
2767 buf += bytes;
2768 addr += bytes;
2769 }
2770 up_read(&mm->mmap_sem);
2771 mmput(mm);
2772
2773 return buf - old_buf;
2774}
Andi Kleen03252912008-01-30 13:33:18 +01002775
2776/*
2777 * Print the name of a VMA.
2778 */
2779void print_vma_addr(char *prefix, unsigned long ip)
2780{
2781 struct mm_struct *mm = current->mm;
2782 struct vm_area_struct *vma;
2783
Ingo Molnare8bff742008-02-13 20:21:06 +01002784 /*
2785 * Do not print if we are in atomic
2786 * contexts (in exception stacks, etc.):
2787 */
2788 if (preempt_count())
2789 return;
2790
Andi Kleen03252912008-01-30 13:33:18 +01002791 down_read(&mm->mmap_sem);
2792 vma = find_vma(mm, ip);
2793 if (vma && vma->vm_file) {
2794 struct file *f = vma->vm_file;
2795 char *buf = (char *)__get_free_page(GFP_KERNEL);
2796 if (buf) {
2797 char *p, *s;
2798
Jan Blunckcf28b482008-02-14 19:38:44 -08002799 p = d_path(&f->f_path, buf, PAGE_SIZE);
Andi Kleen03252912008-01-30 13:33:18 +01002800 if (IS_ERR(p))
2801 p = "?";
2802 s = strrchr(p, '/');
2803 if (s)
2804 p = s+1;
2805 printk("%s%s[%lx+%lx]", prefix, p,
2806 vma->vm_start,
2807 vma->vm_end - vma->vm_start);
2808 free_page((unsigned long)buf);
2809 }
2810 }
2811 up_read(&current->mm->mmap_sem);
2812}