blob: ba86b436b85fadc015bd86138519c4e80038bf59 [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>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070054#include <linux/mmu_notifier.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
56#include <asm/pgalloc.h>
57#include <asm/uaccess.h>
58#include <asm/tlb.h>
59#include <asm/tlbflush.h>
60#include <asm/pgtable.h>
61
62#include <linux/swapops.h>
63#include <linux/elf.h>
64
Jan Beulich42b77722008-07-23 21:27:10 -070065#include "internal.h"
66
Nick Pigginb291f002008-10-18 20:26:44 -070067#include "internal.h"
68
Andy Whitcroftd41dee32005-06-23 00:07:54 -070069#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -070070/* use the per-pgdat data instead for discontigmem - mbligh */
71unsigned long max_mapnr;
72struct page *mem_map;
73
74EXPORT_SYMBOL(max_mapnr);
75EXPORT_SYMBOL(mem_map);
76#endif
77
78unsigned long num_physpages;
79/*
80 * A number of key systems in x86 including ioremap() rely on the assumption
81 * that high_memory defines the upper bound on direct map memory, then end
82 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
83 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
84 * and ZONE_HIGHMEM.
85 */
86void * high_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -070087
88EXPORT_SYMBOL(num_physpages);
89EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Ingo Molnar32a93232008-02-06 22:39:44 +010091/*
92 * Randomize the address space (stacks, mmaps, brk, etc.).
93 *
94 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
95 * as ancient (libc5 based) binaries can segfault. )
96 */
97int randomize_va_space __read_mostly =
98#ifdef CONFIG_COMPAT_BRK
99 1;
100#else
101 2;
102#endif
Andi Kleena62eaf12006-02-16 23:41:58 +0100103
104static int __init disable_randmaps(char *s)
105{
106 randomize_va_space = 0;
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800107 return 1;
Andi Kleena62eaf12006-02-16 23:41:58 +0100108}
109__setup("norandmaps", disable_randmaps);
110
111
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112/*
113 * If a p?d_bad entry is found while walking page tables, report
114 * the error, before resetting entry to p?d_none. Usually (but
115 * very seldom) called out from the p?d_none_or_clear_bad macros.
116 */
117
118void pgd_clear_bad(pgd_t *pgd)
119{
120 pgd_ERROR(*pgd);
121 pgd_clear(pgd);
122}
123
124void pud_clear_bad(pud_t *pud)
125{
126 pud_ERROR(*pud);
127 pud_clear(pud);
128}
129
130void pmd_clear_bad(pmd_t *pmd)
131{
132 pmd_ERROR(*pmd);
133 pmd_clear(pmd);
134}
135
136/*
137 * Note: this doesn't free the actual pages themselves. That
138 * has been handled earlier when unmapping all the memory regions.
139 */
Hugh Dickinse0da3822005-04-19 13:29:15 -0700140static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800142 pgtable_t token = pmd_pgtable(*pmd);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700143 pmd_clear(pmd);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800144 pte_free_tlb(tlb, token);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700145 tlb->mm->nr_ptes--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146}
147
Hugh Dickinse0da3822005-04-19 13:29:15 -0700148static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
149 unsigned long addr, unsigned long end,
150 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151{
152 pmd_t *pmd;
153 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700154 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155
Hugh Dickinse0da3822005-04-19 13:29:15 -0700156 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 do {
159 next = pmd_addr_end(addr, end);
160 if (pmd_none_or_clear_bad(pmd))
161 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700162 free_pte_range(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 } while (pmd++, addr = next, addr != end);
164
Hugh Dickinse0da3822005-04-19 13:29:15 -0700165 start &= PUD_MASK;
166 if (start < floor)
167 return;
168 if (ceiling) {
169 ceiling &= PUD_MASK;
170 if (!ceiling)
171 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700173 if (end - 1 > ceiling - 1)
174 return;
175
176 pmd = pmd_offset(pud, start);
177 pud_clear(pud);
178 pmd_free_tlb(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179}
180
Hugh Dickinse0da3822005-04-19 13:29:15 -0700181static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
182 unsigned long addr, unsigned long end,
183 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184{
185 pud_t *pud;
186 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700187 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
Hugh Dickinse0da3822005-04-19 13:29:15 -0700189 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 pud = pud_offset(pgd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 do {
192 next = pud_addr_end(addr, end);
193 if (pud_none_or_clear_bad(pud))
194 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700195 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 } while (pud++, addr = next, addr != end);
197
Hugh Dickinse0da3822005-04-19 13:29:15 -0700198 start &= PGDIR_MASK;
199 if (start < floor)
200 return;
201 if (ceiling) {
202 ceiling &= PGDIR_MASK;
203 if (!ceiling)
204 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700206 if (end - 1 > ceiling - 1)
207 return;
208
209 pud = pud_offset(pgd, start);
210 pgd_clear(pgd);
211 pud_free_tlb(tlb, pud);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212}
213
214/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700215 * This function frees user-level page tables of a process.
216 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 * Must be called with pagetable lock held.
218 */
Jan Beulich42b77722008-07-23 21:27:10 -0700219void free_pgd_range(struct mmu_gather *tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700220 unsigned long addr, unsigned long end,
221 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222{
223 pgd_t *pgd;
224 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700225 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226
Hugh Dickinse0da3822005-04-19 13:29:15 -0700227 /*
228 * The next few lines have given us lots of grief...
229 *
230 * Why are we testing PMD* at this top level? Because often
231 * there will be no work to do at all, and we'd prefer not to
232 * go all the way down to the bottom just to discover that.
233 *
234 * Why all these "- 1"s? Because 0 represents both the bottom
235 * of the address space and the top of it (using -1 for the
236 * top wouldn't help much: the masks would do the wrong thing).
237 * The rule is that addr 0 and floor 0 refer to the bottom of
238 * the address space, but end 0 and ceiling 0 refer to the top
239 * Comparisons need to use "end - 1" and "ceiling - 1" (though
240 * that end 0 case should be mythical).
241 *
242 * Wherever addr is brought up or ceiling brought down, we must
243 * be careful to reject "the opposite 0" before it confuses the
244 * subsequent tests. But what about where end is brought down
245 * by PMD_SIZE below? no, end can't go down to 0 there.
246 *
247 * Whereas we round start (addr) and ceiling down, by different
248 * masks at different levels, in order to test whether a table
249 * now has no other vmas using it, so can be freed, we don't
250 * bother to round floor or end up - the tests don't need that.
251 */
252
253 addr &= PMD_MASK;
254 if (addr < floor) {
255 addr += PMD_SIZE;
256 if (!addr)
257 return;
258 }
259 if (ceiling) {
260 ceiling &= PMD_MASK;
261 if (!ceiling)
262 return;
263 }
264 if (end - 1 > ceiling - 1)
265 end -= PMD_SIZE;
266 if (addr > end - 1)
267 return;
268
269 start = addr;
Jan Beulich42b77722008-07-23 21:27:10 -0700270 pgd = pgd_offset(tlb->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 do {
272 next = pgd_addr_end(addr, end);
273 if (pgd_none_or_clear_bad(pgd))
274 continue;
Jan Beulich42b77722008-07-23 21:27:10 -0700275 free_pud_range(tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700277}
278
Jan Beulich42b77722008-07-23 21:27:10 -0700279void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700280 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700281{
282 while (vma) {
283 struct vm_area_struct *next = vma->vm_next;
284 unsigned long addr = vma->vm_start;
285
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700286 /*
287 * Hide vma from rmap and vmtruncate before freeing pgtables
288 */
289 anon_vma_unlink(vma);
290 unlink_file_vma(vma);
291
David Gibson9da61ae2006-03-22 00:08:57 -0800292 if (is_vm_hugetlb_page(vma)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700293 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700294 floor, next? next->vm_start: ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700295 } else {
296 /*
297 * Optimization: gather nearby vmas into one call down
298 */
299 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
David Gibson48669202006-03-22 00:08:58 -0800300 && !is_vm_hugetlb_page(next)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700301 vma = next;
302 next = vma->vm_next;
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700303 anon_vma_unlink(vma);
304 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700305 }
306 free_pgd_range(tlb, addr, vma->vm_end,
307 floor, next? next->vm_start: ceiling);
308 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700309 vma = next;
310 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311}
312
Hugh Dickins1bb36302005-10-29 18:16:22 -0700313int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800315 pgtable_t new = pte_alloc_one(mm, address);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700316 if (!new)
317 return -ENOMEM;
318
Nick Piggin362a61a2008-05-14 06:37:36 +0200319 /*
320 * Ensure all pte setup (eg. pte page lock and page clearing) are
321 * visible before the pte is made visible to other CPUs by being
322 * put into page tables.
323 *
324 * The other side of the story is the pointer chasing in the page
325 * table walking code (when walking the page table without locking;
326 * ie. most of the time). Fortunately, these data accesses consist
327 * of a chain of data-dependent loads, meaning most CPUs (alpha
328 * being the notable exception) will already guarantee loads are
329 * seen in-order. See the alpha page table accessors for the
330 * smp_read_barrier_depends() barriers in page table walking code.
331 */
332 smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
333
Hugh Dickinsc74df322005-10-29 18:16:23 -0700334 spin_lock(&mm->page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800335 if (!pmd_present(*pmd)) { /* Has another populated it ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 mm->nr_ptes++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 pmd_populate(mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800338 new = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 }
Hugh Dickinsc74df322005-10-29 18:16:23 -0700340 spin_unlock(&mm->page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800341 if (new)
342 pte_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700343 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344}
345
Hugh Dickins1bb36302005-10-29 18:16:22 -0700346int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700348 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
349 if (!new)
350 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351
Nick Piggin362a61a2008-05-14 06:37:36 +0200352 smp_wmb(); /* See comment in __pte_alloc */
353
Hugh Dickins1bb36302005-10-29 18:16:22 -0700354 spin_lock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800355 if (!pmd_present(*pmd)) { /* Has another populated it ? */
Hugh Dickins1bb36302005-10-29 18:16:22 -0700356 pmd_populate_kernel(&init_mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800357 new = NULL;
358 }
Hugh Dickins1bb36302005-10-29 18:16:22 -0700359 spin_unlock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800360 if (new)
361 pte_free_kernel(&init_mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700362 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363}
364
Hugh Dickinsae859762005-10-29 18:16:05 -0700365static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
366{
367 if (file_rss)
368 add_mm_counter(mm, file_rss, file_rss);
369 if (anon_rss)
370 add_mm_counter(mm, anon_rss, anon_rss);
371}
372
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800374 * This function is called to print an error when a bad pte
375 * is found. For example, we might have a PFN-mapped pte in
376 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700377 *
378 * The calling function must still handle the error.
379 */
Adrian Bunk15f59ad2008-07-25 19:46:23 -0700380static void print_bad_pte(struct vm_area_struct *vma, pte_t pte,
381 unsigned long vaddr)
Nick Pigginb5810032005-10-29 18:16:12 -0700382{
383 printk(KERN_ERR "Bad pte = %08llx, process = %s, "
384 "vm_flags = %lx, vaddr = %lx\n",
385 (long long)pte_val(pte),
386 (vma->vm_mm == current->mm ? current->comm : "???"),
387 vma->vm_flags, vaddr);
388 dump_stack();
389}
390
Linus Torvalds67121172005-12-11 20:38:17 -0800391static inline int is_cow_mapping(unsigned int flags)
392{
393 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
394}
395
Nick Pigginb5810032005-10-29 18:16:12 -0700396/*
Nick Piggin7e675132008-04-28 02:13:00 -0700397 * vm_normal_page -- This function gets the "struct page" associated with a pte.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800398 *
Nick Piggin7e675132008-04-28 02:13:00 -0700399 * "Special" mappings do not wish to be associated with a "struct page" (either
400 * it doesn't exist, or it exists but they don't want to touch it). In this
401 * case, NULL is returned here. "Normal" mappings do have a struct page.
Jared Hulbertb379d792008-04-28 02:12:58 -0700402 *
Nick Piggin7e675132008-04-28 02:13:00 -0700403 * There are 2 broad cases. Firstly, an architecture may define a pte_special()
404 * pte bit, in which case this function is trivial. Secondly, an architecture
405 * may not have a spare pte bit, which requires a more complicated scheme,
406 * described below.
407 *
408 * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
409 * special mapping (even if there are underlying and valid "struct pages").
410 * COWed pages of a VM_PFNMAP are always normal.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800411 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700412 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
413 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
Nick Piggin7e675132008-04-28 02:13:00 -0700414 * set, and the vm_pgoff will point to the first PFN mapped: thus every special
415 * mapping will always honor the rule
Linus Torvalds6aab3412005-11-28 14:34:23 -0800416 *
417 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
418 *
Nick Piggin7e675132008-04-28 02:13:00 -0700419 * And for normal mappings this is false.
Jared Hulbertb379d792008-04-28 02:12:58 -0700420 *
Nick Piggin7e675132008-04-28 02:13:00 -0700421 * This restricts such mappings to be a linear translation from virtual address
422 * to pfn. To get around this restriction, we allow arbitrary mappings so long
423 * as the vma is not a COW mapping; in that case, we know that all ptes are
424 * special (because none can have been COWed).
Jared Hulbertb379d792008-04-28 02:12:58 -0700425 *
426 *
Nick Piggin7e675132008-04-28 02:13:00 -0700427 * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
428 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700429 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
430 * page" backing, however the difference is that _all_ pages with a struct
431 * page (that is, those where pfn_valid is true) are refcounted and considered
432 * normal pages by the VM. The disadvantage is that pages are refcounted
433 * (which can be slower and simply not an option for some PFNMAP users). The
434 * advantage is that we don't have to follow the strict linearity rule of
435 * PFNMAP mappings in order to support COWable mappings.
436 *
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800437 */
Nick Piggin7e675132008-04-28 02:13:00 -0700438#ifdef __HAVE_ARCH_PTE_SPECIAL
439# define HAVE_PTE_SPECIAL 1
440#else
441# define HAVE_PTE_SPECIAL 0
442#endif
443struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
444 pte_t pte)
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800445{
Nick Piggin7e675132008-04-28 02:13:00 -0700446 unsigned long pfn;
447
448 if (HAVE_PTE_SPECIAL) {
449 if (likely(!pte_special(pte))) {
450 VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
451 return pte_page(pte);
452 }
453 VM_BUG_ON(!(vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)));
454 return NULL;
455 }
456
457 /* !HAVE_PTE_SPECIAL case follows: */
458
459 pfn = pte_pfn(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800460
Jared Hulbertb379d792008-04-28 02:12:58 -0700461 if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
462 if (vma->vm_flags & VM_MIXEDMAP) {
463 if (!pfn_valid(pfn))
464 return NULL;
465 goto out;
466 } else {
Nick Piggin7e675132008-04-28 02:13:00 -0700467 unsigned long off;
468 off = (addr - vma->vm_start) >> PAGE_SHIFT;
Jared Hulbertb379d792008-04-28 02:12:58 -0700469 if (pfn == vma->vm_pgoff + off)
470 return NULL;
471 if (!is_cow_mapping(vma->vm_flags))
472 return NULL;
473 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800474 }
475
Nick Piggin7e675132008-04-28 02:13:00 -0700476 VM_BUG_ON(!pfn_valid(pfn));
Linus Torvalds6aab3412005-11-28 14:34:23 -0800477
478 /*
Nick Piggin7e675132008-04-28 02:13:00 -0700479 * NOTE! We still have PageReserved() pages in the page tables.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800480 *
Nick Piggin7e675132008-04-28 02:13:00 -0700481 * eg. VDSO mappings can cause them to exist.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800482 */
Jared Hulbertb379d792008-04-28 02:12:58 -0700483out:
Linus Torvalds6aab3412005-11-28 14:34:23 -0800484 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800485}
486
487/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 * copy one vm_area from one task to the other. Assumes the page tables
489 * already present in the new task to be cleared in the whole range
490 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 */
492
Hugh Dickins8c103762005-10-29 18:16:13 -0700493static inline void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700495 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700496 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497{
Nick Pigginb5810032005-10-29 18:16:12 -0700498 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 pte_t pte = *src_pte;
500 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501
502 /* pte contains position in swap or file, so copy. */
503 if (unlikely(!pte_present(pte))) {
504 if (!pte_file(pte)) {
Christoph Lameter06972122006-06-23 02:03:35 -0700505 swp_entry_t entry = pte_to_swp_entry(pte);
506
507 swap_duplicate(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 /* make sure dst_mm is on swapoff's mmlist. */
509 if (unlikely(list_empty(&dst_mm->mmlist))) {
510 spin_lock(&mmlist_lock);
Hugh Dickinsf412ac02005-10-29 18:16:41 -0700511 if (list_empty(&dst_mm->mmlist))
512 list_add(&dst_mm->mmlist,
513 &src_mm->mmlist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 spin_unlock(&mmlist_lock);
515 }
Christoph Lameter06972122006-06-23 02:03:35 -0700516 if (is_write_migration_entry(entry) &&
517 is_cow_mapping(vm_flags)) {
518 /*
519 * COW mappings require pages in both parent
520 * and child to be set to read.
521 */
522 make_migration_entry_read(&entry);
523 pte = swp_entry_to_pte(entry);
524 set_pte_at(src_mm, addr, src_pte, pte);
525 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700527 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 }
529
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 /*
531 * If it's a COW mapping, write protect it both
532 * in the parent and the child
533 */
Linus Torvalds67121172005-12-11 20:38:17 -0800534 if (is_cow_mapping(vm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 ptep_set_wrprotect(src_mm, addr, src_pte);
Zachary Amsden3dc90792006-09-30 23:29:30 -0700536 pte = pte_wrprotect(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 }
538
539 /*
540 * If it's a shared mapping, mark it clean in
541 * the child
542 */
543 if (vm_flags & VM_SHARED)
544 pte = pte_mkclean(pte);
545 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800546
547 page = vm_normal_page(vma, addr, pte);
548 if (page) {
549 get_page(page);
Nick Pigginc97a9e12007-05-16 22:11:21 -0700550 page_dup_rmap(page, vma, addr);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800551 rss[!!PageAnon(page)]++;
552 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700553
554out_set_pte:
555 set_pte_at(dst_mm, addr, dst_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556}
557
558static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
559 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
560 unsigned long addr, unsigned long end)
561{
562 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700563 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -0700564 int progress = 0;
Hugh Dickins8c103762005-10-29 18:16:13 -0700565 int rss[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
567again:
Hugh Dickinsae859762005-10-29 18:16:05 -0700568 rss[1] = rss[0] = 0;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700569 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 if (!dst_pte)
571 return -ENOMEM;
572 src_pte = pte_offset_map_nested(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700573 src_ptl = pte_lockptr(src_mm, src_pmd);
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700574 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700575 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 do {
578 /*
579 * We are holding two locks at this point - either of them
580 * could generate latencies in another task on another CPU.
581 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700582 if (progress >= 32) {
583 progress = 0;
584 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +0100585 spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700586 break;
587 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 if (pte_none(*src_pte)) {
589 progress++;
590 continue;
591 }
Hugh Dickins8c103762005-10-29 18:16:13 -0700592 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 progress += 8;
594 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700596 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -0700597 spin_unlock(src_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 pte_unmap_nested(src_pte - 1);
Hugh Dickinsae859762005-10-29 18:16:05 -0700599 add_mm_rss(dst_mm, rss[0], rss[1]);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700600 pte_unmap_unlock(dst_pte - 1, dst_ptl);
601 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 if (addr != end)
603 goto again;
604 return 0;
605}
606
607static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
608 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
609 unsigned long addr, unsigned long end)
610{
611 pmd_t *src_pmd, *dst_pmd;
612 unsigned long next;
613
614 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
615 if (!dst_pmd)
616 return -ENOMEM;
617 src_pmd = pmd_offset(src_pud, addr);
618 do {
619 next = pmd_addr_end(addr, end);
620 if (pmd_none_or_clear_bad(src_pmd))
621 continue;
622 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
623 vma, addr, next))
624 return -ENOMEM;
625 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
626 return 0;
627}
628
629static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
630 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
631 unsigned long addr, unsigned long end)
632{
633 pud_t *src_pud, *dst_pud;
634 unsigned long next;
635
636 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
637 if (!dst_pud)
638 return -ENOMEM;
639 src_pud = pud_offset(src_pgd, addr);
640 do {
641 next = pud_addr_end(addr, end);
642 if (pud_none_or_clear_bad(src_pud))
643 continue;
644 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
645 vma, addr, next))
646 return -ENOMEM;
647 } while (dst_pud++, src_pud++, addr = next, addr != end);
648 return 0;
649}
650
651int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
652 struct vm_area_struct *vma)
653{
654 pgd_t *src_pgd, *dst_pgd;
655 unsigned long next;
656 unsigned long addr = vma->vm_start;
657 unsigned long end = vma->vm_end;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700658 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
Nick Piggind9928952005-08-28 16:49:11 +1000660 /*
661 * Don't copy ptes where a page fault will fill them correctly.
662 * Fork becomes much lighter when there are big shared or private
663 * readonly mappings. The tradeoff is that copy_page_range is more
664 * efficient than faulting.
665 */
Linus Torvalds4d7672b2005-12-16 10:21:23 -0800666 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
Nick Piggind9928952005-08-28 16:49:11 +1000667 if (!vma->anon_vma)
668 return 0;
669 }
670
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 if (is_vm_hugetlb_page(vma))
672 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
673
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700674 /*
675 * We need to invalidate the secondary MMU mappings only when
676 * there could be a permission downgrade on the ptes of the
677 * parent mm. And a permission downgrade will only happen if
678 * is_cow_mapping() returns true.
679 */
680 if (is_cow_mapping(vma->vm_flags))
681 mmu_notifier_invalidate_range_start(src_mm, addr, end);
682
683 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 dst_pgd = pgd_offset(dst_mm, addr);
685 src_pgd = pgd_offset(src_mm, addr);
686 do {
687 next = pgd_addr_end(addr, end);
688 if (pgd_none_or_clear_bad(src_pgd))
689 continue;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700690 if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
691 vma, addr, next))) {
692 ret = -ENOMEM;
693 break;
694 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700696
697 if (is_cow_mapping(vma->vm_flags))
698 mmu_notifier_invalidate_range_end(src_mm,
699 vma->vm_start, end);
700 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701}
702
Robin Holt51c6f662005-11-13 16:06:42 -0800703static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700704 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800706 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707{
Nick Pigginb5810032005-10-29 18:16:12 -0700708 struct mm_struct *mm = tlb->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 pte_t *pte;
Hugh Dickins508034a2005-10-29 18:16:30 -0700710 spinlock_t *ptl;
Hugh Dickinsae859762005-10-29 18:16:05 -0700711 int file_rss = 0;
712 int anon_rss = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713
Hugh Dickins508034a2005-10-29 18:16:30 -0700714 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700715 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 do {
717 pte_t ptent = *pte;
Robin Holt51c6f662005-11-13 16:06:42 -0800718 if (pte_none(ptent)) {
719 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800721 }
Hugh Dickins6f5e6b92006-03-16 23:04:09 -0800722
723 (*zap_work) -= PAGE_SIZE;
724
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800726 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -0800727
Linus Torvalds6aab3412005-11-28 14:34:23 -0800728 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 if (unlikely(details) && page) {
730 /*
731 * unmap_shared_mapping_pages() wants to
732 * invalidate cache without truncating:
733 * unmap shared but keep private pages.
734 */
735 if (details->check_mapping &&
736 details->check_mapping != page->mapping)
737 continue;
738 /*
739 * Each page->index must be checked when
740 * invalidating or truncating nonlinear.
741 */
742 if (details->nonlinear_vma &&
743 (page->index < details->first_index ||
744 page->index > details->last_index))
745 continue;
746 }
Nick Pigginb5810032005-10-29 18:16:12 -0700747 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -0700748 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 tlb_remove_tlb_entry(tlb, pte, addr);
750 if (unlikely(!page))
751 continue;
752 if (unlikely(details) && details->nonlinear_vma
753 && linear_page_index(details->nonlinear_vma,
754 addr) != page->index)
Nick Pigginb5810032005-10-29 18:16:12 -0700755 set_pte_at(mm, addr, pte,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 pgoff_to_pte(page->index));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 if (PageAnon(page))
Hugh Dickins86d912f2005-10-29 18:16:14 -0700758 anon_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700759 else {
760 if (pte_dirty(ptent))
761 set_page_dirty(page);
762 if (pte_young(ptent))
Ken Chendaa88c82007-02-10 01:43:18 -0800763 SetPageReferenced(page);
Hugh Dickins86d912f2005-10-29 18:16:14 -0700764 file_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700765 }
Nick Piggin7de6b802006-12-22 01:09:33 -0800766 page_remove_rmap(page, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 tlb_remove_page(tlb, page);
768 continue;
769 }
770 /*
771 * If details->check_mapping, we leave swap entries;
772 * if details->nonlinear_vma, we leave file entries.
773 */
774 if (unlikely(details))
775 continue;
776 if (!pte_file(ptent))
777 free_swap_and_cache(pte_to_swp_entry(ptent));
Zachary Amsden9888a1c2006-09-30 23:29:31 -0700778 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
Robin Holt51c6f662005-11-13 16:06:42 -0800779 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
Hugh Dickinsae859762005-10-29 18:16:05 -0700780
Hugh Dickins86d912f2005-10-29 18:16:14 -0700781 add_mm_rss(mm, file_rss, anon_rss);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700782 arch_leave_lazy_mmu_mode();
Hugh Dickins508034a2005-10-29 18:16:30 -0700783 pte_unmap_unlock(pte - 1, ptl);
Robin Holt51c6f662005-11-13 16:06:42 -0800784
785 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786}
787
Robin Holt51c6f662005-11-13 16:06:42 -0800788static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700789 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800791 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792{
793 pmd_t *pmd;
794 unsigned long next;
795
796 pmd = pmd_offset(pud, addr);
797 do {
798 next = pmd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800799 if (pmd_none_or_clear_bad(pmd)) {
800 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800802 }
803 next = zap_pte_range(tlb, vma, pmd, addr, next,
804 zap_work, details);
805 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
806
807 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808}
809
Robin Holt51c6f662005-11-13 16:06:42 -0800810static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700811 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800813 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814{
815 pud_t *pud;
816 unsigned long next;
817
818 pud = pud_offset(pgd, addr);
819 do {
820 next = pud_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800821 if (pud_none_or_clear_bad(pud)) {
822 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800824 }
825 next = zap_pmd_range(tlb, vma, pud, addr, next,
826 zap_work, details);
827 } while (pud++, addr = next, (addr != end && *zap_work > 0));
828
829 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830}
831
Robin Holt51c6f662005-11-13 16:06:42 -0800832static unsigned long unmap_page_range(struct mmu_gather *tlb,
833 struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800835 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836{
837 pgd_t *pgd;
838 unsigned long next;
839
840 if (details && !details->check_mapping && !details->nonlinear_vma)
841 details = NULL;
842
843 BUG_ON(addr >= end);
844 tlb_start_vma(tlb, vma);
845 pgd = pgd_offset(vma->vm_mm, addr);
846 do {
847 next = pgd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800848 if (pgd_none_or_clear_bad(pgd)) {
849 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800851 }
852 next = zap_pud_range(tlb, vma, pgd, addr, next,
853 zap_work, details);
854 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 tlb_end_vma(tlb, vma);
Robin Holt51c6f662005-11-13 16:06:42 -0800856
857 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858}
859
860#ifdef CONFIG_PREEMPT
861# define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
862#else
863/* No preempt: go for improved straight-line efficiency */
864# define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
865#endif
866
867/**
868 * unmap_vmas - unmap a range of memory covered by a list of vma's
869 * @tlbp: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 * @vma: the starting vma
871 * @start_addr: virtual address at which to start unmapping
872 * @end_addr: virtual address at which to end unmapping
873 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
874 * @details: details of nonlinear truncation or shared cache invalidation
875 *
Hugh Dickinsee39b372005-04-19 13:29:15 -0700876 * Returns the end address of the unmapping (restart addr if interrupted).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700878 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700880 * We aim to not hold locks for too long (for scheduling latency reasons).
881 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 * return the ending mmu_gather to the caller.
883 *
884 * Only addresses between `start' and `end' will be unmapped.
885 *
886 * The VMA list must be sorted in ascending virtual address order.
887 *
888 * unmap_vmas() assumes that the caller will flush the whole unmapped address
889 * range after unmap_vmas() returns. So the only responsibility here is to
890 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
891 * drops the lock and schedules.
892 */
Hugh Dickins508034a2005-10-29 18:16:30 -0700893unsigned long unmap_vmas(struct mmu_gather **tlbp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 struct vm_area_struct *vma, unsigned long start_addr,
895 unsigned long end_addr, unsigned long *nr_accounted,
896 struct zap_details *details)
897{
Robin Holt51c6f662005-11-13 16:06:42 -0800898 long zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
900 int tlb_start_valid = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -0700901 unsigned long start = start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700903 int fullmm = (*tlbp)->fullmm;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700904 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700906 mmu_notifier_invalidate_range_start(mm, start_addr, end_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 unsigned long end;
909
910 start = max(vma->vm_start, start_addr);
911 if (start >= vma->vm_end)
912 continue;
913 end = min(vma->vm_end, end_addr);
914 if (end <= vma->vm_start)
915 continue;
916
917 if (vma->vm_flags & VM_ACCOUNT)
918 *nr_accounted += (end - start) >> PAGE_SHIFT;
919
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 while (start != end) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 if (!tlb_start_valid) {
922 tlb_start = start;
923 tlb_start_valid = 1;
924 }
925
Robin Holt51c6f662005-11-13 16:06:42 -0800926 if (unlikely(is_vm_hugetlb_page(vma))) {
Andi Kleena137e1c2008-07-23 21:27:43 -0700927 /*
928 * It is undesirable to test vma->vm_file as it
929 * should be non-null for valid hugetlb area.
930 * However, vm_file will be NULL in the error
931 * cleanup path of do_mmap_pgoff. When
932 * hugetlbfs ->mmap method fails,
933 * do_mmap_pgoff() nullifies vma->vm_file
934 * before calling this function to clean up.
935 * Since no pte has actually been setup, it is
936 * safe to do nothing in this case.
937 */
938 if (vma->vm_file) {
939 unmap_hugepage_range(vma, start, end, NULL);
940 zap_work -= (end - start) /
Andi Kleena5516432008-07-23 21:27:41 -0700941 pages_per_huge_page(hstate_vma(vma));
Andi Kleena137e1c2008-07-23 21:27:43 -0700942 }
943
Robin Holt51c6f662005-11-13 16:06:42 -0800944 start = end;
945 } else
946 start = unmap_page_range(*tlbp, vma,
947 start, end, &zap_work, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948
Robin Holt51c6f662005-11-13 16:06:42 -0800949 if (zap_work > 0) {
950 BUG_ON(start != end);
951 break;
952 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953
954 tlb_finish_mmu(*tlbp, tlb_start, start);
955
956 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +0100957 (i_mmap_lock && spin_needbreak(i_mmap_lock))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 if (i_mmap_lock) {
Hugh Dickins508034a2005-10-29 18:16:30 -0700959 *tlbp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 goto out;
961 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 }
964
Hugh Dickins508034a2005-10-29 18:16:30 -0700965 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 tlb_start_valid = 0;
Robin Holt51c6f662005-11-13 16:06:42 -0800967 zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 }
969 }
970out:
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700971 mmu_notifier_invalidate_range_end(mm, start_addr, end_addr);
Hugh Dickinsee39b372005-04-19 13:29:15 -0700972 return start; /* which is now the end (or restart) address */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973}
974
975/**
976 * zap_page_range - remove user pages in a given range
977 * @vma: vm_area_struct holding the applicable pages
978 * @address: starting address of pages to zap
979 * @size: number of bytes to zap
980 * @details: details of nonlinear truncation or shared cache invalidation
981 */
Hugh Dickinsee39b372005-04-19 13:29:15 -0700982unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 unsigned long size, struct zap_details *details)
984{
985 struct mm_struct *mm = vma->vm_mm;
986 struct mmu_gather *tlb;
987 unsigned long end = address + size;
988 unsigned long nr_accounted = 0;
989
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -0700992 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -0700993 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
994 if (tlb)
995 tlb_finish_mmu(tlb, address, end);
Hugh Dickinsee39b372005-04-19 13:29:15 -0700996 return end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997}
998
Jack Steinerc627f9c2008-07-29 22:33:53 -0700999/**
1000 * zap_vma_ptes - remove ptes mapping the vma
1001 * @vma: vm_area_struct holding ptes to be zapped
1002 * @address: starting address of pages to zap
1003 * @size: number of bytes to zap
1004 *
1005 * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1006 *
1007 * The entire address range must be fully contained within the vma.
1008 *
1009 * Returns 0 if successful.
1010 */
1011int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
1012 unsigned long size)
1013{
1014 if (address < vma->vm_start || address + size > vma->vm_end ||
1015 !(vma->vm_flags & VM_PFNMAP))
1016 return -1;
1017 zap_page_range(vma, address, size, NULL);
1018 return 0;
1019}
1020EXPORT_SYMBOL_GPL(zap_vma_ptes);
1021
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022/*
1023 * Do a quick page-table lookup for a single page.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 */
Linus Torvalds6aab3412005-11-28 14:34:23 -08001025struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001026 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027{
1028 pgd_t *pgd;
1029 pud_t *pud;
1030 pmd_t *pmd;
1031 pte_t *ptep, pte;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001032 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 struct page *page;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001034 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001036 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
1037 if (!IS_ERR(page)) {
1038 BUG_ON(flags & FOLL_GET);
1039 goto out;
1040 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001042 page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 pgd = pgd_offset(mm, address);
1044 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001045 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046
1047 pud = pud_offset(pgd, address);
Andi Kleenceb86872008-07-23 21:27:50 -07001048 if (pud_none(*pud))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001049 goto no_page_table;
Andi Kleenceb86872008-07-23 21:27:50 -07001050 if (pud_huge(*pud)) {
1051 BUG_ON(flags & FOLL_GET);
1052 page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE);
1053 goto out;
1054 }
1055 if (unlikely(pud_bad(*pud)))
1056 goto no_page_table;
1057
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 pmd = pmd_offset(pud, address);
Hugh Dickinsaeed5fc2008-05-06 20:49:23 +01001059 if (pmd_none(*pmd))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001060 goto no_page_table;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001061 if (pmd_huge(*pmd)) {
1062 BUG_ON(flags & FOLL_GET);
1063 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
1064 goto out;
1065 }
Hugh Dickinsaeed5fc2008-05-06 20:49:23 +01001066 if (unlikely(pmd_bad(*pmd)))
1067 goto no_page_table;
1068
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001069 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070
1071 pte = *ptep;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001072 if (!pte_present(pte))
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001073 goto no_page;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001074 if ((flags & FOLL_WRITE) && !pte_write(pte))
1075 goto unlock;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001076 page = vm_normal_page(vma, address, pte);
1077 if (unlikely(!page))
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001078 goto bad_page;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001079
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001080 if (flags & FOLL_GET)
1081 get_page(page);
1082 if (flags & FOLL_TOUCH) {
1083 if ((flags & FOLL_WRITE) &&
1084 !pte_dirty(pte) && !PageDirty(page))
1085 set_page_dirty(page);
1086 mark_page_accessed(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001088unlock:
1089 pte_unmap_unlock(ptep, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090out:
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001091 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001093bad_page:
1094 pte_unmap_unlock(ptep, ptl);
1095 return ERR_PTR(-EFAULT);
1096
1097no_page:
1098 pte_unmap_unlock(ptep, ptl);
1099 if (!pte_none(pte))
1100 return page;
1101 /* Fall through to ZERO_PAGE handling */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001102no_page_table:
1103 /*
1104 * When core dumping an enormous anonymous area that nobody
1105 * has touched so far, we don't want to allocate page tables.
1106 */
1107 if (flags & FOLL_ANON) {
Nick Piggin557ed1f2007-10-16 01:24:40 -07001108 page = ZERO_PAGE(0);
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001109 if (flags & FOLL_GET)
1110 get_page(page);
1111 BUG_ON(flags & FOLL_WRITE);
1112 }
1113 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114}
1115
Linus Torvalds672ca282008-06-23 11:21:37 -07001116/* Can we do the FOLL_ANON optimization? */
1117static inline int use_zero_page(struct vm_area_struct *vma)
1118{
1119 /*
1120 * We don't want to optimize FOLL_ANON for make_pages_present()
1121 * when it tries to page in a VM_LOCKED region. As to VM_SHARED,
1122 * we want to get the page from the page tables to make sure
1123 * that we serialize and update with any other user of that
1124 * mapping.
1125 */
1126 if (vma->vm_flags & (VM_LOCKED | VM_SHARED))
1127 return 0;
1128 /*
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001129 * And if we have a fault routine, it's not an anonymous region.
Linus Torvalds672ca282008-06-23 11:21:37 -07001130 */
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001131 return !vma->vm_ops || !vma->vm_ops->fault;
Linus Torvalds672ca282008-06-23 11:21:37 -07001132}
1133
Nick Pigginb291f002008-10-18 20:26:44 -07001134
1135
1136int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
1137 unsigned long start, int len, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 struct page **pages, struct vm_area_struct **vmas)
1139{
1140 int i;
Nick Pigginb291f002008-10-18 20:26:44 -07001141 unsigned int vm_flags = 0;
1142 int write = !!(flags & GUP_FLAGS_WRITE);
1143 int force = !!(flags & GUP_FLAGS_FORCE);
1144 int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
Jonathan Corbet900cf082008-02-11 16:17:33 -07001146 if (len <= 0)
1147 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 /*
1149 * Require read or write permissions.
1150 * If 'force' is set, we only require the "MAY" flags.
1151 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001152 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
1153 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 i = 0;
1155
1156 do {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001157 struct vm_area_struct *vma;
1158 unsigned int foll_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159
1160 vma = find_extend_vma(mm, start);
1161 if (!vma && in_gate_area(tsk, start)) {
1162 unsigned long pg = start & PAGE_MASK;
1163 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
1164 pgd_t *pgd;
1165 pud_t *pud;
1166 pmd_t *pmd;
1167 pte_t *pte;
Nick Pigginb291f002008-10-18 20:26:44 -07001168
1169 /* user gate pages are read-only */
1170 if (!ignore && write)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 return i ? : -EFAULT;
1172 if (pg > TASK_SIZE)
1173 pgd = pgd_offset_k(pg);
1174 else
1175 pgd = pgd_offset_gate(mm, pg);
1176 BUG_ON(pgd_none(*pgd));
1177 pud = pud_offset(pgd, pg);
1178 BUG_ON(pud_none(*pud));
1179 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001180 if (pmd_none(*pmd))
1181 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001183 if (pte_none(*pte)) {
1184 pte_unmap(pte);
1185 return i ? : -EFAULT;
1186 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 if (pages) {
Nick Pigginfa2a4552005-11-29 18:43:17 +11001188 struct page *page = vm_normal_page(gate_vma, start, *pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001189 pages[i] = page;
1190 if (page)
1191 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 }
1193 pte_unmap(pte);
1194 if (vmas)
1195 vmas[i] = gate_vma;
1196 i++;
1197 start += PAGE_SIZE;
1198 len--;
1199 continue;
1200 }
1201
Nick Pigginb291f002008-10-18 20:26:44 -07001202 if (!vma ||
1203 (vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
1204 (!ignore && !(vm_flags & vma->vm_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 return i ? : -EFAULT;
1206
1207 if (is_vm_hugetlb_page(vma)) {
1208 i = follow_hugetlb_page(mm, vma, pages, vmas,
Adam Litke5b23dbe2007-11-14 16:59:33 -08001209 &start, &len, i, write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 continue;
1211 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001212
1213 foll_flags = FOLL_TOUCH;
1214 if (pages)
1215 foll_flags |= FOLL_GET;
Linus Torvalds672ca282008-06-23 11:21:37 -07001216 if (!write && use_zero_page(vma))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001217 foll_flags |= FOLL_ANON;
1218
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 do {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001220 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221
Ethan Solomita462e00c2007-07-15 23:38:16 -07001222 /*
1223 * If tsk is ooming, cut off its access to large memory
1224 * allocations. It has a pending SIGKILL, but it can't
1225 * be processed until returning to user space.
1226 */
1227 if (unlikely(test_tsk_thread_flag(tsk, TIF_MEMDIE)))
Oleg Nesterov7a36a752008-07-04 09:59:28 -07001228 return i ? i : -ENOMEM;
Ethan Solomita462e00c2007-07-15 23:38:16 -07001229
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001230 if (write)
1231 foll_flags |= FOLL_WRITE;
1232
1233 cond_resched();
Linus Torvalds6aab3412005-11-28 14:34:23 -08001234 while (!(page = follow_page(vma, start, foll_flags))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001235 int ret;
Nick Piggin83c54072007-07-19 01:47:05 -07001236 ret = handle_mm_fault(mm, vma, start,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001237 foll_flags & FOLL_WRITE);
Nick Piggin83c54072007-07-19 01:47:05 -07001238 if (ret & VM_FAULT_ERROR) {
1239 if (ret & VM_FAULT_OOM)
1240 return i ? i : -ENOMEM;
1241 else if (ret & VM_FAULT_SIGBUS)
1242 return i ? i : -EFAULT;
1243 BUG();
1244 }
1245 if (ret & VM_FAULT_MAJOR)
1246 tsk->maj_flt++;
1247 else
1248 tsk->min_flt++;
1249
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001250 /*
Nick Piggin83c54072007-07-19 01:47:05 -07001251 * The VM_FAULT_WRITE bit tells us that
1252 * do_wp_page has broken COW when necessary,
1253 * even if maybe_mkwrite decided not to set
1254 * pte_write. We can thus safely do subsequent
1255 * page lookups as if they were reads.
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001256 */
1257 if (ret & VM_FAULT_WRITE)
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001258 foll_flags &= ~FOLL_WRITE;
Nick Piggin83c54072007-07-19 01:47:05 -07001259
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07001260 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 }
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001262 if (IS_ERR(page))
1263 return i ? i : PTR_ERR(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001265 pages[i] = page;
James Bottomley03beb072006-03-26 01:36:57 -08001266
Russell Kinga6f36be2006-12-30 22:24:19 +00001267 flush_anon_page(vma, page, start);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001268 flush_dcache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 }
1270 if (vmas)
1271 vmas[i] = vma;
1272 i++;
1273 start += PAGE_SIZE;
1274 len--;
Hugh Dickins08ef4722005-06-21 17:15:10 -07001275 } while (len && start < vma->vm_end);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001276 } while (len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 return i;
1278}
Nick Pigginb291f002008-10-18 20:26:44 -07001279
1280int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
1281 unsigned long start, int len, int write, int force,
1282 struct page **pages, struct vm_area_struct **vmas)
1283{
1284 int flags = 0;
1285
1286 if (write)
1287 flags |= GUP_FLAGS_WRITE;
1288 if (force)
1289 flags |= GUP_FLAGS_FORCE;
1290
1291 return __get_user_pages(tsk, mm,
1292 start, len, flags,
1293 pages, vmas);
1294}
1295
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296EXPORT_SYMBOL(get_user_pages);
1297
Harvey Harrison920c7a52008-02-04 22:29:26 -08001298pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr,
1299 spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001300{
1301 pgd_t * pgd = pgd_offset(mm, addr);
1302 pud_t * pud = pud_alloc(mm, pgd, addr);
1303 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001304 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001305 if (pmd)
1306 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1307 }
1308 return NULL;
1309}
1310
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001312 * This is the old fallback for page remapping.
1313 *
1314 * For historical reasons, it only allows reserved pages. Only
1315 * old drivers should use this, and they needed to mark their
1316 * pages reserved for the old functions anyway.
1317 */
Nick Piggin423bad62008-04-28 02:13:01 -07001318static int insert_page(struct vm_area_struct *vma, unsigned long addr,
1319 struct page *page, pgprot_t prot)
Linus Torvalds238f58d2005-11-29 13:01:56 -08001320{
Nick Piggin423bad62008-04-28 02:13:01 -07001321 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001322 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001323 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001324 spinlock_t *ptl;
1325
Balbir Singhe1a1cd52008-02-07 00:14:02 -08001326 retval = mem_cgroup_charge(page, mm, GFP_KERNEL);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001327 if (retval)
1328 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001329
1330 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001331 if (PageAnon(page))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001332 goto out_uncharge;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001333 retval = -ENOMEM;
1334 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001335 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001336 if (!pte)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001337 goto out_uncharge;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001338 retval = -EBUSY;
1339 if (!pte_none(*pte))
1340 goto out_unlock;
1341
1342 /* Ok, finally just insert the thing.. */
1343 get_page(page);
1344 inc_mm_counter(mm, file_rss);
1345 page_add_file_rmap(page);
1346 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1347
1348 retval = 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001349 pte_unmap_unlock(pte, ptl);
1350 return retval;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001351out_unlock:
1352 pte_unmap_unlock(pte, ptl);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001353out_uncharge:
1354 mem_cgroup_uncharge_page(page);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001355out:
1356 return retval;
1357}
1358
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001359/**
1360 * vm_insert_page - insert single page into user vma
1361 * @vma: user vma to map to
1362 * @addr: target user address of this page
1363 * @page: source kernel page
1364 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001365 * This allows drivers to insert individual pages they've allocated
1366 * into a user vma.
1367 *
1368 * The page has to be a nice clean _individual_ kernel allocation.
1369 * If you allocate a compound page, you need to have marked it as
1370 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001371 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001372 *
1373 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1374 * took an arbitrary page protection parameter. This doesn't allow
1375 * that. Your vma protection will have to be set up correctly, which
1376 * means that if you want a shared writable mapping, you'd better
1377 * ask for a shared writable mapping!
1378 *
1379 * The page does not need to be reserved.
1380 */
Nick Piggin423bad62008-04-28 02:13:01 -07001381int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
1382 struct page *page)
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001383{
1384 if (addr < vma->vm_start || addr >= vma->vm_end)
1385 return -EFAULT;
1386 if (!page_count(page))
1387 return -EINVAL;
Linus Torvalds4d7672b2005-12-16 10:21:23 -08001388 vma->vm_flags |= VM_INSERTPAGE;
Nick Piggin423bad62008-04-28 02:13:01 -07001389 return insert_page(vma, addr, page, vma->vm_page_prot);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001390}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001391EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001392
Nick Piggin423bad62008-04-28 02:13:01 -07001393static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1394 unsigned long pfn, pgprot_t prot)
1395{
1396 struct mm_struct *mm = vma->vm_mm;
1397 int retval;
1398 pte_t *pte, entry;
1399 spinlock_t *ptl;
1400
1401 retval = -ENOMEM;
1402 pte = get_locked_pte(mm, addr, &ptl);
1403 if (!pte)
1404 goto out;
1405 retval = -EBUSY;
1406 if (!pte_none(*pte))
1407 goto out_unlock;
1408
1409 /* Ok, finally just insert the thing.. */
1410 entry = pte_mkspecial(pfn_pte(pfn, prot));
1411 set_pte_at(mm, addr, pte, entry);
1412 update_mmu_cache(vma, addr, entry); /* XXX: why not for insert_page? */
1413
1414 retval = 0;
1415out_unlock:
1416 pte_unmap_unlock(pte, ptl);
1417out:
1418 return retval;
1419}
1420
Nick Piggine0dc0d82007-02-12 00:51:36 -08001421/**
1422 * vm_insert_pfn - insert single pfn into user vma
1423 * @vma: user vma to map to
1424 * @addr: target user address of this page
1425 * @pfn: source kernel pfn
1426 *
1427 * Similar to vm_inert_page, this allows drivers to insert individual pages
1428 * they've allocated into a user vma. Same comments apply.
1429 *
1430 * This function should only be called from a vm_ops->fault handler, and
1431 * in that case the handler should return NULL.
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001432 *
1433 * vma cannot be a COW mapping.
1434 *
1435 * As this is called only for pages that do not currently exist, we
1436 * do not need to flush old virtual caches or the TLB.
Nick Piggine0dc0d82007-02-12 00:51:36 -08001437 */
1438int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
Nick Piggin423bad62008-04-28 02:13:01 -07001439 unsigned long pfn)
Nick Piggine0dc0d82007-02-12 00:51:36 -08001440{
Nick Piggin7e675132008-04-28 02:13:00 -07001441 /*
1442 * Technically, architectures with pte_special can avoid all these
1443 * restrictions (same for remap_pfn_range). However we would like
1444 * consistency in testing and feature parity among all, so we should
1445 * try to keep these invariants in place for everybody.
1446 */
Jared Hulbertb379d792008-04-28 02:12:58 -07001447 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
1448 BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
1449 (VM_PFNMAP|VM_MIXEDMAP));
1450 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1451 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
Nick Piggine0dc0d82007-02-12 00:51:36 -08001452
Nick Piggin423bad62008-04-28 02:13:01 -07001453 if (addr < vma->vm_start || addr >= vma->vm_end)
1454 return -EFAULT;
1455 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
Nick Piggine0dc0d82007-02-12 00:51:36 -08001456}
1457EXPORT_SYMBOL(vm_insert_pfn);
1458
Nick Piggin423bad62008-04-28 02:13:01 -07001459int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
1460 unsigned long pfn)
1461{
1462 BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
1463
1464 if (addr < vma->vm_start || addr >= vma->vm_end)
1465 return -EFAULT;
1466
1467 /*
1468 * If we don't have pte special, then we have to use the pfn_valid()
1469 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1470 * refcount the page if pfn_valid is true (hence insert_page rather
1471 * than insert_pfn).
1472 */
1473 if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) {
1474 struct page *page;
1475
1476 page = pfn_to_page(pfn);
1477 return insert_page(vma, addr, page, vma->vm_page_prot);
1478 }
1479 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
1480}
1481EXPORT_SYMBOL(vm_insert_mixed);
1482
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001483/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 * maps a range of physical memory into the requested pages. the old
1485 * mappings are removed. any references to nonexistent pages results
1486 * in null mappings (currently treated as "copy-on-access")
1487 */
1488static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1489 unsigned long addr, unsigned long end,
1490 unsigned long pfn, pgprot_t prot)
1491{
1492 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001493 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494
Hugh Dickinsc74df322005-10-29 18:16:23 -07001495 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 if (!pte)
1497 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001498 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 do {
1500 BUG_ON(!pte_none(*pte));
Nick Piggin7e675132008-04-28 02:13:00 -07001501 set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 pfn++;
1503 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001504 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001505 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 return 0;
1507}
1508
1509static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1510 unsigned long addr, unsigned long end,
1511 unsigned long pfn, pgprot_t prot)
1512{
1513 pmd_t *pmd;
1514 unsigned long next;
1515
1516 pfn -= addr >> PAGE_SHIFT;
1517 pmd = pmd_alloc(mm, pud, addr);
1518 if (!pmd)
1519 return -ENOMEM;
1520 do {
1521 next = pmd_addr_end(addr, end);
1522 if (remap_pte_range(mm, pmd, addr, next,
1523 pfn + (addr >> PAGE_SHIFT), prot))
1524 return -ENOMEM;
1525 } while (pmd++, addr = next, addr != end);
1526 return 0;
1527}
1528
1529static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1530 unsigned long addr, unsigned long end,
1531 unsigned long pfn, pgprot_t prot)
1532{
1533 pud_t *pud;
1534 unsigned long next;
1535
1536 pfn -= addr >> PAGE_SHIFT;
1537 pud = pud_alloc(mm, pgd, addr);
1538 if (!pud)
1539 return -ENOMEM;
1540 do {
1541 next = pud_addr_end(addr, end);
1542 if (remap_pmd_range(mm, pud, addr, next,
1543 pfn + (addr >> PAGE_SHIFT), prot))
1544 return -ENOMEM;
1545 } while (pud++, addr = next, addr != end);
1546 return 0;
1547}
1548
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001549/**
1550 * remap_pfn_range - remap kernel memory to userspace
1551 * @vma: user vma to map to
1552 * @addr: target user address to start at
1553 * @pfn: physical address of kernel memory
1554 * @size: size of map area
1555 * @prot: page protection flags for this mapping
1556 *
1557 * Note: this is only safe if the mm semaphore is held when called.
1558 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1560 unsigned long pfn, unsigned long size, pgprot_t prot)
1561{
1562 pgd_t *pgd;
1563 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001564 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 struct mm_struct *mm = vma->vm_mm;
1566 int err;
1567
1568 /*
1569 * Physically remapped pages are special. Tell the
1570 * rest of the world about it:
1571 * VM_IO tells people not to look at these pages
1572 * (accesses can have side effects).
Hugh Dickins0b14c172005-11-21 21:32:15 -08001573 * VM_RESERVED is specified all over the place, because
1574 * in 2.4 it kept swapout's vma scan off this vma; but
1575 * in 2.6 the LRU scan won't even find its pages, so this
1576 * flag means no more than count its pages in reserved_vm,
1577 * and omit it from core dump, even when VM_IO turned off.
Linus Torvalds6aab3412005-11-28 14:34:23 -08001578 * VM_PFNMAP tells the core MM that the base pages are just
1579 * raw PFN mappings, and do not have a "struct page" associated
1580 * with them.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001581 *
1582 * There's a horrible special case to handle copy-on-write
1583 * behaviour that some programs depend on. We mark the "original"
1584 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 */
Linus Torvalds67121172005-12-11 20:38:17 -08001586 if (is_cow_mapping(vma->vm_flags)) {
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001587 if (addr != vma->vm_start || end != vma->vm_end)
Linus Torvalds7fc7e2e2005-12-11 19:57:52 -08001588 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001589 vma->vm_pgoff = pfn;
1590 }
1591
Linus Torvalds6aab3412005-11-28 14:34:23 -08001592 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593
1594 BUG_ON(addr >= end);
1595 pfn -= addr >> PAGE_SHIFT;
1596 pgd = pgd_offset(mm, addr);
1597 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 do {
1599 next = pgd_addr_end(addr, end);
1600 err = remap_pud_range(mm, pgd, addr, next,
1601 pfn + (addr >> PAGE_SHIFT), prot);
1602 if (err)
1603 break;
1604 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 return err;
1606}
1607EXPORT_SYMBOL(remap_pfn_range);
1608
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001609static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1610 unsigned long addr, unsigned long end,
1611 pte_fn_t fn, void *data)
1612{
1613 pte_t *pte;
1614 int err;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001615 pgtable_t token;
Borislav Petkov94909912007-05-06 14:49:17 -07001616 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001617
1618 pte = (mm == &init_mm) ?
1619 pte_alloc_kernel(pmd, addr) :
1620 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1621 if (!pte)
1622 return -ENOMEM;
1623
1624 BUG_ON(pmd_huge(*pmd));
1625
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001626 token = pmd_pgtable(*pmd);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001627
1628 do {
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001629 err = fn(pte, token, addr, data);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001630 if (err)
1631 break;
1632 } while (pte++, addr += PAGE_SIZE, addr != end);
1633
1634 if (mm != &init_mm)
1635 pte_unmap_unlock(pte-1, ptl);
1636 return err;
1637}
1638
1639static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1640 unsigned long addr, unsigned long end,
1641 pte_fn_t fn, void *data)
1642{
1643 pmd_t *pmd;
1644 unsigned long next;
1645 int err;
1646
Andi Kleenceb86872008-07-23 21:27:50 -07001647 BUG_ON(pud_huge(*pud));
1648
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001649 pmd = pmd_alloc(mm, pud, addr);
1650 if (!pmd)
1651 return -ENOMEM;
1652 do {
1653 next = pmd_addr_end(addr, end);
1654 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1655 if (err)
1656 break;
1657 } while (pmd++, addr = next, addr != end);
1658 return err;
1659}
1660
1661static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1662 unsigned long addr, unsigned long end,
1663 pte_fn_t fn, void *data)
1664{
1665 pud_t *pud;
1666 unsigned long next;
1667 int err;
1668
1669 pud = pud_alloc(mm, pgd, addr);
1670 if (!pud)
1671 return -ENOMEM;
1672 do {
1673 next = pud_addr_end(addr, end);
1674 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1675 if (err)
1676 break;
1677 } while (pud++, addr = next, addr != end);
1678 return err;
1679}
1680
1681/*
1682 * Scan a region of virtual memory, filling in page tables as necessary
1683 * and calling a provided function on each leaf page table.
1684 */
1685int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1686 unsigned long size, pte_fn_t fn, void *data)
1687{
1688 pgd_t *pgd;
1689 unsigned long next;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001690 unsigned long start = addr, end = addr + size;
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001691 int err;
1692
1693 BUG_ON(addr >= end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001694 mmu_notifier_invalidate_range_start(mm, start, end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001695 pgd = pgd_offset(mm, addr);
1696 do {
1697 next = pgd_addr_end(addr, end);
1698 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1699 if (err)
1700 break;
1701 } while (pgd++, addr = next, addr != end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001702 mmu_notifier_invalidate_range_end(mm, start, end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001703 return err;
1704}
1705EXPORT_SYMBOL_GPL(apply_to_page_range);
1706
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001708 * handle_pte_fault chooses page fault handler according to an entry
1709 * which was read non-atomically. Before making any commitment, on
1710 * those architectures or configurations (e.g. i386 with PAE) which
1711 * might give a mix of unmatched parts, do_swap_page and do_file_page
1712 * must check under lock before unmapping the pte and proceeding
1713 * (but do_wp_page is only called after already making such a check;
1714 * and do_anonymous_page and do_no_page can safely check later on).
1715 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001716static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001717 pte_t *page_table, pte_t orig_pte)
1718{
1719 int same = 1;
1720#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1721 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001722 spinlock_t *ptl = pte_lockptr(mm, pmd);
1723 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001724 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001725 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001726 }
1727#endif
1728 pte_unmap(page_table);
1729 return same;
1730}
1731
1732/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1734 * servicing faults for write access. In the normal case, do always want
1735 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1736 * that do not have writing enabled, when used by access_process_vm.
1737 */
1738static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1739{
1740 if (likely(vma->vm_flags & VM_WRITE))
1741 pte = pte_mkwrite(pte);
1742 return pte;
1743}
1744
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001745static 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 -08001746{
1747 /*
1748 * If the source page was a PFN mapping, we don't have
1749 * a "struct page" for it. We do a best-effort copy by
1750 * just copying from the original user address. If that
1751 * fails, we just zero-fill it. Live with it.
1752 */
1753 if (unlikely(!src)) {
1754 void *kaddr = kmap_atomic(dst, KM_USER0);
Linus Torvalds5d2a2dbbc2005-11-29 14:07:55 -08001755 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1756
1757 /*
1758 * This really shouldn't fail, because the page is there
1759 * in the page tables. But it might just be unreadable,
1760 * in which case we just give up and fill the result with
1761 * zeroes.
1762 */
1763 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Linus Torvalds6aab3412005-11-28 14:34:23 -08001764 memset(kaddr, 0, PAGE_SIZE);
1765 kunmap_atomic(kaddr, KM_USER0);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07001766 flush_dcache_page(dst);
Nick Piggin0ed361d2008-02-04 22:29:34 -08001767 } else
1768 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001769}
1770
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 * This routine handles present pages, when users try to write
1773 * to a shared page. It is done by copying the page to a new address
1774 * and decrementing the shared-page counter for the old page.
1775 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 * Note that this routine assumes that the protection checks have been
1777 * done by the caller (the low-level page fault routine in most cases).
1778 * Thus we can safely just mark it writable once we've done any necessary
1779 * COW.
1780 *
1781 * We also mark the page dirty at this point even though the page will
1782 * change only once the write actually happens. This avoids a few races,
1783 * and potentially makes it more efficient.
1784 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001785 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1786 * but allow concurrent faults), with pte both mapped and locked.
1787 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001789static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1790 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001791 spinlock_t *ptl, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792{
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001793 struct page *old_page, *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 pte_t entry;
Nick Piggin83c54072007-07-19 01:47:05 -07001795 int reuse = 0, ret = 0;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001796 int page_mkwrite = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001797 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798
Linus Torvalds6aab3412005-11-28 14:34:23 -08001799 old_page = vm_normal_page(vma, address, orig_pte);
Peter Zijlstra251b97f2008-07-04 09:59:24 -07001800 if (!old_page) {
1801 /*
1802 * VM_MIXEDMAP !pfn_valid() case
1803 *
1804 * We should not cow pages in a shared writeable mapping.
1805 * Just mark the pages writable as we can't do any dirty
1806 * accounting on raw pfn maps.
1807 */
1808 if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
1809 (VM_WRITE|VM_SHARED))
1810 goto reuse;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001811 goto gotten;
Peter Zijlstra251b97f2008-07-04 09:59:24 -07001812 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001814 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001815 * Take out anonymous pages first, anonymous shared vmas are
1816 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001817 */
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001818 if (PageAnon(old_page)) {
Nick Piggin529ae9a2008-08-02 12:01:03 +02001819 if (trylock_page(old_page)) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001820 reuse = can_share_swap_page(old_page);
1821 unlock_page(old_page);
1822 }
1823 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001824 (VM_WRITE|VM_SHARED))) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001825 /*
1826 * Only catch write-faults on shared writable pages,
1827 * read-only shared pages can get COWed by
1828 * get_user_pages(.write=1, .force=1).
1829 */
David Howells9637a5e2006-06-23 02:03:43 -07001830 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
1831 /*
1832 * Notify the address space that the page is about to
1833 * become writable so that it can prohibit this or wait
1834 * for the page to get into an appropriate state.
1835 *
1836 * We do this without the lock held, so that it can
1837 * sleep if it needs to.
1838 */
1839 page_cache_get(old_page);
1840 pte_unmap_unlock(page_table, ptl);
1841
1842 if (vma->vm_ops->page_mkwrite(vma, old_page) < 0)
1843 goto unwritable_page;
1844
David Howells9637a5e2006-06-23 02:03:43 -07001845 /*
1846 * Since we dropped the lock we need to revalidate
1847 * the PTE as someone else may have changed it. If
1848 * they did, we just return, as we can count on the
1849 * MMU to tell us if they didn't also make it writable.
1850 */
1851 page_table = pte_offset_map_lock(mm, pmd, address,
1852 &ptl);
Hugh Dickinsc3704ce2007-02-10 01:43:00 -08001853 page_cache_release(old_page);
David Howells9637a5e2006-06-23 02:03:43 -07001854 if (!pte_same(*page_table, orig_pte))
1855 goto unlock;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001856
1857 page_mkwrite = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001859 dirty_page = old_page;
1860 get_page(dirty_page);
David Howells9637a5e2006-06-23 02:03:43 -07001861 reuse = 1;
David Howells9637a5e2006-06-23 02:03:43 -07001862 }
1863
1864 if (reuse) {
Peter Zijlstra251b97f2008-07-04 09:59:24 -07001865reuse:
David Howells9637a5e2006-06-23 02:03:43 -07001866 flush_cache_page(vma, address, pte_pfn(orig_pte));
1867 entry = pte_mkyoung(orig_pte);
1868 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
KAMEZAWA Hiroyuki954ffcb2007-10-16 01:25:44 -07001869 if (ptep_set_access_flags(vma, address, page_table, entry,1))
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07001870 update_mmu_cache(vma, address, entry);
David Howells9637a5e2006-06-23 02:03:43 -07001871 ret |= VM_FAULT_WRITE;
1872 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874
1875 /*
1876 * Ok, we need to copy. Oh, well..
1877 */
Nick Pigginb5810032005-10-29 18:16:12 -07001878 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08001879gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001880 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881
1882 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07001883 goto oom;
Nick Piggin557ed1f2007-10-16 01:24:40 -07001884 VM_BUG_ON(old_page == ZERO_PAGE(0));
1885 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
1886 if (!new_page)
1887 goto oom;
Nick Pigginb291f002008-10-18 20:26:44 -07001888 /*
1889 * Don't let another task, with possibly unlocked vma,
1890 * keep the mlocked page.
1891 */
1892 if (vma->vm_flags & VM_LOCKED) {
1893 lock_page(old_page); /* for LRU manipulation */
1894 clear_page_mlock(old_page);
1895 unlock_page(old_page);
1896 }
Nick Piggin557ed1f2007-10-16 01:24:40 -07001897 cow_user_page(new_page, old_page, address, vma);
Nick Piggin0ed361d2008-02-04 22:29:34 -08001898 __SetPageUptodate(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001899
Balbir Singhe1a1cd52008-02-07 00:14:02 -08001900 if (mem_cgroup_charge(new_page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001901 goto oom_free_new;
1902
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 /*
1904 * Re-check the pte - we dropped the lock
1905 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001906 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07001907 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08001908 if (old_page) {
Hugh Dickins920fc352005-11-21 21:32:17 -08001909 if (!PageAnon(old_page)) {
1910 dec_mm_counter(mm, file_rss);
1911 inc_mm_counter(mm, anon_rss);
1912 }
1913 } else
Hugh Dickins42946212005-10-29 18:16:05 -07001914 inc_mm_counter(mm, anon_rss);
Ben Collinseca35132005-11-29 11:45:26 -08001915 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07001916 entry = mk_pte(new_page, vma->vm_page_prot);
1917 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Siddha, Suresh B4ce072f2006-09-29 01:58:42 -07001918 /*
1919 * Clear the pte entry and flush it first, before updating the
1920 * pte with the new entry. This will avoid a race condition
1921 * seen in the presence of one thread doing SMC and another
1922 * thread doing COW.
1923 */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001924 ptep_clear_flush_notify(vma, address, page_table);
Rik van Rielb2e18532008-10-18 20:26:30 -07001925 SetPageSwapBacked(new_page);
Lee Schermerhorn64d65192008-10-18 20:26:52 -07001926 lru_cache_add_active_or_unevictable(new_page, vma);
Nick Piggin9617d952006-01-06 00:11:12 -08001927 page_add_new_anon_rmap(new_page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928
Lee Schermerhorn64d65192008-10-18 20:26:52 -07001929//TODO: is this safe? do_anonymous_page() does it this way.
1930 set_pte_at(mm, address, page_table, entry);
1931 update_mmu_cache(vma, address, entry);
Nick Piggin945754a2008-06-23 14:30:30 +02001932 if (old_page) {
1933 /*
1934 * Only after switching the pte to the new page may
1935 * we remove the mapcount here. Otherwise another
1936 * process may come and find the rmap count decremented
1937 * before the pte is switched to the new page, and
1938 * "reuse" the old page writing into it while our pte
1939 * here still points into it and can be read by other
1940 * threads.
1941 *
1942 * The critical issue is to order this
1943 * page_remove_rmap with the ptp_clear_flush above.
1944 * Those stores are ordered by (if nothing else,)
1945 * the barrier present in the atomic_add_negative
1946 * in page_remove_rmap.
1947 *
1948 * Then the TLB flush in ptep_clear_flush ensures that
1949 * no process can access the old page before the
1950 * decremented mapcount is visible. And the old page
1951 * cannot be reused until after the decremented
1952 * mapcount is visible. So transitively, TLBs to
1953 * old page will be flushed before it can be reused.
1954 */
1955 page_remove_rmap(old_page, vma);
1956 }
1957
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 /* Free the old page.. */
1959 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001960 ret |= VM_FAULT_WRITE;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001961 } else
1962 mem_cgroup_uncharge_page(new_page);
1963
Hugh Dickins920fc352005-11-21 21:32:17 -08001964 if (new_page)
1965 page_cache_release(new_page);
1966 if (old_page)
1967 page_cache_release(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001968unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001969 pte_unmap_unlock(page_table, ptl);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001970 if (dirty_page) {
Anton Salikhmetov8f7b3d12008-01-23 02:21:18 +03001971 if (vma->vm_file)
1972 file_update_time(vma->vm_file);
1973
Nick Piggin79352892007-07-19 01:47:22 -07001974 /*
1975 * Yes, Virginia, this is actually required to prevent a race
1976 * with clear_page_dirty_for_io() from clearing the page dirty
1977 * bit after it clear all dirty ptes, but before a racing
1978 * do_wp_page installs a dirty pte.
1979 *
1980 * do_no_page is protected similarly.
1981 */
1982 wait_on_page_locked(dirty_page);
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001983 set_page_dirty_balance(dirty_page, page_mkwrite);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001984 put_page(dirty_page);
1985 }
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001986 return ret;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001987oom_free_new:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08001988 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001989oom:
Hugh Dickins920fc352005-11-21 21:32:17 -08001990 if (old_page)
1991 page_cache_release(old_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 return VM_FAULT_OOM;
David Howells9637a5e2006-06-23 02:03:43 -07001993
1994unwritable_page:
1995 page_cache_release(old_page);
1996 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997}
1998
1999/*
2000 * Helper functions for unmap_mapping_range().
2001 *
2002 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
2003 *
2004 * We have to restart searching the prio_tree whenever we drop the lock,
2005 * since the iterator is only valid while the lock is held, and anyway
2006 * a later vma might be split and reinserted earlier while lock dropped.
2007 *
2008 * The list of nonlinear vmas could be handled more efficiently, using
2009 * a placeholder, but handle it in the same way until a need is shown.
2010 * It is important to search the prio_tree before nonlinear list: a vma
2011 * may become nonlinear and be shifted from prio_tree to nonlinear list
2012 * while the lock is dropped; but never shifted from list to prio_tree.
2013 *
2014 * In order to make forward progress despite restarting the search,
2015 * vm_truncate_count is used to mark a vma as now dealt with, so we can
2016 * quickly skip it next time around. Since the prio_tree search only
2017 * shows us those vmas affected by unmapping the range in question, we
2018 * can't efficiently keep all vmas in step with mapping->truncate_count:
2019 * so instead reset them all whenever it wraps back to 0 (then go to 1).
2020 * mapping->truncate_count and vma->vm_truncate_count are protected by
2021 * i_mmap_lock.
2022 *
2023 * In order to make forward progress despite repeatedly restarting some
Hugh Dickinsee39b372005-04-19 13:29:15 -07002024 * large vma, note the restart_addr from unmap_vmas when it breaks out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 * and restart from that address when we reach that vma again. It might
2026 * have been split or merged, shrunk or extended, but never shifted: so
2027 * restart_addr remains valid so long as it remains in the vma's range.
2028 * unmap_mapping_range forces truncate_count to leap over page-aligned
2029 * values so we can save vma's restart_addr in its truncate_count field.
2030 */
2031#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
2032
2033static void reset_vma_truncate_counts(struct address_space *mapping)
2034{
2035 struct vm_area_struct *vma;
2036 struct prio_tree_iter iter;
2037
2038 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
2039 vma->vm_truncate_count = 0;
2040 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
2041 vma->vm_truncate_count = 0;
2042}
2043
2044static int unmap_mapping_range_vma(struct vm_area_struct *vma,
2045 unsigned long start_addr, unsigned long end_addr,
2046 struct zap_details *details)
2047{
2048 unsigned long restart_addr;
2049 int need_break;
2050
Nick Piggind00806b2007-07-19 01:46:57 -07002051 /*
2052 * files that support invalidating or truncating portions of the
Nick Piggind0217ac2007-07-19 01:47:03 -07002053 * file from under mmaped areas must have their ->fault function
Nick Piggin83c54072007-07-19 01:47:05 -07002054 * return a locked page (and set VM_FAULT_LOCKED in the return).
2055 * This provides synchronisation against concurrent unmapping here.
Nick Piggind00806b2007-07-19 01:46:57 -07002056 */
Nick Piggind00806b2007-07-19 01:46:57 -07002057
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058again:
2059 restart_addr = vma->vm_truncate_count;
2060 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
2061 start_addr = restart_addr;
2062 if (start_addr >= end_addr) {
2063 /* Top of vma has been split off since last time */
2064 vma->vm_truncate_count = details->truncate_count;
2065 return 0;
2066 }
2067 }
2068
Hugh Dickinsee39b372005-04-19 13:29:15 -07002069 restart_addr = zap_page_range(vma, start_addr,
2070 end_addr - start_addr, details);
Nick Piggin95c354f2008-01-30 13:31:20 +01002071 need_break = need_resched() || spin_needbreak(details->i_mmap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072
Hugh Dickinsee39b372005-04-19 13:29:15 -07002073 if (restart_addr >= end_addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 /* We have now completed this vma: mark it so */
2075 vma->vm_truncate_count = details->truncate_count;
2076 if (!need_break)
2077 return 0;
2078 } else {
2079 /* Note restart_addr in vma's truncate_count field */
Hugh Dickinsee39b372005-04-19 13:29:15 -07002080 vma->vm_truncate_count = restart_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 if (!need_break)
2082 goto again;
2083 }
2084
2085 spin_unlock(details->i_mmap_lock);
2086 cond_resched();
2087 spin_lock(details->i_mmap_lock);
2088 return -EINTR;
2089}
2090
2091static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
2092 struct zap_details *details)
2093{
2094 struct vm_area_struct *vma;
2095 struct prio_tree_iter iter;
2096 pgoff_t vba, vea, zba, zea;
2097
2098restart:
2099 vma_prio_tree_foreach(vma, &iter, root,
2100 details->first_index, details->last_index) {
2101 /* Skip quickly over those we have already dealt with */
2102 if (vma->vm_truncate_count == details->truncate_count)
2103 continue;
2104
2105 vba = vma->vm_pgoff;
2106 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
2107 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
2108 zba = details->first_index;
2109 if (zba < vba)
2110 zba = vba;
2111 zea = details->last_index;
2112 if (zea > vea)
2113 zea = vea;
2114
2115 if (unmap_mapping_range_vma(vma,
2116 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
2117 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
2118 details) < 0)
2119 goto restart;
2120 }
2121}
2122
2123static inline void unmap_mapping_range_list(struct list_head *head,
2124 struct zap_details *details)
2125{
2126 struct vm_area_struct *vma;
2127
2128 /*
2129 * In nonlinear VMAs there is no correspondence between virtual address
2130 * offset and file offset. So we must perform an exhaustive search
2131 * across *all* the pages in each nonlinear VMA, not just the pages
2132 * whose virtual address lies outside the file truncation point.
2133 */
2134restart:
2135 list_for_each_entry(vma, head, shared.vm_set.list) {
2136 /* Skip quickly over those we have already dealt with */
2137 if (vma->vm_truncate_count == details->truncate_count)
2138 continue;
2139 details->nonlinear_vma = vma;
2140 if (unmap_mapping_range_vma(vma, vma->vm_start,
2141 vma->vm_end, details) < 0)
2142 goto restart;
2143 }
2144}
2145
2146/**
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002147 * 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 -07002148 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 * @holebegin: byte in first page to unmap, relative to the start of
2150 * the underlying file. This will be rounded down to a PAGE_SIZE
2151 * boundary. Note that this is different from vmtruncate(), which
2152 * must keep the partial page. In contrast, we must get rid of
2153 * partial pages.
2154 * @holelen: size of prospective hole in bytes. This will be rounded
2155 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2156 * end of the file.
2157 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2158 * but 0 when invalidating pagecache, don't throw away private data.
2159 */
2160void unmap_mapping_range(struct address_space *mapping,
2161 loff_t const holebegin, loff_t const holelen, int even_cows)
2162{
2163 struct zap_details details;
2164 pgoff_t hba = holebegin >> PAGE_SHIFT;
2165 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2166
2167 /* Check for overflow. */
2168 if (sizeof(holelen) > sizeof(hlen)) {
2169 long long holeend =
2170 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2171 if (holeend & ~(long long)ULONG_MAX)
2172 hlen = ULONG_MAX - hba + 1;
2173 }
2174
2175 details.check_mapping = even_cows? NULL: mapping;
2176 details.nonlinear_vma = NULL;
2177 details.first_index = hba;
2178 details.last_index = hba + hlen - 1;
2179 if (details.last_index < details.first_index)
2180 details.last_index = ULONG_MAX;
2181 details.i_mmap_lock = &mapping->i_mmap_lock;
2182
2183 spin_lock(&mapping->i_mmap_lock);
2184
Nick Piggind00806b2007-07-19 01:46:57 -07002185 /* Protect against endless unmapping loops */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 mapping->truncate_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 if (unlikely(is_restart_addr(mapping->truncate_count))) {
2188 if (mapping->truncate_count == 0)
2189 reset_vma_truncate_counts(mapping);
2190 mapping->truncate_count++;
2191 }
2192 details.truncate_count = mapping->truncate_count;
2193
2194 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
2195 unmap_mapping_range_tree(&mapping->i_mmap, &details);
2196 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
2197 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
2198 spin_unlock(&mapping->i_mmap_lock);
2199}
2200EXPORT_SYMBOL(unmap_mapping_range);
2201
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07002202/**
2203 * vmtruncate - unmap mappings "freed" by truncate() syscall
2204 * @inode: inode of the file used
2205 * @offset: file offset to start truncating
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 *
2207 * NOTE! We have to be ready to update the memory sharing
2208 * between the file and the memory map for a potential last
2209 * incomplete page. Ugly, but necessary.
2210 */
2211int vmtruncate(struct inode * inode, loff_t offset)
2212{
Christoph Hellwig61d50482008-02-04 22:28:56 -08002213 if (inode->i_size < offset) {
2214 unsigned long limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215
Christoph Hellwig61d50482008-02-04 22:28:56 -08002216 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
2217 if (limit != RLIM_INFINITY && offset > limit)
2218 goto out_sig;
2219 if (offset > inode->i_sb->s_maxbytes)
2220 goto out_big;
2221 i_size_write(inode, offset);
2222 } else {
2223 struct address_space *mapping = inode->i_mapping;
Nick Piggind00806b2007-07-19 01:46:57 -07002224
Christoph Hellwig61d50482008-02-04 22:28:56 -08002225 /*
2226 * truncation of in-use swapfiles is disallowed - it would
2227 * cause subsequent swapout to scribble on the now-freed
2228 * blocks.
2229 */
2230 if (IS_SWAPFILE(inode))
2231 return -ETXTBSY;
2232 i_size_write(inode, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233
Christoph Hellwig61d50482008-02-04 22:28:56 -08002234 /*
2235 * unmap_mapping_range is called twice, first simply for
2236 * efficiency so that truncate_inode_pages does fewer
2237 * single-page unmaps. However after this first call, and
2238 * before truncate_inode_pages finishes, it is possible for
2239 * private pages to be COWed, which remain after
2240 * truncate_inode_pages finishes, hence the second
2241 * unmap_mapping_range call must be made for correctness.
2242 */
2243 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2244 truncate_inode_pages(mapping, offset);
2245 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2246 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 if (inode->i_op && inode->i_op->truncate)
2249 inode->i_op->truncate(inode);
2250 return 0;
Christoph Hellwig61d50482008-02-04 22:28:56 -08002251
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252out_sig:
2253 send_sig(SIGXFSZ, current, 0);
2254out_big:
2255 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257EXPORT_SYMBOL(vmtruncate);
2258
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002259int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
2260{
2261 struct address_space *mapping = inode->i_mapping;
2262
2263 /*
2264 * If the underlying filesystem is not going to provide
2265 * a way to truncate a range of blocks (punch a hole) -
2266 * we should return failure right now.
2267 */
2268 if (!inode->i_op || !inode->i_op->truncate_range)
2269 return -ENOSYS;
2270
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002271 mutex_lock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002272 down_write(&inode->i_alloc_sem);
2273 unmap_mapping_range(mapping, offset, (end - offset), 1);
2274 truncate_inode_pages_range(mapping, offset, end);
Nick Piggind00806b2007-07-19 01:46:57 -07002275 unmap_mapping_range(mapping, offset, (end - offset), 1);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002276 inode->i_op->truncate_range(inode, offset, end);
2277 up_write(&inode->i_alloc_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002278 mutex_unlock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002279
2280 return 0;
2281}
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002282
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002284 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2285 * but allow concurrent faults), and pte mapped but not yet locked.
2286 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002288static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2289 unsigned long address, pte_t *page_table, pmd_t *pmd,
2290 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002292 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 struct page *page;
Hugh Dickins65500d22005-10-29 18:15:59 -07002294 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 pte_t pte;
Nick Piggin83c54072007-07-19 01:47:05 -07002296 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002298 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002299 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002300
2301 entry = pte_to_swp_entry(orig_pte);
Christoph Lameter06972122006-06-23 02:03:35 -07002302 if (is_migration_entry(entry)) {
2303 migration_entry_wait(mm, pmd, address);
2304 goto out;
2305 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002306 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 page = lookup_swap_cache(entry);
2308 if (!page) {
Ashwin Chaugule098fe652006-12-06 20:31:54 -08002309 grab_swap_token(); /* Contend for token _before_ read-in */
Hugh Dickins02098fe2008-02-04 22:28:42 -08002310 page = swapin_readahead(entry,
2311 GFP_HIGHUSER_MOVABLE, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 if (!page) {
2313 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002314 * Back out if somebody else faulted in this pte
2315 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002317 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 if (likely(pte_same(*page_table, orig_pte)))
2319 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002320 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002321 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 }
2323
2324 /* Had to read the page from swap area: Major fault */
2325 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002326 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 }
2328
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002329 if (mem_cgroup_charge(page, mm, GFP_KERNEL)) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002330 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2331 ret = VM_FAULT_OOM;
2332 goto out;
2333 }
2334
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 mark_page_accessed(page);
2336 lock_page(page);
Balbir Singh20a10222007-11-14 17:00:33 -08002337 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338
2339 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002340 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002342 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002343 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002344 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002345
2346 if (unlikely(!PageUptodate(page))) {
2347 ret = VM_FAULT_SIGBUS;
2348 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 }
2350
2351 /* The page isn't present yet, go ahead with the fault. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352
Hugh Dickins42946212005-10-29 18:16:05 -07002353 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354 pte = mk_pte(page, vma->vm_page_prot);
2355 if (write_access && can_share_swap_page(page)) {
2356 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2357 write_access = 0;
2358 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359
2360 flush_icache_page(vma, page);
2361 set_pte_at(mm, address, page_table, pte);
2362 page_add_anon_rmap(page, vma, address);
2363
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002364 swap_free(entry);
Nick Pigginb291f002008-10-18 20:26:44 -07002365 if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002366 remove_exclusive_swap_page(page);
2367 unlock_page(page);
2368
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 if (write_access) {
Hugh Dickins61469f12008-03-04 14:29:04 -08002370 ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
2371 if (ret & VM_FAULT_ERROR)
2372 ret &= VM_FAULT_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373 goto out;
2374 }
2375
2376 /* No need to invalidate - it was non-present before */
2377 update_mmu_cache(vma, address, pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07002378unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002379 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380out:
2381 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002382out_nomap:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002383 mem_cgroup_uncharge_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002384 pte_unmap_unlock(page_table, ptl);
Kirill Korotaevb8107482005-05-16 21:53:50 -07002385 unlock_page(page);
2386 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002387 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388}
2389
2390/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002391 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2392 * but allow concurrent faults), and pte mapped but not yet locked.
2393 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002395static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2396 unsigned long address, pte_t *page_table, pmd_t *pmd,
2397 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002399 struct page *page;
2400 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402
Nick Piggin557ed1f2007-10-16 01:24:40 -07002403 /* Allocate our own private page. */
2404 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405
Nick Piggin557ed1f2007-10-16 01:24:40 -07002406 if (unlikely(anon_vma_prepare(vma)))
2407 goto oom;
2408 page = alloc_zeroed_user_highpage_movable(vma, address);
2409 if (!page)
2410 goto oom;
Nick Piggin0ed361d2008-02-04 22:29:34 -08002411 __SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002413 if (mem_cgroup_charge(page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002414 goto oom_free_page;
2415
Nick Piggin557ed1f2007-10-16 01:24:40 -07002416 entry = mk_pte(page, vma->vm_page_prot);
2417 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002418
Nick Piggin557ed1f2007-10-16 01:24:40 -07002419 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2420 if (!pte_none(*page_table))
2421 goto release;
2422 inc_mm_counter(mm, anon_rss);
Rik van Rielb2e18532008-10-18 20:26:30 -07002423 SetPageSwapBacked(page);
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002424 lru_cache_add_active_or_unevictable(page, vma);
Nick Piggin557ed1f2007-10-16 01:24:40 -07002425 page_add_new_anon_rmap(page, vma, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002426 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427
2428 /* No need to invalidate - it was non-present before */
Hugh Dickins65500d22005-10-29 18:15:59 -07002429 update_mmu_cache(vma, address, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002430unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002431 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002432 return 0;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002433release:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002434 mem_cgroup_uncharge_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002435 page_cache_release(page);
2436 goto unlock;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002437oom_free_page:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002438 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002439oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 return VM_FAULT_OOM;
2441}
2442
2443/*
Nick Piggin54cb8822007-07-19 01:46:59 -07002444 * __do_fault() tries to create a new page mapping. It aggressively
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445 * tries to share with existing pages, but makes a separate copy if
Nick Piggin54cb8822007-07-19 01:46:59 -07002446 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2447 * the next page fault.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448 *
2449 * As this is called only for pages that do not currently exist, we
2450 * do not need to flush old virtual caches or the TLB.
2451 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002452 * We enter with non-exclusive mmap_sem (to exclude vma changes,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002453 * but allow concurrent faults), and pte neither mapped nor locked.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002454 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455 */
Nick Piggin54cb8822007-07-19 01:46:59 -07002456static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002457 unsigned long address, pmd_t *pmd,
Nick Piggin54cb8822007-07-19 01:46:59 -07002458 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459{
Hugh Dickins16abfa02007-10-04 16:56:06 +01002460 pte_t *page_table;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002461 spinlock_t *ptl;
Nick Piggind0217ac2007-07-19 01:47:03 -07002462 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464 int anon = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002465 struct page *dirty_page = NULL;
Nick Piggind0217ac2007-07-19 01:47:03 -07002466 struct vm_fault vmf;
2467 int ret;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002468 int page_mkwrite = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07002469
Nick Piggind0217ac2007-07-19 01:47:03 -07002470 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2471 vmf.pgoff = pgoff;
2472 vmf.flags = flags;
2473 vmf.page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474
Nick Piggin3c18ddd2008-04-28 02:12:10 -07002475 ret = vma->vm_ops->fault(vma, &vmf);
2476 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2477 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478
Nick Piggind00806b2007-07-19 01:46:57 -07002479 /*
Nick Piggind0217ac2007-07-19 01:47:03 -07002480 * For consistency in subsequent calls, make the faulted page always
Nick Piggind00806b2007-07-19 01:46:57 -07002481 * locked.
2482 */
Nick Piggin83c54072007-07-19 01:47:05 -07002483 if (unlikely(!(ret & VM_FAULT_LOCKED)))
Nick Piggind0217ac2007-07-19 01:47:03 -07002484 lock_page(vmf.page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002485 else
Nick Piggind0217ac2007-07-19 01:47:03 -07002486 VM_BUG_ON(!PageLocked(vmf.page));
Nick Piggind00806b2007-07-19 01:46:57 -07002487
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 /*
2489 * Should we do an early C-O-W break?
2490 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002491 page = vmf.page;
Nick Piggin54cb8822007-07-19 01:46:59 -07002492 if (flags & FAULT_FLAG_WRITE) {
David Howells9637a5e2006-06-23 02:03:43 -07002493 if (!(vma->vm_flags & VM_SHARED)) {
Nick Piggin54cb8822007-07-19 01:46:59 -07002494 anon = 1;
Nick Piggind00806b2007-07-19 01:46:57 -07002495 if (unlikely(anon_vma_prepare(vma))) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002496 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002497 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002498 }
Nick Piggin83c54072007-07-19 01:47:05 -07002499 page = alloc_page_vma(GFP_HIGHUSER_MOVABLE,
2500 vma, address);
Nick Piggind00806b2007-07-19 01:46:57 -07002501 if (!page) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002502 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002503 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002504 }
Nick Pigginb291f002008-10-18 20:26:44 -07002505 /*
2506 * Don't let another task, with possibly unlocked vma,
2507 * keep the mlocked page.
2508 */
2509 if (vma->vm_flags & VM_LOCKED)
2510 clear_page_mlock(vmf.page);
Nick Piggind0217ac2007-07-19 01:47:03 -07002511 copy_user_highpage(page, vmf.page, address, vma);
Nick Piggin0ed361d2008-02-04 22:29:34 -08002512 __SetPageUptodate(page);
David Howells9637a5e2006-06-23 02:03:43 -07002513 } else {
Nick Piggin54cb8822007-07-19 01:46:59 -07002514 /*
2515 * If the page will be shareable, see if the backing
David Howells9637a5e2006-06-23 02:03:43 -07002516 * address space wants to know that the page is about
Nick Piggin54cb8822007-07-19 01:46:59 -07002517 * to become writable
2518 */
Mark Fasheh69676142007-07-19 01:47:00 -07002519 if (vma->vm_ops->page_mkwrite) {
2520 unlock_page(page);
2521 if (vma->vm_ops->page_mkwrite(vma, page) < 0) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002522 ret = VM_FAULT_SIGBUS;
2523 anon = 1; /* no anon but release vmf.page */
Mark Fasheh69676142007-07-19 01:47:00 -07002524 goto out_unlocked;
2525 }
2526 lock_page(page);
Nick Piggind0217ac2007-07-19 01:47:03 -07002527 /*
2528 * XXX: this is not quite right (racy vs
2529 * invalidate) to unlock and relock the page
2530 * like this, however a better fix requires
2531 * reworking page_mkwrite locking API, which
2532 * is better done later.
2533 */
2534 if (!page->mapping) {
Nick Piggin83c54072007-07-19 01:47:05 -07002535 ret = 0;
Nick Piggind0217ac2007-07-19 01:47:03 -07002536 anon = 1; /* no anon but release vmf.page */
2537 goto out;
2538 }
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002539 page_mkwrite = 1;
David Howells9637a5e2006-06-23 02:03:43 -07002540 }
2541 }
Nick Piggin54cb8822007-07-19 01:46:59 -07002542
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 }
2544
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002545 if (mem_cgroup_charge(page, mm, GFP_KERNEL)) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002546 ret = VM_FAULT_OOM;
2547 goto out;
2548 }
2549
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002550 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551
2552 /*
2553 * This silly early PAGE_DIRTY setting removes a race
2554 * due to the bad i386 page protection. But it's valid
2555 * for other architectures too.
2556 *
2557 * Note that if write_access is true, we either now have
2558 * an exclusive copy of the page, or this is a shared mapping,
2559 * so we can make it writable and dirty to avoid having to
2560 * handle that later.
2561 */
2562 /* Only go through if we didn't race with anybody else... */
Nick Piggin54cb8822007-07-19 01:46:59 -07002563 if (likely(pte_same(*page_table, orig_pte))) {
Nick Piggind00806b2007-07-19 01:46:57 -07002564 flush_icache_page(vma, page);
2565 entry = mk_pte(page, vma->vm_page_prot);
Nick Piggin54cb8822007-07-19 01:46:59 -07002566 if (flags & FAULT_FLAG_WRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 if (anon) {
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002569 inc_mm_counter(mm, anon_rss);
Rik van Rielb2e18532008-10-18 20:26:30 -07002570 SetPageSwapBacked(page);
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002571 lru_cache_add_active_or_unevictable(page, vma);
2572 page_add_new_anon_rmap(page, vma, address);
Hugh Dickinsf57e88a2005-11-21 21:32:19 -08002573 } else {
Hugh Dickins42946212005-10-29 18:16:05 -07002574 inc_mm_counter(mm, file_rss);
Nick Piggind00806b2007-07-19 01:46:57 -07002575 page_add_file_rmap(page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002576 if (flags & FAULT_FLAG_WRITE) {
Nick Piggind00806b2007-07-19 01:46:57 -07002577 dirty_page = page;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002578 get_page(dirty_page);
2579 }
Hugh Dickins42946212005-10-29 18:16:05 -07002580 }
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002581//TODO: is this safe? do_anonymous_page() does it this way.
2582 set_pte_at(mm, address, page_table, entry);
Nick Piggind00806b2007-07-19 01:46:57 -07002583
2584 /* no need to invalidate: a not-present page won't be cached */
2585 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 } else {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002587 mem_cgroup_uncharge_page(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002588 if (anon)
2589 page_cache_release(page);
2590 else
Nick Piggin54cb8822007-07-19 01:46:59 -07002591 anon = 1; /* no anon but release faulted_page */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592 }
2593
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002594 pte_unmap_unlock(page_table, ptl);
Nick Piggind00806b2007-07-19 01:46:57 -07002595
2596out:
Nick Piggind0217ac2007-07-19 01:47:03 -07002597 unlock_page(vmf.page);
Mark Fasheh69676142007-07-19 01:47:00 -07002598out_unlocked:
Nick Piggind00806b2007-07-19 01:46:57 -07002599 if (anon)
Nick Piggind0217ac2007-07-19 01:47:03 -07002600 page_cache_release(vmf.page);
Nick Piggind00806b2007-07-19 01:46:57 -07002601 else if (dirty_page) {
Anton Salikhmetov8f7b3d12008-01-23 02:21:18 +03002602 if (vma->vm_file)
2603 file_update_time(vma->vm_file);
2604
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002605 set_page_dirty_balance(dirty_page, page_mkwrite);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002606 put_page(dirty_page);
2607 }
Nick Piggind00806b2007-07-19 01:46:57 -07002608
Nick Piggin83c54072007-07-19 01:47:05 -07002609 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07002610}
Nick Piggind00806b2007-07-19 01:46:57 -07002611
Nick Piggin54cb8822007-07-19 01:46:59 -07002612static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2613 unsigned long address, pte_t *page_table, pmd_t *pmd,
2614 int write_access, pte_t orig_pte)
2615{
2616 pgoff_t pgoff = (((address & PAGE_MASK)
Dean Nelson0da7e012007-10-16 01:24:45 -07002617 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
Nick Piggin54cb8822007-07-19 01:46:59 -07002618 unsigned int flags = (write_access ? FAULT_FLAG_WRITE : 0);
2619
Hugh Dickins16abfa02007-10-04 16:56:06 +01002620 pte_unmap(page_table);
2621 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Nick Piggin54cb8822007-07-19 01:46:59 -07002622}
2623
Jes Sorensenf4b81802006-09-27 01:50:10 -07002624/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 * Fault of a previously existing named mapping. Repopulate the pte
2626 * from the encoded file_pte if possible. This enables swappable
2627 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002628 *
2629 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2630 * but allow concurrent faults), and pte mapped but not yet locked.
2631 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002633static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins65500d22005-10-29 18:15:59 -07002634 unsigned long address, pte_t *page_table, pmd_t *pmd,
2635 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636{
Nick Piggind0217ac2007-07-19 01:47:03 -07002637 unsigned int flags = FAULT_FLAG_NONLINEAR |
2638 (write_access ? FAULT_FLAG_WRITE : 0);
Hugh Dickins65500d22005-10-29 18:15:59 -07002639 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002641 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Nick Piggin83c54072007-07-19 01:47:05 -07002642 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643
Nick Piggind0217ac2007-07-19 01:47:03 -07002644 if (unlikely(!(vma->vm_flags & VM_NONLINEAR) ||
2645 !(vma->vm_flags & VM_CAN_NONLINEAR))) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002646 /*
2647 * Page table corrupted: show pte and kill process.
2648 */
Nick Pigginb5810032005-10-29 18:16:12 -07002649 print_bad_pte(vma, orig_pte, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002650 return VM_FAULT_OOM;
2651 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002652
2653 pgoff = pte_to_pgoff(orig_pte);
Hugh Dickins16abfa02007-10-04 16:56:06 +01002654 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655}
2656
2657/*
2658 * These routines also need to handle stuff like marking pages dirty
2659 * and/or accessed for architectures that don't do it in hardware (most
2660 * RISC architectures). The early dirtying is also good on the i386.
2661 *
2662 * There is also a hook called "update_mmu_cache()" that architectures
2663 * with external mmu caches can use to update those (ie the Sparc or
2664 * PowerPC hashed page tables that act as extended TLBs).
2665 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07002666 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2667 * but allow concurrent faults), and pte mapped but not yet locked.
2668 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 */
2670static inline int handle_pte_fault(struct mm_struct *mm,
Hugh Dickins65500d22005-10-29 18:15:59 -07002671 struct vm_area_struct *vma, unsigned long address,
2672 pte_t *pte, pmd_t *pmd, int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673{
2674 pte_t entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002675 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002677 entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002679 if (pte_none(entry)) {
Jes Sorensenf4b81802006-09-27 01:50:10 -07002680 if (vma->vm_ops) {
Nick Piggin3c18ddd2008-04-28 02:12:10 -07002681 if (likely(vma->vm_ops->fault))
Nick Piggin54cb8822007-07-19 01:46:59 -07002682 return do_linear_fault(mm, vma, address,
2683 pte, pmd, write_access, entry);
Jes Sorensenf4b81802006-09-27 01:50:10 -07002684 }
2685 return do_anonymous_page(mm, vma, address,
2686 pte, pmd, write_access);
Hugh Dickins65500d22005-10-29 18:15:59 -07002687 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 if (pte_file(entry))
Nick Piggind0217ac2007-07-19 01:47:03 -07002689 return do_nonlinear_fault(mm, vma, address,
Hugh Dickins65500d22005-10-29 18:15:59 -07002690 pte, pmd, write_access, entry);
2691 return do_swap_page(mm, vma, address,
2692 pte, pmd, write_access, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 }
2694
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002695 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002696 spin_lock(ptl);
2697 if (unlikely(!pte_same(*pte, entry)))
2698 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699 if (write_access) {
2700 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002701 return do_wp_page(mm, vma, address,
2702 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703 entry = pte_mkdirty(entry);
2704 }
2705 entry = pte_mkyoung(entry);
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002706 if (ptep_set_access_flags(vma, address, pte, entry, write_access)) {
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002707 update_mmu_cache(vma, address, entry);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002708 } else {
2709 /*
2710 * This is needed only for protection faults but the arch code
2711 * is not yet telling us if this is a protection fault or not.
2712 * This still avoids useless tlb flushes for .text page faults
2713 * with threads.
2714 */
2715 if (write_access)
2716 flush_tlb_page(vma, address);
2717 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002718unlock:
2719 pte_unmap_unlock(pte, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002720 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721}
2722
2723/*
2724 * By the time we get here, we already hold the mm semaphore
2725 */
Nick Piggin83c54072007-07-19 01:47:05 -07002726int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727 unsigned long address, int write_access)
2728{
2729 pgd_t *pgd;
2730 pud_t *pud;
2731 pmd_t *pmd;
2732 pte_t *pte;
2733
2734 __set_current_state(TASK_RUNNING);
2735
Christoph Lameterf8891e52006-06-30 01:55:45 -07002736 count_vm_event(PGFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01002738 if (unlikely(is_vm_hugetlb_page(vma)))
2739 return hugetlb_fault(mm, vma, address, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740
Linus Torvalds1da177e2005-04-16 15:20:36 -07002741 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 pud = pud_alloc(mm, pgd, address);
2743 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002744 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002745 pmd = pmd_alloc(mm, pud, address);
2746 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002747 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 pte = pte_alloc_map(mm, pmd, address);
2749 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002750 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751
Hugh Dickinsc74df322005-10-29 18:16:23 -07002752 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753}
2754
2755#ifndef __PAGETABLE_PUD_FOLDED
2756/*
2757 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002758 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002760int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002762 pud_t *new = pud_alloc_one(mm, address);
2763 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002764 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765
Nick Piggin362a61a2008-05-14 06:37:36 +02002766 smp_wmb(); /* See comment in __pte_alloc */
2767
Hugh Dickins872fec12005-10-29 18:16:21 -07002768 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002769 if (pgd_present(*pgd)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002770 pud_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002771 else
2772 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07002773 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002774 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775}
2776#endif /* __PAGETABLE_PUD_FOLDED */
2777
2778#ifndef __PAGETABLE_PMD_FOLDED
2779/*
2780 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002781 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002783int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002785 pmd_t *new = pmd_alloc_one(mm, address);
2786 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002787 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788
Nick Piggin362a61a2008-05-14 06:37:36 +02002789 smp_wmb(); /* See comment in __pte_alloc */
2790
Hugh Dickins872fec12005-10-29 18:16:21 -07002791 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07002793 if (pud_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002794 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002795 else
2796 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07002798 if (pgd_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002799 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002800 else
2801 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07002803 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002804 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805}
2806#endif /* __PAGETABLE_PMD_FOLDED */
2807
2808int make_pages_present(unsigned long addr, unsigned long end)
2809{
2810 int ret, len, write;
2811 struct vm_area_struct * vma;
2812
2813 vma = find_vma(current->mm, addr);
2814 if (!vma)
KOSAKI Motohiroa4770972008-08-04 13:41:14 -07002815 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 write = (vma->vm_flags & VM_WRITE) != 0;
Eric Sesterhenn5bcb28b2006-03-26 18:30:52 +02002817 BUG_ON(addr >= end);
2818 BUG_ON(end > vma->vm_end);
Rolf Eike Beer68e116a2007-07-15 23:38:03 -07002819 len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820 ret = get_user_pages(current, current->mm, addr,
2821 len, write, 0, NULL, NULL);
Lee Schermerhornc11d69d2008-10-18 20:26:56 -07002822 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823 return ret;
Lee Schermerhorn9978ad52008-10-18 20:26:56 -07002824 return ret == len ? 0 : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825}
2826
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827#if !defined(__HAVE_ARCH_GATE_AREA)
2828
2829#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07002830static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831
2832static int __init gate_vma_init(void)
2833{
2834 gate_vma.vm_mm = NULL;
2835 gate_vma.vm_start = FIXADDR_USER_START;
2836 gate_vma.vm_end = FIXADDR_USER_END;
Roland McGrathb6558c42007-01-26 00:56:47 -08002837 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
2838 gate_vma.vm_page_prot = __P101;
Roland McGrathf47aef52007-01-26 00:56:49 -08002839 /*
2840 * Make sure the vDSO gets into every core dump.
2841 * Dumping its contents makes post-mortem fully interpretable later
2842 * without matching up the same kernel and hardware config to see
2843 * what PC values meant.
2844 */
2845 gate_vma.vm_flags |= VM_ALWAYSDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846 return 0;
2847}
2848__initcall(gate_vma_init);
2849#endif
2850
2851struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2852{
2853#ifdef AT_SYSINFO_EHDR
2854 return &gate_vma;
2855#else
2856 return NULL;
2857#endif
2858}
2859
2860int in_gate_area_no_task(unsigned long addr)
2861{
2862#ifdef AT_SYSINFO_EHDR
2863 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2864 return 1;
2865#endif
2866 return 0;
2867}
2868
2869#endif /* __HAVE_ARCH_GATE_AREA */
David Howells0ec76a12006-09-27 01:50:15 -07002870
Rik van Riel28b2ee22008-07-23 21:27:05 -07002871#ifdef CONFIG_HAVE_IOREMAP_PROT
2872static resource_size_t follow_phys(struct vm_area_struct *vma,
2873 unsigned long address, unsigned int flags,
2874 unsigned long *prot)
2875{
2876 pgd_t *pgd;
2877 pud_t *pud;
2878 pmd_t *pmd;
2879 pte_t *ptep, pte;
2880 spinlock_t *ptl;
2881 resource_size_t phys_addr = 0;
2882 struct mm_struct *mm = vma->vm_mm;
2883
2884 VM_BUG_ON(!(vma->vm_flags & (VM_IO | VM_PFNMAP)));
2885
2886 pgd = pgd_offset(mm, address);
2887 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
2888 goto no_page_table;
2889
2890 pud = pud_offset(pgd, address);
2891 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
2892 goto no_page_table;
2893
2894 pmd = pmd_offset(pud, address);
2895 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
2896 goto no_page_table;
2897
2898 /* We cannot handle huge page PFN maps. Luckily they don't exist. */
2899 if (pmd_huge(*pmd))
2900 goto no_page_table;
2901
2902 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
2903 if (!ptep)
2904 goto out;
2905
2906 pte = *ptep;
2907 if (!pte_present(pte))
2908 goto unlock;
2909 if ((flags & FOLL_WRITE) && !pte_write(pte))
2910 goto unlock;
2911 phys_addr = pte_pfn(pte);
2912 phys_addr <<= PAGE_SHIFT; /* Shift here to avoid overflow on PAE */
2913
2914 *prot = pgprot_val(pte_pgprot(pte));
2915
2916unlock:
2917 pte_unmap_unlock(ptep, ptl);
2918out:
2919 return phys_addr;
2920no_page_table:
2921 return 0;
2922}
2923
2924int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
2925 void *buf, int len, int write)
2926{
2927 resource_size_t phys_addr;
2928 unsigned long prot = 0;
2929 void *maddr;
2930 int offset = addr & (PAGE_SIZE-1);
2931
2932 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
2933 return -EINVAL;
2934
2935 phys_addr = follow_phys(vma, addr, write, &prot);
2936
2937 if (!phys_addr)
2938 return -EINVAL;
2939
2940 maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot);
2941 if (write)
2942 memcpy_toio(maddr + offset, buf, len);
2943 else
2944 memcpy_fromio(buf, maddr + offset, len);
2945 iounmap(maddr);
2946
2947 return len;
2948}
2949#endif
2950
David Howells0ec76a12006-09-27 01:50:15 -07002951/*
2952 * Access another process' address space.
2953 * Source/target buffer must be kernel space,
2954 * Do not walk the page table directly, use get_user_pages
2955 */
2956int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2957{
2958 struct mm_struct *mm;
2959 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07002960 void *old_buf = buf;
2961
2962 mm = get_task_mm(tsk);
2963 if (!mm)
2964 return 0;
2965
2966 down_read(&mm->mmap_sem);
Simon Arlott183ff222007-10-20 01:27:18 +02002967 /* ignore errors, just check how much was successfully transferred */
David Howells0ec76a12006-09-27 01:50:15 -07002968 while (len) {
2969 int bytes, ret, offset;
2970 void *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07002971 struct page *page = NULL;
David Howells0ec76a12006-09-27 01:50:15 -07002972
2973 ret = get_user_pages(tsk, mm, addr, 1,
2974 write, 1, &page, &vma);
Rik van Riel28b2ee22008-07-23 21:27:05 -07002975 if (ret <= 0) {
2976 /*
2977 * Check if this is a VM_IO | VM_PFNMAP VMA, which
2978 * we can access using slightly different code.
2979 */
2980#ifdef CONFIG_HAVE_IOREMAP_PROT
2981 vma = find_vma(mm, addr);
2982 if (!vma)
2983 break;
2984 if (vma->vm_ops && vma->vm_ops->access)
2985 ret = vma->vm_ops->access(vma, addr, buf,
2986 len, write);
2987 if (ret <= 0)
2988#endif
2989 break;
2990 bytes = ret;
David Howells0ec76a12006-09-27 01:50:15 -07002991 } else {
Rik van Riel28b2ee22008-07-23 21:27:05 -07002992 bytes = len;
2993 offset = addr & (PAGE_SIZE-1);
2994 if (bytes > PAGE_SIZE-offset)
2995 bytes = PAGE_SIZE-offset;
2996
2997 maddr = kmap(page);
2998 if (write) {
2999 copy_to_user_page(vma, page, addr,
3000 maddr + offset, buf, bytes);
3001 set_page_dirty_lock(page);
3002 } else {
3003 copy_from_user_page(vma, page, addr,
3004 buf, maddr + offset, bytes);
3005 }
3006 kunmap(page);
3007 page_cache_release(page);
David Howells0ec76a12006-09-27 01:50:15 -07003008 }
David Howells0ec76a12006-09-27 01:50:15 -07003009 len -= bytes;
3010 buf += bytes;
3011 addr += bytes;
3012 }
3013 up_read(&mm->mmap_sem);
3014 mmput(mm);
3015
3016 return buf - old_buf;
3017}
Andi Kleen03252912008-01-30 13:33:18 +01003018
3019/*
3020 * Print the name of a VMA.
3021 */
3022void print_vma_addr(char *prefix, unsigned long ip)
3023{
3024 struct mm_struct *mm = current->mm;
3025 struct vm_area_struct *vma;
3026
Ingo Molnare8bff742008-02-13 20:21:06 +01003027 /*
3028 * Do not print if we are in atomic
3029 * contexts (in exception stacks, etc.):
3030 */
3031 if (preempt_count())
3032 return;
3033
Andi Kleen03252912008-01-30 13:33:18 +01003034 down_read(&mm->mmap_sem);
3035 vma = find_vma(mm, ip);
3036 if (vma && vma->vm_file) {
3037 struct file *f = vma->vm_file;
3038 char *buf = (char *)__get_free_page(GFP_KERNEL);
3039 if (buf) {
3040 char *p, *s;
3041
Jan Blunckcf28b482008-02-14 19:38:44 -08003042 p = d_path(&f->f_path, buf, PAGE_SIZE);
Andi Kleen03252912008-01-30 13:33:18 +01003043 if (IS_ERR(p))
3044 p = "?";
3045 s = strrchr(p, '/');
3046 if (s)
3047 p = s+1;
3048 printk("%s%s[%lx+%lx]", prefix, p,
3049 vma->vm_start,
3050 vma->vm_end - vma->vm_start);
3051 free_page((unsigned long)buf);
3052 }
3053 }
3054 up_read(&current->mm->mmap_sem);
3055}