blob: 164951c473058a25c081d5e47260d872068cdbb7 [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
Andy Whitcroftd41dee32005-06-23 00:07:54 -070067#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -070068/* use the per-pgdat data instead for discontigmem - mbligh */
69unsigned long max_mapnr;
70struct page *mem_map;
71
72EXPORT_SYMBOL(max_mapnr);
73EXPORT_SYMBOL(mem_map);
74#endif
75
76unsigned long num_physpages;
77/*
78 * A number of key systems in x86 including ioremap() rely on the assumption
79 * that high_memory defines the upper bound on direct map memory, then end
80 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
81 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
82 * and ZONE_HIGHMEM.
83 */
84void * high_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
86EXPORT_SYMBOL(num_physpages);
87EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Ingo Molnar32a93232008-02-06 22:39:44 +010089/*
90 * Randomize the address space (stacks, mmaps, brk, etc.).
91 *
92 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
93 * as ancient (libc5 based) binaries can segfault. )
94 */
95int randomize_va_space __read_mostly =
96#ifdef CONFIG_COMPAT_BRK
97 1;
98#else
99 2;
100#endif
Andi Kleena62eaf12006-02-16 23:41:58 +0100101
102static int __init disable_randmaps(char *s)
103{
104 randomize_va_space = 0;
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800105 return 1;
Andi Kleena62eaf12006-02-16 23:41:58 +0100106}
107__setup("norandmaps", disable_randmaps);
108
109
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110/*
111 * If a p?d_bad entry is found while walking page tables, report
112 * the error, before resetting entry to p?d_none. Usually (but
113 * very seldom) called out from the p?d_none_or_clear_bad macros.
114 */
115
116void pgd_clear_bad(pgd_t *pgd)
117{
118 pgd_ERROR(*pgd);
119 pgd_clear(pgd);
120}
121
122void pud_clear_bad(pud_t *pud)
123{
124 pud_ERROR(*pud);
125 pud_clear(pud);
126}
127
128void pmd_clear_bad(pmd_t *pmd)
129{
130 pmd_ERROR(*pmd);
131 pmd_clear(pmd);
132}
133
134/*
135 * Note: this doesn't free the actual pages themselves. That
136 * has been handled earlier when unmapping all the memory regions.
137 */
Hugh Dickinse0da3822005-04-19 13:29:15 -0700138static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800140 pgtable_t token = pmd_pgtable(*pmd);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700141 pmd_clear(pmd);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800142 pte_free_tlb(tlb, token);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700143 tlb->mm->nr_ptes--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144}
145
Hugh Dickinse0da3822005-04-19 13:29:15 -0700146static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
147 unsigned long addr, unsigned long end,
148 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149{
150 pmd_t *pmd;
151 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700152 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
Hugh Dickinse0da3822005-04-19 13:29:15 -0700154 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 do {
157 next = pmd_addr_end(addr, end);
158 if (pmd_none_or_clear_bad(pmd))
159 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700160 free_pte_range(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 } while (pmd++, addr = next, addr != end);
162
Hugh Dickinse0da3822005-04-19 13:29:15 -0700163 start &= PUD_MASK;
164 if (start < floor)
165 return;
166 if (ceiling) {
167 ceiling &= PUD_MASK;
168 if (!ceiling)
169 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700171 if (end - 1 > ceiling - 1)
172 return;
173
174 pmd = pmd_offset(pud, start);
175 pud_clear(pud);
176 pmd_free_tlb(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177}
178
Hugh Dickinse0da3822005-04-19 13:29:15 -0700179static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
180 unsigned long addr, unsigned long end,
181 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182{
183 pud_t *pud;
184 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700185 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
Hugh Dickinse0da3822005-04-19 13:29:15 -0700187 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 pud = pud_offset(pgd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 do {
190 next = pud_addr_end(addr, end);
191 if (pud_none_or_clear_bad(pud))
192 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700193 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 } while (pud++, addr = next, addr != end);
195
Hugh Dickinse0da3822005-04-19 13:29:15 -0700196 start &= PGDIR_MASK;
197 if (start < floor)
198 return;
199 if (ceiling) {
200 ceiling &= PGDIR_MASK;
201 if (!ceiling)
202 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700204 if (end - 1 > ceiling - 1)
205 return;
206
207 pud = pud_offset(pgd, start);
208 pgd_clear(pgd);
209 pud_free_tlb(tlb, pud);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210}
211
212/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700213 * This function frees user-level page tables of a process.
214 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 * Must be called with pagetable lock held.
216 */
Jan Beulich42b77722008-07-23 21:27:10 -0700217void free_pgd_range(struct mmu_gather *tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700218 unsigned long addr, unsigned long end,
219 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220{
221 pgd_t *pgd;
222 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700223 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224
Hugh Dickinse0da3822005-04-19 13:29:15 -0700225 /*
226 * The next few lines have given us lots of grief...
227 *
228 * Why are we testing PMD* at this top level? Because often
229 * there will be no work to do at all, and we'd prefer not to
230 * go all the way down to the bottom just to discover that.
231 *
232 * Why all these "- 1"s? Because 0 represents both the bottom
233 * of the address space and the top of it (using -1 for the
234 * top wouldn't help much: the masks would do the wrong thing).
235 * The rule is that addr 0 and floor 0 refer to the bottom of
236 * the address space, but end 0 and ceiling 0 refer to the top
237 * Comparisons need to use "end - 1" and "ceiling - 1" (though
238 * that end 0 case should be mythical).
239 *
240 * Wherever addr is brought up or ceiling brought down, we must
241 * be careful to reject "the opposite 0" before it confuses the
242 * subsequent tests. But what about where end is brought down
243 * by PMD_SIZE below? no, end can't go down to 0 there.
244 *
245 * Whereas we round start (addr) and ceiling down, by different
246 * masks at different levels, in order to test whether a table
247 * now has no other vmas using it, so can be freed, we don't
248 * bother to round floor or end up - the tests don't need that.
249 */
250
251 addr &= PMD_MASK;
252 if (addr < floor) {
253 addr += PMD_SIZE;
254 if (!addr)
255 return;
256 }
257 if (ceiling) {
258 ceiling &= PMD_MASK;
259 if (!ceiling)
260 return;
261 }
262 if (end - 1 > ceiling - 1)
263 end -= PMD_SIZE;
264 if (addr > end - 1)
265 return;
266
267 start = addr;
Jan Beulich42b77722008-07-23 21:27:10 -0700268 pgd = pgd_offset(tlb->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 do {
270 next = pgd_addr_end(addr, end);
271 if (pgd_none_or_clear_bad(pgd))
272 continue;
Jan Beulich42b77722008-07-23 21:27:10 -0700273 free_pud_range(tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700275}
276
Jan Beulich42b77722008-07-23 21:27:10 -0700277void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700278 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700279{
280 while (vma) {
281 struct vm_area_struct *next = vma->vm_next;
282 unsigned long addr = vma->vm_start;
283
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700284 /*
285 * Hide vma from rmap and vmtruncate before freeing pgtables
286 */
287 anon_vma_unlink(vma);
288 unlink_file_vma(vma);
289
David Gibson9da61ae2006-03-22 00:08:57 -0800290 if (is_vm_hugetlb_page(vma)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700291 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700292 floor, next? next->vm_start: ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700293 } else {
294 /*
295 * Optimization: gather nearby vmas into one call down
296 */
297 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
David Gibson48669202006-03-22 00:08:58 -0800298 && !is_vm_hugetlb_page(next)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700299 vma = next;
300 next = vma->vm_next;
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700301 anon_vma_unlink(vma);
302 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700303 }
304 free_pgd_range(tlb, addr, vma->vm_end,
305 floor, next? next->vm_start: ceiling);
306 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700307 vma = next;
308 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309}
310
Hugh Dickins1bb36302005-10-29 18:16:22 -0700311int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800313 pgtable_t new = pte_alloc_one(mm, address);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700314 if (!new)
315 return -ENOMEM;
316
Nick Piggin362a61a2008-05-14 06:37:36 +0200317 /*
318 * Ensure all pte setup (eg. pte page lock and page clearing) are
319 * visible before the pte is made visible to other CPUs by being
320 * put into page tables.
321 *
322 * The other side of the story is the pointer chasing in the page
323 * table walking code (when walking the page table without locking;
324 * ie. most of the time). Fortunately, these data accesses consist
325 * of a chain of data-dependent loads, meaning most CPUs (alpha
326 * being the notable exception) will already guarantee loads are
327 * seen in-order. See the alpha page table accessors for the
328 * smp_read_barrier_depends() barriers in page table walking code.
329 */
330 smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
331
Hugh Dickinsc74df322005-10-29 18:16:23 -0700332 spin_lock(&mm->page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800333 if (!pmd_present(*pmd)) { /* Has another populated it ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 mm->nr_ptes++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 pmd_populate(mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800336 new = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 }
Hugh Dickinsc74df322005-10-29 18:16:23 -0700338 spin_unlock(&mm->page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800339 if (new)
340 pte_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700341 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342}
343
Hugh Dickins1bb36302005-10-29 18:16:22 -0700344int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700346 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
347 if (!new)
348 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349
Nick Piggin362a61a2008-05-14 06:37:36 +0200350 smp_wmb(); /* See comment in __pte_alloc */
351
Hugh Dickins1bb36302005-10-29 18:16:22 -0700352 spin_lock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800353 if (!pmd_present(*pmd)) { /* Has another populated it ? */
Hugh Dickins1bb36302005-10-29 18:16:22 -0700354 pmd_populate_kernel(&init_mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800355 new = NULL;
356 }
Hugh Dickins1bb36302005-10-29 18:16:22 -0700357 spin_unlock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800358 if (new)
359 pte_free_kernel(&init_mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700360 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361}
362
Hugh Dickinsae859762005-10-29 18:16:05 -0700363static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
364{
365 if (file_rss)
366 add_mm_counter(mm, file_rss, file_rss);
367 if (anon_rss)
368 add_mm_counter(mm, anon_rss, anon_rss);
369}
370
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800372 * This function is called to print an error when a bad pte
373 * is found. For example, we might have a PFN-mapped pte in
374 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700375 *
376 * The calling function must still handle the error.
377 */
Adrian Bunk15f59ad2008-07-25 19:46:23 -0700378static void print_bad_pte(struct vm_area_struct *vma, pte_t pte,
379 unsigned long vaddr)
Nick Pigginb5810032005-10-29 18:16:12 -0700380{
381 printk(KERN_ERR "Bad pte = %08llx, process = %s, "
382 "vm_flags = %lx, vaddr = %lx\n",
383 (long long)pte_val(pte),
384 (vma->vm_mm == current->mm ? current->comm : "???"),
385 vma->vm_flags, vaddr);
386 dump_stack();
387}
388
Linus Torvalds67121172005-12-11 20:38:17 -0800389static inline int is_cow_mapping(unsigned int flags)
390{
391 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
392}
393
Nick Pigginb5810032005-10-29 18:16:12 -0700394/*
Nick Piggin7e675132008-04-28 02:13:00 -0700395 * vm_normal_page -- This function gets the "struct page" associated with a pte.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800396 *
Nick Piggin7e675132008-04-28 02:13:00 -0700397 * "Special" mappings do not wish to be associated with a "struct page" (either
398 * it doesn't exist, or it exists but they don't want to touch it). In this
399 * case, NULL is returned here. "Normal" mappings do have a struct page.
Jared Hulbertb379d792008-04-28 02:12:58 -0700400 *
Nick Piggin7e675132008-04-28 02:13:00 -0700401 * There are 2 broad cases. Firstly, an architecture may define a pte_special()
402 * pte bit, in which case this function is trivial. Secondly, an architecture
403 * may not have a spare pte bit, which requires a more complicated scheme,
404 * described below.
405 *
406 * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
407 * special mapping (even if there are underlying and valid "struct pages").
408 * COWed pages of a VM_PFNMAP are always normal.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800409 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700410 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
411 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
Nick Piggin7e675132008-04-28 02:13:00 -0700412 * set, and the vm_pgoff will point to the first PFN mapped: thus every special
413 * mapping will always honor the rule
Linus Torvalds6aab3412005-11-28 14:34:23 -0800414 *
415 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
416 *
Nick Piggin7e675132008-04-28 02:13:00 -0700417 * And for normal mappings this is false.
Jared Hulbertb379d792008-04-28 02:12:58 -0700418 *
Nick Piggin7e675132008-04-28 02:13:00 -0700419 * This restricts such mappings to be a linear translation from virtual address
420 * to pfn. To get around this restriction, we allow arbitrary mappings so long
421 * as the vma is not a COW mapping; in that case, we know that all ptes are
422 * special (because none can have been COWed).
Jared Hulbertb379d792008-04-28 02:12:58 -0700423 *
424 *
Nick Piggin7e675132008-04-28 02:13:00 -0700425 * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
426 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700427 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
428 * page" backing, however the difference is that _all_ pages with a struct
429 * page (that is, those where pfn_valid is true) are refcounted and considered
430 * normal pages by the VM. The disadvantage is that pages are refcounted
431 * (which can be slower and simply not an option for some PFNMAP users). The
432 * advantage is that we don't have to follow the strict linearity rule of
433 * PFNMAP mappings in order to support COWable mappings.
434 *
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800435 */
Nick Piggin7e675132008-04-28 02:13:00 -0700436#ifdef __HAVE_ARCH_PTE_SPECIAL
437# define HAVE_PTE_SPECIAL 1
438#else
439# define HAVE_PTE_SPECIAL 0
440#endif
441struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
442 pte_t pte)
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800443{
Nick Piggin7e675132008-04-28 02:13:00 -0700444 unsigned long pfn;
445
446 if (HAVE_PTE_SPECIAL) {
447 if (likely(!pte_special(pte))) {
448 VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
449 return pte_page(pte);
450 }
451 VM_BUG_ON(!(vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)));
452 return NULL;
453 }
454
455 /* !HAVE_PTE_SPECIAL case follows: */
456
457 pfn = pte_pfn(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800458
Jared Hulbertb379d792008-04-28 02:12:58 -0700459 if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
460 if (vma->vm_flags & VM_MIXEDMAP) {
461 if (!pfn_valid(pfn))
462 return NULL;
463 goto out;
464 } else {
Nick Piggin7e675132008-04-28 02:13:00 -0700465 unsigned long off;
466 off = (addr - vma->vm_start) >> PAGE_SHIFT;
Jared Hulbertb379d792008-04-28 02:12:58 -0700467 if (pfn == vma->vm_pgoff + off)
468 return NULL;
469 if (!is_cow_mapping(vma->vm_flags))
470 return NULL;
471 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800472 }
473
Nick Piggin7e675132008-04-28 02:13:00 -0700474 VM_BUG_ON(!pfn_valid(pfn));
Linus Torvalds6aab3412005-11-28 14:34:23 -0800475
476 /*
Nick Piggin7e675132008-04-28 02:13:00 -0700477 * NOTE! We still have PageReserved() pages in the page tables.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800478 *
Nick Piggin7e675132008-04-28 02:13:00 -0700479 * eg. VDSO mappings can cause them to exist.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800480 */
Jared Hulbertb379d792008-04-28 02:12:58 -0700481out:
Linus Torvalds6aab3412005-11-28 14:34:23 -0800482 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800483}
484
485/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 * copy one vm_area from one task to the other. Assumes the page tables
487 * already present in the new task to be cleared in the whole range
488 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 */
490
Hugh Dickins8c103762005-10-29 18:16:13 -0700491static inline void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700493 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700494 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495{
Nick Pigginb5810032005-10-29 18:16:12 -0700496 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 pte_t pte = *src_pte;
498 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499
500 /* pte contains position in swap or file, so copy. */
501 if (unlikely(!pte_present(pte))) {
502 if (!pte_file(pte)) {
Christoph Lameter06972122006-06-23 02:03:35 -0700503 swp_entry_t entry = pte_to_swp_entry(pte);
504
505 swap_duplicate(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 /* make sure dst_mm is on swapoff's mmlist. */
507 if (unlikely(list_empty(&dst_mm->mmlist))) {
508 spin_lock(&mmlist_lock);
Hugh Dickinsf412ac02005-10-29 18:16:41 -0700509 if (list_empty(&dst_mm->mmlist))
510 list_add(&dst_mm->mmlist,
511 &src_mm->mmlist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 spin_unlock(&mmlist_lock);
513 }
Christoph Lameter06972122006-06-23 02:03:35 -0700514 if (is_write_migration_entry(entry) &&
515 is_cow_mapping(vm_flags)) {
516 /*
517 * COW mappings require pages in both parent
518 * and child to be set to read.
519 */
520 make_migration_entry_read(&entry);
521 pte = swp_entry_to_pte(entry);
522 set_pte_at(src_mm, addr, src_pte, pte);
523 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700525 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 }
527
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 /*
529 * If it's a COW mapping, write protect it both
530 * in the parent and the child
531 */
Linus Torvalds67121172005-12-11 20:38:17 -0800532 if (is_cow_mapping(vm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 ptep_set_wrprotect(src_mm, addr, src_pte);
Zachary Amsden3dc90792006-09-30 23:29:30 -0700534 pte = pte_wrprotect(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 }
536
537 /*
538 * If it's a shared mapping, mark it clean in
539 * the child
540 */
541 if (vm_flags & VM_SHARED)
542 pte = pte_mkclean(pte);
543 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800544
545 page = vm_normal_page(vma, addr, pte);
546 if (page) {
547 get_page(page);
Nick Pigginc97a9e12007-05-16 22:11:21 -0700548 page_dup_rmap(page, vma, addr);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800549 rss[!!PageAnon(page)]++;
550 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700551
552out_set_pte:
553 set_pte_at(dst_mm, addr, dst_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554}
555
556static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
557 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
558 unsigned long addr, unsigned long end)
559{
560 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700561 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -0700562 int progress = 0;
Hugh Dickins8c103762005-10-29 18:16:13 -0700563 int rss[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564
565again:
Hugh Dickinsae859762005-10-29 18:16:05 -0700566 rss[1] = rss[0] = 0;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700567 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 if (!dst_pte)
569 return -ENOMEM;
570 src_pte = pte_offset_map_nested(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700571 src_ptl = pte_lockptr(src_mm, src_pmd);
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700572 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700573 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 do {
576 /*
577 * We are holding two locks at this point - either of them
578 * could generate latencies in another task on another CPU.
579 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700580 if (progress >= 32) {
581 progress = 0;
582 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +0100583 spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700584 break;
585 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 if (pte_none(*src_pte)) {
587 progress++;
588 continue;
589 }
Hugh Dickins8c103762005-10-29 18:16:13 -0700590 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 progress += 8;
592 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700594 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -0700595 spin_unlock(src_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 pte_unmap_nested(src_pte - 1);
Hugh Dickinsae859762005-10-29 18:16:05 -0700597 add_mm_rss(dst_mm, rss[0], rss[1]);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700598 pte_unmap_unlock(dst_pte - 1, dst_ptl);
599 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 if (addr != end)
601 goto again;
602 return 0;
603}
604
605static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
606 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
607 unsigned long addr, unsigned long end)
608{
609 pmd_t *src_pmd, *dst_pmd;
610 unsigned long next;
611
612 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
613 if (!dst_pmd)
614 return -ENOMEM;
615 src_pmd = pmd_offset(src_pud, addr);
616 do {
617 next = pmd_addr_end(addr, end);
618 if (pmd_none_or_clear_bad(src_pmd))
619 continue;
620 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
621 vma, addr, next))
622 return -ENOMEM;
623 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
624 return 0;
625}
626
627static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
628 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
629 unsigned long addr, unsigned long end)
630{
631 pud_t *src_pud, *dst_pud;
632 unsigned long next;
633
634 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
635 if (!dst_pud)
636 return -ENOMEM;
637 src_pud = pud_offset(src_pgd, addr);
638 do {
639 next = pud_addr_end(addr, end);
640 if (pud_none_or_clear_bad(src_pud))
641 continue;
642 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
643 vma, addr, next))
644 return -ENOMEM;
645 } while (dst_pud++, src_pud++, addr = next, addr != end);
646 return 0;
647}
648
649int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
650 struct vm_area_struct *vma)
651{
652 pgd_t *src_pgd, *dst_pgd;
653 unsigned long next;
654 unsigned long addr = vma->vm_start;
655 unsigned long end = vma->vm_end;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700656 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
Nick Piggind9928952005-08-28 16:49:11 +1000658 /*
659 * Don't copy ptes where a page fault will fill them correctly.
660 * Fork becomes much lighter when there are big shared or private
661 * readonly mappings. The tradeoff is that copy_page_range is more
662 * efficient than faulting.
663 */
Linus Torvalds4d7672b2005-12-16 10:21:23 -0800664 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
Nick Piggind9928952005-08-28 16:49:11 +1000665 if (!vma->anon_vma)
666 return 0;
667 }
668
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 if (is_vm_hugetlb_page(vma))
670 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
671
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700672 /*
673 * We need to invalidate the secondary MMU mappings only when
674 * there could be a permission downgrade on the ptes of the
675 * parent mm. And a permission downgrade will only happen if
676 * is_cow_mapping() returns true.
677 */
678 if (is_cow_mapping(vma->vm_flags))
679 mmu_notifier_invalidate_range_start(src_mm, addr, end);
680
681 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 dst_pgd = pgd_offset(dst_mm, addr);
683 src_pgd = pgd_offset(src_mm, addr);
684 do {
685 next = pgd_addr_end(addr, end);
686 if (pgd_none_or_clear_bad(src_pgd))
687 continue;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700688 if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
689 vma, addr, next))) {
690 ret = -ENOMEM;
691 break;
692 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700694
695 if (is_cow_mapping(vma->vm_flags))
696 mmu_notifier_invalidate_range_end(src_mm,
697 vma->vm_start, end);
698 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699}
700
Robin Holt51c6f662005-11-13 16:06:42 -0800701static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700702 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800704 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705{
Nick Pigginb5810032005-10-29 18:16:12 -0700706 struct mm_struct *mm = tlb->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 pte_t *pte;
Hugh Dickins508034a2005-10-29 18:16:30 -0700708 spinlock_t *ptl;
Hugh Dickinsae859762005-10-29 18:16:05 -0700709 int file_rss = 0;
710 int anon_rss = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711
Hugh Dickins508034a2005-10-29 18:16:30 -0700712 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700713 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 do {
715 pte_t ptent = *pte;
Robin Holt51c6f662005-11-13 16:06:42 -0800716 if (pte_none(ptent)) {
717 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800719 }
Hugh Dickins6f5e6b92006-03-16 23:04:09 -0800720
721 (*zap_work) -= PAGE_SIZE;
722
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800724 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -0800725
Linus Torvalds6aab3412005-11-28 14:34:23 -0800726 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 if (unlikely(details) && page) {
728 /*
729 * unmap_shared_mapping_pages() wants to
730 * invalidate cache without truncating:
731 * unmap shared but keep private pages.
732 */
733 if (details->check_mapping &&
734 details->check_mapping != page->mapping)
735 continue;
736 /*
737 * Each page->index must be checked when
738 * invalidating or truncating nonlinear.
739 */
740 if (details->nonlinear_vma &&
741 (page->index < details->first_index ||
742 page->index > details->last_index))
743 continue;
744 }
Nick Pigginb5810032005-10-29 18:16:12 -0700745 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -0700746 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 tlb_remove_tlb_entry(tlb, pte, addr);
748 if (unlikely(!page))
749 continue;
750 if (unlikely(details) && details->nonlinear_vma
751 && linear_page_index(details->nonlinear_vma,
752 addr) != page->index)
Nick Pigginb5810032005-10-29 18:16:12 -0700753 set_pte_at(mm, addr, pte,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 pgoff_to_pte(page->index));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 if (PageAnon(page))
Hugh Dickins86d912f2005-10-29 18:16:14 -0700756 anon_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700757 else {
758 if (pte_dirty(ptent))
759 set_page_dirty(page);
760 if (pte_young(ptent))
Ken Chendaa88c82007-02-10 01:43:18 -0800761 SetPageReferenced(page);
Hugh Dickins86d912f2005-10-29 18:16:14 -0700762 file_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700763 }
Nick Piggin7de6b802006-12-22 01:09:33 -0800764 page_remove_rmap(page, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 tlb_remove_page(tlb, page);
766 continue;
767 }
768 /*
769 * If details->check_mapping, we leave swap entries;
770 * if details->nonlinear_vma, we leave file entries.
771 */
772 if (unlikely(details))
773 continue;
774 if (!pte_file(ptent))
775 free_swap_and_cache(pte_to_swp_entry(ptent));
Zachary Amsden9888a1c2006-09-30 23:29:31 -0700776 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
Robin Holt51c6f662005-11-13 16:06:42 -0800777 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
Hugh Dickinsae859762005-10-29 18:16:05 -0700778
Hugh Dickins86d912f2005-10-29 18:16:14 -0700779 add_mm_rss(mm, file_rss, anon_rss);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700780 arch_leave_lazy_mmu_mode();
Hugh Dickins508034a2005-10-29 18:16:30 -0700781 pte_unmap_unlock(pte - 1, ptl);
Robin Holt51c6f662005-11-13 16:06:42 -0800782
783 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784}
785
Robin Holt51c6f662005-11-13 16:06:42 -0800786static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700787 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800789 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790{
791 pmd_t *pmd;
792 unsigned long next;
793
794 pmd = pmd_offset(pud, addr);
795 do {
796 next = pmd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800797 if (pmd_none_or_clear_bad(pmd)) {
798 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800800 }
801 next = zap_pte_range(tlb, vma, pmd, addr, next,
802 zap_work, details);
803 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
804
805 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806}
807
Robin Holt51c6f662005-11-13 16:06:42 -0800808static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700809 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800811 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812{
813 pud_t *pud;
814 unsigned long next;
815
816 pud = pud_offset(pgd, addr);
817 do {
818 next = pud_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800819 if (pud_none_or_clear_bad(pud)) {
820 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800822 }
823 next = zap_pmd_range(tlb, vma, pud, addr, next,
824 zap_work, details);
825 } while (pud++, addr = next, (addr != end && *zap_work > 0));
826
827 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828}
829
Robin Holt51c6f662005-11-13 16:06:42 -0800830static unsigned long unmap_page_range(struct mmu_gather *tlb,
831 struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800833 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834{
835 pgd_t *pgd;
836 unsigned long next;
837
838 if (details && !details->check_mapping && !details->nonlinear_vma)
839 details = NULL;
840
841 BUG_ON(addr >= end);
842 tlb_start_vma(tlb, vma);
843 pgd = pgd_offset(vma->vm_mm, addr);
844 do {
845 next = pgd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800846 if (pgd_none_or_clear_bad(pgd)) {
847 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800849 }
850 next = zap_pud_range(tlb, vma, pgd, addr, next,
851 zap_work, details);
852 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 tlb_end_vma(tlb, vma);
Robin Holt51c6f662005-11-13 16:06:42 -0800854
855 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856}
857
858#ifdef CONFIG_PREEMPT
859# define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
860#else
861/* No preempt: go for improved straight-line efficiency */
862# define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
863#endif
864
865/**
866 * unmap_vmas - unmap a range of memory covered by a list of vma's
867 * @tlbp: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 * @vma: the starting vma
869 * @start_addr: virtual address at which to start unmapping
870 * @end_addr: virtual address at which to end unmapping
871 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
872 * @details: details of nonlinear truncation or shared cache invalidation
873 *
Hugh Dickinsee39b372005-04-19 13:29:15 -0700874 * Returns the end address of the unmapping (restart addr if interrupted).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700876 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700878 * We aim to not hold locks for too long (for scheduling latency reasons).
879 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 * return the ending mmu_gather to the caller.
881 *
882 * Only addresses between `start' and `end' will be unmapped.
883 *
884 * The VMA list must be sorted in ascending virtual address order.
885 *
886 * unmap_vmas() assumes that the caller will flush the whole unmapped address
887 * range after unmap_vmas() returns. So the only responsibility here is to
888 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
889 * drops the lock and schedules.
890 */
Hugh Dickins508034a2005-10-29 18:16:30 -0700891unsigned long unmap_vmas(struct mmu_gather **tlbp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 struct vm_area_struct *vma, unsigned long start_addr,
893 unsigned long end_addr, unsigned long *nr_accounted,
894 struct zap_details *details)
895{
Robin Holt51c6f662005-11-13 16:06:42 -0800896 long zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
898 int tlb_start_valid = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -0700899 unsigned long start = start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700901 int fullmm = (*tlbp)->fullmm;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700902 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700904 mmu_notifier_invalidate_range_start(mm, start_addr, end_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 unsigned long end;
907
908 start = max(vma->vm_start, start_addr);
909 if (start >= vma->vm_end)
910 continue;
911 end = min(vma->vm_end, end_addr);
912 if (end <= vma->vm_start)
913 continue;
914
915 if (vma->vm_flags & VM_ACCOUNT)
916 *nr_accounted += (end - start) >> PAGE_SHIFT;
917
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 while (start != end) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 if (!tlb_start_valid) {
920 tlb_start = start;
921 tlb_start_valid = 1;
922 }
923
Robin Holt51c6f662005-11-13 16:06:42 -0800924 if (unlikely(is_vm_hugetlb_page(vma))) {
Andi Kleena137e1c2008-07-23 21:27:43 -0700925 /*
926 * It is undesirable to test vma->vm_file as it
927 * should be non-null for valid hugetlb area.
928 * However, vm_file will be NULL in the error
929 * cleanup path of do_mmap_pgoff. When
930 * hugetlbfs ->mmap method fails,
931 * do_mmap_pgoff() nullifies vma->vm_file
932 * before calling this function to clean up.
933 * Since no pte has actually been setup, it is
934 * safe to do nothing in this case.
935 */
936 if (vma->vm_file) {
937 unmap_hugepage_range(vma, start, end, NULL);
938 zap_work -= (end - start) /
Andi Kleena5516432008-07-23 21:27:41 -0700939 pages_per_huge_page(hstate_vma(vma));
Andi Kleena137e1c2008-07-23 21:27:43 -0700940 }
941
Robin Holt51c6f662005-11-13 16:06:42 -0800942 start = end;
943 } else
944 start = unmap_page_range(*tlbp, vma,
945 start, end, &zap_work, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946
Robin Holt51c6f662005-11-13 16:06:42 -0800947 if (zap_work > 0) {
948 BUG_ON(start != end);
949 break;
950 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951
952 tlb_finish_mmu(*tlbp, tlb_start, start);
953
954 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +0100955 (i_mmap_lock && spin_needbreak(i_mmap_lock))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 if (i_mmap_lock) {
Hugh Dickins508034a2005-10-29 18:16:30 -0700957 *tlbp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 goto out;
959 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 }
962
Hugh Dickins508034a2005-10-29 18:16:30 -0700963 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 tlb_start_valid = 0;
Robin Holt51c6f662005-11-13 16:06:42 -0800965 zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 }
967 }
968out:
Andrea Arcangelicddb8a52008-07-28 15:46:29 -0700969 mmu_notifier_invalidate_range_end(mm, start_addr, end_addr);
Hugh Dickinsee39b372005-04-19 13:29:15 -0700970 return start; /* which is now the end (or restart) address */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971}
972
973/**
974 * zap_page_range - remove user pages in a given range
975 * @vma: vm_area_struct holding the applicable pages
976 * @address: starting address of pages to zap
977 * @size: number of bytes to zap
978 * @details: details of nonlinear truncation or shared cache invalidation
979 */
Hugh Dickinsee39b372005-04-19 13:29:15 -0700980unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 unsigned long size, struct zap_details *details)
982{
983 struct mm_struct *mm = vma->vm_mm;
984 struct mmu_gather *tlb;
985 unsigned long end = address + size;
986 unsigned long nr_accounted = 0;
987
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -0700990 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -0700991 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
992 if (tlb)
993 tlb_finish_mmu(tlb, address, end);
Hugh Dickinsee39b372005-04-19 13:29:15 -0700994 return end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995}
996
Jack Steinerc627f9c2008-07-29 22:33:53 -0700997/**
998 * zap_vma_ptes - remove ptes mapping the vma
999 * @vma: vm_area_struct holding ptes to be zapped
1000 * @address: starting address of pages to zap
1001 * @size: number of bytes to zap
1002 *
1003 * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1004 *
1005 * The entire address range must be fully contained within the vma.
1006 *
1007 * Returns 0 if successful.
1008 */
1009int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
1010 unsigned long size)
1011{
1012 if (address < vma->vm_start || address + size > vma->vm_end ||
1013 !(vma->vm_flags & VM_PFNMAP))
1014 return -1;
1015 zap_page_range(vma, address, size, NULL);
1016 return 0;
1017}
1018EXPORT_SYMBOL_GPL(zap_vma_ptes);
1019
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020/*
1021 * Do a quick page-table lookup for a single page.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 */
Linus Torvalds6aab3412005-11-28 14:34:23 -08001023struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001024 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025{
1026 pgd_t *pgd;
1027 pud_t *pud;
1028 pmd_t *pmd;
1029 pte_t *ptep, pte;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001030 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 struct page *page;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001032 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001034 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
1035 if (!IS_ERR(page)) {
1036 BUG_ON(flags & FOLL_GET);
1037 goto out;
1038 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001040 page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 pgd = pgd_offset(mm, address);
1042 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001043 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
1045 pud = pud_offset(pgd, address);
Andi Kleenceb86872008-07-23 21:27:50 -07001046 if (pud_none(*pud))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001047 goto no_page_table;
Andi Kleenceb86872008-07-23 21:27:50 -07001048 if (pud_huge(*pud)) {
1049 BUG_ON(flags & FOLL_GET);
1050 page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE);
1051 goto out;
1052 }
1053 if (unlikely(pud_bad(*pud)))
1054 goto no_page_table;
1055
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 pmd = pmd_offset(pud, address);
Hugh Dickinsaeed5fc2008-05-06 20:49:23 +01001057 if (pmd_none(*pmd))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001058 goto no_page_table;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001059 if (pmd_huge(*pmd)) {
1060 BUG_ON(flags & FOLL_GET);
1061 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
1062 goto out;
1063 }
Hugh Dickinsaeed5fc2008-05-06 20:49:23 +01001064 if (unlikely(pmd_bad(*pmd)))
1065 goto no_page_table;
1066
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001067 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068
1069 pte = *ptep;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001070 if (!pte_present(pte))
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001071 goto no_page;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001072 if ((flags & FOLL_WRITE) && !pte_write(pte))
1073 goto unlock;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001074 page = vm_normal_page(vma, address, pte);
1075 if (unlikely(!page))
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001076 goto bad_page;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001077
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001078 if (flags & FOLL_GET)
1079 get_page(page);
1080 if (flags & FOLL_TOUCH) {
1081 if ((flags & FOLL_WRITE) &&
1082 !pte_dirty(pte) && !PageDirty(page))
1083 set_page_dirty(page);
1084 mark_page_accessed(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001086unlock:
1087 pte_unmap_unlock(ptep, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088out:
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001089 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001091bad_page:
1092 pte_unmap_unlock(ptep, ptl);
1093 return ERR_PTR(-EFAULT);
1094
1095no_page:
1096 pte_unmap_unlock(ptep, ptl);
1097 if (!pte_none(pte))
1098 return page;
1099 /* Fall through to ZERO_PAGE handling */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001100no_page_table:
1101 /*
1102 * When core dumping an enormous anonymous area that nobody
1103 * has touched so far, we don't want to allocate page tables.
1104 */
1105 if (flags & FOLL_ANON) {
Nick Piggin557ed1f2007-10-16 01:24:40 -07001106 page = ZERO_PAGE(0);
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001107 if (flags & FOLL_GET)
1108 get_page(page);
1109 BUG_ON(flags & FOLL_WRITE);
1110 }
1111 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112}
1113
Linus Torvalds672ca282008-06-23 11:21:37 -07001114/* Can we do the FOLL_ANON optimization? */
1115static inline int use_zero_page(struct vm_area_struct *vma)
1116{
1117 /*
1118 * We don't want to optimize FOLL_ANON for make_pages_present()
1119 * when it tries to page in a VM_LOCKED region. As to VM_SHARED,
1120 * we want to get the page from the page tables to make sure
1121 * that we serialize and update with any other user of that
1122 * mapping.
1123 */
1124 if (vma->vm_flags & (VM_LOCKED | VM_SHARED))
1125 return 0;
1126 /*
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001127 * And if we have a fault routine, it's not an anonymous region.
Linus Torvalds672ca282008-06-23 11:21:37 -07001128 */
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001129 return !vma->vm_ops || !vma->vm_ops->fault;
Linus Torvalds672ca282008-06-23 11:21:37 -07001130}
1131
Nick Pigginb291f002008-10-18 20:26:44 -07001132
1133
1134int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
1135 unsigned long start, int len, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 struct page **pages, struct vm_area_struct **vmas)
1137{
1138 int i;
Nick Pigginb291f002008-10-18 20:26:44 -07001139 unsigned int vm_flags = 0;
1140 int write = !!(flags & GUP_FLAGS_WRITE);
1141 int force = !!(flags & GUP_FLAGS_FORCE);
1142 int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
Jonathan Corbet900cf082008-02-11 16:17:33 -07001144 if (len <= 0)
1145 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 /*
1147 * Require read or write permissions.
1148 * If 'force' is set, we only require the "MAY" flags.
1149 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001150 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
1151 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 i = 0;
1153
1154 do {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001155 struct vm_area_struct *vma;
1156 unsigned int foll_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157
1158 vma = find_extend_vma(mm, start);
1159 if (!vma && in_gate_area(tsk, start)) {
1160 unsigned long pg = start & PAGE_MASK;
1161 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
1162 pgd_t *pgd;
1163 pud_t *pud;
1164 pmd_t *pmd;
1165 pte_t *pte;
Nick Pigginb291f002008-10-18 20:26:44 -07001166
1167 /* user gate pages are read-only */
1168 if (!ignore && write)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 return i ? : -EFAULT;
1170 if (pg > TASK_SIZE)
1171 pgd = pgd_offset_k(pg);
1172 else
1173 pgd = pgd_offset_gate(mm, pg);
1174 BUG_ON(pgd_none(*pgd));
1175 pud = pud_offset(pgd, pg);
1176 BUG_ON(pud_none(*pud));
1177 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001178 if (pmd_none(*pmd))
1179 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001181 if (pte_none(*pte)) {
1182 pte_unmap(pte);
1183 return i ? : -EFAULT;
1184 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 if (pages) {
Nick Pigginfa2a4552005-11-29 18:43:17 +11001186 struct page *page = vm_normal_page(gate_vma, start, *pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001187 pages[i] = page;
1188 if (page)
1189 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 }
1191 pte_unmap(pte);
1192 if (vmas)
1193 vmas[i] = gate_vma;
1194 i++;
1195 start += PAGE_SIZE;
1196 len--;
1197 continue;
1198 }
1199
Nick Pigginb291f002008-10-18 20:26:44 -07001200 if (!vma ||
1201 (vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
1202 (!ignore && !(vm_flags & vma->vm_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 return i ? : -EFAULT;
1204
1205 if (is_vm_hugetlb_page(vma)) {
1206 i = follow_hugetlb_page(mm, vma, pages, vmas,
Adam Litke5b23dbe2007-11-14 16:59:33 -08001207 &start, &len, i, write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 continue;
1209 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001210
1211 foll_flags = FOLL_TOUCH;
1212 if (pages)
1213 foll_flags |= FOLL_GET;
Linus Torvalds672ca282008-06-23 11:21:37 -07001214 if (!write && use_zero_page(vma))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001215 foll_flags |= FOLL_ANON;
1216
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 do {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001218 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219
Ethan Solomita462e00c2007-07-15 23:38:16 -07001220 /*
1221 * If tsk is ooming, cut off its access to large memory
1222 * allocations. It has a pending SIGKILL, but it can't
1223 * be processed until returning to user space.
1224 */
1225 if (unlikely(test_tsk_thread_flag(tsk, TIF_MEMDIE)))
Oleg Nesterov7a36a752008-07-04 09:59:28 -07001226 return i ? i : -ENOMEM;
Ethan Solomita462e00c2007-07-15 23:38:16 -07001227
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001228 if (write)
1229 foll_flags |= FOLL_WRITE;
1230
1231 cond_resched();
Linus Torvalds6aab3412005-11-28 14:34:23 -08001232 while (!(page = follow_page(vma, start, foll_flags))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001233 int ret;
Nick Piggin83c54072007-07-19 01:47:05 -07001234 ret = handle_mm_fault(mm, vma, start,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001235 foll_flags & FOLL_WRITE);
Nick Piggin83c54072007-07-19 01:47:05 -07001236 if (ret & VM_FAULT_ERROR) {
1237 if (ret & VM_FAULT_OOM)
1238 return i ? i : -ENOMEM;
1239 else if (ret & VM_FAULT_SIGBUS)
1240 return i ? i : -EFAULT;
1241 BUG();
1242 }
1243 if (ret & VM_FAULT_MAJOR)
1244 tsk->maj_flt++;
1245 else
1246 tsk->min_flt++;
1247
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001248 /*
Nick Piggin83c54072007-07-19 01:47:05 -07001249 * The VM_FAULT_WRITE bit tells us that
1250 * do_wp_page has broken COW when necessary,
1251 * even if maybe_mkwrite decided not to set
1252 * pte_write. We can thus safely do subsequent
1253 * page lookups as if they were reads.
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001254 */
1255 if (ret & VM_FAULT_WRITE)
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001256 foll_flags &= ~FOLL_WRITE;
Nick Piggin83c54072007-07-19 01:47:05 -07001257
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07001258 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 }
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001260 if (IS_ERR(page))
1261 return i ? i : PTR_ERR(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001263 pages[i] = page;
James Bottomley03beb072006-03-26 01:36:57 -08001264
Russell Kinga6f36be2006-12-30 22:24:19 +00001265 flush_anon_page(vma, page, start);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001266 flush_dcache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 }
1268 if (vmas)
1269 vmas[i] = vma;
1270 i++;
1271 start += PAGE_SIZE;
1272 len--;
Hugh Dickins08ef4722005-06-21 17:15:10 -07001273 } while (len && start < vma->vm_end);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001274 } while (len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 return i;
1276}
Nick Pigginb291f002008-10-18 20:26:44 -07001277
1278int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
1279 unsigned long start, int len, int write, int force,
1280 struct page **pages, struct vm_area_struct **vmas)
1281{
1282 int flags = 0;
1283
1284 if (write)
1285 flags |= GUP_FLAGS_WRITE;
1286 if (force)
1287 flags |= GUP_FLAGS_FORCE;
1288
1289 return __get_user_pages(tsk, mm,
1290 start, len, flags,
1291 pages, vmas);
1292}
1293
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294EXPORT_SYMBOL(get_user_pages);
1295
Harvey Harrison920c7a52008-02-04 22:29:26 -08001296pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr,
1297 spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001298{
1299 pgd_t * pgd = pgd_offset(mm, addr);
1300 pud_t * pud = pud_alloc(mm, pgd, addr);
1301 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001302 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001303 if (pmd)
1304 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1305 }
1306 return NULL;
1307}
1308
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001310 * This is the old fallback for page remapping.
1311 *
1312 * For historical reasons, it only allows reserved pages. Only
1313 * old drivers should use this, and they needed to mark their
1314 * pages reserved for the old functions anyway.
1315 */
Nick Piggin423bad62008-04-28 02:13:01 -07001316static int insert_page(struct vm_area_struct *vma, unsigned long addr,
1317 struct page *page, pgprot_t prot)
Linus Torvalds238f58d2005-11-29 13:01:56 -08001318{
Nick Piggin423bad62008-04-28 02:13:01 -07001319 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001320 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001321 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001322 spinlock_t *ptl;
1323
Linus Torvalds238f58d2005-11-29 13:01:56 -08001324 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001325 if (PageAnon(page))
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001326 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001327 retval = -ENOMEM;
1328 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001329 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001330 if (!pte)
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07001331 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001332 retval = -EBUSY;
1333 if (!pte_none(*pte))
1334 goto out_unlock;
1335
1336 /* Ok, finally just insert the thing.. */
1337 get_page(page);
1338 inc_mm_counter(mm, file_rss);
1339 page_add_file_rmap(page);
1340 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1341
1342 retval = 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001343 pte_unmap_unlock(pte, ptl);
1344 return retval;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001345out_unlock:
1346 pte_unmap_unlock(pte, ptl);
1347out:
1348 return retval;
1349}
1350
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001351/**
1352 * vm_insert_page - insert single page into user vma
1353 * @vma: user vma to map to
1354 * @addr: target user address of this page
1355 * @page: source kernel page
1356 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001357 * This allows drivers to insert individual pages they've allocated
1358 * into a user vma.
1359 *
1360 * The page has to be a nice clean _individual_ kernel allocation.
1361 * If you allocate a compound page, you need to have marked it as
1362 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001363 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001364 *
1365 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1366 * took an arbitrary page protection parameter. This doesn't allow
1367 * that. Your vma protection will have to be set up correctly, which
1368 * means that if you want a shared writable mapping, you'd better
1369 * ask for a shared writable mapping!
1370 *
1371 * The page does not need to be reserved.
1372 */
Nick Piggin423bad62008-04-28 02:13:01 -07001373int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
1374 struct page *page)
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001375{
1376 if (addr < vma->vm_start || addr >= vma->vm_end)
1377 return -EFAULT;
1378 if (!page_count(page))
1379 return -EINVAL;
Linus Torvalds4d7672b2005-12-16 10:21:23 -08001380 vma->vm_flags |= VM_INSERTPAGE;
Nick Piggin423bad62008-04-28 02:13:01 -07001381 return insert_page(vma, addr, page, vma->vm_page_prot);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001382}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001383EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001384
Nick Piggin423bad62008-04-28 02:13:01 -07001385static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1386 unsigned long pfn, pgprot_t prot)
1387{
1388 struct mm_struct *mm = vma->vm_mm;
1389 int retval;
1390 pte_t *pte, entry;
1391 spinlock_t *ptl;
1392
1393 retval = -ENOMEM;
1394 pte = get_locked_pte(mm, addr, &ptl);
1395 if (!pte)
1396 goto out;
1397 retval = -EBUSY;
1398 if (!pte_none(*pte))
1399 goto out_unlock;
1400
1401 /* Ok, finally just insert the thing.. */
1402 entry = pte_mkspecial(pfn_pte(pfn, prot));
1403 set_pte_at(mm, addr, pte, entry);
1404 update_mmu_cache(vma, addr, entry); /* XXX: why not for insert_page? */
1405
1406 retval = 0;
1407out_unlock:
1408 pte_unmap_unlock(pte, ptl);
1409out:
1410 return retval;
1411}
1412
Nick Piggine0dc0d82007-02-12 00:51:36 -08001413/**
1414 * vm_insert_pfn - insert single pfn into user vma
1415 * @vma: user vma to map to
1416 * @addr: target user address of this page
1417 * @pfn: source kernel pfn
1418 *
1419 * Similar to vm_inert_page, this allows drivers to insert individual pages
1420 * they've allocated into a user vma. Same comments apply.
1421 *
1422 * This function should only be called from a vm_ops->fault handler, and
1423 * in that case the handler should return NULL.
Nick Piggin0d71d10a2008-07-23 21:27:05 -07001424 *
1425 * vma cannot be a COW mapping.
1426 *
1427 * As this is called only for pages that do not currently exist, we
1428 * do not need to flush old virtual caches or the TLB.
Nick Piggine0dc0d82007-02-12 00:51:36 -08001429 */
1430int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
Nick Piggin423bad62008-04-28 02:13:01 -07001431 unsigned long pfn)
Nick Piggine0dc0d82007-02-12 00:51:36 -08001432{
Nick Piggin7e675132008-04-28 02:13:00 -07001433 /*
1434 * Technically, architectures with pte_special can avoid all these
1435 * restrictions (same for remap_pfn_range). However we would like
1436 * consistency in testing and feature parity among all, so we should
1437 * try to keep these invariants in place for everybody.
1438 */
Jared Hulbertb379d792008-04-28 02:12:58 -07001439 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
1440 BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
1441 (VM_PFNMAP|VM_MIXEDMAP));
1442 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1443 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
Nick Piggine0dc0d82007-02-12 00:51:36 -08001444
Nick Piggin423bad62008-04-28 02:13:01 -07001445 if (addr < vma->vm_start || addr >= vma->vm_end)
1446 return -EFAULT;
1447 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
Nick Piggine0dc0d82007-02-12 00:51:36 -08001448}
1449EXPORT_SYMBOL(vm_insert_pfn);
1450
Nick Piggin423bad62008-04-28 02:13:01 -07001451int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
1452 unsigned long pfn)
1453{
1454 BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
1455
1456 if (addr < vma->vm_start || addr >= vma->vm_end)
1457 return -EFAULT;
1458
1459 /*
1460 * If we don't have pte special, then we have to use the pfn_valid()
1461 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1462 * refcount the page if pfn_valid is true (hence insert_page rather
1463 * than insert_pfn).
1464 */
1465 if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) {
1466 struct page *page;
1467
1468 page = pfn_to_page(pfn);
1469 return insert_page(vma, addr, page, vma->vm_page_prot);
1470 }
1471 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
1472}
1473EXPORT_SYMBOL(vm_insert_mixed);
1474
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001475/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 * maps a range of physical memory into the requested pages. the old
1477 * mappings are removed. any references to nonexistent pages results
1478 * in null mappings (currently treated as "copy-on-access")
1479 */
1480static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1481 unsigned long addr, unsigned long end,
1482 unsigned long pfn, pgprot_t prot)
1483{
1484 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001485 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486
Hugh Dickinsc74df322005-10-29 18:16:23 -07001487 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 if (!pte)
1489 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001490 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 do {
1492 BUG_ON(!pte_none(*pte));
Nick Piggin7e675132008-04-28 02:13:00 -07001493 set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 pfn++;
1495 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001496 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001497 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 return 0;
1499}
1500
1501static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1502 unsigned long addr, unsigned long end,
1503 unsigned long pfn, pgprot_t prot)
1504{
1505 pmd_t *pmd;
1506 unsigned long next;
1507
1508 pfn -= addr >> PAGE_SHIFT;
1509 pmd = pmd_alloc(mm, pud, addr);
1510 if (!pmd)
1511 return -ENOMEM;
1512 do {
1513 next = pmd_addr_end(addr, end);
1514 if (remap_pte_range(mm, pmd, addr, next,
1515 pfn + (addr >> PAGE_SHIFT), prot))
1516 return -ENOMEM;
1517 } while (pmd++, addr = next, addr != end);
1518 return 0;
1519}
1520
1521static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1522 unsigned long addr, unsigned long end,
1523 unsigned long pfn, pgprot_t prot)
1524{
1525 pud_t *pud;
1526 unsigned long next;
1527
1528 pfn -= addr >> PAGE_SHIFT;
1529 pud = pud_alloc(mm, pgd, addr);
1530 if (!pud)
1531 return -ENOMEM;
1532 do {
1533 next = pud_addr_end(addr, end);
1534 if (remap_pmd_range(mm, pud, addr, next,
1535 pfn + (addr >> PAGE_SHIFT), prot))
1536 return -ENOMEM;
1537 } while (pud++, addr = next, addr != end);
1538 return 0;
1539}
1540
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001541/**
1542 * remap_pfn_range - remap kernel memory to userspace
1543 * @vma: user vma to map to
1544 * @addr: target user address to start at
1545 * @pfn: physical address of kernel memory
1546 * @size: size of map area
1547 * @prot: page protection flags for this mapping
1548 *
1549 * Note: this is only safe if the mm semaphore is held when called.
1550 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1552 unsigned long pfn, unsigned long size, pgprot_t prot)
1553{
1554 pgd_t *pgd;
1555 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001556 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 struct mm_struct *mm = vma->vm_mm;
1558 int err;
1559
1560 /*
1561 * Physically remapped pages are special. Tell the
1562 * rest of the world about it:
1563 * VM_IO tells people not to look at these pages
1564 * (accesses can have side effects).
Hugh Dickins0b14c172005-11-21 21:32:15 -08001565 * VM_RESERVED is specified all over the place, because
1566 * in 2.4 it kept swapout's vma scan off this vma; but
1567 * in 2.6 the LRU scan won't even find its pages, so this
1568 * flag means no more than count its pages in reserved_vm,
1569 * and omit it from core dump, even when VM_IO turned off.
Linus Torvalds6aab3412005-11-28 14:34:23 -08001570 * VM_PFNMAP tells the core MM that the base pages are just
1571 * raw PFN mappings, and do not have a "struct page" associated
1572 * with them.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001573 *
1574 * There's a horrible special case to handle copy-on-write
1575 * behaviour that some programs depend on. We mark the "original"
1576 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 */
Linus Torvalds67121172005-12-11 20:38:17 -08001578 if (is_cow_mapping(vma->vm_flags)) {
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001579 if (addr != vma->vm_start || end != vma->vm_end)
Linus Torvalds7fc7e2e2005-12-11 19:57:52 -08001580 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001581 vma->vm_pgoff = pfn;
1582 }
1583
Linus Torvalds6aab3412005-11-28 14:34:23 -08001584 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585
1586 BUG_ON(addr >= end);
1587 pfn -= addr >> PAGE_SHIFT;
1588 pgd = pgd_offset(mm, addr);
1589 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 do {
1591 next = pgd_addr_end(addr, end);
1592 err = remap_pud_range(mm, pgd, addr, next,
1593 pfn + (addr >> PAGE_SHIFT), prot);
1594 if (err)
1595 break;
1596 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 return err;
1598}
1599EXPORT_SYMBOL(remap_pfn_range);
1600
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001601static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1602 unsigned long addr, unsigned long end,
1603 pte_fn_t fn, void *data)
1604{
1605 pte_t *pte;
1606 int err;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001607 pgtable_t token;
Borislav Petkov94909912007-05-06 14:49:17 -07001608 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001609
1610 pte = (mm == &init_mm) ?
1611 pte_alloc_kernel(pmd, addr) :
1612 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1613 if (!pte)
1614 return -ENOMEM;
1615
1616 BUG_ON(pmd_huge(*pmd));
1617
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001618 token = pmd_pgtable(*pmd);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001619
1620 do {
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001621 err = fn(pte, token, addr, data);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001622 if (err)
1623 break;
1624 } while (pte++, addr += PAGE_SIZE, addr != end);
1625
1626 if (mm != &init_mm)
1627 pte_unmap_unlock(pte-1, ptl);
1628 return err;
1629}
1630
1631static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1632 unsigned long addr, unsigned long end,
1633 pte_fn_t fn, void *data)
1634{
1635 pmd_t *pmd;
1636 unsigned long next;
1637 int err;
1638
Andi Kleenceb86872008-07-23 21:27:50 -07001639 BUG_ON(pud_huge(*pud));
1640
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001641 pmd = pmd_alloc(mm, pud, addr);
1642 if (!pmd)
1643 return -ENOMEM;
1644 do {
1645 next = pmd_addr_end(addr, end);
1646 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1647 if (err)
1648 break;
1649 } while (pmd++, addr = next, addr != end);
1650 return err;
1651}
1652
1653static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1654 unsigned long addr, unsigned long end,
1655 pte_fn_t fn, void *data)
1656{
1657 pud_t *pud;
1658 unsigned long next;
1659 int err;
1660
1661 pud = pud_alloc(mm, pgd, addr);
1662 if (!pud)
1663 return -ENOMEM;
1664 do {
1665 next = pud_addr_end(addr, end);
1666 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1667 if (err)
1668 break;
1669 } while (pud++, addr = next, addr != end);
1670 return err;
1671}
1672
1673/*
1674 * Scan a region of virtual memory, filling in page tables as necessary
1675 * and calling a provided function on each leaf page table.
1676 */
1677int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1678 unsigned long size, pte_fn_t fn, void *data)
1679{
1680 pgd_t *pgd;
1681 unsigned long next;
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001682 unsigned long start = addr, end = addr + size;
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001683 int err;
1684
1685 BUG_ON(addr >= end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001686 mmu_notifier_invalidate_range_start(mm, start, end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001687 pgd = pgd_offset(mm, addr);
1688 do {
1689 next = pgd_addr_end(addr, end);
1690 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1691 if (err)
1692 break;
1693 } while (pgd++, addr = next, addr != end);
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001694 mmu_notifier_invalidate_range_end(mm, start, end);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001695 return err;
1696}
1697EXPORT_SYMBOL_GPL(apply_to_page_range);
1698
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001700 * handle_pte_fault chooses page fault handler according to an entry
1701 * which was read non-atomically. Before making any commitment, on
1702 * those architectures or configurations (e.g. i386 with PAE) which
1703 * might give a mix of unmatched parts, do_swap_page and do_file_page
1704 * must check under lock before unmapping the pte and proceeding
1705 * (but do_wp_page is only called after already making such a check;
1706 * and do_anonymous_page and do_no_page can safely check later on).
1707 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001708static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001709 pte_t *page_table, pte_t orig_pte)
1710{
1711 int same = 1;
1712#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1713 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001714 spinlock_t *ptl = pte_lockptr(mm, pmd);
1715 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001716 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001717 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001718 }
1719#endif
1720 pte_unmap(page_table);
1721 return same;
1722}
1723
1724/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1726 * servicing faults for write access. In the normal case, do always want
1727 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1728 * that do not have writing enabled, when used by access_process_vm.
1729 */
1730static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1731{
1732 if (likely(vma->vm_flags & VM_WRITE))
1733 pte = pte_mkwrite(pte);
1734 return pte;
1735}
1736
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001737static 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 -08001738{
1739 /*
1740 * If the source page was a PFN mapping, we don't have
1741 * a "struct page" for it. We do a best-effort copy by
1742 * just copying from the original user address. If that
1743 * fails, we just zero-fill it. Live with it.
1744 */
1745 if (unlikely(!src)) {
1746 void *kaddr = kmap_atomic(dst, KM_USER0);
Linus Torvalds5d2a2dbbc2005-11-29 14:07:55 -08001747 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1748
1749 /*
1750 * This really shouldn't fail, because the page is there
1751 * in the page tables. But it might just be unreadable,
1752 * in which case we just give up and fill the result with
1753 * zeroes.
1754 */
1755 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Linus Torvalds6aab3412005-11-28 14:34:23 -08001756 memset(kaddr, 0, PAGE_SIZE);
1757 kunmap_atomic(kaddr, KM_USER0);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07001758 flush_dcache_page(dst);
Nick Piggin0ed361d2008-02-04 22:29:34 -08001759 } else
1760 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001761}
1762
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 * This routine handles present pages, when users try to write
1765 * to a shared page. It is done by copying the page to a new address
1766 * and decrementing the shared-page counter for the old page.
1767 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 * Note that this routine assumes that the protection checks have been
1769 * done by the caller (the low-level page fault routine in most cases).
1770 * Thus we can safely just mark it writable once we've done any necessary
1771 * COW.
1772 *
1773 * We also mark the page dirty at this point even though the page will
1774 * change only once the write actually happens. This avoids a few races,
1775 * and potentially makes it more efficient.
1776 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001777 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1778 * but allow concurrent faults), with pte both mapped and locked.
1779 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001781static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1782 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001783 spinlock_t *ptl, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784{
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001785 struct page *old_page, *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 pte_t entry;
Nick Piggin83c54072007-07-19 01:47:05 -07001787 int reuse = 0, ret = 0;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001788 int page_mkwrite = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001789 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790
Linus Torvalds6aab3412005-11-28 14:34:23 -08001791 old_page = vm_normal_page(vma, address, orig_pte);
Peter Zijlstra251b97f2008-07-04 09:59:24 -07001792 if (!old_page) {
1793 /*
1794 * VM_MIXEDMAP !pfn_valid() case
1795 *
1796 * We should not cow pages in a shared writeable mapping.
1797 * Just mark the pages writable as we can't do any dirty
1798 * accounting on raw pfn maps.
1799 */
1800 if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
1801 (VM_WRITE|VM_SHARED))
1802 goto reuse;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001803 goto gotten;
Peter Zijlstra251b97f2008-07-04 09:59:24 -07001804 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001806 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001807 * Take out anonymous pages first, anonymous shared vmas are
1808 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001809 */
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001810 if (PageAnon(old_page)) {
Nick Piggin529ae9a2008-08-02 12:01:03 +02001811 if (trylock_page(old_page)) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001812 reuse = can_share_swap_page(old_page);
1813 unlock_page(old_page);
1814 }
1815 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001816 (VM_WRITE|VM_SHARED))) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001817 /*
1818 * Only catch write-faults on shared writable pages,
1819 * read-only shared pages can get COWed by
1820 * get_user_pages(.write=1, .force=1).
1821 */
David Howells9637a5e2006-06-23 02:03:43 -07001822 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
1823 /*
1824 * Notify the address space that the page is about to
1825 * become writable so that it can prohibit this or wait
1826 * for the page to get into an appropriate state.
1827 *
1828 * We do this without the lock held, so that it can
1829 * sleep if it needs to.
1830 */
1831 page_cache_get(old_page);
1832 pte_unmap_unlock(page_table, ptl);
1833
1834 if (vma->vm_ops->page_mkwrite(vma, old_page) < 0)
1835 goto unwritable_page;
1836
David Howells9637a5e2006-06-23 02:03:43 -07001837 /*
1838 * Since we dropped the lock we need to revalidate
1839 * the PTE as someone else may have changed it. If
1840 * they did, we just return, as we can count on the
1841 * MMU to tell us if they didn't also make it writable.
1842 */
1843 page_table = pte_offset_map_lock(mm, pmd, address,
1844 &ptl);
Hugh Dickinsc3704ce2007-02-10 01:43:00 -08001845 page_cache_release(old_page);
David Howells9637a5e2006-06-23 02:03:43 -07001846 if (!pte_same(*page_table, orig_pte))
1847 goto unlock;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001848
1849 page_mkwrite = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001851 dirty_page = old_page;
1852 get_page(dirty_page);
David Howells9637a5e2006-06-23 02:03:43 -07001853 reuse = 1;
David Howells9637a5e2006-06-23 02:03:43 -07001854 }
1855
1856 if (reuse) {
Peter Zijlstra251b97f2008-07-04 09:59:24 -07001857reuse:
David Howells9637a5e2006-06-23 02:03:43 -07001858 flush_cache_page(vma, address, pte_pfn(orig_pte));
1859 entry = pte_mkyoung(orig_pte);
1860 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
KAMEZAWA Hiroyuki954ffcb2007-10-16 01:25:44 -07001861 if (ptep_set_access_flags(vma, address, page_table, entry,1))
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07001862 update_mmu_cache(vma, address, entry);
David Howells9637a5e2006-06-23 02:03:43 -07001863 ret |= VM_FAULT_WRITE;
1864 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866
1867 /*
1868 * Ok, we need to copy. Oh, well..
1869 */
Nick Pigginb5810032005-10-29 18:16:12 -07001870 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08001871gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001872 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873
1874 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07001875 goto oom;
Nick Piggin557ed1f2007-10-16 01:24:40 -07001876 VM_BUG_ON(old_page == ZERO_PAGE(0));
1877 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
1878 if (!new_page)
1879 goto oom;
Nick Pigginb291f002008-10-18 20:26:44 -07001880 /*
1881 * Don't let another task, with possibly unlocked vma,
1882 * keep the mlocked page.
1883 */
1884 if (vma->vm_flags & VM_LOCKED) {
1885 lock_page(old_page); /* for LRU manipulation */
1886 clear_page_mlock(old_page);
1887 unlock_page(old_page);
1888 }
Nick Piggin557ed1f2007-10-16 01:24:40 -07001889 cow_user_page(new_page, old_page, address, vma);
Nick Piggin0ed361d2008-02-04 22:29:34 -08001890 __SetPageUptodate(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001891
Balbir Singhe1a1cd52008-02-07 00:14:02 -08001892 if (mem_cgroup_charge(new_page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001893 goto oom_free_new;
1894
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 /*
1896 * Re-check the pte - we dropped the lock
1897 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001898 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07001899 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08001900 if (old_page) {
Hugh Dickins920fc352005-11-21 21:32:17 -08001901 if (!PageAnon(old_page)) {
1902 dec_mm_counter(mm, file_rss);
1903 inc_mm_counter(mm, anon_rss);
1904 }
1905 } else
Hugh Dickins42946212005-10-29 18:16:05 -07001906 inc_mm_counter(mm, anon_rss);
Ben Collinseca35132005-11-29 11:45:26 -08001907 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07001908 entry = mk_pte(new_page, vma->vm_page_prot);
1909 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Siddha, Suresh B4ce072f2006-09-29 01:58:42 -07001910 /*
1911 * Clear the pte entry and flush it first, before updating the
1912 * pte with the new entry. This will avoid a race condition
1913 * seen in the presence of one thread doing SMC and another
1914 * thread doing COW.
1915 */
Andrea Arcangelicddb8a52008-07-28 15:46:29 -07001916 ptep_clear_flush_notify(vma, address, page_table);
Rik van Rielb2e18532008-10-18 20:26:30 -07001917 SetPageSwapBacked(new_page);
Lee Schermerhorn64d65192008-10-18 20:26:52 -07001918 lru_cache_add_active_or_unevictable(new_page, vma);
Nick Piggin9617d952006-01-06 00:11:12 -08001919 page_add_new_anon_rmap(new_page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920
Lee Schermerhorn64d65192008-10-18 20:26:52 -07001921//TODO: is this safe? do_anonymous_page() does it this way.
1922 set_pte_at(mm, address, page_table, entry);
1923 update_mmu_cache(vma, address, entry);
Nick Piggin945754a2008-06-23 14:30:30 +02001924 if (old_page) {
1925 /*
1926 * Only after switching the pte to the new page may
1927 * we remove the mapcount here. Otherwise another
1928 * process may come and find the rmap count decremented
1929 * before the pte is switched to the new page, and
1930 * "reuse" the old page writing into it while our pte
1931 * here still points into it and can be read by other
1932 * threads.
1933 *
1934 * The critical issue is to order this
1935 * page_remove_rmap with the ptp_clear_flush above.
1936 * Those stores are ordered by (if nothing else,)
1937 * the barrier present in the atomic_add_negative
1938 * in page_remove_rmap.
1939 *
1940 * Then the TLB flush in ptep_clear_flush ensures that
1941 * no process can access the old page before the
1942 * decremented mapcount is visible. And the old page
1943 * cannot be reused until after the decremented
1944 * mapcount is visible. So transitively, TLBs to
1945 * old page will be flushed before it can be reused.
1946 */
1947 page_remove_rmap(old_page, vma);
1948 }
1949
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 /* Free the old page.. */
1951 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001952 ret |= VM_FAULT_WRITE;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001953 } else
1954 mem_cgroup_uncharge_page(new_page);
1955
Hugh Dickins920fc352005-11-21 21:32:17 -08001956 if (new_page)
1957 page_cache_release(new_page);
1958 if (old_page)
1959 page_cache_release(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001960unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001961 pte_unmap_unlock(page_table, ptl);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001962 if (dirty_page) {
Anton Salikhmetov8f7b3d12008-01-23 02:21:18 +03001963 if (vma->vm_file)
1964 file_update_time(vma->vm_file);
1965
Nick Piggin79352892007-07-19 01:47:22 -07001966 /*
1967 * Yes, Virginia, this is actually required to prevent a race
1968 * with clear_page_dirty_for_io() from clearing the page dirty
1969 * bit after it clear all dirty ptes, but before a racing
1970 * do_wp_page installs a dirty pte.
1971 *
1972 * do_no_page is protected similarly.
1973 */
1974 wait_on_page_locked(dirty_page);
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001975 set_page_dirty_balance(dirty_page, page_mkwrite);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001976 put_page(dirty_page);
1977 }
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001978 return ret;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001979oom_free_new:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08001980 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001981oom:
Hugh Dickins920fc352005-11-21 21:32:17 -08001982 if (old_page)
1983 page_cache_release(old_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 return VM_FAULT_OOM;
David Howells9637a5e2006-06-23 02:03:43 -07001985
1986unwritable_page:
1987 page_cache_release(old_page);
1988 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989}
1990
1991/*
1992 * Helper functions for unmap_mapping_range().
1993 *
1994 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1995 *
1996 * We have to restart searching the prio_tree whenever we drop the lock,
1997 * since the iterator is only valid while the lock is held, and anyway
1998 * a later vma might be split and reinserted earlier while lock dropped.
1999 *
2000 * The list of nonlinear vmas could be handled more efficiently, using
2001 * a placeholder, but handle it in the same way until a need is shown.
2002 * It is important to search the prio_tree before nonlinear list: a vma
2003 * may become nonlinear and be shifted from prio_tree to nonlinear list
2004 * while the lock is dropped; but never shifted from list to prio_tree.
2005 *
2006 * In order to make forward progress despite restarting the search,
2007 * vm_truncate_count is used to mark a vma as now dealt with, so we can
2008 * quickly skip it next time around. Since the prio_tree search only
2009 * shows us those vmas affected by unmapping the range in question, we
2010 * can't efficiently keep all vmas in step with mapping->truncate_count:
2011 * so instead reset them all whenever it wraps back to 0 (then go to 1).
2012 * mapping->truncate_count and vma->vm_truncate_count are protected by
2013 * i_mmap_lock.
2014 *
2015 * In order to make forward progress despite repeatedly restarting some
Hugh Dickinsee39b372005-04-19 13:29:15 -07002016 * large vma, note the restart_addr from unmap_vmas when it breaks out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 * and restart from that address when we reach that vma again. It might
2018 * have been split or merged, shrunk or extended, but never shifted: so
2019 * restart_addr remains valid so long as it remains in the vma's range.
2020 * unmap_mapping_range forces truncate_count to leap over page-aligned
2021 * values so we can save vma's restart_addr in its truncate_count field.
2022 */
2023#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
2024
2025static void reset_vma_truncate_counts(struct address_space *mapping)
2026{
2027 struct vm_area_struct *vma;
2028 struct prio_tree_iter iter;
2029
2030 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
2031 vma->vm_truncate_count = 0;
2032 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
2033 vma->vm_truncate_count = 0;
2034}
2035
2036static int unmap_mapping_range_vma(struct vm_area_struct *vma,
2037 unsigned long start_addr, unsigned long end_addr,
2038 struct zap_details *details)
2039{
2040 unsigned long restart_addr;
2041 int need_break;
2042
Nick Piggind00806b2007-07-19 01:46:57 -07002043 /*
2044 * files that support invalidating or truncating portions of the
Nick Piggind0217ac2007-07-19 01:47:03 -07002045 * file from under mmaped areas must have their ->fault function
Nick Piggin83c54072007-07-19 01:47:05 -07002046 * return a locked page (and set VM_FAULT_LOCKED in the return).
2047 * This provides synchronisation against concurrent unmapping here.
Nick Piggind00806b2007-07-19 01:46:57 -07002048 */
Nick Piggind00806b2007-07-19 01:46:57 -07002049
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050again:
2051 restart_addr = vma->vm_truncate_count;
2052 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
2053 start_addr = restart_addr;
2054 if (start_addr >= end_addr) {
2055 /* Top of vma has been split off since last time */
2056 vma->vm_truncate_count = details->truncate_count;
2057 return 0;
2058 }
2059 }
2060
Hugh Dickinsee39b372005-04-19 13:29:15 -07002061 restart_addr = zap_page_range(vma, start_addr,
2062 end_addr - start_addr, details);
Nick Piggin95c354f2008-01-30 13:31:20 +01002063 need_break = need_resched() || spin_needbreak(details->i_mmap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064
Hugh Dickinsee39b372005-04-19 13:29:15 -07002065 if (restart_addr >= end_addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 /* We have now completed this vma: mark it so */
2067 vma->vm_truncate_count = details->truncate_count;
2068 if (!need_break)
2069 return 0;
2070 } else {
2071 /* Note restart_addr in vma's truncate_count field */
Hugh Dickinsee39b372005-04-19 13:29:15 -07002072 vma->vm_truncate_count = restart_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 if (!need_break)
2074 goto again;
2075 }
2076
2077 spin_unlock(details->i_mmap_lock);
2078 cond_resched();
2079 spin_lock(details->i_mmap_lock);
2080 return -EINTR;
2081}
2082
2083static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
2084 struct zap_details *details)
2085{
2086 struct vm_area_struct *vma;
2087 struct prio_tree_iter iter;
2088 pgoff_t vba, vea, zba, zea;
2089
2090restart:
2091 vma_prio_tree_foreach(vma, &iter, root,
2092 details->first_index, details->last_index) {
2093 /* Skip quickly over those we have already dealt with */
2094 if (vma->vm_truncate_count == details->truncate_count)
2095 continue;
2096
2097 vba = vma->vm_pgoff;
2098 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
2099 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
2100 zba = details->first_index;
2101 if (zba < vba)
2102 zba = vba;
2103 zea = details->last_index;
2104 if (zea > vea)
2105 zea = vea;
2106
2107 if (unmap_mapping_range_vma(vma,
2108 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
2109 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
2110 details) < 0)
2111 goto restart;
2112 }
2113}
2114
2115static inline void unmap_mapping_range_list(struct list_head *head,
2116 struct zap_details *details)
2117{
2118 struct vm_area_struct *vma;
2119
2120 /*
2121 * In nonlinear VMAs there is no correspondence between virtual address
2122 * offset and file offset. So we must perform an exhaustive search
2123 * across *all* the pages in each nonlinear VMA, not just the pages
2124 * whose virtual address lies outside the file truncation point.
2125 */
2126restart:
2127 list_for_each_entry(vma, head, shared.vm_set.list) {
2128 /* Skip quickly over those we have already dealt with */
2129 if (vma->vm_truncate_count == details->truncate_count)
2130 continue;
2131 details->nonlinear_vma = vma;
2132 if (unmap_mapping_range_vma(vma, vma->vm_start,
2133 vma->vm_end, details) < 0)
2134 goto restart;
2135 }
2136}
2137
2138/**
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002139 * 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 -07002140 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 * @holebegin: byte in first page to unmap, relative to the start of
2142 * the underlying file. This will be rounded down to a PAGE_SIZE
2143 * boundary. Note that this is different from vmtruncate(), which
2144 * must keep the partial page. In contrast, we must get rid of
2145 * partial pages.
2146 * @holelen: size of prospective hole in bytes. This will be rounded
2147 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2148 * end of the file.
2149 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2150 * but 0 when invalidating pagecache, don't throw away private data.
2151 */
2152void unmap_mapping_range(struct address_space *mapping,
2153 loff_t const holebegin, loff_t const holelen, int even_cows)
2154{
2155 struct zap_details details;
2156 pgoff_t hba = holebegin >> PAGE_SHIFT;
2157 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2158
2159 /* Check for overflow. */
2160 if (sizeof(holelen) > sizeof(hlen)) {
2161 long long holeend =
2162 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2163 if (holeend & ~(long long)ULONG_MAX)
2164 hlen = ULONG_MAX - hba + 1;
2165 }
2166
2167 details.check_mapping = even_cows? NULL: mapping;
2168 details.nonlinear_vma = NULL;
2169 details.first_index = hba;
2170 details.last_index = hba + hlen - 1;
2171 if (details.last_index < details.first_index)
2172 details.last_index = ULONG_MAX;
2173 details.i_mmap_lock = &mapping->i_mmap_lock;
2174
2175 spin_lock(&mapping->i_mmap_lock);
2176
Nick Piggind00806b2007-07-19 01:46:57 -07002177 /* Protect against endless unmapping loops */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 mapping->truncate_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 if (unlikely(is_restart_addr(mapping->truncate_count))) {
2180 if (mapping->truncate_count == 0)
2181 reset_vma_truncate_counts(mapping);
2182 mapping->truncate_count++;
2183 }
2184 details.truncate_count = mapping->truncate_count;
2185
2186 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
2187 unmap_mapping_range_tree(&mapping->i_mmap, &details);
2188 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
2189 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
2190 spin_unlock(&mapping->i_mmap_lock);
2191}
2192EXPORT_SYMBOL(unmap_mapping_range);
2193
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07002194/**
2195 * vmtruncate - unmap mappings "freed" by truncate() syscall
2196 * @inode: inode of the file used
2197 * @offset: file offset to start truncating
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 *
2199 * NOTE! We have to be ready to update the memory sharing
2200 * between the file and the memory map for a potential last
2201 * incomplete page. Ugly, but necessary.
2202 */
2203int vmtruncate(struct inode * inode, loff_t offset)
2204{
Christoph Hellwig61d50482008-02-04 22:28:56 -08002205 if (inode->i_size < offset) {
2206 unsigned long limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207
Christoph Hellwig61d50482008-02-04 22:28:56 -08002208 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
2209 if (limit != RLIM_INFINITY && offset > limit)
2210 goto out_sig;
2211 if (offset > inode->i_sb->s_maxbytes)
2212 goto out_big;
2213 i_size_write(inode, offset);
2214 } else {
2215 struct address_space *mapping = inode->i_mapping;
Nick Piggind00806b2007-07-19 01:46:57 -07002216
Christoph Hellwig61d50482008-02-04 22:28:56 -08002217 /*
2218 * truncation of in-use swapfiles is disallowed - it would
2219 * cause subsequent swapout to scribble on the now-freed
2220 * blocks.
2221 */
2222 if (IS_SWAPFILE(inode))
2223 return -ETXTBSY;
2224 i_size_write(inode, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225
Christoph Hellwig61d50482008-02-04 22:28:56 -08002226 /*
2227 * unmap_mapping_range is called twice, first simply for
2228 * efficiency so that truncate_inode_pages does fewer
2229 * single-page unmaps. However after this first call, and
2230 * before truncate_inode_pages finishes, it is possible for
2231 * private pages to be COWed, which remain after
2232 * truncate_inode_pages finishes, hence the second
2233 * unmap_mapping_range call must be made for correctness.
2234 */
2235 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2236 truncate_inode_pages(mapping, offset);
2237 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2238 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 if (inode->i_op && inode->i_op->truncate)
2241 inode->i_op->truncate(inode);
2242 return 0;
Christoph Hellwig61d50482008-02-04 22:28:56 -08002243
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244out_sig:
2245 send_sig(SIGXFSZ, current, 0);
2246out_big:
2247 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249EXPORT_SYMBOL(vmtruncate);
2250
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002251int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
2252{
2253 struct address_space *mapping = inode->i_mapping;
2254
2255 /*
2256 * If the underlying filesystem is not going to provide
2257 * a way to truncate a range of blocks (punch a hole) -
2258 * we should return failure right now.
2259 */
2260 if (!inode->i_op || !inode->i_op->truncate_range)
2261 return -ENOSYS;
2262
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002263 mutex_lock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002264 down_write(&inode->i_alloc_sem);
2265 unmap_mapping_range(mapping, offset, (end - offset), 1);
2266 truncate_inode_pages_range(mapping, offset, end);
Nick Piggind00806b2007-07-19 01:46:57 -07002267 unmap_mapping_range(mapping, offset, (end - offset), 1);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002268 inode->i_op->truncate_range(inode, offset, end);
2269 up_write(&inode->i_alloc_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002270 mutex_unlock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002271
2272 return 0;
2273}
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002274
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002276 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2277 * but allow concurrent faults), and pte mapped but not yet locked.
2278 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002280static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2281 unsigned long address, pte_t *page_table, pmd_t *pmd,
2282 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002284 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 struct page *page;
Hugh Dickins65500d22005-10-29 18:15:59 -07002286 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 pte_t pte;
Nick Piggin83c54072007-07-19 01:47:05 -07002288 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002290 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002291 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002292
2293 entry = pte_to_swp_entry(orig_pte);
Christoph Lameter06972122006-06-23 02:03:35 -07002294 if (is_migration_entry(entry)) {
2295 migration_entry_wait(mm, pmd, address);
2296 goto out;
2297 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002298 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 page = lookup_swap_cache(entry);
2300 if (!page) {
Ashwin Chaugule098fe652006-12-06 20:31:54 -08002301 grab_swap_token(); /* Contend for token _before_ read-in */
Hugh Dickins02098fe2008-02-04 22:28:42 -08002302 page = swapin_readahead(entry,
2303 GFP_HIGHUSER_MOVABLE, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 if (!page) {
2305 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002306 * Back out if somebody else faulted in this pte
2307 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002309 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310 if (likely(pte_same(*page_table, orig_pte)))
2311 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002312 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002313 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 }
2315
2316 /* Had to read the page from swap area: Major fault */
2317 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002318 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 }
2320
2321 mark_page_accessed(page);
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002322
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 lock_page(page);
Balbir Singh20a10222007-11-14 17:00:33 -08002324 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325
KAMEZAWA Hiroyuki073e5872008-10-18 20:28:08 -07002326 if (mem_cgroup_charge(page, mm, GFP_KERNEL)) {
2327 ret = VM_FAULT_OOM;
2328 unlock_page(page);
2329 goto out;
2330 }
2331
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002333 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002335 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002336 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002337 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002338
2339 if (unlikely(!PageUptodate(page))) {
2340 ret = VM_FAULT_SIGBUS;
2341 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 }
2343
2344 /* The page isn't present yet, go ahead with the fault. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345
Hugh Dickins42946212005-10-29 18:16:05 -07002346 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 pte = mk_pte(page, vma->vm_page_prot);
2348 if (write_access && can_share_swap_page(page)) {
2349 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2350 write_access = 0;
2351 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352
2353 flush_icache_page(vma, page);
2354 set_pte_at(mm, address, page_table, pte);
2355 page_add_anon_rmap(page, vma, address);
2356
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002357 swap_free(entry);
Nick Pigginb291f002008-10-18 20:26:44 -07002358 if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002359 remove_exclusive_swap_page(page);
2360 unlock_page(page);
2361
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 if (write_access) {
Hugh Dickins61469f12008-03-04 14:29:04 -08002363 ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
2364 if (ret & VM_FAULT_ERROR)
2365 ret &= VM_FAULT_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366 goto out;
2367 }
2368
2369 /* No need to invalidate - it was non-present before */
2370 update_mmu_cache(vma, address, pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07002371unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002372 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373out:
2374 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002375out_nomap:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002376 mem_cgroup_uncharge_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002377 pte_unmap_unlock(page_table, ptl);
Kirill Korotaevb8107482005-05-16 21:53:50 -07002378 unlock_page(page);
2379 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002380 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381}
2382
2383/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002384 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2385 * but allow concurrent faults), and pte mapped but not yet locked.
2386 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002388static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2389 unsigned long address, pte_t *page_table, pmd_t *pmd,
2390 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002392 struct page *page;
2393 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395
Nick Piggin557ed1f2007-10-16 01:24:40 -07002396 /* Allocate our own private page. */
2397 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398
Nick Piggin557ed1f2007-10-16 01:24:40 -07002399 if (unlikely(anon_vma_prepare(vma)))
2400 goto oom;
2401 page = alloc_zeroed_user_highpage_movable(vma, address);
2402 if (!page)
2403 goto oom;
Nick Piggin0ed361d2008-02-04 22:29:34 -08002404 __SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002406 if (mem_cgroup_charge(page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002407 goto oom_free_page;
2408
Nick Piggin557ed1f2007-10-16 01:24:40 -07002409 entry = mk_pte(page, vma->vm_page_prot);
2410 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002411
Nick Piggin557ed1f2007-10-16 01:24:40 -07002412 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2413 if (!pte_none(*page_table))
2414 goto release;
2415 inc_mm_counter(mm, anon_rss);
Rik van Rielb2e18532008-10-18 20:26:30 -07002416 SetPageSwapBacked(page);
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002417 lru_cache_add_active_or_unevictable(page, vma);
Nick Piggin557ed1f2007-10-16 01:24:40 -07002418 page_add_new_anon_rmap(page, vma, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002419 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420
2421 /* No need to invalidate - it was non-present before */
Hugh Dickins65500d22005-10-29 18:15:59 -07002422 update_mmu_cache(vma, address, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002423unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002424 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002425 return 0;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002426release:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002427 mem_cgroup_uncharge_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002428 page_cache_release(page);
2429 goto unlock;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002430oom_free_page:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002431 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002432oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433 return VM_FAULT_OOM;
2434}
2435
2436/*
Nick Piggin54cb8822007-07-19 01:46:59 -07002437 * __do_fault() tries to create a new page mapping. It aggressively
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 * tries to share with existing pages, but makes a separate copy if
Nick Piggin54cb8822007-07-19 01:46:59 -07002439 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2440 * the next page fault.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441 *
2442 * As this is called only for pages that do not currently exist, we
2443 * do not need to flush old virtual caches or the TLB.
2444 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002445 * We enter with non-exclusive mmap_sem (to exclude vma changes,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002446 * but allow concurrent faults), and pte neither mapped nor locked.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002447 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448 */
Nick Piggin54cb8822007-07-19 01:46:59 -07002449static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002450 unsigned long address, pmd_t *pmd,
Nick Piggin54cb8822007-07-19 01:46:59 -07002451 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452{
Hugh Dickins16abfa02007-10-04 16:56:06 +01002453 pte_t *page_table;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002454 spinlock_t *ptl;
Nick Piggind0217ac2007-07-19 01:47:03 -07002455 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 int anon = 0;
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002458 int charged = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002459 struct page *dirty_page = NULL;
Nick Piggind0217ac2007-07-19 01:47:03 -07002460 struct vm_fault vmf;
2461 int ret;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002462 int page_mkwrite = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07002463
Nick Piggind0217ac2007-07-19 01:47:03 -07002464 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2465 vmf.pgoff = pgoff;
2466 vmf.flags = flags;
2467 vmf.page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468
Nick Piggin3c18ddd2008-04-28 02:12:10 -07002469 ret = vma->vm_ops->fault(vma, &vmf);
2470 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2471 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472
Nick Piggind00806b2007-07-19 01:46:57 -07002473 /*
Nick Piggind0217ac2007-07-19 01:47:03 -07002474 * For consistency in subsequent calls, make the faulted page always
Nick Piggind00806b2007-07-19 01:46:57 -07002475 * locked.
2476 */
Nick Piggin83c54072007-07-19 01:47:05 -07002477 if (unlikely(!(ret & VM_FAULT_LOCKED)))
Nick Piggind0217ac2007-07-19 01:47:03 -07002478 lock_page(vmf.page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002479 else
Nick Piggind0217ac2007-07-19 01:47:03 -07002480 VM_BUG_ON(!PageLocked(vmf.page));
Nick Piggind00806b2007-07-19 01:46:57 -07002481
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 /*
2483 * Should we do an early C-O-W break?
2484 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002485 page = vmf.page;
Nick Piggin54cb8822007-07-19 01:46:59 -07002486 if (flags & FAULT_FLAG_WRITE) {
David Howells9637a5e2006-06-23 02:03:43 -07002487 if (!(vma->vm_flags & VM_SHARED)) {
Nick Piggin54cb8822007-07-19 01:46:59 -07002488 anon = 1;
Nick Piggind00806b2007-07-19 01:46:57 -07002489 if (unlikely(anon_vma_prepare(vma))) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002490 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002491 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002492 }
Nick Piggin83c54072007-07-19 01:47:05 -07002493 page = alloc_page_vma(GFP_HIGHUSER_MOVABLE,
2494 vma, address);
Nick Piggind00806b2007-07-19 01:46:57 -07002495 if (!page) {
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 }
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002499 if (mem_cgroup_charge(page, mm, GFP_KERNEL)) {
2500 ret = VM_FAULT_OOM;
2501 page_cache_release(page);
2502 goto out;
2503 }
2504 charged = 1;
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
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002545 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546
2547 /*
2548 * This silly early PAGE_DIRTY setting removes a race
2549 * due to the bad i386 page protection. But it's valid
2550 * for other architectures too.
2551 *
2552 * Note that if write_access is true, we either now have
2553 * an exclusive copy of the page, or this is a shared mapping,
2554 * so we can make it writable and dirty to avoid having to
2555 * handle that later.
2556 */
2557 /* Only go through if we didn't race with anybody else... */
Nick Piggin54cb8822007-07-19 01:46:59 -07002558 if (likely(pte_same(*page_table, orig_pte))) {
Nick Piggind00806b2007-07-19 01:46:57 -07002559 flush_icache_page(vma, page);
2560 entry = mk_pte(page, vma->vm_page_prot);
Nick Piggin54cb8822007-07-19 01:46:59 -07002561 if (flags & FAULT_FLAG_WRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 if (anon) {
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002564 inc_mm_counter(mm, anon_rss);
Rik van Rielb2e18532008-10-18 20:26:30 -07002565 SetPageSwapBacked(page);
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002566 lru_cache_add_active_or_unevictable(page, vma);
2567 page_add_new_anon_rmap(page, vma, address);
Hugh Dickinsf57e88a2005-11-21 21:32:19 -08002568 } else {
Hugh Dickins42946212005-10-29 18:16:05 -07002569 inc_mm_counter(mm, file_rss);
Nick Piggind00806b2007-07-19 01:46:57 -07002570 page_add_file_rmap(page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002571 if (flags & FAULT_FLAG_WRITE) {
Nick Piggind00806b2007-07-19 01:46:57 -07002572 dirty_page = page;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002573 get_page(dirty_page);
2574 }
Hugh Dickins42946212005-10-29 18:16:05 -07002575 }
Lee Schermerhorn64d65192008-10-18 20:26:52 -07002576//TODO: is this safe? do_anonymous_page() does it this way.
2577 set_pte_at(mm, address, page_table, entry);
Nick Piggind00806b2007-07-19 01:46:57 -07002578
2579 /* no need to invalidate: a not-present page won't be cached */
2580 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 } else {
KAMEZAWA Hiroyuki5b4e6552008-10-18 20:28:10 -07002582 if (charged)
2583 mem_cgroup_uncharge_page(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002584 if (anon)
2585 page_cache_release(page);
2586 else
Nick Piggin54cb8822007-07-19 01:46:59 -07002587 anon = 1; /* no anon but release faulted_page */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588 }
2589
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002590 pte_unmap_unlock(page_table, ptl);
Nick Piggind00806b2007-07-19 01:46:57 -07002591
2592out:
Nick Piggind0217ac2007-07-19 01:47:03 -07002593 unlock_page(vmf.page);
Mark Fasheh69676142007-07-19 01:47:00 -07002594out_unlocked:
Nick Piggind00806b2007-07-19 01:46:57 -07002595 if (anon)
Nick Piggind0217ac2007-07-19 01:47:03 -07002596 page_cache_release(vmf.page);
Nick Piggind00806b2007-07-19 01:46:57 -07002597 else if (dirty_page) {
Anton Salikhmetov8f7b3d12008-01-23 02:21:18 +03002598 if (vma->vm_file)
2599 file_update_time(vma->vm_file);
2600
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002601 set_page_dirty_balance(dirty_page, page_mkwrite);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002602 put_page(dirty_page);
2603 }
Nick Piggind00806b2007-07-19 01:46:57 -07002604
Nick Piggin83c54072007-07-19 01:47:05 -07002605 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07002606}
Nick Piggind00806b2007-07-19 01:46:57 -07002607
Nick Piggin54cb8822007-07-19 01:46:59 -07002608static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2609 unsigned long address, pte_t *page_table, pmd_t *pmd,
2610 int write_access, pte_t orig_pte)
2611{
2612 pgoff_t pgoff = (((address & PAGE_MASK)
Dean Nelson0da7e012007-10-16 01:24:45 -07002613 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
Nick Piggin54cb8822007-07-19 01:46:59 -07002614 unsigned int flags = (write_access ? FAULT_FLAG_WRITE : 0);
2615
Hugh Dickins16abfa02007-10-04 16:56:06 +01002616 pte_unmap(page_table);
2617 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Nick Piggin54cb8822007-07-19 01:46:59 -07002618}
2619
Jes Sorensenf4b81802006-09-27 01:50:10 -07002620/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 * Fault of a previously existing named mapping. Repopulate the pte
2622 * from the encoded file_pte if possible. This enables swappable
2623 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002624 *
2625 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2626 * but allow concurrent faults), and pte mapped but not yet locked.
2627 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002629static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins65500d22005-10-29 18:15:59 -07002630 unsigned long address, pte_t *page_table, pmd_t *pmd,
2631 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632{
Nick Piggind0217ac2007-07-19 01:47:03 -07002633 unsigned int flags = FAULT_FLAG_NONLINEAR |
2634 (write_access ? FAULT_FLAG_WRITE : 0);
Hugh Dickins65500d22005-10-29 18:15:59 -07002635 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002637 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Nick Piggin83c54072007-07-19 01:47:05 -07002638 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639
Nick Piggind0217ac2007-07-19 01:47:03 -07002640 if (unlikely(!(vma->vm_flags & VM_NONLINEAR) ||
2641 !(vma->vm_flags & VM_CAN_NONLINEAR))) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002642 /*
2643 * Page table corrupted: show pte and kill process.
2644 */
Nick Pigginb5810032005-10-29 18:16:12 -07002645 print_bad_pte(vma, orig_pte, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002646 return VM_FAULT_OOM;
2647 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002648
2649 pgoff = pte_to_pgoff(orig_pte);
Hugh Dickins16abfa02007-10-04 16:56:06 +01002650 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651}
2652
2653/*
2654 * These routines also need to handle stuff like marking pages dirty
2655 * and/or accessed for architectures that don't do it in hardware (most
2656 * RISC architectures). The early dirtying is also good on the i386.
2657 *
2658 * There is also a hook called "update_mmu_cache()" that architectures
2659 * with external mmu caches can use to update those (ie the Sparc or
2660 * PowerPC hashed page tables that act as extended TLBs).
2661 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07002662 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2663 * but allow concurrent faults), and pte mapped but not yet locked.
2664 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665 */
2666static inline int handle_pte_fault(struct mm_struct *mm,
Hugh Dickins65500d22005-10-29 18:15:59 -07002667 struct vm_area_struct *vma, unsigned long address,
2668 pte_t *pte, pmd_t *pmd, int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669{
2670 pte_t entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002671 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002673 entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002675 if (pte_none(entry)) {
Jes Sorensenf4b81802006-09-27 01:50:10 -07002676 if (vma->vm_ops) {
Nick Piggin3c18ddd2008-04-28 02:12:10 -07002677 if (likely(vma->vm_ops->fault))
Nick Piggin54cb8822007-07-19 01:46:59 -07002678 return do_linear_fault(mm, vma, address,
2679 pte, pmd, write_access, entry);
Jes Sorensenf4b81802006-09-27 01:50:10 -07002680 }
2681 return do_anonymous_page(mm, vma, address,
2682 pte, pmd, write_access);
Hugh Dickins65500d22005-10-29 18:15:59 -07002683 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684 if (pte_file(entry))
Nick Piggind0217ac2007-07-19 01:47:03 -07002685 return do_nonlinear_fault(mm, vma, address,
Hugh Dickins65500d22005-10-29 18:15:59 -07002686 pte, pmd, write_access, entry);
2687 return do_swap_page(mm, vma, address,
2688 pte, pmd, write_access, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689 }
2690
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002691 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002692 spin_lock(ptl);
2693 if (unlikely(!pte_same(*pte, entry)))
2694 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695 if (write_access) {
2696 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002697 return do_wp_page(mm, vma, address,
2698 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699 entry = pte_mkdirty(entry);
2700 }
2701 entry = pte_mkyoung(entry);
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002702 if (ptep_set_access_flags(vma, address, pte, entry, write_access)) {
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002703 update_mmu_cache(vma, address, entry);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002704 } else {
2705 /*
2706 * This is needed only for protection faults but the arch code
2707 * is not yet telling us if this is a protection fault or not.
2708 * This still avoids useless tlb flushes for .text page faults
2709 * with threads.
2710 */
2711 if (write_access)
2712 flush_tlb_page(vma, address);
2713 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002714unlock:
2715 pte_unmap_unlock(pte, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002716 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717}
2718
2719/*
2720 * By the time we get here, we already hold the mm semaphore
2721 */
Nick Piggin83c54072007-07-19 01:47:05 -07002722int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723 unsigned long address, int write_access)
2724{
2725 pgd_t *pgd;
2726 pud_t *pud;
2727 pmd_t *pmd;
2728 pte_t *pte;
2729
2730 __set_current_state(TASK_RUNNING);
2731
Christoph Lameterf8891e52006-06-30 01:55:45 -07002732 count_vm_event(PGFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01002734 if (unlikely(is_vm_hugetlb_page(vma)))
2735 return hugetlb_fault(mm, vma, address, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738 pud = pud_alloc(mm, pgd, address);
2739 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002740 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002741 pmd = pmd_alloc(mm, pud, address);
2742 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002743 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 pte = pte_alloc_map(mm, pmd, address);
2745 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002746 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747
Hugh Dickinsc74df322005-10-29 18:16:23 -07002748 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749}
2750
2751#ifndef __PAGETABLE_PUD_FOLDED
2752/*
2753 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002754 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002756int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002758 pud_t *new = pud_alloc_one(mm, address);
2759 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002760 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761
Nick Piggin362a61a2008-05-14 06:37:36 +02002762 smp_wmb(); /* See comment in __pte_alloc */
2763
Hugh Dickins872fec12005-10-29 18:16:21 -07002764 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002765 if (pgd_present(*pgd)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002766 pud_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002767 else
2768 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07002769 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002770 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771}
2772#endif /* __PAGETABLE_PUD_FOLDED */
2773
2774#ifndef __PAGETABLE_PMD_FOLDED
2775/*
2776 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002777 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002779int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002781 pmd_t *new = pmd_alloc_one(mm, address);
2782 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002783 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784
Nick Piggin362a61a2008-05-14 06:37:36 +02002785 smp_wmb(); /* See comment in __pte_alloc */
2786
Hugh Dickins872fec12005-10-29 18:16:21 -07002787 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07002789 if (pud_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002790 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002791 else
2792 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07002794 if (pgd_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002795 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002796 else
2797 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07002799 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002800 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801}
2802#endif /* __PAGETABLE_PMD_FOLDED */
2803
2804int make_pages_present(unsigned long addr, unsigned long end)
2805{
2806 int ret, len, write;
2807 struct vm_area_struct * vma;
2808
2809 vma = find_vma(current->mm, addr);
2810 if (!vma)
KOSAKI Motohiroa4770972008-08-04 13:41:14 -07002811 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812 write = (vma->vm_flags & VM_WRITE) != 0;
Eric Sesterhenn5bcb28b2006-03-26 18:30:52 +02002813 BUG_ON(addr >= end);
2814 BUG_ON(end > vma->vm_end);
Rolf Eike Beer68e116a2007-07-15 23:38:03 -07002815 len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 ret = get_user_pages(current, current->mm, addr,
2817 len, write, 0, NULL, NULL);
Lee Schermerhornc11d69d2008-10-18 20:26:56 -07002818 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819 return ret;
Lee Schermerhorn9978ad52008-10-18 20:26:56 -07002820 return ret == len ? 0 : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821}
2822
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823#if !defined(__HAVE_ARCH_GATE_AREA)
2824
2825#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07002826static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827
2828static int __init gate_vma_init(void)
2829{
2830 gate_vma.vm_mm = NULL;
2831 gate_vma.vm_start = FIXADDR_USER_START;
2832 gate_vma.vm_end = FIXADDR_USER_END;
Roland McGrathb6558c42007-01-26 00:56:47 -08002833 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
2834 gate_vma.vm_page_prot = __P101;
Roland McGrathf47aef52007-01-26 00:56:49 -08002835 /*
2836 * Make sure the vDSO gets into every core dump.
2837 * Dumping its contents makes post-mortem fully interpretable later
2838 * without matching up the same kernel and hardware config to see
2839 * what PC values meant.
2840 */
2841 gate_vma.vm_flags |= VM_ALWAYSDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842 return 0;
2843}
2844__initcall(gate_vma_init);
2845#endif
2846
2847struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2848{
2849#ifdef AT_SYSINFO_EHDR
2850 return &gate_vma;
2851#else
2852 return NULL;
2853#endif
2854}
2855
2856int in_gate_area_no_task(unsigned long addr)
2857{
2858#ifdef AT_SYSINFO_EHDR
2859 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2860 return 1;
2861#endif
2862 return 0;
2863}
2864
2865#endif /* __HAVE_ARCH_GATE_AREA */
David Howells0ec76a12006-09-27 01:50:15 -07002866
Rik van Riel28b2ee22008-07-23 21:27:05 -07002867#ifdef CONFIG_HAVE_IOREMAP_PROT
2868static resource_size_t follow_phys(struct vm_area_struct *vma,
2869 unsigned long address, unsigned int flags,
2870 unsigned long *prot)
2871{
2872 pgd_t *pgd;
2873 pud_t *pud;
2874 pmd_t *pmd;
2875 pte_t *ptep, pte;
2876 spinlock_t *ptl;
2877 resource_size_t phys_addr = 0;
2878 struct mm_struct *mm = vma->vm_mm;
2879
2880 VM_BUG_ON(!(vma->vm_flags & (VM_IO | VM_PFNMAP)));
2881
2882 pgd = pgd_offset(mm, address);
2883 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
2884 goto no_page_table;
2885
2886 pud = pud_offset(pgd, address);
2887 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
2888 goto no_page_table;
2889
2890 pmd = pmd_offset(pud, address);
2891 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
2892 goto no_page_table;
2893
2894 /* We cannot handle huge page PFN maps. Luckily they don't exist. */
2895 if (pmd_huge(*pmd))
2896 goto no_page_table;
2897
2898 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
2899 if (!ptep)
2900 goto out;
2901
2902 pte = *ptep;
2903 if (!pte_present(pte))
2904 goto unlock;
2905 if ((flags & FOLL_WRITE) && !pte_write(pte))
2906 goto unlock;
2907 phys_addr = pte_pfn(pte);
2908 phys_addr <<= PAGE_SHIFT; /* Shift here to avoid overflow on PAE */
2909
2910 *prot = pgprot_val(pte_pgprot(pte));
2911
2912unlock:
2913 pte_unmap_unlock(ptep, ptl);
2914out:
2915 return phys_addr;
2916no_page_table:
2917 return 0;
2918}
2919
2920int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
2921 void *buf, int len, int write)
2922{
2923 resource_size_t phys_addr;
2924 unsigned long prot = 0;
2925 void *maddr;
2926 int offset = addr & (PAGE_SIZE-1);
2927
2928 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
2929 return -EINVAL;
2930
2931 phys_addr = follow_phys(vma, addr, write, &prot);
2932
2933 if (!phys_addr)
2934 return -EINVAL;
2935
2936 maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot);
2937 if (write)
2938 memcpy_toio(maddr + offset, buf, len);
2939 else
2940 memcpy_fromio(buf, maddr + offset, len);
2941 iounmap(maddr);
2942
2943 return len;
2944}
2945#endif
2946
David Howells0ec76a12006-09-27 01:50:15 -07002947/*
2948 * Access another process' address space.
2949 * Source/target buffer must be kernel space,
2950 * Do not walk the page table directly, use get_user_pages
2951 */
2952int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2953{
2954 struct mm_struct *mm;
2955 struct vm_area_struct *vma;
David Howells0ec76a12006-09-27 01:50:15 -07002956 void *old_buf = buf;
2957
2958 mm = get_task_mm(tsk);
2959 if (!mm)
2960 return 0;
2961
2962 down_read(&mm->mmap_sem);
Simon Arlott183ff222007-10-20 01:27:18 +02002963 /* ignore errors, just check how much was successfully transferred */
David Howells0ec76a12006-09-27 01:50:15 -07002964 while (len) {
2965 int bytes, ret, offset;
2966 void *maddr;
Rik van Riel28b2ee22008-07-23 21:27:05 -07002967 struct page *page = NULL;
David Howells0ec76a12006-09-27 01:50:15 -07002968
2969 ret = get_user_pages(tsk, mm, addr, 1,
2970 write, 1, &page, &vma);
Rik van Riel28b2ee22008-07-23 21:27:05 -07002971 if (ret <= 0) {
2972 /*
2973 * Check if this is a VM_IO | VM_PFNMAP VMA, which
2974 * we can access using slightly different code.
2975 */
2976#ifdef CONFIG_HAVE_IOREMAP_PROT
2977 vma = find_vma(mm, addr);
2978 if (!vma)
2979 break;
2980 if (vma->vm_ops && vma->vm_ops->access)
2981 ret = vma->vm_ops->access(vma, addr, buf,
2982 len, write);
2983 if (ret <= 0)
2984#endif
2985 break;
2986 bytes = ret;
David Howells0ec76a12006-09-27 01:50:15 -07002987 } else {
Rik van Riel28b2ee22008-07-23 21:27:05 -07002988 bytes = len;
2989 offset = addr & (PAGE_SIZE-1);
2990 if (bytes > PAGE_SIZE-offset)
2991 bytes = PAGE_SIZE-offset;
2992
2993 maddr = kmap(page);
2994 if (write) {
2995 copy_to_user_page(vma, page, addr,
2996 maddr + offset, buf, bytes);
2997 set_page_dirty_lock(page);
2998 } else {
2999 copy_from_user_page(vma, page, addr,
3000 buf, maddr + offset, bytes);
3001 }
3002 kunmap(page);
3003 page_cache_release(page);
David Howells0ec76a12006-09-27 01:50:15 -07003004 }
David Howells0ec76a12006-09-27 01:50:15 -07003005 len -= bytes;
3006 buf += bytes;
3007 addr += bytes;
3008 }
3009 up_read(&mm->mmap_sem);
3010 mmput(mm);
3011
3012 return buf - old_buf;
3013}
Andi Kleen03252912008-01-30 13:33:18 +01003014
3015/*
3016 * Print the name of a VMA.
3017 */
3018void print_vma_addr(char *prefix, unsigned long ip)
3019{
3020 struct mm_struct *mm = current->mm;
3021 struct vm_area_struct *vma;
3022
Ingo Molnare8bff742008-02-13 20:21:06 +01003023 /*
3024 * Do not print if we are in atomic
3025 * contexts (in exception stacks, etc.):
3026 */
3027 if (preempt_count())
3028 return;
3029
Andi Kleen03252912008-01-30 13:33:18 +01003030 down_read(&mm->mmap_sem);
3031 vma = find_vma(mm, ip);
3032 if (vma && vma->vm_file) {
3033 struct file *f = vma->vm_file;
3034 char *buf = (char *)__get_free_page(GFP_KERNEL);
3035 if (buf) {
3036 char *p, *s;
3037
Jan Blunckcf28b482008-02-14 19:38:44 -08003038 p = d_path(&f->f_path, buf, PAGE_SIZE);
Andi Kleen03252912008-01-30 13:33:18 +01003039 if (IS_ERR(p))
3040 p = "?";
3041 s = strrchr(p, '/');
3042 if (s)
3043 p = s+1;
3044 printk("%s%s[%lx+%lx]", prefix, p,
3045 vma->vm_start,
3046 vma->vm_end - vma->vm_start);
3047 free_page((unsigned long)buf);
3048 }
3049 }
3050 up_read(&current->mm->mmap_sem);
3051}