blob: 9aefaae4685812e12505551ade19ed6183220422 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/memory.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 */
6
7/*
8 * demand-loading started 01.12.91 - seems it is high on the list of
9 * things wanted, and it should be easy to implement. - Linus
10 */
11
12/*
13 * Ok, demand-loading was easy, shared pages a little bit tricker. Shared
14 * pages started 02.12.91, seems to work. - Linus.
15 *
16 * Tested sharing by executing about 30 /bin/sh: under the old kernel it
17 * would have taken more than the 6M I have free, but it worked well as
18 * far as I could see.
19 *
20 * Also corrected some "invalidate()"s - I wasn't doing enough of them.
21 */
22
23/*
24 * Real VM (paging to/from disk) started 18.12.91. Much more work and
25 * thought has to go into this. Oh, well..
26 * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why.
27 * Found it. Everything seems to work now.
28 * 20.12.91 - Ok, making the swap-device changeable like the root.
29 */
30
31/*
32 * 05.04.94 - Multi-page memory management added for v1.1.
33 * Idea by Alex Bligh (alex@cconcepts.co.uk)
34 *
35 * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG
36 * (Gerhard.Wichert@pdb.siemens.de)
37 *
38 * Aug/Sep 2004 Changed to four level page tables (Andi Kleen)
39 */
40
41#include <linux/kernel_stat.h>
42#include <linux/mm.h>
43#include <linux/hugetlb.h>
44#include <linux/mman.h>
45#include <linux/swap.h>
46#include <linux/highmem.h>
47#include <linux/pagemap.h>
48#include <linux/rmap.h>
49#include <linux/module.h>
Shailabh Nagar0ff92242006-07-14 00:24:37 -070050#include <linux/delayacct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include <linux/init.h>
Peter Zijlstraedc79b22006-09-25 23:30:58 -070052#include <linux/writeback.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080053#include <linux/memcontrol.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
55#include <asm/pgalloc.h>
56#include <asm/uaccess.h>
57#include <asm/tlb.h>
58#include <asm/tlbflush.h>
59#include <asm/pgtable.h>
60
61#include <linux/swapops.h>
62#include <linux/elf.h>
63
Andy Whitcroftd41dee32005-06-23 00:07:54 -070064#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -070065/* use the per-pgdat data instead for discontigmem - mbligh */
66unsigned long max_mapnr;
67struct page *mem_map;
68
69EXPORT_SYMBOL(max_mapnr);
70EXPORT_SYMBOL(mem_map);
71#endif
72
73unsigned long num_physpages;
74/*
75 * A number of key systems in x86 including ioremap() rely on the assumption
76 * that high_memory defines the upper bound on direct map memory, then end
77 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
78 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
79 * and ZONE_HIGHMEM.
80 */
81void * high_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
83EXPORT_SYMBOL(num_physpages);
84EXPORT_SYMBOL(high_memory);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
Ingo Molnar32a93232008-02-06 22:39:44 +010086/*
87 * Randomize the address space (stacks, mmaps, brk, etc.).
88 *
89 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
90 * as ancient (libc5 based) binaries can segfault. )
91 */
92int randomize_va_space __read_mostly =
93#ifdef CONFIG_COMPAT_BRK
94 1;
95#else
96 2;
97#endif
Andi Kleena62eaf12006-02-16 23:41:58 +010098
99static int __init disable_randmaps(char *s)
100{
101 randomize_va_space = 0;
OGAWA Hirofumi9b410462006-03-31 02:30:33 -0800102 return 1;
Andi Kleena62eaf12006-02-16 23:41:58 +0100103}
104__setup("norandmaps", disable_randmaps);
105
106
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107/*
108 * If a p?d_bad entry is found while walking page tables, report
109 * the error, before resetting entry to p?d_none. Usually (but
110 * very seldom) called out from the p?d_none_or_clear_bad macros.
111 */
112
113void pgd_clear_bad(pgd_t *pgd)
114{
115 pgd_ERROR(*pgd);
116 pgd_clear(pgd);
117}
118
119void pud_clear_bad(pud_t *pud)
120{
121 pud_ERROR(*pud);
122 pud_clear(pud);
123}
124
125void pmd_clear_bad(pmd_t *pmd)
126{
127 pmd_ERROR(*pmd);
128 pmd_clear(pmd);
129}
130
131/*
132 * Note: this doesn't free the actual pages themselves. That
133 * has been handled earlier when unmapping all the memory regions.
134 */
Hugh Dickinse0da3822005-04-19 13:29:15 -0700135static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800137 pgtable_t token = pmd_pgtable(*pmd);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700138 pmd_clear(pmd);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800139 pte_free_tlb(tlb, token);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700140 tlb->mm->nr_ptes--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141}
142
Hugh Dickinse0da3822005-04-19 13:29:15 -0700143static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
144 unsigned long addr, unsigned long end,
145 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146{
147 pmd_t *pmd;
148 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700149 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150
Hugh Dickinse0da3822005-04-19 13:29:15 -0700151 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 pmd = pmd_offset(pud, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 do {
154 next = pmd_addr_end(addr, end);
155 if (pmd_none_or_clear_bad(pmd))
156 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700157 free_pte_range(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 } while (pmd++, addr = next, addr != end);
159
Hugh Dickinse0da3822005-04-19 13:29:15 -0700160 start &= PUD_MASK;
161 if (start < floor)
162 return;
163 if (ceiling) {
164 ceiling &= PUD_MASK;
165 if (!ceiling)
166 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700168 if (end - 1 > ceiling - 1)
169 return;
170
171 pmd = pmd_offset(pud, start);
172 pud_clear(pud);
173 pmd_free_tlb(tlb, pmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174}
175
Hugh Dickinse0da3822005-04-19 13:29:15 -0700176static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
177 unsigned long addr, unsigned long end,
178 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179{
180 pud_t *pud;
181 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700182 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183
Hugh Dickinse0da3822005-04-19 13:29:15 -0700184 start = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 pud = pud_offset(pgd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 do {
187 next = pud_addr_end(addr, end);
188 if (pud_none_or_clear_bad(pud))
189 continue;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700190 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 } while (pud++, addr = next, addr != end);
192
Hugh Dickinse0da3822005-04-19 13:29:15 -0700193 start &= PGDIR_MASK;
194 if (start < floor)
195 return;
196 if (ceiling) {
197 ceiling &= PGDIR_MASK;
198 if (!ceiling)
199 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700201 if (end - 1 > ceiling - 1)
202 return;
203
204 pud = pud_offset(pgd, start);
205 pgd_clear(pgd);
206 pud_free_tlb(tlb, pud);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207}
208
209/*
Hugh Dickinse0da3822005-04-19 13:29:15 -0700210 * This function frees user-level page tables of a process.
211 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 * Must be called with pagetable lock held.
213 */
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700214void free_pgd_range(struct mmu_gather **tlb,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700215 unsigned long addr, unsigned long end,
216 unsigned long floor, unsigned long ceiling)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217{
218 pgd_t *pgd;
219 unsigned long next;
Hugh Dickinse0da3822005-04-19 13:29:15 -0700220 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221
Hugh Dickinse0da3822005-04-19 13:29:15 -0700222 /*
223 * The next few lines have given us lots of grief...
224 *
225 * Why are we testing PMD* at this top level? Because often
226 * there will be no work to do at all, and we'd prefer not to
227 * go all the way down to the bottom just to discover that.
228 *
229 * Why all these "- 1"s? Because 0 represents both the bottom
230 * of the address space and the top of it (using -1 for the
231 * top wouldn't help much: the masks would do the wrong thing).
232 * The rule is that addr 0 and floor 0 refer to the bottom of
233 * the address space, but end 0 and ceiling 0 refer to the top
234 * Comparisons need to use "end - 1" and "ceiling - 1" (though
235 * that end 0 case should be mythical).
236 *
237 * Wherever addr is brought up or ceiling brought down, we must
238 * be careful to reject "the opposite 0" before it confuses the
239 * subsequent tests. But what about where end is brought down
240 * by PMD_SIZE below? no, end can't go down to 0 there.
241 *
242 * Whereas we round start (addr) and ceiling down, by different
243 * masks at different levels, in order to test whether a table
244 * now has no other vmas using it, so can be freed, we don't
245 * bother to round floor or end up - the tests don't need that.
246 */
247
248 addr &= PMD_MASK;
249 if (addr < floor) {
250 addr += PMD_SIZE;
251 if (!addr)
252 return;
253 }
254 if (ceiling) {
255 ceiling &= PMD_MASK;
256 if (!ceiling)
257 return;
258 }
259 if (end - 1 > ceiling - 1)
260 end -= PMD_SIZE;
261 if (addr > end - 1)
262 return;
263
264 start = addr;
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700265 pgd = pgd_offset((*tlb)->mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 do {
267 next = pgd_addr_end(addr, end);
268 if (pgd_none_or_clear_bad(pgd))
269 continue;
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700270 free_pud_range(*tlb, pgd, addr, next, floor, ceiling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 } while (pgd++, addr = next, addr != end);
Hugh Dickinse0da3822005-04-19 13:29:15 -0700272}
273
274void free_pgtables(struct mmu_gather **tlb, struct vm_area_struct *vma,
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700275 unsigned long floor, unsigned long ceiling)
Hugh Dickinse0da3822005-04-19 13:29:15 -0700276{
277 while (vma) {
278 struct vm_area_struct *next = vma->vm_next;
279 unsigned long addr = vma->vm_start;
280
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700281 /*
282 * Hide vma from rmap and vmtruncate before freeing pgtables
283 */
284 anon_vma_unlink(vma);
285 unlink_file_vma(vma);
286
David Gibson9da61ae2006-03-22 00:08:57 -0800287 if (is_vm_hugetlb_page(vma)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700288 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
Hugh Dickinse0da3822005-04-19 13:29:15 -0700289 floor, next? next->vm_start: ceiling);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700290 } else {
291 /*
292 * Optimization: gather nearby vmas into one call down
293 */
294 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
David Gibson48669202006-03-22 00:08:58 -0800295 && !is_vm_hugetlb_page(next)) {
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700296 vma = next;
297 next = vma->vm_next;
Hugh Dickins8f4f8c12005-10-29 18:16:29 -0700298 anon_vma_unlink(vma);
299 unlink_file_vma(vma);
Hugh Dickins3bf5ee92005-04-19 13:29:16 -0700300 }
301 free_pgd_range(tlb, addr, vma->vm_end,
302 floor, next? next->vm_start: ceiling);
303 }
Hugh Dickinse0da3822005-04-19 13:29:15 -0700304 vma = next;
305 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306}
307
Hugh Dickins1bb36302005-10-29 18:16:22 -0700308int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309{
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800310 pgtable_t new = pte_alloc_one(mm, address);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700311 if (!new)
312 return -ENOMEM;
313
Nick Piggin362a61a2008-05-14 06:37:36 +0200314 /*
315 * Ensure all pte setup (eg. pte page lock and page clearing) are
316 * visible before the pte is made visible to other CPUs by being
317 * put into page tables.
318 *
319 * The other side of the story is the pointer chasing in the page
320 * table walking code (when walking the page table without locking;
321 * ie. most of the time). Fortunately, these data accesses consist
322 * of a chain of data-dependent loads, meaning most CPUs (alpha
323 * being the notable exception) will already guarantee loads are
324 * seen in-order. See the alpha page table accessors for the
325 * smp_read_barrier_depends() barriers in page table walking code.
326 */
327 smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
328
Hugh Dickinsc74df322005-10-29 18:16:23 -0700329 spin_lock(&mm->page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800330 if (!pmd_present(*pmd)) { /* Has another populated it ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 mm->nr_ptes++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 pmd_populate(mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800333 new = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 }
Hugh Dickinsc74df322005-10-29 18:16:23 -0700335 spin_unlock(&mm->page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800336 if (new)
337 pte_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700338 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339}
340
Hugh Dickins1bb36302005-10-29 18:16:22 -0700341int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342{
Hugh Dickins1bb36302005-10-29 18:16:22 -0700343 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
344 if (!new)
345 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346
Nick Piggin362a61a2008-05-14 06:37:36 +0200347 smp_wmb(); /* See comment in __pte_alloc */
348
Hugh Dickins1bb36302005-10-29 18:16:22 -0700349 spin_lock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800350 if (!pmd_present(*pmd)) { /* Has another populated it ? */
Hugh Dickins1bb36302005-10-29 18:16:22 -0700351 pmd_populate_kernel(&init_mm, pmd, new);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800352 new = NULL;
353 }
Hugh Dickins1bb36302005-10-29 18:16:22 -0700354 spin_unlock(&init_mm.page_table_lock);
Martin Schwidefsky2f569af2008-02-08 04:22:04 -0800355 if (new)
356 pte_free_kernel(&init_mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -0700357 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358}
359
Hugh Dickinsae859762005-10-29 18:16:05 -0700360static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
361{
362 if (file_rss)
363 add_mm_counter(mm, file_rss, file_rss);
364 if (anon_rss)
365 add_mm_counter(mm, anon_rss, anon_rss);
366}
367
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368/*
Linus Torvalds6aab3412005-11-28 14:34:23 -0800369 * This function is called to print an error when a bad pte
370 * is found. For example, we might have a PFN-mapped pte in
371 * a region that doesn't allow it.
Nick Pigginb5810032005-10-29 18:16:12 -0700372 *
373 * The calling function must still handle the error.
374 */
375void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr)
376{
377 printk(KERN_ERR "Bad pte = %08llx, process = %s, "
378 "vm_flags = %lx, vaddr = %lx\n",
379 (long long)pte_val(pte),
380 (vma->vm_mm == current->mm ? current->comm : "???"),
381 vma->vm_flags, vaddr);
382 dump_stack();
383}
384
Linus Torvalds67121172005-12-11 20:38:17 -0800385static inline int is_cow_mapping(unsigned int flags)
386{
387 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
388}
389
Nick Pigginb5810032005-10-29 18:16:12 -0700390/*
Nick Piggin7e675132008-04-28 02:13:00 -0700391 * vm_normal_page -- This function gets the "struct page" associated with a pte.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800392 *
Nick Piggin7e675132008-04-28 02:13:00 -0700393 * "Special" mappings do not wish to be associated with a "struct page" (either
394 * it doesn't exist, or it exists but they don't want to touch it). In this
395 * case, NULL is returned here. "Normal" mappings do have a struct page.
Jared Hulbertb379d792008-04-28 02:12:58 -0700396 *
Nick Piggin7e675132008-04-28 02:13:00 -0700397 * There are 2 broad cases. Firstly, an architecture may define a pte_special()
398 * pte bit, in which case this function is trivial. Secondly, an architecture
399 * may not have a spare pte bit, which requires a more complicated scheme,
400 * described below.
401 *
402 * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
403 * special mapping (even if there are underlying and valid "struct pages").
404 * COWed pages of a VM_PFNMAP are always normal.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800405 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700406 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
407 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
Nick Piggin7e675132008-04-28 02:13:00 -0700408 * set, and the vm_pgoff will point to the first PFN mapped: thus every special
409 * mapping will always honor the rule
Linus Torvalds6aab3412005-11-28 14:34:23 -0800410 *
411 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
412 *
Nick Piggin7e675132008-04-28 02:13:00 -0700413 * And for normal mappings this is false.
Jared Hulbertb379d792008-04-28 02:12:58 -0700414 *
Nick Piggin7e675132008-04-28 02:13:00 -0700415 * This restricts such mappings to be a linear translation from virtual address
416 * to pfn. To get around this restriction, we allow arbitrary mappings so long
417 * as the vma is not a COW mapping; in that case, we know that all ptes are
418 * special (because none can have been COWed).
Jared Hulbertb379d792008-04-28 02:12:58 -0700419 *
420 *
Nick Piggin7e675132008-04-28 02:13:00 -0700421 * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
422 *
Jared Hulbertb379d792008-04-28 02:12:58 -0700423 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
424 * page" backing, however the difference is that _all_ pages with a struct
425 * page (that is, those where pfn_valid is true) are refcounted and considered
426 * normal pages by the VM. The disadvantage is that pages are refcounted
427 * (which can be slower and simply not an option for some PFNMAP users). The
428 * advantage is that we don't have to follow the strict linearity rule of
429 * PFNMAP mappings in order to support COWable mappings.
430 *
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800431 */
Nick Piggin7e675132008-04-28 02:13:00 -0700432#ifdef __HAVE_ARCH_PTE_SPECIAL
433# define HAVE_PTE_SPECIAL 1
434#else
435# define HAVE_PTE_SPECIAL 0
436#endif
437struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
438 pte_t pte)
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800439{
Nick Piggin7e675132008-04-28 02:13:00 -0700440 unsigned long pfn;
441
442 if (HAVE_PTE_SPECIAL) {
443 if (likely(!pte_special(pte))) {
444 VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
445 return pte_page(pte);
446 }
447 VM_BUG_ON(!(vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)));
448 return NULL;
449 }
450
451 /* !HAVE_PTE_SPECIAL case follows: */
452
453 pfn = pte_pfn(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800454
Jared Hulbertb379d792008-04-28 02:12:58 -0700455 if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
456 if (vma->vm_flags & VM_MIXEDMAP) {
457 if (!pfn_valid(pfn))
458 return NULL;
459 goto out;
460 } else {
Nick Piggin7e675132008-04-28 02:13:00 -0700461 unsigned long off;
462 off = (addr - vma->vm_start) >> PAGE_SHIFT;
Jared Hulbertb379d792008-04-28 02:12:58 -0700463 if (pfn == vma->vm_pgoff + off)
464 return NULL;
465 if (!is_cow_mapping(vma->vm_flags))
466 return NULL;
467 }
Linus Torvalds6aab3412005-11-28 14:34:23 -0800468 }
469
Nick Piggin7e675132008-04-28 02:13:00 -0700470 VM_BUG_ON(!pfn_valid(pfn));
Linus Torvalds6aab3412005-11-28 14:34:23 -0800471
472 /*
Nick Piggin7e675132008-04-28 02:13:00 -0700473 * NOTE! We still have PageReserved() pages in the page tables.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800474 *
Nick Piggin7e675132008-04-28 02:13:00 -0700475 * eg. VDSO mappings can cause them to exist.
Linus Torvalds6aab3412005-11-28 14:34:23 -0800476 */
Jared Hulbertb379d792008-04-28 02:12:58 -0700477out:
Linus Torvalds6aab3412005-11-28 14:34:23 -0800478 return pfn_to_page(pfn);
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800479}
480
481/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 * copy one vm_area from one task to the other. Assumes the page tables
483 * already present in the new task to be cleared in the whole range
484 * covered by this vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 */
486
Hugh Dickins8c103762005-10-29 18:16:13 -0700487static inline void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
Nick Pigginb5810032005-10-29 18:16:12 -0700489 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
Hugh Dickins8c103762005-10-29 18:16:13 -0700490 unsigned long addr, int *rss)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491{
Nick Pigginb5810032005-10-29 18:16:12 -0700492 unsigned long vm_flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 pte_t pte = *src_pte;
494 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
496 /* pte contains position in swap or file, so copy. */
497 if (unlikely(!pte_present(pte))) {
498 if (!pte_file(pte)) {
Christoph Lameter06972122006-06-23 02:03:35 -0700499 swp_entry_t entry = pte_to_swp_entry(pte);
500
501 swap_duplicate(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 /* make sure dst_mm is on swapoff's mmlist. */
503 if (unlikely(list_empty(&dst_mm->mmlist))) {
504 spin_lock(&mmlist_lock);
Hugh Dickinsf412ac02005-10-29 18:16:41 -0700505 if (list_empty(&dst_mm->mmlist))
506 list_add(&dst_mm->mmlist,
507 &src_mm->mmlist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 spin_unlock(&mmlist_lock);
509 }
Christoph Lameter06972122006-06-23 02:03:35 -0700510 if (is_write_migration_entry(entry) &&
511 is_cow_mapping(vm_flags)) {
512 /*
513 * COW mappings require pages in both parent
514 * and child to be set to read.
515 */
516 make_migration_entry_read(&entry);
517 pte = swp_entry_to_pte(entry);
518 set_pte_at(src_mm, addr, src_pte, pte);
519 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700521 goto out_set_pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 }
523
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 /*
525 * If it's a COW mapping, write protect it both
526 * in the parent and the child
527 */
Linus Torvalds67121172005-12-11 20:38:17 -0800528 if (is_cow_mapping(vm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 ptep_set_wrprotect(src_mm, addr, src_pte);
Zachary Amsden3dc90792006-09-30 23:29:30 -0700530 pte = pte_wrprotect(pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 }
532
533 /*
534 * If it's a shared mapping, mark it clean in
535 * the child
536 */
537 if (vm_flags & VM_SHARED)
538 pte = pte_mkclean(pte);
539 pte = pte_mkold(pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800540
541 page = vm_normal_page(vma, addr, pte);
542 if (page) {
543 get_page(page);
Nick Pigginc97a9e12007-05-16 22:11:21 -0700544 page_dup_rmap(page, vma, addr);
Linus Torvalds6aab3412005-11-28 14:34:23 -0800545 rss[!!PageAnon(page)]++;
546 }
Hugh Dickinsae859762005-10-29 18:16:05 -0700547
548out_set_pte:
549 set_pte_at(dst_mm, addr, dst_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550}
551
552static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
553 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
554 unsigned long addr, unsigned long end)
555{
556 pte_t *src_pte, *dst_pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700557 spinlock_t *src_ptl, *dst_ptl;
Hugh Dickinse040f212005-10-29 18:15:53 -0700558 int progress = 0;
Hugh Dickins8c103762005-10-29 18:16:13 -0700559 int rss[2];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
561again:
Hugh Dickinsae859762005-10-29 18:16:05 -0700562 rss[1] = rss[0] = 0;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700563 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 if (!dst_pte)
565 return -ENOMEM;
566 src_pte = pte_offset_map_nested(src_pmd, addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700567 src_ptl = pte_lockptr(src_mm, src_pmd);
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700568 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700569 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 do {
572 /*
573 * We are holding two locks at this point - either of them
574 * could generate latencies in another task on another CPU.
575 */
Hugh Dickinse040f212005-10-29 18:15:53 -0700576 if (progress >= 32) {
577 progress = 0;
578 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +0100579 spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
Hugh Dickinse040f212005-10-29 18:15:53 -0700580 break;
581 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 if (pte_none(*src_pte)) {
583 progress++;
584 continue;
585 }
Hugh Dickins8c103762005-10-29 18:16:13 -0700586 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 progress += 8;
588 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700590 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -0700591 spin_unlock(src_ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 pte_unmap_nested(src_pte - 1);
Hugh Dickinsae859762005-10-29 18:16:05 -0700593 add_mm_rss(dst_mm, rss[0], rss[1]);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700594 pte_unmap_unlock(dst_pte - 1, dst_ptl);
595 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 if (addr != end)
597 goto again;
598 return 0;
599}
600
601static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
602 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
603 unsigned long addr, unsigned long end)
604{
605 pmd_t *src_pmd, *dst_pmd;
606 unsigned long next;
607
608 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
609 if (!dst_pmd)
610 return -ENOMEM;
611 src_pmd = pmd_offset(src_pud, addr);
612 do {
613 next = pmd_addr_end(addr, end);
614 if (pmd_none_or_clear_bad(src_pmd))
615 continue;
616 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
617 vma, addr, next))
618 return -ENOMEM;
619 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
620 return 0;
621}
622
623static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
624 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
625 unsigned long addr, unsigned long end)
626{
627 pud_t *src_pud, *dst_pud;
628 unsigned long next;
629
630 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
631 if (!dst_pud)
632 return -ENOMEM;
633 src_pud = pud_offset(src_pgd, addr);
634 do {
635 next = pud_addr_end(addr, end);
636 if (pud_none_or_clear_bad(src_pud))
637 continue;
638 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
639 vma, addr, next))
640 return -ENOMEM;
641 } while (dst_pud++, src_pud++, addr = next, addr != end);
642 return 0;
643}
644
645int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
646 struct vm_area_struct *vma)
647{
648 pgd_t *src_pgd, *dst_pgd;
649 unsigned long next;
650 unsigned long addr = vma->vm_start;
651 unsigned long end = vma->vm_end;
652
Nick Piggind9928952005-08-28 16:49:11 +1000653 /*
654 * Don't copy ptes where a page fault will fill them correctly.
655 * Fork becomes much lighter when there are big shared or private
656 * readonly mappings. The tradeoff is that copy_page_range is more
657 * efficient than faulting.
658 */
Linus Torvalds4d7672b2005-12-16 10:21:23 -0800659 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
Nick Piggind9928952005-08-28 16:49:11 +1000660 if (!vma->anon_vma)
661 return 0;
662 }
663
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 if (is_vm_hugetlb_page(vma))
665 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
666
667 dst_pgd = pgd_offset(dst_mm, addr);
668 src_pgd = pgd_offset(src_mm, addr);
669 do {
670 next = pgd_addr_end(addr, end);
671 if (pgd_none_or_clear_bad(src_pgd))
672 continue;
673 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
674 vma, addr, next))
675 return -ENOMEM;
676 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
677 return 0;
678}
679
Robin Holt51c6f662005-11-13 16:06:42 -0800680static unsigned long zap_pte_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700681 struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800683 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684{
Nick Pigginb5810032005-10-29 18:16:12 -0700685 struct mm_struct *mm = tlb->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 pte_t *pte;
Hugh Dickins508034a2005-10-29 18:16:30 -0700687 spinlock_t *ptl;
Hugh Dickinsae859762005-10-29 18:16:05 -0700688 int file_rss = 0;
689 int anon_rss = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690
Hugh Dickins508034a2005-10-29 18:16:30 -0700691 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700692 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 do {
694 pte_t ptent = *pte;
Robin Holt51c6f662005-11-13 16:06:42 -0800695 if (pte_none(ptent)) {
696 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800698 }
Hugh Dickins6f5e6b92006-03-16 23:04:09 -0800699
700 (*zap_work) -= PAGE_SIZE;
701
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 if (pte_present(ptent)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800703 struct page *page;
Robin Holt51c6f662005-11-13 16:06:42 -0800704
Linus Torvalds6aab3412005-11-28 14:34:23 -0800705 page = vm_normal_page(vma, addr, ptent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 if (unlikely(details) && page) {
707 /*
708 * unmap_shared_mapping_pages() wants to
709 * invalidate cache without truncating:
710 * unmap shared but keep private pages.
711 */
712 if (details->check_mapping &&
713 details->check_mapping != page->mapping)
714 continue;
715 /*
716 * Each page->index must be checked when
717 * invalidating or truncating nonlinear.
718 */
719 if (details->nonlinear_vma &&
720 (page->index < details->first_index ||
721 page->index > details->last_index))
722 continue;
723 }
Nick Pigginb5810032005-10-29 18:16:12 -0700724 ptent = ptep_get_and_clear_full(mm, addr, pte,
Zachary Amsdena6003882005-09-03 15:55:04 -0700725 tlb->fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 tlb_remove_tlb_entry(tlb, pte, addr);
727 if (unlikely(!page))
728 continue;
729 if (unlikely(details) && details->nonlinear_vma
730 && linear_page_index(details->nonlinear_vma,
731 addr) != page->index)
Nick Pigginb5810032005-10-29 18:16:12 -0700732 set_pte_at(mm, addr, pte,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 pgoff_to_pte(page->index));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 if (PageAnon(page))
Hugh Dickins86d912f2005-10-29 18:16:14 -0700735 anon_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700736 else {
737 if (pte_dirty(ptent))
738 set_page_dirty(page);
739 if (pte_young(ptent))
Ken Chendaa88c82007-02-10 01:43:18 -0800740 SetPageReferenced(page);
Hugh Dickins86d912f2005-10-29 18:16:14 -0700741 file_rss--;
Hugh Dickins6237bcd2005-10-29 18:15:54 -0700742 }
Nick Piggin7de6b802006-12-22 01:09:33 -0800743 page_remove_rmap(page, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 tlb_remove_page(tlb, page);
745 continue;
746 }
747 /*
748 * If details->check_mapping, we leave swap entries;
749 * if details->nonlinear_vma, we leave file entries.
750 */
751 if (unlikely(details))
752 continue;
753 if (!pte_file(ptent))
754 free_swap_and_cache(pte_to_swp_entry(ptent));
Zachary Amsden9888a1c2006-09-30 23:29:31 -0700755 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
Robin Holt51c6f662005-11-13 16:06:42 -0800756 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
Hugh Dickinsae859762005-10-29 18:16:05 -0700757
Hugh Dickins86d912f2005-10-29 18:16:14 -0700758 add_mm_rss(mm, file_rss, anon_rss);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700759 arch_leave_lazy_mmu_mode();
Hugh Dickins508034a2005-10-29 18:16:30 -0700760 pte_unmap_unlock(pte - 1, ptl);
Robin Holt51c6f662005-11-13 16:06:42 -0800761
762 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763}
764
Robin Holt51c6f662005-11-13 16:06:42 -0800765static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700766 struct vm_area_struct *vma, pud_t *pud,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800768 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769{
770 pmd_t *pmd;
771 unsigned long next;
772
773 pmd = pmd_offset(pud, addr);
774 do {
775 next = pmd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800776 if (pmd_none_or_clear_bad(pmd)) {
777 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800779 }
780 next = zap_pte_range(tlb, vma, pmd, addr, next,
781 zap_work, details);
782 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
783
784 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785}
786
Robin Holt51c6f662005-11-13 16:06:42 -0800787static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
Nick Pigginb5810032005-10-29 18:16:12 -0700788 struct vm_area_struct *vma, pgd_t *pgd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800790 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791{
792 pud_t *pud;
793 unsigned long next;
794
795 pud = pud_offset(pgd, addr);
796 do {
797 next = pud_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800798 if (pud_none_or_clear_bad(pud)) {
799 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800801 }
802 next = zap_pmd_range(tlb, vma, pud, addr, next,
803 zap_work, details);
804 } while (pud++, addr = next, (addr != end && *zap_work > 0));
805
806 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807}
808
Robin Holt51c6f662005-11-13 16:06:42 -0800809static unsigned long unmap_page_range(struct mmu_gather *tlb,
810 struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 unsigned long addr, unsigned long end,
Robin Holt51c6f662005-11-13 16:06:42 -0800812 long *zap_work, struct zap_details *details)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813{
814 pgd_t *pgd;
815 unsigned long next;
816
817 if (details && !details->check_mapping && !details->nonlinear_vma)
818 details = NULL;
819
820 BUG_ON(addr >= end);
821 tlb_start_vma(tlb, vma);
822 pgd = pgd_offset(vma->vm_mm, addr);
823 do {
824 next = pgd_addr_end(addr, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800825 if (pgd_none_or_clear_bad(pgd)) {
826 (*zap_work)--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 continue;
Robin Holt51c6f662005-11-13 16:06:42 -0800828 }
829 next = zap_pud_range(tlb, vma, pgd, addr, next,
830 zap_work, details);
831 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 tlb_end_vma(tlb, vma);
Robin Holt51c6f662005-11-13 16:06:42 -0800833
834 return addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835}
836
837#ifdef CONFIG_PREEMPT
838# define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
839#else
840/* No preempt: go for improved straight-line efficiency */
841# define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
842#endif
843
844/**
845 * unmap_vmas - unmap a range of memory covered by a list of vma's
846 * @tlbp: address of the caller's struct mmu_gather
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 * @vma: the starting vma
848 * @start_addr: virtual address at which to start unmapping
849 * @end_addr: virtual address at which to end unmapping
850 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
851 * @details: details of nonlinear truncation or shared cache invalidation
852 *
Hugh Dickinsee39b372005-04-19 13:29:15 -0700853 * Returns the end address of the unmapping (restart addr if interrupted).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700855 * Unmap all pages in the vma list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 *
Hugh Dickins508034a2005-10-29 18:16:30 -0700857 * We aim to not hold locks for too long (for scheduling latency reasons).
858 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 * return the ending mmu_gather to the caller.
860 *
861 * Only addresses between `start' and `end' will be unmapped.
862 *
863 * The VMA list must be sorted in ascending virtual address order.
864 *
865 * unmap_vmas() assumes that the caller will flush the whole unmapped address
866 * range after unmap_vmas() returns. So the only responsibility here is to
867 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
868 * drops the lock and schedules.
869 */
Hugh Dickins508034a2005-10-29 18:16:30 -0700870unsigned long unmap_vmas(struct mmu_gather **tlbp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 struct vm_area_struct *vma, unsigned long start_addr,
872 unsigned long end_addr, unsigned long *nr_accounted,
873 struct zap_details *details)
874{
Robin Holt51c6f662005-11-13 16:06:42 -0800875 long zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
877 int tlb_start_valid = 0;
Hugh Dickinsee39b372005-04-19 13:29:15 -0700878 unsigned long start = start_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
Hugh Dickins4d6ddfa2005-10-29 18:16:02 -0700880 int fullmm = (*tlbp)->fullmm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881
882 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 unsigned long end;
884
885 start = max(vma->vm_start, start_addr);
886 if (start >= vma->vm_end)
887 continue;
888 end = min(vma->vm_end, end_addr);
889 if (end <= vma->vm_start)
890 continue;
891
892 if (vma->vm_flags & VM_ACCOUNT)
893 *nr_accounted += (end - start) >> PAGE_SHIFT;
894
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 while (start != end) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 if (!tlb_start_valid) {
897 tlb_start = start;
898 tlb_start_valid = 1;
899 }
900
Robin Holt51c6f662005-11-13 16:06:42 -0800901 if (unlikely(is_vm_hugetlb_page(vma))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 unmap_hugepage_range(vma, start, end);
Robin Holt51c6f662005-11-13 16:06:42 -0800903 zap_work -= (end - start) /
904 (HPAGE_SIZE / PAGE_SIZE);
905 start = end;
906 } else
907 start = unmap_page_range(*tlbp, vma,
908 start, end, &zap_work, details);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909
Robin Holt51c6f662005-11-13 16:06:42 -0800910 if (zap_work > 0) {
911 BUG_ON(start != end);
912 break;
913 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914
915 tlb_finish_mmu(*tlbp, tlb_start, start);
916
917 if (need_resched() ||
Nick Piggin95c354f2008-01-30 13:31:20 +0100918 (i_mmap_lock && spin_needbreak(i_mmap_lock))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 if (i_mmap_lock) {
Hugh Dickins508034a2005-10-29 18:16:30 -0700920 *tlbp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 goto out;
922 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 }
925
Hugh Dickins508034a2005-10-29 18:16:30 -0700926 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 tlb_start_valid = 0;
Robin Holt51c6f662005-11-13 16:06:42 -0800928 zap_work = ZAP_BLOCK_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 }
930 }
931out:
Hugh Dickinsee39b372005-04-19 13:29:15 -0700932 return start; /* which is now the end (or restart) address */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933}
934
935/**
936 * zap_page_range - remove user pages in a given range
937 * @vma: vm_area_struct holding the applicable pages
938 * @address: starting address of pages to zap
939 * @size: number of bytes to zap
940 * @details: details of nonlinear truncation or shared cache invalidation
941 */
Hugh Dickinsee39b372005-04-19 13:29:15 -0700942unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 unsigned long size, struct zap_details *details)
944{
945 struct mm_struct *mm = vma->vm_mm;
946 struct mmu_gather *tlb;
947 unsigned long end = address + size;
948 unsigned long nr_accounted = 0;
949
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 lru_add_drain();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 tlb = tlb_gather_mmu(mm, 0);
Hugh Dickins365e9c872005-10-29 18:16:18 -0700952 update_hiwater_rss(mm);
Hugh Dickins508034a2005-10-29 18:16:30 -0700953 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
954 if (tlb)
955 tlb_finish_mmu(tlb, address, end);
Hugh Dickinsee39b372005-04-19 13:29:15 -0700956 return end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957}
958
959/*
960 * Do a quick page-table lookup for a single page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 */
Linus Torvalds6aab3412005-11-28 14:34:23 -0800962struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700963 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964{
965 pgd_t *pgd;
966 pud_t *pud;
967 pmd_t *pmd;
968 pte_t *ptep, pte;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700969 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 struct page *page;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800971 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700973 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
974 if (!IS_ERR(page)) {
975 BUG_ON(flags & FOLL_GET);
976 goto out;
977 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700979 page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 pgd = pgd_offset(mm, address);
981 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700982 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983
984 pud = pud_offset(pgd, address);
985 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700986 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987
988 pmd = pmd_offset(pud, address);
Hugh Dickinsaeed5fc2008-05-06 20:49:23 +0100989 if (pmd_none(*pmd))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700990 goto no_page_table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -0700992 if (pmd_huge(*pmd)) {
993 BUG_ON(flags & FOLL_GET);
994 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
995 goto out;
996 }
997
Hugh Dickinsaeed5fc2008-05-06 20:49:23 +0100998 if (unlikely(pmd_bad(*pmd)))
999 goto no_page_table;
1000
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001001 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002
1003 pte = *ptep;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001004 if (!pte_present(pte))
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001005 goto no_page;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001006 if ((flags & FOLL_WRITE) && !pte_write(pte))
1007 goto unlock;
Linus Torvalds6aab3412005-11-28 14:34:23 -08001008 page = vm_normal_page(vma, address, pte);
1009 if (unlikely(!page))
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001010 goto bad_page;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001011
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001012 if (flags & FOLL_GET)
1013 get_page(page);
1014 if (flags & FOLL_TOUCH) {
1015 if ((flags & FOLL_WRITE) &&
1016 !pte_dirty(pte) && !PageDirty(page))
1017 set_page_dirty(page);
1018 mark_page_accessed(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001020unlock:
1021 pte_unmap_unlock(ptep, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022out:
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001023 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001025bad_page:
1026 pte_unmap_unlock(ptep, ptl);
1027 return ERR_PTR(-EFAULT);
1028
1029no_page:
1030 pte_unmap_unlock(ptep, ptl);
1031 if (!pte_none(pte))
1032 return page;
1033 /* Fall through to ZERO_PAGE handling */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001034no_page_table:
1035 /*
1036 * When core dumping an enormous anonymous area that nobody
1037 * has touched so far, we don't want to allocate page tables.
1038 */
1039 if (flags & FOLL_ANON) {
Nick Piggin557ed1f2007-10-16 01:24:40 -07001040 page = ZERO_PAGE(0);
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001041 if (flags & FOLL_GET)
1042 get_page(page);
1043 BUG_ON(flags & FOLL_WRITE);
1044 }
1045 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046}
1047
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
1049 unsigned long start, int len, int write, int force,
1050 struct page **pages, struct vm_area_struct **vmas)
1051{
1052 int i;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001053 unsigned int vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054
Jonathan Corbet900cf082008-02-11 16:17:33 -07001055 if (len <= 0)
1056 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 /*
1058 * Require read or write permissions.
1059 * If 'force' is set, we only require the "MAY" flags.
1060 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001061 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
1062 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 i = 0;
1064
1065 do {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001066 struct vm_area_struct *vma;
1067 unsigned int foll_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068
1069 vma = find_extend_vma(mm, start);
1070 if (!vma && in_gate_area(tsk, start)) {
1071 unsigned long pg = start & PAGE_MASK;
1072 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
1073 pgd_t *pgd;
1074 pud_t *pud;
1075 pmd_t *pmd;
1076 pte_t *pte;
1077 if (write) /* user gate pages are read-only */
1078 return i ? : -EFAULT;
1079 if (pg > TASK_SIZE)
1080 pgd = pgd_offset_k(pg);
1081 else
1082 pgd = pgd_offset_gate(mm, pg);
1083 BUG_ON(pgd_none(*pgd));
1084 pud = pud_offset(pgd, pg);
1085 BUG_ON(pud_none(*pud));
1086 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001087 if (pmd_none(*pmd))
1088 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001090 if (pte_none(*pte)) {
1091 pte_unmap(pte);
1092 return i ? : -EFAULT;
1093 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 if (pages) {
Nick Pigginfa2a4552005-11-29 18:43:17 +11001095 struct page *page = vm_normal_page(gate_vma, start, *pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001096 pages[i] = page;
1097 if (page)
1098 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 }
1100 pte_unmap(pte);
1101 if (vmas)
1102 vmas[i] = gate_vma;
1103 i++;
1104 start += PAGE_SIZE;
1105 len--;
1106 continue;
1107 }
1108
Linus Torvalds1ff80382005-12-12 16:24:33 -08001109 if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001110 || !(vm_flags & vma->vm_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 return i ? : -EFAULT;
1112
1113 if (is_vm_hugetlb_page(vma)) {
1114 i = follow_hugetlb_page(mm, vma, pages, vmas,
Adam Litke5b23dbe2007-11-14 16:59:33 -08001115 &start, &len, i, write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 continue;
1117 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001118
1119 foll_flags = FOLL_TOUCH;
1120 if (pages)
1121 foll_flags |= FOLL_GET;
1122 if (!write && !(vma->vm_flags & VM_LOCKED) &&
Nick Piggin3c18ddd2008-04-28 02:12:10 -07001123 (!vma->vm_ops || !vma->vm_ops->fault))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001124 foll_flags |= FOLL_ANON;
1125
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 do {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001127 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128
Ethan Solomita462e00c2007-07-15 23:38:16 -07001129 /*
1130 * If tsk is ooming, cut off its access to large memory
1131 * allocations. It has a pending SIGKILL, but it can't
1132 * be processed until returning to user space.
1133 */
1134 if (unlikely(test_tsk_thread_flag(tsk, TIF_MEMDIE)))
1135 return -ENOMEM;
1136
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001137 if (write)
1138 foll_flags |= FOLL_WRITE;
1139
1140 cond_resched();
Linus Torvalds6aab3412005-11-28 14:34:23 -08001141 while (!(page = follow_page(vma, start, foll_flags))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001142 int ret;
Nick Piggin83c54072007-07-19 01:47:05 -07001143 ret = handle_mm_fault(mm, vma, start,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001144 foll_flags & FOLL_WRITE);
Nick Piggin83c54072007-07-19 01:47:05 -07001145 if (ret & VM_FAULT_ERROR) {
1146 if (ret & VM_FAULT_OOM)
1147 return i ? i : -ENOMEM;
1148 else if (ret & VM_FAULT_SIGBUS)
1149 return i ? i : -EFAULT;
1150 BUG();
1151 }
1152 if (ret & VM_FAULT_MAJOR)
1153 tsk->maj_flt++;
1154 else
1155 tsk->min_flt++;
1156
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001157 /*
Nick Piggin83c54072007-07-19 01:47:05 -07001158 * The VM_FAULT_WRITE bit tells us that
1159 * do_wp_page has broken COW when necessary,
1160 * even if maybe_mkwrite decided not to set
1161 * pte_write. We can thus safely do subsequent
1162 * page lookups as if they were reads.
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001163 */
1164 if (ret & VM_FAULT_WRITE)
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001165 foll_flags &= ~FOLL_WRITE;
Nick Piggin83c54072007-07-19 01:47:05 -07001166
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07001167 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 }
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001169 if (IS_ERR(page))
1170 return i ? i : PTR_ERR(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001172 pages[i] = page;
James Bottomley03beb072006-03-26 01:36:57 -08001173
Russell Kinga6f36be2006-12-30 22:24:19 +00001174 flush_anon_page(vma, page, start);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001175 flush_dcache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 }
1177 if (vmas)
1178 vmas[i] = vma;
1179 i++;
1180 start += PAGE_SIZE;
1181 len--;
Hugh Dickins08ef4722005-06-21 17:15:10 -07001182 } while (len && start < vma->vm_end);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001183 } while (len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 return i;
1185}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186EXPORT_SYMBOL(get_user_pages);
1187
Harvey Harrison920c7a52008-02-04 22:29:26 -08001188pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr,
1189 spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001190{
1191 pgd_t * pgd = pgd_offset(mm, addr);
1192 pud_t * pud = pud_alloc(mm, pgd, addr);
1193 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001194 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001195 if (pmd)
1196 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1197 }
1198 return NULL;
1199}
1200
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001202 * This is the old fallback for page remapping.
1203 *
1204 * For historical reasons, it only allows reserved pages. Only
1205 * old drivers should use this, and they needed to mark their
1206 * pages reserved for the old functions anyway.
1207 */
Nick Piggin423bad62008-04-28 02:13:01 -07001208static int insert_page(struct vm_area_struct *vma, unsigned long addr,
1209 struct page *page, pgprot_t prot)
Linus Torvalds238f58d2005-11-29 13:01:56 -08001210{
Nick Piggin423bad62008-04-28 02:13:01 -07001211 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001212 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001213 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001214 spinlock_t *ptl;
1215
Balbir Singhe1a1cd52008-02-07 00:14:02 -08001216 retval = mem_cgroup_charge(page, mm, GFP_KERNEL);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001217 if (retval)
1218 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001219
1220 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001221 if (PageAnon(page))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001222 goto out_uncharge;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001223 retval = -ENOMEM;
1224 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001225 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001226 if (!pte)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001227 goto out_uncharge;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001228 retval = -EBUSY;
1229 if (!pte_none(*pte))
1230 goto out_unlock;
1231
1232 /* Ok, finally just insert the thing.. */
1233 get_page(page);
1234 inc_mm_counter(mm, file_rss);
1235 page_add_file_rmap(page);
1236 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1237
1238 retval = 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001239 pte_unmap_unlock(pte, ptl);
1240 return retval;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001241out_unlock:
1242 pte_unmap_unlock(pte, ptl);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001243out_uncharge:
1244 mem_cgroup_uncharge_page(page);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001245out:
1246 return retval;
1247}
1248
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001249/**
1250 * vm_insert_page - insert single page into user vma
1251 * @vma: user vma to map to
1252 * @addr: target user address of this page
1253 * @page: source kernel page
1254 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001255 * This allows drivers to insert individual pages they've allocated
1256 * into a user vma.
1257 *
1258 * The page has to be a nice clean _individual_ kernel allocation.
1259 * If you allocate a compound page, you need to have marked it as
1260 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001261 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001262 *
1263 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1264 * took an arbitrary page protection parameter. This doesn't allow
1265 * that. Your vma protection will have to be set up correctly, which
1266 * means that if you want a shared writable mapping, you'd better
1267 * ask for a shared writable mapping!
1268 *
1269 * The page does not need to be reserved.
1270 */
Nick Piggin423bad62008-04-28 02:13:01 -07001271int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
1272 struct page *page)
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001273{
1274 if (addr < vma->vm_start || addr >= vma->vm_end)
1275 return -EFAULT;
1276 if (!page_count(page))
1277 return -EINVAL;
Linus Torvalds4d7672b2005-12-16 10:21:23 -08001278 vma->vm_flags |= VM_INSERTPAGE;
Nick Piggin423bad62008-04-28 02:13:01 -07001279 return insert_page(vma, addr, page, vma->vm_page_prot);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001280}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001281EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001282
Nick Piggin423bad62008-04-28 02:13:01 -07001283static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1284 unsigned long pfn, pgprot_t prot)
1285{
1286 struct mm_struct *mm = vma->vm_mm;
1287 int retval;
1288 pte_t *pte, entry;
1289 spinlock_t *ptl;
1290
1291 retval = -ENOMEM;
1292 pte = get_locked_pte(mm, addr, &ptl);
1293 if (!pte)
1294 goto out;
1295 retval = -EBUSY;
1296 if (!pte_none(*pte))
1297 goto out_unlock;
1298
1299 /* Ok, finally just insert the thing.. */
1300 entry = pte_mkspecial(pfn_pte(pfn, prot));
1301 set_pte_at(mm, addr, pte, entry);
1302 update_mmu_cache(vma, addr, entry); /* XXX: why not for insert_page? */
1303
1304 retval = 0;
1305out_unlock:
1306 pte_unmap_unlock(pte, ptl);
1307out:
1308 return retval;
1309}
1310
Nick Piggine0dc0d82007-02-12 00:51:36 -08001311/**
1312 * vm_insert_pfn - insert single pfn into user vma
1313 * @vma: user vma to map to
1314 * @addr: target user address of this page
1315 * @pfn: source kernel pfn
1316 *
1317 * Similar to vm_inert_page, this allows drivers to insert individual pages
1318 * they've allocated into a user vma. Same comments apply.
1319 *
1320 * This function should only be called from a vm_ops->fault handler, and
1321 * in that case the handler should return NULL.
1322 */
1323int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
Nick Piggin423bad62008-04-28 02:13:01 -07001324 unsigned long pfn)
Nick Piggine0dc0d82007-02-12 00:51:36 -08001325{
Nick Piggin7e675132008-04-28 02:13:00 -07001326 /*
1327 * Technically, architectures with pte_special can avoid all these
1328 * restrictions (same for remap_pfn_range). However we would like
1329 * consistency in testing and feature parity among all, so we should
1330 * try to keep these invariants in place for everybody.
1331 */
Jared Hulbertb379d792008-04-28 02:12:58 -07001332 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
1333 BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
1334 (VM_PFNMAP|VM_MIXEDMAP));
1335 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1336 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
Nick Piggine0dc0d82007-02-12 00:51:36 -08001337
Nick Piggin423bad62008-04-28 02:13:01 -07001338 if (addr < vma->vm_start || addr >= vma->vm_end)
1339 return -EFAULT;
1340 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
Nick Piggine0dc0d82007-02-12 00:51:36 -08001341}
1342EXPORT_SYMBOL(vm_insert_pfn);
1343
Nick Piggin423bad62008-04-28 02:13:01 -07001344int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
1345 unsigned long pfn)
1346{
1347 BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
1348
1349 if (addr < vma->vm_start || addr >= vma->vm_end)
1350 return -EFAULT;
1351
1352 /*
1353 * If we don't have pte special, then we have to use the pfn_valid()
1354 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1355 * refcount the page if pfn_valid is true (hence insert_page rather
1356 * than insert_pfn).
1357 */
1358 if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) {
1359 struct page *page;
1360
1361 page = pfn_to_page(pfn);
1362 return insert_page(vma, addr, page, vma->vm_page_prot);
1363 }
1364 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
1365}
1366EXPORT_SYMBOL(vm_insert_mixed);
1367
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001368/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 * maps a range of physical memory into the requested pages. the old
1370 * mappings are removed. any references to nonexistent pages results
1371 * in null mappings (currently treated as "copy-on-access")
1372 */
1373static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1374 unsigned long addr, unsigned long end,
1375 unsigned long pfn, pgprot_t prot)
1376{
1377 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001378 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379
Hugh Dickinsc74df322005-10-29 18:16:23 -07001380 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 if (!pte)
1382 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001383 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 do {
1385 BUG_ON(!pte_none(*pte));
Nick Piggin7e675132008-04-28 02:13:00 -07001386 set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 pfn++;
1388 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001389 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001390 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 return 0;
1392}
1393
1394static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1395 unsigned long addr, unsigned long end,
1396 unsigned long pfn, pgprot_t prot)
1397{
1398 pmd_t *pmd;
1399 unsigned long next;
1400
1401 pfn -= addr >> PAGE_SHIFT;
1402 pmd = pmd_alloc(mm, pud, addr);
1403 if (!pmd)
1404 return -ENOMEM;
1405 do {
1406 next = pmd_addr_end(addr, end);
1407 if (remap_pte_range(mm, pmd, addr, next,
1408 pfn + (addr >> PAGE_SHIFT), prot))
1409 return -ENOMEM;
1410 } while (pmd++, addr = next, addr != end);
1411 return 0;
1412}
1413
1414static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1415 unsigned long addr, unsigned long end,
1416 unsigned long pfn, pgprot_t prot)
1417{
1418 pud_t *pud;
1419 unsigned long next;
1420
1421 pfn -= addr >> PAGE_SHIFT;
1422 pud = pud_alloc(mm, pgd, addr);
1423 if (!pud)
1424 return -ENOMEM;
1425 do {
1426 next = pud_addr_end(addr, end);
1427 if (remap_pmd_range(mm, pud, addr, next,
1428 pfn + (addr >> PAGE_SHIFT), prot))
1429 return -ENOMEM;
1430 } while (pud++, addr = next, addr != end);
1431 return 0;
1432}
1433
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001434/**
1435 * remap_pfn_range - remap kernel memory to userspace
1436 * @vma: user vma to map to
1437 * @addr: target user address to start at
1438 * @pfn: physical address of kernel memory
1439 * @size: size of map area
1440 * @prot: page protection flags for this mapping
1441 *
1442 * Note: this is only safe if the mm semaphore is held when called.
1443 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1445 unsigned long pfn, unsigned long size, pgprot_t prot)
1446{
1447 pgd_t *pgd;
1448 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001449 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 struct mm_struct *mm = vma->vm_mm;
1451 int err;
1452
1453 /*
1454 * Physically remapped pages are special. Tell the
1455 * rest of the world about it:
1456 * VM_IO tells people not to look at these pages
1457 * (accesses can have side effects).
Hugh Dickins0b14c172005-11-21 21:32:15 -08001458 * VM_RESERVED is specified all over the place, because
1459 * in 2.4 it kept swapout's vma scan off this vma; but
1460 * in 2.6 the LRU scan won't even find its pages, so this
1461 * flag means no more than count its pages in reserved_vm,
1462 * and omit it from core dump, even when VM_IO turned off.
Linus Torvalds6aab3412005-11-28 14:34:23 -08001463 * VM_PFNMAP tells the core MM that the base pages are just
1464 * raw PFN mappings, and do not have a "struct page" associated
1465 * with them.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001466 *
1467 * There's a horrible special case to handle copy-on-write
1468 * behaviour that some programs depend on. We mark the "original"
1469 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 */
Linus Torvalds67121172005-12-11 20:38:17 -08001471 if (is_cow_mapping(vma->vm_flags)) {
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001472 if (addr != vma->vm_start || end != vma->vm_end)
Linus Torvalds7fc7e2e2005-12-11 19:57:52 -08001473 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001474 vma->vm_pgoff = pfn;
1475 }
1476
Linus Torvalds6aab3412005-11-28 14:34:23 -08001477 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478
1479 BUG_ON(addr >= end);
1480 pfn -= addr >> PAGE_SHIFT;
1481 pgd = pgd_offset(mm, addr);
1482 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 do {
1484 next = pgd_addr_end(addr, end);
1485 err = remap_pud_range(mm, pgd, addr, next,
1486 pfn + (addr >> PAGE_SHIFT), prot);
1487 if (err)
1488 break;
1489 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 return err;
1491}
1492EXPORT_SYMBOL(remap_pfn_range);
1493
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001494static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1495 unsigned long addr, unsigned long end,
1496 pte_fn_t fn, void *data)
1497{
1498 pte_t *pte;
1499 int err;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001500 pgtable_t token;
Borislav Petkov94909912007-05-06 14:49:17 -07001501 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001502
1503 pte = (mm == &init_mm) ?
1504 pte_alloc_kernel(pmd, addr) :
1505 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1506 if (!pte)
1507 return -ENOMEM;
1508
1509 BUG_ON(pmd_huge(*pmd));
1510
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001511 token = pmd_pgtable(*pmd);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001512
1513 do {
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001514 err = fn(pte, token, addr, data);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001515 if (err)
1516 break;
1517 } while (pte++, addr += PAGE_SIZE, addr != end);
1518
1519 if (mm != &init_mm)
1520 pte_unmap_unlock(pte-1, ptl);
1521 return err;
1522}
1523
1524static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1525 unsigned long addr, unsigned long end,
1526 pte_fn_t fn, void *data)
1527{
1528 pmd_t *pmd;
1529 unsigned long next;
1530 int err;
1531
1532 pmd = pmd_alloc(mm, pud, addr);
1533 if (!pmd)
1534 return -ENOMEM;
1535 do {
1536 next = pmd_addr_end(addr, end);
1537 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1538 if (err)
1539 break;
1540 } while (pmd++, addr = next, addr != end);
1541 return err;
1542}
1543
1544static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1545 unsigned long addr, unsigned long end,
1546 pte_fn_t fn, void *data)
1547{
1548 pud_t *pud;
1549 unsigned long next;
1550 int err;
1551
1552 pud = pud_alloc(mm, pgd, addr);
1553 if (!pud)
1554 return -ENOMEM;
1555 do {
1556 next = pud_addr_end(addr, end);
1557 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1558 if (err)
1559 break;
1560 } while (pud++, addr = next, addr != end);
1561 return err;
1562}
1563
1564/*
1565 * Scan a region of virtual memory, filling in page tables as necessary
1566 * and calling a provided function on each leaf page table.
1567 */
1568int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1569 unsigned long size, pte_fn_t fn, void *data)
1570{
1571 pgd_t *pgd;
1572 unsigned long next;
1573 unsigned long end = addr + size;
1574 int err;
1575
1576 BUG_ON(addr >= end);
1577 pgd = pgd_offset(mm, addr);
1578 do {
1579 next = pgd_addr_end(addr, end);
1580 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1581 if (err)
1582 break;
1583 } while (pgd++, addr = next, addr != end);
1584 return err;
1585}
1586EXPORT_SYMBOL_GPL(apply_to_page_range);
1587
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001589 * handle_pte_fault chooses page fault handler according to an entry
1590 * which was read non-atomically. Before making any commitment, on
1591 * those architectures or configurations (e.g. i386 with PAE) which
1592 * might give a mix of unmatched parts, do_swap_page and do_file_page
1593 * must check under lock before unmapping the pte and proceeding
1594 * (but do_wp_page is only called after already making such a check;
1595 * and do_anonymous_page and do_no_page can safely check later on).
1596 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001597static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001598 pte_t *page_table, pte_t orig_pte)
1599{
1600 int same = 1;
1601#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1602 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001603 spinlock_t *ptl = pte_lockptr(mm, pmd);
1604 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001605 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001606 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001607 }
1608#endif
1609 pte_unmap(page_table);
1610 return same;
1611}
1612
1613/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1615 * servicing faults for write access. In the normal case, do always want
1616 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1617 * that do not have writing enabled, when used by access_process_vm.
1618 */
1619static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1620{
1621 if (likely(vma->vm_flags & VM_WRITE))
1622 pte = pte_mkwrite(pte);
1623 return pte;
1624}
1625
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001626static 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 -08001627{
1628 /*
1629 * If the source page was a PFN mapping, we don't have
1630 * a "struct page" for it. We do a best-effort copy by
1631 * just copying from the original user address. If that
1632 * fails, we just zero-fill it. Live with it.
1633 */
1634 if (unlikely(!src)) {
1635 void *kaddr = kmap_atomic(dst, KM_USER0);
Linus Torvalds5d2a2db2005-11-29 14:07:55 -08001636 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1637
1638 /*
1639 * This really shouldn't fail, because the page is there
1640 * in the page tables. But it might just be unreadable,
1641 * in which case we just give up and fill the result with
1642 * zeroes.
1643 */
1644 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Linus Torvalds6aab3412005-11-28 14:34:23 -08001645 memset(kaddr, 0, PAGE_SIZE);
1646 kunmap_atomic(kaddr, KM_USER0);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07001647 flush_dcache_page(dst);
Nick Piggin0ed361d2008-02-04 22:29:34 -08001648 } else
1649 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001650}
1651
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 * This routine handles present pages, when users try to write
1654 * to a shared page. It is done by copying the page to a new address
1655 * and decrementing the shared-page counter for the old page.
1656 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 * Note that this routine assumes that the protection checks have been
1658 * done by the caller (the low-level page fault routine in most cases).
1659 * Thus we can safely just mark it writable once we've done any necessary
1660 * COW.
1661 *
1662 * We also mark the page dirty at this point even though the page will
1663 * change only once the write actually happens. This avoids a few races,
1664 * and potentially makes it more efficient.
1665 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001666 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1667 * but allow concurrent faults), with pte both mapped and locked.
1668 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001670static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1671 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001672 spinlock_t *ptl, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673{
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001674 struct page *old_page, *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 pte_t entry;
Nick Piggin83c54072007-07-19 01:47:05 -07001676 int reuse = 0, ret = 0;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001677 int page_mkwrite = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001678 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679
Linus Torvalds6aab3412005-11-28 14:34:23 -08001680 old_page = vm_normal_page(vma, address, orig_pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001681 if (!old_page)
1682 goto gotten;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001684 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001685 * Take out anonymous pages first, anonymous shared vmas are
1686 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001687 */
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001688 if (PageAnon(old_page)) {
1689 if (!TestSetPageLocked(old_page)) {
1690 reuse = can_share_swap_page(old_page);
1691 unlock_page(old_page);
1692 }
1693 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001694 (VM_WRITE|VM_SHARED))) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001695 /*
1696 * Only catch write-faults on shared writable pages,
1697 * read-only shared pages can get COWed by
1698 * get_user_pages(.write=1, .force=1).
1699 */
David Howells9637a5e2006-06-23 02:03:43 -07001700 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
1701 /*
1702 * Notify the address space that the page is about to
1703 * become writable so that it can prohibit this or wait
1704 * for the page to get into an appropriate state.
1705 *
1706 * We do this without the lock held, so that it can
1707 * sleep if it needs to.
1708 */
1709 page_cache_get(old_page);
1710 pte_unmap_unlock(page_table, ptl);
1711
1712 if (vma->vm_ops->page_mkwrite(vma, old_page) < 0)
1713 goto unwritable_page;
1714
David Howells9637a5e2006-06-23 02:03:43 -07001715 /*
1716 * Since we dropped the lock we need to revalidate
1717 * the PTE as someone else may have changed it. If
1718 * they did, we just return, as we can count on the
1719 * MMU to tell us if they didn't also make it writable.
1720 */
1721 page_table = pte_offset_map_lock(mm, pmd, address,
1722 &ptl);
Hugh Dickinsc3704ce2007-02-10 01:43:00 -08001723 page_cache_release(old_page);
David Howells9637a5e2006-06-23 02:03:43 -07001724 if (!pte_same(*page_table, orig_pte))
1725 goto unlock;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001726
1727 page_mkwrite = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001729 dirty_page = old_page;
1730 get_page(dirty_page);
David Howells9637a5e2006-06-23 02:03:43 -07001731 reuse = 1;
David Howells9637a5e2006-06-23 02:03:43 -07001732 }
1733
1734 if (reuse) {
1735 flush_cache_page(vma, address, pte_pfn(orig_pte));
1736 entry = pte_mkyoung(orig_pte);
1737 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
KAMEZAWA Hiroyuki954ffcb2007-10-16 01:25:44 -07001738 if (ptep_set_access_flags(vma, address, page_table, entry,1))
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07001739 update_mmu_cache(vma, address, entry);
David Howells9637a5e2006-06-23 02:03:43 -07001740 ret |= VM_FAULT_WRITE;
1741 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743
1744 /*
1745 * Ok, we need to copy. Oh, well..
1746 */
Nick Pigginb5810032005-10-29 18:16:12 -07001747 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08001748gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001749 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750
1751 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07001752 goto oom;
Nick Piggin557ed1f2007-10-16 01:24:40 -07001753 VM_BUG_ON(old_page == ZERO_PAGE(0));
1754 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
1755 if (!new_page)
1756 goto oom;
1757 cow_user_page(new_page, old_page, address, vma);
Nick Piggin0ed361d2008-02-04 22:29:34 -08001758 __SetPageUptodate(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001759
Balbir Singhe1a1cd52008-02-07 00:14:02 -08001760 if (mem_cgroup_charge(new_page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001761 goto oom_free_new;
1762
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 /*
1764 * Re-check the pte - we dropped the lock
1765 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001766 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07001767 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08001768 if (old_page) {
Nick Piggin7de6b802006-12-22 01:09:33 -08001769 page_remove_rmap(old_page, vma);
Hugh Dickins920fc352005-11-21 21:32:17 -08001770 if (!PageAnon(old_page)) {
1771 dec_mm_counter(mm, file_rss);
1772 inc_mm_counter(mm, anon_rss);
1773 }
1774 } else
Hugh Dickins42946212005-10-29 18:16:05 -07001775 inc_mm_counter(mm, anon_rss);
Ben Collinseca35132005-11-29 11:45:26 -08001776 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07001777 entry = mk_pte(new_page, vma->vm_page_prot);
1778 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Siddha, Suresh B4ce072f2006-09-29 01:58:42 -07001779 /*
1780 * Clear the pte entry and flush it first, before updating the
1781 * pte with the new entry. This will avoid a race condition
1782 * seen in the presence of one thread doing SMC and another
1783 * thread doing COW.
1784 */
1785 ptep_clear_flush(vma, address, page_table);
1786 set_pte_at(mm, address, page_table, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07001787 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 lru_cache_add_active(new_page);
Nick Piggin9617d952006-01-06 00:11:12 -08001789 page_add_new_anon_rmap(new_page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790
1791 /* Free the old page.. */
1792 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001793 ret |= VM_FAULT_WRITE;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001794 } else
1795 mem_cgroup_uncharge_page(new_page);
1796
Hugh Dickins920fc352005-11-21 21:32:17 -08001797 if (new_page)
1798 page_cache_release(new_page);
1799 if (old_page)
1800 page_cache_release(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001801unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001802 pte_unmap_unlock(page_table, ptl);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001803 if (dirty_page) {
Anton Salikhmetov8f7b3d12008-01-23 02:21:18 +03001804 if (vma->vm_file)
1805 file_update_time(vma->vm_file);
1806
Nick Piggin79352892007-07-19 01:47:22 -07001807 /*
1808 * Yes, Virginia, this is actually required to prevent a race
1809 * with clear_page_dirty_for_io() from clearing the page dirty
1810 * bit after it clear all dirty ptes, but before a racing
1811 * do_wp_page installs a dirty pte.
1812 *
1813 * do_no_page is protected similarly.
1814 */
1815 wait_on_page_locked(dirty_page);
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001816 set_page_dirty_balance(dirty_page, page_mkwrite);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001817 put_page(dirty_page);
1818 }
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001819 return ret;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001820oom_free_new:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08001821 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001822oom:
Hugh Dickins920fc352005-11-21 21:32:17 -08001823 if (old_page)
1824 page_cache_release(old_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 return VM_FAULT_OOM;
David Howells9637a5e2006-06-23 02:03:43 -07001826
1827unwritable_page:
1828 page_cache_release(old_page);
1829 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830}
1831
1832/*
1833 * Helper functions for unmap_mapping_range().
1834 *
1835 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1836 *
1837 * We have to restart searching the prio_tree whenever we drop the lock,
1838 * since the iterator is only valid while the lock is held, and anyway
1839 * a later vma might be split and reinserted earlier while lock dropped.
1840 *
1841 * The list of nonlinear vmas could be handled more efficiently, using
1842 * a placeholder, but handle it in the same way until a need is shown.
1843 * It is important to search the prio_tree before nonlinear list: a vma
1844 * may become nonlinear and be shifted from prio_tree to nonlinear list
1845 * while the lock is dropped; but never shifted from list to prio_tree.
1846 *
1847 * In order to make forward progress despite restarting the search,
1848 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1849 * quickly skip it next time around. Since the prio_tree search only
1850 * shows us those vmas affected by unmapping the range in question, we
1851 * can't efficiently keep all vmas in step with mapping->truncate_count:
1852 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1853 * mapping->truncate_count and vma->vm_truncate_count are protected by
1854 * i_mmap_lock.
1855 *
1856 * In order to make forward progress despite repeatedly restarting some
Hugh Dickinsee39b372005-04-19 13:29:15 -07001857 * large vma, note the restart_addr from unmap_vmas when it breaks out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 * and restart from that address when we reach that vma again. It might
1859 * have been split or merged, shrunk or extended, but never shifted: so
1860 * restart_addr remains valid so long as it remains in the vma's range.
1861 * unmap_mapping_range forces truncate_count to leap over page-aligned
1862 * values so we can save vma's restart_addr in its truncate_count field.
1863 */
1864#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1865
1866static void reset_vma_truncate_counts(struct address_space *mapping)
1867{
1868 struct vm_area_struct *vma;
1869 struct prio_tree_iter iter;
1870
1871 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1872 vma->vm_truncate_count = 0;
1873 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1874 vma->vm_truncate_count = 0;
1875}
1876
1877static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1878 unsigned long start_addr, unsigned long end_addr,
1879 struct zap_details *details)
1880{
1881 unsigned long restart_addr;
1882 int need_break;
1883
Nick Piggind00806b2007-07-19 01:46:57 -07001884 /*
1885 * files that support invalidating or truncating portions of the
Nick Piggind0217ac2007-07-19 01:47:03 -07001886 * file from under mmaped areas must have their ->fault function
Nick Piggin83c54072007-07-19 01:47:05 -07001887 * return a locked page (and set VM_FAULT_LOCKED in the return).
1888 * This provides synchronisation against concurrent unmapping here.
Nick Piggind00806b2007-07-19 01:46:57 -07001889 */
Nick Piggind00806b2007-07-19 01:46:57 -07001890
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891again:
1892 restart_addr = vma->vm_truncate_count;
1893 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1894 start_addr = restart_addr;
1895 if (start_addr >= end_addr) {
1896 /* Top of vma has been split off since last time */
1897 vma->vm_truncate_count = details->truncate_count;
1898 return 0;
1899 }
1900 }
1901
Hugh Dickinsee39b372005-04-19 13:29:15 -07001902 restart_addr = zap_page_range(vma, start_addr,
1903 end_addr - start_addr, details);
Nick Piggin95c354f2008-01-30 13:31:20 +01001904 need_break = need_resched() || spin_needbreak(details->i_mmap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905
Hugh Dickinsee39b372005-04-19 13:29:15 -07001906 if (restart_addr >= end_addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 /* We have now completed this vma: mark it so */
1908 vma->vm_truncate_count = details->truncate_count;
1909 if (!need_break)
1910 return 0;
1911 } else {
1912 /* Note restart_addr in vma's truncate_count field */
Hugh Dickinsee39b372005-04-19 13:29:15 -07001913 vma->vm_truncate_count = restart_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 if (!need_break)
1915 goto again;
1916 }
1917
1918 spin_unlock(details->i_mmap_lock);
1919 cond_resched();
1920 spin_lock(details->i_mmap_lock);
1921 return -EINTR;
1922}
1923
1924static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1925 struct zap_details *details)
1926{
1927 struct vm_area_struct *vma;
1928 struct prio_tree_iter iter;
1929 pgoff_t vba, vea, zba, zea;
1930
1931restart:
1932 vma_prio_tree_foreach(vma, &iter, root,
1933 details->first_index, details->last_index) {
1934 /* Skip quickly over those we have already dealt with */
1935 if (vma->vm_truncate_count == details->truncate_count)
1936 continue;
1937
1938 vba = vma->vm_pgoff;
1939 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1940 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1941 zba = details->first_index;
1942 if (zba < vba)
1943 zba = vba;
1944 zea = details->last_index;
1945 if (zea > vea)
1946 zea = vea;
1947
1948 if (unmap_mapping_range_vma(vma,
1949 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1950 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1951 details) < 0)
1952 goto restart;
1953 }
1954}
1955
1956static inline void unmap_mapping_range_list(struct list_head *head,
1957 struct zap_details *details)
1958{
1959 struct vm_area_struct *vma;
1960
1961 /*
1962 * In nonlinear VMAs there is no correspondence between virtual address
1963 * offset and file offset. So we must perform an exhaustive search
1964 * across *all* the pages in each nonlinear VMA, not just the pages
1965 * whose virtual address lies outside the file truncation point.
1966 */
1967restart:
1968 list_for_each_entry(vma, head, shared.vm_set.list) {
1969 /* Skip quickly over those we have already dealt with */
1970 if (vma->vm_truncate_count == details->truncate_count)
1971 continue;
1972 details->nonlinear_vma = vma;
1973 if (unmap_mapping_range_vma(vma, vma->vm_start,
1974 vma->vm_end, details) < 0)
1975 goto restart;
1976 }
1977}
1978
1979/**
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001980 * 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 -07001981 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 * @holebegin: byte in first page to unmap, relative to the start of
1983 * the underlying file. This will be rounded down to a PAGE_SIZE
1984 * boundary. Note that this is different from vmtruncate(), which
1985 * must keep the partial page. In contrast, we must get rid of
1986 * partial pages.
1987 * @holelen: size of prospective hole in bytes. This will be rounded
1988 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1989 * end of the file.
1990 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1991 * but 0 when invalidating pagecache, don't throw away private data.
1992 */
1993void unmap_mapping_range(struct address_space *mapping,
1994 loff_t const holebegin, loff_t const holelen, int even_cows)
1995{
1996 struct zap_details details;
1997 pgoff_t hba = holebegin >> PAGE_SHIFT;
1998 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1999
2000 /* Check for overflow. */
2001 if (sizeof(holelen) > sizeof(hlen)) {
2002 long long holeend =
2003 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2004 if (holeend & ~(long long)ULONG_MAX)
2005 hlen = ULONG_MAX - hba + 1;
2006 }
2007
2008 details.check_mapping = even_cows? NULL: mapping;
2009 details.nonlinear_vma = NULL;
2010 details.first_index = hba;
2011 details.last_index = hba + hlen - 1;
2012 if (details.last_index < details.first_index)
2013 details.last_index = ULONG_MAX;
2014 details.i_mmap_lock = &mapping->i_mmap_lock;
2015
2016 spin_lock(&mapping->i_mmap_lock);
2017
Nick Piggind00806b2007-07-19 01:46:57 -07002018 /* Protect against endless unmapping loops */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 mapping->truncate_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 if (unlikely(is_restart_addr(mapping->truncate_count))) {
2021 if (mapping->truncate_count == 0)
2022 reset_vma_truncate_counts(mapping);
2023 mapping->truncate_count++;
2024 }
2025 details.truncate_count = mapping->truncate_count;
2026
2027 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
2028 unmap_mapping_range_tree(&mapping->i_mmap, &details);
2029 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
2030 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
2031 spin_unlock(&mapping->i_mmap_lock);
2032}
2033EXPORT_SYMBOL(unmap_mapping_range);
2034
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07002035/**
2036 * vmtruncate - unmap mappings "freed" by truncate() syscall
2037 * @inode: inode of the file used
2038 * @offset: file offset to start truncating
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 *
2040 * NOTE! We have to be ready to update the memory sharing
2041 * between the file and the memory map for a potential last
2042 * incomplete page. Ugly, but necessary.
2043 */
2044int vmtruncate(struct inode * inode, loff_t offset)
2045{
Christoph Hellwig61d50482008-02-04 22:28:56 -08002046 if (inode->i_size < offset) {
2047 unsigned long limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048
Christoph Hellwig61d50482008-02-04 22:28:56 -08002049 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
2050 if (limit != RLIM_INFINITY && offset > limit)
2051 goto out_sig;
2052 if (offset > inode->i_sb->s_maxbytes)
2053 goto out_big;
2054 i_size_write(inode, offset);
2055 } else {
2056 struct address_space *mapping = inode->i_mapping;
Nick Piggind00806b2007-07-19 01:46:57 -07002057
Christoph Hellwig61d50482008-02-04 22:28:56 -08002058 /*
2059 * truncation of in-use swapfiles is disallowed - it would
2060 * cause subsequent swapout to scribble on the now-freed
2061 * blocks.
2062 */
2063 if (IS_SWAPFILE(inode))
2064 return -ETXTBSY;
2065 i_size_write(inode, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066
Christoph Hellwig61d50482008-02-04 22:28:56 -08002067 /*
2068 * unmap_mapping_range is called twice, first simply for
2069 * efficiency so that truncate_inode_pages does fewer
2070 * single-page unmaps. However after this first call, and
2071 * before truncate_inode_pages finishes, it is possible for
2072 * private pages to be COWed, which remain after
2073 * truncate_inode_pages finishes, hence the second
2074 * unmap_mapping_range call must be made for correctness.
2075 */
2076 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2077 truncate_inode_pages(mapping, offset);
2078 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2079 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 if (inode->i_op && inode->i_op->truncate)
2082 inode->i_op->truncate(inode);
2083 return 0;
Christoph Hellwig61d50482008-02-04 22:28:56 -08002084
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085out_sig:
2086 send_sig(SIGXFSZ, current, 0);
2087out_big:
2088 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090EXPORT_SYMBOL(vmtruncate);
2091
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002092int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
2093{
2094 struct address_space *mapping = inode->i_mapping;
2095
2096 /*
2097 * If the underlying filesystem is not going to provide
2098 * a way to truncate a range of blocks (punch a hole) -
2099 * we should return failure right now.
2100 */
2101 if (!inode->i_op || !inode->i_op->truncate_range)
2102 return -ENOSYS;
2103
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002104 mutex_lock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002105 down_write(&inode->i_alloc_sem);
2106 unmap_mapping_range(mapping, offset, (end - offset), 1);
2107 truncate_inode_pages_range(mapping, offset, end);
Nick Piggind00806b2007-07-19 01:46:57 -07002108 unmap_mapping_range(mapping, offset, (end - offset), 1);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002109 inode->i_op->truncate_range(inode, offset, end);
2110 up_write(&inode->i_alloc_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002111 mutex_unlock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002112
2113 return 0;
2114}
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002115
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002117 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2118 * but allow concurrent faults), and pte mapped but not yet locked.
2119 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002121static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2122 unsigned long address, pte_t *page_table, pmd_t *pmd,
2123 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002125 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 struct page *page;
Hugh Dickins65500d22005-10-29 18:15:59 -07002127 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 pte_t pte;
Nick Piggin83c54072007-07-19 01:47:05 -07002129 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002131 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002132 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002133
2134 entry = pte_to_swp_entry(orig_pte);
Christoph Lameter06972122006-06-23 02:03:35 -07002135 if (is_migration_entry(entry)) {
2136 migration_entry_wait(mm, pmd, address);
2137 goto out;
2138 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002139 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 page = lookup_swap_cache(entry);
2141 if (!page) {
Ashwin Chaugule098fe652006-12-06 20:31:54 -08002142 grab_swap_token(); /* Contend for token _before_ read-in */
Hugh Dickins02098fe2008-02-04 22:28:42 -08002143 page = swapin_readahead(entry,
2144 GFP_HIGHUSER_MOVABLE, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 if (!page) {
2146 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002147 * Back out if somebody else faulted in this pte
2148 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002150 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 if (likely(pte_same(*page_table, orig_pte)))
2152 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002153 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002154 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 }
2156
2157 /* Had to read the page from swap area: Major fault */
2158 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002159 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 }
2161
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002162 if (mem_cgroup_charge(page, mm, GFP_KERNEL)) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002163 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2164 ret = VM_FAULT_OOM;
2165 goto out;
2166 }
2167
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 mark_page_accessed(page);
2169 lock_page(page);
Balbir Singh20a10222007-11-14 17:00:33 -08002170 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171
2172 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002173 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002175 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002176 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002177 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002178
2179 if (unlikely(!PageUptodate(page))) {
2180 ret = VM_FAULT_SIGBUS;
2181 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 }
2183
2184 /* The page isn't present yet, go ahead with the fault. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185
Hugh Dickins42946212005-10-29 18:16:05 -07002186 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 pte = mk_pte(page, vma->vm_page_prot);
2188 if (write_access && can_share_swap_page(page)) {
2189 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2190 write_access = 0;
2191 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192
2193 flush_icache_page(vma, page);
2194 set_pte_at(mm, address, page_table, pte);
2195 page_add_anon_rmap(page, vma, address);
2196
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002197 swap_free(entry);
2198 if (vm_swap_full())
2199 remove_exclusive_swap_page(page);
2200 unlock_page(page);
2201
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 if (write_access) {
Hugh Dickins61469f12008-03-04 14:29:04 -08002203 ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
2204 if (ret & VM_FAULT_ERROR)
2205 ret &= VM_FAULT_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 goto out;
2207 }
2208
2209 /* No need to invalidate - it was non-present before */
2210 update_mmu_cache(vma, address, pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07002211unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002212 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213out:
2214 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002215out_nomap:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002216 mem_cgroup_uncharge_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002217 pte_unmap_unlock(page_table, ptl);
Kirill Korotaevb8107482005-05-16 21:53:50 -07002218 unlock_page(page);
2219 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002220 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221}
2222
2223/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002224 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2225 * but allow concurrent faults), and pte mapped but not yet locked.
2226 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002228static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2229 unsigned long address, pte_t *page_table, pmd_t *pmd,
2230 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002232 struct page *page;
2233 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235
Nick Piggin557ed1f2007-10-16 01:24:40 -07002236 /* Allocate our own private page. */
2237 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238
Nick Piggin557ed1f2007-10-16 01:24:40 -07002239 if (unlikely(anon_vma_prepare(vma)))
2240 goto oom;
2241 page = alloc_zeroed_user_highpage_movable(vma, address);
2242 if (!page)
2243 goto oom;
Nick Piggin0ed361d2008-02-04 22:29:34 -08002244 __SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002246 if (mem_cgroup_charge(page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002247 goto oom_free_page;
2248
Nick Piggin557ed1f2007-10-16 01:24:40 -07002249 entry = mk_pte(page, vma->vm_page_prot);
2250 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002251
Nick Piggin557ed1f2007-10-16 01:24:40 -07002252 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2253 if (!pte_none(*page_table))
2254 goto release;
2255 inc_mm_counter(mm, anon_rss);
2256 lru_cache_add_active(page);
2257 page_add_new_anon_rmap(page, vma, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002258 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259
2260 /* No need to invalidate - it was non-present before */
Hugh Dickins65500d22005-10-29 18:15:59 -07002261 update_mmu_cache(vma, address, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002262unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002263 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002264 return 0;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002265release:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002266 mem_cgroup_uncharge_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002267 page_cache_release(page);
2268 goto unlock;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002269oom_free_page:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002270 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002271oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 return VM_FAULT_OOM;
2273}
2274
2275/*
Nick Piggin54cb8822007-07-19 01:46:59 -07002276 * __do_fault() tries to create a new page mapping. It aggressively
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277 * tries to share with existing pages, but makes a separate copy if
Nick Piggin54cb8822007-07-19 01:46:59 -07002278 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2279 * the next page fault.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 *
2281 * As this is called only for pages that do not currently exist, we
2282 * do not need to flush old virtual caches or the TLB.
2283 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002284 * We enter with non-exclusive mmap_sem (to exclude vma changes,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002285 * but allow concurrent faults), and pte neither mapped nor locked.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002286 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 */
Nick Piggin54cb8822007-07-19 01:46:59 -07002288static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002289 unsigned long address, pmd_t *pmd,
Nick Piggin54cb8822007-07-19 01:46:59 -07002290 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291{
Hugh Dickins16abfa02007-10-04 16:56:06 +01002292 pte_t *page_table;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002293 spinlock_t *ptl;
Nick Piggind0217ac2007-07-19 01:47:03 -07002294 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 int anon = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002297 struct page *dirty_page = NULL;
Nick Piggind0217ac2007-07-19 01:47:03 -07002298 struct vm_fault vmf;
2299 int ret;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002300 int page_mkwrite = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07002301
Nick Piggind0217ac2007-07-19 01:47:03 -07002302 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2303 vmf.pgoff = pgoff;
2304 vmf.flags = flags;
2305 vmf.page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306
Nick Piggin3c18ddd2008-04-28 02:12:10 -07002307 ret = vma->vm_ops->fault(vma, &vmf);
2308 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2309 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310
Nick Piggind00806b2007-07-19 01:46:57 -07002311 /*
Nick Piggind0217ac2007-07-19 01:47:03 -07002312 * For consistency in subsequent calls, make the faulted page always
Nick Piggind00806b2007-07-19 01:46:57 -07002313 * locked.
2314 */
Nick Piggin83c54072007-07-19 01:47:05 -07002315 if (unlikely(!(ret & VM_FAULT_LOCKED)))
Nick Piggind0217ac2007-07-19 01:47:03 -07002316 lock_page(vmf.page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002317 else
Nick Piggind0217ac2007-07-19 01:47:03 -07002318 VM_BUG_ON(!PageLocked(vmf.page));
Nick Piggind00806b2007-07-19 01:46:57 -07002319
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 /*
2321 * Should we do an early C-O-W break?
2322 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002323 page = vmf.page;
Nick Piggin54cb8822007-07-19 01:46:59 -07002324 if (flags & FAULT_FLAG_WRITE) {
David Howells9637a5e2006-06-23 02:03:43 -07002325 if (!(vma->vm_flags & VM_SHARED)) {
Nick Piggin54cb8822007-07-19 01:46:59 -07002326 anon = 1;
Nick Piggind00806b2007-07-19 01:46:57 -07002327 if (unlikely(anon_vma_prepare(vma))) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002328 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002329 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002330 }
Nick Piggin83c54072007-07-19 01:47:05 -07002331 page = alloc_page_vma(GFP_HIGHUSER_MOVABLE,
2332 vma, address);
Nick Piggind00806b2007-07-19 01:46:57 -07002333 if (!page) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002334 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002335 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002336 }
Nick Piggind0217ac2007-07-19 01:47:03 -07002337 copy_user_highpage(page, vmf.page, address, vma);
Nick Piggin0ed361d2008-02-04 22:29:34 -08002338 __SetPageUptodate(page);
David Howells9637a5e2006-06-23 02:03:43 -07002339 } else {
Nick Piggin54cb8822007-07-19 01:46:59 -07002340 /*
2341 * If the page will be shareable, see if the backing
David Howells9637a5e2006-06-23 02:03:43 -07002342 * address space wants to know that the page is about
Nick Piggin54cb8822007-07-19 01:46:59 -07002343 * to become writable
2344 */
Mark Fasheh69676142007-07-19 01:47:00 -07002345 if (vma->vm_ops->page_mkwrite) {
2346 unlock_page(page);
2347 if (vma->vm_ops->page_mkwrite(vma, page) < 0) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002348 ret = VM_FAULT_SIGBUS;
2349 anon = 1; /* no anon but release vmf.page */
Mark Fasheh69676142007-07-19 01:47:00 -07002350 goto out_unlocked;
2351 }
2352 lock_page(page);
Nick Piggind0217ac2007-07-19 01:47:03 -07002353 /*
2354 * XXX: this is not quite right (racy vs
2355 * invalidate) to unlock and relock the page
2356 * like this, however a better fix requires
2357 * reworking page_mkwrite locking API, which
2358 * is better done later.
2359 */
2360 if (!page->mapping) {
Nick Piggin83c54072007-07-19 01:47:05 -07002361 ret = 0;
Nick Piggind0217ac2007-07-19 01:47:03 -07002362 anon = 1; /* no anon but release vmf.page */
2363 goto out;
2364 }
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002365 page_mkwrite = 1;
David Howells9637a5e2006-06-23 02:03:43 -07002366 }
2367 }
Nick Piggin54cb8822007-07-19 01:46:59 -07002368
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 }
2370
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002371 if (mem_cgroup_charge(page, mm, GFP_KERNEL)) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002372 ret = VM_FAULT_OOM;
2373 goto out;
2374 }
2375
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002376 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377
2378 /*
2379 * This silly early PAGE_DIRTY setting removes a race
2380 * due to the bad i386 page protection. But it's valid
2381 * for other architectures too.
2382 *
2383 * Note that if write_access is true, we either now have
2384 * an exclusive copy of the page, or this is a shared mapping,
2385 * so we can make it writable and dirty to avoid having to
2386 * handle that later.
2387 */
2388 /* Only go through if we didn't race with anybody else... */
Nick Piggin54cb8822007-07-19 01:46:59 -07002389 if (likely(pte_same(*page_table, orig_pte))) {
Nick Piggind00806b2007-07-19 01:46:57 -07002390 flush_icache_page(vma, page);
2391 entry = mk_pte(page, vma->vm_page_prot);
Nick Piggin54cb8822007-07-19 01:46:59 -07002392 if (flags & FAULT_FLAG_WRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2394 set_pte_at(mm, address, page_table, entry);
2395 if (anon) {
Nick Piggind00806b2007-07-19 01:46:57 -07002396 inc_mm_counter(mm, anon_rss);
2397 lru_cache_add_active(page);
2398 page_add_new_anon_rmap(page, vma, address);
Hugh Dickinsf57e88a2005-11-21 21:32:19 -08002399 } else {
Hugh Dickins42946212005-10-29 18:16:05 -07002400 inc_mm_counter(mm, file_rss);
Nick Piggind00806b2007-07-19 01:46:57 -07002401 page_add_file_rmap(page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002402 if (flags & FAULT_FLAG_WRITE) {
Nick Piggind00806b2007-07-19 01:46:57 -07002403 dirty_page = page;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002404 get_page(dirty_page);
2405 }
Hugh Dickins42946212005-10-29 18:16:05 -07002406 }
Nick Piggind00806b2007-07-19 01:46:57 -07002407
2408 /* no need to invalidate: a not-present page won't be cached */
2409 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 } else {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002411 mem_cgroup_uncharge_page(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002412 if (anon)
2413 page_cache_release(page);
2414 else
Nick Piggin54cb8822007-07-19 01:46:59 -07002415 anon = 1; /* no anon but release faulted_page */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 }
2417
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002418 pte_unmap_unlock(page_table, ptl);
Nick Piggind00806b2007-07-19 01:46:57 -07002419
2420out:
Nick Piggind0217ac2007-07-19 01:47:03 -07002421 unlock_page(vmf.page);
Mark Fasheh69676142007-07-19 01:47:00 -07002422out_unlocked:
Nick Piggind00806b2007-07-19 01:46:57 -07002423 if (anon)
Nick Piggind0217ac2007-07-19 01:47:03 -07002424 page_cache_release(vmf.page);
Nick Piggind00806b2007-07-19 01:46:57 -07002425 else if (dirty_page) {
Anton Salikhmetov8f7b3d12008-01-23 02:21:18 +03002426 if (vma->vm_file)
2427 file_update_time(vma->vm_file);
2428
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002429 set_page_dirty_balance(dirty_page, page_mkwrite);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002430 put_page(dirty_page);
2431 }
Nick Piggind00806b2007-07-19 01:46:57 -07002432
Nick Piggin83c54072007-07-19 01:47:05 -07002433 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07002434}
Nick Piggind00806b2007-07-19 01:46:57 -07002435
Nick Piggin54cb8822007-07-19 01:46:59 -07002436static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2437 unsigned long address, pte_t *page_table, pmd_t *pmd,
2438 int write_access, pte_t orig_pte)
2439{
2440 pgoff_t pgoff = (((address & PAGE_MASK)
Dean Nelson0da7e012007-10-16 01:24:45 -07002441 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
Nick Piggin54cb8822007-07-19 01:46:59 -07002442 unsigned int flags = (write_access ? FAULT_FLAG_WRITE : 0);
2443
Hugh Dickins16abfa02007-10-04 16:56:06 +01002444 pte_unmap(page_table);
2445 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Nick Piggin54cb8822007-07-19 01:46:59 -07002446}
2447
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448
2449/*
Jes Sorensenf4b81802006-09-27 01:50:10 -07002450 * do_no_pfn() tries to create a new page mapping for a page without
2451 * a struct_page backing it
2452 *
2453 * As this is called only for pages that do not currently exist, we
2454 * do not need to flush old virtual caches or the TLB.
2455 *
2456 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2457 * but allow concurrent faults), and pte mapped but not yet locked.
2458 * We return with mmap_sem still held, but pte unmapped and unlocked.
2459 *
2460 * It is expected that the ->nopfn handler always returns the same pfn
2461 * for a given virtual mapping.
2462 *
2463 * Mark this `noinline' to prevent it from bloating the main pagefault code.
2464 */
2465static noinline int do_no_pfn(struct mm_struct *mm, struct vm_area_struct *vma,
2466 unsigned long address, pte_t *page_table, pmd_t *pmd,
2467 int write_access)
2468{
2469 spinlock_t *ptl;
2470 pte_t entry;
2471 unsigned long pfn;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002472
2473 pte_unmap(page_table);
Jared Hulbertb379d792008-04-28 02:12:58 -07002474 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
2475 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
Jes Sorensenf4b81802006-09-27 01:50:10 -07002476
2477 pfn = vma->vm_ops->nopfn(vma, address & PAGE_MASK);
Jared Hulbertb379d792008-04-28 02:12:58 -07002478
2479 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
2480
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002481 if (unlikely(pfn == NOPFN_OOM))
Jes Sorensenf4b81802006-09-27 01:50:10 -07002482 return VM_FAULT_OOM;
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002483 else if (unlikely(pfn == NOPFN_SIGBUS))
Jes Sorensenf4b81802006-09-27 01:50:10 -07002484 return VM_FAULT_SIGBUS;
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002485 else if (unlikely(pfn == NOPFN_REFAULT))
Nick Piggin83c54072007-07-19 01:47:05 -07002486 return 0;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002487
2488 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2489
2490 /* Only go through if we didn't race with anybody else... */
2491 if (pte_none(*page_table)) {
2492 entry = pfn_pte(pfn, vma->vm_page_prot);
2493 if (write_access)
2494 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2495 set_pte_at(mm, address, page_table, entry);
2496 }
2497 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002498 return 0;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002499}
2500
2501/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 * Fault of a previously existing named mapping. Repopulate the pte
2503 * from the encoded file_pte if possible. This enables swappable
2504 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002505 *
2506 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2507 * but allow concurrent faults), and pte mapped but not yet locked.
2508 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002510static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins65500d22005-10-29 18:15:59 -07002511 unsigned long address, pte_t *page_table, pmd_t *pmd,
2512 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513{
Nick Piggind0217ac2007-07-19 01:47:03 -07002514 unsigned int flags = FAULT_FLAG_NONLINEAR |
2515 (write_access ? FAULT_FLAG_WRITE : 0);
Hugh Dickins65500d22005-10-29 18:15:59 -07002516 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002518 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Nick Piggin83c54072007-07-19 01:47:05 -07002519 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520
Nick Piggind0217ac2007-07-19 01:47:03 -07002521 if (unlikely(!(vma->vm_flags & VM_NONLINEAR) ||
2522 !(vma->vm_flags & VM_CAN_NONLINEAR))) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002523 /*
2524 * Page table corrupted: show pte and kill process.
2525 */
Nick Pigginb5810032005-10-29 18:16:12 -07002526 print_bad_pte(vma, orig_pte, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002527 return VM_FAULT_OOM;
2528 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002529
2530 pgoff = pte_to_pgoff(orig_pte);
Hugh Dickins16abfa02007-10-04 16:56:06 +01002531 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532}
2533
2534/*
2535 * These routines also need to handle stuff like marking pages dirty
2536 * and/or accessed for architectures that don't do it in hardware (most
2537 * RISC architectures). The early dirtying is also good on the i386.
2538 *
2539 * There is also a hook called "update_mmu_cache()" that architectures
2540 * with external mmu caches can use to update those (ie the Sparc or
2541 * PowerPC hashed page tables that act as extended TLBs).
2542 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07002543 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2544 * but allow concurrent faults), and pte mapped but not yet locked.
2545 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 */
2547static inline int handle_pte_fault(struct mm_struct *mm,
Hugh Dickins65500d22005-10-29 18:15:59 -07002548 struct vm_area_struct *vma, unsigned long address,
2549 pte_t *pte, pmd_t *pmd, int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550{
2551 pte_t entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002552 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002554 entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002556 if (pte_none(entry)) {
Jes Sorensenf4b81802006-09-27 01:50:10 -07002557 if (vma->vm_ops) {
Nick Piggin3c18ddd2008-04-28 02:12:10 -07002558 if (likely(vma->vm_ops->fault))
Nick Piggin54cb8822007-07-19 01:46:59 -07002559 return do_linear_fault(mm, vma, address,
2560 pte, pmd, write_access, entry);
Jes Sorensenf4b81802006-09-27 01:50:10 -07002561 if (unlikely(vma->vm_ops->nopfn))
2562 return do_no_pfn(mm, vma, address, pte,
2563 pmd, write_access);
2564 }
2565 return do_anonymous_page(mm, vma, address,
2566 pte, pmd, write_access);
Hugh Dickins65500d22005-10-29 18:15:59 -07002567 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 if (pte_file(entry))
Nick Piggind0217ac2007-07-19 01:47:03 -07002569 return do_nonlinear_fault(mm, vma, address,
Hugh Dickins65500d22005-10-29 18:15:59 -07002570 pte, pmd, write_access, entry);
2571 return do_swap_page(mm, vma, address,
2572 pte, pmd, write_access, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573 }
2574
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002575 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002576 spin_lock(ptl);
2577 if (unlikely(!pte_same(*pte, entry)))
2578 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 if (write_access) {
2580 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002581 return do_wp_page(mm, vma, address,
2582 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 entry = pte_mkdirty(entry);
2584 }
2585 entry = pte_mkyoung(entry);
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002586 if (ptep_set_access_flags(vma, address, pte, entry, write_access)) {
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002587 update_mmu_cache(vma, address, entry);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002588 } else {
2589 /*
2590 * This is needed only for protection faults but the arch code
2591 * is not yet telling us if this is a protection fault or not.
2592 * This still avoids useless tlb flushes for .text page faults
2593 * with threads.
2594 */
2595 if (write_access)
2596 flush_tlb_page(vma, address);
2597 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002598unlock:
2599 pte_unmap_unlock(pte, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002600 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601}
2602
2603/*
2604 * By the time we get here, we already hold the mm semaphore
2605 */
Nick Piggin83c54072007-07-19 01:47:05 -07002606int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 unsigned long address, int write_access)
2608{
2609 pgd_t *pgd;
2610 pud_t *pud;
2611 pmd_t *pmd;
2612 pte_t *pte;
2613
2614 __set_current_state(TASK_RUNNING);
2615
Christoph Lameterf8891e52006-06-30 01:55:45 -07002616 count_vm_event(PGFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01002618 if (unlikely(is_vm_hugetlb_page(vma)))
2619 return hugetlb_fault(mm, vma, address, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622 pud = pud_alloc(mm, pgd, address);
2623 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002624 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 pmd = pmd_alloc(mm, pud, address);
2626 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002627 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 pte = pte_alloc_map(mm, pmd, address);
2629 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002630 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631
Hugh Dickinsc74df322005-10-29 18:16:23 -07002632 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633}
2634
2635#ifndef __PAGETABLE_PUD_FOLDED
2636/*
2637 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002638 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002640int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002642 pud_t *new = pud_alloc_one(mm, address);
2643 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002644 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645
Nick Piggin362a61a2008-05-14 06:37:36 +02002646 smp_wmb(); /* See comment in __pte_alloc */
2647
Hugh Dickins872fec12005-10-29 18:16:21 -07002648 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002649 if (pgd_present(*pgd)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002650 pud_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002651 else
2652 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07002653 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002654 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655}
2656#endif /* __PAGETABLE_PUD_FOLDED */
2657
2658#ifndef __PAGETABLE_PMD_FOLDED
2659/*
2660 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002661 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002663int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002665 pmd_t *new = pmd_alloc_one(mm, address);
2666 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002667 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668
Nick Piggin362a61a2008-05-14 06:37:36 +02002669 smp_wmb(); /* See comment in __pte_alloc */
2670
Hugh Dickins872fec12005-10-29 18:16:21 -07002671 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07002673 if (pud_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002674 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002675 else
2676 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07002678 if (pgd_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002679 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002680 else
2681 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07002683 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002684 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685}
2686#endif /* __PAGETABLE_PMD_FOLDED */
2687
2688int make_pages_present(unsigned long addr, unsigned long end)
2689{
2690 int ret, len, write;
2691 struct vm_area_struct * vma;
2692
2693 vma = find_vma(current->mm, addr);
2694 if (!vma)
2695 return -1;
2696 write = (vma->vm_flags & VM_WRITE) != 0;
Eric Sesterhenn5bcb28b2006-03-26 18:30:52 +02002697 BUG_ON(addr >= end);
2698 BUG_ON(end > vma->vm_end);
Rolf Eike Beer68e116a2007-07-15 23:38:03 -07002699 len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 ret = get_user_pages(current, current->mm, addr,
2701 len, write, 0, NULL, NULL);
2702 if (ret < 0)
2703 return ret;
2704 return ret == len ? 0 : -1;
2705}
2706
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707#if !defined(__HAVE_ARCH_GATE_AREA)
2708
2709#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07002710static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711
2712static int __init gate_vma_init(void)
2713{
2714 gate_vma.vm_mm = NULL;
2715 gate_vma.vm_start = FIXADDR_USER_START;
2716 gate_vma.vm_end = FIXADDR_USER_END;
Roland McGrathb6558c42007-01-26 00:56:47 -08002717 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
2718 gate_vma.vm_page_prot = __P101;
Roland McGrathf47aef52007-01-26 00:56:49 -08002719 /*
2720 * Make sure the vDSO gets into every core dump.
2721 * Dumping its contents makes post-mortem fully interpretable later
2722 * without matching up the same kernel and hardware config to see
2723 * what PC values meant.
2724 */
2725 gate_vma.vm_flags |= VM_ALWAYSDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726 return 0;
2727}
2728__initcall(gate_vma_init);
2729#endif
2730
2731struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2732{
2733#ifdef AT_SYSINFO_EHDR
2734 return &gate_vma;
2735#else
2736 return NULL;
2737#endif
2738}
2739
2740int in_gate_area_no_task(unsigned long addr)
2741{
2742#ifdef AT_SYSINFO_EHDR
2743 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2744 return 1;
2745#endif
2746 return 0;
2747}
2748
2749#endif /* __HAVE_ARCH_GATE_AREA */
David Howells0ec76a12006-09-27 01:50:15 -07002750
2751/*
2752 * Access another process' address space.
2753 * Source/target buffer must be kernel space,
2754 * Do not walk the page table directly, use get_user_pages
2755 */
2756int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2757{
2758 struct mm_struct *mm;
2759 struct vm_area_struct *vma;
2760 struct page *page;
2761 void *old_buf = buf;
2762
2763 mm = get_task_mm(tsk);
2764 if (!mm)
2765 return 0;
2766
2767 down_read(&mm->mmap_sem);
Simon Arlott183ff222007-10-20 01:27:18 +02002768 /* ignore errors, just check how much was successfully transferred */
David Howells0ec76a12006-09-27 01:50:15 -07002769 while (len) {
2770 int bytes, ret, offset;
2771 void *maddr;
2772
2773 ret = get_user_pages(tsk, mm, addr, 1,
2774 write, 1, &page, &vma);
2775 if (ret <= 0)
2776 break;
2777
2778 bytes = len;
2779 offset = addr & (PAGE_SIZE-1);
2780 if (bytes > PAGE_SIZE-offset)
2781 bytes = PAGE_SIZE-offset;
2782
2783 maddr = kmap(page);
2784 if (write) {
2785 copy_to_user_page(vma, page, addr,
2786 maddr + offset, buf, bytes);
2787 set_page_dirty_lock(page);
2788 } else {
2789 copy_from_user_page(vma, page, addr,
2790 buf, maddr + offset, bytes);
2791 }
2792 kunmap(page);
2793 page_cache_release(page);
2794 len -= bytes;
2795 buf += bytes;
2796 addr += bytes;
2797 }
2798 up_read(&mm->mmap_sem);
2799 mmput(mm);
2800
2801 return buf - old_buf;
2802}
Andi Kleen03252912008-01-30 13:33:18 +01002803
2804/*
2805 * Print the name of a VMA.
2806 */
2807void print_vma_addr(char *prefix, unsigned long ip)
2808{
2809 struct mm_struct *mm = current->mm;
2810 struct vm_area_struct *vma;
2811
Ingo Molnare8bff742008-02-13 20:21:06 +01002812 /*
2813 * Do not print if we are in atomic
2814 * contexts (in exception stacks, etc.):
2815 */
2816 if (preempt_count())
2817 return;
2818
Andi Kleen03252912008-01-30 13:33:18 +01002819 down_read(&mm->mmap_sem);
2820 vma = find_vma(mm, ip);
2821 if (vma && vma->vm_file) {
2822 struct file *f = vma->vm_file;
2823 char *buf = (char *)__get_free_page(GFP_KERNEL);
2824 if (buf) {
2825 char *p, *s;
2826
Jan Blunckcf28b482008-02-14 19:38:44 -08002827 p = d_path(&f->f_path, buf, PAGE_SIZE);
Andi Kleen03252912008-01-30 13:33:18 +01002828 if (IS_ERR(p))
2829 p = "?";
2830 s = strrchr(p, '/');
2831 if (s)
2832 p = s+1;
2833 printk("%s%s[%lx+%lx]", prefix, p,
2834 vma->vm_start,
2835 vma->vm_end - vma->vm_start);
2836 free_page((unsigned long)buf);
2837 }
2838 }
2839 up_read(&current->mm->mmap_sem);
2840}