blob: d14b251a25a638dec1b5327a6261a67660ba67d5 [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 Torvalds672ca282008-06-23 11:21:37 -07001048/* Can we do the FOLL_ANON optimization? */
1049static inline int use_zero_page(struct vm_area_struct *vma)
1050{
1051 /*
1052 * We don't want to optimize FOLL_ANON for make_pages_present()
1053 * when it tries to page in a VM_LOCKED region. As to VM_SHARED,
1054 * we want to get the page from the page tables to make sure
1055 * that we serialize and update with any other user of that
1056 * mapping.
1057 */
1058 if (vma->vm_flags & (VM_LOCKED | VM_SHARED))
1059 return 0;
1060 /*
1061 * And if we have a fault or a nopfn routine, it's not an
1062 * anonymous region.
1063 */
1064 return !vma->vm_ops ||
1065 (!vma->vm_ops->fault && !vma->vm_ops->nopfn);
1066}
1067
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
1069 unsigned long start, int len, int write, int force,
1070 struct page **pages, struct vm_area_struct **vmas)
1071{
1072 int i;
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001073 unsigned int vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074
Jonathan Corbet900cf082008-02-11 16:17:33 -07001075 if (len <= 0)
1076 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 /*
1078 * Require read or write permissions.
1079 * If 'force' is set, we only require the "MAY" flags.
1080 */
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001081 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
1082 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 i = 0;
1084
1085 do {
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001086 struct vm_area_struct *vma;
1087 unsigned int foll_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088
1089 vma = find_extend_vma(mm, start);
1090 if (!vma && in_gate_area(tsk, start)) {
1091 unsigned long pg = start & PAGE_MASK;
1092 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
1093 pgd_t *pgd;
1094 pud_t *pud;
1095 pmd_t *pmd;
1096 pte_t *pte;
1097 if (write) /* user gate pages are read-only */
1098 return i ? : -EFAULT;
1099 if (pg > TASK_SIZE)
1100 pgd = pgd_offset_k(pg);
1101 else
1102 pgd = pgd_offset_gate(mm, pg);
1103 BUG_ON(pgd_none(*pgd));
1104 pud = pud_offset(pgd, pg);
1105 BUG_ON(pud_none(*pud));
1106 pmd = pmd_offset(pud, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001107 if (pmd_none(*pmd))
1108 return i ? : -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 pte = pte_offset_map(pmd, pg);
Hugh Dickins690dbe12005-08-01 21:11:42 -07001110 if (pte_none(*pte)) {
1111 pte_unmap(pte);
1112 return i ? : -EFAULT;
1113 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 if (pages) {
Nick Pigginfa2a4552005-11-29 18:43:17 +11001115 struct page *page = vm_normal_page(gate_vma, start, *pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001116 pages[i] = page;
1117 if (page)
1118 get_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 }
1120 pte_unmap(pte);
1121 if (vmas)
1122 vmas[i] = gate_vma;
1123 i++;
1124 start += PAGE_SIZE;
1125 len--;
1126 continue;
1127 }
1128
Linus Torvalds1ff80382005-12-12 16:24:33 -08001129 if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001130 || !(vm_flags & vma->vm_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 return i ? : -EFAULT;
1132
1133 if (is_vm_hugetlb_page(vma)) {
1134 i = follow_hugetlb_page(mm, vma, pages, vmas,
Adam Litke5b23dbe2007-11-14 16:59:33 -08001135 &start, &len, i, write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 continue;
1137 }
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001138
1139 foll_flags = FOLL_TOUCH;
1140 if (pages)
1141 foll_flags |= FOLL_GET;
Linus Torvalds672ca282008-06-23 11:21:37 -07001142 if (!write && use_zero_page(vma))
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001143 foll_flags |= FOLL_ANON;
1144
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 do {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001146 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147
Ethan Solomita462e00c2007-07-15 23:38:16 -07001148 /*
1149 * If tsk is ooming, cut off its access to large memory
1150 * allocations. It has a pending SIGKILL, but it can't
1151 * be processed until returning to user space.
1152 */
1153 if (unlikely(test_tsk_thread_flag(tsk, TIF_MEMDIE)))
1154 return -ENOMEM;
1155
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001156 if (write)
1157 foll_flags |= FOLL_WRITE;
1158
1159 cond_resched();
Linus Torvalds6aab3412005-11-28 14:34:23 -08001160 while (!(page = follow_page(vma, start, foll_flags))) {
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001161 int ret;
Nick Piggin83c54072007-07-19 01:47:05 -07001162 ret = handle_mm_fault(mm, vma, start,
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001163 foll_flags & FOLL_WRITE);
Nick Piggin83c54072007-07-19 01:47:05 -07001164 if (ret & VM_FAULT_ERROR) {
1165 if (ret & VM_FAULT_OOM)
1166 return i ? i : -ENOMEM;
1167 else if (ret & VM_FAULT_SIGBUS)
1168 return i ? i : -EFAULT;
1169 BUG();
1170 }
1171 if (ret & VM_FAULT_MAJOR)
1172 tsk->maj_flt++;
1173 else
1174 tsk->min_flt++;
1175
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001176 /*
Nick Piggin83c54072007-07-19 01:47:05 -07001177 * The VM_FAULT_WRITE bit tells us that
1178 * do_wp_page has broken COW when necessary,
1179 * even if maybe_mkwrite decided not to set
1180 * pte_write. We can thus safely do subsequent
1181 * page lookups as if they were reads.
Linus Torvaldsa68d2eb2005-08-03 10:07:09 -07001182 */
1183 if (ret & VM_FAULT_WRITE)
Hugh Dickinsdeceb6c2005-10-29 18:16:33 -07001184 foll_flags &= ~FOLL_WRITE;
Nick Piggin83c54072007-07-19 01:47:05 -07001185
Benjamin Herrenschmidt7f7bbbe2006-10-06 00:43:53 -07001186 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 }
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001188 if (IS_ERR(page))
1189 return i ? i : PTR_ERR(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 if (pages) {
Hugh Dickins08ef4722005-06-21 17:15:10 -07001191 pages[i] = page;
James Bottomley03beb072006-03-26 01:36:57 -08001192
Russell Kinga6f36be2006-12-30 22:24:19 +00001193 flush_anon_page(vma, page, start);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001194 flush_dcache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 }
1196 if (vmas)
1197 vmas[i] = vma;
1198 i++;
1199 start += PAGE_SIZE;
1200 len--;
Hugh Dickins08ef4722005-06-21 17:15:10 -07001201 } while (len && start < vma->vm_end);
Hugh Dickins08ef4722005-06-21 17:15:10 -07001202 } while (len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 return i;
1204}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205EXPORT_SYMBOL(get_user_pages);
1206
Harvey Harrison920c7a52008-02-04 22:29:26 -08001207pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr,
1208 spinlock_t **ptl)
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001209{
1210 pgd_t * pgd = pgd_offset(mm, addr);
1211 pud_t * pud = pud_alloc(mm, pgd, addr);
1212 if (pud) {
Trond Myklebust49c91fb2005-11-29 19:27:22 -05001213 pmd_t * pmd = pmd_alloc(mm, pud, addr);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001214 if (pmd)
1215 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1216 }
1217 return NULL;
1218}
1219
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220/*
Linus Torvalds238f58d2005-11-29 13:01:56 -08001221 * This is the old fallback for page remapping.
1222 *
1223 * For historical reasons, it only allows reserved pages. Only
1224 * old drivers should use this, and they needed to mark their
1225 * pages reserved for the old functions anyway.
1226 */
Nick Piggin423bad62008-04-28 02:13:01 -07001227static int insert_page(struct vm_area_struct *vma, unsigned long addr,
1228 struct page *page, pgprot_t prot)
Linus Torvalds238f58d2005-11-29 13:01:56 -08001229{
Nick Piggin423bad62008-04-28 02:13:01 -07001230 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001231 int retval;
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001232 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001233 spinlock_t *ptl;
1234
Balbir Singhe1a1cd52008-02-07 00:14:02 -08001235 retval = mem_cgroup_charge(page, mm, GFP_KERNEL);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001236 if (retval)
1237 goto out;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001238
1239 retval = -EINVAL;
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001240 if (PageAnon(page))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001241 goto out_uncharge;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001242 retval = -ENOMEM;
1243 flush_dcache_page(page);
Linus Torvaldsc9cfcdd2005-11-29 14:03:14 -08001244 pte = get_locked_pte(mm, addr, &ptl);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001245 if (!pte)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001246 goto out_uncharge;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001247 retval = -EBUSY;
1248 if (!pte_none(*pte))
1249 goto out_unlock;
1250
1251 /* Ok, finally just insert the thing.. */
1252 get_page(page);
1253 inc_mm_counter(mm, file_rss);
1254 page_add_file_rmap(page);
1255 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1256
1257 retval = 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001258 pte_unmap_unlock(pte, ptl);
1259 return retval;
Linus Torvalds238f58d2005-11-29 13:01:56 -08001260out_unlock:
1261 pte_unmap_unlock(pte, ptl);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001262out_uncharge:
1263 mem_cgroup_uncharge_page(page);
Linus Torvalds238f58d2005-11-29 13:01:56 -08001264out:
1265 return retval;
1266}
1267
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001268/**
1269 * vm_insert_page - insert single page into user vma
1270 * @vma: user vma to map to
1271 * @addr: target user address of this page
1272 * @page: source kernel page
1273 *
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001274 * This allows drivers to insert individual pages they've allocated
1275 * into a user vma.
1276 *
1277 * The page has to be a nice clean _individual_ kernel allocation.
1278 * If you allocate a compound page, you need to have marked it as
1279 * such (__GFP_COMP), or manually just split the page up yourself
Nick Piggin8dfcc9b2006-03-22 00:08:05 -08001280 * (see split_page()).
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001281 *
1282 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1283 * took an arbitrary page protection parameter. This doesn't allow
1284 * that. Your vma protection will have to be set up correctly, which
1285 * means that if you want a shared writable mapping, you'd better
1286 * ask for a shared writable mapping!
1287 *
1288 * The page does not need to be reserved.
1289 */
Nick Piggin423bad62008-04-28 02:13:01 -07001290int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
1291 struct page *page)
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001292{
1293 if (addr < vma->vm_start || addr >= vma->vm_end)
1294 return -EFAULT;
1295 if (!page_count(page))
1296 return -EINVAL;
Linus Torvalds4d7672b2005-12-16 10:21:23 -08001297 vma->vm_flags |= VM_INSERTPAGE;
Nick Piggin423bad62008-04-28 02:13:01 -07001298 return insert_page(vma, addr, page, vma->vm_page_prot);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001299}
Linus Torvaldse3c33742005-12-03 20:48:11 -08001300EXPORT_SYMBOL(vm_insert_page);
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001301
Nick Piggin423bad62008-04-28 02:13:01 -07001302static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1303 unsigned long pfn, pgprot_t prot)
1304{
1305 struct mm_struct *mm = vma->vm_mm;
1306 int retval;
1307 pte_t *pte, entry;
1308 spinlock_t *ptl;
1309
1310 retval = -ENOMEM;
1311 pte = get_locked_pte(mm, addr, &ptl);
1312 if (!pte)
1313 goto out;
1314 retval = -EBUSY;
1315 if (!pte_none(*pte))
1316 goto out_unlock;
1317
1318 /* Ok, finally just insert the thing.. */
1319 entry = pte_mkspecial(pfn_pte(pfn, prot));
1320 set_pte_at(mm, addr, pte, entry);
1321 update_mmu_cache(vma, addr, entry); /* XXX: why not for insert_page? */
1322
1323 retval = 0;
1324out_unlock:
1325 pte_unmap_unlock(pte, ptl);
1326out:
1327 return retval;
1328}
1329
Nick Piggine0dc0d82007-02-12 00:51:36 -08001330/**
1331 * vm_insert_pfn - insert single pfn into user vma
1332 * @vma: user vma to map to
1333 * @addr: target user address of this page
1334 * @pfn: source kernel pfn
1335 *
1336 * Similar to vm_inert_page, this allows drivers to insert individual pages
1337 * they've allocated into a user vma. Same comments apply.
1338 *
1339 * This function should only be called from a vm_ops->fault handler, and
1340 * in that case the handler should return NULL.
1341 */
1342int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
Nick Piggin423bad62008-04-28 02:13:01 -07001343 unsigned long pfn)
Nick Piggine0dc0d82007-02-12 00:51:36 -08001344{
Nick Piggin7e675132008-04-28 02:13:00 -07001345 /*
1346 * Technically, architectures with pte_special can avoid all these
1347 * restrictions (same for remap_pfn_range). However we would like
1348 * consistency in testing and feature parity among all, so we should
1349 * try to keep these invariants in place for everybody.
1350 */
Jared Hulbertb379d792008-04-28 02:12:58 -07001351 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
1352 BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
1353 (VM_PFNMAP|VM_MIXEDMAP));
1354 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1355 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
Nick Piggine0dc0d82007-02-12 00:51:36 -08001356
Nick Piggin423bad62008-04-28 02:13:01 -07001357 if (addr < vma->vm_start || addr >= vma->vm_end)
1358 return -EFAULT;
1359 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
Nick Piggine0dc0d82007-02-12 00:51:36 -08001360}
1361EXPORT_SYMBOL(vm_insert_pfn);
1362
Nick Piggin423bad62008-04-28 02:13:01 -07001363int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
1364 unsigned long pfn)
1365{
1366 BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
1367
1368 if (addr < vma->vm_start || addr >= vma->vm_end)
1369 return -EFAULT;
1370
1371 /*
1372 * If we don't have pte special, then we have to use the pfn_valid()
1373 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1374 * refcount the page if pfn_valid is true (hence insert_page rather
1375 * than insert_pfn).
1376 */
1377 if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) {
1378 struct page *page;
1379
1380 page = pfn_to_page(pfn);
1381 return insert_page(vma, addr, page, vma->vm_page_prot);
1382 }
1383 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
1384}
1385EXPORT_SYMBOL(vm_insert_mixed);
1386
Linus Torvaldsa145dd42005-11-30 09:35:19 -08001387/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 * maps a range of physical memory into the requested pages. the old
1389 * mappings are removed. any references to nonexistent pages results
1390 * in null mappings (currently treated as "copy-on-access")
1391 */
1392static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1393 unsigned long addr, unsigned long end,
1394 unsigned long pfn, pgprot_t prot)
1395{
1396 pte_t *pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -07001397 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398
Hugh Dickinsc74df322005-10-29 18:16:23 -07001399 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 if (!pte)
1401 return -ENOMEM;
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001402 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 do {
1404 BUG_ON(!pte_none(*pte));
Nick Piggin7e675132008-04-28 02:13:00 -07001405 set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 pfn++;
1407 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -07001408 arch_leave_lazy_mmu_mode();
Hugh Dickinsc74df322005-10-29 18:16:23 -07001409 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 return 0;
1411}
1412
1413static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1414 unsigned long addr, unsigned long end,
1415 unsigned long pfn, pgprot_t prot)
1416{
1417 pmd_t *pmd;
1418 unsigned long next;
1419
1420 pfn -= addr >> PAGE_SHIFT;
1421 pmd = pmd_alloc(mm, pud, addr);
1422 if (!pmd)
1423 return -ENOMEM;
1424 do {
1425 next = pmd_addr_end(addr, end);
1426 if (remap_pte_range(mm, pmd, addr, next,
1427 pfn + (addr >> PAGE_SHIFT), prot))
1428 return -ENOMEM;
1429 } while (pmd++, addr = next, addr != end);
1430 return 0;
1431}
1432
1433static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1434 unsigned long addr, unsigned long end,
1435 unsigned long pfn, pgprot_t prot)
1436{
1437 pud_t *pud;
1438 unsigned long next;
1439
1440 pfn -= addr >> PAGE_SHIFT;
1441 pud = pud_alloc(mm, pgd, addr);
1442 if (!pud)
1443 return -ENOMEM;
1444 do {
1445 next = pud_addr_end(addr, end);
1446 if (remap_pmd_range(mm, pud, addr, next,
1447 pfn + (addr >> PAGE_SHIFT), prot))
1448 return -ENOMEM;
1449 } while (pud++, addr = next, addr != end);
1450 return 0;
1451}
1452
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07001453/**
1454 * remap_pfn_range - remap kernel memory to userspace
1455 * @vma: user vma to map to
1456 * @addr: target user address to start at
1457 * @pfn: physical address of kernel memory
1458 * @size: size of map area
1459 * @prot: page protection flags for this mapping
1460 *
1461 * Note: this is only safe if the mm semaphore is held when called.
1462 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1464 unsigned long pfn, unsigned long size, pgprot_t prot)
1465{
1466 pgd_t *pgd;
1467 unsigned long next;
Hugh Dickins2d15cab2005-06-25 14:54:33 -07001468 unsigned long end = addr + PAGE_ALIGN(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 struct mm_struct *mm = vma->vm_mm;
1470 int err;
1471
1472 /*
1473 * Physically remapped pages are special. Tell the
1474 * rest of the world about it:
1475 * VM_IO tells people not to look at these pages
1476 * (accesses can have side effects).
Hugh Dickins0b14c172005-11-21 21:32:15 -08001477 * VM_RESERVED is specified all over the place, because
1478 * in 2.4 it kept swapout's vma scan off this vma; but
1479 * in 2.6 the LRU scan won't even find its pages, so this
1480 * flag means no more than count its pages in reserved_vm,
1481 * and omit it from core dump, even when VM_IO turned off.
Linus Torvalds6aab3412005-11-28 14:34:23 -08001482 * VM_PFNMAP tells the core MM that the base pages are just
1483 * raw PFN mappings, and do not have a "struct page" associated
1484 * with them.
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001485 *
1486 * There's a horrible special case to handle copy-on-write
1487 * behaviour that some programs depend on. We mark the "original"
1488 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 */
Linus Torvalds67121172005-12-11 20:38:17 -08001490 if (is_cow_mapping(vma->vm_flags)) {
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001491 if (addr != vma->vm_start || end != vma->vm_end)
Linus Torvalds7fc7e2e2005-12-11 19:57:52 -08001492 return -EINVAL;
Linus Torvaldsfb155c12005-12-11 19:46:02 -08001493 vma->vm_pgoff = pfn;
1494 }
1495
Linus Torvalds6aab3412005-11-28 14:34:23 -08001496 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497
1498 BUG_ON(addr >= end);
1499 pfn -= addr >> PAGE_SHIFT;
1500 pgd = pgd_offset(mm, addr);
1501 flush_cache_range(vma, addr, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 do {
1503 next = pgd_addr_end(addr, end);
1504 err = remap_pud_range(mm, pgd, addr, next,
1505 pfn + (addr >> PAGE_SHIFT), prot);
1506 if (err)
1507 break;
1508 } while (pgd++, addr = next, addr != end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 return err;
1510}
1511EXPORT_SYMBOL(remap_pfn_range);
1512
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001513static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1514 unsigned long addr, unsigned long end,
1515 pte_fn_t fn, void *data)
1516{
1517 pte_t *pte;
1518 int err;
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001519 pgtable_t token;
Borislav Petkov94909912007-05-06 14:49:17 -07001520 spinlock_t *uninitialized_var(ptl);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001521
1522 pte = (mm == &init_mm) ?
1523 pte_alloc_kernel(pmd, addr) :
1524 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1525 if (!pte)
1526 return -ENOMEM;
1527
1528 BUG_ON(pmd_huge(*pmd));
1529
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001530 token = pmd_pgtable(*pmd);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001531
1532 do {
Martin Schwidefsky2f569af2008-02-08 04:22:04 -08001533 err = fn(pte, token, addr, data);
Jeremy Fitzhardingeaee16b32007-05-06 14:48:54 -07001534 if (err)
1535 break;
1536 } while (pte++, addr += PAGE_SIZE, addr != end);
1537
1538 if (mm != &init_mm)
1539 pte_unmap_unlock(pte-1, ptl);
1540 return err;
1541}
1542
1543static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1544 unsigned long addr, unsigned long end,
1545 pte_fn_t fn, void *data)
1546{
1547 pmd_t *pmd;
1548 unsigned long next;
1549 int err;
1550
1551 pmd = pmd_alloc(mm, pud, addr);
1552 if (!pmd)
1553 return -ENOMEM;
1554 do {
1555 next = pmd_addr_end(addr, end);
1556 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1557 if (err)
1558 break;
1559 } while (pmd++, addr = next, addr != end);
1560 return err;
1561}
1562
1563static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1564 unsigned long addr, unsigned long end,
1565 pte_fn_t fn, void *data)
1566{
1567 pud_t *pud;
1568 unsigned long next;
1569 int err;
1570
1571 pud = pud_alloc(mm, pgd, addr);
1572 if (!pud)
1573 return -ENOMEM;
1574 do {
1575 next = pud_addr_end(addr, end);
1576 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1577 if (err)
1578 break;
1579 } while (pud++, addr = next, addr != end);
1580 return err;
1581}
1582
1583/*
1584 * Scan a region of virtual memory, filling in page tables as necessary
1585 * and calling a provided function on each leaf page table.
1586 */
1587int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1588 unsigned long size, pte_fn_t fn, void *data)
1589{
1590 pgd_t *pgd;
1591 unsigned long next;
1592 unsigned long end = addr + size;
1593 int err;
1594
1595 BUG_ON(addr >= end);
1596 pgd = pgd_offset(mm, addr);
1597 do {
1598 next = pgd_addr_end(addr, end);
1599 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1600 if (err)
1601 break;
1602 } while (pgd++, addr = next, addr != end);
1603 return err;
1604}
1605EXPORT_SYMBOL_GPL(apply_to_page_range);
1606
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001608 * handle_pte_fault chooses page fault handler according to an entry
1609 * which was read non-atomically. Before making any commitment, on
1610 * those architectures or configurations (e.g. i386 with PAE) which
1611 * might give a mix of unmatched parts, do_swap_page and do_file_page
1612 * must check under lock before unmapping the pte and proceeding
1613 * (but do_wp_page is only called after already making such a check;
1614 * and do_anonymous_page and do_no_page can safely check later on).
1615 */
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001616static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001617 pte_t *page_table, pte_t orig_pte)
1618{
1619 int same = 1;
1620#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1621 if (sizeof(pte_t) > sizeof(unsigned long)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001622 spinlock_t *ptl = pte_lockptr(mm, pmd);
1623 spin_lock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001624 same = pte_same(*page_table, orig_pte);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001625 spin_unlock(ptl);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001626 }
1627#endif
1628 pte_unmap(page_table);
1629 return same;
1630}
1631
1632/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1634 * servicing faults for write access. In the normal case, do always want
1635 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1636 * that do not have writing enabled, when used by access_process_vm.
1637 */
1638static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1639{
1640 if (likely(vma->vm_flags & VM_WRITE))
1641 pte = pte_mkwrite(pte);
1642 return pte;
1643}
1644
Atsushi Nemoto9de455b2006-12-12 17:14:55 +00001645static 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 -08001646{
1647 /*
1648 * If the source page was a PFN mapping, we don't have
1649 * a "struct page" for it. We do a best-effort copy by
1650 * just copying from the original user address. If that
1651 * fails, we just zero-fill it. Live with it.
1652 */
1653 if (unlikely(!src)) {
1654 void *kaddr = kmap_atomic(dst, KM_USER0);
Linus Torvalds5d2a2db2005-11-29 14:07:55 -08001655 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1656
1657 /*
1658 * This really shouldn't fail, because the page is there
1659 * in the page tables. But it might just be unreadable,
1660 * in which case we just give up and fill the result with
1661 * zeroes.
1662 */
1663 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
Linus Torvalds6aab3412005-11-28 14:34:23 -08001664 memset(kaddr, 0, PAGE_SIZE);
1665 kunmap_atomic(kaddr, KM_USER0);
Dmitriy Monakhovc4ec7b02006-10-19 23:29:08 -07001666 flush_dcache_page(dst);
Nick Piggin0ed361d2008-02-04 22:29:34 -08001667 } else
1668 copy_user_highpage(dst, src, va, vma);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001669}
1670
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 * This routine handles present pages, when users try to write
1673 * to a shared page. It is done by copying the page to a new address
1674 * and decrementing the shared-page counter for the old page.
1675 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 * Note that this routine assumes that the protection checks have been
1677 * done by the caller (the low-level page fault routine in most cases).
1678 * Thus we can safely just mark it writable once we've done any necessary
1679 * COW.
1680 *
1681 * We also mark the page dirty at this point even though the page will
1682 * change only once the write actually happens. This avoids a few races,
1683 * and potentially makes it more efficient.
1684 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001685 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1686 * but allow concurrent faults), with pte both mapped and locked.
1687 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 */
Hugh Dickins65500d22005-10-29 18:15:59 -07001689static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1690 unsigned long address, pte_t *page_table, pmd_t *pmd,
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001691 spinlock_t *ptl, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692{
Hugh Dickinse5bbe4d2005-11-29 16:54:51 +00001693 struct page *old_page, *new_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 pte_t entry;
Nick Piggin83c54072007-07-19 01:47:05 -07001695 int reuse = 0, ret = 0;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001696 int page_mkwrite = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001697 struct page *dirty_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698
Linus Torvalds6aab3412005-11-28 14:34:23 -08001699 old_page = vm_normal_page(vma, address, orig_pte);
Linus Torvalds6aab3412005-11-28 14:34:23 -08001700 if (!old_page)
1701 goto gotten;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001703 /*
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001704 * Take out anonymous pages first, anonymous shared vmas are
1705 * not dirty accountable.
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001706 */
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001707 if (PageAnon(old_page)) {
1708 if (!TestSetPageLocked(old_page)) {
1709 reuse = can_share_swap_page(old_page);
1710 unlock_page(old_page);
1711 }
1712 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001713 (VM_WRITE|VM_SHARED))) {
Peter Zijlstraee6a6452006-09-25 23:31:00 -07001714 /*
1715 * Only catch write-faults on shared writable pages,
1716 * read-only shared pages can get COWed by
1717 * get_user_pages(.write=1, .force=1).
1718 */
David Howells9637a5e2006-06-23 02:03:43 -07001719 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
1720 /*
1721 * Notify the address space that the page is about to
1722 * become writable so that it can prohibit this or wait
1723 * for the page to get into an appropriate state.
1724 *
1725 * We do this without the lock held, so that it can
1726 * sleep if it needs to.
1727 */
1728 page_cache_get(old_page);
1729 pte_unmap_unlock(page_table, ptl);
1730
1731 if (vma->vm_ops->page_mkwrite(vma, old_page) < 0)
1732 goto unwritable_page;
1733
David Howells9637a5e2006-06-23 02:03:43 -07001734 /*
1735 * Since we dropped the lock we need to revalidate
1736 * the PTE as someone else may have changed it. If
1737 * they did, we just return, as we can count on the
1738 * MMU to tell us if they didn't also make it writable.
1739 */
1740 page_table = pte_offset_map_lock(mm, pmd, address,
1741 &ptl);
Hugh Dickinsc3704ce2007-02-10 01:43:00 -08001742 page_cache_release(old_page);
David Howells9637a5e2006-06-23 02:03:43 -07001743 if (!pte_same(*page_table, orig_pte))
1744 goto unlock;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001745
1746 page_mkwrite = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001748 dirty_page = old_page;
1749 get_page(dirty_page);
David Howells9637a5e2006-06-23 02:03:43 -07001750 reuse = 1;
David Howells9637a5e2006-06-23 02:03:43 -07001751 }
1752
1753 if (reuse) {
1754 flush_cache_page(vma, address, pte_pfn(orig_pte));
1755 entry = pte_mkyoung(orig_pte);
1756 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
KAMEZAWA Hiroyuki954ffcb2007-10-16 01:25:44 -07001757 if (ptep_set_access_flags(vma, address, page_table, entry,1))
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07001758 update_mmu_cache(vma, address, entry);
David Howells9637a5e2006-06-23 02:03:43 -07001759 ret |= VM_FAULT_WRITE;
1760 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762
1763 /*
1764 * Ok, we need to copy. Oh, well..
1765 */
Nick Pigginb5810032005-10-29 18:16:12 -07001766 page_cache_get(old_page);
Hugh Dickins920fc352005-11-21 21:32:17 -08001767gotten:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001768 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
1770 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickins65500d22005-10-29 18:15:59 -07001771 goto oom;
Nick Piggin557ed1f2007-10-16 01:24:40 -07001772 VM_BUG_ON(old_page == ZERO_PAGE(0));
1773 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
1774 if (!new_page)
1775 goto oom;
1776 cow_user_page(new_page, old_page, address, vma);
Nick Piggin0ed361d2008-02-04 22:29:34 -08001777 __SetPageUptodate(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001778
Balbir Singhe1a1cd52008-02-07 00:14:02 -08001779 if (mem_cgroup_charge(new_page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001780 goto oom_free_new;
1781
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 /*
1783 * Re-check the pte - we dropped the lock
1784 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001785 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins65500d22005-10-29 18:15:59 -07001786 if (likely(pte_same(*page_table, orig_pte))) {
Hugh Dickins920fc352005-11-21 21:32:17 -08001787 if (old_page) {
Hugh Dickins920fc352005-11-21 21:32:17 -08001788 if (!PageAnon(old_page)) {
1789 dec_mm_counter(mm, file_rss);
1790 inc_mm_counter(mm, anon_rss);
1791 }
1792 } else
Hugh Dickins42946212005-10-29 18:16:05 -07001793 inc_mm_counter(mm, anon_rss);
Ben Collinseca35132005-11-29 11:45:26 -08001794 flush_cache_page(vma, address, pte_pfn(orig_pte));
Hugh Dickins65500d22005-10-29 18:15:59 -07001795 entry = mk_pte(new_page, vma->vm_page_prot);
1796 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Siddha, Suresh B4ce072f2006-09-29 01:58:42 -07001797 /*
1798 * Clear the pte entry and flush it first, before updating the
1799 * pte with the new entry. This will avoid a race condition
1800 * seen in the presence of one thread doing SMC and another
1801 * thread doing COW.
1802 */
1803 ptep_clear_flush(vma, address, page_table);
1804 set_pte_at(mm, address, page_table, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07001805 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 lru_cache_add_active(new_page);
Nick Piggin9617d952006-01-06 00:11:12 -08001807 page_add_new_anon_rmap(new_page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808
Nick Piggin945754a2008-06-23 14:30:30 +02001809 if (old_page) {
1810 /*
1811 * Only after switching the pte to the new page may
1812 * we remove the mapcount here. Otherwise another
1813 * process may come and find the rmap count decremented
1814 * before the pte is switched to the new page, and
1815 * "reuse" the old page writing into it while our pte
1816 * here still points into it and can be read by other
1817 * threads.
1818 *
1819 * The critical issue is to order this
1820 * page_remove_rmap with the ptp_clear_flush above.
1821 * Those stores are ordered by (if nothing else,)
1822 * the barrier present in the atomic_add_negative
1823 * in page_remove_rmap.
1824 *
1825 * Then the TLB flush in ptep_clear_flush ensures that
1826 * no process can access the old page before the
1827 * decremented mapcount is visible. And the old page
1828 * cannot be reused until after the decremented
1829 * mapcount is visible. So transitively, TLBs to
1830 * old page will be flushed before it can be reused.
1831 */
1832 page_remove_rmap(old_page, vma);
1833 }
1834
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 /* Free the old page.. */
1836 new_page = old_page;
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001837 ret |= VM_FAULT_WRITE;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001838 } else
1839 mem_cgroup_uncharge_page(new_page);
1840
Hugh Dickins920fc352005-11-21 21:32:17 -08001841 if (new_page)
1842 page_cache_release(new_page);
1843 if (old_page)
1844 page_cache_release(old_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001845unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07001846 pte_unmap_unlock(page_table, ptl);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001847 if (dirty_page) {
Anton Salikhmetov8f7b3d12008-01-23 02:21:18 +03001848 if (vma->vm_file)
1849 file_update_time(vma->vm_file);
1850
Nick Piggin79352892007-07-19 01:47:22 -07001851 /*
1852 * Yes, Virginia, this is actually required to prevent a race
1853 * with clear_page_dirty_for_io() from clearing the page dirty
1854 * bit after it clear all dirty ptes, but before a racing
1855 * do_wp_page installs a dirty pte.
1856 *
1857 * do_no_page is protected similarly.
1858 */
1859 wait_on_page_locked(dirty_page);
Peter Zijlstraa200ee12007-10-08 18:54:37 +02001860 set_page_dirty_balance(dirty_page, page_mkwrite);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07001861 put_page(dirty_page);
1862 }
Nick Pigginf33ea7f2005-08-03 20:24:01 +10001863 return ret;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001864oom_free_new:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08001865 page_cache_release(new_page);
Hugh Dickins65500d22005-10-29 18:15:59 -07001866oom:
Hugh Dickins920fc352005-11-21 21:32:17 -08001867 if (old_page)
1868 page_cache_release(old_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 return VM_FAULT_OOM;
David Howells9637a5e2006-06-23 02:03:43 -07001870
1871unwritable_page:
1872 page_cache_release(old_page);
1873 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874}
1875
1876/*
1877 * Helper functions for unmap_mapping_range().
1878 *
1879 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1880 *
1881 * We have to restart searching the prio_tree whenever we drop the lock,
1882 * since the iterator is only valid while the lock is held, and anyway
1883 * a later vma might be split and reinserted earlier while lock dropped.
1884 *
1885 * The list of nonlinear vmas could be handled more efficiently, using
1886 * a placeholder, but handle it in the same way until a need is shown.
1887 * It is important to search the prio_tree before nonlinear list: a vma
1888 * may become nonlinear and be shifted from prio_tree to nonlinear list
1889 * while the lock is dropped; but never shifted from list to prio_tree.
1890 *
1891 * In order to make forward progress despite restarting the search,
1892 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1893 * quickly skip it next time around. Since the prio_tree search only
1894 * shows us those vmas affected by unmapping the range in question, we
1895 * can't efficiently keep all vmas in step with mapping->truncate_count:
1896 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1897 * mapping->truncate_count and vma->vm_truncate_count are protected by
1898 * i_mmap_lock.
1899 *
1900 * In order to make forward progress despite repeatedly restarting some
Hugh Dickinsee39b372005-04-19 13:29:15 -07001901 * large vma, note the restart_addr from unmap_vmas when it breaks out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 * and restart from that address when we reach that vma again. It might
1903 * have been split or merged, shrunk or extended, but never shifted: so
1904 * restart_addr remains valid so long as it remains in the vma's range.
1905 * unmap_mapping_range forces truncate_count to leap over page-aligned
1906 * values so we can save vma's restart_addr in its truncate_count field.
1907 */
1908#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1909
1910static void reset_vma_truncate_counts(struct address_space *mapping)
1911{
1912 struct vm_area_struct *vma;
1913 struct prio_tree_iter iter;
1914
1915 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1916 vma->vm_truncate_count = 0;
1917 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1918 vma->vm_truncate_count = 0;
1919}
1920
1921static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1922 unsigned long start_addr, unsigned long end_addr,
1923 struct zap_details *details)
1924{
1925 unsigned long restart_addr;
1926 int need_break;
1927
Nick Piggind00806b2007-07-19 01:46:57 -07001928 /*
1929 * files that support invalidating or truncating portions of the
Nick Piggind0217ac2007-07-19 01:47:03 -07001930 * file from under mmaped areas must have their ->fault function
Nick Piggin83c54072007-07-19 01:47:05 -07001931 * return a locked page (and set VM_FAULT_LOCKED in the return).
1932 * This provides synchronisation against concurrent unmapping here.
Nick Piggind00806b2007-07-19 01:46:57 -07001933 */
Nick Piggind00806b2007-07-19 01:46:57 -07001934
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935again:
1936 restart_addr = vma->vm_truncate_count;
1937 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1938 start_addr = restart_addr;
1939 if (start_addr >= end_addr) {
1940 /* Top of vma has been split off since last time */
1941 vma->vm_truncate_count = details->truncate_count;
1942 return 0;
1943 }
1944 }
1945
Hugh Dickinsee39b372005-04-19 13:29:15 -07001946 restart_addr = zap_page_range(vma, start_addr,
1947 end_addr - start_addr, details);
Nick Piggin95c354f2008-01-30 13:31:20 +01001948 need_break = need_resched() || spin_needbreak(details->i_mmap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949
Hugh Dickinsee39b372005-04-19 13:29:15 -07001950 if (restart_addr >= end_addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 /* We have now completed this vma: mark it so */
1952 vma->vm_truncate_count = details->truncate_count;
1953 if (!need_break)
1954 return 0;
1955 } else {
1956 /* Note restart_addr in vma's truncate_count field */
Hugh Dickinsee39b372005-04-19 13:29:15 -07001957 vma->vm_truncate_count = restart_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 if (!need_break)
1959 goto again;
1960 }
1961
1962 spin_unlock(details->i_mmap_lock);
1963 cond_resched();
1964 spin_lock(details->i_mmap_lock);
1965 return -EINTR;
1966}
1967
1968static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1969 struct zap_details *details)
1970{
1971 struct vm_area_struct *vma;
1972 struct prio_tree_iter iter;
1973 pgoff_t vba, vea, zba, zea;
1974
1975restart:
1976 vma_prio_tree_foreach(vma, &iter, root,
1977 details->first_index, details->last_index) {
1978 /* Skip quickly over those we have already dealt with */
1979 if (vma->vm_truncate_count == details->truncate_count)
1980 continue;
1981
1982 vba = vma->vm_pgoff;
1983 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1984 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1985 zba = details->first_index;
1986 if (zba < vba)
1987 zba = vba;
1988 zea = details->last_index;
1989 if (zea > vea)
1990 zea = vea;
1991
1992 if (unmap_mapping_range_vma(vma,
1993 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1994 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1995 details) < 0)
1996 goto restart;
1997 }
1998}
1999
2000static inline void unmap_mapping_range_list(struct list_head *head,
2001 struct zap_details *details)
2002{
2003 struct vm_area_struct *vma;
2004
2005 /*
2006 * In nonlinear VMAs there is no correspondence between virtual address
2007 * offset and file offset. So we must perform an exhaustive search
2008 * across *all* the pages in each nonlinear VMA, not just the pages
2009 * whose virtual address lies outside the file truncation point.
2010 */
2011restart:
2012 list_for_each_entry(vma, head, shared.vm_set.list) {
2013 /* Skip quickly over those we have already dealt with */
2014 if (vma->vm_truncate_count == details->truncate_count)
2015 continue;
2016 details->nonlinear_vma = vma;
2017 if (unmap_mapping_range_vma(vma, vma->vm_start,
2018 vma->vm_end, details) < 0)
2019 goto restart;
2020 }
2021}
2022
2023/**
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08002024 * 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 -07002025 * @mapping: the address space containing mmaps to be unmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 * @holebegin: byte in first page to unmap, relative to the start of
2027 * the underlying file. This will be rounded down to a PAGE_SIZE
2028 * boundary. Note that this is different from vmtruncate(), which
2029 * must keep the partial page. In contrast, we must get rid of
2030 * partial pages.
2031 * @holelen: size of prospective hole in bytes. This will be rounded
2032 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2033 * end of the file.
2034 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2035 * but 0 when invalidating pagecache, don't throw away private data.
2036 */
2037void unmap_mapping_range(struct address_space *mapping,
2038 loff_t const holebegin, loff_t const holelen, int even_cows)
2039{
2040 struct zap_details details;
2041 pgoff_t hba = holebegin >> PAGE_SHIFT;
2042 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2043
2044 /* Check for overflow. */
2045 if (sizeof(holelen) > sizeof(hlen)) {
2046 long long holeend =
2047 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2048 if (holeend & ~(long long)ULONG_MAX)
2049 hlen = ULONG_MAX - hba + 1;
2050 }
2051
2052 details.check_mapping = even_cows? NULL: mapping;
2053 details.nonlinear_vma = NULL;
2054 details.first_index = hba;
2055 details.last_index = hba + hlen - 1;
2056 if (details.last_index < details.first_index)
2057 details.last_index = ULONG_MAX;
2058 details.i_mmap_lock = &mapping->i_mmap_lock;
2059
2060 spin_lock(&mapping->i_mmap_lock);
2061
Nick Piggind00806b2007-07-19 01:46:57 -07002062 /* Protect against endless unmapping loops */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 mapping->truncate_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 if (unlikely(is_restart_addr(mapping->truncate_count))) {
2065 if (mapping->truncate_count == 0)
2066 reset_vma_truncate_counts(mapping);
2067 mapping->truncate_count++;
2068 }
2069 details.truncate_count = mapping->truncate_count;
2070
2071 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
2072 unmap_mapping_range_tree(&mapping->i_mmap, &details);
2073 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
2074 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
2075 spin_unlock(&mapping->i_mmap_lock);
2076}
2077EXPORT_SYMBOL(unmap_mapping_range);
2078
Rolf Eike Beerbfa5bf62006-09-25 23:31:22 -07002079/**
2080 * vmtruncate - unmap mappings "freed" by truncate() syscall
2081 * @inode: inode of the file used
2082 * @offset: file offset to start truncating
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 *
2084 * NOTE! We have to be ready to update the memory sharing
2085 * between the file and the memory map for a potential last
2086 * incomplete page. Ugly, but necessary.
2087 */
2088int vmtruncate(struct inode * inode, loff_t offset)
2089{
Christoph Hellwig61d50482008-02-04 22:28:56 -08002090 if (inode->i_size < offset) {
2091 unsigned long limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092
Christoph Hellwig61d50482008-02-04 22:28:56 -08002093 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
2094 if (limit != RLIM_INFINITY && offset > limit)
2095 goto out_sig;
2096 if (offset > inode->i_sb->s_maxbytes)
2097 goto out_big;
2098 i_size_write(inode, offset);
2099 } else {
2100 struct address_space *mapping = inode->i_mapping;
Nick Piggind00806b2007-07-19 01:46:57 -07002101
Christoph Hellwig61d50482008-02-04 22:28:56 -08002102 /*
2103 * truncation of in-use swapfiles is disallowed - it would
2104 * cause subsequent swapout to scribble on the now-freed
2105 * blocks.
2106 */
2107 if (IS_SWAPFILE(inode))
2108 return -ETXTBSY;
2109 i_size_write(inode, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110
Christoph Hellwig61d50482008-02-04 22:28:56 -08002111 /*
2112 * unmap_mapping_range is called twice, first simply for
2113 * efficiency so that truncate_inode_pages does fewer
2114 * single-page unmaps. However after this first call, and
2115 * before truncate_inode_pages finishes, it is possible for
2116 * private pages to be COWed, which remain after
2117 * truncate_inode_pages finishes, hence the second
2118 * unmap_mapping_range call must be made for correctness.
2119 */
2120 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2121 truncate_inode_pages(mapping, offset);
2122 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2123 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 if (inode->i_op && inode->i_op->truncate)
2126 inode->i_op->truncate(inode);
2127 return 0;
Christoph Hellwig61d50482008-02-04 22:28:56 -08002128
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129out_sig:
2130 send_sig(SIGXFSZ, current, 0);
2131out_big:
2132 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134EXPORT_SYMBOL(vmtruncate);
2135
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002136int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
2137{
2138 struct address_space *mapping = inode->i_mapping;
2139
2140 /*
2141 * If the underlying filesystem is not going to provide
2142 * a way to truncate a range of blocks (punch a hole) -
2143 * we should return failure right now.
2144 */
2145 if (!inode->i_op || !inode->i_op->truncate_range)
2146 return -ENOSYS;
2147
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002148 mutex_lock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002149 down_write(&inode->i_alloc_sem);
2150 unmap_mapping_range(mapping, offset, (end - offset), 1);
2151 truncate_inode_pages_range(mapping, offset, end);
Nick Piggind00806b2007-07-19 01:46:57 -07002152 unmap_mapping_range(mapping, offset, (end - offset), 1);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002153 inode->i_op->truncate_range(inode, offset, end);
2154 up_write(&inode->i_alloc_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002155 mutex_unlock(&inode->i_mutex);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002156
2157 return 0;
2158}
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002159
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002161 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2162 * but allow concurrent faults), and pte mapped but not yet locked.
2163 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002165static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2166 unsigned long address, pte_t *page_table, pmd_t *pmd,
2167 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002169 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 struct page *page;
Hugh Dickins65500d22005-10-29 18:15:59 -07002171 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 pte_t pte;
Nick Piggin83c54072007-07-19 01:47:05 -07002173 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002175 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002176 goto out;
Hugh Dickins65500d22005-10-29 18:15:59 -07002177
2178 entry = pte_to_swp_entry(orig_pte);
Christoph Lameter06972122006-06-23 02:03:35 -07002179 if (is_migration_entry(entry)) {
2180 migration_entry_wait(mm, pmd, address);
2181 goto out;
2182 }
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002183 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184 page = lookup_swap_cache(entry);
2185 if (!page) {
Ashwin Chaugule098fe652006-12-06 20:31:54 -08002186 grab_swap_token(); /* Contend for token _before_ read-in */
Hugh Dickins02098fe2008-02-04 22:28:42 -08002187 page = swapin_readahead(entry,
2188 GFP_HIGHUSER_MOVABLE, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 if (!page) {
2190 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002191 * Back out if somebody else faulted in this pte
2192 * while we released the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002194 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 if (likely(pte_same(*page_table, orig_pte)))
2196 ret = VM_FAULT_OOM;
Shailabh Nagar0ff92242006-07-14 00:24:37 -07002197 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Hugh Dickins65500d22005-10-29 18:15:59 -07002198 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 }
2200
2201 /* Had to read the page from swap area: Major fault */
2202 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07002203 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 }
2205
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002206 if (mem_cgroup_charge(page, mm, GFP_KERNEL)) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002207 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2208 ret = VM_FAULT_OOM;
2209 goto out;
2210 }
2211
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 mark_page_accessed(page);
2213 lock_page(page);
Balbir Singh20a10222007-11-14 17:00:33 -08002214 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215
2216 /*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002217 * Back out if somebody else already faulted in this pte.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 */
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002219 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Hugh Dickins9e9bef02005-10-29 18:16:15 -07002220 if (unlikely(!pte_same(*page_table, orig_pte)))
Kirill Korotaevb8107482005-05-16 21:53:50 -07002221 goto out_nomap;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002222
2223 if (unlikely(!PageUptodate(page))) {
2224 ret = VM_FAULT_SIGBUS;
2225 goto out_nomap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 }
2227
2228 /* The page isn't present yet, go ahead with the fault. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229
Hugh Dickins42946212005-10-29 18:16:05 -07002230 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 pte = mk_pte(page, vma->vm_page_prot);
2232 if (write_access && can_share_swap_page(page)) {
2233 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2234 write_access = 0;
2235 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236
2237 flush_icache_page(vma, page);
2238 set_pte_at(mm, address, page_table, pte);
2239 page_add_anon_rmap(page, vma, address);
2240
Hugh Dickinsc475a8a2005-06-21 17:15:12 -07002241 swap_free(entry);
2242 if (vm_swap_full())
2243 remove_exclusive_swap_page(page);
2244 unlock_page(page);
2245
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 if (write_access) {
Hugh Dickins61469f12008-03-04 14:29:04 -08002247 ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
2248 if (ret & VM_FAULT_ERROR)
2249 ret &= VM_FAULT_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 goto out;
2251 }
2252
2253 /* No need to invalidate - it was non-present before */
2254 update_mmu_cache(vma, address, pte);
Hugh Dickins65500d22005-10-29 18:15:59 -07002255unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002256 pte_unmap_unlock(page_table, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257out:
2258 return ret;
Kirill Korotaevb8107482005-05-16 21:53:50 -07002259out_nomap:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002260 mem_cgroup_uncharge_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002261 pte_unmap_unlock(page_table, ptl);
Kirill Korotaevb8107482005-05-16 21:53:50 -07002262 unlock_page(page);
2263 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002264 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265}
2266
2267/*
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002268 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2269 * but allow concurrent faults), and pte mapped but not yet locked.
2270 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271 */
Hugh Dickins65500d22005-10-29 18:15:59 -07002272static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2273 unsigned long address, pte_t *page_table, pmd_t *pmd,
2274 int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275{
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002276 struct page *page;
2277 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279
Nick Piggin557ed1f2007-10-16 01:24:40 -07002280 /* Allocate our own private page. */
2281 pte_unmap(page_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282
Nick Piggin557ed1f2007-10-16 01:24:40 -07002283 if (unlikely(anon_vma_prepare(vma)))
2284 goto oom;
2285 page = alloc_zeroed_user_highpage_movable(vma, address);
2286 if (!page)
2287 goto oom;
Nick Piggin0ed361d2008-02-04 22:29:34 -08002288 __SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002290 if (mem_cgroup_charge(page, mm, GFP_KERNEL))
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002291 goto oom_free_page;
2292
Nick Piggin557ed1f2007-10-16 01:24:40 -07002293 entry = mk_pte(page, vma->vm_page_prot);
2294 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002295
Nick Piggin557ed1f2007-10-16 01:24:40 -07002296 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2297 if (!pte_none(*page_table))
2298 goto release;
2299 inc_mm_counter(mm, anon_rss);
2300 lru_cache_add_active(page);
2301 page_add_new_anon_rmap(page, vma, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002302 set_pte_at(mm, address, page_table, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303
2304 /* No need to invalidate - it was non-present before */
Hugh Dickins65500d22005-10-29 18:15:59 -07002305 update_mmu_cache(vma, address, entry);
Hugh Dickins65500d22005-10-29 18:15:59 -07002306unlock:
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002307 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002308 return 0;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002309release:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002310 mem_cgroup_uncharge_page(page);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002311 page_cache_release(page);
2312 goto unlock;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002313oom_free_page:
Hugh Dickins6dbf6d32008-03-04 14:29:04 -08002314 page_cache_release(page);
Hugh Dickins65500d22005-10-29 18:15:59 -07002315oom:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 return VM_FAULT_OOM;
2317}
2318
2319/*
Nick Piggin54cb8822007-07-19 01:46:59 -07002320 * __do_fault() tries to create a new page mapping. It aggressively
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 * tries to share with existing pages, but makes a separate copy if
Nick Piggin54cb8822007-07-19 01:46:59 -07002322 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2323 * the next page fault.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 *
2325 * As this is called only for pages that do not currently exist, we
2326 * do not need to flush old virtual caches or the TLB.
2327 *
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002328 * We enter with non-exclusive mmap_sem (to exclude vma changes,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002329 * but allow concurrent faults), and pte neither mapped nor locked.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002330 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 */
Nick Piggin54cb8822007-07-19 01:46:59 -07002332static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins16abfa02007-10-04 16:56:06 +01002333 unsigned long address, pmd_t *pmd,
Nick Piggin54cb8822007-07-19 01:46:59 -07002334 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335{
Hugh Dickins16abfa02007-10-04 16:56:06 +01002336 pte_t *page_table;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002337 spinlock_t *ptl;
Nick Piggind0217ac2007-07-19 01:47:03 -07002338 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 pte_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340 int anon = 0;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002341 struct page *dirty_page = NULL;
Nick Piggind0217ac2007-07-19 01:47:03 -07002342 struct vm_fault vmf;
2343 int ret;
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002344 int page_mkwrite = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07002345
Nick Piggind0217ac2007-07-19 01:47:03 -07002346 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2347 vmf.pgoff = pgoff;
2348 vmf.flags = flags;
2349 vmf.page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350
Nick Piggin3c18ddd2008-04-28 02:12:10 -07002351 ret = vma->vm_ops->fault(vma, &vmf);
2352 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2353 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354
Nick Piggind00806b2007-07-19 01:46:57 -07002355 /*
Nick Piggind0217ac2007-07-19 01:47:03 -07002356 * For consistency in subsequent calls, make the faulted page always
Nick Piggind00806b2007-07-19 01:46:57 -07002357 * locked.
2358 */
Nick Piggin83c54072007-07-19 01:47:05 -07002359 if (unlikely(!(ret & VM_FAULT_LOCKED)))
Nick Piggind0217ac2007-07-19 01:47:03 -07002360 lock_page(vmf.page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002361 else
Nick Piggind0217ac2007-07-19 01:47:03 -07002362 VM_BUG_ON(!PageLocked(vmf.page));
Nick Piggind00806b2007-07-19 01:46:57 -07002363
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 /*
2365 * Should we do an early C-O-W break?
2366 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002367 page = vmf.page;
Nick Piggin54cb8822007-07-19 01:46:59 -07002368 if (flags & FAULT_FLAG_WRITE) {
David Howells9637a5e2006-06-23 02:03:43 -07002369 if (!(vma->vm_flags & VM_SHARED)) {
Nick Piggin54cb8822007-07-19 01:46:59 -07002370 anon = 1;
Nick Piggind00806b2007-07-19 01:46:57 -07002371 if (unlikely(anon_vma_prepare(vma))) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002372 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002373 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002374 }
Nick Piggin83c54072007-07-19 01:47:05 -07002375 page = alloc_page_vma(GFP_HIGHUSER_MOVABLE,
2376 vma, address);
Nick Piggind00806b2007-07-19 01:46:57 -07002377 if (!page) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002378 ret = VM_FAULT_OOM;
Nick Piggin54cb8822007-07-19 01:46:59 -07002379 goto out;
Nick Piggind00806b2007-07-19 01:46:57 -07002380 }
Nick Piggind0217ac2007-07-19 01:47:03 -07002381 copy_user_highpage(page, vmf.page, address, vma);
Nick Piggin0ed361d2008-02-04 22:29:34 -08002382 __SetPageUptodate(page);
David Howells9637a5e2006-06-23 02:03:43 -07002383 } else {
Nick Piggin54cb8822007-07-19 01:46:59 -07002384 /*
2385 * If the page will be shareable, see if the backing
David Howells9637a5e2006-06-23 02:03:43 -07002386 * address space wants to know that the page is about
Nick Piggin54cb8822007-07-19 01:46:59 -07002387 * to become writable
2388 */
Mark Fasheh69676142007-07-19 01:47:00 -07002389 if (vma->vm_ops->page_mkwrite) {
2390 unlock_page(page);
2391 if (vma->vm_ops->page_mkwrite(vma, page) < 0) {
Nick Piggind0217ac2007-07-19 01:47:03 -07002392 ret = VM_FAULT_SIGBUS;
2393 anon = 1; /* no anon but release vmf.page */
Mark Fasheh69676142007-07-19 01:47:00 -07002394 goto out_unlocked;
2395 }
2396 lock_page(page);
Nick Piggind0217ac2007-07-19 01:47:03 -07002397 /*
2398 * XXX: this is not quite right (racy vs
2399 * invalidate) to unlock and relock the page
2400 * like this, however a better fix requires
2401 * reworking page_mkwrite locking API, which
2402 * is better done later.
2403 */
2404 if (!page->mapping) {
Nick Piggin83c54072007-07-19 01:47:05 -07002405 ret = 0;
Nick Piggind0217ac2007-07-19 01:47:03 -07002406 anon = 1; /* no anon but release vmf.page */
2407 goto out;
2408 }
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002409 page_mkwrite = 1;
David Howells9637a5e2006-06-23 02:03:43 -07002410 }
2411 }
Nick Piggin54cb8822007-07-19 01:46:59 -07002412
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 }
2414
Balbir Singhe1a1cd52008-02-07 00:14:02 -08002415 if (mem_cgroup_charge(page, mm, GFP_KERNEL)) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002416 ret = VM_FAULT_OOM;
2417 goto out;
2418 }
2419
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002420 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421
2422 /*
2423 * This silly early PAGE_DIRTY setting removes a race
2424 * due to the bad i386 page protection. But it's valid
2425 * for other architectures too.
2426 *
2427 * Note that if write_access is true, we either now have
2428 * an exclusive copy of the page, or this is a shared mapping,
2429 * so we can make it writable and dirty to avoid having to
2430 * handle that later.
2431 */
2432 /* Only go through if we didn't race with anybody else... */
Nick Piggin54cb8822007-07-19 01:46:59 -07002433 if (likely(pte_same(*page_table, orig_pte))) {
Nick Piggind00806b2007-07-19 01:46:57 -07002434 flush_icache_page(vma, page);
2435 entry = mk_pte(page, vma->vm_page_prot);
Nick Piggin54cb8822007-07-19 01:46:59 -07002436 if (flags & FAULT_FLAG_WRITE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2438 set_pte_at(mm, address, page_table, entry);
2439 if (anon) {
Nick Piggind00806b2007-07-19 01:46:57 -07002440 inc_mm_counter(mm, anon_rss);
2441 lru_cache_add_active(page);
2442 page_add_new_anon_rmap(page, vma, address);
Hugh Dickinsf57e88a2005-11-21 21:32:19 -08002443 } else {
Hugh Dickins42946212005-10-29 18:16:05 -07002444 inc_mm_counter(mm, file_rss);
Nick Piggind00806b2007-07-19 01:46:57 -07002445 page_add_file_rmap(page);
Nick Piggin54cb8822007-07-19 01:46:59 -07002446 if (flags & FAULT_FLAG_WRITE) {
Nick Piggind00806b2007-07-19 01:46:57 -07002447 dirty_page = page;
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002448 get_page(dirty_page);
2449 }
Hugh Dickins42946212005-10-29 18:16:05 -07002450 }
Nick Piggind00806b2007-07-19 01:46:57 -07002451
2452 /* no need to invalidate: a not-present page won't be cached */
2453 update_mmu_cache(vma, address, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 } else {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002455 mem_cgroup_uncharge_page(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002456 if (anon)
2457 page_cache_release(page);
2458 else
Nick Piggin54cb8822007-07-19 01:46:59 -07002459 anon = 1; /* no anon but release faulted_page */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460 }
2461
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002462 pte_unmap_unlock(page_table, ptl);
Nick Piggind00806b2007-07-19 01:46:57 -07002463
2464out:
Nick Piggind0217ac2007-07-19 01:47:03 -07002465 unlock_page(vmf.page);
Mark Fasheh69676142007-07-19 01:47:00 -07002466out_unlocked:
Nick Piggind00806b2007-07-19 01:46:57 -07002467 if (anon)
Nick Piggind0217ac2007-07-19 01:47:03 -07002468 page_cache_release(vmf.page);
Nick Piggind00806b2007-07-19 01:46:57 -07002469 else if (dirty_page) {
Anton Salikhmetov8f7b3d12008-01-23 02:21:18 +03002470 if (vma->vm_file)
2471 file_update_time(vma->vm_file);
2472
Peter Zijlstraa200ee12007-10-08 18:54:37 +02002473 set_page_dirty_balance(dirty_page, page_mkwrite);
Peter Zijlstrad08b3852006-09-25 23:30:57 -07002474 put_page(dirty_page);
2475 }
Nick Piggind00806b2007-07-19 01:46:57 -07002476
Nick Piggin83c54072007-07-19 01:47:05 -07002477 return ret;
Nick Piggin54cb8822007-07-19 01:46:59 -07002478}
Nick Piggind00806b2007-07-19 01:46:57 -07002479
Nick Piggin54cb8822007-07-19 01:46:59 -07002480static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2481 unsigned long address, pte_t *page_table, pmd_t *pmd,
2482 int write_access, pte_t orig_pte)
2483{
2484 pgoff_t pgoff = (((address & PAGE_MASK)
Dean Nelson0da7e012007-10-16 01:24:45 -07002485 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
Nick Piggin54cb8822007-07-19 01:46:59 -07002486 unsigned int flags = (write_access ? FAULT_FLAG_WRITE : 0);
2487
Hugh Dickins16abfa02007-10-04 16:56:06 +01002488 pte_unmap(page_table);
2489 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Nick Piggin54cb8822007-07-19 01:46:59 -07002490}
2491
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492
2493/*
Jes Sorensenf4b81802006-09-27 01:50:10 -07002494 * do_no_pfn() tries to create a new page mapping for a page without
2495 * a struct_page backing it
2496 *
2497 * As this is called only for pages that do not currently exist, we
2498 * do not need to flush old virtual caches or the TLB.
2499 *
2500 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2501 * but allow concurrent faults), and pte mapped but not yet locked.
2502 * We return with mmap_sem still held, but pte unmapped and unlocked.
2503 *
2504 * It is expected that the ->nopfn handler always returns the same pfn
2505 * for a given virtual mapping.
2506 *
2507 * Mark this `noinline' to prevent it from bloating the main pagefault code.
2508 */
2509static noinline int do_no_pfn(struct mm_struct *mm, struct vm_area_struct *vma,
2510 unsigned long address, pte_t *page_table, pmd_t *pmd,
2511 int write_access)
2512{
2513 spinlock_t *ptl;
2514 pte_t entry;
2515 unsigned long pfn;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002516
2517 pte_unmap(page_table);
Jared Hulbertb379d792008-04-28 02:12:58 -07002518 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
2519 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
Jes Sorensenf4b81802006-09-27 01:50:10 -07002520
2521 pfn = vma->vm_ops->nopfn(vma, address & PAGE_MASK);
Jared Hulbertb379d792008-04-28 02:12:58 -07002522
2523 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
2524
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002525 if (unlikely(pfn == NOPFN_OOM))
Jes Sorensenf4b81802006-09-27 01:50:10 -07002526 return VM_FAULT_OOM;
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002527 else if (unlikely(pfn == NOPFN_SIGBUS))
Jes Sorensenf4b81802006-09-27 01:50:10 -07002528 return VM_FAULT_SIGBUS;
Benjamin Herrenschmidt22cd25e2007-02-12 00:51:38 -08002529 else if (unlikely(pfn == NOPFN_REFAULT))
Nick Piggin83c54072007-07-19 01:47:05 -07002530 return 0;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002531
2532 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2533
2534 /* Only go through if we didn't race with anybody else... */
2535 if (pte_none(*page_table)) {
2536 entry = pfn_pte(pfn, vma->vm_page_prot);
2537 if (write_access)
2538 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2539 set_pte_at(mm, address, page_table, entry);
2540 }
2541 pte_unmap_unlock(page_table, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002542 return 0;
Jes Sorensenf4b81802006-09-27 01:50:10 -07002543}
2544
2545/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 * Fault of a previously existing named mapping. Repopulate the pte
2547 * from the encoded file_pte if possible. This enables swappable
2548 * nonlinear vmas.
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002549 *
2550 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2551 * but allow concurrent faults), and pte mapped but not yet locked.
2552 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002554static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Hugh Dickins65500d22005-10-29 18:15:59 -07002555 unsigned long address, pte_t *page_table, pmd_t *pmd,
2556 int write_access, pte_t orig_pte)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557{
Nick Piggind0217ac2007-07-19 01:47:03 -07002558 unsigned int flags = FAULT_FLAG_NONLINEAR |
2559 (write_access ? FAULT_FLAG_WRITE : 0);
Hugh Dickins65500d22005-10-29 18:15:59 -07002560 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002562 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
Nick Piggin83c54072007-07-19 01:47:05 -07002563 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564
Nick Piggind0217ac2007-07-19 01:47:03 -07002565 if (unlikely(!(vma->vm_flags & VM_NONLINEAR) ||
2566 !(vma->vm_flags & VM_CAN_NONLINEAR))) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002567 /*
2568 * Page table corrupted: show pte and kill process.
2569 */
Nick Pigginb5810032005-10-29 18:16:12 -07002570 print_bad_pte(vma, orig_pte, address);
Hugh Dickins65500d22005-10-29 18:15:59 -07002571 return VM_FAULT_OOM;
2572 }
Hugh Dickins65500d22005-10-29 18:15:59 -07002573
2574 pgoff = pte_to_pgoff(orig_pte);
Hugh Dickins16abfa02007-10-04 16:56:06 +01002575 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576}
2577
2578/*
2579 * These routines also need to handle stuff like marking pages dirty
2580 * and/or accessed for architectures that don't do it in hardware (most
2581 * RISC architectures). The early dirtying is also good on the i386.
2582 *
2583 * There is also a hook called "update_mmu_cache()" that architectures
2584 * with external mmu caches can use to update those (ie the Sparc or
2585 * PowerPC hashed page tables that act as extended TLBs).
2586 *
Hugh Dickinsc74df322005-10-29 18:16:23 -07002587 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2588 * but allow concurrent faults), and pte mapped but not yet locked.
2589 * We return with mmap_sem still held, but pte unmapped and unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 */
2591static inline int handle_pte_fault(struct mm_struct *mm,
Hugh Dickins65500d22005-10-29 18:15:59 -07002592 struct vm_area_struct *vma, unsigned long address,
2593 pte_t *pte, pmd_t *pmd, int write_access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594{
2595 pte_t entry;
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002596 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002598 entry = *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599 if (!pte_present(entry)) {
Hugh Dickins65500d22005-10-29 18:15:59 -07002600 if (pte_none(entry)) {
Jes Sorensenf4b81802006-09-27 01:50:10 -07002601 if (vma->vm_ops) {
Nick Piggin3c18ddd2008-04-28 02:12:10 -07002602 if (likely(vma->vm_ops->fault))
Nick Piggin54cb8822007-07-19 01:46:59 -07002603 return do_linear_fault(mm, vma, address,
2604 pte, pmd, write_access, entry);
Jes Sorensenf4b81802006-09-27 01:50:10 -07002605 if (unlikely(vma->vm_ops->nopfn))
2606 return do_no_pfn(mm, vma, address, pte,
2607 pmd, write_access);
2608 }
2609 return do_anonymous_page(mm, vma, address,
2610 pte, pmd, write_access);
Hugh Dickins65500d22005-10-29 18:15:59 -07002611 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 if (pte_file(entry))
Nick Piggind0217ac2007-07-19 01:47:03 -07002613 return do_nonlinear_fault(mm, vma, address,
Hugh Dickins65500d22005-10-29 18:15:59 -07002614 pte, pmd, write_access, entry);
2615 return do_swap_page(mm, vma, address,
2616 pte, pmd, write_access, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617 }
2618
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07002619 ptl = pte_lockptr(mm, pmd);
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002620 spin_lock(ptl);
2621 if (unlikely(!pte_same(*pte, entry)))
2622 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 if (write_access) {
2624 if (!pte_write(entry))
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002625 return do_wp_page(mm, vma, address,
2626 pte, pmd, ptl, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 entry = pte_mkdirty(entry);
2628 }
2629 entry = pte_mkyoung(entry);
Benjamin Herrenschmidt8dab5242007-06-16 10:16:12 -07002630 if (ptep_set_access_flags(vma, address, pte, entry, write_access)) {
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002631 update_mmu_cache(vma, address, entry);
Andrea Arcangeli1a44e142005-10-29 18:16:48 -07002632 } else {
2633 /*
2634 * This is needed only for protection faults but the arch code
2635 * is not yet telling us if this is a protection fault or not.
2636 * This still avoids useless tlb flushes for .text page faults
2637 * with threads.
2638 */
2639 if (write_access)
2640 flush_tlb_page(vma, address);
2641 }
Hugh Dickins8f4e2102005-10-29 18:16:26 -07002642unlock:
2643 pte_unmap_unlock(pte, ptl);
Nick Piggin83c54072007-07-19 01:47:05 -07002644 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645}
2646
2647/*
2648 * By the time we get here, we already hold the mm semaphore
2649 */
Nick Piggin83c54072007-07-19 01:47:05 -07002650int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 unsigned long address, int write_access)
2652{
2653 pgd_t *pgd;
2654 pud_t *pud;
2655 pmd_t *pmd;
2656 pte_t *pte;
2657
2658 __set_current_state(TASK_RUNNING);
2659
Christoph Lameterf8891e52006-06-30 01:55:45 -07002660 count_vm_event(PGFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661
Hugh Dickinsac9b9c62005-10-20 16:24:28 +01002662 if (unlikely(is_vm_hugetlb_page(vma)))
2663 return hugetlb_fault(mm, vma, address, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 pud = pud_alloc(mm, pgd, address);
2667 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002668 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 pmd = pmd_alloc(mm, pud, address);
2670 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002671 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672 pte = pte_alloc_map(mm, pmd, address);
2673 if (!pte)
Hugh Dickinsc74df322005-10-29 18:16:23 -07002674 return VM_FAULT_OOM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675
Hugh Dickinsc74df322005-10-29 18:16:23 -07002676 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677}
2678
2679#ifndef __PAGETABLE_PUD_FOLDED
2680/*
2681 * Allocate page upper directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002682 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002684int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002686 pud_t *new = pud_alloc_one(mm, address);
2687 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002688 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689
Nick Piggin362a61a2008-05-14 06:37:36 +02002690 smp_wmb(); /* See comment in __pte_alloc */
2691
Hugh Dickins872fec12005-10-29 18:16:21 -07002692 spin_lock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002693 if (pgd_present(*pgd)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002694 pud_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002695 else
2696 pgd_populate(mm, pgd, new);
Hugh Dickinsc74df322005-10-29 18:16:23 -07002697 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002698 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699}
2700#endif /* __PAGETABLE_PUD_FOLDED */
2701
2702#ifndef __PAGETABLE_PMD_FOLDED
2703/*
2704 * Allocate page middle directory.
Hugh Dickins872fec12005-10-29 18:16:21 -07002705 * We've already handled the fast-path in-line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 */
Hugh Dickins1bb36302005-10-29 18:16:22 -07002707int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708{
Hugh Dickinsc74df322005-10-29 18:16:23 -07002709 pmd_t *new = pmd_alloc_one(mm, address);
2710 if (!new)
Hugh Dickins1bb36302005-10-29 18:16:22 -07002711 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712
Nick Piggin362a61a2008-05-14 06:37:36 +02002713 smp_wmb(); /* See comment in __pte_alloc */
2714
Hugh Dickins872fec12005-10-29 18:16:21 -07002715 spin_lock(&mm->page_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716#ifndef __ARCH_HAS_4LEVEL_HACK
Hugh Dickins1bb36302005-10-29 18:16:22 -07002717 if (pud_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002718 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002719 else
2720 pud_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721#else
Hugh Dickins1bb36302005-10-29 18:16:22 -07002722 if (pgd_present(*pud)) /* Another has populated it */
Benjamin Herrenschmidt5e541972008-02-04 22:29:14 -08002723 pmd_free(mm, new);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002724 else
2725 pgd_populate(mm, pud, new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726#endif /* __ARCH_HAS_4LEVEL_HACK */
Hugh Dickinsc74df322005-10-29 18:16:23 -07002727 spin_unlock(&mm->page_table_lock);
Hugh Dickins1bb36302005-10-29 18:16:22 -07002728 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729}
2730#endif /* __PAGETABLE_PMD_FOLDED */
2731
2732int make_pages_present(unsigned long addr, unsigned long end)
2733{
2734 int ret, len, write;
2735 struct vm_area_struct * vma;
2736
2737 vma = find_vma(current->mm, addr);
2738 if (!vma)
2739 return -1;
2740 write = (vma->vm_flags & VM_WRITE) != 0;
Eric Sesterhenn5bcb28b2006-03-26 18:30:52 +02002741 BUG_ON(addr >= end);
2742 BUG_ON(end > vma->vm_end);
Rolf Eike Beer68e116a2007-07-15 23:38:03 -07002743 len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 ret = get_user_pages(current, current->mm, addr,
2745 len, write, 0, NULL, NULL);
2746 if (ret < 0)
2747 return ret;
2748 return ret == len ? 0 : -1;
2749}
2750
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751#if !defined(__HAVE_ARCH_GATE_AREA)
2752
2753#if defined(AT_SYSINFO_EHDR)
Adrian Bunk5ce78522005-09-10 00:26:28 -07002754static struct vm_area_struct gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755
2756static int __init gate_vma_init(void)
2757{
2758 gate_vma.vm_mm = NULL;
2759 gate_vma.vm_start = FIXADDR_USER_START;
2760 gate_vma.vm_end = FIXADDR_USER_END;
Roland McGrathb6558c42007-01-26 00:56:47 -08002761 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
2762 gate_vma.vm_page_prot = __P101;
Roland McGrathf47aef52007-01-26 00:56:49 -08002763 /*
2764 * Make sure the vDSO gets into every core dump.
2765 * Dumping its contents makes post-mortem fully interpretable later
2766 * without matching up the same kernel and hardware config to see
2767 * what PC values meant.
2768 */
2769 gate_vma.vm_flags |= VM_ALWAYSDUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770 return 0;
2771}
2772__initcall(gate_vma_init);
2773#endif
2774
2775struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2776{
2777#ifdef AT_SYSINFO_EHDR
2778 return &gate_vma;
2779#else
2780 return NULL;
2781#endif
2782}
2783
2784int in_gate_area_no_task(unsigned long addr)
2785{
2786#ifdef AT_SYSINFO_EHDR
2787 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2788 return 1;
2789#endif
2790 return 0;
2791}
2792
2793#endif /* __HAVE_ARCH_GATE_AREA */
David Howells0ec76a12006-09-27 01:50:15 -07002794
2795/*
2796 * Access another process' address space.
2797 * Source/target buffer must be kernel space,
2798 * Do not walk the page table directly, use get_user_pages
2799 */
2800int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2801{
2802 struct mm_struct *mm;
2803 struct vm_area_struct *vma;
2804 struct page *page;
2805 void *old_buf = buf;
2806
2807 mm = get_task_mm(tsk);
2808 if (!mm)
2809 return 0;
2810
2811 down_read(&mm->mmap_sem);
Simon Arlott183ff222007-10-20 01:27:18 +02002812 /* ignore errors, just check how much was successfully transferred */
David Howells0ec76a12006-09-27 01:50:15 -07002813 while (len) {
2814 int bytes, ret, offset;
2815 void *maddr;
2816
2817 ret = get_user_pages(tsk, mm, addr, 1,
2818 write, 1, &page, &vma);
2819 if (ret <= 0)
2820 break;
2821
2822 bytes = len;
2823 offset = addr & (PAGE_SIZE-1);
2824 if (bytes > PAGE_SIZE-offset)
2825 bytes = PAGE_SIZE-offset;
2826
2827 maddr = kmap(page);
2828 if (write) {
2829 copy_to_user_page(vma, page, addr,
2830 maddr + offset, buf, bytes);
2831 set_page_dirty_lock(page);
2832 } else {
2833 copy_from_user_page(vma, page, addr,
2834 buf, maddr + offset, bytes);
2835 }
2836 kunmap(page);
2837 page_cache_release(page);
2838 len -= bytes;
2839 buf += bytes;
2840 addr += bytes;
2841 }
2842 up_read(&mm->mmap_sem);
2843 mmput(mm);
2844
2845 return buf - old_buf;
2846}
Andi Kleen03252912008-01-30 13:33:18 +01002847
2848/*
2849 * Print the name of a VMA.
2850 */
2851void print_vma_addr(char *prefix, unsigned long ip)
2852{
2853 struct mm_struct *mm = current->mm;
2854 struct vm_area_struct *vma;
2855
Ingo Molnare8bff742008-02-13 20:21:06 +01002856 /*
2857 * Do not print if we are in atomic
2858 * contexts (in exception stacks, etc.):
2859 */
2860 if (preempt_count())
2861 return;
2862
Andi Kleen03252912008-01-30 13:33:18 +01002863 down_read(&mm->mmap_sem);
2864 vma = find_vma(mm, ip);
2865 if (vma && vma->vm_file) {
2866 struct file *f = vma->vm_file;
2867 char *buf = (char *)__get_free_page(GFP_KERNEL);
2868 if (buf) {
2869 char *p, *s;
2870
Jan Blunckcf28b482008-02-14 19:38:44 -08002871 p = d_path(&f->f_path, buf, PAGE_SIZE);
Andi Kleen03252912008-01-30 13:33:18 +01002872 if (IS_ERR(p))
2873 p = "?";
2874 s = strrchr(p, '/');
2875 if (s)
2876 p = s+1;
2877 printk("%s%s[%lx+%lx]", prefix, p,
2878 vma->vm_start,
2879 vma->vm_end - vma->vm_start);
2880 free_page((unsigned long)buf);
2881 }
2882 }
2883 up_read(&current->mm->mmap_sem);
2884}