4 * Explicit pagetable population and nonlinear (random) mappings support.
6 * started by Ingo Molnar, Copyright (C) 2002, 2003
10 #include <linux/swap.h>
11 #include <linux/file.h>
12 #include <linux/mman.h>
13 #include <linux/pagemap.h>
14 #include <linux/swapops.h>
15 #include <linux/rmap.h>
16 #include <linux/module.h>
17 #include <linux/syscalls.h>
19 #include <asm/mmu_context.h>
20 #include <asm/cacheflush.h>
21 #include <asm/tlbflush.h>
23 static int zap_pte(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
24 unsigned long addr
, pte_t
*ptep
)
27 struct page
*page
= NULL
;
29 if (pte_present(pte
)) {
30 unsigned long pfn
= pte_pfn(pte
);
31 flush_cache_page(vma
, addr
, pfn
);
32 pte
= ptep_clear_flush(vma
, addr
, ptep
);
33 if (unlikely(!pfn_valid(pfn
))) {
34 print_bad_pte(vma
, pte
, addr
);
37 page
= pfn_to_page(pfn
);
40 page_remove_rmap(page
);
41 page_cache_release(page
);
44 free_swap_and_cache(pte_to_swp_entry(pte
));
45 pte_clear(mm
, addr
, ptep
);
52 * Install a file page to a given virtual memory address, release any
53 * previously existing mapping.
55 int install_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
56 unsigned long addr
, struct page
*page
, pgprot_t prot
)
67 BUG_ON(vma
->vm_flags
& VM_RESERVED
);
69 pgd
= pgd_offset(mm
, addr
);
70 spin_lock(&mm
->page_table_lock
);
72 pud
= pud_alloc(mm
, pgd
, addr
);
76 pmd
= pmd_alloc(mm
, pud
, addr
);
80 pte
= pte_alloc_map(mm
, pmd
, addr
);
85 * This page may have been truncated. Tell the
89 inode
= vma
->vm_file
->f_mapping
->host
;
90 size
= (i_size_read(inode
) + PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
91 if (!page
->mapping
|| page
->index
>= size
)
94 if (page_mapcount(page
) > INT_MAX
/2)
97 if (pte_none(*pte
) || !zap_pte(mm
, vma
, addr
, pte
))
98 inc_mm_counter(mm
, file_rss
);
100 flush_icache_page(vma
, page
);
101 set_pte_at(mm
, addr
, pte
, mk_pte(page
, prot
));
102 page_add_file_rmap(page
);
105 update_mmu_cache(vma
, addr
, pte_val
);
109 spin_unlock(&mm
->page_table_lock
);
112 EXPORT_SYMBOL(install_page
);
116 * Install a file pte to a given virtual memory address, release any
117 * previously existing mapping.
119 int install_file_pte(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
120 unsigned long addr
, unsigned long pgoff
, pgprot_t prot
)
129 BUG_ON(vma
->vm_flags
& VM_RESERVED
);
131 pgd
= pgd_offset(mm
, addr
);
132 spin_lock(&mm
->page_table_lock
);
134 pud
= pud_alloc(mm
, pgd
, addr
);
138 pmd
= pmd_alloc(mm
, pud
, addr
);
142 pte
= pte_alloc_map(mm
, pmd
, addr
);
146 if (!pte_none(*pte
) && zap_pte(mm
, vma
, addr
, pte
)) {
147 update_hiwater_rss(mm
);
148 dec_mm_counter(mm
, file_rss
);
151 set_pte_at(mm
, addr
, pte
, pgoff_to_pte(pgoff
));
154 update_mmu_cache(vma
, addr
, pte_val
);
155 spin_unlock(&mm
->page_table_lock
);
159 spin_unlock(&mm
->page_table_lock
);
165 * sys_remap_file_pages - remap arbitrary pages of a shared backing store
166 * file within an existing vma.
167 * @start: start of the remapped virtual memory range
168 * @size: size of the remapped virtual memory range
169 * @prot: new protection bits of the range
170 * @pgoff: to be mapped page of the backing store file
171 * @flags: 0 or MAP_NONBLOCKED - the later will cause no IO.
173 * this syscall works purely via pagetables, so it's the most efficient
174 * way to map the same (large) file into a given virtual window. Unlike
175 * mmap()/mremap() it does not create any new vmas. The new mappings are
176 * also safe across swapout.
178 * NOTE: the 'prot' parameter right now is ignored, and the vma's default
179 * protection is used. Arbitrary protections might be implemented in the
182 asmlinkage
long sys_remap_file_pages(unsigned long start
, unsigned long size
,
183 unsigned long __prot
, unsigned long pgoff
, unsigned long flags
)
185 struct mm_struct
*mm
= current
->mm
;
186 struct address_space
*mapping
;
187 unsigned long end
= start
+ size
;
188 struct vm_area_struct
*vma
;
190 int has_write_lock
= 0;
195 * Sanitize the syscall parameters:
197 start
= start
& PAGE_MASK
;
198 size
= size
& PAGE_MASK
;
200 /* Does the address range wrap, or is the span zero-sized? */
201 if (start
+ size
<= start
)
204 /* Can we represent this offset inside this architecture's pte's? */
205 #if PTE_FILE_MAX_BITS < BITS_PER_LONG
206 if (pgoff
+ (size
>> PAGE_SHIFT
) >= (1UL << PTE_FILE_MAX_BITS
))
210 /* We need down_write() to change vma->vm_flags. */
211 down_read(&mm
->mmap_sem
);
213 vma
= find_vma(mm
, start
);
216 * Make sure the vma is shared, that it supports prefaulting,
217 * and that the remapped range is valid and fully within
218 * the single existing vma. vm_private_data is used as a
219 * swapout cursor in a VM_NONLINEAR vma (unless VM_RESERVED
220 * or VM_LOCKED, but VM_LOCKED could be revoked later on).
222 if (vma
&& (vma
->vm_flags
& VM_SHARED
) &&
223 (!vma
->vm_private_data
||
224 (vma
->vm_flags
& (VM_NONLINEAR
|VM_RESERVED
))) &&
225 vma
->vm_ops
&& vma
->vm_ops
->populate
&&
226 end
> start
&& start
>= vma
->vm_start
&&
227 end
<= vma
->vm_end
) {
229 /* Must set VM_NONLINEAR before any pages are populated. */
230 if (pgoff
!= linear_page_index(vma
, start
) &&
231 !(vma
->vm_flags
& VM_NONLINEAR
)) {
232 if (!has_write_lock
) {
233 up_read(&mm
->mmap_sem
);
234 down_write(&mm
->mmap_sem
);
238 mapping
= vma
->vm_file
->f_mapping
;
239 spin_lock(&mapping
->i_mmap_lock
);
240 flush_dcache_mmap_lock(mapping
);
241 vma
->vm_flags
|= VM_NONLINEAR
;
242 vma_prio_tree_remove(vma
, &mapping
->i_mmap
);
243 vma_nonlinear_insert(vma
, &mapping
->i_mmap_nonlinear
);
244 flush_dcache_mmap_unlock(mapping
);
245 spin_unlock(&mapping
->i_mmap_lock
);
248 err
= vma
->vm_ops
->populate(vma
, start
, size
,
250 pgoff
, flags
& MAP_NONBLOCK
);
253 * We can't clear VM_NONLINEAR because we'd have to do
254 * it after ->populate completes, and that would prevent
255 * downgrading the lock. (Locks can't be upgraded).
258 if (likely(!has_write_lock
))
259 up_read(&mm
->mmap_sem
);
261 up_write(&mm
->mmap_sem
);