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 flush_cache_page(vma
, addr
, pte_pfn(pte
));
31 pte
= ptep_clear_flush(vma
, addr
, ptep
);
32 page
= vm_normal_page(vma
, addr
, pte
);
36 page_remove_rmap(page
, vma
);
37 page_cache_release(page
);
41 free_swap_and_cache(pte_to_swp_entry(pte
));
42 pte_clear_not_present_full(mm
, addr
, ptep
, 0);
48 * Install a file page to a given virtual memory address, release any
49 * previously existing mapping.
51 int install_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
52 unsigned long addr
, struct page
*page
, pgprot_t prot
)
61 pte
= get_locked_pte(mm
, addr
, &ptl
);
66 * This page may have been truncated. Tell the
70 inode
= vma
->vm_file
->f_mapping
->host
;
71 size
= (i_size_read(inode
) + PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
72 if (!page
->mapping
|| page
->index
>= size
)
75 if (page_mapcount(page
) > INT_MAX
/2)
78 if (pte_none(*pte
) || !zap_pte(mm
, vma
, addr
, pte
))
79 inc_mm_counter(mm
, file_rss
);
81 flush_icache_page(vma
, page
);
82 pte_val
= mk_pte(page
, prot
);
83 set_pte_at(mm
, addr
, pte
, pte_val
);
84 page_add_file_rmap(page
);
85 update_mmu_cache(vma
, addr
, pte_val
);
86 lazy_mmu_prot_update(pte_val
);
89 pte_unmap_unlock(pte
, ptl
);
93 EXPORT_SYMBOL(install_page
);
96 * Install a file pte to a given virtual memory address, release any
97 * previously existing mapping.
99 int install_file_pte(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
100 unsigned long addr
, unsigned long pgoff
, pgprot_t prot
)
106 pte
= get_locked_pte(mm
, addr
, &ptl
);
110 if (!pte_none(*pte
) && zap_pte(mm
, vma
, addr
, pte
)) {
111 update_hiwater_rss(mm
);
112 dec_mm_counter(mm
, file_rss
);
115 set_pte_at(mm
, addr
, pte
, pgoff_to_pte(pgoff
));
117 * We don't need to run update_mmu_cache() here because the "file pte"
118 * being installed by install_file_pte() is not a real pte - it's a
119 * non-present entry (like a swap entry), noting what file offset should
120 * be mapped there when there's a fault (in a non-linear vma where
121 * that's not obvious).
123 pte_unmap_unlock(pte
, ptl
);
129 static int populate_range(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
130 unsigned long addr
, unsigned long size
, pgoff_t pgoff
)
135 err
= install_file_pte(mm
, vma
, addr
, pgoff
, vma
->vm_page_prot
);
149 * sys_remap_file_pages - remap arbitrary pages of a shared backing store
150 * file within an existing vma.
151 * @start: start of the remapped virtual memory range
152 * @size: size of the remapped virtual memory range
153 * @prot: new protection bits of the range
154 * @pgoff: to be mapped page of the backing store file
155 * @flags: 0 or MAP_NONBLOCKED - the later will cause no IO.
157 * this syscall works purely via pagetables, so it's the most efficient
158 * way to map the same (large) file into a given virtual window. Unlike
159 * mmap()/mremap() it does not create any new vmas. The new mappings are
160 * also safe across swapout.
162 * NOTE: the 'prot' parameter right now is ignored, and the vma's default
163 * protection is used. Arbitrary protections might be implemented in the
166 asmlinkage
long sys_remap_file_pages(unsigned long start
, unsigned long size
,
167 unsigned long __prot
, unsigned long pgoff
, unsigned long flags
)
169 struct mm_struct
*mm
= current
->mm
;
170 struct address_space
*mapping
;
171 unsigned long end
= start
+ size
;
172 struct vm_area_struct
*vma
;
174 int has_write_lock
= 0;
179 * Sanitize the syscall parameters:
181 start
= start
& PAGE_MASK
;
182 size
= size
& PAGE_MASK
;
184 /* Does the address range wrap, or is the span zero-sized? */
185 if (start
+ size
<= start
)
188 /* Can we represent this offset inside this architecture's pte's? */
189 #if PTE_FILE_MAX_BITS < BITS_PER_LONG
190 if (pgoff
+ (size
>> PAGE_SHIFT
) >= (1UL << PTE_FILE_MAX_BITS
))
194 /* We need down_write() to change vma->vm_flags. */
195 down_read(&mm
->mmap_sem
);
197 vma
= find_vma(mm
, start
);
200 * Make sure the vma is shared, that it supports prefaulting,
201 * and that the remapped range is valid and fully within
202 * the single existing vma. vm_private_data is used as a
203 * swapout cursor in a VM_NONLINEAR vma.
205 if (!vma
|| !(vma
->vm_flags
& VM_SHARED
))
208 if (vma
->vm_private_data
&& !(vma
->vm_flags
& VM_NONLINEAR
))
211 if ((!vma
->vm_ops
|| !vma
->vm_ops
->populate
) &&
212 !(vma
->vm_flags
& VM_CAN_NONLINEAR
))
215 if (end
<= start
|| start
< vma
->vm_start
|| end
> vma
->vm_end
)
218 /* Must set VM_NONLINEAR before any pages are populated. */
219 if (!(vma
->vm_flags
& VM_NONLINEAR
)) {
220 /* Don't need a nonlinear mapping, exit success */
221 if (pgoff
== linear_page_index(vma
, start
)) {
226 if (!has_write_lock
) {
227 up_read(&mm
->mmap_sem
);
228 down_write(&mm
->mmap_sem
);
232 mapping
= vma
->vm_file
->f_mapping
;
233 spin_lock(&mapping
->i_mmap_lock
);
234 flush_dcache_mmap_lock(mapping
);
235 vma
->vm_flags
|= VM_NONLINEAR
;
236 vma_prio_tree_remove(vma
, &mapping
->i_mmap
);
237 vma_nonlinear_insert(vma
, &mapping
->i_mmap_nonlinear
);
238 flush_dcache_mmap_unlock(mapping
);
239 spin_unlock(&mapping
->i_mmap_lock
);
242 if (vma
->vm_flags
& VM_CAN_NONLINEAR
) {
243 err
= populate_range(mm
, vma
, start
, size
, pgoff
);
244 if (!err
&& !(flags
& MAP_NONBLOCK
)) {
245 if (unlikely(has_write_lock
)) {
246 downgrade_write(&mm
->mmap_sem
);
249 make_pages_present(start
, start
+size
);
252 err
= vma
->vm_ops
->populate(vma
, start
, size
, vma
->vm_page_prot
,
253 pgoff
, flags
& MAP_NONBLOCK
);
256 * We can't clear VM_NONLINEAR because we'd have to do
257 * it after ->populate completes, and that would prevent
258 * downgrading the lock. (Locks can't be upgraded).
262 if (likely(!has_write_lock
))
263 up_read(&mm
->mmap_sem
);
265 up_write(&mm
->mmap_sem
);