4 * Explicit pagetable population and nonlinear (random) mappings support.
6 * started by Ingo Molnar, Copyright (C) 2002, 2003
8 #include <linux/backing-dev.h>
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/syscalls.h>
17 #include <linux/mmu_notifier.h>
19 #include <asm/mmu_context.h>
20 #include <asm/cacheflush.h>
21 #include <asm/tlbflush.h>
25 static void zap_pte(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
26 unsigned long addr
, pte_t
*ptep
)
30 if (pte_present(pte
)) {
33 flush_cache_page(vma
, addr
, pte_pfn(pte
));
34 pte
= ptep_clear_flush(vma
, addr
, ptep
);
35 page
= vm_normal_page(vma
, addr
, pte
);
39 page_remove_rmap(page
);
40 page_cache_release(page
);
41 update_hiwater_rss(mm
);
42 dec_mm_counter(mm
, MM_FILEPAGES
);
46 free_swap_and_cache(pte_to_swp_entry(pte
));
47 pte_clear_not_present_full(mm
, addr
, ptep
, 0);
52 * Install a file pte to a given virtual memory address, release any
53 * previously existing mapping.
55 static int install_file_pte(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
56 unsigned long addr
, unsigned long pgoff
, pgprot_t prot
)
62 pte
= get_locked_pte(mm
, addr
, &ptl
);
67 zap_pte(mm
, vma
, addr
, pte
);
69 set_pte_at(mm
, addr
, pte
, pgoff_to_pte(pgoff
));
71 * We don't need to run update_mmu_cache() here because the "file pte"
72 * being installed by install_file_pte() is not a real pte - it's a
73 * non-present entry (like a swap entry), noting what file offset should
74 * be mapped there when there's a fault (in a non-linear vma where
75 * that's not obvious).
77 pte_unmap_unlock(pte
, ptl
);
83 static int populate_range(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
84 unsigned long addr
, unsigned long size
, pgoff_t pgoff
)
89 err
= install_file_pte(mm
, vma
, addr
, pgoff
, vma
->vm_page_prot
);
103 * sys_remap_file_pages - remap arbitrary pages of an existing VM_SHARED vma
104 * @start: start of the remapped virtual memory range
105 * @size: size of the remapped virtual memory range
106 * @prot: new protection bits of the range (see NOTE)
107 * @pgoff: to-be-mapped page of the backing store file
108 * @flags: 0 or MAP_NONBLOCKED - the later will cause no IO.
110 * sys_remap_file_pages remaps arbitrary pages of an existing VM_SHARED vma
111 * (shared backing store file).
113 * This syscall works purely via pagetables, so it's the most efficient
114 * way to map the same (large) file into a given virtual window. Unlike
115 * mmap()/mremap() it does not create any new vmas. The new mappings are
116 * also safe across swapout.
118 * NOTE: the @prot parameter right now is ignored (but must be zero),
119 * and the vma's default protection is used. Arbitrary protections
120 * might be implemented in the future.
122 SYSCALL_DEFINE5(remap_file_pages
, unsigned long, start
, unsigned long, size
,
123 unsigned long, prot
, unsigned long, pgoff
, unsigned long, flags
)
125 struct mm_struct
*mm
= current
->mm
;
126 struct address_space
*mapping
;
127 struct vm_area_struct
*vma
;
129 int has_write_lock
= 0;
134 * Sanitize the syscall parameters:
136 start
= start
& PAGE_MASK
;
137 size
= size
& PAGE_MASK
;
139 /* Does the address range wrap, or is the span zero-sized? */
140 if (start
+ size
<= start
)
143 /* Does pgoff wrap? */
144 if (pgoff
+ (size
>> PAGE_SHIFT
) < pgoff
)
147 /* Can we represent this offset inside this architecture's pte's? */
148 #if PTE_FILE_MAX_BITS < BITS_PER_LONG
149 if (pgoff
+ (size
>> PAGE_SHIFT
) >= (1UL << PTE_FILE_MAX_BITS
))
153 /* We need down_write() to change vma->vm_flags. */
154 down_read(&mm
->mmap_sem
);
156 vma
= find_vma(mm
, start
);
159 * Make sure the vma is shared, that it supports prefaulting,
160 * and that the remapped range is valid and fully within
161 * the single existing vma. vm_private_data is used as a
162 * swapout cursor in a VM_NONLINEAR vma.
164 if (!vma
|| !(vma
->vm_flags
& VM_SHARED
))
167 if (vma
->vm_private_data
&& !(vma
->vm_flags
& VM_NONLINEAR
))
170 if (!(vma
->vm_flags
& VM_CAN_NONLINEAR
))
173 if (start
< vma
->vm_start
|| start
+ size
> vma
->vm_end
)
176 /* Must set VM_NONLINEAR before any pages are populated. */
177 if (!(vma
->vm_flags
& VM_NONLINEAR
)) {
178 /* Don't need a nonlinear mapping, exit success */
179 if (pgoff
== linear_page_index(vma
, start
)) {
184 if (!has_write_lock
) {
185 up_read(&mm
->mmap_sem
);
186 down_write(&mm
->mmap_sem
);
190 mapping
= vma
->vm_file
->f_mapping
;
192 * page_mkclean doesn't work on nonlinear vmas, so if
193 * dirty pages need to be accounted, emulate with linear
196 if (mapping_cap_account_dirty(mapping
)) {
198 struct file
*file
= vma
->vm_file
;
200 flags
&= MAP_NONBLOCK
;
202 addr
= mmap_region(file
, start
, size
,
203 flags
, vma
->vm_flags
, pgoff
);
205 if (IS_ERR_VALUE(addr
)) {
208 BUG_ON(addr
!= start
);
213 mutex_lock(&mapping
->i_mmap_mutex
);
214 flush_dcache_mmap_lock(mapping
);
215 vma
->vm_flags
|= VM_NONLINEAR
;
216 vma_prio_tree_remove(vma
, &mapping
->i_mmap
);
217 vma_nonlinear_insert(vma
, &mapping
->i_mmap_nonlinear
);
218 flush_dcache_mmap_unlock(mapping
);
219 mutex_unlock(&mapping
->i_mmap_mutex
);
222 if (vma
->vm_flags
& VM_LOCKED
) {
224 * drop PG_Mlocked flag for over-mapped range
226 vm_flags_t saved_flags
= vma
->vm_flags
;
227 munlock_vma_pages_range(vma
, start
, start
+ size
);
228 vma
->vm_flags
= saved_flags
;
231 mmu_notifier_invalidate_range_start(mm
, start
, start
+ size
);
232 err
= populate_range(mm
, vma
, start
, size
, pgoff
);
233 mmu_notifier_invalidate_range_end(mm
, start
, start
+ size
);
234 if (!err
&& !(flags
& MAP_NONBLOCK
)) {
235 if (vma
->vm_flags
& VM_LOCKED
) {
237 * might be mapping previously unmapped range of file
239 mlock_vma_pages_range(vma
, start
, start
+ size
);
241 if (unlikely(has_write_lock
)) {
242 downgrade_write(&mm
->mmap_sem
);
245 make_pages_present(start
, start
+size
);
250 * We can't clear VM_NONLINEAR because we'd have to do
251 * it after ->populate completes, and that would prevent
252 * downgrading the lock. (Locks can't be upgraded).
256 if (likely(!has_write_lock
))
257 up_read(&mm
->mmap_sem
);
259 up_write(&mm
->mmap_sem
);
This page took 0.035171 seconds and 5 git commands to generate.