]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/fremap.c | |
3 | * | |
4 | * Explicit pagetable population and nonlinear (random) mappings support. | |
5 | * | |
6 | * started by Ingo Molnar, Copyright (C) 2002, 2003 | |
7 | */ | |
4af3c9cc | 8 | #include <linux/backing-dev.h> |
1da177e4 LT |
9 | #include <linux/mm.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/module.h> | |
17 | #include <linux/syscalls.h> | |
18 | ||
19 | #include <asm/mmu_context.h> | |
20 | #include <asm/cacheflush.h> | |
21 | #include <asm/tlbflush.h> | |
22 | ||
d0217ac0 | 23 | static void zap_pte(struct mm_struct *mm, struct vm_area_struct *vma, |
1da177e4 LT |
24 | unsigned long addr, pte_t *ptep) |
25 | { | |
26 | pte_t pte = *ptep; | |
27 | ||
1da177e4 | 28 | if (pte_present(pte)) { |
d0217ac0 NP |
29 | struct page *page; |
30 | ||
6aab341e | 31 | flush_cache_page(vma, addr, pte_pfn(pte)); |
1da177e4 | 32 | pte = ptep_clear_flush(vma, addr, ptep); |
6aab341e LT |
33 | page = vm_normal_page(vma, addr, pte); |
34 | if (page) { | |
35 | if (pte_dirty(pte)) | |
36 | set_page_dirty(page); | |
7de6b805 | 37 | page_remove_rmap(page, vma); |
6aab341e | 38 | page_cache_release(page); |
d0217ac0 NP |
39 | update_hiwater_rss(mm); |
40 | dec_mm_counter(mm, file_rss); | |
1da177e4 LT |
41 | } |
42 | } else { | |
43 | if (!pte_file(pte)) | |
44 | free_swap_and_cache(pte_to_swp_entry(pte)); | |
9888a1ca | 45 | pte_clear_not_present_full(mm, addr, ptep, 0); |
1da177e4 LT |
46 | } |
47 | } | |
48 | ||
1da177e4 LT |
49 | /* |
50 | * Install a file pte to a given virtual memory address, release any | |
51 | * previously existing mapping. | |
52 | */ | |
d0217ac0 | 53 | static int install_file_pte(struct mm_struct *mm, struct vm_area_struct *vma, |
1da177e4 LT |
54 | unsigned long addr, unsigned long pgoff, pgprot_t prot) |
55 | { | |
56 | int err = -ENOMEM; | |
57 | pte_t *pte; | |
c74df32c | 58 | spinlock_t *ptl; |
1da177e4 | 59 | |
c9cfcddf | 60 | pte = get_locked_pte(mm, addr, &ptl); |
1da177e4 | 61 | if (!pte) |
c74df32c | 62 | goto out; |
1da177e4 | 63 | |
d0217ac0 NP |
64 | if (!pte_none(*pte)) |
65 | zap_pte(mm, vma, addr, pte); | |
1da177e4 LT |
66 | |
67 | set_pte_at(mm, addr, pte, pgoff_to_pte(pgoff)); | |
668e0d8f HD |
68 | /* |
69 | * We don't need to run update_mmu_cache() here because the "file pte" | |
70 | * being installed by install_file_pte() is not a real pte - it's a | |
71 | * non-present entry (like a swap entry), noting what file offset should | |
72 | * be mapped there when there's a fault (in a non-linear vma where | |
73 | * that's not obvious). | |
74 | */ | |
c74df32c HD |
75 | pte_unmap_unlock(pte, ptl); |
76 | err = 0; | |
77 | out: | |
1da177e4 LT |
78 | return err; |
79 | } | |
80 | ||
54cb8821 NP |
81 | static int populate_range(struct mm_struct *mm, struct vm_area_struct *vma, |
82 | unsigned long addr, unsigned long size, pgoff_t pgoff) | |
83 | { | |
84 | int err; | |
85 | ||
86 | do { | |
87 | err = install_file_pte(mm, vma, addr, pgoff, vma->vm_page_prot); | |
88 | if (err) | |
89 | return err; | |
90 | ||
91 | size -= PAGE_SIZE; | |
92 | addr += PAGE_SIZE; | |
93 | pgoff++; | |
94 | } while (size); | |
95 | ||
96 | return 0; | |
97 | ||
98 | } | |
99 | ||
8d63494f RD |
100 | /** |
101 | * sys_remap_file_pages - remap arbitrary pages of an existing VM_SHARED vma | |
1da177e4 LT |
102 | * @start: start of the remapped virtual memory range |
103 | * @size: size of the remapped virtual memory range | |
8d63494f RD |
104 | * @prot: new protection bits of the range (see NOTE) |
105 | * @pgoff: to-be-mapped page of the backing store file | |
1da177e4 LT |
106 | * @flags: 0 or MAP_NONBLOCKED - the later will cause no IO. |
107 | * | |
8d63494f RD |
108 | * sys_remap_file_pages remaps arbitrary pages of an existing VM_SHARED vma |
109 | * (shared backing store file). | |
110 | * | |
111 | * This syscall works purely via pagetables, so it's the most efficient | |
1da177e4 LT |
112 | * way to map the same (large) file into a given virtual window. Unlike |
113 | * mmap()/mremap() it does not create any new vmas. The new mappings are | |
114 | * also safe across swapout. | |
115 | * | |
7682486b | 116 | * NOTE: the @prot parameter right now is ignored (but must be zero), |
8d63494f RD |
117 | * and the vma's default protection is used. Arbitrary protections |
118 | * might be implemented in the future. | |
1da177e4 LT |
119 | */ |
120 | asmlinkage long sys_remap_file_pages(unsigned long start, unsigned long size, | |
8d63494f | 121 | unsigned long prot, unsigned long pgoff, unsigned long flags) |
1da177e4 LT |
122 | { |
123 | struct mm_struct *mm = current->mm; | |
124 | struct address_space *mapping; | |
125 | unsigned long end = start + size; | |
126 | struct vm_area_struct *vma; | |
127 | int err = -EINVAL; | |
128 | int has_write_lock = 0; | |
129 | ||
8d63494f | 130 | if (prot) |
1da177e4 LT |
131 | return err; |
132 | /* | |
133 | * Sanitize the syscall parameters: | |
134 | */ | |
135 | start = start & PAGE_MASK; | |
136 | size = size & PAGE_MASK; | |
137 | ||
138 | /* Does the address range wrap, or is the span zero-sized? */ | |
139 | if (start + size <= start) | |
140 | return err; | |
141 | ||
142 | /* Can we represent this offset inside this architecture's pte's? */ | |
143 | #if PTE_FILE_MAX_BITS < BITS_PER_LONG | |
144 | if (pgoff + (size >> PAGE_SHIFT) >= (1UL << PTE_FILE_MAX_BITS)) | |
145 | return err; | |
146 | #endif | |
147 | ||
148 | /* We need down_write() to change vma->vm_flags. */ | |
149 | down_read(&mm->mmap_sem); | |
150 | retry: | |
151 | vma = find_vma(mm, start); | |
152 | ||
153 | /* | |
154 | * Make sure the vma is shared, that it supports prefaulting, | |
155 | * and that the remapped range is valid and fully within | |
156 | * the single existing vma. vm_private_data is used as a | |
101d2be7 | 157 | * swapout cursor in a VM_NONLINEAR vma. |
1da177e4 | 158 | */ |
54cb8821 NP |
159 | if (!vma || !(vma->vm_flags & VM_SHARED)) |
160 | goto out; | |
161 | ||
162 | if (vma->vm_private_data && !(vma->vm_flags & VM_NONLINEAR)) | |
163 | goto out; | |
164 | ||
dd204d63 | 165 | if (!(vma->vm_flags & VM_CAN_NONLINEAR)) |
54cb8821 NP |
166 | goto out; |
167 | ||
168 | if (end <= start || start < vma->vm_start || end > vma->vm_end) | |
169 | goto out; | |
170 | ||
171 | /* Must set VM_NONLINEAR before any pages are populated. */ | |
172 | if (!(vma->vm_flags & VM_NONLINEAR)) { | |
173 | /* Don't need a nonlinear mapping, exit success */ | |
174 | if (pgoff == linear_page_index(vma, start)) { | |
175 | err = 0; | |
176 | goto out; | |
177 | } | |
178 | ||
179 | if (!has_write_lock) { | |
180 | up_read(&mm->mmap_sem); | |
181 | down_write(&mm->mmap_sem); | |
182 | has_write_lock = 1; | |
183 | goto retry; | |
184 | } | |
185 | mapping = vma->vm_file->f_mapping; | |
3ee6dafc MS |
186 | /* |
187 | * page_mkclean doesn't work on nonlinear vmas, so if | |
188 | * dirty pages need to be accounted, emulate with linear | |
189 | * vmas. | |
190 | */ | |
191 | if (mapping_cap_account_dirty(mapping)) { | |
192 | unsigned long addr; | |
8a459e44 | 193 | struct file *file = vma->vm_file; |
3ee6dafc MS |
194 | |
195 | flags &= MAP_NONBLOCK; | |
8a459e44 ON |
196 | get_file(file); |
197 | addr = mmap_region(file, start, size, | |
3ee6dafc | 198 | flags, vma->vm_flags, pgoff, 1); |
8a459e44 | 199 | fput(file); |
3ee6dafc MS |
200 | if (IS_ERR_VALUE(addr)) { |
201 | err = addr; | |
202 | } else { | |
203 | BUG_ON(addr != start); | |
204 | err = 0; | |
205 | } | |
206 | goto out; | |
207 | } | |
54cb8821 NP |
208 | spin_lock(&mapping->i_mmap_lock); |
209 | flush_dcache_mmap_lock(mapping); | |
210 | vma->vm_flags |= VM_NONLINEAR; | |
211 | vma_prio_tree_remove(vma, &mapping->i_mmap); | |
212 | vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear); | |
213 | flush_dcache_mmap_unlock(mapping); | |
214 | spin_unlock(&mapping->i_mmap_lock); | |
215 | } | |
216 | ||
d0217ac0 NP |
217 | err = populate_range(mm, vma, start, size, pgoff); |
218 | if (!err && !(flags & MAP_NONBLOCK)) { | |
219 | if (unlikely(has_write_lock)) { | |
220 | downgrade_write(&mm->mmap_sem); | |
221 | has_write_lock = 0; | |
1da177e4 | 222 | } |
d0217ac0 NP |
223 | make_pages_present(start, start+size); |
224 | } | |
1da177e4 | 225 | |
54cb8821 NP |
226 | /* |
227 | * We can't clear VM_NONLINEAR because we'd have to do | |
228 | * it after ->populate completes, and that would prevent | |
229 | * downgrading the lock. (Locks can't be upgraded). | |
230 | */ | |
1da177e4 | 231 | |
54cb8821 | 232 | out: |
1da177e4 LT |
233 | if (likely(!has_write_lock)) |
234 | up_read(&mm->mmap_sem); | |
235 | else | |
236 | up_write(&mm->mmap_sem); | |
237 | ||
238 | return err; | |
239 | } | |
240 |