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 | */ | |
0b173bc4 | 8 | #include <linux/export.h> |
4af3c9cc | 9 | #include <linux/backing-dev.h> |
1da177e4 LT |
10 | #include <linux/mm.h> |
11 | #include <linux/swap.h> | |
12 | #include <linux/file.h> | |
13 | #include <linux/mman.h> | |
14 | #include <linux/pagemap.h> | |
15 | #include <linux/swapops.h> | |
16 | #include <linux/rmap.h> | |
1da177e4 | 17 | #include <linux/syscalls.h> |
cddb8a5c | 18 | #include <linux/mmu_notifier.h> |
1da177e4 LT |
19 | |
20 | #include <asm/mmu_context.h> | |
21 | #include <asm/cacheflush.h> | |
22 | #include <asm/tlbflush.h> | |
23 | ||
ba470de4 RR |
24 | #include "internal.h" |
25 | ||
88784396 HD |
26 | static int mm_counter(struct page *page) |
27 | { | |
28 | return PageAnon(page) ? MM_ANONPAGES : MM_FILEPAGES; | |
29 | } | |
30 | ||
d0217ac0 | 31 | static void zap_pte(struct mm_struct *mm, struct vm_area_struct *vma, |
1da177e4 LT |
32 | unsigned long addr, pte_t *ptep) |
33 | { | |
34 | pte_t pte = *ptep; | |
88784396 HD |
35 | struct page *page; |
36 | swp_entry_t entry; | |
1da177e4 | 37 | |
1da177e4 | 38 | if (pte_present(pte)) { |
6aab341e | 39 | flush_cache_page(vma, addr, pte_pfn(pte)); |
1da177e4 | 40 | pte = ptep_clear_flush(vma, addr, ptep); |
6aab341e LT |
41 | page = vm_normal_page(vma, addr, pte); |
42 | if (page) { | |
43 | if (pte_dirty(pte)) | |
44 | set_page_dirty(page); | |
88784396 HD |
45 | update_hiwater_rss(mm); |
46 | dec_mm_counter(mm, mm_counter(page)); | |
edc315fd | 47 | page_remove_rmap(page); |
6aab341e | 48 | page_cache_release(page); |
88784396 HD |
49 | } |
50 | } else { /* zap_pte() is not called when pte_none() */ | |
51 | if (!pte_file(pte)) { | |
d0217ac0 | 52 | update_hiwater_rss(mm); |
88784396 HD |
53 | entry = pte_to_swp_entry(pte); |
54 | if (non_swap_entry(entry)) { | |
55 | if (is_migration_entry(entry)) { | |
56 | page = migration_entry_to_page(entry); | |
57 | dec_mm_counter(mm, mm_counter(page)); | |
58 | } | |
59 | } else { | |
60 | free_swap_and_cache(entry); | |
61 | dec_mm_counter(mm, MM_SWAPENTS); | |
62 | } | |
1da177e4 | 63 | } |
9888a1ca | 64 | pte_clear_not_present_full(mm, addr, ptep, 0); |
1da177e4 LT |
65 | } |
66 | } | |
67 | ||
1da177e4 LT |
68 | /* |
69 | * Install a file pte to a given virtual memory address, release any | |
70 | * previously existing mapping. | |
71 | */ | |
d0217ac0 | 72 | static int install_file_pte(struct mm_struct *mm, struct vm_area_struct *vma, |
1da177e4 LT |
73 | unsigned long addr, unsigned long pgoff, pgprot_t prot) |
74 | { | |
75 | int err = -ENOMEM; | |
41bb3476 | 76 | pte_t *pte, ptfile; |
c74df32c | 77 | spinlock_t *ptl; |
1da177e4 | 78 | |
c9cfcddf | 79 | pte = get_locked_pte(mm, addr, &ptl); |
1da177e4 | 80 | if (!pte) |
c74df32c | 81 | goto out; |
1da177e4 | 82 | |
41bb3476 CG |
83 | ptfile = pgoff_to_pte(pgoff); |
84 | ||
85 | if (!pte_none(*pte)) { | |
86 | if (pte_present(*pte) && pte_soft_dirty(*pte)) | |
87 | pte_file_mksoft_dirty(ptfile); | |
d0217ac0 | 88 | zap_pte(mm, vma, addr, pte); |
41bb3476 | 89 | } |
1da177e4 | 90 | |
41bb3476 | 91 | set_pte_at(mm, addr, pte, ptfile); |
668e0d8f HD |
92 | /* |
93 | * We don't need to run update_mmu_cache() here because the "file pte" | |
94 | * being installed by install_file_pte() is not a real pte - it's a | |
95 | * non-present entry (like a swap entry), noting what file offset should | |
96 | * be mapped there when there's a fault (in a non-linear vma where | |
97 | * that's not obvious). | |
98 | */ | |
c74df32c HD |
99 | pte_unmap_unlock(pte, ptl); |
100 | err = 0; | |
101 | out: | |
1da177e4 LT |
102 | return err; |
103 | } | |
104 | ||
0b173bc4 KK |
105 | int generic_file_remap_pages(struct vm_area_struct *vma, unsigned long addr, |
106 | unsigned long size, pgoff_t pgoff) | |
54cb8821 | 107 | { |
0b173bc4 | 108 | struct mm_struct *mm = vma->vm_mm; |
54cb8821 NP |
109 | int err; |
110 | ||
111 | do { | |
112 | err = install_file_pte(mm, vma, addr, pgoff, vma->vm_page_prot); | |
113 | if (err) | |
114 | return err; | |
115 | ||
116 | size -= PAGE_SIZE; | |
117 | addr += PAGE_SIZE; | |
118 | pgoff++; | |
119 | } while (size); | |
120 | ||
0b173bc4 | 121 | return 0; |
54cb8821 | 122 | } |
0b173bc4 | 123 | EXPORT_SYMBOL(generic_file_remap_pages); |
54cb8821 | 124 | |
8d63494f RD |
125 | /** |
126 | * sys_remap_file_pages - remap arbitrary pages of an existing VM_SHARED vma | |
1da177e4 LT |
127 | * @start: start of the remapped virtual memory range |
128 | * @size: size of the remapped virtual memory range | |
8d63494f RD |
129 | * @prot: new protection bits of the range (see NOTE) |
130 | * @pgoff: to-be-mapped page of the backing store file | |
1da177e4 LT |
131 | * @flags: 0 or MAP_NONBLOCKED - the later will cause no IO. |
132 | * | |
8d63494f RD |
133 | * sys_remap_file_pages remaps arbitrary pages of an existing VM_SHARED vma |
134 | * (shared backing store file). | |
135 | * | |
136 | * This syscall works purely via pagetables, so it's the most efficient | |
1da177e4 LT |
137 | * way to map the same (large) file into a given virtual window. Unlike |
138 | * mmap()/mremap() it does not create any new vmas. The new mappings are | |
139 | * also safe across swapout. | |
140 | * | |
7682486b | 141 | * NOTE: the @prot parameter right now is ignored (but must be zero), |
8d63494f RD |
142 | * and the vma's default protection is used. Arbitrary protections |
143 | * might be implemented in the future. | |
1da177e4 | 144 | */ |
6a6160a7 HC |
145 | SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, |
146 | unsigned long, prot, unsigned long, pgoff, unsigned long, flags) | |
1da177e4 LT |
147 | { |
148 | struct mm_struct *mm = current->mm; | |
149 | struct address_space *mapping; | |
1da177e4 LT |
150 | struct vm_area_struct *vma; |
151 | int err = -EINVAL; | |
152 | int has_write_lock = 0; | |
a2362d24 | 153 | vm_flags_t vm_flags = 0; |
1da177e4 | 154 | |
8d63494f | 155 | if (prot) |
1da177e4 LT |
156 | return err; |
157 | /* | |
158 | * Sanitize the syscall parameters: | |
159 | */ | |
160 | start = start & PAGE_MASK; | |
161 | size = size & PAGE_MASK; | |
162 | ||
163 | /* Does the address range wrap, or is the span zero-sized? */ | |
164 | if (start + size <= start) | |
165 | return err; | |
166 | ||
5ec1055a LW |
167 | /* Does pgoff wrap? */ |
168 | if (pgoff + (size >> PAGE_SHIFT) < pgoff) | |
169 | return err; | |
170 | ||
1da177e4 LT |
171 | /* Can we represent this offset inside this architecture's pte's? */ |
172 | #if PTE_FILE_MAX_BITS < BITS_PER_LONG | |
173 | if (pgoff + (size >> PAGE_SHIFT) >= (1UL << PTE_FILE_MAX_BITS)) | |
174 | return err; | |
175 | #endif | |
176 | ||
177 | /* We need down_write() to change vma->vm_flags. */ | |
178 | down_read(&mm->mmap_sem); | |
179 | retry: | |
180 | vma = find_vma(mm, start); | |
181 | ||
182 | /* | |
183 | * Make sure the vma is shared, that it supports prefaulting, | |
184 | * and that the remapped range is valid and fully within | |
940e7da5 | 185 | * the single existing vma. |
1da177e4 | 186 | */ |
a2362d24 | 187 | if (!vma || !(vma->vm_flags & VM_SHARED)) |
54cb8821 NP |
188 | goto out; |
189 | ||
deb521c4 | 190 | if (!vma->vm_ops || !vma->vm_ops->remap_pages) |
54cb8821 NP |
191 | goto out; |
192 | ||
e92b05de | 193 | if (start < vma->vm_start || start + size > vma->vm_end) |
54cb8821 NP |
194 | goto out; |
195 | ||
196 | /* Must set VM_NONLINEAR before any pages are populated. */ | |
197 | if (!(vma->vm_flags & VM_NONLINEAR)) { | |
940e7da5 ML |
198 | /* |
199 | * vm_private_data is used as a swapout cursor | |
200 | * in a VM_NONLINEAR vma. | |
201 | */ | |
202 | if (vma->vm_private_data) | |
203 | goto out; | |
204 | ||
54cb8821 NP |
205 | /* Don't need a nonlinear mapping, exit success */ |
206 | if (pgoff == linear_page_index(vma, start)) { | |
207 | err = 0; | |
208 | goto out; | |
209 | } | |
210 | ||
211 | if (!has_write_lock) { | |
940e7da5 | 212 | get_write_lock: |
54cb8821 NP |
213 | up_read(&mm->mmap_sem); |
214 | down_write(&mm->mmap_sem); | |
215 | has_write_lock = 1; | |
216 | goto retry; | |
217 | } | |
218 | mapping = vma->vm_file->f_mapping; | |
3ee6dafc MS |
219 | /* |
220 | * page_mkclean doesn't work on nonlinear vmas, so if | |
221 | * dirty pages need to be accounted, emulate with linear | |
222 | * vmas. | |
223 | */ | |
224 | if (mapping_cap_account_dirty(mapping)) { | |
225 | unsigned long addr; | |
cb0942b8 | 226 | struct file *file = get_file(vma->vm_file); |
4eb91982 RR |
227 | /* mmap_region may free vma; grab the info now */ |
228 | vm_flags = vma->vm_flags; | |
3ee6dafc | 229 | |
4eb91982 | 230 | addr = mmap_region(file, start, size, vm_flags, pgoff); |
8a459e44 | 231 | fput(file); |
3ee6dafc MS |
232 | if (IS_ERR_VALUE(addr)) { |
233 | err = addr; | |
234 | } else { | |
235 | BUG_ON(addr != start); | |
236 | err = 0; | |
237 | } | |
4eb91982 | 238 | goto out_freed; |
3ee6dafc | 239 | } |
3d48ae45 | 240 | mutex_lock(&mapping->i_mmap_mutex); |
54cb8821 NP |
241 | flush_dcache_mmap_lock(mapping); |
242 | vma->vm_flags |= VM_NONLINEAR; | |
6b2dbba8 | 243 | vma_interval_tree_remove(vma, &mapping->i_mmap); |
54cb8821 NP |
244 | vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear); |
245 | flush_dcache_mmap_unlock(mapping); | |
3d48ae45 | 246 | mutex_unlock(&mapping->i_mmap_mutex); |
54cb8821 NP |
247 | } |
248 | ||
ba470de4 RR |
249 | if (vma->vm_flags & VM_LOCKED) { |
250 | /* | |
251 | * drop PG_Mlocked flag for over-mapped range | |
252 | */ | |
940e7da5 ML |
253 | if (!has_write_lock) |
254 | goto get_write_lock; | |
a1ea9549 | 255 | vm_flags = vma->vm_flags; |
ba470de4 | 256 | munlock_vma_pages_range(vma, start, start + size); |
a1ea9549 | 257 | vma->vm_flags = vm_flags; |
ba470de4 RR |
258 | } |
259 | ||
cddb8a5c | 260 | mmu_notifier_invalidate_range_start(mm, start, start + size); |
0b173bc4 | 261 | err = vma->vm_ops->remap_pages(vma, start, size, pgoff); |
cddb8a5c | 262 | mmu_notifier_invalidate_range_end(mm, start, start + size); |
1da177e4 | 263 | |
54cb8821 NP |
264 | /* |
265 | * We can't clear VM_NONLINEAR because we'd have to do | |
266 | * it after ->populate completes, and that would prevent | |
267 | * downgrading the lock. (Locks can't be upgraded). | |
268 | */ | |
1da177e4 | 269 | |
54cb8821 | 270 | out: |
6d7825b1 AM |
271 | if (vma) |
272 | vm_flags = vma->vm_flags; | |
4eb91982 | 273 | out_freed: |
1da177e4 LT |
274 | if (likely(!has_write_lock)) |
275 | up_read(&mm->mmap_sem); | |
276 | else | |
277 | up_write(&mm->mmap_sem); | |
a1ea9549 ML |
278 | if (!err && ((vm_flags & VM_LOCKED) || !(flags & MAP_NONBLOCK))) |
279 | mm_populate(start, size); | |
1da177e4 LT |
280 | |
281 | return err; | |
282 | } |