]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * mm/mremap.c | |
3 | * | |
4 | * (C) Copyright 1996 Linus Torvalds | |
5 | * | |
6 | * Address space accounting code <alan@redhat.com> | |
7 | * (C) Copyright 2002 Red Hat Inc, All Rights Reserved | |
8 | */ | |
9 | ||
10 | #include <linux/mm.h> | |
11 | #include <linux/hugetlb.h> | |
12 | #include <linux/slab.h> | |
13 | #include <linux/shm.h> | |
14 | #include <linux/mman.h> | |
15 | #include <linux/swap.h> | |
16 | #include <linux/fs.h> | |
17 | #include <linux/highmem.h> | |
18 | #include <linux/security.h> | |
19 | #include <linux/syscalls.h> | |
20 | ||
21 | #include <asm/uaccess.h> | |
22 | #include <asm/cacheflush.h> | |
23 | #include <asm/tlbflush.h> | |
24 | ||
7be7a546 | 25 | static pmd_t *get_old_pmd(struct mm_struct *mm, unsigned long addr) |
1da177e4 LT |
26 | { |
27 | pgd_t *pgd; | |
28 | pud_t *pud; | |
29 | pmd_t *pmd; | |
30 | ||
7be7a546 HD |
31 | /* |
32 | * We don't need page_table_lock: we have mmap_sem exclusively. | |
33 | */ | |
1da177e4 LT |
34 | pgd = pgd_offset(mm, addr); |
35 | if (pgd_none_or_clear_bad(pgd)) | |
36 | return NULL; | |
37 | ||
38 | pud = pud_offset(pgd, addr); | |
39 | if (pud_none_or_clear_bad(pud)) | |
40 | return NULL; | |
41 | ||
42 | pmd = pmd_offset(pud, addr); | |
43 | if (pmd_none_or_clear_bad(pmd)) | |
44 | return NULL; | |
45 | ||
7be7a546 | 46 | return pmd; |
1da177e4 LT |
47 | } |
48 | ||
7be7a546 | 49 | static pmd_t *alloc_new_pmd(struct mm_struct *mm, unsigned long addr) |
1da177e4 LT |
50 | { |
51 | pgd_t *pgd; | |
52 | pud_t *pud; | |
7be7a546 HD |
53 | pmd_t *pmd = NULL; |
54 | pte_t *pte; | |
1da177e4 | 55 | |
7be7a546 HD |
56 | /* |
57 | * We do need page_table_lock: because allocators expect that. | |
58 | */ | |
59 | spin_lock(&mm->page_table_lock); | |
1da177e4 | 60 | pgd = pgd_offset(mm, addr); |
1da177e4 LT |
61 | pud = pud_alloc(mm, pgd, addr); |
62 | if (!pud) | |
7be7a546 HD |
63 | goto out; |
64 | ||
1da177e4 | 65 | pmd = pmd_alloc(mm, pud, addr); |
7be7a546 HD |
66 | if (!pmd) |
67 | goto out; | |
68 | ||
69 | pte = pte_alloc_map(mm, pmd, addr); | |
70 | if (!pte) { | |
71 | pmd = NULL; | |
72 | goto out; | |
73 | } | |
74 | pte_unmap(pte); | |
75 | out: | |
76 | spin_unlock(&mm->page_table_lock); | |
77 | return pmd; | |
1da177e4 LT |
78 | } |
79 | ||
7be7a546 HD |
80 | static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd, |
81 | unsigned long old_addr, unsigned long old_end, | |
82 | struct vm_area_struct *new_vma, pmd_t *new_pmd, | |
83 | unsigned long new_addr) | |
1da177e4 LT |
84 | { |
85 | struct address_space *mapping = NULL; | |
86 | struct mm_struct *mm = vma->vm_mm; | |
7be7a546 | 87 | pte_t *old_pte, *new_pte, pte; |
1da177e4 LT |
88 | |
89 | if (vma->vm_file) { | |
90 | /* | |
91 | * Subtle point from Rajesh Venkatasubramanian: before | |
92 | * moving file-based ptes, we must lock vmtruncate out, | |
93 | * since it might clean the dst vma before the src vma, | |
94 | * and we propagate stale pages into the dst afterward. | |
95 | */ | |
96 | mapping = vma->vm_file->f_mapping; | |
97 | spin_lock(&mapping->i_mmap_lock); | |
98 | if (new_vma->vm_truncate_count && | |
99 | new_vma->vm_truncate_count != vma->vm_truncate_count) | |
100 | new_vma->vm_truncate_count = 0; | |
101 | } | |
1da177e4 | 102 | |
7be7a546 HD |
103 | spin_lock(&mm->page_table_lock); |
104 | old_pte = pte_offset_map(old_pmd, old_addr); | |
105 | new_pte = pte_offset_map_nested(new_pmd, new_addr); | |
106 | ||
107 | for (; old_addr < old_end; old_pte++, old_addr += PAGE_SIZE, | |
108 | new_pte++, new_addr += PAGE_SIZE) { | |
109 | if (pte_none(*old_pte)) | |
110 | continue; | |
111 | pte = ptep_clear_flush(vma, old_addr, old_pte); | |
112 | /* ZERO_PAGE can be dependant on virtual addr */ | |
113 | pte = move_pte(pte, new_vma->vm_page_prot, old_addr, new_addr); | |
114 | set_pte_at(mm, new_addr, new_pte, pte); | |
1da177e4 | 115 | } |
7be7a546 HD |
116 | |
117 | pte_unmap_nested(new_pte - 1); | |
118 | pte_unmap(old_pte - 1); | |
1da177e4 LT |
119 | spin_unlock(&mm->page_table_lock); |
120 | if (mapping) | |
121 | spin_unlock(&mapping->i_mmap_lock); | |
1da177e4 LT |
122 | } |
123 | ||
7be7a546 HD |
124 | #define LATENCY_LIMIT (64 * PAGE_SIZE) |
125 | ||
1da177e4 LT |
126 | static unsigned long move_page_tables(struct vm_area_struct *vma, |
127 | unsigned long old_addr, struct vm_area_struct *new_vma, | |
128 | unsigned long new_addr, unsigned long len) | |
129 | { | |
7be7a546 HD |
130 | unsigned long extent, next, old_end; |
131 | pmd_t *old_pmd, *new_pmd; | |
1da177e4 | 132 | |
7be7a546 HD |
133 | old_end = old_addr + len; |
134 | flush_cache_range(vma, old_addr, old_end); | |
1da177e4 | 135 | |
7be7a546 | 136 | for (; old_addr < old_end; old_addr += extent, new_addr += extent) { |
1da177e4 | 137 | cond_resched(); |
7be7a546 HD |
138 | next = (old_addr + PMD_SIZE) & PMD_MASK; |
139 | if (next - 1 > old_end) | |
140 | next = old_end; | |
141 | extent = next - old_addr; | |
142 | old_pmd = get_old_pmd(vma->vm_mm, old_addr); | |
143 | if (!old_pmd) | |
144 | continue; | |
145 | new_pmd = alloc_new_pmd(vma->vm_mm, new_addr); | |
146 | if (!new_pmd) | |
147 | break; | |
148 | next = (new_addr + PMD_SIZE) & PMD_MASK; | |
149 | if (extent > next - new_addr) | |
150 | extent = next - new_addr; | |
151 | if (extent > LATENCY_LIMIT) | |
152 | extent = LATENCY_LIMIT; | |
153 | move_ptes(vma, old_pmd, old_addr, old_addr + extent, | |
154 | new_vma, new_pmd, new_addr); | |
1da177e4 | 155 | } |
7be7a546 HD |
156 | |
157 | return len + old_addr - old_end; /* how much done */ | |
1da177e4 LT |
158 | } |
159 | ||
160 | static unsigned long move_vma(struct vm_area_struct *vma, | |
161 | unsigned long old_addr, unsigned long old_len, | |
162 | unsigned long new_len, unsigned long new_addr) | |
163 | { | |
164 | struct mm_struct *mm = vma->vm_mm; | |
165 | struct vm_area_struct *new_vma; | |
166 | unsigned long vm_flags = vma->vm_flags; | |
167 | unsigned long new_pgoff; | |
168 | unsigned long moved_len; | |
169 | unsigned long excess = 0; | |
170 | int split = 0; | |
171 | ||
172 | /* | |
173 | * We'd prefer to avoid failure later on in do_munmap: | |
174 | * which may split one vma into three before unmapping. | |
175 | */ | |
176 | if (mm->map_count >= sysctl_max_map_count - 3) | |
177 | return -ENOMEM; | |
178 | ||
179 | new_pgoff = vma->vm_pgoff + ((old_addr - vma->vm_start) >> PAGE_SHIFT); | |
180 | new_vma = copy_vma(&vma, new_addr, new_len, new_pgoff); | |
181 | if (!new_vma) | |
182 | return -ENOMEM; | |
183 | ||
184 | moved_len = move_page_tables(vma, old_addr, new_vma, new_addr, old_len); | |
185 | if (moved_len < old_len) { | |
186 | /* | |
187 | * On error, move entries back from new area to old, | |
188 | * which will succeed since page tables still there, | |
189 | * and then proceed to unmap new area instead of old. | |
190 | */ | |
191 | move_page_tables(new_vma, new_addr, vma, old_addr, moved_len); | |
192 | vma = new_vma; | |
193 | old_len = new_len; | |
194 | old_addr = new_addr; | |
195 | new_addr = -ENOMEM; | |
196 | } | |
197 | ||
198 | /* Conceal VM_ACCOUNT so old reservation is not undone */ | |
199 | if (vm_flags & VM_ACCOUNT) { | |
200 | vma->vm_flags &= ~VM_ACCOUNT; | |
201 | excess = vma->vm_end - vma->vm_start - old_len; | |
202 | if (old_addr > vma->vm_start && | |
203 | old_addr + old_len < vma->vm_end) | |
204 | split = 1; | |
205 | } | |
206 | ||
71799062 KK |
207 | /* |
208 | * if we failed to move page tables we still do total_vm increment | |
209 | * since do_munmap() will decrement it by old_len == new_len | |
210 | */ | |
211 | mm->total_vm += new_len >> PAGE_SHIFT; | |
ab50b8ed | 212 | vm_stat_account(mm, vma->vm_flags, vma->vm_file, new_len>>PAGE_SHIFT); |
71799062 | 213 | |
1da177e4 LT |
214 | if (do_munmap(mm, old_addr, old_len) < 0) { |
215 | /* OOM: unable to split vma, just get accounts right */ | |
216 | vm_unacct_memory(excess >> PAGE_SHIFT); | |
217 | excess = 0; | |
218 | } | |
219 | ||
220 | /* Restore VM_ACCOUNT if one or two pieces of vma left */ | |
221 | if (excess) { | |
222 | vma->vm_flags |= VM_ACCOUNT; | |
223 | if (split) | |
224 | vma->vm_next->vm_flags |= VM_ACCOUNT; | |
225 | } | |
226 | ||
1da177e4 LT |
227 | if (vm_flags & VM_LOCKED) { |
228 | mm->locked_vm += new_len >> PAGE_SHIFT; | |
229 | if (new_len > old_len) | |
230 | make_pages_present(new_addr + old_len, | |
231 | new_addr + new_len); | |
232 | } | |
233 | ||
234 | return new_addr; | |
235 | } | |
236 | ||
237 | /* | |
238 | * Expand (or shrink) an existing mapping, potentially moving it at the | |
239 | * same time (controlled by the MREMAP_MAYMOVE flag and available VM space) | |
240 | * | |
241 | * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise | |
242 | * This option implies MREMAP_MAYMOVE. | |
243 | */ | |
244 | unsigned long do_mremap(unsigned long addr, | |
245 | unsigned long old_len, unsigned long new_len, | |
246 | unsigned long flags, unsigned long new_addr) | |
247 | { | |
d0de32d9 | 248 | struct mm_struct *mm = current->mm; |
1da177e4 LT |
249 | struct vm_area_struct *vma; |
250 | unsigned long ret = -EINVAL; | |
251 | unsigned long charged = 0; | |
252 | ||
253 | if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE)) | |
254 | goto out; | |
255 | ||
256 | if (addr & ~PAGE_MASK) | |
257 | goto out; | |
258 | ||
259 | old_len = PAGE_ALIGN(old_len); | |
260 | new_len = PAGE_ALIGN(new_len); | |
261 | ||
262 | /* | |
263 | * We allow a zero old-len as a special case | |
264 | * for DOS-emu "duplicate shm area" thing. But | |
265 | * a zero new-len is nonsensical. | |
266 | */ | |
267 | if (!new_len) | |
268 | goto out; | |
269 | ||
270 | /* new_addr is only valid if MREMAP_FIXED is specified */ | |
271 | if (flags & MREMAP_FIXED) { | |
272 | if (new_addr & ~PAGE_MASK) | |
273 | goto out; | |
274 | if (!(flags & MREMAP_MAYMOVE)) | |
275 | goto out; | |
276 | ||
277 | if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len) | |
278 | goto out; | |
279 | ||
280 | /* Check if the location we're moving into overlaps the | |
281 | * old location at all, and fail if it does. | |
282 | */ | |
283 | if ((new_addr <= addr) && (new_addr+new_len) > addr) | |
284 | goto out; | |
285 | ||
286 | if ((addr <= new_addr) && (addr+old_len) > new_addr) | |
287 | goto out; | |
288 | ||
d0de32d9 | 289 | ret = do_munmap(mm, new_addr, new_len); |
1da177e4 LT |
290 | if (ret) |
291 | goto out; | |
292 | } | |
293 | ||
294 | /* | |
295 | * Always allow a shrinking remap: that just unmaps | |
296 | * the unnecessary pages.. | |
297 | * do_munmap does all the needed commit accounting | |
298 | */ | |
299 | if (old_len >= new_len) { | |
d0de32d9 | 300 | ret = do_munmap(mm, addr+new_len, old_len - new_len); |
1da177e4 LT |
301 | if (ret && old_len != new_len) |
302 | goto out; | |
303 | ret = addr; | |
304 | if (!(flags & MREMAP_FIXED) || (new_addr == addr)) | |
305 | goto out; | |
306 | old_len = new_len; | |
307 | } | |
308 | ||
309 | /* | |
310 | * Ok, we need to grow.. or relocate. | |
311 | */ | |
312 | ret = -EFAULT; | |
d0de32d9 | 313 | vma = find_vma(mm, addr); |
1da177e4 LT |
314 | if (!vma || vma->vm_start > addr) |
315 | goto out; | |
316 | if (is_vm_hugetlb_page(vma)) { | |
317 | ret = -EINVAL; | |
318 | goto out; | |
319 | } | |
320 | /* We can't remap across vm area boundaries */ | |
321 | if (old_len > vma->vm_end - addr) | |
322 | goto out; | |
323 | if (vma->vm_flags & VM_DONTEXPAND) { | |
324 | if (new_len > old_len) | |
325 | goto out; | |
326 | } | |
327 | if (vma->vm_flags & VM_LOCKED) { | |
328 | unsigned long locked, lock_limit; | |
d0de32d9 | 329 | locked = mm->locked_vm << PAGE_SHIFT; |
1da177e4 LT |
330 | lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur; |
331 | locked += new_len - old_len; | |
332 | ret = -EAGAIN; | |
333 | if (locked > lock_limit && !capable(CAP_IPC_LOCK)) | |
334 | goto out; | |
335 | } | |
d0de32d9 | 336 | if (!may_expand_vm(mm, (new_len - old_len) >> PAGE_SHIFT)) { |
119f657c | 337 | ret = -ENOMEM; |
1da177e4 | 338 | goto out; |
119f657c | 339 | } |
1da177e4 LT |
340 | |
341 | if (vma->vm_flags & VM_ACCOUNT) { | |
342 | charged = (new_len - old_len) >> PAGE_SHIFT; | |
343 | if (security_vm_enough_memory(charged)) | |
344 | goto out_nc; | |
345 | } | |
346 | ||
347 | /* old_len exactly to the end of the area.. | |
348 | * And we're not relocating the area. | |
349 | */ | |
350 | if (old_len == vma->vm_end - addr && | |
351 | !((flags & MREMAP_FIXED) && (addr != new_addr)) && | |
352 | (old_len != new_len || !(flags & MREMAP_MAYMOVE))) { | |
353 | unsigned long max_addr = TASK_SIZE; | |
354 | if (vma->vm_next) | |
355 | max_addr = vma->vm_next->vm_start; | |
356 | /* can we just expand the current mapping? */ | |
357 | if (max_addr - addr >= new_len) { | |
358 | int pages = (new_len - old_len) >> PAGE_SHIFT; | |
359 | ||
360 | vma_adjust(vma, vma->vm_start, | |
361 | addr + new_len, vma->vm_pgoff, NULL); | |
362 | ||
d0de32d9 HD |
363 | mm->total_vm += pages; |
364 | vm_stat_account(mm, vma->vm_flags, vma->vm_file, pages); | |
1da177e4 | 365 | if (vma->vm_flags & VM_LOCKED) { |
d0de32d9 | 366 | mm->locked_vm += pages; |
1da177e4 LT |
367 | make_pages_present(addr + old_len, |
368 | addr + new_len); | |
369 | } | |
370 | ret = addr; | |
371 | goto out; | |
372 | } | |
373 | } | |
374 | ||
375 | /* | |
376 | * We weren't able to just expand or shrink the area, | |
377 | * we need to create a new one and move it.. | |
378 | */ | |
379 | ret = -ENOMEM; | |
380 | if (flags & MREMAP_MAYMOVE) { | |
381 | if (!(flags & MREMAP_FIXED)) { | |
382 | unsigned long map_flags = 0; | |
383 | if (vma->vm_flags & VM_MAYSHARE) | |
384 | map_flags |= MAP_SHARED; | |
385 | ||
386 | new_addr = get_unmapped_area(vma->vm_file, 0, new_len, | |
387 | vma->vm_pgoff, map_flags); | |
388 | ret = new_addr; | |
389 | if (new_addr & ~PAGE_MASK) | |
390 | goto out; | |
391 | } | |
392 | ret = move_vma(vma, addr, old_len, new_len, new_addr); | |
393 | } | |
394 | out: | |
395 | if (ret & ~PAGE_MASK) | |
396 | vm_unacct_memory(charged); | |
397 | out_nc: | |
398 | return ret; | |
399 | } | |
400 | ||
401 | asmlinkage unsigned long sys_mremap(unsigned long addr, | |
402 | unsigned long old_len, unsigned long new_len, | |
403 | unsigned long flags, unsigned long new_addr) | |
404 | { | |
405 | unsigned long ret; | |
406 | ||
407 | down_write(¤t->mm->mmap_sem); | |
408 | ret = do_mremap(addr, old_len, new_len, flags, new_addr); | |
409 | up_write(¤t->mm->mmap_sem); | |
410 | return ret; | |
411 | } |