]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * mm/mremap.c | |
3 | * | |
4 | * (C) Copyright 1996 Linus Torvalds | |
5 | * | |
046c6884 | 6 | * Address space accounting code <alan@lxorguk.ukuu.org.uk> |
1da177e4 LT |
7 | * (C) Copyright 2002 Red Hat Inc, All Rights Reserved |
8 | */ | |
9 | ||
10 | #include <linux/mm.h> | |
11 | #include <linux/hugetlb.h> | |
1da177e4 | 12 | #include <linux/shm.h> |
1ff82995 | 13 | #include <linux/ksm.h> |
1da177e4 LT |
14 | #include <linux/mman.h> |
15 | #include <linux/swap.h> | |
c59ede7b | 16 | #include <linux/capability.h> |
1da177e4 | 17 | #include <linux/fs.h> |
6dec97dc | 18 | #include <linux/swapops.h> |
1da177e4 LT |
19 | #include <linux/highmem.h> |
20 | #include <linux/security.h> | |
21 | #include <linux/syscalls.h> | |
cddb8a5c | 22 | #include <linux/mmu_notifier.h> |
2581d202 | 23 | #include <linux/uaccess.h> |
4abad2ca | 24 | #include <linux/mm-arch-hooks.h> |
1da177e4 | 25 | |
1da177e4 LT |
26 | #include <asm/cacheflush.h> |
27 | #include <asm/tlbflush.h> | |
28 | ||
ba470de4 RR |
29 | #include "internal.h" |
30 | ||
7be7a546 | 31 | static pmd_t *get_old_pmd(struct mm_struct *mm, unsigned long addr) |
1da177e4 LT |
32 | { |
33 | pgd_t *pgd; | |
34 | pud_t *pud; | |
35 | pmd_t *pmd; | |
36 | ||
37 | pgd = pgd_offset(mm, addr); | |
38 | if (pgd_none_or_clear_bad(pgd)) | |
39 | return NULL; | |
40 | ||
41 | pud = pud_offset(pgd, addr); | |
42 | if (pud_none_or_clear_bad(pud)) | |
43 | return NULL; | |
44 | ||
45 | pmd = pmd_offset(pud, addr); | |
37a1c49a | 46 | if (pmd_none(*pmd)) |
1da177e4 LT |
47 | return NULL; |
48 | ||
7be7a546 | 49 | return pmd; |
1da177e4 LT |
50 | } |
51 | ||
8ac1f832 AA |
52 | static pmd_t *alloc_new_pmd(struct mm_struct *mm, struct vm_area_struct *vma, |
53 | unsigned long addr) | |
1da177e4 LT |
54 | { |
55 | pgd_t *pgd; | |
56 | pud_t *pud; | |
c74df32c | 57 | pmd_t *pmd; |
1da177e4 LT |
58 | |
59 | pgd = pgd_offset(mm, addr); | |
1da177e4 LT |
60 | pud = pud_alloc(mm, pgd, addr); |
61 | if (!pud) | |
c74df32c | 62 | return NULL; |
7be7a546 | 63 | |
1da177e4 | 64 | pmd = pmd_alloc(mm, pud, addr); |
57a8f0cd | 65 | if (!pmd) |
c74df32c | 66 | return NULL; |
7be7a546 | 67 | |
8ac1f832 | 68 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
c74df32c | 69 | |
7be7a546 | 70 | return pmd; |
1da177e4 LT |
71 | } |
72 | ||
1d069b7d HD |
73 | static void take_rmap_locks(struct vm_area_struct *vma) |
74 | { | |
75 | if (vma->vm_file) | |
76 | i_mmap_lock_write(vma->vm_file->f_mapping); | |
77 | if (vma->anon_vma) | |
78 | anon_vma_lock_write(vma->anon_vma); | |
79 | } | |
80 | ||
81 | static void drop_rmap_locks(struct vm_area_struct *vma) | |
82 | { | |
83 | if (vma->anon_vma) | |
84 | anon_vma_unlock_write(vma->anon_vma); | |
85 | if (vma->vm_file) | |
86 | i_mmap_unlock_write(vma->vm_file->f_mapping); | |
87 | } | |
88 | ||
6dec97dc CG |
89 | static pte_t move_soft_dirty_pte(pte_t pte) |
90 | { | |
91 | /* | |
92 | * Set soft dirty bit so we can notice | |
93 | * in userspace the ptes were moved. | |
94 | */ | |
95 | #ifdef CONFIG_MEM_SOFT_DIRTY | |
96 | if (pte_present(pte)) | |
97 | pte = pte_mksoft_dirty(pte); | |
98 | else if (is_swap_pte(pte)) | |
99 | pte = pte_swp_mksoft_dirty(pte); | |
6dec97dc CG |
100 | #endif |
101 | return pte; | |
102 | } | |
103 | ||
7be7a546 HD |
104 | static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd, |
105 | unsigned long old_addr, unsigned long old_end, | |
106 | struct vm_area_struct *new_vma, pmd_t *new_pmd, | |
5d190420 | 107 | unsigned long new_addr, bool need_rmap_locks, bool *need_flush) |
1da177e4 | 108 | { |
1da177e4 | 109 | struct mm_struct *mm = vma->vm_mm; |
7be7a546 | 110 | pte_t *old_pte, *new_pte, pte; |
4c21e2f2 | 111 | spinlock_t *old_ptl, *new_ptl; |
5d190420 AL |
112 | bool force_flush = false; |
113 | unsigned long len = old_end - old_addr; | |
1da177e4 | 114 | |
38a76013 | 115 | /* |
c8c06efa | 116 | * When need_rmap_locks is true, we take the i_mmap_rwsem and anon_vma |
38a76013 ML |
117 | * locks to ensure that rmap will always observe either the old or the |
118 | * new ptes. This is the easiest way to avoid races with | |
119 | * truncate_pagecache(), page migration, etc... | |
120 | * | |
121 | * When need_rmap_locks is false, we use other ways to avoid | |
122 | * such races: | |
123 | * | |
124 | * - During exec() shift_arg_pages(), we use a specially tagged vma | |
125 | * which rmap call sites look for using is_vma_temporary_stack(). | |
126 | * | |
127 | * - During mremap(), new_vma is often known to be placed after vma | |
128 | * in rmap traversal order. This ensures rmap will always observe | |
129 | * either the old pte, or the new pte, or both (the page table locks | |
130 | * serialize access to individual ptes, but only rmap traversal | |
131 | * order guarantees that we won't miss both the old and new ptes). | |
132 | */ | |
1d069b7d HD |
133 | if (need_rmap_locks) |
134 | take_rmap_locks(vma); | |
1da177e4 | 135 | |
4c21e2f2 HD |
136 | /* |
137 | * We don't have to worry about the ordering of src and dst | |
138 | * pte locks because exclusive mmap_sem prevents deadlock. | |
139 | */ | |
c74df32c | 140 | old_pte = pte_offset_map_lock(mm, old_pmd, old_addr, &old_ptl); |
ece0e2b6 | 141 | new_pte = pte_offset_map(new_pmd, new_addr); |
4c21e2f2 HD |
142 | new_ptl = pte_lockptr(mm, new_pmd); |
143 | if (new_ptl != old_ptl) | |
f20dc5f7 | 144 | spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING); |
6606c3e0 | 145 | arch_enter_lazy_mmu_mode(); |
7be7a546 HD |
146 | |
147 | for (; old_addr < old_end; old_pte++, old_addr += PAGE_SIZE, | |
148 | new_pte++, new_addr += PAGE_SIZE) { | |
149 | if (pte_none(*old_pte)) | |
150 | continue; | |
5d190420 AL |
151 | |
152 | /* | |
153 | * We are remapping a dirty PTE, make sure to | |
154 | * flush TLB before we drop the PTL for the | |
155 | * old PTE or we may race with page_mkclean(). | |
156 | */ | |
157 | if (pte_present(*old_pte) && pte_dirty(*old_pte)) | |
158 | force_flush = true; | |
7b6efc2b | 159 | pte = ptep_get_and_clear(mm, old_addr, old_pte); |
7be7a546 | 160 | pte = move_pte(pte, new_vma->vm_page_prot, old_addr, new_addr); |
6dec97dc CG |
161 | pte = move_soft_dirty_pte(pte); |
162 | set_pte_at(mm, new_addr, new_pte, pte); | |
1da177e4 | 163 | } |
7be7a546 | 164 | |
6606c3e0 | 165 | arch_leave_lazy_mmu_mode(); |
4c21e2f2 HD |
166 | if (new_ptl != old_ptl) |
167 | spin_unlock(new_ptl); | |
ece0e2b6 | 168 | pte_unmap(new_pte - 1); |
5d190420 AL |
169 | if (force_flush) |
170 | flush_tlb_range(vma, old_end - len, old_end); | |
171 | else | |
172 | *need_flush = true; | |
c74df32c | 173 | pte_unmap_unlock(old_pte - 1, old_ptl); |
1d069b7d HD |
174 | if (need_rmap_locks) |
175 | drop_rmap_locks(vma); | |
1da177e4 LT |
176 | } |
177 | ||
7be7a546 HD |
178 | #define LATENCY_LIMIT (64 * PAGE_SIZE) |
179 | ||
b6a2fea3 | 180 | unsigned long move_page_tables(struct vm_area_struct *vma, |
1da177e4 | 181 | unsigned long old_addr, struct vm_area_struct *new_vma, |
38a76013 ML |
182 | unsigned long new_addr, unsigned long len, |
183 | bool need_rmap_locks) | |
1da177e4 | 184 | { |
7be7a546 HD |
185 | unsigned long extent, next, old_end; |
186 | pmd_t *old_pmd, *new_pmd; | |
7b6efc2b | 187 | bool need_flush = false; |
2ec74c3e SG |
188 | unsigned long mmun_start; /* For mmu_notifiers */ |
189 | unsigned long mmun_end; /* For mmu_notifiers */ | |
1da177e4 | 190 | |
7be7a546 HD |
191 | old_end = old_addr + len; |
192 | flush_cache_range(vma, old_addr, old_end); | |
1da177e4 | 193 | |
2ec74c3e SG |
194 | mmun_start = old_addr; |
195 | mmun_end = old_end; | |
196 | mmu_notifier_invalidate_range_start(vma->vm_mm, mmun_start, mmun_end); | |
7b6efc2b | 197 | |
7be7a546 | 198 | for (; old_addr < old_end; old_addr += extent, new_addr += extent) { |
1da177e4 | 199 | cond_resched(); |
7be7a546 | 200 | next = (old_addr + PMD_SIZE) & PMD_MASK; |
ebed4846 | 201 | /* even if next overflowed, extent below will be ok */ |
7be7a546 | 202 | extent = next - old_addr; |
ebed4846 AA |
203 | if (extent > old_end - old_addr) |
204 | extent = old_end - old_addr; | |
7be7a546 HD |
205 | old_pmd = get_old_pmd(vma->vm_mm, old_addr); |
206 | if (!old_pmd) | |
207 | continue; | |
8ac1f832 | 208 | new_pmd = alloc_new_pmd(vma->vm_mm, vma, new_addr); |
7be7a546 HD |
209 | if (!new_pmd) |
210 | break; | |
37a1c49a | 211 | if (pmd_trans_huge(*old_pmd)) { |
dd18dbc2 | 212 | if (extent == HPAGE_PMD_SIZE) { |
4b471e88 | 213 | bool moved; |
dd18dbc2 KS |
214 | /* See comment in move_ptes() */ |
215 | if (need_rmap_locks) | |
1d069b7d | 216 | take_rmap_locks(vma); |
bf8616d5 | 217 | moved = move_huge_pmd(vma, old_addr, new_addr, |
5d190420 AL |
218 | old_end, old_pmd, new_pmd, |
219 | &need_flush); | |
dd18dbc2 | 220 | if (need_rmap_locks) |
1d069b7d | 221 | drop_rmap_locks(vma); |
5d190420 | 222 | if (moved) |
4b471e88 | 223 | continue; |
dd18dbc2 | 224 | } |
4b471e88 | 225 | split_huge_pmd(vma, old_pmd, old_addr); |
337d9abf | 226 | if (pmd_trans_unstable(old_pmd)) |
6b9116a6 | 227 | continue; |
37a1c49a | 228 | } |
3ed3a4f0 | 229 | if (pte_alloc(new_vma->vm_mm, new_pmd, new_addr)) |
37a1c49a | 230 | break; |
7be7a546 HD |
231 | next = (new_addr + PMD_SIZE) & PMD_MASK; |
232 | if (extent > next - new_addr) | |
233 | extent = next - new_addr; | |
234 | if (extent > LATENCY_LIMIT) | |
235 | extent = LATENCY_LIMIT; | |
5d190420 AL |
236 | move_ptes(vma, old_pmd, old_addr, old_addr + extent, new_vma, |
237 | new_pmd, new_addr, need_rmap_locks, &need_flush); | |
1da177e4 | 238 | } |
5d190420 | 239 | if (need_flush) |
7b6efc2b AA |
240 | flush_tlb_range(vma, old_end-len, old_addr); |
241 | ||
2ec74c3e | 242 | mmu_notifier_invalidate_range_end(vma->vm_mm, mmun_start, mmun_end); |
7be7a546 HD |
243 | |
244 | return len + old_addr - old_end; /* how much done */ | |
1da177e4 LT |
245 | } |
246 | ||
247 | static unsigned long move_vma(struct vm_area_struct *vma, | |
248 | unsigned long old_addr, unsigned long old_len, | |
81909b84 | 249 | unsigned long new_len, unsigned long new_addr, bool *locked) |
1da177e4 LT |
250 | { |
251 | struct mm_struct *mm = vma->vm_mm; | |
252 | struct vm_area_struct *new_vma; | |
253 | unsigned long vm_flags = vma->vm_flags; | |
254 | unsigned long new_pgoff; | |
255 | unsigned long moved_len; | |
256 | unsigned long excess = 0; | |
365e9c87 | 257 | unsigned long hiwater_vm; |
1da177e4 | 258 | int split = 0; |
7103ad32 | 259 | int err; |
38a76013 | 260 | bool need_rmap_locks; |
1da177e4 LT |
261 | |
262 | /* | |
263 | * We'd prefer to avoid failure later on in do_munmap: | |
264 | * which may split one vma into three before unmapping. | |
265 | */ | |
266 | if (mm->map_count >= sysctl_max_map_count - 3) | |
267 | return -ENOMEM; | |
268 | ||
1ff82995 HD |
269 | /* |
270 | * Advise KSM to break any KSM pages in the area to be moved: | |
271 | * it would be confusing if they were to turn up at the new | |
272 | * location, where they happen to coincide with different KSM | |
273 | * pages recently unmapped. But leave vma->vm_flags as it was, | |
274 | * so KSM can come around to merge on vma and new_vma afterwards. | |
275 | */ | |
7103ad32 HD |
276 | err = ksm_madvise(vma, old_addr, old_addr + old_len, |
277 | MADV_UNMERGEABLE, &vm_flags); | |
278 | if (err) | |
279 | return err; | |
1ff82995 | 280 | |
1da177e4 | 281 | new_pgoff = vma->vm_pgoff + ((old_addr - vma->vm_start) >> PAGE_SHIFT); |
38a76013 ML |
282 | new_vma = copy_vma(&vma, new_addr, new_len, new_pgoff, |
283 | &need_rmap_locks); | |
1da177e4 LT |
284 | if (!new_vma) |
285 | return -ENOMEM; | |
286 | ||
38a76013 ML |
287 | moved_len = move_page_tables(vma, old_addr, new_vma, new_addr, old_len, |
288 | need_rmap_locks); | |
1da177e4 | 289 | if (moved_len < old_len) { |
df1eab30 | 290 | err = -ENOMEM; |
5477e70a ON |
291 | } else if (vma->vm_ops && vma->vm_ops->mremap) { |
292 | err = vma->vm_ops->mremap(new_vma); | |
df1eab30 ON |
293 | } |
294 | ||
295 | if (unlikely(err)) { | |
1da177e4 LT |
296 | /* |
297 | * On error, move entries back from new area to old, | |
298 | * which will succeed since page tables still there, | |
299 | * and then proceed to unmap new area instead of old. | |
300 | */ | |
38a76013 ML |
301 | move_page_tables(new_vma, new_addr, vma, old_addr, moved_len, |
302 | true); | |
1da177e4 LT |
303 | vma = new_vma; |
304 | old_len = new_len; | |
305 | old_addr = new_addr; | |
df1eab30 | 306 | new_addr = err; |
4abad2ca | 307 | } else { |
4abad2ca LD |
308 | arch_remap(mm, old_addr, old_addr + old_len, |
309 | new_addr, new_addr + new_len); | |
b2edffdd | 310 | } |
1da177e4 LT |
311 | |
312 | /* Conceal VM_ACCOUNT so old reservation is not undone */ | |
313 | if (vm_flags & VM_ACCOUNT) { | |
314 | vma->vm_flags &= ~VM_ACCOUNT; | |
315 | excess = vma->vm_end - vma->vm_start - old_len; | |
316 | if (old_addr > vma->vm_start && | |
317 | old_addr + old_len < vma->vm_end) | |
318 | split = 1; | |
319 | } | |
320 | ||
71799062 | 321 | /* |
365e9c87 HD |
322 | * If we failed to move page tables we still do total_vm increment |
323 | * since do_munmap() will decrement it by old_len == new_len. | |
324 | * | |
325 | * Since total_vm is about to be raised artificially high for a | |
326 | * moment, we need to restore high watermark afterwards: if stats | |
327 | * are taken meanwhile, total_vm and hiwater_vm appear too high. | |
328 | * If this were a serious issue, we'd add a flag to do_munmap(). | |
71799062 | 329 | */ |
365e9c87 | 330 | hiwater_vm = mm->hiwater_vm; |
84638335 | 331 | vm_stat_account(mm, vma->vm_flags, new_len >> PAGE_SHIFT); |
71799062 | 332 | |
d9fe4fab TK |
333 | /* Tell pfnmap has moved from this vma */ |
334 | if (unlikely(vma->vm_flags & VM_PFNMAP)) | |
335 | untrack_pfn_moved(vma); | |
336 | ||
1da177e4 LT |
337 | if (do_munmap(mm, old_addr, old_len) < 0) { |
338 | /* OOM: unable to split vma, just get accounts right */ | |
339 | vm_unacct_memory(excess >> PAGE_SHIFT); | |
340 | excess = 0; | |
341 | } | |
365e9c87 | 342 | mm->hiwater_vm = hiwater_vm; |
1da177e4 LT |
343 | |
344 | /* Restore VM_ACCOUNT if one or two pieces of vma left */ | |
345 | if (excess) { | |
346 | vma->vm_flags |= VM_ACCOUNT; | |
347 | if (split) | |
348 | vma->vm_next->vm_flags |= VM_ACCOUNT; | |
349 | } | |
350 | ||
1da177e4 LT |
351 | if (vm_flags & VM_LOCKED) { |
352 | mm->locked_vm += new_len >> PAGE_SHIFT; | |
81909b84 | 353 | *locked = true; |
1da177e4 LT |
354 | } |
355 | ||
356 | return new_addr; | |
357 | } | |
358 | ||
54f5de70 AV |
359 | static struct vm_area_struct *vma_to_resize(unsigned long addr, |
360 | unsigned long old_len, unsigned long new_len, unsigned long *p) | |
361 | { | |
362 | struct mm_struct *mm = current->mm; | |
363 | struct vm_area_struct *vma = find_vma(mm, addr); | |
1d391686 | 364 | unsigned long pgoff; |
54f5de70 AV |
365 | |
366 | if (!vma || vma->vm_start > addr) | |
6cd57613 | 367 | return ERR_PTR(-EFAULT); |
54f5de70 AV |
368 | |
369 | if (is_vm_hugetlb_page(vma)) | |
6cd57613 | 370 | return ERR_PTR(-EINVAL); |
54f5de70 AV |
371 | |
372 | /* We can't remap across vm area boundaries */ | |
373 | if (old_len > vma->vm_end - addr) | |
6cd57613 | 374 | return ERR_PTR(-EFAULT); |
54f5de70 | 375 | |
1d391686 ON |
376 | if (new_len == old_len) |
377 | return vma; | |
378 | ||
982134ba | 379 | /* Need to be careful about a growing mapping */ |
1d391686 ON |
380 | pgoff = (addr - vma->vm_start) >> PAGE_SHIFT; |
381 | pgoff += vma->vm_pgoff; | |
382 | if (pgoff + (new_len >> PAGE_SHIFT) < pgoff) | |
383 | return ERR_PTR(-EINVAL); | |
384 | ||
385 | if (vma->vm_flags & (VM_DONTEXPAND | VM_PFNMAP)) | |
386 | return ERR_PTR(-EFAULT); | |
54f5de70 AV |
387 | |
388 | if (vma->vm_flags & VM_LOCKED) { | |
389 | unsigned long locked, lock_limit; | |
390 | locked = mm->locked_vm << PAGE_SHIFT; | |
59e99e5b | 391 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
54f5de70 AV |
392 | locked += new_len - old_len; |
393 | if (locked > lock_limit && !capable(CAP_IPC_LOCK)) | |
6cd57613 | 394 | return ERR_PTR(-EAGAIN); |
54f5de70 AV |
395 | } |
396 | ||
84638335 KK |
397 | if (!may_expand_vm(mm, vma->vm_flags, |
398 | (new_len - old_len) >> PAGE_SHIFT)) | |
6cd57613 | 399 | return ERR_PTR(-ENOMEM); |
54f5de70 AV |
400 | |
401 | if (vma->vm_flags & VM_ACCOUNT) { | |
402 | unsigned long charged = (new_len - old_len) >> PAGE_SHIFT; | |
191c5424 | 403 | if (security_vm_enough_memory_mm(mm, charged)) |
6cd57613 | 404 | return ERR_PTR(-ENOMEM); |
54f5de70 AV |
405 | *p = charged; |
406 | } | |
407 | ||
408 | return vma; | |
54f5de70 AV |
409 | } |
410 | ||
81909b84 ML |
411 | static unsigned long mremap_to(unsigned long addr, unsigned long old_len, |
412 | unsigned long new_addr, unsigned long new_len, bool *locked) | |
ecc1a899 AV |
413 | { |
414 | struct mm_struct *mm = current->mm; | |
415 | struct vm_area_struct *vma; | |
416 | unsigned long ret = -EINVAL; | |
417 | unsigned long charged = 0; | |
097eed10 | 418 | unsigned long map_flags; |
ecc1a899 | 419 | |
f19cb115 | 420 | if (offset_in_page(new_addr)) |
ecc1a899 AV |
421 | goto out; |
422 | ||
423 | if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len) | |
424 | goto out; | |
425 | ||
9943242c ON |
426 | /* Ensure the old/new locations do not overlap */ |
427 | if (addr + old_len > new_addr && new_addr + new_len > addr) | |
ecc1a899 AV |
428 | goto out; |
429 | ||
ecc1a899 AV |
430 | ret = do_munmap(mm, new_addr, new_len); |
431 | if (ret) | |
432 | goto out; | |
433 | ||
434 | if (old_len >= new_len) { | |
435 | ret = do_munmap(mm, addr+new_len, old_len - new_len); | |
436 | if (ret && old_len != new_len) | |
437 | goto out; | |
438 | old_len = new_len; | |
439 | } | |
440 | ||
441 | vma = vma_to_resize(addr, old_len, new_len, &charged); | |
442 | if (IS_ERR(vma)) { | |
443 | ret = PTR_ERR(vma); | |
444 | goto out; | |
445 | } | |
446 | ||
097eed10 AV |
447 | map_flags = MAP_FIXED; |
448 | if (vma->vm_flags & VM_MAYSHARE) | |
449 | map_flags |= MAP_SHARED; | |
9206de95 | 450 | |
097eed10 AV |
451 | ret = get_unmapped_area(vma->vm_file, new_addr, new_len, vma->vm_pgoff + |
452 | ((addr - vma->vm_start) >> PAGE_SHIFT), | |
453 | map_flags); | |
f19cb115 | 454 | if (offset_in_page(ret)) |
097eed10 AV |
455 | goto out1; |
456 | ||
81909b84 | 457 | ret = move_vma(vma, addr, old_len, new_len, new_addr, locked); |
f19cb115 | 458 | if (!(offset_in_page(ret))) |
097eed10 AV |
459 | goto out; |
460 | out1: | |
461 | vm_unacct_memory(charged); | |
ecc1a899 AV |
462 | |
463 | out: | |
464 | return ret; | |
465 | } | |
466 | ||
1a0ef85f AV |
467 | static int vma_expandable(struct vm_area_struct *vma, unsigned long delta) |
468 | { | |
f106af4e | 469 | unsigned long end = vma->vm_end + delta; |
9206de95 | 470 | if (end < vma->vm_end) /* overflow */ |
f106af4e | 471 | return 0; |
9206de95 | 472 | if (vma->vm_next && vma->vm_next->vm_start < end) /* intersection */ |
f106af4e AV |
473 | return 0; |
474 | if (get_unmapped_area(NULL, vma->vm_start, end - vma->vm_start, | |
475 | 0, MAP_FIXED) & ~PAGE_MASK) | |
1a0ef85f | 476 | return 0; |
1a0ef85f AV |
477 | return 1; |
478 | } | |
479 | ||
1da177e4 LT |
480 | /* |
481 | * Expand (or shrink) an existing mapping, potentially moving it at the | |
482 | * same time (controlled by the MREMAP_MAYMOVE flag and available VM space) | |
483 | * | |
484 | * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise | |
485 | * This option implies MREMAP_MAYMOVE. | |
486 | */ | |
63a81db1 AV |
487 | SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len, |
488 | unsigned long, new_len, unsigned long, flags, | |
489 | unsigned long, new_addr) | |
1da177e4 | 490 | { |
d0de32d9 | 491 | struct mm_struct *mm = current->mm; |
1da177e4 LT |
492 | struct vm_area_struct *vma; |
493 | unsigned long ret = -EINVAL; | |
494 | unsigned long charged = 0; | |
81909b84 | 495 | bool locked = false; |
1da177e4 LT |
496 | |
497 | if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE)) | |
9a2458a6 RV |
498 | return ret; |
499 | ||
500 | if (flags & MREMAP_FIXED && !(flags & MREMAP_MAYMOVE)) | |
501 | return ret; | |
1da177e4 | 502 | |
f19cb115 | 503 | if (offset_in_page(addr)) |
9a2458a6 | 504 | return ret; |
1da177e4 LT |
505 | |
506 | old_len = PAGE_ALIGN(old_len); | |
507 | new_len = PAGE_ALIGN(new_len); | |
508 | ||
509 | /* | |
510 | * We allow a zero old-len as a special case | |
511 | * for DOS-emu "duplicate shm area" thing. But | |
512 | * a zero new-len is nonsensical. | |
513 | */ | |
514 | if (!new_len) | |
9a2458a6 RV |
515 | return ret; |
516 | ||
dc0ef0df MH |
517 | if (down_write_killable(¤t->mm->mmap_sem)) |
518 | return -EINTR; | |
1da177e4 | 519 | |
1da177e4 | 520 | if (flags & MREMAP_FIXED) { |
9a2458a6 RV |
521 | ret = mremap_to(addr, old_len, new_addr, new_len, |
522 | &locked); | |
ecc1a899 | 523 | goto out; |
1da177e4 LT |
524 | } |
525 | ||
526 | /* | |
527 | * Always allow a shrinking remap: that just unmaps | |
528 | * the unnecessary pages.. | |
529 | * do_munmap does all the needed commit accounting | |
530 | */ | |
531 | if (old_len >= new_len) { | |
d0de32d9 | 532 | ret = do_munmap(mm, addr+new_len, old_len - new_len); |
1da177e4 LT |
533 | if (ret && old_len != new_len) |
534 | goto out; | |
535 | ret = addr; | |
ecc1a899 | 536 | goto out; |
1da177e4 LT |
537 | } |
538 | ||
539 | /* | |
ecc1a899 | 540 | * Ok, we need to grow.. |
1da177e4 | 541 | */ |
54f5de70 AV |
542 | vma = vma_to_resize(addr, old_len, new_len, &charged); |
543 | if (IS_ERR(vma)) { | |
544 | ret = PTR_ERR(vma); | |
1da177e4 | 545 | goto out; |
119f657c | 546 | } |
1da177e4 | 547 | |
1da177e4 | 548 | /* old_len exactly to the end of the area.. |
1da177e4 | 549 | */ |
ecc1a899 | 550 | if (old_len == vma->vm_end - addr) { |
1da177e4 | 551 | /* can we just expand the current mapping? */ |
1a0ef85f | 552 | if (vma_expandable(vma, new_len - old_len)) { |
1da177e4 LT |
553 | int pages = (new_len - old_len) >> PAGE_SHIFT; |
554 | ||
5beb4930 RR |
555 | if (vma_adjust(vma, vma->vm_start, addr + new_len, |
556 | vma->vm_pgoff, NULL)) { | |
557 | ret = -ENOMEM; | |
558 | goto out; | |
559 | } | |
1da177e4 | 560 | |
84638335 | 561 | vm_stat_account(mm, vma->vm_flags, pages); |
1da177e4 | 562 | if (vma->vm_flags & VM_LOCKED) { |
d0de32d9 | 563 | mm->locked_vm += pages; |
81909b84 ML |
564 | locked = true; |
565 | new_addr = addr; | |
1da177e4 LT |
566 | } |
567 | ret = addr; | |
568 | goto out; | |
569 | } | |
570 | } | |
571 | ||
572 | /* | |
573 | * We weren't able to just expand or shrink the area, | |
574 | * we need to create a new one and move it.. | |
575 | */ | |
576 | ret = -ENOMEM; | |
577 | if (flags & MREMAP_MAYMOVE) { | |
ecc1a899 AV |
578 | unsigned long map_flags = 0; |
579 | if (vma->vm_flags & VM_MAYSHARE) | |
580 | map_flags |= MAP_SHARED; | |
581 | ||
582 | new_addr = get_unmapped_area(vma->vm_file, 0, new_len, | |
93587414 AV |
583 | vma->vm_pgoff + |
584 | ((addr - vma->vm_start) >> PAGE_SHIFT), | |
585 | map_flags); | |
f19cb115 | 586 | if (offset_in_page(new_addr)) { |
ecc1a899 AV |
587 | ret = new_addr; |
588 | goto out; | |
1da177e4 | 589 | } |
ecc1a899 | 590 | |
81909b84 | 591 | ret = move_vma(vma, addr, old_len, new_len, new_addr, &locked); |
1da177e4 LT |
592 | } |
593 | out: | |
f19cb115 | 594 | if (offset_in_page(ret)) { |
1da177e4 | 595 | vm_unacct_memory(charged); |
d456fb9e ON |
596 | locked = 0; |
597 | } | |
1da177e4 | 598 | up_write(¤t->mm->mmap_sem); |
81909b84 ML |
599 | if (locked && new_len > old_len) |
600 | mm_populate(new_addr + old_len, new_len - old_len); | |
1da177e4 LT |
601 | return ret; |
602 | } |