blob: 457d34ef3bf2fd6000e74d778270d5a02a4bb6e9 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * mm/mremap.c
3 *
4 * (C) Copyright 1996 Linus Torvalds
5 *
Alan Cox046c6882009-01-05 14:06:29 +00006 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * (C) Copyright 2002 Red Hat Inc, All Rights Reserved
8 */
9
10#include <linux/mm.h>
11#include <linux/hugetlb.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/shm.h>
Hugh Dickins1ff82992009-09-21 17:02:05 -070013#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/mman.h>
15#include <linux/swap.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080016#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/fs.h>
18#include <linux/highmem.h>
19#include <linux/security.h>
20#include <linux/syscalls.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070021#include <linux/mmu_notifier.h>
Clark Williamscf4aebc22013-02-07 09:46:59 -060022#include <linux/sched/sysctl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023
24#include <asm/uaccess.h>
25#include <asm/cacheflush.h>
26#include <asm/tlbflush.h>
27
Rik van Rielba470de2008-10-18 20:26:50 -070028#include "internal.h"
29
Hugh Dickins7be7a542005-10-29 18:16:00 -070030static pmd_t *get_old_pmd(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070031{
32 pgd_t *pgd;
33 pud_t *pud;
34 pmd_t *pmd;
35
36 pgd = pgd_offset(mm, addr);
37 if (pgd_none_or_clear_bad(pgd))
38 return NULL;
39
40 pud = pud_offset(pgd, addr);
41 if (pud_none_or_clear_bad(pud))
42 return NULL;
43
44 pmd = pmd_offset(pud, addr);
Andrea Arcangeli37a1c492011-10-31 17:08:30 -070045 if (pmd_none(*pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -070046 return NULL;
47
Hugh Dickins7be7a542005-10-29 18:16:00 -070048 return pmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -070049}
50
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -080051static pmd_t *alloc_new_pmd(struct mm_struct *mm, struct vm_area_struct *vma,
52 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070053{
54 pgd_t *pgd;
55 pud_t *pud;
Hugh Dickinsc74df322005-10-29 18:16:23 -070056 pmd_t *pmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
58 pgd = pgd_offset(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 pud = pud_alloc(mm, pgd, addr);
60 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -070061 return NULL;
Hugh Dickins7be7a542005-10-29 18:16:00 -070062
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 pmd = pmd_alloc(mm, pud, addr);
Hugh Dickins7be7a542005-10-29 18:16:00 -070064 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -070065 return NULL;
Hugh Dickins7be7a542005-10-29 18:16:00 -070066
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -080067 VM_BUG_ON(pmd_trans_huge(*pmd));
Hugh Dickinsc74df322005-10-29 18:16:23 -070068
Hugh Dickins7be7a542005-10-29 18:16:00 -070069 return pmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -070070}
71
Hugh Dickins7be7a542005-10-29 18:16:00 -070072static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd,
73 unsigned long old_addr, unsigned long old_end,
74 struct vm_area_struct *new_vma, pmd_t *new_pmd,
Michel Lespinasse38a76012012-10-08 16:31:50 -070075 unsigned long new_addr, bool need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -070076{
77 struct address_space *mapping = NULL;
Michel Lespinasse38a76012012-10-08 16:31:50 -070078 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 struct mm_struct *mm = vma->vm_mm;
Hugh Dickins7be7a542005-10-29 18:16:00 -070080 pte_t *old_pte, *new_pte, pte;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -070081 spinlock_t *old_ptl, *new_ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
Michel Lespinasse38a76012012-10-08 16:31:50 -070083 /*
84 * When need_rmap_locks is true, we take the i_mmap_mutex and anon_vma
85 * locks to ensure that rmap will always observe either the old or the
86 * new ptes. This is the easiest way to avoid races with
87 * truncate_pagecache(), page migration, etc...
88 *
89 * When need_rmap_locks is false, we use other ways to avoid
90 * such races:
91 *
92 * - During exec() shift_arg_pages(), we use a specially tagged vma
93 * which rmap call sites look for using is_vma_temporary_stack().
94 *
95 * - During mremap(), new_vma is often known to be placed after vma
96 * in rmap traversal order. This ensures rmap will always observe
97 * either the old pte, or the new pte, or both (the page table locks
98 * serialize access to individual ptes, but only rmap traversal
99 * order guarantees that we won't miss both the old and new ptes).
100 */
101 if (need_rmap_locks) {
102 if (vma->vm_file) {
103 mapping = vma->vm_file->f_mapping;
104 mutex_lock(&mapping->i_mmap_mutex);
105 }
106 if (vma->anon_vma) {
107 anon_vma = vma->anon_vma;
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000108 anon_vma_lock_write(anon_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -0700109 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 }
Hugh Dickins7be7a542005-10-29 18:16:00 -0700111
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700112 /*
113 * We don't have to worry about the ordering of src and dst
114 * pte locks because exclusive mmap_sem prevents deadlock.
115 */
Hugh Dickinsc74df322005-10-29 18:16:23 -0700116 old_pte = pte_offset_map_lock(mm, old_pmd, old_addr, &old_ptl);
Peter Zijlstraece0e2b2010-10-26 14:21:52 -0700117 new_pte = pte_offset_map(new_pmd, new_addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700118 new_ptl = pte_lockptr(mm, new_pmd);
119 if (new_ptl != old_ptl)
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700120 spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700121 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
Hugh Dickins7be7a542005-10-29 18:16:00 -0700123 for (; old_addr < old_end; old_pte++, old_addr += PAGE_SIZE,
124 new_pte++, new_addr += PAGE_SIZE) {
125 if (pte_none(*old_pte))
126 continue;
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700127 pte = ptep_get_and_clear(mm, old_addr, old_pte);
Hugh Dickins7be7a542005-10-29 18:16:00 -0700128 pte = move_pte(pte, new_vma->vm_page_prot, old_addr, new_addr);
Pavel Emelyanov0f8975e2013-07-03 15:01:20 -0700129 set_pte_at(mm, new_addr, new_pte, pte_mksoft_dirty(pte));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 }
Hugh Dickins7be7a542005-10-29 18:16:00 -0700131
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700132 arch_leave_lazy_mmu_mode();
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700133 if (new_ptl != old_ptl)
134 spin_unlock(new_ptl);
Peter Zijlstraece0e2b2010-10-26 14:21:52 -0700135 pte_unmap(new_pte - 1);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700136 pte_unmap_unlock(old_pte - 1, old_ptl);
Michel Lespinasse108d6642012-10-08 16:31:36 -0700137 if (anon_vma)
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800138 anon_vma_unlock_write(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700140 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141}
142
Hugh Dickins7be7a542005-10-29 18:16:00 -0700143#define LATENCY_LIMIT (64 * PAGE_SIZE)
144
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700145unsigned long move_page_tables(struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 unsigned long old_addr, struct vm_area_struct *new_vma,
Michel Lespinasse38a76012012-10-08 16:31:50 -0700147 unsigned long new_addr, unsigned long len,
148 bool need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149{
Hugh Dickins7be7a542005-10-29 18:16:00 -0700150 unsigned long extent, next, old_end;
151 pmd_t *old_pmd, *new_pmd;
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700152 bool need_flush = false;
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700153 unsigned long mmun_start; /* For mmu_notifiers */
154 unsigned long mmun_end; /* For mmu_notifiers */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155
Hugh Dickins7be7a542005-10-29 18:16:00 -0700156 old_end = old_addr + len;
157 flush_cache_range(vma, old_addr, old_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700159 mmun_start = old_addr;
160 mmun_end = old_end;
161 mmu_notifier_invalidate_range_start(vma->vm_mm, mmun_start, mmun_end);
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700162
Hugh Dickins7be7a542005-10-29 18:16:00 -0700163 for (; old_addr < old_end; old_addr += extent, new_addr += extent) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 cond_resched();
Hugh Dickins7be7a542005-10-29 18:16:00 -0700165 next = (old_addr + PMD_SIZE) & PMD_MASK;
Andrea Arcangeliebed4842011-10-31 17:08:22 -0700166 /* even if next overflowed, extent below will be ok */
Hugh Dickins7be7a542005-10-29 18:16:00 -0700167 extent = next - old_addr;
Andrea Arcangeliebed4842011-10-31 17:08:22 -0700168 if (extent > old_end - old_addr)
169 extent = old_end - old_addr;
Hugh Dickins7be7a542005-10-29 18:16:00 -0700170 old_pmd = get_old_pmd(vma->vm_mm, old_addr);
171 if (!old_pmd)
172 continue;
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800173 new_pmd = alloc_new_pmd(vma->vm_mm, vma, new_addr);
Hugh Dickins7be7a542005-10-29 18:16:00 -0700174 if (!new_pmd)
175 break;
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700176 if (pmd_trans_huge(*old_pmd)) {
177 int err = 0;
178 if (extent == HPAGE_PMD_SIZE)
179 err = move_huge_pmd(vma, new_vma, old_addr,
180 new_addr, old_end,
181 old_pmd, new_pmd);
182 if (err > 0) {
183 need_flush = true;
184 continue;
185 } else if (!err) {
Kirill A. Shutemove1803772012-12-12 13:50:59 -0800186 split_huge_page_pmd(vma, old_addr, old_pmd);
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700187 }
188 VM_BUG_ON(pmd_trans_huge(*old_pmd));
189 }
190 if (pmd_none(*new_pmd) && __pte_alloc(new_vma->vm_mm, new_vma,
191 new_pmd, new_addr))
192 break;
Hugh Dickins7be7a542005-10-29 18:16:00 -0700193 next = (new_addr + PMD_SIZE) & PMD_MASK;
194 if (extent > next - new_addr)
195 extent = next - new_addr;
196 if (extent > LATENCY_LIMIT)
197 extent = LATENCY_LIMIT;
198 move_ptes(vma, old_pmd, old_addr, old_addr + extent,
Michel Lespinasse38a76012012-10-08 16:31:50 -0700199 new_vma, new_pmd, new_addr, need_rmap_locks);
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700200 need_flush = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 }
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700202 if (likely(need_flush))
203 flush_tlb_range(vma, old_end-len, old_addr);
204
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700205 mmu_notifier_invalidate_range_end(vma->vm_mm, mmun_start, mmun_end);
Hugh Dickins7be7a542005-10-29 18:16:00 -0700206
207 return len + old_addr - old_end; /* how much done */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208}
209
210static unsigned long move_vma(struct vm_area_struct *vma,
211 unsigned long old_addr, unsigned long old_len,
Michel Lespinasse81909b82013-02-22 16:32:41 -0800212 unsigned long new_len, unsigned long new_addr, bool *locked)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213{
214 struct mm_struct *mm = vma->vm_mm;
215 struct vm_area_struct *new_vma;
216 unsigned long vm_flags = vma->vm_flags;
217 unsigned long new_pgoff;
218 unsigned long moved_len;
219 unsigned long excess = 0;
Hugh Dickins365e9c872005-10-29 18:16:18 -0700220 unsigned long hiwater_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 int split = 0;
Hugh Dickins7103ad32009-09-21 17:02:28 -0700222 int err;
Michel Lespinasse38a76012012-10-08 16:31:50 -0700223 bool need_rmap_locks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224
225 /*
226 * We'd prefer to avoid failure later on in do_munmap:
227 * which may split one vma into three before unmapping.
228 */
229 if (mm->map_count >= sysctl_max_map_count - 3)
230 return -ENOMEM;
231
Hugh Dickins1ff82992009-09-21 17:02:05 -0700232 /*
233 * Advise KSM to break any KSM pages in the area to be moved:
234 * it would be confusing if they were to turn up at the new
235 * location, where they happen to coincide with different KSM
236 * pages recently unmapped. But leave vma->vm_flags as it was,
237 * so KSM can come around to merge on vma and new_vma afterwards.
238 */
Hugh Dickins7103ad32009-09-21 17:02:28 -0700239 err = ksm_madvise(vma, old_addr, old_addr + old_len,
240 MADV_UNMERGEABLE, &vm_flags);
241 if (err)
242 return err;
Hugh Dickins1ff82992009-09-21 17:02:05 -0700243
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 new_pgoff = vma->vm_pgoff + ((old_addr - vma->vm_start) >> PAGE_SHIFT);
Michel Lespinasse38a76012012-10-08 16:31:50 -0700245 new_vma = copy_vma(&vma, new_addr, new_len, new_pgoff,
246 &need_rmap_locks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 if (!new_vma)
248 return -ENOMEM;
249
Michel Lespinasse38a76012012-10-08 16:31:50 -0700250 moved_len = move_page_tables(vma, old_addr, new_vma, new_addr, old_len,
251 need_rmap_locks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 if (moved_len < old_len) {
253 /*
254 * On error, move entries back from new area to old,
255 * which will succeed since page tables still there,
256 * and then proceed to unmap new area instead of old.
257 */
Michel Lespinasse38a76012012-10-08 16:31:50 -0700258 move_page_tables(new_vma, new_addr, vma, old_addr, moved_len,
259 true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 vma = new_vma;
261 old_len = new_len;
262 old_addr = new_addr;
263 new_addr = -ENOMEM;
264 }
265
266 /* Conceal VM_ACCOUNT so old reservation is not undone */
267 if (vm_flags & VM_ACCOUNT) {
268 vma->vm_flags &= ~VM_ACCOUNT;
269 excess = vma->vm_end - vma->vm_start - old_len;
270 if (old_addr > vma->vm_start &&
271 old_addr + old_len < vma->vm_end)
272 split = 1;
273 }
274
Kirill Korotaev71799062005-05-16 21:53:18 -0700275 /*
Hugh Dickins365e9c872005-10-29 18:16:18 -0700276 * If we failed to move page tables we still do total_vm increment
277 * since do_munmap() will decrement it by old_len == new_len.
278 *
279 * Since total_vm is about to be raised artificially high for a
280 * moment, we need to restore high watermark afterwards: if stats
281 * are taken meanwhile, total_vm and hiwater_vm appear too high.
282 * If this were a serious issue, we'd add a flag to do_munmap().
Kirill Korotaev71799062005-05-16 21:53:18 -0700283 */
Hugh Dickins365e9c872005-10-29 18:16:18 -0700284 hiwater_vm = mm->hiwater_vm;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -0700285 vm_stat_account(mm, vma->vm_flags, vma->vm_file, new_len>>PAGE_SHIFT);
Kirill Korotaev71799062005-05-16 21:53:18 -0700286
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 if (do_munmap(mm, old_addr, old_len) < 0) {
288 /* OOM: unable to split vma, just get accounts right */
289 vm_unacct_memory(excess >> PAGE_SHIFT);
290 excess = 0;
291 }
Hugh Dickins365e9c872005-10-29 18:16:18 -0700292 mm->hiwater_vm = hiwater_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293
294 /* Restore VM_ACCOUNT if one or two pieces of vma left */
295 if (excess) {
296 vma->vm_flags |= VM_ACCOUNT;
297 if (split)
298 vma->vm_next->vm_flags |= VM_ACCOUNT;
299 }
300
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 if (vm_flags & VM_LOCKED) {
302 mm->locked_vm += new_len >> PAGE_SHIFT;
Michel Lespinasse81909b82013-02-22 16:32:41 -0800303 *locked = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 }
305
306 return new_addr;
307}
308
Al Viro54f5de72009-11-24 07:17:46 -0500309static struct vm_area_struct *vma_to_resize(unsigned long addr,
310 unsigned long old_len, unsigned long new_len, unsigned long *p)
311{
312 struct mm_struct *mm = current->mm;
313 struct vm_area_struct *vma = find_vma(mm, addr);
314
315 if (!vma || vma->vm_start > addr)
316 goto Efault;
317
318 if (is_vm_hugetlb_page(vma))
319 goto Einval;
320
321 /* We can't remap across vm area boundaries */
322 if (old_len > vma->vm_end - addr)
323 goto Efault;
324
Linus Torvalds982134b2011-04-07 07:35:50 -0700325 /* Need to be careful about a growing mapping */
326 if (new_len > old_len) {
327 unsigned long pgoff;
328
329 if (vma->vm_flags & (VM_DONTEXPAND | VM_PFNMAP))
Al Viro54f5de72009-11-24 07:17:46 -0500330 goto Efault;
Linus Torvalds982134b2011-04-07 07:35:50 -0700331 pgoff = (addr - vma->vm_start) >> PAGE_SHIFT;
332 pgoff += vma->vm_pgoff;
333 if (pgoff + (new_len >> PAGE_SHIFT) < pgoff)
334 goto Einval;
Al Viro54f5de72009-11-24 07:17:46 -0500335 }
336
337 if (vma->vm_flags & VM_LOCKED) {
338 unsigned long locked, lock_limit;
339 locked = mm->locked_vm << PAGE_SHIFT;
Jiri Slaby59e99e52010-03-05 13:41:44 -0800340 lock_limit = rlimit(RLIMIT_MEMLOCK);
Al Viro54f5de72009-11-24 07:17:46 -0500341 locked += new_len - old_len;
342 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
343 goto Eagain;
344 }
345
346 if (!may_expand_vm(mm, (new_len - old_len) >> PAGE_SHIFT))
347 goto Enomem;
348
349 if (vma->vm_flags & VM_ACCOUNT) {
350 unsigned long charged = (new_len - old_len) >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +0000351 if (security_vm_enough_memory_mm(mm, charged))
Al Viro54f5de72009-11-24 07:17:46 -0500352 goto Efault;
353 *p = charged;
354 }
355
356 return vma;
357
358Efault: /* very odd choice for most of the cases, but... */
359 return ERR_PTR(-EFAULT);
360Einval:
361 return ERR_PTR(-EINVAL);
362Enomem:
363 return ERR_PTR(-ENOMEM);
364Eagain:
365 return ERR_PTR(-EAGAIN);
366}
367
Michel Lespinasse81909b82013-02-22 16:32:41 -0800368static unsigned long mremap_to(unsigned long addr, unsigned long old_len,
369 unsigned long new_addr, unsigned long new_len, bool *locked)
Al Viroecc1a892009-11-24 07:28:07 -0500370{
371 struct mm_struct *mm = current->mm;
372 struct vm_area_struct *vma;
373 unsigned long ret = -EINVAL;
374 unsigned long charged = 0;
Al Viro097eed12009-11-24 08:43:52 -0500375 unsigned long map_flags;
Al Viroecc1a892009-11-24 07:28:07 -0500376
377 if (new_addr & ~PAGE_MASK)
378 goto out;
379
380 if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len)
381 goto out;
382
383 /* Check if the location we're moving into overlaps the
384 * old location at all, and fail if it does.
385 */
386 if ((new_addr <= addr) && (new_addr+new_len) > addr)
387 goto out;
388
389 if ((addr <= new_addr) && (addr+old_len) > new_addr)
390 goto out;
391
Al Viroecc1a892009-11-24 07:28:07 -0500392 ret = do_munmap(mm, new_addr, new_len);
393 if (ret)
394 goto out;
395
396 if (old_len >= new_len) {
397 ret = do_munmap(mm, addr+new_len, old_len - new_len);
398 if (ret && old_len != new_len)
399 goto out;
400 old_len = new_len;
401 }
402
403 vma = vma_to_resize(addr, old_len, new_len, &charged);
404 if (IS_ERR(vma)) {
405 ret = PTR_ERR(vma);
406 goto out;
407 }
408
Al Viro097eed12009-11-24 08:43:52 -0500409 map_flags = MAP_FIXED;
410 if (vma->vm_flags & VM_MAYSHARE)
411 map_flags |= MAP_SHARED;
Al Viro9206de92009-12-03 15:23:11 -0500412
Al Viro097eed12009-11-24 08:43:52 -0500413 ret = get_unmapped_area(vma->vm_file, new_addr, new_len, vma->vm_pgoff +
414 ((addr - vma->vm_start) >> PAGE_SHIFT),
415 map_flags);
Al Viroecc1a892009-11-24 07:28:07 -0500416 if (ret & ~PAGE_MASK)
Al Viro097eed12009-11-24 08:43:52 -0500417 goto out1;
418
Michel Lespinasse81909b82013-02-22 16:32:41 -0800419 ret = move_vma(vma, addr, old_len, new_len, new_addr, locked);
Al Viro097eed12009-11-24 08:43:52 -0500420 if (!(ret & ~PAGE_MASK))
421 goto out;
422out1:
423 vm_unacct_memory(charged);
Al Viroecc1a892009-11-24 07:28:07 -0500424
425out:
426 return ret;
427}
428
Al Viro1a0ef852009-11-24 07:43:18 -0500429static int vma_expandable(struct vm_area_struct *vma, unsigned long delta)
430{
Al Virof106af42009-11-24 08:25:18 -0500431 unsigned long end = vma->vm_end + delta;
Al Viro9206de92009-12-03 15:23:11 -0500432 if (end < vma->vm_end) /* overflow */
Al Viro1a0ef852009-11-24 07:43:18 -0500433 return 0;
Al Viro9206de92009-12-03 15:23:11 -0500434 if (vma->vm_next && vma->vm_next->vm_start < end) /* intersection */
Al Virof106af42009-11-24 08:25:18 -0500435 return 0;
436 if (get_unmapped_area(NULL, vma->vm_start, end - vma->vm_start,
437 0, MAP_FIXED) & ~PAGE_MASK)
438 return 0;
Al Viro1a0ef852009-11-24 07:43:18 -0500439 return 1;
440}
441
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442/*
443 * Expand (or shrink) an existing mapping, potentially moving it at the
444 * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
445 *
446 * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
447 * This option implies MREMAP_MAYMOVE.
448 */
Al Viro63a81db2012-05-30 11:32:04 -0400449SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
450 unsigned long, new_len, unsigned long, flags,
451 unsigned long, new_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452{
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700453 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 struct vm_area_struct *vma;
455 unsigned long ret = -EINVAL;
456 unsigned long charged = 0;
Michel Lespinasse81909b82013-02-22 16:32:41 -0800457 bool locked = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458
459 if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE))
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700460 return ret;
461
462 if (flags & MREMAP_FIXED && !(flags & MREMAP_MAYMOVE))
463 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
465 if (addr & ~PAGE_MASK)
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700466 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
468 old_len = PAGE_ALIGN(old_len);
469 new_len = PAGE_ALIGN(new_len);
470
471 /*
472 * We allow a zero old-len as a special case
473 * for DOS-emu "duplicate shm area" thing. But
474 * a zero new-len is nonsensical.
475 */
476 if (!new_len)
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700477 return ret;
478
479 down_write(&current->mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 if (flags & MREMAP_FIXED) {
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700482 ret = mremap_to(addr, old_len, new_addr, new_len,
483 &locked);
Al Viroecc1a892009-11-24 07:28:07 -0500484 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 }
486
487 /*
488 * Always allow a shrinking remap: that just unmaps
489 * the unnecessary pages..
490 * do_munmap does all the needed commit accounting
491 */
492 if (old_len >= new_len) {
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700493 ret = do_munmap(mm, addr+new_len, old_len - new_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 if (ret && old_len != new_len)
495 goto out;
496 ret = addr;
Al Viroecc1a892009-11-24 07:28:07 -0500497 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 }
499
500 /*
Al Viroecc1a892009-11-24 07:28:07 -0500501 * Ok, we need to grow..
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 */
Al Viro54f5de72009-11-24 07:17:46 -0500503 vma = vma_to_resize(addr, old_len, new_len, &charged);
504 if (IS_ERR(vma)) {
505 ret = PTR_ERR(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 }
508
509 /* old_len exactly to the end of the area..
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 */
Al Viroecc1a892009-11-24 07:28:07 -0500511 if (old_len == vma->vm_end - addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 /* can we just expand the current mapping? */
Al Viro1a0ef852009-11-24 07:43:18 -0500513 if (vma_expandable(vma, new_len - old_len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 int pages = (new_len - old_len) >> PAGE_SHIFT;
515
Rik van Riel5beb4932010-03-05 13:42:07 -0800516 if (vma_adjust(vma, vma->vm_start, addr + new_len,
517 vma->vm_pgoff, NULL)) {
518 ret = -ENOMEM;
519 goto out;
520 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700522 vm_stat_account(mm, vma->vm_flags, vma->vm_file, pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 if (vma->vm_flags & VM_LOCKED) {
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700524 mm->locked_vm += pages;
Michel Lespinasse81909b82013-02-22 16:32:41 -0800525 locked = true;
526 new_addr = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 }
528 ret = addr;
529 goto out;
530 }
531 }
532
533 /*
534 * We weren't able to just expand or shrink the area,
535 * we need to create a new one and move it..
536 */
537 ret = -ENOMEM;
538 if (flags & MREMAP_MAYMOVE) {
Al Viroecc1a892009-11-24 07:28:07 -0500539 unsigned long map_flags = 0;
540 if (vma->vm_flags & VM_MAYSHARE)
541 map_flags |= MAP_SHARED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542
Al Viroecc1a892009-11-24 07:28:07 -0500543 new_addr = get_unmapped_area(vma->vm_file, 0, new_len,
Al Viro93587412009-11-24 08:45:24 -0500544 vma->vm_pgoff +
545 ((addr - vma->vm_start) >> PAGE_SHIFT),
546 map_flags);
Al Viroecc1a892009-11-24 07:28:07 -0500547 if (new_addr & ~PAGE_MASK) {
548 ret = new_addr;
549 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 }
Al Viroecc1a892009-11-24 07:28:07 -0500551
Michel Lespinasse81909b82013-02-22 16:32:41 -0800552 ret = move_vma(vma, addr, old_len, new_len, new_addr, &locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 }
554out:
555 if (ret & ~PAGE_MASK)
556 vm_unacct_memory(charged);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 up_write(&current->mm->mmap_sem);
Michel Lespinasse81909b82013-02-22 16:32:41 -0800558 if (locked && new_len > old_len)
559 mm_populate(new_addr + old_len, new_len - old_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 return ret;
561}