blob: e1031e1f6a61b68ba5055a162ea0d9161b510ba2 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022
23#include <asm/uaccess.h>
24#include <asm/cacheflush.h>
25#include <asm/tlbflush.h>
26
Rik van Rielba470de2008-10-18 20:26:50 -070027#include "internal.h"
28
Hugh Dickins7be7a542005-10-29 18:16:00 -070029static pmd_t *get_old_pmd(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070030{
31 pgd_t *pgd;
32 pud_t *pud;
33 pmd_t *pmd;
34
35 pgd = pgd_offset(mm, addr);
36 if (pgd_none_or_clear_bad(pgd))
37 return NULL;
38
39 pud = pud_offset(pgd, addr);
40 if (pud_none_or_clear_bad(pud))
41 return NULL;
42
43 pmd = pmd_offset(pud, addr);
Andrea Arcangeli37a1c492011-10-31 17:08:30 -070044 if (pmd_none(*pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -070045 return NULL;
46
Hugh Dickins7be7a542005-10-29 18:16:00 -070047 return pmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048}
49
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -080050static pmd_t *alloc_new_pmd(struct mm_struct *mm, struct vm_area_struct *vma,
51 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070052{
53 pgd_t *pgd;
54 pud_t *pud;
Hugh Dickinsc74df322005-10-29 18:16:23 -070055 pmd_t *pmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
57 pgd = pgd_offset(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 pud = pud_alloc(mm, pgd, addr);
59 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -070060 return NULL;
Hugh Dickins7be7a542005-10-29 18:16:00 -070061
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 pmd = pmd_alloc(mm, pud, addr);
Hugh Dickins7be7a542005-10-29 18:16:00 -070063 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -070064 return NULL;
Hugh Dickins7be7a542005-10-29 18:16:00 -070065
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -080066 VM_BUG_ON(pmd_trans_huge(*pmd));
Hugh Dickinsc74df322005-10-29 18:16:23 -070067
Hugh Dickins7be7a542005-10-29 18:16:00 -070068 return pmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -070069}
70
Hugh Dickins7be7a542005-10-29 18:16:00 -070071static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd,
72 unsigned long old_addr, unsigned long old_end,
73 struct vm_area_struct *new_vma, pmd_t *new_pmd,
Michel Lespinasse38a76012012-10-08 16:31:50 -070074 unsigned long new_addr, bool need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -070075{
76 struct address_space *mapping = NULL;
Michel Lespinasse38a76012012-10-08 16:31:50 -070077 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 struct mm_struct *mm = vma->vm_mm;
Hugh Dickins7be7a542005-10-29 18:16:00 -070079 pte_t *old_pte, *new_pte, pte;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -070080 spinlock_t *old_ptl, *new_ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
Michel Lespinasse38a76012012-10-08 16:31:50 -070082 /*
83 * When need_rmap_locks is true, we take the i_mmap_mutex and anon_vma
84 * locks to ensure that rmap will always observe either the old or the
85 * new ptes. This is the easiest way to avoid races with
86 * truncate_pagecache(), page migration, etc...
87 *
88 * When need_rmap_locks is false, we use other ways to avoid
89 * such races:
90 *
91 * - During exec() shift_arg_pages(), we use a specially tagged vma
92 * which rmap call sites look for using is_vma_temporary_stack().
93 *
94 * - During mremap(), new_vma is often known to be placed after vma
95 * in rmap traversal order. This ensures rmap will always observe
96 * either the old pte, or the new pte, or both (the page table locks
97 * serialize access to individual ptes, but only rmap traversal
98 * order guarantees that we won't miss both the old and new ptes).
99 */
100 if (need_rmap_locks) {
101 if (vma->vm_file) {
102 mapping = vma->vm_file->f_mapping;
103 mutex_lock(&mapping->i_mmap_mutex);
104 }
105 if (vma->anon_vma) {
106 anon_vma = vma->anon_vma;
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000107 anon_vma_lock_write(anon_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -0700108 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 }
Hugh Dickins7be7a542005-10-29 18:16:00 -0700110
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700111 /*
112 * We don't have to worry about the ordering of src and dst
113 * pte locks because exclusive mmap_sem prevents deadlock.
114 */
Hugh Dickinsc74df322005-10-29 18:16:23 -0700115 old_pte = pte_offset_map_lock(mm, old_pmd, old_addr, &old_ptl);
Peter Zijlstraece0e2b2010-10-26 14:21:52 -0700116 new_pte = pte_offset_map(new_pmd, new_addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700117 new_ptl = pte_lockptr(mm, new_pmd);
118 if (new_ptl != old_ptl)
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700119 spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700120 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121
Hugh Dickins7be7a542005-10-29 18:16:00 -0700122 for (; old_addr < old_end; old_pte++, old_addr += PAGE_SIZE,
123 new_pte++, new_addr += PAGE_SIZE) {
124 if (pte_none(*old_pte))
125 continue;
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700126 pte = ptep_get_and_clear(mm, old_addr, old_pte);
Hugh Dickins7be7a542005-10-29 18:16:00 -0700127 pte = move_pte(pte, new_vma->vm_page_prot, old_addr, new_addr);
128 set_pte_at(mm, new_addr, new_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 }
Hugh Dickins7be7a542005-10-29 18:16:00 -0700130
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700131 arch_leave_lazy_mmu_mode();
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700132 if (new_ptl != old_ptl)
133 spin_unlock(new_ptl);
Peter Zijlstraece0e2b2010-10-26 14:21:52 -0700134 pte_unmap(new_pte - 1);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700135 pte_unmap_unlock(old_pte - 1, old_ptl);
Michel Lespinasse108d6642012-10-08 16:31:36 -0700136 if (anon_vma)
137 anon_vma_unlock(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 if (mapping)
Peter Zijlstra3d48ae42011-05-24 17:12:06 -0700139 mutex_unlock(&mapping->i_mmap_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140}
141
Hugh Dickins7be7a542005-10-29 18:16:00 -0700142#define LATENCY_LIMIT (64 * PAGE_SIZE)
143
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700144unsigned long move_page_tables(struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 unsigned long old_addr, struct vm_area_struct *new_vma,
Michel Lespinasse38a76012012-10-08 16:31:50 -0700146 unsigned long new_addr, unsigned long len,
147 bool need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148{
Hugh Dickins7be7a542005-10-29 18:16:00 -0700149 unsigned long extent, next, old_end;
150 pmd_t *old_pmd, *new_pmd;
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700151 bool need_flush = false;
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700152 unsigned long mmun_start; /* For mmu_notifiers */
153 unsigned long mmun_end; /* For mmu_notifiers */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154
Hugh Dickins7be7a542005-10-29 18:16:00 -0700155 old_end = old_addr + len;
156 flush_cache_range(vma, old_addr, old_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700158 mmun_start = old_addr;
159 mmun_end = old_end;
160 mmu_notifier_invalidate_range_start(vma->vm_mm, mmun_start, mmun_end);
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700161
Hugh Dickins7be7a542005-10-29 18:16:00 -0700162 for (; old_addr < old_end; old_addr += extent, new_addr += extent) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 cond_resched();
Hugh Dickins7be7a542005-10-29 18:16:00 -0700164 next = (old_addr + PMD_SIZE) & PMD_MASK;
Andrea Arcangeliebed4842011-10-31 17:08:22 -0700165 /* even if next overflowed, extent below will be ok */
Hugh Dickins7be7a542005-10-29 18:16:00 -0700166 extent = next - old_addr;
Andrea Arcangeliebed4842011-10-31 17:08:22 -0700167 if (extent > old_end - old_addr)
168 extent = old_end - old_addr;
Hugh Dickins7be7a542005-10-29 18:16:00 -0700169 old_pmd = get_old_pmd(vma->vm_mm, old_addr);
170 if (!old_pmd)
171 continue;
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800172 new_pmd = alloc_new_pmd(vma->vm_mm, vma, new_addr);
Hugh Dickins7be7a542005-10-29 18:16:00 -0700173 if (!new_pmd)
174 break;
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700175 if (pmd_trans_huge(*old_pmd)) {
176 int err = 0;
177 if (extent == HPAGE_PMD_SIZE)
178 err = move_huge_pmd(vma, new_vma, old_addr,
179 new_addr, old_end,
180 old_pmd, new_pmd);
181 if (err > 0) {
182 need_flush = true;
183 continue;
184 } else if (!err) {
Kirill A. Shutemove1803772012-12-12 13:50:59 -0800185 split_huge_page_pmd(vma, old_addr, old_pmd);
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700186 }
187 VM_BUG_ON(pmd_trans_huge(*old_pmd));
188 }
189 if (pmd_none(*new_pmd) && __pte_alloc(new_vma->vm_mm, new_vma,
190 new_pmd, new_addr))
191 break;
Hugh Dickins7be7a542005-10-29 18:16:00 -0700192 next = (new_addr + PMD_SIZE) & PMD_MASK;
193 if (extent > next - new_addr)
194 extent = next - new_addr;
195 if (extent > LATENCY_LIMIT)
196 extent = LATENCY_LIMIT;
197 move_ptes(vma, old_pmd, old_addr, old_addr + extent,
Michel Lespinasse38a76012012-10-08 16:31:50 -0700198 new_vma, new_pmd, new_addr, need_rmap_locks);
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700199 need_flush = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 }
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700201 if (likely(need_flush))
202 flush_tlb_range(vma, old_end-len, old_addr);
203
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700204 mmu_notifier_invalidate_range_end(vma->vm_mm, mmun_start, mmun_end);
Hugh Dickins7be7a542005-10-29 18:16:00 -0700205
206 return len + old_addr - old_end; /* how much done */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207}
208
209static unsigned long move_vma(struct vm_area_struct *vma,
210 unsigned long old_addr, unsigned long old_len,
211 unsigned long new_len, unsigned long new_addr)
212{
213 struct mm_struct *mm = vma->vm_mm;
214 struct vm_area_struct *new_vma;
215 unsigned long vm_flags = vma->vm_flags;
216 unsigned long new_pgoff;
217 unsigned long moved_len;
218 unsigned long excess = 0;
Hugh Dickins365e9c872005-10-29 18:16:18 -0700219 unsigned long hiwater_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 int split = 0;
Hugh Dickins7103ad32009-09-21 17:02:28 -0700221 int err;
Michel Lespinasse38a76012012-10-08 16:31:50 -0700222 bool need_rmap_locks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223
224 /*
225 * We'd prefer to avoid failure later on in do_munmap:
226 * which may split one vma into three before unmapping.
227 */
228 if (mm->map_count >= sysctl_max_map_count - 3)
229 return -ENOMEM;
230
Hugh Dickins1ff82992009-09-21 17:02:05 -0700231 /*
232 * Advise KSM to break any KSM pages in the area to be moved:
233 * it would be confusing if they were to turn up at the new
234 * location, where they happen to coincide with different KSM
235 * pages recently unmapped. But leave vma->vm_flags as it was,
236 * so KSM can come around to merge on vma and new_vma afterwards.
237 */
Hugh Dickins7103ad32009-09-21 17:02:28 -0700238 err = ksm_madvise(vma, old_addr, old_addr + old_len,
239 MADV_UNMERGEABLE, &vm_flags);
240 if (err)
241 return err;
Hugh Dickins1ff82992009-09-21 17:02:05 -0700242
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 new_pgoff = vma->vm_pgoff + ((old_addr - vma->vm_start) >> PAGE_SHIFT);
Michel Lespinasse38a76012012-10-08 16:31:50 -0700244 new_vma = copy_vma(&vma, new_addr, new_len, new_pgoff,
245 &need_rmap_locks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 if (!new_vma)
247 return -ENOMEM;
248
Michel Lespinasse38a76012012-10-08 16:31:50 -0700249 moved_len = move_page_tables(vma, old_addr, new_vma, new_addr, old_len,
250 need_rmap_locks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 if (moved_len < old_len) {
252 /*
253 * On error, move entries back from new area to old,
254 * which will succeed since page tables still there,
255 * and then proceed to unmap new area instead of old.
256 */
Michel Lespinasse38a76012012-10-08 16:31:50 -0700257 move_page_tables(new_vma, new_addr, vma, old_addr, moved_len,
258 true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 vma = new_vma;
260 old_len = new_len;
261 old_addr = new_addr;
262 new_addr = -ENOMEM;
263 }
264
265 /* Conceal VM_ACCOUNT so old reservation is not undone */
266 if (vm_flags & VM_ACCOUNT) {
267 vma->vm_flags &= ~VM_ACCOUNT;
268 excess = vma->vm_end - vma->vm_start - old_len;
269 if (old_addr > vma->vm_start &&
270 old_addr + old_len < vma->vm_end)
271 split = 1;
272 }
273
Kirill Korotaev71799062005-05-16 21:53:18 -0700274 /*
Hugh Dickins365e9c872005-10-29 18:16:18 -0700275 * If we failed to move page tables we still do total_vm increment
276 * since do_munmap() will decrement it by old_len == new_len.
277 *
278 * Since total_vm is about to be raised artificially high for a
279 * moment, we need to restore high watermark afterwards: if stats
280 * are taken meanwhile, total_vm and hiwater_vm appear too high.
281 * If this were a serious issue, we'd add a flag to do_munmap().
Kirill Korotaev71799062005-05-16 21:53:18 -0700282 */
Hugh Dickins365e9c872005-10-29 18:16:18 -0700283 hiwater_vm = mm->hiwater_vm;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -0700284 vm_stat_account(mm, vma->vm_flags, vma->vm_file, new_len>>PAGE_SHIFT);
Kirill Korotaev71799062005-05-16 21:53:18 -0700285
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 if (do_munmap(mm, old_addr, old_len) < 0) {
287 /* OOM: unable to split vma, just get accounts right */
288 vm_unacct_memory(excess >> PAGE_SHIFT);
289 excess = 0;
290 }
Hugh Dickins365e9c872005-10-29 18:16:18 -0700291 mm->hiwater_vm = hiwater_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292
293 /* Restore VM_ACCOUNT if one or two pieces of vma left */
294 if (excess) {
295 vma->vm_flags |= VM_ACCOUNT;
296 if (split)
297 vma->vm_next->vm_flags |= VM_ACCOUNT;
298 }
299
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 if (vm_flags & VM_LOCKED) {
301 mm->locked_vm += new_len >> PAGE_SHIFT;
302 if (new_len > old_len)
Rik van Rielba470de2008-10-18 20:26:50 -0700303 mlock_vma_pages_range(new_vma, new_addr + old_len,
304 new_addr + new_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 }
306
307 return new_addr;
308}
309
Al Viro54f5de72009-11-24 07:17:46 -0500310static struct vm_area_struct *vma_to_resize(unsigned long addr,
311 unsigned long old_len, unsigned long new_len, unsigned long *p)
312{
313 struct mm_struct *mm = current->mm;
314 struct vm_area_struct *vma = find_vma(mm, addr);
315
316 if (!vma || vma->vm_start > addr)
317 goto Efault;
318
319 if (is_vm_hugetlb_page(vma))
320 goto Einval;
321
322 /* We can't remap across vm area boundaries */
323 if (old_len > vma->vm_end - addr)
324 goto Efault;
325
Linus Torvalds982134b2011-04-07 07:35:50 -0700326 /* Need to be careful about a growing mapping */
327 if (new_len > old_len) {
328 unsigned long pgoff;
329
330 if (vma->vm_flags & (VM_DONTEXPAND | VM_PFNMAP))
Al Viro54f5de72009-11-24 07:17:46 -0500331 goto Efault;
Linus Torvalds982134b2011-04-07 07:35:50 -0700332 pgoff = (addr - vma->vm_start) >> PAGE_SHIFT;
333 pgoff += vma->vm_pgoff;
334 if (pgoff + (new_len >> PAGE_SHIFT) < pgoff)
335 goto Einval;
Al Viro54f5de72009-11-24 07:17:46 -0500336 }
337
338 if (vma->vm_flags & VM_LOCKED) {
339 unsigned long locked, lock_limit;
340 locked = mm->locked_vm << PAGE_SHIFT;
Jiri Slaby59e99e52010-03-05 13:41:44 -0800341 lock_limit = rlimit(RLIMIT_MEMLOCK);
Al Viro54f5de72009-11-24 07:17:46 -0500342 locked += new_len - old_len;
343 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
344 goto Eagain;
345 }
346
347 if (!may_expand_vm(mm, (new_len - old_len) >> PAGE_SHIFT))
348 goto Enomem;
349
350 if (vma->vm_flags & VM_ACCOUNT) {
351 unsigned long charged = (new_len - old_len) >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +0000352 if (security_vm_enough_memory_mm(mm, charged))
Al Viro54f5de72009-11-24 07:17:46 -0500353 goto Efault;
354 *p = charged;
355 }
356
357 return vma;
358
359Efault: /* very odd choice for most of the cases, but... */
360 return ERR_PTR(-EFAULT);
361Einval:
362 return ERR_PTR(-EINVAL);
363Enomem:
364 return ERR_PTR(-ENOMEM);
365Eagain:
366 return ERR_PTR(-EAGAIN);
367}
368
Al Viroecc1a892009-11-24 07:28:07 -0500369static unsigned long mremap_to(unsigned long addr,
370 unsigned long old_len, unsigned long new_addr,
371 unsigned long new_len)
372{
373 struct mm_struct *mm = current->mm;
374 struct vm_area_struct *vma;
375 unsigned long ret = -EINVAL;
376 unsigned long charged = 0;
Al Viro097eed12009-11-24 08:43:52 -0500377 unsigned long map_flags;
Al Viroecc1a892009-11-24 07:28:07 -0500378
379 if (new_addr & ~PAGE_MASK)
380 goto out;
381
382 if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len)
383 goto out;
384
385 /* Check if the location we're moving into overlaps the
386 * old location at all, and fail if it does.
387 */
388 if ((new_addr <= addr) && (new_addr+new_len) > addr)
389 goto out;
390
391 if ((addr <= new_addr) && (addr+old_len) > new_addr)
392 goto out;
393
Al Viroecc1a892009-11-24 07:28:07 -0500394 ret = do_munmap(mm, new_addr, new_len);
395 if (ret)
396 goto out;
397
398 if (old_len >= new_len) {
399 ret = do_munmap(mm, addr+new_len, old_len - new_len);
400 if (ret && old_len != new_len)
401 goto out;
402 old_len = new_len;
403 }
404
405 vma = vma_to_resize(addr, old_len, new_len, &charged);
406 if (IS_ERR(vma)) {
407 ret = PTR_ERR(vma);
408 goto out;
409 }
410
Al Viro097eed12009-11-24 08:43:52 -0500411 map_flags = MAP_FIXED;
412 if (vma->vm_flags & VM_MAYSHARE)
413 map_flags |= MAP_SHARED;
Al Viro9206de92009-12-03 15:23:11 -0500414
Al Viro097eed12009-11-24 08:43:52 -0500415 ret = get_unmapped_area(vma->vm_file, new_addr, new_len, vma->vm_pgoff +
416 ((addr - vma->vm_start) >> PAGE_SHIFT),
417 map_flags);
Al Viroecc1a892009-11-24 07:28:07 -0500418 if (ret & ~PAGE_MASK)
Al Viro097eed12009-11-24 08:43:52 -0500419 goto out1;
420
421 ret = move_vma(vma, addr, old_len, new_len, new_addr);
422 if (!(ret & ~PAGE_MASK))
423 goto out;
424out1:
425 vm_unacct_memory(charged);
Al Viroecc1a892009-11-24 07:28:07 -0500426
427out:
428 return ret;
429}
430
Al Viro1a0ef852009-11-24 07:43:18 -0500431static int vma_expandable(struct vm_area_struct *vma, unsigned long delta)
432{
Al Virof106af42009-11-24 08:25:18 -0500433 unsigned long end = vma->vm_end + delta;
Al Viro9206de92009-12-03 15:23:11 -0500434 if (end < vma->vm_end) /* overflow */
Al Viro1a0ef852009-11-24 07:43:18 -0500435 return 0;
Al Viro9206de92009-12-03 15:23:11 -0500436 if (vma->vm_next && vma->vm_next->vm_start < end) /* intersection */
Al Virof106af42009-11-24 08:25:18 -0500437 return 0;
438 if (get_unmapped_area(NULL, vma->vm_start, end - vma->vm_start,
439 0, MAP_FIXED) & ~PAGE_MASK)
440 return 0;
Al Viro1a0ef852009-11-24 07:43:18 -0500441 return 1;
442}
443
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444/*
445 * Expand (or shrink) an existing mapping, potentially moving it at the
446 * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
447 *
448 * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
449 * This option implies MREMAP_MAYMOVE.
450 */
Al Viro63a81db2012-05-30 11:32:04 -0400451SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
452 unsigned long, new_len, unsigned long, flags,
453 unsigned long, new_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454{
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700455 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 struct vm_area_struct *vma;
457 unsigned long ret = -EINVAL;
458 unsigned long charged = 0;
459
Al Viro63a81db2012-05-30 11:32:04 -0400460 down_write(&current->mm->mmap_sem);
461
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE))
463 goto out;
464
465 if (addr & ~PAGE_MASK)
466 goto out;
467
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)
477 goto out;
478
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 if (flags & MREMAP_FIXED) {
Al Viroecc1a892009-11-24 07:28:07 -0500480 if (flags & MREMAP_MAYMOVE)
481 ret = mremap_to(addr, old_len, new_addr, new_len);
482 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 }
484
485 /*
486 * Always allow a shrinking remap: that just unmaps
487 * the unnecessary pages..
488 * do_munmap does all the needed commit accounting
489 */
490 if (old_len >= new_len) {
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700491 ret = do_munmap(mm, addr+new_len, old_len - new_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 if (ret && old_len != new_len)
493 goto out;
494 ret = addr;
Al Viroecc1a892009-11-24 07:28:07 -0500495 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 }
497
498 /*
Al Viroecc1a892009-11-24 07:28:07 -0500499 * Ok, we need to grow..
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 */
Al Viro54f5de72009-11-24 07:17:46 -0500501 vma = vma_to_resize(addr, old_len, new_len, &charged);
502 if (IS_ERR(vma)) {
503 ret = PTR_ERR(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 }
506
507 /* old_len exactly to the end of the area..
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 */
Al Viroecc1a892009-11-24 07:28:07 -0500509 if (old_len == vma->vm_end - addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 /* can we just expand the current mapping? */
Al Viro1a0ef852009-11-24 07:43:18 -0500511 if (vma_expandable(vma, new_len - old_len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 int pages = (new_len - old_len) >> PAGE_SHIFT;
513
Rik van Riel5beb4932010-03-05 13:42:07 -0800514 if (vma_adjust(vma, vma->vm_start, addr + new_len,
515 vma->vm_pgoff, NULL)) {
516 ret = -ENOMEM;
517 goto out;
518 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700520 vm_stat_account(mm, vma->vm_flags, vma->vm_file, pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 if (vma->vm_flags & VM_LOCKED) {
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700522 mm->locked_vm += pages;
Rik van Rielba470de2008-10-18 20:26:50 -0700523 mlock_vma_pages_range(vma, addr + old_len,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 addr + new_len);
525 }
526 ret = addr;
527 goto out;
528 }
529 }
530
531 /*
532 * We weren't able to just expand or shrink the area,
533 * we need to create a new one and move it..
534 */
535 ret = -ENOMEM;
536 if (flags & MREMAP_MAYMOVE) {
Al Viroecc1a892009-11-24 07:28:07 -0500537 unsigned long map_flags = 0;
538 if (vma->vm_flags & VM_MAYSHARE)
539 map_flags |= MAP_SHARED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
Al Viroecc1a892009-11-24 07:28:07 -0500541 new_addr = get_unmapped_area(vma->vm_file, 0, new_len,
Al Viro93587412009-11-24 08:45:24 -0500542 vma->vm_pgoff +
543 ((addr - vma->vm_start) >> PAGE_SHIFT),
544 map_flags);
Al Viroecc1a892009-11-24 07:28:07 -0500545 if (new_addr & ~PAGE_MASK) {
546 ret = new_addr;
547 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 }
Al Viroecc1a892009-11-24 07:28:07 -0500549
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 ret = move_vma(vma, addr, old_len, new_len, new_addr);
551 }
552out:
553 if (ret & ~PAGE_MASK)
554 vm_unacct_memory(charged);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 up_write(&current->mm->mmap_sem);
556 return ret;
557}