blob: 57dadc025c6444dd1038691d160803745db124f8 [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>
Cyrill Gorcunov6dec97d2013-08-27 12:37:18 +040018#include <linux/swapops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/highmem.h>
20#include <linux/security.h>
21#include <linux/syscalls.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070022#include <linux/mmu_notifier.h>
Clark Williamscf4aebc22013-02-07 09:46:59 -060023#include <linux/sched/sysctl.h>
Paul McQuade2581d202014-10-09 15:29:01 -070024#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <asm/cacheflush.h>
27#include <asm/tlbflush.h>
28
Rik van Rielba470de2008-10-18 20:26:50 -070029#include "internal.h"
30
Hugh Dickins7be7a542005-10-29 18:16:00 -070031static pmd_t *get_old_pmd(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070032{
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);
Andrea Arcangeli37a1c492011-10-31 17:08:30 -070046 if (pmd_none(*pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -070047 return NULL;
48
Hugh Dickins7be7a542005-10-29 18:16:00 -070049 return pmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050}
51
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -080052static pmd_t *alloc_new_pmd(struct mm_struct *mm, struct vm_area_struct *vma,
53 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070054{
55 pgd_t *pgd;
56 pud_t *pud;
Hugh Dickinsc74df322005-10-29 18:16:23 -070057 pmd_t *pmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
59 pgd = pgd_offset(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -070060 pud = pud_alloc(mm, pgd, addr);
61 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -070062 return NULL;
Hugh Dickins7be7a542005-10-29 18:16:00 -070063
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 pmd = pmd_alloc(mm, pud, addr);
Hugh Dickins57a8f0c2013-10-16 13:47:09 -070065 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -070066 return NULL;
Hugh Dickins7be7a542005-10-29 18:16:00 -070067
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -080068 VM_BUG_ON(pmd_trans_huge(*pmd));
Hugh Dickinsc74df322005-10-29 18:16:23 -070069
Hugh Dickins7be7a542005-10-29 18:16:00 -070070 return pmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -070071}
72
Cyrill Gorcunov6dec97d2013-08-27 12:37:18 +040073static pte_t move_soft_dirty_pte(pte_t pte)
74{
75 /*
76 * Set soft dirty bit so we can notice
77 * in userspace the ptes were moved.
78 */
79#ifdef CONFIG_MEM_SOFT_DIRTY
80 if (pte_present(pte))
81 pte = pte_mksoft_dirty(pte);
82 else if (is_swap_pte(pte))
83 pte = pte_swp_mksoft_dirty(pte);
Cyrill Gorcunov6dec97d2013-08-27 12:37:18 +040084#endif
85 return pte;
86}
87
Hugh Dickins7be7a542005-10-29 18:16:00 -070088static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd,
89 unsigned long old_addr, unsigned long old_end,
90 struct vm_area_struct *new_vma, pmd_t *new_pmd,
Michel Lespinasse38a76012012-10-08 16:31:50 -070091 unsigned long new_addr, bool need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -070092{
93 struct address_space *mapping = NULL;
Michel Lespinasse38a76012012-10-08 16:31:50 -070094 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 struct mm_struct *mm = vma->vm_mm;
Hugh Dickins7be7a542005-10-29 18:16:00 -070096 pte_t *old_pte, *new_pte, pte;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -070097 spinlock_t *old_ptl, *new_ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -070098
Michel Lespinasse38a76012012-10-08 16:31:50 -070099 /*
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800100 * When need_rmap_locks is true, we take the i_mmap_rwsem and anon_vma
Michel Lespinasse38a76012012-10-08 16:31:50 -0700101 * locks to ensure that rmap will always observe either the old or the
102 * new ptes. This is the easiest way to avoid races with
103 * truncate_pagecache(), page migration, etc...
104 *
105 * When need_rmap_locks is false, we use other ways to avoid
106 * such races:
107 *
108 * - During exec() shift_arg_pages(), we use a specially tagged vma
109 * which rmap call sites look for using is_vma_temporary_stack().
110 *
111 * - During mremap(), new_vma is often known to be placed after vma
112 * in rmap traversal order. This ensures rmap will always observe
113 * either the old pte, or the new pte, or both (the page table locks
114 * serialize access to individual ptes, but only rmap traversal
115 * order guarantees that we won't miss both the old and new ptes).
116 */
117 if (need_rmap_locks) {
118 if (vma->vm_file) {
119 mapping = vma->vm_file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800120 i_mmap_lock_write(mapping);
Michel Lespinasse38a76012012-10-08 16:31:50 -0700121 }
122 if (vma->anon_vma) {
123 anon_vma = vma->anon_vma;
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000124 anon_vma_lock_write(anon_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -0700125 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 }
Hugh Dickins7be7a542005-10-29 18:16:00 -0700127
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700128 /*
129 * We don't have to worry about the ordering of src and dst
130 * pte locks because exclusive mmap_sem prevents deadlock.
131 */
Hugh Dickinsc74df322005-10-29 18:16:23 -0700132 old_pte = pte_offset_map_lock(mm, old_pmd, old_addr, &old_ptl);
Peter Zijlstraece0e2b2010-10-26 14:21:52 -0700133 new_pte = pte_offset_map(new_pmd, new_addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700134 new_ptl = pte_lockptr(mm, new_pmd);
135 if (new_ptl != old_ptl)
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700136 spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700137 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138
Hugh Dickins7be7a542005-10-29 18:16:00 -0700139 for (; old_addr < old_end; old_pte++, old_addr += PAGE_SIZE,
140 new_pte++, new_addr += PAGE_SIZE) {
141 if (pte_none(*old_pte))
142 continue;
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700143 pte = ptep_get_and_clear(mm, old_addr, old_pte);
Hugh Dickins7be7a542005-10-29 18:16:00 -0700144 pte = move_pte(pte, new_vma->vm_page_prot, old_addr, new_addr);
Cyrill Gorcunov6dec97d2013-08-27 12:37:18 +0400145 pte = move_soft_dirty_pte(pte);
146 set_pte_at(mm, new_addr, new_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 }
Hugh Dickins7be7a542005-10-29 18:16:00 -0700148
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700149 arch_leave_lazy_mmu_mode();
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700150 if (new_ptl != old_ptl)
151 spin_unlock(new_ptl);
Peter Zijlstraece0e2b2010-10-26 14:21:52 -0700152 pte_unmap(new_pte - 1);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700153 pte_unmap_unlock(old_pte - 1, old_ptl);
Michel Lespinasse108d6642012-10-08 16:31:36 -0700154 if (anon_vma)
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800155 anon_vma_unlock_write(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800157 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158}
159
Hugh Dickins7be7a542005-10-29 18:16:00 -0700160#define LATENCY_LIMIT (64 * PAGE_SIZE)
161
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700162unsigned long move_page_tables(struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 unsigned long old_addr, struct vm_area_struct *new_vma,
Michel Lespinasse38a76012012-10-08 16:31:50 -0700164 unsigned long new_addr, unsigned long len,
165 bool need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166{
Hugh Dickins7be7a542005-10-29 18:16:00 -0700167 unsigned long extent, next, old_end;
168 pmd_t *old_pmd, *new_pmd;
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700169 bool need_flush = false;
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700170 unsigned long mmun_start; /* For mmu_notifiers */
171 unsigned long mmun_end; /* For mmu_notifiers */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
Hugh Dickins7be7a542005-10-29 18:16:00 -0700173 old_end = old_addr + len;
174 flush_cache_range(vma, old_addr, old_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700176 mmun_start = old_addr;
177 mmun_end = old_end;
178 mmu_notifier_invalidate_range_start(vma->vm_mm, mmun_start, mmun_end);
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700179
Hugh Dickins7be7a542005-10-29 18:16:00 -0700180 for (; old_addr < old_end; old_addr += extent, new_addr += extent) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 cond_resched();
Hugh Dickins7be7a542005-10-29 18:16:00 -0700182 next = (old_addr + PMD_SIZE) & PMD_MASK;
Andrea Arcangeliebed4842011-10-31 17:08:22 -0700183 /* even if next overflowed, extent below will be ok */
Hugh Dickins7be7a542005-10-29 18:16:00 -0700184 extent = next - old_addr;
Andrea Arcangeliebed4842011-10-31 17:08:22 -0700185 if (extent > old_end - old_addr)
186 extent = old_end - old_addr;
Hugh Dickins7be7a542005-10-29 18:16:00 -0700187 old_pmd = get_old_pmd(vma->vm_mm, old_addr);
188 if (!old_pmd)
189 continue;
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800190 new_pmd = alloc_new_pmd(vma->vm_mm, vma, new_addr);
Hugh Dickins7be7a542005-10-29 18:16:00 -0700191 if (!new_pmd)
192 break;
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700193 if (pmd_trans_huge(*old_pmd)) {
194 int err = 0;
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700195 if (extent == HPAGE_PMD_SIZE) {
Sasha Levin81d1b092014-10-09 15:28:10 -0700196 VM_BUG_ON_VMA(vma->vm_file || !vma->anon_vma,
197 vma);
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700198 /* See comment in move_ptes() */
199 if (need_rmap_locks)
200 anon_vma_lock_write(vma->anon_vma);
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700201 err = move_huge_pmd(vma, new_vma, old_addr,
202 new_addr, old_end,
203 old_pmd, new_pmd);
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700204 if (need_rmap_locks)
205 anon_vma_unlock_write(vma->anon_vma);
206 }
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700207 if (err > 0) {
208 need_flush = true;
209 continue;
210 } else if (!err) {
Kirill A. Shutemove1803772012-12-12 13:50:59 -0800211 split_huge_page_pmd(vma, old_addr, old_pmd);
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700212 }
213 VM_BUG_ON(pmd_trans_huge(*old_pmd));
214 }
215 if (pmd_none(*new_pmd) && __pte_alloc(new_vma->vm_mm, new_vma,
216 new_pmd, new_addr))
217 break;
Hugh Dickins7be7a542005-10-29 18:16:00 -0700218 next = (new_addr + PMD_SIZE) & PMD_MASK;
219 if (extent > next - new_addr)
220 extent = next - new_addr;
221 if (extent > LATENCY_LIMIT)
222 extent = LATENCY_LIMIT;
223 move_ptes(vma, old_pmd, old_addr, old_addr + extent,
Michel Lespinasse38a76012012-10-08 16:31:50 -0700224 new_vma, new_pmd, new_addr, need_rmap_locks);
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700225 need_flush = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 }
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700227 if (likely(need_flush))
228 flush_tlb_range(vma, old_end-len, old_addr);
229
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700230 mmu_notifier_invalidate_range_end(vma->vm_mm, mmun_start, mmun_end);
Hugh Dickins7be7a542005-10-29 18:16:00 -0700231
232 return len + old_addr - old_end; /* how much done */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233}
234
235static unsigned long move_vma(struct vm_area_struct *vma,
236 unsigned long old_addr, unsigned long old_len,
Michel Lespinasse81909b82013-02-22 16:32:41 -0800237 unsigned long new_len, unsigned long new_addr, bool *locked)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238{
239 struct mm_struct *mm = vma->vm_mm;
240 struct vm_area_struct *new_vma;
241 unsigned long vm_flags = vma->vm_flags;
242 unsigned long new_pgoff;
243 unsigned long moved_len;
244 unsigned long excess = 0;
Hugh Dickins365e9c872005-10-29 18:16:18 -0700245 unsigned long hiwater_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 int split = 0;
Hugh Dickins7103ad32009-09-21 17:02:28 -0700247 int err;
Michel Lespinasse38a76012012-10-08 16:31:50 -0700248 bool need_rmap_locks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
250 /*
251 * We'd prefer to avoid failure later on in do_munmap:
252 * which may split one vma into three before unmapping.
253 */
254 if (mm->map_count >= sysctl_max_map_count - 3)
255 return -ENOMEM;
256
Hugh Dickins1ff82992009-09-21 17:02:05 -0700257 /*
258 * Advise KSM to break any KSM pages in the area to be moved:
259 * it would be confusing if they were to turn up at the new
260 * location, where they happen to coincide with different KSM
261 * pages recently unmapped. But leave vma->vm_flags as it was,
262 * so KSM can come around to merge on vma and new_vma afterwards.
263 */
Hugh Dickins7103ad32009-09-21 17:02:28 -0700264 err = ksm_madvise(vma, old_addr, old_addr + old_len,
265 MADV_UNMERGEABLE, &vm_flags);
266 if (err)
267 return err;
Hugh Dickins1ff82992009-09-21 17:02:05 -0700268
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 new_pgoff = vma->vm_pgoff + ((old_addr - vma->vm_start) >> PAGE_SHIFT);
Michel Lespinasse38a76012012-10-08 16:31:50 -0700270 new_vma = copy_vma(&vma, new_addr, new_len, new_pgoff,
271 &need_rmap_locks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 if (!new_vma)
273 return -ENOMEM;
274
Michel Lespinasse38a76012012-10-08 16:31:50 -0700275 moved_len = move_page_tables(vma, old_addr, new_vma, new_addr, old_len,
276 need_rmap_locks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 if (moved_len < old_len) {
278 /*
279 * On error, move entries back from new area to old,
280 * which will succeed since page tables still there,
281 * and then proceed to unmap new area instead of old.
282 */
Michel Lespinasse38a76012012-10-08 16:31:50 -0700283 move_page_tables(new_vma, new_addr, vma, old_addr, moved_len,
284 true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 vma = new_vma;
286 old_len = new_len;
287 old_addr = new_addr;
288 new_addr = -ENOMEM;
Pavel Emelyanove4a0d3e2014-09-18 19:56:17 +0400289 } else if (vma->vm_file && vma->vm_file->f_op->mremap)
290 vma->vm_file->f_op->mremap(vma->vm_file, new_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291
292 /* Conceal VM_ACCOUNT so old reservation is not undone */
293 if (vm_flags & VM_ACCOUNT) {
294 vma->vm_flags &= ~VM_ACCOUNT;
295 excess = vma->vm_end - vma->vm_start - old_len;
296 if (old_addr > vma->vm_start &&
297 old_addr + old_len < vma->vm_end)
298 split = 1;
299 }
300
Kirill Korotaev71799062005-05-16 21:53:18 -0700301 /*
Hugh Dickins365e9c872005-10-29 18:16:18 -0700302 * If we failed to move page tables we still do total_vm increment
303 * since do_munmap() will decrement it by old_len == new_len.
304 *
305 * Since total_vm is about to be raised artificially high for a
306 * moment, we need to restore high watermark afterwards: if stats
307 * are taken meanwhile, total_vm and hiwater_vm appear too high.
308 * If this were a serious issue, we'd add a flag to do_munmap().
Kirill Korotaev71799062005-05-16 21:53:18 -0700309 */
Hugh Dickins365e9c872005-10-29 18:16:18 -0700310 hiwater_vm = mm->hiwater_vm;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -0700311 vm_stat_account(mm, vma->vm_flags, vma->vm_file, new_len>>PAGE_SHIFT);
Kirill Korotaev71799062005-05-16 21:53:18 -0700312
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 if (do_munmap(mm, old_addr, old_len) < 0) {
314 /* OOM: unable to split vma, just get accounts right */
315 vm_unacct_memory(excess >> PAGE_SHIFT);
316 excess = 0;
317 }
Hugh Dickins365e9c872005-10-29 18:16:18 -0700318 mm->hiwater_vm = hiwater_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319
320 /* Restore VM_ACCOUNT if one or two pieces of vma left */
321 if (excess) {
322 vma->vm_flags |= VM_ACCOUNT;
323 if (split)
324 vma->vm_next->vm_flags |= VM_ACCOUNT;
325 }
326
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 if (vm_flags & VM_LOCKED) {
328 mm->locked_vm += new_len >> PAGE_SHIFT;
Michel Lespinasse81909b82013-02-22 16:32:41 -0800329 *locked = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 }
331
332 return new_addr;
333}
334
Al Viro54f5de72009-11-24 07:17:46 -0500335static struct vm_area_struct *vma_to_resize(unsigned long addr,
336 unsigned long old_len, unsigned long new_len, unsigned long *p)
337{
338 struct mm_struct *mm = current->mm;
339 struct vm_area_struct *vma = find_vma(mm, addr);
340
341 if (!vma || vma->vm_start > addr)
342 goto Efault;
343
344 if (is_vm_hugetlb_page(vma))
345 goto Einval;
346
347 /* We can't remap across vm area boundaries */
348 if (old_len > vma->vm_end - addr)
349 goto Efault;
350
Linus Torvalds982134b2011-04-07 07:35:50 -0700351 /* Need to be careful about a growing mapping */
352 if (new_len > old_len) {
353 unsigned long pgoff;
354
355 if (vma->vm_flags & (VM_DONTEXPAND | VM_PFNMAP))
Al Viro54f5de72009-11-24 07:17:46 -0500356 goto Efault;
Linus Torvalds982134b2011-04-07 07:35:50 -0700357 pgoff = (addr - vma->vm_start) >> PAGE_SHIFT;
358 pgoff += vma->vm_pgoff;
359 if (pgoff + (new_len >> PAGE_SHIFT) < pgoff)
360 goto Einval;
Al Viro54f5de72009-11-24 07:17:46 -0500361 }
362
363 if (vma->vm_flags & VM_LOCKED) {
364 unsigned long locked, lock_limit;
365 locked = mm->locked_vm << PAGE_SHIFT;
Jiri Slaby59e99e52010-03-05 13:41:44 -0800366 lock_limit = rlimit(RLIMIT_MEMLOCK);
Al Viro54f5de72009-11-24 07:17:46 -0500367 locked += new_len - old_len;
368 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
369 goto Eagain;
370 }
371
372 if (!may_expand_vm(mm, (new_len - old_len) >> PAGE_SHIFT))
373 goto Enomem;
374
375 if (vma->vm_flags & VM_ACCOUNT) {
376 unsigned long charged = (new_len - old_len) >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +0000377 if (security_vm_enough_memory_mm(mm, charged))
Al Viro54f5de72009-11-24 07:17:46 -0500378 goto Efault;
379 *p = charged;
380 }
381
382 return vma;
383
384Efault: /* very odd choice for most of the cases, but... */
385 return ERR_PTR(-EFAULT);
386Einval:
387 return ERR_PTR(-EINVAL);
388Enomem:
389 return ERR_PTR(-ENOMEM);
390Eagain:
391 return ERR_PTR(-EAGAIN);
392}
393
Michel Lespinasse81909b82013-02-22 16:32:41 -0800394static unsigned long mremap_to(unsigned long addr, unsigned long old_len,
395 unsigned long new_addr, unsigned long new_len, bool *locked)
Al Viroecc1a892009-11-24 07:28:07 -0500396{
397 struct mm_struct *mm = current->mm;
398 struct vm_area_struct *vma;
399 unsigned long ret = -EINVAL;
400 unsigned long charged = 0;
Al Viro097eed12009-11-24 08:43:52 -0500401 unsigned long map_flags;
Al Viroecc1a892009-11-24 07:28:07 -0500402
403 if (new_addr & ~PAGE_MASK)
404 goto out;
405
406 if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len)
407 goto out;
408
409 /* Check if the location we're moving into overlaps the
410 * old location at all, and fail if it does.
411 */
412 if ((new_addr <= addr) && (new_addr+new_len) > addr)
413 goto out;
414
415 if ((addr <= new_addr) && (addr+old_len) > new_addr)
416 goto out;
417
Al Viroecc1a892009-11-24 07:28:07 -0500418 ret = do_munmap(mm, new_addr, new_len);
419 if (ret)
420 goto out;
421
422 if (old_len >= new_len) {
423 ret = do_munmap(mm, addr+new_len, old_len - new_len);
424 if (ret && old_len != new_len)
425 goto out;
426 old_len = new_len;
427 }
428
429 vma = vma_to_resize(addr, old_len, new_len, &charged);
430 if (IS_ERR(vma)) {
431 ret = PTR_ERR(vma);
432 goto out;
433 }
434
Al Viro097eed12009-11-24 08:43:52 -0500435 map_flags = MAP_FIXED;
436 if (vma->vm_flags & VM_MAYSHARE)
437 map_flags |= MAP_SHARED;
Al Viro9206de92009-12-03 15:23:11 -0500438
Al Viro097eed12009-11-24 08:43:52 -0500439 ret = get_unmapped_area(vma->vm_file, new_addr, new_len, vma->vm_pgoff +
440 ((addr - vma->vm_start) >> PAGE_SHIFT),
441 map_flags);
Al Viroecc1a892009-11-24 07:28:07 -0500442 if (ret & ~PAGE_MASK)
Al Viro097eed12009-11-24 08:43:52 -0500443 goto out1;
444
Michel Lespinasse81909b82013-02-22 16:32:41 -0800445 ret = move_vma(vma, addr, old_len, new_len, new_addr, locked);
Al Viro097eed12009-11-24 08:43:52 -0500446 if (!(ret & ~PAGE_MASK))
447 goto out;
448out1:
449 vm_unacct_memory(charged);
Al Viroecc1a892009-11-24 07:28:07 -0500450
451out:
452 return ret;
453}
454
Al Viro1a0ef852009-11-24 07:43:18 -0500455static int vma_expandable(struct vm_area_struct *vma, unsigned long delta)
456{
Al Virof106af42009-11-24 08:25:18 -0500457 unsigned long end = vma->vm_end + delta;
Al Viro9206de92009-12-03 15:23:11 -0500458 if (end < vma->vm_end) /* overflow */
Al Viro1a0ef852009-11-24 07:43:18 -0500459 return 0;
Al Viro9206de92009-12-03 15:23:11 -0500460 if (vma->vm_next && vma->vm_next->vm_start < end) /* intersection */
Al Virof106af42009-11-24 08:25:18 -0500461 return 0;
462 if (get_unmapped_area(NULL, vma->vm_start, end - vma->vm_start,
463 0, MAP_FIXED) & ~PAGE_MASK)
464 return 0;
Al Viro1a0ef852009-11-24 07:43:18 -0500465 return 1;
466}
467
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468/*
469 * Expand (or shrink) an existing mapping, potentially moving it at the
470 * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
471 *
472 * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
473 * This option implies MREMAP_MAYMOVE.
474 */
Al Viro63a81db2012-05-30 11:32:04 -0400475SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
476 unsigned long, new_len, unsigned long, flags,
477 unsigned long, new_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478{
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700479 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 struct vm_area_struct *vma;
481 unsigned long ret = -EINVAL;
482 unsigned long charged = 0;
Michel Lespinasse81909b82013-02-22 16:32:41 -0800483 bool locked = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
485 if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE))
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700486 return ret;
487
488 if (flags & MREMAP_FIXED && !(flags & MREMAP_MAYMOVE))
489 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490
491 if (addr & ~PAGE_MASK)
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700492 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
494 old_len = PAGE_ALIGN(old_len);
495 new_len = PAGE_ALIGN(new_len);
496
497 /*
498 * We allow a zero old-len as a special case
499 * for DOS-emu "duplicate shm area" thing. But
500 * a zero new-len is nonsensical.
501 */
502 if (!new_len)
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700503 return ret;
504
505 down_write(&current->mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 if (flags & MREMAP_FIXED) {
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700508 ret = mremap_to(addr, old_len, new_addr, new_len,
509 &locked);
Al Viroecc1a892009-11-24 07:28:07 -0500510 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 }
512
513 /*
514 * Always allow a shrinking remap: that just unmaps
515 * the unnecessary pages..
516 * do_munmap does all the needed commit accounting
517 */
518 if (old_len >= new_len) {
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700519 ret = do_munmap(mm, addr+new_len, old_len - new_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 if (ret && old_len != new_len)
521 goto out;
522 ret = addr;
Al Viroecc1a892009-11-24 07:28:07 -0500523 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 }
525
526 /*
Al Viroecc1a892009-11-24 07:28:07 -0500527 * Ok, we need to grow..
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 */
Al Viro54f5de72009-11-24 07:17:46 -0500529 vma = vma_to_resize(addr, old_len, new_len, &charged);
530 if (IS_ERR(vma)) {
531 ret = PTR_ERR(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 }
534
535 /* old_len exactly to the end of the area..
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 */
Al Viroecc1a892009-11-24 07:28:07 -0500537 if (old_len == vma->vm_end - addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 /* can we just expand the current mapping? */
Al Viro1a0ef852009-11-24 07:43:18 -0500539 if (vma_expandable(vma, new_len - old_len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 int pages = (new_len - old_len) >> PAGE_SHIFT;
541
Rik van Riel5beb4932010-03-05 13:42:07 -0800542 if (vma_adjust(vma, vma->vm_start, addr + new_len,
543 vma->vm_pgoff, NULL)) {
544 ret = -ENOMEM;
545 goto out;
546 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700548 vm_stat_account(mm, vma->vm_flags, vma->vm_file, pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 if (vma->vm_flags & VM_LOCKED) {
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700550 mm->locked_vm += pages;
Michel Lespinasse81909b82013-02-22 16:32:41 -0800551 locked = true;
552 new_addr = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 }
554 ret = addr;
555 goto out;
556 }
557 }
558
559 /*
560 * We weren't able to just expand or shrink the area,
561 * we need to create a new one and move it..
562 */
563 ret = -ENOMEM;
564 if (flags & MREMAP_MAYMOVE) {
Al Viroecc1a892009-11-24 07:28:07 -0500565 unsigned long map_flags = 0;
566 if (vma->vm_flags & VM_MAYSHARE)
567 map_flags |= MAP_SHARED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568
Al Viroecc1a892009-11-24 07:28:07 -0500569 new_addr = get_unmapped_area(vma->vm_file, 0, new_len,
Al Viro93587412009-11-24 08:45:24 -0500570 vma->vm_pgoff +
571 ((addr - vma->vm_start) >> PAGE_SHIFT),
572 map_flags);
Al Viroecc1a892009-11-24 07:28:07 -0500573 if (new_addr & ~PAGE_MASK) {
574 ret = new_addr;
575 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 }
Al Viroecc1a892009-11-24 07:28:07 -0500577
Michel Lespinasse81909b82013-02-22 16:32:41 -0800578 ret = move_vma(vma, addr, old_len, new_len, new_addr, &locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 }
580out:
581 if (ret & ~PAGE_MASK)
582 vm_unacct_memory(charged);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 up_write(&current->mm->mmap_sem);
Michel Lespinasse81909b82013-02-22 16:32:41 -0800584 if (locked && new_len > old_len)
585 mm_populate(new_addr + old_len, new_len - old_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 return ret;
587}