blob: 30d7d2482eea1d04cf72b03bcd54d1d293f666e4 [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>
Paul McQuade2581d202014-10-09 15:29:01 -070023#include <linux/uaccess.h>
Laurent Dufour4abad2c2015-06-24 16:56:19 -070024#include <linux/mm-arch-hooks.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
Hugh Dickins1d069b72016-05-19 17:12:57 -070073static 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
81static 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
Cyrill Gorcunov6dec97d2013-08-27 12:37:18 +040089static 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);
Cyrill Gorcunov6dec97d2013-08-27 12:37:18 +0400100#endif
101 return pte;
102}
103
Hugh Dickins7be7a542005-10-29 18:16:00 -0700104static 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,
Aaron Lu5d190422016-11-10 17:16:33 +0800107 unsigned long new_addr, bool need_rmap_locks, bool *need_flush)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 struct mm_struct *mm = vma->vm_mm;
Hugh Dickins7be7a542005-10-29 18:16:00 -0700110 pte_t *old_pte, *new_pte, pte;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700111 spinlock_t *old_ptl, *new_ptl;
Aaron Lu5d190422016-11-10 17:16:33 +0800112 bool force_flush = false;
113 unsigned long len = old_end - old_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114
Michel Lespinasse38a76012012-10-08 16:31:50 -0700115 /*
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800116 * When need_rmap_locks is true, we take the i_mmap_rwsem and anon_vma
Michel Lespinasse38a76012012-10-08 16:31:50 -0700117 * 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 */
Hugh Dickins1d069b72016-05-19 17:12:57 -0700133 if (need_rmap_locks)
134 take_rmap_locks(vma);
Hugh Dickins7be7a542005-10-29 18:16:00 -0700135
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700136 /*
137 * We don't have to worry about the ordering of src and dst
138 * pte locks because exclusive mmap_sem prevents deadlock.
139 */
Hugh Dickinsc74df322005-10-29 18:16:23 -0700140 old_pte = pte_offset_map_lock(mm, old_pmd, old_addr, &old_ptl);
Peter Zijlstraece0e2b2010-10-26 14:21:52 -0700141 new_pte = pte_offset_map(new_pmd, new_addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700142 new_ptl = pte_lockptr(mm, new_pmd);
143 if (new_ptl != old_ptl)
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700144 spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700145 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
Hugh Dickins7be7a542005-10-29 18:16:00 -0700147 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;
Aaron Lu5d190422016-11-10 17:16:33 +0800151
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700152 pte = ptep_get_and_clear(mm, old_addr, old_pte);
Aaron Lua2ce2662016-11-29 13:27:31 +0800153 /*
154 * If we are remapping a dirty PTE, make sure
155 * to flush TLB before we drop the PTL for the
156 * old PTE or we may race with page_mkclean().
157 *
158 * This check has to be done after we removed the
159 * old PTE from page tables or another thread may
160 * dirty it after the check and before the removal.
161 */
162 if (pte_present(pte) && pte_dirty(pte))
163 force_flush = true;
Hugh Dickins7be7a542005-10-29 18:16:00 -0700164 pte = move_pte(pte, new_vma->vm_page_prot, old_addr, new_addr);
Cyrill Gorcunov6dec97d2013-08-27 12:37:18 +0400165 pte = move_soft_dirty_pte(pte);
166 set_pte_at(mm, new_addr, new_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 }
Hugh Dickins7be7a542005-10-29 18:16:00 -0700168
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700169 arch_leave_lazy_mmu_mode();
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700170 if (new_ptl != old_ptl)
171 spin_unlock(new_ptl);
Peter Zijlstraece0e2b2010-10-26 14:21:52 -0700172 pte_unmap(new_pte - 1);
Aaron Lu5d190422016-11-10 17:16:33 +0800173 if (force_flush)
174 flush_tlb_range(vma, old_end - len, old_end);
175 else
176 *need_flush = true;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700177 pte_unmap_unlock(old_pte - 1, old_ptl);
Hugh Dickins1d069b72016-05-19 17:12:57 -0700178 if (need_rmap_locks)
179 drop_rmap_locks(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180}
181
Hugh Dickins7be7a542005-10-29 18:16:00 -0700182#define LATENCY_LIMIT (64 * PAGE_SIZE)
183
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700184unsigned long move_page_tables(struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 unsigned long old_addr, struct vm_area_struct *new_vma,
Michel Lespinasse38a76012012-10-08 16:31:50 -0700186 unsigned long new_addr, unsigned long len,
187 bool need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188{
Hugh Dickins7be7a542005-10-29 18:16:00 -0700189 unsigned long extent, next, old_end;
190 pmd_t *old_pmd, *new_pmd;
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700191 bool need_flush = false;
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700192 unsigned long mmun_start; /* For mmu_notifiers */
193 unsigned long mmun_end; /* For mmu_notifiers */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
Hugh Dickins7be7a542005-10-29 18:16:00 -0700195 old_end = old_addr + len;
196 flush_cache_range(vma, old_addr, old_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700198 mmun_start = old_addr;
199 mmun_end = old_end;
200 mmu_notifier_invalidate_range_start(vma->vm_mm, mmun_start, mmun_end);
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700201
Hugh Dickins7be7a542005-10-29 18:16:00 -0700202 for (; old_addr < old_end; old_addr += extent, new_addr += extent) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 cond_resched();
Hugh Dickins7be7a542005-10-29 18:16:00 -0700204 next = (old_addr + PMD_SIZE) & PMD_MASK;
Andrea Arcangeliebed4842011-10-31 17:08:22 -0700205 /* even if next overflowed, extent below will be ok */
Hugh Dickins7be7a542005-10-29 18:16:00 -0700206 extent = next - old_addr;
Andrea Arcangeliebed4842011-10-31 17:08:22 -0700207 if (extent > old_end - old_addr)
208 extent = old_end - old_addr;
Hugh Dickins7be7a542005-10-29 18:16:00 -0700209 old_pmd = get_old_pmd(vma->vm_mm, old_addr);
210 if (!old_pmd)
211 continue;
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800212 new_pmd = alloc_new_pmd(vma->vm_mm, vma, new_addr);
Hugh Dickins7be7a542005-10-29 18:16:00 -0700213 if (!new_pmd)
214 break;
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700215 if (pmd_trans_huge(*old_pmd)) {
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700216 if (extent == HPAGE_PMD_SIZE) {
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800217 bool moved;
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700218 /* See comment in move_ptes() */
219 if (need_rmap_locks)
Hugh Dickins1d069b72016-05-19 17:12:57 -0700220 take_rmap_locks(vma);
Hugh Dickinsbf8616d2016-05-19 17:12:54 -0700221 moved = move_huge_pmd(vma, old_addr, new_addr,
Aaron Lu5d190422016-11-10 17:16:33 +0800222 old_end, old_pmd, new_pmd,
223 &need_flush);
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700224 if (need_rmap_locks)
Hugh Dickins1d069b72016-05-19 17:12:57 -0700225 drop_rmap_locks(vma);
Aaron Lu5d190422016-11-10 17:16:33 +0800226 if (moved)
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800227 continue;
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700228 }
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800229 split_huge_pmd(vma, old_pmd, old_addr);
Naoya Horiguchi337d9ab2016-07-26 15:24:03 -0700230 if (pmd_trans_unstable(old_pmd))
Kirill A. Shutemov6b9116a2016-02-11 16:13:03 -0800231 continue;
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700232 }
Kirill A. Shutemov3ed3a4f2016-03-17 14:19:11 -0700233 if (pte_alloc(new_vma->vm_mm, new_pmd, new_addr))
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700234 break;
Hugh Dickins7be7a542005-10-29 18:16:00 -0700235 next = (new_addr + PMD_SIZE) & PMD_MASK;
236 if (extent > next - new_addr)
237 extent = next - new_addr;
238 if (extent > LATENCY_LIMIT)
239 extent = LATENCY_LIMIT;
Aaron Lu5d190422016-11-10 17:16:33 +0800240 move_ptes(vma, old_pmd, old_addr, old_addr + extent, new_vma,
241 new_pmd, new_addr, need_rmap_locks, &need_flush);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 }
Aaron Lu5d190422016-11-10 17:16:33 +0800243 if (need_flush)
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700244 flush_tlb_range(vma, old_end-len, old_addr);
245
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700246 mmu_notifier_invalidate_range_end(vma->vm_mm, mmun_start, mmun_end);
Hugh Dickins7be7a542005-10-29 18:16:00 -0700247
248 return len + old_addr - old_end; /* how much done */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249}
250
251static unsigned long move_vma(struct vm_area_struct *vma,
252 unsigned long old_addr, unsigned long old_len,
Michel Lespinasse81909b82013-02-22 16:32:41 -0800253 unsigned long new_len, unsigned long new_addr, bool *locked)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254{
255 struct mm_struct *mm = vma->vm_mm;
256 struct vm_area_struct *new_vma;
257 unsigned long vm_flags = vma->vm_flags;
258 unsigned long new_pgoff;
259 unsigned long moved_len;
260 unsigned long excess = 0;
Hugh Dickins365e9c872005-10-29 18:16:18 -0700261 unsigned long hiwater_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 int split = 0;
Hugh Dickins7103ad32009-09-21 17:02:28 -0700263 int err;
Michel Lespinasse38a76012012-10-08 16:31:50 -0700264 bool need_rmap_locks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
266 /*
267 * We'd prefer to avoid failure later on in do_munmap:
268 * which may split one vma into three before unmapping.
269 */
270 if (mm->map_count >= sysctl_max_map_count - 3)
271 return -ENOMEM;
272
Hugh Dickins1ff82992009-09-21 17:02:05 -0700273 /*
274 * Advise KSM to break any KSM pages in the area to be moved:
275 * it would be confusing if they were to turn up at the new
276 * location, where they happen to coincide with different KSM
277 * pages recently unmapped. But leave vma->vm_flags as it was,
278 * so KSM can come around to merge on vma and new_vma afterwards.
279 */
Hugh Dickins7103ad32009-09-21 17:02:28 -0700280 err = ksm_madvise(vma, old_addr, old_addr + old_len,
281 MADV_UNMERGEABLE, &vm_flags);
282 if (err)
283 return err;
Hugh Dickins1ff82992009-09-21 17:02:05 -0700284
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 new_pgoff = vma->vm_pgoff + ((old_addr - vma->vm_start) >> PAGE_SHIFT);
Michel Lespinasse38a76012012-10-08 16:31:50 -0700286 new_vma = copy_vma(&vma, new_addr, new_len, new_pgoff,
287 &need_rmap_locks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 if (!new_vma)
289 return -ENOMEM;
290
Michel Lespinasse38a76012012-10-08 16:31:50 -0700291 moved_len = move_page_tables(vma, old_addr, new_vma, new_addr, old_len,
292 need_rmap_locks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 if (moved_len < old_len) {
Oleg Nesterovdf1eab32015-09-04 15:48:01 -0700294 err = -ENOMEM;
Oleg Nesterov5477e702015-09-04 15:48:04 -0700295 } else if (vma->vm_ops && vma->vm_ops->mremap) {
296 err = vma->vm_ops->mremap(new_vma);
Oleg Nesterovdf1eab32015-09-04 15:48:01 -0700297 }
298
299 if (unlikely(err)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 /*
301 * On error, move entries back from new area to old,
302 * which will succeed since page tables still there,
303 * and then proceed to unmap new area instead of old.
304 */
Michel Lespinasse38a76012012-10-08 16:31:50 -0700305 move_page_tables(new_vma, new_addr, vma, old_addr, moved_len,
306 true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 vma = new_vma;
308 old_len = new_len;
309 old_addr = new_addr;
Oleg Nesterovdf1eab32015-09-04 15:48:01 -0700310 new_addr = err;
Laurent Dufour4abad2c2015-06-24 16:56:19 -0700311 } else {
Laurent Dufour4abad2c2015-06-24 16:56:19 -0700312 arch_remap(mm, old_addr, old_addr + old_len,
313 new_addr, new_addr + new_len);
Al Virob2edffd2015-04-06 17:48:54 -0400314 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315
316 /* Conceal VM_ACCOUNT so old reservation is not undone */
317 if (vm_flags & VM_ACCOUNT) {
318 vma->vm_flags &= ~VM_ACCOUNT;
319 excess = vma->vm_end - vma->vm_start - old_len;
320 if (old_addr > vma->vm_start &&
321 old_addr + old_len < vma->vm_end)
322 split = 1;
323 }
324
Kirill Korotaev71799062005-05-16 21:53:18 -0700325 /*
Hugh Dickins365e9c872005-10-29 18:16:18 -0700326 * If we failed to move page tables we still do total_vm increment
327 * since do_munmap() will decrement it by old_len == new_len.
328 *
329 * Since total_vm is about to be raised artificially high for a
330 * moment, we need to restore high watermark afterwards: if stats
331 * are taken meanwhile, total_vm and hiwater_vm appear too high.
332 * If this were a serious issue, we'd add a flag to do_munmap().
Kirill Korotaev71799062005-05-16 21:53:18 -0700333 */
Hugh Dickins365e9c872005-10-29 18:16:18 -0700334 hiwater_vm = mm->hiwater_vm;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -0800335 vm_stat_account(mm, vma->vm_flags, new_len >> PAGE_SHIFT);
Kirill Korotaev71799062005-05-16 21:53:18 -0700336
Toshi Kanid9fe4fa2015-12-22 17:54:23 -0700337 /* Tell pfnmap has moved from this vma */
338 if (unlikely(vma->vm_flags & VM_PFNMAP))
339 untrack_pfn_moved(vma);
340
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 if (do_munmap(mm, old_addr, old_len) < 0) {
342 /* OOM: unable to split vma, just get accounts right */
343 vm_unacct_memory(excess >> PAGE_SHIFT);
344 excess = 0;
345 }
Hugh Dickins365e9c872005-10-29 18:16:18 -0700346 mm->hiwater_vm = hiwater_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347
348 /* Restore VM_ACCOUNT if one or two pieces of vma left */
349 if (excess) {
350 vma->vm_flags |= VM_ACCOUNT;
351 if (split)
352 vma->vm_next->vm_flags |= VM_ACCOUNT;
353 }
354
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 if (vm_flags & VM_LOCKED) {
356 mm->locked_vm += new_len >> PAGE_SHIFT;
Michel Lespinasse81909b82013-02-22 16:32:41 -0800357 *locked = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 }
359
360 return new_addr;
361}
362
Al Viro54f5de72009-11-24 07:17:46 -0500363static struct vm_area_struct *vma_to_resize(unsigned long addr,
364 unsigned long old_len, unsigned long new_len, unsigned long *p)
365{
366 struct mm_struct *mm = current->mm;
367 struct vm_area_struct *vma = find_vma(mm, addr);
Oleg Nesterov1d391682015-09-04 15:48:10 -0700368 unsigned long pgoff;
Al Viro54f5de72009-11-24 07:17:46 -0500369
370 if (!vma || vma->vm_start > addr)
Derek6cd576132015-04-15 16:14:02 -0700371 return ERR_PTR(-EFAULT);
Al Viro54f5de72009-11-24 07:17:46 -0500372
373 if (is_vm_hugetlb_page(vma))
Derek6cd576132015-04-15 16:14:02 -0700374 return ERR_PTR(-EINVAL);
Al Viro54f5de72009-11-24 07:17:46 -0500375
376 /* We can't remap across vm area boundaries */
377 if (old_len > vma->vm_end - addr)
Derek6cd576132015-04-15 16:14:02 -0700378 return ERR_PTR(-EFAULT);
Al Viro54f5de72009-11-24 07:17:46 -0500379
Oleg Nesterov1d391682015-09-04 15:48:10 -0700380 if (new_len == old_len)
381 return vma;
Linus Torvalds982134b2011-04-07 07:35:50 -0700382
Oleg Nesterov1d391682015-09-04 15:48:10 -0700383 /* Need to be careful about a growing mapping */
384 pgoff = (addr - vma->vm_start) >> PAGE_SHIFT;
385 pgoff += vma->vm_pgoff;
386 if (pgoff + (new_len >> PAGE_SHIFT) < pgoff)
387 return ERR_PTR(-EINVAL);
388
389 if (vma->vm_flags & (VM_DONTEXPAND | VM_PFNMAP))
390 return ERR_PTR(-EFAULT);
Al Viro54f5de72009-11-24 07:17:46 -0500391
392 if (vma->vm_flags & VM_LOCKED) {
393 unsigned long locked, lock_limit;
394 locked = mm->locked_vm << PAGE_SHIFT;
Jiri Slaby59e99e52010-03-05 13:41:44 -0800395 lock_limit = rlimit(RLIMIT_MEMLOCK);
Al Viro54f5de72009-11-24 07:17:46 -0500396 locked += new_len - old_len;
397 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
Derek6cd576132015-04-15 16:14:02 -0700398 return ERR_PTR(-EAGAIN);
Al Viro54f5de72009-11-24 07:17:46 -0500399 }
400
Konstantin Khlebnikov84638332016-01-14 15:22:07 -0800401 if (!may_expand_vm(mm, vma->vm_flags,
402 (new_len - old_len) >> PAGE_SHIFT))
Derek6cd576132015-04-15 16:14:02 -0700403 return ERR_PTR(-ENOMEM);
Al Viro54f5de72009-11-24 07:17:46 -0500404
405 if (vma->vm_flags & VM_ACCOUNT) {
406 unsigned long charged = (new_len - old_len) >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +0000407 if (security_vm_enough_memory_mm(mm, charged))
Derek6cd576132015-04-15 16:14:02 -0700408 return ERR_PTR(-ENOMEM);
Al Viro54f5de72009-11-24 07:17:46 -0500409 *p = charged;
410 }
411
412 return vma;
Al Viro54f5de72009-11-24 07:17:46 -0500413}
414
Michel Lespinasse81909b82013-02-22 16:32:41 -0800415static unsigned long mremap_to(unsigned long addr, unsigned long old_len,
416 unsigned long new_addr, unsigned long new_len, bool *locked)
Al Viroecc1a892009-11-24 07:28:07 -0500417{
418 struct mm_struct *mm = current->mm;
419 struct vm_area_struct *vma;
420 unsigned long ret = -EINVAL;
421 unsigned long charged = 0;
Al Viro097eed12009-11-24 08:43:52 -0500422 unsigned long map_flags;
Al Viroecc1a892009-11-24 07:28:07 -0500423
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800424 if (offset_in_page(new_addr))
Al Viroecc1a892009-11-24 07:28:07 -0500425 goto out;
426
427 if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len)
428 goto out;
429
Oleg Nesterov99432422015-09-04 15:48:13 -0700430 /* Ensure the old/new locations do not overlap */
431 if (addr + old_len > new_addr && new_addr + new_len > addr)
Al Viroecc1a892009-11-24 07:28:07 -0500432 goto out;
433
Al Viroecc1a892009-11-24 07:28:07 -0500434 ret = do_munmap(mm, new_addr, new_len);
435 if (ret)
436 goto out;
437
438 if (old_len >= new_len) {
439 ret = do_munmap(mm, addr+new_len, old_len - new_len);
440 if (ret && old_len != new_len)
441 goto out;
442 old_len = new_len;
443 }
444
445 vma = vma_to_resize(addr, old_len, new_len, &charged);
446 if (IS_ERR(vma)) {
447 ret = PTR_ERR(vma);
448 goto out;
449 }
450
Al Viro097eed12009-11-24 08:43:52 -0500451 map_flags = MAP_FIXED;
452 if (vma->vm_flags & VM_MAYSHARE)
453 map_flags |= MAP_SHARED;
Al Viro9206de92009-12-03 15:23:11 -0500454
Al Viro097eed12009-11-24 08:43:52 -0500455 ret = get_unmapped_area(vma->vm_file, new_addr, new_len, vma->vm_pgoff +
456 ((addr - vma->vm_start) >> PAGE_SHIFT),
457 map_flags);
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800458 if (offset_in_page(ret))
Al Viro097eed12009-11-24 08:43:52 -0500459 goto out1;
460
Michel Lespinasse81909b82013-02-22 16:32:41 -0800461 ret = move_vma(vma, addr, old_len, new_len, new_addr, locked);
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800462 if (!(offset_in_page(ret)))
Al Viro097eed12009-11-24 08:43:52 -0500463 goto out;
464out1:
465 vm_unacct_memory(charged);
Al Viroecc1a892009-11-24 07:28:07 -0500466
467out:
468 return ret;
469}
470
Al Viro1a0ef852009-11-24 07:43:18 -0500471static int vma_expandable(struct vm_area_struct *vma, unsigned long delta)
472{
Al Virof106af42009-11-24 08:25:18 -0500473 unsigned long end = vma->vm_end + delta;
Al Viro9206de92009-12-03 15:23:11 -0500474 if (end < vma->vm_end) /* overflow */
Al Viro1a0ef852009-11-24 07:43:18 -0500475 return 0;
Al Viro9206de92009-12-03 15:23:11 -0500476 if (vma->vm_next && vma->vm_next->vm_start < end) /* intersection */
Al Virof106af42009-11-24 08:25:18 -0500477 return 0;
478 if (get_unmapped_area(NULL, vma->vm_start, end - vma->vm_start,
479 0, MAP_FIXED) & ~PAGE_MASK)
480 return 0;
Al Viro1a0ef852009-11-24 07:43:18 -0500481 return 1;
482}
483
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484/*
485 * Expand (or shrink) an existing mapping, potentially moving it at the
486 * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
487 *
488 * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
489 * This option implies MREMAP_MAYMOVE.
490 */
Al Viro63a81db2012-05-30 11:32:04 -0400491SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
492 unsigned long, new_len, unsigned long, flags,
493 unsigned long, new_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494{
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700495 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 struct vm_area_struct *vma;
497 unsigned long ret = -EINVAL;
498 unsigned long charged = 0;
Michel Lespinasse81909b82013-02-22 16:32:41 -0800499 bool locked = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500
501 if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE))
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700502 return ret;
503
504 if (flags & MREMAP_FIXED && !(flags & MREMAP_MAYMOVE))
505 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800507 if (offset_in_page(addr))
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700508 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
510 old_len = PAGE_ALIGN(old_len);
511 new_len = PAGE_ALIGN(new_len);
512
513 /*
514 * We allow a zero old-len as a special case
515 * for DOS-emu "duplicate shm area" thing. But
516 * a zero new-len is nonsensical.
517 */
518 if (!new_len)
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700519 return ret;
520
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700521 if (down_write_killable(&current->mm->mmap_sem))
522 return -EINTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 if (flags & MREMAP_FIXED) {
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700525 ret = mremap_to(addr, old_len, new_addr, new_len,
526 &locked);
Al Viroecc1a892009-11-24 07:28:07 -0500527 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 }
529
530 /*
531 * Always allow a shrinking remap: that just unmaps
532 * the unnecessary pages..
533 * do_munmap does all the needed commit accounting
534 */
535 if (old_len >= new_len) {
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700536 ret = do_munmap(mm, addr+new_len, old_len - new_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 if (ret && old_len != new_len)
538 goto out;
539 ret = addr;
Al Viroecc1a892009-11-24 07:28:07 -0500540 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 }
542
543 /*
Al Viroecc1a892009-11-24 07:28:07 -0500544 * Ok, we need to grow..
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 */
Al Viro54f5de72009-11-24 07:17:46 -0500546 vma = vma_to_resize(addr, old_len, new_len, &charged);
547 if (IS_ERR(vma)) {
548 ret = PTR_ERR(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 }
551
552 /* old_len exactly to the end of the area..
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 */
Al Viroecc1a892009-11-24 07:28:07 -0500554 if (old_len == vma->vm_end - addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 /* can we just expand the current mapping? */
Al Viro1a0ef852009-11-24 07:43:18 -0500556 if (vma_expandable(vma, new_len - old_len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 int pages = (new_len - old_len) >> PAGE_SHIFT;
558
Rik van Riel5beb4932010-03-05 13:42:07 -0800559 if (vma_adjust(vma, vma->vm_start, addr + new_len,
560 vma->vm_pgoff, NULL)) {
561 ret = -ENOMEM;
562 goto out;
563 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564
Konstantin Khlebnikov84638332016-01-14 15:22:07 -0800565 vm_stat_account(mm, vma->vm_flags, pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 if (vma->vm_flags & VM_LOCKED) {
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700567 mm->locked_vm += pages;
Michel Lespinasse81909b82013-02-22 16:32:41 -0800568 locked = true;
569 new_addr = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 }
571 ret = addr;
572 goto out;
573 }
574 }
575
576 /*
577 * We weren't able to just expand or shrink the area,
578 * we need to create a new one and move it..
579 */
580 ret = -ENOMEM;
581 if (flags & MREMAP_MAYMOVE) {
Al Viroecc1a892009-11-24 07:28:07 -0500582 unsigned long map_flags = 0;
583 if (vma->vm_flags & VM_MAYSHARE)
584 map_flags |= MAP_SHARED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
Al Viroecc1a892009-11-24 07:28:07 -0500586 new_addr = get_unmapped_area(vma->vm_file, 0, new_len,
Al Viro93587412009-11-24 08:45:24 -0500587 vma->vm_pgoff +
588 ((addr - vma->vm_start) >> PAGE_SHIFT),
589 map_flags);
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800590 if (offset_in_page(new_addr)) {
Al Viroecc1a892009-11-24 07:28:07 -0500591 ret = new_addr;
592 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 }
Al Viroecc1a892009-11-24 07:28:07 -0500594
Michel Lespinasse81909b82013-02-22 16:32:41 -0800595 ret = move_vma(vma, addr, old_len, new_len, new_addr, &locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 }
597out:
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800598 if (offset_in_page(ret)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 vm_unacct_memory(charged);
Oleg Nesterovd456fb92015-09-04 15:48:07 -0700600 locked = 0;
601 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 up_write(&current->mm->mmap_sem);
Michel Lespinasse81909b82013-02-22 16:32:41 -0800603 if (locked && new_len > old_len)
604 mm_populate(new_addr + old_len, new_len - old_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 return ret;
606}