blob: 15976716dd40dcece73b3a985e82f292776baf6b [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);
Mel Gorman5a1eef72017-08-02 13:31:52 -0700145 flush_tlb_batched_pending(vma->vm_mm);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700146 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147
Hugh Dickins7be7a542005-10-29 18:16:00 -0700148 for (; old_addr < old_end; old_pte++, old_addr += PAGE_SIZE,
149 new_pte++, new_addr += PAGE_SIZE) {
150 if (pte_none(*old_pte))
151 continue;
Aaron Lu5d190422016-11-10 17:16:33 +0800152
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700153 pte = ptep_get_and_clear(mm, old_addr, old_pte);
Aaron Lua2ce2662016-11-29 13:27:31 +0800154 /*
155 * If we are remapping a dirty PTE, make sure
156 * to flush TLB before we drop the PTL for the
157 * old PTE or we may race with page_mkclean().
158 *
159 * This check has to be done after we removed the
160 * old PTE from page tables or another thread may
161 * dirty it after the check and before the removal.
162 */
163 if (pte_present(pte) && pte_dirty(pte))
164 force_flush = true;
Hugh Dickins7be7a542005-10-29 18:16:00 -0700165 pte = move_pte(pte, new_vma->vm_page_prot, old_addr, new_addr);
Cyrill Gorcunov6dec97d2013-08-27 12:37:18 +0400166 pte = move_soft_dirty_pte(pte);
167 set_pte_at(mm, new_addr, new_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 }
Hugh Dickins7be7a542005-10-29 18:16:00 -0700169
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700170 arch_leave_lazy_mmu_mode();
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700171 if (new_ptl != old_ptl)
172 spin_unlock(new_ptl);
Peter Zijlstraece0e2b2010-10-26 14:21:52 -0700173 pte_unmap(new_pte - 1);
Aaron Lu5d190422016-11-10 17:16:33 +0800174 if (force_flush)
175 flush_tlb_range(vma, old_end - len, old_end);
176 else
177 *need_flush = true;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700178 pte_unmap_unlock(old_pte - 1, old_ptl);
Hugh Dickins1d069b72016-05-19 17:12:57 -0700179 if (need_rmap_locks)
180 drop_rmap_locks(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181}
182
Hugh Dickins7be7a542005-10-29 18:16:00 -0700183#define LATENCY_LIMIT (64 * PAGE_SIZE)
184
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700185unsigned long move_page_tables(struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 unsigned long old_addr, struct vm_area_struct *new_vma,
Michel Lespinasse38a76012012-10-08 16:31:50 -0700187 unsigned long new_addr, unsigned long len,
188 bool need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189{
Hugh Dickins7be7a542005-10-29 18:16:00 -0700190 unsigned long extent, next, old_end;
191 pmd_t *old_pmd, *new_pmd;
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700192 bool need_flush = false;
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700193 unsigned long mmun_start; /* For mmu_notifiers */
194 unsigned long mmun_end; /* For mmu_notifiers */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
Hugh Dickins7be7a542005-10-29 18:16:00 -0700196 old_end = old_addr + len;
197 flush_cache_range(vma, old_addr, old_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700199 mmun_start = old_addr;
200 mmun_end = old_end;
201 mmu_notifier_invalidate_range_start(vma->vm_mm, mmun_start, mmun_end);
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700202
Hugh Dickins7be7a542005-10-29 18:16:00 -0700203 for (; old_addr < old_end; old_addr += extent, new_addr += extent) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 cond_resched();
Hugh Dickins7be7a542005-10-29 18:16:00 -0700205 next = (old_addr + PMD_SIZE) & PMD_MASK;
Andrea Arcangeliebed4842011-10-31 17:08:22 -0700206 /* even if next overflowed, extent below will be ok */
Hugh Dickins7be7a542005-10-29 18:16:00 -0700207 extent = next - old_addr;
Andrea Arcangeliebed4842011-10-31 17:08:22 -0700208 if (extent > old_end - old_addr)
209 extent = old_end - old_addr;
Hugh Dickins7be7a542005-10-29 18:16:00 -0700210 old_pmd = get_old_pmd(vma->vm_mm, old_addr);
211 if (!old_pmd)
212 continue;
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800213 new_pmd = alloc_new_pmd(vma->vm_mm, vma, new_addr);
Hugh Dickins7be7a542005-10-29 18:16:00 -0700214 if (!new_pmd)
215 break;
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700216 if (pmd_trans_huge(*old_pmd)) {
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700217 if (extent == HPAGE_PMD_SIZE) {
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800218 bool moved;
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700219 /* See comment in move_ptes() */
220 if (need_rmap_locks)
Hugh Dickins1d069b72016-05-19 17:12:57 -0700221 take_rmap_locks(vma);
Hugh Dickinsbf8616d2016-05-19 17:12:54 -0700222 moved = move_huge_pmd(vma, old_addr, new_addr,
Aaron Lu5d190422016-11-10 17:16:33 +0800223 old_end, old_pmd, new_pmd,
224 &need_flush);
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700225 if (need_rmap_locks)
Hugh Dickins1d069b72016-05-19 17:12:57 -0700226 drop_rmap_locks(vma);
Aaron Lu5d190422016-11-10 17:16:33 +0800227 if (moved)
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800228 continue;
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700229 }
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800230 split_huge_pmd(vma, old_pmd, old_addr);
Naoya Horiguchi337d9ab2016-07-26 15:24:03 -0700231 if (pmd_trans_unstable(old_pmd))
Kirill A. Shutemov6b9116a2016-02-11 16:13:03 -0800232 continue;
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700233 }
Kirill A. Shutemov3ed3a4f2016-03-17 14:19:11 -0700234 if (pte_alloc(new_vma->vm_mm, new_pmd, new_addr))
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700235 break;
Hugh Dickins7be7a542005-10-29 18:16:00 -0700236 next = (new_addr + PMD_SIZE) & PMD_MASK;
237 if (extent > next - new_addr)
238 extent = next - new_addr;
239 if (extent > LATENCY_LIMIT)
240 extent = LATENCY_LIMIT;
Aaron Lu5d190422016-11-10 17:16:33 +0800241 move_ptes(vma, old_pmd, old_addr, old_addr + extent, new_vma,
242 new_pmd, new_addr, need_rmap_locks, &need_flush);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 }
Aaron Lu5d190422016-11-10 17:16:33 +0800244 if (need_flush)
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700245 flush_tlb_range(vma, old_end-len, old_addr);
246
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700247 mmu_notifier_invalidate_range_end(vma->vm_mm, mmun_start, mmun_end);
Hugh Dickins7be7a542005-10-29 18:16:00 -0700248
249 return len + old_addr - old_end; /* how much done */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250}
251
252static unsigned long move_vma(struct vm_area_struct *vma,
253 unsigned long old_addr, unsigned long old_len,
Michel Lespinasse81909b82013-02-22 16:32:41 -0800254 unsigned long new_len, unsigned long new_addr, bool *locked)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255{
256 struct mm_struct *mm = vma->vm_mm;
257 struct vm_area_struct *new_vma;
258 unsigned long vm_flags = vma->vm_flags;
259 unsigned long new_pgoff;
260 unsigned long moved_len;
261 unsigned long excess = 0;
Hugh Dickins365e9c872005-10-29 18:16:18 -0700262 unsigned long hiwater_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 int split = 0;
Hugh Dickins7103ad32009-09-21 17:02:28 -0700264 int err;
Michel Lespinasse38a76012012-10-08 16:31:50 -0700265 bool need_rmap_locks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266
267 /*
268 * We'd prefer to avoid failure later on in do_munmap:
269 * which may split one vma into three before unmapping.
270 */
271 if (mm->map_count >= sysctl_max_map_count - 3)
272 return -ENOMEM;
273
Hugh Dickins1ff82992009-09-21 17:02:05 -0700274 /*
275 * Advise KSM to break any KSM pages in the area to be moved:
276 * it would be confusing if they were to turn up at the new
277 * location, where they happen to coincide with different KSM
278 * pages recently unmapped. But leave vma->vm_flags as it was,
279 * so KSM can come around to merge on vma and new_vma afterwards.
280 */
Hugh Dickins7103ad32009-09-21 17:02:28 -0700281 err = ksm_madvise(vma, old_addr, old_addr + old_len,
282 MADV_UNMERGEABLE, &vm_flags);
283 if (err)
284 return err;
Hugh Dickins1ff82992009-09-21 17:02:05 -0700285
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 new_pgoff = vma->vm_pgoff + ((old_addr - vma->vm_start) >> PAGE_SHIFT);
Michel Lespinasse38a76012012-10-08 16:31:50 -0700287 new_vma = copy_vma(&vma, new_addr, new_len, new_pgoff,
288 &need_rmap_locks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 if (!new_vma)
290 return -ENOMEM;
291
Michel Lespinasse38a76012012-10-08 16:31:50 -0700292 moved_len = move_page_tables(vma, old_addr, new_vma, new_addr, old_len,
293 need_rmap_locks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 if (moved_len < old_len) {
Oleg Nesterovdf1eab32015-09-04 15:48:01 -0700295 err = -ENOMEM;
Oleg Nesterov5477e702015-09-04 15:48:04 -0700296 } else if (vma->vm_ops && vma->vm_ops->mremap) {
297 err = vma->vm_ops->mremap(new_vma);
Oleg Nesterovdf1eab32015-09-04 15:48:01 -0700298 }
299
300 if (unlikely(err)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 /*
302 * On error, move entries back from new area to old,
303 * which will succeed since page tables still there,
304 * and then proceed to unmap new area instead of old.
305 */
Michel Lespinasse38a76012012-10-08 16:31:50 -0700306 move_page_tables(new_vma, new_addr, vma, old_addr, moved_len,
307 true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 vma = new_vma;
309 old_len = new_len;
310 old_addr = new_addr;
Oleg Nesterovdf1eab32015-09-04 15:48:01 -0700311 new_addr = err;
Laurent Dufour4abad2c2015-06-24 16:56:19 -0700312 } else {
Laurent Dufour4abad2c2015-06-24 16:56:19 -0700313 arch_remap(mm, old_addr, old_addr + old_len,
314 new_addr, new_addr + new_len);
Al Virob2edffd2015-04-06 17:48:54 -0400315 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
317 /* Conceal VM_ACCOUNT so old reservation is not undone */
318 if (vm_flags & VM_ACCOUNT) {
319 vma->vm_flags &= ~VM_ACCOUNT;
320 excess = vma->vm_end - vma->vm_start - old_len;
321 if (old_addr > vma->vm_start &&
322 old_addr + old_len < vma->vm_end)
323 split = 1;
324 }
325
Kirill Korotaev71799062005-05-16 21:53:18 -0700326 /*
Hugh Dickins365e9c872005-10-29 18:16:18 -0700327 * If we failed to move page tables we still do total_vm increment
328 * since do_munmap() will decrement it by old_len == new_len.
329 *
330 * Since total_vm is about to be raised artificially high for a
331 * moment, we need to restore high watermark afterwards: if stats
332 * are taken meanwhile, total_vm and hiwater_vm appear too high.
333 * If this were a serious issue, we'd add a flag to do_munmap().
Kirill Korotaev71799062005-05-16 21:53:18 -0700334 */
Hugh Dickins365e9c872005-10-29 18:16:18 -0700335 hiwater_vm = mm->hiwater_vm;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -0800336 vm_stat_account(mm, vma->vm_flags, new_len >> PAGE_SHIFT);
Kirill Korotaev71799062005-05-16 21:53:18 -0700337
Toshi Kanid9fe4fa2015-12-22 17:54:23 -0700338 /* Tell pfnmap has moved from this vma */
339 if (unlikely(vma->vm_flags & VM_PFNMAP))
340 untrack_pfn_moved(vma);
341
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 if (do_munmap(mm, old_addr, old_len) < 0) {
343 /* OOM: unable to split vma, just get accounts right */
344 vm_unacct_memory(excess >> PAGE_SHIFT);
345 excess = 0;
346 }
Hugh Dickins365e9c872005-10-29 18:16:18 -0700347 mm->hiwater_vm = hiwater_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348
349 /* Restore VM_ACCOUNT if one or two pieces of vma left */
350 if (excess) {
351 vma->vm_flags |= VM_ACCOUNT;
352 if (split)
353 vma->vm_next->vm_flags |= VM_ACCOUNT;
354 }
355
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 if (vm_flags & VM_LOCKED) {
357 mm->locked_vm += new_len >> PAGE_SHIFT;
Michel Lespinasse81909b82013-02-22 16:32:41 -0800358 *locked = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 }
360
361 return new_addr;
362}
363
Al Viro54f5de72009-11-24 07:17:46 -0500364static struct vm_area_struct *vma_to_resize(unsigned long addr,
365 unsigned long old_len, unsigned long new_len, unsigned long *p)
366{
367 struct mm_struct *mm = current->mm;
368 struct vm_area_struct *vma = find_vma(mm, addr);
Oleg Nesterov1d391682015-09-04 15:48:10 -0700369 unsigned long pgoff;
Al Viro54f5de72009-11-24 07:17:46 -0500370
371 if (!vma || vma->vm_start > addr)
Derek6cd576132015-04-15 16:14:02 -0700372 return ERR_PTR(-EFAULT);
Al Viro54f5de72009-11-24 07:17:46 -0500373
374 if (is_vm_hugetlb_page(vma))
Derek6cd576132015-04-15 16:14:02 -0700375 return ERR_PTR(-EINVAL);
Al Viro54f5de72009-11-24 07:17:46 -0500376
377 /* We can't remap across vm area boundaries */
378 if (old_len > vma->vm_end - addr)
Derek6cd576132015-04-15 16:14:02 -0700379 return ERR_PTR(-EFAULT);
Al Viro54f5de72009-11-24 07:17:46 -0500380
Oleg Nesterov1d391682015-09-04 15:48:10 -0700381 if (new_len == old_len)
382 return vma;
Linus Torvalds982134b2011-04-07 07:35:50 -0700383
Oleg Nesterov1d391682015-09-04 15:48:10 -0700384 /* Need to be careful about a growing mapping */
385 pgoff = (addr - vma->vm_start) >> PAGE_SHIFT;
386 pgoff += vma->vm_pgoff;
387 if (pgoff + (new_len >> PAGE_SHIFT) < pgoff)
388 return ERR_PTR(-EINVAL);
389
390 if (vma->vm_flags & (VM_DONTEXPAND | VM_PFNMAP))
391 return ERR_PTR(-EFAULT);
Al Viro54f5de72009-11-24 07:17:46 -0500392
393 if (vma->vm_flags & VM_LOCKED) {
394 unsigned long locked, lock_limit;
395 locked = mm->locked_vm << PAGE_SHIFT;
Jiri Slaby59e99e52010-03-05 13:41:44 -0800396 lock_limit = rlimit(RLIMIT_MEMLOCK);
Al Viro54f5de72009-11-24 07:17:46 -0500397 locked += new_len - old_len;
398 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
Derek6cd576132015-04-15 16:14:02 -0700399 return ERR_PTR(-EAGAIN);
Al Viro54f5de72009-11-24 07:17:46 -0500400 }
401
Konstantin Khlebnikov84638332016-01-14 15:22:07 -0800402 if (!may_expand_vm(mm, vma->vm_flags,
403 (new_len - old_len) >> PAGE_SHIFT))
Derek6cd576132015-04-15 16:14:02 -0700404 return ERR_PTR(-ENOMEM);
Al Viro54f5de72009-11-24 07:17:46 -0500405
406 if (vma->vm_flags & VM_ACCOUNT) {
407 unsigned long charged = (new_len - old_len) >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +0000408 if (security_vm_enough_memory_mm(mm, charged))
Derek6cd576132015-04-15 16:14:02 -0700409 return ERR_PTR(-ENOMEM);
Al Viro54f5de72009-11-24 07:17:46 -0500410 *p = charged;
411 }
412
413 return vma;
Al Viro54f5de72009-11-24 07:17:46 -0500414}
415
Michel Lespinasse81909b82013-02-22 16:32:41 -0800416static unsigned long mremap_to(unsigned long addr, unsigned long old_len,
417 unsigned long new_addr, unsigned long new_len, bool *locked)
Al Viroecc1a892009-11-24 07:28:07 -0500418{
419 struct mm_struct *mm = current->mm;
420 struct vm_area_struct *vma;
421 unsigned long ret = -EINVAL;
422 unsigned long charged = 0;
Al Viro097eed12009-11-24 08:43:52 -0500423 unsigned long map_flags;
Al Viroecc1a892009-11-24 07:28:07 -0500424
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800425 if (offset_in_page(new_addr))
Al Viroecc1a892009-11-24 07:28:07 -0500426 goto out;
427
428 if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len)
429 goto out;
430
Oleg Nesterov99432422015-09-04 15:48:13 -0700431 /* Ensure the old/new locations do not overlap */
432 if (addr + old_len > new_addr && new_addr + new_len > addr)
Al Viroecc1a892009-11-24 07:28:07 -0500433 goto out;
434
Al Viroecc1a892009-11-24 07:28:07 -0500435 ret = do_munmap(mm, new_addr, new_len);
436 if (ret)
437 goto out;
438
439 if (old_len >= new_len) {
440 ret = do_munmap(mm, addr+new_len, old_len - new_len);
441 if (ret && old_len != new_len)
442 goto out;
443 old_len = new_len;
444 }
445
446 vma = vma_to_resize(addr, old_len, new_len, &charged);
447 if (IS_ERR(vma)) {
448 ret = PTR_ERR(vma);
449 goto out;
450 }
451
Al Viro097eed12009-11-24 08:43:52 -0500452 map_flags = MAP_FIXED;
453 if (vma->vm_flags & VM_MAYSHARE)
454 map_flags |= MAP_SHARED;
Al Viro9206de92009-12-03 15:23:11 -0500455
Al Viro097eed12009-11-24 08:43:52 -0500456 ret = get_unmapped_area(vma->vm_file, new_addr, new_len, vma->vm_pgoff +
457 ((addr - vma->vm_start) >> PAGE_SHIFT),
458 map_flags);
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800459 if (offset_in_page(ret))
Al Viro097eed12009-11-24 08:43:52 -0500460 goto out1;
461
Michel Lespinasse81909b82013-02-22 16:32:41 -0800462 ret = move_vma(vma, addr, old_len, new_len, new_addr, locked);
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800463 if (!(offset_in_page(ret)))
Al Viro097eed12009-11-24 08:43:52 -0500464 goto out;
465out1:
466 vm_unacct_memory(charged);
Al Viroecc1a892009-11-24 07:28:07 -0500467
468out:
469 return ret;
470}
471
Al Viro1a0ef852009-11-24 07:43:18 -0500472static int vma_expandable(struct vm_area_struct *vma, unsigned long delta)
473{
Al Virof106af42009-11-24 08:25:18 -0500474 unsigned long end = vma->vm_end + delta;
Al Viro9206de92009-12-03 15:23:11 -0500475 if (end < vma->vm_end) /* overflow */
Al Viro1a0ef852009-11-24 07:43:18 -0500476 return 0;
Al Viro9206de92009-12-03 15:23:11 -0500477 if (vma->vm_next && vma->vm_next->vm_start < end) /* intersection */
Al Virof106af42009-11-24 08:25:18 -0500478 return 0;
479 if (get_unmapped_area(NULL, vma->vm_start, end - vma->vm_start,
480 0, MAP_FIXED) & ~PAGE_MASK)
481 return 0;
Al Viro1a0ef852009-11-24 07:43:18 -0500482 return 1;
483}
484
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485/*
486 * Expand (or shrink) an existing mapping, potentially moving it at the
487 * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
488 *
489 * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
490 * This option implies MREMAP_MAYMOVE.
491 */
Al Viro63a81db2012-05-30 11:32:04 -0400492SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
493 unsigned long, new_len, unsigned long, flags,
494 unsigned long, new_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495{
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700496 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 struct vm_area_struct *vma;
498 unsigned long ret = -EINVAL;
499 unsigned long charged = 0;
Michel Lespinasse81909b82013-02-22 16:32:41 -0800500 bool locked = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501
502 if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE))
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700503 return ret;
504
505 if (flags & MREMAP_FIXED && !(flags & MREMAP_MAYMOVE))
506 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800508 if (offset_in_page(addr))
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700509 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510
511 old_len = PAGE_ALIGN(old_len);
512 new_len = PAGE_ALIGN(new_len);
513
514 /*
515 * We allow a zero old-len as a special case
516 * for DOS-emu "duplicate shm area" thing. But
517 * a zero new-len is nonsensical.
518 */
519 if (!new_len)
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700520 return ret;
521
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700522 if (down_write_killable(&current->mm->mmap_sem))
523 return -EINTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 if (flags & MREMAP_FIXED) {
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700526 ret = mremap_to(addr, old_len, new_addr, new_len,
527 &locked);
Al Viroecc1a892009-11-24 07:28:07 -0500528 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 }
530
531 /*
532 * Always allow a shrinking remap: that just unmaps
533 * the unnecessary pages..
534 * do_munmap does all the needed commit accounting
535 */
536 if (old_len >= new_len) {
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700537 ret = do_munmap(mm, addr+new_len, old_len - new_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 if (ret && old_len != new_len)
539 goto out;
540 ret = addr;
Al Viroecc1a892009-11-24 07:28:07 -0500541 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 }
543
544 /*
Al Viroecc1a892009-11-24 07:28:07 -0500545 * Ok, we need to grow..
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 */
Al Viro54f5de72009-11-24 07:17:46 -0500547 vma = vma_to_resize(addr, old_len, new_len, &charged);
548 if (IS_ERR(vma)) {
549 ret = PTR_ERR(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 }
552
553 /* old_len exactly to the end of the area..
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 */
Al Viroecc1a892009-11-24 07:28:07 -0500555 if (old_len == vma->vm_end - addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 /* can we just expand the current mapping? */
Al Viro1a0ef852009-11-24 07:43:18 -0500557 if (vma_expandable(vma, new_len - old_len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 int pages = (new_len - old_len) >> PAGE_SHIFT;
559
Rik van Riel5beb4932010-03-05 13:42:07 -0800560 if (vma_adjust(vma, vma->vm_start, addr + new_len,
561 vma->vm_pgoff, NULL)) {
562 ret = -ENOMEM;
563 goto out;
564 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565
Konstantin Khlebnikov84638332016-01-14 15:22:07 -0800566 vm_stat_account(mm, vma->vm_flags, pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 if (vma->vm_flags & VM_LOCKED) {
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700568 mm->locked_vm += pages;
Michel Lespinasse81909b82013-02-22 16:32:41 -0800569 locked = true;
570 new_addr = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 }
572 ret = addr;
573 goto out;
574 }
575 }
576
577 /*
578 * We weren't able to just expand or shrink the area,
579 * we need to create a new one and move it..
580 */
581 ret = -ENOMEM;
582 if (flags & MREMAP_MAYMOVE) {
Al Viroecc1a892009-11-24 07:28:07 -0500583 unsigned long map_flags = 0;
584 if (vma->vm_flags & VM_MAYSHARE)
585 map_flags |= MAP_SHARED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
Al Viroecc1a892009-11-24 07:28:07 -0500587 new_addr = get_unmapped_area(vma->vm_file, 0, new_len,
Al Viro93587412009-11-24 08:45:24 -0500588 vma->vm_pgoff +
589 ((addr - vma->vm_start) >> PAGE_SHIFT),
590 map_flags);
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800591 if (offset_in_page(new_addr)) {
Al Viroecc1a892009-11-24 07:28:07 -0500592 ret = new_addr;
593 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 }
Al Viroecc1a892009-11-24 07:28:07 -0500595
Michel Lespinasse81909b82013-02-22 16:32:41 -0800596 ret = move_vma(vma, addr, old_len, new_len, new_addr, &locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 }
598out:
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800599 if (offset_in_page(ret)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 vm_unacct_memory(charged);
Oleg Nesterovd456fb92015-09-04 15:48:07 -0700601 locked = 0;
602 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 up_write(&current->mm->mmap_sem);
Michel Lespinasse81909b82013-02-22 16:32:41 -0800604 if (locked && new_len > old_len)
605 mm_populate(new_addr + old_len, new_len - old_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 return ret;
607}