blob: d77946a997f798ecc25ccaee2886139ff9d9a587 [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>
Laurent Dufour4abad2c2015-06-24 16:56:19 -070025#include <linux/mm-arch-hooks.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <asm/cacheflush.h>
28#include <asm/tlbflush.h>
29
Rik van Rielba470de2008-10-18 20:26:50 -070030#include "internal.h"
31
Hugh Dickins7be7a542005-10-29 18:16:00 -070032static pmd_t *get_old_pmd(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070033{
34 pgd_t *pgd;
35 pud_t *pud;
36 pmd_t *pmd;
37
38 pgd = pgd_offset(mm, addr);
39 if (pgd_none_or_clear_bad(pgd))
40 return NULL;
41
42 pud = pud_offset(pgd, addr);
43 if (pud_none_or_clear_bad(pud))
44 return NULL;
45
46 pmd = pmd_offset(pud, addr);
Andrea Arcangeli37a1c492011-10-31 17:08:30 -070047 if (pmd_none(*pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -070048 return NULL;
49
Hugh Dickins7be7a542005-10-29 18:16:00 -070050 return pmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -070051}
52
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -080053static pmd_t *alloc_new_pmd(struct mm_struct *mm, struct vm_area_struct *vma,
54 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070055{
56 pgd_t *pgd;
57 pud_t *pud;
Hugh Dickinsc74df322005-10-29 18:16:23 -070058 pmd_t *pmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
60 pgd = pgd_offset(mm, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 pud = pud_alloc(mm, pgd, addr);
62 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -070063 return NULL;
Hugh Dickins7be7a542005-10-29 18:16:00 -070064
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 pmd = pmd_alloc(mm, pud, addr);
Hugh Dickins57a8f0c2013-10-16 13:47:09 -070066 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -070067 return NULL;
Hugh Dickins7be7a542005-10-29 18:16:00 -070068
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -080069 VM_BUG_ON(pmd_trans_huge(*pmd));
Hugh Dickinsc74df322005-10-29 18:16:23 -070070
Hugh Dickins7be7a542005-10-29 18:16:00 -070071 return pmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -070072}
73
Cyrill Gorcunov6dec97d2013-08-27 12:37:18 +040074static pte_t move_soft_dirty_pte(pte_t pte)
75{
76 /*
77 * Set soft dirty bit so we can notice
78 * in userspace the ptes were moved.
79 */
80#ifdef CONFIG_MEM_SOFT_DIRTY
81 if (pte_present(pte))
82 pte = pte_mksoft_dirty(pte);
83 else if (is_swap_pte(pte))
84 pte = pte_swp_mksoft_dirty(pte);
Cyrill Gorcunov6dec97d2013-08-27 12:37:18 +040085#endif
86 return pte;
87}
88
Hugh Dickins7be7a542005-10-29 18:16:00 -070089static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd,
90 unsigned long old_addr, unsigned long old_end,
91 struct vm_area_struct *new_vma, pmd_t *new_pmd,
Michel Lespinasse38a76012012-10-08 16:31:50 -070092 unsigned long new_addr, bool need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -070093{
94 struct address_space *mapping = NULL;
Michel Lespinasse38a76012012-10-08 16:31:50 -070095 struct anon_vma *anon_vma = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 struct mm_struct *mm = vma->vm_mm;
Hugh Dickins7be7a542005-10-29 18:16:00 -070097 pte_t *old_pte, *new_pte, pte;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -070098 spinlock_t *old_ptl, *new_ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
Michel Lespinasse38a76012012-10-08 16:31:50 -0700100 /*
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800101 * When need_rmap_locks is true, we take the i_mmap_rwsem and anon_vma
Michel Lespinasse38a76012012-10-08 16:31:50 -0700102 * locks to ensure that rmap will always observe either the old or the
103 * new ptes. This is the easiest way to avoid races with
104 * truncate_pagecache(), page migration, etc...
105 *
106 * When need_rmap_locks is false, we use other ways to avoid
107 * such races:
108 *
109 * - During exec() shift_arg_pages(), we use a specially tagged vma
110 * which rmap call sites look for using is_vma_temporary_stack().
111 *
112 * - During mremap(), new_vma is often known to be placed after vma
113 * in rmap traversal order. This ensures rmap will always observe
114 * either the old pte, or the new pte, or both (the page table locks
115 * serialize access to individual ptes, but only rmap traversal
116 * order guarantees that we won't miss both the old and new ptes).
117 */
118 if (need_rmap_locks) {
119 if (vma->vm_file) {
120 mapping = vma->vm_file->f_mapping;
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800121 i_mmap_lock_write(mapping);
Michel Lespinasse38a76012012-10-08 16:31:50 -0700122 }
123 if (vma->anon_vma) {
124 anon_vma = vma->anon_vma;
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000125 anon_vma_lock_write(anon_vma);
Michel Lespinasse38a76012012-10-08 16:31:50 -0700126 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 }
Hugh Dickins7be7a542005-10-29 18:16:00 -0700128
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700129 /*
130 * We don't have to worry about the ordering of src and dst
131 * pte locks because exclusive mmap_sem prevents deadlock.
132 */
Hugh Dickinsc74df322005-10-29 18:16:23 -0700133 old_pte = pte_offset_map_lock(mm, old_pmd, old_addr, &old_ptl);
Peter Zijlstraece0e2b2010-10-26 14:21:52 -0700134 new_pte = pte_offset_map(new_pmd, new_addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700135 new_ptl = pte_lockptr(mm, new_pmd);
136 if (new_ptl != old_ptl)
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700137 spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700138 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139
Hugh Dickins7be7a542005-10-29 18:16:00 -0700140 for (; old_addr < old_end; old_pte++, old_addr += PAGE_SIZE,
141 new_pte++, new_addr += PAGE_SIZE) {
142 if (pte_none(*old_pte))
143 continue;
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700144 pte = ptep_get_and_clear(mm, old_addr, old_pte);
Hugh Dickins7be7a542005-10-29 18:16:00 -0700145 pte = move_pte(pte, new_vma->vm_page_prot, old_addr, new_addr);
Cyrill Gorcunov6dec97d2013-08-27 12:37:18 +0400146 pte = move_soft_dirty_pte(pte);
147 set_pte_at(mm, new_addr, new_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 }
Hugh Dickins7be7a542005-10-29 18:16:00 -0700149
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700150 arch_leave_lazy_mmu_mode();
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700151 if (new_ptl != old_ptl)
152 spin_unlock(new_ptl);
Peter Zijlstraece0e2b2010-10-26 14:21:52 -0700153 pte_unmap(new_pte - 1);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700154 pte_unmap_unlock(old_pte - 1, old_ptl);
Michel Lespinasse108d6642012-10-08 16:31:36 -0700155 if (anon_vma)
Konstantin Khlebnikov08b52702013-02-22 16:34:40 -0800156 anon_vma_unlock_write(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 if (mapping)
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800158 i_mmap_unlock_write(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159}
160
Hugh Dickins7be7a542005-10-29 18:16:00 -0700161#define LATENCY_LIMIT (64 * PAGE_SIZE)
162
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700163unsigned long move_page_tables(struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 unsigned long old_addr, struct vm_area_struct *new_vma,
Michel Lespinasse38a76012012-10-08 16:31:50 -0700165 unsigned long new_addr, unsigned long len,
166 bool need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167{
Hugh Dickins7be7a542005-10-29 18:16:00 -0700168 unsigned long extent, next, old_end;
169 pmd_t *old_pmd, *new_pmd;
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700170 bool need_flush = false;
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700171 unsigned long mmun_start; /* For mmu_notifiers */
172 unsigned long mmun_end; /* For mmu_notifiers */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173
Hugh Dickins7be7a542005-10-29 18:16:00 -0700174 old_end = old_addr + len;
175 flush_cache_range(vma, old_addr, old_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700177 mmun_start = old_addr;
178 mmun_end = old_end;
179 mmu_notifier_invalidate_range_start(vma->vm_mm, mmun_start, mmun_end);
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700180
Hugh Dickins7be7a542005-10-29 18:16:00 -0700181 for (; old_addr < old_end; old_addr += extent, new_addr += extent) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 cond_resched();
Hugh Dickins7be7a542005-10-29 18:16:00 -0700183 next = (old_addr + PMD_SIZE) & PMD_MASK;
Andrea Arcangeliebed4842011-10-31 17:08:22 -0700184 /* even if next overflowed, extent below will be ok */
Hugh Dickins7be7a542005-10-29 18:16:00 -0700185 extent = next - old_addr;
Andrea Arcangeliebed4842011-10-31 17:08:22 -0700186 if (extent > old_end - old_addr)
187 extent = old_end - old_addr;
Hugh Dickins7be7a542005-10-29 18:16:00 -0700188 old_pmd = get_old_pmd(vma->vm_mm, old_addr);
189 if (!old_pmd)
190 continue;
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800191 new_pmd = alloc_new_pmd(vma->vm_mm, vma, new_addr);
Hugh Dickins7be7a542005-10-29 18:16:00 -0700192 if (!new_pmd)
193 break;
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700194 if (pmd_trans_huge(*old_pmd)) {
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700195 if (extent == HPAGE_PMD_SIZE) {
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800196 bool moved;
Sasha Levin81d1b092014-10-09 15:28:10 -0700197 VM_BUG_ON_VMA(vma->vm_file || !vma->anon_vma,
198 vma);
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700199 /* See comment in move_ptes() */
200 if (need_rmap_locks)
201 anon_vma_lock_write(vma->anon_vma);
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800202 moved = move_huge_pmd(vma, new_vma, old_addr,
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700203 new_addr, old_end,
204 old_pmd, new_pmd);
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700205 if (need_rmap_locks)
206 anon_vma_unlock_write(vma->anon_vma);
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800207 if (moved) {
208 need_flush = true;
209 continue;
210 }
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700211 }
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800212 split_huge_pmd(vma, old_pmd, old_addr);
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700213 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) {
Oleg Nesterovdf1eab32015-09-04 15:48:01 -0700278 err = -ENOMEM;
Oleg Nesterov5477e702015-09-04 15:48:04 -0700279 } else if (vma->vm_ops && vma->vm_ops->mremap) {
280 err = vma->vm_ops->mremap(new_vma);
Oleg Nesterovdf1eab32015-09-04 15:48:01 -0700281 }
282
283 if (unlikely(err)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 /*
285 * On error, move entries back from new area to old,
286 * which will succeed since page tables still there,
287 * and then proceed to unmap new area instead of old.
288 */
Michel Lespinasse38a76012012-10-08 16:31:50 -0700289 move_page_tables(new_vma, new_addr, vma, old_addr, moved_len,
290 true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 vma = new_vma;
292 old_len = new_len;
293 old_addr = new_addr;
Oleg Nesterovdf1eab32015-09-04 15:48:01 -0700294 new_addr = err;
Laurent Dufour4abad2c2015-06-24 16:56:19 -0700295 } else {
Laurent Dufour4abad2c2015-06-24 16:56:19 -0700296 arch_remap(mm, old_addr, old_addr + old_len,
297 new_addr, new_addr + new_len);
Al Virob2edffd2015-04-06 17:48:54 -0400298 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299
300 /* Conceal VM_ACCOUNT so old reservation is not undone */
301 if (vm_flags & VM_ACCOUNT) {
302 vma->vm_flags &= ~VM_ACCOUNT;
303 excess = vma->vm_end - vma->vm_start - old_len;
304 if (old_addr > vma->vm_start &&
305 old_addr + old_len < vma->vm_end)
306 split = 1;
307 }
308
Kirill Korotaev71799062005-05-16 21:53:18 -0700309 /*
Hugh Dickins365e9c872005-10-29 18:16:18 -0700310 * If we failed to move page tables we still do total_vm increment
311 * since do_munmap() will decrement it by old_len == new_len.
312 *
313 * Since total_vm is about to be raised artificially high for a
314 * moment, we need to restore high watermark afterwards: if stats
315 * are taken meanwhile, total_vm and hiwater_vm appear too high.
316 * If this were a serious issue, we'd add a flag to do_munmap().
Kirill Korotaev71799062005-05-16 21:53:18 -0700317 */
Hugh Dickins365e9c872005-10-29 18:16:18 -0700318 hiwater_vm = mm->hiwater_vm;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -0800319 vm_stat_account(mm, vma->vm_flags, new_len >> PAGE_SHIFT);
Kirill Korotaev71799062005-05-16 21:53:18 -0700320
Toshi Kanid9fe4fa2015-12-22 17:54:23 -0700321 /* Tell pfnmap has moved from this vma */
322 if (unlikely(vma->vm_flags & VM_PFNMAP))
323 untrack_pfn_moved(vma);
324
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 if (do_munmap(mm, old_addr, old_len) < 0) {
326 /* OOM: unable to split vma, just get accounts right */
327 vm_unacct_memory(excess >> PAGE_SHIFT);
328 excess = 0;
329 }
Hugh Dickins365e9c872005-10-29 18:16:18 -0700330 mm->hiwater_vm = hiwater_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331
332 /* Restore VM_ACCOUNT if one or two pieces of vma left */
333 if (excess) {
334 vma->vm_flags |= VM_ACCOUNT;
335 if (split)
336 vma->vm_next->vm_flags |= VM_ACCOUNT;
337 }
338
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 if (vm_flags & VM_LOCKED) {
340 mm->locked_vm += new_len >> PAGE_SHIFT;
Michel Lespinasse81909b82013-02-22 16:32:41 -0800341 *locked = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 }
343
344 return new_addr;
345}
346
Al Viro54f5de72009-11-24 07:17:46 -0500347static struct vm_area_struct *vma_to_resize(unsigned long addr,
348 unsigned long old_len, unsigned long new_len, unsigned long *p)
349{
350 struct mm_struct *mm = current->mm;
351 struct vm_area_struct *vma = find_vma(mm, addr);
Oleg Nesterov1d391682015-09-04 15:48:10 -0700352 unsigned long pgoff;
Al Viro54f5de72009-11-24 07:17:46 -0500353
354 if (!vma || vma->vm_start > addr)
Derek6cd576132015-04-15 16:14:02 -0700355 return ERR_PTR(-EFAULT);
Al Viro54f5de72009-11-24 07:17:46 -0500356
357 if (is_vm_hugetlb_page(vma))
Derek6cd576132015-04-15 16:14:02 -0700358 return ERR_PTR(-EINVAL);
Al Viro54f5de72009-11-24 07:17:46 -0500359
360 /* We can't remap across vm area boundaries */
361 if (old_len > vma->vm_end - addr)
Derek6cd576132015-04-15 16:14:02 -0700362 return ERR_PTR(-EFAULT);
Al Viro54f5de72009-11-24 07:17:46 -0500363
Oleg Nesterov1d391682015-09-04 15:48:10 -0700364 if (new_len == old_len)
365 return vma;
Linus Torvalds982134b2011-04-07 07:35:50 -0700366
Oleg Nesterov1d391682015-09-04 15:48:10 -0700367 /* Need to be careful about a growing mapping */
368 pgoff = (addr - vma->vm_start) >> PAGE_SHIFT;
369 pgoff += vma->vm_pgoff;
370 if (pgoff + (new_len >> PAGE_SHIFT) < pgoff)
371 return ERR_PTR(-EINVAL);
372
373 if (vma->vm_flags & (VM_DONTEXPAND | VM_PFNMAP))
374 return ERR_PTR(-EFAULT);
Al Viro54f5de72009-11-24 07:17:46 -0500375
376 if (vma->vm_flags & VM_LOCKED) {
377 unsigned long locked, lock_limit;
378 locked = mm->locked_vm << PAGE_SHIFT;
Jiri Slaby59e99e52010-03-05 13:41:44 -0800379 lock_limit = rlimit(RLIMIT_MEMLOCK);
Al Viro54f5de72009-11-24 07:17:46 -0500380 locked += new_len - old_len;
381 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
Derek6cd576132015-04-15 16:14:02 -0700382 return ERR_PTR(-EAGAIN);
Al Viro54f5de72009-11-24 07:17:46 -0500383 }
384
Konstantin Khlebnikov84638332016-01-14 15:22:07 -0800385 if (!may_expand_vm(mm, vma->vm_flags,
386 (new_len - old_len) >> PAGE_SHIFT))
Derek6cd576132015-04-15 16:14:02 -0700387 return ERR_PTR(-ENOMEM);
Al Viro54f5de72009-11-24 07:17:46 -0500388
389 if (vma->vm_flags & VM_ACCOUNT) {
390 unsigned long charged = (new_len - old_len) >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +0000391 if (security_vm_enough_memory_mm(mm, charged))
Derek6cd576132015-04-15 16:14:02 -0700392 return ERR_PTR(-ENOMEM);
Al Viro54f5de72009-11-24 07:17:46 -0500393 *p = charged;
394 }
395
396 return vma;
Al Viro54f5de72009-11-24 07:17:46 -0500397}
398
Michel Lespinasse81909b82013-02-22 16:32:41 -0800399static unsigned long mremap_to(unsigned long addr, unsigned long old_len,
400 unsigned long new_addr, unsigned long new_len, bool *locked)
Al Viroecc1a892009-11-24 07:28:07 -0500401{
402 struct mm_struct *mm = current->mm;
403 struct vm_area_struct *vma;
404 unsigned long ret = -EINVAL;
405 unsigned long charged = 0;
Al Viro097eed12009-11-24 08:43:52 -0500406 unsigned long map_flags;
Al Viroecc1a892009-11-24 07:28:07 -0500407
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800408 if (offset_in_page(new_addr))
Al Viroecc1a892009-11-24 07:28:07 -0500409 goto out;
410
411 if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len)
412 goto out;
413
Oleg Nesterov99432422015-09-04 15:48:13 -0700414 /* Ensure the old/new locations do not overlap */
415 if (addr + old_len > new_addr && new_addr + new_len > addr)
Al Viroecc1a892009-11-24 07:28:07 -0500416 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);
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800442 if (offset_in_page(ret))
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);
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800446 if (!(offset_in_page(ret)))
Al Viro097eed12009-11-24 08:43:52 -0500447 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
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800491 if (offset_in_page(addr))
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
Konstantin Khlebnikov84638332016-01-14 15:22:07 -0800548 vm_stat_account(mm, vma->vm_flags, 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);
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800573 if (offset_in_page(new_addr)) {
Al Viroecc1a892009-11-24 07:28:07 -0500574 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:
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800581 if (offset_in_page(ret)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 vm_unacct_memory(charged);
Oleg Nesterovd456fb92015-09-04 15:48:07 -0700583 locked = 0;
584 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 up_write(&current->mm->mmap_sem);
Michel Lespinasse81909b82013-02-22 16:32:41 -0800586 if (locked && new_len > old_len)
587 mm_populate(new_addr + old_len, new_len - old_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 return ret;
589}