blob: a7c93eceb1c8d1ce59235d47ec7e49df4efefb86 [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)) {
195 int err = 0;
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700196 if (extent == HPAGE_PMD_SIZE) {
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);
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700202 err = move_huge_pmd(vma, new_vma, old_addr,
203 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);
207 }
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700208 if (err > 0) {
209 need_flush = true;
210 continue;
211 } else if (!err) {
Kirill A. Shutemove1803772012-12-12 13:50:59 -0800212 split_huge_page_pmd(vma, old_addr, old_pmd);
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700213 }
214 VM_BUG_ON(pmd_trans_huge(*old_pmd));
215 }
216 if (pmd_none(*new_pmd) && __pte_alloc(new_vma->vm_mm, new_vma,
217 new_pmd, new_addr))
218 break;
Hugh Dickins7be7a542005-10-29 18:16:00 -0700219 next = (new_addr + PMD_SIZE) & PMD_MASK;
220 if (extent > next - new_addr)
221 extent = next - new_addr;
222 if (extent > LATENCY_LIMIT)
223 extent = LATENCY_LIMIT;
224 move_ptes(vma, old_pmd, old_addr, old_addr + extent,
Michel Lespinasse38a76012012-10-08 16:31:50 -0700225 new_vma, new_pmd, new_addr, need_rmap_locks);
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700226 need_flush = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 }
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700228 if (likely(need_flush))
229 flush_tlb_range(vma, old_end-len, old_addr);
230
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700231 mmu_notifier_invalidate_range_end(vma->vm_mm, mmun_start, mmun_end);
Hugh Dickins7be7a542005-10-29 18:16:00 -0700232
233 return len + old_addr - old_end; /* how much done */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234}
235
236static unsigned long move_vma(struct vm_area_struct *vma,
237 unsigned long old_addr, unsigned long old_len,
Michel Lespinasse81909b82013-02-22 16:32:41 -0800238 unsigned long new_len, unsigned long new_addr, bool *locked)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239{
240 struct mm_struct *mm = vma->vm_mm;
241 struct vm_area_struct *new_vma;
242 unsigned long vm_flags = vma->vm_flags;
243 unsigned long new_pgoff;
244 unsigned long moved_len;
245 unsigned long excess = 0;
Hugh Dickins365e9c872005-10-29 18:16:18 -0700246 unsigned long hiwater_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 int split = 0;
Hugh Dickins7103ad32009-09-21 17:02:28 -0700248 int err;
Michel Lespinasse38a76012012-10-08 16:31:50 -0700249 bool need_rmap_locks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250
251 /*
252 * We'd prefer to avoid failure later on in do_munmap:
253 * which may split one vma into three before unmapping.
254 */
255 if (mm->map_count >= sysctl_max_map_count - 3)
256 return -ENOMEM;
257
Hugh Dickins1ff82992009-09-21 17:02:05 -0700258 /*
259 * Advise KSM to break any KSM pages in the area to be moved:
260 * it would be confusing if they were to turn up at the new
261 * location, where they happen to coincide with different KSM
262 * pages recently unmapped. But leave vma->vm_flags as it was,
263 * so KSM can come around to merge on vma and new_vma afterwards.
264 */
Hugh Dickins7103ad32009-09-21 17:02:28 -0700265 err = ksm_madvise(vma, old_addr, old_addr + old_len,
266 MADV_UNMERGEABLE, &vm_flags);
267 if (err)
268 return err;
Hugh Dickins1ff82992009-09-21 17:02:05 -0700269
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 new_pgoff = vma->vm_pgoff + ((old_addr - vma->vm_start) >> PAGE_SHIFT);
Michel Lespinasse38a76012012-10-08 16:31:50 -0700271 new_vma = copy_vma(&vma, new_addr, new_len, new_pgoff,
272 &need_rmap_locks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 if (!new_vma)
274 return -ENOMEM;
275
Michel Lespinasse38a76012012-10-08 16:31:50 -0700276 moved_len = move_page_tables(vma, old_addr, new_vma, new_addr, old_len,
277 need_rmap_locks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 if (moved_len < old_len) {
279 /*
280 * On error, move entries back from new area to old,
281 * which will succeed since page tables still there,
282 * and then proceed to unmap new area instead of old.
283 */
Michel Lespinasse38a76012012-10-08 16:31:50 -0700284 move_page_tables(new_vma, new_addr, vma, old_addr, moved_len,
285 true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 vma = new_vma;
287 old_len = new_len;
288 old_addr = new_addr;
289 new_addr = -ENOMEM;
Laurent Dufour4abad2c2015-06-24 16:56:19 -0700290 } else {
291 if (vma->vm_file && vma->vm_file->f_op->mremap) {
292 err = vma->vm_file->f_op->mremap(vma->vm_file, new_vma);
293 if (err < 0) {
294 move_page_tables(new_vma, new_addr, vma,
295 old_addr, moved_len, true);
296 return err;
297 }
Al Virob2edffd2015-04-06 17:48:54 -0400298 }
Laurent Dufour4abad2c2015-06-24 16:56:19 -0700299 arch_remap(mm, old_addr, old_addr + old_len,
300 new_addr, new_addr + new_len);
Al Virob2edffd2015-04-06 17:48:54 -0400301 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302
303 /* Conceal VM_ACCOUNT so old reservation is not undone */
304 if (vm_flags & VM_ACCOUNT) {
305 vma->vm_flags &= ~VM_ACCOUNT;
306 excess = vma->vm_end - vma->vm_start - old_len;
307 if (old_addr > vma->vm_start &&
308 old_addr + old_len < vma->vm_end)
309 split = 1;
310 }
311
Kirill Korotaev71799062005-05-16 21:53:18 -0700312 /*
Hugh Dickins365e9c872005-10-29 18:16:18 -0700313 * If we failed to move page tables we still do total_vm increment
314 * since do_munmap() will decrement it by old_len == new_len.
315 *
316 * Since total_vm is about to be raised artificially high for a
317 * moment, we need to restore high watermark afterwards: if stats
318 * are taken meanwhile, total_vm and hiwater_vm appear too high.
319 * If this were a serious issue, we'd add a flag to do_munmap().
Kirill Korotaev71799062005-05-16 21:53:18 -0700320 */
Hugh Dickins365e9c872005-10-29 18:16:18 -0700321 hiwater_vm = mm->hiwater_vm;
Hugh Dickinsab50b8e2005-10-29 18:15:56 -0700322 vm_stat_account(mm, vma->vm_flags, vma->vm_file, new_len>>PAGE_SHIFT);
Kirill Korotaev71799062005-05-16 21:53:18 -0700323
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 if (do_munmap(mm, old_addr, old_len) < 0) {
325 /* OOM: unable to split vma, just get accounts right */
326 vm_unacct_memory(excess >> PAGE_SHIFT);
327 excess = 0;
328 }
Hugh Dickins365e9c872005-10-29 18:16:18 -0700329 mm->hiwater_vm = hiwater_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330
331 /* Restore VM_ACCOUNT if one or two pieces of vma left */
332 if (excess) {
333 vma->vm_flags |= VM_ACCOUNT;
334 if (split)
335 vma->vm_next->vm_flags |= VM_ACCOUNT;
336 }
337
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 if (vm_flags & VM_LOCKED) {
339 mm->locked_vm += new_len >> PAGE_SHIFT;
Michel Lespinasse81909b82013-02-22 16:32:41 -0800340 *locked = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 }
342
343 return new_addr;
344}
345
Al Viro54f5de72009-11-24 07:17:46 -0500346static struct vm_area_struct *vma_to_resize(unsigned long addr,
347 unsigned long old_len, unsigned long new_len, unsigned long *p)
348{
349 struct mm_struct *mm = current->mm;
350 struct vm_area_struct *vma = find_vma(mm, addr);
351
352 if (!vma || vma->vm_start > addr)
Derek6cd576132015-04-15 16:14:02 -0700353 return ERR_PTR(-EFAULT);
Al Viro54f5de72009-11-24 07:17:46 -0500354
355 if (is_vm_hugetlb_page(vma))
Derek6cd576132015-04-15 16:14:02 -0700356 return ERR_PTR(-EINVAL);
Al Viro54f5de72009-11-24 07:17:46 -0500357
358 /* We can't remap across vm area boundaries */
359 if (old_len > vma->vm_end - addr)
Derek6cd576132015-04-15 16:14:02 -0700360 return ERR_PTR(-EFAULT);
Al Viro54f5de72009-11-24 07:17:46 -0500361
Linus Torvalds982134b2011-04-07 07:35:50 -0700362 /* Need to be careful about a growing mapping */
363 if (new_len > old_len) {
364 unsigned long pgoff;
365
366 if (vma->vm_flags & (VM_DONTEXPAND | VM_PFNMAP))
Derek6cd576132015-04-15 16:14:02 -0700367 return ERR_PTR(-EFAULT);
Linus Torvalds982134b2011-04-07 07:35:50 -0700368 pgoff = (addr - vma->vm_start) >> PAGE_SHIFT;
369 pgoff += vma->vm_pgoff;
370 if (pgoff + (new_len >> PAGE_SHIFT) < pgoff)
Derek6cd576132015-04-15 16:14:02 -0700371 return ERR_PTR(-EINVAL);
Al Viro54f5de72009-11-24 07:17:46 -0500372 }
373
374 if (vma->vm_flags & VM_LOCKED) {
375 unsigned long locked, lock_limit;
376 locked = mm->locked_vm << PAGE_SHIFT;
Jiri Slaby59e99e52010-03-05 13:41:44 -0800377 lock_limit = rlimit(RLIMIT_MEMLOCK);
Al Viro54f5de72009-11-24 07:17:46 -0500378 locked += new_len - old_len;
379 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
Derek6cd576132015-04-15 16:14:02 -0700380 return ERR_PTR(-EAGAIN);
Al Viro54f5de72009-11-24 07:17:46 -0500381 }
382
383 if (!may_expand_vm(mm, (new_len - old_len) >> PAGE_SHIFT))
Derek6cd576132015-04-15 16:14:02 -0700384 return ERR_PTR(-ENOMEM);
Al Viro54f5de72009-11-24 07:17:46 -0500385
386 if (vma->vm_flags & VM_ACCOUNT) {
387 unsigned long charged = (new_len - old_len) >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +0000388 if (security_vm_enough_memory_mm(mm, charged))
Derek6cd576132015-04-15 16:14:02 -0700389 return ERR_PTR(-ENOMEM);
Al Viro54f5de72009-11-24 07:17:46 -0500390 *p = charged;
391 }
392
393 return vma;
Al Viro54f5de72009-11-24 07:17:46 -0500394}
395
Michel Lespinasse81909b82013-02-22 16:32:41 -0800396static unsigned long mremap_to(unsigned long addr, unsigned long old_len,
397 unsigned long new_addr, unsigned long new_len, bool *locked)
Al Viroecc1a892009-11-24 07:28:07 -0500398{
399 struct mm_struct *mm = current->mm;
400 struct vm_area_struct *vma;
401 unsigned long ret = -EINVAL;
402 unsigned long charged = 0;
Al Viro097eed12009-11-24 08:43:52 -0500403 unsigned long map_flags;
Al Viroecc1a892009-11-24 07:28:07 -0500404
405 if (new_addr & ~PAGE_MASK)
406 goto out;
407
408 if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len)
409 goto out;
410
411 /* Check if the location we're moving into overlaps the
412 * old location at all, and fail if it does.
413 */
414 if ((new_addr <= addr) && (new_addr+new_len) > addr)
415 goto out;
416
417 if ((addr <= new_addr) && (addr+old_len) > new_addr)
418 goto out;
419
Al Viroecc1a892009-11-24 07:28:07 -0500420 ret = do_munmap(mm, new_addr, new_len);
421 if (ret)
422 goto out;
423
424 if (old_len >= new_len) {
425 ret = do_munmap(mm, addr+new_len, old_len - new_len);
426 if (ret && old_len != new_len)
427 goto out;
428 old_len = new_len;
429 }
430
431 vma = vma_to_resize(addr, old_len, new_len, &charged);
432 if (IS_ERR(vma)) {
433 ret = PTR_ERR(vma);
434 goto out;
435 }
436
Al Viro097eed12009-11-24 08:43:52 -0500437 map_flags = MAP_FIXED;
438 if (vma->vm_flags & VM_MAYSHARE)
439 map_flags |= MAP_SHARED;
Al Viro9206de92009-12-03 15:23:11 -0500440
Al Viro097eed12009-11-24 08:43:52 -0500441 ret = get_unmapped_area(vma->vm_file, new_addr, new_len, vma->vm_pgoff +
442 ((addr - vma->vm_start) >> PAGE_SHIFT),
443 map_flags);
Al Viroecc1a892009-11-24 07:28:07 -0500444 if (ret & ~PAGE_MASK)
Al Viro097eed12009-11-24 08:43:52 -0500445 goto out1;
446
Michel Lespinasse81909b82013-02-22 16:32:41 -0800447 ret = move_vma(vma, addr, old_len, new_len, new_addr, locked);
Al Viro097eed12009-11-24 08:43:52 -0500448 if (!(ret & ~PAGE_MASK))
449 goto out;
450out1:
451 vm_unacct_memory(charged);
Al Viroecc1a892009-11-24 07:28:07 -0500452
453out:
454 return ret;
455}
456
Al Viro1a0ef852009-11-24 07:43:18 -0500457static int vma_expandable(struct vm_area_struct *vma, unsigned long delta)
458{
Al Virof106af42009-11-24 08:25:18 -0500459 unsigned long end = vma->vm_end + delta;
Al Viro9206de92009-12-03 15:23:11 -0500460 if (end < vma->vm_end) /* overflow */
Al Viro1a0ef852009-11-24 07:43:18 -0500461 return 0;
Al Viro9206de92009-12-03 15:23:11 -0500462 if (vma->vm_next && vma->vm_next->vm_start < end) /* intersection */
Al Virof106af42009-11-24 08:25:18 -0500463 return 0;
464 if (get_unmapped_area(NULL, vma->vm_start, end - vma->vm_start,
465 0, MAP_FIXED) & ~PAGE_MASK)
466 return 0;
Al Viro1a0ef852009-11-24 07:43:18 -0500467 return 1;
468}
469
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470/*
471 * Expand (or shrink) an existing mapping, potentially moving it at the
472 * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
473 *
474 * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
475 * This option implies MREMAP_MAYMOVE.
476 */
Al Viro63a81db2012-05-30 11:32:04 -0400477SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
478 unsigned long, new_len, unsigned long, flags,
479 unsigned long, new_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480{
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700481 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 struct vm_area_struct *vma;
483 unsigned long ret = -EINVAL;
484 unsigned long charged = 0;
Michel Lespinasse81909b82013-02-22 16:32:41 -0800485 bool locked = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
487 if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE))
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700488 return ret;
489
490 if (flags & MREMAP_FIXED && !(flags & MREMAP_MAYMOVE))
491 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
493 if (addr & ~PAGE_MASK)
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700494 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
496 old_len = PAGE_ALIGN(old_len);
497 new_len = PAGE_ALIGN(new_len);
498
499 /*
500 * We allow a zero old-len as a special case
501 * for DOS-emu "duplicate shm area" thing. But
502 * a zero new-len is nonsensical.
503 */
504 if (!new_len)
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700505 return ret;
506
507 down_write(&current->mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 if (flags & MREMAP_FIXED) {
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700510 ret = mremap_to(addr, old_len, new_addr, new_len,
511 &locked);
Al Viroecc1a892009-11-24 07:28:07 -0500512 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 }
514
515 /*
516 * Always allow a shrinking remap: that just unmaps
517 * the unnecessary pages..
518 * do_munmap does all the needed commit accounting
519 */
520 if (old_len >= new_len) {
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700521 ret = do_munmap(mm, addr+new_len, old_len - new_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 if (ret && old_len != new_len)
523 goto out;
524 ret = addr;
Al Viroecc1a892009-11-24 07:28:07 -0500525 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 }
527
528 /*
Al Viroecc1a892009-11-24 07:28:07 -0500529 * Ok, we need to grow..
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 */
Al Viro54f5de72009-11-24 07:17:46 -0500531 vma = vma_to_resize(addr, old_len, new_len, &charged);
532 if (IS_ERR(vma)) {
533 ret = PTR_ERR(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 }
536
537 /* old_len exactly to the end of the area..
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 */
Al Viroecc1a892009-11-24 07:28:07 -0500539 if (old_len == vma->vm_end - addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 /* can we just expand the current mapping? */
Al Viro1a0ef852009-11-24 07:43:18 -0500541 if (vma_expandable(vma, new_len - old_len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 int pages = (new_len - old_len) >> PAGE_SHIFT;
543
Rik van Riel5beb4932010-03-05 13:42:07 -0800544 if (vma_adjust(vma, vma->vm_start, addr + new_len,
545 vma->vm_pgoff, NULL)) {
546 ret = -ENOMEM;
547 goto out;
548 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700550 vm_stat_account(mm, vma->vm_flags, vma->vm_file, pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 if (vma->vm_flags & VM_LOCKED) {
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700552 mm->locked_vm += pages;
Michel Lespinasse81909b82013-02-22 16:32:41 -0800553 locked = true;
554 new_addr = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 }
556 ret = addr;
557 goto out;
558 }
559 }
560
561 /*
562 * We weren't able to just expand or shrink the area,
563 * we need to create a new one and move it..
564 */
565 ret = -ENOMEM;
566 if (flags & MREMAP_MAYMOVE) {
Al Viroecc1a892009-11-24 07:28:07 -0500567 unsigned long map_flags = 0;
568 if (vma->vm_flags & VM_MAYSHARE)
569 map_flags |= MAP_SHARED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
Al Viroecc1a892009-11-24 07:28:07 -0500571 new_addr = get_unmapped_area(vma->vm_file, 0, new_len,
Al Viro93587412009-11-24 08:45:24 -0500572 vma->vm_pgoff +
573 ((addr - vma->vm_start) >> PAGE_SHIFT),
574 map_flags);
Al Viroecc1a892009-11-24 07:28:07 -0500575 if (new_addr & ~PAGE_MASK) {
576 ret = new_addr;
577 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 }
Al Viroecc1a892009-11-24 07:28:07 -0500579
Michel Lespinasse81909b82013-02-22 16:32:41 -0800580 ret = move_vma(vma, addr, old_len, new_len, new_addr, &locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 }
582out:
583 if (ret & ~PAGE_MASK)
584 vm_unacct_memory(charged);
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}