blob: 049470aa1e3eefc88407e9a35f1ca252fd01912d [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * mm/mremap.c
4 *
5 * (C) Copyright 1996 Linus Torvalds
6 *
Alan Cox046c6882009-01-05 14:06:29 +00007 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * (C) Copyright 2002 Red Hat Inc, All Rights Reserved
9 */
10
11#include <linux/mm.h>
12#include <linux/hugetlb.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/shm.h>
Hugh Dickins1ff82992009-09-21 17:02:05 -070014#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/mman.h>
16#include <linux/swap.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080017#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/fs.h>
Cyrill Gorcunov6dec97d2013-08-27 12:37:18 +040019#include <linux/swapops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/highmem.h>
21#include <linux/security.h>
22#include <linux/syscalls.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070023#include <linux/mmu_notifier.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>
Pavel Emelyanov72f87652017-02-22 15:42:34 -080026#include <linux/userfaultfd_k.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <asm/cacheflush.h>
29#include <asm/tlbflush.h>
30
Rik van Rielba470de2008-10-18 20:26:50 -070031#include "internal.h"
32
Hugh Dickins7be7a542005-10-29 18:16:00 -070033static pmd_t *get_old_pmd(struct mm_struct *mm, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070034{
35 pgd_t *pgd;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +030036 p4d_t *p4d;
Linus Torvalds1da177e2005-04-16 15:20:36 -070037 pud_t *pud;
38 pmd_t *pmd;
39
40 pgd = pgd_offset(mm, addr);
41 if (pgd_none_or_clear_bad(pgd))
42 return NULL;
43
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +030044 p4d = p4d_offset(pgd, addr);
45 if (p4d_none_or_clear_bad(p4d))
46 return NULL;
47
48 pud = pud_offset(p4d, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -070049 if (pud_none_or_clear_bad(pud))
50 return NULL;
51
52 pmd = pmd_offset(pud, addr);
Andrea Arcangeli37a1c492011-10-31 17:08:30 -070053 if (pmd_none(*pmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -070054 return NULL;
55
Hugh Dickins7be7a542005-10-29 18:16:00 -070056 return pmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -070057}
58
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -080059static pmd_t *alloc_new_pmd(struct mm_struct *mm, struct vm_area_struct *vma,
60 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070061{
62 pgd_t *pgd;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +030063 p4d_t *p4d;
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 pud_t *pud;
Hugh Dickinsc74df322005-10-29 18:16:23 -070065 pmd_t *pmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
67 pgd = pgd_offset(mm, addr);
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +030068 p4d = p4d_alloc(mm, pgd, addr);
69 if (!p4d)
70 return NULL;
71 pud = pud_alloc(mm, p4d, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 if (!pud)
Hugh Dickinsc74df322005-10-29 18:16:23 -070073 return NULL;
Hugh Dickins7be7a542005-10-29 18:16:00 -070074
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 pmd = pmd_alloc(mm, pud, addr);
Hugh Dickins57a8f0c2013-10-16 13:47:09 -070076 if (!pmd)
Hugh Dickinsc74df322005-10-29 18:16:23 -070077 return NULL;
Hugh Dickins7be7a542005-10-29 18:16:00 -070078
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -080079 VM_BUG_ON(pmd_trans_huge(*pmd));
Hugh Dickinsc74df322005-10-29 18:16:23 -070080
Hugh Dickins7be7a542005-10-29 18:16:00 -070081 return pmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -070082}
83
Hugh Dickins1d069b72016-05-19 17:12:57 -070084static void take_rmap_locks(struct vm_area_struct *vma)
85{
86 if (vma->vm_file)
87 i_mmap_lock_write(vma->vm_file->f_mapping);
88 if (vma->anon_vma)
89 anon_vma_lock_write(vma->anon_vma);
90}
91
92static void drop_rmap_locks(struct vm_area_struct *vma)
93{
94 if (vma->anon_vma)
95 anon_vma_unlock_write(vma->anon_vma);
96 if (vma->vm_file)
97 i_mmap_unlock_write(vma->vm_file->f_mapping);
98}
99
Cyrill Gorcunov6dec97d2013-08-27 12:37:18 +0400100static pte_t move_soft_dirty_pte(pte_t pte)
101{
102 /*
103 * Set soft dirty bit so we can notice
104 * in userspace the ptes were moved.
105 */
106#ifdef CONFIG_MEM_SOFT_DIRTY
107 if (pte_present(pte))
108 pte = pte_mksoft_dirty(pte);
109 else if (is_swap_pte(pte))
110 pte = pte_swp_mksoft_dirty(pte);
Cyrill Gorcunov6dec97d2013-08-27 12:37:18 +0400111#endif
112 return pte;
113}
114
Hugh Dickins7be7a542005-10-29 18:16:00 -0700115static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd,
116 unsigned long old_addr, unsigned long old_end,
117 struct vm_area_struct *new_vma, pmd_t *new_pmd,
Aaron Lu5d190422016-11-10 17:16:33 +0800118 unsigned long new_addr, bool need_rmap_locks, bool *need_flush)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 struct mm_struct *mm = vma->vm_mm;
Hugh Dickins7be7a542005-10-29 18:16:00 -0700121 pte_t *old_pte, *new_pte, pte;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700122 spinlock_t *old_ptl, *new_ptl;
Aaron Lu5d190422016-11-10 17:16:33 +0800123 bool force_flush = false;
124 unsigned long len = old_end - old_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125
Michel Lespinasse38a76012012-10-08 16:31:50 -0700126 /*
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800127 * When need_rmap_locks is true, we take the i_mmap_rwsem and anon_vma
Michel Lespinasse38a76012012-10-08 16:31:50 -0700128 * locks to ensure that rmap will always observe either the old or the
129 * new ptes. This is the easiest way to avoid races with
130 * truncate_pagecache(), page migration, etc...
131 *
132 * When need_rmap_locks is false, we use other ways to avoid
133 * such races:
134 *
135 * - During exec() shift_arg_pages(), we use a specially tagged vma
136 * which rmap call sites look for using is_vma_temporary_stack().
137 *
138 * - During mremap(), new_vma is often known to be placed after vma
139 * in rmap traversal order. This ensures rmap will always observe
140 * either the old pte, or the new pte, or both (the page table locks
141 * serialize access to individual ptes, but only rmap traversal
142 * order guarantees that we won't miss both the old and new ptes).
143 */
Hugh Dickins1d069b72016-05-19 17:12:57 -0700144 if (need_rmap_locks)
145 take_rmap_locks(vma);
Hugh Dickins7be7a542005-10-29 18:16:00 -0700146
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700147 /*
148 * We don't have to worry about the ordering of src and dst
149 * pte locks because exclusive mmap_sem prevents deadlock.
150 */
Hugh Dickinsc74df322005-10-29 18:16:23 -0700151 old_pte = pte_offset_map_lock(mm, old_pmd, old_addr, &old_ptl);
Peter Zijlstraece0e2b2010-10-26 14:21:52 -0700152 new_pte = pte_offset_map(new_pmd, new_addr);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700153 new_ptl = pte_lockptr(mm, new_pmd);
154 if (new_ptl != old_ptl)
Ingo Molnarf20dc5f2006-07-03 00:25:08 -0700155 spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
Mel Gorman3ea27712017-08-02 13:31:52 -0700156 flush_tlb_batched_pending(vma->vm_mm);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700157 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158
Hugh Dickins7be7a542005-10-29 18:16:00 -0700159 for (; old_addr < old_end; old_pte++, old_addr += PAGE_SIZE,
160 new_pte++, new_addr += PAGE_SIZE) {
161 if (pte_none(*old_pte))
162 continue;
Aaron Lu5d190422016-11-10 17:16:33 +0800163
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700164 pte = ptep_get_and_clear(mm, old_addr, old_pte);
Aaron Lua2ce2662016-11-29 13:27:31 +0800165 /*
166 * If we are remapping a dirty PTE, make sure
167 * to flush TLB before we drop the PTL for the
168 * old PTE or we may race with page_mkclean().
169 *
170 * This check has to be done after we removed the
171 * old PTE from page tables or another thread may
172 * dirty it after the check and before the removal.
173 */
174 if (pte_present(pte) && pte_dirty(pte))
175 force_flush = true;
Hugh Dickins7be7a542005-10-29 18:16:00 -0700176 pte = move_pte(pte, new_vma->vm_page_prot, old_addr, new_addr);
Cyrill Gorcunov6dec97d2013-08-27 12:37:18 +0400177 pte = move_soft_dirty_pte(pte);
178 set_pte_at(mm, new_addr, new_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 }
Hugh Dickins7be7a542005-10-29 18:16:00 -0700180
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700181 arch_leave_lazy_mmu_mode();
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700182 if (new_ptl != old_ptl)
183 spin_unlock(new_ptl);
Peter Zijlstraece0e2b2010-10-26 14:21:52 -0700184 pte_unmap(new_pte - 1);
Aaron Lu5d190422016-11-10 17:16:33 +0800185 if (force_flush)
186 flush_tlb_range(vma, old_end - len, old_end);
187 else
188 *need_flush = true;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700189 pte_unmap_unlock(old_pte - 1, old_ptl);
Hugh Dickins1d069b72016-05-19 17:12:57 -0700190 if (need_rmap_locks)
191 drop_rmap_locks(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192}
193
Hugh Dickins7be7a542005-10-29 18:16:00 -0700194#define LATENCY_LIMIT (64 * PAGE_SIZE)
195
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700196unsigned long move_page_tables(struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 unsigned long old_addr, struct vm_area_struct *new_vma,
Michel Lespinasse38a76012012-10-08 16:31:50 -0700198 unsigned long new_addr, unsigned long len,
199 bool need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200{
Hugh Dickins7be7a542005-10-29 18:16:00 -0700201 unsigned long extent, next, old_end;
202 pmd_t *old_pmd, *new_pmd;
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700203 bool need_flush = false;
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700204 unsigned long mmun_start; /* For mmu_notifiers */
205 unsigned long mmun_end; /* For mmu_notifiers */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206
Hugh Dickins7be7a542005-10-29 18:16:00 -0700207 old_end = old_addr + len;
208 flush_cache_range(vma, old_addr, old_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700210 mmun_start = old_addr;
211 mmun_end = old_end;
212 mmu_notifier_invalidate_range_start(vma->vm_mm, mmun_start, mmun_end);
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700213
Hugh Dickins7be7a542005-10-29 18:16:00 -0700214 for (; old_addr < old_end; old_addr += extent, new_addr += extent) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 cond_resched();
Hugh Dickins7be7a542005-10-29 18:16:00 -0700216 next = (old_addr + PMD_SIZE) & PMD_MASK;
Andrea Arcangeliebed4842011-10-31 17:08:22 -0700217 /* even if next overflowed, extent below will be ok */
Hugh Dickins7be7a542005-10-29 18:16:00 -0700218 extent = next - old_addr;
Andrea Arcangeliebed4842011-10-31 17:08:22 -0700219 if (extent > old_end - old_addr)
220 extent = old_end - old_addr;
Hugh Dickins7be7a542005-10-29 18:16:00 -0700221 old_pmd = get_old_pmd(vma->vm_mm, old_addr);
222 if (!old_pmd)
223 continue;
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800224 new_pmd = alloc_new_pmd(vma->vm_mm, vma, new_addr);
Hugh Dickins7be7a542005-10-29 18:16:00 -0700225 if (!new_pmd)
226 break;
Zi Yan84c3fc42017-09-08 16:11:01 -0700227 if (is_swap_pmd(*old_pmd) || pmd_trans_huge(*old_pmd)) {
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700228 if (extent == HPAGE_PMD_SIZE) {
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800229 bool moved;
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700230 /* See comment in move_ptes() */
231 if (need_rmap_locks)
Hugh Dickins1d069b72016-05-19 17:12:57 -0700232 take_rmap_locks(vma);
Hugh Dickinsbf8616d2016-05-19 17:12:54 -0700233 moved = move_huge_pmd(vma, old_addr, new_addr,
Aaron Lu5d190422016-11-10 17:16:33 +0800234 old_end, old_pmd, new_pmd,
235 &need_flush);
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700236 if (need_rmap_locks)
Hugh Dickins1d069b72016-05-19 17:12:57 -0700237 drop_rmap_locks(vma);
Aaron Lu5d190422016-11-10 17:16:33 +0800238 if (moved)
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800239 continue;
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700240 }
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800241 split_huge_pmd(vma, old_pmd, old_addr);
Naoya Horiguchi337d9ab2016-07-26 15:24:03 -0700242 if (pmd_trans_unstable(old_pmd))
Kirill A. Shutemov6b9116a2016-02-11 16:13:03 -0800243 continue;
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700244 }
Kirill A. Shutemov3ed3a4f2016-03-17 14:19:11 -0700245 if (pte_alloc(new_vma->vm_mm, new_pmd, new_addr))
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700246 break;
Hugh Dickins7be7a542005-10-29 18:16:00 -0700247 next = (new_addr + PMD_SIZE) & PMD_MASK;
248 if (extent > next - new_addr)
249 extent = next - new_addr;
250 if (extent > LATENCY_LIMIT)
251 extent = LATENCY_LIMIT;
Aaron Lu5d190422016-11-10 17:16:33 +0800252 move_ptes(vma, old_pmd, old_addr, old_addr + extent, new_vma,
253 new_pmd, new_addr, need_rmap_locks, &need_flush);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 }
Aaron Lu5d190422016-11-10 17:16:33 +0800255 if (need_flush)
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700256 flush_tlb_range(vma, old_end-len, old_addr);
257
Sagi Grimberg2ec74c32012-10-08 16:33:33 -0700258 mmu_notifier_invalidate_range_end(vma->vm_mm, mmun_start, mmun_end);
Hugh Dickins7be7a542005-10-29 18:16:00 -0700259
260 return len + old_addr - old_end; /* how much done */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261}
262
263static unsigned long move_vma(struct vm_area_struct *vma,
264 unsigned long old_addr, unsigned long old_len,
Pavel Emelyanov72f87652017-02-22 15:42:34 -0800265 unsigned long new_len, unsigned long new_addr,
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800266 bool *locked, struct vm_userfaultfd_ctx *uf,
267 struct list_head *uf_unmap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268{
269 struct mm_struct *mm = vma->vm_mm;
270 struct vm_area_struct *new_vma;
271 unsigned long vm_flags = vma->vm_flags;
272 unsigned long new_pgoff;
273 unsigned long moved_len;
274 unsigned long excess = 0;
Hugh Dickins365e9c872005-10-29 18:16:18 -0700275 unsigned long hiwater_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 int split = 0;
Hugh Dickins7103ad32009-09-21 17:02:28 -0700277 int err;
Michel Lespinasse38a76012012-10-08 16:31:50 -0700278 bool need_rmap_locks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279
280 /*
281 * We'd prefer to avoid failure later on in do_munmap:
282 * which may split one vma into three before unmapping.
283 */
284 if (mm->map_count >= sysctl_max_map_count - 3)
285 return -ENOMEM;
286
Hugh Dickins1ff82992009-09-21 17:02:05 -0700287 /*
288 * Advise KSM to break any KSM pages in the area to be moved:
289 * it would be confusing if they were to turn up at the new
290 * location, where they happen to coincide with different KSM
291 * pages recently unmapped. But leave vma->vm_flags as it was,
292 * so KSM can come around to merge on vma and new_vma afterwards.
293 */
Hugh Dickins7103ad32009-09-21 17:02:28 -0700294 err = ksm_madvise(vma, old_addr, old_addr + old_len,
295 MADV_UNMERGEABLE, &vm_flags);
296 if (err)
297 return err;
Hugh Dickins1ff82992009-09-21 17:02:05 -0700298
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 new_pgoff = vma->vm_pgoff + ((old_addr - vma->vm_start) >> PAGE_SHIFT);
Michel Lespinasse38a76012012-10-08 16:31:50 -0700300 new_vma = copy_vma(&vma, new_addr, new_len, new_pgoff,
301 &need_rmap_locks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 if (!new_vma)
303 return -ENOMEM;
304
Michel Lespinasse38a76012012-10-08 16:31:50 -0700305 moved_len = move_page_tables(vma, old_addr, new_vma, new_addr, old_len,
306 need_rmap_locks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 if (moved_len < old_len) {
Oleg Nesterovdf1eab32015-09-04 15:48:01 -0700308 err = -ENOMEM;
Oleg Nesterov5477e702015-09-04 15:48:04 -0700309 } else if (vma->vm_ops && vma->vm_ops->mremap) {
310 err = vma->vm_ops->mremap(new_vma);
Oleg Nesterovdf1eab32015-09-04 15:48:01 -0700311 }
312
313 if (unlikely(err)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 /*
315 * On error, move entries back from new area to old,
316 * which will succeed since page tables still there,
317 * and then proceed to unmap new area instead of old.
318 */
Michel Lespinasse38a76012012-10-08 16:31:50 -0700319 move_page_tables(new_vma, new_addr, vma, old_addr, moved_len,
320 true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 vma = new_vma;
322 old_len = new_len;
323 old_addr = new_addr;
Oleg Nesterovdf1eab32015-09-04 15:48:01 -0700324 new_addr = err;
Laurent Dufour4abad2c2015-06-24 16:56:19 -0700325 } else {
Pavel Emelyanov72f87652017-02-22 15:42:34 -0800326 mremap_userfaultfd_prep(new_vma, uf);
Laurent Dufour4abad2c2015-06-24 16:56:19 -0700327 arch_remap(mm, old_addr, old_addr + old_len,
328 new_addr, new_addr + new_len);
Al Virob2edffd2015-04-06 17:48:54 -0400329 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330
331 /* Conceal VM_ACCOUNT so old reservation is not undone */
332 if (vm_flags & VM_ACCOUNT) {
333 vma->vm_flags &= ~VM_ACCOUNT;
334 excess = vma->vm_end - vma->vm_start - old_len;
335 if (old_addr > vma->vm_start &&
336 old_addr + old_len < vma->vm_end)
337 split = 1;
338 }
339
Kirill Korotaev71799062005-05-16 21:53:18 -0700340 /*
Hugh Dickins365e9c872005-10-29 18:16:18 -0700341 * If we failed to move page tables we still do total_vm increment
342 * since do_munmap() will decrement it by old_len == new_len.
343 *
344 * Since total_vm is about to be raised artificially high for a
345 * moment, we need to restore high watermark afterwards: if stats
346 * are taken meanwhile, total_vm and hiwater_vm appear too high.
347 * If this were a serious issue, we'd add a flag to do_munmap().
Kirill Korotaev71799062005-05-16 21:53:18 -0700348 */
Hugh Dickins365e9c872005-10-29 18:16:18 -0700349 hiwater_vm = mm->hiwater_vm;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -0800350 vm_stat_account(mm, vma->vm_flags, new_len >> PAGE_SHIFT);
Kirill Korotaev71799062005-05-16 21:53:18 -0700351
Toshi Kanid9fe4fa2015-12-22 17:54:23 -0700352 /* Tell pfnmap has moved from this vma */
353 if (unlikely(vma->vm_flags & VM_PFNMAP))
354 untrack_pfn_moved(vma);
355
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800356 if (do_munmap(mm, old_addr, old_len, uf_unmap) < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 /* OOM: unable to split vma, just get accounts right */
358 vm_unacct_memory(excess >> PAGE_SHIFT);
359 excess = 0;
360 }
Hugh Dickins365e9c872005-10-29 18:16:18 -0700361 mm->hiwater_vm = hiwater_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362
363 /* Restore VM_ACCOUNT if one or two pieces of vma left */
364 if (excess) {
365 vma->vm_flags |= VM_ACCOUNT;
366 if (split)
367 vma->vm_next->vm_flags |= VM_ACCOUNT;
368 }
369
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 if (vm_flags & VM_LOCKED) {
371 mm->locked_vm += new_len >> PAGE_SHIFT;
Michel Lespinasse81909b82013-02-22 16:32:41 -0800372 *locked = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 }
374
375 return new_addr;
376}
377
Al Viro54f5de72009-11-24 07:17:46 -0500378static struct vm_area_struct *vma_to_resize(unsigned long addr,
379 unsigned long old_len, unsigned long new_len, unsigned long *p)
380{
381 struct mm_struct *mm = current->mm;
382 struct vm_area_struct *vma = find_vma(mm, addr);
Oleg Nesterov1d391682015-09-04 15:48:10 -0700383 unsigned long pgoff;
Al Viro54f5de72009-11-24 07:17:46 -0500384
385 if (!vma || vma->vm_start > addr)
Derek6cd576132015-04-15 16:14:02 -0700386 return ERR_PTR(-EFAULT);
Al Viro54f5de72009-11-24 07:17:46 -0500387
Mike Kravetzdba58d32017-09-06 16:20:55 -0700388 /*
389 * !old_len is a special case where an attempt is made to 'duplicate'
390 * a mapping. This makes no sense for private mappings as it will
391 * instead create a fresh/new mapping unrelated to the original. This
392 * is contrary to the basic idea of mremap which creates new mappings
393 * based on the original. There are no known use cases for this
394 * behavior. As a result, fail such attempts.
395 */
396 if (!old_len && !(vma->vm_flags & (VM_SHARED | VM_MAYSHARE))) {
397 pr_warn_once("%s (%d): attempted to duplicate a private mapping with mremap. This is not supported.\n", current->comm, current->pid);
398 return ERR_PTR(-EINVAL);
399 }
400
Al Viro54f5de72009-11-24 07:17:46 -0500401 if (is_vm_hugetlb_page(vma))
Derek6cd576132015-04-15 16:14:02 -0700402 return ERR_PTR(-EINVAL);
Al Viro54f5de72009-11-24 07:17:46 -0500403
404 /* We can't remap across vm area boundaries */
405 if (old_len > vma->vm_end - addr)
Derek6cd576132015-04-15 16:14:02 -0700406 return ERR_PTR(-EFAULT);
Al Viro54f5de72009-11-24 07:17:46 -0500407
Oleg Nesterov1d391682015-09-04 15:48:10 -0700408 if (new_len == old_len)
409 return vma;
Linus Torvalds982134b2011-04-07 07:35:50 -0700410
Oleg Nesterov1d391682015-09-04 15:48:10 -0700411 /* Need to be careful about a growing mapping */
412 pgoff = (addr - vma->vm_start) >> PAGE_SHIFT;
413 pgoff += vma->vm_pgoff;
414 if (pgoff + (new_len >> PAGE_SHIFT) < pgoff)
415 return ERR_PTR(-EINVAL);
416
417 if (vma->vm_flags & (VM_DONTEXPAND | VM_PFNMAP))
418 return ERR_PTR(-EFAULT);
Al Viro54f5de72009-11-24 07:17:46 -0500419
420 if (vma->vm_flags & VM_LOCKED) {
421 unsigned long locked, lock_limit;
422 locked = mm->locked_vm << PAGE_SHIFT;
Jiri Slaby59e99e52010-03-05 13:41:44 -0800423 lock_limit = rlimit(RLIMIT_MEMLOCK);
Al Viro54f5de72009-11-24 07:17:46 -0500424 locked += new_len - old_len;
425 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
Derek6cd576132015-04-15 16:14:02 -0700426 return ERR_PTR(-EAGAIN);
Al Viro54f5de72009-11-24 07:17:46 -0500427 }
428
Konstantin Khlebnikov84638332016-01-14 15:22:07 -0800429 if (!may_expand_vm(mm, vma->vm_flags,
430 (new_len - old_len) >> PAGE_SHIFT))
Derek6cd576132015-04-15 16:14:02 -0700431 return ERR_PTR(-ENOMEM);
Al Viro54f5de72009-11-24 07:17:46 -0500432
433 if (vma->vm_flags & VM_ACCOUNT) {
434 unsigned long charged = (new_len - old_len) >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +0000435 if (security_vm_enough_memory_mm(mm, charged))
Derek6cd576132015-04-15 16:14:02 -0700436 return ERR_PTR(-ENOMEM);
Al Viro54f5de72009-11-24 07:17:46 -0500437 *p = charged;
438 }
439
440 return vma;
Al Viro54f5de72009-11-24 07:17:46 -0500441}
442
Michel Lespinasse81909b82013-02-22 16:32:41 -0800443static unsigned long mremap_to(unsigned long addr, unsigned long old_len,
Pavel Emelyanov72f87652017-02-22 15:42:34 -0800444 unsigned long new_addr, unsigned long new_len, bool *locked,
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800445 struct vm_userfaultfd_ctx *uf,
Mike Rapoportb2282372017-08-02 13:31:55 -0700446 struct list_head *uf_unmap_early,
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800447 struct list_head *uf_unmap)
Al Viroecc1a892009-11-24 07:28:07 -0500448{
449 struct mm_struct *mm = current->mm;
450 struct vm_area_struct *vma;
451 unsigned long ret = -EINVAL;
452 unsigned long charged = 0;
Al Viro097eed12009-11-24 08:43:52 -0500453 unsigned long map_flags;
Al Viroecc1a892009-11-24 07:28:07 -0500454
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800455 if (offset_in_page(new_addr))
Al Viroecc1a892009-11-24 07:28:07 -0500456 goto out;
457
458 if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len)
459 goto out;
460
Oleg Nesterov99432422015-09-04 15:48:13 -0700461 /* Ensure the old/new locations do not overlap */
462 if (addr + old_len > new_addr && new_addr + new_len > addr)
Al Viroecc1a892009-11-24 07:28:07 -0500463 goto out;
464
Mike Rapoportb2282372017-08-02 13:31:55 -0700465 ret = do_munmap(mm, new_addr, new_len, uf_unmap_early);
Al Viroecc1a892009-11-24 07:28:07 -0500466 if (ret)
467 goto out;
468
469 if (old_len >= new_len) {
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800470 ret = do_munmap(mm, addr+new_len, old_len - new_len, uf_unmap);
Al Viroecc1a892009-11-24 07:28:07 -0500471 if (ret && old_len != new_len)
472 goto out;
473 old_len = new_len;
474 }
475
476 vma = vma_to_resize(addr, old_len, new_len, &charged);
477 if (IS_ERR(vma)) {
478 ret = PTR_ERR(vma);
479 goto out;
480 }
481
Al Viro097eed12009-11-24 08:43:52 -0500482 map_flags = MAP_FIXED;
483 if (vma->vm_flags & VM_MAYSHARE)
484 map_flags |= MAP_SHARED;
Al Viro9206de92009-12-03 15:23:11 -0500485
Al Viro097eed12009-11-24 08:43:52 -0500486 ret = get_unmapped_area(vma->vm_file, new_addr, new_len, vma->vm_pgoff +
487 ((addr - vma->vm_start) >> PAGE_SHIFT),
488 map_flags);
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800489 if (offset_in_page(ret))
Al Viro097eed12009-11-24 08:43:52 -0500490 goto out1;
491
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800492 ret = move_vma(vma, addr, old_len, new_len, new_addr, locked, uf,
493 uf_unmap);
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800494 if (!(offset_in_page(ret)))
Al Viro097eed12009-11-24 08:43:52 -0500495 goto out;
496out1:
497 vm_unacct_memory(charged);
Al Viroecc1a892009-11-24 07:28:07 -0500498
499out:
500 return ret;
501}
502
Al Viro1a0ef852009-11-24 07:43:18 -0500503static int vma_expandable(struct vm_area_struct *vma, unsigned long delta)
504{
Al Virof106af42009-11-24 08:25:18 -0500505 unsigned long end = vma->vm_end + delta;
Al Viro9206de92009-12-03 15:23:11 -0500506 if (end < vma->vm_end) /* overflow */
Al Viro1a0ef852009-11-24 07:43:18 -0500507 return 0;
Al Viro9206de92009-12-03 15:23:11 -0500508 if (vma->vm_next && vma->vm_next->vm_start < end) /* intersection */
Al Virof106af42009-11-24 08:25:18 -0500509 return 0;
510 if (get_unmapped_area(NULL, vma->vm_start, end - vma->vm_start,
511 0, MAP_FIXED) & ~PAGE_MASK)
512 return 0;
Al Viro1a0ef852009-11-24 07:43:18 -0500513 return 1;
514}
515
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516/*
517 * Expand (or shrink) an existing mapping, potentially moving it at the
518 * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
519 *
520 * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
521 * This option implies MREMAP_MAYMOVE.
522 */
Al Viro63a81db2012-05-30 11:32:04 -0400523SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
524 unsigned long, new_len, unsigned long, flags,
525 unsigned long, new_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526{
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700527 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 struct vm_area_struct *vma;
529 unsigned long ret = -EINVAL;
530 unsigned long charged = 0;
Michel Lespinasse81909b82013-02-22 16:32:41 -0800531 bool locked = false;
Pavel Emelyanov72f87652017-02-22 15:42:34 -0800532 struct vm_userfaultfd_ctx uf = NULL_VM_UFFD_CTX;
Mike Rapoportb2282372017-08-02 13:31:55 -0700533 LIST_HEAD(uf_unmap_early);
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800534 LIST_HEAD(uf_unmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535
536 if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE))
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700537 return ret;
538
539 if (flags & MREMAP_FIXED && !(flags & MREMAP_MAYMOVE))
540 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800542 if (offset_in_page(addr))
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700543 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544
545 old_len = PAGE_ALIGN(old_len);
546 new_len = PAGE_ALIGN(new_len);
547
548 /*
549 * We allow a zero old-len as a special case
550 * for DOS-emu "duplicate shm area" thing. But
551 * a zero new-len is nonsensical.
552 */
553 if (!new_len)
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700554 return ret;
555
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700556 if (down_write_killable(&current->mm->mmap_sem))
557 return -EINTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 if (flags & MREMAP_FIXED) {
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700560 ret = mremap_to(addr, old_len, new_addr, new_len,
Mike Rapoportb2282372017-08-02 13:31:55 -0700561 &locked, &uf, &uf_unmap_early, &uf_unmap);
Al Viroecc1a892009-11-24 07:28:07 -0500562 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 }
564
565 /*
566 * Always allow a shrinking remap: that just unmaps
567 * the unnecessary pages..
568 * do_munmap does all the needed commit accounting
569 */
570 if (old_len >= new_len) {
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800571 ret = do_munmap(mm, addr+new_len, old_len - new_len, &uf_unmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 if (ret && old_len != new_len)
573 goto out;
574 ret = addr;
Al Viroecc1a892009-11-24 07:28:07 -0500575 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 }
577
578 /*
Al Viroecc1a892009-11-24 07:28:07 -0500579 * Ok, we need to grow..
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 */
Al Viro54f5de72009-11-24 07:17:46 -0500581 vma = vma_to_resize(addr, old_len, new_len, &charged);
582 if (IS_ERR(vma)) {
583 ret = PTR_ERR(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 }
586
587 /* old_len exactly to the end of the area..
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 */
Al Viroecc1a892009-11-24 07:28:07 -0500589 if (old_len == vma->vm_end - addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 /* can we just expand the current mapping? */
Al Viro1a0ef852009-11-24 07:43:18 -0500591 if (vma_expandable(vma, new_len - old_len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 int pages = (new_len - old_len) >> PAGE_SHIFT;
593
Rik van Riel5beb4932010-03-05 13:42:07 -0800594 if (vma_adjust(vma, vma->vm_start, addr + new_len,
595 vma->vm_pgoff, NULL)) {
596 ret = -ENOMEM;
597 goto out;
598 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599
Konstantin Khlebnikov84638332016-01-14 15:22:07 -0800600 vm_stat_account(mm, vma->vm_flags, pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 if (vma->vm_flags & VM_LOCKED) {
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700602 mm->locked_vm += pages;
Michel Lespinasse81909b82013-02-22 16:32:41 -0800603 locked = true;
604 new_addr = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 }
606 ret = addr;
607 goto out;
608 }
609 }
610
611 /*
612 * We weren't able to just expand or shrink the area,
613 * we need to create a new one and move it..
614 */
615 ret = -ENOMEM;
616 if (flags & MREMAP_MAYMOVE) {
Al Viroecc1a892009-11-24 07:28:07 -0500617 unsigned long map_flags = 0;
618 if (vma->vm_flags & VM_MAYSHARE)
619 map_flags |= MAP_SHARED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
Al Viroecc1a892009-11-24 07:28:07 -0500621 new_addr = get_unmapped_area(vma->vm_file, 0, new_len,
Al Viro93587412009-11-24 08:45:24 -0500622 vma->vm_pgoff +
623 ((addr - vma->vm_start) >> PAGE_SHIFT),
624 map_flags);
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800625 if (offset_in_page(new_addr)) {
Al Viroecc1a892009-11-24 07:28:07 -0500626 ret = new_addr;
627 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 }
Al Viroecc1a892009-11-24 07:28:07 -0500629
Pavel Emelyanov72f87652017-02-22 15:42:34 -0800630 ret = move_vma(vma, addr, old_len, new_len, new_addr,
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800631 &locked, &uf, &uf_unmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 }
633out:
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800634 if (offset_in_page(ret)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 vm_unacct_memory(charged);
Oleg Nesterovd456fb92015-09-04 15:48:07 -0700636 locked = 0;
637 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 up_write(&current->mm->mmap_sem);
Michel Lespinasse81909b82013-02-22 16:32:41 -0800639 if (locked && new_len > old_len)
640 mm_populate(new_addr + old_len, new_len - old_len);
Mike Rapoportb2282372017-08-02 13:31:55 -0700641 userfaultfd_unmap_complete(mm, &uf_unmap_early);
Andrea Arcangeli90794bf2017-02-22 15:42:37 -0800642 mremap_userfaultfd_complete(&uf, addr, new_addr, old_len);
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800643 userfaultfd_unmap_complete(mm, &uf_unmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 return ret;
645}