blob: 138abbae4f758763c5be19578b2909d5d003db28 [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,
Linus Torvaldseb66ae02018-10-12 15:22:59 -0700118 unsigned long new_addr, bool need_rmap_locks)
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
Anshuman Khandual222100e2020-04-01 21:07:52 -0700136 * which rmap call sites look for using vma_is_temporary_stack().
Michel Lespinasse38a76012012-10-08 16:31:50 -0700137 *
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
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -0700149 * pte locks because exclusive mmap_lock prevents deadlock.
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700150 */
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 /*
Linus Torvaldseb66ae02018-10-12 15:22:59 -0700166 * If we are remapping a valid PTE, make sure
Aaron Lua2ce2662016-11-29 13:27:31 +0800167 * to flush TLB before we drop the PTL for the
Linus Torvaldseb66ae02018-10-12 15:22:59 -0700168 * PTE.
Aaron Lua2ce2662016-11-29 13:27:31 +0800169 *
Linus Torvaldseb66ae02018-10-12 15:22:59 -0700170 * NOTE! Both old and new PTL matter: the old one
171 * for racing with page_mkclean(), the new one to
172 * make sure the physical page stays valid until
173 * the TLB entry for the old mapping has been
174 * flushed.
Aaron Lua2ce2662016-11-29 13:27:31 +0800175 */
Linus Torvaldseb66ae02018-10-12 15:22:59 -0700176 if (pte_present(pte))
Aaron Lua2ce2662016-11-29 13:27:31 +0800177 force_flush = true;
Hugh Dickins7be7a542005-10-29 18:16:00 -0700178 pte = move_pte(pte, new_vma->vm_page_prot, old_addr, new_addr);
Cyrill Gorcunov6dec97d2013-08-27 12:37:18 +0400179 pte = move_soft_dirty_pte(pte);
180 set_pte_at(mm, new_addr, new_pte, pte);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 }
Hugh Dickins7be7a542005-10-29 18:16:00 -0700182
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700183 arch_leave_lazy_mmu_mode();
Linus Torvaldseb66ae02018-10-12 15:22:59 -0700184 if (force_flush)
185 flush_tlb_range(vma, old_end - len, old_end);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700186 if (new_ptl != old_ptl)
187 spin_unlock(new_ptl);
Peter Zijlstraece0e2b2010-10-26 14:21:52 -0700188 pte_unmap(new_pte - 1);
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
Joel Fernandes (Google)2c91bd42019-01-03 15:28:38 -0800194#ifdef CONFIG_HAVE_MOVE_PMD
195static bool move_normal_pmd(struct vm_area_struct *vma, unsigned long old_addr,
Wei Yangb8aa9d92020-08-06 23:23:40 -0700196 unsigned long new_addr, pmd_t *old_pmd, pmd_t *new_pmd)
Joel Fernandes (Google)2c91bd42019-01-03 15:28:38 -0800197{
198 spinlock_t *old_ptl, *new_ptl;
199 struct mm_struct *mm = vma->vm_mm;
200 pmd_t pmd;
201
Joel Fernandes (Google)2c91bd42019-01-03 15:28:38 -0800202 /*
203 * The destination pmd shouldn't be established, free_pgtables()
Linus Torvaldsf81fdd02020-07-13 11:37:39 -0700204 * should have released it.
205 *
206 * However, there's a case during execve() where we use mremap
207 * to move the initial stack, and in that case the target area
208 * may overlap the source area (always moving down).
209 *
210 * If everything is PMD-aligned, that works fine, as moving
211 * each pmd down will clear the source pmd. But if we first
212 * have a few 4kB-only pages that get moved down, and then
213 * hit the "now the rest is PMD-aligned, let's do everything
214 * one pmd at a time", we will still have the old (now empty
215 * of any 4kB pages, but still there) PMD in the page table
216 * tree.
217 *
218 * Warn on it once - because we really should try to figure
219 * out how to do this better - but then say "I won't move
220 * this pmd".
221 *
222 * One alternative might be to just unmap the target pmd at
223 * this point, and verify that it really is empty. We'll see.
Joel Fernandes (Google)2c91bd42019-01-03 15:28:38 -0800224 */
Linus Torvaldsf81fdd02020-07-13 11:37:39 -0700225 if (WARN_ON_ONCE(!pmd_none(*new_pmd)))
Joel Fernandes (Google)2c91bd42019-01-03 15:28:38 -0800226 return false;
227
228 /*
229 * We don't have to worry about the ordering of src and dst
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -0700230 * ptlocks because exclusive mmap_lock prevents deadlock.
Joel Fernandes (Google)2c91bd42019-01-03 15:28:38 -0800231 */
232 old_ptl = pmd_lock(vma->vm_mm, old_pmd);
233 new_ptl = pmd_lockptr(mm, new_pmd);
234 if (new_ptl != old_ptl)
235 spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
236
237 /* Clear the pmd */
238 pmd = *old_pmd;
239 pmd_clear(old_pmd);
240
241 VM_BUG_ON(!pmd_none(*new_pmd));
242
243 /* Set the new pmd */
244 set_pmd_at(mm, new_addr, new_pmd, pmd);
245 flush_tlb_range(vma, old_addr, old_addr + PMD_SIZE);
246 if (new_ptl != old_ptl)
247 spin_unlock(new_ptl);
248 spin_unlock(old_ptl);
249
250 return true;
251}
252#endif
253
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700254unsigned long move_page_tables(struct vm_area_struct *vma,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 unsigned long old_addr, struct vm_area_struct *new_vma,
Michel Lespinasse38a76012012-10-08 16:31:50 -0700256 unsigned long new_addr, unsigned long len,
257 bool need_rmap_locks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258{
Hugh Dickins7be7a542005-10-29 18:16:00 -0700259 unsigned long extent, next, old_end;
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800260 struct mmu_notifier_range range;
Hugh Dickins7be7a542005-10-29 18:16:00 -0700261 pmd_t *old_pmd, *new_pmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262
Hugh Dickins7be7a542005-10-29 18:16:00 -0700263 old_end = old_addr + len;
264 flush_cache_range(vma, old_addr, old_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
Jérôme Glisse6f4f13e2019-05-13 17:20:49 -0700266 mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0, vma, vma->vm_mm,
267 old_addr, old_end);
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800268 mmu_notifier_invalidate_range_start(&range);
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700269
Hugh Dickins7be7a542005-10-29 18:16:00 -0700270 for (; old_addr < old_end; old_addr += extent, new_addr += extent) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 cond_resched();
Hugh Dickins7be7a542005-10-29 18:16:00 -0700272 next = (old_addr + PMD_SIZE) & PMD_MASK;
Andrea Arcangeliebed4842011-10-31 17:08:22 -0700273 /* even if next overflowed, extent below will be ok */
Hugh Dickins7be7a542005-10-29 18:16:00 -0700274 extent = next - old_addr;
Andrea Arcangeliebed4842011-10-31 17:08:22 -0700275 if (extent > old_end - old_addr)
276 extent = old_end - old_addr;
Wei Yang9ad97182020-08-06 23:23:44 -0700277 next = (new_addr + PMD_SIZE) & PMD_MASK;
278 if (extent > next - new_addr)
279 extent = next - new_addr;
Hugh Dickins7be7a542005-10-29 18:16:00 -0700280 old_pmd = get_old_pmd(vma->vm_mm, old_addr);
281 if (!old_pmd)
282 continue;
Andrea Arcangeli8ac1f832011-01-13 15:46:43 -0800283 new_pmd = alloc_new_pmd(vma->vm_mm, vma, new_addr);
Hugh Dickins7be7a542005-10-29 18:16:00 -0700284 if (!new_pmd)
285 break;
Fan Yang5bfea2d2020-06-04 18:22:07 +0800286 if (is_swap_pmd(*old_pmd) || pmd_trans_huge(*old_pmd) || pmd_devmap(*old_pmd)) {
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700287 if (extent == HPAGE_PMD_SIZE) {
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800288 bool moved;
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700289 /* See comment in move_ptes() */
290 if (need_rmap_locks)
Hugh Dickins1d069b72016-05-19 17:12:57 -0700291 take_rmap_locks(vma);
Hugh Dickinsbf8616d2016-05-19 17:12:54 -0700292 moved = move_huge_pmd(vma, old_addr, new_addr,
Wei Yangb8aa9d92020-08-06 23:23:40 -0700293 old_pmd, new_pmd);
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700294 if (need_rmap_locks)
Hugh Dickins1d069b72016-05-19 17:12:57 -0700295 drop_rmap_locks(vma);
Aaron Lu5d190422016-11-10 17:16:33 +0800296 if (moved)
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800297 continue;
Kirill A. Shutemovdd18dbc2014-05-09 15:37:00 -0700298 }
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -0800299 split_huge_pmd(vma, old_pmd, old_addr);
Naoya Horiguchi337d9ab2016-07-26 15:24:03 -0700300 if (pmd_trans_unstable(old_pmd))
Kirill A. Shutemov6b9116a2016-02-11 16:13:03 -0800301 continue;
Joel Fernandes (Google)2c91bd42019-01-03 15:28:38 -0800302 } else if (extent == PMD_SIZE) {
303#ifdef CONFIG_HAVE_MOVE_PMD
304 /*
305 * If the extent is PMD-sized, try to speed the move by
306 * moving at the PMD level if possible.
307 */
308 bool moved;
309
310 if (need_rmap_locks)
311 take_rmap_locks(vma);
312 moved = move_normal_pmd(vma, old_addr, new_addr,
Wei Yangb8aa9d92020-08-06 23:23:40 -0700313 old_pmd, new_pmd);
Joel Fernandes (Google)2c91bd42019-01-03 15:28:38 -0800314 if (need_rmap_locks)
315 drop_rmap_locks(vma);
316 if (moved)
317 continue;
318#endif
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700319 }
Joel Fernandes (Google)2c91bd42019-01-03 15:28:38 -0800320
Joel Fernandes (Google)4cf58922019-01-03 15:28:34 -0800321 if (pte_alloc(new_vma->vm_mm, new_pmd))
Andrea Arcangeli37a1c492011-10-31 17:08:30 -0700322 break;
Aaron Lu5d190422016-11-10 17:16:33 +0800323 move_ptes(vma, old_pmd, old_addr, old_addr + extent, new_vma,
Linus Torvaldseb66ae02018-10-12 15:22:59 -0700324 new_pmd, new_addr, need_rmap_locks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 }
Andrea Arcangeli7b6efc22011-10-31 17:08:26 -0700326
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800327 mmu_notifier_invalidate_range_end(&range);
Hugh Dickins7be7a542005-10-29 18:16:00 -0700328
329 return len + old_addr - old_end; /* how much done */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330}
331
332static unsigned long move_vma(struct vm_area_struct *vma,
333 unsigned long old_addr, unsigned long old_len,
Pavel Emelyanov72f87652017-02-22 15:42:34 -0800334 unsigned long new_len, unsigned long new_addr,
Brian Geffone346b382020-04-01 21:09:17 -0700335 bool *locked, unsigned long flags,
336 struct vm_userfaultfd_ctx *uf, struct list_head *uf_unmap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337{
338 struct mm_struct *mm = vma->vm_mm;
339 struct vm_area_struct *new_vma;
340 unsigned long vm_flags = vma->vm_flags;
341 unsigned long new_pgoff;
342 unsigned long moved_len;
343 unsigned long excess = 0;
Hugh Dickins365e9c872005-10-29 18:16:18 -0700344 unsigned long hiwater_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 int split = 0;
Hugh Dickins7103ad32009-09-21 17:02:28 -0700346 int err;
Michel Lespinasse38a76012012-10-08 16:31:50 -0700347 bool need_rmap_locks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348
349 /*
350 * We'd prefer to avoid failure later on in do_munmap:
351 * which may split one vma into three before unmapping.
352 */
353 if (mm->map_count >= sysctl_max_map_count - 3)
354 return -ENOMEM;
355
Hugh Dickins1ff82992009-09-21 17:02:05 -0700356 /*
357 * Advise KSM to break any KSM pages in the area to be moved:
358 * it would be confusing if they were to turn up at the new
359 * location, where they happen to coincide with different KSM
360 * pages recently unmapped. But leave vma->vm_flags as it was,
361 * so KSM can come around to merge on vma and new_vma afterwards.
362 */
Hugh Dickins7103ad32009-09-21 17:02:28 -0700363 err = ksm_madvise(vma, old_addr, old_addr + old_len,
364 MADV_UNMERGEABLE, &vm_flags);
365 if (err)
366 return err;
Hugh Dickins1ff82992009-09-21 17:02:05 -0700367
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 new_pgoff = vma->vm_pgoff + ((old_addr - vma->vm_start) >> PAGE_SHIFT);
Michel Lespinasse38a76012012-10-08 16:31:50 -0700369 new_vma = copy_vma(&vma, new_addr, new_len, new_pgoff,
370 &need_rmap_locks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 if (!new_vma)
372 return -ENOMEM;
373
Michel Lespinasse38a76012012-10-08 16:31:50 -0700374 moved_len = move_page_tables(vma, old_addr, new_vma, new_addr, old_len,
375 need_rmap_locks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 if (moved_len < old_len) {
Oleg Nesterovdf1eab32015-09-04 15:48:01 -0700377 err = -ENOMEM;
Oleg Nesterov5477e702015-09-04 15:48:04 -0700378 } else if (vma->vm_ops && vma->vm_ops->mremap) {
379 err = vma->vm_ops->mremap(new_vma);
Oleg Nesterovdf1eab32015-09-04 15:48:01 -0700380 }
381
382 if (unlikely(err)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 /*
384 * On error, move entries back from new area to old,
385 * which will succeed since page tables still there,
386 * and then proceed to unmap new area instead of old.
387 */
Michel Lespinasse38a76012012-10-08 16:31:50 -0700388 move_page_tables(new_vma, new_addr, vma, old_addr, moved_len,
389 true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 vma = new_vma;
391 old_len = new_len;
392 old_addr = new_addr;
Oleg Nesterovdf1eab32015-09-04 15:48:01 -0700393 new_addr = err;
Laurent Dufour4abad2c2015-06-24 16:56:19 -0700394 } else {
Pavel Emelyanov72f87652017-02-22 15:42:34 -0800395 mremap_userfaultfd_prep(new_vma, uf);
Laurent Dufour4abad2c2015-06-24 16:56:19 -0700396 arch_remap(mm, old_addr, old_addr + old_len,
397 new_addr, new_addr + new_len);
Al Virob2edffd2015-04-06 17:48:54 -0400398 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399
400 /* Conceal VM_ACCOUNT so old reservation is not undone */
401 if (vm_flags & VM_ACCOUNT) {
402 vma->vm_flags &= ~VM_ACCOUNT;
403 excess = vma->vm_end - vma->vm_start - old_len;
404 if (old_addr > vma->vm_start &&
405 old_addr + old_len < vma->vm_end)
406 split = 1;
407 }
408
Kirill Korotaev71799062005-05-16 21:53:18 -0700409 /*
Hugh Dickins365e9c872005-10-29 18:16:18 -0700410 * If we failed to move page tables we still do total_vm increment
411 * since do_munmap() will decrement it by old_len == new_len.
412 *
413 * Since total_vm is about to be raised artificially high for a
414 * moment, we need to restore high watermark afterwards: if stats
415 * are taken meanwhile, total_vm and hiwater_vm appear too high.
416 * If this were a serious issue, we'd add a flag to do_munmap().
Kirill Korotaev71799062005-05-16 21:53:18 -0700417 */
Hugh Dickins365e9c872005-10-29 18:16:18 -0700418 hiwater_vm = mm->hiwater_vm;
Konstantin Khlebnikov84638332016-01-14 15:22:07 -0800419 vm_stat_account(mm, vma->vm_flags, new_len >> PAGE_SHIFT);
Kirill Korotaev71799062005-05-16 21:53:18 -0700420
Toshi Kanid9fe4fa2015-12-22 17:54:23 -0700421 /* Tell pfnmap has moved from this vma */
422 if (unlikely(vma->vm_flags & VM_PFNMAP))
423 untrack_pfn_moved(vma);
424
Brian Geffone346b382020-04-01 21:09:17 -0700425 if (unlikely(!err && (flags & MREMAP_DONTUNMAP))) {
426 if (vm_flags & VM_ACCOUNT) {
427 /* Always put back VM_ACCOUNT since we won't unmap */
428 vma->vm_flags |= VM_ACCOUNT;
429
Brian Geffondadbd852020-04-17 10:25:56 -0700430 vm_acct_memory(new_len >> PAGE_SHIFT);
Brian Geffone346b382020-04-01 21:09:17 -0700431 }
432
Brian Geffondadbd852020-04-17 10:25:56 -0700433 /*
434 * VMAs can actually be merged back together in copy_vma
435 * calling merge_vma. This can happen with anonymous vmas
436 * which have not yet been faulted, so if we were to consider
437 * this VMA split we'll end up adding VM_ACCOUNT on the
438 * next VMA, which is completely unrelated if this VMA
439 * was re-merged.
440 */
441 if (split && new_vma == vma)
442 split = 0;
443
Brian Geffone346b382020-04-01 21:09:17 -0700444 /* We always clear VM_LOCKED[ONFAULT] on the old vma */
445 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
446
447 /* Because we won't unmap we don't need to touch locked_vm */
448 goto out;
449 }
450
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800451 if (do_munmap(mm, old_addr, old_len, uf_unmap) < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 /* OOM: unable to split vma, just get accounts right */
453 vm_unacct_memory(excess >> PAGE_SHIFT);
454 excess = 0;
455 }
Brian Geffone346b382020-04-01 21:09:17 -0700456
457 if (vm_flags & VM_LOCKED) {
458 mm->locked_vm += new_len >> PAGE_SHIFT;
459 *locked = true;
460 }
461out:
Hugh Dickins365e9c872005-10-29 18:16:18 -0700462 mm->hiwater_vm = hiwater_vm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463
464 /* Restore VM_ACCOUNT if one or two pieces of vma left */
465 if (excess) {
466 vma->vm_flags |= VM_ACCOUNT;
467 if (split)
468 vma->vm_next->vm_flags |= VM_ACCOUNT;
469 }
470
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 return new_addr;
472}
473
Al Viro54f5de72009-11-24 07:17:46 -0500474static struct vm_area_struct *vma_to_resize(unsigned long addr,
Brian Geffone346b382020-04-01 21:09:17 -0700475 unsigned long old_len, unsigned long new_len, unsigned long flags,
476 unsigned long *p)
Al Viro54f5de72009-11-24 07:17:46 -0500477{
478 struct mm_struct *mm = current->mm;
479 struct vm_area_struct *vma = find_vma(mm, addr);
Oleg Nesterov1d391682015-09-04 15:48:10 -0700480 unsigned long pgoff;
Al Viro54f5de72009-11-24 07:17:46 -0500481
482 if (!vma || vma->vm_start > addr)
Derek6cd576132015-04-15 16:14:02 -0700483 return ERR_PTR(-EFAULT);
Al Viro54f5de72009-11-24 07:17:46 -0500484
Mike Kravetzdba58d32017-09-06 16:20:55 -0700485 /*
486 * !old_len is a special case where an attempt is made to 'duplicate'
487 * a mapping. This makes no sense for private mappings as it will
488 * instead create a fresh/new mapping unrelated to the original. This
489 * is contrary to the basic idea of mremap which creates new mappings
490 * based on the original. There are no known use cases for this
491 * behavior. As a result, fail such attempts.
492 */
493 if (!old_len && !(vma->vm_flags & (VM_SHARED | VM_MAYSHARE))) {
494 pr_warn_once("%s (%d): attempted to duplicate a private mapping with mremap. This is not supported.\n", current->comm, current->pid);
495 return ERR_PTR(-EINVAL);
496 }
497
Brian Geffone346b382020-04-01 21:09:17 -0700498 if (flags & MREMAP_DONTUNMAP && (!vma_is_anonymous(vma) ||
499 vma->vm_flags & VM_SHARED))
500 return ERR_PTR(-EINVAL);
501
Al Viro54f5de72009-11-24 07:17:46 -0500502 if (is_vm_hugetlb_page(vma))
Derek6cd576132015-04-15 16:14:02 -0700503 return ERR_PTR(-EINVAL);
Al Viro54f5de72009-11-24 07:17:46 -0500504
505 /* We can't remap across vm area boundaries */
506 if (old_len > vma->vm_end - addr)
Derek6cd576132015-04-15 16:14:02 -0700507 return ERR_PTR(-EFAULT);
Al Viro54f5de72009-11-24 07:17:46 -0500508
Oleg Nesterov1d391682015-09-04 15:48:10 -0700509 if (new_len == old_len)
510 return vma;
Linus Torvalds982134b2011-04-07 07:35:50 -0700511
Oleg Nesterov1d391682015-09-04 15:48:10 -0700512 /* Need to be careful about a growing mapping */
513 pgoff = (addr - vma->vm_start) >> PAGE_SHIFT;
514 pgoff += vma->vm_pgoff;
515 if (pgoff + (new_len >> PAGE_SHIFT) < pgoff)
516 return ERR_PTR(-EINVAL);
517
518 if (vma->vm_flags & (VM_DONTEXPAND | VM_PFNMAP))
519 return ERR_PTR(-EFAULT);
Al Viro54f5de72009-11-24 07:17:46 -0500520
521 if (vma->vm_flags & VM_LOCKED) {
522 unsigned long locked, lock_limit;
523 locked = mm->locked_vm << PAGE_SHIFT;
Jiri Slaby59e99e52010-03-05 13:41:44 -0800524 lock_limit = rlimit(RLIMIT_MEMLOCK);
Al Viro54f5de72009-11-24 07:17:46 -0500525 locked += new_len - old_len;
526 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
Derek6cd576132015-04-15 16:14:02 -0700527 return ERR_PTR(-EAGAIN);
Al Viro54f5de72009-11-24 07:17:46 -0500528 }
529
Konstantin Khlebnikov84638332016-01-14 15:22:07 -0800530 if (!may_expand_vm(mm, vma->vm_flags,
531 (new_len - old_len) >> PAGE_SHIFT))
Derek6cd576132015-04-15 16:14:02 -0700532 return ERR_PTR(-ENOMEM);
Al Viro54f5de72009-11-24 07:17:46 -0500533
534 if (vma->vm_flags & VM_ACCOUNT) {
535 unsigned long charged = (new_len - old_len) >> PAGE_SHIFT;
Al Viro191c5422012-02-13 03:58:52 +0000536 if (security_vm_enough_memory_mm(mm, charged))
Derek6cd576132015-04-15 16:14:02 -0700537 return ERR_PTR(-ENOMEM);
Al Viro54f5de72009-11-24 07:17:46 -0500538 *p = charged;
539 }
540
541 return vma;
Al Viro54f5de72009-11-24 07:17:46 -0500542}
543
Michel Lespinasse81909b82013-02-22 16:32:41 -0800544static unsigned long mremap_to(unsigned long addr, unsigned long old_len,
Pavel Emelyanov72f87652017-02-22 15:42:34 -0800545 unsigned long new_addr, unsigned long new_len, bool *locked,
Brian Geffone346b382020-04-01 21:09:17 -0700546 unsigned long flags, struct vm_userfaultfd_ctx *uf,
Mike Rapoportb2282372017-08-02 13:31:55 -0700547 struct list_head *uf_unmap_early,
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800548 struct list_head *uf_unmap)
Al Viroecc1a892009-11-24 07:28:07 -0500549{
550 struct mm_struct *mm = current->mm;
551 struct vm_area_struct *vma;
552 unsigned long ret = -EINVAL;
553 unsigned long charged = 0;
Brian Geffone346b382020-04-01 21:09:17 -0700554 unsigned long map_flags = 0;
Al Viroecc1a892009-11-24 07:28:07 -0500555
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800556 if (offset_in_page(new_addr))
Al Viroecc1a892009-11-24 07:28:07 -0500557 goto out;
558
559 if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len)
560 goto out;
561
Oleg Nesterov99432422015-09-04 15:48:13 -0700562 /* Ensure the old/new locations do not overlap */
563 if (addr + old_len > new_addr && new_addr + new_len > addr)
Al Viroecc1a892009-11-24 07:28:07 -0500564 goto out;
565
Oscar Salvadorea2c3f62019-03-05 15:50:14 -0800566 /*
567 * move_vma() need us to stay 4 maps below the threshold, otherwise
568 * it will bail out at the very beginning.
569 * That is a problem if we have already unmaped the regions here
570 * (new_addr, and old_addr), because userspace will not know the
571 * state of the vma's after it gets -ENOMEM.
572 * So, to avoid such scenario we can pre-compute if the whole
573 * operation has high chances to success map-wise.
574 * Worst-scenario case is when both vma's (new_addr and old_addr) get
575 * split in 3 before unmaping it.
576 * That means 2 more maps (1 for each) to the ones we already hold.
577 * Check whether current map count plus 2 still leads us to 4 maps below
578 * the threshold, otherwise return -ENOMEM here to be more safe.
579 */
580 if ((mm->map_count + 2) >= sysctl_max_map_count - 3)
581 return -ENOMEM;
582
Brian Geffone346b382020-04-01 21:09:17 -0700583 if (flags & MREMAP_FIXED) {
584 ret = do_munmap(mm, new_addr, new_len, uf_unmap_early);
585 if (ret)
586 goto out;
587 }
Al Viroecc1a892009-11-24 07:28:07 -0500588
589 if (old_len >= new_len) {
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800590 ret = do_munmap(mm, addr+new_len, old_len - new_len, uf_unmap);
Al Viroecc1a892009-11-24 07:28:07 -0500591 if (ret && old_len != new_len)
592 goto out;
593 old_len = new_len;
594 }
595
Brian Geffone346b382020-04-01 21:09:17 -0700596 vma = vma_to_resize(addr, old_len, new_len, flags, &charged);
Al Viroecc1a892009-11-24 07:28:07 -0500597 if (IS_ERR(vma)) {
598 ret = PTR_ERR(vma);
599 goto out;
600 }
601
Brian Geffone346b382020-04-01 21:09:17 -0700602 /* MREMAP_DONTUNMAP expands by old_len since old_len == new_len */
603 if (flags & MREMAP_DONTUNMAP &&
604 !may_expand_vm(mm, vma->vm_flags, old_len >> PAGE_SHIFT)) {
605 ret = -ENOMEM;
606 goto out;
607 }
608
609 if (flags & MREMAP_FIXED)
610 map_flags |= MAP_FIXED;
611
Al Viro097eed12009-11-24 08:43:52 -0500612 if (vma->vm_flags & VM_MAYSHARE)
613 map_flags |= MAP_SHARED;
Al Viro9206de92009-12-03 15:23:11 -0500614
Al Viro097eed12009-11-24 08:43:52 -0500615 ret = get_unmapped_area(vma->vm_file, new_addr, new_len, vma->vm_pgoff +
616 ((addr - vma->vm_start) >> PAGE_SHIFT),
617 map_flags);
Gaowei Puff68dac2019-11-30 17:51:03 -0800618 if (IS_ERR_VALUE(ret))
Al Viro097eed12009-11-24 08:43:52 -0500619 goto out1;
620
Brian Geffone346b382020-04-01 21:09:17 -0700621 /* We got a new mapping */
622 if (!(flags & MREMAP_FIXED))
623 new_addr = ret;
624
625 ret = move_vma(vma, addr, old_len, new_len, new_addr, locked, flags, uf,
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800626 uf_unmap);
Brian Geffone346b382020-04-01 21:09:17 -0700627
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800628 if (!(offset_in_page(ret)))
Al Viro097eed12009-11-24 08:43:52 -0500629 goto out;
Brian Geffone346b382020-04-01 21:09:17 -0700630
Al Viro097eed12009-11-24 08:43:52 -0500631out1:
632 vm_unacct_memory(charged);
Al Viroecc1a892009-11-24 07:28:07 -0500633
634out:
635 return ret;
636}
637
Al Viro1a0ef852009-11-24 07:43:18 -0500638static int vma_expandable(struct vm_area_struct *vma, unsigned long delta)
639{
Al Virof106af42009-11-24 08:25:18 -0500640 unsigned long end = vma->vm_end + delta;
Al Viro9206de92009-12-03 15:23:11 -0500641 if (end < vma->vm_end) /* overflow */
Al Viro1a0ef852009-11-24 07:43:18 -0500642 return 0;
Al Viro9206de92009-12-03 15:23:11 -0500643 if (vma->vm_next && vma->vm_next->vm_start < end) /* intersection */
Al Virof106af42009-11-24 08:25:18 -0500644 return 0;
645 if (get_unmapped_area(NULL, vma->vm_start, end - vma->vm_start,
646 0, MAP_FIXED) & ~PAGE_MASK)
647 return 0;
Al Viro1a0ef852009-11-24 07:43:18 -0500648 return 1;
649}
650
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651/*
652 * Expand (or shrink) an existing mapping, potentially moving it at the
653 * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
654 *
655 * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
656 * This option implies MREMAP_MAYMOVE.
657 */
Al Viro63a81db2012-05-30 11:32:04 -0400658SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
659 unsigned long, new_len, unsigned long, flags,
660 unsigned long, new_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661{
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700662 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 struct vm_area_struct *vma;
664 unsigned long ret = -EINVAL;
665 unsigned long charged = 0;
Michel Lespinasse81909b82013-02-22 16:32:41 -0800666 bool locked = false;
Yang Shi85a06832018-10-26 15:08:50 -0700667 bool downgraded = false;
Pavel Emelyanov72f87652017-02-22 15:42:34 -0800668 struct vm_userfaultfd_ctx uf = NULL_VM_UFFD_CTX;
Mike Rapoportb2282372017-08-02 13:31:55 -0700669 LIST_HEAD(uf_unmap_early);
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800670 LIST_HEAD(uf_unmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671
Will Deaconb2a84de2020-03-25 11:13:46 +0000672 /*
673 * There is a deliberate asymmetry here: we strip the pointer tag
674 * from the old address but leave the new address alone. This is
675 * for consistency with mmap(), where we prevent the creation of
676 * aliasing mappings in userspace by leaving the tag bits of the
677 * mapping address intact. A non-zero tag will cause the subsequent
678 * range checks to reject the address as invalid.
679 *
680 * See Documentation/arm64/tagged-address-abi.rst for more information.
681 */
Andrey Konovalov057d33892019-09-25 16:48:30 -0700682 addr = untagged_addr(addr);
683
Brian Geffone346b382020-04-01 21:09:17 -0700684 if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE | MREMAP_DONTUNMAP))
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700685 return ret;
686
687 if (flags & MREMAP_FIXED && !(flags & MREMAP_MAYMOVE))
688 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689
Brian Geffone346b382020-04-01 21:09:17 -0700690 /*
691 * MREMAP_DONTUNMAP is always a move and it does not allow resizing
692 * in the process.
693 */
694 if (flags & MREMAP_DONTUNMAP &&
695 (!(flags & MREMAP_MAYMOVE) || old_len != new_len))
696 return ret;
697
698
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800699 if (offset_in_page(addr))
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700700 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701
702 old_len = PAGE_ALIGN(old_len);
703 new_len = PAGE_ALIGN(new_len);
704
705 /*
706 * We allow a zero old-len as a special case
707 * for DOS-emu "duplicate shm area" thing. But
708 * a zero new-len is nonsensical.
709 */
710 if (!new_len)
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700711 return ret;
712
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700713 if (mmap_write_lock_killable(current->mm))
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700714 return -EINTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715
Brian Geffone346b382020-04-01 21:09:17 -0700716 if (flags & (MREMAP_FIXED | MREMAP_DONTUNMAP)) {
Rasmus Villemoes9a2458a2013-07-08 15:59:48 -0700717 ret = mremap_to(addr, old_len, new_addr, new_len,
Brian Geffone346b382020-04-01 21:09:17 -0700718 &locked, flags, &uf, &uf_unmap_early,
719 &uf_unmap);
Al Viroecc1a892009-11-24 07:28:07 -0500720 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 }
722
723 /*
724 * Always allow a shrinking remap: that just unmaps
725 * the unnecessary pages..
Yang Shi85a06832018-10-26 15:08:50 -0700726 * __do_munmap does all the needed commit accounting, and
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -0700727 * downgrades mmap_lock to read if so directed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 */
729 if (old_len >= new_len) {
Yang Shi85a06832018-10-26 15:08:50 -0700730 int retval;
731
732 retval = __do_munmap(mm, addr+new_len, old_len - new_len,
733 &uf_unmap, true);
734 if (retval < 0 && old_len != new_len) {
735 ret = retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 goto out;
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -0700737 /* Returning 1 indicates mmap_lock is downgraded to read. */
Yang Shi85a06832018-10-26 15:08:50 -0700738 } else if (retval == 1)
739 downgraded = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 ret = addr;
Al Viroecc1a892009-11-24 07:28:07 -0500741 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 }
743
744 /*
Al Viroecc1a892009-11-24 07:28:07 -0500745 * Ok, we need to grow..
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 */
Brian Geffone346b382020-04-01 21:09:17 -0700747 vma = vma_to_resize(addr, old_len, new_len, flags, &charged);
Al Viro54f5de72009-11-24 07:17:46 -0500748 if (IS_ERR(vma)) {
749 ret = PTR_ERR(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 }
752
753 /* old_len exactly to the end of the area..
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 */
Al Viroecc1a892009-11-24 07:28:07 -0500755 if (old_len == vma->vm_end - addr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 /* can we just expand the current mapping? */
Al Viro1a0ef852009-11-24 07:43:18 -0500757 if (vma_expandable(vma, new_len - old_len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 int pages = (new_len - old_len) >> PAGE_SHIFT;
759
Rik van Riel5beb4932010-03-05 13:42:07 -0800760 if (vma_adjust(vma, vma->vm_start, addr + new_len,
761 vma->vm_pgoff, NULL)) {
762 ret = -ENOMEM;
763 goto out;
764 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
Konstantin Khlebnikov84638332016-01-14 15:22:07 -0800766 vm_stat_account(mm, vma->vm_flags, pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 if (vma->vm_flags & VM_LOCKED) {
Hugh Dickinsd0de32d2005-10-29 18:16:16 -0700768 mm->locked_vm += pages;
Michel Lespinasse81909b82013-02-22 16:32:41 -0800769 locked = true;
770 new_addr = addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 }
772 ret = addr;
773 goto out;
774 }
775 }
776
777 /*
778 * We weren't able to just expand or shrink the area,
779 * we need to create a new one and move it..
780 */
781 ret = -ENOMEM;
782 if (flags & MREMAP_MAYMOVE) {
Al Viroecc1a892009-11-24 07:28:07 -0500783 unsigned long map_flags = 0;
784 if (vma->vm_flags & VM_MAYSHARE)
785 map_flags |= MAP_SHARED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786
Al Viroecc1a892009-11-24 07:28:07 -0500787 new_addr = get_unmapped_area(vma->vm_file, 0, new_len,
Al Viro93587412009-11-24 08:45:24 -0500788 vma->vm_pgoff +
789 ((addr - vma->vm_start) >> PAGE_SHIFT),
790 map_flags);
Gaowei Puff68dac2019-11-30 17:51:03 -0800791 if (IS_ERR_VALUE(new_addr)) {
Al Viroecc1a892009-11-24 07:28:07 -0500792 ret = new_addr;
793 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 }
Al Viroecc1a892009-11-24 07:28:07 -0500795
Pavel Emelyanov72f87652017-02-22 15:42:34 -0800796 ret = move_vma(vma, addr, old_len, new_len, new_addr,
Brian Geffone346b382020-04-01 21:09:17 -0700797 &locked, flags, &uf, &uf_unmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 }
799out:
Alexander Kuleshovf19cb112015-11-05 18:46:57 -0800800 if (offset_in_page(ret)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 vm_unacct_memory(charged);
Zou Weifa1f68c2020-06-04 16:49:46 -0700802 locked = false;
Oleg Nesterovd456fb92015-09-04 15:48:07 -0700803 }
Yang Shi85a06832018-10-26 15:08:50 -0700804 if (downgraded)
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700805 mmap_read_unlock(current->mm);
Yang Shi85a06832018-10-26 15:08:50 -0700806 else
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700807 mmap_write_unlock(current->mm);
Michel Lespinasse81909b82013-02-22 16:32:41 -0800808 if (locked && new_len > old_len)
809 mm_populate(new_addr + old_len, new_len - old_len);
Mike Rapoportb2282372017-08-02 13:31:55 -0700810 userfaultfd_unmap_complete(mm, &uf_unmap_early);
Brian Geffond15649262020-05-13 17:50:44 -0700811 mremap_userfaultfd_complete(&uf, addr, ret, old_len);
Mike Rapoport897ab3e2017-02-24 14:58:22 -0800812 userfaultfd_unmap_complete(mm, &uf_unmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 return ret;
814}