1b2441318SGreg Kroah-Hartman // SPDX-License-Identifier: GPL-2.0 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * mm/mremap.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * (C) Copyright 1996 Linus Torvalds 61da177e4SLinus Torvalds * 7046c6884SAlan Cox * Address space accounting code <alan@lxorguk.ukuu.org.uk> 81da177e4SLinus Torvalds * (C) Copyright 2002 Red Hat Inc, All Rights Reserved 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds #include <linux/mm.h> 121da177e4SLinus Torvalds #include <linux/hugetlb.h> 131da177e4SLinus Torvalds #include <linux/shm.h> 141ff82995SHugh Dickins #include <linux/ksm.h> 151da177e4SLinus Torvalds #include <linux/mman.h> 161da177e4SLinus Torvalds #include <linux/swap.h> 17c59ede7bSRandy.Dunlap #include <linux/capability.h> 181da177e4SLinus Torvalds #include <linux/fs.h> 196dec97dcSCyrill Gorcunov #include <linux/swapops.h> 201da177e4SLinus Torvalds #include <linux/highmem.h> 211da177e4SLinus Torvalds #include <linux/security.h> 221da177e4SLinus Torvalds #include <linux/syscalls.h> 23cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 242581d202SPaul McQuade #include <linux/uaccess.h> 2572f87654SPavel Emelyanov #include <linux/userfaultfd_k.h> 261da177e4SLinus Torvalds 271da177e4SLinus Torvalds #include <asm/cacheflush.h> 281da177e4SLinus Torvalds #include <asm/tlbflush.h> 291da177e4SLinus Torvalds 30ba470de4SRik van Riel #include "internal.h" 31ba470de4SRik van Riel 32c49dd340SKalesh Singh static pud_t *get_old_pud(struct mm_struct *mm, unsigned long addr) 331da177e4SLinus Torvalds { 341da177e4SLinus Torvalds pgd_t *pgd; 35c2febafcSKirill A. Shutemov p4d_t *p4d; 361da177e4SLinus Torvalds pud_t *pud; 371da177e4SLinus Torvalds 381da177e4SLinus Torvalds pgd = pgd_offset(mm, addr); 391da177e4SLinus Torvalds if (pgd_none_or_clear_bad(pgd)) 401da177e4SLinus Torvalds return NULL; 411da177e4SLinus Torvalds 42c2febafcSKirill A. Shutemov p4d = p4d_offset(pgd, addr); 43c2febafcSKirill A. Shutemov if (p4d_none_or_clear_bad(p4d)) 44c2febafcSKirill A. Shutemov return NULL; 45c2febafcSKirill A. Shutemov 46c2febafcSKirill A. Shutemov pud = pud_offset(p4d, addr); 471da177e4SLinus Torvalds if (pud_none_or_clear_bad(pud)) 481da177e4SLinus Torvalds return NULL; 491da177e4SLinus Torvalds 50c49dd340SKalesh Singh return pud; 51c49dd340SKalesh Singh } 52c49dd340SKalesh Singh 53c49dd340SKalesh Singh static pmd_t *get_old_pmd(struct mm_struct *mm, unsigned long addr) 54c49dd340SKalesh Singh { 55c49dd340SKalesh Singh pud_t *pud; 56c49dd340SKalesh Singh pmd_t *pmd; 57c49dd340SKalesh Singh 58c49dd340SKalesh Singh pud = get_old_pud(mm, addr); 59c49dd340SKalesh Singh if (!pud) 60c49dd340SKalesh Singh return NULL; 61c49dd340SKalesh Singh 621da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 6337a1c49aSAndrea Arcangeli if (pmd_none(*pmd)) 641da177e4SLinus Torvalds return NULL; 651da177e4SLinus Torvalds 667be7a546SHugh Dickins return pmd; 671da177e4SLinus Torvalds } 681da177e4SLinus Torvalds 69c49dd340SKalesh Singh static pud_t *alloc_new_pud(struct mm_struct *mm, struct vm_area_struct *vma, 708ac1f832SAndrea Arcangeli unsigned long addr) 711da177e4SLinus Torvalds { 721da177e4SLinus Torvalds pgd_t *pgd; 73c2febafcSKirill A. Shutemov p4d_t *p4d; 741da177e4SLinus Torvalds 751da177e4SLinus Torvalds pgd = pgd_offset(mm, addr); 76c2febafcSKirill A. Shutemov p4d = p4d_alloc(mm, pgd, addr); 77c2febafcSKirill A. Shutemov if (!p4d) 78c2febafcSKirill A. Shutemov return NULL; 79c49dd340SKalesh Singh 80c49dd340SKalesh Singh return pud_alloc(mm, p4d, addr); 81c49dd340SKalesh Singh } 82c49dd340SKalesh Singh 83c49dd340SKalesh Singh static pmd_t *alloc_new_pmd(struct mm_struct *mm, struct vm_area_struct *vma, 84c49dd340SKalesh Singh unsigned long addr) 85c49dd340SKalesh Singh { 86c49dd340SKalesh Singh pud_t *pud; 87c49dd340SKalesh Singh pmd_t *pmd; 88c49dd340SKalesh Singh 89c49dd340SKalesh Singh pud = alloc_new_pud(mm, vma, addr); 901da177e4SLinus Torvalds if (!pud) 91c74df32cSHugh Dickins return NULL; 927be7a546SHugh Dickins 931da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, addr); 9457a8f0cdSHugh Dickins if (!pmd) 95c74df32cSHugh Dickins return NULL; 967be7a546SHugh Dickins 978ac1f832SAndrea Arcangeli VM_BUG_ON(pmd_trans_huge(*pmd)); 98c74df32cSHugh Dickins 997be7a546SHugh Dickins return pmd; 1001da177e4SLinus Torvalds } 1011da177e4SLinus Torvalds 1021d069b7dSHugh Dickins static void take_rmap_locks(struct vm_area_struct *vma) 1031d069b7dSHugh Dickins { 1041d069b7dSHugh Dickins if (vma->vm_file) 1051d069b7dSHugh Dickins i_mmap_lock_write(vma->vm_file->f_mapping); 1061d069b7dSHugh Dickins if (vma->anon_vma) 1071d069b7dSHugh Dickins anon_vma_lock_write(vma->anon_vma); 1081d069b7dSHugh Dickins } 1091d069b7dSHugh Dickins 1101d069b7dSHugh Dickins static void drop_rmap_locks(struct vm_area_struct *vma) 1111d069b7dSHugh Dickins { 1121d069b7dSHugh Dickins if (vma->anon_vma) 1131d069b7dSHugh Dickins anon_vma_unlock_write(vma->anon_vma); 1141d069b7dSHugh Dickins if (vma->vm_file) 1151d069b7dSHugh Dickins i_mmap_unlock_write(vma->vm_file->f_mapping); 1161d069b7dSHugh Dickins } 1171d069b7dSHugh Dickins 1186dec97dcSCyrill Gorcunov static pte_t move_soft_dirty_pte(pte_t pte) 1196dec97dcSCyrill Gorcunov { 1206dec97dcSCyrill Gorcunov /* 1216dec97dcSCyrill Gorcunov * Set soft dirty bit so we can notice 1226dec97dcSCyrill Gorcunov * in userspace the ptes were moved. 1236dec97dcSCyrill Gorcunov */ 1246dec97dcSCyrill Gorcunov #ifdef CONFIG_MEM_SOFT_DIRTY 1256dec97dcSCyrill Gorcunov if (pte_present(pte)) 1266dec97dcSCyrill Gorcunov pte = pte_mksoft_dirty(pte); 1276dec97dcSCyrill Gorcunov else if (is_swap_pte(pte)) 1286dec97dcSCyrill Gorcunov pte = pte_swp_mksoft_dirty(pte); 1296dec97dcSCyrill Gorcunov #endif 1306dec97dcSCyrill Gorcunov return pte; 1316dec97dcSCyrill Gorcunov } 1326dec97dcSCyrill Gorcunov 1337be7a546SHugh Dickins static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd, 1347be7a546SHugh Dickins unsigned long old_addr, unsigned long old_end, 1357be7a546SHugh Dickins struct vm_area_struct *new_vma, pmd_t *new_pmd, 136eb66ae03SLinus Torvalds unsigned long new_addr, bool need_rmap_locks) 1371da177e4SLinus Torvalds { 1381da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 1397be7a546SHugh Dickins pte_t *old_pte, *new_pte, pte; 1404c21e2f2SHugh Dickins spinlock_t *old_ptl, *new_ptl; 1415d190420SAaron Lu bool force_flush = false; 1425d190420SAaron Lu unsigned long len = old_end - old_addr; 1431da177e4SLinus Torvalds 1441da177e4SLinus Torvalds /* 145c8c06efaSDavidlohr Bueso * When need_rmap_locks is true, we take the i_mmap_rwsem and anon_vma 14638a76013SMichel Lespinasse * locks to ensure that rmap will always observe either the old or the 14738a76013SMichel Lespinasse * new ptes. This is the easiest way to avoid races with 14838a76013SMichel Lespinasse * truncate_pagecache(), page migration, etc... 14938a76013SMichel Lespinasse * 15038a76013SMichel Lespinasse * When need_rmap_locks is false, we use other ways to avoid 15138a76013SMichel Lespinasse * such races: 15238a76013SMichel Lespinasse * 15338a76013SMichel Lespinasse * - During exec() shift_arg_pages(), we use a specially tagged vma 154222100eeSAnshuman Khandual * which rmap call sites look for using vma_is_temporary_stack(). 15538a76013SMichel Lespinasse * 15638a76013SMichel Lespinasse * - During mremap(), new_vma is often known to be placed after vma 15738a76013SMichel Lespinasse * in rmap traversal order. This ensures rmap will always observe 15838a76013SMichel Lespinasse * either the old pte, or the new pte, or both (the page table locks 15938a76013SMichel Lespinasse * serialize access to individual ptes, but only rmap traversal 16038a76013SMichel Lespinasse * order guarantees that we won't miss both the old and new ptes). 1611da177e4SLinus Torvalds */ 1621d069b7dSHugh Dickins if (need_rmap_locks) 1631d069b7dSHugh Dickins take_rmap_locks(vma); 1641da177e4SLinus Torvalds 1654c21e2f2SHugh Dickins /* 1664c21e2f2SHugh Dickins * We don't have to worry about the ordering of src and dst 167c1e8d7c6SMichel Lespinasse * pte locks because exclusive mmap_lock prevents deadlock. 1684c21e2f2SHugh Dickins */ 169c74df32cSHugh Dickins old_pte = pte_offset_map_lock(mm, old_pmd, old_addr, &old_ptl); 170ece0e2b6SPeter Zijlstra new_pte = pte_offset_map(new_pmd, new_addr); 1714c21e2f2SHugh Dickins new_ptl = pte_lockptr(mm, new_pmd); 1724c21e2f2SHugh Dickins if (new_ptl != old_ptl) 173f20dc5f7SIngo Molnar spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING); 1743ea27719SMel Gorman flush_tlb_batched_pending(vma->vm_mm); 1756606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 1768b1f3124SNick Piggin 1777be7a546SHugh Dickins for (; old_addr < old_end; old_pte++, old_addr += PAGE_SIZE, 1787be7a546SHugh Dickins new_pte++, new_addr += PAGE_SIZE) { 1797be7a546SHugh Dickins if (pte_none(*old_pte)) 1807be7a546SHugh Dickins continue; 1815d190420SAaron Lu 1827b6efc2bSAndrea Arcangeli pte = ptep_get_and_clear(mm, old_addr, old_pte); 183a2ce2666SAaron Lu /* 184eb66ae03SLinus Torvalds * If we are remapping a valid PTE, make sure 185a2ce2666SAaron Lu * to flush TLB before we drop the PTL for the 186eb66ae03SLinus Torvalds * PTE. 187a2ce2666SAaron Lu * 188eb66ae03SLinus Torvalds * NOTE! Both old and new PTL matter: the old one 189eb66ae03SLinus Torvalds * for racing with page_mkclean(), the new one to 190eb66ae03SLinus Torvalds * make sure the physical page stays valid until 191eb66ae03SLinus Torvalds * the TLB entry for the old mapping has been 192eb66ae03SLinus Torvalds * flushed. 193a2ce2666SAaron Lu */ 194eb66ae03SLinus Torvalds if (pte_present(pte)) 195a2ce2666SAaron Lu force_flush = true; 1967be7a546SHugh Dickins pte = move_pte(pte, new_vma->vm_page_prot, old_addr, new_addr); 1976dec97dcSCyrill Gorcunov pte = move_soft_dirty_pte(pte); 1986dec97dcSCyrill Gorcunov set_pte_at(mm, new_addr, new_pte, pte); 1991da177e4SLinus Torvalds } 2007be7a546SHugh Dickins 2016606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 202eb66ae03SLinus Torvalds if (force_flush) 203eb66ae03SLinus Torvalds flush_tlb_range(vma, old_end - len, old_end); 2044c21e2f2SHugh Dickins if (new_ptl != old_ptl) 2054c21e2f2SHugh Dickins spin_unlock(new_ptl); 206ece0e2b6SPeter Zijlstra pte_unmap(new_pte - 1); 207c74df32cSHugh Dickins pte_unmap_unlock(old_pte - 1, old_ptl); 2081d069b7dSHugh Dickins if (need_rmap_locks) 2091d069b7dSHugh Dickins drop_rmap_locks(vma); 2101da177e4SLinus Torvalds } 2111da177e4SLinus Torvalds 2122c91bd4aSJoel Fernandes (Google) #ifdef CONFIG_HAVE_MOVE_PMD 2132c91bd4aSJoel Fernandes (Google) static bool move_normal_pmd(struct vm_area_struct *vma, unsigned long old_addr, 214b8aa9d9dSWei Yang unsigned long new_addr, pmd_t *old_pmd, pmd_t *new_pmd) 2152c91bd4aSJoel Fernandes (Google) { 2162c91bd4aSJoel Fernandes (Google) spinlock_t *old_ptl, *new_ptl; 2172c91bd4aSJoel Fernandes (Google) struct mm_struct *mm = vma->vm_mm; 2182c91bd4aSJoel Fernandes (Google) pmd_t pmd; 2192c91bd4aSJoel Fernandes (Google) 2202c91bd4aSJoel Fernandes (Google) /* 2212c91bd4aSJoel Fernandes (Google) * The destination pmd shouldn't be established, free_pgtables() 222f81fdd0cSLinus Torvalds * should have released it. 223f81fdd0cSLinus Torvalds * 224f81fdd0cSLinus Torvalds * However, there's a case during execve() where we use mremap 225f81fdd0cSLinus Torvalds * to move the initial stack, and in that case the target area 226f81fdd0cSLinus Torvalds * may overlap the source area (always moving down). 227f81fdd0cSLinus Torvalds * 228f81fdd0cSLinus Torvalds * If everything is PMD-aligned, that works fine, as moving 229f81fdd0cSLinus Torvalds * each pmd down will clear the source pmd. But if we first 230f81fdd0cSLinus Torvalds * have a few 4kB-only pages that get moved down, and then 231f81fdd0cSLinus Torvalds * hit the "now the rest is PMD-aligned, let's do everything 232f81fdd0cSLinus Torvalds * one pmd at a time", we will still have the old (now empty 233f81fdd0cSLinus Torvalds * of any 4kB pages, but still there) PMD in the page table 234f81fdd0cSLinus Torvalds * tree. 235f81fdd0cSLinus Torvalds * 236f81fdd0cSLinus Torvalds * Warn on it once - because we really should try to figure 237f81fdd0cSLinus Torvalds * out how to do this better - but then say "I won't move 238f81fdd0cSLinus Torvalds * this pmd". 239f81fdd0cSLinus Torvalds * 240f81fdd0cSLinus Torvalds * One alternative might be to just unmap the target pmd at 241f81fdd0cSLinus Torvalds * this point, and verify that it really is empty. We'll see. 2422c91bd4aSJoel Fernandes (Google) */ 243f81fdd0cSLinus Torvalds if (WARN_ON_ONCE(!pmd_none(*new_pmd))) 2442c91bd4aSJoel Fernandes (Google) return false; 2452c91bd4aSJoel Fernandes (Google) 2462c91bd4aSJoel Fernandes (Google) /* 2472c91bd4aSJoel Fernandes (Google) * We don't have to worry about the ordering of src and dst 248c1e8d7c6SMichel Lespinasse * ptlocks because exclusive mmap_lock prevents deadlock. 2492c91bd4aSJoel Fernandes (Google) */ 2502c91bd4aSJoel Fernandes (Google) old_ptl = pmd_lock(vma->vm_mm, old_pmd); 2512c91bd4aSJoel Fernandes (Google) new_ptl = pmd_lockptr(mm, new_pmd); 2522c91bd4aSJoel Fernandes (Google) if (new_ptl != old_ptl) 2532c91bd4aSJoel Fernandes (Google) spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING); 2542c91bd4aSJoel Fernandes (Google) 2552c91bd4aSJoel Fernandes (Google) /* Clear the pmd */ 2562c91bd4aSJoel Fernandes (Google) pmd = *old_pmd; 2572c91bd4aSJoel Fernandes (Google) pmd_clear(old_pmd); 2582c91bd4aSJoel Fernandes (Google) 2592c91bd4aSJoel Fernandes (Google) VM_BUG_ON(!pmd_none(*new_pmd)); 2602c91bd4aSJoel Fernandes (Google) 2612c91bd4aSJoel Fernandes (Google) /* Set the new pmd */ 2622c91bd4aSJoel Fernandes (Google) set_pmd_at(mm, new_addr, new_pmd, pmd); 2632c91bd4aSJoel Fernandes (Google) flush_tlb_range(vma, old_addr, old_addr + PMD_SIZE); 2642c91bd4aSJoel Fernandes (Google) if (new_ptl != old_ptl) 2652c91bd4aSJoel Fernandes (Google) spin_unlock(new_ptl); 2662c91bd4aSJoel Fernandes (Google) spin_unlock(old_ptl); 2672c91bd4aSJoel Fernandes (Google) 2682c91bd4aSJoel Fernandes (Google) return true; 2692c91bd4aSJoel Fernandes (Google) } 270c49dd340SKalesh Singh #else 271c49dd340SKalesh Singh static inline bool move_normal_pmd(struct vm_area_struct *vma, 272c49dd340SKalesh Singh unsigned long old_addr, unsigned long new_addr, pmd_t *old_pmd, 273c49dd340SKalesh Singh pmd_t *new_pmd) 274c49dd340SKalesh Singh { 275c49dd340SKalesh Singh return false; 276c49dd340SKalesh Singh } 2772c91bd4aSJoel Fernandes (Google) #endif 2782c91bd4aSJoel Fernandes (Google) 279c49dd340SKalesh Singh #ifdef CONFIG_HAVE_MOVE_PUD 280c49dd340SKalesh Singh static bool move_normal_pud(struct vm_area_struct *vma, unsigned long old_addr, 281c49dd340SKalesh Singh unsigned long new_addr, pud_t *old_pud, pud_t *new_pud) 282c49dd340SKalesh Singh { 283c49dd340SKalesh Singh spinlock_t *old_ptl, *new_ptl; 284c49dd340SKalesh Singh struct mm_struct *mm = vma->vm_mm; 285c49dd340SKalesh Singh pud_t pud; 286c49dd340SKalesh Singh 287c49dd340SKalesh Singh /* 288c49dd340SKalesh Singh * The destination pud shouldn't be established, free_pgtables() 289c49dd340SKalesh Singh * should have released it. 290c49dd340SKalesh Singh */ 291c49dd340SKalesh Singh if (WARN_ON_ONCE(!pud_none(*new_pud))) 292c49dd340SKalesh Singh return false; 293c49dd340SKalesh Singh 294c49dd340SKalesh Singh /* 295c49dd340SKalesh Singh * We don't have to worry about the ordering of src and dst 296c49dd340SKalesh Singh * ptlocks because exclusive mmap_lock prevents deadlock. 297c49dd340SKalesh Singh */ 298c49dd340SKalesh Singh old_ptl = pud_lock(vma->vm_mm, old_pud); 299c49dd340SKalesh Singh new_ptl = pud_lockptr(mm, new_pud); 300c49dd340SKalesh Singh if (new_ptl != old_ptl) 301c49dd340SKalesh Singh spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING); 302c49dd340SKalesh Singh 303c49dd340SKalesh Singh /* Clear the pud */ 304c49dd340SKalesh Singh pud = *old_pud; 305c49dd340SKalesh Singh pud_clear(old_pud); 306c49dd340SKalesh Singh 307c49dd340SKalesh Singh VM_BUG_ON(!pud_none(*new_pud)); 308c49dd340SKalesh Singh 309c49dd340SKalesh Singh /* Set the new pud */ 310c49dd340SKalesh Singh set_pud_at(mm, new_addr, new_pud, pud); 311c49dd340SKalesh Singh flush_tlb_range(vma, old_addr, old_addr + PUD_SIZE); 312c49dd340SKalesh Singh if (new_ptl != old_ptl) 313c49dd340SKalesh Singh spin_unlock(new_ptl); 314c49dd340SKalesh Singh spin_unlock(old_ptl); 315c49dd340SKalesh Singh 316c49dd340SKalesh Singh return true; 317c49dd340SKalesh Singh } 318c49dd340SKalesh Singh #else 319c49dd340SKalesh Singh static inline bool move_normal_pud(struct vm_area_struct *vma, 320c49dd340SKalesh Singh unsigned long old_addr, unsigned long new_addr, pud_t *old_pud, 321c49dd340SKalesh Singh pud_t *new_pud) 322c49dd340SKalesh Singh { 323c49dd340SKalesh Singh return false; 324c49dd340SKalesh Singh } 325c49dd340SKalesh Singh #endif 326c49dd340SKalesh Singh 327c49dd340SKalesh Singh enum pgt_entry { 328c49dd340SKalesh Singh NORMAL_PMD, 329c49dd340SKalesh Singh HPAGE_PMD, 330c49dd340SKalesh Singh NORMAL_PUD, 331c49dd340SKalesh Singh }; 332c49dd340SKalesh Singh 333c49dd340SKalesh Singh /* 334c49dd340SKalesh Singh * Returns an extent of the corresponding size for the pgt_entry specified if 335c49dd340SKalesh Singh * valid. Else returns a smaller extent bounded by the end of the source and 336c49dd340SKalesh Singh * destination pgt_entry. 337c49dd340SKalesh Singh */ 338a30a2909SArnd Bergmann static __always_inline unsigned long get_extent(enum pgt_entry entry, 339a30a2909SArnd Bergmann unsigned long old_addr, unsigned long old_end, 340a30a2909SArnd Bergmann unsigned long new_addr) 341c49dd340SKalesh Singh { 342c49dd340SKalesh Singh unsigned long next, extent, mask, size; 343c49dd340SKalesh Singh 344c49dd340SKalesh Singh switch (entry) { 345c49dd340SKalesh Singh case HPAGE_PMD: 346c49dd340SKalesh Singh case NORMAL_PMD: 347c49dd340SKalesh Singh mask = PMD_MASK; 348c49dd340SKalesh Singh size = PMD_SIZE; 349c49dd340SKalesh Singh break; 350c49dd340SKalesh Singh case NORMAL_PUD: 351c49dd340SKalesh Singh mask = PUD_MASK; 352c49dd340SKalesh Singh size = PUD_SIZE; 353c49dd340SKalesh Singh break; 354c49dd340SKalesh Singh default: 355c49dd340SKalesh Singh BUILD_BUG(); 356c49dd340SKalesh Singh break; 357c49dd340SKalesh Singh } 358c49dd340SKalesh Singh 359c49dd340SKalesh Singh next = (old_addr + size) & mask; 360c49dd340SKalesh Singh /* even if next overflowed, extent below will be ok */ 361e05986eeSKalesh Singh extent = next - old_addr; 362e05986eeSKalesh Singh if (extent > old_end - old_addr) 363e05986eeSKalesh Singh extent = old_end - old_addr; 364c49dd340SKalesh Singh next = (new_addr + size) & mask; 365c49dd340SKalesh Singh if (extent > next - new_addr) 366c49dd340SKalesh Singh extent = next - new_addr; 367c49dd340SKalesh Singh return extent; 368c49dd340SKalesh Singh } 369c49dd340SKalesh Singh 370c49dd340SKalesh Singh /* 371c49dd340SKalesh Singh * Attempts to speedup the move by moving entry at the level corresponding to 372c49dd340SKalesh Singh * pgt_entry. Returns true if the move was successful, else false. 373c49dd340SKalesh Singh */ 374c49dd340SKalesh Singh static bool move_pgt_entry(enum pgt_entry entry, struct vm_area_struct *vma, 375c49dd340SKalesh Singh unsigned long old_addr, unsigned long new_addr, 376c49dd340SKalesh Singh void *old_entry, void *new_entry, bool need_rmap_locks) 377c49dd340SKalesh Singh { 378c49dd340SKalesh Singh bool moved = false; 379c49dd340SKalesh Singh 380c49dd340SKalesh Singh /* See comment in move_ptes() */ 381c49dd340SKalesh Singh if (need_rmap_locks) 382c49dd340SKalesh Singh take_rmap_locks(vma); 383c49dd340SKalesh Singh 384c49dd340SKalesh Singh switch (entry) { 385c49dd340SKalesh Singh case NORMAL_PMD: 386c49dd340SKalesh Singh moved = move_normal_pmd(vma, old_addr, new_addr, old_entry, 387c49dd340SKalesh Singh new_entry); 388c49dd340SKalesh Singh break; 389c49dd340SKalesh Singh case NORMAL_PUD: 390c49dd340SKalesh Singh moved = move_normal_pud(vma, old_addr, new_addr, old_entry, 391c49dd340SKalesh Singh new_entry); 392c49dd340SKalesh Singh break; 393c49dd340SKalesh Singh case HPAGE_PMD: 394c49dd340SKalesh Singh moved = IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && 395c49dd340SKalesh Singh move_huge_pmd(vma, old_addr, new_addr, old_entry, 396c49dd340SKalesh Singh new_entry); 397c49dd340SKalesh Singh break; 398c49dd340SKalesh Singh default: 399c49dd340SKalesh Singh WARN_ON_ONCE(1); 400c49dd340SKalesh Singh break; 401c49dd340SKalesh Singh } 402c49dd340SKalesh Singh 403c49dd340SKalesh Singh if (need_rmap_locks) 404c49dd340SKalesh Singh drop_rmap_locks(vma); 405c49dd340SKalesh Singh 406c49dd340SKalesh Singh return moved; 407c49dd340SKalesh Singh } 408c49dd340SKalesh Singh 409b6a2fea3SOllie Wild unsigned long move_page_tables(struct vm_area_struct *vma, 4101da177e4SLinus Torvalds unsigned long old_addr, struct vm_area_struct *new_vma, 41138a76013SMichel Lespinasse unsigned long new_addr, unsigned long len, 41238a76013SMichel Lespinasse bool need_rmap_locks) 4131da177e4SLinus Torvalds { 414c49dd340SKalesh Singh unsigned long extent, old_end; 415ac46d4f3SJérôme Glisse struct mmu_notifier_range range; 4167be7a546SHugh Dickins pmd_t *old_pmd, *new_pmd; 4171da177e4SLinus Torvalds 4187be7a546SHugh Dickins old_end = old_addr + len; 4197be7a546SHugh Dickins flush_cache_range(vma, old_addr, old_end); 4201da177e4SLinus Torvalds 4216f4f13e8SJérôme Glisse mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0, vma, vma->vm_mm, 4226f4f13e8SJérôme Glisse old_addr, old_end); 423ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_start(&range); 4247b6efc2bSAndrea Arcangeli 4257be7a546SHugh Dickins for (; old_addr < old_end; old_addr += extent, new_addr += extent) { 4261da177e4SLinus Torvalds cond_resched(); 427c49dd340SKalesh Singh /* 428c49dd340SKalesh Singh * If extent is PUD-sized try to speed up the move by moving at the 429c49dd340SKalesh Singh * PUD level if possible. 430c49dd340SKalesh Singh */ 431c49dd340SKalesh Singh extent = get_extent(NORMAL_PUD, old_addr, old_end, new_addr); 432c49dd340SKalesh Singh if (IS_ENABLED(CONFIG_HAVE_MOVE_PUD) && extent == PUD_SIZE) { 433c49dd340SKalesh Singh pud_t *old_pud, *new_pud; 434c49dd340SKalesh Singh 435c49dd340SKalesh Singh old_pud = get_old_pud(vma->vm_mm, old_addr); 436c49dd340SKalesh Singh if (!old_pud) 437c49dd340SKalesh Singh continue; 438c49dd340SKalesh Singh new_pud = alloc_new_pud(vma->vm_mm, vma, new_addr); 439c49dd340SKalesh Singh if (!new_pud) 440c49dd340SKalesh Singh break; 441c49dd340SKalesh Singh if (move_pgt_entry(NORMAL_PUD, vma, old_addr, new_addr, 442c49dd340SKalesh Singh old_pud, new_pud, need_rmap_locks)) 443c49dd340SKalesh Singh continue; 444c49dd340SKalesh Singh } 445c49dd340SKalesh Singh 446c49dd340SKalesh Singh extent = get_extent(NORMAL_PMD, old_addr, old_end, new_addr); 4477be7a546SHugh Dickins old_pmd = get_old_pmd(vma->vm_mm, old_addr); 4487be7a546SHugh Dickins if (!old_pmd) 4497be7a546SHugh Dickins continue; 4508ac1f832SAndrea Arcangeli new_pmd = alloc_new_pmd(vma->vm_mm, vma, new_addr); 4517be7a546SHugh Dickins if (!new_pmd) 4527be7a546SHugh Dickins break; 453c49dd340SKalesh Singh if (is_swap_pmd(*old_pmd) || pmd_trans_huge(*old_pmd) || 454c49dd340SKalesh Singh pmd_devmap(*old_pmd)) { 455c49dd340SKalesh Singh if (extent == HPAGE_PMD_SIZE && 456c49dd340SKalesh Singh move_pgt_entry(HPAGE_PMD, vma, old_addr, new_addr, 457c49dd340SKalesh Singh old_pmd, new_pmd, need_rmap_locks)) 45837a1c49aSAndrea Arcangeli continue; 4594b471e88SKirill A. Shutemov split_huge_pmd(vma, old_pmd, old_addr); 460337d9abfSNaoya Horiguchi if (pmd_trans_unstable(old_pmd)) 4616b9116a6SKirill A. Shutemov continue; 462c49dd340SKalesh Singh } else if (IS_ENABLED(CONFIG_HAVE_MOVE_PMD) && 463c49dd340SKalesh Singh extent == PMD_SIZE) { 4642c91bd4aSJoel Fernandes (Google) /* 4652c91bd4aSJoel Fernandes (Google) * If the extent is PMD-sized, try to speed the move by 4662c91bd4aSJoel Fernandes (Google) * moving at the PMD level if possible. 4672c91bd4aSJoel Fernandes (Google) */ 468c49dd340SKalesh Singh if (move_pgt_entry(NORMAL_PMD, vma, old_addr, new_addr, 469c49dd340SKalesh Singh old_pmd, new_pmd, need_rmap_locks)) 4702c91bd4aSJoel Fernandes (Google) continue; 47137a1c49aSAndrea Arcangeli } 4722c91bd4aSJoel Fernandes (Google) 4734cf58924SJoel Fernandes (Google) if (pte_alloc(new_vma->vm_mm, new_pmd)) 47437a1c49aSAndrea Arcangeli break; 4755d190420SAaron Lu move_ptes(vma, old_pmd, old_addr, old_addr + extent, new_vma, 476eb66ae03SLinus Torvalds new_pmd, new_addr, need_rmap_locks); 4771da177e4SLinus Torvalds } 4787b6efc2bSAndrea Arcangeli 479ac46d4f3SJérôme Glisse mmu_notifier_invalidate_range_end(&range); 4807be7a546SHugh Dickins 4817be7a546SHugh Dickins return len + old_addr - old_end; /* how much done */ 4821da177e4SLinus Torvalds } 4831da177e4SLinus Torvalds 4841da177e4SLinus Torvalds static unsigned long move_vma(struct vm_area_struct *vma, 4851da177e4SLinus Torvalds unsigned long old_addr, unsigned long old_len, 48672f87654SPavel Emelyanov unsigned long new_len, unsigned long new_addr, 487e346b381SBrian Geffon bool *locked, unsigned long flags, 488e346b381SBrian Geffon struct vm_userfaultfd_ctx *uf, struct list_head *uf_unmap) 4891da177e4SLinus Torvalds { 4901da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 4911da177e4SLinus Torvalds struct vm_area_struct *new_vma; 4921da177e4SLinus Torvalds unsigned long vm_flags = vma->vm_flags; 4931da177e4SLinus Torvalds unsigned long new_pgoff; 4941da177e4SLinus Torvalds unsigned long moved_len; 4951da177e4SLinus Torvalds unsigned long excess = 0; 496365e9c87SHugh Dickins unsigned long hiwater_vm; 4971da177e4SLinus Torvalds int split = 0; 49873d5e062SDmitry Safonov int err = 0; 49938a76013SMichel Lespinasse bool need_rmap_locks; 5001da177e4SLinus Torvalds 5011da177e4SLinus Torvalds /* 5021da177e4SLinus Torvalds * We'd prefer to avoid failure later on in do_munmap: 5031da177e4SLinus Torvalds * which may split one vma into three before unmapping. 5041da177e4SLinus Torvalds */ 5051da177e4SLinus Torvalds if (mm->map_count >= sysctl_max_map_count - 3) 5061da177e4SLinus Torvalds return -ENOMEM; 5071da177e4SLinus Torvalds 50873d5e062SDmitry Safonov if (vma->vm_ops && vma->vm_ops->may_split) { 50973d5e062SDmitry Safonov if (vma->vm_start != old_addr) 51073d5e062SDmitry Safonov err = vma->vm_ops->may_split(vma, old_addr); 51173d5e062SDmitry Safonov if (!err && vma->vm_end != old_addr + old_len) 51273d5e062SDmitry Safonov err = vma->vm_ops->may_split(vma, old_addr + old_len); 51373d5e062SDmitry Safonov if (err) 51473d5e062SDmitry Safonov return err; 51573d5e062SDmitry Safonov } 51673d5e062SDmitry Safonov 5171ff82995SHugh Dickins /* 5181ff82995SHugh Dickins * Advise KSM to break any KSM pages in the area to be moved: 5191ff82995SHugh Dickins * it would be confusing if they were to turn up at the new 5201ff82995SHugh Dickins * location, where they happen to coincide with different KSM 5211ff82995SHugh Dickins * pages recently unmapped. But leave vma->vm_flags as it was, 5221ff82995SHugh Dickins * so KSM can come around to merge on vma and new_vma afterwards. 5231ff82995SHugh Dickins */ 5247103ad32SHugh Dickins err = ksm_madvise(vma, old_addr, old_addr + old_len, 5257103ad32SHugh Dickins MADV_UNMERGEABLE, &vm_flags); 5267103ad32SHugh Dickins if (err) 5277103ad32SHugh Dickins return err; 5281ff82995SHugh Dickins 529ad8ee77eSDmitry Safonov if (unlikely(flags & MREMAP_DONTUNMAP && vm_flags & VM_ACCOUNT)) { 530ad8ee77eSDmitry Safonov if (security_vm_enough_memory_mm(mm, new_len >> PAGE_SHIFT)) 531ad8ee77eSDmitry Safonov return -ENOMEM; 532ad8ee77eSDmitry Safonov } 533ad8ee77eSDmitry Safonov 5341da177e4SLinus Torvalds new_pgoff = vma->vm_pgoff + ((old_addr - vma->vm_start) >> PAGE_SHIFT); 53538a76013SMichel Lespinasse new_vma = copy_vma(&vma, new_addr, new_len, new_pgoff, 53638a76013SMichel Lespinasse &need_rmap_locks); 537ad8ee77eSDmitry Safonov if (!new_vma) { 538ad8ee77eSDmitry Safonov if (unlikely(flags & MREMAP_DONTUNMAP && vm_flags & VM_ACCOUNT)) 539ad8ee77eSDmitry Safonov vm_unacct_memory(new_len >> PAGE_SHIFT); 5401da177e4SLinus Torvalds return -ENOMEM; 541ad8ee77eSDmitry Safonov } 5421da177e4SLinus Torvalds 54338a76013SMichel Lespinasse moved_len = move_page_tables(vma, old_addr, new_vma, new_addr, old_len, 54438a76013SMichel Lespinasse need_rmap_locks); 5451da177e4SLinus Torvalds if (moved_len < old_len) { 546df1eab30SOleg Nesterov err = -ENOMEM; 5475477e70aSOleg Nesterov } else if (vma->vm_ops && vma->vm_ops->mremap) { 54814d07113SBrian Geffon err = vma->vm_ops->mremap(new_vma); 549df1eab30SOleg Nesterov } 550df1eab30SOleg Nesterov 551df1eab30SOleg Nesterov if (unlikely(err)) { 5521da177e4SLinus Torvalds /* 5531da177e4SLinus Torvalds * On error, move entries back from new area to old, 5541da177e4SLinus Torvalds * which will succeed since page tables still there, 5551da177e4SLinus Torvalds * and then proceed to unmap new area instead of old. 5561da177e4SLinus Torvalds */ 55738a76013SMichel Lespinasse move_page_tables(new_vma, new_addr, vma, old_addr, moved_len, 55838a76013SMichel Lespinasse true); 5591da177e4SLinus Torvalds vma = new_vma; 5601da177e4SLinus Torvalds old_len = new_len; 5611da177e4SLinus Torvalds old_addr = new_addr; 562df1eab30SOleg Nesterov new_addr = err; 5634abad2caSLaurent Dufour } else { 56472f87654SPavel Emelyanov mremap_userfaultfd_prep(new_vma, uf); 5654abad2caSLaurent Dufour } 5661da177e4SLinus Torvalds 5671da177e4SLinus Torvalds /* Conceal VM_ACCOUNT so old reservation is not undone */ 568ad8ee77eSDmitry Safonov if (vm_flags & VM_ACCOUNT && !(flags & MREMAP_DONTUNMAP)) { 5691da177e4SLinus Torvalds vma->vm_flags &= ~VM_ACCOUNT; 5701da177e4SLinus Torvalds excess = vma->vm_end - vma->vm_start - old_len; 5711da177e4SLinus Torvalds if (old_addr > vma->vm_start && 5721da177e4SLinus Torvalds old_addr + old_len < vma->vm_end) 5731da177e4SLinus Torvalds split = 1; 5741da177e4SLinus Torvalds } 5751da177e4SLinus Torvalds 57671799062SKirill Korotaev /* 577365e9c87SHugh Dickins * If we failed to move page tables we still do total_vm increment 578365e9c87SHugh Dickins * since do_munmap() will decrement it by old_len == new_len. 579365e9c87SHugh Dickins * 580365e9c87SHugh Dickins * Since total_vm is about to be raised artificially high for a 581365e9c87SHugh Dickins * moment, we need to restore high watermark afterwards: if stats 582365e9c87SHugh Dickins * are taken meanwhile, total_vm and hiwater_vm appear too high. 583365e9c87SHugh Dickins * If this were a serious issue, we'd add a flag to do_munmap(). 58471799062SKirill Korotaev */ 585365e9c87SHugh Dickins hiwater_vm = mm->hiwater_vm; 58684638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, new_len >> PAGE_SHIFT); 58771799062SKirill Korotaev 588d9fe4fabSToshi Kani /* Tell pfnmap has moved from this vma */ 589d9fe4fabSToshi Kani if (unlikely(vma->vm_flags & VM_PFNMAP)) 590d9fe4fabSToshi Kani untrack_pfn_moved(vma); 591d9fe4fabSToshi Kani 592e346b381SBrian Geffon if (unlikely(!err && (flags & MREMAP_DONTUNMAP))) { 593e346b381SBrian Geffon /* We always clear VM_LOCKED[ONFAULT] on the old vma */ 594e346b381SBrian Geffon vma->vm_flags &= VM_LOCKED_CLEAR_MASK; 595e346b381SBrian Geffon 5961583aa27SLi Xinhai /* 5971583aa27SLi Xinhai * anon_vma links of the old vma is no longer needed after its page 5981583aa27SLi Xinhai * table has been moved. 5991583aa27SLi Xinhai */ 6001583aa27SLi Xinhai if (new_vma != vma && vma->vm_start == old_addr && 6011583aa27SLi Xinhai vma->vm_end == (old_addr + old_len)) 6021583aa27SLi Xinhai unlink_anon_vmas(vma); 6031583aa27SLi Xinhai 604e346b381SBrian Geffon /* Because we won't unmap we don't need to touch locked_vm */ 605ad8ee77eSDmitry Safonov return new_addr; 606e346b381SBrian Geffon } 607e346b381SBrian Geffon 608897ab3e0SMike Rapoport if (do_munmap(mm, old_addr, old_len, uf_unmap) < 0) { 6091da177e4SLinus Torvalds /* OOM: unable to split vma, just get accounts right */ 610ad8ee77eSDmitry Safonov if (vm_flags & VM_ACCOUNT && !(flags & MREMAP_DONTUNMAP)) 61151df7bcbSDmitry Safonov vm_acct_memory(new_len >> PAGE_SHIFT); 6121da177e4SLinus Torvalds excess = 0; 6131da177e4SLinus Torvalds } 614e346b381SBrian Geffon 615e346b381SBrian Geffon if (vm_flags & VM_LOCKED) { 616e346b381SBrian Geffon mm->locked_vm += new_len >> PAGE_SHIFT; 617e346b381SBrian Geffon *locked = true; 618e346b381SBrian Geffon } 619ad8ee77eSDmitry Safonov 620365e9c87SHugh Dickins mm->hiwater_vm = hiwater_vm; 6211da177e4SLinus Torvalds 6221da177e4SLinus Torvalds /* Restore VM_ACCOUNT if one or two pieces of vma left */ 6231da177e4SLinus Torvalds if (excess) { 6241da177e4SLinus Torvalds vma->vm_flags |= VM_ACCOUNT; 6251da177e4SLinus Torvalds if (split) 6261da177e4SLinus Torvalds vma->vm_next->vm_flags |= VM_ACCOUNT; 6271da177e4SLinus Torvalds } 6281da177e4SLinus Torvalds 6291da177e4SLinus Torvalds return new_addr; 6301da177e4SLinus Torvalds } 6311da177e4SLinus Torvalds 63254f5de70SAl Viro static struct vm_area_struct *vma_to_resize(unsigned long addr, 633e346b381SBrian Geffon unsigned long old_len, unsigned long new_len, unsigned long flags, 634e346b381SBrian Geffon unsigned long *p) 63554f5de70SAl Viro { 63654f5de70SAl Viro struct mm_struct *mm = current->mm; 63754f5de70SAl Viro struct vm_area_struct *vma = find_vma(mm, addr); 6381d391686SOleg Nesterov unsigned long pgoff; 63954f5de70SAl Viro 64054f5de70SAl Viro if (!vma || vma->vm_start > addr) 6416cd57613SDerek return ERR_PTR(-EFAULT); 64254f5de70SAl Viro 643dba58d3bSMike Kravetz /* 644dba58d3bSMike Kravetz * !old_len is a special case where an attempt is made to 'duplicate' 645dba58d3bSMike Kravetz * a mapping. This makes no sense for private mappings as it will 646dba58d3bSMike Kravetz * instead create a fresh/new mapping unrelated to the original. This 647dba58d3bSMike Kravetz * is contrary to the basic idea of mremap which creates new mappings 648dba58d3bSMike Kravetz * based on the original. There are no known use cases for this 649dba58d3bSMike Kravetz * behavior. As a result, fail such attempts. 650dba58d3bSMike Kravetz */ 651dba58d3bSMike Kravetz if (!old_len && !(vma->vm_flags & (VM_SHARED | VM_MAYSHARE))) { 652dba58d3bSMike Kravetz pr_warn_once("%s (%d): attempted to duplicate a private mapping with mremap. This is not supported.\n", current->comm, current->pid); 653dba58d3bSMike Kravetz return ERR_PTR(-EINVAL); 654dba58d3bSMike Kravetz } 655dba58d3bSMike Kravetz 656a4609387SBrian Geffon if ((flags & MREMAP_DONTUNMAP) && 657a4609387SBrian Geffon (vma->vm_flags & (VM_DONTEXPAND | VM_PFNMAP))) 658e346b381SBrian Geffon return ERR_PTR(-EINVAL); 659e346b381SBrian Geffon 66054f5de70SAl Viro if (is_vm_hugetlb_page(vma)) 6616cd57613SDerek return ERR_PTR(-EINVAL); 66254f5de70SAl Viro 66354f5de70SAl Viro /* We can't remap across vm area boundaries */ 66454f5de70SAl Viro if (old_len > vma->vm_end - addr) 6656cd57613SDerek return ERR_PTR(-EFAULT); 66654f5de70SAl Viro 6671d391686SOleg Nesterov if (new_len == old_len) 6681d391686SOleg Nesterov return vma; 669982134baSLinus Torvalds 6701d391686SOleg Nesterov /* Need to be careful about a growing mapping */ 671982134baSLinus Torvalds pgoff = (addr - vma->vm_start) >> PAGE_SHIFT; 672982134baSLinus Torvalds pgoff += vma->vm_pgoff; 673982134baSLinus Torvalds if (pgoff + (new_len >> PAGE_SHIFT) < pgoff) 6746cd57613SDerek return ERR_PTR(-EINVAL); 6751d391686SOleg Nesterov 6761d391686SOleg Nesterov if (vma->vm_flags & (VM_DONTEXPAND | VM_PFNMAP)) 6771d391686SOleg Nesterov return ERR_PTR(-EFAULT); 67854f5de70SAl Viro 67954f5de70SAl Viro if (vma->vm_flags & VM_LOCKED) { 68054f5de70SAl Viro unsigned long locked, lock_limit; 68154f5de70SAl Viro locked = mm->locked_vm << PAGE_SHIFT; 68259e99e5bSJiri Slaby lock_limit = rlimit(RLIMIT_MEMLOCK); 68354f5de70SAl Viro locked += new_len - old_len; 68454f5de70SAl Viro if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 6856cd57613SDerek return ERR_PTR(-EAGAIN); 68654f5de70SAl Viro } 68754f5de70SAl Viro 68884638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vma->vm_flags, 68984638335SKonstantin Khlebnikov (new_len - old_len) >> PAGE_SHIFT)) 6906cd57613SDerek return ERR_PTR(-ENOMEM); 69154f5de70SAl Viro 69254f5de70SAl Viro if (vma->vm_flags & VM_ACCOUNT) { 69354f5de70SAl Viro unsigned long charged = (new_len - old_len) >> PAGE_SHIFT; 694191c5424SAl Viro if (security_vm_enough_memory_mm(mm, charged)) 6956cd57613SDerek return ERR_PTR(-ENOMEM); 69654f5de70SAl Viro *p = charged; 69754f5de70SAl Viro } 69854f5de70SAl Viro 69954f5de70SAl Viro return vma; 70054f5de70SAl Viro } 70154f5de70SAl Viro 70281909b84SMichel Lespinasse static unsigned long mremap_to(unsigned long addr, unsigned long old_len, 70372f87654SPavel Emelyanov unsigned long new_addr, unsigned long new_len, bool *locked, 704e346b381SBrian Geffon unsigned long flags, struct vm_userfaultfd_ctx *uf, 705b2282371SMike Rapoport struct list_head *uf_unmap_early, 706897ab3e0SMike Rapoport struct list_head *uf_unmap) 707ecc1a899SAl Viro { 708ecc1a899SAl Viro struct mm_struct *mm = current->mm; 709ecc1a899SAl Viro struct vm_area_struct *vma; 710ecc1a899SAl Viro unsigned long ret = -EINVAL; 711ecc1a899SAl Viro unsigned long charged = 0; 712e346b381SBrian Geffon unsigned long map_flags = 0; 713ecc1a899SAl Viro 714f19cb115SAlexander Kuleshov if (offset_in_page(new_addr)) 715ecc1a899SAl Viro goto out; 716ecc1a899SAl Viro 717ecc1a899SAl Viro if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len) 718ecc1a899SAl Viro goto out; 719ecc1a899SAl Viro 7209943242cSOleg Nesterov /* Ensure the old/new locations do not overlap */ 7219943242cSOleg Nesterov if (addr + old_len > new_addr && new_addr + new_len > addr) 722ecc1a899SAl Viro goto out; 723ecc1a899SAl Viro 724ea2c3f6fSOscar Salvador /* 725ea2c3f6fSOscar Salvador * move_vma() need us to stay 4 maps below the threshold, otherwise 726ea2c3f6fSOscar Salvador * it will bail out at the very beginning. 727ea2c3f6fSOscar Salvador * That is a problem if we have already unmaped the regions here 728ea2c3f6fSOscar Salvador * (new_addr, and old_addr), because userspace will not know the 729ea2c3f6fSOscar Salvador * state of the vma's after it gets -ENOMEM. 730ea2c3f6fSOscar Salvador * So, to avoid such scenario we can pre-compute if the whole 731ea2c3f6fSOscar Salvador * operation has high chances to success map-wise. 732ea2c3f6fSOscar Salvador * Worst-scenario case is when both vma's (new_addr and old_addr) get 733*f0953a1bSIngo Molnar * split in 3 before unmapping it. 734ea2c3f6fSOscar Salvador * That means 2 more maps (1 for each) to the ones we already hold. 735ea2c3f6fSOscar Salvador * Check whether current map count plus 2 still leads us to 4 maps below 736ea2c3f6fSOscar Salvador * the threshold, otherwise return -ENOMEM here to be more safe. 737ea2c3f6fSOscar Salvador */ 738ea2c3f6fSOscar Salvador if ((mm->map_count + 2) >= sysctl_max_map_count - 3) 739ea2c3f6fSOscar Salvador return -ENOMEM; 740ea2c3f6fSOscar Salvador 741e346b381SBrian Geffon if (flags & MREMAP_FIXED) { 742b2282371SMike Rapoport ret = do_munmap(mm, new_addr, new_len, uf_unmap_early); 743ecc1a899SAl Viro if (ret) 744ecc1a899SAl Viro goto out; 745e346b381SBrian Geffon } 746ecc1a899SAl Viro 747ecc1a899SAl Viro if (old_len >= new_len) { 748897ab3e0SMike Rapoport ret = do_munmap(mm, addr+new_len, old_len - new_len, uf_unmap); 749ecc1a899SAl Viro if (ret && old_len != new_len) 750ecc1a899SAl Viro goto out; 751ecc1a899SAl Viro old_len = new_len; 752ecc1a899SAl Viro } 753ecc1a899SAl Viro 754e346b381SBrian Geffon vma = vma_to_resize(addr, old_len, new_len, flags, &charged); 755ecc1a899SAl Viro if (IS_ERR(vma)) { 756ecc1a899SAl Viro ret = PTR_ERR(vma); 757ecc1a899SAl Viro goto out; 758ecc1a899SAl Viro } 759ecc1a899SAl Viro 760e346b381SBrian Geffon /* MREMAP_DONTUNMAP expands by old_len since old_len == new_len */ 761e346b381SBrian Geffon if (flags & MREMAP_DONTUNMAP && 762e346b381SBrian Geffon !may_expand_vm(mm, vma->vm_flags, old_len >> PAGE_SHIFT)) { 763e346b381SBrian Geffon ret = -ENOMEM; 764e346b381SBrian Geffon goto out; 765e346b381SBrian Geffon } 766e346b381SBrian Geffon 767e346b381SBrian Geffon if (flags & MREMAP_FIXED) 768e346b381SBrian Geffon map_flags |= MAP_FIXED; 769e346b381SBrian Geffon 770097eed10SAl Viro if (vma->vm_flags & VM_MAYSHARE) 771097eed10SAl Viro map_flags |= MAP_SHARED; 7729206de95SAl Viro 773097eed10SAl Viro ret = get_unmapped_area(vma->vm_file, new_addr, new_len, vma->vm_pgoff + 774097eed10SAl Viro ((addr - vma->vm_start) >> PAGE_SHIFT), 775097eed10SAl Viro map_flags); 776ff68dac6SGaowei Pu if (IS_ERR_VALUE(ret)) 777097eed10SAl Viro goto out1; 778097eed10SAl Viro 779e346b381SBrian Geffon /* We got a new mapping */ 780e346b381SBrian Geffon if (!(flags & MREMAP_FIXED)) 781e346b381SBrian Geffon new_addr = ret; 782e346b381SBrian Geffon 783e346b381SBrian Geffon ret = move_vma(vma, addr, old_len, new_len, new_addr, locked, flags, uf, 784897ab3e0SMike Rapoport uf_unmap); 785e346b381SBrian Geffon 786f19cb115SAlexander Kuleshov if (!(offset_in_page(ret))) 787097eed10SAl Viro goto out; 788e346b381SBrian Geffon 789097eed10SAl Viro out1: 790ecc1a899SAl Viro vm_unacct_memory(charged); 791ecc1a899SAl Viro 792ecc1a899SAl Viro out: 793ecc1a899SAl Viro return ret; 794ecc1a899SAl Viro } 795ecc1a899SAl Viro 7961a0ef85fSAl Viro static int vma_expandable(struct vm_area_struct *vma, unsigned long delta) 7971a0ef85fSAl Viro { 798f106af4eSAl Viro unsigned long end = vma->vm_end + delta; 7999206de95SAl Viro if (end < vma->vm_end) /* overflow */ 8001a0ef85fSAl Viro return 0; 8019206de95SAl Viro if (vma->vm_next && vma->vm_next->vm_start < end) /* intersection */ 802f106af4eSAl Viro return 0; 803f106af4eSAl Viro if (get_unmapped_area(NULL, vma->vm_start, end - vma->vm_start, 804f106af4eSAl Viro 0, MAP_FIXED) & ~PAGE_MASK) 805f106af4eSAl Viro return 0; 8061a0ef85fSAl Viro return 1; 8071a0ef85fSAl Viro } 8081a0ef85fSAl Viro 8091da177e4SLinus Torvalds /* 8101da177e4SLinus Torvalds * Expand (or shrink) an existing mapping, potentially moving it at the 8111da177e4SLinus Torvalds * same time (controlled by the MREMAP_MAYMOVE flag and available VM space) 8121da177e4SLinus Torvalds * 8131da177e4SLinus Torvalds * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise 8141da177e4SLinus Torvalds * This option implies MREMAP_MAYMOVE. 8151da177e4SLinus Torvalds */ 81663a81db1SAl Viro SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len, 81763a81db1SAl Viro unsigned long, new_len, unsigned long, flags, 81863a81db1SAl Viro unsigned long, new_addr) 8191da177e4SLinus Torvalds { 820d0de32d9SHugh Dickins struct mm_struct *mm = current->mm; 8211da177e4SLinus Torvalds struct vm_area_struct *vma; 8221da177e4SLinus Torvalds unsigned long ret = -EINVAL; 8231da177e4SLinus Torvalds unsigned long charged = 0; 82481909b84SMichel Lespinasse bool locked = false; 82585a06835SYang Shi bool downgraded = false; 82672f87654SPavel Emelyanov struct vm_userfaultfd_ctx uf = NULL_VM_UFFD_CTX; 827b2282371SMike Rapoport LIST_HEAD(uf_unmap_early); 828897ab3e0SMike Rapoport LIST_HEAD(uf_unmap); 8291da177e4SLinus Torvalds 830b2a84de2SWill Deacon /* 831b2a84de2SWill Deacon * There is a deliberate asymmetry here: we strip the pointer tag 832b2a84de2SWill Deacon * from the old address but leave the new address alone. This is 833b2a84de2SWill Deacon * for consistency with mmap(), where we prevent the creation of 834b2a84de2SWill Deacon * aliasing mappings in userspace by leaving the tag bits of the 835b2a84de2SWill Deacon * mapping address intact. A non-zero tag will cause the subsequent 836b2a84de2SWill Deacon * range checks to reject the address as invalid. 837b2a84de2SWill Deacon * 838b2a84de2SWill Deacon * See Documentation/arm64/tagged-address-abi.rst for more information. 839b2a84de2SWill Deacon */ 840057d3389SAndrey Konovalov addr = untagged_addr(addr); 841057d3389SAndrey Konovalov 842e346b381SBrian Geffon if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE | MREMAP_DONTUNMAP)) 8439a2458a6SRasmus Villemoes return ret; 8449a2458a6SRasmus Villemoes 8459a2458a6SRasmus Villemoes if (flags & MREMAP_FIXED && !(flags & MREMAP_MAYMOVE)) 8469a2458a6SRasmus Villemoes return ret; 8471da177e4SLinus Torvalds 848e346b381SBrian Geffon /* 849e346b381SBrian Geffon * MREMAP_DONTUNMAP is always a move and it does not allow resizing 850e346b381SBrian Geffon * in the process. 851e346b381SBrian Geffon */ 852e346b381SBrian Geffon if (flags & MREMAP_DONTUNMAP && 853e346b381SBrian Geffon (!(flags & MREMAP_MAYMOVE) || old_len != new_len)) 854e346b381SBrian Geffon return ret; 855e346b381SBrian Geffon 856e346b381SBrian Geffon 857f19cb115SAlexander Kuleshov if (offset_in_page(addr)) 8589a2458a6SRasmus Villemoes return ret; 8591da177e4SLinus Torvalds 8601da177e4SLinus Torvalds old_len = PAGE_ALIGN(old_len); 8611da177e4SLinus Torvalds new_len = PAGE_ALIGN(new_len); 8621da177e4SLinus Torvalds 8631da177e4SLinus Torvalds /* 8641da177e4SLinus Torvalds * We allow a zero old-len as a special case 8651da177e4SLinus Torvalds * for DOS-emu "duplicate shm area" thing. But 8661da177e4SLinus Torvalds * a zero new-len is nonsensical. 8671da177e4SLinus Torvalds */ 8681da177e4SLinus Torvalds if (!new_len) 8699a2458a6SRasmus Villemoes return ret; 8709a2458a6SRasmus Villemoes 871d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(current->mm)) 872dc0ef0dfSMichal Hocko return -EINTR; 8731da177e4SLinus Torvalds 874e346b381SBrian Geffon if (flags & (MREMAP_FIXED | MREMAP_DONTUNMAP)) { 87581909b84SMichel Lespinasse ret = mremap_to(addr, old_len, new_addr, new_len, 876e346b381SBrian Geffon &locked, flags, &uf, &uf_unmap_early, 877e346b381SBrian Geffon &uf_unmap); 8781da177e4SLinus Torvalds goto out; 8791da177e4SLinus Torvalds } 8801da177e4SLinus Torvalds 8811da177e4SLinus Torvalds /* 8821da177e4SLinus Torvalds * Always allow a shrinking remap: that just unmaps 8831da177e4SLinus Torvalds * the unnecessary pages.. 88485a06835SYang Shi * __do_munmap does all the needed commit accounting, and 885c1e8d7c6SMichel Lespinasse * downgrades mmap_lock to read if so directed. 8861da177e4SLinus Torvalds */ 8871da177e4SLinus Torvalds if (old_len >= new_len) { 88885a06835SYang Shi int retval; 88985a06835SYang Shi 89085a06835SYang Shi retval = __do_munmap(mm, addr+new_len, old_len - new_len, 89185a06835SYang Shi &uf_unmap, true); 89285a06835SYang Shi if (retval < 0 && old_len != new_len) { 89385a06835SYang Shi ret = retval; 8941da177e4SLinus Torvalds goto out; 895c1e8d7c6SMichel Lespinasse /* Returning 1 indicates mmap_lock is downgraded to read. */ 89685a06835SYang Shi } else if (retval == 1) 89785a06835SYang Shi downgraded = true; 8981da177e4SLinus Torvalds ret = addr; 8991da177e4SLinus Torvalds goto out; 9001da177e4SLinus Torvalds } 9011da177e4SLinus Torvalds 9021da177e4SLinus Torvalds /* 903ecc1a899SAl Viro * Ok, we need to grow.. 9041da177e4SLinus Torvalds */ 905e346b381SBrian Geffon vma = vma_to_resize(addr, old_len, new_len, flags, &charged); 90654f5de70SAl Viro if (IS_ERR(vma)) { 90754f5de70SAl Viro ret = PTR_ERR(vma); 9081da177e4SLinus Torvalds goto out; 9091da177e4SLinus Torvalds } 9101da177e4SLinus Torvalds 9111da177e4SLinus Torvalds /* old_len exactly to the end of the area.. 9121da177e4SLinus Torvalds */ 913ecc1a899SAl Viro if (old_len == vma->vm_end - addr) { 9141da177e4SLinus Torvalds /* can we just expand the current mapping? */ 9151a0ef85fSAl Viro if (vma_expandable(vma, new_len - old_len)) { 9161da177e4SLinus Torvalds int pages = (new_len - old_len) >> PAGE_SHIFT; 9171da177e4SLinus Torvalds 9185beb4930SRik van Riel if (vma_adjust(vma, vma->vm_start, addr + new_len, 9195beb4930SRik van Riel vma->vm_pgoff, NULL)) { 9205beb4930SRik van Riel ret = -ENOMEM; 9215beb4930SRik van Riel goto out; 9225beb4930SRik van Riel } 9231da177e4SLinus Torvalds 92484638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, pages); 9251da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) { 926d0de32d9SHugh Dickins mm->locked_vm += pages; 92781909b84SMichel Lespinasse locked = true; 92881909b84SMichel Lespinasse new_addr = addr; 9291da177e4SLinus Torvalds } 9301da177e4SLinus Torvalds ret = addr; 9311da177e4SLinus Torvalds goto out; 9321da177e4SLinus Torvalds } 9331da177e4SLinus Torvalds } 9341da177e4SLinus Torvalds 9351da177e4SLinus Torvalds /* 9361da177e4SLinus Torvalds * We weren't able to just expand or shrink the area, 9371da177e4SLinus Torvalds * we need to create a new one and move it.. 9381da177e4SLinus Torvalds */ 9391da177e4SLinus Torvalds ret = -ENOMEM; 9401da177e4SLinus Torvalds if (flags & MREMAP_MAYMOVE) { 9411da177e4SLinus Torvalds unsigned long map_flags = 0; 9421da177e4SLinus Torvalds if (vma->vm_flags & VM_MAYSHARE) 9431da177e4SLinus Torvalds map_flags |= MAP_SHARED; 9441da177e4SLinus Torvalds 9451da177e4SLinus Torvalds new_addr = get_unmapped_area(vma->vm_file, 0, new_len, 94693587414SAl Viro vma->vm_pgoff + 94793587414SAl Viro ((addr - vma->vm_start) >> PAGE_SHIFT), 94893587414SAl Viro map_flags); 949ff68dac6SGaowei Pu if (IS_ERR_VALUE(new_addr)) { 9501da177e4SLinus Torvalds ret = new_addr; 951ed032189SEric Paris goto out; 952ed032189SEric Paris } 953ed032189SEric Paris 95472f87654SPavel Emelyanov ret = move_vma(vma, addr, old_len, new_len, new_addr, 955e346b381SBrian Geffon &locked, flags, &uf, &uf_unmap); 9561da177e4SLinus Torvalds } 9571da177e4SLinus Torvalds out: 958f19cb115SAlexander Kuleshov if (offset_in_page(ret)) { 9591da177e4SLinus Torvalds vm_unacct_memory(charged); 960fa1f68ccSZou Wei locked = false; 961d456fb9eSOleg Nesterov } 96285a06835SYang Shi if (downgraded) 963d8ed45c5SMichel Lespinasse mmap_read_unlock(current->mm); 96485a06835SYang Shi else 965d8ed45c5SMichel Lespinasse mmap_write_unlock(current->mm); 96681909b84SMichel Lespinasse if (locked && new_len > old_len) 96781909b84SMichel Lespinasse mm_populate(new_addr + old_len, new_len - old_len); 968b2282371SMike Rapoport userfaultfd_unmap_complete(mm, &uf_unmap_early); 969d1564926SBrian Geffon mremap_userfaultfd_complete(&uf, addr, ret, old_len); 970897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf_unmap); 9711da177e4SLinus Torvalds return ret; 9721da177e4SLinus Torvalds } 973