4 * (C) Copyright 1994 Linus Torvalds
5 * (C) Copyright 2002 Christoph Hellwig
7 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
8 * (C) Copyright 2002 Red Hat Inc, All Rights Reserved
12 #include <linux/hugetlb.h>
13 #include <linux/shm.h>
14 #include <linux/mman.h>
16 #include <linux/highmem.h>
17 #include <linux/security.h>
18 #include <linux/mempolicy.h>
19 #include <linux/personality.h>
20 #include <linux/syscalls.h>
21 #include <linux/swap.h>
22 #include <linux/swapops.h>
23 #include <linux/mmu_notifier.h>
24 #include <linux/migrate.h>
25 #include <linux/perf_event.h>
26 #include <asm/uaccess.h>
27 #include <asm/pgtable.h>
28 #include <asm/cacheflush.h>
29 #include <asm/tlbflush.h>
32 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
38 static void change_pte_range(struct mm_struct *mm, pmd_t *pmd,
39 unsigned long addr, unsigned long end, pgprot_t newprot,
40 int dirty_accountable)
45 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
46 arch_enter_lazy_mmu_mode();
49 if (pte_present(oldpte)) {
52 ptent = ptep_modify_prot_start(mm, addr, pte);
53 ptent = pte_modify(ptent, newprot);
56 * Avoid taking write faults for pages we know to be
59 if (dirty_accountable && pte_dirty(ptent))
60 ptent = pte_mkwrite(ptent);
62 ptep_modify_prot_commit(mm, addr, pte, ptent);
63 } else if (IS_ENABLED(CONFIG_MIGRATION) && !pte_file(oldpte)) {
64 swp_entry_t entry = pte_to_swp_entry(oldpte);
66 if (is_write_migration_entry(entry)) {
68 * A protection check is difficult so
69 * just be safe and disable write
71 make_migration_entry_read(&entry);
72 set_pte_at(mm, addr, pte,
73 swp_entry_to_pte(entry));
76 } while (pte++, addr += PAGE_SIZE, addr != end);
77 arch_leave_lazy_mmu_mode();
78 pte_unmap_unlock(pte - 1, ptl);
81 static inline void change_pmd_range(struct vm_area_struct *vma, pud_t *pud,
82 unsigned long addr, unsigned long end, pgprot_t newprot,
83 int dirty_accountable)
88 pmd = pmd_offset(pud, addr);
90 next = pmd_addr_end(addr, end);
91 if (pmd_trans_huge(*pmd)) {
92 if (next - addr != HPAGE_PMD_SIZE)
93 split_huge_page_pmd(vma->vm_mm, pmd);
94 else if (change_huge_pmd(vma, pmd, addr, newprot))
98 if (pmd_none_or_clear_bad(pmd))
100 change_pte_range(vma->vm_mm, pmd, addr, next, newprot,
102 } while (pmd++, addr = next, addr != end);
105 static inline void change_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
106 unsigned long addr, unsigned long end, pgprot_t newprot,
107 int dirty_accountable)
112 pud = pud_offset(pgd, addr);
114 next = pud_addr_end(addr, end);
115 if (pud_none_or_clear_bad(pud))
117 change_pmd_range(vma, pud, addr, next, newprot,
119 } while (pud++, addr = next, addr != end);
122 static void change_protection(struct vm_area_struct *vma,
123 unsigned long addr, unsigned long end, pgprot_t newprot,
124 int dirty_accountable)
126 struct mm_struct *mm = vma->vm_mm;
129 unsigned long start = addr;
132 pgd = pgd_offset(mm, addr);
133 flush_cache_range(vma, addr, end);
135 next = pgd_addr_end(addr, end);
136 if (pgd_none_or_clear_bad(pgd))
138 change_pud_range(vma, pgd, addr, next, newprot,
140 } while (pgd++, addr = next, addr != end);
141 flush_tlb_range(vma, start, end);
145 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
146 unsigned long start, unsigned long end, unsigned long newflags)
148 struct mm_struct *mm = vma->vm_mm;
149 unsigned long oldflags = vma->vm_flags;
150 long nrpages = (end - start) >> PAGE_SHIFT;
151 unsigned long charged = 0;
154 int dirty_accountable = 0;
156 if (newflags == oldflags) {
162 * If we make a private mapping writable we increase our commit;
163 * but (without finer accounting) cannot reduce our commit if we
164 * make it unwritable again. hugetlb mapping were accounted for
165 * even if read-only so there is no need to account for them here
167 if (newflags & VM_WRITE) {
168 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
169 VM_SHARED|VM_NORESERVE))) {
171 if (security_vm_enough_memory_mm(mm, charged))
173 newflags |= VM_ACCOUNT;
178 * First try to merge with previous and/or next vma.
180 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
181 *pprev = vma_merge(mm, *pprev, start, end, newflags,
182 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
190 if (start != vma->vm_start) {
191 error = split_vma(mm, vma, start, 1);
196 if (end != vma->vm_end) {
197 error = split_vma(mm, vma, end, 0);
204 * vm_flags and vm_page_prot are protected by the mmap_sem
205 * held in write mode.
207 vma->vm_flags = newflags;
208 vma->vm_page_prot = pgprot_modify(vma->vm_page_prot,
209 vm_get_page_prot(newflags));
211 if (vma_wants_writenotify(vma)) {
212 vma->vm_page_prot = vm_get_page_prot(newflags & ~VM_SHARED);
213 dirty_accountable = 1;
216 mmu_notifier_invalidate_range_start(mm, start, end);
217 if (is_vm_hugetlb_page(vma))
218 hugetlb_change_protection(vma, start, end, vma->vm_page_prot);
220 change_protection(vma, start, end, vma->vm_page_prot, dirty_accountable);
221 mmu_notifier_invalidate_range_end(mm, start, end);
222 vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
223 vm_stat_account(mm, newflags, vma->vm_file, nrpages);
224 perf_event_mmap(vma);
228 vm_unacct_memory(charged);
232 SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
235 unsigned long vm_flags, nstart, end, tmp, reqprot;
236 struct vm_area_struct *vma, *prev;
238 const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
239 prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
240 if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
243 if (start & ~PAGE_MASK)
247 len = PAGE_ALIGN(len);
251 if (!arch_validate_prot(prot))
256 * Does the application expect PROT_READ to imply PROT_EXEC:
258 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
261 vm_flags = calc_vm_prot_bits(prot);
263 down_write(¤t->mm->mmap_sem);
265 vma = find_vma(current->mm, start);
270 if (unlikely(grows & PROT_GROWSDOWN)) {
271 if (vma->vm_start >= end)
273 start = vma->vm_start;
275 if (!(vma->vm_flags & VM_GROWSDOWN))
279 if (vma->vm_start > start)
281 if (unlikely(grows & PROT_GROWSUP)) {
284 if (!(vma->vm_flags & VM_GROWSUP))
288 if (start > vma->vm_start)
291 for (nstart = start ; ; ) {
292 unsigned long newflags;
294 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
296 newflags = vm_flags | (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
298 /* newflags >> 4 shift VM_MAY% in place of VM_% */
299 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
304 error = security_file_mprotect(vma, reqprot, prot);
311 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
316 if (nstart < prev->vm_end)
317 nstart = prev->vm_end;
322 if (!vma || vma->vm_start != nstart) {
328 up_write(¤t->mm->mmap_sem);