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 unsigned long change_pte_range(struct mm_struct *mm, pmd_t *pmd,
39 unsigned long addr, unsigned long end, pgprot_t newprot,
40 int dirty_accountable)
44 unsigned long pages = 0;
46 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
47 arch_enter_lazy_mmu_mode();
50 if (pte_present(oldpte)) {
53 ptent = ptep_modify_prot_start(mm, addr, pte);
54 ptent = pte_modify(ptent, newprot);
57 * Avoid taking write faults for pages we know to be
60 if (dirty_accountable && pte_dirty(ptent))
61 ptent = pte_mkwrite(ptent);
63 ptep_modify_prot_commit(mm, addr, pte, ptent);
65 } else if (IS_ENABLED(CONFIG_MIGRATION) && !pte_file(oldpte)) {
66 swp_entry_t entry = pte_to_swp_entry(oldpte);
68 if (is_write_migration_entry(entry)) {
70 * A protection check is difficult so
71 * just be safe and disable write
73 make_migration_entry_read(&entry);
74 set_pte_at(mm, addr, pte,
75 swp_entry_to_pte(entry));
79 } while (pte++, addr += PAGE_SIZE, addr != end);
80 arch_leave_lazy_mmu_mode();
81 pte_unmap_unlock(pte - 1, ptl);
86 static inline unsigned long change_pmd_range(struct vm_area_struct *vma, pud_t *pud,
87 unsigned long addr, unsigned long end, pgprot_t newprot,
88 int dirty_accountable)
92 unsigned long pages = 0;
94 pmd = pmd_offset(pud, addr);
96 next = pmd_addr_end(addr, end);
97 if (pmd_trans_huge(*pmd)) {
98 if (next - addr != HPAGE_PMD_SIZE)
99 split_huge_page_pmd(vma->vm_mm, pmd);
100 else if (change_huge_pmd(vma, pmd, addr, newprot)) {
101 pages += HPAGE_PMD_NR;
106 if (pmd_none_or_clear_bad(pmd))
108 pages += change_pte_range(vma->vm_mm, pmd, addr, next, newprot,
110 } while (pmd++, addr = next, addr != end);
115 static inline unsigned long change_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
116 unsigned long addr, unsigned long end, pgprot_t newprot,
117 int dirty_accountable)
121 unsigned long pages = 0;
123 pud = pud_offset(pgd, addr);
125 next = pud_addr_end(addr, end);
126 if (pud_none_or_clear_bad(pud))
128 pages += change_pmd_range(vma, pud, addr, next, newprot,
130 } while (pud++, addr = next, addr != end);
135 static unsigned long change_protection_range(struct vm_area_struct *vma,
136 unsigned long addr, unsigned long end, pgprot_t newprot,
137 int dirty_accountable)
139 struct mm_struct *mm = vma->vm_mm;
142 unsigned long start = addr;
143 unsigned long pages = 0;
146 pgd = pgd_offset(mm, addr);
147 flush_cache_range(vma, addr, end);
149 next = pgd_addr_end(addr, end);
150 if (pgd_none_or_clear_bad(pgd))
152 pages += change_pud_range(vma, pgd, addr, next, newprot,
154 } while (pgd++, addr = next, addr != end);
156 flush_tlb_range(vma, start, end);
161 unsigned long change_protection(struct vm_area_struct *vma, unsigned long start,
162 unsigned long end, pgprot_t newprot,
163 int dirty_accountable)
165 struct mm_struct *mm = vma->vm_mm;
168 mmu_notifier_invalidate_range_start(mm, start, end);
169 if (is_vm_hugetlb_page(vma))
170 pages = hugetlb_change_protection(vma, start, end, newprot);
172 pages = change_protection_range(vma, start, end, newprot, dirty_accountable);
173 mmu_notifier_invalidate_range_end(mm, start, end);
179 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
180 unsigned long start, unsigned long end, unsigned long newflags)
182 struct mm_struct *mm = vma->vm_mm;
183 unsigned long oldflags = vma->vm_flags;
184 long nrpages = (end - start) >> PAGE_SHIFT;
185 unsigned long charged = 0;
188 int dirty_accountable = 0;
190 if (newflags == oldflags) {
196 * If we make a private mapping writable we increase our commit;
197 * but (without finer accounting) cannot reduce our commit if we
198 * make it unwritable again. hugetlb mapping were accounted for
199 * even if read-only so there is no need to account for them here
201 if (newflags & VM_WRITE) {
202 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
203 VM_SHARED|VM_NORESERVE))) {
205 if (security_vm_enough_memory_mm(mm, charged))
207 newflags |= VM_ACCOUNT;
212 * First try to merge with previous and/or next vma.
214 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
215 *pprev = vma_merge(mm, *pprev, start, end, newflags,
216 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
224 if (start != vma->vm_start) {
225 error = split_vma(mm, vma, start, 1);
230 if (end != vma->vm_end) {
231 error = split_vma(mm, vma, end, 0);
238 * vm_flags and vm_page_prot are protected by the mmap_sem
239 * held in write mode.
241 vma->vm_flags = newflags;
242 vma->vm_page_prot = pgprot_modify(vma->vm_page_prot,
243 vm_get_page_prot(newflags));
245 if (vma_wants_writenotify(vma)) {
246 vma->vm_page_prot = vm_get_page_prot(newflags & ~VM_SHARED);
247 dirty_accountable = 1;
250 change_protection(vma, start, end, vma->vm_page_prot, dirty_accountable);
252 vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
253 vm_stat_account(mm, newflags, vma->vm_file, nrpages);
254 perf_event_mmap(vma);
258 vm_unacct_memory(charged);
262 SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
265 unsigned long vm_flags, nstart, end, tmp, reqprot;
266 struct vm_area_struct *vma, *prev;
268 const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
269 prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
270 if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
273 if (start & ~PAGE_MASK)
277 len = PAGE_ALIGN(len);
281 if (!arch_validate_prot(prot))
286 * Does the application expect PROT_READ to imply PROT_EXEC:
288 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
291 vm_flags = calc_vm_prot_bits(prot);
293 down_write(¤t->mm->mmap_sem);
295 vma = find_vma(current->mm, start);
300 if (unlikely(grows & PROT_GROWSDOWN)) {
301 if (vma->vm_start >= end)
303 start = vma->vm_start;
305 if (!(vma->vm_flags & VM_GROWSDOWN))
309 if (vma->vm_start > start)
311 if (unlikely(grows & PROT_GROWSUP)) {
314 if (!(vma->vm_flags & VM_GROWSUP))
318 if (start > vma->vm_start)
321 for (nstart = start ; ; ) {
322 unsigned long newflags;
324 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
326 newflags = vm_flags | (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
328 /* newflags >> 4 shift VM_MAY% in place of VM_% */
329 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
334 error = security_file_mprotect(vma, reqprot, prot);
341 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
346 if (nstart < prev->vm_end)
347 nstart = prev->vm_end;
352 if (!vma || vma->vm_start != nstart) {
358 up_write(¤t->mm->mmap_sem);