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/pgalloc.h>
29 #include <asm/cacheflush.h>
30 #include <asm/tlbflush.h>
32 #ifndef arch_remove_exec_range
33 #define arch_remove_exec_range(mm, limit) do { ; } while (0)
37 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
43 static void change_pte_range(struct mm_struct *mm, pmd_t *pmd,
44 unsigned long addr, unsigned long end, pgprot_t newprot,
45 int dirty_accountable)
50 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
51 arch_enter_lazy_mmu_mode();
54 if (pte_present(oldpte)) {
57 ptent = ptep_modify_prot_start(mm, addr, pte);
58 ptent = pte_modify(ptent, newprot);
61 * Avoid taking write faults for pages we know to be
64 if (dirty_accountable && pte_dirty(ptent))
65 ptent = pte_mkwrite(ptent);
67 ptep_modify_prot_commit(mm, addr, pte, ptent);
68 } else if (PAGE_MIGRATION && !pte_file(oldpte)) {
69 swp_entry_t entry = pte_to_swp_entry(oldpte);
71 if (is_write_migration_entry(entry)) {
73 * A protection check is difficult so
74 * just be safe and disable write
76 make_migration_entry_read(&entry);
77 set_pte_at(mm, addr, pte,
78 swp_entry_to_pte(entry));
81 } while (pte++, addr += PAGE_SIZE, addr != end);
82 arch_leave_lazy_mmu_mode();
83 pte_unmap_unlock(pte - 1, ptl);
86 static inline void 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)
93 pmd = pmd_offset(pud, addr);
95 next = pmd_addr_end(addr, end);
96 if (pmd_trans_huge(*pmd)) {
97 if (next - addr != HPAGE_PMD_SIZE)
98 split_huge_page_pmd(vma->vm_mm, pmd);
99 else if (change_huge_pmd(vma, pmd, addr, newprot))
103 if (pmd_none_or_clear_bad(pmd))
105 change_pte_range(vma->vm_mm, pmd, addr, next, newprot,
107 } while (pmd++, addr = next, addr != end);
110 static inline void change_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
111 unsigned long addr, unsigned long end, pgprot_t newprot,
112 int dirty_accountable)
117 pud = pud_offset(pgd, addr);
119 next = pud_addr_end(addr, end);
120 if (pud_none_or_clear_bad(pud))
122 change_pmd_range(vma, pud, addr, next, newprot,
124 } while (pud++, addr = next, addr != end);
127 static void change_protection(struct vm_area_struct *vma,
128 unsigned long addr, unsigned long end, pgprot_t newprot,
129 int dirty_accountable)
131 struct mm_struct *mm = vma->vm_mm;
134 unsigned long start = addr;
137 pgd = pgd_offset(mm, addr);
138 flush_cache_range(vma, addr, end);
140 next = pgd_addr_end(addr, end);
141 if (pgd_none_or_clear_bad(pgd))
143 change_pud_range(vma, pgd, addr, next, newprot,
145 } while (pgd++, addr = next, addr != end);
146 flush_tlb_range(vma, start, end);
150 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
151 unsigned long start, unsigned long end, unsigned long newflags)
153 struct mm_struct *mm = vma->vm_mm;
154 unsigned long oldflags = vma->vm_flags;
155 long nrpages = (end - start) >> PAGE_SHIFT;
156 unsigned long charged = 0, old_end = vma->vm_end;
159 int dirty_accountable = 0;
161 if (newflags == oldflags) {
167 * If we make a private mapping writable we increase our commit;
168 * but (without finer accounting) cannot reduce our commit if we
169 * make it unwritable again. hugetlb mapping were accounted for
170 * even if read-only so there is no need to account for them here
172 if (newflags & VM_WRITE) {
173 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
174 VM_SHARED|VM_NORESERVE))) {
176 if (security_vm_enough_memory(charged))
178 newflags |= VM_ACCOUNT;
183 * First try to merge with previous and/or next vma.
185 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
186 *pprev = vma_merge(mm, *pprev, start, end, newflags,
187 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
195 if (start != vma->vm_start) {
196 error = split_vma(mm, vma, start, 1);
201 if (end != vma->vm_end) {
202 error = split_vma(mm, vma, end, 0);
209 * vm_flags and vm_page_prot are protected by the mmap_sem
210 * held in write mode.
212 vma->vm_flags = newflags;
213 vma->vm_page_prot = pgprot_modify(vma->vm_page_prot,
214 vm_get_page_prot(newflags));
216 if (vma_wants_writenotify(vma)) {
217 vma->vm_page_prot = vm_get_page_prot(newflags & ~VM_SHARED);
218 dirty_accountable = 1;
221 if (oldflags & VM_EXEC)
222 arch_remove_exec_range(current->mm, old_end);
224 mmu_notifier_invalidate_range_start(mm, start, end);
225 if (is_vm_hugetlb_page(vma))
226 hugetlb_change_protection(vma, start, end, vma->vm_page_prot);
228 change_protection(vma, start, end, vma->vm_page_prot, dirty_accountable);
229 mmu_notifier_invalidate_range_end(mm, start, end);
230 vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
231 vm_stat_account(mm, newflags, vma->vm_file, nrpages);
232 perf_event_mmap(vma);
236 vm_unacct_memory(charged);
240 SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
243 unsigned long vm_flags, nstart, end, tmp, reqprot;
244 struct vm_area_struct *vma, *prev;
246 const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
247 prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
248 if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
251 if (start & ~PAGE_MASK)
255 len = PAGE_ALIGN(len);
259 if (!arch_validate_prot(prot))
264 * Does the application expect PROT_READ to imply PROT_EXEC:
266 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
269 vm_flags = calc_vm_prot_bits(prot);
271 down_write(¤t->mm->mmap_sem);
273 vma = find_vma_prev(current->mm, start, &prev);
277 if (unlikely(grows & PROT_GROWSDOWN)) {
278 if (vma->vm_start >= end)
280 start = vma->vm_start;
282 if (!(vma->vm_flags & VM_GROWSDOWN))
286 if (vma->vm_start > start)
288 if (unlikely(grows & PROT_GROWSUP)) {
291 if (!(vma->vm_flags & VM_GROWSUP))
295 if (start > vma->vm_start)
298 for (nstart = start ; ; ) {
299 unsigned long newflags;
301 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
303 newflags = vm_flags | (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
305 /* newflags >> 4 shift VM_MAY% in place of VM_% */
306 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
311 error = security_file_mprotect(vma, reqprot, prot);
318 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
323 if (nstart < prev->vm_end)
324 nstart = prev->vm_end;
329 if (!vma || vma->vm_start != nstart) {
335 up_write(¤t->mm->mmap_sem);