4 * (C) Copyright 1994 Linus Torvalds
5 * (C) Copyright 2002 Christoph Hellwig
7 * Address space accounting code <alan@redhat.com>
8 * (C) Copyright 2002 Red Hat Inc, All Rights Reserved
12 #include <linux/hugetlb.h>
13 #include <linux/slab.h>
14 #include <linux/shm.h>
15 #include <linux/mman.h>
17 #include <linux/highmem.h>
18 #include <linux/security.h>
19 #include <linux/mempolicy.h>
20 #include <linux/personality.h>
21 #include <linux/syscalls.h>
23 #include <asm/uaccess.h>
24 #include <asm/pgtable.h>
25 #include <asm/pgalloc.h>
26 #include <asm/cacheflush.h>
27 #include <asm/tlbflush.h>
29 static void change_pte_range(struct mm_struct *mm, pmd_t *pmd,
30 unsigned long addr, unsigned long end, pgprot_t newprot)
35 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
37 if (pte_present(*pte)) {
40 /* Avoid an SMP race with hardware updated dirty/clean
41 * bits by wiping the pte and then setting the new pte
44 ptent = pte_modify(ptep_get_and_clear(mm, addr, pte), newprot);
45 set_pte_at(mm, addr, pte, ptent);
46 lazy_mmu_prot_update(ptent);
48 } while (pte++, addr += PAGE_SIZE, addr != end);
49 pte_unmap_unlock(pte - 1, ptl);
52 static inline void change_pmd_range(struct mm_struct *mm, pud_t *pud,
53 unsigned long addr, unsigned long end, pgprot_t newprot)
58 pmd = pmd_offset(pud, addr);
60 next = pmd_addr_end(addr, end);
61 if (pmd_none_or_clear_bad(pmd))
63 change_pte_range(mm, pmd, addr, next, newprot);
64 } while (pmd++, addr = next, addr != end);
67 static inline void change_pud_range(struct mm_struct *mm, pgd_t *pgd,
68 unsigned long addr, unsigned long end, pgprot_t newprot)
73 pud = pud_offset(pgd, addr);
75 next = pud_addr_end(addr, end);
76 if (pud_none_or_clear_bad(pud))
78 change_pmd_range(mm, pud, addr, next, newprot);
79 } while (pud++, addr = next, addr != end);
82 static void change_protection(struct vm_area_struct *vma,
83 unsigned long addr, unsigned long end, pgprot_t newprot)
85 struct mm_struct *mm = vma->vm_mm;
88 unsigned long start = addr;
91 pgd = pgd_offset(mm, addr);
92 flush_cache_range(vma, addr, end);
94 next = pgd_addr_end(addr, end);
95 if (pgd_none_or_clear_bad(pgd))
97 change_pud_range(mm, pgd, addr, next, newprot);
98 } while (pgd++, addr = next, addr != end);
99 flush_tlb_range(vma, start, end);
103 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
104 unsigned long start, unsigned long end, unsigned long newflags)
106 struct mm_struct *mm = vma->vm_mm;
107 unsigned long oldflags = vma->vm_flags;
108 long nrpages = (end - start) >> PAGE_SHIFT;
109 unsigned long charged = 0, old_end = vma->vm_end;
114 if (newflags == oldflags) {
120 * If we make a private mapping writable we increase our commit;
121 * but (without finer accounting) cannot reduce our commit if we
122 * make it unwritable again.
124 * FIXME? We haven't defined a VM_NORESERVE flag, so mprotecting
125 * a MAP_NORESERVE private mapping to writable will now reserve.
127 if (newflags & VM_WRITE) {
128 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_SHARED))) {
130 if (security_vm_enough_memory(charged))
132 newflags |= VM_ACCOUNT;
136 newprot = protection_map[newflags & 0xf];
139 * First try to merge with previous and/or next vma.
141 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
142 *pprev = vma_merge(mm, *pprev, start, end, newflags,
143 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
151 if (start != vma->vm_start) {
152 error = split_vma(mm, vma, start, 1);
157 if (end != vma->vm_end) {
158 error = split_vma(mm, vma, end, 0);
165 * vm_flags and vm_page_prot are protected by the mmap_sem
166 * held in write mode.
168 vma->vm_flags = newflags;
169 vma->vm_page_prot = newprot;
170 if (oldflags & VM_EXEC)
171 arch_remove_exec_range(current->mm, old_end);
172 if (is_vm_hugetlb_page(vma))
173 hugetlb_change_protection(vma, start, end, newprot);
175 change_protection(vma, start, end, newprot);
176 vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
177 vm_stat_account(mm, newflags, vma->vm_file, nrpages);
181 vm_unacct_memory(charged);
186 sys_mprotect(unsigned long start, size_t len, unsigned long prot)
188 unsigned long vm_flags, nstart, end, tmp, reqprot;
189 struct vm_area_struct *vma, *prev;
191 const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
192 prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
193 if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
196 if (start & ~PAGE_MASK)
200 len = PAGE_ALIGN(len);
204 if (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC | PROT_SEM))
209 * Does the application expect PROT_READ to imply PROT_EXEC:
211 if (unlikely((prot & PROT_READ) &&
212 (current->personality & READ_IMPLIES_EXEC)))
215 vm_flags = calc_vm_prot_bits(prot);
217 down_write(¤t->mm->mmap_sem);
219 vma = find_vma_prev(current->mm, start, &prev);
223 if (unlikely(grows & PROT_GROWSDOWN)) {
224 if (vma->vm_start >= end)
226 start = vma->vm_start;
228 if (!(vma->vm_flags & VM_GROWSDOWN))
232 if (vma->vm_start > start)
234 if (unlikely(grows & PROT_GROWSUP)) {
237 if (!(vma->vm_flags & VM_GROWSUP))
241 if (start > vma->vm_start)
244 for (nstart = start ; ; ) {
245 unsigned long newflags;
247 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
249 newflags = vm_flags | (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
251 /* newflags >> 4 shift VM_MAY% in place of VM_% */
252 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
257 error = security_file_mprotect(vma, reqprot, prot);
264 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
269 if (nstart < prev->vm_end)
270 nstart = prev->vm_end;
275 if (!vma || vma->vm_start != nstart) {
281 up_write(¤t->mm->mmap_sem);