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>
22 #include <linux/swap.h>
23 #include <linux/swapops.h>
24 #include <asm/uaccess.h>
25 #include <asm/pgtable.h>
26 #include <asm/pgalloc.h>
27 #include <asm/cacheflush.h>
28 #include <asm/tlbflush.h>
30 static void change_pte_range(struct mm_struct *mm, pmd_t *pmd,
31 unsigned long addr, unsigned long end, pgprot_t newprot,
32 int dirty_accountable)
37 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
40 if (pte_present(oldpte)) {
43 /* Avoid an SMP race with hardware updated dirty/clean
44 * bits by wiping the pte and then setting the new pte
47 ptent = ptep_get_and_clear(mm, addr, pte);
48 ptent = pte_modify(ptent, newprot);
50 * Avoid taking write faults for pages we know to be
53 if (dirty_accountable && pte_dirty(ptent))
54 ptent = pte_mkwrite(ptent);
55 set_pte_at(mm, addr, pte, ptent);
56 lazy_mmu_prot_update(ptent);
57 #ifdef CONFIG_MIGRATION
58 } else if (!pte_file(oldpte)) {
59 swp_entry_t entry = pte_to_swp_entry(oldpte);
61 if (is_write_migration_entry(entry)) {
63 * A protection check is difficult so
64 * just be safe and disable write
66 make_migration_entry_read(&entry);
67 set_pte_at(mm, addr, pte,
68 swp_entry_to_pte(entry));
73 } while (pte++, addr += PAGE_SIZE, addr != end);
74 pte_unmap_unlock(pte - 1, ptl);
77 static inline void change_pmd_range(struct mm_struct *mm, pud_t *pud,
78 unsigned long addr, unsigned long end, pgprot_t newprot,
79 int dirty_accountable)
84 pmd = pmd_offset(pud, addr);
86 next = pmd_addr_end(addr, end);
87 if (pmd_none_or_clear_bad(pmd))
89 change_pte_range(mm, pmd, addr, next, newprot, dirty_accountable);
90 } while (pmd++, addr = next, addr != end);
93 static inline void change_pud_range(struct mm_struct *mm, pgd_t *pgd,
94 unsigned long addr, unsigned long end, pgprot_t newprot,
95 int dirty_accountable)
100 pud = pud_offset(pgd, addr);
102 next = pud_addr_end(addr, end);
103 if (pud_none_or_clear_bad(pud))
105 change_pmd_range(mm, pud, addr, next, newprot, dirty_accountable);
106 } while (pud++, addr = next, addr != end);
109 static void change_protection(struct vm_area_struct *vma,
110 unsigned long addr, unsigned long end, pgprot_t newprot,
111 int dirty_accountable)
113 struct mm_struct *mm = vma->vm_mm;
116 unsigned long start = addr;
119 pgd = pgd_offset(mm, addr);
120 flush_cache_range(vma, addr, end);
122 next = pgd_addr_end(addr, end);
123 if (pgd_none_or_clear_bad(pgd))
125 change_pud_range(mm, pgd, addr, next, newprot, dirty_accountable);
126 } while (pgd++, addr = next, addr != end);
127 flush_tlb_range(vma, start, end);
131 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
132 unsigned long start, unsigned long end, unsigned long newflags)
134 struct mm_struct *mm = vma->vm_mm;
135 unsigned long oldflags = vma->vm_flags;
136 long nrpages = (end - start) >> PAGE_SHIFT;
137 unsigned long charged = 0, old_end = vma->vm_end;
140 int dirty_accountable = 0;
142 if (newflags == oldflags) {
148 * If we make a private mapping writable we increase our commit;
149 * but (without finer accounting) cannot reduce our commit if we
150 * make it unwritable again.
152 * FIXME? We haven't defined a VM_NORESERVE flag, so mprotecting
153 * a MAP_NORESERVE private mapping to writable will now reserve.
155 if (newflags & VM_WRITE) {
156 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_SHARED))) {
158 if (security_vm_enough_memory(charged))
160 newflags |= VM_ACCOUNT;
165 * First try to merge with previous and/or next vma.
167 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
168 *pprev = vma_merge(mm, *pprev, start, end, newflags,
169 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
177 if (start != vma->vm_start) {
178 error = split_vma(mm, vma, start, 1);
183 if (end != vma->vm_end) {
184 error = split_vma(mm, vma, end, 0);
191 * vm_flags and vm_page_prot are protected by the mmap_sem
192 * held in write mode.
194 vma->vm_flags = newflags;
195 if (oldflags & VM_EXEC)
196 arch_remove_exec_range(current->mm, old_end);
197 vma->vm_page_prot = protection_map[newflags &
198 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)];
199 if (vma_wants_writenotify(vma)) {
200 vma->vm_page_prot = protection_map[newflags &
201 (VM_READ|VM_WRITE|VM_EXEC)];
202 dirty_accountable = 1;
205 if (is_vm_hugetlb_page(vma))
206 hugetlb_change_protection(vma, start, end, vma->vm_page_prot);
208 change_protection(vma, start, end, vma->vm_page_prot, dirty_accountable);
209 vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
210 vm_stat_account(mm, newflags, vma->vm_file, nrpages);
214 vm_unacct_memory(charged);
219 sys_mprotect(unsigned long start, size_t len, unsigned long prot)
221 unsigned long vm_flags, nstart, end, tmp, reqprot;
222 struct vm_area_struct *vma, *prev;
224 const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
225 prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
226 if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
229 if (start & ~PAGE_MASK)
233 len = PAGE_ALIGN(len);
237 if (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC | PROT_SEM))
242 * Does the application expect PROT_READ to imply PROT_EXEC:
244 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
247 vm_flags = calc_vm_prot_bits(prot);
249 down_write(¤t->mm->mmap_sem);
251 vma = find_vma_prev(current->mm, start, &prev);
255 if (unlikely(grows & PROT_GROWSDOWN)) {
256 if (vma->vm_start >= end)
258 start = vma->vm_start;
260 if (!(vma->vm_flags & VM_GROWSDOWN))
264 if (vma->vm_start > start)
266 if (unlikely(grows & PROT_GROWSUP)) {
269 if (!(vma->vm_flags & VM_GROWSUP))
273 if (start > vma->vm_start)
276 for (nstart = start ; ; ) {
277 unsigned long newflags;
279 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
281 newflags = vm_flags | (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
283 /* newflags >> 4 shift VM_MAY% in place of VM_% */
284 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
289 error = security_file_mprotect(vma, reqprot, prot);
296 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
301 if (nstart < prev->vm_end)
302 nstart = prev->vm_end;
307 if (!vma || vma->vm_start != nstart) {
313 up_write(¤t->mm->mmap_sem);