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);
38 arch_enter_lazy_mmu_mode();
41 if (pte_present(oldpte)) {
44 /* Avoid an SMP race with hardware updated dirty/clean
45 * bits by wiping the pte and then setting the new pte
48 ptent = ptep_get_and_clear(mm, addr, pte);
49 ptent = pte_modify(ptent, newprot);
51 * Avoid taking write faults for pages we know to be
54 if (dirty_accountable && pte_dirty(ptent))
55 ptent = pte_mkwrite(ptent);
56 set_pte_at(mm, addr, pte, ptent);
57 lazy_mmu_prot_update(ptent);
58 #ifdef CONFIG_MIGRATION
59 } else if (!pte_file(oldpte)) {
60 swp_entry_t entry = pte_to_swp_entry(oldpte);
62 if (is_write_migration_entry(entry)) {
64 * A protection check is difficult so
65 * just be safe and disable write
67 make_migration_entry_read(&entry);
68 set_pte_at(mm, addr, pte,
69 swp_entry_to_pte(entry));
74 } while (pte++, addr += PAGE_SIZE, addr != end);
75 arch_leave_lazy_mmu_mode();
76 pte_unmap_unlock(pte - 1, ptl);
79 static inline void change_pmd_range(struct mm_struct *mm, pud_t *pud,
80 unsigned long addr, unsigned long end, pgprot_t newprot,
81 int dirty_accountable)
86 pmd = pmd_offset(pud, addr);
88 next = pmd_addr_end(addr, end);
89 if (pmd_none_or_clear_bad(pmd))
91 change_pte_range(mm, pmd, addr, next, newprot, dirty_accountable);
92 } while (pmd++, addr = next, addr != end);
95 static inline void change_pud_range(struct mm_struct *mm, pgd_t *pgd,
96 unsigned long addr, unsigned long end, pgprot_t newprot,
97 int dirty_accountable)
102 pud = pud_offset(pgd, addr);
104 next = pud_addr_end(addr, end);
105 if (pud_none_or_clear_bad(pud))
107 change_pmd_range(mm, pud, addr, next, newprot, dirty_accountable);
108 } while (pud++, addr = next, addr != end);
111 static void change_protection(struct vm_area_struct *vma,
112 unsigned long addr, unsigned long end, pgprot_t newprot,
113 int dirty_accountable)
115 struct mm_struct *mm = vma->vm_mm;
118 unsigned long start = addr;
121 pgd = pgd_offset(mm, addr);
122 flush_cache_range(vma, addr, end);
124 next = pgd_addr_end(addr, end);
125 if (pgd_none_or_clear_bad(pgd))
127 change_pud_range(mm, pgd, addr, next, newprot, dirty_accountable);
128 } while (pgd++, addr = next, addr != end);
129 flush_tlb_range(vma, start, end);
133 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
134 unsigned long start, unsigned long end, unsigned long newflags)
136 struct mm_struct *mm = vma->vm_mm;
137 unsigned long oldflags = vma->vm_flags;
138 long nrpages = (end - start) >> PAGE_SHIFT;
139 unsigned long charged = 0, old_end = vma->vm_end;
142 int dirty_accountable = 0;
144 if (newflags == oldflags) {
150 * If we make a private mapping writable we increase our commit;
151 * but (without finer accounting) cannot reduce our commit if we
152 * make it unwritable again.
154 * FIXME? We haven't defined a VM_NORESERVE flag, so mprotecting
155 * a MAP_NORESERVE private mapping to writable will now reserve.
157 if (newflags & VM_WRITE) {
158 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_SHARED))) {
160 if (security_vm_enough_memory(charged))
162 newflags |= VM_ACCOUNT;
167 * First try to merge with previous and/or next vma.
169 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
170 *pprev = vma_merge(mm, *pprev, start, end, newflags,
171 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
179 if (start != vma->vm_start) {
180 error = split_vma(mm, vma, start, 1);
185 if (end != vma->vm_end) {
186 error = split_vma(mm, vma, end, 0);
193 * vm_flags and vm_page_prot are protected by the mmap_sem
194 * held in write mode.
196 vma->vm_flags = newflags;
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 (oldflags & VM_EXEC)
206 arch_remove_exec_range(current->mm, old_end);
208 if (is_vm_hugetlb_page(vma))
209 hugetlb_change_protection(vma, start, end, vma->vm_page_prot);
211 change_protection(vma, start, end, vma->vm_page_prot, dirty_accountable);
212 vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
213 vm_stat_account(mm, newflags, vma->vm_file, nrpages);
217 vm_unacct_memory(charged);
222 sys_mprotect(unsigned long start, size_t len, unsigned long prot)
224 unsigned long vm_flags, nstart, end, tmp, reqprot;
225 struct vm_area_struct *vma, *prev;
227 const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
228 prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
229 if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
232 if (start & ~PAGE_MASK)
236 len = PAGE_ALIGN(len);
240 if (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC | PROT_SEM))
245 * Does the application expect PROT_READ to imply PROT_EXEC:
247 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
250 vm_flags = calc_vm_prot_bits(prot);
252 down_write(¤t->mm->mmap_sem);
254 vma = find_vma_prev(current->mm, start, &prev);
258 if (unlikely(grows & PROT_GROWSDOWN)) {
259 if (vma->vm_start >= end)
261 start = vma->vm_start;
263 if (!(vma->vm_flags & VM_GROWSDOWN))
267 if (vma->vm_start > start)
269 if (unlikely(grows & PROT_GROWSUP)) {
272 if (!(vma->vm_flags & VM_GROWSUP))
276 if (start > vma->vm_start)
279 for (nstart = start ; ; ) {
280 unsigned long newflags;
282 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
284 newflags = vm_flags | (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
286 /* newflags >> 4 shift VM_MAY% in place of VM_% */
287 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
292 error = security_file_mprotect(vma, reqprot, prot);
299 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
304 if (nstart < prev->vm_end)
305 nstart = prev->vm_end;
310 if (!vma || vma->vm_start != nstart) {
316 up_write(¤t->mm->mmap_sem);