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 <linux/ksm.h>
27 #include <asm/uaccess.h>
28 #include <asm/pgtable.h>
29 #include <asm/cacheflush.h>
30 #include <asm/tlbflush.h>
33 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
39 static unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
40 unsigned long addr, unsigned long end, pgprot_t newprot,
41 int dirty_accountable, int prot_numa)
43 struct mm_struct *mm = vma->vm_mm;
46 unsigned long pages = 0;
48 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
49 arch_enter_lazy_mmu_mode();
52 if (pte_present(oldpte)) {
57 ptent = ptep_modify_prot_start(mm, addr, pte);
59 ptent = pte_mknonnuma(ptent);
60 ptent = pte_modify(ptent, newprot);
62 * Avoid taking write faults for pages we
65 if (dirty_accountable && pte_dirty(ptent))
66 ptent = pte_mkwrite(ptent);
67 ptep_modify_prot_commit(mm, addr, pte, ptent);
73 page = vm_normal_page(vma, addr, oldpte);
74 if (page && !PageKsm(page)) {
75 if (!pte_numa(oldpte)) {
76 ptent = pte_mknuma(ptent);
77 set_pte_at(mm, addr, pte, ptent);
84 } else if (IS_ENABLED(CONFIG_MIGRATION) && !pte_file(oldpte)) {
85 swp_entry_t entry = pte_to_swp_entry(oldpte);
87 if (is_write_migration_entry(entry)) {
90 * A protection check is difficult so
91 * just be safe and disable write
93 make_migration_entry_read(&entry);
94 newpte = swp_entry_to_pte(entry);
95 if (pte_swp_soft_dirty(oldpte))
96 newpte = pte_swp_mksoft_dirty(newpte);
97 set_pte_at(mm, addr, pte, newpte);
102 } while (pte++, addr += PAGE_SIZE, addr != end);
103 arch_leave_lazy_mmu_mode();
104 pte_unmap_unlock(pte - 1, ptl);
109 static inline unsigned long change_pmd_range(struct vm_area_struct *vma,
110 pud_t *pud, unsigned long addr, unsigned long end,
111 pgprot_t newprot, int dirty_accountable, int prot_numa)
115 unsigned long pages = 0;
116 unsigned long nr_huge_updates = 0;
118 pmd = pmd_offset(pud, addr);
120 unsigned long this_pages;
122 next = pmd_addr_end(addr, end);
123 if (pmd_trans_huge(*pmd)) {
124 if (next - addr != HPAGE_PMD_SIZE)
125 split_huge_page_pmd(vma, addr, pmd);
127 int nr_ptes = change_huge_pmd(vma, pmd, addr,
131 if (nr_ptes == HPAGE_PMD_NR) {
132 pages += HPAGE_PMD_NR;
140 if (pmd_none_or_clear_bad(pmd))
142 this_pages = change_pte_range(vma, pmd, addr, next, newprot,
143 dirty_accountable, prot_numa);
145 } while (pmd++, addr = next, addr != end);
148 count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates);
152 static inline unsigned long change_pud_range(struct vm_area_struct *vma,
153 pgd_t *pgd, unsigned long addr, unsigned long end,
154 pgprot_t newprot, int dirty_accountable, int prot_numa)
158 unsigned long pages = 0;
160 pud = pud_offset(pgd, addr);
162 next = pud_addr_end(addr, end);
163 if (pud_none_or_clear_bad(pud))
165 pages += change_pmd_range(vma, pud, addr, next, newprot,
166 dirty_accountable, prot_numa);
167 } while (pud++, addr = next, addr != end);
172 static unsigned long change_protection_range(struct vm_area_struct *vma,
173 unsigned long addr, unsigned long end, pgprot_t newprot,
174 int dirty_accountable, int prot_numa)
176 struct mm_struct *mm = vma->vm_mm;
179 unsigned long start = addr;
180 unsigned long pages = 0;
183 pgd = pgd_offset(mm, addr);
184 flush_cache_range(vma, addr, end);
185 set_tlb_flush_pending(mm);
187 next = pgd_addr_end(addr, end);
188 if (pgd_none_or_clear_bad(pgd))
190 pages += change_pud_range(vma, pgd, addr, next, newprot,
191 dirty_accountable, prot_numa);
192 } while (pgd++, addr = next, addr != end);
194 /* Only flush the TLB if we actually modified any entries: */
196 flush_tlb_range(vma, start, end);
197 clear_tlb_flush_pending(mm);
202 unsigned long change_protection(struct vm_area_struct *vma, unsigned long start,
203 unsigned long end, pgprot_t newprot,
204 int dirty_accountable, int prot_numa)
206 struct mm_struct *mm = vma->vm_mm;
209 mmu_notifier_invalidate_range_start(mm, start, end);
210 if (is_vm_hugetlb_page(vma))
211 pages = hugetlb_change_protection(vma, start, end, newprot);
213 pages = change_protection_range(vma, start, end, newprot, dirty_accountable, prot_numa);
214 mmu_notifier_invalidate_range_end(mm, start, end);
220 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
221 unsigned long start, unsigned long end, unsigned long newflags)
223 struct mm_struct *mm = vma->vm_mm;
224 unsigned long oldflags = vma->vm_flags;
225 long nrpages = (end - start) >> PAGE_SHIFT;
226 unsigned long charged = 0;
229 int dirty_accountable = 0;
231 if (newflags == oldflags) {
237 * If we make a private mapping writable we increase our commit;
238 * but (without finer accounting) cannot reduce our commit if we
239 * make it unwritable again. hugetlb mapping were accounted for
240 * even if read-only so there is no need to account for them here
242 if (newflags & VM_WRITE) {
243 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
244 VM_SHARED|VM_NORESERVE))) {
246 if (security_vm_enough_memory_mm(mm, charged))
248 newflags |= VM_ACCOUNT;
253 * First try to merge with previous and/or next vma.
255 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
256 *pprev = vma_merge(mm, *pprev, start, end, newflags,
257 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
265 if (start != vma->vm_start) {
266 error = split_vma(mm, vma, start, 1);
271 if (end != vma->vm_end) {
272 error = split_vma(mm, vma, end, 0);
279 * vm_flags and vm_page_prot are protected by the mmap_sem
280 * held in write mode.
282 vma->vm_flags = newflags;
283 vma->vm_page_prot = pgprot_modify(vma->vm_page_prot,
284 vm_get_page_prot(newflags));
286 if (vma_wants_writenotify(vma)) {
287 vma->vm_page_prot = vm_get_page_prot(newflags & ~VM_SHARED);
288 dirty_accountable = 1;
291 change_protection(vma, start, end, vma->vm_page_prot,
292 dirty_accountable, 0);
294 vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
295 vm_stat_account(mm, newflags, vma->vm_file, nrpages);
296 perf_event_mmap(vma);
300 vm_unacct_memory(charged);
304 SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
307 unsigned long vm_flags, nstart, end, tmp, reqprot;
308 struct vm_area_struct *vma, *prev;
310 const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
311 prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
312 if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
315 if (start & ~PAGE_MASK)
319 len = PAGE_ALIGN(len);
323 if (!arch_validate_prot(prot))
328 * Does the application expect PROT_READ to imply PROT_EXEC:
330 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
333 vm_flags = calc_vm_prot_bits(prot);
335 down_write(¤t->mm->mmap_sem);
337 vma = find_vma(current->mm, start);
342 if (unlikely(grows & PROT_GROWSDOWN)) {
343 if (vma->vm_start >= end)
345 start = vma->vm_start;
347 if (!(vma->vm_flags & VM_GROWSDOWN))
350 if (vma->vm_start > start)
352 if (unlikely(grows & PROT_GROWSUP)) {
355 if (!(vma->vm_flags & VM_GROWSUP))
359 if (start > vma->vm_start)
362 for (nstart = start ; ; ) {
363 unsigned long newflags;
365 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
368 newflags |= (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
370 /* newflags >> 4 shift VM_MAY% in place of VM_% */
371 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
376 error = security_file_mprotect(vma, reqprot, prot);
383 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
388 if (nstart < prev->vm_end)
389 nstart = prev->vm_end;
394 if (!vma || vma->vm_start != nstart) {
400 up_write(¤t->mm->mmap_sem);