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 * For a prot_numa update we only hold mmap_sem for read so there is a
34 * potential race with faulting where a pmd was temporarily none. This
35 * function checks for a transhuge pmd under the appropriate lock. It
36 * returns a pte if it was successfully locked or NULL if it raced with
37 * a transhuge insertion.
39 static pte_t *lock_pte_protection(struct vm_area_struct *vma, pmd_t *pmd,
40 unsigned long addr, int prot_numa, spinlock_t **ptl)
45 /* !prot_numa is protected by mmap_sem held for write */
47 return pte_offset_map_lock(vma->vm_mm, pmd, addr, ptl);
49 pmdl = pmd_lock(vma->vm_mm, pmd);
50 if (unlikely(pmd_trans_huge(*pmd) || pmd_none(*pmd))) {
55 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, ptl);
60 static unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
61 unsigned long addr, unsigned long end, pgprot_t newprot,
62 int dirty_accountable, int prot_numa)
64 struct mm_struct *mm = vma->vm_mm;
67 unsigned long pages = 0;
69 pte = lock_pte_protection(vma, pmd, addr, prot_numa, &ptl);
73 arch_enter_lazy_mmu_mode();
76 if (pte_present(oldpte)) {
79 ptent = ptep_modify_prot_start(mm, addr, pte);
80 ptent = pte_modify(ptent, newprot);
82 /* Avoid taking write faults for known dirty pages */
83 if (dirty_accountable && pte_dirty(ptent) &&
84 (pte_soft_dirty(ptent) ||
85 !(vma->vm_flags & VM_SOFTDIRTY))) {
86 ptent = pte_mkwrite(ptent);
88 ptep_modify_prot_commit(mm, addr, pte, ptent);
90 } else if (IS_ENABLED(CONFIG_MIGRATION)) {
91 swp_entry_t entry = pte_to_swp_entry(oldpte);
93 if (is_write_migration_entry(entry)) {
96 * A protection check is difficult so
97 * just be safe and disable write
99 make_migration_entry_read(&entry);
100 newpte = swp_entry_to_pte(entry);
101 if (pte_swp_soft_dirty(oldpte))
102 newpte = pte_swp_mksoft_dirty(newpte);
103 set_pte_at(mm, addr, pte, newpte);
108 } while (pte++, addr += PAGE_SIZE, addr != end);
109 arch_leave_lazy_mmu_mode();
110 pte_unmap_unlock(pte - 1, ptl);
115 static inline unsigned long change_pmd_range(struct vm_area_struct *vma,
116 pud_t *pud, unsigned long addr, unsigned long end,
117 pgprot_t newprot, int dirty_accountable, int prot_numa)
120 struct mm_struct *mm = vma->vm_mm;
122 unsigned long pages = 0;
123 unsigned long nr_huge_updates = 0;
124 unsigned long mni_start = 0;
126 pmd = pmd_offset(pud, addr);
128 unsigned long this_pages;
130 next = pmd_addr_end(addr, end);
131 if (!pmd_trans_huge(*pmd) && pmd_none_or_clear_bad(pmd))
134 /* invoke the mmu notifier if the pmd is populated */
137 mmu_notifier_invalidate_range_start(mm, mni_start, end);
140 if (pmd_trans_huge(*pmd)) {
141 if (next - addr != HPAGE_PMD_SIZE)
142 split_huge_page_pmd(vma, addr, pmd);
144 int nr_ptes = change_huge_pmd(vma, pmd, addr,
148 if (nr_ptes == HPAGE_PMD_NR) {
149 pages += HPAGE_PMD_NR;
153 /* huge pmd was handled */
157 /* fall through, the trans huge pmd just split */
159 this_pages = change_pte_range(vma, pmd, addr, next, newprot,
160 dirty_accountable, prot_numa);
162 } while (pmd++, addr = next, addr != end);
165 mmu_notifier_invalidate_range_end(mm, mni_start, end);
168 count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates);
172 static inline unsigned long change_pud_range(struct vm_area_struct *vma,
173 pgd_t *pgd, unsigned long addr, unsigned long end,
174 pgprot_t newprot, int dirty_accountable, int prot_numa)
178 unsigned long pages = 0;
180 pud = pud_offset(pgd, addr);
182 next = pud_addr_end(addr, end);
183 if (pud_none_or_clear_bad(pud))
185 pages += change_pmd_range(vma, pud, addr, next, newprot,
186 dirty_accountable, prot_numa);
187 } while (pud++, addr = next, addr != end);
192 static unsigned long change_protection_range(struct vm_area_struct *vma,
193 unsigned long addr, unsigned long end, pgprot_t newprot,
194 int dirty_accountable, int prot_numa)
196 struct mm_struct *mm = vma->vm_mm;
199 unsigned long start = addr;
200 unsigned long pages = 0;
203 pgd = pgd_offset(mm, addr);
204 flush_cache_range(vma, addr, end);
205 set_tlb_flush_pending(mm);
207 next = pgd_addr_end(addr, end);
208 if (pgd_none_or_clear_bad(pgd))
210 pages += change_pud_range(vma, pgd, addr, next, newprot,
211 dirty_accountable, prot_numa);
212 } while (pgd++, addr = next, addr != end);
214 /* Only flush the TLB if we actually modified any entries: */
216 flush_tlb_range(vma, start, end);
217 clear_tlb_flush_pending(mm);
222 unsigned long change_protection(struct vm_area_struct *vma, unsigned long start,
223 unsigned long end, pgprot_t newprot,
224 int dirty_accountable, int prot_numa)
228 if (is_vm_hugetlb_page(vma))
229 pages = hugetlb_change_protection(vma, start, end, newprot);
231 pages = change_protection_range(vma, start, end, newprot, dirty_accountable, prot_numa);
237 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
238 unsigned long start, unsigned long end, unsigned long newflags)
240 struct mm_struct *mm = vma->vm_mm;
241 unsigned long oldflags = vma->vm_flags;
242 long nrpages = (end - start) >> PAGE_SHIFT;
243 unsigned long charged = 0;
246 int dirty_accountable = 0;
248 if (newflags == oldflags) {
254 * If we make a private mapping writable we increase our commit;
255 * but (without finer accounting) cannot reduce our commit if we
256 * make it unwritable again. hugetlb mapping were accounted for
257 * even if read-only so there is no need to account for them here
259 if (newflags & VM_WRITE) {
260 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
261 VM_SHARED|VM_NORESERVE))) {
263 if (security_vm_enough_memory_mm(mm, charged))
265 newflags |= VM_ACCOUNT;
270 * First try to merge with previous and/or next vma.
272 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
273 *pprev = vma_merge(mm, *pprev, start, end, newflags,
274 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
282 if (start != vma->vm_start) {
283 error = split_vma(mm, vma, start, 1);
288 if (end != vma->vm_end) {
289 error = split_vma(mm, vma, end, 0);
296 * vm_flags and vm_page_prot are protected by the mmap_sem
297 * held in write mode.
299 vma->vm_flags = newflags;
300 dirty_accountable = vma_wants_writenotify(vma);
301 vma_set_page_prot(vma);
303 change_protection(vma, start, end, vma->vm_page_prot,
304 dirty_accountable, 0);
306 vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
307 vm_stat_account(mm, newflags, vma->vm_file, nrpages);
308 perf_event_mmap(vma);
312 vm_unacct_memory(charged);
316 SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
319 unsigned long vm_flags, nstart, end, tmp, reqprot;
320 struct vm_area_struct *vma, *prev;
322 const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
323 prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
324 if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
327 if (start & ~PAGE_MASK)
331 len = PAGE_ALIGN(len);
335 if (!arch_validate_prot(prot))
340 * Does the application expect PROT_READ to imply PROT_EXEC:
342 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
345 vm_flags = calc_vm_prot_bits(prot);
347 down_write(¤t->mm->mmap_sem);
349 vma = find_vma(current->mm, start);
354 if (unlikely(grows & PROT_GROWSDOWN)) {
355 if (vma->vm_start >= end)
357 start = vma->vm_start;
359 if (!(vma->vm_flags & VM_GROWSDOWN))
362 if (vma->vm_start > start)
364 if (unlikely(grows & PROT_GROWSUP)) {
367 if (!(vma->vm_flags & VM_GROWSUP))
371 if (start > vma->vm_start)
374 for (nstart = start ; ; ) {
375 unsigned long newflags;
377 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
380 newflags |= (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
382 /* newflags >> 4 shift VM_MAY% in place of VM_% */
383 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
388 error = security_file_mprotect(vma, reqprot, prot);
395 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
400 if (nstart < prev->vm_end)
401 nstart = prev->vm_end;
406 if (!vma || vma->vm_start != nstart) {
412 up_write(¤t->mm->mmap_sem);