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 <asm/uaccess.h>
27 #include <asm/pgtable.h>
28 #include <asm/cacheflush.h>
29 #include <asm/tlbflush.h>
32 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
38 static unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
39 unsigned long addr, unsigned long end, pgprot_t newprot,
40 int dirty_accountable, int prot_numa, bool *ret_all_same_node)
42 struct mm_struct *mm = vma->vm_mm;
45 unsigned long pages = 0;
46 bool all_same_node = true;
49 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
50 arch_enter_lazy_mmu_mode();
53 if (pte_present(oldpte)) {
57 ptent = ptep_modify_prot_start(mm, addr, pte);
59 ptent = pte_modify(ptent, newprot);
64 page = vm_normal_page(vma, addr, oldpte);
66 int this_nid = page_to_nid(page);
69 if (last_nid != this_nid)
70 all_same_node = false;
72 /* only check non-shared pages */
73 if (!pte_numa(oldpte) &&
74 page_mapcount(page) == 1) {
75 ptent = pte_mknuma(ptent);
82 * Avoid taking write faults for pages we know to be
85 if (dirty_accountable && pte_dirty(ptent)) {
86 ptent = pte_mkwrite(ptent);
92 ptep_modify_prot_commit(mm, addr, pte, ptent);
93 } else if (IS_ENABLED(CONFIG_MIGRATION) && !pte_file(oldpte)) {
94 swp_entry_t entry = pte_to_swp_entry(oldpte);
96 if (is_write_migration_entry(entry)) {
98 * A protection check is difficult so
99 * just be safe and disable write
101 make_migration_entry_read(&entry);
102 set_pte_at(mm, addr, pte,
103 swp_entry_to_pte(entry));
107 } while (pte++, addr += PAGE_SIZE, addr != end);
108 arch_leave_lazy_mmu_mode();
109 pte_unmap_unlock(pte - 1, ptl);
111 *ret_all_same_node = all_same_node;
115 #ifdef CONFIG_NUMA_BALANCING
116 static inline void change_pmd_protnuma(struct mm_struct *mm, unsigned long addr,
119 spin_lock(&mm->page_table_lock);
120 set_pmd_at(mm, addr & PMD_MASK, pmd, pmd_mknuma(*pmd));
121 spin_unlock(&mm->page_table_lock);
124 static inline void change_pmd_protnuma(struct mm_struct *mm, unsigned long addr,
129 #endif /* CONFIG_NUMA_BALANCING */
131 static inline unsigned long change_pmd_range(struct vm_area_struct *vma,
132 pud_t *pud, unsigned long addr, unsigned long end,
133 pgprot_t newprot, int dirty_accountable, int prot_numa)
137 unsigned long pages = 0;
138 unsigned long nr_huge_updates = 0;
141 pmd = pmd_offset(pud, addr);
143 next = pmd_addr_end(addr, end);
144 if (pmd_trans_huge(*pmd)) {
145 if (next - addr != HPAGE_PMD_SIZE)
146 split_huge_page_pmd(vma, addr, pmd);
147 else if (change_huge_pmd(vma, pmd, addr, newprot,
149 pages += HPAGE_PMD_NR;
155 if (pmd_none_or_clear_bad(pmd))
157 pages += change_pte_range(vma, pmd, addr, next, newprot,
158 dirty_accountable, prot_numa, &all_same_node);
161 * If we are changing protections for NUMA hinting faults then
162 * set pmd_numa if the examined pages were all on the same
163 * node. This allows a regular PMD to be handled as one fault
164 * and effectively batches the taking of the PTL
166 if (prot_numa && all_same_node)
167 change_pmd_protnuma(vma->vm_mm, addr, pmd);
168 } while (pmd++, addr = next, addr != end);
171 count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates);
176 static inline unsigned long change_pud_range(struct vm_area_struct *vma,
177 pgd_t *pgd, unsigned long addr, unsigned long end,
178 pgprot_t newprot, int dirty_accountable, int prot_numa)
182 unsigned long pages = 0;
184 pud = pud_offset(pgd, addr);
186 next = pud_addr_end(addr, end);
187 if (pud_none_or_clear_bad(pud))
189 pages += change_pmd_range(vma, pud, addr, next, newprot,
190 dirty_accountable, prot_numa);
191 } while (pud++, addr = next, addr != end);
196 static unsigned long change_protection_range(struct vm_area_struct *vma,
197 unsigned long addr, unsigned long end, pgprot_t newprot,
198 int dirty_accountable, int prot_numa)
200 struct mm_struct *mm = vma->vm_mm;
203 unsigned long start = addr;
204 unsigned long pages = 0;
207 pgd = pgd_offset(mm, addr);
208 flush_cache_range(vma, addr, end);
210 next = pgd_addr_end(addr, end);
211 if (pgd_none_or_clear_bad(pgd))
213 pages += change_pud_range(vma, pgd, addr, next, newprot,
214 dirty_accountable, prot_numa);
215 } while (pgd++, addr = next, addr != end);
217 /* Only flush the TLB if we actually modified any entries: */
219 flush_tlb_range(vma, start, end);
224 unsigned long change_protection(struct vm_area_struct *vma, unsigned long start,
225 unsigned long end, pgprot_t newprot,
226 int dirty_accountable, int prot_numa)
228 struct mm_struct *mm = vma->vm_mm;
231 mmu_notifier_invalidate_range_start(mm, start, end);
232 if (is_vm_hugetlb_page(vma))
233 pages = hugetlb_change_protection(vma, start, end, newprot);
235 pages = change_protection_range(vma, start, end, newprot, dirty_accountable, prot_numa);
236 mmu_notifier_invalidate_range_end(mm, start, end);
242 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
243 unsigned long start, unsigned long end, unsigned long newflags)
245 struct mm_struct *mm = vma->vm_mm;
246 unsigned long oldflags = vma->vm_flags;
247 long nrpages = (end - start) >> PAGE_SHIFT;
248 unsigned long charged = 0;
251 int dirty_accountable = 0;
253 if (newflags == oldflags) {
259 * If we make a private mapping writable we increase our commit;
260 * but (without finer accounting) cannot reduce our commit if we
261 * make it unwritable again. hugetlb mapping were accounted for
262 * even if read-only so there is no need to account for them here
264 if (newflags & VM_WRITE) {
265 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
266 VM_SHARED|VM_NORESERVE))) {
268 if (security_vm_enough_memory_mm(mm, charged))
270 newflags |= VM_ACCOUNT;
275 * First try to merge with previous and/or next vma.
277 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
278 *pprev = vma_merge(mm, *pprev, start, end, newflags,
279 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
287 if (start != vma->vm_start) {
288 error = split_vma(mm, vma, start, 1);
293 if (end != vma->vm_end) {
294 error = split_vma(mm, vma, end, 0);
301 * vm_flags and vm_page_prot are protected by the mmap_sem
302 * held in write mode.
304 vma->vm_flags = newflags;
305 vma->vm_page_prot = pgprot_modify(vma->vm_page_prot,
306 vm_get_page_prot(newflags));
308 if (vma_wants_writenotify(vma)) {
309 vma->vm_page_prot = vm_get_page_prot(newflags & ~VM_SHARED);
310 dirty_accountable = 1;
313 change_protection(vma, start, end, vma->vm_page_prot,
314 dirty_accountable, 0);
316 vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
317 vm_stat_account(mm, newflags, vma->vm_file, nrpages);
318 perf_event_mmap(vma);
322 vm_unacct_memory(charged);
326 SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
329 unsigned long vm_flags, nstart, end, tmp, reqprot;
330 struct vm_area_struct *vma, *prev;
332 const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
333 prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
334 if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
337 if (start & ~PAGE_MASK)
341 len = PAGE_ALIGN(len);
345 if (!arch_validate_prot(prot))
350 * Does the application expect PROT_READ to imply PROT_EXEC:
352 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
355 vm_flags = calc_vm_prot_bits(prot);
357 down_write(¤t->mm->mmap_sem);
359 vma = find_vma(current->mm, start);
364 if (unlikely(grows & PROT_GROWSDOWN)) {
365 if (vma->vm_start >= end)
367 start = vma->vm_start;
369 if (!(vma->vm_flags & VM_GROWSDOWN))
372 if (vma->vm_start > start)
374 if (unlikely(grows & PROT_GROWSUP)) {
377 if (!(vma->vm_flags & VM_GROWSUP))
381 if (start > vma->vm_start)
384 for (nstart = start ; ; ) {
385 unsigned long newflags;
387 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
390 newflags |= (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
392 /* newflags >> 4 shift VM_MAY% in place of VM_% */
393 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
398 error = security_file_mprotect(vma, reqprot, prot);
405 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
410 if (nstart < prev->vm_end)
411 nstart = prev->vm_end;
416 if (!vma || vma->vm_start != nstart) {
422 up_write(¤t->mm->mmap_sem);