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_nidpid)
42 struct mm_struct *mm = vma->vm_mm;
45 unsigned long pages = 0;
46 bool all_same_nidpid = true;
50 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
51 arch_enter_lazy_mmu_mode();
54 if (pte_present(oldpte)) {
58 ptent = ptep_modify_prot_start(mm, addr, pte);
60 ptent = pte_modify(ptent, newprot);
65 page = vm_normal_page(vma, addr, oldpte);
67 int nidpid = page_nidpid_last(page);
68 int this_nid = nidpid_to_nid(nidpid);
69 int this_pid = nidpid_to_pid(nidpid);
75 if (last_nid != this_nid ||
76 last_pid != this_pid) {
77 all_same_nidpid = false;
80 if (!pte_numa(oldpte)) {
81 ptent = pte_mknuma(ptent);
88 * Avoid taking write faults for pages we know to be
91 if (dirty_accountable && pte_dirty(ptent)) {
92 ptent = pte_mkwrite(ptent);
98 ptep_modify_prot_commit(mm, addr, pte, ptent);
99 } else if (IS_ENABLED(CONFIG_MIGRATION) && !pte_file(oldpte)) {
100 swp_entry_t entry = pte_to_swp_entry(oldpte);
102 if (is_write_migration_entry(entry)) {
104 * A protection check is difficult so
105 * just be safe and disable write
107 make_migration_entry_read(&entry);
108 set_pte_at(mm, addr, pte,
109 swp_entry_to_pte(entry));
114 } while (pte++, addr += PAGE_SIZE, addr != end);
115 arch_leave_lazy_mmu_mode();
116 pte_unmap_unlock(pte - 1, ptl);
118 *ret_all_same_nidpid = all_same_nidpid;
122 #ifdef CONFIG_NUMA_BALANCING
123 static inline void change_pmd_protnuma(struct mm_struct *mm, unsigned long addr,
126 spin_lock(&mm->page_table_lock);
127 set_pmd_at(mm, addr & PMD_MASK, pmd, pmd_mknuma(*pmd));
128 spin_unlock(&mm->page_table_lock);
131 static inline void change_pmd_protnuma(struct mm_struct *mm, unsigned long addr,
136 #endif /* CONFIG_NUMA_BALANCING */
138 static inline unsigned long change_pmd_range(struct vm_area_struct *vma,
139 pud_t *pud, unsigned long addr, unsigned long end,
140 pgprot_t newprot, int dirty_accountable, int prot_numa)
144 unsigned long pages = 0;
145 bool all_same_nidpid;
147 pmd = pmd_offset(pud, addr);
149 unsigned long this_pages;
151 next = pmd_addr_end(addr, end);
152 if (pmd_trans_huge(*pmd)) {
153 if (next - addr != HPAGE_PMD_SIZE)
154 split_huge_page_pmd(vma, addr, pmd);
156 int nr_ptes = change_huge_pmd(vma, pmd, addr,
160 if (nr_ptes == HPAGE_PMD_NR)
168 if (pmd_none_or_clear_bad(pmd))
170 this_pages = change_pte_range(vma, pmd, addr, next, newprot,
171 dirty_accountable, prot_numa, &all_same_nidpid);
175 * If we are changing protections for NUMA hinting faults then
176 * set pmd_numa if the examined pages were all on the same
177 * node. This allows a regular PMD to be handled as one fault
178 * and effectively batches the taking of the PTL
180 if (prot_numa && this_pages && all_same_nidpid)
181 change_pmd_protnuma(vma->vm_mm, addr, pmd);
182 } while (pmd++, addr = next, addr != end);
187 static inline unsigned long change_pud_range(struct vm_area_struct *vma,
188 pgd_t *pgd, unsigned long addr, unsigned long end,
189 pgprot_t newprot, int dirty_accountable, int prot_numa)
193 unsigned long pages = 0;
195 pud = pud_offset(pgd, addr);
197 next = pud_addr_end(addr, end);
198 if (pud_none_or_clear_bad(pud))
200 pages += change_pmd_range(vma, pud, addr, next, newprot,
201 dirty_accountable, prot_numa);
202 } while (pud++, addr = next, addr != end);
207 static unsigned long change_protection_range(struct vm_area_struct *vma,
208 unsigned long addr, unsigned long end, pgprot_t newprot,
209 int dirty_accountable, int prot_numa)
211 struct mm_struct *mm = vma->vm_mm;
214 unsigned long start = addr;
215 unsigned long pages = 0;
218 pgd = pgd_offset(mm, addr);
219 flush_cache_range(vma, addr, end);
221 next = pgd_addr_end(addr, end);
222 if (pgd_none_or_clear_bad(pgd))
224 pages += change_pud_range(vma, pgd, addr, next, newprot,
225 dirty_accountable, prot_numa);
226 } while (pgd++, addr = next, addr != end);
228 /* Only flush the TLB if we actually modified any entries: */
230 flush_tlb_range(vma, start, end);
235 unsigned long change_protection(struct vm_area_struct *vma, unsigned long start,
236 unsigned long end, pgprot_t newprot,
237 int dirty_accountable, int prot_numa)
239 struct mm_struct *mm = vma->vm_mm;
242 mmu_notifier_invalidate_range_start(mm, start, end);
243 if (is_vm_hugetlb_page(vma))
244 pages = hugetlb_change_protection(vma, start, end, newprot);
246 pages = change_protection_range(vma, start, end, newprot, dirty_accountable, prot_numa);
247 mmu_notifier_invalidate_range_end(mm, start, end);
253 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
254 unsigned long start, unsigned long end, unsigned long newflags)
256 struct mm_struct *mm = vma->vm_mm;
257 unsigned long oldflags = vma->vm_flags;
258 long nrpages = (end - start) >> PAGE_SHIFT;
259 unsigned long charged = 0;
262 int dirty_accountable = 0;
264 if (newflags == oldflags) {
270 * If we make a private mapping writable we increase our commit;
271 * but (without finer accounting) cannot reduce our commit if we
272 * make it unwritable again. hugetlb mapping were accounted for
273 * even if read-only so there is no need to account for them here
275 if (newflags & VM_WRITE) {
276 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
277 VM_SHARED|VM_NORESERVE))) {
279 if (security_vm_enough_memory_mm(mm, charged))
281 newflags |= VM_ACCOUNT;
286 * First try to merge with previous and/or next vma.
288 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
289 *pprev = vma_merge(mm, *pprev, start, end, newflags,
290 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
298 if (start != vma->vm_start) {
299 error = split_vma(mm, vma, start, 1);
304 if (end != vma->vm_end) {
305 error = split_vma(mm, vma, end, 0);
312 * vm_flags and vm_page_prot are protected by the mmap_sem
313 * held in write mode.
315 vma->vm_flags = newflags;
316 vma->vm_page_prot = pgprot_modify(vma->vm_page_prot,
317 vm_get_page_prot(newflags));
319 if (vma_wants_writenotify(vma)) {
320 vma->vm_page_prot = vm_get_page_prot(newflags & ~VM_SHARED);
321 dirty_accountable = 1;
324 change_protection(vma, start, end, vma->vm_page_prot,
325 dirty_accountable, 0);
327 vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
328 vm_stat_account(mm, newflags, vma->vm_file, nrpages);
329 perf_event_mmap(vma);
333 vm_unacct_memory(charged);
337 SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
340 unsigned long vm_flags, nstart, end, tmp, reqprot;
341 struct vm_area_struct *vma, *prev;
343 const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
344 prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
345 if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
348 if (start & ~PAGE_MASK)
352 len = PAGE_ALIGN(len);
356 if (!arch_validate_prot(prot))
361 * Does the application expect PROT_READ to imply PROT_EXEC:
363 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
366 vm_flags = calc_vm_prot_bits(prot);
368 down_write(¤t->mm->mmap_sem);
370 vma = find_vma(current->mm, start);
375 if (unlikely(grows & PROT_GROWSDOWN)) {
376 if (vma->vm_start >= end)
378 start = vma->vm_start;
380 if (!(vma->vm_flags & VM_GROWSDOWN))
383 if (vma->vm_start > start)
385 if (unlikely(grows & PROT_GROWSUP)) {
388 if (!(vma->vm_flags & VM_GROWSUP))
392 if (start > vma->vm_start)
395 for (nstart = start ; ; ) {
396 unsigned long newflags;
398 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
401 newflags |= (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
403 /* newflags >> 4 shift VM_MAY% in place of VM_% */
404 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
409 error = security_file_mprotect(vma, reqprot, prot);
416 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
421 if (nstart < prev->vm_end)
422 nstart = prev->vm_end;
427 if (!vma || vma->vm_start != nstart) {
433 up_write(¤t->mm->mmap_sem);