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)) {
80 * Avoid trapping faults against the zero or KSM
81 * pages. See similar comment in change_huge_pmd.
86 page = vm_normal_page(vma, addr, oldpte);
87 if (!page || PageKsm(page))
91 ptent = ptep_modify_prot_start(mm, addr, pte);
92 ptent = pte_modify(ptent, newprot);
94 /* Avoid taking write faults for known dirty pages */
95 if (dirty_accountable && pte_dirty(ptent) &&
96 (pte_soft_dirty(ptent) ||
97 !(vma->vm_flags & VM_SOFTDIRTY))) {
98 ptent = pte_mkwrite(ptent);
100 ptep_modify_prot_commit(mm, addr, pte, ptent);
102 } else if (IS_ENABLED(CONFIG_MIGRATION)) {
103 swp_entry_t entry = pte_to_swp_entry(oldpte);
105 if (is_write_migration_entry(entry)) {
108 * A protection check is difficult so
109 * just be safe and disable write
111 make_migration_entry_read(&entry);
112 newpte = swp_entry_to_pte(entry);
113 if (pte_swp_soft_dirty(oldpte))
114 newpte = pte_swp_mksoft_dirty(newpte);
115 set_pte_at(mm, addr, pte, newpte);
120 } while (pte++, addr += PAGE_SIZE, addr != end);
121 arch_leave_lazy_mmu_mode();
122 pte_unmap_unlock(pte - 1, ptl);
127 static inline unsigned long change_pmd_range(struct vm_area_struct *vma,
128 pud_t *pud, unsigned long addr, unsigned long end,
129 pgprot_t newprot, int dirty_accountable, int prot_numa)
132 struct mm_struct *mm = vma->vm_mm;
134 unsigned long pages = 0;
135 unsigned long nr_huge_updates = 0;
136 unsigned long mni_start = 0;
138 pmd = pmd_offset(pud, addr);
140 unsigned long this_pages;
142 next = pmd_addr_end(addr, end);
143 if (!pmd_trans_huge(*pmd) && pmd_none_or_clear_bad(pmd))
146 /* invoke the mmu notifier if the pmd is populated */
149 mmu_notifier_invalidate_range_start(mm, mni_start, 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) {
161 pages += HPAGE_PMD_NR;
165 /* huge pmd was handled */
169 /* fall through, the trans huge pmd just split */
171 this_pages = change_pte_range(vma, pmd, addr, next, newprot,
172 dirty_accountable, prot_numa);
174 } while (pmd++, addr = next, addr != end);
177 mmu_notifier_invalidate_range_end(mm, mni_start, end);
180 count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates);
184 static inline unsigned long change_pud_range(struct vm_area_struct *vma,
185 pgd_t *pgd, unsigned long addr, unsigned long end,
186 pgprot_t newprot, int dirty_accountable, int prot_numa)
190 unsigned long pages = 0;
192 pud = pud_offset(pgd, addr);
194 next = pud_addr_end(addr, end);
195 if (pud_none_or_clear_bad(pud))
197 pages += change_pmd_range(vma, pud, addr, next, newprot,
198 dirty_accountable, prot_numa);
199 } while (pud++, addr = next, addr != end);
204 static unsigned long change_protection_range(struct vm_area_struct *vma,
205 unsigned long addr, unsigned long end, pgprot_t newprot,
206 int dirty_accountable, int prot_numa)
208 struct mm_struct *mm = vma->vm_mm;
211 unsigned long start = addr;
212 unsigned long pages = 0;
215 pgd = pgd_offset(mm, addr);
216 flush_cache_range(vma, addr, end);
217 set_tlb_flush_pending(mm);
219 next = pgd_addr_end(addr, end);
220 if (pgd_none_or_clear_bad(pgd))
222 pages += change_pud_range(vma, pgd, addr, next, newprot,
223 dirty_accountable, prot_numa);
224 } while (pgd++, addr = next, addr != end);
226 /* Only flush the TLB if we actually modified any entries: */
228 flush_tlb_range(vma, start, end);
229 clear_tlb_flush_pending(mm);
234 unsigned long change_protection(struct vm_area_struct *vma, unsigned long start,
235 unsigned long end, pgprot_t newprot,
236 int dirty_accountable, int prot_numa)
240 if (is_vm_hugetlb_page(vma))
241 pages = hugetlb_change_protection(vma, start, end, newprot);
243 pages = change_protection_range(vma, start, end, newprot, dirty_accountable, prot_numa);
249 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
250 unsigned long start, unsigned long end, unsigned long newflags)
252 struct mm_struct *mm = vma->vm_mm;
253 unsigned long oldflags = vma->vm_flags;
254 long nrpages = (end - start) >> PAGE_SHIFT;
255 unsigned long charged = 0;
258 int dirty_accountable = 0;
260 if (newflags == oldflags) {
266 * If we make a private mapping writable we increase our commit;
267 * but (without finer accounting) cannot reduce our commit if we
268 * make it unwritable again. hugetlb mapping were accounted for
269 * even if read-only so there is no need to account for them here
271 if (newflags & VM_WRITE) {
272 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
273 VM_SHARED|VM_NORESERVE))) {
275 if (security_vm_enough_memory_mm(mm, charged))
277 newflags |= VM_ACCOUNT;
282 * First try to merge with previous and/or next vma.
284 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
285 *pprev = vma_merge(mm, *pprev, start, end, newflags,
286 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
294 if (start != vma->vm_start) {
295 error = split_vma(mm, vma, start, 1);
300 if (end != vma->vm_end) {
301 error = split_vma(mm, vma, end, 0);
308 * vm_flags and vm_page_prot are protected by the mmap_sem
309 * held in write mode.
311 vma->vm_flags = newflags;
312 dirty_accountable = vma_wants_writenotify(vma);
313 vma_set_page_prot(vma);
315 change_protection(vma, start, end, vma->vm_page_prot,
316 dirty_accountable, 0);
318 vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
319 vm_stat_account(mm, newflags, vma->vm_file, nrpages);
320 perf_event_mmap(vma);
324 vm_unacct_memory(charged);
328 SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
331 unsigned long vm_flags, nstart, end, tmp, reqprot;
332 struct vm_area_struct *vma, *prev;
334 const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
335 prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
336 if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
339 if (start & ~PAGE_MASK)
343 len = PAGE_ALIGN(len);
347 if (!arch_validate_prot(prot))
352 * Does the application expect PROT_READ to imply PROT_EXEC:
354 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
357 vm_flags = calc_vm_prot_bits(prot);
359 down_write(¤t->mm->mmap_sem);
361 vma = find_vma(current->mm, start);
366 if (unlikely(grows & PROT_GROWSDOWN)) {
367 if (vma->vm_start >= end)
369 start = vma->vm_start;
371 if (!(vma->vm_flags & VM_GROWSDOWN))
374 if (vma->vm_start > start)
376 if (unlikely(grows & PROT_GROWSUP)) {
379 if (!(vma->vm_flags & VM_GROWSUP))
383 if (start > vma->vm_start)
386 for (nstart = start ; ; ) {
387 unsigned long newflags;
389 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
392 newflags |= (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
394 /* newflags >> 4 shift VM_MAY% in place of VM_% */
395 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
400 error = security_file_mprotect(vma, reqprot, prot);
407 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
412 if (nstart < prev->vm_end)
413 nstart = prev->vm_end;
418 if (!vma || vma->vm_start != nstart) {
424 up_write(¤t->mm->mmap_sem);