4 * (C) Copyright 1996 Linus Torvalds
6 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
7 * (C) Copyright 2002 Red Hat Inc, All Rights Reserved
11 #include <linux/hugetlb.h>
12 #include <linux/shm.h>
13 #include <linux/ksm.h>
14 #include <linux/mman.h>
15 #include <linux/swap.h>
16 #include <linux/capability.h>
18 #include <linux/highmem.h>
19 #include <linux/security.h>
20 #include <linux/syscalls.h>
21 #include <linux/mmu_notifier.h>
22 #include <linux/sched/sysctl.h>
24 #include <asm/uaccess.h>
25 #include <asm/cacheflush.h>
26 #include <asm/tlbflush.h>
30 static pmd_t *get_old_pmd(struct mm_struct *mm, unsigned long addr)
36 pgd = pgd_offset(mm, addr);
37 if (pgd_none_or_clear_bad(pgd))
40 pud = pud_offset(pgd, addr);
41 if (pud_none_or_clear_bad(pud))
44 pmd = pmd_offset(pud, addr);
51 static pmd_t *alloc_new_pmd(struct mm_struct *mm, struct vm_area_struct *vma,
58 pgd = pgd_offset(mm, addr);
59 pud = pud_alloc(mm, pgd, addr);
63 pmd = pmd_alloc(mm, pud, addr);
67 VM_BUG_ON(pmd_trans_huge(*pmd));
72 static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd,
73 unsigned long old_addr, unsigned long old_end,
74 struct vm_area_struct *new_vma, pmd_t *new_pmd,
75 unsigned long new_addr, bool need_rmap_locks)
77 struct address_space *mapping = NULL;
78 struct anon_vma *anon_vma = NULL;
79 struct mm_struct *mm = vma->vm_mm;
80 pte_t *old_pte, *new_pte, pte;
81 spinlock_t *old_ptl, *new_ptl;
84 * When need_rmap_locks is true, we take the i_mmap_mutex and anon_vma
85 * locks to ensure that rmap will always observe either the old or the
86 * new ptes. This is the easiest way to avoid races with
87 * truncate_pagecache(), page migration, etc...
89 * When need_rmap_locks is false, we use other ways to avoid
92 * - During exec() shift_arg_pages(), we use a specially tagged vma
93 * which rmap call sites look for using is_vma_temporary_stack().
95 * - During mremap(), new_vma is often known to be placed after vma
96 * in rmap traversal order. This ensures rmap will always observe
97 * either the old pte, or the new pte, or both (the page table locks
98 * serialize access to individual ptes, but only rmap traversal
99 * order guarantees that we won't miss both the old and new ptes).
101 if (need_rmap_locks) {
103 mapping = vma->vm_file->f_mapping;
104 mutex_lock(&mapping->i_mmap_mutex);
107 anon_vma = vma->anon_vma;
108 anon_vma_lock_write(anon_vma);
113 * We don't have to worry about the ordering of src and dst
114 * pte locks because exclusive mmap_sem prevents deadlock.
116 old_pte = pte_offset_map_lock(mm, old_pmd, old_addr, &old_ptl);
117 new_pte = pte_offset_map(new_pmd, new_addr);
118 new_ptl = pte_lockptr(mm, new_pmd);
119 if (new_ptl != old_ptl)
120 spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
121 arch_enter_lazy_mmu_mode();
123 for (; old_addr < old_end; old_pte++, old_addr += PAGE_SIZE,
124 new_pte++, new_addr += PAGE_SIZE) {
125 if (pte_none(*old_pte))
127 pte = ptep_get_and_clear(mm, old_addr, old_pte);
128 pte = move_pte(pte, new_vma->vm_page_prot, old_addr, new_addr);
129 set_pte_at(mm, new_addr, new_pte, pte);
132 arch_leave_lazy_mmu_mode();
133 if (new_ptl != old_ptl)
134 spin_unlock(new_ptl);
135 pte_unmap(new_pte - 1);
136 pte_unmap_unlock(old_pte - 1, old_ptl);
138 anon_vma_unlock_write(anon_vma);
140 mutex_unlock(&mapping->i_mmap_mutex);
143 #define LATENCY_LIMIT (64 * PAGE_SIZE)
145 unsigned long move_page_tables(struct vm_area_struct *vma,
146 unsigned long old_addr, struct vm_area_struct *new_vma,
147 unsigned long new_addr, unsigned long len,
148 bool need_rmap_locks)
150 unsigned long extent, next, old_end;
151 pmd_t *old_pmd, *new_pmd;
152 bool need_flush = false;
153 unsigned long mmun_start; /* For mmu_notifiers */
154 unsigned long mmun_end; /* For mmu_notifiers */
156 old_end = old_addr + len;
157 flush_cache_range(vma, old_addr, old_end);
159 mmun_start = old_addr;
161 mmu_notifier_invalidate_range_start(vma->vm_mm, mmun_start, mmun_end);
163 for (; old_addr < old_end; old_addr += extent, new_addr += extent) {
165 next = (old_addr + PMD_SIZE) & PMD_MASK;
166 /* even if next overflowed, extent below will be ok */
167 extent = next - old_addr;
168 if (extent > old_end - old_addr)
169 extent = old_end - old_addr;
170 old_pmd = get_old_pmd(vma->vm_mm, old_addr);
173 new_pmd = alloc_new_pmd(vma->vm_mm, vma, new_addr);
176 if (pmd_trans_huge(*old_pmd)) {
178 if (extent == HPAGE_PMD_SIZE) {
179 VM_BUG_ON(vma->vm_file || !vma->anon_vma);
180 /* See comment in move_ptes() */
182 anon_vma_lock_write(vma->anon_vma);
183 err = move_huge_pmd(vma, new_vma, old_addr,
187 anon_vma_unlock_write(vma->anon_vma);
193 split_huge_page_pmd(vma, old_addr, old_pmd);
195 VM_BUG_ON(pmd_trans_huge(*old_pmd));
197 if (pmd_none(*new_pmd) && __pte_alloc(new_vma->vm_mm, new_vma,
200 next = (new_addr + PMD_SIZE) & PMD_MASK;
201 if (extent > next - new_addr)
202 extent = next - new_addr;
203 if (extent > LATENCY_LIMIT)
204 extent = LATENCY_LIMIT;
205 move_ptes(vma, old_pmd, old_addr, old_addr + extent,
206 new_vma, new_pmd, new_addr, need_rmap_locks);
209 if (likely(need_flush))
210 flush_tlb_range(vma, old_end-len, old_addr);
212 mmu_notifier_invalidate_range_end(vma->vm_mm, mmun_start, mmun_end);
214 return len + old_addr - old_end; /* how much done */
217 static unsigned long move_vma(struct vm_area_struct *vma,
218 unsigned long old_addr, unsigned long old_len,
219 unsigned long new_len, unsigned long new_addr, bool *locked)
221 struct mm_struct *mm = vma->vm_mm;
222 struct vm_area_struct *new_vma;
223 unsigned long vm_flags = vma->vm_flags;
224 unsigned long new_pgoff;
225 unsigned long moved_len;
226 unsigned long excess = 0;
227 unsigned long hiwater_vm;
230 bool need_rmap_locks;
233 * We'd prefer to avoid failure later on in do_munmap:
234 * which may split one vma into three before unmapping.
236 if (mm->map_count >= sysctl_max_map_count - 3)
240 * Advise KSM to break any KSM pages in the area to be moved:
241 * it would be confusing if they were to turn up at the new
242 * location, where they happen to coincide with different KSM
243 * pages recently unmapped. But leave vma->vm_flags as it was,
244 * so KSM can come around to merge on vma and new_vma afterwards.
246 err = ksm_madvise(vma, old_addr, old_addr + old_len,
247 MADV_UNMERGEABLE, &vm_flags);
251 new_pgoff = vma->vm_pgoff + ((old_addr - vma->vm_start) >> PAGE_SHIFT);
252 new_vma = copy_vma(&vma, new_addr, new_len, new_pgoff,
257 moved_len = move_page_tables(vma, old_addr, new_vma, new_addr, old_len,
259 if (moved_len < old_len) {
261 * On error, move entries back from new area to old,
262 * which will succeed since page tables still there,
263 * and then proceed to unmap new area instead of old.
265 move_page_tables(new_vma, new_addr, vma, old_addr, moved_len,
273 /* Conceal VM_ACCOUNT so old reservation is not undone */
274 if (vm_flags & VM_ACCOUNT) {
275 vma->vm_flags &= ~VM_ACCOUNT;
276 excess = vma->vm_end - vma->vm_start - old_len;
277 if (old_addr > vma->vm_start &&
278 old_addr + old_len < vma->vm_end)
283 * If we failed to move page tables we still do total_vm increment
284 * since do_munmap() will decrement it by old_len == new_len.
286 * Since total_vm is about to be raised artificially high for a
287 * moment, we need to restore high watermark afterwards: if stats
288 * are taken meanwhile, total_vm and hiwater_vm appear too high.
289 * If this were a serious issue, we'd add a flag to do_munmap().
291 hiwater_vm = mm->hiwater_vm;
292 vm_stat_account(mm, vma->vm_flags, vma->vm_file, new_len>>PAGE_SHIFT);
294 if (do_munmap(mm, old_addr, old_len) < 0) {
295 /* OOM: unable to split vma, just get accounts right */
296 vm_unacct_memory(excess >> PAGE_SHIFT);
299 mm->hiwater_vm = hiwater_vm;
301 /* Restore VM_ACCOUNT if one or two pieces of vma left */
303 vma->vm_flags |= VM_ACCOUNT;
305 vma->vm_next->vm_flags |= VM_ACCOUNT;
308 if (vm_flags & VM_LOCKED) {
309 mm->locked_vm += new_len >> PAGE_SHIFT;
316 static struct vm_area_struct *vma_to_resize(unsigned long addr,
317 unsigned long old_len, unsigned long new_len, unsigned long *p)
319 struct mm_struct *mm = current->mm;
320 struct vm_area_struct *vma = find_vma(mm, addr);
322 if (!vma || vma->vm_start > addr)
325 if (is_vm_hugetlb_page(vma))
328 /* We can't remap across vm area boundaries */
329 if (old_len > vma->vm_end - addr)
332 /* Need to be careful about a growing mapping */
333 if (new_len > old_len) {
336 if (vma->vm_flags & (VM_DONTEXPAND | VM_PFNMAP))
338 pgoff = (addr - vma->vm_start) >> PAGE_SHIFT;
339 pgoff += vma->vm_pgoff;
340 if (pgoff + (new_len >> PAGE_SHIFT) < pgoff)
344 if (vma->vm_flags & VM_LOCKED) {
345 unsigned long locked, lock_limit;
346 locked = mm->locked_vm << PAGE_SHIFT;
347 lock_limit = rlimit(RLIMIT_MEMLOCK);
348 locked += new_len - old_len;
349 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
353 if (!may_expand_vm(mm, (new_len - old_len) >> PAGE_SHIFT))
356 if (vma->vm_flags & VM_ACCOUNT) {
357 unsigned long charged = (new_len - old_len) >> PAGE_SHIFT;
358 if (security_vm_enough_memory_mm(mm, charged))
365 Efault: /* very odd choice for most of the cases, but... */
366 return ERR_PTR(-EFAULT);
368 return ERR_PTR(-EINVAL);
370 return ERR_PTR(-ENOMEM);
372 return ERR_PTR(-EAGAIN);
375 static unsigned long mremap_to(unsigned long addr, unsigned long old_len,
376 unsigned long new_addr, unsigned long new_len, bool *locked)
378 struct mm_struct *mm = current->mm;
379 struct vm_area_struct *vma;
380 unsigned long ret = -EINVAL;
381 unsigned long charged = 0;
382 unsigned long map_flags;
384 if (new_addr & ~PAGE_MASK)
387 if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len)
390 /* Check if the location we're moving into overlaps the
391 * old location at all, and fail if it does.
393 if ((new_addr <= addr) && (new_addr+new_len) > addr)
396 if ((addr <= new_addr) && (addr+old_len) > new_addr)
399 ret = do_munmap(mm, new_addr, new_len);
403 if (old_len >= new_len) {
404 ret = do_munmap(mm, addr+new_len, old_len - new_len);
405 if (ret && old_len != new_len)
410 vma = vma_to_resize(addr, old_len, new_len, &charged);
416 map_flags = MAP_FIXED;
417 if (vma->vm_flags & VM_MAYSHARE)
418 map_flags |= MAP_SHARED;
420 ret = get_unmapped_area(vma->vm_file, new_addr, new_len, vma->vm_pgoff +
421 ((addr - vma->vm_start) >> PAGE_SHIFT),
423 if (ret & ~PAGE_MASK)
426 ret = move_vma(vma, addr, old_len, new_len, new_addr, locked);
427 if (!(ret & ~PAGE_MASK))
430 vm_unacct_memory(charged);
436 static int vma_expandable(struct vm_area_struct *vma, unsigned long delta)
438 unsigned long end = vma->vm_end + delta;
439 if (end < vma->vm_end) /* overflow */
441 if (vma->vm_next && vma->vm_next->vm_start < end) /* intersection */
443 if (get_unmapped_area(NULL, vma->vm_start, end - vma->vm_start,
444 0, MAP_FIXED) & ~PAGE_MASK)
450 * Expand (or shrink) an existing mapping, potentially moving it at the
451 * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
453 * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
454 * This option implies MREMAP_MAYMOVE.
456 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
457 unsigned long, new_len, unsigned long, flags,
458 unsigned long, new_addr)
460 struct mm_struct *mm = current->mm;
461 struct vm_area_struct *vma;
462 unsigned long ret = -EINVAL;
463 unsigned long charged = 0;
466 down_write(¤t->mm->mmap_sem);
468 if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE))
471 if (addr & ~PAGE_MASK)
474 old_len = PAGE_ALIGN(old_len);
475 new_len = PAGE_ALIGN(new_len);
478 * We allow a zero old-len as a special case
479 * for DOS-emu "duplicate shm area" thing. But
480 * a zero new-len is nonsensical.
485 if (flags & MREMAP_FIXED) {
486 if (flags & MREMAP_MAYMOVE)
487 ret = mremap_to(addr, old_len, new_addr, new_len,
493 * Always allow a shrinking remap: that just unmaps
494 * the unnecessary pages..
495 * do_munmap does all the needed commit accounting
497 if (old_len >= new_len) {
498 ret = do_munmap(mm, addr+new_len, old_len - new_len);
499 if (ret && old_len != new_len)
506 * Ok, we need to grow..
508 vma = vma_to_resize(addr, old_len, new_len, &charged);
514 /* old_len exactly to the end of the area..
516 if (old_len == vma->vm_end - addr) {
517 /* can we just expand the current mapping? */
518 if (vma_expandable(vma, new_len - old_len)) {
519 int pages = (new_len - old_len) >> PAGE_SHIFT;
521 if (vma_adjust(vma, vma->vm_start, addr + new_len,
522 vma->vm_pgoff, NULL)) {
527 vm_stat_account(mm, vma->vm_flags, vma->vm_file, pages);
528 if (vma->vm_flags & VM_LOCKED) {
529 mm->locked_vm += pages;
539 * We weren't able to just expand or shrink the area,
540 * we need to create a new one and move it..
543 if (flags & MREMAP_MAYMOVE) {
544 unsigned long map_flags = 0;
545 if (vma->vm_flags & VM_MAYSHARE)
546 map_flags |= MAP_SHARED;
548 new_addr = get_unmapped_area(vma->vm_file, 0, new_len,
550 ((addr - vma->vm_start) >> PAGE_SHIFT),
552 if (new_addr & ~PAGE_MASK) {
557 ret = move_vma(vma, addr, old_len, new_len, new_addr, &locked);
560 if (ret & ~PAGE_MASK)
561 vm_unacct_memory(charged);
562 up_write(¤t->mm->mmap_sem);
563 if (locked && new_len > old_len)
564 mm_populate(new_addr + old_len, new_len - old_len);