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/slab.h>
13 #include <linux/shm.h>
14 #include <linux/ksm.h>
15 #include <linux/mman.h>
16 #include <linux/swap.h>
17 #include <linux/capability.h>
19 #include <linux/highmem.h>
20 #include <linux/security.h>
21 #include <linux/syscalls.h>
22 #include <linux/mmu_notifier.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);
45 if (pmd_none_or_clear_bad(pmd))
51 static pmd_t *alloc_new_pmd(struct mm_struct *mm, unsigned long addr)
57 pgd = pgd_offset(mm, addr);
58 pud = pud_alloc(mm, pgd, addr);
62 pmd = pmd_alloc(mm, pud, addr);
66 if (!pmd_present(*pmd) && __pte_alloc(mm, pmd, addr))
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)
77 struct address_space *mapping = NULL;
78 struct mm_struct *mm = vma->vm_mm;
79 pte_t *old_pte, *new_pte, pte;
80 spinlock_t *old_ptl, *new_ptl;
81 unsigned long old_start;
84 mmu_notifier_invalidate_range_start(vma->vm_mm,
88 * Subtle point from Rajesh Venkatasubramanian: before
89 * moving file-based ptes, we must lock vmtruncate out,
90 * since it might clean the dst vma before the src vma,
91 * and we propagate stale pages into the dst afterward.
93 mapping = vma->vm_file->f_mapping;
94 spin_lock(&mapping->i_mmap_lock);
95 if (new_vma->vm_truncate_count &&
96 new_vma->vm_truncate_count != vma->vm_truncate_count)
97 new_vma->vm_truncate_count = 0;
101 * We don't have to worry about the ordering of src and dst
102 * pte locks because exclusive mmap_sem prevents deadlock.
104 old_pte = pte_offset_map_lock(mm, old_pmd, old_addr, &old_ptl);
105 new_pte = pte_offset_map_nested(new_pmd, new_addr);
106 new_ptl = pte_lockptr(mm, new_pmd);
107 if (new_ptl != old_ptl)
108 spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
109 arch_enter_lazy_mmu_mode();
111 for (; old_addr < old_end; old_pte++, old_addr += PAGE_SIZE,
112 new_pte++, new_addr += PAGE_SIZE) {
113 if (pte_none(*old_pte))
115 pte = ptep_clear_flush(vma, old_addr, old_pte);
116 pte = move_pte(pte, new_vma->vm_page_prot, old_addr, new_addr);
117 set_pte_at(mm, new_addr, new_pte, pte);
120 arch_leave_lazy_mmu_mode();
121 if (new_ptl != old_ptl)
122 spin_unlock(new_ptl);
123 pte_unmap_nested(new_pte - 1);
124 pte_unmap_unlock(old_pte - 1, old_ptl);
126 spin_unlock(&mapping->i_mmap_lock);
127 mmu_notifier_invalidate_range_end(vma->vm_mm, old_start, old_end);
130 #define LATENCY_LIMIT (64 * PAGE_SIZE)
132 unsigned long move_page_tables(struct vm_area_struct *vma,
133 unsigned long old_addr, struct vm_area_struct *new_vma,
134 unsigned long new_addr, unsigned long len)
136 unsigned long extent, next, old_end;
137 pmd_t *old_pmd, *new_pmd;
139 old_end = old_addr + len;
140 flush_cache_range(vma, old_addr, old_end);
142 for (; old_addr < old_end; old_addr += extent, new_addr += extent) {
144 next = (old_addr + PMD_SIZE) & PMD_MASK;
145 if (next - 1 > old_end)
147 extent = next - old_addr;
148 old_pmd = get_old_pmd(vma->vm_mm, old_addr);
151 new_pmd = alloc_new_pmd(vma->vm_mm, new_addr);
154 next = (new_addr + PMD_SIZE) & PMD_MASK;
155 if (extent > next - new_addr)
156 extent = next - new_addr;
157 if (extent > LATENCY_LIMIT)
158 extent = LATENCY_LIMIT;
159 move_ptes(vma, old_pmd, old_addr, old_addr + extent,
160 new_vma, new_pmd, new_addr);
163 return len + old_addr - old_end; /* how much done */
166 static unsigned long move_vma(struct vm_area_struct *vma,
167 unsigned long old_addr, unsigned long old_len,
168 unsigned long new_len, unsigned long new_addr)
170 struct mm_struct *mm = vma->vm_mm;
171 struct vm_area_struct *new_vma;
172 unsigned long vm_flags = vma->vm_flags;
173 unsigned long new_pgoff;
174 unsigned long moved_len;
175 unsigned long excess = 0;
176 unsigned long hiwater_vm;
181 * We'd prefer to avoid failure later on in do_munmap:
182 * which may split one vma into three before unmapping.
184 if (mm->map_count >= sysctl_max_map_count - 3)
188 * Advise KSM to break any KSM pages in the area to be moved:
189 * it would be confusing if they were to turn up at the new
190 * location, where they happen to coincide with different KSM
191 * pages recently unmapped. But leave vma->vm_flags as it was,
192 * so KSM can come around to merge on vma and new_vma afterwards.
194 err = ksm_madvise(vma, old_addr, old_addr + old_len,
195 MADV_UNMERGEABLE, &vm_flags);
199 new_pgoff = vma->vm_pgoff + ((old_addr - vma->vm_start) >> PAGE_SHIFT);
200 new_vma = copy_vma(&vma, new_addr, new_len, new_pgoff);
204 moved_len = move_page_tables(vma, old_addr, new_vma, new_addr, old_len);
205 if (moved_len < old_len) {
207 * On error, move entries back from new area to old,
208 * which will succeed since page tables still there,
209 * and then proceed to unmap new area instead of old.
211 move_page_tables(new_vma, new_addr, vma, old_addr, moved_len);
218 /* Conceal VM_ACCOUNT so old reservation is not undone */
219 if (vm_flags & VM_ACCOUNT) {
220 vma->vm_flags &= ~VM_ACCOUNT;
221 excess = vma->vm_end - vma->vm_start - old_len;
222 if (old_addr > vma->vm_start &&
223 old_addr + old_len < vma->vm_end)
228 * If we failed to move page tables we still do total_vm increment
229 * since do_munmap() will decrement it by old_len == new_len.
231 * Since total_vm is about to be raised artificially high for a
232 * moment, we need to restore high watermark afterwards: if stats
233 * are taken meanwhile, total_vm and hiwater_vm appear too high.
234 * If this were a serious issue, we'd add a flag to do_munmap().
236 hiwater_vm = mm->hiwater_vm;
237 mm->total_vm += new_len >> PAGE_SHIFT;
238 vm_stat_account(mm, vma->vm_flags, vma->vm_file, new_len>>PAGE_SHIFT);
240 if (do_munmap(mm, old_addr, old_len) < 0) {
241 /* OOM: unable to split vma, just get accounts right */
242 vm_unacct_memory(excess >> PAGE_SHIFT);
245 mm->hiwater_vm = hiwater_vm;
247 /* Restore VM_ACCOUNT if one or two pieces of vma left */
249 vma->vm_flags |= VM_ACCOUNT;
251 vma->vm_next->vm_flags |= VM_ACCOUNT;
254 if (vm_flags & VM_LOCKED) {
255 mm->locked_vm += new_len >> PAGE_SHIFT;
256 if (new_len > old_len)
257 mlock_vma_pages_range(new_vma, new_addr + old_len,
265 * Expand (or shrink) an existing mapping, potentially moving it at the
266 * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
268 * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
269 * This option implies MREMAP_MAYMOVE.
271 unsigned long do_mremap(unsigned long addr,
272 unsigned long old_len, unsigned long new_len,
273 unsigned long flags, unsigned long new_addr)
275 struct mm_struct *mm = current->mm;
276 struct vm_area_struct *vma;
277 unsigned long ret = -EINVAL;
278 unsigned long charged = 0;
280 if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE))
283 if (addr & ~PAGE_MASK)
286 old_len = PAGE_ALIGN(old_len);
287 new_len = PAGE_ALIGN(new_len);
290 * We allow a zero old-len as a special case
291 * for DOS-emu "duplicate shm area" thing. But
292 * a zero new-len is nonsensical.
297 /* new_addr is only valid if MREMAP_FIXED is specified */
298 if (flags & MREMAP_FIXED) {
299 if (new_addr & ~PAGE_MASK)
301 if (!(flags & MREMAP_MAYMOVE))
304 if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len)
307 /* Check if the location we're moving into overlaps the
308 * old location at all, and fail if it does.
310 if ((new_addr <= addr) && (new_addr+new_len) > addr)
313 if ((addr <= new_addr) && (addr+old_len) > new_addr)
316 ret = security_file_mmap(NULL, 0, 0, 0, new_addr, 1);
320 ret = do_munmap(mm, new_addr, new_len);
326 * Always allow a shrinking remap: that just unmaps
327 * the unnecessary pages..
328 * do_munmap does all the needed commit accounting
330 if (old_len >= new_len) {
331 ret = do_munmap(mm, addr+new_len, old_len - new_len);
332 if (ret && old_len != new_len)
335 if (!(flags & MREMAP_FIXED) || (new_addr == addr))
341 * Ok, we need to grow.. or relocate.
344 vma = find_vma(mm, addr);
345 if (!vma || vma->vm_start > addr)
347 if (is_vm_hugetlb_page(vma)) {
351 /* We can't remap across vm area boundaries */
352 if (old_len > vma->vm_end - addr)
354 if (vma->vm_flags & (VM_DONTEXPAND | VM_PFNMAP)) {
355 if (new_len > old_len)
358 if (vma->vm_flags & VM_LOCKED) {
359 unsigned long locked, lock_limit;
360 locked = mm->locked_vm << PAGE_SHIFT;
361 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
362 locked += new_len - old_len;
364 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
367 if (!may_expand_vm(mm, (new_len - old_len) >> PAGE_SHIFT)) {
372 if (vma->vm_flags & VM_ACCOUNT) {
373 charged = (new_len - old_len) >> PAGE_SHIFT;
374 if (security_vm_enough_memory(charged))
378 /* old_len exactly to the end of the area..
379 * And we're not relocating the area.
381 if (old_len == vma->vm_end - addr &&
382 !((flags & MREMAP_FIXED) && (addr != new_addr)) &&
383 (old_len != new_len || !(flags & MREMAP_MAYMOVE))) {
384 unsigned long max_addr = TASK_SIZE;
386 max_addr = vma->vm_next->vm_start;
387 /* can we just expand the current mapping? */
388 if (max_addr - addr >= new_len) {
389 int pages = (new_len - old_len) >> PAGE_SHIFT;
391 vma_adjust(vma, vma->vm_start,
392 addr + new_len, vma->vm_pgoff, NULL);
394 mm->total_vm += pages;
395 vm_stat_account(mm, vma->vm_flags, vma->vm_file, pages);
396 if (vma->vm_flags & VM_LOCKED) {
397 mm->locked_vm += pages;
398 mlock_vma_pages_range(vma, addr + old_len,
407 * We weren't able to just expand or shrink the area,
408 * we need to create a new one and move it..
411 if (flags & MREMAP_MAYMOVE) {
412 if (!(flags & MREMAP_FIXED)) {
413 unsigned long map_flags = 0;
414 if (vma->vm_flags & VM_MAYSHARE)
415 map_flags |= MAP_SHARED;
417 new_addr = get_unmapped_area(vma->vm_file, 0, new_len,
418 vma->vm_pgoff, map_flags);
419 if (new_addr & ~PAGE_MASK) {
424 ret = security_file_mmap(NULL, 0, 0, 0, new_addr, 1);
428 ret = move_vma(vma, addr, old_len, new_len, new_addr);
431 if (ret & ~PAGE_MASK)
432 vm_unacct_memory(charged);
437 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
438 unsigned long, new_len, unsigned long, flags,
439 unsigned long, new_addr)
443 down_write(¤t->mm->mmap_sem);
444 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
445 up_write(¤t->mm->mmap_sem);