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>
23 #include <asm/uaccess.h>
24 #include <asm/cacheflush.h>
25 #include <asm/tlbflush.h>
29 static pmd_t *get_old_pmd(struct mm_struct *mm, unsigned long addr)
35 pgd = pgd_offset(mm, addr);
36 if (pgd_none_or_clear_bad(pgd))
39 pud = pud_offset(pgd, addr);
40 if (pud_none_or_clear_bad(pud))
43 pmd = pmd_offset(pud, addr);
44 if (pmd_none_or_clear_bad(pmd))
50 static pmd_t *alloc_new_pmd(struct mm_struct *mm, unsigned long addr)
56 pgd = pgd_offset(mm, addr);
57 pud = pud_alloc(mm, pgd, addr);
61 pmd = pmd_alloc(mm, pud, addr);
65 if (!pmd_present(*pmd) && __pte_alloc(mm, pmd, addr))
71 static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd,
72 unsigned long old_addr, unsigned long old_end,
73 struct vm_area_struct *new_vma, pmd_t *new_pmd,
74 unsigned long new_addr)
76 struct address_space *mapping = NULL;
77 struct mm_struct *mm = vma->vm_mm;
78 pte_t *old_pte, *new_pte, pte;
79 spinlock_t *old_ptl, *new_ptl;
80 unsigned long old_start;
83 mmu_notifier_invalidate_range_start(vma->vm_mm,
87 * Subtle point from Rajesh Venkatasubramanian: before
88 * moving file-based ptes, we must lock truncate_pagecache
89 * out, since it might clean the dst vma before the src vma,
90 * and we propagate stale pages into the dst afterward.
92 mapping = vma->vm_file->f_mapping;
93 spin_lock(&mapping->i_mmap_lock);
94 if (new_vma->vm_truncate_count &&
95 new_vma->vm_truncate_count != vma->vm_truncate_count)
96 new_vma->vm_truncate_count = 0;
100 * We don't have to worry about the ordering of src and dst
101 * pte locks because exclusive mmap_sem prevents deadlock.
103 old_pte = pte_offset_map_lock(mm, old_pmd, old_addr, &old_ptl);
104 new_pte = pte_offset_map_nested(new_pmd, new_addr);
105 new_ptl = pte_lockptr(mm, new_pmd);
106 if (new_ptl != old_ptl)
107 spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
108 arch_enter_lazy_mmu_mode();
110 for (; old_addr < old_end; old_pte++, old_addr += PAGE_SIZE,
111 new_pte++, new_addr += PAGE_SIZE) {
112 if (pte_none(*old_pte))
114 pte = ptep_clear_flush(vma, old_addr, old_pte);
115 pte = move_pte(pte, new_vma->vm_page_prot, old_addr, new_addr);
116 set_pte_at(mm, new_addr, new_pte, pte);
119 arch_leave_lazy_mmu_mode();
120 if (new_ptl != old_ptl)
121 spin_unlock(new_ptl);
122 pte_unmap_nested(new_pte - 1);
123 pte_unmap_unlock(old_pte - 1, old_ptl);
125 spin_unlock(&mapping->i_mmap_lock);
126 mmu_notifier_invalidate_range_end(vma->vm_mm, old_start, old_end);
129 #define LATENCY_LIMIT (64 * PAGE_SIZE)
131 unsigned long move_page_tables(struct vm_area_struct *vma,
132 unsigned long old_addr, struct vm_area_struct *new_vma,
133 unsigned long new_addr, unsigned long len)
135 unsigned long extent, next, old_end;
136 pmd_t *old_pmd, *new_pmd;
138 old_end = old_addr + len;
139 flush_cache_range(vma, old_addr, old_end);
141 for (; old_addr < old_end; old_addr += extent, new_addr += extent) {
143 next = (old_addr + PMD_SIZE) & PMD_MASK;
144 if (next - 1 > old_end)
146 extent = next - old_addr;
147 old_pmd = get_old_pmd(vma->vm_mm, old_addr);
150 new_pmd = alloc_new_pmd(vma->vm_mm, new_addr);
153 next = (new_addr + PMD_SIZE) & PMD_MASK;
154 if (extent > next - new_addr)
155 extent = next - new_addr;
156 if (extent > LATENCY_LIMIT)
157 extent = LATENCY_LIMIT;
158 move_ptes(vma, old_pmd, old_addr, old_addr + extent,
159 new_vma, new_pmd, new_addr);
162 return len + old_addr - old_end; /* how much done */
165 static unsigned long move_vma(struct vm_area_struct *vma,
166 unsigned long old_addr, unsigned long old_len,
167 unsigned long new_len, unsigned long new_addr)
169 struct mm_struct *mm = vma->vm_mm;
170 struct vm_area_struct *new_vma;
171 unsigned long vm_flags = vma->vm_flags;
172 unsigned long new_pgoff;
173 unsigned long moved_len;
174 unsigned long excess = 0;
175 unsigned long hiwater_vm;
180 * We'd prefer to avoid failure later on in do_munmap:
181 * which may split one vma into three before unmapping.
183 if (mm->map_count >= sysctl_max_map_count - 3)
187 * Advise KSM to break any KSM pages in the area to be moved:
188 * it would be confusing if they were to turn up at the new
189 * location, where they happen to coincide with different KSM
190 * pages recently unmapped. But leave vma->vm_flags as it was,
191 * so KSM can come around to merge on vma and new_vma afterwards.
193 err = ksm_madvise(vma, old_addr, old_addr + old_len,
194 MADV_UNMERGEABLE, &vm_flags);
198 new_pgoff = vma->vm_pgoff + ((old_addr - vma->vm_start) >> PAGE_SHIFT);
199 new_vma = copy_vma(&vma, new_addr, new_len, new_pgoff);
203 moved_len = move_page_tables(vma, old_addr, new_vma, new_addr, old_len);
204 if (moved_len < old_len) {
206 * On error, move entries back from new area to old,
207 * which will succeed since page tables still there,
208 * and then proceed to unmap new area instead of old.
210 move_page_tables(new_vma, new_addr, vma, old_addr, moved_len);
217 /* Conceal VM_ACCOUNT so old reservation is not undone */
218 if (vm_flags & VM_ACCOUNT) {
219 vma->vm_flags &= ~VM_ACCOUNT;
220 excess = vma->vm_end - vma->vm_start - old_len;
221 if (old_addr > vma->vm_start &&
222 old_addr + old_len < vma->vm_end)
227 * If we failed to move page tables we still do total_vm increment
228 * since do_munmap() will decrement it by old_len == new_len.
230 * Since total_vm is about to be raised artificially high for a
231 * moment, we need to restore high watermark afterwards: if stats
232 * are taken meanwhile, total_vm and hiwater_vm appear too high.
233 * If this were a serious issue, we'd add a flag to do_munmap().
235 hiwater_vm = mm->hiwater_vm;
236 mm->total_vm += new_len >> PAGE_SHIFT;
237 vm_stat_account(mm, vma->vm_flags, vma->vm_file, new_len>>PAGE_SHIFT);
239 if (do_munmap(mm, old_addr, old_len) < 0) {
240 /* OOM: unable to split vma, just get accounts right */
241 vm_unacct_memory(excess >> PAGE_SHIFT);
244 mm->hiwater_vm = hiwater_vm;
246 /* Restore VM_ACCOUNT if one or two pieces of vma left */
248 vma->vm_flags |= VM_ACCOUNT;
250 vma->vm_next->vm_flags |= VM_ACCOUNT;
253 if (vm_flags & VM_LOCKED) {
254 mm->locked_vm += new_len >> PAGE_SHIFT;
255 if (new_len > old_len)
256 mlock_vma_pages_range(new_vma, new_addr + old_len,
263 static struct vm_area_struct *vma_to_resize(unsigned long addr,
264 unsigned long old_len, unsigned long new_len, unsigned long *p)
266 struct mm_struct *mm = current->mm;
267 struct vm_area_struct *vma = find_vma(mm, addr);
269 if (!vma || vma->vm_start > addr)
272 if (is_vm_hugetlb_page(vma))
275 /* We can't remap across vm area boundaries */
276 if (old_len > vma->vm_end - addr)
279 if (vma->vm_flags & (VM_DONTEXPAND | VM_PFNMAP)) {
280 if (new_len > old_len)
284 if (vma->vm_flags & VM_LOCKED) {
285 unsigned long locked, lock_limit;
286 locked = mm->locked_vm << PAGE_SHIFT;
287 lock_limit = rlimit(RLIMIT_MEMLOCK);
288 locked += new_len - old_len;
289 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
293 if (!may_expand_vm(mm, (new_len - old_len) >> PAGE_SHIFT))
296 if (vma->vm_flags & VM_ACCOUNT) {
297 unsigned long charged = (new_len - old_len) >> PAGE_SHIFT;
298 if (security_vm_enough_memory(charged))
305 Efault: /* very odd choice for most of the cases, but... */
306 return ERR_PTR(-EFAULT);
308 return ERR_PTR(-EINVAL);
310 return ERR_PTR(-ENOMEM);
312 return ERR_PTR(-EAGAIN);
315 static unsigned long mremap_to(unsigned long addr,
316 unsigned long old_len, unsigned long new_addr,
317 unsigned long new_len)
319 struct mm_struct *mm = current->mm;
320 struct vm_area_struct *vma;
321 unsigned long ret = -EINVAL;
322 unsigned long charged = 0;
323 unsigned long map_flags;
325 if (new_addr & ~PAGE_MASK)
328 if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len)
331 /* Check if the location we're moving into overlaps the
332 * old location at all, and fail if it does.
334 if ((new_addr <= addr) && (new_addr+new_len) > addr)
337 if ((addr <= new_addr) && (addr+old_len) > new_addr)
340 ret = security_file_mmap(NULL, 0, 0, 0, new_addr, 1);
344 ret = do_munmap(mm, new_addr, new_len);
348 if (old_len >= new_len) {
349 ret = do_munmap(mm, addr+new_len, old_len - new_len);
350 if (ret && old_len != new_len)
355 vma = vma_to_resize(addr, old_len, new_len, &charged);
361 map_flags = MAP_FIXED;
362 if (vma->vm_flags & VM_MAYSHARE)
363 map_flags |= MAP_SHARED;
365 ret = get_unmapped_area(vma->vm_file, new_addr, new_len, vma->vm_pgoff +
366 ((addr - vma->vm_start) >> PAGE_SHIFT),
368 if (ret & ~PAGE_MASK)
371 ret = move_vma(vma, addr, old_len, new_len, new_addr);
372 if (!(ret & ~PAGE_MASK))
375 vm_unacct_memory(charged);
381 static int vma_expandable(struct vm_area_struct *vma, unsigned long delta)
383 unsigned long end = vma->vm_end + delta;
384 if (end < vma->vm_end) /* overflow */
386 if (vma->vm_next && vma->vm_next->vm_start < end) /* intersection */
388 if (get_unmapped_area(NULL, vma->vm_start, end - vma->vm_start,
389 0, MAP_FIXED) & ~PAGE_MASK)
395 * Expand (or shrink) an existing mapping, potentially moving it at the
396 * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
398 * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
399 * This option implies MREMAP_MAYMOVE.
401 unsigned long do_mremap(unsigned long addr,
402 unsigned long old_len, unsigned long new_len,
403 unsigned long flags, unsigned long new_addr)
405 struct mm_struct *mm = current->mm;
406 struct vm_area_struct *vma;
407 unsigned long ret = -EINVAL;
408 unsigned long charged = 0;
410 if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE))
413 if (addr & ~PAGE_MASK)
416 old_len = PAGE_ALIGN(old_len);
417 new_len = PAGE_ALIGN(new_len);
420 * We allow a zero old-len as a special case
421 * for DOS-emu "duplicate shm area" thing. But
422 * a zero new-len is nonsensical.
427 if (flags & MREMAP_FIXED) {
428 if (flags & MREMAP_MAYMOVE)
429 ret = mremap_to(addr, old_len, new_addr, new_len);
434 * Always allow a shrinking remap: that just unmaps
435 * the unnecessary pages..
436 * do_munmap does all the needed commit accounting
438 if (old_len >= new_len) {
439 ret = do_munmap(mm, addr+new_len, old_len - new_len);
440 if (ret && old_len != new_len)
447 * Ok, we need to grow..
449 vma = vma_to_resize(addr, old_len, new_len, &charged);
455 /* old_len exactly to the end of the area..
457 if (old_len == vma->vm_end - addr) {
458 /* can we just expand the current mapping? */
459 if (vma_expandable(vma, new_len - old_len)) {
460 int pages = (new_len - old_len) >> PAGE_SHIFT;
462 if (vma_adjust(vma, vma->vm_start, addr + new_len,
463 vma->vm_pgoff, NULL)) {
468 mm->total_vm += pages;
469 vm_stat_account(mm, vma->vm_flags, vma->vm_file, pages);
470 if (vma->vm_flags & VM_LOCKED) {
471 mm->locked_vm += pages;
472 mlock_vma_pages_range(vma, addr + old_len,
481 * We weren't able to just expand or shrink the area,
482 * we need to create a new one and move it..
485 if (flags & MREMAP_MAYMOVE) {
486 unsigned long map_flags = 0;
487 if (vma->vm_flags & VM_MAYSHARE)
488 map_flags |= MAP_SHARED;
490 new_addr = get_unmapped_area(vma->vm_file, 0, new_len,
492 ((addr - vma->vm_start) >> PAGE_SHIFT),
494 if (new_addr & ~PAGE_MASK) {
499 ret = security_file_mmap(NULL, 0, 0, 0, new_addr, 1);
502 ret = move_vma(vma, addr, old_len, new_len, new_addr);
505 if (ret & ~PAGE_MASK)
506 vm_unacct_memory(charged);
510 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
511 unsigned long, new_len, unsigned long, flags,
512 unsigned long, new_addr)
516 down_write(¤t->mm->mmap_sem);
517 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
518 up_write(¤t->mm->mmap_sem);