4 * (C) Copyright 1996 Linus Torvalds
6 * Address space accounting code <alan@redhat.com>
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/mman.h>
15 #include <linux/swap.h>
17 #include <linux/highmem.h>
18 #include <linux/security.h>
20 #include <asm/uaccess.h>
21 #include <asm/cacheflush.h>
22 #include <asm/tlbflush.h>
24 static pte_t *get_one_pte_map_nested(struct mm_struct *mm, unsigned long addr)
30 pgd = pgd_offset(mm, addr);
39 pmd = pmd_offset(pgd, addr);
48 pte = pte_offset_map_nested(pmd, addr);
50 pte_unmap_nested(pte);
57 static pte_t *get_one_pte_map(struct mm_struct *mm, unsigned long addr)
62 pgd = pgd_offset(mm, addr);
65 pmd = pmd_offset(pgd, addr);
66 if (!pmd_present(*pmd))
68 return pte_offset_map(pmd, addr);
71 static inline pte_t *alloc_one_pte_map(struct mm_struct *mm, unsigned long addr)
76 pmd = pmd_alloc(mm, pgd_offset(mm, addr), addr);
78 pte = pte_alloc_map(mm, pmd, addr);
83 move_one_page(struct vm_area_struct *vma, unsigned long old_addr,
84 unsigned long new_addr)
86 struct address_space *mapping = NULL;
87 struct mm_struct *mm = vma->vm_mm;
93 * Subtle point from Rajesh Venkatasubramanian: before
94 * moving file-based ptes, we must lock vmtruncate out,
95 * since it might clean the dst vma before the src vma,
96 * and we propagate stale pages into the dst afterward.
98 mapping = vma->vm_file->f_mapping;
99 spin_lock(&mapping->i_mmap_lock);
101 spin_lock(&mm->page_table_lock);
103 src = get_one_pte_map_nested(mm, old_addr);
106 * Look to see whether alloc_one_pte_map needs to perform a
107 * memory allocation. If it does then we need to drop the
110 dst = get_one_pte_map(mm, new_addr);
111 if (unlikely(!dst)) {
112 pte_unmap_nested(src);
114 spin_unlock(&mapping->i_mmap_lock);
115 dst = alloc_one_pte_map(mm, new_addr);
116 if (mapping && !spin_trylock(&mapping->i_mmap_lock)) {
117 spin_unlock(&mm->page_table_lock);
118 spin_lock(&mapping->i_mmap_lock);
119 spin_lock(&mm->page_table_lock);
121 src = get_one_pte_map_nested(mm, old_addr);
124 * Since alloc_one_pte_map can drop and re-acquire
125 * page_table_lock, we should re-check the src entry...
130 pte = ptep_clear_flush(vma, old_addr, src);
134 pte_unmap_nested(src);
139 spin_unlock(&mm->page_table_lock);
141 spin_unlock(&mapping->i_mmap_lock);
145 static unsigned long move_page_tables(struct vm_area_struct *vma,
146 unsigned long new_addr, unsigned long old_addr,
149 unsigned long offset;
151 flush_cache_range(vma, old_addr, old_addr + len);
154 * This is not the clever way to do this, but we're taking the
155 * easy way out on the assumption that most remappings will be
156 * only a few pages.. This also makes error recovery easier.
158 for (offset = 0; offset < len; offset += PAGE_SIZE) {
159 if (move_one_page(vma, old_addr+offset, new_addr+offset) < 0)
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;
179 * We'd prefer to avoid failure later on in do_munmap:
180 * which may split one vma into three before unmapping.
182 if (mm->map_count >= sysctl_max_map_count - 3)
185 new_pgoff = vma->vm_pgoff + ((old_addr - vma->vm_start) >> PAGE_SHIFT);
186 new_vma = copy_vma(&vma, new_addr, new_len, new_pgoff);
190 moved_len = move_page_tables(vma, new_addr, old_addr, old_len);
191 if (moved_len < old_len) {
193 * On error, move entries back from new area to old,
194 * which will succeed since page tables still there,
195 * and then proceed to unmap new area instead of old.
197 move_page_tables(new_vma, old_addr, new_addr, moved_len);
204 /* Conceal VM_ACCOUNT so old reservation is not undone */
205 if (vm_flags & VM_ACCOUNT) {
206 vma->vm_flags &= ~VM_ACCOUNT;
207 excess = vma->vm_end - vma->vm_start - old_len;
208 if (old_addr > vma->vm_start &&
209 old_addr + old_len < vma->vm_end)
213 if (do_munmap(mm, old_addr, old_len) < 0) {
214 /* OOM: unable to split vma, just get accounts right */
215 vm_unacct_memory(excess >> PAGE_SHIFT);
219 /* Restore VM_ACCOUNT if one or two pieces of vma left */
221 vma->vm_flags |= VM_ACCOUNT;
223 vma->vm_next->vm_flags |= VM_ACCOUNT;
226 mm->total_vm += new_len >> PAGE_SHIFT;
227 if (vm_flags & VM_LOCKED) {
228 mm->locked_vm += new_len >> PAGE_SHIFT;
229 if (new_len > old_len)
230 make_pages_present(new_addr + old_len,
238 * Expand (or shrink) an existing mapping, potentially moving it at the
239 * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
241 * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
242 * This option implies MREMAP_MAYMOVE.
244 unsigned long do_mremap(unsigned long addr,
245 unsigned long old_len, unsigned long new_len,
246 unsigned long flags, unsigned long new_addr)
248 struct vm_area_struct *vma;
249 unsigned long ret = -EINVAL;
250 unsigned long charged = 0;
252 if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE))
255 if (addr & ~PAGE_MASK)
258 old_len = PAGE_ALIGN(old_len);
259 new_len = PAGE_ALIGN(new_len);
262 * We allow a zero old-len as a special case
263 * for DOS-emu "duplicate shm area" thing. But
264 * a zero new-len is nonsensical.
269 /* new_addr is only valid if MREMAP_FIXED is specified */
270 if (flags & MREMAP_FIXED) {
271 if (new_addr & ~PAGE_MASK)
273 if (!(flags & MREMAP_MAYMOVE))
276 if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len)
279 /* Check if the location we're moving into overlaps the
280 * old location at all, and fail if it does.
282 if ((new_addr <= addr) && (new_addr+new_len) > addr)
285 if ((addr <= new_addr) && (addr+old_len) > new_addr)
288 ret = do_munmap(current->mm, new_addr, new_len);
294 * Always allow a shrinking remap: that just unmaps
295 * the unnecessary pages..
296 * do_munmap does all the needed commit accounting
298 if (old_len >= new_len) {
299 ret = do_munmap(current->mm, addr+new_len, old_len - new_len);
300 if (ret && old_len != new_len)
303 if (!(flags & MREMAP_FIXED) || (new_addr == addr))
309 * Ok, we need to grow.. or relocate.
312 vma = find_vma(current->mm, addr);
313 if (!vma || vma->vm_start > addr)
315 if (is_vm_hugetlb_page(vma)) {
319 /* We can't remap across vm area boundaries */
320 if (old_len > vma->vm_end - addr)
322 if (vma->vm_flags & VM_DONTEXPAND) {
323 if (new_len > old_len)
326 if (vma->vm_flags & VM_LOCKED) {
327 unsigned long locked = current->mm->locked_vm << PAGE_SHIFT;
328 locked += new_len - old_len;
330 if (locked > current->rlim[RLIMIT_MEMLOCK].rlim_cur)
334 if ((current->mm->total_vm << PAGE_SHIFT) + (new_len - old_len)
335 > current->rlim[RLIMIT_AS].rlim_cur)
338 if (vma->vm_flags & VM_ACCOUNT) {
339 charged = (new_len - old_len) >> PAGE_SHIFT;
340 if (security_vm_enough_memory(charged))
344 /* old_len exactly to the end of the area..
345 * And we're not relocating the area.
347 if (old_len == vma->vm_end - addr &&
348 !((flags & MREMAP_FIXED) && (addr != new_addr)) &&
349 (old_len != new_len || !(flags & MREMAP_MAYMOVE))) {
350 unsigned long max_addr = TASK_SIZE;
352 max_addr = vma->vm_next->vm_start;
353 /* can we just expand the current mapping? */
354 if (max_addr - addr >= new_len) {
355 int pages = (new_len - old_len) >> PAGE_SHIFT;
357 vma_adjust(vma, vma->vm_start,
358 addr + new_len, vma->vm_pgoff, NULL);
360 current->mm->total_vm += pages;
361 if (vma->vm_flags & VM_LOCKED) {
362 current->mm->locked_vm += pages;
363 make_pages_present(addr + old_len,
372 * We weren't able to just expand or shrink the area,
373 * we need to create a new one and move it..
376 if (flags & MREMAP_MAYMOVE) {
377 if (!(flags & MREMAP_FIXED)) {
378 unsigned long map_flags = 0;
379 if (vma->vm_flags & VM_MAYSHARE)
380 map_flags |= MAP_SHARED;
382 new_addr = get_unmapped_area(vma->vm_file, 0, new_len,
383 vma->vm_pgoff, map_flags);
385 if (new_addr & ~PAGE_MASK)
388 ret = move_vma(vma, addr, old_len, new_len, new_addr);
391 if (ret & ~PAGE_MASK)
392 vm_unacct_memory(charged);
397 asmlinkage unsigned long sys_mremap(unsigned long addr,
398 unsigned long old_len, unsigned long new_len,
399 unsigned long flags, unsigned long new_addr)
403 down_write(¤t->mm->mmap_sem);
404 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
405 up_write(¤t->mm->mmap_sem);