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/rmap-locking.h>
19 #include <linux/security.h>
21 #include <asm/uaccess.h>
22 #include <asm/pgalloc.h>
23 #include <asm/cacheflush.h>
24 #include <asm/tlbflush.h>
26 static pte_t *get_one_pte_map_nested(struct mm_struct *mm, unsigned long addr)
32 pgd = pgd_offset(mm, addr);
41 pmd = pmd_offset(pgd, addr);
50 pte = pte_offset_map_nested(pmd, addr);
52 pte_unmap_nested(pte);
59 static inline int page_table_present(struct mm_struct *mm, unsigned long addr)
64 pgd = pgd_offset(mm, addr);
67 pmd = pmd_offset(pgd, addr);
68 return pmd_present(*pmd);
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 copy_one_pte(struct vm_area_struct *vma, unsigned long old_addr,
84 pte_t *src, pte_t *dst, struct pte_chain **pte_chainp)
86 pte_t pte = ptep_clear_flush(vma, old_addr, src);
89 if (pte_present(pte)) {
90 unsigned long pfn = pte_pfn(pte);
92 struct page *page = pfn_to_page(pfn);
93 page_remove_rmap(page, src);
94 *pte_chainp = page_add_rmap(page, dst, *pte_chainp);
100 move_one_page(struct vm_area_struct *vma, unsigned long old_addr,
101 unsigned long new_addr)
103 struct mm_struct *mm = vma->vm_mm;
106 struct pte_chain *pte_chain;
108 pte_chain = pte_chain_alloc(GFP_KERNEL);
113 spin_lock(&mm->page_table_lock);
114 src = get_one_pte_map_nested(mm, old_addr);
117 * Look to see whether alloc_one_pte_map needs to perform a
118 * memory allocation. If it does then we need to drop the
121 if (!page_table_present(mm, new_addr)) {
122 pte_unmap_nested(src);
125 dst = alloc_one_pte_map(mm, new_addr);
127 src = get_one_pte_map_nested(mm, old_addr);
129 * Since alloc_one_pte_map can drop and re-acquire
130 * page_table_lock, we should re-check the src entry...
134 copy_one_pte(vma, old_addr, src,
138 pte_unmap_nested(src);
142 spin_unlock(&mm->page_table_lock);
143 pte_chain_free(pte_chain);
148 static int move_page_tables(struct vm_area_struct *vma,
149 unsigned long new_addr, unsigned long old_addr, unsigned long len)
151 unsigned long offset;
153 flush_cache_range(vma, old_addr, old_addr + len);
156 * This is not the clever way to do this, but we're taking the
157 * easy way out on the assumption that most remappings will be
158 * only a few pages.. This also makes error recovery easier.
160 for (offset = 0; offset < len; offset += PAGE_SIZE) {
161 if (move_one_page(vma, old_addr+offset, new_addr+offset) < 0)
167 static unsigned long move_vma(struct vm_area_struct *vma,
168 unsigned long old_addr, unsigned long old_len,
169 unsigned long new_len, unsigned long new_addr)
171 struct mm_struct *mm = vma->vm_mm;
172 struct vm_area_struct *new_vma;
173 unsigned long vm_flags = vma->vm_flags;
174 unsigned long new_pgoff;
175 unsigned long moved_len;
176 unsigned long excess = 0;
179 new_pgoff = vma->vm_pgoff + ((old_addr - vma->vm_start) >> PAGE_SHIFT);
180 new_vma = copy_vma(vma, new_addr, new_len, new_pgoff);
184 moved_len = move_page_tables(vma, new_addr, old_addr, old_len);
185 if (moved_len < old_len) {
187 * On error, move entries back from new area to old,
188 * which will succeed since page tables still there,
189 * and then proceed to unmap new area instead of old.
191 * Subtle point from Rajesh Venkatasubramanian: before
192 * moving file-based ptes, move new_vma before old vma
193 * in the i_mmap or i_mmap_shared list, so when racing
194 * against vmtruncate we cannot propagate pages to be
195 * truncated back from new_vma into just cleaned old.
197 vma_relink_file(vma, new_vma);
198 move_page_tables(new_vma, old_addr, new_addr, moved_len);
205 /* Conceal VM_ACCOUNT so old reservation is not undone */
206 if (vm_flags & VM_ACCOUNT) {
207 vma->vm_flags &= ~VM_ACCOUNT;
208 excess = vma->vm_end - vma->vm_start - old_len;
209 if (old_addr > vma->vm_start &&
210 old_addr + old_len < vma->vm_end)
214 if (do_munmap(mm, old_addr, old_len) < 0) {
215 /* OOM: unable to split vma, just get accounts right */
216 vm_unacct_memory(excess >> PAGE_SHIFT);
220 /* Restore VM_ACCOUNT if one or two pieces of vma left */
222 vma->vm_flags |= VM_ACCOUNT;
224 vma->vm_next->vm_flags |= VM_ACCOUNT;
227 mm->total_vm += new_len >> PAGE_SHIFT;
228 if (vm_flags & VM_LOCKED) {
229 mm->locked_vm += new_len >> PAGE_SHIFT;
230 if (new_len > old_len)
231 make_pages_present(new_addr + old_len,
239 * Expand (or shrink) an existing mapping, potentially moving it at the
240 * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
242 * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
243 * This option implies MREMAP_MAYMOVE.
245 unsigned long do_mremap(unsigned long addr,
246 unsigned long old_len, unsigned long new_len,
247 unsigned long flags, unsigned long new_addr)
249 struct vm_area_struct *vma;
250 unsigned long ret = -EINVAL;
251 unsigned long charged = 0;
253 if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE))
256 if (addr & ~PAGE_MASK)
259 old_len = PAGE_ALIGN(old_len);
260 new_len = PAGE_ALIGN(new_len);
263 * We allow a zero old-len as a special case
264 * for DOS-emu "duplicate shm area" thing. But
265 * a zero new-len is nonsensical.
270 /* new_addr is only valid if MREMAP_FIXED is specified */
271 if (flags & MREMAP_FIXED) {
272 if (new_addr & ~PAGE_MASK)
274 if (!(flags & MREMAP_MAYMOVE))
277 if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len)
280 /* Check if the location we're moving into overlaps the
281 * old location at all, and fail if it does.
283 if ((new_addr <= addr) && (new_addr+new_len) > addr)
286 if ((addr <= new_addr) && (addr+old_len) > new_addr)
289 ret = do_munmap(current->mm, new_addr, new_len);
295 * Always allow a shrinking remap: that just unmaps
296 * the unnecessary pages..
297 * do_munmap does all the needed commit accounting
299 if (old_len >= new_len) {
300 ret = do_munmap(current->mm, addr+new_len, old_len - new_len);
301 if (ret && old_len != new_len)
304 if (!(flags & MREMAP_FIXED) || (new_addr == addr))
310 * Ok, we need to grow.. or relocate.
313 vma = find_vma(current->mm, addr);
314 if (!vma || vma->vm_start > addr)
316 if (is_vm_hugetlb_page(vma)) {
320 /* We can't remap across vm area boundaries */
321 if (old_len > vma->vm_end - addr)
323 if (vma->vm_flags & VM_DONTEXPAND) {
324 if (new_len > old_len)
327 if (vma->vm_flags & VM_LOCKED) {
328 unsigned long locked = current->mm->locked_vm << PAGE_SHIFT;
329 locked += new_len - old_len;
331 if (locked > current->rlim[RLIMIT_MEMLOCK].rlim_cur)
335 if ((current->mm->total_vm << PAGE_SHIFT) + (new_len - old_len)
336 > current->rlim[RLIMIT_AS].rlim_cur)
339 if (vma->vm_flags & VM_ACCOUNT) {
340 charged = (new_len - old_len) >> PAGE_SHIFT;
341 if (security_vm_enough_memory(charged))
345 /* old_len exactly to the end of the area..
346 * And we're not relocating the area.
348 if (old_len == vma->vm_end - addr &&
349 !((flags & MREMAP_FIXED) && (addr != new_addr)) &&
350 (old_len != new_len || !(flags & MREMAP_MAYMOVE))) {
351 unsigned long max_addr = TASK_SIZE;
353 max_addr = vma->vm_next->vm_start;
354 /* can we just expand the current mapping? */
355 if (max_addr - addr >= new_len) {
356 int pages = (new_len - old_len) >> PAGE_SHIFT;
357 spin_lock(&vma->vm_mm->page_table_lock);
358 vma->vm_end = addr + new_len;
359 spin_unlock(&vma->vm_mm->page_table_lock);
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);