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>
19 #include <linux/syscalls.h>
20 #include <linux/vs_memory.h>
22 #include <asm/uaccess.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)
33 pgd = pgd_offset(mm, addr);
34 if (pgd_none_or_clear_bad(pgd))
37 pud = pud_offset(pgd, addr);
38 if (pud_none_or_clear_bad(pud))
41 pmd = pmd_offset(pud, addr);
42 if (pmd_none_or_clear_bad(pmd))
45 pte = pte_offset_map_nested(pmd, addr);
47 pte_unmap_nested(pte);
54 static pte_t *get_one_pte_map(struct mm_struct *mm, unsigned long addr)
60 pgd = pgd_offset(mm, addr);
61 if (pgd_none_or_clear_bad(pgd))
64 pud = pud_offset(pgd, addr);
65 if (pud_none_or_clear_bad(pud))
68 pmd = pmd_offset(pud, addr);
69 if (pmd_none_or_clear_bad(pmd))
72 return pte_offset_map(pmd, addr);
75 static inline pte_t *alloc_one_pte_map(struct mm_struct *mm, unsigned long addr)
82 pgd = pgd_offset(mm, addr);
84 pud = pud_alloc(mm, pgd, addr);
87 pmd = pmd_alloc(mm, pud, addr);
89 pte = pte_alloc_map(mm, pmd, addr);
94 move_one_page(struct vm_area_struct *vma, unsigned long old_addr,
95 struct vm_area_struct *new_vma, unsigned long new_addr)
97 struct address_space *mapping = NULL;
98 struct mm_struct *mm = vma->vm_mm;
104 * Subtle point from Rajesh Venkatasubramanian: before
105 * moving file-based ptes, we must lock vmtruncate out,
106 * since it might clean the dst vma before the src vma,
107 * and we propagate stale pages into the dst afterward.
109 mapping = vma->vm_file->f_mapping;
110 spin_lock(&mapping->i_mmap_lock);
111 if (new_vma->vm_truncate_count &&
112 new_vma->vm_truncate_count != vma->vm_truncate_count)
113 new_vma->vm_truncate_count = 0;
115 spin_lock(&mm->page_table_lock);
117 src = get_one_pte_map_nested(mm, old_addr);
120 * Look to see whether alloc_one_pte_map needs to perform a
121 * memory allocation. If it does then we need to drop the
124 dst = get_one_pte_map(mm, new_addr);
125 if (unlikely(!dst)) {
126 pte_unmap_nested(src);
128 spin_unlock(&mapping->i_mmap_lock);
129 dst = alloc_one_pte_map(mm, new_addr);
130 if (mapping && !spin_trylock(&mapping->i_mmap_lock)) {
131 spin_unlock(&mm->page_table_lock);
132 spin_lock(&mapping->i_mmap_lock);
133 spin_lock(&mm->page_table_lock);
135 src = get_one_pte_map_nested(mm, old_addr);
138 * Since alloc_one_pte_map can drop and re-acquire
139 * page_table_lock, we should re-check the src entry...
144 pte = ptep_clear_flush(vma, old_addr, src);
145 set_pte_at(mm, new_addr, dst, pte);
148 pte_unmap_nested(src);
153 spin_unlock(&mm->page_table_lock);
155 spin_unlock(&mapping->i_mmap_lock);
159 static unsigned long move_page_tables(struct vm_area_struct *vma,
160 unsigned long old_addr, struct vm_area_struct *new_vma,
161 unsigned long new_addr, unsigned long len)
163 unsigned long offset;
165 flush_cache_range(vma, old_addr, old_addr + len);
168 * This is not the clever way to do this, but we're taking the
169 * easy way out on the assumption that most remappings will be
170 * only a few pages.. This also makes error recovery easier.
172 for (offset = 0; offset < len; offset += PAGE_SIZE) {
173 if (move_one_page(vma, old_addr + offset,
174 new_vma, new_addr + offset) < 0)
181 static unsigned long move_vma(struct vm_area_struct *vma,
182 unsigned long old_addr, unsigned long old_len,
183 unsigned long new_len, unsigned long new_addr)
185 struct mm_struct *mm = vma->vm_mm;
186 struct vm_area_struct *new_vma;
187 unsigned long vm_flags = vma->vm_flags;
188 unsigned long new_pgoff;
189 unsigned long moved_len;
190 unsigned long excess = 0;
194 * We'd prefer to avoid failure later on in do_munmap:
195 * which may split one vma into three before unmapping.
197 if (mm->map_count >= sysctl_max_map_count - 3)
200 new_pgoff = vma->vm_pgoff + ((old_addr - vma->vm_start) >> PAGE_SHIFT);
201 new_vma = copy_vma(&vma, new_addr, new_len, new_pgoff);
205 moved_len = move_page_tables(vma, old_addr, new_vma, new_addr, old_len);
206 if (moved_len < old_len) {
208 * On error, move entries back from new area to old,
209 * which will succeed since page tables still there,
210 * and then proceed to unmap new area instead of old.
212 move_page_tables(new_vma, new_addr, vma, old_addr, moved_len);
219 /* Conceal VM_ACCOUNT so old reservation is not undone */
220 if (vm_flags & VM_ACCOUNT) {
221 vma->vm_flags &= ~VM_ACCOUNT;
222 excess = vma->vm_end - vma->vm_start - old_len;
223 if (old_addr > vma->vm_start &&
224 old_addr + old_len < vma->vm_end)
229 * if we failed to move page tables we still do total_vm increment
230 * since do_munmap() will decrement it by old_len == new_len
232 vx_vmpages_add(mm, new_len >> PAGE_SHIFT);
234 if (do_munmap(mm, old_addr, old_len) < 0) {
235 /* OOM: unable to split vma, just get accounts right */
236 vm_unacct_memory(excess >> PAGE_SHIFT);
240 /* Restore VM_ACCOUNT if one or two pieces of vma left */
242 vma->vm_flags |= VM_ACCOUNT;
244 vma->vm_next->vm_flags |= VM_ACCOUNT;
247 vx_vmpages_add(mm, new_len >> PAGE_SHIFT);
248 __vm_stat_account(mm, vma->vm_flags, vma->vm_file, new_len>>PAGE_SHIFT);
249 if (vm_flags & VM_LOCKED) {
250 vx_vmlocked_add(mm, new_len >> PAGE_SHIFT);
251 if (new_len > old_len)
252 make_pages_present(new_addr + old_len,
260 * Expand (or shrink) an existing mapping, potentially moving it at the
261 * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
263 * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
264 * This option implies MREMAP_MAYMOVE.
266 unsigned long do_mremap(unsigned long addr,
267 unsigned long old_len, unsigned long new_len,
268 unsigned long flags, unsigned long new_addr)
270 struct vm_area_struct *vma;
271 unsigned long ret = -EINVAL;
272 unsigned long charged = 0;
274 if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE))
277 if (addr & ~PAGE_MASK)
280 old_len = PAGE_ALIGN(old_len);
281 new_len = PAGE_ALIGN(new_len);
284 * We allow a zero old-len as a special case
285 * for DOS-emu "duplicate shm area" thing. But
286 * a zero new-len is nonsensical.
291 /* new_addr is only valid if MREMAP_FIXED is specified */
292 if (flags & MREMAP_FIXED) {
293 if (new_addr & ~PAGE_MASK)
295 if (!(flags & MREMAP_MAYMOVE))
298 if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len)
301 /* Check if the location we're moving into overlaps the
302 * old location at all, and fail if it does.
304 if ((new_addr <= addr) && (new_addr+new_len) > addr)
307 if ((addr <= new_addr) && (addr+old_len) > new_addr)
310 ret = do_munmap(current->mm, new_addr, new_len);
316 * Always allow a shrinking remap: that just unmaps
317 * the unnecessary pages..
318 * do_munmap does all the needed commit accounting
320 if (old_len >= new_len) {
321 ret = do_munmap(current->mm, addr+new_len, old_len - new_len);
322 if (ret && old_len != new_len)
325 if (!(flags & MREMAP_FIXED) || (new_addr == addr))
331 * Ok, we need to grow.. or relocate.
334 vma = find_vma(current->mm, addr);
335 if (!vma || vma->vm_start > addr)
337 if (is_vm_hugetlb_page(vma)) {
341 /* We can't remap across vm area boundaries */
342 if (old_len > vma->vm_end - addr)
344 if (vma->vm_flags & VM_DONTEXPAND) {
345 if (new_len > old_len)
348 if (vma->vm_flags & VM_LOCKED) {
349 unsigned long locked, lock_limit;
350 locked = current->mm->locked_vm << PAGE_SHIFT;
351 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
352 locked += new_len - old_len;
354 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
356 if (!vx_vmlocked_avail(current->mm,
357 (new_len - old_len) >> PAGE_SHIFT))
360 if (!may_expand_vm(current->mm, (new_len - old_len) >> PAGE_SHIFT)) {
365 /* check context space, maybe only Private writable mapping? */
366 if (!vx_vmpages_avail(current->mm, (new_len - old_len) >> PAGE_SHIFT))
369 if (vma->vm_flags & VM_ACCOUNT) {
370 charged = (new_len - old_len) >> PAGE_SHIFT;
371 if (security_vm_enough_memory(charged))
375 /* old_len exactly to the end of the area..
376 * And we're not relocating the area.
378 if (old_len == vma->vm_end - addr &&
379 !((flags & MREMAP_FIXED) && (addr != new_addr)) &&
380 (old_len != new_len || !(flags & MREMAP_MAYMOVE))) {
381 unsigned long max_addr = TASK_SIZE;
383 max_addr = vma->vm_next->vm_start;
384 /* can we just expand the current mapping? */
385 if (max_addr - addr >= new_len) {
386 int pages = (new_len - old_len) >> PAGE_SHIFT;
388 vma_adjust(vma, vma->vm_start,
389 addr + new_len, vma->vm_pgoff, NULL);
391 vx_vmpages_add(current->mm, pages);
392 __vm_stat_account(vma->vm_mm, vma->vm_flags,
393 vma->vm_file, pages);
394 if (vma->vm_flags & VM_LOCKED) {
395 vx_vmlocked_add(vma->vm_mm, pages);
396 make_pages_present(addr + old_len,
405 * We weren't able to just expand or shrink the area,
406 * we need to create a new one and move it..
409 if (flags & MREMAP_MAYMOVE) {
410 if (!(flags & MREMAP_FIXED)) {
411 unsigned long map_flags = 0;
412 if (vma->vm_flags & VM_MAYSHARE)
413 map_flags |= MAP_SHARED;
415 new_addr = get_unmapped_area_prot(vma->vm_file, 0, new_len,
416 vma->vm_pgoff, map_flags, vma->vm_flags & VM_EXEC);
418 if (new_addr & ~PAGE_MASK)
421 ret = move_vma(vma, addr, old_len, new_len, new_addr);
424 if (ret & ~PAGE_MASK)
425 vm_unacct_memory(charged);
430 asmlinkage unsigned long sys_mremap(unsigned long addr,
431 unsigned long old_len, unsigned long new_len,
432 unsigned long flags, unsigned long new_addr)
436 down_write(¤t->mm->mmap_sem);
437 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
438 up_write(¤t->mm->mmap_sem);