[PARISC] Update bitops from parisc tree
[linux-2.6.22.y-op.git] / mm / mmap.c
blobfa11d91242e8ca4b0d25135b801a0cbc7855dd8c
1 /*
2 * mm/mmap.c
4 * Written by obz.
6 * Address space accounting code <alan@redhat.com>
7 */
9 #include <linux/slab.h>
10 #include <linux/mm.h>
11 #include <linux/shm.h>
12 #include <linux/mman.h>
13 #include <linux/pagemap.h>
14 #include <linux/swap.h>
15 #include <linux/syscalls.h>
16 #include <linux/init.h>
17 #include <linux/file.h>
18 #include <linux/fs.h>
19 #include <linux/personality.h>
20 #include <linux/security.h>
21 #include <linux/hugetlb.h>
22 #include <linux/profile.h>
23 #include <linux/module.h>
24 #include <linux/mount.h>
25 #include <linux/mempolicy.h>
26 #include <linux/rmap.h>
28 #include <asm/uaccess.h>
29 #include <asm/cacheflush.h>
30 #include <asm/tlb.h>
32 static void unmap_region(struct mm_struct *mm,
33 struct vm_area_struct *vma, struct vm_area_struct *prev,
34 unsigned long start, unsigned long end);
37 * WARNING: the debugging will use recursive algorithms so never enable this
38 * unless you know what you are doing.
40 #undef DEBUG_MM_RB
42 /* description of effects of mapping type and prot in current implementation.
43 * this is due to the limited x86 page protection hardware. The expected
44 * behavior is in parens:
46 * map_type prot
47 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
48 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
49 * w: (no) no w: (no) no w: (yes) yes w: (no) no
50 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
52 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
53 * w: (no) no w: (no) no w: (copy) copy w: (no) no
54 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
57 pgprot_t protection_map[16] = {
58 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
59 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
62 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
63 int sysctl_overcommit_ratio = 50; /* default is 50% */
64 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
65 atomic_t vm_committed_space = ATOMIC_INIT(0);
68 * Check that a process has enough memory to allocate a new virtual
69 * mapping. 0 means there is enough memory for the allocation to
70 * succeed and -ENOMEM implies there is not.
72 * We currently support three overcommit policies, which are set via the
73 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
75 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
76 * Additional code 2002 Jul 20 by Robert Love.
78 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
80 * Note this is a helper function intended to be used by LSMs which
81 * wish to use this logic.
83 int __vm_enough_memory(long pages, int cap_sys_admin)
85 unsigned long free, allowed;
87 vm_acct_memory(pages);
90 * Sometimes we want to use more memory than we have
92 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
93 return 0;
95 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
96 unsigned long n;
98 free = get_page_cache_size();
99 free += nr_swap_pages;
102 * Any slabs which are created with the
103 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
104 * which are reclaimable, under pressure. The dentry
105 * cache and most inode caches should fall into this
107 free += atomic_read(&slab_reclaim_pages);
110 * Leave the last 3% for root
112 if (!cap_sys_admin)
113 free -= free / 32;
115 if (free > pages)
116 return 0;
119 * nr_free_pages() is very expensive on large systems,
120 * only call if we're about to fail.
122 n = nr_free_pages();
123 if (!cap_sys_admin)
124 n -= n / 32;
125 free += n;
127 if (free > pages)
128 return 0;
129 vm_unacct_memory(pages);
130 return -ENOMEM;
133 allowed = (totalram_pages - hugetlb_total_pages())
134 * sysctl_overcommit_ratio / 100;
136 * Leave the last 3% for root
138 if (!cap_sys_admin)
139 allowed -= allowed / 32;
140 allowed += total_swap_pages;
142 /* Don't let a single process grow too big:
143 leave 3% of the size of this process for other processes */
144 allowed -= current->mm->total_vm / 32;
147 * cast `allowed' as a signed long because vm_committed_space
148 * sometimes has a negative value
150 if (atomic_read(&vm_committed_space) < (long)allowed)
151 return 0;
153 vm_unacct_memory(pages);
155 return -ENOMEM;
158 EXPORT_SYMBOL(sysctl_overcommit_memory);
159 EXPORT_SYMBOL(sysctl_overcommit_ratio);
160 EXPORT_SYMBOL(sysctl_max_map_count);
161 EXPORT_SYMBOL(vm_committed_space);
162 EXPORT_SYMBOL(__vm_enough_memory);
165 * Requires inode->i_mapping->i_mmap_lock
167 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
168 struct file *file, struct address_space *mapping)
170 if (vma->vm_flags & VM_DENYWRITE)
171 atomic_inc(&file->f_dentry->d_inode->i_writecount);
172 if (vma->vm_flags & VM_SHARED)
173 mapping->i_mmap_writable--;
175 flush_dcache_mmap_lock(mapping);
176 if (unlikely(vma->vm_flags & VM_NONLINEAR))
177 list_del_init(&vma->shared.vm_set.list);
178 else
179 vma_prio_tree_remove(vma, &mapping->i_mmap);
180 flush_dcache_mmap_unlock(mapping);
184 * Remove one vm structure and free it.
186 static void remove_vm_struct(struct vm_area_struct *vma)
188 struct file *file = vma->vm_file;
190 might_sleep();
191 if (file) {
192 struct address_space *mapping = file->f_mapping;
193 spin_lock(&mapping->i_mmap_lock);
194 __remove_shared_vm_struct(vma, file, mapping);
195 spin_unlock(&mapping->i_mmap_lock);
197 if (vma->vm_ops && vma->vm_ops->close)
198 vma->vm_ops->close(vma);
199 if (file)
200 fput(file);
201 anon_vma_unlink(vma);
202 mpol_free(vma_policy(vma));
203 kmem_cache_free(vm_area_cachep, vma);
206 asmlinkage unsigned long sys_brk(unsigned long brk)
208 unsigned long rlim, retval;
209 unsigned long newbrk, oldbrk;
210 struct mm_struct *mm = current->mm;
212 down_write(&mm->mmap_sem);
214 if (brk < mm->end_code)
215 goto out;
216 newbrk = PAGE_ALIGN(brk);
217 oldbrk = PAGE_ALIGN(mm->brk);
218 if (oldbrk == newbrk)
219 goto set_brk;
221 /* Always allow shrinking brk. */
222 if (brk <= mm->brk) {
223 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
224 goto set_brk;
225 goto out;
228 /* Check against rlimit.. */
229 rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
230 if (rlim < RLIM_INFINITY && brk - mm->start_data > rlim)
231 goto out;
233 /* Check against existing mmap mappings. */
234 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
235 goto out;
237 /* Ok, looks good - let it rip. */
238 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
239 goto out;
240 set_brk:
241 mm->brk = brk;
242 out:
243 retval = mm->brk;
244 up_write(&mm->mmap_sem);
245 return retval;
248 #ifdef DEBUG_MM_RB
249 static int browse_rb(struct rb_root *root)
251 int i = 0, j;
252 struct rb_node *nd, *pn = NULL;
253 unsigned long prev = 0, pend = 0;
255 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
256 struct vm_area_struct *vma;
257 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
258 if (vma->vm_start < prev)
259 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
260 if (vma->vm_start < pend)
261 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
262 if (vma->vm_start > vma->vm_end)
263 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
264 i++;
265 pn = nd;
267 j = 0;
268 for (nd = pn; nd; nd = rb_prev(nd)) {
269 j++;
271 if (i != j)
272 printk("backwards %d, forwards %d\n", j, i), i = 0;
273 return i;
276 void validate_mm(struct mm_struct *mm)
278 int bug = 0;
279 int i = 0;
280 struct vm_area_struct *tmp = mm->mmap;
281 while (tmp) {
282 tmp = tmp->vm_next;
283 i++;
285 if (i != mm->map_count)
286 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
287 i = browse_rb(&mm->mm_rb);
288 if (i != mm->map_count)
289 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
290 if (bug)
291 BUG();
293 #else
294 #define validate_mm(mm) do { } while (0)
295 #endif
297 static struct vm_area_struct *
298 find_vma_prepare(struct mm_struct *mm, unsigned long addr,
299 struct vm_area_struct **pprev, struct rb_node ***rb_link,
300 struct rb_node ** rb_parent)
302 struct vm_area_struct * vma;
303 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
305 __rb_link = &mm->mm_rb.rb_node;
306 rb_prev = __rb_parent = NULL;
307 vma = NULL;
309 while (*__rb_link) {
310 struct vm_area_struct *vma_tmp;
312 __rb_parent = *__rb_link;
313 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
315 if (vma_tmp->vm_end > addr) {
316 vma = vma_tmp;
317 if (vma_tmp->vm_start <= addr)
318 return vma;
319 __rb_link = &__rb_parent->rb_left;
320 } else {
321 rb_prev = __rb_parent;
322 __rb_link = &__rb_parent->rb_right;
326 *pprev = NULL;
327 if (rb_prev)
328 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
329 *rb_link = __rb_link;
330 *rb_parent = __rb_parent;
331 return vma;
334 static inline void
335 __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
336 struct vm_area_struct *prev, struct rb_node *rb_parent)
338 if (prev) {
339 vma->vm_next = prev->vm_next;
340 prev->vm_next = vma;
341 } else {
342 mm->mmap = vma;
343 if (rb_parent)
344 vma->vm_next = rb_entry(rb_parent,
345 struct vm_area_struct, vm_rb);
346 else
347 vma->vm_next = NULL;
351 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
352 struct rb_node **rb_link, struct rb_node *rb_parent)
354 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
355 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
358 static inline void __vma_link_file(struct vm_area_struct *vma)
360 struct file * file;
362 file = vma->vm_file;
363 if (file) {
364 struct address_space *mapping = file->f_mapping;
366 if (vma->vm_flags & VM_DENYWRITE)
367 atomic_dec(&file->f_dentry->d_inode->i_writecount);
368 if (vma->vm_flags & VM_SHARED)
369 mapping->i_mmap_writable++;
371 flush_dcache_mmap_lock(mapping);
372 if (unlikely(vma->vm_flags & VM_NONLINEAR))
373 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
374 else
375 vma_prio_tree_insert(vma, &mapping->i_mmap);
376 flush_dcache_mmap_unlock(mapping);
380 static void
381 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
382 struct vm_area_struct *prev, struct rb_node **rb_link,
383 struct rb_node *rb_parent)
385 __vma_link_list(mm, vma, prev, rb_parent);
386 __vma_link_rb(mm, vma, rb_link, rb_parent);
387 __anon_vma_link(vma);
390 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
391 struct vm_area_struct *prev, struct rb_node **rb_link,
392 struct rb_node *rb_parent)
394 struct address_space *mapping = NULL;
396 if (vma->vm_file)
397 mapping = vma->vm_file->f_mapping;
399 if (mapping) {
400 spin_lock(&mapping->i_mmap_lock);
401 vma->vm_truncate_count = mapping->truncate_count;
403 anon_vma_lock(vma);
405 __vma_link(mm, vma, prev, rb_link, rb_parent);
406 __vma_link_file(vma);
408 anon_vma_unlock(vma);
409 if (mapping)
410 spin_unlock(&mapping->i_mmap_lock);
412 mm->map_count++;
413 validate_mm(mm);
417 * Helper for vma_adjust in the split_vma insert case:
418 * insert vm structure into list and rbtree and anon_vma,
419 * but it has already been inserted into prio_tree earlier.
421 static void
422 __insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
424 struct vm_area_struct * __vma, * prev;
425 struct rb_node ** rb_link, * rb_parent;
427 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
428 if (__vma && __vma->vm_start < vma->vm_end)
429 BUG();
430 __vma_link(mm, vma, prev, rb_link, rb_parent);
431 mm->map_count++;
434 static inline void
435 __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
436 struct vm_area_struct *prev)
438 prev->vm_next = vma->vm_next;
439 rb_erase(&vma->vm_rb, &mm->mm_rb);
440 if (mm->mmap_cache == vma)
441 mm->mmap_cache = prev;
445 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
446 * is already present in an i_mmap tree without adjusting the tree.
447 * The following helper function should be used when such adjustments
448 * are necessary. The "insert" vma (if any) is to be inserted
449 * before we drop the necessary locks.
451 void vma_adjust(struct vm_area_struct *vma, unsigned long start,
452 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
454 struct mm_struct *mm = vma->vm_mm;
455 struct vm_area_struct *next = vma->vm_next;
456 struct vm_area_struct *importer = NULL;
457 struct address_space *mapping = NULL;
458 struct prio_tree_root *root = NULL;
459 struct file *file = vma->vm_file;
460 struct anon_vma *anon_vma = NULL;
461 long adjust_next = 0;
462 int remove_next = 0;
464 if (next && !insert) {
465 if (end >= next->vm_end) {
467 * vma expands, overlapping all the next, and
468 * perhaps the one after too (mprotect case 6).
470 again: remove_next = 1 + (end > next->vm_end);
471 end = next->vm_end;
472 anon_vma = next->anon_vma;
473 importer = vma;
474 } else if (end > next->vm_start) {
476 * vma expands, overlapping part of the next:
477 * mprotect case 5 shifting the boundary up.
479 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
480 anon_vma = next->anon_vma;
481 importer = vma;
482 } else if (end < vma->vm_end) {
484 * vma shrinks, and !insert tells it's not
485 * split_vma inserting another: so it must be
486 * mprotect case 4 shifting the boundary down.
488 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
489 anon_vma = next->anon_vma;
490 importer = next;
494 if (file) {
495 mapping = file->f_mapping;
496 if (!(vma->vm_flags & VM_NONLINEAR))
497 root = &mapping->i_mmap;
498 spin_lock(&mapping->i_mmap_lock);
499 if (importer &&
500 vma->vm_truncate_count != next->vm_truncate_count) {
502 * unmap_mapping_range might be in progress:
503 * ensure that the expanding vma is rescanned.
505 importer->vm_truncate_count = 0;
507 if (insert) {
508 insert->vm_truncate_count = vma->vm_truncate_count;
510 * Put into prio_tree now, so instantiated pages
511 * are visible to arm/parisc __flush_dcache_page
512 * throughout; but we cannot insert into address
513 * space until vma start or end is updated.
515 __vma_link_file(insert);
520 * When changing only vma->vm_end, we don't really need
521 * anon_vma lock: but is that case worth optimizing out?
523 if (vma->anon_vma)
524 anon_vma = vma->anon_vma;
525 if (anon_vma) {
526 spin_lock(&anon_vma->lock);
528 * Easily overlooked: when mprotect shifts the boundary,
529 * make sure the expanding vma has anon_vma set if the
530 * shrinking vma had, to cover any anon pages imported.
532 if (importer && !importer->anon_vma) {
533 importer->anon_vma = anon_vma;
534 __anon_vma_link(importer);
538 if (root) {
539 flush_dcache_mmap_lock(mapping);
540 vma_prio_tree_remove(vma, root);
541 if (adjust_next)
542 vma_prio_tree_remove(next, root);
545 vma->vm_start = start;
546 vma->vm_end = end;
547 vma->vm_pgoff = pgoff;
548 if (adjust_next) {
549 next->vm_start += adjust_next << PAGE_SHIFT;
550 next->vm_pgoff += adjust_next;
553 if (root) {
554 if (adjust_next)
555 vma_prio_tree_insert(next, root);
556 vma_prio_tree_insert(vma, root);
557 flush_dcache_mmap_unlock(mapping);
560 if (remove_next) {
562 * vma_merge has merged next into vma, and needs
563 * us to remove next before dropping the locks.
565 __vma_unlink(mm, next, vma);
566 if (file)
567 __remove_shared_vm_struct(next, file, mapping);
568 if (next->anon_vma)
569 __anon_vma_merge(vma, next);
570 } else if (insert) {
572 * split_vma has split insert from vma, and needs
573 * us to insert it before dropping the locks
574 * (it may either follow vma or precede it).
576 __insert_vm_struct(mm, insert);
579 if (anon_vma)
580 spin_unlock(&anon_vma->lock);
581 if (mapping)
582 spin_unlock(&mapping->i_mmap_lock);
584 if (remove_next) {
585 if (file)
586 fput(file);
587 mm->map_count--;
588 mpol_free(vma_policy(next));
589 kmem_cache_free(vm_area_cachep, next);
591 * In mprotect's case 6 (see comments on vma_merge),
592 * we must remove another next too. It would clutter
593 * up the code too much to do both in one go.
595 if (remove_next == 2) {
596 next = vma->vm_next;
597 goto again;
601 validate_mm(mm);
605 * If the vma has a ->close operation then the driver probably needs to release
606 * per-vma resources, so we don't attempt to merge those.
608 #define VM_SPECIAL (VM_IO | VM_DONTCOPY | VM_DONTEXPAND | VM_RESERVED)
610 static inline int is_mergeable_vma(struct vm_area_struct *vma,
611 struct file *file, unsigned long vm_flags)
613 if (vma->vm_flags != vm_flags)
614 return 0;
615 if (vma->vm_file != file)
616 return 0;
617 if (vma->vm_ops && vma->vm_ops->close)
618 return 0;
619 return 1;
622 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
623 struct anon_vma *anon_vma2)
625 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
629 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
630 * in front of (at a lower virtual address and file offset than) the vma.
632 * We cannot merge two vmas if they have differently assigned (non-NULL)
633 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
635 * We don't check here for the merged mmap wrapping around the end of pagecache
636 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
637 * wrap, nor mmaps which cover the final page at index -1UL.
639 static int
640 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
641 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
643 if (is_mergeable_vma(vma, file, vm_flags) &&
644 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
645 if (vma->vm_pgoff == vm_pgoff)
646 return 1;
648 return 0;
652 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
653 * beyond (at a higher virtual address and file offset than) the vma.
655 * We cannot merge two vmas if they have differently assigned (non-NULL)
656 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
658 static int
659 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
660 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
662 if (is_mergeable_vma(vma, file, vm_flags) &&
663 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
664 pgoff_t vm_pglen;
665 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
666 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
667 return 1;
669 return 0;
673 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
674 * whether that can be merged with its predecessor or its successor.
675 * Or both (it neatly fills a hole).
677 * In most cases - when called for mmap, brk or mremap - [addr,end) is
678 * certain not to be mapped by the time vma_merge is called; but when
679 * called for mprotect, it is certain to be already mapped (either at
680 * an offset within prev, or at the start of next), and the flags of
681 * this area are about to be changed to vm_flags - and the no-change
682 * case has already been eliminated.
684 * The following mprotect cases have to be considered, where AAAA is
685 * the area passed down from mprotect_fixup, never extending beyond one
686 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
688 * AAAA AAAA AAAA AAAA
689 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
690 * cannot merge might become might become might become
691 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
692 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
693 * mremap move: PPPPNNNNNNNN 8
694 * AAAA
695 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
696 * might become case 1 below case 2 below case 3 below
698 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
699 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
701 struct vm_area_struct *vma_merge(struct mm_struct *mm,
702 struct vm_area_struct *prev, unsigned long addr,
703 unsigned long end, unsigned long vm_flags,
704 struct anon_vma *anon_vma, struct file *file,
705 pgoff_t pgoff, struct mempolicy *policy)
707 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
708 struct vm_area_struct *area, *next;
711 * We later require that vma->vm_flags == vm_flags,
712 * so this tests vma->vm_flags & VM_SPECIAL, too.
714 if (vm_flags & VM_SPECIAL)
715 return NULL;
717 if (prev)
718 next = prev->vm_next;
719 else
720 next = mm->mmap;
721 area = next;
722 if (next && next->vm_end == end) /* cases 6, 7, 8 */
723 next = next->vm_next;
726 * Can it merge with the predecessor?
728 if (prev && prev->vm_end == addr &&
729 mpol_equal(vma_policy(prev), policy) &&
730 can_vma_merge_after(prev, vm_flags,
731 anon_vma, file, pgoff)) {
733 * OK, it can. Can we now merge in the successor as well?
735 if (next && end == next->vm_start &&
736 mpol_equal(policy, vma_policy(next)) &&
737 can_vma_merge_before(next, vm_flags,
738 anon_vma, file, pgoff+pglen) &&
739 is_mergeable_anon_vma(prev->anon_vma,
740 next->anon_vma)) {
741 /* cases 1, 6 */
742 vma_adjust(prev, prev->vm_start,
743 next->vm_end, prev->vm_pgoff, NULL);
744 } else /* cases 2, 5, 7 */
745 vma_adjust(prev, prev->vm_start,
746 end, prev->vm_pgoff, NULL);
747 return prev;
751 * Can this new request be merged in front of next?
753 if (next && end == next->vm_start &&
754 mpol_equal(policy, vma_policy(next)) &&
755 can_vma_merge_before(next, vm_flags,
756 anon_vma, file, pgoff+pglen)) {
757 if (prev && addr < prev->vm_end) /* case 4 */
758 vma_adjust(prev, prev->vm_start,
759 addr, prev->vm_pgoff, NULL);
760 else /* cases 3, 8 */
761 vma_adjust(area, addr, next->vm_end,
762 next->vm_pgoff - pglen, NULL);
763 return area;
766 return NULL;
770 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
771 * neighbouring vmas for a suitable anon_vma, before it goes off
772 * to allocate a new anon_vma. It checks because a repetitive
773 * sequence of mprotects and faults may otherwise lead to distinct
774 * anon_vmas being allocated, preventing vma merge in subsequent
775 * mprotect.
777 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
779 struct vm_area_struct *near;
780 unsigned long vm_flags;
782 near = vma->vm_next;
783 if (!near)
784 goto try_prev;
787 * Since only mprotect tries to remerge vmas, match flags
788 * which might be mprotected into each other later on.
789 * Neither mlock nor madvise tries to remerge at present,
790 * so leave their flags as obstructing a merge.
792 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
793 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
795 if (near->anon_vma && vma->vm_end == near->vm_start &&
796 mpol_equal(vma_policy(vma), vma_policy(near)) &&
797 can_vma_merge_before(near, vm_flags,
798 NULL, vma->vm_file, vma->vm_pgoff +
799 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
800 return near->anon_vma;
801 try_prev:
803 * It is potentially slow to have to call find_vma_prev here.
804 * But it's only on the first write fault on the vma, not
805 * every time, and we could devise a way to avoid it later
806 * (e.g. stash info in next's anon_vma_node when assigning
807 * an anon_vma, or when trying vma_merge). Another time.
809 if (find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma)
810 BUG();
811 if (!near)
812 goto none;
814 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
815 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
817 if (near->anon_vma && near->vm_end == vma->vm_start &&
818 mpol_equal(vma_policy(near), vma_policy(vma)) &&
819 can_vma_merge_after(near, vm_flags,
820 NULL, vma->vm_file, vma->vm_pgoff))
821 return near->anon_vma;
822 none:
824 * There's no absolute need to look only at touching neighbours:
825 * we could search further afield for "compatible" anon_vmas.
826 * But it would probably just be a waste of time searching,
827 * or lead to too many vmas hanging off the same anon_vma.
828 * We're trying to allow mprotect remerging later on,
829 * not trying to minimize memory used for anon_vmas.
831 return NULL;
834 #ifdef CONFIG_PROC_FS
835 void __vm_stat_account(struct mm_struct *mm, unsigned long flags,
836 struct file *file, long pages)
838 const unsigned long stack_flags
839 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
841 #ifdef CONFIG_HUGETLB
842 if (flags & VM_HUGETLB) {
843 if (!(flags & VM_DONTCOPY))
844 mm->shared_vm += pages;
845 return;
847 #endif /* CONFIG_HUGETLB */
849 if (file) {
850 mm->shared_vm += pages;
851 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
852 mm->exec_vm += pages;
853 } else if (flags & stack_flags)
854 mm->stack_vm += pages;
855 if (flags & (VM_RESERVED|VM_IO))
856 mm->reserved_vm += pages;
858 #endif /* CONFIG_PROC_FS */
861 * The caller must hold down_write(current->mm->mmap_sem).
864 unsigned long do_mmap_pgoff(struct file * file, unsigned long addr,
865 unsigned long len, unsigned long prot,
866 unsigned long flags, unsigned long pgoff)
868 struct mm_struct * mm = current->mm;
869 struct vm_area_struct * vma, * prev;
870 struct inode *inode;
871 unsigned int vm_flags;
872 int correct_wcount = 0;
873 int error;
874 struct rb_node ** rb_link, * rb_parent;
875 int accountable = 1;
876 unsigned long charged = 0, reqprot = prot;
878 if (file) {
879 if (is_file_hugepages(file))
880 accountable = 0;
882 if (!file->f_op || !file->f_op->mmap)
883 return -ENODEV;
885 if ((prot & PROT_EXEC) &&
886 (file->f_vfsmnt->mnt_flags & MNT_NOEXEC))
887 return -EPERM;
890 * Does the application expect PROT_READ to imply PROT_EXEC?
892 * (the exception is when the underlying filesystem is noexec
893 * mounted, in which case we dont add PROT_EXEC.)
895 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
896 if (!(file && (file->f_vfsmnt->mnt_flags & MNT_NOEXEC)))
897 prot |= PROT_EXEC;
899 if (!len)
900 return -EINVAL;
902 /* Careful about overflows.. */
903 len = PAGE_ALIGN(len);
904 if (!len || len > TASK_SIZE)
905 return -ENOMEM;
907 /* offset overflow? */
908 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
909 return -EOVERFLOW;
911 /* Too many mappings? */
912 if (mm->map_count > sysctl_max_map_count)
913 return -ENOMEM;
915 /* Obtain the address to map to. we verify (or select) it and ensure
916 * that it represents a valid section of the address space.
918 addr = get_unmapped_area(file, addr, len, pgoff, flags);
919 if (addr & ~PAGE_MASK)
920 return addr;
922 /* Do simple checking here so the lower-level routines won't have
923 * to. we assume access permissions have been handled by the open
924 * of the memory object, so we don't do any here.
926 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
927 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
929 if (flags & MAP_LOCKED) {
930 if (!can_do_mlock())
931 return -EPERM;
932 vm_flags |= VM_LOCKED;
934 /* mlock MCL_FUTURE? */
935 if (vm_flags & VM_LOCKED) {
936 unsigned long locked, lock_limit;
937 locked = len >> PAGE_SHIFT;
938 locked += mm->locked_vm;
939 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
940 lock_limit >>= PAGE_SHIFT;
941 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
942 return -EAGAIN;
945 inode = file ? file->f_dentry->d_inode : NULL;
947 if (file) {
948 switch (flags & MAP_TYPE) {
949 case MAP_SHARED:
950 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
951 return -EACCES;
954 * Make sure we don't allow writing to an append-only
955 * file..
957 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
958 return -EACCES;
961 * Make sure there are no mandatory locks on the file.
963 if (locks_verify_locked(inode))
964 return -EAGAIN;
966 vm_flags |= VM_SHARED | VM_MAYSHARE;
967 if (!(file->f_mode & FMODE_WRITE))
968 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
970 /* fall through */
971 case MAP_PRIVATE:
972 if (!(file->f_mode & FMODE_READ))
973 return -EACCES;
974 break;
976 default:
977 return -EINVAL;
979 } else {
980 switch (flags & MAP_TYPE) {
981 case MAP_SHARED:
982 vm_flags |= VM_SHARED | VM_MAYSHARE;
983 break;
984 case MAP_PRIVATE:
986 * Set pgoff according to addr for anon_vma.
988 pgoff = addr >> PAGE_SHIFT;
989 break;
990 default:
991 return -EINVAL;
995 error = security_file_mmap(file, reqprot, prot, flags);
996 if (error)
997 return error;
999 /* Clear old maps */
1000 error = -ENOMEM;
1001 munmap_back:
1002 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1003 if (vma && vma->vm_start < addr + len) {
1004 if (do_munmap(mm, addr, len))
1005 return -ENOMEM;
1006 goto munmap_back;
1009 /* Check against address space limit. */
1010 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1011 return -ENOMEM;
1013 if (accountable && (!(flags & MAP_NORESERVE) ||
1014 sysctl_overcommit_memory == OVERCOMMIT_NEVER)) {
1015 if (vm_flags & VM_SHARED) {
1016 /* Check memory availability in shmem_file_setup? */
1017 vm_flags |= VM_ACCOUNT;
1018 } else if (vm_flags & VM_WRITE) {
1020 * Private writable mapping: check memory availability
1022 charged = len >> PAGE_SHIFT;
1023 if (security_vm_enough_memory(charged))
1024 return -ENOMEM;
1025 vm_flags |= VM_ACCOUNT;
1030 * Can we just expand an old private anonymous mapping?
1031 * The VM_SHARED test is necessary because shmem_zero_setup
1032 * will create the file object for a shared anonymous map below.
1034 if (!file && !(vm_flags & VM_SHARED) &&
1035 vma_merge(mm, prev, addr, addr + len, vm_flags,
1036 NULL, NULL, pgoff, NULL))
1037 goto out;
1040 * Determine the object being mapped and call the appropriate
1041 * specific mapper. the address has already been validated, but
1042 * not unmapped, but the maps are removed from the list.
1044 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1045 if (!vma) {
1046 error = -ENOMEM;
1047 goto unacct_error;
1049 memset(vma, 0, sizeof(*vma));
1051 vma->vm_mm = mm;
1052 vma->vm_start = addr;
1053 vma->vm_end = addr + len;
1054 vma->vm_flags = vm_flags;
1055 vma->vm_page_prot = protection_map[vm_flags & 0x0f];
1056 vma->vm_pgoff = pgoff;
1058 if (file) {
1059 error = -EINVAL;
1060 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1061 goto free_vma;
1062 if (vm_flags & VM_DENYWRITE) {
1063 error = deny_write_access(file);
1064 if (error)
1065 goto free_vma;
1066 correct_wcount = 1;
1068 vma->vm_file = file;
1069 get_file(file);
1070 error = file->f_op->mmap(file, vma);
1071 if (error)
1072 goto unmap_and_free_vma;
1073 } else if (vm_flags & VM_SHARED) {
1074 error = shmem_zero_setup(vma);
1075 if (error)
1076 goto free_vma;
1079 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
1080 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
1081 * that memory reservation must be checked; but that reservation
1082 * belongs to shared memory object, not to vma: so now clear it.
1084 if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT))
1085 vma->vm_flags &= ~VM_ACCOUNT;
1087 /* Can addr have changed??
1089 * Answer: Yes, several device drivers can do it in their
1090 * f_op->mmap method. -DaveM
1092 addr = vma->vm_start;
1093 pgoff = vma->vm_pgoff;
1094 vm_flags = vma->vm_flags;
1096 if (!file || !vma_merge(mm, prev, addr, vma->vm_end,
1097 vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
1098 file = vma->vm_file;
1099 vma_link(mm, vma, prev, rb_link, rb_parent);
1100 if (correct_wcount)
1101 atomic_inc(&inode->i_writecount);
1102 } else {
1103 if (file) {
1104 if (correct_wcount)
1105 atomic_inc(&inode->i_writecount);
1106 fput(file);
1108 mpol_free(vma_policy(vma));
1109 kmem_cache_free(vm_area_cachep, vma);
1111 out:
1112 mm->total_vm += len >> PAGE_SHIFT;
1113 __vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1114 if (vm_flags & VM_LOCKED) {
1115 mm->locked_vm += len >> PAGE_SHIFT;
1116 make_pages_present(addr, addr + len);
1118 if (flags & MAP_POPULATE) {
1119 up_write(&mm->mmap_sem);
1120 sys_remap_file_pages(addr, len, 0,
1121 pgoff, flags & MAP_NONBLOCK);
1122 down_write(&mm->mmap_sem);
1124 return addr;
1126 unmap_and_free_vma:
1127 if (correct_wcount)
1128 atomic_inc(&inode->i_writecount);
1129 vma->vm_file = NULL;
1130 fput(file);
1132 /* Undo any partial mapping done by a device driver. */
1133 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1134 charged = 0;
1135 free_vma:
1136 kmem_cache_free(vm_area_cachep, vma);
1137 unacct_error:
1138 if (charged)
1139 vm_unacct_memory(charged);
1140 return error;
1143 EXPORT_SYMBOL(do_mmap_pgoff);
1145 /* Get an address range which is currently unmapped.
1146 * For shmat() with addr=0.
1148 * Ugly calling convention alert:
1149 * Return value with the low bits set means error value,
1150 * ie
1151 * if (ret & ~PAGE_MASK)
1152 * error = ret;
1154 * This function "knows" that -ENOMEM has the bits set.
1156 #ifndef HAVE_ARCH_UNMAPPED_AREA
1157 unsigned long
1158 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1159 unsigned long len, unsigned long pgoff, unsigned long flags)
1161 struct mm_struct *mm = current->mm;
1162 struct vm_area_struct *vma;
1163 unsigned long start_addr;
1165 if (len > TASK_SIZE)
1166 return -ENOMEM;
1168 if (addr) {
1169 addr = PAGE_ALIGN(addr);
1170 vma = find_vma(mm, addr);
1171 if (TASK_SIZE - len >= addr &&
1172 (!vma || addr + len <= vma->vm_start))
1173 return addr;
1175 if (len > mm->cached_hole_size) {
1176 start_addr = addr = mm->free_area_cache;
1177 } else {
1178 start_addr = addr = TASK_UNMAPPED_BASE;
1179 mm->cached_hole_size = 0;
1182 full_search:
1183 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1184 /* At this point: (!vma || addr < vma->vm_end). */
1185 if (TASK_SIZE - len < addr) {
1187 * Start a new search - just in case we missed
1188 * some holes.
1190 if (start_addr != TASK_UNMAPPED_BASE) {
1191 addr = TASK_UNMAPPED_BASE;
1192 start_addr = addr;
1193 mm->cached_hole_size = 0;
1194 goto full_search;
1196 return -ENOMEM;
1198 if (!vma || addr + len <= vma->vm_start) {
1200 * Remember the place where we stopped the search:
1202 mm->free_area_cache = addr + len;
1203 return addr;
1205 if (addr + mm->cached_hole_size < vma->vm_start)
1206 mm->cached_hole_size = vma->vm_start - addr;
1207 addr = vma->vm_end;
1210 #endif
1212 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1215 * Is this a new hole at the lowest possible address?
1217 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1218 mm->free_area_cache = addr;
1219 mm->cached_hole_size = ~0UL;
1224 * This mmap-allocator allocates new areas top-down from below the
1225 * stack's low limit (the base):
1227 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1228 unsigned long
1229 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1230 const unsigned long len, const unsigned long pgoff,
1231 const unsigned long flags)
1233 struct vm_area_struct *vma;
1234 struct mm_struct *mm = current->mm;
1235 unsigned long addr = addr0;
1237 /* requested length too big for entire address space */
1238 if (len > TASK_SIZE)
1239 return -ENOMEM;
1241 /* requesting a specific address */
1242 if (addr) {
1243 addr = PAGE_ALIGN(addr);
1244 vma = find_vma(mm, addr);
1245 if (TASK_SIZE - len >= addr &&
1246 (!vma || addr + len <= vma->vm_start))
1247 return addr;
1250 /* check if free_area_cache is useful for us */
1251 if (len <= mm->cached_hole_size) {
1252 mm->cached_hole_size = 0;
1253 mm->free_area_cache = mm->mmap_base;
1256 /* either no address requested or can't fit in requested address hole */
1257 addr = mm->free_area_cache;
1259 /* make sure it can fit in the remaining address space */
1260 if (addr > len) {
1261 vma = find_vma(mm, addr-len);
1262 if (!vma || addr <= vma->vm_start)
1263 /* remember the address as a hint for next time */
1264 return (mm->free_area_cache = addr-len);
1267 if (mm->mmap_base < len)
1268 goto bottomup;
1270 addr = mm->mmap_base-len;
1272 do {
1274 * Lookup failure means no vma is above this address,
1275 * else if new region fits below vma->vm_start,
1276 * return with success:
1278 vma = find_vma(mm, addr);
1279 if (!vma || addr+len <= vma->vm_start)
1280 /* remember the address as a hint for next time */
1281 return (mm->free_area_cache = addr);
1283 /* remember the largest hole we saw so far */
1284 if (addr + mm->cached_hole_size < vma->vm_start)
1285 mm->cached_hole_size = vma->vm_start - addr;
1287 /* try just below the current vma->vm_start */
1288 addr = vma->vm_start-len;
1289 } while (len < vma->vm_start);
1291 bottomup:
1293 * A failed mmap() very likely causes application failure,
1294 * so fall back to the bottom-up function here. This scenario
1295 * can happen with large stack limits and large mmap()
1296 * allocations.
1298 mm->cached_hole_size = ~0UL;
1299 mm->free_area_cache = TASK_UNMAPPED_BASE;
1300 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1302 * Restore the topdown base:
1304 mm->free_area_cache = mm->mmap_base;
1305 mm->cached_hole_size = ~0UL;
1307 return addr;
1309 #endif
1311 void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
1314 * Is this a new hole at the highest possible address?
1316 if (addr > mm->free_area_cache)
1317 mm->free_area_cache = addr;
1319 /* dont allow allocations above current base */
1320 if (mm->free_area_cache > mm->mmap_base)
1321 mm->free_area_cache = mm->mmap_base;
1324 unsigned long
1325 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1326 unsigned long pgoff, unsigned long flags)
1328 unsigned long ret;
1330 if (!(flags & MAP_FIXED)) {
1331 unsigned long (*get_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
1333 get_area = current->mm->get_unmapped_area;
1334 if (file && file->f_op && file->f_op->get_unmapped_area)
1335 get_area = file->f_op->get_unmapped_area;
1336 addr = get_area(file, addr, len, pgoff, flags);
1337 if (IS_ERR_VALUE(addr))
1338 return addr;
1341 if (addr > TASK_SIZE - len)
1342 return -ENOMEM;
1343 if (addr & ~PAGE_MASK)
1344 return -EINVAL;
1345 if (file && is_file_hugepages(file)) {
1347 * Check if the given range is hugepage aligned, and
1348 * can be made suitable for hugepages.
1350 ret = prepare_hugepage_range(addr, len);
1351 } else {
1353 * Ensure that a normal request is not falling in a
1354 * reserved hugepage range. For some archs like IA-64,
1355 * there is a separate region for hugepages.
1357 ret = is_hugepage_only_range(current->mm, addr, len);
1359 if (ret)
1360 return -EINVAL;
1361 return addr;
1364 EXPORT_SYMBOL(get_unmapped_area);
1366 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1367 struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
1369 struct vm_area_struct *vma = NULL;
1371 if (mm) {
1372 /* Check the cache first. */
1373 /* (Cache hit rate is typically around 35%.) */
1374 vma = mm->mmap_cache;
1375 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1376 struct rb_node * rb_node;
1378 rb_node = mm->mm_rb.rb_node;
1379 vma = NULL;
1381 while (rb_node) {
1382 struct vm_area_struct * vma_tmp;
1384 vma_tmp = rb_entry(rb_node,
1385 struct vm_area_struct, vm_rb);
1387 if (vma_tmp->vm_end > addr) {
1388 vma = vma_tmp;
1389 if (vma_tmp->vm_start <= addr)
1390 break;
1391 rb_node = rb_node->rb_left;
1392 } else
1393 rb_node = rb_node->rb_right;
1395 if (vma)
1396 mm->mmap_cache = vma;
1399 return vma;
1402 EXPORT_SYMBOL(find_vma);
1404 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1405 struct vm_area_struct *
1406 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1407 struct vm_area_struct **pprev)
1409 struct vm_area_struct *vma = NULL, *prev = NULL;
1410 struct rb_node * rb_node;
1411 if (!mm)
1412 goto out;
1414 /* Guard against addr being lower than the first VMA */
1415 vma = mm->mmap;
1417 /* Go through the RB tree quickly. */
1418 rb_node = mm->mm_rb.rb_node;
1420 while (rb_node) {
1421 struct vm_area_struct *vma_tmp;
1422 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1424 if (addr < vma_tmp->vm_end) {
1425 rb_node = rb_node->rb_left;
1426 } else {
1427 prev = vma_tmp;
1428 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1429 break;
1430 rb_node = rb_node->rb_right;
1434 out:
1435 *pprev = prev;
1436 return prev ? prev->vm_next : vma;
1440 * Verify that the stack growth is acceptable and
1441 * update accounting. This is shared with both the
1442 * grow-up and grow-down cases.
1444 static int acct_stack_growth(struct vm_area_struct * vma, unsigned long size, unsigned long grow)
1446 struct mm_struct *mm = vma->vm_mm;
1447 struct rlimit *rlim = current->signal->rlim;
1449 /* address space limit tests */
1450 if (!may_expand_vm(mm, grow))
1451 return -ENOMEM;
1453 /* Stack limit test */
1454 if (size > rlim[RLIMIT_STACK].rlim_cur)
1455 return -ENOMEM;
1457 /* mlock limit tests */
1458 if (vma->vm_flags & VM_LOCKED) {
1459 unsigned long locked;
1460 unsigned long limit;
1461 locked = mm->locked_vm + grow;
1462 limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
1463 if (locked > limit && !capable(CAP_IPC_LOCK))
1464 return -ENOMEM;
1468 * Overcommit.. This must be the final test, as it will
1469 * update security statistics.
1471 if (security_vm_enough_memory(grow))
1472 return -ENOMEM;
1474 /* Ok, everything looks good - let it rip */
1475 mm->total_vm += grow;
1476 if (vma->vm_flags & VM_LOCKED)
1477 mm->locked_vm += grow;
1478 __vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1479 return 0;
1482 #ifdef CONFIG_STACK_GROWSUP
1484 * vma is the first one with address > vma->vm_end. Have to extend vma.
1486 int expand_stack(struct vm_area_struct * vma, unsigned long address)
1488 int error;
1490 if (!(vma->vm_flags & VM_GROWSUP))
1491 return -EFAULT;
1494 * We must make sure the anon_vma is allocated
1495 * so that the anon_vma locking is not a noop.
1497 if (unlikely(anon_vma_prepare(vma)))
1498 return -ENOMEM;
1499 anon_vma_lock(vma);
1502 * vma->vm_start/vm_end cannot change under us because the caller
1503 * is required to hold the mmap_sem in read mode. We need the
1504 * anon_vma lock to serialize against concurrent expand_stacks.
1506 address += 4 + PAGE_SIZE - 1;
1507 address &= PAGE_MASK;
1508 error = 0;
1510 /* Somebody else might have raced and expanded it already */
1511 if (address > vma->vm_end) {
1512 unsigned long size, grow;
1514 size = address - vma->vm_start;
1515 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1517 error = acct_stack_growth(vma, size, grow);
1518 if (!error)
1519 vma->vm_end = address;
1521 anon_vma_unlock(vma);
1522 return error;
1525 struct vm_area_struct *
1526 find_extend_vma(struct mm_struct *mm, unsigned long addr)
1528 struct vm_area_struct *vma, *prev;
1530 addr &= PAGE_MASK;
1531 vma = find_vma_prev(mm, addr, &prev);
1532 if (vma && (vma->vm_start <= addr))
1533 return vma;
1534 if (!prev || expand_stack(prev, addr))
1535 return NULL;
1536 if (prev->vm_flags & VM_LOCKED) {
1537 make_pages_present(addr, prev->vm_end);
1539 return prev;
1541 #else
1543 * vma is the first one with address < vma->vm_start. Have to extend vma.
1545 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1547 int error;
1550 * We must make sure the anon_vma is allocated
1551 * so that the anon_vma locking is not a noop.
1553 if (unlikely(anon_vma_prepare(vma)))
1554 return -ENOMEM;
1555 anon_vma_lock(vma);
1558 * vma->vm_start/vm_end cannot change under us because the caller
1559 * is required to hold the mmap_sem in read mode. We need the
1560 * anon_vma lock to serialize against concurrent expand_stacks.
1562 address &= PAGE_MASK;
1563 error = 0;
1565 /* Somebody else might have raced and expanded it already */
1566 if (address < vma->vm_start) {
1567 unsigned long size, grow;
1569 size = vma->vm_end - address;
1570 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1572 error = acct_stack_growth(vma, size, grow);
1573 if (!error) {
1574 vma->vm_start = address;
1575 vma->vm_pgoff -= grow;
1578 anon_vma_unlock(vma);
1579 return error;
1582 struct vm_area_struct *
1583 find_extend_vma(struct mm_struct * mm, unsigned long addr)
1585 struct vm_area_struct * vma;
1586 unsigned long start;
1588 addr &= PAGE_MASK;
1589 vma = find_vma(mm,addr);
1590 if (!vma)
1591 return NULL;
1592 if (vma->vm_start <= addr)
1593 return vma;
1594 if (!(vma->vm_flags & VM_GROWSDOWN))
1595 return NULL;
1596 start = vma->vm_start;
1597 if (expand_stack(vma, addr))
1598 return NULL;
1599 if (vma->vm_flags & VM_LOCKED) {
1600 make_pages_present(addr, start);
1602 return vma;
1604 #endif
1606 /* Normal function to fix up a mapping
1607 * This function is the default for when an area has no specific
1608 * function. This may be used as part of a more specific routine.
1610 * By the time this function is called, the area struct has been
1611 * removed from the process mapping list.
1613 static void unmap_vma(struct mm_struct *mm, struct vm_area_struct *area)
1615 size_t len = area->vm_end - area->vm_start;
1617 area->vm_mm->total_vm -= len >> PAGE_SHIFT;
1618 if (area->vm_flags & VM_LOCKED)
1619 area->vm_mm->locked_vm -= len >> PAGE_SHIFT;
1620 vm_stat_unaccount(area);
1621 remove_vm_struct(area);
1625 * Update the VMA and inode share lists.
1627 * Ok - we have the memory areas we should free on the 'free' list,
1628 * so release them, and do the vma updates.
1630 static void unmap_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
1632 do {
1633 struct vm_area_struct *next = vma->vm_next;
1634 unmap_vma(mm, vma);
1635 vma = next;
1636 } while (vma);
1637 validate_mm(mm);
1641 * Get rid of page table information in the indicated region.
1643 * Called with the mm semaphore held.
1645 static void unmap_region(struct mm_struct *mm,
1646 struct vm_area_struct *vma, struct vm_area_struct *prev,
1647 unsigned long start, unsigned long end)
1649 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
1650 struct mmu_gather *tlb;
1651 unsigned long nr_accounted = 0;
1653 lru_add_drain();
1654 spin_lock(&mm->page_table_lock);
1655 tlb = tlb_gather_mmu(mm, 0);
1656 unmap_vmas(&tlb, mm, vma, start, end, &nr_accounted, NULL);
1657 vm_unacct_memory(nr_accounted);
1658 free_pgtables(&tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
1659 next? next->vm_start: 0);
1660 tlb_finish_mmu(tlb, start, end);
1661 spin_unlock(&mm->page_table_lock);
1665 * Create a list of vma's touched by the unmap, removing them from the mm's
1666 * vma list as we go..
1668 static void
1669 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1670 struct vm_area_struct *prev, unsigned long end)
1672 struct vm_area_struct **insertion_point;
1673 struct vm_area_struct *tail_vma = NULL;
1674 unsigned long addr;
1676 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1677 do {
1678 rb_erase(&vma->vm_rb, &mm->mm_rb);
1679 mm->map_count--;
1680 tail_vma = vma;
1681 vma = vma->vm_next;
1682 } while (vma && vma->vm_start < end);
1683 *insertion_point = vma;
1684 tail_vma->vm_next = NULL;
1685 if (mm->unmap_area == arch_unmap_area)
1686 addr = prev ? prev->vm_end : mm->mmap_base;
1687 else
1688 addr = vma ? vma->vm_start : mm->mmap_base;
1689 mm->unmap_area(mm, addr);
1690 mm->mmap_cache = NULL; /* Kill the cache. */
1694 * Split a vma into two pieces at address 'addr', a new vma is allocated
1695 * either for the first part or the the tail.
1697 int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1698 unsigned long addr, int new_below)
1700 struct mempolicy *pol;
1701 struct vm_area_struct *new;
1703 if (is_vm_hugetlb_page(vma) && (addr & ~HPAGE_MASK))
1704 return -EINVAL;
1706 if (mm->map_count >= sysctl_max_map_count)
1707 return -ENOMEM;
1709 new = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1710 if (!new)
1711 return -ENOMEM;
1713 /* most fields are the same, copy all, and then fixup */
1714 *new = *vma;
1716 if (new_below)
1717 new->vm_end = addr;
1718 else {
1719 new->vm_start = addr;
1720 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1723 pol = mpol_copy(vma_policy(vma));
1724 if (IS_ERR(pol)) {
1725 kmem_cache_free(vm_area_cachep, new);
1726 return PTR_ERR(pol);
1728 vma_set_policy(new, pol);
1730 if (new->vm_file)
1731 get_file(new->vm_file);
1733 if (new->vm_ops && new->vm_ops->open)
1734 new->vm_ops->open(new);
1736 if (new_below)
1737 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1738 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1739 else
1740 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1742 return 0;
1745 /* Munmap is split into 2 main parts -- this part which finds
1746 * what needs doing, and the areas themselves, which do the
1747 * work. This now handles partial unmappings.
1748 * Jeremy Fitzhardinge <jeremy@goop.org>
1750 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1752 unsigned long end;
1753 struct vm_area_struct *vma, *prev, *last;
1755 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1756 return -EINVAL;
1758 if ((len = PAGE_ALIGN(len)) == 0)
1759 return -EINVAL;
1761 /* Find the first overlapping VMA */
1762 vma = find_vma_prev(mm, start, &prev);
1763 if (!vma)
1764 return 0;
1765 /* we have start < vma->vm_end */
1767 /* if it doesn't overlap, we have nothing.. */
1768 end = start + len;
1769 if (vma->vm_start >= end)
1770 return 0;
1773 * If we need to split any vma, do it now to save pain later.
1775 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1776 * unmapped vm_area_struct will remain in use: so lower split_vma
1777 * places tmp vma above, and higher split_vma places tmp vma below.
1779 if (start > vma->vm_start) {
1780 int error = split_vma(mm, vma, start, 0);
1781 if (error)
1782 return error;
1783 prev = vma;
1786 /* Does it split the last one? */
1787 last = find_vma(mm, end);
1788 if (last && end > last->vm_start) {
1789 int error = split_vma(mm, last, end, 1);
1790 if (error)
1791 return error;
1793 vma = prev? prev->vm_next: mm->mmap;
1796 * Remove the vma's, and unmap the actual pages
1798 detach_vmas_to_be_unmapped(mm, vma, prev, end);
1799 unmap_region(mm, vma, prev, start, end);
1801 /* Fix up all other VM information */
1802 unmap_vma_list(mm, vma);
1804 return 0;
1807 EXPORT_SYMBOL(do_munmap);
1809 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1811 int ret;
1812 struct mm_struct *mm = current->mm;
1814 profile_munmap(addr);
1816 down_write(&mm->mmap_sem);
1817 ret = do_munmap(mm, addr, len);
1818 up_write(&mm->mmap_sem);
1819 return ret;
1822 static inline void verify_mm_writelocked(struct mm_struct *mm)
1824 #ifdef CONFIG_DEBUG_KERNEL
1825 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
1826 WARN_ON(1);
1827 up_read(&mm->mmap_sem);
1829 #endif
1833 * this is really a simplified "do_mmap". it only handles
1834 * anonymous maps. eventually we may be able to do some
1835 * brk-specific accounting here.
1837 unsigned long do_brk(unsigned long addr, unsigned long len)
1839 struct mm_struct * mm = current->mm;
1840 struct vm_area_struct * vma, * prev;
1841 unsigned long flags;
1842 struct rb_node ** rb_link, * rb_parent;
1843 pgoff_t pgoff = addr >> PAGE_SHIFT;
1845 len = PAGE_ALIGN(len);
1846 if (!len)
1847 return addr;
1849 if ((addr + len) > TASK_SIZE || (addr + len) < addr)
1850 return -EINVAL;
1853 * mlock MCL_FUTURE?
1855 if (mm->def_flags & VM_LOCKED) {
1856 unsigned long locked, lock_limit;
1857 locked = len >> PAGE_SHIFT;
1858 locked += mm->locked_vm;
1859 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
1860 lock_limit >>= PAGE_SHIFT;
1861 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1862 return -EAGAIN;
1866 * mm->mmap_sem is required to protect against another thread
1867 * changing the mappings in case we sleep.
1869 verify_mm_writelocked(mm);
1872 * Clear old maps. this also does some error checking for us
1874 munmap_back:
1875 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1876 if (vma && vma->vm_start < addr + len) {
1877 if (do_munmap(mm, addr, len))
1878 return -ENOMEM;
1879 goto munmap_back;
1882 /* Check against address space limits *after* clearing old maps... */
1883 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1884 return -ENOMEM;
1886 if (mm->map_count > sysctl_max_map_count)
1887 return -ENOMEM;
1889 if (security_vm_enough_memory(len >> PAGE_SHIFT))
1890 return -ENOMEM;
1892 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
1894 /* Can we just expand an old private anonymous mapping? */
1895 if (vma_merge(mm, prev, addr, addr + len, flags,
1896 NULL, NULL, pgoff, NULL))
1897 goto out;
1900 * create a vma struct for an anonymous mapping
1902 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1903 if (!vma) {
1904 vm_unacct_memory(len >> PAGE_SHIFT);
1905 return -ENOMEM;
1907 memset(vma, 0, sizeof(*vma));
1909 vma->vm_mm = mm;
1910 vma->vm_start = addr;
1911 vma->vm_end = addr + len;
1912 vma->vm_pgoff = pgoff;
1913 vma->vm_flags = flags;
1914 vma->vm_page_prot = protection_map[flags & 0x0f];
1915 vma_link(mm, vma, prev, rb_link, rb_parent);
1916 out:
1917 mm->total_vm += len >> PAGE_SHIFT;
1918 if (flags & VM_LOCKED) {
1919 mm->locked_vm += len >> PAGE_SHIFT;
1920 make_pages_present(addr, addr + len);
1922 return addr;
1925 EXPORT_SYMBOL(do_brk);
1927 /* Release all mmaps. */
1928 void exit_mmap(struct mm_struct *mm)
1930 struct mmu_gather *tlb;
1931 struct vm_area_struct *vma = mm->mmap;
1932 unsigned long nr_accounted = 0;
1933 unsigned long end;
1935 lru_add_drain();
1937 spin_lock(&mm->page_table_lock);
1939 flush_cache_mm(mm);
1940 tlb = tlb_gather_mmu(mm, 1);
1941 /* Use -1 here to ensure all VMAs in the mm are unmapped */
1942 end = unmap_vmas(&tlb, mm, vma, 0, -1, &nr_accounted, NULL);
1943 vm_unacct_memory(nr_accounted);
1944 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
1945 tlb_finish_mmu(tlb, 0, end);
1947 mm->mmap = mm->mmap_cache = NULL;
1948 mm->mm_rb = RB_ROOT;
1949 set_mm_counter(mm, rss, 0);
1950 mm->total_vm = 0;
1951 mm->locked_vm = 0;
1953 spin_unlock(&mm->page_table_lock);
1956 * Walk the list again, actually closing and freeing it
1957 * without holding any MM locks.
1959 while (vma) {
1960 struct vm_area_struct *next = vma->vm_next;
1961 remove_vm_struct(vma);
1962 vma = next;
1965 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
1968 /* Insert vm structure into process list sorted by address
1969 * and into the inode's i_mmap tree. If vm_file is non-NULL
1970 * then i_mmap_lock is taken here.
1972 int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
1974 struct vm_area_struct * __vma, * prev;
1975 struct rb_node ** rb_link, * rb_parent;
1978 * The vm_pgoff of a purely anonymous vma should be irrelevant
1979 * until its first write fault, when page's anon_vma and index
1980 * are set. But now set the vm_pgoff it will almost certainly
1981 * end up with (unless mremap moves it elsewhere before that
1982 * first wfault), so /proc/pid/maps tells a consistent story.
1984 * By setting it to reflect the virtual start address of the
1985 * vma, merges and splits can happen in a seamless way, just
1986 * using the existing file pgoff checks and manipulations.
1987 * Similarly in do_mmap_pgoff and in do_brk.
1989 if (!vma->vm_file) {
1990 BUG_ON(vma->anon_vma);
1991 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
1993 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
1994 if (__vma && __vma->vm_start < vma->vm_end)
1995 return -ENOMEM;
1996 if ((vma->vm_flags & VM_ACCOUNT) &&
1997 security_vm_enough_memory(vma_pages(vma)))
1998 return -ENOMEM;
1999 vma_link(mm, vma, prev, rb_link, rb_parent);
2000 return 0;
2004 * Copy the vma structure to a new location in the same mm,
2005 * prior to moving page table entries, to effect an mremap move.
2007 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2008 unsigned long addr, unsigned long len, pgoff_t pgoff)
2010 struct vm_area_struct *vma = *vmap;
2011 unsigned long vma_start = vma->vm_start;
2012 struct mm_struct *mm = vma->vm_mm;
2013 struct vm_area_struct *new_vma, *prev;
2014 struct rb_node **rb_link, *rb_parent;
2015 struct mempolicy *pol;
2018 * If anonymous vma has not yet been faulted, update new pgoff
2019 * to match new location, to increase its chance of merging.
2021 if (!vma->vm_file && !vma->anon_vma)
2022 pgoff = addr >> PAGE_SHIFT;
2024 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2025 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2026 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2027 if (new_vma) {
2029 * Source vma may have been merged into new_vma
2031 if (vma_start >= new_vma->vm_start &&
2032 vma_start < new_vma->vm_end)
2033 *vmap = new_vma;
2034 } else {
2035 new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
2036 if (new_vma) {
2037 *new_vma = *vma;
2038 pol = mpol_copy(vma_policy(vma));
2039 if (IS_ERR(pol)) {
2040 kmem_cache_free(vm_area_cachep, new_vma);
2041 return NULL;
2043 vma_set_policy(new_vma, pol);
2044 new_vma->vm_start = addr;
2045 new_vma->vm_end = addr + len;
2046 new_vma->vm_pgoff = pgoff;
2047 if (new_vma->vm_file)
2048 get_file(new_vma->vm_file);
2049 if (new_vma->vm_ops && new_vma->vm_ops->open)
2050 new_vma->vm_ops->open(new_vma);
2051 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2054 return new_vma;
2058 * Return true if the calling process may expand its vm space by the passed
2059 * number of pages
2061 int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2063 unsigned long cur = mm->total_vm; /* pages */
2064 unsigned long lim;
2066 lim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
2068 if (cur + npages > lim)
2069 return 0;
2070 return 1;