6 * Address space accounting code <alan@redhat.com>
9 #include <linux/slab.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/capability.h>
17 #include <linux/init.h>
18 #include <linux/file.h>
20 #include <linux/personality.h>
21 #include <linux/security.h>
22 #include <linux/hugetlb.h>
23 #include <linux/profile.h>
24 #include <linux/module.h>
25 #include <linux/mount.h>
26 #include <linux/mempolicy.h>
27 #include <linux/rmap.h>
28 #include <linux/random.h>
30 #include <asm/uaccess.h>
31 #include <asm/cacheflush.h>
34 static void unmap_region(struct mm_struct *mm,
35 struct vm_area_struct *vma, struct vm_area_struct *prev,
36 unsigned long start, unsigned long end);
39 * WARNING: the debugging will use recursive algorithms so never enable this
40 * unless you know what you are doing.
44 /* description of effects of mapping type and prot in current implementation.
45 * this is due to the limited x86 page protection hardware. The expected
46 * behavior is in parens:
49 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
50 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
51 * w: (no) no w: (no) no w: (yes) yes w: (no) no
52 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
54 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
55 * w: (no) no w: (no) no w: (copy) copy w: (no) no
56 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
59 pgprot_t protection_map[16] = {
60 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
61 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
64 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
65 int sysctl_overcommit_ratio = 50; /* default is 50% */
66 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
67 atomic_t vm_committed_space = ATOMIC_INIT(0);
70 * Check that a process has enough memory to allocate a new virtual
71 * mapping. 0 means there is enough memory for the allocation to
72 * succeed and -ENOMEM implies there is not.
74 * We currently support three overcommit policies, which are set via the
75 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
77 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
78 * Additional code 2002 Jul 20 by Robert Love.
80 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
82 * Note this is a helper function intended to be used by LSMs which
83 * wish to use this logic.
85 int __vm_enough_memory(long pages, int cap_sys_admin)
87 unsigned long free, allowed;
89 vm_acct_memory(pages);
92 * Sometimes we want to use more memory than we have
94 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
97 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
100 free = get_page_cache_size();
101 free += nr_swap_pages;
104 * Any slabs which are created with the
105 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
106 * which are reclaimable, under pressure. The dentry
107 * cache and most inode caches should fall into this
109 free += atomic_read(&slab_reclaim_pages);
112 * Leave the last 3% for root
121 * nr_free_pages() is very expensive on large systems,
122 * only call if we're about to fail.
127 * Leave reserved pages. The pages are not for anonymous pages.
129 if (n <= totalreserve_pages)
132 n -= totalreserve_pages;
135 * Leave the last 3% for root
147 allowed = (totalram_pages - hugetlb_total_pages())
148 * sysctl_overcommit_ratio / 100;
150 * Leave the last 3% for root
153 allowed -= allowed / 32;
154 allowed += total_swap_pages;
156 /* Don't let a single process grow too big:
157 leave 3% of the size of this process for other processes */
158 allowed -= current->mm->total_vm / 32;
161 * cast `allowed' as a signed long because vm_committed_space
162 * sometimes has a negative value
164 if (atomic_read(&vm_committed_space) < (long)allowed)
167 vm_unacct_memory(pages);
172 EXPORT_SYMBOL(__vm_enough_memory);
175 * Requires inode->i_mapping->i_mmap_lock
177 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
178 struct file *file, struct address_space *mapping)
180 if (vma->vm_flags & VM_DENYWRITE)
181 atomic_inc(&file->f_dentry->d_inode->i_writecount);
182 if (vma->vm_flags & VM_SHARED)
183 mapping->i_mmap_writable--;
185 flush_dcache_mmap_lock(mapping);
186 if (unlikely(vma->vm_flags & VM_NONLINEAR))
187 list_del_init(&vma->shared.vm_set.list);
189 vma_prio_tree_remove(vma, &mapping->i_mmap);
190 flush_dcache_mmap_unlock(mapping);
194 * Unlink a file-based vm structure from its prio_tree, to hide
195 * vma from rmap and vmtruncate before freeing its page tables.
197 void unlink_file_vma(struct vm_area_struct *vma)
199 struct file *file = vma->vm_file;
202 struct address_space *mapping = file->f_mapping;
203 spin_lock(&mapping->i_mmap_lock);
204 __remove_shared_vm_struct(vma, file, mapping);
205 spin_unlock(&mapping->i_mmap_lock);
210 * Close a vm structure and free it, returning the next.
212 static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
214 struct vm_area_struct *next = vma->vm_next;
217 if (vma->vm_ops && vma->vm_ops->close)
218 vma->vm_ops->close(vma);
221 mpol_free(vma_policy(vma));
222 kmem_cache_free(vm_area_cachep, vma);
226 asmlinkage unsigned long sys_brk(unsigned long brk)
228 unsigned long rlim, retval;
229 unsigned long newbrk, oldbrk;
230 struct mm_struct *mm = current->mm;
232 down_write(&mm->mmap_sem);
234 if (brk < mm->end_code)
238 * Check against rlimit here. If this check is done later after the test
239 * of oldbrk with newbrk then it can escape the test and let the data
240 * segment grow beyond its set limit the in case where the limit is
241 * not page aligned -Ram Gupta
243 rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
244 if (rlim < RLIM_INFINITY && brk - mm->start_data > rlim)
247 newbrk = PAGE_ALIGN(brk);
248 oldbrk = PAGE_ALIGN(mm->brk);
249 if (oldbrk == newbrk)
252 /* Always allow shrinking brk. */
253 if (brk <= mm->brk) {
254 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
259 /* Check against existing mmap mappings. */
260 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
263 /* Ok, looks good - let it rip. */
264 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
270 up_write(&mm->mmap_sem);
275 static int browse_rb(struct rb_root *root)
278 struct rb_node *nd, *pn = NULL;
279 unsigned long prev = 0, pend = 0;
281 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
282 struct vm_area_struct *vma;
283 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
284 if (vma->vm_start < prev)
285 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
286 if (vma->vm_start < pend)
287 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
288 if (vma->vm_start > vma->vm_end)
289 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
294 for (nd = pn; nd; nd = rb_prev(nd)) {
298 printk("backwards %d, forwards %d\n", j, i), i = 0;
302 void validate_mm(struct mm_struct *mm)
306 struct vm_area_struct *tmp = mm->mmap;
311 if (i != mm->map_count)
312 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
313 i = browse_rb(&mm->mm_rb);
314 if (i != mm->map_count)
315 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
319 #define validate_mm(mm) do { } while (0)
322 static struct vm_area_struct *
323 find_vma_prepare(struct mm_struct *mm, unsigned long addr,
324 struct vm_area_struct **pprev, struct rb_node ***rb_link,
325 struct rb_node ** rb_parent)
327 struct vm_area_struct * vma;
328 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
330 __rb_link = &mm->mm_rb.rb_node;
331 rb_prev = __rb_parent = NULL;
335 struct vm_area_struct *vma_tmp;
337 __rb_parent = *__rb_link;
338 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
340 if (vma_tmp->vm_end > addr) {
342 if (vma_tmp->vm_start <= addr)
344 __rb_link = &__rb_parent->rb_left;
346 rb_prev = __rb_parent;
347 __rb_link = &__rb_parent->rb_right;
353 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
354 *rb_link = __rb_link;
355 *rb_parent = __rb_parent;
360 __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
361 struct vm_area_struct *prev, struct rb_node *rb_parent)
363 if (vma->vm_flags & VM_EXEC)
364 arch_add_exec_range(mm, vma->vm_end);
366 vma->vm_next = prev->vm_next;
371 vma->vm_next = rb_entry(rb_parent,
372 struct vm_area_struct, vm_rb);
378 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
379 struct rb_node **rb_link, struct rb_node *rb_parent)
381 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
382 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
385 static inline void __vma_link_file(struct vm_area_struct *vma)
391 struct address_space *mapping = file->f_mapping;
393 if (vma->vm_flags & VM_DENYWRITE)
394 atomic_dec(&file->f_dentry->d_inode->i_writecount);
395 if (vma->vm_flags & VM_SHARED)
396 mapping->i_mmap_writable++;
398 flush_dcache_mmap_lock(mapping);
399 if (unlikely(vma->vm_flags & VM_NONLINEAR))
400 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
402 vma_prio_tree_insert(vma, &mapping->i_mmap);
403 flush_dcache_mmap_unlock(mapping);
408 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
409 struct vm_area_struct *prev, struct rb_node **rb_link,
410 struct rb_node *rb_parent)
412 __vma_link_list(mm, vma, prev, rb_parent);
413 __vma_link_rb(mm, vma, rb_link, rb_parent);
414 __anon_vma_link(vma);
417 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
418 struct vm_area_struct *prev, struct rb_node **rb_link,
419 struct rb_node *rb_parent)
421 struct address_space *mapping = NULL;
424 mapping = vma->vm_file->f_mapping;
427 spin_lock(&mapping->i_mmap_lock);
428 vma->vm_truncate_count = mapping->truncate_count;
432 __vma_link(mm, vma, prev, rb_link, rb_parent);
433 __vma_link_file(vma);
435 anon_vma_unlock(vma);
437 spin_unlock(&mapping->i_mmap_lock);
444 * Helper for vma_adjust in the split_vma insert case:
445 * insert vm structure into list and rbtree and anon_vma,
446 * but it has already been inserted into prio_tree earlier.
449 __insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
451 struct vm_area_struct * __vma, * prev;
452 struct rb_node ** rb_link, * rb_parent;
454 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
455 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
456 __vma_link(mm, vma, prev, rb_link, rb_parent);
461 __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
462 struct vm_area_struct *prev)
464 prev->vm_next = vma->vm_next;
465 rb_erase(&vma->vm_rb, &mm->mm_rb);
466 if (mm->mmap_cache == vma)
467 mm->mmap_cache = prev;
468 if (vma->vm_flags & VM_EXEC)
469 arch_remove_exec_range(mm, vma->vm_end);
473 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
474 * is already present in an i_mmap tree without adjusting the tree.
475 * The following helper function should be used when such adjustments
476 * are necessary. The "insert" vma (if any) is to be inserted
477 * before we drop the necessary locks.
479 void vma_adjust(struct vm_area_struct *vma, unsigned long start,
480 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
482 struct mm_struct *mm = vma->vm_mm;
483 struct vm_area_struct *next = vma->vm_next;
484 struct vm_area_struct *importer = NULL;
485 struct address_space *mapping = NULL;
486 struct prio_tree_root *root = NULL;
487 struct file *file = vma->vm_file;
488 struct anon_vma *anon_vma = NULL;
489 long adjust_next = 0;
492 if (next && !insert) {
493 if (end >= next->vm_end) {
495 * vma expands, overlapping all the next, and
496 * perhaps the one after too (mprotect case 6).
498 again: remove_next = 1 + (end > next->vm_end);
500 anon_vma = next->anon_vma;
502 } else if (end > next->vm_start) {
504 * vma expands, overlapping part of the next:
505 * mprotect case 5 shifting the boundary up.
507 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
508 anon_vma = next->anon_vma;
510 } else if (end < vma->vm_end) {
512 * vma shrinks, and !insert tells it's not
513 * split_vma inserting another: so it must be
514 * mprotect case 4 shifting the boundary down.
516 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
517 anon_vma = next->anon_vma;
523 mapping = file->f_mapping;
524 if (!(vma->vm_flags & VM_NONLINEAR))
525 root = &mapping->i_mmap;
526 spin_lock(&mapping->i_mmap_lock);
528 vma->vm_truncate_count != next->vm_truncate_count) {
530 * unmap_mapping_range might be in progress:
531 * ensure that the expanding vma is rescanned.
533 importer->vm_truncate_count = 0;
536 insert->vm_truncate_count = vma->vm_truncate_count;
538 * Put into prio_tree now, so instantiated pages
539 * are visible to arm/parisc __flush_dcache_page
540 * throughout; but we cannot insert into address
541 * space until vma start or end is updated.
543 __vma_link_file(insert);
548 * When changing only vma->vm_end, we don't really need
549 * anon_vma lock: but is that case worth optimizing out?
552 anon_vma = vma->anon_vma;
554 spin_lock(&anon_vma->lock);
556 * Easily overlooked: when mprotect shifts the boundary,
557 * make sure the expanding vma has anon_vma set if the
558 * shrinking vma had, to cover any anon pages imported.
560 if (importer && !importer->anon_vma) {
561 importer->anon_vma = anon_vma;
562 __anon_vma_link(importer);
567 flush_dcache_mmap_lock(mapping);
568 vma_prio_tree_remove(vma, root);
570 vma_prio_tree_remove(next, root);
573 vma->vm_start = start;
575 vma->vm_pgoff = pgoff;
577 next->vm_start += adjust_next << PAGE_SHIFT;
578 next->vm_pgoff += adjust_next;
583 vma_prio_tree_insert(next, root);
584 vma_prio_tree_insert(vma, root);
585 flush_dcache_mmap_unlock(mapping);
590 * vma_merge has merged next into vma, and needs
591 * us to remove next before dropping the locks.
593 __vma_unlink(mm, next, vma);
595 __remove_shared_vm_struct(next, file, mapping);
597 __anon_vma_merge(vma, next);
600 * split_vma has split insert from vma, and needs
601 * us to insert it before dropping the locks
602 * (it may either follow vma or precede it).
604 __insert_vm_struct(mm, insert);
608 spin_unlock(&anon_vma->lock);
610 spin_unlock(&mapping->i_mmap_lock);
616 mpol_free(vma_policy(next));
617 kmem_cache_free(vm_area_cachep, next);
619 * In mprotect's case 6 (see comments on vma_merge),
620 * we must remove another next too. It would clutter
621 * up the code too much to do both in one go.
623 if (remove_next == 2) {
633 * If the vma has a ->close operation then the driver probably needs to release
634 * per-vma resources, so we don't attempt to merge those.
636 #define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_RESERVED | VM_PFNMAP)
638 static inline int is_mergeable_vma(struct vm_area_struct *vma,
639 struct file *file, unsigned long vm_flags)
641 if (vma->vm_flags != vm_flags)
643 if (vma->vm_file != file)
645 if (vma->vm_ops && vma->vm_ops->close)
650 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
651 struct anon_vma *anon_vma2)
653 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
657 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
658 * in front of (at a lower virtual address and file offset than) the vma.
660 * We cannot merge two vmas if they have differently assigned (non-NULL)
661 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
663 * We don't check here for the merged mmap wrapping around the end of pagecache
664 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
665 * wrap, nor mmaps which cover the final page at index -1UL.
668 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
669 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
671 if (is_mergeable_vma(vma, file, vm_flags) &&
672 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
673 if (vma->vm_pgoff == vm_pgoff)
680 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
681 * beyond (at a higher virtual address and file offset than) the vma.
683 * We cannot merge two vmas if they have differently assigned (non-NULL)
684 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
687 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
688 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
690 if (is_mergeable_vma(vma, file, vm_flags) &&
691 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
693 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
694 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
701 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
702 * whether that can be merged with its predecessor or its successor.
703 * Or both (it neatly fills a hole).
705 * In most cases - when called for mmap, brk or mremap - [addr,end) is
706 * certain not to be mapped by the time vma_merge is called; but when
707 * called for mprotect, it is certain to be already mapped (either at
708 * an offset within prev, or at the start of next), and the flags of
709 * this area are about to be changed to vm_flags - and the no-change
710 * case has already been eliminated.
712 * The following mprotect cases have to be considered, where AAAA is
713 * the area passed down from mprotect_fixup, never extending beyond one
714 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
716 * AAAA AAAA AAAA AAAA
717 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
718 * cannot merge might become might become might become
719 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
720 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
721 * mremap move: PPPPNNNNNNNN 8
723 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
724 * might become case 1 below case 2 below case 3 below
726 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
727 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
729 struct vm_area_struct *vma_merge(struct mm_struct *mm,
730 struct vm_area_struct *prev, unsigned long addr,
731 unsigned long end, unsigned long vm_flags,
732 struct anon_vma *anon_vma, struct file *file,
733 pgoff_t pgoff, struct mempolicy *policy)
735 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
736 struct vm_area_struct *area, *next;
739 * We later require that vma->vm_flags == vm_flags,
740 * so this tests vma->vm_flags & VM_SPECIAL, too.
742 if (vm_flags & VM_SPECIAL)
746 next = prev->vm_next;
750 if (next && next->vm_end == end) /* cases 6, 7, 8 */
751 next = next->vm_next;
754 * Can it merge with the predecessor?
756 if (prev && prev->vm_end == addr &&
757 mpol_equal(vma_policy(prev), policy) &&
758 can_vma_merge_after(prev, vm_flags,
759 anon_vma, file, pgoff)) {
761 * OK, it can. Can we now merge in the successor as well?
763 if (next && end == next->vm_start &&
764 mpol_equal(policy, vma_policy(next)) &&
765 can_vma_merge_before(next, vm_flags,
766 anon_vma, file, pgoff+pglen) &&
767 is_mergeable_anon_vma(prev->anon_vma,
770 vma_adjust(prev, prev->vm_start,
771 next->vm_end, prev->vm_pgoff, NULL);
772 } else /* cases 2, 5, 7 */
773 vma_adjust(prev, prev->vm_start,
774 end, prev->vm_pgoff, NULL);
775 if (prev->vm_flags & VM_EXEC)
776 arch_add_exec_range(mm, prev->vm_end);
781 * Can this new request be merged in front of next?
783 if (next && end == next->vm_start &&
784 mpol_equal(policy, vma_policy(next)) &&
785 can_vma_merge_before(next, vm_flags,
786 anon_vma, file, pgoff+pglen)) {
787 if (prev && addr < prev->vm_end) /* case 4 */
788 vma_adjust(prev, prev->vm_start,
789 addr, prev->vm_pgoff, NULL);
790 else /* cases 3, 8 */
791 vma_adjust(area, addr, next->vm_end,
792 next->vm_pgoff - pglen, NULL);
800 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
801 * neighbouring vmas for a suitable anon_vma, before it goes off
802 * to allocate a new anon_vma. It checks because a repetitive
803 * sequence of mprotects and faults may otherwise lead to distinct
804 * anon_vmas being allocated, preventing vma merge in subsequent
807 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
809 struct vm_area_struct *near;
810 unsigned long vm_flags;
817 * Since only mprotect tries to remerge vmas, match flags
818 * which might be mprotected into each other later on.
819 * Neither mlock nor madvise tries to remerge at present,
820 * so leave their flags as obstructing a merge.
822 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
823 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
825 if (near->anon_vma && vma->vm_end == near->vm_start &&
826 mpol_equal(vma_policy(vma), vma_policy(near)) &&
827 can_vma_merge_before(near, vm_flags,
828 NULL, vma->vm_file, vma->vm_pgoff +
829 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
830 return near->anon_vma;
833 * It is potentially slow to have to call find_vma_prev here.
834 * But it's only on the first write fault on the vma, not
835 * every time, and we could devise a way to avoid it later
836 * (e.g. stash info in next's anon_vma_node when assigning
837 * an anon_vma, or when trying vma_merge). Another time.
839 BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma);
843 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
844 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
846 if (near->anon_vma && near->vm_end == vma->vm_start &&
847 mpol_equal(vma_policy(near), vma_policy(vma)) &&
848 can_vma_merge_after(near, vm_flags,
849 NULL, vma->vm_file, vma->vm_pgoff))
850 return near->anon_vma;
853 * There's no absolute need to look only at touching neighbours:
854 * we could search further afield for "compatible" anon_vmas.
855 * But it would probably just be a waste of time searching,
856 * or lead to too many vmas hanging off the same anon_vma.
857 * We're trying to allow mprotect remerging later on,
858 * not trying to minimize memory used for anon_vmas.
863 #ifdef CONFIG_PROC_FS
864 void vm_stat_account(struct mm_struct *mm, unsigned long flags,
865 struct file *file, long pages)
867 const unsigned long stack_flags
868 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
871 mm->shared_vm += pages;
872 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
873 mm->exec_vm += pages;
874 } else if (flags & stack_flags)
875 mm->stack_vm += pages;
876 if (flags & (VM_RESERVED|VM_IO))
877 mm->reserved_vm += pages;
879 #endif /* CONFIG_PROC_FS */
882 * The caller must hold down_write(current->mm->mmap_sem).
885 unsigned long do_mmap_pgoff(struct file * file, unsigned long addr,
886 unsigned long len, unsigned long prot,
887 unsigned long flags, unsigned long pgoff)
889 struct mm_struct * mm = current->mm;
890 struct vm_area_struct * vma, * prev;
892 unsigned int vm_flags;
893 int correct_wcount = 0;
895 struct rb_node ** rb_link, * rb_parent;
897 unsigned long charged = 0, reqprot = prot;
900 if (is_file_hugepages(file))
903 if (!file->f_op || !file->f_op->mmap)
906 if ((prot & PROT_EXEC) &&
907 (file->f_vfsmnt->mnt_flags & MNT_NOEXEC))
911 * Does the application expect PROT_READ to imply PROT_EXEC?
913 * (the exception is when the underlying filesystem is noexec
914 * mounted, in which case we dont add PROT_EXEC.)
916 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
917 if (!(file && (file->f_vfsmnt->mnt_flags & MNT_NOEXEC)))
923 error = arch_mmap_check(addr, len, flags);
927 /* Careful about overflows.. */
928 len = PAGE_ALIGN(len);
929 if (!len || len > TASK_SIZE)
932 /* offset overflow? */
933 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
936 /* Too many mappings? */
937 if (mm->map_count > sysctl_max_map_count)
940 /* Obtain the address to map to. we verify (or select) it and ensure
941 * that it represents a valid section of the address space.
943 addr = get_unmapped_area_prot(file, addr, len, pgoff, flags, prot & PROT_EXEC);
944 if (addr & ~PAGE_MASK)
947 /* Do simple checking here so the lower-level routines won't have
948 * to. we assume access permissions have been handled by the open
949 * of the memory object, so we don't do any here.
951 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
952 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
954 if (flags & MAP_LOCKED) {
957 vm_flags |= VM_LOCKED;
959 /* mlock MCL_FUTURE? */
960 if (vm_flags & VM_LOCKED) {
961 unsigned long locked, lock_limit;
962 locked = len >> PAGE_SHIFT;
963 locked += mm->locked_vm;
964 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
965 lock_limit >>= PAGE_SHIFT;
966 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
970 inode = file ? file->f_dentry->d_inode : NULL;
973 switch (flags & MAP_TYPE) {
975 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
979 * Make sure we don't allow writing to an append-only
982 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
986 * Make sure there are no mandatory locks on the file.
988 if (locks_verify_locked(inode))
991 vm_flags |= VM_SHARED | VM_MAYSHARE;
992 if (!(file->f_mode & FMODE_WRITE))
993 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
997 if (!(file->f_mode & FMODE_READ))
1005 switch (flags & MAP_TYPE) {
1007 vm_flags |= VM_SHARED | VM_MAYSHARE;
1011 * Set pgoff according to addr for anon_vma.
1013 pgoff = addr >> PAGE_SHIFT;
1020 error = security_file_mmap(file, reqprot, prot, flags);
1024 /* Clear old maps */
1027 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1028 if (vma && vma->vm_start < addr + len) {
1029 if (do_munmap(mm, addr, len))
1034 /* Check against address space limit. */
1035 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1038 if (accountable && (!(flags & MAP_NORESERVE) ||
1039 sysctl_overcommit_memory == OVERCOMMIT_NEVER)) {
1040 if (vm_flags & VM_SHARED) {
1041 /* Check memory availability in shmem_file_setup? */
1042 vm_flags |= VM_ACCOUNT;
1043 } else if (vm_flags & VM_WRITE) {
1045 * Private writable mapping: check memory availability
1047 charged = len >> PAGE_SHIFT;
1048 if (security_vm_enough_memory(charged))
1050 vm_flags |= VM_ACCOUNT;
1055 * Can we just expand an old private anonymous mapping?
1056 * The VM_SHARED test is necessary because shmem_zero_setup
1057 * will create the file object for a shared anonymous map below.
1059 if (!file && !(vm_flags & VM_SHARED) &&
1060 vma_merge(mm, prev, addr, addr + len, vm_flags,
1061 NULL, NULL, pgoff, NULL))
1065 * Determine the object being mapped and call the appropriate
1066 * specific mapper. the address has already been validated, but
1067 * not unmapped, but the maps are removed from the list.
1069 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1076 vma->vm_start = addr;
1077 vma->vm_end = addr + len;
1078 vma->vm_flags = vm_flags;
1079 vma->vm_page_prot = protection_map[vm_flags & 0x0f];
1080 vma->vm_pgoff = pgoff;
1084 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1086 if (vm_flags & VM_DENYWRITE) {
1087 error = deny_write_access(file);
1092 vma->vm_file = file;
1094 error = file->f_op->mmap(file, vma);
1096 goto unmap_and_free_vma;
1097 } else if (vm_flags & VM_SHARED) {
1098 error = shmem_zero_setup(vma);
1103 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
1104 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
1105 * that memory reservation must be checked; but that reservation
1106 * belongs to shared memory object, not to vma: so now clear it.
1108 if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT))
1109 vma->vm_flags &= ~VM_ACCOUNT;
1111 /* Can addr have changed??
1113 * Answer: Yes, several device drivers can do it in their
1114 * f_op->mmap method. -DaveM
1116 addr = vma->vm_start;
1117 pgoff = vma->vm_pgoff;
1118 vm_flags = vma->vm_flags;
1120 if (!file || !vma_merge(mm, prev, addr, vma->vm_end,
1121 vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
1122 file = vma->vm_file;
1123 vma_link(mm, vma, prev, rb_link, rb_parent);
1125 atomic_inc(&inode->i_writecount);
1129 atomic_inc(&inode->i_writecount);
1132 mpol_free(vma_policy(vma));
1133 kmem_cache_free(vm_area_cachep, vma);
1136 vx_vmpages_add(mm, len >> PAGE_SHIFT);
1137 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1138 if (vm_flags & VM_LOCKED) {
1139 vx_vmlocked_add(mm, len >> PAGE_SHIFT);
1140 make_pages_present(addr, addr + len);
1142 if (flags & MAP_POPULATE) {
1143 up_write(&mm->mmap_sem);
1144 sys_remap_file_pages(addr, len, 0,
1145 pgoff, flags & MAP_NONBLOCK);
1146 down_write(&mm->mmap_sem);
1152 atomic_inc(&inode->i_writecount);
1153 vma->vm_file = NULL;
1156 /* Undo any partial mapping done by a device driver. */
1157 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1160 kmem_cache_free(vm_area_cachep, vma);
1163 vm_unacct_memory(charged);
1167 EXPORT_SYMBOL(do_mmap_pgoff);
1169 /* Get an address range which is currently unmapped.
1170 * For shmat() with addr=0.
1172 * Ugly calling convention alert:
1173 * Return value with the low bits set means error value,
1175 * if (ret & ~PAGE_MASK)
1178 * This function "knows" that -ENOMEM has the bits set.
1180 #ifndef HAVE_ARCH_UNMAPPED_AREA
1182 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1183 unsigned long len, unsigned long pgoff, unsigned long flags)
1185 struct mm_struct *mm = current->mm;
1186 struct vm_area_struct *vma;
1187 unsigned long start_addr;
1189 if (len > TASK_SIZE)
1193 addr = PAGE_ALIGN(addr);
1194 vma = find_vma(mm, addr);
1195 if (TASK_SIZE - len >= addr &&
1196 (!vma || addr + len <= vma->vm_start))
1199 if (len > mm->cached_hole_size) {
1200 start_addr = addr = mm->free_area_cache;
1202 start_addr = addr = TASK_UNMAPPED_BASE;
1203 mm->cached_hole_size = 0;
1207 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1208 /* At this point: (!vma || addr < vma->vm_end). */
1209 if (TASK_SIZE - len < addr) {
1211 * Start a new search - just in case we missed
1214 if (start_addr != TASK_UNMAPPED_BASE) {
1215 addr = TASK_UNMAPPED_BASE;
1217 mm->cached_hole_size = 0;
1222 if (!vma || addr + len <= vma->vm_start) {
1224 * Remember the place where we stopped the search:
1226 mm->free_area_cache = addr + len;
1229 if (addr + mm->cached_hole_size < vma->vm_start)
1230 mm->cached_hole_size = vma->vm_start - addr;
1236 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1239 * Is this a new hole at the lowest possible address?
1241 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1242 mm->free_area_cache = addr;
1243 mm->cached_hole_size = ~0UL;
1248 * This mmap-allocator allocates new areas top-down from below the
1249 * stack's low limit (the base):
1251 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1253 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1254 const unsigned long len, const unsigned long pgoff,
1255 const unsigned long flags)
1257 struct vm_area_struct *vma;
1258 struct mm_struct *mm = current->mm;
1259 unsigned long addr = addr0;
1261 /* requested length too big for entire address space */
1262 if (len > TASK_SIZE)
1265 /* requesting a specific address */
1267 addr = PAGE_ALIGN(addr);
1268 vma = find_vma(mm, addr);
1269 if (TASK_SIZE - len >= addr &&
1270 (!vma || addr + len <= vma->vm_start))
1274 /* check if free_area_cache is useful for us */
1275 if (len <= mm->cached_hole_size) {
1276 mm->cached_hole_size = 0;
1277 mm->free_area_cache = mm->mmap_base;
1280 /* either no address requested or can't fit in requested address hole */
1281 addr = mm->free_area_cache;
1283 /* make sure it can fit in the remaining address space */
1285 vma = find_vma(mm, addr-len);
1286 if (!vma || addr <= vma->vm_start)
1287 /* remember the address as a hint for next time */
1288 return (mm->free_area_cache = addr-len);
1291 if (mm->mmap_base < len)
1294 addr = mm->mmap_base-len;
1298 * Lookup failure means no vma is above this address,
1299 * else if new region fits below vma->vm_start,
1300 * return with success:
1302 vma = find_vma(mm, addr);
1303 if (!vma || addr+len <= vma->vm_start)
1304 /* remember the address as a hint for next time */
1305 return (mm->free_area_cache = addr);
1307 /* remember the largest hole we saw so far */
1308 if (addr + mm->cached_hole_size < vma->vm_start)
1309 mm->cached_hole_size = vma->vm_start - addr;
1311 /* try just below the current vma->vm_start */
1312 addr = vma->vm_start-len;
1313 } while (len < vma->vm_start);
1317 * A failed mmap() very likely causes application failure,
1318 * so fall back to the bottom-up function here. This scenario
1319 * can happen with large stack limits and large mmap()
1322 mm->cached_hole_size = ~0UL;
1323 mm->free_area_cache = TASK_UNMAPPED_BASE;
1324 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1326 * Restore the topdown base:
1328 mm->free_area_cache = mm->mmap_base;
1329 mm->cached_hole_size = ~0UL;
1335 void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
1338 * Is this a new hole at the highest possible address?
1340 if (addr > mm->free_area_cache)
1341 mm->free_area_cache = addr;
1343 /* dont allow allocations above current base */
1344 if (mm->free_area_cache > mm->mmap_base)
1345 mm->free_area_cache = mm->mmap_base;
1350 get_unmapped_area_prot(struct file *file, unsigned long addr, unsigned long len,
1351 unsigned long pgoff, unsigned long flags, int exec)
1355 if (!(flags & MAP_FIXED)) {
1356 unsigned long (*get_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
1358 if (exec && current->mm->get_unmapped_exec_area)
1359 get_area = current->mm->get_unmapped_exec_area;
1361 get_area = current->mm->get_unmapped_area;
1363 if (file && file->f_op && file->f_op->get_unmapped_area)
1364 get_area = file->f_op->get_unmapped_area;
1365 addr = get_area(file, addr, len, pgoff, flags);
1366 if (IS_ERR_VALUE(addr))
1370 if (addr > TASK_SIZE - len)
1372 if (addr & ~PAGE_MASK)
1374 if (file && is_file_hugepages(file)) {
1376 * Check if the given range is hugepage aligned, and
1377 * can be made suitable for hugepages.
1379 ret = prepare_hugepage_range(addr, len);
1382 * Ensure that a normal request is not falling in a
1383 * reserved hugepage range. For some archs like IA-64,
1384 * there is a separate region for hugepages.
1386 ret = is_hugepage_only_range(current->mm, addr, len);
1393 EXPORT_SYMBOL(get_unmapped_area_prot);
1395 #define SHLIB_BASE 0x00111000
1397 unsigned long arch_get_unmapped_exec_area(struct file *filp, unsigned long addr0,
1398 unsigned long len0, unsigned long pgoff, unsigned long flags)
1400 unsigned long addr = addr0, len = len0;
1401 struct mm_struct *mm = current->mm;
1402 struct vm_area_struct *vma;
1405 if (len > TASK_SIZE)
1408 if (!addr && !(flags & MAP_FIXED))
1409 addr = randomize_range(SHLIB_BASE, 0x01000000, len);
1412 addr = PAGE_ALIGN(addr);
1413 vma = find_vma(mm, addr);
1414 if (TASK_SIZE - len >= addr &&
1415 (!vma || addr + len <= vma->vm_start)) {
1421 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1422 /* At this point: (!vma || addr < vma->vm_end). */
1423 if (TASK_SIZE - len < addr)
1426 if (!vma || addr + len <= vma->vm_start) {
1428 * Must not let a PROT_EXEC mapping get into the
1431 if (addr + len > mm->brk)
1435 * Up until the brk area we randomize addresses
1436 * as much as possible:
1438 if (addr >= 0x01000000) {
1439 tmp = randomize_range(0x01000000, PAGE_ALIGN(max(mm->start_brk, (unsigned long)0x08000000)), len);
1440 vma = find_vma(mm, tmp);
1441 if (TASK_SIZE - len >= tmp &&
1442 (!vma || tmp + len <= vma->vm_start))
1446 * Ok, randomization didnt work out - return
1447 * the result of the linear search:
1455 return current->mm->get_unmapped_area(filp, addr0, len0, pgoff, flags);
1459 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1460 struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
1462 struct vm_area_struct *vma = NULL;
1465 /* Check the cache first. */
1466 /* (Cache hit rate is typically around 35%.) */
1467 vma = mm->mmap_cache;
1468 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1469 struct rb_node * rb_node;
1471 rb_node = mm->mm_rb.rb_node;
1475 struct vm_area_struct * vma_tmp;
1477 vma_tmp = rb_entry(rb_node,
1478 struct vm_area_struct, vm_rb);
1480 if (vma_tmp->vm_end > addr) {
1482 if (vma_tmp->vm_start <= addr)
1484 rb_node = rb_node->rb_left;
1486 rb_node = rb_node->rb_right;
1489 mm->mmap_cache = vma;
1495 EXPORT_SYMBOL(find_vma);
1497 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1498 struct vm_area_struct *
1499 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1500 struct vm_area_struct **pprev)
1502 struct vm_area_struct *vma = NULL, *prev = NULL;
1503 struct rb_node * rb_node;
1507 /* Guard against addr being lower than the first VMA */
1510 /* Go through the RB tree quickly. */
1511 rb_node = mm->mm_rb.rb_node;
1514 struct vm_area_struct *vma_tmp;
1515 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1517 if (addr < vma_tmp->vm_end) {
1518 rb_node = rb_node->rb_left;
1521 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1523 rb_node = rb_node->rb_right;
1529 return prev ? prev->vm_next : vma;
1532 static int over_stack_limit(unsigned long sz)
1534 if (sz < EXEC_STACK_BIAS)
1536 return (sz - EXEC_STACK_BIAS) >
1537 current->signal->rlim[RLIMIT_STACK].rlim_cur;
1541 * Verify that the stack growth is acceptable and
1542 * update accounting. This is shared with both the
1543 * grow-up and grow-down cases.
1545 static int acct_stack_growth(struct vm_area_struct * vma, unsigned long size, unsigned long grow)
1547 struct mm_struct *mm = vma->vm_mm;
1548 struct rlimit *rlim = current->signal->rlim;
1550 /* address space limit tests */
1551 if (!may_expand_vm(mm, grow))
1554 /* Stack limit test */
1555 if (over_stack_limit(size))
1558 /* mlock limit tests */
1559 if (vma->vm_flags & VM_LOCKED) {
1560 unsigned long locked;
1561 unsigned long limit;
1562 locked = mm->locked_vm + grow;
1563 limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
1564 if (locked > limit && !capable(CAP_IPC_LOCK))
1569 * Overcommit.. This must be the final test, as it will
1570 * update security statistics.
1572 if (security_vm_enough_memory(grow))
1575 /* Ok, everything looks good - let it rip */
1576 vx_vmpages_add(mm, grow);
1577 if (vma->vm_flags & VM_LOCKED)
1578 vx_vmlocked_add(mm, grow);
1579 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1583 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1585 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1586 * vma is the last one with address > vma->vm_end. Have to extend vma.
1591 int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1595 if (!(vma->vm_flags & VM_GROWSUP))
1599 * We must make sure the anon_vma is allocated
1600 * so that the anon_vma locking is not a noop.
1602 if (unlikely(anon_vma_prepare(vma)))
1607 * vma->vm_start/vm_end cannot change under us because the caller
1608 * is required to hold the mmap_sem in read mode. We need the
1609 * anon_vma lock to serialize against concurrent expand_stacks.
1611 address += 4 + PAGE_SIZE - 1;
1612 address &= PAGE_MASK;
1615 /* Somebody else might have raced and expanded it already */
1616 if (address > vma->vm_end) {
1617 unsigned long size, grow;
1619 size = address - vma->vm_start;
1620 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1622 error = acct_stack_growth(vma, size, grow);
1624 vma->vm_end = address;
1626 anon_vma_unlock(vma);
1629 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1631 #ifdef CONFIG_STACK_GROWSUP
1632 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1634 return expand_upwards(vma, address);
1637 struct vm_area_struct *
1638 find_extend_vma(struct mm_struct *mm, unsigned long addr)
1640 struct vm_area_struct *vma, *prev;
1643 vma = find_vma_prev(mm, addr, &prev);
1644 if (vma && (vma->vm_start <= addr))
1646 if (!prev || expand_stack(prev, addr))
1648 if (prev->vm_flags & VM_LOCKED) {
1649 make_pages_present(addr, prev->vm_end);
1655 * vma is the first one with address < vma->vm_start. Have to extend vma.
1657 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1662 * We must make sure the anon_vma is allocated
1663 * so that the anon_vma locking is not a noop.
1665 if (unlikely(anon_vma_prepare(vma)))
1670 * vma->vm_start/vm_end cannot change under us because the caller
1671 * is required to hold the mmap_sem in read mode. We need the
1672 * anon_vma lock to serialize against concurrent expand_stacks.
1674 address &= PAGE_MASK;
1677 /* Somebody else might have raced and expanded it already */
1678 if (address < vma->vm_start) {
1679 unsigned long size, grow;
1681 size = vma->vm_end - address;
1682 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1684 error = acct_stack_growth(vma, size, grow);
1686 vma->vm_start = address;
1687 vma->vm_pgoff -= grow;
1690 anon_vma_unlock(vma);
1694 struct vm_area_struct *
1695 find_extend_vma(struct mm_struct * mm, unsigned long addr)
1697 struct vm_area_struct * vma;
1698 unsigned long start;
1701 vma = find_vma(mm,addr);
1704 if (vma->vm_start <= addr)
1706 if (!(vma->vm_flags & VM_GROWSDOWN))
1708 start = vma->vm_start;
1709 if (expand_stack(vma, addr))
1711 if (vma->vm_flags & VM_LOCKED) {
1712 make_pages_present(addr, start);
1719 * Ok - we have the memory areas we should free on the vma list,
1720 * so release them, and do the vma updates.
1722 * Called with the mm semaphore held.
1724 static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
1726 /* Update high watermark before we lower total_vm */
1727 update_hiwater_vm(mm);
1729 long nrpages = vma_pages(vma);
1731 vx_vmpages_sub(mm, nrpages);
1732 if (vma->vm_flags & VM_LOCKED)
1733 vx_vmlocked_sub(mm, nrpages);
1734 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
1735 vma = remove_vma(vma);
1741 * Get rid of page table information in the indicated region.
1743 * Called with the mm semaphore held.
1745 static void unmap_region(struct mm_struct *mm,
1746 struct vm_area_struct *vma, struct vm_area_struct *prev,
1747 unsigned long start, unsigned long end)
1749 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
1750 struct mmu_gather *tlb;
1751 unsigned long nr_accounted = 0;
1754 tlb = tlb_gather_mmu(mm, 0);
1755 update_hiwater_rss(mm);
1756 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
1757 vm_unacct_memory(nr_accounted);
1758 free_pgtables(&tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
1759 next? next->vm_start: 0);
1760 tlb_finish_mmu(tlb, start, end);
1764 * Create a list of vma's touched by the unmap, removing them from the mm's
1765 * vma list as we go..
1768 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1769 struct vm_area_struct *prev, unsigned long end)
1771 struct vm_area_struct **insertion_point;
1772 struct vm_area_struct *tail_vma = NULL;
1775 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1777 rb_erase(&vma->vm_rb, &mm->mm_rb);
1781 } while (vma && vma->vm_start < end);
1782 *insertion_point = vma;
1783 tail_vma->vm_next = NULL;
1784 if (mm->unmap_area == arch_unmap_area)
1785 addr = prev ? prev->vm_end : mm->mmap_base;
1787 addr = vma ? vma->vm_start : mm->mmap_base;
1788 mm->unmap_area(mm, addr);
1789 mm->mmap_cache = NULL; /* Kill the cache. */
1793 * Split a vma into two pieces at address 'addr', a new vma is allocated
1794 * either for the first part or the the tail.
1796 int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1797 unsigned long addr, int new_below)
1799 struct mempolicy *pol;
1800 struct vm_area_struct *new;
1802 if (is_vm_hugetlb_page(vma) && (addr & ~HPAGE_MASK))
1805 if (mm->map_count >= sysctl_max_map_count)
1808 new = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1812 /* most fields are the same, copy all, and then fixup */
1818 new->vm_start = addr;
1819 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1822 pol = mpol_copy(vma_policy(vma));
1824 kmem_cache_free(vm_area_cachep, new);
1825 return PTR_ERR(pol);
1827 vma_set_policy(new, pol);
1830 get_file(new->vm_file);
1832 if (new->vm_ops && new->vm_ops->open)
1833 new->vm_ops->open(new);
1836 unsigned long old_end = vma->vm_end;
1838 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1839 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1840 if (vma->vm_flags & VM_EXEC)
1841 arch_remove_exec_range(mm, old_end);
1843 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1848 /* Munmap is split into 2 main parts -- this part which finds
1849 * what needs doing, and the areas themselves, which do the
1850 * work. This now handles partial unmappings.
1851 * Jeremy Fitzhardinge <jeremy@goop.org>
1853 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1856 struct vm_area_struct *vma, *prev, *last;
1858 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1861 if ((len = PAGE_ALIGN(len)) == 0)
1864 /* Find the first overlapping VMA */
1865 vma = find_vma_prev(mm, start, &prev);
1868 /* we have start < vma->vm_end */
1870 /* if it doesn't overlap, we have nothing.. */
1872 if (vma->vm_start >= end)
1876 * If we need to split any vma, do it now to save pain later.
1878 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1879 * unmapped vm_area_struct will remain in use: so lower split_vma
1880 * places tmp vma above, and higher split_vma places tmp vma below.
1882 if (start > vma->vm_start) {
1883 int error = split_vma(mm, vma, start, 0);
1889 /* Does it split the last one? */
1890 last = find_vma(mm, end);
1891 if (last && end > last->vm_start) {
1892 int error = split_vma(mm, last, end, 1);
1896 vma = prev? prev->vm_next: mm->mmap;
1899 * Remove the vma's, and unmap the actual pages
1901 detach_vmas_to_be_unmapped(mm, vma, prev, end);
1902 unmap_region(mm, vma, prev, start, end);
1904 /* Fix up all other VM information */
1905 remove_vma_list(mm, vma);
1910 EXPORT_SYMBOL(do_munmap);
1912 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1915 struct mm_struct *mm = current->mm;
1917 profile_munmap(addr);
1919 down_write(&mm->mmap_sem);
1920 ret = do_munmap(mm, addr, len);
1921 up_write(&mm->mmap_sem);
1925 static inline void verify_mm_writelocked(struct mm_struct *mm)
1927 #ifdef CONFIG_DEBUG_VM
1928 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
1930 up_read(&mm->mmap_sem);
1936 * this is really a simplified "do_mmap". it only handles
1937 * anonymous maps. eventually we may be able to do some
1938 * brk-specific accounting here.
1940 unsigned long do_brk(unsigned long addr, unsigned long len)
1942 struct mm_struct * mm = current->mm;
1943 struct vm_area_struct * vma, * prev;
1944 unsigned long flags;
1945 struct rb_node ** rb_link, * rb_parent;
1946 pgoff_t pgoff = addr >> PAGE_SHIFT;
1949 len = PAGE_ALIGN(len);
1953 if ((addr + len) > TASK_SIZE || (addr + len) < addr)
1956 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
1958 error = arch_mmap_check(addr, len, flags);
1965 if (mm->def_flags & VM_LOCKED) {
1966 unsigned long locked, lock_limit;
1967 locked = len >> PAGE_SHIFT;
1968 locked += mm->locked_vm;
1969 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
1970 lock_limit >>= PAGE_SHIFT;
1971 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1973 if (!vx_vmlocked_avail(mm, len >> PAGE_SHIFT))
1978 * mm->mmap_sem is required to protect against another thread
1979 * changing the mappings in case we sleep.
1981 verify_mm_writelocked(mm);
1984 * Clear old maps. this also does some error checking for us
1987 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1988 if (vma && vma->vm_start < addr + len) {
1989 if (do_munmap(mm, addr, len))
1994 /* Check against address space limits *after* clearing old maps... */
1995 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1998 if (mm->map_count > sysctl_max_map_count)
2001 if (security_vm_enough_memory(len >> PAGE_SHIFT) ||
2002 !vx_vmpages_avail(mm, len >> PAGE_SHIFT))
2005 /* Can we just expand an old private anonymous mapping? */
2006 if (vma_merge(mm, prev, addr, addr + len, flags,
2007 NULL, NULL, pgoff, NULL))
2011 * create a vma struct for an anonymous mapping
2013 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2015 vm_unacct_memory(len >> PAGE_SHIFT);
2020 vma->vm_start = addr;
2021 vma->vm_end = addr + len;
2022 vma->vm_pgoff = pgoff;
2023 vma->vm_flags = flags;
2024 vma->vm_page_prot = protection_map[flags & 0x0f];
2025 vma_link(mm, vma, prev, rb_link, rb_parent);
2027 vx_vmpages_add(mm, len >> PAGE_SHIFT);
2028 if (flags & VM_LOCKED) {
2029 vx_vmlocked_add(mm, len >> PAGE_SHIFT);
2030 make_pages_present(addr, addr + len);
2035 EXPORT_SYMBOL(do_brk);
2037 /* Release all mmaps. */
2038 void exit_mmap(struct mm_struct *mm)
2040 struct mmu_gather *tlb;
2041 struct vm_area_struct *vma = mm->mmap;
2042 unsigned long nr_accounted = 0;
2045 #ifdef arch_exit_mmap
2051 tlb = tlb_gather_mmu(mm, 1);
2052 /* Don't update_hiwater_rss(mm) here, do_exit already did */
2053 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2054 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
2055 vm_unacct_memory(nr_accounted);
2056 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
2057 tlb_finish_mmu(tlb, 0, end);
2058 arch_flush_exec_range(mm);
2060 set_mm_counter(mm, file_rss, 0);
2061 set_mm_counter(mm, anon_rss, 0);
2062 vx_vmpages_sub(mm, mm->total_vm);
2063 vx_vmlocked_sub(mm, mm->locked_vm);
2066 * Walk the list again, actually closing and freeing it,
2067 * with preemption enabled, without holding any MM locks.
2070 vma = remove_vma(vma);
2072 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
2075 /* Insert vm structure into process list sorted by address
2076 * and into the inode's i_mmap tree. If vm_file is non-NULL
2077 * then i_mmap_lock is taken here.
2079 int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2081 struct vm_area_struct * __vma, * prev;
2082 struct rb_node ** rb_link, * rb_parent;
2085 * The vm_pgoff of a purely anonymous vma should be irrelevant
2086 * until its first write fault, when page's anon_vma and index
2087 * are set. But now set the vm_pgoff it will almost certainly
2088 * end up with (unless mremap moves it elsewhere before that
2089 * first wfault), so /proc/pid/maps tells a consistent story.
2091 * By setting it to reflect the virtual start address of the
2092 * vma, merges and splits can happen in a seamless way, just
2093 * using the existing file pgoff checks and manipulations.
2094 * Similarly in do_mmap_pgoff and in do_brk.
2096 if (!vma->vm_file) {
2097 BUG_ON(vma->anon_vma);
2098 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2100 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2101 if (__vma && __vma->vm_start < vma->vm_end)
2103 if ((vma->vm_flags & VM_ACCOUNT) &&
2104 (security_vm_enough_memory(vma_pages(vma)) ||
2105 !vx_vmpages_avail(mm, vma_pages(vma))))
2107 vma_link(mm, vma, prev, rb_link, rb_parent);
2112 * Copy the vma structure to a new location in the same mm,
2113 * prior to moving page table entries, to effect an mremap move.
2115 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2116 unsigned long addr, unsigned long len, pgoff_t pgoff)
2118 struct vm_area_struct *vma = *vmap;
2119 unsigned long vma_start = vma->vm_start;
2120 struct mm_struct *mm = vma->vm_mm;
2121 struct vm_area_struct *new_vma, *prev;
2122 struct rb_node **rb_link, *rb_parent;
2123 struct mempolicy *pol;
2126 * If anonymous vma has not yet been faulted, update new pgoff
2127 * to match new location, to increase its chance of merging.
2129 if (!vma->vm_file && !vma->anon_vma)
2130 pgoff = addr >> PAGE_SHIFT;
2132 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2133 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2134 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2137 * Source vma may have been merged into new_vma
2139 if (vma_start >= new_vma->vm_start &&
2140 vma_start < new_vma->vm_end)
2143 new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
2146 pol = mpol_copy(vma_policy(vma));
2148 kmem_cache_free(vm_area_cachep, new_vma);
2151 vma_set_policy(new_vma, pol);
2152 new_vma->vm_start = addr;
2153 new_vma->vm_end = addr + len;
2154 new_vma->vm_pgoff = pgoff;
2155 if (new_vma->vm_file)
2156 get_file(new_vma->vm_file);
2157 if (new_vma->vm_ops && new_vma->vm_ops->open)
2158 new_vma->vm_ops->open(new_vma);
2159 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2166 * Return true if the calling process may expand its vm space by the passed
2169 int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2171 unsigned long cur = mm->total_vm; /* pages */
2174 lim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
2176 if (cur + npages > lim)
2178 if (!vx_vmpages_avail(mm, npages))
2184 static struct page *
2185 special_mapping_nopage(struct vm_area_struct *vma,
2186 unsigned long address, int *type)
2188 struct page **pages;
2190 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
2192 address -= vma->vm_start;
2193 for (pages = vma->vm_private_data; address > 0 && *pages; ++pages)
2194 address -= PAGE_SIZE;
2201 return NOPAGE_SIGBUS;
2204 static struct vm_operations_struct special_mapping_vmops = {
2205 .nopage = special_mapping_nopage,
2208 unsigned int vdso_populate = 1;
2211 * Insert a new vma covering the given region, with the given flags and
2212 * protections. Its pages are supplied by the given null-terminated array.
2213 * The region past the last page supplied will always produce SIGBUS.
2214 * The array pointer and the pages it points to are assumed to stay alive
2215 * for as long as this mapping might exist.
2217 int install_special_mapping(struct mm_struct *mm,
2218 unsigned long addr, unsigned long len,
2219 unsigned long vm_flags, pgprot_t pgprot,
2220 struct page **pages)
2222 struct vm_area_struct *vma;
2225 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
2226 if (unlikely(vma == NULL))
2228 memset(vma, 0, sizeof(*vma));
2231 vma->vm_start = addr;
2232 vma->vm_end = addr + len;
2234 vma->vm_flags = vm_flags;
2235 vma->vm_page_prot = pgprot;
2237 vma->vm_ops = &special_mapping_vmops;
2238 vma->vm_private_data = pages;
2240 insert_vm_struct(mm, vma);
2241 mm->total_vm += len >> PAGE_SHIFT;
2248 struct page *page = *pages++;
2250 err = install_page(mm, vma, addr, page, vma->vm_page_prot);