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>
29 #include <linux/vs_base.h>
30 #include <linux/vs_memory.h>
32 #include <asm/uaccess.h>
33 #include <asm/cacheflush.h>
36 static void unmap_region(struct mm_struct *mm,
37 struct vm_area_struct *vma, struct vm_area_struct *prev,
38 unsigned long start, unsigned long end);
41 * WARNING: the debugging will use recursive algorithms so never enable this
42 * unless you know what you are doing.
46 /* description of effects of mapping type and prot in current implementation.
47 * this is due to the limited x86 page protection hardware. The expected
48 * behavior is in parens:
51 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
52 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
53 * w: (no) no w: (no) no w: (yes) yes w: (no) no
54 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
56 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
57 * w: (no) no w: (no) no w: (copy) copy w: (no) no
58 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
61 pgprot_t protection_map[16] = {
62 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
63 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
66 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
67 int sysctl_overcommit_ratio = 50; /* default is 50% */
68 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
69 atomic_t vm_committed_space = ATOMIC_INIT(0);
72 * Check that a process has enough memory to allocate a new virtual
73 * mapping. 0 means there is enough memory for the allocation to
74 * succeed and -ENOMEM implies there is not.
76 * We currently support three overcommit policies, which are set via the
77 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
79 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
80 * Additional code 2002 Jul 20 by Robert Love.
82 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
84 * Note this is a helper function intended to be used by LSMs which
85 * wish to use this logic.
87 int __vm_enough_memory(long pages, int cap_sys_admin)
89 unsigned long free, allowed;
91 vm_acct_memory(pages);
94 * Sometimes we want to use more memory than we have
96 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
99 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
102 free = get_page_cache_size();
103 free += nr_swap_pages;
106 * Any slabs which are created with the
107 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
108 * which are reclaimable, under pressure. The dentry
109 * cache and most inode caches should fall into this
111 free += atomic_read(&slab_reclaim_pages);
114 * Leave the last 3% for root
123 * nr_free_pages() is very expensive on large systems,
124 * only call if we're about to fail.
129 * Leave reserved pages. The pages are not for anonymous pages.
131 if (n <= totalreserve_pages)
134 n -= totalreserve_pages;
137 * Leave the last 3% for root
149 allowed = (totalram_pages - hugetlb_total_pages())
150 * sysctl_overcommit_ratio / 100;
152 * Leave the last 3% for root
155 allowed -= allowed / 32;
156 allowed += total_swap_pages;
158 /* Don't let a single process grow too big:
159 leave 3% of the size of this process for other processes */
160 allowed -= current->mm->total_vm / 32;
163 * cast `allowed' as a signed long because vm_committed_space
164 * sometimes has a negative value
166 if (atomic_read(&vm_committed_space) < (long)allowed)
169 vm_unacct_memory(pages);
174 EXPORT_SYMBOL(__vm_enough_memory);
177 * Requires inode->i_mapping->i_mmap_lock
179 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
180 struct file *file, struct address_space *mapping)
182 if (vma->vm_flags & VM_DENYWRITE)
183 atomic_inc(&file->f_dentry->d_inode->i_writecount);
184 if (vma->vm_flags & VM_SHARED)
185 mapping->i_mmap_writable--;
187 flush_dcache_mmap_lock(mapping);
188 if (unlikely(vma->vm_flags & VM_NONLINEAR))
189 list_del_init(&vma->shared.vm_set.list);
191 vma_prio_tree_remove(vma, &mapping->i_mmap);
192 flush_dcache_mmap_unlock(mapping);
196 * Unlink a file-based vm structure from its prio_tree, to hide
197 * vma from rmap and vmtruncate before freeing its page tables.
199 void unlink_file_vma(struct vm_area_struct *vma)
201 struct file *file = vma->vm_file;
204 struct address_space *mapping = file->f_mapping;
205 spin_lock(&mapping->i_mmap_lock);
206 __remove_shared_vm_struct(vma, file, mapping);
207 spin_unlock(&mapping->i_mmap_lock);
212 * Close a vm structure and free it, returning the next.
214 static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
216 struct vm_area_struct *next = vma->vm_next;
219 if (vma->vm_ops && vma->vm_ops->close)
220 vma->vm_ops->close(vma);
223 mpol_free(vma_policy(vma));
224 kmem_cache_free(vm_area_cachep, vma);
228 asmlinkage unsigned long sys_brk(unsigned long brk)
230 unsigned long rlim, retval;
231 unsigned long newbrk, oldbrk;
232 struct mm_struct *mm = current->mm;
234 down_write(&mm->mmap_sem);
236 if (brk < mm->end_code)
240 * Check against rlimit here. If this check is done later after the test
241 * of oldbrk with newbrk then it can escape the test and let the data
242 * segment grow beyond its set limit the in case where the limit is
243 * not page aligned -Ram Gupta
245 rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
246 if (rlim < RLIM_INFINITY && brk - mm->start_data > rlim)
249 newbrk = PAGE_ALIGN(brk);
250 oldbrk = PAGE_ALIGN(mm->brk);
251 if (oldbrk == newbrk)
254 /* Always allow shrinking brk. */
255 if (brk <= mm->brk) {
256 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
261 /* Check against existing mmap mappings. */
262 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
265 /* Ok, looks good - let it rip. */
266 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
272 up_write(&mm->mmap_sem);
277 static int browse_rb(struct rb_root *root)
280 struct rb_node *nd, *pn = NULL;
281 unsigned long prev = 0, pend = 0;
283 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
284 struct vm_area_struct *vma;
285 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
286 if (vma->vm_start < prev)
287 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
288 if (vma->vm_start < pend)
289 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
290 if (vma->vm_start > vma->vm_end)
291 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
296 for (nd = pn; nd; nd = rb_prev(nd)) {
300 printk("backwards %d, forwards %d\n", j, i), i = 0;
304 void validate_mm(struct mm_struct *mm)
308 struct vm_area_struct *tmp = mm->mmap;
313 if (i != mm->map_count)
314 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
315 i = browse_rb(&mm->mm_rb);
316 if (i != mm->map_count)
317 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
321 #define validate_mm(mm) do { } while (0)
324 static struct vm_area_struct *
325 find_vma_prepare(struct mm_struct *mm, unsigned long addr,
326 struct vm_area_struct **pprev, struct rb_node ***rb_link,
327 struct rb_node ** rb_parent)
329 struct vm_area_struct * vma;
330 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
332 __rb_link = &mm->mm_rb.rb_node;
333 rb_prev = __rb_parent = NULL;
337 struct vm_area_struct *vma_tmp;
339 __rb_parent = *__rb_link;
340 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
342 if (vma_tmp->vm_end > addr) {
344 if (vma_tmp->vm_start <= addr)
346 __rb_link = &__rb_parent->rb_left;
348 rb_prev = __rb_parent;
349 __rb_link = &__rb_parent->rb_right;
355 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
356 *rb_link = __rb_link;
357 *rb_parent = __rb_parent;
362 __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
363 struct vm_area_struct *prev, struct rb_node *rb_parent)
365 if (vma->vm_flags & VM_EXEC)
366 arch_add_exec_range(mm, vma->vm_end);
368 vma->vm_next = prev->vm_next;
373 vma->vm_next = rb_entry(rb_parent,
374 struct vm_area_struct, vm_rb);
380 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
381 struct rb_node **rb_link, struct rb_node *rb_parent)
383 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
384 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
387 static inline void __vma_link_file(struct vm_area_struct *vma)
393 struct address_space *mapping = file->f_mapping;
395 if (vma->vm_flags & VM_DENYWRITE)
396 atomic_dec(&file->f_dentry->d_inode->i_writecount);
397 if (vma->vm_flags & VM_SHARED)
398 mapping->i_mmap_writable++;
400 flush_dcache_mmap_lock(mapping);
401 if (unlikely(vma->vm_flags & VM_NONLINEAR))
402 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
404 vma_prio_tree_insert(vma, &mapping->i_mmap);
405 flush_dcache_mmap_unlock(mapping);
410 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
411 struct vm_area_struct *prev, struct rb_node **rb_link,
412 struct rb_node *rb_parent)
414 __vma_link_list(mm, vma, prev, rb_parent);
415 __vma_link_rb(mm, vma, rb_link, rb_parent);
416 __anon_vma_link(vma);
419 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
420 struct vm_area_struct *prev, struct rb_node **rb_link,
421 struct rb_node *rb_parent)
423 struct address_space *mapping = NULL;
426 mapping = vma->vm_file->f_mapping;
429 spin_lock(&mapping->i_mmap_lock);
430 vma->vm_truncate_count = mapping->truncate_count;
434 __vma_link(mm, vma, prev, rb_link, rb_parent);
435 __vma_link_file(vma);
437 anon_vma_unlock(vma);
439 spin_unlock(&mapping->i_mmap_lock);
446 * Helper for vma_adjust in the split_vma insert case:
447 * insert vm structure into list and rbtree and anon_vma,
448 * but it has already been inserted into prio_tree earlier.
451 __insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
453 struct vm_area_struct * __vma, * prev;
454 struct rb_node ** rb_link, * rb_parent;
456 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
457 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
458 __vma_link(mm, vma, prev, rb_link, rb_parent);
463 __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
464 struct vm_area_struct *prev)
466 prev->vm_next = vma->vm_next;
467 rb_erase(&vma->vm_rb, &mm->mm_rb);
468 if (mm->mmap_cache == vma)
469 mm->mmap_cache = prev;
470 if (vma->vm_flags & VM_EXEC)
471 arch_remove_exec_range(mm, vma->vm_end);
475 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
476 * is already present in an i_mmap tree without adjusting the tree.
477 * The following helper function should be used when such adjustments
478 * are necessary. The "insert" vma (if any) is to be inserted
479 * before we drop the necessary locks.
481 void vma_adjust(struct vm_area_struct *vma, unsigned long start,
482 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
484 struct mm_struct *mm = vma->vm_mm;
485 struct vm_area_struct *next = vma->vm_next;
486 struct vm_area_struct *importer = NULL;
487 struct address_space *mapping = NULL;
488 struct prio_tree_root *root = NULL;
489 struct file *file = vma->vm_file;
490 struct anon_vma *anon_vma = NULL;
491 long adjust_next = 0;
494 if (next && !insert) {
495 if (end >= next->vm_end) {
497 * vma expands, overlapping all the next, and
498 * perhaps the one after too (mprotect case 6).
500 again: remove_next = 1 + (end > next->vm_end);
502 anon_vma = next->anon_vma;
504 } else if (end > next->vm_start) {
506 * vma expands, overlapping part of the next:
507 * mprotect case 5 shifting the boundary up.
509 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
510 anon_vma = next->anon_vma;
512 } else if (end < vma->vm_end) {
514 * vma shrinks, and !insert tells it's not
515 * split_vma inserting another: so it must be
516 * mprotect case 4 shifting the boundary down.
518 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
519 anon_vma = next->anon_vma;
525 mapping = file->f_mapping;
526 if (!(vma->vm_flags & VM_NONLINEAR))
527 root = &mapping->i_mmap;
528 spin_lock(&mapping->i_mmap_lock);
530 vma->vm_truncate_count != next->vm_truncate_count) {
532 * unmap_mapping_range might be in progress:
533 * ensure that the expanding vma is rescanned.
535 importer->vm_truncate_count = 0;
538 insert->vm_truncate_count = vma->vm_truncate_count;
540 * Put into prio_tree now, so instantiated pages
541 * are visible to arm/parisc __flush_dcache_page
542 * throughout; but we cannot insert into address
543 * space until vma start or end is updated.
545 __vma_link_file(insert);
550 * When changing only vma->vm_end, we don't really need
551 * anon_vma lock: but is that case worth optimizing out?
554 anon_vma = vma->anon_vma;
556 spin_lock(&anon_vma->lock);
558 * Easily overlooked: when mprotect shifts the boundary,
559 * make sure the expanding vma has anon_vma set if the
560 * shrinking vma had, to cover any anon pages imported.
562 if (importer && !importer->anon_vma) {
563 importer->anon_vma = anon_vma;
564 __anon_vma_link(importer);
569 flush_dcache_mmap_lock(mapping);
570 vma_prio_tree_remove(vma, root);
572 vma_prio_tree_remove(next, root);
575 vma->vm_start = start;
577 vma->vm_pgoff = pgoff;
579 next->vm_start += adjust_next << PAGE_SHIFT;
580 next->vm_pgoff += adjust_next;
585 vma_prio_tree_insert(next, root);
586 vma_prio_tree_insert(vma, root);
587 flush_dcache_mmap_unlock(mapping);
592 * vma_merge has merged next into vma, and needs
593 * us to remove next before dropping the locks.
595 __vma_unlink(mm, next, vma);
597 __remove_shared_vm_struct(next, file, mapping);
599 __anon_vma_merge(vma, next);
602 * split_vma has split insert from vma, and needs
603 * us to insert it before dropping the locks
604 * (it may either follow vma or precede it).
606 __insert_vm_struct(mm, insert);
610 spin_unlock(&anon_vma->lock);
612 spin_unlock(&mapping->i_mmap_lock);
618 mpol_free(vma_policy(next));
619 kmem_cache_free(vm_area_cachep, next);
621 * In mprotect's case 6 (see comments on vma_merge),
622 * we must remove another next too. It would clutter
623 * up the code too much to do both in one go.
625 if (remove_next == 2) {
635 * If the vma has a ->close operation then the driver probably needs to release
636 * per-vma resources, so we don't attempt to merge those.
638 #define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_RESERVED | VM_PFNMAP)
640 static inline int is_mergeable_vma(struct vm_area_struct *vma,
641 struct file *file, unsigned long vm_flags)
643 if (vma->vm_flags != vm_flags)
645 if (vma->vm_file != file)
647 if (vma->vm_ops && vma->vm_ops->close)
652 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
653 struct anon_vma *anon_vma2)
655 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
659 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
660 * in front of (at a lower virtual address and file offset than) the vma.
662 * We cannot merge two vmas if they have differently assigned (non-NULL)
663 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
665 * We don't check here for the merged mmap wrapping around the end of pagecache
666 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
667 * wrap, nor mmaps which cover the final page at index -1UL.
670 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
671 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
673 if (is_mergeable_vma(vma, file, vm_flags) &&
674 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
675 if (vma->vm_pgoff == vm_pgoff)
682 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
683 * beyond (at a higher virtual address and file offset than) the vma.
685 * We cannot merge two vmas if they have differently assigned (non-NULL)
686 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
689 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
690 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
692 if (is_mergeable_vma(vma, file, vm_flags) &&
693 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
695 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
696 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
703 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
704 * whether that can be merged with its predecessor or its successor.
705 * Or both (it neatly fills a hole).
707 * In most cases - when called for mmap, brk or mremap - [addr,end) is
708 * certain not to be mapped by the time vma_merge is called; but when
709 * called for mprotect, it is certain to be already mapped (either at
710 * an offset within prev, or at the start of next), and the flags of
711 * this area are about to be changed to vm_flags - and the no-change
712 * case has already been eliminated.
714 * The following mprotect cases have to be considered, where AAAA is
715 * the area passed down from mprotect_fixup, never extending beyond one
716 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
718 * AAAA AAAA AAAA AAAA
719 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
720 * cannot merge might become might become might become
721 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
722 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
723 * mremap move: PPPPNNNNNNNN 8
725 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
726 * might become case 1 below case 2 below case 3 below
728 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
729 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
731 struct vm_area_struct *vma_merge(struct mm_struct *mm,
732 struct vm_area_struct *prev, unsigned long addr,
733 unsigned long end, unsigned long vm_flags,
734 struct anon_vma *anon_vma, struct file *file,
735 pgoff_t pgoff, struct mempolicy *policy)
737 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
738 struct vm_area_struct *area, *next;
741 * We later require that vma->vm_flags == vm_flags,
742 * so this tests vma->vm_flags & VM_SPECIAL, too.
744 if (vm_flags & VM_SPECIAL)
748 next = prev->vm_next;
752 if (next && next->vm_end == end) /* cases 6, 7, 8 */
753 next = next->vm_next;
756 * Can it merge with the predecessor?
758 if (prev && prev->vm_end == addr &&
759 mpol_equal(vma_policy(prev), policy) &&
760 can_vma_merge_after(prev, vm_flags,
761 anon_vma, file, pgoff)) {
763 * OK, it can. Can we now merge in the successor as well?
765 if (next && end == next->vm_start &&
766 mpol_equal(policy, vma_policy(next)) &&
767 can_vma_merge_before(next, vm_flags,
768 anon_vma, file, pgoff+pglen) &&
769 is_mergeable_anon_vma(prev->anon_vma,
772 vma_adjust(prev, prev->vm_start,
773 next->vm_end, prev->vm_pgoff, NULL);
774 } else /* cases 2, 5, 7 */
775 vma_adjust(prev, prev->vm_start,
776 end, prev->vm_pgoff, NULL);
777 if (prev->vm_flags & VM_EXEC)
778 arch_add_exec_range(mm, prev->vm_end);
783 * Can this new request be merged in front of next?
785 if (next && end == next->vm_start &&
786 mpol_equal(policy, vma_policy(next)) &&
787 can_vma_merge_before(next, vm_flags,
788 anon_vma, file, pgoff+pglen)) {
789 if (prev && addr < prev->vm_end) /* case 4 */
790 vma_adjust(prev, prev->vm_start,
791 addr, prev->vm_pgoff, NULL);
792 else /* cases 3, 8 */
793 vma_adjust(area, addr, next->vm_end,
794 next->vm_pgoff - pglen, NULL);
802 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
803 * neighbouring vmas for a suitable anon_vma, before it goes off
804 * to allocate a new anon_vma. It checks because a repetitive
805 * sequence of mprotects and faults may otherwise lead to distinct
806 * anon_vmas being allocated, preventing vma merge in subsequent
809 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
811 struct vm_area_struct *near;
812 unsigned long vm_flags;
819 * Since only mprotect tries to remerge vmas, match flags
820 * which might be mprotected into each other later on.
821 * Neither mlock nor madvise tries to remerge at present,
822 * so leave their flags as obstructing a merge.
824 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
825 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
827 if (near->anon_vma && vma->vm_end == near->vm_start &&
828 mpol_equal(vma_policy(vma), vma_policy(near)) &&
829 can_vma_merge_before(near, vm_flags,
830 NULL, vma->vm_file, vma->vm_pgoff +
831 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
832 return near->anon_vma;
835 * It is potentially slow to have to call find_vma_prev here.
836 * But it's only on the first write fault on the vma, not
837 * every time, and we could devise a way to avoid it later
838 * (e.g. stash info in next's anon_vma_node when assigning
839 * an anon_vma, or when trying vma_merge). Another time.
841 BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma);
845 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
846 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
848 if (near->anon_vma && near->vm_end == vma->vm_start &&
849 mpol_equal(vma_policy(near), vma_policy(vma)) &&
850 can_vma_merge_after(near, vm_flags,
851 NULL, vma->vm_file, vma->vm_pgoff))
852 return near->anon_vma;
855 * There's no absolute need to look only at touching neighbours:
856 * we could search further afield for "compatible" anon_vmas.
857 * But it would probably just be a waste of time searching,
858 * or lead to too many vmas hanging off the same anon_vma.
859 * We're trying to allow mprotect remerging later on,
860 * not trying to minimize memory used for anon_vmas.
865 #ifdef CONFIG_PROC_FS
866 void vm_stat_account(struct mm_struct *mm, unsigned long flags,
867 struct file *file, long pages)
869 const unsigned long stack_flags
870 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
873 mm->shared_vm += pages;
874 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
875 mm->exec_vm += pages;
876 } else if (flags & stack_flags)
877 mm->stack_vm += pages;
878 if (flags & (VM_RESERVED|VM_IO))
879 mm->reserved_vm += pages;
881 #endif /* CONFIG_PROC_FS */
884 * The caller must hold down_write(current->mm->mmap_sem).
887 unsigned long do_mmap_pgoff(struct file * file, unsigned long addr,
888 unsigned long len, unsigned long prot,
889 unsigned long flags, unsigned long pgoff)
891 struct mm_struct * mm = current->mm;
892 struct vm_area_struct * vma, * prev;
894 unsigned int vm_flags;
895 int correct_wcount = 0;
897 struct rb_node ** rb_link, * rb_parent;
899 unsigned long charged = 0, reqprot = prot;
902 if (is_file_hugepages(file))
905 if (!file->f_op || !file->f_op->mmap)
908 if ((prot & PROT_EXEC) &&
909 (file->f_vfsmnt->mnt_flags & MNT_NOEXEC))
913 * Does the application expect PROT_READ to imply PROT_EXEC?
915 * (the exception is when the underlying filesystem is noexec
916 * mounted, in which case we dont add PROT_EXEC.)
918 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
919 if (!(file && (file->f_vfsmnt->mnt_flags & MNT_NOEXEC)))
925 error = arch_mmap_check(addr, len, flags);
929 /* Careful about overflows.. */
930 len = PAGE_ALIGN(len);
931 if (!len || len > TASK_SIZE)
934 /* offset overflow? */
935 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
938 /* Too many mappings? */
939 if (mm->map_count > sysctl_max_map_count)
942 /* Obtain the address to map to. we verify (or select) it and ensure
943 * that it represents a valid section of the address space.
945 addr = get_unmapped_area_prot(file, addr, len, pgoff, flags, prot & PROT_EXEC);
946 if (addr & ~PAGE_MASK)
949 /* Do simple checking here so the lower-level routines won't have
950 * to. we assume access permissions have been handled by the open
951 * of the memory object, so we don't do any here.
953 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
954 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
956 if (flags & MAP_LOCKED) {
959 vm_flags |= VM_LOCKED;
961 /* mlock MCL_FUTURE? */
962 if (vm_flags & VM_LOCKED) {
963 unsigned long locked, lock_limit;
964 locked = len >> PAGE_SHIFT;
965 locked += mm->locked_vm;
966 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
967 lock_limit >>= PAGE_SHIFT;
968 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
972 inode = file ? file->f_dentry->d_inode : NULL;
975 switch (flags & MAP_TYPE) {
977 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
981 * Make sure we don't allow writing to an append-only
984 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
988 * Make sure there are no mandatory locks on the file.
990 if (locks_verify_locked(inode))
993 vm_flags |= VM_SHARED | VM_MAYSHARE;
994 if (!(file->f_mode & FMODE_WRITE))
995 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
999 if (!(file->f_mode & FMODE_READ))
1007 switch (flags & MAP_TYPE) {
1009 vm_flags |= VM_SHARED | VM_MAYSHARE;
1013 * Set pgoff according to addr for anon_vma.
1015 pgoff = addr >> PAGE_SHIFT;
1022 error = security_file_mmap(file, reqprot, prot, flags);
1026 /* Clear old maps */
1029 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1030 if (vma && vma->vm_start < addr + len) {
1031 if (do_munmap(mm, addr, len))
1036 /* Check against address space limit. */
1037 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1040 if (accountable && (!(flags & MAP_NORESERVE) ||
1041 sysctl_overcommit_memory == OVERCOMMIT_NEVER)) {
1042 if (vm_flags & VM_SHARED) {
1043 /* Check memory availability in shmem_file_setup? */
1044 vm_flags |= VM_ACCOUNT;
1045 } else if (vm_flags & VM_WRITE) {
1047 * Private writable mapping: check memory availability
1049 charged = len >> PAGE_SHIFT;
1050 if (security_vm_enough_memory(charged))
1052 vm_flags |= VM_ACCOUNT;
1057 * Can we just expand an old private anonymous mapping?
1058 * The VM_SHARED test is necessary because shmem_zero_setup
1059 * will create the file object for a shared anonymous map below.
1061 if (!file && !(vm_flags & VM_SHARED) &&
1062 vma_merge(mm, prev, addr, addr + len, vm_flags,
1063 NULL, NULL, pgoff, NULL))
1067 * Determine the object being mapped and call the appropriate
1068 * specific mapper. the address has already been validated, but
1069 * not unmapped, but the maps are removed from the list.
1071 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1078 vma->vm_start = addr;
1079 vma->vm_end = addr + len;
1080 vma->vm_flags = vm_flags;
1081 vma->vm_page_prot = protection_map[vm_flags & 0x0f];
1082 vma->vm_pgoff = pgoff;
1086 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1088 if (vm_flags & VM_DENYWRITE) {
1089 error = deny_write_access(file);
1094 vma->vm_file = file;
1096 error = file->f_op->mmap(file, vma);
1098 goto unmap_and_free_vma;
1099 } else if (vm_flags & VM_SHARED) {
1100 error = shmem_zero_setup(vma);
1105 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
1106 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
1107 * that memory reservation must be checked; but that reservation
1108 * belongs to shared memory object, not to vma: so now clear it.
1110 if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT))
1111 vma->vm_flags &= ~VM_ACCOUNT;
1113 /* Can addr have changed??
1115 * Answer: Yes, several device drivers can do it in their
1116 * f_op->mmap method. -DaveM
1118 addr = vma->vm_start;
1119 pgoff = vma->vm_pgoff;
1120 vm_flags = vma->vm_flags;
1122 if (!file || !vma_merge(mm, prev, addr, vma->vm_end,
1123 vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
1124 file = vma->vm_file;
1125 vma_link(mm, vma, prev, rb_link, rb_parent);
1127 atomic_inc(&inode->i_writecount);
1131 atomic_inc(&inode->i_writecount);
1134 mpol_free(vma_policy(vma));
1135 kmem_cache_free(vm_area_cachep, vma);
1138 vx_vmpages_add(mm, len >> PAGE_SHIFT);
1139 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1140 if (vm_flags & VM_LOCKED) {
1141 vx_vmlocked_add(mm, len >> PAGE_SHIFT);
1142 make_pages_present(addr, addr + len);
1144 if (flags & MAP_POPULATE) {
1145 up_write(&mm->mmap_sem);
1146 sys_remap_file_pages(addr, len, 0,
1147 pgoff, flags & MAP_NONBLOCK);
1148 down_write(&mm->mmap_sem);
1154 atomic_inc(&inode->i_writecount);
1155 vma->vm_file = NULL;
1158 /* Undo any partial mapping done by a device driver. */
1159 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1162 kmem_cache_free(vm_area_cachep, vma);
1165 vm_unacct_memory(charged);
1169 EXPORT_SYMBOL(do_mmap_pgoff);
1171 /* Get an address range which is currently unmapped.
1172 * For shmat() with addr=0.
1174 * Ugly calling convention alert:
1175 * Return value with the low bits set means error value,
1177 * if (ret & ~PAGE_MASK)
1180 * This function "knows" that -ENOMEM has the bits set.
1182 #ifndef HAVE_ARCH_UNMAPPED_AREA
1184 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1185 unsigned long len, unsigned long pgoff, unsigned long flags)
1187 struct mm_struct *mm = current->mm;
1188 struct vm_area_struct *vma;
1189 unsigned long start_addr;
1191 if (len > TASK_SIZE)
1195 addr = PAGE_ALIGN(addr);
1196 vma = find_vma(mm, addr);
1197 if (TASK_SIZE - len >= addr &&
1198 (!vma || addr + len <= vma->vm_start))
1201 if (len > mm->cached_hole_size) {
1202 start_addr = addr = mm->free_area_cache;
1204 start_addr = addr = TASK_UNMAPPED_BASE;
1205 mm->cached_hole_size = 0;
1209 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1210 /* At this point: (!vma || addr < vma->vm_end). */
1211 if (TASK_SIZE - len < addr) {
1213 * Start a new search - just in case we missed
1216 if (start_addr != TASK_UNMAPPED_BASE) {
1217 addr = TASK_UNMAPPED_BASE;
1219 mm->cached_hole_size = 0;
1224 if (!vma || addr + len <= vma->vm_start) {
1226 * Remember the place where we stopped the search:
1228 mm->free_area_cache = addr + len;
1231 if (addr + mm->cached_hole_size < vma->vm_start)
1232 mm->cached_hole_size = vma->vm_start - addr;
1238 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1241 * Is this a new hole at the lowest possible address?
1243 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1244 mm->free_area_cache = addr;
1245 mm->cached_hole_size = ~0UL;
1250 * This mmap-allocator allocates new areas top-down from below the
1251 * stack's low limit (the base):
1253 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1255 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1256 const unsigned long len, const unsigned long pgoff,
1257 const unsigned long flags)
1259 struct vm_area_struct *vma;
1260 struct mm_struct *mm = current->mm;
1261 unsigned long addr = addr0;
1263 /* requested length too big for entire address space */
1264 if (len > TASK_SIZE)
1267 /* requesting a specific address */
1269 addr = PAGE_ALIGN(addr);
1270 vma = find_vma(mm, addr);
1271 if (TASK_SIZE - len >= addr &&
1272 (!vma || addr + len <= vma->vm_start))
1276 /* check if free_area_cache is useful for us */
1277 if (len <= mm->cached_hole_size) {
1278 mm->cached_hole_size = 0;
1279 mm->free_area_cache = mm->mmap_base;
1282 /* either no address requested or can't fit in requested address hole */
1283 addr = mm->free_area_cache;
1285 /* make sure it can fit in the remaining address space */
1287 vma = find_vma(mm, addr-len);
1288 if (!vma || addr <= vma->vm_start)
1289 /* remember the address as a hint for next time */
1290 return (mm->free_area_cache = addr-len);
1293 if (mm->mmap_base < len)
1296 addr = mm->mmap_base-len;
1300 * Lookup failure means no vma is above this address,
1301 * else if new region fits below vma->vm_start,
1302 * return with success:
1304 vma = find_vma(mm, addr);
1305 if (!vma || addr+len <= vma->vm_start)
1306 /* remember the address as a hint for next time */
1307 return (mm->free_area_cache = addr);
1309 /* remember the largest hole we saw so far */
1310 if (addr + mm->cached_hole_size < vma->vm_start)
1311 mm->cached_hole_size = vma->vm_start - addr;
1313 /* try just below the current vma->vm_start */
1314 addr = vma->vm_start-len;
1315 } while (len < vma->vm_start);
1319 * A failed mmap() very likely causes application failure,
1320 * so fall back to the bottom-up function here. This scenario
1321 * can happen with large stack limits and large mmap()
1324 mm->cached_hole_size = ~0UL;
1325 mm->free_area_cache = TASK_UNMAPPED_BASE;
1326 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1328 * Restore the topdown base:
1330 mm->free_area_cache = mm->mmap_base;
1331 mm->cached_hole_size = ~0UL;
1337 void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
1340 * Is this a new hole at the highest possible address?
1342 if (addr > mm->free_area_cache)
1343 mm->free_area_cache = addr;
1345 /* dont allow allocations above current base */
1346 if (mm->free_area_cache > mm->mmap_base)
1347 mm->free_area_cache = mm->mmap_base;
1352 get_unmapped_area_prot(struct file *file, unsigned long addr, unsigned long len,
1353 unsigned long pgoff, unsigned long flags, int exec)
1357 if (!(flags & MAP_FIXED)) {
1358 unsigned long (*get_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
1360 if (exec && current->mm->get_unmapped_exec_area)
1361 get_area = current->mm->get_unmapped_exec_area;
1363 get_area = current->mm->get_unmapped_area;
1365 if (file && file->f_op && file->f_op->get_unmapped_area)
1366 get_area = file->f_op->get_unmapped_area;
1367 addr = get_area(file, addr, len, pgoff, flags);
1368 if (IS_ERR_VALUE(addr))
1372 if (addr > TASK_SIZE - len)
1374 if (addr & ~PAGE_MASK)
1376 if (file && is_file_hugepages(file)) {
1378 * Check if the given range is hugepage aligned, and
1379 * can be made suitable for hugepages.
1381 ret = prepare_hugepage_range(addr, len);
1384 * Ensure that a normal request is not falling in a
1385 * reserved hugepage range. For some archs like IA-64,
1386 * there is a separate region for hugepages.
1388 ret = is_hugepage_only_range(current->mm, addr, len);
1395 EXPORT_SYMBOL(get_unmapped_area_prot);
1397 #define SHLIB_BASE 0x00111000
1399 unsigned long arch_get_unmapped_exec_area(struct file *filp, unsigned long addr0,
1400 unsigned long len0, unsigned long pgoff, unsigned long flags)
1402 unsigned long addr = addr0, len = len0;
1403 struct mm_struct *mm = current->mm;
1404 struct vm_area_struct *vma;
1407 if (len > TASK_SIZE)
1410 if (!addr && !(flags & MAP_FIXED))
1411 addr = randomize_range(SHLIB_BASE, 0x01000000, len);
1414 addr = PAGE_ALIGN(addr);
1415 vma = find_vma(mm, addr);
1416 if (TASK_SIZE - len >= addr &&
1417 (!vma || addr + len <= vma->vm_start)) {
1423 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1424 /* At this point: (!vma || addr < vma->vm_end). */
1425 if (TASK_SIZE - len < addr)
1428 if (!vma || addr + len <= vma->vm_start) {
1430 * Must not let a PROT_EXEC mapping get into the
1433 if (addr + len > mm->brk)
1437 * Up until the brk area we randomize addresses
1438 * as much as possible:
1440 if (addr >= 0x01000000) {
1441 tmp = randomize_range(0x01000000, PAGE_ALIGN(max(mm->start_brk, (unsigned long)0x08000000)), len);
1442 vma = find_vma(mm, tmp);
1443 if (TASK_SIZE - len >= tmp &&
1444 (!vma || tmp + len <= vma->vm_start))
1448 * Ok, randomization didnt work out - return
1449 * the result of the linear search:
1457 return current->mm->get_unmapped_area(filp, addr0, len0, pgoff, flags);
1461 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1462 struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
1464 struct vm_area_struct *vma = NULL;
1467 /* Check the cache first. */
1468 /* (Cache hit rate is typically around 35%.) */
1469 vma = mm->mmap_cache;
1470 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1471 struct rb_node * rb_node;
1473 rb_node = mm->mm_rb.rb_node;
1477 struct vm_area_struct * vma_tmp;
1479 vma_tmp = rb_entry(rb_node,
1480 struct vm_area_struct, vm_rb);
1482 if (vma_tmp->vm_end > addr) {
1484 if (vma_tmp->vm_start <= addr)
1486 rb_node = rb_node->rb_left;
1488 rb_node = rb_node->rb_right;
1491 mm->mmap_cache = vma;
1497 EXPORT_SYMBOL(find_vma);
1499 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1500 struct vm_area_struct *
1501 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1502 struct vm_area_struct **pprev)
1504 struct vm_area_struct *vma = NULL, *prev = NULL;
1505 struct rb_node * rb_node;
1509 /* Guard against addr being lower than the first VMA */
1512 /* Go through the RB tree quickly. */
1513 rb_node = mm->mm_rb.rb_node;
1516 struct vm_area_struct *vma_tmp;
1517 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1519 if (addr < vma_tmp->vm_end) {
1520 rb_node = rb_node->rb_left;
1523 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1525 rb_node = rb_node->rb_right;
1531 return prev ? prev->vm_next : vma;
1534 static int over_stack_limit(unsigned long sz)
1536 if (sz < EXEC_STACK_BIAS)
1538 return (sz - EXEC_STACK_BIAS) >
1539 current->signal->rlim[RLIMIT_STACK].rlim_cur;
1543 * Verify that the stack growth is acceptable and
1544 * update accounting. This is shared with both the
1545 * grow-up and grow-down cases.
1547 static int acct_stack_growth(struct vm_area_struct * vma, unsigned long size, unsigned long grow)
1549 struct mm_struct *mm = vma->vm_mm;
1550 struct rlimit *rlim = current->signal->rlim;
1552 /* address space limit tests */
1553 if (!may_expand_vm(mm, grow))
1556 /* Stack limit test */
1557 if (over_stack_limit(size))
1560 /* mlock limit tests */
1561 if (vma->vm_flags & VM_LOCKED) {
1562 unsigned long locked;
1563 unsigned long limit;
1564 locked = mm->locked_vm + grow;
1565 limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
1566 if (locked > limit && !capable(CAP_IPC_LOCK))
1571 * Overcommit.. This must be the final test, as it will
1572 * update security statistics.
1574 if (security_vm_enough_memory(grow))
1577 /* Ok, everything looks good - let it rip */
1578 vx_vmpages_add(mm, grow);
1579 if (vma->vm_flags & VM_LOCKED)
1580 vx_vmlocked_add(mm, grow);
1581 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1585 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1587 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1588 * vma is the last one with address > vma->vm_end. Have to extend vma.
1593 int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1597 if (!(vma->vm_flags & VM_GROWSUP))
1601 * We must make sure the anon_vma is allocated
1602 * so that the anon_vma locking is not a noop.
1604 if (unlikely(anon_vma_prepare(vma)))
1609 * vma->vm_start/vm_end cannot change under us because the caller
1610 * is required to hold the mmap_sem in read mode. We need the
1611 * anon_vma lock to serialize against concurrent expand_stacks.
1613 address += 4 + PAGE_SIZE - 1;
1614 address &= PAGE_MASK;
1617 /* Somebody else might have raced and expanded it already */
1618 if (address > vma->vm_end) {
1619 unsigned long size, grow;
1621 size = address - vma->vm_start;
1622 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1624 error = acct_stack_growth(vma, size, grow);
1626 vma->vm_end = address;
1628 anon_vma_unlock(vma);
1631 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1633 #ifdef CONFIG_STACK_GROWSUP
1634 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1636 return expand_upwards(vma, address);
1639 struct vm_area_struct *
1640 find_extend_vma(struct mm_struct *mm, unsigned long addr)
1642 struct vm_area_struct *vma, *prev;
1645 vma = find_vma_prev(mm, addr, &prev);
1646 if (vma && (vma->vm_start <= addr))
1648 if (!prev || expand_stack(prev, addr))
1650 if (prev->vm_flags & VM_LOCKED) {
1651 make_pages_present(addr, prev->vm_end);
1657 * vma is the first one with address < vma->vm_start. Have to extend vma.
1659 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1664 * We must make sure the anon_vma is allocated
1665 * so that the anon_vma locking is not a noop.
1667 if (unlikely(anon_vma_prepare(vma)))
1672 * vma->vm_start/vm_end cannot change under us because the caller
1673 * is required to hold the mmap_sem in read mode. We need the
1674 * anon_vma lock to serialize against concurrent expand_stacks.
1676 address &= PAGE_MASK;
1679 /* Somebody else might have raced and expanded it already */
1680 if (address < vma->vm_start) {
1681 unsigned long size, grow;
1683 size = vma->vm_end - address;
1684 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1686 error = acct_stack_growth(vma, size, grow);
1688 vma->vm_start = address;
1689 vma->vm_pgoff -= grow;
1692 anon_vma_unlock(vma);
1696 struct vm_area_struct *
1697 find_extend_vma(struct mm_struct * mm, unsigned long addr)
1699 struct vm_area_struct * vma;
1700 unsigned long start;
1703 vma = find_vma(mm,addr);
1706 if (vma->vm_start <= addr)
1708 if (!(vma->vm_flags & VM_GROWSDOWN))
1710 start = vma->vm_start;
1711 if (expand_stack(vma, addr))
1713 if (vma->vm_flags & VM_LOCKED) {
1714 make_pages_present(addr, start);
1721 * Ok - we have the memory areas we should free on the vma list,
1722 * so release them, and do the vma updates.
1724 * Called with the mm semaphore held.
1726 static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
1728 /* Update high watermark before we lower total_vm */
1729 update_hiwater_vm(mm);
1731 long nrpages = vma_pages(vma);
1733 vx_vmpages_sub(mm, nrpages);
1734 if (vma->vm_flags & VM_LOCKED)
1735 vx_vmlocked_sub(mm, nrpages);
1736 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
1737 vma = remove_vma(vma);
1743 * Get rid of page table information in the indicated region.
1745 * Called with the mm semaphore held.
1747 static void unmap_region(struct mm_struct *mm,
1748 struct vm_area_struct *vma, struct vm_area_struct *prev,
1749 unsigned long start, unsigned long end)
1751 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
1752 struct mmu_gather *tlb;
1753 unsigned long nr_accounted = 0;
1756 tlb = tlb_gather_mmu(mm, 0);
1757 update_hiwater_rss(mm);
1758 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
1759 vm_unacct_memory(nr_accounted);
1760 free_pgtables(&tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
1761 next? next->vm_start: 0);
1762 tlb_finish_mmu(tlb, start, end);
1766 * Create a list of vma's touched by the unmap, removing them from the mm's
1767 * vma list as we go..
1770 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1771 struct vm_area_struct *prev, unsigned long end)
1773 struct vm_area_struct **insertion_point;
1774 struct vm_area_struct *tail_vma = NULL;
1777 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1779 rb_erase(&vma->vm_rb, &mm->mm_rb);
1783 } while (vma && vma->vm_start < end);
1784 *insertion_point = vma;
1785 tail_vma->vm_next = NULL;
1786 if (mm->unmap_area == arch_unmap_area)
1787 addr = prev ? prev->vm_end : mm->mmap_base;
1789 addr = vma ? vma->vm_start : mm->mmap_base;
1790 mm->unmap_area(mm, addr);
1791 mm->mmap_cache = NULL; /* Kill the cache. */
1795 * Split a vma into two pieces at address 'addr', a new vma is allocated
1796 * either for the first part or the the tail.
1798 int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1799 unsigned long addr, int new_below)
1801 struct mempolicy *pol;
1802 struct vm_area_struct *new;
1804 if (is_vm_hugetlb_page(vma) && (addr & ~HPAGE_MASK))
1807 if (mm->map_count >= sysctl_max_map_count)
1810 new = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
1814 /* most fields are the same, copy all, and then fixup */
1820 new->vm_start = addr;
1821 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1824 pol = mpol_copy(vma_policy(vma));
1826 kmem_cache_free(vm_area_cachep, new);
1827 return PTR_ERR(pol);
1829 vma_set_policy(new, pol);
1832 get_file(new->vm_file);
1834 if (new->vm_ops && new->vm_ops->open)
1835 new->vm_ops->open(new);
1838 unsigned long old_end = vma->vm_end;
1840 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1841 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1842 if (vma->vm_flags & VM_EXEC)
1843 arch_remove_exec_range(mm, old_end);
1845 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1850 /* Munmap is split into 2 main parts -- this part which finds
1851 * what needs doing, and the areas themselves, which do the
1852 * work. This now handles partial unmappings.
1853 * Jeremy Fitzhardinge <jeremy@goop.org>
1855 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1858 struct vm_area_struct *vma, *prev, *last;
1860 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1863 if ((len = PAGE_ALIGN(len)) == 0)
1866 /* Find the first overlapping VMA */
1867 vma = find_vma_prev(mm, start, &prev);
1870 /* we have start < vma->vm_end */
1872 /* if it doesn't overlap, we have nothing.. */
1874 if (vma->vm_start >= end)
1878 * If we need to split any vma, do it now to save pain later.
1880 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1881 * unmapped vm_area_struct will remain in use: so lower split_vma
1882 * places tmp vma above, and higher split_vma places tmp vma below.
1884 if (start > vma->vm_start) {
1885 int error = split_vma(mm, vma, start, 0);
1891 /* Does it split the last one? */
1892 last = find_vma(mm, end);
1893 if (last && end > last->vm_start) {
1894 int error = split_vma(mm, last, end, 1);
1898 vma = prev? prev->vm_next: mm->mmap;
1901 * Remove the vma's, and unmap the actual pages
1903 detach_vmas_to_be_unmapped(mm, vma, prev, end);
1904 unmap_region(mm, vma, prev, start, end);
1906 /* Fix up all other VM information */
1907 remove_vma_list(mm, vma);
1912 EXPORT_SYMBOL(do_munmap);
1914 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1917 struct mm_struct *mm = current->mm;
1919 profile_munmap(addr);
1921 down_write(&mm->mmap_sem);
1922 ret = do_munmap(mm, addr, len);
1923 up_write(&mm->mmap_sem);
1927 static inline void verify_mm_writelocked(struct mm_struct *mm)
1929 #ifdef CONFIG_DEBUG_VM
1930 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
1932 up_read(&mm->mmap_sem);
1938 * this is really a simplified "do_mmap". it only handles
1939 * anonymous maps. eventually we may be able to do some
1940 * brk-specific accounting here.
1942 unsigned long do_brk(unsigned long addr, unsigned long len)
1944 struct mm_struct * mm = current->mm;
1945 struct vm_area_struct * vma, * prev;
1946 unsigned long flags;
1947 struct rb_node ** rb_link, * rb_parent;
1948 pgoff_t pgoff = addr >> PAGE_SHIFT;
1951 len = PAGE_ALIGN(len);
1955 if ((addr + len) > TASK_SIZE || (addr + len) < addr)
1958 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
1960 error = arch_mmap_check(addr, len, flags);
1967 if (mm->def_flags & VM_LOCKED) {
1968 unsigned long locked, lock_limit;
1969 locked = len >> PAGE_SHIFT;
1970 locked += mm->locked_vm;
1971 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
1972 lock_limit >>= PAGE_SHIFT;
1973 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1975 if (!vx_vmlocked_avail(mm, len >> PAGE_SHIFT))
1980 * mm->mmap_sem is required to protect against another thread
1981 * changing the mappings in case we sleep.
1983 verify_mm_writelocked(mm);
1986 * Clear old maps. this also does some error checking for us
1989 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1990 if (vma && vma->vm_start < addr + len) {
1991 if (do_munmap(mm, addr, len))
1996 /* Check against address space limits *after* clearing old maps... */
1997 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
2000 if (mm->map_count > sysctl_max_map_count)
2003 if (security_vm_enough_memory(len >> PAGE_SHIFT) ||
2004 !vx_vmpages_avail(mm, len >> PAGE_SHIFT))
2007 /* Can we just expand an old private anonymous mapping? */
2008 if (vma_merge(mm, prev, addr, addr + len, flags,
2009 NULL, NULL, pgoff, NULL))
2013 * create a vma struct for an anonymous mapping
2015 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2017 vm_unacct_memory(len >> PAGE_SHIFT);
2022 vma->vm_start = addr;
2023 vma->vm_end = addr + len;
2024 vma->vm_pgoff = pgoff;
2025 vma->vm_flags = flags;
2026 vma->vm_page_prot = protection_map[flags & 0x0f];
2027 vma_link(mm, vma, prev, rb_link, rb_parent);
2029 vx_vmpages_add(mm, len >> PAGE_SHIFT);
2030 if (flags & VM_LOCKED) {
2031 vx_vmlocked_add(mm, len >> PAGE_SHIFT);
2032 make_pages_present(addr, addr + len);
2037 EXPORT_SYMBOL(do_brk);
2039 /* Release all mmaps. */
2040 void exit_mmap(struct mm_struct *mm)
2042 struct mmu_gather *tlb;
2043 struct vm_area_struct *vma = mm->mmap;
2044 unsigned long nr_accounted = 0;
2047 #ifdef arch_exit_mmap
2053 tlb = tlb_gather_mmu(mm, 1);
2054 /* Don't update_hiwater_rss(mm) here, do_exit already did */
2055 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2056 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
2057 vm_unacct_memory(nr_accounted);
2058 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
2059 tlb_finish_mmu(tlb, 0, end);
2060 arch_flush_exec_range(mm);
2062 set_mm_counter(mm, file_rss, 0);
2063 set_mm_counter(mm, anon_rss, 0);
2064 vx_vmpages_sub(mm, mm->total_vm);
2065 vx_vmlocked_sub(mm, mm->locked_vm);
2068 * Walk the list again, actually closing and freeing it,
2069 * with preemption enabled, without holding any MM locks.
2072 vma = remove_vma(vma);
2074 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
2077 /* Insert vm structure into process list sorted by address
2078 * and into the inode's i_mmap tree. If vm_file is non-NULL
2079 * then i_mmap_lock is taken here.
2081 int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2083 struct vm_area_struct * __vma, * prev;
2084 struct rb_node ** rb_link, * rb_parent;
2087 * The vm_pgoff of a purely anonymous vma should be irrelevant
2088 * until its first write fault, when page's anon_vma and index
2089 * are set. But now set the vm_pgoff it will almost certainly
2090 * end up with (unless mremap moves it elsewhere before that
2091 * first wfault), so /proc/pid/maps tells a consistent story.
2093 * By setting it to reflect the virtual start address of the
2094 * vma, merges and splits can happen in a seamless way, just
2095 * using the existing file pgoff checks and manipulations.
2096 * Similarly in do_mmap_pgoff and in do_brk.
2098 if (!vma->vm_file) {
2099 BUG_ON(vma->anon_vma);
2100 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2102 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2103 if (__vma && __vma->vm_start < vma->vm_end)
2105 if ((vma->vm_flags & VM_ACCOUNT) &&
2106 (security_vm_enough_memory(vma_pages(vma)) ||
2107 !vx_vmpages_avail(mm, vma_pages(vma))))
2109 vma_link(mm, vma, prev, rb_link, rb_parent);
2114 * Copy the vma structure to a new location in the same mm,
2115 * prior to moving page table entries, to effect an mremap move.
2117 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2118 unsigned long addr, unsigned long len, pgoff_t pgoff)
2120 struct vm_area_struct *vma = *vmap;
2121 unsigned long vma_start = vma->vm_start;
2122 struct mm_struct *mm = vma->vm_mm;
2123 struct vm_area_struct *new_vma, *prev;
2124 struct rb_node **rb_link, *rb_parent;
2125 struct mempolicy *pol;
2128 * If anonymous vma has not yet been faulted, update new pgoff
2129 * to match new location, to increase its chance of merging.
2131 if (!vma->vm_file && !vma->anon_vma)
2132 pgoff = addr >> PAGE_SHIFT;
2134 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2135 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2136 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2139 * Source vma may have been merged into new_vma
2141 if (vma_start >= new_vma->vm_start &&
2142 vma_start < new_vma->vm_end)
2145 new_vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
2148 pol = mpol_copy(vma_policy(vma));
2150 kmem_cache_free(vm_area_cachep, new_vma);
2153 vma_set_policy(new_vma, pol);
2154 new_vma->vm_start = addr;
2155 new_vma->vm_end = addr + len;
2156 new_vma->vm_pgoff = pgoff;
2157 if (new_vma->vm_file)
2158 get_file(new_vma->vm_file);
2159 if (new_vma->vm_ops && new_vma->vm_ops->open)
2160 new_vma->vm_ops->open(new_vma);
2161 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2168 * Return true if the calling process may expand its vm space by the passed
2171 int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2173 unsigned long cur = mm->total_vm; /* pages */
2176 lim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
2178 if (cur + npages > lim)
2180 if (!vx_vmpages_avail(mm, npages))
2186 static struct page *
2187 special_mapping_nopage(struct vm_area_struct *vma,
2188 unsigned long address, int *type)
2190 struct page **pages;
2192 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
2194 address -= vma->vm_start;
2195 for (pages = vma->vm_private_data; address > 0 && *pages; ++pages)
2196 address -= PAGE_SIZE;
2203 return NOPAGE_SIGBUS;
2206 static struct vm_operations_struct special_mapping_vmops = {
2207 .nopage = special_mapping_nopage,
2210 unsigned int vdso_populate = 1;
2213 * Insert a new vma covering the given region, with the given flags and
2214 * protections. Its pages are supplied by the given null-terminated array.
2215 * The region past the last page supplied will always produce SIGBUS.
2216 * The array pointer and the pages it points to are assumed to stay alive
2217 * for as long as this mapping might exist.
2219 int install_special_mapping(struct mm_struct *mm,
2220 unsigned long addr, unsigned long len,
2221 unsigned long vm_flags, pgprot_t pgprot,
2222 struct page **pages)
2224 struct vm_area_struct *vma;
2227 vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
2228 if (unlikely(vma == NULL))
2230 memset(vma, 0, sizeof(*vma));
2233 vma->vm_start = addr;
2234 vma->vm_end = addr + len;
2236 vma->vm_flags = vm_flags;
2237 vma->vm_page_prot = pgprot;
2239 vma->vm_ops = &special_mapping_vmops;
2240 vma->vm_private_data = pages;
2242 insert_vm_struct(mm, vma);
2243 mm->total_vm += len >> PAGE_SHIFT;
2250 struct page *page = *pages++;
2252 err = install_page(mm, vma, addr, page, vma->vm_page_prot);